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