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