sys/cam/ctl/ctl.c

   1 /*-
   2  * Copyright (c) 2003-2009 Silicon Graphics International Corp.
   3  * Copyright (c) 2012 The FreeBSD Foundation
   4  * All rights reserved.
   5  *
   6  * Portions of this software were developed by Edward Tomasz Napierala
   7  * under sponsorship from the FreeBSD Foundation.
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions
  11  * are met:
  12  * 1. Redistributions of source code must retain the above copyright
  13  *    notice, this list of conditions, and the following disclaimer,
  14  *    without modification.
  15  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
  16  *    substantially similar to the "NO WARRANTY" disclaimer below
  17  *    ("Disclaimer") and any redistribution must be conditioned upon
  18  *    including a substantially similar Disclaimer requirement for further
  19  *    binary redistribution.
  20  *
  21  * NO WARRANTY
  22  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  23  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  24  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
  25  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  26  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
  30  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
  31  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  32  * POSSIBILITY OF SUCH DAMAGES.
  33  *
  34  * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl.c#8 $
  35  */
  36 /*
  37  * CAM Target Layer, a SCSI device emulation subsystem.
  38  *
  39  * Author: Ken Merry <ken@FreeBSD.org>
  40  */
  41
  42 #define _CTL_C
  43
  44 #include <sys/cdefs.h>
  45 __FBSDID("$FreeBSD$");
  46
  47 #include <sys/param.h>
  48 #include <sys/systm.h>
  49 #include <sys/kernel.h>
  50 #include <sys/types.h>
  51 #include <sys/kthread.h>
  52 #include <sys/bio.h>
  53 #include <sys/fcntl.h>
  54 #include <sys/lock.h>
  55 #include <sys/module.h>
  56 #include <sys/mutex.h>
  57 #include <sys/condvar.h>
  58 #include <sys/malloc.h>
  59 #include <sys/conf.h>
  60 #include <sys/ioccom.h>
  61 #include <sys/queue.h>
  62 #include <sys/sbuf.h>
  63 #include <sys/smp.h>
  64 #include <sys/endian.h>
  65 #include <sys/sysctl.h>
  66
  67 #include <cam/cam.h>
  68 #include <cam/scsi/scsi_all.h>
  69 #include <cam/scsi/scsi_da.h>
  70 #include <cam/ctl/ctl_io.h>
  71 #include <cam/ctl/ctl.h>
  72 #include <cam/ctl/ctl_frontend.h>
  73 #include <cam/ctl/ctl_frontend_internal.h>
  74 #include <cam/ctl/ctl_util.h>
  75 #include <cam/ctl/ctl_backend.h>
  76 #include <cam/ctl/ctl_ioctl.h>
  77 #include <cam/ctl/ctl_ha.h>
  78 #include <cam/ctl/ctl_private.h>
  79 #include <cam/ctl/ctl_debug.h>
  80 #include <cam/ctl/ctl_scsi_all.h>
  81 #include <cam/ctl/ctl_error.h>
  82
  83 struct ctl_softc *control_softc = NULL;
  84
  85 /*
  86  * Size and alignment macros needed for Copan-specific HA hardware.  These
  87  * can go away when the HA code is re-written, and uses busdma for any
  88  * hardware.
  89  */
  90 #define CTL_ALIGN_8B(target, source, type)                              \
  91         if (((uint32_t)source & 0x7) != 0)                              \
  92                 target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
  93         else                                                            \
  94                 target = (type)source;
  95
  96 #define CTL_SIZE_8B(target, size)                                       \
  97         if ((size & 0x7) != 0)                                          \
  98                 target = size + (0x8 - (size & 0x7));                   \
  99         else                                                            \
 100                 target = size;
 101
 102 #define CTL_ALIGN_8B_MARGIN     16
 103
 104 /*
 105  * Template mode pages.
 106  */
 107
 108 /*
 109  * Note that these are default values only.  The actual values will be
 110  * filled in when the user does a mode sense.
 111  */
 112 static struct copan_power_subpage power_page_default = {
 113         /*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
 114         /*subpage*/ PWR_SUBPAGE_CODE,
 115         /*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
 116                          (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
 117         /*page_version*/ PWR_VERSION,
 118         /* total_luns */ 26,
 119         /* max_active_luns*/ PWR_DFLT_MAX_LUNS,
 120         /*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
 121                       0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
 122                       0, 0, 0, 0, 0, 0}
 123 };
 124
 125 static struct copan_power_subpage power_page_changeable = {
 126         /*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
 127         /*subpage*/ PWR_SUBPAGE_CODE,
 128         /*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
 129                          (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
 130         /*page_version*/ 0,
 131         /* total_luns */ 0,
 132         /* max_active_luns*/ 0,
 133         /*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
 134                       0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
 135                       0, 0, 0, 0, 0, 0}
 136 };
 137
 138 static struct copan_aps_subpage aps_page_default = {
 139         APS_PAGE_CODE | SMPH_SPF, //page_code
 140         APS_SUBPAGE_CODE, //subpage
 141         {(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
 142          (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
 143         APS_VERSION, //page_version
 144         0, //lock_active
 145         {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
 146         0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
 147         0, 0, 0, 0, 0} //reserved
 148 };
 149
 150 static struct copan_aps_subpage aps_page_changeable = {
 151         APS_PAGE_CODE | SMPH_SPF, //page_code
 152         APS_SUBPAGE_CODE, //subpage
 153         {(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
 154          (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
 155         0, //page_version
 156         0, //lock_active
 157         {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
 158         0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
 159         0, 0, 0, 0, 0} //reserved
 160 };
 161
 162 static struct copan_debugconf_subpage debugconf_page_default = {
 163         DBGCNF_PAGE_CODE | SMPH_SPF,    /* page_code */
 164         DBGCNF_SUBPAGE_CODE,            /* subpage */
 165         {(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
 166          (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
 167         DBGCNF_VERSION,                 /* page_version */
 168         {CTL_TIME_IO_DEFAULT_SECS>>8,
 169          CTL_TIME_IO_DEFAULT_SECS>>0},  /* ctl_time_io_secs */
 170 };
 171
 172 static struct copan_debugconf_subpage debugconf_page_changeable = {
 173         DBGCNF_PAGE_CODE | SMPH_SPF,    /* page_code */
 174         DBGCNF_SUBPAGE_CODE,            /* subpage */
 175         {(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
 176          (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
 177         0,                              /* page_version */
 178         {0xff,0xff},                    /* ctl_time_io_secs */
 179 };
 180
 181 static struct scsi_format_page format_page_default = {
 182         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
 183         /*page_length*/sizeof(struct scsi_format_page) - 2,
 184         /*tracks_per_zone*/ {0, 0},
 185         /*alt_sectors_per_zone*/ {0, 0},
 186         /*alt_tracks_per_zone*/ {0, 0},
 187         /*alt_tracks_per_lun*/ {0, 0},
 188         /*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
 189                                 CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
 190         /*bytes_per_sector*/ {0, 0},
 191         /*interleave*/ {0, 0},
 192         /*track_skew*/ {0, 0},
 193         /*cylinder_skew*/ {0, 0},
 194         /*flags*/ SFP_HSEC,
 195         /*reserved*/ {0, 0, 0}
 196 };
 197
 198 static struct scsi_format_page format_page_changeable = {
 199         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
 200         /*page_length*/sizeof(struct scsi_format_page) - 2,
 201         /*tracks_per_zone*/ {0, 0},
 202         /*alt_sectors_per_zone*/ {0, 0},
 203         /*alt_tracks_per_zone*/ {0, 0},
 204         /*alt_tracks_per_lun*/ {0, 0},
 205         /*sectors_per_track*/ {0, 0},
 206         /*bytes_per_sector*/ {0, 0},
 207         /*interleave*/ {0, 0},
 208         /*track_skew*/ {0, 0},
 209         /*cylinder_skew*/ {0, 0},
 210         /*flags*/ 0,
 211         /*reserved*/ {0, 0, 0}
 212 };
 213
 214 static struct scsi_rigid_disk_page rigid_disk_page_default = {
 215         /*page_code*/SMS_RIGID_DISK_PAGE,
 216         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
 217         /*cylinders*/ {0, 0, 0},
 218         /*heads*/ CTL_DEFAULT_HEADS,
 219         /*start_write_precomp*/ {0, 0, 0},
 220         /*start_reduced_current*/ {0, 0, 0},
 221         /*step_rate*/ {0, 0},
 222         /*landing_zone_cylinder*/ {0, 0, 0},
 223         /*rpl*/ SRDP_RPL_DISABLED,
 224         /*rotational_offset*/ 0,
 225         /*reserved1*/ 0,
 226         /*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
 227                            CTL_DEFAULT_ROTATION_RATE & 0xff},
 228         /*reserved2*/ {0, 0}
 229 };
 230
 231 static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
 232         /*page_code*/SMS_RIGID_DISK_PAGE,
 233         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
 234         /*cylinders*/ {0, 0, 0},
 235         /*heads*/ 0,
 236         /*start_write_precomp*/ {0, 0, 0},
 237         /*start_reduced_current*/ {0, 0, 0},
 238         /*step_rate*/ {0, 0},
 239         /*landing_zone_cylinder*/ {0, 0, 0},
 240         /*rpl*/ 0,
 241         /*rotational_offset*/ 0,
 242         /*reserved1*/ 0,
 243         /*rotation_rate*/ {0, 0},
 244         /*reserved2*/ {0, 0}
 245 };
 246
 247 static struct scsi_caching_page caching_page_default = {
 248         /*page_code*/SMS_CACHING_PAGE,
 249         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 250         /*flags1*/ SCP_DISC | SCP_WCE,
 251         /*ret_priority*/ 0,
 252         /*disable_pf_transfer_len*/ {0xff, 0xff},
 253         /*min_prefetch*/ {0, 0},
 254         /*max_prefetch*/ {0xff, 0xff},
 255         /*max_pf_ceiling*/ {0xff, 0xff},
 256         /*flags2*/ 0,
 257         /*cache_segments*/ 0,
 258         /*cache_seg_size*/ {0, 0},
 259         /*reserved*/ 0,
 260         /*non_cache_seg_size*/ {0, 0, 0}
 261 };
 262
 263 static struct scsi_caching_page caching_page_changeable = {
 264         /*page_code*/SMS_CACHING_PAGE,
 265         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 266         /*flags1*/ 0,
 267         /*ret_priority*/ 0,
 268         /*disable_pf_transfer_len*/ {0, 0},
 269         /*min_prefetch*/ {0, 0},
 270         /*max_prefetch*/ {0, 0},
 271         /*max_pf_ceiling*/ {0, 0},
 272         /*flags2*/ 0,
 273         /*cache_segments*/ 0,
 274         /*cache_seg_size*/ {0, 0},
 275         /*reserved*/ 0,
 276         /*non_cache_seg_size*/ {0, 0, 0}
 277 };
 278
 279 static struct scsi_control_page control_page_default = {
 280         /*page_code*/SMS_CONTROL_MODE_PAGE,
 281         /*page_length*/sizeof(struct scsi_control_page) - 2,
 282         /*rlec*/0,
 283         /*queue_flags*/0,
 284         /*eca_and_aen*/0,
 285         /*reserved*/0,
 286         /*aen_holdoff_period*/{0, 0}
 287 };
 288
 289 static struct scsi_control_page control_page_changeable = {
 290         /*page_code*/SMS_CONTROL_MODE_PAGE,
 291         /*page_length*/sizeof(struct scsi_control_page) - 2,
 292         /*rlec*/SCP_DSENSE,
 293         /*queue_flags*/0,
 294         /*eca_and_aen*/0,
 295         /*reserved*/0,
 296         /*aen_holdoff_period*/{0, 0}
 297 };
 298
 299
 300 /*
 301  * XXX KDM move these into the softc.
 302  */
 303 static int rcv_sync_msg;
 304 static int persis_offset;
 305 static uint8_t ctl_pause_rtr;
 306 static int     ctl_is_single = 1;
 307 static int     index_to_aps_page;
 308
 309 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
 310 static int worker_threads = -1;
 311 TUNABLE_INT("kern.cam.ctl.worker_threads", &worker_threads);
 312 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
 313     &worker_threads, 1, "Number of worker threads");
 314 static int verbose = 0;
 315 TUNABLE_INT("kern.cam.ctl.verbose", &verbose);
 316 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, verbose, CTLFLAG_RWTUN,
 317     &verbose, 0, "Show SCSI errors returned to initiator");
 318
 319 /*
 320  * Serial number (0x80), device id (0x83), supported pages (0x00),
 321  * Block limits (0xB0) and Logical Block Provisioning (0xB2)
 322  */
 323 #define SCSI_EVPD_NUM_SUPPORTED_PAGES   5
 324
 325 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
 326                                   int param);
 327 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
 328 static int ctl_init(void);
 329 void ctl_shutdown(void);
 330 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
 331 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
 332 static void ctl_ioctl_online(void *arg);
 333 static void ctl_ioctl_offline(void *arg);
 334 static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
 335 static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
 336 static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
 337 static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
 338 static int ctl_ioctl_submit_wait(union ctl_io *io);
 339 static void ctl_ioctl_datamove(union ctl_io *io);
 340 static void ctl_ioctl_done(union ctl_io *io);
 341 static void ctl_ioctl_hard_startstop_callback(void *arg,
 342                                               struct cfi_metatask *metatask);
 343 static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
 344 static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
 345                               struct ctl_ooa *ooa_hdr,
 346                               struct ctl_ooa_entry *kern_entries);
 347 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
 348                      struct thread *td);
 349 uint32_t ctl_get_resindex(struct ctl_nexus *nexus);
 350 uint32_t ctl_port_idx(int port_num);
 351 #ifdef unused
 352 static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
 353                                    uint32_t targ_target, uint32_t targ_lun,
 354                                    int can_wait);
 355 static void ctl_kfree_io(union ctl_io *io);
 356 #endif /* unused */
 357 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
 358                          struct ctl_be_lun *be_lun, struct ctl_id target_id);
 359 static int ctl_free_lun(struct ctl_lun *lun);
 360 static void ctl_create_lun(struct ctl_be_lun *be_lun);
 361 /**
 362 static void ctl_failover_change_pages(struct ctl_softc *softc,
 363                                       struct ctl_scsiio *ctsio, int master);
 364 **/
 365
 366 static int ctl_do_mode_select(union ctl_io *io);
 367 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
 368                            uint64_t res_key, uint64_t sa_res_key,
 369                            uint8_t type, uint32_t residx,
 370                            struct ctl_scsiio *ctsio,
 371                            struct scsi_per_res_out *cdb,
 372                            struct scsi_per_res_out_parms* param);
 373 static void ctl_pro_preempt_other(struct ctl_lun *lun,
 374                                   union ctl_ha_msg *msg);
 375 static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
 376 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
 377 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
 378 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
 379 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
 380                                          int alloc_len);
 381 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
 382 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
 383 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
 384 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len);
 385 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
 386 static ctl_action ctl_check_for_blockage(union ctl_io *pending_io,
 387                                          union ctl_io *ooa_io);
 388 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
 389                                 union ctl_io *starting_io);
 390 static int ctl_check_blocked(struct ctl_lun *lun);
 391 static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
 392                                 struct ctl_lun *lun,
 393                                 const struct ctl_cmd_entry *entry,
 394                                 struct ctl_scsiio *ctsio);
 395 //static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
 396 static void ctl_failover(void);
 397 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
 398                                struct ctl_scsiio *ctsio);
 399 static int ctl_scsiio(struct ctl_scsiio *ctsio);
 400
 401 static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
 402 static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
 403                             ctl_ua_type ua_type);
 404 static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
 405                          ctl_ua_type ua_type);
 406 static int ctl_abort_task(union ctl_io *io);
 407 static void ctl_run_task(union ctl_io *io);
 408 #ifdef CTL_IO_DELAY
 409 static void ctl_datamove_timer_wakeup(void *arg);
 410 static void ctl_done_timer_wakeup(void *arg);
 411 #endif /* CTL_IO_DELAY */
 412
 413 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
 414 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
 415 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
 416 static void ctl_datamove_remote_write(union ctl_io *io);
 417 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
 418 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
 419 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
 420 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
 421                                     ctl_ha_dt_cb callback);
 422 static void ctl_datamove_remote_read(union ctl_io *io);
 423 static void ctl_datamove_remote(union ctl_io *io);
 424 static int ctl_process_done(union ctl_io *io);
 425 static void ctl_lun_thread(void *arg);
 426 static void ctl_work_thread(void *arg);
 427 static void ctl_enqueue_incoming(union ctl_io *io);
 428 static void ctl_enqueue_rtr(union ctl_io *io);
 429 static void ctl_enqueue_done(union ctl_io *io);
 430 static void ctl_enqueue_isc(union ctl_io *io);
 431 static const struct ctl_cmd_entry *
 432     ctl_get_cmd_entry(struct ctl_scsiio *ctsio);
 433 static const struct ctl_cmd_entry *
 434     ctl_validate_command(struct ctl_scsiio *ctsio);
 435 static int ctl_cmd_applicable(uint8_t lun_type,
 436     const struct ctl_cmd_entry *entry);
 437
 438 /*
 439  * Load the serialization table.  This isn't very pretty, but is probably
 440  * the easiest way to do it.
 441  */
 442 #include "ctl_ser_table.c"
 443
 444 /*
 445  * We only need to define open, close and ioctl routines for this driver.
 446  */
 447 static struct cdevsw ctl_cdevsw = {
 448         .d_version =    D_VERSION,
 449         .d_flags =      0,
 450         .d_open =       ctl_open,
 451         .d_close =      ctl_close,
 452         .d_ioctl =      ctl_ioctl,
 453         .d_name =       "ctl",
 454 };
 455
 456
 457 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
 458 MALLOC_DEFINE(M_CTLIO, "ctlio", "Memory used for CTL requests");
 459
 460 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
 461
 462 static moduledata_t ctl_moduledata = {
 463         "ctl",
 464         ctl_module_event_handler,
 465         NULL
 466 };
 467
 468 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
 469 MODULE_VERSION(ctl, 1);
 470
 471 static struct ctl_frontend ioctl_frontend =
 472 {
 473         .name = "ioctl",
 474 };
 475
 476 static void
 477 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
 478                             union ctl_ha_msg *msg_info)
 479 {
 480         struct ctl_scsiio *ctsio;
 481
 482         if (msg_info->hdr.original_sc == NULL) {
 483                 printf("%s: original_sc == NULL!\n", __func__);
 484                 /* XXX KDM now what? */
 485                 return;
 486         }
 487
 488         ctsio = &msg_info->hdr.original_sc->scsiio;
 489         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 490         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 491         ctsio->io_hdr.status = msg_info->hdr.status;
 492         ctsio->scsi_status = msg_info->scsi.scsi_status;
 493         ctsio->sense_len = msg_info->scsi.sense_len;
 494         ctsio->sense_residual = msg_info->scsi.sense_residual;
 495         ctsio->residual = msg_info->scsi.residual;
 496         memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
 497                sizeof(ctsio->sense_data));
 498         memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
 499                &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
 500         ctl_enqueue_isc((union ctl_io *)ctsio);
 501 }
 502
 503 static void
 504 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
 505                                 union ctl_ha_msg *msg_info)
 506 {
 507         struct ctl_scsiio *ctsio;
 508
 509         if (msg_info->hdr.serializing_sc == NULL) {
 510                 printf("%s: serializing_sc == NULL!\n", __func__);
 511                 /* XXX KDM now what? */
 512                 return;
 513         }
 514
 515         ctsio = &msg_info->hdr.serializing_sc->scsiio;
 516 #if 0
 517         /*
 518          * Attempt to catch the situation where an I/O has
 519          * been freed, and we're using it again.
 520          */
 521         if (ctsio->io_hdr.io_type == 0xff) {
 522                 union ctl_io *tmp_io;
 523                 tmp_io = (union ctl_io *)ctsio;
 524                 printf("%s: %p use after free!\n", __func__,
 525                        ctsio);
 526                 printf("%s: type %d msg %d cdb %x iptl: "
 527                        "%d:%d:%d:%d tag 0x%04x "
 528                        "flag %#x status %x\n",
 529                         __func__,
 530                         tmp_io->io_hdr.io_type,
 531                         tmp_io->io_hdr.msg_type,
 532                         tmp_io->scsiio.cdb[0],
 533                         tmp_io->io_hdr.nexus.initid.id,
 534                         tmp_io->io_hdr.nexus.targ_port,
 535                         tmp_io->io_hdr.nexus.targ_target.id,
 536                         tmp_io->io_hdr.nexus.targ_lun,
 537                         (tmp_io->io_hdr.io_type ==
 538                         CTL_IO_TASK) ?
 539                         tmp_io->taskio.tag_num :
 540                         tmp_io->scsiio.tag_num,
 541                         tmp_io->io_hdr.flags,
 542                         tmp_io->io_hdr.status);
 543         }
 544 #endif
 545         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 546         ctl_enqueue_isc((union ctl_io *)ctsio);
 547 }
 548
 549 /*
 550  * ISC (Inter Shelf Communication) event handler.  Events from the HA
 551  * subsystem come in here.
 552  */
 553 static void
 554 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
 555 {
 556         struct ctl_softc *ctl_softc;
 557         union ctl_io *io;
 558         struct ctl_prio *presio;
 559         ctl_ha_status isc_status;
 560
 561         ctl_softc = control_softc;
 562         io = NULL;
 563
 564
 565 #if 0
 566         printf("CTL: Isc Msg event %d\n", event);
 567 #endif
 568         if (event == CTL_HA_EVT_MSG_RECV) {
 569                 union ctl_ha_msg msg_info;
 570
 571                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
 572                                              sizeof(msg_info), /*wait*/ 0);
 573 #if 0
 574                 printf("CTL: msg_type %d\n", msg_info.msg_type);
 575 #endif
 576                 if (isc_status != 0) {
 577                         printf("Error receiving message, status = %d\n",
 578                                isc_status);
 579                         return;
 580                 }
 581
 582                 switch (msg_info.hdr.msg_type) {
 583                 case CTL_MSG_SERIALIZE:
 584 #if 0
 585                         printf("Serialize\n");
 586 #endif
 587                         io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
 588                         if (io == NULL) {
 589                                 printf("ctl_isc_event_handler: can't allocate "
 590                                        "ctl_io!\n");
 591                                 /* Bad Juju */
 592                                 /* Need to set busy and send msg back */
 593                                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
 594                                 msg_info.hdr.status = CTL_SCSI_ERROR;
 595                                 msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
 596                                 msg_info.scsi.sense_len = 0;
 597                                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
 598                                     sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
 599                                 }
 600                                 goto bailout;
 601                         }
 602                         ctl_zero_io(io);
 603                         // populate ctsio from msg_info
 604                         io->io_hdr.io_type = CTL_IO_SCSI;
 605                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
 606                         io->io_hdr.original_sc = msg_info.hdr.original_sc;
 607 #if 0
 608                         printf("pOrig %x\n", (int)msg_info.original_sc);
 609 #endif
 610                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
 611                                             CTL_FLAG_IO_ACTIVE;
 612                         /*
 613                          * If we're in serialization-only mode, we don't
 614                          * want to go through full done processing.  Thus
 615                          * the COPY flag.
 616                          *
 617                          * XXX KDM add another flag that is more specific.
 618                          */
 619                         if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
 620                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
 621                         io->io_hdr.nexus = msg_info.hdr.nexus;
 622 #if 0
 623                         printf("targ %d, port %d, iid %d, lun %d\n",
 624                                io->io_hdr.nexus.targ_target.id,
 625                                io->io_hdr.nexus.targ_port,
 626                                io->io_hdr.nexus.initid.id,
 627                                io->io_hdr.nexus.targ_lun);
 628 #endif
 629                         io->scsiio.tag_num = msg_info.scsi.tag_num;
 630                         io->scsiio.tag_type = msg_info.scsi.tag_type;
 631                         memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
 632                                CTL_MAX_CDBLEN);
 633                         if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
 634                                 const struct ctl_cmd_entry *entry;
 635
 636                                 entry = ctl_get_cmd_entry(&io->scsiio);
 637                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
 638                                 io->io_hdr.flags |=
 639                                         entry->flags & CTL_FLAG_DATA_MASK;
 640                         }
 641                         ctl_enqueue_isc(io);
 642                         break;
 643
 644                 /* Performed on the Originating SC, XFER mode only */
 645                 case CTL_MSG_DATAMOVE: {
 646                         struct ctl_sg_entry *sgl;
 647                         int i, j;
 648
 649                         io = msg_info.hdr.original_sc;
 650                         if (io == NULL) {
 651                                 printf("%s: original_sc == NULL!\n", __func__);
 652                                 /* XXX KDM do something here */
 653                                 break;
 654                         }
 655                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
 656                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 657                         /*
 658                          * Keep track of this, we need to send it back over
 659                          * when the datamove is complete.
 660                          */
 661                         io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
 662
 663                         if (msg_info.dt.sg_sequence == 0) {
 664                                 /*
 665                                  * XXX KDM we use the preallocated S/G list
 666                                  * here, but we'll need to change this to
 667                                  * dynamic allocation if we need larger S/G
 668                                  * lists.
 669                                  */
 670                                 if (msg_info.dt.kern_sg_entries >
 671                                     sizeof(io->io_hdr.remote_sglist) /
 672                                     sizeof(io->io_hdr.remote_sglist[0])) {
 673                                         printf("%s: number of S/G entries "
 674                                             "needed %u > allocated num %zd\n",
 675                                             __func__,
 676                                             msg_info.dt.kern_sg_entries,
 677                                             sizeof(io->io_hdr.remote_sglist)/
 678                                             sizeof(io->io_hdr.remote_sglist[0]));
 679
 680                                         /*
 681                                          * XXX KDM send a message back to
 682                                          * the other side to shut down the
 683                                          * DMA.  The error will come back
 684                                          * through via the normal channel.
 685                                          */
 686                                         break;
 687                                 }
 688                                 sgl = io->io_hdr.remote_sglist;
 689                                 memset(sgl, 0,
 690                                        sizeof(io->io_hdr.remote_sglist));
 691
 692                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
 693
 694                                 io->scsiio.kern_sg_entries =
 695                                         msg_info.dt.kern_sg_entries;
 696                                 io->scsiio.rem_sg_entries =
 697                                         msg_info.dt.kern_sg_entries;
 698                                 io->scsiio.kern_data_len =
 699                                         msg_info.dt.kern_data_len;
 700                                 io->scsiio.kern_total_len =
 701                                         msg_info.dt.kern_total_len;
 702                                 io->scsiio.kern_data_resid =
 703                                         msg_info.dt.kern_data_resid;
 704                                 io->scsiio.kern_rel_offset =
 705                                         msg_info.dt.kern_rel_offset;
 706                                 /*
 707                                  * Clear out per-DMA flags.
 708                                  */
 709                                 io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
 710                                 /*
 711                                  * Add per-DMA flags that are set for this
 712                                  * particular DMA request.
 713                                  */
 714                                 io->io_hdr.flags |= msg_info.dt.flags &
 715                                                     CTL_FLAG_RDMA_MASK;
 716                         } else
 717                                 sgl = (struct ctl_sg_entry *)
 718                                         io->scsiio.kern_data_ptr;
 719
 720                         for (i = msg_info.dt.sent_sg_entries, j = 0;
 721                              i < (msg_info.dt.sent_sg_entries +
 722                              msg_info.dt.cur_sg_entries); i++, j++) {
 723                                 sgl[i].addr = msg_info.dt.sg_list[j].addr;
 724                                 sgl[i].len = msg_info.dt.sg_list[j].len;
 725
 726 #if 0
 727                                 printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
 728                                        __func__,
 729                                        msg_info.dt.sg_list[j].addr,
 730                                        msg_info.dt.sg_list[j].len,
 731                                        sgl[i].addr, sgl[i].len, j, i);
 732 #endif
 733                         }
 734 #if 0
 735                         memcpy(&sgl[msg_info.dt.sent_sg_entries],
 736                                msg_info.dt.sg_list,
 737                                sizeof(*sgl) * msg_info.dt.cur_sg_entries);
 738 #endif
 739
 740                         /*
 741                          * If this is the last piece of the I/O, we've got
 742                          * the full S/G list.  Queue processing in the thread.
 743                          * Otherwise wait for the next piece.
 744                          */
 745                         if (msg_info.dt.sg_last != 0)
 746                                 ctl_enqueue_isc(io);
 747                         break;
 748                 }
 749                 /* Performed on the Serializing (primary) SC, XFER mode only */
 750                 case CTL_MSG_DATAMOVE_DONE: {
 751                         if (msg_info.hdr.serializing_sc == NULL) {
 752                                 printf("%s: serializing_sc == NULL!\n",
 753                                        __func__);
 754                                 /* XXX KDM now what? */
 755                                 break;
 756                         }
 757                         /*
 758                          * We grab the sense information here in case
 759                          * there was a failure, so we can return status
 760                          * back to the initiator.
 761                          */
 762                         io = msg_info.hdr.serializing_sc;
 763                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
 764                         io->io_hdr.status = msg_info.hdr.status;
 765                         io->scsiio.scsi_status = msg_info.scsi.scsi_status;
 766                         io->scsiio.sense_len = msg_info.scsi.sense_len;
 767                         io->scsiio.sense_residual =msg_info.scsi.sense_residual;
 768                         io->io_hdr.port_status = msg_info.scsi.fetd_status;
 769                         io->scsiio.residual = msg_info.scsi.residual;
 770                         memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
 771                                sizeof(io->scsiio.sense_data));
 772                         ctl_enqueue_isc(io);
 773                         break;
 774                 }
 775
 776                 /* Preformed on Originating SC, SER_ONLY mode */
 777                 case CTL_MSG_R2R:
 778                         io = msg_info.hdr.original_sc;
 779                         if (io == NULL) {
 780                                 printf("%s: Major Bummer\n", __func__);
 781                                 return;
 782                         } else {
 783 #if 0
 784                                 printf("pOrig %x\n",(int) ctsio);
 785 #endif
 786                         }
 787                         io->io_hdr.msg_type = CTL_MSG_R2R;
 788                         io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
 789                         ctl_enqueue_isc(io);
 790                         break;
 791
 792                 /*
 793                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
 794                  * mode.
 795                  * Performed on the Originating (i.e. secondary) SC in XFER
 796                  * mode
 797                  */
 798                 case CTL_MSG_FINISH_IO:
 799                         if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
 800                                 ctl_isc_handler_finish_xfer(ctl_softc,
 801                                                             &msg_info);
 802                         else
 803                                 ctl_isc_handler_finish_ser_only(ctl_softc,
 804                                                                 &msg_info);
 805                         break;
 806
 807                 /* Preformed on Originating SC */
 808                 case CTL_MSG_BAD_JUJU:
 809                         io = msg_info.hdr.original_sc;
 810                         if (io == NULL) {
 811                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
 812                                        __func__);
 813                                 break;
 814                         }
 815                         ctl_copy_sense_data(&msg_info, io);
 816                         /*
 817                          * IO should have already been cleaned up on other
 818                          * SC so clear this flag so we won't send a message
 819                          * back to finish the IO there.
 820                          */
 821                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
 822                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 823
 824                         /* io = msg_info.hdr.serializing_sc; */
 825                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
 826                         ctl_enqueue_isc(io);
 827                         break;
 828
 829                 /* Handle resets sent from the other side */
 830                 case CTL_MSG_MANAGE_TASKS: {
 831                         struct ctl_taskio *taskio;
 832                         taskio = (struct ctl_taskio *)ctl_alloc_io(
 833                                 (void *)ctl_softc->othersc_pool);
 834                         if (taskio == NULL) {
 835                                 printf("ctl_isc_event_handler: can't allocate "
 836                                        "ctl_io!\n");
 837                                 /* Bad Juju */
 838                                 /* should I just call the proper reset func
 839                                    here??? */
 840                                 goto bailout;
 841                         }
 842                         ctl_zero_io((union ctl_io *)taskio);
 843                         taskio->io_hdr.io_type = CTL_IO_TASK;
 844                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
 845                         taskio->io_hdr.nexus = msg_info.hdr.nexus;
 846                         taskio->task_action = msg_info.task.task_action;
 847                         taskio->tag_num = msg_info.task.tag_num;
 848                         taskio->tag_type = msg_info.task.tag_type;
 849 #ifdef CTL_TIME_IO
 850                         taskio->io_hdr.start_time = time_uptime;
 851                         getbintime(&taskio->io_hdr.start_bt);
 852 #if 0
 853                         cs_prof_gettime(&taskio->io_hdr.start_ticks);
 854 #endif
 855 #endif /* CTL_TIME_IO */
 856                         ctl_run_task((union ctl_io *)taskio);
 857                         break;
 858                 }
 859                 /* Persistent Reserve action which needs attention */
 860                 case CTL_MSG_PERS_ACTION:
 861                         presio = (struct ctl_prio *)ctl_alloc_io(
 862                                 (void *)ctl_softc->othersc_pool);
 863                         if (presio == NULL) {
 864                                 printf("ctl_isc_event_handler: can't allocate "
 865                                        "ctl_io!\n");
 866                                 /* Bad Juju */
 867                                 /* Need to set busy and send msg back */
 868                                 goto bailout;
 869                         }
 870                         ctl_zero_io((union ctl_io *)presio);
 871                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
 872                         presio->pr_msg = msg_info.pr;
 873                         ctl_enqueue_isc((union ctl_io *)presio);
 874                         break;
 875                 case CTL_MSG_SYNC_FE:
 876                         rcv_sync_msg = 1;
 877                         break;
 878                 case CTL_MSG_APS_LOCK: {
 879                         // It's quicker to execute this then to
 880                         // queue it.
 881                         struct ctl_lun *lun;
 882                         struct ctl_page_index *page_index;
 883                         struct copan_aps_subpage *current_sp;
 884                         uint32_t targ_lun;
 885
 886                         targ_lun = msg_info.hdr.nexus.targ_mapped_lun;
 887                         lun = ctl_softc->ctl_luns[targ_lun];
 888                         mtx_lock(&lun->lun_lock);
 889                         page_index = &lun->mode_pages.index[index_to_aps_page];
 890                         current_sp = (struct copan_aps_subpage *)
 891                                      (page_index->page_data +
 892                                      (page_index->page_len * CTL_PAGE_CURRENT));
 893
 894                         current_sp->lock_active = msg_info.aps.lock_flag;
 895                         mtx_unlock(&lun->lun_lock);
 896                         break;
 897                 }
 898                 default:
 899                         printf("How did I get here?\n");
 900                 }
 901         } else if (event == CTL_HA_EVT_MSG_SENT) {
 902                 if (param != CTL_HA_STATUS_SUCCESS) {
 903                         printf("Bad status from ctl_ha_msg_send status %d\n",
 904                                param);
 905                 }
 906                 return;
 907         } else if (event == CTL_HA_EVT_DISCONNECT) {
 908                 printf("CTL: Got a disconnect from Isc\n");
 909                 return;
 910         } else {
 911                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
 912                 return;
 913         }
 914
 915 bailout:
 916         return;
 917 }
 918
 919 static void
 920 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
 921 {
 922         struct scsi_sense_data *sense;
 923
 924         sense = &dest->scsiio.sense_data;
 925         bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
 926         dest->scsiio.scsi_status = src->scsi.scsi_status;
 927         dest->scsiio.sense_len = src->scsi.sense_len;
 928         dest->io_hdr.status = src->hdr.status;
 929 }
 930
 931 static int
 932 ctl_init(void)
 933 {
 934         struct ctl_softc *softc;
 935         struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
 936         struct ctl_port *port;
 937         uint8_t sc_id =0;
 938         int i, error, retval;
 939         //int isc_retval;
 940
 941         retval = 0;
 942         ctl_pause_rtr = 0;
 943         rcv_sync_msg = 0;
 944
 945         control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
 946                                M_WAITOK | M_ZERO);
 947         softc = control_softc;
 948
 949         softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
 950                               "cam/ctl");
 951
 952         softc->dev->si_drv1 = softc;
 953
 954         /*
 955          * By default, return a "bad LUN" peripheral qualifier for unknown
 956          * LUNs.  The user can override this default using the tunable or
 957          * sysctl.  See the comment in ctl_inquiry_std() for more details.
 958          */
 959         softc->inquiry_pq_no_lun = 1;
 960         TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
 961                           &softc->inquiry_pq_no_lun);
 962         sysctl_ctx_init(&softc->sysctl_ctx);
 963         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
 964                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
 965                 CTLFLAG_RD, 0, "CAM Target Layer");
 966
 967         if (softc->sysctl_tree == NULL) {
 968                 printf("%s: unable to allocate sysctl tree\n", __func__);
 969                 destroy_dev(softc->dev);
 970                 free(control_softc, M_DEVBUF);
 971                 control_softc = NULL;
 972                 return (ENOMEM);
 973         }
 974
 975         SYSCTL_ADD_INT(&softc->sysctl_ctx,
 976                        SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
 977                        "inquiry_pq_no_lun", CTLFLAG_RW,
 978                        &softc->inquiry_pq_no_lun, 0,
 979                        "Report no lun possible for invalid LUNs");
 980
 981         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
 982         mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
 983         softc->open_count = 0;
 984
 985         /*
 986          * Default to actually sending a SYNCHRONIZE CACHE command down to
 987          * the drive.
 988          */
 989         softc->flags = CTL_FLAG_REAL_SYNC;
 990
 991         /*
 992          * In Copan's HA scheme, the "master" and "slave" roles are
 993          * figured out through the slot the controller is in.  Although it
 994          * is an active/active system, someone has to be in charge.
 995          */
 996 #ifdef NEEDTOPORT
 997         scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
 998 #endif
 999
1000         if (sc_id == 0) {
1001                 softc->flags |= CTL_FLAG_MASTER_SHELF;
1002                 persis_offset = 0;
1003         } else
1004                 persis_offset = CTL_MAX_INITIATORS;
1005
1006         /*
1007          * XXX KDM need to figure out where we want to get our target ID
1008          * and WWID.  Is it different on each port?
1009          */
1010         softc->target.id = 0;
1011         softc->target.wwid[0] = 0x12345678;
1012         softc->target.wwid[1] = 0x87654321;
1013         STAILQ_INIT(&softc->lun_list);
1014         STAILQ_INIT(&softc->pending_lun_queue);
1015         STAILQ_INIT(&softc->fe_list);
1016         STAILQ_INIT(&softc->port_list);
1017         STAILQ_INIT(&softc->be_list);
1018         STAILQ_INIT(&softc->io_pools);
1019
1020         if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1021                             &internal_pool)!= 0){
1022                 printf("ctl: can't allocate %d entry internal pool, "
1023                        "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1024                 return (ENOMEM);
1025         }
1026
1027         if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1028                             CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1029                 printf("ctl: can't allocate %d entry emergency pool, "
1030                        "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1031                 ctl_pool_free(internal_pool);
1032                 return (ENOMEM);
1033         }
1034
1035         if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1036                             &other_pool) != 0)
1037         {
1038                 printf("ctl: can't allocate %d entry other SC pool, "
1039                        "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1040                 ctl_pool_free(internal_pool);
1041                 ctl_pool_free(emergency_pool);
1042                 return (ENOMEM);
1043         }
1044
1045         softc->internal_pool = internal_pool;
1046         softc->emergency_pool = emergency_pool;
1047         softc->othersc_pool = other_pool;
1048
1049         if (worker_threads <= 0)
1050                 worker_threads = max(1, mp_ncpus / 4);
1051         if (worker_threads > CTL_MAX_THREADS)
1052                 worker_threads = CTL_MAX_THREADS;
1053
1054         for (i = 0; i < worker_threads; i++) {
1055                 struct ctl_thread *thr = &softc->threads[i];
1056
1057                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1058                 thr->ctl_softc = softc;
1059                 STAILQ_INIT(&thr->incoming_queue);
1060                 STAILQ_INIT(&thr->rtr_queue);
1061                 STAILQ_INIT(&thr->done_queue);
1062                 STAILQ_INIT(&thr->isc_queue);
1063
1064                 error = kproc_kthread_add(ctl_work_thread, thr,
1065                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1066                 if (error != 0) {
1067                         printf("error creating CTL work thread!\n");
1068                         ctl_pool_free(internal_pool);
1069                         ctl_pool_free(emergency_pool);
1070                         ctl_pool_free(other_pool);
1071                         return (error);
1072                 }
1073         }
1074         error = kproc_kthread_add(ctl_lun_thread, softc,
1075             &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1076         if (error != 0) {
1077                 printf("error creating CTL lun thread!\n");
1078                 ctl_pool_free(internal_pool);
1079                 ctl_pool_free(emergency_pool);
1080                 ctl_pool_free(other_pool);
1081                 return (error);
1082         }
1083         if (bootverbose)
1084                 printf("ctl: CAM Target Layer loaded\n");
1085
1086         /*
1087          * Initialize the initiator and portname mappings
1088          */
1089         memset(softc->wwpn_iid, 0, sizeof(softc->wwpn_iid));
1090
1091         /*
1092          * Initialize the ioctl front end.
1093          */
1094         ctl_frontend_register(&ioctl_frontend);
1095         port = &softc->ioctl_info.port;
1096         port->frontend = &ioctl_frontend;
1097         sprintf(softc->ioctl_info.port_name, "ioctl");
1098         port->port_type = CTL_PORT_IOCTL;
1099         port->num_requested_ctl_io = 100;
1100         port->port_name = softc->ioctl_info.port_name;
1101         port->port_online = ctl_ioctl_online;
1102         port->port_offline = ctl_ioctl_offline;
1103         port->onoff_arg = &softc->ioctl_info;
1104         port->lun_enable = ctl_ioctl_lun_enable;
1105         port->lun_disable = ctl_ioctl_lun_disable;
1106         port->targ_lun_arg = &softc->ioctl_info;
1107         port->fe_datamove = ctl_ioctl_datamove;
1108         port->fe_done = ctl_ioctl_done;
1109         port->max_targets = 15;
1110         port->max_target_id = 15;
1111
1112         if (ctl_port_register(&softc->ioctl_info.port,
1113                           (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1114                 printf("ctl: ioctl front end registration failed, will "
1115                        "continue anyway\n");
1116         }
1117
1118 #ifdef CTL_IO_DELAY
1119         if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1120                 printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1121                        sizeof(struct callout), CTL_TIMER_BYTES);
1122                 return (EINVAL);
1123         }
1124 #endif /* CTL_IO_DELAY */
1125
1126         return (0);
1127 }
1128
1129 void
1130 ctl_shutdown(void)
1131 {
1132         struct ctl_softc *softc;
1133         struct ctl_lun *lun, *next_lun;
1134         struct ctl_io_pool *pool;
1135
1136         softc = (struct ctl_softc *)control_softc;
1137
1138         if (ctl_port_deregister(&softc->ioctl_info.port) != 0)
1139                 printf("ctl: ioctl front end deregistration failed\n");
1140
1141         mtx_lock(&softc->ctl_lock);
1142
1143         /*
1144          * Free up each LUN.
1145          */
1146         for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1147                 next_lun = STAILQ_NEXT(lun, links);
1148                 ctl_free_lun(lun);
1149         }
1150
1151         mtx_unlock(&softc->ctl_lock);
1152
1153         ctl_frontend_deregister(&ioctl_frontend);
1154
1155         /*
1156          * This will rip the rug out from under any FETDs or anyone else
1157          * that has a pool allocated.  Since we increment our module
1158          * refcount any time someone outside the main CTL module allocates
1159          * a pool, we shouldn't have any problems here.  The user won't be
1160          * able to unload the CTL module until client modules have
1161          * successfully unloaded.
1162          */
1163         while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1164                 ctl_pool_free(pool);
1165
1166 #if 0
1167         ctl_shutdown_thread(softc->work_thread);
1168         mtx_destroy(&softc->queue_lock);
1169 #endif
1170
1171         mtx_destroy(&softc->pool_lock);
1172         mtx_destroy(&softc->ctl_lock);
1173
1174         destroy_dev(softc->dev);
1175
1176         sysctl_ctx_free(&softc->sysctl_ctx);
1177
1178         free(control_softc, M_DEVBUF);
1179         control_softc = NULL;
1180
1181         if (bootverbose)
1182                 printf("ctl: CAM Target Layer unloaded\n");
1183 }
1184
1185 static int
1186 ctl_module_event_handler(module_t mod, int what, void *arg)
1187 {
1188
1189         switch (what) {
1190         case MOD_LOAD:
1191                 return (ctl_init());
1192         case MOD_UNLOAD:
1193                 return (EBUSY);
1194         default:
1195                 return (EOPNOTSUPP);
1196         }
1197 }
1198
1199 /*
1200  * XXX KDM should we do some access checks here?  Bump a reference count to
1201  * prevent a CTL module from being unloaded while someone has it open?
1202  */
1203 static int
1204 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1205 {
1206         return (0);
1207 }
1208
1209 static int
1210 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1211 {
1212         return (0);
1213 }
1214
1215 int
1216 ctl_port_enable(ctl_port_type port_type)
1217 {
1218         struct ctl_softc *softc;
1219         struct ctl_port *port;
1220
1221         if (ctl_is_single == 0) {
1222                 union ctl_ha_msg msg_info;
1223                 int isc_retval;
1224
1225 #if 0
1226                 printf("%s: HA mode, synchronizing frontend enable\n",
1227                         __func__);
1228 #endif
1229                 msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1230                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1231                         sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1232                         printf("Sync msg send error retval %d\n", isc_retval);
1233                 }
1234                 if (!rcv_sync_msg) {
1235                         isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1236                                 sizeof(msg_info), 1);
1237                 }
1238 #if 0
1239                 printf("CTL:Frontend Enable\n");
1240         } else {
1241                 printf("%s: single mode, skipping frontend synchronization\n",
1242                         __func__);
1243 #endif
1244         }
1245
1246         softc = control_softc;
1247
1248         STAILQ_FOREACH(port, &softc->port_list, links) {
1249                 if (port_type & port->port_type)
1250                 {
1251 #if 0
1252                         printf("port %d\n", port->targ_port);
1253 #endif
1254                         ctl_port_online(port);
1255                 }
1256         }
1257
1258         return (0);
1259 }
1260
1261 int
1262 ctl_port_disable(ctl_port_type port_type)
1263 {
1264         struct ctl_softc *softc;
1265         struct ctl_port *port;
1266
1267         softc = control_softc;
1268
1269         STAILQ_FOREACH(port, &softc->port_list, links) {
1270                 if (port_type & port->port_type)
1271                         ctl_port_offline(port);
1272         }
1273
1274         return (0);
1275 }
1276
1277 /*
1278  * Returns 0 for success, 1 for failure.
1279  * Currently the only failure mode is if there aren't enough entries
1280  * allocated.  So, in case of a failure, look at num_entries_dropped,
1281  * reallocate and try again.
1282  */
1283 int
1284 ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1285               int *num_entries_filled, int *num_entries_dropped,
1286               ctl_port_type port_type, int no_virtual)
1287 {
1288         struct ctl_softc *softc;
1289         struct ctl_port *port;
1290         int entries_dropped, entries_filled;
1291         int retval;
1292         int i;
1293
1294         softc = control_softc;
1295
1296         retval = 0;
1297         entries_filled = 0;
1298         entries_dropped = 0;
1299
1300         i = 0;
1301         mtx_lock(&softc->ctl_lock);
1302         STAILQ_FOREACH(port, &softc->port_list, links) {
1303                 struct ctl_port_entry *entry;
1304
1305                 if ((port->port_type & port_type) == 0)
1306                         continue;
1307
1308                 if ((no_virtual != 0)
1309                  && (port->virtual_port != 0))
1310                         continue;
1311
1312                 if (entries_filled >= num_entries_alloced) {
1313                         entries_dropped++;
1314                         continue;
1315                 }
1316                 entry = &entries[i];
1317
1318                 entry->port_type = port->port_type;
1319                 strlcpy(entry->port_name, port->port_name,
1320                         sizeof(entry->port_name));
1321                 entry->physical_port = port->physical_port;
1322                 entry->virtual_port = port->virtual_port;
1323                 entry->wwnn = port->wwnn;
1324                 entry->wwpn = port->wwpn;
1325
1326                 i++;
1327                 entries_filled++;
1328         }
1329
1330         mtx_unlock(&softc->ctl_lock);
1331
1332         if (entries_dropped > 0)
1333                 retval = 1;
1334
1335         *num_entries_dropped = entries_dropped;
1336         *num_entries_filled = entries_filled;
1337
1338         return (retval);
1339 }
1340
1341 static void
1342 ctl_ioctl_online(void *arg)
1343 {
1344         struct ctl_ioctl_info *ioctl_info;
1345
1346         ioctl_info = (struct ctl_ioctl_info *)arg;
1347
1348         ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1349 }
1350
1351 static void
1352 ctl_ioctl_offline(void *arg)
1353 {
1354         struct ctl_ioctl_info *ioctl_info;
1355
1356         ioctl_info = (struct ctl_ioctl_info *)arg;
1357
1358         ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1359 }
1360
1361 /*
1362  * Remove an initiator by port number and initiator ID.
1363  * Returns 0 for success, 1 for failure.
1364  */
1365 int
1366 ctl_remove_initiator(int32_t targ_port, uint32_t iid)
1367 {
1368         struct ctl_softc *softc;
1369
1370         softc = control_softc;
1371
1372         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1373
1374         if ((targ_port < 0)
1375          || (targ_port > CTL_MAX_PORTS)) {
1376                 printf("%s: invalid port number %d\n", __func__, targ_port);
1377                 return (1);
1378         }
1379         if (iid > CTL_MAX_INIT_PER_PORT) {
1380                 printf("%s: initiator ID %u > maximun %u!\n",
1381                        __func__, iid, CTL_MAX_INIT_PER_PORT);
1382                 return (1);
1383         }
1384
1385         mtx_lock(&softc->ctl_lock);
1386
1387         softc->wwpn_iid[targ_port][iid].in_use = 0;
1388
1389         mtx_unlock(&softc->ctl_lock);
1390
1391         return (0);
1392 }
1393
1394 /*
1395  * Add an initiator to the initiator map.
1396  * Returns 0 for success, 1 for failure.
1397  */
1398 int
1399 ctl_add_initiator(uint64_t wwpn, int32_t targ_port, uint32_t iid)
1400 {
1401         struct ctl_softc *softc;
1402         int retval;
1403
1404         softc = control_softc;
1405
1406         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1407
1408         retval = 0;
1409
1410         if ((targ_port < 0)
1411          || (targ_port > CTL_MAX_PORTS)) {
1412                 printf("%s: invalid port number %d\n", __func__, targ_port);
1413                 return (1);
1414         }
1415         if (iid > CTL_MAX_INIT_PER_PORT) {
1416                 printf("%s: WWPN %#jx initiator ID %u > maximun %u!\n",
1417                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1418                 return (1);
1419         }
1420
1421         mtx_lock(&softc->ctl_lock);
1422
1423         if (softc->wwpn_iid[targ_port][iid].in_use != 0) {
1424                 /*
1425                  * We don't treat this as an error.
1426                  */
1427                 if (softc->wwpn_iid[targ_port][iid].wwpn == wwpn) {
1428                         printf("%s: port %d iid %u WWPN %#jx arrived again?\n",
1429                                __func__, targ_port, iid, (uintmax_t)wwpn);
1430                         goto bailout;
1431                 }
1432
1433                 /*
1434                  * This is an error, but what do we do about it?  The
1435                  * driver is telling us we have a new WWPN for this
1436                  * initiator ID, so we pretty much need to use it.
1437                  */
1438                 printf("%s: port %d iid %u WWPN %#jx arrived, WWPN %#jx is "
1439                        "still at that address\n", __func__, targ_port, iid,
1440                        (uintmax_t)wwpn,
1441                        (uintmax_t)softc->wwpn_iid[targ_port][iid].wwpn);
1442
1443                 /*
1444                  * XXX KDM clear have_ca and ua_pending on each LUN for
1445                  * this initiator.
1446                  */
1447         }
1448         softc->wwpn_iid[targ_port][iid].in_use = 1;
1449         softc->wwpn_iid[targ_port][iid].iid = iid;
1450         softc->wwpn_iid[targ_port][iid].wwpn = wwpn;
1451         softc->wwpn_iid[targ_port][iid].port = targ_port;
1452
1453 bailout:
1454
1455         mtx_unlock(&softc->ctl_lock);
1456
1457         return (retval);
1458 }
1459
1460 static int
1461 ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1462 {
1463         return (0);
1464 }
1465
1466 static int
1467 ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1468 {
1469         return (0);
1470 }
1471
1472 /*
1473  * Data movement routine for the CTL ioctl frontend port.
1474  */
1475 static int
1476 ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1477 {
1478         struct ctl_sg_entry *ext_sglist, *kern_sglist;
1479         struct ctl_sg_entry ext_entry, kern_entry;
1480         int ext_sglen, ext_sg_entries, kern_sg_entries;
1481         int ext_sg_start, ext_offset;
1482         int len_to_copy, len_copied;
1483         int kern_watermark, ext_watermark;
1484         int ext_sglist_malloced;
1485         int i, j;
1486
1487         ext_sglist_malloced = 0;
1488         ext_sg_start = 0;
1489         ext_offset = 0;
1490
1491         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1492
1493         /*
1494          * If this flag is set, fake the data transfer.
1495          */
1496         if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1497                 ctsio->ext_data_filled = ctsio->ext_data_len;
1498                 goto bailout;
1499         }
1500
1501         /*
1502          * To simplify things here, if we have a single buffer, stick it in
1503          * a S/G entry and just make it a single entry S/G list.
1504          */
1505         if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1506                 int len_seen;
1507
1508                 ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1509
1510                 ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1511                                                            M_WAITOK);
1512                 ext_sglist_malloced = 1;
1513                 if (copyin(ctsio->ext_data_ptr, ext_sglist,
1514                                    ext_sglen) != 0) {
1515                         ctl_set_internal_failure(ctsio,
1516                                                  /*sks_valid*/ 0,
1517                                                  /*retry_count*/ 0);
1518                         goto bailout;
1519                 }
1520                 ext_sg_entries = ctsio->ext_sg_entries;
1521                 len_seen = 0;
1522                 for (i = 0; i < ext_sg_entries; i++) {
1523                         if ((len_seen + ext_sglist[i].len) >=
1524                              ctsio->ext_data_filled) {
1525                                 ext_sg_start = i;
1526                                 ext_offset = ctsio->ext_data_filled - len_seen;
1527                                 break;
1528                         }
1529                         len_seen += ext_sglist[i].len;
1530                 }
1531         } else {
1532                 ext_sglist = &ext_entry;
1533                 ext_sglist->addr = ctsio->ext_data_ptr;
1534                 ext_sglist->len = ctsio->ext_data_len;
1535                 ext_sg_entries = 1;
1536                 ext_sg_start = 0;
1537                 ext_offset = ctsio->ext_data_filled;
1538         }
1539
1540         if (ctsio->kern_sg_entries > 0) {
1541                 kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1542                 kern_sg_entries = ctsio->kern_sg_entries;
1543         } else {
1544                 kern_sglist = &kern_entry;
1545                 kern_sglist->addr = ctsio->kern_data_ptr;
1546                 kern_sglist->len = ctsio->kern_data_len;
1547                 kern_sg_entries = 1;
1548         }
1549
1550
1551         kern_watermark = 0;
1552         ext_watermark = ext_offset;
1553         len_copied = 0;
1554         for (i = ext_sg_start, j = 0;
1555              i < ext_sg_entries && j < kern_sg_entries;) {
1556                 uint8_t *ext_ptr, *kern_ptr;
1557
1558                 len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1559                                       kern_sglist[j].len - kern_watermark);
1560
1561                 ext_ptr = (uint8_t *)ext_sglist[i].addr;
1562                 ext_ptr = ext_ptr + ext_watermark;
1563                 if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1564                         /*
1565                          * XXX KDM fix this!
1566                          */
1567                         panic("need to implement bus address support");
1568 #if 0
1569                         kern_ptr = bus_to_virt(kern_sglist[j].addr);
1570 #endif
1571                 } else
1572                         kern_ptr = (uint8_t *)kern_sglist[j].addr;
1573                 kern_ptr = kern_ptr + kern_watermark;
1574
1575                 kern_watermark += len_to_copy;
1576                 ext_watermark += len_to_copy;
1577
1578                 if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1579                      CTL_FLAG_DATA_IN) {
1580                         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1581                                          "bytes to user\n", len_to_copy));
1582                         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1583                                          "to %p\n", kern_ptr, ext_ptr));
1584                         if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1585                                 ctl_set_internal_failure(ctsio,
1586                                                          /*sks_valid*/ 0,
1587                                                          /*retry_count*/ 0);
1588                                 goto bailout;
1589                         }
1590                 } else {
1591                         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1592                                          "bytes from user\n", len_to_copy));
1593                         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1594                                          "to %p\n", ext_ptr, kern_ptr));
1595                         if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1596                                 ctl_set_internal_failure(ctsio,
1597                                                          /*sks_valid*/ 0,
1598                                                          /*retry_count*/0);
1599                                 goto bailout;
1600                         }
1601                 }
1602
1603                 len_copied += len_to_copy;
1604
1605                 if (ext_sglist[i].len == ext_watermark) {
1606                         i++;
1607                         ext_watermark = 0;
1608                 }
1609
1610                 if (kern_sglist[j].len == kern_watermark) {
1611                         j++;
1612                         kern_watermark = 0;
1613                 }
1614         }
1615
1616         ctsio->ext_data_filled += len_copied;
1617
1618         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1619                          "kern_sg_entries: %d\n", ext_sg_entries,
1620                          kern_sg_entries));
1621         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1622                          "kern_data_len = %d\n", ctsio->ext_data_len,
1623                          ctsio->kern_data_len));
1624
1625
1626         /* XXX KDM set residual?? */
1627 bailout:
1628
1629         if (ext_sglist_malloced != 0)
1630                 free(ext_sglist, M_CTL);
1631
1632         return (CTL_RETVAL_COMPLETE);
1633 }
1634
1635 /*
1636  * Serialize a command that went down the "wrong" side, and so was sent to
1637  * this controller for execution.  The logic is a little different than the
1638  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1639  * sent back to the other side, but in the success case, we execute the
1640  * command on this side (XFER mode) or tell the other side to execute it
1641  * (SER_ONLY mode).
1642  */
1643 static int
1644 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1645 {
1646         struct ctl_softc *ctl_softc;
1647         union ctl_ha_msg msg_info;
1648         struct ctl_lun *lun;
1649         int retval = 0;
1650         uint32_t targ_lun;
1651
1652         ctl_softc = control_softc;
1653
1654         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1655         lun = ctl_softc->ctl_luns[targ_lun];
1656         if (lun==NULL)
1657         {
1658                 /*
1659                  * Why isn't LUN defined? The other side wouldn't
1660                  * send a cmd if the LUN is undefined.
1661                  */
1662                 printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1663
1664                 /* "Logical unit not supported" */
1665                 ctl_set_sense_data(&msg_info.scsi.sense_data,
1666                                    lun,
1667                                    /*sense_format*/SSD_TYPE_NONE,
1668                                    /*current_error*/ 1,
1669                                    /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1670                                    /*asc*/ 0x25,
1671                                    /*ascq*/ 0x00,
1672                                    SSD_ELEM_NONE);
1673
1674                 msg_info.scsi.sense_len = SSD_FULL_SIZE;
1675                 msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1676                 msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1677                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1678                 msg_info.hdr.serializing_sc = NULL;
1679                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1680                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1681                                 sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1682                 }
1683                 return(1);
1684
1685         }
1686
1687         mtx_lock(&lun->lun_lock);
1688         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1689
1690         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1691                 (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1692                  ooa_links))) {
1693         case CTL_ACTION_BLOCK:
1694                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1695                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1696                                   blocked_links);
1697                 break;
1698         case CTL_ACTION_PASS:
1699         case CTL_ACTION_SKIP:
1700                 if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1701                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1702                         ctl_enqueue_rtr((union ctl_io *)ctsio);
1703                 } else {
1704
1705                         /* send msg back to other side */
1706                         msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1707                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1708                         msg_info.hdr.msg_type = CTL_MSG_R2R;
1709 #if 0
1710                         printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1711 #endif
1712                         if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1713                             sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1714                         }
1715                 }
1716                 break;
1717         case CTL_ACTION_OVERLAP:
1718                 /* OVERLAPPED COMMANDS ATTEMPTED */
1719                 ctl_set_sense_data(&msg_info.scsi.sense_data,
1720                                    lun,
1721                                    /*sense_format*/SSD_TYPE_NONE,
1722                                    /*current_error*/ 1,
1723                                    /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1724                                    /*asc*/ 0x4E,
1725                                    /*ascq*/ 0x00,
1726                                    SSD_ELEM_NONE);
1727
1728                 msg_info.scsi.sense_len = SSD_FULL_SIZE;
1729                 msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1730                 msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1731                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1732                 msg_info.hdr.serializing_sc = NULL;
1733                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1734 #if 0
1735                 printf("BAD JUJU:Major Bummer Overlap\n");
1736 #endif
1737                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1738                 retval = 1;
1739                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1740                     sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1741                 }
1742                 break;
1743         case CTL_ACTION_OVERLAP_TAG:
1744                 /* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1745                 ctl_set_sense_data(&msg_info.scsi.sense_data,
1746                                    lun,
1747                                    /*sense_format*/SSD_TYPE_NONE,
1748                                    /*current_error*/ 1,
1749                                    /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1750                                    /*asc*/ 0x4D,
1751                                    /*ascq*/ ctsio->tag_num & 0xff,
1752                                    SSD_ELEM_NONE);
1753
1754                 msg_info.scsi.sense_len = SSD_FULL_SIZE;
1755                 msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1756                 msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1757                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1758                 msg_info.hdr.serializing_sc = NULL;
1759                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1760 #if 0
1761                 printf("BAD JUJU:Major Bummer Overlap Tag\n");
1762 #endif
1763                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1764                 retval = 1;
1765                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1766                     sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1767                 }
1768                 break;
1769         case CTL_ACTION_ERROR:
1770         default:
1771                 /* "Internal target failure" */
1772                 ctl_set_sense_data(&msg_info.scsi.sense_data,
1773                                    lun,
1774                                    /*sense_format*/SSD_TYPE_NONE,
1775                                    /*current_error*/ 1,
1776                                    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1777                                    /*asc*/ 0x44,
1778                                    /*ascq*/ 0x00,
1779                                    SSD_ELEM_NONE);
1780
1781                 msg_info.scsi.sense_len = SSD_FULL_SIZE;
1782                 msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1783                 msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1784                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1785                 msg_info.hdr.serializing_sc = NULL;
1786                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1787 #if 0
1788                 printf("BAD JUJU:Major Bummer HW Error\n");
1789 #endif
1790                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1791                 retval = 1;
1792                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1793                     sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1794                 }
1795                 break;
1796         }
1797         mtx_unlock(&lun->lun_lock);
1798         return (retval);
1799 }
1800
1801 static int
1802 ctl_ioctl_submit_wait(union ctl_io *io)
1803 {
1804         struct ctl_fe_ioctl_params params;
1805         ctl_fe_ioctl_state last_state;
1806         int done, retval;
1807
1808         retval = 0;
1809
1810         bzero(&params, sizeof(params));
1811
1812         mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1813         cv_init(&params.sem, "ctlioccv");
1814         params.state = CTL_IOCTL_INPROG;
1815         last_state = params.state;
1816
1817         io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1818
1819         CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1820
1821         /* This shouldn't happen */
1822         if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1823                 return (retval);
1824
1825         done = 0;
1826
1827         do {
1828                 mtx_lock(&params.ioctl_mtx);
1829                 /*
1830                  * Check the state here, and don't sleep if the state has
1831                  * already changed (i.e. wakeup has already occured, but we
1832                  * weren't waiting yet).
1833                  */
1834                 if (params.state == last_state) {
1835                         /* XXX KDM cv_wait_sig instead? */
1836                         cv_wait(&params.sem, &params.ioctl_mtx);
1837                 }
1838                 last_state = params.state;
1839
1840                 switch (params.state) {
1841                 case CTL_IOCTL_INPROG:
1842                         /* Why did we wake up? */
1843                         /* XXX KDM error here? */
1844                         mtx_unlock(&params.ioctl_mtx);
1845                         break;
1846                 case CTL_IOCTL_DATAMOVE:
1847                         CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1848
1849                         /*
1850                          * change last_state back to INPROG to avoid
1851                          * deadlock on subsequent data moves.
1852                          */
1853                         params.state = last_state = CTL_IOCTL_INPROG;
1854
1855                         mtx_unlock(&params.ioctl_mtx);
1856                         ctl_ioctl_do_datamove(&io->scsiio);
1857                         /*
1858                          * Note that in some cases, most notably writes,
1859                          * this will queue the I/O and call us back later.
1860                          * In other cases, generally reads, this routine
1861                          * will immediately call back and wake us up,
1862                          * probably using our own context.
1863                          */
1864                         io->scsiio.be_move_done(io);
1865                         break;
1866                 case CTL_IOCTL_DONE:
1867                         mtx_unlock(&params.ioctl_mtx);
1868                         CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1869                         done = 1;
1870                         break;
1871                 default:
1872                         mtx_unlock(&params.ioctl_mtx);
1873                         /* XXX KDM error here? */
1874                         break;
1875                 }
1876         } while (done == 0);
1877
1878         mtx_destroy(&params.ioctl_mtx);
1879         cv_destroy(&params.sem);
1880
1881         return (CTL_RETVAL_COMPLETE);
1882 }
1883
1884 static void
1885 ctl_ioctl_datamove(union ctl_io *io)
1886 {
1887         struct ctl_fe_ioctl_params *params;
1888
1889         params = (struct ctl_fe_ioctl_params *)
1890                 io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1891
1892         mtx_lock(&params->ioctl_mtx);
1893         params->state = CTL_IOCTL_DATAMOVE;
1894         cv_broadcast(&params->sem);
1895         mtx_unlock(&params->ioctl_mtx);
1896 }
1897
1898 static void
1899 ctl_ioctl_done(union ctl_io *io)
1900 {
1901         struct ctl_fe_ioctl_params *params;
1902
1903         params = (struct ctl_fe_ioctl_params *)
1904                 io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1905
1906         mtx_lock(&params->ioctl_mtx);
1907         params->state = CTL_IOCTL_DONE;
1908         cv_broadcast(&params->sem);
1909         mtx_unlock(&params->ioctl_mtx);
1910 }
1911
1912 static void
1913 ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
1914 {
1915         struct ctl_fe_ioctl_startstop_info *sd_info;
1916
1917         sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
1918
1919         sd_info->hs_info.status = metatask->status;
1920         sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
1921         sd_info->hs_info.luns_complete =
1922                 metatask->taskinfo.startstop.luns_complete;
1923         sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
1924
1925         cv_broadcast(&sd_info->sem);
1926 }
1927
1928 static void
1929 ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
1930 {
1931         struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
1932
1933         fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
1934
1935         mtx_lock(fe_bbr_info->lock);
1936         fe_bbr_info->bbr_info->status = metatask->status;
1937         fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
1938         fe_bbr_info->wakeup_done = 1;
1939         mtx_unlock(fe_bbr_info->lock);
1940
1941         cv_broadcast(&fe_bbr_info->sem);
1942 }
1943
1944 /*
1945  * Returns 0 for success, errno for failure.
1946  */
1947 static int
1948 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
1949                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
1950 {
1951         union ctl_io *io;
1952         int retval;
1953
1954         retval = 0;
1955
1956         mtx_lock(&lun->lun_lock);
1957         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
1958              (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
1959              ooa_links)) {
1960                 struct ctl_ooa_entry *entry;
1961
1962                 /*
1963                  * If we've got more than we can fit, just count the
1964                  * remaining entries.
1965                  */
1966                 if (*cur_fill_num >= ooa_hdr->alloc_num)
1967                         continue;
1968
1969                 entry = &kern_entries[*cur_fill_num];
1970
1971                 entry->tag_num = io->scsiio.tag_num;
1972                 entry->lun_num = lun->lun;
1973 #ifdef CTL_TIME_IO
1974                 entry->start_bt = io->io_hdr.start_bt;
1975 #endif
1976                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
1977                 entry->cdb_len = io->scsiio.cdb_len;
1978                 if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
1979                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
1980
1981                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
1982                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
1983
1984                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
1985                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
1986
1987                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
1988                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
1989
1990                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
1991                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
1992         }
1993         mtx_unlock(&lun->lun_lock);
1994
1995         return (retval);
1996 }
1997
1998 static void *
1999 ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2000                  size_t error_str_len)
2001 {
2002         void *kptr;
2003
2004         kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2005
2006         if (copyin(user_addr, kptr, len) != 0) {
2007                 snprintf(error_str, error_str_len, "Error copying %d bytes "
2008                          "from user address %p to kernel address %p", len,
2009                          user_addr, kptr);
2010                 free(kptr, M_CTL);
2011                 return (NULL);
2012         }
2013
2014         return (kptr);
2015 }
2016
2017 static void
2018 ctl_free_args(int num_be_args, struct ctl_be_arg *be_args)
2019 {
2020         int i;
2021
2022         if (be_args == NULL)
2023                 return;
2024
2025         for (i = 0; i < num_be_args; i++) {
2026                 free(be_args[i].kname, M_CTL);
2027                 free(be_args[i].kvalue, M_CTL);
2028         }
2029
2030         free(be_args, M_CTL);
2031 }
2032
2033 static struct ctl_be_arg *
2034 ctl_copyin_args(int num_be_args, struct ctl_be_arg *be_args,
2035                 char *error_str, size_t error_str_len)
2036 {
2037         struct ctl_be_arg *args;
2038         int i;
2039
2040         args = ctl_copyin_alloc(be_args, num_be_args * sizeof(*be_args),
2041                                 error_str, error_str_len);
2042
2043         if (args == NULL)
2044                 goto bailout;
2045
2046         for (i = 0; i < num_be_args; i++) {
2047                 args[i].kname = NULL;
2048                 args[i].kvalue = NULL;
2049         }
2050
2051         for (i = 0; i < num_be_args; i++) {
2052                 uint8_t *tmpptr;
2053
2054                 args[i].kname = ctl_copyin_alloc(args[i].name,
2055                         args[i].namelen, error_str, error_str_len);
2056                 if (args[i].kname == NULL)
2057                         goto bailout;
2058
2059                 if (args[i].kname[args[i].namelen - 1] != '\0') {
2060                         snprintf(error_str, error_str_len, "Argument %d "
2061                                  "name is not NUL-terminated", i);
2062                         goto bailout;
2063                 }
2064
2065                 args[i].kvalue = NULL;
2066
2067                 tmpptr = ctl_copyin_alloc(args[i].value,
2068                         args[i].vallen, error_str, error_str_len);
2069                 if (tmpptr == NULL)
2070                         goto bailout;
2071
2072                 args[i].kvalue = tmpptr;
2073
2074                 if ((args[i].flags & CTL_BEARG_ASCII)
2075                  && (tmpptr[args[i].vallen - 1] != '\0')) {
2076                         snprintf(error_str, error_str_len, "Argument %d "
2077                                  "value is not NUL-terminated", i);
2078                         goto bailout;
2079                 }
2080         }
2081
2082         return (args);
2083 bailout:
2084
2085         ctl_free_args(num_be_args, args);
2086
2087         return (NULL);
2088 }
2089
2090 /*
2091  * Escape characters that are illegal or not recommended in XML.
2092  */
2093 int
2094 ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2095 {
2096         int retval;
2097
2098         retval = 0;
2099
2100         for (; *str; str++) {
2101                 switch (*str) {
2102                 case '&':
2103                         retval = sbuf_printf(sb, "&amp;");
2104                         break;
2105                 case '>':
2106                         retval = sbuf_printf(sb, "&gt;");
2107                         break;
2108                 case '<':
2109                         retval = sbuf_printf(sb, "&lt;");
2110                         break;
2111                 default:
2112                         retval = sbuf_putc(sb, *str);
2113                         break;
2114                 }
2115
2116                 if (retval != 0)
2117                         break;
2118
2119         }
2120
2121         return (retval);
2122 }
2123
2124 static int
2125 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2126           struct thread *td)
2127 {
2128         struct ctl_softc *softc;
2129         int retval;
2130
2131         softc = control_softc;
2132
2133         retval = 0;
2134
2135         switch (cmd) {
2136         case CTL_IO: {
2137                 union ctl_io *io;
2138                 void *pool_tmp;
2139
2140                 /*
2141                  * If we haven't been "enabled", don't allow any SCSI I/O
2142                  * to this FETD.
2143                  */
2144                 if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2145                         retval = EPERM;
2146                         break;
2147                 }
2148
2149                 io = ctl_alloc_io(softc->ioctl_info.port.ctl_pool_ref);
2150                 if (io == NULL) {
2151                         printf("ctl_ioctl: can't allocate ctl_io!\n");
2152                         retval = ENOSPC;
2153                         break;
2154                 }
2155
2156                 /*
2157                  * Need to save the pool reference so it doesn't get
2158                  * spammed by the user's ctl_io.
2159                  */
2160                 pool_tmp = io->io_hdr.pool;
2161
2162                 memcpy(io, (void *)addr, sizeof(*io));
2163
2164                 io->io_hdr.pool = pool_tmp;
2165                 /*
2166                  * No status yet, so make sure the status is set properly.
2167                  */
2168                 io->io_hdr.status = CTL_STATUS_NONE;
2169
2170                 /*
2171                  * The user sets the initiator ID, target and LUN IDs.
2172                  */
2173                 io->io_hdr.nexus.targ_port = softc->ioctl_info.port.targ_port;
2174                 io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2175                 if ((io->io_hdr.io_type == CTL_IO_SCSI)
2176                  && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2177                         io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2178
2179                 retval = ctl_ioctl_submit_wait(io);
2180
2181                 if (retval != 0) {
2182                         ctl_free_io(io);
2183                         break;
2184                 }
2185
2186                 memcpy((void *)addr, io, sizeof(*io));
2187
2188                 /* return this to our pool */
2189                 ctl_free_io(io);
2190
2191                 break;
2192         }
2193         case CTL_ENABLE_PORT:
2194         case CTL_DISABLE_PORT:
2195         case CTL_SET_PORT_WWNS: {
2196                 struct ctl_port *port;
2197                 struct ctl_port_entry *entry;
2198
2199                 entry = (struct ctl_port_entry *)addr;
2200
2201                 mtx_lock(&softc->ctl_lock);
2202                 STAILQ_FOREACH(port, &softc->port_list, links) {
2203                         int action, done;
2204
2205                         action = 0;
2206                         done = 0;
2207
2208                         if ((entry->port_type == CTL_PORT_NONE)
2209                          && (entry->targ_port == port->targ_port)) {
2210                                 /*
2211                                  * If the user only wants to enable or
2212                                  * disable or set WWNs on a specific port,
2213                                  * do the operation and we're done.
2214                                  */
2215                                 action = 1;
2216                                 done = 1;
2217                         } else if (entry->port_type & port->port_type) {
2218                                 /*
2219                                  * Compare the user's type mask with the
2220                                  * particular frontend type to see if we
2221                                  * have a match.
2222                                  */
2223                                 action = 1;
2224                                 done = 0;
2225
2226                                 /*
2227                                  * Make sure the user isn't trying to set
2228                                  * WWNs on multiple ports at the same time.
2229                                  */
2230                                 if (cmd == CTL_SET_PORT_WWNS) {
2231                                         printf("%s: Can't set WWNs on "
2232                                                "multiple ports\n", __func__);
2233                                         retval = EINVAL;
2234                                         break;
2235                                 }
2236                         }
2237                         if (action != 0) {
2238                                 /*
2239                                  * XXX KDM we have to drop the lock here,
2240                                  * because the online/offline operations
2241                                  * can potentially block.  We need to
2242                                  * reference count the frontends so they
2243                                  * can't go away,
2244                                  */
2245                                 mtx_unlock(&softc->ctl_lock);
2246
2247                                 if (cmd == CTL_ENABLE_PORT) {
2248                                         struct ctl_lun *lun;
2249
2250                                         STAILQ_FOREACH(lun, &softc->lun_list,
2251                                                        links) {
2252                                                 port->lun_enable(port->targ_lun_arg,
2253                                                     lun->target,
2254                                                     lun->lun);
2255                                         }
2256
2257                                         ctl_port_online(port);
2258                                 } else if (cmd == CTL_DISABLE_PORT) {
2259                                         struct ctl_lun *lun;
2260
2261                                         ctl_port_offline(port);
2262
2263                                         STAILQ_FOREACH(lun, &softc->lun_list,
2264                                                        links) {
2265                                                 port->lun_disable(
2266                                                     port->targ_lun_arg,
2267                                                     lun->target,
2268                                                     lun->lun);
2269                                         }
2270                                 }
2271
2272                                 mtx_lock(&softc->ctl_lock);
2273
2274                                 if (cmd == CTL_SET_PORT_WWNS)
2275                                         ctl_port_set_wwns(port,
2276                                             (entry->flags & CTL_PORT_WWNN_VALID) ?
2277                                             1 : 0, entry->wwnn,
2278                                             (entry->flags & CTL_PORT_WWPN_VALID) ?
2279                                             1 : 0, entry->wwpn);
2280                         }
2281                         if (done != 0)
2282                                 break;
2283                 }
2284                 mtx_unlock(&softc->ctl_lock);
2285                 break;
2286         }
2287         case CTL_GET_PORT_LIST: {
2288                 struct ctl_port *port;
2289                 struct ctl_port_list *list;
2290                 int i;
2291
2292                 list = (struct ctl_port_list *)addr;
2293
2294                 if (list->alloc_len != (list->alloc_num *
2295                     sizeof(struct ctl_port_entry))) {
2296                         printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2297                                "alloc_num %u * sizeof(struct ctl_port_entry) "
2298                                "%zu\n", __func__, list->alloc_len,
2299                                list->alloc_num, sizeof(struct ctl_port_entry));
2300                         retval = EINVAL;
2301                         break;
2302                 }
2303                 list->fill_len = 0;
2304                 list->fill_num = 0;
2305                 list->dropped_num = 0;
2306                 i = 0;
2307                 mtx_lock(&softc->ctl_lock);
2308                 STAILQ_FOREACH(port, &softc->port_list, links) {
2309                         struct ctl_port_entry entry, *list_entry;
2310
2311                         if (list->fill_num >= list->alloc_num) {
2312                                 list->dropped_num++;
2313                                 continue;
2314                         }
2315
2316                         entry.port_type = port->port_type;
2317                         strlcpy(entry.port_name, port->port_name,
2318                                 sizeof(entry.port_name));
2319                         entry.targ_port = port->targ_port;
2320                         entry.physical_port = port->physical_port;
2321                         entry.virtual_port = port->virtual_port;
2322                         entry.wwnn = port->wwnn;
2323                         entry.wwpn = port->wwpn;
2324                         if (port->status & CTL_PORT_STATUS_ONLINE)
2325                                 entry.online = 1;
2326                         else
2327                                 entry.online = 0;
2328
2329                         list_entry = &list->entries[i];
2330
2331                         retval = copyout(&entry, list_entry, sizeof(entry));
2332                         if (retval != 0) {
2333                                 printf("%s: CTL_GET_PORT_LIST: copyout "
2334                                        "returned %d\n", __func__, retval);
2335                                 break;
2336                         }
2337                         i++;
2338                         list->fill_num++;
2339                         list->fill_len += sizeof(entry);
2340                 }
2341                 mtx_unlock(&softc->ctl_lock);
2342
2343                 /*
2344                  * If this is non-zero, we had a copyout fault, so there's
2345                  * probably no point in attempting to set the status inside
2346                  * the structure.
2347                  */
2348                 if (retval != 0)
2349                         break;
2350
2351                 if (list->dropped_num > 0)
2352                         list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2353                 else
2354                         list->status = CTL_PORT_LIST_OK;
2355                 break;
2356         }
2357         case CTL_DUMP_OOA: {
2358                 struct ctl_lun *lun;
2359                 union ctl_io *io;
2360                 char printbuf[128];
2361                 struct sbuf sb;
2362
2363                 mtx_lock(&softc->ctl_lock);
2364                 printf("Dumping OOA queues:\n");
2365                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2366                         mtx_lock(&lun->lun_lock);
2367                         for (io = (union ctl_io *)TAILQ_FIRST(
2368                              &lun->ooa_queue); io != NULL;
2369                              io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2370                              ooa_links)) {
2371                                 sbuf_new(&sb, printbuf, sizeof(printbuf),
2372                                          SBUF_FIXEDLEN);
2373                                 sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2374                                             (intmax_t)lun->lun,
2375                                             io->scsiio.tag_num,
2376                                             (io->io_hdr.flags &
2377                                             CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2378                                             (io->io_hdr.flags &
2379                                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2380                                             (io->io_hdr.flags &
2381                                             CTL_FLAG_ABORT) ? " ABORT" : "",
2382                                             (io->io_hdr.flags &
2383                                         CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2384                                 ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2385                                 sbuf_finish(&sb);
2386                                 printf("%s\n", sbuf_data(&sb));
2387                         }
2388                         mtx_unlock(&lun->lun_lock);
2389                 }
2390                 printf("OOA queues dump done\n");
2391                 mtx_unlock(&softc->ctl_lock);
2392                 break;
2393         }
2394         case CTL_GET_OOA: {
2395                 struct ctl_lun *lun;
2396                 struct ctl_ooa *ooa_hdr;
2397                 struct ctl_ooa_entry *entries;
2398                 uint32_t cur_fill_num;
2399
2400                 ooa_hdr = (struct ctl_ooa *)addr;
2401
2402                 if ((ooa_hdr->alloc_len == 0)
2403                  || (ooa_hdr->alloc_num == 0)) {
2404                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2405                                "must be non-zero\n", __func__,
2406                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2407                         retval = EINVAL;
2408                         break;
2409                 }
2410
2411                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2412                     sizeof(struct ctl_ooa_entry))) {
2413                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2414                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2415                                __func__, ooa_hdr->alloc_len,
2416                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2417                         retval = EINVAL;
2418                         break;
2419                 }
2420
2421                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2422                 if (entries == NULL) {
2423                         printf("%s: could not allocate %d bytes for OOA "
2424                                "dump\n", __func__, ooa_hdr->alloc_len);
2425                         retval = ENOMEM;
2426                         break;
2427                 }
2428
2429                 mtx_lock(&softc->ctl_lock);
2430                 if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2431                  && ((ooa_hdr->lun_num > CTL_MAX_LUNS)
2432                   || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2433                         mtx_unlock(&softc->ctl_lock);
2434                         free(entries, M_CTL);
2435                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2436                                __func__, (uintmax_t)ooa_hdr->lun_num);
2437                         retval = EINVAL;
2438                         break;
2439                 }
2440
2441                 cur_fill_num = 0;
2442
2443                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2444                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2445                                 retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2446                                         ooa_hdr, entries);
2447                                 if (retval != 0)
2448                                         break;
2449                         }
2450                         if (retval != 0) {
2451                                 mtx_unlock(&softc->ctl_lock);
2452                                 free(entries, M_CTL);
2453                                 break;
2454                         }
2455                 } else {
2456                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2457
2458                         retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2459                                                     entries);
2460                 }
2461                 mtx_unlock(&softc->ctl_lock);
2462
2463                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2464                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2465                         sizeof(struct ctl_ooa_entry);
2466                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2467                 if (retval != 0) {
2468                         printf("%s: error copying out %d bytes for OOA dump\n",
2469                                __func__, ooa_hdr->fill_len);
2470                 }
2471
2472                 getbintime(&ooa_hdr->cur_bt);
2473
2474                 if (cur_fill_num > ooa_hdr->alloc_num) {
2475                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2476                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2477                 } else {
2478                         ooa_hdr->dropped_num = 0;
2479                         ooa_hdr->status = CTL_OOA_OK;
2480                 }
2481
2482                 free(entries, M_CTL);
2483                 break;
2484         }
2485         case CTL_CHECK_OOA: {
2486                 union ctl_io *io;
2487                 struct ctl_lun *lun;
2488                 struct ctl_ooa_info *ooa_info;
2489
2490
2491                 ooa_info = (struct ctl_ooa_info *)addr;
2492
2493                 if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2494                         ooa_info->status = CTL_OOA_INVALID_LUN;
2495                         break;
2496                 }
2497                 mtx_lock(&softc->ctl_lock);
2498                 lun = softc->ctl_luns[ooa_info->lun_id];
2499                 if (lun == NULL) {
2500                         mtx_unlock(&softc->ctl_lock);
2501                         ooa_info->status = CTL_OOA_INVALID_LUN;
2502                         break;
2503                 }
2504                 mtx_lock(&lun->lun_lock);
2505                 mtx_unlock(&softc->ctl_lock);
2506                 ooa_info->num_entries = 0;
2507                 for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2508                      io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2509                      &io->io_hdr, ooa_links)) {
2510                         ooa_info->num_entries++;
2511                 }
2512                 mtx_unlock(&lun->lun_lock);
2513
2514                 ooa_info->status = CTL_OOA_SUCCESS;
2515
2516                 break;
2517         }
2518         case CTL_HARD_START:
2519         case CTL_HARD_STOP: {
2520                 struct ctl_fe_ioctl_startstop_info ss_info;
2521                 struct cfi_metatask *metatask;
2522                 struct mtx hs_mtx;
2523
2524                 mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2525
2526                 cv_init(&ss_info.sem, "hard start/stop cv" );
2527
2528                 metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2529                 if (metatask == NULL) {
2530                         retval = ENOMEM;
2531                         mtx_destroy(&hs_mtx);
2532                         break;
2533                 }
2534
2535                 if (cmd == CTL_HARD_START)
2536                         metatask->tasktype = CFI_TASK_STARTUP;
2537                 else
2538                         metatask->tasktype = CFI_TASK_SHUTDOWN;
2539
2540                 metatask->callback = ctl_ioctl_hard_startstop_callback;
2541                 metatask->callback_arg = &ss_info;
2542
2543                 cfi_action(metatask);
2544
2545                 /* Wait for the callback */
2546                 mtx_lock(&hs_mtx);
2547                 cv_wait_sig(&ss_info.sem, &hs_mtx);
2548                 mtx_unlock(&hs_mtx);
2549
2550                 /*
2551                  * All information has been copied from the metatask by the
2552                  * time cv_broadcast() is called, so we free the metatask here.
2553                  */
2554                 cfi_free_metatask(metatask);
2555
2556                 memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2557
2558                 mtx_destroy(&hs_mtx);
2559                 break;
2560         }
2561         case CTL_BBRREAD: {
2562                 struct ctl_bbrread_info *bbr_info;
2563                 struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2564                 struct mtx bbr_mtx;
2565                 struct cfi_metatask *metatask;
2566
2567                 bbr_info = (struct ctl_bbrread_info *)addr;
2568
2569                 bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2570
2571                 bzero(&bbr_mtx, sizeof(bbr_mtx));
2572                 mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2573
2574                 fe_bbr_info.bbr_info = bbr_info;
2575                 fe_bbr_info.lock = &bbr_mtx;
2576
2577                 cv_init(&fe_bbr_info.sem, "BBR read cv");
2578                 metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2579
2580                 if (metatask == NULL) {
2581                         mtx_destroy(&bbr_mtx);
2582                         cv_destroy(&fe_bbr_info.sem);
2583                         retval = ENOMEM;
2584                         break;
2585                 }
2586                 metatask->tasktype = CFI_TASK_BBRREAD;
2587                 metatask->callback = ctl_ioctl_bbrread_callback;
2588                 metatask->callback_arg = &fe_bbr_info;
2589                 metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2590                 metatask->taskinfo.bbrread.lba = bbr_info->lba;
2591                 metatask->taskinfo.bbrread.len = bbr_info->len;
2592
2593                 cfi_action(metatask);
2594
2595                 mtx_lock(&bbr_mtx);
2596                 while (fe_bbr_info.wakeup_done == 0)
2597                         cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2598                 mtx_unlock(&bbr_mtx);
2599
2600                 bbr_info->status = metatask->status;
2601                 bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2602                 bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2603                 memcpy(&bbr_info->sense_data,
2604                        &metatask->taskinfo.bbrread.sense_data,
2605                        ctl_min(sizeof(bbr_info->sense_data),
2606                                sizeof(metatask->taskinfo.bbrread.sense_data)));
2607
2608                 cfi_free_metatask(metatask);
2609
2610                 mtx_destroy(&bbr_mtx);
2611                 cv_destroy(&fe_bbr_info.sem);
2612
2613                 break;
2614         }
2615         case CTL_DELAY_IO: {
2616                 struct ctl_io_delay_info *delay_info;
2617 #ifdef CTL_IO_DELAY
2618                 struct ctl_lun *lun;
2619 #endif /* CTL_IO_DELAY */
2620
2621                 delay_info = (struct ctl_io_delay_info *)addr;
2622
2623 #ifdef CTL_IO_DELAY
2624                 mtx_lock(&softc->ctl_lock);
2625
2626                 if ((delay_info->lun_id > CTL_MAX_LUNS)
2627                  || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2628                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2629                 } else {
2630                         lun = softc->ctl_luns[delay_info->lun_id];
2631                         mtx_lock(&lun->lun_lock);
2632
2633                         delay_info->status = CTL_DELAY_STATUS_OK;
2634
2635                         switch (delay_info->delay_type) {
2636                         case CTL_DELAY_TYPE_CONT:
2637                                 break;
2638                         case CTL_DELAY_TYPE_ONESHOT:
2639                                 break;
2640                         default:
2641                                 delay_info->status =
2642                                         CTL_DELAY_STATUS_INVALID_TYPE;
2643                                 break;
2644                         }
2645
2646                         switch (delay_info->delay_loc) {
2647                         case CTL_DELAY_LOC_DATAMOVE:
2648                                 lun->delay_info.datamove_type =
2649                                         delay_info->delay_type;
2650                                 lun->delay_info.datamove_delay =
2651                                         delay_info->delay_secs;
2652                                 break;
2653                         case CTL_DELAY_LOC_DONE:
2654                                 lun->delay_info.done_type =
2655                                         delay_info->delay_type;
2656                                 lun->delay_info.done_delay =
2657                                         delay_info->delay_secs;
2658                                 break;
2659                         default:
2660                                 delay_info->status =
2661                                         CTL_DELAY_STATUS_INVALID_LOC;
2662                                 break;
2663                         }
2664                         mtx_unlock(&lun->lun_lock);
2665                 }
2666
2667                 mtx_unlock(&softc->ctl_lock);
2668 #else
2669                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2670 #endif /* CTL_IO_DELAY */
2671                 break;
2672         }
2673         case CTL_REALSYNC_SET: {
2674                 int *syncstate;
2675
2676                 syncstate = (int *)addr;
2677
2678                 mtx_lock(&softc->ctl_lock);
2679                 switch (*syncstate) {
2680                 case 0:
2681                         softc->flags &= ~CTL_FLAG_REAL_SYNC;
2682                         break;
2683                 case 1:
2684                         softc->flags |= CTL_FLAG_REAL_SYNC;
2685                         break;
2686                 default:
2687                         retval = EINVAL;
2688                         break;
2689                 }
2690                 mtx_unlock(&softc->ctl_lock);
2691                 break;
2692         }
2693         case CTL_REALSYNC_GET: {
2694                 int *syncstate;
2695
2696                 syncstate = (int*)addr;
2697
2698                 mtx_lock(&softc->ctl_lock);
2699                 if (softc->flags & CTL_FLAG_REAL_SYNC)
2700                         *syncstate = 1;
2701                 else
2702                         *syncstate = 0;
2703                 mtx_unlock(&softc->ctl_lock);
2704
2705                 break;
2706         }
2707         case CTL_SETSYNC:
2708         case CTL_GETSYNC: {
2709                 struct ctl_sync_info *sync_info;
2710                 struct ctl_lun *lun;
2711
2712                 sync_info = (struct ctl_sync_info *)addr;
2713
2714                 mtx_lock(&softc->ctl_lock);
2715                 lun = softc->ctl_luns[sync_info->lun_id];
2716                 if (lun == NULL) {
2717                         mtx_unlock(&softc->ctl_lock);
2718                         sync_info->status = CTL_GS_SYNC_NO_LUN;
2719                 }
2720                 /*
2721                  * Get or set the sync interval.  We're not bounds checking
2722                  * in the set case, hopefully the user won't do something
2723                  * silly.
2724                  */
2725                 mtx_lock(&lun->lun_lock);
2726                 mtx_unlock(&softc->ctl_lock);
2727                 if (cmd == CTL_GETSYNC)
2728                         sync_info->sync_interval = lun->sync_interval;
2729                 else
2730                         lun->sync_interval = sync_info->sync_interval;
2731                 mtx_unlock(&lun->lun_lock);
2732
2733                 sync_info->status = CTL_GS_SYNC_OK;
2734
2735                 break;
2736         }
2737         case CTL_GETSTATS: {
2738                 struct ctl_stats *stats;
2739                 struct ctl_lun *lun;
2740                 int i;
2741
2742                 stats = (struct ctl_stats *)addr;
2743
2744                 if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2745                      stats->alloc_len) {
2746                         stats->status = CTL_SS_NEED_MORE_SPACE;
2747                         stats->num_luns = softc->num_luns;
2748                         break;
2749                 }
2750                 /*
2751                  * XXX KDM no locking here.  If the LUN list changes,
2752                  * things can blow up.
2753                  */
2754                 for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2755                      i++, lun = STAILQ_NEXT(lun, links)) {
2756                         retval = copyout(&lun->stats, &stats->lun_stats[i],
2757                                          sizeof(lun->stats));
2758                         if (retval != 0)
2759                                 break;
2760                 }
2761                 stats->num_luns = softc->num_luns;
2762                 stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2763                                  softc->num_luns;
2764                 stats->status = CTL_SS_OK;
2765 #ifdef CTL_TIME_IO
2766                 stats->flags = CTL_STATS_FLAG_TIME_VALID;
2767 #else
2768                 stats->flags = CTL_STATS_FLAG_NONE;
2769 #endif
2770                 getnanouptime(&stats->timestamp);
2771                 break;
2772         }
2773         case CTL_ERROR_INJECT: {
2774                 struct ctl_error_desc *err_desc, *new_err_desc;
2775                 struct ctl_lun *lun;
2776
2777                 err_desc = (struct ctl_error_desc *)addr;
2778
2779                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2780                                       M_WAITOK | M_ZERO);
2781                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2782
2783                 mtx_lock(&softc->ctl_lock);
2784                 lun = softc->ctl_luns[err_desc->lun_id];
2785                 if (lun == NULL) {
2786                         mtx_unlock(&softc->ctl_lock);
2787                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2788                                __func__, (uintmax_t)err_desc->lun_id);
2789                         retval = EINVAL;
2790                         break;
2791                 }
2792                 mtx_lock(&lun->lun_lock);
2793                 mtx_unlock(&softc->ctl_lock);
2794
2795                 /*
2796                  * We could do some checking here to verify the validity
2797                  * of the request, but given the complexity of error
2798                  * injection requests, the checking logic would be fairly
2799                  * complex.
2800                  *
2801                  * For now, if the request is invalid, it just won't get
2802                  * executed and might get deleted.
2803                  */
2804                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2805
2806                 /*
2807                  * XXX KDM check to make sure the serial number is unique,
2808                  * in case we somehow manage to wrap.  That shouldn't
2809                  * happen for a very long time, but it's the right thing to
2810                  * do.
2811                  */
2812                 new_err_desc->serial = lun->error_serial;
2813                 err_desc->serial = lun->error_serial;
2814                 lun->error_serial++;
2815
2816                 mtx_unlock(&lun->lun_lock);
2817                 break;
2818         }
2819         case CTL_ERROR_INJECT_DELETE: {
2820                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2821                 struct ctl_lun *lun;
2822                 int delete_done;
2823
2824                 delete_desc = (struct ctl_error_desc *)addr;
2825                 delete_done = 0;
2826
2827                 mtx_lock(&softc->ctl_lock);
2828                 lun = softc->ctl_luns[delete_desc->lun_id];
2829                 if (lun == NULL) {
2830                         mtx_unlock(&softc->ctl_lock);
2831                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2832                                __func__, (uintmax_t)delete_desc->lun_id);
2833                         retval = EINVAL;
2834                         break;
2835                 }
2836                 mtx_lock(&lun->lun_lock);
2837                 mtx_unlock(&softc->ctl_lock);
2838                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2839                         if (desc->serial != delete_desc->serial)
2840                                 continue;
2841
2842                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2843                                       links);
2844                         free(desc, M_CTL);
2845                         delete_done = 1;
2846                 }
2847                 mtx_unlock(&lun->lun_lock);
2848                 if (delete_done == 0) {
2849                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2850                                "error serial %ju on LUN %u\n", __func__,
2851                                delete_desc->serial, delete_desc->lun_id);
2852                         retval = EINVAL;
2853                         break;
2854                 }
2855                 break;
2856         }
2857         case CTL_DUMP_STRUCTS: {
2858                 int i, j, k;
2859                 struct ctl_port *port;
2860                 struct ctl_frontend *fe;
2861
2862                 printf("CTL IID to WWPN map start:\n");
2863                 for (i = 0; i < CTL_MAX_PORTS; i++) {
2864                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2865                                 if (softc->wwpn_iid[i][j].in_use == 0)
2866                                         continue;
2867
2868                                 printf("port %d iid %u WWPN %#jx\n",
2869                                        softc->wwpn_iid[i][j].port,
2870                                        softc->wwpn_iid[i][j].iid,
2871                                        (uintmax_t)softc->wwpn_iid[i][j].wwpn);
2872                         }
2873                 }
2874                 printf("CTL IID to WWPN map end\n");
2875                 printf("CTL Persistent Reservation information start:\n");
2876                 for (i = 0; i < CTL_MAX_LUNS; i++) {
2877                         struct ctl_lun *lun;
2878
2879                         lun = softc->ctl_luns[i];
2880
2881                         if ((lun == NULL)
2882                          || ((lun->flags & CTL_LUN_DISABLED) != 0))
2883                                 continue;
2884
2885                         for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2886                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2887                                         if (lun->per_res[j+k].registered == 0)
2888                                                 continue;
2889                                         printf("LUN %d port %d iid %d key "
2890                                                "%#jx\n", i, j, k,
2891                                                (uintmax_t)scsi_8btou64(
2892                                                lun->per_res[j+k].res_key.key));
2893                                 }
2894                         }
2895                 }
2896                 printf("CTL Persistent Reservation information end\n");
2897                 printf("CTL Ports:\n");
2898                 /*
2899                  * XXX KDM calling this without a lock.  We'd likely want
2900                  * to drop the lock before calling the frontend's dump
2901                  * routine anyway.
2902                  */
2903                 STAILQ_FOREACH(port, &softc->port_list, links) {
2904                         printf("Port %s Frontend %s Type %u pport %d vport %d WWNN "
2905                                "%#jx WWPN %#jx\n", port->port_name,
2906                                port->frontend->name, port->port_type,
2907                                port->physical_port, port->virtual_port,
2908                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2909                 }
2910                 printf("CTL Port information end\n");
2911                 printf("CTL Frontends:\n");
2912                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
2913                         printf("Frontend %s\n", fe->name);
2914                         if (fe->fe_dump != NULL)
2915                                 fe->fe_dump();
2916                 }
2917                 printf("CTL Frontend information end\n");
2918                 break;
2919         }
2920         case CTL_LUN_REQ: {
2921                 struct ctl_lun_req *lun_req;
2922                 struct ctl_backend_driver *backend;
2923
2924                 lun_req = (struct ctl_lun_req *)addr;
2925
2926                 backend = ctl_backend_find(lun_req->backend);
2927                 if (backend == NULL) {
2928                         lun_req->status = CTL_LUN_ERROR;
2929                         snprintf(lun_req->error_str,
2930                                  sizeof(lun_req->error_str),
2931                                  "Backend \"%s\" not found.",
2932                                  lun_req->backend);
2933                         break;
2934                 }
2935                 if (lun_req->num_be_args > 0) {
2936                         lun_req->kern_be_args = ctl_copyin_args(
2937                                 lun_req->num_be_args,
2938                                 lun_req->be_args,
2939                                 lun_req->error_str,
2940                                 sizeof(lun_req->error_str));
2941                         if (lun_req->kern_be_args == NULL) {
2942                                 lun_req->status = CTL_LUN_ERROR;
2943                                 break;
2944                         }
2945                 }
2946
2947                 retval = backend->ioctl(dev, cmd, addr, flag, td);
2948
2949                 if (lun_req->num_be_args > 0) {
2950                         ctl_free_args(lun_req->num_be_args,
2951                                       lun_req->kern_be_args);
2952                 }
2953                 break;
2954         }
2955         case CTL_LUN_LIST: {
2956                 struct sbuf *sb;
2957                 struct ctl_lun *lun;
2958                 struct ctl_lun_list *list;
2959                 struct ctl_option *opt;
2960
2961                 list = (struct ctl_lun_list *)addr;
2962
2963                 /*
2964                  * Allocate a fixed length sbuf here, based on the length
2965                  * of the user's buffer.  We could allocate an auto-extending
2966                  * buffer, and then tell the user how much larger our
2967                  * amount of data is than his buffer, but that presents
2968                  * some problems:
2969                  *
2970                  * 1.  The sbuf(9) routines use a blocking malloc, and so
2971                  *     we can't hold a lock while calling them with an
2972                  *     auto-extending buffer.
2973                  *
2974                  * 2.  There is not currently a LUN reference counting
2975                  *     mechanism, outside of outstanding transactions on
2976                  *     the LUN's OOA queue.  So a LUN could go away on us
2977                  *     while we're getting the LUN number, backend-specific
2978                  *     information, etc.  Thus, given the way things
2979                  *     currently work, we need to hold the CTL lock while
2980                  *     grabbing LUN information.
2981                  *
2982                  * So, from the user's standpoint, the best thing to do is
2983                  * allocate what he thinks is a reasonable buffer length,
2984                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
2985                  * double the buffer length and try again.  (And repeat
2986                  * that until he succeeds.)
2987                  */
2988                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
2989                 if (sb == NULL) {
2990                         list->status = CTL_LUN_LIST_ERROR;
2991                         snprintf(list->error_str, sizeof(list->error_str),
2992                                  "Unable to allocate %d bytes for LUN list",
2993                                  list->alloc_len);
2994                         break;
2995                 }
2996
2997                 sbuf_printf(sb, "<ctllunlist>\n");
2998
2999                 mtx_lock(&softc->ctl_lock);
3000                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3001                         mtx_lock(&lun->lun_lock);
3002                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3003                                              (uintmax_t)lun->lun);
3004
3005                         /*
3006                          * Bail out as soon as we see that we've overfilled
3007                          * the buffer.
3008                          */
3009                         if (retval != 0)
3010                                 break;
3011
3012                         retval = sbuf_printf(sb, "\t<backend_type>%s"
3013                                              "</backend_type>\n",
3014                                              (lun->backend == NULL) ?  "none" :
3015                                              lun->backend->name);
3016
3017                         if (retval != 0)
3018                                 break;
3019
3020                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3021                                              lun->be_lun->lun_type);
3022
3023                         if (retval != 0)
3024                                 break;
3025
3026                         if (lun->backend == NULL) {
3027                                 retval = sbuf_printf(sb, "</lun>\n");
3028                                 if (retval != 0)
3029                                         break;
3030                                 continue;
3031                         }
3032
3033                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3034                                              (lun->be_lun->maxlba > 0) ?
3035                                              lun->be_lun->maxlba + 1 : 0);
3036
3037                         if (retval != 0)
3038                                 break;
3039
3040                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3041                                              lun->be_lun->blocksize);
3042
3043                         if (retval != 0)
3044                                 break;
3045
3046                         retval = sbuf_printf(sb, "\t<serial_number>");
3047
3048                         if (retval != 0)
3049                                 break;
3050
3051                         retval = ctl_sbuf_printf_esc(sb,
3052                                                      lun->be_lun->serial_num);
3053
3054                         if (retval != 0)
3055                                 break;
3056
3057                         retval = sbuf_printf(sb, "</serial_number>\n");
3058
3059                         if (retval != 0)
3060                                 break;
3061
3062                         retval = sbuf_printf(sb, "\t<device_id>");
3063
3064                         if (retval != 0)
3065                                 break;
3066
3067                         retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3068
3069                         if (retval != 0)
3070                                 break;
3071
3072                         retval = sbuf_printf(sb, "</device_id>\n");
3073
3074                         if (retval != 0)
3075                                 break;
3076
3077                         if (lun->backend->lun_info != NULL) {
3078                                 retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3079                                 if (retval != 0)
3080                                         break;
3081                         }
3082                         STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3083                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3084                                     opt->name, opt->value, opt->name);
3085                                 if (retval != 0)
3086                                         break;
3087                         }
3088
3089                         retval = sbuf_printf(sb, "</lun>\n");
3090
3091                         if (retval != 0)
3092                                 break;
3093                         mtx_unlock(&lun->lun_lock);
3094                 }
3095                 if (lun != NULL)
3096                         mtx_unlock(&lun->lun_lock);
3097                 mtx_unlock(&softc->ctl_lock);
3098
3099                 if ((retval != 0)
3100                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3101                         retval = 0;
3102                         sbuf_delete(sb);
3103                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3104                         snprintf(list->error_str, sizeof(list->error_str),
3105                                  "Out of space, %d bytes is too small",
3106                                  list->alloc_len);
3107                         break;
3108                 }
3109
3110                 sbuf_finish(sb);
3111
3112                 retval = copyout(sbuf_data(sb), list->lun_xml,
3113                                  sbuf_len(sb) + 1);
3114
3115                 list->fill_len = sbuf_len(sb) + 1;
3116                 list->status = CTL_LUN_LIST_OK;
3117                 sbuf_delete(sb);
3118                 break;
3119         }
3120         case CTL_ISCSI: {
3121                 struct ctl_iscsi *ci;
3122                 struct ctl_frontend *fe;
3123
3124                 ci = (struct ctl_iscsi *)addr;
3125
3126                 mtx_lock(&softc->ctl_lock);
3127                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
3128                         if (strcmp(fe->name, "iscsi") == 0)
3129                                 break;
3130                 }
3131                 mtx_unlock(&softc->ctl_lock);
3132
3133                 if (fe == NULL) {
3134                         ci->status = CTL_ISCSI_ERROR;
3135                         snprintf(ci->error_str, sizeof(ci->error_str),
3136                             "Frontend \"iscsi\" not found.");
3137                         break;
3138                 }
3139
3140                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3141                 break;
3142         }
3143         default: {
3144                 /* XXX KDM should we fix this? */
3145 #if 0
3146                 struct ctl_backend_driver *backend;
3147                 unsigned int type;
3148                 int found;
3149
3150                 found = 0;
3151
3152                 /*
3153                  * We encode the backend type as the ioctl type for backend
3154                  * ioctls.  So parse it out here, and then search for a
3155                  * backend of this type.
3156                  */
3157                 type = _IOC_TYPE(cmd);
3158
3159                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3160                         if (backend->type == type) {
3161                                 found = 1;
3162                                 break;
3163                         }
3164                 }
3165                 if (found == 0) {
3166                         printf("ctl: unknown ioctl command %#lx or backend "
3167                                "%d\n", cmd, type);
3168                         retval = EINVAL;
3169                         break;
3170                 }
3171                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3172 #endif
3173                 retval = ENOTTY;
3174                 break;
3175         }
3176         }
3177         return (retval);
3178 }
3179
3180 uint32_t
3181 ctl_get_initindex(struct ctl_nexus *nexus)
3182 {
3183         if (nexus->targ_port < CTL_MAX_PORTS)
3184                 return (nexus->initid.id +
3185                         (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3186         else
3187                 return (nexus->initid.id +
3188                        ((nexus->targ_port - CTL_MAX_PORTS) *
3189                         CTL_MAX_INIT_PER_PORT));
3190 }
3191
3192 uint32_t
3193 ctl_get_resindex(struct ctl_nexus *nexus)
3194 {
3195         return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3196 }
3197
3198 uint32_t
3199 ctl_port_idx(int port_num)
3200 {
3201         if (port_num < CTL_MAX_PORTS)
3202                 return(port_num);
3203         else
3204                 return(port_num - CTL_MAX_PORTS);
3205 }
3206
3207 /*
3208  * Note:  This only works for bitmask sizes that are at least 32 bits, and
3209  * that are a power of 2.
3210  */
3211 int
3212 ctl_ffz(uint32_t *mask, uint32_t size)
3213 {
3214         uint32_t num_chunks, num_pieces;
3215         int i, j;
3216
3217         num_chunks = (size >> 5);
3218         if (num_chunks == 0)
3219                 num_chunks++;
3220         num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3221
3222         for (i = 0; i < num_chunks; i++) {
3223                 for (j = 0; j < num_pieces; j++) {
3224                         if ((mask[i] & (1 << j)) == 0)
3225                                 return ((i << 5) + j);
3226                 }
3227         }
3228
3229         return (-1);
3230 }
3231
3232 int
3233 ctl_set_mask(uint32_t *mask, uint32_t bit)
3234 {
3235         uint32_t chunk, piece;
3236
3237         chunk = bit >> 5;
3238         piece = bit % (sizeof(uint32_t) * 8);
3239
3240         if ((mask[chunk] & (1 << piece)) != 0)
3241                 return (-1);
3242         else
3243                 mask[chunk] |= (1 << piece);
3244
3245         return (0);
3246 }
3247
3248 int
3249 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3250 {
3251         uint32_t chunk, piece;
3252
3253         chunk = bit >> 5;
3254         piece = bit % (sizeof(uint32_t) * 8);
3255
3256         if ((mask[chunk] & (1 << piece)) == 0)
3257                 return (-1);
3258         else
3259                 mask[chunk] &= ~(1 << piece);
3260
3261         return (0);
3262 }
3263
3264 int
3265 ctl_is_set(uint32_t *mask, uint32_t bit)
3266 {
3267         uint32_t chunk, piece;
3268
3269         chunk = bit >> 5;
3270         piece = bit % (sizeof(uint32_t) * 8);
3271
3272         if ((mask[chunk] & (1 << piece)) == 0)
3273                 return (0);
3274         else
3275                 return (1);
3276 }
3277
3278 #ifdef unused
3279 /*
3280  * The bus, target and lun are optional, they can be filled in later.
3281  * can_wait is used to determine whether we can wait on the malloc or not.
3282  */
3283 union ctl_io*
3284 ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3285               uint32_t targ_lun, int can_wait)
3286 {
3287         union ctl_io *io;
3288
3289         if (can_wait)
3290                 io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3291         else
3292                 io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3293
3294         if (io != NULL) {
3295                 io->io_hdr.io_type = io_type;
3296                 io->io_hdr.targ_port = targ_port;
3297                 /*
3298                  * XXX KDM this needs to change/go away.  We need to move
3299                  * to a preallocated pool of ctl_scsiio structures.
3300                  */
3301                 io->io_hdr.nexus.targ_target.id = targ_target;
3302                 io->io_hdr.nexus.targ_lun = targ_lun;
3303         }
3304
3305         return (io);
3306 }
3307
3308 void
3309 ctl_kfree_io(union ctl_io *io)
3310 {
3311         free(io, M_CTL);
3312 }
3313 #endif /* unused */
3314
3315 /*
3316  * ctl_softc, pool_type, total_ctl_io are passed in.
3317  * npool is passed out.
3318  */
3319 int
3320 ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3321                 uint32_t total_ctl_io, struct ctl_io_pool **npool)
3322 {
3323         uint32_t i;
3324         union ctl_io *cur_io, *next_io;
3325         struct ctl_io_pool *pool;
3326         int retval;
3327
3328         retval = 0;
3329
3330         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3331                                             M_NOWAIT | M_ZERO);
3332         if (pool == NULL) {
3333                 retval = ENOMEM;
3334                 goto bailout;
3335         }
3336
3337         pool->type = pool_type;
3338         pool->ctl_softc = ctl_softc;
3339
3340         mtx_lock(&ctl_softc->pool_lock);
3341         pool->id = ctl_softc->cur_pool_id++;
3342         mtx_unlock(&ctl_softc->pool_lock);
3343
3344         pool->flags = CTL_POOL_FLAG_NONE;
3345         pool->refcount = 1;             /* Reference for validity. */
3346         STAILQ_INIT(&pool->free_queue);
3347
3348         /*
3349          * XXX KDM other options here:
3350          * - allocate a page at a time
3351          * - allocate one big chunk of memory.
3352          * Page allocation might work well, but would take a little more
3353          * tracking.
3354          */
3355         for (i = 0; i < total_ctl_io; i++) {
3356                 cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTLIO,
3357                                                 M_NOWAIT);
3358                 if (cur_io == NULL) {
3359                         retval = ENOMEM;
3360                         break;
3361                 }
3362                 cur_io->io_hdr.pool = pool;
3363                 STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3364                 pool->total_ctl_io++;
3365                 pool->free_ctl_io++;
3366         }
3367
3368         if (retval != 0) {
3369                 for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3370                      cur_io != NULL; cur_io = next_io) {
3371                         next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3372                                                               links);
3373                         STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3374                                       ctl_io_hdr, links);
3375                         free(cur_io, M_CTLIO);
3376                 }
3377
3378                 free(pool, M_CTL);
3379                 goto bailout;
3380         }
3381         mtx_lock(&ctl_softc->pool_lock);
3382         ctl_softc->num_pools++;
3383         STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3384         /*
3385          * Increment our usage count if this is an external consumer, so we
3386          * can't get unloaded until the external consumer (most likely a
3387          * FETD) unloads and frees his pool.
3388          *
3389          * XXX KDM will this increment the caller's module use count, or
3390          * mine?
3391          */
3392 #if 0
3393         if ((pool_type != CTL_POOL_EMERGENCY)
3394          && (pool_type != CTL_POOL_INTERNAL)
3395          && (pool_type != CTL_POOL_4OTHERSC))
3396                 MOD_INC_USE_COUNT;
3397 #endif
3398
3399         mtx_unlock(&ctl_softc->pool_lock);
3400
3401         *npool = pool;
3402
3403 bailout:
3404
3405         return (retval);
3406 }
3407
3408 static int
3409 ctl_pool_acquire(struct ctl_io_pool *pool)
3410 {
3411
3412         mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3413
3414         if (pool->flags & CTL_POOL_FLAG_INVALID)
3415                 return (EINVAL);
3416
3417         pool->refcount++;
3418
3419         return (0);
3420 }
3421
3422 static void
3423 ctl_pool_release(struct ctl_io_pool *pool)
3424 {
3425         struct ctl_softc *ctl_softc = pool->ctl_softc;
3426         union ctl_io *io;
3427
3428         mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3429
3430         if (--pool->refcount != 0)
3431                 return;
3432
3433         while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3434                 STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3435                               links);
3436                 free(io, M_CTLIO);
3437         }
3438
3439         STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3440         ctl_softc->num_pools--;
3441
3442         /*
3443          * XXX KDM will this decrement the caller's usage count or mine?
3444          */
3445 #if 0
3446         if ((pool->type != CTL_POOL_EMERGENCY)
3447          && (pool->type != CTL_POOL_INTERNAL)
3448          && (pool->type != CTL_POOL_4OTHERSC))
3449                 MOD_DEC_USE_COUNT;
3450 #endif
3451
3452         free(pool, M_CTL);
3453 }
3454
3455 void
3456 ctl_pool_free(struct ctl_io_pool *pool)
3457 {
3458         struct ctl_softc *ctl_softc;
3459
3460         if (pool == NULL)
3461                 return;
3462
3463         ctl_softc = pool->ctl_softc;
3464         mtx_lock(&ctl_softc->pool_lock);
3465         pool->flags |= CTL_POOL_FLAG_INVALID;
3466         ctl_pool_release(pool);
3467         mtx_unlock(&ctl_softc->pool_lock);
3468 }
3469
3470 /*
3471  * This routine does not block (except for spinlocks of course).
3472  * It tries to allocate a ctl_io union from the caller's pool as quickly as
3473  * possible.
3474  */
3475 union ctl_io *
3476 ctl_alloc_io(void *pool_ref)
3477 {
3478         union ctl_io *io;
3479         struct ctl_softc *ctl_softc;
3480         struct ctl_io_pool *pool, *npool;
3481         struct ctl_io_pool *emergency_pool;
3482
3483         pool = (struct ctl_io_pool *)pool_ref;
3484
3485         if (pool == NULL) {
3486                 printf("%s: pool is NULL\n", __func__);
3487                 return (NULL);
3488         }
3489
3490         emergency_pool = NULL;
3491
3492         ctl_softc = pool->ctl_softc;
3493
3494         mtx_lock(&ctl_softc->pool_lock);
3495         /*
3496          * First, try to get the io structure from the user's pool.
3497          */
3498         if (ctl_pool_acquire(pool) == 0) {
3499                 io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3500                 if (io != NULL) {
3501                         STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3502                         pool->total_allocated++;
3503                         pool->free_ctl_io--;
3504                         mtx_unlock(&ctl_softc->pool_lock);
3505                         return (io);
3506                 } else
3507                         ctl_pool_release(pool);
3508         }
3509         /*
3510          * If he doesn't have any io structures left, search for an
3511          * emergency pool and grab one from there.
3512          */
3513         STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3514                 if (npool->type != CTL_POOL_EMERGENCY)
3515                         continue;
3516
3517                 if (ctl_pool_acquire(npool) != 0)
3518                         continue;
3519
3520                 emergency_pool = npool;
3521
3522                 io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3523                 if (io != NULL) {
3524                         STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3525                         npool->total_allocated++;
3526                         npool->free_ctl_io--;
3527                         mtx_unlock(&ctl_softc->pool_lock);
3528                         return (io);
3529                 } else
3530                         ctl_pool_release(npool);
3531         }
3532
3533         /* Drop the spinlock before we malloc */
3534         mtx_unlock(&ctl_softc->pool_lock);
3535
3536         /*
3537          * The emergency pool (if it exists) didn't have one, so try an
3538          * atomic (i.e. nonblocking) malloc and see if we get lucky.
3539          */
3540         io = (union ctl_io *)malloc(sizeof(*io), M_CTLIO, M_NOWAIT);
3541         if (io != NULL) {
3542                 /*
3543                  * If the emergency pool exists but is empty, add this
3544                  * ctl_io to its list when it gets freed.
3545                  */
3546                 if (emergency_pool != NULL) {
3547                         mtx_lock(&ctl_softc->pool_lock);
3548                         if (ctl_pool_acquire(emergency_pool) == 0) {
3549                                 io->io_hdr.pool = emergency_pool;
3550                                 emergency_pool->total_ctl_io++;
3551                                 /*
3552                                  * Need to bump this, otherwise
3553                                  * total_allocated and total_freed won't
3554                                  * match when we no longer have anything
3555                                  * outstanding.
3556                                  */
3557                                 emergency_pool->total_allocated++;
3558                         }
3559                         mtx_unlock(&ctl_softc->pool_lock);
3560                 } else
3561                         io->io_hdr.pool = NULL;
3562         }
3563
3564         return (io);
3565 }
3566
3567 void
3568 ctl_free_io(union ctl_io *io)
3569 {
3570         if (io == NULL)
3571                 return;
3572
3573         /*
3574          * If this ctl_io has a pool, return it to that pool.
3575          */
3576         if (io->io_hdr.pool != NULL) {
3577                 struct ctl_io_pool *pool;
3578
3579                 pool = (struct ctl_io_pool *)io->io_hdr.pool;
3580                 mtx_lock(&pool->ctl_softc->pool_lock);
3581                 io->io_hdr.io_type = 0xff;
3582                 STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3583                 pool->total_freed++;
3584                 pool->free_ctl_io++;
3585                 ctl_pool_release(pool);
3586                 mtx_unlock(&pool->ctl_softc->pool_lock);
3587         } else {
3588                 /*
3589                  * Otherwise, just free it.  We probably malloced it and
3590                  * the emergency pool wasn't available.
3591                  */
3592                 free(io, M_CTLIO);
3593         }
3594
3595 }
3596
3597 void
3598 ctl_zero_io(union ctl_io *io)
3599 {
3600         void *pool_ref;
3601
3602         if (io == NULL)
3603                 return;
3604
3605         /*
3606          * May need to preserve linked list pointers at some point too.
3607          */
3608         pool_ref = io->io_hdr.pool;
3609
3610         memset(io, 0, sizeof(*io));
3611
3612         io->io_hdr.pool = pool_ref;
3613 }
3614
3615 /*
3616  * This routine is currently used for internal copies of ctl_ios that need
3617  * to persist for some reason after we've already returned status to the
3618  * FETD.  (Thus the flag set.)
3619  *
3620  * XXX XXX
3621  * Note that this makes a blind copy of all fields in the ctl_io, except
3622  * for the pool reference.  This includes any memory that has been
3623  * allocated!  That memory will no longer be valid after done has been
3624  * called, so this would be VERY DANGEROUS for command that actually does
3625  * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3626  * start and stop commands, which don't transfer any data, so this is not a
3627  * problem.  If it is used for anything else, the caller would also need to
3628  * allocate data buffer space and this routine would need to be modified to
3629  * copy the data buffer(s) as well.
3630  */
3631 void
3632 ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3633 {
3634         void *pool_ref;
3635
3636         if ((src == NULL)
3637          || (dest == NULL))
3638                 return;
3639
3640         /*
3641          * May need to preserve linked list pointers at some point too.
3642          */
3643         pool_ref = dest->io_hdr.pool;
3644
3645         memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3646
3647         dest->io_hdr.pool = pool_ref;
3648         /*
3649          * We need to know that this is an internal copy, and doesn't need
3650          * to get passed back to the FETD that allocated it.
3651          */
3652         dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3653 }
3654
3655 #ifdef NEEDTOPORT
3656 static void
3657 ctl_update_power_subpage(struct copan_power_subpage *page)
3658 {
3659         int num_luns, num_partitions, config_type;
3660         struct ctl_softc *softc;
3661         cs_BOOL_t aor_present, shelf_50pct_power;
3662         cs_raidset_personality_t rs_type;
3663         int max_active_luns;
3664
3665         softc = control_softc;
3666
3667         /* subtract out the processor LUN */
3668         num_luns = softc->num_luns - 1;
3669         /*
3670          * Default to 7 LUNs active, which was the only number we allowed
3671          * in the past.
3672          */
3673         max_active_luns = 7;
3674
3675         num_partitions = config_GetRsPartitionInfo();
3676         config_type = config_GetConfigType();
3677         shelf_50pct_power = config_GetShelfPowerMode();
3678         aor_present = config_IsAorRsPresent();
3679
3680         rs_type = ddb_GetRsRaidType(1);
3681         if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3682          && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3683                 EPRINT(0, "Unsupported RS type %d!", rs_type);
3684         }
3685
3686
3687         page->total_luns = num_luns;
3688
3689         switch (config_type) {
3690         case 40:
3691                 /*
3692                  * In a 40 drive configuration, it doesn't matter what DC
3693                  * cards we have, whether we have AOR enabled or not,
3694                  * partitioning or not, or what type of RAIDset we have.
3695                  * In that scenario, we can power up every LUN we present
3696                  * to the user.
3697                  */
3698                 max_active_luns = num_luns;
3699
3700                 break;
3701         case 64:
3702                 if (shelf_50pct_power == CS_FALSE) {
3703                         /* 25% power */
3704                         if (aor_present == CS_TRUE) {
3705                                 if (rs_type ==
3706                                      CS_RAIDSET_PERSONALITY_RAID5) {
3707                                         max_active_luns = 7;
3708                                 } else if (rs_type ==
3709                                          CS_RAIDSET_PERSONALITY_RAID1){
3710                                         max_active_luns = 14;
3711                                 } else {
3712                                         /* XXX KDM now what?? */
3713                                 }
3714                         } else {
3715                                 if (rs_type ==
3716                                      CS_RAIDSET_PERSONALITY_RAID5) {
3717                                         max_active_luns = 8;
3718                                 } else if (rs_type ==
3719                                          CS_RAIDSET_PERSONALITY_RAID1){
3720                                         max_active_luns = 16;
3721                                 } else {
3722                                         /* XXX KDM now what?? */
3723                                 }
3724                         }
3725                 } else {
3726                         /* 50% power */
3727                         /*
3728                          * With 50% power in a 64 drive configuration, we
3729                          * can power all LUNs we present.
3730                          */
3731                         max_active_luns = num_luns;
3732                 }
3733                 break;
3734         case 112:
3735                 if (shelf_50pct_power == CS_FALSE) {
3736                         /* 25% power */
3737                         if (aor_present == CS_TRUE) {
3738                                 if (rs_type ==
3739                                      CS_RAIDSET_PERSONALITY_RAID5) {
3740                                         max_active_luns = 7;
3741                                 } else if (rs_type ==
3742                                          CS_RAIDSET_PERSONALITY_RAID1){
3743                                         max_active_luns = 14;
3744                                 } else {
3745                                         /* XXX KDM now what?? */
3746                                 }
3747                         } else {
3748                                 if (rs_type ==
3749                                      CS_RAIDSET_PERSONALITY_RAID5) {
3750                                         max_active_luns = 8;
3751                                 } else if (rs_type ==
3752                                          CS_RAIDSET_PERSONALITY_RAID1){
3753                                         max_active_luns = 16;
3754                                 } else {
3755                                         /* XXX KDM now what?? */
3756                                 }
3757                         }
3758                 } else {
3759                         /* 50% power */
3760                         if (aor_present == CS_TRUE) {
3761                                 if (rs_type ==
3762                                      CS_RAIDSET_PERSONALITY_RAID5) {
3763                                         max_active_luns = 14;
3764                                 } else if (rs_type ==
3765                                          CS_RAIDSET_PERSONALITY_RAID1){
3766                                         /*
3767                                          * We're assuming here that disk
3768                                          * caching is enabled, and so we're
3769                                          * able to power up half of each
3770                                          * LUN, and cache all writes.
3771                                          */
3772                                         max_active_luns = num_luns;
3773                                 } else {
3774                                         /* XXX KDM now what?? */
3775                                 }
3776                         } else {
3777                                 if (rs_type ==
3778                                      CS_RAIDSET_PERSONALITY_RAID5) {
3779                                         max_active_luns = 15;
3780                                 } else if (rs_type ==
3781                                          CS_RAIDSET_PERSONALITY_RAID1){
3782                                         max_active_luns = 30;
3783                                 } else {
3784                                         /* XXX KDM now what?? */
3785                                 }
3786                         }
3787                 }
3788                 break;
3789         default:
3790                 /*
3791                  * In this case, we have an unknown configuration, so we
3792                  * just use the default from above.
3793                  */
3794                 break;
3795         }
3796
3797         page->max_active_luns = max_active_luns;
3798 #if 0
3799         printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
3800                page->total_luns, page->max_active_luns);
3801 #endif
3802 }
3803 #endif /* NEEDTOPORT */
3804
3805 /*
3806  * This routine could be used in the future to load default and/or saved
3807  * mode page parameters for a particuar lun.
3808  */
3809 static int
3810 ctl_init_page_index(struct ctl_lun *lun)
3811 {
3812         int i;
3813         struct ctl_page_index *page_index;
3814         struct ctl_softc *softc;
3815
3816         memcpy(&lun->mode_pages.index, page_index_template,
3817                sizeof(page_index_template));
3818
3819         softc = lun->ctl_softc;
3820
3821         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3822
3823                 page_index = &lun->mode_pages.index[i];
3824                 /*
3825                  * If this is a disk-only mode page, there's no point in
3826                  * setting it up.  For some pages, we have to have some
3827                  * basic information about the disk in order to calculate the
3828                  * mode page data.
3829                  */
3830                 if ((lun->be_lun->lun_type != T_DIRECT)
3831                  && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3832                         continue;
3833
3834                 switch (page_index->page_code & SMPH_PC_MASK) {
3835                 case SMS_FORMAT_DEVICE_PAGE: {
3836                         struct scsi_format_page *format_page;
3837
3838                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3839                                 panic("subpage is incorrect!");
3840
3841                         /*
3842                          * Sectors per track are set above.  Bytes per
3843                          * sector need to be set here on a per-LUN basis.
3844                          */
3845                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3846                                &format_page_default,
3847                                sizeof(format_page_default));
3848                         memcpy(&lun->mode_pages.format_page[
3849                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
3850                                sizeof(format_page_changeable));
3851                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3852                                &format_page_default,
3853                                sizeof(format_page_default));
3854                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3855                                &format_page_default,
3856                                sizeof(format_page_default));
3857
3858                         format_page = &lun->mode_pages.format_page[
3859                                 CTL_PAGE_CURRENT];
3860                         scsi_ulto2b(lun->be_lun->blocksize,
3861                                     format_page->bytes_per_sector);
3862
3863                         format_page = &lun->mode_pages.format_page[
3864                                 CTL_PAGE_DEFAULT];
3865                         scsi_ulto2b(lun->be_lun->blocksize,
3866                                     format_page->bytes_per_sector);
3867
3868                         format_page = &lun->mode_pages.format_page[
3869                                 CTL_PAGE_SAVED];
3870                         scsi_ulto2b(lun->be_lun->blocksize,
3871                                     format_page->bytes_per_sector);
3872
3873                         page_index->page_data =
3874                                 (uint8_t *)lun->mode_pages.format_page;
3875                         break;
3876                 }
3877                 case SMS_RIGID_DISK_PAGE: {
3878                         struct scsi_rigid_disk_page *rigid_disk_page;
3879                         uint32_t sectors_per_cylinder;
3880                         uint64_t cylinders;
3881 #ifndef __XSCALE__
3882                         int shift;
3883 #endif /* !__XSCALE__ */
3884
3885                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3886                                 panic("invalid subpage value %d",
3887                                       page_index->subpage);
3888
3889                         /*
3890                          * Rotation rate and sectors per track are set
3891                          * above.  We calculate the cylinders here based on
3892                          * capacity.  Due to the number of heads and
3893                          * sectors per track we're using, smaller arrays
3894                          * may turn out to have 0 cylinders.  Linux and
3895                          * FreeBSD don't pay attention to these mode pages
3896                          * to figure out capacity, but Solaris does.  It
3897                          * seems to deal with 0 cylinders just fine, and
3898                          * works out a fake geometry based on the capacity.
3899                          */
3900                         memcpy(&lun->mode_pages.rigid_disk_page[
3901                                CTL_PAGE_CURRENT], &rigid_disk_page_default,
3902                                sizeof(rigid_disk_page_default));
3903                         memcpy(&lun->mode_pages.rigid_disk_page[
3904                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
3905                                sizeof(rigid_disk_page_changeable));
3906                         memcpy(&lun->mode_pages.rigid_disk_page[
3907                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
3908                                sizeof(rigid_disk_page_default));
3909                         memcpy(&lun->mode_pages.rigid_disk_page[
3910                                CTL_PAGE_SAVED], &rigid_disk_page_default,
3911                                sizeof(rigid_disk_page_default));
3912
3913                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
3914                                 CTL_DEFAULT_HEADS;
3915
3916                         /*
3917                          * The divide method here will be more accurate,
3918                          * probably, but results in floating point being
3919                          * used in the kernel on i386 (__udivdi3()).  On the
3920                          * XScale, though, __udivdi3() is implemented in
3921                          * software.
3922                          *
3923                          * The shift method for cylinder calculation is
3924                          * accurate if sectors_per_cylinder is a power of
3925                          * 2.  Otherwise it might be slightly off -- you
3926                          * might have a bit of a truncation problem.
3927                          */
3928 #ifdef  __XSCALE__
3929                         cylinders = (lun->be_lun->maxlba + 1) /
3930                                 sectors_per_cylinder;
3931 #else
3932                         for (shift = 31; shift > 0; shift--) {
3933                                 if (sectors_per_cylinder & (1 << shift))
3934                                         break;
3935                         }
3936                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
3937 #endif
3938
3939                         /*
3940                          * We've basically got 3 bytes, or 24 bits for the
3941                          * cylinder size in the mode page.  If we're over,
3942                          * just round down to 2^24.
3943                          */
3944                         if (cylinders > 0xffffff)
3945                                 cylinders = 0xffffff;
3946
3947                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3948                                 CTL_PAGE_CURRENT];
3949                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3950
3951                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3952                                 CTL_PAGE_DEFAULT];
3953                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3954
3955                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3956                                 CTL_PAGE_SAVED];
3957                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3958
3959                         page_index->page_data =
3960                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
3961                         break;
3962                 }
3963                 case SMS_CACHING_PAGE: {
3964
3965                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3966                                 panic("invalid subpage value %d",
3967                                       page_index->subpage);
3968                         /*
3969                          * Defaults should be okay here, no calculations
3970                          * needed.
3971                          */
3972                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
3973                                &caching_page_default,
3974                                sizeof(caching_page_default));
3975                         memcpy(&lun->mode_pages.caching_page[
3976                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
3977                                sizeof(caching_page_changeable));
3978                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
3979                                &caching_page_default,
3980                                sizeof(caching_page_default));
3981                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
3982                                &caching_page_default,
3983                                sizeof(caching_page_default));
3984                         page_index->page_data =
3985                                 (uint8_t *)lun->mode_pages.caching_page;
3986                         break;
3987                 }
3988                 case SMS_CONTROL_MODE_PAGE: {
3989
3990                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3991                                 panic("invalid subpage value %d",
3992                                       page_index->subpage);
3993
3994                         /*
3995                          * Defaults should be okay here, no calculations
3996                          * needed.
3997                          */
3998                         memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
3999                                &control_page_default,
4000                                sizeof(control_page_default));
4001                         memcpy(&lun->mode_pages.control_page[
4002                                CTL_PAGE_CHANGEABLE], &control_page_changeable,
4003                                sizeof(control_page_changeable));
4004                         memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4005                                &control_page_default,
4006                                sizeof(control_page_default));
4007                         memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4008                                &control_page_default,
4009                                sizeof(control_page_default));
4010                         page_index->page_data =
4011                                 (uint8_t *)lun->mode_pages.control_page;
4012                         break;
4013
4014                 }
4015                 case SMS_VENDOR_SPECIFIC_PAGE:{
4016                         switch (page_index->subpage) {
4017                         case PWR_SUBPAGE_CODE: {
4018                                 struct copan_power_subpage *current_page,
4019                                                            *saved_page;
4020
4021                                 memcpy(&lun->mode_pages.power_subpage[
4022                                        CTL_PAGE_CURRENT],
4023                                        &power_page_default,
4024                                        sizeof(power_page_default));
4025                                 memcpy(&lun->mode_pages.power_subpage[
4026                                        CTL_PAGE_CHANGEABLE],
4027                                        &power_page_changeable,
4028                                        sizeof(power_page_changeable));
4029                                 memcpy(&lun->mode_pages.power_subpage[
4030                                        CTL_PAGE_DEFAULT],
4031                                        &power_page_default,
4032                                        sizeof(power_page_default));
4033                                 memcpy(&lun->mode_pages.power_subpage[
4034                                        CTL_PAGE_SAVED],
4035                                        &power_page_default,
4036                                        sizeof(power_page_default));
4037                                 page_index->page_data =
4038                                     (uint8_t *)lun->mode_pages.power_subpage;
4039
4040                                 current_page = (struct copan_power_subpage *)
4041                                         (page_index->page_data +
4042                                          (page_index->page_len *
4043                                           CTL_PAGE_CURRENT));
4044                                 saved_page = (struct copan_power_subpage *)
4045                                         (page_index->page_data +
4046                                          (page_index->page_len *
4047                                           CTL_PAGE_SAVED));
4048                                 break;
4049                         }
4050                         case APS_SUBPAGE_CODE: {
4051                                 struct copan_aps_subpage *current_page,
4052                                                          *saved_page;
4053
4054                                 // This gets set multiple times but
4055                                 // it should always be the same. It's
4056                                 // only done during init so who cares.
4057                                 index_to_aps_page = i;
4058
4059                                 memcpy(&lun->mode_pages.aps_subpage[
4060                                        CTL_PAGE_CURRENT],
4061                                        &aps_page_default,
4062                                        sizeof(aps_page_default));
4063                                 memcpy(&lun->mode_pages.aps_subpage[
4064                                        CTL_PAGE_CHANGEABLE],
4065                                        &aps_page_changeable,
4066                                        sizeof(aps_page_changeable));
4067                                 memcpy(&lun->mode_pages.aps_subpage[
4068                                        CTL_PAGE_DEFAULT],
4069                                        &aps_page_default,
4070                                        sizeof(aps_page_default));
4071                                 memcpy(&lun->mode_pages.aps_subpage[
4072                                        CTL_PAGE_SAVED],
4073                                        &aps_page_default,
4074                                        sizeof(aps_page_default));
4075                                 page_index->page_data =
4076                                         (uint8_t *)lun->mode_pages.aps_subpage;
4077
4078                                 current_page = (struct copan_aps_subpage *)
4079                                         (page_index->page_data +
4080                                          (page_index->page_len *
4081                                           CTL_PAGE_CURRENT));
4082                                 saved_page = (struct copan_aps_subpage *)
4083                                         (page_index->page_data +
4084                                          (page_index->page_len *
4085                                           CTL_PAGE_SAVED));
4086                                 break;
4087                         }
4088                         case DBGCNF_SUBPAGE_CODE: {
4089                                 struct copan_debugconf_subpage *current_page,
4090                                                                *saved_page;
4091
4092                                 memcpy(&lun->mode_pages.debugconf_subpage[
4093                                        CTL_PAGE_CURRENT],
4094                                        &debugconf_page_default,
4095                                        sizeof(debugconf_page_default));
4096                                 memcpy(&lun->mode_pages.debugconf_subpage[
4097                                        CTL_PAGE_CHANGEABLE],
4098                                        &debugconf_page_changeable,
4099                                        sizeof(debugconf_page_changeable));
4100                                 memcpy(&lun->mode_pages.debugconf_subpage[
4101                                        CTL_PAGE_DEFAULT],
4102                                        &debugconf_page_default,
4103                                        sizeof(debugconf_page_default));
4104                                 memcpy(&lun->mode_pages.debugconf_subpage[
4105                                        CTL_PAGE_SAVED],
4106                                        &debugconf_page_default,
4107                                        sizeof(debugconf_page_default));
4108                                 page_index->page_data =
4109                                         (uint8_t *)lun->mode_pages.debugconf_subpage;
4110
4111                                 current_page = (struct copan_debugconf_subpage *)
4112                                         (page_index->page_data +
4113                                          (page_index->page_len *
4114                                           CTL_PAGE_CURRENT));
4115                                 saved_page = (struct copan_debugconf_subpage *)
4116                                         (page_index->page_data +
4117                                          (page_index->page_len *
4118                                           CTL_PAGE_SAVED));
4119                                 break;
4120                         }
4121                         default:
4122                                 panic("invalid subpage value %d",
4123                                       page_index->subpage);
4124                                 break;
4125                         }
4126                         break;
4127                 }
4128                 default:
4129                         panic("invalid page value %d",
4130                               page_index->page_code & SMPH_PC_MASK);
4131                         break;
4132         }
4133         }
4134
4135         return (CTL_RETVAL_COMPLETE);
4136 }
4137
4138 /*
4139  * LUN allocation.
4140  *
4141  * Requirements:
4142  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4143  *   wants us to allocate the LUN and he can block.
4144  * - ctl_softc is always set
4145  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4146  *
4147  * Returns 0 for success, non-zero (errno) for failure.
4148  */
4149 static int
4150 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4151               struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4152 {
4153         struct ctl_lun *nlun, *lun;
4154         struct ctl_port *port;
4155         int lun_number, i, lun_malloced;
4156
4157         if (be_lun == NULL)
4158                 return (EINVAL);
4159
4160         /*
4161          * We currently only support Direct Access or Processor LUN types.
4162          */
4163         switch (be_lun->lun_type) {
4164         case T_DIRECT:
4165                 break;
4166         case T_PROCESSOR:
4167                 break;
4168         case T_SEQUENTIAL:
4169         case T_CHANGER:
4170         default:
4171                 be_lun->lun_config_status(be_lun->be_lun,
4172                                           CTL_LUN_CONFIG_FAILURE);
4173                 break;
4174         }
4175         if (ctl_lun == NULL) {
4176                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4177                 lun_malloced = 1;
4178         } else {
4179                 lun_malloced = 0;
4180                 lun = ctl_lun;
4181         }
4182
4183         memset(lun, 0, sizeof(*lun));
4184         if (lun_malloced)
4185                 lun->flags = CTL_LUN_MALLOCED;
4186
4187         mtx_lock(&ctl_softc->ctl_lock);
4188         /*
4189          * See if the caller requested a particular LUN number.  If so, see
4190          * if it is available.  Otherwise, allocate the first available LUN.
4191          */
4192         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4193                 if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4194                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4195                         mtx_unlock(&ctl_softc->ctl_lock);
4196                         if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4197                                 printf("ctl: requested LUN ID %d is higher "
4198                                        "than CTL_MAX_LUNS - 1 (%d)\n",
4199                                        be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4200                         } else {
4201                                 /*
4202                                  * XXX KDM return an error, or just assign
4203                                  * another LUN ID in this case??
4204                                  */
4205                                 printf("ctl: requested LUN ID %d is already "
4206                                        "in use\n", be_lun->req_lun_id);
4207                         }
4208                         if (lun->flags & CTL_LUN_MALLOCED)
4209                                 free(lun, M_CTL);
4210                         be_lun->lun_config_status(be_lun->be_lun,
4211                                                   CTL_LUN_CONFIG_FAILURE);
4212                         return (ENOSPC);
4213                 }
4214                 lun_number = be_lun->req_lun_id;
4215         } else {
4216                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4217                 if (lun_number == -1) {
4218                         mtx_unlock(&ctl_softc->ctl_lock);
4219                         printf("ctl: can't allocate LUN on target %ju, out of "
4220                                "LUNs\n", (uintmax_t)target_id.id);
4221                         if (lun->flags & CTL_LUN_MALLOCED)
4222                                 free(lun, M_CTL);
4223                         be_lun->lun_config_status(be_lun->be_lun,
4224                                                   CTL_LUN_CONFIG_FAILURE);
4225                         return (ENOSPC);
4226                 }
4227         }
4228         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4229
4230         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4231         lun->target = target_id;
4232         lun->lun = lun_number;
4233         lun->be_lun = be_lun;
4234         /*
4235          * The processor LUN is always enabled.  Disk LUNs come on line
4236          * disabled, and must be enabled by the backend.
4237          */
4238         lun->flags |= CTL_LUN_DISABLED;
4239         lun->backend = be_lun->be;
4240         be_lun->ctl_lun = lun;
4241         be_lun->lun_id = lun_number;
4242         atomic_add_int(&be_lun->be->num_luns, 1);
4243         if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4244                 lun->flags |= CTL_LUN_STOPPED;
4245
4246         if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4247                 lun->flags |= CTL_LUN_INOPERABLE;
4248
4249         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4250                 lun->flags |= CTL_LUN_PRIMARY_SC;
4251
4252         lun->ctl_softc = ctl_softc;
4253         TAILQ_INIT(&lun->ooa_queue);
4254         TAILQ_INIT(&lun->blocked_queue);
4255         STAILQ_INIT(&lun->error_list);
4256
4257         /*
4258          * Initialize the mode page index.
4259          */
4260         ctl_init_page_index(lun);
4261
4262         /*
4263          * Set the poweron UA for all initiators on this LUN only.
4264          */
4265         for (i = 0; i < CTL_MAX_INITIATORS; i++)
4266                 lun->pending_sense[i].ua_pending = CTL_UA_POWERON;
4267
4268         /*
4269          * Now, before we insert this lun on the lun list, set the lun
4270          * inventory changed UA for all other luns.
4271          */
4272         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4273                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4274                         nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4275                 }
4276         }
4277
4278         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4279
4280         ctl_softc->ctl_luns[lun_number] = lun;
4281
4282         ctl_softc->num_luns++;
4283
4284         /* Setup statistics gathering */
4285         lun->stats.device_type = be_lun->lun_type;
4286         lun->stats.lun_number = lun_number;
4287         if (lun->stats.device_type == T_DIRECT)
4288                 lun->stats.blocksize = be_lun->blocksize;
4289         else
4290                 lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4291         for (i = 0;i < CTL_MAX_PORTS;i++)
4292                 lun->stats.ports[i].targ_port = i;
4293
4294         mtx_unlock(&ctl_softc->ctl_lock);
4295
4296         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4297
4298         /*
4299          * Run through each registered FETD and bring it online if it isn't
4300          * already.  Enable the target ID if it hasn't been enabled, and
4301          * enable this particular LUN.
4302          */
4303         STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4304                 int retval;
4305
4306                 retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4307                 if (retval != 0) {
4308                         printf("ctl_alloc_lun: FETD %s port %d returned error "
4309                                "%d for lun_enable on target %ju lun %d\n",
4310                                port->port_name, port->targ_port, retval,
4311                                (uintmax_t)target_id.id, lun_number);
4312                 } else
4313                         port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4314         }
4315         return (0);
4316 }
4317
4318 /*
4319  * Delete a LUN.
4320  * Assumptions:
4321  * - LUN has already been marked invalid and any pending I/O has been taken
4322  *   care of.
4323  */
4324 static int
4325 ctl_free_lun(struct ctl_lun *lun)
4326 {
4327         struct ctl_softc *softc;
4328 #if 0
4329         struct ctl_port *port;
4330 #endif
4331         struct ctl_lun *nlun;
4332         int i;
4333
4334         softc = lun->ctl_softc;
4335
4336         mtx_assert(&softc->ctl_lock, MA_OWNED);
4337
4338         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4339
4340         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4341
4342         softc->ctl_luns[lun->lun] = NULL;
4343
4344         if (!TAILQ_EMPTY(&lun->ooa_queue))
4345                 panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4346
4347         softc->num_luns--;
4348
4349         /*
4350          * XXX KDM this scheme only works for a single target/multiple LUN
4351          * setup.  It needs to be revamped for a multiple target scheme.
4352          *
4353          * XXX KDM this results in port->lun_disable() getting called twice,
4354          * once when ctl_disable_lun() is called, and a second time here.
4355          * We really need to re-think the LUN disable semantics.  There
4356          * should probably be several steps/levels to LUN removal:
4357          *  - disable
4358          *  - invalidate
4359          *  - free
4360          *
4361          * Right now we only have a disable method when communicating to
4362          * the front end ports, at least for individual LUNs.
4363          */
4364 #if 0
4365         STAILQ_FOREACH(port, &softc->port_list, links) {
4366                 int retval;
4367
4368                 retval = port->lun_disable(port->targ_lun_arg, lun->target,
4369                                          lun->lun);
4370                 if (retval != 0) {
4371                         printf("ctl_free_lun: FETD %s port %d returned error "
4372                                "%d for lun_disable on target %ju lun %jd\n",
4373                                port->port_name, port->targ_port, retval,
4374                                (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4375                 }
4376
4377                 if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4378                         port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4379
4380                         retval = port->targ_disable(port->targ_lun_arg,lun->target);
4381                         if (retval != 0) {
4382                                 printf("ctl_free_lun: FETD %s port %d "
4383                                        "returned error %d for targ_disable on "
4384                                        "target %ju\n", port->port_name,
4385                                        port->targ_port, retval,
4386                                        (uintmax_t)lun->target.id);
4387                         } else
4388                                 port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4389
4390                         if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4391                                 continue;
4392
4393 #if 0
4394                         port->port_offline(port->onoff_arg);
4395                         port->status &= ~CTL_PORT_STATUS_ONLINE;
4396 #endif
4397                 }
4398         }
4399 #endif
4400
4401         /*
4402          * Tell the backend to free resources, if this LUN has a backend.
4403          */
4404         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4405         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4406
4407         mtx_destroy(&lun->lun_lock);
4408         if (lun->flags & CTL_LUN_MALLOCED)
4409                 free(lun, M_CTL);
4410
4411         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4412                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4413                         nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4414                 }
4415         }
4416
4417         return (0);
4418 }
4419
4420 static void
4421 ctl_create_lun(struct ctl_be_lun *be_lun)
4422 {
4423         struct ctl_softc *ctl_softc;
4424
4425         ctl_softc = control_softc;
4426
4427         /*
4428          * ctl_alloc_lun() should handle all potential failure cases.
4429          */
4430         ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4431 }
4432
4433 int
4434 ctl_add_lun(struct ctl_be_lun *be_lun)
4435 {
4436         struct ctl_softc *ctl_softc = control_softc;
4437
4438         mtx_lock(&ctl_softc->ctl_lock);
4439         STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4440         mtx_unlock(&ctl_softc->ctl_lock);
4441         wakeup(&ctl_softc->pending_lun_queue);
4442
4443         return (0);
4444 }
4445
4446 int
4447 ctl_enable_lun(struct ctl_be_lun *be_lun)
4448 {
4449         struct ctl_softc *ctl_softc;
4450         struct ctl_port *port, *nport;
4451         struct ctl_lun *lun;
4452         int retval;
4453
4454         ctl_softc = control_softc;
4455
4456         lun = (struct ctl_lun *)be_lun->ctl_lun;
4457
4458         mtx_lock(&ctl_softc->ctl_lock);
4459         mtx_lock(&lun->lun_lock);
4460         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4461                 /*
4462                  * eh?  Why did we get called if the LUN is already
4463                  * enabled?
4464                  */
4465                 mtx_unlock(&lun->lun_lock);
4466                 mtx_unlock(&ctl_softc->ctl_lock);
4467                 return (0);
4468         }
4469         lun->flags &= ~CTL_LUN_DISABLED;
4470         mtx_unlock(&lun->lun_lock);
4471
4472         for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4473                 nport = STAILQ_NEXT(port, links);
4474
4475                 /*
4476                  * Drop the lock while we call the FETD's enable routine.
4477                  * This can lead to a callback into CTL (at least in the
4478                  * case of the internal initiator frontend.
4479                  */
4480                 mtx_unlock(&ctl_softc->ctl_lock);
4481                 retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4482                 mtx_lock(&ctl_softc->ctl_lock);
4483                 if (retval != 0) {
4484                         printf("%s: FETD %s port %d returned error "
4485                                "%d for lun_enable on target %ju lun %jd\n",
4486                                __func__, port->port_name, port->targ_port, retval,
4487                                (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4488                 }
4489 #if 0
4490                  else {
4491             /* NOTE:  TODO:  why does lun enable affect port status? */
4492                         port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4493                 }
4494 #endif
4495         }
4496
4497         mtx_unlock(&ctl_softc->ctl_lock);
4498
4499         return (0);
4500 }
4501
4502 int
4503 ctl_disable_lun(struct ctl_be_lun *be_lun)
4504 {
4505         struct ctl_softc *ctl_softc;
4506         struct ctl_port *port;
4507         struct ctl_lun *lun;
4508         int retval;
4509
4510         ctl_softc = control_softc;
4511
4512         lun = (struct ctl_lun *)be_lun->ctl_lun;
4513
4514         mtx_lock(&ctl_softc->ctl_lock);
4515         mtx_lock(&lun->lun_lock);
4516         if (lun->flags & CTL_LUN_DISABLED) {
4517                 mtx_unlock(&lun->lun_lock);
4518                 mtx_unlock(&ctl_softc->ctl_lock);
4519                 return (0);
4520         }
4521         lun->flags |= CTL_LUN_DISABLED;
4522         mtx_unlock(&lun->lun_lock);
4523
4524         STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4525                 mtx_unlock(&ctl_softc->ctl_lock);
4526                 /*
4527                  * Drop the lock before we call the frontend's disable
4528                  * routine, to avoid lock order reversals.
4529                  *
4530                  * XXX KDM what happens if the frontend list changes while
4531                  * we're traversing it?  It's unlikely, but should be handled.
4532                  */
4533                 retval = port->lun_disable(port->targ_lun_arg, lun->target,
4534                                          lun->lun);
4535                 mtx_lock(&ctl_softc->ctl_lock);
4536                 if (retval != 0) {
4537                         printf("ctl_alloc_lun: FETD %s port %d returned error "
4538                                "%d for lun_disable on target %ju lun %jd\n",
4539                                port->port_name, port->targ_port, retval,
4540                                (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4541                 }
4542         }
4543
4544         mtx_unlock(&ctl_softc->ctl_lock);
4545
4546         return (0);
4547 }
4548
4549 int
4550 ctl_start_lun(struct ctl_be_lun *be_lun)
4551 {
4552         struct ctl_softc *ctl_softc;
4553         struct ctl_lun *lun;
4554
4555         ctl_softc = control_softc;
4556
4557         lun = (struct ctl_lun *)be_lun->ctl_lun;
4558
4559         mtx_lock(&lun->lun_lock);
4560         lun->flags &= ~CTL_LUN_STOPPED;
4561         mtx_unlock(&lun->lun_lock);
4562
4563         return (0);
4564 }
4565
4566 int
4567 ctl_stop_lun(struct ctl_be_lun *be_lun)
4568 {
4569         struct ctl_softc *ctl_softc;
4570         struct ctl_lun *lun;
4571
4572         ctl_softc = control_softc;
4573
4574         lun = (struct ctl_lun *)be_lun->ctl_lun;
4575
4576         mtx_lock(&lun->lun_lock);
4577         lun->flags |= CTL_LUN_STOPPED;
4578         mtx_unlock(&lun->lun_lock);
4579
4580         return (0);
4581 }
4582
4583 int
4584 ctl_lun_offline(struct ctl_be_lun *be_lun)
4585 {
4586         struct ctl_softc *ctl_softc;
4587         struct ctl_lun *lun;
4588
4589         ctl_softc = control_softc;
4590
4591         lun = (struct ctl_lun *)be_lun->ctl_lun;
4592
4593         mtx_lock(&lun->lun_lock);
4594         lun->flags |= CTL_LUN_OFFLINE;
4595         mtx_unlock(&lun->lun_lock);
4596
4597         return (0);
4598 }
4599
4600 int
4601 ctl_lun_online(struct ctl_be_lun *be_lun)
4602 {
4603         struct ctl_softc *ctl_softc;
4604         struct ctl_lun *lun;
4605
4606         ctl_softc = control_softc;
4607
4608         lun = (struct ctl_lun *)be_lun->ctl_lun;
4609
4610         mtx_lock(&lun->lun_lock);
4611         lun->flags &= ~CTL_LUN_OFFLINE;
4612         mtx_unlock(&lun->lun_lock);
4613
4614         return (0);
4615 }
4616
4617 int
4618 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4619 {
4620         struct ctl_softc *ctl_softc;
4621         struct ctl_lun *lun;
4622
4623         ctl_softc = control_softc;
4624
4625         lun = (struct ctl_lun *)be_lun->ctl_lun;
4626
4627         mtx_lock(&lun->lun_lock);
4628
4629         /*
4630          * The LUN needs to be disabled before it can be marked invalid.
4631          */
4632         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4633                 mtx_unlock(&lun->lun_lock);
4634                 return (-1);
4635         }
4636         /*
4637          * Mark the LUN invalid.
4638          */
4639         lun->flags |= CTL_LUN_INVALID;
4640
4641         /*
4642          * If there is nothing in the OOA queue, go ahead and free the LUN.
4643          * If we have something in the OOA queue, we'll free it when the
4644          * last I/O completes.
4645          */
4646         if (TAILQ_EMPTY(&lun->ooa_queue)) {
4647                 mtx_unlock(&lun->lun_lock);
4648                 mtx_lock(&ctl_softc->ctl_lock);
4649                 ctl_free_lun(lun);
4650                 mtx_unlock(&ctl_softc->ctl_lock);
4651         } else
4652                 mtx_unlock(&lun->lun_lock);
4653
4654         return (0);
4655 }
4656
4657 int
4658 ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4659 {
4660         struct ctl_softc *ctl_softc;
4661         struct ctl_lun *lun;
4662
4663         ctl_softc = control_softc;
4664         lun = (struct ctl_lun *)be_lun->ctl_lun;
4665
4666         mtx_lock(&lun->lun_lock);
4667         lun->flags |= CTL_LUN_INOPERABLE;
4668         mtx_unlock(&lun->lun_lock);
4669
4670         return (0);
4671 }
4672
4673 int
4674 ctl_lun_operable(struct ctl_be_lun *be_lun)
4675 {
4676         struct ctl_softc *ctl_softc;
4677         struct ctl_lun *lun;
4678
4679         ctl_softc = control_softc;
4680         lun = (struct ctl_lun *)be_lun->ctl_lun;
4681
4682         mtx_lock(&lun->lun_lock);
4683         lun->flags &= ~CTL_LUN_INOPERABLE;
4684         mtx_unlock(&lun->lun_lock);
4685
4686         return (0);
4687 }
4688
4689 int
4690 ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
4691                    int lock)
4692 {
4693         struct ctl_softc *softc;
4694         struct ctl_lun *lun;
4695         struct copan_aps_subpage *current_sp;
4696         struct ctl_page_index *page_index;
4697         int i;
4698
4699         softc = control_softc;
4700
4701         mtx_lock(&softc->ctl_lock);
4702
4703         lun = (struct ctl_lun *)be_lun->ctl_lun;
4704         mtx_lock(&lun->lun_lock);
4705
4706         page_index = NULL;
4707         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4708                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
4709                      APS_PAGE_CODE)
4710                         continue;
4711
4712                 if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
4713                         continue;
4714                 page_index = &lun->mode_pages.index[i];
4715         }
4716
4717         if (page_index == NULL) {
4718                 mtx_unlock(&lun->lun_lock);
4719                 mtx_unlock(&softc->ctl_lock);
4720                 printf("%s: APS subpage not found for lun %ju!\n", __func__,
4721                        (uintmax_t)lun->lun);
4722                 return (1);
4723         }
4724 #if 0
4725         if ((softc->aps_locked_lun != 0)
4726          && (softc->aps_locked_lun != lun->lun)) {
4727                 printf("%s: attempt to lock LUN %llu when %llu is already "
4728                        "locked\n");
4729                 mtx_unlock(&lun->lun_lock);
4730                 mtx_unlock(&softc->ctl_lock);
4731                 return (1);
4732         }
4733 #endif
4734
4735         current_sp = (struct copan_aps_subpage *)(page_index->page_data +
4736                 (page_index->page_len * CTL_PAGE_CURRENT));
4737
4738         if (lock != 0) {
4739                 current_sp->lock_active = APS_LOCK_ACTIVE;
4740                 softc->aps_locked_lun = lun->lun;
4741         } else {
4742                 current_sp->lock_active = 0;
4743                 softc->aps_locked_lun = 0;
4744         }
4745
4746
4747         /*
4748          * If we're in HA mode, try to send the lock message to the other
4749          * side.
4750          */
4751         if (ctl_is_single == 0) {
4752                 int isc_retval;
4753                 union ctl_ha_msg lock_msg;
4754
4755                 lock_msg.hdr.nexus = *nexus;
4756                 lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
4757                 if (lock != 0)
4758                         lock_msg.aps.lock_flag = 1;
4759                 else
4760                         lock_msg.aps.lock_flag = 0;
4761                 isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
4762                                          sizeof(lock_msg), 0);
4763                 if (isc_retval > CTL_HA_STATUS_SUCCESS) {
4764                         printf("%s: APS (lock=%d) error returned from "
4765                                "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
4766                         mtx_unlock(&lun->lun_lock);
4767                         mtx_unlock(&softc->ctl_lock);
4768                         return (1);
4769                 }
4770         }
4771
4772         mtx_unlock(&lun->lun_lock);
4773         mtx_unlock(&softc->ctl_lock);
4774
4775         return (0);
4776 }
4777
4778 void
4779 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4780 {
4781         struct ctl_lun *lun;
4782         struct ctl_softc *softc;
4783         int i;
4784
4785         softc = control_softc;
4786
4787         lun = (struct ctl_lun *)be_lun->ctl_lun;
4788
4789         mtx_lock(&lun->lun_lock);
4790
4791         for (i = 0; i < CTL_MAX_INITIATORS; i++)
4792                 lun->pending_sense[i].ua_pending |= CTL_UA_CAPACITY_CHANGED;
4793
4794         mtx_unlock(&lun->lun_lock);
4795 }
4796
4797 /*
4798  * Backend "memory move is complete" callback for requests that never
4799  * make it down to say RAIDCore's configuration code.
4800  */
4801 int
4802 ctl_config_move_done(union ctl_io *io)
4803 {
4804         int retval;
4805
4806         retval = CTL_RETVAL_COMPLETE;
4807
4808
4809         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4810         /*
4811          * XXX KDM this shouldn't happen, but what if it does?
4812          */
4813         if (io->io_hdr.io_type != CTL_IO_SCSI)
4814                 panic("I/O type isn't CTL_IO_SCSI!");
4815
4816         if ((io->io_hdr.port_status == 0)
4817          && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4818          && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
4819                 io->io_hdr.status = CTL_SUCCESS;
4820         else if ((io->io_hdr.port_status != 0)
4821               && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4822               && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
4823                 /*
4824                  * For hardware error sense keys, the sense key
4825                  * specific value is defined to be a retry count,
4826                  * but we use it to pass back an internal FETD
4827                  * error code.  XXX KDM  Hopefully the FETD is only
4828                  * using 16 bits for an error code, since that's
4829                  * all the space we have in the sks field.
4830                  */
4831                 ctl_set_internal_failure(&io->scsiio,
4832                                          /*sks_valid*/ 1,
4833                                          /*retry_count*/
4834                                          io->io_hdr.port_status);
4835                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4836                         free(io->scsiio.kern_data_ptr, M_CTL);
4837                 ctl_done(io);
4838                 goto bailout;
4839         }
4840
4841         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
4842          || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
4843          || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4844                 /*
4845                  * XXX KDM just assuming a single pointer here, and not a
4846                  * S/G list.  If we start using S/G lists for config data,
4847                  * we'll need to know how to clean them up here as well.
4848                  */
4849                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4850                         free(io->scsiio.kern_data_ptr, M_CTL);
4851                 /* Hopefully the user has already set the status... */
4852                 ctl_done(io);
4853         } else {
4854                 /*
4855                  * XXX KDM now we need to continue data movement.  Some
4856                  * options:
4857                  * - call ctl_scsiio() again?  We don't do this for data
4858                  *   writes, because for those at least we know ahead of
4859                  *   time where the write will go and how long it is.  For
4860                  *   config writes, though, that information is largely
4861                  *   contained within the write itself, thus we need to
4862                  *   parse out the data again.
4863                  *
4864                  * - Call some other function once the data is in?
4865                  */
4866
4867                 /*
4868                  * XXX KDM call ctl_scsiio() again for now, and check flag
4869                  * bits to see whether we're allocated or not.
4870                  */
4871                 retval = ctl_scsiio(&io->scsiio);
4872         }
4873 bailout:
4874         return (retval);
4875 }
4876
4877 /*
4878  * This gets called by a backend driver when it is done with a
4879  * data_submit method.
4880  */
4881 void
4882 ctl_data_submit_done(union ctl_io *io)
4883 {
4884         /*
4885          * If the IO_CONT flag is set, we need to call the supplied
4886          * function to continue processing the I/O, instead of completing
4887          * the I/O just yet.
4888          *
4889          * If there is an error, though, we don't want to keep processing.
4890          * Instead, just send status back to the initiator.
4891          */
4892         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
4893             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
4894             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4895              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4896                 io->scsiio.io_cont(io);
4897                 return;
4898         }
4899         ctl_done(io);
4900 }
4901
4902 /*
4903  * This gets called by a backend driver when it is done with a
4904  * configuration write.
4905  */
4906 void
4907 ctl_config_write_done(union ctl_io *io)
4908 {
4909         /*
4910          * If the IO_CONT flag is set, we need to call the supplied
4911          * function to continue processing the I/O, instead of completing
4912          * the I/O just yet.
4913          *
4914          * If there is an error, though, we don't want to keep processing.
4915          * Instead, just send status back to the initiator.
4916          */
4917         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
4918          && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
4919           || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
4920                 io->scsiio.io_cont(io);
4921                 return;
4922         }
4923         /*
4924          * Since a configuration write can be done for commands that actually
4925          * have data allocated, like write buffer, and commands that have
4926          * no data, like start/stop unit, we need to check here.
4927          */
4928         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
4929                 free(io->scsiio.kern_data_ptr, M_CTL);
4930         ctl_done(io);
4931 }
4932
4933 /*
4934  * SCSI release command.
4935  */
4936 int
4937 ctl_scsi_release(struct ctl_scsiio *ctsio)
4938 {
4939         int length, longid, thirdparty_id, resv_id;
4940         struct ctl_softc *ctl_softc;
4941         struct ctl_lun *lun;
4942
4943         length = 0;
4944         resv_id = 0;
4945
4946         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
4947
4948         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
4949         ctl_softc = control_softc;
4950
4951         switch (ctsio->cdb[0]) {
4952         case RELEASE_10: {
4953                 struct scsi_release_10 *cdb;
4954
4955                 cdb = (struct scsi_release_10 *)ctsio->cdb;
4956
4957                 if (cdb->byte2 & SR10_LONGID)
4958                         longid = 1;
4959                 else
4960                         thirdparty_id = cdb->thirdparty_id;
4961
4962                 resv_id = cdb->resv_id;
4963                 length = scsi_2btoul(cdb->length);
4964                 break;
4965         }
4966         }
4967
4968
4969         /*
4970          * XXX KDM right now, we only support LUN reservation.  We don't
4971          * support 3rd party reservations, or extent reservations, which
4972          * might actually need the parameter list.  If we've gotten this
4973          * far, we've got a LUN reservation.  Anything else got kicked out
4974          * above.  So, according to SPC, ignore the length.
4975          */
4976         length = 0;
4977
4978         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
4979          && (length > 0)) {
4980                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
4981                 ctsio->kern_data_len = length;
4982                 ctsio->kern_total_len = length;
4983                 ctsio->kern_data_resid = 0;
4984                 ctsio->kern_rel_offset = 0;
4985                 ctsio->kern_sg_entries = 0;
4986                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
4987                 ctsio->be_move_done = ctl_config_move_done;
4988                 ctl_datamove((union ctl_io *)ctsio);
4989
4990                 return (CTL_RETVAL_COMPLETE);
4991         }
4992
4993         if (length > 0)
4994                 thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
4995
4996         mtx_lock(&lun->lun_lock);
4997
4998         /*
4999          * According to SPC, it is not an error for an intiator to attempt
5000          * to release a reservation on a LUN that isn't reserved, or that
5001          * is reserved by another initiator.  The reservation can only be
5002          * released, though, by the initiator who made it or by one of
5003          * several reset type events.
5004          */
5005         if (lun->flags & CTL_LUN_RESERVED) {
5006                 if ((ctsio->io_hdr.nexus.initid.id == lun->rsv_nexus.initid.id)
5007                  && (ctsio->io_hdr.nexus.targ_port == lun->rsv_nexus.targ_port)
5008                  && (ctsio->io_hdr.nexus.targ_target.id ==
5009                      lun->rsv_nexus.targ_target.id)) {
5010                         lun->flags &= ~CTL_LUN_RESERVED;
5011                 }
5012         }
5013
5014         mtx_unlock(&lun->lun_lock);
5015
5016         ctsio->scsi_status = SCSI_STATUS_OK;
5017         ctsio->io_hdr.status = CTL_SUCCESS;
5018
5019         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5020                 free(ctsio->kern_data_ptr, M_CTL);
5021                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5022         }
5023
5024         ctl_done((union ctl_io *)ctsio);
5025         return (CTL_RETVAL_COMPLETE);
5026 }
5027
5028 int
5029 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5030 {
5031         int extent, thirdparty, longid;
5032         int resv_id, length;
5033         uint64_t thirdparty_id;
5034         struct ctl_softc *ctl_softc;
5035         struct ctl_lun *lun;
5036
5037         extent = 0;
5038         thirdparty = 0;
5039         longid = 0;
5040         resv_id = 0;
5041         length = 0;
5042         thirdparty_id = 0;
5043
5044         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5045
5046         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5047         ctl_softc = control_softc;
5048
5049         switch (ctsio->cdb[0]) {
5050         case RESERVE_10: {
5051                 struct scsi_reserve_10 *cdb;
5052
5053                 cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5054
5055                 if (cdb->byte2 & SR10_LONGID)
5056                         longid = 1;
5057                 else
5058                         thirdparty_id = cdb->thirdparty_id;
5059
5060                 resv_id = cdb->resv_id;
5061                 length = scsi_2btoul(cdb->length);
5062                 break;
5063         }
5064         }
5065
5066         /*
5067          * XXX KDM right now, we only support LUN reservation.  We don't
5068          * support 3rd party reservations, or extent reservations, which
5069          * might actually need the parameter list.  If we've gotten this
5070          * far, we've got a LUN reservation.  Anything else got kicked out
5071          * above.  So, according to SPC, ignore the length.
5072          */
5073         length = 0;
5074
5075         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5076          && (length > 0)) {
5077                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5078                 ctsio->kern_data_len = length;
5079                 ctsio->kern_total_len = length;
5080                 ctsio->kern_data_resid = 0;
5081                 ctsio->kern_rel_offset = 0;
5082                 ctsio->kern_sg_entries = 0;
5083                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5084                 ctsio->be_move_done = ctl_config_move_done;
5085                 ctl_datamove((union ctl_io *)ctsio);
5086
5087                 return (CTL_RETVAL_COMPLETE);
5088         }
5089
5090         if (length > 0)
5091                 thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5092
5093         mtx_lock(&lun->lun_lock);
5094         if (lun->flags & CTL_LUN_RESERVED) {
5095                 if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
5096                  || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
5097                  || (ctsio->io_hdr.nexus.targ_target.id !=
5098                      lun->rsv_nexus.targ_target.id)) {
5099                         ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5100                         ctsio->io_hdr.status = CTL_SCSI_ERROR;
5101                         goto bailout;
5102                 }
5103         }
5104
5105         lun->flags |= CTL_LUN_RESERVED;
5106         lun->rsv_nexus = ctsio->io_hdr.nexus;
5107
5108         ctsio->scsi_status = SCSI_STATUS_OK;
5109         ctsio->io_hdr.status = CTL_SUCCESS;
5110
5111 bailout:
5112         mtx_unlock(&lun->lun_lock);
5113
5114         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5115                 free(ctsio->kern_data_ptr, M_CTL);
5116                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5117         }
5118
5119         ctl_done((union ctl_io *)ctsio);
5120         return (CTL_RETVAL_COMPLETE);
5121 }
5122
5123 int
5124 ctl_start_stop(struct ctl_scsiio *ctsio)
5125 {
5126         struct scsi_start_stop_unit *cdb;
5127         struct ctl_lun *lun;
5128         struct ctl_softc *ctl_softc;
5129         int retval;
5130
5131         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5132
5133         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5134         ctl_softc = control_softc;
5135         retval = 0;
5136
5137         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5138
5139         /*
5140          * XXX KDM
5141          * We don't support the immediate bit on a stop unit.  In order to
5142          * do that, we would need to code up a way to know that a stop is
5143          * pending, and hold off any new commands until it completes, one
5144          * way or another.  Then we could accept or reject those commands
5145          * depending on its status.  We would almost need to do the reverse
5146          * of what we do below for an immediate start -- return the copy of
5147          * the ctl_io to the FETD with status to send to the host (and to
5148          * free the copy!) and then free the original I/O once the stop
5149          * actually completes.  That way, the OOA queue mechanism can work
5150          * to block commands that shouldn't proceed.  Another alternative
5151          * would be to put the copy in the queue in place of the original,
5152          * and return the original back to the caller.  That could be
5153          * slightly safer..
5154          */
5155         if ((cdb->byte2 & SSS_IMMED)
5156          && ((cdb->how & SSS_START) == 0)) {
5157                 ctl_set_invalid_field(ctsio,
5158                                       /*sks_valid*/ 1,
5159                                       /*command*/ 1,
5160                                       /*field*/ 1,
5161                                       /*bit_valid*/ 1,
5162                                       /*bit*/ 0);
5163                 ctl_done((union ctl_io *)ctsio);
5164                 return (CTL_RETVAL_COMPLETE);
5165         }
5166
5167         if ((lun->flags & CTL_LUN_PR_RESERVED)
5168          && ((cdb->how & SSS_START)==0)) {
5169                 uint32_t residx;
5170
5171                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5172                 if (!lun->per_res[residx].registered
5173                  || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5174
5175                         ctl_set_reservation_conflict(ctsio);
5176                         ctl_done((union ctl_io *)ctsio);
5177                         return (CTL_RETVAL_COMPLETE);
5178                 }
5179         }
5180
5181         /*
5182          * If there is no backend on this device, we can't start or stop
5183          * it.  In theory we shouldn't get any start/stop commands in the
5184          * first place at this level if the LUN doesn't have a backend.
5185          * That should get stopped by the command decode code.
5186          */
5187         if (lun->backend == NULL) {
5188                 ctl_set_invalid_opcode(ctsio);
5189                 ctl_done((union ctl_io *)ctsio);
5190                 return (CTL_RETVAL_COMPLETE);
5191         }
5192
5193         /*
5194          * XXX KDM Copan-specific offline behavior.
5195          * Figure out a reasonable way to port this?
5196          */
5197 #ifdef NEEDTOPORT
5198         mtx_lock(&lun->lun_lock);
5199
5200         if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5201          && (lun->flags & CTL_LUN_OFFLINE)) {
5202                 /*
5203                  * If the LUN is offline, and the on/offline bit isn't set,
5204                  * reject the start or stop.  Otherwise, let it through.
5205                  */
5206                 mtx_unlock(&lun->lun_lock);
5207                 ctl_set_lun_not_ready(ctsio);
5208                 ctl_done((union ctl_io *)ctsio);
5209         } else {
5210                 mtx_unlock(&lun->lun_lock);
5211 #endif /* NEEDTOPORT */
5212                 /*
5213                  * This could be a start or a stop when we're online,
5214                  * or a stop/offline or start/online.  A start or stop when
5215                  * we're offline is covered in the case above.
5216                  */
5217                 /*
5218                  * In the non-immediate case, we send the request to
5219                  * the backend and return status to the user when
5220                  * it is done.
5221                  *
5222                  * In the immediate case, we allocate a new ctl_io
5223                  * to hold a copy of the request, and send that to
5224                  * the backend.  We then set good status on the
5225                  * user's request and return it immediately.
5226                  */
5227                 if (cdb->byte2 & SSS_IMMED) {
5228                         union ctl_io *new_io;
5229
5230                         new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5231                         if (new_io == NULL) {
5232                                 ctl_set_busy(ctsio);
5233                                 ctl_done((union ctl_io *)ctsio);
5234                         } else {
5235                                 ctl_copy_io((union ctl_io *)ctsio,
5236                                             new_io);
5237                                 retval = lun->backend->config_write(new_io);
5238                                 ctl_set_success(ctsio);
5239                                 ctl_done((union ctl_io *)ctsio);
5240                         }
5241                 } else {
5242                         retval = lun->backend->config_write(
5243                                 (union ctl_io *)ctsio);
5244                 }
5245 #ifdef NEEDTOPORT
5246         }
5247 #endif
5248         return (retval);
5249 }
5250
5251 /*
5252  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5253  * we don't really do anything with the LBA and length fields if the user
5254  * passes them in.  Instead we'll just flush out the cache for the entire
5255  * LUN.
5256  */
5257 int
5258 ctl_sync_cache(struct ctl_scsiio *ctsio)
5259 {
5260         struct ctl_lun *lun;
5261         struct ctl_softc *ctl_softc;
5262         uint64_t starting_lba;
5263         uint32_t block_count;
5264         int retval;
5265
5266         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5267
5268         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5269         ctl_softc = control_softc;
5270         retval = 0;
5271
5272         switch (ctsio->cdb[0]) {
5273         case SYNCHRONIZE_CACHE: {
5274                 struct scsi_sync_cache *cdb;
5275                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5276
5277                 starting_lba = scsi_4btoul(cdb->begin_lba);
5278                 block_count = scsi_2btoul(cdb->lb_count);
5279                 break;
5280         }
5281         case SYNCHRONIZE_CACHE_16: {
5282                 struct scsi_sync_cache_16 *cdb;
5283                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5284
5285                 starting_lba = scsi_8btou64(cdb->begin_lba);
5286                 block_count = scsi_4btoul(cdb->lb_count);
5287                 break;
5288         }
5289         default:
5290                 ctl_set_invalid_opcode(ctsio);
5291                 ctl_done((union ctl_io *)ctsio);
5292                 goto bailout;
5293                 break; /* NOTREACHED */
5294         }
5295
5296         /*
5297          * We check the LBA and length, but don't do anything with them.
5298          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5299          * get flushed.  This check will just help satisfy anyone who wants
5300          * to see an error for an out of range LBA.
5301          */
5302         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5303                 ctl_set_lba_out_of_range(ctsio);
5304                 ctl_done((union ctl_io *)ctsio);
5305                 goto bailout;
5306         }
5307
5308         /*
5309          * If this LUN has no backend, we can't flush the cache anyway.
5310          */
5311         if (lun->backend == NULL) {
5312                 ctl_set_invalid_opcode(ctsio);
5313                 ctl_done((union ctl_io *)ctsio);
5314                 goto bailout;
5315         }
5316
5317         /*
5318          * Check to see whether we're configured to send the SYNCHRONIZE
5319          * CACHE command directly to the back end.
5320          */
5321         mtx_lock(&lun->lun_lock);
5322         if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5323          && (++(lun->sync_count) >= lun->sync_interval)) {
5324                 lun->sync_count = 0;
5325                 mtx_unlock(&lun->lun_lock);
5326                 retval = lun->backend->config_write((union ctl_io *)ctsio);
5327         } else {
5328                 mtx_unlock(&lun->lun_lock);
5329                 ctl_set_success(ctsio);
5330                 ctl_done((union ctl_io *)ctsio);
5331         }
5332
5333 bailout:
5334
5335         return (retval);
5336 }
5337
5338 int
5339 ctl_format(struct ctl_scsiio *ctsio)
5340 {
5341         struct scsi_format *cdb;
5342         struct ctl_lun *lun;
5343         struct ctl_softc *ctl_softc;
5344         int length, defect_list_len;
5345
5346         CTL_DEBUG_PRINT(("ctl_format\n"));
5347
5348         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5349         ctl_softc = control_softc;
5350
5351         cdb = (struct scsi_format *)ctsio->cdb;
5352
5353         length = 0;
5354         if (cdb->byte2 & SF_FMTDATA) {
5355                 if (cdb->byte2 & SF_LONGLIST)
5356                         length = sizeof(struct scsi_format_header_long);
5357                 else
5358                         length = sizeof(struct scsi_format_header_short);
5359         }
5360
5361         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5362          && (length > 0)) {
5363                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5364                 ctsio->kern_data_len = length;
5365                 ctsio->kern_total_len = length;
5366                 ctsio->kern_data_resid = 0;
5367                 ctsio->kern_rel_offset = 0;
5368                 ctsio->kern_sg_entries = 0;
5369                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5370                 ctsio->be_move_done = ctl_config_move_done;
5371                 ctl_datamove((union ctl_io *)ctsio);
5372
5373                 return (CTL_RETVAL_COMPLETE);
5374         }
5375
5376         defect_list_len = 0;
5377
5378         if (cdb->byte2 & SF_FMTDATA) {
5379                 if (cdb->byte2 & SF_LONGLIST) {
5380                         struct scsi_format_header_long *header;
5381
5382                         header = (struct scsi_format_header_long *)
5383                                 ctsio->kern_data_ptr;
5384
5385                         defect_list_len = scsi_4btoul(header->defect_list_len);
5386                         if (defect_list_len != 0) {
5387                                 ctl_set_invalid_field(ctsio,
5388                                                       /*sks_valid*/ 1,
5389                                                       /*command*/ 0,
5390                                                       /*field*/ 2,
5391                                                       /*bit_valid*/ 0,
5392                                                       /*bit*/ 0);
5393                                 goto bailout;
5394                         }
5395                 } else {
5396                         struct scsi_format_header_short *header;
5397
5398                         header = (struct scsi_format_header_short *)
5399                                 ctsio->kern_data_ptr;
5400
5401                         defect_list_len = scsi_2btoul(header->defect_list_len);
5402                         if (defect_list_len != 0) {
5403                                 ctl_set_invalid_field(ctsio,
5404                                                       /*sks_valid*/ 1,
5405                                                       /*command*/ 0,
5406                                                       /*field*/ 2,
5407                                                       /*bit_valid*/ 0,
5408                                                       /*bit*/ 0);
5409                                 goto bailout;
5410                         }
5411                 }
5412         }
5413
5414         /*
5415          * The format command will clear out the "Medium format corrupted"
5416          * status if set by the configuration code.  That status is really
5417          * just a way to notify the host that we have lost the media, and
5418          * get them to issue a command that will basically make them think
5419          * they're blowing away the media.
5420          */
5421         mtx_lock(&lun->lun_lock);
5422         lun->flags &= ~CTL_LUN_INOPERABLE;
5423         mtx_unlock(&lun->lun_lock);
5424
5425         ctsio->scsi_status = SCSI_STATUS_OK;
5426         ctsio->io_hdr.status = CTL_SUCCESS;
5427 bailout:
5428
5429         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5430                 free(ctsio->kern_data_ptr, M_CTL);
5431                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5432         }
5433
5434         ctl_done((union ctl_io *)ctsio);
5435         return (CTL_RETVAL_COMPLETE);
5436 }
5437
5438 int
5439 ctl_read_buffer(struct ctl_scsiio *ctsio)
5440 {
5441         struct scsi_read_buffer *cdb;
5442         struct ctl_lun *lun;
5443         int buffer_offset, len;
5444         static uint8_t descr[4];
5445         static uint8_t echo_descr[4] = { 0 };
5446
5447         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5448
5449         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5450         cdb = (struct scsi_read_buffer *)ctsio->cdb;
5451
5452         if (lun->flags & CTL_LUN_PR_RESERVED) {
5453                 uint32_t residx;
5454
5455                 /*
5456                  * XXX KDM need a lock here.
5457                  */
5458                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5459                 if ((lun->res_type == SPR_TYPE_EX_AC
5460                   && residx != lun->pr_res_idx)
5461                  || ((lun->res_type == SPR_TYPE_EX_AC_RO
5462                    || lun->res_type == SPR_TYPE_EX_AC_AR)
5463                   && !lun->per_res[residx].registered)) {
5464                         ctl_set_reservation_conflict(ctsio);
5465                         ctl_done((union ctl_io *)ctsio);
5466                         return (CTL_RETVAL_COMPLETE);
5467                 }
5468         }
5469
5470         if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5471             (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5472             (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5473                 ctl_set_invalid_field(ctsio,
5474                                       /*sks_valid*/ 1,
5475                                       /*command*/ 1,
5476                                       /*field*/ 1,
5477                                       /*bit_valid*/ 1,
5478                                       /*bit*/ 4);
5479                 ctl_done((union ctl_io *)ctsio);
5480                 return (CTL_RETVAL_COMPLETE);
5481         }
5482
5483         len = scsi_3btoul(cdb->length);
5484         buffer_offset = scsi_3btoul(cdb->offset);
5485
5486         if (buffer_offset + len > sizeof(lun->write_buffer)) {
5487                 ctl_set_invalid_field(ctsio,
5488                                       /*sks_valid*/ 1,
5489                                       /*command*/ 1,
5490                                       /*field*/ 6,
5491                                       /*bit_valid*/ 0,
5492                                       /*bit*/ 0);
5493                 ctl_done((union ctl_io *)ctsio);
5494                 return (CTL_RETVAL_COMPLETE);
5495         }
5496
5497         if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5498                 descr[0] = 0;
5499                 scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5500                 ctsio->kern_data_ptr = descr;
5501                 len = min(len, sizeof(descr));
5502         } else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5503                 ctsio->kern_data_ptr = echo_descr;
5504                 len = min(len, sizeof(echo_descr));
5505         } else
5506                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5507         ctsio->kern_data_len = len;
5508         ctsio->kern_total_len = len;
5509         ctsio->kern_data_resid = 0;
5510         ctsio->kern_rel_offset = 0;
5511         ctsio->kern_sg_entries = 0;
5512         ctsio->be_move_done = ctl_config_move_done;
5513         ctl_datamove((union ctl_io *)ctsio);
5514
5515         return (CTL_RETVAL_COMPLETE);
5516 }
5517
5518 int
5519 ctl_write_buffer(struct ctl_scsiio *ctsio)
5520 {
5521         struct scsi_write_buffer *cdb;
5522         struct ctl_lun *lun;
5523         int buffer_offset, len;
5524
5525         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5526
5527         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5528         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5529
5530         if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5531                 ctl_set_invalid_field(ctsio,
5532                                       /*sks_valid*/ 1,
5533                                       /*command*/ 1,
5534                                       /*field*/ 1,
5535                                       /*bit_valid*/ 1,
5536                                       /*bit*/ 4);
5537                 ctl_done((union ctl_io *)ctsio);
5538                 return (CTL_RETVAL_COMPLETE);
5539         }
5540
5541         len = scsi_3btoul(cdb->length);
5542         buffer_offset = scsi_3btoul(cdb->offset);
5543
5544         if (buffer_offset + len > sizeof(lun->write_buffer)) {
5545                 ctl_set_invalid_field(ctsio,
5546                                       /*sks_valid*/ 1,
5547                                       /*command*/ 1,
5548                                       /*field*/ 6,
5549                                       /*bit_valid*/ 0,
5550                                       /*bit*/ 0);
5551                 ctl_done((union ctl_io *)ctsio);
5552                 return (CTL_RETVAL_COMPLETE);
5553         }
5554
5555         /*
5556          * If we've got a kernel request that hasn't been malloced yet,
5557          * malloc it and tell the caller the data buffer is here.
5558          */
5559         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5560                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5561                 ctsio->kern_data_len = len;
5562                 ctsio->kern_total_len = len;
5563                 ctsio->kern_data_resid = 0;
5564                 ctsio->kern_rel_offset = 0;
5565                 ctsio->kern_sg_entries = 0;
5566                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5567                 ctsio->be_move_done = ctl_config_move_done;
5568                 ctl_datamove((union ctl_io *)ctsio);
5569
5570                 return (CTL_RETVAL_COMPLETE);
5571         }
5572
5573         ctl_done((union ctl_io *)ctsio);
5574
5575         return (CTL_RETVAL_COMPLETE);
5576 }
5577
5578 int
5579 ctl_write_same(struct ctl_scsiio *ctsio)
5580 {
5581         struct ctl_lun *lun;
5582         struct ctl_lba_len_flags *lbalen;
5583         uint64_t lba;
5584         uint32_t num_blocks;
5585         int len, retval;
5586         uint8_t byte2;
5587
5588         retval = CTL_RETVAL_COMPLETE;
5589
5590         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5591
5592         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5593
5594         switch (ctsio->cdb[0]) {
5595         case WRITE_SAME_10: {
5596                 struct scsi_write_same_10 *cdb;
5597
5598                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5599
5600                 lba = scsi_4btoul(cdb->addr);
5601                 num_blocks = scsi_2btoul(cdb->length);
5602                 byte2 = cdb->byte2;
5603                 break;
5604         }
5605         case WRITE_SAME_16: {
5606                 struct scsi_write_same_16 *cdb;
5607
5608                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5609
5610                 lba = scsi_8btou64(cdb->addr);
5611                 num_blocks = scsi_4btoul(cdb->length);
5612                 byte2 = cdb->byte2;
5613                 break;
5614         }
5615         default:
5616                 /*
5617                  * We got a command we don't support.  This shouldn't
5618                  * happen, commands should be filtered out above us.
5619                  */
5620                 ctl_set_invalid_opcode(ctsio);
5621                 ctl_done((union ctl_io *)ctsio);
5622
5623                 return (CTL_RETVAL_COMPLETE);
5624                 break; /* NOTREACHED */
5625         }
5626
5627         /*
5628          * The first check is to make sure we're in bounds, the second
5629          * check is to catch wrap-around problems.  If the lba + num blocks
5630          * is less than the lba, then we've wrapped around and the block
5631          * range is invalid anyway.
5632          */
5633         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5634          || ((lba + num_blocks) < lba)) {
5635                 ctl_set_lba_out_of_range(ctsio);
5636                 ctl_done((union ctl_io *)ctsio);
5637                 return (CTL_RETVAL_COMPLETE);
5638         }
5639
5640         /* Zero number of blocks means "to the last logical block" */
5641         if (num_blocks == 0) {
5642                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5643                         ctl_set_invalid_field(ctsio,
5644                                               /*sks_valid*/ 0,
5645                                               /*command*/ 1,
5646                                               /*field*/ 0,
5647                                               /*bit_valid*/ 0,
5648                                               /*bit*/ 0);
5649                         ctl_done((union ctl_io *)ctsio);
5650                         return (CTL_RETVAL_COMPLETE);
5651                 }
5652                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
5653         }
5654
5655         len = lun->be_lun->blocksize;
5656
5657         /*
5658          * If we've got a kernel request that hasn't been malloced yet,
5659          * malloc it and tell the caller the data buffer is here.
5660          */
5661         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5662                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5663                 ctsio->kern_data_len = len;
5664                 ctsio->kern_total_len = len;
5665                 ctsio->kern_data_resid = 0;
5666                 ctsio->kern_rel_offset = 0;
5667                 ctsio->kern_sg_entries = 0;
5668                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5669                 ctsio->be_move_done = ctl_config_move_done;
5670                 ctl_datamove((union ctl_io *)ctsio);
5671
5672                 return (CTL_RETVAL_COMPLETE);
5673         }
5674
5675         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5676         lbalen->lba = lba;
5677         lbalen->len = num_blocks;
5678         lbalen->flags = byte2;
5679         retval = lun->backend->config_write((union ctl_io *)ctsio);
5680
5681         return (retval);
5682 }
5683
5684 int
5685 ctl_unmap(struct ctl_scsiio *ctsio)
5686 {
5687         struct ctl_lun *lun;
5688         struct scsi_unmap *cdb;
5689         struct ctl_ptr_len_flags *ptrlen;
5690         struct scsi_unmap_header *hdr;
5691         struct scsi_unmap_desc *buf, *end;
5692         uint64_t lba;
5693         uint32_t num_blocks;
5694         int len, retval;
5695         uint8_t byte2;
5696
5697         retval = CTL_RETVAL_COMPLETE;
5698
5699         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5700
5701         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5702         cdb = (struct scsi_unmap *)ctsio->cdb;
5703
5704         len = scsi_2btoul(cdb->length);
5705         byte2 = cdb->byte2;
5706
5707         /*
5708          * If we've got a kernel request that hasn't been malloced yet,
5709          * malloc it and tell the caller the data buffer is here.
5710          */
5711         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5712                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5713                 ctsio->kern_data_len = len;
5714                 ctsio->kern_total_len = len;
5715                 ctsio->kern_data_resid = 0;
5716                 ctsio->kern_rel_offset = 0;
5717                 ctsio->kern_sg_entries = 0;
5718                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5719                 ctsio->be_move_done = ctl_config_move_done;
5720                 ctl_datamove((union ctl_io *)ctsio);
5721
5722                 return (CTL_RETVAL_COMPLETE);
5723         }
5724
5725         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5726         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5727         if (len < sizeof (*hdr) ||
5728             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5729             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5730             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5731                 ctl_set_invalid_field(ctsio,
5732                                       /*sks_valid*/ 0,
5733                                       /*command*/ 0,
5734                                       /*field*/ 0,
5735                                       /*bit_valid*/ 0,
5736                                       /*bit*/ 0);
5737                 ctl_done((union ctl_io *)ctsio);
5738                 return (CTL_RETVAL_COMPLETE);
5739         }
5740         len = scsi_2btoul(hdr->desc_length);
5741         buf = (struct scsi_unmap_desc *)(hdr + 1);
5742         end = buf + len / sizeof(*buf);
5743
5744         ptrlen = (struct ctl_ptr_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5745         ptrlen->ptr = (void *)buf;
5746         ptrlen->len = len;
5747         ptrlen->flags = byte2;
5748
5749         for (; buf < end; buf++) {
5750                 lba = scsi_8btou64(buf->lba);
5751                 num_blocks = scsi_4btoul(buf->length);
5752                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5753                  || ((lba + num_blocks) < lba)) {
5754                         ctl_set_lba_out_of_range(ctsio);
5755                         ctl_done((union ctl_io *)ctsio);
5756                         return (CTL_RETVAL_COMPLETE);
5757                 }
5758         }
5759
5760         retval = lun->backend->config_write((union ctl_io *)ctsio);
5761
5762         return (retval);
5763 }
5764
5765 /*
5766  * Note that this function currently doesn't actually do anything inside
5767  * CTL to enforce things if the DQue bit is turned on.
5768  *
5769  * Also note that this function can't be used in the default case, because
5770  * the DQue bit isn't set in the changeable mask for the control mode page
5771  * anyway.  This is just here as an example for how to implement a page
5772  * handler, and a placeholder in case we want to allow the user to turn
5773  * tagged queueing on and off.
5774  *
5775  * The D_SENSE bit handling is functional, however, and will turn
5776  * descriptor sense on and off for a given LUN.
5777  */
5778 int
5779 ctl_control_page_handler(struct ctl_scsiio *ctsio,
5780                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5781 {
5782         struct scsi_control_page *current_cp, *saved_cp, *user_cp;
5783         struct ctl_lun *lun;
5784         struct ctl_softc *softc;
5785         int set_ua;
5786         uint32_t initidx;
5787
5788         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5789         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5790         set_ua = 0;
5791
5792         user_cp = (struct scsi_control_page *)page_ptr;
5793         current_cp = (struct scsi_control_page *)
5794                 (page_index->page_data + (page_index->page_len *
5795                 CTL_PAGE_CURRENT));
5796         saved_cp = (struct scsi_control_page *)
5797                 (page_index->page_data + (page_index->page_len *
5798                 CTL_PAGE_SAVED));
5799
5800         softc = control_softc;
5801
5802         mtx_lock(&lun->lun_lock);
5803         if (((current_cp->rlec & SCP_DSENSE) == 0)
5804          && ((user_cp->rlec & SCP_DSENSE) != 0)) {
5805                 /*
5806                  * Descriptor sense is currently turned off and the user
5807                  * wants to turn it on.
5808                  */
5809                 current_cp->rlec |= SCP_DSENSE;
5810                 saved_cp->rlec |= SCP_DSENSE;
5811                 lun->flags |= CTL_LUN_SENSE_DESC;
5812                 set_ua = 1;
5813         } else if (((current_cp->rlec & SCP_DSENSE) != 0)
5814                 && ((user_cp->rlec & SCP_DSENSE) == 0)) {
5815                 /*
5816                  * Descriptor sense is currently turned on, and the user
5817                  * wants to turn it off.
5818                  */
5819                 current_cp->rlec &= ~SCP_DSENSE;
5820                 saved_cp->rlec &= ~SCP_DSENSE;
5821                 lun->flags &= ~CTL_LUN_SENSE_DESC;
5822                 set_ua = 1;
5823         }
5824         if (current_cp->queue_flags & SCP_QUEUE_DQUE) {
5825                 if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
5826 #ifdef NEEDTOPORT
5827                         csevent_log(CSC_CTL | CSC_SHELF_SW |
5828                                     CTL_UNTAG_TO_UNTAG,
5829                                     csevent_LogType_Trace,
5830                                     csevent_Severity_Information,
5831                                     csevent_AlertLevel_Green,
5832                                     csevent_FRU_Firmware,
5833                                     csevent_FRU_Unknown,
5834                                     "Received untagged to untagged transition");
5835 #endif /* NEEDTOPORT */
5836                 } else {
5837 #ifdef NEEDTOPORT
5838                         csevent_log(CSC_CTL | CSC_SHELF_SW |
5839                                     CTL_UNTAG_TO_TAG,
5840                                     csevent_LogType_ConfigChange,
5841                                     csevent_Severity_Information,
5842                                     csevent_AlertLevel_Green,
5843                                     csevent_FRU_Firmware,
5844                                     csevent_FRU_Unknown,
5845                                     "Received untagged to tagged "
5846                                     "queueing transition");
5847 #endif /* NEEDTOPORT */
5848
5849                         current_cp->queue_flags &= ~SCP_QUEUE_DQUE;
5850                         saved_cp->queue_flags &= ~SCP_QUEUE_DQUE;
5851                         set_ua = 1;
5852                 }
5853         } else {
5854                 if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
5855 #ifdef NEEDTOPORT
5856                         csevent_log(CSC_CTL | CSC_SHELF_SW |
5857                                     CTL_TAG_TO_UNTAG,
5858                                     csevent_LogType_ConfigChange,
5859                                     csevent_Severity_Warning,
5860                                     csevent_AlertLevel_Yellow,
5861                                     csevent_FRU_Firmware,
5862                                     csevent_FRU_Unknown,
5863                                     "Received tagged queueing to untagged "
5864                                     "transition");
5865 #endif /* NEEDTOPORT */
5866
5867                         current_cp->queue_flags |= SCP_QUEUE_DQUE;
5868                         saved_cp->queue_flags |= SCP_QUEUE_DQUE;
5869                         set_ua = 1;
5870                 } else {
5871 #ifdef NEEDTOPORT
5872                         csevent_log(CSC_CTL | CSC_SHELF_SW |
5873                                     CTL_TAG_TO_TAG,
5874                                     csevent_LogType_Trace,
5875                                     csevent_Severity_Information,
5876                                     csevent_AlertLevel_Green,
5877                                     csevent_FRU_Firmware,
5878                                     csevent_FRU_Unknown,
5879                                     "Received tagged queueing to tagged "
5880                                     "queueing transition");
5881 #endif /* NEEDTOPORT */
5882                 }
5883         }
5884         if (set_ua != 0) {
5885                 int i;
5886                 /*
5887                  * Let other initiators know that the mode
5888                  * parameters for this LUN have changed.
5889                  */
5890                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
5891                         if (i == initidx)
5892                                 continue;
5893
5894                         lun->pending_sense[i].ua_pending |=
5895                                 CTL_UA_MODE_CHANGE;
5896                 }
5897         }
5898         mtx_unlock(&lun->lun_lock);
5899
5900         return (0);
5901 }
5902
5903 int
5904 ctl_power_sp_handler(struct ctl_scsiio *ctsio,
5905                      struct ctl_page_index *page_index, uint8_t *page_ptr)
5906 {
5907         return (0);
5908 }
5909
5910 int
5911 ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
5912                            struct ctl_page_index *page_index, int pc)
5913 {
5914         struct copan_power_subpage *page;
5915
5916         page = (struct copan_power_subpage *)page_index->page_data +
5917                 (page_index->page_len * pc);
5918
5919         switch (pc) {
5920         case SMS_PAGE_CTRL_CHANGEABLE >> 6:
5921                 /*
5922                  * We don't update the changable bits for this page.
5923                  */
5924                 break;
5925         case SMS_PAGE_CTRL_CURRENT >> 6:
5926         case SMS_PAGE_CTRL_DEFAULT >> 6:
5927         case SMS_PAGE_CTRL_SAVED >> 6:
5928 #ifdef NEEDTOPORT
5929                 ctl_update_power_subpage(page);
5930 #endif
5931                 break;
5932         default:
5933 #ifdef NEEDTOPORT
5934                 EPRINT(0, "Invalid PC %d!!", pc);
5935 #endif
5936                 break;
5937         }
5938         return (0);
5939 }
5940
5941
5942 int
5943 ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
5944                    struct ctl_page_index *page_index, uint8_t *page_ptr)
5945 {
5946         struct copan_aps_subpage *user_sp;
5947         struct copan_aps_subpage *current_sp;
5948         union ctl_modepage_info *modepage_info;
5949         struct ctl_softc *softc;
5950         struct ctl_lun *lun;
5951         int retval;
5952
5953         retval = CTL_RETVAL_COMPLETE;
5954         current_sp = (struct copan_aps_subpage *)(page_index->page_data +
5955                      (page_index->page_len * CTL_PAGE_CURRENT));
5956         softc = control_softc;
5957         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5958
5959         user_sp = (struct copan_aps_subpage *)page_ptr;
5960
5961         modepage_info = (union ctl_modepage_info *)
5962                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5963
5964         modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
5965         modepage_info->header.subpage = page_index->subpage;
5966         modepage_info->aps.lock_active = user_sp->lock_active;
5967
5968         mtx_lock(&softc->ctl_lock);
5969
5970         /*
5971          * If there is a request to lock the LUN and another LUN is locked
5972          * this is an error. If the requested LUN is already locked ignore
5973          * the request. If no LUN is locked attempt to lock it.
5974          * if there is a request to unlock the LUN and the LUN is currently
5975          * locked attempt to unlock it. Otherwise ignore the request. i.e.
5976          * if another LUN is locked or no LUN is locked.
5977          */
5978         if (user_sp->lock_active & APS_LOCK_ACTIVE) {
5979                 if (softc->aps_locked_lun == lun->lun) {
5980                         /*
5981                          * This LUN is already locked, so we're done.
5982                          */
5983                         retval = CTL_RETVAL_COMPLETE;
5984                 } else if (softc->aps_locked_lun == 0) {
5985                         /*
5986                          * No one has the lock, pass the request to the
5987                          * backend.
5988                          */
5989                         retval = lun->backend->config_write(
5990                                 (union ctl_io *)ctsio);
5991                 } else {
5992                         /*
5993                          * Someone else has the lock, throw out the request.
5994                          */
5995                         ctl_set_already_locked(ctsio);
5996                         free(ctsio->kern_data_ptr, M_CTL);
5997                         ctl_done((union ctl_io *)ctsio);
5998
5999                         /*
6000                          * Set the return value so that ctl_do_mode_select()
6001                          * won't try to complete the command.  We already
6002                          * completed it here.
6003                          */
6004                         retval = CTL_RETVAL_ERROR;
6005                 }
6006         } else if (softc->aps_locked_lun == lun->lun) {
6007                 /*
6008                  * This LUN is locked, so pass the unlock request to the
6009                  * backend.
6010                  */
6011                 retval = lun->backend->config_write((union ctl_io *)ctsio);
6012         }
6013         mtx_unlock(&softc->ctl_lock);
6014
6015         return (retval);
6016 }
6017
6018 int
6019 ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6020                                 struct ctl_page_index *page_index,
6021                                 uint8_t *page_ptr)
6022 {
6023         uint8_t *c;
6024         int i;
6025
6026         c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6027         ctl_time_io_secs =
6028                 (c[0] << 8) |
6029                 (c[1] << 0) |
6030                 0;
6031         CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6032         printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6033         printf("page data:");
6034         for (i=0; i<8; i++)
6035                 printf(" %.2x",page_ptr[i]);
6036         printf("\n");
6037         return (0);
6038 }
6039
6040 int
6041 ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6042                                struct ctl_page_index *page_index,
6043                                int pc)
6044 {
6045         struct copan_debugconf_subpage *page;
6046
6047         page = (struct copan_debugconf_subpage *)page_index->page_data +
6048                 (page_index->page_len * pc);
6049
6050         switch (pc) {
6051         case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6052         case SMS_PAGE_CTRL_DEFAULT >> 6:
6053         case SMS_PAGE_CTRL_SAVED >> 6:
6054                 /*
6055                  * We don't update the changable or default bits for this page.
6056                  */
6057                 break;
6058         case SMS_PAGE_CTRL_CURRENT >> 6:
6059                 page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6060                 page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6061                 break;
6062         default:
6063 #ifdef NEEDTOPORT
6064                 EPRINT(0, "Invalid PC %d!!", pc);
6065 #endif /* NEEDTOPORT */
6066                 break;
6067         }
6068         return (0);
6069 }
6070
6071
6072 static int
6073 ctl_do_mode_select(union ctl_io *io)
6074 {
6075         struct scsi_mode_page_header *page_header;
6076         struct ctl_page_index *page_index;
6077         struct ctl_scsiio *ctsio;
6078         int control_dev, page_len;
6079         int page_len_offset, page_len_size;
6080         union ctl_modepage_info *modepage_info;
6081         struct ctl_lun *lun;
6082         int *len_left, *len_used;
6083         int retval, i;
6084
6085         ctsio = &io->scsiio;
6086         page_index = NULL;
6087         page_len = 0;
6088         retval = CTL_RETVAL_COMPLETE;
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         modepage_info = (union ctl_modepage_info *)
6098                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6099         len_left = &modepage_info->header.len_left;
6100         len_used = &modepage_info->header.len_used;
6101
6102 do_next_page:
6103
6104         page_header = (struct scsi_mode_page_header *)
6105                 (ctsio->kern_data_ptr + *len_used);
6106
6107         if (*len_left == 0) {
6108                 free(ctsio->kern_data_ptr, M_CTL);
6109                 ctl_set_success(ctsio);
6110                 ctl_done((union ctl_io *)ctsio);
6111                 return (CTL_RETVAL_COMPLETE);
6112         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6113
6114                 free(ctsio->kern_data_ptr, M_CTL);
6115                 ctl_set_param_len_error(ctsio);
6116                 ctl_done((union ctl_io *)ctsio);
6117                 return (CTL_RETVAL_COMPLETE);
6118
6119         } else if ((page_header->page_code & SMPH_SPF)
6120                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6121
6122                 free(ctsio->kern_data_ptr, M_CTL);
6123                 ctl_set_param_len_error(ctsio);
6124                 ctl_done((union ctl_io *)ctsio);
6125                 return (CTL_RETVAL_COMPLETE);
6126         }
6127
6128
6129         /*
6130          * XXX KDM should we do something with the block descriptor?
6131          */
6132         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6133
6134                 if ((control_dev != 0)
6135                  && (lun->mode_pages.index[i].page_flags &
6136                      CTL_PAGE_FLAG_DISK_ONLY))
6137                         continue;
6138
6139                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6140                     (page_header->page_code & SMPH_PC_MASK))
6141                         continue;
6142
6143                 /*
6144                  * If neither page has a subpage code, then we've got a
6145                  * match.
6146                  */
6147                 if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6148                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6149                         page_index = &lun->mode_pages.index[i];
6150                         page_len = page_header->page_length;
6151                         break;
6152                 }
6153
6154                 /*
6155                  * If both pages have subpages, then the subpage numbers
6156                  * have to match.
6157                  */
6158                 if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6159                   && (page_header->page_code & SMPH_SPF)) {
6160                         struct scsi_mode_page_header_sp *sph;
6161
6162                         sph = (struct scsi_mode_page_header_sp *)page_header;
6163
6164                         if (lun->mode_pages.index[i].subpage ==
6165                             sph->subpage) {
6166                                 page_index = &lun->mode_pages.index[i];
6167                                 page_len = scsi_2btoul(sph->page_length);
6168                                 break;
6169                         }
6170                 }
6171         }
6172
6173         /*
6174          * If we couldn't find the page, or if we don't have a mode select
6175          * handler for it, send back an error to the user.
6176          */
6177         if ((page_index == NULL)
6178          || (page_index->select_handler == NULL)) {
6179                 ctl_set_invalid_field(ctsio,
6180                                       /*sks_valid*/ 1,
6181                                       /*command*/ 0,
6182                                       /*field*/ *len_used,
6183                                       /*bit_valid*/ 0,
6184                                       /*bit*/ 0);
6185                 free(ctsio->kern_data_ptr, M_CTL);
6186                 ctl_done((union ctl_io *)ctsio);
6187                 return (CTL_RETVAL_COMPLETE);
6188         }
6189
6190         if (page_index->page_code & SMPH_SPF) {
6191                 page_len_offset = 2;
6192                 page_len_size = 2;
6193         } else {
6194                 page_len_size = 1;
6195                 page_len_offset = 1;
6196         }
6197
6198         /*
6199          * If the length the initiator gives us isn't the one we specify in
6200          * the mode page header, or if they didn't specify enough data in
6201          * the CDB to avoid truncating this page, kick out the request.
6202          */
6203         if ((page_len != (page_index->page_len - page_len_offset -
6204                           page_len_size))
6205          || (*len_left < page_index->page_len)) {
6206
6207
6208                 ctl_set_invalid_field(ctsio,
6209                                       /*sks_valid*/ 1,
6210                                       /*command*/ 0,
6211                                       /*field*/ *len_used + page_len_offset,
6212                                       /*bit_valid*/ 0,
6213                                       /*bit*/ 0);
6214                 free(ctsio->kern_data_ptr, M_CTL);
6215                 ctl_done((union ctl_io *)ctsio);
6216                 return (CTL_RETVAL_COMPLETE);
6217         }
6218
6219         /*
6220          * Run through the mode page, checking to make sure that the bits
6221          * the user changed are actually legal for him to change.
6222          */
6223         for (i = 0; i < page_index->page_len; i++) {
6224                 uint8_t *user_byte, *change_mask, *current_byte;
6225                 int bad_bit;
6226                 int j;
6227
6228                 user_byte = (uint8_t *)page_header + i;
6229                 change_mask = page_index->page_data +
6230                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6231                 current_byte = page_index->page_data +
6232                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6233
6234                 /*
6235                  * Check to see whether the user set any bits in this byte
6236                  * that he is not allowed to set.
6237                  */
6238                 if ((*user_byte & ~(*change_mask)) ==
6239                     (*current_byte & ~(*change_mask)))
6240                         continue;
6241
6242                 /*
6243                  * Go through bit by bit to determine which one is illegal.
6244                  */
6245                 bad_bit = 0;
6246                 for (j = 7; j >= 0; j--) {
6247                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6248                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6249                                 bad_bit = i;
6250                                 break;
6251                         }
6252                 }
6253                 ctl_set_invalid_field(ctsio,
6254                                       /*sks_valid*/ 1,
6255                                       /*command*/ 0,
6256                                       /*field*/ *len_used + i,
6257                                       /*bit_valid*/ 1,
6258                                       /*bit*/ bad_bit);
6259                 free(ctsio->kern_data_ptr, M_CTL);
6260                 ctl_done((union ctl_io *)ctsio);
6261                 return (CTL_RETVAL_COMPLETE);
6262         }
6263
6264         /*
6265          * Decrement these before we call the page handler, since we may
6266          * end up getting called back one way or another before the handler
6267          * returns to this context.
6268          */
6269         *len_left -= page_index->page_len;
6270         *len_used += page_index->page_len;
6271
6272         retval = page_index->select_handler(ctsio, page_index,
6273                                             (uint8_t *)page_header);
6274
6275         /*
6276          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6277          * wait until this queued command completes to finish processing
6278          * the mode page.  If it returns anything other than
6279          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6280          * already set the sense information, freed the data pointer, and
6281          * completed the io for us.
6282          */
6283         if (retval != CTL_RETVAL_COMPLETE)
6284                 goto bailout_no_done;
6285
6286         /*
6287          * If the initiator sent us more than one page, parse the next one.
6288          */
6289         if (*len_left > 0)
6290                 goto do_next_page;
6291
6292         ctl_set_success(ctsio);
6293         free(ctsio->kern_data_ptr, M_CTL);
6294         ctl_done((union ctl_io *)ctsio);
6295
6296 bailout_no_done:
6297
6298         return (CTL_RETVAL_COMPLETE);
6299
6300 }
6301
6302 int
6303 ctl_mode_select(struct ctl_scsiio *ctsio)
6304 {
6305         int param_len, pf, sp;
6306         int header_size, bd_len;
6307         int len_left, len_used;
6308         struct ctl_page_index *page_index;
6309         struct ctl_lun *lun;
6310         int control_dev, page_len;
6311         union ctl_modepage_info *modepage_info;
6312         int retval;
6313
6314         pf = 0;
6315         sp = 0;
6316         page_len = 0;
6317         len_used = 0;
6318         len_left = 0;
6319         retval = 0;
6320         bd_len = 0;
6321         page_index = NULL;
6322
6323         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6324
6325         if (lun->be_lun->lun_type != T_DIRECT)
6326                 control_dev = 1;
6327         else
6328                 control_dev = 0;
6329
6330         switch (ctsio->cdb[0]) {
6331         case MODE_SELECT_6: {
6332                 struct scsi_mode_select_6 *cdb;
6333
6334                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6335
6336                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6337                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6338
6339                 param_len = cdb->length;
6340                 header_size = sizeof(struct scsi_mode_header_6);
6341                 break;
6342         }
6343         case MODE_SELECT_10: {
6344                 struct scsi_mode_select_10 *cdb;
6345
6346                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6347
6348                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6349                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6350
6351                 param_len = scsi_2btoul(cdb->length);
6352                 header_size = sizeof(struct scsi_mode_header_10);
6353                 break;
6354         }
6355         default:
6356                 ctl_set_invalid_opcode(ctsio);
6357                 ctl_done((union ctl_io *)ctsio);
6358                 return (CTL_RETVAL_COMPLETE);
6359                 break; /* NOTREACHED */
6360         }
6361
6362         /*
6363          * From SPC-3:
6364          * "A parameter list length of zero indicates that the Data-Out Buffer
6365          * shall be empty. This condition shall not be considered as an error."
6366          */
6367         if (param_len == 0) {
6368                 ctl_set_success(ctsio);
6369                 ctl_done((union ctl_io *)ctsio);
6370                 return (CTL_RETVAL_COMPLETE);
6371         }
6372
6373         /*
6374          * Since we'll hit this the first time through, prior to
6375          * allocation, we don't need to free a data buffer here.
6376          */
6377         if (param_len < header_size) {
6378                 ctl_set_param_len_error(ctsio);
6379                 ctl_done((union ctl_io *)ctsio);
6380                 return (CTL_RETVAL_COMPLETE);
6381         }
6382
6383         /*
6384          * Allocate the data buffer and grab the user's data.  In theory,
6385          * we shouldn't have to sanity check the parameter list length here
6386          * because the maximum size is 64K.  We should be able to malloc
6387          * that much without too many problems.
6388          */
6389         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6390                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6391                 ctsio->kern_data_len = param_len;
6392                 ctsio->kern_total_len = param_len;
6393                 ctsio->kern_data_resid = 0;
6394                 ctsio->kern_rel_offset = 0;
6395                 ctsio->kern_sg_entries = 0;
6396                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6397                 ctsio->be_move_done = ctl_config_move_done;
6398                 ctl_datamove((union ctl_io *)ctsio);
6399
6400                 return (CTL_RETVAL_COMPLETE);
6401         }
6402
6403         switch (ctsio->cdb[0]) {
6404         case MODE_SELECT_6: {
6405                 struct scsi_mode_header_6 *mh6;
6406
6407                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6408                 bd_len = mh6->blk_desc_len;
6409                 break;
6410         }
6411         case MODE_SELECT_10: {
6412                 struct scsi_mode_header_10 *mh10;
6413
6414                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6415                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6416                 break;
6417         }
6418         default:
6419                 panic("Invalid CDB type %#x", ctsio->cdb[0]);
6420                 break;
6421         }
6422
6423         if (param_len < (header_size + bd_len)) {
6424                 free(ctsio->kern_data_ptr, M_CTL);
6425                 ctl_set_param_len_error(ctsio);
6426                 ctl_done((union ctl_io *)ctsio);
6427                 return (CTL_RETVAL_COMPLETE);
6428         }
6429
6430         /*
6431          * Set the IO_CONT flag, so that if this I/O gets passed to
6432          * ctl_config_write_done(), it'll get passed back to
6433          * ctl_do_mode_select() for further processing, or completion if
6434          * we're all done.
6435          */
6436         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6437         ctsio->io_cont = ctl_do_mode_select;
6438
6439         modepage_info = (union ctl_modepage_info *)
6440                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6441
6442         memset(modepage_info, 0, sizeof(*modepage_info));
6443
6444         len_left = param_len - header_size - bd_len;
6445         len_used = header_size + bd_len;
6446
6447         modepage_info->header.len_left = len_left;
6448         modepage_info->header.len_used = len_used;
6449
6450         return (ctl_do_mode_select((union ctl_io *)ctsio));
6451 }
6452
6453 int
6454 ctl_mode_sense(struct ctl_scsiio *ctsio)
6455 {
6456         struct ctl_lun *lun;
6457         int pc, page_code, dbd, llba, subpage;
6458         int alloc_len, page_len, header_len, total_len;
6459         struct scsi_mode_block_descr *block_desc;
6460         struct ctl_page_index *page_index;
6461         int control_dev;
6462
6463         dbd = 0;
6464         llba = 0;
6465         block_desc = NULL;
6466         page_index = NULL;
6467
6468         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6469
6470         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6471
6472         if (lun->be_lun->lun_type != T_DIRECT)
6473                 control_dev = 1;
6474         else
6475                 control_dev = 0;
6476
6477         if (lun->flags & CTL_LUN_PR_RESERVED) {
6478                 uint32_t residx;
6479
6480                 /*
6481                  * XXX KDM need a lock here.
6482                  */
6483                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6484                 if ((lun->res_type == SPR_TYPE_EX_AC
6485                   && residx != lun->pr_res_idx)
6486                  || ((lun->res_type == SPR_TYPE_EX_AC_RO
6487                    || lun->res_type == SPR_TYPE_EX_AC_AR)
6488                   && !lun->per_res[residx].registered)) {
6489                         ctl_set_reservation_conflict(ctsio);
6490                         ctl_done((union ctl_io *)ctsio);
6491                         return (CTL_RETVAL_COMPLETE);
6492                 }
6493         }
6494
6495         switch (ctsio->cdb[0]) {
6496         case MODE_SENSE_6: {
6497                 struct scsi_mode_sense_6 *cdb;
6498
6499                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6500
6501                 header_len = sizeof(struct scsi_mode_hdr_6);
6502                 if (cdb->byte2 & SMS_DBD)
6503                         dbd = 1;
6504                 else
6505                         header_len += sizeof(struct scsi_mode_block_descr);
6506
6507                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6508                 page_code = cdb->page & SMS_PAGE_CODE;
6509                 subpage = cdb->subpage;
6510                 alloc_len = cdb->length;
6511                 break;
6512         }
6513         case MODE_SENSE_10: {
6514                 struct scsi_mode_sense_10 *cdb;
6515
6516                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6517
6518                 header_len = sizeof(struct scsi_mode_hdr_10);
6519
6520                 if (cdb->byte2 & SMS_DBD)
6521                         dbd = 1;
6522                 else
6523                         header_len += sizeof(struct scsi_mode_block_descr);
6524                 if (cdb->byte2 & SMS10_LLBAA)
6525                         llba = 1;
6526                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6527                 page_code = cdb->page & SMS_PAGE_CODE;
6528                 subpage = cdb->subpage;
6529                 alloc_len = scsi_2btoul(cdb->length);
6530                 break;
6531         }
6532         default:
6533                 ctl_set_invalid_opcode(ctsio);
6534                 ctl_done((union ctl_io *)ctsio);
6535                 return (CTL_RETVAL_COMPLETE);
6536                 break; /* NOTREACHED */
6537         }
6538
6539         /*
6540          * We have to make a first pass through to calculate the size of
6541          * the pages that match the user's query.  Then we allocate enough
6542          * memory to hold it, and actually copy the data into the buffer.
6543          */
6544         switch (page_code) {
6545         case SMS_ALL_PAGES_PAGE: {
6546                 int i;
6547
6548                 page_len = 0;
6549
6550                 /*
6551                  * At the moment, values other than 0 and 0xff here are
6552                  * reserved according to SPC-3.
6553                  */
6554                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6555                  && (subpage != SMS_SUBPAGE_ALL)) {
6556                         ctl_set_invalid_field(ctsio,
6557                                               /*sks_valid*/ 1,
6558                                               /*command*/ 1,
6559                                               /*field*/ 3,
6560                                               /*bit_valid*/ 0,
6561                                               /*bit*/ 0);
6562                         ctl_done((union ctl_io *)ctsio);
6563                         return (CTL_RETVAL_COMPLETE);
6564                 }
6565
6566                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6567                         if ((control_dev != 0)
6568                          && (lun->mode_pages.index[i].page_flags &
6569                              CTL_PAGE_FLAG_DISK_ONLY))
6570                                 continue;
6571
6572                         /*
6573                          * We don't use this subpage if the user didn't
6574                          * request all subpages.
6575                          */
6576                         if ((lun->mode_pages.index[i].subpage != 0)
6577                          && (subpage == SMS_SUBPAGE_PAGE_0))
6578                                 continue;
6579
6580 #if 0
6581                         printf("found page %#x len %d\n",
6582                                lun->mode_pages.index[i].page_code &
6583                                SMPH_PC_MASK,
6584                                lun->mode_pages.index[i].page_len);
6585 #endif
6586                         page_len += lun->mode_pages.index[i].page_len;
6587                 }
6588                 break;
6589         }
6590         default: {
6591                 int i;
6592
6593                 page_len = 0;
6594
6595                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6596                         /* Look for the right page code */
6597                         if ((lun->mode_pages.index[i].page_code &
6598                              SMPH_PC_MASK) != page_code)
6599                                 continue;
6600
6601                         /* Look for the right subpage or the subpage wildcard*/
6602                         if ((lun->mode_pages.index[i].subpage != subpage)
6603                          && (subpage != SMS_SUBPAGE_ALL))
6604                                 continue;
6605
6606                         /* Make sure the page is supported for this dev type */
6607                         if ((control_dev != 0)
6608                          && (lun->mode_pages.index[i].page_flags &
6609                              CTL_PAGE_FLAG_DISK_ONLY))
6610                                 continue;
6611
6612 #if 0
6613                         printf("found page %#x len %d\n",
6614                                lun->mode_pages.index[i].page_code &
6615                                SMPH_PC_MASK,
6616                                lun->mode_pages.index[i].page_len);
6617 #endif
6618
6619                         page_len += lun->mode_pages.index[i].page_len;
6620                 }
6621
6622                 if (page_len == 0) {
6623                         ctl_set_invalid_field(ctsio,
6624                                               /*sks_valid*/ 1,
6625                                               /*command*/ 1,
6626                                               /*field*/ 2,
6627                                               /*bit_valid*/ 1,
6628                                               /*bit*/ 5);
6629                         ctl_done((union ctl_io *)ctsio);
6630                         return (CTL_RETVAL_COMPLETE);
6631                 }
6632                 break;
6633         }
6634         }
6635
6636         total_len = header_len + page_len;
6637 #if 0
6638         printf("header_len = %d, page_len = %d, total_len = %d\n",
6639                header_len, page_len, total_len);
6640 #endif
6641
6642         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6643         ctsio->kern_sg_entries = 0;
6644         ctsio->kern_data_resid = 0;
6645         ctsio->kern_rel_offset = 0;
6646         if (total_len < alloc_len) {
6647                 ctsio->residual = alloc_len - total_len;
6648                 ctsio->kern_data_len = total_len;
6649                 ctsio->kern_total_len = total_len;
6650         } else {
6651                 ctsio->residual = 0;
6652                 ctsio->kern_data_len = alloc_len;
6653                 ctsio->kern_total_len = alloc_len;
6654         }
6655
6656         switch (ctsio->cdb[0]) {
6657         case MODE_SENSE_6: {
6658                 struct scsi_mode_hdr_6 *header;
6659
6660                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6661
6662                 header->datalen = ctl_min(total_len - 1, 254);
6663
6664                 if (dbd)
6665                         header->block_descr_len = 0;
6666                 else
6667                         header->block_descr_len =
6668                                 sizeof(struct scsi_mode_block_descr);
6669                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6670                 break;
6671         }
6672         case MODE_SENSE_10: {
6673                 struct scsi_mode_hdr_10 *header;
6674                 int datalen;
6675
6676                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6677
6678                 datalen = ctl_min(total_len - 2, 65533);
6679                 scsi_ulto2b(datalen, header->datalen);
6680                 if (dbd)
6681                         scsi_ulto2b(0, header->block_descr_len);
6682                 else
6683                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6684                                     header->block_descr_len);
6685                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6686                 break;
6687         }
6688         default:
6689                 panic("invalid CDB type %#x", ctsio->cdb[0]);
6690                 break; /* NOTREACHED */
6691         }
6692
6693         /*
6694          * If we've got a disk, use its blocksize in the block
6695          * descriptor.  Otherwise, just set it to 0.
6696          */
6697         if (dbd == 0) {
6698                 if (control_dev != 0)
6699                         scsi_ulto3b(lun->be_lun->blocksize,
6700                                     block_desc->block_len);
6701                 else
6702                         scsi_ulto3b(0, block_desc->block_len);
6703         }
6704
6705         switch (page_code) {
6706         case SMS_ALL_PAGES_PAGE: {
6707                 int i, data_used;
6708
6709                 data_used = header_len;
6710                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6711                         struct ctl_page_index *page_index;
6712
6713                         page_index = &lun->mode_pages.index[i];
6714
6715                         if ((control_dev != 0)
6716                          && (page_index->page_flags &
6717                             CTL_PAGE_FLAG_DISK_ONLY))
6718                                 continue;
6719
6720                         /*
6721                          * We don't use this subpage if the user didn't
6722                          * request all subpages.  We already checked (above)
6723                          * to make sure the user only specified a subpage
6724                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6725                          */
6726                         if ((page_index->subpage != 0)
6727                          && (subpage == SMS_SUBPAGE_PAGE_0))
6728                                 continue;
6729
6730                         /*
6731                          * Call the handler, if it exists, to update the
6732                          * page to the latest values.
6733                          */
6734                         if (page_index->sense_handler != NULL)
6735                                 page_index->sense_handler(ctsio, page_index,pc);
6736
6737                         memcpy(ctsio->kern_data_ptr + data_used,
6738                                page_index->page_data +
6739                                (page_index->page_len * pc),
6740                                page_index->page_len);
6741                         data_used += page_index->page_len;
6742                 }
6743                 break;
6744         }
6745         default: {
6746                 int i, data_used;
6747
6748                 data_used = header_len;
6749
6750                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6751                         struct ctl_page_index *page_index;
6752
6753                         page_index = &lun->mode_pages.index[i];
6754
6755                         /* Look for the right page code */
6756                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6757                                 continue;
6758
6759                         /* Look for the right subpage or the subpage wildcard*/
6760                         if ((page_index->subpage != subpage)
6761                          && (subpage != SMS_SUBPAGE_ALL))
6762                                 continue;
6763
6764                         /* Make sure the page is supported for this dev type */
6765                         if ((control_dev != 0)
6766                          && (page_index->page_flags &
6767                              CTL_PAGE_FLAG_DISK_ONLY))
6768                                 continue;
6769
6770                         /*
6771                          * Call the handler, if it exists, to update the
6772                          * page to the latest values.
6773                          */
6774                         if (page_index->sense_handler != NULL)
6775                                 page_index->sense_handler(ctsio, page_index,pc);
6776
6777                         memcpy(ctsio->kern_data_ptr + data_used,
6778                                page_index->page_data +
6779                                (page_index->page_len * pc),
6780                                page_index->page_len);
6781                         data_used += page_index->page_len;
6782                 }
6783                 break;
6784         }
6785         }
6786
6787         ctsio->scsi_status = SCSI_STATUS_OK;
6788
6789         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6790         ctsio->be_move_done = ctl_config_move_done;
6791         ctl_datamove((union ctl_io *)ctsio);
6792
6793         return (CTL_RETVAL_COMPLETE);
6794 }
6795
6796 int
6797 ctl_read_capacity(struct ctl_scsiio *ctsio)
6798 {
6799         struct scsi_read_capacity *cdb;
6800         struct scsi_read_capacity_data *data;
6801         struct ctl_lun *lun;
6802         uint32_t lba;
6803
6804         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6805
6806         cdb = (struct scsi_read_capacity *)ctsio->cdb;
6807
6808         lba = scsi_4btoul(cdb->addr);
6809         if (((cdb->pmi & SRC_PMI) == 0)
6810          && (lba != 0)) {
6811                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6812                                       /*sks_valid*/ 1,
6813                                       /*command*/ 1,
6814                                       /*field*/ 2,
6815                                       /*bit_valid*/ 0,
6816                                       /*bit*/ 0);
6817                 ctl_done((union ctl_io *)ctsio);
6818                 return (CTL_RETVAL_COMPLETE);
6819         }
6820
6821         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6822
6823         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6824         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6825         ctsio->residual = 0;
6826         ctsio->kern_data_len = sizeof(*data);
6827         ctsio->kern_total_len = sizeof(*data);
6828         ctsio->kern_data_resid = 0;
6829         ctsio->kern_rel_offset = 0;
6830         ctsio->kern_sg_entries = 0;
6831
6832         /*
6833          * If the maximum LBA is greater than 0xfffffffe, the user must
6834          * issue a SERVICE ACTION IN (16) command, with the read capacity
6835          * serivce action set.
6836          */
6837         if (lun->be_lun->maxlba > 0xfffffffe)
6838                 scsi_ulto4b(0xffffffff, data->addr);
6839         else
6840                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6841
6842         /*
6843          * XXX KDM this may not be 512 bytes...
6844          */
6845         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6846
6847         ctsio->scsi_status = SCSI_STATUS_OK;
6848
6849         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6850         ctsio->be_move_done = ctl_config_move_done;
6851         ctl_datamove((union ctl_io *)ctsio);
6852
6853         return (CTL_RETVAL_COMPLETE);
6854 }
6855
6856 int
6857 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6858 {
6859         struct scsi_read_capacity_16 *cdb;
6860         struct scsi_read_capacity_data_long *data;
6861         struct ctl_lun *lun;
6862         uint64_t lba;
6863         uint32_t alloc_len;
6864
6865         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6866
6867         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6868
6869         alloc_len = scsi_4btoul(cdb->alloc_len);
6870         lba = scsi_8btou64(cdb->addr);
6871
6872         if ((cdb->reladr & SRC16_PMI)
6873          && (lba != 0)) {
6874                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6875                                       /*sks_valid*/ 1,
6876                                       /*command*/ 1,
6877                                       /*field*/ 2,
6878                                       /*bit_valid*/ 0,
6879                                       /*bit*/ 0);
6880                 ctl_done((union ctl_io *)ctsio);
6881                 return (CTL_RETVAL_COMPLETE);
6882         }
6883
6884         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6885
6886         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6887         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6888
6889         if (sizeof(*data) < alloc_len) {
6890                 ctsio->residual = alloc_len - sizeof(*data);
6891                 ctsio->kern_data_len = sizeof(*data);
6892                 ctsio->kern_total_len = sizeof(*data);
6893         } else {
6894                 ctsio->residual = 0;
6895                 ctsio->kern_data_len = alloc_len;
6896                 ctsio->kern_total_len = alloc_len;
6897         }
6898         ctsio->kern_data_resid = 0;
6899         ctsio->kern_rel_offset = 0;
6900         ctsio->kern_sg_entries = 0;
6901
6902         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6903         /* XXX KDM this may not be 512 bytes... */
6904         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6905         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
6906         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
6907         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
6908                 data->lalba_lbp[0] |= SRC16_LBPME;
6909
6910         ctsio->scsi_status = SCSI_STATUS_OK;
6911
6912         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6913         ctsio->be_move_done = ctl_config_move_done;
6914         ctl_datamove((union ctl_io *)ctsio);
6915
6916         return (CTL_RETVAL_COMPLETE);
6917 }
6918
6919 int
6920 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
6921 {
6922         struct scsi_maintenance_in *cdb;
6923         int retval;
6924         int alloc_len, total_len = 0;
6925         int num_target_port_groups, single;
6926         struct ctl_lun *lun;
6927         struct ctl_softc *softc;
6928         struct scsi_target_group_data *rtg_ptr;
6929         struct scsi_target_port_group_descriptor *tpg_desc_ptr1, *tpg_desc_ptr2;
6930         struct scsi_target_port_descriptor  *tp_desc_ptr1_1, *tp_desc_ptr1_2,
6931                                             *tp_desc_ptr2_1, *tp_desc_ptr2_2;
6932
6933         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
6934
6935         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
6936         softc = control_softc;
6937         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6938
6939         retval = CTL_RETVAL_COMPLETE;
6940
6941         single = ctl_is_single;
6942         if (single)
6943                 num_target_port_groups = NUM_TARGET_PORT_GROUPS - 1;
6944         else
6945                 num_target_port_groups = NUM_TARGET_PORT_GROUPS;
6946
6947         total_len = sizeof(struct scsi_target_group_data) +
6948                 sizeof(struct scsi_target_port_group_descriptor) *
6949                 num_target_port_groups +
6950                 sizeof(struct scsi_target_port_descriptor) *
6951                 NUM_PORTS_PER_GRP * num_target_port_groups;
6952
6953         alloc_len = scsi_4btoul(cdb->length);
6954
6955         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6956
6957         ctsio->kern_sg_entries = 0;
6958
6959         if (total_len < alloc_len) {
6960                 ctsio->residual = alloc_len - total_len;
6961                 ctsio->kern_data_len = total_len;
6962                 ctsio->kern_total_len = total_len;
6963         } else {
6964                 ctsio->residual = 0;
6965                 ctsio->kern_data_len = alloc_len;
6966                 ctsio->kern_total_len = alloc_len;
6967         }
6968         ctsio->kern_data_resid = 0;
6969         ctsio->kern_rel_offset = 0;
6970
6971         rtg_ptr = (struct scsi_target_group_data *)ctsio->kern_data_ptr;
6972
6973         tpg_desc_ptr1 = &rtg_ptr->groups[0];
6974         tp_desc_ptr1_1 = &tpg_desc_ptr1->descriptors[0];
6975         tp_desc_ptr1_2 = (struct scsi_target_port_descriptor *)
6976                 &tp_desc_ptr1_1->desc_list[0];
6977
6978         if (single == 0) {
6979                 tpg_desc_ptr2 = (struct scsi_target_port_group_descriptor *)
6980                         &tp_desc_ptr1_2->desc_list[0];
6981                 tp_desc_ptr2_1 = &tpg_desc_ptr2->descriptors[0];
6982                 tp_desc_ptr2_2 = (struct scsi_target_port_descriptor *)
6983                         &tp_desc_ptr2_1->desc_list[0];
6984         } else {
6985                 tpg_desc_ptr2 = NULL;
6986                 tp_desc_ptr2_1 = NULL;
6987                 tp_desc_ptr2_2 = NULL;
6988         }
6989
6990         scsi_ulto4b(total_len - 4, rtg_ptr->length);
6991         if (single == 0) {
6992                 if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
6993                         if (lun->flags & CTL_LUN_PRIMARY_SC) {
6994                                 tpg_desc_ptr1->pref_state = TPG_PRIMARY;
6995                                 tpg_desc_ptr2->pref_state =
6996                                         TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
6997                         } else {
6998                                 tpg_desc_ptr1->pref_state =
6999                                         TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7000                                 tpg_desc_ptr2->pref_state = TPG_PRIMARY;
7001                         }
7002                 } else {
7003                         if (lun->flags & CTL_LUN_PRIMARY_SC) {
7004                                 tpg_desc_ptr1->pref_state =
7005                                         TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7006                                 tpg_desc_ptr2->pref_state = TPG_PRIMARY;
7007                         } else {
7008                                 tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7009                                 tpg_desc_ptr2->pref_state =
7010                                         TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7011                         }
7012                 }
7013         } else {
7014                 tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7015         }
7016         tpg_desc_ptr1->support = 0;
7017         tpg_desc_ptr1->target_port_group[1] = 1;
7018         tpg_desc_ptr1->status = TPG_IMPLICIT;
7019         tpg_desc_ptr1->target_port_count= NUM_PORTS_PER_GRP;
7020
7021         if (single == 0) {
7022                 tpg_desc_ptr2->support = 0;
7023                 tpg_desc_ptr2->target_port_group[1] = 2;
7024                 tpg_desc_ptr2->status = TPG_IMPLICIT;
7025                 tpg_desc_ptr2->target_port_count = NUM_PORTS_PER_GRP;
7026
7027                 tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7028                 tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7029
7030                 tp_desc_ptr2_1->relative_target_port_identifier[1] = 9;
7031                 tp_desc_ptr2_2->relative_target_port_identifier[1] = 10;
7032         } else {
7033                 if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
7034                         tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7035                         tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7036                 } else {
7037                         tp_desc_ptr1_1->relative_target_port_identifier[1] = 9;
7038                         tp_desc_ptr1_2->relative_target_port_identifier[1] = 10;
7039                 }
7040         }
7041
7042         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7043         ctsio->be_move_done = ctl_config_move_done;
7044
7045         CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7046                          ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7047                          ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7048                          ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7049                          ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7050
7051         ctl_datamove((union ctl_io *)ctsio);
7052         return(retval);
7053 }
7054
7055 int
7056 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7057 {
7058         struct ctl_lun *lun;
7059         struct scsi_report_supported_opcodes *cdb;
7060         const struct ctl_cmd_entry *entry, *sentry;
7061         struct scsi_report_supported_opcodes_all *all;
7062         struct scsi_report_supported_opcodes_descr *descr;
7063         struct scsi_report_supported_opcodes_one *one;
7064         int retval;
7065         int alloc_len, total_len;
7066         int opcode, service_action, i, j, num;
7067
7068         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7069
7070         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7071         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7072
7073         retval = CTL_RETVAL_COMPLETE;
7074
7075         opcode = cdb->requested_opcode;
7076         service_action = scsi_2btoul(cdb->requested_service_action);
7077         switch (cdb->options & RSO_OPTIONS_MASK) {
7078         case RSO_OPTIONS_ALL:
7079                 num = 0;
7080                 for (i = 0; i < 256; i++) {
7081                         entry = &ctl_cmd_table[i];
7082                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7083                                 for (j = 0; j < 32; j++) {
7084                                         sentry = &((const struct ctl_cmd_entry *)
7085                                             entry->execute)[j];
7086                                         if (ctl_cmd_applicable(
7087                                             lun->be_lun->lun_type, sentry))
7088                                                 num++;
7089                                 }
7090                         } else {
7091                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7092                                     entry))
7093                                         num++;
7094                         }
7095                 }
7096                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7097                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7098                 break;
7099         case RSO_OPTIONS_OC:
7100                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7101                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7102                                               /*sks_valid*/ 1,
7103                                               /*command*/ 1,
7104                                               /*field*/ 2,
7105                                               /*bit_valid*/ 1,
7106                                               /*bit*/ 2);
7107                         ctl_done((union ctl_io *)ctsio);
7108                         return (CTL_RETVAL_COMPLETE);
7109                 }
7110                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7111                 break;
7112         case RSO_OPTIONS_OC_SA:
7113                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7114                     service_action >= 32) {
7115                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7116                                               /*sks_valid*/ 1,
7117                                               /*command*/ 1,
7118                                               /*field*/ 2,
7119                                               /*bit_valid*/ 1,
7120                                               /*bit*/ 2);
7121                         ctl_done((union ctl_io *)ctsio);
7122                         return (CTL_RETVAL_COMPLETE);
7123                 }
7124                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7125                 break;
7126         default:
7127                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7128                                       /*sks_valid*/ 1,
7129                                       /*command*/ 1,
7130                                       /*field*/ 2,
7131                                       /*bit_valid*/ 1,
7132                                       /*bit*/ 2);
7133                 ctl_done((union ctl_io *)ctsio);
7134                 return (CTL_RETVAL_COMPLETE);
7135         }
7136
7137         alloc_len = scsi_4btoul(cdb->length);
7138
7139         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7140
7141         ctsio->kern_sg_entries = 0;
7142
7143         if (total_len < alloc_len) {
7144                 ctsio->residual = alloc_len - total_len;
7145                 ctsio->kern_data_len = total_len;
7146                 ctsio->kern_total_len = total_len;
7147         } else {
7148                 ctsio->residual = 0;
7149                 ctsio->kern_data_len = alloc_len;
7150                 ctsio->kern_total_len = alloc_len;
7151         }
7152         ctsio->kern_data_resid = 0;
7153         ctsio->kern_rel_offset = 0;
7154
7155         switch (cdb->options & RSO_OPTIONS_MASK) {
7156         case RSO_OPTIONS_ALL:
7157                 all = (struct scsi_report_supported_opcodes_all *)
7158                     ctsio->kern_data_ptr;
7159                 num = 0;
7160                 for (i = 0; i < 256; i++) {
7161                         entry = &ctl_cmd_table[i];
7162                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7163                                 for (j = 0; j < 32; j++) {
7164                                         sentry = &((const struct ctl_cmd_entry *)
7165                                             entry->execute)[j];
7166                                         if (!ctl_cmd_applicable(
7167                                             lun->be_lun->lun_type, sentry))
7168                                                 continue;
7169                                         descr = &all->descr[num++];
7170                                         descr->opcode = i;
7171                                         scsi_ulto2b(j, descr->service_action);
7172                                         descr->flags = RSO_SERVACTV;
7173                                         scsi_ulto2b(sentry->length,
7174                                             descr->cdb_length);
7175                                 }
7176                         } else {
7177                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7178                                     entry))
7179                                         continue;
7180                                 descr = &all->descr[num++];
7181                                 descr->opcode = i;
7182                                 scsi_ulto2b(0, descr->service_action);
7183                                 descr->flags = 0;
7184                                 scsi_ulto2b(entry->length, descr->cdb_length);
7185                         }
7186                 }
7187                 scsi_ulto4b(
7188                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7189                     all->length);
7190                 break;
7191         case RSO_OPTIONS_OC:
7192                 one = (struct scsi_report_supported_opcodes_one *)
7193                     ctsio->kern_data_ptr;
7194                 entry = &ctl_cmd_table[opcode];
7195                 goto fill_one;
7196         case RSO_OPTIONS_OC_SA:
7197                 one = (struct scsi_report_supported_opcodes_one *)
7198                     ctsio->kern_data_ptr;
7199                 entry = &ctl_cmd_table[opcode];
7200                 entry = &((const struct ctl_cmd_entry *)
7201                     entry->execute)[service_action];
7202 fill_one:
7203                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7204                         one->support = 3;
7205                         scsi_ulto2b(entry->length, one->cdb_length);
7206                         one->cdb_usage[0] = opcode;
7207                         memcpy(&one->cdb_usage[1], entry->usage,
7208                             entry->length - 1);
7209                 } else
7210                         one->support = 1;
7211                 break;
7212         }
7213
7214         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7215         ctsio->be_move_done = ctl_config_move_done;
7216
7217         ctl_datamove((union ctl_io *)ctsio);
7218         return(retval);
7219 }
7220
7221 int
7222 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7223 {
7224         struct ctl_lun *lun;
7225         struct scsi_report_supported_tmf *cdb;
7226         struct scsi_report_supported_tmf_data *data;
7227         int retval;
7228         int alloc_len, total_len;
7229
7230         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7231
7232         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7233         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7234
7235         retval = CTL_RETVAL_COMPLETE;
7236
7237         total_len = sizeof(struct scsi_report_supported_tmf_data);
7238         alloc_len = scsi_4btoul(cdb->length);
7239
7240         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7241
7242         ctsio->kern_sg_entries = 0;
7243
7244         if (total_len < alloc_len) {
7245                 ctsio->residual = alloc_len - total_len;
7246                 ctsio->kern_data_len = total_len;
7247                 ctsio->kern_total_len = total_len;
7248         } else {
7249                 ctsio->residual = 0;
7250                 ctsio->kern_data_len = alloc_len;
7251                 ctsio->kern_total_len = alloc_len;
7252         }
7253         ctsio->kern_data_resid = 0;
7254         ctsio->kern_rel_offset = 0;
7255
7256         data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7257         data->byte1 |= RST_ATS | RST_LURS | RST_TRS;
7258
7259         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7260         ctsio->be_move_done = ctl_config_move_done;
7261
7262         ctl_datamove((union ctl_io *)ctsio);
7263         return (retval);
7264 }
7265
7266 int
7267 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7268 {
7269         struct ctl_lun *lun;
7270         struct scsi_report_timestamp *cdb;
7271         struct scsi_report_timestamp_data *data;
7272         struct timeval tv;
7273         int64_t timestamp;
7274         int retval;
7275         int alloc_len, total_len;
7276
7277         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7278
7279         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7280         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7281
7282         retval = CTL_RETVAL_COMPLETE;
7283
7284         total_len = sizeof(struct scsi_report_timestamp_data);
7285         alloc_len = scsi_4btoul(cdb->length);
7286
7287         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7288
7289         ctsio->kern_sg_entries = 0;
7290
7291         if (total_len < alloc_len) {
7292                 ctsio->residual = alloc_len - total_len;
7293                 ctsio->kern_data_len = total_len;
7294                 ctsio->kern_total_len = total_len;
7295         } else {
7296                 ctsio->residual = 0;
7297                 ctsio->kern_data_len = alloc_len;
7298                 ctsio->kern_total_len = alloc_len;
7299         }
7300         ctsio->kern_data_resid = 0;
7301         ctsio->kern_rel_offset = 0;
7302
7303         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7304         scsi_ulto2b(sizeof(*data) - 2, data->length);
7305         data->origin = RTS_ORIG_OUTSIDE;
7306         getmicrotime(&tv);
7307         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7308         scsi_ulto4b(timestamp >> 16, data->timestamp);
7309         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7310
7311         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7312         ctsio->be_move_done = ctl_config_move_done;
7313
7314         ctl_datamove((union ctl_io *)ctsio);
7315         return (retval);
7316 }
7317
7318 int
7319 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7320 {
7321         struct scsi_per_res_in *cdb;
7322         int alloc_len, total_len = 0;
7323         /* struct scsi_per_res_in_rsrv in_data; */
7324         struct ctl_lun *lun;
7325         struct ctl_softc *softc;
7326
7327         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7328
7329         softc = control_softc;
7330
7331         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7332
7333         alloc_len = scsi_2btoul(cdb->length);
7334
7335         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7336
7337 retry:
7338         mtx_lock(&lun->lun_lock);
7339         switch (cdb->action) {
7340         case SPRI_RK: /* read keys */
7341                 total_len = sizeof(struct scsi_per_res_in_keys) +
7342                         lun->pr_key_count *
7343                         sizeof(struct scsi_per_res_key);
7344                 break;
7345         case SPRI_RR: /* read reservation */
7346                 if (lun->flags & CTL_LUN_PR_RESERVED)
7347                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7348                 else
7349                         total_len = sizeof(struct scsi_per_res_in_header);
7350                 break;
7351         case SPRI_RC: /* report capabilities */
7352                 total_len = sizeof(struct scsi_per_res_cap);
7353                 break;
7354         default:
7355                 panic("Invalid PR type %x", cdb->action);
7356         }
7357         mtx_unlock(&lun->lun_lock);
7358
7359         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7360
7361         if (total_len < alloc_len) {
7362                 ctsio->residual = alloc_len - total_len;
7363                 ctsio->kern_data_len = total_len;
7364                 ctsio->kern_total_len = total_len;
7365         } else {
7366                 ctsio->residual = 0;
7367                 ctsio->kern_data_len = alloc_len;
7368                 ctsio->kern_total_len = alloc_len;
7369         }
7370
7371         ctsio->kern_data_resid = 0;
7372         ctsio->kern_rel_offset = 0;
7373         ctsio->kern_sg_entries = 0;
7374
7375         mtx_lock(&lun->lun_lock);
7376         switch (cdb->action) {
7377         case SPRI_RK: { // read keys
7378         struct scsi_per_res_in_keys *res_keys;
7379                 int i, key_count;
7380
7381                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7382
7383                 /*
7384                  * We had to drop the lock to allocate our buffer, which
7385                  * leaves time for someone to come in with another
7386                  * persistent reservation.  (That is unlikely, though,
7387                  * since this should be the only persistent reservation
7388                  * command active right now.)
7389                  */
7390                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7391                     (lun->pr_key_count *
7392                      sizeof(struct scsi_per_res_key)))){
7393                         mtx_unlock(&lun->lun_lock);
7394                         free(ctsio->kern_data_ptr, M_CTL);
7395                         printf("%s: reservation length changed, retrying\n",
7396                                __func__);
7397                         goto retry;
7398                 }
7399
7400                 scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7401
7402                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7403                              lun->pr_key_count, res_keys->header.length);
7404
7405                 for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7406                         if (!lun->per_res[i].registered)
7407                                 continue;
7408
7409                         /*
7410                          * We used lun->pr_key_count to calculate the
7411                          * size to allocate.  If it turns out the number of
7412                          * initiators with the registered flag set is
7413                          * larger than that (i.e. they haven't been kept in
7414                          * sync), we've got a problem.
7415                          */
7416                         if (key_count >= lun->pr_key_count) {
7417 #ifdef NEEDTOPORT
7418                                 csevent_log(CSC_CTL | CSC_SHELF_SW |
7419                                             CTL_PR_ERROR,
7420                                             csevent_LogType_Fault,
7421                                             csevent_AlertLevel_Yellow,
7422                                             csevent_FRU_ShelfController,
7423                                             csevent_FRU_Firmware,
7424                                         csevent_FRU_Unknown,
7425                                             "registered keys %d >= key "
7426                                             "count %d", key_count,
7427                                             lun->pr_key_count);
7428 #endif
7429                                 key_count++;
7430                                 continue;
7431                         }
7432                         memcpy(res_keys->keys[key_count].key,
7433                                lun->per_res[i].res_key.key,
7434                                ctl_min(sizeof(res_keys->keys[key_count].key),
7435                                sizeof(lun->per_res[i].res_key)));
7436                         key_count++;
7437                 }
7438                 break;
7439         }
7440         case SPRI_RR: { // read reservation
7441                 struct scsi_per_res_in_rsrv *res;
7442                 int tmp_len, header_only;
7443
7444                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7445
7446                 scsi_ulto4b(lun->PRGeneration, res->header.generation);
7447
7448                 if (lun->flags & CTL_LUN_PR_RESERVED)
7449                 {
7450                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7451                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7452                                     res->header.length);
7453                         header_only = 0;
7454                 } else {
7455                         tmp_len = sizeof(struct scsi_per_res_in_header);
7456                         scsi_ulto4b(0, res->header.length);
7457                         header_only = 1;
7458                 }
7459
7460                 /*
7461                  * We had to drop the lock to allocate our buffer, which
7462                  * leaves time for someone to come in with another
7463                  * persistent reservation.  (That is unlikely, though,
7464                  * since this should be the only persistent reservation
7465                  * command active right now.)
7466                  */
7467                 if (tmp_len != total_len) {
7468                         mtx_unlock(&lun->lun_lock);
7469                         free(ctsio->kern_data_ptr, M_CTL);
7470                         printf("%s: reservation status changed, retrying\n",
7471                                __func__);
7472                         goto retry;
7473                 }
7474
7475                 /*
7476                  * No reservation held, so we're done.
7477                  */
7478                 if (header_only != 0)
7479                         break;
7480
7481                 /*
7482                  * If the registration is an All Registrants type, the key
7483                  * is 0, since it doesn't really matter.
7484                  */
7485                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7486                         memcpy(res->data.reservation,
7487                                &lun->per_res[lun->pr_res_idx].res_key,
7488                                sizeof(struct scsi_per_res_key));
7489                 }
7490                 res->data.scopetype = lun->res_type;
7491                 break;
7492         }
7493         case SPRI_RC:     //report capabilities
7494         {
7495                 struct scsi_per_res_cap *res_cap;
7496                 uint16_t type_mask;
7497
7498                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7499                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7500                 res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_3;
7501                 type_mask = SPRI_TM_WR_EX_AR |
7502                             SPRI_TM_EX_AC_RO |
7503                             SPRI_TM_WR_EX_RO |
7504                             SPRI_TM_EX_AC |
7505                             SPRI_TM_WR_EX |
7506                             SPRI_TM_EX_AC_AR;
7507                 scsi_ulto2b(type_mask, res_cap->type_mask);
7508                 break;
7509         }
7510         case SPRI_RS: //read full status
7511         default:
7512                 /*
7513                  * This is a bug, because we just checked for this above,
7514                  * and should have returned an error.
7515                  */
7516                 panic("Invalid PR type %x", cdb->action);
7517                 break; /* NOTREACHED */
7518         }
7519         mtx_unlock(&lun->lun_lock);
7520
7521         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7522         ctsio->be_move_done = ctl_config_move_done;
7523
7524         CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7525                          ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7526                          ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7527                          ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7528                          ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7529
7530         ctl_datamove((union ctl_io *)ctsio);
7531
7532         return (CTL_RETVAL_COMPLETE);
7533 }
7534
7535 /*
7536  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7537  * it should return.
7538  */
7539 static int
7540 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7541                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7542                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7543                 struct scsi_per_res_out_parms* param)
7544 {
7545         union ctl_ha_msg persis_io;
7546         int retval, i;
7547         int isc_retval;
7548
7549         retval = 0;
7550
7551         mtx_lock(&lun->lun_lock);
7552         if (sa_res_key == 0) {
7553                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7554                         /* validate scope and type */
7555                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7556                              SPR_LU_SCOPE) {
7557                                 mtx_unlock(&lun->lun_lock);
7558                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7559                                                       /*sks_valid*/ 1,
7560                                                       /*command*/ 1,
7561                                                       /*field*/ 2,
7562                                                       /*bit_valid*/ 1,
7563                                                       /*bit*/ 4);
7564                                 ctl_done((union ctl_io *)ctsio);
7565                                 return (1);
7566                         }
7567
7568                         if (type>8 || type==2 || type==4 || type==0) {
7569                                 mtx_unlock(&lun->lun_lock);
7570                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7571                                                       /*sks_valid*/ 1,
7572                                                       /*command*/ 1,
7573                                                       /*field*/ 2,
7574                                                       /*bit_valid*/ 1,
7575                                                       /*bit*/ 0);
7576                                 ctl_done((union ctl_io *)ctsio);
7577                                 return (1);
7578                         }
7579
7580                         /* temporarily unregister this nexus */
7581                         lun->per_res[residx].registered = 0;
7582
7583                         /*
7584                          * Unregister everybody else and build UA for
7585                          * them
7586                          */
7587                         for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7588                                 if (lun->per_res[i].registered == 0)
7589                                         continue;
7590
7591                                 if (!persis_offset
7592                                  && i <CTL_MAX_INITIATORS)
7593                                         lun->pending_sense[i].ua_pending |=
7594                                                 CTL_UA_REG_PREEMPT;
7595                                 else if (persis_offset
7596                                       && i >= persis_offset)
7597                                         lun->pending_sense[i-persis_offset
7598                                                 ].ua_pending |=
7599                                                 CTL_UA_REG_PREEMPT;
7600                                 lun->per_res[i].registered = 0;
7601                                 memset(&lun->per_res[i].res_key, 0,
7602                                        sizeof(struct scsi_per_res_key));
7603                         }
7604                         lun->per_res[residx].registered = 1;
7605                         lun->pr_key_count = 1;
7606                         lun->res_type = type;
7607                         if (lun->res_type != SPR_TYPE_WR_EX_AR
7608                          && lun->res_type != SPR_TYPE_EX_AC_AR)
7609                                 lun->pr_res_idx = residx;
7610
7611                         /* send msg to other side */
7612                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7613                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7614                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7615                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7616                         persis_io.pr.pr_info.res_type = type;
7617                         memcpy(persis_io.pr.pr_info.sa_res_key,
7618                                param->serv_act_res_key,
7619                                sizeof(param->serv_act_res_key));
7620                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7621                              &persis_io, sizeof(persis_io), 0)) >
7622                              CTL_HA_STATUS_SUCCESS) {
7623                                 printf("CTL:Persis Out error returned "
7624                                        "from ctl_ha_msg_send %d\n",
7625                                        isc_retval);
7626                         }
7627                 } else {
7628                         /* not all registrants */
7629                         mtx_unlock(&lun->lun_lock);
7630                         free(ctsio->kern_data_ptr, M_CTL);
7631                         ctl_set_invalid_field(ctsio,
7632                                               /*sks_valid*/ 1,
7633                                               /*command*/ 0,
7634                                               /*field*/ 8,
7635                                               /*bit_valid*/ 0,
7636                                               /*bit*/ 0);
7637                         ctl_done((union ctl_io *)ctsio);
7638                         return (1);
7639                 }
7640         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7641                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
7642                 int found = 0;
7643
7644                 if (res_key == sa_res_key) {
7645                         /* special case */
7646                         /*
7647                          * The spec implies this is not good but doesn't
7648                          * say what to do. There are two choices either
7649                          * generate a res conflict or check condition
7650                          * with illegal field in parameter data. Since
7651                          * that is what is done when the sa_res_key is
7652                          * zero I'll take that approach since this has
7653                          * to do with the sa_res_key.
7654                          */
7655                         mtx_unlock(&lun->lun_lock);
7656                         free(ctsio->kern_data_ptr, M_CTL);
7657                         ctl_set_invalid_field(ctsio,
7658                                               /*sks_valid*/ 1,
7659                                               /*command*/ 0,
7660                                               /*field*/ 8,
7661                                               /*bit_valid*/ 0,
7662                                               /*bit*/ 0);
7663                         ctl_done((union ctl_io *)ctsio);
7664                         return (1);
7665                 }
7666
7667                 for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7668                         if (lun->per_res[i].registered
7669                          && memcmp(param->serv_act_res_key,
7670                             lun->per_res[i].res_key.key,
7671                             sizeof(struct scsi_per_res_key)) != 0)
7672                                 continue;
7673
7674                         found = 1;
7675                         lun->per_res[i].registered = 0;
7676                         memset(&lun->per_res[i].res_key, 0,
7677                                sizeof(struct scsi_per_res_key));
7678                         lun->pr_key_count--;
7679
7680                         if (!persis_offset
7681                          && i < CTL_MAX_INITIATORS)
7682                                 lun->pending_sense[i].ua_pending |=
7683                                         CTL_UA_REG_PREEMPT;
7684                         else if (persis_offset
7685                               && i >= persis_offset)
7686                                 lun->pending_sense[i-persis_offset].ua_pending|=
7687                                         CTL_UA_REG_PREEMPT;
7688                 }
7689                 if (!found) {
7690                         mtx_unlock(&lun->lun_lock);
7691                         free(ctsio->kern_data_ptr, M_CTL);
7692                         ctl_set_reservation_conflict(ctsio);
7693                         ctl_done((union ctl_io *)ctsio);
7694                         return (CTL_RETVAL_COMPLETE);
7695                 }
7696                 /* send msg to other side */
7697                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7698                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7699                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7700                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
7701                 persis_io.pr.pr_info.res_type = type;
7702                 memcpy(persis_io.pr.pr_info.sa_res_key,
7703                        param->serv_act_res_key,
7704                        sizeof(param->serv_act_res_key));
7705                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7706                      &persis_io, sizeof(persis_io), 0)) >
7707                      CTL_HA_STATUS_SUCCESS) {
7708                         printf("CTL:Persis Out error returned from "
7709                                "ctl_ha_msg_send %d\n", isc_retval);
7710                 }
7711         } else {
7712                 /* Reserved but not all registrants */
7713                 /* sa_res_key is res holder */
7714                 if (memcmp(param->serv_act_res_key,
7715                    lun->per_res[lun->pr_res_idx].res_key.key,
7716                    sizeof(struct scsi_per_res_key)) == 0) {
7717                         /* validate scope and type */
7718                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7719                              SPR_LU_SCOPE) {
7720                                 mtx_unlock(&lun->lun_lock);
7721                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7722                                                       /*sks_valid*/ 1,
7723                                                       /*command*/ 1,
7724                                                       /*field*/ 2,
7725                                                       /*bit_valid*/ 1,
7726                                                       /*bit*/ 4);
7727                                 ctl_done((union ctl_io *)ctsio);
7728                                 return (1);
7729                         }
7730
7731                         if (type>8 || type==2 || type==4 || type==0) {
7732                                 mtx_unlock(&lun->lun_lock);
7733                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7734                                                       /*sks_valid*/ 1,
7735                                                       /*command*/ 1,
7736                                                       /*field*/ 2,
7737                                                       /*bit_valid*/ 1,
7738                                                       /*bit*/ 0);
7739                                 ctl_done((union ctl_io *)ctsio);
7740                                 return (1);
7741                         }
7742
7743                         /*
7744                          * Do the following:
7745                          * if sa_res_key != res_key remove all
7746                          * registrants w/sa_res_key and generate UA
7747                          * for these registrants(Registrations
7748                          * Preempted) if it wasn't an exclusive
7749                          * reservation generate UA(Reservations
7750                          * Preempted) for all other registered nexuses
7751                          * if the type has changed. Establish the new
7752                          * reservation and holder. If res_key and
7753                          * sa_res_key are the same do the above
7754                          * except don't unregister the res holder.
7755                          */
7756
7757                         /*
7758                          * Temporarily unregister so it won't get
7759                          * removed or UA generated
7760                          */
7761                         lun->per_res[residx].registered = 0;
7762                         for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7763                                 if (lun->per_res[i].registered == 0)
7764                                         continue;
7765
7766                                 if (memcmp(param->serv_act_res_key,
7767                                     lun->per_res[i].res_key.key,
7768                                     sizeof(struct scsi_per_res_key)) == 0) {
7769                                         lun->per_res[i].registered = 0;
7770                                         memset(&lun->per_res[i].res_key,
7771                                                0,
7772                                                sizeof(struct scsi_per_res_key));
7773                                         lun->pr_key_count--;
7774
7775                                         if (!persis_offset
7776                                          && i < CTL_MAX_INITIATORS)
7777                                                 lun->pending_sense[i
7778                                                         ].ua_pending |=
7779                                                         CTL_UA_REG_PREEMPT;
7780                                         else if (persis_offset
7781                                               && i >= persis_offset)
7782                                                 lun->pending_sense[
7783                                                   i-persis_offset].ua_pending |=
7784                                                   CTL_UA_REG_PREEMPT;
7785                                 } else if (type != lun->res_type
7786                                         && (lun->res_type == SPR_TYPE_WR_EX_RO
7787                                          || lun->res_type ==SPR_TYPE_EX_AC_RO)){
7788                                                 if (!persis_offset
7789                                                  && i < CTL_MAX_INITIATORS)
7790                                                         lun->pending_sense[i
7791                                                         ].ua_pending |=
7792                                                         CTL_UA_RES_RELEASE;
7793                                                 else if (persis_offset
7794                                                       && i >= persis_offset)
7795                                                         lun->pending_sense[
7796                                                         i-persis_offset
7797                                                         ].ua_pending |=
7798                                                         CTL_UA_RES_RELEASE;
7799                                 }
7800                         }
7801                         lun->per_res[residx].registered = 1;
7802                         lun->res_type = type;
7803                         if (lun->res_type != SPR_TYPE_WR_EX_AR
7804                          && lun->res_type != SPR_TYPE_EX_AC_AR)
7805                                 lun->pr_res_idx = residx;
7806                         else
7807                                 lun->pr_res_idx =
7808                                         CTL_PR_ALL_REGISTRANTS;
7809
7810                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7811                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7812                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7813                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7814                         persis_io.pr.pr_info.res_type = type;
7815                         memcpy(persis_io.pr.pr_info.sa_res_key,
7816                                param->serv_act_res_key,
7817                                sizeof(param->serv_act_res_key));
7818                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7819                              &persis_io, sizeof(persis_io), 0)) >
7820                              CTL_HA_STATUS_SUCCESS) {
7821                                 printf("CTL:Persis Out error returned "
7822                                        "from ctl_ha_msg_send %d\n",
7823                                        isc_retval);
7824                         }
7825                 } else {
7826                         /*
7827                          * sa_res_key is not the res holder just
7828                          * remove registrants
7829                          */
7830                         int found=0;
7831
7832                         for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7833                                 if (memcmp(param->serv_act_res_key,
7834                                     lun->per_res[i].res_key.key,
7835                                     sizeof(struct scsi_per_res_key)) != 0)
7836                                         continue;
7837
7838                                 found = 1;
7839                                 lun->per_res[i].registered = 0;
7840                                 memset(&lun->per_res[i].res_key, 0,
7841                                        sizeof(struct scsi_per_res_key));
7842                                 lun->pr_key_count--;
7843
7844                                 if (!persis_offset
7845                                  && i < CTL_MAX_INITIATORS)
7846                                         lun->pending_sense[i].ua_pending |=
7847                                                 CTL_UA_REG_PREEMPT;
7848                                 else if (persis_offset
7849                                       && i >= persis_offset)
7850                                         lun->pending_sense[
7851                                                 i-persis_offset].ua_pending |=
7852                                                 CTL_UA_REG_PREEMPT;
7853                         }
7854
7855                         if (!found) {
7856                                 mtx_unlock(&lun->lun_lock);
7857                                 free(ctsio->kern_data_ptr, M_CTL);
7858                                 ctl_set_reservation_conflict(ctsio);
7859                                 ctl_done((union ctl_io *)ctsio);
7860                                 return (1);
7861                         }
7862                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7863                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7864                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7865                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7866                         persis_io.pr.pr_info.res_type = type;
7867                         memcpy(persis_io.pr.pr_info.sa_res_key,
7868                                param->serv_act_res_key,
7869                                sizeof(param->serv_act_res_key));
7870                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7871                              &persis_io, sizeof(persis_io), 0)) >
7872                              CTL_HA_STATUS_SUCCESS) {
7873                                 printf("CTL:Persis Out error returned "
7874                                        "from ctl_ha_msg_send %d\n",
7875                                 isc_retval);
7876                         }
7877                 }
7878         }
7879
7880         lun->PRGeneration++;
7881         mtx_unlock(&lun->lun_lock);
7882
7883         return (retval);
7884 }
7885
7886 static void
7887 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
7888 {
7889         int i;
7890
7891         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7892          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
7893          || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
7894                    msg->pr.pr_info.sa_res_key,
7895                    sizeof(struct scsi_per_res_key)) != 0) {
7896                 uint64_t sa_res_key;
7897                 sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
7898
7899                 if (sa_res_key == 0) {
7900                         /* temporarily unregister this nexus */
7901                         lun->per_res[msg->pr.pr_info.residx].registered = 0;
7902
7903                         /*
7904                          * Unregister everybody else and build UA for
7905                          * them
7906                          */
7907                         for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7908                                 if (lun->per_res[i].registered == 0)
7909                                         continue;
7910
7911                                 if (!persis_offset
7912                                  && i < CTL_MAX_INITIATORS)
7913                                         lun->pending_sense[i].ua_pending |=
7914                                                 CTL_UA_REG_PREEMPT;
7915                                 else if (persis_offset && i >= persis_offset)
7916                                         lun->pending_sense[i -
7917                                                 persis_offset].ua_pending |=
7918                                                 CTL_UA_REG_PREEMPT;
7919                                 lun->per_res[i].registered = 0;
7920                                 memset(&lun->per_res[i].res_key, 0,
7921                                        sizeof(struct scsi_per_res_key));
7922                         }
7923
7924                         lun->per_res[msg->pr.pr_info.residx].registered = 1;
7925                         lun->pr_key_count = 1;
7926                         lun->res_type = msg->pr.pr_info.res_type;
7927                         if (lun->res_type != SPR_TYPE_WR_EX_AR
7928                          && lun->res_type != SPR_TYPE_EX_AC_AR)
7929                                 lun->pr_res_idx = msg->pr.pr_info.residx;
7930                 } else {
7931                         for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7932                                 if (memcmp(msg->pr.pr_info.sa_res_key,
7933                                    lun->per_res[i].res_key.key,
7934                                    sizeof(struct scsi_per_res_key)) != 0)
7935                                         continue;
7936
7937                                 lun->per_res[i].registered = 0;
7938                                 memset(&lun->per_res[i].res_key, 0,
7939                                        sizeof(struct scsi_per_res_key));
7940                                 lun->pr_key_count--;
7941
7942                                 if (!persis_offset
7943                                  && i < persis_offset)
7944                                         lun->pending_sense[i].ua_pending |=
7945                                                 CTL_UA_REG_PREEMPT;
7946                                 else if (persis_offset
7947                                       && i >= persis_offset)
7948                                         lun->pending_sense[i -
7949                                                 persis_offset].ua_pending |=
7950                                                 CTL_UA_REG_PREEMPT;
7951                         }
7952                 }
7953         } else {
7954                 /*
7955                  * Temporarily unregister so it won't get removed
7956                  * or UA generated
7957                  */
7958                 lun->per_res[msg->pr.pr_info.residx].registered = 0;
7959                 for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7960                         if (lun->per_res[i].registered == 0)
7961                                 continue;
7962
7963                         if (memcmp(msg->pr.pr_info.sa_res_key,
7964                            lun->per_res[i].res_key.key,
7965                            sizeof(struct scsi_per_res_key)) == 0) {
7966                                 lun->per_res[i].registered = 0;
7967                                 memset(&lun->per_res[i].res_key, 0,
7968                                        sizeof(struct scsi_per_res_key));
7969                                 lun->pr_key_count--;
7970                                 if (!persis_offset
7971                                  && i < CTL_MAX_INITIATORS)
7972                                         lun->pending_sense[i].ua_pending |=
7973                                                 CTL_UA_REG_PREEMPT;
7974                                 else if (persis_offset
7975                                       && i >= persis_offset)
7976                                         lun->pending_sense[i -
7977                                                 persis_offset].ua_pending |=
7978                                                 CTL_UA_REG_PREEMPT;
7979                         } else if (msg->pr.pr_info.res_type != lun->res_type
7980                                 && (lun->res_type == SPR_TYPE_WR_EX_RO
7981                                  || lun->res_type == SPR_TYPE_EX_AC_RO)) {
7982                                         if (!persis_offset
7983                                          && i < persis_offset)
7984                                                 lun->pending_sense[i
7985                                                         ].ua_pending |=
7986                                                         CTL_UA_RES_RELEASE;
7987                                         else if (persis_offset
7988                                               && i >= persis_offset)
7989                                         lun->pending_sense[i -
7990                                                 persis_offset].ua_pending |=
7991                                                 CTL_UA_RES_RELEASE;
7992                         }
7993                 }
7994                 lun->per_res[msg->pr.pr_info.residx].registered = 1;
7995                 lun->res_type = msg->pr.pr_info.res_type;
7996                 if (lun->res_type != SPR_TYPE_WR_EX_AR
7997                  && lun->res_type != SPR_TYPE_EX_AC_AR)
7998                         lun->pr_res_idx = msg->pr.pr_info.residx;
7999                 else
8000                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8001         }
8002         lun->PRGeneration++;
8003
8004 }
8005
8006
8007 int
8008 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8009 {
8010         int retval;
8011         int isc_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;
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         softc = control_softc;
8028
8029         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8030         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8031
8032         /*
8033          * We only support whole-LUN scope.  The scope & type are ignored for
8034          * register, register and ignore existing key and clear.
8035          * We sometimes ignore scope and type on preempts too!!
8036          * Verify reservation type here as well.
8037          */
8038         type = cdb->scope_type & SPR_TYPE_MASK;
8039         if ((cdb->action == SPRO_RESERVE)
8040          || (cdb->action == SPRO_RELEASE)) {
8041                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8042                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8043                                               /*sks_valid*/ 1,
8044                                               /*command*/ 1,
8045                                               /*field*/ 2,
8046                                               /*bit_valid*/ 1,
8047                                               /*bit*/ 4);
8048                         ctl_done((union ctl_io *)ctsio);
8049                         return (CTL_RETVAL_COMPLETE);
8050                 }
8051
8052                 if (type>8 || type==2 || type==4 || type==0) {
8053                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8054                                               /*sks_valid*/ 1,
8055                                               /*command*/ 1,
8056                                               /*field*/ 2,
8057                                               /*bit_valid*/ 1,
8058                                               /*bit*/ 0);
8059                         ctl_done((union ctl_io *)ctsio);
8060                         return (CTL_RETVAL_COMPLETE);
8061                 }
8062         }
8063
8064         param_len = scsi_4btoul(cdb->length);
8065
8066         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8067                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8068                 ctsio->kern_data_len = param_len;
8069                 ctsio->kern_total_len = param_len;
8070                 ctsio->kern_data_resid = 0;
8071                 ctsio->kern_rel_offset = 0;
8072                 ctsio->kern_sg_entries = 0;
8073                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8074                 ctsio->be_move_done = ctl_config_move_done;
8075                 ctl_datamove((union ctl_io *)ctsio);
8076
8077                 return (CTL_RETVAL_COMPLETE);
8078         }
8079
8080         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8081
8082         residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8083         res_key = scsi_8btou64(param->res_key.key);
8084         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8085
8086         /*
8087          * Validate the reservation key here except for SPRO_REG_IGNO
8088          * This must be done for all other service actions
8089          */
8090         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8091                 mtx_lock(&lun->lun_lock);
8092                 if (lun->per_res[residx].registered) {
8093                     if (memcmp(param->res_key.key,
8094                                lun->per_res[residx].res_key.key,
8095                                ctl_min(sizeof(param->res_key),
8096                                sizeof(lun->per_res[residx].res_key))) != 0) {
8097                                 /*
8098                                  * The current key passed in doesn't match
8099                                  * the one the initiator previously
8100                                  * registered.
8101                                  */
8102                                 mtx_unlock(&lun->lun_lock);
8103                                 free(ctsio->kern_data_ptr, M_CTL);
8104                                 ctl_set_reservation_conflict(ctsio);
8105                                 ctl_done((union ctl_io *)ctsio);
8106                                 return (CTL_RETVAL_COMPLETE);
8107                         }
8108                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8109                         /*
8110                          * We are not registered
8111                          */
8112                         mtx_unlock(&lun->lun_lock);
8113                         free(ctsio->kern_data_ptr, M_CTL);
8114                         ctl_set_reservation_conflict(ctsio);
8115                         ctl_done((union ctl_io *)ctsio);
8116                         return (CTL_RETVAL_COMPLETE);
8117                 } else if (res_key != 0) {
8118                         /*
8119                          * We are not registered and trying to register but
8120                          * the register key isn't zero.
8121                          */
8122                         mtx_unlock(&lun->lun_lock);
8123                         free(ctsio->kern_data_ptr, M_CTL);
8124                         ctl_set_reservation_conflict(ctsio);
8125                         ctl_done((union ctl_io *)ctsio);
8126                         return (CTL_RETVAL_COMPLETE);
8127                 }
8128                 mtx_unlock(&lun->lun_lock);
8129         }
8130
8131         switch (cdb->action & SPRO_ACTION_MASK) {
8132         case SPRO_REGISTER:
8133         case SPRO_REG_IGNO: {
8134
8135 #if 0
8136                 printf("Registration received\n");
8137 #endif
8138
8139                 /*
8140                  * We don't support any of these options, as we report in
8141                  * the read capabilities request (see
8142                  * ctl_persistent_reserve_in(), above).
8143                  */
8144                 if ((param->flags & SPR_SPEC_I_PT)
8145                  || (param->flags & SPR_ALL_TG_PT)
8146                  || (param->flags & SPR_APTPL)) {
8147                         int bit_ptr;
8148
8149                         if (param->flags & SPR_APTPL)
8150                                 bit_ptr = 0;
8151                         else if (param->flags & SPR_ALL_TG_PT)
8152                                 bit_ptr = 2;
8153                         else /* SPR_SPEC_I_PT */
8154                                 bit_ptr = 3;
8155
8156                         free(ctsio->kern_data_ptr, M_CTL);
8157                         ctl_set_invalid_field(ctsio,
8158                                               /*sks_valid*/ 1,
8159                                               /*command*/ 0,
8160                                               /*field*/ 20,
8161                                               /*bit_valid*/ 1,
8162                                               /*bit*/ bit_ptr);
8163                         ctl_done((union ctl_io *)ctsio);
8164                         return (CTL_RETVAL_COMPLETE);
8165                 }
8166
8167                 mtx_lock(&lun->lun_lock);
8168
8169                 /*
8170                  * The initiator wants to clear the
8171                  * key/unregister.
8172                  */
8173                 if (sa_res_key == 0) {
8174                         if ((res_key == 0
8175                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8176                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8177                           && !lun->per_res[residx].registered)) {
8178                                 mtx_unlock(&lun->lun_lock);
8179                                 goto done;
8180                         }
8181
8182                         lun->per_res[residx].registered = 0;
8183                         memset(&lun->per_res[residx].res_key,
8184                                0, sizeof(lun->per_res[residx].res_key));
8185                         lun->pr_key_count--;
8186
8187                         if (residx == lun->pr_res_idx) {
8188                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8189                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8190
8191                                 if ((lun->res_type == SPR_TYPE_WR_EX_RO
8192                                   || lun->res_type == SPR_TYPE_EX_AC_RO)
8193                                  && lun->pr_key_count) {
8194                                         /*
8195                                          * If the reservation is a registrants
8196                                          * only type we need to generate a UA
8197                                          * for other registered inits.  The
8198                                          * sense code should be RESERVATIONS
8199                                          * RELEASED
8200                                          */
8201
8202                                         for (i = 0; i < CTL_MAX_INITIATORS;i++){
8203                                                 if (lun->per_res[
8204                                                     i+persis_offset].registered
8205                                                     == 0)
8206                                                         continue;
8207                                                 lun->pending_sense[i
8208                                                         ].ua_pending |=
8209                                                         CTL_UA_RES_RELEASE;
8210                                         }
8211                                 }
8212                                 lun->res_type = 0;
8213                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8214                                 if (lun->pr_key_count==0) {
8215                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8216                                         lun->res_type = 0;
8217                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8218                                 }
8219                         }
8220                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8221                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8222                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8223                         persis_io.pr.pr_info.residx = residx;
8224                         if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8225                              &persis_io, sizeof(persis_io), 0 )) >
8226                              CTL_HA_STATUS_SUCCESS) {
8227                                 printf("CTL:Persis Out error returned from "
8228                                        "ctl_ha_msg_send %d\n", isc_retval);
8229                         }
8230                 } else /* sa_res_key != 0 */ {
8231
8232                         /*
8233                          * If we aren't registered currently then increment
8234                          * the key count and set the registered flag.
8235                          */
8236                         if (!lun->per_res[residx].registered) {
8237                                 lun->pr_key_count++;
8238                                 lun->per_res[residx].registered = 1;
8239                         }
8240
8241                         memcpy(&lun->per_res[residx].res_key,
8242                                param->serv_act_res_key,
8243                                ctl_min(sizeof(param->serv_act_res_key),
8244                                sizeof(lun->per_res[residx].res_key)));
8245
8246                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8247                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8248                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8249                         persis_io.pr.pr_info.residx = residx;
8250                         memcpy(persis_io.pr.pr_info.sa_res_key,
8251                                param->serv_act_res_key,
8252                                sizeof(param->serv_act_res_key));
8253                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8254                              &persis_io, sizeof(persis_io), 0)) >
8255                              CTL_HA_STATUS_SUCCESS) {
8256                                 printf("CTL:Persis Out error returned from "
8257                                        "ctl_ha_msg_send %d\n", isc_retval);
8258                         }
8259                 }
8260                 lun->PRGeneration++;
8261                 mtx_unlock(&lun->lun_lock);
8262
8263                 break;
8264         }
8265         case SPRO_RESERVE:
8266 #if 0
8267                 printf("Reserve executed type %d\n", type);
8268 #endif
8269                 mtx_lock(&lun->lun_lock);
8270                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8271                         /*
8272                          * if this isn't the reservation holder and it's
8273                          * not a "all registrants" type or if the type is
8274                          * different then we have a conflict
8275                          */
8276                         if ((lun->pr_res_idx != residx
8277                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8278                          || lun->res_type != type) {
8279                                 mtx_unlock(&lun->lun_lock);
8280                                 free(ctsio->kern_data_ptr, M_CTL);
8281                                 ctl_set_reservation_conflict(ctsio);
8282                                 ctl_done((union ctl_io *)ctsio);
8283                                 return (CTL_RETVAL_COMPLETE);
8284                         }
8285                         mtx_unlock(&lun->lun_lock);
8286                 } else /* create a reservation */ {
8287                         /*
8288                          * If it's not an "all registrants" type record
8289                          * reservation holder
8290                          */
8291                         if (type != SPR_TYPE_WR_EX_AR
8292                          && type != SPR_TYPE_EX_AC_AR)
8293                                 lun->pr_res_idx = residx; /* Res holder */
8294                         else
8295                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8296
8297                         lun->flags |= CTL_LUN_PR_RESERVED;
8298                         lun->res_type = type;
8299
8300                         mtx_unlock(&lun->lun_lock);
8301
8302                         /* send msg to other side */
8303                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8304                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8305                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8306                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8307                         persis_io.pr.pr_info.res_type = type;
8308                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8309                              &persis_io, sizeof(persis_io), 0)) >
8310                              CTL_HA_STATUS_SUCCESS) {
8311                                 printf("CTL:Persis Out error returned from "
8312                                        "ctl_ha_msg_send %d\n", isc_retval);
8313                         }
8314                 }
8315                 break;
8316
8317         case SPRO_RELEASE:
8318                 mtx_lock(&lun->lun_lock);
8319                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8320                         /* No reservation exists return good status */
8321                         mtx_unlock(&lun->lun_lock);
8322                         goto done;
8323                 }
8324                 /*
8325                  * Is this nexus a reservation holder?
8326                  */
8327                 if (lun->pr_res_idx != residx
8328                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8329                         /*
8330                          * not a res holder return good status but
8331                          * do nothing
8332                          */
8333                         mtx_unlock(&lun->lun_lock);
8334                         goto done;
8335                 }
8336
8337                 if (lun->res_type != type) {
8338                         mtx_unlock(&lun->lun_lock);
8339                         free(ctsio->kern_data_ptr, M_CTL);
8340                         ctl_set_illegal_pr_release(ctsio);
8341                         ctl_done((union ctl_io *)ctsio);
8342                         return (CTL_RETVAL_COMPLETE);
8343                 }
8344
8345                 /* okay to release */
8346                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8347                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8348                 lun->res_type = 0;
8349
8350                 /*
8351                  * if this isn't an exclusive access
8352                  * res generate UA for all other
8353                  * registrants.
8354                  */
8355                 if (type != SPR_TYPE_EX_AC
8356                  && type != SPR_TYPE_WR_EX) {
8357                         /*
8358                          * temporarily unregister so we don't generate UA
8359                          */
8360                         lun->per_res[residx].registered = 0;
8361
8362                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8363                                 if (lun->per_res[i+persis_offset].registered
8364                                     == 0)
8365                                         continue;
8366                                 lun->pending_sense[i].ua_pending |=
8367                                         CTL_UA_RES_RELEASE;
8368                         }
8369
8370                         lun->per_res[residx].registered = 1;
8371                 }
8372                 mtx_unlock(&lun->lun_lock);
8373                 /* Send msg to other side */
8374                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8375                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8376                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8377                 if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8378                      sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8379                         printf("CTL:Persis Out error returned from "
8380                                "ctl_ha_msg_send %d\n", isc_retval);
8381                 }
8382                 break;
8383
8384         case SPRO_CLEAR:
8385                 /* send msg to other side */
8386
8387                 mtx_lock(&lun->lun_lock);
8388                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8389                 lun->res_type = 0;
8390                 lun->pr_key_count = 0;
8391                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8392
8393
8394                 memset(&lun->per_res[residx].res_key,
8395                        0, sizeof(lun->per_res[residx].res_key));
8396                 lun->per_res[residx].registered = 0;
8397
8398                 for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8399                         if (lun->per_res[i].registered) {
8400                                 if (!persis_offset && i < CTL_MAX_INITIATORS)
8401                                         lun->pending_sense[i].ua_pending |=
8402                                                 CTL_UA_RES_PREEMPT;
8403                                 else if (persis_offset && i >= persis_offset)
8404                                         lun->pending_sense[i-persis_offset
8405                                             ].ua_pending |= CTL_UA_RES_PREEMPT;
8406
8407                                 memset(&lun->per_res[i].res_key,
8408                                        0, sizeof(struct scsi_per_res_key));
8409                                 lun->per_res[i].registered = 0;
8410                         }
8411                 lun->PRGeneration++;
8412                 mtx_unlock(&lun->lun_lock);
8413                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8414                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8415                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8416                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8417                      sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8418                         printf("CTL:Persis Out error returned from "
8419                                "ctl_ha_msg_send %d\n", isc_retval);
8420                 }
8421                 break;
8422
8423         case SPRO_PREEMPT: {
8424                 int nretval;
8425
8426                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8427                                           residx, ctsio, cdb, param);
8428                 if (nretval != 0)
8429                         return (CTL_RETVAL_COMPLETE);
8430                 break;
8431         }
8432         default:
8433                 panic("Invalid PR type %x", cdb->action);
8434         }
8435
8436 done:
8437         free(ctsio->kern_data_ptr, M_CTL);
8438         ctl_set_success(ctsio);
8439         ctl_done((union ctl_io *)ctsio);
8440
8441         return (retval);
8442 }
8443
8444 /*
8445  * This routine is for handling a message from the other SC pertaining to
8446  * persistent reserve out. All the error checking will have been done
8447  * so only perorming the action need be done here to keep the two
8448  * in sync.
8449  */
8450 static void
8451 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8452 {
8453         struct ctl_lun *lun;
8454         struct ctl_softc *softc;
8455         int i;
8456         uint32_t targ_lun;
8457
8458         softc = control_softc;
8459
8460         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8461         lun = softc->ctl_luns[targ_lun];
8462         mtx_lock(&lun->lun_lock);
8463         switch(msg->pr.pr_info.action) {
8464         case CTL_PR_REG_KEY:
8465                 if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8466                         lun->per_res[msg->pr.pr_info.residx].registered = 1;
8467                         lun->pr_key_count++;
8468                 }
8469                 lun->PRGeneration++;
8470                 memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8471                        msg->pr.pr_info.sa_res_key,
8472                        sizeof(struct scsi_per_res_key));
8473                 break;
8474
8475         case CTL_PR_UNREG_KEY:
8476                 lun->per_res[msg->pr.pr_info.residx].registered = 0;
8477                 memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8478                        0, sizeof(struct scsi_per_res_key));
8479                 lun->pr_key_count--;
8480
8481                 /* XXX Need to see if the reservation has been released */
8482                 /* if so do we need to generate UA? */
8483                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8484                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8485                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8486
8487                         if ((lun->res_type == SPR_TYPE_WR_EX_RO
8488                           || lun->res_type == SPR_TYPE_EX_AC_RO)
8489                          && lun->pr_key_count) {
8490                                 /*
8491                                  * If the reservation is a registrants
8492                                  * only type we need to generate a UA
8493                                  * for other registered inits.  The
8494                                  * sense code should be RESERVATIONS
8495                                  * RELEASED
8496                                  */
8497
8498                                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8499                                         if (lun->per_res[i+
8500                                             persis_offset].registered == 0)
8501                                                 continue;
8502
8503                                         lun->pending_sense[i
8504                                                 ].ua_pending |=
8505                                                 CTL_UA_RES_RELEASE;
8506                                 }
8507                         }
8508                         lun->res_type = 0;
8509                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8510                         if (lun->pr_key_count==0) {
8511                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8512                                 lun->res_type = 0;
8513                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8514                         }
8515                 }
8516                 lun->PRGeneration++;
8517                 break;
8518
8519         case CTL_PR_RESERVE:
8520                 lun->flags |= CTL_LUN_PR_RESERVED;
8521                 lun->res_type = msg->pr.pr_info.res_type;
8522                 lun->pr_res_idx = msg->pr.pr_info.residx;
8523
8524                 break;
8525
8526         case CTL_PR_RELEASE:
8527                 /*
8528                  * if this isn't an exclusive access res generate UA for all
8529                  * other registrants.
8530                  */
8531                 if (lun->res_type != SPR_TYPE_EX_AC
8532                  && lun->res_type != SPR_TYPE_WR_EX) {
8533                         for (i = 0; i < CTL_MAX_INITIATORS; i++)
8534                                 if (lun->per_res[i+persis_offset].registered)
8535                                         lun->pending_sense[i].ua_pending |=
8536                                                 CTL_UA_RES_RELEASE;
8537                 }
8538
8539                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8540                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8541                 lun->res_type = 0;
8542                 break;
8543
8544         case CTL_PR_PREEMPT:
8545                 ctl_pro_preempt_other(lun, msg);
8546                 break;
8547         case CTL_PR_CLEAR:
8548                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8549                 lun->res_type = 0;
8550                 lun->pr_key_count = 0;
8551                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8552
8553                 for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8554                         if (lun->per_res[i].registered == 0)
8555                                 continue;
8556                         if (!persis_offset
8557                          && i < CTL_MAX_INITIATORS)
8558                                 lun->pending_sense[i].ua_pending |=
8559                                         CTL_UA_RES_PREEMPT;
8560                         else if (persis_offset
8561                               && i >= persis_offset)
8562                                 lun->pending_sense[i-persis_offset].ua_pending|=
8563                                         CTL_UA_RES_PREEMPT;
8564                         memset(&lun->per_res[i].res_key, 0,
8565                                sizeof(struct scsi_per_res_key));
8566                         lun->per_res[i].registered = 0;
8567                 }
8568                 lun->PRGeneration++;
8569                 break;
8570         }
8571
8572         mtx_unlock(&lun->lun_lock);
8573 }
8574
8575 int
8576 ctl_read_write(struct ctl_scsiio *ctsio)
8577 {
8578         struct ctl_lun *lun;
8579         struct ctl_lba_len_flags *lbalen;
8580         uint64_t lba;
8581         uint32_t num_blocks;
8582         int fua, dpo;
8583         int retval;
8584         int isread;
8585
8586         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8587
8588         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8589
8590         fua = 0;
8591         dpo = 0;
8592
8593         retval = CTL_RETVAL_COMPLETE;
8594
8595         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8596               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8597         if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
8598                 uint32_t residx;
8599
8600                 /*
8601                  * XXX KDM need a lock here.
8602                  */
8603                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8604                 if ((lun->res_type == SPR_TYPE_EX_AC
8605                   && residx != lun->pr_res_idx)
8606                  || ((lun->res_type == SPR_TYPE_EX_AC_RO
8607                    || lun->res_type == SPR_TYPE_EX_AC_AR)
8608                   && !lun->per_res[residx].registered)) {
8609                         ctl_set_reservation_conflict(ctsio);
8610                         ctl_done((union ctl_io *)ctsio);
8611                         return (CTL_RETVAL_COMPLETE);
8612                 }
8613         }
8614
8615         switch (ctsio->cdb[0]) {
8616         case READ_6:
8617         case WRITE_6: {
8618                 struct scsi_rw_6 *cdb;
8619
8620                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8621
8622                 lba = scsi_3btoul(cdb->addr);
8623                 /* only 5 bits are valid in the most significant address byte */
8624                 lba &= 0x1fffff;
8625                 num_blocks = cdb->length;
8626                 /*
8627                  * This is correct according to SBC-2.
8628                  */
8629                 if (num_blocks == 0)
8630                         num_blocks = 256;
8631                 break;
8632         }
8633         case READ_10:
8634         case WRITE_10: {
8635                 struct scsi_rw_10 *cdb;
8636
8637                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8638
8639                 if (cdb->byte2 & SRW10_FUA)
8640                         fua = 1;
8641                 if (cdb->byte2 & SRW10_DPO)
8642                         dpo = 1;
8643
8644                 lba = scsi_4btoul(cdb->addr);
8645                 num_blocks = scsi_2btoul(cdb->length);
8646                 break;
8647         }
8648         case WRITE_VERIFY_10: {
8649                 struct scsi_write_verify_10 *cdb;
8650
8651                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8652
8653                 /*
8654                  * XXX KDM we should do actual write verify support at some
8655                  * point.  This is obviously fake, we're just translating
8656                  * things to a write.  So we don't even bother checking the
8657                  * BYTCHK field, since we don't do any verification.  If
8658                  * the user asks for it, we'll just pretend we did it.
8659                  */
8660                 if (cdb->byte2 & SWV_DPO)
8661                         dpo = 1;
8662
8663                 lba = scsi_4btoul(cdb->addr);
8664                 num_blocks = scsi_2btoul(cdb->length);
8665                 break;
8666         }
8667         case READ_12:
8668         case WRITE_12: {
8669                 struct scsi_rw_12 *cdb;
8670
8671                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8672
8673                 if (cdb->byte2 & SRW12_FUA)
8674                         fua = 1;
8675                 if (cdb->byte2 & SRW12_DPO)
8676                         dpo = 1;
8677                 lba = scsi_4btoul(cdb->addr);
8678                 num_blocks = scsi_4btoul(cdb->length);
8679                 break;
8680         }
8681         case WRITE_VERIFY_12: {
8682                 struct scsi_write_verify_12 *cdb;
8683
8684                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8685
8686                 if (cdb->byte2 & SWV_DPO)
8687                         dpo = 1;
8688
8689                 lba = scsi_4btoul(cdb->addr);
8690                 num_blocks = scsi_4btoul(cdb->length);
8691
8692                 break;
8693         }
8694         case READ_16:
8695         case WRITE_16: {
8696                 struct scsi_rw_16 *cdb;
8697
8698                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8699
8700                 if (cdb->byte2 & SRW12_FUA)
8701                         fua = 1;
8702                 if (cdb->byte2 & SRW12_DPO)
8703                         dpo = 1;
8704
8705                 lba = scsi_8btou64(cdb->addr);
8706                 num_blocks = scsi_4btoul(cdb->length);
8707                 break;
8708         }
8709         case WRITE_VERIFY_16: {
8710                 struct scsi_write_verify_16 *cdb;
8711
8712                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8713
8714                 if (cdb->byte2 & SWV_DPO)
8715                         dpo = 1;
8716
8717                 lba = scsi_8btou64(cdb->addr);
8718                 num_blocks = scsi_4btoul(cdb->length);
8719                 break;
8720         }
8721         default:
8722                 /*
8723                  * We got a command we don't support.  This shouldn't
8724                  * happen, commands should be filtered out above us.
8725                  */
8726                 ctl_set_invalid_opcode(ctsio);
8727                 ctl_done((union ctl_io *)ctsio);
8728
8729                 return (CTL_RETVAL_COMPLETE);
8730                 break; /* NOTREACHED */
8731         }
8732
8733         /*
8734          * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
8735          * interesting for us, but if RAIDCore is in write-back mode,
8736          * getting it to do write-through for a particular transaction may
8737          * not be possible.
8738          */
8739
8740         /*
8741          * The first check is to make sure we're in bounds, the second
8742          * check is to catch wrap-around problems.  If the lba + num blocks
8743          * is less than the lba, then we've wrapped around and the block
8744          * range is invalid anyway.
8745          */
8746         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8747          || ((lba + num_blocks) < lba)) {
8748                 ctl_set_lba_out_of_range(ctsio);
8749                 ctl_done((union ctl_io *)ctsio);
8750                 return (CTL_RETVAL_COMPLETE);
8751         }
8752
8753         /*
8754          * According to SBC-3, a transfer length of 0 is not an error.
8755          * Note that this cannot happen with WRITE(6) or READ(6), since 0
8756          * translates to 256 blocks for those commands.
8757          */
8758         if (num_blocks == 0) {
8759                 ctl_set_success(ctsio);
8760                 ctl_done((union ctl_io *)ctsio);
8761                 return (CTL_RETVAL_COMPLETE);
8762         }
8763
8764         lbalen = (struct ctl_lba_len_flags *)
8765             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8766         lbalen->lba = lba;
8767         lbalen->len = num_blocks;
8768         lbalen->flags = isread ? CTL_LLF_READ : CTL_LLF_WRITE;
8769
8770         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8771         ctsio->kern_rel_offset = 0;
8772
8773         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8774
8775         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8776
8777         return (retval);
8778 }
8779
8780 static int
8781 ctl_cnw_cont(union ctl_io *io)
8782 {
8783         struct ctl_scsiio *ctsio;
8784         struct ctl_lun *lun;
8785         struct ctl_lba_len_flags *lbalen;
8786         int retval;
8787
8788         ctsio = &io->scsiio;
8789         ctsio->io_hdr.status = CTL_STATUS_NONE;
8790         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8791         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8792         lbalen = (struct ctl_lba_len_flags *)
8793             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8794         lbalen->flags = CTL_LLF_WRITE;
8795
8796         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8797         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8798         return (retval);
8799 }
8800
8801 int
8802 ctl_cnw(struct ctl_scsiio *ctsio)
8803 {
8804         struct ctl_lun *lun;
8805         struct ctl_lba_len_flags *lbalen;
8806         uint64_t lba;
8807         uint32_t num_blocks;
8808         int fua, dpo;
8809         int retval;
8810
8811         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8812
8813         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8814
8815         fua = 0;
8816         dpo = 0;
8817
8818         retval = CTL_RETVAL_COMPLETE;
8819
8820         switch (ctsio->cdb[0]) {
8821         case COMPARE_AND_WRITE: {
8822                 struct scsi_compare_and_write *cdb;
8823
8824                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8825
8826                 if (cdb->byte2 & SRW10_FUA)
8827                         fua = 1;
8828                 if (cdb->byte2 & SRW10_DPO)
8829                         dpo = 1;
8830                 lba = scsi_8btou64(cdb->addr);
8831                 num_blocks = cdb->length;
8832                 break;
8833         }
8834         default:
8835                 /*
8836                  * We got a command we don't support.  This shouldn't
8837                  * happen, commands should be filtered out above us.
8838                  */
8839                 ctl_set_invalid_opcode(ctsio);
8840                 ctl_done((union ctl_io *)ctsio);
8841
8842                 return (CTL_RETVAL_COMPLETE);
8843                 break; /* NOTREACHED */
8844         }
8845
8846         /*
8847          * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
8848          * interesting for us, but if RAIDCore is in write-back mode,
8849          * getting it to do write-through for a particular transaction may
8850          * not be possible.
8851          */
8852
8853         /*
8854          * The first check is to make sure we're in bounds, the second
8855          * check is to catch wrap-around problems.  If the lba + num blocks
8856          * is less than the lba, then we've wrapped around and the block
8857          * range is invalid anyway.
8858          */
8859         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8860          || ((lba + num_blocks) < lba)) {
8861                 ctl_set_lba_out_of_range(ctsio);
8862                 ctl_done((union ctl_io *)ctsio);
8863                 return (CTL_RETVAL_COMPLETE);
8864         }
8865
8866         /*
8867          * According to SBC-3, a transfer length of 0 is not an error.
8868          */
8869         if (num_blocks == 0) {
8870                 ctl_set_success(ctsio);
8871                 ctl_done((union ctl_io *)ctsio);
8872                 return (CTL_RETVAL_COMPLETE);
8873         }
8874
8875         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8876         ctsio->kern_rel_offset = 0;
8877
8878         /*
8879          * Set the IO_CONT flag, so that if this I/O gets passed to
8880          * ctl_data_submit_done(), it'll get passed back to
8881          * ctl_ctl_cnw_cont() for further processing.
8882          */
8883         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8884         ctsio->io_cont = ctl_cnw_cont;
8885
8886         lbalen = (struct ctl_lba_len_flags *)
8887             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8888         lbalen->lba = lba;
8889         lbalen->len = num_blocks;
8890         lbalen->flags = CTL_LLF_COMPARE;
8891
8892         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8893         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8894         return (retval);
8895 }
8896
8897 int
8898 ctl_verify(struct ctl_scsiio *ctsio)
8899 {
8900         struct ctl_lun *lun;
8901         struct ctl_lba_len_flags *lbalen;
8902         uint64_t lba;
8903         uint32_t num_blocks;
8904         int bytchk, dpo;
8905         int retval;
8906
8907         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8908
8909         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8910
8911         bytchk = 0;
8912         dpo = 0;
8913         retval = CTL_RETVAL_COMPLETE;
8914
8915         switch (ctsio->cdb[0]) {
8916         case VERIFY_10: {
8917                 struct scsi_verify_10 *cdb;
8918
8919                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
8920                 if (cdb->byte2 & SVFY_BYTCHK)
8921                         bytchk = 1;
8922                 if (cdb->byte2 & SVFY_DPO)
8923                         dpo = 1;
8924                 lba = scsi_4btoul(cdb->addr);
8925                 num_blocks = scsi_2btoul(cdb->length);
8926                 break;
8927         }
8928         case VERIFY_12: {
8929                 struct scsi_verify_12 *cdb;
8930
8931                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
8932                 if (cdb->byte2 & SVFY_BYTCHK)
8933                         bytchk = 1;
8934                 if (cdb->byte2 & SVFY_DPO)
8935                         dpo = 1;
8936                 lba = scsi_4btoul(cdb->addr);
8937                 num_blocks = scsi_4btoul(cdb->length);
8938                 break;
8939         }
8940         case VERIFY_16: {
8941                 struct scsi_rw_16 *cdb;
8942
8943                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8944                 if (cdb->byte2 & SVFY_BYTCHK)
8945                         bytchk = 1;
8946                 if (cdb->byte2 & SVFY_DPO)
8947                         dpo = 1;
8948                 lba = scsi_8btou64(cdb->addr);
8949                 num_blocks = scsi_4btoul(cdb->length);
8950                 break;
8951         }
8952         default:
8953                 /*
8954                  * We got a command we don't support.  This shouldn't
8955                  * happen, commands should be filtered out above us.
8956                  */
8957                 ctl_set_invalid_opcode(ctsio);
8958                 ctl_done((union ctl_io *)ctsio);
8959                 return (CTL_RETVAL_COMPLETE);
8960         }
8961
8962         /*
8963          * The first check is to make sure we're in bounds, the second
8964          * check is to catch wrap-around problems.  If the lba + num blocks
8965          * is less than the lba, then we've wrapped around and the block
8966          * range is invalid anyway.
8967          */
8968         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8969          || ((lba + num_blocks) < lba)) {
8970                 ctl_set_lba_out_of_range(ctsio);
8971                 ctl_done((union ctl_io *)ctsio);
8972                 return (CTL_RETVAL_COMPLETE);
8973         }
8974
8975         /*
8976          * According to SBC-3, a transfer length of 0 is not an error.
8977          */
8978         if (num_blocks == 0) {
8979                 ctl_set_success(ctsio);
8980                 ctl_done((union ctl_io *)ctsio);
8981                 return (CTL_RETVAL_COMPLETE);
8982         }
8983
8984         lbalen = (struct ctl_lba_len_flags *)
8985             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8986         lbalen->lba = lba;
8987         lbalen->len = num_blocks;
8988         if (bytchk) {
8989                 lbalen->flags = CTL_LLF_COMPARE;
8990                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8991         } else {
8992                 lbalen->flags = CTL_LLF_VERIFY;
8993                 ctsio->kern_total_len = 0;
8994         }
8995         ctsio->kern_rel_offset = 0;
8996
8997         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
8998         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8999         return (retval);
9000 }
9001
9002 int
9003 ctl_report_luns(struct ctl_scsiio *ctsio)
9004 {
9005         struct scsi_report_luns *cdb;
9006         struct scsi_report_luns_data *lun_data;
9007         struct ctl_lun *lun, *request_lun;
9008         int num_luns, retval;
9009         uint32_t alloc_len, lun_datalen;
9010         int num_filled, well_known;
9011         uint32_t initidx, targ_lun_id, lun_id;
9012
9013         retval = CTL_RETVAL_COMPLETE;
9014         well_known = 0;
9015
9016         cdb = (struct scsi_report_luns *)ctsio->cdb;
9017
9018         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9019
9020         mtx_lock(&control_softc->ctl_lock);
9021         num_luns = control_softc->num_luns;
9022         mtx_unlock(&control_softc->ctl_lock);
9023
9024         switch (cdb->select_report) {
9025         case RPL_REPORT_DEFAULT:
9026         case RPL_REPORT_ALL:
9027                 break;
9028         case RPL_REPORT_WELLKNOWN:
9029                 well_known = 1;
9030                 num_luns = 0;
9031                 break;
9032         default:
9033                 ctl_set_invalid_field(ctsio,
9034                                       /*sks_valid*/ 1,
9035                                       /*command*/ 1,
9036                                       /*field*/ 2,
9037                                       /*bit_valid*/ 0,
9038                                       /*bit*/ 0);
9039                 ctl_done((union ctl_io *)ctsio);
9040                 return (retval);
9041                 break; /* NOTREACHED */
9042         }
9043
9044         alloc_len = scsi_4btoul(cdb->length);
9045         /*
9046          * The initiator has to allocate at least 16 bytes for this request,
9047          * so he can at least get the header and the first LUN.  Otherwise
9048          * we reject the request (per SPC-3 rev 14, section 6.21).
9049          */
9050         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9051             sizeof(struct scsi_report_luns_lundata))) {
9052                 ctl_set_invalid_field(ctsio,
9053                                       /*sks_valid*/ 1,
9054                                       /*command*/ 1,
9055                                       /*field*/ 6,
9056                                       /*bit_valid*/ 0,
9057                                       /*bit*/ 0);
9058                 ctl_done((union ctl_io *)ctsio);
9059                 return (retval);
9060         }
9061
9062         request_lun = (struct ctl_lun *)
9063                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9064
9065         lun_datalen = sizeof(*lun_data) +
9066                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9067
9068         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9069         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9070         ctsio->kern_sg_entries = 0;
9071
9072         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9073
9074         mtx_lock(&control_softc->ctl_lock);
9075         for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9076                 lun_id = targ_lun_id;
9077                 if (ctsio->io_hdr.nexus.lun_map_fn != NULL)
9078                         lun_id = ctsio->io_hdr.nexus.lun_map_fn(ctsio->io_hdr.nexus.lun_map_arg, lun_id);
9079                 if (lun_id >= CTL_MAX_LUNS)
9080                         continue;
9081                 lun = control_softc->ctl_luns[lun_id];
9082                 if (lun == NULL)
9083                         continue;
9084
9085                 if (targ_lun_id <= 0xff) {
9086                         /*
9087                          * Peripheral addressing method, bus number 0.
9088                          */
9089                         lun_data->luns[num_filled].lundata[0] =
9090                                 RPL_LUNDATA_ATYP_PERIPH;
9091                         lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9092                         num_filled++;
9093                 } else if (targ_lun_id <= 0x3fff) {
9094                         /*
9095                          * Flat addressing method.
9096                          */
9097                         lun_data->luns[num_filled].lundata[0] =
9098                                 RPL_LUNDATA_ATYP_FLAT |
9099                                 (targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9100 #ifdef OLDCTLHEADERS
9101                                 (SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9102                                 (targ_lun_id & SRLD_BUS_LUN_MASK);
9103 #endif
9104                         lun_data->luns[num_filled].lundata[1] =
9105 #ifdef OLDCTLHEADERS
9106                                 targ_lun_id >> SRLD_BUS_LUN_BITS;
9107 #endif
9108                                 targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9109                         num_filled++;
9110                 } else {
9111                         printf("ctl_report_luns: bogus LUN number %jd, "
9112                                "skipping\n", (intmax_t)targ_lun_id);
9113                 }
9114                 /*
9115                  * According to SPC-3, rev 14 section 6.21:
9116                  *
9117                  * "The execution of a REPORT LUNS command to any valid and
9118                  * installed logical unit shall clear the REPORTED LUNS DATA
9119                  * HAS CHANGED unit attention condition for all logical
9120                  * units of that target with respect to the requesting
9121                  * initiator. A valid and installed logical unit is one
9122                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9123                  * INQUIRY data (see 6.4.2)."
9124                  *
9125                  * If request_lun is NULL, the LUN this report luns command
9126                  * was issued to is either disabled or doesn't exist. In that
9127                  * case, we shouldn't clear any pending lun change unit
9128                  * attention.
9129                  */
9130                 if (request_lun != NULL) {
9131                         mtx_lock(&lun->lun_lock);
9132                         lun->pending_sense[initidx].ua_pending &=
9133                                 ~CTL_UA_LUN_CHANGE;
9134                         mtx_unlock(&lun->lun_lock);
9135                 }
9136         }
9137         mtx_unlock(&control_softc->ctl_lock);
9138
9139         /*
9140          * It's quite possible that we've returned fewer LUNs than we allocated
9141          * space for.  Trim it.
9142          */
9143         lun_datalen = sizeof(*lun_data) +
9144                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9145
9146         if (lun_datalen < alloc_len) {
9147                 ctsio->residual = alloc_len - lun_datalen;
9148                 ctsio->kern_data_len = lun_datalen;
9149                 ctsio->kern_total_len = lun_datalen;
9150         } else {
9151                 ctsio->residual = 0;
9152                 ctsio->kern_data_len = alloc_len;
9153                 ctsio->kern_total_len = alloc_len;
9154         }
9155         ctsio->kern_data_resid = 0;
9156         ctsio->kern_rel_offset = 0;
9157         ctsio->kern_sg_entries = 0;
9158
9159         /*
9160          * We set this to the actual data length, regardless of how much
9161          * space we actually have to return results.  If the user looks at
9162          * this value, he'll know whether or not he allocated enough space
9163          * and reissue the command if necessary.  We don't support well
9164          * known logical units, so if the user asks for that, return none.
9165          */
9166         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9167
9168         /*
9169          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9170          * this request.
9171          */
9172         ctsio->scsi_status = SCSI_STATUS_OK;
9173
9174         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9175         ctsio->be_move_done = ctl_config_move_done;
9176         ctl_datamove((union ctl_io *)ctsio);
9177
9178         return (retval);
9179 }
9180
9181 int
9182 ctl_request_sense(struct ctl_scsiio *ctsio)
9183 {
9184         struct scsi_request_sense *cdb;
9185         struct scsi_sense_data *sense_ptr;
9186         struct ctl_lun *lun;
9187         uint32_t initidx;
9188         int have_error;
9189         scsi_sense_data_type sense_format;
9190
9191         cdb = (struct scsi_request_sense *)ctsio->cdb;
9192
9193         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9194
9195         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9196
9197         /*
9198          * Determine which sense format the user wants.
9199          */
9200         if (cdb->byte2 & SRS_DESC)
9201                 sense_format = SSD_TYPE_DESC;
9202         else
9203                 sense_format = SSD_TYPE_FIXED;
9204
9205         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9206         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9207         ctsio->kern_sg_entries = 0;
9208
9209         /*
9210          * struct scsi_sense_data, which is currently set to 256 bytes, is
9211          * larger than the largest allowed value for the length field in the
9212          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9213          */
9214         ctsio->residual = 0;
9215         ctsio->kern_data_len = cdb->length;
9216         ctsio->kern_total_len = cdb->length;
9217
9218         ctsio->kern_data_resid = 0;
9219         ctsio->kern_rel_offset = 0;
9220         ctsio->kern_sg_entries = 0;
9221
9222         /*
9223          * If we don't have a LUN, we don't have any pending sense.
9224          */
9225         if (lun == NULL)
9226                 goto no_sense;
9227
9228         have_error = 0;
9229         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9230         /*
9231          * Check for pending sense, and then for pending unit attentions.
9232          * Pending sense gets returned first, then pending unit attentions.
9233          */
9234         mtx_lock(&lun->lun_lock);
9235         if (ctl_is_set(lun->have_ca, initidx)) {
9236                 scsi_sense_data_type stored_format;
9237
9238                 /*
9239                  * Check to see which sense format was used for the stored
9240                  * sense data.
9241                  */
9242                 stored_format = scsi_sense_type(
9243                     &lun->pending_sense[initidx].sense);
9244
9245                 /*
9246                  * If the user requested a different sense format than the
9247                  * one we stored, then we need to convert it to the other
9248                  * format.  If we're going from descriptor to fixed format
9249                  * sense data, we may lose things in translation, depending
9250                  * on what options were used.
9251                  *
9252                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9253                  * for some reason we'll just copy it out as-is.
9254                  */
9255                 if ((stored_format == SSD_TYPE_FIXED)
9256                  && (sense_format == SSD_TYPE_DESC))
9257                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9258                             &lun->pending_sense[initidx].sense,
9259                             (struct scsi_sense_data_desc *)sense_ptr);
9260                 else if ((stored_format == SSD_TYPE_DESC)
9261                       && (sense_format == SSD_TYPE_FIXED))
9262                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9263                             &lun->pending_sense[initidx].sense,
9264                             (struct scsi_sense_data_fixed *)sense_ptr);
9265                 else
9266                         memcpy(sense_ptr, &lun->pending_sense[initidx].sense,
9267                                ctl_min(sizeof(*sense_ptr),
9268                                sizeof(lun->pending_sense[initidx].sense)));
9269
9270                 ctl_clear_mask(lun->have_ca, initidx);
9271                 have_error = 1;
9272         } else if (lun->pending_sense[initidx].ua_pending != CTL_UA_NONE) {
9273                 ctl_ua_type ua_type;
9274
9275                 ua_type = ctl_build_ua(lun->pending_sense[initidx].ua_pending,
9276                                        sense_ptr, sense_format);
9277                 if (ua_type != CTL_UA_NONE) {
9278                         have_error = 1;
9279                         /* We're reporting this UA, so clear it */
9280                         lun->pending_sense[initidx].ua_pending &= ~ua_type;
9281                 }
9282         }
9283         mtx_unlock(&lun->lun_lock);
9284
9285         /*
9286          * We already have a pending error, return it.
9287          */
9288         if (have_error != 0) {
9289                 /*
9290                  * We report the SCSI status as OK, since the status of the
9291                  * request sense command itself is OK.
9292                  */
9293                 ctsio->scsi_status = SCSI_STATUS_OK;
9294
9295                 /*
9296                  * We report 0 for the sense length, because we aren't doing
9297                  * autosense in this case.  We're reporting sense as
9298                  * parameter data.
9299                  */
9300                 ctsio->sense_len = 0;
9301                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9302                 ctsio->be_move_done = ctl_config_move_done;
9303                 ctl_datamove((union ctl_io *)ctsio);
9304
9305                 return (CTL_RETVAL_COMPLETE);
9306         }
9307
9308 no_sense:
9309
9310         /*
9311          * No sense information to report, so we report that everything is
9312          * okay.
9313          */
9314         ctl_set_sense_data(sense_ptr,
9315                            lun,
9316                            sense_format,
9317                            /*current_error*/ 1,
9318                            /*sense_key*/ SSD_KEY_NO_SENSE,
9319                            /*asc*/ 0x00,
9320                            /*ascq*/ 0x00,
9321                            SSD_ELEM_NONE);
9322
9323         ctsio->scsi_status = SCSI_STATUS_OK;
9324
9325         /*
9326          * We report 0 for the sense length, because we aren't doing
9327          * autosense in this case.  We're reporting sense as parameter data.
9328          */
9329         ctsio->sense_len = 0;
9330         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9331         ctsio->be_move_done = ctl_config_move_done;
9332         ctl_datamove((union ctl_io *)ctsio);
9333
9334         return (CTL_RETVAL_COMPLETE);
9335 }
9336
9337 int
9338 ctl_tur(struct ctl_scsiio *ctsio)
9339 {
9340         struct ctl_lun *lun;
9341
9342         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9343
9344         CTL_DEBUG_PRINT(("ctl_tur\n"));
9345
9346         if (lun == NULL)
9347                 return (EINVAL);
9348
9349         ctsio->scsi_status = SCSI_STATUS_OK;
9350         ctsio->io_hdr.status = CTL_SUCCESS;
9351
9352         ctl_done((union ctl_io *)ctsio);
9353
9354         return (CTL_RETVAL_COMPLETE);
9355 }
9356
9357 #ifdef notyet
9358 static int
9359 ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9360 {
9361
9362 }
9363 #endif
9364
9365 static int
9366 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9367 {
9368         struct scsi_vpd_supported_pages *pages;
9369         int sup_page_size;
9370         struct ctl_lun *lun;
9371
9372         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9373
9374         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9375             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9376         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9377         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9378         ctsio->kern_sg_entries = 0;
9379
9380         if (sup_page_size < alloc_len) {
9381                 ctsio->residual = alloc_len - sup_page_size;
9382                 ctsio->kern_data_len = sup_page_size;
9383                 ctsio->kern_total_len = sup_page_size;
9384         } else {
9385                 ctsio->residual = 0;
9386                 ctsio->kern_data_len = alloc_len;
9387                 ctsio->kern_total_len = alloc_len;
9388         }
9389         ctsio->kern_data_resid = 0;
9390         ctsio->kern_rel_offset = 0;
9391         ctsio->kern_sg_entries = 0;
9392
9393         /*
9394          * The control device is always connected.  The disk device, on the
9395          * other hand, may not be online all the time.  Need to change this
9396          * to figure out whether the disk device is actually online or not.
9397          */
9398         if (lun != NULL)
9399                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9400                                 lun->be_lun->lun_type;
9401         else
9402                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9403
9404         pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9405         /* Supported VPD pages */
9406         pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9407         /* Serial Number */
9408         pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9409         /* Device Identification */
9410         pages->page_list[2] = SVPD_DEVICE_ID;
9411         /* Block limits */
9412         pages->page_list[3] = SVPD_BLOCK_LIMITS;
9413         /* Logical Block Provisioning */
9414         pages->page_list[4] = SVPD_LBP;
9415
9416         ctsio->scsi_status = SCSI_STATUS_OK;
9417
9418         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9419         ctsio->be_move_done = ctl_config_move_done;
9420         ctl_datamove((union ctl_io *)ctsio);
9421
9422         return (CTL_RETVAL_COMPLETE);
9423 }
9424
9425 static int
9426 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9427 {
9428         struct scsi_vpd_unit_serial_number *sn_ptr;
9429         struct ctl_lun *lun;
9430
9431         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9432
9433         ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9434         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9435         ctsio->kern_sg_entries = 0;
9436
9437         if (sizeof(*sn_ptr) < alloc_len) {
9438                 ctsio->residual = alloc_len - sizeof(*sn_ptr);
9439                 ctsio->kern_data_len = sizeof(*sn_ptr);
9440                 ctsio->kern_total_len = sizeof(*sn_ptr);
9441         } else {
9442                 ctsio->residual = 0;
9443                 ctsio->kern_data_len = alloc_len;
9444                 ctsio->kern_total_len = alloc_len;
9445         }
9446         ctsio->kern_data_resid = 0;
9447         ctsio->kern_rel_offset = 0;
9448         ctsio->kern_sg_entries = 0;
9449
9450         /*
9451          * The control device is always connected.  The disk device, on the
9452          * other hand, may not be online all the time.  Need to change this
9453          * to figure out whether the disk device is actually online or not.
9454          */
9455         if (lun != NULL)
9456                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9457                                   lun->be_lun->lun_type;
9458         else
9459                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9460
9461         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9462         sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9463         /*
9464          * If we don't have a LUN, we just leave the serial number as
9465          * all spaces.
9466          */
9467         memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9468         if (lun != NULL) {
9469                 strncpy((char *)sn_ptr->serial_num,
9470                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9471         }
9472         ctsio->scsi_status = SCSI_STATUS_OK;
9473
9474         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9475         ctsio->be_move_done = ctl_config_move_done;
9476         ctl_datamove((union ctl_io *)ctsio);
9477
9478         return (CTL_RETVAL_COMPLETE);
9479 }
9480
9481
9482 static int
9483 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9484 {
9485         struct scsi_vpd_device_id *devid_ptr;
9486         struct scsi_vpd_id_descriptor *desc, *desc1;
9487         struct scsi_vpd_id_descriptor *desc2, *desc3; /* for types 4h and 5h */
9488         struct scsi_vpd_id_t10 *t10id;
9489         struct ctl_softc *ctl_softc;
9490         struct ctl_lun *lun;
9491         struct ctl_port *port;
9492         char *val;
9493         int data_len, devid_len;
9494
9495         ctl_softc = control_softc;
9496
9497         port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9498
9499         if (port->devid != NULL)
9500                 return ((port->devid)(ctsio, alloc_len));
9501
9502         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9503
9504         if (lun == NULL) {
9505                 devid_len = CTL_DEVID_MIN_LEN;
9506         } else {
9507                 devid_len = max(CTL_DEVID_MIN_LEN,
9508                     strnlen(lun->be_lun->device_id, CTL_DEVID_LEN));
9509         }
9510
9511         data_len = sizeof(struct scsi_vpd_device_id) +
9512                 sizeof(struct scsi_vpd_id_descriptor) +
9513                 sizeof(struct scsi_vpd_id_t10) + devid_len +
9514                 sizeof(struct scsi_vpd_id_descriptor) + CTL_WWPN_LEN +
9515                 sizeof(struct scsi_vpd_id_descriptor) +
9516                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9517                 sizeof(struct scsi_vpd_id_descriptor) +
9518                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9519
9520         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9521         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9522         ctsio->kern_sg_entries = 0;
9523
9524         if (data_len < alloc_len) {
9525                 ctsio->residual = alloc_len - data_len;
9526                 ctsio->kern_data_len = data_len;
9527                 ctsio->kern_total_len = data_len;
9528         } else {
9529                 ctsio->residual = 0;
9530                 ctsio->kern_data_len = alloc_len;
9531                 ctsio->kern_total_len = alloc_len;
9532         }
9533         ctsio->kern_data_resid = 0;
9534         ctsio->kern_rel_offset = 0;
9535         ctsio->kern_sg_entries = 0;
9536
9537         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9538         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
9539         desc1 = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9540                 sizeof(struct scsi_vpd_id_t10) + devid_len);
9541         desc2 = (struct scsi_vpd_id_descriptor *)(&desc1->identifier[0] +
9542                   CTL_WWPN_LEN);
9543         desc3 = (struct scsi_vpd_id_descriptor *)(&desc2->identifier[0] +
9544                  sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9545
9546         /*
9547          * The control device is always connected.  The disk device, on the
9548          * other hand, may not be online all the time.
9549          */
9550         if (lun != NULL)
9551                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9552                                      lun->be_lun->lun_type;
9553         else
9554                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9555
9556         devid_ptr->page_code = SVPD_DEVICE_ID;
9557
9558         scsi_ulto2b(data_len - 4, devid_ptr->length);
9559
9560         /*
9561          * For Fibre channel,
9562          */
9563         if (port->port_type == CTL_PORT_FC)
9564         {
9565                 desc->proto_codeset = (SCSI_PROTO_FC << 4) |
9566                                       SVPD_ID_CODESET_ASCII;
9567                 desc1->proto_codeset = (SCSI_PROTO_FC << 4) |
9568                               SVPD_ID_CODESET_BINARY;
9569         }
9570         else
9571         {
9572                 desc->proto_codeset = (SCSI_PROTO_SPI << 4) |
9573                                       SVPD_ID_CODESET_ASCII;
9574                 desc1->proto_codeset = (SCSI_PROTO_SPI << 4) |
9575                               SVPD_ID_CODESET_BINARY;
9576         }
9577         desc2->proto_codeset = desc3->proto_codeset = desc1->proto_codeset;
9578
9579         /*
9580          * We're using a LUN association here.  i.e., this device ID is a
9581          * per-LUN identifier.
9582          */
9583         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
9584         desc->length = sizeof(*t10id) + devid_len;
9585         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
9586             "vendor")) == NULL) {
9587                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
9588         } else {
9589                 memset(t10id->vendor, ' ', sizeof(t10id->vendor));
9590                 strncpy(t10id->vendor, val,
9591                     min(sizeof(t10id->vendor), strlen(val)));
9592         }
9593
9594         /*
9595          * desc1 is for the WWPN which is a port asscociation.
9596          */
9597         desc1->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT | SVPD_ID_TYPE_NAA;
9598         desc1->length = CTL_WWPN_LEN;
9599         /* XXX Call Reggie's get_WWNN func here then add port # to the end */
9600         /* For testing just create the WWPN */
9601 #if 0
9602         ddb_GetWWNN((char *)desc1->identifier);
9603
9604         /* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9605         /* This is so Copancontrol will return something sane */
9606         if (ctsio->io_hdr.nexus.targ_port!=0 &&
9607             ctsio->io_hdr.nexus.targ_port!=8)
9608                 desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port-1;
9609         else
9610                 desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port;
9611 #endif
9612
9613         be64enc(desc1->identifier, port->wwpn);
9614
9615         /*
9616          * desc2 is for the Relative Target Port(type 4h) identifier
9617          */
9618         desc2->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9619                          | SVPD_ID_TYPE_RELTARG;
9620         desc2->length = 4;
9621 //#if 0
9622         /* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9623         /* This is so Copancontrol will return something sane */
9624         if (ctsio->io_hdr.nexus.targ_port!=0 &&
9625             ctsio->io_hdr.nexus.targ_port!=8)
9626                 desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port - 1;
9627         else
9628                 desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port;
9629 //#endif
9630
9631         /*
9632          * desc3 is for the Target Port Group(type 5h) identifier
9633          */
9634         desc3->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9635                          | SVPD_ID_TYPE_TPORTGRP;
9636         desc3->length = 4;
9637         if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS || ctl_is_single)
9638                 desc3->identifier[3] = 1;
9639         else
9640                 desc3->identifier[3] = 2;
9641
9642         /*
9643          * If we've actually got a backend, copy the device id from the
9644          * per-LUN data.  Otherwise, set it to all spaces.
9645          */
9646         if (lun != NULL) {
9647                 /*
9648                  * Copy the backend's LUN ID.
9649                  */
9650                 strncpy((char *)t10id->vendor_spec_id,
9651                         (char *)lun->be_lun->device_id, devid_len);
9652         } else {
9653                 /*
9654                  * No backend, set this to spaces.
9655                  */
9656                 memset(t10id->vendor_spec_id, 0x20, devid_len);
9657         }
9658
9659         ctsio->scsi_status = SCSI_STATUS_OK;
9660
9661         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9662         ctsio->be_move_done = ctl_config_move_done;
9663         ctl_datamove((union ctl_io *)ctsio);
9664
9665         return (CTL_RETVAL_COMPLETE);
9666 }
9667
9668 static int
9669 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9670 {
9671         struct scsi_vpd_block_limits *bl_ptr;
9672         struct ctl_lun *lun;
9673         int bs;
9674
9675         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9676         bs = lun->be_lun->blocksize;
9677
9678         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9679         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9680         ctsio->kern_sg_entries = 0;
9681
9682         if (sizeof(*bl_ptr) < alloc_len) {
9683                 ctsio->residual = alloc_len - sizeof(*bl_ptr);
9684                 ctsio->kern_data_len = sizeof(*bl_ptr);
9685                 ctsio->kern_total_len = sizeof(*bl_ptr);
9686         } else {
9687                 ctsio->residual = 0;
9688                 ctsio->kern_data_len = alloc_len;
9689                 ctsio->kern_total_len = alloc_len;
9690         }
9691         ctsio->kern_data_resid = 0;
9692         ctsio->kern_rel_offset = 0;
9693         ctsio->kern_sg_entries = 0;
9694
9695         /*
9696          * The control device is always connected.  The disk device, on the
9697          * other hand, may not be online all the time.  Need to change this
9698          * to figure out whether the disk device is actually online or not.
9699          */
9700         if (lun != NULL)
9701                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9702                                   lun->be_lun->lun_type;
9703         else
9704                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9705
9706         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9707         scsi_ulto2b(sizeof(*bl_ptr), bl_ptr->page_length);
9708         bl_ptr->max_cmp_write_len = 0xff;
9709         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9710         scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
9711         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9712                 scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
9713                 scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
9714         }
9715         scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
9716
9717         ctsio->scsi_status = SCSI_STATUS_OK;
9718         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9719         ctsio->be_move_done = ctl_config_move_done;
9720         ctl_datamove((union ctl_io *)ctsio);
9721
9722         return (CTL_RETVAL_COMPLETE);
9723 }
9724
9725 static int
9726 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9727 {
9728         struct scsi_vpd_logical_block_prov *lbp_ptr;
9729         struct ctl_lun *lun;
9730         int bs;
9731
9732         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9733         bs = lun->be_lun->blocksize;
9734
9735         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9736         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9737         ctsio->kern_sg_entries = 0;
9738
9739         if (sizeof(*lbp_ptr) < alloc_len) {
9740                 ctsio->residual = alloc_len - sizeof(*lbp_ptr);
9741                 ctsio->kern_data_len = sizeof(*lbp_ptr);
9742                 ctsio->kern_total_len = sizeof(*lbp_ptr);
9743         } else {
9744                 ctsio->residual = 0;
9745                 ctsio->kern_data_len = alloc_len;
9746                 ctsio->kern_total_len = alloc_len;
9747         }
9748         ctsio->kern_data_resid = 0;
9749         ctsio->kern_rel_offset = 0;
9750         ctsio->kern_sg_entries = 0;
9751
9752         /*
9753          * The control device is always connected.  The disk device, on the
9754          * other hand, may not be online all the time.  Need to change this
9755          * to figure out whether the disk device is actually online or not.
9756          */
9757         if (lun != NULL)
9758                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9759                                   lun->be_lun->lun_type;
9760         else
9761                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9762
9763         lbp_ptr->page_code = SVPD_LBP;
9764         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
9765                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 | SVPD_LBP_WS10;
9766
9767         ctsio->scsi_status = SCSI_STATUS_OK;
9768         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9769         ctsio->be_move_done = ctl_config_move_done;
9770         ctl_datamove((union ctl_io *)ctsio);
9771
9772         return (CTL_RETVAL_COMPLETE);
9773 }
9774
9775 static int
9776 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9777 {
9778         struct scsi_inquiry *cdb;
9779         struct ctl_lun *lun;
9780         int alloc_len, retval;
9781
9782         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9783         cdb = (struct scsi_inquiry *)ctsio->cdb;
9784
9785         retval = CTL_RETVAL_COMPLETE;
9786
9787         alloc_len = scsi_2btoul(cdb->length);
9788
9789         switch (cdb->page_code) {
9790         case SVPD_SUPPORTED_PAGES:
9791                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9792                 break;
9793         case SVPD_UNIT_SERIAL_NUMBER:
9794                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9795                 break;
9796         case SVPD_DEVICE_ID:
9797                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9798                 break;
9799         case SVPD_BLOCK_LIMITS:
9800                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
9801                 break;
9802         case SVPD_LBP:
9803                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
9804                 break;
9805         default:
9806                 ctl_set_invalid_field(ctsio,
9807                                       /*sks_valid*/ 1,
9808                                       /*command*/ 1,
9809                                       /*field*/ 2,
9810                                       /*bit_valid*/ 0,
9811                                       /*bit*/ 0);
9812                 ctl_done((union ctl_io *)ctsio);
9813                 retval = CTL_RETVAL_COMPLETE;
9814                 break;
9815         }
9816
9817         return (retval);
9818 }
9819
9820 static int
9821 ctl_inquiry_std(struct ctl_scsiio *ctsio)
9822 {
9823         struct scsi_inquiry_data *inq_ptr;
9824         struct scsi_inquiry *cdb;
9825         struct ctl_softc *ctl_softc;
9826         struct ctl_lun *lun;
9827         char *val;
9828         uint32_t alloc_len;
9829         int is_fc;
9830
9831         ctl_softc = control_softc;
9832
9833         /*
9834          * Figure out whether we're talking to a Fibre Channel port or not.
9835          * We treat the ioctl front end, and any SCSI adapters, as packetized
9836          * SCSI front ends.
9837          */
9838         if (ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type !=
9839             CTL_PORT_FC)
9840                 is_fc = 0;
9841         else
9842                 is_fc = 1;
9843
9844         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9845         cdb = (struct scsi_inquiry *)ctsio->cdb;
9846         alloc_len = scsi_2btoul(cdb->length);
9847
9848         /*
9849          * We malloc the full inquiry data size here and fill it
9850          * in.  If the user only asks for less, we'll give him
9851          * that much.
9852          */
9853         ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
9854         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
9855         ctsio->kern_sg_entries = 0;
9856         ctsio->kern_data_resid = 0;
9857         ctsio->kern_rel_offset = 0;
9858
9859         if (sizeof(*inq_ptr) < alloc_len) {
9860                 ctsio->residual = alloc_len - sizeof(*inq_ptr);
9861                 ctsio->kern_data_len = sizeof(*inq_ptr);
9862                 ctsio->kern_total_len = sizeof(*inq_ptr);
9863         } else {
9864                 ctsio->residual = 0;
9865                 ctsio->kern_data_len = alloc_len;
9866                 ctsio->kern_total_len = alloc_len;
9867         }
9868
9869         /*
9870          * If we have a LUN configured, report it as connected.  Otherwise,
9871          * report that it is offline or no device is supported, depending
9872          * on the value of inquiry_pq_no_lun.
9873          *
9874          * According to the spec (SPC-4 r34), the peripheral qualifier
9875          * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
9876          *
9877          * "A peripheral device having the specified peripheral device type
9878          * is not connected to this logical unit. However, the device
9879          * server is capable of supporting the specified peripheral device
9880          * type on this logical unit."
9881          *
9882          * According to the same spec, the peripheral qualifier
9883          * SID_QUAL_BAD_LU (011b) is used in this scenario:
9884          *
9885          * "The device server is not capable of supporting a peripheral
9886          * device on this logical unit. For this peripheral qualifier the
9887          * peripheral device type shall be set to 1Fh. All other peripheral
9888          * device type values are reserved for this peripheral qualifier."
9889          *
9890          * Given the text, it would seem that we probably want to report that
9891          * the LUN is offline here.  There is no LUN connected, but we can
9892          * support a LUN at the given LUN number.
9893          *
9894          * In the real world, though, it sounds like things are a little
9895          * different:
9896          *
9897          * - Linux, when presented with a LUN with the offline peripheral
9898          *   qualifier, will create an sg driver instance for it.  So when
9899          *   you attach it to CTL, you wind up with a ton of sg driver
9900          *   instances.  (One for every LUN that Linux bothered to probe.)
9901          *   Linux does this despite the fact that it issues a REPORT LUNs
9902          *   to LUN 0 to get the inventory of supported LUNs.
9903          *
9904          * - There is other anecdotal evidence (from Emulex folks) about
9905          *   arrays that use the offline peripheral qualifier for LUNs that
9906          *   are on the "passive" path in an active/passive array.
9907          *
9908          * So the solution is provide a hopefully reasonable default
9909          * (return bad/no LUN) and allow the user to change the behavior
9910          * with a tunable/sysctl variable.
9911          */
9912         if (lun != NULL)
9913                 inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9914                                   lun->be_lun->lun_type;
9915         else if (ctl_softc->inquiry_pq_no_lun == 0)
9916                 inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9917         else
9918                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
9919
9920         /* RMB in byte 2 is 0 */
9921         inq_ptr->version = SCSI_REV_SPC3;
9922
9923         /*
9924          * According to SAM-3, even if a device only supports a single
9925          * level of LUN addressing, it should still set the HISUP bit:
9926          *
9927          * 4.9.1 Logical unit numbers overview
9928          *
9929          * All logical unit number formats described in this standard are
9930          * hierarchical in structure even when only a single level in that
9931          * hierarchy is used. The HISUP bit shall be set to one in the
9932          * standard INQUIRY data (see SPC-2) when any logical unit number
9933          * format described in this standard is used.  Non-hierarchical
9934          * formats are outside the scope of this standard.
9935          *
9936          * Therefore we set the HiSup bit here.
9937          *
9938          * The reponse format is 2, per SPC-3.
9939          */
9940         inq_ptr->response_format = SID_HiSup | 2;
9941
9942         inq_ptr->additional_length = sizeof(*inq_ptr) - 4;
9943         CTL_DEBUG_PRINT(("additional_length = %d\n",
9944                          inq_ptr->additional_length));
9945
9946         inq_ptr->spc3_flags = SPC3_SID_TPGS_IMPLICIT;
9947         /* 16 bit addressing */
9948         if (is_fc == 0)
9949                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
9950         /* XXX set the SID_MultiP bit here if we're actually going to
9951            respond on multiple ports */
9952         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
9953
9954         /* 16 bit data bus, synchronous transfers */
9955         /* XXX these flags don't apply for FC */
9956         if (is_fc == 0)
9957                 inq_ptr->flags = SID_WBus16 | SID_Sync;
9958         /*
9959          * XXX KDM do we want to support tagged queueing on the control
9960          * device at all?
9961          */
9962         if ((lun == NULL)
9963          || (lun->be_lun->lun_type != T_PROCESSOR))
9964                 inq_ptr->flags |= SID_CmdQue;
9965         /*
9966          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
9967          * We have 8 bytes for the vendor name, and 16 bytes for the device
9968          * name and 4 bytes for the revision.
9969          */
9970         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
9971             "vendor")) == NULL) {
9972                 strcpy(inq_ptr->vendor, CTL_VENDOR);
9973         } else {
9974                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
9975                 strncpy(inq_ptr->vendor, val,
9976                     min(sizeof(inq_ptr->vendor), strlen(val)));
9977         }
9978         if (lun == NULL) {
9979                 strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
9980         } else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
9981                 switch (lun->be_lun->lun_type) {
9982                 case T_DIRECT:
9983                         strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
9984                         break;
9985                 case T_PROCESSOR:
9986                         strcpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT);
9987                         break;
9988                 default:
9989                         strcpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT);
9990                         break;
9991                 }
9992         } else {
9993                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
9994                 strncpy(inq_ptr->product, val,
9995                     min(sizeof(inq_ptr->product), strlen(val)));
9996         }
9997
9998         /*
9999          * XXX make this a macro somewhere so it automatically gets
10000          * incremented when we make changes.
10001          */
10002         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10003             "revision")) == NULL) {
10004                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10005         } else {
10006                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10007                 strncpy(inq_ptr->revision, val,
10008                     min(sizeof(inq_ptr->revision), strlen(val)));
10009         }
10010
10011         /*
10012          * For parallel SCSI, we support double transition and single
10013          * transition clocking.  We also support QAS (Quick Arbitration
10014          * and Selection) and Information Unit transfers on both the
10015          * control and array devices.
10016          */
10017         if (is_fc == 0)
10018                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10019                                     SID_SPI_IUS;
10020
10021         /* SAM-3 */
10022         scsi_ulto2b(0x0060, inq_ptr->version1);
10023         /* SPC-3 (no version claimed) XXX should we claim a version? */
10024         scsi_ulto2b(0x0300, inq_ptr->version2);
10025         if (is_fc) {
10026                 /* FCP-2 ANSI INCITS.350:2003 */
10027                 scsi_ulto2b(0x0917, inq_ptr->version3);
10028         } else {
10029                 /* SPI-4 ANSI INCITS.362:200x */
10030                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10031         }
10032
10033         if (lun == NULL) {
10034                 /* SBC-2 (no version claimed) XXX should we claim a version? */
10035                 scsi_ulto2b(0x0320, inq_ptr->version4);
10036         } else {
10037                 switch (lun->be_lun->lun_type) {
10038                 case T_DIRECT:
10039                         /*
10040                          * SBC-2 (no version claimed) XXX should we claim a
10041                          * version?
10042                          */
10043                         scsi_ulto2b(0x0320, inq_ptr->version4);
10044                         break;
10045                 case T_PROCESSOR:
10046                 default:
10047                         break;
10048                 }
10049         }
10050
10051         ctsio->scsi_status = SCSI_STATUS_OK;
10052         if (ctsio->kern_data_len > 0) {
10053                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10054                 ctsio->be_move_done = ctl_config_move_done;
10055                 ctl_datamove((union ctl_io *)ctsio);
10056         } else {
10057                 ctsio->io_hdr.status = CTL_SUCCESS;
10058                 ctl_done((union ctl_io *)ctsio);
10059         }
10060
10061         return (CTL_RETVAL_COMPLETE);
10062 }
10063
10064 int
10065 ctl_inquiry(struct ctl_scsiio *ctsio)
10066 {
10067         struct scsi_inquiry *cdb;
10068         int retval;
10069
10070         cdb = (struct scsi_inquiry *)ctsio->cdb;
10071
10072         retval = 0;
10073
10074         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10075
10076         /*
10077          * Right now, we don't support the CmdDt inquiry information.
10078          * This would be nice to support in the future.  When we do
10079          * support it, we should change this test so that it checks to make
10080          * sure SI_EVPD and SI_CMDDT aren't both set at the same time.
10081          */
10082 #ifdef notyet
10083         if (((cdb->byte2 & SI_EVPD)
10084          && (cdb->byte2 & SI_CMDDT)))
10085 #endif
10086         if (cdb->byte2 & SI_CMDDT) {
10087                 /*
10088                  * Point to the SI_CMDDT bit.  We might change this
10089                  * when we support SI_CMDDT, but since both bits would be
10090                  * "wrong", this should probably just stay as-is then.
10091                  */
10092                 ctl_set_invalid_field(ctsio,
10093                                       /*sks_valid*/ 1,
10094                                       /*command*/ 1,
10095                                       /*field*/ 1,
10096                                       /*bit_valid*/ 1,
10097                                       /*bit*/ 1);
10098                 ctl_done((union ctl_io *)ctsio);
10099                 return (CTL_RETVAL_COMPLETE);
10100         }
10101         if (cdb->byte2 & SI_EVPD)
10102                 retval = ctl_inquiry_evpd(ctsio);
10103 #ifdef notyet
10104         else if (cdb->byte2 & SI_CMDDT)
10105                 retval = ctl_inquiry_cmddt(ctsio);
10106 #endif
10107         else
10108                 retval = ctl_inquiry_std(ctsio);
10109
10110         return (retval);
10111 }
10112
10113 /*
10114  * For known CDB types, parse the LBA and length.
10115  */
10116 static int
10117 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len)
10118 {
10119         if (io->io_hdr.io_type != CTL_IO_SCSI)
10120                 return (1);
10121
10122         switch (io->scsiio.cdb[0]) {
10123         case COMPARE_AND_WRITE: {
10124                 struct scsi_compare_and_write *cdb;
10125
10126                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10127
10128                 *lba = scsi_8btou64(cdb->addr);
10129                 *len = cdb->length;
10130                 break;
10131         }
10132         case READ_6:
10133         case WRITE_6: {
10134                 struct scsi_rw_6 *cdb;
10135
10136                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10137
10138                 *lba = scsi_3btoul(cdb->addr);
10139                 /* only 5 bits are valid in the most significant address byte */
10140                 *lba &= 0x1fffff;
10141                 *len = cdb->length;
10142                 break;
10143         }
10144         case READ_10:
10145         case WRITE_10: {
10146                 struct scsi_rw_10 *cdb;
10147
10148                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10149
10150                 *lba = scsi_4btoul(cdb->addr);
10151                 *len = scsi_2btoul(cdb->length);
10152                 break;
10153         }
10154         case WRITE_VERIFY_10: {
10155                 struct scsi_write_verify_10 *cdb;
10156
10157                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10158
10159                 *lba = scsi_4btoul(cdb->addr);
10160                 *len = scsi_2btoul(cdb->length);
10161                 break;
10162         }
10163         case READ_12:
10164         case WRITE_12: {
10165                 struct scsi_rw_12 *cdb;
10166
10167                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10168
10169                 *lba = scsi_4btoul(cdb->addr);
10170                 *len = scsi_4btoul(cdb->length);
10171                 break;
10172         }
10173         case WRITE_VERIFY_12: {
10174                 struct scsi_write_verify_12 *cdb;
10175
10176                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10177
10178                 *lba = scsi_4btoul(cdb->addr);
10179                 *len = scsi_4btoul(cdb->length);
10180                 break;
10181         }
10182         case READ_16:
10183         case WRITE_16: {
10184                 struct scsi_rw_16 *cdb;
10185
10186                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10187
10188                 *lba = scsi_8btou64(cdb->addr);
10189                 *len = scsi_4btoul(cdb->length);
10190                 break;
10191         }
10192         case WRITE_VERIFY_16: {
10193                 struct scsi_write_verify_16 *cdb;
10194
10195                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10196
10197
10198                 *lba = scsi_8btou64(cdb->addr);
10199                 *len = scsi_4btoul(cdb->length);
10200                 break;
10201         }
10202         case WRITE_SAME_10: {
10203                 struct scsi_write_same_10 *cdb;
10204
10205                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10206
10207                 *lba = scsi_4btoul(cdb->addr);
10208                 *len = scsi_2btoul(cdb->length);
10209                 break;
10210         }
10211         case WRITE_SAME_16: {
10212                 struct scsi_write_same_16 *cdb;
10213
10214                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10215
10216                 *lba = scsi_8btou64(cdb->addr);
10217                 *len = scsi_4btoul(cdb->length);
10218                 break;
10219         }
10220         case VERIFY_10: {
10221                 struct scsi_verify_10 *cdb;
10222
10223                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10224
10225                 *lba = scsi_4btoul(cdb->addr);
10226                 *len = scsi_2btoul(cdb->length);
10227                 break;
10228         }
10229         case VERIFY_12: {
10230                 struct scsi_verify_12 *cdb;
10231
10232                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10233
10234                 *lba = scsi_4btoul(cdb->addr);
10235                 *len = scsi_4btoul(cdb->length);
10236                 break;
10237         }
10238         case VERIFY_16: {
10239                 struct scsi_verify_16 *cdb;
10240
10241                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10242
10243                 *lba = scsi_8btou64(cdb->addr);
10244                 *len = scsi_4btoul(cdb->length);
10245                 break;
10246         }
10247         default:
10248                 return (1);
10249                 break; /* NOTREACHED */
10250         }
10251
10252         return (0);
10253 }
10254
10255 static ctl_action
10256 ctl_extent_check_lba(uint64_t lba1, uint32_t len1, uint64_t lba2, uint32_t len2)
10257 {
10258         uint64_t endlba1, endlba2;
10259
10260         endlba1 = lba1 + len1 - 1;
10261         endlba2 = lba2 + len2 - 1;
10262
10263         if ((endlba1 < lba2)
10264          || (endlba2 < lba1))
10265                 return (CTL_ACTION_PASS);
10266         else
10267                 return (CTL_ACTION_BLOCK);
10268 }
10269
10270 static ctl_action
10271 ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10272 {
10273         uint64_t lba1, lba2;
10274         uint32_t len1, len2;
10275         int retval;
10276
10277         retval = ctl_get_lba_len(io1, &lba1, &len1);
10278         if (retval != 0)
10279                 return (CTL_ACTION_ERROR);
10280
10281         retval = ctl_get_lba_len(io2, &lba2, &len2);
10282         if (retval != 0)
10283                 return (CTL_ACTION_ERROR);
10284
10285         return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10286 }
10287
10288 static ctl_action
10289 ctl_check_for_blockage(union ctl_io *pending_io, union ctl_io *ooa_io)
10290 {
10291         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10292         ctl_serialize_action *serialize_row;
10293
10294         /*
10295          * The initiator attempted multiple untagged commands at the same
10296          * time.  Can't do that.
10297          */
10298         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10299          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10300          && ((pending_io->io_hdr.nexus.targ_port ==
10301               ooa_io->io_hdr.nexus.targ_port)
10302           && (pending_io->io_hdr.nexus.initid.id ==
10303               ooa_io->io_hdr.nexus.initid.id))
10304          && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10305                 return (CTL_ACTION_OVERLAP);
10306
10307         /*
10308          * The initiator attempted to send multiple tagged commands with
10309          * the same ID.  (It's fine if different initiators have the same
10310          * tag ID.)
10311          *
10312          * Even if all of those conditions are true, we don't kill the I/O
10313          * if the command ahead of us has been aborted.  We won't end up
10314          * sending it to the FETD, and it's perfectly legal to resend a
10315          * command with the same tag number as long as the previous
10316          * instance of this tag number has been aborted somehow.
10317          */
10318         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10319          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10320          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10321          && ((pending_io->io_hdr.nexus.targ_port ==
10322               ooa_io->io_hdr.nexus.targ_port)
10323           && (pending_io->io_hdr.nexus.initid.id ==
10324               ooa_io->io_hdr.nexus.initid.id))
10325          && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10326                 return (CTL_ACTION_OVERLAP_TAG);
10327
10328         /*
10329          * If we get a head of queue tag, SAM-3 says that we should
10330          * immediately execute it.
10331          *
10332          * What happens if this command would normally block for some other
10333          * reason?  e.g. a request sense with a head of queue tag
10334          * immediately after a write.  Normally that would block, but this
10335          * will result in its getting executed immediately...
10336          *
10337          * We currently return "pass" instead of "skip", so we'll end up
10338          * going through the rest of the queue to check for overlapped tags.
10339          *
10340          * XXX KDM check for other types of blockage first??
10341          */
10342         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10343                 return (CTL_ACTION_PASS);
10344
10345         /*
10346          * Ordered tags have to block until all items ahead of them
10347          * have completed.  If we get called with an ordered tag, we always
10348          * block, if something else is ahead of us in the queue.
10349          */
10350         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10351                 return (CTL_ACTION_BLOCK);
10352
10353         /*
10354          * Simple tags get blocked until all head of queue and ordered tags
10355          * ahead of them have completed.  I'm lumping untagged commands in
10356          * with simple tags here.  XXX KDM is that the right thing to do?
10357          */
10358         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10359           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10360          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10361           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10362                 return (CTL_ACTION_BLOCK);
10363
10364         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio);
10365         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio);
10366
10367         serialize_row = ctl_serialize_table[ooa_entry->seridx];
10368
10369         switch (serialize_row[pending_entry->seridx]) {
10370         case CTL_SER_BLOCK:
10371                 return (CTL_ACTION_BLOCK);
10372                 break; /* NOTREACHED */
10373         case CTL_SER_EXTENT:
10374                 return (ctl_extent_check(pending_io, ooa_io));
10375                 break; /* NOTREACHED */
10376         case CTL_SER_PASS:
10377                 return (CTL_ACTION_PASS);
10378                 break; /* NOTREACHED */
10379         case CTL_SER_SKIP:
10380                 return (CTL_ACTION_SKIP);
10381                 break;
10382         default:
10383                 panic("invalid serialization value %d",
10384                       serialize_row[pending_entry->seridx]);
10385                 break; /* NOTREACHED */
10386         }
10387
10388         return (CTL_ACTION_ERROR);
10389 }
10390
10391 /*
10392  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10393  * Assumptions:
10394  * - pending_io is generally either incoming, or on the blocked queue
10395  * - starting I/O is the I/O we want to start the check with.
10396  */
10397 static ctl_action
10398 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10399               union ctl_io *starting_io)
10400 {
10401         union ctl_io *ooa_io;
10402         ctl_action action;
10403
10404         mtx_assert(&lun->lun_lock, MA_OWNED);
10405
10406         /*
10407          * Run back along the OOA queue, starting with the current
10408          * blocked I/O and going through every I/O before it on the
10409          * queue.  If starting_io is NULL, we'll just end up returning
10410          * CTL_ACTION_PASS.
10411          */
10412         for (ooa_io = starting_io; ooa_io != NULL;
10413              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10414              ooa_links)){
10415
10416                 /*
10417                  * This routine just checks to see whether
10418                  * cur_blocked is blocked by ooa_io, which is ahead
10419                  * of it in the queue.  It doesn't queue/dequeue
10420                  * cur_blocked.
10421                  */
10422                 action = ctl_check_for_blockage(pending_io, ooa_io);
10423                 switch (action) {
10424                 case CTL_ACTION_BLOCK:
10425                 case CTL_ACTION_OVERLAP:
10426                 case CTL_ACTION_OVERLAP_TAG:
10427                 case CTL_ACTION_SKIP:
10428                 case CTL_ACTION_ERROR:
10429                         return (action);
10430                         break; /* NOTREACHED */
10431                 case CTL_ACTION_PASS:
10432                         break;
10433                 default:
10434                         panic("invalid action %d", action);
10435                         break;  /* NOTREACHED */
10436                 }
10437         }
10438
10439         return (CTL_ACTION_PASS);
10440 }
10441
10442 /*
10443  * Assumptions:
10444  * - An I/O has just completed, and has been removed from the per-LUN OOA
10445  *   queue, so some items on the blocked queue may now be unblocked.
10446  */
10447 static int
10448 ctl_check_blocked(struct ctl_lun *lun)
10449 {
10450         union ctl_io *cur_blocked, *next_blocked;
10451
10452         mtx_assert(&lun->lun_lock, MA_OWNED);
10453
10454         /*
10455          * Run forward from the head of the blocked queue, checking each
10456          * entry against the I/Os prior to it on the OOA queue to see if
10457          * there is still any blockage.
10458          *
10459          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10460          * with our removing a variable on it while it is traversing the
10461          * list.
10462          */
10463         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10464              cur_blocked != NULL; cur_blocked = next_blocked) {
10465                 union ctl_io *prev_ooa;
10466                 ctl_action action;
10467
10468                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10469                                                           blocked_links);
10470
10471                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10472                                                       ctl_ooaq, ooa_links);
10473
10474                 /*
10475                  * If cur_blocked happens to be the first item in the OOA
10476                  * queue now, prev_ooa will be NULL, and the action
10477                  * returned will just be CTL_ACTION_PASS.
10478                  */
10479                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10480
10481                 switch (action) {
10482                 case CTL_ACTION_BLOCK:
10483                         /* Nothing to do here, still blocked */
10484                         break;
10485                 case CTL_ACTION_OVERLAP:
10486                 case CTL_ACTION_OVERLAP_TAG:
10487                         /*
10488                          * This shouldn't happen!  In theory we've already
10489                          * checked this command for overlap...
10490                          */
10491                         break;
10492                 case CTL_ACTION_PASS:
10493                 case CTL_ACTION_SKIP: {
10494                         struct ctl_softc *softc;
10495                         const struct ctl_cmd_entry *entry;
10496                         uint32_t initidx;
10497                         int isc_retval;
10498
10499                         /*
10500                          * The skip case shouldn't happen, this transaction
10501                          * should have never made it onto the blocked queue.
10502                          */
10503                         /*
10504                          * This I/O is no longer blocked, we can remove it
10505                          * from the blocked queue.  Since this is a TAILQ
10506                          * (doubly linked list), we can do O(1) removals
10507                          * from any place on the list.
10508                          */
10509                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
10510                                      blocked_links);
10511                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10512
10513                         if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
10514                                 /*
10515                                  * Need to send IO back to original side to
10516                                  * run
10517                                  */
10518                                 union ctl_ha_msg msg_info;
10519
10520                                 msg_info.hdr.original_sc =
10521                                         cur_blocked->io_hdr.original_sc;
10522                                 msg_info.hdr.serializing_sc = cur_blocked;
10523                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
10524                                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10525                                      &msg_info, sizeof(msg_info), 0)) >
10526                                      CTL_HA_STATUS_SUCCESS) {
10527                                         printf("CTL:Check Blocked error from "
10528                                                "ctl_ha_msg_send %d\n",
10529                                                isc_retval);
10530                                 }
10531                                 break;
10532                         }
10533                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio);
10534                         softc = control_softc;
10535
10536                         initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
10537
10538                         /*
10539                          * Check this I/O for LUN state changes that may
10540                          * have happened while this command was blocked.
10541                          * The LUN state may have been changed by a command
10542                          * ahead of us in the queue, so we need to re-check
10543                          * for any states that can be caused by SCSI
10544                          * commands.
10545                          */
10546                         if (ctl_scsiio_lun_check(softc, lun, entry,
10547                                                  &cur_blocked->scsiio) == 0) {
10548                                 cur_blocked->io_hdr.flags |=
10549                                                       CTL_FLAG_IS_WAS_ON_RTR;
10550                                 ctl_enqueue_rtr(cur_blocked);
10551                         } else
10552                                 ctl_done(cur_blocked);
10553                         break;
10554                 }
10555                 default:
10556                         /*
10557                          * This probably shouldn't happen -- we shouldn't
10558                          * get CTL_ACTION_ERROR, or anything else.
10559                          */
10560                         break;
10561                 }
10562         }
10563
10564         return (CTL_RETVAL_COMPLETE);
10565 }
10566
10567 /*
10568  * This routine (with one exception) checks LUN flags that can be set by
10569  * commands ahead of us in the OOA queue.  These flags have to be checked
10570  * when a command initially comes in, and when we pull a command off the
10571  * blocked queue and are preparing to execute it.  The reason we have to
10572  * check these flags for commands on the blocked queue is that the LUN
10573  * state may have been changed by a command ahead of us while we're on the
10574  * blocked queue.
10575  *
10576  * Ordering is somewhat important with these checks, so please pay
10577  * careful attention to the placement of any new checks.
10578  */
10579 static int
10580 ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
10581     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
10582 {
10583         int retval;
10584
10585         retval = 0;
10586
10587         mtx_assert(&lun->lun_lock, MA_OWNED);
10588
10589         /*
10590          * If this shelf is a secondary shelf controller, we have to reject
10591          * any media access commands.
10592          */
10593 #if 0
10594         /* No longer needed for HA */
10595         if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
10596          && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
10597                 ctl_set_lun_standby(ctsio);
10598                 retval = 1;
10599                 goto bailout;
10600         }
10601 #endif
10602
10603         /*
10604          * Check for a reservation conflict.  If this command isn't allowed
10605          * even on reserved LUNs, and if this initiator isn't the one who
10606          * reserved us, reject the command with a reservation conflict.
10607          */
10608         if ((lun->flags & CTL_LUN_RESERVED)
10609          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
10610                 if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
10611                  || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
10612                  || (ctsio->io_hdr.nexus.targ_target.id !=
10613                      lun->rsv_nexus.targ_target.id)) {
10614                         ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10615                         ctsio->io_hdr.status = CTL_SCSI_ERROR;
10616                         retval = 1;
10617                         goto bailout;
10618                 }
10619         }
10620
10621         if ( (lun->flags & CTL_LUN_PR_RESERVED)
10622          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
10623                 uint32_t residx;
10624
10625                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
10626                 /*
10627                  * if we aren't registered or it's a res holder type
10628                  * reservation and this isn't the res holder then set a
10629                  * conflict.
10630                  * NOTE: Commands which might be allowed on write exclusive
10631                  * type reservations are checked in the particular command
10632                  * for a conflict. Read and SSU are the only ones.
10633                  */
10634                 if (!lun->per_res[residx].registered
10635                  || (residx != lun->pr_res_idx && lun->res_type < 4)) {
10636                         ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10637                         ctsio->io_hdr.status = CTL_SCSI_ERROR;
10638                         retval = 1;
10639                         goto bailout;
10640                 }
10641
10642         }
10643
10644         if ((lun->flags & CTL_LUN_OFFLINE)
10645          && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
10646                 ctl_set_lun_not_ready(ctsio);
10647                 retval = 1;
10648                 goto bailout;
10649         }
10650
10651         /*
10652          * If the LUN is stopped, see if this particular command is allowed
10653          * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
10654          */
10655         if ((lun->flags & CTL_LUN_STOPPED)
10656          && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
10657                 /* "Logical unit not ready, initializing cmd. required" */
10658                 ctl_set_lun_stopped(ctsio);
10659                 retval = 1;
10660                 goto bailout;
10661         }
10662
10663         if ((lun->flags & CTL_LUN_INOPERABLE)
10664          && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
10665                 /* "Medium format corrupted" */
10666                 ctl_set_medium_format_corrupted(ctsio);
10667                 retval = 1;
10668                 goto bailout;
10669         }
10670
10671 bailout:
10672         return (retval);
10673
10674 }
10675
10676 static void
10677 ctl_failover_io(union ctl_io *io, int have_lock)
10678 {
10679         ctl_set_busy(&io->scsiio);
10680         ctl_done(io);
10681 }
10682
10683 static void
10684 ctl_failover(void)
10685 {
10686         struct ctl_lun *lun;
10687         struct ctl_softc *ctl_softc;
10688         union ctl_io *next_io, *pending_io;
10689         union ctl_io *io;
10690         int lun_idx;
10691         int i;
10692
10693         ctl_softc = control_softc;
10694
10695         mtx_lock(&ctl_softc->ctl_lock);
10696         /*
10697          * Remove any cmds from the other SC from the rtr queue.  These
10698          * will obviously only be for LUNs for which we're the primary.
10699          * We can't send status or get/send data for these commands.
10700          * Since they haven't been executed yet, we can just remove them.
10701          * We'll either abort them or delete them below, depending on
10702          * which HA mode we're in.
10703          */
10704 #ifdef notyet
10705         mtx_lock(&ctl_softc->queue_lock);
10706         for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
10707              io != NULL; io = next_io) {
10708                 next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10709                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10710                         STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
10711                                       ctl_io_hdr, links);
10712         }
10713         mtx_unlock(&ctl_softc->queue_lock);
10714 #endif
10715
10716         for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
10717                 lun = ctl_softc->ctl_luns[lun_idx];
10718                 if (lun==NULL)
10719                         continue;
10720
10721                 /*
10722                  * Processor LUNs are primary on both sides.
10723                  * XXX will this always be true?
10724                  */
10725                 if (lun->be_lun->lun_type == T_PROCESSOR)
10726                         continue;
10727
10728                 if ((lun->flags & CTL_LUN_PRIMARY_SC)
10729                  && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10730                         printf("FAILOVER: primary lun %d\n", lun_idx);
10731                         /*
10732                          * Remove all commands from the other SC. First from the
10733                          * blocked queue then from the ooa queue. Once we have
10734                          * removed them. Call ctl_check_blocked to see if there
10735                          * is anything that can run.
10736                          */
10737                         for (io = (union ctl_io *)TAILQ_FIRST(
10738                              &lun->blocked_queue); io != NULL; io = next_io) {
10739
10740                                 next_io = (union ctl_io *)TAILQ_NEXT(
10741                                     &io->io_hdr, blocked_links);
10742
10743                                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10744                                         TAILQ_REMOVE(&lun->blocked_queue,
10745                                                      &io->io_hdr,blocked_links);
10746                                         io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10747                                         TAILQ_REMOVE(&lun->ooa_queue,
10748                                                      &io->io_hdr, ooa_links);
10749
10750                                         ctl_free_io(io);
10751                                 }
10752                         }
10753
10754                         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10755                              io != NULL; io = next_io) {
10756
10757                                 next_io = (union ctl_io *)TAILQ_NEXT(
10758                                     &io->io_hdr, ooa_links);
10759
10760                                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10761
10762                                         TAILQ_REMOVE(&lun->ooa_queue,
10763                                                 &io->io_hdr,
10764                                                 ooa_links);
10765
10766                                         ctl_free_io(io);
10767                                 }
10768                         }
10769                         ctl_check_blocked(lun);
10770                 } else if ((lun->flags & CTL_LUN_PRIMARY_SC)
10771                         && (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10772
10773                         printf("FAILOVER: primary lun %d\n", lun_idx);
10774                         /*
10775                          * Abort all commands from the other SC.  We can't
10776                          * send status back for them now.  These should get
10777                          * cleaned up when they are completed or come out
10778                          * for a datamove operation.
10779                          */
10780                         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10781                              io != NULL; io = next_io) {
10782                                 next_io = (union ctl_io *)TAILQ_NEXT(
10783                                         &io->io_hdr, ooa_links);
10784
10785                                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10786                                         io->io_hdr.flags |= CTL_FLAG_ABORT;
10787                         }
10788                 } else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10789                         && (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10790
10791                         printf("FAILOVER: secondary lun %d\n", lun_idx);
10792
10793                         lun->flags |= CTL_LUN_PRIMARY_SC;
10794
10795                         /*
10796                          * We send all I/O that was sent to this controller
10797                          * and redirected to the other side back with
10798                          * busy status, and have the initiator retry it.
10799                          * Figuring out how much data has been transferred,
10800                          * etc. and picking up where we left off would be
10801                          * very tricky.
10802                          *
10803                          * XXX KDM need to remove I/O from the blocked
10804                          * queue as well!
10805                          */
10806                         for (pending_io = (union ctl_io *)TAILQ_FIRST(
10807                              &lun->ooa_queue); pending_io != NULL;
10808                              pending_io = next_io) {
10809
10810                                 next_io =  (union ctl_io *)TAILQ_NEXT(
10811                                         &pending_io->io_hdr, ooa_links);
10812
10813                                 pending_io->io_hdr.flags &=
10814                                         ~CTL_FLAG_SENT_2OTHER_SC;
10815
10816                                 if (pending_io->io_hdr.flags &
10817                                     CTL_FLAG_IO_ACTIVE) {
10818                                         pending_io->io_hdr.flags |=
10819                                                 CTL_FLAG_FAILOVER;
10820                                 } else {
10821                                         ctl_set_busy(&pending_io->scsiio);
10822                                         ctl_done(pending_io);
10823                                 }
10824                         }
10825
10826                         /*
10827                          * Build Unit Attention
10828                          */
10829                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10830                                 lun->pending_sense[i].ua_pending |=
10831                                                      CTL_UA_ASYM_ACC_CHANGE;
10832                         }
10833                 } else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10834                         && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10835                         printf("FAILOVER: secondary lun %d\n", lun_idx);
10836                         /*
10837                          * if the first io on the OOA is not on the RtR queue
10838                          * add it.
10839                          */
10840                         lun->flags |= CTL_LUN_PRIMARY_SC;
10841
10842                         pending_io = (union ctl_io *)TAILQ_FIRST(
10843                             &lun->ooa_queue);
10844                         if (pending_io==NULL) {
10845                                 printf("Nothing on OOA queue\n");
10846                                 continue;
10847                         }
10848
10849                         pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
10850                         if ((pending_io->io_hdr.flags &
10851                              CTL_FLAG_IS_WAS_ON_RTR) == 0) {
10852                                 pending_io->io_hdr.flags |=
10853                                     CTL_FLAG_IS_WAS_ON_RTR;
10854                                 ctl_enqueue_rtr(pending_io);
10855                         }
10856 #if 0
10857                         else
10858                         {
10859                                 printf("Tag 0x%04x is running\n",
10860                                       pending_io->scsiio.tag_num);
10861                         }
10862 #endif
10863
10864                         next_io = (union ctl_io *)TAILQ_NEXT(
10865                             &pending_io->io_hdr, ooa_links);
10866                         for (pending_io=next_io; pending_io != NULL;
10867                              pending_io = next_io) {
10868                                 pending_io->io_hdr.flags &=
10869                                     ~CTL_FLAG_SENT_2OTHER_SC;
10870                                 next_io = (union ctl_io *)TAILQ_NEXT(
10871                                         &pending_io->io_hdr, ooa_links);
10872                                 if (pending_io->io_hdr.flags &
10873                                     CTL_FLAG_IS_WAS_ON_RTR) {
10874 #if 0
10875                                         printf("Tag 0x%04x is running\n",
10876                                                 pending_io->scsiio.tag_num);
10877 #endif
10878                                         continue;
10879                                 }
10880
10881                                 switch (ctl_check_ooa(lun, pending_io,
10882                                     (union ctl_io *)TAILQ_PREV(
10883                                     &pending_io->io_hdr, ctl_ooaq,
10884                                     ooa_links))) {
10885
10886                                 case CTL_ACTION_BLOCK:
10887                                         TAILQ_INSERT_TAIL(&lun->blocked_queue,
10888                                                           &pending_io->io_hdr,
10889                                                           blocked_links);
10890                                         pending_io->io_hdr.flags |=
10891                                             CTL_FLAG_BLOCKED;
10892                                         break;
10893                                 case CTL_ACTION_PASS:
10894                                 case CTL_ACTION_SKIP:
10895                                         pending_io->io_hdr.flags |=
10896                                             CTL_FLAG_IS_WAS_ON_RTR;
10897                                         ctl_enqueue_rtr(pending_io);
10898                                         break;
10899                                 case CTL_ACTION_OVERLAP:
10900                                         ctl_set_overlapped_cmd(
10901                                             (struct ctl_scsiio *)pending_io);
10902                                         ctl_done(pending_io);
10903                                         break;
10904                                 case CTL_ACTION_OVERLAP_TAG:
10905                                         ctl_set_overlapped_tag(
10906                                             (struct ctl_scsiio *)pending_io,
10907                                             pending_io->scsiio.tag_num & 0xff);
10908                                         ctl_done(pending_io);
10909                                         break;
10910                                 case CTL_ACTION_ERROR:
10911                                 default:
10912                                         ctl_set_internal_failure(
10913                                                 (struct ctl_scsiio *)pending_io,
10914                                                 0,  // sks_valid
10915                                                 0); //retry count
10916                                         ctl_done(pending_io);
10917                                         break;
10918                                 }
10919                         }
10920
10921                         /*
10922                          * Build Unit Attention
10923                          */
10924                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10925                                 lun->pending_sense[i].ua_pending |=
10926                                                      CTL_UA_ASYM_ACC_CHANGE;
10927                         }
10928                 } else {
10929                         panic("Unhandled HA mode failover, LUN flags = %#x, "
10930                               "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
10931                 }
10932         }
10933         ctl_pause_rtr = 0;
10934         mtx_unlock(&ctl_softc->ctl_lock);
10935 }
10936
10937 static int
10938 ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
10939 {
10940         struct ctl_lun *lun;
10941         const struct ctl_cmd_entry *entry;
10942         uint32_t initidx, targ_lun;
10943         int retval;
10944
10945         retval = 0;
10946
10947         lun = NULL;
10948
10949         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
10950         if ((targ_lun < CTL_MAX_LUNS)
10951          && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
10952                 lun = ctl_softc->ctl_luns[targ_lun];
10953                 /*
10954                  * If the LUN is invalid, pretend that it doesn't exist.
10955                  * It will go away as soon as all pending I/O has been
10956                  * completed.
10957                  */
10958                 if (lun->flags & CTL_LUN_DISABLED) {
10959                         lun = NULL;
10960                 } else {
10961                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
10962                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
10963                                 lun->be_lun;
10964                         if (lun->be_lun->lun_type == T_PROCESSOR) {
10965                                 ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
10966                         }
10967
10968                         /*
10969                          * Every I/O goes into the OOA queue for a
10970                          * particular LUN, and stays there until completion.
10971                          */
10972                         mtx_lock(&lun->lun_lock);
10973                         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
10974                             ooa_links);
10975                 }
10976         } else {
10977                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
10978                 ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
10979         }
10980
10981         /* Get command entry and return error if it is unsuppotyed. */
10982         entry = ctl_validate_command(ctsio);
10983         if (entry == NULL) {
10984                 if (lun)
10985                         mtx_unlock(&lun->lun_lock);
10986                 return (retval);
10987         }
10988
10989         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
10990         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
10991
10992         /*
10993          * Check to see whether we can send this command to LUNs that don't
10994          * exist.  This should pretty much only be the case for inquiry
10995          * and request sense.  Further checks, below, really require having
10996          * a LUN, so we can't really check the command anymore.  Just put
10997          * it on the rtr queue.
10998          */
10999         if (lun == NULL) {
11000                 if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11001                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11002                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11003                         return (retval);
11004                 }
11005
11006                 ctl_set_unsupported_lun(ctsio);
11007                 ctl_done((union ctl_io *)ctsio);
11008                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11009                 return (retval);
11010         } else {
11011                 /*
11012                  * Make sure we support this particular command on this LUN.
11013                  * e.g., we don't support writes to the control LUN.
11014                  */
11015                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11016                         mtx_unlock(&lun->lun_lock);
11017                         ctl_set_invalid_opcode(ctsio);
11018                         ctl_done((union ctl_io *)ctsio);
11019                         return (retval);
11020                 }
11021         }
11022
11023         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11024
11025         /*
11026          * If we've got a request sense, it'll clear the contingent
11027          * allegiance condition.  Otherwise, if we have a CA condition for
11028          * this initiator, clear it, because it sent down a command other
11029          * than request sense.
11030          */
11031         if ((ctsio->cdb[0] != REQUEST_SENSE)
11032          && (ctl_is_set(lun->have_ca, initidx)))
11033                 ctl_clear_mask(lun->have_ca, initidx);
11034
11035         /*
11036          * If the command has this flag set, it handles its own unit
11037          * attention reporting, we shouldn't do anything.  Otherwise we
11038          * check for any pending unit attentions, and send them back to the
11039          * initiator.  We only do this when a command initially comes in,
11040          * not when we pull it off the blocked queue.
11041          *
11042          * According to SAM-3, section 5.3.2, the order that things get
11043          * presented back to the host is basically unit attentions caused
11044          * by some sort of reset event, busy status, reservation conflicts
11045          * or task set full, and finally any other status.
11046          *
11047          * One issue here is that some of the unit attentions we report
11048          * don't fall into the "reset" category (e.g. "reported luns data
11049          * has changed").  So reporting it here, before the reservation
11050          * check, may be technically wrong.  I guess the only thing to do
11051          * would be to check for and report the reset events here, and then
11052          * check for the other unit attention types after we check for a
11053          * reservation conflict.
11054          *
11055          * XXX KDM need to fix this
11056          */
11057         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11058                 ctl_ua_type ua_type;
11059
11060                 ua_type = lun->pending_sense[initidx].ua_pending;
11061                 if (ua_type != CTL_UA_NONE) {
11062                         scsi_sense_data_type sense_format;
11063
11064                         if (lun != NULL)
11065                                 sense_format = (lun->flags &
11066                                     CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11067                                     SSD_TYPE_FIXED;
11068                         else
11069                                 sense_format = SSD_TYPE_FIXED;
11070
11071                         ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
11072                                                sense_format);
11073                         if (ua_type != CTL_UA_NONE) {
11074                                 ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11075                                 ctsio->io_hdr.status = CTL_SCSI_ERROR |
11076                                                        CTL_AUTOSENSE;
11077                                 ctsio->sense_len = SSD_FULL_SIZE;
11078                                 lun->pending_sense[initidx].ua_pending &=
11079                                         ~ua_type;
11080                                 mtx_unlock(&lun->lun_lock);
11081                                 ctl_done((union ctl_io *)ctsio);
11082                                 return (retval);
11083                         }
11084                 }
11085         }
11086
11087
11088         if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11089                 mtx_unlock(&lun->lun_lock);
11090                 ctl_done((union ctl_io *)ctsio);
11091                 return (retval);
11092         }
11093
11094         /*
11095          * XXX CHD this is where we want to send IO to other side if
11096          * this LUN is secondary on this SC. We will need to make a copy
11097          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11098          * the copy we send as FROM_OTHER.
11099          * We also need to stuff the address of the original IO so we can
11100          * find it easily. Something similar will need be done on the other
11101          * side so when we are done we can find the copy.
11102          */
11103         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11104                 union ctl_ha_msg msg_info;
11105                 int isc_retval;
11106
11107                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11108
11109                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11110                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11111 #if 0
11112                 printf("1. ctsio %p\n", ctsio);
11113 #endif
11114                 msg_info.hdr.serializing_sc = NULL;
11115                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11116                 msg_info.scsi.tag_num = ctsio->tag_num;
11117                 msg_info.scsi.tag_type = ctsio->tag_type;
11118                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11119
11120                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11121
11122                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11123                     (void *)&msg_info, sizeof(msg_info), 0)) >
11124                     CTL_HA_STATUS_SUCCESS) {
11125                         printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11126                                isc_retval);
11127                         printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11128                 } else {
11129 #if 0
11130                         printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11131 #endif
11132                 }
11133
11134                 /*
11135                  * XXX KDM this I/O is off the incoming queue, but hasn't
11136                  * been inserted on any other queue.  We may need to come
11137                  * up with a holding queue while we wait for serialization
11138                  * so that we have an idea of what we're waiting for from
11139                  * the other side.
11140                  */
11141                 mtx_unlock(&lun->lun_lock);
11142                 return (retval);
11143         }
11144
11145         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11146                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11147                               ctl_ooaq, ooa_links))) {
11148         case CTL_ACTION_BLOCK:
11149                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11150                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11151                                   blocked_links);
11152                 mtx_unlock(&lun->lun_lock);
11153                 return (retval);
11154         case CTL_ACTION_PASS:
11155         case CTL_ACTION_SKIP:
11156                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11157                 mtx_unlock(&lun->lun_lock);
11158                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11159                 break;
11160         case CTL_ACTION_OVERLAP:
11161                 mtx_unlock(&lun->lun_lock);
11162                 ctl_set_overlapped_cmd(ctsio);
11163                 ctl_done((union ctl_io *)ctsio);
11164                 break;
11165         case CTL_ACTION_OVERLAP_TAG:
11166                 mtx_unlock(&lun->lun_lock);
11167                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11168                 ctl_done((union ctl_io *)ctsio);
11169                 break;
11170         case CTL_ACTION_ERROR:
11171         default:
11172                 mtx_unlock(&lun->lun_lock);
11173                 ctl_set_internal_failure(ctsio,
11174                                          /*sks_valid*/ 0,
11175                                          /*retry_count*/ 0);
11176                 ctl_done((union ctl_io *)ctsio);
11177                 break;
11178         }
11179         return (retval);
11180 }
11181
11182 const struct ctl_cmd_entry *
11183 ctl_get_cmd_entry(struct ctl_scsiio *ctsio)
11184 {
11185         const struct ctl_cmd_entry *entry;
11186         int service_action;
11187
11188         entry = &ctl_cmd_table[ctsio->cdb[0]];
11189         if (entry->flags & CTL_CMD_FLAG_SA5) {
11190                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11191                 entry = &((const struct ctl_cmd_entry *)
11192                     entry->execute)[service_action];
11193         }
11194         return (entry);
11195 }
11196
11197 const struct ctl_cmd_entry *
11198 ctl_validate_command(struct ctl_scsiio *ctsio)
11199 {
11200         const struct ctl_cmd_entry *entry;
11201         int i;
11202         uint8_t diff;
11203
11204         entry = ctl_get_cmd_entry(ctsio);
11205         if (entry->execute == NULL) {
11206                 ctl_set_invalid_opcode(ctsio);
11207                 ctl_done((union ctl_io *)ctsio);
11208                 return (NULL);
11209         }
11210         KASSERT(entry->length > 0,
11211             ("Not defined length for command 0x%02x/0x%02x",
11212              ctsio->cdb[0], ctsio->cdb[1]));
11213         for (i = 1; i < entry->length; i++) {
11214                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11215                 if (diff == 0)
11216                         continue;
11217                 ctl_set_invalid_field(ctsio,
11218                                       /*sks_valid*/ 1,
11219                                       /*command*/ 1,
11220                                       /*field*/ i,
11221                                       /*bit_valid*/ 1,
11222                                       /*bit*/ fls(diff) - 1);
11223                 ctl_done((union ctl_io *)ctsio);
11224                 return (NULL);
11225         }
11226         return (entry);
11227 }
11228
11229 static int
11230 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11231 {
11232
11233         switch (lun_type) {
11234         case T_PROCESSOR:
11235                 if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11236                     ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11237                         return (0);
11238                 break;
11239         case T_DIRECT:
11240                 if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11241                     ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11242                         return (0);
11243                 break;
11244         default:
11245                 return (0);
11246         }
11247         return (1);
11248 }
11249
11250 static int
11251 ctl_scsiio(struct ctl_scsiio *ctsio)
11252 {
11253         int retval;
11254         const struct ctl_cmd_entry *entry;
11255
11256         retval = CTL_RETVAL_COMPLETE;
11257
11258         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11259
11260         entry = ctl_get_cmd_entry(ctsio);
11261
11262         /*
11263          * If this I/O has been aborted, just send it straight to
11264          * ctl_done() without executing it.
11265          */
11266         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11267                 ctl_done((union ctl_io *)ctsio);
11268                 goto bailout;
11269         }
11270
11271         /*
11272          * All the checks should have been handled by ctl_scsiio_precheck().
11273          * We should be clear now to just execute the I/O.
11274          */
11275         retval = entry->execute(ctsio);
11276
11277 bailout:
11278         return (retval);
11279 }
11280
11281 /*
11282  * Since we only implement one target right now, a bus reset simply resets
11283  * our single target.
11284  */
11285 static int
11286 ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11287 {
11288         return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11289 }
11290
11291 static int
11292 ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11293                  ctl_ua_type ua_type)
11294 {
11295         struct ctl_lun *lun;
11296         int retval;
11297
11298         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11299                 union ctl_ha_msg msg_info;
11300
11301                 io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11302                 msg_info.hdr.nexus = io->io_hdr.nexus;
11303                 if (ua_type==CTL_UA_TARG_RESET)
11304                         msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11305                 else
11306                         msg_info.task.task_action = CTL_TASK_BUS_RESET;
11307                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11308                 msg_info.hdr.original_sc = NULL;
11309                 msg_info.hdr.serializing_sc = NULL;
11310                 if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11311                     (void *)&msg_info, sizeof(msg_info), 0)) {
11312                 }
11313         }
11314         retval = 0;
11315
11316         mtx_lock(&ctl_softc->ctl_lock);
11317         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11318                 retval += ctl_lun_reset(lun, io, ua_type);
11319         mtx_unlock(&ctl_softc->ctl_lock);
11320
11321         return (retval);
11322 }
11323
11324 /*
11325  * The LUN should always be set.  The I/O is optional, and is used to
11326  * distinguish between I/Os sent by this initiator, and by other
11327  * initiators.  We set unit attention for initiators other than this one.
11328  * SAM-3 is vague on this point.  It does say that a unit attention should
11329  * be established for other initiators when a LUN is reset (see section
11330  * 5.7.3), but it doesn't specifically say that the unit attention should
11331  * be established for this particular initiator when a LUN is reset.  Here
11332  * is the relevant text, from SAM-3 rev 8:
11333  *
11334  * 5.7.2 When a SCSI initiator port aborts its own tasks
11335  *
11336  * When a SCSI initiator port causes its own task(s) to be aborted, no
11337  * notification that the task(s) have been aborted shall be returned to
11338  * the SCSI initiator port other than the completion response for the
11339  * command or task management function action that caused the task(s) to
11340  * be aborted and notification(s) associated with related effects of the
11341  * action (e.g., a reset unit attention condition).
11342  *
11343  * XXX KDM for now, we're setting unit attention for all initiators.
11344  */
11345 static int
11346 ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11347 {
11348         union ctl_io *xio;
11349 #if 0
11350         uint32_t initindex;
11351 #endif
11352         int i;
11353
11354         mtx_lock(&lun->lun_lock);
11355         /*
11356          * Run through the OOA queue and abort each I/O.
11357          */
11358 #if 0
11359         TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11360 #endif
11361         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11362              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11363                 xio->io_hdr.flags |= CTL_FLAG_ABORT;
11364         }
11365
11366         /*
11367          * This version sets unit attention for every
11368          */
11369 #if 0
11370         initindex = ctl_get_initindex(&io->io_hdr.nexus);
11371         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11372                 if (initindex == i)
11373                         continue;
11374                 lun->pending_sense[i].ua_pending |= ua_type;
11375         }
11376 #endif
11377
11378         /*
11379          * A reset (any kind, really) clears reservations established with
11380          * RESERVE/RELEASE.  It does not clear reservations established
11381          * with PERSISTENT RESERVE OUT, but we don't support that at the
11382          * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11383          * reservations made with the RESERVE/RELEASE commands, because
11384          * those commands are obsolete in SPC-3.
11385          */
11386         lun->flags &= ~CTL_LUN_RESERVED;
11387
11388         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11389                 ctl_clear_mask(lun->have_ca, i);
11390                 lun->pending_sense[i].ua_pending |= ua_type;
11391         }
11392         mtx_unlock(&lun->lun_lock);
11393
11394         return (0);
11395 }
11396
11397 static int
11398 ctl_abort_task(union ctl_io *io)
11399 {
11400         union ctl_io *xio;
11401         struct ctl_lun *lun;
11402         struct ctl_softc *ctl_softc;
11403 #if 0
11404         struct sbuf sb;
11405         char printbuf[128];
11406 #endif
11407         int found;
11408         uint32_t targ_lun;
11409
11410         ctl_softc = control_softc;
11411         found = 0;
11412
11413         /*
11414          * Look up the LUN.
11415          */
11416         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11417         mtx_lock(&ctl_softc->ctl_lock);
11418         if ((targ_lun < CTL_MAX_LUNS)
11419          && (ctl_softc->ctl_luns[targ_lun] != NULL))
11420                 lun = ctl_softc->ctl_luns[targ_lun];
11421         else {
11422                 mtx_unlock(&ctl_softc->ctl_lock);
11423                 goto bailout;
11424         }
11425
11426 #if 0
11427         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11428                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11429 #endif
11430
11431         mtx_lock(&lun->lun_lock);
11432         mtx_unlock(&ctl_softc->ctl_lock);
11433         /*
11434          * Run through the OOA queue and attempt to find the given I/O.
11435          * The target port, initiator ID, tag type and tag number have to
11436          * match the values that we got from the initiator.  If we have an
11437          * untagged command to abort, simply abort the first untagged command
11438          * we come to.  We only allow one untagged command at a time of course.
11439          */
11440 #if 0
11441         TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11442 #endif
11443         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11444              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11445 #if 0
11446                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11447
11448                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11449                             lun->lun, xio->scsiio.tag_num,
11450                             xio->scsiio.tag_type,
11451                             (xio->io_hdr.blocked_links.tqe_prev
11452                             == NULL) ? "" : " BLOCKED",
11453                             (xio->io_hdr.flags &
11454                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11455                             (xio->io_hdr.flags &
11456                             CTL_FLAG_ABORT) ? " ABORT" : "",
11457                             (xio->io_hdr.flags &
11458                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11459                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11460                 sbuf_finish(&sb);
11461                 printf("%s\n", sbuf_data(&sb));
11462 #endif
11463
11464                 if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
11465                  && (xio->io_hdr.nexus.initid.id ==
11466                      io->io_hdr.nexus.initid.id)) {
11467                         /*
11468                          * If the abort says that the task is untagged, the
11469                          * task in the queue must be untagged.  Otherwise,
11470                          * we just check to see whether the tag numbers
11471                          * match.  This is because the QLogic firmware
11472                          * doesn't pass back the tag type in an abort
11473                          * request.
11474                          */
11475 #if 0
11476                         if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
11477                           && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
11478                          || (xio->scsiio.tag_num == io->taskio.tag_num)) {
11479 #endif
11480                         /*
11481                          * XXX KDM we've got problems with FC, because it
11482                          * doesn't send down a tag type with aborts.  So we
11483                          * can only really go by the tag number...
11484                          * This may cause problems with parallel SCSI.
11485                          * Need to figure that out!!
11486                          */
11487                         if (xio->scsiio.tag_num == io->taskio.tag_num) {
11488                                 xio->io_hdr.flags |= CTL_FLAG_ABORT;
11489                                 found = 1;
11490                                 if ((io->io_hdr.flags &
11491                                      CTL_FLAG_FROM_OTHER_SC) == 0 &&
11492                                     !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11493                                         union ctl_ha_msg msg_info;
11494
11495                                         io->io_hdr.flags |=
11496                                                         CTL_FLAG_SENT_2OTHER_SC;
11497                                         msg_info.hdr.nexus = io->io_hdr.nexus;
11498                                         msg_info.task.task_action =
11499                                                 CTL_TASK_ABORT_TASK;
11500                                         msg_info.task.tag_num =
11501                                                 io->taskio.tag_num;
11502                                         msg_info.task.tag_type =
11503                                                 io->taskio.tag_type;
11504                                         msg_info.hdr.msg_type =
11505                                                 CTL_MSG_MANAGE_TASKS;
11506                                         msg_info.hdr.original_sc = NULL;
11507                                         msg_info.hdr.serializing_sc = NULL;
11508 #if 0
11509                                         printf("Sent Abort to other side\n");
11510 #endif
11511                                         if (CTL_HA_STATUS_SUCCESS !=
11512                                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11513                                                 (void *)&msg_info,
11514                                                 sizeof(msg_info), 0)) {
11515                                         }
11516                                 }
11517 #if 0
11518                                 printf("ctl_abort_task: found I/O to abort\n");
11519 #endif
11520                                 break;
11521                         }
11522                 }
11523         }
11524         mtx_unlock(&lun->lun_lock);
11525
11526 bailout:
11527
11528         if (found == 0) {
11529                 /*
11530                  * This isn't really an error.  It's entirely possible for
11531                  * the abort and command completion to cross on the wire.
11532                  * This is more of an informative/diagnostic error.
11533                  */
11534 #if 0
11535                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
11536                        "%d:%d:%d:%d tag %d type %d\n",
11537                        io->io_hdr.nexus.initid.id,
11538                        io->io_hdr.nexus.targ_port,
11539                        io->io_hdr.nexus.targ_target.id,
11540                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
11541                        io->taskio.tag_type);
11542 #endif
11543                 return (1);
11544         } else
11545                 return (0);
11546 }
11547
11548 static void
11549 ctl_run_task(union ctl_io *io)
11550 {
11551         struct ctl_softc *ctl_softc;
11552         int retval;
11553         const char *task_desc;
11554
11555         CTL_DEBUG_PRINT(("ctl_run_task\n"));
11556
11557         ctl_softc = control_softc;
11558         retval = 0;
11559
11560         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
11561             ("ctl_run_task: Unextected io_type %d\n",
11562              io->io_hdr.io_type));
11563
11564         task_desc = ctl_scsi_task_string(&io->taskio);
11565         if (task_desc != NULL) {
11566 #ifdef NEEDTOPORT
11567                 csevent_log(CSC_CTL | CSC_SHELF_SW |
11568                             CTL_TASK_REPORT,
11569                             csevent_LogType_Trace,
11570                             csevent_Severity_Information,
11571                             csevent_AlertLevel_Green,
11572                             csevent_FRU_Firmware,
11573                             csevent_FRU_Unknown,
11574                             "CTL: received task: %s",task_desc);
11575 #endif
11576         } else {
11577 #ifdef NEEDTOPORT
11578                 csevent_log(CSC_CTL | CSC_SHELF_SW |
11579                             CTL_TASK_REPORT,
11580                             csevent_LogType_Trace,
11581                             csevent_Severity_Information,
11582                             csevent_AlertLevel_Green,
11583                             csevent_FRU_Firmware,
11584                             csevent_FRU_Unknown,
11585                             "CTL: received unknown task "
11586                             "type: %d (%#x)",
11587                             io->taskio.task_action,
11588                             io->taskio.task_action);
11589 #endif
11590         }
11591         switch (io->taskio.task_action) {
11592         case CTL_TASK_ABORT_TASK:
11593                 retval = ctl_abort_task(io);
11594                 break;
11595         case CTL_TASK_ABORT_TASK_SET:
11596                 break;
11597         case CTL_TASK_CLEAR_ACA:
11598                 break;
11599         case CTL_TASK_CLEAR_TASK_SET:
11600                 break;
11601         case CTL_TASK_LUN_RESET: {
11602                 struct ctl_lun *lun;
11603                 uint32_t targ_lun;
11604                 int retval;
11605
11606                 targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11607                 mtx_lock(&ctl_softc->ctl_lock);
11608                 if ((targ_lun < CTL_MAX_LUNS)
11609                  && (ctl_softc->ctl_luns[targ_lun] != NULL))
11610                         lun = ctl_softc->ctl_luns[targ_lun];
11611                 else {
11612                         mtx_unlock(&ctl_softc->ctl_lock);
11613                         retval = 1;
11614                         break;
11615                 }
11616
11617                 if (!(io->io_hdr.flags &
11618                     CTL_FLAG_FROM_OTHER_SC)) {
11619                         union ctl_ha_msg msg_info;
11620
11621                         io->io_hdr.flags |=
11622                                 CTL_FLAG_SENT_2OTHER_SC;
11623                         msg_info.hdr.msg_type =
11624                                 CTL_MSG_MANAGE_TASKS;
11625                         msg_info.hdr.nexus = io->io_hdr.nexus;
11626                         msg_info.task.task_action =
11627                                 CTL_TASK_LUN_RESET;
11628                         msg_info.hdr.original_sc = NULL;
11629                         msg_info.hdr.serializing_sc = NULL;
11630                         if (CTL_HA_STATUS_SUCCESS !=
11631                             ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11632                             (void *)&msg_info,
11633                             sizeof(msg_info), 0)) {
11634                         }
11635                 }
11636
11637                 retval = ctl_lun_reset(lun, io,
11638                                        CTL_UA_LUN_RESET);
11639                 mtx_unlock(&ctl_softc->ctl_lock);
11640                 break;
11641         }
11642         case CTL_TASK_TARGET_RESET:
11643                 retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
11644                 break;
11645         case CTL_TASK_BUS_RESET:
11646                 retval = ctl_bus_reset(ctl_softc, io);
11647                 break;
11648         case CTL_TASK_PORT_LOGIN:
11649                 break;
11650         case CTL_TASK_PORT_LOGOUT:
11651                 break;
11652         default:
11653                 printf("ctl_run_task: got unknown task management event %d\n",
11654                        io->taskio.task_action);
11655                 break;
11656         }
11657         if (retval == 0)
11658                 io->io_hdr.status = CTL_SUCCESS;
11659         else
11660                 io->io_hdr.status = CTL_ERROR;
11661
11662         /*
11663          * This will queue this I/O to the done queue, but the
11664          * work thread won't be able to process it until we
11665          * return and the lock is released.
11666          */
11667         ctl_done(io);
11668 }
11669
11670 /*
11671  * For HA operation.  Handle commands that come in from the other
11672  * controller.
11673  */
11674 static void
11675 ctl_handle_isc(union ctl_io *io)
11676 {
11677         int free_io;
11678         struct ctl_lun *lun;
11679         struct ctl_softc *ctl_softc;
11680         uint32_t targ_lun;
11681
11682         ctl_softc = control_softc;
11683
11684         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11685         lun = ctl_softc->ctl_luns[targ_lun];
11686
11687         switch (io->io_hdr.msg_type) {
11688         case CTL_MSG_SERIALIZE:
11689                 free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
11690                 break;
11691         case CTL_MSG_R2R: {
11692                 const struct ctl_cmd_entry *entry;
11693
11694                 /*
11695                  * This is only used in SER_ONLY mode.
11696                  */
11697                 free_io = 0;
11698                 entry = ctl_get_cmd_entry(&io->scsiio);
11699                 mtx_lock(&lun->lun_lock);
11700                 if (ctl_scsiio_lun_check(ctl_softc, lun,
11701                     entry, (struct ctl_scsiio *)io) != 0) {
11702                         mtx_unlock(&lun->lun_lock);
11703                         ctl_done(io);
11704                         break;
11705                 }
11706                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11707                 mtx_unlock(&lun->lun_lock);
11708                 ctl_enqueue_rtr(io);
11709                 break;
11710         }
11711         case CTL_MSG_FINISH_IO:
11712                 if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
11713                         free_io = 0;
11714                         ctl_done(io);
11715                 } else {
11716                         free_io = 1;
11717                         mtx_lock(&lun->lun_lock);
11718                         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
11719                                      ooa_links);
11720                         ctl_check_blocked(lun);
11721                         mtx_unlock(&lun->lun_lock);
11722                 }
11723                 break;
11724         case CTL_MSG_PERS_ACTION:
11725                 ctl_hndl_per_res_out_on_other_sc(
11726                         (union ctl_ha_msg *)&io->presio.pr_msg);
11727                 free_io = 1;
11728                 break;
11729         case CTL_MSG_BAD_JUJU:
11730                 free_io = 0;
11731                 ctl_done(io);
11732                 break;
11733         case CTL_MSG_DATAMOVE:
11734                 /* Only used in XFER mode */
11735                 free_io = 0;
11736                 ctl_datamove_remote(io);
11737                 break;
11738         case CTL_MSG_DATAMOVE_DONE:
11739                 /* Only used in XFER mode */
11740                 free_io = 0;
11741                 io->scsiio.be_move_done(io);
11742                 break;
11743         default:
11744                 free_io = 1;
11745                 printf("%s: Invalid message type %d\n",
11746                        __func__, io->io_hdr.msg_type);
11747                 break;
11748         }
11749         if (free_io)
11750                 ctl_free_io(io);
11751
11752 }
11753
11754
11755 /*
11756  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
11757  * there is no match.
11758  */
11759 static ctl_lun_error_pattern
11760 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
11761 {
11762         const struct ctl_cmd_entry *entry;
11763         ctl_lun_error_pattern filtered_pattern, pattern;
11764
11765         pattern = desc->error_pattern;
11766
11767         /*
11768          * XXX KDM we need more data passed into this function to match a
11769          * custom pattern, and we actually need to implement custom pattern
11770          * matching.
11771          */
11772         if (pattern & CTL_LUN_PAT_CMD)
11773                 return (CTL_LUN_PAT_CMD);
11774
11775         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
11776                 return (CTL_LUN_PAT_ANY);
11777
11778         entry = ctl_get_cmd_entry(ctsio);
11779
11780         filtered_pattern = entry->pattern & pattern;
11781
11782         /*
11783          * If the user requested specific flags in the pattern (e.g.
11784          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
11785          * flags.
11786          *
11787          * If the user did not specify any flags, it doesn't matter whether
11788          * or not the command supports the flags.
11789          */
11790         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
11791              (pattern & ~CTL_LUN_PAT_MASK))
11792                 return (CTL_LUN_PAT_NONE);
11793
11794         /*
11795          * If the user asked for a range check, see if the requested LBA
11796          * range overlaps with this command's LBA range.
11797          */
11798         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
11799                 uint64_t lba1;
11800                 uint32_t len1;
11801                 ctl_action action;
11802                 int retval;
11803
11804                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
11805                 if (retval != 0)
11806                         return (CTL_LUN_PAT_NONE);
11807
11808                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
11809                                               desc->lba_range.len);
11810                 /*
11811                  * A "pass" means that the LBA ranges don't overlap, so
11812                  * this doesn't match the user's range criteria.
11813                  */
11814                 if (action == CTL_ACTION_PASS)
11815                         return (CTL_LUN_PAT_NONE);
11816         }
11817
11818         return (filtered_pattern);
11819 }
11820
11821 static void
11822 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
11823 {
11824         struct ctl_error_desc *desc, *desc2;
11825
11826         mtx_assert(&lun->lun_lock, MA_OWNED);
11827
11828         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
11829                 ctl_lun_error_pattern pattern;
11830                 /*
11831                  * Check to see whether this particular command matches
11832                  * the pattern in the descriptor.
11833                  */
11834                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
11835                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
11836                         continue;
11837
11838                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
11839                 case CTL_LUN_INJ_ABORTED:
11840                         ctl_set_aborted(&io->scsiio);
11841                         break;
11842                 case CTL_LUN_INJ_MEDIUM_ERR:
11843                         ctl_set_medium_error(&io->scsiio);
11844                         break;
11845                 case CTL_LUN_INJ_UA:
11846                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
11847                          * OCCURRED */
11848                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
11849                         break;
11850                 case CTL_LUN_INJ_CUSTOM:
11851                         /*
11852                          * We're assuming the user knows what he is doing.
11853                          * Just copy the sense information without doing
11854                          * checks.
11855                          */
11856                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
11857                               ctl_min(sizeof(desc->custom_sense),
11858                                       sizeof(io->scsiio.sense_data)));
11859                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
11860                         io->scsiio.sense_len = SSD_FULL_SIZE;
11861                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11862                         break;
11863                 case CTL_LUN_INJ_NONE:
11864                 default:
11865                         /*
11866                          * If this is an error injection type we don't know
11867                          * about, clear the continuous flag (if it is set)
11868                          * so it will get deleted below.
11869                          */
11870                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
11871                         break;
11872                 }
11873                 /*
11874                  * By default, each error injection action is a one-shot
11875                  */
11876                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
11877                         continue;
11878
11879                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
11880
11881                 free(desc, M_CTL);
11882         }
11883 }
11884
11885 #ifdef CTL_IO_DELAY
11886 static void
11887 ctl_datamove_timer_wakeup(void *arg)
11888 {
11889         union ctl_io *io;
11890
11891         io = (union ctl_io *)arg;
11892
11893         ctl_datamove(io);
11894 }
11895 #endif /* CTL_IO_DELAY */
11896
11897 void
11898 ctl_datamove(union ctl_io *io)
11899 {
11900         void (*fe_datamove)(union ctl_io *io);
11901
11902         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
11903
11904         CTL_DEBUG_PRINT(("ctl_datamove\n"));
11905
11906 #ifdef CTL_TIME_IO
11907         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
11908                 char str[256];
11909                 char path_str[64];
11910                 struct sbuf sb;
11911
11912                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
11913                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
11914
11915                 sbuf_cat(&sb, path_str);
11916                 switch (io->io_hdr.io_type) {
11917                 case CTL_IO_SCSI:
11918                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
11919                         sbuf_printf(&sb, "\n");
11920                         sbuf_cat(&sb, path_str);
11921                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
11922                                     io->scsiio.tag_num, io->scsiio.tag_type);
11923                         break;
11924                 case CTL_IO_TASK:
11925                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
11926                                     "Tag Type: %d\n", io->taskio.task_action,
11927                                     io->taskio.tag_num, io->taskio.tag_type);
11928                         break;
11929                 default:
11930                         printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
11931                         panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
11932                         break;
11933                 }
11934                 sbuf_cat(&sb, path_str);
11935                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
11936                             (intmax_t)time_uptime - io->io_hdr.start_time);
11937                 sbuf_finish(&sb);
11938                 printf("%s", sbuf_data(&sb));
11939         }
11940 #endif /* CTL_TIME_IO */
11941
11942 #ifdef CTL_IO_DELAY
11943         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
11944                 struct ctl_lun *lun;
11945
11946                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
11947
11948                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
11949         } else {
11950                 struct ctl_lun *lun;
11951
11952                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
11953                 if ((lun != NULL)
11954                  && (lun->delay_info.datamove_delay > 0)) {
11955                         struct callout *callout;
11956
11957                         callout = (struct callout *)&io->io_hdr.timer_bytes;
11958                         callout_init(callout, /*mpsafe*/ 1);
11959                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
11960                         callout_reset(callout,
11961                                       lun->delay_info.datamove_delay * hz,
11962                                       ctl_datamove_timer_wakeup, io);
11963                         if (lun->delay_info.datamove_type ==
11964                             CTL_DELAY_TYPE_ONESHOT)
11965                                 lun->delay_info.datamove_delay = 0;
11966                         return;
11967                 }
11968         }
11969 #endif
11970
11971         /*
11972          * This command has been aborted.  Set the port status, so we fail
11973          * the data move.
11974          */
11975         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
11976                 printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
11977                        io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
11978                        io->io_hdr.nexus.targ_port,
11979                        (uintmax_t)io->io_hdr.nexus.targ_target.id,
11980                        io->io_hdr.nexus.targ_lun);
11981                 io->io_hdr.status = CTL_CMD_ABORTED;
11982                 io->io_hdr.port_status = 31337;
11983                 /*
11984                  * Note that the backend, in this case, will get the
11985                  * callback in its context.  In other cases it may get
11986                  * called in the frontend's interrupt thread context.
11987                  */
11988                 io->scsiio.be_move_done(io);
11989                 return;
11990         }
11991
11992         /*
11993          * If we're in XFER mode and this I/O is from the other shelf
11994          * controller, we need to send the DMA to the other side to
11995          * actually transfer the data to/from the host.  In serialize only
11996          * mode the transfer happens below CTL and ctl_datamove() is only
11997          * called on the machine that originally received the I/O.
11998          */
11999         if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12000          && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12001                 union ctl_ha_msg msg;
12002                 uint32_t sg_entries_sent;
12003                 int do_sg_copy;
12004                 int i;
12005
12006                 memset(&msg, 0, sizeof(msg));
12007                 msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12008                 msg.hdr.original_sc = io->io_hdr.original_sc;
12009                 msg.hdr.serializing_sc = io;
12010                 msg.hdr.nexus = io->io_hdr.nexus;
12011                 msg.dt.flags = io->io_hdr.flags;
12012                 /*
12013                  * We convert everything into a S/G list here.  We can't
12014                  * pass by reference, only by value between controllers.
12015                  * So we can't pass a pointer to the S/G list, only as many
12016                  * S/G entries as we can fit in here.  If it's possible for
12017                  * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12018                  * then we need to break this up into multiple transfers.
12019                  */
12020                 if (io->scsiio.kern_sg_entries == 0) {
12021                         msg.dt.kern_sg_entries = 1;
12022                         /*
12023                          * If this is in cached memory, flush the cache
12024                          * before we send the DMA request to the other
12025                          * controller.  We want to do this in either the
12026                          * read or the write case.  The read case is
12027                          * straightforward.  In the write case, we want to
12028                          * make sure nothing is in the local cache that
12029                          * could overwrite the DMAed data.
12030                          */
12031                         if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12032                                 /*
12033                                  * XXX KDM use bus_dmamap_sync() here.
12034                                  */
12035                         }
12036
12037                         /*
12038                          * Convert to a physical address if this is a
12039                          * virtual address.
12040                          */
12041                         if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12042                                 msg.dt.sg_list[0].addr =
12043                                         io->scsiio.kern_data_ptr;
12044                         } else {
12045                                 /*
12046                                  * XXX KDM use busdma here!
12047                                  */
12048 #if 0
12049                                 msg.dt.sg_list[0].addr = (void *)
12050                                         vtophys(io->scsiio.kern_data_ptr);
12051 #endif
12052                         }
12053
12054                         msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12055                         do_sg_copy = 0;
12056                 } else {
12057                         struct ctl_sg_entry *sgl;
12058
12059                         do_sg_copy = 1;
12060                         msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12061                         sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12062                         if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12063                                 /*
12064                                  * XXX KDM use bus_dmamap_sync() here.
12065                                  */
12066                         }
12067                 }
12068
12069                 msg.dt.kern_data_len = io->scsiio.kern_data_len;
12070                 msg.dt.kern_total_len = io->scsiio.kern_total_len;
12071                 msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12072                 msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12073                 msg.dt.sg_sequence = 0;
12074
12075                 /*
12076                  * Loop until we've sent all of the S/G entries.  On the
12077                  * other end, we'll recompose these S/G entries into one
12078                  * contiguous list before passing it to the
12079                  */
12080                 for (sg_entries_sent = 0; sg_entries_sent <
12081                      msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12082                         msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12083                                 sizeof(msg.dt.sg_list[0])),
12084                                 msg.dt.kern_sg_entries - sg_entries_sent);
12085
12086                         if (do_sg_copy != 0) {
12087                                 struct ctl_sg_entry *sgl;
12088                                 int j;
12089
12090                                 sgl = (struct ctl_sg_entry *)
12091                                         io->scsiio.kern_data_ptr;
12092                                 /*
12093                                  * If this is in cached memory, flush the cache
12094                                  * before we send the DMA request to the other
12095                                  * controller.  We want to do this in either
12096                                  * the * read or the write case.  The read
12097                                  * case is straightforward.  In the write
12098                                  * case, we want to make sure nothing is
12099                                  * in the local cache that could overwrite
12100                                  * the DMAed data.
12101                                  */
12102
12103                                 for (i = sg_entries_sent, j = 0;
12104                                      i < msg.dt.cur_sg_entries; i++, j++) {
12105                                         if ((io->io_hdr.flags &
12106                                              CTL_FLAG_NO_DATASYNC) == 0) {
12107                                                 /*
12108                                                  * XXX KDM use bus_dmamap_sync()
12109                                                  */
12110                                         }
12111                                         if ((io->io_hdr.flags &
12112                                              CTL_FLAG_BUS_ADDR) == 0) {
12113                                                 /*
12114                                                  * XXX KDM use busdma.
12115                                                  */
12116 #if 0
12117                                                 msg.dt.sg_list[j].addr =(void *)
12118                                                        vtophys(sgl[i].addr);
12119 #endif
12120                                         } else {
12121                                                 msg.dt.sg_list[j].addr =
12122                                                         sgl[i].addr;
12123                                         }
12124                                         msg.dt.sg_list[j].len = sgl[i].len;
12125                                 }
12126                         }
12127
12128                         sg_entries_sent += msg.dt.cur_sg_entries;
12129                         if (sg_entries_sent >= msg.dt.kern_sg_entries)
12130                                 msg.dt.sg_last = 1;
12131                         else
12132                                 msg.dt.sg_last = 0;
12133
12134                         /*
12135                          * XXX KDM drop and reacquire the lock here?
12136                          */
12137                         if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12138                             sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12139                                 /*
12140                                  * XXX do something here.
12141                                  */
12142                         }
12143
12144                         msg.dt.sent_sg_entries = sg_entries_sent;
12145                 }
12146                 io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12147                 if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12148                         ctl_failover_io(io, /*have_lock*/ 0);
12149
12150         } else {
12151
12152                 /*
12153                  * Lookup the fe_datamove() function for this particular
12154                  * front end.
12155                  */
12156                 fe_datamove =
12157                     control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12158
12159                 fe_datamove(io);
12160         }
12161 }
12162
12163 static void
12164 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12165 {
12166         union ctl_ha_msg msg;
12167         int isc_status;
12168
12169         memset(&msg, 0, sizeof(msg));
12170
12171         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12172         msg.hdr.original_sc = io;
12173         msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12174         msg.hdr.nexus = io->io_hdr.nexus;
12175         msg.hdr.status = io->io_hdr.status;
12176         msg.scsi.tag_num = io->scsiio.tag_num;
12177         msg.scsi.tag_type = io->scsiio.tag_type;
12178         msg.scsi.scsi_status = io->scsiio.scsi_status;
12179         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12180                sizeof(io->scsiio.sense_data));
12181         msg.scsi.sense_len = io->scsiio.sense_len;
12182         msg.scsi.sense_residual = io->scsiio.sense_residual;
12183         msg.scsi.fetd_status = io->io_hdr.port_status;
12184         msg.scsi.residual = io->scsiio.residual;
12185         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12186
12187         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12188                 ctl_failover_io(io, /*have_lock*/ have_lock);
12189                 return;
12190         }
12191
12192         isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12193         if (isc_status > CTL_HA_STATUS_SUCCESS) {
12194                 /* XXX do something if this fails */
12195         }
12196
12197 }
12198
12199 /*
12200  * The DMA to the remote side is done, now we need to tell the other side
12201  * we're done so it can continue with its data movement.
12202  */
12203 static void
12204 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12205 {
12206         union ctl_io *io;
12207
12208         io = rq->context;
12209
12210         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12211                 printf("%s: ISC DMA write failed with error %d", __func__,
12212                        rq->ret);
12213                 ctl_set_internal_failure(&io->scsiio,
12214                                          /*sks_valid*/ 1,
12215                                          /*retry_count*/ rq->ret);
12216         }
12217
12218         ctl_dt_req_free(rq);
12219
12220         /*
12221          * In this case, we had to malloc the memory locally.  Free it.
12222          */
12223         if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12224                 int i;
12225                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12226                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
12227         }
12228         /*
12229          * The data is in local and remote memory, so now we need to send
12230          * status (good or back) back to the other side.
12231          */
12232         ctl_send_datamove_done(io, /*have_lock*/ 0);
12233 }
12234
12235 /*
12236  * We've moved the data from the host/controller into local memory.  Now we
12237  * need to push it over to the remote controller's memory.
12238  */
12239 static int
12240 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12241 {
12242         int retval;
12243
12244         retval = 0;
12245
12246         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12247                                           ctl_datamove_remote_write_cb);
12248
12249         return (retval);
12250 }
12251
12252 static void
12253 ctl_datamove_remote_write(union ctl_io *io)
12254 {
12255         int retval;
12256         void (*fe_datamove)(union ctl_io *io);
12257
12258         /*
12259          * - Get the data from the host/HBA into local memory.
12260          * - DMA memory from the local controller to the remote controller.
12261          * - Send status back to the remote controller.
12262          */
12263
12264         retval = ctl_datamove_remote_sgl_setup(io);
12265         if (retval != 0)
12266                 return;
12267
12268         /* Switch the pointer over so the FETD knows what to do */
12269         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12270
12271         /*
12272          * Use a custom move done callback, since we need to send completion
12273          * back to the other controller, not to the backend on this side.
12274          */
12275         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12276
12277         fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12278
12279         fe_datamove(io);
12280
12281         return;
12282
12283 }
12284
12285 static int
12286 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12287 {
12288 #if 0
12289         char str[256];
12290         char path_str[64];
12291         struct sbuf sb;
12292 #endif
12293
12294         /*
12295          * In this case, we had to malloc the memory locally.  Free it.
12296          */
12297         if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12298                 int i;
12299                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12300                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
12301         }
12302
12303 #if 0
12304         scsi_path_string(io, path_str, sizeof(path_str));
12305         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12306         sbuf_cat(&sb, path_str);
12307         scsi_command_string(&io->scsiio, NULL, &sb);
12308         sbuf_printf(&sb, "\n");
12309         sbuf_cat(&sb, path_str);
12310         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12311                     io->scsiio.tag_num, io->scsiio.tag_type);
12312         sbuf_cat(&sb, path_str);
12313         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12314                     io->io_hdr.flags, io->io_hdr.status);
12315         sbuf_finish(&sb);
12316         printk("%s", sbuf_data(&sb));
12317 #endif
12318
12319
12320         /*
12321          * The read is done, now we need to send status (good or bad) back
12322          * to the other side.
12323          */
12324         ctl_send_datamove_done(io, /*have_lock*/ 0);
12325
12326         return (0);
12327 }
12328
12329 static void
12330 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12331 {
12332         union ctl_io *io;
12333         void (*fe_datamove)(union ctl_io *io);
12334
12335         io = rq->context;
12336
12337         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12338                 printf("%s: ISC DMA read failed with error %d", __func__,
12339                        rq->ret);
12340                 ctl_set_internal_failure(&io->scsiio,
12341                                          /*sks_valid*/ 1,
12342                                          /*retry_count*/ rq->ret);
12343         }
12344
12345         ctl_dt_req_free(rq);
12346
12347         /* Switch the pointer over so the FETD knows what to do */
12348         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12349
12350         /*
12351          * Use a custom move done callback, since we need to send completion
12352          * back to the other controller, not to the backend on this side.
12353          */
12354         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12355
12356         /* XXX KDM add checks like the ones in ctl_datamove? */
12357
12358         fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12359
12360         fe_datamove(io);
12361 }
12362
12363 static int
12364 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12365 {
12366         struct ctl_sg_entry *local_sglist, *remote_sglist;
12367         struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
12368         struct ctl_softc *softc;
12369         int retval;
12370         int i;
12371
12372         retval = 0;
12373         softc = control_softc;
12374
12375         local_sglist = io->io_hdr.local_sglist;
12376         local_dma_sglist = io->io_hdr.local_dma_sglist;
12377         remote_sglist = io->io_hdr.remote_sglist;
12378         remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12379
12380         if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
12381                 for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
12382                         local_sglist[i].len = remote_sglist[i].len;
12383
12384                         /*
12385                          * XXX Detect the situation where the RS-level I/O
12386                          * redirector on the other side has already read the
12387                          * data off of the AOR RS on this side, and
12388                          * transferred it to remote (mirror) memory on the
12389                          * other side.  Since we already have the data in
12390                          * memory here, we just need to use it.
12391                          *
12392                          * XXX KDM this can probably be removed once we
12393                          * get the cache device code in and take the
12394                          * current AOR implementation out.
12395                          */
12396 #ifdef NEEDTOPORT
12397                         if ((remote_sglist[i].addr >=
12398                              (void *)vtophys(softc->mirr->addr))
12399                          && (remote_sglist[i].addr <
12400                              ((void *)vtophys(softc->mirr->addr) +
12401                              CacheMirrorOffset))) {
12402                                 local_sglist[i].addr = remote_sglist[i].addr -
12403                                         CacheMirrorOffset;
12404                                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12405                                      CTL_FLAG_DATA_IN)
12406                                         io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
12407                         } else {
12408                                 local_sglist[i].addr = remote_sglist[i].addr +
12409                                         CacheMirrorOffset;
12410                         }
12411 #endif
12412 #if 0
12413                         printf("%s: local %p, remote %p, len %d\n",
12414                                __func__, local_sglist[i].addr,
12415                                remote_sglist[i].addr, local_sglist[i].len);
12416 #endif
12417                 }
12418         } else {
12419                 uint32_t len_to_go;
12420
12421                 /*
12422                  * In this case, we don't have automatically allocated
12423                  * memory for this I/O on this controller.  This typically
12424                  * happens with internal CTL I/O -- e.g. inquiry, mode
12425                  * sense, etc.  Anything coming from RAIDCore will have
12426                  * a mirror area available.
12427                  */
12428                 len_to_go = io->scsiio.kern_data_len;
12429
12430                 /*
12431                  * Clear the no datasync flag, we have to use malloced
12432                  * buffers.
12433                  */
12434                 io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
12435
12436                 /*
12437                  * The difficult thing here is that the size of the various
12438                  * S/G segments may be different than the size from the
12439                  * remote controller.  That'll make it harder when DMAing
12440                  * the data back to the other side.
12441                  */
12442                 for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
12443                      sizeof(io->io_hdr.remote_sglist[0])) &&
12444                      (len_to_go > 0); i++) {
12445                         local_sglist[i].len = ctl_min(len_to_go, 131072);
12446                         CTL_SIZE_8B(local_dma_sglist[i].len,
12447                                     local_sglist[i].len);
12448                         local_sglist[i].addr =
12449                                 malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
12450
12451                         local_dma_sglist[i].addr = local_sglist[i].addr;
12452
12453                         if (local_sglist[i].addr == NULL) {
12454                                 int j;
12455
12456                                 printf("malloc failed for %zd bytes!",
12457                                        local_dma_sglist[i].len);
12458                                 for (j = 0; j < i; j++) {
12459                                         free(local_sglist[j].addr, M_CTL);
12460                                 }
12461                                 ctl_set_internal_failure(&io->scsiio,
12462                                                          /*sks_valid*/ 1,
12463                                                          /*retry_count*/ 4857);
12464                                 retval = 1;
12465                                 goto bailout_error;
12466
12467                         }
12468                         /* XXX KDM do we need a sync here? */
12469
12470                         len_to_go -= local_sglist[i].len;
12471                 }
12472                 /*
12473                  * Reset the number of S/G entries accordingly.  The
12474                  * original number of S/G entries is available in
12475                  * 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 (DMA: %d)\n", __func__, i,
12484                                local_sglist[i].addr, local_sglist[i].len,
12485                                local_dma_sglist[i].len);
12486 #endif
12487         }
12488
12489
12490         return (retval);
12491
12492 bailout_error:
12493
12494         ctl_send_datamove_done(io, /*have_lock*/ 0);
12495
12496         return (retval);
12497 }
12498
12499 static int
12500 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12501                          ctl_ha_dt_cb callback)
12502 {
12503         struct ctl_ha_dt_req *rq;
12504         struct ctl_sg_entry *remote_sglist, *local_sglist;
12505         struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
12506         uint32_t local_used, remote_used, total_used;
12507         int retval;
12508         int i, j;
12509
12510         retval = 0;
12511
12512         rq = ctl_dt_req_alloc();
12513
12514         /*
12515          * If we failed to allocate the request, and if the DMA didn't fail
12516          * anyway, set busy status.  This is just a resource allocation
12517          * failure.
12518          */
12519         if ((rq == NULL)
12520          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
12521                 ctl_set_busy(&io->scsiio);
12522
12523         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
12524
12525                 if (rq != NULL)
12526                         ctl_dt_req_free(rq);
12527
12528                 /*
12529                  * The data move failed.  We need to return status back
12530                  * to the other controller.  No point in trying to DMA
12531                  * data to the remote controller.
12532                  */
12533
12534                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12535
12536                 retval = 1;
12537
12538                 goto bailout;
12539         }
12540
12541         local_sglist = io->io_hdr.local_sglist;
12542         local_dma_sglist = io->io_hdr.local_dma_sglist;
12543         remote_sglist = io->io_hdr.remote_sglist;
12544         remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12545         local_used = 0;
12546         remote_used = 0;
12547         total_used = 0;
12548
12549         if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
12550                 rq->ret = CTL_HA_STATUS_SUCCESS;
12551                 rq->context = io;
12552                 callback(rq);
12553                 goto bailout;
12554         }
12555
12556         /*
12557          * Pull/push the data over the wire from/to the other controller.
12558          * This takes into account the possibility that the local and
12559          * remote sglists may not be identical in terms of the size of
12560          * the elements and the number of elements.
12561          *
12562          * One fundamental assumption here is that the length allocated for
12563          * both the local and remote sglists is identical.  Otherwise, we've
12564          * essentially got a coding error of some sort.
12565          */
12566         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12567                 int isc_ret;
12568                 uint32_t cur_len, dma_length;
12569                 uint8_t *tmp_ptr;
12570
12571                 rq->id = CTL_HA_DATA_CTL;
12572                 rq->command = command;
12573                 rq->context = io;
12574
12575                 /*
12576                  * Both pointers should be aligned.  But it is possible
12577                  * that the allocation length is not.  They should both
12578                  * also have enough slack left over at the end, though,
12579                  * to round up to the next 8 byte boundary.
12580                  */
12581                 cur_len = ctl_min(local_sglist[i].len - local_used,
12582                                   remote_sglist[j].len - remote_used);
12583
12584                 /*
12585                  * In this case, we have a size issue and need to decrease
12586                  * the size, except in the case where we actually have less
12587                  * than 8 bytes left.  In that case, we need to increase
12588                  * the DMA length to get the last bit.
12589                  */
12590                 if ((cur_len & 0x7) != 0) {
12591                         if (cur_len > 0x7) {
12592                                 cur_len = cur_len - (cur_len & 0x7);
12593                                 dma_length = cur_len;
12594                         } else {
12595                                 CTL_SIZE_8B(dma_length, cur_len);
12596                         }
12597
12598                 } else
12599                         dma_length = cur_len;
12600
12601                 /*
12602                  * If we had to allocate memory for this I/O, instead of using
12603                  * the non-cached mirror memory, we'll need to flush the cache
12604                  * before trying to DMA to the other controller.
12605                  *
12606                  * We could end up doing this multiple times for the same
12607                  * segment if we have a larger local segment than remote
12608                  * segment.  That shouldn't be an issue.
12609                  */
12610                 if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12611                         /*
12612                          * XXX KDM use bus_dmamap_sync() here.
12613                          */
12614                 }
12615
12616                 rq->size = dma_length;
12617
12618                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
12619                 tmp_ptr += local_used;
12620
12621                 /* Use physical addresses when talking to ISC hardware */
12622                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12623                         /* XXX KDM use busdma */
12624 #if 0
12625                         rq->local = vtophys(tmp_ptr);
12626 #endif
12627                 } else
12628                         rq->local = tmp_ptr;
12629
12630                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12631                 tmp_ptr += remote_used;
12632                 rq->remote = tmp_ptr;
12633
12634                 rq->callback = NULL;
12635
12636                 local_used += cur_len;
12637                 if (local_used >= local_sglist[i].len) {
12638                         i++;
12639                         local_used = 0;
12640                 }
12641
12642                 remote_used += cur_len;
12643                 if (remote_used >= remote_sglist[j].len) {
12644                         j++;
12645                         remote_used = 0;
12646                 }
12647                 total_used += cur_len;
12648
12649                 if (total_used >= io->scsiio.kern_data_len)
12650                         rq->callback = callback;
12651
12652                 if ((rq->size & 0x7) != 0) {
12653                         printf("%s: warning: size %d is not on 8b boundary\n",
12654                                __func__, rq->size);
12655                 }
12656                 if (((uintptr_t)rq->local & 0x7) != 0) {
12657                         printf("%s: warning: local %p not on 8b boundary\n",
12658                                __func__, rq->local);
12659                 }
12660                 if (((uintptr_t)rq->remote & 0x7) != 0) {
12661                         printf("%s: warning: remote %p not on 8b boundary\n",
12662                                __func__, rq->local);
12663                 }
12664 #if 0
12665                 printf("%s: %s: local %#x remote %#x size %d\n", __func__,
12666                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12667                        rq->local, rq->remote, rq->size);
12668 #endif
12669
12670                 isc_ret = ctl_dt_single(rq);
12671                 if (isc_ret == CTL_HA_STATUS_WAIT)
12672                         continue;
12673
12674                 if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
12675                         rq->ret = CTL_HA_STATUS_SUCCESS;
12676                 } else {
12677                         rq->ret = isc_ret;
12678                 }
12679                 callback(rq);
12680                 goto bailout;
12681         }
12682
12683 bailout:
12684         return (retval);
12685
12686 }
12687
12688 static void
12689 ctl_datamove_remote_read(union ctl_io *io)
12690 {
12691         int retval;
12692         int i;
12693
12694         /*
12695          * This will send an error to the other controller in the case of a
12696          * failure.
12697          */
12698         retval = ctl_datamove_remote_sgl_setup(io);
12699         if (retval != 0)
12700                 return;
12701
12702         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12703                                           ctl_datamove_remote_read_cb);
12704         if ((retval != 0)
12705          && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
12706                 /*
12707                  * Make sure we free memory if there was an error..  The
12708                  * ctl_datamove_remote_xfer() function will send the
12709                  * datamove done message, or call the callback with an
12710                  * error if there is a problem.
12711                  */
12712                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12713                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
12714         }
12715
12716         return;
12717 }
12718
12719 /*
12720  * Process a datamove request from the other controller.  This is used for
12721  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12722  * first.  Once that is complete, the data gets DMAed into the remote
12723  * controller's memory.  For reads, we DMA from the remote controller's
12724  * memory into our memory first, and then move it out to the FETD.
12725  */
12726 static void
12727 ctl_datamove_remote(union ctl_io *io)
12728 {
12729         struct ctl_softc *softc;
12730
12731         softc = control_softc;
12732
12733         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
12734
12735         /*
12736          * Note that we look for an aborted I/O here, but don't do some of
12737          * the other checks that ctl_datamove() normally does.  We don't
12738          * need to run the task queue, because this I/O is on the ISC
12739          * queue, which is executed by the work thread after the task queue.
12740          * We don't need to run the datamove delay code, since that should
12741          * have been done if need be on the other controller.
12742          */
12743         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12744
12745                 printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
12746                        io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
12747                        io->io_hdr.nexus.targ_port,
12748                        io->io_hdr.nexus.targ_target.id,
12749                        io->io_hdr.nexus.targ_lun);
12750                 io->io_hdr.status = CTL_CMD_ABORTED;
12751                 io->io_hdr.port_status = 31338;
12752
12753                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12754
12755                 return;
12756         }
12757
12758         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
12759                 ctl_datamove_remote_write(io);
12760         } else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
12761                 ctl_datamove_remote_read(io);
12762         } else {
12763                 union ctl_ha_msg msg;
12764                 struct scsi_sense_data *sense;
12765                 uint8_t sks[3];
12766                 int retry_count;
12767
12768                 memset(&msg, 0, sizeof(msg));
12769
12770                 msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
12771                 msg.hdr.status = CTL_SCSI_ERROR;
12772                 msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
12773
12774                 retry_count = 4243;
12775
12776                 sense = &msg.scsi.sense_data;
12777                 sks[0] = SSD_SCS_VALID;
12778                 sks[1] = (retry_count >> 8) & 0xff;
12779                 sks[2] = retry_count & 0xff;
12780
12781                 /* "Internal target failure" */
12782                 scsi_set_sense_data(sense,
12783                                     /*sense_format*/ SSD_TYPE_NONE,
12784                                     /*current_error*/ 1,
12785                                     /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
12786                                     /*asc*/ 0x44,
12787                                     /*ascq*/ 0x00,
12788                                     /*type*/ SSD_ELEM_SKS,
12789                                     /*size*/ sizeof(sks),
12790                                     /*data*/ sks,
12791                                     SSD_ELEM_NONE);
12792
12793                 io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12794                 if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12795                         ctl_failover_io(io, /*have_lock*/ 1);
12796                         return;
12797                 }
12798
12799                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
12800                     CTL_HA_STATUS_SUCCESS) {
12801                         /* XXX KDM what to do if this fails? */
12802                 }
12803                 return;
12804         }
12805
12806 }
12807
12808 static int
12809 ctl_process_done(union ctl_io *io)
12810 {
12811         struct ctl_lun *lun;
12812         struct ctl_softc *ctl_softc;
12813         void (*fe_done)(union ctl_io *io);
12814         uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
12815
12816         CTL_DEBUG_PRINT(("ctl_process_done\n"));
12817
12818         fe_done =
12819             control_softc->ctl_ports[targ_port]->fe_done;
12820
12821 #ifdef CTL_TIME_IO
12822         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12823                 char str[256];
12824                 char path_str[64];
12825                 struct sbuf sb;
12826
12827                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12828                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12829
12830                 sbuf_cat(&sb, path_str);
12831                 switch (io->io_hdr.io_type) {
12832                 case CTL_IO_SCSI:
12833                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12834                         sbuf_printf(&sb, "\n");
12835                         sbuf_cat(&sb, path_str);
12836                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12837                                     io->scsiio.tag_num, io->scsiio.tag_type);
12838                         break;
12839                 case CTL_IO_TASK:
12840                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12841                                     "Tag Type: %d\n", io->taskio.task_action,
12842                                     io->taskio.tag_num, io->taskio.tag_type);
12843                         break;
12844                 default:
12845                         printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12846                         panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12847                         break;
12848                 }
12849                 sbuf_cat(&sb, path_str);
12850                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12851                             (intmax_t)time_uptime - io->io_hdr.start_time);
12852                 sbuf_finish(&sb);
12853                 printf("%s", sbuf_data(&sb));
12854         }
12855 #endif /* CTL_TIME_IO */
12856
12857         switch (io->io_hdr.io_type) {
12858         case CTL_IO_SCSI:
12859                 break;
12860         case CTL_IO_TASK:
12861                 if (bootverbose || verbose > 0)
12862                         ctl_io_error_print(io, NULL);
12863                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
12864                         ctl_free_io(io);
12865                 else
12866                         fe_done(io);
12867                 return (CTL_RETVAL_COMPLETE);
12868                 break;
12869         default:
12870                 printf("ctl_process_done: invalid io type %d\n",
12871                        io->io_hdr.io_type);
12872                 panic("ctl_process_done: invalid io type %d\n",
12873                       io->io_hdr.io_type);
12874                 break; /* NOTREACHED */
12875         }
12876
12877         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12878         if (lun == NULL) {
12879                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
12880                                  io->io_hdr.nexus.targ_mapped_lun));
12881                 fe_done(io);
12882                 goto bailout;
12883         }
12884         ctl_softc = lun->ctl_softc;
12885
12886         mtx_lock(&lun->lun_lock);
12887
12888         /*
12889          * Check to see if we have any errors to inject here.  We only
12890          * inject errors for commands that don't already have errors set.
12891          */
12892         if ((STAILQ_FIRST(&lun->error_list) != NULL)
12893          && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
12894                 ctl_inject_error(lun, io);
12895
12896         /*
12897          * XXX KDM how do we treat commands that aren't completed
12898          * successfully?
12899          *
12900          * XXX KDM should we also track I/O latency?
12901          */
12902         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
12903             io->io_hdr.io_type == CTL_IO_SCSI) {
12904 #ifdef CTL_TIME_IO
12905                 struct bintime cur_bt;
12906 #endif
12907                 int type;
12908
12909                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12910                     CTL_FLAG_DATA_IN)
12911                         type = CTL_STATS_READ;
12912                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12913                     CTL_FLAG_DATA_OUT)
12914                         type = CTL_STATS_WRITE;
12915                 else
12916                         type = CTL_STATS_NO_IO;
12917
12918                 lun->stats.ports[targ_port].bytes[type] +=
12919                     io->scsiio.kern_total_len;
12920                 lun->stats.ports[targ_port].operations[type]++;
12921 #ifdef CTL_TIME_IO
12922                 bintime_add(&lun->stats.ports[targ_port].dma_time[type],
12923                    &io->io_hdr.dma_bt);
12924                 lun->stats.ports[targ_port].num_dmas[type] +=
12925                     io->io_hdr.num_dmas;
12926                 getbintime(&cur_bt);
12927                 bintime_sub(&cur_bt, &io->io_hdr.start_bt);
12928                 bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
12929 #endif
12930         }
12931
12932         /*
12933          * Remove this from the OOA queue.
12934          */
12935         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
12936
12937         /*
12938          * Run through the blocked queue on this LUN and see if anything
12939          * has become unblocked, now that this transaction is done.
12940          */
12941         ctl_check_blocked(lun);
12942
12943         /*
12944          * If the LUN has been invalidated, free it if there is nothing
12945          * left on its OOA queue.
12946          */
12947         if ((lun->flags & CTL_LUN_INVALID)
12948          && TAILQ_EMPTY(&lun->ooa_queue)) {
12949                 mtx_unlock(&lun->lun_lock);
12950                 mtx_lock(&ctl_softc->ctl_lock);
12951                 ctl_free_lun(lun);
12952                 mtx_unlock(&ctl_softc->ctl_lock);
12953         } else
12954                 mtx_unlock(&lun->lun_lock);
12955
12956         /*
12957          * If this command has been aborted, make sure we set the status
12958          * properly.  The FETD is responsible for freeing the I/O and doing
12959          * whatever it needs to do to clean up its state.
12960          */
12961         if (io->io_hdr.flags & CTL_FLAG_ABORT)
12962                 io->io_hdr.status = CTL_CMD_ABORTED;
12963
12964         /*
12965          * We print out status for every task management command.  For SCSI
12966          * commands, we filter out any unit attention errors; they happen
12967          * on every boot, and would clutter up the log.  Note:  task
12968          * management commands aren't printed here, they are printed above,
12969          * since they should never even make it down here.
12970          */
12971         switch (io->io_hdr.io_type) {
12972         case CTL_IO_SCSI: {
12973                 int error_code, sense_key, asc, ascq;
12974
12975                 sense_key = 0;
12976
12977                 if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
12978                  && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
12979                         /*
12980                          * Since this is just for printing, no need to
12981                          * show errors here.
12982                          */
12983                         scsi_extract_sense_len(&io->scsiio.sense_data,
12984                                                io->scsiio.sense_len,
12985                                                &error_code,
12986                                                &sense_key,
12987                                                &asc,
12988                                                &ascq,
12989                                                /*show_errors*/ 0);
12990                 }
12991
12992                 if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
12993                  && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
12994                   || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
12995                   || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
12996
12997                         if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
12998                                 ctl_softc->skipped_prints++;
12999                         } else {
13000                                 uint32_t skipped_prints;
13001
13002                                 skipped_prints = ctl_softc->skipped_prints;
13003
13004                                 ctl_softc->skipped_prints = 0;
13005                                 ctl_softc->last_print_jiffies = time_uptime;
13006
13007                                 if (skipped_prints > 0) {
13008 #ifdef NEEDTOPORT
13009                                         csevent_log(CSC_CTL | CSC_SHELF_SW |
13010                                             CTL_ERROR_REPORT,
13011                                             csevent_LogType_Trace,
13012                                             csevent_Severity_Information,
13013                                             csevent_AlertLevel_Green,
13014                                             csevent_FRU_Firmware,
13015                                             csevent_FRU_Unknown,
13016                                             "High CTL error volume, %d prints "
13017                                             "skipped", skipped_prints);
13018 #endif
13019                                 }
13020                                 if (bootverbose || verbose > 0)
13021                                         ctl_io_error_print(io, NULL);
13022                         }
13023                 }
13024                 break;
13025         }
13026         case CTL_IO_TASK:
13027                 if (bootverbose || verbose > 0)
13028                         ctl_io_error_print(io, NULL);
13029                 break;
13030         default:
13031                 break;
13032         }
13033
13034         /*
13035          * Tell the FETD or the other shelf controller we're done with this
13036          * command.  Note that only SCSI commands get to this point.  Task
13037          * management commands are completed above.
13038          *
13039          * We only send status to the other controller if we're in XFER
13040          * mode.  In SER_ONLY mode, the I/O is done on the controller that
13041          * received the I/O (from CTL's perspective), and so the status is
13042          * generated there.
13043          *
13044          * XXX KDM if we hold the lock here, we could cause a deadlock
13045          * if the frontend comes back in in this context to queue
13046          * something.
13047          */
13048         if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13049          && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13050                 union ctl_ha_msg msg;
13051
13052                 memset(&msg, 0, sizeof(msg));
13053                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13054                 msg.hdr.original_sc = io->io_hdr.original_sc;
13055                 msg.hdr.nexus = io->io_hdr.nexus;
13056                 msg.hdr.status = io->io_hdr.status;
13057                 msg.scsi.scsi_status = io->scsiio.scsi_status;
13058                 msg.scsi.tag_num = io->scsiio.tag_num;
13059                 msg.scsi.tag_type = io->scsiio.tag_type;
13060                 msg.scsi.sense_len = io->scsiio.sense_len;
13061                 msg.scsi.sense_residual = io->scsiio.sense_residual;
13062                 msg.scsi.residual = io->scsiio.residual;
13063                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13064                        sizeof(io->scsiio.sense_data));
13065                 /*
13066                  * We copy this whether or not this is an I/O-related
13067                  * command.  Otherwise, we'd have to go and check to see
13068                  * whether it's a read/write command, and it really isn't
13069                  * worth it.
13070                  */
13071                 memcpy(&msg.scsi.lbalen,
13072                        &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13073                        sizeof(msg.scsi.lbalen));
13074
13075                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13076                                 sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13077                         /* XXX do something here */
13078                 }
13079
13080                 ctl_free_io(io);
13081         } else
13082                 fe_done(io);
13083
13084 bailout:
13085
13086         return (CTL_RETVAL_COMPLETE);
13087 }
13088
13089 /*
13090  * Front end should call this if it doesn't do autosense.  When the request
13091  * sense comes back in from the initiator, we'll dequeue this and send it.
13092  */
13093 int
13094 ctl_queue_sense(union ctl_io *io)
13095 {
13096         struct ctl_lun *lun;
13097         struct ctl_softc *ctl_softc;
13098         uint32_t initidx, targ_lun;
13099
13100         ctl_softc = control_softc;
13101
13102         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13103
13104         /*
13105          * LUN lookup will likely move to the ctl_work_thread() once we
13106          * have our new queueing infrastructure (that doesn't put things on
13107          * a per-LUN queue initially).  That is so that we can handle
13108          * things like an INQUIRY to a LUN that we don't have enabled.  We
13109          * can't deal with that right now.
13110          */
13111         mtx_lock(&ctl_softc->ctl_lock);
13112
13113         /*
13114          * If we don't have a LUN for this, just toss the sense
13115          * information.
13116          */
13117         targ_lun = io->io_hdr.nexus.targ_lun;
13118         if (io->io_hdr.nexus.lun_map_fn != NULL)
13119                 targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
13120         if ((targ_lun < CTL_MAX_LUNS)
13121          && (ctl_softc->ctl_luns[targ_lun] != NULL))
13122                 lun = ctl_softc->ctl_luns[targ_lun];
13123         else
13124                 goto bailout;
13125
13126         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13127
13128         mtx_lock(&lun->lun_lock);
13129         /*
13130          * Already have CA set for this LUN...toss the sense information.
13131          */
13132         if (ctl_is_set(lun->have_ca, initidx)) {
13133                 mtx_unlock(&lun->lun_lock);
13134                 goto bailout;
13135         }
13136
13137         memcpy(&lun->pending_sense[initidx].sense, &io->scsiio.sense_data,
13138                ctl_min(sizeof(lun->pending_sense[initidx].sense),
13139                sizeof(io->scsiio.sense_data)));
13140         ctl_set_mask(lun->have_ca, initidx);
13141         mtx_unlock(&lun->lun_lock);
13142
13143 bailout:
13144         mtx_unlock(&ctl_softc->ctl_lock);
13145
13146         ctl_free_io(io);
13147
13148         return (CTL_RETVAL_COMPLETE);
13149 }
13150
13151 /*
13152  * Primary command inlet from frontend ports.  All SCSI and task I/O
13153  * requests must go through this function.
13154  */
13155 int
13156 ctl_queue(union ctl_io *io)
13157 {
13158         struct ctl_softc *ctl_softc;
13159
13160         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13161
13162         ctl_softc = control_softc;
13163
13164 #ifdef CTL_TIME_IO
13165         io->io_hdr.start_time = time_uptime;
13166         getbintime(&io->io_hdr.start_bt);
13167 #endif /* CTL_TIME_IO */
13168
13169         /* Map FE-specific LUN ID into global one. */
13170         if (io->io_hdr.nexus.lun_map_fn != NULL)
13171                 io->io_hdr.nexus.targ_mapped_lun = io->io_hdr.nexus.lun_map_fn(
13172                     io->io_hdr.nexus.lun_map_arg, io->io_hdr.nexus.targ_lun);
13173         else
13174                 io->io_hdr.nexus.targ_mapped_lun = io->io_hdr.nexus.targ_lun;
13175
13176         switch (io->io_hdr.io_type) {
13177         case CTL_IO_SCSI:
13178         case CTL_IO_TASK:
13179                 ctl_enqueue_incoming(io);
13180                 break;
13181         default:
13182                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13183                 return (EINVAL);
13184         }
13185
13186         return (CTL_RETVAL_COMPLETE);
13187 }
13188
13189 #ifdef CTL_IO_DELAY
13190 static void
13191 ctl_done_timer_wakeup(void *arg)
13192 {
13193         union ctl_io *io;
13194
13195         io = (union ctl_io *)arg;
13196         ctl_done(io);
13197 }
13198 #endif /* CTL_IO_DELAY */
13199
13200 void
13201 ctl_done(union ctl_io *io)
13202 {
13203         struct ctl_softc *ctl_softc;
13204
13205         ctl_softc = control_softc;
13206
13207         /*
13208          * Enable this to catch duplicate completion issues.
13209          */
13210 #if 0
13211         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13212                 printf("%s: type %d msg %d cdb %x iptl: "
13213                        "%d:%d:%d:%d tag 0x%04x "
13214                        "flag %#x status %x\n",
13215                         __func__,
13216                         io->io_hdr.io_type,
13217                         io->io_hdr.msg_type,
13218                         io->scsiio.cdb[0],
13219                         io->io_hdr.nexus.initid.id,
13220                         io->io_hdr.nexus.targ_port,
13221                         io->io_hdr.nexus.targ_target.id,
13222                         io->io_hdr.nexus.targ_lun,
13223                         (io->io_hdr.io_type ==
13224                         CTL_IO_TASK) ?
13225                         io->taskio.tag_num :
13226                         io->scsiio.tag_num,
13227                         io->io_hdr.flags,
13228                         io->io_hdr.status);
13229         } else
13230                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13231 #endif
13232
13233         /*
13234          * This is an internal copy of an I/O, and should not go through
13235          * the normal done processing logic.
13236          */
13237         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13238                 return;
13239
13240         /*
13241          * We need to send a msg to the serializing shelf to finish the IO
13242          * as well.  We don't send a finish message to the other shelf if
13243          * this is a task management command.  Task management commands
13244          * aren't serialized in the OOA queue, but rather just executed on
13245          * both shelf controllers for commands that originated on that
13246          * controller.
13247          */
13248         if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13249          && (io->io_hdr.io_type != CTL_IO_TASK)) {
13250                 union ctl_ha_msg msg_io;
13251
13252                 msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13253                 msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13254                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13255                     sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13256                 }
13257                 /* continue on to finish IO */
13258         }
13259 #ifdef CTL_IO_DELAY
13260         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13261                 struct ctl_lun *lun;
13262
13263                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13264
13265                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13266         } else {
13267                 struct ctl_lun *lun;
13268
13269                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13270
13271                 if ((lun != NULL)
13272                  && (lun->delay_info.done_delay > 0)) {
13273                         struct callout *callout;
13274
13275                         callout = (struct callout *)&io->io_hdr.timer_bytes;
13276                         callout_init(callout, /*mpsafe*/ 1);
13277                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13278                         callout_reset(callout,
13279                                       lun->delay_info.done_delay * hz,
13280                                       ctl_done_timer_wakeup, io);
13281                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13282                                 lun->delay_info.done_delay = 0;
13283                         return;
13284                 }
13285         }
13286 #endif /* CTL_IO_DELAY */
13287
13288         ctl_enqueue_done(io);
13289 }
13290
13291 int
13292 ctl_isc(struct ctl_scsiio *ctsio)
13293 {
13294         struct ctl_lun *lun;
13295         int retval;
13296
13297         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13298
13299         CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13300
13301         CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13302
13303         retval = lun->backend->data_submit((union ctl_io *)ctsio);
13304
13305         return (retval);
13306 }
13307
13308
13309 static void
13310 ctl_work_thread(void *arg)
13311 {
13312         struct ctl_thread *thr = (struct ctl_thread *)arg;
13313         struct ctl_softc *softc = thr->ctl_softc;
13314         union ctl_io *io;
13315         int retval;
13316
13317         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13318
13319         for (;;) {
13320                 retval = 0;
13321
13322                 /*
13323                  * We handle the queues in this order:
13324                  * - ISC
13325                  * - done queue (to free up resources, unblock other commands)
13326                  * - RtR queue
13327                  * - incoming queue
13328                  *
13329                  * If those queues are empty, we break out of the loop and
13330                  * go to sleep.
13331                  */
13332                 mtx_lock(&thr->queue_lock);
13333                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13334                 if (io != NULL) {
13335                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13336                         mtx_unlock(&thr->queue_lock);
13337                         ctl_handle_isc(io);
13338                         continue;
13339                 }
13340                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13341                 if (io != NULL) {
13342                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13343                         /* clear any blocked commands, call fe_done */
13344                         mtx_unlock(&thr->queue_lock);
13345                         retval = ctl_process_done(io);
13346                         continue;
13347                 }
13348                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13349                 if (io != NULL) {
13350                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13351                         mtx_unlock(&thr->queue_lock);
13352                         if (io->io_hdr.io_type == CTL_IO_TASK)
13353                                 ctl_run_task(io);
13354                         else
13355                                 ctl_scsiio_precheck(softc, &io->scsiio);
13356                         continue;
13357                 }
13358                 if (!ctl_pause_rtr) {
13359                         io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13360                         if (io != NULL) {
13361                                 STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13362                                 mtx_unlock(&thr->queue_lock);
13363                                 retval = ctl_scsiio(&io->scsiio);
13364                                 if (retval != CTL_RETVAL_COMPLETE)
13365                                         CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13366                                 continue;
13367                         }
13368                 }
13369
13370                 /* Sleep until we have something to do. */
13371                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13372         }
13373 }
13374
13375 static void
13376 ctl_lun_thread(void *arg)
13377 {
13378         struct ctl_softc *softc = (struct ctl_softc *)arg;
13379         struct ctl_be_lun *be_lun;
13380         int retval;
13381
13382         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13383
13384         for (;;) {
13385                 retval = 0;
13386                 mtx_lock(&softc->ctl_lock);
13387                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13388                 if (be_lun != NULL) {
13389                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13390                         mtx_unlock(&softc->ctl_lock);
13391                         ctl_create_lun(be_lun);
13392                         continue;
13393                 }
13394
13395                 /* Sleep until we have something to do. */
13396                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13397                     PDROP | PRIBIO, "-", 0);
13398         }
13399 }
13400
13401 static void
13402 ctl_enqueue_incoming(union ctl_io *io)
13403 {
13404         struct ctl_softc *softc = control_softc;
13405         struct ctl_thread *thr;
13406         u_int idx;
13407
13408         idx = (io->io_hdr.nexus.targ_port * 127 +
13409                io->io_hdr.nexus.initid.id) % worker_threads;
13410         thr = &softc->threads[idx];
13411         mtx_lock(&thr->queue_lock);
13412         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13413         mtx_unlock(&thr->queue_lock);
13414         wakeup(thr);
13415 }
13416
13417 static void
13418 ctl_enqueue_rtr(union ctl_io *io)
13419 {
13420         struct ctl_softc *softc = control_softc;
13421         struct ctl_thread *thr;
13422
13423         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13424         mtx_lock(&thr->queue_lock);
13425         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13426         mtx_unlock(&thr->queue_lock);
13427         wakeup(thr);
13428 }
13429
13430 static void
13431 ctl_enqueue_done(union ctl_io *io)
13432 {
13433         struct ctl_softc *softc = control_softc;
13434         struct ctl_thread *thr;
13435
13436         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13437         mtx_lock(&thr->queue_lock);
13438         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13439         mtx_unlock(&thr->queue_lock);
13440         wakeup(thr);
13441 }
13442
13443 static void
13444 ctl_enqueue_isc(union ctl_io *io)
13445 {
13446         struct ctl_softc *softc = control_softc;
13447         struct ctl_thread *thr;
13448
13449         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13450         mtx_lock(&thr->queue_lock);
13451         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13452         mtx_unlock(&thr->queue_lock);
13453         wakeup(thr);
13454 }
13455
13456 /* Initialization and failover */
13457
13458 void
13459 ctl_init_isc_msg(void)
13460 {
13461         printf("CTL: Still calling this thing\n");
13462 }
13463
13464 /*
13465  * Init component
13466  *      Initializes component into configuration defined by bootMode
13467  *      (see hasc-sv.c)
13468  *      returns hasc_Status:
13469  *              OK
13470  *              ERROR - fatal error
13471  */
13472 static ctl_ha_comp_status
13473 ctl_isc_init(struct ctl_ha_component *c)
13474 {
13475         ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13476
13477         c->status = ret;
13478         return ret;
13479 }
13480
13481 /* Start component
13482  *      Starts component in state requested. If component starts successfully,
13483  *      it must set its own state to the requestrd state
13484  *      When requested state is HASC_STATE_HA, the component may refine it
13485  *      by adding _SLAVE or _MASTER flags.
13486  *      Currently allowed state transitions are:
13487  *      UNKNOWN->HA             - initial startup
13488  *      UNKNOWN->SINGLE - initial startup when no parter detected
13489  *      HA->SINGLE              - failover
13490  * returns ctl_ha_comp_status:
13491  *              OK      - component successfully started in requested state
13492  *              FAILED  - could not start the requested state, failover may
13493  *                        be possible
13494  *              ERROR   - fatal error detected, no future startup possible
13495  */
13496 static ctl_ha_comp_status
13497 ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
13498 {
13499         ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13500
13501         printf("%s: go\n", __func__);
13502
13503         // UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
13504         if (c->state == CTL_HA_STATE_UNKNOWN ) {
13505                 ctl_is_single = 0;
13506                 if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
13507                     != CTL_HA_STATUS_SUCCESS) {
13508                         printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
13509                         ret = CTL_HA_COMP_STATUS_ERROR;
13510                 }
13511         } else if (CTL_HA_STATE_IS_HA(c->state)
13512                 && CTL_HA_STATE_IS_SINGLE(state)){
13513                 // HA->SINGLE transition
13514                 ctl_failover();
13515                 ctl_is_single = 1;
13516         } else {
13517                 printf("ctl_isc_start:Invalid state transition %X->%X\n",
13518                        c->state, state);
13519                 ret = CTL_HA_COMP_STATUS_ERROR;
13520         }
13521         if (CTL_HA_STATE_IS_SINGLE(state))
13522                 ctl_is_single = 1;
13523
13524         c->state = state;
13525         c->status = ret;
13526         return ret;
13527 }
13528
13529 /*
13530  * Quiesce component
13531  * The component must clear any error conditions (set status to OK) and
13532  * prepare itself to another Start call
13533  * returns ctl_ha_comp_status:
13534  *      OK
13535  *      ERROR
13536  */
13537 static ctl_ha_comp_status
13538 ctl_isc_quiesce(struct ctl_ha_component *c)
13539 {
13540         int ret = CTL_HA_COMP_STATUS_OK;
13541
13542         ctl_pause_rtr = 1;
13543         c->status = ret;
13544         return ret;
13545 }
13546
13547 struct ctl_ha_component ctl_ha_component_ctlisc =
13548 {
13549         .name = "CTL ISC",
13550         .state = CTL_HA_STATE_UNKNOWN,
13551         .init = ctl_isc_init,
13552         .start = ctl_isc_start,
13553         .quiesce = ctl_isc_quiesce
13554 };
13555
13556 /*
13557  *  vim: ts=8
13558  */