block/block-backend.c

   1 /*
   2  * QEMU Block backends
   3  *
   4  * Copyright (C) 2014 Red Hat, Inc.
   5  *
   6  * Authors:
   7  *  Markus Armbruster <armbru@redhat.com>,
   8  *
   9  * This work is licensed under the terms of the GNU LGPL, version 2.1
  10  * or later.  See the COPYING.LIB file in the top-level directory.
  11  */
  12
  13 #include "qemu/osdep.h"
  14 #include "sysemu/block-backend.h"
  15 #include "block/block_int.h"
  16 #include "block/blockjob.h"
  17 #include "block/throttle-groups.h"
  18 #include "sysemu/blockdev.h"
  19 #include "sysemu/sysemu.h"
  20 #include "qapi-event.h"
  21 #include "qemu/id.h"
  22
  23 /* Number of coroutines to reserve per attached device model */
  24 #define COROUTINE_POOL_RESERVATION 64
  25
  26 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
  27
  28 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb);
  29
  30 struct BlockBackend {
  31     char *name;
  32     int refcnt;
  33     BdrvChild *root;
  34     DriveInfo *legacy_dinfo;    /* null unless created by drive_new() */
  35     QTAILQ_ENTRY(BlockBackend) link;         /* for block_backends */
  36     QTAILQ_ENTRY(BlockBackend) monitor_link; /* for monitor_block_backends */
  37
  38     void *dev;                  /* attached device model, if any */
  39     /* TODO change to DeviceState when all users are qdevified */
  40     const BlockDevOps *dev_ops;
  41     void *dev_opaque;
  42
  43     /* the block size for which the guest device expects atomicity */
  44     int guest_block_size;
  45
  46     /* If the BDS tree is removed, some of its options are stored here (which
  47      * can be used to restore those options in the new BDS on insert) */
  48     BlockBackendRootState root_state;
  49
  50     bool enable_write_cache;
  51
  52     /* I/O stats (display with "info blockstats"). */
  53     BlockAcctStats stats;
  54
  55     BlockdevOnError on_read_error, on_write_error;
  56     bool iostatus_enabled;
  57     BlockDeviceIoStatus iostatus;
  58
  59     bool allow_write_beyond_eof;
  60
  61     NotifierList remove_bs_notifiers, insert_bs_notifiers;
  62 };
  63
  64 typedef struct BlockBackendAIOCB {
  65     BlockAIOCB common;
  66     QEMUBH *bh;
  67     BlockBackend *blk;
  68     int ret;
  69 } BlockBackendAIOCB;
  70
  71 static const AIOCBInfo block_backend_aiocb_info = {
  72     .get_aio_context = blk_aiocb_get_aio_context,
  73     .aiocb_size = sizeof(BlockBackendAIOCB),
  74 };
  75
  76 static void drive_info_del(DriveInfo *dinfo);
  77
  78 /* All BlockBackends */
  79 static QTAILQ_HEAD(, BlockBackend) block_backends =
  80     QTAILQ_HEAD_INITIALIZER(block_backends);
  81
  82 /* All BlockBackends referenced by the monitor and which are iterated through by
  83  * blk_next() */
  84 static QTAILQ_HEAD(, BlockBackend) monitor_block_backends =
  85     QTAILQ_HEAD_INITIALIZER(monitor_block_backends);
  86
  87 static void blk_root_inherit_options(int *child_flags, QDict *child_options,
  88                                      int parent_flags, QDict *parent_options)
  89 {
  90     /* We're not supposed to call this function for root nodes */
  91     abort();
  92 }
  93
  94 static const BdrvChildRole child_root = {
  95     .inherit_options = blk_root_inherit_options,
  96 };
  97
  98 /*
  99  * Create a new BlockBackend with a reference count of one.
 100  * Store an error through @errp on failure, unless it's null.
 101  * Return the new BlockBackend on success, null on failure.
 102  */
 103 BlockBackend *blk_new(Error **errp)
 104 {
 105     BlockBackend *blk;
 106
 107     blk = g_new0(BlockBackend, 1);
 108     blk->refcnt = 1;
 109     notifier_list_init(&blk->remove_bs_notifiers);
 110     notifier_list_init(&blk->insert_bs_notifiers);
 111     QTAILQ_INSERT_TAIL(&block_backends, blk, link);
 112     return blk;
 113 }
 114
 115 /*
 116  * Create a new BlockBackend with a new BlockDriverState attached.
 117  * Otherwise just like blk_new(), which see.
 118  */
 119 BlockBackend *blk_new_with_bs(Error **errp)
 120 {
 121     BlockBackend *blk;
 122     BlockDriverState *bs;
 123
 124     blk = blk_new(errp);
 125     if (!blk) {
 126         return NULL;
 127     }
 128
 129     bs = bdrv_new_root();
 130     blk->root = bdrv_root_attach_child(bs, "root", &child_root);
 131     bs->blk = blk;
 132     return blk;
 133 }
 134
 135 /*
 136  * Calls blk_new_with_bs() and then calls bdrv_open() on the BlockDriverState.
 137  *
 138  * Just as with bdrv_open(), after having called this function the reference to
 139  * @options belongs to the block layer (even on failure).
 140  *
 141  * TODO: Remove @filename and @flags; it should be possible to specify a whole
 142  * BDS tree just by specifying the @options QDict (or @reference,
 143  * alternatively). At the time of adding this function, this is not possible,
 144  * though, so callers of this function have to be able to specify @filename and
 145  * @flags.
 146  */
 147 BlockBackend *blk_new_open(const char *filename, const char *reference,
 148                            QDict *options, int flags, Error **errp)
 149 {
 150     BlockBackend *blk;
 151     int ret;
 152
 153     blk = blk_new_with_bs(errp);
 154     if (!blk) {
 155         QDECREF(options);
 156         return NULL;
 157     }
 158
 159     ret = bdrv_open(&blk->root->bs, filename, reference, options, flags, errp);
 160     if (ret < 0) {
 161         blk_unref(blk);
 162         return NULL;
 163     }
 164
 165     blk_set_enable_write_cache(blk, true);
 166
 167     return blk;
 168 }
 169
 170 static void blk_delete(BlockBackend *blk)
 171 {
 172     assert(!blk->refcnt);
 173     assert(!blk->name);
 174     assert(!blk->dev);
 175     if (blk->root) {
 176         blk_remove_bs(blk);
 177     }
 178     assert(QLIST_EMPTY(&blk->remove_bs_notifiers.notifiers));
 179     assert(QLIST_EMPTY(&blk->insert_bs_notifiers.notifiers));
 180     if (blk->root_state.throttle_state) {
 181         g_free(blk->root_state.throttle_group);
 182         throttle_group_unref(blk->root_state.throttle_state);
 183     }
 184     QTAILQ_REMOVE(&block_backends, blk, link);
 185     drive_info_del(blk->legacy_dinfo);
 186     block_acct_cleanup(&blk->stats);
 187     g_free(blk);
 188 }
 189
 190 static void drive_info_del(DriveInfo *dinfo)
 191 {
 192     if (!dinfo) {
 193         return;
 194     }
 195     qemu_opts_del(dinfo->opts);
 196     g_free(dinfo->serial);
 197     g_free(dinfo);
 198 }
 199
 200 int blk_get_refcnt(BlockBackend *blk)
 201 {
 202     return blk ? blk->refcnt : 0;
 203 }
 204
 205 /*
 206  * Increment @blk's reference count.
 207  * @blk must not be null.
 208  */
 209 void blk_ref(BlockBackend *blk)
 210 {
 211     blk->refcnt++;
 212 }
 213
 214 /*
 215  * Decrement @blk's reference count.
 216  * If this drops it to zero, destroy @blk.
 217  * For convenience, do nothing if @blk is null.
 218  */
 219 void blk_unref(BlockBackend *blk)
 220 {
 221     if (blk) {
 222         assert(blk->refcnt > 0);
 223         if (!--blk->refcnt) {
 224             blk_delete(blk);
 225         }
 226     }
 227 }
 228
 229 /*
 230  * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
 231  * ones which are hidden (i.e. are not referenced by the monitor).
 232  */
 233 static BlockBackend *blk_all_next(BlockBackend *blk)
 234 {
 235     return blk ? QTAILQ_NEXT(blk, link)
 236                : QTAILQ_FIRST(&block_backends);
 237 }
 238
 239 void blk_remove_all_bs(void)
 240 {
 241     BlockBackend *blk = NULL;
 242
 243     while ((blk = blk_all_next(blk)) != NULL) {
 244         AioContext *ctx = blk_get_aio_context(blk);
 245
 246         aio_context_acquire(ctx);
 247         if (blk->root) {
 248             blk_remove_bs(blk);
 249         }
 250         aio_context_release(ctx);
 251     }
 252 }
 253
 254 /*
 255  * Return the monitor-owned BlockBackend after @blk.
 256  * If @blk is null, return the first one.
 257  * Else, return @blk's next sibling, which may be null.
 258  *
 259  * To iterate over all BlockBackends, do
 260  * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
 261  *     ...
 262  * }
 263  */
 264 BlockBackend *blk_next(BlockBackend *blk)
 265 {
 266     return blk ? QTAILQ_NEXT(blk, monitor_link)
 267                : QTAILQ_FIRST(&monitor_block_backends);
 268 }
 269
 270 /*
 271  * Iterates over all BlockDriverStates which are attached to a BlockBackend.
 272  * This function is for use by bdrv_next().
 273  *
 274  * @bs must be NULL or a BDS that is attached to a BB.
 275  */
 276 BlockDriverState *blk_next_root_bs(BlockDriverState *bs)
 277 {
 278     BlockBackend *blk;
 279
 280     if (bs) {
 281         assert(bs->blk);
 282         blk = bs->blk;
 283     } else {
 284         blk = NULL;
 285     }
 286
 287     do {
 288         blk = blk_all_next(blk);
 289     } while (blk && !blk->root);
 290
 291     return blk ? blk->root->bs : NULL;
 292 }
 293
 294 /*
 295  * Add a BlockBackend into the list of backends referenced by the monitor, with
 296  * the given @name acting as the handle for the monitor.
 297  * Strictly for use by blockdev.c.
 298  *
 299  * @name must not be null or empty.
 300  *
 301  * Returns true on success and false on failure. In the latter case, an Error
 302  * object is returned through @errp.
 303  */
 304 bool monitor_add_blk(BlockBackend *blk, const char *name, Error **errp)
 305 {
 306     assert(!blk->name);
 307     assert(name && name[0]);
 308
 309     if (!id_wellformed(name)) {
 310         error_setg(errp, "Invalid device name");
 311         return false;
 312     }
 313     if (blk_by_name(name)) {
 314         error_setg(errp, "Device with id '%s' already exists", name);
 315         return false;
 316     }
 317     if (bdrv_find_node(name)) {
 318         error_setg(errp,
 319                    "Device name '%s' conflicts with an existing node name",
 320                    name);
 321         return false;
 322     }
 323
 324     blk->name = g_strdup(name);
 325     QTAILQ_INSERT_TAIL(&monitor_block_backends, blk, monitor_link);
 326     return true;
 327 }
 328
 329 /*
 330  * Remove a BlockBackend from the list of backends referenced by the monitor.
 331  * Strictly for use by blockdev.c.
 332  */
 333 void monitor_remove_blk(BlockBackend *blk)
 334 {
 335     if (!blk->name) {
 336         return;
 337     }
 338
 339     QTAILQ_REMOVE(&monitor_block_backends, blk, monitor_link);
 340     g_free(blk->name);
 341     blk->name = NULL;
 342 }
 343
 344 /*
 345  * Return @blk's name, a non-null string.
 346  * Returns an empty string iff @blk is not referenced by the monitor.
 347  */
 348 const char *blk_name(BlockBackend *blk)
 349 {
 350     return blk->name ?: "";
 351 }
 352
 353 /*
 354  * Return the BlockBackend with name @name if it exists, else null.
 355  * @name must not be null.
 356  */
 357 BlockBackend *blk_by_name(const char *name)
 358 {
 359     BlockBackend *blk = NULL;
 360
 361     assert(name);
 362     while ((blk = blk_next(blk)) != NULL) {
 363         if (!strcmp(name, blk->name)) {
 364             return blk;
 365         }
 366     }
 367     return NULL;
 368 }
 369
 370 /*
 371  * Return the BlockDriverState attached to @blk if any, else null.
 372  */
 373 BlockDriverState *blk_bs(BlockBackend *blk)
 374 {
 375     return blk->root ? blk->root->bs : NULL;
 376 }
 377
 378 /*
 379  * Return @blk's DriveInfo if any, else null.
 380  */
 381 DriveInfo *blk_legacy_dinfo(BlockBackend *blk)
 382 {
 383     return blk->legacy_dinfo;
 384 }
 385
 386 /*
 387  * Set @blk's DriveInfo to @dinfo, and return it.
 388  * @blk must not have a DriveInfo set already.
 389  * No other BlockBackend may have the same DriveInfo set.
 390  */
 391 DriveInfo *blk_set_legacy_dinfo(BlockBackend *blk, DriveInfo *dinfo)
 392 {
 393     assert(!blk->legacy_dinfo);
 394     return blk->legacy_dinfo = dinfo;
 395 }
 396
 397 /*
 398  * Return the BlockBackend with DriveInfo @dinfo.
 399  * It must exist.
 400  */
 401 BlockBackend *blk_by_legacy_dinfo(DriveInfo *dinfo)
 402 {
 403     BlockBackend *blk = NULL;
 404
 405     while ((blk = blk_next(blk)) != NULL) {
 406         if (blk->legacy_dinfo == dinfo) {
 407             return blk;
 408         }
 409     }
 410     abort();
 411 }
 412
 413 /*
 414  * Disassociates the currently associated BlockDriverState from @blk.
 415  */
 416 void blk_remove_bs(BlockBackend *blk)
 417 {
 418     assert(blk->root->bs->blk == blk);
 419
 420     notifier_list_notify(&blk->remove_bs_notifiers, blk);
 421
 422     blk_update_root_state(blk);
 423
 424     blk->root->bs->blk = NULL;
 425     bdrv_root_unref_child(blk->root);
 426     blk->root = NULL;
 427 }
 428
 429 /*
 430  * Associates a new BlockDriverState with @blk.
 431  */
 432 void blk_insert_bs(BlockBackend *blk, BlockDriverState *bs)
 433 {
 434     assert(!blk->root && !bs->blk);
 435     bdrv_ref(bs);
 436     blk->root = bdrv_root_attach_child(bs, "root", &child_root);
 437     bs->blk = blk;
 438
 439     notifier_list_notify(&blk->insert_bs_notifiers, blk);
 440 }
 441
 442 /*
 443  * Attach device model @dev to @blk.
 444  * Return 0 on success, -EBUSY when a device model is attached already.
 445  */
 446 int blk_attach_dev(BlockBackend *blk, void *dev)
 447 /* TODO change to DeviceState *dev when all users are qdevified */
 448 {
 449     if (blk->dev) {
 450         return -EBUSY;
 451     }
 452     blk_ref(blk);
 453     blk->dev = dev;
 454     blk_iostatus_reset(blk);
 455     return 0;
 456 }
 457
 458 /*
 459  * Attach device model @dev to @blk.
 460  * @blk must not have a device model attached already.
 461  * TODO qdevified devices don't use this, remove when devices are qdevified
 462  */
 463 void blk_attach_dev_nofail(BlockBackend *blk, void *dev)
 464 {
 465     if (blk_attach_dev(blk, dev) < 0) {
 466         abort();
 467     }
 468 }
 469
 470 /*
 471  * Detach device model @dev from @blk.
 472  * @dev must be currently attached to @blk.
 473  */
 474 void blk_detach_dev(BlockBackend *blk, void *dev)
 475 /* TODO change to DeviceState *dev when all users are qdevified */
 476 {
 477     assert(blk->dev == dev);
 478     blk->dev = NULL;
 479     blk->dev_ops = NULL;
 480     blk->dev_opaque = NULL;
 481     blk->guest_block_size = 512;
 482     blk_unref(blk);
 483 }
 484
 485 /*
 486  * Return the device model attached to @blk if any, else null.
 487  */
 488 void *blk_get_attached_dev(BlockBackend *blk)
 489 /* TODO change to return DeviceState * when all users are qdevified */
 490 {
 491     return blk->dev;
 492 }
 493
 494 /*
 495  * Set @blk's device model callbacks to @ops.
 496  * @opaque is the opaque argument to pass to the callbacks.
 497  * This is for use by device models.
 498  */
 499 void blk_set_dev_ops(BlockBackend *blk, const BlockDevOps *ops,
 500                      void *opaque)
 501 {
 502     blk->dev_ops = ops;
 503     blk->dev_opaque = opaque;
 504 }
 505
 506 /*
 507  * Notify @blk's attached device model of media change.
 508  * If @load is true, notify of media load.
 509  * Else, notify of media eject.
 510  * Also send DEVICE_TRAY_MOVED events as appropriate.
 511  */
 512 void blk_dev_change_media_cb(BlockBackend *blk, bool load)
 513 {
 514     if (blk->dev_ops && blk->dev_ops->change_media_cb) {
 515         bool tray_was_open, tray_is_open;
 516
 517         tray_was_open = blk_dev_is_tray_open(blk);
 518         blk->dev_ops->change_media_cb(blk->dev_opaque, load);
 519         tray_is_open = blk_dev_is_tray_open(blk);
 520
 521         if (tray_was_open != tray_is_open) {
 522             qapi_event_send_device_tray_moved(blk_name(blk), tray_is_open,
 523                                               &error_abort);
 524         }
 525     }
 526 }
 527
 528 /*
 529  * Does @blk's attached device model have removable media?
 530  * %true if no device model is attached.
 531  */
 532 bool blk_dev_has_removable_media(BlockBackend *blk)
 533 {
 534     return !blk->dev || (blk->dev_ops && blk->dev_ops->change_media_cb);
 535 }
 536
 537 /*
 538  * Does @blk's attached device model have a tray?
 539  */
 540 bool blk_dev_has_tray(BlockBackend *blk)
 541 {
 542     return blk->dev_ops && blk->dev_ops->is_tray_open;
 543 }
 544
 545 /*
 546  * Notify @blk's attached device model of a media eject request.
 547  * If @force is true, the medium is about to be yanked out forcefully.
 548  */
 549 void blk_dev_eject_request(BlockBackend *blk, bool force)
 550 {
 551     if (blk->dev_ops && blk->dev_ops->eject_request_cb) {
 552         blk->dev_ops->eject_request_cb(blk->dev_opaque, force);
 553     }
 554 }
 555
 556 /*
 557  * Does @blk's attached device model have a tray, and is it open?
 558  */
 559 bool blk_dev_is_tray_open(BlockBackend *blk)
 560 {
 561     if (blk_dev_has_tray(blk)) {
 562         return blk->dev_ops->is_tray_open(blk->dev_opaque);
 563     }
 564     return false;
 565 }
 566
 567 /*
 568  * Does @blk's attached device model have the medium locked?
 569  * %false if the device model has no such lock.
 570  */
 571 bool blk_dev_is_medium_locked(BlockBackend *blk)
 572 {
 573     if (blk->dev_ops && blk->dev_ops->is_medium_locked) {
 574         return blk->dev_ops->is_medium_locked(blk->dev_opaque);
 575     }
 576     return false;
 577 }
 578
 579 /*
 580  * Notify @blk's attached device model of a backend size change.
 581  */
 582 void blk_dev_resize_cb(BlockBackend *blk)
 583 {
 584     if (blk->dev_ops && blk->dev_ops->resize_cb) {
 585         blk->dev_ops->resize_cb(blk->dev_opaque);
 586     }
 587 }
 588
 589 void blk_iostatus_enable(BlockBackend *blk)
 590 {
 591     blk->iostatus_enabled = true;
 592     blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 593 }
 594
 595 /* The I/O status is only enabled if the drive explicitly
 596  * enables it _and_ the VM is configured to stop on errors */
 597 bool blk_iostatus_is_enabled(const BlockBackend *blk)
 598 {
 599     return (blk->iostatus_enabled &&
 600            (blk->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
 601             blk->on_write_error == BLOCKDEV_ON_ERROR_STOP   ||
 602             blk->on_read_error == BLOCKDEV_ON_ERROR_STOP));
 603 }
 604
 605 BlockDeviceIoStatus blk_iostatus(const BlockBackend *blk)
 606 {
 607     return blk->iostatus;
 608 }
 609
 610 void blk_iostatus_disable(BlockBackend *blk)
 611 {
 612     blk->iostatus_enabled = false;
 613 }
 614
 615 void blk_iostatus_reset(BlockBackend *blk)
 616 {
 617     if (blk_iostatus_is_enabled(blk)) {
 618         BlockDriverState *bs = blk_bs(blk);
 619         blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 620         if (bs && bs->job) {
 621             block_job_iostatus_reset(bs->job);
 622         }
 623     }
 624 }
 625
 626 void blk_iostatus_set_err(BlockBackend *blk, int error)
 627 {
 628     assert(blk_iostatus_is_enabled(blk));
 629     if (blk->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 630         blk->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 631                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 632     }
 633 }
 634
 635 void blk_set_allow_write_beyond_eof(BlockBackend *blk, bool allow)
 636 {
 637     blk->allow_write_beyond_eof = allow;
 638 }
 639
 640 static int blk_check_byte_request(BlockBackend *blk, int64_t offset,
 641                                   size_t size)
 642 {
 643     int64_t len;
 644
 645     if (size > INT_MAX) {
 646         return -EIO;
 647     }
 648
 649     if (!blk_is_available(blk)) {
 650         return -ENOMEDIUM;
 651     }
 652
 653     if (offset < 0) {
 654         return -EIO;
 655     }
 656
 657     if (!blk->allow_write_beyond_eof) {
 658         len = blk_getlength(blk);
 659         if (len < 0) {
 660             return len;
 661         }
 662
 663         if (offset > len || len - offset < size) {
 664             return -EIO;
 665         }
 666     }
 667
 668     return 0;
 669 }
 670
 671 static int blk_check_request(BlockBackend *blk, int64_t sector_num,
 672                              int nb_sectors)
 673 {
 674     if (sector_num < 0 || sector_num > INT64_MAX / BDRV_SECTOR_SIZE) {
 675         return -EIO;
 676     }
 677
 678     if (nb_sectors < 0 || nb_sectors > INT_MAX / BDRV_SECTOR_SIZE) {
 679         return -EIO;
 680     }
 681
 682     return blk_check_byte_request(blk, sector_num * BDRV_SECTOR_SIZE,
 683                                   nb_sectors * BDRV_SECTOR_SIZE);
 684 }
 685
 686 static int coroutine_fn blk_co_preadv(BlockBackend *blk, int64_t offset,
 687                                       unsigned int bytes, QEMUIOVector *qiov,
 688                                       BdrvRequestFlags flags)
 689 {
 690     int ret = blk_check_byte_request(blk, offset, bytes);
 691     if (ret < 0) {
 692         return ret;
 693     }
 694
 695     return bdrv_co_do_preadv(blk_bs(blk), offset, bytes, qiov, flags);
 696 }
 697
 698 static int coroutine_fn blk_co_pwritev(BlockBackend *blk, int64_t offset,
 699                                       unsigned int bytes, QEMUIOVector *qiov,
 700                                       BdrvRequestFlags flags)
 701 {
 702     int ret;
 703
 704     ret = blk_check_byte_request(blk, offset, bytes);
 705     if (ret < 0) {
 706         return ret;
 707     }
 708
 709     if (!blk->enable_write_cache) {
 710         flags |= BDRV_REQ_FUA;
 711     }
 712
 713     return bdrv_co_do_pwritev(blk_bs(blk), offset, bytes, qiov, flags);
 714 }
 715
 716 typedef struct BlkRwCo {
 717     BlockBackend *blk;
 718     int64_t offset;
 719     QEMUIOVector *qiov;
 720     int ret;
 721     BdrvRequestFlags flags;
 722 } BlkRwCo;
 723
 724 static void blk_read_entry(void *opaque)
 725 {
 726     BlkRwCo *rwco = opaque;
 727
 728     rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, rwco->qiov->size,
 729                               rwco->qiov, rwco->flags);
 730 }
 731
 732 static void blk_write_entry(void *opaque)
 733 {
 734     BlkRwCo *rwco = opaque;
 735
 736     rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, rwco->qiov->size,
 737                                rwco->qiov, rwco->flags);
 738 }
 739
 740 static int blk_prw(BlockBackend *blk, int64_t offset, uint8_t *buf,
 741                    int64_t bytes, CoroutineEntry co_entry,
 742                    BdrvRequestFlags flags)
 743 {
 744     AioContext *aio_context;
 745     QEMUIOVector qiov;
 746     struct iovec iov;
 747     Coroutine *co;
 748     BlkRwCo rwco;
 749
 750     iov = (struct iovec) {
 751         .iov_base = buf,
 752         .iov_len = bytes,
 753     };
 754     qemu_iovec_init_external(&qiov, &iov, 1);
 755
 756     rwco = (BlkRwCo) {
 757         .blk    = blk,
 758         .offset = offset,
 759         .qiov   = &qiov,
 760         .flags  = flags,
 761         .ret    = NOT_DONE,
 762     };
 763
 764     co = qemu_coroutine_create(co_entry);
 765     qemu_coroutine_enter(co, &rwco);
 766
 767     aio_context = blk_get_aio_context(blk);
 768     while (rwco.ret == NOT_DONE) {
 769         aio_poll(aio_context, true);
 770     }
 771
 772     return rwco.ret;
 773 }
 774
 775 static int blk_rw(BlockBackend *blk, int64_t sector_num, uint8_t *buf,
 776                   int nb_sectors, CoroutineEntry co_entry,
 777                   BdrvRequestFlags flags)
 778 {
 779     if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
 780         return -EINVAL;
 781     }
 782
 783     return blk_prw(blk, sector_num << BDRV_SECTOR_BITS, buf,
 784                    nb_sectors << BDRV_SECTOR_BITS, co_entry, flags);
 785 }
 786
 787 int blk_read(BlockBackend *blk, int64_t sector_num, uint8_t *buf,
 788              int nb_sectors)
 789 {
 790     return blk_rw(blk, sector_num, buf, nb_sectors, blk_read_entry, 0);
 791 }
 792
 793 int blk_read_unthrottled(BlockBackend *blk, int64_t sector_num, uint8_t *buf,
 794                          int nb_sectors)
 795 {
 796     BlockDriverState *bs = blk_bs(blk);
 797     bool enabled;
 798     int ret;
 799
 800     ret = blk_check_request(blk, sector_num, nb_sectors);
 801     if (ret < 0) {
 802         return ret;
 803     }
 804
 805     enabled = bs->io_limits_enabled;
 806     bs->io_limits_enabled = false;
 807     ret = blk_read(blk, sector_num, buf, nb_sectors);
 808     bs->io_limits_enabled = enabled;
 809     return ret;
 810 }
 811
 812 int blk_write(BlockBackend *blk, int64_t sector_num, const uint8_t *buf,
 813               int nb_sectors)
 814 {
 815     return blk_rw(blk, sector_num, (uint8_t*) buf, nb_sectors,
 816                   blk_write_entry, 0);
 817 }
 818
 819 int blk_write_zeroes(BlockBackend *blk, int64_t sector_num,
 820                      int nb_sectors, BdrvRequestFlags flags)
 821 {
 822     return blk_rw(blk, sector_num, NULL, nb_sectors, blk_write_entry,
 823                   flags | BDRV_REQ_ZERO_WRITE);
 824 }
 825
 826 static void error_callback_bh(void *opaque)
 827 {
 828     struct BlockBackendAIOCB *acb = opaque;
 829     qemu_bh_delete(acb->bh);
 830     acb->common.cb(acb->common.opaque, acb->ret);
 831     qemu_aio_unref(acb);
 832 }
 833
 834 BlockAIOCB *blk_abort_aio_request(BlockBackend *blk,
 835                                   BlockCompletionFunc *cb,
 836                                   void *opaque, int ret)
 837 {
 838     struct BlockBackendAIOCB *acb;
 839     QEMUBH *bh;
 840
 841     acb = blk_aio_get(&block_backend_aiocb_info, blk, cb, opaque);
 842     acb->blk = blk;
 843     acb->ret = ret;
 844
 845     bh = aio_bh_new(blk_get_aio_context(blk), error_callback_bh, acb);
 846     acb->bh = bh;
 847     qemu_bh_schedule(bh);
 848
 849     return &acb->common;
 850 }
 851
 852 typedef struct BlkAioEmAIOCB {
 853     BlockAIOCB common;
 854     BlkRwCo rwco;
 855     int bytes;
 856     bool has_returned;
 857     QEMUBH* bh;
 858 } BlkAioEmAIOCB;
 859
 860 static const AIOCBInfo blk_aio_em_aiocb_info = {
 861     .aiocb_size         = sizeof(BlkAioEmAIOCB),
 862 };
 863
 864 static void blk_aio_complete(BlkAioEmAIOCB *acb)
 865 {
 866     if (acb->bh) {
 867         assert(acb->has_returned);
 868         qemu_bh_delete(acb->bh);
 869     }
 870     if (acb->has_returned) {
 871         acb->common.cb(acb->common.opaque, acb->rwco.ret);
 872         qemu_aio_unref(acb);
 873     }
 874 }
 875
 876 static void blk_aio_complete_bh(void *opaque)
 877 {
 878     blk_aio_complete(opaque);
 879 }
 880
 881 static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset, int bytes,
 882                                 QEMUIOVector *qiov, CoroutineEntry co_entry,
 883                                 BdrvRequestFlags flags,
 884                                 BlockCompletionFunc *cb, void *opaque)
 885 {
 886     BlkAioEmAIOCB *acb;
 887     Coroutine *co;
 888
 889     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
 890     acb->rwco = (BlkRwCo) {
 891         .blk    = blk,
 892         .offset = offset,
 893         .qiov   = qiov,
 894         .flags  = flags,
 895         .ret    = NOT_DONE,
 896     };
 897     acb->bytes = bytes;
 898     acb->bh = NULL;
 899     acb->has_returned = false;
 900
 901     co = qemu_coroutine_create(co_entry);
 902     qemu_coroutine_enter(co, acb);
 903
 904     acb->has_returned = true;
 905     if (acb->rwco.ret != NOT_DONE) {
 906         acb->bh = aio_bh_new(blk_get_aio_context(blk), blk_aio_complete_bh, acb);
 907         qemu_bh_schedule(acb->bh);
 908     }
 909
 910     return &acb->common;
 911 }
 912
 913 static void blk_aio_read_entry(void *opaque)
 914 {
 915     BlkAioEmAIOCB *acb = opaque;
 916     BlkRwCo *rwco = &acb->rwco;
 917
 918     assert(rwco->qiov->size == acb->bytes);
 919     rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, acb->bytes,
 920                               rwco->qiov, rwco->flags);
 921     blk_aio_complete(acb);
 922 }
 923
 924 static void blk_aio_write_entry(void *opaque)
 925 {
 926     BlkAioEmAIOCB *acb = opaque;
 927     BlkRwCo *rwco = &acb->rwco;
 928
 929     assert(!rwco->qiov || rwco->qiov->size == acb->bytes);
 930     rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, acb->bytes,
 931                                rwco->qiov, rwco->flags);
 932     blk_aio_complete(acb);
 933 }
 934
 935 BlockAIOCB *blk_aio_write_zeroes(BlockBackend *blk, int64_t sector_num,
 936                                  int nb_sectors, BdrvRequestFlags flags,
 937                                  BlockCompletionFunc *cb, void *opaque)
 938 {
 939     if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
 940         return blk_abort_aio_request(blk, cb, opaque, -EINVAL);
 941     }
 942
 943     return blk_aio_prwv(blk, sector_num << BDRV_SECTOR_BITS,
 944                         nb_sectors << BDRV_SECTOR_BITS, NULL,
 945                         blk_aio_write_entry, flags | BDRV_REQ_ZERO_WRITE,
 946                         cb, opaque);
 947 }
 948
 949 int blk_pread(BlockBackend *blk, int64_t offset, void *buf, int count)
 950 {
 951     int ret = blk_prw(blk, offset, buf, count, blk_read_entry, 0);
 952     if (ret < 0) {
 953         return ret;
 954     }
 955     return count;
 956 }
 957
 958 int blk_pwrite(BlockBackend *blk, int64_t offset, const void *buf, int count)
 959 {
 960     int ret = blk_prw(blk, offset, (void*) buf, count, blk_write_entry, 0);
 961     if (ret < 0) {
 962         return ret;
 963     }
 964     return count;
 965 }
 966
 967 int64_t blk_getlength(BlockBackend *blk)
 968 {
 969     if (!blk_is_available(blk)) {
 970         return -ENOMEDIUM;
 971     }
 972
 973     return bdrv_getlength(blk_bs(blk));
 974 }
 975
 976 void blk_get_geometry(BlockBackend *blk, uint64_t *nb_sectors_ptr)
 977 {
 978     if (!blk_bs(blk)) {
 979         *nb_sectors_ptr = 0;
 980     } else {
 981         bdrv_get_geometry(blk_bs(blk), nb_sectors_ptr);
 982     }
 983 }
 984
 985 int64_t blk_nb_sectors(BlockBackend *blk)
 986 {
 987     if (!blk_is_available(blk)) {
 988         return -ENOMEDIUM;
 989     }
 990
 991     return bdrv_nb_sectors(blk_bs(blk));
 992 }
 993
 994 BlockAIOCB *blk_aio_readv(BlockBackend *blk, int64_t sector_num,
 995                           QEMUIOVector *iov, int nb_sectors,
 996                           BlockCompletionFunc *cb, void *opaque)
 997 {
 998     if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
 999         return blk_abort_aio_request(blk, cb, opaque, -EINVAL);
1000     }
1001
1002     assert(nb_sectors << BDRV_SECTOR_BITS == iov->size);
1003     return blk_aio_prwv(blk, sector_num << BDRV_SECTOR_BITS, iov->size, iov,
1004                         blk_aio_read_entry, 0, cb, opaque);
1005 }
1006
1007 BlockAIOCB *blk_aio_writev(BlockBackend *blk, int64_t sector_num,
1008                            QEMUIOVector *iov, int nb_sectors,
1009                            BlockCompletionFunc *cb, void *opaque)
1010 {
1011     if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
1012         return blk_abort_aio_request(blk, cb, opaque, -EINVAL);
1013     }
1014
1015     assert(nb_sectors << BDRV_SECTOR_BITS == iov->size);
1016     return blk_aio_prwv(blk, sector_num << BDRV_SECTOR_BITS, iov->size, iov,
1017                         blk_aio_write_entry, 0, cb, opaque);
1018 }
1019
1020 BlockAIOCB *blk_aio_flush(BlockBackend *blk,
1021                           BlockCompletionFunc *cb, void *opaque)
1022 {
1023     if (!blk_is_available(blk)) {
1024         return blk_abort_aio_request(blk, cb, opaque, -ENOMEDIUM);
1025     }
1026
1027     return bdrv_aio_flush(blk_bs(blk), cb, opaque);
1028 }
1029
1030 BlockAIOCB *blk_aio_discard(BlockBackend *blk,
1031                             int64_t sector_num, int nb_sectors,
1032                             BlockCompletionFunc *cb, void *opaque)
1033 {
1034     int ret = blk_check_request(blk, sector_num, nb_sectors);
1035     if (ret < 0) {
1036         return blk_abort_aio_request(blk, cb, opaque, ret);
1037     }
1038
1039     return bdrv_aio_discard(blk_bs(blk), sector_num, nb_sectors, cb, opaque);
1040 }
1041
1042 void blk_aio_cancel(BlockAIOCB *acb)
1043 {
1044     bdrv_aio_cancel(acb);
1045 }
1046
1047 void blk_aio_cancel_async(BlockAIOCB *acb)
1048 {
1049     bdrv_aio_cancel_async(acb);
1050 }
1051
1052 int blk_aio_multiwrite(BlockBackend *blk, BlockRequest *reqs, int num_reqs)
1053 {
1054     int i, ret;
1055
1056     for (i = 0; i < num_reqs; i++) {
1057         ret = blk_check_request(blk, reqs[i].sector, reqs[i].nb_sectors);
1058         if (ret < 0) {
1059             return ret;
1060         }
1061     }
1062
1063     return bdrv_aio_multiwrite(blk_bs(blk), reqs, num_reqs);
1064 }
1065
1066 int blk_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1067 {
1068     if (!blk_is_available(blk)) {
1069         return -ENOMEDIUM;
1070     }
1071
1072     return bdrv_ioctl(blk_bs(blk), req, buf);
1073 }
1074
1075 BlockAIOCB *blk_aio_ioctl(BlockBackend *blk, unsigned long int req, void *buf,
1076                           BlockCompletionFunc *cb, void *opaque)
1077 {
1078     if (!blk_is_available(blk)) {
1079         return blk_abort_aio_request(blk, cb, opaque, -ENOMEDIUM);
1080     }
1081
1082     return bdrv_aio_ioctl(blk_bs(blk), req, buf, cb, opaque);
1083 }
1084
1085 int blk_co_discard(BlockBackend *blk, int64_t sector_num, int nb_sectors)
1086 {
1087     int ret = blk_check_request(blk, sector_num, nb_sectors);
1088     if (ret < 0) {
1089         return ret;
1090     }
1091
1092     return bdrv_co_discard(blk_bs(blk), sector_num, nb_sectors);
1093 }
1094
1095 int blk_co_flush(BlockBackend *blk)
1096 {
1097     if (!blk_is_available(blk)) {
1098         return -ENOMEDIUM;
1099     }
1100
1101     return bdrv_co_flush(blk_bs(blk));
1102 }
1103
1104 int blk_flush(BlockBackend *blk)
1105 {
1106     if (!blk_is_available(blk)) {
1107         return -ENOMEDIUM;
1108     }
1109
1110     return bdrv_flush(blk_bs(blk));
1111 }
1112
1113 void blk_drain(BlockBackend *blk)
1114 {
1115     if (blk_bs(blk)) {
1116         bdrv_drain(blk_bs(blk));
1117     }
1118 }
1119
1120 void blk_drain_all(void)
1121 {
1122     bdrv_drain_all();
1123 }
1124
1125 void blk_set_on_error(BlockBackend *blk, BlockdevOnError on_read_error,
1126                       BlockdevOnError on_write_error)
1127 {
1128     blk->on_read_error = on_read_error;
1129     blk->on_write_error = on_write_error;
1130 }
1131
1132 BlockdevOnError blk_get_on_error(BlockBackend *blk, bool is_read)
1133 {
1134     return is_read ? blk->on_read_error : blk->on_write_error;
1135 }
1136
1137 BlockErrorAction blk_get_error_action(BlockBackend *blk, bool is_read,
1138                                       int error)
1139 {
1140     BlockdevOnError on_err = blk_get_on_error(blk, is_read);
1141
1142     switch (on_err) {
1143     case BLOCKDEV_ON_ERROR_ENOSPC:
1144         return (error == ENOSPC) ?
1145                BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
1146     case BLOCKDEV_ON_ERROR_STOP:
1147         return BLOCK_ERROR_ACTION_STOP;
1148     case BLOCKDEV_ON_ERROR_REPORT:
1149         return BLOCK_ERROR_ACTION_REPORT;
1150     case BLOCKDEV_ON_ERROR_IGNORE:
1151         return BLOCK_ERROR_ACTION_IGNORE;
1152     default:
1153         abort();
1154     }
1155 }
1156
1157 static void send_qmp_error_event(BlockBackend *blk,
1158                                  BlockErrorAction action,
1159                                  bool is_read, int error)
1160 {
1161     IoOperationType optype;
1162
1163     optype = is_read ? IO_OPERATION_TYPE_READ : IO_OPERATION_TYPE_WRITE;
1164     qapi_event_send_block_io_error(blk_name(blk), optype, action,
1165                                    blk_iostatus_is_enabled(blk),
1166                                    error == ENOSPC, strerror(error),
1167                                    &error_abort);
1168 }
1169
1170 /* This is done by device models because, while the block layer knows
1171  * about the error, it does not know whether an operation comes from
1172  * the device or the block layer (from a job, for example).
1173  */
1174 void blk_error_action(BlockBackend *blk, BlockErrorAction action,
1175                       bool is_read, int error)
1176 {
1177     assert(error >= 0);
1178
1179     if (action == BLOCK_ERROR_ACTION_STOP) {
1180         /* First set the iostatus, so that "info block" returns an iostatus
1181          * that matches the events raised so far (an additional error iostatus
1182          * is fine, but not a lost one).
1183          */
1184         blk_iostatus_set_err(blk, error);
1185
1186         /* Then raise the request to stop the VM and the event.
1187          * qemu_system_vmstop_request_prepare has two effects.  First,
1188          * it ensures that the STOP event always comes after the
1189          * BLOCK_IO_ERROR event.  Second, it ensures that even if management
1190          * can observe the STOP event and do a "cont" before the STOP
1191          * event is issued, the VM will not stop.  In this case, vm_start()
1192          * also ensures that the STOP/RESUME pair of events is emitted.
1193          */
1194         qemu_system_vmstop_request_prepare();
1195         send_qmp_error_event(blk, action, is_read, error);
1196         qemu_system_vmstop_request(RUN_STATE_IO_ERROR);
1197     } else {
1198         send_qmp_error_event(blk, action, is_read, error);
1199     }
1200 }
1201
1202 int blk_is_read_only(BlockBackend *blk)
1203 {
1204     BlockDriverState *bs = blk_bs(blk);
1205
1206     if (bs) {
1207         return bdrv_is_read_only(bs);
1208     } else {
1209         return blk->root_state.read_only;
1210     }
1211 }
1212
1213 int blk_is_sg(BlockBackend *blk)
1214 {
1215     BlockDriverState *bs = blk_bs(blk);
1216
1217     if (!bs) {
1218         return 0;
1219     }
1220
1221     return bdrv_is_sg(bs);
1222 }
1223
1224 int blk_enable_write_cache(BlockBackend *blk)
1225 {
1226     return blk->enable_write_cache;
1227 }
1228
1229 void blk_set_enable_write_cache(BlockBackend *blk, bool wce)
1230 {
1231     blk->enable_write_cache = wce;
1232 }
1233
1234 void blk_invalidate_cache(BlockBackend *blk, Error **errp)
1235 {
1236     BlockDriverState *bs = blk_bs(blk);
1237
1238     if (!bs) {
1239         error_setg(errp, "Device '%s' has no medium", blk->name);
1240         return;
1241     }
1242
1243     bdrv_invalidate_cache(bs, errp);
1244 }
1245
1246 bool blk_is_inserted(BlockBackend *blk)
1247 {
1248     BlockDriverState *bs = blk_bs(blk);
1249
1250     return bs && bdrv_is_inserted(bs);
1251 }
1252
1253 bool blk_is_available(BlockBackend *blk)
1254 {
1255     return blk_is_inserted(blk) && !blk_dev_is_tray_open(blk);
1256 }
1257
1258 void blk_lock_medium(BlockBackend *blk, bool locked)
1259 {
1260     BlockDriverState *bs = blk_bs(blk);
1261
1262     if (bs) {
1263         bdrv_lock_medium(bs, locked);
1264     }
1265 }
1266
1267 void blk_eject(BlockBackend *blk, bool eject_flag)
1268 {
1269     BlockDriverState *bs = blk_bs(blk);
1270
1271     if (bs) {
1272         bdrv_eject(bs, eject_flag);
1273     }
1274 }
1275
1276 int blk_get_flags(BlockBackend *blk)
1277 {
1278     BlockDriverState *bs = blk_bs(blk);
1279
1280     if (bs) {
1281         return bdrv_get_flags(bs);
1282     } else {
1283         return blk->root_state.open_flags;
1284     }
1285 }
1286
1287 int blk_get_max_transfer_length(BlockBackend *blk)
1288 {
1289     BlockDriverState *bs = blk_bs(blk);
1290
1291     if (bs) {
1292         return bs->bl.max_transfer_length;
1293     } else {
1294         return 0;
1295     }
1296 }
1297
1298 int blk_get_max_iov(BlockBackend *blk)
1299 {
1300     return blk->root->bs->bl.max_iov;
1301 }
1302
1303 void blk_set_guest_block_size(BlockBackend *blk, int align)
1304 {
1305     blk->guest_block_size = align;
1306 }
1307
1308 void *blk_try_blockalign(BlockBackend *blk, size_t size)
1309 {
1310     return qemu_try_blockalign(blk ? blk_bs(blk) : NULL, size);
1311 }
1312
1313 void *blk_blockalign(BlockBackend *blk, size_t size)
1314 {
1315     return qemu_blockalign(blk ? blk_bs(blk) : NULL, size);
1316 }
1317
1318 bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp)
1319 {
1320     BlockDriverState *bs = blk_bs(blk);
1321
1322     if (!bs) {
1323         return false;
1324     }
1325
1326     return bdrv_op_is_blocked(bs, op, errp);
1327 }
1328
1329 void blk_op_unblock(BlockBackend *blk, BlockOpType op, Error *reason)
1330 {
1331     BlockDriverState *bs = blk_bs(blk);
1332
1333     if (bs) {
1334         bdrv_op_unblock(bs, op, reason);
1335     }
1336 }
1337
1338 void blk_op_block_all(BlockBackend *blk, Error *reason)
1339 {
1340     BlockDriverState *bs = blk_bs(blk);
1341
1342     if (bs) {
1343         bdrv_op_block_all(bs, reason);
1344     }
1345 }
1346
1347 void blk_op_unblock_all(BlockBackend *blk, Error *reason)
1348 {
1349     BlockDriverState *bs = blk_bs(blk);
1350
1351     if (bs) {
1352         bdrv_op_unblock_all(bs, reason);
1353     }
1354 }
1355
1356 AioContext *blk_get_aio_context(BlockBackend *blk)
1357 {
1358     BlockDriverState *bs = blk_bs(blk);
1359
1360     if (bs) {
1361         return bdrv_get_aio_context(bs);
1362     } else {
1363         return qemu_get_aio_context();
1364     }
1365 }
1366
1367 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb)
1368 {
1369     BlockBackendAIOCB *blk_acb = DO_UPCAST(BlockBackendAIOCB, common, acb);
1370     return blk_get_aio_context(blk_acb->blk);
1371 }
1372
1373 void blk_set_aio_context(BlockBackend *blk, AioContext *new_context)
1374 {
1375     BlockDriverState *bs = blk_bs(blk);
1376
1377     if (bs) {
1378         bdrv_set_aio_context(bs, new_context);
1379     }
1380 }
1381
1382 void blk_add_aio_context_notifier(BlockBackend *blk,
1383         void (*attached_aio_context)(AioContext *new_context, void *opaque),
1384         void (*detach_aio_context)(void *opaque), void *opaque)
1385 {
1386     BlockDriverState *bs = blk_bs(blk);
1387
1388     if (bs) {
1389         bdrv_add_aio_context_notifier(bs, attached_aio_context,
1390                                       detach_aio_context, opaque);
1391     }
1392 }
1393
1394 void blk_remove_aio_context_notifier(BlockBackend *blk,
1395                                      void (*attached_aio_context)(AioContext *,
1396                                                                   void *),
1397                                      void (*detach_aio_context)(void *),
1398                                      void *opaque)
1399 {
1400     BlockDriverState *bs = blk_bs(blk);
1401
1402     if (bs) {
1403         bdrv_remove_aio_context_notifier(bs, attached_aio_context,
1404                                          detach_aio_context, opaque);
1405     }
1406 }
1407
1408 void blk_add_remove_bs_notifier(BlockBackend *blk, Notifier *notify)
1409 {
1410     notifier_list_add(&blk->remove_bs_notifiers, notify);
1411 }
1412
1413 void blk_add_insert_bs_notifier(BlockBackend *blk, Notifier *notify)
1414 {
1415     notifier_list_add(&blk->insert_bs_notifiers, notify);
1416 }
1417
1418 void blk_io_plug(BlockBackend *blk)
1419 {
1420     BlockDriverState *bs = blk_bs(blk);
1421
1422     if (bs) {
1423         bdrv_io_plug(bs);
1424     }
1425 }
1426
1427 void blk_io_unplug(BlockBackend *blk)
1428 {
1429     BlockDriverState *bs = blk_bs(blk);
1430
1431     if (bs) {
1432         bdrv_io_unplug(bs);
1433     }
1434 }
1435
1436 BlockAcctStats *blk_get_stats(BlockBackend *blk)
1437 {
1438     return &blk->stats;
1439 }
1440
1441 void *blk_aio_get(const AIOCBInfo *aiocb_info, BlockBackend *blk,
1442                   BlockCompletionFunc *cb, void *opaque)
1443 {
1444     return qemu_aio_get(aiocb_info, blk_bs(blk), cb, opaque);
1445 }
1446
1447 int coroutine_fn blk_co_write_zeroes(BlockBackend *blk, int64_t sector_num,
1448                                      int nb_sectors, BdrvRequestFlags flags)
1449 {
1450     if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
1451         return -EINVAL;
1452     }
1453
1454     return blk_co_pwritev(blk, sector_num << BDRV_SECTOR_BITS,
1455                           nb_sectors << BDRV_SECTOR_BITS, NULL,
1456                           flags | BDRV_REQ_ZERO_WRITE);
1457 }
1458
1459 int blk_write_compressed(BlockBackend *blk, int64_t sector_num,
1460                          const uint8_t *buf, int nb_sectors)
1461 {
1462     int ret = blk_check_request(blk, sector_num, nb_sectors);
1463     if (ret < 0) {
1464         return ret;
1465     }
1466
1467     return bdrv_write_compressed(blk_bs(blk), sector_num, buf, nb_sectors);
1468 }
1469
1470 int blk_truncate(BlockBackend *blk, int64_t offset)
1471 {
1472     if (!blk_is_available(blk)) {
1473         return -ENOMEDIUM;
1474     }
1475
1476     return bdrv_truncate(blk_bs(blk), offset);
1477 }
1478
1479 int blk_discard(BlockBackend *blk, int64_t sector_num, int nb_sectors)
1480 {
1481     int ret = blk_check_request(blk, sector_num, nb_sectors);
1482     if (ret < 0) {
1483         return ret;
1484     }
1485
1486     return bdrv_discard(blk_bs(blk), sector_num, nb_sectors);
1487 }
1488
1489 int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
1490                      int64_t pos, int size)
1491 {
1492     int ret;
1493
1494     if (!blk_is_available(blk)) {
1495         return -ENOMEDIUM;
1496     }
1497
1498     ret = bdrv_save_vmstate(blk_bs(blk), buf, pos, size);
1499     if (ret < 0) {
1500         return ret;
1501     }
1502
1503     if (ret == size && !blk->enable_write_cache) {
1504         ret = bdrv_flush(blk_bs(blk));
1505     }
1506
1507     return ret < 0 ? ret : size;
1508 }
1509
1510 int blk_load_vmstate(BlockBackend *blk, uint8_t *buf, int64_t pos, int size)
1511 {
1512     if (!blk_is_available(blk)) {
1513         return -ENOMEDIUM;
1514     }
1515
1516     return bdrv_load_vmstate(blk_bs(blk), buf, pos, size);
1517 }
1518
1519 int blk_probe_blocksizes(BlockBackend *blk, BlockSizes *bsz)
1520 {
1521     if (!blk_is_available(blk)) {
1522         return -ENOMEDIUM;
1523     }
1524
1525     return bdrv_probe_blocksizes(blk_bs(blk), bsz);
1526 }
1527
1528 int blk_probe_geometry(BlockBackend *blk, HDGeometry *geo)
1529 {
1530     if (!blk_is_available(blk)) {
1531         return -ENOMEDIUM;
1532     }
1533
1534     return bdrv_probe_geometry(blk_bs(blk), geo);
1535 }
1536
1537 /*
1538  * Updates the BlockBackendRootState object with data from the currently
1539  * attached BlockDriverState.
1540  */
1541 void blk_update_root_state(BlockBackend *blk)
1542 {
1543     assert(blk->root);
1544
1545     blk->root_state.open_flags    = blk->root->bs->open_flags;
1546     blk->root_state.read_only     = blk->root->bs->read_only;
1547     blk->root_state.detect_zeroes = blk->root->bs->detect_zeroes;
1548
1549     if (blk->root_state.throttle_group) {
1550         g_free(blk->root_state.throttle_group);
1551         throttle_group_unref(blk->root_state.throttle_state);
1552     }
1553     if (blk->root->bs->throttle_state) {
1554         const char *name = throttle_group_get_name(blk->root->bs);
1555         blk->root_state.throttle_group = g_strdup(name);
1556         blk->root_state.throttle_state = throttle_group_incref(name);
1557     } else {
1558         blk->root_state.throttle_group = NULL;
1559         blk->root_state.throttle_state = NULL;
1560     }
1561 }
1562
1563 /*
1564  * Applies the information in the root state to the given BlockDriverState. This
1565  * does not include the flags which have to be specified for bdrv_open(), use
1566  * blk_get_open_flags_from_root_state() to inquire them.
1567  */
1568 void blk_apply_root_state(BlockBackend *blk, BlockDriverState *bs)
1569 {
1570     bs->detect_zeroes = blk->root_state.detect_zeroes;
1571     if (blk->root_state.throttle_group) {
1572         bdrv_io_limits_enable(bs, blk->root_state.throttle_group);
1573     }
1574 }
1575
1576 /*
1577  * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
1578  * supposed to inherit the root state.
1579  */
1580 int blk_get_open_flags_from_root_state(BlockBackend *blk)
1581 {
1582     int bs_flags;
1583
1584     bs_flags = blk->root_state.read_only ? 0 : BDRV_O_RDWR;
1585     bs_flags |= blk->root_state.open_flags & ~BDRV_O_RDWR;
1586
1587     return bs_flags;
1588 }
1589
1590 BlockBackendRootState *blk_get_root_state(BlockBackend *blk)
1591 {
1592     return &blk->root_state;
1593 }
1594
1595 int blk_commit_all(void)
1596 {
1597     BlockBackend *blk = NULL;
1598
1599     while ((blk = blk_all_next(blk)) != NULL) {
1600         AioContext *aio_context = blk_get_aio_context(blk);
1601
1602         aio_context_acquire(aio_context);
1603         if (blk_is_inserted(blk) && blk->root->bs->backing) {
1604             int ret = bdrv_commit(blk->root->bs);
1605             if (ret < 0) {
1606                 aio_context_release(aio_context);
1607                 return ret;
1608             }
1609         }
1610         aio_context_release(aio_context);
1611     }
1612     return 0;
1613 }
1614
1615 int blk_flush_all(void)
1616 {
1617     BlockBackend *blk = NULL;
1618     int result = 0;
1619
1620     while ((blk = blk_all_next(blk)) != NULL) {
1621         AioContext *aio_context = blk_get_aio_context(blk);
1622         int ret;
1623
1624         aio_context_acquire(aio_context);
1625         if (blk_is_inserted(blk)) {
1626             ret = blk_flush(blk);
1627             if (ret < 0 && !result) {
1628                 result = ret;
1629             }
1630         }
1631         aio_context_release(aio_context);
1632     }
1633
1634     return result;
1635 }