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