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/coroutines.h"
  18 #include "block/throttle-groups.h"
  19 #include "hw/qdev-core.h"
  20 #include "sysemu/blockdev.h"
  21 #include "sysemu/runstate.h"
  22 #include "sysemu/replay.h"
  23 #include "qapi/error.h"
  24 #include "qapi/qapi-events-block.h"
  25 #include "qemu/id.h"
  26 #include "qemu/main-loop.h"
  27 #include "qemu/option.h"
  28 #include "trace.h"
  29 #include "migration/misc.h"
  30
  31 /* Number of coroutines to reserve per attached device model */
  32 #define COROUTINE_POOL_RESERVATION 64
  33
  34 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
  35
  36 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb);
  37
  38 typedef struct BlockBackendAioNotifier {
  39     void (*attached_aio_context)(AioContext *new_context, void *opaque);
  40     void (*detach_aio_context)(void *opaque);
  41     void *opaque;
  42     QLIST_ENTRY(BlockBackendAioNotifier) list;
  43 } BlockBackendAioNotifier;
  44
  45 struct BlockBackend {
  46     char *name;
  47     int refcnt;
  48     BdrvChild *root;
  49     AioContext *ctx;
  50     DriveInfo *legacy_dinfo;    /* null unless created by drive_new() */
  51     QTAILQ_ENTRY(BlockBackend) link;         /* for block_backends */
  52     QTAILQ_ENTRY(BlockBackend) monitor_link; /* for monitor_block_backends */
  53     BlockBackendPublic public;
  54
  55     DeviceState *dev;           /* attached device model, if any */
  56     const BlockDevOps *dev_ops;
  57     void *dev_opaque;
  58
  59     /* If the BDS tree is removed, some of its options are stored here (which
  60      * can be used to restore those options in the new BDS on insert) */
  61     BlockBackendRootState root_state;
  62
  63     bool enable_write_cache;
  64
  65     /* I/O stats (display with "info blockstats"). */
  66     BlockAcctStats stats;
  67
  68     BlockdevOnError on_read_error, on_write_error;
  69     bool iostatus_enabled;
  70     BlockDeviceIoStatus iostatus;
  71
  72     uint64_t perm;
  73     uint64_t shared_perm;
  74     bool disable_perm;
  75
  76     bool allow_aio_context_change;
  77     bool allow_write_beyond_eof;
  78
  79     /* Protected by BQL */
  80     NotifierList remove_bs_notifiers, insert_bs_notifiers;
  81     QLIST_HEAD(, BlockBackendAioNotifier) aio_notifiers;
  82
  83     int quiesce_counter; /* atomic: written under BQL, read by other threads */
  84     QemuMutex queued_requests_lock; /* protects queued_requests */
  85     CoQueue queued_requests;
  86     bool disable_request_queuing; /* atomic */
  87
  88     VMChangeStateEntry *vmsh;
  89     bool force_allow_inactivate;
  90
  91     /* Number of in-flight aio requests.  BlockDriverState also counts
  92      * in-flight requests but aio requests can exist even when blk->root is
  93      * NULL, so we cannot rely on its counter for that case.
  94      * Accessed with atomic ops.
  95      */
  96     unsigned int in_flight;
  97 };
  98
  99 typedef struct BlockBackendAIOCB {
 100     BlockAIOCB common;
 101     BlockBackend *blk;
 102     int ret;
 103 } BlockBackendAIOCB;
 104
 105 static const AIOCBInfo block_backend_aiocb_info = {
 106     .get_aio_context = blk_aiocb_get_aio_context,
 107     .aiocb_size = sizeof(BlockBackendAIOCB),
 108 };
 109
 110 static void drive_info_del(DriveInfo *dinfo);
 111 static BlockBackend *bdrv_first_blk(BlockDriverState *bs);
 112
 113 /* All BlockBackends. Protected by BQL. */
 114 static QTAILQ_HEAD(, BlockBackend) block_backends =
 115     QTAILQ_HEAD_INITIALIZER(block_backends);
 116
 117 /*
 118  * All BlockBackends referenced by the monitor and which are iterated through by
 119  * blk_next(). Protected by BQL.
 120  */
 121 static QTAILQ_HEAD(, BlockBackend) monitor_block_backends =
 122     QTAILQ_HEAD_INITIALIZER(monitor_block_backends);
 123
 124 static int coroutine_mixed_fn GRAPH_RDLOCK
 125 blk_set_perm_locked(BlockBackend *blk, uint64_t perm, uint64_t shared_perm,
 126                     Error **errp);
 127
 128 static void blk_root_inherit_options(BdrvChildRole role, bool parent_is_format,
 129                                      int *child_flags, QDict *child_options,
 130                                      int parent_flags, QDict *parent_options)
 131 {
 132     /* We're not supposed to call this function for root nodes */
 133     abort();
 134 }
 135 static void blk_root_drained_begin(BdrvChild *child);
 136 static bool blk_root_drained_poll(BdrvChild *child);
 137 static void blk_root_drained_end(BdrvChild *child);
 138
 139 static void blk_root_change_media(BdrvChild *child, bool load);
 140 static void blk_root_resize(BdrvChild *child);
 141
 142 static bool blk_root_change_aio_ctx(BdrvChild *child, AioContext *ctx,
 143                                     GHashTable *visited, Transaction *tran,
 144                                     Error **errp);
 145
 146 static char *blk_root_get_parent_desc(BdrvChild *child)
 147 {
 148     BlockBackend *blk = child->opaque;
 149     g_autofree char *dev_id = NULL;
 150
 151     if (blk->name) {
 152         return g_strdup_printf("block device '%s'", blk->name);
 153     }
 154
 155     dev_id = blk_get_attached_dev_id(blk);
 156     if (*dev_id) {
 157         return g_strdup_printf("block device '%s'", dev_id);
 158     } else {
 159         /* TODO Callback into the BB owner for something more detailed */
 160         return g_strdup("an unnamed block device");
 161     }
 162 }
 163
 164 static const char *blk_root_get_name(BdrvChild *child)
 165 {
 166     return blk_name(child->opaque);
 167 }
 168
 169 static void blk_vm_state_changed(void *opaque, bool running, RunState state)
 170 {
 171     Error *local_err = NULL;
 172     BlockBackend *blk = opaque;
 173
 174     if (state == RUN_STATE_INMIGRATE) {
 175         return;
 176     }
 177
 178     qemu_del_vm_change_state_handler(blk->vmsh);
 179     blk->vmsh = NULL;
 180     blk_set_perm(blk, blk->perm, blk->shared_perm, &local_err);
 181     if (local_err) {
 182         error_report_err(local_err);
 183     }
 184 }
 185
 186 /*
 187  * Notifies the user of the BlockBackend that migration has completed. qdev
 188  * devices can tighten their permissions in response (specifically revoke
 189  * shared write permissions that we needed for storage migration).
 190  *
 191  * If an error is returned, the VM cannot be allowed to be resumed.
 192  */
 193 static void GRAPH_RDLOCK blk_root_activate(BdrvChild *child, Error **errp)
 194 {
 195     BlockBackend *blk = child->opaque;
 196     Error *local_err = NULL;
 197     uint64_t saved_shared_perm;
 198
 199     if (!blk->disable_perm) {
 200         return;
 201     }
 202
 203     blk->disable_perm = false;
 204
 205     /*
 206      * blk->shared_perm contains the permissions we want to share once
 207      * migration is really completely done.  For now, we need to share
 208      * all; but we also need to retain blk->shared_perm, which is
 209      * overwritten by a successful blk_set_perm() call.  Save it and
 210      * restore it below.
 211      */
 212     saved_shared_perm = blk->shared_perm;
 213
 214     blk_set_perm_locked(blk, blk->perm, BLK_PERM_ALL, &local_err);
 215     if (local_err) {
 216         error_propagate(errp, local_err);
 217         blk->disable_perm = true;
 218         return;
 219     }
 220     blk->shared_perm = saved_shared_perm;
 221
 222     if (runstate_check(RUN_STATE_INMIGRATE)) {
 223         /* Activation can happen when migration process is still active, for
 224          * example when nbd_server_add is called during non-shared storage
 225          * migration. Defer the shared_perm update to migration completion. */
 226         if (!blk->vmsh) {
 227             blk->vmsh = qemu_add_vm_change_state_handler(blk_vm_state_changed,
 228                                                          blk);
 229         }
 230         return;
 231     }
 232
 233     blk_set_perm_locked(blk, blk->perm, blk->shared_perm, &local_err);
 234     if (local_err) {
 235         error_propagate(errp, local_err);
 236         blk->disable_perm = true;
 237         return;
 238     }
 239 }
 240
 241 void blk_set_force_allow_inactivate(BlockBackend *blk)
 242 {
 243     GLOBAL_STATE_CODE();
 244     blk->force_allow_inactivate = true;
 245 }
 246
 247 static bool blk_can_inactivate(BlockBackend *blk)
 248 {
 249     /* If it is a guest device, inactivate is ok. */
 250     if (blk->dev || blk_name(blk)[0]) {
 251         return true;
 252     }
 253
 254     /* Inactivating means no more writes to the image can be done,
 255      * even if those writes would be changes invisible to the
 256      * guest.  For block job BBs that satisfy this, we can just allow
 257      * it.  This is the case for mirror job source, which is required
 258      * by libvirt non-shared block migration. */
 259     if (!(blk->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED))) {
 260         return true;
 261     }
 262
 263     return blk->force_allow_inactivate;
 264 }
 265
 266 static int GRAPH_RDLOCK blk_root_inactivate(BdrvChild *child)
 267 {
 268     BlockBackend *blk = child->opaque;
 269
 270     if (blk->disable_perm) {
 271         return 0;
 272     }
 273
 274     if (!blk_can_inactivate(blk)) {
 275         return -EPERM;
 276     }
 277
 278     blk->disable_perm = true;
 279     if (blk->root) {
 280         bdrv_child_try_set_perm(blk->root, 0, BLK_PERM_ALL, &error_abort);
 281     }
 282
 283     return 0;
 284 }
 285
 286 static void blk_root_attach(BdrvChild *child)
 287 {
 288     BlockBackend *blk = child->opaque;
 289     BlockBackendAioNotifier *notifier;
 290
 291     trace_blk_root_attach(child, blk, child->bs);
 292
 293     QLIST_FOREACH(notifier, &blk->aio_notifiers, list) {
 294         bdrv_add_aio_context_notifier(child->bs,
 295                 notifier->attached_aio_context,
 296                 notifier->detach_aio_context,
 297                 notifier->opaque);
 298     }
 299 }
 300
 301 static void blk_root_detach(BdrvChild *child)
 302 {
 303     BlockBackend *blk = child->opaque;
 304     BlockBackendAioNotifier *notifier;
 305
 306     trace_blk_root_detach(child, blk, child->bs);
 307
 308     QLIST_FOREACH(notifier, &blk->aio_notifiers, list) {
 309         bdrv_remove_aio_context_notifier(child->bs,
 310                 notifier->attached_aio_context,
 311                 notifier->detach_aio_context,
 312                 notifier->opaque);
 313     }
 314 }
 315
 316 static AioContext *blk_root_get_parent_aio_context(BdrvChild *c)
 317 {
 318     BlockBackend *blk = c->opaque;
 319     IO_CODE();
 320
 321     return blk_get_aio_context(blk);
 322 }
 323
 324 static const BdrvChildClass child_root = {
 325     .inherit_options    = blk_root_inherit_options,
 326
 327     .change_media       = blk_root_change_media,
 328     .resize             = blk_root_resize,
 329     .get_name           = blk_root_get_name,
 330     .get_parent_desc    = blk_root_get_parent_desc,
 331
 332     .drained_begin      = blk_root_drained_begin,
 333     .drained_poll       = blk_root_drained_poll,
 334     .drained_end        = blk_root_drained_end,
 335
 336     .activate           = blk_root_activate,
 337     .inactivate         = blk_root_inactivate,
 338
 339     .attach             = blk_root_attach,
 340     .detach             = blk_root_detach,
 341
 342     .change_aio_ctx     = blk_root_change_aio_ctx,
 343
 344     .get_parent_aio_context = blk_root_get_parent_aio_context,
 345 };
 346
 347 /*
 348  * Create a new BlockBackend with a reference count of one.
 349  *
 350  * @perm is a bitmasks of BLK_PERM_* constants which describes the permissions
 351  * to request for a block driver node that is attached to this BlockBackend.
 352  * @shared_perm is a bitmask which describes which permissions may be granted
 353  * to other users of the attached node.
 354  * Both sets of permissions can be changed later using blk_set_perm().
 355  *
 356  * Return the new BlockBackend on success, null on failure.
 357  */
 358 BlockBackend *blk_new(AioContext *ctx, uint64_t perm, uint64_t shared_perm)
 359 {
 360     BlockBackend *blk;
 361
 362     GLOBAL_STATE_CODE();
 363
 364     blk = g_new0(BlockBackend, 1);
 365     blk->refcnt = 1;
 366     blk->ctx = ctx;
 367     blk->perm = perm;
 368     blk->shared_perm = shared_perm;
 369     blk_set_enable_write_cache(blk, true);
 370
 371     blk->on_read_error = BLOCKDEV_ON_ERROR_REPORT;
 372     blk->on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
 373
 374     block_acct_init(&blk->stats);
 375
 376     qemu_mutex_init(&blk->queued_requests_lock);
 377     qemu_co_queue_init(&blk->queued_requests);
 378     notifier_list_init(&blk->remove_bs_notifiers);
 379     notifier_list_init(&blk->insert_bs_notifiers);
 380     QLIST_INIT(&blk->aio_notifiers);
 381
 382     QTAILQ_INSERT_TAIL(&block_backends, blk, link);
 383     return blk;
 384 }
 385
 386 /*
 387  * Create a new BlockBackend connected to an existing BlockDriverState.
 388  *
 389  * @perm is a bitmasks of BLK_PERM_* constants which describes the
 390  * permissions to request for @bs that is attached to this
 391  * BlockBackend.  @shared_perm is a bitmask which describes which
 392  * permissions may be granted to other users of the attached node.
 393  * Both sets of permissions can be changed later using blk_set_perm().
 394  *
 395  * Return the new BlockBackend on success, null on failure.
 396  *
 397  * Callers must hold the AioContext lock of @bs.
 398  */
 399 BlockBackend *blk_new_with_bs(BlockDriverState *bs, uint64_t perm,
 400                               uint64_t shared_perm, Error **errp)
 401 {
 402     BlockBackend *blk = blk_new(bdrv_get_aio_context(bs), perm, shared_perm);
 403
 404     GLOBAL_STATE_CODE();
 405
 406     if (blk_insert_bs(blk, bs, errp) < 0) {
 407         blk_unref(blk);
 408         return NULL;
 409     }
 410     return blk;
 411 }
 412
 413 /*
 414  * Creates a new BlockBackend, opens a new BlockDriverState, and connects both.
 415  * By default, the new BlockBackend is in the main AioContext, but if the
 416  * parameters connect it with any existing node in a different AioContext, it
 417  * may end up there instead.
 418  *
 419  * Just as with bdrv_open(), after having called this function the reference to
 420  * @options belongs to the block layer (even on failure).
 421  *
 422  * Called without holding an AioContext lock.
 423  *
 424  * TODO: Remove @filename and @flags; it should be possible to specify a whole
 425  * BDS tree just by specifying the @options QDict (or @reference,
 426  * alternatively). At the time of adding this function, this is not possible,
 427  * though, so callers of this function have to be able to specify @filename and
 428  * @flags.
 429  */
 430 BlockBackend *blk_new_open(const char *filename, const char *reference,
 431                            QDict *options, int flags, Error **errp)
 432 {
 433     BlockBackend *blk;
 434     BlockDriverState *bs;
 435     AioContext *ctx;
 436     uint64_t perm = 0;
 437     uint64_t shared = BLK_PERM_ALL;
 438
 439     GLOBAL_STATE_CODE();
 440
 441     /*
 442      * blk_new_open() is mainly used in .bdrv_create implementations and the
 443      * tools where sharing isn't a major concern because the BDS stays private
 444      * and the file is generally not supposed to be used by a second process,
 445      * so we just request permission according to the flags.
 446      *
 447      * The exceptions are xen_disk and blockdev_init(); in these cases, the
 448      * caller of blk_new_open() doesn't make use of the permissions, but they
 449      * shouldn't hurt either. We can still share everything here because the
 450      * guest devices will add their own blockers if they can't share.
 451      */
 452     if ((flags & BDRV_O_NO_IO) == 0) {
 453         perm |= BLK_PERM_CONSISTENT_READ;
 454         if (flags & BDRV_O_RDWR) {
 455             perm |= BLK_PERM_WRITE;
 456         }
 457     }
 458     if (flags & BDRV_O_RESIZE) {
 459         perm |= BLK_PERM_RESIZE;
 460     }
 461     if (flags & BDRV_O_NO_SHARE) {
 462         shared = BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
 463     }
 464
 465     aio_context_acquire(qemu_get_aio_context());
 466     bs = bdrv_open(filename, reference, options, flags, errp);
 467     aio_context_release(qemu_get_aio_context());
 468     if (!bs) {
 469         return NULL;
 470     }
 471
 472     /* bdrv_open() could have moved bs to a different AioContext */
 473     ctx = bdrv_get_aio_context(bs);
 474     blk = blk_new(bdrv_get_aio_context(bs), perm, shared);
 475     blk->perm = perm;
 476     blk->shared_perm = shared;
 477
 478     aio_context_acquire(ctx);
 479     blk_insert_bs(blk, bs, errp);
 480     bdrv_unref(bs);
 481     aio_context_release(ctx);
 482
 483     if (!blk->root) {
 484         blk_unref(blk);
 485         return NULL;
 486     }
 487
 488     return blk;
 489 }
 490
 491 static void blk_delete(BlockBackend *blk)
 492 {
 493     assert(!blk->refcnt);
 494     assert(!blk->name);
 495     assert(!blk->dev);
 496     if (blk->public.throttle_group_member.throttle_state) {
 497         blk_io_limits_disable(blk);
 498     }
 499     if (blk->root) {
 500         blk_remove_bs(blk);
 501     }
 502     if (blk->vmsh) {
 503         qemu_del_vm_change_state_handler(blk->vmsh);
 504         blk->vmsh = NULL;
 505     }
 506     assert(QLIST_EMPTY(&blk->remove_bs_notifiers.notifiers));
 507     assert(QLIST_EMPTY(&blk->insert_bs_notifiers.notifiers));
 508     assert(QLIST_EMPTY(&blk->aio_notifiers));
 509     assert(qemu_co_queue_empty(&blk->queued_requests));
 510     qemu_mutex_destroy(&blk->queued_requests_lock);
 511     QTAILQ_REMOVE(&block_backends, blk, link);
 512     drive_info_del(blk->legacy_dinfo);
 513     block_acct_cleanup(&blk->stats);
 514     g_free(blk);
 515 }
 516
 517 static void drive_info_del(DriveInfo *dinfo)
 518 {
 519     if (!dinfo) {
 520         return;
 521     }
 522     qemu_opts_del(dinfo->opts);
 523     g_free(dinfo);
 524 }
 525
 526 int blk_get_refcnt(BlockBackend *blk)
 527 {
 528     GLOBAL_STATE_CODE();
 529     return blk ? blk->refcnt : 0;
 530 }
 531
 532 /*
 533  * Increment @blk's reference count.
 534  * @blk must not be null.
 535  */
 536 void blk_ref(BlockBackend *blk)
 537 {
 538     assert(blk->refcnt > 0);
 539     GLOBAL_STATE_CODE();
 540     blk->refcnt++;
 541 }
 542
 543 /*
 544  * Decrement @blk's reference count.
 545  * If this drops it to zero, destroy @blk.
 546  * For convenience, do nothing if @blk is null.
 547  */
 548 void blk_unref(BlockBackend *blk)
 549 {
 550     GLOBAL_STATE_CODE();
 551     if (blk) {
 552         assert(blk->refcnt > 0);
 553         if (blk->refcnt > 1) {
 554             blk->refcnt--;
 555         } else {
 556             blk_drain(blk);
 557             /* blk_drain() cannot resurrect blk, nobody held a reference */
 558             assert(blk->refcnt == 1);
 559             blk->refcnt = 0;
 560             blk_delete(blk);
 561         }
 562     }
 563 }
 564
 565 /*
 566  * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
 567  * ones which are hidden (i.e. are not referenced by the monitor).
 568  */
 569 BlockBackend *blk_all_next(BlockBackend *blk)
 570 {
 571     GLOBAL_STATE_CODE();
 572     return blk ? QTAILQ_NEXT(blk, link)
 573                : QTAILQ_FIRST(&block_backends);
 574 }
 575
 576 void blk_remove_all_bs(void)
 577 {
 578     BlockBackend *blk = NULL;
 579
 580     GLOBAL_STATE_CODE();
 581
 582     while ((blk = blk_all_next(blk)) != NULL) {
 583         AioContext *ctx = blk_get_aio_context(blk);
 584
 585         aio_context_acquire(ctx);
 586         if (blk->root) {
 587             blk_remove_bs(blk);
 588         }
 589         aio_context_release(ctx);
 590     }
 591 }
 592
 593 /*
 594  * Return the monitor-owned BlockBackend after @blk.
 595  * If @blk is null, return the first one.
 596  * Else, return @blk's next sibling, which may be null.
 597  *
 598  * To iterate over all BlockBackends, do
 599  * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
 600  *     ...
 601  * }
 602  */
 603 BlockBackend *blk_next(BlockBackend *blk)
 604 {
 605     GLOBAL_STATE_CODE();
 606     return blk ? QTAILQ_NEXT(blk, monitor_link)
 607                : QTAILQ_FIRST(&monitor_block_backends);
 608 }
 609
 610 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
 611  * the monitor or attached to a BlockBackend */
 612 BlockDriverState *bdrv_next(BdrvNextIterator *it)
 613 {
 614     BlockDriverState *bs, *old_bs;
 615
 616     /* Must be called from the main loop */
 617     assert(qemu_get_current_aio_context() == qemu_get_aio_context());
 618
 619     /* First, return all root nodes of BlockBackends. In order to avoid
 620      * returning a BDS twice when multiple BBs refer to it, we only return it
 621      * if the BB is the first one in the parent list of the BDS. */
 622     if (it->phase == BDRV_NEXT_BACKEND_ROOTS) {
 623         BlockBackend *old_blk = it->blk;
 624
 625         old_bs = old_blk ? blk_bs(old_blk) : NULL;
 626
 627         do {
 628             it->blk = blk_all_next(it->blk);
 629             bs = it->blk ? blk_bs(it->blk) : NULL;
 630         } while (it->blk && (bs == NULL || bdrv_first_blk(bs) != it->blk));
 631
 632         if (it->blk) {
 633             blk_ref(it->blk);
 634         }
 635         blk_unref(old_blk);
 636
 637         if (bs) {
 638             bdrv_ref(bs);
 639             bdrv_unref(old_bs);
 640             return bs;
 641         }
 642         it->phase = BDRV_NEXT_MONITOR_OWNED;
 643     } else {
 644         old_bs = it->bs;
 645     }
 646
 647     /* Then return the monitor-owned BDSes without a BB attached. Ignore all
 648      * BDSes that are attached to a BlockBackend here; they have been handled
 649      * by the above block already */
 650     do {
 651         it->bs = bdrv_next_monitor_owned(it->bs);
 652         bs = it->bs;
 653     } while (bs && bdrv_has_blk(bs));
 654
 655     if (bs) {
 656         bdrv_ref(bs);
 657     }
 658     bdrv_unref(old_bs);
 659
 660     return bs;
 661 }
 662
 663 static void bdrv_next_reset(BdrvNextIterator *it)
 664 {
 665     *it = (BdrvNextIterator) {
 666         .phase = BDRV_NEXT_BACKEND_ROOTS,
 667     };
 668 }
 669
 670 BlockDriverState *bdrv_first(BdrvNextIterator *it)
 671 {
 672     GLOBAL_STATE_CODE();
 673     bdrv_next_reset(it);
 674     return bdrv_next(it);
 675 }
 676
 677 /* Must be called when aborting a bdrv_next() iteration before
 678  * bdrv_next() returns NULL */
 679 void bdrv_next_cleanup(BdrvNextIterator *it)
 680 {
 681     /* Must be called from the main loop */
 682     assert(qemu_get_current_aio_context() == qemu_get_aio_context());
 683
 684     if (it->phase == BDRV_NEXT_BACKEND_ROOTS) {
 685         if (it->blk) {
 686             bdrv_unref(blk_bs(it->blk));
 687             blk_unref(it->blk);
 688         }
 689     } else {
 690         bdrv_unref(it->bs);
 691     }
 692
 693     bdrv_next_reset(it);
 694 }
 695
 696 /*
 697  * Add a BlockBackend into the list of backends referenced by the monitor, with
 698  * the given @name acting as the handle for the monitor.
 699  * Strictly for use by blockdev.c.
 700  *
 701  * @name must not be null or empty.
 702  *
 703  * Returns true on success and false on failure. In the latter case, an Error
 704  * object is returned through @errp.
 705  */
 706 bool monitor_add_blk(BlockBackend *blk, const char *name, Error **errp)
 707 {
 708     assert(!blk->name);
 709     assert(name && name[0]);
 710     GLOBAL_STATE_CODE();
 711
 712     if (!id_wellformed(name)) {
 713         error_setg(errp, "Invalid device name");
 714         return false;
 715     }
 716     if (blk_by_name(name)) {
 717         error_setg(errp, "Device with id '%s' already exists", name);
 718         return false;
 719     }
 720     if (bdrv_find_node(name)) {
 721         error_setg(errp,
 722                    "Device name '%s' conflicts with an existing node name",
 723                    name);
 724         return false;
 725     }
 726
 727     blk->name = g_strdup(name);
 728     QTAILQ_INSERT_TAIL(&monitor_block_backends, blk, monitor_link);
 729     return true;
 730 }
 731
 732 /*
 733  * Remove a BlockBackend from the list of backends referenced by the monitor.
 734  * Strictly for use by blockdev.c.
 735  */
 736 void monitor_remove_blk(BlockBackend *blk)
 737 {
 738     GLOBAL_STATE_CODE();
 739
 740     if (!blk->name) {
 741         return;
 742     }
 743
 744     QTAILQ_REMOVE(&monitor_block_backends, blk, monitor_link);
 745     g_free(blk->name);
 746     blk->name = NULL;
 747 }
 748
 749 /*
 750  * Return @blk's name, a non-null string.
 751  * Returns an empty string iff @blk is not referenced by the monitor.
 752  */
 753 const char *blk_name(const BlockBackend *blk)
 754 {
 755     IO_CODE();
 756     return blk->name ?: "";
 757 }
 758
 759 /*
 760  * Return the BlockBackend with name @name if it exists, else null.
 761  * @name must not be null.
 762  */
 763 BlockBackend *blk_by_name(const char *name)
 764 {
 765     BlockBackend *blk = NULL;
 766
 767     GLOBAL_STATE_CODE();
 768     assert(name);
 769     while ((blk = blk_next(blk)) != NULL) {
 770         if (!strcmp(name, blk->name)) {
 771             return blk;
 772         }
 773     }
 774     return NULL;
 775 }
 776
 777 /*
 778  * Return the BlockDriverState attached to @blk if any, else null.
 779  */
 780 BlockDriverState *blk_bs(BlockBackend *blk)
 781 {
 782     IO_CODE();
 783     return blk->root ? blk->root->bs : NULL;
 784 }
 785
 786 static BlockBackend *bdrv_first_blk(BlockDriverState *bs)
 787 {
 788     BdrvChild *child;
 789
 790     GLOBAL_STATE_CODE();
 791
 792     QLIST_FOREACH(child, &bs->parents, next_parent) {
 793         if (child->klass == &child_root) {
 794             return child->opaque;
 795         }
 796     }
 797
 798     return NULL;
 799 }
 800
 801 /*
 802  * Returns true if @bs has an associated BlockBackend.
 803  */
 804 bool bdrv_has_blk(BlockDriverState *bs)
 805 {
 806     GLOBAL_STATE_CODE();
 807     return bdrv_first_blk(bs) != NULL;
 808 }
 809
 810 /*
 811  * Returns true if @bs has only BlockBackends as parents.
 812  */
 813 bool bdrv_is_root_node(BlockDriverState *bs)
 814 {
 815     BdrvChild *c;
 816
 817     GLOBAL_STATE_CODE();
 818     QLIST_FOREACH(c, &bs->parents, next_parent) {
 819         if (c->klass != &child_root) {
 820             return false;
 821         }
 822     }
 823
 824     return true;
 825 }
 826
 827 /*
 828  * Return @blk's DriveInfo if any, else null.
 829  */
 830 DriveInfo *blk_legacy_dinfo(BlockBackend *blk)
 831 {
 832     GLOBAL_STATE_CODE();
 833     return blk->legacy_dinfo;
 834 }
 835
 836 /*
 837  * Set @blk's DriveInfo to @dinfo, and return it.
 838  * @blk must not have a DriveInfo set already.
 839  * No other BlockBackend may have the same DriveInfo set.
 840  */
 841 DriveInfo *blk_set_legacy_dinfo(BlockBackend *blk, DriveInfo *dinfo)
 842 {
 843     assert(!blk->legacy_dinfo);
 844     GLOBAL_STATE_CODE();
 845     return blk->legacy_dinfo = dinfo;
 846 }
 847
 848 /*
 849  * Return the BlockBackend with DriveInfo @dinfo.
 850  * It must exist.
 851  */
 852 BlockBackend *blk_by_legacy_dinfo(DriveInfo *dinfo)
 853 {
 854     BlockBackend *blk = NULL;
 855     GLOBAL_STATE_CODE();
 856
 857     while ((blk = blk_next(blk)) != NULL) {
 858         if (blk->legacy_dinfo == dinfo) {
 859             return blk;
 860         }
 861     }
 862     abort();
 863 }
 864
 865 /*
 866  * Returns a pointer to the publicly accessible fields of @blk.
 867  */
 868 BlockBackendPublic *blk_get_public(BlockBackend *blk)
 869 {
 870     GLOBAL_STATE_CODE();
 871     return &blk->public;
 872 }
 873
 874 /*
 875  * Returns a BlockBackend given the associated @public fields.
 876  */
 877 BlockBackend *blk_by_public(BlockBackendPublic *public)
 878 {
 879     GLOBAL_STATE_CODE();
 880     return container_of(public, BlockBackend, public);
 881 }
 882
 883 /*
 884  * Disassociates the currently associated BlockDriverState from @blk.
 885  */
 886 void blk_remove_bs(BlockBackend *blk)
 887 {
 888     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
 889     BdrvChild *root;
 890
 891     GLOBAL_STATE_CODE();
 892
 893     notifier_list_notify(&blk->remove_bs_notifiers, blk);
 894     if (tgm->throttle_state) {
 895         BlockDriverState *bs = blk_bs(blk);
 896
 897         /*
 898          * Take a ref in case blk_bs() changes across bdrv_drained_begin(), for
 899          * example, if a temporary filter node is removed by a blockjob.
 900          */
 901         bdrv_ref(bs);
 902         bdrv_drained_begin(bs);
 903         throttle_group_detach_aio_context(tgm);
 904         throttle_group_attach_aio_context(tgm, qemu_get_aio_context());
 905         bdrv_drained_end(bs);
 906         bdrv_unref(bs);
 907     }
 908
 909     blk_update_root_state(blk);
 910
 911     /* bdrv_root_unref_child() will cause blk->root to become stale and may
 912      * switch to a completion coroutine later on. Let's drain all I/O here
 913      * to avoid that and a potential QEMU crash.
 914      */
 915     blk_drain(blk);
 916     root = blk->root;
 917     blk->root = NULL;
 918     bdrv_root_unref_child(root);
 919 }
 920
 921 /*
 922  * Associates a new BlockDriverState with @blk.
 923  *
 924  * Callers must hold the AioContext lock of @bs.
 925  */
 926 int blk_insert_bs(BlockBackend *blk, BlockDriverState *bs, Error **errp)
 927 {
 928     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
 929     GLOBAL_STATE_CODE();
 930     bdrv_ref(bs);
 931     blk->root = bdrv_root_attach_child(bs, "root", &child_root,
 932                                        BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY,
 933                                        blk->perm, blk->shared_perm,
 934                                        blk, errp);
 935     if (blk->root == NULL) {
 936         return -EPERM;
 937     }
 938
 939     notifier_list_notify(&blk->insert_bs_notifiers, blk);
 940     if (tgm->throttle_state) {
 941         throttle_group_detach_aio_context(tgm);
 942         throttle_group_attach_aio_context(tgm, bdrv_get_aio_context(bs));
 943     }
 944
 945     return 0;
 946 }
 947
 948 /*
 949  * Change BlockDriverState associated with @blk.
 950  */
 951 int blk_replace_bs(BlockBackend *blk, BlockDriverState *new_bs, Error **errp)
 952 {
 953     GLOBAL_STATE_CODE();
 954     return bdrv_replace_child_bs(blk->root, new_bs, errp);
 955 }
 956
 957 /*
 958  * Sets the permission bitmasks that the user of the BlockBackend needs.
 959  */
 960 static int coroutine_mixed_fn GRAPH_RDLOCK
 961 blk_set_perm_locked(BlockBackend *blk, uint64_t perm, uint64_t shared_perm,
 962                     Error **errp)
 963 {
 964     int ret;
 965     GLOBAL_STATE_CODE();
 966
 967     if (blk->root && !blk->disable_perm) {
 968         ret = bdrv_child_try_set_perm(blk->root, perm, shared_perm, errp);
 969         if (ret < 0) {
 970             return ret;
 971         }
 972     }
 973
 974     blk->perm = perm;
 975     blk->shared_perm = shared_perm;
 976
 977     return 0;
 978 }
 979
 980 int blk_set_perm(BlockBackend *blk, uint64_t perm, uint64_t shared_perm,
 981                  Error **errp)
 982 {
 983     GLOBAL_STATE_CODE();
 984     GRAPH_RDLOCK_GUARD_MAINLOOP();
 985
 986     return blk_set_perm_locked(blk, perm, shared_perm, errp);
 987 }
 988
 989 void blk_get_perm(BlockBackend *blk, uint64_t *perm, uint64_t *shared_perm)
 990 {
 991     GLOBAL_STATE_CODE();
 992     *perm = blk->perm;
 993     *shared_perm = blk->shared_perm;
 994 }
 995
 996 /*
 997  * Attach device model @dev to @blk.
 998  * Return 0 on success, -EBUSY when a device model is attached already.
 999  */
1000 int blk_attach_dev(BlockBackend *blk, DeviceState *dev)
1001 {
1002     GLOBAL_STATE_CODE();
1003     if (blk->dev) {
1004         return -EBUSY;
1005     }
1006
1007     /* While migration is still incoming, we don't need to apply the
1008      * permissions of guest device BlockBackends. We might still have a block
1009      * job or NBD server writing to the image for storage migration. */
1010     if (runstate_check(RUN_STATE_INMIGRATE)) {
1011         blk->disable_perm = true;
1012     }
1013
1014     blk_ref(blk);
1015     blk->dev = dev;
1016     blk_iostatus_reset(blk);
1017
1018     return 0;
1019 }
1020
1021 /*
1022  * Detach device model @dev from @blk.
1023  * @dev must be currently attached to @blk.
1024  */
1025 void blk_detach_dev(BlockBackend *blk, DeviceState *dev)
1026 {
1027     assert(blk->dev == dev);
1028     GLOBAL_STATE_CODE();
1029     blk->dev = NULL;
1030     blk->dev_ops = NULL;
1031     blk->dev_opaque = NULL;
1032     blk_set_perm(blk, 0, BLK_PERM_ALL, &error_abort);
1033     blk_unref(blk);
1034 }
1035
1036 /*
1037  * Return the device model attached to @blk if any, else null.
1038  */
1039 DeviceState *blk_get_attached_dev(BlockBackend *blk)
1040 {
1041     GLOBAL_STATE_CODE();
1042     return blk->dev;
1043 }
1044
1045 /* Return the qdev ID, or if no ID is assigned the QOM path, of the block
1046  * device attached to the BlockBackend. */
1047 char *blk_get_attached_dev_id(BlockBackend *blk)
1048 {
1049     DeviceState *dev = blk->dev;
1050     IO_CODE();
1051
1052     if (!dev) {
1053         return g_strdup("");
1054     } else if (dev->id) {
1055         return g_strdup(dev->id);
1056     }
1057
1058     return object_get_canonical_path(OBJECT(dev)) ?: g_strdup("");
1059 }
1060
1061 /*
1062  * Return the BlockBackend which has the device model @dev attached if it
1063  * exists, else null.
1064  *
1065  * @dev must not be null.
1066  */
1067 BlockBackend *blk_by_dev(void *dev)
1068 {
1069     BlockBackend *blk = NULL;
1070
1071     GLOBAL_STATE_CODE();
1072
1073     assert(dev != NULL);
1074     while ((blk = blk_all_next(blk)) != NULL) {
1075         if (blk->dev == dev) {
1076             return blk;
1077         }
1078     }
1079     return NULL;
1080 }
1081
1082 /*
1083  * Set @blk's device model callbacks to @ops.
1084  * @opaque is the opaque argument to pass to the callbacks.
1085  * This is for use by device models.
1086  */
1087 void blk_set_dev_ops(BlockBackend *blk, const BlockDevOps *ops,
1088                      void *opaque)
1089 {
1090     GLOBAL_STATE_CODE();
1091     blk->dev_ops = ops;
1092     blk->dev_opaque = opaque;
1093
1094     /* Are we currently quiesced? Should we enforce this right now? */
1095     if (qatomic_read(&blk->quiesce_counter) && ops && ops->drained_begin) {
1096         ops->drained_begin(opaque);
1097     }
1098 }
1099
1100 /*
1101  * Notify @blk's attached device model of media change.
1102  *
1103  * If @load is true, notify of media load. This action can fail, meaning that
1104  * the medium cannot be loaded. @errp is set then.
1105  *
1106  * If @load is false, notify of media eject. This can never fail.
1107  *
1108  * Also send DEVICE_TRAY_MOVED events as appropriate.
1109  */
1110 void blk_dev_change_media_cb(BlockBackend *blk, bool load, Error **errp)
1111 {
1112     GLOBAL_STATE_CODE();
1113     if (blk->dev_ops && blk->dev_ops->change_media_cb) {
1114         bool tray_was_open, tray_is_open;
1115         Error *local_err = NULL;
1116
1117         tray_was_open = blk_dev_is_tray_open(blk);
1118         blk->dev_ops->change_media_cb(blk->dev_opaque, load, &local_err);
1119         if (local_err) {
1120             assert(load == true);
1121             error_propagate(errp, local_err);
1122             return;
1123         }
1124         tray_is_open = blk_dev_is_tray_open(blk);
1125
1126         if (tray_was_open != tray_is_open) {
1127             char *id = blk_get_attached_dev_id(blk);
1128             qapi_event_send_device_tray_moved(blk_name(blk), id, tray_is_open);
1129             g_free(id);
1130         }
1131     }
1132 }
1133
1134 static void blk_root_change_media(BdrvChild *child, bool load)
1135 {
1136     blk_dev_change_media_cb(child->opaque, load, NULL);
1137 }
1138
1139 /*
1140  * Does @blk's attached device model have removable media?
1141  * %true if no device model is attached.
1142  */
1143 bool blk_dev_has_removable_media(BlockBackend *blk)
1144 {
1145     GLOBAL_STATE_CODE();
1146     return !blk->dev || (blk->dev_ops && blk->dev_ops->change_media_cb);
1147 }
1148
1149 /*
1150  * Does @blk's attached device model have a tray?
1151  */
1152 bool blk_dev_has_tray(BlockBackend *blk)
1153 {
1154     IO_CODE();
1155     return blk->dev_ops && blk->dev_ops->is_tray_open;
1156 }
1157
1158 /*
1159  * Notify @blk's attached device model of a media eject request.
1160  * If @force is true, the medium is about to be yanked out forcefully.
1161  */
1162 void blk_dev_eject_request(BlockBackend *blk, bool force)
1163 {
1164     GLOBAL_STATE_CODE();
1165     if (blk->dev_ops && blk->dev_ops->eject_request_cb) {
1166         blk->dev_ops->eject_request_cb(blk->dev_opaque, force);
1167     }
1168 }
1169
1170 /*
1171  * Does @blk's attached device model have a tray, and is it open?
1172  */
1173 bool blk_dev_is_tray_open(BlockBackend *blk)
1174 {
1175     IO_CODE();
1176     if (blk_dev_has_tray(blk)) {
1177         return blk->dev_ops->is_tray_open(blk->dev_opaque);
1178     }
1179     return false;
1180 }
1181
1182 /*
1183  * Does @blk's attached device model have the medium locked?
1184  * %false if the device model has no such lock.
1185  */
1186 bool blk_dev_is_medium_locked(BlockBackend *blk)
1187 {
1188     GLOBAL_STATE_CODE();
1189     if (blk->dev_ops && blk->dev_ops->is_medium_locked) {
1190         return blk->dev_ops->is_medium_locked(blk->dev_opaque);
1191     }
1192     return false;
1193 }
1194
1195 /*
1196  * Notify @blk's attached device model of a backend size change.
1197  */
1198 static void blk_root_resize(BdrvChild *child)
1199 {
1200     BlockBackend *blk = child->opaque;
1201
1202     if (blk->dev_ops && blk->dev_ops->resize_cb) {
1203         blk->dev_ops->resize_cb(blk->dev_opaque);
1204     }
1205 }
1206
1207 void blk_iostatus_enable(BlockBackend *blk)
1208 {
1209     GLOBAL_STATE_CODE();
1210     blk->iostatus_enabled = true;
1211     blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
1212 }
1213
1214 /* The I/O status is only enabled if the drive explicitly
1215  * enables it _and_ the VM is configured to stop on errors */
1216 bool blk_iostatus_is_enabled(const BlockBackend *blk)
1217 {
1218     IO_CODE();
1219     return (blk->iostatus_enabled &&
1220            (blk->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
1221             blk->on_write_error == BLOCKDEV_ON_ERROR_STOP   ||
1222             blk->on_read_error == BLOCKDEV_ON_ERROR_STOP));
1223 }
1224
1225 BlockDeviceIoStatus blk_iostatus(const BlockBackend *blk)
1226 {
1227     GLOBAL_STATE_CODE();
1228     return blk->iostatus;
1229 }
1230
1231 void blk_iostatus_disable(BlockBackend *blk)
1232 {
1233     GLOBAL_STATE_CODE();
1234     blk->iostatus_enabled = false;
1235 }
1236
1237 void blk_iostatus_reset(BlockBackend *blk)
1238 {
1239     GLOBAL_STATE_CODE();
1240     if (blk_iostatus_is_enabled(blk)) {
1241         blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
1242     }
1243 }
1244
1245 void blk_iostatus_set_err(BlockBackend *blk, int error)
1246 {
1247     IO_CODE();
1248     assert(blk_iostatus_is_enabled(blk));
1249     if (blk->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
1250         blk->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
1251                                           BLOCK_DEVICE_IO_STATUS_FAILED;
1252     }
1253 }
1254
1255 void blk_set_allow_write_beyond_eof(BlockBackend *blk, bool allow)
1256 {
1257     IO_CODE();
1258     blk->allow_write_beyond_eof = allow;
1259 }
1260
1261 void blk_set_allow_aio_context_change(BlockBackend *blk, bool allow)
1262 {
1263     IO_CODE();
1264     blk->allow_aio_context_change = allow;
1265 }
1266
1267 void blk_set_disable_request_queuing(BlockBackend *blk, bool disable)
1268 {
1269     IO_CODE();
1270     qatomic_set(&blk->disable_request_queuing, disable);
1271 }
1272
1273 static int coroutine_fn GRAPH_RDLOCK
1274 blk_check_byte_request(BlockBackend *blk, int64_t offset, int64_t bytes)
1275 {
1276     int64_t len;
1277
1278     if (bytes < 0) {
1279         return -EIO;
1280     }
1281
1282     if (!blk_co_is_available(blk)) {
1283         return -ENOMEDIUM;
1284     }
1285
1286     if (offset < 0) {
1287         return -EIO;
1288     }
1289
1290     if (!blk->allow_write_beyond_eof) {
1291         len = bdrv_co_getlength(blk_bs(blk));
1292         if (len < 0) {
1293             return len;
1294         }
1295
1296         if (offset > len || len - offset < bytes) {
1297             return -EIO;
1298         }
1299     }
1300
1301     return 0;
1302 }
1303
1304 /* Are we currently in a drained section? */
1305 bool blk_in_drain(BlockBackend *blk)
1306 {
1307     GLOBAL_STATE_CODE(); /* change to IO_OR_GS_CODE(), if necessary */
1308     return qatomic_read(&blk->quiesce_counter);
1309 }
1310
1311 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1312 static void coroutine_fn blk_wait_while_drained(BlockBackend *blk)
1313 {
1314     assert(blk->in_flight > 0);
1315
1316     if (qatomic_read(&blk->quiesce_counter) &&
1317         !qatomic_read(&blk->disable_request_queuing)) {
1318         /*
1319          * Take lock before decrementing in flight counter so main loop thread
1320          * waits for us to enqueue ourselves before it can leave the drained
1321          * section.
1322          */
1323         qemu_mutex_lock(&blk->queued_requests_lock);
1324         blk_dec_in_flight(blk);
1325         qemu_co_queue_wait(&blk->queued_requests, &blk->queued_requests_lock);
1326         blk_inc_in_flight(blk);
1327         qemu_mutex_unlock(&blk->queued_requests_lock);
1328     }
1329 }
1330
1331 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1332 static int coroutine_fn
1333 blk_co_do_preadv_part(BlockBackend *blk, int64_t offset, int64_t bytes,
1334                       QEMUIOVector *qiov, size_t qiov_offset,
1335                       BdrvRequestFlags flags)
1336 {
1337     int ret;
1338     BlockDriverState *bs;
1339     IO_CODE();
1340
1341     blk_wait_while_drained(blk);
1342     GRAPH_RDLOCK_GUARD();
1343
1344     /* Call blk_bs() only after waiting, the graph may have changed */
1345     bs = blk_bs(blk);
1346     trace_blk_co_preadv(blk, bs, offset, bytes, flags);
1347
1348     ret = blk_check_byte_request(blk, offset, bytes);
1349     if (ret < 0) {
1350         return ret;
1351     }
1352
1353     bdrv_inc_in_flight(bs);
1354
1355     /* throttling disk I/O */
1356     if (blk->public.throttle_group_member.throttle_state) {
1357         throttle_group_co_io_limits_intercept(&blk->public.throttle_group_member,
1358                 bytes, false);
1359     }
1360
1361     ret = bdrv_co_preadv_part(blk->root, offset, bytes, qiov, qiov_offset,
1362                               flags);
1363     bdrv_dec_in_flight(bs);
1364     return ret;
1365 }
1366
1367 int coroutine_fn blk_co_pread(BlockBackend *blk, int64_t offset, int64_t bytes,
1368                               void *buf, BdrvRequestFlags flags)
1369 {
1370     QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
1371     IO_OR_GS_CODE();
1372
1373     assert(bytes <= SIZE_MAX);
1374
1375     return blk_co_preadv(blk, offset, bytes, &qiov, flags);
1376 }
1377
1378 int coroutine_fn blk_co_preadv(BlockBackend *blk, int64_t offset,
1379                                int64_t bytes, QEMUIOVector *qiov,
1380                                BdrvRequestFlags flags)
1381 {
1382     int ret;
1383     IO_OR_GS_CODE();
1384
1385     blk_inc_in_flight(blk);
1386     ret = blk_co_do_preadv_part(blk, offset, bytes, qiov, 0, flags);
1387     blk_dec_in_flight(blk);
1388
1389     return ret;
1390 }
1391
1392 int coroutine_fn blk_co_preadv_part(BlockBackend *blk, int64_t offset,
1393                                     int64_t bytes, QEMUIOVector *qiov,
1394                                     size_t qiov_offset, BdrvRequestFlags flags)
1395 {
1396     int ret;
1397     IO_OR_GS_CODE();
1398
1399     blk_inc_in_flight(blk);
1400     ret = blk_co_do_preadv_part(blk, offset, bytes, qiov, qiov_offset, flags);
1401     blk_dec_in_flight(blk);
1402
1403     return ret;
1404 }
1405
1406 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1407 static int coroutine_fn
1408 blk_co_do_pwritev_part(BlockBackend *blk, int64_t offset, int64_t bytes,
1409                        QEMUIOVector *qiov, size_t qiov_offset,
1410                        BdrvRequestFlags flags)
1411 {
1412     int ret;
1413     BlockDriverState *bs;
1414     IO_CODE();
1415
1416     blk_wait_while_drained(blk);
1417     GRAPH_RDLOCK_GUARD();
1418
1419     /* Call blk_bs() only after waiting, the graph may have changed */
1420     bs = blk_bs(blk);
1421     trace_blk_co_pwritev(blk, bs, offset, bytes, flags);
1422
1423     ret = blk_check_byte_request(blk, offset, bytes);
1424     if (ret < 0) {
1425         return ret;
1426     }
1427
1428     bdrv_inc_in_flight(bs);
1429     /* throttling disk I/O */
1430     if (blk->public.throttle_group_member.throttle_state) {
1431         throttle_group_co_io_limits_intercept(&blk->public.throttle_group_member,
1432                 bytes, true);
1433     }
1434
1435     if (!blk->enable_write_cache) {
1436         flags |= BDRV_REQ_FUA;
1437     }
1438
1439     ret = bdrv_co_pwritev_part(blk->root, offset, bytes, qiov, qiov_offset,
1440                                flags);
1441     bdrv_dec_in_flight(bs);
1442     return ret;
1443 }
1444
1445 int coroutine_fn blk_co_pwritev_part(BlockBackend *blk, int64_t offset,
1446                                      int64_t bytes,
1447                                      QEMUIOVector *qiov, size_t qiov_offset,
1448                                      BdrvRequestFlags flags)
1449 {
1450     int ret;
1451     IO_OR_GS_CODE();
1452
1453     blk_inc_in_flight(blk);
1454     ret = blk_co_do_pwritev_part(blk, offset, bytes, qiov, qiov_offset, flags);
1455     blk_dec_in_flight(blk);
1456
1457     return ret;
1458 }
1459
1460 int coroutine_fn blk_co_pwrite(BlockBackend *blk, int64_t offset, int64_t bytes,
1461                                const void *buf, BdrvRequestFlags flags)
1462 {
1463     QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
1464     IO_OR_GS_CODE();
1465
1466     assert(bytes <= SIZE_MAX);
1467
1468     return blk_co_pwritev(blk, offset, bytes, &qiov, flags);
1469 }
1470
1471 int coroutine_fn blk_co_pwritev(BlockBackend *blk, int64_t offset,
1472                                 int64_t bytes, QEMUIOVector *qiov,
1473                                 BdrvRequestFlags flags)
1474 {
1475     IO_OR_GS_CODE();
1476     return blk_co_pwritev_part(blk, offset, bytes, qiov, 0, flags);
1477 }
1478
1479 int coroutine_fn blk_co_block_status_above(BlockBackend *blk,
1480                                            BlockDriverState *base,
1481                                            int64_t offset, int64_t bytes,
1482                                            int64_t *pnum, int64_t *map,
1483                                            BlockDriverState **file)
1484 {
1485     IO_CODE();
1486     GRAPH_RDLOCK_GUARD();
1487     return bdrv_co_block_status_above(blk_bs(blk), base, offset, bytes, pnum,
1488                                       map, file);
1489 }
1490
1491 int coroutine_fn blk_co_is_allocated_above(BlockBackend *blk,
1492                                            BlockDriverState *base,
1493                                            bool include_base, int64_t offset,
1494                                            int64_t bytes, int64_t *pnum)
1495 {
1496     IO_CODE();
1497     GRAPH_RDLOCK_GUARD();
1498     return bdrv_co_is_allocated_above(blk_bs(blk), base, include_base, offset,
1499                                       bytes, pnum);
1500 }
1501
1502 typedef struct BlkRwCo {
1503     BlockBackend *blk;
1504     int64_t offset;
1505     void *iobuf;
1506     int ret;
1507     BdrvRequestFlags flags;
1508 } BlkRwCo;
1509
1510 int blk_make_zero(BlockBackend *blk, BdrvRequestFlags flags)
1511 {
1512     GLOBAL_STATE_CODE();
1513     return bdrv_make_zero(blk->root, flags);
1514 }
1515
1516 void blk_inc_in_flight(BlockBackend *blk)
1517 {
1518     IO_CODE();
1519     qatomic_inc(&blk->in_flight);
1520 }
1521
1522 void blk_dec_in_flight(BlockBackend *blk)
1523 {
1524     IO_CODE();
1525     qatomic_dec(&blk->in_flight);
1526     aio_wait_kick();
1527 }
1528
1529 static void error_callback_bh(void *opaque)
1530 {
1531     struct BlockBackendAIOCB *acb = opaque;
1532
1533     blk_dec_in_flight(acb->blk);
1534     acb->common.cb(acb->common.opaque, acb->ret);
1535     qemu_aio_unref(acb);
1536 }
1537
1538 BlockAIOCB *blk_abort_aio_request(BlockBackend *blk,
1539                                   BlockCompletionFunc *cb,
1540                                   void *opaque, int ret)
1541 {
1542     struct BlockBackendAIOCB *acb;
1543     IO_CODE();
1544
1545     blk_inc_in_flight(blk);
1546     acb = blk_aio_get(&block_backend_aiocb_info, blk, cb, opaque);
1547     acb->blk = blk;
1548     acb->ret = ret;
1549
1550     replay_bh_schedule_oneshot_event(blk_get_aio_context(blk),
1551                                      error_callback_bh, acb);
1552     return &acb->common;
1553 }
1554
1555 typedef struct BlkAioEmAIOCB {
1556     BlockAIOCB common;
1557     BlkRwCo rwco;
1558     int64_t bytes;
1559     bool has_returned;
1560 } BlkAioEmAIOCB;
1561
1562 static AioContext *blk_aio_em_aiocb_get_aio_context(BlockAIOCB *acb_)
1563 {
1564     BlkAioEmAIOCB *acb = container_of(acb_, BlkAioEmAIOCB, common);
1565
1566     return blk_get_aio_context(acb->rwco.blk);
1567 }
1568
1569 static const AIOCBInfo blk_aio_em_aiocb_info = {
1570     .aiocb_size         = sizeof(BlkAioEmAIOCB),
1571     .get_aio_context    = blk_aio_em_aiocb_get_aio_context,
1572 };
1573
1574 static void blk_aio_complete(BlkAioEmAIOCB *acb)
1575 {
1576     if (acb->has_returned) {
1577         acb->common.cb(acb->common.opaque, acb->rwco.ret);
1578         blk_dec_in_flight(acb->rwco.blk);
1579         qemu_aio_unref(acb);
1580     }
1581 }
1582
1583 static void blk_aio_complete_bh(void *opaque)
1584 {
1585     BlkAioEmAIOCB *acb = opaque;
1586     assert(acb->has_returned);
1587     blk_aio_complete(acb);
1588 }
1589
1590 static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset,
1591                                 int64_t bytes,
1592                                 void *iobuf, CoroutineEntry co_entry,
1593                                 BdrvRequestFlags flags,
1594                                 BlockCompletionFunc *cb, void *opaque)
1595 {
1596     BlkAioEmAIOCB *acb;
1597     Coroutine *co;
1598
1599     blk_inc_in_flight(blk);
1600     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1601     acb->rwco = (BlkRwCo) {
1602         .blk    = blk,
1603         .offset = offset,
1604         .iobuf  = iobuf,
1605         .flags  = flags,
1606         .ret    = NOT_DONE,
1607     };
1608     acb->bytes = bytes;
1609     acb->has_returned = false;
1610
1611     co = qemu_coroutine_create(co_entry, acb);
1612     aio_co_enter(blk_get_aio_context(blk), co);
1613
1614     acb->has_returned = true;
1615     if (acb->rwco.ret != NOT_DONE) {
1616         replay_bh_schedule_oneshot_event(blk_get_aio_context(blk),
1617                                          blk_aio_complete_bh, acb);
1618     }
1619
1620     return &acb->common;
1621 }
1622
1623 static void coroutine_fn blk_aio_read_entry(void *opaque)
1624 {
1625     BlkAioEmAIOCB *acb = opaque;
1626     BlkRwCo *rwco = &acb->rwco;
1627     QEMUIOVector *qiov = rwco->iobuf;
1628
1629     assert(qiov->size == acb->bytes);
1630     rwco->ret = blk_co_do_preadv_part(rwco->blk, rwco->offset, acb->bytes, qiov,
1631                                       0, rwco->flags);
1632     blk_aio_complete(acb);
1633 }
1634
1635 static void coroutine_fn blk_aio_write_entry(void *opaque)
1636 {
1637     BlkAioEmAIOCB *acb = opaque;
1638     BlkRwCo *rwco = &acb->rwco;
1639     QEMUIOVector *qiov = rwco->iobuf;
1640
1641     assert(!qiov || qiov->size == acb->bytes);
1642     rwco->ret = blk_co_do_pwritev_part(rwco->blk, rwco->offset, acb->bytes,
1643                                        qiov, 0, rwco->flags);
1644     blk_aio_complete(acb);
1645 }
1646
1647 BlockAIOCB *blk_aio_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1648                                   int64_t bytes, BdrvRequestFlags flags,
1649                                   BlockCompletionFunc *cb, void *opaque)
1650 {
1651     IO_CODE();
1652     return blk_aio_prwv(blk, offset, bytes, NULL, blk_aio_write_entry,
1653                         flags | BDRV_REQ_ZERO_WRITE, cb, opaque);
1654 }
1655
1656 int64_t coroutine_fn blk_co_getlength(BlockBackend *blk)
1657 {
1658     IO_CODE();
1659     GRAPH_RDLOCK_GUARD();
1660
1661     if (!blk_co_is_available(blk)) {
1662         return -ENOMEDIUM;
1663     }
1664
1665     return bdrv_co_getlength(blk_bs(blk));
1666 }
1667
1668 int64_t coroutine_fn blk_co_nb_sectors(BlockBackend *blk)
1669 {
1670     BlockDriverState *bs = blk_bs(blk);
1671
1672     IO_CODE();
1673     GRAPH_RDLOCK_GUARD();
1674
1675     if (!bs) {
1676         return -ENOMEDIUM;
1677     } else {
1678         return bdrv_co_nb_sectors(bs);
1679     }
1680 }
1681
1682 /*
1683  * This wrapper is written by hand because this function is in the hot I/O path,
1684  * via blk_get_geometry.
1685  */
1686 int64_t coroutine_mixed_fn blk_nb_sectors(BlockBackend *blk)
1687 {
1688     BlockDriverState *bs = blk_bs(blk);
1689
1690     IO_CODE();
1691
1692     if (!bs) {
1693         return -ENOMEDIUM;
1694     } else {
1695         return bdrv_nb_sectors(bs);
1696     }
1697 }
1698
1699 /* return 0 as number of sectors if no device present or error */
1700 void coroutine_fn blk_co_get_geometry(BlockBackend *blk,
1701                                       uint64_t *nb_sectors_ptr)
1702 {
1703     int64_t ret = blk_co_nb_sectors(blk);
1704     *nb_sectors_ptr = ret < 0 ? 0 : ret;
1705 }
1706
1707 /*
1708  * This wrapper is written by hand because this function is in the hot I/O path.
1709  */
1710 void coroutine_mixed_fn blk_get_geometry(BlockBackend *blk,
1711                                          uint64_t *nb_sectors_ptr)
1712 {
1713     int64_t ret = blk_nb_sectors(blk);
1714     *nb_sectors_ptr = ret < 0 ? 0 : ret;
1715 }
1716
1717 BlockAIOCB *blk_aio_preadv(BlockBackend *blk, int64_t offset,
1718                            QEMUIOVector *qiov, BdrvRequestFlags flags,
1719                            BlockCompletionFunc *cb, void *opaque)
1720 {
1721     IO_CODE();
1722     assert((uint64_t)qiov->size <= INT64_MAX);
1723     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1724                         blk_aio_read_entry, flags, cb, opaque);
1725 }
1726
1727 BlockAIOCB *blk_aio_pwritev(BlockBackend *blk, int64_t offset,
1728                             QEMUIOVector *qiov, BdrvRequestFlags flags,
1729                             BlockCompletionFunc *cb, void *opaque)
1730 {
1731     IO_CODE();
1732     assert((uint64_t)qiov->size <= INT64_MAX);
1733     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1734                         blk_aio_write_entry, flags, cb, opaque);
1735 }
1736
1737 void blk_aio_cancel(BlockAIOCB *acb)
1738 {
1739     GLOBAL_STATE_CODE();
1740     bdrv_aio_cancel(acb);
1741 }
1742
1743 void blk_aio_cancel_async(BlockAIOCB *acb)
1744 {
1745     IO_CODE();
1746     bdrv_aio_cancel_async(acb);
1747 }
1748
1749 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1750 static int coroutine_fn
1751 blk_co_do_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1752 {
1753     IO_CODE();
1754
1755     blk_wait_while_drained(blk);
1756     GRAPH_RDLOCK_GUARD();
1757
1758     if (!blk_co_is_available(blk)) {
1759         return -ENOMEDIUM;
1760     }
1761
1762     return bdrv_co_ioctl(blk_bs(blk), req, buf);
1763 }
1764
1765 int coroutine_fn blk_co_ioctl(BlockBackend *blk, unsigned long int req,
1766                               void *buf)
1767 {
1768     int ret;
1769     IO_OR_GS_CODE();
1770
1771     blk_inc_in_flight(blk);
1772     ret = blk_co_do_ioctl(blk, req, buf);
1773     blk_dec_in_flight(blk);
1774
1775     return ret;
1776 }
1777
1778 static void coroutine_fn blk_aio_ioctl_entry(void *opaque)
1779 {
1780     BlkAioEmAIOCB *acb = opaque;
1781     BlkRwCo *rwco = &acb->rwco;
1782
1783     rwco->ret = blk_co_do_ioctl(rwco->blk, rwco->offset, rwco->iobuf);
1784
1785     blk_aio_complete(acb);
1786 }
1787
1788 BlockAIOCB *blk_aio_ioctl(BlockBackend *blk, unsigned long int req, void *buf,
1789                           BlockCompletionFunc *cb, void *opaque)
1790 {
1791     IO_CODE();
1792     return blk_aio_prwv(blk, req, 0, buf, blk_aio_ioctl_entry, 0, cb, opaque);
1793 }
1794
1795 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1796 static int coroutine_fn
1797 blk_co_do_pdiscard(BlockBackend *blk, int64_t offset, int64_t bytes)
1798 {
1799     int ret;
1800     IO_CODE();
1801
1802     blk_wait_while_drained(blk);
1803     GRAPH_RDLOCK_GUARD();
1804
1805     ret = blk_check_byte_request(blk, offset, bytes);
1806     if (ret < 0) {
1807         return ret;
1808     }
1809
1810     return bdrv_co_pdiscard(blk->root, offset, bytes);
1811 }
1812
1813 static void coroutine_fn blk_aio_pdiscard_entry(void *opaque)
1814 {
1815     BlkAioEmAIOCB *acb = opaque;
1816     BlkRwCo *rwco = &acb->rwco;
1817
1818     rwco->ret = blk_co_do_pdiscard(rwco->blk, rwco->offset, acb->bytes);
1819     blk_aio_complete(acb);
1820 }
1821
1822 BlockAIOCB *blk_aio_pdiscard(BlockBackend *blk,
1823                              int64_t offset, int64_t bytes,
1824                              BlockCompletionFunc *cb, void *opaque)
1825 {
1826     IO_CODE();
1827     return blk_aio_prwv(blk, offset, bytes, NULL, blk_aio_pdiscard_entry, 0,
1828                         cb, opaque);
1829 }
1830
1831 int coroutine_fn blk_co_pdiscard(BlockBackend *blk, int64_t offset,
1832                                  int64_t bytes)
1833 {
1834     int ret;
1835     IO_OR_GS_CODE();
1836
1837     blk_inc_in_flight(blk);
1838     ret = blk_co_do_pdiscard(blk, offset, bytes);
1839     blk_dec_in_flight(blk);
1840
1841     return ret;
1842 }
1843
1844 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1845 static int coroutine_fn blk_co_do_flush(BlockBackend *blk)
1846 {
1847     IO_CODE();
1848     blk_wait_while_drained(blk);
1849     GRAPH_RDLOCK_GUARD();
1850
1851     if (!blk_co_is_available(blk)) {
1852         return -ENOMEDIUM;
1853     }
1854
1855     return bdrv_co_flush(blk_bs(blk));
1856 }
1857
1858 static void coroutine_fn blk_aio_flush_entry(void *opaque)
1859 {
1860     BlkAioEmAIOCB *acb = opaque;
1861     BlkRwCo *rwco = &acb->rwco;
1862
1863     rwco->ret = blk_co_do_flush(rwco->blk);
1864     blk_aio_complete(acb);
1865 }
1866
1867 BlockAIOCB *blk_aio_flush(BlockBackend *blk,
1868                           BlockCompletionFunc *cb, void *opaque)
1869 {
1870     IO_CODE();
1871     return blk_aio_prwv(blk, 0, 0, NULL, blk_aio_flush_entry, 0, cb, opaque);
1872 }
1873
1874 int coroutine_fn blk_co_flush(BlockBackend *blk)
1875 {
1876     int ret;
1877     IO_OR_GS_CODE();
1878
1879     blk_inc_in_flight(blk);
1880     ret = blk_co_do_flush(blk);
1881     blk_dec_in_flight(blk);
1882
1883     return ret;
1884 }
1885
1886 static void coroutine_fn blk_aio_zone_report_entry(void *opaque)
1887 {
1888     BlkAioEmAIOCB *acb = opaque;
1889     BlkRwCo *rwco = &acb->rwco;
1890
1891     rwco->ret = blk_co_zone_report(rwco->blk, rwco->offset,
1892                                    (unsigned int*)(uintptr_t)acb->bytes,
1893                                    rwco->iobuf);
1894     blk_aio_complete(acb);
1895 }
1896
1897 BlockAIOCB *blk_aio_zone_report(BlockBackend *blk, int64_t offset,
1898                                 unsigned int *nr_zones,
1899                                 BlockZoneDescriptor  *zones,
1900                                 BlockCompletionFunc *cb, void *opaque)
1901 {
1902     BlkAioEmAIOCB *acb;
1903     Coroutine *co;
1904     IO_CODE();
1905
1906     blk_inc_in_flight(blk);
1907     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1908     acb->rwco = (BlkRwCo) {
1909         .blk    = blk,
1910         .offset = offset,
1911         .iobuf  = zones,
1912         .ret    = NOT_DONE,
1913     };
1914     acb->bytes = (int64_t)(uintptr_t)nr_zones,
1915     acb->has_returned = false;
1916
1917     co = qemu_coroutine_create(blk_aio_zone_report_entry, acb);
1918     aio_co_enter(blk_get_aio_context(blk), co);
1919
1920     acb->has_returned = true;
1921     if (acb->rwco.ret != NOT_DONE) {
1922         replay_bh_schedule_oneshot_event(blk_get_aio_context(blk),
1923                                          blk_aio_complete_bh, acb);
1924     }
1925
1926     return &acb->common;
1927 }
1928
1929 static void coroutine_fn blk_aio_zone_mgmt_entry(void *opaque)
1930 {
1931     BlkAioEmAIOCB *acb = opaque;
1932     BlkRwCo *rwco = &acb->rwco;
1933
1934     rwco->ret = blk_co_zone_mgmt(rwco->blk,
1935                                  (BlockZoneOp)(uintptr_t)rwco->iobuf,
1936                                  rwco->offset, acb->bytes);
1937     blk_aio_complete(acb);
1938 }
1939
1940 BlockAIOCB *blk_aio_zone_mgmt(BlockBackend *blk, BlockZoneOp op,
1941                               int64_t offset, int64_t len,
1942                               BlockCompletionFunc *cb, void *opaque) {
1943     BlkAioEmAIOCB *acb;
1944     Coroutine *co;
1945     IO_CODE();
1946
1947     blk_inc_in_flight(blk);
1948     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1949     acb->rwco = (BlkRwCo) {
1950         .blk    = blk,
1951         .offset = offset,
1952         .iobuf  = (void *)(uintptr_t)op,
1953         .ret    = NOT_DONE,
1954     };
1955     acb->bytes = len;
1956     acb->has_returned = false;
1957
1958     co = qemu_coroutine_create(blk_aio_zone_mgmt_entry, acb);
1959     aio_co_enter(blk_get_aio_context(blk), co);
1960
1961     acb->has_returned = true;
1962     if (acb->rwco.ret != NOT_DONE) {
1963         replay_bh_schedule_oneshot_event(blk_get_aio_context(blk),
1964                                          blk_aio_complete_bh, acb);
1965     }
1966
1967     return &acb->common;
1968 }
1969
1970 static void coroutine_fn blk_aio_zone_append_entry(void *opaque)
1971 {
1972     BlkAioEmAIOCB *acb = opaque;
1973     BlkRwCo *rwco = &acb->rwco;
1974
1975     rwco->ret = blk_co_zone_append(rwco->blk, (int64_t *)(uintptr_t)acb->bytes,
1976                                    rwco->iobuf, rwco->flags);
1977     blk_aio_complete(acb);
1978 }
1979
1980 BlockAIOCB *blk_aio_zone_append(BlockBackend *blk, int64_t *offset,
1981                                 QEMUIOVector *qiov, BdrvRequestFlags flags,
1982                                 BlockCompletionFunc *cb, void *opaque) {
1983     BlkAioEmAIOCB *acb;
1984     Coroutine *co;
1985     IO_CODE();
1986
1987     blk_inc_in_flight(blk);
1988     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1989     acb->rwco = (BlkRwCo) {
1990         .blk    = blk,
1991         .ret    = NOT_DONE,
1992         .flags  = flags,
1993         .iobuf  = qiov,
1994     };
1995     acb->bytes = (int64_t)(uintptr_t)offset;
1996     acb->has_returned = false;
1997
1998     co = qemu_coroutine_create(blk_aio_zone_append_entry, acb);
1999     aio_co_enter(blk_get_aio_context(blk), co);
2000     acb->has_returned = true;
2001     if (acb->rwco.ret != NOT_DONE) {
2002         replay_bh_schedule_oneshot_event(blk_get_aio_context(blk),
2003                                          blk_aio_complete_bh, acb);
2004     }
2005
2006     return &acb->common;
2007 }
2008
2009 /*
2010  * Send a zone_report command.
2011  * offset is a byte offset from the start of the device. No alignment
2012  * required for offset.
2013  * nr_zones represents IN maximum and OUT actual.
2014  */
2015 int coroutine_fn blk_co_zone_report(BlockBackend *blk, int64_t offset,
2016                                     unsigned int *nr_zones,
2017                                     BlockZoneDescriptor *zones)
2018 {
2019     int ret;
2020     IO_CODE();
2021
2022     blk_inc_in_flight(blk); /* increase before waiting */
2023     blk_wait_while_drained(blk);
2024     GRAPH_RDLOCK_GUARD();
2025     if (!blk_is_available(blk)) {
2026         blk_dec_in_flight(blk);
2027         return -ENOMEDIUM;
2028     }
2029     ret = bdrv_co_zone_report(blk_bs(blk), offset, nr_zones, zones);
2030     blk_dec_in_flight(blk);
2031     return ret;
2032 }
2033
2034 /*
2035  * Send a zone_management command.
2036  * op is the zone operation;
2037  * offset is the byte offset from the start of the zoned device;
2038  * len is the maximum number of bytes the command should operate on. It
2039  * should be aligned with the device zone size.
2040  */
2041 int coroutine_fn blk_co_zone_mgmt(BlockBackend *blk, BlockZoneOp op,
2042         int64_t offset, int64_t len)
2043 {
2044     int ret;
2045     IO_CODE();
2046
2047     blk_inc_in_flight(blk);
2048     blk_wait_while_drained(blk);
2049     GRAPH_RDLOCK_GUARD();
2050
2051     ret = blk_check_byte_request(blk, offset, len);
2052     if (ret < 0) {
2053         blk_dec_in_flight(blk);
2054         return ret;
2055     }
2056
2057     ret = bdrv_co_zone_mgmt(blk_bs(blk), op, offset, len);
2058     blk_dec_in_flight(blk);
2059     return ret;
2060 }
2061
2062 /*
2063  * Send a zone_append command.
2064  */
2065 int coroutine_fn blk_co_zone_append(BlockBackend *blk, int64_t *offset,
2066         QEMUIOVector *qiov, BdrvRequestFlags flags)
2067 {
2068     int ret;
2069     IO_CODE();
2070
2071     blk_inc_in_flight(blk);
2072     blk_wait_while_drained(blk);
2073     GRAPH_RDLOCK_GUARD();
2074     if (!blk_is_available(blk)) {
2075         blk_dec_in_flight(blk);
2076         return -ENOMEDIUM;
2077     }
2078
2079     ret = bdrv_co_zone_append(blk_bs(blk), offset, qiov, flags);
2080     blk_dec_in_flight(blk);
2081     return ret;
2082 }
2083
2084 void blk_drain(BlockBackend *blk)
2085 {
2086     BlockDriverState *bs = blk_bs(blk);
2087     GLOBAL_STATE_CODE();
2088
2089     if (bs) {
2090         bdrv_ref(bs);
2091         bdrv_drained_begin(bs);
2092     }
2093
2094     /* We may have -ENOMEDIUM completions in flight */
2095     AIO_WAIT_WHILE(blk_get_aio_context(blk),
2096                    qatomic_read(&blk->in_flight) > 0);
2097
2098     if (bs) {
2099         bdrv_drained_end(bs);
2100         bdrv_unref(bs);
2101     }
2102 }
2103
2104 void blk_drain_all(void)
2105 {
2106     BlockBackend *blk = NULL;
2107
2108     GLOBAL_STATE_CODE();
2109
2110     bdrv_drain_all_begin();
2111
2112     while ((blk = blk_all_next(blk)) != NULL) {
2113         /* We may have -ENOMEDIUM completions in flight */
2114         AIO_WAIT_WHILE_UNLOCKED(NULL, qatomic_read(&blk->in_flight) > 0);
2115     }
2116
2117     bdrv_drain_all_end();
2118 }
2119
2120 void blk_set_on_error(BlockBackend *blk, BlockdevOnError on_read_error,
2121                       BlockdevOnError on_write_error)
2122 {
2123     GLOBAL_STATE_CODE();
2124     blk->on_read_error = on_read_error;
2125     blk->on_write_error = on_write_error;
2126 }
2127
2128 BlockdevOnError blk_get_on_error(BlockBackend *blk, bool is_read)
2129 {
2130     IO_CODE();
2131     return is_read ? blk->on_read_error : blk->on_write_error;
2132 }
2133
2134 BlockErrorAction blk_get_error_action(BlockBackend *blk, bool is_read,
2135                                       int error)
2136 {
2137     BlockdevOnError on_err = blk_get_on_error(blk, is_read);
2138     IO_CODE();
2139
2140     switch (on_err) {
2141     case BLOCKDEV_ON_ERROR_ENOSPC:
2142         return (error == ENOSPC) ?
2143                BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
2144     case BLOCKDEV_ON_ERROR_STOP:
2145         return BLOCK_ERROR_ACTION_STOP;
2146     case BLOCKDEV_ON_ERROR_REPORT:
2147         return BLOCK_ERROR_ACTION_REPORT;
2148     case BLOCKDEV_ON_ERROR_IGNORE:
2149         return BLOCK_ERROR_ACTION_IGNORE;
2150     case BLOCKDEV_ON_ERROR_AUTO:
2151     default:
2152         abort();
2153     }
2154 }
2155
2156 static void send_qmp_error_event(BlockBackend *blk,
2157                                  BlockErrorAction action,
2158                                  bool is_read, int error)
2159 {
2160     IoOperationType optype;
2161     BlockDriverState *bs = blk_bs(blk);
2162
2163     optype = is_read ? IO_OPERATION_TYPE_READ : IO_OPERATION_TYPE_WRITE;
2164     qapi_event_send_block_io_error(blk_name(blk),
2165                                    bs ? bdrv_get_node_name(bs) : NULL, optype,
2166                                    action, blk_iostatus_is_enabled(blk),
2167                                    error == ENOSPC, strerror(error));
2168 }
2169
2170 /* This is done by device models because, while the block layer knows
2171  * about the error, it does not know whether an operation comes from
2172  * the device or the block layer (from a job, for example).
2173  */
2174 void blk_error_action(BlockBackend *blk, BlockErrorAction action,
2175                       bool is_read, int error)
2176 {
2177     assert(error >= 0);
2178     IO_CODE();
2179
2180     if (action == BLOCK_ERROR_ACTION_STOP) {
2181         /* First set the iostatus, so that "info block" returns an iostatus
2182          * that matches the events raised so far (an additional error iostatus
2183          * is fine, but not a lost one).
2184          */
2185         blk_iostatus_set_err(blk, error);
2186
2187         /* Then raise the request to stop the VM and the event.
2188          * qemu_system_vmstop_request_prepare has two effects.  First,
2189          * it ensures that the STOP event always comes after the
2190          * BLOCK_IO_ERROR event.  Second, it ensures that even if management
2191          * can observe the STOP event and do a "cont" before the STOP
2192          * event is issued, the VM will not stop.  In this case, vm_start()
2193          * also ensures that the STOP/RESUME pair of events is emitted.
2194          */
2195         qemu_system_vmstop_request_prepare();
2196         send_qmp_error_event(blk, action, is_read, error);
2197         qemu_system_vmstop_request(RUN_STATE_IO_ERROR);
2198     } else {
2199         send_qmp_error_event(blk, action, is_read, error);
2200     }
2201 }
2202
2203 /*
2204  * Returns true if the BlockBackend can support taking write permissions
2205  * (because its root node is not read-only).
2206  */
2207 bool blk_supports_write_perm(BlockBackend *blk)
2208 {
2209     BlockDriverState *bs = blk_bs(blk);
2210     GLOBAL_STATE_CODE();
2211
2212     if (bs) {
2213         return !bdrv_is_read_only(bs);
2214     } else {
2215         return blk->root_state.open_flags & BDRV_O_RDWR;
2216     }
2217 }
2218
2219 /*
2220  * Returns true if the BlockBackend can be written to in its current
2221  * configuration (i.e. if write permission have been requested)
2222  */
2223 bool blk_is_writable(BlockBackend *blk)
2224 {
2225     IO_CODE();
2226     return blk->perm & BLK_PERM_WRITE;
2227 }
2228
2229 bool blk_is_sg(BlockBackend *blk)
2230 {
2231     BlockDriverState *bs = blk_bs(blk);
2232     GLOBAL_STATE_CODE();
2233
2234     if (!bs) {
2235         return false;
2236     }
2237
2238     return bdrv_is_sg(bs);
2239 }
2240
2241 bool blk_enable_write_cache(BlockBackend *blk)
2242 {
2243     IO_CODE();
2244     return blk->enable_write_cache;
2245 }
2246
2247 void blk_set_enable_write_cache(BlockBackend *blk, bool wce)
2248 {
2249     IO_CODE();
2250     blk->enable_write_cache = wce;
2251 }
2252
2253 void blk_activate(BlockBackend *blk, Error **errp)
2254 {
2255     BlockDriverState *bs = blk_bs(blk);
2256     GLOBAL_STATE_CODE();
2257
2258     if (!bs) {
2259         error_setg(errp, "Device '%s' has no medium", blk->name);
2260         return;
2261     }
2262
2263     /*
2264      * Migration code can call this function in coroutine context, so leave
2265      * coroutine context if necessary.
2266      */
2267     if (qemu_in_coroutine()) {
2268         bdrv_co_activate(bs, errp);
2269     } else {
2270         bdrv_activate(bs, errp);
2271     }
2272 }
2273
2274 bool coroutine_fn blk_co_is_inserted(BlockBackend *blk)
2275 {
2276     BlockDriverState *bs = blk_bs(blk);
2277     IO_CODE();
2278     assert_bdrv_graph_readable();
2279
2280     return bs && bdrv_co_is_inserted(bs);
2281 }
2282
2283 bool coroutine_fn blk_co_is_available(BlockBackend *blk)
2284 {
2285     IO_CODE();
2286     return blk_co_is_inserted(blk) && !blk_dev_is_tray_open(blk);
2287 }
2288
2289 void coroutine_fn blk_co_lock_medium(BlockBackend *blk, bool locked)
2290 {
2291     BlockDriverState *bs = blk_bs(blk);
2292     IO_CODE();
2293     GRAPH_RDLOCK_GUARD();
2294
2295     if (bs) {
2296         bdrv_co_lock_medium(bs, locked);
2297     }
2298 }
2299
2300 void coroutine_fn blk_co_eject(BlockBackend *blk, bool eject_flag)
2301 {
2302     BlockDriverState *bs = blk_bs(blk);
2303     char *id;
2304     IO_CODE();
2305     GRAPH_RDLOCK_GUARD();
2306
2307     if (bs) {
2308         bdrv_co_eject(bs, eject_flag);
2309     }
2310
2311     /* Whether or not we ejected on the backend,
2312      * the frontend experienced a tray event. */
2313     id = blk_get_attached_dev_id(blk);
2314     qapi_event_send_device_tray_moved(blk_name(blk), id,
2315                                       eject_flag);
2316     g_free(id);
2317 }
2318
2319 int blk_get_flags(BlockBackend *blk)
2320 {
2321     BlockDriverState *bs = blk_bs(blk);
2322     GLOBAL_STATE_CODE();
2323
2324     if (bs) {
2325         return bdrv_get_flags(bs);
2326     } else {
2327         return blk->root_state.open_flags;
2328     }
2329 }
2330
2331 /* Returns the minimum request alignment, in bytes; guaranteed nonzero */
2332 uint32_t blk_get_request_alignment(BlockBackend *blk)
2333 {
2334     BlockDriverState *bs = blk_bs(blk);
2335     IO_CODE();
2336     return bs ? bs->bl.request_alignment : BDRV_SECTOR_SIZE;
2337 }
2338
2339 /* Returns the maximum hardware transfer length, in bytes; guaranteed nonzero */
2340 uint64_t blk_get_max_hw_transfer(BlockBackend *blk)
2341 {
2342     BlockDriverState *bs = blk_bs(blk);
2343     uint64_t max = INT_MAX;
2344     IO_CODE();
2345
2346     if (bs) {
2347         max = MIN_NON_ZERO(max, bs->bl.max_hw_transfer);
2348         max = MIN_NON_ZERO(max, bs->bl.max_transfer);
2349     }
2350     return ROUND_DOWN(max, blk_get_request_alignment(blk));
2351 }
2352
2353 /* Returns the maximum transfer length, in bytes; guaranteed nonzero */
2354 uint32_t blk_get_max_transfer(BlockBackend *blk)
2355 {
2356     BlockDriverState *bs = blk_bs(blk);
2357     uint32_t max = INT_MAX;
2358     IO_CODE();
2359
2360     if (bs) {
2361         max = MIN_NON_ZERO(max, bs->bl.max_transfer);
2362     }
2363     return ROUND_DOWN(max, blk_get_request_alignment(blk));
2364 }
2365
2366 int blk_get_max_hw_iov(BlockBackend *blk)
2367 {
2368     IO_CODE();
2369     return MIN_NON_ZERO(blk->root->bs->bl.max_hw_iov,
2370                         blk->root->bs->bl.max_iov);
2371 }
2372
2373 int blk_get_max_iov(BlockBackend *blk)
2374 {
2375     IO_CODE();
2376     return blk->root->bs->bl.max_iov;
2377 }
2378
2379 void *blk_try_blockalign(BlockBackend *blk, size_t size)
2380 {
2381     IO_CODE();
2382     return qemu_try_blockalign(blk ? blk_bs(blk) : NULL, size);
2383 }
2384
2385 void *blk_blockalign(BlockBackend *blk, size_t size)
2386 {
2387     IO_CODE();
2388     return qemu_blockalign(blk ? blk_bs(blk) : NULL, size);
2389 }
2390
2391 bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp)
2392 {
2393     BlockDriverState *bs = blk_bs(blk);
2394     GLOBAL_STATE_CODE();
2395
2396     if (!bs) {
2397         return false;
2398     }
2399
2400     return bdrv_op_is_blocked(bs, op, errp);
2401 }
2402
2403 void blk_op_unblock(BlockBackend *blk, BlockOpType op, Error *reason)
2404 {
2405     BlockDriverState *bs = blk_bs(blk);
2406     GLOBAL_STATE_CODE();
2407
2408     if (bs) {
2409         bdrv_op_unblock(bs, op, reason);
2410     }
2411 }
2412
2413 void blk_op_block_all(BlockBackend *blk, Error *reason)
2414 {
2415     BlockDriverState *bs = blk_bs(blk);
2416     GLOBAL_STATE_CODE();
2417
2418     if (bs) {
2419         bdrv_op_block_all(bs, reason);
2420     }
2421 }
2422
2423 void blk_op_unblock_all(BlockBackend *blk, Error *reason)
2424 {
2425     BlockDriverState *bs = blk_bs(blk);
2426     GLOBAL_STATE_CODE();
2427
2428     if (bs) {
2429         bdrv_op_unblock_all(bs, reason);
2430     }
2431 }
2432
2433 AioContext *blk_get_aio_context(BlockBackend *blk)
2434 {
2435     BlockDriverState *bs;
2436     IO_CODE();
2437
2438     if (!blk) {
2439         return qemu_get_aio_context();
2440     }
2441
2442     bs = blk_bs(blk);
2443     if (bs) {
2444         AioContext *ctx = bdrv_get_aio_context(blk_bs(blk));
2445         assert(ctx == blk->ctx);
2446     }
2447
2448     return blk->ctx;
2449 }
2450
2451 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb)
2452 {
2453     BlockBackendAIOCB *blk_acb = DO_UPCAST(BlockBackendAIOCB, common, acb);
2454     return blk_get_aio_context(blk_acb->blk);
2455 }
2456
2457 int blk_set_aio_context(BlockBackend *blk, AioContext *new_context,
2458                         Error **errp)
2459 {
2460     bool old_allow_change;
2461     BlockDriverState *bs = blk_bs(blk);
2462     int ret;
2463
2464     GLOBAL_STATE_CODE();
2465
2466     if (!bs) {
2467         blk->ctx = new_context;
2468         return 0;
2469     }
2470
2471     bdrv_ref(bs);
2472
2473     old_allow_change = blk->allow_aio_context_change;
2474     blk->allow_aio_context_change = true;
2475
2476     ret = bdrv_try_change_aio_context(bs, new_context, NULL, errp);
2477
2478     blk->allow_aio_context_change = old_allow_change;
2479
2480     bdrv_unref(bs);
2481     return ret;
2482 }
2483
2484 typedef struct BdrvStateBlkRootContext {
2485     AioContext *new_ctx;
2486     BlockBackend *blk;
2487 } BdrvStateBlkRootContext;
2488
2489 static void blk_root_set_aio_ctx_commit(void *opaque)
2490 {
2491     BdrvStateBlkRootContext *s = opaque;
2492     BlockBackend *blk = s->blk;
2493     AioContext *new_context = s->new_ctx;
2494     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
2495
2496     blk->ctx = new_context;
2497     if (tgm->throttle_state) {
2498         throttle_group_detach_aio_context(tgm);
2499         throttle_group_attach_aio_context(tgm, new_context);
2500     }
2501 }
2502
2503 static TransactionActionDrv set_blk_root_context = {
2504     .commit = blk_root_set_aio_ctx_commit,
2505     .clean = g_free,
2506 };
2507
2508 static bool blk_root_change_aio_ctx(BdrvChild *child, AioContext *ctx,
2509                                     GHashTable *visited, Transaction *tran,
2510                                     Error **errp)
2511 {
2512     BlockBackend *blk = child->opaque;
2513     BdrvStateBlkRootContext *s;
2514
2515     if (!blk->allow_aio_context_change) {
2516         /*
2517          * Manually created BlockBackends (those with a name) that are not
2518          * attached to anything can change their AioContext without updating
2519          * their user; return an error for others.
2520          */
2521         if (!blk->name || blk->dev) {
2522             /* TODO Add BB name/QOM path */
2523             error_setg(errp, "Cannot change iothread of active block backend");
2524             return false;
2525         }
2526     }
2527
2528     s = g_new(BdrvStateBlkRootContext, 1);
2529     *s = (BdrvStateBlkRootContext) {
2530         .new_ctx = ctx,
2531         .blk = blk,
2532     };
2533
2534     tran_add(tran, &set_blk_root_context, s);
2535     return true;
2536 }
2537
2538 void blk_add_aio_context_notifier(BlockBackend *blk,
2539         void (*attached_aio_context)(AioContext *new_context, void *opaque),
2540         void (*detach_aio_context)(void *opaque), void *opaque)
2541 {
2542     BlockBackendAioNotifier *notifier;
2543     BlockDriverState *bs = blk_bs(blk);
2544     GLOBAL_STATE_CODE();
2545
2546     notifier = g_new(BlockBackendAioNotifier, 1);
2547     notifier->attached_aio_context = attached_aio_context;
2548     notifier->detach_aio_context = detach_aio_context;
2549     notifier->opaque = opaque;
2550     QLIST_INSERT_HEAD(&blk->aio_notifiers, notifier, list);
2551
2552     if (bs) {
2553         bdrv_add_aio_context_notifier(bs, attached_aio_context,
2554                                       detach_aio_context, opaque);
2555     }
2556 }
2557
2558 void blk_remove_aio_context_notifier(BlockBackend *blk,
2559                                      void (*attached_aio_context)(AioContext *,
2560                                                                   void *),
2561                                      void (*detach_aio_context)(void *),
2562                                      void *opaque)
2563 {
2564     BlockBackendAioNotifier *notifier;
2565     BlockDriverState *bs = blk_bs(blk);
2566
2567     GLOBAL_STATE_CODE();
2568
2569     if (bs) {
2570         bdrv_remove_aio_context_notifier(bs, attached_aio_context,
2571                                          detach_aio_context, opaque);
2572     }
2573
2574     QLIST_FOREACH(notifier, &blk->aio_notifiers, list) {
2575         if (notifier->attached_aio_context == attached_aio_context &&
2576             notifier->detach_aio_context == detach_aio_context &&
2577             notifier->opaque == opaque) {
2578             QLIST_REMOVE(notifier, list);
2579             g_free(notifier);
2580             return;
2581         }
2582     }
2583
2584     abort();
2585 }
2586
2587 void blk_add_remove_bs_notifier(BlockBackend *blk, Notifier *notify)
2588 {
2589     GLOBAL_STATE_CODE();
2590     notifier_list_add(&blk->remove_bs_notifiers, notify);
2591 }
2592
2593 void blk_add_insert_bs_notifier(BlockBackend *blk, Notifier *notify)
2594 {
2595     GLOBAL_STATE_CODE();
2596     notifier_list_add(&blk->insert_bs_notifiers, notify);
2597 }
2598
2599 BlockAcctStats *blk_get_stats(BlockBackend *blk)
2600 {
2601     IO_CODE();
2602     return &blk->stats;
2603 }
2604
2605 void *blk_aio_get(const AIOCBInfo *aiocb_info, BlockBackend *blk,
2606                   BlockCompletionFunc *cb, void *opaque)
2607 {
2608     IO_CODE();
2609     return qemu_aio_get(aiocb_info, blk_bs(blk), cb, opaque);
2610 }
2611
2612 int coroutine_fn blk_co_pwrite_zeroes(BlockBackend *blk, int64_t offset,
2613                                       int64_t bytes, BdrvRequestFlags flags)
2614 {
2615     IO_OR_GS_CODE();
2616     return blk_co_pwritev(blk, offset, bytes, NULL,
2617                           flags | BDRV_REQ_ZERO_WRITE);
2618 }
2619
2620 int coroutine_fn blk_co_pwrite_compressed(BlockBackend *blk, int64_t offset,
2621                                           int64_t bytes, const void *buf)
2622 {
2623     QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
2624     IO_OR_GS_CODE();
2625     return blk_co_pwritev_part(blk, offset, bytes, &qiov, 0,
2626                                BDRV_REQ_WRITE_COMPRESSED);
2627 }
2628
2629 int coroutine_fn blk_co_truncate(BlockBackend *blk, int64_t offset, bool exact,
2630                                  PreallocMode prealloc, BdrvRequestFlags flags,
2631                                  Error **errp)
2632 {
2633     IO_OR_GS_CODE();
2634     GRAPH_RDLOCK_GUARD();
2635     if (!blk_co_is_available(blk)) {
2636         error_setg(errp, "No medium inserted");
2637         return -ENOMEDIUM;
2638     }
2639
2640     return bdrv_co_truncate(blk->root, offset, exact, prealloc, flags, errp);
2641 }
2642
2643 int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
2644                      int64_t pos, int size)
2645 {
2646     int ret;
2647     GLOBAL_STATE_CODE();
2648
2649     if (!blk_is_available(blk)) {
2650         return -ENOMEDIUM;
2651     }
2652
2653     ret = bdrv_save_vmstate(blk_bs(blk), buf, pos, size);
2654     if (ret < 0) {
2655         return ret;
2656     }
2657
2658     if (ret == size && !blk->enable_write_cache) {
2659         ret = bdrv_flush(blk_bs(blk));
2660     }
2661
2662     return ret < 0 ? ret : size;
2663 }
2664
2665 int blk_load_vmstate(BlockBackend *blk, uint8_t *buf, int64_t pos, int size)
2666 {
2667     GLOBAL_STATE_CODE();
2668     if (!blk_is_available(blk)) {
2669         return -ENOMEDIUM;
2670     }
2671
2672     return bdrv_load_vmstate(blk_bs(blk), buf, pos, size);
2673 }
2674
2675 int blk_probe_blocksizes(BlockBackend *blk, BlockSizes *bsz)
2676 {
2677     GLOBAL_STATE_CODE();
2678     if (!blk_is_available(blk)) {
2679         return -ENOMEDIUM;
2680     }
2681
2682     return bdrv_probe_blocksizes(blk_bs(blk), bsz);
2683 }
2684
2685 int blk_probe_geometry(BlockBackend *blk, HDGeometry *geo)
2686 {
2687     GLOBAL_STATE_CODE();
2688     if (!blk_is_available(blk)) {
2689         return -ENOMEDIUM;
2690     }
2691
2692     return bdrv_probe_geometry(blk_bs(blk), geo);
2693 }
2694
2695 /*
2696  * Updates the BlockBackendRootState object with data from the currently
2697  * attached BlockDriverState.
2698  */
2699 void blk_update_root_state(BlockBackend *blk)
2700 {
2701     GLOBAL_STATE_CODE();
2702     assert(blk->root);
2703
2704     blk->root_state.open_flags    = blk->root->bs->open_flags;
2705     blk->root_state.detect_zeroes = blk->root->bs->detect_zeroes;
2706 }
2707
2708 /*
2709  * Returns the detect-zeroes setting to be used for bdrv_open() of a
2710  * BlockDriverState which is supposed to inherit the root state.
2711  */
2712 bool blk_get_detect_zeroes_from_root_state(BlockBackend *blk)
2713 {
2714     GLOBAL_STATE_CODE();
2715     return blk->root_state.detect_zeroes;
2716 }
2717
2718 /*
2719  * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
2720  * supposed to inherit the root state.
2721  */
2722 int blk_get_open_flags_from_root_state(BlockBackend *blk)
2723 {
2724     GLOBAL_STATE_CODE();
2725     return blk->root_state.open_flags;
2726 }
2727
2728 BlockBackendRootState *blk_get_root_state(BlockBackend *blk)
2729 {
2730     GLOBAL_STATE_CODE();
2731     return &blk->root_state;
2732 }
2733
2734 int blk_commit_all(void)
2735 {
2736     BlockBackend *blk = NULL;
2737     GLOBAL_STATE_CODE();
2738
2739     while ((blk = blk_all_next(blk)) != NULL) {
2740         AioContext *aio_context = blk_get_aio_context(blk);
2741         BlockDriverState *unfiltered_bs = bdrv_skip_filters(blk_bs(blk));
2742
2743         aio_context_acquire(aio_context);
2744         if (blk_is_inserted(blk) && bdrv_cow_child(unfiltered_bs)) {
2745             int ret;
2746
2747             ret = bdrv_commit(unfiltered_bs);
2748             if (ret < 0) {
2749                 aio_context_release(aio_context);
2750                 return ret;
2751             }
2752         }
2753         aio_context_release(aio_context);
2754     }
2755     return 0;
2756 }
2757
2758
2759 /* throttling disk I/O limits */
2760 void blk_set_io_limits(BlockBackend *blk, ThrottleConfig *cfg)
2761 {
2762     GLOBAL_STATE_CODE();
2763     throttle_group_config(&blk->public.throttle_group_member, cfg);
2764 }
2765
2766 void blk_io_limits_disable(BlockBackend *blk)
2767 {
2768     BlockDriverState *bs = blk_bs(blk);
2769     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
2770     assert(tgm->throttle_state);
2771     GLOBAL_STATE_CODE();
2772     if (bs) {
2773         bdrv_ref(bs);
2774         bdrv_drained_begin(bs);
2775     }
2776     throttle_group_unregister_tgm(tgm);
2777     if (bs) {
2778         bdrv_drained_end(bs);
2779         bdrv_unref(bs);
2780     }
2781 }
2782
2783 /* should be called before blk_set_io_limits if a limit is set */
2784 void blk_io_limits_enable(BlockBackend *blk, const char *group)
2785 {
2786     assert(!blk->public.throttle_group_member.throttle_state);
2787     GLOBAL_STATE_CODE();
2788     throttle_group_register_tgm(&blk->public.throttle_group_member,
2789                                 group, blk_get_aio_context(blk));
2790 }
2791
2792 void blk_io_limits_update_group(BlockBackend *blk, const char *group)
2793 {
2794     GLOBAL_STATE_CODE();
2795     /* this BB is not part of any group */
2796     if (!blk->public.throttle_group_member.throttle_state) {
2797         return;
2798     }
2799
2800     /* this BB is a part of the same group than the one we want */
2801     if (!g_strcmp0(throttle_group_get_name(&blk->public.throttle_group_member),
2802                 group)) {
2803         return;
2804     }
2805
2806     /* need to change the group this bs belong to */
2807     blk_io_limits_disable(blk);
2808     blk_io_limits_enable(blk, group);
2809 }
2810
2811 static void blk_root_drained_begin(BdrvChild *child)
2812 {
2813     BlockBackend *blk = child->opaque;
2814     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
2815
2816     if (qatomic_fetch_inc(&blk->quiesce_counter) == 0) {
2817         if (blk->dev_ops && blk->dev_ops->drained_begin) {
2818             blk->dev_ops->drained_begin(blk->dev_opaque);
2819         }
2820     }
2821
2822     /* Note that blk->root may not be accessible here yet if we are just
2823      * attaching to a BlockDriverState that is drained. Use child instead. */
2824
2825     if (qatomic_fetch_inc(&tgm->io_limits_disabled) == 0) {
2826         throttle_group_restart_tgm(tgm);
2827     }
2828 }
2829
2830 static bool blk_root_drained_poll(BdrvChild *child)
2831 {
2832     BlockBackend *blk = child->opaque;
2833     bool busy = false;
2834     assert(qatomic_read(&blk->quiesce_counter));
2835
2836     if (blk->dev_ops && blk->dev_ops->drained_poll) {
2837         busy = blk->dev_ops->drained_poll(blk->dev_opaque);
2838     }
2839     return busy || !!blk->in_flight;
2840 }
2841
2842 static void blk_root_drained_end(BdrvChild *child)
2843 {
2844     BlockBackend *blk = child->opaque;
2845     assert(qatomic_read(&blk->quiesce_counter));
2846
2847     assert(blk->public.throttle_group_member.io_limits_disabled);
2848     qatomic_dec(&blk->public.throttle_group_member.io_limits_disabled);
2849
2850     if (qatomic_fetch_dec(&blk->quiesce_counter) == 1) {
2851         if (blk->dev_ops && blk->dev_ops->drained_end) {
2852             blk->dev_ops->drained_end(blk->dev_opaque);
2853         }
2854         qemu_mutex_lock(&blk->queued_requests_lock);
2855         while (qemu_co_enter_next(&blk->queued_requests,
2856                                   &blk->queued_requests_lock)) {
2857             /* Resume all queued requests */
2858         }
2859         qemu_mutex_unlock(&blk->queued_requests_lock);
2860     }
2861 }
2862
2863 bool blk_register_buf(BlockBackend *blk, void *host, size_t size, Error **errp)
2864 {
2865     BlockDriverState *bs = blk_bs(blk);
2866
2867     GLOBAL_STATE_CODE();
2868
2869     if (bs) {
2870         return bdrv_register_buf(bs, host, size, errp);
2871     }
2872     return true;
2873 }
2874
2875 void blk_unregister_buf(BlockBackend *blk, void *host, size_t size)
2876 {
2877     BlockDriverState *bs = blk_bs(blk);
2878
2879     GLOBAL_STATE_CODE();
2880
2881     if (bs) {
2882         bdrv_unregister_buf(bs, host, size);
2883     }
2884 }
2885
2886 int coroutine_fn blk_co_copy_range(BlockBackend *blk_in, int64_t off_in,
2887                                    BlockBackend *blk_out, int64_t off_out,
2888                                    int64_t bytes, BdrvRequestFlags read_flags,
2889                                    BdrvRequestFlags write_flags)
2890 {
2891     int r;
2892     IO_CODE();
2893     GRAPH_RDLOCK_GUARD();
2894
2895     r = blk_check_byte_request(blk_in, off_in, bytes);
2896     if (r) {
2897         return r;
2898     }
2899     r = blk_check_byte_request(blk_out, off_out, bytes);
2900     if (r) {
2901         return r;
2902     }
2903
2904     return bdrv_co_copy_range(blk_in->root, off_in,
2905                               blk_out->root, off_out,
2906                               bytes, read_flags, write_flags);
2907 }
2908
2909 const BdrvChild *blk_root(BlockBackend *blk)
2910 {
2911     GLOBAL_STATE_CODE();
2912     return blk->root;
2913 }
2914
2915 int blk_make_empty(BlockBackend *blk, Error **errp)
2916 {
2917     GLOBAL_STATE_CODE();
2918     if (!blk_is_available(blk)) {
2919         error_setg(errp, "No medium inserted");
2920         return -ENOMEDIUM;
2921     }
2922
2923     return bdrv_make_empty(blk->root, errp);
2924 }