sys/cam/ctl/ctl_backend_block.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
   3  *
   4  * Copyright (c) 2003 Silicon Graphics International Corp.
   5  * Copyright (c) 2009-2011 Spectra Logic Corporation
   6  * Copyright (c) 2012 The FreeBSD Foundation
   7  * Copyright (c) 2014-2015 Alexander Motin <mav@FreeBSD.org>
   8  * All rights reserved.
   9  *
  10  * Portions of this software were developed by Edward Tomasz Napierala
  11  * under sponsorship from the FreeBSD Foundation.
  12  *
  13  * Redistribution and use in source and binary forms, with or without
  14  * modification, are permitted provided that the following conditions
  15  * are met:
  16  * 1. Redistributions of source code must retain the above copyright
  17  *    notice, this list of conditions, and the following disclaimer,
  18  *    without modification.
  19  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
  20  *    substantially similar to the "NO WARRANTY" disclaimer below
  21  *    ("Disclaimer") and any redistribution must be conditioned upon
  22  *    including a substantially similar Disclaimer requirement for further
  23  *    binary redistribution.
  24  *
  25  * NO WARRANTY
  26  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  27  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  28  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
  29  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  30  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
  34  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
  35  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  36  * POSSIBILITY OF SUCH DAMAGES.
  37  *
  38  * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_block.c#5 $
  39  */
  40 /*
  41  * CAM Target Layer driver backend for block devices.
  42  *
  43  * Author: Ken Merry <ken@FreeBSD.org>
  44  */
  45 #include <sys/cdefs.h>
  46 __FBSDID("$FreeBSD$");
  47
  48 #include <sys/param.h>
  49 #include <sys/systm.h>
  50 #include <sys/kernel.h>
  51 #include <sys/types.h>
  52 #include <sys/kthread.h>
  53 #include <sys/bio.h>
  54 #include <sys/fcntl.h>
  55 #include <sys/limits.h>
  56 #include <sys/lock.h>
  57 #include <sys/mutex.h>
  58 #include <sys/condvar.h>
  59 #include <sys/malloc.h>
  60 #include <sys/conf.h>
  61 #include <sys/ioccom.h>
  62 #include <sys/queue.h>
  63 #include <sys/sbuf.h>
  64 #include <sys/endian.h>
  65 #include <sys/uio.h>
  66 #include <sys/buf.h>
  67 #include <sys/taskqueue.h>
  68 #include <sys/vnode.h>
  69 #include <sys/namei.h>
  70 #include <sys/mount.h>
  71 #include <sys/disk.h>
  72 #include <sys/fcntl.h>
  73 #include <sys/filedesc.h>
  74 #include <sys/filio.h>
  75 #include <sys/proc.h>
  76 #include <sys/pcpu.h>
  77 #include <sys/module.h>
  78 #include <sys/sdt.h>
  79 #include <sys/devicestat.h>
  80 #include <sys/sysctl.h>
  81 #include <sys/nv.h>
  82 #include <sys/dnv.h>
  83 #include <sys/sx.h>
  84
  85 #include <geom/geom.h>
  86
  87 #include <cam/cam.h>
  88 #include <cam/scsi/scsi_all.h>
  89 #include <cam/scsi/scsi_da.h>
  90 #include <cam/ctl/ctl_io.h>
  91 #include <cam/ctl/ctl.h>
  92 #include <cam/ctl/ctl_backend.h>
  93 #include <cam/ctl/ctl_ioctl.h>
  94 #include <cam/ctl/ctl_ha.h>
  95 #include <cam/ctl/ctl_scsi_all.h>
  96 #include <cam/ctl/ctl_private.h>
  97 #include <cam/ctl/ctl_error.h>
  98
  99 /*
 100  * The idea here is that we'll allocate enough S/G space to hold a 1MB
 101  * I/O.  If we get an I/O larger than that, we'll split it.
 102  */
 103 #define CTLBLK_HALF_IO_SIZE     (512 * 1024)
 104 #define CTLBLK_MAX_IO_SIZE      (CTLBLK_HALF_IO_SIZE * 2)
 105 #define CTLBLK_MIN_SEG          (128 * 1024)
 106 #define CTLBLK_MAX_SEG          MIN(CTLBLK_HALF_IO_SIZE, maxphys)
 107 #define CTLBLK_HALF_SEGS        MAX(CTLBLK_HALF_IO_SIZE / CTLBLK_MIN_SEG, 1)
 108 #define CTLBLK_MAX_SEGS         (CTLBLK_HALF_SEGS * 2)
 109 #define CTLBLK_NUM_SEGS         (CTLBLK_MAX_IO_SIZE / CTLBLK_MAX_SEG)
 110
 111 #ifdef CTLBLK_DEBUG
 112 #define DPRINTF(fmt, args...) \
 113     printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args)
 114 #else
 115 #define DPRINTF(fmt, args...) do {} while(0)
 116 #endif
 117
 118 #define PRIV(io)        \
 119     ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND])
 120 #define ARGS(io)        \
 121     ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN])
 122
 123 SDT_PROVIDER_DEFINE(cbb);
 124
 125 typedef enum {
 126         CTL_BE_BLOCK_LUN_UNCONFIGURED   = 0x01,
 127         CTL_BE_BLOCK_LUN_WAITING        = 0x04,
 128 } ctl_be_block_lun_flags;
 129
 130 typedef enum {
 131         CTL_BE_BLOCK_NONE,
 132         CTL_BE_BLOCK_DEV,
 133         CTL_BE_BLOCK_FILE
 134 } ctl_be_block_type;
 135
 136 struct ctl_be_block_filedata {
 137         struct ucred *cred;
 138 };
 139
 140 union ctl_be_block_bedata {
 141         struct ctl_be_block_filedata file;
 142 };
 143
 144 struct ctl_be_block_io;
 145 struct ctl_be_block_lun;
 146
 147 typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun,
 148                                struct ctl_be_block_io *beio);
 149 typedef uint64_t (*cbb_getattr_t)(struct ctl_be_block_lun *be_lun,
 150                                   const char *attrname);
 151
 152 /*
 153  * Backend LUN structure.  There is a 1:1 mapping between a block device
 154  * and a backend block LUN, and between a backend block LUN and a CTL LUN.
 155  */
 156 struct ctl_be_block_lun {
 157         struct ctl_be_lun cbe_lun;              /* Must be first element. */
 158         struct ctl_lun_create_params params;
 159         char *dev_path;
 160         ctl_be_block_type dev_type;
 161         struct vnode *vn;
 162         union ctl_be_block_bedata backend;
 163         cbb_dispatch_t dispatch;
 164         cbb_dispatch_t lun_flush;
 165         cbb_dispatch_t unmap;
 166         cbb_dispatch_t get_lba_status;
 167         cbb_getattr_t getattr;
 168         uint64_t size_blocks;
 169         uint64_t size_bytes;
 170         struct ctl_be_block_softc *softc;
 171         struct devstat *disk_stats;
 172         ctl_be_block_lun_flags flags;
 173         SLIST_ENTRY(ctl_be_block_lun) links;
 174         struct taskqueue *io_taskqueue;
 175         struct task io_task;
 176         int num_threads;
 177         STAILQ_HEAD(, ctl_io_hdr) input_queue;
 178         STAILQ_HEAD(, ctl_io_hdr) config_read_queue;
 179         STAILQ_HEAD(, ctl_io_hdr) config_write_queue;
 180         STAILQ_HEAD(, ctl_io_hdr) datamove_queue;
 181         struct mtx_padalign io_lock;
 182         struct mtx_padalign queue_lock;
 183 };
 184
 185 /*
 186  * Overall softc structure for the block backend module.
 187  */
 188 struct ctl_be_block_softc {
 189         struct sx                        modify_lock;
 190         struct mtx                       lock;
 191         int                              num_luns;
 192         SLIST_HEAD(, ctl_be_block_lun)   lun_list;
 193         uma_zone_t                       beio_zone;
 194         uma_zone_t                       bufmin_zone;
 195         uma_zone_t                       bufmax_zone;
 196 };
 197
 198 static struct ctl_be_block_softc backend_block_softc;
 199
 200 /*
 201  * Per-I/O information.
 202  */
 203 struct ctl_be_block_io {
 204         union ctl_io                    *io;
 205         struct ctl_sg_entry             sg_segs[CTLBLK_MAX_SEGS];
 206         struct iovec                    xiovecs[CTLBLK_MAX_SEGS];
 207         int                             refcnt;
 208         int                             bio_cmd;
 209         int                             two_sglists;
 210         int                             num_segs;
 211         int                             num_bios_sent;
 212         int                             num_bios_done;
 213         int                             send_complete;
 214         int                             first_error;
 215         uint64_t                        first_error_offset;
 216         struct bintime                  ds_t0;
 217         devstat_tag_type                ds_tag_type;
 218         devstat_trans_flags             ds_trans_type;
 219         uint64_t                        io_len;
 220         uint64_t                        io_offset;
 221         int                             io_arg;
 222         struct ctl_be_block_softc       *softc;
 223         struct ctl_be_block_lun         *lun;
 224         void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */
 225 };
 226
 227 extern struct ctl_softc *control_softc;
 228
 229 static int cbb_num_threads = 14;
 230 SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
 231             "CAM Target Layer Block Backend");
 232 SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RWTUN,
 233            &cbb_num_threads, 0, "Number of threads per backing file");
 234
 235 static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc);
 236 static void ctl_free_beio(struct ctl_be_block_io *beio);
 237 static void ctl_complete_beio(struct ctl_be_block_io *beio);
 238 static int ctl_be_block_move_done(union ctl_io *io);
 239 static void ctl_be_block_biodone(struct bio *bio);
 240 static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun,
 241                                     struct ctl_be_block_io *beio);
 242 static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun,
 243                                        struct ctl_be_block_io *beio);
 244 static void ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun,
 245                                   struct ctl_be_block_io *beio);
 246 static uint64_t ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun,
 247                                          const char *attrname);
 248 static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun,
 249                                    struct ctl_be_block_io *beio);
 250 static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun,
 251                                    struct ctl_be_block_io *beio);
 252 static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun,
 253                                       struct ctl_be_block_io *beio);
 254 static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun,
 255                                          const char *attrname);
 256 static void ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun,
 257                                     union ctl_io *io);
 258 static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun,
 259                                     union ctl_io *io);
 260 static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun,
 261                                   union ctl_io *io);
 262 static void ctl_be_block_worker(void *context, int pending);
 263 static int ctl_be_block_submit(union ctl_io *io);
 264 static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
 265                                    int flag, struct thread *td);
 266 static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun,
 267                                   struct ctl_lun_req *req);
 268 static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun,
 269                                  struct ctl_lun_req *req);
 270 static int ctl_be_block_close(struct ctl_be_block_lun *be_lun);
 271 static int ctl_be_block_open(struct ctl_be_block_lun *be_lun,
 272                              struct ctl_lun_req *req);
 273 static int ctl_be_block_create(struct ctl_be_block_softc *softc,
 274                                struct ctl_lun_req *req);
 275 static int ctl_be_block_rm(struct ctl_be_block_softc *softc,
 276                            struct ctl_lun_req *req);
 277 static int ctl_be_block_modify(struct ctl_be_block_softc *softc,
 278                            struct ctl_lun_req *req);
 279 static void ctl_be_block_lun_shutdown(struct ctl_be_lun *cbe_lun);
 280 static int ctl_be_block_config_write(union ctl_io *io);
 281 static int ctl_be_block_config_read(union ctl_io *io);
 282 static int ctl_be_block_lun_info(struct ctl_be_lun *cbe_lun, struct sbuf *sb);
 283 static uint64_t ctl_be_block_lun_attr(struct ctl_be_lun *cbe_lun, const char *attrname);
 284 static int ctl_be_block_init(void);
 285 static int ctl_be_block_shutdown(void);
 286
 287 static struct ctl_backend_driver ctl_be_block_driver =
 288 {
 289         .name = "block",
 290         .flags = CTL_BE_FLAG_HAS_CONFIG,
 291         .init = ctl_be_block_init,
 292         .shutdown = ctl_be_block_shutdown,
 293         .data_submit = ctl_be_block_submit,
 294         .data_move_done = ctl_be_block_move_done,
 295         .config_read = ctl_be_block_config_read,
 296         .config_write = ctl_be_block_config_write,
 297         .ioctl = ctl_be_block_ioctl,
 298         .lun_info = ctl_be_block_lun_info,
 299         .lun_attr = ctl_be_block_lun_attr
 300 };
 301
 302 MALLOC_DEFINE(M_CTLBLK, "ctlblock", "Memory used for CTL block backend");
 303 CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver);
 304
 305 static void
 306 ctl_alloc_seg(struct ctl_be_block_softc *softc, struct ctl_sg_entry *sg,
 307     size_t len)
 308 {
 309
 310         if (len <= CTLBLK_MIN_SEG) {
 311                 sg->addr = uma_zalloc(softc->bufmin_zone, M_WAITOK);
 312         } else {
 313                 KASSERT(len <= CTLBLK_MAX_SEG,
 314                     ("Too large alloc %zu > %lu", len, CTLBLK_MAX_SEG));
 315                 sg->addr = uma_zalloc(softc->bufmax_zone, M_WAITOK);
 316         }
 317         sg->len = len;
 318 }
 319
 320 static void
 321 ctl_free_seg(struct ctl_be_block_softc *softc, struct ctl_sg_entry *sg)
 322 {
 323
 324         if (sg->len <= CTLBLK_MIN_SEG) {
 325                 uma_zfree(softc->bufmin_zone, sg->addr);
 326         } else {
 327                 KASSERT(sg->len <= CTLBLK_MAX_SEG,
 328                     ("Too large free %zu > %lu", sg->len, CTLBLK_MAX_SEG));
 329                 uma_zfree(softc->bufmax_zone, sg->addr);
 330         }
 331 }
 332
 333 static struct ctl_be_block_io *
 334 ctl_alloc_beio(struct ctl_be_block_softc *softc)
 335 {
 336         struct ctl_be_block_io *beio;
 337
 338         beio = uma_zalloc(softc->beio_zone, M_WAITOK | M_ZERO);
 339         beio->softc = softc;
 340         beio->refcnt = 1;
 341         return (beio);
 342 }
 343
 344 static void
 345 ctl_real_free_beio(struct ctl_be_block_io *beio)
 346 {
 347         struct ctl_be_block_softc *softc = beio->softc;
 348         int i;
 349
 350         for (i = 0; i < beio->num_segs; i++) {
 351                 ctl_free_seg(softc, &beio->sg_segs[i]);
 352
 353                 /* For compare we had two equal S/G lists. */
 354                 if (beio->two_sglists) {
 355                         ctl_free_seg(softc,
 356                             &beio->sg_segs[i + CTLBLK_HALF_SEGS]);
 357                 }
 358         }
 359
 360         uma_zfree(softc->beio_zone, beio);
 361 }
 362
 363 static void
 364 ctl_refcnt_beio(void *arg, int diff)
 365 {
 366         struct ctl_be_block_io *beio = arg;
 367
 368         if (atomic_fetchadd_int(&beio->refcnt, diff) + diff == 0)
 369                 ctl_real_free_beio(beio);
 370 }
 371
 372 static void
 373 ctl_free_beio(struct ctl_be_block_io *beio)
 374 {
 375
 376         ctl_refcnt_beio(beio, -1);
 377 }
 378
 379 static void
 380 ctl_complete_beio(struct ctl_be_block_io *beio)
 381 {
 382         union ctl_io *io = beio->io;
 383
 384         if (beio->beio_cont != NULL) {
 385                 beio->beio_cont(beio);
 386         } else {
 387                 ctl_free_beio(beio);
 388                 ctl_data_submit_done(io);
 389         }
 390 }
 391
 392 static size_t
 393 cmp(uint8_t *a, uint8_t *b, size_t size)
 394 {
 395         size_t i;
 396
 397         for (i = 0; i < size; i++) {
 398                 if (a[i] != b[i])
 399                         break;
 400         }
 401         return (i);
 402 }
 403
 404 static void
 405 ctl_be_block_compare(union ctl_io *io)
 406 {
 407         struct ctl_be_block_io *beio;
 408         uint64_t off, res;
 409         int i;
 410         uint8_t info[8];
 411
 412         beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
 413         off = 0;
 414         for (i = 0; i < beio->num_segs; i++) {
 415                 res = cmp(beio->sg_segs[i].addr,
 416                     beio->sg_segs[i + CTLBLK_HALF_SEGS].addr,
 417                     beio->sg_segs[i].len);
 418                 off += res;
 419                 if (res < beio->sg_segs[i].len)
 420                         break;
 421         }
 422         if (i < beio->num_segs) {
 423                 scsi_u64to8b(off, info);
 424                 ctl_set_sense(&io->scsiio, /*current_error*/ 1,
 425                     /*sense_key*/ SSD_KEY_MISCOMPARE,
 426                     /*asc*/ 0x1D, /*ascq*/ 0x00,
 427                     /*type*/ SSD_ELEM_INFO,
 428                     /*size*/ sizeof(info), /*data*/ &info,
 429                     /*type*/ SSD_ELEM_NONE);
 430         } else
 431                 ctl_set_success(&io->scsiio);
 432 }
 433
 434 static int
 435 ctl_be_block_move_done(union ctl_io *io)
 436 {
 437         struct ctl_be_block_io *beio;
 438         struct ctl_be_block_lun *be_lun;
 439         struct ctl_lba_len_flags *lbalen;
 440 #ifdef CTL_TIME_IO
 441         struct bintime cur_bt;
 442 #endif
 443
 444         beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
 445         be_lun = beio->lun;
 446
 447         DPRINTF("entered\n");
 448
 449 #ifdef CTL_TIME_IO
 450         getbinuptime(&cur_bt);
 451         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
 452         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
 453 #endif
 454         io->io_hdr.num_dmas++;
 455         io->scsiio.kern_rel_offset += io->scsiio.kern_data_len;
 456
 457         /*
 458          * We set status at this point for read commands, and write
 459          * commands with errors.
 460          */
 461         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
 462                 ;
 463         } else if ((io->io_hdr.port_status != 0) &&
 464             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
 465              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
 466                 ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1,
 467                     /*retry_count*/ io->io_hdr.port_status);
 468         } else if (io->scsiio.kern_data_resid != 0 &&
 469             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT &&
 470             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
 471              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
 472                 ctl_set_invalid_field_ciu(&io->scsiio);
 473         } else if ((io->io_hdr.port_status == 0) &&
 474             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) {
 475                 lbalen = ARGS(beio->io);
 476                 if (lbalen->flags & CTL_LLF_READ) {
 477                         ctl_set_success(&io->scsiio);
 478                 } else if (lbalen->flags & CTL_LLF_COMPARE) {
 479                         /* We have two data blocks ready for comparison. */
 480                         ctl_be_block_compare(io);
 481                 }
 482         }
 483
 484         /*
 485          * If this is a read, or a write with errors, it is done.
 486          */
 487         if ((beio->bio_cmd == BIO_READ)
 488          || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)
 489          || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) {
 490                 ctl_complete_beio(beio);
 491                 return (0);
 492         }
 493
 494         /*
 495          * At this point, we have a write and the DMA completed
 496          * successfully.  We now have to queue it to the task queue to
 497          * execute the backend I/O.  That is because we do blocking
 498          * memory allocations, and in the file backing case, blocking I/O.
 499          * This move done routine is generally called in the SIM's
 500          * interrupt context, and therefore we cannot block.
 501          */
 502         mtx_lock(&be_lun->queue_lock);
 503         STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links);
 504         mtx_unlock(&be_lun->queue_lock);
 505         taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
 506
 507         return (0);
 508 }
 509
 510 static void
 511 ctl_be_block_biodone(struct bio *bio)
 512 {
 513         struct ctl_be_block_io *beio;
 514         struct ctl_be_block_lun *be_lun;
 515         union ctl_io *io;
 516         int error;
 517
 518         beio = bio->bio_caller1;
 519         be_lun = beio->lun;
 520         io = beio->io;
 521
 522         DPRINTF("entered\n");
 523
 524         error = bio->bio_error;
 525         mtx_lock(&be_lun->io_lock);
 526         if (error != 0 &&
 527             (beio->first_error == 0 ||
 528              bio->bio_offset < beio->first_error_offset)) {
 529                 beio->first_error = error;
 530                 beio->first_error_offset = bio->bio_offset;
 531         }
 532
 533         beio->num_bios_done++;
 534
 535         /*
 536          * XXX KDM will this cause WITNESS to complain?  Holding a lock
 537          * during the free might cause it to complain.
 538          */
 539         g_destroy_bio(bio);
 540
 541         /*
 542          * If the send complete bit isn't set, or we aren't the last I/O to
 543          * complete, then we're done.
 544          */
 545         if ((beio->send_complete == 0)
 546          || (beio->num_bios_done < beio->num_bios_sent)) {
 547                 mtx_unlock(&be_lun->io_lock);
 548                 return;
 549         }
 550
 551         /*
 552          * At this point, we've verified that we are the last I/O to
 553          * complete, so it's safe to drop the lock.
 554          */
 555         devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
 556             beio->ds_tag_type, beio->ds_trans_type,
 557             /*now*/ NULL, /*then*/&beio->ds_t0);
 558         mtx_unlock(&be_lun->io_lock);
 559
 560         /*
 561          * If there are any errors from the backing device, we fail the
 562          * entire I/O with a medium error.
 563          */
 564         error = beio->first_error;
 565         if (error != 0) {
 566                 if (error == EOPNOTSUPP) {
 567                         ctl_set_invalid_opcode(&io->scsiio);
 568                 } else if (error == ENOSPC || error == EDQUOT) {
 569                         ctl_set_space_alloc_fail(&io->scsiio);
 570                 } else if (error == EROFS || error == EACCES) {
 571                         ctl_set_hw_write_protected(&io->scsiio);
 572                 } else if (beio->bio_cmd == BIO_FLUSH) {
 573                         /* XXX KDM is there is a better error here? */
 574                         ctl_set_internal_failure(&io->scsiio,
 575                                                  /*sks_valid*/ 1,
 576                                                  /*retry_count*/ 0xbad2);
 577                 } else {
 578                         ctl_set_medium_error(&io->scsiio,
 579                             beio->bio_cmd == BIO_READ);
 580                 }
 581                 ctl_complete_beio(beio);
 582                 return;
 583         }
 584
 585         /*
 586          * If this is a write, a flush, a delete or verify, we're all done.
 587          * If this is a read, we can now send the data to the user.
 588          */
 589         if ((beio->bio_cmd == BIO_WRITE)
 590          || (beio->bio_cmd == BIO_FLUSH)
 591          || (beio->bio_cmd == BIO_DELETE)
 592          || (ARGS(io)->flags & CTL_LLF_VERIFY)) {
 593                 ctl_set_success(&io->scsiio);
 594                 ctl_complete_beio(beio);
 595         } else {
 596                 if ((ARGS(io)->flags & CTL_LLF_READ) &&
 597                     beio->beio_cont == NULL) {
 598                         ctl_set_success(&io->scsiio);
 599                         ctl_serseq_done(io);
 600                 }
 601 #ifdef CTL_TIME_IO
 602                 getbinuptime(&io->io_hdr.dma_start_bt);
 603 #endif
 604                 ctl_datamove(io);
 605         }
 606 }
 607
 608 static void
 609 ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun,
 610                         struct ctl_be_block_io *beio)
 611 {
 612         union ctl_io *io = beio->io;
 613         struct mount *mountpoint;
 614         int error, lock_flags;
 615
 616         DPRINTF("entered\n");
 617
 618         binuptime(&beio->ds_t0);
 619         devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
 620
 621         (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT);
 622
 623         if (MNT_SHARED_WRITES(mountpoint) ||
 624             ((mountpoint == NULL) && MNT_SHARED_WRITES(be_lun->vn->v_mount)))
 625                 lock_flags = LK_SHARED;
 626         else
 627                 lock_flags = LK_EXCLUSIVE;
 628         vn_lock(be_lun->vn, lock_flags | LK_RETRY);
 629         error = VOP_FSYNC(be_lun->vn, beio->io_arg ? MNT_NOWAIT : MNT_WAIT,
 630             curthread);
 631         VOP_UNLOCK(be_lun->vn);
 632
 633         vn_finished_write(mountpoint);
 634
 635         mtx_lock(&be_lun->io_lock);
 636         devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
 637             beio->ds_tag_type, beio->ds_trans_type,
 638             /*now*/ NULL, /*then*/&beio->ds_t0);
 639         mtx_unlock(&be_lun->io_lock);
 640
 641         if (error == 0)
 642                 ctl_set_success(&io->scsiio);
 643         else {
 644                 /* XXX KDM is there is a better error here? */
 645                 ctl_set_internal_failure(&io->scsiio,
 646                                          /*sks_valid*/ 1,
 647                                          /*retry_count*/ 0xbad1);
 648         }
 649
 650         ctl_complete_beio(beio);
 651 }
 652
 653 SDT_PROBE_DEFINE1(cbb, , read, file_start, "uint64_t");
 654 SDT_PROBE_DEFINE1(cbb, , write, file_start, "uint64_t");
 655 SDT_PROBE_DEFINE1(cbb, , read, file_done,"uint64_t");
 656 SDT_PROBE_DEFINE1(cbb, , write, file_done, "uint64_t");
 657
 658 static void
 659 ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun,
 660                            struct ctl_be_block_io *beio)
 661 {
 662         struct ctl_be_block_filedata *file_data;
 663         union ctl_io *io;
 664         struct uio xuio;
 665         struct iovec *xiovec;
 666         size_t s;
 667         int error, flags, i;
 668
 669         DPRINTF("entered\n");
 670
 671         file_data = &be_lun->backend.file;
 672         io = beio->io;
 673         flags = 0;
 674         if (ARGS(io)->flags & CTL_LLF_DPO)
 675                 flags |= IO_DIRECT;
 676         if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA)
 677                 flags |= IO_SYNC;
 678
 679         bzero(&xuio, sizeof(xuio));
 680         if (beio->bio_cmd == BIO_READ) {
 681                 SDT_PROBE0(cbb, , read, file_start);
 682                 xuio.uio_rw = UIO_READ;
 683         } else {
 684                 SDT_PROBE0(cbb, , write, file_start);
 685                 xuio.uio_rw = UIO_WRITE;
 686         }
 687         xuio.uio_offset = beio->io_offset;
 688         xuio.uio_resid = beio->io_len;
 689         xuio.uio_segflg = UIO_SYSSPACE;
 690         xuio.uio_iov = beio->xiovecs;
 691         xuio.uio_iovcnt = beio->num_segs;
 692         xuio.uio_td = curthread;
 693
 694         for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) {
 695                 xiovec->iov_base = beio->sg_segs[i].addr;
 696                 xiovec->iov_len = beio->sg_segs[i].len;
 697         }
 698
 699         binuptime(&beio->ds_t0);
 700         devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
 701
 702         if (beio->bio_cmd == BIO_READ) {
 703                 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
 704
 705                 /*
 706                  * UFS pays attention to IO_DIRECT for reads.  If the
 707                  * DIRECTIO option is configured into the kernel, it calls
 708                  * ffs_rawread().  But that only works for single-segment
 709                  * uios with user space addresses.  In our case, with a
 710                  * kernel uio, it still reads into the buffer cache, but it
 711                  * will just try to release the buffer from the cache later
 712                  * on in ffs_read().
 713                  *
 714                  * ZFS does not pay attention to IO_DIRECT for reads.
 715                  *
 716                  * UFS does not pay attention to IO_SYNC for reads.
 717                  *
 718                  * ZFS pays attention to IO_SYNC (which translates into the
 719                  * Solaris define FRSYNC for zfs_read()) for reads.  It
 720                  * attempts to sync the file before reading.
 721                  */
 722                 error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred);
 723
 724                 VOP_UNLOCK(be_lun->vn);
 725                 SDT_PROBE0(cbb, , read, file_done);
 726                 if (error == 0 && xuio.uio_resid > 0) {
 727                         /*
 728                          * If we red less then requested (EOF), then
 729                          * we should clean the rest of the buffer.
 730                          */
 731                         s = beio->io_len - xuio.uio_resid;
 732                         for (i = 0; i < beio->num_segs; i++) {
 733                                 if (s >= beio->sg_segs[i].len) {
 734                                         s -= beio->sg_segs[i].len;
 735                                         continue;
 736                                 }
 737                                 bzero((uint8_t *)beio->sg_segs[i].addr + s,
 738                                     beio->sg_segs[i].len - s);
 739                                 s = 0;
 740                         }
 741                 }
 742         } else {
 743                 struct mount *mountpoint;
 744                 int lock_flags;
 745
 746                 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT);
 747
 748                 if (MNT_SHARED_WRITES(mountpoint) || ((mountpoint == NULL)
 749                   && MNT_SHARED_WRITES(be_lun->vn->v_mount)))
 750                         lock_flags = LK_SHARED;
 751                 else
 752                         lock_flags = LK_EXCLUSIVE;
 753                 vn_lock(be_lun->vn, lock_flags | LK_RETRY);
 754
 755                 /*
 756                  * UFS pays attention to IO_DIRECT for writes.  The write
 757                  * is done asynchronously.  (Normally the write would just
 758                  * get put into cache.
 759                  *
 760                  * UFS pays attention to IO_SYNC for writes.  It will
 761                  * attempt to write the buffer out synchronously if that
 762                  * flag is set.
 763                  *
 764                  * ZFS does not pay attention to IO_DIRECT for writes.
 765                  *
 766                  * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC)
 767                  * for writes.  It will flush the transaction from the
 768                  * cache before returning.
 769                  */
 770                 error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred);
 771                 VOP_UNLOCK(be_lun->vn);
 772
 773                 vn_finished_write(mountpoint);
 774                 SDT_PROBE0(cbb, , write, file_done);
 775         }
 776
 777         mtx_lock(&be_lun->io_lock);
 778         devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
 779             beio->ds_tag_type, beio->ds_trans_type,
 780             /*now*/ NULL, /*then*/&beio->ds_t0);
 781         mtx_unlock(&be_lun->io_lock);
 782
 783         /*
 784          * If we got an error, set the sense data to "MEDIUM ERROR" and
 785          * return the I/O to the user.
 786          */
 787         if (error != 0) {
 788                 if (error == ENOSPC || error == EDQUOT) {
 789                         ctl_set_space_alloc_fail(&io->scsiio);
 790                 } else if (error == EROFS || error == EACCES) {
 791                         ctl_set_hw_write_protected(&io->scsiio);
 792                 } else {
 793                         ctl_set_medium_error(&io->scsiio,
 794                             beio->bio_cmd == BIO_READ);
 795                 }
 796                 ctl_complete_beio(beio);
 797                 return;
 798         }
 799
 800         /*
 801          * If this is a write or a verify, we're all done.
 802          * If this is a read, we can now send the data to the user.
 803          */
 804         if ((beio->bio_cmd == BIO_WRITE) ||
 805             (ARGS(io)->flags & CTL_LLF_VERIFY)) {
 806                 ctl_set_success(&io->scsiio);
 807                 ctl_complete_beio(beio);
 808         } else {
 809                 if ((ARGS(io)->flags & CTL_LLF_READ) &&
 810                     beio->beio_cont == NULL) {
 811                         ctl_set_success(&io->scsiio);
 812                         ctl_serseq_done(io);
 813                 }
 814 #ifdef CTL_TIME_IO
 815                 getbinuptime(&io->io_hdr.dma_start_bt);
 816 #endif
 817                 ctl_datamove(io);
 818         }
 819 }
 820
 821 static void
 822 ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun,
 823                         struct ctl_be_block_io *beio)
 824 {
 825         union ctl_io *io = beio->io;
 826         struct ctl_lba_len_flags *lbalen = ARGS(io);
 827         struct scsi_get_lba_status_data *data;
 828         off_t roff, off;
 829         int error, status;
 830
 831         DPRINTF("entered\n");
 832
 833         off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize;
 834         vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
 835         error = VOP_IOCTL(be_lun->vn, FIOSEEKHOLE, &off,
 836             0, curthread->td_ucred, curthread);
 837         if (error == 0 && off > roff)
 838                 status = 0;     /* mapped up to off */
 839         else {
 840                 error = VOP_IOCTL(be_lun->vn, FIOSEEKDATA, &off,
 841                     0, curthread->td_ucred, curthread);
 842                 if (error == 0 && off > roff)
 843                         status = 1;     /* deallocated up to off */
 844                 else {
 845                         status = 0;     /* unknown up to the end */
 846                         off = be_lun->size_bytes;
 847                 }
 848         }
 849         VOP_UNLOCK(be_lun->vn);
 850
 851         data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr;
 852         scsi_u64to8b(lbalen->lba, data->descr[0].addr);
 853         scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize -
 854             lbalen->lba), data->descr[0].length);
 855         data->descr[0].status = status;
 856
 857         ctl_complete_beio(beio);
 858 }
 859
 860 static uint64_t
 861 ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, const char *attrname)
 862 {
 863         struct vattr            vattr;
 864         struct statfs           statfs;
 865         uint64_t                val;
 866         int                     error;
 867
 868         val = UINT64_MAX;
 869         if (be_lun->vn == NULL)
 870                 return (val);
 871         vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
 872         if (strcmp(attrname, "blocksused") == 0) {
 873                 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred);
 874                 if (error == 0)
 875                         val = vattr.va_bytes / be_lun->cbe_lun.blocksize;
 876         }
 877         if (strcmp(attrname, "blocksavail") == 0 &&
 878             !VN_IS_DOOMED(be_lun->vn)) {
 879                 error = VFS_STATFS(be_lun->vn->v_mount, &statfs);
 880                 if (error == 0)
 881                         val = statfs.f_bavail * statfs.f_bsize /
 882                             be_lun->cbe_lun.blocksize;
 883         }
 884         VOP_UNLOCK(be_lun->vn);
 885         return (val);
 886 }
 887
 888 static void
 889 ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun,
 890                            struct ctl_be_block_io *beio)
 891 {
 892         union ctl_io *io;
 893         struct cdevsw *csw;
 894         struct cdev *dev;
 895         struct uio xuio;
 896         struct iovec *xiovec;
 897         int error, flags, i, ref;
 898
 899         DPRINTF("entered\n");
 900
 901         io = beio->io;
 902         flags = 0;
 903         if (ARGS(io)->flags & CTL_LLF_DPO)
 904                 flags |= IO_DIRECT;
 905         if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA)
 906                 flags |= IO_SYNC;
 907
 908         bzero(&xuio, sizeof(xuio));
 909         if (beio->bio_cmd == BIO_READ) {
 910                 SDT_PROBE0(cbb, , read, file_start);
 911                 xuio.uio_rw = UIO_READ;
 912         } else {
 913                 SDT_PROBE0(cbb, , write, file_start);
 914                 xuio.uio_rw = UIO_WRITE;
 915         }
 916         xuio.uio_offset = beio->io_offset;
 917         xuio.uio_resid = beio->io_len;
 918         xuio.uio_segflg = UIO_SYSSPACE;
 919         xuio.uio_iov = beio->xiovecs;
 920         xuio.uio_iovcnt = beio->num_segs;
 921         xuio.uio_td = curthread;
 922
 923         for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) {
 924                 xiovec->iov_base = beio->sg_segs[i].addr;
 925                 xiovec->iov_len = beio->sg_segs[i].len;
 926         }
 927
 928         binuptime(&beio->ds_t0);
 929         devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
 930
 931         csw = devvn_refthread(be_lun->vn, &dev, &ref);
 932         if (csw) {
 933                 if (beio->bio_cmd == BIO_READ)
 934                         error = csw->d_read(dev, &xuio, flags);
 935                 else
 936                         error = csw->d_write(dev, &xuio, flags);
 937                 dev_relthread(dev, ref);
 938         } else
 939                 error = ENXIO;
 940
 941         if (beio->bio_cmd == BIO_READ)
 942                 SDT_PROBE0(cbb, , read, file_done);
 943         else
 944                 SDT_PROBE0(cbb, , write, file_done);
 945
 946         mtx_lock(&be_lun->io_lock);
 947         devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
 948             beio->ds_tag_type, beio->ds_trans_type,
 949             /*now*/ NULL, /*then*/&beio->ds_t0);
 950         mtx_unlock(&be_lun->io_lock);
 951
 952         /*
 953          * If we got an error, set the sense data to "MEDIUM ERROR" and
 954          * return the I/O to the user.
 955          */
 956         if (error != 0) {
 957                 if (error == ENOSPC || error == EDQUOT) {
 958                         ctl_set_space_alloc_fail(&io->scsiio);
 959                 } else if (error == EROFS || error == EACCES) {
 960                         ctl_set_hw_write_protected(&io->scsiio);
 961                 } else {
 962                         ctl_set_medium_error(&io->scsiio,
 963                             beio->bio_cmd == BIO_READ);
 964                 }
 965                 ctl_complete_beio(beio);
 966                 return;
 967         }
 968
 969         /*
 970          * If this is a write or a verify, we're all done.
 971          * If this is a read, we can now send the data to the user.
 972          */
 973         if ((beio->bio_cmd == BIO_WRITE) ||
 974             (ARGS(io)->flags & CTL_LLF_VERIFY)) {
 975                 ctl_set_success(&io->scsiio);
 976                 ctl_complete_beio(beio);
 977         } else {
 978                 if ((ARGS(io)->flags & CTL_LLF_READ) &&
 979                     beio->beio_cont == NULL) {
 980                         ctl_set_success(&io->scsiio);
 981                         ctl_serseq_done(io);
 982                 }
 983 #ifdef CTL_TIME_IO
 984                 getbinuptime(&io->io_hdr.dma_start_bt);
 985 #endif
 986                 ctl_datamove(io);
 987         }
 988 }
 989
 990 static void
 991 ctl_be_block_gls_zvol(struct ctl_be_block_lun *be_lun,
 992                         struct ctl_be_block_io *beio)
 993 {
 994         union ctl_io *io = beio->io;
 995         struct cdevsw *csw;
 996         struct cdev *dev;
 997         struct ctl_lba_len_flags *lbalen = ARGS(io);
 998         struct scsi_get_lba_status_data *data;
 999         off_t roff, off;
1000         int error, ref, status;
1001
1002         DPRINTF("entered\n");
1003
1004         csw = devvn_refthread(be_lun->vn, &dev, &ref);
1005         if (csw == NULL) {
1006                 status = 0;     /* unknown up to the end */
1007                 off = be_lun->size_bytes;
1008                 goto done;
1009         }
1010         off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize;
1011         error = csw->d_ioctl(dev, FIOSEEKHOLE, (caddr_t)&off, FREAD,
1012             curthread);
1013         if (error == 0 && off > roff)
1014                 status = 0;     /* mapped up to off */
1015         else {
1016                 error = csw->d_ioctl(dev, FIOSEEKDATA, (caddr_t)&off, FREAD,
1017                     curthread);
1018                 if (error == 0 && off > roff)
1019                         status = 1;     /* deallocated up to off */
1020                 else {
1021                         status = 0;     /* unknown up to the end */
1022                         off = be_lun->size_bytes;
1023                 }
1024         }
1025         dev_relthread(dev, ref);
1026
1027 done:
1028         data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr;
1029         scsi_u64to8b(lbalen->lba, data->descr[0].addr);
1030         scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize -
1031             lbalen->lba), data->descr[0].length);
1032         data->descr[0].status = status;
1033
1034         ctl_complete_beio(beio);
1035 }
1036
1037 static void
1038 ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun,
1039                        struct ctl_be_block_io *beio)
1040 {
1041         struct bio *bio;
1042         struct cdevsw *csw;
1043         struct cdev *dev;
1044         int ref;
1045
1046         DPRINTF("entered\n");
1047
1048         /* This can't fail, it's a blocking allocation. */
1049         bio = g_alloc_bio();
1050
1051         bio->bio_cmd        = BIO_FLUSH;
1052         bio->bio_offset     = 0;
1053         bio->bio_data       = 0;
1054         bio->bio_done       = ctl_be_block_biodone;
1055         bio->bio_caller1    = beio;
1056         bio->bio_pblkno     = 0;
1057
1058         /*
1059          * We don't need to acquire the LUN lock here, because we are only
1060          * sending one bio, and so there is no other context to synchronize
1061          * with.
1062          */
1063         beio->num_bios_sent = 1;
1064         beio->send_complete = 1;
1065
1066         binuptime(&beio->ds_t0);
1067         devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1068
1069         csw = devvn_refthread(be_lun->vn, &dev, &ref);
1070         if (csw) {
1071                 bio->bio_dev = dev;
1072                 csw->d_strategy(bio);
1073                 dev_relthread(dev, ref);
1074         } else {
1075                 bio->bio_error = ENXIO;
1076                 ctl_be_block_biodone(bio);
1077         }
1078 }
1079
1080 static void
1081 ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun,
1082                        struct ctl_be_block_io *beio,
1083                        uint64_t off, uint64_t len, int last)
1084 {
1085         struct bio *bio;
1086         uint64_t maxlen;
1087         struct cdevsw *csw;
1088         struct cdev *dev;
1089         int ref;
1090
1091         csw = devvn_refthread(be_lun->vn, &dev, &ref);
1092         maxlen = LONG_MAX - (LONG_MAX % be_lun->cbe_lun.blocksize);
1093         while (len > 0) {
1094                 bio = g_alloc_bio();
1095                 bio->bio_cmd        = BIO_DELETE;
1096                 bio->bio_dev        = dev;
1097                 bio->bio_offset     = off;
1098                 bio->bio_length     = MIN(len, maxlen);
1099                 bio->bio_data       = 0;
1100                 bio->bio_done       = ctl_be_block_biodone;
1101                 bio->bio_caller1    = beio;
1102                 bio->bio_pblkno     = off / be_lun->cbe_lun.blocksize;
1103
1104                 off += bio->bio_length;
1105                 len -= bio->bio_length;
1106
1107                 mtx_lock(&be_lun->io_lock);
1108                 beio->num_bios_sent++;
1109                 if (last && len == 0)
1110                         beio->send_complete = 1;
1111                 mtx_unlock(&be_lun->io_lock);
1112
1113                 if (csw) {
1114                         csw->d_strategy(bio);
1115                 } else {
1116                         bio->bio_error = ENXIO;
1117                         ctl_be_block_biodone(bio);
1118                 }
1119         }
1120         if (csw)
1121                 dev_relthread(dev, ref);
1122 }
1123
1124 static void
1125 ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun,
1126                        struct ctl_be_block_io *beio)
1127 {
1128         union ctl_io *io;
1129         struct ctl_ptr_len_flags *ptrlen;
1130         struct scsi_unmap_desc *buf, *end;
1131         uint64_t len;
1132
1133         io = beio->io;
1134
1135         DPRINTF("entered\n");
1136
1137         binuptime(&beio->ds_t0);
1138         devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1139
1140         if (beio->io_offset == -1) {
1141                 beio->io_len = 0;
1142                 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1143                 buf = (struct scsi_unmap_desc *)ptrlen->ptr;
1144                 end = buf + ptrlen->len / sizeof(*buf);
1145                 for (; buf < end; buf++) {
1146                         len = (uint64_t)scsi_4btoul(buf->length) *
1147                             be_lun->cbe_lun.blocksize;
1148                         beio->io_len += len;
1149                         ctl_be_block_unmap_dev_range(be_lun, beio,
1150                             scsi_8btou64(buf->lba) * be_lun->cbe_lun.blocksize,
1151                             len, (end - buf < 2) ? TRUE : FALSE);
1152                 }
1153         } else
1154                 ctl_be_block_unmap_dev_range(be_lun, beio,
1155                     beio->io_offset, beio->io_len, TRUE);
1156 }
1157
1158 static void
1159 ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun,
1160                           struct ctl_be_block_io *beio)
1161 {
1162         TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue);
1163         struct bio *bio;
1164         struct cdevsw *csw;
1165         struct cdev *dev;
1166         off_t cur_offset;
1167         int i, max_iosize, ref;
1168
1169         DPRINTF("entered\n");
1170         csw = devvn_refthread(be_lun->vn, &dev, &ref);
1171
1172         /*
1173          * We have to limit our I/O size to the maximum supported by the
1174          * backend device.
1175          */
1176         if (csw) {
1177                 max_iosize = dev->si_iosize_max;
1178                 if (max_iosize < PAGE_SIZE)
1179                         max_iosize = DFLTPHYS;
1180         } else
1181                 max_iosize = DFLTPHYS;
1182
1183         cur_offset = beio->io_offset;
1184         for (i = 0; i < beio->num_segs; i++) {
1185                 size_t cur_size;
1186                 uint8_t *cur_ptr;
1187
1188                 cur_size = beio->sg_segs[i].len;
1189                 cur_ptr = beio->sg_segs[i].addr;
1190
1191                 while (cur_size > 0) {
1192                         /* This can't fail, it's a blocking allocation. */
1193                         bio = g_alloc_bio();
1194
1195                         KASSERT(bio != NULL, ("g_alloc_bio() failed!\n"));
1196
1197                         bio->bio_cmd = beio->bio_cmd;
1198                         bio->bio_dev = dev;
1199                         bio->bio_caller1 = beio;
1200                         bio->bio_length = min(cur_size, max_iosize);
1201                         bio->bio_offset = cur_offset;
1202                         bio->bio_data = cur_ptr;
1203                         bio->bio_done = ctl_be_block_biodone;
1204                         bio->bio_pblkno = cur_offset / be_lun->cbe_lun.blocksize;
1205
1206                         cur_offset += bio->bio_length;
1207                         cur_ptr += bio->bio_length;
1208                         cur_size -= bio->bio_length;
1209
1210                         TAILQ_INSERT_TAIL(&queue, bio, bio_queue);
1211                         beio->num_bios_sent++;
1212                 }
1213         }
1214         beio->send_complete = 1;
1215         binuptime(&beio->ds_t0);
1216         devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1217
1218         /*
1219          * Fire off all allocated requests!
1220          */
1221         while ((bio = TAILQ_FIRST(&queue)) != NULL) {
1222                 TAILQ_REMOVE(&queue, bio, bio_queue);
1223                 if (csw)
1224                         csw->d_strategy(bio);
1225                 else {
1226                         bio->bio_error = ENXIO;
1227                         ctl_be_block_biodone(bio);
1228                 }
1229         }
1230         if (csw)
1231                 dev_relthread(dev, ref);
1232 }
1233
1234 static uint64_t
1235 ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname)
1236 {
1237         struct diocgattr_arg    arg;
1238         struct cdevsw *csw;
1239         struct cdev *dev;
1240         int error, ref;
1241
1242         csw = devvn_refthread(be_lun->vn, &dev, &ref);
1243         if (csw == NULL)
1244                 return (UINT64_MAX);
1245         strlcpy(arg.name, attrname, sizeof(arg.name));
1246         arg.len = sizeof(arg.value.off);
1247         if (csw->d_ioctl) {
1248                 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD,
1249                     curthread);
1250         } else
1251                 error = ENODEV;
1252         dev_relthread(dev, ref);
1253         if (error != 0)
1254                 return (UINT64_MAX);
1255         return (arg.value.off);
1256 }
1257
1258 static void
1259 ctl_be_block_cw_dispatch_sync(struct ctl_be_block_lun *be_lun,
1260                             union ctl_io *io)
1261 {
1262         struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1263         struct ctl_be_block_io *beio;
1264         struct ctl_lba_len_flags *lbalen;
1265
1266         DPRINTF("entered\n");
1267         beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1268         lbalen = (struct ctl_lba_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1269
1270         beio->io_len = lbalen->len * cbe_lun->blocksize;
1271         beio->io_offset = lbalen->lba * cbe_lun->blocksize;
1272         beio->io_arg = (lbalen->flags & SSC_IMMED) != 0;
1273         beio->bio_cmd = BIO_FLUSH;
1274         beio->ds_trans_type = DEVSTAT_NO_DATA;
1275         DPRINTF("SYNC\n");
1276         be_lun->lun_flush(be_lun, beio);
1277 }
1278
1279 static void
1280 ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio)
1281 {
1282         union ctl_io *io;
1283
1284         io = beio->io;
1285         ctl_free_beio(beio);
1286         if ((io->io_hdr.flags & CTL_FLAG_ABORT) ||
1287             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
1288              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
1289                 ctl_config_write_done(io);
1290                 return;
1291         }
1292
1293         ctl_be_block_config_write(io);
1294 }
1295
1296 static void
1297 ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun,
1298                             union ctl_io *io)
1299 {
1300         struct ctl_be_block_softc *softc = be_lun->softc;
1301         struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1302         struct ctl_be_block_io *beio;
1303         struct ctl_lba_len_flags *lbalen;
1304         uint64_t len_left, lba;
1305         uint32_t pb, pbo, adj;
1306         int i, seglen;
1307         uint8_t *buf, *end;
1308
1309         DPRINTF("entered\n");
1310
1311         beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1312         lbalen = ARGS(beio->io);
1313
1314         if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) ||
1315             (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) {
1316                 ctl_free_beio(beio);
1317                 ctl_set_invalid_field(&io->scsiio,
1318                                       /*sks_valid*/ 1,
1319                                       /*command*/ 1,
1320                                       /*field*/ 1,
1321                                       /*bit_valid*/ 0,
1322                                       /*bit*/ 0);
1323                 ctl_config_write_done(io);
1324                 return;
1325         }
1326
1327         if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) {
1328                 beio->io_offset = lbalen->lba * cbe_lun->blocksize;
1329                 beio->io_len = (uint64_t)lbalen->len * cbe_lun->blocksize;
1330                 beio->bio_cmd = BIO_DELETE;
1331                 beio->ds_trans_type = DEVSTAT_FREE;
1332
1333                 be_lun->unmap(be_lun, beio);
1334                 return;
1335         }
1336
1337         beio->bio_cmd = BIO_WRITE;
1338         beio->ds_trans_type = DEVSTAT_WRITE;
1339
1340         DPRINTF("WRITE SAME at LBA %jx len %u\n",
1341                (uintmax_t)lbalen->lba, lbalen->len);
1342
1343         pb = cbe_lun->blocksize << be_lun->cbe_lun.pblockexp;
1344         if (be_lun->cbe_lun.pblockoff > 0)
1345                 pbo = pb - cbe_lun->blocksize * be_lun->cbe_lun.pblockoff;
1346         else
1347                 pbo = 0;
1348         len_left = (uint64_t)lbalen->len * cbe_lun->blocksize;
1349         for (i = 0, lba = 0; i < CTLBLK_NUM_SEGS && len_left > 0; i++) {
1350                 /*
1351                  * Setup the S/G entry for this chunk.
1352                  */
1353                 seglen = MIN(CTLBLK_MAX_SEG, len_left);
1354                 if (pb > cbe_lun->blocksize) {
1355                         adj = ((lbalen->lba + lba) * cbe_lun->blocksize +
1356                             seglen - pbo) % pb;
1357                         if (seglen > adj)
1358                                 seglen -= adj;
1359                         else
1360                                 seglen -= seglen % cbe_lun->blocksize;
1361                 } else
1362                         seglen -= seglen % cbe_lun->blocksize;
1363                 ctl_alloc_seg(softc, &beio->sg_segs[i], seglen);
1364
1365                 DPRINTF("segment %d addr %p len %zd\n", i,
1366                         beio->sg_segs[i].addr, beio->sg_segs[i].len);
1367
1368                 beio->num_segs++;
1369                 len_left -= seglen;
1370
1371                 buf = beio->sg_segs[i].addr;
1372                 end = buf + seglen;
1373                 for (; buf < end; buf += cbe_lun->blocksize) {
1374                         if (lbalen->flags & SWS_NDOB) {
1375                                 memset(buf, 0, cbe_lun->blocksize);
1376                         } else {
1377                                 memcpy(buf, io->scsiio.kern_data_ptr,
1378                                     cbe_lun->blocksize);
1379                         }
1380                         if (lbalen->flags & SWS_LBDATA)
1381                                 scsi_ulto4b(lbalen->lba + lba, buf);
1382                         lba++;
1383                 }
1384         }
1385
1386         beio->io_offset = lbalen->lba * cbe_lun->blocksize;
1387         beio->io_len = lba * cbe_lun->blocksize;
1388
1389         /* We can not do all in one run. Correct and schedule rerun. */
1390         if (len_left > 0) {
1391                 lbalen->lba += lba;
1392                 lbalen->len -= lba;
1393                 beio->beio_cont = ctl_be_block_cw_done_ws;
1394         }
1395
1396         be_lun->dispatch(be_lun, beio);
1397 }
1398
1399 static void
1400 ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun,
1401                             union ctl_io *io)
1402 {
1403         struct ctl_be_block_io *beio;
1404         struct ctl_ptr_len_flags *ptrlen;
1405
1406         DPRINTF("entered\n");
1407
1408         beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1409         ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1410
1411         if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) {
1412                 ctl_free_beio(beio);
1413                 ctl_set_invalid_field(&io->scsiio,
1414                                       /*sks_valid*/ 0,
1415                                       /*command*/ 1,
1416                                       /*field*/ 0,
1417                                       /*bit_valid*/ 0,
1418                                       /*bit*/ 0);
1419                 ctl_config_write_done(io);
1420                 return;
1421         }
1422
1423         beio->io_len = 0;
1424         beio->io_offset = -1;
1425         beio->bio_cmd = BIO_DELETE;
1426         beio->ds_trans_type = DEVSTAT_FREE;
1427         DPRINTF("UNMAP\n");
1428         be_lun->unmap(be_lun, beio);
1429 }
1430
1431 static void
1432 ctl_be_block_cr_done(struct ctl_be_block_io *beio)
1433 {
1434         union ctl_io *io;
1435
1436         io = beio->io;
1437         ctl_free_beio(beio);
1438         ctl_config_read_done(io);
1439 }
1440
1441 static void
1442 ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun,
1443                          union ctl_io *io)
1444 {
1445         struct ctl_be_block_io *beio;
1446         struct ctl_be_block_softc *softc;
1447
1448         DPRINTF("entered\n");
1449
1450         softc = be_lun->softc;
1451         beio = ctl_alloc_beio(softc);
1452         beio->io = io;
1453         beio->lun = be_lun;
1454         beio->beio_cont = ctl_be_block_cr_done;
1455         PRIV(io)->ptr = (void *)beio;
1456
1457         switch (io->scsiio.cdb[0]) {
1458         case SERVICE_ACTION_IN:         /* GET LBA STATUS */
1459                 beio->bio_cmd = -1;
1460                 beio->ds_trans_type = DEVSTAT_NO_DATA;
1461                 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1462                 beio->io_len = 0;
1463                 if (be_lun->get_lba_status)
1464                         be_lun->get_lba_status(be_lun, beio);
1465                 else
1466                         ctl_be_block_cr_done(beio);
1467                 break;
1468         default:
1469                 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]);
1470                 break;
1471         }
1472 }
1473
1474 static void
1475 ctl_be_block_cw_done(struct ctl_be_block_io *beio)
1476 {
1477         union ctl_io *io;
1478
1479         io = beio->io;
1480         ctl_free_beio(beio);
1481         ctl_config_write_done(io);
1482 }
1483
1484 static void
1485 ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun,
1486                          union ctl_io *io)
1487 {
1488         struct ctl_be_block_io *beio;
1489         struct ctl_be_block_softc *softc;
1490
1491         DPRINTF("entered\n");
1492
1493         softc = be_lun->softc;
1494         beio = ctl_alloc_beio(softc);
1495         beio->io = io;
1496         beio->lun = be_lun;
1497         beio->beio_cont = ctl_be_block_cw_done;
1498         switch (io->scsiio.tag_type) {
1499         case CTL_TAG_ORDERED:
1500                 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1501                 break;
1502         case CTL_TAG_HEAD_OF_QUEUE:
1503                 beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1504                 break;
1505         case CTL_TAG_UNTAGGED:
1506         case CTL_TAG_SIMPLE:
1507         case CTL_TAG_ACA:
1508         default:
1509                 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1510                 break;
1511         }
1512         PRIV(io)->ptr = (void *)beio;
1513
1514         switch (io->scsiio.cdb[0]) {
1515         case SYNCHRONIZE_CACHE:
1516         case SYNCHRONIZE_CACHE_16:
1517                 ctl_be_block_cw_dispatch_sync(be_lun, io);
1518                 break;
1519         case WRITE_SAME_10:
1520         case WRITE_SAME_16:
1521                 ctl_be_block_cw_dispatch_ws(be_lun, io);
1522                 break;
1523         case UNMAP:
1524                 ctl_be_block_cw_dispatch_unmap(be_lun, io);
1525                 break;
1526         default:
1527                 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]);
1528                 break;
1529         }
1530 }
1531
1532 SDT_PROBE_DEFINE1(cbb, , read, start, "uint64_t");
1533 SDT_PROBE_DEFINE1(cbb, , write, start, "uint64_t");
1534 SDT_PROBE_DEFINE1(cbb, , read, alloc_done, "uint64_t");
1535 SDT_PROBE_DEFINE1(cbb, , write, alloc_done, "uint64_t");
1536
1537 static void
1538 ctl_be_block_next(struct ctl_be_block_io *beio)
1539 {
1540         struct ctl_be_block_lun *be_lun;
1541         union ctl_io *io;
1542
1543         io = beio->io;
1544         be_lun = beio->lun;
1545         ctl_free_beio(beio);
1546         if ((io->io_hdr.flags & CTL_FLAG_ABORT) ||
1547             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
1548              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
1549                 ctl_data_submit_done(io);
1550                 return;
1551         }
1552
1553         io->io_hdr.status &= ~CTL_STATUS_MASK;
1554         io->io_hdr.status |= CTL_STATUS_NONE;
1555
1556         mtx_lock(&be_lun->queue_lock);
1557         STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links);
1558         mtx_unlock(&be_lun->queue_lock);
1559         taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
1560 }
1561
1562 static void
1563 ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun,
1564                            union ctl_io *io)
1565 {
1566         struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1567         struct ctl_be_block_io *beio;
1568         struct ctl_be_block_softc *softc;
1569         struct ctl_lba_len_flags *lbalen;
1570         struct ctl_ptr_len_flags *bptrlen;
1571         uint64_t len_left, lbas;
1572         int i;
1573
1574         softc = be_lun->softc;
1575
1576         DPRINTF("entered\n");
1577
1578         lbalen = ARGS(io);
1579         if (lbalen->flags & CTL_LLF_WRITE) {
1580                 SDT_PROBE0(cbb, , write, start);
1581         } else {
1582                 SDT_PROBE0(cbb, , read, start);
1583         }
1584
1585         beio = ctl_alloc_beio(softc);
1586         beio->io = io;
1587         beio->lun = be_lun;
1588         bptrlen = PRIV(io);
1589         bptrlen->ptr = (void *)beio;
1590
1591         switch (io->scsiio.tag_type) {
1592         case CTL_TAG_ORDERED:
1593                 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1594                 break;
1595         case CTL_TAG_HEAD_OF_QUEUE:
1596                 beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1597                 break;
1598         case CTL_TAG_UNTAGGED:
1599         case CTL_TAG_SIMPLE:
1600         case CTL_TAG_ACA:
1601         default:
1602                 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1603                 break;
1604         }
1605
1606         if (lbalen->flags & CTL_LLF_WRITE) {
1607                 beio->bio_cmd = BIO_WRITE;
1608                 beio->ds_trans_type = DEVSTAT_WRITE;
1609         } else {
1610                 beio->bio_cmd = BIO_READ;
1611                 beio->ds_trans_type = DEVSTAT_READ;
1612         }
1613
1614         DPRINTF("%s at LBA %jx len %u @%ju\n",
1615                (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE",
1616                (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len);
1617         if (lbalen->flags & CTL_LLF_COMPARE) {
1618                 beio->two_sglists = 1;
1619                 lbas = CTLBLK_HALF_IO_SIZE;
1620         } else {
1621                 lbas = CTLBLK_MAX_IO_SIZE;
1622         }
1623         lbas = MIN(lbalen->len - bptrlen->len, lbas / cbe_lun->blocksize);
1624         beio->io_offset = (lbalen->lba + bptrlen->len) * cbe_lun->blocksize;
1625         beio->io_len = lbas * cbe_lun->blocksize;
1626         bptrlen->len += lbas;
1627
1628         for (i = 0, len_left = beio->io_len; len_left > 0; i++) {
1629                 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)",
1630                     i, CTLBLK_MAX_SEGS));
1631
1632                 /*
1633                  * Setup the S/G entry for this chunk.
1634                  */
1635                 ctl_alloc_seg(softc, &beio->sg_segs[i],
1636                     MIN(CTLBLK_MAX_SEG, len_left));
1637
1638                 DPRINTF("segment %d addr %p len %zd\n", i,
1639                         beio->sg_segs[i].addr, beio->sg_segs[i].len);
1640
1641                 /* Set up second segment for compare operation. */
1642                 if (beio->two_sglists) {
1643                         ctl_alloc_seg(softc,
1644                             &beio->sg_segs[i + CTLBLK_HALF_SEGS],
1645                             beio->sg_segs[i].len);
1646                 }
1647
1648                 beio->num_segs++;
1649                 len_left -= beio->sg_segs[i].len;
1650         }
1651         if (bptrlen->len < lbalen->len)
1652                 beio->beio_cont = ctl_be_block_next;
1653         io->scsiio.be_move_done = ctl_be_block_move_done;
1654         /* For compare we have separate S/G lists for read and datamove. */
1655         if (beio->two_sglists)
1656                 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS];
1657         else
1658                 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs;
1659         io->scsiio.kern_data_len = beio->io_len;
1660         io->scsiio.kern_sg_entries = beio->num_segs;
1661         io->scsiio.kern_data_ref = ctl_refcnt_beio;
1662         io->scsiio.kern_data_arg = beio;
1663         io->io_hdr.flags |= CTL_FLAG_ALLOCATED;
1664
1665         /*
1666          * For the read case, we need to read the data into our buffers and
1667          * then we can send it back to the user.  For the write case, we
1668          * need to get the data from the user first.
1669          */
1670         if (beio->bio_cmd == BIO_READ) {
1671                 SDT_PROBE0(cbb, , read, alloc_done);
1672                 be_lun->dispatch(be_lun, beio);
1673         } else {
1674                 SDT_PROBE0(cbb, , write, alloc_done);
1675 #ifdef CTL_TIME_IO
1676                 getbinuptime(&io->io_hdr.dma_start_bt);
1677 #endif
1678                 ctl_datamove(io);
1679         }
1680 }
1681
1682 static void
1683 ctl_be_block_worker(void *context, int pending)
1684 {
1685         struct ctl_be_block_lun *be_lun = (struct ctl_be_block_lun *)context;
1686         struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1687         union ctl_io *io;
1688         struct ctl_be_block_io *beio;
1689
1690         DPRINTF("entered\n");
1691         /*
1692          * Fetch and process I/Os from all queues.  If we detect LUN
1693          * CTL_LUN_FLAG_NO_MEDIA status here -- it is result of a race,
1694          * so make response maximally opaque to not confuse initiator.
1695          */
1696         for (;;) {
1697                 mtx_lock(&be_lun->queue_lock);
1698                 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue);
1699                 if (io != NULL) {
1700                         DPRINTF("datamove queue\n");
1701                         STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr,
1702                                       ctl_io_hdr, links);
1703                         mtx_unlock(&be_lun->queue_lock);
1704                         beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1705                         if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) {
1706                                 ctl_set_busy(&io->scsiio);
1707                                 ctl_complete_beio(beio);
1708                                 continue;
1709                         }
1710                         be_lun->dispatch(be_lun, beio);
1711                         continue;
1712                 }
1713                 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue);
1714                 if (io != NULL) {
1715                         DPRINTF("config write queue\n");
1716                         STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr,
1717                                       ctl_io_hdr, links);
1718                         mtx_unlock(&be_lun->queue_lock);
1719                         if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) {
1720                                 ctl_set_busy(&io->scsiio);
1721                                 ctl_config_write_done(io);
1722                                 continue;
1723                         }
1724                         ctl_be_block_cw_dispatch(be_lun, io);
1725                         continue;
1726                 }
1727                 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_read_queue);
1728                 if (io != NULL) {
1729                         DPRINTF("config read queue\n");
1730                         STAILQ_REMOVE(&be_lun->config_read_queue, &io->io_hdr,
1731                                       ctl_io_hdr, links);
1732                         mtx_unlock(&be_lun->queue_lock);
1733                         if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) {
1734                                 ctl_set_busy(&io->scsiio);
1735                                 ctl_config_read_done(io);
1736                                 continue;
1737                         }
1738                         ctl_be_block_cr_dispatch(be_lun, io);
1739                         continue;
1740                 }
1741                 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue);
1742                 if (io != NULL) {
1743                         DPRINTF("input queue\n");
1744                         STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr,
1745                                       ctl_io_hdr, links);
1746                         mtx_unlock(&be_lun->queue_lock);
1747                         if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) {
1748                                 ctl_set_busy(&io->scsiio);
1749                                 ctl_data_submit_done(io);
1750                                 continue;
1751                         }
1752                         ctl_be_block_dispatch(be_lun, io);
1753                         continue;
1754                 }
1755
1756                 /*
1757                  * If we get here, there is no work left in the queues, so
1758                  * just break out and let the task queue go to sleep.
1759                  */
1760                 mtx_unlock(&be_lun->queue_lock);
1761                 break;
1762         }
1763 }
1764
1765 /*
1766  * Entry point from CTL to the backend for I/O.  We queue everything to a
1767  * work thread, so this just puts the I/O on a queue and wakes up the
1768  * thread.
1769  */
1770 static int
1771 ctl_be_block_submit(union ctl_io *io)
1772 {
1773         struct ctl_be_block_lun *be_lun;
1774
1775         DPRINTF("entered\n");
1776
1777         be_lun = (struct ctl_be_block_lun *)CTL_BACKEND_LUN(io);
1778
1779         /*
1780          * Make sure we only get SCSI I/O.
1781          */
1782         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type "
1783                 "%#x) encountered", io->io_hdr.io_type));
1784
1785         PRIV(io)->len = 0;
1786
1787         mtx_lock(&be_lun->queue_lock);
1788         STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links);
1789         mtx_unlock(&be_lun->queue_lock);
1790         taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
1791
1792         return (CTL_RETVAL_COMPLETE);
1793 }
1794
1795 static int
1796 ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
1797                         int flag, struct thread *td)
1798 {
1799         struct ctl_be_block_softc *softc = &backend_block_softc;
1800         int error;
1801
1802         error = 0;
1803         switch (cmd) {
1804         case CTL_LUN_REQ: {
1805                 struct ctl_lun_req *lun_req;
1806
1807                 lun_req = (struct ctl_lun_req *)addr;
1808
1809                 switch (lun_req->reqtype) {
1810                 case CTL_LUNREQ_CREATE:
1811                         error = ctl_be_block_create(softc, lun_req);
1812                         break;
1813                 case CTL_LUNREQ_RM:
1814                         error = ctl_be_block_rm(softc, lun_req);
1815                         break;
1816                 case CTL_LUNREQ_MODIFY:
1817                         error = ctl_be_block_modify(softc, lun_req);
1818                         break;
1819                 default:
1820                         lun_req->status = CTL_LUN_ERROR;
1821                         snprintf(lun_req->error_str, sizeof(lun_req->error_str),
1822                                  "invalid LUN request type %d",
1823                                  lun_req->reqtype);
1824                         break;
1825                 }
1826                 break;
1827         }
1828         default:
1829                 error = ENOTTY;
1830                 break;
1831         }
1832
1833         return (error);
1834 }
1835
1836 static int
1837 ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
1838 {
1839         struct ctl_be_lun *cbe_lun;
1840         struct ctl_be_block_filedata *file_data;
1841         struct ctl_lun_create_params *params;
1842         const char                   *value;
1843         struct vattr                  vattr;
1844         off_t                         ps, pss, po, pos, us, uss, uo, uos;
1845         int                           error;
1846
1847         cbe_lun = &be_lun->cbe_lun;
1848         file_data = &be_lun->backend.file;
1849         params = &be_lun->params;
1850
1851         be_lun->dev_type = CTL_BE_BLOCK_FILE;
1852         be_lun->dispatch = ctl_be_block_dispatch_file;
1853         be_lun->lun_flush = ctl_be_block_flush_file;
1854         be_lun->get_lba_status = ctl_be_block_gls_file;
1855         be_lun->getattr = ctl_be_block_getattr_file;
1856         be_lun->unmap = NULL;
1857         cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP;
1858
1859         error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred);
1860         if (error != 0) {
1861                 snprintf(req->error_str, sizeof(req->error_str),
1862                          "error calling VOP_GETATTR() for file %s",
1863                          be_lun->dev_path);
1864                 return (error);
1865         }
1866
1867         file_data->cred = crhold(curthread->td_ucred);
1868         if (params->lun_size_bytes != 0)
1869                 be_lun->size_bytes = params->lun_size_bytes;
1870         else
1871                 be_lun->size_bytes = vattr.va_size;
1872
1873         /*
1874          * For files we can use any logical block size.  Prefer 512 bytes
1875          * for compatibility reasons.  If file's vattr.va_blocksize
1876          * (preferred I/O block size) is bigger and multiple to chosen
1877          * logical block size -- report it as physical block size.
1878          */
1879         if (params->blocksize_bytes != 0)
1880                 cbe_lun->blocksize = params->blocksize_bytes;
1881         else if (cbe_lun->lun_type == T_CDROM)
1882                 cbe_lun->blocksize = 2048;
1883         else
1884                 cbe_lun->blocksize = 512;
1885         be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize;
1886         cbe_lun->maxlba = (be_lun->size_blocks == 0) ?
1887             0 : (be_lun->size_blocks - 1);
1888
1889         us = ps = vattr.va_blocksize;
1890         uo = po = 0;
1891
1892         value = dnvlist_get_string(cbe_lun->options, "pblocksize", NULL);
1893         if (value != NULL)
1894                 ctl_expand_number(value, &ps);
1895         value = dnvlist_get_string(cbe_lun->options, "pblockoffset", NULL);
1896         if (value != NULL)
1897                 ctl_expand_number(value, &po);
1898         pss = ps / cbe_lun->blocksize;
1899         pos = po / cbe_lun->blocksize;
1900         if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) &&
1901             ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) {
1902                 cbe_lun->pblockexp = fls(pss) - 1;
1903                 cbe_lun->pblockoff = (pss - pos) % pss;
1904         }
1905
1906         value = dnvlist_get_string(cbe_lun->options, "ublocksize", NULL);
1907         if (value != NULL)
1908                 ctl_expand_number(value, &us);
1909         value = dnvlist_get_string(cbe_lun->options, "ublockoffset", NULL);
1910         if (value != NULL)
1911                 ctl_expand_number(value, &uo);
1912         uss = us / cbe_lun->blocksize;
1913         uos = uo / cbe_lun->blocksize;
1914         if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) &&
1915             ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) {
1916                 cbe_lun->ublockexp = fls(uss) - 1;
1917                 cbe_lun->ublockoff = (uss - uos) % uss;
1918         }
1919
1920         /*
1921          * Sanity check.  The media size has to be at least one
1922          * sector long.
1923          */
1924         if (be_lun->size_bytes < cbe_lun->blocksize) {
1925                 error = EINVAL;
1926                 snprintf(req->error_str, sizeof(req->error_str),
1927                          "file %s size %ju < block size %u", be_lun->dev_path,
1928                          (uintmax_t)be_lun->size_bytes, cbe_lun->blocksize);
1929         }
1930
1931         cbe_lun->opttxferlen = CTLBLK_MAX_IO_SIZE / cbe_lun->blocksize;
1932         return (error);
1933 }
1934
1935 static int
1936 ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
1937 {
1938         struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
1939         struct ctl_lun_create_params *params;
1940         struct cdevsw                *csw;
1941         struct cdev                  *dev;
1942         const char                   *value;
1943         int                           error, atomic, maxio, ref, unmap, tmp;
1944         off_t                         ps, pss, po, pos, us, uss, uo, uos, otmp;
1945
1946         params = &be_lun->params;
1947
1948         be_lun->dev_type = CTL_BE_BLOCK_DEV;
1949         csw = devvn_refthread(be_lun->vn, &dev, &ref);
1950         if (csw == NULL)
1951                 return (ENXIO);
1952         if (strcmp(csw->d_name, "zvol") == 0) {
1953                 be_lun->dispatch = ctl_be_block_dispatch_zvol;
1954                 be_lun->get_lba_status = ctl_be_block_gls_zvol;
1955                 atomic = maxio = CTLBLK_MAX_IO_SIZE;
1956         } else {
1957                 be_lun->dispatch = ctl_be_block_dispatch_dev;
1958                 be_lun->get_lba_status = NULL;
1959                 atomic = 0;
1960                 maxio = dev->si_iosize_max;
1961                 if (maxio <= 0)
1962                         maxio = DFLTPHYS;
1963                 if (maxio > CTLBLK_MAX_SEG)
1964                         maxio = CTLBLK_MAX_SEG;
1965         }
1966         be_lun->lun_flush = ctl_be_block_flush_dev;
1967         be_lun->getattr = ctl_be_block_getattr_dev;
1968         be_lun->unmap = ctl_be_block_unmap_dev;
1969
1970         if (!csw->d_ioctl) {
1971                 dev_relthread(dev, ref);
1972                 snprintf(req->error_str, sizeof(req->error_str),
1973                          "no d_ioctl for device %s!", be_lun->dev_path);
1974                 return (ENODEV);
1975         }
1976
1977         error = csw->d_ioctl(dev, DIOCGSECTORSIZE, (caddr_t)&tmp, FREAD,
1978                                curthread);
1979         if (error) {
1980                 dev_relthread(dev, ref);
1981                 snprintf(req->error_str, sizeof(req->error_str),
1982                          "error %d returned for DIOCGSECTORSIZE ioctl "
1983                          "on %s!", error, be_lun->dev_path);
1984                 return (error);
1985         }
1986
1987         /*
1988          * If the user has asked for a blocksize that is greater than the
1989          * backing device's blocksize, we can do it only if the blocksize
1990          * the user is asking for is an even multiple of the underlying
1991          * device's blocksize.
1992          */
1993         if ((params->blocksize_bytes != 0) &&
1994             (params->blocksize_bytes >= tmp)) {
1995                 if (params->blocksize_bytes % tmp == 0) {
1996                         cbe_lun->blocksize = params->blocksize_bytes;
1997                 } else {
1998                         dev_relthread(dev, ref);
1999                         snprintf(req->error_str, sizeof(req->error_str),
2000                                  "requested blocksize %u is not an even "
2001                                  "multiple of backing device blocksize %u",
2002                                  params->blocksize_bytes, tmp);
2003                         return (EINVAL);
2004                 }
2005         } else if (params->blocksize_bytes != 0) {
2006                 dev_relthread(dev, ref);
2007                 snprintf(req->error_str, sizeof(req->error_str),
2008                          "requested blocksize %u < backing device "
2009                          "blocksize %u", params->blocksize_bytes, tmp);
2010                 return (EINVAL);
2011         } else if (cbe_lun->lun_type == T_CDROM)
2012                 cbe_lun->blocksize = MAX(tmp, 2048);
2013         else
2014                 cbe_lun->blocksize = tmp;
2015
2016         error = csw->d_ioctl(dev, DIOCGMEDIASIZE, (caddr_t)&otmp, FREAD,
2017                              curthread);
2018         if (error) {
2019                 dev_relthread(dev, ref);
2020                 snprintf(req->error_str, sizeof(req->error_str),
2021                          "error %d returned for DIOCGMEDIASIZE "
2022                          " ioctl on %s!", error,
2023                          be_lun->dev_path);
2024                 return (error);
2025         }
2026
2027         if (params->lun_size_bytes != 0) {
2028                 if (params->lun_size_bytes > otmp) {
2029                         dev_relthread(dev, ref);
2030                         snprintf(req->error_str, sizeof(req->error_str),
2031                                  "requested LUN size %ju > backing device "
2032                                  "size %ju",
2033                                  (uintmax_t)params->lun_size_bytes,
2034                                  (uintmax_t)otmp);
2035                         return (EINVAL);
2036                 }
2037
2038                 be_lun->size_bytes = params->lun_size_bytes;
2039         } else
2040                 be_lun->size_bytes = otmp;
2041         be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize;
2042         cbe_lun->maxlba = (be_lun->size_blocks == 0) ?
2043             0 : (be_lun->size_blocks - 1);
2044
2045         error = csw->d_ioctl(dev, DIOCGSTRIPESIZE, (caddr_t)&ps, FREAD,
2046             curthread);
2047         if (error)
2048                 ps = po = 0;
2049         else {
2050                 error = csw->d_ioctl(dev, DIOCGSTRIPEOFFSET, (caddr_t)&po,
2051                     FREAD, curthread);
2052                 if (error)
2053                         po = 0;
2054         }
2055         us = ps;
2056         uo = po;
2057
2058         value = dnvlist_get_string(cbe_lun->options, "pblocksize", NULL);
2059         if (value != NULL)
2060                 ctl_expand_number(value, &ps);
2061         value = dnvlist_get_string(cbe_lun->options, "pblockoffset", NULL);
2062         if (value != NULL)
2063                 ctl_expand_number(value, &po);
2064         pss = ps / cbe_lun->blocksize;
2065         pos = po / cbe_lun->blocksize;
2066         if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) &&
2067             ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) {
2068                 cbe_lun->pblockexp = fls(pss) - 1;
2069                 cbe_lun->pblockoff = (pss - pos) % pss;
2070         }
2071
2072         value = dnvlist_get_string(cbe_lun->options, "ublocksize", NULL);
2073         if (value != NULL)
2074                 ctl_expand_number(value, &us);
2075         value = dnvlist_get_string(cbe_lun->options, "ublockoffset", NULL);
2076         if (value != NULL)
2077                 ctl_expand_number(value, &uo);
2078         uss = us / cbe_lun->blocksize;
2079         uos = uo / cbe_lun->blocksize;
2080         if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) &&
2081             ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) {
2082                 cbe_lun->ublockexp = fls(uss) - 1;
2083                 cbe_lun->ublockoff = (uss - uos) % uss;
2084         }
2085
2086         cbe_lun->atomicblock = atomic / cbe_lun->blocksize;
2087         cbe_lun->opttxferlen = maxio / cbe_lun->blocksize;
2088
2089         if (be_lun->dispatch == ctl_be_block_dispatch_zvol) {
2090                 unmap = 1;
2091         } else {
2092                 struct diocgattr_arg    arg;
2093
2094                 strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name));
2095                 arg.len = sizeof(arg.value.i);
2096                 error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD,
2097                     curthread);
2098                 unmap = (error == 0) ? arg.value.i : 0;
2099         }
2100         value = dnvlist_get_string(cbe_lun->options, "unmap", NULL);
2101         if (value != NULL)
2102                 unmap = (strcmp(value, "on") == 0);
2103         if (unmap)
2104                 cbe_lun->flags |= CTL_LUN_FLAG_UNMAP;
2105         else
2106                 cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP;
2107
2108         dev_relthread(dev, ref);
2109         return (0);
2110 }
2111
2112 static int
2113 ctl_be_block_close(struct ctl_be_block_lun *be_lun)
2114 {
2115         struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
2116         int flags;
2117
2118         if (be_lun->vn) {
2119                 flags = FREAD;
2120                 if ((cbe_lun->flags & CTL_LUN_FLAG_READONLY) == 0)
2121                         flags |= FWRITE;
2122                 (void)vn_close(be_lun->vn, flags, NOCRED, curthread);
2123                 be_lun->vn = NULL;
2124
2125                 switch (be_lun->dev_type) {
2126                 case CTL_BE_BLOCK_DEV:
2127                         break;
2128                 case CTL_BE_BLOCK_FILE:
2129                         if (be_lun->backend.file.cred != NULL) {
2130                                 crfree(be_lun->backend.file.cred);
2131                                 be_lun->backend.file.cred = NULL;
2132                         }
2133                         break;
2134                 case CTL_BE_BLOCK_NONE:
2135                         break;
2136                 default:
2137                         panic("Unexpected backend type %d", be_lun->dev_type);
2138                         break;
2139                 }
2140                 be_lun->dev_type = CTL_BE_BLOCK_NONE;
2141         }
2142         return (0);
2143 }
2144
2145 static int
2146 ctl_be_block_open(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
2147 {
2148         struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun;
2149         struct nameidata nd;
2150         const char      *value;
2151         int              error, flags;
2152
2153         error = 0;
2154         if (rootvnode == NULL) {
2155                 snprintf(req->error_str, sizeof(req->error_str),
2156                          "Root filesystem is not mounted");
2157                 return (1);
2158         }
2159         pwd_ensure_dirs();
2160
2161         value = dnvlist_get_string(cbe_lun->options, "file", NULL);
2162         if (value == NULL) {
2163                 snprintf(req->error_str, sizeof(req->error_str),
2164                          "no file argument specified");
2165                 return (1);
2166         }
2167         free(be_lun->dev_path, M_CTLBLK);
2168         be_lun->dev_path = strdup(value, M_CTLBLK);
2169
2170         flags = FREAD;
2171         value = dnvlist_get_string(cbe_lun->options, "readonly", NULL);
2172         if (value != NULL) {
2173                 if (strcmp(value, "on") != 0)
2174                         flags |= FWRITE;
2175         } else if (cbe_lun->lun_type == T_DIRECT)
2176                 flags |= FWRITE;
2177
2178 again:
2179         NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread);
2180         error = vn_open(&nd, &flags, 0, NULL);
2181         if ((error == EROFS || error == EACCES) && (flags & FWRITE)) {
2182                 flags &= ~FWRITE;
2183                 goto again;
2184         }
2185         if (error) {
2186                 /*
2187                  * This is the only reasonable guess we can make as far as
2188                  * path if the user doesn't give us a fully qualified path.
2189                  * If they want to specify a file, they need to specify the
2190                  * full path.
2191                  */
2192                 if (be_lun->dev_path[0] != '/') {
2193                         char *dev_name;
2194
2195                         asprintf(&dev_name, M_CTLBLK, "/dev/%s",
2196                                 be_lun->dev_path);
2197                         free(be_lun->dev_path, M_CTLBLK);
2198                         be_lun->dev_path = dev_name;
2199                         goto again;
2200                 }
2201                 snprintf(req->error_str, sizeof(req->error_str),
2202                     "error opening %s: %d", be_lun->dev_path, error);
2203                 return (error);
2204         }
2205         if (flags & FWRITE)
2206                 cbe_lun->flags &= ~CTL_LUN_FLAG_READONLY;
2207         else
2208                 cbe_lun->flags |= CTL_LUN_FLAG_READONLY;
2209
2210         NDFREE(&nd, NDF_ONLY_PNBUF);
2211         be_lun->vn = nd.ni_vp;
2212
2213         /* We only support disks and files. */
2214         if (vn_isdisk_error(be_lun->vn, &error)) {
2215                 error = ctl_be_block_open_dev(be_lun, req);
2216         } else if (be_lun->vn->v_type == VREG) {
2217                 error = ctl_be_block_open_file(be_lun, req);
2218         } else {
2219                 error = EINVAL;
2220                 snprintf(req->error_str, sizeof(req->error_str),
2221                          "%s is not a disk or plain file", be_lun->dev_path);
2222         }
2223         VOP_UNLOCK(be_lun->vn);
2224
2225         if (error != 0)
2226                 ctl_be_block_close(be_lun);
2227         cbe_lun->serseq = CTL_LUN_SERSEQ_OFF;
2228         if (be_lun->dispatch != ctl_be_block_dispatch_dev)
2229                 cbe_lun->serseq = CTL_LUN_SERSEQ_READ;
2230         value = dnvlist_get_string(cbe_lun->options, "serseq", NULL);
2231         if (value != NULL && strcmp(value, "on") == 0)
2232                 cbe_lun->serseq = CTL_LUN_SERSEQ_ON;
2233         else if (value != NULL && strcmp(value, "read") == 0)
2234                 cbe_lun->serseq = CTL_LUN_SERSEQ_READ;
2235         else if (value != NULL && strcmp(value, "off") == 0)
2236                 cbe_lun->serseq = CTL_LUN_SERSEQ_OFF;
2237         return (0);
2238 }
2239
2240 static int
2241 ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2242 {
2243         struct ctl_be_lun *cbe_lun;
2244         struct ctl_be_block_lun *be_lun;
2245         struct ctl_lun_create_params *params;
2246         char num_thread_str[16];
2247         char tmpstr[32];
2248         const char *value;
2249         int retval, num_threads;
2250         int tmp_num_threads;
2251
2252         params = &req->reqdata.create;
2253         retval = 0;
2254         req->status = CTL_LUN_OK;
2255
2256         be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK);
2257         cbe_lun = &be_lun->cbe_lun;
2258         be_lun->params = req->reqdata.create;
2259         be_lun->softc = softc;
2260         STAILQ_INIT(&be_lun->input_queue);
2261         STAILQ_INIT(&be_lun->config_read_queue);
2262         STAILQ_INIT(&be_lun->config_write_queue);
2263         STAILQ_INIT(&be_lun->datamove_queue);
2264         mtx_init(&be_lun->io_lock, "ctlblock io", NULL, MTX_DEF);
2265         mtx_init(&be_lun->queue_lock, "ctlblock queue", NULL, MTX_DEF);
2266         cbe_lun->options = nvlist_clone(req->args_nvl);
2267
2268         if (params->flags & CTL_LUN_FLAG_DEV_TYPE)
2269                 cbe_lun->lun_type = params->device_type;
2270         else
2271                 cbe_lun->lun_type = T_DIRECT;
2272         be_lun->flags = 0;
2273         cbe_lun->flags = 0;
2274         value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL);
2275         if (value != NULL) {
2276                 if (strcmp(value, "primary") == 0)
2277                         cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
2278         } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF)
2279                 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
2280
2281         if (cbe_lun->lun_type == T_DIRECT ||
2282             cbe_lun->lun_type == T_CDROM) {
2283                 be_lun->size_bytes = params->lun_size_bytes;
2284                 if (params->blocksize_bytes != 0)
2285                         cbe_lun->blocksize = params->blocksize_bytes;
2286                 else if (cbe_lun->lun_type == T_CDROM)
2287                         cbe_lun->blocksize = 2048;
2288                 else
2289                         cbe_lun->blocksize = 512;
2290                 be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize;
2291                 cbe_lun->maxlba = (be_lun->size_blocks == 0) ?
2292                     0 : (be_lun->size_blocks - 1);
2293
2294                 if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) ||
2295                     control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) {
2296                         retval = ctl_be_block_open(be_lun, req);
2297                         if (retval != 0) {
2298                                 retval = 0;
2299                                 req->status = CTL_LUN_WARNING;
2300                         }
2301                 }
2302                 num_threads = cbb_num_threads;
2303         } else {
2304                 num_threads = 1;
2305         }
2306
2307         value = dnvlist_get_string(cbe_lun->options, "num_threads", NULL);
2308         if (value != NULL) {
2309                 tmp_num_threads = strtol(value, NULL, 0);
2310
2311                 /*
2312                  * We don't let the user specify less than one
2313                  * thread, but hope he's clueful enough not to
2314                  * specify 1000 threads.
2315                  */
2316                 if (tmp_num_threads < 1) {
2317                         snprintf(req->error_str, sizeof(req->error_str),
2318                                  "invalid number of threads %s",
2319                                  num_thread_str);
2320                         goto bailout_error;
2321                 }
2322                 num_threads = tmp_num_threads;
2323         }
2324
2325         if (be_lun->vn == NULL)
2326                 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2327         /* Tell the user the blocksize we ended up using */
2328         params->lun_size_bytes = be_lun->size_bytes;
2329         params->blocksize_bytes = cbe_lun->blocksize;
2330         if (params->flags & CTL_LUN_FLAG_ID_REQ) {
2331                 cbe_lun->req_lun_id = params->req_lun_id;
2332                 cbe_lun->flags |= CTL_LUN_FLAG_ID_REQ;
2333         } else
2334                 cbe_lun->req_lun_id = 0;
2335
2336         cbe_lun->lun_shutdown = ctl_be_block_lun_shutdown;
2337         cbe_lun->be = &ctl_be_block_driver;
2338
2339         if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) {
2340                 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%04d",
2341                          softc->num_luns);
2342                 strncpy((char *)cbe_lun->serial_num, tmpstr,
2343                         MIN(sizeof(cbe_lun->serial_num), sizeof(tmpstr)));
2344
2345                 /* Tell the user what we used for a serial number */
2346                 strncpy((char *)params->serial_num, tmpstr,
2347                         MIN(sizeof(params->serial_num), sizeof(tmpstr)));
2348         } else {
2349                 strncpy((char *)cbe_lun->serial_num, params->serial_num,
2350                         MIN(sizeof(cbe_lun->serial_num),
2351                         sizeof(params->serial_num)));
2352         }
2353         if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) {
2354                 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%04d", softc->num_luns);
2355                 strncpy((char *)cbe_lun->device_id, tmpstr,
2356                         MIN(sizeof(cbe_lun->device_id), sizeof(tmpstr)));
2357
2358                 /* Tell the user what we used for a device ID */
2359                 strncpy((char *)params->device_id, tmpstr,
2360                         MIN(sizeof(params->device_id), sizeof(tmpstr)));
2361         } else {
2362                 strncpy((char *)cbe_lun->device_id, params->device_id,
2363                         MIN(sizeof(cbe_lun->device_id),
2364                             sizeof(params->device_id)));
2365         }
2366
2367         TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun);
2368
2369         be_lun->io_taskqueue = taskqueue_create("ctlblocktq", M_WAITOK,
2370             taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue);
2371
2372         if (be_lun->io_taskqueue == NULL) {
2373                 snprintf(req->error_str, sizeof(req->error_str),
2374                          "unable to create taskqueue");
2375                 goto bailout_error;
2376         }
2377
2378         /*
2379          * Note that we start the same number of threads by default for
2380          * both the file case and the block device case.  For the file
2381          * case, we need multiple threads to allow concurrency, because the
2382          * vnode interface is designed to be a blocking interface.  For the
2383          * block device case, ZFS zvols at least will block the caller's
2384          * context in many instances, and so we need multiple threads to
2385          * overcome that problem.  Other block devices don't need as many
2386          * threads, but they shouldn't cause too many problems.
2387          *
2388          * If the user wants to just have a single thread for a block
2389          * device, he can specify that when the LUN is created, or change
2390          * the tunable/sysctl to alter the default number of threads.
2391          */
2392         retval = taskqueue_start_threads_in_proc(&be_lun->io_taskqueue,
2393                                          /*num threads*/num_threads,
2394                                          /*priority*/PUSER,
2395                                          /*proc*/control_softc->ctl_proc,
2396                                          /*thread name*/"block");
2397
2398         if (retval != 0)
2399                 goto bailout_error;
2400
2401         be_lun->num_threads = num_threads;
2402
2403         retval = ctl_add_lun(&be_lun->cbe_lun);
2404         if (retval != 0) {
2405                 snprintf(req->error_str, sizeof(req->error_str),
2406                          "ctl_add_lun() returned error %d, see dmesg for "
2407                          "details", retval);
2408                 retval = 0;
2409                 goto bailout_error;
2410         }
2411
2412         be_lun->disk_stats = devstat_new_entry("cbb", cbe_lun->lun_id,
2413                                                cbe_lun->blocksize,
2414                                                DEVSTAT_ALL_SUPPORTED,
2415                                                cbe_lun->lun_type
2416                                                | DEVSTAT_TYPE_IF_OTHER,
2417                                                DEVSTAT_PRIORITY_OTHER);
2418
2419         mtx_lock(&softc->lock);
2420         softc->num_luns++;
2421         SLIST_INSERT_HEAD(&softc->lun_list, be_lun, links);
2422         mtx_unlock(&softc->lock);
2423
2424         params->req_lun_id = cbe_lun->lun_id;
2425
2426         return (retval);
2427
2428 bailout_error:
2429         req->status = CTL_LUN_ERROR;
2430
2431         if (be_lun->io_taskqueue != NULL)
2432                 taskqueue_free(be_lun->io_taskqueue);
2433         ctl_be_block_close(be_lun);
2434         if (be_lun->dev_path != NULL)
2435                 free(be_lun->dev_path, M_CTLBLK);
2436         nvlist_destroy(cbe_lun->options);
2437         mtx_destroy(&be_lun->queue_lock);
2438         mtx_destroy(&be_lun->io_lock);
2439         free(be_lun, M_CTLBLK);
2440
2441         return (retval);
2442 }
2443
2444 static int
2445 ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2446 {
2447         struct ctl_lun_rm_params *params;
2448         struct ctl_be_block_lun *be_lun;
2449         struct ctl_be_lun *cbe_lun;
2450         int retval;
2451
2452         params = &req->reqdata.rm;
2453
2454         sx_xlock(&softc->modify_lock);
2455         mtx_lock(&softc->lock);
2456         SLIST_FOREACH(be_lun, &softc->lun_list, links) {
2457                 if (be_lun->cbe_lun.lun_id == params->lun_id) {
2458                         SLIST_REMOVE(&softc->lun_list, be_lun,
2459                             ctl_be_block_lun, links);
2460                         softc->num_luns--;
2461                         break;
2462                 }
2463         }
2464         mtx_unlock(&softc->lock);
2465         sx_xunlock(&softc->modify_lock);
2466         if (be_lun == NULL) {
2467                 snprintf(req->error_str, sizeof(req->error_str),
2468                          "LUN %u is not managed by the block backend",
2469                          params->lun_id);
2470                 goto bailout_error;
2471         }
2472         cbe_lun = &be_lun->cbe_lun;
2473
2474         if (be_lun->vn != NULL) {
2475                 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2476                 ctl_lun_no_media(cbe_lun);
2477                 taskqueue_drain_all(be_lun->io_taskqueue);
2478                 ctl_be_block_close(be_lun);
2479         }
2480
2481         mtx_lock(&softc->lock);
2482         be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING;
2483         mtx_unlock(&softc->lock);
2484
2485         retval = ctl_remove_lun(cbe_lun);
2486         if (retval != 0) {
2487                 snprintf(req->error_str, sizeof(req->error_str),
2488                          "error %d returned from ctl_remove_lun() for "
2489                          "LUN %d", retval, params->lun_id);
2490                 mtx_lock(&softc->lock);
2491                 be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING;
2492                 mtx_unlock(&softc->lock);
2493                 goto bailout_error;
2494         }
2495
2496         mtx_lock(&softc->lock);
2497         while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) {
2498                 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblockrm", 0);
2499                 if (retval == EINTR)
2500                         break;
2501         }
2502         be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING;
2503         if (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) {
2504                 mtx_unlock(&softc->lock);
2505                 free(be_lun, M_CTLBLK);
2506         } else {
2507                 mtx_unlock(&softc->lock);
2508                 return (EINTR);
2509         }
2510
2511         req->status = CTL_LUN_OK;
2512         return (0);
2513
2514 bailout_error:
2515         req->status = CTL_LUN_ERROR;
2516         return (0);
2517 }
2518
2519 static int
2520 ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2521 {
2522         struct ctl_lun_modify_params *params;
2523         struct ctl_be_block_lun *be_lun;
2524         struct ctl_be_lun *cbe_lun;
2525         const char *value;
2526         uint64_t oldsize;
2527         int error, wasprim;
2528
2529         params = &req->reqdata.modify;
2530
2531         sx_xlock(&softc->modify_lock);
2532         mtx_lock(&softc->lock);
2533         SLIST_FOREACH(be_lun, &softc->lun_list, links) {
2534                 if (be_lun->cbe_lun.lun_id == params->lun_id)
2535                         break;
2536         }
2537         mtx_unlock(&softc->lock);
2538         if (be_lun == NULL) {
2539                 snprintf(req->error_str, sizeof(req->error_str),
2540                          "LUN %u is not managed by the block backend",
2541                          params->lun_id);
2542                 goto bailout_error;
2543         }
2544         cbe_lun = &be_lun->cbe_lun;
2545
2546         if (params->lun_size_bytes != 0)
2547                 be_lun->params.lun_size_bytes = params->lun_size_bytes;
2548
2549         if (req->args_nvl != NULL) {
2550                 nvlist_destroy(cbe_lun->options);
2551                 cbe_lun->options = nvlist_clone(req->args_nvl);
2552         }
2553
2554         wasprim = (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY);
2555         value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL);
2556         if (value != NULL) {
2557                 if (strcmp(value, "primary") == 0)
2558                         cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
2559                 else
2560                         cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY;
2561         } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF)
2562                 cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY;
2563         else
2564                 cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY;
2565         if (wasprim != (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)) {
2566                 if (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)
2567                         ctl_lun_primary(cbe_lun);
2568                 else
2569                         ctl_lun_secondary(cbe_lun);
2570         }
2571
2572         oldsize = be_lun->size_blocks;
2573         if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) ||
2574             control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) {
2575                 if (be_lun->vn == NULL)
2576                         error = ctl_be_block_open(be_lun, req);
2577                 else if (vn_isdisk_error(be_lun->vn, &error))
2578                         error = ctl_be_block_open_dev(be_lun, req);
2579                 else if (be_lun->vn->v_type == VREG) {
2580                         vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
2581                         error = ctl_be_block_open_file(be_lun, req);
2582                         VOP_UNLOCK(be_lun->vn);
2583                 } else
2584                         error = EINVAL;
2585                 if ((cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) &&
2586                     be_lun->vn != NULL) {
2587                         cbe_lun->flags &= ~CTL_LUN_FLAG_NO_MEDIA;
2588                         ctl_lun_has_media(cbe_lun);
2589                 } else if ((cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) == 0 &&
2590                     be_lun->vn == NULL) {
2591                         cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2592                         ctl_lun_no_media(cbe_lun);
2593                 }
2594                 cbe_lun->flags &= ~CTL_LUN_FLAG_EJECTED;
2595         } else {
2596                 if (be_lun->vn != NULL) {
2597                         cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2598                         ctl_lun_no_media(cbe_lun);
2599                         taskqueue_drain_all(be_lun->io_taskqueue);
2600                         error = ctl_be_block_close(be_lun);
2601                 } else
2602                         error = 0;
2603         }
2604         if (be_lun->size_blocks != oldsize)
2605                 ctl_lun_capacity_changed(cbe_lun);
2606
2607         /* Tell the user the exact size we ended up using */
2608         params->lun_size_bytes = be_lun->size_bytes;
2609
2610         sx_xunlock(&softc->modify_lock);
2611         req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK;
2612         return (0);
2613
2614 bailout_error:
2615         sx_xunlock(&softc->modify_lock);
2616         req->status = CTL_LUN_ERROR;
2617         return (0);
2618 }
2619
2620 static void
2621 ctl_be_block_lun_shutdown(struct ctl_be_lun *cbe_lun)
2622 {
2623         struct ctl_be_block_lun *be_lun = (struct ctl_be_block_lun *)cbe_lun;
2624         struct ctl_be_block_softc *softc = be_lun->softc;
2625
2626         taskqueue_drain_all(be_lun->io_taskqueue);
2627         taskqueue_free(be_lun->io_taskqueue);
2628         if (be_lun->disk_stats != NULL)
2629                 devstat_remove_entry(be_lun->disk_stats);
2630         nvlist_destroy(be_lun->cbe_lun.options);
2631         free(be_lun->dev_path, M_CTLBLK);
2632         mtx_destroy(&be_lun->queue_lock);
2633         mtx_destroy(&be_lun->io_lock);
2634
2635         mtx_lock(&softc->lock);
2636         be_lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED;
2637         if (be_lun->flags & CTL_BE_BLOCK_LUN_WAITING)
2638                 wakeup(be_lun);
2639         else
2640                 free(be_lun, M_CTLBLK);
2641         mtx_unlock(&softc->lock);
2642 }
2643
2644 static int
2645 ctl_be_block_config_write(union ctl_io *io)
2646 {
2647         struct ctl_be_block_lun *be_lun;
2648         struct ctl_be_lun *cbe_lun;
2649         int retval;
2650
2651         DPRINTF("entered\n");
2652
2653         cbe_lun = CTL_BACKEND_LUN(io);
2654         be_lun = (struct ctl_be_block_lun *)cbe_lun;
2655
2656         retval = 0;
2657         switch (io->scsiio.cdb[0]) {
2658         case SYNCHRONIZE_CACHE:
2659         case SYNCHRONIZE_CACHE_16:
2660         case WRITE_SAME_10:
2661         case WRITE_SAME_16:
2662         case UNMAP:
2663                 /*
2664                  * The upper level CTL code will filter out any CDBs with
2665                  * the immediate bit set and return the proper error.
2666                  *
2667                  * We don't really need to worry about what LBA range the
2668                  * user asked to be synced out.  When they issue a sync
2669                  * cache command, we'll sync out the whole thing.
2670                  */
2671                 mtx_lock(&be_lun->queue_lock);
2672                 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr,
2673                                    links);
2674                 mtx_unlock(&be_lun->queue_lock);
2675                 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
2676                 break;
2677         case START_STOP_UNIT: {
2678                 struct scsi_start_stop_unit *cdb;
2679                 struct ctl_lun_req req;
2680
2681                 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb;
2682                 if ((cdb->how & SSS_PC_MASK) != 0) {
2683                         ctl_set_success(&io->scsiio);
2684                         ctl_config_write_done(io);
2685                         break;
2686                 }
2687                 if (cdb->how & SSS_START) {
2688                         if ((cdb->how & SSS_LOEJ) && be_lun->vn == NULL) {
2689                                 retval = ctl_be_block_open(be_lun, &req);
2690                                 cbe_lun->flags &= ~CTL_LUN_FLAG_EJECTED;
2691                                 if (retval == 0) {
2692                                         cbe_lun->flags &= ~CTL_LUN_FLAG_NO_MEDIA;
2693                                         ctl_lun_has_media(cbe_lun);
2694                                 } else {
2695                                         cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2696                                         ctl_lun_no_media(cbe_lun);
2697                                 }
2698                         }
2699                         ctl_start_lun(cbe_lun);
2700                 } else {
2701                         ctl_stop_lun(cbe_lun);
2702                         if (cdb->how & SSS_LOEJ) {
2703                                 cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA;
2704                                 cbe_lun->flags |= CTL_LUN_FLAG_EJECTED;
2705                                 ctl_lun_ejected(cbe_lun);
2706                                 if (be_lun->vn != NULL)
2707                                         ctl_be_block_close(be_lun);
2708                         }
2709                 }
2710
2711                 ctl_set_success(&io->scsiio);
2712                 ctl_config_write_done(io);
2713                 break;
2714         }
2715         case PREVENT_ALLOW:
2716                 ctl_set_success(&io->scsiio);
2717                 ctl_config_write_done(io);
2718                 break;
2719         default:
2720                 ctl_set_invalid_opcode(&io->scsiio);
2721                 ctl_config_write_done(io);
2722                 retval = CTL_RETVAL_COMPLETE;
2723                 break;
2724         }
2725
2726         return (retval);
2727 }
2728
2729 static int
2730 ctl_be_block_config_read(union ctl_io *io)
2731 {
2732         struct ctl_be_block_lun *be_lun;
2733         int retval = 0;
2734
2735         DPRINTF("entered\n");
2736
2737         be_lun = (struct ctl_be_block_lun *)CTL_BACKEND_LUN(io);
2738
2739         switch (io->scsiio.cdb[0]) {
2740         case SERVICE_ACTION_IN:
2741                 if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) {
2742                         mtx_lock(&be_lun->queue_lock);
2743                         STAILQ_INSERT_TAIL(&be_lun->config_read_queue,
2744                             &io->io_hdr, links);
2745                         mtx_unlock(&be_lun->queue_lock);
2746                         taskqueue_enqueue(be_lun->io_taskqueue,
2747                             &be_lun->io_task);
2748                         retval = CTL_RETVAL_QUEUED;
2749                         break;
2750                 }
2751                 ctl_set_invalid_field(&io->scsiio,
2752                                       /*sks_valid*/ 1,
2753                                       /*command*/ 1,
2754                                       /*field*/ 1,
2755                                       /*bit_valid*/ 1,
2756                                       /*bit*/ 4);
2757                 ctl_config_read_done(io);
2758                 retval = CTL_RETVAL_COMPLETE;
2759                 break;
2760         default:
2761                 ctl_set_invalid_opcode(&io->scsiio);
2762                 ctl_config_read_done(io);
2763                 retval = CTL_RETVAL_COMPLETE;
2764                 break;
2765         }
2766
2767         return (retval);
2768 }
2769
2770 static int
2771 ctl_be_block_lun_info(struct ctl_be_lun *cbe_lun, struct sbuf *sb)
2772 {
2773         struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)cbe_lun;
2774         int retval;
2775
2776         retval = sbuf_printf(sb, "\t<num_threads>");
2777         if (retval != 0)
2778                 goto bailout;
2779         retval = sbuf_printf(sb, "%d", lun->num_threads);
2780         if (retval != 0)
2781                 goto bailout;
2782         retval = sbuf_printf(sb, "</num_threads>\n");
2783
2784 bailout:
2785         return (retval);
2786 }
2787
2788 static uint64_t
2789 ctl_be_block_lun_attr(struct ctl_be_lun *cbe_lun, const char *attrname)
2790 {
2791         struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)cbe_lun;
2792
2793         if (lun->getattr == NULL)
2794                 return (UINT64_MAX);
2795         return (lun->getattr(lun, attrname));
2796 }
2797
2798 static int
2799 ctl_be_block_init(void)
2800 {
2801         struct ctl_be_block_softc *softc = &backend_block_softc;
2802
2803         sx_init(&softc->modify_lock, "ctlblock modify");
2804         mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF);
2805         softc->beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io),
2806             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
2807         softc->bufmin_zone = uma_zcreate("ctlblockmin", CTLBLK_MIN_SEG,
2808             NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0);
2809         if (CTLBLK_MIN_SEG < CTLBLK_MAX_SEG)
2810                 softc->bufmax_zone = uma_zcreate("ctlblockmax", CTLBLK_MAX_SEG,
2811                     NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0);
2812         SLIST_INIT(&softc->lun_list);
2813         return (0);
2814 }
2815
2816 static int
2817 ctl_be_block_shutdown(void)
2818 {
2819         struct ctl_be_block_softc *softc = &backend_block_softc;
2820         struct ctl_be_block_lun *lun;
2821
2822         mtx_lock(&softc->lock);
2823         while ((lun = SLIST_FIRST(&softc->lun_list)) != NULL) {
2824                 SLIST_REMOVE_HEAD(&softc->lun_list, links);
2825                 softc->num_luns--;
2826                 /*
2827                  * Drop our lock here.  Since ctl_remove_lun() can call
2828                  * back into us, this could potentially lead to a recursive
2829                  * lock of the same mutex, which would cause a hang.
2830                  */
2831                 mtx_unlock(&softc->lock);
2832                 ctl_remove_lun(&lun->cbe_lun);
2833                 mtx_lock(&softc->lock);
2834         }
2835         mtx_unlock(&softc->lock);
2836         uma_zdestroy(softc->bufmin_zone);
2837         if (CTLBLK_MIN_SEG < CTLBLK_MAX_SEG)
2838                 uma_zdestroy(softc->bufmax_zone);
2839         uma_zdestroy(softc->beio_zone);
2840         mtx_destroy(&softc->lock);
2841         sx_destroy(&softc->modify_lock);
2842         return (0);
2843 }