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->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 int coroutine_fn
1284 blk_co_do_preadv(BlockBackend *blk, int64_t offset, int64_t bytes,
1285                  QEMUIOVector *qiov, BdrvRequestFlags flags)
1286 {
1287     int ret;
1288     BlockDriverState *bs;
1289     IO_CODE();
1290
1291     blk_wait_while_drained(blk);
1292
1293     /* Call blk_bs() only after waiting, the graph may have changed */
1294     bs = blk_bs(blk);
1295     trace_blk_co_preadv(blk, bs, offset, bytes, flags);
1296
1297     ret = blk_check_byte_request(blk, offset, bytes);
1298     if (ret < 0) {
1299         return ret;
1300     }
1301
1302     bdrv_inc_in_flight(bs);
1303
1304     /* throttling disk I/O */
1305     if (blk->public.throttle_group_member.throttle_state) {
1306         throttle_group_co_io_limits_intercept(&blk->public.throttle_group_member,
1307                 bytes, false);
1308     }
1309
1310     ret = bdrv_co_preadv(blk->root, offset, bytes, qiov, flags);
1311     bdrv_dec_in_flight(bs);
1312     return ret;
1313 }
1314
1315 int coroutine_fn blk_co_preadv(BlockBackend *blk, int64_t offset,
1316                                int64_t bytes, QEMUIOVector *qiov,
1317                                BdrvRequestFlags flags)
1318 {
1319     int ret;
1320     IO_OR_GS_CODE();
1321
1322     blk_inc_in_flight(blk);
1323     ret = blk_co_do_preadv(blk, offset, bytes, qiov, flags);
1324     blk_dec_in_flight(blk);
1325
1326     return ret;
1327 }
1328
1329 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1330 int coroutine_fn
1331 blk_co_do_pwritev_part(BlockBackend *blk, int64_t offset, int64_t bytes,
1332                        QEMUIOVector *qiov, size_t qiov_offset,
1333                        BdrvRequestFlags flags)
1334 {
1335     int ret;
1336     BlockDriverState *bs;
1337     IO_CODE();
1338
1339     blk_wait_while_drained(blk);
1340
1341     /* Call blk_bs() only after waiting, the graph may have changed */
1342     bs = blk_bs(blk);
1343     trace_blk_co_pwritev(blk, bs, offset, bytes, flags);
1344
1345     ret = blk_check_byte_request(blk, offset, bytes);
1346     if (ret < 0) {
1347         return ret;
1348     }
1349
1350     bdrv_inc_in_flight(bs);
1351     /* throttling disk I/O */
1352     if (blk->public.throttle_group_member.throttle_state) {
1353         throttle_group_co_io_limits_intercept(&blk->public.throttle_group_member,
1354                 bytes, true);
1355     }
1356
1357     if (!blk->enable_write_cache) {
1358         flags |= BDRV_REQ_FUA;
1359     }
1360
1361     ret = bdrv_co_pwritev_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_pwritev_part(BlockBackend *blk, int64_t offset,
1368                                      int64_t bytes,
1369                                      QEMUIOVector *qiov, size_t qiov_offset,
1370                                      BdrvRequestFlags flags)
1371 {
1372     int ret;
1373     IO_OR_GS_CODE();
1374
1375     blk_inc_in_flight(blk);
1376     ret = blk_co_do_pwritev_part(blk, offset, bytes, qiov, qiov_offset, flags);
1377     blk_dec_in_flight(blk);
1378
1379     return ret;
1380 }
1381
1382 int coroutine_fn blk_co_pwritev(BlockBackend *blk, int64_t offset,
1383                                 int64_t bytes, QEMUIOVector *qiov,
1384                                 BdrvRequestFlags flags)
1385 {
1386     IO_OR_GS_CODE();
1387     return blk_co_pwritev_part(blk, offset, bytes, qiov, 0, flags);
1388 }
1389
1390 static int coroutine_fn blk_pwritev_part(BlockBackend *blk, int64_t offset,
1391                                          int64_t bytes,
1392                                          QEMUIOVector *qiov, size_t qiov_offset,
1393                                          BdrvRequestFlags flags)
1394 {
1395     int ret;
1396
1397     blk_inc_in_flight(blk);
1398     ret = blk_do_pwritev_part(blk, offset, bytes, qiov, qiov_offset, flags);
1399     blk_dec_in_flight(blk);
1400
1401     return ret;
1402 }
1403
1404 typedef struct BlkRwCo {
1405     BlockBackend *blk;
1406     int64_t offset;
1407     void *iobuf;
1408     int ret;
1409     BdrvRequestFlags flags;
1410 } BlkRwCo;
1411
1412 int blk_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1413                       int64_t bytes, BdrvRequestFlags flags)
1414 {
1415     IO_OR_GS_CODE();
1416     return blk_pwritev_part(blk, offset, bytes, NULL, 0,
1417                             flags | BDRV_REQ_ZERO_WRITE);
1418 }
1419
1420 int blk_make_zero(BlockBackend *blk, BdrvRequestFlags flags)
1421 {
1422     GLOBAL_STATE_CODE();
1423     return bdrv_make_zero(blk->root, flags);
1424 }
1425
1426 void blk_inc_in_flight(BlockBackend *blk)
1427 {
1428     IO_CODE();
1429     qatomic_inc(&blk->in_flight);
1430 }
1431
1432 void blk_dec_in_flight(BlockBackend *blk)
1433 {
1434     IO_CODE();
1435     qatomic_dec(&blk->in_flight);
1436     aio_wait_kick();
1437 }
1438
1439 static void error_callback_bh(void *opaque)
1440 {
1441     struct BlockBackendAIOCB *acb = opaque;
1442
1443     blk_dec_in_flight(acb->blk);
1444     acb->common.cb(acb->common.opaque, acb->ret);
1445     qemu_aio_unref(acb);
1446 }
1447
1448 BlockAIOCB *blk_abort_aio_request(BlockBackend *blk,
1449                                   BlockCompletionFunc *cb,
1450                                   void *opaque, int ret)
1451 {
1452     struct BlockBackendAIOCB *acb;
1453     IO_CODE();
1454
1455     blk_inc_in_flight(blk);
1456     acb = blk_aio_get(&block_backend_aiocb_info, blk, cb, opaque);
1457     acb->blk = blk;
1458     acb->ret = ret;
1459
1460     replay_bh_schedule_oneshot_event(blk_get_aio_context(blk),
1461                                      error_callback_bh, acb);
1462     return &acb->common;
1463 }
1464
1465 typedef struct BlkAioEmAIOCB {
1466     BlockAIOCB common;
1467     BlkRwCo rwco;
1468     int64_t bytes;
1469     bool has_returned;
1470 } BlkAioEmAIOCB;
1471
1472 static AioContext *blk_aio_em_aiocb_get_aio_context(BlockAIOCB *acb_)
1473 {
1474     BlkAioEmAIOCB *acb = container_of(acb_, BlkAioEmAIOCB, common);
1475
1476     return blk_get_aio_context(acb->rwco.blk);
1477 }
1478
1479 static const AIOCBInfo blk_aio_em_aiocb_info = {
1480     .aiocb_size         = sizeof(BlkAioEmAIOCB),
1481     .get_aio_context    = blk_aio_em_aiocb_get_aio_context,
1482 };
1483
1484 static void blk_aio_complete(BlkAioEmAIOCB *acb)
1485 {
1486     if (acb->has_returned) {
1487         acb->common.cb(acb->common.opaque, acb->rwco.ret);
1488         blk_dec_in_flight(acb->rwco.blk);
1489         qemu_aio_unref(acb);
1490     }
1491 }
1492
1493 static void blk_aio_complete_bh(void *opaque)
1494 {
1495     BlkAioEmAIOCB *acb = opaque;
1496     assert(acb->has_returned);
1497     blk_aio_complete(acb);
1498 }
1499
1500 static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset,
1501                                 int64_t bytes,
1502                                 void *iobuf, CoroutineEntry co_entry,
1503                                 BdrvRequestFlags flags,
1504                                 BlockCompletionFunc *cb, void *opaque)
1505 {
1506     BlkAioEmAIOCB *acb;
1507     Coroutine *co;
1508
1509     blk_inc_in_flight(blk);
1510     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1511     acb->rwco = (BlkRwCo) {
1512         .blk    = blk,
1513         .offset = offset,
1514         .iobuf  = iobuf,
1515         .flags  = flags,
1516         .ret    = NOT_DONE,
1517     };
1518     acb->bytes = bytes;
1519     acb->has_returned = false;
1520
1521     co = qemu_coroutine_create(co_entry, acb);
1522     bdrv_coroutine_enter(blk_bs(blk), co);
1523
1524     acb->has_returned = true;
1525     if (acb->rwco.ret != NOT_DONE) {
1526         replay_bh_schedule_oneshot_event(blk_get_aio_context(blk),
1527                                          blk_aio_complete_bh, acb);
1528     }
1529
1530     return &acb->common;
1531 }
1532
1533 static void blk_aio_read_entry(void *opaque)
1534 {
1535     BlkAioEmAIOCB *acb = opaque;
1536     BlkRwCo *rwco = &acb->rwco;
1537     QEMUIOVector *qiov = rwco->iobuf;
1538
1539     assert(qiov->size == acb->bytes);
1540     rwco->ret = blk_co_do_preadv(rwco->blk, rwco->offset, acb->bytes,
1541                                  qiov, rwco->flags);
1542     blk_aio_complete(acb);
1543 }
1544
1545 static void blk_aio_write_entry(void *opaque)
1546 {
1547     BlkAioEmAIOCB *acb = opaque;
1548     BlkRwCo *rwco = &acb->rwco;
1549     QEMUIOVector *qiov = rwco->iobuf;
1550
1551     assert(!qiov || qiov->size == acb->bytes);
1552     rwco->ret = blk_co_do_pwritev_part(rwco->blk, rwco->offset, acb->bytes,
1553                                        qiov, 0, rwco->flags);
1554     blk_aio_complete(acb);
1555 }
1556
1557 BlockAIOCB *blk_aio_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1558                                   int64_t bytes, BdrvRequestFlags flags,
1559                                   BlockCompletionFunc *cb, void *opaque)
1560 {
1561     IO_CODE();
1562     return blk_aio_prwv(blk, offset, bytes, NULL, blk_aio_write_entry,
1563                         flags | BDRV_REQ_ZERO_WRITE, cb, opaque);
1564 }
1565
1566 int blk_pread(BlockBackend *blk, int64_t offset, void *buf, int bytes)
1567 {
1568     int ret;
1569     QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
1570     IO_OR_GS_CODE();
1571
1572     blk_inc_in_flight(blk);
1573     ret = blk_do_preadv(blk, offset, bytes, &qiov, 0);
1574     blk_dec_in_flight(blk);
1575
1576     return ret < 0 ? ret : bytes;
1577 }
1578
1579 int blk_pwrite(BlockBackend *blk, int64_t offset, const void *buf, int bytes,
1580                BdrvRequestFlags flags)
1581 {
1582     int ret;
1583     QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
1584     IO_OR_GS_CODE();
1585
1586     ret = blk_pwritev_part(blk, offset, bytes, &qiov, 0, flags);
1587
1588     return ret < 0 ? ret : bytes;
1589 }
1590
1591 int64_t blk_getlength(BlockBackend *blk)
1592 {
1593     IO_CODE();
1594     if (!blk_is_available(blk)) {
1595         return -ENOMEDIUM;
1596     }
1597
1598     return bdrv_getlength(blk_bs(blk));
1599 }
1600
1601 void blk_get_geometry(BlockBackend *blk, uint64_t *nb_sectors_ptr)
1602 {
1603     IO_CODE();
1604     if (!blk_bs(blk)) {
1605         *nb_sectors_ptr = 0;
1606     } else {
1607         bdrv_get_geometry(blk_bs(blk), nb_sectors_ptr);
1608     }
1609 }
1610
1611 int64_t blk_nb_sectors(BlockBackend *blk)
1612 {
1613     IO_CODE();
1614     if (!blk_is_available(blk)) {
1615         return -ENOMEDIUM;
1616     }
1617
1618     return bdrv_nb_sectors(blk_bs(blk));
1619 }
1620
1621 BlockAIOCB *blk_aio_preadv(BlockBackend *blk, int64_t offset,
1622                            QEMUIOVector *qiov, BdrvRequestFlags flags,
1623                            BlockCompletionFunc *cb, void *opaque)
1624 {
1625     IO_CODE();
1626     assert((uint64_t)qiov->size <= INT64_MAX);
1627     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1628                         blk_aio_read_entry, flags, cb, opaque);
1629 }
1630
1631 BlockAIOCB *blk_aio_pwritev(BlockBackend *blk, int64_t offset,
1632                             QEMUIOVector *qiov, BdrvRequestFlags flags,
1633                             BlockCompletionFunc *cb, void *opaque)
1634 {
1635     IO_CODE();
1636     assert((uint64_t)qiov->size <= INT64_MAX);
1637     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1638                         blk_aio_write_entry, flags, cb, opaque);
1639 }
1640
1641 void blk_aio_cancel(BlockAIOCB *acb)
1642 {
1643     GLOBAL_STATE_CODE();
1644     bdrv_aio_cancel(acb);
1645 }
1646
1647 void blk_aio_cancel_async(BlockAIOCB *acb)
1648 {
1649     IO_CODE();
1650     bdrv_aio_cancel_async(acb);
1651 }
1652
1653 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1654 int coroutine_fn
1655 blk_co_do_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1656 {
1657     IO_CODE();
1658
1659     blk_wait_while_drained(blk);
1660
1661     if (!blk_is_available(blk)) {
1662         return -ENOMEDIUM;
1663     }
1664
1665     return bdrv_co_ioctl(blk_bs(blk), req, buf);
1666 }
1667
1668 int blk_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1669 {
1670     int ret;
1671     IO_OR_GS_CODE();
1672
1673     blk_inc_in_flight(blk);
1674     ret = blk_do_ioctl(blk, req, buf);
1675     blk_dec_in_flight(blk);
1676
1677     return ret;
1678 }
1679
1680 static void blk_aio_ioctl_entry(void *opaque)
1681 {
1682     BlkAioEmAIOCB *acb = opaque;
1683     BlkRwCo *rwco = &acb->rwco;
1684
1685     rwco->ret = blk_co_do_ioctl(rwco->blk, rwco->offset, rwco->iobuf);
1686
1687     blk_aio_complete(acb);
1688 }
1689
1690 BlockAIOCB *blk_aio_ioctl(BlockBackend *blk, unsigned long int req, void *buf,
1691                           BlockCompletionFunc *cb, void *opaque)
1692 {
1693     IO_CODE();
1694     return blk_aio_prwv(blk, req, 0, buf, blk_aio_ioctl_entry, 0, cb, opaque);
1695 }
1696
1697 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1698 int coroutine_fn
1699 blk_co_do_pdiscard(BlockBackend *blk, int64_t offset, int64_t bytes)
1700 {
1701     int ret;
1702     IO_CODE();
1703
1704     blk_wait_while_drained(blk);
1705
1706     ret = blk_check_byte_request(blk, offset, bytes);
1707     if (ret < 0) {
1708         return ret;
1709     }
1710
1711     return bdrv_co_pdiscard(blk->root, offset, bytes);
1712 }
1713
1714 static void blk_aio_pdiscard_entry(void *opaque)
1715 {
1716     BlkAioEmAIOCB *acb = opaque;
1717     BlkRwCo *rwco = &acb->rwco;
1718
1719     rwco->ret = blk_co_do_pdiscard(rwco->blk, rwco->offset, acb->bytes);
1720     blk_aio_complete(acb);
1721 }
1722
1723 BlockAIOCB *blk_aio_pdiscard(BlockBackend *blk,
1724                              int64_t offset, int64_t bytes,
1725                              BlockCompletionFunc *cb, void *opaque)
1726 {
1727     IO_CODE();
1728     return blk_aio_prwv(blk, offset, bytes, NULL, blk_aio_pdiscard_entry, 0,
1729                         cb, opaque);
1730 }
1731
1732 int coroutine_fn blk_co_pdiscard(BlockBackend *blk, int64_t offset,
1733                                  int64_t bytes)
1734 {
1735     int ret;
1736     IO_OR_GS_CODE();
1737
1738     blk_inc_in_flight(blk);
1739     ret = blk_co_do_pdiscard(blk, offset, bytes);
1740     blk_dec_in_flight(blk);
1741
1742     return ret;
1743 }
1744
1745 int blk_pdiscard(BlockBackend *blk, int64_t offset, int64_t bytes)
1746 {
1747     int ret;
1748     IO_OR_GS_CODE();
1749
1750     blk_inc_in_flight(blk);
1751     ret = blk_do_pdiscard(blk, offset, bytes);
1752     blk_dec_in_flight(blk);
1753
1754     return ret;
1755 }
1756
1757 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1758 int coroutine_fn blk_co_do_flush(BlockBackend *blk)
1759 {
1760     blk_wait_while_drained(blk);
1761     IO_CODE();
1762
1763     if (!blk_is_available(blk)) {
1764         return -ENOMEDIUM;
1765     }
1766
1767     return bdrv_co_flush(blk_bs(blk));
1768 }
1769
1770 static void blk_aio_flush_entry(void *opaque)
1771 {
1772     BlkAioEmAIOCB *acb = opaque;
1773     BlkRwCo *rwco = &acb->rwco;
1774
1775     rwco->ret = blk_co_do_flush(rwco->blk);
1776     blk_aio_complete(acb);
1777 }
1778
1779 BlockAIOCB *blk_aio_flush(BlockBackend *blk,
1780                           BlockCompletionFunc *cb, void *opaque)
1781 {
1782     IO_CODE();
1783     return blk_aio_prwv(blk, 0, 0, NULL, blk_aio_flush_entry, 0, cb, opaque);
1784 }
1785
1786 int coroutine_fn blk_co_flush(BlockBackend *blk)
1787 {
1788     int ret;
1789     IO_OR_GS_CODE();
1790
1791     blk_inc_in_flight(blk);
1792     ret = blk_co_do_flush(blk);
1793     blk_dec_in_flight(blk);
1794
1795     return ret;
1796 }
1797
1798 int blk_flush(BlockBackend *blk)
1799 {
1800     int ret;
1801
1802     blk_inc_in_flight(blk);
1803     ret = blk_do_flush(blk);
1804     blk_dec_in_flight(blk);
1805
1806     return ret;
1807 }
1808
1809 void blk_drain(BlockBackend *blk)
1810 {
1811     BlockDriverState *bs = blk_bs(blk);
1812     GLOBAL_STATE_CODE();
1813
1814     if (bs) {
1815         bdrv_ref(bs);
1816         bdrv_drained_begin(bs);
1817     }
1818
1819     /* We may have -ENOMEDIUM completions in flight */
1820     AIO_WAIT_WHILE(blk_get_aio_context(blk),
1821                    qatomic_mb_read(&blk->in_flight) > 0);
1822
1823     if (bs) {
1824         bdrv_drained_end(bs);
1825         bdrv_unref(bs);
1826     }
1827 }
1828
1829 void blk_drain_all(void)
1830 {
1831     BlockBackend *blk = NULL;
1832
1833     GLOBAL_STATE_CODE();
1834
1835     bdrv_drain_all_begin();
1836
1837     while ((blk = blk_all_next(blk)) != NULL) {
1838         AioContext *ctx = blk_get_aio_context(blk);
1839
1840         aio_context_acquire(ctx);
1841
1842         /* We may have -ENOMEDIUM completions in flight */
1843         AIO_WAIT_WHILE(ctx, qatomic_mb_read(&blk->in_flight) > 0);
1844
1845         aio_context_release(ctx);
1846     }
1847
1848     bdrv_drain_all_end();
1849 }
1850
1851 void blk_set_on_error(BlockBackend *blk, BlockdevOnError on_read_error,
1852                       BlockdevOnError on_write_error)
1853 {
1854     GLOBAL_STATE_CODE();
1855     blk->on_read_error = on_read_error;
1856     blk->on_write_error = on_write_error;
1857 }
1858
1859 BlockdevOnError blk_get_on_error(BlockBackend *blk, bool is_read)
1860 {
1861     IO_CODE();
1862     return is_read ? blk->on_read_error : blk->on_write_error;
1863 }
1864
1865 BlockErrorAction blk_get_error_action(BlockBackend *blk, bool is_read,
1866                                       int error)
1867 {
1868     BlockdevOnError on_err = blk_get_on_error(blk, is_read);
1869     IO_CODE();
1870
1871     switch (on_err) {
1872     case BLOCKDEV_ON_ERROR_ENOSPC:
1873         return (error == ENOSPC) ?
1874                BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
1875     case BLOCKDEV_ON_ERROR_STOP:
1876         return BLOCK_ERROR_ACTION_STOP;
1877     case BLOCKDEV_ON_ERROR_REPORT:
1878         return BLOCK_ERROR_ACTION_REPORT;
1879     case BLOCKDEV_ON_ERROR_IGNORE:
1880         return BLOCK_ERROR_ACTION_IGNORE;
1881     case BLOCKDEV_ON_ERROR_AUTO:
1882     default:
1883         abort();
1884     }
1885 }
1886
1887 static void send_qmp_error_event(BlockBackend *blk,
1888                                  BlockErrorAction action,
1889                                  bool is_read, int error)
1890 {
1891     IoOperationType optype;
1892     BlockDriverState *bs = blk_bs(blk);
1893
1894     optype = is_read ? IO_OPERATION_TYPE_READ : IO_OPERATION_TYPE_WRITE;
1895     qapi_event_send_block_io_error(blk_name(blk), !!bs,
1896                                    bs ? bdrv_get_node_name(bs) : NULL, optype,
1897                                    action, blk_iostatus_is_enabled(blk),
1898                                    error == ENOSPC, strerror(error));
1899 }
1900
1901 /* This is done by device models because, while the block layer knows
1902  * about the error, it does not know whether an operation comes from
1903  * the device or the block layer (from a job, for example).
1904  */
1905 void blk_error_action(BlockBackend *blk, BlockErrorAction action,
1906                       bool is_read, int error)
1907 {
1908     assert(error >= 0);
1909     IO_CODE();
1910
1911     if (action == BLOCK_ERROR_ACTION_STOP) {
1912         /* First set the iostatus, so that "info block" returns an iostatus
1913          * that matches the events raised so far (an additional error iostatus
1914          * is fine, but not a lost one).
1915          */
1916         blk_iostatus_set_err(blk, error);
1917
1918         /* Then raise the request to stop the VM and the event.
1919          * qemu_system_vmstop_request_prepare has two effects.  First,
1920          * it ensures that the STOP event always comes after the
1921          * BLOCK_IO_ERROR event.  Second, it ensures that even if management
1922          * can observe the STOP event and do a "cont" before the STOP
1923          * event is issued, the VM will not stop.  In this case, vm_start()
1924          * also ensures that the STOP/RESUME pair of events is emitted.
1925          */
1926         qemu_system_vmstop_request_prepare();
1927         send_qmp_error_event(blk, action, is_read, error);
1928         qemu_system_vmstop_request(RUN_STATE_IO_ERROR);
1929     } else {
1930         send_qmp_error_event(blk, action, is_read, error);
1931     }
1932 }
1933
1934 /*
1935  * Returns true if the BlockBackend can support taking write permissions
1936  * (because its root node is not read-only).
1937  */
1938 bool blk_supports_write_perm(BlockBackend *blk)
1939 {
1940     BlockDriverState *bs = blk_bs(blk);
1941     GLOBAL_STATE_CODE();
1942
1943     if (bs) {
1944         return !bdrv_is_read_only(bs);
1945     } else {
1946         return blk->root_state.open_flags & BDRV_O_RDWR;
1947     }
1948 }
1949
1950 /*
1951  * Returns true if the BlockBackend can be written to in its current
1952  * configuration (i.e. if write permission have been requested)
1953  */
1954 bool blk_is_writable(BlockBackend *blk)
1955 {
1956     IO_CODE();
1957     return blk->perm & BLK_PERM_WRITE;
1958 }
1959
1960 bool blk_is_sg(BlockBackend *blk)
1961 {
1962     BlockDriverState *bs = blk_bs(blk);
1963     GLOBAL_STATE_CODE();
1964
1965     if (!bs) {
1966         return false;
1967     }
1968
1969     return bdrv_is_sg(bs);
1970 }
1971
1972 bool blk_enable_write_cache(BlockBackend *blk)
1973 {
1974     IO_CODE();
1975     return blk->enable_write_cache;
1976 }
1977
1978 void blk_set_enable_write_cache(BlockBackend *blk, bool wce)
1979 {
1980     GLOBAL_STATE_CODE();
1981     blk->enable_write_cache = wce;
1982 }
1983
1984 void blk_activate(BlockBackend *blk, Error **errp)
1985 {
1986     BlockDriverState *bs = blk_bs(blk);
1987     GLOBAL_STATE_CODE();
1988
1989     if (!bs) {
1990         error_setg(errp, "Device '%s' has no medium", blk->name);
1991         return;
1992     }
1993
1994     bdrv_activate(bs, errp);
1995 }
1996
1997 bool blk_is_inserted(BlockBackend *blk)
1998 {
1999     BlockDriverState *bs = blk_bs(blk);
2000     IO_CODE();
2001
2002     return bs && bdrv_is_inserted(bs);
2003 }
2004
2005 bool blk_is_available(BlockBackend *blk)
2006 {
2007     IO_CODE();
2008     return blk_is_inserted(blk) && !blk_dev_is_tray_open(blk);
2009 }
2010
2011 void blk_lock_medium(BlockBackend *blk, bool locked)
2012 {
2013     BlockDriverState *bs = blk_bs(blk);
2014     IO_CODE();
2015
2016     if (bs) {
2017         bdrv_lock_medium(bs, locked);
2018     }
2019 }
2020
2021 void blk_eject(BlockBackend *blk, bool eject_flag)
2022 {
2023     BlockDriverState *bs = blk_bs(blk);
2024     char *id;
2025     IO_CODE();
2026
2027     if (bs) {
2028         bdrv_eject(bs, eject_flag);
2029     }
2030
2031     /* Whether or not we ejected on the backend,
2032      * the frontend experienced a tray event. */
2033     id = blk_get_attached_dev_id(blk);
2034     qapi_event_send_device_tray_moved(blk_name(blk), id,
2035                                       eject_flag);
2036     g_free(id);
2037 }
2038
2039 int blk_get_flags(BlockBackend *blk)
2040 {
2041     BlockDriverState *bs = blk_bs(blk);
2042     GLOBAL_STATE_CODE();
2043
2044     if (bs) {
2045         return bdrv_get_flags(bs);
2046     } else {
2047         return blk->root_state.open_flags;
2048     }
2049 }
2050
2051 /* Returns the minimum request alignment, in bytes; guaranteed nonzero */
2052 uint32_t blk_get_request_alignment(BlockBackend *blk)
2053 {
2054     BlockDriverState *bs = blk_bs(blk);
2055     IO_CODE();
2056     return bs ? bs->bl.request_alignment : BDRV_SECTOR_SIZE;
2057 }
2058
2059 /* Returns the maximum hardware transfer length, in bytes; guaranteed nonzero */
2060 uint64_t blk_get_max_hw_transfer(BlockBackend *blk)
2061 {
2062     BlockDriverState *bs = blk_bs(blk);
2063     uint64_t max = INT_MAX;
2064     IO_CODE();
2065
2066     if (bs) {
2067         max = MIN_NON_ZERO(max, bs->bl.max_hw_transfer);
2068         max = MIN_NON_ZERO(max, bs->bl.max_transfer);
2069     }
2070     return ROUND_DOWN(max, blk_get_request_alignment(blk));
2071 }
2072
2073 /* Returns the maximum transfer length, in bytes; guaranteed nonzero */
2074 uint32_t blk_get_max_transfer(BlockBackend *blk)
2075 {
2076     BlockDriverState *bs = blk_bs(blk);
2077     uint32_t max = INT_MAX;
2078     IO_CODE();
2079
2080     if (bs) {
2081         max = MIN_NON_ZERO(max, bs->bl.max_transfer);
2082     }
2083     return ROUND_DOWN(max, blk_get_request_alignment(blk));
2084 }
2085
2086 int blk_get_max_hw_iov(BlockBackend *blk)
2087 {
2088     IO_CODE();
2089     return MIN_NON_ZERO(blk->root->bs->bl.max_hw_iov,
2090                         blk->root->bs->bl.max_iov);
2091 }
2092
2093 int blk_get_max_iov(BlockBackend *blk)
2094 {
2095     IO_CODE();
2096     return blk->root->bs->bl.max_iov;
2097 }
2098
2099 void *blk_try_blockalign(BlockBackend *blk, size_t size)
2100 {
2101     IO_CODE();
2102     return qemu_try_blockalign(blk ? blk_bs(blk) : NULL, size);
2103 }
2104
2105 void *blk_blockalign(BlockBackend *blk, size_t size)
2106 {
2107     IO_CODE();
2108     return qemu_blockalign(blk ? blk_bs(blk) : NULL, size);
2109 }
2110
2111 bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp)
2112 {
2113     BlockDriverState *bs = blk_bs(blk);
2114     GLOBAL_STATE_CODE();
2115
2116     if (!bs) {
2117         return false;
2118     }
2119
2120     return bdrv_op_is_blocked(bs, op, errp);
2121 }
2122
2123 void blk_op_unblock(BlockBackend *blk, BlockOpType op, Error *reason)
2124 {
2125     BlockDriverState *bs = blk_bs(blk);
2126     GLOBAL_STATE_CODE();
2127
2128     if (bs) {
2129         bdrv_op_unblock(bs, op, reason);
2130     }
2131 }
2132
2133 void blk_op_block_all(BlockBackend *blk, Error *reason)
2134 {
2135     BlockDriverState *bs = blk_bs(blk);
2136     GLOBAL_STATE_CODE();
2137
2138     if (bs) {
2139         bdrv_op_block_all(bs, reason);
2140     }
2141 }
2142
2143 void blk_op_unblock_all(BlockBackend *blk, Error *reason)
2144 {
2145     BlockDriverState *bs = blk_bs(blk);
2146     GLOBAL_STATE_CODE();
2147
2148     if (bs) {
2149         bdrv_op_unblock_all(bs, reason);
2150     }
2151 }
2152
2153 AioContext *blk_get_aio_context(BlockBackend *blk)
2154 {
2155     BlockDriverState *bs = blk_bs(blk);
2156     IO_CODE();
2157
2158     if (bs) {
2159         AioContext *ctx = bdrv_get_aio_context(blk_bs(blk));
2160         assert(ctx == blk->ctx);
2161     }
2162
2163     return blk->ctx;
2164 }
2165
2166 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb)
2167 {
2168     BlockBackendAIOCB *blk_acb = DO_UPCAST(BlockBackendAIOCB, common, acb);
2169     return blk_get_aio_context(blk_acb->blk);
2170 }
2171
2172 static int blk_do_set_aio_context(BlockBackend *blk, AioContext *new_context,
2173                                   bool update_root_node, Error **errp)
2174 {
2175     BlockDriverState *bs = blk_bs(blk);
2176     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
2177     int ret;
2178
2179     if (bs) {
2180         bdrv_ref(bs);
2181
2182         if (update_root_node) {
2183             ret = bdrv_child_try_set_aio_context(bs, new_context, blk->root,
2184                                                  errp);
2185             if (ret < 0) {
2186                 bdrv_unref(bs);
2187                 return ret;
2188             }
2189         }
2190         if (tgm->throttle_state) {
2191             bdrv_drained_begin(bs);
2192             throttle_group_detach_aio_context(tgm);
2193             throttle_group_attach_aio_context(tgm, new_context);
2194             bdrv_drained_end(bs);
2195         }
2196
2197         bdrv_unref(bs);
2198     }
2199
2200     blk->ctx = new_context;
2201     return 0;
2202 }
2203
2204 int blk_set_aio_context(BlockBackend *blk, AioContext *new_context,
2205                         Error **errp)
2206 {
2207     GLOBAL_STATE_CODE();
2208     return blk_do_set_aio_context(blk, new_context, true, errp);
2209 }
2210
2211 static bool blk_root_can_set_aio_ctx(BdrvChild *child, AioContext *ctx,
2212                                      GSList **ignore, Error **errp)
2213 {
2214     BlockBackend *blk = child->opaque;
2215
2216     if (blk->allow_aio_context_change) {
2217         return true;
2218     }
2219
2220     /* Only manually created BlockBackends that are not attached to anything
2221      * can change their AioContext without updating their user. */
2222     if (!blk->name || blk->dev) {
2223         /* TODO Add BB name/QOM path */
2224         error_setg(errp, "Cannot change iothread of active block backend");
2225         return false;
2226     }
2227
2228     return true;
2229 }
2230
2231 static void blk_root_set_aio_ctx(BdrvChild *child, AioContext *ctx,
2232                                  GSList **ignore)
2233 {
2234     BlockBackend *blk = child->opaque;
2235     blk_do_set_aio_context(blk, ctx, false, &error_abort);
2236 }
2237
2238 void blk_add_aio_context_notifier(BlockBackend *blk,
2239         void (*attached_aio_context)(AioContext *new_context, void *opaque),
2240         void (*detach_aio_context)(void *opaque), void *opaque)
2241 {
2242     BlockBackendAioNotifier *notifier;
2243     BlockDriverState *bs = blk_bs(blk);
2244     GLOBAL_STATE_CODE();
2245
2246     notifier = g_new(BlockBackendAioNotifier, 1);
2247     notifier->attached_aio_context = attached_aio_context;
2248     notifier->detach_aio_context = detach_aio_context;
2249     notifier->opaque = opaque;
2250     QLIST_INSERT_HEAD(&blk->aio_notifiers, notifier, list);
2251
2252     if (bs) {
2253         bdrv_add_aio_context_notifier(bs, attached_aio_context,
2254                                       detach_aio_context, opaque);
2255     }
2256 }
2257
2258 void blk_remove_aio_context_notifier(BlockBackend *blk,
2259                                      void (*attached_aio_context)(AioContext *,
2260                                                                   void *),
2261                                      void (*detach_aio_context)(void *),
2262                                      void *opaque)
2263 {
2264     BlockBackendAioNotifier *notifier;
2265     BlockDriverState *bs = blk_bs(blk);
2266
2267     GLOBAL_STATE_CODE();
2268
2269     if (bs) {
2270         bdrv_remove_aio_context_notifier(bs, attached_aio_context,
2271                                          detach_aio_context, opaque);
2272     }
2273
2274     QLIST_FOREACH(notifier, &blk->aio_notifiers, list) {
2275         if (notifier->attached_aio_context == attached_aio_context &&
2276             notifier->detach_aio_context == detach_aio_context &&
2277             notifier->opaque == opaque) {
2278             QLIST_REMOVE(notifier, list);
2279             g_free(notifier);
2280             return;
2281         }
2282     }
2283
2284     abort();
2285 }
2286
2287 void blk_add_remove_bs_notifier(BlockBackend *blk, Notifier *notify)
2288 {
2289     GLOBAL_STATE_CODE();
2290     notifier_list_add(&blk->remove_bs_notifiers, notify);
2291 }
2292
2293 void blk_add_insert_bs_notifier(BlockBackend *blk, Notifier *notify)
2294 {
2295     GLOBAL_STATE_CODE();
2296     notifier_list_add(&blk->insert_bs_notifiers, notify);
2297 }
2298
2299 void blk_io_plug(BlockBackend *blk)
2300 {
2301     BlockDriverState *bs = blk_bs(blk);
2302     IO_CODE();
2303
2304     if (bs) {
2305         bdrv_io_plug(bs);
2306     }
2307 }
2308
2309 void blk_io_unplug(BlockBackend *blk)
2310 {
2311     BlockDriverState *bs = blk_bs(blk);
2312     IO_CODE();
2313
2314     if (bs) {
2315         bdrv_io_unplug(bs);
2316     }
2317 }
2318
2319 BlockAcctStats *blk_get_stats(BlockBackend *blk)
2320 {
2321     IO_CODE();
2322     return &blk->stats;
2323 }
2324
2325 void *blk_aio_get(const AIOCBInfo *aiocb_info, BlockBackend *blk,
2326                   BlockCompletionFunc *cb, void *opaque)
2327 {
2328     IO_CODE();
2329     return qemu_aio_get(aiocb_info, blk_bs(blk), cb, opaque);
2330 }
2331
2332 int coroutine_fn blk_co_pwrite_zeroes(BlockBackend *blk, int64_t offset,
2333                                       int64_t bytes, BdrvRequestFlags flags)
2334 {
2335     IO_OR_GS_CODE();
2336     return blk_co_pwritev(blk, offset, bytes, NULL,
2337                           flags | BDRV_REQ_ZERO_WRITE);
2338 }
2339
2340 int blk_pwrite_compressed(BlockBackend *blk, int64_t offset, const void *buf,
2341                           int64_t bytes)
2342 {
2343     QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
2344     IO_OR_GS_CODE();
2345     return blk_pwritev_part(blk, offset, bytes, &qiov, 0,
2346                             BDRV_REQ_WRITE_COMPRESSED);
2347 }
2348
2349 int blk_truncate(BlockBackend *blk, int64_t offset, bool exact,
2350                  PreallocMode prealloc, BdrvRequestFlags flags, Error **errp)
2351 {
2352     IO_OR_GS_CODE();
2353     if (!blk_is_available(blk)) {
2354         error_setg(errp, "No medium inserted");
2355         return -ENOMEDIUM;
2356     }
2357
2358     return bdrv_truncate(blk->root, offset, exact, prealloc, flags, errp);
2359 }
2360
2361 int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
2362                      int64_t pos, int size)
2363 {
2364     int ret;
2365     GLOBAL_STATE_CODE();
2366
2367     if (!blk_is_available(blk)) {
2368         return -ENOMEDIUM;
2369     }
2370
2371     ret = bdrv_save_vmstate(blk_bs(blk), buf, pos, size);
2372     if (ret < 0) {
2373         return ret;
2374     }
2375
2376     if (ret == size && !blk->enable_write_cache) {
2377         ret = bdrv_flush(blk_bs(blk));
2378     }
2379
2380     return ret < 0 ? ret : size;
2381 }
2382
2383 int blk_load_vmstate(BlockBackend *blk, uint8_t *buf, int64_t pos, int size)
2384 {
2385     GLOBAL_STATE_CODE();
2386     if (!blk_is_available(blk)) {
2387         return -ENOMEDIUM;
2388     }
2389
2390     return bdrv_load_vmstate(blk_bs(blk), buf, pos, size);
2391 }
2392
2393 int blk_probe_blocksizes(BlockBackend *blk, BlockSizes *bsz)
2394 {
2395     GLOBAL_STATE_CODE();
2396     if (!blk_is_available(blk)) {
2397         return -ENOMEDIUM;
2398     }
2399
2400     return bdrv_probe_blocksizes(blk_bs(blk), bsz);
2401 }
2402
2403 int blk_probe_geometry(BlockBackend *blk, HDGeometry *geo)
2404 {
2405     GLOBAL_STATE_CODE();
2406     if (!blk_is_available(blk)) {
2407         return -ENOMEDIUM;
2408     }
2409
2410     return bdrv_probe_geometry(blk_bs(blk), geo);
2411 }
2412
2413 /*
2414  * Updates the BlockBackendRootState object with data from the currently
2415  * attached BlockDriverState.
2416  */
2417 void blk_update_root_state(BlockBackend *blk)
2418 {
2419     GLOBAL_STATE_CODE();
2420     assert(blk->root);
2421
2422     blk->root_state.open_flags    = blk->root->bs->open_flags;
2423     blk->root_state.detect_zeroes = blk->root->bs->detect_zeroes;
2424 }
2425
2426 /*
2427  * Returns the detect-zeroes setting to be used for bdrv_open() of a
2428  * BlockDriverState which is supposed to inherit the root state.
2429  */
2430 bool blk_get_detect_zeroes_from_root_state(BlockBackend *blk)
2431 {
2432     GLOBAL_STATE_CODE();
2433     return blk->root_state.detect_zeroes;
2434 }
2435
2436 /*
2437  * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
2438  * supposed to inherit the root state.
2439  */
2440 int blk_get_open_flags_from_root_state(BlockBackend *blk)
2441 {
2442     GLOBAL_STATE_CODE();
2443     return blk->root_state.open_flags;
2444 }
2445
2446 BlockBackendRootState *blk_get_root_state(BlockBackend *blk)
2447 {
2448     GLOBAL_STATE_CODE();
2449     return &blk->root_state;
2450 }
2451
2452 int blk_commit_all(void)
2453 {
2454     BlockBackend *blk = NULL;
2455     GLOBAL_STATE_CODE();
2456
2457     while ((blk = blk_all_next(blk)) != NULL) {
2458         AioContext *aio_context = blk_get_aio_context(blk);
2459         BlockDriverState *unfiltered_bs = bdrv_skip_filters(blk_bs(blk));
2460
2461         aio_context_acquire(aio_context);
2462         if (blk_is_inserted(blk) && bdrv_cow_child(unfiltered_bs)) {
2463             int ret;
2464
2465             ret = bdrv_commit(unfiltered_bs);
2466             if (ret < 0) {
2467                 aio_context_release(aio_context);
2468                 return ret;
2469             }
2470         }
2471         aio_context_release(aio_context);
2472     }
2473     return 0;
2474 }
2475
2476
2477 /* throttling disk I/O limits */
2478 void blk_set_io_limits(BlockBackend *blk, ThrottleConfig *cfg)
2479 {
2480     GLOBAL_STATE_CODE();
2481     throttle_group_config(&blk->public.throttle_group_member, cfg);
2482 }
2483
2484 void blk_io_limits_disable(BlockBackend *blk)
2485 {
2486     BlockDriverState *bs = blk_bs(blk);
2487     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
2488     assert(tgm->throttle_state);
2489     GLOBAL_STATE_CODE();
2490     if (bs) {
2491         bdrv_ref(bs);
2492         bdrv_drained_begin(bs);
2493     }
2494     throttle_group_unregister_tgm(tgm);
2495     if (bs) {
2496         bdrv_drained_end(bs);
2497         bdrv_unref(bs);
2498     }
2499 }
2500
2501 /* should be called before blk_set_io_limits if a limit is set */
2502 void blk_io_limits_enable(BlockBackend *blk, const char *group)
2503 {
2504     assert(!blk->public.throttle_group_member.throttle_state);
2505     GLOBAL_STATE_CODE();
2506     throttle_group_register_tgm(&blk->public.throttle_group_member,
2507                                 group, blk_get_aio_context(blk));
2508 }
2509
2510 void blk_io_limits_update_group(BlockBackend *blk, const char *group)
2511 {
2512     GLOBAL_STATE_CODE();
2513     /* this BB is not part of any group */
2514     if (!blk->public.throttle_group_member.throttle_state) {
2515         return;
2516     }
2517
2518     /* this BB is a part of the same group than the one we want */
2519     if (!g_strcmp0(throttle_group_get_name(&blk->public.throttle_group_member),
2520                 group)) {
2521         return;
2522     }
2523
2524     /* need to change the group this bs belong to */
2525     blk_io_limits_disable(blk);
2526     blk_io_limits_enable(blk, group);
2527 }
2528
2529 static void blk_root_drained_begin(BdrvChild *child)
2530 {
2531     BlockBackend *blk = child->opaque;
2532     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
2533
2534     if (++blk->quiesce_counter == 1) {
2535         if (blk->dev_ops && blk->dev_ops->drained_begin) {
2536             blk->dev_ops->drained_begin(blk->dev_opaque);
2537         }
2538     }
2539
2540     /* Note that blk->root may not be accessible here yet if we are just
2541      * attaching to a BlockDriverState that is drained. Use child instead. */
2542
2543     if (qatomic_fetch_inc(&tgm->io_limits_disabled) == 0) {
2544         throttle_group_restart_tgm(tgm);
2545     }
2546 }
2547
2548 static bool blk_root_drained_poll(BdrvChild *child)
2549 {
2550     BlockBackend *blk = child->opaque;
2551     bool busy = false;
2552     assert(blk->quiesce_counter);
2553
2554     if (blk->dev_ops && blk->dev_ops->drained_poll) {
2555         busy = blk->dev_ops->drained_poll(blk->dev_opaque);
2556     }
2557     return busy || !!blk->in_flight;
2558 }
2559
2560 static void blk_root_drained_end(BdrvChild *child, int *drained_end_counter)
2561 {
2562     BlockBackend *blk = child->opaque;
2563     assert(blk->quiesce_counter);
2564
2565     assert(blk->public.throttle_group_member.io_limits_disabled);
2566     qatomic_dec(&blk->public.throttle_group_member.io_limits_disabled);
2567
2568     if (--blk->quiesce_counter == 0) {
2569         if (blk->dev_ops && blk->dev_ops->drained_end) {
2570             blk->dev_ops->drained_end(blk->dev_opaque);
2571         }
2572         while (qemu_co_enter_next(&blk->queued_requests, NULL)) {
2573             /* Resume all queued requests */
2574         }
2575     }
2576 }
2577
2578 void blk_register_buf(BlockBackend *blk, void *host, size_t size)
2579 {
2580     GLOBAL_STATE_CODE();
2581     bdrv_register_buf(blk_bs(blk), host, size);
2582 }
2583
2584 void blk_unregister_buf(BlockBackend *blk, void *host)
2585 {
2586     GLOBAL_STATE_CODE();
2587     bdrv_unregister_buf(blk_bs(blk), host);
2588 }
2589
2590 int coroutine_fn blk_co_copy_range(BlockBackend *blk_in, int64_t off_in,
2591                                    BlockBackend *blk_out, int64_t off_out,
2592                                    int64_t bytes, BdrvRequestFlags read_flags,
2593                                    BdrvRequestFlags write_flags)
2594 {
2595     int r;
2596     IO_CODE();
2597
2598     r = blk_check_byte_request(blk_in, off_in, bytes);
2599     if (r) {
2600         return r;
2601     }
2602     r = blk_check_byte_request(blk_out, off_out, bytes);
2603     if (r) {
2604         return r;
2605     }
2606     return bdrv_co_copy_range(blk_in->root, off_in,
2607                               blk_out->root, off_out,
2608                               bytes, read_flags, write_flags);
2609 }
2610
2611 const BdrvChild *blk_root(BlockBackend *blk)
2612 {
2613     GLOBAL_STATE_CODE();
2614     return blk->root;
2615 }
2616
2617 int blk_make_empty(BlockBackend *blk, Error **errp)
2618 {
2619     GLOBAL_STATE_CODE();
2620     if (!blk_is_available(blk)) {
2621         error_setg(errp, "No medium inserted");
2622         return -ENOMEDIUM;
2623     }
2624
2625     return bdrv_make_empty(blk->root, errp);
2626 }