block/block-backend.c

   1 /*
   2  * QEMU Block backends
   3  *
   4  * Copyright (C) 2014-2016 Red Hat, Inc.
   5  *
   6  * Authors:
   7  *  Markus Armbruster <armbru@redhat.com>,
   8  *
   9  * This work is licensed under the terms of the GNU LGPL, version 2.1
  10  * or later.  See the COPYING.LIB file in the top-level directory.
  11  */
  12
  13 #include "qemu/osdep.h"
  14 #include "sysemu/block-backend.h"
  15 #include "block/block_int.h"
  16 #include "block/blockjob.h"
  17 #include "block/throttle-groups.h"
  18 #include "sysemu/blockdev.h"
  19 #include "sysemu/sysemu.h"
  20 #include "qapi-event.h"
  21 #include "qemu/id.h"
  22
  23 /* Number of coroutines to reserve per attached device model */
  24 #define COROUTINE_POOL_RESERVATION 64
  25
  26 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
  27
  28 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb);
  29
  30 struct BlockBackend {
  31     char *name;
  32     int refcnt;
  33     BdrvChild *root;
  34     DriveInfo *legacy_dinfo;    /* null unless created by drive_new() */
  35     QTAILQ_ENTRY(BlockBackend) link;         /* for block_backends */
  36     QTAILQ_ENTRY(BlockBackend) monitor_link; /* for monitor_block_backends */
  37     BlockBackendPublic public;
  38
  39     void *dev;                  /* attached device model, if any */
  40     /* TODO change to DeviceState when all users are qdevified */
  41     const BlockDevOps *dev_ops;
  42     void *dev_opaque;
  43
  44     /* the block size for which the guest device expects atomicity */
  45     int guest_block_size;
  46
  47     /* If the BDS tree is removed, some of its options are stored here (which
  48      * can be used to restore those options in the new BDS on insert) */
  49     BlockBackendRootState root_state;
  50
  51     bool enable_write_cache;
  52
  53     /* I/O stats (display with "info blockstats"). */
  54     BlockAcctStats stats;
  55
  56     BlockdevOnError on_read_error, on_write_error;
  57     bool iostatus_enabled;
  58     BlockDeviceIoStatus iostatus;
  59
  60     bool allow_write_beyond_eof;
  61
  62     NotifierList remove_bs_notifiers, insert_bs_notifiers;
  63 };
  64
  65 typedef struct BlockBackendAIOCB {
  66     BlockAIOCB common;
  67     QEMUBH *bh;
  68     BlockBackend *blk;
  69     int ret;
  70 } BlockBackendAIOCB;
  71
  72 static const AIOCBInfo block_backend_aiocb_info = {
  73     .get_aio_context = blk_aiocb_get_aio_context,
  74     .aiocb_size = sizeof(BlockBackendAIOCB),
  75 };
  76
  77 static void drive_info_del(DriveInfo *dinfo);
  78 static BlockBackend *bdrv_first_blk(BlockDriverState *bs);
  79
  80 /* All BlockBackends */
  81 static QTAILQ_HEAD(, BlockBackend) block_backends =
  82     QTAILQ_HEAD_INITIALIZER(block_backends);
  83
  84 /* All BlockBackends referenced by the monitor and which are iterated through by
  85  * blk_next() */
  86 static QTAILQ_HEAD(, BlockBackend) monitor_block_backends =
  87     QTAILQ_HEAD_INITIALIZER(monitor_block_backends);
  88
  89 static void blk_root_inherit_options(int *child_flags, QDict *child_options,
  90                                      int parent_flags, QDict *parent_options)
  91 {
  92     /* We're not supposed to call this function for root nodes */
  93     abort();
  94 }
  95 static void blk_root_drained_begin(BdrvChild *child);
  96 static void blk_root_drained_end(BdrvChild *child);
  97
  98 static void blk_root_change_media(BdrvChild *child, bool load);
  99 static void blk_root_resize(BdrvChild *child);
 100
 101 static const char *blk_root_get_name(BdrvChild *child)
 102 {
 103     return blk_name(child->opaque);
 104 }
 105
 106 static const BdrvChildRole child_root = {
 107     .inherit_options    = blk_root_inherit_options,
 108
 109     .change_media       = blk_root_change_media,
 110     .resize             = blk_root_resize,
 111     .get_name           = blk_root_get_name,
 112
 113     .drained_begin      = blk_root_drained_begin,
 114     .drained_end        = blk_root_drained_end,
 115 };
 116
 117 /*
 118  * Create a new BlockBackend with a reference count of one.
 119  * Store an error through @errp on failure, unless it's null.
 120  * Return the new BlockBackend on success, null on failure.
 121  */
 122 BlockBackend *blk_new(void)
 123 {
 124     BlockBackend *blk;
 125
 126     blk = g_new0(BlockBackend, 1);
 127     blk->refcnt = 1;
 128     qemu_co_queue_init(&blk->public.throttled_reqs[0]);
 129     qemu_co_queue_init(&blk->public.throttled_reqs[1]);
 130
 131     notifier_list_init(&blk->remove_bs_notifiers);
 132     notifier_list_init(&blk->insert_bs_notifiers);
 133
 134     QTAILQ_INSERT_TAIL(&block_backends, blk, link);
 135     return blk;
 136 }
 137
 138 /*
 139  * Creates a new BlockBackend, opens a new BlockDriverState, and connects both.
 140  *
 141  * Just as with bdrv_open(), after having called this function the reference to
 142  * @options belongs to the block layer (even on failure).
 143  *
 144  * TODO: Remove @filename and @flags; it should be possible to specify a whole
 145  * BDS tree just by specifying the @options QDict (or @reference,
 146  * alternatively). At the time of adding this function, this is not possible,
 147  * though, so callers of this function have to be able to specify @filename and
 148  * @flags.
 149  */
 150 BlockBackend *blk_new_open(const char *filename, const char *reference,
 151                            QDict *options, int flags, Error **errp)
 152 {
 153     BlockBackend *blk;
 154     BlockDriverState *bs;
 155
 156     blk = blk_new();
 157     bs = bdrv_open(filename, reference, options, flags, errp);
 158     if (!bs) {
 159         blk_unref(blk);
 160         return NULL;
 161     }
 162
 163     blk_set_enable_write_cache(blk, true);
 164     blk->root = bdrv_root_attach_child(bs, "root", &child_root);
 165     blk->root->opaque = blk;
 166
 167     return blk;
 168 }
 169
 170 static void blk_delete(BlockBackend *blk)
 171 {
 172     assert(!blk->refcnt);
 173     assert(!blk->name);
 174     assert(!blk->dev);
 175     if (blk->root) {
 176         blk_remove_bs(blk);
 177     }
 178     assert(QLIST_EMPTY(&blk->remove_bs_notifiers.notifiers));
 179     assert(QLIST_EMPTY(&blk->insert_bs_notifiers.notifiers));
 180     QTAILQ_REMOVE(&block_backends, blk, link);
 181     drive_info_del(blk->legacy_dinfo);
 182     block_acct_cleanup(&blk->stats);
 183     g_free(blk);
 184 }
 185
 186 static void drive_info_del(DriveInfo *dinfo)
 187 {
 188     if (!dinfo) {
 189         return;
 190     }
 191     qemu_opts_del(dinfo->opts);
 192     g_free(dinfo->serial);
 193     g_free(dinfo);
 194 }
 195
 196 int blk_get_refcnt(BlockBackend *blk)
 197 {
 198     return blk ? blk->refcnt : 0;
 199 }
 200
 201 /*
 202  * Increment @blk's reference count.
 203  * @blk must not be null.
 204  */
 205 void blk_ref(BlockBackend *blk)
 206 {
 207     blk->refcnt++;
 208 }
 209
 210 /*
 211  * Decrement @blk's reference count.
 212  * If this drops it to zero, destroy @blk.
 213  * For convenience, do nothing if @blk is null.
 214  */
 215 void blk_unref(BlockBackend *blk)
 216 {
 217     if (blk) {
 218         assert(blk->refcnt > 0);
 219         if (!--blk->refcnt) {
 220             blk_delete(blk);
 221         }
 222     }
 223 }
 224
 225 /*
 226  * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
 227  * ones which are hidden (i.e. are not referenced by the monitor).
 228  */
 229 static BlockBackend *blk_all_next(BlockBackend *blk)
 230 {
 231     return blk ? QTAILQ_NEXT(blk, link)
 232                : QTAILQ_FIRST(&block_backends);
 233 }
 234
 235 void blk_remove_all_bs(void)
 236 {
 237     BlockBackend *blk = NULL;
 238
 239     while ((blk = blk_all_next(blk)) != NULL) {
 240         AioContext *ctx = blk_get_aio_context(blk);
 241
 242         aio_context_acquire(ctx);
 243         if (blk->root) {
 244             blk_remove_bs(blk);
 245         }
 246         aio_context_release(ctx);
 247     }
 248 }
 249
 250 /*
 251  * Return the monitor-owned BlockBackend after @blk.
 252  * If @blk is null, return the first one.
 253  * Else, return @blk's next sibling, which may be null.
 254  *
 255  * To iterate over all BlockBackends, do
 256  * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
 257  *     ...
 258  * }
 259  */
 260 BlockBackend *blk_next(BlockBackend *blk)
 261 {
 262     return blk ? QTAILQ_NEXT(blk, monitor_link)
 263                : QTAILQ_FIRST(&monitor_block_backends);
 264 }
 265
 266 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
 267  * the monitor or attached to a BlockBackend */
 268 BlockDriverState *bdrv_next(BdrvNextIterator *it)
 269 {
 270     BlockDriverState *bs;
 271
 272     /* First, return all root nodes of BlockBackends. In order to avoid
 273      * returning a BDS twice when multiple BBs refer to it, we only return it
 274      * if the BB is the first one in the parent list of the BDS. */
 275     if (it->phase == BDRV_NEXT_BACKEND_ROOTS) {
 276         do {
 277             it->blk = blk_all_next(it->blk);
 278             bs = it->blk ? blk_bs(it->blk) : NULL;
 279         } while (it->blk && (bs == NULL || bdrv_first_blk(bs) != it->blk));
 280
 281         if (bs) {
 282             return bs;
 283         }
 284         it->phase = BDRV_NEXT_MONITOR_OWNED;
 285     }
 286
 287     /* Then return the monitor-owned BDSes without a BB attached. Ignore all
 288      * BDSes that are attached to a BlockBackend here; they have been handled
 289      * by the above block already */
 290     do {
 291         it->bs = bdrv_next_monitor_owned(it->bs);
 292         bs = it->bs;
 293     } while (bs && bdrv_has_blk(bs));
 294
 295     return bs;
 296 }
 297
 298 BlockDriverState *bdrv_first(BdrvNextIterator *it)
 299 {
 300     *it = (BdrvNextIterator) {
 301         .phase = BDRV_NEXT_BACKEND_ROOTS,
 302     };
 303
 304     return bdrv_next(it);
 305 }
 306
 307 /*
 308  * Add a BlockBackend into the list of backends referenced by the monitor, with
 309  * the given @name acting as the handle for the monitor.
 310  * Strictly for use by blockdev.c.
 311  *
 312  * @name must not be null or empty.
 313  *
 314  * Returns true on success and false on failure. In the latter case, an Error
 315  * object is returned through @errp.
 316  */
 317 bool monitor_add_blk(BlockBackend *blk, const char *name, Error **errp)
 318 {
 319     assert(!blk->name);
 320     assert(name && name[0]);
 321
 322     if (!id_wellformed(name)) {
 323         error_setg(errp, "Invalid device name");
 324         return false;
 325     }
 326     if (blk_by_name(name)) {
 327         error_setg(errp, "Device with id '%s' already exists", name);
 328         return false;
 329     }
 330     if (bdrv_find_node(name)) {
 331         error_setg(errp,
 332                    "Device name '%s' conflicts with an existing node name",
 333                    name);
 334         return false;
 335     }
 336
 337     blk->name = g_strdup(name);
 338     QTAILQ_INSERT_TAIL(&monitor_block_backends, blk, monitor_link);
 339     return true;
 340 }
 341
 342 /*
 343  * Remove a BlockBackend from the list of backends referenced by the monitor.
 344  * Strictly for use by blockdev.c.
 345  */
 346 void monitor_remove_blk(BlockBackend *blk)
 347 {
 348     if (!blk->name) {
 349         return;
 350     }
 351
 352     QTAILQ_REMOVE(&monitor_block_backends, blk, monitor_link);
 353     g_free(blk->name);
 354     blk->name = NULL;
 355 }
 356
 357 /*
 358  * Return @blk's name, a non-null string.
 359  * Returns an empty string iff @blk is not referenced by the monitor.
 360  */
 361 const char *blk_name(BlockBackend *blk)
 362 {
 363     return blk->name ?: "";
 364 }
 365
 366 /*
 367  * Return the BlockBackend with name @name if it exists, else null.
 368  * @name must not be null.
 369  */
 370 BlockBackend *blk_by_name(const char *name)
 371 {
 372     BlockBackend *blk = NULL;
 373
 374     assert(name);
 375     while ((blk = blk_next(blk)) != NULL) {
 376         if (!strcmp(name, blk->name)) {
 377             return blk;
 378         }
 379     }
 380     return NULL;
 381 }
 382
 383 /*
 384  * Return the BlockDriverState attached to @blk if any, else null.
 385  */
 386 BlockDriverState *blk_bs(BlockBackend *blk)
 387 {
 388     return blk->root ? blk->root->bs : NULL;
 389 }
 390
 391 static BlockBackend *bdrv_first_blk(BlockDriverState *bs)
 392 {
 393     BdrvChild *child;
 394     QLIST_FOREACH(child, &bs->parents, next_parent) {
 395         if (child->role == &child_root) {
 396             return child->opaque;
 397         }
 398     }
 399
 400     return NULL;
 401 }
 402
 403 /*
 404  * Returns true if @bs has an associated BlockBackend.
 405  */
 406 bool bdrv_has_blk(BlockDriverState *bs)
 407 {
 408     return bdrv_first_blk(bs) != NULL;
 409 }
 410
 411 /*
 412  * Return @blk's DriveInfo if any, else null.
 413  */
 414 DriveInfo *blk_legacy_dinfo(BlockBackend *blk)
 415 {
 416     return blk->legacy_dinfo;
 417 }
 418
 419 /*
 420  * Set @blk's DriveInfo to @dinfo, and return it.
 421  * @blk must not have a DriveInfo set already.
 422  * No other BlockBackend may have the same DriveInfo set.
 423  */
 424 DriveInfo *blk_set_legacy_dinfo(BlockBackend *blk, DriveInfo *dinfo)
 425 {
 426     assert(!blk->legacy_dinfo);
 427     return blk->legacy_dinfo = dinfo;
 428 }
 429
 430 /*
 431  * Return the BlockBackend with DriveInfo @dinfo.
 432  * It must exist.
 433  */
 434 BlockBackend *blk_by_legacy_dinfo(DriveInfo *dinfo)
 435 {
 436     BlockBackend *blk = NULL;
 437
 438     while ((blk = blk_next(blk)) != NULL) {
 439         if (blk->legacy_dinfo == dinfo) {
 440             return blk;
 441         }
 442     }
 443     abort();
 444 }
 445
 446 /*
 447  * Returns a pointer to the publicly accessible fields of @blk.
 448  */
 449 BlockBackendPublic *blk_get_public(BlockBackend *blk)
 450 {
 451     return &blk->public;
 452 }
 453
 454 /*
 455  * Returns a BlockBackend given the associated @public fields.
 456  */
 457 BlockBackend *blk_by_public(BlockBackendPublic *public)
 458 {
 459     return container_of(public, BlockBackend, public);
 460 }
 461
 462 /*
 463  * Disassociates the currently associated BlockDriverState from @blk.
 464  */
 465 void blk_remove_bs(BlockBackend *blk)
 466 {
 467     notifier_list_notify(&blk->remove_bs_notifiers, blk);
 468     if (blk->public.throttle_state) {
 469         throttle_timers_detach_aio_context(&blk->public.throttle_timers);
 470     }
 471
 472     blk_update_root_state(blk);
 473
 474     bdrv_root_unref_child(blk->root);
 475     blk->root = NULL;
 476 }
 477
 478 /*
 479  * Associates a new BlockDriverState with @blk.
 480  */
 481 void blk_insert_bs(BlockBackend *blk, BlockDriverState *bs)
 482 {
 483     bdrv_ref(bs);
 484     blk->root = bdrv_root_attach_child(bs, "root", &child_root);
 485     blk->root->opaque = blk;
 486
 487     notifier_list_notify(&blk->insert_bs_notifiers, blk);
 488     if (blk->public.throttle_state) {
 489         throttle_timers_attach_aio_context(
 490             &blk->public.throttle_timers, bdrv_get_aio_context(bs));
 491     }
 492 }
 493
 494 /*
 495  * Attach device model @dev to @blk.
 496  * Return 0 on success, -EBUSY when a device model is attached already.
 497  */
 498 int blk_attach_dev(BlockBackend *blk, void *dev)
 499 /* TODO change to DeviceState *dev when all users are qdevified */
 500 {
 501     if (blk->dev) {
 502         return -EBUSY;
 503     }
 504     blk_ref(blk);
 505     blk->dev = dev;
 506     blk_iostatus_reset(blk);
 507     return 0;
 508 }
 509
 510 /*
 511  * Attach device model @dev to @blk.
 512  * @blk must not have a device model attached already.
 513  * TODO qdevified devices don't use this, remove when devices are qdevified
 514  */
 515 void blk_attach_dev_nofail(BlockBackend *blk, void *dev)
 516 {
 517     if (blk_attach_dev(blk, dev) < 0) {
 518         abort();
 519     }
 520 }
 521
 522 /*
 523  * Detach device model @dev from @blk.
 524  * @dev must be currently attached to @blk.
 525  */
 526 void blk_detach_dev(BlockBackend *blk, void *dev)
 527 /* TODO change to DeviceState *dev when all users are qdevified */
 528 {
 529     assert(blk->dev == dev);
 530     blk->dev = NULL;
 531     blk->dev_ops = NULL;
 532     blk->dev_opaque = NULL;
 533     blk->guest_block_size = 512;
 534     blk_unref(blk);
 535 }
 536
 537 /*
 538  * Return the device model attached to @blk if any, else null.
 539  */
 540 void *blk_get_attached_dev(BlockBackend *blk)
 541 /* TODO change to return DeviceState * when all users are qdevified */
 542 {
 543     return blk->dev;
 544 }
 545
 546 /*
 547  * Set @blk's device model callbacks to @ops.
 548  * @opaque is the opaque argument to pass to the callbacks.
 549  * This is for use by device models.
 550  */
 551 void blk_set_dev_ops(BlockBackend *blk, const BlockDevOps *ops,
 552                      void *opaque)
 553 {
 554     blk->dev_ops = ops;
 555     blk->dev_opaque = opaque;
 556 }
 557
 558 /*
 559  * Notify @blk's attached device model of media change.
 560  * If @load is true, notify of media load.
 561  * Else, notify of media eject.
 562  * Also send DEVICE_TRAY_MOVED events as appropriate.
 563  */
 564 void blk_dev_change_media_cb(BlockBackend *blk, bool load)
 565 {
 566     if (blk->dev_ops && blk->dev_ops->change_media_cb) {
 567         bool tray_was_open, tray_is_open;
 568
 569         tray_was_open = blk_dev_is_tray_open(blk);
 570         blk->dev_ops->change_media_cb(blk->dev_opaque, load);
 571         tray_is_open = blk_dev_is_tray_open(blk);
 572
 573         if (tray_was_open != tray_is_open) {
 574             qapi_event_send_device_tray_moved(blk_name(blk), tray_is_open,
 575                                               &error_abort);
 576         }
 577     }
 578 }
 579
 580 static void blk_root_change_media(BdrvChild *child, bool load)
 581 {
 582     blk_dev_change_media_cb(child->opaque, load);
 583 }
 584
 585 /*
 586  * Does @blk's attached device model have removable media?
 587  * %true if no device model is attached.
 588  */
 589 bool blk_dev_has_removable_media(BlockBackend *blk)
 590 {
 591     return !blk->dev || (blk->dev_ops && blk->dev_ops->change_media_cb);
 592 }
 593
 594 /*
 595  * Does @blk's attached device model have a tray?
 596  */
 597 bool blk_dev_has_tray(BlockBackend *blk)
 598 {
 599     return blk->dev_ops && blk->dev_ops->is_tray_open;
 600 }
 601
 602 /*
 603  * Notify @blk's attached device model of a media eject request.
 604  * If @force is true, the medium is about to be yanked out forcefully.
 605  */
 606 void blk_dev_eject_request(BlockBackend *blk, bool force)
 607 {
 608     if (blk->dev_ops && blk->dev_ops->eject_request_cb) {
 609         blk->dev_ops->eject_request_cb(blk->dev_opaque, force);
 610     }
 611 }
 612
 613 /*
 614  * Does @blk's attached device model have a tray, and is it open?
 615  */
 616 bool blk_dev_is_tray_open(BlockBackend *blk)
 617 {
 618     if (blk_dev_has_tray(blk)) {
 619         return blk->dev_ops->is_tray_open(blk->dev_opaque);
 620     }
 621     return false;
 622 }
 623
 624 /*
 625  * Does @blk's attached device model have the medium locked?
 626  * %false if the device model has no such lock.
 627  */
 628 bool blk_dev_is_medium_locked(BlockBackend *blk)
 629 {
 630     if (blk->dev_ops && blk->dev_ops->is_medium_locked) {
 631         return blk->dev_ops->is_medium_locked(blk->dev_opaque);
 632     }
 633     return false;
 634 }
 635
 636 /*
 637  * Notify @blk's attached device model of a backend size change.
 638  */
 639 static void blk_root_resize(BdrvChild *child)
 640 {
 641     BlockBackend *blk = child->opaque;
 642
 643     if (blk->dev_ops && blk->dev_ops->resize_cb) {
 644         blk->dev_ops->resize_cb(blk->dev_opaque);
 645     }
 646 }
 647
 648 void blk_iostatus_enable(BlockBackend *blk)
 649 {
 650     blk->iostatus_enabled = true;
 651     blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 652 }
 653
 654 /* The I/O status is only enabled if the drive explicitly
 655  * enables it _and_ the VM is configured to stop on errors */
 656 bool blk_iostatus_is_enabled(const BlockBackend *blk)
 657 {
 658     return (blk->iostatus_enabled &&
 659            (blk->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
 660             blk->on_write_error == BLOCKDEV_ON_ERROR_STOP   ||
 661             blk->on_read_error == BLOCKDEV_ON_ERROR_STOP));
 662 }
 663
 664 BlockDeviceIoStatus blk_iostatus(const BlockBackend *blk)
 665 {
 666     return blk->iostatus;
 667 }
 668
 669 void blk_iostatus_disable(BlockBackend *blk)
 670 {
 671     blk->iostatus_enabled = false;
 672 }
 673
 674 void blk_iostatus_reset(BlockBackend *blk)
 675 {
 676     if (blk_iostatus_is_enabled(blk)) {
 677         BlockDriverState *bs = blk_bs(blk);
 678         blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 679         if (bs && bs->job) {
 680             block_job_iostatus_reset(bs->job);
 681         }
 682     }
 683 }
 684
 685 void blk_iostatus_set_err(BlockBackend *blk, int error)
 686 {
 687     assert(blk_iostatus_is_enabled(blk));
 688     if (blk->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 689         blk->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 690                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 691     }
 692 }
 693
 694 void blk_set_allow_write_beyond_eof(BlockBackend *blk, bool allow)
 695 {
 696     blk->allow_write_beyond_eof = allow;
 697 }
 698
 699 static int blk_check_byte_request(BlockBackend *blk, int64_t offset,
 700                                   size_t size)
 701 {
 702     int64_t len;
 703
 704     if (size > INT_MAX) {
 705         return -EIO;
 706     }
 707
 708     if (!blk_is_available(blk)) {
 709         return -ENOMEDIUM;
 710     }
 711
 712     if (offset < 0) {
 713         return -EIO;
 714     }
 715
 716     if (!blk->allow_write_beyond_eof) {
 717         len = blk_getlength(blk);
 718         if (len < 0) {
 719             return len;
 720         }
 721
 722         if (offset > len || len - offset < size) {
 723             return -EIO;
 724         }
 725     }
 726
 727     return 0;
 728 }
 729
 730 static int blk_check_request(BlockBackend *blk, int64_t sector_num,
 731                              int nb_sectors)
 732 {
 733     if (sector_num < 0 || sector_num > INT64_MAX / BDRV_SECTOR_SIZE) {
 734         return -EIO;
 735     }
 736
 737     if (nb_sectors < 0 || nb_sectors > INT_MAX / BDRV_SECTOR_SIZE) {
 738         return -EIO;
 739     }
 740
 741     return blk_check_byte_request(blk, sector_num * BDRV_SECTOR_SIZE,
 742                                   nb_sectors * BDRV_SECTOR_SIZE);
 743 }
 744
 745 static int coroutine_fn blk_co_preadv(BlockBackend *blk, int64_t offset,
 746                                       unsigned int bytes, QEMUIOVector *qiov,
 747                                       BdrvRequestFlags flags)
 748 {
 749     int ret = blk_check_byte_request(blk, offset, bytes);
 750     if (ret < 0) {
 751         return ret;
 752     }
 753
 754     /* throttling disk I/O */
 755     if (blk->public.throttle_state) {
 756         throttle_group_co_io_limits_intercept(blk, bytes, false);
 757     }
 758
 759     return bdrv_co_preadv(blk_bs(blk), offset, bytes, qiov, flags);
 760 }
 761
 762 static int coroutine_fn blk_co_pwritev(BlockBackend *blk, int64_t offset,
 763                                       unsigned int bytes, QEMUIOVector *qiov,
 764                                       BdrvRequestFlags flags)
 765 {
 766     int ret;
 767
 768     ret = blk_check_byte_request(blk, offset, bytes);
 769     if (ret < 0) {
 770         return ret;
 771     }
 772
 773     /* throttling disk I/O */
 774     if (blk->public.throttle_state) {
 775         throttle_group_co_io_limits_intercept(blk, bytes, true);
 776     }
 777
 778     if (!blk->enable_write_cache) {
 779         flags |= BDRV_REQ_FUA;
 780     }
 781
 782     return bdrv_co_pwritev(blk_bs(blk), offset, bytes, qiov, flags);
 783 }
 784
 785 typedef struct BlkRwCo {
 786     BlockBackend *blk;
 787     int64_t offset;
 788     QEMUIOVector *qiov;
 789     int ret;
 790     BdrvRequestFlags flags;
 791 } BlkRwCo;
 792
 793 static void blk_read_entry(void *opaque)
 794 {
 795     BlkRwCo *rwco = opaque;
 796
 797     rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, rwco->qiov->size,
 798                               rwco->qiov, rwco->flags);
 799 }
 800
 801 static void blk_write_entry(void *opaque)
 802 {
 803     BlkRwCo *rwco = opaque;
 804
 805     rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, rwco->qiov->size,
 806                                rwco->qiov, rwco->flags);
 807 }
 808
 809 static int blk_prw(BlockBackend *blk, int64_t offset, uint8_t *buf,
 810                    int64_t bytes, CoroutineEntry co_entry,
 811                    BdrvRequestFlags flags)
 812 {
 813     AioContext *aio_context;
 814     QEMUIOVector qiov;
 815     struct iovec iov;
 816     Coroutine *co;
 817     BlkRwCo rwco;
 818
 819     iov = (struct iovec) {
 820         .iov_base = buf,
 821         .iov_len = bytes,
 822     };
 823     qemu_iovec_init_external(&qiov, &iov, 1);
 824
 825     rwco = (BlkRwCo) {
 826         .blk    = blk,
 827         .offset = offset,
 828         .qiov   = &qiov,
 829         .flags  = flags,
 830         .ret    = NOT_DONE,
 831     };
 832
 833     co = qemu_coroutine_create(co_entry);
 834     qemu_coroutine_enter(co, &rwco);
 835
 836     aio_context = blk_get_aio_context(blk);
 837     while (rwco.ret == NOT_DONE) {
 838         aio_poll(aio_context, true);
 839     }
 840
 841     return rwco.ret;
 842 }
 843
 844 int blk_pread_unthrottled(BlockBackend *blk, int64_t offset, uint8_t *buf,
 845                           int count)
 846 {
 847     int ret;
 848
 849     ret = blk_check_byte_request(blk, offset, count);
 850     if (ret < 0) {
 851         return ret;
 852     }
 853
 854     blk_root_drained_begin(blk->root);
 855     ret = blk_pread(blk, offset, buf, count);
 856     blk_root_drained_end(blk->root);
 857     return ret;
 858 }
 859
 860 int blk_write_zeroes(BlockBackend *blk, int64_t offset,
 861                      int count, BdrvRequestFlags flags)
 862 {
 863     return blk_prw(blk, offset, NULL, count, blk_write_entry,
 864                    flags | BDRV_REQ_ZERO_WRITE);
 865 }
 866
 867 static void error_callback_bh(void *opaque)
 868 {
 869     struct BlockBackendAIOCB *acb = opaque;
 870     qemu_bh_delete(acb->bh);
 871     acb->common.cb(acb->common.opaque, acb->ret);
 872     qemu_aio_unref(acb);
 873 }
 874
 875 BlockAIOCB *blk_abort_aio_request(BlockBackend *blk,
 876                                   BlockCompletionFunc *cb,
 877                                   void *opaque, int ret)
 878 {
 879     struct BlockBackendAIOCB *acb;
 880     QEMUBH *bh;
 881
 882     acb = blk_aio_get(&block_backend_aiocb_info, blk, cb, opaque);
 883     acb->blk = blk;
 884     acb->ret = ret;
 885
 886     bh = aio_bh_new(blk_get_aio_context(blk), error_callback_bh, acb);
 887     acb->bh = bh;
 888     qemu_bh_schedule(bh);
 889
 890     return &acb->common;
 891 }
 892
 893 typedef struct BlkAioEmAIOCB {
 894     BlockAIOCB common;
 895     BlkRwCo rwco;
 896     int bytes;
 897     bool has_returned;
 898     QEMUBH* bh;
 899 } BlkAioEmAIOCB;
 900
 901 static const AIOCBInfo blk_aio_em_aiocb_info = {
 902     .aiocb_size         = sizeof(BlkAioEmAIOCB),
 903 };
 904
 905 static void blk_aio_complete(BlkAioEmAIOCB *acb)
 906 {
 907     if (acb->bh) {
 908         assert(acb->has_returned);
 909         qemu_bh_delete(acb->bh);
 910     }
 911     if (acb->has_returned) {
 912         acb->common.cb(acb->common.opaque, acb->rwco.ret);
 913         qemu_aio_unref(acb);
 914     }
 915 }
 916
 917 static void blk_aio_complete_bh(void *opaque)
 918 {
 919     blk_aio_complete(opaque);
 920 }
 921
 922 static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset, int bytes,
 923                                 QEMUIOVector *qiov, CoroutineEntry co_entry,
 924                                 BdrvRequestFlags flags,
 925                                 BlockCompletionFunc *cb, void *opaque)
 926 {
 927     BlkAioEmAIOCB *acb;
 928     Coroutine *co;
 929
 930     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
 931     acb->rwco = (BlkRwCo) {
 932         .blk    = blk,
 933         .offset = offset,
 934         .qiov   = qiov,
 935         .flags  = flags,
 936         .ret    = NOT_DONE,
 937     };
 938     acb->bytes = bytes;
 939     acb->bh = NULL;
 940     acb->has_returned = false;
 941
 942     co = qemu_coroutine_create(co_entry);
 943     qemu_coroutine_enter(co, acb);
 944
 945     acb->has_returned = true;
 946     if (acb->rwco.ret != NOT_DONE) {
 947         acb->bh = aio_bh_new(blk_get_aio_context(blk), blk_aio_complete_bh, acb);
 948         qemu_bh_schedule(acb->bh);
 949     }
 950
 951     return &acb->common;
 952 }
 953
 954 static void blk_aio_read_entry(void *opaque)
 955 {
 956     BlkAioEmAIOCB *acb = opaque;
 957     BlkRwCo *rwco = &acb->rwco;
 958
 959     assert(rwco->qiov->size == acb->bytes);
 960     rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, acb->bytes,
 961                               rwco->qiov, rwco->flags);
 962     blk_aio_complete(acb);
 963 }
 964
 965 static void blk_aio_write_entry(void *opaque)
 966 {
 967     BlkAioEmAIOCB *acb = opaque;
 968     BlkRwCo *rwco = &acb->rwco;
 969
 970     assert(!rwco->qiov || rwco->qiov->size == acb->bytes);
 971     rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, acb->bytes,
 972                                rwco->qiov, rwco->flags);
 973     blk_aio_complete(acb);
 974 }
 975
 976 BlockAIOCB *blk_aio_write_zeroes(BlockBackend *blk, int64_t offset,
 977                                  int count, BdrvRequestFlags flags,
 978                                  BlockCompletionFunc *cb, void *opaque)
 979 {
 980     return blk_aio_prwv(blk, offset, count, NULL, blk_aio_write_entry,
 981                         flags | BDRV_REQ_ZERO_WRITE, cb, opaque);
 982 }
 983
 984 int blk_pread(BlockBackend *blk, int64_t offset, void *buf, int count)
 985 {
 986     int ret = blk_prw(blk, offset, buf, count, blk_read_entry, 0);
 987     if (ret < 0) {
 988         return ret;
 989     }
 990     return count;
 991 }
 992
 993 int blk_pwrite(BlockBackend *blk, int64_t offset, const void *buf, int count,
 994                BdrvRequestFlags flags)
 995 {
 996     int ret = blk_prw(blk, offset, (void *) buf, count, blk_write_entry,
 997                       flags);
 998     if (ret < 0) {
 999         return ret;
1000     }
1001     return count;
1002 }
1003
1004 int64_t blk_getlength(BlockBackend *blk)
1005 {
1006     if (!blk_is_available(blk)) {
1007         return -ENOMEDIUM;
1008     }
1009
1010     return bdrv_getlength(blk_bs(blk));
1011 }
1012
1013 void blk_get_geometry(BlockBackend *blk, uint64_t *nb_sectors_ptr)
1014 {
1015     if (!blk_bs(blk)) {
1016         *nb_sectors_ptr = 0;
1017     } else {
1018         bdrv_get_geometry(blk_bs(blk), nb_sectors_ptr);
1019     }
1020 }
1021
1022 int64_t blk_nb_sectors(BlockBackend *blk)
1023 {
1024     if (!blk_is_available(blk)) {
1025         return -ENOMEDIUM;
1026     }
1027
1028     return bdrv_nb_sectors(blk_bs(blk));
1029 }
1030
1031 BlockAIOCB *blk_aio_preadv(BlockBackend *blk, int64_t offset,
1032                            QEMUIOVector *qiov, BdrvRequestFlags flags,
1033                            BlockCompletionFunc *cb, void *opaque)
1034 {
1035     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1036                         blk_aio_read_entry, flags, cb, opaque);
1037 }
1038
1039 BlockAIOCB *blk_aio_pwritev(BlockBackend *blk, int64_t offset,
1040                             QEMUIOVector *qiov, BdrvRequestFlags flags,
1041                             BlockCompletionFunc *cb, void *opaque)
1042 {
1043     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1044                         blk_aio_write_entry, flags, cb, opaque);
1045 }
1046
1047 BlockAIOCB *blk_aio_flush(BlockBackend *blk,
1048                           BlockCompletionFunc *cb, void *opaque)
1049 {
1050     if (!blk_is_available(blk)) {
1051         return blk_abort_aio_request(blk, cb, opaque, -ENOMEDIUM);
1052     }
1053
1054     return bdrv_aio_flush(blk_bs(blk), cb, opaque);
1055 }
1056
1057 BlockAIOCB *blk_aio_discard(BlockBackend *blk,
1058                             int64_t sector_num, int nb_sectors,
1059                             BlockCompletionFunc *cb, void *opaque)
1060 {
1061     int ret = blk_check_request(blk, sector_num, nb_sectors);
1062     if (ret < 0) {
1063         return blk_abort_aio_request(blk, cb, opaque, ret);
1064     }
1065
1066     return bdrv_aio_discard(blk_bs(blk), sector_num, nb_sectors, cb, opaque);
1067 }
1068
1069 void blk_aio_cancel(BlockAIOCB *acb)
1070 {
1071     bdrv_aio_cancel(acb);
1072 }
1073
1074 void blk_aio_cancel_async(BlockAIOCB *acb)
1075 {
1076     bdrv_aio_cancel_async(acb);
1077 }
1078
1079 int blk_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1080 {
1081     if (!blk_is_available(blk)) {
1082         return -ENOMEDIUM;
1083     }
1084
1085     return bdrv_ioctl(blk_bs(blk), req, buf);
1086 }
1087
1088 BlockAIOCB *blk_aio_ioctl(BlockBackend *blk, unsigned long int req, void *buf,
1089                           BlockCompletionFunc *cb, void *opaque)
1090 {
1091     if (!blk_is_available(blk)) {
1092         return blk_abort_aio_request(blk, cb, opaque, -ENOMEDIUM);
1093     }
1094
1095     return bdrv_aio_ioctl(blk_bs(blk), req, buf, cb, opaque);
1096 }
1097
1098 int blk_co_discard(BlockBackend *blk, int64_t sector_num, int nb_sectors)
1099 {
1100     int ret = blk_check_request(blk, sector_num, nb_sectors);
1101     if (ret < 0) {
1102         return ret;
1103     }
1104
1105     return bdrv_co_discard(blk_bs(blk), sector_num, nb_sectors);
1106 }
1107
1108 int blk_co_flush(BlockBackend *blk)
1109 {
1110     if (!blk_is_available(blk)) {
1111         return -ENOMEDIUM;
1112     }
1113
1114     return bdrv_co_flush(blk_bs(blk));
1115 }
1116
1117 int blk_flush(BlockBackend *blk)
1118 {
1119     if (!blk_is_available(blk)) {
1120         return -ENOMEDIUM;
1121     }
1122
1123     return bdrv_flush(blk_bs(blk));
1124 }
1125
1126 void blk_drain(BlockBackend *blk)
1127 {
1128     if (blk_bs(blk)) {
1129         bdrv_drain(blk_bs(blk));
1130     }
1131 }
1132
1133 void blk_drain_all(void)
1134 {
1135     bdrv_drain_all();
1136 }
1137
1138 void blk_set_on_error(BlockBackend *blk, BlockdevOnError on_read_error,
1139                       BlockdevOnError on_write_error)
1140 {
1141     blk->on_read_error = on_read_error;
1142     blk->on_write_error = on_write_error;
1143 }
1144
1145 BlockdevOnError blk_get_on_error(BlockBackend *blk, bool is_read)
1146 {
1147     return is_read ? blk->on_read_error : blk->on_write_error;
1148 }
1149
1150 BlockErrorAction blk_get_error_action(BlockBackend *blk, bool is_read,
1151                                       int error)
1152 {
1153     BlockdevOnError on_err = blk_get_on_error(blk, is_read);
1154
1155     switch (on_err) {
1156     case BLOCKDEV_ON_ERROR_ENOSPC:
1157         return (error == ENOSPC) ?
1158                BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
1159     case BLOCKDEV_ON_ERROR_STOP:
1160         return BLOCK_ERROR_ACTION_STOP;
1161     case BLOCKDEV_ON_ERROR_REPORT:
1162         return BLOCK_ERROR_ACTION_REPORT;
1163     case BLOCKDEV_ON_ERROR_IGNORE:
1164         return BLOCK_ERROR_ACTION_IGNORE;
1165     default:
1166         abort();
1167     }
1168 }
1169
1170 static void send_qmp_error_event(BlockBackend *blk,
1171                                  BlockErrorAction action,
1172                                  bool is_read, int error)
1173 {
1174     IoOperationType optype;
1175
1176     optype = is_read ? IO_OPERATION_TYPE_READ : IO_OPERATION_TYPE_WRITE;
1177     qapi_event_send_block_io_error(blk_name(blk), optype, action,
1178                                    blk_iostatus_is_enabled(blk),
1179                                    error == ENOSPC, strerror(error),
1180                                    &error_abort);
1181 }
1182
1183 /* This is done by device models because, while the block layer knows
1184  * about the error, it does not know whether an operation comes from
1185  * the device or the block layer (from a job, for example).
1186  */
1187 void blk_error_action(BlockBackend *blk, BlockErrorAction action,
1188                       bool is_read, int error)
1189 {
1190     assert(error >= 0);
1191
1192     if (action == BLOCK_ERROR_ACTION_STOP) {
1193         /* First set the iostatus, so that "info block" returns an iostatus
1194          * that matches the events raised so far (an additional error iostatus
1195          * is fine, but not a lost one).
1196          */
1197         blk_iostatus_set_err(blk, error);
1198
1199         /* Then raise the request to stop the VM and the event.
1200          * qemu_system_vmstop_request_prepare has two effects.  First,
1201          * it ensures that the STOP event always comes after the
1202          * BLOCK_IO_ERROR event.  Second, it ensures that even if management
1203          * can observe the STOP event and do a "cont" before the STOP
1204          * event is issued, the VM will not stop.  In this case, vm_start()
1205          * also ensures that the STOP/RESUME pair of events is emitted.
1206          */
1207         qemu_system_vmstop_request_prepare();
1208         send_qmp_error_event(blk, action, is_read, error);
1209         qemu_system_vmstop_request(RUN_STATE_IO_ERROR);
1210     } else {
1211         send_qmp_error_event(blk, action, is_read, error);
1212     }
1213 }
1214
1215 int blk_is_read_only(BlockBackend *blk)
1216 {
1217     BlockDriverState *bs = blk_bs(blk);
1218
1219     if (bs) {
1220         return bdrv_is_read_only(bs);
1221     } else {
1222         return blk->root_state.read_only;
1223     }
1224 }
1225
1226 int blk_is_sg(BlockBackend *blk)
1227 {
1228     BlockDriverState *bs = blk_bs(blk);
1229
1230     if (!bs) {
1231         return 0;
1232     }
1233
1234     return bdrv_is_sg(bs);
1235 }
1236
1237 int blk_enable_write_cache(BlockBackend *blk)
1238 {
1239     return blk->enable_write_cache;
1240 }
1241
1242 void blk_set_enable_write_cache(BlockBackend *blk, bool wce)
1243 {
1244     blk->enable_write_cache = wce;
1245 }
1246
1247 void blk_invalidate_cache(BlockBackend *blk, Error **errp)
1248 {
1249     BlockDriverState *bs = blk_bs(blk);
1250
1251     if (!bs) {
1252         error_setg(errp, "Device '%s' has no medium", blk->name);
1253         return;
1254     }
1255
1256     bdrv_invalidate_cache(bs, errp);
1257 }
1258
1259 bool blk_is_inserted(BlockBackend *blk)
1260 {
1261     BlockDriverState *bs = blk_bs(blk);
1262
1263     return bs && bdrv_is_inserted(bs);
1264 }
1265
1266 bool blk_is_available(BlockBackend *blk)
1267 {
1268     return blk_is_inserted(blk) && !blk_dev_is_tray_open(blk);
1269 }
1270
1271 void blk_lock_medium(BlockBackend *blk, bool locked)
1272 {
1273     BlockDriverState *bs = blk_bs(blk);
1274
1275     if (bs) {
1276         bdrv_lock_medium(bs, locked);
1277     }
1278 }
1279
1280 void blk_eject(BlockBackend *blk, bool eject_flag)
1281 {
1282     BlockDriverState *bs = blk_bs(blk);
1283
1284     if (bs) {
1285         bdrv_eject(bs, eject_flag);
1286     }
1287 }
1288
1289 int blk_get_flags(BlockBackend *blk)
1290 {
1291     BlockDriverState *bs = blk_bs(blk);
1292
1293     if (bs) {
1294         return bdrv_get_flags(bs);
1295     } else {
1296         return blk->root_state.open_flags;
1297     }
1298 }
1299
1300 int blk_get_max_transfer_length(BlockBackend *blk)
1301 {
1302     BlockDriverState *bs = blk_bs(blk);
1303
1304     if (bs) {
1305         return bs->bl.max_transfer_length;
1306     } else {
1307         return 0;
1308     }
1309 }
1310
1311 int blk_get_max_iov(BlockBackend *blk)
1312 {
1313     return blk->root->bs->bl.max_iov;
1314 }
1315
1316 void blk_set_guest_block_size(BlockBackend *blk, int align)
1317 {
1318     blk->guest_block_size = align;
1319 }
1320
1321 void *blk_try_blockalign(BlockBackend *blk, size_t size)
1322 {
1323     return qemu_try_blockalign(blk ? blk_bs(blk) : NULL, size);
1324 }
1325
1326 void *blk_blockalign(BlockBackend *blk, size_t size)
1327 {
1328     return qemu_blockalign(blk ? blk_bs(blk) : NULL, size);
1329 }
1330
1331 bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp)
1332 {
1333     BlockDriverState *bs = blk_bs(blk);
1334
1335     if (!bs) {
1336         return false;
1337     }
1338
1339     return bdrv_op_is_blocked(bs, op, errp);
1340 }
1341
1342 void blk_op_unblock(BlockBackend *blk, BlockOpType op, Error *reason)
1343 {
1344     BlockDriverState *bs = blk_bs(blk);
1345
1346     if (bs) {
1347         bdrv_op_unblock(bs, op, reason);
1348     }
1349 }
1350
1351 void blk_op_block_all(BlockBackend *blk, Error *reason)
1352 {
1353     BlockDriverState *bs = blk_bs(blk);
1354
1355     if (bs) {
1356         bdrv_op_block_all(bs, reason);
1357     }
1358 }
1359
1360 void blk_op_unblock_all(BlockBackend *blk, Error *reason)
1361 {
1362     BlockDriverState *bs = blk_bs(blk);
1363
1364     if (bs) {
1365         bdrv_op_unblock_all(bs, reason);
1366     }
1367 }
1368
1369 AioContext *blk_get_aio_context(BlockBackend *blk)
1370 {
1371     BlockDriverState *bs = blk_bs(blk);
1372
1373     if (bs) {
1374         return bdrv_get_aio_context(bs);
1375     } else {
1376         return qemu_get_aio_context();
1377     }
1378 }
1379
1380 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb)
1381 {
1382     BlockBackendAIOCB *blk_acb = DO_UPCAST(BlockBackendAIOCB, common, acb);
1383     return blk_get_aio_context(blk_acb->blk);
1384 }
1385
1386 void blk_set_aio_context(BlockBackend *blk, AioContext *new_context)
1387 {
1388     BlockDriverState *bs = blk_bs(blk);
1389
1390     if (bs) {
1391         if (blk->public.throttle_state) {
1392             throttle_timers_detach_aio_context(&blk->public.throttle_timers);
1393         }
1394         bdrv_set_aio_context(bs, new_context);
1395         if (blk->public.throttle_state) {
1396             throttle_timers_attach_aio_context(&blk->public.throttle_timers,
1397                                                new_context);
1398         }
1399     }
1400 }
1401
1402 void blk_add_aio_context_notifier(BlockBackend *blk,
1403         void (*attached_aio_context)(AioContext *new_context, void *opaque),
1404         void (*detach_aio_context)(void *opaque), void *opaque)
1405 {
1406     BlockDriverState *bs = blk_bs(blk);
1407
1408     if (bs) {
1409         bdrv_add_aio_context_notifier(bs, attached_aio_context,
1410                                       detach_aio_context, opaque);
1411     }
1412 }
1413
1414 void blk_remove_aio_context_notifier(BlockBackend *blk,
1415                                      void (*attached_aio_context)(AioContext *,
1416                                                                   void *),
1417                                      void (*detach_aio_context)(void *),
1418                                      void *opaque)
1419 {
1420     BlockDriverState *bs = blk_bs(blk);
1421
1422     if (bs) {
1423         bdrv_remove_aio_context_notifier(bs, attached_aio_context,
1424                                          detach_aio_context, opaque);
1425     }
1426 }
1427
1428 void blk_add_remove_bs_notifier(BlockBackend *blk, Notifier *notify)
1429 {
1430     notifier_list_add(&blk->remove_bs_notifiers, notify);
1431 }
1432
1433 void blk_add_insert_bs_notifier(BlockBackend *blk, Notifier *notify)
1434 {
1435     notifier_list_add(&blk->insert_bs_notifiers, notify);
1436 }
1437
1438 void blk_io_plug(BlockBackend *blk)
1439 {
1440     BlockDriverState *bs = blk_bs(blk);
1441
1442     if (bs) {
1443         bdrv_io_plug(bs);
1444     }
1445 }
1446
1447 void blk_io_unplug(BlockBackend *blk)
1448 {
1449     BlockDriverState *bs = blk_bs(blk);
1450
1451     if (bs) {
1452         bdrv_io_unplug(bs);
1453     }
1454 }
1455
1456 BlockAcctStats *blk_get_stats(BlockBackend *blk)
1457 {
1458     return &blk->stats;
1459 }
1460
1461 void *blk_aio_get(const AIOCBInfo *aiocb_info, BlockBackend *blk,
1462                   BlockCompletionFunc *cb, void *opaque)
1463 {
1464     return qemu_aio_get(aiocb_info, blk_bs(blk), cb, opaque);
1465 }
1466
1467 int coroutine_fn blk_co_write_zeroes(BlockBackend *blk, int64_t offset,
1468                                      int count, BdrvRequestFlags flags)
1469 {
1470     return blk_co_pwritev(blk, offset, count, NULL,
1471                           flags | BDRV_REQ_ZERO_WRITE);
1472 }
1473
1474 int blk_write_compressed(BlockBackend *blk, int64_t sector_num,
1475                          const uint8_t *buf, int nb_sectors)
1476 {
1477     int ret = blk_check_request(blk, sector_num, nb_sectors);
1478     if (ret < 0) {
1479         return ret;
1480     }
1481
1482     return bdrv_write_compressed(blk_bs(blk), sector_num, buf, nb_sectors);
1483 }
1484
1485 int blk_truncate(BlockBackend *blk, int64_t offset)
1486 {
1487     if (!blk_is_available(blk)) {
1488         return -ENOMEDIUM;
1489     }
1490
1491     return bdrv_truncate(blk_bs(blk), offset);
1492 }
1493
1494 int blk_discard(BlockBackend *blk, int64_t sector_num, int nb_sectors)
1495 {
1496     int ret = blk_check_request(blk, sector_num, nb_sectors);
1497     if (ret < 0) {
1498         return ret;
1499     }
1500
1501     return bdrv_discard(blk_bs(blk), sector_num, nb_sectors);
1502 }
1503
1504 int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
1505                      int64_t pos, int size)
1506 {
1507     int ret;
1508
1509     if (!blk_is_available(blk)) {
1510         return -ENOMEDIUM;
1511     }
1512
1513     ret = bdrv_save_vmstate(blk_bs(blk), buf, pos, size);
1514     if (ret < 0) {
1515         return ret;
1516     }
1517
1518     if (ret == size && !blk->enable_write_cache) {
1519         ret = bdrv_flush(blk_bs(blk));
1520     }
1521
1522     return ret < 0 ? ret : size;
1523 }
1524
1525 int blk_load_vmstate(BlockBackend *blk, uint8_t *buf, int64_t pos, int size)
1526 {
1527     if (!blk_is_available(blk)) {
1528         return -ENOMEDIUM;
1529     }
1530
1531     return bdrv_load_vmstate(blk_bs(blk), buf, pos, size);
1532 }
1533
1534 int blk_probe_blocksizes(BlockBackend *blk, BlockSizes *bsz)
1535 {
1536     if (!blk_is_available(blk)) {
1537         return -ENOMEDIUM;
1538     }
1539
1540     return bdrv_probe_blocksizes(blk_bs(blk), bsz);
1541 }
1542
1543 int blk_probe_geometry(BlockBackend *blk, HDGeometry *geo)
1544 {
1545     if (!blk_is_available(blk)) {
1546         return -ENOMEDIUM;
1547     }
1548
1549     return bdrv_probe_geometry(blk_bs(blk), geo);
1550 }
1551
1552 /*
1553  * Updates the BlockBackendRootState object with data from the currently
1554  * attached BlockDriverState.
1555  */
1556 void blk_update_root_state(BlockBackend *blk)
1557 {
1558     assert(blk->root);
1559
1560     blk->root_state.open_flags    = blk->root->bs->open_flags;
1561     blk->root_state.read_only     = blk->root->bs->read_only;
1562     blk->root_state.detect_zeroes = blk->root->bs->detect_zeroes;
1563 }
1564
1565 /*
1566  * Applies the information in the root state to the given BlockDriverState. This
1567  * does not include the flags which have to be specified for bdrv_open(), use
1568  * blk_get_open_flags_from_root_state() to inquire them.
1569  */
1570 void blk_apply_root_state(BlockBackend *blk, BlockDriverState *bs)
1571 {
1572     bs->detect_zeroes = blk->root_state.detect_zeroes;
1573 }
1574
1575 /*
1576  * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
1577  * supposed to inherit the root state.
1578  */
1579 int blk_get_open_flags_from_root_state(BlockBackend *blk)
1580 {
1581     int bs_flags;
1582
1583     bs_flags = blk->root_state.read_only ? 0 : BDRV_O_RDWR;
1584     bs_flags |= blk->root_state.open_flags & ~BDRV_O_RDWR;
1585
1586     return bs_flags;
1587 }
1588
1589 BlockBackendRootState *blk_get_root_state(BlockBackend *blk)
1590 {
1591     return &blk->root_state;
1592 }
1593
1594 int blk_commit_all(void)
1595 {
1596     BlockBackend *blk = NULL;
1597
1598     while ((blk = blk_all_next(blk)) != NULL) {
1599         AioContext *aio_context = blk_get_aio_context(blk);
1600
1601         aio_context_acquire(aio_context);
1602         if (blk_is_inserted(blk) && blk->root->bs->backing) {
1603             int ret = bdrv_commit(blk->root->bs);
1604             if (ret < 0) {
1605                 aio_context_release(aio_context);
1606                 return ret;
1607             }
1608         }
1609         aio_context_release(aio_context);
1610     }
1611     return 0;
1612 }
1613
1614 int blk_flush_all(void)
1615 {
1616     BlockBackend *blk = NULL;
1617     int result = 0;
1618
1619     while ((blk = blk_all_next(blk)) != NULL) {
1620         AioContext *aio_context = blk_get_aio_context(blk);
1621         int ret;
1622
1623         aio_context_acquire(aio_context);
1624         if (blk_is_inserted(blk)) {
1625             ret = blk_flush(blk);
1626             if (ret < 0 && !result) {
1627                 result = ret;
1628             }
1629         }
1630         aio_context_release(aio_context);
1631     }
1632
1633     return result;
1634 }
1635
1636
1637 /* throttling disk I/O limits */
1638 void blk_set_io_limits(BlockBackend *blk, ThrottleConfig *cfg)
1639 {
1640     throttle_group_config(blk, cfg);
1641 }
1642
1643 void blk_io_limits_disable(BlockBackend *blk)
1644 {
1645     assert(blk->public.throttle_state);
1646     bdrv_drained_begin(blk_bs(blk));
1647     throttle_group_unregister_blk(blk);
1648     bdrv_drained_end(blk_bs(blk));
1649 }
1650
1651 /* should be called before blk_set_io_limits if a limit is set */
1652 void blk_io_limits_enable(BlockBackend *blk, const char *group)
1653 {
1654     assert(!blk->public.throttle_state);
1655     throttle_group_register_blk(blk, group);
1656 }
1657
1658 void blk_io_limits_update_group(BlockBackend *blk, const char *group)
1659 {
1660     /* this BB is not part of any group */
1661     if (!blk->public.throttle_state) {
1662         return;
1663     }
1664
1665     /* this BB is a part of the same group than the one we want */
1666     if (!g_strcmp0(throttle_group_get_name(blk), group)) {
1667         return;
1668     }
1669
1670     /* need to change the group this bs belong to */
1671     blk_io_limits_disable(blk);
1672     blk_io_limits_enable(blk, group);
1673 }
1674
1675 static void blk_root_drained_begin(BdrvChild *child)
1676 {
1677     BlockBackend *blk = child->opaque;
1678
1679     if (blk->public.io_limits_disabled++ == 0) {
1680         throttle_group_restart_blk(blk);
1681     }
1682 }
1683
1684 static void blk_root_drained_end(BdrvChild *child)
1685 {
1686     BlockBackend *blk = child->opaque;
1687
1688     assert(blk->public.io_limits_disabled);
1689     --blk->public.io_limits_disabled;
1690 }