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     bool legacy_dev;            /* true if dev is not a DeviceState */
  42     /* TODO change to DeviceState when all users are qdevified */
  43     const BlockDevOps *dev_ops;
  44     void *dev_opaque;
  45
  46     /* the block size for which the guest device expects atomicity */
  47     int guest_block_size;
  48
  49     /* If the BDS tree is removed, some of its options are stored here (which
  50      * can be used to restore those options in the new BDS on insert) */
  51     BlockBackendRootState root_state;
  52
  53     bool enable_write_cache;
  54
  55     /* I/O stats (display with "info blockstats"). */
  56     BlockAcctStats stats;
  57
  58     BlockdevOnError on_read_error, on_write_error;
  59     bool iostatus_enabled;
  60     BlockDeviceIoStatus iostatus;
  61
  62     uint64_t perm;
  63     uint64_t shared_perm;
  64     bool disable_perm;
  65
  66     bool allow_write_beyond_eof;
  67
  68     NotifierList remove_bs_notifiers, insert_bs_notifiers;
  69
  70     int quiesce_counter;
  71 };
  72
  73 typedef struct BlockBackendAIOCB {
  74     BlockAIOCB common;
  75     BlockBackend *blk;
  76     int ret;
  77 } BlockBackendAIOCB;
  78
  79 static const AIOCBInfo block_backend_aiocb_info = {
  80     .get_aio_context = blk_aiocb_get_aio_context,
  81     .aiocb_size = sizeof(BlockBackendAIOCB),
  82 };
  83
  84 static void drive_info_del(DriveInfo *dinfo);
  85 static BlockBackend *bdrv_first_blk(BlockDriverState *bs);
  86 static char *blk_get_attached_dev_id(BlockBackend *blk);
  87
  88 /* All BlockBackends */
  89 static QTAILQ_HEAD(, BlockBackend) block_backends =
  90     QTAILQ_HEAD_INITIALIZER(block_backends);
  91
  92 /* All BlockBackends referenced by the monitor and which are iterated through by
  93  * blk_next() */
  94 static QTAILQ_HEAD(, BlockBackend) monitor_block_backends =
  95     QTAILQ_HEAD_INITIALIZER(monitor_block_backends);
  96
  97 static void blk_root_inherit_options(int *child_flags, QDict *child_options,
  98                                      int parent_flags, QDict *parent_options)
  99 {
 100     /* We're not supposed to call this function for root nodes */
 101     abort();
 102 }
 103 static void blk_root_drained_begin(BdrvChild *child);
 104 static void blk_root_drained_end(BdrvChild *child);
 105
 106 static void blk_root_change_media(BdrvChild *child, bool load);
 107 static void blk_root_resize(BdrvChild *child);
 108
 109 static char *blk_root_get_parent_desc(BdrvChild *child)
 110 {
 111     BlockBackend *blk = child->opaque;
 112     char *dev_id;
 113
 114     if (blk->name) {
 115         return g_strdup(blk->name);
 116     }
 117
 118     dev_id = blk_get_attached_dev_id(blk);
 119     if (*dev_id) {
 120         return dev_id;
 121     } else {
 122         /* TODO Callback into the BB owner for something more detailed */
 123         g_free(dev_id);
 124         return g_strdup("a block device");
 125     }
 126 }
 127
 128 static const char *blk_root_get_name(BdrvChild *child)
 129 {
 130     return blk_name(child->opaque);
 131 }
 132
 133 /*
 134  * Notifies the user of the BlockBackend that migration has completed. qdev
 135  * devices can tighten their permissions in response (specifically revoke
 136  * shared write permissions that we needed for storage migration).
 137  *
 138  * If an error is returned, the VM cannot be allowed to be resumed.
 139  */
 140 static void blk_root_activate(BdrvChild *child, Error **errp)
 141 {
 142     BlockBackend *blk = child->opaque;
 143     Error *local_err = NULL;
 144
 145     if (!blk->disable_perm) {
 146         return;
 147     }
 148
 149     blk->disable_perm = false;
 150
 151     blk_set_perm(blk, blk->perm, blk->shared_perm, &local_err);
 152     if (local_err) {
 153         error_propagate(errp, local_err);
 154         blk->disable_perm = true;
 155         return;
 156     }
 157 }
 158
 159 static int blk_root_inactivate(BdrvChild *child)
 160 {
 161     BlockBackend *blk = child->opaque;
 162
 163     if (blk->disable_perm) {
 164         return 0;
 165     }
 166
 167     /* Only inactivate BlockBackends for guest devices (which are inactive at
 168      * this point because the VM is stopped) and unattached monitor-owned
 169      * BlockBackends. If there is still any other user like a block job, then
 170      * we simply can't inactivate the image. */
 171     if (!blk->dev && !blk_name(blk)[0]) {
 172         return -EPERM;
 173     }
 174
 175     blk->disable_perm = true;
 176     if (blk->root) {
 177         bdrv_child_try_set_perm(blk->root, 0, BLK_PERM_ALL, &error_abort);
 178     }
 179
 180     return 0;
 181 }
 182
 183 static const BdrvChildRole child_root = {
 184     .inherit_options    = blk_root_inherit_options,
 185
 186     .change_media       = blk_root_change_media,
 187     .resize             = blk_root_resize,
 188     .get_name           = blk_root_get_name,
 189     .get_parent_desc    = blk_root_get_parent_desc,
 190
 191     .drained_begin      = blk_root_drained_begin,
 192     .drained_end        = blk_root_drained_end,
 193
 194     .activate           = blk_root_activate,
 195     .inactivate         = blk_root_inactivate,
 196 };
 197
 198 /*
 199  * Create a new BlockBackend with a reference count of one.
 200  *
 201  * @perm is a bitmasks of BLK_PERM_* constants which describes the permissions
 202  * to request for a block driver node that is attached to this BlockBackend.
 203  * @shared_perm is a bitmask which describes which permissions may be granted
 204  * to other users of the attached node.
 205  * Both sets of permissions can be changed later using blk_set_perm().
 206  *
 207  * Return the new BlockBackend on success, null on failure.
 208  */
 209 BlockBackend *blk_new(uint64_t perm, uint64_t shared_perm)
 210 {
 211     BlockBackend *blk;
 212
 213     blk = g_new0(BlockBackend, 1);
 214     blk->refcnt = 1;
 215     blk->perm = perm;
 216     blk->shared_perm = shared_perm;
 217     blk_set_enable_write_cache(blk, true);
 218
 219     qemu_co_mutex_init(&blk->public.throttled_reqs_lock);
 220     qemu_co_queue_init(&blk->public.throttled_reqs[0]);
 221     qemu_co_queue_init(&blk->public.throttled_reqs[1]);
 222     block_acct_init(&blk->stats);
 223
 224     notifier_list_init(&blk->remove_bs_notifiers);
 225     notifier_list_init(&blk->insert_bs_notifiers);
 226
 227     QTAILQ_INSERT_TAIL(&block_backends, blk, link);
 228     return blk;
 229 }
 230
 231 /*
 232  * Creates a new BlockBackend, opens a new BlockDriverState, and connects both.
 233  *
 234  * Just as with bdrv_open(), after having called this function the reference to
 235  * @options belongs to the block layer (even on failure).
 236  *
 237  * TODO: Remove @filename and @flags; it should be possible to specify a whole
 238  * BDS tree just by specifying the @options QDict (or @reference,
 239  * alternatively). At the time of adding this function, this is not possible,
 240  * though, so callers of this function have to be able to specify @filename and
 241  * @flags.
 242  */
 243 BlockBackend *blk_new_open(const char *filename, const char *reference,
 244                            QDict *options, int flags, Error **errp)
 245 {
 246     BlockBackend *blk;
 247     BlockDriverState *bs;
 248     uint64_t perm;
 249
 250     /* blk_new_open() is mainly used in .bdrv_create implementations and the
 251      * tools where sharing isn't a concern because the BDS stays private, so we
 252      * just request permission according to the flags.
 253      *
 254      * The exceptions are xen_disk and blockdev_init(); in these cases, the
 255      * caller of blk_new_open() doesn't make use of the permissions, but they
 256      * shouldn't hurt either. We can still share everything here because the
 257      * guest devices will add their own blockers if they can't share. */
 258     perm = BLK_PERM_CONSISTENT_READ;
 259     if (flags & BDRV_O_RDWR) {
 260         perm |= BLK_PERM_WRITE;
 261     }
 262     if (flags & BDRV_O_RESIZE) {
 263         perm |= BLK_PERM_RESIZE;
 264     }
 265
 266     blk = blk_new(perm, BLK_PERM_ALL);
 267     bs = bdrv_open(filename, reference, options, flags, errp);
 268     if (!bs) {
 269         blk_unref(blk);
 270         return NULL;
 271     }
 272
 273     blk->root = bdrv_root_attach_child(bs, "root", &child_root,
 274                                        perm, BLK_PERM_ALL, blk, errp);
 275     if (!blk->root) {
 276         bdrv_unref(bs);
 277         blk_unref(blk);
 278         return NULL;
 279     }
 280
 281     return blk;
 282 }
 283
 284 static void blk_delete(BlockBackend *blk)
 285 {
 286     assert(!blk->refcnt);
 287     assert(!blk->name);
 288     assert(!blk->dev);
 289     if (blk->public.throttle_state) {
 290         blk_io_limits_disable(blk);
 291     }
 292     if (blk->root) {
 293         blk_remove_bs(blk);
 294     }
 295     assert(QLIST_EMPTY(&blk->remove_bs_notifiers.notifiers));
 296     assert(QLIST_EMPTY(&blk->insert_bs_notifiers.notifiers));
 297     QTAILQ_REMOVE(&block_backends, blk, link);
 298     drive_info_del(blk->legacy_dinfo);
 299     block_acct_cleanup(&blk->stats);
 300     g_free(blk);
 301 }
 302
 303 static void drive_info_del(DriveInfo *dinfo)
 304 {
 305     if (!dinfo) {
 306         return;
 307     }
 308     qemu_opts_del(dinfo->opts);
 309     g_free(dinfo->serial);
 310     g_free(dinfo);
 311 }
 312
 313 int blk_get_refcnt(BlockBackend *blk)
 314 {
 315     return blk ? blk->refcnt : 0;
 316 }
 317
 318 /*
 319  * Increment @blk's reference count.
 320  * @blk must not be null.
 321  */
 322 void blk_ref(BlockBackend *blk)
 323 {
 324     blk->refcnt++;
 325 }
 326
 327 /*
 328  * Decrement @blk's reference count.
 329  * If this drops it to zero, destroy @blk.
 330  * For convenience, do nothing if @blk is null.
 331  */
 332 void blk_unref(BlockBackend *blk)
 333 {
 334     if (blk) {
 335         assert(blk->refcnt > 0);
 336         if (!--blk->refcnt) {
 337             blk_delete(blk);
 338         }
 339     }
 340 }
 341
 342 /*
 343  * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
 344  * ones which are hidden (i.e. are not referenced by the monitor).
 345  */
 346 static BlockBackend *blk_all_next(BlockBackend *blk)
 347 {
 348     return blk ? QTAILQ_NEXT(blk, link)
 349                : QTAILQ_FIRST(&block_backends);
 350 }
 351
 352 void blk_remove_all_bs(void)
 353 {
 354     BlockBackend *blk = NULL;
 355
 356     while ((blk = blk_all_next(blk)) != NULL) {
 357         AioContext *ctx = blk_get_aio_context(blk);
 358
 359         aio_context_acquire(ctx);
 360         if (blk->root) {
 361             blk_remove_bs(blk);
 362         }
 363         aio_context_release(ctx);
 364     }
 365 }
 366
 367 /*
 368  * Return the monitor-owned BlockBackend after @blk.
 369  * If @blk is null, return the first one.
 370  * Else, return @blk's next sibling, which may be null.
 371  *
 372  * To iterate over all BlockBackends, do
 373  * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
 374  *     ...
 375  * }
 376  */
 377 BlockBackend *blk_next(BlockBackend *blk)
 378 {
 379     return blk ? QTAILQ_NEXT(blk, monitor_link)
 380                : QTAILQ_FIRST(&monitor_block_backends);
 381 }
 382
 383 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
 384  * the monitor or attached to a BlockBackend */
 385 BlockDriverState *bdrv_next(BdrvNextIterator *it)
 386 {
 387     BlockDriverState *bs;
 388
 389     /* First, return all root nodes of BlockBackends. In order to avoid
 390      * returning a BDS twice when multiple BBs refer to it, we only return it
 391      * if the BB is the first one in the parent list of the BDS. */
 392     if (it->phase == BDRV_NEXT_BACKEND_ROOTS) {
 393         do {
 394             it->blk = blk_all_next(it->blk);
 395             bs = it->blk ? blk_bs(it->blk) : NULL;
 396         } while (it->blk && (bs == NULL || bdrv_first_blk(bs) != it->blk));
 397
 398         if (bs) {
 399             return bs;
 400         }
 401         it->phase = BDRV_NEXT_MONITOR_OWNED;
 402     }
 403
 404     /* Then return the monitor-owned BDSes without a BB attached. Ignore all
 405      * BDSes that are attached to a BlockBackend here; they have been handled
 406      * by the above block already */
 407     do {
 408         it->bs = bdrv_next_monitor_owned(it->bs);
 409         bs = it->bs;
 410     } while (bs && bdrv_has_blk(bs));
 411
 412     return bs;
 413 }
 414
 415 BlockDriverState *bdrv_first(BdrvNextIterator *it)
 416 {
 417     *it = (BdrvNextIterator) {
 418         .phase = BDRV_NEXT_BACKEND_ROOTS,
 419     };
 420
 421     return bdrv_next(it);
 422 }
 423
 424 /*
 425  * Add a BlockBackend into the list of backends referenced by the monitor, with
 426  * the given @name acting as the handle for the monitor.
 427  * Strictly for use by blockdev.c.
 428  *
 429  * @name must not be null or empty.
 430  *
 431  * Returns true on success and false on failure. In the latter case, an Error
 432  * object is returned through @errp.
 433  */
 434 bool monitor_add_blk(BlockBackend *blk, const char *name, Error **errp)
 435 {
 436     assert(!blk->name);
 437     assert(name && name[0]);
 438
 439     if (!id_wellformed(name)) {
 440         error_setg(errp, "Invalid device name");
 441         return false;
 442     }
 443     if (blk_by_name(name)) {
 444         error_setg(errp, "Device with id '%s' already exists", name);
 445         return false;
 446     }
 447     if (bdrv_find_node(name)) {
 448         error_setg(errp,
 449                    "Device name '%s' conflicts with an existing node name",
 450                    name);
 451         return false;
 452     }
 453
 454     blk->name = g_strdup(name);
 455     QTAILQ_INSERT_TAIL(&monitor_block_backends, blk, monitor_link);
 456     return true;
 457 }
 458
 459 /*
 460  * Remove a BlockBackend from the list of backends referenced by the monitor.
 461  * Strictly for use by blockdev.c.
 462  */
 463 void monitor_remove_blk(BlockBackend *blk)
 464 {
 465     if (!blk->name) {
 466         return;
 467     }
 468
 469     QTAILQ_REMOVE(&monitor_block_backends, blk, monitor_link);
 470     g_free(blk->name);
 471     blk->name = NULL;
 472 }
 473
 474 /*
 475  * Return @blk's name, a non-null string.
 476  * Returns an empty string iff @blk is not referenced by the monitor.
 477  */
 478 const char *blk_name(const BlockBackend *blk)
 479 {
 480     return blk->name ?: "";
 481 }
 482
 483 /*
 484  * Return the BlockBackend with name @name if it exists, else null.
 485  * @name must not be null.
 486  */
 487 BlockBackend *blk_by_name(const char *name)
 488 {
 489     BlockBackend *blk = NULL;
 490
 491     assert(name);
 492     while ((blk = blk_next(blk)) != NULL) {
 493         if (!strcmp(name, blk->name)) {
 494             return blk;
 495         }
 496     }
 497     return NULL;
 498 }
 499
 500 /*
 501  * Return the BlockDriverState attached to @blk if any, else null.
 502  */
 503 BlockDriverState *blk_bs(BlockBackend *blk)
 504 {
 505     return blk->root ? blk->root->bs : NULL;
 506 }
 507
 508 static BlockBackend *bdrv_first_blk(BlockDriverState *bs)
 509 {
 510     BdrvChild *child;
 511     QLIST_FOREACH(child, &bs->parents, next_parent) {
 512         if (child->role == &child_root) {
 513             return child->opaque;
 514         }
 515     }
 516
 517     return NULL;
 518 }
 519
 520 /*
 521  * Returns true if @bs has an associated BlockBackend.
 522  */
 523 bool bdrv_has_blk(BlockDriverState *bs)
 524 {
 525     return bdrv_first_blk(bs) != NULL;
 526 }
 527
 528 /*
 529  * Returns true if @bs has only BlockBackends as parents.
 530  */
 531 bool bdrv_is_root_node(BlockDriverState *bs)
 532 {
 533     BdrvChild *c;
 534
 535     QLIST_FOREACH(c, &bs->parents, next_parent) {
 536         if (c->role != &child_root) {
 537             return false;
 538         }
 539     }
 540
 541     return true;
 542 }
 543
 544 /*
 545  * Return @blk's DriveInfo if any, else null.
 546  */
 547 DriveInfo *blk_legacy_dinfo(BlockBackend *blk)
 548 {
 549     return blk->legacy_dinfo;
 550 }
 551
 552 /*
 553  * Set @blk's DriveInfo to @dinfo, and return it.
 554  * @blk must not have a DriveInfo set already.
 555  * No other BlockBackend may have the same DriveInfo set.
 556  */
 557 DriveInfo *blk_set_legacy_dinfo(BlockBackend *blk, DriveInfo *dinfo)
 558 {
 559     assert(!blk->legacy_dinfo);
 560     return blk->legacy_dinfo = dinfo;
 561 }
 562
 563 /*
 564  * Return the BlockBackend with DriveInfo @dinfo.
 565  * It must exist.
 566  */
 567 BlockBackend *blk_by_legacy_dinfo(DriveInfo *dinfo)
 568 {
 569     BlockBackend *blk = NULL;
 570
 571     while ((blk = blk_next(blk)) != NULL) {
 572         if (blk->legacy_dinfo == dinfo) {
 573             return blk;
 574         }
 575     }
 576     abort();
 577 }
 578
 579 /*
 580  * Returns a pointer to the publicly accessible fields of @blk.
 581  */
 582 BlockBackendPublic *blk_get_public(BlockBackend *blk)
 583 {
 584     return &blk->public;
 585 }
 586
 587 /*
 588  * Returns a BlockBackend given the associated @public fields.
 589  */
 590 BlockBackend *blk_by_public(BlockBackendPublic *public)
 591 {
 592     return container_of(public, BlockBackend, public);
 593 }
 594
 595 /*
 596  * Disassociates the currently associated BlockDriverState from @blk.
 597  */
 598 void blk_remove_bs(BlockBackend *blk)
 599 {
 600     notifier_list_notify(&blk->remove_bs_notifiers, blk);
 601     if (blk->public.throttle_state) {
 602         throttle_timers_detach_aio_context(&blk->public.throttle_timers);
 603     }
 604
 605     blk_update_root_state(blk);
 606
 607     bdrv_root_unref_child(blk->root);
 608     blk->root = NULL;
 609 }
 610
 611 /*
 612  * Associates a new BlockDriverState with @blk.
 613  */
 614 int blk_insert_bs(BlockBackend *blk, BlockDriverState *bs, Error **errp)
 615 {
 616     blk->root = bdrv_root_attach_child(bs, "root", &child_root,
 617                                        blk->perm, blk->shared_perm, blk, errp);
 618     if (blk->root == NULL) {
 619         return -EPERM;
 620     }
 621     bdrv_ref(bs);
 622
 623     notifier_list_notify(&blk->insert_bs_notifiers, blk);
 624     if (blk->public.throttle_state) {
 625         throttle_timers_attach_aio_context(
 626             &blk->public.throttle_timers, bdrv_get_aio_context(bs));
 627     }
 628
 629     return 0;
 630 }
 631
 632 /*
 633  * Sets the permission bitmasks that the user of the BlockBackend needs.
 634  */
 635 int blk_set_perm(BlockBackend *blk, uint64_t perm, uint64_t shared_perm,
 636                  Error **errp)
 637 {
 638     int ret;
 639
 640     if (blk->root && !blk->disable_perm) {
 641         ret = bdrv_child_try_set_perm(blk->root, perm, shared_perm, errp);
 642         if (ret < 0) {
 643             return ret;
 644         }
 645     }
 646
 647     blk->perm = perm;
 648     blk->shared_perm = shared_perm;
 649
 650     return 0;
 651 }
 652
 653 void blk_get_perm(BlockBackend *blk, uint64_t *perm, uint64_t *shared_perm)
 654 {
 655     *perm = blk->perm;
 656     *shared_perm = blk->shared_perm;
 657 }
 658
 659 static int blk_do_attach_dev(BlockBackend *blk, void *dev)
 660 {
 661     if (blk->dev) {
 662         return -EBUSY;
 663     }
 664
 665     /* While migration is still incoming, we don't need to apply the
 666      * permissions of guest device BlockBackends. We might still have a block
 667      * job or NBD server writing to the image for storage migration. */
 668     if (runstate_check(RUN_STATE_INMIGRATE)) {
 669         blk->disable_perm = true;
 670     }
 671
 672     blk_ref(blk);
 673     blk->dev = dev;
 674     blk->legacy_dev = false;
 675     blk_iostatus_reset(blk);
 676
 677     return 0;
 678 }
 679
 680 /*
 681  * Attach device model @dev to @blk.
 682  * Return 0 on success, -EBUSY when a device model is attached already.
 683  */
 684 int blk_attach_dev(BlockBackend *blk, DeviceState *dev)
 685 {
 686     return blk_do_attach_dev(blk, dev);
 687 }
 688
 689 /*
 690  * Attach device model @dev to @blk.
 691  * @blk must not have a device model attached already.
 692  * TODO qdevified devices don't use this, remove when devices are qdevified
 693  */
 694 void blk_attach_dev_legacy(BlockBackend *blk, void *dev)
 695 {
 696     if (blk_do_attach_dev(blk, dev) < 0) {
 697         abort();
 698     }
 699     blk->legacy_dev = true;
 700 }
 701
 702 /*
 703  * Detach device model @dev from @blk.
 704  * @dev must be currently attached to @blk.
 705  */
 706 void blk_detach_dev(BlockBackend *blk, void *dev)
 707 /* TODO change to DeviceState *dev when all users are qdevified */
 708 {
 709     assert(blk->dev == dev);
 710     blk->dev = NULL;
 711     blk->dev_ops = NULL;
 712     blk->dev_opaque = NULL;
 713     blk->guest_block_size = 512;
 714     blk_set_perm(blk, 0, BLK_PERM_ALL, &error_abort);
 715     blk_unref(blk);
 716 }
 717
 718 /*
 719  * Return the device model attached to @blk if any, else null.
 720  */
 721 void *blk_get_attached_dev(BlockBackend *blk)
 722 /* TODO change to return DeviceState * when all users are qdevified */
 723 {
 724     return blk->dev;
 725 }
 726
 727 /* Return the qdev ID, or if no ID is assigned the QOM path, of the block
 728  * device attached to the BlockBackend. */
 729 static char *blk_get_attached_dev_id(BlockBackend *blk)
 730 {
 731     DeviceState *dev;
 732
 733     assert(!blk->legacy_dev);
 734     dev = blk->dev;
 735
 736     if (!dev) {
 737         return g_strdup("");
 738     } else if (dev->id) {
 739         return g_strdup(dev->id);
 740     }
 741     return object_get_canonical_path(OBJECT(dev));
 742 }
 743
 744 /*
 745  * Return the BlockBackend which has the device model @dev attached if it
 746  * exists, else null.
 747  *
 748  * @dev must not be null.
 749  */
 750 BlockBackend *blk_by_dev(void *dev)
 751 {
 752     BlockBackend *blk = NULL;
 753
 754     assert(dev != NULL);
 755     while ((blk = blk_all_next(blk)) != NULL) {
 756         if (blk->dev == dev) {
 757             return blk;
 758         }
 759     }
 760     return NULL;
 761 }
 762
 763 /*
 764  * Set @blk's device model callbacks to @ops.
 765  * @opaque is the opaque argument to pass to the callbacks.
 766  * This is for use by device models.
 767  */
 768 void blk_set_dev_ops(BlockBackend *blk, const BlockDevOps *ops,
 769                      void *opaque)
 770 {
 771     /* All drivers that use blk_set_dev_ops() are qdevified and we want to keep
 772      * it that way, so we can assume blk->dev, if present, is a DeviceState if
 773      * blk->dev_ops is set. Non-device users may use dev_ops without device. */
 774     assert(!blk->legacy_dev);
 775
 776     blk->dev_ops = ops;
 777     blk->dev_opaque = opaque;
 778
 779     /* Are we currently quiesced? Should we enforce this right now? */
 780     if (blk->quiesce_counter && ops->drained_begin) {
 781         ops->drained_begin(opaque);
 782     }
 783 }
 784
 785 /*
 786  * Notify @blk's attached device model of media change.
 787  *
 788  * If @load is true, notify of media load. This action can fail, meaning that
 789  * the medium cannot be loaded. @errp is set then.
 790  *
 791  * If @load is false, notify of media eject. This can never fail.
 792  *
 793  * Also send DEVICE_TRAY_MOVED events as appropriate.
 794  */
 795 void blk_dev_change_media_cb(BlockBackend *blk, bool load, Error **errp)
 796 {
 797     if (blk->dev_ops && blk->dev_ops->change_media_cb) {
 798         bool tray_was_open, tray_is_open;
 799         Error *local_err = NULL;
 800
 801         assert(!blk->legacy_dev);
 802
 803         tray_was_open = blk_dev_is_tray_open(blk);
 804         blk->dev_ops->change_media_cb(blk->dev_opaque, load, &local_err);
 805         if (local_err) {
 806             assert(load == true);
 807             error_propagate(errp, local_err);
 808             return;
 809         }
 810         tray_is_open = blk_dev_is_tray_open(blk);
 811
 812         if (tray_was_open != tray_is_open) {
 813             char *id = blk_get_attached_dev_id(blk);
 814             qapi_event_send_device_tray_moved(blk_name(blk), id, tray_is_open,
 815                                               &error_abort);
 816             g_free(id);
 817         }
 818     }
 819 }
 820
 821 static void blk_root_change_media(BdrvChild *child, bool load)
 822 {
 823     blk_dev_change_media_cb(child->opaque, load, NULL);
 824 }
 825
 826 /*
 827  * Does @blk's attached device model have removable media?
 828  * %true if no device model is attached.
 829  */
 830 bool blk_dev_has_removable_media(BlockBackend *blk)
 831 {
 832     return !blk->dev || (blk->dev_ops && blk->dev_ops->change_media_cb);
 833 }
 834
 835 /*
 836  * Does @blk's attached device model have a tray?
 837  */
 838 bool blk_dev_has_tray(BlockBackend *blk)
 839 {
 840     return blk->dev_ops && blk->dev_ops->is_tray_open;
 841 }
 842
 843 /*
 844  * Notify @blk's attached device model of a media eject request.
 845  * If @force is true, the medium is about to be yanked out forcefully.
 846  */
 847 void blk_dev_eject_request(BlockBackend *blk, bool force)
 848 {
 849     if (blk->dev_ops && blk->dev_ops->eject_request_cb) {
 850         blk->dev_ops->eject_request_cb(blk->dev_opaque, force);
 851     }
 852 }
 853
 854 /*
 855  * Does @blk's attached device model have a tray, and is it open?
 856  */
 857 bool blk_dev_is_tray_open(BlockBackend *blk)
 858 {
 859     if (blk_dev_has_tray(blk)) {
 860         return blk->dev_ops->is_tray_open(blk->dev_opaque);
 861     }
 862     return false;
 863 }
 864
 865 /*
 866  * Does @blk's attached device model have the medium locked?
 867  * %false if the device model has no such lock.
 868  */
 869 bool blk_dev_is_medium_locked(BlockBackend *blk)
 870 {
 871     if (blk->dev_ops && blk->dev_ops->is_medium_locked) {
 872         return blk->dev_ops->is_medium_locked(blk->dev_opaque);
 873     }
 874     return false;
 875 }
 876
 877 /*
 878  * Notify @blk's attached device model of a backend size change.
 879  */
 880 static void blk_root_resize(BdrvChild *child)
 881 {
 882     BlockBackend *blk = child->opaque;
 883
 884     if (blk->dev_ops && blk->dev_ops->resize_cb) {
 885         blk->dev_ops->resize_cb(blk->dev_opaque);
 886     }
 887 }
 888
 889 void blk_iostatus_enable(BlockBackend *blk)
 890 {
 891     blk->iostatus_enabled = true;
 892     blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 893 }
 894
 895 /* The I/O status is only enabled if the drive explicitly
 896  * enables it _and_ the VM is configured to stop on errors */
 897 bool blk_iostatus_is_enabled(const BlockBackend *blk)
 898 {
 899     return (blk->iostatus_enabled &&
 900            (blk->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
 901             blk->on_write_error == BLOCKDEV_ON_ERROR_STOP   ||
 902             blk->on_read_error == BLOCKDEV_ON_ERROR_STOP));
 903 }
 904
 905 BlockDeviceIoStatus blk_iostatus(const BlockBackend *blk)
 906 {
 907     return blk->iostatus;
 908 }
 909
 910 void blk_iostatus_disable(BlockBackend *blk)
 911 {
 912     blk->iostatus_enabled = false;
 913 }
 914
 915 void blk_iostatus_reset(BlockBackend *blk)
 916 {
 917     if (blk_iostatus_is_enabled(blk)) {
 918         BlockDriverState *bs = blk_bs(blk);
 919         blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 920         if (bs && bs->job) {
 921             block_job_iostatus_reset(bs->job);
 922         }
 923     }
 924 }
 925
 926 void blk_iostatus_set_err(BlockBackend *blk, int error)
 927 {
 928     assert(blk_iostatus_is_enabled(blk));
 929     if (blk->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 930         blk->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 931                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 932     }
 933 }
 934
 935 void blk_set_allow_write_beyond_eof(BlockBackend *blk, bool allow)
 936 {
 937     blk->allow_write_beyond_eof = allow;
 938 }
 939
 940 static int blk_check_byte_request(BlockBackend *blk, int64_t offset,
 941                                   size_t size)
 942 {
 943     int64_t len;
 944
 945     if (size > INT_MAX) {
 946         return -EIO;
 947     }
 948
 949     if (!blk_is_available(blk)) {
 950         return -ENOMEDIUM;
 951     }
 952
 953     if (offset < 0) {
 954         return -EIO;
 955     }
 956
 957     if (!blk->allow_write_beyond_eof) {
 958         len = blk_getlength(blk);
 959         if (len < 0) {
 960             return len;
 961         }
 962
 963         if (offset > len || len - offset < size) {
 964             return -EIO;
 965         }
 966     }
 967
 968     return 0;
 969 }
 970
 971 int coroutine_fn blk_co_preadv(BlockBackend *blk, int64_t offset,
 972                                unsigned int bytes, QEMUIOVector *qiov,
 973                                BdrvRequestFlags flags)
 974 {
 975     int ret;
 976     BlockDriverState *bs = blk_bs(blk);
 977
 978     trace_blk_co_preadv(blk, bs, offset, bytes, flags);
 979
 980     ret = blk_check_byte_request(blk, offset, bytes);
 981     if (ret < 0) {
 982         return ret;
 983     }
 984
 985     bdrv_inc_in_flight(bs);
 986
 987     /* throttling disk I/O */
 988     if (blk->public.throttle_state) {
 989         throttle_group_co_io_limits_intercept(blk, bytes, false);
 990     }
 991
 992     ret = bdrv_co_preadv(blk->root, offset, bytes, qiov, flags);
 993     bdrv_dec_in_flight(bs);
 994     return ret;
 995 }
 996
 997 int coroutine_fn blk_co_pwritev(BlockBackend *blk, int64_t offset,
 998                                 unsigned int bytes, QEMUIOVector *qiov,
 999                                 BdrvRequestFlags flags)
1000 {
1001     int ret;
1002     BlockDriverState *bs = blk_bs(blk);
1003
1004     trace_blk_co_pwritev(blk, bs, offset, bytes, flags);
1005
1006     ret = blk_check_byte_request(blk, offset, bytes);
1007     if (ret < 0) {
1008         return ret;
1009     }
1010
1011     bdrv_inc_in_flight(bs);
1012
1013     /* throttling disk I/O */
1014     if (blk->public.throttle_state) {
1015         throttle_group_co_io_limits_intercept(blk, bytes, true);
1016     }
1017
1018     if (!blk->enable_write_cache) {
1019         flags |= BDRV_REQ_FUA;
1020     }
1021
1022     ret = bdrv_co_pwritev(blk->root, offset, bytes, qiov, flags);
1023     bdrv_dec_in_flight(bs);
1024     return ret;
1025 }
1026
1027 typedef struct BlkRwCo {
1028     BlockBackend *blk;
1029     int64_t offset;
1030     QEMUIOVector *qiov;
1031     int ret;
1032     BdrvRequestFlags flags;
1033 } BlkRwCo;
1034
1035 static void blk_read_entry(void *opaque)
1036 {
1037     BlkRwCo *rwco = opaque;
1038
1039     rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, rwco->qiov->size,
1040                               rwco->qiov, rwco->flags);
1041 }
1042
1043 static void blk_write_entry(void *opaque)
1044 {
1045     BlkRwCo *rwco = opaque;
1046
1047     rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, rwco->qiov->size,
1048                                rwco->qiov, rwco->flags);
1049 }
1050
1051 static int blk_prw(BlockBackend *blk, int64_t offset, uint8_t *buf,
1052                    int64_t bytes, CoroutineEntry co_entry,
1053                    BdrvRequestFlags flags)
1054 {
1055     QEMUIOVector qiov;
1056     struct iovec iov;
1057     BlkRwCo rwco;
1058
1059     iov = (struct iovec) {
1060         .iov_base = buf,
1061         .iov_len = bytes,
1062     };
1063     qemu_iovec_init_external(&qiov, &iov, 1);
1064
1065     rwco = (BlkRwCo) {
1066         .blk    = blk,
1067         .offset = offset,
1068         .qiov   = &qiov,
1069         .flags  = flags,
1070         .ret    = NOT_DONE,
1071     };
1072
1073     if (qemu_in_coroutine()) {
1074         /* Fast-path if already in coroutine context */
1075         co_entry(&rwco);
1076     } else {
1077         Coroutine *co = qemu_coroutine_create(co_entry, &rwco);
1078         bdrv_coroutine_enter(blk_bs(blk), co);
1079         BDRV_POLL_WHILE(blk_bs(blk), rwco.ret == NOT_DONE);
1080     }
1081
1082     return rwco.ret;
1083 }
1084
1085 int blk_pread_unthrottled(BlockBackend *blk, int64_t offset, uint8_t *buf,
1086                           int count)
1087 {
1088     int ret;
1089
1090     ret = blk_check_byte_request(blk, offset, count);
1091     if (ret < 0) {
1092         return ret;
1093     }
1094
1095     blk_root_drained_begin(blk->root);
1096     ret = blk_pread(blk, offset, buf, count);
1097     blk_root_drained_end(blk->root);
1098     return ret;
1099 }
1100
1101 int blk_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1102                       int bytes, BdrvRequestFlags flags)
1103 {
1104     return blk_prw(blk, offset, NULL, bytes, blk_write_entry,
1105                    flags | BDRV_REQ_ZERO_WRITE);
1106 }
1107
1108 int blk_make_zero(BlockBackend *blk, BdrvRequestFlags flags)
1109 {
1110     return bdrv_make_zero(blk->root, flags);
1111 }
1112
1113 static void error_callback_bh(void *opaque)
1114 {
1115     struct BlockBackendAIOCB *acb = opaque;
1116
1117     bdrv_dec_in_flight(acb->common.bs);
1118     acb->common.cb(acb->common.opaque, acb->ret);
1119     qemu_aio_unref(acb);
1120 }
1121
1122 BlockAIOCB *blk_abort_aio_request(BlockBackend *blk,
1123                                   BlockCompletionFunc *cb,
1124                                   void *opaque, int ret)
1125 {
1126     struct BlockBackendAIOCB *acb;
1127
1128     bdrv_inc_in_flight(blk_bs(blk));
1129     acb = blk_aio_get(&block_backend_aiocb_info, blk, cb, opaque);
1130     acb->blk = blk;
1131     acb->ret = ret;
1132
1133     aio_bh_schedule_oneshot(blk_get_aio_context(blk), error_callback_bh, acb);
1134     return &acb->common;
1135 }
1136
1137 typedef struct BlkAioEmAIOCB {
1138     BlockAIOCB common;
1139     BlkRwCo rwco;
1140     int bytes;
1141     bool has_returned;
1142 } BlkAioEmAIOCB;
1143
1144 static const AIOCBInfo blk_aio_em_aiocb_info = {
1145     .aiocb_size         = sizeof(BlkAioEmAIOCB),
1146 };
1147
1148 static void blk_aio_complete(BlkAioEmAIOCB *acb)
1149 {
1150     if (acb->has_returned) {
1151         bdrv_dec_in_flight(acb->common.bs);
1152         acb->common.cb(acb->common.opaque, acb->rwco.ret);
1153         qemu_aio_unref(acb);
1154     }
1155 }
1156
1157 static void blk_aio_complete_bh(void *opaque)
1158 {
1159     BlkAioEmAIOCB *acb = opaque;
1160     assert(acb->has_returned);
1161     blk_aio_complete(acb);
1162 }
1163
1164 static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset, int bytes,
1165                                 QEMUIOVector *qiov, CoroutineEntry co_entry,
1166                                 BdrvRequestFlags flags,
1167                                 BlockCompletionFunc *cb, void *opaque)
1168 {
1169     BlkAioEmAIOCB *acb;
1170     Coroutine *co;
1171
1172     bdrv_inc_in_flight(blk_bs(blk));
1173     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1174     acb->rwco = (BlkRwCo) {
1175         .blk    = blk,
1176         .offset = offset,
1177         .qiov   = qiov,
1178         .flags  = flags,
1179         .ret    = NOT_DONE,
1180     };
1181     acb->bytes = bytes;
1182     acb->has_returned = false;
1183
1184     co = qemu_coroutine_create(co_entry, acb);
1185     bdrv_coroutine_enter(blk_bs(blk), co);
1186
1187     acb->has_returned = true;
1188     if (acb->rwco.ret != NOT_DONE) {
1189         aio_bh_schedule_oneshot(blk_get_aio_context(blk),
1190                                 blk_aio_complete_bh, acb);
1191     }
1192
1193     return &acb->common;
1194 }
1195
1196 static void blk_aio_read_entry(void *opaque)
1197 {
1198     BlkAioEmAIOCB *acb = opaque;
1199     BlkRwCo *rwco = &acb->rwco;
1200
1201     assert(rwco->qiov->size == acb->bytes);
1202     rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, acb->bytes,
1203                               rwco->qiov, rwco->flags);
1204     blk_aio_complete(acb);
1205 }
1206
1207 static void blk_aio_write_entry(void *opaque)
1208 {
1209     BlkAioEmAIOCB *acb = opaque;
1210     BlkRwCo *rwco = &acb->rwco;
1211
1212     assert(!rwco->qiov || rwco->qiov->size == acb->bytes);
1213     rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, acb->bytes,
1214                                rwco->qiov, rwco->flags);
1215     blk_aio_complete(acb);
1216 }
1217
1218 BlockAIOCB *blk_aio_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1219                                   int count, BdrvRequestFlags flags,
1220                                   BlockCompletionFunc *cb, void *opaque)
1221 {
1222     return blk_aio_prwv(blk, offset, count, NULL, blk_aio_write_entry,
1223                         flags | BDRV_REQ_ZERO_WRITE, cb, opaque);
1224 }
1225
1226 int blk_pread(BlockBackend *blk, int64_t offset, void *buf, int count)
1227 {
1228     int ret = blk_prw(blk, offset, buf, count, blk_read_entry, 0);
1229     if (ret < 0) {
1230         return ret;
1231     }
1232     return count;
1233 }
1234
1235 int blk_pwrite(BlockBackend *blk, int64_t offset, const void *buf, int count,
1236                BdrvRequestFlags flags)
1237 {
1238     int ret = blk_prw(blk, offset, (void *) buf, count, blk_write_entry,
1239                       flags);
1240     if (ret < 0) {
1241         return ret;
1242     }
1243     return count;
1244 }
1245
1246 int64_t blk_getlength(BlockBackend *blk)
1247 {
1248     if (!blk_is_available(blk)) {
1249         return -ENOMEDIUM;
1250     }
1251
1252     return bdrv_getlength(blk_bs(blk));
1253 }
1254
1255 void blk_get_geometry(BlockBackend *blk, uint64_t *nb_sectors_ptr)
1256 {
1257     if (!blk_bs(blk)) {
1258         *nb_sectors_ptr = 0;
1259     } else {
1260         bdrv_get_geometry(blk_bs(blk), nb_sectors_ptr);
1261     }
1262 }
1263
1264 int64_t blk_nb_sectors(BlockBackend *blk)
1265 {
1266     if (!blk_is_available(blk)) {
1267         return -ENOMEDIUM;
1268     }
1269
1270     return bdrv_nb_sectors(blk_bs(blk));
1271 }
1272
1273 BlockAIOCB *blk_aio_preadv(BlockBackend *blk, int64_t offset,
1274                            QEMUIOVector *qiov, BdrvRequestFlags flags,
1275                            BlockCompletionFunc *cb, void *opaque)
1276 {
1277     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1278                         blk_aio_read_entry, flags, cb, opaque);
1279 }
1280
1281 BlockAIOCB *blk_aio_pwritev(BlockBackend *blk, int64_t offset,
1282                             QEMUIOVector *qiov, BdrvRequestFlags flags,
1283                             BlockCompletionFunc *cb, void *opaque)
1284 {
1285     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1286                         blk_aio_write_entry, flags, cb, opaque);
1287 }
1288
1289 static void blk_aio_flush_entry(void *opaque)
1290 {
1291     BlkAioEmAIOCB *acb = opaque;
1292     BlkRwCo *rwco = &acb->rwco;
1293
1294     rwco->ret = blk_co_flush(rwco->blk);
1295     blk_aio_complete(acb);
1296 }
1297
1298 BlockAIOCB *blk_aio_flush(BlockBackend *blk,
1299                           BlockCompletionFunc *cb, void *opaque)
1300 {
1301     return blk_aio_prwv(blk, 0, 0, NULL, blk_aio_flush_entry, 0, cb, opaque);
1302 }
1303
1304 static void blk_aio_pdiscard_entry(void *opaque)
1305 {
1306     BlkAioEmAIOCB *acb = opaque;
1307     BlkRwCo *rwco = &acb->rwco;
1308
1309     rwco->ret = blk_co_pdiscard(rwco->blk, rwco->offset, acb->bytes);
1310     blk_aio_complete(acb);
1311 }
1312
1313 BlockAIOCB *blk_aio_pdiscard(BlockBackend *blk,
1314                              int64_t offset, int bytes,
1315                              BlockCompletionFunc *cb, void *opaque)
1316 {
1317     return blk_aio_prwv(blk, offset, bytes, NULL, blk_aio_pdiscard_entry, 0,
1318                         cb, opaque);
1319 }
1320
1321 void blk_aio_cancel(BlockAIOCB *acb)
1322 {
1323     bdrv_aio_cancel(acb);
1324 }
1325
1326 void blk_aio_cancel_async(BlockAIOCB *acb)
1327 {
1328     bdrv_aio_cancel_async(acb);
1329 }
1330
1331 int blk_co_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1332 {
1333     if (!blk_is_available(blk)) {
1334         return -ENOMEDIUM;
1335     }
1336
1337     return bdrv_co_ioctl(blk_bs(blk), req, buf);
1338 }
1339
1340 static void blk_ioctl_entry(void *opaque)
1341 {
1342     BlkRwCo *rwco = opaque;
1343     rwco->ret = blk_co_ioctl(rwco->blk, rwco->offset,
1344                              rwco->qiov->iov[0].iov_base);
1345 }
1346
1347 int blk_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1348 {
1349     return blk_prw(blk, req, buf, 0, blk_ioctl_entry, 0);
1350 }
1351
1352 static void blk_aio_ioctl_entry(void *opaque)
1353 {
1354     BlkAioEmAIOCB *acb = opaque;
1355     BlkRwCo *rwco = &acb->rwco;
1356
1357     rwco->ret = blk_co_ioctl(rwco->blk, rwco->offset,
1358                              rwco->qiov->iov[0].iov_base);
1359     blk_aio_complete(acb);
1360 }
1361
1362 BlockAIOCB *blk_aio_ioctl(BlockBackend *blk, unsigned long int req, void *buf,
1363                           BlockCompletionFunc *cb, void *opaque)
1364 {
1365     QEMUIOVector qiov;
1366     struct iovec iov;
1367
1368     iov = (struct iovec) {
1369         .iov_base = buf,
1370         .iov_len = 0,
1371     };
1372     qemu_iovec_init_external(&qiov, &iov, 1);
1373
1374     return blk_aio_prwv(blk, req, 0, &qiov, blk_aio_ioctl_entry, 0, cb, opaque);
1375 }
1376
1377 int blk_co_pdiscard(BlockBackend *blk, int64_t offset, int bytes)
1378 {
1379     int ret = blk_check_byte_request(blk, offset, bytes);
1380     if (ret < 0) {
1381         return ret;
1382     }
1383
1384     return bdrv_co_pdiscard(blk_bs(blk), offset, bytes);
1385 }
1386
1387 int blk_co_flush(BlockBackend *blk)
1388 {
1389     if (!blk_is_available(blk)) {
1390         return -ENOMEDIUM;
1391     }
1392
1393     return bdrv_co_flush(blk_bs(blk));
1394 }
1395
1396 static void blk_flush_entry(void *opaque)
1397 {
1398     BlkRwCo *rwco = opaque;
1399     rwco->ret = blk_co_flush(rwco->blk);
1400 }
1401
1402 int blk_flush(BlockBackend *blk)
1403 {
1404     return blk_prw(blk, 0, NULL, 0, blk_flush_entry, 0);
1405 }
1406
1407 void blk_drain(BlockBackend *blk)
1408 {
1409     if (blk_bs(blk)) {
1410         bdrv_drain(blk_bs(blk));
1411     }
1412 }
1413
1414 void blk_drain_all(void)
1415 {
1416     bdrv_drain_all();
1417 }
1418
1419 void blk_set_on_error(BlockBackend *blk, BlockdevOnError on_read_error,
1420                       BlockdevOnError on_write_error)
1421 {
1422     blk->on_read_error = on_read_error;
1423     blk->on_write_error = on_write_error;
1424 }
1425
1426 BlockdevOnError blk_get_on_error(BlockBackend *blk, bool is_read)
1427 {
1428     return is_read ? blk->on_read_error : blk->on_write_error;
1429 }
1430
1431 BlockErrorAction blk_get_error_action(BlockBackend *blk, bool is_read,
1432                                       int error)
1433 {
1434     BlockdevOnError on_err = blk_get_on_error(blk, is_read);
1435
1436     switch (on_err) {
1437     case BLOCKDEV_ON_ERROR_ENOSPC:
1438         return (error == ENOSPC) ?
1439                BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
1440     case BLOCKDEV_ON_ERROR_STOP:
1441         return BLOCK_ERROR_ACTION_STOP;
1442     case BLOCKDEV_ON_ERROR_REPORT:
1443         return BLOCK_ERROR_ACTION_REPORT;
1444     case BLOCKDEV_ON_ERROR_IGNORE:
1445         return BLOCK_ERROR_ACTION_IGNORE;
1446     case BLOCKDEV_ON_ERROR_AUTO:
1447     default:
1448         abort();
1449     }
1450 }
1451
1452 static void send_qmp_error_event(BlockBackend *blk,
1453                                  BlockErrorAction action,
1454                                  bool is_read, int error)
1455 {
1456     IoOperationType optype;
1457
1458     optype = is_read ? IO_OPERATION_TYPE_READ : IO_OPERATION_TYPE_WRITE;
1459     qapi_event_send_block_io_error(blk_name(blk),
1460                                    bdrv_get_node_name(blk_bs(blk)), optype,
1461                                    action, blk_iostatus_is_enabled(blk),
1462                                    error == ENOSPC, strerror(error),
1463                                    &error_abort);
1464 }
1465
1466 /* This is done by device models because, while the block layer knows
1467  * about the error, it does not know whether an operation comes from
1468  * the device or the block layer (from a job, for example).
1469  */
1470 void blk_error_action(BlockBackend *blk, BlockErrorAction action,
1471                       bool is_read, int error)
1472 {
1473     assert(error >= 0);
1474
1475     if (action == BLOCK_ERROR_ACTION_STOP) {
1476         /* First set the iostatus, so that "info block" returns an iostatus
1477          * that matches the events raised so far (an additional error iostatus
1478          * is fine, but not a lost one).
1479          */
1480         blk_iostatus_set_err(blk, error);
1481
1482         /* Then raise the request to stop the VM and the event.
1483          * qemu_system_vmstop_request_prepare has two effects.  First,
1484          * it ensures that the STOP event always comes after the
1485          * BLOCK_IO_ERROR event.  Second, it ensures that even if management
1486          * can observe the STOP event and do a "cont" before the STOP
1487          * event is issued, the VM will not stop.  In this case, vm_start()
1488          * also ensures that the STOP/RESUME pair of events is emitted.
1489          */
1490         qemu_system_vmstop_request_prepare();
1491         send_qmp_error_event(blk, action, is_read, error);
1492         qemu_system_vmstop_request(RUN_STATE_IO_ERROR);
1493     } else {
1494         send_qmp_error_event(blk, action, is_read, error);
1495     }
1496 }
1497
1498 int blk_is_read_only(BlockBackend *blk)
1499 {
1500     BlockDriverState *bs = blk_bs(blk);
1501
1502     if (bs) {
1503         return bdrv_is_read_only(bs);
1504     } else {
1505         return blk->root_state.read_only;
1506     }
1507 }
1508
1509 int blk_is_sg(BlockBackend *blk)
1510 {
1511     BlockDriverState *bs = blk_bs(blk);
1512
1513     if (!bs) {
1514         return 0;
1515     }
1516
1517     return bdrv_is_sg(bs);
1518 }
1519
1520 int blk_enable_write_cache(BlockBackend *blk)
1521 {
1522     return blk->enable_write_cache;
1523 }
1524
1525 void blk_set_enable_write_cache(BlockBackend *blk, bool wce)
1526 {
1527     blk->enable_write_cache = wce;
1528 }
1529
1530 void blk_invalidate_cache(BlockBackend *blk, Error **errp)
1531 {
1532     BlockDriverState *bs = blk_bs(blk);
1533
1534     if (!bs) {
1535         error_setg(errp, "Device '%s' has no medium", blk->name);
1536         return;
1537     }
1538
1539     bdrv_invalidate_cache(bs, errp);
1540 }
1541
1542 bool blk_is_inserted(BlockBackend *blk)
1543 {
1544     BlockDriverState *bs = blk_bs(blk);
1545
1546     return bs && bdrv_is_inserted(bs);
1547 }
1548
1549 bool blk_is_available(BlockBackend *blk)
1550 {
1551     return blk_is_inserted(blk) && !blk_dev_is_tray_open(blk);
1552 }
1553
1554 void blk_lock_medium(BlockBackend *blk, bool locked)
1555 {
1556     BlockDriverState *bs = blk_bs(blk);
1557
1558     if (bs) {
1559         bdrv_lock_medium(bs, locked);
1560     }
1561 }
1562
1563 void blk_eject(BlockBackend *blk, bool eject_flag)
1564 {
1565     BlockDriverState *bs = blk_bs(blk);
1566     char *id;
1567
1568     /* blk_eject is only called by qdevified devices */
1569     assert(!blk->legacy_dev);
1570
1571     if (bs) {
1572         bdrv_eject(bs, eject_flag);
1573     }
1574
1575     /* Whether or not we ejected on the backend,
1576      * the frontend experienced a tray event. */
1577     id = blk_get_attached_dev_id(blk);
1578     qapi_event_send_device_tray_moved(blk_name(blk), id,
1579                                       eject_flag, &error_abort);
1580     g_free(id);
1581 }
1582
1583 int blk_get_flags(BlockBackend *blk)
1584 {
1585     BlockDriverState *bs = blk_bs(blk);
1586
1587     if (bs) {
1588         return bdrv_get_flags(bs);
1589     } else {
1590         return blk->root_state.open_flags;
1591     }
1592 }
1593
1594 /* Returns the maximum transfer length, in bytes; guaranteed nonzero */
1595 uint32_t blk_get_max_transfer(BlockBackend *blk)
1596 {
1597     BlockDriverState *bs = blk_bs(blk);
1598     uint32_t max = 0;
1599
1600     if (bs) {
1601         max = bs->bl.max_transfer;
1602     }
1603     return MIN_NON_ZERO(max, INT_MAX);
1604 }
1605
1606 int blk_get_max_iov(BlockBackend *blk)
1607 {
1608     return blk->root->bs->bl.max_iov;
1609 }
1610
1611 void blk_set_guest_block_size(BlockBackend *blk, int align)
1612 {
1613     blk->guest_block_size = align;
1614 }
1615
1616 void *blk_try_blockalign(BlockBackend *blk, size_t size)
1617 {
1618     return qemu_try_blockalign(blk ? blk_bs(blk) : NULL, size);
1619 }
1620
1621 void *blk_blockalign(BlockBackend *blk, size_t size)
1622 {
1623     return qemu_blockalign(blk ? blk_bs(blk) : NULL, size);
1624 }
1625
1626 bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp)
1627 {
1628     BlockDriverState *bs = blk_bs(blk);
1629
1630     if (!bs) {
1631         return false;
1632     }
1633
1634     return bdrv_op_is_blocked(bs, op, errp);
1635 }
1636
1637 void blk_op_unblock(BlockBackend *blk, BlockOpType op, Error *reason)
1638 {
1639     BlockDriverState *bs = blk_bs(blk);
1640
1641     if (bs) {
1642         bdrv_op_unblock(bs, op, reason);
1643     }
1644 }
1645
1646 void blk_op_block_all(BlockBackend *blk, Error *reason)
1647 {
1648     BlockDriverState *bs = blk_bs(blk);
1649
1650     if (bs) {
1651         bdrv_op_block_all(bs, reason);
1652     }
1653 }
1654
1655 void blk_op_unblock_all(BlockBackend *blk, Error *reason)
1656 {
1657     BlockDriverState *bs = blk_bs(blk);
1658
1659     if (bs) {
1660         bdrv_op_unblock_all(bs, reason);
1661     }
1662 }
1663
1664 AioContext *blk_get_aio_context(BlockBackend *blk)
1665 {
1666     BlockDriverState *bs = blk_bs(blk);
1667
1668     if (bs) {
1669         return bdrv_get_aio_context(bs);
1670     } else {
1671         return qemu_get_aio_context();
1672     }
1673 }
1674
1675 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb)
1676 {
1677     BlockBackendAIOCB *blk_acb = DO_UPCAST(BlockBackendAIOCB, common, acb);
1678     return blk_get_aio_context(blk_acb->blk);
1679 }
1680
1681 void blk_set_aio_context(BlockBackend *blk, AioContext *new_context)
1682 {
1683     BlockDriverState *bs = blk_bs(blk);
1684
1685     if (bs) {
1686         if (blk->public.throttle_state) {
1687             throttle_timers_detach_aio_context(&blk->public.throttle_timers);
1688         }
1689         bdrv_set_aio_context(bs, new_context);
1690         if (blk->public.throttle_state) {
1691             throttle_timers_attach_aio_context(&blk->public.throttle_timers,
1692                                                new_context);
1693         }
1694     }
1695 }
1696
1697 void blk_add_aio_context_notifier(BlockBackend *blk,
1698         void (*attached_aio_context)(AioContext *new_context, void *opaque),
1699         void (*detach_aio_context)(void *opaque), void *opaque)
1700 {
1701     BlockDriverState *bs = blk_bs(blk);
1702
1703     if (bs) {
1704         bdrv_add_aio_context_notifier(bs, attached_aio_context,
1705                                       detach_aio_context, opaque);
1706     }
1707 }
1708
1709 void blk_remove_aio_context_notifier(BlockBackend *blk,
1710                                      void (*attached_aio_context)(AioContext *,
1711                                                                   void *),
1712                                      void (*detach_aio_context)(void *),
1713                                      void *opaque)
1714 {
1715     BlockDriverState *bs = blk_bs(blk);
1716
1717     if (bs) {
1718         bdrv_remove_aio_context_notifier(bs, attached_aio_context,
1719                                          detach_aio_context, opaque);
1720     }
1721 }
1722
1723 void blk_add_remove_bs_notifier(BlockBackend *blk, Notifier *notify)
1724 {
1725     notifier_list_add(&blk->remove_bs_notifiers, notify);
1726 }
1727
1728 void blk_add_insert_bs_notifier(BlockBackend *blk, Notifier *notify)
1729 {
1730     notifier_list_add(&blk->insert_bs_notifiers, notify);
1731 }
1732
1733 void blk_io_plug(BlockBackend *blk)
1734 {
1735     BlockDriverState *bs = blk_bs(blk);
1736
1737     if (bs) {
1738         bdrv_io_plug(bs);
1739     }
1740 }
1741
1742 void blk_io_unplug(BlockBackend *blk)
1743 {
1744     BlockDriverState *bs = blk_bs(blk);
1745
1746     if (bs) {
1747         bdrv_io_unplug(bs);
1748     }
1749 }
1750
1751 BlockAcctStats *blk_get_stats(BlockBackend *blk)
1752 {
1753     return &blk->stats;
1754 }
1755
1756 void *blk_aio_get(const AIOCBInfo *aiocb_info, BlockBackend *blk,
1757                   BlockCompletionFunc *cb, void *opaque)
1758 {
1759     return qemu_aio_get(aiocb_info, blk_bs(blk), cb, opaque);
1760 }
1761
1762 int coroutine_fn blk_co_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1763                                       int bytes, BdrvRequestFlags flags)
1764 {
1765     return blk_co_pwritev(blk, offset, bytes, NULL,
1766                           flags | BDRV_REQ_ZERO_WRITE);
1767 }
1768
1769 int blk_pwrite_compressed(BlockBackend *blk, int64_t offset, const void *buf,
1770                           int count)
1771 {
1772     return blk_prw(blk, offset, (void *) buf, count, blk_write_entry,
1773                    BDRV_REQ_WRITE_COMPRESSED);
1774 }
1775
1776 int blk_truncate(BlockBackend *blk, int64_t offset, PreallocMode prealloc,
1777                  Error **errp)
1778 {
1779     if (!blk_is_available(blk)) {
1780         error_setg(errp, "No medium inserted");
1781         return -ENOMEDIUM;
1782     }
1783
1784     return bdrv_truncate(blk->root, offset, prealloc, errp);
1785 }
1786
1787 static void blk_pdiscard_entry(void *opaque)
1788 {
1789     BlkRwCo *rwco = opaque;
1790     rwco->ret = blk_co_pdiscard(rwco->blk, rwco->offset, rwco->qiov->size);
1791 }
1792
1793 int blk_pdiscard(BlockBackend *blk, int64_t offset, int bytes)
1794 {
1795     return blk_prw(blk, offset, NULL, bytes, blk_pdiscard_entry, 0);
1796 }
1797
1798 int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
1799                      int64_t pos, int size)
1800 {
1801     int ret;
1802
1803     if (!blk_is_available(blk)) {
1804         return -ENOMEDIUM;
1805     }
1806
1807     ret = bdrv_save_vmstate(blk_bs(blk), buf, pos, size);
1808     if (ret < 0) {
1809         return ret;
1810     }
1811
1812     if (ret == size && !blk->enable_write_cache) {
1813         ret = bdrv_flush(blk_bs(blk));
1814     }
1815
1816     return ret < 0 ? ret : size;
1817 }
1818
1819 int blk_load_vmstate(BlockBackend *blk, uint8_t *buf, int64_t pos, int size)
1820 {
1821     if (!blk_is_available(blk)) {
1822         return -ENOMEDIUM;
1823     }
1824
1825     return bdrv_load_vmstate(blk_bs(blk), buf, pos, size);
1826 }
1827
1828 int blk_probe_blocksizes(BlockBackend *blk, BlockSizes *bsz)
1829 {
1830     if (!blk_is_available(blk)) {
1831         return -ENOMEDIUM;
1832     }
1833
1834     return bdrv_probe_blocksizes(blk_bs(blk), bsz);
1835 }
1836
1837 int blk_probe_geometry(BlockBackend *blk, HDGeometry *geo)
1838 {
1839     if (!blk_is_available(blk)) {
1840         return -ENOMEDIUM;
1841     }
1842
1843     return bdrv_probe_geometry(blk_bs(blk), geo);
1844 }
1845
1846 /*
1847  * Updates the BlockBackendRootState object with data from the currently
1848  * attached BlockDriverState.
1849  */
1850 void blk_update_root_state(BlockBackend *blk)
1851 {
1852     assert(blk->root);
1853
1854     blk->root_state.open_flags    = blk->root->bs->open_flags;
1855     blk->root_state.read_only     = blk->root->bs->read_only;
1856     blk->root_state.detect_zeroes = blk->root->bs->detect_zeroes;
1857 }
1858
1859 /*
1860  * Returns the detect-zeroes setting to be used for bdrv_open() of a
1861  * BlockDriverState which is supposed to inherit the root state.
1862  */
1863 bool blk_get_detect_zeroes_from_root_state(BlockBackend *blk)
1864 {
1865     return blk->root_state.detect_zeroes;
1866 }
1867
1868 /*
1869  * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
1870  * supposed to inherit the root state.
1871  */
1872 int blk_get_open_flags_from_root_state(BlockBackend *blk)
1873 {
1874     int bs_flags;
1875
1876     bs_flags = blk->root_state.read_only ? 0 : BDRV_O_RDWR;
1877     bs_flags |= blk->root_state.open_flags & ~BDRV_O_RDWR;
1878
1879     return bs_flags;
1880 }
1881
1882 BlockBackendRootState *blk_get_root_state(BlockBackend *blk)
1883 {
1884     return &blk->root_state;
1885 }
1886
1887 int blk_commit_all(void)
1888 {
1889     BlockBackend *blk = NULL;
1890
1891     while ((blk = blk_all_next(blk)) != NULL) {
1892         AioContext *aio_context = blk_get_aio_context(blk);
1893
1894         aio_context_acquire(aio_context);
1895         if (blk_is_inserted(blk) && blk->root->bs->backing) {
1896             int ret = bdrv_commit(blk->root->bs);
1897             if (ret < 0) {
1898                 aio_context_release(aio_context);
1899                 return ret;
1900             }
1901         }
1902         aio_context_release(aio_context);
1903     }
1904     return 0;
1905 }
1906
1907
1908 /* throttling disk I/O limits */
1909 void blk_set_io_limits(BlockBackend *blk, ThrottleConfig *cfg)
1910 {
1911     throttle_group_config(blk, cfg);
1912 }
1913
1914 void blk_io_limits_disable(BlockBackend *blk)
1915 {
1916     assert(blk->public.throttle_state);
1917     bdrv_drained_begin(blk_bs(blk));
1918     throttle_group_unregister_blk(blk);
1919     bdrv_drained_end(blk_bs(blk));
1920 }
1921
1922 /* should be called before blk_set_io_limits if a limit is set */
1923 void blk_io_limits_enable(BlockBackend *blk, const char *group)
1924 {
1925     assert(!blk->public.throttle_state);
1926     throttle_group_register_blk(blk, group);
1927 }
1928
1929 void blk_io_limits_update_group(BlockBackend *blk, const char *group)
1930 {
1931     /* this BB is not part of any group */
1932     if (!blk->public.throttle_state) {
1933         return;
1934     }
1935
1936     /* this BB is a part of the same group than the one we want */
1937     if (!g_strcmp0(throttle_group_get_name(blk), group)) {
1938         return;
1939     }
1940
1941     /* need to change the group this bs belong to */
1942     blk_io_limits_disable(blk);
1943     blk_io_limits_enable(blk, group);
1944 }
1945
1946 static void blk_root_drained_begin(BdrvChild *child)
1947 {
1948     BlockBackend *blk = child->opaque;
1949
1950     if (++blk->quiesce_counter == 1) {
1951         if (blk->dev_ops && blk->dev_ops->drained_begin) {
1952             blk->dev_ops->drained_begin(blk->dev_opaque);
1953         }
1954     }
1955
1956     /* Note that blk->root may not be accessible here yet if we are just
1957      * attaching to a BlockDriverState that is drained. Use child instead. */
1958
1959     if (atomic_fetch_inc(&blk->public.io_limits_disabled) == 0) {
1960         throttle_group_restart_blk(blk);
1961     }
1962 }
1963
1964 static void blk_root_drained_end(BdrvChild *child)
1965 {
1966     BlockBackend *blk = child->opaque;
1967     assert(blk->quiesce_counter);
1968
1969     assert(blk->public.io_limits_disabled);
1970     atomic_dec(&blk->public.io_limits_disabled);
1971
1972     if (--blk->quiesce_counter == 0) {
1973         if (blk->dev_ops && blk->dev_ops->drained_end) {
1974             blk->dev_ops->drained_end(blk->dev_opaque);
1975         }
1976     }
1977 }