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  * Use the serial number and device ID provided by the backend, rather than
  87  * making up our own.
  88  */
  89 #define CTL_USE_BACKEND_SN
  90
  91 /*
  92  * Size and alignment macros needed for Copan-specific HA hardware.  These
  93  * can go away when the HA code is re-written, and uses busdma for any
  94  * hardware.
  95  */
  96 #define CTL_ALIGN_8B(target, source, type)                              \
  97         if (((uint32_t)source & 0x7) != 0)                              \
  98                 target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
  99         else                                                            \
 100                 target = (type)source;
 101
 102 #define CTL_SIZE_8B(target, size)                                       \
 103         if ((size & 0x7) != 0)                                          \
 104                 target = size + (0x8 - (size & 0x7));                   \
 105         else                                                            \
 106                 target = size;
 107
 108 #define CTL_ALIGN_8B_MARGIN     16
 109
 110 /*
 111  * Template mode pages.
 112  */
 113
 114 /*
 115  * Note that these are default values only.  The actual values will be
 116  * filled in when the user does a mode sense.
 117  */
 118 static struct copan_power_subpage power_page_default = {
 119         /*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
 120         /*subpage*/ PWR_SUBPAGE_CODE,
 121         /*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
 122                          (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
 123         /*page_version*/ PWR_VERSION,
 124         /* total_luns */ 26,
 125         /* max_active_luns*/ PWR_DFLT_MAX_LUNS,
 126         /*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
 127                       0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
 128                       0, 0, 0, 0, 0, 0}
 129 };
 130
 131 static struct copan_power_subpage power_page_changeable = {
 132         /*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
 133         /*subpage*/ PWR_SUBPAGE_CODE,
 134         /*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
 135                          (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
 136         /*page_version*/ 0,
 137         /* total_luns */ 0,
 138         /* max_active_luns*/ 0,
 139         /*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
 140                       0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
 141                       0, 0, 0, 0, 0, 0}
 142 };
 143
 144 static struct copan_aps_subpage aps_page_default = {
 145         APS_PAGE_CODE | SMPH_SPF, //page_code
 146         APS_SUBPAGE_CODE, //subpage
 147         {(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
 148          (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
 149         APS_VERSION, //page_version
 150         0, //lock_active
 151         {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
 152         0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
 153         0, 0, 0, 0, 0} //reserved
 154 };
 155
 156 static struct copan_aps_subpage aps_page_changeable = {
 157         APS_PAGE_CODE | SMPH_SPF, //page_code
 158         APS_SUBPAGE_CODE, //subpage
 159         {(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
 160          (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
 161         0, //page_version
 162         0, //lock_active
 163         {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
 164         0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
 165         0, 0, 0, 0, 0} //reserved
 166 };
 167
 168 static struct copan_debugconf_subpage debugconf_page_default = {
 169         DBGCNF_PAGE_CODE | SMPH_SPF,    /* page_code */
 170         DBGCNF_SUBPAGE_CODE,            /* subpage */
 171         {(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
 172          (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
 173         DBGCNF_VERSION,                 /* page_version */
 174         {CTL_TIME_IO_DEFAULT_SECS>>8,
 175          CTL_TIME_IO_DEFAULT_SECS>>0},  /* ctl_time_io_secs */
 176 };
 177
 178 static struct copan_debugconf_subpage debugconf_page_changeable = {
 179         DBGCNF_PAGE_CODE | SMPH_SPF,    /* page_code */
 180         DBGCNF_SUBPAGE_CODE,            /* subpage */
 181         {(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
 182          (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
 183         0,                              /* page_version */
 184         {0xff,0xff},                    /* ctl_time_io_secs */
 185 };
 186
 187 static struct scsi_format_page format_page_default = {
 188         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
 189         /*page_length*/sizeof(struct scsi_format_page) - 2,
 190         /*tracks_per_zone*/ {0, 0},
 191         /*alt_sectors_per_zone*/ {0, 0},
 192         /*alt_tracks_per_zone*/ {0, 0},
 193         /*alt_tracks_per_lun*/ {0, 0},
 194         /*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
 195                                 CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
 196         /*bytes_per_sector*/ {0, 0},
 197         /*interleave*/ {0, 0},
 198         /*track_skew*/ {0, 0},
 199         /*cylinder_skew*/ {0, 0},
 200         /*flags*/ SFP_HSEC,
 201         /*reserved*/ {0, 0, 0}
 202 };
 203
 204 static struct scsi_format_page format_page_changeable = {
 205         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
 206         /*page_length*/sizeof(struct scsi_format_page) - 2,
 207         /*tracks_per_zone*/ {0, 0},
 208         /*alt_sectors_per_zone*/ {0, 0},
 209         /*alt_tracks_per_zone*/ {0, 0},
 210         /*alt_tracks_per_lun*/ {0, 0},
 211         /*sectors_per_track*/ {0, 0},
 212         /*bytes_per_sector*/ {0, 0},
 213         /*interleave*/ {0, 0},
 214         /*track_skew*/ {0, 0},
 215         /*cylinder_skew*/ {0, 0},
 216         /*flags*/ 0,
 217         /*reserved*/ {0, 0, 0}
 218 };
 219
 220 static struct scsi_rigid_disk_page rigid_disk_page_default = {
 221         /*page_code*/SMS_RIGID_DISK_PAGE,
 222         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
 223         /*cylinders*/ {0, 0, 0},
 224         /*heads*/ CTL_DEFAULT_HEADS,
 225         /*start_write_precomp*/ {0, 0, 0},
 226         /*start_reduced_current*/ {0, 0, 0},
 227         /*step_rate*/ {0, 0},
 228         /*landing_zone_cylinder*/ {0, 0, 0},
 229         /*rpl*/ SRDP_RPL_DISABLED,
 230         /*rotational_offset*/ 0,
 231         /*reserved1*/ 0,
 232         /*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
 233                            CTL_DEFAULT_ROTATION_RATE & 0xff},
 234         /*reserved2*/ {0, 0}
 235 };
 236
 237 static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
 238         /*page_code*/SMS_RIGID_DISK_PAGE,
 239         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
 240         /*cylinders*/ {0, 0, 0},
 241         /*heads*/ 0,
 242         /*start_write_precomp*/ {0, 0, 0},
 243         /*start_reduced_current*/ {0, 0, 0},
 244         /*step_rate*/ {0, 0},
 245         /*landing_zone_cylinder*/ {0, 0, 0},
 246         /*rpl*/ 0,
 247         /*rotational_offset*/ 0,
 248         /*reserved1*/ 0,
 249         /*rotation_rate*/ {0, 0},
 250         /*reserved2*/ {0, 0}
 251 };
 252
 253 static struct scsi_caching_page caching_page_default = {
 254         /*page_code*/SMS_CACHING_PAGE,
 255         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 256         /*flags1*/ SCP_DISC | SCP_WCE,
 257         /*ret_priority*/ 0,
 258         /*disable_pf_transfer_len*/ {0xff, 0xff},
 259         /*min_prefetch*/ {0, 0},
 260         /*max_prefetch*/ {0xff, 0xff},
 261         /*max_pf_ceiling*/ {0xff, 0xff},
 262         /*flags2*/ 0,
 263         /*cache_segments*/ 0,
 264         /*cache_seg_size*/ {0, 0},
 265         /*reserved*/ 0,
 266         /*non_cache_seg_size*/ {0, 0, 0}
 267 };
 268
 269 static struct scsi_caching_page caching_page_changeable = {
 270         /*page_code*/SMS_CACHING_PAGE,
 271         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 272         /*flags1*/ 0,
 273         /*ret_priority*/ 0,
 274         /*disable_pf_transfer_len*/ {0, 0},
 275         /*min_prefetch*/ {0, 0},
 276         /*max_prefetch*/ {0, 0},
 277         /*max_pf_ceiling*/ {0, 0},
 278         /*flags2*/ 0,
 279         /*cache_segments*/ 0,
 280         /*cache_seg_size*/ {0, 0},
 281         /*reserved*/ 0,
 282         /*non_cache_seg_size*/ {0, 0, 0}
 283 };
 284
 285 static struct scsi_control_page control_page_default = {
 286         /*page_code*/SMS_CONTROL_MODE_PAGE,
 287         /*page_length*/sizeof(struct scsi_control_page) - 2,
 288         /*rlec*/0,
 289         /*queue_flags*/0,
 290         /*eca_and_aen*/0,
 291         /*reserved*/0,
 292         /*aen_holdoff_period*/{0, 0}
 293 };
 294
 295 static struct scsi_control_page control_page_changeable = {
 296         /*page_code*/SMS_CONTROL_MODE_PAGE,
 297         /*page_length*/sizeof(struct scsi_control_page) - 2,
 298         /*rlec*/SCP_DSENSE,
 299         /*queue_flags*/0,
 300         /*eca_and_aen*/0,
 301         /*reserved*/0,
 302         /*aen_holdoff_period*/{0, 0}
 303 };
 304
 305
 306 /*
 307  * XXX KDM move these into the softc.
 308  */
 309 static int rcv_sync_msg;
 310 static int persis_offset;
 311 static uint8_t ctl_pause_rtr;
 312 static int     ctl_is_single = 1;
 313 static int     index_to_aps_page;
 314
 315 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
 316 static int worker_threads = -1;
 317 TUNABLE_INT("kern.cam.ctl.worker_threads", &worker_threads);
 318 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
 319     &worker_threads, 1, "Number of worker threads");
 320 static int verbose = 0;
 321 TUNABLE_INT("kern.cam.ctl.verbose", &verbose);
 322 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, verbose, CTLFLAG_RWTUN,
 323     &verbose, 0, "Show SCSI errors returned to initiator");
 324
 325 /*
 326  * Serial number (0x80), device id (0x83), supported pages (0x00),
 327  * Block limits (0xB0) and Logical Block Provisioning (0xB2)
 328  */
 329 #define SCSI_EVPD_NUM_SUPPORTED_PAGES   5
 330
 331 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
 332                                   int param);
 333 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
 334 static int ctl_init(void);
 335 void ctl_shutdown(void);
 336 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
 337 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
 338 static void ctl_ioctl_online(void *arg);
 339 static void ctl_ioctl_offline(void *arg);
 340 static int ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id);
 341 static int ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id);
 342 static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
 343 static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
 344 static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
 345 static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
 346 static int ctl_ioctl_submit_wait(union ctl_io *io);
 347 static void ctl_ioctl_datamove(union ctl_io *io);
 348 static void ctl_ioctl_done(union ctl_io *io);
 349 static void ctl_ioctl_hard_startstop_callback(void *arg,
 350                                               struct cfi_metatask *metatask);
 351 static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
 352 static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
 353                               struct ctl_ooa *ooa_hdr,
 354                               struct ctl_ooa_entry *kern_entries);
 355 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
 356                      struct thread *td);
 357 uint32_t ctl_get_resindex(struct ctl_nexus *nexus);
 358 uint32_t ctl_port_idx(int port_num);
 359 #ifdef unused
 360 static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
 361                                    uint32_t targ_target, uint32_t targ_lun,
 362                                    int can_wait);
 363 static void ctl_kfree_io(union ctl_io *io);
 364 #endif /* unused */
 365 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
 366                          struct ctl_be_lun *be_lun, struct ctl_id target_id);
 367 static int ctl_free_lun(struct ctl_lun *lun);
 368 static void ctl_create_lun(struct ctl_be_lun *be_lun);
 369 /**
 370 static void ctl_failover_change_pages(struct ctl_softc *softc,
 371                                       struct ctl_scsiio *ctsio, int master);
 372 **/
 373
 374 static int ctl_do_mode_select(union ctl_io *io);
 375 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
 376                            uint64_t res_key, uint64_t sa_res_key,
 377                            uint8_t type, uint32_t residx,
 378                            struct ctl_scsiio *ctsio,
 379                            struct scsi_per_res_out *cdb,
 380                            struct scsi_per_res_out_parms* param);
 381 static void ctl_pro_preempt_other(struct ctl_lun *lun,
 382                                   union ctl_ha_msg *msg);
 383 static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
 384 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
 385 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
 386 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
 387 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
 388                                          int alloc_len);
 389 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
 390 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
 391 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
 392 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len);
 393 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
 394 static ctl_action ctl_check_for_blockage(union ctl_io *pending_io,
 395                                          union ctl_io *ooa_io);
 396 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
 397                                 union ctl_io *starting_io);
 398 static int ctl_check_blocked(struct ctl_lun *lun);
 399 static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
 400                                 struct ctl_lun *lun,
 401                                 struct ctl_cmd_entry *entry,
 402                                 struct ctl_scsiio *ctsio);
 403 //static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
 404 static void ctl_failover(void);
 405 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
 406                                struct ctl_scsiio *ctsio);
 407 static int ctl_scsiio(struct ctl_scsiio *ctsio);
 408
 409 static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
 410 static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
 411                             ctl_ua_type ua_type);
 412 static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
 413                          ctl_ua_type ua_type);
 414 static int ctl_abort_task(union ctl_io *io);
 415 static void ctl_run_task(union ctl_io *io);
 416 #ifdef CTL_IO_DELAY
 417 static void ctl_datamove_timer_wakeup(void *arg);
 418 static void ctl_done_timer_wakeup(void *arg);
 419 #endif /* CTL_IO_DELAY */
 420
 421 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
 422 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
 423 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
 424 static void ctl_datamove_remote_write(union ctl_io *io);
 425 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
 426 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
 427 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
 428 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
 429                                     ctl_ha_dt_cb callback);
 430 static void ctl_datamove_remote_read(union ctl_io *io);
 431 static void ctl_datamove_remote(union ctl_io *io);
 432 static int ctl_process_done(union ctl_io *io);
 433 static void ctl_lun_thread(void *arg);
 434 static void ctl_work_thread(void *arg);
 435 static void ctl_enqueue_incoming(union ctl_io *io);
 436 static void ctl_enqueue_rtr(union ctl_io *io);
 437 static void ctl_enqueue_done(union ctl_io *io);
 438 static void ctl_enqueue_isc(union ctl_io *io);
 439
 440 /*
 441  * Load the serialization table.  This isn't very pretty, but is probably
 442  * the easiest way to do it.
 443  */
 444 #include "ctl_ser_table.c"
 445
 446 /*
 447  * We only need to define open, close and ioctl routines for this driver.
 448  */
 449 static struct cdevsw ctl_cdevsw = {
 450         .d_version =    D_VERSION,
 451         .d_flags =      0,
 452         .d_open =       ctl_open,
 453         .d_close =      ctl_close,
 454         .d_ioctl =      ctl_ioctl,
 455         .d_name =       "ctl",
 456 };
 457
 458
 459 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
 460 MALLOC_DEFINE(M_CTLIO, "ctlio", "Memory used for CTL requests");
 461
 462 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
 463
 464 static moduledata_t ctl_moduledata = {
 465         "ctl",
 466         ctl_module_event_handler,
 467         NULL
 468 };
 469
 470 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
 471 MODULE_VERSION(ctl, 1);
 472
 473 static void
 474 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
 475                             union ctl_ha_msg *msg_info)
 476 {
 477         struct ctl_scsiio *ctsio;
 478
 479         if (msg_info->hdr.original_sc == NULL) {
 480                 printf("%s: original_sc == NULL!\n", __func__);
 481                 /* XXX KDM now what? */
 482                 return;
 483         }
 484
 485         ctsio = &msg_info->hdr.original_sc->scsiio;
 486         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 487         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 488         ctsio->io_hdr.status = msg_info->hdr.status;
 489         ctsio->scsi_status = msg_info->scsi.scsi_status;
 490         ctsio->sense_len = msg_info->scsi.sense_len;
 491         ctsio->sense_residual = msg_info->scsi.sense_residual;
 492         ctsio->residual = msg_info->scsi.residual;
 493         memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
 494                sizeof(ctsio->sense_data));
 495         memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
 496                &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
 497         ctl_enqueue_isc((union ctl_io *)ctsio);
 498 }
 499
 500 static void
 501 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
 502                                 union ctl_ha_msg *msg_info)
 503 {
 504         struct ctl_scsiio *ctsio;
 505
 506         if (msg_info->hdr.serializing_sc == NULL) {
 507                 printf("%s: serializing_sc == NULL!\n", __func__);
 508                 /* XXX KDM now what? */
 509                 return;
 510         }
 511
 512         ctsio = &msg_info->hdr.serializing_sc->scsiio;
 513 #if 0
 514         /*
 515          * Attempt to catch the situation where an I/O has
 516          * been freed, and we're using it again.
 517          */
 518         if (ctsio->io_hdr.io_type == 0xff) {
 519                 union ctl_io *tmp_io;
 520                 tmp_io = (union ctl_io *)ctsio;
 521                 printf("%s: %p use after free!\n", __func__,
 522                        ctsio);
 523                 printf("%s: type %d msg %d cdb %x iptl: "
 524                        "%d:%d:%d:%d tag 0x%04x "
 525                        "flag %#x status %x\n",
 526                         __func__,
 527                         tmp_io->io_hdr.io_type,
 528                         tmp_io->io_hdr.msg_type,
 529                         tmp_io->scsiio.cdb[0],
 530                         tmp_io->io_hdr.nexus.initid.id,
 531                         tmp_io->io_hdr.nexus.targ_port,
 532                         tmp_io->io_hdr.nexus.targ_target.id,
 533                         tmp_io->io_hdr.nexus.targ_lun,
 534                         (tmp_io->io_hdr.io_type ==
 535                         CTL_IO_TASK) ?
 536                         tmp_io->taskio.tag_num :
 537                         tmp_io->scsiio.tag_num,
 538                         tmp_io->io_hdr.flags,
 539                         tmp_io->io_hdr.status);
 540         }
 541 #endif
 542         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 543         ctl_enqueue_isc((union ctl_io *)ctsio);
 544 }
 545
 546 /*
 547  * ISC (Inter Shelf Communication) event handler.  Events from the HA
 548  * subsystem come in here.
 549  */
 550 static void
 551 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
 552 {
 553         struct ctl_softc *ctl_softc;
 554         union ctl_io *io;
 555         struct ctl_prio *presio;
 556         ctl_ha_status isc_status;
 557
 558         ctl_softc = control_softc;
 559         io = NULL;
 560
 561
 562 #if 0
 563         printf("CTL: Isc Msg event %d\n", event);
 564 #endif
 565         if (event == CTL_HA_EVT_MSG_RECV) {
 566                 union ctl_ha_msg msg_info;
 567
 568                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
 569                                              sizeof(msg_info), /*wait*/ 0);
 570 #if 0
 571                 printf("CTL: msg_type %d\n", msg_info.msg_type);
 572 #endif
 573                 if (isc_status != 0) {
 574                         printf("Error receiving message, status = %d\n",
 575                                isc_status);
 576                         return;
 577                 }
 578
 579                 switch (msg_info.hdr.msg_type) {
 580                 case CTL_MSG_SERIALIZE:
 581 #if 0
 582                         printf("Serialize\n");
 583 #endif
 584                         io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
 585                         if (io == NULL) {
 586                                 printf("ctl_isc_event_handler: can't allocate "
 587                                        "ctl_io!\n");
 588                                 /* Bad Juju */
 589                                 /* Need to set busy and send msg back */
 590                                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
 591                                 msg_info.hdr.status = CTL_SCSI_ERROR;
 592                                 msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
 593                                 msg_info.scsi.sense_len = 0;
 594                                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
 595                                     sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
 596                                 }
 597                                 goto bailout;
 598                         }
 599                         ctl_zero_io(io);
 600                         // populate ctsio from msg_info
 601                         io->io_hdr.io_type = CTL_IO_SCSI;
 602                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
 603                         io->io_hdr.original_sc = msg_info.hdr.original_sc;
 604 #if 0
 605                         printf("pOrig %x\n", (int)msg_info.original_sc);
 606 #endif
 607                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
 608                                             CTL_FLAG_IO_ACTIVE;
 609                         /*
 610                          * If we're in serialization-only mode, we don't
 611                          * want to go through full done processing.  Thus
 612                          * the COPY flag.
 613                          *
 614                          * XXX KDM add another flag that is more specific.
 615                          */
 616                         if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
 617                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
 618                         io->io_hdr.nexus = msg_info.hdr.nexus;
 619 #if 0
 620                         printf("targ %d, port %d, iid %d, lun %d\n",
 621                                io->io_hdr.nexus.targ_target.id,
 622                                io->io_hdr.nexus.targ_port,
 623                                io->io_hdr.nexus.initid.id,
 624                                io->io_hdr.nexus.targ_lun);
 625 #endif
 626                         io->scsiio.tag_num = msg_info.scsi.tag_num;
 627                         io->scsiio.tag_type = msg_info.scsi.tag_type;
 628                         memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
 629                                CTL_MAX_CDBLEN);
 630                         if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
 631                                 struct ctl_cmd_entry *entry;
 632                                 uint8_t opcode;
 633
 634                                 opcode = io->scsiio.cdb[0];
 635                                 entry = &ctl_cmd_table[opcode];
 636                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
 637                                 io->io_hdr.flags |=
 638                                         entry->flags & CTL_FLAG_DATA_MASK;
 639                         }
 640                         ctl_enqueue_isc(io);
 641                         break;
 642
 643                 /* Performed on the Originating SC, XFER mode only */
 644                 case CTL_MSG_DATAMOVE: {
 645                         struct ctl_sg_entry *sgl;
 646                         int i, j;
 647
 648                         io = msg_info.hdr.original_sc;
 649                         if (io == NULL) {
 650                                 printf("%s: original_sc == NULL!\n", __func__);
 651                                 /* XXX KDM do something here */
 652                                 break;
 653                         }
 654                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
 655                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 656                         /*
 657                          * Keep track of this, we need to send it back over
 658                          * when the datamove is complete.
 659                          */
 660                         io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
 661
 662                         if (msg_info.dt.sg_sequence == 0) {
 663                                 /*
 664                                  * XXX KDM we use the preallocated S/G list
 665                                  * here, but we'll need to change this to
 666                                  * dynamic allocation if we need larger S/G
 667                                  * lists.
 668                                  */
 669                                 if (msg_info.dt.kern_sg_entries >
 670                                     sizeof(io->io_hdr.remote_sglist) /
 671                                     sizeof(io->io_hdr.remote_sglist[0])) {
 672                                         printf("%s: number of S/G entries "
 673                                             "needed %u > allocated num %zd\n",
 674                                             __func__,
 675                                             msg_info.dt.kern_sg_entries,
 676                                             sizeof(io->io_hdr.remote_sglist)/
 677                                             sizeof(io->io_hdr.remote_sglist[0]));
 678
 679                                         /*
 680                                          * XXX KDM send a message back to
 681                                          * the other side to shut down the
 682                                          * DMA.  The error will come back
 683                                          * through via the normal channel.
 684                                          */
 685                                         break;
 686                                 }
 687                                 sgl = io->io_hdr.remote_sglist;
 688                                 memset(sgl, 0,
 689                                        sizeof(io->io_hdr.remote_sglist));
 690
 691                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
 692
 693                                 io->scsiio.kern_sg_entries =
 694                                         msg_info.dt.kern_sg_entries;
 695                                 io->scsiio.rem_sg_entries =
 696                                         msg_info.dt.kern_sg_entries;
 697                                 io->scsiio.kern_data_len =
 698                                         msg_info.dt.kern_data_len;
 699                                 io->scsiio.kern_total_len =
 700                                         msg_info.dt.kern_total_len;
 701                                 io->scsiio.kern_data_resid =
 702                                         msg_info.dt.kern_data_resid;
 703                                 io->scsiio.kern_rel_offset =
 704                                         msg_info.dt.kern_rel_offset;
 705                                 /*
 706                                  * Clear out per-DMA flags.
 707                                  */
 708                                 io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
 709                                 /*
 710                                  * Add per-DMA flags that are set for this
 711                                  * particular DMA request.
 712                                  */
 713                                 io->io_hdr.flags |= msg_info.dt.flags &
 714                                                     CTL_FLAG_RDMA_MASK;
 715                         } else
 716                                 sgl = (struct ctl_sg_entry *)
 717                                         io->scsiio.kern_data_ptr;
 718
 719                         for (i = msg_info.dt.sent_sg_entries, j = 0;
 720                              i < (msg_info.dt.sent_sg_entries +
 721                              msg_info.dt.cur_sg_entries); i++, j++) {
 722                                 sgl[i].addr = msg_info.dt.sg_list[j].addr;
 723                                 sgl[i].len = msg_info.dt.sg_list[j].len;
 724
 725 #if 0
 726                                 printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
 727                                        __func__,
 728                                        msg_info.dt.sg_list[j].addr,
 729                                        msg_info.dt.sg_list[j].len,
 730                                        sgl[i].addr, sgl[i].len, j, i);
 731 #endif
 732                         }
 733 #if 0
 734                         memcpy(&sgl[msg_info.dt.sent_sg_entries],
 735                                msg_info.dt.sg_list,
 736                                sizeof(*sgl) * msg_info.dt.cur_sg_entries);
 737 #endif
 738
 739                         /*
 740                          * If this is the last piece of the I/O, we've got
 741                          * the full S/G list.  Queue processing in the thread.
 742                          * Otherwise wait for the next piece.
 743                          */
 744                         if (msg_info.dt.sg_last != 0)
 745                                 ctl_enqueue_isc(io);
 746                         break;
 747                 }
 748                 /* Performed on the Serializing (primary) SC, XFER mode only */
 749                 case CTL_MSG_DATAMOVE_DONE: {
 750                         if (msg_info.hdr.serializing_sc == NULL) {
 751                                 printf("%s: serializing_sc == NULL!\n",
 752                                        __func__);
 753                                 /* XXX KDM now what? */
 754                                 break;
 755                         }
 756                         /*
 757                          * We grab the sense information here in case
 758                          * there was a failure, so we can return status
 759                          * back to the initiator.
 760                          */
 761                         io = msg_info.hdr.serializing_sc;
 762                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
 763                         io->io_hdr.status = msg_info.hdr.status;
 764                         io->scsiio.scsi_status = msg_info.scsi.scsi_status;
 765                         io->scsiio.sense_len = msg_info.scsi.sense_len;
 766                         io->scsiio.sense_residual =msg_info.scsi.sense_residual;
 767                         io->io_hdr.port_status = msg_info.scsi.fetd_status;
 768                         io->scsiio.residual = msg_info.scsi.residual;
 769                         memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
 770                                sizeof(io->scsiio.sense_data));
 771                         ctl_enqueue_isc(io);
 772                         break;
 773                 }
 774
 775                 /* Preformed on Originating SC, SER_ONLY mode */
 776                 case CTL_MSG_R2R:
 777                         io = msg_info.hdr.original_sc;
 778                         if (io == NULL) {
 779                                 printf("%s: Major Bummer\n", __func__);
 780                                 return;
 781                         } else {
 782 #if 0
 783                                 printf("pOrig %x\n",(int) ctsio);
 784 #endif
 785                         }
 786                         io->io_hdr.msg_type = CTL_MSG_R2R;
 787                         io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
 788                         ctl_enqueue_isc(io);
 789                         break;
 790
 791                 /*
 792                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
 793                  * mode.
 794                  * Performed on the Originating (i.e. secondary) SC in XFER
 795                  * mode
 796                  */
 797                 case CTL_MSG_FINISH_IO:
 798                         if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
 799                                 ctl_isc_handler_finish_xfer(ctl_softc,
 800                                                             &msg_info);
 801                         else
 802                                 ctl_isc_handler_finish_ser_only(ctl_softc,
 803                                                                 &msg_info);
 804                         break;
 805
 806                 /* Preformed on Originating SC */
 807                 case CTL_MSG_BAD_JUJU:
 808                         io = msg_info.hdr.original_sc;
 809                         if (io == NULL) {
 810                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
 811                                        __func__);
 812                                 break;
 813                         }
 814                         ctl_copy_sense_data(&msg_info, io);
 815                         /*
 816                          * IO should have already been cleaned up on other
 817                          * SC so clear this flag so we won't send a message
 818                          * back to finish the IO there.
 819                          */
 820                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
 821                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 822
 823                         /* io = msg_info.hdr.serializing_sc; */
 824                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
 825                         ctl_enqueue_isc(io);
 826                         break;
 827
 828                 /* Handle resets sent from the other side */
 829                 case CTL_MSG_MANAGE_TASKS: {
 830                         struct ctl_taskio *taskio;
 831                         taskio = (struct ctl_taskio *)ctl_alloc_io(
 832                                 (void *)ctl_softc->othersc_pool);
 833                         if (taskio == NULL) {
 834                                 printf("ctl_isc_event_handler: can't allocate "
 835                                        "ctl_io!\n");
 836                                 /* Bad Juju */
 837                                 /* should I just call the proper reset func
 838                                    here??? */
 839                                 goto bailout;
 840                         }
 841                         ctl_zero_io((union ctl_io *)taskio);
 842                         taskio->io_hdr.io_type = CTL_IO_TASK;
 843                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
 844                         taskio->io_hdr.nexus = msg_info.hdr.nexus;
 845                         taskio->task_action = msg_info.task.task_action;
 846                         taskio->tag_num = msg_info.task.tag_num;
 847                         taskio->tag_type = msg_info.task.tag_type;
 848 #ifdef CTL_TIME_IO
 849                         taskio->io_hdr.start_time = time_uptime;
 850                         getbintime(&taskio->io_hdr.start_bt);
 851 #if 0
 852                         cs_prof_gettime(&taskio->io_hdr.start_ticks);
 853 #endif
 854 #endif /* CTL_TIME_IO */
 855                         ctl_run_task((union ctl_io *)taskio);
 856                         break;
 857                 }
 858                 /* Persistent Reserve action which needs attention */
 859                 case CTL_MSG_PERS_ACTION:
 860                         presio = (struct ctl_prio *)ctl_alloc_io(
 861                                 (void *)ctl_softc->othersc_pool);
 862                         if (presio == NULL) {
 863                                 printf("ctl_isc_event_handler: can't allocate "
 864                                        "ctl_io!\n");
 865                                 /* Bad Juju */
 866                                 /* Need to set busy and send msg back */
 867                                 goto bailout;
 868                         }
 869                         ctl_zero_io((union ctl_io *)presio);
 870                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
 871                         presio->pr_msg = msg_info.pr;
 872                         ctl_enqueue_isc((union ctl_io *)presio);
 873                         break;
 874                 case CTL_MSG_SYNC_FE:
 875                         rcv_sync_msg = 1;
 876                         break;
 877                 case CTL_MSG_APS_LOCK: {
 878                         // It's quicker to execute this then to
 879                         // queue it.
 880                         struct ctl_lun *lun;
 881                         struct ctl_page_index *page_index;
 882                         struct copan_aps_subpage *current_sp;
 883                         uint32_t targ_lun;
 884
 885                         targ_lun = msg_info.hdr.nexus.targ_mapped_lun;
 886                         lun = ctl_softc->ctl_luns[targ_lun];
 887                         mtx_lock(&lun->lun_lock);
 888                         page_index = &lun->mode_pages.index[index_to_aps_page];
 889                         current_sp = (struct copan_aps_subpage *)
 890                                      (page_index->page_data +
 891                                      (page_index->page_len * CTL_PAGE_CURRENT));
 892
 893                         current_sp->lock_active = msg_info.aps.lock_flag;
 894                         mtx_unlock(&lun->lun_lock);
 895                         break;
 896                 }
 897                 default:
 898                         printf("How did I get here?\n");
 899                 }
 900         } else if (event == CTL_HA_EVT_MSG_SENT) {
 901                 if (param != CTL_HA_STATUS_SUCCESS) {
 902                         printf("Bad status from ctl_ha_msg_send status %d\n",
 903                                param);
 904                 }
 905                 return;
 906         } else if (event == CTL_HA_EVT_DISCONNECT) {
 907                 printf("CTL: Got a disconnect from Isc\n");
 908                 return;
 909         } else {
 910                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
 911                 return;
 912         }
 913
 914 bailout:
 915         return;
 916 }
 917
 918 static void
 919 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
 920 {
 921         struct scsi_sense_data *sense;
 922
 923         sense = &dest->scsiio.sense_data;
 924         bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
 925         dest->scsiio.scsi_status = src->scsi.scsi_status;
 926         dest->scsiio.sense_len = src->scsi.sense_len;
 927         dest->io_hdr.status = src->hdr.status;
 928 }
 929
 930 static int
 931 ctl_init(void)
 932 {
 933         struct ctl_softc *softc;
 934         struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
 935         struct ctl_frontend *fe;
 936         uint8_t sc_id =0;
 937         int i, error, retval;
 938         //int isc_retval;
 939
 940         retval = 0;
 941         ctl_pause_rtr = 0;
 942         rcv_sync_msg = 0;
 943
 944         control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
 945                                M_WAITOK | M_ZERO);
 946         softc = control_softc;
 947
 948         softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
 949                               "cam/ctl");
 950
 951         softc->dev->si_drv1 = softc;
 952
 953         /*
 954          * By default, return a "bad LUN" peripheral qualifier for unknown
 955          * LUNs.  The user can override this default using the tunable or
 956          * sysctl.  See the comment in ctl_inquiry_std() for more details.
 957          */
 958         softc->inquiry_pq_no_lun = 1;
 959         TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
 960                           &softc->inquiry_pq_no_lun);
 961         sysctl_ctx_init(&softc->sysctl_ctx);
 962         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
 963                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
 964                 CTLFLAG_RD, 0, "CAM Target Layer");
 965
 966         if (softc->sysctl_tree == NULL) {
 967                 printf("%s: unable to allocate sysctl tree\n", __func__);
 968                 destroy_dev(softc->dev);
 969                 free(control_softc, M_DEVBUF);
 970                 control_softc = NULL;
 971                 return (ENOMEM);
 972         }
 973
 974         SYSCTL_ADD_INT(&softc->sysctl_ctx,
 975                        SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
 976                        "inquiry_pq_no_lun", CTLFLAG_RW,
 977                        &softc->inquiry_pq_no_lun, 0,
 978                        "Report no lun possible for invalid LUNs");
 979
 980         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
 981         mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
 982         softc->open_count = 0;
 983
 984         /*
 985          * Default to actually sending a SYNCHRONIZE CACHE command down to
 986          * the drive.
 987          */
 988         softc->flags = CTL_FLAG_REAL_SYNC;
 989
 990         /*
 991          * In Copan's HA scheme, the "master" and "slave" roles are
 992          * figured out through the slot the controller is in.  Although it
 993          * is an active/active system, someone has to be in charge.
 994          */
 995 #ifdef NEEDTOPORT
 996         scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
 997 #endif
 998
 999         if (sc_id == 0) {
1000                 softc->flags |= CTL_FLAG_MASTER_SHELF;
1001                 persis_offset = 0;
1002         } else
1003                 persis_offset = CTL_MAX_INITIATORS;
1004
1005         /*
1006          * XXX KDM need to figure out where we want to get our target ID
1007          * and WWID.  Is it different on each port?
1008          */
1009         softc->target.id = 0;
1010         softc->target.wwid[0] = 0x12345678;
1011         softc->target.wwid[1] = 0x87654321;
1012         STAILQ_INIT(&softc->lun_list);
1013         STAILQ_INIT(&softc->pending_lun_queue);
1014         STAILQ_INIT(&softc->fe_list);
1015         STAILQ_INIT(&softc->be_list);
1016         STAILQ_INIT(&softc->io_pools);
1017
1018         if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1019                             &internal_pool)!= 0){
1020                 printf("ctl: can't allocate %d entry internal pool, "
1021                        "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1022                 return (ENOMEM);
1023         }
1024
1025         if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1026                             CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1027                 printf("ctl: can't allocate %d entry emergency pool, "
1028                        "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1029                 ctl_pool_free(internal_pool);
1030                 return (ENOMEM);
1031         }
1032
1033         if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1034                             &other_pool) != 0)
1035         {
1036                 printf("ctl: can't allocate %d entry other SC pool, "
1037                        "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1038                 ctl_pool_free(internal_pool);
1039                 ctl_pool_free(emergency_pool);
1040                 return (ENOMEM);
1041         }
1042
1043         softc->internal_pool = internal_pool;
1044         softc->emergency_pool = emergency_pool;
1045         softc->othersc_pool = other_pool;
1046
1047         if (worker_threads <= 0)
1048                 worker_threads = max(1, mp_ncpus / 4);
1049         if (worker_threads > CTL_MAX_THREADS)
1050                 worker_threads = CTL_MAX_THREADS;
1051
1052         for (i = 0; i < worker_threads; i++) {
1053                 struct ctl_thread *thr = &softc->threads[i];
1054
1055                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1056                 thr->ctl_softc = softc;
1057                 STAILQ_INIT(&thr->incoming_queue);
1058                 STAILQ_INIT(&thr->rtr_queue);
1059                 STAILQ_INIT(&thr->done_queue);
1060                 STAILQ_INIT(&thr->isc_queue);
1061
1062                 error = kproc_kthread_add(ctl_work_thread, thr,
1063                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1064                 if (error != 0) {
1065                         printf("error creating CTL work thread!\n");
1066                         ctl_pool_free(internal_pool);
1067                         ctl_pool_free(emergency_pool);
1068                         ctl_pool_free(other_pool);
1069                         return (error);
1070                 }
1071         }
1072         error = kproc_kthread_add(ctl_lun_thread, softc,
1073             &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1074         if (error != 0) {
1075                 printf("error creating CTL lun thread!\n");
1076                 ctl_pool_free(internal_pool);
1077                 ctl_pool_free(emergency_pool);
1078                 ctl_pool_free(other_pool);
1079                 return (error);
1080         }
1081         if (bootverbose)
1082                 printf("ctl: CAM Target Layer loaded\n");
1083
1084         /*
1085          * Initialize the initiator and portname mappings
1086          */
1087         memset(softc->wwpn_iid, 0, sizeof(softc->wwpn_iid));
1088
1089         /*
1090          * Initialize the ioctl front end.
1091          */
1092         fe = &softc->ioctl_info.fe;
1093         sprintf(softc->ioctl_info.port_name, "CTL ioctl");
1094         fe->port_type = CTL_PORT_IOCTL;
1095         fe->num_requested_ctl_io = 100;
1096         fe->port_name = softc->ioctl_info.port_name;
1097         fe->port_online = ctl_ioctl_online;
1098         fe->port_offline = ctl_ioctl_offline;
1099         fe->onoff_arg = &softc->ioctl_info;
1100         fe->targ_enable = ctl_ioctl_targ_enable;
1101         fe->targ_disable = ctl_ioctl_targ_disable;
1102         fe->lun_enable = ctl_ioctl_lun_enable;
1103         fe->lun_disable = ctl_ioctl_lun_disable;
1104         fe->targ_lun_arg = &softc->ioctl_info;
1105         fe->fe_datamove = ctl_ioctl_datamove;
1106         fe->fe_done = ctl_ioctl_done;
1107         fe->max_targets = 15;
1108         fe->max_target_id = 15;
1109
1110         if (ctl_frontend_register(&softc->ioctl_info.fe,
1111                           (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1112                 printf("ctl: ioctl front end registration failed, will "
1113                        "continue anyway\n");
1114         }
1115
1116 #ifdef CTL_IO_DELAY
1117         if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1118                 printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1119                        sizeof(struct callout), CTL_TIMER_BYTES);
1120                 return (EINVAL);
1121         }
1122 #endif /* CTL_IO_DELAY */
1123
1124         return (0);
1125 }
1126
1127 void
1128 ctl_shutdown(void)
1129 {
1130         struct ctl_softc *softc;
1131         struct ctl_lun *lun, *next_lun;
1132         struct ctl_io_pool *pool;
1133
1134         softc = (struct ctl_softc *)control_softc;
1135
1136         if (ctl_frontend_deregister(&softc->ioctl_info.fe) != 0)
1137                 printf("ctl: ioctl front end deregistration failed\n");
1138
1139         mtx_lock(&softc->ctl_lock);
1140
1141         /*
1142          * Free up each LUN.
1143          */
1144         for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1145                 next_lun = STAILQ_NEXT(lun, links);
1146                 ctl_free_lun(lun);
1147         }
1148
1149         mtx_unlock(&softc->ctl_lock);
1150
1151         /*
1152          * This will rip the rug out from under any FETDs or anyone else
1153          * that has a pool allocated.  Since we increment our module
1154          * refcount any time someone outside the main CTL module allocates
1155          * a pool, we shouldn't have any problems here.  The user won't be
1156          * able to unload the CTL module until client modules have
1157          * successfully unloaded.
1158          */
1159         while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1160                 ctl_pool_free(pool);
1161
1162 #if 0
1163         ctl_shutdown_thread(softc->work_thread);
1164         mtx_destroy(&softc->queue_lock);
1165 #endif
1166
1167         mtx_destroy(&softc->pool_lock);
1168         mtx_destroy(&softc->ctl_lock);
1169
1170         destroy_dev(softc->dev);
1171
1172         sysctl_ctx_free(&softc->sysctl_ctx);
1173
1174         free(control_softc, M_DEVBUF);
1175         control_softc = NULL;
1176
1177         if (bootverbose)
1178                 printf("ctl: CAM Target Layer unloaded\n");
1179 }
1180
1181 static int
1182 ctl_module_event_handler(module_t mod, int what, void *arg)
1183 {
1184
1185         switch (what) {
1186         case MOD_LOAD:
1187                 return (ctl_init());
1188         case MOD_UNLOAD:
1189                 return (EBUSY);
1190         default:
1191                 return (EOPNOTSUPP);
1192         }
1193 }
1194
1195 /*
1196  * XXX KDM should we do some access checks here?  Bump a reference count to
1197  * prevent a CTL module from being unloaded while someone has it open?
1198  */
1199 static int
1200 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1201 {
1202         return (0);
1203 }
1204
1205 static int
1206 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1207 {
1208         return (0);
1209 }
1210
1211 int
1212 ctl_port_enable(ctl_port_type port_type)
1213 {
1214         struct ctl_softc *softc;
1215         struct ctl_frontend *fe;
1216
1217         if (ctl_is_single == 0) {
1218                 union ctl_ha_msg msg_info;
1219                 int isc_retval;
1220
1221 #if 0
1222                 printf("%s: HA mode, synchronizing frontend enable\n",
1223                         __func__);
1224 #endif
1225                 msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1226                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1227                         sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1228                         printf("Sync msg send error retval %d\n", isc_retval);
1229                 }
1230                 if (!rcv_sync_msg) {
1231                         isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1232                                 sizeof(msg_info), 1);
1233                 }
1234 #if 0
1235                 printf("CTL:Frontend Enable\n");
1236         } else {
1237                 printf("%s: single mode, skipping frontend synchronization\n",
1238                         __func__);
1239 #endif
1240         }
1241
1242         softc = control_softc;
1243
1244         STAILQ_FOREACH(fe, &softc->fe_list, links) {
1245                 if (port_type & fe->port_type)
1246                 {
1247 #if 0
1248                         printf("port %d\n", fe->targ_port);
1249 #endif
1250                         ctl_frontend_online(fe);
1251                 }
1252         }
1253
1254         return (0);
1255 }
1256
1257 int
1258 ctl_port_disable(ctl_port_type port_type)
1259 {
1260         struct ctl_softc *softc;
1261         struct ctl_frontend *fe;
1262
1263         softc = control_softc;
1264
1265         STAILQ_FOREACH(fe, &softc->fe_list, links) {
1266                 if (port_type & fe->port_type)
1267                         ctl_frontend_offline(fe);
1268         }
1269
1270         return (0);
1271 }
1272
1273 /*
1274  * Returns 0 for success, 1 for failure.
1275  * Currently the only failure mode is if there aren't enough entries
1276  * allocated.  So, in case of a failure, look at num_entries_dropped,
1277  * reallocate and try again.
1278  */
1279 int
1280 ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1281               int *num_entries_filled, int *num_entries_dropped,
1282               ctl_port_type port_type, int no_virtual)
1283 {
1284         struct ctl_softc *softc;
1285         struct ctl_frontend *fe;
1286         int entries_dropped, entries_filled;
1287         int retval;
1288         int i;
1289
1290         softc = control_softc;
1291
1292         retval = 0;
1293         entries_filled = 0;
1294         entries_dropped = 0;
1295
1296         i = 0;
1297         mtx_lock(&softc->ctl_lock);
1298         STAILQ_FOREACH(fe, &softc->fe_list, links) {
1299                 struct ctl_port_entry *entry;
1300
1301                 if ((fe->port_type & port_type) == 0)
1302                         continue;
1303
1304                 if ((no_virtual != 0)
1305                  && (fe->virtual_port != 0))
1306                         continue;
1307
1308                 if (entries_filled >= num_entries_alloced) {
1309                         entries_dropped++;
1310                         continue;
1311                 }
1312                 entry = &entries[i];
1313
1314                 entry->port_type = fe->port_type;
1315                 strlcpy(entry->port_name, fe->port_name,
1316                         sizeof(entry->port_name));
1317                 entry->physical_port = fe->physical_port;
1318                 entry->virtual_port = fe->virtual_port;
1319                 entry->wwnn = fe->wwnn;
1320                 entry->wwpn = fe->wwpn;
1321
1322                 i++;
1323                 entries_filled++;
1324         }
1325
1326         mtx_unlock(&softc->ctl_lock);
1327
1328         if (entries_dropped > 0)
1329                 retval = 1;
1330
1331         *num_entries_dropped = entries_dropped;
1332         *num_entries_filled = entries_filled;
1333
1334         return (retval);
1335 }
1336
1337 static void
1338 ctl_ioctl_online(void *arg)
1339 {
1340         struct ctl_ioctl_info *ioctl_info;
1341
1342         ioctl_info = (struct ctl_ioctl_info *)arg;
1343
1344         ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1345 }
1346
1347 static void
1348 ctl_ioctl_offline(void *arg)
1349 {
1350         struct ctl_ioctl_info *ioctl_info;
1351
1352         ioctl_info = (struct ctl_ioctl_info *)arg;
1353
1354         ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1355 }
1356
1357 /*
1358  * Remove an initiator by port number and initiator ID.
1359  * Returns 0 for success, 1 for failure.
1360  */
1361 int
1362 ctl_remove_initiator(int32_t targ_port, uint32_t iid)
1363 {
1364         struct ctl_softc *softc;
1365
1366         softc = control_softc;
1367
1368         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1369
1370         if ((targ_port < 0)
1371          || (targ_port > CTL_MAX_PORTS)) {
1372                 printf("%s: invalid port number %d\n", __func__, targ_port);
1373                 return (1);
1374         }
1375         if (iid > CTL_MAX_INIT_PER_PORT) {
1376                 printf("%s: initiator ID %u > maximun %u!\n",
1377                        __func__, iid, CTL_MAX_INIT_PER_PORT);
1378                 return (1);
1379         }
1380
1381         mtx_lock(&softc->ctl_lock);
1382
1383         softc->wwpn_iid[targ_port][iid].in_use = 0;
1384
1385         mtx_unlock(&softc->ctl_lock);
1386
1387         return (0);
1388 }
1389
1390 /*
1391  * Add an initiator to the initiator map.
1392  * Returns 0 for success, 1 for failure.
1393  */
1394 int
1395 ctl_add_initiator(uint64_t wwpn, int32_t targ_port, uint32_t iid)
1396 {
1397         struct ctl_softc *softc;
1398         int retval;
1399
1400         softc = control_softc;
1401
1402         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1403
1404         retval = 0;
1405
1406         if ((targ_port < 0)
1407          || (targ_port > CTL_MAX_PORTS)) {
1408                 printf("%s: invalid port number %d\n", __func__, targ_port);
1409                 return (1);
1410         }
1411         if (iid > CTL_MAX_INIT_PER_PORT) {
1412                 printf("%s: WWPN %#jx initiator ID %u > maximun %u!\n",
1413                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1414                 return (1);
1415         }
1416
1417         mtx_lock(&softc->ctl_lock);
1418
1419         if (softc->wwpn_iid[targ_port][iid].in_use != 0) {
1420                 /*
1421                  * We don't treat this as an error.
1422                  */
1423                 if (softc->wwpn_iid[targ_port][iid].wwpn == wwpn) {
1424                         printf("%s: port %d iid %u WWPN %#jx arrived again?\n",
1425                                __func__, targ_port, iid, (uintmax_t)wwpn);
1426                         goto bailout;
1427                 }
1428
1429                 /*
1430                  * This is an error, but what do we do about it?  The
1431                  * driver is telling us we have a new WWPN for this
1432                  * initiator ID, so we pretty much need to use it.
1433                  */
1434                 printf("%s: port %d iid %u WWPN %#jx arrived, WWPN %#jx is "
1435                        "still at that address\n", __func__, targ_port, iid,
1436                        (uintmax_t)wwpn,
1437                        (uintmax_t)softc->wwpn_iid[targ_port][iid].wwpn);
1438
1439                 /*
1440                  * XXX KDM clear have_ca and ua_pending on each LUN for
1441                  * this initiator.
1442                  */
1443         }
1444         softc->wwpn_iid[targ_port][iid].in_use = 1;
1445         softc->wwpn_iid[targ_port][iid].iid = iid;
1446         softc->wwpn_iid[targ_port][iid].wwpn = wwpn;
1447         softc->wwpn_iid[targ_port][iid].port = targ_port;
1448
1449 bailout:
1450
1451         mtx_unlock(&softc->ctl_lock);
1452
1453         return (retval);
1454 }
1455
1456 /*
1457  * XXX KDM should we pretend to do something in the target/lun
1458  * enable/disable functions?
1459  */
1460 static int
1461 ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id)
1462 {
1463         return (0);
1464 }
1465
1466 static int
1467 ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id)
1468 {
1469         return (0);
1470 }
1471
1472 static int
1473 ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1474 {
1475         return (0);
1476 }
1477
1478 static int
1479 ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1480 {
1481         return (0);
1482 }
1483
1484 /*
1485  * Data movement routine for the CTL ioctl frontend port.
1486  */
1487 static int
1488 ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1489 {
1490         struct ctl_sg_entry *ext_sglist, *kern_sglist;
1491         struct ctl_sg_entry ext_entry, kern_entry;
1492         int ext_sglen, ext_sg_entries, kern_sg_entries;
1493         int ext_sg_start, ext_offset;
1494         int len_to_copy, len_copied;
1495         int kern_watermark, ext_watermark;
1496         int ext_sglist_malloced;
1497         int i, j;
1498
1499         ext_sglist_malloced = 0;
1500         ext_sg_start = 0;
1501         ext_offset = 0;
1502
1503         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1504
1505         /*
1506          * If this flag is set, fake the data transfer.
1507          */
1508         if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1509                 ctsio->ext_data_filled = ctsio->ext_data_len;
1510                 goto bailout;
1511         }
1512
1513         /*
1514          * To simplify things here, if we have a single buffer, stick it in
1515          * a S/G entry and just make it a single entry S/G list.
1516          */
1517         if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1518                 int len_seen;
1519
1520                 ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1521
1522                 ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1523                                                            M_WAITOK);
1524                 ext_sglist_malloced = 1;
1525                 if (copyin(ctsio->ext_data_ptr, ext_sglist,
1526                                    ext_sglen) != 0) {
1527                         ctl_set_internal_failure(ctsio,
1528                                                  /*sks_valid*/ 0,
1529                                                  /*retry_count*/ 0);
1530                         goto bailout;
1531                 }
1532                 ext_sg_entries = ctsio->ext_sg_entries;
1533                 len_seen = 0;
1534                 for (i = 0; i < ext_sg_entries; i++) {
1535                         if ((len_seen + ext_sglist[i].len) >=
1536                              ctsio->ext_data_filled) {
1537                                 ext_sg_start = i;
1538                                 ext_offset = ctsio->ext_data_filled - len_seen;
1539                                 break;
1540                         }
1541                         len_seen += ext_sglist[i].len;
1542                 }
1543         } else {
1544                 ext_sglist = &ext_entry;
1545                 ext_sglist->addr = ctsio->ext_data_ptr;
1546                 ext_sglist->len = ctsio->ext_data_len;
1547                 ext_sg_entries = 1;
1548                 ext_sg_start = 0;
1549                 ext_offset = ctsio->ext_data_filled;
1550         }
1551
1552         if (ctsio->kern_sg_entries > 0) {
1553                 kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1554                 kern_sg_entries = ctsio->kern_sg_entries;
1555         } else {
1556                 kern_sglist = &kern_entry;
1557                 kern_sglist->addr = ctsio->kern_data_ptr;
1558                 kern_sglist->len = ctsio->kern_data_len;
1559                 kern_sg_entries = 1;
1560         }
1561
1562
1563         kern_watermark = 0;
1564         ext_watermark = ext_offset;
1565         len_copied = 0;
1566         for (i = ext_sg_start, j = 0;
1567              i < ext_sg_entries && j < kern_sg_entries;) {
1568                 uint8_t *ext_ptr, *kern_ptr;
1569
1570                 len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1571                                       kern_sglist[j].len - kern_watermark);
1572
1573                 ext_ptr = (uint8_t *)ext_sglist[i].addr;
1574                 ext_ptr = ext_ptr + ext_watermark;
1575                 if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1576                         /*
1577                          * XXX KDM fix this!
1578                          */
1579                         panic("need to implement bus address support");
1580 #if 0
1581                         kern_ptr = bus_to_virt(kern_sglist[j].addr);
1582 #endif
1583                 } else
1584                         kern_ptr = (uint8_t *)kern_sglist[j].addr;
1585                 kern_ptr = kern_ptr + kern_watermark;
1586
1587                 kern_watermark += len_to_copy;
1588                 ext_watermark += len_to_copy;
1589
1590                 if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1591                      CTL_FLAG_DATA_IN) {
1592                         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1593                                          "bytes to user\n", len_to_copy));
1594                         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1595                                          "to %p\n", kern_ptr, ext_ptr));
1596                         if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1597                                 ctl_set_internal_failure(ctsio,
1598                                                          /*sks_valid*/ 0,
1599                                                          /*retry_count*/ 0);
1600                                 goto bailout;
1601                         }
1602                 } else {
1603                         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1604                                          "bytes from user\n", len_to_copy));
1605                         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1606                                          "to %p\n", ext_ptr, kern_ptr));
1607                         if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1608                                 ctl_set_internal_failure(ctsio,
1609                                                          /*sks_valid*/ 0,
1610                                                          /*retry_count*/0);
1611                                 goto bailout;
1612                         }
1613                 }
1614
1615                 len_copied += len_to_copy;
1616
1617                 if (ext_sglist[i].len == ext_watermark) {
1618                         i++;
1619                         ext_watermark = 0;
1620                 }
1621
1622                 if (kern_sglist[j].len == kern_watermark) {
1623                         j++;
1624                         kern_watermark = 0;
1625                 }
1626         }
1627
1628         ctsio->ext_data_filled += len_copied;
1629
1630         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1631                          "kern_sg_entries: %d\n", ext_sg_entries,
1632                          kern_sg_entries));
1633         CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1634                          "kern_data_len = %d\n", ctsio->ext_data_len,
1635                          ctsio->kern_data_len));
1636
1637
1638         /* XXX KDM set residual?? */
1639 bailout:
1640
1641         if (ext_sglist_malloced != 0)
1642                 free(ext_sglist, M_CTL);
1643
1644         return (CTL_RETVAL_COMPLETE);
1645 }
1646
1647 /*
1648  * Serialize a command that went down the "wrong" side, and so was sent to
1649  * this controller for execution.  The logic is a little different than the
1650  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1651  * sent back to the other side, but in the success case, we execute the
1652  * command on this side (XFER mode) or tell the other side to execute it
1653  * (SER_ONLY mode).
1654  */
1655 static int
1656 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1657 {
1658         struct ctl_softc *ctl_softc;
1659         union ctl_ha_msg msg_info;
1660         struct ctl_lun *lun;
1661         int retval = 0;
1662         uint32_t targ_lun;
1663
1664         ctl_softc = control_softc;
1665
1666         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1667         lun = ctl_softc->ctl_luns[targ_lun];
1668         if (lun==NULL)
1669         {
1670                 /*
1671                  * Why isn't LUN defined? The other side wouldn't
1672                  * send a cmd if the LUN is undefined.
1673                  */
1674                 printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1675
1676                 /* "Logical unit not supported" */
1677                 ctl_set_sense_data(&msg_info.scsi.sense_data,
1678                                    lun,
1679                                    /*sense_format*/SSD_TYPE_NONE,
1680                                    /*current_error*/ 1,
1681                                    /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1682                                    /*asc*/ 0x25,
1683                                    /*ascq*/ 0x00,
1684                                    SSD_ELEM_NONE);
1685
1686                 msg_info.scsi.sense_len = SSD_FULL_SIZE;
1687                 msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1688                 msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1689                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1690                 msg_info.hdr.serializing_sc = NULL;
1691                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1692                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1693                                 sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1694                 }
1695                 return(1);
1696
1697         }
1698
1699         mtx_lock(&lun->lun_lock);
1700         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1701
1702         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1703                 (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1704                  ooa_links))) {
1705         case CTL_ACTION_BLOCK:
1706                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1707                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1708                                   blocked_links);
1709                 break;
1710         case CTL_ACTION_PASS:
1711         case CTL_ACTION_SKIP:
1712                 if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1713                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1714                         ctl_enqueue_rtr((union ctl_io *)ctsio);
1715                 } else {
1716
1717                         /* send msg back to other side */
1718                         msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1719                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1720                         msg_info.hdr.msg_type = CTL_MSG_R2R;
1721 #if 0
1722                         printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1723 #endif
1724                         if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1725                             sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1726                         }
1727                 }
1728                 break;
1729         case CTL_ACTION_OVERLAP:
1730                 /* OVERLAPPED COMMANDS ATTEMPTED */
1731                 ctl_set_sense_data(&msg_info.scsi.sense_data,
1732                                    lun,
1733                                    /*sense_format*/SSD_TYPE_NONE,
1734                                    /*current_error*/ 1,
1735                                    /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1736                                    /*asc*/ 0x4E,
1737                                    /*ascq*/ 0x00,
1738                                    SSD_ELEM_NONE);
1739
1740                 msg_info.scsi.sense_len = SSD_FULL_SIZE;
1741                 msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1742                 msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1743                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1744                 msg_info.hdr.serializing_sc = NULL;
1745                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1746 #if 0
1747                 printf("BAD JUJU:Major Bummer Overlap\n");
1748 #endif
1749                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1750                 retval = 1;
1751                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1752                     sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1753                 }
1754                 break;
1755         case CTL_ACTION_OVERLAP_TAG:
1756                 /* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1757                 ctl_set_sense_data(&msg_info.scsi.sense_data,
1758                                    lun,
1759                                    /*sense_format*/SSD_TYPE_NONE,
1760                                    /*current_error*/ 1,
1761                                    /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1762                                    /*asc*/ 0x4D,
1763                                    /*ascq*/ ctsio->tag_num & 0xff,
1764                                    SSD_ELEM_NONE);
1765
1766                 msg_info.scsi.sense_len = SSD_FULL_SIZE;
1767                 msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1768                 msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1769                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1770                 msg_info.hdr.serializing_sc = NULL;
1771                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1772 #if 0
1773                 printf("BAD JUJU:Major Bummer Overlap Tag\n");
1774 #endif
1775                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1776                 retval = 1;
1777                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1778                     sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1779                 }
1780                 break;
1781         case CTL_ACTION_ERROR:
1782         default:
1783                 /* "Internal target failure" */
1784                 ctl_set_sense_data(&msg_info.scsi.sense_data,
1785                                    lun,
1786                                    /*sense_format*/SSD_TYPE_NONE,
1787                                    /*current_error*/ 1,
1788                                    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1789                                    /*asc*/ 0x44,
1790                                    /*ascq*/ 0x00,
1791                                    SSD_ELEM_NONE);
1792
1793                 msg_info.scsi.sense_len = SSD_FULL_SIZE;
1794                 msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1795                 msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1796                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1797                 msg_info.hdr.serializing_sc = NULL;
1798                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1799 #if 0
1800                 printf("BAD JUJU:Major Bummer HW Error\n");
1801 #endif
1802                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1803                 retval = 1;
1804                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1805                     sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1806                 }
1807                 break;
1808         }
1809         mtx_unlock(&lun->lun_lock);
1810         return (retval);
1811 }
1812
1813 static int
1814 ctl_ioctl_submit_wait(union ctl_io *io)
1815 {
1816         struct ctl_fe_ioctl_params params;
1817         ctl_fe_ioctl_state last_state;
1818         int done, retval;
1819
1820         retval = 0;
1821
1822         bzero(&params, sizeof(params));
1823
1824         mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1825         cv_init(&params.sem, "ctlioccv");
1826         params.state = CTL_IOCTL_INPROG;
1827         last_state = params.state;
1828
1829         io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1830
1831         CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1832
1833         /* This shouldn't happen */
1834         if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1835                 return (retval);
1836
1837         done = 0;
1838
1839         do {
1840                 mtx_lock(&params.ioctl_mtx);
1841                 /*
1842                  * Check the state here, and don't sleep if the state has
1843                  * already changed (i.e. wakeup has already occured, but we
1844                  * weren't waiting yet).
1845                  */
1846                 if (params.state == last_state) {
1847                         /* XXX KDM cv_wait_sig instead? */
1848                         cv_wait(&params.sem, &params.ioctl_mtx);
1849                 }
1850                 last_state = params.state;
1851
1852                 switch (params.state) {
1853                 case CTL_IOCTL_INPROG:
1854                         /* Why did we wake up? */
1855                         /* XXX KDM error here? */
1856                         mtx_unlock(&params.ioctl_mtx);
1857                         break;
1858                 case CTL_IOCTL_DATAMOVE:
1859                         CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1860
1861                         /*
1862                          * change last_state back to INPROG to avoid
1863                          * deadlock on subsequent data moves.
1864                          */
1865                         params.state = last_state = CTL_IOCTL_INPROG;
1866
1867                         mtx_unlock(&params.ioctl_mtx);
1868                         ctl_ioctl_do_datamove(&io->scsiio);
1869                         /*
1870                          * Note that in some cases, most notably writes,
1871                          * this will queue the I/O and call us back later.
1872                          * In other cases, generally reads, this routine
1873                          * will immediately call back and wake us up,
1874                          * probably using our own context.
1875                          */
1876                         io->scsiio.be_move_done(io);
1877                         break;
1878                 case CTL_IOCTL_DONE:
1879                         mtx_unlock(&params.ioctl_mtx);
1880                         CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1881                         done = 1;
1882                         break;
1883                 default:
1884                         mtx_unlock(&params.ioctl_mtx);
1885                         /* XXX KDM error here? */
1886                         break;
1887                 }
1888         } while (done == 0);
1889
1890         mtx_destroy(&params.ioctl_mtx);
1891         cv_destroy(&params.sem);
1892
1893         return (CTL_RETVAL_COMPLETE);
1894 }
1895
1896 static void
1897 ctl_ioctl_datamove(union ctl_io *io)
1898 {
1899         struct ctl_fe_ioctl_params *params;
1900
1901         params = (struct ctl_fe_ioctl_params *)
1902                 io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1903
1904         mtx_lock(&params->ioctl_mtx);
1905         params->state = CTL_IOCTL_DATAMOVE;
1906         cv_broadcast(&params->sem);
1907         mtx_unlock(&params->ioctl_mtx);
1908 }
1909
1910 static void
1911 ctl_ioctl_done(union ctl_io *io)
1912 {
1913         struct ctl_fe_ioctl_params *params;
1914
1915         params = (struct ctl_fe_ioctl_params *)
1916                 io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1917
1918         mtx_lock(&params->ioctl_mtx);
1919         params->state = CTL_IOCTL_DONE;
1920         cv_broadcast(&params->sem);
1921         mtx_unlock(&params->ioctl_mtx);
1922 }
1923
1924 static void
1925 ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
1926 {
1927         struct ctl_fe_ioctl_startstop_info *sd_info;
1928
1929         sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
1930
1931         sd_info->hs_info.status = metatask->status;
1932         sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
1933         sd_info->hs_info.luns_complete =
1934                 metatask->taskinfo.startstop.luns_complete;
1935         sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
1936
1937         cv_broadcast(&sd_info->sem);
1938 }
1939
1940 static void
1941 ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
1942 {
1943         struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
1944
1945         fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
1946
1947         mtx_lock(fe_bbr_info->lock);
1948         fe_bbr_info->bbr_info->status = metatask->status;
1949         fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
1950         fe_bbr_info->wakeup_done = 1;
1951         mtx_unlock(fe_bbr_info->lock);
1952
1953         cv_broadcast(&fe_bbr_info->sem);
1954 }
1955
1956 /*
1957  * Returns 0 for success, errno for failure.
1958  */
1959 static int
1960 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
1961                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
1962 {
1963         union ctl_io *io;
1964         int retval;
1965
1966         retval = 0;
1967
1968         mtx_lock(&lun->lun_lock);
1969         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
1970              (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
1971              ooa_links)) {
1972                 struct ctl_ooa_entry *entry;
1973
1974                 /*
1975                  * If we've got more than we can fit, just count the
1976                  * remaining entries.
1977                  */
1978                 if (*cur_fill_num >= ooa_hdr->alloc_num)
1979                         continue;
1980
1981                 entry = &kern_entries[*cur_fill_num];
1982
1983                 entry->tag_num = io->scsiio.tag_num;
1984                 entry->lun_num = lun->lun;
1985 #ifdef CTL_TIME_IO
1986                 entry->start_bt = io->io_hdr.start_bt;
1987 #endif
1988                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
1989                 entry->cdb_len = io->scsiio.cdb_len;
1990                 if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
1991                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
1992
1993                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
1994                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
1995
1996                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
1997                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
1998
1999                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2000                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2001
2002                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2003                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2004         }
2005         mtx_unlock(&lun->lun_lock);
2006
2007         return (retval);
2008 }
2009
2010 static void *
2011 ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2012                  size_t error_str_len)
2013 {
2014         void *kptr;
2015
2016         kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2017
2018         if (copyin(user_addr, kptr, len) != 0) {
2019                 snprintf(error_str, error_str_len, "Error copying %d bytes "
2020                          "from user address %p to kernel address %p", len,
2021                          user_addr, kptr);
2022                 free(kptr, M_CTL);
2023                 return (NULL);
2024         }
2025
2026         return (kptr);
2027 }
2028
2029 static void
2030 ctl_free_args(int num_be_args, struct ctl_be_arg *be_args)
2031 {
2032         int i;
2033
2034         if (be_args == NULL)
2035                 return;
2036
2037         for (i = 0; i < num_be_args; i++) {
2038                 free(be_args[i].kname, M_CTL);
2039                 free(be_args[i].kvalue, M_CTL);
2040         }
2041
2042         free(be_args, M_CTL);
2043 }
2044
2045 static struct ctl_be_arg *
2046 ctl_copyin_args(int num_be_args, struct ctl_be_arg *be_args,
2047                 char *error_str, size_t error_str_len)
2048 {
2049         struct ctl_be_arg *args;
2050         int i;
2051
2052         args = ctl_copyin_alloc(be_args, num_be_args * sizeof(*be_args),
2053                                 error_str, error_str_len);
2054
2055         if (args == NULL)
2056                 goto bailout;
2057
2058         for (i = 0; i < num_be_args; i++) {
2059                 args[i].kname = NULL;
2060                 args[i].kvalue = NULL;
2061         }
2062
2063         for (i = 0; i < num_be_args; i++) {
2064                 uint8_t *tmpptr;
2065
2066                 args[i].kname = ctl_copyin_alloc(args[i].name,
2067                         args[i].namelen, error_str, error_str_len);
2068                 if (args[i].kname == NULL)
2069                         goto bailout;
2070
2071                 if (args[i].kname[args[i].namelen - 1] != '\0') {
2072                         snprintf(error_str, error_str_len, "Argument %d "
2073                                  "name is not NUL-terminated", i);
2074                         goto bailout;
2075                 }
2076
2077                 args[i].kvalue = NULL;
2078
2079                 tmpptr = ctl_copyin_alloc(args[i].value,
2080                         args[i].vallen, error_str, error_str_len);
2081                 if (tmpptr == NULL)
2082                         goto bailout;
2083
2084                 args[i].kvalue = tmpptr;
2085
2086                 if ((args[i].flags & CTL_BEARG_ASCII)
2087                  && (tmpptr[args[i].vallen - 1] != '\0')) {
2088                         snprintf(error_str, error_str_len, "Argument %d "
2089                                  "value is not NUL-terminated", i);
2090                         goto bailout;
2091                 }
2092         }
2093
2094         return (args);
2095 bailout:
2096
2097         ctl_free_args(num_be_args, args);
2098
2099         return (NULL);
2100 }
2101
2102 /*
2103  * Escape characters that are illegal or not recommended in XML.
2104  */
2105 int
2106 ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2107 {
2108         int retval;
2109
2110         retval = 0;
2111
2112         for (; *str; str++) {
2113                 switch (*str) {
2114                 case '&':
2115                         retval = sbuf_printf(sb, "&amp;");
2116                         break;
2117                 case '>':
2118                         retval = sbuf_printf(sb, "&gt;");
2119                         break;
2120                 case '<':
2121                         retval = sbuf_printf(sb, "&lt;");
2122                         break;
2123                 default:
2124                         retval = sbuf_putc(sb, *str);
2125                         break;
2126                 }
2127
2128                 if (retval != 0)
2129                         break;
2130
2131         }
2132
2133         return (retval);
2134 }
2135
2136 static int
2137 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2138           struct thread *td)
2139 {
2140         struct ctl_softc *softc;
2141         int retval;
2142
2143         softc = control_softc;
2144
2145         retval = 0;
2146
2147         switch (cmd) {
2148         case CTL_IO: {
2149                 union ctl_io *io;
2150                 void *pool_tmp;
2151
2152                 /*
2153                  * If we haven't been "enabled", don't allow any SCSI I/O
2154                  * to this FETD.
2155                  */
2156                 if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2157                         retval = EPERM;
2158                         break;
2159                 }
2160
2161                 io = ctl_alloc_io(softc->ioctl_info.fe.ctl_pool_ref);
2162                 if (io == NULL) {
2163                         printf("ctl_ioctl: can't allocate ctl_io!\n");
2164                         retval = ENOSPC;
2165                         break;
2166                 }
2167
2168                 /*
2169                  * Need to save the pool reference so it doesn't get
2170                  * spammed by the user's ctl_io.
2171                  */
2172                 pool_tmp = io->io_hdr.pool;
2173
2174                 memcpy(io, (void *)addr, sizeof(*io));
2175
2176                 io->io_hdr.pool = pool_tmp;
2177                 /*
2178                  * No status yet, so make sure the status is set properly.
2179                  */
2180                 io->io_hdr.status = CTL_STATUS_NONE;
2181
2182                 /*
2183                  * The user sets the initiator ID, target and LUN IDs.
2184                  */
2185                 io->io_hdr.nexus.targ_port = softc->ioctl_info.fe.targ_port;
2186                 io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2187                 if ((io->io_hdr.io_type == CTL_IO_SCSI)
2188                  && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2189                         io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2190
2191                 retval = ctl_ioctl_submit_wait(io);
2192
2193                 if (retval != 0) {
2194                         ctl_free_io(io);
2195                         break;
2196                 }
2197
2198                 memcpy((void *)addr, io, sizeof(*io));
2199
2200                 /* return this to our pool */
2201                 ctl_free_io(io);
2202
2203                 break;
2204         }
2205         case CTL_ENABLE_PORT:
2206         case CTL_DISABLE_PORT:
2207         case CTL_SET_PORT_WWNS: {
2208                 struct ctl_frontend *fe;
2209                 struct ctl_port_entry *entry;
2210
2211                 entry = (struct ctl_port_entry *)addr;
2212
2213                 mtx_lock(&softc->ctl_lock);
2214                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
2215                         int action, done;
2216
2217                         action = 0;
2218                         done = 0;
2219
2220                         if ((entry->port_type == CTL_PORT_NONE)
2221                          && (entry->targ_port == fe->targ_port)) {
2222                                 /*
2223                                  * If the user only wants to enable or
2224                                  * disable or set WWNs on a specific port,
2225                                  * do the operation and we're done.
2226                                  */
2227                                 action = 1;
2228                                 done = 1;
2229                         } else if (entry->port_type & fe->port_type) {
2230                                 /*
2231                                  * Compare the user's type mask with the
2232                                  * particular frontend type to see if we
2233                                  * have a match.
2234                                  */
2235                                 action = 1;
2236                                 done = 0;
2237
2238                                 /*
2239                                  * Make sure the user isn't trying to set
2240                                  * WWNs on multiple ports at the same time.
2241                                  */
2242                                 if (cmd == CTL_SET_PORT_WWNS) {
2243                                         printf("%s: Can't set WWNs on "
2244                                                "multiple ports\n", __func__);
2245                                         retval = EINVAL;
2246                                         break;
2247                                 }
2248                         }
2249                         if (action != 0) {
2250                                 /*
2251                                  * XXX KDM we have to drop the lock here,
2252                                  * because the online/offline operations
2253                                  * can potentially block.  We need to
2254                                  * reference count the frontends so they
2255                                  * can't go away,
2256                                  */
2257                                 mtx_unlock(&softc->ctl_lock);
2258
2259                                 if (cmd == CTL_ENABLE_PORT) {
2260                                         struct ctl_lun *lun;
2261
2262                                         STAILQ_FOREACH(lun, &softc->lun_list,
2263                                                        links) {
2264                                                 fe->lun_enable(fe->targ_lun_arg,
2265                                                     lun->target,
2266                                                     lun->lun);
2267                                         }
2268
2269                                         ctl_frontend_online(fe);
2270                                 } else if (cmd == CTL_DISABLE_PORT) {
2271                                         struct ctl_lun *lun;
2272
2273                                         ctl_frontend_offline(fe);
2274
2275                                         STAILQ_FOREACH(lun, &softc->lun_list,
2276                                                        links) {
2277                                                 fe->lun_disable(
2278                                                     fe->targ_lun_arg,
2279                                                     lun->target,
2280                                                     lun->lun);
2281                                         }
2282                                 }
2283
2284                                 mtx_lock(&softc->ctl_lock);
2285
2286                                 if (cmd == CTL_SET_PORT_WWNS)
2287                                         ctl_frontend_set_wwns(fe,
2288                                             (entry->flags & CTL_PORT_WWNN_VALID) ?
2289                                             1 : 0, entry->wwnn,
2290                                             (entry->flags & CTL_PORT_WWPN_VALID) ?
2291                                             1 : 0, entry->wwpn);
2292                         }
2293                         if (done != 0)
2294                                 break;
2295                 }
2296                 mtx_unlock(&softc->ctl_lock);
2297                 break;
2298         }
2299         case CTL_GET_PORT_LIST: {
2300                 struct ctl_frontend *fe;
2301                 struct ctl_port_list *list;
2302                 int i;
2303
2304                 list = (struct ctl_port_list *)addr;
2305
2306                 if (list->alloc_len != (list->alloc_num *
2307                     sizeof(struct ctl_port_entry))) {
2308                         printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2309                                "alloc_num %u * sizeof(struct ctl_port_entry) "
2310                                "%zu\n", __func__, list->alloc_len,
2311                                list->alloc_num, sizeof(struct ctl_port_entry));
2312                         retval = EINVAL;
2313                         break;
2314                 }
2315                 list->fill_len = 0;
2316                 list->fill_num = 0;
2317                 list->dropped_num = 0;
2318                 i = 0;
2319                 mtx_lock(&softc->ctl_lock);
2320                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
2321                         struct ctl_port_entry entry, *list_entry;
2322
2323                         if (list->fill_num >= list->alloc_num) {
2324                                 list->dropped_num++;
2325                                 continue;
2326                         }
2327
2328                         entry.port_type = fe->port_type;
2329                         strlcpy(entry.port_name, fe->port_name,
2330                                 sizeof(entry.port_name));
2331                         entry.targ_port = fe->targ_port;
2332                         entry.physical_port = fe->physical_port;
2333                         entry.virtual_port = fe->virtual_port;
2334                         entry.wwnn = fe->wwnn;
2335                         entry.wwpn = fe->wwpn;
2336                         if (fe->status & CTL_PORT_STATUS_ONLINE)
2337                                 entry.online = 1;
2338                         else
2339                                 entry.online = 0;
2340
2341                         list_entry = &list->entries[i];
2342
2343                         retval = copyout(&entry, list_entry, sizeof(entry));
2344                         if (retval != 0) {
2345                                 printf("%s: CTL_GET_PORT_LIST: copyout "
2346                                        "returned %d\n", __func__, retval);
2347                                 break;
2348                         }
2349                         i++;
2350                         list->fill_num++;
2351                         list->fill_len += sizeof(entry);
2352                 }
2353                 mtx_unlock(&softc->ctl_lock);
2354
2355                 /*
2356                  * If this is non-zero, we had a copyout fault, so there's
2357                  * probably no point in attempting to set the status inside
2358                  * the structure.
2359                  */
2360                 if (retval != 0)
2361                         break;
2362
2363                 if (list->dropped_num > 0)
2364                         list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2365                 else
2366                         list->status = CTL_PORT_LIST_OK;
2367                 break;
2368         }
2369         case CTL_DUMP_OOA: {
2370                 struct ctl_lun *lun;
2371                 union ctl_io *io;
2372                 char printbuf[128];
2373                 struct sbuf sb;
2374
2375                 mtx_lock(&softc->ctl_lock);
2376                 printf("Dumping OOA queues:\n");
2377                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2378                         mtx_lock(&lun->lun_lock);
2379                         for (io = (union ctl_io *)TAILQ_FIRST(
2380                              &lun->ooa_queue); io != NULL;
2381                              io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2382                              ooa_links)) {
2383                                 sbuf_new(&sb, printbuf, sizeof(printbuf),
2384                                          SBUF_FIXEDLEN);
2385                                 sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2386                                             (intmax_t)lun->lun,
2387                                             io->scsiio.tag_num,
2388                                             (io->io_hdr.flags &
2389                                             CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2390                                             (io->io_hdr.flags &
2391                                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2392                                             (io->io_hdr.flags &
2393                                             CTL_FLAG_ABORT) ? " ABORT" : "",
2394                                             (io->io_hdr.flags &
2395                                         CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2396                                 ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2397                                 sbuf_finish(&sb);
2398                                 printf("%s\n", sbuf_data(&sb));
2399                         }
2400                         mtx_unlock(&lun->lun_lock);
2401                 }
2402                 printf("OOA queues dump done\n");
2403                 mtx_unlock(&softc->ctl_lock);
2404                 break;
2405         }
2406         case CTL_GET_OOA: {
2407                 struct ctl_lun *lun;
2408                 struct ctl_ooa *ooa_hdr;
2409                 struct ctl_ooa_entry *entries;
2410                 uint32_t cur_fill_num;
2411
2412                 ooa_hdr = (struct ctl_ooa *)addr;
2413
2414                 if ((ooa_hdr->alloc_len == 0)
2415                  || (ooa_hdr->alloc_num == 0)) {
2416                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2417                                "must be non-zero\n", __func__,
2418                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2419                         retval = EINVAL;
2420                         break;
2421                 }
2422
2423                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2424                     sizeof(struct ctl_ooa_entry))) {
2425                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2426                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2427                                __func__, ooa_hdr->alloc_len,
2428                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2429                         retval = EINVAL;
2430                         break;
2431                 }
2432
2433                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2434                 if (entries == NULL) {
2435                         printf("%s: could not allocate %d bytes for OOA "
2436                                "dump\n", __func__, ooa_hdr->alloc_len);
2437                         retval = ENOMEM;
2438                         break;
2439                 }
2440
2441                 mtx_lock(&softc->ctl_lock);
2442                 if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2443                  && ((ooa_hdr->lun_num > CTL_MAX_LUNS)
2444                   || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2445                         mtx_unlock(&softc->ctl_lock);
2446                         free(entries, M_CTL);
2447                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2448                                __func__, (uintmax_t)ooa_hdr->lun_num);
2449                         retval = EINVAL;
2450                         break;
2451                 }
2452
2453                 cur_fill_num = 0;
2454
2455                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2456                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2457                                 retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2458                                         ooa_hdr, entries);
2459                                 if (retval != 0)
2460                                         break;
2461                         }
2462                         if (retval != 0) {
2463                                 mtx_unlock(&softc->ctl_lock);
2464                                 free(entries, M_CTL);
2465                                 break;
2466                         }
2467                 } else {
2468                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2469
2470                         retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2471                                                     entries);
2472                 }
2473                 mtx_unlock(&softc->ctl_lock);
2474
2475                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2476                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2477                         sizeof(struct ctl_ooa_entry);
2478                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2479                 if (retval != 0) {
2480                         printf("%s: error copying out %d bytes for OOA dump\n",
2481                                __func__, ooa_hdr->fill_len);
2482                 }
2483
2484                 getbintime(&ooa_hdr->cur_bt);
2485
2486                 if (cur_fill_num > ooa_hdr->alloc_num) {
2487                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2488                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2489                 } else {
2490                         ooa_hdr->dropped_num = 0;
2491                         ooa_hdr->status = CTL_OOA_OK;
2492                 }
2493
2494                 free(entries, M_CTL);
2495                 break;
2496         }
2497         case CTL_CHECK_OOA: {
2498                 union ctl_io *io;
2499                 struct ctl_lun *lun;
2500                 struct ctl_ooa_info *ooa_info;
2501
2502
2503                 ooa_info = (struct ctl_ooa_info *)addr;
2504
2505                 if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2506                         ooa_info->status = CTL_OOA_INVALID_LUN;
2507                         break;
2508                 }
2509                 mtx_lock(&softc->ctl_lock);
2510                 lun = softc->ctl_luns[ooa_info->lun_id];
2511                 if (lun == NULL) {
2512                         mtx_unlock(&softc->ctl_lock);
2513                         ooa_info->status = CTL_OOA_INVALID_LUN;
2514                         break;
2515                 }
2516                 mtx_lock(&lun->lun_lock);
2517                 mtx_unlock(&softc->ctl_lock);
2518                 ooa_info->num_entries = 0;
2519                 for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2520                      io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2521                      &io->io_hdr, ooa_links)) {
2522                         ooa_info->num_entries++;
2523                 }
2524                 mtx_unlock(&lun->lun_lock);
2525
2526                 ooa_info->status = CTL_OOA_SUCCESS;
2527
2528                 break;
2529         }
2530         case CTL_HARD_START:
2531         case CTL_HARD_STOP: {
2532                 struct ctl_fe_ioctl_startstop_info ss_info;
2533                 struct cfi_metatask *metatask;
2534                 struct mtx hs_mtx;
2535
2536                 mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2537
2538                 cv_init(&ss_info.sem, "hard start/stop cv" );
2539
2540                 metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2541                 if (metatask == NULL) {
2542                         retval = ENOMEM;
2543                         mtx_destroy(&hs_mtx);
2544                         break;
2545                 }
2546
2547                 if (cmd == CTL_HARD_START)
2548                         metatask->tasktype = CFI_TASK_STARTUP;
2549                 else
2550                         metatask->tasktype = CFI_TASK_SHUTDOWN;
2551
2552                 metatask->callback = ctl_ioctl_hard_startstop_callback;
2553                 metatask->callback_arg = &ss_info;
2554
2555                 cfi_action(metatask);
2556
2557                 /* Wait for the callback */
2558                 mtx_lock(&hs_mtx);
2559                 cv_wait_sig(&ss_info.sem, &hs_mtx);
2560                 mtx_unlock(&hs_mtx);
2561
2562                 /*
2563                  * All information has been copied from the metatask by the
2564                  * time cv_broadcast() is called, so we free the metatask here.
2565                  */
2566                 cfi_free_metatask(metatask);
2567
2568                 memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2569
2570                 mtx_destroy(&hs_mtx);
2571                 break;
2572         }
2573         case CTL_BBRREAD: {
2574                 struct ctl_bbrread_info *bbr_info;
2575                 struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2576                 struct mtx bbr_mtx;
2577                 struct cfi_metatask *metatask;
2578
2579                 bbr_info = (struct ctl_bbrread_info *)addr;
2580
2581                 bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2582
2583                 bzero(&bbr_mtx, sizeof(bbr_mtx));
2584                 mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2585
2586                 fe_bbr_info.bbr_info = bbr_info;
2587                 fe_bbr_info.lock = &bbr_mtx;
2588
2589                 cv_init(&fe_bbr_info.sem, "BBR read cv");
2590                 metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2591
2592                 if (metatask == NULL) {
2593                         mtx_destroy(&bbr_mtx);
2594                         cv_destroy(&fe_bbr_info.sem);
2595                         retval = ENOMEM;
2596                         break;
2597                 }
2598                 metatask->tasktype = CFI_TASK_BBRREAD;
2599                 metatask->callback = ctl_ioctl_bbrread_callback;
2600                 metatask->callback_arg = &fe_bbr_info;
2601                 metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2602                 metatask->taskinfo.bbrread.lba = bbr_info->lba;
2603                 metatask->taskinfo.bbrread.len = bbr_info->len;
2604
2605                 cfi_action(metatask);
2606
2607                 mtx_lock(&bbr_mtx);
2608                 while (fe_bbr_info.wakeup_done == 0)
2609                         cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2610                 mtx_unlock(&bbr_mtx);
2611
2612                 bbr_info->status = metatask->status;
2613                 bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2614                 bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2615                 memcpy(&bbr_info->sense_data,
2616                        &metatask->taskinfo.bbrread.sense_data,
2617                        ctl_min(sizeof(bbr_info->sense_data),
2618                                sizeof(metatask->taskinfo.bbrread.sense_data)));
2619
2620                 cfi_free_metatask(metatask);
2621
2622                 mtx_destroy(&bbr_mtx);
2623                 cv_destroy(&fe_bbr_info.sem);
2624
2625                 break;
2626         }
2627         case CTL_DELAY_IO: {
2628                 struct ctl_io_delay_info *delay_info;
2629 #ifdef CTL_IO_DELAY
2630                 struct ctl_lun *lun;
2631 #endif /* CTL_IO_DELAY */
2632
2633                 delay_info = (struct ctl_io_delay_info *)addr;
2634
2635 #ifdef CTL_IO_DELAY
2636                 mtx_lock(&softc->ctl_lock);
2637
2638                 if ((delay_info->lun_id > CTL_MAX_LUNS)
2639                  || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2640                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2641                 } else {
2642                         lun = softc->ctl_luns[delay_info->lun_id];
2643                         mtx_lock(&lun->lun_lock);
2644
2645                         delay_info->status = CTL_DELAY_STATUS_OK;
2646
2647                         switch (delay_info->delay_type) {
2648                         case CTL_DELAY_TYPE_CONT:
2649                                 break;
2650                         case CTL_DELAY_TYPE_ONESHOT:
2651                                 break;
2652                         default:
2653                                 delay_info->status =
2654                                         CTL_DELAY_STATUS_INVALID_TYPE;
2655                                 break;
2656                         }
2657
2658                         switch (delay_info->delay_loc) {
2659                         case CTL_DELAY_LOC_DATAMOVE:
2660                                 lun->delay_info.datamove_type =
2661                                         delay_info->delay_type;
2662                                 lun->delay_info.datamove_delay =
2663                                         delay_info->delay_secs;
2664                                 break;
2665                         case CTL_DELAY_LOC_DONE:
2666                                 lun->delay_info.done_type =
2667                                         delay_info->delay_type;
2668                                 lun->delay_info.done_delay =
2669                                         delay_info->delay_secs;
2670                                 break;
2671                         default:
2672                                 delay_info->status =
2673                                         CTL_DELAY_STATUS_INVALID_LOC;
2674                                 break;
2675                         }
2676                         mtx_unlock(&lun->lun_lock);
2677                 }
2678
2679                 mtx_unlock(&softc->ctl_lock);
2680 #else
2681                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2682 #endif /* CTL_IO_DELAY */
2683                 break;
2684         }
2685         case CTL_REALSYNC_SET: {
2686                 int *syncstate;
2687
2688                 syncstate = (int *)addr;
2689
2690                 mtx_lock(&softc->ctl_lock);
2691                 switch (*syncstate) {
2692                 case 0:
2693                         softc->flags &= ~CTL_FLAG_REAL_SYNC;
2694                         break;
2695                 case 1:
2696                         softc->flags |= CTL_FLAG_REAL_SYNC;
2697                         break;
2698                 default:
2699                         retval = EINVAL;
2700                         break;
2701                 }
2702                 mtx_unlock(&softc->ctl_lock);
2703                 break;
2704         }
2705         case CTL_REALSYNC_GET: {
2706                 int *syncstate;
2707
2708                 syncstate = (int*)addr;
2709
2710                 mtx_lock(&softc->ctl_lock);
2711                 if (softc->flags & CTL_FLAG_REAL_SYNC)
2712                         *syncstate = 1;
2713                 else
2714                         *syncstate = 0;
2715                 mtx_unlock(&softc->ctl_lock);
2716
2717                 break;
2718         }
2719         case CTL_SETSYNC:
2720         case CTL_GETSYNC: {
2721                 struct ctl_sync_info *sync_info;
2722                 struct ctl_lun *lun;
2723
2724                 sync_info = (struct ctl_sync_info *)addr;
2725
2726                 mtx_lock(&softc->ctl_lock);
2727                 lun = softc->ctl_luns[sync_info->lun_id];
2728                 if (lun == NULL) {
2729                         mtx_unlock(&softc->ctl_lock);
2730                         sync_info->status = CTL_GS_SYNC_NO_LUN;
2731                 }
2732                 /*
2733                  * Get or set the sync interval.  We're not bounds checking
2734                  * in the set case, hopefully the user won't do something
2735                  * silly.
2736                  */
2737                 mtx_lock(&lun->lun_lock);
2738                 mtx_unlock(&softc->ctl_lock);
2739                 if (cmd == CTL_GETSYNC)
2740                         sync_info->sync_interval = lun->sync_interval;
2741                 else
2742                         lun->sync_interval = sync_info->sync_interval;
2743                 mtx_unlock(&lun->lun_lock);
2744
2745                 sync_info->status = CTL_GS_SYNC_OK;
2746
2747                 break;
2748         }
2749         case CTL_GETSTATS: {
2750                 struct ctl_stats *stats;
2751                 struct ctl_lun *lun;
2752                 int i;
2753
2754                 stats = (struct ctl_stats *)addr;
2755
2756                 if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2757                      stats->alloc_len) {
2758                         stats->status = CTL_SS_NEED_MORE_SPACE;
2759                         stats->num_luns = softc->num_luns;
2760                         break;
2761                 }
2762                 /*
2763                  * XXX KDM no locking here.  If the LUN list changes,
2764                  * things can blow up.
2765                  */
2766                 for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2767                      i++, lun = STAILQ_NEXT(lun, links)) {
2768                         retval = copyout(&lun->stats, &stats->lun_stats[i],
2769                                          sizeof(lun->stats));
2770                         if (retval != 0)
2771                                 break;
2772                 }
2773                 stats->num_luns = softc->num_luns;
2774                 stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2775                                  softc->num_luns;
2776                 stats->status = CTL_SS_OK;
2777 #ifdef CTL_TIME_IO
2778                 stats->flags = CTL_STATS_FLAG_TIME_VALID;
2779 #else
2780                 stats->flags = CTL_STATS_FLAG_NONE;
2781 #endif
2782                 getnanouptime(&stats->timestamp);
2783                 break;
2784         }
2785         case CTL_ERROR_INJECT: {
2786                 struct ctl_error_desc *err_desc, *new_err_desc;
2787                 struct ctl_lun *lun;
2788
2789                 err_desc = (struct ctl_error_desc *)addr;
2790
2791                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2792                                       M_WAITOK | M_ZERO);
2793                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2794
2795                 mtx_lock(&softc->ctl_lock);
2796                 lun = softc->ctl_luns[err_desc->lun_id];
2797                 if (lun == NULL) {
2798                         mtx_unlock(&softc->ctl_lock);
2799                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2800                                __func__, (uintmax_t)err_desc->lun_id);
2801                         retval = EINVAL;
2802                         break;
2803                 }
2804                 mtx_lock(&lun->lun_lock);
2805                 mtx_unlock(&softc->ctl_lock);
2806
2807                 /*
2808                  * We could do some checking here to verify the validity
2809                  * of the request, but given the complexity of error
2810                  * injection requests, the checking logic would be fairly
2811                  * complex.
2812                  *
2813                  * For now, if the request is invalid, it just won't get
2814                  * executed and might get deleted.
2815                  */
2816                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2817
2818                 /*
2819                  * XXX KDM check to make sure the serial number is unique,
2820                  * in case we somehow manage to wrap.  That shouldn't
2821                  * happen for a very long time, but it's the right thing to
2822                  * do.
2823                  */
2824                 new_err_desc->serial = lun->error_serial;
2825                 err_desc->serial = lun->error_serial;
2826                 lun->error_serial++;
2827
2828                 mtx_unlock(&lun->lun_lock);
2829                 break;
2830         }
2831         case CTL_ERROR_INJECT_DELETE: {
2832                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2833                 struct ctl_lun *lun;
2834                 int delete_done;
2835
2836                 delete_desc = (struct ctl_error_desc *)addr;
2837                 delete_done = 0;
2838
2839                 mtx_lock(&softc->ctl_lock);
2840                 lun = softc->ctl_luns[delete_desc->lun_id];
2841                 if (lun == NULL) {
2842                         mtx_unlock(&softc->ctl_lock);
2843                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2844                                __func__, (uintmax_t)delete_desc->lun_id);
2845                         retval = EINVAL;
2846                         break;
2847                 }
2848                 mtx_lock(&lun->lun_lock);
2849                 mtx_unlock(&softc->ctl_lock);
2850                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2851                         if (desc->serial != delete_desc->serial)
2852                                 continue;
2853
2854                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2855                                       links);
2856                         free(desc, M_CTL);
2857                         delete_done = 1;
2858                 }
2859                 mtx_unlock(&lun->lun_lock);
2860                 if (delete_done == 0) {
2861                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2862                                "error serial %ju on LUN %u\n", __func__,
2863                                delete_desc->serial, delete_desc->lun_id);
2864                         retval = EINVAL;
2865                         break;
2866                 }
2867                 break;
2868         }
2869         case CTL_DUMP_STRUCTS: {
2870                 int i, j, k;
2871                 struct ctl_frontend *fe;
2872
2873                 printf("CTL IID to WWPN map start:\n");
2874                 for (i = 0; i < CTL_MAX_PORTS; i++) {
2875                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2876                                 if (softc->wwpn_iid[i][j].in_use == 0)
2877                                         continue;
2878
2879                                 printf("port %d iid %u WWPN %#jx\n",
2880                                        softc->wwpn_iid[i][j].port,
2881                                        softc->wwpn_iid[i][j].iid,
2882                                        (uintmax_t)softc->wwpn_iid[i][j].wwpn);
2883                         }
2884                 }
2885                 printf("CTL IID to WWPN map end\n");
2886                 printf("CTL Persistent Reservation information start:\n");
2887                 for (i = 0; i < CTL_MAX_LUNS; i++) {
2888                         struct ctl_lun *lun;
2889
2890                         lun = softc->ctl_luns[i];
2891
2892                         if ((lun == NULL)
2893                          || ((lun->flags & CTL_LUN_DISABLED) != 0))
2894                                 continue;
2895
2896                         for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2897                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2898                                         if (lun->per_res[j+k].registered == 0)
2899                                                 continue;
2900                                         printf("LUN %d port %d iid %d key "
2901                                                "%#jx\n", i, j, k,
2902                                                (uintmax_t)scsi_8btou64(
2903                                                lun->per_res[j+k].res_key.key));
2904                                 }
2905                         }
2906                 }
2907                 printf("CTL Persistent Reservation information end\n");
2908                 printf("CTL Frontends:\n");
2909                 /*
2910                  * XXX KDM calling this without a lock.  We'd likely want
2911                  * to drop the lock before calling the frontend's dump
2912                  * routine anyway.
2913                  */
2914                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
2915                         printf("Frontend %s Type %u pport %d vport %d WWNN "
2916                                "%#jx WWPN %#jx\n", fe->port_name, fe->port_type,
2917                                fe->physical_port, fe->virtual_port,
2918                                (uintmax_t)fe->wwnn, (uintmax_t)fe->wwpn);
2919
2920                         /*
2921                          * Frontends are not required to support the dump
2922                          * routine.
2923                          */
2924                         if (fe->fe_dump == NULL)
2925                                 continue;
2926
2927                         fe->fe_dump();
2928                 }
2929                 printf("CTL Frontend information end\n");
2930                 break;
2931         }
2932         case CTL_LUN_REQ: {
2933                 struct ctl_lun_req *lun_req;
2934                 struct ctl_backend_driver *backend;
2935
2936                 lun_req = (struct ctl_lun_req *)addr;
2937
2938                 backend = ctl_backend_find(lun_req->backend);
2939                 if (backend == NULL) {
2940                         lun_req->status = CTL_LUN_ERROR;
2941                         snprintf(lun_req->error_str,
2942                                  sizeof(lun_req->error_str),
2943                                  "Backend \"%s\" not found.",
2944                                  lun_req->backend);
2945                         break;
2946                 }
2947                 if (lun_req->num_be_args > 0) {
2948                         lun_req->kern_be_args = ctl_copyin_args(
2949                                 lun_req->num_be_args,
2950                                 lun_req->be_args,
2951                                 lun_req->error_str,
2952                                 sizeof(lun_req->error_str));
2953                         if (lun_req->kern_be_args == NULL) {
2954                                 lun_req->status = CTL_LUN_ERROR;
2955                                 break;
2956                         }
2957                 }
2958
2959                 retval = backend->ioctl(dev, cmd, addr, flag, td);
2960
2961                 if (lun_req->num_be_args > 0) {
2962                         ctl_free_args(lun_req->num_be_args,
2963                                       lun_req->kern_be_args);
2964                 }
2965                 break;
2966         }
2967         case CTL_LUN_LIST: {
2968                 struct sbuf *sb;
2969                 struct ctl_lun *lun;
2970                 struct ctl_lun_list *list;
2971                 struct ctl_be_lun_option *opt;
2972
2973                 list = (struct ctl_lun_list *)addr;
2974
2975                 /*
2976                  * Allocate a fixed length sbuf here, based on the length
2977                  * of the user's buffer.  We could allocate an auto-extending
2978                  * buffer, and then tell the user how much larger our
2979                  * amount of data is than his buffer, but that presents
2980                  * some problems:
2981                  *
2982                  * 1.  The sbuf(9) routines use a blocking malloc, and so
2983                  *     we can't hold a lock while calling them with an
2984                  *     auto-extending buffer.
2985                  *
2986                  * 2.  There is not currently a LUN reference counting
2987                  *     mechanism, outside of outstanding transactions on
2988                  *     the LUN's OOA queue.  So a LUN could go away on us
2989                  *     while we're getting the LUN number, backend-specific
2990                  *     information, etc.  Thus, given the way things
2991                  *     currently work, we need to hold the CTL lock while
2992                  *     grabbing LUN information.
2993                  *
2994                  * So, from the user's standpoint, the best thing to do is
2995                  * allocate what he thinks is a reasonable buffer length,
2996                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
2997                  * double the buffer length and try again.  (And repeat
2998                  * that until he succeeds.)
2999                  */
3000                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3001                 if (sb == NULL) {
3002                         list->status = CTL_LUN_LIST_ERROR;
3003                         snprintf(list->error_str, sizeof(list->error_str),
3004                                  "Unable to allocate %d bytes for LUN list",
3005                                  list->alloc_len);
3006                         break;
3007                 }
3008
3009                 sbuf_printf(sb, "<ctllunlist>\n");
3010
3011                 mtx_lock(&softc->ctl_lock);
3012                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3013                         mtx_lock(&lun->lun_lock);
3014                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3015                                              (uintmax_t)lun->lun);
3016
3017                         /*
3018                          * Bail out as soon as we see that we've overfilled
3019                          * the buffer.
3020                          */
3021                         if (retval != 0)
3022                                 break;
3023
3024                         retval = sbuf_printf(sb, "\t<backend_type>%s"
3025                                              "</backend_type>\n",
3026                                              (lun->backend == NULL) ?  "none" :
3027                                              lun->backend->name);
3028
3029                         if (retval != 0)
3030                                 break;
3031
3032                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3033                                              lun->be_lun->lun_type);
3034
3035                         if (retval != 0)
3036                                 break;
3037
3038                         if (lun->backend == NULL) {
3039                                 retval = sbuf_printf(sb, "</lun>\n");
3040                                 if (retval != 0)
3041                                         break;
3042                                 continue;
3043                         }
3044
3045                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3046                                              (lun->be_lun->maxlba > 0) ?
3047                                              lun->be_lun->maxlba + 1 : 0);
3048
3049                         if (retval != 0)
3050                                 break;
3051
3052                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3053                                              lun->be_lun->blocksize);
3054
3055                         if (retval != 0)
3056                                 break;
3057
3058                         retval = sbuf_printf(sb, "\t<serial_number>");
3059
3060                         if (retval != 0)
3061                                 break;
3062
3063                         retval = ctl_sbuf_printf_esc(sb,
3064                                                      lun->be_lun->serial_num);
3065
3066                         if (retval != 0)
3067                                 break;
3068
3069                         retval = sbuf_printf(sb, "</serial_number>\n");
3070
3071                         if (retval != 0)
3072                                 break;
3073
3074                         retval = sbuf_printf(sb, "\t<device_id>");
3075
3076                         if (retval != 0)
3077                                 break;
3078
3079                         retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3080
3081                         if (retval != 0)
3082                                 break;
3083
3084                         retval = sbuf_printf(sb, "</device_id>\n");
3085
3086                         if (retval != 0)
3087                                 break;
3088
3089                         if (lun->backend->lun_info != NULL) {
3090                                 retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3091                                 if (retval != 0)
3092                                         break;
3093                         }
3094                         STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3095                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3096                                     opt->name, opt->value, opt->name);
3097                                 if (retval != 0)
3098                                         break;
3099                         }
3100
3101                         retval = sbuf_printf(sb, "</lun>\n");
3102
3103                         if (retval != 0)
3104                                 break;
3105                         mtx_unlock(&lun->lun_lock);
3106                 }
3107                 if (lun != NULL)
3108                         mtx_unlock(&lun->lun_lock);
3109                 mtx_unlock(&softc->ctl_lock);
3110
3111                 if ((retval != 0)
3112                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3113                         retval = 0;
3114                         sbuf_delete(sb);
3115                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3116                         snprintf(list->error_str, sizeof(list->error_str),
3117                                  "Out of space, %d bytes is too small",
3118                                  list->alloc_len);
3119                         break;
3120                 }
3121
3122                 sbuf_finish(sb);
3123
3124                 retval = copyout(sbuf_data(sb), list->lun_xml,
3125                                  sbuf_len(sb) + 1);
3126
3127                 list->fill_len = sbuf_len(sb) + 1;
3128                 list->status = CTL_LUN_LIST_OK;
3129                 sbuf_delete(sb);
3130                 break;
3131         }
3132         case CTL_ISCSI: {
3133                 struct ctl_iscsi *ci;
3134                 struct ctl_frontend *fe;
3135
3136                 ci = (struct ctl_iscsi *)addr;
3137
3138                 mtx_lock(&softc->ctl_lock);
3139                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
3140                         if (strcmp(fe->port_name, "iscsi") == 0)
3141                                 break;
3142                 }
3143                 mtx_unlock(&softc->ctl_lock);
3144
3145                 if (fe == NULL) {
3146                         ci->status = CTL_ISCSI_ERROR;
3147                         snprintf(ci->error_str, sizeof(ci->error_str), "Backend \"iscsi\" not found.");
3148                         break;
3149                 }
3150
3151                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3152                 break;
3153         }
3154         default: {
3155                 /* XXX KDM should we fix this? */
3156 #if 0
3157                 struct ctl_backend_driver *backend;
3158                 unsigned int type;
3159                 int found;
3160
3161                 found = 0;
3162
3163                 /*
3164                  * We encode the backend type as the ioctl type for backend
3165                  * ioctls.  So parse it out here, and then search for a
3166                  * backend of this type.
3167                  */
3168                 type = _IOC_TYPE(cmd);
3169
3170                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3171                         if (backend->type == type) {
3172                                 found = 1;
3173                                 break;
3174                         }
3175                 }
3176                 if (found == 0) {
3177                         printf("ctl: unknown ioctl command %#lx or backend "
3178                                "%d\n", cmd, type);
3179                         retval = EINVAL;
3180                         break;
3181                 }
3182                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3183 #endif
3184                 retval = ENOTTY;
3185                 break;
3186         }
3187         }
3188         return (retval);
3189 }
3190
3191 uint32_t
3192 ctl_get_initindex(struct ctl_nexus *nexus)
3193 {
3194         if (nexus->targ_port < CTL_MAX_PORTS)
3195                 return (nexus->initid.id +
3196                         (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3197         else
3198                 return (nexus->initid.id +
3199                        ((nexus->targ_port - CTL_MAX_PORTS) *
3200                         CTL_MAX_INIT_PER_PORT));
3201 }
3202
3203 uint32_t
3204 ctl_get_resindex(struct ctl_nexus *nexus)
3205 {
3206         return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3207 }
3208
3209 uint32_t
3210 ctl_port_idx(int port_num)
3211 {
3212         if (port_num < CTL_MAX_PORTS)
3213                 return(port_num);
3214         else
3215                 return(port_num - CTL_MAX_PORTS);
3216 }
3217
3218 /*
3219  * Note:  This only works for bitmask sizes that are at least 32 bits, and
3220  * that are a power of 2.
3221  */
3222 int
3223 ctl_ffz(uint32_t *mask, uint32_t size)
3224 {
3225         uint32_t num_chunks, num_pieces;
3226         int i, j;
3227
3228         num_chunks = (size >> 5);
3229         if (num_chunks == 0)
3230                 num_chunks++;
3231         num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3232
3233         for (i = 0; i < num_chunks; i++) {
3234                 for (j = 0; j < num_pieces; j++) {
3235                         if ((mask[i] & (1 << j)) == 0)
3236                                 return ((i << 5) + j);
3237                 }
3238         }
3239
3240         return (-1);
3241 }
3242
3243 int
3244 ctl_set_mask(uint32_t *mask, uint32_t bit)
3245 {
3246         uint32_t chunk, piece;
3247
3248         chunk = bit >> 5;
3249         piece = bit % (sizeof(uint32_t) * 8);
3250
3251         if ((mask[chunk] & (1 << piece)) != 0)
3252                 return (-1);
3253         else
3254                 mask[chunk] |= (1 << piece);
3255
3256         return (0);
3257 }
3258
3259 int
3260 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3261 {
3262         uint32_t chunk, piece;
3263
3264         chunk = bit >> 5;
3265         piece = bit % (sizeof(uint32_t) * 8);
3266
3267         if ((mask[chunk] & (1 << piece)) == 0)
3268                 return (-1);
3269         else
3270                 mask[chunk] &= ~(1 << piece);
3271
3272         return (0);
3273 }
3274
3275 int
3276 ctl_is_set(uint32_t *mask, uint32_t bit)
3277 {
3278         uint32_t chunk, piece;
3279
3280         chunk = bit >> 5;
3281         piece = bit % (sizeof(uint32_t) * 8);
3282
3283         if ((mask[chunk] & (1 << piece)) == 0)
3284                 return (0);
3285         else
3286                 return (1);
3287 }
3288
3289 #ifdef unused
3290 /*
3291  * The bus, target and lun are optional, they can be filled in later.
3292  * can_wait is used to determine whether we can wait on the malloc or not.
3293  */
3294 union ctl_io*
3295 ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3296               uint32_t targ_lun, int can_wait)
3297 {
3298         union ctl_io *io;
3299
3300         if (can_wait)
3301                 io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3302         else
3303                 io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3304
3305         if (io != NULL) {
3306                 io->io_hdr.io_type = io_type;
3307                 io->io_hdr.targ_port = targ_port;
3308                 /*
3309                  * XXX KDM this needs to change/go away.  We need to move
3310                  * to a preallocated pool of ctl_scsiio structures.
3311                  */
3312                 io->io_hdr.nexus.targ_target.id = targ_target;
3313                 io->io_hdr.nexus.targ_lun = targ_lun;
3314         }
3315
3316         return (io);
3317 }
3318
3319 void
3320 ctl_kfree_io(union ctl_io *io)
3321 {
3322         free(io, M_CTL);
3323 }
3324 #endif /* unused */
3325
3326 /*
3327  * ctl_softc, pool_type, total_ctl_io are passed in.
3328  * npool is passed out.
3329  */
3330 int
3331 ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3332                 uint32_t total_ctl_io, struct ctl_io_pool **npool)
3333 {
3334         uint32_t i;
3335         union ctl_io *cur_io, *next_io;
3336         struct ctl_io_pool *pool;
3337         int retval;
3338
3339         retval = 0;
3340
3341         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3342                                             M_NOWAIT | M_ZERO);
3343         if (pool == NULL) {
3344                 retval = ENOMEM;
3345                 goto bailout;
3346         }
3347
3348         pool->type = pool_type;
3349         pool->ctl_softc = ctl_softc;
3350
3351         mtx_lock(&ctl_softc->pool_lock);
3352         pool->id = ctl_softc->cur_pool_id++;
3353         mtx_unlock(&ctl_softc->pool_lock);
3354
3355         pool->flags = CTL_POOL_FLAG_NONE;
3356         pool->refcount = 1;             /* Reference for validity. */
3357         STAILQ_INIT(&pool->free_queue);
3358
3359         /*
3360          * XXX KDM other options here:
3361          * - allocate a page at a time
3362          * - allocate one big chunk of memory.
3363          * Page allocation might work well, but would take a little more
3364          * tracking.
3365          */
3366         for (i = 0; i < total_ctl_io; i++) {
3367                 cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTLIO,
3368                                                 M_NOWAIT);
3369                 if (cur_io == NULL) {
3370                         retval = ENOMEM;
3371                         break;
3372                 }
3373                 cur_io->io_hdr.pool = pool;
3374                 STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3375                 pool->total_ctl_io++;
3376                 pool->free_ctl_io++;
3377         }
3378
3379         if (retval != 0) {
3380                 for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3381                      cur_io != NULL; cur_io = next_io) {
3382                         next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3383                                                               links);
3384                         STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3385                                       ctl_io_hdr, links);
3386                         free(cur_io, M_CTLIO);
3387                 }
3388
3389                 free(pool, M_CTL);
3390                 goto bailout;
3391         }
3392         mtx_lock(&ctl_softc->pool_lock);
3393         ctl_softc->num_pools++;
3394         STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3395         /*
3396          * Increment our usage count if this is an external consumer, so we
3397          * can't get unloaded until the external consumer (most likely a
3398          * FETD) unloads and frees his pool.
3399          *
3400          * XXX KDM will this increment the caller's module use count, or
3401          * mine?
3402          */
3403 #if 0
3404         if ((pool_type != CTL_POOL_EMERGENCY)
3405          && (pool_type != CTL_POOL_INTERNAL)
3406          && (pool_type != CTL_POOL_IOCTL)
3407          && (pool_type != CTL_POOL_4OTHERSC))
3408                 MOD_INC_USE_COUNT;
3409 #endif
3410
3411         mtx_unlock(&ctl_softc->pool_lock);
3412
3413         *npool = pool;
3414
3415 bailout:
3416
3417         return (retval);
3418 }
3419
3420 static int
3421 ctl_pool_acquire(struct ctl_io_pool *pool)
3422 {
3423
3424         mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3425
3426         if (pool->flags & CTL_POOL_FLAG_INVALID)
3427                 return (EINVAL);
3428
3429         pool->refcount++;
3430
3431         return (0);
3432 }
3433
3434 static void
3435 ctl_pool_release(struct ctl_io_pool *pool)
3436 {
3437         struct ctl_softc *ctl_softc = pool->ctl_softc;
3438         union ctl_io *io;
3439
3440         mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3441
3442         if (--pool->refcount != 0)
3443                 return;
3444
3445         while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3446                 STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3447                               links);
3448                 free(io, M_CTLIO);
3449         }
3450
3451         STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3452         ctl_softc->num_pools--;
3453
3454         /*
3455          * XXX KDM will this decrement the caller's usage count or mine?
3456          */
3457 #if 0
3458         if ((pool->type != CTL_POOL_EMERGENCY)
3459          && (pool->type != CTL_POOL_INTERNAL)
3460          && (pool->type != CTL_POOL_IOCTL))
3461                 MOD_DEC_USE_COUNT;
3462 #endif
3463
3464         free(pool, M_CTL);
3465 }
3466
3467 void
3468 ctl_pool_free(struct ctl_io_pool *pool)
3469 {
3470         struct ctl_softc *ctl_softc;
3471
3472         if (pool == NULL)
3473                 return;
3474
3475         ctl_softc = pool->ctl_softc;
3476         mtx_lock(&ctl_softc->pool_lock);
3477         pool->flags |= CTL_POOL_FLAG_INVALID;
3478         ctl_pool_release(pool);
3479         mtx_unlock(&ctl_softc->pool_lock);
3480 }
3481
3482 /*
3483  * This routine does not block (except for spinlocks of course).
3484  * It tries to allocate a ctl_io union from the caller's pool as quickly as
3485  * possible.
3486  */
3487 union ctl_io *
3488 ctl_alloc_io(void *pool_ref)
3489 {
3490         union ctl_io *io;
3491         struct ctl_softc *ctl_softc;
3492         struct ctl_io_pool *pool, *npool;
3493         struct ctl_io_pool *emergency_pool;
3494
3495         pool = (struct ctl_io_pool *)pool_ref;
3496
3497         if (pool == NULL) {
3498                 printf("%s: pool is NULL\n", __func__);
3499                 return (NULL);
3500         }
3501
3502         emergency_pool = NULL;
3503
3504         ctl_softc = pool->ctl_softc;
3505
3506         mtx_lock(&ctl_softc->pool_lock);
3507         /*
3508          * First, try to get the io structure from the user's pool.
3509          */
3510         if (ctl_pool_acquire(pool) == 0) {
3511                 io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3512                 if (io != NULL) {
3513                         STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3514                         pool->total_allocated++;
3515                         pool->free_ctl_io--;
3516                         mtx_unlock(&ctl_softc->pool_lock);
3517                         return (io);
3518                 } else
3519                         ctl_pool_release(pool);
3520         }
3521         /*
3522          * If he doesn't have any io structures left, search for an
3523          * emergency pool and grab one from there.
3524          */
3525         STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3526                 if (npool->type != CTL_POOL_EMERGENCY)
3527                         continue;
3528
3529                 if (ctl_pool_acquire(npool) != 0)
3530                         continue;
3531
3532                 emergency_pool = npool;
3533
3534                 io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3535                 if (io != NULL) {
3536                         STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3537                         npool->total_allocated++;
3538                         npool->free_ctl_io--;
3539                         mtx_unlock(&ctl_softc->pool_lock);
3540                         return (io);
3541                 } else
3542                         ctl_pool_release(npool);
3543         }
3544
3545         /* Drop the spinlock before we malloc */
3546         mtx_unlock(&ctl_softc->pool_lock);
3547
3548         /*
3549          * The emergency pool (if it exists) didn't have one, so try an
3550          * atomic (i.e. nonblocking) malloc and see if we get lucky.
3551          */
3552         io = (union ctl_io *)malloc(sizeof(*io), M_CTLIO, M_NOWAIT);
3553         if (io != NULL) {
3554                 /*
3555                  * If the emergency pool exists but is empty, add this
3556                  * ctl_io to its list when it gets freed.
3557                  */
3558                 if (emergency_pool != NULL) {
3559                         mtx_lock(&ctl_softc->pool_lock);
3560                         if (ctl_pool_acquire(emergency_pool) == 0) {
3561                                 io->io_hdr.pool = emergency_pool;
3562                                 emergency_pool->total_ctl_io++;
3563                                 /*
3564                                  * Need to bump this, otherwise
3565                                  * total_allocated and total_freed won't
3566                                  * match when we no longer have anything
3567                                  * outstanding.
3568                                  */
3569                                 emergency_pool->total_allocated++;
3570                         }
3571                         mtx_unlock(&ctl_softc->pool_lock);
3572                 } else
3573                         io->io_hdr.pool = NULL;
3574         }
3575
3576         return (io);
3577 }
3578
3579 void
3580 ctl_free_io(union ctl_io *io)
3581 {
3582         if (io == NULL)
3583                 return;
3584
3585         /*
3586          * If this ctl_io has a pool, return it to that pool.
3587          */
3588         if (io->io_hdr.pool != NULL) {
3589                 struct ctl_io_pool *pool;
3590
3591                 pool = (struct ctl_io_pool *)io->io_hdr.pool;
3592                 mtx_lock(&pool->ctl_softc->pool_lock);
3593                 io->io_hdr.io_type = 0xff;
3594                 STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3595                 pool->total_freed++;
3596                 pool->free_ctl_io++;
3597                 ctl_pool_release(pool);
3598                 mtx_unlock(&pool->ctl_softc->pool_lock);
3599         } else {
3600                 /*
3601                  * Otherwise, just free it.  We probably malloced it and
3602                  * the emergency pool wasn't available.
3603                  */
3604                 free(io, M_CTLIO);
3605         }
3606
3607 }
3608
3609 void
3610 ctl_zero_io(union ctl_io *io)
3611 {
3612         void *pool_ref;
3613
3614         if (io == NULL)
3615                 return;
3616
3617         /*
3618          * May need to preserve linked list pointers at some point too.
3619          */
3620         pool_ref = io->io_hdr.pool;
3621
3622         memset(io, 0, sizeof(*io));
3623
3624         io->io_hdr.pool = pool_ref;
3625 }
3626
3627 /*
3628  * This routine is currently used for internal copies of ctl_ios that need
3629  * to persist for some reason after we've already returned status to the
3630  * FETD.  (Thus the flag set.)
3631  *
3632  * XXX XXX
3633  * Note that this makes a blind copy of all fields in the ctl_io, except
3634  * for the pool reference.  This includes any memory that has been
3635  * allocated!  That memory will no longer be valid after done has been
3636  * called, so this would be VERY DANGEROUS for command that actually does
3637  * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3638  * start and stop commands, which don't transfer any data, so this is not a
3639  * problem.  If it is used for anything else, the caller would also need to
3640  * allocate data buffer space and this routine would need to be modified to
3641  * copy the data buffer(s) as well.
3642  */
3643 void
3644 ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3645 {
3646         void *pool_ref;
3647
3648         if ((src == NULL)
3649          || (dest == NULL))
3650                 return;
3651
3652         /*
3653          * May need to preserve linked list pointers at some point too.
3654          */
3655         pool_ref = dest->io_hdr.pool;
3656
3657         memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3658
3659         dest->io_hdr.pool = pool_ref;
3660         /*
3661          * We need to know that this is an internal copy, and doesn't need
3662          * to get passed back to the FETD that allocated it.
3663          */
3664         dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3665 }
3666
3667 #ifdef NEEDTOPORT
3668 static void
3669 ctl_update_power_subpage(struct copan_power_subpage *page)
3670 {
3671         int num_luns, num_partitions, config_type;
3672         struct ctl_softc *softc;
3673         cs_BOOL_t aor_present, shelf_50pct_power;
3674         cs_raidset_personality_t rs_type;
3675         int max_active_luns;
3676
3677         softc = control_softc;
3678
3679         /* subtract out the processor LUN */
3680         num_luns = softc->num_luns - 1;
3681         /*
3682          * Default to 7 LUNs active, which was the only number we allowed
3683          * in the past.
3684          */
3685         max_active_luns = 7;
3686
3687         num_partitions = config_GetRsPartitionInfo();
3688         config_type = config_GetConfigType();
3689         shelf_50pct_power = config_GetShelfPowerMode();
3690         aor_present = config_IsAorRsPresent();
3691
3692         rs_type = ddb_GetRsRaidType(1);
3693         if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3694          && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3695                 EPRINT(0, "Unsupported RS type %d!", rs_type);
3696         }
3697
3698
3699         page->total_luns = num_luns;
3700
3701         switch (config_type) {
3702         case 40:
3703                 /*
3704                  * In a 40 drive configuration, it doesn't matter what DC
3705                  * cards we have, whether we have AOR enabled or not,
3706                  * partitioning or not, or what type of RAIDset we have.
3707                  * In that scenario, we can power up every LUN we present
3708                  * to the user.
3709                  */
3710                 max_active_luns = num_luns;
3711
3712                 break;
3713         case 64:
3714                 if (shelf_50pct_power == CS_FALSE) {
3715                         /* 25% power */
3716                         if (aor_present == CS_TRUE) {
3717                                 if (rs_type ==
3718                                      CS_RAIDSET_PERSONALITY_RAID5) {
3719                                         max_active_luns = 7;
3720                                 } else if (rs_type ==
3721                                          CS_RAIDSET_PERSONALITY_RAID1){
3722                                         max_active_luns = 14;
3723                                 } else {
3724                                         /* XXX KDM now what?? */
3725                                 }
3726                         } else {
3727                                 if (rs_type ==
3728                                      CS_RAIDSET_PERSONALITY_RAID5) {
3729                                         max_active_luns = 8;
3730                                 } else if (rs_type ==
3731                                          CS_RAIDSET_PERSONALITY_RAID1){
3732                                         max_active_luns = 16;
3733                                 } else {
3734                                         /* XXX KDM now what?? */
3735                                 }
3736                         }
3737                 } else {
3738                         /* 50% power */
3739                         /*
3740                          * With 50% power in a 64 drive configuration, we
3741                          * can power all LUNs we present.
3742                          */
3743                         max_active_luns = num_luns;
3744                 }
3745                 break;
3746         case 112:
3747                 if (shelf_50pct_power == CS_FALSE) {
3748                         /* 25% power */
3749                         if (aor_present == CS_TRUE) {
3750                                 if (rs_type ==
3751                                      CS_RAIDSET_PERSONALITY_RAID5) {
3752                                         max_active_luns = 7;
3753                                 } else if (rs_type ==
3754                                          CS_RAIDSET_PERSONALITY_RAID1){
3755                                         max_active_luns = 14;
3756                                 } else {
3757                                         /* XXX KDM now what?? */
3758                                 }
3759                         } else {
3760                                 if (rs_type ==
3761                                      CS_RAIDSET_PERSONALITY_RAID5) {
3762                                         max_active_luns = 8;
3763                                 } else if (rs_type ==
3764                                          CS_RAIDSET_PERSONALITY_RAID1){
3765                                         max_active_luns = 16;
3766                                 } else {
3767                                         /* XXX KDM now what?? */
3768                                 }
3769                         }
3770                 } else {
3771                         /* 50% power */
3772                         if (aor_present == CS_TRUE) {
3773                                 if (rs_type ==
3774                                      CS_RAIDSET_PERSONALITY_RAID5) {
3775                                         max_active_luns = 14;
3776                                 } else if (rs_type ==
3777                                          CS_RAIDSET_PERSONALITY_RAID1){
3778                                         /*
3779                                          * We're assuming here that disk
3780                                          * caching is enabled, and so we're
3781                                          * able to power up half of each
3782                                          * LUN, and cache all writes.
3783                                          */
3784                                         max_active_luns = num_luns;
3785                                 } else {
3786                                         /* XXX KDM now what?? */
3787                                 }
3788                         } else {
3789                                 if (rs_type ==
3790                                      CS_RAIDSET_PERSONALITY_RAID5) {
3791                                         max_active_luns = 15;
3792                                 } else if (rs_type ==
3793                                          CS_RAIDSET_PERSONALITY_RAID1){
3794                                         max_active_luns = 30;
3795                                 } else {
3796                                         /* XXX KDM now what?? */
3797                                 }
3798                         }
3799                 }
3800                 break;
3801         default:
3802                 /*
3803                  * In this case, we have an unknown configuration, so we
3804                  * just use the default from above.
3805                  */
3806                 break;
3807         }
3808
3809         page->max_active_luns = max_active_luns;
3810 #if 0
3811         printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
3812                page->total_luns, page->max_active_luns);
3813 #endif
3814 }
3815 #endif /* NEEDTOPORT */
3816
3817 /*
3818  * This routine could be used in the future to load default and/or saved
3819  * mode page parameters for a particuar lun.
3820  */
3821 static int
3822 ctl_init_page_index(struct ctl_lun *lun)
3823 {
3824         int i;
3825         struct ctl_page_index *page_index;
3826         struct ctl_softc *softc;
3827
3828         memcpy(&lun->mode_pages.index, page_index_template,
3829                sizeof(page_index_template));
3830
3831         softc = lun->ctl_softc;
3832
3833         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3834
3835                 page_index = &lun->mode_pages.index[i];
3836                 /*
3837                  * If this is a disk-only mode page, there's no point in
3838                  * setting it up.  For some pages, we have to have some
3839                  * basic information about the disk in order to calculate the
3840                  * mode page data.
3841                  */
3842                 if ((lun->be_lun->lun_type != T_DIRECT)
3843                  && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3844                         continue;
3845
3846                 switch (page_index->page_code & SMPH_PC_MASK) {
3847                 case SMS_FORMAT_DEVICE_PAGE: {
3848                         struct scsi_format_page *format_page;
3849
3850                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3851                                 panic("subpage is incorrect!");
3852
3853                         /*
3854                          * Sectors per track are set above.  Bytes per
3855                          * sector need to be set here on a per-LUN basis.
3856                          */
3857                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3858                                &format_page_default,
3859                                sizeof(format_page_default));
3860                         memcpy(&lun->mode_pages.format_page[
3861                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
3862                                sizeof(format_page_changeable));
3863                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3864                                &format_page_default,
3865                                sizeof(format_page_default));
3866                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3867                                &format_page_default,
3868                                sizeof(format_page_default));
3869
3870                         format_page = &lun->mode_pages.format_page[
3871                                 CTL_PAGE_CURRENT];
3872                         scsi_ulto2b(lun->be_lun->blocksize,
3873                                     format_page->bytes_per_sector);
3874
3875                         format_page = &lun->mode_pages.format_page[
3876                                 CTL_PAGE_DEFAULT];
3877                         scsi_ulto2b(lun->be_lun->blocksize,
3878                                     format_page->bytes_per_sector);
3879
3880                         format_page = &lun->mode_pages.format_page[
3881                                 CTL_PAGE_SAVED];
3882                         scsi_ulto2b(lun->be_lun->blocksize,
3883                                     format_page->bytes_per_sector);
3884
3885                         page_index->page_data =
3886                                 (uint8_t *)lun->mode_pages.format_page;
3887                         break;
3888                 }
3889                 case SMS_RIGID_DISK_PAGE: {
3890                         struct scsi_rigid_disk_page *rigid_disk_page;
3891                         uint32_t sectors_per_cylinder;
3892                         uint64_t cylinders;
3893 #ifndef __XSCALE__
3894                         int shift;
3895 #endif /* !__XSCALE__ */
3896
3897                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3898                                 panic("invalid subpage value %d",
3899                                       page_index->subpage);
3900
3901                         /*
3902                          * Rotation rate and sectors per track are set
3903                          * above.  We calculate the cylinders here based on
3904                          * capacity.  Due to the number of heads and
3905                          * sectors per track we're using, smaller arrays
3906                          * may turn out to have 0 cylinders.  Linux and
3907                          * FreeBSD don't pay attention to these mode pages
3908                          * to figure out capacity, but Solaris does.  It
3909                          * seems to deal with 0 cylinders just fine, and
3910                          * works out a fake geometry based on the capacity.
3911                          */
3912                         memcpy(&lun->mode_pages.rigid_disk_page[
3913                                CTL_PAGE_CURRENT], &rigid_disk_page_default,
3914                                sizeof(rigid_disk_page_default));
3915                         memcpy(&lun->mode_pages.rigid_disk_page[
3916                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
3917                                sizeof(rigid_disk_page_changeable));
3918                         memcpy(&lun->mode_pages.rigid_disk_page[
3919                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
3920                                sizeof(rigid_disk_page_default));
3921                         memcpy(&lun->mode_pages.rigid_disk_page[
3922                                CTL_PAGE_SAVED], &rigid_disk_page_default,
3923                                sizeof(rigid_disk_page_default));
3924
3925                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
3926                                 CTL_DEFAULT_HEADS;
3927
3928                         /*
3929                          * The divide method here will be more accurate,
3930                          * probably, but results in floating point being
3931                          * used in the kernel on i386 (__udivdi3()).  On the
3932                          * XScale, though, __udivdi3() is implemented in
3933                          * software.
3934                          *
3935                          * The shift method for cylinder calculation is
3936                          * accurate if sectors_per_cylinder is a power of
3937                          * 2.  Otherwise it might be slightly off -- you
3938                          * might have a bit of a truncation problem.
3939                          */
3940 #ifdef  __XSCALE__
3941                         cylinders = (lun->be_lun->maxlba + 1) /
3942                                 sectors_per_cylinder;
3943 #else
3944                         for (shift = 31; shift > 0; shift--) {
3945                                 if (sectors_per_cylinder & (1 << shift))
3946                                         break;
3947                         }
3948                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
3949 #endif
3950
3951                         /*
3952                          * We've basically got 3 bytes, or 24 bits for the
3953                          * cylinder size in the mode page.  If we're over,
3954                          * just round down to 2^24.
3955                          */
3956                         if (cylinders > 0xffffff)
3957                                 cylinders = 0xffffff;
3958
3959                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3960                                 CTL_PAGE_CURRENT];
3961                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3962
3963                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3964                                 CTL_PAGE_DEFAULT];
3965                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3966
3967                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3968                                 CTL_PAGE_SAVED];
3969                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3970
3971                         page_index->page_data =
3972                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
3973                         break;
3974                 }
3975                 case SMS_CACHING_PAGE: {
3976
3977                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3978                                 panic("invalid subpage value %d",
3979                                       page_index->subpage);
3980                         /*
3981                          * Defaults should be okay here, no calculations
3982                          * needed.
3983                          */
3984                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
3985                                &caching_page_default,
3986                                sizeof(caching_page_default));
3987                         memcpy(&lun->mode_pages.caching_page[
3988                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
3989                                sizeof(caching_page_changeable));
3990                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
3991                                &caching_page_default,
3992                                sizeof(caching_page_default));
3993                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
3994                                &caching_page_default,
3995                                sizeof(caching_page_default));
3996                         page_index->page_data =
3997                                 (uint8_t *)lun->mode_pages.caching_page;
3998                         break;
3999                 }
4000                 case SMS_CONTROL_MODE_PAGE: {
4001
4002                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4003                                 panic("invalid subpage value %d",
4004                                       page_index->subpage);
4005
4006                         /*
4007                          * Defaults should be okay here, no calculations
4008                          * needed.
4009                          */
4010                         memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4011                                &control_page_default,
4012                                sizeof(control_page_default));
4013                         memcpy(&lun->mode_pages.control_page[
4014                                CTL_PAGE_CHANGEABLE], &control_page_changeable,
4015                                sizeof(control_page_changeable));
4016                         memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4017                                &control_page_default,
4018                                sizeof(control_page_default));
4019                         memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4020                                &control_page_default,
4021                                sizeof(control_page_default));
4022                         page_index->page_data =
4023                                 (uint8_t *)lun->mode_pages.control_page;
4024                         break;
4025
4026                 }
4027                 case SMS_VENDOR_SPECIFIC_PAGE:{
4028                         switch (page_index->subpage) {
4029                         case PWR_SUBPAGE_CODE: {
4030                                 struct copan_power_subpage *current_page,
4031                                                            *saved_page;
4032
4033                                 memcpy(&lun->mode_pages.power_subpage[
4034                                        CTL_PAGE_CURRENT],
4035                                        &power_page_default,
4036                                        sizeof(power_page_default));
4037                                 memcpy(&lun->mode_pages.power_subpage[
4038                                        CTL_PAGE_CHANGEABLE],
4039                                        &power_page_changeable,
4040                                        sizeof(power_page_changeable));
4041                                 memcpy(&lun->mode_pages.power_subpage[
4042                                        CTL_PAGE_DEFAULT],
4043                                        &power_page_default,
4044                                        sizeof(power_page_default));
4045                                 memcpy(&lun->mode_pages.power_subpage[
4046                                        CTL_PAGE_SAVED],
4047                                        &power_page_default,
4048                                        sizeof(power_page_default));
4049                                 page_index->page_data =
4050                                     (uint8_t *)lun->mode_pages.power_subpage;
4051
4052                                 current_page = (struct copan_power_subpage *)
4053                                         (page_index->page_data +
4054                                          (page_index->page_len *
4055                                           CTL_PAGE_CURRENT));
4056                                 saved_page = (struct copan_power_subpage *)
4057                                         (page_index->page_data +
4058                                          (page_index->page_len *
4059                                           CTL_PAGE_SAVED));
4060                                 break;
4061                         }
4062                         case APS_SUBPAGE_CODE: {
4063                                 struct copan_aps_subpage *current_page,
4064                                                          *saved_page;
4065
4066                                 // This gets set multiple times but
4067                                 // it should always be the same. It's
4068                                 // only done during init so who cares.
4069                                 index_to_aps_page = i;
4070
4071                                 memcpy(&lun->mode_pages.aps_subpage[
4072                                        CTL_PAGE_CURRENT],
4073                                        &aps_page_default,
4074                                        sizeof(aps_page_default));
4075                                 memcpy(&lun->mode_pages.aps_subpage[
4076                                        CTL_PAGE_CHANGEABLE],
4077                                        &aps_page_changeable,
4078                                        sizeof(aps_page_changeable));
4079                                 memcpy(&lun->mode_pages.aps_subpage[
4080                                        CTL_PAGE_DEFAULT],
4081                                        &aps_page_default,
4082                                        sizeof(aps_page_default));
4083                                 memcpy(&lun->mode_pages.aps_subpage[
4084                                        CTL_PAGE_SAVED],
4085                                        &aps_page_default,
4086                                        sizeof(aps_page_default));
4087                                 page_index->page_data =
4088                                         (uint8_t *)lun->mode_pages.aps_subpage;
4089
4090                                 current_page = (struct copan_aps_subpage *)
4091                                         (page_index->page_data +
4092                                          (page_index->page_len *
4093                                           CTL_PAGE_CURRENT));
4094                                 saved_page = (struct copan_aps_subpage *)
4095                                         (page_index->page_data +
4096                                          (page_index->page_len *
4097                                           CTL_PAGE_SAVED));
4098                                 break;
4099                         }
4100                         case DBGCNF_SUBPAGE_CODE: {
4101                                 struct copan_debugconf_subpage *current_page,
4102                                                                *saved_page;
4103
4104                                 memcpy(&lun->mode_pages.debugconf_subpage[
4105                                        CTL_PAGE_CURRENT],
4106                                        &debugconf_page_default,
4107                                        sizeof(debugconf_page_default));
4108                                 memcpy(&lun->mode_pages.debugconf_subpage[
4109                                        CTL_PAGE_CHANGEABLE],
4110                                        &debugconf_page_changeable,
4111                                        sizeof(debugconf_page_changeable));
4112                                 memcpy(&lun->mode_pages.debugconf_subpage[
4113                                        CTL_PAGE_DEFAULT],
4114                                        &debugconf_page_default,
4115                                        sizeof(debugconf_page_default));
4116                                 memcpy(&lun->mode_pages.debugconf_subpage[
4117                                        CTL_PAGE_SAVED],
4118                                        &debugconf_page_default,
4119                                        sizeof(debugconf_page_default));
4120                                 page_index->page_data =
4121                                         (uint8_t *)lun->mode_pages.debugconf_subpage;
4122
4123                                 current_page = (struct copan_debugconf_subpage *)
4124                                         (page_index->page_data +
4125                                          (page_index->page_len *
4126                                           CTL_PAGE_CURRENT));
4127                                 saved_page = (struct copan_debugconf_subpage *)
4128                                         (page_index->page_data +
4129                                          (page_index->page_len *
4130                                           CTL_PAGE_SAVED));
4131                                 break;
4132                         }
4133                         default:
4134                                 panic("invalid subpage value %d",
4135                                       page_index->subpage);
4136                                 break;
4137                         }
4138                         break;
4139                 }
4140                 default:
4141                         panic("invalid page value %d",
4142                               page_index->page_code & SMPH_PC_MASK);
4143                         break;
4144         }
4145         }
4146
4147         return (CTL_RETVAL_COMPLETE);
4148 }
4149
4150 /*
4151  * LUN allocation.
4152  *
4153  * Requirements:
4154  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4155  *   wants us to allocate the LUN and he can block.
4156  * - ctl_softc is always set
4157  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4158  *
4159  * Returns 0 for success, non-zero (errno) for failure.
4160  */
4161 static int
4162 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4163               struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4164 {
4165         struct ctl_lun *nlun, *lun;
4166         struct ctl_frontend *fe;
4167         int lun_number, i, lun_malloced;
4168
4169         if (be_lun == NULL)
4170                 return (EINVAL);
4171
4172         /*
4173          * We currently only support Direct Access or Processor LUN types.
4174          */
4175         switch (be_lun->lun_type) {
4176         case T_DIRECT:
4177                 break;
4178         case T_PROCESSOR:
4179                 break;
4180         case T_SEQUENTIAL:
4181         case T_CHANGER:
4182         default:
4183                 be_lun->lun_config_status(be_lun->be_lun,
4184                                           CTL_LUN_CONFIG_FAILURE);
4185                 break;
4186         }
4187         if (ctl_lun == NULL) {
4188                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4189                 lun_malloced = 1;
4190         } else {
4191                 lun_malloced = 0;
4192                 lun = ctl_lun;
4193         }
4194
4195         memset(lun, 0, sizeof(*lun));
4196         if (lun_malloced)
4197                 lun->flags = CTL_LUN_MALLOCED;
4198
4199         mtx_lock(&ctl_softc->ctl_lock);
4200         /*
4201          * See if the caller requested a particular LUN number.  If so, see
4202          * if it is available.  Otherwise, allocate the first available LUN.
4203          */
4204         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4205                 if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4206                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4207                         mtx_unlock(&ctl_softc->ctl_lock);
4208                         if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4209                                 printf("ctl: requested LUN ID %d is higher "
4210                                        "than CTL_MAX_LUNS - 1 (%d)\n",
4211                                        be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4212                         } else {
4213                                 /*
4214                                  * XXX KDM return an error, or just assign
4215                                  * another LUN ID in this case??
4216                                  */
4217                                 printf("ctl: requested LUN ID %d is already "
4218                                        "in use\n", be_lun->req_lun_id);
4219                         }
4220                         if (lun->flags & CTL_LUN_MALLOCED)
4221                                 free(lun, M_CTL);
4222                         be_lun->lun_config_status(be_lun->be_lun,
4223                                                   CTL_LUN_CONFIG_FAILURE);
4224                         return (ENOSPC);
4225                 }
4226                 lun_number = be_lun->req_lun_id;
4227         } else {
4228                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4229                 if (lun_number == -1) {
4230                         mtx_unlock(&ctl_softc->ctl_lock);
4231                         printf("ctl: can't allocate LUN on target %ju, out of "
4232                                "LUNs\n", (uintmax_t)target_id.id);
4233                         if (lun->flags & CTL_LUN_MALLOCED)
4234                                 free(lun, M_CTL);
4235                         be_lun->lun_config_status(be_lun->be_lun,
4236                                                   CTL_LUN_CONFIG_FAILURE);
4237                         return (ENOSPC);
4238                 }
4239         }
4240         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4241
4242         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4243         lun->target = target_id;
4244         lun->lun = lun_number;
4245         lun->be_lun = be_lun;
4246         /*
4247          * The processor LUN is always enabled.  Disk LUNs come on line
4248          * disabled, and must be enabled by the backend.
4249          */
4250         lun->flags |= CTL_LUN_DISABLED;
4251         lun->backend = be_lun->be;
4252         be_lun->ctl_lun = lun;
4253         be_lun->lun_id = lun_number;
4254         atomic_add_int(&be_lun->be->num_luns, 1);
4255         if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4256                 lun->flags |= CTL_LUN_STOPPED;
4257
4258         if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4259                 lun->flags |= CTL_LUN_INOPERABLE;
4260
4261         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4262                 lun->flags |= CTL_LUN_PRIMARY_SC;
4263
4264         lun->ctl_softc = ctl_softc;
4265         TAILQ_INIT(&lun->ooa_queue);
4266         TAILQ_INIT(&lun->blocked_queue);
4267         STAILQ_INIT(&lun->error_list);
4268
4269         /*
4270          * Initialize the mode page index.
4271          */
4272         ctl_init_page_index(lun);
4273
4274         /*
4275          * Set the poweron UA for all initiators on this LUN only.
4276          */
4277         for (i = 0; i < CTL_MAX_INITIATORS; i++)
4278                 lun->pending_sense[i].ua_pending = CTL_UA_POWERON;
4279
4280         /*
4281          * Now, before we insert this lun on the lun list, set the lun
4282          * inventory changed UA for all other luns.
4283          */
4284         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4285                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4286                         nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4287                 }
4288         }
4289
4290         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4291
4292         ctl_softc->ctl_luns[lun_number] = lun;
4293
4294         ctl_softc->num_luns++;
4295
4296         /* Setup statistics gathering */
4297         lun->stats.device_type = be_lun->lun_type;
4298         lun->stats.lun_number = lun_number;
4299         if (lun->stats.device_type == T_DIRECT)
4300                 lun->stats.blocksize = be_lun->blocksize;
4301         else
4302                 lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4303         for (i = 0;i < CTL_MAX_PORTS;i++)
4304                 lun->stats.ports[i].targ_port = i;
4305
4306         mtx_unlock(&ctl_softc->ctl_lock);
4307
4308         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4309
4310         /*
4311          * Run through each registered FETD and bring it online if it isn't
4312          * already.  Enable the target ID if it hasn't been enabled, and
4313          * enable this particular LUN.
4314          */
4315         STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4316                 int retval;
4317
4318                 /*
4319                  * XXX KDM this only works for ONE TARGET ID.  We'll need
4320                  * to do things differently if we go to a multiple target
4321                  * ID scheme.
4322                  */
4323                 if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) == 0) {
4324
4325                         retval = fe->targ_enable(fe->targ_lun_arg, target_id);
4326                         if (retval != 0) {
4327                                 printf("ctl_alloc_lun: FETD %s port %d "
4328                                        "returned error %d for targ_enable on "
4329                                        "target %ju\n", fe->port_name,
4330                                        fe->targ_port, retval,
4331                                        (uintmax_t)target_id.id);
4332                         } else
4333                                 fe->status |= CTL_PORT_STATUS_TARG_ONLINE;
4334                 }
4335
4336                 retval = fe->lun_enable(fe->targ_lun_arg, target_id,lun_number);
4337                 if (retval != 0) {
4338                         printf("ctl_alloc_lun: FETD %s port %d returned error "
4339                                "%d for lun_enable on target %ju lun %d\n",
4340                                fe->port_name, fe->targ_port, retval,
4341                                (uintmax_t)target_id.id, lun_number);
4342                 } else
4343                         fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4344         }
4345         return (0);
4346 }
4347
4348 /*
4349  * Delete a LUN.
4350  * Assumptions:
4351  * - LUN has already been marked invalid and any pending I/O has been taken
4352  *   care of.
4353  */
4354 static int
4355 ctl_free_lun(struct ctl_lun *lun)
4356 {
4357         struct ctl_softc *softc;
4358 #if 0
4359         struct ctl_frontend *fe;
4360 #endif
4361         struct ctl_lun *nlun;
4362         int i;
4363
4364         softc = lun->ctl_softc;
4365
4366         mtx_assert(&softc->ctl_lock, MA_OWNED);
4367
4368         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4369
4370         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4371
4372         softc->ctl_luns[lun->lun] = NULL;
4373
4374         if (!TAILQ_EMPTY(&lun->ooa_queue))
4375                 panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4376
4377         softc->num_luns--;
4378
4379         /*
4380          * XXX KDM this scheme only works for a single target/multiple LUN
4381          * setup.  It needs to be revamped for a multiple target scheme.
4382          *
4383          * XXX KDM this results in fe->lun_disable() getting called twice,
4384          * once when ctl_disable_lun() is called, and a second time here.
4385          * We really need to re-think the LUN disable semantics.  There
4386          * should probably be several steps/levels to LUN removal:
4387          *  - disable
4388          *  - invalidate
4389          *  - free
4390          *
4391          * Right now we only have a disable method when communicating to
4392          * the front end ports, at least for individual LUNs.
4393          */
4394 #if 0
4395         STAILQ_FOREACH(fe, &softc->fe_list, links) {
4396                 int retval;
4397
4398                 retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4399                                          lun->lun);
4400                 if (retval != 0) {
4401                         printf("ctl_free_lun: FETD %s port %d returned error "
4402                                "%d for lun_disable on target %ju lun %jd\n",
4403                                fe->port_name, fe->targ_port, retval,
4404                                (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4405                 }
4406
4407                 if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4408                         fe->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4409
4410                         retval = fe->targ_disable(fe->targ_lun_arg,lun->target);
4411                         if (retval != 0) {
4412                                 printf("ctl_free_lun: FETD %s port %d "
4413                                        "returned error %d for targ_disable on "
4414                                        "target %ju\n", fe->port_name,
4415                                        fe->targ_port, retval,
4416                                        (uintmax_t)lun->target.id);
4417                         } else
4418                                 fe->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4419
4420                         if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4421                                 continue;
4422
4423 #if 0
4424                         fe->port_offline(fe->onoff_arg);
4425                         fe->status &= ~CTL_PORT_STATUS_ONLINE;
4426 #endif
4427                 }
4428         }
4429 #endif
4430
4431         /*
4432          * Tell the backend to free resources, if this LUN has a backend.
4433          */
4434         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4435         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4436
4437         mtx_destroy(&lun->lun_lock);
4438         if (lun->flags & CTL_LUN_MALLOCED)
4439                 free(lun, M_CTL);
4440
4441         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4442                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4443                         nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4444                 }
4445         }
4446
4447         return (0);
4448 }
4449
4450 static void
4451 ctl_create_lun(struct ctl_be_lun *be_lun)
4452 {
4453         struct ctl_softc *ctl_softc;
4454
4455         ctl_softc = control_softc;
4456
4457         /*
4458          * ctl_alloc_lun() should handle all potential failure cases.
4459          */
4460         ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4461 }
4462
4463 int
4464 ctl_add_lun(struct ctl_be_lun *be_lun)
4465 {
4466         struct ctl_softc *ctl_softc = control_softc;
4467
4468         mtx_lock(&ctl_softc->ctl_lock);
4469         STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4470         mtx_unlock(&ctl_softc->ctl_lock);
4471         wakeup(&ctl_softc->pending_lun_queue);
4472
4473         return (0);
4474 }
4475
4476 int
4477 ctl_enable_lun(struct ctl_be_lun *be_lun)
4478 {
4479         struct ctl_softc *ctl_softc;
4480         struct ctl_frontend *fe, *nfe;
4481         struct ctl_lun *lun;
4482         int retval;
4483
4484         ctl_softc = control_softc;
4485
4486         lun = (struct ctl_lun *)be_lun->ctl_lun;
4487
4488         mtx_lock(&ctl_softc->ctl_lock);
4489         mtx_lock(&lun->lun_lock);
4490         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4491                 /*
4492                  * eh?  Why did we get called if the LUN is already
4493                  * enabled?
4494                  */
4495                 mtx_unlock(&lun->lun_lock);
4496                 mtx_unlock(&ctl_softc->ctl_lock);
4497                 return (0);
4498         }
4499         lun->flags &= ~CTL_LUN_DISABLED;
4500         mtx_unlock(&lun->lun_lock);
4501
4502         for (fe = STAILQ_FIRST(&ctl_softc->fe_list); fe != NULL; fe = nfe) {
4503                 nfe = STAILQ_NEXT(fe, links);
4504
4505                 /*
4506                  * Drop the lock while we call the FETD's enable routine.
4507                  * This can lead to a callback into CTL (at least in the
4508                  * case of the internal initiator frontend.
4509                  */
4510                 mtx_unlock(&ctl_softc->ctl_lock);
4511                 retval = fe->lun_enable(fe->targ_lun_arg, lun->target,lun->lun);
4512                 mtx_lock(&ctl_softc->ctl_lock);
4513                 if (retval != 0) {
4514                         printf("%s: FETD %s port %d returned error "
4515                                "%d for lun_enable on target %ju lun %jd\n",
4516                                __func__, fe->port_name, fe->targ_port, retval,
4517                                (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4518                 }
4519 #if 0
4520                  else {
4521             /* NOTE:  TODO:  why does lun enable affect port status? */
4522                         fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4523                 }
4524 #endif
4525         }
4526
4527         mtx_unlock(&ctl_softc->ctl_lock);
4528
4529         return (0);
4530 }
4531
4532 int
4533 ctl_disable_lun(struct ctl_be_lun *be_lun)
4534 {
4535         struct ctl_softc *ctl_softc;
4536         struct ctl_frontend *fe;
4537         struct ctl_lun *lun;
4538         int retval;
4539
4540         ctl_softc = control_softc;
4541
4542         lun = (struct ctl_lun *)be_lun->ctl_lun;
4543
4544         mtx_lock(&ctl_softc->ctl_lock);
4545         mtx_lock(&lun->lun_lock);
4546         if (lun->flags & CTL_LUN_DISABLED) {
4547                 mtx_unlock(&lun->lun_lock);
4548                 mtx_unlock(&ctl_softc->ctl_lock);
4549                 return (0);
4550         }
4551         lun->flags |= CTL_LUN_DISABLED;
4552         mtx_unlock(&lun->lun_lock);
4553
4554         STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4555                 mtx_unlock(&ctl_softc->ctl_lock);
4556                 /*
4557                  * Drop the lock before we call the frontend's disable
4558                  * routine, to avoid lock order reversals.
4559                  *
4560                  * XXX KDM what happens if the frontend list changes while
4561                  * we're traversing it?  It's unlikely, but should be handled.
4562                  */
4563                 retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4564                                          lun->lun);
4565                 mtx_lock(&ctl_softc->ctl_lock);
4566                 if (retval != 0) {
4567                         printf("ctl_alloc_lun: FETD %s port %d returned error "
4568                                "%d for lun_disable on target %ju lun %jd\n",
4569                                fe->port_name, fe->targ_port, retval,
4570                                (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4571                 }
4572         }
4573
4574         mtx_unlock(&ctl_softc->ctl_lock);
4575
4576         return (0);
4577 }
4578
4579 int
4580 ctl_start_lun(struct ctl_be_lun *be_lun)
4581 {
4582         struct ctl_softc *ctl_softc;
4583         struct ctl_lun *lun;
4584
4585         ctl_softc = control_softc;
4586
4587         lun = (struct ctl_lun *)be_lun->ctl_lun;
4588
4589         mtx_lock(&lun->lun_lock);
4590         lun->flags &= ~CTL_LUN_STOPPED;
4591         mtx_unlock(&lun->lun_lock);
4592
4593         return (0);
4594 }
4595
4596 int
4597 ctl_stop_lun(struct ctl_be_lun *be_lun)
4598 {
4599         struct ctl_softc *ctl_softc;
4600         struct ctl_lun *lun;
4601
4602         ctl_softc = control_softc;
4603
4604         lun = (struct ctl_lun *)be_lun->ctl_lun;
4605
4606         mtx_lock(&lun->lun_lock);
4607         lun->flags |= CTL_LUN_STOPPED;
4608         mtx_unlock(&lun->lun_lock);
4609
4610         return (0);
4611 }
4612
4613 int
4614 ctl_lun_offline(struct ctl_be_lun *be_lun)
4615 {
4616         struct ctl_softc *ctl_softc;
4617         struct ctl_lun *lun;
4618
4619         ctl_softc = control_softc;
4620
4621         lun = (struct ctl_lun *)be_lun->ctl_lun;
4622
4623         mtx_lock(&lun->lun_lock);
4624         lun->flags |= CTL_LUN_OFFLINE;
4625         mtx_unlock(&lun->lun_lock);
4626
4627         return (0);
4628 }
4629
4630 int
4631 ctl_lun_online(struct ctl_be_lun *be_lun)
4632 {
4633         struct ctl_softc *ctl_softc;
4634         struct ctl_lun *lun;
4635
4636         ctl_softc = control_softc;
4637
4638         lun = (struct ctl_lun *)be_lun->ctl_lun;
4639
4640         mtx_lock(&lun->lun_lock);
4641         lun->flags &= ~CTL_LUN_OFFLINE;
4642         mtx_unlock(&lun->lun_lock);
4643
4644         return (0);
4645 }
4646
4647 int
4648 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4649 {
4650         struct ctl_softc *ctl_softc;
4651         struct ctl_lun *lun;
4652
4653         ctl_softc = control_softc;
4654
4655         lun = (struct ctl_lun *)be_lun->ctl_lun;
4656
4657         mtx_lock(&lun->lun_lock);
4658
4659         /*
4660          * The LUN needs to be disabled before it can be marked invalid.
4661          */
4662         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4663                 mtx_unlock(&lun->lun_lock);
4664                 return (-1);
4665         }
4666         /*
4667          * Mark the LUN invalid.
4668          */
4669         lun->flags |= CTL_LUN_INVALID;
4670
4671         /*
4672          * If there is nothing in the OOA queue, go ahead and free the LUN.
4673          * If we have something in the OOA queue, we'll free it when the
4674          * last I/O completes.
4675          */
4676         if (TAILQ_EMPTY(&lun->ooa_queue)) {
4677                 mtx_unlock(&lun->lun_lock);
4678                 mtx_lock(&ctl_softc->ctl_lock);
4679                 ctl_free_lun(lun);
4680                 mtx_unlock(&ctl_softc->ctl_lock);
4681         } else
4682                 mtx_unlock(&lun->lun_lock);
4683
4684         return (0);
4685 }
4686
4687 int
4688 ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4689 {
4690         struct ctl_softc *ctl_softc;
4691         struct ctl_lun *lun;
4692
4693         ctl_softc = control_softc;
4694         lun = (struct ctl_lun *)be_lun->ctl_lun;
4695
4696         mtx_lock(&lun->lun_lock);
4697         lun->flags |= CTL_LUN_INOPERABLE;
4698         mtx_unlock(&lun->lun_lock);
4699
4700         return (0);
4701 }
4702
4703 int
4704 ctl_lun_operable(struct ctl_be_lun *be_lun)
4705 {
4706         struct ctl_softc *ctl_softc;
4707         struct ctl_lun *lun;
4708
4709         ctl_softc = control_softc;
4710         lun = (struct ctl_lun *)be_lun->ctl_lun;
4711
4712         mtx_lock(&lun->lun_lock);
4713         lun->flags &= ~CTL_LUN_INOPERABLE;
4714         mtx_unlock(&lun->lun_lock);
4715
4716         return (0);
4717 }
4718
4719 int
4720 ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
4721                    int lock)
4722 {
4723         struct ctl_softc *softc;
4724         struct ctl_lun *lun;
4725         struct copan_aps_subpage *current_sp;
4726         struct ctl_page_index *page_index;
4727         int i;
4728
4729         softc = control_softc;
4730
4731         mtx_lock(&softc->ctl_lock);
4732
4733         lun = (struct ctl_lun *)be_lun->ctl_lun;
4734         mtx_lock(&lun->lun_lock);
4735
4736         page_index = NULL;
4737         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4738                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
4739                      APS_PAGE_CODE)
4740                         continue;
4741
4742                 if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
4743                         continue;
4744                 page_index = &lun->mode_pages.index[i];
4745         }
4746
4747         if (page_index == NULL) {
4748                 mtx_unlock(&lun->lun_lock);
4749                 mtx_unlock(&softc->ctl_lock);
4750                 printf("%s: APS subpage not found for lun %ju!\n", __func__,
4751                        (uintmax_t)lun->lun);
4752                 return (1);
4753         }
4754 #if 0
4755         if ((softc->aps_locked_lun != 0)
4756          && (softc->aps_locked_lun != lun->lun)) {
4757                 printf("%s: attempt to lock LUN %llu when %llu is already "
4758                        "locked\n");
4759                 mtx_unlock(&lun->lun_lock);
4760                 mtx_unlock(&softc->ctl_lock);
4761                 return (1);
4762         }
4763 #endif
4764
4765         current_sp = (struct copan_aps_subpage *)(page_index->page_data +
4766                 (page_index->page_len * CTL_PAGE_CURRENT));
4767
4768         if (lock != 0) {
4769                 current_sp->lock_active = APS_LOCK_ACTIVE;
4770                 softc->aps_locked_lun = lun->lun;
4771         } else {
4772                 current_sp->lock_active = 0;
4773                 softc->aps_locked_lun = 0;
4774         }
4775
4776
4777         /*
4778          * If we're in HA mode, try to send the lock message to the other
4779          * side.
4780          */
4781         if (ctl_is_single == 0) {
4782                 int isc_retval;
4783                 union ctl_ha_msg lock_msg;
4784
4785                 lock_msg.hdr.nexus = *nexus;
4786                 lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
4787                 if (lock != 0)
4788                         lock_msg.aps.lock_flag = 1;
4789                 else
4790                         lock_msg.aps.lock_flag = 0;
4791                 isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
4792                                          sizeof(lock_msg), 0);
4793                 if (isc_retval > CTL_HA_STATUS_SUCCESS) {
4794                         printf("%s: APS (lock=%d) error returned from "
4795                                "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
4796                         mtx_unlock(&lun->lun_lock);
4797                         mtx_unlock(&softc->ctl_lock);
4798                         return (1);
4799                 }
4800         }
4801
4802         mtx_unlock(&lun->lun_lock);
4803         mtx_unlock(&softc->ctl_lock);
4804
4805         return (0);
4806 }
4807
4808 void
4809 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4810 {
4811         struct ctl_lun *lun;
4812         struct ctl_softc *softc;
4813         int i;
4814
4815         softc = control_softc;
4816
4817         lun = (struct ctl_lun *)be_lun->ctl_lun;
4818
4819         mtx_lock(&lun->lun_lock);
4820
4821         for (i = 0; i < CTL_MAX_INITIATORS; i++)
4822                 lun->pending_sense[i].ua_pending |= CTL_UA_CAPACITY_CHANGED;
4823
4824         mtx_unlock(&lun->lun_lock);
4825 }
4826
4827 /*
4828  * Backend "memory move is complete" callback for requests that never
4829  * make it down to say RAIDCore's configuration code.
4830  */
4831 int
4832 ctl_config_move_done(union ctl_io *io)
4833 {
4834         int retval;
4835
4836         retval = CTL_RETVAL_COMPLETE;
4837
4838
4839         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4840         /*
4841          * XXX KDM this shouldn't happen, but what if it does?
4842          */
4843         if (io->io_hdr.io_type != CTL_IO_SCSI)
4844                 panic("I/O type isn't CTL_IO_SCSI!");
4845
4846         if ((io->io_hdr.port_status == 0)
4847          && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4848          && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
4849                 io->io_hdr.status = CTL_SUCCESS;
4850         else if ((io->io_hdr.port_status != 0)
4851               && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4852               && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
4853                 /*
4854                  * For hardware error sense keys, the sense key
4855                  * specific value is defined to be a retry count,
4856                  * but we use it to pass back an internal FETD
4857                  * error code.  XXX KDM  Hopefully the FETD is only
4858                  * using 16 bits for an error code, since that's
4859                  * all the space we have in the sks field.
4860                  */
4861                 ctl_set_internal_failure(&io->scsiio,
4862                                          /*sks_valid*/ 1,
4863                                          /*retry_count*/
4864                                          io->io_hdr.port_status);
4865                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4866                         free(io->scsiio.kern_data_ptr, M_CTL);
4867                 ctl_done(io);
4868                 goto bailout;
4869         }
4870
4871         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
4872          || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
4873          || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4874                 /*
4875                  * XXX KDM just assuming a single pointer here, and not a
4876                  * S/G list.  If we start using S/G lists for config data,
4877                  * we'll need to know how to clean them up here as well.
4878                  */
4879                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4880                         free(io->scsiio.kern_data_ptr, M_CTL);
4881                 /* Hopefully the user has already set the status... */
4882                 ctl_done(io);
4883         } else {
4884                 /*
4885                  * XXX KDM now we need to continue data movement.  Some
4886                  * options:
4887                  * - call ctl_scsiio() again?  We don't do this for data
4888                  *   writes, because for those at least we know ahead of
4889                  *   time where the write will go and how long it is.  For
4890                  *   config writes, though, that information is largely
4891                  *   contained within the write itself, thus we need to
4892                  *   parse out the data again.
4893                  *
4894                  * - Call some other function once the data is in?
4895                  */
4896
4897                 /*
4898                  * XXX KDM call ctl_scsiio() again for now, and check flag
4899                  * bits to see whether we're allocated or not.
4900                  */
4901                 retval = ctl_scsiio(&io->scsiio);
4902         }
4903 bailout:
4904         return (retval);
4905 }
4906
4907 /*
4908  * This gets called by a backend driver when it is done with a
4909  * data_submit method.
4910  */
4911 void
4912 ctl_data_submit_done(union ctl_io *io)
4913 {
4914         /*
4915          * If the IO_CONT flag is set, we need to call the supplied
4916          * function to continue processing the I/O, instead of completing
4917          * the I/O just yet.
4918          *
4919          * If there is an error, though, we don't want to keep processing.
4920          * Instead, just send status back to the initiator.
4921          */
4922         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
4923             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
4924             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4925              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4926                 io->scsiio.io_cont(io);
4927                 return;
4928         }
4929         ctl_done(io);
4930 }
4931
4932 /*
4933  * This gets called by a backend driver when it is done with a
4934  * configuration write.
4935  */
4936 void
4937 ctl_config_write_done(union ctl_io *io)
4938 {
4939         /*
4940          * If the IO_CONT flag is set, we need to call the supplied
4941          * function to continue processing the I/O, instead of completing
4942          * the I/O just yet.
4943          *
4944          * If there is an error, though, we don't want to keep processing.
4945          * Instead, just send status back to the initiator.
4946          */
4947         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
4948          && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
4949           || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
4950                 io->scsiio.io_cont(io);
4951                 return;
4952         }
4953         /*
4954          * Since a configuration write can be done for commands that actually
4955          * have data allocated, like write buffer, and commands that have
4956          * no data, like start/stop unit, we need to check here.
4957          */
4958         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
4959                 free(io->scsiio.kern_data_ptr, M_CTL);
4960         ctl_done(io);
4961 }
4962
4963 /*
4964  * SCSI release command.
4965  */
4966 int
4967 ctl_scsi_release(struct ctl_scsiio *ctsio)
4968 {
4969         int length, longid, thirdparty_id, resv_id;
4970         struct ctl_softc *ctl_softc;
4971         struct ctl_lun *lun;
4972
4973         length = 0;
4974         resv_id = 0;
4975
4976         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
4977
4978         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
4979         ctl_softc = control_softc;
4980
4981         switch (ctsio->cdb[0]) {
4982         case RELEASE: {
4983                 struct scsi_release *cdb;
4984
4985                 cdb = (struct scsi_release *)ctsio->cdb;
4986                 if ((cdb->byte2 & 0x1f) != 0) {
4987                         ctl_set_invalid_field(ctsio,
4988                                               /*sks_valid*/ 1,
4989                                               /*command*/ 1,
4990                                               /*field*/ 1,
4991                                               /*bit_valid*/ 0,
4992                                               /*bit*/ 0);
4993                         ctl_done((union ctl_io *)ctsio);
4994                         return (CTL_RETVAL_COMPLETE);
4995                 }
4996                 break;
4997         }
4998         case RELEASE_10: {
4999                 struct scsi_release_10 *cdb;
5000
5001                 cdb = (struct scsi_release_10 *)ctsio->cdb;
5002
5003                 if ((cdb->byte2 & SR10_EXTENT) != 0) {
5004                         ctl_set_invalid_field(ctsio,
5005                                               /*sks_valid*/ 1,
5006                                               /*command*/ 1,
5007                                               /*field*/ 1,
5008                                               /*bit_valid*/ 1,
5009                                               /*bit*/ 0);
5010                         ctl_done((union ctl_io *)ctsio);
5011                         return (CTL_RETVAL_COMPLETE);
5012
5013                 }
5014
5015                 if ((cdb->byte2 & SR10_3RDPTY) != 0) {
5016                         ctl_set_invalid_field(ctsio,
5017                                               /*sks_valid*/ 1,
5018                                               /*command*/ 1,
5019                                               /*field*/ 1,
5020                                               /*bit_valid*/ 1,
5021                                               /*bit*/ 4);
5022                         ctl_done((union ctl_io *)ctsio);
5023                         return (CTL_RETVAL_COMPLETE);
5024                 }
5025
5026                 if (cdb->byte2 & SR10_LONGID)
5027                         longid = 1;
5028                 else
5029                         thirdparty_id = cdb->thirdparty_id;
5030
5031                 resv_id = cdb->resv_id;
5032                 length = scsi_2btoul(cdb->length);
5033                 break;
5034         }
5035         }
5036
5037
5038         /*
5039          * XXX KDM right now, we only support LUN reservation.  We don't
5040          * support 3rd party reservations, or extent reservations, which
5041          * might actually need the parameter list.  If we've gotten this
5042          * far, we've got a LUN reservation.  Anything else got kicked out
5043          * above.  So, according to SPC, ignore the length.
5044          */
5045         length = 0;
5046
5047         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5048          && (length > 0)) {
5049                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5050                 ctsio->kern_data_len = length;
5051                 ctsio->kern_total_len = length;
5052                 ctsio->kern_data_resid = 0;
5053                 ctsio->kern_rel_offset = 0;
5054                 ctsio->kern_sg_entries = 0;
5055                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5056                 ctsio->be_move_done = ctl_config_move_done;
5057                 ctl_datamove((union ctl_io *)ctsio);
5058
5059                 return (CTL_RETVAL_COMPLETE);
5060         }
5061
5062         if (length > 0)
5063                 thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5064
5065         mtx_lock(&lun->lun_lock);
5066
5067         /*
5068          * According to SPC, it is not an error for an intiator to attempt
5069          * to release a reservation on a LUN that isn't reserved, or that
5070          * is reserved by another initiator.  The reservation can only be
5071          * released, though, by the initiator who made it or by one of
5072          * several reset type events.
5073          */
5074         if (lun->flags & CTL_LUN_RESERVED) {
5075                 if ((ctsio->io_hdr.nexus.initid.id == lun->rsv_nexus.initid.id)
5076                  && (ctsio->io_hdr.nexus.targ_port == lun->rsv_nexus.targ_port)
5077                  && (ctsio->io_hdr.nexus.targ_target.id ==
5078                      lun->rsv_nexus.targ_target.id)) {
5079                         lun->flags &= ~CTL_LUN_RESERVED;
5080                 }
5081         }
5082
5083         mtx_unlock(&lun->lun_lock);
5084
5085         ctsio->scsi_status = SCSI_STATUS_OK;
5086         ctsio->io_hdr.status = CTL_SUCCESS;
5087
5088         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5089                 free(ctsio->kern_data_ptr, M_CTL);
5090                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5091         }
5092
5093         ctl_done((union ctl_io *)ctsio);
5094         return (CTL_RETVAL_COMPLETE);
5095 }
5096
5097 int
5098 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5099 {
5100         int extent, thirdparty, longid;
5101         int resv_id, length;
5102         uint64_t thirdparty_id;
5103         struct ctl_softc *ctl_softc;
5104         struct ctl_lun *lun;
5105
5106         extent = 0;
5107         thirdparty = 0;
5108         longid = 0;
5109         resv_id = 0;
5110         length = 0;
5111         thirdparty_id = 0;
5112
5113         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5114
5115         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5116         ctl_softc = control_softc;
5117
5118         switch (ctsio->cdb[0]) {
5119         case RESERVE: {
5120                 struct scsi_reserve *cdb;
5121
5122                 cdb = (struct scsi_reserve *)ctsio->cdb;
5123                 if ((cdb->byte2 & 0x1f) != 0) {
5124                         ctl_set_invalid_field(ctsio,
5125                                               /*sks_valid*/ 1,
5126                                               /*command*/ 1,
5127                                               /*field*/ 1,
5128                                               /*bit_valid*/ 0,
5129                                               /*bit*/ 0);
5130                         ctl_done((union ctl_io *)ctsio);
5131                         return (CTL_RETVAL_COMPLETE);
5132                 }
5133                 resv_id = cdb->resv_id;
5134                 length = scsi_2btoul(cdb->length);
5135                 break;
5136         }
5137         case RESERVE_10: {
5138                 struct scsi_reserve_10 *cdb;
5139
5140                 cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5141
5142                 if ((cdb->byte2 & SR10_EXTENT) != 0) {
5143                         ctl_set_invalid_field(ctsio,
5144                                               /*sks_valid*/ 1,
5145                                               /*command*/ 1,
5146                                               /*field*/ 1,
5147                                               /*bit_valid*/ 1,
5148                                               /*bit*/ 0);
5149                         ctl_done((union ctl_io *)ctsio);
5150                         return (CTL_RETVAL_COMPLETE);
5151                 }
5152                 if ((cdb->byte2 & SR10_3RDPTY) != 0) {
5153                         ctl_set_invalid_field(ctsio,
5154                                               /*sks_valid*/ 1,
5155                                               /*command*/ 1,
5156                                               /*field*/ 1,
5157                                               /*bit_valid*/ 1,
5158                                               /*bit*/ 4);
5159                         ctl_done((union ctl_io *)ctsio);
5160                         return (CTL_RETVAL_COMPLETE);
5161                 }
5162                 if (cdb->byte2 & SR10_LONGID)
5163                         longid = 1;
5164                 else
5165                         thirdparty_id = cdb->thirdparty_id;
5166
5167                 resv_id = cdb->resv_id;
5168                 length = scsi_2btoul(cdb->length);
5169                 break;
5170         }
5171         }
5172
5173         /*
5174          * XXX KDM right now, we only support LUN reservation.  We don't
5175          * support 3rd party reservations, or extent reservations, which
5176          * might actually need the parameter list.  If we've gotten this
5177          * far, we've got a LUN reservation.  Anything else got kicked out
5178          * above.  So, according to SPC, ignore the length.
5179          */
5180         length = 0;
5181
5182         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5183          && (length > 0)) {
5184                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5185                 ctsio->kern_data_len = length;
5186                 ctsio->kern_total_len = length;
5187                 ctsio->kern_data_resid = 0;
5188                 ctsio->kern_rel_offset = 0;
5189                 ctsio->kern_sg_entries = 0;
5190                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5191                 ctsio->be_move_done = ctl_config_move_done;
5192                 ctl_datamove((union ctl_io *)ctsio);
5193
5194                 return (CTL_RETVAL_COMPLETE);
5195         }
5196
5197         if (length > 0)
5198                 thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5199
5200         mtx_lock(&lun->lun_lock);
5201         if (lun->flags & CTL_LUN_RESERVED) {
5202                 if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
5203                  || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
5204                  || (ctsio->io_hdr.nexus.targ_target.id !=
5205                      lun->rsv_nexus.targ_target.id)) {
5206                         ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5207                         ctsio->io_hdr.status = CTL_SCSI_ERROR;
5208                         goto bailout;
5209                 }
5210         }
5211
5212         lun->flags |= CTL_LUN_RESERVED;
5213         lun->rsv_nexus = ctsio->io_hdr.nexus;
5214
5215         ctsio->scsi_status = SCSI_STATUS_OK;
5216         ctsio->io_hdr.status = CTL_SUCCESS;
5217
5218 bailout:
5219         mtx_unlock(&lun->lun_lock);
5220
5221         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5222                 free(ctsio->kern_data_ptr, M_CTL);
5223                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5224         }
5225
5226         ctl_done((union ctl_io *)ctsio);
5227         return (CTL_RETVAL_COMPLETE);
5228 }
5229
5230 int
5231 ctl_start_stop(struct ctl_scsiio *ctsio)
5232 {
5233         struct scsi_start_stop_unit *cdb;
5234         struct ctl_lun *lun;
5235         struct ctl_softc *ctl_softc;
5236         int retval;
5237
5238         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5239
5240         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5241         ctl_softc = control_softc;
5242         retval = 0;
5243
5244         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5245
5246         /*
5247          * XXX KDM
5248          * We don't support the immediate bit on a stop unit.  In order to
5249          * do that, we would need to code up a way to know that a stop is
5250          * pending, and hold off any new commands until it completes, one
5251          * way or another.  Then we could accept or reject those commands
5252          * depending on its status.  We would almost need to do the reverse
5253          * of what we do below for an immediate start -- return the copy of
5254          * the ctl_io to the FETD with status to send to the host (and to
5255          * free the copy!) and then free the original I/O once the stop
5256          * actually completes.  That way, the OOA queue mechanism can work
5257          * to block commands that shouldn't proceed.  Another alternative
5258          * would be to put the copy in the queue in place of the original,
5259          * and return the original back to the caller.  That could be
5260          * slightly safer..
5261          */
5262         if ((cdb->byte2 & SSS_IMMED)
5263          && ((cdb->how & SSS_START) == 0)) {
5264                 ctl_set_invalid_field(ctsio,
5265                                       /*sks_valid*/ 1,
5266                                       /*command*/ 1,
5267                                       /*field*/ 1,
5268                                       /*bit_valid*/ 1,
5269                                       /*bit*/ 0);
5270                 ctl_done((union ctl_io *)ctsio);
5271                 return (CTL_RETVAL_COMPLETE);
5272         }
5273
5274         /*
5275          * We don't support the power conditions field.  We need to check
5276          * this prior to checking the load/eject and start/stop bits.
5277          */
5278         if ((cdb->how & SSS_PC_MASK) != SSS_PC_START_VALID) {
5279                 ctl_set_invalid_field(ctsio,
5280                                       /*sks_valid*/ 1,
5281                                       /*command*/ 1,
5282                                       /*field*/ 4,
5283                                       /*bit_valid*/ 1,
5284                                       /*bit*/ 4);
5285                 ctl_done((union ctl_io *)ctsio);
5286                 return (CTL_RETVAL_COMPLETE);
5287         }
5288
5289         /*
5290          * Media isn't removable, so we can't load or eject it.
5291          */
5292         if ((cdb->how & SSS_LOEJ) != 0) {
5293                 ctl_set_invalid_field(ctsio,
5294                                       /*sks_valid*/ 1,
5295                                       /*command*/ 1,
5296                                       /*field*/ 4,
5297                                       /*bit_valid*/ 1,
5298                                       /*bit*/ 1);
5299                 ctl_done((union ctl_io *)ctsio);
5300                 return (CTL_RETVAL_COMPLETE);
5301         }
5302
5303         if ((lun->flags & CTL_LUN_PR_RESERVED)
5304          && ((cdb->how & SSS_START)==0)) {
5305                 uint32_t residx;
5306
5307                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5308                 if (!lun->per_res[residx].registered
5309                  || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5310
5311                         ctl_set_reservation_conflict(ctsio);
5312                         ctl_done((union ctl_io *)ctsio);
5313                         return (CTL_RETVAL_COMPLETE);
5314                 }
5315         }
5316
5317         /*
5318          * If there is no backend on this device, we can't start or stop
5319          * it.  In theory we shouldn't get any start/stop commands in the
5320          * first place at this level if the LUN doesn't have a backend.
5321          * That should get stopped by the command decode code.
5322          */
5323         if (lun->backend == NULL) {
5324                 ctl_set_invalid_opcode(ctsio);
5325                 ctl_done((union ctl_io *)ctsio);
5326                 return (CTL_RETVAL_COMPLETE);
5327         }
5328
5329         /*
5330          * XXX KDM Copan-specific offline behavior.
5331          * Figure out a reasonable way to port this?
5332          */
5333 #ifdef NEEDTOPORT
5334         mtx_lock(&lun->lun_lock);
5335
5336         if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5337          && (lun->flags & CTL_LUN_OFFLINE)) {
5338                 /*
5339                  * If the LUN is offline, and the on/offline bit isn't set,
5340                  * reject the start or stop.  Otherwise, let it through.
5341                  */
5342                 mtx_unlock(&lun->lun_lock);
5343                 ctl_set_lun_not_ready(ctsio);
5344                 ctl_done((union ctl_io *)ctsio);
5345         } else {
5346                 mtx_unlock(&lun->lun_lock);
5347 #endif /* NEEDTOPORT */
5348                 /*
5349                  * This could be a start or a stop when we're online,
5350                  * or a stop/offline or start/online.  A start or stop when
5351                  * we're offline is covered in the case above.
5352                  */
5353                 /*
5354                  * In the non-immediate case, we send the request to
5355                  * the backend and return status to the user when
5356                  * it is done.
5357                  *
5358                  * In the immediate case, we allocate a new ctl_io
5359                  * to hold a copy of the request, and send that to
5360                  * the backend.  We then set good status on the
5361                  * user's request and return it immediately.
5362                  */
5363                 if (cdb->byte2 & SSS_IMMED) {
5364                         union ctl_io *new_io;
5365
5366                         new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5367                         if (new_io == NULL) {
5368                                 ctl_set_busy(ctsio);
5369                                 ctl_done((union ctl_io *)ctsio);
5370                         } else {
5371                                 ctl_copy_io((union ctl_io *)ctsio,
5372                                             new_io);
5373                                 retval = lun->backend->config_write(new_io);
5374                                 ctl_set_success(ctsio);
5375                                 ctl_done((union ctl_io *)ctsio);
5376                         }
5377                 } else {
5378                         retval = lun->backend->config_write(
5379                                 (union ctl_io *)ctsio);
5380                 }
5381 #ifdef NEEDTOPORT
5382         }
5383 #endif
5384         return (retval);
5385 }
5386
5387 /*
5388  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5389  * we don't really do anything with the LBA and length fields if the user
5390  * passes them in.  Instead we'll just flush out the cache for the entire
5391  * LUN.
5392  */
5393 int
5394 ctl_sync_cache(struct ctl_scsiio *ctsio)
5395 {
5396         struct ctl_lun *lun;
5397         struct ctl_softc *ctl_softc;
5398         uint64_t starting_lba;
5399         uint32_t block_count;
5400         int reladr, immed;
5401         int retval;
5402
5403         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5404
5405         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5406         ctl_softc = control_softc;
5407         retval = 0;
5408         reladr = 0;
5409         immed = 0;
5410
5411         switch (ctsio->cdb[0]) {
5412         case SYNCHRONIZE_CACHE: {
5413                 struct scsi_sync_cache *cdb;
5414                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5415
5416                 if (cdb->byte2 & SSC_RELADR)
5417                         reladr = 1;
5418
5419                 if (cdb->byte2 & SSC_IMMED)
5420                         immed = 1;
5421
5422                 starting_lba = scsi_4btoul(cdb->begin_lba);
5423                 block_count = scsi_2btoul(cdb->lb_count);
5424                 break;
5425         }
5426         case SYNCHRONIZE_CACHE_16: {
5427                 struct scsi_sync_cache_16 *cdb;
5428                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5429
5430                 if (cdb->byte2 & SSC_RELADR)
5431                         reladr = 1;
5432
5433                 if (cdb->byte2 & SSC_IMMED)
5434                         immed = 1;
5435
5436                 starting_lba = scsi_8btou64(cdb->begin_lba);
5437                 block_count = scsi_4btoul(cdb->lb_count);
5438                 break;
5439         }
5440         default:
5441                 ctl_set_invalid_opcode(ctsio);
5442                 ctl_done((union ctl_io *)ctsio);
5443                 goto bailout;
5444                 break; /* NOTREACHED */
5445         }
5446
5447         if (immed) {
5448                 /*
5449                  * We don't support the immediate bit.  Since it's in the
5450                  * same place for the 10 and 16 byte SYNCHRONIZE CACHE
5451                  * commands, we can just return the same error in either
5452                  * case.
5453                  */
5454                 ctl_set_invalid_field(ctsio,
5455                                       /*sks_valid*/ 1,
5456                                       /*command*/ 1,
5457                                       /*field*/ 1,
5458                                       /*bit_valid*/ 1,
5459                                       /*bit*/ 1);
5460                 ctl_done((union ctl_io *)ctsio);
5461                 goto bailout;
5462         }
5463
5464         if (reladr) {
5465                 /*
5466                  * We don't support the reladr bit either.  It can only be
5467                  * used with linked commands, and we don't support linked
5468                  * commands.  Since the bit is in the same place for the
5469                  * 10 and 16 byte SYNCHRONIZE CACHE * commands, we can
5470                  * just return the same error in either case.
5471                  */
5472                 ctl_set_invalid_field(ctsio,
5473                                       /*sks_valid*/ 1,
5474                                       /*command*/ 1,
5475                                       /*field*/ 1,
5476                                       /*bit_valid*/ 1,
5477                                       /*bit*/ 0);
5478                 ctl_done((union ctl_io *)ctsio);
5479                 goto bailout;
5480         }
5481
5482         /*
5483          * We check the LBA and length, but don't do anything with them.
5484          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5485          * get flushed.  This check will just help satisfy anyone who wants
5486          * to see an error for an out of range LBA.
5487          */
5488         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5489                 ctl_set_lba_out_of_range(ctsio);
5490                 ctl_done((union ctl_io *)ctsio);
5491                 goto bailout;
5492         }
5493
5494         /*
5495          * If this LUN has no backend, we can't flush the cache anyway.
5496          */
5497         if (lun->backend == NULL) {
5498                 ctl_set_invalid_opcode(ctsio);
5499                 ctl_done((union ctl_io *)ctsio);
5500                 goto bailout;
5501         }
5502
5503         /*
5504          * Check to see whether we're configured to send the SYNCHRONIZE
5505          * CACHE command directly to the back end.
5506          */
5507         mtx_lock(&lun->lun_lock);
5508         if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5509          && (++(lun->sync_count) >= lun->sync_interval)) {
5510                 lun->sync_count = 0;
5511                 mtx_unlock(&lun->lun_lock);
5512                 retval = lun->backend->config_write((union ctl_io *)ctsio);
5513         } else {
5514                 mtx_unlock(&lun->lun_lock);
5515                 ctl_set_success(ctsio);
5516                 ctl_done((union ctl_io *)ctsio);
5517         }
5518
5519 bailout:
5520
5521         return (retval);
5522 }
5523
5524 int
5525 ctl_format(struct ctl_scsiio *ctsio)
5526 {
5527         struct scsi_format *cdb;
5528         struct ctl_lun *lun;
5529         struct ctl_softc *ctl_softc;
5530         int length, defect_list_len;
5531
5532         CTL_DEBUG_PRINT(("ctl_format\n"));
5533
5534         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5535         ctl_softc = control_softc;
5536
5537         cdb = (struct scsi_format *)ctsio->cdb;
5538
5539         length = 0;
5540         if (cdb->byte2 & SF_FMTDATA) {
5541                 if (cdb->byte2 & SF_LONGLIST)
5542                         length = sizeof(struct scsi_format_header_long);
5543                 else
5544                         length = sizeof(struct scsi_format_header_short);
5545         }
5546
5547         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5548          && (length > 0)) {
5549                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5550                 ctsio->kern_data_len = length;
5551                 ctsio->kern_total_len = length;
5552                 ctsio->kern_data_resid = 0;
5553                 ctsio->kern_rel_offset = 0;
5554                 ctsio->kern_sg_entries = 0;
5555                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5556                 ctsio->be_move_done = ctl_config_move_done;
5557                 ctl_datamove((union ctl_io *)ctsio);
5558
5559                 return (CTL_RETVAL_COMPLETE);
5560         }
5561
5562         defect_list_len = 0;
5563
5564         if (cdb->byte2 & SF_FMTDATA) {
5565                 if (cdb->byte2 & SF_LONGLIST) {
5566                         struct scsi_format_header_long *header;
5567
5568                         header = (struct scsi_format_header_long *)
5569                                 ctsio->kern_data_ptr;
5570
5571                         defect_list_len = scsi_4btoul(header->defect_list_len);
5572                         if (defect_list_len != 0) {
5573                                 ctl_set_invalid_field(ctsio,
5574                                                       /*sks_valid*/ 1,
5575                                                       /*command*/ 0,
5576                                                       /*field*/ 2,
5577                                                       /*bit_valid*/ 0,
5578                                                       /*bit*/ 0);
5579                                 goto bailout;
5580                         }
5581                 } else {
5582                         struct scsi_format_header_short *header;
5583
5584                         header = (struct scsi_format_header_short *)
5585                                 ctsio->kern_data_ptr;
5586
5587                         defect_list_len = scsi_2btoul(header->defect_list_len);
5588                         if (defect_list_len != 0) {
5589                                 ctl_set_invalid_field(ctsio,
5590                                                       /*sks_valid*/ 1,
5591                                                       /*command*/ 0,
5592                                                       /*field*/ 2,
5593                                                       /*bit_valid*/ 0,
5594                                                       /*bit*/ 0);
5595                                 goto bailout;
5596                         }
5597                 }
5598         }
5599
5600         /*
5601          * The format command will clear out the "Medium format corrupted"
5602          * status if set by the configuration code.  That status is really
5603          * just a way to notify the host that we have lost the media, and
5604          * get them to issue a command that will basically make them think
5605          * they're blowing away the media.
5606          */
5607         mtx_lock(&lun->lun_lock);
5608         lun->flags &= ~CTL_LUN_INOPERABLE;
5609         mtx_unlock(&lun->lun_lock);
5610
5611         ctsio->scsi_status = SCSI_STATUS_OK;
5612         ctsio->io_hdr.status = CTL_SUCCESS;
5613 bailout:
5614
5615         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5616                 free(ctsio->kern_data_ptr, M_CTL);
5617                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5618         }
5619
5620         ctl_done((union ctl_io *)ctsio);
5621         return (CTL_RETVAL_COMPLETE);
5622 }
5623
5624 int
5625 ctl_read_buffer(struct ctl_scsiio *ctsio)
5626 {
5627         struct scsi_read_buffer *cdb;
5628         struct ctl_lun *lun;
5629         int buffer_offset, len;
5630         static uint8_t descr[4];
5631         static uint8_t echo_descr[4] = { 0 };
5632
5633         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5634
5635         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5636         cdb = (struct scsi_read_buffer *)ctsio->cdb;
5637
5638         if (lun->flags & CTL_LUN_PR_RESERVED) {
5639                 uint32_t residx;
5640
5641                 /*
5642                  * XXX KDM need a lock here.
5643                  */
5644                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5645                 if ((lun->res_type == SPR_TYPE_EX_AC
5646                   && residx != lun->pr_res_idx)
5647                  || ((lun->res_type == SPR_TYPE_EX_AC_RO
5648                    || lun->res_type == SPR_TYPE_EX_AC_AR)
5649                   && !lun->per_res[residx].registered)) {
5650                         ctl_set_reservation_conflict(ctsio);
5651                         ctl_done((union ctl_io *)ctsio);
5652                         return (CTL_RETVAL_COMPLETE);
5653                 }
5654         }
5655
5656         if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5657             (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5658             (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5659                 ctl_set_invalid_field(ctsio,
5660                                       /*sks_valid*/ 1,
5661                                       /*command*/ 1,
5662                                       /*field*/ 1,
5663                                       /*bit_valid*/ 1,
5664                                       /*bit*/ 4);
5665                 ctl_done((union ctl_io *)ctsio);
5666                 return (CTL_RETVAL_COMPLETE);
5667         }
5668         if (cdb->buffer_id != 0) {
5669                 ctl_set_invalid_field(ctsio,
5670                                       /*sks_valid*/ 1,
5671                                       /*command*/ 1,
5672                                       /*field*/ 2,
5673                                       /*bit_valid*/ 0,
5674                                       /*bit*/ 0);
5675                 ctl_done((union ctl_io *)ctsio);
5676                 return (CTL_RETVAL_COMPLETE);
5677         }
5678
5679         len = scsi_3btoul(cdb->length);
5680         buffer_offset = scsi_3btoul(cdb->offset);
5681
5682         if (buffer_offset + len > sizeof(lun->write_buffer)) {
5683                 ctl_set_invalid_field(ctsio,
5684                                       /*sks_valid*/ 1,
5685                                       /*command*/ 1,
5686                                       /*field*/ 6,
5687                                       /*bit_valid*/ 0,
5688                                       /*bit*/ 0);
5689                 ctl_done((union ctl_io *)ctsio);
5690                 return (CTL_RETVAL_COMPLETE);
5691         }
5692
5693         if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5694                 descr[0] = 0;
5695                 scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5696                 ctsio->kern_data_ptr = descr;
5697                 len = min(len, sizeof(descr));
5698         } else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5699                 ctsio->kern_data_ptr = echo_descr;
5700                 len = min(len, sizeof(echo_descr));
5701         } else
5702                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5703         ctsio->kern_data_len = len;
5704         ctsio->kern_total_len = len;
5705         ctsio->kern_data_resid = 0;
5706         ctsio->kern_rel_offset = 0;
5707         ctsio->kern_sg_entries = 0;
5708         ctsio->be_move_done = ctl_config_move_done;
5709         ctl_datamove((union ctl_io *)ctsio);
5710
5711         return (CTL_RETVAL_COMPLETE);
5712 }
5713
5714 int
5715 ctl_write_buffer(struct ctl_scsiio *ctsio)
5716 {
5717         struct scsi_write_buffer *cdb;
5718         struct ctl_lun *lun;
5719         int buffer_offset, len;
5720
5721         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5722
5723         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5724         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5725
5726         if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5727                 ctl_set_invalid_field(ctsio,
5728                                       /*sks_valid*/ 1,
5729                                       /*command*/ 1,
5730                                       /*field*/ 1,
5731                                       /*bit_valid*/ 1,
5732                                       /*bit*/ 4);
5733                 ctl_done((union ctl_io *)ctsio);
5734                 return (CTL_RETVAL_COMPLETE);
5735         }
5736         if (cdb->buffer_id != 0) {
5737                 ctl_set_invalid_field(ctsio,
5738                                       /*sks_valid*/ 1,
5739                                       /*command*/ 1,
5740                                       /*field*/ 2,
5741                                       /*bit_valid*/ 0,
5742                                       /*bit*/ 0);
5743                 ctl_done((union ctl_io *)ctsio);
5744                 return (CTL_RETVAL_COMPLETE);
5745         }
5746
5747         len = scsi_3btoul(cdb->length);
5748         buffer_offset = scsi_3btoul(cdb->offset);
5749
5750         if (buffer_offset + len > sizeof(lun->write_buffer)) {
5751                 ctl_set_invalid_field(ctsio,
5752                                       /*sks_valid*/ 1,
5753                                       /*command*/ 1,
5754                                       /*field*/ 6,
5755                                       /*bit_valid*/ 0,
5756                                       /*bit*/ 0);
5757                 ctl_done((union ctl_io *)ctsio);
5758                 return (CTL_RETVAL_COMPLETE);
5759         }
5760
5761         /*
5762          * If we've got a kernel request that hasn't been malloced yet,
5763          * malloc it and tell the caller the data buffer is here.
5764          */
5765         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5766                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5767                 ctsio->kern_data_len = len;
5768                 ctsio->kern_total_len = len;
5769                 ctsio->kern_data_resid = 0;
5770                 ctsio->kern_rel_offset = 0;
5771                 ctsio->kern_sg_entries = 0;
5772                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5773                 ctsio->be_move_done = ctl_config_move_done;
5774                 ctl_datamove((union ctl_io *)ctsio);
5775
5776                 return (CTL_RETVAL_COMPLETE);
5777         }
5778
5779         ctl_done((union ctl_io *)ctsio);
5780
5781         return (CTL_RETVAL_COMPLETE);
5782 }
5783
5784 int
5785 ctl_write_same(struct ctl_scsiio *ctsio)
5786 {
5787         struct ctl_lun *lun;
5788         struct ctl_lba_len_flags *lbalen;
5789         uint64_t lba;
5790         uint32_t num_blocks;
5791         int len, retval;
5792         uint8_t byte2;
5793
5794         retval = CTL_RETVAL_COMPLETE;
5795
5796         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5797
5798         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5799
5800         switch (ctsio->cdb[0]) {
5801         case WRITE_SAME_10: {
5802                 struct scsi_write_same_10 *cdb;
5803
5804                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5805
5806                 lba = scsi_4btoul(cdb->addr);
5807                 num_blocks = scsi_2btoul(cdb->length);
5808                 byte2 = cdb->byte2;
5809                 break;
5810         }
5811         case WRITE_SAME_16: {
5812                 struct scsi_write_same_16 *cdb;
5813
5814                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5815
5816                 lba = scsi_8btou64(cdb->addr);
5817                 num_blocks = scsi_4btoul(cdb->length);
5818                 byte2 = cdb->byte2;
5819                 break;
5820         }
5821         default:
5822                 /*
5823                  * We got a command we don't support.  This shouldn't
5824                  * happen, commands should be filtered out above us.
5825                  */
5826                 ctl_set_invalid_opcode(ctsio);
5827                 ctl_done((union ctl_io *)ctsio);
5828
5829                 return (CTL_RETVAL_COMPLETE);
5830                 break; /* NOTREACHED */
5831         }
5832
5833         /*
5834          * The first check is to make sure we're in bounds, the second
5835          * check is to catch wrap-around problems.  If the lba + num blocks
5836          * is less than the lba, then we've wrapped around and the block
5837          * range is invalid anyway.
5838          */
5839         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5840          || ((lba + num_blocks) < lba)) {
5841                 ctl_set_lba_out_of_range(ctsio);
5842                 ctl_done((union ctl_io *)ctsio);
5843                 return (CTL_RETVAL_COMPLETE);
5844         }
5845
5846         /* Zero number of blocks means "to the last logical block" */
5847         if (num_blocks == 0) {
5848                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5849                         ctl_set_invalid_field(ctsio,
5850                                               /*sks_valid*/ 0,
5851                                               /*command*/ 1,
5852                                               /*field*/ 0,
5853                                               /*bit_valid*/ 0,
5854                                               /*bit*/ 0);
5855                         ctl_done((union ctl_io *)ctsio);
5856                         return (CTL_RETVAL_COMPLETE);
5857                 }
5858                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
5859         }
5860
5861         len = lun->be_lun->blocksize;
5862
5863         /*
5864          * If we've got a kernel request that hasn't been malloced yet,
5865          * malloc it and tell the caller the data buffer is here.
5866          */
5867         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5868                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5869                 ctsio->kern_data_len = len;
5870                 ctsio->kern_total_len = len;
5871                 ctsio->kern_data_resid = 0;
5872                 ctsio->kern_rel_offset = 0;
5873                 ctsio->kern_sg_entries = 0;
5874                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5875                 ctsio->be_move_done = ctl_config_move_done;
5876                 ctl_datamove((union ctl_io *)ctsio);
5877
5878                 return (CTL_RETVAL_COMPLETE);
5879         }
5880
5881         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5882         lbalen->lba = lba;
5883         lbalen->len = num_blocks;
5884         lbalen->flags = byte2;
5885         retval = lun->backend->config_write((union ctl_io *)ctsio);
5886
5887         return (retval);
5888 }
5889
5890 int
5891 ctl_unmap(struct ctl_scsiio *ctsio)
5892 {
5893         struct ctl_lun *lun;
5894         struct scsi_unmap *cdb;
5895         struct ctl_ptr_len_flags *ptrlen;
5896         struct scsi_unmap_header *hdr;
5897         struct scsi_unmap_desc *buf, *end;
5898         uint64_t lba;
5899         uint32_t num_blocks;
5900         int len, retval;
5901         uint8_t byte2;
5902
5903         retval = CTL_RETVAL_COMPLETE;
5904
5905         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5906
5907         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5908         cdb = (struct scsi_unmap *)ctsio->cdb;
5909
5910         len = scsi_2btoul(cdb->length);
5911         byte2 = cdb->byte2;
5912
5913         /*
5914          * If we've got a kernel request that hasn't been malloced yet,
5915          * malloc it and tell the caller the data buffer is here.
5916          */
5917         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5918                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5919                 ctsio->kern_data_len = len;
5920                 ctsio->kern_total_len = len;
5921                 ctsio->kern_data_resid = 0;
5922                 ctsio->kern_rel_offset = 0;
5923                 ctsio->kern_sg_entries = 0;
5924                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5925                 ctsio->be_move_done = ctl_config_move_done;
5926                 ctl_datamove((union ctl_io *)ctsio);
5927
5928                 return (CTL_RETVAL_COMPLETE);
5929         }
5930
5931         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5932         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5933         if (len < sizeof (*hdr) ||
5934             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5935             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5936             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5937                 ctl_set_invalid_field(ctsio,
5938                                       /*sks_valid*/ 0,
5939                                       /*command*/ 0,
5940                                       /*field*/ 0,
5941                                       /*bit_valid*/ 0,
5942                                       /*bit*/ 0);
5943                 ctl_done((union ctl_io *)ctsio);
5944                 return (CTL_RETVAL_COMPLETE);
5945         }
5946         len = scsi_2btoul(hdr->desc_length);
5947         buf = (struct scsi_unmap_desc *)(hdr + 1);
5948         end = buf + len / sizeof(*buf);
5949
5950         ptrlen = (struct ctl_ptr_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5951         ptrlen->ptr = (void *)buf;
5952         ptrlen->len = len;
5953         ptrlen->flags = byte2;
5954
5955         for (; buf < end; buf++) {
5956                 lba = scsi_8btou64(buf->lba);
5957                 num_blocks = scsi_4btoul(buf->length);
5958                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5959                  || ((lba + num_blocks) < lba)) {
5960                         ctl_set_lba_out_of_range(ctsio);
5961                         ctl_done((union ctl_io *)ctsio);
5962                         return (CTL_RETVAL_COMPLETE);
5963                 }
5964         }
5965
5966         retval = lun->backend->config_write((union ctl_io *)ctsio);
5967
5968         return (retval);
5969 }
5970
5971 /*
5972  * Note that this function currently doesn't actually do anything inside
5973  * CTL to enforce things if the DQue bit is turned on.
5974  *
5975  * Also note that this function can't be used in the default case, because
5976  * the DQue bit isn't set in the changeable mask for the control mode page
5977  * anyway.  This is just here as an example for how to implement a page
5978  * handler, and a placeholder in case we want to allow the user to turn
5979  * tagged queueing on and off.
5980  *
5981  * The D_SENSE bit handling is functional, however, and will turn
5982  * descriptor sense on and off for a given LUN.
5983  */
5984 int
5985 ctl_control_page_handler(struct ctl_scsiio *ctsio,
5986                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5987 {
5988         struct scsi_control_page *current_cp, *saved_cp, *user_cp;
5989         struct ctl_lun *lun;
5990         struct ctl_softc *softc;
5991         int set_ua;
5992         uint32_t initidx;
5993
5994         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5995         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5996         set_ua = 0;
5997
5998         user_cp = (struct scsi_control_page *)page_ptr;
5999         current_cp = (struct scsi_control_page *)
6000                 (page_index->page_data + (page_index->page_len *
6001                 CTL_PAGE_CURRENT));
6002         saved_cp = (struct scsi_control_page *)
6003                 (page_index->page_data + (page_index->page_len *
6004                 CTL_PAGE_SAVED));
6005
6006         softc = control_softc;
6007
6008         mtx_lock(&lun->lun_lock);
6009         if (((current_cp->rlec & SCP_DSENSE) == 0)
6010          && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6011                 /*
6012                  * Descriptor sense is currently turned off and the user
6013                  * wants to turn it on.
6014                  */
6015                 current_cp->rlec |= SCP_DSENSE;
6016                 saved_cp->rlec |= SCP_DSENSE;
6017                 lun->flags |= CTL_LUN_SENSE_DESC;
6018                 set_ua = 1;
6019         } else if (((current_cp->rlec & SCP_DSENSE) != 0)
6020                 && ((user_cp->rlec & SCP_DSENSE) == 0)) {
6021                 /*
6022                  * Descriptor sense is currently turned on, and the user
6023                  * wants to turn it off.
6024                  */
6025                 current_cp->rlec &= ~SCP_DSENSE;
6026                 saved_cp->rlec &= ~SCP_DSENSE;
6027                 lun->flags &= ~CTL_LUN_SENSE_DESC;
6028                 set_ua = 1;
6029         }
6030         if (current_cp->queue_flags & SCP_QUEUE_DQUE) {
6031                 if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6032 #ifdef NEEDTOPORT
6033                         csevent_log(CSC_CTL | CSC_SHELF_SW |
6034                                     CTL_UNTAG_TO_UNTAG,
6035                                     csevent_LogType_Trace,
6036                                     csevent_Severity_Information,
6037                                     csevent_AlertLevel_Green,
6038                                     csevent_FRU_Firmware,
6039                                     csevent_FRU_Unknown,
6040                                     "Received untagged to untagged transition");
6041 #endif /* NEEDTOPORT */
6042                 } else {
6043 #ifdef NEEDTOPORT
6044                         csevent_log(CSC_CTL | CSC_SHELF_SW |
6045                                     CTL_UNTAG_TO_TAG,
6046                                     csevent_LogType_ConfigChange,
6047                                     csevent_Severity_Information,
6048                                     csevent_AlertLevel_Green,
6049                                     csevent_FRU_Firmware,
6050                                     csevent_FRU_Unknown,
6051                                     "Received untagged to tagged "
6052                                     "queueing transition");
6053 #endif /* NEEDTOPORT */
6054
6055                         current_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6056                         saved_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6057                         set_ua = 1;
6058                 }
6059         } else {
6060                 if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6061 #ifdef NEEDTOPORT
6062                         csevent_log(CSC_CTL | CSC_SHELF_SW |
6063                                     CTL_TAG_TO_UNTAG,
6064                                     csevent_LogType_ConfigChange,
6065                                     csevent_Severity_Warning,
6066                                     csevent_AlertLevel_Yellow,
6067                                     csevent_FRU_Firmware,
6068                                     csevent_FRU_Unknown,
6069                                     "Received tagged queueing to untagged "
6070                                     "transition");
6071 #endif /* NEEDTOPORT */
6072
6073                         current_cp->queue_flags |= SCP_QUEUE_DQUE;
6074                         saved_cp->queue_flags |= SCP_QUEUE_DQUE;
6075                         set_ua = 1;
6076                 } else {
6077 #ifdef NEEDTOPORT
6078                         csevent_log(CSC_CTL | CSC_SHELF_SW |
6079                                     CTL_TAG_TO_TAG,
6080                                     csevent_LogType_Trace,
6081                                     csevent_Severity_Information,
6082                                     csevent_AlertLevel_Green,
6083                                     csevent_FRU_Firmware,
6084                                     csevent_FRU_Unknown,
6085                                     "Received tagged queueing to tagged "
6086                                     "queueing transition");
6087 #endif /* NEEDTOPORT */
6088                 }
6089         }
6090         if (set_ua != 0) {
6091                 int i;
6092                 /*
6093                  * Let other initiators know that the mode
6094                  * parameters for this LUN have changed.
6095                  */
6096                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6097                         if (i == initidx)
6098                                 continue;
6099
6100                         lun->pending_sense[i].ua_pending |=
6101                                 CTL_UA_MODE_CHANGE;
6102                 }
6103         }
6104         mtx_unlock(&lun->lun_lock);
6105
6106         return (0);
6107 }
6108
6109 int
6110 ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6111                      struct ctl_page_index *page_index, uint8_t *page_ptr)
6112 {
6113         return (0);
6114 }
6115
6116 int
6117 ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6118                            struct ctl_page_index *page_index, int pc)
6119 {
6120         struct copan_power_subpage *page;
6121
6122         page = (struct copan_power_subpage *)page_index->page_data +
6123                 (page_index->page_len * pc);
6124
6125         switch (pc) {
6126         case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6127                 /*
6128                  * We don't update the changable bits for this page.
6129                  */
6130                 break;
6131         case SMS_PAGE_CTRL_CURRENT >> 6:
6132         case SMS_PAGE_CTRL_DEFAULT >> 6:
6133         case SMS_PAGE_CTRL_SAVED >> 6:
6134 #ifdef NEEDTOPORT
6135                 ctl_update_power_subpage(page);
6136 #endif
6137                 break;
6138         default:
6139 #ifdef NEEDTOPORT
6140                 EPRINT(0, "Invalid PC %d!!", pc);
6141 #endif
6142                 break;
6143         }
6144         return (0);
6145 }
6146
6147
6148 int
6149 ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6150                    struct ctl_page_index *page_index, uint8_t *page_ptr)
6151 {
6152         struct copan_aps_subpage *user_sp;
6153         struct copan_aps_subpage *current_sp;
6154         union ctl_modepage_info *modepage_info;
6155         struct ctl_softc *softc;
6156         struct ctl_lun *lun;
6157         int retval;
6158
6159         retval = CTL_RETVAL_COMPLETE;
6160         current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6161                      (page_index->page_len * CTL_PAGE_CURRENT));
6162         softc = control_softc;
6163         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6164
6165         user_sp = (struct copan_aps_subpage *)page_ptr;
6166
6167         modepage_info = (union ctl_modepage_info *)
6168                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6169
6170         modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6171         modepage_info->header.subpage = page_index->subpage;
6172         modepage_info->aps.lock_active = user_sp->lock_active;
6173
6174         mtx_lock(&softc->ctl_lock);
6175
6176         /*
6177          * If there is a request to lock the LUN and another LUN is locked
6178          * this is an error. If the requested LUN is already locked ignore
6179          * the request. If no LUN is locked attempt to lock it.
6180          * if there is a request to unlock the LUN and the LUN is currently
6181          * locked attempt to unlock it. Otherwise ignore the request. i.e.
6182          * if another LUN is locked or no LUN is locked.
6183          */
6184         if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6185                 if (softc->aps_locked_lun == lun->lun) {
6186                         /*
6187                          * This LUN is already locked, so we're done.
6188                          */
6189                         retval = CTL_RETVAL_COMPLETE;
6190                 } else if (softc->aps_locked_lun == 0) {
6191                         /*
6192                          * No one has the lock, pass the request to the
6193                          * backend.
6194                          */
6195                         retval = lun->backend->config_write(
6196                                 (union ctl_io *)ctsio);
6197                 } else {
6198                         /*
6199                          * Someone else has the lock, throw out the request.
6200                          */
6201                         ctl_set_already_locked(ctsio);
6202                         free(ctsio->kern_data_ptr, M_CTL);
6203                         ctl_done((union ctl_io *)ctsio);
6204
6205                         /*
6206                          * Set the return value so that ctl_do_mode_select()
6207                          * won't try to complete the command.  We already
6208                          * completed it here.
6209                          */
6210                         retval = CTL_RETVAL_ERROR;
6211                 }
6212         } else if (softc->aps_locked_lun == lun->lun) {
6213                 /*
6214                  * This LUN is locked, so pass the unlock request to the
6215                  * backend.
6216                  */
6217                 retval = lun->backend->config_write((union ctl_io *)ctsio);
6218         }
6219         mtx_unlock(&softc->ctl_lock);
6220
6221         return (retval);
6222 }
6223
6224 int
6225 ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6226                                 struct ctl_page_index *page_index,
6227                                 uint8_t *page_ptr)
6228 {
6229         uint8_t *c;
6230         int i;
6231
6232         c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6233         ctl_time_io_secs =
6234                 (c[0] << 8) |
6235                 (c[1] << 0) |
6236                 0;
6237         CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6238         printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6239         printf("page data:");
6240         for (i=0; i<8; i++)
6241                 printf(" %.2x",page_ptr[i]);
6242         printf("\n");
6243         return (0);
6244 }
6245
6246 int
6247 ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6248                                struct ctl_page_index *page_index,
6249                                int pc)
6250 {
6251         struct copan_debugconf_subpage *page;
6252
6253         page = (struct copan_debugconf_subpage *)page_index->page_data +
6254                 (page_index->page_len * pc);
6255
6256         switch (pc) {
6257         case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6258         case SMS_PAGE_CTRL_DEFAULT >> 6:
6259         case SMS_PAGE_CTRL_SAVED >> 6:
6260                 /*
6261                  * We don't update the changable or default bits for this page.
6262                  */
6263                 break;
6264         case SMS_PAGE_CTRL_CURRENT >> 6:
6265                 page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6266                 page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6267                 break;
6268         default:
6269 #ifdef NEEDTOPORT
6270                 EPRINT(0, "Invalid PC %d!!", pc);
6271 #endif /* NEEDTOPORT */
6272                 break;
6273         }
6274         return (0);
6275 }
6276
6277
6278 static int
6279 ctl_do_mode_select(union ctl_io *io)
6280 {
6281         struct scsi_mode_page_header *page_header;
6282         struct ctl_page_index *page_index;
6283         struct ctl_scsiio *ctsio;
6284         int control_dev, page_len;
6285         int page_len_offset, page_len_size;
6286         union ctl_modepage_info *modepage_info;
6287         struct ctl_lun *lun;
6288         int *len_left, *len_used;
6289         int retval, i;
6290
6291         ctsio = &io->scsiio;
6292         page_index = NULL;
6293         page_len = 0;
6294         retval = CTL_RETVAL_COMPLETE;
6295
6296         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6297
6298         if (lun->be_lun->lun_type != T_DIRECT)
6299                 control_dev = 1;
6300         else
6301                 control_dev = 0;
6302
6303         modepage_info = (union ctl_modepage_info *)
6304                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6305         len_left = &modepage_info->header.len_left;
6306         len_used = &modepage_info->header.len_used;
6307
6308 do_next_page:
6309
6310         page_header = (struct scsi_mode_page_header *)
6311                 (ctsio->kern_data_ptr + *len_used);
6312
6313         if (*len_left == 0) {
6314                 free(ctsio->kern_data_ptr, M_CTL);
6315                 ctl_set_success(ctsio);
6316                 ctl_done((union ctl_io *)ctsio);
6317                 return (CTL_RETVAL_COMPLETE);
6318         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6319
6320                 free(ctsio->kern_data_ptr, M_CTL);
6321                 ctl_set_param_len_error(ctsio);
6322                 ctl_done((union ctl_io *)ctsio);
6323                 return (CTL_RETVAL_COMPLETE);
6324
6325         } else if ((page_header->page_code & SMPH_SPF)
6326                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6327
6328                 free(ctsio->kern_data_ptr, M_CTL);
6329                 ctl_set_param_len_error(ctsio);
6330                 ctl_done((union ctl_io *)ctsio);
6331                 return (CTL_RETVAL_COMPLETE);
6332         }
6333
6334
6335         /*
6336          * XXX KDM should we do something with the block descriptor?
6337          */
6338         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6339
6340                 if ((control_dev != 0)
6341                  && (lun->mode_pages.index[i].page_flags &
6342                      CTL_PAGE_FLAG_DISK_ONLY))
6343                         continue;
6344
6345                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6346                     (page_header->page_code & SMPH_PC_MASK))
6347                         continue;
6348
6349                 /*
6350                  * If neither page has a subpage code, then we've got a
6351                  * match.
6352                  */
6353                 if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6354                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6355                         page_index = &lun->mode_pages.index[i];
6356                         page_len = page_header->page_length;
6357                         break;
6358                 }
6359
6360                 /*
6361                  * If both pages have subpages, then the subpage numbers
6362                  * have to match.
6363                  */
6364                 if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6365                   && (page_header->page_code & SMPH_SPF)) {
6366                         struct scsi_mode_page_header_sp *sph;
6367
6368                         sph = (struct scsi_mode_page_header_sp *)page_header;
6369
6370                         if (lun->mode_pages.index[i].subpage ==
6371                             sph->subpage) {
6372                                 page_index = &lun->mode_pages.index[i];
6373                                 page_len = scsi_2btoul(sph->page_length);
6374                                 break;
6375                         }
6376                 }
6377         }
6378
6379         /*
6380          * If we couldn't find the page, or if we don't have a mode select
6381          * handler for it, send back an error to the user.
6382          */
6383         if ((page_index == NULL)
6384          || (page_index->select_handler == NULL)) {
6385                 ctl_set_invalid_field(ctsio,
6386                                       /*sks_valid*/ 1,
6387                                       /*command*/ 0,
6388                                       /*field*/ *len_used,
6389                                       /*bit_valid*/ 0,
6390                                       /*bit*/ 0);
6391                 free(ctsio->kern_data_ptr, M_CTL);
6392                 ctl_done((union ctl_io *)ctsio);
6393                 return (CTL_RETVAL_COMPLETE);
6394         }
6395
6396         if (page_index->page_code & SMPH_SPF) {
6397                 page_len_offset = 2;
6398                 page_len_size = 2;
6399         } else {
6400                 page_len_size = 1;
6401                 page_len_offset = 1;
6402         }
6403
6404         /*
6405          * If the length the initiator gives us isn't the one we specify in
6406          * the mode page header, or if they didn't specify enough data in
6407          * the CDB to avoid truncating this page, kick out the request.
6408          */
6409         if ((page_len != (page_index->page_len - page_len_offset -
6410                           page_len_size))
6411          || (*len_left < page_index->page_len)) {
6412
6413
6414                 ctl_set_invalid_field(ctsio,
6415                                       /*sks_valid*/ 1,
6416                                       /*command*/ 0,
6417                                       /*field*/ *len_used + page_len_offset,
6418                                       /*bit_valid*/ 0,
6419                                       /*bit*/ 0);
6420                 free(ctsio->kern_data_ptr, M_CTL);
6421                 ctl_done((union ctl_io *)ctsio);
6422                 return (CTL_RETVAL_COMPLETE);
6423         }
6424
6425         /*
6426          * Run through the mode page, checking to make sure that the bits
6427          * the user changed are actually legal for him to change.
6428          */
6429         for (i = 0; i < page_index->page_len; i++) {
6430                 uint8_t *user_byte, *change_mask, *current_byte;
6431                 int bad_bit;
6432                 int j;
6433
6434                 user_byte = (uint8_t *)page_header + i;
6435                 change_mask = page_index->page_data +
6436                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6437                 current_byte = page_index->page_data +
6438                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6439
6440                 /*
6441                  * Check to see whether the user set any bits in this byte
6442                  * that he is not allowed to set.
6443                  */
6444                 if ((*user_byte & ~(*change_mask)) ==
6445                     (*current_byte & ~(*change_mask)))
6446                         continue;
6447
6448                 /*
6449                  * Go through bit by bit to determine which one is illegal.
6450                  */
6451                 bad_bit = 0;
6452                 for (j = 7; j >= 0; j--) {
6453                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6454                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6455                                 bad_bit = i;
6456                                 break;
6457                         }
6458                 }
6459                 ctl_set_invalid_field(ctsio,
6460                                       /*sks_valid*/ 1,
6461                                       /*command*/ 0,
6462                                       /*field*/ *len_used + i,
6463                                       /*bit_valid*/ 1,
6464                                       /*bit*/ bad_bit);
6465                 free(ctsio->kern_data_ptr, M_CTL);
6466                 ctl_done((union ctl_io *)ctsio);
6467                 return (CTL_RETVAL_COMPLETE);
6468         }
6469
6470         /*
6471          * Decrement these before we call the page handler, since we may
6472          * end up getting called back one way or another before the handler
6473          * returns to this context.
6474          */
6475         *len_left -= page_index->page_len;
6476         *len_used += page_index->page_len;
6477
6478         retval = page_index->select_handler(ctsio, page_index,
6479                                             (uint8_t *)page_header);
6480
6481         /*
6482          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6483          * wait until this queued command completes to finish processing
6484          * the mode page.  If it returns anything other than
6485          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6486          * already set the sense information, freed the data pointer, and
6487          * completed the io for us.
6488          */
6489         if (retval != CTL_RETVAL_COMPLETE)
6490                 goto bailout_no_done;
6491
6492         /*
6493          * If the initiator sent us more than one page, parse the next one.
6494          */
6495         if (*len_left > 0)
6496                 goto do_next_page;
6497
6498         ctl_set_success(ctsio);
6499         free(ctsio->kern_data_ptr, M_CTL);
6500         ctl_done((union ctl_io *)ctsio);
6501
6502 bailout_no_done:
6503
6504         return (CTL_RETVAL_COMPLETE);
6505
6506 }
6507
6508 int
6509 ctl_mode_select(struct ctl_scsiio *ctsio)
6510 {
6511         int param_len, pf, sp;
6512         int header_size, bd_len;
6513         int len_left, len_used;
6514         struct ctl_page_index *page_index;
6515         struct ctl_lun *lun;
6516         int control_dev, page_len;
6517         union ctl_modepage_info *modepage_info;
6518         int retval;
6519
6520         pf = 0;
6521         sp = 0;
6522         page_len = 0;
6523         len_used = 0;
6524         len_left = 0;
6525         retval = 0;
6526         bd_len = 0;
6527         page_index = NULL;
6528
6529         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6530
6531         if (lun->be_lun->lun_type != T_DIRECT)
6532                 control_dev = 1;
6533         else
6534                 control_dev = 0;
6535
6536         switch (ctsio->cdb[0]) {
6537         case MODE_SELECT_6: {
6538                 struct scsi_mode_select_6 *cdb;
6539
6540                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6541
6542                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6543                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6544
6545                 param_len = cdb->length;
6546                 header_size = sizeof(struct scsi_mode_header_6);
6547                 break;
6548         }
6549         case MODE_SELECT_10: {
6550                 struct scsi_mode_select_10 *cdb;
6551
6552                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6553
6554                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6555                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6556
6557                 param_len = scsi_2btoul(cdb->length);
6558                 header_size = sizeof(struct scsi_mode_header_10);
6559                 break;
6560         }
6561         default:
6562                 ctl_set_invalid_opcode(ctsio);
6563                 ctl_done((union ctl_io *)ctsio);
6564                 return (CTL_RETVAL_COMPLETE);
6565                 break; /* NOTREACHED */
6566         }
6567
6568         /*
6569          * From SPC-3:
6570          * "A parameter list length of zero indicates that the Data-Out Buffer
6571          * shall be empty. This condition shall not be considered as an error."
6572          */
6573         if (param_len == 0) {
6574                 ctl_set_success(ctsio);
6575                 ctl_done((union ctl_io *)ctsio);
6576                 return (CTL_RETVAL_COMPLETE);
6577         }
6578
6579         /*
6580          * Since we'll hit this the first time through, prior to
6581          * allocation, we don't need to free a data buffer here.
6582          */
6583         if (param_len < header_size) {
6584                 ctl_set_param_len_error(ctsio);
6585                 ctl_done((union ctl_io *)ctsio);
6586                 return (CTL_RETVAL_COMPLETE);
6587         }
6588
6589         /*
6590          * Allocate the data buffer and grab the user's data.  In theory,
6591          * we shouldn't have to sanity check the parameter list length here
6592          * because the maximum size is 64K.  We should be able to malloc
6593          * that much without too many problems.
6594          */
6595         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6596                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6597                 ctsio->kern_data_len = param_len;
6598                 ctsio->kern_total_len = param_len;
6599                 ctsio->kern_data_resid = 0;
6600                 ctsio->kern_rel_offset = 0;
6601                 ctsio->kern_sg_entries = 0;
6602                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6603                 ctsio->be_move_done = ctl_config_move_done;
6604                 ctl_datamove((union ctl_io *)ctsio);
6605
6606                 return (CTL_RETVAL_COMPLETE);
6607         }
6608
6609         switch (ctsio->cdb[0]) {
6610         case MODE_SELECT_6: {
6611                 struct scsi_mode_header_6 *mh6;
6612
6613                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6614                 bd_len = mh6->blk_desc_len;
6615                 break;
6616         }
6617         case MODE_SELECT_10: {
6618                 struct scsi_mode_header_10 *mh10;
6619
6620                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6621                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6622                 break;
6623         }
6624         default:
6625                 panic("Invalid CDB type %#x", ctsio->cdb[0]);
6626                 break;
6627         }
6628
6629         if (param_len < (header_size + bd_len)) {
6630                 free(ctsio->kern_data_ptr, M_CTL);
6631                 ctl_set_param_len_error(ctsio);
6632                 ctl_done((union ctl_io *)ctsio);
6633                 return (CTL_RETVAL_COMPLETE);
6634         }
6635
6636         /*
6637          * Set the IO_CONT flag, so that if this I/O gets passed to
6638          * ctl_config_write_done(), it'll get passed back to
6639          * ctl_do_mode_select() for further processing, or completion if
6640          * we're all done.
6641          */
6642         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6643         ctsio->io_cont = ctl_do_mode_select;
6644
6645         modepage_info = (union ctl_modepage_info *)
6646                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6647
6648         memset(modepage_info, 0, sizeof(*modepage_info));
6649
6650         len_left = param_len - header_size - bd_len;
6651         len_used = header_size + bd_len;
6652
6653         modepage_info->header.len_left = len_left;
6654         modepage_info->header.len_used = len_used;
6655
6656         return (ctl_do_mode_select((union ctl_io *)ctsio));
6657 }
6658
6659 int
6660 ctl_mode_sense(struct ctl_scsiio *ctsio)
6661 {
6662         struct ctl_lun *lun;
6663         int pc, page_code, dbd, llba, subpage;
6664         int alloc_len, page_len, header_len, total_len;
6665         struct scsi_mode_block_descr *block_desc;
6666         struct ctl_page_index *page_index;
6667         int control_dev;
6668
6669         dbd = 0;
6670         llba = 0;
6671         block_desc = NULL;
6672         page_index = NULL;
6673
6674         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6675
6676         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6677
6678         if (lun->be_lun->lun_type != T_DIRECT)
6679                 control_dev = 1;
6680         else
6681                 control_dev = 0;
6682
6683         if (lun->flags & CTL_LUN_PR_RESERVED) {
6684                 uint32_t residx;
6685
6686                 /*
6687                  * XXX KDM need a lock here.
6688                  */
6689                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6690                 if ((lun->res_type == SPR_TYPE_EX_AC
6691                   && residx != lun->pr_res_idx)
6692                  || ((lun->res_type == SPR_TYPE_EX_AC_RO
6693                    || lun->res_type == SPR_TYPE_EX_AC_AR)
6694                   && !lun->per_res[residx].registered)) {
6695                         ctl_set_reservation_conflict(ctsio);
6696                         ctl_done((union ctl_io *)ctsio);
6697                         return (CTL_RETVAL_COMPLETE);
6698                 }
6699         }
6700
6701         switch (ctsio->cdb[0]) {
6702         case MODE_SENSE_6: {
6703                 struct scsi_mode_sense_6 *cdb;
6704
6705                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6706
6707                 header_len = sizeof(struct scsi_mode_hdr_6);
6708                 if (cdb->byte2 & SMS_DBD)
6709                         dbd = 1;
6710                 else
6711                         header_len += sizeof(struct scsi_mode_block_descr);
6712
6713                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6714                 page_code = cdb->page & SMS_PAGE_CODE;
6715                 subpage = cdb->subpage;
6716                 alloc_len = cdb->length;
6717                 break;
6718         }
6719         case MODE_SENSE_10: {
6720                 struct scsi_mode_sense_10 *cdb;
6721
6722                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6723
6724                 header_len = sizeof(struct scsi_mode_hdr_10);
6725
6726                 if (cdb->byte2 & SMS_DBD)
6727                         dbd = 1;
6728                 else
6729                         header_len += sizeof(struct scsi_mode_block_descr);
6730                 if (cdb->byte2 & SMS10_LLBAA)
6731                         llba = 1;
6732                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6733                 page_code = cdb->page & SMS_PAGE_CODE;
6734                 subpage = cdb->subpage;
6735                 alloc_len = scsi_2btoul(cdb->length);
6736                 break;
6737         }
6738         default:
6739                 ctl_set_invalid_opcode(ctsio);
6740                 ctl_done((union ctl_io *)ctsio);
6741                 return (CTL_RETVAL_COMPLETE);
6742                 break; /* NOTREACHED */
6743         }
6744
6745         /*
6746          * We have to make a first pass through to calculate the size of
6747          * the pages that match the user's query.  Then we allocate enough
6748          * memory to hold it, and actually copy the data into the buffer.
6749          */
6750         switch (page_code) {
6751         case SMS_ALL_PAGES_PAGE: {
6752                 int i;
6753
6754                 page_len = 0;
6755
6756                 /*
6757                  * At the moment, values other than 0 and 0xff here are
6758                  * reserved according to SPC-3.
6759                  */
6760                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6761                  && (subpage != SMS_SUBPAGE_ALL)) {
6762                         ctl_set_invalid_field(ctsio,
6763                                               /*sks_valid*/ 1,
6764                                               /*command*/ 1,
6765                                               /*field*/ 3,
6766                                               /*bit_valid*/ 0,
6767                                               /*bit*/ 0);
6768                         ctl_done((union ctl_io *)ctsio);
6769                         return (CTL_RETVAL_COMPLETE);
6770                 }
6771
6772                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6773                         if ((control_dev != 0)
6774                          && (lun->mode_pages.index[i].page_flags &
6775                              CTL_PAGE_FLAG_DISK_ONLY))
6776                                 continue;
6777
6778                         /*
6779                          * We don't use this subpage if the user didn't
6780                          * request all subpages.
6781                          */
6782                         if ((lun->mode_pages.index[i].subpage != 0)
6783                          && (subpage == SMS_SUBPAGE_PAGE_0))
6784                                 continue;
6785
6786 #if 0
6787                         printf("found page %#x len %d\n",
6788                                lun->mode_pages.index[i].page_code &
6789                                SMPH_PC_MASK,
6790                                lun->mode_pages.index[i].page_len);
6791 #endif
6792                         page_len += lun->mode_pages.index[i].page_len;
6793                 }
6794                 break;
6795         }
6796         default: {
6797                 int i;
6798
6799                 page_len = 0;
6800
6801                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6802                         /* Look for the right page code */
6803                         if ((lun->mode_pages.index[i].page_code &
6804                              SMPH_PC_MASK) != page_code)
6805                                 continue;
6806
6807                         /* Look for the right subpage or the subpage wildcard*/
6808                         if ((lun->mode_pages.index[i].subpage != subpage)
6809                          && (subpage != SMS_SUBPAGE_ALL))
6810                                 continue;
6811
6812                         /* Make sure the page is supported for this dev type */
6813                         if ((control_dev != 0)
6814                          && (lun->mode_pages.index[i].page_flags &
6815                              CTL_PAGE_FLAG_DISK_ONLY))
6816                                 continue;
6817
6818 #if 0
6819                         printf("found page %#x len %d\n",
6820                                lun->mode_pages.index[i].page_code &
6821                                SMPH_PC_MASK,
6822                                lun->mode_pages.index[i].page_len);
6823 #endif
6824
6825                         page_len += lun->mode_pages.index[i].page_len;
6826                 }
6827
6828                 if (page_len == 0) {
6829                         ctl_set_invalid_field(ctsio,
6830                                               /*sks_valid*/ 1,
6831                                               /*command*/ 1,
6832                                               /*field*/ 2,
6833                                               /*bit_valid*/ 1,
6834                                               /*bit*/ 5);
6835                         ctl_done((union ctl_io *)ctsio);
6836                         return (CTL_RETVAL_COMPLETE);
6837                 }
6838                 break;
6839         }
6840         }
6841
6842         total_len = header_len + page_len;
6843 #if 0
6844         printf("header_len = %d, page_len = %d, total_len = %d\n",
6845                header_len, page_len, total_len);
6846 #endif
6847
6848         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6849         ctsio->kern_sg_entries = 0;
6850         ctsio->kern_data_resid = 0;
6851         ctsio->kern_rel_offset = 0;
6852         if (total_len < alloc_len) {
6853                 ctsio->residual = alloc_len - total_len;
6854                 ctsio->kern_data_len = total_len;
6855                 ctsio->kern_total_len = total_len;
6856         } else {
6857                 ctsio->residual = 0;
6858                 ctsio->kern_data_len = alloc_len;
6859                 ctsio->kern_total_len = alloc_len;
6860         }
6861
6862         switch (ctsio->cdb[0]) {
6863         case MODE_SENSE_6: {
6864                 struct scsi_mode_hdr_6 *header;
6865
6866                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6867
6868                 header->datalen = ctl_min(total_len - 1, 254);
6869
6870                 if (dbd)
6871                         header->block_descr_len = 0;
6872                 else
6873                         header->block_descr_len =
6874                                 sizeof(struct scsi_mode_block_descr);
6875                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6876                 break;
6877         }
6878         case MODE_SENSE_10: {
6879                 struct scsi_mode_hdr_10 *header;
6880                 int datalen;
6881
6882                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6883
6884                 datalen = ctl_min(total_len - 2, 65533);
6885                 scsi_ulto2b(datalen, header->datalen);
6886                 if (dbd)
6887                         scsi_ulto2b(0, header->block_descr_len);
6888                 else
6889                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6890                                     header->block_descr_len);
6891                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6892                 break;
6893         }
6894         default:
6895                 panic("invalid CDB type %#x", ctsio->cdb[0]);
6896                 break; /* NOTREACHED */
6897         }
6898
6899         /*
6900          * If we've got a disk, use its blocksize in the block
6901          * descriptor.  Otherwise, just set it to 0.
6902          */
6903         if (dbd == 0) {
6904                 if (control_dev != 0)
6905                         scsi_ulto3b(lun->be_lun->blocksize,
6906                                     block_desc->block_len);
6907                 else
6908                         scsi_ulto3b(0, block_desc->block_len);
6909         }
6910
6911         switch (page_code) {
6912         case SMS_ALL_PAGES_PAGE: {
6913                 int i, data_used;
6914
6915                 data_used = header_len;
6916                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6917                         struct ctl_page_index *page_index;
6918
6919                         page_index = &lun->mode_pages.index[i];
6920
6921                         if ((control_dev != 0)
6922                          && (page_index->page_flags &
6923                             CTL_PAGE_FLAG_DISK_ONLY))
6924                                 continue;
6925
6926                         /*
6927                          * We don't use this subpage if the user didn't
6928                          * request all subpages.  We already checked (above)
6929                          * to make sure the user only specified a subpage
6930                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6931                          */
6932                         if ((page_index->subpage != 0)
6933                          && (subpage == SMS_SUBPAGE_PAGE_0))
6934                                 continue;
6935
6936                         /*
6937                          * Call the handler, if it exists, to update the
6938                          * page to the latest values.
6939                          */
6940                         if (page_index->sense_handler != NULL)
6941                                 page_index->sense_handler(ctsio, page_index,pc);
6942
6943                         memcpy(ctsio->kern_data_ptr + data_used,
6944                                page_index->page_data +
6945                                (page_index->page_len * pc),
6946                                page_index->page_len);
6947                         data_used += page_index->page_len;
6948                 }
6949                 break;
6950         }
6951         default: {
6952                 int i, data_used;
6953
6954                 data_used = header_len;
6955
6956                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6957                         struct ctl_page_index *page_index;
6958
6959                         page_index = &lun->mode_pages.index[i];
6960
6961                         /* Look for the right page code */
6962                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6963                                 continue;
6964
6965                         /* Look for the right subpage or the subpage wildcard*/
6966                         if ((page_index->subpage != subpage)
6967                          && (subpage != SMS_SUBPAGE_ALL))
6968                                 continue;
6969
6970                         /* Make sure the page is supported for this dev type */
6971                         if ((control_dev != 0)
6972                          && (page_index->page_flags &
6973                              CTL_PAGE_FLAG_DISK_ONLY))
6974                                 continue;
6975
6976                         /*
6977                          * Call the handler, if it exists, to update the
6978                          * page to the latest values.
6979                          */
6980                         if (page_index->sense_handler != NULL)
6981                                 page_index->sense_handler(ctsio, page_index,pc);
6982
6983                         memcpy(ctsio->kern_data_ptr + data_used,
6984                                page_index->page_data +
6985                                (page_index->page_len * pc),
6986                                page_index->page_len);
6987                         data_used += page_index->page_len;
6988                 }
6989                 break;
6990         }
6991         }
6992
6993         ctsio->scsi_status = SCSI_STATUS_OK;
6994
6995         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6996         ctsio->be_move_done = ctl_config_move_done;
6997         ctl_datamove((union ctl_io *)ctsio);
6998
6999         return (CTL_RETVAL_COMPLETE);
7000 }
7001
7002 int
7003 ctl_read_capacity(struct ctl_scsiio *ctsio)
7004 {
7005         struct scsi_read_capacity *cdb;
7006         struct scsi_read_capacity_data *data;
7007         struct ctl_lun *lun;
7008         uint32_t lba;
7009
7010         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7011
7012         cdb = (struct scsi_read_capacity *)ctsio->cdb;
7013
7014         lba = scsi_4btoul(cdb->addr);
7015         if (((cdb->pmi & SRC_PMI) == 0)
7016          && (lba != 0)) {
7017                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7018                                       /*sks_valid*/ 1,
7019                                       /*command*/ 1,
7020                                       /*field*/ 2,
7021                                       /*bit_valid*/ 0,
7022                                       /*bit*/ 0);
7023                 ctl_done((union ctl_io *)ctsio);
7024                 return (CTL_RETVAL_COMPLETE);
7025         }
7026
7027         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7028
7029         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7030         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7031         ctsio->residual = 0;
7032         ctsio->kern_data_len = sizeof(*data);
7033         ctsio->kern_total_len = sizeof(*data);
7034         ctsio->kern_data_resid = 0;
7035         ctsio->kern_rel_offset = 0;
7036         ctsio->kern_sg_entries = 0;
7037
7038         /*
7039          * If the maximum LBA is greater than 0xfffffffe, the user must
7040          * issue a SERVICE ACTION IN (16) command, with the read capacity
7041          * serivce action set.
7042          */
7043         if (lun->be_lun->maxlba > 0xfffffffe)
7044                 scsi_ulto4b(0xffffffff, data->addr);
7045         else
7046                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7047
7048         /*
7049          * XXX KDM this may not be 512 bytes...
7050          */
7051         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7052
7053         ctsio->scsi_status = SCSI_STATUS_OK;
7054
7055         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7056         ctsio->be_move_done = ctl_config_move_done;
7057         ctl_datamove((union ctl_io *)ctsio);
7058
7059         return (CTL_RETVAL_COMPLETE);
7060 }
7061
7062 static int
7063 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7064 {
7065         struct scsi_read_capacity_16 *cdb;
7066         struct scsi_read_capacity_data_long *data;
7067         struct ctl_lun *lun;
7068         uint64_t lba;
7069         uint32_t alloc_len;
7070
7071         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7072
7073         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7074
7075         alloc_len = scsi_4btoul(cdb->alloc_len);
7076         lba = scsi_8btou64(cdb->addr);
7077
7078         if ((cdb->reladr & SRC16_PMI)
7079          && (lba != 0)) {
7080                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7081                                       /*sks_valid*/ 1,
7082                                       /*command*/ 1,
7083                                       /*field*/ 2,
7084                                       /*bit_valid*/ 0,
7085                                       /*bit*/ 0);
7086                 ctl_done((union ctl_io *)ctsio);
7087                 return (CTL_RETVAL_COMPLETE);
7088         }
7089
7090         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7091
7092         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7093         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7094
7095         if (sizeof(*data) < alloc_len) {
7096                 ctsio->residual = alloc_len - sizeof(*data);
7097                 ctsio->kern_data_len = sizeof(*data);
7098                 ctsio->kern_total_len = sizeof(*data);
7099         } else {
7100                 ctsio->residual = 0;
7101                 ctsio->kern_data_len = alloc_len;
7102                 ctsio->kern_total_len = alloc_len;
7103         }
7104         ctsio->kern_data_resid = 0;
7105         ctsio->kern_rel_offset = 0;
7106         ctsio->kern_sg_entries = 0;
7107
7108         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7109         /* XXX KDM this may not be 512 bytes... */
7110         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7111         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7112         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7113         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7114                 data->lalba_lbp[0] |= SRC16_LBPME;
7115
7116         ctsio->scsi_status = SCSI_STATUS_OK;
7117
7118         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7119         ctsio->be_move_done = ctl_config_move_done;
7120         ctl_datamove((union ctl_io *)ctsio);
7121
7122         return (CTL_RETVAL_COMPLETE);
7123 }
7124
7125 int
7126 ctl_service_action_in(struct ctl_scsiio *ctsio)
7127 {
7128         struct scsi_service_action_in *cdb;
7129         int retval;
7130
7131         CTL_DEBUG_PRINT(("ctl_service_action_in\n"));
7132
7133         cdb = (struct scsi_service_action_in *)ctsio->cdb;
7134
7135         retval = CTL_RETVAL_COMPLETE;
7136
7137         switch (cdb->service_action) {
7138         case SRC16_SERVICE_ACTION:
7139                 retval = ctl_read_capacity_16(ctsio);
7140                 break;
7141         default:
7142                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7143                                       /*sks_valid*/ 1,
7144                                       /*command*/ 1,
7145                                       /*field*/ 1,
7146                                       /*bit_valid*/ 1,
7147                                       /*bit*/ 4);
7148                 ctl_done((union ctl_io *)ctsio);
7149                 break;
7150         }
7151
7152         return (retval);
7153 }
7154
7155 int
7156 ctl_maintenance_in(struct ctl_scsiio *ctsio)
7157 {
7158         struct scsi_maintenance_in *cdb;
7159         int retval;
7160         int alloc_len, total_len = 0;
7161         int num_target_port_groups, single;
7162         struct ctl_lun *lun;
7163         struct ctl_softc *softc;
7164         struct scsi_target_group_data *rtg_ptr;
7165         struct scsi_target_port_group_descriptor *tpg_desc_ptr1, *tpg_desc_ptr2;
7166         struct scsi_target_port_descriptor  *tp_desc_ptr1_1, *tp_desc_ptr1_2,
7167                                             *tp_desc_ptr2_1, *tp_desc_ptr2_2;
7168
7169         CTL_DEBUG_PRINT(("ctl_maintenance_in\n"));
7170
7171         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7172         softc = control_softc;
7173         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7174
7175         retval = CTL_RETVAL_COMPLETE;
7176
7177         if ((cdb->byte2 & SERVICE_ACTION_MASK) != SA_RPRT_TRGT_GRP) {
7178                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7179                                       /*sks_valid*/ 1,
7180                                       /*command*/ 1,
7181                                       /*field*/ 1,
7182                                       /*bit_valid*/ 1,
7183                                       /*bit*/ 4);
7184                 ctl_done((union ctl_io *)ctsio);
7185                 return(retval);
7186         }
7187
7188         single = ctl_is_single;
7189         if (single)
7190                 num_target_port_groups = NUM_TARGET_PORT_GROUPS - 1;
7191         else
7192                 num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7193
7194         total_len = sizeof(struct scsi_target_group_data) +
7195                 sizeof(struct scsi_target_port_group_descriptor) *
7196                 num_target_port_groups +
7197                 sizeof(struct scsi_target_port_descriptor) *
7198                 NUM_PORTS_PER_GRP * num_target_port_groups;
7199
7200         alloc_len = scsi_4btoul(cdb->length);
7201
7202         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7203
7204         ctsio->kern_sg_entries = 0;
7205
7206         if (total_len < alloc_len) {
7207                 ctsio->residual = alloc_len - total_len;
7208                 ctsio->kern_data_len = total_len;
7209                 ctsio->kern_total_len = total_len;
7210         } else {
7211                 ctsio->residual = 0;
7212                 ctsio->kern_data_len = alloc_len;
7213                 ctsio->kern_total_len = alloc_len;
7214         }
7215         ctsio->kern_data_resid = 0;
7216         ctsio->kern_rel_offset = 0;
7217
7218         rtg_ptr = (struct scsi_target_group_data *)ctsio->kern_data_ptr;
7219
7220         tpg_desc_ptr1 = &rtg_ptr->groups[0];
7221         tp_desc_ptr1_1 = &tpg_desc_ptr1->descriptors[0];
7222         tp_desc_ptr1_2 = (struct scsi_target_port_descriptor *)
7223                 &tp_desc_ptr1_1->desc_list[0];
7224
7225         if (single == 0) {
7226                 tpg_desc_ptr2 = (struct scsi_target_port_group_descriptor *)
7227                         &tp_desc_ptr1_2->desc_list[0];
7228                 tp_desc_ptr2_1 = &tpg_desc_ptr2->descriptors[0];
7229                 tp_desc_ptr2_2 = (struct scsi_target_port_descriptor *)
7230                         &tp_desc_ptr2_1->desc_list[0];
7231         } else {
7232                 tpg_desc_ptr2 = NULL;
7233                 tp_desc_ptr2_1 = NULL;
7234                 tp_desc_ptr2_2 = NULL;
7235         }
7236
7237         scsi_ulto4b(total_len - 4, rtg_ptr->length);
7238         if (single == 0) {
7239                 if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
7240                         if (lun->flags & CTL_LUN_PRIMARY_SC) {
7241                                 tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7242                                 tpg_desc_ptr2->pref_state =
7243                                         TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7244                         } else {
7245                                 tpg_desc_ptr1->pref_state =
7246                                         TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7247                                 tpg_desc_ptr2->pref_state = TPG_PRIMARY;
7248                         }
7249                 } else {
7250                         if (lun->flags & CTL_LUN_PRIMARY_SC) {
7251                                 tpg_desc_ptr1->pref_state =
7252                                         TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7253                                 tpg_desc_ptr2->pref_state = TPG_PRIMARY;
7254                         } else {
7255                                 tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7256                                 tpg_desc_ptr2->pref_state =
7257                                         TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7258                         }
7259                 }
7260         } else {
7261                 tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7262         }
7263         tpg_desc_ptr1->support = 0;
7264         tpg_desc_ptr1->target_port_group[1] = 1;
7265         tpg_desc_ptr1->status = TPG_IMPLICIT;
7266         tpg_desc_ptr1->target_port_count= NUM_PORTS_PER_GRP;
7267
7268         if (single == 0) {
7269                 tpg_desc_ptr2->support = 0;
7270                 tpg_desc_ptr2->target_port_group[1] = 2;
7271                 tpg_desc_ptr2->status = TPG_IMPLICIT;
7272                 tpg_desc_ptr2->target_port_count = NUM_PORTS_PER_GRP;
7273
7274                 tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7275                 tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7276
7277                 tp_desc_ptr2_1->relative_target_port_identifier[1] = 9;
7278                 tp_desc_ptr2_2->relative_target_port_identifier[1] = 10;
7279         } else {
7280                 if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
7281                         tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7282                         tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7283                 } else {
7284                         tp_desc_ptr1_1->relative_target_port_identifier[1] = 9;
7285                         tp_desc_ptr1_2->relative_target_port_identifier[1] = 10;
7286                 }
7287         }
7288
7289         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7290         ctsio->be_move_done = ctl_config_move_done;
7291
7292         CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7293                          ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7294                          ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7295                          ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7296                          ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7297
7298         ctl_datamove((union ctl_io *)ctsio);
7299         return(retval);
7300 }
7301
7302 int
7303 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7304 {
7305         struct scsi_per_res_in *cdb;
7306         int alloc_len, total_len = 0;
7307         /* struct scsi_per_res_in_rsrv in_data; */
7308         struct ctl_lun *lun;
7309         struct ctl_softc *softc;
7310
7311         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7312
7313         softc = control_softc;
7314
7315         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7316
7317         alloc_len = scsi_2btoul(cdb->length);
7318
7319         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7320
7321 retry:
7322         mtx_lock(&lun->lun_lock);
7323         switch (cdb->action) {
7324         case SPRI_RK: /* read keys */
7325                 total_len = sizeof(struct scsi_per_res_in_keys) +
7326                         lun->pr_key_count *
7327                         sizeof(struct scsi_per_res_key);
7328                 break;
7329         case SPRI_RR: /* read reservation */
7330                 if (lun->flags & CTL_LUN_PR_RESERVED)
7331                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7332                 else
7333                         total_len = sizeof(struct scsi_per_res_in_header);
7334                 break;
7335         case SPRI_RC: /* report capabilities */
7336                 total_len = sizeof(struct scsi_per_res_cap);
7337                 break;
7338         case SPRI_RS: /* read full status */
7339         default:
7340                 mtx_unlock(&lun->lun_lock);
7341                 ctl_set_invalid_field(ctsio,
7342                                       /*sks_valid*/ 1,
7343                                       /*command*/ 1,
7344                                       /*field*/ 1,
7345                                       /*bit_valid*/ 1,
7346                                       /*bit*/ 0);
7347                 ctl_done((union ctl_io *)ctsio);
7348                 return (CTL_RETVAL_COMPLETE);
7349                 break; /* NOTREACHED */
7350         }
7351         mtx_unlock(&lun->lun_lock);
7352
7353         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7354
7355         if (total_len < alloc_len) {
7356                 ctsio->residual = alloc_len - total_len;
7357                 ctsio->kern_data_len = total_len;
7358                 ctsio->kern_total_len = total_len;
7359         } else {
7360                 ctsio->residual = 0;
7361                 ctsio->kern_data_len = alloc_len;
7362                 ctsio->kern_total_len = alloc_len;
7363         }
7364
7365         ctsio->kern_data_resid = 0;
7366         ctsio->kern_rel_offset = 0;
7367         ctsio->kern_sg_entries = 0;
7368
7369         mtx_lock(&lun->lun_lock);
7370         switch (cdb->action) {
7371         case SPRI_RK: { // read keys
7372         struct scsi_per_res_in_keys *res_keys;
7373                 int i, key_count;
7374
7375                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7376
7377                 /*
7378                  * We had to drop the lock to allocate our buffer, which
7379                  * leaves time for someone to come in with another
7380                  * persistent reservation.  (That is unlikely, though,
7381                  * since this should be the only persistent reservation
7382                  * command active right now.)
7383                  */
7384                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7385                     (lun->pr_key_count *
7386                      sizeof(struct scsi_per_res_key)))){
7387                         mtx_unlock(&lun->lun_lock);
7388                         free(ctsio->kern_data_ptr, M_CTL);
7389                         printf("%s: reservation length changed, retrying\n",
7390                                __func__);
7391                         goto retry;
7392                 }
7393
7394                 scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7395
7396                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7397                              lun->pr_key_count, res_keys->header.length);
7398
7399                 for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7400                         if (!lun->per_res[i].registered)
7401                                 continue;
7402
7403                         /*
7404                          * We used lun->pr_key_count to calculate the
7405                          * size to allocate.  If it turns out the number of
7406                          * initiators with the registered flag set is
7407                          * larger than that (i.e. they haven't been kept in
7408                          * sync), we've got a problem.
7409                          */
7410                         if (key_count >= lun->pr_key_count) {
7411 #ifdef NEEDTOPORT
7412                                 csevent_log(CSC_CTL | CSC_SHELF_SW |
7413                                             CTL_PR_ERROR,
7414                                             csevent_LogType_Fault,
7415                                             csevent_AlertLevel_Yellow,
7416                                             csevent_FRU_ShelfController,
7417                                             csevent_FRU_Firmware,
7418                                         csevent_FRU_Unknown,
7419                                             "registered keys %d >= key "
7420                                             "count %d", key_count,
7421                                             lun->pr_key_count);
7422 #endif
7423                                 key_count++;
7424                                 continue;
7425                         }
7426                         memcpy(res_keys->keys[key_count].key,
7427                                lun->per_res[i].res_key.key,
7428                                ctl_min(sizeof(res_keys->keys[key_count].key),
7429                                sizeof(lun->per_res[i].res_key)));
7430                         key_count++;
7431                 }
7432                 break;
7433         }
7434         case SPRI_RR: { // read reservation
7435                 struct scsi_per_res_in_rsrv *res;
7436                 int tmp_len, header_only;
7437
7438                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7439
7440                 scsi_ulto4b(lun->PRGeneration, res->header.generation);
7441
7442                 if (lun->flags & CTL_LUN_PR_RESERVED)
7443                 {
7444                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7445                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7446                                     res->header.length);
7447                         header_only = 0;
7448                 } else {
7449                         tmp_len = sizeof(struct scsi_per_res_in_header);
7450                         scsi_ulto4b(0, res->header.length);
7451                         header_only = 1;
7452                 }
7453
7454                 /*
7455                  * We had to drop the lock to allocate our buffer, which
7456                  * leaves time for someone to come in with another
7457                  * persistent reservation.  (That is unlikely, though,
7458                  * since this should be the only persistent reservation
7459                  * command active right now.)
7460                  */
7461                 if (tmp_len != total_len) {
7462                         mtx_unlock(&lun->lun_lock);
7463                         free(ctsio->kern_data_ptr, M_CTL);
7464                         printf("%s: reservation status changed, retrying\n",
7465                                __func__);
7466                         goto retry;
7467                 }
7468
7469                 /*
7470                  * No reservation held, so we're done.
7471                  */
7472                 if (header_only != 0)
7473                         break;
7474
7475                 /*
7476                  * If the registration is an All Registrants type, the key
7477                  * is 0, since it doesn't really matter.
7478                  */
7479                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7480                         memcpy(res->data.reservation,
7481                                &lun->per_res[lun->pr_res_idx].res_key,
7482                                sizeof(struct scsi_per_res_key));
7483                 }
7484                 res->data.scopetype = lun->res_type;
7485                 break;
7486         }
7487         case SPRI_RC:     //report capabilities
7488         {
7489                 struct scsi_per_res_cap *res_cap;
7490                 uint16_t type_mask;
7491
7492                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7493                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7494                 res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_3;
7495                 type_mask = SPRI_TM_WR_EX_AR |
7496                             SPRI_TM_EX_AC_RO |
7497                             SPRI_TM_WR_EX_RO |
7498                             SPRI_TM_EX_AC |
7499                             SPRI_TM_WR_EX |
7500                             SPRI_TM_EX_AC_AR;
7501                 scsi_ulto2b(type_mask, res_cap->type_mask);
7502                 break;
7503         }
7504         case SPRI_RS: //read full status
7505         default:
7506                 /*
7507                  * This is a bug, because we just checked for this above,
7508                  * and should have returned an error.
7509                  */
7510                 panic("Invalid PR type %x", cdb->action);
7511                 break; /* NOTREACHED */
7512         }
7513         mtx_unlock(&lun->lun_lock);
7514
7515         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7516         ctsio->be_move_done = ctl_config_move_done;
7517
7518         CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7519                          ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7520                          ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7521                          ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7522                          ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7523
7524         ctl_datamove((union ctl_io *)ctsio);
7525
7526         return (CTL_RETVAL_COMPLETE);
7527 }
7528
7529 /*
7530  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7531  * it should return.
7532  */
7533 static int
7534 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7535                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7536                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7537                 struct scsi_per_res_out_parms* param)
7538 {
7539         union ctl_ha_msg persis_io;
7540         int retval, i;
7541         int isc_retval;
7542
7543         retval = 0;
7544
7545         mtx_lock(&lun->lun_lock);
7546         if (sa_res_key == 0) {
7547                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7548                         /* validate scope and type */
7549                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7550                              SPR_LU_SCOPE) {
7551                                 mtx_unlock(&lun->lun_lock);
7552                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7553                                                       /*sks_valid*/ 1,
7554                                                       /*command*/ 1,
7555                                                       /*field*/ 2,
7556                                                       /*bit_valid*/ 1,
7557                                                       /*bit*/ 4);
7558                                 ctl_done((union ctl_io *)ctsio);
7559                                 return (1);
7560                         }
7561
7562                         if (type>8 || type==2 || type==4 || type==0) {
7563                                 mtx_unlock(&lun->lun_lock);
7564                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7565                                                       /*sks_valid*/ 1,
7566                                                       /*command*/ 1,
7567                                                       /*field*/ 2,
7568                                                       /*bit_valid*/ 1,
7569                                                       /*bit*/ 0);
7570                                 ctl_done((union ctl_io *)ctsio);
7571                                 return (1);
7572                         }
7573
7574                         /* temporarily unregister this nexus */
7575                         lun->per_res[residx].registered = 0;
7576
7577                         /*
7578                          * Unregister everybody else and build UA for
7579                          * them
7580                          */
7581                         for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7582                                 if (lun->per_res[i].registered == 0)
7583                                         continue;
7584
7585                                 if (!persis_offset
7586                                  && i <CTL_MAX_INITIATORS)
7587                                         lun->pending_sense[i].ua_pending |=
7588                                                 CTL_UA_REG_PREEMPT;
7589                                 else if (persis_offset
7590                                       && i >= persis_offset)
7591                                         lun->pending_sense[i-persis_offset
7592                                                 ].ua_pending |=
7593                                                 CTL_UA_REG_PREEMPT;
7594                                 lun->per_res[i].registered = 0;
7595                                 memset(&lun->per_res[i].res_key, 0,
7596                                        sizeof(struct scsi_per_res_key));
7597                         }
7598                         lun->per_res[residx].registered = 1;
7599                         lun->pr_key_count = 1;
7600                         lun->res_type = type;
7601                         if (lun->res_type != SPR_TYPE_WR_EX_AR
7602                          && lun->res_type != SPR_TYPE_EX_AC_AR)
7603                                 lun->pr_res_idx = residx;
7604
7605                         /* send msg to other side */
7606                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7607                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7608                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7609                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7610                         persis_io.pr.pr_info.res_type = type;
7611                         memcpy(persis_io.pr.pr_info.sa_res_key,
7612                                param->serv_act_res_key,
7613                                sizeof(param->serv_act_res_key));
7614                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7615                              &persis_io, sizeof(persis_io), 0)) >
7616                              CTL_HA_STATUS_SUCCESS) {
7617                                 printf("CTL:Persis Out error returned "
7618                                        "from ctl_ha_msg_send %d\n",
7619                                        isc_retval);
7620                         }
7621                 } else {
7622                         /* not all registrants */
7623                         mtx_unlock(&lun->lun_lock);
7624                         free(ctsio->kern_data_ptr, M_CTL);
7625                         ctl_set_invalid_field(ctsio,
7626                                               /*sks_valid*/ 1,
7627                                               /*command*/ 0,
7628                                               /*field*/ 8,
7629                                               /*bit_valid*/ 0,
7630                                               /*bit*/ 0);
7631                         ctl_done((union ctl_io *)ctsio);
7632                         return (1);
7633                 }
7634         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7635                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
7636                 int found = 0;
7637
7638                 if (res_key == sa_res_key) {
7639                         /* special case */
7640                         /*
7641                          * The spec implies this is not good but doesn't
7642                          * say what to do. There are two choices either
7643                          * generate a res conflict or check condition
7644                          * with illegal field in parameter data. Since
7645                          * that is what is done when the sa_res_key is
7646                          * zero I'll take that approach since this has
7647                          * to do with the sa_res_key.
7648                          */
7649                         mtx_unlock(&lun->lun_lock);
7650                         free(ctsio->kern_data_ptr, M_CTL);
7651                         ctl_set_invalid_field(ctsio,
7652                                               /*sks_valid*/ 1,
7653                                               /*command*/ 0,
7654                                               /*field*/ 8,
7655                                               /*bit_valid*/ 0,
7656                                               /*bit*/ 0);
7657                         ctl_done((union ctl_io *)ctsio);
7658                         return (1);
7659                 }
7660
7661                 for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7662                         if (lun->per_res[i].registered
7663                          && memcmp(param->serv_act_res_key,
7664                             lun->per_res[i].res_key.key,
7665                             sizeof(struct scsi_per_res_key)) != 0)
7666                                 continue;
7667
7668                         found = 1;
7669                         lun->per_res[i].registered = 0;
7670                         memset(&lun->per_res[i].res_key, 0,
7671                                sizeof(struct scsi_per_res_key));
7672                         lun->pr_key_count--;
7673
7674                         if (!persis_offset
7675                          && i < CTL_MAX_INITIATORS)
7676                                 lun->pending_sense[i].ua_pending |=
7677                                         CTL_UA_REG_PREEMPT;
7678                         else if (persis_offset
7679                               && i >= persis_offset)
7680                                 lun->pending_sense[i-persis_offset].ua_pending|=
7681                                         CTL_UA_REG_PREEMPT;
7682                 }
7683                 if (!found) {
7684                         mtx_unlock(&lun->lun_lock);
7685                         free(ctsio->kern_data_ptr, M_CTL);
7686                         ctl_set_reservation_conflict(ctsio);
7687                         ctl_done((union ctl_io *)ctsio);
7688                         return (CTL_RETVAL_COMPLETE);
7689                 }
7690                 /* send msg to other side */
7691                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7692                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7693                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7694                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
7695                 persis_io.pr.pr_info.res_type = type;
7696                 memcpy(persis_io.pr.pr_info.sa_res_key,
7697                        param->serv_act_res_key,
7698                        sizeof(param->serv_act_res_key));
7699                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7700                      &persis_io, sizeof(persis_io), 0)) >
7701                      CTL_HA_STATUS_SUCCESS) {
7702                         printf("CTL:Persis Out error returned from "
7703                                "ctl_ha_msg_send %d\n", isc_retval);
7704                 }
7705         } else {
7706                 /* Reserved but not all registrants */
7707                 /* sa_res_key is res holder */
7708                 if (memcmp(param->serv_act_res_key,
7709                    lun->per_res[lun->pr_res_idx].res_key.key,
7710                    sizeof(struct scsi_per_res_key)) == 0) {
7711                         /* validate scope and type */
7712                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7713                              SPR_LU_SCOPE) {
7714                                 mtx_unlock(&lun->lun_lock);
7715                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7716                                                       /*sks_valid*/ 1,
7717                                                       /*command*/ 1,
7718                                                       /*field*/ 2,
7719                                                       /*bit_valid*/ 1,
7720                                                       /*bit*/ 4);
7721                                 ctl_done((union ctl_io *)ctsio);
7722                                 return (1);
7723                         }
7724
7725                         if (type>8 || type==2 || type==4 || type==0) {
7726                                 mtx_unlock(&lun->lun_lock);
7727                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7728                                                       /*sks_valid*/ 1,
7729                                                       /*command*/ 1,
7730                                                       /*field*/ 2,
7731                                                       /*bit_valid*/ 1,
7732                                                       /*bit*/ 0);
7733                                 ctl_done((union ctl_io *)ctsio);
7734                                 return (1);
7735                         }
7736
7737                         /*
7738                          * Do the following:
7739                          * if sa_res_key != res_key remove all
7740                          * registrants w/sa_res_key and generate UA
7741                          * for these registrants(Registrations
7742                          * Preempted) if it wasn't an exclusive
7743                          * reservation generate UA(Reservations
7744                          * Preempted) for all other registered nexuses
7745                          * if the type has changed. Establish the new
7746                          * reservation and holder. If res_key and
7747                          * sa_res_key are the same do the above
7748                          * except don't unregister the res holder.
7749                          */
7750
7751                         /*
7752                          * Temporarily unregister so it won't get
7753                          * removed or UA generated
7754                          */
7755                         lun->per_res[residx].registered = 0;
7756                         for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7757                                 if (lun->per_res[i].registered == 0)
7758                                         continue;
7759
7760                                 if (memcmp(param->serv_act_res_key,
7761                                     lun->per_res[i].res_key.key,
7762                                     sizeof(struct scsi_per_res_key)) == 0) {
7763                                         lun->per_res[i].registered = 0;
7764                                         memset(&lun->per_res[i].res_key,
7765                                                0,
7766                                                sizeof(struct scsi_per_res_key));
7767                                         lun->pr_key_count--;
7768
7769                                         if (!persis_offset
7770                                          && i < CTL_MAX_INITIATORS)
7771                                                 lun->pending_sense[i
7772                                                         ].ua_pending |=
7773                                                         CTL_UA_REG_PREEMPT;
7774                                         else if (persis_offset
7775                                               && i >= persis_offset)
7776                                                 lun->pending_sense[
7777                                                   i-persis_offset].ua_pending |=
7778                                                   CTL_UA_REG_PREEMPT;
7779                                 } else if (type != lun->res_type
7780                                         && (lun->res_type == SPR_TYPE_WR_EX_RO
7781                                          || lun->res_type ==SPR_TYPE_EX_AC_RO)){
7782                                                 if (!persis_offset
7783                                                  && i < CTL_MAX_INITIATORS)
7784                                                         lun->pending_sense[i
7785                                                         ].ua_pending |=
7786                                                         CTL_UA_RES_RELEASE;
7787                                                 else if (persis_offset
7788                                                       && i >= persis_offset)
7789                                                         lun->pending_sense[
7790                                                         i-persis_offset
7791                                                         ].ua_pending |=
7792                                                         CTL_UA_RES_RELEASE;
7793                                 }
7794                         }
7795                         lun->per_res[residx].registered = 1;
7796                         lun->res_type = type;
7797                         if (lun->res_type != SPR_TYPE_WR_EX_AR
7798                          && lun->res_type != SPR_TYPE_EX_AC_AR)
7799                                 lun->pr_res_idx = residx;
7800                         else
7801                                 lun->pr_res_idx =
7802                                         CTL_PR_ALL_REGISTRANTS;
7803
7804                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7805                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7806                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7807                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7808                         persis_io.pr.pr_info.res_type = type;
7809                         memcpy(persis_io.pr.pr_info.sa_res_key,
7810                                param->serv_act_res_key,
7811                                sizeof(param->serv_act_res_key));
7812                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7813                              &persis_io, sizeof(persis_io), 0)) >
7814                              CTL_HA_STATUS_SUCCESS) {
7815                                 printf("CTL:Persis Out error returned "
7816                                        "from ctl_ha_msg_send %d\n",
7817                                        isc_retval);
7818                         }
7819                 } else {
7820                         /*
7821                          * sa_res_key is not the res holder just
7822                          * remove registrants
7823                          */
7824                         int found=0;
7825
7826                         for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7827                                 if (memcmp(param->serv_act_res_key,
7828                                     lun->per_res[i].res_key.key,
7829                                     sizeof(struct scsi_per_res_key)) != 0)
7830                                         continue;
7831
7832                                 found = 1;
7833                                 lun->per_res[i].registered = 0;
7834                                 memset(&lun->per_res[i].res_key, 0,
7835                                        sizeof(struct scsi_per_res_key));
7836                                 lun->pr_key_count--;
7837
7838                                 if (!persis_offset
7839                                  && i < CTL_MAX_INITIATORS)
7840                                         lun->pending_sense[i].ua_pending |=
7841                                                 CTL_UA_REG_PREEMPT;
7842                                 else if (persis_offset
7843                                       && i >= persis_offset)
7844                                         lun->pending_sense[
7845                                                 i-persis_offset].ua_pending |=
7846                                                 CTL_UA_REG_PREEMPT;
7847                         }
7848
7849                         if (!found) {
7850                                 mtx_unlock(&lun->lun_lock);
7851                                 free(ctsio->kern_data_ptr, M_CTL);
7852                                 ctl_set_reservation_conflict(ctsio);
7853                                 ctl_done((union ctl_io *)ctsio);
7854                                 return (1);
7855                         }
7856                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7857                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7858                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7859                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7860                         persis_io.pr.pr_info.res_type = type;
7861                         memcpy(persis_io.pr.pr_info.sa_res_key,
7862                                param->serv_act_res_key,
7863                                sizeof(param->serv_act_res_key));
7864                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7865                              &persis_io, sizeof(persis_io), 0)) >
7866                              CTL_HA_STATUS_SUCCESS) {
7867                                 printf("CTL:Persis Out error returned "
7868                                        "from ctl_ha_msg_send %d\n",
7869                                 isc_retval);
7870                         }
7871                 }
7872         }
7873
7874         lun->PRGeneration++;
7875         mtx_unlock(&lun->lun_lock);
7876
7877         return (retval);
7878 }
7879
7880 static void
7881 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
7882 {
7883         int i;
7884
7885         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7886          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
7887          || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
7888                    msg->pr.pr_info.sa_res_key,
7889                    sizeof(struct scsi_per_res_key)) != 0) {
7890                 uint64_t sa_res_key;
7891                 sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
7892
7893                 if (sa_res_key == 0) {
7894                         /* temporarily unregister this nexus */
7895                         lun->per_res[msg->pr.pr_info.residx].registered = 0;
7896
7897                         /*
7898                          * Unregister everybody else and build UA for
7899                          * them
7900                          */
7901                         for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7902                                 if (lun->per_res[i].registered == 0)
7903                                         continue;
7904
7905                                 if (!persis_offset
7906                                  && i < CTL_MAX_INITIATORS)
7907                                         lun->pending_sense[i].ua_pending |=
7908                                                 CTL_UA_REG_PREEMPT;
7909                                 else if (persis_offset && i >= persis_offset)
7910                                         lun->pending_sense[i -
7911                                                 persis_offset].ua_pending |=
7912                                                 CTL_UA_REG_PREEMPT;
7913                                 lun->per_res[i].registered = 0;
7914                                 memset(&lun->per_res[i].res_key, 0,
7915                                        sizeof(struct scsi_per_res_key));
7916                         }
7917
7918                         lun->per_res[msg->pr.pr_info.residx].registered = 1;
7919                         lun->pr_key_count = 1;
7920                         lun->res_type = msg->pr.pr_info.res_type;
7921                         if (lun->res_type != SPR_TYPE_WR_EX_AR
7922                          && lun->res_type != SPR_TYPE_EX_AC_AR)
7923                                 lun->pr_res_idx = msg->pr.pr_info.residx;
7924                 } else {
7925                         for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7926                                 if (memcmp(msg->pr.pr_info.sa_res_key,
7927                                    lun->per_res[i].res_key.key,
7928                                    sizeof(struct scsi_per_res_key)) != 0)
7929                                         continue;
7930
7931                                 lun->per_res[i].registered = 0;
7932                                 memset(&lun->per_res[i].res_key, 0,
7933                                        sizeof(struct scsi_per_res_key));
7934                                 lun->pr_key_count--;
7935
7936                                 if (!persis_offset
7937                                  && i < persis_offset)
7938                                         lun->pending_sense[i].ua_pending |=
7939                                                 CTL_UA_REG_PREEMPT;
7940                                 else if (persis_offset
7941                                       && i >= persis_offset)
7942                                         lun->pending_sense[i -
7943                                                 persis_offset].ua_pending |=
7944                                                 CTL_UA_REG_PREEMPT;
7945                         }
7946                 }
7947         } else {
7948                 /*
7949                  * Temporarily unregister so it won't get removed
7950                  * or UA generated
7951                  */
7952                 lun->per_res[msg->pr.pr_info.residx].registered = 0;
7953                 for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7954                         if (lun->per_res[i].registered == 0)
7955                                 continue;
7956
7957                         if (memcmp(msg->pr.pr_info.sa_res_key,
7958                            lun->per_res[i].res_key.key,
7959                            sizeof(struct scsi_per_res_key)) == 0) {
7960                                 lun->per_res[i].registered = 0;
7961                                 memset(&lun->per_res[i].res_key, 0,
7962                                        sizeof(struct scsi_per_res_key));
7963                                 lun->pr_key_count--;
7964                                 if (!persis_offset
7965                                  && i < CTL_MAX_INITIATORS)
7966                                         lun->pending_sense[i].ua_pending |=
7967                                                 CTL_UA_REG_PREEMPT;
7968                                 else if (persis_offset
7969                                       && i >= persis_offset)
7970                                         lun->pending_sense[i -
7971                                                 persis_offset].ua_pending |=
7972                                                 CTL_UA_REG_PREEMPT;
7973                         } else if (msg->pr.pr_info.res_type != lun->res_type
7974                                 && (lun->res_type == SPR_TYPE_WR_EX_RO
7975                                  || lun->res_type == SPR_TYPE_EX_AC_RO)) {
7976                                         if (!persis_offset
7977                                          && i < persis_offset)
7978                                                 lun->pending_sense[i
7979                                                         ].ua_pending |=
7980                                                         CTL_UA_RES_RELEASE;
7981                                         else if (persis_offset
7982                                               && i >= persis_offset)
7983                                         lun->pending_sense[i -
7984                                                 persis_offset].ua_pending |=
7985                                                 CTL_UA_RES_RELEASE;
7986                         }
7987                 }
7988                 lun->per_res[msg->pr.pr_info.residx].registered = 1;
7989                 lun->res_type = msg->pr.pr_info.res_type;
7990                 if (lun->res_type != SPR_TYPE_WR_EX_AR
7991                  && lun->res_type != SPR_TYPE_EX_AC_AR)
7992                         lun->pr_res_idx = msg->pr.pr_info.residx;
7993                 else
7994                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7995         }
7996         lun->PRGeneration++;
7997
7998 }
7999
8000
8001 int
8002 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8003 {
8004         int retval;
8005         int isc_retval;
8006         u_int32_t param_len;
8007         struct scsi_per_res_out *cdb;
8008         struct ctl_lun *lun;
8009         struct scsi_per_res_out_parms* param;
8010         struct ctl_softc *softc;
8011         uint32_t residx;
8012         uint64_t res_key, sa_res_key;
8013         uint8_t type;
8014         union ctl_ha_msg persis_io;
8015         int    i;
8016
8017         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8018
8019         retval = CTL_RETVAL_COMPLETE;
8020
8021         softc = control_softc;
8022
8023         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8024         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8025
8026         /*
8027          * We only support whole-LUN scope.  The scope & type are ignored for
8028          * register, register and ignore existing key and clear.
8029          * We sometimes ignore scope and type on preempts too!!
8030          * Verify reservation type here as well.
8031          */
8032         type = cdb->scope_type & SPR_TYPE_MASK;
8033         if ((cdb->action == SPRO_RESERVE)
8034          || (cdb->action == SPRO_RELEASE)) {
8035                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8036                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8037                                               /*sks_valid*/ 1,
8038                                               /*command*/ 1,
8039                                               /*field*/ 2,
8040                                               /*bit_valid*/ 1,
8041                                               /*bit*/ 4);
8042                         ctl_done((union ctl_io *)ctsio);
8043                         return (CTL_RETVAL_COMPLETE);
8044                 }
8045
8046                 if (type>8 || type==2 || type==4 || type==0) {
8047                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8048                                               /*sks_valid*/ 1,
8049                                               /*command*/ 1,
8050                                               /*field*/ 2,
8051                                               /*bit_valid*/ 1,
8052                                               /*bit*/ 0);
8053                         ctl_done((union ctl_io *)ctsio);
8054                         return (CTL_RETVAL_COMPLETE);
8055                 }
8056         }
8057
8058         switch (cdb->action & SPRO_ACTION_MASK) {
8059         case SPRO_REGISTER:
8060         case SPRO_RESERVE:
8061         case SPRO_RELEASE:
8062         case SPRO_CLEAR:
8063         case SPRO_PREEMPT:
8064         case SPRO_REG_IGNO:
8065                 break;
8066         case SPRO_REG_MOVE:
8067         case SPRO_PRE_ABO:
8068         default:
8069                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8070                                       /*sks_valid*/ 1,
8071                                       /*command*/ 1,
8072                                       /*field*/ 1,
8073                                       /*bit_valid*/ 1,
8074                                       /*bit*/ 0);
8075                 ctl_done((union ctl_io *)ctsio);
8076                 return (CTL_RETVAL_COMPLETE);
8077                 break; /* NOTREACHED */
8078         }
8079
8080         param_len = scsi_4btoul(cdb->length);
8081
8082         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8083                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8084                 ctsio->kern_data_len = param_len;
8085                 ctsio->kern_total_len = param_len;
8086                 ctsio->kern_data_resid = 0;
8087                 ctsio->kern_rel_offset = 0;
8088                 ctsio->kern_sg_entries = 0;
8089                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8090                 ctsio->be_move_done = ctl_config_move_done;
8091                 ctl_datamove((union ctl_io *)ctsio);
8092
8093                 return (CTL_RETVAL_COMPLETE);
8094         }
8095
8096         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8097
8098         residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8099         res_key = scsi_8btou64(param->res_key.key);
8100         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8101
8102         /*
8103          * Validate the reservation key here except for SPRO_REG_IGNO
8104          * This must be done for all other service actions
8105          */
8106         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8107                 mtx_lock(&lun->lun_lock);
8108                 if (lun->per_res[residx].registered) {
8109                     if (memcmp(param->res_key.key,
8110                                lun->per_res[residx].res_key.key,
8111                                ctl_min(sizeof(param->res_key),
8112                                sizeof(lun->per_res[residx].res_key))) != 0) {
8113                                 /*
8114                                  * The current key passed in doesn't match
8115                                  * the one the initiator previously
8116                                  * registered.
8117                                  */
8118                                 mtx_unlock(&lun->lun_lock);
8119                                 free(ctsio->kern_data_ptr, M_CTL);
8120                                 ctl_set_reservation_conflict(ctsio);
8121                                 ctl_done((union ctl_io *)ctsio);
8122                                 return (CTL_RETVAL_COMPLETE);
8123                         }
8124                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8125                         /*
8126                          * We are not registered
8127                          */
8128                         mtx_unlock(&lun->lun_lock);
8129                         free(ctsio->kern_data_ptr, M_CTL);
8130                         ctl_set_reservation_conflict(ctsio);
8131                         ctl_done((union ctl_io *)ctsio);
8132                         return (CTL_RETVAL_COMPLETE);
8133                 } else if (res_key != 0) {
8134                         /*
8135                          * We are not registered and trying to register but
8136                          * the register key isn't zero.
8137                          */
8138                         mtx_unlock(&lun->lun_lock);
8139                         free(ctsio->kern_data_ptr, M_CTL);
8140                         ctl_set_reservation_conflict(ctsio);
8141                         ctl_done((union ctl_io *)ctsio);
8142                         return (CTL_RETVAL_COMPLETE);
8143                 }
8144                 mtx_unlock(&lun->lun_lock);
8145         }
8146
8147         switch (cdb->action & SPRO_ACTION_MASK) {
8148         case SPRO_REGISTER:
8149         case SPRO_REG_IGNO: {
8150
8151 #if 0
8152                 printf("Registration received\n");
8153 #endif
8154
8155                 /*
8156                  * We don't support any of these options, as we report in
8157                  * the read capabilities request (see
8158                  * ctl_persistent_reserve_in(), above).
8159                  */
8160                 if ((param->flags & SPR_SPEC_I_PT)
8161                  || (param->flags & SPR_ALL_TG_PT)
8162                  || (param->flags & SPR_APTPL)) {
8163                         int bit_ptr;
8164
8165                         if (param->flags & SPR_APTPL)
8166                                 bit_ptr = 0;
8167                         else if (param->flags & SPR_ALL_TG_PT)
8168                                 bit_ptr = 2;
8169                         else /* SPR_SPEC_I_PT */
8170                                 bit_ptr = 3;
8171
8172                         free(ctsio->kern_data_ptr, M_CTL);
8173                         ctl_set_invalid_field(ctsio,
8174                                               /*sks_valid*/ 1,
8175                                               /*command*/ 0,
8176                                               /*field*/ 20,
8177                                               /*bit_valid*/ 1,
8178                                               /*bit*/ bit_ptr);
8179                         ctl_done((union ctl_io *)ctsio);
8180                         return (CTL_RETVAL_COMPLETE);
8181                 }
8182
8183                 mtx_lock(&lun->lun_lock);
8184
8185                 /*
8186                  * The initiator wants to clear the
8187                  * key/unregister.
8188                  */
8189                 if (sa_res_key == 0) {
8190                         if ((res_key == 0
8191                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8192                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8193                           && !lun->per_res[residx].registered)) {
8194                                 mtx_unlock(&lun->lun_lock);
8195                                 goto done;
8196                         }
8197
8198                         lun->per_res[residx].registered = 0;
8199                         memset(&lun->per_res[residx].res_key,
8200                                0, sizeof(lun->per_res[residx].res_key));
8201                         lun->pr_key_count--;
8202
8203                         if (residx == lun->pr_res_idx) {
8204                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8205                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8206
8207                                 if ((lun->res_type == SPR_TYPE_WR_EX_RO
8208                                   || lun->res_type == SPR_TYPE_EX_AC_RO)
8209                                  && lun->pr_key_count) {
8210                                         /*
8211                                          * If the reservation is a registrants
8212                                          * only type we need to generate a UA
8213                                          * for other registered inits.  The
8214                                          * sense code should be RESERVATIONS
8215                                          * RELEASED
8216                                          */
8217
8218                                         for (i = 0; i < CTL_MAX_INITIATORS;i++){
8219                                                 if (lun->per_res[
8220                                                     i+persis_offset].registered
8221                                                     == 0)
8222                                                         continue;
8223                                                 lun->pending_sense[i
8224                                                         ].ua_pending |=
8225                                                         CTL_UA_RES_RELEASE;
8226                                         }
8227                                 }
8228                                 lun->res_type = 0;
8229                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8230                                 if (lun->pr_key_count==0) {
8231                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8232                                         lun->res_type = 0;
8233                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8234                                 }
8235                         }
8236                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8237                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8238                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8239                         persis_io.pr.pr_info.residx = residx;
8240                         if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8241                              &persis_io, sizeof(persis_io), 0 )) >
8242                              CTL_HA_STATUS_SUCCESS) {
8243                                 printf("CTL:Persis Out error returned from "
8244                                        "ctl_ha_msg_send %d\n", isc_retval);
8245                         }
8246                 } else /* sa_res_key != 0 */ {
8247
8248                         /*
8249                          * If we aren't registered currently then increment
8250                          * the key count and set the registered flag.
8251                          */
8252                         if (!lun->per_res[residx].registered) {
8253                                 lun->pr_key_count++;
8254                                 lun->per_res[residx].registered = 1;
8255                         }
8256
8257                         memcpy(&lun->per_res[residx].res_key,
8258                                param->serv_act_res_key,
8259                                ctl_min(sizeof(param->serv_act_res_key),
8260                                sizeof(lun->per_res[residx].res_key)));
8261
8262                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8263                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8264                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8265                         persis_io.pr.pr_info.residx = residx;
8266                         memcpy(persis_io.pr.pr_info.sa_res_key,
8267                                param->serv_act_res_key,
8268                                sizeof(param->serv_act_res_key));
8269                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8270                              &persis_io, sizeof(persis_io), 0)) >
8271                              CTL_HA_STATUS_SUCCESS) {
8272                                 printf("CTL:Persis Out error returned from "
8273                                        "ctl_ha_msg_send %d\n", isc_retval);
8274                         }
8275                 }
8276                 lun->PRGeneration++;
8277                 mtx_unlock(&lun->lun_lock);
8278
8279                 break;
8280         }
8281         case SPRO_RESERVE:
8282 #if 0
8283                 printf("Reserve executed type %d\n", type);
8284 #endif
8285                 mtx_lock(&lun->lun_lock);
8286                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8287                         /*
8288                          * if this isn't the reservation holder and it's
8289                          * not a "all registrants" type or if the type is
8290                          * different then we have a conflict
8291                          */
8292                         if ((lun->pr_res_idx != residx
8293                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8294                          || lun->res_type != type) {
8295                                 mtx_unlock(&lun->lun_lock);
8296                                 free(ctsio->kern_data_ptr, M_CTL);
8297                                 ctl_set_reservation_conflict(ctsio);
8298                                 ctl_done((union ctl_io *)ctsio);
8299                                 return (CTL_RETVAL_COMPLETE);
8300                         }
8301                         mtx_unlock(&lun->lun_lock);
8302                 } else /* create a reservation */ {
8303                         /*
8304                          * If it's not an "all registrants" type record
8305                          * reservation holder
8306                          */
8307                         if (type != SPR_TYPE_WR_EX_AR
8308                          && type != SPR_TYPE_EX_AC_AR)
8309                                 lun->pr_res_idx = residx; /* Res holder */
8310                         else
8311                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8312
8313                         lun->flags |= CTL_LUN_PR_RESERVED;
8314                         lun->res_type = type;
8315
8316                         mtx_unlock(&lun->lun_lock);
8317
8318                         /* send msg to other side */
8319                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8320                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8321                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8322                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8323                         persis_io.pr.pr_info.res_type = type;
8324                         if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8325                              &persis_io, sizeof(persis_io), 0)) >
8326                              CTL_HA_STATUS_SUCCESS) {
8327                                 printf("CTL:Persis Out error returned from "
8328                                        "ctl_ha_msg_send %d\n", isc_retval);
8329                         }
8330                 }
8331                 break;
8332
8333         case SPRO_RELEASE:
8334                 mtx_lock(&lun->lun_lock);
8335                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8336                         /* No reservation exists return good status */
8337                         mtx_unlock(&lun->lun_lock);
8338                         goto done;
8339                 }
8340                 /*
8341                  * Is this nexus a reservation holder?
8342                  */
8343                 if (lun->pr_res_idx != residx
8344                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8345                         /*
8346                          * not a res holder return good status but
8347                          * do nothing
8348                          */
8349                         mtx_unlock(&lun->lun_lock);
8350                         goto done;
8351                 }
8352
8353                 if (lun->res_type != type) {
8354                         mtx_unlock(&lun->lun_lock);
8355                         free(ctsio->kern_data_ptr, M_CTL);
8356                         ctl_set_illegal_pr_release(ctsio);
8357                         ctl_done((union ctl_io *)ctsio);
8358                         return (CTL_RETVAL_COMPLETE);
8359                 }
8360
8361                 /* okay to release */
8362                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8363                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8364                 lun->res_type = 0;
8365
8366                 /*
8367                  * if this isn't an exclusive access
8368                  * res generate UA for all other
8369                  * registrants.
8370                  */
8371                 if (type != SPR_TYPE_EX_AC
8372                  && type != SPR_TYPE_WR_EX) {
8373                         /*
8374                          * temporarily unregister so we don't generate UA
8375                          */
8376                         lun->per_res[residx].registered = 0;
8377
8378                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8379                                 if (lun->per_res[i+persis_offset].registered
8380                                     == 0)
8381                                         continue;
8382                                 lun->pending_sense[i].ua_pending |=
8383                                         CTL_UA_RES_RELEASE;
8384                         }
8385
8386                         lun->per_res[residx].registered = 1;
8387                 }
8388                 mtx_unlock(&lun->lun_lock);
8389                 /* Send msg to other side */
8390                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8391                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8392                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8393                 if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8394                      sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8395                         printf("CTL:Persis Out error returned from "
8396                                "ctl_ha_msg_send %d\n", isc_retval);
8397                 }
8398                 break;
8399
8400         case SPRO_CLEAR:
8401                 /* send msg to other side */
8402
8403                 mtx_lock(&lun->lun_lock);
8404                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8405                 lun->res_type = 0;
8406                 lun->pr_key_count = 0;
8407                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8408
8409
8410                 memset(&lun->per_res[residx].res_key,
8411                        0, sizeof(lun->per_res[residx].res_key));
8412                 lun->per_res[residx].registered = 0;
8413
8414                 for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8415                         if (lun->per_res[i].registered) {
8416                                 if (!persis_offset && i < CTL_MAX_INITIATORS)
8417                                         lun->pending_sense[i].ua_pending |=
8418                                                 CTL_UA_RES_PREEMPT;
8419                                 else if (persis_offset && i >= persis_offset)
8420                                         lun->pending_sense[i-persis_offset
8421                                             ].ua_pending |= CTL_UA_RES_PREEMPT;
8422
8423                                 memset(&lun->per_res[i].res_key,
8424                                        0, sizeof(struct scsi_per_res_key));
8425                                 lun->per_res[i].registered = 0;
8426                         }
8427                 lun->PRGeneration++;
8428                 mtx_unlock(&lun->lun_lock);
8429                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8430                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8431                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8432                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8433                      sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8434                         printf("CTL:Persis Out error returned from "
8435                                "ctl_ha_msg_send %d\n", isc_retval);
8436                 }
8437                 break;
8438
8439         case SPRO_PREEMPT: {
8440                 int nretval;
8441
8442                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8443                                           residx, ctsio, cdb, param);
8444                 if (nretval != 0)
8445                         return (CTL_RETVAL_COMPLETE);
8446                 break;
8447         }
8448         case SPRO_REG_MOVE:
8449         case SPRO_PRE_ABO:
8450         default:
8451                 free(ctsio->kern_data_ptr, M_CTL);
8452                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8453                                       /*sks_valid*/ 1,
8454                                       /*command*/ 1,
8455                                       /*field*/ 1,
8456                                       /*bit_valid*/ 1,
8457                                       /*bit*/ 0);
8458                 ctl_done((union ctl_io *)ctsio);
8459                 return (CTL_RETVAL_COMPLETE);
8460                 break; /* NOTREACHED */
8461         }
8462
8463 done:
8464         free(ctsio->kern_data_ptr, M_CTL);
8465         ctl_set_success(ctsio);
8466         ctl_done((union ctl_io *)ctsio);
8467
8468         return (retval);
8469 }
8470
8471 /*
8472  * This routine is for handling a message from the other SC pertaining to
8473  * persistent reserve out. All the error checking will have been done
8474  * so only perorming the action need be done here to keep the two
8475  * in sync.
8476  */
8477 static void
8478 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8479 {
8480         struct ctl_lun *lun;
8481         struct ctl_softc *softc;
8482         int i;
8483         uint32_t targ_lun;
8484
8485         softc = control_softc;
8486
8487         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8488         lun = softc->ctl_luns[targ_lun];
8489         mtx_lock(&lun->lun_lock);
8490         switch(msg->pr.pr_info.action) {
8491         case CTL_PR_REG_KEY:
8492                 if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8493                         lun->per_res[msg->pr.pr_info.residx].registered = 1;
8494                         lun->pr_key_count++;
8495                 }
8496                 lun->PRGeneration++;
8497                 memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8498                        msg->pr.pr_info.sa_res_key,
8499                        sizeof(struct scsi_per_res_key));
8500                 break;
8501
8502         case CTL_PR_UNREG_KEY:
8503                 lun->per_res[msg->pr.pr_info.residx].registered = 0;
8504                 memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8505                        0, sizeof(struct scsi_per_res_key));
8506                 lun->pr_key_count--;
8507
8508                 /* XXX Need to see if the reservation has been released */
8509                 /* if so do we need to generate UA? */
8510                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8511                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8512                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8513
8514                         if ((lun->res_type == SPR_TYPE_WR_EX_RO
8515                           || lun->res_type == SPR_TYPE_EX_AC_RO)
8516                          && lun->pr_key_count) {
8517                                 /*
8518                                  * If the reservation is a registrants
8519                                  * only type we need to generate a UA
8520                                  * for other registered inits.  The
8521                                  * sense code should be RESERVATIONS
8522                                  * RELEASED
8523                                  */
8524
8525                                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8526                                         if (lun->per_res[i+
8527                                             persis_offset].registered == 0)
8528                                                 continue;
8529
8530                                         lun->pending_sense[i
8531                                                 ].ua_pending |=
8532                                                 CTL_UA_RES_RELEASE;
8533                                 }
8534                         }
8535                         lun->res_type = 0;
8536                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8537                         if (lun->pr_key_count==0) {
8538                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8539                                 lun->res_type = 0;
8540                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8541                         }
8542                 }
8543                 lun->PRGeneration++;
8544                 break;
8545
8546         case CTL_PR_RESERVE:
8547                 lun->flags |= CTL_LUN_PR_RESERVED;
8548                 lun->res_type = msg->pr.pr_info.res_type;
8549                 lun->pr_res_idx = msg->pr.pr_info.residx;
8550
8551                 break;
8552
8553         case CTL_PR_RELEASE:
8554                 /*
8555                  * if this isn't an exclusive access res generate UA for all
8556                  * other registrants.
8557                  */
8558                 if (lun->res_type != SPR_TYPE_EX_AC
8559                  && lun->res_type != SPR_TYPE_WR_EX) {
8560                         for (i = 0; i < CTL_MAX_INITIATORS; i++)
8561                                 if (lun->per_res[i+persis_offset].registered)
8562                                         lun->pending_sense[i].ua_pending |=
8563                                                 CTL_UA_RES_RELEASE;
8564                 }
8565
8566                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8567                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8568                 lun->res_type = 0;
8569                 break;
8570
8571         case CTL_PR_PREEMPT:
8572                 ctl_pro_preempt_other(lun, msg);
8573                 break;
8574         case CTL_PR_CLEAR:
8575                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8576                 lun->res_type = 0;
8577                 lun->pr_key_count = 0;
8578                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8579
8580                 for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8581                         if (lun->per_res[i].registered == 0)
8582                                 continue;
8583                         if (!persis_offset
8584                          && i < CTL_MAX_INITIATORS)
8585                                 lun->pending_sense[i].ua_pending |=
8586                                         CTL_UA_RES_PREEMPT;
8587                         else if (persis_offset
8588                               && i >= persis_offset)
8589                                 lun->pending_sense[i-persis_offset].ua_pending|=
8590                                         CTL_UA_RES_PREEMPT;
8591                         memset(&lun->per_res[i].res_key, 0,
8592                                sizeof(struct scsi_per_res_key));
8593                         lun->per_res[i].registered = 0;
8594                 }
8595                 lun->PRGeneration++;
8596                 break;
8597         }
8598
8599         mtx_unlock(&lun->lun_lock);
8600 }
8601
8602 int
8603 ctl_read_write(struct ctl_scsiio *ctsio)
8604 {
8605         struct ctl_lun *lun;
8606         struct ctl_lba_len_flags *lbalen;
8607         uint64_t lba;
8608         uint32_t num_blocks;
8609         int reladdr, fua, dpo, ebp;
8610         int retval;
8611         int isread;
8612
8613         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8614
8615         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8616
8617         reladdr = 0;
8618         fua = 0;
8619         dpo = 0;
8620         ebp = 0;
8621
8622         retval = CTL_RETVAL_COMPLETE;
8623
8624         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8625               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8626         if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
8627                 uint32_t residx;
8628
8629                 /*
8630                  * XXX KDM need a lock here.
8631                  */
8632                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8633                 if ((lun->res_type == SPR_TYPE_EX_AC
8634                   && residx != lun->pr_res_idx)
8635                  || ((lun->res_type == SPR_TYPE_EX_AC_RO
8636                    || lun->res_type == SPR_TYPE_EX_AC_AR)
8637                   && !lun->per_res[residx].registered)) {
8638                         ctl_set_reservation_conflict(ctsio);
8639                         ctl_done((union ctl_io *)ctsio);
8640                         return (CTL_RETVAL_COMPLETE);
8641                 }
8642         }
8643
8644         switch (ctsio->cdb[0]) {
8645         case READ_6:
8646         case WRITE_6: {
8647                 struct scsi_rw_6 *cdb;
8648
8649                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8650
8651                 lba = scsi_3btoul(cdb->addr);
8652                 /* only 5 bits are valid in the most significant address byte */
8653                 lba &= 0x1fffff;
8654                 num_blocks = cdb->length;
8655                 /*
8656                  * This is correct according to SBC-2.
8657                  */
8658                 if (num_blocks == 0)
8659                         num_blocks = 256;
8660                 break;
8661         }
8662         case READ_10:
8663         case WRITE_10: {
8664                 struct scsi_rw_10 *cdb;
8665
8666                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8667
8668                 if (cdb->byte2 & SRW10_RELADDR)
8669                         reladdr = 1;
8670                 if (cdb->byte2 & SRW10_FUA)
8671                         fua = 1;
8672                 if (cdb->byte2 & SRW10_DPO)
8673                         dpo = 1;
8674
8675                 if ((cdb->opcode == WRITE_10)
8676                  && (cdb->byte2 & SRW10_EBP))
8677                         ebp = 1;
8678
8679                 lba = scsi_4btoul(cdb->addr);
8680                 num_blocks = scsi_2btoul(cdb->length);
8681                 break;
8682         }
8683         case WRITE_VERIFY_10: {
8684                 struct scsi_write_verify_10 *cdb;
8685
8686                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8687
8688                 /*
8689                  * XXX KDM we should do actual write verify support at some
8690                  * point.  This is obviously fake, we're just translating
8691                  * things to a write.  So we don't even bother checking the
8692                  * BYTCHK field, since we don't do any verification.  If
8693                  * the user asks for it, we'll just pretend we did it.
8694                  */
8695                 if (cdb->byte2 & SWV_DPO)
8696                         dpo = 1;
8697
8698                 lba = scsi_4btoul(cdb->addr);
8699                 num_blocks = scsi_2btoul(cdb->length);
8700                 break;
8701         }
8702         case READ_12:
8703         case WRITE_12: {
8704                 struct scsi_rw_12 *cdb;
8705
8706                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8707
8708                 if (cdb->byte2 & SRW12_RELADDR)
8709                         reladdr = 1;
8710                 if (cdb->byte2 & SRW12_FUA)
8711                         fua = 1;
8712                 if (cdb->byte2 & SRW12_DPO)
8713                         dpo = 1;
8714                 lba = scsi_4btoul(cdb->addr);
8715                 num_blocks = scsi_4btoul(cdb->length);
8716                 break;
8717         }
8718         case WRITE_VERIFY_12: {
8719                 struct scsi_write_verify_12 *cdb;
8720
8721                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8722
8723                 if (cdb->byte2 & SWV_DPO)
8724                         dpo = 1;
8725
8726                 lba = scsi_4btoul(cdb->addr);
8727                 num_blocks = scsi_4btoul(cdb->length);
8728
8729                 break;
8730         }
8731         case READ_16:
8732         case WRITE_16: {
8733                 struct scsi_rw_16 *cdb;
8734
8735                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8736
8737                 if (cdb->byte2 & SRW12_RELADDR)
8738                         reladdr = 1;
8739                 if (cdb->byte2 & SRW12_FUA)
8740                         fua = 1;
8741                 if (cdb->byte2 & SRW12_DPO)
8742                         dpo = 1;
8743
8744                 lba = scsi_8btou64(cdb->addr);
8745                 num_blocks = scsi_4btoul(cdb->length);
8746                 break;
8747         }
8748         case WRITE_VERIFY_16: {
8749                 struct scsi_write_verify_16 *cdb;
8750
8751                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8752
8753                 if (cdb->byte2 & SWV_DPO)
8754                         dpo = 1;
8755
8756                 lba = scsi_8btou64(cdb->addr);
8757                 num_blocks = scsi_4btoul(cdb->length);
8758                 break;
8759         }
8760         default:
8761                 /*
8762                  * We got a command we don't support.  This shouldn't
8763                  * happen, commands should be filtered out above us.
8764                  */
8765                 ctl_set_invalid_opcode(ctsio);
8766                 ctl_done((union ctl_io *)ctsio);
8767
8768                 return (CTL_RETVAL_COMPLETE);
8769                 break; /* NOTREACHED */
8770         }
8771
8772         /*
8773          * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
8774          * interesting for us, but if RAIDCore is in write-back mode,
8775          * getting it to do write-through for a particular transaction may
8776          * not be possible.
8777          */
8778         /*
8779          * We don't support relative addressing.  That also requires
8780          * supporting linked commands, which we don't do.
8781          */
8782         if (reladdr != 0) {
8783                 ctl_set_invalid_field(ctsio,
8784                                       /*sks_valid*/ 1,
8785                                       /*command*/ 1,
8786                                       /*field*/ 1,
8787                                       /*bit_valid*/ 1,
8788                                       /*bit*/ 0);
8789                 ctl_done((union ctl_io *)ctsio);
8790                 return (CTL_RETVAL_COMPLETE);
8791         }
8792
8793         /*
8794          * The first check is to make sure we're in bounds, the second
8795          * check is to catch wrap-around problems.  If the lba + num blocks
8796          * is less than the lba, then we've wrapped around and the block
8797          * range is invalid anyway.
8798          */
8799         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8800          || ((lba + num_blocks) < lba)) {
8801                 ctl_set_lba_out_of_range(ctsio);
8802                 ctl_done((union ctl_io *)ctsio);
8803                 return (CTL_RETVAL_COMPLETE);
8804         }
8805
8806         /*
8807          * According to SBC-3, a transfer length of 0 is not an error.
8808          * Note that this cannot happen with WRITE(6) or READ(6), since 0
8809          * translates to 256 blocks for those commands.
8810          */
8811         if (num_blocks == 0) {
8812                 ctl_set_success(ctsio);
8813                 ctl_done((union ctl_io *)ctsio);
8814                 return (CTL_RETVAL_COMPLETE);
8815         }
8816
8817         lbalen = (struct ctl_lba_len_flags *)
8818             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8819         lbalen->lba = lba;
8820         lbalen->len = num_blocks;
8821         lbalen->flags = isread ? CTL_LLF_READ : CTL_LLF_WRITE;
8822
8823         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8824         ctsio->kern_rel_offset = 0;
8825
8826         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8827
8828         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8829
8830         return (retval);
8831 }
8832
8833 static int
8834 ctl_cnw_cont(union ctl_io *io)
8835 {
8836         struct ctl_scsiio *ctsio;
8837         struct ctl_lun *lun;
8838         struct ctl_lba_len_flags *lbalen;
8839         int retval;
8840
8841         ctsio = &io->scsiio;
8842         ctsio->io_hdr.status = CTL_STATUS_NONE;
8843         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8844         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8845         lbalen = (struct ctl_lba_len_flags *)
8846             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8847         lbalen->flags = CTL_LLF_WRITE;
8848
8849         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8850         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8851         return (retval);
8852 }
8853
8854 int
8855 ctl_cnw(struct ctl_scsiio *ctsio)
8856 {
8857         struct ctl_lun *lun;
8858         struct ctl_lba_len_flags *lbalen;
8859         uint64_t lba;
8860         uint32_t num_blocks;
8861         int fua, dpo;
8862         int retval;
8863
8864         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8865
8866         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8867
8868         fua = 0;
8869         dpo = 0;
8870
8871         retval = CTL_RETVAL_COMPLETE;
8872
8873         switch (ctsio->cdb[0]) {
8874         case COMPARE_AND_WRITE: {
8875                 struct scsi_compare_and_write *cdb;
8876
8877                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8878
8879                 if (cdb->byte2 & SRW10_FUA)
8880                         fua = 1;
8881                 if (cdb->byte2 & SRW10_DPO)
8882                         dpo = 1;
8883                 lba = scsi_8btou64(cdb->addr);
8884                 num_blocks = cdb->length;
8885                 break;
8886         }
8887         default:
8888                 /*
8889                  * We got a command we don't support.  This shouldn't
8890                  * happen, commands should be filtered out above us.
8891                  */
8892                 ctl_set_invalid_opcode(ctsio);
8893                 ctl_done((union ctl_io *)ctsio);
8894
8895                 return (CTL_RETVAL_COMPLETE);
8896                 break; /* NOTREACHED */
8897         }
8898
8899         /*
8900          * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
8901          * interesting for us, but if RAIDCore is in write-back mode,
8902          * getting it to do write-through for a particular transaction may
8903          * not be possible.
8904          */
8905
8906         /*
8907          * The first check is to make sure we're in bounds, the second
8908          * check is to catch wrap-around problems.  If the lba + num blocks
8909          * is less than the lba, then we've wrapped around and the block
8910          * range is invalid anyway.
8911          */
8912         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8913          || ((lba + num_blocks) < lba)) {
8914                 ctl_set_lba_out_of_range(ctsio);
8915                 ctl_done((union ctl_io *)ctsio);
8916                 return (CTL_RETVAL_COMPLETE);
8917         }
8918
8919         /*
8920          * According to SBC-3, a transfer length of 0 is not an error.
8921          */
8922         if (num_blocks == 0) {
8923                 ctl_set_success(ctsio);
8924                 ctl_done((union ctl_io *)ctsio);
8925                 return (CTL_RETVAL_COMPLETE);
8926         }
8927
8928         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8929         ctsio->kern_rel_offset = 0;
8930
8931         /*
8932          * Set the IO_CONT flag, so that if this I/O gets passed to
8933          * ctl_data_submit_done(), it'll get passed back to
8934          * ctl_ctl_cnw_cont() for further processing.
8935          */
8936         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8937         ctsio->io_cont = ctl_cnw_cont;
8938
8939         lbalen = (struct ctl_lba_len_flags *)
8940             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8941         lbalen->lba = lba;
8942         lbalen->len = num_blocks;
8943         lbalen->flags = CTL_LLF_COMPARE;
8944
8945         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8946         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8947         return (retval);
8948 }
8949
8950 int
8951 ctl_verify(struct ctl_scsiio *ctsio)
8952 {
8953         struct ctl_lun *lun;
8954         struct ctl_lba_len_flags *lbalen;
8955         uint64_t lba;
8956         uint32_t num_blocks;
8957         int bytchk, dpo;
8958         int retval;
8959
8960         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8961
8962         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8963
8964         bytchk = 0;
8965         dpo = 0;
8966         retval = CTL_RETVAL_COMPLETE;
8967
8968         switch (ctsio->cdb[0]) {
8969         case VERIFY_10: {
8970                 struct scsi_verify_10 *cdb;
8971
8972                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
8973                 if (cdb->byte2 & SVFY_BYTCHK)
8974                         bytchk = 1;
8975                 if (cdb->byte2 & SVFY_DPO)
8976                         dpo = 1;
8977                 lba = scsi_4btoul(cdb->addr);
8978                 num_blocks = scsi_2btoul(cdb->length);
8979                 break;
8980         }
8981         case VERIFY_12: {
8982                 struct scsi_verify_12 *cdb;
8983
8984                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
8985                 if (cdb->byte2 & SVFY_BYTCHK)
8986                         bytchk = 1;
8987                 if (cdb->byte2 & SVFY_DPO)
8988                         dpo = 1;
8989                 lba = scsi_4btoul(cdb->addr);
8990                 num_blocks = scsi_4btoul(cdb->length);
8991                 break;
8992         }
8993         case VERIFY_16: {
8994                 struct scsi_rw_16 *cdb;
8995
8996                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8997                 if (cdb->byte2 & SVFY_BYTCHK)
8998                         bytchk = 1;
8999                 if (cdb->byte2 & SVFY_DPO)
9000                         dpo = 1;
9001                 lba = scsi_8btou64(cdb->addr);
9002                 num_blocks = scsi_4btoul(cdb->length);
9003                 break;
9004         }
9005         default:
9006                 /*
9007                  * We got a command we don't support.  This shouldn't
9008                  * happen, commands should be filtered out above us.
9009                  */
9010                 ctl_set_invalid_opcode(ctsio);
9011                 ctl_done((union ctl_io *)ctsio);
9012                 return (CTL_RETVAL_COMPLETE);
9013         }
9014
9015         /*
9016          * The first check is to make sure we're in bounds, the second
9017          * check is to catch wrap-around problems.  If the lba + num blocks
9018          * is less than the lba, then we've wrapped around and the block
9019          * range is invalid anyway.
9020          */
9021         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9022          || ((lba + num_blocks) < lba)) {
9023                 ctl_set_lba_out_of_range(ctsio);
9024                 ctl_done((union ctl_io *)ctsio);
9025                 return (CTL_RETVAL_COMPLETE);
9026         }
9027
9028         /*
9029          * According to SBC-3, a transfer length of 0 is not an error.
9030          */
9031         if (num_blocks == 0) {
9032                 ctl_set_success(ctsio);
9033                 ctl_done((union ctl_io *)ctsio);
9034                 return (CTL_RETVAL_COMPLETE);
9035         }
9036
9037         lbalen = (struct ctl_lba_len_flags *)
9038             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9039         lbalen->lba = lba;
9040         lbalen->len = num_blocks;
9041         if (bytchk) {
9042                 lbalen->flags = CTL_LLF_COMPARE;
9043                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9044         } else {
9045                 lbalen->flags = CTL_LLF_VERIFY;
9046                 ctsio->kern_total_len = 0;
9047         }
9048         ctsio->kern_rel_offset = 0;
9049
9050         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9051         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9052         return (retval);
9053 }
9054
9055 int
9056 ctl_report_luns(struct ctl_scsiio *ctsio)
9057 {
9058         struct scsi_report_luns *cdb;
9059         struct scsi_report_luns_data *lun_data;
9060         struct ctl_lun *lun, *request_lun;
9061         int num_luns, retval;
9062         uint32_t alloc_len, lun_datalen;
9063         int num_filled, well_known;
9064         uint32_t initidx, targ_lun_id, lun_id;
9065
9066         retval = CTL_RETVAL_COMPLETE;
9067         well_known = 0;
9068
9069         cdb = (struct scsi_report_luns *)ctsio->cdb;
9070
9071         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9072
9073         mtx_lock(&control_softc->ctl_lock);
9074         num_luns = control_softc->num_luns;
9075         mtx_unlock(&control_softc->ctl_lock);
9076
9077         switch (cdb->select_report) {
9078         case RPL_REPORT_DEFAULT:
9079         case RPL_REPORT_ALL:
9080                 break;
9081         case RPL_REPORT_WELLKNOWN:
9082                 well_known = 1;
9083                 num_luns = 0;
9084                 break;
9085         default:
9086                 ctl_set_invalid_field(ctsio,
9087                                       /*sks_valid*/ 1,
9088                                       /*command*/ 1,
9089                                       /*field*/ 2,
9090                                       /*bit_valid*/ 0,
9091                                       /*bit*/ 0);
9092                 ctl_done((union ctl_io *)ctsio);
9093                 return (retval);
9094                 break; /* NOTREACHED */
9095         }
9096
9097         alloc_len = scsi_4btoul(cdb->length);
9098         /*
9099          * The initiator has to allocate at least 16 bytes for this request,
9100          * so he can at least get the header and the first LUN.  Otherwise
9101          * we reject the request (per SPC-3 rev 14, section 6.21).
9102          */
9103         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9104             sizeof(struct scsi_report_luns_lundata))) {
9105                 ctl_set_invalid_field(ctsio,
9106                                       /*sks_valid*/ 1,
9107                                       /*command*/ 1,
9108                                       /*field*/ 6,
9109                                       /*bit_valid*/ 0,
9110                                       /*bit*/ 0);
9111                 ctl_done((union ctl_io *)ctsio);
9112                 return (retval);
9113         }
9114
9115         request_lun = (struct ctl_lun *)
9116                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9117
9118         lun_datalen = sizeof(*lun_data) +
9119                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9120
9121         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9122         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9123         ctsio->kern_sg_entries = 0;
9124
9125         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9126
9127         mtx_lock(&control_softc->ctl_lock);
9128         for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9129                 lun_id = targ_lun_id;
9130                 if (ctsio->io_hdr.nexus.lun_map_fn != NULL)
9131                         lun_id = ctsio->io_hdr.nexus.lun_map_fn(ctsio->io_hdr.nexus.lun_map_arg, lun_id);
9132                 if (lun_id >= CTL_MAX_LUNS)
9133                         continue;
9134                 lun = control_softc->ctl_luns[lun_id];
9135                 if (lun == NULL)
9136                         continue;
9137
9138                 if (targ_lun_id <= 0xff) {
9139                         /*
9140                          * Peripheral addressing method, bus number 0.
9141                          */
9142                         lun_data->luns[num_filled].lundata[0] =
9143                                 RPL_LUNDATA_ATYP_PERIPH;
9144                         lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9145                         num_filled++;
9146                 } else if (targ_lun_id <= 0x3fff) {
9147                         /*
9148                          * Flat addressing method.
9149                          */
9150                         lun_data->luns[num_filled].lundata[0] =
9151                                 RPL_LUNDATA_ATYP_FLAT |
9152                                 (targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9153 #ifdef OLDCTLHEADERS
9154                                 (SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9155                                 (targ_lun_id & SRLD_BUS_LUN_MASK);
9156 #endif
9157                         lun_data->luns[num_filled].lundata[1] =
9158 #ifdef OLDCTLHEADERS
9159                                 targ_lun_id >> SRLD_BUS_LUN_BITS;
9160 #endif
9161                                 targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9162                         num_filled++;
9163                 } else {
9164                         printf("ctl_report_luns: bogus LUN number %jd, "
9165                                "skipping\n", (intmax_t)targ_lun_id);
9166                 }
9167                 /*
9168                  * According to SPC-3, rev 14 section 6.21:
9169                  *
9170                  * "The execution of a REPORT LUNS command to any valid and
9171                  * installed logical unit shall clear the REPORTED LUNS DATA
9172                  * HAS CHANGED unit attention condition for all logical
9173                  * units of that target with respect to the requesting
9174                  * initiator. A valid and installed logical unit is one
9175                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9176                  * INQUIRY data (see 6.4.2)."
9177                  *
9178                  * If request_lun is NULL, the LUN this report luns command
9179                  * was issued to is either disabled or doesn't exist. In that
9180                  * case, we shouldn't clear any pending lun change unit
9181                  * attention.
9182                  */
9183                 if (request_lun != NULL) {
9184                         mtx_lock(&lun->lun_lock);
9185                         lun->pending_sense[initidx].ua_pending &=
9186                                 ~CTL_UA_LUN_CHANGE;
9187                         mtx_unlock(&lun->lun_lock);
9188                 }
9189         }
9190         mtx_unlock(&control_softc->ctl_lock);
9191
9192         /*
9193          * It's quite possible that we've returned fewer LUNs than we allocated
9194          * space for.  Trim it.
9195          */
9196         lun_datalen = sizeof(*lun_data) +
9197                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9198
9199         if (lun_datalen < alloc_len) {
9200                 ctsio->residual = alloc_len - lun_datalen;
9201                 ctsio->kern_data_len = lun_datalen;
9202                 ctsio->kern_total_len = lun_datalen;
9203         } else {
9204                 ctsio->residual = 0;
9205                 ctsio->kern_data_len = alloc_len;
9206                 ctsio->kern_total_len = alloc_len;
9207         }
9208         ctsio->kern_data_resid = 0;
9209         ctsio->kern_rel_offset = 0;
9210         ctsio->kern_sg_entries = 0;
9211
9212         /*
9213          * We set this to the actual data length, regardless of how much
9214          * space we actually have to return results.  If the user looks at
9215          * this value, he'll know whether or not he allocated enough space
9216          * and reissue the command if necessary.  We don't support well
9217          * known logical units, so if the user asks for that, return none.
9218          */
9219         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9220
9221         /*
9222          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9223          * this request.
9224          */
9225         ctsio->scsi_status = SCSI_STATUS_OK;
9226
9227         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9228         ctsio->be_move_done = ctl_config_move_done;
9229         ctl_datamove((union ctl_io *)ctsio);
9230
9231         return (retval);
9232 }
9233
9234 int
9235 ctl_request_sense(struct ctl_scsiio *ctsio)
9236 {
9237         struct scsi_request_sense *cdb;
9238         struct scsi_sense_data *sense_ptr;
9239         struct ctl_lun *lun;
9240         uint32_t initidx;
9241         int have_error;
9242         scsi_sense_data_type sense_format;
9243
9244         cdb = (struct scsi_request_sense *)ctsio->cdb;
9245
9246         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9247
9248         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9249
9250         /*
9251          * Determine which sense format the user wants.
9252          */
9253         if (cdb->byte2 & SRS_DESC)
9254                 sense_format = SSD_TYPE_DESC;
9255         else
9256                 sense_format = SSD_TYPE_FIXED;
9257
9258         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9259         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9260         ctsio->kern_sg_entries = 0;
9261
9262         /*
9263          * struct scsi_sense_data, which is currently set to 256 bytes, is
9264          * larger than the largest allowed value for the length field in the
9265          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9266          */
9267         ctsio->residual = 0;
9268         ctsio->kern_data_len = cdb->length;
9269         ctsio->kern_total_len = cdb->length;
9270
9271         ctsio->kern_data_resid = 0;
9272         ctsio->kern_rel_offset = 0;
9273         ctsio->kern_sg_entries = 0;
9274
9275         /*
9276          * If we don't have a LUN, we don't have any pending sense.
9277          */
9278         if (lun == NULL)
9279                 goto no_sense;
9280
9281         have_error = 0;
9282         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9283         /*
9284          * Check for pending sense, and then for pending unit attentions.
9285          * Pending sense gets returned first, then pending unit attentions.
9286          */
9287         mtx_lock(&lun->lun_lock);
9288         if (ctl_is_set(lun->have_ca, initidx)) {
9289                 scsi_sense_data_type stored_format;
9290
9291                 /*
9292                  * Check to see which sense format was used for the stored
9293                  * sense data.
9294                  */
9295                 stored_format = scsi_sense_type(
9296                     &lun->pending_sense[initidx].sense);
9297
9298                 /*
9299                  * If the user requested a different sense format than the
9300                  * one we stored, then we need to convert it to the other
9301                  * format.  If we're going from descriptor to fixed format
9302                  * sense data, we may lose things in translation, depending
9303                  * on what options were used.
9304                  *
9305                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9306                  * for some reason we'll just copy it out as-is.
9307                  */
9308                 if ((stored_format == SSD_TYPE_FIXED)
9309                  && (sense_format == SSD_TYPE_DESC))
9310                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9311                             &lun->pending_sense[initidx].sense,
9312                             (struct scsi_sense_data_desc *)sense_ptr);
9313                 else if ((stored_format == SSD_TYPE_DESC)
9314                       && (sense_format == SSD_TYPE_FIXED))
9315                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9316                             &lun->pending_sense[initidx].sense,
9317                             (struct scsi_sense_data_fixed *)sense_ptr);
9318                 else
9319                         memcpy(sense_ptr, &lun->pending_sense[initidx].sense,
9320                                ctl_min(sizeof(*sense_ptr),
9321                                sizeof(lun->pending_sense[initidx].sense)));
9322
9323                 ctl_clear_mask(lun->have_ca, initidx);
9324                 have_error = 1;
9325         } else if (lun->pending_sense[initidx].ua_pending != CTL_UA_NONE) {
9326                 ctl_ua_type ua_type;
9327
9328                 ua_type = ctl_build_ua(lun->pending_sense[initidx].ua_pending,
9329                                        sense_ptr, sense_format);
9330                 if (ua_type != CTL_UA_NONE) {
9331                         have_error = 1;
9332                         /* We're reporting this UA, so clear it */
9333                         lun->pending_sense[initidx].ua_pending &= ~ua_type;
9334                 }
9335         }
9336         mtx_unlock(&lun->lun_lock);
9337
9338         /*
9339          * We already have a pending error, return it.
9340          */
9341         if (have_error != 0) {
9342                 /*
9343                  * We report the SCSI status as OK, since the status of the
9344                  * request sense command itself is OK.
9345                  */
9346                 ctsio->scsi_status = SCSI_STATUS_OK;
9347
9348                 /*
9349                  * We report 0 for the sense length, because we aren't doing
9350                  * autosense in this case.  We're reporting sense as
9351                  * parameter data.
9352                  */
9353                 ctsio->sense_len = 0;
9354                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9355                 ctsio->be_move_done = ctl_config_move_done;
9356                 ctl_datamove((union ctl_io *)ctsio);
9357
9358                 return (CTL_RETVAL_COMPLETE);
9359         }
9360
9361 no_sense:
9362
9363         /*
9364          * No sense information to report, so we report that everything is
9365          * okay.
9366          */
9367         ctl_set_sense_data(sense_ptr,
9368                            lun,
9369                            sense_format,
9370                            /*current_error*/ 1,
9371                            /*sense_key*/ SSD_KEY_NO_SENSE,
9372                            /*asc*/ 0x00,
9373                            /*ascq*/ 0x00,
9374                            SSD_ELEM_NONE);
9375
9376         ctsio->scsi_status = SCSI_STATUS_OK;
9377
9378         /*
9379          * We report 0 for the sense length, because we aren't doing
9380          * autosense in this case.  We're reporting sense as parameter data.
9381          */
9382         ctsio->sense_len = 0;
9383         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9384         ctsio->be_move_done = ctl_config_move_done;
9385         ctl_datamove((union ctl_io *)ctsio);
9386
9387         return (CTL_RETVAL_COMPLETE);
9388 }
9389
9390 int
9391 ctl_tur(struct ctl_scsiio *ctsio)
9392 {
9393         struct ctl_lun *lun;
9394
9395         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9396
9397         CTL_DEBUG_PRINT(("ctl_tur\n"));
9398
9399         if (lun == NULL)
9400                 return (EINVAL);
9401
9402         ctsio->scsi_status = SCSI_STATUS_OK;
9403         ctsio->io_hdr.status = CTL_SUCCESS;
9404
9405         ctl_done((union ctl_io *)ctsio);
9406
9407         return (CTL_RETVAL_COMPLETE);
9408 }
9409
9410 #ifdef notyet
9411 static int
9412 ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9413 {
9414
9415 }
9416 #endif
9417
9418 static int
9419 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9420 {
9421         struct scsi_vpd_supported_pages *pages;
9422         int sup_page_size;
9423         struct ctl_lun *lun;
9424
9425         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9426
9427         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9428             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9429         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9430         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9431         ctsio->kern_sg_entries = 0;
9432
9433         if (sup_page_size < alloc_len) {
9434                 ctsio->residual = alloc_len - sup_page_size;
9435                 ctsio->kern_data_len = sup_page_size;
9436                 ctsio->kern_total_len = sup_page_size;
9437         } else {
9438                 ctsio->residual = 0;
9439                 ctsio->kern_data_len = alloc_len;
9440                 ctsio->kern_total_len = alloc_len;
9441         }
9442         ctsio->kern_data_resid = 0;
9443         ctsio->kern_rel_offset = 0;
9444         ctsio->kern_sg_entries = 0;
9445
9446         /*
9447          * The control device is always connected.  The disk device, on the
9448          * other hand, may not be online all the time.  Need to change this
9449          * to figure out whether the disk device is actually online or not.
9450          */
9451         if (lun != NULL)
9452                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9453                                 lun->be_lun->lun_type;
9454         else
9455                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9456
9457         pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9458         /* Supported VPD pages */
9459         pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9460         /* Serial Number */
9461         pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9462         /* Device Identification */
9463         pages->page_list[2] = SVPD_DEVICE_ID;
9464         /* Block limits */
9465         pages->page_list[3] = SVPD_BLOCK_LIMITS;
9466         /* Logical Block Provisioning */
9467         pages->page_list[4] = SVPD_LBP;
9468
9469         ctsio->scsi_status = SCSI_STATUS_OK;
9470
9471         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9472         ctsio->be_move_done = ctl_config_move_done;
9473         ctl_datamove((union ctl_io *)ctsio);
9474
9475         return (CTL_RETVAL_COMPLETE);
9476 }
9477
9478 static int
9479 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9480 {
9481         struct scsi_vpd_unit_serial_number *sn_ptr;
9482         struct ctl_lun *lun;
9483 #ifndef CTL_USE_BACKEND_SN
9484         char tmpstr[32];
9485 #endif
9486
9487         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9488
9489         ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9490         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9491         ctsio->kern_sg_entries = 0;
9492
9493         if (sizeof(*sn_ptr) < alloc_len) {
9494                 ctsio->residual = alloc_len - sizeof(*sn_ptr);
9495                 ctsio->kern_data_len = sizeof(*sn_ptr);
9496                 ctsio->kern_total_len = sizeof(*sn_ptr);
9497         } else {
9498                 ctsio->residual = 0;
9499                 ctsio->kern_data_len = alloc_len;
9500                 ctsio->kern_total_len = alloc_len;
9501         }
9502         ctsio->kern_data_resid = 0;
9503         ctsio->kern_rel_offset = 0;
9504         ctsio->kern_sg_entries = 0;
9505
9506         /*
9507          * The control device is always connected.  The disk device, on the
9508          * other hand, may not be online all the time.  Need to change this
9509          * to figure out whether the disk device is actually online or not.
9510          */
9511         if (lun != NULL)
9512                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9513                                   lun->be_lun->lun_type;
9514         else
9515                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9516
9517         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9518         sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9519 #ifdef CTL_USE_BACKEND_SN
9520         /*
9521          * If we don't have a LUN, we just leave the serial number as
9522          * all spaces.
9523          */
9524         memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9525         if (lun != NULL) {
9526                 strncpy((char *)sn_ptr->serial_num,
9527                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9528         }
9529 #else
9530         /*
9531          * Note that we're using a non-unique serial number here,
9532          */
9533         snprintf(tmpstr, sizeof(tmpstr), "MYSERIALNUMIS000");
9534         memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9535         strncpy(sn_ptr->serial_num, tmpstr, ctl_min(CTL_SN_LEN,
9536                 ctl_min(sizeof(tmpstr), sizeof(*sn_ptr) - 4)));
9537 #endif
9538         ctsio->scsi_status = SCSI_STATUS_OK;
9539
9540         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9541         ctsio->be_move_done = ctl_config_move_done;
9542         ctl_datamove((union ctl_io *)ctsio);
9543
9544         return (CTL_RETVAL_COMPLETE);
9545 }
9546
9547
9548 static int
9549 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9550 {
9551         struct scsi_vpd_device_id *devid_ptr;
9552         struct scsi_vpd_id_descriptor *desc, *desc1;
9553         struct scsi_vpd_id_descriptor *desc2, *desc3; /* for types 4h and 5h */
9554         struct scsi_vpd_id_t10 *t10id;
9555         struct ctl_softc *ctl_softc;
9556         struct ctl_lun *lun;
9557         struct ctl_frontend *fe;
9558         char *val;
9559 #ifndef CTL_USE_BACKEND_SN
9560         char tmpstr[32];
9561 #endif /* CTL_USE_BACKEND_SN */
9562         int devid_len;
9563
9564         ctl_softc = control_softc;
9565
9566         fe = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9567
9568         if (fe->devid != NULL)
9569                 return ((fe->devid)(ctsio, alloc_len));
9570
9571         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9572
9573         devid_len = sizeof(struct scsi_vpd_device_id) +
9574                 sizeof(struct scsi_vpd_id_descriptor) +
9575                 sizeof(struct scsi_vpd_id_t10) + CTL_DEVID_LEN +
9576                 sizeof(struct scsi_vpd_id_descriptor) + CTL_WWPN_LEN +
9577                 sizeof(struct scsi_vpd_id_descriptor) +
9578                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9579                 sizeof(struct scsi_vpd_id_descriptor) +
9580                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9581
9582         ctsio->kern_data_ptr = malloc(devid_len, M_CTL, M_WAITOK | M_ZERO);
9583         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9584         ctsio->kern_sg_entries = 0;
9585
9586         if (devid_len < alloc_len) {
9587                 ctsio->residual = alloc_len - devid_len;
9588                 ctsio->kern_data_len = devid_len;
9589                 ctsio->kern_total_len = devid_len;
9590         } else {
9591                 ctsio->residual = 0;
9592                 ctsio->kern_data_len = alloc_len;
9593                 ctsio->kern_total_len = alloc_len;
9594         }
9595         ctsio->kern_data_resid = 0;
9596         ctsio->kern_rel_offset = 0;
9597         ctsio->kern_sg_entries = 0;
9598
9599         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9600         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
9601         desc1 = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9602                 sizeof(struct scsi_vpd_id_t10) + CTL_DEVID_LEN);
9603         desc2 = (struct scsi_vpd_id_descriptor *)(&desc1->identifier[0] +
9604                   CTL_WWPN_LEN);
9605         desc3 = (struct scsi_vpd_id_descriptor *)(&desc2->identifier[0] +
9606                  sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9607
9608         /*
9609          * The control device is always connected.  The disk device, on the
9610          * other hand, may not be online all the time.
9611          */
9612         if (lun != NULL)
9613                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9614                                      lun->be_lun->lun_type;
9615         else
9616                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9617
9618         devid_ptr->page_code = SVPD_DEVICE_ID;
9619
9620         scsi_ulto2b(devid_len - 4, devid_ptr->length);
9621
9622         /*
9623          * For Fibre channel,
9624          */
9625         if (fe->port_type == CTL_PORT_FC)
9626         {
9627                 desc->proto_codeset = (SCSI_PROTO_FC << 4) |
9628                                       SVPD_ID_CODESET_ASCII;
9629                 desc1->proto_codeset = (SCSI_PROTO_FC << 4) |
9630                               SVPD_ID_CODESET_BINARY;
9631         }
9632         else
9633         {
9634                 desc->proto_codeset = (SCSI_PROTO_SPI << 4) |
9635                                       SVPD_ID_CODESET_ASCII;
9636                 desc1->proto_codeset = (SCSI_PROTO_SPI << 4) |
9637                               SVPD_ID_CODESET_BINARY;
9638         }
9639         desc2->proto_codeset = desc3->proto_codeset = desc1->proto_codeset;
9640
9641         /*
9642          * We're using a LUN association here.  i.e., this device ID is a
9643          * per-LUN identifier.
9644          */
9645         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
9646         desc->length = sizeof(*t10id) + CTL_DEVID_LEN;
9647         if (lun == NULL || (val = ctl_get_opt(lun->be_lun, "vendor")) == NULL) {
9648                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
9649         } else {
9650                 memset(t10id->vendor, ' ', sizeof(t10id->vendor));
9651                 strncpy(t10id->vendor, val,
9652                     min(sizeof(t10id->vendor), strlen(val)));
9653         }
9654
9655         /*
9656          * desc1 is for the WWPN which is a port asscociation.
9657          */
9658         desc1->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT | SVPD_ID_TYPE_NAA;
9659         desc1->length = CTL_WWPN_LEN;
9660         /* XXX Call Reggie's get_WWNN func here then add port # to the end */
9661         /* For testing just create the WWPN */
9662 #if 0
9663         ddb_GetWWNN((char *)desc1->identifier);
9664
9665         /* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9666         /* This is so Copancontrol will return something sane */
9667         if (ctsio->io_hdr.nexus.targ_port!=0 &&
9668             ctsio->io_hdr.nexus.targ_port!=8)
9669                 desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port-1;
9670         else
9671                 desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port;
9672 #endif
9673
9674         be64enc(desc1->identifier, fe->wwpn);
9675
9676         /*
9677          * desc2 is for the Relative Target Port(type 4h) identifier
9678          */
9679         desc2->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9680                          | SVPD_ID_TYPE_RELTARG;
9681         desc2->length = 4;
9682 //#if 0
9683         /* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9684         /* This is so Copancontrol will return something sane */
9685         if (ctsio->io_hdr.nexus.targ_port!=0 &&
9686             ctsio->io_hdr.nexus.targ_port!=8)
9687                 desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port - 1;
9688         else
9689                 desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port;
9690 //#endif
9691
9692         /*
9693          * desc3 is for the Target Port Group(type 5h) identifier
9694          */
9695         desc3->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9696                          | SVPD_ID_TYPE_TPORTGRP;
9697         desc3->length = 4;
9698         if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS || ctl_is_single)
9699                 desc3->identifier[3] = 1;
9700         else
9701                 desc3->identifier[3] = 2;
9702
9703 #ifdef CTL_USE_BACKEND_SN
9704         /*
9705          * If we've actually got a backend, copy the device id from the
9706          * per-LUN data.  Otherwise, set it to all spaces.
9707          */
9708         if (lun != NULL) {
9709                 /*
9710                  * Copy the backend's LUN ID.
9711                  */
9712                 strncpy((char *)t10id->vendor_spec_id,
9713                         (char *)lun->be_lun->device_id, CTL_DEVID_LEN);
9714         } else {
9715                 /*
9716                  * No backend, set this to spaces.
9717                  */
9718                 memset(t10id->vendor_spec_id, 0x20, CTL_DEVID_LEN);
9719         }
9720 #else
9721         snprintf(tmpstr, sizeof(tmpstr), "MYDEVICEIDIS%4d",
9722                  (lun != NULL) ?  (int)lun->lun : 0);
9723         strncpy(t10id->vendor_spec_id, tmpstr, ctl_min(CTL_DEVID_LEN,
9724                 sizeof(tmpstr)));
9725 #endif
9726
9727         ctsio->scsi_status = SCSI_STATUS_OK;
9728
9729         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9730         ctsio->be_move_done = ctl_config_move_done;
9731         ctl_datamove((union ctl_io *)ctsio);
9732
9733         return (CTL_RETVAL_COMPLETE);
9734 }
9735
9736 static int
9737 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9738 {
9739         struct scsi_vpd_block_limits *bl_ptr;
9740         struct ctl_lun *lun;
9741         int bs;
9742
9743         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9744         bs = lun->be_lun->blocksize;
9745
9746         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9747         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9748         ctsio->kern_sg_entries = 0;
9749
9750         if (sizeof(*bl_ptr) < alloc_len) {
9751                 ctsio->residual = alloc_len - sizeof(*bl_ptr);
9752                 ctsio->kern_data_len = sizeof(*bl_ptr);
9753                 ctsio->kern_total_len = sizeof(*bl_ptr);
9754         } else {
9755                 ctsio->residual = 0;
9756                 ctsio->kern_data_len = alloc_len;
9757                 ctsio->kern_total_len = alloc_len;
9758         }
9759         ctsio->kern_data_resid = 0;
9760         ctsio->kern_rel_offset = 0;
9761         ctsio->kern_sg_entries = 0;
9762
9763         /*
9764          * The control device is always connected.  The disk device, on the
9765          * other hand, may not be online all the time.  Need to change this
9766          * to figure out whether the disk device is actually online or not.
9767          */
9768         if (lun != NULL)
9769                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9770                                   lun->be_lun->lun_type;
9771         else
9772                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9773
9774         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9775         scsi_ulto2b(sizeof(*bl_ptr), bl_ptr->page_length);
9776         bl_ptr->max_cmp_write_len = 0xff;
9777         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9778         scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
9779         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9780                 scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
9781                 scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
9782         }
9783         scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
9784
9785         ctsio->scsi_status = SCSI_STATUS_OK;
9786         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9787         ctsio->be_move_done = ctl_config_move_done;
9788         ctl_datamove((union ctl_io *)ctsio);
9789
9790         return (CTL_RETVAL_COMPLETE);
9791 }
9792
9793 static int
9794 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9795 {
9796         struct scsi_vpd_logical_block_prov *lbp_ptr;
9797         struct ctl_lun *lun;
9798         int bs;
9799
9800         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9801         bs = lun->be_lun->blocksize;
9802
9803         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9804         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9805         ctsio->kern_sg_entries = 0;
9806
9807         if (sizeof(*lbp_ptr) < alloc_len) {
9808                 ctsio->residual = alloc_len - sizeof(*lbp_ptr);
9809                 ctsio->kern_data_len = sizeof(*lbp_ptr);
9810                 ctsio->kern_total_len = sizeof(*lbp_ptr);
9811         } else {
9812                 ctsio->residual = 0;
9813                 ctsio->kern_data_len = alloc_len;
9814                 ctsio->kern_total_len = alloc_len;
9815         }
9816         ctsio->kern_data_resid = 0;
9817         ctsio->kern_rel_offset = 0;
9818         ctsio->kern_sg_entries = 0;
9819
9820         /*
9821          * The control device is always connected.  The disk device, on the
9822          * other hand, may not be online all the time.  Need to change this
9823          * to figure out whether the disk device is actually online or not.
9824          */
9825         if (lun != NULL)
9826                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9827                                   lun->be_lun->lun_type;
9828         else
9829                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9830
9831         lbp_ptr->page_code = SVPD_LBP;
9832         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
9833                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 | SVPD_LBP_WS10;
9834
9835         ctsio->scsi_status = SCSI_STATUS_OK;
9836         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9837         ctsio->be_move_done = ctl_config_move_done;
9838         ctl_datamove((union ctl_io *)ctsio);
9839
9840         return (CTL_RETVAL_COMPLETE);
9841 }
9842
9843 static int
9844 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9845 {
9846         struct scsi_inquiry *cdb;
9847         struct ctl_lun *lun;
9848         int alloc_len, retval;
9849
9850         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9851         cdb = (struct scsi_inquiry *)ctsio->cdb;
9852
9853         retval = CTL_RETVAL_COMPLETE;
9854
9855         alloc_len = scsi_2btoul(cdb->length);
9856
9857         switch (cdb->page_code) {
9858         case SVPD_SUPPORTED_PAGES:
9859                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9860                 break;
9861         case SVPD_UNIT_SERIAL_NUMBER:
9862                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9863                 break;
9864         case SVPD_DEVICE_ID:
9865                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9866                 break;
9867         case SVPD_BLOCK_LIMITS:
9868                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
9869                 break;
9870         case SVPD_LBP:
9871                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
9872                 break;
9873         default:
9874                 ctl_set_invalid_field(ctsio,
9875                                       /*sks_valid*/ 1,
9876                                       /*command*/ 1,
9877                                       /*field*/ 2,
9878                                       /*bit_valid*/ 0,
9879                                       /*bit*/ 0);
9880                 ctl_done((union ctl_io *)ctsio);
9881                 retval = CTL_RETVAL_COMPLETE;
9882                 break;
9883         }
9884
9885         return (retval);
9886 }
9887
9888 static int
9889 ctl_inquiry_std(struct ctl_scsiio *ctsio)
9890 {
9891         struct scsi_inquiry_data *inq_ptr;
9892         struct scsi_inquiry *cdb;
9893         struct ctl_softc *ctl_softc;
9894         struct ctl_lun *lun;
9895         char *val;
9896         uint32_t alloc_len;
9897         int is_fc;
9898
9899         ctl_softc = control_softc;
9900
9901         /*
9902          * Figure out whether we're talking to a Fibre Channel port or not.
9903          * We treat the ioctl front end, and any SCSI adapters, as packetized
9904          * SCSI front ends.
9905          */
9906         if (ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type !=
9907             CTL_PORT_FC)
9908                 is_fc = 0;
9909         else
9910                 is_fc = 1;
9911
9912         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9913         cdb = (struct scsi_inquiry *)ctsio->cdb;
9914         alloc_len = scsi_2btoul(cdb->length);
9915
9916         /*
9917          * We malloc the full inquiry data size here and fill it
9918          * in.  If the user only asks for less, we'll give him
9919          * that much.
9920          */
9921         ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
9922         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
9923         ctsio->kern_sg_entries = 0;
9924         ctsio->kern_data_resid = 0;
9925         ctsio->kern_rel_offset = 0;
9926
9927         if (sizeof(*inq_ptr) < alloc_len) {
9928                 ctsio->residual = alloc_len - sizeof(*inq_ptr);
9929                 ctsio->kern_data_len = sizeof(*inq_ptr);
9930                 ctsio->kern_total_len = sizeof(*inq_ptr);
9931         } else {
9932                 ctsio->residual = 0;
9933                 ctsio->kern_data_len = alloc_len;
9934                 ctsio->kern_total_len = alloc_len;
9935         }
9936
9937         /*
9938          * If we have a LUN configured, report it as connected.  Otherwise,
9939          * report that it is offline or no device is supported, depending
9940          * on the value of inquiry_pq_no_lun.
9941          *
9942          * According to the spec (SPC-4 r34), the peripheral qualifier
9943          * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
9944          *
9945          * "A peripheral device having the specified peripheral device type
9946          * is not connected to this logical unit. However, the device
9947          * server is capable of supporting the specified peripheral device
9948          * type on this logical unit."
9949          *
9950          * According to the same spec, the peripheral qualifier
9951          * SID_QUAL_BAD_LU (011b) is used in this scenario:
9952          *
9953          * "The device server is not capable of supporting a peripheral
9954          * device on this logical unit. For this peripheral qualifier the
9955          * peripheral device type shall be set to 1Fh. All other peripheral
9956          * device type values are reserved for this peripheral qualifier."
9957          *
9958          * Given the text, it would seem that we probably want to report that
9959          * the LUN is offline here.  There is no LUN connected, but we can
9960          * support a LUN at the given LUN number.
9961          *
9962          * In the real world, though, it sounds like things are a little
9963          * different:
9964          *
9965          * - Linux, when presented with a LUN with the offline peripheral
9966          *   qualifier, will create an sg driver instance for it.  So when
9967          *   you attach it to CTL, you wind up with a ton of sg driver
9968          *   instances.  (One for every LUN that Linux bothered to probe.)
9969          *   Linux does this despite the fact that it issues a REPORT LUNs
9970          *   to LUN 0 to get the inventory of supported LUNs.
9971          *
9972          * - There is other anecdotal evidence (from Emulex folks) about
9973          *   arrays that use the offline peripheral qualifier for LUNs that
9974          *   are on the "passive" path in an active/passive array.
9975          *
9976          * So the solution is provide a hopefully reasonable default
9977          * (return bad/no LUN) and allow the user to change the behavior
9978          * with a tunable/sysctl variable.
9979          */
9980         if (lun != NULL)
9981                 inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9982                                   lun->be_lun->lun_type;
9983         else if (ctl_softc->inquiry_pq_no_lun == 0)
9984                 inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9985         else
9986                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
9987
9988         /* RMB in byte 2 is 0 */
9989         inq_ptr->version = SCSI_REV_SPC3;
9990
9991         /*
9992          * According to SAM-3, even if a device only supports a single
9993          * level of LUN addressing, it should still set the HISUP bit:
9994          *
9995          * 4.9.1 Logical unit numbers overview
9996          *
9997          * All logical unit number formats described in this standard are
9998          * hierarchical in structure even when only a single level in that
9999          * hierarchy is used. The HISUP bit shall be set to one in the
10000          * standard INQUIRY data (see SPC-2) when any logical unit number
10001          * format described in this standard is used.  Non-hierarchical
10002          * formats are outside the scope of this standard.
10003          *
10004          * Therefore we set the HiSup bit here.
10005          *
10006          * The reponse format is 2, per SPC-3.
10007          */
10008         inq_ptr->response_format = SID_HiSup | 2;
10009
10010         inq_ptr->additional_length = sizeof(*inq_ptr) - 4;
10011         CTL_DEBUG_PRINT(("additional_length = %d\n",
10012                          inq_ptr->additional_length));
10013
10014         inq_ptr->spc3_flags = SPC3_SID_TPGS_IMPLICIT;
10015         /* 16 bit addressing */
10016         if (is_fc == 0)
10017                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10018         /* XXX set the SID_MultiP bit here if we're actually going to
10019            respond on multiple ports */
10020         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10021
10022         /* 16 bit data bus, synchronous transfers */
10023         /* XXX these flags don't apply for FC */
10024         if (is_fc == 0)
10025                 inq_ptr->flags = SID_WBus16 | SID_Sync;
10026         /*
10027          * XXX KDM do we want to support tagged queueing on the control
10028          * device at all?
10029          */
10030         if ((lun == NULL)
10031          || (lun->be_lun->lun_type != T_PROCESSOR))
10032                 inq_ptr->flags |= SID_CmdQue;
10033         /*
10034          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10035          * We have 8 bytes for the vendor name, and 16 bytes for the device
10036          * name and 4 bytes for the revision.
10037          */
10038         if (lun == NULL || (val = ctl_get_opt(lun->be_lun, "vendor")) == NULL) {
10039                 strcpy(inq_ptr->vendor, CTL_VENDOR);
10040         } else {
10041                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10042                 strncpy(inq_ptr->vendor, val,
10043                     min(sizeof(inq_ptr->vendor), strlen(val)));
10044         }
10045         if (lun == NULL) {
10046                 strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10047         } else if ((val = ctl_get_opt(lun->be_lun, "product")) == NULL) {
10048                 switch (lun->be_lun->lun_type) {
10049                 case T_DIRECT:
10050                         strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10051                         break;
10052                 case T_PROCESSOR:
10053                         strcpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT);
10054                         break;
10055                 default:
10056                         strcpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT);
10057                         break;
10058                 }
10059         } else {
10060                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10061                 strncpy(inq_ptr->product, val,
10062                     min(sizeof(inq_ptr->product), strlen(val)));
10063         }
10064
10065         /*
10066          * XXX make this a macro somewhere so it automatically gets
10067          * incremented when we make changes.
10068          */
10069         if (lun == NULL || (val = ctl_get_opt(lun->be_lun, "revision")) == NULL) {
10070                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10071         } else {
10072                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10073                 strncpy(inq_ptr->revision, val,
10074                     min(sizeof(inq_ptr->revision), strlen(val)));
10075         }
10076
10077         /*
10078          * For parallel SCSI, we support double transition and single
10079          * transition clocking.  We also support QAS (Quick Arbitration
10080          * and Selection) and Information Unit transfers on both the
10081          * control and array devices.
10082          */
10083         if (is_fc == 0)
10084                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10085                                     SID_SPI_IUS;
10086
10087         /* SAM-3 */
10088         scsi_ulto2b(0x0060, inq_ptr->version1);
10089         /* SPC-3 (no version claimed) XXX should we claim a version? */
10090         scsi_ulto2b(0x0300, inq_ptr->version2);
10091         if (is_fc) {
10092                 /* FCP-2 ANSI INCITS.350:2003 */
10093                 scsi_ulto2b(0x0917, inq_ptr->version3);
10094         } else {
10095                 /* SPI-4 ANSI INCITS.362:200x */
10096                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10097         }
10098
10099         if (lun == NULL) {
10100                 /* SBC-2 (no version claimed) XXX should we claim a version? */
10101                 scsi_ulto2b(0x0320, inq_ptr->version4);
10102         } else {
10103                 switch (lun->be_lun->lun_type) {
10104                 case T_DIRECT:
10105                         /*
10106                          * SBC-2 (no version claimed) XXX should we claim a
10107                          * version?
10108                          */
10109                         scsi_ulto2b(0x0320, inq_ptr->version4);
10110                         break;
10111                 case T_PROCESSOR:
10112                 default:
10113                         break;
10114                 }
10115         }
10116
10117         ctsio->scsi_status = SCSI_STATUS_OK;
10118         if (ctsio->kern_data_len > 0) {
10119                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10120                 ctsio->be_move_done = ctl_config_move_done;
10121                 ctl_datamove((union ctl_io *)ctsio);
10122         } else {
10123                 ctsio->io_hdr.status = CTL_SUCCESS;
10124                 ctl_done((union ctl_io *)ctsio);
10125         }
10126
10127         return (CTL_RETVAL_COMPLETE);
10128 }
10129
10130 int
10131 ctl_inquiry(struct ctl_scsiio *ctsio)
10132 {
10133         struct scsi_inquiry *cdb;
10134         int retval;
10135
10136         cdb = (struct scsi_inquiry *)ctsio->cdb;
10137
10138         retval = 0;
10139
10140         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10141
10142         /*
10143          * Right now, we don't support the CmdDt inquiry information.
10144          * This would be nice to support in the future.  When we do
10145          * support it, we should change this test so that it checks to make
10146          * sure SI_EVPD and SI_CMDDT aren't both set at the same time.
10147          */
10148 #ifdef notyet
10149         if (((cdb->byte2 & SI_EVPD)
10150          && (cdb->byte2 & SI_CMDDT)))
10151 #endif
10152         if (cdb->byte2 & SI_CMDDT) {
10153                 /*
10154                  * Point to the SI_CMDDT bit.  We might change this
10155                  * when we support SI_CMDDT, but since both bits would be
10156                  * "wrong", this should probably just stay as-is then.
10157                  */
10158                 ctl_set_invalid_field(ctsio,
10159                                       /*sks_valid*/ 1,
10160                                       /*command*/ 1,
10161                                       /*field*/ 1,
10162                                       /*bit_valid*/ 1,
10163                                       /*bit*/ 1);
10164                 ctl_done((union ctl_io *)ctsio);
10165                 return (CTL_RETVAL_COMPLETE);
10166         }
10167         if (cdb->byte2 & SI_EVPD)
10168                 retval = ctl_inquiry_evpd(ctsio);
10169 #ifdef notyet
10170         else if (cdb->byte2 & SI_CMDDT)
10171                 retval = ctl_inquiry_cmddt(ctsio);
10172 #endif
10173         else
10174                 retval = ctl_inquiry_std(ctsio);
10175
10176         return (retval);
10177 }
10178
10179 /*
10180  * For known CDB types, parse the LBA and length.
10181  */
10182 static int
10183 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len)
10184 {
10185         if (io->io_hdr.io_type != CTL_IO_SCSI)
10186                 return (1);
10187
10188         switch (io->scsiio.cdb[0]) {
10189         case COMPARE_AND_WRITE: {
10190                 struct scsi_compare_and_write *cdb;
10191
10192                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10193
10194                 *lba = scsi_8btou64(cdb->addr);
10195                 *len = cdb->length;
10196                 break;
10197         }
10198         case READ_6:
10199         case WRITE_6: {
10200                 struct scsi_rw_6 *cdb;
10201
10202                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10203
10204                 *lba = scsi_3btoul(cdb->addr);
10205                 /* only 5 bits are valid in the most significant address byte */
10206                 *lba &= 0x1fffff;
10207                 *len = cdb->length;
10208                 break;
10209         }
10210         case READ_10:
10211         case WRITE_10: {
10212                 struct scsi_rw_10 *cdb;
10213
10214                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10215
10216                 *lba = scsi_4btoul(cdb->addr);
10217                 *len = scsi_2btoul(cdb->length);
10218                 break;
10219         }
10220         case WRITE_VERIFY_10: {
10221                 struct scsi_write_verify_10 *cdb;
10222
10223                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10224
10225                 *lba = scsi_4btoul(cdb->addr);
10226                 *len = scsi_2btoul(cdb->length);
10227                 break;
10228         }
10229         case READ_12:
10230         case WRITE_12: {
10231                 struct scsi_rw_12 *cdb;
10232
10233                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10234
10235                 *lba = scsi_4btoul(cdb->addr);
10236                 *len = scsi_4btoul(cdb->length);
10237                 break;
10238         }
10239         case WRITE_VERIFY_12: {
10240                 struct scsi_write_verify_12 *cdb;
10241
10242                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10243
10244                 *lba = scsi_4btoul(cdb->addr);
10245                 *len = scsi_4btoul(cdb->length);
10246                 break;
10247         }
10248         case READ_16:
10249         case WRITE_16: {
10250                 struct scsi_rw_16 *cdb;
10251
10252                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10253
10254                 *lba = scsi_8btou64(cdb->addr);
10255                 *len = scsi_4btoul(cdb->length);
10256                 break;
10257         }
10258         case WRITE_VERIFY_16: {
10259                 struct scsi_write_verify_16 *cdb;
10260
10261                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10262
10263
10264                 *lba = scsi_8btou64(cdb->addr);
10265                 *len = scsi_4btoul(cdb->length);
10266                 break;
10267         }
10268         case WRITE_SAME_10: {
10269                 struct scsi_write_same_10 *cdb;
10270
10271                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10272
10273                 *lba = scsi_4btoul(cdb->addr);
10274                 *len = scsi_2btoul(cdb->length);
10275                 break;
10276         }
10277         case WRITE_SAME_16: {
10278                 struct scsi_write_same_16 *cdb;
10279
10280                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10281
10282                 *lba = scsi_8btou64(cdb->addr);
10283                 *len = scsi_4btoul(cdb->length);
10284                 break;
10285         }
10286         case VERIFY_10: {
10287                 struct scsi_verify_10 *cdb;
10288
10289                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10290
10291                 *lba = scsi_4btoul(cdb->addr);
10292                 *len = scsi_2btoul(cdb->length);
10293                 break;
10294         }
10295         case VERIFY_12: {
10296                 struct scsi_verify_12 *cdb;
10297
10298                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10299
10300                 *lba = scsi_4btoul(cdb->addr);
10301                 *len = scsi_4btoul(cdb->length);
10302                 break;
10303         }
10304         case VERIFY_16: {
10305                 struct scsi_verify_16 *cdb;
10306
10307                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10308
10309                 *lba = scsi_8btou64(cdb->addr);
10310                 *len = scsi_4btoul(cdb->length);
10311                 break;
10312         }
10313         default:
10314                 return (1);
10315                 break; /* NOTREACHED */
10316         }
10317
10318         return (0);
10319 }
10320
10321 static ctl_action
10322 ctl_extent_check_lba(uint64_t lba1, uint32_t len1, uint64_t lba2, uint32_t len2)
10323 {
10324         uint64_t endlba1, endlba2;
10325
10326         endlba1 = lba1 + len1 - 1;
10327         endlba2 = lba2 + len2 - 1;
10328
10329         if ((endlba1 < lba2)
10330          || (endlba2 < lba1))
10331                 return (CTL_ACTION_PASS);
10332         else
10333                 return (CTL_ACTION_BLOCK);
10334 }
10335
10336 static ctl_action
10337 ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10338 {
10339         uint64_t lba1, lba2;
10340         uint32_t len1, len2;
10341         int retval;
10342
10343         retval = ctl_get_lba_len(io1, &lba1, &len1);
10344         if (retval != 0)
10345                 return (CTL_ACTION_ERROR);
10346
10347         retval = ctl_get_lba_len(io2, &lba2, &len2);
10348         if (retval != 0)
10349                 return (CTL_ACTION_ERROR);
10350
10351         return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10352 }
10353
10354 static ctl_action
10355 ctl_check_for_blockage(union ctl_io *pending_io, union ctl_io *ooa_io)
10356 {
10357         struct ctl_cmd_entry *pending_entry, *ooa_entry;
10358         ctl_serialize_action *serialize_row;
10359
10360         /*
10361          * The initiator attempted multiple untagged commands at the same
10362          * time.  Can't do that.
10363          */
10364         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10365          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10366          && ((pending_io->io_hdr.nexus.targ_port ==
10367               ooa_io->io_hdr.nexus.targ_port)
10368           && (pending_io->io_hdr.nexus.initid.id ==
10369               ooa_io->io_hdr.nexus.initid.id))
10370          && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10371                 return (CTL_ACTION_OVERLAP);
10372
10373         /*
10374          * The initiator attempted to send multiple tagged commands with
10375          * the same ID.  (It's fine if different initiators have the same
10376          * tag ID.)
10377          *
10378          * Even if all of those conditions are true, we don't kill the I/O
10379          * if the command ahead of us has been aborted.  We won't end up
10380          * sending it to the FETD, and it's perfectly legal to resend a
10381          * command with the same tag number as long as the previous
10382          * instance of this tag number has been aborted somehow.
10383          */
10384         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10385          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10386          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10387          && ((pending_io->io_hdr.nexus.targ_port ==
10388               ooa_io->io_hdr.nexus.targ_port)
10389           && (pending_io->io_hdr.nexus.initid.id ==
10390               ooa_io->io_hdr.nexus.initid.id))
10391          && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10392                 return (CTL_ACTION_OVERLAP_TAG);
10393
10394         /*
10395          * If we get a head of queue tag, SAM-3 says that we should
10396          * immediately execute it.
10397          *
10398          * What happens if this command would normally block for some other
10399          * reason?  e.g. a request sense with a head of queue tag
10400          * immediately after a write.  Normally that would block, but this
10401          * will result in its getting executed immediately...
10402          *
10403          * We currently return "pass" instead of "skip", so we'll end up
10404          * going through the rest of the queue to check for overlapped tags.
10405          *
10406          * XXX KDM check for other types of blockage first??
10407          */
10408         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10409                 return (CTL_ACTION_PASS);
10410
10411         /*
10412          * Ordered tags have to block until all items ahead of them
10413          * have completed.  If we get called with an ordered tag, we always
10414          * block, if something else is ahead of us in the queue.
10415          */
10416         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10417                 return (CTL_ACTION_BLOCK);
10418
10419         /*
10420          * Simple tags get blocked until all head of queue and ordered tags
10421          * ahead of them have completed.  I'm lumping untagged commands in
10422          * with simple tags here.  XXX KDM is that the right thing to do?
10423          */
10424         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10425           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10426          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10427           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10428                 return (CTL_ACTION_BLOCK);
10429
10430         pending_entry = &ctl_cmd_table[pending_io->scsiio.cdb[0]];
10431         ooa_entry = &ctl_cmd_table[ooa_io->scsiio.cdb[0]];
10432
10433         serialize_row = ctl_serialize_table[ooa_entry->seridx];
10434
10435         switch (serialize_row[pending_entry->seridx]) {
10436         case CTL_SER_BLOCK:
10437                 return (CTL_ACTION_BLOCK);
10438                 break; /* NOTREACHED */
10439         case CTL_SER_EXTENT:
10440                 return (ctl_extent_check(pending_io, ooa_io));
10441                 break; /* NOTREACHED */
10442         case CTL_SER_PASS:
10443                 return (CTL_ACTION_PASS);
10444                 break; /* NOTREACHED */
10445         case CTL_SER_SKIP:
10446                 return (CTL_ACTION_SKIP);
10447                 break;
10448         default:
10449                 panic("invalid serialization value %d",
10450                       serialize_row[pending_entry->seridx]);
10451                 break; /* NOTREACHED */
10452         }
10453
10454         return (CTL_ACTION_ERROR);
10455 }
10456
10457 /*
10458  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10459  * Assumptions:
10460  * - pending_io is generally either incoming, or on the blocked queue
10461  * - starting I/O is the I/O we want to start the check with.
10462  */
10463 static ctl_action
10464 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10465               union ctl_io *starting_io)
10466 {
10467         union ctl_io *ooa_io;
10468         ctl_action action;
10469
10470         mtx_assert(&lun->lun_lock, MA_OWNED);
10471
10472         /*
10473          * Run back along the OOA queue, starting with the current
10474          * blocked I/O and going through every I/O before it on the
10475          * queue.  If starting_io is NULL, we'll just end up returning
10476          * CTL_ACTION_PASS.
10477          */
10478         for (ooa_io = starting_io; ooa_io != NULL;
10479              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10480              ooa_links)){
10481
10482                 /*
10483                  * This routine just checks to see whether
10484                  * cur_blocked is blocked by ooa_io, which is ahead
10485                  * of it in the queue.  It doesn't queue/dequeue
10486                  * cur_blocked.
10487                  */
10488                 action = ctl_check_for_blockage(pending_io, ooa_io);
10489                 switch (action) {
10490                 case CTL_ACTION_BLOCK:
10491                 case CTL_ACTION_OVERLAP:
10492                 case CTL_ACTION_OVERLAP_TAG:
10493                 case CTL_ACTION_SKIP:
10494                 case CTL_ACTION_ERROR:
10495                         return (action);
10496                         break; /* NOTREACHED */
10497                 case CTL_ACTION_PASS:
10498                         break;
10499                 default:
10500                         panic("invalid action %d", action);
10501                         break;  /* NOTREACHED */
10502                 }
10503         }
10504
10505         return (CTL_ACTION_PASS);
10506 }
10507
10508 /*
10509  * Assumptions:
10510  * - An I/O has just completed, and has been removed from the per-LUN OOA
10511  *   queue, so some items on the blocked queue may now be unblocked.
10512  */
10513 static int
10514 ctl_check_blocked(struct ctl_lun *lun)
10515 {
10516         union ctl_io *cur_blocked, *next_blocked;
10517
10518         mtx_assert(&lun->lun_lock, MA_OWNED);
10519
10520         /*
10521          * Run forward from the head of the blocked queue, checking each
10522          * entry against the I/Os prior to it on the OOA queue to see if
10523          * there is still any blockage.
10524          *
10525          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10526          * with our removing a variable on it while it is traversing the
10527          * list.
10528          */
10529         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10530              cur_blocked != NULL; cur_blocked = next_blocked) {
10531                 union ctl_io *prev_ooa;
10532                 ctl_action action;
10533
10534                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10535                                                           blocked_links);
10536
10537                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10538                                                       ctl_ooaq, ooa_links);
10539
10540                 /*
10541                  * If cur_blocked happens to be the first item in the OOA
10542                  * queue now, prev_ooa will be NULL, and the action
10543                  * returned will just be CTL_ACTION_PASS.
10544                  */
10545                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10546
10547                 switch (action) {
10548                 case CTL_ACTION_BLOCK:
10549                         /* Nothing to do here, still blocked */
10550                         break;
10551                 case CTL_ACTION_OVERLAP:
10552                 case CTL_ACTION_OVERLAP_TAG:
10553                         /*
10554                          * This shouldn't happen!  In theory we've already
10555                          * checked this command for overlap...
10556                          */
10557                         break;
10558                 case CTL_ACTION_PASS:
10559                 case CTL_ACTION_SKIP: {
10560                         struct ctl_softc *softc;
10561                         struct ctl_cmd_entry *entry;
10562                         uint32_t initidx;
10563                         uint8_t opcode;
10564                         int isc_retval;
10565
10566                         /*
10567                          * The skip case shouldn't happen, this transaction
10568                          * should have never made it onto the blocked queue.
10569                          */
10570                         /*
10571                          * This I/O is no longer blocked, we can remove it
10572                          * from the blocked queue.  Since this is a TAILQ
10573                          * (doubly linked list), we can do O(1) removals
10574                          * from any place on the list.
10575                          */
10576                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
10577                                      blocked_links);
10578                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10579
10580                         if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
10581                                 /*
10582                                  * Need to send IO back to original side to
10583                                  * run
10584                                  */
10585                                 union ctl_ha_msg msg_info;
10586
10587                                 msg_info.hdr.original_sc =
10588                                         cur_blocked->io_hdr.original_sc;
10589                                 msg_info.hdr.serializing_sc = cur_blocked;
10590                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
10591                                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10592                                      &msg_info, sizeof(msg_info), 0)) >
10593                                      CTL_HA_STATUS_SUCCESS) {
10594                                         printf("CTL:Check Blocked error from "
10595                                                "ctl_ha_msg_send %d\n",
10596                                                isc_retval);
10597                                 }
10598                                 break;
10599                         }
10600                         opcode = cur_blocked->scsiio.cdb[0];
10601                         entry = &ctl_cmd_table[opcode];
10602                         softc = control_softc;
10603
10604                         initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
10605
10606                         /*
10607                          * Check this I/O for LUN state changes that may
10608                          * have happened while this command was blocked.
10609                          * The LUN state may have been changed by a command
10610                          * ahead of us in the queue, so we need to re-check
10611                          * for any states that can be caused by SCSI
10612                          * commands.
10613                          */
10614                         if (ctl_scsiio_lun_check(softc, lun, entry,
10615                                                  &cur_blocked->scsiio) == 0) {
10616                                 cur_blocked->io_hdr.flags |=
10617                                                       CTL_FLAG_IS_WAS_ON_RTR;
10618                                 ctl_enqueue_rtr(cur_blocked);
10619                         } else
10620                                 ctl_done(cur_blocked);
10621                         break;
10622                 }
10623                 default:
10624                         /*
10625                          * This probably shouldn't happen -- we shouldn't
10626                          * get CTL_ACTION_ERROR, or anything else.
10627                          */
10628                         break;
10629                 }
10630         }
10631
10632         return (CTL_RETVAL_COMPLETE);
10633 }
10634
10635 /*
10636  * This routine (with one exception) checks LUN flags that can be set by
10637  * commands ahead of us in the OOA queue.  These flags have to be checked
10638  * when a command initially comes in, and when we pull a command off the
10639  * blocked queue and are preparing to execute it.  The reason we have to
10640  * check these flags for commands on the blocked queue is that the LUN
10641  * state may have been changed by a command ahead of us while we're on the
10642  * blocked queue.
10643  *
10644  * Ordering is somewhat important with these checks, so please pay
10645  * careful attention to the placement of any new checks.
10646  */
10647 static int
10648 ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
10649                      struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
10650 {
10651         int retval;
10652
10653         retval = 0;
10654
10655         mtx_assert(&lun->lun_lock, MA_OWNED);
10656
10657         /*
10658          * If this shelf is a secondary shelf controller, we have to reject
10659          * any media access commands.
10660          */
10661 #if 0
10662         /* No longer needed for HA */
10663         if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
10664          && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
10665                 ctl_set_lun_standby(ctsio);
10666                 retval = 1;
10667                 goto bailout;
10668         }
10669 #endif
10670
10671         /*
10672          * Check for a reservation conflict.  If this command isn't allowed
10673          * even on reserved LUNs, and if this initiator isn't the one who
10674          * reserved us, reject the command with a reservation conflict.
10675          */
10676         if ((lun->flags & CTL_LUN_RESERVED)
10677          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
10678                 if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
10679                  || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
10680                  || (ctsio->io_hdr.nexus.targ_target.id !=
10681                      lun->rsv_nexus.targ_target.id)) {
10682                         ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10683                         ctsio->io_hdr.status = CTL_SCSI_ERROR;
10684                         retval = 1;
10685                         goto bailout;
10686                 }
10687         }
10688
10689         if ( (lun->flags & CTL_LUN_PR_RESERVED)
10690          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
10691                 uint32_t residx;
10692
10693                 residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
10694                 /*
10695                  * if we aren't registered or it's a res holder type
10696                  * reservation and this isn't the res holder then set a
10697                  * conflict.
10698                  * NOTE: Commands which might be allowed on write exclusive
10699                  * type reservations are checked in the particular command
10700                  * for a conflict. Read and SSU are the only ones.
10701                  */
10702                 if (!lun->per_res[residx].registered
10703                  || (residx != lun->pr_res_idx && lun->res_type < 4)) {
10704                         ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10705                         ctsio->io_hdr.status = CTL_SCSI_ERROR;
10706                         retval = 1;
10707                         goto bailout;
10708                 }
10709
10710         }
10711
10712         if ((lun->flags & CTL_LUN_OFFLINE)
10713          && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
10714                 ctl_set_lun_not_ready(ctsio);
10715                 retval = 1;
10716                 goto bailout;
10717         }
10718
10719         /*
10720          * If the LUN is stopped, see if this particular command is allowed
10721          * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
10722          */
10723         if ((lun->flags & CTL_LUN_STOPPED)
10724          && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
10725                 /* "Logical unit not ready, initializing cmd. required" */
10726                 ctl_set_lun_stopped(ctsio);
10727                 retval = 1;
10728                 goto bailout;
10729         }
10730
10731         if ((lun->flags & CTL_LUN_INOPERABLE)
10732          && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
10733                 /* "Medium format corrupted" */
10734                 ctl_set_medium_format_corrupted(ctsio);
10735                 retval = 1;
10736                 goto bailout;
10737         }
10738
10739 bailout:
10740         return (retval);
10741
10742 }
10743
10744 static void
10745 ctl_failover_io(union ctl_io *io, int have_lock)
10746 {
10747         ctl_set_busy(&io->scsiio);
10748         ctl_done(io);
10749 }
10750
10751 static void
10752 ctl_failover(void)
10753 {
10754         struct ctl_lun *lun;
10755         struct ctl_softc *ctl_softc;
10756         union ctl_io *next_io, *pending_io;
10757         union ctl_io *io;
10758         int lun_idx;
10759         int i;
10760
10761         ctl_softc = control_softc;
10762
10763         mtx_lock(&ctl_softc->ctl_lock);
10764         /*
10765          * Remove any cmds from the other SC from the rtr queue.  These
10766          * will obviously only be for LUNs for which we're the primary.
10767          * We can't send status or get/send data for these commands.
10768          * Since they haven't been executed yet, we can just remove them.
10769          * We'll either abort them or delete them below, depending on
10770          * which HA mode we're in.
10771          */
10772 #ifdef notyet
10773         mtx_lock(&ctl_softc->queue_lock);
10774         for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
10775              io != NULL; io = next_io) {
10776                 next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10777                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10778                         STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
10779                                       ctl_io_hdr, links);
10780         }
10781         mtx_unlock(&ctl_softc->queue_lock);
10782 #endif
10783
10784         for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
10785                 lun = ctl_softc->ctl_luns[lun_idx];
10786                 if (lun==NULL)
10787                         continue;
10788
10789                 /*
10790                  * Processor LUNs are primary on both sides.
10791                  * XXX will this always be true?
10792                  */
10793                 if (lun->be_lun->lun_type == T_PROCESSOR)
10794                         continue;
10795
10796                 if ((lun->flags & CTL_LUN_PRIMARY_SC)
10797                  && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10798                         printf("FAILOVER: primary lun %d\n", lun_idx);
10799                         /*
10800                          * Remove all commands from the other SC. First from the
10801                          * blocked queue then from the ooa queue. Once we have
10802                          * removed them. Call ctl_check_blocked to see if there
10803                          * is anything that can run.
10804                          */
10805                         for (io = (union ctl_io *)TAILQ_FIRST(
10806                              &lun->blocked_queue); io != NULL; io = next_io) {
10807
10808                                 next_io = (union ctl_io *)TAILQ_NEXT(
10809                                     &io->io_hdr, blocked_links);
10810
10811                                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10812                                         TAILQ_REMOVE(&lun->blocked_queue,
10813                                                      &io->io_hdr,blocked_links);
10814                                         io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10815                                         TAILQ_REMOVE(&lun->ooa_queue,
10816                                                      &io->io_hdr, ooa_links);
10817
10818                                         ctl_free_io(io);
10819                                 }
10820                         }
10821
10822                         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10823                              io != NULL; io = next_io) {
10824
10825                                 next_io = (union ctl_io *)TAILQ_NEXT(
10826                                     &io->io_hdr, ooa_links);
10827
10828                                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10829
10830                                         TAILQ_REMOVE(&lun->ooa_queue,
10831                                                 &io->io_hdr,
10832                                                 ooa_links);
10833
10834                                         ctl_free_io(io);
10835                                 }
10836                         }
10837                         ctl_check_blocked(lun);
10838                 } else if ((lun->flags & CTL_LUN_PRIMARY_SC)
10839                         && (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10840
10841                         printf("FAILOVER: primary lun %d\n", lun_idx);
10842                         /*
10843                          * Abort all commands from the other SC.  We can't
10844                          * send status back for them now.  These should get
10845                          * cleaned up when they are completed or come out
10846                          * for a datamove operation.
10847                          */
10848                         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10849                              io != NULL; io = next_io) {
10850                                 next_io = (union ctl_io *)TAILQ_NEXT(
10851                                         &io->io_hdr, ooa_links);
10852
10853                                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10854                                         io->io_hdr.flags |= CTL_FLAG_ABORT;
10855                         }
10856                 } else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10857                         && (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10858
10859                         printf("FAILOVER: secondary lun %d\n", lun_idx);
10860
10861                         lun->flags |= CTL_LUN_PRIMARY_SC;
10862
10863                         /*
10864                          * We send all I/O that was sent to this controller
10865                          * and redirected to the other side back with
10866                          * busy status, and have the initiator retry it.
10867                          * Figuring out how much data has been transferred,
10868                          * etc. and picking up where we left off would be
10869                          * very tricky.
10870                          *
10871                          * XXX KDM need to remove I/O from the blocked
10872                          * queue as well!
10873                          */
10874                         for (pending_io = (union ctl_io *)TAILQ_FIRST(
10875                              &lun->ooa_queue); pending_io != NULL;
10876                              pending_io = next_io) {
10877
10878                                 next_io =  (union ctl_io *)TAILQ_NEXT(
10879                                         &pending_io->io_hdr, ooa_links);
10880
10881                                 pending_io->io_hdr.flags &=
10882                                         ~CTL_FLAG_SENT_2OTHER_SC;
10883
10884                                 if (pending_io->io_hdr.flags &
10885                                     CTL_FLAG_IO_ACTIVE) {
10886                                         pending_io->io_hdr.flags |=
10887                                                 CTL_FLAG_FAILOVER;
10888                                 } else {
10889                                         ctl_set_busy(&pending_io->scsiio);
10890                                         ctl_done(pending_io);
10891                                 }
10892                         }
10893
10894                         /*
10895                          * Build Unit Attention
10896                          */
10897                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10898                                 lun->pending_sense[i].ua_pending |=
10899                                                      CTL_UA_ASYM_ACC_CHANGE;
10900                         }
10901                 } else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10902                         && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10903                         printf("FAILOVER: secondary lun %d\n", lun_idx);
10904                         /*
10905                          * if the first io on the OOA is not on the RtR queue
10906                          * add it.
10907                          */
10908                         lun->flags |= CTL_LUN_PRIMARY_SC;
10909
10910                         pending_io = (union ctl_io *)TAILQ_FIRST(
10911                             &lun->ooa_queue);
10912                         if (pending_io==NULL) {
10913                                 printf("Nothing on OOA queue\n");
10914                                 continue;
10915                         }
10916
10917                         pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
10918                         if ((pending_io->io_hdr.flags &
10919                              CTL_FLAG_IS_WAS_ON_RTR) == 0) {
10920                                 pending_io->io_hdr.flags |=
10921                                     CTL_FLAG_IS_WAS_ON_RTR;
10922                                 ctl_enqueue_rtr(pending_io);
10923                         }
10924 #if 0
10925                         else
10926                         {
10927                                 printf("Tag 0x%04x is running\n",
10928                                       pending_io->scsiio.tag_num);
10929                         }
10930 #endif
10931
10932                         next_io = (union ctl_io *)TAILQ_NEXT(
10933                             &pending_io->io_hdr, ooa_links);
10934                         for (pending_io=next_io; pending_io != NULL;
10935                              pending_io = next_io) {
10936                                 pending_io->io_hdr.flags &=
10937                                     ~CTL_FLAG_SENT_2OTHER_SC;
10938                                 next_io = (union ctl_io *)TAILQ_NEXT(
10939                                         &pending_io->io_hdr, ooa_links);
10940                                 if (pending_io->io_hdr.flags &
10941                                     CTL_FLAG_IS_WAS_ON_RTR) {
10942 #if 0
10943                                         printf("Tag 0x%04x is running\n",
10944                                                 pending_io->scsiio.tag_num);
10945 #endif
10946                                         continue;
10947                                 }
10948
10949                                 switch (ctl_check_ooa(lun, pending_io,
10950                                     (union ctl_io *)TAILQ_PREV(
10951                                     &pending_io->io_hdr, ctl_ooaq,
10952                                     ooa_links))) {
10953
10954                                 case CTL_ACTION_BLOCK:
10955                                         TAILQ_INSERT_TAIL(&lun->blocked_queue,
10956                                                           &pending_io->io_hdr,
10957                                                           blocked_links);
10958                                         pending_io->io_hdr.flags |=
10959                                             CTL_FLAG_BLOCKED;
10960                                         break;
10961                                 case CTL_ACTION_PASS:
10962                                 case CTL_ACTION_SKIP:
10963                                         pending_io->io_hdr.flags |=
10964                                             CTL_FLAG_IS_WAS_ON_RTR;
10965                                         ctl_enqueue_rtr(pending_io);
10966                                         break;
10967                                 case CTL_ACTION_OVERLAP:
10968                                         ctl_set_overlapped_cmd(
10969                                             (struct ctl_scsiio *)pending_io);
10970                                         ctl_done(pending_io);
10971                                         break;
10972                                 case CTL_ACTION_OVERLAP_TAG:
10973                                         ctl_set_overlapped_tag(
10974                                             (struct ctl_scsiio *)pending_io,
10975                                             pending_io->scsiio.tag_num & 0xff);
10976                                         ctl_done(pending_io);
10977                                         break;
10978                                 case CTL_ACTION_ERROR:
10979                                 default:
10980                                         ctl_set_internal_failure(
10981                                                 (struct ctl_scsiio *)pending_io,
10982                                                 0,  // sks_valid
10983                                                 0); //retry count
10984                                         ctl_done(pending_io);
10985                                         break;
10986                                 }
10987                         }
10988
10989                         /*
10990                          * Build Unit Attention
10991                          */
10992                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10993                                 lun->pending_sense[i].ua_pending |=
10994                                                      CTL_UA_ASYM_ACC_CHANGE;
10995                         }
10996                 } else {
10997                         panic("Unhandled HA mode failover, LUN flags = %#x, "
10998                               "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
10999                 }
11000         }
11001         ctl_pause_rtr = 0;
11002         mtx_unlock(&ctl_softc->ctl_lock);
11003 }
11004
11005 static int
11006 ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11007 {
11008         struct ctl_lun *lun;
11009         struct ctl_cmd_entry *entry;
11010         uint8_t opcode;
11011         uint32_t initidx, targ_lun;
11012         int retval;
11013
11014         retval = 0;
11015
11016         lun = NULL;
11017
11018         opcode = ctsio->cdb[0];
11019
11020         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11021         if ((targ_lun < CTL_MAX_LUNS)
11022          && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11023                 lun = ctl_softc->ctl_luns[targ_lun];
11024                 /*
11025                  * If the LUN is invalid, pretend that it doesn't exist.
11026                  * It will go away as soon as all pending I/O has been
11027                  * completed.
11028                  */
11029                 if (lun->flags & CTL_LUN_DISABLED) {
11030                         lun = NULL;
11031                 } else {
11032                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11033                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11034                                 lun->be_lun;
11035                         if (lun->be_lun->lun_type == T_PROCESSOR) {
11036                                 ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11037                         }
11038                 }
11039         } else {
11040                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11041                 ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11042         }
11043
11044         entry = &ctl_cmd_table[opcode];
11045
11046         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11047         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11048
11049         /*
11050          * Check to see whether we can send this command to LUNs that don't
11051          * exist.  This should pretty much only be the case for inquiry
11052          * and request sense.  Further checks, below, really require having
11053          * a LUN, so we can't really check the command anymore.  Just put
11054          * it on the rtr queue.
11055          */
11056         if (lun == NULL) {
11057                 if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11058                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11059                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11060                         return (retval);
11061                 }
11062
11063                 ctl_set_unsupported_lun(ctsio);
11064                 ctl_done((union ctl_io *)ctsio);
11065                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11066                 return (retval);
11067         } else {
11068                 mtx_lock(&lun->lun_lock);
11069
11070                 /*
11071                  * Every I/O goes into the OOA queue for a particular LUN, and
11072                  * stays there until completion.
11073                  */
11074                 TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
11075
11076                 /*
11077                  * Make sure we support this particular command on this LUN.
11078                  * e.g., we don't support writes to the control LUN.
11079                  */
11080                 switch (lun->be_lun->lun_type) {
11081                 case T_PROCESSOR:
11082                         if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11083                          && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
11084                               == 0)) {
11085                                 mtx_unlock(&lun->lun_lock);
11086                                 ctl_set_invalid_opcode(ctsio);
11087                                 ctl_done((union ctl_io *)ctsio);
11088                                 return (retval);
11089                         }
11090                         break;
11091                 case T_DIRECT:
11092                         if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0)
11093                          && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
11094                               == 0)){
11095                                 mtx_unlock(&lun->lun_lock);
11096                                 ctl_set_invalid_opcode(ctsio);
11097                                 ctl_done((union ctl_io *)ctsio);
11098                                 return (retval);
11099                         }
11100                         break;
11101                 default:
11102                         mtx_unlock(&lun->lun_lock);
11103                         panic("Unsupported CTL LUN type %d\n",
11104                               lun->be_lun->lun_type);
11105                 }
11106         }
11107
11108         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11109
11110         /*
11111          * If we've got a request sense, it'll clear the contingent
11112          * allegiance condition.  Otherwise, if we have a CA condition for
11113          * this initiator, clear it, because it sent down a command other
11114          * than request sense.
11115          */
11116         if ((opcode != REQUEST_SENSE)
11117          && (ctl_is_set(lun->have_ca, initidx)))
11118                 ctl_clear_mask(lun->have_ca, initidx);
11119
11120         /*
11121          * If the command has this flag set, it handles its own unit
11122          * attention reporting, we shouldn't do anything.  Otherwise we
11123          * check for any pending unit attentions, and send them back to the
11124          * initiator.  We only do this when a command initially comes in,
11125          * not when we pull it off the blocked queue.
11126          *
11127          * According to SAM-3, section 5.3.2, the order that things get
11128          * presented back to the host is basically unit attentions caused
11129          * by some sort of reset event, busy status, reservation conflicts
11130          * or task set full, and finally any other status.
11131          *
11132          * One issue here is that some of the unit attentions we report
11133          * don't fall into the "reset" category (e.g. "reported luns data
11134          * has changed").  So reporting it here, before the reservation
11135          * check, may be technically wrong.  I guess the only thing to do
11136          * would be to check for and report the reset events here, and then
11137          * check for the other unit attention types after we check for a
11138          * reservation conflict.
11139          *
11140          * XXX KDM need to fix this
11141          */
11142         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11143                 ctl_ua_type ua_type;
11144
11145                 ua_type = lun->pending_sense[initidx].ua_pending;
11146                 if (ua_type != CTL_UA_NONE) {
11147                         scsi_sense_data_type sense_format;
11148
11149                         if (lun != NULL)
11150                                 sense_format = (lun->flags &
11151                                     CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11152                                     SSD_TYPE_FIXED;
11153                         else
11154                                 sense_format = SSD_TYPE_FIXED;
11155
11156                         ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
11157                                                sense_format);
11158                         if (ua_type != CTL_UA_NONE) {
11159                                 ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11160                                 ctsio->io_hdr.status = CTL_SCSI_ERROR |
11161                                                        CTL_AUTOSENSE;
11162                                 ctsio->sense_len = SSD_FULL_SIZE;
11163                                 lun->pending_sense[initidx].ua_pending &=
11164                                         ~ua_type;
11165                                 mtx_unlock(&lun->lun_lock);
11166                                 ctl_done((union ctl_io *)ctsio);
11167                                 return (retval);
11168                         }
11169                 }
11170         }
11171
11172
11173         if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11174                 mtx_unlock(&lun->lun_lock);
11175                 ctl_done((union ctl_io *)ctsio);
11176                 return (retval);
11177         }
11178
11179         /*
11180          * XXX CHD this is where we want to send IO to other side if
11181          * this LUN is secondary on this SC. We will need to make a copy
11182          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11183          * the copy we send as FROM_OTHER.
11184          * We also need to stuff the address of the original IO so we can
11185          * find it easily. Something similar will need be done on the other
11186          * side so when we are done we can find the copy.
11187          */
11188         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11189                 union ctl_ha_msg msg_info;
11190                 int isc_retval;
11191
11192                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11193
11194                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11195                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11196 #if 0
11197                 printf("1. ctsio %p\n", ctsio);
11198 #endif
11199                 msg_info.hdr.serializing_sc = NULL;
11200                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11201                 msg_info.scsi.tag_num = ctsio->tag_num;
11202                 msg_info.scsi.tag_type = ctsio->tag_type;
11203                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11204
11205                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11206
11207                 if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11208                     (void *)&msg_info, sizeof(msg_info), 0)) >
11209                     CTL_HA_STATUS_SUCCESS) {
11210                         printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11211                                isc_retval);
11212                         printf("CTL:opcode is %x\n",opcode);
11213                 } else {
11214 #if 0
11215                         printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11216 #endif
11217                 }
11218
11219                 /*
11220                  * XXX KDM this I/O is off the incoming queue, but hasn't
11221                  * been inserted on any other queue.  We may need to come
11222                  * up with a holding queue while we wait for serialization
11223                  * so that we have an idea of what we're waiting for from
11224                  * the other side.
11225                  */
11226                 mtx_unlock(&lun->lun_lock);
11227                 return (retval);
11228         }
11229
11230         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11231                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11232                               ctl_ooaq, ooa_links))) {
11233         case CTL_ACTION_BLOCK:
11234                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11235                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11236                                   blocked_links);
11237                 mtx_unlock(&lun->lun_lock);
11238                 return (retval);
11239         case CTL_ACTION_PASS:
11240         case CTL_ACTION_SKIP:
11241                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11242                 mtx_unlock(&lun->lun_lock);
11243                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11244                 break;
11245         case CTL_ACTION_OVERLAP:
11246                 mtx_unlock(&lun->lun_lock);
11247                 ctl_set_overlapped_cmd(ctsio);
11248                 ctl_done((union ctl_io *)ctsio);
11249                 break;
11250         case CTL_ACTION_OVERLAP_TAG:
11251                 mtx_unlock(&lun->lun_lock);
11252                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11253                 ctl_done((union ctl_io *)ctsio);
11254                 break;
11255         case CTL_ACTION_ERROR:
11256         default:
11257                 mtx_unlock(&lun->lun_lock);
11258                 ctl_set_internal_failure(ctsio,
11259                                          /*sks_valid*/ 0,
11260                                          /*retry_count*/ 0);
11261                 ctl_done((union ctl_io *)ctsio);
11262                 break;
11263         }
11264         return (retval);
11265 }
11266
11267 static int
11268 ctl_scsiio(struct ctl_scsiio *ctsio)
11269 {
11270         int retval;
11271         struct ctl_cmd_entry *entry;
11272
11273         retval = CTL_RETVAL_COMPLETE;
11274
11275         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11276
11277         entry = &ctl_cmd_table[ctsio->cdb[0]];
11278
11279         /*
11280          * If this I/O has been aborted, just send it straight to
11281          * ctl_done() without executing it.
11282          */
11283         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11284                 ctl_done((union ctl_io *)ctsio);
11285                 goto bailout;
11286         }
11287
11288         /*
11289          * All the checks should have been handled by ctl_scsiio_precheck().
11290          * We should be clear now to just execute the I/O.
11291          */
11292         retval = entry->execute(ctsio);
11293
11294 bailout:
11295         return (retval);
11296 }
11297
11298 /*
11299  * Since we only implement one target right now, a bus reset simply resets
11300  * our single target.
11301  */
11302 static int
11303 ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11304 {
11305         return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11306 }
11307
11308 static int
11309 ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11310                  ctl_ua_type ua_type)
11311 {
11312         struct ctl_lun *lun;
11313         int retval;
11314
11315         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11316                 union ctl_ha_msg msg_info;
11317
11318                 io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11319                 msg_info.hdr.nexus = io->io_hdr.nexus;
11320                 if (ua_type==CTL_UA_TARG_RESET)
11321                         msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11322                 else
11323                         msg_info.task.task_action = CTL_TASK_BUS_RESET;
11324                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11325                 msg_info.hdr.original_sc = NULL;
11326                 msg_info.hdr.serializing_sc = NULL;
11327                 if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11328                     (void *)&msg_info, sizeof(msg_info), 0)) {
11329                 }
11330         }
11331         retval = 0;
11332
11333         mtx_lock(&ctl_softc->ctl_lock);
11334         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11335                 retval += ctl_lun_reset(lun, io, ua_type);
11336         mtx_unlock(&ctl_softc->ctl_lock);
11337
11338         return (retval);
11339 }
11340
11341 /*
11342  * The LUN should always be set.  The I/O is optional, and is used to
11343  * distinguish between I/Os sent by this initiator, and by other
11344  * initiators.  We set unit attention for initiators other than this one.
11345  * SAM-3 is vague on this point.  It does say that a unit attention should
11346  * be established for other initiators when a LUN is reset (see section
11347  * 5.7.3), but it doesn't specifically say that the unit attention should
11348  * be established for this particular initiator when a LUN is reset.  Here
11349  * is the relevant text, from SAM-3 rev 8:
11350  *
11351  * 5.7.2 When a SCSI initiator port aborts its own tasks
11352  *
11353  * When a SCSI initiator port causes its own task(s) to be aborted, no
11354  * notification that the task(s) have been aborted shall be returned to
11355  * the SCSI initiator port other than the completion response for the
11356  * command or task management function action that caused the task(s) to
11357  * be aborted and notification(s) associated with related effects of the
11358  * action (e.g., a reset unit attention condition).
11359  *
11360  * XXX KDM for now, we're setting unit attention for all initiators.
11361  */
11362 static int
11363 ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11364 {
11365         union ctl_io *xio;
11366 #if 0
11367         uint32_t initindex;
11368 #endif
11369         int i;
11370
11371         mtx_lock(&lun->lun_lock);
11372         /*
11373          * Run through the OOA queue and abort each I/O.
11374          */
11375 #if 0
11376         TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11377 #endif
11378         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11379              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11380                 xio->io_hdr.flags |= CTL_FLAG_ABORT;
11381         }
11382
11383         /*
11384          * This version sets unit attention for every
11385          */
11386 #if 0
11387         initindex = ctl_get_initindex(&io->io_hdr.nexus);
11388         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11389                 if (initindex == i)
11390                         continue;
11391                 lun->pending_sense[i].ua_pending |= ua_type;
11392         }
11393 #endif
11394
11395         /*
11396          * A reset (any kind, really) clears reservations established with
11397          * RESERVE/RELEASE.  It does not clear reservations established
11398          * with PERSISTENT RESERVE OUT, but we don't support that at the
11399          * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11400          * reservations made with the RESERVE/RELEASE commands, because
11401          * those commands are obsolete in SPC-3.
11402          */
11403         lun->flags &= ~CTL_LUN_RESERVED;
11404
11405         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11406                 ctl_clear_mask(lun->have_ca, i);
11407                 lun->pending_sense[i].ua_pending |= ua_type;
11408         }
11409         mtx_unlock(&lun->lun_lock);
11410
11411         return (0);
11412 }
11413
11414 static int
11415 ctl_abort_task(union ctl_io *io)
11416 {
11417         union ctl_io *xio;
11418         struct ctl_lun *lun;
11419         struct ctl_softc *ctl_softc;
11420 #if 0
11421         struct sbuf sb;
11422         char printbuf[128];
11423 #endif
11424         int found;
11425         uint32_t targ_lun;
11426
11427         ctl_softc = control_softc;
11428         found = 0;
11429
11430         /*
11431          * Look up the LUN.
11432          */
11433         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11434         mtx_lock(&ctl_softc->ctl_lock);
11435         if ((targ_lun < CTL_MAX_LUNS)
11436          && (ctl_softc->ctl_luns[targ_lun] != NULL))
11437                 lun = ctl_softc->ctl_luns[targ_lun];
11438         else {
11439                 mtx_unlock(&ctl_softc->ctl_lock);
11440                 goto bailout;
11441         }
11442
11443 #if 0
11444         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11445                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11446 #endif
11447
11448         mtx_lock(&lun->lun_lock);
11449         mtx_unlock(&ctl_softc->ctl_lock);
11450         /*
11451          * Run through the OOA queue and attempt to find the given I/O.
11452          * The target port, initiator ID, tag type and tag number have to
11453          * match the values that we got from the initiator.  If we have an
11454          * untagged command to abort, simply abort the first untagged command
11455          * we come to.  We only allow one untagged command at a time of course.
11456          */
11457 #if 0
11458         TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11459 #endif
11460         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11461              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11462 #if 0
11463                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11464
11465                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11466                             lun->lun, xio->scsiio.tag_num,
11467                             xio->scsiio.tag_type,
11468                             (xio->io_hdr.blocked_links.tqe_prev
11469                             == NULL) ? "" : " BLOCKED",
11470                             (xio->io_hdr.flags &
11471                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11472                             (xio->io_hdr.flags &
11473                             CTL_FLAG_ABORT) ? " ABORT" : "",
11474                             (xio->io_hdr.flags &
11475                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11476                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11477                 sbuf_finish(&sb);
11478                 printf("%s\n", sbuf_data(&sb));
11479 #endif
11480
11481                 if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
11482                  && (xio->io_hdr.nexus.initid.id ==
11483                      io->io_hdr.nexus.initid.id)) {
11484                         /*
11485                          * If the abort says that the task is untagged, the
11486                          * task in the queue must be untagged.  Otherwise,
11487                          * we just check to see whether the tag numbers
11488                          * match.  This is because the QLogic firmware
11489                          * doesn't pass back the tag type in an abort
11490                          * request.
11491                          */
11492 #if 0
11493                         if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
11494                           && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
11495                          || (xio->scsiio.tag_num == io->taskio.tag_num)) {
11496 #endif
11497                         /*
11498                          * XXX KDM we've got problems with FC, because it
11499                          * doesn't send down a tag type with aborts.  So we
11500                          * can only really go by the tag number...
11501                          * This may cause problems with parallel SCSI.
11502                          * Need to figure that out!!
11503                          */
11504                         if (xio->scsiio.tag_num == io->taskio.tag_num) {
11505                                 xio->io_hdr.flags |= CTL_FLAG_ABORT;
11506                                 found = 1;
11507                                 if ((io->io_hdr.flags &
11508                                      CTL_FLAG_FROM_OTHER_SC) == 0 &&
11509                                     !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11510                                         union ctl_ha_msg msg_info;
11511
11512                                         io->io_hdr.flags |=
11513                                                         CTL_FLAG_SENT_2OTHER_SC;
11514                                         msg_info.hdr.nexus = io->io_hdr.nexus;
11515                                         msg_info.task.task_action =
11516                                                 CTL_TASK_ABORT_TASK;
11517                                         msg_info.task.tag_num =
11518                                                 io->taskio.tag_num;
11519                                         msg_info.task.tag_type =
11520                                                 io->taskio.tag_type;
11521                                         msg_info.hdr.msg_type =
11522                                                 CTL_MSG_MANAGE_TASKS;
11523                                         msg_info.hdr.original_sc = NULL;
11524                                         msg_info.hdr.serializing_sc = NULL;
11525 #if 0
11526                                         printf("Sent Abort to other side\n");
11527 #endif
11528                                         if (CTL_HA_STATUS_SUCCESS !=
11529                                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11530                                                 (void *)&msg_info,
11531                                                 sizeof(msg_info), 0)) {
11532                                         }
11533                                 }
11534 #if 0
11535                                 printf("ctl_abort_task: found I/O to abort\n");
11536 #endif
11537                                 break;
11538                         }
11539                 }
11540         }
11541         mtx_unlock(&lun->lun_lock);
11542
11543 bailout:
11544
11545         if (found == 0) {
11546                 /*
11547                  * This isn't really an error.  It's entirely possible for
11548                  * the abort and command completion to cross on the wire.
11549                  * This is more of an informative/diagnostic error.
11550                  */
11551 #if 0
11552                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
11553                        "%d:%d:%d:%d tag %d type %d\n",
11554                        io->io_hdr.nexus.initid.id,
11555                        io->io_hdr.nexus.targ_port,
11556                        io->io_hdr.nexus.targ_target.id,
11557                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
11558                        io->taskio.tag_type);
11559 #endif
11560                 return (1);
11561         } else
11562                 return (0);
11563 }
11564
11565 static void
11566 ctl_run_task(union ctl_io *io)
11567 {
11568         struct ctl_softc *ctl_softc;
11569         int retval;
11570         const char *task_desc;
11571
11572         CTL_DEBUG_PRINT(("ctl_run_task\n"));
11573
11574         ctl_softc = control_softc;
11575         retval = 0;
11576
11577         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
11578             ("ctl_run_task: Unextected io_type %d\n",
11579              io->io_hdr.io_type));
11580
11581         task_desc = ctl_scsi_task_string(&io->taskio);
11582         if (task_desc != NULL) {
11583 #ifdef NEEDTOPORT
11584                 csevent_log(CSC_CTL | CSC_SHELF_SW |
11585                             CTL_TASK_REPORT,
11586                             csevent_LogType_Trace,
11587                             csevent_Severity_Information,
11588                             csevent_AlertLevel_Green,
11589                             csevent_FRU_Firmware,
11590                             csevent_FRU_Unknown,
11591                             "CTL: received task: %s",task_desc);
11592 #endif
11593         } else {
11594 #ifdef NEEDTOPORT
11595                 csevent_log(CSC_CTL | CSC_SHELF_SW |
11596                             CTL_TASK_REPORT,
11597                             csevent_LogType_Trace,
11598                             csevent_Severity_Information,
11599                             csevent_AlertLevel_Green,
11600                             csevent_FRU_Firmware,
11601                             csevent_FRU_Unknown,
11602                             "CTL: received unknown task "
11603                             "type: %d (%#x)",
11604                             io->taskio.task_action,
11605                             io->taskio.task_action);
11606 #endif
11607         }
11608         switch (io->taskio.task_action) {
11609         case CTL_TASK_ABORT_TASK:
11610                 retval = ctl_abort_task(io);
11611                 break;
11612         case CTL_TASK_ABORT_TASK_SET:
11613                 break;
11614         case CTL_TASK_CLEAR_ACA:
11615                 break;
11616         case CTL_TASK_CLEAR_TASK_SET:
11617                 break;
11618         case CTL_TASK_LUN_RESET: {
11619                 struct ctl_lun *lun;
11620                 uint32_t targ_lun;
11621                 int retval;
11622
11623                 targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11624                 mtx_lock(&ctl_softc->ctl_lock);
11625                 if ((targ_lun < CTL_MAX_LUNS)
11626                  && (ctl_softc->ctl_luns[targ_lun] != NULL))
11627                         lun = ctl_softc->ctl_luns[targ_lun];
11628                 else {
11629                         mtx_unlock(&ctl_softc->ctl_lock);
11630                         retval = 1;
11631                         break;
11632                 }
11633
11634                 if (!(io->io_hdr.flags &
11635                     CTL_FLAG_FROM_OTHER_SC)) {
11636                         union ctl_ha_msg msg_info;
11637
11638                         io->io_hdr.flags |=
11639                                 CTL_FLAG_SENT_2OTHER_SC;
11640                         msg_info.hdr.msg_type =
11641                                 CTL_MSG_MANAGE_TASKS;
11642                         msg_info.hdr.nexus = io->io_hdr.nexus;
11643                         msg_info.task.task_action =
11644                                 CTL_TASK_LUN_RESET;
11645                         msg_info.hdr.original_sc = NULL;
11646                         msg_info.hdr.serializing_sc = NULL;
11647                         if (CTL_HA_STATUS_SUCCESS !=
11648                             ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11649                             (void *)&msg_info,
11650                             sizeof(msg_info), 0)) {
11651                         }
11652                 }
11653
11654                 retval = ctl_lun_reset(lun, io,
11655                                        CTL_UA_LUN_RESET);
11656                 mtx_unlock(&ctl_softc->ctl_lock);
11657                 break;
11658         }
11659         case CTL_TASK_TARGET_RESET:
11660                 retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
11661                 break;
11662         case CTL_TASK_BUS_RESET:
11663                 retval = ctl_bus_reset(ctl_softc, io);
11664                 break;
11665         case CTL_TASK_PORT_LOGIN:
11666                 break;
11667         case CTL_TASK_PORT_LOGOUT:
11668                 break;
11669         default:
11670                 printf("ctl_run_task: got unknown task management event %d\n",
11671                        io->taskio.task_action);
11672                 break;
11673         }
11674         if (retval == 0)
11675                 io->io_hdr.status = CTL_SUCCESS;
11676         else
11677                 io->io_hdr.status = CTL_ERROR;
11678
11679         /*
11680          * This will queue this I/O to the done queue, but the
11681          * work thread won't be able to process it until we
11682          * return and the lock is released.
11683          */
11684         ctl_done(io);
11685 }
11686
11687 /*
11688  * For HA operation.  Handle commands that come in from the other
11689  * controller.
11690  */
11691 static void
11692 ctl_handle_isc(union ctl_io *io)
11693 {
11694         int free_io;
11695         struct ctl_lun *lun;
11696         struct ctl_softc *ctl_softc;
11697         uint32_t targ_lun;
11698
11699         ctl_softc = control_softc;
11700
11701         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11702         lun = ctl_softc->ctl_luns[targ_lun];
11703
11704         switch (io->io_hdr.msg_type) {
11705         case CTL_MSG_SERIALIZE:
11706                 free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
11707                 break;
11708         case CTL_MSG_R2R: {
11709                 uint8_t opcode;
11710                 struct ctl_cmd_entry *entry;
11711
11712                 /*
11713                  * This is only used in SER_ONLY mode.
11714                  */
11715                 free_io = 0;
11716                 opcode = io->scsiio.cdb[0];
11717                 entry = &ctl_cmd_table[opcode];
11718                 mtx_lock(&lun->lun_lock);
11719                 if (ctl_scsiio_lun_check(ctl_softc, lun,
11720                     entry, (struct ctl_scsiio *)io) != 0) {
11721                         mtx_unlock(&lun->lun_lock);
11722                         ctl_done(io);
11723                         break;
11724                 }
11725                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11726                 mtx_unlock(&lun->lun_lock);
11727                 ctl_enqueue_rtr(io);
11728                 break;
11729         }
11730         case CTL_MSG_FINISH_IO:
11731                 if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
11732                         free_io = 0;
11733                         ctl_done(io);
11734                 } else {
11735                         free_io = 1;
11736                         mtx_lock(&lun->lun_lock);
11737                         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
11738                                      ooa_links);
11739                         ctl_check_blocked(lun);
11740                         mtx_unlock(&lun->lun_lock);
11741                 }
11742                 break;
11743         case CTL_MSG_PERS_ACTION:
11744                 ctl_hndl_per_res_out_on_other_sc(
11745                         (union ctl_ha_msg *)&io->presio.pr_msg);
11746                 free_io = 1;
11747                 break;
11748         case CTL_MSG_BAD_JUJU:
11749                 free_io = 0;
11750                 ctl_done(io);
11751                 break;
11752         case CTL_MSG_DATAMOVE:
11753                 /* Only used in XFER mode */
11754                 free_io = 0;
11755                 ctl_datamove_remote(io);
11756                 break;
11757         case CTL_MSG_DATAMOVE_DONE:
11758                 /* Only used in XFER mode */
11759                 free_io = 0;
11760                 io->scsiio.be_move_done(io);
11761                 break;
11762         default:
11763                 free_io = 1;
11764                 printf("%s: Invalid message type %d\n",
11765                        __func__, io->io_hdr.msg_type);
11766                 break;
11767         }
11768         if (free_io)
11769                 ctl_free_io(io);
11770
11771 }
11772
11773
11774 /*
11775  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
11776  * there is no match.
11777  */
11778 static ctl_lun_error_pattern
11779 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
11780 {
11781         struct ctl_cmd_entry *entry;
11782         ctl_lun_error_pattern filtered_pattern, pattern;
11783         uint8_t opcode;
11784
11785         pattern = desc->error_pattern;
11786
11787         /*
11788          * XXX KDM we need more data passed into this function to match a
11789          * custom pattern, and we actually need to implement custom pattern
11790          * matching.
11791          */
11792         if (pattern & CTL_LUN_PAT_CMD)
11793                 return (CTL_LUN_PAT_CMD);
11794
11795         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
11796                 return (CTL_LUN_PAT_ANY);
11797
11798         opcode = ctsio->cdb[0];
11799         entry = &ctl_cmd_table[opcode];
11800
11801         filtered_pattern = entry->pattern & pattern;
11802
11803         /*
11804          * If the user requested specific flags in the pattern (e.g.
11805          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
11806          * flags.
11807          *
11808          * If the user did not specify any flags, it doesn't matter whether
11809          * or not the command supports the flags.
11810          */
11811         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
11812              (pattern & ~CTL_LUN_PAT_MASK))
11813                 return (CTL_LUN_PAT_NONE);
11814
11815         /*
11816          * If the user asked for a range check, see if the requested LBA
11817          * range overlaps with this command's LBA range.
11818          */
11819         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
11820                 uint64_t lba1;
11821                 uint32_t len1;
11822                 ctl_action action;
11823                 int retval;
11824
11825                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
11826                 if (retval != 0)
11827                         return (CTL_LUN_PAT_NONE);
11828
11829                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
11830                                               desc->lba_range.len);
11831                 /*
11832                  * A "pass" means that the LBA ranges don't overlap, so
11833                  * this doesn't match the user's range criteria.
11834                  */
11835                 if (action == CTL_ACTION_PASS)
11836                         return (CTL_LUN_PAT_NONE);
11837         }
11838
11839         return (filtered_pattern);
11840 }
11841
11842 static void
11843 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
11844 {
11845         struct ctl_error_desc *desc, *desc2;
11846
11847         mtx_assert(&lun->lun_lock, MA_OWNED);
11848
11849         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
11850                 ctl_lun_error_pattern pattern;
11851                 /*
11852                  * Check to see whether this particular command matches
11853                  * the pattern in the descriptor.
11854                  */
11855                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
11856                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
11857                         continue;
11858
11859                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
11860                 case CTL_LUN_INJ_ABORTED:
11861                         ctl_set_aborted(&io->scsiio);
11862                         break;
11863                 case CTL_LUN_INJ_MEDIUM_ERR:
11864                         ctl_set_medium_error(&io->scsiio);
11865                         break;
11866                 case CTL_LUN_INJ_UA:
11867                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
11868                          * OCCURRED */
11869                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
11870                         break;
11871                 case CTL_LUN_INJ_CUSTOM:
11872                         /*
11873                          * We're assuming the user knows what he is doing.
11874                          * Just copy the sense information without doing
11875                          * checks.
11876                          */
11877                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
11878                               ctl_min(sizeof(desc->custom_sense),
11879                                       sizeof(io->scsiio.sense_data)));
11880                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
11881                         io->scsiio.sense_len = SSD_FULL_SIZE;
11882                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11883                         break;
11884                 case CTL_LUN_INJ_NONE:
11885                 default:
11886                         /*
11887                          * If this is an error injection type we don't know
11888                          * about, clear the continuous flag (if it is set)
11889                          * so it will get deleted below.
11890                          */
11891                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
11892                         break;
11893                 }
11894                 /*
11895                  * By default, each error injection action is a one-shot
11896                  */
11897                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
11898                         continue;
11899
11900                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
11901
11902                 free(desc, M_CTL);
11903         }
11904 }
11905
11906 #ifdef CTL_IO_DELAY
11907 static void
11908 ctl_datamove_timer_wakeup(void *arg)
11909 {
11910         union ctl_io *io;
11911
11912         io = (union ctl_io *)arg;
11913
11914         ctl_datamove(io);
11915 }
11916 #endif /* CTL_IO_DELAY */
11917
11918 void
11919 ctl_datamove(union ctl_io *io)
11920 {
11921         void (*fe_datamove)(union ctl_io *io);
11922
11923         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
11924
11925         CTL_DEBUG_PRINT(("ctl_datamove\n"));
11926
11927 #ifdef CTL_TIME_IO
11928         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
11929                 char str[256];
11930                 char path_str[64];
11931                 struct sbuf sb;
11932
11933                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
11934                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
11935
11936                 sbuf_cat(&sb, path_str);
11937                 switch (io->io_hdr.io_type) {
11938                 case CTL_IO_SCSI:
11939                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
11940                         sbuf_printf(&sb, "\n");
11941                         sbuf_cat(&sb, path_str);
11942                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
11943                                     io->scsiio.tag_num, io->scsiio.tag_type);
11944                         break;
11945                 case CTL_IO_TASK:
11946                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
11947                                     "Tag Type: %d\n", io->taskio.task_action,
11948                                     io->taskio.tag_num, io->taskio.tag_type);
11949                         break;
11950                 default:
11951                         printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
11952                         panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
11953                         break;
11954                 }
11955                 sbuf_cat(&sb, path_str);
11956                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
11957                             (intmax_t)time_uptime - io->io_hdr.start_time);
11958                 sbuf_finish(&sb);
11959                 printf("%s", sbuf_data(&sb));
11960         }
11961 #endif /* CTL_TIME_IO */
11962
11963 #ifdef CTL_IO_DELAY
11964         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
11965                 struct ctl_lun *lun;
11966
11967                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
11968
11969                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
11970         } else {
11971                 struct ctl_lun *lun;
11972
11973                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
11974                 if ((lun != NULL)
11975                  && (lun->delay_info.datamove_delay > 0)) {
11976                         struct callout *callout;
11977
11978                         callout = (struct callout *)&io->io_hdr.timer_bytes;
11979                         callout_init(callout, /*mpsafe*/ 1);
11980                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
11981                         callout_reset(callout,
11982                                       lun->delay_info.datamove_delay * hz,
11983                                       ctl_datamove_timer_wakeup, io);
11984                         if (lun->delay_info.datamove_type ==
11985                             CTL_DELAY_TYPE_ONESHOT)
11986                                 lun->delay_info.datamove_delay = 0;
11987                         return;
11988                 }
11989         }
11990 #endif
11991
11992         /*
11993          * This command has been aborted.  Set the port status, so we fail
11994          * the data move.
11995          */
11996         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
11997                 printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
11998                        io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
11999                        io->io_hdr.nexus.targ_port,
12000                        (uintmax_t)io->io_hdr.nexus.targ_target.id,
12001                        io->io_hdr.nexus.targ_lun);
12002                 io->io_hdr.status = CTL_CMD_ABORTED;
12003                 io->io_hdr.port_status = 31337;
12004                 /*
12005                  * Note that the backend, in this case, will get the
12006                  * callback in its context.  In other cases it may get
12007                  * called in the frontend's interrupt thread context.
12008                  */
12009                 io->scsiio.be_move_done(io);
12010                 return;
12011         }
12012
12013         /*
12014          * If we're in XFER mode and this I/O is from the other shelf
12015          * controller, we need to send the DMA to the other side to
12016          * actually transfer the data to/from the host.  In serialize only
12017          * mode the transfer happens below CTL and ctl_datamove() is only
12018          * called on the machine that originally received the I/O.
12019          */
12020         if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12021          && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12022                 union ctl_ha_msg msg;
12023                 uint32_t sg_entries_sent;
12024                 int do_sg_copy;
12025                 int i;
12026
12027                 memset(&msg, 0, sizeof(msg));
12028                 msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12029                 msg.hdr.original_sc = io->io_hdr.original_sc;
12030                 msg.hdr.serializing_sc = io;
12031                 msg.hdr.nexus = io->io_hdr.nexus;
12032                 msg.dt.flags = io->io_hdr.flags;
12033                 /*
12034                  * We convert everything into a S/G list here.  We can't
12035                  * pass by reference, only by value between controllers.
12036                  * So we can't pass a pointer to the S/G list, only as many
12037                  * S/G entries as we can fit in here.  If it's possible for
12038                  * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12039                  * then we need to break this up into multiple transfers.
12040                  */
12041                 if (io->scsiio.kern_sg_entries == 0) {
12042                         msg.dt.kern_sg_entries = 1;
12043                         /*
12044                          * If this is in cached memory, flush the cache
12045                          * before we send the DMA request to the other
12046                          * controller.  We want to do this in either the
12047                          * read or the write case.  The read case is
12048                          * straightforward.  In the write case, we want to
12049                          * make sure nothing is in the local cache that
12050                          * could overwrite the DMAed data.
12051                          */
12052                         if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12053                                 /*
12054                                  * XXX KDM use bus_dmamap_sync() here.
12055                                  */
12056                         }
12057
12058                         /*
12059                          * Convert to a physical address if this is a
12060                          * virtual address.
12061                          */
12062                         if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12063                                 msg.dt.sg_list[0].addr =
12064                                         io->scsiio.kern_data_ptr;
12065                         } else {
12066                                 /*
12067                                  * XXX KDM use busdma here!
12068                                  */
12069 #if 0
12070                                 msg.dt.sg_list[0].addr = (void *)
12071                                         vtophys(io->scsiio.kern_data_ptr);
12072 #endif
12073                         }
12074
12075                         msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12076                         do_sg_copy = 0;
12077                 } else {
12078                         struct ctl_sg_entry *sgl;
12079
12080                         do_sg_copy = 1;
12081                         msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12082                         sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12083                         if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12084                                 /*
12085                                  * XXX KDM use bus_dmamap_sync() here.
12086                                  */
12087                         }
12088                 }
12089
12090                 msg.dt.kern_data_len = io->scsiio.kern_data_len;
12091                 msg.dt.kern_total_len = io->scsiio.kern_total_len;
12092                 msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12093                 msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12094                 msg.dt.sg_sequence = 0;
12095
12096                 /*
12097                  * Loop until we've sent all of the S/G entries.  On the
12098                  * other end, we'll recompose these S/G entries into one
12099                  * contiguous list before passing it to the
12100                  */
12101                 for (sg_entries_sent = 0; sg_entries_sent <
12102                      msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12103                         msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12104                                 sizeof(msg.dt.sg_list[0])),
12105                                 msg.dt.kern_sg_entries - sg_entries_sent);
12106
12107                         if (do_sg_copy != 0) {
12108                                 struct ctl_sg_entry *sgl;
12109                                 int j;
12110
12111                                 sgl = (struct ctl_sg_entry *)
12112                                         io->scsiio.kern_data_ptr;
12113                                 /*
12114                                  * If this is in cached memory, flush the cache
12115                                  * before we send the DMA request to the other
12116                                  * controller.  We want to do this in either
12117                                  * the * read or the write case.  The read
12118                                  * case is straightforward.  In the write
12119                                  * case, we want to make sure nothing is
12120                                  * in the local cache that could overwrite
12121                                  * the DMAed data.
12122                                  */
12123
12124                                 for (i = sg_entries_sent, j = 0;
12125                                      i < msg.dt.cur_sg_entries; i++, j++) {
12126                                         if ((io->io_hdr.flags &
12127                                              CTL_FLAG_NO_DATASYNC) == 0) {
12128                                                 /*
12129                                                  * XXX KDM use bus_dmamap_sync()
12130                                                  */
12131                                         }
12132                                         if ((io->io_hdr.flags &
12133                                              CTL_FLAG_BUS_ADDR) == 0) {
12134                                                 /*
12135                                                  * XXX KDM use busdma.
12136                                                  */
12137 #if 0
12138                                                 msg.dt.sg_list[j].addr =(void *)
12139                                                        vtophys(sgl[i].addr);
12140 #endif
12141                                         } else {
12142                                                 msg.dt.sg_list[j].addr =
12143                                                         sgl[i].addr;
12144                                         }
12145                                         msg.dt.sg_list[j].len = sgl[i].len;
12146                                 }
12147                         }
12148
12149                         sg_entries_sent += msg.dt.cur_sg_entries;
12150                         if (sg_entries_sent >= msg.dt.kern_sg_entries)
12151                                 msg.dt.sg_last = 1;
12152                         else
12153                                 msg.dt.sg_last = 0;
12154
12155                         /*
12156                          * XXX KDM drop and reacquire the lock here?
12157                          */
12158                         if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12159                             sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12160                                 /*
12161                                  * XXX do something here.
12162                                  */
12163                         }
12164
12165                         msg.dt.sent_sg_entries = sg_entries_sent;
12166                 }
12167                 io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12168                 if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12169                         ctl_failover_io(io, /*have_lock*/ 0);
12170
12171         } else {
12172
12173                 /*
12174                  * Lookup the fe_datamove() function for this particular
12175                  * front end.
12176                  */
12177                 fe_datamove =
12178                     control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12179
12180                 fe_datamove(io);
12181         }
12182 }
12183
12184 static void
12185 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12186 {
12187         union ctl_ha_msg msg;
12188         int isc_status;
12189
12190         memset(&msg, 0, sizeof(msg));
12191
12192         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12193         msg.hdr.original_sc = io;
12194         msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12195         msg.hdr.nexus = io->io_hdr.nexus;
12196         msg.hdr.status = io->io_hdr.status;
12197         msg.scsi.tag_num = io->scsiio.tag_num;
12198         msg.scsi.tag_type = io->scsiio.tag_type;
12199         msg.scsi.scsi_status = io->scsiio.scsi_status;
12200         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12201                sizeof(io->scsiio.sense_data));
12202         msg.scsi.sense_len = io->scsiio.sense_len;
12203         msg.scsi.sense_residual = io->scsiio.sense_residual;
12204         msg.scsi.fetd_status = io->io_hdr.port_status;
12205         msg.scsi.residual = io->scsiio.residual;
12206         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12207
12208         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12209                 ctl_failover_io(io, /*have_lock*/ have_lock);
12210                 return;
12211         }
12212
12213         isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12214         if (isc_status > CTL_HA_STATUS_SUCCESS) {
12215                 /* XXX do something if this fails */
12216         }
12217
12218 }
12219
12220 /*
12221  * The DMA to the remote side is done, now we need to tell the other side
12222  * we're done so it can continue with its data movement.
12223  */
12224 static void
12225 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12226 {
12227         union ctl_io *io;
12228
12229         io = rq->context;
12230
12231         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12232                 printf("%s: ISC DMA write failed with error %d", __func__,
12233                        rq->ret);
12234                 ctl_set_internal_failure(&io->scsiio,
12235                                          /*sks_valid*/ 1,
12236                                          /*retry_count*/ rq->ret);
12237         }
12238
12239         ctl_dt_req_free(rq);
12240
12241         /*
12242          * In this case, we had to malloc the memory locally.  Free it.
12243          */
12244         if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12245                 int i;
12246                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12247                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
12248         }
12249         /*
12250          * The data is in local and remote memory, so now we need to send
12251          * status (good or back) back to the other side.
12252          */
12253         ctl_send_datamove_done(io, /*have_lock*/ 0);
12254 }
12255
12256 /*
12257  * We've moved the data from the host/controller into local memory.  Now we
12258  * need to push it over to the remote controller's memory.
12259  */
12260 static int
12261 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12262 {
12263         int retval;
12264
12265         retval = 0;
12266
12267         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12268                                           ctl_datamove_remote_write_cb);
12269
12270         return (retval);
12271 }
12272
12273 static void
12274 ctl_datamove_remote_write(union ctl_io *io)
12275 {
12276         int retval;
12277         void (*fe_datamove)(union ctl_io *io);
12278
12279         /*
12280          * - Get the data from the host/HBA into local memory.
12281          * - DMA memory from the local controller to the remote controller.
12282          * - Send status back to the remote controller.
12283          */
12284
12285         retval = ctl_datamove_remote_sgl_setup(io);
12286         if (retval != 0)
12287                 return;
12288
12289         /* Switch the pointer over so the FETD knows what to do */
12290         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12291
12292         /*
12293          * Use a custom move done callback, since we need to send completion
12294          * back to the other controller, not to the backend on this side.
12295          */
12296         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12297
12298         fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12299
12300         fe_datamove(io);
12301
12302         return;
12303
12304 }
12305
12306 static int
12307 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12308 {
12309 #if 0
12310         char str[256];
12311         char path_str[64];
12312         struct sbuf sb;
12313 #endif
12314
12315         /*
12316          * In this case, we had to malloc the memory locally.  Free it.
12317          */
12318         if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12319                 int i;
12320                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12321                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
12322         }
12323
12324 #if 0
12325         scsi_path_string(io, path_str, sizeof(path_str));
12326         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12327         sbuf_cat(&sb, path_str);
12328         scsi_command_string(&io->scsiio, NULL, &sb);
12329         sbuf_printf(&sb, "\n");
12330         sbuf_cat(&sb, path_str);
12331         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12332                     io->scsiio.tag_num, io->scsiio.tag_type);
12333         sbuf_cat(&sb, path_str);
12334         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12335                     io->io_hdr.flags, io->io_hdr.status);
12336         sbuf_finish(&sb);
12337         printk("%s", sbuf_data(&sb));
12338 #endif
12339
12340
12341         /*
12342          * The read is done, now we need to send status (good or bad) back
12343          * to the other side.
12344          */
12345         ctl_send_datamove_done(io, /*have_lock*/ 0);
12346
12347         return (0);
12348 }
12349
12350 static void
12351 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12352 {
12353         union ctl_io *io;
12354         void (*fe_datamove)(union ctl_io *io);
12355
12356         io = rq->context;
12357
12358         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12359                 printf("%s: ISC DMA read failed with error %d", __func__,
12360                        rq->ret);
12361                 ctl_set_internal_failure(&io->scsiio,
12362                                          /*sks_valid*/ 1,
12363                                          /*retry_count*/ rq->ret);
12364         }
12365
12366         ctl_dt_req_free(rq);
12367
12368         /* Switch the pointer over so the FETD knows what to do */
12369         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12370
12371         /*
12372          * Use a custom move done callback, since we need to send completion
12373          * back to the other controller, not to the backend on this side.
12374          */
12375         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12376
12377         /* XXX KDM add checks like the ones in ctl_datamove? */
12378
12379         fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12380
12381         fe_datamove(io);
12382 }
12383
12384 static int
12385 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12386 {
12387         struct ctl_sg_entry *local_sglist, *remote_sglist;
12388         struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
12389         struct ctl_softc *softc;
12390         int retval;
12391         int i;
12392
12393         retval = 0;
12394         softc = control_softc;
12395
12396         local_sglist = io->io_hdr.local_sglist;
12397         local_dma_sglist = io->io_hdr.local_dma_sglist;
12398         remote_sglist = io->io_hdr.remote_sglist;
12399         remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12400
12401         if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
12402                 for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
12403                         local_sglist[i].len = remote_sglist[i].len;
12404
12405                         /*
12406                          * XXX Detect the situation where the RS-level I/O
12407                          * redirector on the other side has already read the
12408                          * data off of the AOR RS on this side, and
12409                          * transferred it to remote (mirror) memory on the
12410                          * other side.  Since we already have the data in
12411                          * memory here, we just need to use it.
12412                          *
12413                          * XXX KDM this can probably be removed once we
12414                          * get the cache device code in and take the
12415                          * current AOR implementation out.
12416                          */
12417 #ifdef NEEDTOPORT
12418                         if ((remote_sglist[i].addr >=
12419                              (void *)vtophys(softc->mirr->addr))
12420                          && (remote_sglist[i].addr <
12421                              ((void *)vtophys(softc->mirr->addr) +
12422                              CacheMirrorOffset))) {
12423                                 local_sglist[i].addr = remote_sglist[i].addr -
12424                                         CacheMirrorOffset;
12425                                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12426                                      CTL_FLAG_DATA_IN)
12427                                         io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
12428                         } else {
12429                                 local_sglist[i].addr = remote_sglist[i].addr +
12430                                         CacheMirrorOffset;
12431                         }
12432 #endif
12433 #if 0
12434                         printf("%s: local %p, remote %p, len %d\n",
12435                                __func__, local_sglist[i].addr,
12436                                remote_sglist[i].addr, local_sglist[i].len);
12437 #endif
12438                 }
12439         } else {
12440                 uint32_t len_to_go;
12441
12442                 /*
12443                  * In this case, we don't have automatically allocated
12444                  * memory for this I/O on this controller.  This typically
12445                  * happens with internal CTL I/O -- e.g. inquiry, mode
12446                  * sense, etc.  Anything coming from RAIDCore will have
12447                  * a mirror area available.
12448                  */
12449                 len_to_go = io->scsiio.kern_data_len;
12450
12451                 /*
12452                  * Clear the no datasync flag, we have to use malloced
12453                  * buffers.
12454                  */
12455                 io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
12456
12457                 /*
12458                  * The difficult thing here is that the size of the various
12459                  * S/G segments may be different than the size from the
12460                  * remote controller.  That'll make it harder when DMAing
12461                  * the data back to the other side.
12462                  */
12463                 for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
12464                      sizeof(io->io_hdr.remote_sglist[0])) &&
12465                      (len_to_go > 0); i++) {
12466                         local_sglist[i].len = ctl_min(len_to_go, 131072);
12467                         CTL_SIZE_8B(local_dma_sglist[i].len,
12468                                     local_sglist[i].len);
12469                         local_sglist[i].addr =
12470                                 malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
12471
12472                         local_dma_sglist[i].addr = local_sglist[i].addr;
12473
12474                         if (local_sglist[i].addr == NULL) {
12475                                 int j;
12476
12477                                 printf("malloc failed for %zd bytes!",
12478                                        local_dma_sglist[i].len);
12479                                 for (j = 0; j < i; j++) {
12480                                         free(local_sglist[j].addr, M_CTL);
12481                                 }
12482                                 ctl_set_internal_failure(&io->scsiio,
12483                                                          /*sks_valid*/ 1,
12484                                                          /*retry_count*/ 4857);
12485                                 retval = 1;
12486                                 goto bailout_error;
12487
12488                         }
12489                         /* XXX KDM do we need a sync here? */
12490
12491                         len_to_go -= local_sglist[i].len;
12492                 }
12493                 /*
12494                  * Reset the number of S/G entries accordingly.  The
12495                  * original number of S/G entries is available in
12496                  * rem_sg_entries.
12497                  */
12498                 io->scsiio.kern_sg_entries = i;
12499
12500 #if 0
12501                 printf("%s: kern_sg_entries = %d\n", __func__,
12502                        io->scsiio.kern_sg_entries);
12503                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12504                         printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
12505                                local_sglist[i].addr, local_sglist[i].len,
12506                                local_dma_sglist[i].len);
12507 #endif
12508         }
12509
12510
12511         return (retval);
12512
12513 bailout_error:
12514
12515         ctl_send_datamove_done(io, /*have_lock*/ 0);
12516
12517         return (retval);
12518 }
12519
12520 static int
12521 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12522                          ctl_ha_dt_cb callback)
12523 {
12524         struct ctl_ha_dt_req *rq;
12525         struct ctl_sg_entry *remote_sglist, *local_sglist;
12526         struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
12527         uint32_t local_used, remote_used, total_used;
12528         int retval;
12529         int i, j;
12530
12531         retval = 0;
12532
12533         rq = ctl_dt_req_alloc();
12534
12535         /*
12536          * If we failed to allocate the request, and if the DMA didn't fail
12537          * anyway, set busy status.  This is just a resource allocation
12538          * failure.
12539          */
12540         if ((rq == NULL)
12541          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
12542                 ctl_set_busy(&io->scsiio);
12543
12544         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
12545
12546                 if (rq != NULL)
12547                         ctl_dt_req_free(rq);
12548
12549                 /*
12550                  * The data move failed.  We need to return status back
12551                  * to the other controller.  No point in trying to DMA
12552                  * data to the remote controller.
12553                  */
12554
12555                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12556
12557                 retval = 1;
12558
12559                 goto bailout;
12560         }
12561
12562         local_sglist = io->io_hdr.local_sglist;
12563         local_dma_sglist = io->io_hdr.local_dma_sglist;
12564         remote_sglist = io->io_hdr.remote_sglist;
12565         remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12566         local_used = 0;
12567         remote_used = 0;
12568         total_used = 0;
12569
12570         if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
12571                 rq->ret = CTL_HA_STATUS_SUCCESS;
12572                 rq->context = io;
12573                 callback(rq);
12574                 goto bailout;
12575         }
12576
12577         /*
12578          * Pull/push the data over the wire from/to the other controller.
12579          * This takes into account the possibility that the local and
12580          * remote sglists may not be identical in terms of the size of
12581          * the elements and the number of elements.
12582          *
12583          * One fundamental assumption here is that the length allocated for
12584          * both the local and remote sglists is identical.  Otherwise, we've
12585          * essentially got a coding error of some sort.
12586          */
12587         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12588                 int isc_ret;
12589                 uint32_t cur_len, dma_length;
12590                 uint8_t *tmp_ptr;
12591
12592                 rq->id = CTL_HA_DATA_CTL;
12593                 rq->command = command;
12594                 rq->context = io;
12595
12596                 /*
12597                  * Both pointers should be aligned.  But it is possible
12598                  * that the allocation length is not.  They should both
12599                  * also have enough slack left over at the end, though,
12600                  * to round up to the next 8 byte boundary.
12601                  */
12602                 cur_len = ctl_min(local_sglist[i].len - local_used,
12603                                   remote_sglist[j].len - remote_used);
12604
12605                 /*
12606                  * In this case, we have a size issue and need to decrease
12607                  * the size, except in the case where we actually have less
12608                  * than 8 bytes left.  In that case, we need to increase
12609                  * the DMA length to get the last bit.
12610                  */
12611                 if ((cur_len & 0x7) != 0) {
12612                         if (cur_len > 0x7) {
12613                                 cur_len = cur_len - (cur_len & 0x7);
12614                                 dma_length = cur_len;
12615                         } else {
12616                                 CTL_SIZE_8B(dma_length, cur_len);
12617                         }
12618
12619                 } else
12620                         dma_length = cur_len;
12621
12622                 /*
12623                  * If we had to allocate memory for this I/O, instead of using
12624                  * the non-cached mirror memory, we'll need to flush the cache
12625                  * before trying to DMA to the other controller.
12626                  *
12627                  * We could end up doing this multiple times for the same
12628                  * segment if we have a larger local segment than remote
12629                  * segment.  That shouldn't be an issue.
12630                  */
12631                 if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12632                         /*
12633                          * XXX KDM use bus_dmamap_sync() here.
12634                          */
12635                 }
12636
12637                 rq->size = dma_length;
12638
12639                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
12640                 tmp_ptr += local_used;
12641
12642                 /* Use physical addresses when talking to ISC hardware */
12643                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12644                         /* XXX KDM use busdma */
12645 #if 0
12646                         rq->local = vtophys(tmp_ptr);
12647 #endif
12648                 } else
12649                         rq->local = tmp_ptr;
12650
12651                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12652                 tmp_ptr += remote_used;
12653                 rq->remote = tmp_ptr;
12654
12655                 rq->callback = NULL;
12656
12657                 local_used += cur_len;
12658                 if (local_used >= local_sglist[i].len) {
12659                         i++;
12660                         local_used = 0;
12661                 }
12662
12663                 remote_used += cur_len;
12664                 if (remote_used >= remote_sglist[j].len) {
12665                         j++;
12666                         remote_used = 0;
12667                 }
12668                 total_used += cur_len;
12669
12670                 if (total_used >= io->scsiio.kern_data_len)
12671                         rq->callback = callback;
12672
12673                 if ((rq->size & 0x7) != 0) {
12674                         printf("%s: warning: size %d is not on 8b boundary\n",
12675                                __func__, rq->size);
12676                 }
12677                 if (((uintptr_t)rq->local & 0x7) != 0) {
12678                         printf("%s: warning: local %p not on 8b boundary\n",
12679                                __func__, rq->local);
12680                 }
12681                 if (((uintptr_t)rq->remote & 0x7) != 0) {
12682                         printf("%s: warning: remote %p not on 8b boundary\n",
12683                                __func__, rq->local);
12684                 }
12685 #if 0
12686                 printf("%s: %s: local %#x remote %#x size %d\n", __func__,
12687                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12688                        rq->local, rq->remote, rq->size);
12689 #endif
12690
12691                 isc_ret = ctl_dt_single(rq);
12692                 if (isc_ret == CTL_HA_STATUS_WAIT)
12693                         continue;
12694
12695                 if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
12696                         rq->ret = CTL_HA_STATUS_SUCCESS;
12697                 } else {
12698                         rq->ret = isc_ret;
12699                 }
12700                 callback(rq);
12701                 goto bailout;
12702         }
12703
12704 bailout:
12705         return (retval);
12706
12707 }
12708
12709 static void
12710 ctl_datamove_remote_read(union ctl_io *io)
12711 {
12712         int retval;
12713         int i;
12714
12715         /*
12716          * This will send an error to the other controller in the case of a
12717          * failure.
12718          */
12719         retval = ctl_datamove_remote_sgl_setup(io);
12720         if (retval != 0)
12721                 return;
12722
12723         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12724                                           ctl_datamove_remote_read_cb);
12725         if ((retval != 0)
12726          && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
12727                 /*
12728                  * Make sure we free memory if there was an error..  The
12729                  * ctl_datamove_remote_xfer() function will send the
12730                  * datamove done message, or call the callback with an
12731                  * error if there is a problem.
12732                  */
12733                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12734                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
12735         }
12736
12737         return;
12738 }
12739
12740 /*
12741  * Process a datamove request from the other controller.  This is used for
12742  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12743  * first.  Once that is complete, the data gets DMAed into the remote
12744  * controller's memory.  For reads, we DMA from the remote controller's
12745  * memory into our memory first, and then move it out to the FETD.
12746  */
12747 static void
12748 ctl_datamove_remote(union ctl_io *io)
12749 {
12750         struct ctl_softc *softc;
12751
12752         softc = control_softc;
12753
12754         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
12755
12756         /*
12757          * Note that we look for an aborted I/O here, but don't do some of
12758          * the other checks that ctl_datamove() normally does.  We don't
12759          * need to run the task queue, because this I/O is on the ISC
12760          * queue, which is executed by the work thread after the task queue.
12761          * We don't need to run the datamove delay code, since that should
12762          * have been done if need be on the other controller.
12763          */
12764         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12765
12766                 printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
12767                        io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
12768                        io->io_hdr.nexus.targ_port,
12769                        io->io_hdr.nexus.targ_target.id,
12770                        io->io_hdr.nexus.targ_lun);
12771                 io->io_hdr.status = CTL_CMD_ABORTED;
12772                 io->io_hdr.port_status = 31338;
12773
12774                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12775
12776                 return;
12777         }
12778
12779         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
12780                 ctl_datamove_remote_write(io);
12781         } else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
12782                 ctl_datamove_remote_read(io);
12783         } else {
12784                 union ctl_ha_msg msg;
12785                 struct scsi_sense_data *sense;
12786                 uint8_t sks[3];
12787                 int retry_count;
12788
12789                 memset(&msg, 0, sizeof(msg));
12790
12791                 msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
12792                 msg.hdr.status = CTL_SCSI_ERROR;
12793                 msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
12794
12795                 retry_count = 4243;
12796
12797                 sense = &msg.scsi.sense_data;
12798                 sks[0] = SSD_SCS_VALID;
12799                 sks[1] = (retry_count >> 8) & 0xff;
12800                 sks[2] = retry_count & 0xff;
12801
12802                 /* "Internal target failure" */
12803                 scsi_set_sense_data(sense,
12804                                     /*sense_format*/ SSD_TYPE_NONE,
12805                                     /*current_error*/ 1,
12806                                     /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
12807                                     /*asc*/ 0x44,
12808                                     /*ascq*/ 0x00,
12809                                     /*type*/ SSD_ELEM_SKS,
12810                                     /*size*/ sizeof(sks),
12811                                     /*data*/ sks,
12812                                     SSD_ELEM_NONE);
12813
12814                 io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12815                 if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12816                         ctl_failover_io(io, /*have_lock*/ 1);
12817                         return;
12818                 }
12819
12820                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
12821                     CTL_HA_STATUS_SUCCESS) {
12822                         /* XXX KDM what to do if this fails? */
12823                 }
12824                 return;
12825         }
12826
12827 }
12828
12829 static int
12830 ctl_process_done(union ctl_io *io)
12831 {
12832         struct ctl_lun *lun;
12833         struct ctl_softc *ctl_softc;
12834         void (*fe_done)(union ctl_io *io);
12835         uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
12836
12837         CTL_DEBUG_PRINT(("ctl_process_done\n"));
12838
12839         fe_done =
12840             control_softc->ctl_ports[targ_port]->fe_done;
12841
12842 #ifdef CTL_TIME_IO
12843         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12844                 char str[256];
12845                 char path_str[64];
12846                 struct sbuf sb;
12847
12848                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12849                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12850
12851                 sbuf_cat(&sb, path_str);
12852                 switch (io->io_hdr.io_type) {
12853                 case CTL_IO_SCSI:
12854                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12855                         sbuf_printf(&sb, "\n");
12856                         sbuf_cat(&sb, path_str);
12857                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12858                                     io->scsiio.tag_num, io->scsiio.tag_type);
12859                         break;
12860                 case CTL_IO_TASK:
12861                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12862                                     "Tag Type: %d\n", io->taskio.task_action,
12863                                     io->taskio.tag_num, io->taskio.tag_type);
12864                         break;
12865                 default:
12866                         printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12867                         panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12868                         break;
12869                 }
12870                 sbuf_cat(&sb, path_str);
12871                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12872                             (intmax_t)time_uptime - io->io_hdr.start_time);
12873                 sbuf_finish(&sb);
12874                 printf("%s", sbuf_data(&sb));
12875         }
12876 #endif /* CTL_TIME_IO */
12877
12878         switch (io->io_hdr.io_type) {
12879         case CTL_IO_SCSI:
12880                 break;
12881         case CTL_IO_TASK:
12882                 if (bootverbose || verbose > 0)
12883                         ctl_io_error_print(io, NULL);
12884                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
12885                         ctl_free_io(io);
12886                 else
12887                         fe_done(io);
12888                 return (CTL_RETVAL_COMPLETE);
12889                 break;
12890         default:
12891                 printf("ctl_process_done: invalid io type %d\n",
12892                        io->io_hdr.io_type);
12893                 panic("ctl_process_done: invalid io type %d\n",
12894                       io->io_hdr.io_type);
12895                 break; /* NOTREACHED */
12896         }
12897
12898         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12899         if (lun == NULL) {
12900                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
12901                                  io->io_hdr.nexus.targ_mapped_lun));
12902                 fe_done(io);
12903                 goto bailout;
12904         }
12905         ctl_softc = lun->ctl_softc;
12906
12907         mtx_lock(&lun->lun_lock);
12908
12909         /*
12910          * Check to see if we have any errors to inject here.  We only
12911          * inject errors for commands that don't already have errors set.
12912          */
12913         if ((STAILQ_FIRST(&lun->error_list) != NULL)
12914          && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
12915                 ctl_inject_error(lun, io);
12916
12917         /*
12918          * XXX KDM how do we treat commands that aren't completed
12919          * successfully?
12920          *
12921          * XXX KDM should we also track I/O latency?
12922          */
12923         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
12924             io->io_hdr.io_type == CTL_IO_SCSI) {
12925 #ifdef CTL_TIME_IO
12926                 struct bintime cur_bt;
12927 #endif
12928                 int type;
12929
12930                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12931                     CTL_FLAG_DATA_IN)
12932                         type = CTL_STATS_READ;
12933                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12934                     CTL_FLAG_DATA_OUT)
12935                         type = CTL_STATS_WRITE;
12936                 else
12937                         type = CTL_STATS_NO_IO;
12938
12939                 lun->stats.ports[targ_port].bytes[type] +=
12940                     io->scsiio.kern_total_len;
12941                 lun->stats.ports[targ_port].operations[type]++;
12942 #ifdef CTL_TIME_IO
12943                 bintime_add(&lun->stats.ports[targ_port].dma_time[type],
12944                    &io->io_hdr.dma_bt);
12945                 lun->stats.ports[targ_port].num_dmas[type] +=
12946                     io->io_hdr.num_dmas;
12947                 getbintime(&cur_bt);
12948                 bintime_sub(&cur_bt, &io->io_hdr.start_bt);
12949                 bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
12950 #endif
12951         }
12952
12953         /*
12954          * Remove this from the OOA queue.
12955          */
12956         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
12957
12958         /*
12959          * Run through the blocked queue on this LUN and see if anything
12960          * has become unblocked, now that this transaction is done.
12961          */
12962         ctl_check_blocked(lun);
12963
12964         /*
12965          * If the LUN has been invalidated, free it if there is nothing
12966          * left on its OOA queue.
12967          */
12968         if ((lun->flags & CTL_LUN_INVALID)
12969          && TAILQ_EMPTY(&lun->ooa_queue)) {
12970                 mtx_unlock(&lun->lun_lock);
12971                 mtx_lock(&ctl_softc->ctl_lock);
12972                 ctl_free_lun(lun);
12973                 mtx_unlock(&ctl_softc->ctl_lock);
12974         } else
12975                 mtx_unlock(&lun->lun_lock);
12976
12977         /*
12978          * If this command has been aborted, make sure we set the status
12979          * properly.  The FETD is responsible for freeing the I/O and doing
12980          * whatever it needs to do to clean up its state.
12981          */
12982         if (io->io_hdr.flags & CTL_FLAG_ABORT)
12983                 io->io_hdr.status = CTL_CMD_ABORTED;
12984
12985         /*
12986          * We print out status for every task management command.  For SCSI
12987          * commands, we filter out any unit attention errors; they happen
12988          * on every boot, and would clutter up the log.  Note:  task
12989          * management commands aren't printed here, they are printed above,
12990          * since they should never even make it down here.
12991          */
12992         switch (io->io_hdr.io_type) {
12993         case CTL_IO_SCSI: {
12994                 int error_code, sense_key, asc, ascq;
12995
12996                 sense_key = 0;
12997
12998                 if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
12999                  && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13000                         /*
13001                          * Since this is just for printing, no need to
13002                          * show errors here.
13003                          */
13004                         scsi_extract_sense_len(&io->scsiio.sense_data,
13005                                                io->scsiio.sense_len,
13006                                                &error_code,
13007                                                &sense_key,
13008                                                &asc,
13009                                                &ascq,
13010                                                /*show_errors*/ 0);
13011                 }
13012
13013                 if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
13014                  && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
13015                   || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
13016                   || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
13017
13018                         if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
13019                                 ctl_softc->skipped_prints++;
13020                         } else {
13021                                 uint32_t skipped_prints;
13022
13023                                 skipped_prints = ctl_softc->skipped_prints;
13024
13025                                 ctl_softc->skipped_prints = 0;
13026                                 ctl_softc->last_print_jiffies = time_uptime;
13027
13028                                 if (skipped_prints > 0) {
13029 #ifdef NEEDTOPORT
13030                                         csevent_log(CSC_CTL | CSC_SHELF_SW |
13031                                             CTL_ERROR_REPORT,
13032                                             csevent_LogType_Trace,
13033                                             csevent_Severity_Information,
13034                                             csevent_AlertLevel_Green,
13035                                             csevent_FRU_Firmware,
13036                                             csevent_FRU_Unknown,
13037                                             "High CTL error volume, %d prints "
13038                                             "skipped", skipped_prints);
13039 #endif
13040                                 }
13041                                 if (bootverbose || verbose > 0)
13042                                         ctl_io_error_print(io, NULL);
13043                         }
13044                 }
13045                 break;
13046         }
13047         case CTL_IO_TASK:
13048                 if (bootverbose || verbose > 0)
13049                         ctl_io_error_print(io, NULL);
13050                 break;
13051         default:
13052                 break;
13053         }
13054
13055         /*
13056          * Tell the FETD or the other shelf controller we're done with this
13057          * command.  Note that only SCSI commands get to this point.  Task
13058          * management commands are completed above.
13059          *
13060          * We only send status to the other controller if we're in XFER
13061          * mode.  In SER_ONLY mode, the I/O is done on the controller that
13062          * received the I/O (from CTL's perspective), and so the status is
13063          * generated there.
13064          *
13065          * XXX KDM if we hold the lock here, we could cause a deadlock
13066          * if the frontend comes back in in this context to queue
13067          * something.
13068          */
13069         if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13070          && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13071                 union ctl_ha_msg msg;
13072
13073                 memset(&msg, 0, sizeof(msg));
13074                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13075                 msg.hdr.original_sc = io->io_hdr.original_sc;
13076                 msg.hdr.nexus = io->io_hdr.nexus;
13077                 msg.hdr.status = io->io_hdr.status;
13078                 msg.scsi.scsi_status = io->scsiio.scsi_status;
13079                 msg.scsi.tag_num = io->scsiio.tag_num;
13080                 msg.scsi.tag_type = io->scsiio.tag_type;
13081                 msg.scsi.sense_len = io->scsiio.sense_len;
13082                 msg.scsi.sense_residual = io->scsiio.sense_residual;
13083                 msg.scsi.residual = io->scsiio.residual;
13084                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13085                        sizeof(io->scsiio.sense_data));
13086                 /*
13087                  * We copy this whether or not this is an I/O-related
13088                  * command.  Otherwise, we'd have to go and check to see
13089                  * whether it's a read/write command, and it really isn't
13090                  * worth it.
13091                  */
13092                 memcpy(&msg.scsi.lbalen,
13093                        &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13094                        sizeof(msg.scsi.lbalen));
13095
13096                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13097                                 sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13098                         /* XXX do something here */
13099                 }
13100
13101                 ctl_free_io(io);
13102         } else
13103                 fe_done(io);
13104
13105 bailout:
13106
13107         return (CTL_RETVAL_COMPLETE);
13108 }
13109
13110 /*
13111  * Front end should call this if it doesn't do autosense.  When the request
13112  * sense comes back in from the initiator, we'll dequeue this and send it.
13113  */
13114 int
13115 ctl_queue_sense(union ctl_io *io)
13116 {
13117         struct ctl_lun *lun;
13118         struct ctl_softc *ctl_softc;
13119         uint32_t initidx, targ_lun;
13120
13121         ctl_softc = control_softc;
13122
13123         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13124
13125         /*
13126          * LUN lookup will likely move to the ctl_work_thread() once we
13127          * have our new queueing infrastructure (that doesn't put things on
13128          * a per-LUN queue initially).  That is so that we can handle
13129          * things like an INQUIRY to a LUN that we don't have enabled.  We
13130          * can't deal with that right now.
13131          */
13132         mtx_lock(&ctl_softc->ctl_lock);
13133
13134         /*
13135          * If we don't have a LUN for this, just toss the sense
13136          * information.
13137          */
13138         targ_lun = io->io_hdr.nexus.targ_lun;
13139         if (io->io_hdr.nexus.lun_map_fn != NULL)
13140                 targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
13141         if ((targ_lun < CTL_MAX_LUNS)
13142          && (ctl_softc->ctl_luns[targ_lun] != NULL))
13143                 lun = ctl_softc->ctl_luns[targ_lun];
13144         else
13145                 goto bailout;
13146
13147         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13148
13149         mtx_lock(&lun->lun_lock);
13150         /*
13151          * Already have CA set for this LUN...toss the sense information.
13152          */
13153         if (ctl_is_set(lun->have_ca, initidx)) {
13154                 mtx_unlock(&lun->lun_lock);
13155                 goto bailout;
13156         }
13157
13158         memcpy(&lun->pending_sense[initidx].sense, &io->scsiio.sense_data,
13159                ctl_min(sizeof(lun->pending_sense[initidx].sense),
13160                sizeof(io->scsiio.sense_data)));
13161         ctl_set_mask(lun->have_ca, initidx);
13162         mtx_unlock(&lun->lun_lock);
13163
13164 bailout:
13165         mtx_unlock(&ctl_softc->ctl_lock);
13166
13167         ctl_free_io(io);
13168
13169         return (CTL_RETVAL_COMPLETE);
13170 }
13171
13172 /*
13173  * Primary command inlet from frontend ports.  All SCSI and task I/O
13174  * requests must go through this function.
13175  */
13176 int
13177 ctl_queue(union ctl_io *io)
13178 {
13179         struct ctl_softc *ctl_softc;
13180
13181         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13182
13183         ctl_softc = control_softc;
13184
13185 #ifdef CTL_TIME_IO
13186         io->io_hdr.start_time = time_uptime;
13187         getbintime(&io->io_hdr.start_bt);
13188 #endif /* CTL_TIME_IO */
13189
13190         /* Map FE-specific LUN ID into global one. */
13191         if (io->io_hdr.nexus.lun_map_fn != NULL)
13192                 io->io_hdr.nexus.targ_mapped_lun = io->io_hdr.nexus.lun_map_fn(
13193                     io->io_hdr.nexus.lun_map_arg, io->io_hdr.nexus.targ_lun);
13194         else
13195                 io->io_hdr.nexus.targ_mapped_lun = io->io_hdr.nexus.targ_lun;
13196
13197         switch (io->io_hdr.io_type) {
13198         case CTL_IO_SCSI:
13199         case CTL_IO_TASK:
13200                 ctl_enqueue_incoming(io);
13201                 break;
13202         default:
13203                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13204                 return (EINVAL);
13205         }
13206
13207         return (CTL_RETVAL_COMPLETE);
13208 }
13209
13210 #ifdef CTL_IO_DELAY
13211 static void
13212 ctl_done_timer_wakeup(void *arg)
13213 {
13214         union ctl_io *io;
13215
13216         io = (union ctl_io *)arg;
13217         ctl_done(io);
13218 }
13219 #endif /* CTL_IO_DELAY */
13220
13221 void
13222 ctl_done(union ctl_io *io)
13223 {
13224         struct ctl_softc *ctl_softc;
13225
13226         ctl_softc = control_softc;
13227
13228         /*
13229          * Enable this to catch duplicate completion issues.
13230          */
13231 #if 0
13232         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13233                 printf("%s: type %d msg %d cdb %x iptl: "
13234                        "%d:%d:%d:%d tag 0x%04x "
13235                        "flag %#x status %x\n",
13236                         __func__,
13237                         io->io_hdr.io_type,
13238                         io->io_hdr.msg_type,
13239                         io->scsiio.cdb[0],
13240                         io->io_hdr.nexus.initid.id,
13241                         io->io_hdr.nexus.targ_port,
13242                         io->io_hdr.nexus.targ_target.id,
13243                         io->io_hdr.nexus.targ_lun,
13244                         (io->io_hdr.io_type ==
13245                         CTL_IO_TASK) ?
13246                         io->taskio.tag_num :
13247                         io->scsiio.tag_num,
13248                         io->io_hdr.flags,
13249                         io->io_hdr.status);
13250         } else
13251                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13252 #endif
13253
13254         /*
13255          * This is an internal copy of an I/O, and should not go through
13256          * the normal done processing logic.
13257          */
13258         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13259                 return;
13260
13261         /*
13262          * We need to send a msg to the serializing shelf to finish the IO
13263          * as well.  We don't send a finish message to the other shelf if
13264          * this is a task management command.  Task management commands
13265          * aren't serialized in the OOA queue, but rather just executed on
13266          * both shelf controllers for commands that originated on that
13267          * controller.
13268          */
13269         if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13270          && (io->io_hdr.io_type != CTL_IO_TASK)) {
13271                 union ctl_ha_msg msg_io;
13272
13273                 msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13274                 msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13275                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13276                     sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13277                 }
13278                 /* continue on to finish IO */
13279         }
13280 #ifdef CTL_IO_DELAY
13281         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13282                 struct ctl_lun *lun;
13283
13284                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13285
13286                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13287         } else {
13288                 struct ctl_lun *lun;
13289
13290                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13291
13292                 if ((lun != NULL)
13293                  && (lun->delay_info.done_delay > 0)) {
13294                         struct callout *callout;
13295
13296                         callout = (struct callout *)&io->io_hdr.timer_bytes;
13297                         callout_init(callout, /*mpsafe*/ 1);
13298                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13299                         callout_reset(callout,
13300                                       lun->delay_info.done_delay * hz,
13301                                       ctl_done_timer_wakeup, io);
13302                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13303                                 lun->delay_info.done_delay = 0;
13304                         return;
13305                 }
13306         }
13307 #endif /* CTL_IO_DELAY */
13308
13309         ctl_enqueue_done(io);
13310 }
13311
13312 int
13313 ctl_isc(struct ctl_scsiio *ctsio)
13314 {
13315         struct ctl_lun *lun;
13316         int retval;
13317
13318         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13319
13320         CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13321
13322         CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13323
13324         retval = lun->backend->data_submit((union ctl_io *)ctsio);
13325
13326         return (retval);
13327 }
13328
13329
13330 static void
13331 ctl_work_thread(void *arg)
13332 {
13333         struct ctl_thread *thr = (struct ctl_thread *)arg;
13334         struct ctl_softc *softc = thr->ctl_softc;
13335         union ctl_io *io;
13336         int retval;
13337
13338         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13339
13340         for (;;) {
13341                 retval = 0;
13342
13343                 /*
13344                  * We handle the queues in this order:
13345                  * - ISC
13346                  * - done queue (to free up resources, unblock other commands)
13347                  * - RtR queue
13348                  * - incoming queue
13349                  *
13350                  * If those queues are empty, we break out of the loop and
13351                  * go to sleep.
13352                  */
13353                 mtx_lock(&thr->queue_lock);
13354                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13355                 if (io != NULL) {
13356                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13357                         mtx_unlock(&thr->queue_lock);
13358                         ctl_handle_isc(io);
13359                         continue;
13360                 }
13361                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13362                 if (io != NULL) {
13363                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13364                         /* clear any blocked commands, call fe_done */
13365                         mtx_unlock(&thr->queue_lock);
13366                         retval = ctl_process_done(io);
13367                         continue;
13368                 }
13369                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13370                 if (io != NULL) {
13371                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13372                         mtx_unlock(&thr->queue_lock);
13373                         if (io->io_hdr.io_type == CTL_IO_TASK)
13374                                 ctl_run_task(io);
13375                         else
13376                                 ctl_scsiio_precheck(softc, &io->scsiio);
13377                         continue;
13378                 }
13379                 if (!ctl_pause_rtr) {
13380                         io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13381                         if (io != NULL) {
13382                                 STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13383                                 mtx_unlock(&thr->queue_lock);
13384                                 retval = ctl_scsiio(&io->scsiio);
13385                                 if (retval != CTL_RETVAL_COMPLETE)
13386                                         CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13387                                 continue;
13388                         }
13389                 }
13390
13391                 /* Sleep until we have something to do. */
13392                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13393         }
13394 }
13395
13396 static void
13397 ctl_lun_thread(void *arg)
13398 {
13399         struct ctl_softc *softc = (struct ctl_softc *)arg;
13400         struct ctl_be_lun *be_lun;
13401         int retval;
13402
13403         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13404
13405         for (;;) {
13406                 retval = 0;
13407                 mtx_lock(&softc->ctl_lock);
13408                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13409                 if (be_lun != NULL) {
13410                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13411                         mtx_unlock(&softc->ctl_lock);
13412                         ctl_create_lun(be_lun);
13413                         continue;
13414                 }
13415
13416                 /* Sleep until we have something to do. */
13417                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13418                     PDROP | PRIBIO, "-", 0);
13419         }
13420 }
13421
13422 static void
13423 ctl_enqueue_incoming(union ctl_io *io)
13424 {
13425         struct ctl_softc *softc = control_softc;
13426         struct ctl_thread *thr;
13427         u_int idx;
13428
13429         idx = (io->io_hdr.nexus.targ_port * 127 +
13430                io->io_hdr.nexus.initid.id) % worker_threads;
13431         thr = &softc->threads[idx];
13432         mtx_lock(&thr->queue_lock);
13433         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13434         mtx_unlock(&thr->queue_lock);
13435         wakeup(thr);
13436 }
13437
13438 static void
13439 ctl_enqueue_rtr(union ctl_io *io)
13440 {
13441         struct ctl_softc *softc = control_softc;
13442         struct ctl_thread *thr;
13443
13444         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13445         mtx_lock(&thr->queue_lock);
13446         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13447         mtx_unlock(&thr->queue_lock);
13448         wakeup(thr);
13449 }
13450
13451 static void
13452 ctl_enqueue_done(union ctl_io *io)
13453 {
13454         struct ctl_softc *softc = control_softc;
13455         struct ctl_thread *thr;
13456
13457         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13458         mtx_lock(&thr->queue_lock);
13459         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13460         mtx_unlock(&thr->queue_lock);
13461         wakeup(thr);
13462 }
13463
13464 static void
13465 ctl_enqueue_isc(union ctl_io *io)
13466 {
13467         struct ctl_softc *softc = control_softc;
13468         struct ctl_thread *thr;
13469
13470         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13471         mtx_lock(&thr->queue_lock);
13472         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13473         mtx_unlock(&thr->queue_lock);
13474         wakeup(thr);
13475 }
13476
13477 /* Initialization and failover */
13478
13479 void
13480 ctl_init_isc_msg(void)
13481 {
13482         printf("CTL: Still calling this thing\n");
13483 }
13484
13485 /*
13486  * Init component
13487  *      Initializes component into configuration defined by bootMode
13488  *      (see hasc-sv.c)
13489  *      returns hasc_Status:
13490  *              OK
13491  *              ERROR - fatal error
13492  */
13493 static ctl_ha_comp_status
13494 ctl_isc_init(struct ctl_ha_component *c)
13495 {
13496         ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13497
13498         c->status = ret;
13499         return ret;
13500 }
13501
13502 /* Start component
13503  *      Starts component in state requested. If component starts successfully,
13504  *      it must set its own state to the requestrd state
13505  *      When requested state is HASC_STATE_HA, the component may refine it
13506  *      by adding _SLAVE or _MASTER flags.
13507  *      Currently allowed state transitions are:
13508  *      UNKNOWN->HA             - initial startup
13509  *      UNKNOWN->SINGLE - initial startup when no parter detected
13510  *      HA->SINGLE              - failover
13511  * returns ctl_ha_comp_status:
13512  *              OK      - component successfully started in requested state
13513  *              FAILED  - could not start the requested state, failover may
13514  *                        be possible
13515  *              ERROR   - fatal error detected, no future startup possible
13516  */
13517 static ctl_ha_comp_status
13518 ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
13519 {
13520         ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13521
13522         printf("%s: go\n", __func__);
13523
13524         // UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
13525         if (c->state == CTL_HA_STATE_UNKNOWN ) {
13526                 ctl_is_single = 0;
13527                 if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
13528                     != CTL_HA_STATUS_SUCCESS) {
13529                         printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
13530                         ret = CTL_HA_COMP_STATUS_ERROR;
13531                 }
13532         } else if (CTL_HA_STATE_IS_HA(c->state)
13533                 && CTL_HA_STATE_IS_SINGLE(state)){
13534                 // HA->SINGLE transition
13535                 ctl_failover();
13536                 ctl_is_single = 1;
13537         } else {
13538                 printf("ctl_isc_start:Invalid state transition %X->%X\n",
13539                        c->state, state);
13540                 ret = CTL_HA_COMP_STATUS_ERROR;
13541         }
13542         if (CTL_HA_STATE_IS_SINGLE(state))
13543                 ctl_is_single = 1;
13544
13545         c->state = state;
13546         c->status = ret;
13547         return ret;
13548 }
13549
13550 /*
13551  * Quiesce component
13552  * The component must clear any error conditions (set status to OK) and
13553  * prepare itself to another Start call
13554  * returns ctl_ha_comp_status:
13555  *      OK
13556  *      ERROR
13557  */
13558 static ctl_ha_comp_status
13559 ctl_isc_quiesce(struct ctl_ha_component *c)
13560 {
13561         int ret = CTL_HA_COMP_STATUS_OK;
13562
13563         ctl_pause_rtr = 1;
13564         c->status = ret;
13565         return ret;
13566 }
13567
13568 struct ctl_ha_component ctl_ha_component_ctlisc =
13569 {
13570         .name = "CTL ISC",
13571         .state = CTL_HA_STATE_UNKNOWN,
13572         .init = ctl_isc_init,
13573         .start = ctl_isc_start,
13574         .quiesce = ctl_isc_quiesce
13575 };
13576
13577 /*
13578  *  vim: ts=8
13579  */