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(Error **errp)
 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  * Create a new BlockBackend with a new BlockDriverState attached.
 140  * Otherwise just like blk_new(), which see.
 141  */
 142 BlockBackend *blk_new_with_bs(Error **errp)
 143 {
 144     BlockBackend *blk;
 145     BlockDriverState *bs;
 146
 147     blk = blk_new(errp);
 148     if (!blk) {
 149         return NULL;
 150     }
 151
 152     bs = bdrv_new_root();
 153     blk->root = bdrv_root_attach_child(bs, "root", &child_root);
 154     blk->root->opaque = blk;
 155     return blk;
 156 }
 157
 158 /*
 159  * Calls blk_new_with_bs() and then calls bdrv_open() on the BlockDriverState.
 160  *
 161  * Just as with bdrv_open(), after having called this function the reference to
 162  * @options belongs to the block layer (even on failure).
 163  *
 164  * TODO: Remove @filename and @flags; it should be possible to specify a whole
 165  * BDS tree just by specifying the @options QDict (or @reference,
 166  * alternatively). At the time of adding this function, this is not possible,
 167  * though, so callers of this function have to be able to specify @filename and
 168  * @flags.
 169  */
 170 BlockBackend *blk_new_open(const char *filename, const char *reference,
 171                            QDict *options, int flags, Error **errp)
 172 {
 173     BlockBackend *blk;
 174     int ret;
 175
 176     blk = blk_new_with_bs(errp);
 177     if (!blk) {
 178         QDECREF(options);
 179         return NULL;
 180     }
 181
 182     ret = bdrv_open(&blk->root->bs, filename, reference, options, flags, errp);
 183     if (ret < 0) {
 184         blk_unref(blk);
 185         return NULL;
 186     }
 187
 188     blk_set_enable_write_cache(blk, true);
 189
 190     return blk;
 191 }
 192
 193 static void blk_delete(BlockBackend *blk)
 194 {
 195     assert(!blk->refcnt);
 196     assert(!blk->name);
 197     assert(!blk->dev);
 198     if (blk->root) {
 199         blk_remove_bs(blk);
 200     }
 201     assert(QLIST_EMPTY(&blk->remove_bs_notifiers.notifiers));
 202     assert(QLIST_EMPTY(&blk->insert_bs_notifiers.notifiers));
 203     QTAILQ_REMOVE(&block_backends, blk, link);
 204     drive_info_del(blk->legacy_dinfo);
 205     block_acct_cleanup(&blk->stats);
 206     g_free(blk);
 207 }
 208
 209 static void drive_info_del(DriveInfo *dinfo)
 210 {
 211     if (!dinfo) {
 212         return;
 213     }
 214     qemu_opts_del(dinfo->opts);
 215     g_free(dinfo->serial);
 216     g_free(dinfo);
 217 }
 218
 219 int blk_get_refcnt(BlockBackend *blk)
 220 {
 221     return blk ? blk->refcnt : 0;
 222 }
 223
 224 /*
 225  * Increment @blk's reference count.
 226  * @blk must not be null.
 227  */
 228 void blk_ref(BlockBackend *blk)
 229 {
 230     blk->refcnt++;
 231 }
 232
 233 /*
 234  * Decrement @blk's reference count.
 235  * If this drops it to zero, destroy @blk.
 236  * For convenience, do nothing if @blk is null.
 237  */
 238 void blk_unref(BlockBackend *blk)
 239 {
 240     if (blk) {
 241         assert(blk->refcnt > 0);
 242         if (!--blk->refcnt) {
 243             blk_delete(blk);
 244         }
 245     }
 246 }
 247
 248 /*
 249  * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
 250  * ones which are hidden (i.e. are not referenced by the monitor).
 251  */
 252 static BlockBackend *blk_all_next(BlockBackend *blk)
 253 {
 254     return blk ? QTAILQ_NEXT(blk, link)
 255                : QTAILQ_FIRST(&block_backends);
 256 }
 257
 258 void blk_remove_all_bs(void)
 259 {
 260     BlockBackend *blk = NULL;
 261
 262     while ((blk = blk_all_next(blk)) != NULL) {
 263         AioContext *ctx = blk_get_aio_context(blk);
 264
 265         aio_context_acquire(ctx);
 266         if (blk->root) {
 267             blk_remove_bs(blk);
 268         }
 269         aio_context_release(ctx);
 270     }
 271 }
 272
 273 /*
 274  * Return the monitor-owned BlockBackend after @blk.
 275  * If @blk is null, return the first one.
 276  * Else, return @blk's next sibling, which may be null.
 277  *
 278  * To iterate over all BlockBackends, do
 279  * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
 280  *     ...
 281  * }
 282  */
 283 BlockBackend *blk_next(BlockBackend *blk)
 284 {
 285     return blk ? QTAILQ_NEXT(blk, monitor_link)
 286                : QTAILQ_FIRST(&monitor_block_backends);
 287 }
 288
 289 struct BdrvNextIterator {
 290     enum {
 291         BDRV_NEXT_BACKEND_ROOTS,
 292         BDRV_NEXT_MONITOR_OWNED,
 293     } phase;
 294     BlockBackend *blk;
 295     BlockDriverState *bs;
 296 };
 297
 298 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
 299  * the monitor or attached to a BlockBackend */
 300 BdrvNextIterator *bdrv_next(BdrvNextIterator *it, BlockDriverState **bs)
 301 {
 302     if (!it) {
 303         it = g_new(BdrvNextIterator, 1);
 304         *it = (BdrvNextIterator) {
 305             .phase = BDRV_NEXT_BACKEND_ROOTS,
 306         };
 307     }
 308
 309     /* First, return all root nodes of BlockBackends. In order to avoid
 310      * returning a BDS twice when multiple BBs refer to it, we only return it
 311      * if the BB is the first one in the parent list of the BDS. */
 312     if (it->phase == BDRV_NEXT_BACKEND_ROOTS) {
 313         do {
 314             it->blk = blk_all_next(it->blk);
 315             *bs = it->blk ? blk_bs(it->blk) : NULL;
 316         } while (it->blk && (*bs == NULL || bdrv_first_blk(*bs) != it->blk));
 317
 318         if (*bs) {
 319             return it;
 320         }
 321         it->phase = BDRV_NEXT_MONITOR_OWNED;
 322     }
 323
 324     /* Then return the monitor-owned BDSes without a BB attached. Ignore all
 325      * BDSes that are attached to a BlockBackend here; they have been handled
 326      * by the above block already */
 327     do {
 328         it->bs = bdrv_next_monitor_owned(it->bs);
 329         *bs = it->bs;
 330     } while (*bs && bdrv_has_blk(*bs));
 331
 332     return *bs ? it : NULL;
 333 }
 334
 335 /*
 336  * Add a BlockBackend into the list of backends referenced by the monitor, with
 337  * the given @name acting as the handle for the monitor.
 338  * Strictly for use by blockdev.c.
 339  *
 340  * @name must not be null or empty.
 341  *
 342  * Returns true on success and false on failure. In the latter case, an Error
 343  * object is returned through @errp.
 344  */
 345 bool monitor_add_blk(BlockBackend *blk, const char *name, Error **errp)
 346 {
 347     assert(!blk->name);
 348     assert(name && name[0]);
 349
 350     if (!id_wellformed(name)) {
 351         error_setg(errp, "Invalid device name");
 352         return false;
 353     }
 354     if (blk_by_name(name)) {
 355         error_setg(errp, "Device with id '%s' already exists", name);
 356         return false;
 357     }
 358     if (bdrv_find_node(name)) {
 359         error_setg(errp,
 360                    "Device name '%s' conflicts with an existing node name",
 361                    name);
 362         return false;
 363     }
 364
 365     blk->name = g_strdup(name);
 366     QTAILQ_INSERT_TAIL(&monitor_block_backends, blk, monitor_link);
 367     return true;
 368 }
 369
 370 /*
 371  * Remove a BlockBackend from the list of backends referenced by the monitor.
 372  * Strictly for use by blockdev.c.
 373  */
 374 void monitor_remove_blk(BlockBackend *blk)
 375 {
 376     if (!blk->name) {
 377         return;
 378     }
 379
 380     QTAILQ_REMOVE(&monitor_block_backends, blk, monitor_link);
 381     g_free(blk->name);
 382     blk->name = NULL;
 383 }
 384
 385 /*
 386  * Return @blk's name, a non-null string.
 387  * Returns an empty string iff @blk is not referenced by the monitor.
 388  */
 389 const char *blk_name(BlockBackend *blk)
 390 {
 391     return blk->name ?: "";
 392 }
 393
 394 /*
 395  * Return the BlockBackend with name @name if it exists, else null.
 396  * @name must not be null.
 397  */
 398 BlockBackend *blk_by_name(const char *name)
 399 {
 400     BlockBackend *blk = NULL;
 401
 402     assert(name);
 403     while ((blk = blk_next(blk)) != NULL) {
 404         if (!strcmp(name, blk->name)) {
 405             return blk;
 406         }
 407     }
 408     return NULL;
 409 }
 410
 411 /*
 412  * Return the BlockDriverState attached to @blk if any, else null.
 413  */
 414 BlockDriverState *blk_bs(BlockBackend *blk)
 415 {
 416     return blk->root ? blk->root->bs : NULL;
 417 }
 418
 419 static BlockBackend *bdrv_first_blk(BlockDriverState *bs)
 420 {
 421     BdrvChild *child;
 422     QLIST_FOREACH(child, &bs->parents, next_parent) {
 423         if (child->role == &child_root) {
 424             return child->opaque;
 425         }
 426     }
 427
 428     return NULL;
 429 }
 430
 431 /*
 432  * Returns true if @bs has an associated BlockBackend.
 433  */
 434 bool bdrv_has_blk(BlockDriverState *bs)
 435 {
 436     return bdrv_first_blk(bs) != NULL;
 437 }
 438
 439 /*
 440  * Return @blk's DriveInfo if any, else null.
 441  */
 442 DriveInfo *blk_legacy_dinfo(BlockBackend *blk)
 443 {
 444     return blk->legacy_dinfo;
 445 }
 446
 447 /*
 448  * Set @blk's DriveInfo to @dinfo, and return it.
 449  * @blk must not have a DriveInfo set already.
 450  * No other BlockBackend may have the same DriveInfo set.
 451  */
 452 DriveInfo *blk_set_legacy_dinfo(BlockBackend *blk, DriveInfo *dinfo)
 453 {
 454     assert(!blk->legacy_dinfo);
 455     return blk->legacy_dinfo = dinfo;
 456 }
 457
 458 /*
 459  * Return the BlockBackend with DriveInfo @dinfo.
 460  * It must exist.
 461  */
 462 BlockBackend *blk_by_legacy_dinfo(DriveInfo *dinfo)
 463 {
 464     BlockBackend *blk = NULL;
 465
 466     while ((blk = blk_next(blk)) != NULL) {
 467         if (blk->legacy_dinfo == dinfo) {
 468             return blk;
 469         }
 470     }
 471     abort();
 472 }
 473
 474 /*
 475  * Returns a pointer to the publicly accessible fields of @blk.
 476  */
 477 BlockBackendPublic *blk_get_public(BlockBackend *blk)
 478 {
 479     return &blk->public;
 480 }
 481
 482 /*
 483  * Returns a BlockBackend given the associated @public fields.
 484  */
 485 BlockBackend *blk_by_public(BlockBackendPublic *public)
 486 {
 487     return container_of(public, BlockBackend, public);
 488 }
 489
 490 /*
 491  * Disassociates the currently associated BlockDriverState from @blk.
 492  */
 493 void blk_remove_bs(BlockBackend *blk)
 494 {
 495     notifier_list_notify(&blk->remove_bs_notifiers, blk);
 496     if (blk->public.throttle_state) {
 497         throttle_timers_detach_aio_context(&blk->public.throttle_timers);
 498     }
 499
 500     blk_update_root_state(blk);
 501
 502     bdrv_root_unref_child(blk->root);
 503     blk->root = NULL;
 504 }
 505
 506 /*
 507  * Associates a new BlockDriverState with @blk.
 508  */
 509 void blk_insert_bs(BlockBackend *blk, BlockDriverState *bs)
 510 {
 511     bdrv_ref(bs);
 512     blk->root = bdrv_root_attach_child(bs, "root", &child_root);
 513     blk->root->opaque = blk;
 514
 515     notifier_list_notify(&blk->insert_bs_notifiers, blk);
 516     if (blk->public.throttle_state) {
 517         throttle_timers_attach_aio_context(
 518             &blk->public.throttle_timers, bdrv_get_aio_context(bs));
 519     }
 520 }
 521
 522 /*
 523  * Attach device model @dev to @blk.
 524  * Return 0 on success, -EBUSY when a device model is attached already.
 525  */
 526 int blk_attach_dev(BlockBackend *blk, void *dev)
 527 /* TODO change to DeviceState *dev when all users are qdevified */
 528 {
 529     if (blk->dev) {
 530         return -EBUSY;
 531     }
 532     blk_ref(blk);
 533     blk->dev = dev;
 534     blk_iostatus_reset(blk);
 535     return 0;
 536 }
 537
 538 /*
 539  * Attach device model @dev to @blk.
 540  * @blk must not have a device model attached already.
 541  * TODO qdevified devices don't use this, remove when devices are qdevified
 542  */
 543 void blk_attach_dev_nofail(BlockBackend *blk, void *dev)
 544 {
 545     if (blk_attach_dev(blk, dev) < 0) {
 546         abort();
 547     }
 548 }
 549
 550 /*
 551  * Detach device model @dev from @blk.
 552  * @dev must be currently attached to @blk.
 553  */
 554 void blk_detach_dev(BlockBackend *blk, void *dev)
 555 /* TODO change to DeviceState *dev when all users are qdevified */
 556 {
 557     assert(blk->dev == dev);
 558     blk->dev = NULL;
 559     blk->dev_ops = NULL;
 560     blk->dev_opaque = NULL;
 561     blk->guest_block_size = 512;
 562     blk_unref(blk);
 563 }
 564
 565 /*
 566  * Return the device model attached to @blk if any, else null.
 567  */
 568 void *blk_get_attached_dev(BlockBackend *blk)
 569 /* TODO change to return DeviceState * when all users are qdevified */
 570 {
 571     return blk->dev;
 572 }
 573
 574 /*
 575  * Set @blk's device model callbacks to @ops.
 576  * @opaque is the opaque argument to pass to the callbacks.
 577  * This is for use by device models.
 578  */
 579 void blk_set_dev_ops(BlockBackend *blk, const BlockDevOps *ops,
 580                      void *opaque)
 581 {
 582     blk->dev_ops = ops;
 583     blk->dev_opaque = opaque;
 584 }
 585
 586 /*
 587  * Notify @blk's attached device model of media change.
 588  * If @load is true, notify of media load.
 589  * Else, notify of media eject.
 590  * Also send DEVICE_TRAY_MOVED events as appropriate.
 591  */
 592 void blk_dev_change_media_cb(BlockBackend *blk, bool load)
 593 {
 594     if (blk->dev_ops && blk->dev_ops->change_media_cb) {
 595         bool tray_was_open, tray_is_open;
 596
 597         tray_was_open = blk_dev_is_tray_open(blk);
 598         blk->dev_ops->change_media_cb(blk->dev_opaque, load);
 599         tray_is_open = blk_dev_is_tray_open(blk);
 600
 601         if (tray_was_open != tray_is_open) {
 602             qapi_event_send_device_tray_moved(blk_name(blk), tray_is_open,
 603                                               &error_abort);
 604         }
 605     }
 606 }
 607
 608 static void blk_root_change_media(BdrvChild *child, bool load)
 609 {
 610     blk_dev_change_media_cb(child->opaque, load);
 611 }
 612
 613 /*
 614  * Does @blk's attached device model have removable media?
 615  * %true if no device model is attached.
 616  */
 617 bool blk_dev_has_removable_media(BlockBackend *blk)
 618 {
 619     return !blk->dev || (blk->dev_ops && blk->dev_ops->change_media_cb);
 620 }
 621
 622 /*
 623  * Does @blk's attached device model have a tray?
 624  */
 625 bool blk_dev_has_tray(BlockBackend *blk)
 626 {
 627     return blk->dev_ops && blk->dev_ops->is_tray_open;
 628 }
 629
 630 /*
 631  * Notify @blk's attached device model of a media eject request.
 632  * If @force is true, the medium is about to be yanked out forcefully.
 633  */
 634 void blk_dev_eject_request(BlockBackend *blk, bool force)
 635 {
 636     if (blk->dev_ops && blk->dev_ops->eject_request_cb) {
 637         blk->dev_ops->eject_request_cb(blk->dev_opaque, force);
 638     }
 639 }
 640
 641 /*
 642  * Does @blk's attached device model have a tray, and is it open?
 643  */
 644 bool blk_dev_is_tray_open(BlockBackend *blk)
 645 {
 646     if (blk_dev_has_tray(blk)) {
 647         return blk->dev_ops->is_tray_open(blk->dev_opaque);
 648     }
 649     return false;
 650 }
 651
 652 /*
 653  * Does @blk's attached device model have the medium locked?
 654  * %false if the device model has no such lock.
 655  */
 656 bool blk_dev_is_medium_locked(BlockBackend *blk)
 657 {
 658     if (blk->dev_ops && blk->dev_ops->is_medium_locked) {
 659         return blk->dev_ops->is_medium_locked(blk->dev_opaque);
 660     }
 661     return false;
 662 }
 663
 664 /*
 665  * Notify @blk's attached device model of a backend size change.
 666  */
 667 static void blk_root_resize(BdrvChild *child)
 668 {
 669     BlockBackend *blk = child->opaque;
 670
 671     if (blk->dev_ops && blk->dev_ops->resize_cb) {
 672         blk->dev_ops->resize_cb(blk->dev_opaque);
 673     }
 674 }
 675
 676 void blk_iostatus_enable(BlockBackend *blk)
 677 {
 678     blk->iostatus_enabled = true;
 679     blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 680 }
 681
 682 /* The I/O status is only enabled if the drive explicitly
 683  * enables it _and_ the VM is configured to stop on errors */
 684 bool blk_iostatus_is_enabled(const BlockBackend *blk)
 685 {
 686     return (blk->iostatus_enabled &&
 687            (blk->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
 688             blk->on_write_error == BLOCKDEV_ON_ERROR_STOP   ||
 689             blk->on_read_error == BLOCKDEV_ON_ERROR_STOP));
 690 }
 691
 692 BlockDeviceIoStatus blk_iostatus(const BlockBackend *blk)
 693 {
 694     return blk->iostatus;
 695 }
 696
 697 void blk_iostatus_disable(BlockBackend *blk)
 698 {
 699     blk->iostatus_enabled = false;
 700 }
 701
 702 void blk_iostatus_reset(BlockBackend *blk)
 703 {
 704     if (blk_iostatus_is_enabled(blk)) {
 705         BlockDriverState *bs = blk_bs(blk);
 706         blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 707         if (bs && bs->job) {
 708             block_job_iostatus_reset(bs->job);
 709         }
 710     }
 711 }
 712
 713 void blk_iostatus_set_err(BlockBackend *blk, int error)
 714 {
 715     assert(blk_iostatus_is_enabled(blk));
 716     if (blk->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 717         blk->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 718                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 719     }
 720 }
 721
 722 void blk_set_allow_write_beyond_eof(BlockBackend *blk, bool allow)
 723 {
 724     blk->allow_write_beyond_eof = allow;
 725 }
 726
 727 static int blk_check_byte_request(BlockBackend *blk, int64_t offset,
 728                                   size_t size)
 729 {
 730     int64_t len;
 731
 732     if (size > INT_MAX) {
 733         return -EIO;
 734     }
 735
 736     if (!blk_is_available(blk)) {
 737         return -ENOMEDIUM;
 738     }
 739
 740     if (offset < 0) {
 741         return -EIO;
 742     }
 743
 744     if (!blk->allow_write_beyond_eof) {
 745         len = blk_getlength(blk);
 746         if (len < 0) {
 747             return len;
 748         }
 749
 750         if (offset > len || len - offset < size) {
 751             return -EIO;
 752         }
 753     }
 754
 755     return 0;
 756 }
 757
 758 static int blk_check_request(BlockBackend *blk, int64_t sector_num,
 759                              int nb_sectors)
 760 {
 761     if (sector_num < 0 || sector_num > INT64_MAX / BDRV_SECTOR_SIZE) {
 762         return -EIO;
 763     }
 764
 765     if (nb_sectors < 0 || nb_sectors > INT_MAX / BDRV_SECTOR_SIZE) {
 766         return -EIO;
 767     }
 768
 769     return blk_check_byte_request(blk, sector_num * BDRV_SECTOR_SIZE,
 770                                   nb_sectors * BDRV_SECTOR_SIZE);
 771 }
 772
 773 static int coroutine_fn blk_co_preadv(BlockBackend *blk, int64_t offset,
 774                                       unsigned int bytes, QEMUIOVector *qiov,
 775                                       BdrvRequestFlags flags)
 776 {
 777     int ret = blk_check_byte_request(blk, offset, bytes);
 778     if (ret < 0) {
 779         return ret;
 780     }
 781
 782     /* throttling disk I/O */
 783     if (blk->public.throttle_state) {
 784         throttle_group_co_io_limits_intercept(blk, bytes, false);
 785     }
 786
 787     return bdrv_co_preadv(blk_bs(blk), offset, bytes, qiov, flags);
 788 }
 789
 790 static int coroutine_fn blk_co_pwritev(BlockBackend *blk, int64_t offset,
 791                                       unsigned int bytes, QEMUIOVector *qiov,
 792                                       BdrvRequestFlags flags)
 793 {
 794     int ret;
 795
 796     ret = blk_check_byte_request(blk, offset, bytes);
 797     if (ret < 0) {
 798         return ret;
 799     }
 800
 801     /* throttling disk I/O */
 802     if (blk->public.throttle_state) {
 803         throttle_group_co_io_limits_intercept(blk, bytes, true);
 804     }
 805
 806     if (!blk->enable_write_cache) {
 807         flags |= BDRV_REQ_FUA;
 808     }
 809
 810     return bdrv_co_pwritev(blk_bs(blk), offset, bytes, qiov, flags);
 811 }
 812
 813 typedef struct BlkRwCo {
 814     BlockBackend *blk;
 815     int64_t offset;
 816     QEMUIOVector *qiov;
 817     int ret;
 818     BdrvRequestFlags flags;
 819 } BlkRwCo;
 820
 821 static void blk_read_entry(void *opaque)
 822 {
 823     BlkRwCo *rwco = opaque;
 824
 825     rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, rwco->qiov->size,
 826                               rwco->qiov, rwco->flags);
 827 }
 828
 829 static void blk_write_entry(void *opaque)
 830 {
 831     BlkRwCo *rwco = opaque;
 832
 833     rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, rwco->qiov->size,
 834                                rwco->qiov, rwco->flags);
 835 }
 836
 837 static int blk_prw(BlockBackend *blk, int64_t offset, uint8_t *buf,
 838                    int64_t bytes, CoroutineEntry co_entry,
 839                    BdrvRequestFlags flags)
 840 {
 841     AioContext *aio_context;
 842     QEMUIOVector qiov;
 843     struct iovec iov;
 844     Coroutine *co;
 845     BlkRwCo rwco;
 846
 847     iov = (struct iovec) {
 848         .iov_base = buf,
 849         .iov_len = bytes,
 850     };
 851     qemu_iovec_init_external(&qiov, &iov, 1);
 852
 853     rwco = (BlkRwCo) {
 854         .blk    = blk,
 855         .offset = offset,
 856         .qiov   = &qiov,
 857         .flags  = flags,
 858         .ret    = NOT_DONE,
 859     };
 860
 861     co = qemu_coroutine_create(co_entry);
 862     qemu_coroutine_enter(co, &rwco);
 863
 864     aio_context = blk_get_aio_context(blk);
 865     while (rwco.ret == NOT_DONE) {
 866         aio_poll(aio_context, true);
 867     }
 868
 869     return rwco.ret;
 870 }
 871
 872 int blk_pread_unthrottled(BlockBackend *blk, int64_t offset, uint8_t *buf,
 873                           int count)
 874 {
 875     int ret;
 876
 877     ret = blk_check_byte_request(blk, offset, count);
 878     if (ret < 0) {
 879         return ret;
 880     }
 881
 882     blk_root_drained_begin(blk->root);
 883     ret = blk_pread(blk, offset, buf, count);
 884     blk_root_drained_end(blk->root);
 885     return ret;
 886 }
 887
 888 int blk_write_zeroes(BlockBackend *blk, int64_t offset,
 889                      int count, BdrvRequestFlags flags)
 890 {
 891     return blk_prw(blk, offset, NULL, count, blk_write_entry,
 892                    flags | BDRV_REQ_ZERO_WRITE);
 893 }
 894
 895 static void error_callback_bh(void *opaque)
 896 {
 897     struct BlockBackendAIOCB *acb = opaque;
 898     qemu_bh_delete(acb->bh);
 899     acb->common.cb(acb->common.opaque, acb->ret);
 900     qemu_aio_unref(acb);
 901 }
 902
 903 BlockAIOCB *blk_abort_aio_request(BlockBackend *blk,
 904                                   BlockCompletionFunc *cb,
 905                                   void *opaque, int ret)
 906 {
 907     struct BlockBackendAIOCB *acb;
 908     QEMUBH *bh;
 909
 910     acb = blk_aio_get(&block_backend_aiocb_info, blk, cb, opaque);
 911     acb->blk = blk;
 912     acb->ret = ret;
 913
 914     bh = aio_bh_new(blk_get_aio_context(blk), error_callback_bh, acb);
 915     acb->bh = bh;
 916     qemu_bh_schedule(bh);
 917
 918     return &acb->common;
 919 }
 920
 921 typedef struct BlkAioEmAIOCB {
 922     BlockAIOCB common;
 923     BlkRwCo rwco;
 924     int bytes;
 925     bool has_returned;
 926     QEMUBH* bh;
 927 } BlkAioEmAIOCB;
 928
 929 static const AIOCBInfo blk_aio_em_aiocb_info = {
 930     .aiocb_size         = sizeof(BlkAioEmAIOCB),
 931 };
 932
 933 static void blk_aio_complete(BlkAioEmAIOCB *acb)
 934 {
 935     if (acb->bh) {
 936         assert(acb->has_returned);
 937         qemu_bh_delete(acb->bh);
 938     }
 939     if (acb->has_returned) {
 940         acb->common.cb(acb->common.opaque, acb->rwco.ret);
 941         qemu_aio_unref(acb);
 942     }
 943 }
 944
 945 static void blk_aio_complete_bh(void *opaque)
 946 {
 947     blk_aio_complete(opaque);
 948 }
 949
 950 static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset, int bytes,
 951                                 QEMUIOVector *qiov, CoroutineEntry co_entry,
 952                                 BdrvRequestFlags flags,
 953                                 BlockCompletionFunc *cb, void *opaque)
 954 {
 955     BlkAioEmAIOCB *acb;
 956     Coroutine *co;
 957
 958     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
 959     acb->rwco = (BlkRwCo) {
 960         .blk    = blk,
 961         .offset = offset,
 962         .qiov   = qiov,
 963         .flags  = flags,
 964         .ret    = NOT_DONE,
 965     };
 966     acb->bytes = bytes;
 967     acb->bh = NULL;
 968     acb->has_returned = false;
 969
 970     co = qemu_coroutine_create(co_entry);
 971     qemu_coroutine_enter(co, acb);
 972
 973     acb->has_returned = true;
 974     if (acb->rwco.ret != NOT_DONE) {
 975         acb->bh = aio_bh_new(blk_get_aio_context(blk), blk_aio_complete_bh, acb);
 976         qemu_bh_schedule(acb->bh);
 977     }
 978
 979     return &acb->common;
 980 }
 981
 982 static void blk_aio_read_entry(void *opaque)
 983 {
 984     BlkAioEmAIOCB *acb = opaque;
 985     BlkRwCo *rwco = &acb->rwco;
 986
 987     assert(rwco->qiov->size == acb->bytes);
 988     rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, acb->bytes,
 989                               rwco->qiov, rwco->flags);
 990     blk_aio_complete(acb);
 991 }
 992
 993 static void blk_aio_write_entry(void *opaque)
 994 {
 995     BlkAioEmAIOCB *acb = opaque;
 996     BlkRwCo *rwco = &acb->rwco;
 997
 998     assert(!rwco->qiov || rwco->qiov->size == acb->bytes);
 999     rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, acb->bytes,
1000                                rwco->qiov, rwco->flags);
1001     blk_aio_complete(acb);
1002 }
1003
1004 BlockAIOCB *blk_aio_write_zeroes(BlockBackend *blk, int64_t offset,
1005                                  int count, BdrvRequestFlags flags,
1006                                  BlockCompletionFunc *cb, void *opaque)
1007 {
1008     return blk_aio_prwv(blk, offset, count, NULL, blk_aio_write_entry,
1009                         flags | BDRV_REQ_ZERO_WRITE, cb, opaque);
1010 }
1011
1012 int blk_pread(BlockBackend *blk, int64_t offset, void *buf, int count)
1013 {
1014     int ret = blk_prw(blk, offset, buf, count, blk_read_entry, 0);
1015     if (ret < 0) {
1016         return ret;
1017     }
1018     return count;
1019 }
1020
1021 int blk_pwrite(BlockBackend *blk, int64_t offset, const void *buf, int count,
1022                BdrvRequestFlags flags)
1023 {
1024     int ret = blk_prw(blk, offset, (void *) buf, count, blk_write_entry,
1025                       flags);
1026     if (ret < 0) {
1027         return ret;
1028     }
1029     return count;
1030 }
1031
1032 int64_t blk_getlength(BlockBackend *blk)
1033 {
1034     if (!blk_is_available(blk)) {
1035         return -ENOMEDIUM;
1036     }
1037
1038     return bdrv_getlength(blk_bs(blk));
1039 }
1040
1041 void blk_get_geometry(BlockBackend *blk, uint64_t *nb_sectors_ptr)
1042 {
1043     if (!blk_bs(blk)) {
1044         *nb_sectors_ptr = 0;
1045     } else {
1046         bdrv_get_geometry(blk_bs(blk), nb_sectors_ptr);
1047     }
1048 }
1049
1050 int64_t blk_nb_sectors(BlockBackend *blk)
1051 {
1052     if (!blk_is_available(blk)) {
1053         return -ENOMEDIUM;
1054     }
1055
1056     return bdrv_nb_sectors(blk_bs(blk));
1057 }
1058
1059 BlockAIOCB *blk_aio_preadv(BlockBackend *blk, int64_t offset,
1060                            QEMUIOVector *qiov, BdrvRequestFlags flags,
1061                            BlockCompletionFunc *cb, void *opaque)
1062 {
1063     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1064                         blk_aio_read_entry, flags, cb, opaque);
1065 }
1066
1067 BlockAIOCB *blk_aio_pwritev(BlockBackend *blk, int64_t offset,
1068                             QEMUIOVector *qiov, BdrvRequestFlags flags,
1069                             BlockCompletionFunc *cb, void *opaque)
1070 {
1071     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1072                         blk_aio_write_entry, flags, cb, opaque);
1073 }
1074
1075 BlockAIOCB *blk_aio_flush(BlockBackend *blk,
1076                           BlockCompletionFunc *cb, void *opaque)
1077 {
1078     if (!blk_is_available(blk)) {
1079         return blk_abort_aio_request(blk, cb, opaque, -ENOMEDIUM);
1080     }
1081
1082     return bdrv_aio_flush(blk_bs(blk), cb, opaque);
1083 }
1084
1085 BlockAIOCB *blk_aio_discard(BlockBackend *blk,
1086                             int64_t sector_num, int nb_sectors,
1087                             BlockCompletionFunc *cb, void *opaque)
1088 {
1089     int ret = blk_check_request(blk, sector_num, nb_sectors);
1090     if (ret < 0) {
1091         return blk_abort_aio_request(blk, cb, opaque, ret);
1092     }
1093
1094     return bdrv_aio_discard(blk_bs(blk), sector_num, nb_sectors, cb, opaque);
1095 }
1096
1097 void blk_aio_cancel(BlockAIOCB *acb)
1098 {
1099     bdrv_aio_cancel(acb);
1100 }
1101
1102 void blk_aio_cancel_async(BlockAIOCB *acb)
1103 {
1104     bdrv_aio_cancel_async(acb);
1105 }
1106
1107 int blk_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1108 {
1109     if (!blk_is_available(blk)) {
1110         return -ENOMEDIUM;
1111     }
1112
1113     return bdrv_ioctl(blk_bs(blk), req, buf);
1114 }
1115
1116 BlockAIOCB *blk_aio_ioctl(BlockBackend *blk, unsigned long int req, void *buf,
1117                           BlockCompletionFunc *cb, void *opaque)
1118 {
1119     if (!blk_is_available(blk)) {
1120         return blk_abort_aio_request(blk, cb, opaque, -ENOMEDIUM);
1121     }
1122
1123     return bdrv_aio_ioctl(blk_bs(blk), req, buf, cb, opaque);
1124 }
1125
1126 int blk_co_discard(BlockBackend *blk, int64_t sector_num, int nb_sectors)
1127 {
1128     int ret = blk_check_request(blk, sector_num, nb_sectors);
1129     if (ret < 0) {
1130         return ret;
1131     }
1132
1133     return bdrv_co_discard(blk_bs(blk), sector_num, nb_sectors);
1134 }
1135
1136 int blk_co_flush(BlockBackend *blk)
1137 {
1138     if (!blk_is_available(blk)) {
1139         return -ENOMEDIUM;
1140     }
1141
1142     return bdrv_co_flush(blk_bs(blk));
1143 }
1144
1145 int blk_flush(BlockBackend *blk)
1146 {
1147     if (!blk_is_available(blk)) {
1148         return -ENOMEDIUM;
1149     }
1150
1151     return bdrv_flush(blk_bs(blk));
1152 }
1153
1154 void blk_drain(BlockBackend *blk)
1155 {
1156     if (blk_bs(blk)) {
1157         bdrv_drain(blk_bs(blk));
1158     }
1159 }
1160
1161 void blk_drain_all(void)
1162 {
1163     bdrv_drain_all();
1164 }
1165
1166 void blk_set_on_error(BlockBackend *blk, BlockdevOnError on_read_error,
1167                       BlockdevOnError on_write_error)
1168 {
1169     blk->on_read_error = on_read_error;
1170     blk->on_write_error = on_write_error;
1171 }
1172
1173 BlockdevOnError blk_get_on_error(BlockBackend *blk, bool is_read)
1174 {
1175     return is_read ? blk->on_read_error : blk->on_write_error;
1176 }
1177
1178 BlockErrorAction blk_get_error_action(BlockBackend *blk, bool is_read,
1179                                       int error)
1180 {
1181     BlockdevOnError on_err = blk_get_on_error(blk, is_read);
1182
1183     switch (on_err) {
1184     case BLOCKDEV_ON_ERROR_ENOSPC:
1185         return (error == ENOSPC) ?
1186                BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
1187     case BLOCKDEV_ON_ERROR_STOP:
1188         return BLOCK_ERROR_ACTION_STOP;
1189     case BLOCKDEV_ON_ERROR_REPORT:
1190         return BLOCK_ERROR_ACTION_REPORT;
1191     case BLOCKDEV_ON_ERROR_IGNORE:
1192         return BLOCK_ERROR_ACTION_IGNORE;
1193     default:
1194         abort();
1195     }
1196 }
1197
1198 static void send_qmp_error_event(BlockBackend *blk,
1199                                  BlockErrorAction action,
1200                                  bool is_read, int error)
1201 {
1202     IoOperationType optype;
1203
1204     optype = is_read ? IO_OPERATION_TYPE_READ : IO_OPERATION_TYPE_WRITE;
1205     qapi_event_send_block_io_error(blk_name(blk), optype, action,
1206                                    blk_iostatus_is_enabled(blk),
1207                                    error == ENOSPC, strerror(error),
1208                                    &error_abort);
1209 }
1210
1211 /* This is done by device models because, while the block layer knows
1212  * about the error, it does not know whether an operation comes from
1213  * the device or the block layer (from a job, for example).
1214  */
1215 void blk_error_action(BlockBackend *blk, BlockErrorAction action,
1216                       bool is_read, int error)
1217 {
1218     assert(error >= 0);
1219
1220     if (action == BLOCK_ERROR_ACTION_STOP) {
1221         /* First set the iostatus, so that "info block" returns an iostatus
1222          * that matches the events raised so far (an additional error iostatus
1223          * is fine, but not a lost one).
1224          */
1225         blk_iostatus_set_err(blk, error);
1226
1227         /* Then raise the request to stop the VM and the event.
1228          * qemu_system_vmstop_request_prepare has two effects.  First,
1229          * it ensures that the STOP event always comes after the
1230          * BLOCK_IO_ERROR event.  Second, it ensures that even if management
1231          * can observe the STOP event and do a "cont" before the STOP
1232          * event is issued, the VM will not stop.  In this case, vm_start()
1233          * also ensures that the STOP/RESUME pair of events is emitted.
1234          */
1235         qemu_system_vmstop_request_prepare();
1236         send_qmp_error_event(blk, action, is_read, error);
1237         qemu_system_vmstop_request(RUN_STATE_IO_ERROR);
1238     } else {
1239         send_qmp_error_event(blk, action, is_read, error);
1240     }
1241 }
1242
1243 int blk_is_read_only(BlockBackend *blk)
1244 {
1245     BlockDriverState *bs = blk_bs(blk);
1246
1247     if (bs) {
1248         return bdrv_is_read_only(bs);
1249     } else {
1250         return blk->root_state.read_only;
1251     }
1252 }
1253
1254 int blk_is_sg(BlockBackend *blk)
1255 {
1256     BlockDriverState *bs = blk_bs(blk);
1257
1258     if (!bs) {
1259         return 0;
1260     }
1261
1262     return bdrv_is_sg(bs);
1263 }
1264
1265 int blk_enable_write_cache(BlockBackend *blk)
1266 {
1267     return blk->enable_write_cache;
1268 }
1269
1270 void blk_set_enable_write_cache(BlockBackend *blk, bool wce)
1271 {
1272     blk->enable_write_cache = wce;
1273 }
1274
1275 void blk_invalidate_cache(BlockBackend *blk, Error **errp)
1276 {
1277     BlockDriverState *bs = blk_bs(blk);
1278
1279     if (!bs) {
1280         error_setg(errp, "Device '%s' has no medium", blk->name);
1281         return;
1282     }
1283
1284     bdrv_invalidate_cache(bs, errp);
1285 }
1286
1287 bool blk_is_inserted(BlockBackend *blk)
1288 {
1289     BlockDriverState *bs = blk_bs(blk);
1290
1291     return bs && bdrv_is_inserted(bs);
1292 }
1293
1294 bool blk_is_available(BlockBackend *blk)
1295 {
1296     return blk_is_inserted(blk) && !blk_dev_is_tray_open(blk);
1297 }
1298
1299 void blk_lock_medium(BlockBackend *blk, bool locked)
1300 {
1301     BlockDriverState *bs = blk_bs(blk);
1302
1303     if (bs) {
1304         bdrv_lock_medium(bs, locked);
1305     }
1306 }
1307
1308 void blk_eject(BlockBackend *blk, bool eject_flag)
1309 {
1310     BlockDriverState *bs = blk_bs(blk);
1311
1312     if (bs) {
1313         bdrv_eject(bs, eject_flag);
1314     }
1315 }
1316
1317 int blk_get_flags(BlockBackend *blk)
1318 {
1319     BlockDriverState *bs = blk_bs(blk);
1320
1321     if (bs) {
1322         return bdrv_get_flags(bs);
1323     } else {
1324         return blk->root_state.open_flags;
1325     }
1326 }
1327
1328 int blk_get_max_transfer_length(BlockBackend *blk)
1329 {
1330     BlockDriverState *bs = blk_bs(blk);
1331
1332     if (bs) {
1333         return bs->bl.max_transfer_length;
1334     } else {
1335         return 0;
1336     }
1337 }
1338
1339 int blk_get_max_iov(BlockBackend *blk)
1340 {
1341     return blk->root->bs->bl.max_iov;
1342 }
1343
1344 void blk_set_guest_block_size(BlockBackend *blk, int align)
1345 {
1346     blk->guest_block_size = align;
1347 }
1348
1349 void *blk_try_blockalign(BlockBackend *blk, size_t size)
1350 {
1351     return qemu_try_blockalign(blk ? blk_bs(blk) : NULL, size);
1352 }
1353
1354 void *blk_blockalign(BlockBackend *blk, size_t size)
1355 {
1356     return qemu_blockalign(blk ? blk_bs(blk) : NULL, size);
1357 }
1358
1359 bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp)
1360 {
1361     BlockDriverState *bs = blk_bs(blk);
1362
1363     if (!bs) {
1364         return false;
1365     }
1366
1367     return bdrv_op_is_blocked(bs, op, errp);
1368 }
1369
1370 void blk_op_unblock(BlockBackend *blk, BlockOpType op, Error *reason)
1371 {
1372     BlockDriverState *bs = blk_bs(blk);
1373
1374     if (bs) {
1375         bdrv_op_unblock(bs, op, reason);
1376     }
1377 }
1378
1379 void blk_op_block_all(BlockBackend *blk, Error *reason)
1380 {
1381     BlockDriverState *bs = blk_bs(blk);
1382
1383     if (bs) {
1384         bdrv_op_block_all(bs, reason);
1385     }
1386 }
1387
1388 void blk_op_unblock_all(BlockBackend *blk, Error *reason)
1389 {
1390     BlockDriverState *bs = blk_bs(blk);
1391
1392     if (bs) {
1393         bdrv_op_unblock_all(bs, reason);
1394     }
1395 }
1396
1397 AioContext *blk_get_aio_context(BlockBackend *blk)
1398 {
1399     BlockDriverState *bs = blk_bs(blk);
1400
1401     if (bs) {
1402         return bdrv_get_aio_context(bs);
1403     } else {
1404         return qemu_get_aio_context();
1405     }
1406 }
1407
1408 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb)
1409 {
1410     BlockBackendAIOCB *blk_acb = DO_UPCAST(BlockBackendAIOCB, common, acb);
1411     return blk_get_aio_context(blk_acb->blk);
1412 }
1413
1414 void blk_set_aio_context(BlockBackend *blk, AioContext *new_context)
1415 {
1416     BlockDriverState *bs = blk_bs(blk);
1417
1418     if (bs) {
1419         if (blk->public.throttle_state) {
1420             throttle_timers_detach_aio_context(&blk->public.throttle_timers);
1421         }
1422         bdrv_set_aio_context(bs, new_context);
1423         if (blk->public.throttle_state) {
1424             throttle_timers_attach_aio_context(&blk->public.throttle_timers,
1425                                                new_context);
1426         }
1427     }
1428 }
1429
1430 void blk_add_aio_context_notifier(BlockBackend *blk,
1431         void (*attached_aio_context)(AioContext *new_context, void *opaque),
1432         void (*detach_aio_context)(void *opaque), void *opaque)
1433 {
1434     BlockDriverState *bs = blk_bs(blk);
1435
1436     if (bs) {
1437         bdrv_add_aio_context_notifier(bs, attached_aio_context,
1438                                       detach_aio_context, opaque);
1439     }
1440 }
1441
1442 void blk_remove_aio_context_notifier(BlockBackend *blk,
1443                                      void (*attached_aio_context)(AioContext *,
1444                                                                   void *),
1445                                      void (*detach_aio_context)(void *),
1446                                      void *opaque)
1447 {
1448     BlockDriverState *bs = blk_bs(blk);
1449
1450     if (bs) {
1451         bdrv_remove_aio_context_notifier(bs, attached_aio_context,
1452                                          detach_aio_context, opaque);
1453     }
1454 }
1455
1456 void blk_add_remove_bs_notifier(BlockBackend *blk, Notifier *notify)
1457 {
1458     notifier_list_add(&blk->remove_bs_notifiers, notify);
1459 }
1460
1461 void blk_add_insert_bs_notifier(BlockBackend *blk, Notifier *notify)
1462 {
1463     notifier_list_add(&blk->insert_bs_notifiers, notify);
1464 }
1465
1466 void blk_io_plug(BlockBackend *blk)
1467 {
1468     BlockDriverState *bs = blk_bs(blk);
1469
1470     if (bs) {
1471         bdrv_io_plug(bs);
1472     }
1473 }
1474
1475 void blk_io_unplug(BlockBackend *blk)
1476 {
1477     BlockDriverState *bs = blk_bs(blk);
1478
1479     if (bs) {
1480         bdrv_io_unplug(bs);
1481     }
1482 }
1483
1484 BlockAcctStats *blk_get_stats(BlockBackend *blk)
1485 {
1486     return &blk->stats;
1487 }
1488
1489 void *blk_aio_get(const AIOCBInfo *aiocb_info, BlockBackend *blk,
1490                   BlockCompletionFunc *cb, void *opaque)
1491 {
1492     return qemu_aio_get(aiocb_info, blk_bs(blk), cb, opaque);
1493 }
1494
1495 int coroutine_fn blk_co_write_zeroes(BlockBackend *blk, int64_t offset,
1496                                      int count, BdrvRequestFlags flags)
1497 {
1498     return blk_co_pwritev(blk, offset, count, NULL,
1499                           flags | BDRV_REQ_ZERO_WRITE);
1500 }
1501
1502 int blk_write_compressed(BlockBackend *blk, int64_t sector_num,
1503                          const uint8_t *buf, int nb_sectors)
1504 {
1505     int ret = blk_check_request(blk, sector_num, nb_sectors);
1506     if (ret < 0) {
1507         return ret;
1508     }
1509
1510     return bdrv_write_compressed(blk_bs(blk), sector_num, buf, nb_sectors);
1511 }
1512
1513 int blk_truncate(BlockBackend *blk, int64_t offset)
1514 {
1515     if (!blk_is_available(blk)) {
1516         return -ENOMEDIUM;
1517     }
1518
1519     return bdrv_truncate(blk_bs(blk), offset);
1520 }
1521
1522 int blk_discard(BlockBackend *blk, int64_t sector_num, int nb_sectors)
1523 {
1524     int ret = blk_check_request(blk, sector_num, nb_sectors);
1525     if (ret < 0) {
1526         return ret;
1527     }
1528
1529     return bdrv_discard(blk_bs(blk), sector_num, nb_sectors);
1530 }
1531
1532 int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
1533                      int64_t pos, int size)
1534 {
1535     int ret;
1536
1537     if (!blk_is_available(blk)) {
1538         return -ENOMEDIUM;
1539     }
1540
1541     ret = bdrv_save_vmstate(blk_bs(blk), buf, pos, size);
1542     if (ret < 0) {
1543         return ret;
1544     }
1545
1546     if (ret == size && !blk->enable_write_cache) {
1547         ret = bdrv_flush(blk_bs(blk));
1548     }
1549
1550     return ret < 0 ? ret : size;
1551 }
1552
1553 int blk_load_vmstate(BlockBackend *blk, uint8_t *buf, int64_t pos, int size)
1554 {
1555     if (!blk_is_available(blk)) {
1556         return -ENOMEDIUM;
1557     }
1558
1559     return bdrv_load_vmstate(blk_bs(blk), buf, pos, size);
1560 }
1561
1562 int blk_probe_blocksizes(BlockBackend *blk, BlockSizes *bsz)
1563 {
1564     if (!blk_is_available(blk)) {
1565         return -ENOMEDIUM;
1566     }
1567
1568     return bdrv_probe_blocksizes(blk_bs(blk), bsz);
1569 }
1570
1571 int blk_probe_geometry(BlockBackend *blk, HDGeometry *geo)
1572 {
1573     if (!blk_is_available(blk)) {
1574         return -ENOMEDIUM;
1575     }
1576
1577     return bdrv_probe_geometry(blk_bs(blk), geo);
1578 }
1579
1580 /*
1581  * Updates the BlockBackendRootState object with data from the currently
1582  * attached BlockDriverState.
1583  */
1584 void blk_update_root_state(BlockBackend *blk)
1585 {
1586     assert(blk->root);
1587
1588     blk->root_state.open_flags    = blk->root->bs->open_flags;
1589     blk->root_state.read_only     = blk->root->bs->read_only;
1590     blk->root_state.detect_zeroes = blk->root->bs->detect_zeroes;
1591 }
1592
1593 /*
1594  * Applies the information in the root state to the given BlockDriverState. This
1595  * does not include the flags which have to be specified for bdrv_open(), use
1596  * blk_get_open_flags_from_root_state() to inquire them.
1597  */
1598 void blk_apply_root_state(BlockBackend *blk, BlockDriverState *bs)
1599 {
1600     bs->detect_zeroes = blk->root_state.detect_zeroes;
1601 }
1602
1603 /*
1604  * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
1605  * supposed to inherit the root state.
1606  */
1607 int blk_get_open_flags_from_root_state(BlockBackend *blk)
1608 {
1609     int bs_flags;
1610
1611     bs_flags = blk->root_state.read_only ? 0 : BDRV_O_RDWR;
1612     bs_flags |= blk->root_state.open_flags & ~BDRV_O_RDWR;
1613
1614     return bs_flags;
1615 }
1616
1617 BlockBackendRootState *blk_get_root_state(BlockBackend *blk)
1618 {
1619     return &blk->root_state;
1620 }
1621
1622 int blk_commit_all(void)
1623 {
1624     BlockBackend *blk = NULL;
1625
1626     while ((blk = blk_all_next(blk)) != NULL) {
1627         AioContext *aio_context = blk_get_aio_context(blk);
1628
1629         aio_context_acquire(aio_context);
1630         if (blk_is_inserted(blk) && blk->root->bs->backing) {
1631             int ret = bdrv_commit(blk->root->bs);
1632             if (ret < 0) {
1633                 aio_context_release(aio_context);
1634                 return ret;
1635             }
1636         }
1637         aio_context_release(aio_context);
1638     }
1639     return 0;
1640 }
1641
1642 int blk_flush_all(void)
1643 {
1644     BlockBackend *blk = NULL;
1645     int result = 0;
1646
1647     while ((blk = blk_all_next(blk)) != NULL) {
1648         AioContext *aio_context = blk_get_aio_context(blk);
1649         int ret;
1650
1651         aio_context_acquire(aio_context);
1652         if (blk_is_inserted(blk)) {
1653             ret = blk_flush(blk);
1654             if (ret < 0 && !result) {
1655                 result = ret;
1656             }
1657         }
1658         aio_context_release(aio_context);
1659     }
1660
1661     return result;
1662 }
1663
1664
1665 /* throttling disk I/O limits */
1666 void blk_set_io_limits(BlockBackend *blk, ThrottleConfig *cfg)
1667 {
1668     throttle_group_config(blk, cfg);
1669 }
1670
1671 void blk_io_limits_disable(BlockBackend *blk)
1672 {
1673     assert(blk->public.throttle_state);
1674     bdrv_drained_begin(blk_bs(blk));
1675     throttle_group_unregister_blk(blk);
1676     bdrv_drained_end(blk_bs(blk));
1677 }
1678
1679 /* should be called before blk_set_io_limits if a limit is set */
1680 void blk_io_limits_enable(BlockBackend *blk, const char *group)
1681 {
1682     assert(!blk->public.throttle_state);
1683     throttle_group_register_blk(blk, group);
1684 }
1685
1686 void blk_io_limits_update_group(BlockBackend *blk, const char *group)
1687 {
1688     /* this BB is not part of any group */
1689     if (!blk->public.throttle_state) {
1690         return;
1691     }
1692
1693     /* this BB is a part of the same group than the one we want */
1694     if (!g_strcmp0(throttle_group_get_name(blk), group)) {
1695         return;
1696     }
1697
1698     /* need to change the group this bs belong to */
1699     blk_io_limits_disable(blk);
1700     blk_io_limits_enable(blk, group);
1701 }
1702
1703 static void blk_root_drained_begin(BdrvChild *child)
1704 {
1705     BlockBackend *blk = child->opaque;
1706
1707     if (blk->public.io_limits_disabled++ == 0) {
1708         throttle_group_restart_blk(blk);
1709     }
1710 }
1711
1712 static void blk_root_drained_end(BdrvChild *child)
1713 {
1714     BlockBackend *blk = child->opaque;
1715
1716     assert(blk->public.io_limits_disabled);
1717     --blk->public.io_limits_disabled;
1718 }