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);
 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_change_aio_ctx(BdrvChild *child, AioContext *ctx,
 138                                     GHashTable *visited, Transaction *tran,
 139                                     Error **errp);
 140
 141 static char *blk_root_get_parent_desc(BdrvChild *child)
 142 {
 143     BlockBackend *blk = child->opaque;
 144     g_autofree char *dev_id = NULL;
 145
 146     if (blk->name) {
 147         return g_strdup_printf("block device '%s'", blk->name);
 148     }
 149
 150     dev_id = blk_get_attached_dev_id(blk);
 151     if (*dev_id) {
 152         return g_strdup_printf("block device '%s'", dev_id);
 153     } else {
 154         /* TODO Callback into the BB owner for something more detailed */
 155         return g_strdup("an unnamed block device");
 156     }
 157 }
 158
 159 static const char *blk_root_get_name(BdrvChild *child)
 160 {
 161     return blk_name(child->opaque);
 162 }
 163
 164 static void blk_vm_state_changed(void *opaque, bool running, RunState state)
 165 {
 166     Error *local_err = NULL;
 167     BlockBackend *blk = opaque;
 168
 169     if (state == RUN_STATE_INMIGRATE) {
 170         return;
 171     }
 172
 173     qemu_del_vm_change_state_handler(blk->vmsh);
 174     blk->vmsh = NULL;
 175     blk_set_perm(blk, blk->perm, blk->shared_perm, &local_err);
 176     if (local_err) {
 177         error_report_err(local_err);
 178     }
 179 }
 180
 181 /*
 182  * Notifies the user of the BlockBackend that migration has completed. qdev
 183  * devices can tighten their permissions in response (specifically revoke
 184  * shared write permissions that we needed for storage migration).
 185  *
 186  * If an error is returned, the VM cannot be allowed to be resumed.
 187  */
 188 static void blk_root_activate(BdrvChild *child, Error **errp)
 189 {
 190     BlockBackend *blk = child->opaque;
 191     Error *local_err = NULL;
 192     uint64_t saved_shared_perm;
 193
 194     if (!blk->disable_perm) {
 195         return;
 196     }
 197
 198     blk->disable_perm = false;
 199
 200     /*
 201      * blk->shared_perm contains the permissions we want to share once
 202      * migration is really completely done.  For now, we need to share
 203      * all; but we also need to retain blk->shared_perm, which is
 204      * overwritten by a successful blk_set_perm() call.  Save it and
 205      * restore it below.
 206      */
 207     saved_shared_perm = blk->shared_perm;
 208
 209     blk_set_perm(blk, blk->perm, BLK_PERM_ALL, &local_err);
 210     if (local_err) {
 211         error_propagate(errp, local_err);
 212         blk->disable_perm = true;
 213         return;
 214     }
 215     blk->shared_perm = saved_shared_perm;
 216
 217     if (runstate_check(RUN_STATE_INMIGRATE)) {
 218         /* Activation can happen when migration process is still active, for
 219          * example when nbd_server_add is called during non-shared storage
 220          * migration. Defer the shared_perm update to migration completion. */
 221         if (!blk->vmsh) {
 222             blk->vmsh = qemu_add_vm_change_state_handler(blk_vm_state_changed,
 223                                                          blk);
 224         }
 225         return;
 226     }
 227
 228     blk_set_perm(blk, blk->perm, blk->shared_perm, &local_err);
 229     if (local_err) {
 230         error_propagate(errp, local_err);
 231         blk->disable_perm = true;
 232         return;
 233     }
 234 }
 235
 236 void blk_set_force_allow_inactivate(BlockBackend *blk)
 237 {
 238     GLOBAL_STATE_CODE();
 239     blk->force_allow_inactivate = true;
 240 }
 241
 242 static bool blk_can_inactivate(BlockBackend *blk)
 243 {
 244     /* If it is a guest device, inactivate is ok. */
 245     if (blk->dev || blk_name(blk)[0]) {
 246         return true;
 247     }
 248
 249     /* Inactivating means no more writes to the image can be done,
 250      * even if those writes would be changes invisible to the
 251      * guest.  For block job BBs that satisfy this, we can just allow
 252      * it.  This is the case for mirror job source, which is required
 253      * by libvirt non-shared block migration. */
 254     if (!(blk->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED))) {
 255         return true;
 256     }
 257
 258     return blk->force_allow_inactivate;
 259 }
 260
 261 static int blk_root_inactivate(BdrvChild *child)
 262 {
 263     BlockBackend *blk = child->opaque;
 264
 265     if (blk->disable_perm) {
 266         return 0;
 267     }
 268
 269     if (!blk_can_inactivate(blk)) {
 270         return -EPERM;
 271     }
 272
 273     blk->disable_perm = true;
 274     if (blk->root) {
 275         bdrv_child_try_set_perm(blk->root, 0, BLK_PERM_ALL, &error_abort);
 276     }
 277
 278     return 0;
 279 }
 280
 281 static void blk_root_attach(BdrvChild *child)
 282 {
 283     BlockBackend *blk = child->opaque;
 284     BlockBackendAioNotifier *notifier;
 285
 286     trace_blk_root_attach(child, blk, child->bs);
 287
 288     QLIST_FOREACH(notifier, &blk->aio_notifiers, list) {
 289         bdrv_add_aio_context_notifier(child->bs,
 290                 notifier->attached_aio_context,
 291                 notifier->detach_aio_context,
 292                 notifier->opaque);
 293     }
 294 }
 295
 296 static void blk_root_detach(BdrvChild *child)
 297 {
 298     BlockBackend *blk = child->opaque;
 299     BlockBackendAioNotifier *notifier;
 300
 301     trace_blk_root_detach(child, blk, child->bs);
 302
 303     QLIST_FOREACH(notifier, &blk->aio_notifiers, list) {
 304         bdrv_remove_aio_context_notifier(child->bs,
 305                 notifier->attached_aio_context,
 306                 notifier->detach_aio_context,
 307                 notifier->opaque);
 308     }
 309 }
 310
 311 static AioContext *blk_root_get_parent_aio_context(BdrvChild *c)
 312 {
 313     BlockBackend *blk = c->opaque;
 314     IO_CODE();
 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     .change_aio_ctx     = blk_root_change_aio_ctx,
 338
 339     .get_parent_aio_context = blk_root_get_parent_aio_context,
 340 };
 341
 342 /*
 343  * Create a new BlockBackend with a reference count of one.
 344  *
 345  * @perm is a bitmasks of BLK_PERM_* constants which describes the permissions
 346  * to request for a block driver node that is attached to this BlockBackend.
 347  * @shared_perm is a bitmask which describes which permissions may be granted
 348  * to other users of the attached node.
 349  * Both sets of permissions can be changed later using blk_set_perm().
 350  *
 351  * Return the new BlockBackend on success, null on failure.
 352  */
 353 BlockBackend *blk_new(AioContext *ctx, uint64_t perm, uint64_t shared_perm)
 354 {
 355     BlockBackend *blk;
 356
 357     GLOBAL_STATE_CODE();
 358
 359     blk = g_new0(BlockBackend, 1);
 360     blk->refcnt = 1;
 361     blk->ctx = ctx;
 362     blk->perm = perm;
 363     blk->shared_perm = shared_perm;
 364     blk_set_enable_write_cache(blk, true);
 365
 366     blk->on_read_error = BLOCKDEV_ON_ERROR_REPORT;
 367     blk->on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
 368
 369     block_acct_init(&blk->stats);
 370
 371     qemu_co_queue_init(&blk->queued_requests);
 372     notifier_list_init(&blk->remove_bs_notifiers);
 373     notifier_list_init(&blk->insert_bs_notifiers);
 374     QLIST_INIT(&blk->aio_notifiers);
 375
 376     QTAILQ_INSERT_TAIL(&block_backends, blk, link);
 377     return blk;
 378 }
 379
 380 /*
 381  * Create a new BlockBackend connected to an existing BlockDriverState.
 382  *
 383  * @perm is a bitmasks of BLK_PERM_* constants which describes the
 384  * permissions to request for @bs that is attached to this
 385  * BlockBackend.  @shared_perm is a bitmask which describes which
 386  * permissions may be granted to other users of the attached node.
 387  * Both sets of permissions can be changed later using blk_set_perm().
 388  *
 389  * Return the new BlockBackend on success, null on failure.
 390  */
 391 BlockBackend *blk_new_with_bs(BlockDriverState *bs, uint64_t perm,
 392                               uint64_t shared_perm, Error **errp)
 393 {
 394     BlockBackend *blk = blk_new(bdrv_get_aio_context(bs), perm, shared_perm);
 395
 396     GLOBAL_STATE_CODE();
 397
 398     if (blk_insert_bs(blk, bs, errp) < 0) {
 399         blk_unref(blk);
 400         return NULL;
 401     }
 402     return blk;
 403 }
 404
 405 /*
 406  * Creates a new BlockBackend, opens a new BlockDriverState, and connects both.
 407  * The new BlockBackend is in the main AioContext.
 408  *
 409  * Just as with bdrv_open(), after having called this function the reference to
 410  * @options belongs to the block layer (even on failure).
 411  *
 412  * TODO: Remove @filename and @flags; it should be possible to specify a whole
 413  * BDS tree just by specifying the @options QDict (or @reference,
 414  * alternatively). At the time of adding this function, this is not possible,
 415  * though, so callers of this function have to be able to specify @filename and
 416  * @flags.
 417  */
 418 BlockBackend *blk_new_open(const char *filename, const char *reference,
 419                            QDict *options, int flags, Error **errp)
 420 {
 421     BlockBackend *blk;
 422     BlockDriverState *bs;
 423     uint64_t perm = 0;
 424     uint64_t shared = BLK_PERM_ALL;
 425
 426     GLOBAL_STATE_CODE();
 427
 428     /*
 429      * blk_new_open() is mainly used in .bdrv_create implementations and the
 430      * tools where sharing isn't a major concern because the BDS stays private
 431      * and the file is generally not supposed to be used by a second process,
 432      * so we just request permission according to the flags.
 433      *
 434      * The exceptions are xen_disk and blockdev_init(); in these cases, the
 435      * caller of blk_new_open() doesn't make use of the permissions, but they
 436      * shouldn't hurt either. We can still share everything here because the
 437      * guest devices will add their own blockers if they can't share.
 438      */
 439     if ((flags & BDRV_O_NO_IO) == 0) {
 440         perm |= BLK_PERM_CONSISTENT_READ;
 441         if (flags & BDRV_O_RDWR) {
 442             perm |= BLK_PERM_WRITE;
 443         }
 444     }
 445     if (flags & BDRV_O_RESIZE) {
 446         perm |= BLK_PERM_RESIZE;
 447     }
 448     if (flags & BDRV_O_NO_SHARE) {
 449         shared = BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
 450     }
 451
 452     blk = blk_new(qemu_get_aio_context(), perm, shared);
 453     bs = bdrv_open(filename, reference, options, flags, errp);
 454     if (!bs) {
 455         blk_unref(blk);
 456         return NULL;
 457     }
 458
 459     blk->root = bdrv_root_attach_child(bs, "root", &child_root,
 460                                        BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY,
 461                                        perm, shared, blk, errp);
 462     if (!blk->root) {
 463         blk_unref(blk);
 464         return NULL;
 465     }
 466
 467     return blk;
 468 }
 469
 470 static void blk_delete(BlockBackend *blk)
 471 {
 472     assert(!blk->refcnt);
 473     assert(!blk->name);
 474     assert(!blk->dev);
 475     if (blk->public.throttle_group_member.throttle_state) {
 476         blk_io_limits_disable(blk);
 477     }
 478     if (blk->root) {
 479         blk_remove_bs(blk);
 480     }
 481     if (blk->vmsh) {
 482         qemu_del_vm_change_state_handler(blk->vmsh);
 483         blk->vmsh = NULL;
 484     }
 485     assert(QLIST_EMPTY(&blk->remove_bs_notifiers.notifiers));
 486     assert(QLIST_EMPTY(&blk->insert_bs_notifiers.notifiers));
 487     assert(QLIST_EMPTY(&blk->aio_notifiers));
 488     QTAILQ_REMOVE(&block_backends, blk, link);
 489     drive_info_del(blk->legacy_dinfo);
 490     block_acct_cleanup(&blk->stats);
 491     g_free(blk);
 492 }
 493
 494 static void drive_info_del(DriveInfo *dinfo)
 495 {
 496     if (!dinfo) {
 497         return;
 498     }
 499     qemu_opts_del(dinfo->opts);
 500     g_free(dinfo);
 501 }
 502
 503 int blk_get_refcnt(BlockBackend *blk)
 504 {
 505     GLOBAL_STATE_CODE();
 506     return blk ? blk->refcnt : 0;
 507 }
 508
 509 /*
 510  * Increment @blk's reference count.
 511  * @blk must not be null.
 512  */
 513 void blk_ref(BlockBackend *blk)
 514 {
 515     assert(blk->refcnt > 0);
 516     GLOBAL_STATE_CODE();
 517     blk->refcnt++;
 518 }
 519
 520 /*
 521  * Decrement @blk's reference count.
 522  * If this drops it to zero, destroy @blk.
 523  * For convenience, do nothing if @blk is null.
 524  */
 525 void blk_unref(BlockBackend *blk)
 526 {
 527     GLOBAL_STATE_CODE();
 528     if (blk) {
 529         assert(blk->refcnt > 0);
 530         if (blk->refcnt > 1) {
 531             blk->refcnt--;
 532         } else {
 533             blk_drain(blk);
 534             /* blk_drain() cannot resurrect blk, nobody held a reference */
 535             assert(blk->refcnt == 1);
 536             blk->refcnt = 0;
 537             blk_delete(blk);
 538         }
 539     }
 540 }
 541
 542 /*
 543  * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
 544  * ones which are hidden (i.e. are not referenced by the monitor).
 545  */
 546 BlockBackend *blk_all_next(BlockBackend *blk)
 547 {
 548     GLOBAL_STATE_CODE();
 549     return blk ? QTAILQ_NEXT(blk, link)
 550                : QTAILQ_FIRST(&block_backends);
 551 }
 552
 553 void blk_remove_all_bs(void)
 554 {
 555     BlockBackend *blk = NULL;
 556
 557     GLOBAL_STATE_CODE();
 558
 559     while ((blk = blk_all_next(blk)) != NULL) {
 560         AioContext *ctx = blk_get_aio_context(blk);
 561
 562         aio_context_acquire(ctx);
 563         if (blk->root) {
 564             blk_remove_bs(blk);
 565         }
 566         aio_context_release(ctx);
 567     }
 568 }
 569
 570 /*
 571  * Return the monitor-owned BlockBackend after @blk.
 572  * If @blk is null, return the first one.
 573  * Else, return @blk's next sibling, which may be null.
 574  *
 575  * To iterate over all BlockBackends, do
 576  * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
 577  *     ...
 578  * }
 579  */
 580 BlockBackend *blk_next(BlockBackend *blk)
 581 {
 582     GLOBAL_STATE_CODE();
 583     return blk ? QTAILQ_NEXT(blk, monitor_link)
 584                : QTAILQ_FIRST(&monitor_block_backends);
 585 }
 586
 587 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
 588  * the monitor or attached to a BlockBackend */
 589 BlockDriverState *bdrv_next(BdrvNextIterator *it)
 590 {
 591     BlockDriverState *bs, *old_bs;
 592
 593     /* Must be called from the main loop */
 594     assert(qemu_get_current_aio_context() == qemu_get_aio_context());
 595
 596     /* First, return all root nodes of BlockBackends. In order to avoid
 597      * returning a BDS twice when multiple BBs refer to it, we only return it
 598      * if the BB is the first one in the parent list of the BDS. */
 599     if (it->phase == BDRV_NEXT_BACKEND_ROOTS) {
 600         BlockBackend *old_blk = it->blk;
 601
 602         old_bs = old_blk ? blk_bs(old_blk) : NULL;
 603
 604         do {
 605             it->blk = blk_all_next(it->blk);
 606             bs = it->blk ? blk_bs(it->blk) : NULL;
 607         } while (it->blk && (bs == NULL || bdrv_first_blk(bs) != it->blk));
 608
 609         if (it->blk) {
 610             blk_ref(it->blk);
 611         }
 612         blk_unref(old_blk);
 613
 614         if (bs) {
 615             bdrv_ref(bs);
 616             bdrv_unref(old_bs);
 617             return bs;
 618         }
 619         it->phase = BDRV_NEXT_MONITOR_OWNED;
 620     } else {
 621         old_bs = it->bs;
 622     }
 623
 624     /* Then return the monitor-owned BDSes without a BB attached. Ignore all
 625      * BDSes that are attached to a BlockBackend here; they have been handled
 626      * by the above block already */
 627     do {
 628         it->bs = bdrv_next_monitor_owned(it->bs);
 629         bs = it->bs;
 630     } while (bs && bdrv_has_blk(bs));
 631
 632     if (bs) {
 633         bdrv_ref(bs);
 634     }
 635     bdrv_unref(old_bs);
 636
 637     return bs;
 638 }
 639
 640 static void bdrv_next_reset(BdrvNextIterator *it)
 641 {
 642     *it = (BdrvNextIterator) {
 643         .phase = BDRV_NEXT_BACKEND_ROOTS,
 644     };
 645 }
 646
 647 BlockDriverState *bdrv_first(BdrvNextIterator *it)
 648 {
 649     GLOBAL_STATE_CODE();
 650     bdrv_next_reset(it);
 651     return bdrv_next(it);
 652 }
 653
 654 /* Must be called when aborting a bdrv_next() iteration before
 655  * bdrv_next() returns NULL */
 656 void bdrv_next_cleanup(BdrvNextIterator *it)
 657 {
 658     /* Must be called from the main loop */
 659     assert(qemu_get_current_aio_context() == qemu_get_aio_context());
 660
 661     if (it->phase == BDRV_NEXT_BACKEND_ROOTS) {
 662         if (it->blk) {
 663             bdrv_unref(blk_bs(it->blk));
 664             blk_unref(it->blk);
 665         }
 666     } else {
 667         bdrv_unref(it->bs);
 668     }
 669
 670     bdrv_next_reset(it);
 671 }
 672
 673 /*
 674  * Add a BlockBackend into the list of backends referenced by the monitor, with
 675  * the given @name acting as the handle for the monitor.
 676  * Strictly for use by blockdev.c.
 677  *
 678  * @name must not be null or empty.
 679  *
 680  * Returns true on success and false on failure. In the latter case, an Error
 681  * object is returned through @errp.
 682  */
 683 bool monitor_add_blk(BlockBackend *blk, const char *name, Error **errp)
 684 {
 685     assert(!blk->name);
 686     assert(name && name[0]);
 687     GLOBAL_STATE_CODE();
 688
 689     if (!id_wellformed(name)) {
 690         error_setg(errp, "Invalid device name");
 691         return false;
 692     }
 693     if (blk_by_name(name)) {
 694         error_setg(errp, "Device with id '%s' already exists", name);
 695         return false;
 696     }
 697     if (bdrv_find_node(name)) {
 698         error_setg(errp,
 699                    "Device name '%s' conflicts with an existing node name",
 700                    name);
 701         return false;
 702     }
 703
 704     blk->name = g_strdup(name);
 705     QTAILQ_INSERT_TAIL(&monitor_block_backends, blk, monitor_link);
 706     return true;
 707 }
 708
 709 /*
 710  * Remove a BlockBackend from the list of backends referenced by the monitor.
 711  * Strictly for use by blockdev.c.
 712  */
 713 void monitor_remove_blk(BlockBackend *blk)
 714 {
 715     GLOBAL_STATE_CODE();
 716
 717     if (!blk->name) {
 718         return;
 719     }
 720
 721     QTAILQ_REMOVE(&monitor_block_backends, blk, monitor_link);
 722     g_free(blk->name);
 723     blk->name = NULL;
 724 }
 725
 726 /*
 727  * Return @blk's name, a non-null string.
 728  * Returns an empty string iff @blk is not referenced by the monitor.
 729  */
 730 const char *blk_name(const BlockBackend *blk)
 731 {
 732     IO_CODE();
 733     return blk->name ?: "";
 734 }
 735
 736 /*
 737  * Return the BlockBackend with name @name if it exists, else null.
 738  * @name must not be null.
 739  */
 740 BlockBackend *blk_by_name(const char *name)
 741 {
 742     BlockBackend *blk = NULL;
 743
 744     GLOBAL_STATE_CODE();
 745     assert(name);
 746     while ((blk = blk_next(blk)) != NULL) {
 747         if (!strcmp(name, blk->name)) {
 748             return blk;
 749         }
 750     }
 751     return NULL;
 752 }
 753
 754 /*
 755  * Return the BlockDriverState attached to @blk if any, else null.
 756  */
 757 BlockDriverState *blk_bs(BlockBackend *blk)
 758 {
 759     IO_CODE();
 760     return blk->root ? blk->root->bs : NULL;
 761 }
 762
 763 static BlockBackend *bdrv_first_blk(BlockDriverState *bs)
 764 {
 765     BdrvChild *child;
 766
 767     GLOBAL_STATE_CODE();
 768
 769     QLIST_FOREACH(child, &bs->parents, next_parent) {
 770         if (child->klass == &child_root) {
 771             return child->opaque;
 772         }
 773     }
 774
 775     return NULL;
 776 }
 777
 778 /*
 779  * Returns true if @bs has an associated BlockBackend.
 780  */
 781 bool bdrv_has_blk(BlockDriverState *bs)
 782 {
 783     GLOBAL_STATE_CODE();
 784     return bdrv_first_blk(bs) != NULL;
 785 }
 786
 787 /*
 788  * Returns true if @bs has only BlockBackends as parents.
 789  */
 790 bool bdrv_is_root_node(BlockDriverState *bs)
 791 {
 792     BdrvChild *c;
 793
 794     GLOBAL_STATE_CODE();
 795     QLIST_FOREACH(c, &bs->parents, next_parent) {
 796         if (c->klass != &child_root) {
 797             return false;
 798         }
 799     }
 800
 801     return true;
 802 }
 803
 804 /*
 805  * Return @blk's DriveInfo if any, else null.
 806  */
 807 DriveInfo *blk_legacy_dinfo(BlockBackend *blk)
 808 {
 809     GLOBAL_STATE_CODE();
 810     return blk->legacy_dinfo;
 811 }
 812
 813 /*
 814  * Set @blk's DriveInfo to @dinfo, and return it.
 815  * @blk must not have a DriveInfo set already.
 816  * No other BlockBackend may have the same DriveInfo set.
 817  */
 818 DriveInfo *blk_set_legacy_dinfo(BlockBackend *blk, DriveInfo *dinfo)
 819 {
 820     assert(!blk->legacy_dinfo);
 821     GLOBAL_STATE_CODE();
 822     return blk->legacy_dinfo = dinfo;
 823 }
 824
 825 /*
 826  * Return the BlockBackend with DriveInfo @dinfo.
 827  * It must exist.
 828  */
 829 BlockBackend *blk_by_legacy_dinfo(DriveInfo *dinfo)
 830 {
 831     BlockBackend *blk = NULL;
 832     GLOBAL_STATE_CODE();
 833
 834     while ((blk = blk_next(blk)) != NULL) {
 835         if (blk->legacy_dinfo == dinfo) {
 836             return blk;
 837         }
 838     }
 839     abort();
 840 }
 841
 842 /*
 843  * Returns a pointer to the publicly accessible fields of @blk.
 844  */
 845 BlockBackendPublic *blk_get_public(BlockBackend *blk)
 846 {
 847     GLOBAL_STATE_CODE();
 848     return &blk->public;
 849 }
 850
 851 /*
 852  * Returns a BlockBackend given the associated @public fields.
 853  */
 854 BlockBackend *blk_by_public(BlockBackendPublic *public)
 855 {
 856     GLOBAL_STATE_CODE();
 857     return container_of(public, BlockBackend, public);
 858 }
 859
 860 /*
 861  * Disassociates the currently associated BlockDriverState from @blk.
 862  */
 863 void blk_remove_bs(BlockBackend *blk)
 864 {
 865     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
 866     BdrvChild *root;
 867
 868     GLOBAL_STATE_CODE();
 869
 870     notifier_list_notify(&blk->remove_bs_notifiers, blk);
 871     if (tgm->throttle_state) {
 872         BlockDriverState *bs = blk_bs(blk);
 873
 874         /*
 875          * Take a ref in case blk_bs() changes across bdrv_drained_begin(), for
 876          * example, if a temporary filter node is removed by a blockjob.
 877          */
 878         bdrv_ref(bs);
 879         bdrv_drained_begin(bs);
 880         throttle_group_detach_aio_context(tgm);
 881         throttle_group_attach_aio_context(tgm, qemu_get_aio_context());
 882         bdrv_drained_end(bs);
 883         bdrv_unref(bs);
 884     }
 885
 886     blk_update_root_state(blk);
 887
 888     /* bdrv_root_unref_child() will cause blk->root to become stale and may
 889      * switch to a completion coroutine later on. Let's drain all I/O here
 890      * to avoid that and a potential QEMU crash.
 891      */
 892     blk_drain(blk);
 893     root = blk->root;
 894     blk->root = NULL;
 895     bdrv_root_unref_child(root);
 896 }
 897
 898 /*
 899  * Associates a new BlockDriverState with @blk.
 900  */
 901 int blk_insert_bs(BlockBackend *blk, BlockDriverState *bs, Error **errp)
 902 {
 903     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
 904     GLOBAL_STATE_CODE();
 905     bdrv_ref(bs);
 906     blk->root = bdrv_root_attach_child(bs, "root", &child_root,
 907                                        BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY,
 908                                        blk->perm, blk->shared_perm,
 909                                        blk, errp);
 910     if (blk->root == NULL) {
 911         return -EPERM;
 912     }
 913
 914     notifier_list_notify(&blk->insert_bs_notifiers, blk);
 915     if (tgm->throttle_state) {
 916         throttle_group_detach_aio_context(tgm);
 917         throttle_group_attach_aio_context(tgm, bdrv_get_aio_context(bs));
 918     }
 919
 920     return 0;
 921 }
 922
 923 /*
 924  * Change BlockDriverState associated with @blk.
 925  */
 926 int blk_replace_bs(BlockBackend *blk, BlockDriverState *new_bs, Error **errp)
 927 {
 928     GLOBAL_STATE_CODE();
 929     return bdrv_replace_child_bs(blk->root, new_bs, errp);
 930 }
 931
 932 /*
 933  * Sets the permission bitmasks that the user of the BlockBackend needs.
 934  */
 935 int blk_set_perm(BlockBackend *blk, uint64_t perm, uint64_t shared_perm,
 936                  Error **errp)
 937 {
 938     int ret;
 939     GLOBAL_STATE_CODE();
 940
 941     if (blk->root && !blk->disable_perm) {
 942         ret = bdrv_child_try_set_perm(blk->root, perm, shared_perm, errp);
 943         if (ret < 0) {
 944             return ret;
 945         }
 946     }
 947
 948     blk->perm = perm;
 949     blk->shared_perm = shared_perm;
 950
 951     return 0;
 952 }
 953
 954 void blk_get_perm(BlockBackend *blk, uint64_t *perm, uint64_t *shared_perm)
 955 {
 956     GLOBAL_STATE_CODE();
 957     *perm = blk->perm;
 958     *shared_perm = blk->shared_perm;
 959 }
 960
 961 /*
 962  * Attach device model @dev to @blk.
 963  * Return 0 on success, -EBUSY when a device model is attached already.
 964  */
 965 int blk_attach_dev(BlockBackend *blk, DeviceState *dev)
 966 {
 967     GLOBAL_STATE_CODE();
 968     if (blk->dev) {
 969         return -EBUSY;
 970     }
 971
 972     /* While migration is still incoming, we don't need to apply the
 973      * permissions of guest device BlockBackends. We might still have a block
 974      * job or NBD server writing to the image for storage migration. */
 975     if (runstate_check(RUN_STATE_INMIGRATE)) {
 976         blk->disable_perm = true;
 977     }
 978
 979     blk_ref(blk);
 980     blk->dev = dev;
 981     blk_iostatus_reset(blk);
 982
 983     return 0;
 984 }
 985
 986 /*
 987  * Detach device model @dev from @blk.
 988  * @dev must be currently attached to @blk.
 989  */
 990 void blk_detach_dev(BlockBackend *blk, DeviceState *dev)
 991 {
 992     assert(blk->dev == dev);
 993     GLOBAL_STATE_CODE();
 994     blk->dev = NULL;
 995     blk->dev_ops = NULL;
 996     blk->dev_opaque = NULL;
 997     blk_set_perm(blk, 0, BLK_PERM_ALL, &error_abort);
 998     blk_unref(blk);
 999 }
1000
1001 /*
1002  * Return the device model attached to @blk if any, else null.
1003  */
1004 DeviceState *blk_get_attached_dev(BlockBackend *blk)
1005 {
1006     GLOBAL_STATE_CODE();
1007     return blk->dev;
1008 }
1009
1010 /* Return the qdev ID, or if no ID is assigned the QOM path, of the block
1011  * device attached to the BlockBackend. */
1012 char *blk_get_attached_dev_id(BlockBackend *blk)
1013 {
1014     DeviceState *dev = blk->dev;
1015     IO_CODE();
1016
1017     if (!dev) {
1018         return g_strdup("");
1019     } else if (dev->id) {
1020         return g_strdup(dev->id);
1021     }
1022
1023     return object_get_canonical_path(OBJECT(dev)) ?: g_strdup("");
1024 }
1025
1026 /*
1027  * Return the BlockBackend which has the device model @dev attached if it
1028  * exists, else null.
1029  *
1030  * @dev must not be null.
1031  */
1032 BlockBackend *blk_by_dev(void *dev)
1033 {
1034     BlockBackend *blk = NULL;
1035
1036     GLOBAL_STATE_CODE();
1037
1038     assert(dev != NULL);
1039     while ((blk = blk_all_next(blk)) != NULL) {
1040         if (blk->dev == dev) {
1041             return blk;
1042         }
1043     }
1044     return NULL;
1045 }
1046
1047 /*
1048  * Set @blk's device model callbacks to @ops.
1049  * @opaque is the opaque argument to pass to the callbacks.
1050  * This is for use by device models.
1051  */
1052 void blk_set_dev_ops(BlockBackend *blk, const BlockDevOps *ops,
1053                      void *opaque)
1054 {
1055     GLOBAL_STATE_CODE();
1056     blk->dev_ops = ops;
1057     blk->dev_opaque = opaque;
1058
1059     /* Are we currently quiesced? Should we enforce this right now? */
1060     if (blk->quiesce_counter && ops && ops->drained_begin) {
1061         ops->drained_begin(opaque);
1062     }
1063 }
1064
1065 /*
1066  * Notify @blk's attached device model of media change.
1067  *
1068  * If @load is true, notify of media load. This action can fail, meaning that
1069  * the medium cannot be loaded. @errp is set then.
1070  *
1071  * If @load is false, notify of media eject. This can never fail.
1072  *
1073  * Also send DEVICE_TRAY_MOVED events as appropriate.
1074  */
1075 void blk_dev_change_media_cb(BlockBackend *blk, bool load, Error **errp)
1076 {
1077     GLOBAL_STATE_CODE();
1078     if (blk->dev_ops && blk->dev_ops->change_media_cb) {
1079         bool tray_was_open, tray_is_open;
1080         Error *local_err = NULL;
1081
1082         tray_was_open = blk_dev_is_tray_open(blk);
1083         blk->dev_ops->change_media_cb(blk->dev_opaque, load, &local_err);
1084         if (local_err) {
1085             assert(load == true);
1086             error_propagate(errp, local_err);
1087             return;
1088         }
1089         tray_is_open = blk_dev_is_tray_open(blk);
1090
1091         if (tray_was_open != tray_is_open) {
1092             char *id = blk_get_attached_dev_id(blk);
1093             qapi_event_send_device_tray_moved(blk_name(blk), id, tray_is_open);
1094             g_free(id);
1095         }
1096     }
1097 }
1098
1099 static void blk_root_change_media(BdrvChild *child, bool load)
1100 {
1101     blk_dev_change_media_cb(child->opaque, load, NULL);
1102 }
1103
1104 /*
1105  * Does @blk's attached device model have removable media?
1106  * %true if no device model is attached.
1107  */
1108 bool blk_dev_has_removable_media(BlockBackend *blk)
1109 {
1110     GLOBAL_STATE_CODE();
1111     return !blk->dev || (blk->dev_ops && blk->dev_ops->change_media_cb);
1112 }
1113
1114 /*
1115  * Does @blk's attached device model have a tray?
1116  */
1117 bool blk_dev_has_tray(BlockBackend *blk)
1118 {
1119     IO_CODE();
1120     return blk->dev_ops && blk->dev_ops->is_tray_open;
1121 }
1122
1123 /*
1124  * Notify @blk's attached device model of a media eject request.
1125  * If @force is true, the medium is about to be yanked out forcefully.
1126  */
1127 void blk_dev_eject_request(BlockBackend *blk, bool force)
1128 {
1129     GLOBAL_STATE_CODE();
1130     if (blk->dev_ops && blk->dev_ops->eject_request_cb) {
1131         blk->dev_ops->eject_request_cb(blk->dev_opaque, force);
1132     }
1133 }
1134
1135 /*
1136  * Does @blk's attached device model have a tray, and is it open?
1137  */
1138 bool blk_dev_is_tray_open(BlockBackend *blk)
1139 {
1140     IO_CODE();
1141     if (blk_dev_has_tray(blk)) {
1142         return blk->dev_ops->is_tray_open(blk->dev_opaque);
1143     }
1144     return false;
1145 }
1146
1147 /*
1148  * Does @blk's attached device model have the medium locked?
1149  * %false if the device model has no such lock.
1150  */
1151 bool blk_dev_is_medium_locked(BlockBackend *blk)
1152 {
1153     GLOBAL_STATE_CODE();
1154     if (blk->dev_ops && blk->dev_ops->is_medium_locked) {
1155         return blk->dev_ops->is_medium_locked(blk->dev_opaque);
1156     }
1157     return false;
1158 }
1159
1160 /*
1161  * Notify @blk's attached device model of a backend size change.
1162  */
1163 static void blk_root_resize(BdrvChild *child)
1164 {
1165     BlockBackend *blk = child->opaque;
1166
1167     if (blk->dev_ops && blk->dev_ops->resize_cb) {
1168         blk->dev_ops->resize_cb(blk->dev_opaque);
1169     }
1170 }
1171
1172 void blk_iostatus_enable(BlockBackend *blk)
1173 {
1174     GLOBAL_STATE_CODE();
1175     blk->iostatus_enabled = true;
1176     blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
1177 }
1178
1179 /* The I/O status is only enabled if the drive explicitly
1180  * enables it _and_ the VM is configured to stop on errors */
1181 bool blk_iostatus_is_enabled(const BlockBackend *blk)
1182 {
1183     IO_CODE();
1184     return (blk->iostatus_enabled &&
1185            (blk->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
1186             blk->on_write_error == BLOCKDEV_ON_ERROR_STOP   ||
1187             blk->on_read_error == BLOCKDEV_ON_ERROR_STOP));
1188 }
1189
1190 BlockDeviceIoStatus blk_iostatus(const BlockBackend *blk)
1191 {
1192     GLOBAL_STATE_CODE();
1193     return blk->iostatus;
1194 }
1195
1196 void blk_iostatus_disable(BlockBackend *blk)
1197 {
1198     GLOBAL_STATE_CODE();
1199     blk->iostatus_enabled = false;
1200 }
1201
1202 void blk_iostatus_reset(BlockBackend *blk)
1203 {
1204     GLOBAL_STATE_CODE();
1205     if (blk_iostatus_is_enabled(blk)) {
1206         blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
1207     }
1208 }
1209
1210 void blk_iostatus_set_err(BlockBackend *blk, int error)
1211 {
1212     IO_CODE();
1213     assert(blk_iostatus_is_enabled(blk));
1214     if (blk->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
1215         blk->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
1216                                           BLOCK_DEVICE_IO_STATUS_FAILED;
1217     }
1218 }
1219
1220 void blk_set_allow_write_beyond_eof(BlockBackend *blk, bool allow)
1221 {
1222     IO_CODE();
1223     blk->allow_write_beyond_eof = allow;
1224 }
1225
1226 void blk_set_allow_aio_context_change(BlockBackend *blk, bool allow)
1227 {
1228     IO_CODE();
1229     blk->allow_aio_context_change = allow;
1230 }
1231
1232 void blk_set_disable_request_queuing(BlockBackend *blk, bool disable)
1233 {
1234     IO_CODE();
1235     blk->disable_request_queuing = disable;
1236 }
1237
1238 static int coroutine_fn GRAPH_RDLOCK
1239 blk_check_byte_request(BlockBackend *blk, int64_t offset, int64_t bytes)
1240 {
1241     int64_t len;
1242
1243     if (bytes < 0) {
1244         return -EIO;
1245     }
1246
1247     if (!blk_co_is_available(blk)) {
1248         return -ENOMEDIUM;
1249     }
1250
1251     if (offset < 0) {
1252         return -EIO;
1253     }
1254
1255     if (!blk->allow_write_beyond_eof) {
1256         len = bdrv_co_getlength(blk_bs(blk));
1257         if (len < 0) {
1258             return len;
1259         }
1260
1261         if (offset > len || len - offset < bytes) {
1262             return -EIO;
1263         }
1264     }
1265
1266     return 0;
1267 }
1268
1269 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1270 static void coroutine_fn blk_wait_while_drained(BlockBackend *blk)
1271 {
1272     assert(blk->in_flight > 0);
1273
1274     if (blk->quiesce_counter && !blk->disable_request_queuing) {
1275         blk_dec_in_flight(blk);
1276         qemu_co_queue_wait(&blk->queued_requests, NULL);
1277         blk_inc_in_flight(blk);
1278     }
1279 }
1280
1281 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1282 static int coroutine_fn
1283 blk_co_do_preadv_part(BlockBackend *blk, int64_t offset, int64_t bytes,
1284                       QEMUIOVector *qiov, size_t qiov_offset,
1285                       BdrvRequestFlags flags)
1286 {
1287     int ret;
1288     BlockDriverState *bs;
1289     IO_CODE();
1290
1291     blk_wait_while_drained(blk);
1292     GRAPH_RDLOCK_GUARD();
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     GRAPH_RDLOCK_GUARD();
1368
1369     /* Call blk_bs() only after waiting, the graph may have changed */
1370     bs = blk_bs(blk);
1371     trace_blk_co_pwritev(blk, bs, offset, bytes, flags);
1372
1373     ret = blk_check_byte_request(blk, offset, bytes);
1374     if (ret < 0) {
1375         return ret;
1376     }
1377
1378     bdrv_inc_in_flight(bs);
1379     /* throttling disk I/O */
1380     if (blk->public.throttle_group_member.throttle_state) {
1381         throttle_group_co_io_limits_intercept(&blk->public.throttle_group_member,
1382                 bytes, true);
1383     }
1384
1385     if (!blk->enable_write_cache) {
1386         flags |= BDRV_REQ_FUA;
1387     }
1388
1389     ret = bdrv_co_pwritev_part(blk->root, offset, bytes, qiov, qiov_offset,
1390                                flags);
1391     bdrv_dec_in_flight(bs);
1392     return ret;
1393 }
1394
1395 int coroutine_fn blk_co_pwritev_part(BlockBackend *blk, int64_t offset,
1396                                      int64_t bytes,
1397                                      QEMUIOVector *qiov, size_t qiov_offset,
1398                                      BdrvRequestFlags flags)
1399 {
1400     int ret;
1401     IO_OR_GS_CODE();
1402
1403     blk_inc_in_flight(blk);
1404     ret = blk_co_do_pwritev_part(blk, offset, bytes, qiov, qiov_offset, flags);
1405     blk_dec_in_flight(blk);
1406
1407     return ret;
1408 }
1409
1410 int coroutine_fn blk_co_pwrite(BlockBackend *blk, int64_t offset, int64_t bytes,
1411                                const void *buf, BdrvRequestFlags flags)
1412 {
1413     QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
1414     IO_OR_GS_CODE();
1415
1416     assert(bytes <= SIZE_MAX);
1417
1418     return blk_co_pwritev(blk, offset, bytes, &qiov, flags);
1419 }
1420
1421 int coroutine_fn blk_co_pwritev(BlockBackend *blk, int64_t offset,
1422                                 int64_t bytes, QEMUIOVector *qiov,
1423                                 BdrvRequestFlags flags)
1424 {
1425     IO_OR_GS_CODE();
1426     return blk_co_pwritev_part(blk, offset, bytes, qiov, 0, flags);
1427 }
1428
1429 int coroutine_fn blk_co_block_status_above(BlockBackend *blk,
1430                                            BlockDriverState *base,
1431                                            int64_t offset, int64_t bytes,
1432                                            int64_t *pnum, int64_t *map,
1433                                            BlockDriverState **file)
1434 {
1435     IO_CODE();
1436     GRAPH_RDLOCK_GUARD();
1437     return bdrv_co_block_status_above(blk_bs(blk), base, offset, bytes, pnum,
1438                                       map, file);
1439 }
1440
1441 int coroutine_fn blk_co_is_allocated_above(BlockBackend *blk,
1442                                            BlockDriverState *base,
1443                                            bool include_base, int64_t offset,
1444                                            int64_t bytes, int64_t *pnum)
1445 {
1446     IO_CODE();
1447     GRAPH_RDLOCK_GUARD();
1448     return bdrv_co_is_allocated_above(blk_bs(blk), base, include_base, offset,
1449                                       bytes, pnum);
1450 }
1451
1452 typedef struct BlkRwCo {
1453     BlockBackend *blk;
1454     int64_t offset;
1455     void *iobuf;
1456     int ret;
1457     BdrvRequestFlags flags;
1458 } BlkRwCo;
1459
1460 int blk_make_zero(BlockBackend *blk, BdrvRequestFlags flags)
1461 {
1462     GLOBAL_STATE_CODE();
1463     return bdrv_make_zero(blk->root, flags);
1464 }
1465
1466 void blk_inc_in_flight(BlockBackend *blk)
1467 {
1468     IO_CODE();
1469     qatomic_inc(&blk->in_flight);
1470 }
1471
1472 void blk_dec_in_flight(BlockBackend *blk)
1473 {
1474     IO_CODE();
1475     qatomic_dec(&blk->in_flight);
1476     aio_wait_kick();
1477 }
1478
1479 static void error_callback_bh(void *opaque)
1480 {
1481     struct BlockBackendAIOCB *acb = opaque;
1482
1483     blk_dec_in_flight(acb->blk);
1484     acb->common.cb(acb->common.opaque, acb->ret);
1485     qemu_aio_unref(acb);
1486 }
1487
1488 BlockAIOCB *blk_abort_aio_request(BlockBackend *blk,
1489                                   BlockCompletionFunc *cb,
1490                                   void *opaque, int ret)
1491 {
1492     struct BlockBackendAIOCB *acb;
1493     IO_CODE();
1494
1495     blk_inc_in_flight(blk);
1496     acb = blk_aio_get(&block_backend_aiocb_info, blk, cb, opaque);
1497     acb->blk = blk;
1498     acb->ret = ret;
1499
1500     replay_bh_schedule_oneshot_event(blk_get_aio_context(blk),
1501                                      error_callback_bh, acb);
1502     return &acb->common;
1503 }
1504
1505 typedef struct BlkAioEmAIOCB {
1506     BlockAIOCB common;
1507     BlkRwCo rwco;
1508     int64_t bytes;
1509     bool has_returned;
1510 } BlkAioEmAIOCB;
1511
1512 static AioContext *blk_aio_em_aiocb_get_aio_context(BlockAIOCB *acb_)
1513 {
1514     BlkAioEmAIOCB *acb = container_of(acb_, BlkAioEmAIOCB, common);
1515
1516     return blk_get_aio_context(acb->rwco.blk);
1517 }
1518
1519 static const AIOCBInfo blk_aio_em_aiocb_info = {
1520     .aiocb_size         = sizeof(BlkAioEmAIOCB),
1521     .get_aio_context    = blk_aio_em_aiocb_get_aio_context,
1522 };
1523
1524 static void blk_aio_complete(BlkAioEmAIOCB *acb)
1525 {
1526     if (acb->has_returned) {
1527         acb->common.cb(acb->common.opaque, acb->rwco.ret);
1528         blk_dec_in_flight(acb->rwco.blk);
1529         qemu_aio_unref(acb);
1530     }
1531 }
1532
1533 static void blk_aio_complete_bh(void *opaque)
1534 {
1535     BlkAioEmAIOCB *acb = opaque;
1536     assert(acb->has_returned);
1537     blk_aio_complete(acb);
1538 }
1539
1540 static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset,
1541                                 int64_t bytes,
1542                                 void *iobuf, CoroutineEntry co_entry,
1543                                 BdrvRequestFlags flags,
1544                                 BlockCompletionFunc *cb, void *opaque)
1545 {
1546     BlkAioEmAIOCB *acb;
1547     Coroutine *co;
1548
1549     blk_inc_in_flight(blk);
1550     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1551     acb->rwco = (BlkRwCo) {
1552         .blk    = blk,
1553         .offset = offset,
1554         .iobuf  = iobuf,
1555         .flags  = flags,
1556         .ret    = NOT_DONE,
1557     };
1558     acb->bytes = bytes;
1559     acb->has_returned = false;
1560
1561     co = qemu_coroutine_create(co_entry, acb);
1562     aio_co_enter(blk_get_aio_context(blk), co);
1563
1564     acb->has_returned = true;
1565     if (acb->rwco.ret != NOT_DONE) {
1566         replay_bh_schedule_oneshot_event(blk_get_aio_context(blk),
1567                                          blk_aio_complete_bh, acb);
1568     }
1569
1570     return &acb->common;
1571 }
1572
1573 static void coroutine_fn blk_aio_read_entry(void *opaque)
1574 {
1575     BlkAioEmAIOCB *acb = opaque;
1576     BlkRwCo *rwco = &acb->rwco;
1577     QEMUIOVector *qiov = rwco->iobuf;
1578
1579     assert(qiov->size == acb->bytes);
1580     rwco->ret = blk_co_do_preadv_part(rwco->blk, rwco->offset, acb->bytes, qiov,
1581                                       0, rwco->flags);
1582     blk_aio_complete(acb);
1583 }
1584
1585 static void coroutine_fn blk_aio_write_entry(void *opaque)
1586 {
1587     BlkAioEmAIOCB *acb = opaque;
1588     BlkRwCo *rwco = &acb->rwco;
1589     QEMUIOVector *qiov = rwco->iobuf;
1590
1591     assert(!qiov || qiov->size == acb->bytes);
1592     rwco->ret = blk_co_do_pwritev_part(rwco->blk, rwco->offset, acb->bytes,
1593                                        qiov, 0, rwco->flags);
1594     blk_aio_complete(acb);
1595 }
1596
1597 BlockAIOCB *blk_aio_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1598                                   int64_t bytes, BdrvRequestFlags flags,
1599                                   BlockCompletionFunc *cb, void *opaque)
1600 {
1601     IO_CODE();
1602     return blk_aio_prwv(blk, offset, bytes, NULL, blk_aio_write_entry,
1603                         flags | BDRV_REQ_ZERO_WRITE, cb, opaque);
1604 }
1605
1606 int64_t coroutine_fn blk_co_getlength(BlockBackend *blk)
1607 {
1608     IO_CODE();
1609     GRAPH_RDLOCK_GUARD();
1610
1611     if (!blk_co_is_available(blk)) {
1612         return -ENOMEDIUM;
1613     }
1614
1615     return bdrv_co_getlength(blk_bs(blk));
1616 }
1617
1618 void blk_get_geometry(BlockBackend *blk, uint64_t *nb_sectors_ptr)
1619 {
1620     IO_CODE();
1621     if (!blk_bs(blk)) {
1622         *nb_sectors_ptr = 0;
1623     } else {
1624         bdrv_get_geometry(blk_bs(blk), nb_sectors_ptr);
1625     }
1626 }
1627
1628 int64_t coroutine_fn blk_co_nb_sectors(BlockBackend *blk)
1629 {
1630     IO_CODE();
1631     GRAPH_RDLOCK_GUARD();
1632
1633     if (!blk_co_is_available(blk)) {
1634         return -ENOMEDIUM;
1635     }
1636
1637     return bdrv_co_nb_sectors(blk_bs(blk));
1638 }
1639
1640 BlockAIOCB *blk_aio_preadv(BlockBackend *blk, int64_t offset,
1641                            QEMUIOVector *qiov, BdrvRequestFlags flags,
1642                            BlockCompletionFunc *cb, void *opaque)
1643 {
1644     IO_CODE();
1645     assert((uint64_t)qiov->size <= INT64_MAX);
1646     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1647                         blk_aio_read_entry, flags, cb, opaque);
1648 }
1649
1650 BlockAIOCB *blk_aio_pwritev(BlockBackend *blk, int64_t offset,
1651                             QEMUIOVector *qiov, BdrvRequestFlags flags,
1652                             BlockCompletionFunc *cb, void *opaque)
1653 {
1654     IO_CODE();
1655     assert((uint64_t)qiov->size <= INT64_MAX);
1656     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1657                         blk_aio_write_entry, flags, cb, opaque);
1658 }
1659
1660 void blk_aio_cancel(BlockAIOCB *acb)
1661 {
1662     GLOBAL_STATE_CODE();
1663     bdrv_aio_cancel(acb);
1664 }
1665
1666 void blk_aio_cancel_async(BlockAIOCB *acb)
1667 {
1668     IO_CODE();
1669     bdrv_aio_cancel_async(acb);
1670 }
1671
1672 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1673 static int coroutine_fn
1674 blk_co_do_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1675 {
1676     IO_CODE();
1677
1678     blk_wait_while_drained(blk);
1679     GRAPH_RDLOCK_GUARD();
1680
1681     if (!blk_co_is_available(blk)) {
1682         return -ENOMEDIUM;
1683     }
1684
1685     return bdrv_co_ioctl(blk_bs(blk), req, buf);
1686 }
1687
1688 int coroutine_fn blk_co_ioctl(BlockBackend *blk, unsigned long int req,
1689                               void *buf)
1690 {
1691     int ret;
1692     IO_OR_GS_CODE();
1693
1694     blk_inc_in_flight(blk);
1695     ret = blk_co_do_ioctl(blk, req, buf);
1696     blk_dec_in_flight(blk);
1697
1698     return ret;
1699 }
1700
1701 static void coroutine_fn blk_aio_ioctl_entry(void *opaque)
1702 {
1703     BlkAioEmAIOCB *acb = opaque;
1704     BlkRwCo *rwco = &acb->rwco;
1705
1706     rwco->ret = blk_co_do_ioctl(rwco->blk, rwco->offset, rwco->iobuf);
1707
1708     blk_aio_complete(acb);
1709 }
1710
1711 BlockAIOCB *blk_aio_ioctl(BlockBackend *blk, unsigned long int req, void *buf,
1712                           BlockCompletionFunc *cb, void *opaque)
1713 {
1714     IO_CODE();
1715     return blk_aio_prwv(blk, req, 0, buf, blk_aio_ioctl_entry, 0, cb, opaque);
1716 }
1717
1718 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1719 static int coroutine_fn
1720 blk_co_do_pdiscard(BlockBackend *blk, int64_t offset, int64_t bytes)
1721 {
1722     int ret;
1723     IO_CODE();
1724
1725     blk_wait_while_drained(blk);
1726     GRAPH_RDLOCK_GUARD();
1727
1728     ret = blk_check_byte_request(blk, offset, bytes);
1729     if (ret < 0) {
1730         return ret;
1731     }
1732
1733     return bdrv_co_pdiscard(blk->root, offset, bytes);
1734 }
1735
1736 static void coroutine_fn blk_aio_pdiscard_entry(void *opaque)
1737 {
1738     BlkAioEmAIOCB *acb = opaque;
1739     BlkRwCo *rwco = &acb->rwco;
1740
1741     rwco->ret = blk_co_do_pdiscard(rwco->blk, rwco->offset, acb->bytes);
1742     blk_aio_complete(acb);
1743 }
1744
1745 BlockAIOCB *blk_aio_pdiscard(BlockBackend *blk,
1746                              int64_t offset, int64_t bytes,
1747                              BlockCompletionFunc *cb, void *opaque)
1748 {
1749     IO_CODE();
1750     return blk_aio_prwv(blk, offset, bytes, NULL, blk_aio_pdiscard_entry, 0,
1751                         cb, opaque);
1752 }
1753
1754 int coroutine_fn blk_co_pdiscard(BlockBackend *blk, int64_t offset,
1755                                  int64_t bytes)
1756 {
1757     int ret;
1758     IO_OR_GS_CODE();
1759
1760     blk_inc_in_flight(blk);
1761     ret = blk_co_do_pdiscard(blk, offset, bytes);
1762     blk_dec_in_flight(blk);
1763
1764     return ret;
1765 }
1766
1767 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1768 static int coroutine_fn blk_co_do_flush(BlockBackend *blk)
1769 {
1770     IO_CODE();
1771     blk_wait_while_drained(blk);
1772     GRAPH_RDLOCK_GUARD();
1773
1774     if (!blk_co_is_available(blk)) {
1775         return -ENOMEDIUM;
1776     }
1777
1778     return bdrv_co_flush(blk_bs(blk));
1779 }
1780
1781 static void coroutine_fn blk_aio_flush_entry(void *opaque)
1782 {
1783     BlkAioEmAIOCB *acb = opaque;
1784     BlkRwCo *rwco = &acb->rwco;
1785
1786     rwco->ret = blk_co_do_flush(rwco->blk);
1787     blk_aio_complete(acb);
1788 }
1789
1790 BlockAIOCB *blk_aio_flush(BlockBackend *blk,
1791                           BlockCompletionFunc *cb, void *opaque)
1792 {
1793     IO_CODE();
1794     return blk_aio_prwv(blk, 0, 0, NULL, blk_aio_flush_entry, 0, cb, opaque);
1795 }
1796
1797 int coroutine_fn blk_co_flush(BlockBackend *blk)
1798 {
1799     int ret;
1800     IO_OR_GS_CODE();
1801
1802     blk_inc_in_flight(blk);
1803     ret = blk_co_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),
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     IO_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 coroutine_fn blk_co_is_inserted(BlockBackend *blk)
1998 {
1999     BlockDriverState *bs = blk_bs(blk);
2000     IO_CODE();
2001     assert_bdrv_graph_readable();
2002
2003     return bs && bdrv_co_is_inserted(bs);
2004 }
2005
2006 bool coroutine_fn blk_co_is_available(BlockBackend *blk)
2007 {
2008     IO_CODE();
2009     return blk_co_is_inserted(blk) && !blk_dev_is_tray_open(blk);
2010 }
2011
2012 void coroutine_fn blk_co_lock_medium(BlockBackend *blk, bool locked)
2013 {
2014     BlockDriverState *bs = blk_bs(blk);
2015     IO_CODE();
2016     GRAPH_RDLOCK_GUARD();
2017
2018     if (bs) {
2019         bdrv_co_lock_medium(bs, locked);
2020     }
2021 }
2022
2023 void coroutine_fn blk_co_eject(BlockBackend *blk, bool eject_flag)
2024 {
2025     BlockDriverState *bs = blk_bs(blk);
2026     char *id;
2027     IO_CODE();
2028     GRAPH_RDLOCK_GUARD();
2029
2030     if (bs) {
2031         bdrv_co_eject(bs, eject_flag);
2032     }
2033
2034     /* Whether or not we ejected on the backend,
2035      * the frontend experienced a tray event. */
2036     id = blk_get_attached_dev_id(blk);
2037     qapi_event_send_device_tray_moved(blk_name(blk), id,
2038                                       eject_flag);
2039     g_free(id);
2040 }
2041
2042 int blk_get_flags(BlockBackend *blk)
2043 {
2044     BlockDriverState *bs = blk_bs(blk);
2045     GLOBAL_STATE_CODE();
2046
2047     if (bs) {
2048         return bdrv_get_flags(bs);
2049     } else {
2050         return blk->root_state.open_flags;
2051     }
2052 }
2053
2054 /* Returns the minimum request alignment, in bytes; guaranteed nonzero */
2055 uint32_t blk_get_request_alignment(BlockBackend *blk)
2056 {
2057     BlockDriverState *bs = blk_bs(blk);
2058     IO_CODE();
2059     return bs ? bs->bl.request_alignment : BDRV_SECTOR_SIZE;
2060 }
2061
2062 /* Returns the maximum hardware transfer length, in bytes; guaranteed nonzero */
2063 uint64_t blk_get_max_hw_transfer(BlockBackend *blk)
2064 {
2065     BlockDriverState *bs = blk_bs(blk);
2066     uint64_t max = INT_MAX;
2067     IO_CODE();
2068
2069     if (bs) {
2070         max = MIN_NON_ZERO(max, bs->bl.max_hw_transfer);
2071         max = MIN_NON_ZERO(max, bs->bl.max_transfer);
2072     }
2073     return ROUND_DOWN(max, blk_get_request_alignment(blk));
2074 }
2075
2076 /* Returns the maximum transfer length, in bytes; guaranteed nonzero */
2077 uint32_t blk_get_max_transfer(BlockBackend *blk)
2078 {
2079     BlockDriverState *bs = blk_bs(blk);
2080     uint32_t max = INT_MAX;
2081     IO_CODE();
2082
2083     if (bs) {
2084         max = MIN_NON_ZERO(max, bs->bl.max_transfer);
2085     }
2086     return ROUND_DOWN(max, blk_get_request_alignment(blk));
2087 }
2088
2089 int blk_get_max_hw_iov(BlockBackend *blk)
2090 {
2091     IO_CODE();
2092     return MIN_NON_ZERO(blk->root->bs->bl.max_hw_iov,
2093                         blk->root->bs->bl.max_iov);
2094 }
2095
2096 int blk_get_max_iov(BlockBackend *blk)
2097 {
2098     IO_CODE();
2099     return blk->root->bs->bl.max_iov;
2100 }
2101
2102 void *blk_try_blockalign(BlockBackend *blk, size_t size)
2103 {
2104     IO_CODE();
2105     return qemu_try_blockalign(blk ? blk_bs(blk) : NULL, size);
2106 }
2107
2108 void *blk_blockalign(BlockBackend *blk, size_t size)
2109 {
2110     IO_CODE();
2111     return qemu_blockalign(blk ? blk_bs(blk) : NULL, size);
2112 }
2113
2114 bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp)
2115 {
2116     BlockDriverState *bs = blk_bs(blk);
2117     GLOBAL_STATE_CODE();
2118
2119     if (!bs) {
2120         return false;
2121     }
2122
2123     return bdrv_op_is_blocked(bs, op, errp);
2124 }
2125
2126 void blk_op_unblock(BlockBackend *blk, BlockOpType op, Error *reason)
2127 {
2128     BlockDriverState *bs = blk_bs(blk);
2129     GLOBAL_STATE_CODE();
2130
2131     if (bs) {
2132         bdrv_op_unblock(bs, op, reason);
2133     }
2134 }
2135
2136 void blk_op_block_all(BlockBackend *blk, Error *reason)
2137 {
2138     BlockDriverState *bs = blk_bs(blk);
2139     GLOBAL_STATE_CODE();
2140
2141     if (bs) {
2142         bdrv_op_block_all(bs, reason);
2143     }
2144 }
2145
2146 void blk_op_unblock_all(BlockBackend *blk, Error *reason)
2147 {
2148     BlockDriverState *bs = blk_bs(blk);
2149     GLOBAL_STATE_CODE();
2150
2151     if (bs) {
2152         bdrv_op_unblock_all(bs, reason);
2153     }
2154 }
2155
2156 AioContext *blk_get_aio_context(BlockBackend *blk)
2157 {
2158     BlockDriverState *bs = blk_bs(blk);
2159     IO_CODE();
2160
2161     if (bs) {
2162         AioContext *ctx = bdrv_get_aio_context(blk_bs(blk));
2163         assert(ctx == blk->ctx);
2164     }
2165
2166     return blk->ctx;
2167 }
2168
2169 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb)
2170 {
2171     BlockBackendAIOCB *blk_acb = DO_UPCAST(BlockBackendAIOCB, common, acb);
2172     return blk_get_aio_context(blk_acb->blk);
2173 }
2174
2175 static int blk_do_set_aio_context(BlockBackend *blk, AioContext *new_context,
2176                                   bool update_root_node, Error **errp)
2177 {
2178     BlockDriverState *bs = blk_bs(blk);
2179     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
2180     int ret;
2181
2182     if (bs) {
2183         bdrv_ref(bs);
2184
2185         if (update_root_node) {
2186             /*
2187              * update_root_node MUST be false for blk_root_set_aio_ctx_commit(),
2188              * as we are already in the commit function of a transaction.
2189              */
2190             ret = bdrv_try_change_aio_context(bs, new_context, blk->root, errp);
2191             if (ret < 0) {
2192                 bdrv_unref(bs);
2193                 return ret;
2194             }
2195         }
2196         /*
2197          * Make blk->ctx consistent with the root node before we invoke any
2198          * other operations like drain that might inquire blk->ctx
2199          */
2200         blk->ctx = new_context;
2201         if (tgm->throttle_state) {
2202             bdrv_drained_begin(bs);
2203             throttle_group_detach_aio_context(tgm);
2204             throttle_group_attach_aio_context(tgm, new_context);
2205             bdrv_drained_end(bs);
2206         }
2207
2208         bdrv_unref(bs);
2209     } else {
2210         blk->ctx = new_context;
2211     }
2212
2213     return 0;
2214 }
2215
2216 int blk_set_aio_context(BlockBackend *blk, AioContext *new_context,
2217                         Error **errp)
2218 {
2219     GLOBAL_STATE_CODE();
2220     return blk_do_set_aio_context(blk, new_context, true, errp);
2221 }
2222
2223 typedef struct BdrvStateBlkRootContext {
2224     AioContext *new_ctx;
2225     BlockBackend *blk;
2226 } BdrvStateBlkRootContext;
2227
2228 static void blk_root_set_aio_ctx_commit(void *opaque)
2229 {
2230     BdrvStateBlkRootContext *s = opaque;
2231     BlockBackend *blk = s->blk;
2232
2233     blk_do_set_aio_context(blk, s->new_ctx, false, &error_abort);
2234 }
2235
2236 static TransactionActionDrv set_blk_root_context = {
2237     .commit = blk_root_set_aio_ctx_commit,
2238     .clean = g_free,
2239 };
2240
2241 static bool blk_root_change_aio_ctx(BdrvChild *child, AioContext *ctx,
2242                                     GHashTable *visited, Transaction *tran,
2243                                     Error **errp)
2244 {
2245     BlockBackend *blk = child->opaque;
2246     BdrvStateBlkRootContext *s;
2247
2248     if (!blk->allow_aio_context_change) {
2249         /*
2250          * Manually created BlockBackends (those with a name) that are not
2251          * attached to anything can change their AioContext without updating
2252          * their user; return an error for others.
2253          */
2254         if (!blk->name || blk->dev) {
2255             /* TODO Add BB name/QOM path */
2256             error_setg(errp, "Cannot change iothread of active block backend");
2257             return false;
2258         }
2259     }
2260
2261     s = g_new(BdrvStateBlkRootContext, 1);
2262     *s = (BdrvStateBlkRootContext) {
2263         .new_ctx = ctx,
2264         .blk = blk,
2265     };
2266
2267     tran_add(tran, &set_blk_root_context, s);
2268     return true;
2269 }
2270
2271 void blk_add_aio_context_notifier(BlockBackend *blk,
2272         void (*attached_aio_context)(AioContext *new_context, void *opaque),
2273         void (*detach_aio_context)(void *opaque), void *opaque)
2274 {
2275     BlockBackendAioNotifier *notifier;
2276     BlockDriverState *bs = blk_bs(blk);
2277     GLOBAL_STATE_CODE();
2278
2279     notifier = g_new(BlockBackendAioNotifier, 1);
2280     notifier->attached_aio_context = attached_aio_context;
2281     notifier->detach_aio_context = detach_aio_context;
2282     notifier->opaque = opaque;
2283     QLIST_INSERT_HEAD(&blk->aio_notifiers, notifier, list);
2284
2285     if (bs) {
2286         bdrv_add_aio_context_notifier(bs, attached_aio_context,
2287                                       detach_aio_context, opaque);
2288     }
2289 }
2290
2291 void blk_remove_aio_context_notifier(BlockBackend *blk,
2292                                      void (*attached_aio_context)(AioContext *,
2293                                                                   void *),
2294                                      void (*detach_aio_context)(void *),
2295                                      void *opaque)
2296 {
2297     BlockBackendAioNotifier *notifier;
2298     BlockDriverState *bs = blk_bs(blk);
2299
2300     GLOBAL_STATE_CODE();
2301
2302     if (bs) {
2303         bdrv_remove_aio_context_notifier(bs, attached_aio_context,
2304                                          detach_aio_context, opaque);
2305     }
2306
2307     QLIST_FOREACH(notifier, &blk->aio_notifiers, list) {
2308         if (notifier->attached_aio_context == attached_aio_context &&
2309             notifier->detach_aio_context == detach_aio_context &&
2310             notifier->opaque == opaque) {
2311             QLIST_REMOVE(notifier, list);
2312             g_free(notifier);
2313             return;
2314         }
2315     }
2316
2317     abort();
2318 }
2319
2320 void blk_add_remove_bs_notifier(BlockBackend *blk, Notifier *notify)
2321 {
2322     GLOBAL_STATE_CODE();
2323     notifier_list_add(&blk->remove_bs_notifiers, notify);
2324 }
2325
2326 void blk_add_insert_bs_notifier(BlockBackend *blk, Notifier *notify)
2327 {
2328     GLOBAL_STATE_CODE();
2329     notifier_list_add(&blk->insert_bs_notifiers, notify);
2330 }
2331
2332 void coroutine_fn blk_co_io_plug(BlockBackend *blk)
2333 {
2334     BlockDriverState *bs = blk_bs(blk);
2335     IO_CODE();
2336     GRAPH_RDLOCK_GUARD();
2337
2338     if (bs) {
2339         bdrv_co_io_plug(bs);
2340     }
2341 }
2342
2343 void coroutine_fn blk_co_io_unplug(BlockBackend *blk)
2344 {
2345     BlockDriverState *bs = blk_bs(blk);
2346     IO_CODE();
2347     GRAPH_RDLOCK_GUARD();
2348
2349     if (bs) {
2350         bdrv_co_io_unplug(bs);
2351     }
2352 }
2353
2354 BlockAcctStats *blk_get_stats(BlockBackend *blk)
2355 {
2356     IO_CODE();
2357     return &blk->stats;
2358 }
2359
2360 void *blk_aio_get(const AIOCBInfo *aiocb_info, BlockBackend *blk,
2361                   BlockCompletionFunc *cb, void *opaque)
2362 {
2363     IO_CODE();
2364     return qemu_aio_get(aiocb_info, blk_bs(blk), cb, opaque);
2365 }
2366
2367 int coroutine_fn blk_co_pwrite_zeroes(BlockBackend *blk, int64_t offset,
2368                                       int64_t bytes, BdrvRequestFlags flags)
2369 {
2370     IO_OR_GS_CODE();
2371     return blk_co_pwritev(blk, offset, bytes, NULL,
2372                           flags | BDRV_REQ_ZERO_WRITE);
2373 }
2374
2375 int coroutine_fn blk_co_pwrite_compressed(BlockBackend *blk, int64_t offset,
2376                                           int64_t bytes, const void *buf)
2377 {
2378     QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
2379     IO_OR_GS_CODE();
2380     return blk_co_pwritev_part(blk, offset, bytes, &qiov, 0,
2381                                BDRV_REQ_WRITE_COMPRESSED);
2382 }
2383
2384 int coroutine_fn blk_co_truncate(BlockBackend *blk, int64_t offset, bool exact,
2385                                  PreallocMode prealloc, BdrvRequestFlags flags,
2386                                  Error **errp)
2387 {
2388     IO_OR_GS_CODE();
2389     GRAPH_RDLOCK_GUARD();
2390     if (!blk_co_is_available(blk)) {
2391         error_setg(errp, "No medium inserted");
2392         return -ENOMEDIUM;
2393     }
2394
2395     return bdrv_co_truncate(blk->root, offset, exact, prealloc, flags, errp);
2396 }
2397
2398 int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
2399                      int64_t pos, int size)
2400 {
2401     int ret;
2402     GLOBAL_STATE_CODE();
2403
2404     if (!blk_is_available(blk)) {
2405         return -ENOMEDIUM;
2406     }
2407
2408     ret = bdrv_save_vmstate(blk_bs(blk), buf, pos, size);
2409     if (ret < 0) {
2410         return ret;
2411     }
2412
2413     if (ret == size && !blk->enable_write_cache) {
2414         ret = bdrv_flush(blk_bs(blk));
2415     }
2416
2417     return ret < 0 ? ret : size;
2418 }
2419
2420 int blk_load_vmstate(BlockBackend *blk, uint8_t *buf, int64_t pos, int size)
2421 {
2422     GLOBAL_STATE_CODE();
2423     if (!blk_is_available(blk)) {
2424         return -ENOMEDIUM;
2425     }
2426
2427     return bdrv_load_vmstate(blk_bs(blk), buf, pos, size);
2428 }
2429
2430 int blk_probe_blocksizes(BlockBackend *blk, BlockSizes *bsz)
2431 {
2432     GLOBAL_STATE_CODE();
2433     if (!blk_is_available(blk)) {
2434         return -ENOMEDIUM;
2435     }
2436
2437     return bdrv_probe_blocksizes(blk_bs(blk), bsz);
2438 }
2439
2440 int blk_probe_geometry(BlockBackend *blk, HDGeometry *geo)
2441 {
2442     GLOBAL_STATE_CODE();
2443     if (!blk_is_available(blk)) {
2444         return -ENOMEDIUM;
2445     }
2446
2447     return bdrv_probe_geometry(blk_bs(blk), geo);
2448 }
2449
2450 /*
2451  * Updates the BlockBackendRootState object with data from the currently
2452  * attached BlockDriverState.
2453  */
2454 void blk_update_root_state(BlockBackend *blk)
2455 {
2456     GLOBAL_STATE_CODE();
2457     assert(blk->root);
2458
2459     blk->root_state.open_flags    = blk->root->bs->open_flags;
2460     blk->root_state.detect_zeroes = blk->root->bs->detect_zeroes;
2461 }
2462
2463 /*
2464  * Returns the detect-zeroes setting to be used for bdrv_open() of a
2465  * BlockDriverState which is supposed to inherit the root state.
2466  */
2467 bool blk_get_detect_zeroes_from_root_state(BlockBackend *blk)
2468 {
2469     GLOBAL_STATE_CODE();
2470     return blk->root_state.detect_zeroes;
2471 }
2472
2473 /*
2474  * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
2475  * supposed to inherit the root state.
2476  */
2477 int blk_get_open_flags_from_root_state(BlockBackend *blk)
2478 {
2479     GLOBAL_STATE_CODE();
2480     return blk->root_state.open_flags;
2481 }
2482
2483 BlockBackendRootState *blk_get_root_state(BlockBackend *blk)
2484 {
2485     GLOBAL_STATE_CODE();
2486     return &blk->root_state;
2487 }
2488
2489 int blk_commit_all(void)
2490 {
2491     BlockBackend *blk = NULL;
2492     GLOBAL_STATE_CODE();
2493
2494     while ((blk = blk_all_next(blk)) != NULL) {
2495         AioContext *aio_context = blk_get_aio_context(blk);
2496         BlockDriverState *unfiltered_bs = bdrv_skip_filters(blk_bs(blk));
2497
2498         aio_context_acquire(aio_context);
2499         if (blk_is_inserted(blk) && bdrv_cow_child(unfiltered_bs)) {
2500             int ret;
2501
2502             ret = bdrv_commit(unfiltered_bs);
2503             if (ret < 0) {
2504                 aio_context_release(aio_context);
2505                 return ret;
2506             }
2507         }
2508         aio_context_release(aio_context);
2509     }
2510     return 0;
2511 }
2512
2513
2514 /* throttling disk I/O limits */
2515 void blk_set_io_limits(BlockBackend *blk, ThrottleConfig *cfg)
2516 {
2517     GLOBAL_STATE_CODE();
2518     throttle_group_config(&blk->public.throttle_group_member, cfg);
2519 }
2520
2521 void blk_io_limits_disable(BlockBackend *blk)
2522 {
2523     BlockDriverState *bs = blk_bs(blk);
2524     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
2525     assert(tgm->throttle_state);
2526     GLOBAL_STATE_CODE();
2527     if (bs) {
2528         bdrv_ref(bs);
2529         bdrv_drained_begin(bs);
2530     }
2531     throttle_group_unregister_tgm(tgm);
2532     if (bs) {
2533         bdrv_drained_end(bs);
2534         bdrv_unref(bs);
2535     }
2536 }
2537
2538 /* should be called before blk_set_io_limits if a limit is set */
2539 void blk_io_limits_enable(BlockBackend *blk, const char *group)
2540 {
2541     assert(!blk->public.throttle_group_member.throttle_state);
2542     GLOBAL_STATE_CODE();
2543     throttle_group_register_tgm(&blk->public.throttle_group_member,
2544                                 group, blk_get_aio_context(blk));
2545 }
2546
2547 void blk_io_limits_update_group(BlockBackend *blk, const char *group)
2548 {
2549     GLOBAL_STATE_CODE();
2550     /* this BB is not part of any group */
2551     if (!blk->public.throttle_group_member.throttle_state) {
2552         return;
2553     }
2554
2555     /* this BB is a part of the same group than the one we want */
2556     if (!g_strcmp0(throttle_group_get_name(&blk->public.throttle_group_member),
2557                 group)) {
2558         return;
2559     }
2560
2561     /* need to change the group this bs belong to */
2562     blk_io_limits_disable(blk);
2563     blk_io_limits_enable(blk, group);
2564 }
2565
2566 static void blk_root_drained_begin(BdrvChild *child)
2567 {
2568     BlockBackend *blk = child->opaque;
2569     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
2570
2571     if (++blk->quiesce_counter == 1) {
2572         if (blk->dev_ops && blk->dev_ops->drained_begin) {
2573             blk->dev_ops->drained_begin(blk->dev_opaque);
2574         }
2575     }
2576
2577     /* Note that blk->root may not be accessible here yet if we are just
2578      * attaching to a BlockDriverState that is drained. Use child instead. */
2579
2580     if (qatomic_fetch_inc(&tgm->io_limits_disabled) == 0) {
2581         throttle_group_restart_tgm(tgm);
2582     }
2583 }
2584
2585 static bool blk_root_drained_poll(BdrvChild *child)
2586 {
2587     BlockBackend *blk = child->opaque;
2588     bool busy = false;
2589     assert(blk->quiesce_counter);
2590
2591     if (blk->dev_ops && blk->dev_ops->drained_poll) {
2592         busy = blk->dev_ops->drained_poll(blk->dev_opaque);
2593     }
2594     return busy || !!blk->in_flight;
2595 }
2596
2597 static void blk_root_drained_end(BdrvChild *child)
2598 {
2599     BlockBackend *blk = child->opaque;
2600     assert(blk->quiesce_counter);
2601
2602     assert(blk->public.throttle_group_member.io_limits_disabled);
2603     qatomic_dec(&blk->public.throttle_group_member.io_limits_disabled);
2604
2605     if (--blk->quiesce_counter == 0) {
2606         if (blk->dev_ops && blk->dev_ops->drained_end) {
2607             blk->dev_ops->drained_end(blk->dev_opaque);
2608         }
2609         while (qemu_co_enter_next(&blk->queued_requests, NULL)) {
2610             /* Resume all queued requests */
2611         }
2612     }
2613 }
2614
2615 bool blk_register_buf(BlockBackend *blk, void *host, size_t size, Error **errp)
2616 {
2617     BlockDriverState *bs = blk_bs(blk);
2618
2619     GLOBAL_STATE_CODE();
2620
2621     if (bs) {
2622         return bdrv_register_buf(bs, host, size, errp);
2623     }
2624     return true;
2625 }
2626
2627 void blk_unregister_buf(BlockBackend *blk, void *host, size_t size)
2628 {
2629     BlockDriverState *bs = blk_bs(blk);
2630
2631     GLOBAL_STATE_CODE();
2632
2633     if (bs) {
2634         bdrv_unregister_buf(bs, host, size);
2635     }
2636 }
2637
2638 int coroutine_fn blk_co_copy_range(BlockBackend *blk_in, int64_t off_in,
2639                                    BlockBackend *blk_out, int64_t off_out,
2640                                    int64_t bytes, BdrvRequestFlags read_flags,
2641                                    BdrvRequestFlags write_flags)
2642 {
2643     int r;
2644     IO_CODE();
2645     GRAPH_RDLOCK_GUARD();
2646
2647     r = blk_check_byte_request(blk_in, off_in, bytes);
2648     if (r) {
2649         return r;
2650     }
2651     r = blk_check_byte_request(blk_out, off_out, bytes);
2652     if (r) {
2653         return r;
2654     }
2655
2656     return bdrv_co_copy_range(blk_in->root, off_in,
2657                               blk_out->root, off_out,
2658                               bytes, read_flags, write_flags);
2659 }
2660
2661 const BdrvChild *blk_root(BlockBackend *blk)
2662 {
2663     GLOBAL_STATE_CODE();
2664     return blk->root;
2665 }
2666
2667 int blk_make_empty(BlockBackend *blk, Error **errp)
2668 {
2669     GLOBAL_STATE_CODE();
2670     if (!blk_is_available(blk)) {
2671         error_setg(errp, "No medium inserted");
2672         return -ENOMEDIUM;
2673     }
2674
2675     return bdrv_make_empty(blk->root, errp);
2676 }