sys/cam/ctl/ctl_backend_block.c

   1 /*-
   2  * Copyright (c) 2003 Silicon Graphics International Corp.
   3  * Copyright (c) 2009-2011 Spectra Logic Corporation
   4  * Copyright (c) 2012 The FreeBSD Foundation
   5  * All rights reserved.
   6  *
   7  * Portions of this software were developed by Edward Tomasz Napierala
   8  * under sponsorship from the FreeBSD Foundation.
   9  *
  10  * Redistribution and use in source and binary forms, with or without
  11  * modification, are permitted provided that the following conditions
  12  * are met:
  13  * 1. Redistributions of source code must retain the above copyright
  14  *    notice, this list of conditions, and the following disclaimer,
  15  *    without modification.
  16  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
  17  *    substantially similar to the "NO WARRANTY" disclaimer below
  18  *    ("Disclaimer") and any redistribution must be conditioned upon
  19  *    including a substantially similar Disclaimer requirement for further
  20  *    binary redistribution.
  21  *
  22  * NO WARRANTY
  23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  24  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  25  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
  26  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  27  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
  31  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
  32  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  33  * POSSIBILITY OF SUCH DAMAGES.
  34  *
  35  * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_block.c#5 $
  36  */
  37 /*
  38  * CAM Target Layer driver backend for block devices.
  39  *
  40  * Author: Ken Merry <ken@FreeBSD.org>
  41  */
  42 #include <sys/cdefs.h>
  43 __FBSDID("$FreeBSD$");
  44
  45 #include <sys/param.h>
  46 #include <sys/systm.h>
  47 #include <sys/kernel.h>
  48 #include <sys/types.h>
  49 #include <sys/kthread.h>
  50 #include <sys/bio.h>
  51 #include <sys/fcntl.h>
  52 #include <sys/limits.h>
  53 #include <sys/lock.h>
  54 #include <sys/mutex.h>
  55 #include <sys/condvar.h>
  56 #include <sys/malloc.h>
  57 #include <sys/conf.h>
  58 #include <sys/ioccom.h>
  59 #include <sys/queue.h>
  60 #include <sys/sbuf.h>
  61 #include <sys/endian.h>
  62 #include <sys/uio.h>
  63 #include <sys/buf.h>
  64 #include <sys/taskqueue.h>
  65 #include <sys/vnode.h>
  66 #include <sys/namei.h>
  67 #include <sys/mount.h>
  68 #include <sys/disk.h>
  69 #include <sys/fcntl.h>
  70 #include <sys/filedesc.h>
  71 #include <sys/proc.h>
  72 #include <sys/pcpu.h>
  73 #include <sys/module.h>
  74 #include <sys/sdt.h>
  75 #include <sys/devicestat.h>
  76 #include <sys/sysctl.h>
  77
  78 #include <geom/geom.h>
  79
  80 #include <cam/cam.h>
  81 #include <cam/scsi/scsi_all.h>
  82 #include <cam/scsi/scsi_da.h>
  83 #include <cam/ctl/ctl_io.h>
  84 #include <cam/ctl/ctl.h>
  85 #include <cam/ctl/ctl_backend.h>
  86 #include <cam/ctl/ctl_frontend_internal.h>
  87 #include <cam/ctl/ctl_ioctl.h>
  88 #include <cam/ctl/ctl_scsi_all.h>
  89 #include <cam/ctl/ctl_error.h>
  90
  91 /*
  92  * The idea here is that we'll allocate enough S/G space to hold a 1MB
  93  * I/O.  If we get an I/O larger than that, we'll split it.
  94  */
  95 #define CTLBLK_HALF_IO_SIZE     (512 * 1024)
  96 #define CTLBLK_MAX_IO_SIZE      (CTLBLK_HALF_IO_SIZE * 2)
  97 #define CTLBLK_MAX_SEG          MAXPHYS
  98 #define CTLBLK_HALF_SEGS        MAX(CTLBLK_HALF_IO_SIZE / CTLBLK_MAX_SEG, 1)
  99 #define CTLBLK_MAX_SEGS         (CTLBLK_HALF_SEGS * 2)
 100
 101 #ifdef CTLBLK_DEBUG
 102 #define DPRINTF(fmt, args...) \
 103     printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args)
 104 #else
 105 #define DPRINTF(fmt, args...) do {} while(0)
 106 #endif
 107
 108 #define PRIV(io)        \
 109     ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND])
 110 #define ARGS(io)        \
 111     ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN])
 112
 113 SDT_PROVIDER_DEFINE(cbb);
 114
 115 typedef enum {
 116         CTL_BE_BLOCK_LUN_UNCONFIGURED   = 0x01,
 117         CTL_BE_BLOCK_LUN_CONFIG_ERR     = 0x02,
 118         CTL_BE_BLOCK_LUN_WAITING        = 0x04,
 119         CTL_BE_BLOCK_LUN_MULTI_THREAD   = 0x08
 120 } ctl_be_block_lun_flags;
 121
 122 typedef enum {
 123         CTL_BE_BLOCK_NONE,
 124         CTL_BE_BLOCK_DEV,
 125         CTL_BE_BLOCK_FILE
 126 } ctl_be_block_type;
 127
 128 struct ctl_be_block_devdata {
 129         struct cdev *cdev;
 130         struct cdevsw *csw;
 131         int dev_ref;
 132 };
 133
 134 struct ctl_be_block_filedata {
 135         struct ucred *cred;
 136 };
 137
 138 union ctl_be_block_bedata {
 139         struct ctl_be_block_devdata dev;
 140         struct ctl_be_block_filedata file;
 141 };
 142
 143 struct ctl_be_block_io;
 144 struct ctl_be_block_lun;
 145
 146 typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun,
 147                                struct ctl_be_block_io *beio);
 148 typedef uint64_t (*cbb_getattr_t)(struct ctl_be_block_lun *be_lun,
 149                                   const char *attrname);
 150
 151 /*
 152  * Backend LUN structure.  There is a 1:1 mapping between a block device
 153  * and a backend block LUN, and between a backend block LUN and a CTL LUN.
 154  */
 155 struct ctl_be_block_lun {
 156         struct ctl_lun_create_params params;
 157         struct ctl_block_disk *disk;
 158         char lunname[32];
 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_getattr_t getattr;
 167         uma_zone_t lun_zone;
 168         uint64_t size_blocks;
 169         uint64_t size_bytes;
 170         uint32_t blocksize;
 171         int blocksize_shift;
 172         uint16_t pblockexp;
 173         uint16_t pblockoff;
 174         struct ctl_be_block_softc *softc;
 175         struct devstat *disk_stats;
 176         ctl_be_block_lun_flags flags;
 177         STAILQ_ENTRY(ctl_be_block_lun) links;
 178         struct ctl_be_lun ctl_be_lun;
 179         struct taskqueue *io_taskqueue;
 180         struct task io_task;
 181         int num_threads;
 182         STAILQ_HEAD(, ctl_io_hdr) input_queue;
 183         STAILQ_HEAD(, ctl_io_hdr) config_write_queue;
 184         STAILQ_HEAD(, ctl_io_hdr) datamove_queue;
 185         struct mtx_padalign io_lock;
 186         struct mtx_padalign queue_lock;
 187 };
 188
 189 /*
 190  * Overall softc structure for the block backend module.
 191  */
 192 struct ctl_be_block_softc {
 193         struct mtx                       lock;
 194         int                              num_disks;
 195         STAILQ_HEAD(, ctl_block_disk)    disk_list;
 196         int                              num_luns;
 197         STAILQ_HEAD(, ctl_be_block_lun)  lun_list;
 198 };
 199
 200 static struct ctl_be_block_softc backend_block_softc;
 201
 202 /*
 203  * Per-I/O information.
 204  */
 205 struct ctl_be_block_io {
 206         union ctl_io                    *io;
 207         struct ctl_sg_entry             sg_segs[CTLBLK_MAX_SEGS];
 208         struct iovec                    xiovecs[CTLBLK_MAX_SEGS];
 209         int                             bio_cmd;
 210         int                             num_segs;
 211         int                             num_bios_sent;
 212         int                             num_bios_done;
 213         int                             send_complete;
 214         int                             num_errors;
 215         struct bintime                  ds_t0;
 216         devstat_tag_type                ds_tag_type;
 217         devstat_trans_flags             ds_trans_type;
 218         uint64_t                        io_len;
 219         uint64_t                        io_offset;
 220         struct ctl_be_block_softc       *softc;
 221         struct ctl_be_block_lun         *lun;
 222         void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */
 223 };
 224
 225 static int cbb_num_threads = 14;
 226 SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0,
 227             "CAM Target Layer Block Backend");
 228 SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RWTUN,
 229            &cbb_num_threads, 0, "Number of threads per backing file");
 230
 231 static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc);
 232 static void ctl_free_beio(struct ctl_be_block_io *beio);
 233 static void ctl_complete_beio(struct ctl_be_block_io *beio);
 234 static int ctl_be_block_move_done(union ctl_io *io);
 235 static void ctl_be_block_biodone(struct bio *bio);
 236 static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun,
 237                                     struct ctl_be_block_io *beio);
 238 static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun,
 239                                        struct ctl_be_block_io *beio);
 240 static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun,
 241                                    struct ctl_be_block_io *beio);
 242 static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun,
 243                                    struct ctl_be_block_io *beio);
 244 static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun,
 245                                       struct ctl_be_block_io *beio);
 246 static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun,
 247                                          const char *attrname);
 248 static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun,
 249                                     union ctl_io *io);
 250 static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun,
 251                                   union ctl_io *io);
 252 static void ctl_be_block_worker(void *context, int pending);
 253 static int ctl_be_block_submit(union ctl_io *io);
 254 static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
 255                                    int flag, struct thread *td);
 256 static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun,
 257                                   struct ctl_lun_req *req);
 258 static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun,
 259                                  struct ctl_lun_req *req);
 260 static int ctl_be_block_close(struct ctl_be_block_lun *be_lun);
 261 static int ctl_be_block_open(struct ctl_be_block_softc *softc,
 262                              struct ctl_be_block_lun *be_lun,
 263                              struct ctl_lun_req *req);
 264 static int ctl_be_block_create(struct ctl_be_block_softc *softc,
 265                                struct ctl_lun_req *req);
 266 static int ctl_be_block_rm(struct ctl_be_block_softc *softc,
 267                            struct ctl_lun_req *req);
 268 static int ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun,
 269                                   struct ctl_lun_req *req);
 270 static int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun,
 271                                  struct ctl_lun_req *req);
 272 static int ctl_be_block_modify(struct ctl_be_block_softc *softc,
 273                            struct ctl_lun_req *req);
 274 static void ctl_be_block_lun_shutdown(void *be_lun);
 275 static void ctl_be_block_lun_config_status(void *be_lun,
 276                                            ctl_lun_config_status status);
 277 static int ctl_be_block_config_write(union ctl_io *io);
 278 static int ctl_be_block_config_read(union ctl_io *io);
 279 static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb);
 280 static uint64_t ctl_be_block_lun_attr(void *be_lun, const char *attrname);
 281 int ctl_be_block_init(void);
 282
 283 static struct ctl_backend_driver ctl_be_block_driver =
 284 {
 285         .name = "block",
 286         .flags = CTL_BE_FLAG_HAS_CONFIG,
 287         .init = ctl_be_block_init,
 288         .data_submit = ctl_be_block_submit,
 289         .data_move_done = ctl_be_block_move_done,
 290         .config_read = ctl_be_block_config_read,
 291         .config_write = ctl_be_block_config_write,
 292         .ioctl = ctl_be_block_ioctl,
 293         .lun_info = ctl_be_block_lun_info,
 294         .lun_attr = ctl_be_block_lun_attr
 295 };
 296
 297 MALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend");
 298 CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver);
 299
 300 static uma_zone_t beio_zone;
 301
 302 static struct ctl_be_block_io *
 303 ctl_alloc_beio(struct ctl_be_block_softc *softc)
 304 {
 305         struct ctl_be_block_io *beio;
 306
 307         beio = uma_zalloc(beio_zone, M_WAITOK | M_ZERO);
 308         beio->softc = softc;
 309         return (beio);
 310 }
 311
 312 static void
 313 ctl_free_beio(struct ctl_be_block_io *beio)
 314 {
 315         int duplicate_free;
 316         int i;
 317
 318         duplicate_free = 0;
 319
 320         for (i = 0; i < beio->num_segs; i++) {
 321                 if (beio->sg_segs[i].addr == NULL)
 322                         duplicate_free++;
 323
 324                 uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr);
 325                 beio->sg_segs[i].addr = NULL;
 326
 327                 /* For compare we had two equal S/G lists. */
 328                 if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) {
 329                         uma_zfree(beio->lun->lun_zone,
 330                             beio->sg_segs[i + CTLBLK_HALF_SEGS].addr);
 331                         beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL;
 332                 }
 333         }
 334
 335         if (duplicate_free > 0) {
 336                 printf("%s: %d duplicate frees out of %d segments\n", __func__,
 337                        duplicate_free, beio->num_segs);
 338         }
 339
 340         uma_zfree(beio_zone, beio);
 341 }
 342
 343 static void
 344 ctl_complete_beio(struct ctl_be_block_io *beio)
 345 {
 346         union ctl_io *io = beio->io;
 347
 348         if (beio->beio_cont != NULL) {
 349                 beio->beio_cont(beio);
 350         } else {
 351                 ctl_free_beio(beio);
 352                 ctl_data_submit_done(io);
 353         }
 354 }
 355
 356 static int
 357 ctl_be_block_move_done(union ctl_io *io)
 358 {
 359         struct ctl_be_block_io *beio;
 360         struct ctl_be_block_lun *be_lun;
 361         struct ctl_lba_len_flags *lbalen;
 362 #ifdef CTL_TIME_IO
 363         struct bintime cur_bt;
 364 #endif
 365         int i;
 366
 367         beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
 368         be_lun = beio->lun;
 369
 370         DPRINTF("entered\n");
 371
 372 #ifdef CTL_TIME_IO
 373         getbintime(&cur_bt);
 374         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
 375         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
 376         io->io_hdr.num_dmas++;
 377 #endif
 378         io->scsiio.kern_rel_offset += io->scsiio.kern_data_len;
 379
 380         /*
 381          * We set status at this point for read commands, and write
 382          * commands with errors.
 383          */
 384         if ((io->io_hdr.port_status == 0) &&
 385             ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) &&
 386             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) {
 387                 lbalen = ARGS(beio->io);
 388                 if (lbalen->flags & CTL_LLF_READ) {
 389                         ctl_set_success(&io->scsiio);
 390                 } else if (lbalen->flags & CTL_LLF_COMPARE) {
 391                         /* We have two data blocks ready for comparison. */
 392                         for (i = 0; i < beio->num_segs; i++) {
 393                                 if (memcmp(beio->sg_segs[i].addr,
 394                                     beio->sg_segs[i + CTLBLK_HALF_SEGS].addr,
 395                                     beio->sg_segs[i].len) != 0)
 396                                         break;
 397                         }
 398                         if (i < beio->num_segs)
 399                                 ctl_set_sense(&io->scsiio,
 400                                     /*current_error*/ 1,
 401                                     /*sense_key*/ SSD_KEY_MISCOMPARE,
 402                                     /*asc*/ 0x1D,
 403                                     /*ascq*/ 0x00,
 404                                     SSD_ELEM_NONE);
 405                         else
 406                                 ctl_set_success(&io->scsiio);
 407                 }
 408         }
 409         else if ((io->io_hdr.port_status != 0)
 410               && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
 411               && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) {
 412                 /*
 413                  * For hardware error sense keys, the sense key
 414                  * specific value is defined to be a retry count,
 415                  * but we use it to pass back an internal FETD
 416                  * error code.  XXX KDM  Hopefully the FETD is only
 417                  * using 16 bits for an error code, since that's
 418                  * all the space we have in the sks field.
 419                  */
 420                 ctl_set_internal_failure(&io->scsiio,
 421                                          /*sks_valid*/ 1,
 422                                          /*retry_count*/
 423                                          io->io_hdr.port_status);
 424         }
 425
 426         /*
 427          * If this is a read, or a write with errors, it is done.
 428          */
 429         if ((beio->bio_cmd == BIO_READ)
 430          || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)
 431          || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) {
 432                 ctl_complete_beio(beio);
 433                 return (0);
 434         }
 435
 436         /*
 437          * At this point, we have a write and the DMA completed
 438          * successfully.  We now have to queue it to the task queue to
 439          * execute the backend I/O.  That is because we do blocking
 440          * memory allocations, and in the file backing case, blocking I/O.
 441          * This move done routine is generally called in the SIM's
 442          * interrupt context, and therefore we cannot block.
 443          */
 444         mtx_lock(&be_lun->queue_lock);
 445         /*
 446          * XXX KDM make sure that links is okay to use at this point.
 447          * Otherwise, we either need to add another field to ctl_io_hdr,
 448          * or deal with resource allocation here.
 449          */
 450         STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links);
 451         mtx_unlock(&be_lun->queue_lock);
 452
 453         taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
 454
 455         return (0);
 456 }
 457
 458 static void
 459 ctl_be_block_biodone(struct bio *bio)
 460 {
 461         struct ctl_be_block_io *beio;
 462         struct ctl_be_block_lun *be_lun;
 463         union ctl_io *io;
 464         int error;
 465
 466         beio = bio->bio_caller1;
 467         be_lun = beio->lun;
 468         io = beio->io;
 469
 470         DPRINTF("entered\n");
 471
 472         error = bio->bio_error;
 473         mtx_lock(&be_lun->io_lock);
 474         if (error != 0)
 475                 beio->num_errors++;
 476
 477         beio->num_bios_done++;
 478
 479         /*
 480          * XXX KDM will this cause WITNESS to complain?  Holding a lock
 481          * during the free might cause it to complain.
 482          */
 483         g_destroy_bio(bio);
 484
 485         /*
 486          * If the send complete bit isn't set, or we aren't the last I/O to
 487          * complete, then we're done.
 488          */
 489         if ((beio->send_complete == 0)
 490          || (beio->num_bios_done < beio->num_bios_sent)) {
 491                 mtx_unlock(&be_lun->io_lock);
 492                 return;
 493         }
 494
 495         /*
 496          * At this point, we've verified that we are the last I/O to
 497          * complete, so it's safe to drop the lock.
 498          */
 499         devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
 500             beio->ds_tag_type, beio->ds_trans_type,
 501             /*now*/ NULL, /*then*/&beio->ds_t0);
 502         mtx_unlock(&be_lun->io_lock);
 503
 504         /*
 505          * If there are any errors from the backing device, we fail the
 506          * entire I/O with a medium error.
 507          */
 508         if (beio->num_errors > 0) {
 509                 if (error == EOPNOTSUPP) {
 510                         ctl_set_invalid_opcode(&io->scsiio);
 511                 } else if (error == ENOSPC) {
 512                         ctl_set_space_alloc_fail(&io->scsiio);
 513                 } else if (beio->bio_cmd == BIO_FLUSH) {
 514                         /* XXX KDM is there is a better error here? */
 515                         ctl_set_internal_failure(&io->scsiio,
 516                                                  /*sks_valid*/ 1,
 517                                                  /*retry_count*/ 0xbad2);
 518                 } else
 519                         ctl_set_medium_error(&io->scsiio);
 520                 ctl_complete_beio(beio);
 521                 return;
 522         }
 523
 524         /*
 525          * If this is a write, a flush, a delete or verify, we're all done.
 526          * If this is a read, we can now send the data to the user.
 527          */
 528         if ((beio->bio_cmd == BIO_WRITE)
 529          || (beio->bio_cmd == BIO_FLUSH)
 530          || (beio->bio_cmd == BIO_DELETE)
 531          || (ARGS(io)->flags & CTL_LLF_VERIFY)) {
 532                 ctl_set_success(&io->scsiio);
 533                 ctl_complete_beio(beio);
 534         } else {
 535 #ifdef CTL_TIME_IO
 536                 getbintime(&io->io_hdr.dma_start_bt);
 537 #endif
 538                 ctl_datamove(io);
 539         }
 540 }
 541
 542 static void
 543 ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun,
 544                         struct ctl_be_block_io *beio)
 545 {
 546         union ctl_io *io = beio->io;
 547         struct mount *mountpoint;
 548         int error, lock_flags;
 549
 550         DPRINTF("entered\n");
 551
 552         binuptime(&beio->ds_t0);
 553         mtx_lock(&be_lun->io_lock);
 554         devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
 555         mtx_unlock(&be_lun->io_lock);
 556
 557         (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT);
 558
 559         if (MNT_SHARED_WRITES(mountpoint)
 560          || ((mountpoint == NULL)
 561           && MNT_SHARED_WRITES(be_lun->vn->v_mount)))
 562                 lock_flags = LK_SHARED;
 563         else
 564                 lock_flags = LK_EXCLUSIVE;
 565
 566         vn_lock(be_lun->vn, lock_flags | LK_RETRY);
 567
 568         error = VOP_FSYNC(be_lun->vn, MNT_WAIT, curthread);
 569         VOP_UNLOCK(be_lun->vn, 0);
 570
 571         vn_finished_write(mountpoint);
 572
 573         mtx_lock(&be_lun->io_lock);
 574         devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
 575             beio->ds_tag_type, beio->ds_trans_type,
 576             /*now*/ NULL, /*then*/&beio->ds_t0);
 577         mtx_unlock(&be_lun->io_lock);
 578
 579         if (error == 0)
 580                 ctl_set_success(&io->scsiio);
 581         else {
 582                 /* XXX KDM is there is a better error here? */
 583                 ctl_set_internal_failure(&io->scsiio,
 584                                          /*sks_valid*/ 1,
 585                                          /*retry_count*/ 0xbad1);
 586         }
 587
 588         ctl_complete_beio(beio);
 589 }
 590
 591 SDT_PROBE_DEFINE1(cbb, kernel, read, file_start, "uint64_t");
 592 SDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t");
 593 SDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t");
 594 SDT_PROBE_DEFINE1(cbb, kernel, write, file_done, "uint64_t");
 595
 596 static void
 597 ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun,
 598                            struct ctl_be_block_io *beio)
 599 {
 600         struct ctl_be_block_filedata *file_data;
 601         union ctl_io *io;
 602         struct uio xuio;
 603         struct iovec *xiovec;
 604         int flags;
 605         int error, i;
 606
 607         DPRINTF("entered\n");
 608
 609         file_data = &be_lun->backend.file;
 610         io = beio->io;
 611         flags = 0;
 612         if (ARGS(io)->flags & CTL_LLF_DPO)
 613                 flags |= IO_DIRECT;
 614         if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA)
 615                 flags |= IO_SYNC;
 616
 617         bzero(&xuio, sizeof(xuio));
 618         if (beio->bio_cmd == BIO_READ) {
 619                 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0);
 620                 xuio.uio_rw = UIO_READ;
 621         } else {
 622                 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0);
 623                 xuio.uio_rw = UIO_WRITE;
 624         }
 625         xuio.uio_offset = beio->io_offset;
 626         xuio.uio_resid = beio->io_len;
 627         xuio.uio_segflg = UIO_SYSSPACE;
 628         xuio.uio_iov = beio->xiovecs;
 629         xuio.uio_iovcnt = beio->num_segs;
 630         xuio.uio_td = curthread;
 631
 632         for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) {
 633                 xiovec->iov_base = beio->sg_segs[i].addr;
 634                 xiovec->iov_len = beio->sg_segs[i].len;
 635         }
 636
 637         binuptime(&beio->ds_t0);
 638         mtx_lock(&be_lun->io_lock);
 639         devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
 640         mtx_unlock(&be_lun->io_lock);
 641
 642         if (beio->bio_cmd == BIO_READ) {
 643                 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
 644
 645                 /*
 646                  * UFS pays attention to IO_DIRECT for reads.  If the
 647                  * DIRECTIO option is configured into the kernel, it calls
 648                  * ffs_rawread().  But that only works for single-segment
 649                  * uios with user space addresses.  In our case, with a
 650                  * kernel uio, it still reads into the buffer cache, but it
 651                  * will just try to release the buffer from the cache later
 652                  * on in ffs_read().
 653                  *
 654                  * ZFS does not pay attention to IO_DIRECT for reads.
 655                  *
 656                  * UFS does not pay attention to IO_SYNC for reads.
 657                  *
 658                  * ZFS pays attention to IO_SYNC (which translates into the
 659                  * Solaris define FRSYNC for zfs_read()) for reads.  It
 660                  * attempts to sync the file before reading.
 661                  *
 662                  * So, to attempt to provide some barrier semantics in the
 663                  * BIO_ORDERED case, set both IO_DIRECT and IO_SYNC.
 664                  */
 665                 error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred);
 666
 667                 VOP_UNLOCK(be_lun->vn, 0);
 668                 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0);
 669         } else {
 670                 struct mount *mountpoint;
 671                 int lock_flags;
 672
 673                 (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT);
 674
 675                 if (MNT_SHARED_WRITES(mountpoint)
 676                  || ((mountpoint == NULL)
 677                   && MNT_SHARED_WRITES(be_lun->vn->v_mount)))
 678                         lock_flags = LK_SHARED;
 679                 else
 680                         lock_flags = LK_EXCLUSIVE;
 681
 682                 vn_lock(be_lun->vn, lock_flags | LK_RETRY);
 683
 684                 /*
 685                  * UFS pays attention to IO_DIRECT for writes.  The write
 686                  * is done asynchronously.  (Normally the write would just
 687                  * get put into cache.
 688                  *
 689                  * UFS pays attention to IO_SYNC for writes.  It will
 690                  * attempt to write the buffer out synchronously if that
 691                  * flag is set.
 692                  *
 693                  * ZFS does not pay attention to IO_DIRECT for writes.
 694                  *
 695                  * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC)
 696                  * for writes.  It will flush the transaction from the
 697                  * cache before returning.
 698                  *
 699                  * So if we've got the BIO_ORDERED flag set, we want
 700                  * IO_SYNC in either the UFS or ZFS case.
 701                  */
 702                 error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred);
 703                 VOP_UNLOCK(be_lun->vn, 0);
 704
 705                 vn_finished_write(mountpoint);
 706                 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0);
 707         }
 708
 709         mtx_lock(&be_lun->io_lock);
 710         devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
 711             beio->ds_tag_type, beio->ds_trans_type,
 712             /*now*/ NULL, /*then*/&beio->ds_t0);
 713         mtx_unlock(&be_lun->io_lock);
 714
 715         /*
 716          * If we got an error, set the sense data to "MEDIUM ERROR" and
 717          * return the I/O to the user.
 718          */
 719         if (error != 0) {
 720                 char path_str[32];
 721
 722                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
 723                 printf("%s%s command returned errno %d\n", path_str,
 724                        (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error);
 725                 if (error == ENOSPC) {
 726                         ctl_set_space_alloc_fail(&io->scsiio);
 727                 } else
 728                         ctl_set_medium_error(&io->scsiio);
 729                 ctl_complete_beio(beio);
 730                 return;
 731         }
 732
 733         /*
 734          * If this is a write or a verify, we're all done.
 735          * If this is a read, we can now send the data to the user.
 736          */
 737         if ((beio->bio_cmd == BIO_WRITE) ||
 738             (ARGS(io)->flags & CTL_LLF_VERIFY)) {
 739                 ctl_set_success(&io->scsiio);
 740                 ctl_complete_beio(beio);
 741         } else {
 742 #ifdef CTL_TIME_IO
 743                 getbintime(&io->io_hdr.dma_start_bt);
 744 #endif
 745                 ctl_datamove(io);
 746         }
 747 }
 748
 749 static void
 750 ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun,
 751                            struct ctl_be_block_io *beio)
 752 {
 753         struct ctl_be_block_devdata *dev_data;
 754         union ctl_io *io;
 755         struct uio xuio;
 756         struct iovec *xiovec;
 757         int flags;
 758         int error, i;
 759
 760         DPRINTF("entered\n");
 761
 762         dev_data = &be_lun->backend.dev;
 763         io = beio->io;
 764         flags = 0;
 765         if (ARGS(io)->flags & CTL_LLF_DPO)
 766                 flags |= IO_DIRECT;
 767         if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA)
 768                 flags |= IO_SYNC;
 769
 770         bzero(&xuio, sizeof(xuio));
 771         if (beio->bio_cmd == BIO_READ) {
 772                 SDT_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0);
 773                 xuio.uio_rw = UIO_READ;
 774         } else {
 775                 SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0);
 776                 xuio.uio_rw = UIO_WRITE;
 777         }
 778         xuio.uio_offset = beio->io_offset;
 779         xuio.uio_resid = beio->io_len;
 780         xuio.uio_segflg = UIO_SYSSPACE;
 781         xuio.uio_iov = beio->xiovecs;
 782         xuio.uio_iovcnt = beio->num_segs;
 783         xuio.uio_td = curthread;
 784
 785         for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) {
 786                 xiovec->iov_base = beio->sg_segs[i].addr;
 787                 xiovec->iov_len = beio->sg_segs[i].len;
 788         }
 789
 790         binuptime(&beio->ds_t0);
 791         mtx_lock(&be_lun->io_lock);
 792         devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0);
 793         mtx_unlock(&be_lun->io_lock);
 794
 795         if (beio->bio_cmd == BIO_READ) {
 796                 error = (*dev_data->csw->d_read)(dev_data->cdev, &xuio, flags);
 797                 SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0);
 798         } else {
 799                 error = (*dev_data->csw->d_write)(dev_data->cdev, &xuio, flags);
 800                 SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0);
 801         }
 802
 803         mtx_lock(&be_lun->io_lock);
 804         devstat_end_transaction(beio->lun->disk_stats, beio->io_len,
 805             beio->ds_tag_type, beio->ds_trans_type,
 806             /*now*/ NULL, /*then*/&beio->ds_t0);
 807         mtx_unlock(&be_lun->io_lock);
 808
 809         /*
 810          * If we got an error, set the sense data to "MEDIUM ERROR" and
 811          * return the I/O to the user.
 812          */
 813         if (error != 0) {
 814                 if (error == ENOSPC) {
 815                         ctl_set_space_alloc_fail(&io->scsiio);
 816                 } else
 817                         ctl_set_medium_error(&io->scsiio);
 818                 ctl_complete_beio(beio);
 819                 return;
 820         }
 821
 822         /*
 823          * If this is a write or a verify, we're all done.
 824          * If this is a read, we can now send the data to the user.
 825          */
 826         if ((beio->bio_cmd == BIO_WRITE) ||
 827             (ARGS(io)->flags & CTL_LLF_VERIFY)) {
 828                 ctl_set_success(&io->scsiio);
 829                 ctl_complete_beio(beio);
 830         } else {
 831 #ifdef CTL_TIME_IO
 832                 getbintime(&io->io_hdr.dma_start_bt);
 833 #endif
 834                 ctl_datamove(io);
 835         }
 836 }
 837
 838 static void
 839 ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun,
 840                        struct ctl_be_block_io *beio)
 841 {
 842         struct bio *bio;
 843         union ctl_io *io;
 844         struct ctl_be_block_devdata *dev_data;
 845
 846         dev_data = &be_lun->backend.dev;
 847         io = beio->io;
 848
 849         DPRINTF("entered\n");
 850
 851         /* This can't fail, it's a blocking allocation. */
 852         bio = g_alloc_bio();
 853
 854         bio->bio_cmd        = BIO_FLUSH;
 855         bio->bio_flags     |= BIO_ORDERED;
 856         bio->bio_dev        = dev_data->cdev;
 857         bio->bio_offset     = 0;
 858         bio->bio_data       = 0;
 859         bio->bio_done       = ctl_be_block_biodone;
 860         bio->bio_caller1    = beio;
 861         bio->bio_pblkno     = 0;
 862
 863         /*
 864          * We don't need to acquire the LUN lock here, because we are only
 865          * sending one bio, and so there is no other context to synchronize
 866          * with.
 867          */
 868         beio->num_bios_sent = 1;
 869         beio->send_complete = 1;
 870
 871         binuptime(&beio->ds_t0);
 872         mtx_lock(&be_lun->io_lock);
 873         devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
 874         mtx_unlock(&be_lun->io_lock);
 875
 876         (*dev_data->csw->d_strategy)(bio);
 877 }
 878
 879 static void
 880 ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun,
 881                        struct ctl_be_block_io *beio,
 882                        uint64_t off, uint64_t len, int last)
 883 {
 884         struct bio *bio;
 885         struct ctl_be_block_devdata *dev_data;
 886         uint64_t maxlen;
 887
 888         dev_data = &be_lun->backend.dev;
 889         maxlen = LONG_MAX - (LONG_MAX % be_lun->blocksize);
 890         while (len > 0) {
 891                 bio = g_alloc_bio();
 892                 bio->bio_cmd        = BIO_DELETE;
 893                 bio->bio_dev        = dev_data->cdev;
 894                 bio->bio_offset     = off;
 895                 bio->bio_length     = MIN(len, maxlen);
 896                 bio->bio_data       = 0;
 897                 bio->bio_done       = ctl_be_block_biodone;
 898                 bio->bio_caller1    = beio;
 899                 bio->bio_pblkno     = off / be_lun->blocksize;
 900
 901                 off += bio->bio_length;
 902                 len -= bio->bio_length;
 903
 904                 mtx_lock(&be_lun->io_lock);
 905                 beio->num_bios_sent++;
 906                 if (last && len == 0)
 907                         beio->send_complete = 1;
 908                 mtx_unlock(&be_lun->io_lock);
 909
 910                 (*dev_data->csw->d_strategy)(bio);
 911         }
 912 }
 913
 914 static void
 915 ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun,
 916                        struct ctl_be_block_io *beio)
 917 {
 918         union ctl_io *io;
 919         struct ctl_be_block_devdata *dev_data;
 920         struct ctl_ptr_len_flags *ptrlen;
 921         struct scsi_unmap_desc *buf, *end;
 922         uint64_t len;
 923
 924         dev_data = &be_lun->backend.dev;
 925         io = beio->io;
 926
 927         DPRINTF("entered\n");
 928
 929         binuptime(&beio->ds_t0);
 930         mtx_lock(&be_lun->io_lock);
 931         devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
 932         mtx_unlock(&be_lun->io_lock);
 933
 934         if (beio->io_offset == -1) {
 935                 beio->io_len = 0;
 936                 ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
 937                 buf = (struct scsi_unmap_desc *)ptrlen->ptr;
 938                 end = buf + ptrlen->len / sizeof(*buf);
 939                 for (; buf < end; buf++) {
 940                         len = (uint64_t)scsi_4btoul(buf->length) *
 941                             be_lun->blocksize;
 942                         beio->io_len += len;
 943                         ctl_be_block_unmap_dev_range(be_lun, beio,
 944                             scsi_8btou64(buf->lba) * be_lun->blocksize, len,
 945                             (end - buf < 2) ? TRUE : FALSE);
 946                 }
 947         } else
 948                 ctl_be_block_unmap_dev_range(be_lun, beio,
 949                     beio->io_offset, beio->io_len, TRUE);
 950 }
 951
 952 static void
 953 ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun,
 954                           struct ctl_be_block_io *beio)
 955 {
 956         TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue);
 957         int i;
 958         struct bio *bio;
 959         struct ctl_be_block_devdata *dev_data;
 960         off_t cur_offset;
 961         int max_iosize;
 962
 963         DPRINTF("entered\n");
 964
 965         dev_data = &be_lun->backend.dev;
 966
 967         /*
 968          * We have to limit our I/O size to the maximum supported by the
 969          * backend device.  Hopefully it is MAXPHYS.  If the driver doesn't
 970          * set it properly, use DFLTPHYS.
 971          */
 972         max_iosize = dev_data->cdev->si_iosize_max;
 973         if (max_iosize < PAGE_SIZE)
 974                 max_iosize = DFLTPHYS;
 975
 976         cur_offset = beio->io_offset;
 977         for (i = 0; i < beio->num_segs; i++) {
 978                 size_t cur_size;
 979                 uint8_t *cur_ptr;
 980
 981                 cur_size = beio->sg_segs[i].len;
 982                 cur_ptr = beio->sg_segs[i].addr;
 983
 984                 while (cur_size > 0) {
 985                         /* This can't fail, it's a blocking allocation. */
 986                         bio = g_alloc_bio();
 987
 988                         KASSERT(bio != NULL, ("g_alloc_bio() failed!\n"));
 989
 990                         bio->bio_cmd = beio->bio_cmd;
 991                         bio->bio_dev = dev_data->cdev;
 992                         bio->bio_caller1 = beio;
 993                         bio->bio_length = min(cur_size, max_iosize);
 994                         bio->bio_offset = cur_offset;
 995                         bio->bio_data = cur_ptr;
 996                         bio->bio_done = ctl_be_block_biodone;
 997                         bio->bio_pblkno = cur_offset / be_lun->blocksize;
 998
 999                         cur_offset += bio->bio_length;
1000                         cur_ptr += bio->bio_length;
1001                         cur_size -= bio->bio_length;
1002
1003                         TAILQ_INSERT_TAIL(&queue, bio, bio_queue);
1004                         beio->num_bios_sent++;
1005                 }
1006         }
1007         binuptime(&beio->ds_t0);
1008         mtx_lock(&be_lun->io_lock);
1009         devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0);
1010         beio->send_complete = 1;
1011         mtx_unlock(&be_lun->io_lock);
1012
1013         /*
1014          * Fire off all allocated requests!
1015          */
1016         while ((bio = TAILQ_FIRST(&queue)) != NULL) {
1017                 TAILQ_REMOVE(&queue, bio, bio_queue);
1018                 (*dev_data->csw->d_strategy)(bio);
1019         }
1020 }
1021
1022 static uint64_t
1023 ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname)
1024 {
1025         struct ctl_be_block_devdata     *dev_data = &be_lun->backend.dev;
1026         struct diocgattr_arg    arg;
1027         int                     error;
1028
1029         if (dev_data->csw == NULL || dev_data->csw->d_ioctl == NULL)
1030                 return (UINT64_MAX);
1031         strlcpy(arg.name, attrname, sizeof(arg.name));
1032         arg.len = sizeof(arg.value.off);
1033         error = dev_data->csw->d_ioctl(dev_data->cdev,
1034             DIOCGATTR, (caddr_t)&arg, FREAD, curthread);
1035         if (error != 0)
1036                 return (UINT64_MAX);
1037         return (arg.value.off);
1038 }
1039
1040 static void
1041 ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio)
1042 {
1043         union ctl_io *io;
1044
1045         io = beio->io;
1046         ctl_free_beio(beio);
1047         if ((io->io_hdr.flags & CTL_FLAG_ABORT) ||
1048             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
1049              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
1050                 ctl_config_write_done(io);
1051                 return;
1052         }
1053
1054         ctl_be_block_config_write(io);
1055 }
1056
1057 static void
1058 ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun,
1059                             union ctl_io *io)
1060 {
1061         struct ctl_be_block_io *beio;
1062         struct ctl_be_block_softc *softc;
1063         struct ctl_lba_len_flags *lbalen;
1064         uint64_t len_left, lba;
1065         int i, seglen;
1066         uint8_t *buf, *end;
1067
1068         DPRINTF("entered\n");
1069
1070         beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1071         softc = be_lun->softc;
1072         lbalen = ARGS(beio->io);
1073
1074         if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) ||
1075             (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) {
1076                 ctl_free_beio(beio);
1077                 ctl_set_invalid_field(&io->scsiio,
1078                                       /*sks_valid*/ 1,
1079                                       /*command*/ 1,
1080                                       /*field*/ 1,
1081                                       /*bit_valid*/ 0,
1082                                       /*bit*/ 0);
1083                 ctl_config_write_done(io);
1084                 return;
1085         }
1086
1087         switch (io->scsiio.tag_type) {
1088         case CTL_TAG_ORDERED:
1089                 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1090                 break;
1091         case CTL_TAG_HEAD_OF_QUEUE:
1092                 beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1093                 break;
1094         case CTL_TAG_UNTAGGED:
1095         case CTL_TAG_SIMPLE:
1096         case CTL_TAG_ACA:
1097         default:
1098                 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1099                 break;
1100         }
1101
1102         if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) {
1103                 beio->io_offset = lbalen->lba * be_lun->blocksize;
1104                 beio->io_len = (uint64_t)lbalen->len * be_lun->blocksize;
1105                 beio->bio_cmd = BIO_DELETE;
1106                 beio->ds_trans_type = DEVSTAT_FREE;
1107
1108                 be_lun->unmap(be_lun, beio);
1109                 return;
1110         }
1111
1112         beio->bio_cmd = BIO_WRITE;
1113         beio->ds_trans_type = DEVSTAT_WRITE;
1114
1115         DPRINTF("WRITE SAME at LBA %jx len %u\n",
1116                (uintmax_t)lbalen->lba, lbalen->len);
1117
1118         len_left = (uint64_t)lbalen->len * be_lun->blocksize;
1119         for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) {
1120
1121                 /*
1122                  * Setup the S/G entry for this chunk.
1123                  */
1124                 seglen = MIN(CTLBLK_MAX_SEG, len_left);
1125                 seglen -= seglen % be_lun->blocksize;
1126                 beio->sg_segs[i].len = seglen;
1127                 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK);
1128
1129                 DPRINTF("segment %d addr %p len %zd\n", i,
1130                         beio->sg_segs[i].addr, beio->sg_segs[i].len);
1131
1132                 beio->num_segs++;
1133                 len_left -= seglen;
1134
1135                 buf = beio->sg_segs[i].addr;
1136                 end = buf + seglen;
1137                 for (; buf < end; buf += be_lun->blocksize) {
1138                         memcpy(buf, io->scsiio.kern_data_ptr, be_lun->blocksize);
1139                         if (lbalen->flags & SWS_LBDATA)
1140                                 scsi_ulto4b(lbalen->lba + lba, buf);
1141                         lba++;
1142                 }
1143         }
1144
1145         beio->io_offset = lbalen->lba * be_lun->blocksize;
1146         beio->io_len = lba * be_lun->blocksize;
1147
1148         /* We can not do all in one run. Correct and schedule rerun. */
1149         if (len_left > 0) {
1150                 lbalen->lba += lba;
1151                 lbalen->len -= lba;
1152                 beio->beio_cont = ctl_be_block_cw_done_ws;
1153         }
1154
1155         be_lun->dispatch(be_lun, beio);
1156 }
1157
1158 static void
1159 ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun,
1160                             union ctl_io *io)
1161 {
1162         struct ctl_be_block_io *beio;
1163         struct ctl_be_block_softc *softc;
1164         struct ctl_ptr_len_flags *ptrlen;
1165
1166         DPRINTF("entered\n");
1167
1168         beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1169         softc = be_lun->softc;
1170         ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
1171
1172         if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) {
1173                 ctl_free_beio(beio);
1174                 ctl_set_invalid_field(&io->scsiio,
1175                                       /*sks_valid*/ 0,
1176                                       /*command*/ 1,
1177                                       /*field*/ 0,
1178                                       /*bit_valid*/ 0,
1179                                       /*bit*/ 0);
1180                 ctl_config_write_done(io);
1181                 return;
1182         }
1183
1184         switch (io->scsiio.tag_type) {
1185         case CTL_TAG_ORDERED:
1186                 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1187                 break;
1188         case CTL_TAG_HEAD_OF_QUEUE:
1189                 beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1190                 break;
1191         case CTL_TAG_UNTAGGED:
1192         case CTL_TAG_SIMPLE:
1193         case CTL_TAG_ACA:
1194         default:
1195                 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1196                 break;
1197         }
1198
1199         beio->io_len = 0;
1200         beio->io_offset = -1;
1201
1202         beio->bio_cmd = BIO_DELETE;
1203         beio->ds_trans_type = DEVSTAT_FREE;
1204
1205         DPRINTF("UNMAP\n");
1206
1207         be_lun->unmap(be_lun, beio);
1208 }
1209
1210 static void
1211 ctl_be_block_cw_done(struct ctl_be_block_io *beio)
1212 {
1213         union ctl_io *io;
1214
1215         io = beio->io;
1216         ctl_free_beio(beio);
1217         ctl_config_write_done(io);
1218 }
1219
1220 static void
1221 ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun,
1222                          union ctl_io *io)
1223 {
1224         struct ctl_be_block_io *beio;
1225         struct ctl_be_block_softc *softc;
1226
1227         DPRINTF("entered\n");
1228
1229         softc = be_lun->softc;
1230         beio = ctl_alloc_beio(softc);
1231         beio->io = io;
1232         beio->lun = be_lun;
1233         beio->beio_cont = ctl_be_block_cw_done;
1234         PRIV(io)->ptr = (void *)beio;
1235
1236         switch (io->scsiio.cdb[0]) {
1237         case SYNCHRONIZE_CACHE:
1238         case SYNCHRONIZE_CACHE_16:
1239                 beio->bio_cmd = BIO_FLUSH;
1240                 beio->ds_trans_type = DEVSTAT_NO_DATA;
1241                 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1242                 beio->io_len = 0;
1243                 be_lun->lun_flush(be_lun, beio);
1244                 break;
1245         case WRITE_SAME_10:
1246         case WRITE_SAME_16:
1247                 ctl_be_block_cw_dispatch_ws(be_lun, io);
1248                 break;
1249         case UNMAP:
1250                 ctl_be_block_cw_dispatch_unmap(be_lun, io);
1251                 break;
1252         default:
1253                 panic("Unhandled CDB type %#x", io->scsiio.cdb[0]);
1254                 break;
1255         }
1256 }
1257
1258 SDT_PROBE_DEFINE1(cbb, kernel, read, start, "uint64_t");
1259 SDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t");
1260 SDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t");
1261 SDT_PROBE_DEFINE1(cbb, kernel, write, alloc_done, "uint64_t");
1262
1263 static void
1264 ctl_be_block_next(struct ctl_be_block_io *beio)
1265 {
1266         struct ctl_be_block_lun *be_lun;
1267         union ctl_io *io;
1268
1269         io = beio->io;
1270         be_lun = beio->lun;
1271         ctl_free_beio(beio);
1272         if ((io->io_hdr.flags & CTL_FLAG_ABORT) ||
1273             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
1274              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
1275                 ctl_data_submit_done(io);
1276                 return;
1277         }
1278
1279         io->io_hdr.status &= ~CTL_STATUS_MASK;
1280         io->io_hdr.status |= CTL_STATUS_NONE;
1281
1282         mtx_lock(&be_lun->queue_lock);
1283         /*
1284          * XXX KDM make sure that links is okay to use at this point.
1285          * Otherwise, we either need to add another field to ctl_io_hdr,
1286          * or deal with resource allocation here.
1287          */
1288         STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links);
1289         mtx_unlock(&be_lun->queue_lock);
1290
1291         taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
1292 }
1293
1294 static void
1295 ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun,
1296                            union ctl_io *io)
1297 {
1298         struct ctl_be_block_io *beio;
1299         struct ctl_be_block_softc *softc;
1300         struct ctl_lba_len_flags *lbalen;
1301         struct ctl_ptr_len_flags *bptrlen;
1302         uint64_t len_left, lbas;
1303         int i;
1304
1305         softc = be_lun->softc;
1306
1307         DPRINTF("entered\n");
1308
1309         lbalen = ARGS(io);
1310         if (lbalen->flags & CTL_LLF_WRITE) {
1311                 SDT_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0);
1312         } else {
1313                 SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0);
1314         }
1315
1316         beio = ctl_alloc_beio(softc);
1317         beio->io = io;
1318         beio->lun = be_lun;
1319         bptrlen = PRIV(io);
1320         bptrlen->ptr = (void *)beio;
1321
1322         switch (io->scsiio.tag_type) {
1323         case CTL_TAG_ORDERED:
1324                 beio->ds_tag_type = DEVSTAT_TAG_ORDERED;
1325                 break;
1326         case CTL_TAG_HEAD_OF_QUEUE:
1327                 beio->ds_tag_type = DEVSTAT_TAG_HEAD;
1328                 break;
1329         case CTL_TAG_UNTAGGED:
1330         case CTL_TAG_SIMPLE:
1331         case CTL_TAG_ACA:
1332         default:
1333                 beio->ds_tag_type = DEVSTAT_TAG_SIMPLE;
1334                 break;
1335         }
1336
1337         if (lbalen->flags & CTL_LLF_WRITE) {
1338                 beio->bio_cmd = BIO_WRITE;
1339                 beio->ds_trans_type = DEVSTAT_WRITE;
1340         } else {
1341                 beio->bio_cmd = BIO_READ;
1342                 beio->ds_trans_type = DEVSTAT_READ;
1343         }
1344
1345         DPRINTF("%s at LBA %jx len %u @%ju\n",
1346                (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE",
1347                (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len);
1348         if (lbalen->flags & CTL_LLF_COMPARE)
1349                 lbas = CTLBLK_HALF_IO_SIZE;
1350         else
1351                 lbas = CTLBLK_MAX_IO_SIZE;
1352         lbas = MIN(lbalen->len - bptrlen->len, lbas / be_lun->blocksize);
1353         beio->io_offset = (lbalen->lba + bptrlen->len) * be_lun->blocksize;
1354         beio->io_len = lbas * be_lun->blocksize;
1355         bptrlen->len += lbas;
1356
1357         for (i = 0, len_left = beio->io_len; len_left > 0; i++) {
1358                 KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)",
1359                     i, CTLBLK_MAX_SEGS));
1360
1361                 /*
1362                  * Setup the S/G entry for this chunk.
1363                  */
1364                 beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left);
1365                 beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK);
1366
1367                 DPRINTF("segment %d addr %p len %zd\n", i,
1368                         beio->sg_segs[i].addr, beio->sg_segs[i].len);
1369
1370                 /* Set up second segment for compare operation. */
1371                 if (lbalen->flags & CTL_LLF_COMPARE) {
1372                         beio->sg_segs[i + CTLBLK_HALF_SEGS].len =
1373                             beio->sg_segs[i].len;
1374                         beio->sg_segs[i + CTLBLK_HALF_SEGS].addr =
1375                             uma_zalloc(be_lun->lun_zone, M_WAITOK);
1376                 }
1377
1378                 beio->num_segs++;
1379                 len_left -= beio->sg_segs[i].len;
1380         }
1381         if (bptrlen->len < lbalen->len)
1382                 beio->beio_cont = ctl_be_block_next;
1383         io->scsiio.be_move_done = ctl_be_block_move_done;
1384         /* For compare we have separate S/G lists for read and datamove. */
1385         if (lbalen->flags & CTL_LLF_COMPARE)
1386                 io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS];
1387         else
1388                 io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs;
1389         io->scsiio.kern_data_len = beio->io_len;
1390         io->scsiio.kern_data_resid = 0;
1391         io->scsiio.kern_sg_entries = beio->num_segs;
1392         io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST;
1393
1394         /*
1395          * For the read case, we need to read the data into our buffers and
1396          * then we can send it back to the user.  For the write case, we
1397          * need to get the data from the user first.
1398          */
1399         if (beio->bio_cmd == BIO_READ) {
1400                 SDT_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0);
1401                 be_lun->dispatch(be_lun, beio);
1402         } else {
1403                 SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0);
1404 #ifdef CTL_TIME_IO
1405                 getbintime(&io->io_hdr.dma_start_bt);
1406 #endif
1407                 ctl_datamove(io);
1408         }
1409 }
1410
1411 static void
1412 ctl_be_block_worker(void *context, int pending)
1413 {
1414         struct ctl_be_block_lun *be_lun;
1415         struct ctl_be_block_softc *softc;
1416         union ctl_io *io;
1417
1418         be_lun = (struct ctl_be_block_lun *)context;
1419         softc = be_lun->softc;
1420
1421         DPRINTF("entered\n");
1422
1423         mtx_lock(&be_lun->queue_lock);
1424         for (;;) {
1425                 io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue);
1426                 if (io != NULL) {
1427                         struct ctl_be_block_io *beio;
1428
1429                         DPRINTF("datamove queue\n");
1430
1431                         STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr,
1432                                       ctl_io_hdr, links);
1433
1434                         mtx_unlock(&be_lun->queue_lock);
1435
1436                         beio = (struct ctl_be_block_io *)PRIV(io)->ptr;
1437
1438                         be_lun->dispatch(be_lun, beio);
1439
1440                         mtx_lock(&be_lun->queue_lock);
1441                         continue;
1442                 }
1443                 io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue);
1444                 if (io != NULL) {
1445
1446                         DPRINTF("config write queue\n");
1447
1448                         STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr,
1449                                       ctl_io_hdr, links);
1450
1451                         mtx_unlock(&be_lun->queue_lock);
1452
1453                         ctl_be_block_cw_dispatch(be_lun, io);
1454
1455                         mtx_lock(&be_lun->queue_lock);
1456                         continue;
1457                 }
1458                 io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue);
1459                 if (io != NULL) {
1460                         DPRINTF("input queue\n");
1461
1462                         STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr,
1463                                       ctl_io_hdr, links);
1464                         mtx_unlock(&be_lun->queue_lock);
1465
1466                         /*
1467                          * We must drop the lock, since this routine and
1468                          * its children may sleep.
1469                          */
1470                         ctl_be_block_dispatch(be_lun, io);
1471
1472                         mtx_lock(&be_lun->queue_lock);
1473                         continue;
1474                 }
1475
1476                 /*
1477                  * If we get here, there is no work left in the queues, so
1478                  * just break out and let the task queue go to sleep.
1479                  */
1480                 break;
1481         }
1482         mtx_unlock(&be_lun->queue_lock);
1483 }
1484
1485 /*
1486  * Entry point from CTL to the backend for I/O.  We queue everything to a
1487  * work thread, so this just puts the I/O on a queue and wakes up the
1488  * thread.
1489  */
1490 static int
1491 ctl_be_block_submit(union ctl_io *io)
1492 {
1493         struct ctl_be_block_lun *be_lun;
1494         struct ctl_be_lun *ctl_be_lun;
1495
1496         DPRINTF("entered\n");
1497
1498         ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[
1499                 CTL_PRIV_BACKEND_LUN].ptr;
1500         be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun;
1501
1502         /*
1503          * Make sure we only get SCSI I/O.
1504          */
1505         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type "
1506                 "%#x) encountered", io->io_hdr.io_type));
1507
1508         PRIV(io)->len = 0;
1509
1510         mtx_lock(&be_lun->queue_lock);
1511         /*
1512          * XXX KDM make sure that links is okay to use at this point.
1513          * Otherwise, we either need to add another field to ctl_io_hdr,
1514          * or deal with resource allocation here.
1515          */
1516         STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links);
1517         mtx_unlock(&be_lun->queue_lock);
1518         taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
1519
1520         return (CTL_RETVAL_COMPLETE);
1521 }
1522
1523 static int
1524 ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
1525                         int flag, struct thread *td)
1526 {
1527         struct ctl_be_block_softc *softc;
1528         int error;
1529
1530         softc = &backend_block_softc;
1531
1532         error = 0;
1533
1534         switch (cmd) {
1535         case CTL_LUN_REQ: {
1536                 struct ctl_lun_req *lun_req;
1537
1538                 lun_req = (struct ctl_lun_req *)addr;
1539
1540                 switch (lun_req->reqtype) {
1541                 case CTL_LUNREQ_CREATE:
1542                         error = ctl_be_block_create(softc, lun_req);
1543                         break;
1544                 case CTL_LUNREQ_RM:
1545                         error = ctl_be_block_rm(softc, lun_req);
1546                         break;
1547                 case CTL_LUNREQ_MODIFY:
1548                         error = ctl_be_block_modify(softc, lun_req);
1549                         break;
1550                 default:
1551                         lun_req->status = CTL_LUN_ERROR;
1552                         snprintf(lun_req->error_str, sizeof(lun_req->error_str),
1553                                  "invalid LUN request type %d",
1554                                  lun_req->reqtype);
1555                         break;
1556                 }
1557                 break;
1558         }
1559         default:
1560                 error = ENOTTY;
1561                 break;
1562         }
1563
1564         return (error);
1565 }
1566
1567 static int
1568 ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
1569 {
1570         struct ctl_be_block_filedata *file_data;
1571         struct ctl_lun_create_params *params;
1572         struct vattr                  vattr;
1573         off_t                         pss;
1574         int                           error;
1575
1576         error = 0;
1577         file_data = &be_lun->backend.file;
1578         params = &be_lun->params;
1579
1580         be_lun->dev_type = CTL_BE_BLOCK_FILE;
1581         be_lun->dispatch = ctl_be_block_dispatch_file;
1582         be_lun->lun_flush = ctl_be_block_flush_file;
1583
1584         error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred);
1585         if (error != 0) {
1586                 snprintf(req->error_str, sizeof(req->error_str),
1587                          "error calling VOP_GETATTR() for file %s",
1588                          be_lun->dev_path);
1589                 return (error);
1590         }
1591
1592         /*
1593          * Verify that we have the ability to upgrade to exclusive
1594          * access on this file so we can trap errors at open instead
1595          * of reporting them during first access.
1596          */
1597         if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) {
1598                 vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY);
1599                 if (be_lun->vn->v_iflag & VI_DOOMED) {
1600                         error = EBADF;
1601                         snprintf(req->error_str, sizeof(req->error_str),
1602                                  "error locking file %s", be_lun->dev_path);
1603                         return (error);
1604                 }
1605         }
1606
1607
1608         file_data->cred = crhold(curthread->td_ucred);
1609         if (params->lun_size_bytes != 0)
1610                 be_lun->size_bytes = params->lun_size_bytes;
1611         else
1612                 be_lun->size_bytes = vattr.va_size;
1613         /*
1614          * We set the multi thread flag for file operations because all
1615          * filesystems (in theory) are capable of allowing multiple readers
1616          * of a file at once.  So we want to get the maximum possible
1617          * concurrency.
1618          */
1619         be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD;
1620
1621         /*
1622          * For files we can use any logical block size.  Prefer 512 bytes
1623          * for compatibility reasons.  If file's vattr.va_blocksize
1624          * (preferred I/O block size) is bigger and multiple to chosen
1625          * logical block size -- report it as physical block size.
1626          */
1627         if (params->blocksize_bytes != 0)
1628                 be_lun->blocksize = params->blocksize_bytes;
1629         else
1630                 be_lun->blocksize = 512;
1631         pss = vattr.va_blocksize / be_lun->blocksize;
1632         if ((pss > 0) && (pss * be_lun->blocksize == vattr.va_blocksize) &&
1633             ((pss & (pss - 1)) == 0)) {
1634                 be_lun->pblockexp = fls(pss) - 1;
1635                 be_lun->pblockoff = 0;
1636         }
1637
1638         /*
1639          * Sanity check.  The media size has to be at least one
1640          * sector long.
1641          */
1642         if (be_lun->size_bytes < be_lun->blocksize) {
1643                 error = EINVAL;
1644                 snprintf(req->error_str, sizeof(req->error_str),
1645                          "file %s size %ju < block size %u", be_lun->dev_path,
1646                          (uintmax_t)be_lun->size_bytes, be_lun->blocksize);
1647         }
1648         return (error);
1649 }
1650
1651 static int
1652 ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
1653 {
1654         struct ctl_lun_create_params *params;
1655         struct vattr                  vattr;
1656         struct cdev                  *dev;
1657         struct cdevsw                *devsw;
1658         int                           error;
1659         off_t                         ps, pss, po, pos;
1660
1661         params = &be_lun->params;
1662
1663         be_lun->dev_type = CTL_BE_BLOCK_DEV;
1664         be_lun->backend.dev.cdev = be_lun->vn->v_rdev;
1665         be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev,
1666                                              &be_lun->backend.dev.dev_ref);
1667         if (be_lun->backend.dev.csw == NULL)
1668                 panic("Unable to retrieve device switch");
1669         if (strcmp(be_lun->backend.dev.csw->d_name, "zvol") == 0)
1670                 be_lun->dispatch = ctl_be_block_dispatch_zvol;
1671         else
1672                 be_lun->dispatch = ctl_be_block_dispatch_dev;
1673         be_lun->lun_flush = ctl_be_block_flush_dev;
1674         be_lun->unmap = ctl_be_block_unmap_dev;
1675         be_lun->getattr = ctl_be_block_getattr_dev;
1676
1677         error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED);
1678         if (error) {
1679                 snprintf(req->error_str, sizeof(req->error_str),
1680                          "error getting vnode attributes for device %s",
1681                          be_lun->dev_path);
1682                 return (error);
1683         }
1684
1685         dev = be_lun->vn->v_rdev;
1686         devsw = dev->si_devsw;
1687         if (!devsw->d_ioctl) {
1688                 snprintf(req->error_str, sizeof(req->error_str),
1689                          "no d_ioctl for device %s!",
1690                          be_lun->dev_path);
1691                 return (ENODEV);
1692         }
1693
1694         error = devsw->d_ioctl(dev, DIOCGSECTORSIZE,
1695                                (caddr_t)&be_lun->blocksize, FREAD,
1696                                curthread);
1697         if (error) {
1698                 snprintf(req->error_str, sizeof(req->error_str),
1699                          "error %d returned for DIOCGSECTORSIZE ioctl "
1700                          "on %s!", error, be_lun->dev_path);
1701                 return (error);
1702         }
1703
1704         /*
1705          * If the user has asked for a blocksize that is greater than the
1706          * backing device's blocksize, we can do it only if the blocksize
1707          * the user is asking for is an even multiple of the underlying
1708          * device's blocksize.
1709          */
1710         if ((params->blocksize_bytes != 0)
1711          && (params->blocksize_bytes > be_lun->blocksize)) {
1712                 uint32_t bs_multiple, tmp_blocksize;
1713
1714                 bs_multiple = params->blocksize_bytes / be_lun->blocksize;
1715
1716                 tmp_blocksize = bs_multiple * be_lun->blocksize;
1717
1718                 if (tmp_blocksize == params->blocksize_bytes) {
1719                         be_lun->blocksize = params->blocksize_bytes;
1720                 } else {
1721                         snprintf(req->error_str, sizeof(req->error_str),
1722                                  "requested blocksize %u is not an even "
1723                                  "multiple of backing device blocksize %u",
1724                                  params->blocksize_bytes,
1725                                  be_lun->blocksize);
1726                         return (EINVAL);
1727
1728                 }
1729         } else if ((params->blocksize_bytes != 0)
1730                 && (params->blocksize_bytes != be_lun->blocksize)) {
1731                 snprintf(req->error_str, sizeof(req->error_str),
1732                          "requested blocksize %u < backing device "
1733                          "blocksize %u", params->blocksize_bytes,
1734                          be_lun->blocksize);
1735                 return (EINVAL);
1736         }
1737
1738         error = devsw->d_ioctl(dev, DIOCGMEDIASIZE,
1739                                (caddr_t)&be_lun->size_bytes, FREAD,
1740                                curthread);
1741         if (error) {
1742                 snprintf(req->error_str, sizeof(req->error_str),
1743                          "error %d returned for DIOCGMEDIASIZE "
1744                          " ioctl on %s!", error,
1745                          be_lun->dev_path);
1746                 return (error);
1747         }
1748
1749         if (params->lun_size_bytes != 0) {
1750                 if (params->lun_size_bytes > be_lun->size_bytes) {
1751                         snprintf(req->error_str, sizeof(req->error_str),
1752                                  "requested LUN size %ju > backing device "
1753                                  "size %ju",
1754                                  (uintmax_t)params->lun_size_bytes,
1755                                  (uintmax_t)be_lun->size_bytes);
1756                         return (EINVAL);
1757                 }
1758
1759                 be_lun->size_bytes = params->lun_size_bytes;
1760         }
1761
1762         error = devsw->d_ioctl(dev, DIOCGSTRIPESIZE,
1763                                (caddr_t)&ps, FREAD, curthread);
1764         if (error)
1765                 ps = po = 0;
1766         else {
1767                 error = devsw->d_ioctl(dev, DIOCGSTRIPEOFFSET,
1768                                        (caddr_t)&po, FREAD, curthread);
1769                 if (error)
1770                         po = 0;
1771         }
1772         pss = ps / be_lun->blocksize;
1773         pos = po / be_lun->blocksize;
1774         if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) &&
1775             ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) {
1776                 be_lun->pblockexp = fls(pss) - 1;
1777                 be_lun->pblockoff = (pss - pos) % pss;
1778         }
1779
1780         return (0);
1781 }
1782
1783 static int
1784 ctl_be_block_close(struct ctl_be_block_lun *be_lun)
1785 {
1786         DROP_GIANT();
1787         if (be_lun->vn) {
1788                 int flags = FREAD | FWRITE;
1789
1790                 switch (be_lun->dev_type) {
1791                 case CTL_BE_BLOCK_DEV:
1792                         if (be_lun->backend.dev.csw) {
1793                                 dev_relthread(be_lun->backend.dev.cdev,
1794                                               be_lun->backend.dev.dev_ref);
1795                                 be_lun->backend.dev.csw  = NULL;
1796                                 be_lun->backend.dev.cdev = NULL;
1797                         }
1798                         break;
1799                 case CTL_BE_BLOCK_FILE:
1800                         break;
1801                 case CTL_BE_BLOCK_NONE:
1802                         break;
1803                 default:
1804                         panic("Unexpected backend type.");
1805                         break;
1806                 }
1807
1808                 (void)vn_close(be_lun->vn, flags, NOCRED, curthread);
1809                 be_lun->vn = NULL;
1810
1811                 switch (be_lun->dev_type) {
1812                 case CTL_BE_BLOCK_DEV:
1813                         break;
1814                 case CTL_BE_BLOCK_FILE:
1815                         if (be_lun->backend.file.cred != NULL) {
1816                                 crfree(be_lun->backend.file.cred);
1817                                 be_lun->backend.file.cred = NULL;
1818                         }
1819                         break;
1820                 case CTL_BE_BLOCK_NONE:
1821                         break;
1822                 default:
1823                         panic("Unexpected backend type.");
1824                         break;
1825                 }
1826                 be_lun->dev_type = CTL_BE_BLOCK_NONE;
1827         }
1828         PICKUP_GIANT();
1829
1830         return (0);
1831 }
1832
1833 static int
1834 ctl_be_block_open(struct ctl_be_block_softc *softc,
1835                        struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req)
1836 {
1837         struct nameidata nd;
1838         int              flags;
1839         int              error;
1840
1841         /*
1842          * XXX KDM allow a read-only option?
1843          */
1844         flags = FREAD | FWRITE;
1845         error = 0;
1846
1847         if (rootvnode == NULL) {
1848                 snprintf(req->error_str, sizeof(req->error_str),
1849                          "Root filesystem is not mounted");
1850                 return (1);
1851         }
1852
1853         if (!curthread->td_proc->p_fd->fd_cdir) {
1854                 curthread->td_proc->p_fd->fd_cdir = rootvnode;
1855                 VREF(rootvnode);
1856         }
1857         if (!curthread->td_proc->p_fd->fd_rdir) {
1858                 curthread->td_proc->p_fd->fd_rdir = rootvnode;
1859                 VREF(rootvnode);
1860         }
1861         if (!curthread->td_proc->p_fd->fd_jdir) {
1862                 curthread->td_proc->p_fd->fd_jdir = rootvnode;
1863                 VREF(rootvnode);
1864         }
1865
1866  again:
1867         NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread);
1868         error = vn_open(&nd, &flags, 0, NULL);
1869         if (error) {
1870                 /*
1871                  * This is the only reasonable guess we can make as far as
1872                  * path if the user doesn't give us a fully qualified path.
1873                  * If they want to specify a file, they need to specify the
1874                  * full path.
1875                  */
1876                 if (be_lun->dev_path[0] != '/') {
1877                         char *dev_path = "/dev/";
1878                         char *dev_name;
1879
1880                         /* Try adding device path at beginning of name */
1881                         dev_name = malloc(strlen(be_lun->dev_path)
1882                                         + strlen(dev_path) + 1,
1883                                           M_CTLBLK, M_WAITOK);
1884                         if (dev_name) {
1885                                 sprintf(dev_name, "%s%s", dev_path,
1886                                         be_lun->dev_path);
1887                                 free(be_lun->dev_path, M_CTLBLK);
1888                                 be_lun->dev_path = dev_name;
1889                                 goto again;
1890                         }
1891                 }
1892                 snprintf(req->error_str, sizeof(req->error_str),
1893                     "error opening %s: %d", be_lun->dev_path, error);
1894                 return (error);
1895         }
1896
1897         NDFREE(&nd, NDF_ONLY_PNBUF);
1898
1899         be_lun->vn = nd.ni_vp;
1900
1901         /* We only support disks and files. */
1902         if (vn_isdisk(be_lun->vn, &error)) {
1903                 error = ctl_be_block_open_dev(be_lun, req);
1904         } else if (be_lun->vn->v_type == VREG) {
1905                 error = ctl_be_block_open_file(be_lun, req);
1906         } else {
1907                 error = EINVAL;
1908                 snprintf(req->error_str, sizeof(req->error_str),
1909                          "%s is not a disk or plain file", be_lun->dev_path);
1910         }
1911         VOP_UNLOCK(be_lun->vn, 0);
1912
1913         if (error != 0) {
1914                 ctl_be_block_close(be_lun);
1915                 return (error);
1916         }
1917
1918         be_lun->blocksize_shift = fls(be_lun->blocksize) - 1;
1919         be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift;
1920
1921         return (0);
1922 }
1923
1924 static int
1925 ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
1926 {
1927         struct ctl_be_block_lun *be_lun;
1928         struct ctl_lun_create_params *params;
1929         char num_thread_str[16];
1930         char tmpstr[32];
1931         char *value;
1932         int retval, num_threads, unmap;
1933         int tmp_num_threads;
1934
1935         params = &req->reqdata.create;
1936         retval = 0;
1937         req->status = CTL_LUN_OK;
1938
1939         num_threads = cbb_num_threads;
1940
1941         be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK);
1942
1943         be_lun->params = req->reqdata.create;
1944         be_lun->softc = softc;
1945         STAILQ_INIT(&be_lun->input_queue);
1946         STAILQ_INIT(&be_lun->config_write_queue);
1947         STAILQ_INIT(&be_lun->datamove_queue);
1948         sprintf(be_lun->lunname, "cblk%d", softc->num_luns);
1949         mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF);
1950         mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF);
1951         ctl_init_opts(&be_lun->ctl_be_lun.options,
1952             req->num_be_args, req->kern_be_args);
1953
1954         be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG,
1955             NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0);
1956
1957         if (be_lun->lun_zone == NULL) {
1958                 snprintf(req->error_str, sizeof(req->error_str),
1959                          "error allocating UMA zone");
1960                 goto bailout_error;
1961         }
1962
1963         if (params->flags & CTL_LUN_FLAG_DEV_TYPE)
1964                 be_lun->ctl_be_lun.lun_type = params->device_type;
1965         else
1966                 be_lun->ctl_be_lun.lun_type = T_DIRECT;
1967
1968         if (be_lun->ctl_be_lun.lun_type == T_DIRECT) {
1969                 value = ctl_get_opt(&be_lun->ctl_be_lun.options, "file");
1970                 if (value == NULL) {
1971                         snprintf(req->error_str, sizeof(req->error_str),
1972                                  "no file argument specified");
1973                         goto bailout_error;
1974                 }
1975                 be_lun->dev_path = strdup(value, M_CTLBLK);
1976                 be_lun->blocksize = 512;
1977                 be_lun->blocksize_shift = fls(be_lun->blocksize) - 1;
1978
1979                 retval = ctl_be_block_open(softc, be_lun, req);
1980                 if (retval != 0) {
1981                         retval = 0;
1982                         req->status = CTL_LUN_WARNING;
1983                 }
1984         } else {
1985                 /*
1986                  * For processor devices, we don't have any size.
1987                  */
1988                 be_lun->blocksize = 0;
1989                 be_lun->pblockexp = 0;
1990                 be_lun->pblockoff = 0;
1991                 be_lun->size_blocks = 0;
1992                 be_lun->size_bytes = 0;
1993                 be_lun->ctl_be_lun.maxlba = 0;
1994
1995                 /*
1996                  * Default to just 1 thread for processor devices.
1997                  */
1998                 num_threads = 1;
1999         }
2000
2001         /*
2002          * XXX This searching loop might be refactored to be combined with
2003          * the loop above,
2004          */
2005         value = ctl_get_opt(&be_lun->ctl_be_lun.options, "num_threads");
2006         if (value != NULL) {
2007                 tmp_num_threads = strtol(value, NULL, 0);
2008
2009                 /*
2010                  * We don't let the user specify less than one
2011                  * thread, but hope he's clueful enough not to
2012                  * specify 1000 threads.
2013                  */
2014                 if (tmp_num_threads < 1) {
2015                         snprintf(req->error_str, sizeof(req->error_str),
2016                                  "invalid number of threads %s",
2017                                  num_thread_str);
2018                         goto bailout_error;
2019                 }
2020                 num_threads = tmp_num_threads;
2021         }
2022         unmap = (be_lun->dispatch == ctl_be_block_dispatch_zvol);
2023         value = ctl_get_opt(&be_lun->ctl_be_lun.options, "unmap");
2024         if (value != NULL)
2025                 unmap = (strcmp(value, "on") == 0);
2026
2027         be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED;
2028         be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY;
2029         if (be_lun->vn == NULL)
2030                 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_OFFLINE;
2031         if (unmap)
2032                 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP;
2033         be_lun->ctl_be_lun.be_lun = be_lun;
2034         be_lun->ctl_be_lun.maxlba = (be_lun->size_blocks == 0) ?
2035             0 : (be_lun->size_blocks - 1);
2036         be_lun->ctl_be_lun.blocksize = be_lun->blocksize;
2037         be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp;
2038         be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff;
2039         if (be_lun->dispatch == ctl_be_block_dispatch_zvol &&
2040             be_lun->blocksize != 0)
2041                 be_lun->ctl_be_lun.atomicblock = CTLBLK_MAX_IO_SIZE /
2042                     be_lun->blocksize;
2043         /* Tell the user the blocksize we ended up using */
2044         params->lun_size_bytes = be_lun->size_bytes;
2045         params->blocksize_bytes = be_lun->blocksize;
2046         if (params->flags & CTL_LUN_FLAG_ID_REQ) {
2047                 be_lun->ctl_be_lun.req_lun_id = params->req_lun_id;
2048                 be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ;
2049         } else
2050                 be_lun->ctl_be_lun.req_lun_id = 0;
2051
2052         be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown;
2053         be_lun->ctl_be_lun.lun_config_status =
2054                 ctl_be_block_lun_config_status;
2055         be_lun->ctl_be_lun.be = &ctl_be_block_driver;
2056
2057         if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) {
2058                 snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d",
2059                          softc->num_luns);
2060                 strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr,
2061                         ctl_min(sizeof(be_lun->ctl_be_lun.serial_num),
2062                         sizeof(tmpstr)));
2063
2064                 /* Tell the user what we used for a serial number */
2065                 strncpy((char *)params->serial_num, tmpstr,
2066                         ctl_min(sizeof(params->serial_num), sizeof(tmpstr)));
2067         } else {
2068                 strncpy((char *)be_lun->ctl_be_lun.serial_num,
2069                         params->serial_num,
2070                         ctl_min(sizeof(be_lun->ctl_be_lun.serial_num),
2071                         sizeof(params->serial_num)));
2072         }
2073         if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) {
2074                 snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns);
2075                 strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr,
2076                         ctl_min(sizeof(be_lun->ctl_be_lun.device_id),
2077                         sizeof(tmpstr)));
2078
2079                 /* Tell the user what we used for a device ID */
2080                 strncpy((char *)params->device_id, tmpstr,
2081                         ctl_min(sizeof(params->device_id), sizeof(tmpstr)));
2082         } else {
2083                 strncpy((char *)be_lun->ctl_be_lun.device_id,
2084                         params->device_id,
2085                         ctl_min(sizeof(be_lun->ctl_be_lun.device_id),
2086                                 sizeof(params->device_id)));
2087         }
2088
2089         TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun);
2090
2091         be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK,
2092             taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue);
2093
2094         if (be_lun->io_taskqueue == NULL) {
2095                 snprintf(req->error_str, sizeof(req->error_str),
2096                          "unable to create taskqueue");
2097                 goto bailout_error;
2098         }
2099
2100         /*
2101          * Note that we start the same number of threads by default for
2102          * both the file case and the block device case.  For the file
2103          * case, we need multiple threads to allow concurrency, because the
2104          * vnode interface is designed to be a blocking interface.  For the
2105          * block device case, ZFS zvols at least will block the caller's
2106          * context in many instances, and so we need multiple threads to
2107          * overcome that problem.  Other block devices don't need as many
2108          * threads, but they shouldn't cause too many problems.
2109          *
2110          * If the user wants to just have a single thread for a block
2111          * device, he can specify that when the LUN is created, or change
2112          * the tunable/sysctl to alter the default number of threads.
2113          */
2114         retval = taskqueue_start_threads(&be_lun->io_taskqueue,
2115                                          /*num threads*/num_threads,
2116                                          /*priority*/PWAIT,
2117                                          /*thread name*/
2118                                          "%s taskq", be_lun->lunname);
2119
2120         if (retval != 0)
2121                 goto bailout_error;
2122
2123         be_lun->num_threads = num_threads;
2124
2125         mtx_lock(&softc->lock);
2126         softc->num_luns++;
2127         STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links);
2128
2129         mtx_unlock(&softc->lock);
2130
2131         retval = ctl_add_lun(&be_lun->ctl_be_lun);
2132         if (retval != 0) {
2133                 mtx_lock(&softc->lock);
2134                 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun,
2135                               links);
2136                 softc->num_luns--;
2137                 mtx_unlock(&softc->lock);
2138                 snprintf(req->error_str, sizeof(req->error_str),
2139                          "ctl_add_lun() returned error %d, see dmesg for "
2140                          "details", retval);
2141                 retval = 0;
2142                 goto bailout_error;
2143         }
2144
2145         mtx_lock(&softc->lock);
2146
2147         /*
2148          * Tell the config_status routine that we're waiting so it won't
2149          * clean up the LUN in the event of an error.
2150          */
2151         be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING;
2152
2153         while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) {
2154                 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0);
2155                 if (retval == EINTR)
2156                         break;
2157         }
2158         be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING;
2159
2160         if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) {
2161                 snprintf(req->error_str, sizeof(req->error_str),
2162                          "LUN configuration error, see dmesg for details");
2163                 STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun,
2164                               links);
2165                 softc->num_luns--;
2166                 mtx_unlock(&softc->lock);
2167                 goto bailout_error;
2168         } else {
2169                 params->req_lun_id = be_lun->ctl_be_lun.lun_id;
2170         }
2171
2172         mtx_unlock(&softc->lock);
2173
2174         be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id,
2175                                                be_lun->blocksize,
2176                                                DEVSTAT_ALL_SUPPORTED,
2177                                                be_lun->ctl_be_lun.lun_type
2178                                                | DEVSTAT_TYPE_IF_OTHER,
2179                                                DEVSTAT_PRIORITY_OTHER);
2180
2181         return (retval);
2182
2183 bailout_error:
2184         req->status = CTL_LUN_ERROR;
2185
2186         if (be_lun->io_taskqueue != NULL)
2187                 taskqueue_free(be_lun->io_taskqueue);
2188         ctl_be_block_close(be_lun);
2189         if (be_lun->dev_path != NULL)
2190                 free(be_lun->dev_path, M_CTLBLK);
2191         if (be_lun->lun_zone != NULL)
2192                 uma_zdestroy(be_lun->lun_zone);
2193         ctl_free_opts(&be_lun->ctl_be_lun.options);
2194         mtx_destroy(&be_lun->queue_lock);
2195         mtx_destroy(&be_lun->io_lock);
2196         free(be_lun, M_CTLBLK);
2197
2198         return (retval);
2199 }
2200
2201 static int
2202 ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2203 {
2204         struct ctl_lun_rm_params *params;
2205         struct ctl_be_block_lun *be_lun;
2206         int retval;
2207
2208         params = &req->reqdata.rm;
2209
2210         mtx_lock(&softc->lock);
2211
2212         be_lun = NULL;
2213
2214         STAILQ_FOREACH(be_lun, &softc->lun_list, links) {
2215                 if (be_lun->ctl_be_lun.lun_id == params->lun_id)
2216                         break;
2217         }
2218         mtx_unlock(&softc->lock);
2219
2220         if (be_lun == NULL) {
2221                 snprintf(req->error_str, sizeof(req->error_str),
2222                          "LUN %u is not managed by the block backend",
2223                          params->lun_id);
2224                 goto bailout_error;
2225         }
2226
2227         retval = ctl_disable_lun(&be_lun->ctl_be_lun);
2228
2229         if (retval != 0) {
2230                 snprintf(req->error_str, sizeof(req->error_str),
2231                          "error %d returned from ctl_disable_lun() for "
2232                          "LUN %d", retval, params->lun_id);
2233                 goto bailout_error;
2234
2235         }
2236
2237         retval = ctl_invalidate_lun(&be_lun->ctl_be_lun);
2238         if (retval != 0) {
2239                 snprintf(req->error_str, sizeof(req->error_str),
2240                          "error %d returned from ctl_invalidate_lun() for "
2241                          "LUN %d", retval, params->lun_id);
2242                 goto bailout_error;
2243         }
2244
2245         mtx_lock(&softc->lock);
2246
2247         be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING;
2248
2249         while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) {
2250                 retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0);
2251                 if (retval == EINTR)
2252                         break;
2253         }
2254
2255         be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING;
2256
2257         if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) {
2258                 snprintf(req->error_str, sizeof(req->error_str),
2259                          "interrupted waiting for LUN to be freed");
2260                 mtx_unlock(&softc->lock);
2261                 goto bailout_error;
2262         }
2263
2264         STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links);
2265
2266         softc->num_luns--;
2267         mtx_unlock(&softc->lock);
2268
2269         taskqueue_drain(be_lun->io_taskqueue, &be_lun->io_task);
2270
2271         taskqueue_free(be_lun->io_taskqueue);
2272
2273         ctl_be_block_close(be_lun);
2274
2275         if (be_lun->disk_stats != NULL)
2276                 devstat_remove_entry(be_lun->disk_stats);
2277
2278         uma_zdestroy(be_lun->lun_zone);
2279
2280         ctl_free_opts(&be_lun->ctl_be_lun.options);
2281         free(be_lun->dev_path, M_CTLBLK);
2282         mtx_destroy(&be_lun->queue_lock);
2283         mtx_destroy(&be_lun->io_lock);
2284         free(be_lun, M_CTLBLK);
2285
2286         req->status = CTL_LUN_OK;
2287
2288         return (0);
2289
2290 bailout_error:
2291
2292         req->status = CTL_LUN_ERROR;
2293
2294         return (0);
2295 }
2296
2297 static int
2298 ctl_be_block_modify_file(struct ctl_be_block_lun *be_lun,
2299                          struct ctl_lun_req *req)
2300 {
2301         struct vattr vattr;
2302         int error;
2303         struct ctl_lun_create_params *params = &be_lun->params;
2304
2305         if (params->lun_size_bytes != 0) {
2306                 be_lun->size_bytes = params->lun_size_bytes;
2307         } else  {
2308                 vn_lock(be_lun->vn, LK_SHARED | LK_RETRY);
2309                 error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred);
2310                 VOP_UNLOCK(be_lun->vn, 0);
2311                 if (error != 0) {
2312                         snprintf(req->error_str, sizeof(req->error_str),
2313                                  "error calling VOP_GETATTR() for file %s",
2314                                  be_lun->dev_path);
2315                         return (error);
2316                 }
2317
2318                 be_lun->size_bytes = vattr.va_size;
2319         }
2320
2321         return (0);
2322 }
2323
2324 static int
2325 ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun,
2326                         struct ctl_lun_req *req)
2327 {
2328         struct ctl_be_block_devdata *dev_data;
2329         int error;
2330         struct ctl_lun_create_params *params = &be_lun->params;
2331         uint64_t size_bytes;
2332
2333         dev_data = &be_lun->backend.dev;
2334         if (!dev_data->csw->d_ioctl) {
2335                 snprintf(req->error_str, sizeof(req->error_str),
2336                          "no d_ioctl for device %s!", be_lun->dev_path);
2337                 return (ENODEV);
2338         }
2339
2340         error = dev_data->csw->d_ioctl(dev_data->cdev, DIOCGMEDIASIZE,
2341                                (caddr_t)&size_bytes, FREAD,
2342                                curthread);
2343         if (error) {
2344                 snprintf(req->error_str, sizeof(req->error_str),
2345                          "error %d returned for DIOCGMEDIASIZE ioctl "
2346                          "on %s!", error, be_lun->dev_path);
2347                 return (error);
2348         }
2349
2350         if (params->lun_size_bytes != 0) {
2351                 if (params->lun_size_bytes > size_bytes) {
2352                         snprintf(req->error_str, sizeof(req->error_str),
2353                                  "requested LUN size %ju > backing device "
2354                                  "size %ju",
2355                                  (uintmax_t)params->lun_size_bytes,
2356                                  (uintmax_t)size_bytes);
2357                         return (EINVAL);
2358                 }
2359
2360                 be_lun->size_bytes = params->lun_size_bytes;
2361         } else {
2362                 be_lun->size_bytes = size_bytes;
2363         }
2364
2365         return (0);
2366 }
2367
2368 static int
2369 ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req)
2370 {
2371         struct ctl_lun_modify_params *params;
2372         struct ctl_be_block_lun *be_lun;
2373         uint64_t oldsize;
2374         int error;
2375
2376         params = &req->reqdata.modify;
2377
2378         mtx_lock(&softc->lock);
2379         be_lun = NULL;
2380         STAILQ_FOREACH(be_lun, &softc->lun_list, links) {
2381                 if (be_lun->ctl_be_lun.lun_id == params->lun_id)
2382                         break;
2383         }
2384         mtx_unlock(&softc->lock);
2385
2386         if (be_lun == NULL) {
2387                 snprintf(req->error_str, sizeof(req->error_str),
2388                          "LUN %u is not managed by the block backend",
2389                          params->lun_id);
2390                 goto bailout_error;
2391         }
2392
2393         be_lun->params.lun_size_bytes = params->lun_size_bytes;
2394
2395         oldsize = be_lun->size_bytes;
2396         if (be_lun->vn == NULL)
2397                 error = ctl_be_block_open(softc, be_lun, req);
2398         else if (be_lun->vn->v_type == VREG)
2399                 error = ctl_be_block_modify_file(be_lun, req);
2400         else
2401                 error = ctl_be_block_modify_dev(be_lun, req);
2402
2403         if (error == 0 && be_lun->size_bytes != oldsize) {
2404                 be_lun->size_blocks = be_lun->size_bytes >>
2405                     be_lun->blocksize_shift;
2406
2407                 /*
2408                  * The maximum LBA is the size - 1.
2409                  *
2410                  * XXX: Note that this field is being updated without locking,
2411                  *      which might cause problems on 32-bit architectures.
2412                  */
2413                 be_lun->ctl_be_lun.maxlba = (be_lun->size_blocks == 0) ?
2414                     0 : (be_lun->size_blocks - 1);
2415                 be_lun->ctl_be_lun.blocksize = be_lun->blocksize;
2416                 be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp;
2417                 be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff;
2418                 if (be_lun->dispatch == ctl_be_block_dispatch_zvol &&
2419                     be_lun->blocksize != 0)
2420                         be_lun->ctl_be_lun.atomicblock = CTLBLK_MAX_IO_SIZE /
2421                             be_lun->blocksize;
2422                 ctl_lun_capacity_changed(&be_lun->ctl_be_lun);
2423                 if (oldsize == 0 && be_lun->size_blocks != 0)
2424                         ctl_lun_online(&be_lun->ctl_be_lun);
2425         }
2426
2427         /* Tell the user the exact size we ended up using */
2428         params->lun_size_bytes = be_lun->size_bytes;
2429
2430         req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK;
2431
2432         return (0);
2433
2434 bailout_error:
2435         req->status = CTL_LUN_ERROR;
2436
2437         return (0);
2438 }
2439
2440 static void
2441 ctl_be_block_lun_shutdown(void *be_lun)
2442 {
2443         struct ctl_be_block_lun *lun;
2444         struct ctl_be_block_softc *softc;
2445
2446         lun = (struct ctl_be_block_lun *)be_lun;
2447
2448         softc = lun->softc;
2449
2450         mtx_lock(&softc->lock);
2451         lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED;
2452         if (lun->flags & CTL_BE_BLOCK_LUN_WAITING)
2453                 wakeup(lun);
2454         mtx_unlock(&softc->lock);
2455
2456 }
2457
2458 static void
2459 ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status)
2460 {
2461         struct ctl_be_block_lun *lun;
2462         struct ctl_be_block_softc *softc;
2463
2464         lun = (struct ctl_be_block_lun *)be_lun;
2465         softc = lun->softc;
2466
2467         if (status == CTL_LUN_CONFIG_OK) {
2468                 mtx_lock(&softc->lock);
2469                 lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED;
2470                 if (lun->flags & CTL_BE_BLOCK_LUN_WAITING)
2471                         wakeup(lun);
2472                 mtx_unlock(&softc->lock);
2473
2474                 /*
2475                  * We successfully added the LUN, attempt to enable it.
2476                  */
2477                 if (ctl_enable_lun(&lun->ctl_be_lun) != 0) {
2478                         printf("%s: ctl_enable_lun() failed!\n", __func__);
2479                         if (ctl_invalidate_lun(&lun->ctl_be_lun) != 0) {
2480                                 printf("%s: ctl_invalidate_lun() failed!\n",
2481                                        __func__);
2482                         }
2483                 }
2484
2485                 return;
2486         }
2487
2488
2489         mtx_lock(&softc->lock);
2490         lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED;
2491         lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR;
2492         wakeup(lun);
2493         mtx_unlock(&softc->lock);
2494 }
2495
2496
2497 static int
2498 ctl_be_block_config_write(union ctl_io *io)
2499 {
2500         struct ctl_be_block_lun *be_lun;
2501         struct ctl_be_lun *ctl_be_lun;
2502         int retval;
2503
2504         retval = 0;
2505
2506         DPRINTF("entered\n");
2507
2508         ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[
2509                 CTL_PRIV_BACKEND_LUN].ptr;
2510         be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun;
2511
2512         switch (io->scsiio.cdb[0]) {
2513         case SYNCHRONIZE_CACHE:
2514         case SYNCHRONIZE_CACHE_16:
2515         case WRITE_SAME_10:
2516         case WRITE_SAME_16:
2517         case UNMAP:
2518                 /*
2519                  * The upper level CTL code will filter out any CDBs with
2520                  * the immediate bit set and return the proper error.
2521                  *
2522                  * We don't really need to worry about what LBA range the
2523                  * user asked to be synced out.  When they issue a sync
2524                  * cache command, we'll sync out the whole thing.
2525                  */
2526                 mtx_lock(&be_lun->queue_lock);
2527                 STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr,
2528                                    links);
2529                 mtx_unlock(&be_lun->queue_lock);
2530                 taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task);
2531                 break;
2532         case START_STOP_UNIT: {
2533                 struct scsi_start_stop_unit *cdb;
2534
2535                 cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb;
2536
2537                 if (cdb->how & SSS_START)
2538                         retval = ctl_start_lun(ctl_be_lun);
2539                 else {
2540                         retval = ctl_stop_lun(ctl_be_lun);
2541                         /*
2542                          * XXX KDM Copan-specific offline behavior.
2543                          * Figure out a reasonable way to port this?
2544                          */
2545 #ifdef NEEDTOPORT
2546                         if ((retval == 0)
2547                          && (cdb->byte2 & SSS_ONOFFLINE))
2548                                 retval = ctl_lun_offline(ctl_be_lun);
2549 #endif
2550                 }
2551
2552                 /*
2553                  * In general, the above routines should not fail.  They
2554                  * just set state for the LUN.  So we've got something
2555                  * pretty wrong here if we can't start or stop the LUN.
2556                  */
2557                 if (retval != 0) {
2558                         ctl_set_internal_failure(&io->scsiio,
2559                                                  /*sks_valid*/ 1,
2560                                                  /*retry_count*/ 0xf051);
2561                         retval = CTL_RETVAL_COMPLETE;
2562                 } else {
2563                         ctl_set_success(&io->scsiio);
2564                 }
2565                 ctl_config_write_done(io);
2566                 break;
2567         }
2568         default:
2569                 ctl_set_invalid_opcode(&io->scsiio);
2570                 ctl_config_write_done(io);
2571                 retval = CTL_RETVAL_COMPLETE;
2572                 break;
2573         }
2574
2575         return (retval);
2576
2577 }
2578
2579 static int
2580 ctl_be_block_config_read(union ctl_io *io)
2581 {
2582         return (0);
2583 }
2584
2585 static int
2586 ctl_be_block_lun_info(void *be_lun, struct sbuf *sb)
2587 {
2588         struct ctl_be_block_lun *lun;
2589         int retval;
2590
2591         lun = (struct ctl_be_block_lun *)be_lun;
2592         retval = 0;
2593
2594         retval = sbuf_printf(sb, "\t<num_threads>");
2595
2596         if (retval != 0)
2597                 goto bailout;
2598
2599         retval = sbuf_printf(sb, "%d", lun->num_threads);
2600
2601         if (retval != 0)
2602                 goto bailout;
2603
2604         retval = sbuf_printf(sb, "</num_threads>\n");
2605
2606 bailout:
2607
2608         return (retval);
2609 }
2610
2611 static uint64_t
2612 ctl_be_block_lun_attr(void *be_lun, const char *attrname)
2613 {
2614         struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)be_lun;
2615
2616         if (lun->getattr == NULL)
2617                 return (UINT64_MAX);
2618         return (lun->getattr(lun, attrname));
2619 }
2620
2621 int
2622 ctl_be_block_init(void)
2623 {
2624         struct ctl_be_block_softc *softc;
2625         int retval;
2626
2627         softc = &backend_block_softc;
2628         retval = 0;
2629
2630         mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF);
2631         beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io),
2632             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
2633         STAILQ_INIT(&softc->disk_list);
2634         STAILQ_INIT(&softc->lun_list);
2635
2636         return (retval);
2637 }