block/block-backend.c

   1 /*
   2  * QEMU Block backends
   3  *
   4  * Copyright (C) 2014-2016 Red Hat, Inc.
   5  *
   6  * Authors:
   7  *  Markus Armbruster <armbru@redhat.com>,
   8  *
   9  * This work is licensed under the terms of the GNU LGPL, version 2.1
  10  * or later.  See the COPYING.LIB file in the top-level directory.
  11  */
  12
  13 #include "qemu/osdep.h"
  14 #include "sysemu/block-backend.h"
  15 #include "block/block_int.h"
  16 #include "block/blockjob.h"
  17 #include "block/throttle-groups.h"
  18 #include "sysemu/blockdev.h"
  19 #include "sysemu/sysemu.h"
  20 #include "qapi-event.h"
  21 #include "qemu/id.h"
  22 #include "trace.h"
  23
  24 /* Number of coroutines to reserve per attached device model */
  25 #define COROUTINE_POOL_RESERVATION 64
  26
  27 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
  28
  29 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb);
  30
  31 struct BlockBackend {
  32     char *name;
  33     int refcnt;
  34     BdrvChild *root;
  35     DriveInfo *legacy_dinfo;    /* null unless created by drive_new() */
  36     QTAILQ_ENTRY(BlockBackend) link;         /* for block_backends */
  37     QTAILQ_ENTRY(BlockBackend) monitor_link; /* for monitor_block_backends */
  38     BlockBackendPublic public;
  39
  40     void *dev;                  /* attached device model, if any */
  41     /* TODO change to DeviceState when all users are qdevified */
  42     const BlockDevOps *dev_ops;
  43     void *dev_opaque;
  44
  45     /* the block size for which the guest device expects atomicity */
  46     int guest_block_size;
  47
  48     /* If the BDS tree is removed, some of its options are stored here (which
  49      * can be used to restore those options in the new BDS on insert) */
  50     BlockBackendRootState root_state;
  51
  52     bool enable_write_cache;
  53
  54     /* I/O stats (display with "info blockstats"). */
  55     BlockAcctStats stats;
  56
  57     BlockdevOnError on_read_error, on_write_error;
  58     bool iostatus_enabled;
  59     BlockDeviceIoStatus iostatus;
  60
  61     bool allow_write_beyond_eof;
  62
  63     NotifierList remove_bs_notifiers, insert_bs_notifiers;
  64 };
  65
  66 typedef struct BlockBackendAIOCB {
  67     BlockAIOCB common;
  68     QEMUBH *bh;
  69     BlockBackend *blk;
  70     int ret;
  71 } BlockBackendAIOCB;
  72
  73 static const AIOCBInfo block_backend_aiocb_info = {
  74     .get_aio_context = blk_aiocb_get_aio_context,
  75     .aiocb_size = sizeof(BlockBackendAIOCB),
  76 };
  77
  78 static void drive_info_del(DriveInfo *dinfo);
  79 static BlockBackend *bdrv_first_blk(BlockDriverState *bs);
  80
  81 /* All BlockBackends */
  82 static QTAILQ_HEAD(, BlockBackend) block_backends =
  83     QTAILQ_HEAD_INITIALIZER(block_backends);
  84
  85 /* All BlockBackends referenced by the monitor and which are iterated through by
  86  * blk_next() */
  87 static QTAILQ_HEAD(, BlockBackend) monitor_block_backends =
  88     QTAILQ_HEAD_INITIALIZER(monitor_block_backends);
  89
  90 static void blk_root_inherit_options(int *child_flags, QDict *child_options,
  91                                      int parent_flags, QDict *parent_options)
  92 {
  93     /* We're not supposed to call this function for root nodes */
  94     abort();
  95 }
  96 static void blk_root_drained_begin(BdrvChild *child);
  97 static void blk_root_drained_end(BdrvChild *child);
  98
  99 static void blk_root_change_media(BdrvChild *child, bool load);
 100 static void blk_root_resize(BdrvChild *child);
 101
 102 static const char *blk_root_get_name(BdrvChild *child)
 103 {
 104     return blk_name(child->opaque);
 105 }
 106
 107 static const BdrvChildRole child_root = {
 108     .inherit_options    = blk_root_inherit_options,
 109
 110     .change_media       = blk_root_change_media,
 111     .resize             = blk_root_resize,
 112     .get_name           = blk_root_get_name,
 113
 114     .drained_begin      = blk_root_drained_begin,
 115     .drained_end        = blk_root_drained_end,
 116 };
 117
 118 /*
 119  * Create a new BlockBackend with a reference count of one.
 120  * Store an error through @errp on failure, unless it's null.
 121  * Return the new BlockBackend on success, null on failure.
 122  */
 123 BlockBackend *blk_new(void)
 124 {
 125     BlockBackend *blk;
 126
 127     blk = g_new0(BlockBackend, 1);
 128     blk->refcnt = 1;
 129     blk_set_enable_write_cache(blk, true);
 130
 131     qemu_co_queue_init(&blk->public.throttled_reqs[0]);
 132     qemu_co_queue_init(&blk->public.throttled_reqs[1]);
 133
 134     notifier_list_init(&blk->remove_bs_notifiers);
 135     notifier_list_init(&blk->insert_bs_notifiers);
 136
 137     QTAILQ_INSERT_TAIL(&block_backends, blk, link);
 138     return blk;
 139 }
 140
 141 /*
 142  * Creates a new BlockBackend, opens a new BlockDriverState, and connects both.
 143  *
 144  * Just as with bdrv_open(), after having called this function the reference to
 145  * @options belongs to the block layer (even on failure).
 146  *
 147  * TODO: Remove @filename and @flags; it should be possible to specify a whole
 148  * BDS tree just by specifying the @options QDict (or @reference,
 149  * alternatively). At the time of adding this function, this is not possible,
 150  * though, so callers of this function have to be able to specify @filename and
 151  * @flags.
 152  */
 153 BlockBackend *blk_new_open(const char *filename, const char *reference,
 154                            QDict *options, int flags, Error **errp)
 155 {
 156     BlockBackend *blk;
 157     BlockDriverState *bs;
 158
 159     blk = blk_new();
 160     bs = bdrv_open(filename, reference, options, flags, errp);
 161     if (!bs) {
 162         blk_unref(blk);
 163         return NULL;
 164     }
 165
 166     blk->root = bdrv_root_attach_child(bs, "root", &child_root, blk);
 167
 168     return blk;
 169 }
 170
 171 static void blk_delete(BlockBackend *blk)
 172 {
 173     assert(!blk->refcnt);
 174     assert(!blk->name);
 175     assert(!blk->dev);
 176     if (blk->root) {
 177         blk_remove_bs(blk);
 178     }
 179     assert(QLIST_EMPTY(&blk->remove_bs_notifiers.notifiers));
 180     assert(QLIST_EMPTY(&blk->insert_bs_notifiers.notifiers));
 181     QTAILQ_REMOVE(&block_backends, blk, link);
 182     drive_info_del(blk->legacy_dinfo);
 183     block_acct_cleanup(&blk->stats);
 184     g_free(blk);
 185 }
 186
 187 static void drive_info_del(DriveInfo *dinfo)
 188 {
 189     if (!dinfo) {
 190         return;
 191     }
 192     qemu_opts_del(dinfo->opts);
 193     g_free(dinfo->serial);
 194     g_free(dinfo);
 195 }
 196
 197 int blk_get_refcnt(BlockBackend *blk)
 198 {
 199     return blk ? blk->refcnt : 0;
 200 }
 201
 202 /*
 203  * Increment @blk's reference count.
 204  * @blk must not be null.
 205  */
 206 void blk_ref(BlockBackend *blk)
 207 {
 208     blk->refcnt++;
 209 }
 210
 211 /*
 212  * Decrement @blk's reference count.
 213  * If this drops it to zero, destroy @blk.
 214  * For convenience, do nothing if @blk is null.
 215  */
 216 void blk_unref(BlockBackend *blk)
 217 {
 218     if (blk) {
 219         assert(blk->refcnt > 0);
 220         if (!--blk->refcnt) {
 221             blk_delete(blk);
 222         }
 223     }
 224 }
 225
 226 /*
 227  * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
 228  * ones which are hidden (i.e. are not referenced by the monitor).
 229  */
 230 static BlockBackend *blk_all_next(BlockBackend *blk)
 231 {
 232     return blk ? QTAILQ_NEXT(blk, link)
 233                : QTAILQ_FIRST(&block_backends);
 234 }
 235
 236 void blk_remove_all_bs(void)
 237 {
 238     BlockBackend *blk = NULL;
 239
 240     while ((blk = blk_all_next(blk)) != NULL) {
 241         AioContext *ctx = blk_get_aio_context(blk);
 242
 243         aio_context_acquire(ctx);
 244         if (blk->root) {
 245             blk_remove_bs(blk);
 246         }
 247         aio_context_release(ctx);
 248     }
 249 }
 250
 251 /*
 252  * Return the monitor-owned BlockBackend after @blk.
 253  * If @blk is null, return the first one.
 254  * Else, return @blk's next sibling, which may be null.
 255  *
 256  * To iterate over all BlockBackends, do
 257  * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
 258  *     ...
 259  * }
 260  */
 261 BlockBackend *blk_next(BlockBackend *blk)
 262 {
 263     return blk ? QTAILQ_NEXT(blk, monitor_link)
 264                : QTAILQ_FIRST(&monitor_block_backends);
 265 }
 266
 267 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
 268  * the monitor or attached to a BlockBackend */
 269 BlockDriverState *bdrv_next(BdrvNextIterator *it)
 270 {
 271     BlockDriverState *bs;
 272
 273     /* First, return all root nodes of BlockBackends. In order to avoid
 274      * returning a BDS twice when multiple BBs refer to it, we only return it
 275      * if the BB is the first one in the parent list of the BDS. */
 276     if (it->phase == BDRV_NEXT_BACKEND_ROOTS) {
 277         do {
 278             it->blk = blk_all_next(it->blk);
 279             bs = it->blk ? blk_bs(it->blk) : NULL;
 280         } while (it->blk && (bs == NULL || bdrv_first_blk(bs) != it->blk));
 281
 282         if (bs) {
 283             return bs;
 284         }
 285         it->phase = BDRV_NEXT_MONITOR_OWNED;
 286     }
 287
 288     /* Then return the monitor-owned BDSes without a BB attached. Ignore all
 289      * BDSes that are attached to a BlockBackend here; they have been handled
 290      * by the above block already */
 291     do {
 292         it->bs = bdrv_next_monitor_owned(it->bs);
 293         bs = it->bs;
 294     } while (bs && bdrv_has_blk(bs));
 295
 296     return bs;
 297 }
 298
 299 BlockDriverState *bdrv_first(BdrvNextIterator *it)
 300 {
 301     *it = (BdrvNextIterator) {
 302         .phase = BDRV_NEXT_BACKEND_ROOTS,
 303     };
 304
 305     return bdrv_next(it);
 306 }
 307
 308 /*
 309  * Add a BlockBackend into the list of backends referenced by the monitor, with
 310  * the given @name acting as the handle for the monitor.
 311  * Strictly for use by blockdev.c.
 312  *
 313  * @name must not be null or empty.
 314  *
 315  * Returns true on success and false on failure. In the latter case, an Error
 316  * object is returned through @errp.
 317  */
 318 bool monitor_add_blk(BlockBackend *blk, const char *name, Error **errp)
 319 {
 320     assert(!blk->name);
 321     assert(name && name[0]);
 322
 323     if (!id_wellformed(name)) {
 324         error_setg(errp, "Invalid device name");
 325         return false;
 326     }
 327     if (blk_by_name(name)) {
 328         error_setg(errp, "Device with id '%s' already exists", name);
 329         return false;
 330     }
 331     if (bdrv_find_node(name)) {
 332         error_setg(errp,
 333                    "Device name '%s' conflicts with an existing node name",
 334                    name);
 335         return false;
 336     }
 337
 338     blk->name = g_strdup(name);
 339     QTAILQ_INSERT_TAIL(&monitor_block_backends, blk, monitor_link);
 340     return true;
 341 }
 342
 343 /*
 344  * Remove a BlockBackend from the list of backends referenced by the monitor.
 345  * Strictly for use by blockdev.c.
 346  */
 347 void monitor_remove_blk(BlockBackend *blk)
 348 {
 349     if (!blk->name) {
 350         return;
 351     }
 352
 353     QTAILQ_REMOVE(&monitor_block_backends, blk, monitor_link);
 354     g_free(blk->name);
 355     blk->name = NULL;
 356 }
 357
 358 /*
 359  * Return @blk's name, a non-null string.
 360  * Returns an empty string iff @blk is not referenced by the monitor.
 361  */
 362 const char *blk_name(BlockBackend *blk)
 363 {
 364     return blk->name ?: "";
 365 }
 366
 367 /*
 368  * Return the BlockBackend with name @name if it exists, else null.
 369  * @name must not be null.
 370  */
 371 BlockBackend *blk_by_name(const char *name)
 372 {
 373     BlockBackend *blk = NULL;
 374
 375     assert(name);
 376     while ((blk = blk_next(blk)) != NULL) {
 377         if (!strcmp(name, blk->name)) {
 378             return blk;
 379         }
 380     }
 381     return NULL;
 382 }
 383
 384 /*
 385  * Return the BlockDriverState attached to @blk if any, else null.
 386  */
 387 BlockDriverState *blk_bs(BlockBackend *blk)
 388 {
 389     return blk->root ? blk->root->bs : NULL;
 390 }
 391
 392 static BlockBackend *bdrv_first_blk(BlockDriverState *bs)
 393 {
 394     BdrvChild *child;
 395     QLIST_FOREACH(child, &bs->parents, next_parent) {
 396         if (child->role == &child_root) {
 397             return child->opaque;
 398         }
 399     }
 400
 401     return NULL;
 402 }
 403
 404 /*
 405  * Returns true if @bs has an associated BlockBackend.
 406  */
 407 bool bdrv_has_blk(BlockDriverState *bs)
 408 {
 409     return bdrv_first_blk(bs) != NULL;
 410 }
 411
 412 /*
 413  * Returns true if @bs has only BlockBackends as parents.
 414  */
 415 bool bdrv_is_root_node(BlockDriverState *bs)
 416 {
 417     BdrvChild *c;
 418
 419     QLIST_FOREACH(c, &bs->parents, next_parent) {
 420         if (c->role != &child_root) {
 421             return false;
 422         }
 423     }
 424
 425     return true;
 426 }
 427
 428 /*
 429  * Return @blk's DriveInfo if any, else null.
 430  */
 431 DriveInfo *blk_legacy_dinfo(BlockBackend *blk)
 432 {
 433     return blk->legacy_dinfo;
 434 }
 435
 436 /*
 437  * Set @blk's DriveInfo to @dinfo, and return it.
 438  * @blk must not have a DriveInfo set already.
 439  * No other BlockBackend may have the same DriveInfo set.
 440  */
 441 DriveInfo *blk_set_legacy_dinfo(BlockBackend *blk, DriveInfo *dinfo)
 442 {
 443     assert(!blk->legacy_dinfo);
 444     return blk->legacy_dinfo = dinfo;
 445 }
 446
 447 /*
 448  * Return the BlockBackend with DriveInfo @dinfo.
 449  * It must exist.
 450  */
 451 BlockBackend *blk_by_legacy_dinfo(DriveInfo *dinfo)
 452 {
 453     BlockBackend *blk = NULL;
 454
 455     while ((blk = blk_next(blk)) != NULL) {
 456         if (blk->legacy_dinfo == dinfo) {
 457             return blk;
 458         }
 459     }
 460     abort();
 461 }
 462
 463 /*
 464  * Returns a pointer to the publicly accessible fields of @blk.
 465  */
 466 BlockBackendPublic *blk_get_public(BlockBackend *blk)
 467 {
 468     return &blk->public;
 469 }
 470
 471 /*
 472  * Returns a BlockBackend given the associated @public fields.
 473  */
 474 BlockBackend *blk_by_public(BlockBackendPublic *public)
 475 {
 476     return container_of(public, BlockBackend, public);
 477 }
 478
 479 /*
 480  * Disassociates the currently associated BlockDriverState from @blk.
 481  */
 482 void blk_remove_bs(BlockBackend *blk)
 483 {
 484     notifier_list_notify(&blk->remove_bs_notifiers, blk);
 485     if (blk->public.throttle_state) {
 486         throttle_timers_detach_aio_context(&blk->public.throttle_timers);
 487     }
 488
 489     blk_update_root_state(blk);
 490
 491     bdrv_root_unref_child(blk->root);
 492     blk->root = NULL;
 493 }
 494
 495 /*
 496  * Associates a new BlockDriverState with @blk.
 497  */
 498 void blk_insert_bs(BlockBackend *blk, BlockDriverState *bs)
 499 {
 500     bdrv_ref(bs);
 501     blk->root = bdrv_root_attach_child(bs, "root", &child_root, blk);
 502
 503     notifier_list_notify(&blk->insert_bs_notifiers, blk);
 504     if (blk->public.throttle_state) {
 505         throttle_timers_attach_aio_context(
 506             &blk->public.throttle_timers, bdrv_get_aio_context(bs));
 507     }
 508 }
 509
 510 /*
 511  * Attach device model @dev to @blk.
 512  * Return 0 on success, -EBUSY when a device model is attached already.
 513  */
 514 int blk_attach_dev(BlockBackend *blk, void *dev)
 515 /* TODO change to DeviceState *dev when all users are qdevified */
 516 {
 517     if (blk->dev) {
 518         return -EBUSY;
 519     }
 520     blk_ref(blk);
 521     blk->dev = dev;
 522     blk_iostatus_reset(blk);
 523     return 0;
 524 }
 525
 526 /*
 527  * Attach device model @dev to @blk.
 528  * @blk must not have a device model attached already.
 529  * TODO qdevified devices don't use this, remove when devices are qdevified
 530  */
 531 void blk_attach_dev_nofail(BlockBackend *blk, void *dev)
 532 {
 533     if (blk_attach_dev(blk, dev) < 0) {
 534         abort();
 535     }
 536 }
 537
 538 /*
 539  * Detach device model @dev from @blk.
 540  * @dev must be currently attached to @blk.
 541  */
 542 void blk_detach_dev(BlockBackend *blk, void *dev)
 543 /* TODO change to DeviceState *dev when all users are qdevified */
 544 {
 545     assert(blk->dev == dev);
 546     blk->dev = NULL;
 547     blk->dev_ops = NULL;
 548     blk->dev_opaque = NULL;
 549     blk->guest_block_size = 512;
 550     blk_unref(blk);
 551 }
 552
 553 /*
 554  * Return the device model attached to @blk if any, else null.
 555  */
 556 void *blk_get_attached_dev(BlockBackend *blk)
 557 /* TODO change to return DeviceState * when all users are qdevified */
 558 {
 559     return blk->dev;
 560 }
 561
 562 /*
 563  * Return the BlockBackend which has the device model @dev attached if it
 564  * exists, else null.
 565  *
 566  * @dev must not be null.
 567  */
 568 BlockBackend *blk_by_dev(void *dev)
 569 {
 570     BlockBackend *blk = NULL;
 571
 572     assert(dev != NULL);
 573     while ((blk = blk_all_next(blk)) != NULL) {
 574         if (blk->dev == dev) {
 575             return blk;
 576         }
 577     }
 578     return NULL;
 579 }
 580
 581 /*
 582  * Set @blk's device model callbacks to @ops.
 583  * @opaque is the opaque argument to pass to the callbacks.
 584  * This is for use by device models.
 585  */
 586 void blk_set_dev_ops(BlockBackend *blk, const BlockDevOps *ops,
 587                      void *opaque)
 588 {
 589     blk->dev_ops = ops;
 590     blk->dev_opaque = opaque;
 591 }
 592
 593 /*
 594  * Notify @blk's attached device model of media change.
 595  * If @load is true, notify of media load.
 596  * Else, notify of media eject.
 597  * Also send DEVICE_TRAY_MOVED events as appropriate.
 598  */
 599 void blk_dev_change_media_cb(BlockBackend *blk, bool load)
 600 {
 601     if (blk->dev_ops && blk->dev_ops->change_media_cb) {
 602         bool tray_was_open, tray_is_open;
 603
 604         tray_was_open = blk_dev_is_tray_open(blk);
 605         blk->dev_ops->change_media_cb(blk->dev_opaque, load);
 606         tray_is_open = blk_dev_is_tray_open(blk);
 607
 608         if (tray_was_open != tray_is_open) {
 609             qapi_event_send_device_tray_moved(blk_name(blk), tray_is_open,
 610                                               &error_abort);
 611         }
 612     }
 613 }
 614
 615 static void blk_root_change_media(BdrvChild *child, bool load)
 616 {
 617     blk_dev_change_media_cb(child->opaque, load);
 618 }
 619
 620 /*
 621  * Does @blk's attached device model have removable media?
 622  * %true if no device model is attached.
 623  */
 624 bool blk_dev_has_removable_media(BlockBackend *blk)
 625 {
 626     return !blk->dev || (blk->dev_ops && blk->dev_ops->change_media_cb);
 627 }
 628
 629 /*
 630  * Does @blk's attached device model have a tray?
 631  */
 632 bool blk_dev_has_tray(BlockBackend *blk)
 633 {
 634     return blk->dev_ops && blk->dev_ops->is_tray_open;
 635 }
 636
 637 /*
 638  * Notify @blk's attached device model of a media eject request.
 639  * If @force is true, the medium is about to be yanked out forcefully.
 640  */
 641 void blk_dev_eject_request(BlockBackend *blk, bool force)
 642 {
 643     if (blk->dev_ops && blk->dev_ops->eject_request_cb) {
 644         blk->dev_ops->eject_request_cb(blk->dev_opaque, force);
 645     }
 646 }
 647
 648 /*
 649  * Does @blk's attached device model have a tray, and is it open?
 650  */
 651 bool blk_dev_is_tray_open(BlockBackend *blk)
 652 {
 653     if (blk_dev_has_tray(blk)) {
 654         return blk->dev_ops->is_tray_open(blk->dev_opaque);
 655     }
 656     return false;
 657 }
 658
 659 /*
 660  * Does @blk's attached device model have the medium locked?
 661  * %false if the device model has no such lock.
 662  */
 663 bool blk_dev_is_medium_locked(BlockBackend *blk)
 664 {
 665     if (blk->dev_ops && blk->dev_ops->is_medium_locked) {
 666         return blk->dev_ops->is_medium_locked(blk->dev_opaque);
 667     }
 668     return false;
 669 }
 670
 671 /*
 672  * Notify @blk's attached device model of a backend size change.
 673  */
 674 static void blk_root_resize(BdrvChild *child)
 675 {
 676     BlockBackend *blk = child->opaque;
 677
 678     if (blk->dev_ops && blk->dev_ops->resize_cb) {
 679         blk->dev_ops->resize_cb(blk->dev_opaque);
 680     }
 681 }
 682
 683 void blk_iostatus_enable(BlockBackend *blk)
 684 {
 685     blk->iostatus_enabled = true;
 686     blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 687 }
 688
 689 /* The I/O status is only enabled if the drive explicitly
 690  * enables it _and_ the VM is configured to stop on errors */
 691 bool blk_iostatus_is_enabled(const BlockBackend *blk)
 692 {
 693     return (blk->iostatus_enabled &&
 694            (blk->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
 695             blk->on_write_error == BLOCKDEV_ON_ERROR_STOP   ||
 696             blk->on_read_error == BLOCKDEV_ON_ERROR_STOP));
 697 }
 698
 699 BlockDeviceIoStatus blk_iostatus(const BlockBackend *blk)
 700 {
 701     return blk->iostatus;
 702 }
 703
 704 void blk_iostatus_disable(BlockBackend *blk)
 705 {
 706     blk->iostatus_enabled = false;
 707 }
 708
 709 void blk_iostatus_reset(BlockBackend *blk)
 710 {
 711     if (blk_iostatus_is_enabled(blk)) {
 712         BlockDriverState *bs = blk_bs(blk);
 713         blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 714         if (bs && bs->job) {
 715             block_job_iostatus_reset(bs->job);
 716         }
 717     }
 718 }
 719
 720 void blk_iostatus_set_err(BlockBackend *blk, int error)
 721 {
 722     assert(blk_iostatus_is_enabled(blk));
 723     if (blk->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 724         blk->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 725                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 726     }
 727 }
 728
 729 void blk_set_allow_write_beyond_eof(BlockBackend *blk, bool allow)
 730 {
 731     blk->allow_write_beyond_eof = allow;
 732 }
 733
 734 static int blk_check_byte_request(BlockBackend *blk, int64_t offset,
 735                                   size_t size)
 736 {
 737     int64_t len;
 738
 739     if (size > INT_MAX) {
 740         return -EIO;
 741     }
 742
 743     if (!blk_is_available(blk)) {
 744         return -ENOMEDIUM;
 745     }
 746
 747     if (offset < 0) {
 748         return -EIO;
 749     }
 750
 751     if (!blk->allow_write_beyond_eof) {
 752         len = blk_getlength(blk);
 753         if (len < 0) {
 754             return len;
 755         }
 756
 757         if (offset > len || len - offset < size) {
 758             return -EIO;
 759         }
 760     }
 761
 762     return 0;
 763 }
 764
 765 int coroutine_fn blk_co_preadv(BlockBackend *blk, int64_t offset,
 766                                unsigned int bytes, QEMUIOVector *qiov,
 767                                BdrvRequestFlags flags)
 768 {
 769     int ret;
 770
 771     trace_blk_co_preadv(blk, blk_bs(blk), offset, bytes, flags);
 772
 773     ret = blk_check_byte_request(blk, offset, bytes);
 774     if (ret < 0) {
 775         return ret;
 776     }
 777
 778     /* throttling disk I/O */
 779     if (blk->public.throttle_state) {
 780         throttle_group_co_io_limits_intercept(blk, bytes, false);
 781     }
 782
 783     return bdrv_co_preadv(blk->root, offset, bytes, qiov, flags);
 784 }
 785
 786 int coroutine_fn blk_co_pwritev(BlockBackend *blk, int64_t offset,
 787                                 unsigned int bytes, QEMUIOVector *qiov,
 788                                 BdrvRequestFlags flags)
 789 {
 790     int ret;
 791
 792     trace_blk_co_pwritev(blk, blk_bs(blk), offset, bytes, flags);
 793
 794     ret = blk_check_byte_request(blk, offset, bytes);
 795     if (ret < 0) {
 796         return ret;
 797     }
 798
 799     /* throttling disk I/O */
 800     if (blk->public.throttle_state) {
 801         throttle_group_co_io_limits_intercept(blk, bytes, true);
 802     }
 803
 804     if (!blk->enable_write_cache) {
 805         flags |= BDRV_REQ_FUA;
 806     }
 807
 808     return bdrv_co_pwritev(blk->root, offset, bytes, qiov, flags);
 809 }
 810
 811 typedef struct BlkRwCo {
 812     BlockBackend *blk;
 813     int64_t offset;
 814     QEMUIOVector *qiov;
 815     int ret;
 816     BdrvRequestFlags flags;
 817 } BlkRwCo;
 818
 819 static void blk_read_entry(void *opaque)
 820 {
 821     BlkRwCo *rwco = opaque;
 822
 823     rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, rwco->qiov->size,
 824                               rwco->qiov, rwco->flags);
 825 }
 826
 827 static void blk_write_entry(void *opaque)
 828 {
 829     BlkRwCo *rwco = opaque;
 830
 831     rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, rwco->qiov->size,
 832                                rwco->qiov, rwco->flags);
 833 }
 834
 835 static int blk_prw(BlockBackend *blk, int64_t offset, uint8_t *buf,
 836                    int64_t bytes, CoroutineEntry co_entry,
 837                    BdrvRequestFlags flags)
 838 {
 839     AioContext *aio_context;
 840     QEMUIOVector qiov;
 841     struct iovec iov;
 842     Coroutine *co;
 843     BlkRwCo rwco;
 844
 845     iov = (struct iovec) {
 846         .iov_base = buf,
 847         .iov_len = bytes,
 848     };
 849     qemu_iovec_init_external(&qiov, &iov, 1);
 850
 851     rwco = (BlkRwCo) {
 852         .blk    = blk,
 853         .offset = offset,
 854         .qiov   = &qiov,
 855         .flags  = flags,
 856         .ret    = NOT_DONE,
 857     };
 858
 859     co = qemu_coroutine_create(co_entry, &rwco);
 860     qemu_coroutine_enter(co);
 861
 862     aio_context = blk_get_aio_context(blk);
 863     while (rwco.ret == NOT_DONE) {
 864         aio_poll(aio_context, true);
 865     }
 866
 867     return rwco.ret;
 868 }
 869
 870 int blk_pread_unthrottled(BlockBackend *blk, int64_t offset, uint8_t *buf,
 871                           int count)
 872 {
 873     int ret;
 874
 875     ret = blk_check_byte_request(blk, offset, count);
 876     if (ret < 0) {
 877         return ret;
 878     }
 879
 880     blk_root_drained_begin(blk->root);
 881     ret = blk_pread(blk, offset, buf, count);
 882     blk_root_drained_end(blk->root);
 883     return ret;
 884 }
 885
 886 int blk_pwrite_zeroes(BlockBackend *blk, int64_t offset,
 887                       int count, BdrvRequestFlags flags)
 888 {
 889     return blk_prw(blk, offset, NULL, count, blk_write_entry,
 890                    flags | BDRV_REQ_ZERO_WRITE);
 891 }
 892
 893 int blk_make_zero(BlockBackend *blk, BdrvRequestFlags flags)
 894 {
 895     return bdrv_make_zero(blk->root, flags);
 896 }
 897
 898 static void error_callback_bh(void *opaque)
 899 {
 900     struct BlockBackendAIOCB *acb = opaque;
 901     qemu_bh_delete(acb->bh);
 902     acb->common.cb(acb->common.opaque, acb->ret);
 903     qemu_aio_unref(acb);
 904 }
 905
 906 BlockAIOCB *blk_abort_aio_request(BlockBackend *blk,
 907                                   BlockCompletionFunc *cb,
 908                                   void *opaque, int ret)
 909 {
 910     struct BlockBackendAIOCB *acb;
 911     QEMUBH *bh;
 912
 913     acb = blk_aio_get(&block_backend_aiocb_info, blk, cb, opaque);
 914     acb->blk = blk;
 915     acb->ret = ret;
 916
 917     bh = aio_bh_new(blk_get_aio_context(blk), error_callback_bh, acb);
 918     acb->bh = bh;
 919     qemu_bh_schedule(bh);
 920
 921     return &acb->common;
 922 }
 923
 924 typedef struct BlkAioEmAIOCB {
 925     BlockAIOCB common;
 926     BlkRwCo rwco;
 927     int bytes;
 928     bool has_returned;
 929     QEMUBH* bh;
 930 } BlkAioEmAIOCB;
 931
 932 static const AIOCBInfo blk_aio_em_aiocb_info = {
 933     .aiocb_size         = sizeof(BlkAioEmAIOCB),
 934 };
 935
 936 static void blk_aio_complete(BlkAioEmAIOCB *acb)
 937 {
 938     if (acb->bh) {
 939         assert(acb->has_returned);
 940         qemu_bh_delete(acb->bh);
 941     }
 942     if (acb->has_returned) {
 943         acb->common.cb(acb->common.opaque, acb->rwco.ret);
 944         qemu_aio_unref(acb);
 945     }
 946 }
 947
 948 static void blk_aio_complete_bh(void *opaque)
 949 {
 950     blk_aio_complete(opaque);
 951 }
 952
 953 static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset, int bytes,
 954                                 QEMUIOVector *qiov, CoroutineEntry co_entry,
 955                                 BdrvRequestFlags flags,
 956                                 BlockCompletionFunc *cb, void *opaque)
 957 {
 958     BlkAioEmAIOCB *acb;
 959     Coroutine *co;
 960
 961     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
 962     acb->rwco = (BlkRwCo) {
 963         .blk    = blk,
 964         .offset = offset,
 965         .qiov   = qiov,
 966         .flags  = flags,
 967         .ret    = NOT_DONE,
 968     };
 969     acb->bytes = bytes;
 970     acb->bh = NULL;
 971     acb->has_returned = false;
 972
 973     co = qemu_coroutine_create(co_entry, acb);
 974     qemu_coroutine_enter(co);
 975
 976     acb->has_returned = true;
 977     if (acb->rwco.ret != NOT_DONE) {
 978         acb->bh = aio_bh_new(blk_get_aio_context(blk), blk_aio_complete_bh, acb);
 979         qemu_bh_schedule(acb->bh);
 980     }
 981
 982     return &acb->common;
 983 }
 984
 985 static void blk_aio_read_entry(void *opaque)
 986 {
 987     BlkAioEmAIOCB *acb = opaque;
 988     BlkRwCo *rwco = &acb->rwco;
 989
 990     assert(rwco->qiov->size == acb->bytes);
 991     rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, acb->bytes,
 992                               rwco->qiov, rwco->flags);
 993     blk_aio_complete(acb);
 994 }
 995
 996 static void blk_aio_write_entry(void *opaque)
 997 {
 998     BlkAioEmAIOCB *acb = opaque;
 999     BlkRwCo *rwco = &acb->rwco;
1000
1001     assert(!rwco->qiov || rwco->qiov->size == acb->bytes);
1002     rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, acb->bytes,
1003                                rwco->qiov, rwco->flags);
1004     blk_aio_complete(acb);
1005 }
1006
1007 BlockAIOCB *blk_aio_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1008                                   int count, BdrvRequestFlags flags,
1009                                   BlockCompletionFunc *cb, void *opaque)
1010 {
1011     return blk_aio_prwv(blk, offset, count, NULL, blk_aio_write_entry,
1012                         flags | BDRV_REQ_ZERO_WRITE, cb, opaque);
1013 }
1014
1015 int blk_pread(BlockBackend *blk, int64_t offset, void *buf, int count)
1016 {
1017     int ret = blk_prw(blk, offset, buf, count, blk_read_entry, 0);
1018     if (ret < 0) {
1019         return ret;
1020     }
1021     return count;
1022 }
1023
1024 int blk_pwrite(BlockBackend *blk, int64_t offset, const void *buf, int count,
1025                BdrvRequestFlags flags)
1026 {
1027     int ret = blk_prw(blk, offset, (void *) buf, count, blk_write_entry,
1028                       flags);
1029     if (ret < 0) {
1030         return ret;
1031     }
1032     return count;
1033 }
1034
1035 int64_t blk_getlength(BlockBackend *blk)
1036 {
1037     if (!blk_is_available(blk)) {
1038         return -ENOMEDIUM;
1039     }
1040
1041     return bdrv_getlength(blk_bs(blk));
1042 }
1043
1044 void blk_get_geometry(BlockBackend *blk, uint64_t *nb_sectors_ptr)
1045 {
1046     if (!blk_bs(blk)) {
1047         *nb_sectors_ptr = 0;
1048     } else {
1049         bdrv_get_geometry(blk_bs(blk), nb_sectors_ptr);
1050     }
1051 }
1052
1053 int64_t blk_nb_sectors(BlockBackend *blk)
1054 {
1055     if (!blk_is_available(blk)) {
1056         return -ENOMEDIUM;
1057     }
1058
1059     return bdrv_nb_sectors(blk_bs(blk));
1060 }
1061
1062 BlockAIOCB *blk_aio_preadv(BlockBackend *blk, int64_t offset,
1063                            QEMUIOVector *qiov, BdrvRequestFlags flags,
1064                            BlockCompletionFunc *cb, void *opaque)
1065 {
1066     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1067                         blk_aio_read_entry, flags, cb, opaque);
1068 }
1069
1070 BlockAIOCB *blk_aio_pwritev(BlockBackend *blk, int64_t offset,
1071                             QEMUIOVector *qiov, BdrvRequestFlags flags,
1072                             BlockCompletionFunc *cb, void *opaque)
1073 {
1074     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1075                         blk_aio_write_entry, flags, cb, opaque);
1076 }
1077
1078 BlockAIOCB *blk_aio_flush(BlockBackend *blk,
1079                           BlockCompletionFunc *cb, void *opaque)
1080 {
1081     if (!blk_is_available(blk)) {
1082         return blk_abort_aio_request(blk, cb, opaque, -ENOMEDIUM);
1083     }
1084
1085     return bdrv_aio_flush(blk_bs(blk), cb, opaque);
1086 }
1087
1088 BlockAIOCB *blk_aio_pdiscard(BlockBackend *blk,
1089                              int64_t offset, int count,
1090                              BlockCompletionFunc *cb, void *opaque)
1091 {
1092     int ret = blk_check_byte_request(blk, offset, count);
1093     if (ret < 0) {
1094         return blk_abort_aio_request(blk, cb, opaque, ret);
1095     }
1096
1097     return bdrv_aio_pdiscard(blk_bs(blk), offset, count, cb, opaque);
1098 }
1099
1100 void blk_aio_cancel(BlockAIOCB *acb)
1101 {
1102     bdrv_aio_cancel(acb);
1103 }
1104
1105 void blk_aio_cancel_async(BlockAIOCB *acb)
1106 {
1107     bdrv_aio_cancel_async(acb);
1108 }
1109
1110 int blk_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1111 {
1112     if (!blk_is_available(blk)) {
1113         return -ENOMEDIUM;
1114     }
1115
1116     return bdrv_ioctl(blk_bs(blk), req, buf);
1117 }
1118
1119 BlockAIOCB *blk_aio_ioctl(BlockBackend *blk, unsigned long int req, void *buf,
1120                           BlockCompletionFunc *cb, void *opaque)
1121 {
1122     if (!blk_is_available(blk)) {
1123         return blk_abort_aio_request(blk, cb, opaque, -ENOMEDIUM);
1124     }
1125
1126     return bdrv_aio_ioctl(blk_bs(blk), req, buf, cb, opaque);
1127 }
1128
1129 int blk_co_pdiscard(BlockBackend *blk, int64_t offset, int count)
1130 {
1131     int ret = blk_check_byte_request(blk, offset, count);
1132     if (ret < 0) {
1133         return ret;
1134     }
1135
1136     return bdrv_co_pdiscard(blk_bs(blk), offset, count);
1137 }
1138
1139 int blk_co_flush(BlockBackend *blk)
1140 {
1141     if (!blk_is_available(blk)) {
1142         return -ENOMEDIUM;
1143     }
1144
1145     return bdrv_co_flush(blk_bs(blk));
1146 }
1147
1148 int blk_flush(BlockBackend *blk)
1149 {
1150     if (!blk_is_available(blk)) {
1151         return -ENOMEDIUM;
1152     }
1153
1154     return bdrv_flush(blk_bs(blk));
1155 }
1156
1157 void blk_drain(BlockBackend *blk)
1158 {
1159     if (blk_bs(blk)) {
1160         bdrv_drain(blk_bs(blk));
1161     }
1162 }
1163
1164 void blk_drain_all(void)
1165 {
1166     bdrv_drain_all();
1167 }
1168
1169 void blk_set_on_error(BlockBackend *blk, BlockdevOnError on_read_error,
1170                       BlockdevOnError on_write_error)
1171 {
1172     blk->on_read_error = on_read_error;
1173     blk->on_write_error = on_write_error;
1174 }
1175
1176 BlockdevOnError blk_get_on_error(BlockBackend *blk, bool is_read)
1177 {
1178     return is_read ? blk->on_read_error : blk->on_write_error;
1179 }
1180
1181 BlockErrorAction blk_get_error_action(BlockBackend *blk, bool is_read,
1182                                       int error)
1183 {
1184     BlockdevOnError on_err = blk_get_on_error(blk, is_read);
1185
1186     switch (on_err) {
1187     case BLOCKDEV_ON_ERROR_ENOSPC:
1188         return (error == ENOSPC) ?
1189                BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
1190     case BLOCKDEV_ON_ERROR_STOP:
1191         return BLOCK_ERROR_ACTION_STOP;
1192     case BLOCKDEV_ON_ERROR_REPORT:
1193         return BLOCK_ERROR_ACTION_REPORT;
1194     case BLOCKDEV_ON_ERROR_IGNORE:
1195         return BLOCK_ERROR_ACTION_IGNORE;
1196     case BLOCKDEV_ON_ERROR_AUTO:
1197     default:
1198         abort();
1199     }
1200 }
1201
1202 static void send_qmp_error_event(BlockBackend *blk,
1203                                  BlockErrorAction action,
1204                                  bool is_read, int error)
1205 {
1206     IoOperationType optype;
1207
1208     optype = is_read ? IO_OPERATION_TYPE_READ : IO_OPERATION_TYPE_WRITE;
1209     qapi_event_send_block_io_error(blk_name(blk), optype, action,
1210                                    blk_iostatus_is_enabled(blk),
1211                                    error == ENOSPC, strerror(error),
1212                                    &error_abort);
1213 }
1214
1215 /* This is done by device models because, while the block layer knows
1216  * about the error, it does not know whether an operation comes from
1217  * the device or the block layer (from a job, for example).
1218  */
1219 void blk_error_action(BlockBackend *blk, BlockErrorAction action,
1220                       bool is_read, int error)
1221 {
1222     assert(error >= 0);
1223
1224     if (action == BLOCK_ERROR_ACTION_STOP) {
1225         /* First set the iostatus, so that "info block" returns an iostatus
1226          * that matches the events raised so far (an additional error iostatus
1227          * is fine, but not a lost one).
1228          */
1229         blk_iostatus_set_err(blk, error);
1230
1231         /* Then raise the request to stop the VM and the event.
1232          * qemu_system_vmstop_request_prepare has two effects.  First,
1233          * it ensures that the STOP event always comes after the
1234          * BLOCK_IO_ERROR event.  Second, it ensures that even if management
1235          * can observe the STOP event and do a "cont" before the STOP
1236          * event is issued, the VM will not stop.  In this case, vm_start()
1237          * also ensures that the STOP/RESUME pair of events is emitted.
1238          */
1239         qemu_system_vmstop_request_prepare();
1240         send_qmp_error_event(blk, action, is_read, error);
1241         qemu_system_vmstop_request(RUN_STATE_IO_ERROR);
1242     } else {
1243         send_qmp_error_event(blk, action, is_read, error);
1244     }
1245 }
1246
1247 int blk_is_read_only(BlockBackend *blk)
1248 {
1249     BlockDriverState *bs = blk_bs(blk);
1250
1251     if (bs) {
1252         return bdrv_is_read_only(bs);
1253     } else {
1254         return blk->root_state.read_only;
1255     }
1256 }
1257
1258 int blk_is_sg(BlockBackend *blk)
1259 {
1260     BlockDriverState *bs = blk_bs(blk);
1261
1262     if (!bs) {
1263         return 0;
1264     }
1265
1266     return bdrv_is_sg(bs);
1267 }
1268
1269 int blk_enable_write_cache(BlockBackend *blk)
1270 {
1271     return blk->enable_write_cache;
1272 }
1273
1274 void blk_set_enable_write_cache(BlockBackend *blk, bool wce)
1275 {
1276     blk->enable_write_cache = wce;
1277 }
1278
1279 void blk_invalidate_cache(BlockBackend *blk, Error **errp)
1280 {
1281     BlockDriverState *bs = blk_bs(blk);
1282
1283     if (!bs) {
1284         error_setg(errp, "Device '%s' has no medium", blk->name);
1285         return;
1286     }
1287
1288     bdrv_invalidate_cache(bs, errp);
1289 }
1290
1291 bool blk_is_inserted(BlockBackend *blk)
1292 {
1293     BlockDriverState *bs = blk_bs(blk);
1294
1295     return bs && bdrv_is_inserted(bs);
1296 }
1297
1298 bool blk_is_available(BlockBackend *blk)
1299 {
1300     return blk_is_inserted(blk) && !blk_dev_is_tray_open(blk);
1301 }
1302
1303 void blk_lock_medium(BlockBackend *blk, bool locked)
1304 {
1305     BlockDriverState *bs = blk_bs(blk);
1306
1307     if (bs) {
1308         bdrv_lock_medium(bs, locked);
1309     }
1310 }
1311
1312 void blk_eject(BlockBackend *blk, bool eject_flag)
1313 {
1314     BlockDriverState *bs = blk_bs(blk);
1315
1316     if (bs) {
1317         bdrv_eject(bs, eject_flag);
1318     }
1319 }
1320
1321 int blk_get_flags(BlockBackend *blk)
1322 {
1323     BlockDriverState *bs = blk_bs(blk);
1324
1325     if (bs) {
1326         return bdrv_get_flags(bs);
1327     } else {
1328         return blk->root_state.open_flags;
1329     }
1330 }
1331
1332 /* Returns the maximum transfer length, in bytes; guaranteed nonzero */
1333 uint32_t blk_get_max_transfer(BlockBackend *blk)
1334 {
1335     BlockDriverState *bs = blk_bs(blk);
1336     uint32_t max = 0;
1337
1338     if (bs) {
1339         max = bs->bl.max_transfer;
1340     }
1341     return MIN_NON_ZERO(max, INT_MAX);
1342 }
1343
1344 int blk_get_max_iov(BlockBackend *blk)
1345 {
1346     return blk->root->bs->bl.max_iov;
1347 }
1348
1349 void blk_set_guest_block_size(BlockBackend *blk, int align)
1350 {
1351     blk->guest_block_size = align;
1352 }
1353
1354 void *blk_try_blockalign(BlockBackend *blk, size_t size)
1355 {
1356     return qemu_try_blockalign(blk ? blk_bs(blk) : NULL, size);
1357 }
1358
1359 void *blk_blockalign(BlockBackend *blk, size_t size)
1360 {
1361     return qemu_blockalign(blk ? blk_bs(blk) : NULL, size);
1362 }
1363
1364 bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp)
1365 {
1366     BlockDriverState *bs = blk_bs(blk);
1367
1368     if (!bs) {
1369         return false;
1370     }
1371
1372     return bdrv_op_is_blocked(bs, op, errp);
1373 }
1374
1375 void blk_op_unblock(BlockBackend *blk, BlockOpType op, Error *reason)
1376 {
1377     BlockDriverState *bs = blk_bs(blk);
1378
1379     if (bs) {
1380         bdrv_op_unblock(bs, op, reason);
1381     }
1382 }
1383
1384 void blk_op_block_all(BlockBackend *blk, Error *reason)
1385 {
1386     BlockDriverState *bs = blk_bs(blk);
1387
1388     if (bs) {
1389         bdrv_op_block_all(bs, reason);
1390     }
1391 }
1392
1393 void blk_op_unblock_all(BlockBackend *blk, Error *reason)
1394 {
1395     BlockDriverState *bs = blk_bs(blk);
1396
1397     if (bs) {
1398         bdrv_op_unblock_all(bs, reason);
1399     }
1400 }
1401
1402 AioContext *blk_get_aio_context(BlockBackend *blk)
1403 {
1404     BlockDriverState *bs = blk_bs(blk);
1405
1406     if (bs) {
1407         return bdrv_get_aio_context(bs);
1408     } else {
1409         return qemu_get_aio_context();
1410     }
1411 }
1412
1413 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb)
1414 {
1415     BlockBackendAIOCB *blk_acb = DO_UPCAST(BlockBackendAIOCB, common, acb);
1416     return blk_get_aio_context(blk_acb->blk);
1417 }
1418
1419 void blk_set_aio_context(BlockBackend *blk, AioContext *new_context)
1420 {
1421     BlockDriverState *bs = blk_bs(blk);
1422
1423     if (bs) {
1424         if (blk->public.throttle_state) {
1425             throttle_timers_detach_aio_context(&blk->public.throttle_timers);
1426         }
1427         bdrv_set_aio_context(bs, new_context);
1428         if (blk->public.throttle_state) {
1429             throttle_timers_attach_aio_context(&blk->public.throttle_timers,
1430                                                new_context);
1431         }
1432     }
1433 }
1434
1435 void blk_add_aio_context_notifier(BlockBackend *blk,
1436         void (*attached_aio_context)(AioContext *new_context, void *opaque),
1437         void (*detach_aio_context)(void *opaque), void *opaque)
1438 {
1439     BlockDriverState *bs = blk_bs(blk);
1440
1441     if (bs) {
1442         bdrv_add_aio_context_notifier(bs, attached_aio_context,
1443                                       detach_aio_context, opaque);
1444     }
1445 }
1446
1447 void blk_remove_aio_context_notifier(BlockBackend *blk,
1448                                      void (*attached_aio_context)(AioContext *,
1449                                                                   void *),
1450                                      void (*detach_aio_context)(void *),
1451                                      void *opaque)
1452 {
1453     BlockDriverState *bs = blk_bs(blk);
1454
1455     if (bs) {
1456         bdrv_remove_aio_context_notifier(bs, attached_aio_context,
1457                                          detach_aio_context, opaque);
1458     }
1459 }
1460
1461 void blk_add_remove_bs_notifier(BlockBackend *blk, Notifier *notify)
1462 {
1463     notifier_list_add(&blk->remove_bs_notifiers, notify);
1464 }
1465
1466 void blk_add_insert_bs_notifier(BlockBackend *blk, Notifier *notify)
1467 {
1468     notifier_list_add(&blk->insert_bs_notifiers, notify);
1469 }
1470
1471 void blk_io_plug(BlockBackend *blk)
1472 {
1473     BlockDriverState *bs = blk_bs(blk);
1474
1475     if (bs) {
1476         bdrv_io_plug(bs);
1477     }
1478 }
1479
1480 void blk_io_unplug(BlockBackend *blk)
1481 {
1482     BlockDriverState *bs = blk_bs(blk);
1483
1484     if (bs) {
1485         bdrv_io_unplug(bs);
1486     }
1487 }
1488
1489 BlockAcctStats *blk_get_stats(BlockBackend *blk)
1490 {
1491     return &blk->stats;
1492 }
1493
1494 void *blk_aio_get(const AIOCBInfo *aiocb_info, BlockBackend *blk,
1495                   BlockCompletionFunc *cb, void *opaque)
1496 {
1497     return qemu_aio_get(aiocb_info, blk_bs(blk), cb, opaque);
1498 }
1499
1500 int coroutine_fn blk_co_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1501                                       int count, BdrvRequestFlags flags)
1502 {
1503     return blk_co_pwritev(blk, offset, count, NULL,
1504                           flags | BDRV_REQ_ZERO_WRITE);
1505 }
1506
1507 int blk_pwrite_compressed(BlockBackend *blk, int64_t offset, const void *buf,
1508                           int count)
1509 {
1510     return blk_prw(blk, offset, (void *) buf, count, blk_write_entry,
1511                    BDRV_REQ_WRITE_COMPRESSED);
1512 }
1513
1514 int blk_truncate(BlockBackend *blk, int64_t offset)
1515 {
1516     if (!blk_is_available(blk)) {
1517         return -ENOMEDIUM;
1518     }
1519
1520     return bdrv_truncate(blk_bs(blk), offset);
1521 }
1522
1523 int blk_pdiscard(BlockBackend *blk, int64_t offset, int count)
1524 {
1525     int ret = blk_check_byte_request(blk, offset, count);
1526     if (ret < 0) {
1527         return ret;
1528     }
1529
1530     return bdrv_pdiscard(blk_bs(blk), offset, count);
1531 }
1532
1533 int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
1534                      int64_t pos, int size)
1535 {
1536     int ret;
1537
1538     if (!blk_is_available(blk)) {
1539         return -ENOMEDIUM;
1540     }
1541
1542     ret = bdrv_save_vmstate(blk_bs(blk), buf, pos, size);
1543     if (ret < 0) {
1544         return ret;
1545     }
1546
1547     if (ret == size && !blk->enable_write_cache) {
1548         ret = bdrv_flush(blk_bs(blk));
1549     }
1550
1551     return ret < 0 ? ret : size;
1552 }
1553
1554 int blk_load_vmstate(BlockBackend *blk, uint8_t *buf, int64_t pos, int size)
1555 {
1556     if (!blk_is_available(blk)) {
1557         return -ENOMEDIUM;
1558     }
1559
1560     return bdrv_load_vmstate(blk_bs(blk), buf, pos, size);
1561 }
1562
1563 int blk_probe_blocksizes(BlockBackend *blk, BlockSizes *bsz)
1564 {
1565     if (!blk_is_available(blk)) {
1566         return -ENOMEDIUM;
1567     }
1568
1569     return bdrv_probe_blocksizes(blk_bs(blk), bsz);
1570 }
1571
1572 int blk_probe_geometry(BlockBackend *blk, HDGeometry *geo)
1573 {
1574     if (!blk_is_available(blk)) {
1575         return -ENOMEDIUM;
1576     }
1577
1578     return bdrv_probe_geometry(blk_bs(blk), geo);
1579 }
1580
1581 /*
1582  * Updates the BlockBackendRootState object with data from the currently
1583  * attached BlockDriverState.
1584  */
1585 void blk_update_root_state(BlockBackend *blk)
1586 {
1587     assert(blk->root);
1588
1589     blk->root_state.open_flags    = blk->root->bs->open_flags;
1590     blk->root_state.read_only     = blk->root->bs->read_only;
1591     blk->root_state.detect_zeroes = blk->root->bs->detect_zeroes;
1592 }
1593
1594 /*
1595  * Applies the information in the root state to the given BlockDriverState. This
1596  * does not include the flags which have to be specified for bdrv_open(), use
1597  * blk_get_open_flags_from_root_state() to inquire them.
1598  */
1599 void blk_apply_root_state(BlockBackend *blk, BlockDriverState *bs)
1600 {
1601     bs->detect_zeroes = blk->root_state.detect_zeroes;
1602 }
1603
1604 /*
1605  * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
1606  * supposed to inherit the root state.
1607  */
1608 int blk_get_open_flags_from_root_state(BlockBackend *blk)
1609 {
1610     int bs_flags;
1611
1612     bs_flags = blk->root_state.read_only ? 0 : BDRV_O_RDWR;
1613     bs_flags |= blk->root_state.open_flags & ~BDRV_O_RDWR;
1614
1615     return bs_flags;
1616 }
1617
1618 BlockBackendRootState *blk_get_root_state(BlockBackend *blk)
1619 {
1620     return &blk->root_state;
1621 }
1622
1623 int blk_commit_all(void)
1624 {
1625     BlockBackend *blk = NULL;
1626
1627     while ((blk = blk_all_next(blk)) != NULL) {
1628         AioContext *aio_context = blk_get_aio_context(blk);
1629
1630         aio_context_acquire(aio_context);
1631         if (blk_is_inserted(blk) && blk->root->bs->backing) {
1632             int ret = bdrv_commit(blk->root->bs);
1633             if (ret < 0) {
1634                 aio_context_release(aio_context);
1635                 return ret;
1636             }
1637         }
1638         aio_context_release(aio_context);
1639     }
1640     return 0;
1641 }
1642
1643 int blk_flush_all(void)
1644 {
1645     BlockBackend *blk = NULL;
1646     int result = 0;
1647
1648     while ((blk = blk_all_next(blk)) != NULL) {
1649         AioContext *aio_context = blk_get_aio_context(blk);
1650         int ret;
1651
1652         aio_context_acquire(aio_context);
1653         if (blk_is_inserted(blk)) {
1654             ret = blk_flush(blk);
1655             if (ret < 0 && !result) {
1656                 result = ret;
1657             }
1658         }
1659         aio_context_release(aio_context);
1660     }
1661
1662     return result;
1663 }
1664
1665
1666 /* throttling disk I/O limits */
1667 void blk_set_io_limits(BlockBackend *blk, ThrottleConfig *cfg)
1668 {
1669     throttle_group_config(blk, cfg);
1670 }
1671
1672 void blk_io_limits_disable(BlockBackend *blk)
1673 {
1674     assert(blk->public.throttle_state);
1675     bdrv_drained_begin(blk_bs(blk));
1676     throttle_group_unregister_blk(blk);
1677     bdrv_drained_end(blk_bs(blk));
1678 }
1679
1680 /* should be called before blk_set_io_limits if a limit is set */
1681 void blk_io_limits_enable(BlockBackend *blk, const char *group)
1682 {
1683     assert(!blk->public.throttle_state);
1684     throttle_group_register_blk(blk, group);
1685 }
1686
1687 void blk_io_limits_update_group(BlockBackend *blk, const char *group)
1688 {
1689     /* this BB is not part of any group */
1690     if (!blk->public.throttle_state) {
1691         return;
1692     }
1693
1694     /* this BB is a part of the same group than the one we want */
1695     if (!g_strcmp0(throttle_group_get_name(blk), group)) {
1696         return;
1697     }
1698
1699     /* need to change the group this bs belong to */
1700     blk_io_limits_disable(blk);
1701     blk_io_limits_enable(blk, group);
1702 }
1703
1704 static void blk_root_drained_begin(BdrvChild *child)
1705 {
1706     BlockBackend *blk = child->opaque;
1707
1708     /* Note that blk->root may not be accessible here yet if we are just
1709      * attaching to a BlockDriverState that is drained. Use child instead. */
1710
1711     if (blk->public.io_limits_disabled++ == 0) {
1712         throttle_group_restart_blk(blk);
1713     }
1714 }
1715
1716 static void blk_root_drained_end(BdrvChild *child)
1717 {
1718     BlockBackend *blk = child->opaque;
1719
1720     assert(blk->public.io_limits_disabled);
1721     --blk->public.io_limits_disabled;
1722 }