block/block-backend.c

   1 /*
   2  * QEMU Block backends
   3  *
   4  * Copyright (C) 2014-2016 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 #include "trace.h"
  23
  24 /* Number of coroutines to reserve per attached device model */
  25 #define COROUTINE_POOL_RESERVATION 64
  26
  27 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
  28
  29 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb);
  30
  31 struct BlockBackend {
  32     char *name;
  33     int refcnt;
  34     BdrvChild *root;
  35     DriveInfo *legacy_dinfo;    /* null unless created by drive_new() */
  36     QTAILQ_ENTRY(BlockBackend) link;         /* for block_backends */
  37     QTAILQ_ENTRY(BlockBackend) monitor_link; /* for monitor_block_backends */
  38     BlockBackendPublic public;
  39
  40     void *dev;                  /* attached device model, if any */
  41     /* TODO change to DeviceState when all users are qdevified */
  42     const BlockDevOps *dev_ops;
  43     void *dev_opaque;
  44
  45     /* the block size for which the guest device expects atomicity */
  46     int guest_block_size;
  47
  48     /* If the BDS tree is removed, some of its options are stored here (which
  49      * can be used to restore those options in the new BDS on insert) */
  50     BlockBackendRootState root_state;
  51
  52     bool enable_write_cache;
  53
  54     /* I/O stats (display with "info blockstats"). */
  55     BlockAcctStats stats;
  56
  57     BlockdevOnError on_read_error, on_write_error;
  58     bool iostatus_enabled;
  59     BlockDeviceIoStatus iostatus;
  60
  61     bool allow_write_beyond_eof;
  62
  63     NotifierList remove_bs_notifiers, insert_bs_notifiers;
  64 };
  65
  66 typedef struct BlockBackendAIOCB {
  67     BlockAIOCB common;
  68     QEMUBH *bh;
  69     BlockBackend *blk;
  70     int ret;
  71 } BlockBackendAIOCB;
  72
  73 static const AIOCBInfo block_backend_aiocb_info = {
  74     .get_aio_context = blk_aiocb_get_aio_context,
  75     .aiocb_size = sizeof(BlockBackendAIOCB),
  76 };
  77
  78 static void drive_info_del(DriveInfo *dinfo);
  79 static BlockBackend *bdrv_first_blk(BlockDriverState *bs);
  80
  81 /* All BlockBackends */
  82 static QTAILQ_HEAD(, BlockBackend) block_backends =
  83     QTAILQ_HEAD_INITIALIZER(block_backends);
  84
  85 /* All BlockBackends referenced by the monitor and which are iterated through by
  86  * blk_next() */
  87 static QTAILQ_HEAD(, BlockBackend) monitor_block_backends =
  88     QTAILQ_HEAD_INITIALIZER(monitor_block_backends);
  89
  90 static void blk_root_inherit_options(int *child_flags, QDict *child_options,
  91                                      int parent_flags, QDict *parent_options)
  92 {
  93     /* We're not supposed to call this function for root nodes */
  94     abort();
  95 }
  96 static void blk_root_drained_begin(BdrvChild *child);
  97 static void blk_root_drained_end(BdrvChild *child);
  98
  99 static void blk_root_change_media(BdrvChild *child, bool load);
 100 static void blk_root_resize(BdrvChild *child);
 101
 102 static const char *blk_root_get_name(BdrvChild *child)
 103 {
 104     return blk_name(child->opaque);
 105 }
 106
 107 static const BdrvChildRole child_root = {
 108     .inherit_options    = blk_root_inherit_options,
 109
 110     .change_media       = blk_root_change_media,
 111     .resize             = blk_root_resize,
 112     .get_name           = blk_root_get_name,
 113
 114     .drained_begin      = blk_root_drained_begin,
 115     .drained_end        = blk_root_drained_end,
 116 };
 117
 118 /*
 119  * Create a new BlockBackend with a reference count of one.
 120  * Store an error through @errp on failure, unless it's null.
 121  * Return the new BlockBackend on success, null on failure.
 122  */
 123 BlockBackend *blk_new(void)
 124 {
 125     BlockBackend *blk;
 126
 127     blk = g_new0(BlockBackend, 1);
 128     blk->refcnt = 1;
 129     blk_set_enable_write_cache(blk, true);
 130
 131     qemu_co_queue_init(&blk->public.throttled_reqs[0]);
 132     qemu_co_queue_init(&blk->public.throttled_reqs[1]);
 133
 134     notifier_list_init(&blk->remove_bs_notifiers);
 135     notifier_list_init(&blk->insert_bs_notifiers);
 136
 137     QTAILQ_INSERT_TAIL(&block_backends, blk, link);
 138     return blk;
 139 }
 140
 141 /*
 142  * Creates a new BlockBackend, opens a new BlockDriverState, and connects both.
 143  *
 144  * Just as with bdrv_open(), after having called this function the reference to
 145  * @options belongs to the block layer (even on failure).
 146  *
 147  * TODO: Remove @filename and @flags; it should be possible to specify a whole
 148  * BDS tree just by specifying the @options QDict (or @reference,
 149  * alternatively). At the time of adding this function, this is not possible,
 150  * though, so callers of this function have to be able to specify @filename and
 151  * @flags.
 152  */
 153 BlockBackend *blk_new_open(const char *filename, const char *reference,
 154                            QDict *options, int flags, Error **errp)
 155 {
 156     BlockBackend *blk;
 157     BlockDriverState *bs;
 158
 159     blk = blk_new();
 160     bs = bdrv_open(filename, reference, options, flags, errp);
 161     if (!bs) {
 162         blk_unref(blk);
 163         return NULL;
 164     }
 165
 166     blk->root = bdrv_root_attach_child(bs, "root", &child_root, blk);
 167
 168     return blk;
 169 }
 170
 171 static void blk_delete(BlockBackend *blk)
 172 {
 173     assert(!blk->refcnt);
 174     assert(!blk->name);
 175     assert(!blk->dev);
 176     if (blk->root) {
 177         blk_remove_bs(blk);
 178     }
 179     assert(QLIST_EMPTY(&blk->remove_bs_notifiers.notifiers));
 180     assert(QLIST_EMPTY(&blk->insert_bs_notifiers.notifiers));
 181     QTAILQ_REMOVE(&block_backends, blk, link);
 182     drive_info_del(blk->legacy_dinfo);
 183     block_acct_cleanup(&blk->stats);
 184     g_free(blk);
 185 }
 186
 187 static void drive_info_del(DriveInfo *dinfo)
 188 {
 189     if (!dinfo) {
 190         return;
 191     }
 192     qemu_opts_del(dinfo->opts);
 193     g_free(dinfo->serial);
 194     g_free(dinfo);
 195 }
 196
 197 int blk_get_refcnt(BlockBackend *blk)
 198 {
 199     return blk ? blk->refcnt : 0;
 200 }
 201
 202 /*
 203  * Increment @blk's reference count.
 204  * @blk must not be null.
 205  */
 206 void blk_ref(BlockBackend *blk)
 207 {
 208     blk->refcnt++;
 209 }
 210
 211 /*
 212  * Decrement @blk's reference count.
 213  * If this drops it to zero, destroy @blk.
 214  * For convenience, do nothing if @blk is null.
 215  */
 216 void blk_unref(BlockBackend *blk)
 217 {
 218     if (blk) {
 219         assert(blk->refcnt > 0);
 220         if (!--blk->refcnt) {
 221             blk_delete(blk);
 222         }
 223     }
 224 }
 225
 226 /*
 227  * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
 228  * ones which are hidden (i.e. are not referenced by the monitor).
 229  */
 230 static BlockBackend *blk_all_next(BlockBackend *blk)
 231 {
 232     return blk ? QTAILQ_NEXT(blk, link)
 233                : QTAILQ_FIRST(&block_backends);
 234 }
 235
 236 void blk_remove_all_bs(void)
 237 {
 238     BlockBackend *blk = NULL;
 239
 240     while ((blk = blk_all_next(blk)) != NULL) {
 241         AioContext *ctx = blk_get_aio_context(blk);
 242
 243         aio_context_acquire(ctx);
 244         if (blk->root) {
 245             blk_remove_bs(blk);
 246         }
 247         aio_context_release(ctx);
 248     }
 249 }
 250
 251 /*
 252  * Return the monitor-owned BlockBackend after @blk.
 253  * If @blk is null, return the first one.
 254  * Else, return @blk's next sibling, which may be null.
 255  *
 256  * To iterate over all BlockBackends, do
 257  * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
 258  *     ...
 259  * }
 260  */
 261 BlockBackend *blk_next(BlockBackend *blk)
 262 {
 263     return blk ? QTAILQ_NEXT(blk, monitor_link)
 264                : QTAILQ_FIRST(&monitor_block_backends);
 265 }
 266
 267 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
 268  * the monitor or attached to a BlockBackend */
 269 BlockDriverState *bdrv_next(BdrvNextIterator *it)
 270 {
 271     BlockDriverState *bs;
 272
 273     /* First, return all root nodes of BlockBackends. In order to avoid
 274      * returning a BDS twice when multiple BBs refer to it, we only return it
 275      * if the BB is the first one in the parent list of the BDS. */
 276     if (it->phase == BDRV_NEXT_BACKEND_ROOTS) {
 277         do {
 278             it->blk = blk_all_next(it->blk);
 279             bs = it->blk ? blk_bs(it->blk) : NULL;
 280         } while (it->blk && (bs == NULL || bdrv_first_blk(bs) != it->blk));
 281
 282         if (bs) {
 283             return bs;
 284         }
 285         it->phase = BDRV_NEXT_MONITOR_OWNED;
 286     }
 287
 288     /* Then return the monitor-owned BDSes without a BB attached. Ignore all
 289      * BDSes that are attached to a BlockBackend here; they have been handled
 290      * by the above block already */
 291     do {
 292         it->bs = bdrv_next_monitor_owned(it->bs);
 293         bs = it->bs;
 294     } while (bs && bdrv_has_blk(bs));
 295
 296     return bs;
 297 }
 298
 299 BlockDriverState *bdrv_first(BdrvNextIterator *it)
 300 {
 301     *it = (BdrvNextIterator) {
 302         .phase = BDRV_NEXT_BACKEND_ROOTS,
 303     };
 304
 305     return bdrv_next(it);
 306 }
 307
 308 /*
 309  * Add a BlockBackend into the list of backends referenced by the monitor, with
 310  * the given @name acting as the handle for the monitor.
 311  * Strictly for use by blockdev.c.
 312  *
 313  * @name must not be null or empty.
 314  *
 315  * Returns true on success and false on failure. In the latter case, an Error
 316  * object is returned through @errp.
 317  */
 318 bool monitor_add_blk(BlockBackend *blk, const char *name, Error **errp)
 319 {
 320     assert(!blk->name);
 321     assert(name && name[0]);
 322
 323     if (!id_wellformed(name)) {
 324         error_setg(errp, "Invalid device name");
 325         return false;
 326     }
 327     if (blk_by_name(name)) {
 328         error_setg(errp, "Device with id '%s' already exists", name);
 329         return false;
 330     }
 331     if (bdrv_find_node(name)) {
 332         error_setg(errp,
 333                    "Device name '%s' conflicts with an existing node name",
 334                    name);
 335         return false;
 336     }
 337
 338     blk->name = g_strdup(name);
 339     QTAILQ_INSERT_TAIL(&monitor_block_backends, blk, monitor_link);
 340     return true;
 341 }
 342
 343 /*
 344  * Remove a BlockBackend from the list of backends referenced by the monitor.
 345  * Strictly for use by blockdev.c.
 346  */
 347 void monitor_remove_blk(BlockBackend *blk)
 348 {
 349     if (!blk->name) {
 350         return;
 351     }
 352
 353     QTAILQ_REMOVE(&monitor_block_backends, blk, monitor_link);
 354     g_free(blk->name);
 355     blk->name = NULL;
 356 }
 357
 358 /*
 359  * Return @blk's name, a non-null string.
 360  * Returns an empty string iff @blk is not referenced by the monitor.
 361  */
 362 const char *blk_name(BlockBackend *blk)
 363 {
 364     return blk->name ?: "";
 365 }
 366
 367 /*
 368  * Return the BlockBackend with name @name if it exists, else null.
 369  * @name must not be null.
 370  */
 371 BlockBackend *blk_by_name(const char *name)
 372 {
 373     BlockBackend *blk = NULL;
 374
 375     assert(name);
 376     while ((blk = blk_next(blk)) != NULL) {
 377         if (!strcmp(name, blk->name)) {
 378             return blk;
 379         }
 380     }
 381     return NULL;
 382 }
 383
 384 /*
 385  * Return the BlockDriverState attached to @blk if any, else null.
 386  */
 387 BlockDriverState *blk_bs(BlockBackend *blk)
 388 {
 389     return blk->root ? blk->root->bs : NULL;
 390 }
 391
 392 static BlockBackend *bdrv_first_blk(BlockDriverState *bs)
 393 {
 394     BdrvChild *child;
 395     QLIST_FOREACH(child, &bs->parents, next_parent) {
 396         if (child->role == &child_root) {
 397             return child->opaque;
 398         }
 399     }
 400
 401     return NULL;
 402 }
 403
 404 /*
 405  * Returns true if @bs has an associated BlockBackend.
 406  */
 407 bool bdrv_has_blk(BlockDriverState *bs)
 408 {
 409     return bdrv_first_blk(bs) != NULL;
 410 }
 411
 412 /*
 413  * Return @blk's DriveInfo if any, else null.
 414  */
 415 DriveInfo *blk_legacy_dinfo(BlockBackend *blk)
 416 {
 417     return blk->legacy_dinfo;
 418 }
 419
 420 /*
 421  * Set @blk's DriveInfo to @dinfo, and return it.
 422  * @blk must not have a DriveInfo set already.
 423  * No other BlockBackend may have the same DriveInfo set.
 424  */
 425 DriveInfo *blk_set_legacy_dinfo(BlockBackend *blk, DriveInfo *dinfo)
 426 {
 427     assert(!blk->legacy_dinfo);
 428     return blk->legacy_dinfo = dinfo;
 429 }
 430
 431 /*
 432  * Return the BlockBackend with DriveInfo @dinfo.
 433  * It must exist.
 434  */
 435 BlockBackend *blk_by_legacy_dinfo(DriveInfo *dinfo)
 436 {
 437     BlockBackend *blk = NULL;
 438
 439     while ((blk = blk_next(blk)) != NULL) {
 440         if (blk->legacy_dinfo == dinfo) {
 441             return blk;
 442         }
 443     }
 444     abort();
 445 }
 446
 447 /*
 448  * Returns a pointer to the publicly accessible fields of @blk.
 449  */
 450 BlockBackendPublic *blk_get_public(BlockBackend *blk)
 451 {
 452     return &blk->public;
 453 }
 454
 455 /*
 456  * Returns a BlockBackend given the associated @public fields.
 457  */
 458 BlockBackend *blk_by_public(BlockBackendPublic *public)
 459 {
 460     return container_of(public, BlockBackend, public);
 461 }
 462
 463 /*
 464  * Disassociates the currently associated BlockDriverState from @blk.
 465  */
 466 void blk_remove_bs(BlockBackend *blk)
 467 {
 468     notifier_list_notify(&blk->remove_bs_notifiers, blk);
 469     if (blk->public.throttle_state) {
 470         throttle_timers_detach_aio_context(&blk->public.throttle_timers);
 471     }
 472
 473     blk_update_root_state(blk);
 474
 475     bdrv_root_unref_child(blk->root);
 476     blk->root = NULL;
 477 }
 478
 479 /*
 480  * Associates a new BlockDriverState with @blk.
 481  */
 482 void blk_insert_bs(BlockBackend *blk, BlockDriverState *bs)
 483 {
 484     bdrv_ref(bs);
 485     blk->root = bdrv_root_attach_child(bs, "root", &child_root, blk);
 486
 487     notifier_list_notify(&blk->insert_bs_notifiers, blk);
 488     if (blk->public.throttle_state) {
 489         throttle_timers_attach_aio_context(
 490             &blk->public.throttle_timers, bdrv_get_aio_context(bs));
 491     }
 492 }
 493
 494 /*
 495  * Attach device model @dev to @blk.
 496  * Return 0 on success, -EBUSY when a device model is attached already.
 497  */
 498 int blk_attach_dev(BlockBackend *blk, void *dev)
 499 /* TODO change to DeviceState *dev when all users are qdevified */
 500 {
 501     if (blk->dev) {
 502         return -EBUSY;
 503     }
 504     blk_ref(blk);
 505     blk->dev = dev;
 506     blk_iostatus_reset(blk);
 507     return 0;
 508 }
 509
 510 /*
 511  * Attach device model @dev to @blk.
 512  * @blk must not have a device model attached already.
 513  * TODO qdevified devices don't use this, remove when devices are qdevified
 514  */
 515 void blk_attach_dev_nofail(BlockBackend *blk, void *dev)
 516 {
 517     if (blk_attach_dev(blk, dev) < 0) {
 518         abort();
 519     }
 520 }
 521
 522 /*
 523  * Detach device model @dev from @blk.
 524  * @dev must be currently attached to @blk.
 525  */
 526 void blk_detach_dev(BlockBackend *blk, void *dev)
 527 /* TODO change to DeviceState *dev when all users are qdevified */
 528 {
 529     assert(blk->dev == dev);
 530     blk->dev = NULL;
 531     blk->dev_ops = NULL;
 532     blk->dev_opaque = NULL;
 533     blk->guest_block_size = 512;
 534     blk_unref(blk);
 535 }
 536
 537 /*
 538  * Return the device model attached to @blk if any, else null.
 539  */
 540 void *blk_get_attached_dev(BlockBackend *blk)
 541 /* TODO change to return DeviceState * when all users are qdevified */
 542 {
 543     return blk->dev;
 544 }
 545
 546 /*
 547  * Set @blk's device model callbacks to @ops.
 548  * @opaque is the opaque argument to pass to the callbacks.
 549  * This is for use by device models.
 550  */
 551 void blk_set_dev_ops(BlockBackend *blk, const BlockDevOps *ops,
 552                      void *opaque)
 553 {
 554     blk->dev_ops = ops;
 555     blk->dev_opaque = opaque;
 556 }
 557
 558 /*
 559  * Notify @blk's attached device model of media change.
 560  * If @load is true, notify of media load.
 561  * Else, notify of media eject.
 562  * Also send DEVICE_TRAY_MOVED events as appropriate.
 563  */
 564 void blk_dev_change_media_cb(BlockBackend *blk, bool load)
 565 {
 566     if (blk->dev_ops && blk->dev_ops->change_media_cb) {
 567         bool tray_was_open, tray_is_open;
 568
 569         tray_was_open = blk_dev_is_tray_open(blk);
 570         blk->dev_ops->change_media_cb(blk->dev_opaque, load);
 571         tray_is_open = blk_dev_is_tray_open(blk);
 572
 573         if (tray_was_open != tray_is_open) {
 574             qapi_event_send_device_tray_moved(blk_name(blk), tray_is_open,
 575                                               &error_abort);
 576         }
 577     }
 578 }
 579
 580 static void blk_root_change_media(BdrvChild *child, bool load)
 581 {
 582     blk_dev_change_media_cb(child->opaque, load);
 583 }
 584
 585 /*
 586  * Does @blk's attached device model have removable media?
 587  * %true if no device model is attached.
 588  */
 589 bool blk_dev_has_removable_media(BlockBackend *blk)
 590 {
 591     return !blk->dev || (blk->dev_ops && blk->dev_ops->change_media_cb);
 592 }
 593
 594 /*
 595  * Does @blk's attached device model have a tray?
 596  */
 597 bool blk_dev_has_tray(BlockBackend *blk)
 598 {
 599     return blk->dev_ops && blk->dev_ops->is_tray_open;
 600 }
 601
 602 /*
 603  * Notify @blk's attached device model of a media eject request.
 604  * If @force is true, the medium is about to be yanked out forcefully.
 605  */
 606 void blk_dev_eject_request(BlockBackend *blk, bool force)
 607 {
 608     if (blk->dev_ops && blk->dev_ops->eject_request_cb) {
 609         blk->dev_ops->eject_request_cb(blk->dev_opaque, force);
 610     }
 611 }
 612
 613 /*
 614  * Does @blk's attached device model have a tray, and is it open?
 615  */
 616 bool blk_dev_is_tray_open(BlockBackend *blk)
 617 {
 618     if (blk_dev_has_tray(blk)) {
 619         return blk->dev_ops->is_tray_open(blk->dev_opaque);
 620     }
 621     return false;
 622 }
 623
 624 /*
 625  * Does @blk's attached device model have the medium locked?
 626  * %false if the device model has no such lock.
 627  */
 628 bool blk_dev_is_medium_locked(BlockBackend *blk)
 629 {
 630     if (blk->dev_ops && blk->dev_ops->is_medium_locked) {
 631         return blk->dev_ops->is_medium_locked(blk->dev_opaque);
 632     }
 633     return false;
 634 }
 635
 636 /*
 637  * Notify @blk's attached device model of a backend size change.
 638  */
 639 static void blk_root_resize(BdrvChild *child)
 640 {
 641     BlockBackend *blk = child->opaque;
 642
 643     if (blk->dev_ops && blk->dev_ops->resize_cb) {
 644         blk->dev_ops->resize_cb(blk->dev_opaque);
 645     }
 646 }
 647
 648 void blk_iostatus_enable(BlockBackend *blk)
 649 {
 650     blk->iostatus_enabled = true;
 651     blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 652 }
 653
 654 /* The I/O status is only enabled if the drive explicitly
 655  * enables it _and_ the VM is configured to stop on errors */
 656 bool blk_iostatus_is_enabled(const BlockBackend *blk)
 657 {
 658     return (blk->iostatus_enabled &&
 659            (blk->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
 660             blk->on_write_error == BLOCKDEV_ON_ERROR_STOP   ||
 661             blk->on_read_error == BLOCKDEV_ON_ERROR_STOP));
 662 }
 663
 664 BlockDeviceIoStatus blk_iostatus(const BlockBackend *blk)
 665 {
 666     return blk->iostatus;
 667 }
 668
 669 void blk_iostatus_disable(BlockBackend *blk)
 670 {
 671     blk->iostatus_enabled = false;
 672 }
 673
 674 void blk_iostatus_reset(BlockBackend *blk)
 675 {
 676     if (blk_iostatus_is_enabled(blk)) {
 677         BlockDriverState *bs = blk_bs(blk);
 678         blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 679         if (bs && bs->job) {
 680             block_job_iostatus_reset(bs->job);
 681         }
 682     }
 683 }
 684
 685 void blk_iostatus_set_err(BlockBackend *blk, int error)
 686 {
 687     assert(blk_iostatus_is_enabled(blk));
 688     if (blk->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 689         blk->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 690                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 691     }
 692 }
 693
 694 void blk_set_allow_write_beyond_eof(BlockBackend *blk, bool allow)
 695 {
 696     blk->allow_write_beyond_eof = allow;
 697 }
 698
 699 static int blk_check_byte_request(BlockBackend *blk, int64_t offset,
 700                                   size_t size)
 701 {
 702     int64_t len;
 703
 704     if (size > INT_MAX) {
 705         return -EIO;
 706     }
 707
 708     if (!blk_is_available(blk)) {
 709         return -ENOMEDIUM;
 710     }
 711
 712     if (offset < 0) {
 713         return -EIO;
 714     }
 715
 716     if (!blk->allow_write_beyond_eof) {
 717         len = blk_getlength(blk);
 718         if (len < 0) {
 719             return len;
 720         }
 721
 722         if (offset > len || len - offset < size) {
 723             return -EIO;
 724         }
 725     }
 726
 727     return 0;
 728 }
 729
 730 static int blk_check_request(BlockBackend *blk, int64_t sector_num,
 731                              int nb_sectors)
 732 {
 733     if (sector_num < 0 || sector_num > INT64_MAX / BDRV_SECTOR_SIZE) {
 734         return -EIO;
 735     }
 736
 737     if (nb_sectors < 0 || nb_sectors > INT_MAX / BDRV_SECTOR_SIZE) {
 738         return -EIO;
 739     }
 740
 741     return blk_check_byte_request(blk, sector_num * BDRV_SECTOR_SIZE,
 742                                   nb_sectors * BDRV_SECTOR_SIZE);
 743 }
 744
 745 int coroutine_fn blk_co_preadv(BlockBackend *blk, int64_t offset,
 746                                unsigned int bytes, QEMUIOVector *qiov,
 747                                BdrvRequestFlags flags)
 748 {
 749     int ret;
 750
 751     trace_blk_co_preadv(blk, blk_bs(blk), offset, bytes, flags);
 752
 753     ret = blk_check_byte_request(blk, offset, bytes);
 754     if (ret < 0) {
 755         return ret;
 756     }
 757
 758     /* throttling disk I/O */
 759     if (blk->public.throttle_state) {
 760         throttle_group_co_io_limits_intercept(blk, bytes, false);
 761     }
 762
 763     return bdrv_co_preadv(blk_bs(blk), offset, bytes, qiov, flags);
 764 }
 765
 766 int coroutine_fn blk_co_pwritev(BlockBackend *blk, int64_t offset,
 767                                 unsigned int bytes, QEMUIOVector *qiov,
 768                                 BdrvRequestFlags flags)
 769 {
 770     int ret;
 771
 772     trace_blk_co_pwritev(blk, blk_bs(blk), offset, bytes, flags);
 773
 774     ret = blk_check_byte_request(blk, offset, bytes);
 775     if (ret < 0) {
 776         return ret;
 777     }
 778
 779     /* throttling disk I/O */
 780     if (blk->public.throttle_state) {
 781         throttle_group_co_io_limits_intercept(blk, bytes, true);
 782     }
 783
 784     if (!blk->enable_write_cache) {
 785         flags |= BDRV_REQ_FUA;
 786     }
 787
 788     return bdrv_co_pwritev(blk_bs(blk), offset, bytes, qiov, flags);
 789 }
 790
 791 typedef struct BlkRwCo {
 792     BlockBackend *blk;
 793     int64_t offset;
 794     QEMUIOVector *qiov;
 795     int ret;
 796     BdrvRequestFlags flags;
 797 } BlkRwCo;
 798
 799 static void blk_read_entry(void *opaque)
 800 {
 801     BlkRwCo *rwco = opaque;
 802
 803     rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, rwco->qiov->size,
 804                               rwco->qiov, rwco->flags);
 805 }
 806
 807 static void blk_write_entry(void *opaque)
 808 {
 809     BlkRwCo *rwco = opaque;
 810
 811     rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, rwco->qiov->size,
 812                                rwco->qiov, rwco->flags);
 813 }
 814
 815 static int blk_prw(BlockBackend *blk, int64_t offset, uint8_t *buf,
 816                    int64_t bytes, CoroutineEntry co_entry,
 817                    BdrvRequestFlags flags)
 818 {
 819     AioContext *aio_context;
 820     QEMUIOVector qiov;
 821     struct iovec iov;
 822     Coroutine *co;
 823     BlkRwCo rwco;
 824
 825     iov = (struct iovec) {
 826         .iov_base = buf,
 827         .iov_len = bytes,
 828     };
 829     qemu_iovec_init_external(&qiov, &iov, 1);
 830
 831     rwco = (BlkRwCo) {
 832         .blk    = blk,
 833         .offset = offset,
 834         .qiov   = &qiov,
 835         .flags  = flags,
 836         .ret    = NOT_DONE,
 837     };
 838
 839     co = qemu_coroutine_create(co_entry);
 840     qemu_coroutine_enter(co, &rwco);
 841
 842     aio_context = blk_get_aio_context(blk);
 843     while (rwco.ret == NOT_DONE) {
 844         aio_poll(aio_context, true);
 845     }
 846
 847     return rwco.ret;
 848 }
 849
 850 int blk_pread_unthrottled(BlockBackend *blk, int64_t offset, uint8_t *buf,
 851                           int count)
 852 {
 853     int ret;
 854
 855     ret = blk_check_byte_request(blk, offset, count);
 856     if (ret < 0) {
 857         return ret;
 858     }
 859
 860     blk_root_drained_begin(blk->root);
 861     ret = blk_pread(blk, offset, buf, count);
 862     blk_root_drained_end(blk->root);
 863     return ret;
 864 }
 865
 866 int blk_pwrite_zeroes(BlockBackend *blk, int64_t offset,
 867                       int count, BdrvRequestFlags flags)
 868 {
 869     return blk_prw(blk, offset, NULL, count, blk_write_entry,
 870                    flags | BDRV_REQ_ZERO_WRITE);
 871 }
 872
 873 static void error_callback_bh(void *opaque)
 874 {
 875     struct BlockBackendAIOCB *acb = opaque;
 876     qemu_bh_delete(acb->bh);
 877     acb->common.cb(acb->common.opaque, acb->ret);
 878     qemu_aio_unref(acb);
 879 }
 880
 881 BlockAIOCB *blk_abort_aio_request(BlockBackend *blk,
 882                                   BlockCompletionFunc *cb,
 883                                   void *opaque, int ret)
 884 {
 885     struct BlockBackendAIOCB *acb;
 886     QEMUBH *bh;
 887
 888     acb = blk_aio_get(&block_backend_aiocb_info, blk, cb, opaque);
 889     acb->blk = blk;
 890     acb->ret = ret;
 891
 892     bh = aio_bh_new(blk_get_aio_context(blk), error_callback_bh, acb);
 893     acb->bh = bh;
 894     qemu_bh_schedule(bh);
 895
 896     return &acb->common;
 897 }
 898
 899 typedef struct BlkAioEmAIOCB {
 900     BlockAIOCB common;
 901     BlkRwCo rwco;
 902     int bytes;
 903     bool has_returned;
 904     QEMUBH* bh;
 905 } BlkAioEmAIOCB;
 906
 907 static const AIOCBInfo blk_aio_em_aiocb_info = {
 908     .aiocb_size         = sizeof(BlkAioEmAIOCB),
 909 };
 910
 911 static void blk_aio_complete(BlkAioEmAIOCB *acb)
 912 {
 913     if (acb->bh) {
 914         assert(acb->has_returned);
 915         qemu_bh_delete(acb->bh);
 916     }
 917     if (acb->has_returned) {
 918         acb->common.cb(acb->common.opaque, acb->rwco.ret);
 919         qemu_aio_unref(acb);
 920     }
 921 }
 922
 923 static void blk_aio_complete_bh(void *opaque)
 924 {
 925     blk_aio_complete(opaque);
 926 }
 927
 928 static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset, int bytes,
 929                                 QEMUIOVector *qiov, CoroutineEntry co_entry,
 930                                 BdrvRequestFlags flags,
 931                                 BlockCompletionFunc *cb, void *opaque)
 932 {
 933     BlkAioEmAIOCB *acb;
 934     Coroutine *co;
 935
 936     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
 937     acb->rwco = (BlkRwCo) {
 938         .blk    = blk,
 939         .offset = offset,
 940         .qiov   = qiov,
 941         .flags  = flags,
 942         .ret    = NOT_DONE,
 943     };
 944     acb->bytes = bytes;
 945     acb->bh = NULL;
 946     acb->has_returned = false;
 947
 948     co = qemu_coroutine_create(co_entry);
 949     qemu_coroutine_enter(co, acb);
 950
 951     acb->has_returned = true;
 952     if (acb->rwco.ret != NOT_DONE) {
 953         acb->bh = aio_bh_new(blk_get_aio_context(blk), blk_aio_complete_bh, acb);
 954         qemu_bh_schedule(acb->bh);
 955     }
 956
 957     return &acb->common;
 958 }
 959
 960 static void blk_aio_read_entry(void *opaque)
 961 {
 962     BlkAioEmAIOCB *acb = opaque;
 963     BlkRwCo *rwco = &acb->rwco;
 964
 965     assert(rwco->qiov->size == acb->bytes);
 966     rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, acb->bytes,
 967                               rwco->qiov, rwco->flags);
 968     blk_aio_complete(acb);
 969 }
 970
 971 static void blk_aio_write_entry(void *opaque)
 972 {
 973     BlkAioEmAIOCB *acb = opaque;
 974     BlkRwCo *rwco = &acb->rwco;
 975
 976     assert(!rwco->qiov || rwco->qiov->size == acb->bytes);
 977     rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, acb->bytes,
 978                                rwco->qiov, rwco->flags);
 979     blk_aio_complete(acb);
 980 }
 981
 982 BlockAIOCB *blk_aio_pwrite_zeroes(BlockBackend *blk, int64_t offset,
 983                                   int count, BdrvRequestFlags flags,
 984                                   BlockCompletionFunc *cb, void *opaque)
 985 {
 986     return blk_aio_prwv(blk, offset, count, NULL, blk_aio_write_entry,
 987                         flags | BDRV_REQ_ZERO_WRITE, cb, opaque);
 988 }
 989
 990 int blk_pread(BlockBackend *blk, int64_t offset, void *buf, int count)
 991 {
 992     int ret = blk_prw(blk, offset, buf, count, blk_read_entry, 0);
 993     if (ret < 0) {
 994         return ret;
 995     }
 996     return count;
 997 }
 998
 999 int blk_pwrite(BlockBackend *blk, int64_t offset, const void *buf, int count,
1000                BdrvRequestFlags flags)
1001 {
1002     int ret = blk_prw(blk, offset, (void *) buf, count, blk_write_entry,
1003                       flags);
1004     if (ret < 0) {
1005         return ret;
1006     }
1007     return count;
1008 }
1009
1010 int64_t blk_getlength(BlockBackend *blk)
1011 {
1012     if (!blk_is_available(blk)) {
1013         return -ENOMEDIUM;
1014     }
1015
1016     return bdrv_getlength(blk_bs(blk));
1017 }
1018
1019 void blk_get_geometry(BlockBackend *blk, uint64_t *nb_sectors_ptr)
1020 {
1021     if (!blk_bs(blk)) {
1022         *nb_sectors_ptr = 0;
1023     } else {
1024         bdrv_get_geometry(blk_bs(blk), nb_sectors_ptr);
1025     }
1026 }
1027
1028 int64_t blk_nb_sectors(BlockBackend *blk)
1029 {
1030     if (!blk_is_available(blk)) {
1031         return -ENOMEDIUM;
1032     }
1033
1034     return bdrv_nb_sectors(blk_bs(blk));
1035 }
1036
1037 BlockAIOCB *blk_aio_preadv(BlockBackend *blk, int64_t offset,
1038                            QEMUIOVector *qiov, BdrvRequestFlags flags,
1039                            BlockCompletionFunc *cb, void *opaque)
1040 {
1041     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1042                         blk_aio_read_entry, flags, cb, opaque);
1043 }
1044
1045 BlockAIOCB *blk_aio_pwritev(BlockBackend *blk, int64_t offset,
1046                             QEMUIOVector *qiov, BdrvRequestFlags flags,
1047                             BlockCompletionFunc *cb, void *opaque)
1048 {
1049     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1050                         blk_aio_write_entry, flags, cb, opaque);
1051 }
1052
1053 BlockAIOCB *blk_aio_flush(BlockBackend *blk,
1054                           BlockCompletionFunc *cb, void *opaque)
1055 {
1056     if (!blk_is_available(blk)) {
1057         return blk_abort_aio_request(blk, cb, opaque, -ENOMEDIUM);
1058     }
1059
1060     return bdrv_aio_flush(blk_bs(blk), cb, opaque);
1061 }
1062
1063 BlockAIOCB *blk_aio_discard(BlockBackend *blk,
1064                             int64_t sector_num, int nb_sectors,
1065                             BlockCompletionFunc *cb, void *opaque)
1066 {
1067     int ret = blk_check_request(blk, sector_num, nb_sectors);
1068     if (ret < 0) {
1069         return blk_abort_aio_request(blk, cb, opaque, ret);
1070     }
1071
1072     return bdrv_aio_discard(blk_bs(blk), sector_num, nb_sectors, cb, opaque);
1073 }
1074
1075 void blk_aio_cancel(BlockAIOCB *acb)
1076 {
1077     bdrv_aio_cancel(acb);
1078 }
1079
1080 void blk_aio_cancel_async(BlockAIOCB *acb)
1081 {
1082     bdrv_aio_cancel_async(acb);
1083 }
1084
1085 int blk_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1086 {
1087     if (!blk_is_available(blk)) {
1088         return -ENOMEDIUM;
1089     }
1090
1091     return bdrv_ioctl(blk_bs(blk), req, buf);
1092 }
1093
1094 BlockAIOCB *blk_aio_ioctl(BlockBackend *blk, unsigned long int req, void *buf,
1095                           BlockCompletionFunc *cb, void *opaque)
1096 {
1097     if (!blk_is_available(blk)) {
1098         return blk_abort_aio_request(blk, cb, opaque, -ENOMEDIUM);
1099     }
1100
1101     return bdrv_aio_ioctl(blk_bs(blk), req, buf, cb, opaque);
1102 }
1103
1104 int blk_co_discard(BlockBackend *blk, int64_t sector_num, int nb_sectors)
1105 {
1106     int ret = blk_check_request(blk, sector_num, nb_sectors);
1107     if (ret < 0) {
1108         return ret;
1109     }
1110
1111     return bdrv_co_discard(blk_bs(blk), sector_num, nb_sectors);
1112 }
1113
1114 int blk_co_flush(BlockBackend *blk)
1115 {
1116     if (!blk_is_available(blk)) {
1117         return -ENOMEDIUM;
1118     }
1119
1120     return bdrv_co_flush(blk_bs(blk));
1121 }
1122
1123 int blk_flush(BlockBackend *blk)
1124 {
1125     if (!blk_is_available(blk)) {
1126         return -ENOMEDIUM;
1127     }
1128
1129     return bdrv_flush(blk_bs(blk));
1130 }
1131
1132 void blk_drain(BlockBackend *blk)
1133 {
1134     if (blk_bs(blk)) {
1135         bdrv_drain(blk_bs(blk));
1136     }
1137 }
1138
1139 void blk_drain_all(void)
1140 {
1141     bdrv_drain_all();
1142 }
1143
1144 void blk_set_on_error(BlockBackend *blk, BlockdevOnError on_read_error,
1145                       BlockdevOnError on_write_error)
1146 {
1147     blk->on_read_error = on_read_error;
1148     blk->on_write_error = on_write_error;
1149 }
1150
1151 BlockdevOnError blk_get_on_error(BlockBackend *blk, bool is_read)
1152 {
1153     return is_read ? blk->on_read_error : blk->on_write_error;
1154 }
1155
1156 BlockErrorAction blk_get_error_action(BlockBackend *blk, bool is_read,
1157                                       int error)
1158 {
1159     BlockdevOnError on_err = blk_get_on_error(blk, is_read);
1160
1161     switch (on_err) {
1162     case BLOCKDEV_ON_ERROR_ENOSPC:
1163         return (error == ENOSPC) ?
1164                BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
1165     case BLOCKDEV_ON_ERROR_STOP:
1166         return BLOCK_ERROR_ACTION_STOP;
1167     case BLOCKDEV_ON_ERROR_REPORT:
1168         return BLOCK_ERROR_ACTION_REPORT;
1169     case BLOCKDEV_ON_ERROR_IGNORE:
1170         return BLOCK_ERROR_ACTION_IGNORE;
1171     default:
1172         abort();
1173     }
1174 }
1175
1176 static void send_qmp_error_event(BlockBackend *blk,
1177                                  BlockErrorAction action,
1178                                  bool is_read, int error)
1179 {
1180     IoOperationType optype;
1181
1182     optype = is_read ? IO_OPERATION_TYPE_READ : IO_OPERATION_TYPE_WRITE;
1183     qapi_event_send_block_io_error(blk_name(blk), optype, action,
1184                                    blk_iostatus_is_enabled(blk),
1185                                    error == ENOSPC, strerror(error),
1186                                    &error_abort);
1187 }
1188
1189 /* This is done by device models because, while the block layer knows
1190  * about the error, it does not know whether an operation comes from
1191  * the device or the block layer (from a job, for example).
1192  */
1193 void blk_error_action(BlockBackend *blk, BlockErrorAction action,
1194                       bool is_read, int error)
1195 {
1196     assert(error >= 0);
1197
1198     if (action == BLOCK_ERROR_ACTION_STOP) {
1199         /* First set the iostatus, so that "info block" returns an iostatus
1200          * that matches the events raised so far (an additional error iostatus
1201          * is fine, but not a lost one).
1202          */
1203         blk_iostatus_set_err(blk, error);
1204
1205         /* Then raise the request to stop the VM and the event.
1206          * qemu_system_vmstop_request_prepare has two effects.  First,
1207          * it ensures that the STOP event always comes after the
1208          * BLOCK_IO_ERROR event.  Second, it ensures that even if management
1209          * can observe the STOP event and do a "cont" before the STOP
1210          * event is issued, the VM will not stop.  In this case, vm_start()
1211          * also ensures that the STOP/RESUME pair of events is emitted.
1212          */
1213         qemu_system_vmstop_request_prepare();
1214         send_qmp_error_event(blk, action, is_read, error);
1215         qemu_system_vmstop_request(RUN_STATE_IO_ERROR);
1216     } else {
1217         send_qmp_error_event(blk, action, is_read, error);
1218     }
1219 }
1220
1221 int blk_is_read_only(BlockBackend *blk)
1222 {
1223     BlockDriverState *bs = blk_bs(blk);
1224
1225     if (bs) {
1226         return bdrv_is_read_only(bs);
1227     } else {
1228         return blk->root_state.read_only;
1229     }
1230 }
1231
1232 int blk_is_sg(BlockBackend *blk)
1233 {
1234     BlockDriverState *bs = blk_bs(blk);
1235
1236     if (!bs) {
1237         return 0;
1238     }
1239
1240     return bdrv_is_sg(bs);
1241 }
1242
1243 int blk_enable_write_cache(BlockBackend *blk)
1244 {
1245     return blk->enable_write_cache;
1246 }
1247
1248 void blk_set_enable_write_cache(BlockBackend *blk, bool wce)
1249 {
1250     blk->enable_write_cache = wce;
1251 }
1252
1253 void blk_invalidate_cache(BlockBackend *blk, Error **errp)
1254 {
1255     BlockDriverState *bs = blk_bs(blk);
1256
1257     if (!bs) {
1258         error_setg(errp, "Device '%s' has no medium", blk->name);
1259         return;
1260     }
1261
1262     bdrv_invalidate_cache(bs, errp);
1263 }
1264
1265 bool blk_is_inserted(BlockBackend *blk)
1266 {
1267     BlockDriverState *bs = blk_bs(blk);
1268
1269     return bs && bdrv_is_inserted(bs);
1270 }
1271
1272 bool blk_is_available(BlockBackend *blk)
1273 {
1274     return blk_is_inserted(blk) && !blk_dev_is_tray_open(blk);
1275 }
1276
1277 void blk_lock_medium(BlockBackend *blk, bool locked)
1278 {
1279     BlockDriverState *bs = blk_bs(blk);
1280
1281     if (bs) {
1282         bdrv_lock_medium(bs, locked);
1283     }
1284 }
1285
1286 void blk_eject(BlockBackend *blk, bool eject_flag)
1287 {
1288     BlockDriverState *bs = blk_bs(blk);
1289
1290     if (bs) {
1291         bdrv_eject(bs, eject_flag);
1292     }
1293 }
1294
1295 int blk_get_flags(BlockBackend *blk)
1296 {
1297     BlockDriverState *bs = blk_bs(blk);
1298
1299     if (bs) {
1300         return bdrv_get_flags(bs);
1301     } else {
1302         return blk->root_state.open_flags;
1303     }
1304 }
1305
1306 /* Returns the maximum transfer length, in bytes; guaranteed nonzero */
1307 uint32_t blk_get_max_transfer(BlockBackend *blk)
1308 {
1309     BlockDriverState *bs = blk_bs(blk);
1310     uint32_t max = 0;
1311
1312     if (bs) {
1313         max = bs->bl.max_transfer;
1314     }
1315     return MIN_NON_ZERO(max, INT_MAX);
1316 }
1317
1318 int blk_get_max_iov(BlockBackend *blk)
1319 {
1320     return blk->root->bs->bl.max_iov;
1321 }
1322
1323 void blk_set_guest_block_size(BlockBackend *blk, int align)
1324 {
1325     blk->guest_block_size = align;
1326 }
1327
1328 void *blk_try_blockalign(BlockBackend *blk, size_t size)
1329 {
1330     return qemu_try_blockalign(blk ? blk_bs(blk) : NULL, size);
1331 }
1332
1333 void *blk_blockalign(BlockBackend *blk, size_t size)
1334 {
1335     return qemu_blockalign(blk ? blk_bs(blk) : NULL, size);
1336 }
1337
1338 bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp)
1339 {
1340     BlockDriverState *bs = blk_bs(blk);
1341
1342     if (!bs) {
1343         return false;
1344     }
1345
1346     return bdrv_op_is_blocked(bs, op, errp);
1347 }
1348
1349 void blk_op_unblock(BlockBackend *blk, BlockOpType op, Error *reason)
1350 {
1351     BlockDriverState *bs = blk_bs(blk);
1352
1353     if (bs) {
1354         bdrv_op_unblock(bs, op, reason);
1355     }
1356 }
1357
1358 void blk_op_block_all(BlockBackend *blk, Error *reason)
1359 {
1360     BlockDriverState *bs = blk_bs(blk);
1361
1362     if (bs) {
1363         bdrv_op_block_all(bs, reason);
1364     }
1365 }
1366
1367 void blk_op_unblock_all(BlockBackend *blk, Error *reason)
1368 {
1369     BlockDriverState *bs = blk_bs(blk);
1370
1371     if (bs) {
1372         bdrv_op_unblock_all(bs, reason);
1373     }
1374 }
1375
1376 AioContext *blk_get_aio_context(BlockBackend *blk)
1377 {
1378     BlockDriverState *bs = blk_bs(blk);
1379
1380     if (bs) {
1381         return bdrv_get_aio_context(bs);
1382     } else {
1383         return qemu_get_aio_context();
1384     }
1385 }
1386
1387 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb)
1388 {
1389     BlockBackendAIOCB *blk_acb = DO_UPCAST(BlockBackendAIOCB, common, acb);
1390     return blk_get_aio_context(blk_acb->blk);
1391 }
1392
1393 void blk_set_aio_context(BlockBackend *blk, AioContext *new_context)
1394 {
1395     BlockDriverState *bs = blk_bs(blk);
1396
1397     if (bs) {
1398         if (blk->public.throttle_state) {
1399             throttle_timers_detach_aio_context(&blk->public.throttle_timers);
1400         }
1401         bdrv_set_aio_context(bs, new_context);
1402         if (blk->public.throttle_state) {
1403             throttle_timers_attach_aio_context(&blk->public.throttle_timers,
1404                                                new_context);
1405         }
1406     }
1407 }
1408
1409 void blk_add_aio_context_notifier(BlockBackend *blk,
1410         void (*attached_aio_context)(AioContext *new_context, void *opaque),
1411         void (*detach_aio_context)(void *opaque), void *opaque)
1412 {
1413     BlockDriverState *bs = blk_bs(blk);
1414
1415     if (bs) {
1416         bdrv_add_aio_context_notifier(bs, attached_aio_context,
1417                                       detach_aio_context, opaque);
1418     }
1419 }
1420
1421 void blk_remove_aio_context_notifier(BlockBackend *blk,
1422                                      void (*attached_aio_context)(AioContext *,
1423                                                                   void *),
1424                                      void (*detach_aio_context)(void *),
1425                                      void *opaque)
1426 {
1427     BlockDriverState *bs = blk_bs(blk);
1428
1429     if (bs) {
1430         bdrv_remove_aio_context_notifier(bs, attached_aio_context,
1431                                          detach_aio_context, opaque);
1432     }
1433 }
1434
1435 void blk_add_remove_bs_notifier(BlockBackend *blk, Notifier *notify)
1436 {
1437     notifier_list_add(&blk->remove_bs_notifiers, notify);
1438 }
1439
1440 void blk_add_insert_bs_notifier(BlockBackend *blk, Notifier *notify)
1441 {
1442     notifier_list_add(&blk->insert_bs_notifiers, notify);
1443 }
1444
1445 void blk_io_plug(BlockBackend *blk)
1446 {
1447     BlockDriverState *bs = blk_bs(blk);
1448
1449     if (bs) {
1450         bdrv_io_plug(bs);
1451     }
1452 }
1453
1454 void blk_io_unplug(BlockBackend *blk)
1455 {
1456     BlockDriverState *bs = blk_bs(blk);
1457
1458     if (bs) {
1459         bdrv_io_unplug(bs);
1460     }
1461 }
1462
1463 BlockAcctStats *blk_get_stats(BlockBackend *blk)
1464 {
1465     return &blk->stats;
1466 }
1467
1468 void *blk_aio_get(const AIOCBInfo *aiocb_info, BlockBackend *blk,
1469                   BlockCompletionFunc *cb, void *opaque)
1470 {
1471     return qemu_aio_get(aiocb_info, blk_bs(blk), cb, opaque);
1472 }
1473
1474 int coroutine_fn blk_co_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1475                                       int count, BdrvRequestFlags flags)
1476 {
1477     return blk_co_pwritev(blk, offset, count, NULL,
1478                           flags | BDRV_REQ_ZERO_WRITE);
1479 }
1480
1481 int blk_write_compressed(BlockBackend *blk, int64_t sector_num,
1482                          const uint8_t *buf, int nb_sectors)
1483 {
1484     int ret = blk_check_request(blk, sector_num, nb_sectors);
1485     if (ret < 0) {
1486         return ret;
1487     }
1488
1489     return bdrv_write_compressed(blk_bs(blk), sector_num, buf, nb_sectors);
1490 }
1491
1492 int blk_truncate(BlockBackend *blk, int64_t offset)
1493 {
1494     if (!blk_is_available(blk)) {
1495         return -ENOMEDIUM;
1496     }
1497
1498     return bdrv_truncate(blk_bs(blk), offset);
1499 }
1500
1501 int blk_discard(BlockBackend *blk, int64_t sector_num, int nb_sectors)
1502 {
1503     int ret = blk_check_request(blk, sector_num, nb_sectors);
1504     if (ret < 0) {
1505         return ret;
1506     }
1507
1508     return bdrv_discard(blk_bs(blk), sector_num, nb_sectors);
1509 }
1510
1511 int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
1512                      int64_t pos, int size)
1513 {
1514     int ret;
1515
1516     if (!blk_is_available(blk)) {
1517         return -ENOMEDIUM;
1518     }
1519
1520     ret = bdrv_save_vmstate(blk_bs(blk), buf, pos, size);
1521     if (ret < 0) {
1522         return ret;
1523     }
1524
1525     if (ret == size && !blk->enable_write_cache) {
1526         ret = bdrv_flush(blk_bs(blk));
1527     }
1528
1529     return ret < 0 ? ret : size;
1530 }
1531
1532 int blk_load_vmstate(BlockBackend *blk, uint8_t *buf, int64_t pos, int size)
1533 {
1534     if (!blk_is_available(blk)) {
1535         return -ENOMEDIUM;
1536     }
1537
1538     return bdrv_load_vmstate(blk_bs(blk), buf, pos, size);
1539 }
1540
1541 int blk_probe_blocksizes(BlockBackend *blk, BlockSizes *bsz)
1542 {
1543     if (!blk_is_available(blk)) {
1544         return -ENOMEDIUM;
1545     }
1546
1547     return bdrv_probe_blocksizes(blk_bs(blk), bsz);
1548 }
1549
1550 int blk_probe_geometry(BlockBackend *blk, HDGeometry *geo)
1551 {
1552     if (!blk_is_available(blk)) {
1553         return -ENOMEDIUM;
1554     }
1555
1556     return bdrv_probe_geometry(blk_bs(blk), geo);
1557 }
1558
1559 /*
1560  * Updates the BlockBackendRootState object with data from the currently
1561  * attached BlockDriverState.
1562  */
1563 void blk_update_root_state(BlockBackend *blk)
1564 {
1565     assert(blk->root);
1566
1567     blk->root_state.open_flags    = blk->root->bs->open_flags;
1568     blk->root_state.read_only     = blk->root->bs->read_only;
1569     blk->root_state.detect_zeroes = blk->root->bs->detect_zeroes;
1570 }
1571
1572 /*
1573  * Applies the information in the root state to the given BlockDriverState. This
1574  * does not include the flags which have to be specified for bdrv_open(), use
1575  * blk_get_open_flags_from_root_state() to inquire them.
1576  */
1577 void blk_apply_root_state(BlockBackend *blk, BlockDriverState *bs)
1578 {
1579     bs->detect_zeroes = blk->root_state.detect_zeroes;
1580 }
1581
1582 /*
1583  * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
1584  * supposed to inherit the root state.
1585  */
1586 int blk_get_open_flags_from_root_state(BlockBackend *blk)
1587 {
1588     int bs_flags;
1589
1590     bs_flags = blk->root_state.read_only ? 0 : BDRV_O_RDWR;
1591     bs_flags |= blk->root_state.open_flags & ~BDRV_O_RDWR;
1592
1593     return bs_flags;
1594 }
1595
1596 BlockBackendRootState *blk_get_root_state(BlockBackend *blk)
1597 {
1598     return &blk->root_state;
1599 }
1600
1601 int blk_commit_all(void)
1602 {
1603     BlockBackend *blk = NULL;
1604
1605     while ((blk = blk_all_next(blk)) != NULL) {
1606         AioContext *aio_context = blk_get_aio_context(blk);
1607
1608         aio_context_acquire(aio_context);
1609         if (blk_is_inserted(blk) && blk->root->bs->backing) {
1610             int ret = bdrv_commit(blk->root->bs);
1611             if (ret < 0) {
1612                 aio_context_release(aio_context);
1613                 return ret;
1614             }
1615         }
1616         aio_context_release(aio_context);
1617     }
1618     return 0;
1619 }
1620
1621 int blk_flush_all(void)
1622 {
1623     BlockBackend *blk = NULL;
1624     int result = 0;
1625
1626     while ((blk = blk_all_next(blk)) != NULL) {
1627         AioContext *aio_context = blk_get_aio_context(blk);
1628         int ret;
1629
1630         aio_context_acquire(aio_context);
1631         if (blk_is_inserted(blk)) {
1632             ret = blk_flush(blk);
1633             if (ret < 0 && !result) {
1634                 result = ret;
1635             }
1636         }
1637         aio_context_release(aio_context);
1638     }
1639
1640     return result;
1641 }
1642
1643
1644 /* throttling disk I/O limits */
1645 void blk_set_io_limits(BlockBackend *blk, ThrottleConfig *cfg)
1646 {
1647     throttle_group_config(blk, cfg);
1648 }
1649
1650 void blk_io_limits_disable(BlockBackend *blk)
1651 {
1652     assert(blk->public.throttle_state);
1653     bdrv_drained_begin(blk_bs(blk));
1654     throttle_group_unregister_blk(blk);
1655     bdrv_drained_end(blk_bs(blk));
1656 }
1657
1658 /* should be called before blk_set_io_limits if a limit is set */
1659 void blk_io_limits_enable(BlockBackend *blk, const char *group)
1660 {
1661     assert(!blk->public.throttle_state);
1662     throttle_group_register_blk(blk, group);
1663 }
1664
1665 void blk_io_limits_update_group(BlockBackend *blk, const char *group)
1666 {
1667     /* this BB is not part of any group */
1668     if (!blk->public.throttle_state) {
1669         return;
1670     }
1671
1672     /* this BB is a part of the same group than the one we want */
1673     if (!g_strcmp0(throttle_group_get_name(blk), group)) {
1674         return;
1675     }
1676
1677     /* need to change the group this bs belong to */
1678     blk_io_limits_disable(blk);
1679     blk_io_limits_enable(blk, group);
1680 }
1681
1682 static void blk_root_drained_begin(BdrvChild *child)
1683 {
1684     BlockBackend *blk = child->opaque;
1685
1686     /* Note that blk->root may not be accessible here yet if we are just
1687      * attaching to a BlockDriverState that is drained. Use child instead. */
1688
1689     if (blk->public.io_limits_disabled++ == 0) {
1690         throttle_group_restart_blk(blk);
1691     }
1692 }
1693
1694 static void blk_root_drained_end(BdrvChild *child)
1695 {
1696     BlockBackend *blk = child->opaque;
1697
1698     assert(blk->public.io_limits_disabled);
1699     --blk->public.io_limits_disabled;
1700 }