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_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
 315     return blk_get_aio_context(blk);
 316 }
 317
 318 static const BdrvChildClass child_root = {
 319     .inherit_options    = blk_root_inherit_options,
 320
 321     .change_media       = blk_root_change_media,
 322     .resize             = blk_root_resize,
 323     .get_name           = blk_root_get_name,
 324     .get_parent_desc    = blk_root_get_parent_desc,
 325
 326     .drained_begin      = blk_root_drained_begin,
 327     .drained_poll       = blk_root_drained_poll,
 328     .drained_end        = blk_root_drained_end,
 329
 330     .activate           = blk_root_activate,
 331     .inactivate         = blk_root_inactivate,
 332
 333     .attach             = blk_root_attach,
 334     .detach             = blk_root_detach,
 335
 336     .change_aio_ctx     = blk_root_change_aio_ctx,
 337
 338     .get_parent_aio_context = blk_root_get_parent_aio_context,
 339 };
 340
 341 /*
 342  * Create a new BlockBackend with a reference count of one.
 343  *
 344  * @perm is a bitmasks of BLK_PERM_* constants which describes the permissions
 345  * to request for a block driver node that is attached to this BlockBackend.
 346  * @shared_perm is a bitmask which describes which permissions may be granted
 347  * to other users of the attached node.
 348  * Both sets of permissions can be changed later using blk_set_perm().
 349  *
 350  * Return the new BlockBackend on success, null on failure.
 351  */
 352 BlockBackend *blk_new(AioContext *ctx, uint64_t perm, uint64_t shared_perm)
 353 {
 354     BlockBackend *blk;
 355
 356     GLOBAL_STATE_CODE();
 357
 358     blk = g_new0(BlockBackend, 1);
 359     blk->refcnt = 1;
 360     blk->ctx = ctx;
 361     blk->perm = perm;
 362     blk->shared_perm = shared_perm;
 363     blk_set_enable_write_cache(blk, true);
 364
 365     blk->on_read_error = BLOCKDEV_ON_ERROR_REPORT;
 366     blk->on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
 367
 368     block_acct_init(&blk->stats);
 369
 370     qemu_co_queue_init(&blk->queued_requests);
 371     notifier_list_init(&blk->remove_bs_notifiers);
 372     notifier_list_init(&blk->insert_bs_notifiers);
 373     QLIST_INIT(&blk->aio_notifiers);
 374
 375     QTAILQ_INSERT_TAIL(&block_backends, blk, link);
 376     return blk;
 377 }
 378
 379 /*
 380  * Create a new BlockBackend connected to an existing BlockDriverState.
 381  *
 382  * @perm is a bitmasks of BLK_PERM_* constants which describes the
 383  * permissions to request for @bs that is attached to this
 384  * BlockBackend.  @shared_perm is a bitmask which describes which
 385  * permissions may be granted to other users of the attached node.
 386  * Both sets of permissions can be changed later using blk_set_perm().
 387  *
 388  * Return the new BlockBackend on success, null on failure.
 389  */
 390 BlockBackend *blk_new_with_bs(BlockDriverState *bs, uint64_t perm,
 391                               uint64_t shared_perm, Error **errp)
 392 {
 393     BlockBackend *blk = blk_new(bdrv_get_aio_context(bs), perm, shared_perm);
 394
 395     GLOBAL_STATE_CODE();
 396
 397     if (blk_insert_bs(blk, bs, errp) < 0) {
 398         blk_unref(blk);
 399         return NULL;
 400     }
 401     return blk;
 402 }
 403
 404 /*
 405  * Creates a new BlockBackend, opens a new BlockDriverState, and connects both.
 406  * The new BlockBackend is in the main AioContext.
 407  *
 408  * Just as with bdrv_open(), after having called this function the reference to
 409  * @options belongs to the block layer (even on failure).
 410  *
 411  * TODO: Remove @filename and @flags; it should be possible to specify a whole
 412  * BDS tree just by specifying the @options QDict (or @reference,
 413  * alternatively). At the time of adding this function, this is not possible,
 414  * though, so callers of this function have to be able to specify @filename and
 415  * @flags.
 416  */
 417 BlockBackend *blk_new_open(const char *filename, const char *reference,
 418                            QDict *options, int flags, Error **errp)
 419 {
 420     BlockBackend *blk;
 421     BlockDriverState *bs;
 422     uint64_t perm = 0;
 423     uint64_t shared = BLK_PERM_ALL;
 424
 425     GLOBAL_STATE_CODE();
 426
 427     /*
 428      * blk_new_open() is mainly used in .bdrv_create implementations and the
 429      * tools where sharing isn't a major concern because the BDS stays private
 430      * and the file is generally not supposed to be used by a second process,
 431      * so we just request permission according to the flags.
 432      *
 433      * The exceptions are xen_disk and blockdev_init(); in these cases, the
 434      * caller of blk_new_open() doesn't make use of the permissions, but they
 435      * shouldn't hurt either. We can still share everything here because the
 436      * guest devices will add their own blockers if they can't share.
 437      */
 438     if ((flags & BDRV_O_NO_IO) == 0) {
 439         perm |= BLK_PERM_CONSISTENT_READ;
 440         if (flags & BDRV_O_RDWR) {
 441             perm |= BLK_PERM_WRITE;
 442         }
 443     }
 444     if (flags & BDRV_O_RESIZE) {
 445         perm |= BLK_PERM_RESIZE;
 446     }
 447     if (flags & BDRV_O_NO_SHARE) {
 448         shared = BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
 449     }
 450
 451     blk = blk_new(qemu_get_aio_context(), perm, shared);
 452     bs = bdrv_open(filename, reference, options, flags, errp);
 453     if (!bs) {
 454         blk_unref(blk);
 455         return NULL;
 456     }
 457
 458     blk->root = bdrv_root_attach_child(bs, "root", &child_root,
 459                                        BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY,
 460                                        perm, shared, blk, errp);
 461     if (!blk->root) {
 462         blk_unref(blk);
 463         return NULL;
 464     }
 465
 466     return blk;
 467 }
 468
 469 static void blk_delete(BlockBackend *blk)
 470 {
 471     assert(!blk->refcnt);
 472     assert(!blk->name);
 473     assert(!blk->dev);
 474     if (blk->public.throttle_group_member.throttle_state) {
 475         blk_io_limits_disable(blk);
 476     }
 477     if (blk->root) {
 478         blk_remove_bs(blk);
 479     }
 480     if (blk->vmsh) {
 481         qemu_del_vm_change_state_handler(blk->vmsh);
 482         blk->vmsh = NULL;
 483     }
 484     assert(QLIST_EMPTY(&blk->remove_bs_notifiers.notifiers));
 485     assert(QLIST_EMPTY(&blk->insert_bs_notifiers.notifiers));
 486     assert(QLIST_EMPTY(&blk->aio_notifiers));
 487     QTAILQ_REMOVE(&block_backends, blk, link);
 488     drive_info_del(blk->legacy_dinfo);
 489     block_acct_cleanup(&blk->stats);
 490     g_free(blk);
 491 }
 492
 493 static void drive_info_del(DriveInfo *dinfo)
 494 {
 495     if (!dinfo) {
 496         return;
 497     }
 498     qemu_opts_del(dinfo->opts);
 499     g_free(dinfo);
 500 }
 501
 502 int blk_get_refcnt(BlockBackend *blk)
 503 {
 504     GLOBAL_STATE_CODE();
 505     return blk ? blk->refcnt : 0;
 506 }
 507
 508 /*
 509  * Increment @blk's reference count.
 510  * @blk must not be null.
 511  */
 512 void blk_ref(BlockBackend *blk)
 513 {
 514     assert(blk->refcnt > 0);
 515     GLOBAL_STATE_CODE();
 516     blk->refcnt++;
 517 }
 518
 519 /*
 520  * Decrement @blk's reference count.
 521  * If this drops it to zero, destroy @blk.
 522  * For convenience, do nothing if @blk is null.
 523  */
 524 void blk_unref(BlockBackend *blk)
 525 {
 526     GLOBAL_STATE_CODE();
 527     if (blk) {
 528         assert(blk->refcnt > 0);
 529         if (blk->refcnt > 1) {
 530             blk->refcnt--;
 531         } else {
 532             blk_drain(blk);
 533             /* blk_drain() cannot resurrect blk, nobody held a reference */
 534             assert(blk->refcnt == 1);
 535             blk->refcnt = 0;
 536             blk_delete(blk);
 537         }
 538     }
 539 }
 540
 541 /*
 542  * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
 543  * ones which are hidden (i.e. are not referenced by the monitor).
 544  */
 545 BlockBackend *blk_all_next(BlockBackend *blk)
 546 {
 547     GLOBAL_STATE_CODE();
 548     return blk ? QTAILQ_NEXT(blk, link)
 549                : QTAILQ_FIRST(&block_backends);
 550 }
 551
 552 void blk_remove_all_bs(void)
 553 {
 554     BlockBackend *blk = NULL;
 555
 556     GLOBAL_STATE_CODE();
 557
 558     while ((blk = blk_all_next(blk)) != NULL) {
 559         AioContext *ctx = blk_get_aio_context(blk);
 560
 561         aio_context_acquire(ctx);
 562         if (blk->root) {
 563             blk_remove_bs(blk);
 564         }
 565         aio_context_release(ctx);
 566     }
 567 }
 568
 569 /*
 570  * Return the monitor-owned BlockBackend after @blk.
 571  * If @blk is null, return the first one.
 572  * Else, return @blk's next sibling, which may be null.
 573  *
 574  * To iterate over all BlockBackends, do
 575  * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
 576  *     ...
 577  * }
 578  */
 579 BlockBackend *blk_next(BlockBackend *blk)
 580 {
 581     GLOBAL_STATE_CODE();
 582     return blk ? QTAILQ_NEXT(blk, monitor_link)
 583                : QTAILQ_FIRST(&monitor_block_backends);
 584 }
 585
 586 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
 587  * the monitor or attached to a BlockBackend */
 588 BlockDriverState *bdrv_next(BdrvNextIterator *it)
 589 {
 590     BlockDriverState *bs, *old_bs;
 591
 592     /* Must be called from the main loop */
 593     assert(qemu_get_current_aio_context() == qemu_get_aio_context());
 594
 595     /* First, return all root nodes of BlockBackends. In order to avoid
 596      * returning a BDS twice when multiple BBs refer to it, we only return it
 597      * if the BB is the first one in the parent list of the BDS. */
 598     if (it->phase == BDRV_NEXT_BACKEND_ROOTS) {
 599         BlockBackend *old_blk = it->blk;
 600
 601         old_bs = old_blk ? blk_bs(old_blk) : NULL;
 602
 603         do {
 604             it->blk = blk_all_next(it->blk);
 605             bs = it->blk ? blk_bs(it->blk) : NULL;
 606         } while (it->blk && (bs == NULL || bdrv_first_blk(bs) != it->blk));
 607
 608         if (it->blk) {
 609             blk_ref(it->blk);
 610         }
 611         blk_unref(old_blk);
 612
 613         if (bs) {
 614             bdrv_ref(bs);
 615             bdrv_unref(old_bs);
 616             return bs;
 617         }
 618         it->phase = BDRV_NEXT_MONITOR_OWNED;
 619     } else {
 620         old_bs = it->bs;
 621     }
 622
 623     /* Then return the monitor-owned BDSes without a BB attached. Ignore all
 624      * BDSes that are attached to a BlockBackend here; they have been handled
 625      * by the above block already */
 626     do {
 627         it->bs = bdrv_next_monitor_owned(it->bs);
 628         bs = it->bs;
 629     } while (bs && bdrv_has_blk(bs));
 630
 631     if (bs) {
 632         bdrv_ref(bs);
 633     }
 634     bdrv_unref(old_bs);
 635
 636     return bs;
 637 }
 638
 639 static void bdrv_next_reset(BdrvNextIterator *it)
 640 {
 641     *it = (BdrvNextIterator) {
 642         .phase = BDRV_NEXT_BACKEND_ROOTS,
 643     };
 644 }
 645
 646 BlockDriverState *bdrv_first(BdrvNextIterator *it)
 647 {
 648     GLOBAL_STATE_CODE();
 649     bdrv_next_reset(it);
 650     return bdrv_next(it);
 651 }
 652
 653 /* Must be called when aborting a bdrv_next() iteration before
 654  * bdrv_next() returns NULL */
 655 void bdrv_next_cleanup(BdrvNextIterator *it)
 656 {
 657     /* Must be called from the main loop */
 658     assert(qemu_get_current_aio_context() == qemu_get_aio_context());
 659
 660     if (it->phase == BDRV_NEXT_BACKEND_ROOTS) {
 661         if (it->blk) {
 662             bdrv_unref(blk_bs(it->blk));
 663             blk_unref(it->blk);
 664         }
 665     } else {
 666         bdrv_unref(it->bs);
 667     }
 668
 669     bdrv_next_reset(it);
 670 }
 671
 672 /*
 673  * Add a BlockBackend into the list of backends referenced by the monitor, with
 674  * the given @name acting as the handle for the monitor.
 675  * Strictly for use by blockdev.c.
 676  *
 677  * @name must not be null or empty.
 678  *
 679  * Returns true on success and false on failure. In the latter case, an Error
 680  * object is returned through @errp.
 681  */
 682 bool monitor_add_blk(BlockBackend *blk, const char *name, Error **errp)
 683 {
 684     assert(!blk->name);
 685     assert(name && name[0]);
 686     GLOBAL_STATE_CODE();
 687
 688     if (!id_wellformed(name)) {
 689         error_setg(errp, "Invalid device name");
 690         return false;
 691     }
 692     if (blk_by_name(name)) {
 693         error_setg(errp, "Device with id '%s' already exists", name);
 694         return false;
 695     }
 696     if (bdrv_find_node(name)) {
 697         error_setg(errp,
 698                    "Device name '%s' conflicts with an existing node name",
 699                    name);
 700         return false;
 701     }
 702
 703     blk->name = g_strdup(name);
 704     QTAILQ_INSERT_TAIL(&monitor_block_backends, blk, monitor_link);
 705     return true;
 706 }
 707
 708 /*
 709  * Remove a BlockBackend from the list of backends referenced by the monitor.
 710  * Strictly for use by blockdev.c.
 711  */
 712 void monitor_remove_blk(BlockBackend *blk)
 713 {
 714     GLOBAL_STATE_CODE();
 715
 716     if (!blk->name) {
 717         return;
 718     }
 719
 720     QTAILQ_REMOVE(&monitor_block_backends, blk, monitor_link);
 721     g_free(blk->name);
 722     blk->name = NULL;
 723 }
 724
 725 /*
 726  * Return @blk's name, a non-null string.
 727  * Returns an empty string iff @blk is not referenced by the monitor.
 728  */
 729 const char *blk_name(const BlockBackend *blk)
 730 {
 731     IO_CODE();
 732     return blk->name ?: "";
 733 }
 734
 735 /*
 736  * Return the BlockBackend with name @name if it exists, else null.
 737  * @name must not be null.
 738  */
 739 BlockBackend *blk_by_name(const char *name)
 740 {
 741     BlockBackend *blk = NULL;
 742
 743     GLOBAL_STATE_CODE();
 744     assert(name);
 745     while ((blk = blk_next(blk)) != NULL) {
 746         if (!strcmp(name, blk->name)) {
 747             return blk;
 748         }
 749     }
 750     return NULL;
 751 }
 752
 753 /*
 754  * Return the BlockDriverState attached to @blk if any, else null.
 755  */
 756 BlockDriverState *blk_bs(BlockBackend *blk)
 757 {
 758     IO_CODE();
 759     return blk->root ? blk->root->bs : NULL;
 760 }
 761
 762 static BlockBackend *bdrv_first_blk(BlockDriverState *bs)
 763 {
 764     BdrvChild *child;
 765
 766     GLOBAL_STATE_CODE();
 767
 768     QLIST_FOREACH(child, &bs->parents, next_parent) {
 769         if (child->klass == &child_root) {
 770             return child->opaque;
 771         }
 772     }
 773
 774     return NULL;
 775 }
 776
 777 /*
 778  * Returns true if @bs has an associated BlockBackend.
 779  */
 780 bool bdrv_has_blk(BlockDriverState *bs)
 781 {
 782     GLOBAL_STATE_CODE();
 783     return bdrv_first_blk(bs) != NULL;
 784 }
 785
 786 /*
 787  * Returns true if @bs has only BlockBackends as parents.
 788  */
 789 bool bdrv_is_root_node(BlockDriverState *bs)
 790 {
 791     BdrvChild *c;
 792
 793     GLOBAL_STATE_CODE();
 794     QLIST_FOREACH(c, &bs->parents, next_parent) {
 795         if (c->klass != &child_root) {
 796             return false;
 797         }
 798     }
 799
 800     return true;
 801 }
 802
 803 /*
 804  * Return @blk's DriveInfo if any, else null.
 805  */
 806 DriveInfo *blk_legacy_dinfo(BlockBackend *blk)
 807 {
 808     GLOBAL_STATE_CODE();
 809     return blk->legacy_dinfo;
 810 }
 811
 812 /*
 813  * Set @blk's DriveInfo to @dinfo, and return it.
 814  * @blk must not have a DriveInfo set already.
 815  * No other BlockBackend may have the same DriveInfo set.
 816  */
 817 DriveInfo *blk_set_legacy_dinfo(BlockBackend *blk, DriveInfo *dinfo)
 818 {
 819     assert(!blk->legacy_dinfo);
 820     GLOBAL_STATE_CODE();
 821     return blk->legacy_dinfo = dinfo;
 822 }
 823
 824 /*
 825  * Return the BlockBackend with DriveInfo @dinfo.
 826  * It must exist.
 827  */
 828 BlockBackend *blk_by_legacy_dinfo(DriveInfo *dinfo)
 829 {
 830     BlockBackend *blk = NULL;
 831     GLOBAL_STATE_CODE();
 832
 833     while ((blk = blk_next(blk)) != NULL) {
 834         if (blk->legacy_dinfo == dinfo) {
 835             return blk;
 836         }
 837     }
 838     abort();
 839 }
 840
 841 /*
 842  * Returns a pointer to the publicly accessible fields of @blk.
 843  */
 844 BlockBackendPublic *blk_get_public(BlockBackend *blk)
 845 {
 846     GLOBAL_STATE_CODE();
 847     return &blk->public;
 848 }
 849
 850 /*
 851  * Returns a BlockBackend given the associated @public fields.
 852  */
 853 BlockBackend *blk_by_public(BlockBackendPublic *public)
 854 {
 855     GLOBAL_STATE_CODE();
 856     return container_of(public, BlockBackend, public);
 857 }
 858
 859 /*
 860  * Disassociates the currently associated BlockDriverState from @blk.
 861  */
 862 void blk_remove_bs(BlockBackend *blk)
 863 {
 864     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
 865     BdrvChild *root;
 866
 867     GLOBAL_STATE_CODE();
 868
 869     notifier_list_notify(&blk->remove_bs_notifiers, blk);
 870     if (tgm->throttle_state) {
 871         BlockDriverState *bs = blk_bs(blk);
 872
 873         /*
 874          * Take a ref in case blk_bs() changes across bdrv_drained_begin(), for
 875          * example, if a temporary filter node is removed by a blockjob.
 876          */
 877         bdrv_ref(bs);
 878         bdrv_drained_begin(bs);
 879         throttle_group_detach_aio_context(tgm);
 880         throttle_group_attach_aio_context(tgm, qemu_get_aio_context());
 881         bdrv_drained_end(bs);
 882         bdrv_unref(bs);
 883     }
 884
 885     blk_update_root_state(blk);
 886
 887     /* bdrv_root_unref_child() will cause blk->root to become stale and may
 888      * switch to a completion coroutine later on. Let's drain all I/O here
 889      * to avoid that and a potential QEMU crash.
 890      */
 891     blk_drain(blk);
 892     root = blk->root;
 893     blk->root = NULL;
 894     bdrv_root_unref_child(root);
 895 }
 896
 897 /*
 898  * Associates a new BlockDriverState with @blk.
 899  */
 900 int blk_insert_bs(BlockBackend *blk, BlockDriverState *bs, Error **errp)
 901 {
 902     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
 903     GLOBAL_STATE_CODE();
 904     bdrv_ref(bs);
 905     blk->root = bdrv_root_attach_child(bs, "root", &child_root,
 906                                        BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY,
 907                                        blk->perm, blk->shared_perm,
 908                                        blk, errp);
 909     if (blk->root == NULL) {
 910         return -EPERM;
 911     }
 912
 913     notifier_list_notify(&blk->insert_bs_notifiers, blk);
 914     if (tgm->throttle_state) {
 915         throttle_group_detach_aio_context(tgm);
 916         throttle_group_attach_aio_context(tgm, bdrv_get_aio_context(bs));
 917     }
 918
 919     return 0;
 920 }
 921
 922 /*
 923  * Change BlockDriverState associated with @blk.
 924  */
 925 int blk_replace_bs(BlockBackend *blk, BlockDriverState *new_bs, Error **errp)
 926 {
 927     GLOBAL_STATE_CODE();
 928     return bdrv_replace_child_bs(blk->root, new_bs, errp);
 929 }
 930
 931 /*
 932  * Sets the permission bitmasks that the user of the BlockBackend needs.
 933  */
 934 int blk_set_perm(BlockBackend *blk, uint64_t perm, uint64_t shared_perm,
 935                  Error **errp)
 936 {
 937     int ret;
 938     GLOBAL_STATE_CODE();
 939
 940     if (blk->root && !blk->disable_perm) {
 941         ret = bdrv_child_try_set_perm(blk->root, perm, shared_perm, errp);
 942         if (ret < 0) {
 943             return ret;
 944         }
 945     }
 946
 947     blk->perm = perm;
 948     blk->shared_perm = shared_perm;
 949
 950     return 0;
 951 }
 952
 953 void blk_get_perm(BlockBackend *blk, uint64_t *perm, uint64_t *shared_perm)
 954 {
 955     GLOBAL_STATE_CODE();
 956     *perm = blk->perm;
 957     *shared_perm = blk->shared_perm;
 958 }
 959
 960 /*
 961  * Attach device model @dev to @blk.
 962  * Return 0 on success, -EBUSY when a device model is attached already.
 963  */
 964 int blk_attach_dev(BlockBackend *blk, DeviceState *dev)
 965 {
 966     GLOBAL_STATE_CODE();
 967     if (blk->dev) {
 968         return -EBUSY;
 969     }
 970
 971     /* While migration is still incoming, we don't need to apply the
 972      * permissions of guest device BlockBackends. We might still have a block
 973      * job or NBD server writing to the image for storage migration. */
 974     if (runstate_check(RUN_STATE_INMIGRATE)) {
 975         blk->disable_perm = true;
 976     }
 977
 978     blk_ref(blk);
 979     blk->dev = dev;
 980     blk_iostatus_reset(blk);
 981
 982     return 0;
 983 }
 984
 985 /*
 986  * Detach device model @dev from @blk.
 987  * @dev must be currently attached to @blk.
 988  */
 989 void blk_detach_dev(BlockBackend *blk, DeviceState *dev)
 990 {
 991     assert(blk->dev == dev);
 992     GLOBAL_STATE_CODE();
 993     blk->dev = NULL;
 994     blk->dev_ops = NULL;
 995     blk->dev_opaque = NULL;
 996     blk_set_perm(blk, 0, BLK_PERM_ALL, &error_abort);
 997     blk_unref(blk);
 998 }
 999
1000 /*
1001  * Return the device model attached to @blk if any, else null.
1002  */
1003 DeviceState *blk_get_attached_dev(BlockBackend *blk)
1004 {
1005     GLOBAL_STATE_CODE();
1006     return blk->dev;
1007 }
1008
1009 /* Return the qdev ID, or if no ID is assigned the QOM path, of the block
1010  * device attached to the BlockBackend. */
1011 char *blk_get_attached_dev_id(BlockBackend *blk)
1012 {
1013     DeviceState *dev = blk->dev;
1014     IO_CODE();
1015
1016     if (!dev) {
1017         return g_strdup("");
1018     } else if (dev->id) {
1019         return g_strdup(dev->id);
1020     }
1021
1022     return object_get_canonical_path(OBJECT(dev)) ?: g_strdup("");
1023 }
1024
1025 /*
1026  * Return the BlockBackend which has the device model @dev attached if it
1027  * exists, else null.
1028  *
1029  * @dev must not be null.
1030  */
1031 BlockBackend *blk_by_dev(void *dev)
1032 {
1033     BlockBackend *blk = NULL;
1034
1035     GLOBAL_STATE_CODE();
1036
1037     assert(dev != NULL);
1038     while ((blk = blk_all_next(blk)) != NULL) {
1039         if (blk->dev == dev) {
1040             return blk;
1041         }
1042     }
1043     return NULL;
1044 }
1045
1046 /*
1047  * Set @blk's device model callbacks to @ops.
1048  * @opaque is the opaque argument to pass to the callbacks.
1049  * This is for use by device models.
1050  */
1051 void blk_set_dev_ops(BlockBackend *blk, const BlockDevOps *ops,
1052                      void *opaque)
1053 {
1054     GLOBAL_STATE_CODE();
1055     blk->dev_ops = ops;
1056     blk->dev_opaque = opaque;
1057
1058     /* Are we currently quiesced? Should we enforce this right now? */
1059     if (blk->quiesce_counter && ops && ops->drained_begin) {
1060         ops->drained_begin(opaque);
1061     }
1062 }
1063
1064 /*
1065  * Notify @blk's attached device model of media change.
1066  *
1067  * If @load is true, notify of media load. This action can fail, meaning that
1068  * the medium cannot be loaded. @errp is set then.
1069  *
1070  * If @load is false, notify of media eject. This can never fail.
1071  *
1072  * Also send DEVICE_TRAY_MOVED events as appropriate.
1073  */
1074 void blk_dev_change_media_cb(BlockBackend *blk, bool load, Error **errp)
1075 {
1076     GLOBAL_STATE_CODE();
1077     if (blk->dev_ops && blk->dev_ops->change_media_cb) {
1078         bool tray_was_open, tray_is_open;
1079         Error *local_err = NULL;
1080
1081         tray_was_open = blk_dev_is_tray_open(blk);
1082         blk->dev_ops->change_media_cb(blk->dev_opaque, load, &local_err);
1083         if (local_err) {
1084             assert(load == true);
1085             error_propagate(errp, local_err);
1086             return;
1087         }
1088         tray_is_open = blk_dev_is_tray_open(blk);
1089
1090         if (tray_was_open != tray_is_open) {
1091             char *id = blk_get_attached_dev_id(blk);
1092             qapi_event_send_device_tray_moved(blk_name(blk), id, tray_is_open);
1093             g_free(id);
1094         }
1095     }
1096 }
1097
1098 static void blk_root_change_media(BdrvChild *child, bool load)
1099 {
1100     blk_dev_change_media_cb(child->opaque, load, NULL);
1101 }
1102
1103 /*
1104  * Does @blk's attached device model have removable media?
1105  * %true if no device model is attached.
1106  */
1107 bool blk_dev_has_removable_media(BlockBackend *blk)
1108 {
1109     GLOBAL_STATE_CODE();
1110     return !blk->dev || (blk->dev_ops && blk->dev_ops->change_media_cb);
1111 }
1112
1113 /*
1114  * Does @blk's attached device model have a tray?
1115  */
1116 bool blk_dev_has_tray(BlockBackend *blk)
1117 {
1118     IO_CODE();
1119     return blk->dev_ops && blk->dev_ops->is_tray_open;
1120 }
1121
1122 /*
1123  * Notify @blk's attached device model of a media eject request.
1124  * If @force is true, the medium is about to be yanked out forcefully.
1125  */
1126 void blk_dev_eject_request(BlockBackend *blk, bool force)
1127 {
1128     GLOBAL_STATE_CODE();
1129     if (blk->dev_ops && blk->dev_ops->eject_request_cb) {
1130         blk->dev_ops->eject_request_cb(blk->dev_opaque, force);
1131     }
1132 }
1133
1134 /*
1135  * Does @blk's attached device model have a tray, and is it open?
1136  */
1137 bool blk_dev_is_tray_open(BlockBackend *blk)
1138 {
1139     IO_CODE();
1140     if (blk_dev_has_tray(blk)) {
1141         return blk->dev_ops->is_tray_open(blk->dev_opaque);
1142     }
1143     return false;
1144 }
1145
1146 /*
1147  * Does @blk's attached device model have the medium locked?
1148  * %false if the device model has no such lock.
1149  */
1150 bool blk_dev_is_medium_locked(BlockBackend *blk)
1151 {
1152     GLOBAL_STATE_CODE();
1153     if (blk->dev_ops && blk->dev_ops->is_medium_locked) {
1154         return blk->dev_ops->is_medium_locked(blk->dev_opaque);
1155     }
1156     return false;
1157 }
1158
1159 /*
1160  * Notify @blk's attached device model of a backend size change.
1161  */
1162 static void blk_root_resize(BdrvChild *child)
1163 {
1164     BlockBackend *blk = child->opaque;
1165
1166     if (blk->dev_ops && blk->dev_ops->resize_cb) {
1167         blk->dev_ops->resize_cb(blk->dev_opaque);
1168     }
1169 }
1170
1171 void blk_iostatus_enable(BlockBackend *blk)
1172 {
1173     GLOBAL_STATE_CODE();
1174     blk->iostatus_enabled = true;
1175     blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
1176 }
1177
1178 /* The I/O status is only enabled if the drive explicitly
1179  * enables it _and_ the VM is configured to stop on errors */
1180 bool blk_iostatus_is_enabled(const BlockBackend *blk)
1181 {
1182     IO_CODE();
1183     return (blk->iostatus_enabled &&
1184            (blk->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
1185             blk->on_write_error == BLOCKDEV_ON_ERROR_STOP   ||
1186             blk->on_read_error == BLOCKDEV_ON_ERROR_STOP));
1187 }
1188
1189 BlockDeviceIoStatus blk_iostatus(const BlockBackend *blk)
1190 {
1191     GLOBAL_STATE_CODE();
1192     return blk->iostatus;
1193 }
1194
1195 void blk_iostatus_disable(BlockBackend *blk)
1196 {
1197     GLOBAL_STATE_CODE();
1198     blk->iostatus_enabled = false;
1199 }
1200
1201 void blk_iostatus_reset(BlockBackend *blk)
1202 {
1203     GLOBAL_STATE_CODE();
1204     if (blk_iostatus_is_enabled(blk)) {
1205         blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
1206     }
1207 }
1208
1209 void blk_iostatus_set_err(BlockBackend *blk, int error)
1210 {
1211     IO_CODE();
1212     assert(blk_iostatus_is_enabled(blk));
1213     if (blk->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
1214         blk->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
1215                                           BLOCK_DEVICE_IO_STATUS_FAILED;
1216     }
1217 }
1218
1219 void blk_set_allow_write_beyond_eof(BlockBackend *blk, bool allow)
1220 {
1221     IO_CODE();
1222     blk->allow_write_beyond_eof = allow;
1223 }
1224
1225 void blk_set_allow_aio_context_change(BlockBackend *blk, bool allow)
1226 {
1227     IO_CODE();
1228     blk->allow_aio_context_change = allow;
1229 }
1230
1231 void blk_set_disable_request_queuing(BlockBackend *blk, bool disable)
1232 {
1233     IO_CODE();
1234     blk->disable_request_queuing = disable;
1235 }
1236
1237 static int blk_check_byte_request(BlockBackend *blk, int64_t offset,
1238                                   int64_t bytes)
1239 {
1240     int64_t len;
1241
1242     if (bytes < 0) {
1243         return -EIO;
1244     }
1245
1246     if (!blk_is_available(blk)) {
1247         return -ENOMEDIUM;
1248     }
1249
1250     if (offset < 0) {
1251         return -EIO;
1252     }
1253
1254     if (!blk->allow_write_beyond_eof) {
1255         len = blk_getlength(blk);
1256         if (len < 0) {
1257             return len;
1258         }
1259
1260         if (offset > len || len - offset < bytes) {
1261             return -EIO;
1262         }
1263     }
1264
1265     return 0;
1266 }
1267
1268 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1269 static void coroutine_fn blk_wait_while_drained(BlockBackend *blk)
1270 {
1271     assert(blk->in_flight > 0);
1272
1273     if (blk->quiesce_counter && !blk->disable_request_queuing) {
1274         blk_dec_in_flight(blk);
1275         qemu_co_queue_wait(&blk->queued_requests, NULL);
1276         blk_inc_in_flight(blk);
1277     }
1278 }
1279
1280 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1281 static int coroutine_fn
1282 blk_co_do_preadv_part(BlockBackend *blk, int64_t offset, int64_t bytes,
1283                       QEMUIOVector *qiov, size_t qiov_offset,
1284                       BdrvRequestFlags flags)
1285 {
1286     int ret;
1287     BlockDriverState *bs;
1288     IO_CODE();
1289
1290     blk_wait_while_drained(blk);
1291
1292     /* Call blk_bs() only after waiting, the graph may have changed */
1293     bs = blk_bs(blk);
1294     trace_blk_co_preadv(blk, bs, offset, bytes, flags);
1295
1296     ret = blk_check_byte_request(blk, offset, bytes);
1297     if (ret < 0) {
1298         return ret;
1299     }
1300
1301     bdrv_inc_in_flight(bs);
1302
1303     /* throttling disk I/O */
1304     if (blk->public.throttle_group_member.throttle_state) {
1305         throttle_group_co_io_limits_intercept(&blk->public.throttle_group_member,
1306                 bytes, false);
1307     }
1308
1309     ret = bdrv_co_preadv_part(blk->root, offset, bytes, qiov, qiov_offset,
1310                               flags);
1311     bdrv_dec_in_flight(bs);
1312     return ret;
1313 }
1314
1315 int coroutine_fn blk_co_pread(BlockBackend *blk, int64_t offset, int64_t bytes,
1316                               void *buf, BdrvRequestFlags flags)
1317 {
1318     QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
1319     IO_OR_GS_CODE();
1320
1321     assert(bytes <= SIZE_MAX);
1322
1323     return blk_co_preadv(blk, offset, bytes, &qiov, flags);
1324 }
1325
1326 int coroutine_fn blk_co_preadv(BlockBackend *blk, int64_t offset,
1327                                int64_t bytes, QEMUIOVector *qiov,
1328                                BdrvRequestFlags flags)
1329 {
1330     int ret;
1331     IO_OR_GS_CODE();
1332
1333     blk_inc_in_flight(blk);
1334     ret = blk_co_do_preadv_part(blk, offset, bytes, qiov, 0, flags);
1335     blk_dec_in_flight(blk);
1336
1337     return ret;
1338 }
1339
1340 int coroutine_fn blk_co_preadv_part(BlockBackend *blk, int64_t offset,
1341                                     int64_t bytes, QEMUIOVector *qiov,
1342                                     size_t qiov_offset, BdrvRequestFlags flags)
1343 {
1344     int ret;
1345     IO_OR_GS_CODE();
1346
1347     blk_inc_in_flight(blk);
1348     ret = blk_co_do_preadv_part(blk, offset, bytes, qiov, qiov_offset, flags);
1349     blk_dec_in_flight(blk);
1350
1351     return ret;
1352 }
1353
1354 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1355 static int coroutine_fn
1356 blk_co_do_pwritev_part(BlockBackend *blk, int64_t offset, int64_t bytes,
1357                        QEMUIOVector *qiov, size_t qiov_offset,
1358                        BdrvRequestFlags flags)
1359 {
1360     int ret;
1361     BlockDriverState *bs;
1362     IO_CODE();
1363
1364     blk_wait_while_drained(blk);
1365
1366     /* Call blk_bs() only after waiting, the graph may have changed */
1367     bs = blk_bs(blk);
1368     trace_blk_co_pwritev(blk, bs, offset, bytes, flags);
1369
1370     ret = blk_check_byte_request(blk, offset, bytes);
1371     if (ret < 0) {
1372         return ret;
1373     }
1374
1375     bdrv_inc_in_flight(bs);
1376     /* throttling disk I/O */
1377     if (blk->public.throttle_group_member.throttle_state) {
1378         throttle_group_co_io_limits_intercept(&blk->public.throttle_group_member,
1379                 bytes, true);
1380     }
1381
1382     if (!blk->enable_write_cache) {
1383         flags |= BDRV_REQ_FUA;
1384     }
1385
1386     ret = bdrv_co_pwritev_part(blk->root, offset, bytes, qiov, qiov_offset,
1387                                flags);
1388     bdrv_dec_in_flight(bs);
1389     return ret;
1390 }
1391
1392 int coroutine_fn blk_co_pwritev_part(BlockBackend *blk, int64_t offset,
1393                                      int64_t bytes,
1394                                      QEMUIOVector *qiov, size_t qiov_offset,
1395                                      BdrvRequestFlags flags)
1396 {
1397     int ret;
1398     IO_OR_GS_CODE();
1399
1400     blk_inc_in_flight(blk);
1401     ret = blk_co_do_pwritev_part(blk, offset, bytes, qiov, qiov_offset, flags);
1402     blk_dec_in_flight(blk);
1403
1404     return ret;
1405 }
1406
1407 int coroutine_fn blk_co_pwrite(BlockBackend *blk, int64_t offset, int64_t bytes,
1408                                const void *buf, BdrvRequestFlags flags)
1409 {
1410     QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
1411     IO_OR_GS_CODE();
1412
1413     assert(bytes <= SIZE_MAX);
1414
1415     return blk_co_pwritev(blk, offset, bytes, &qiov, flags);
1416 }
1417
1418 int coroutine_fn blk_co_pwritev(BlockBackend *blk, int64_t offset,
1419                                 int64_t bytes, QEMUIOVector *qiov,
1420                                 BdrvRequestFlags flags)
1421 {
1422     IO_OR_GS_CODE();
1423     return blk_co_pwritev_part(blk, offset, bytes, qiov, 0, flags);
1424 }
1425
1426 typedef struct BlkRwCo {
1427     BlockBackend *blk;
1428     int64_t offset;
1429     void *iobuf;
1430     int ret;
1431     BdrvRequestFlags flags;
1432 } BlkRwCo;
1433
1434 int blk_make_zero(BlockBackend *blk, BdrvRequestFlags flags)
1435 {
1436     GLOBAL_STATE_CODE();
1437     return bdrv_make_zero(blk->root, flags);
1438 }
1439
1440 void blk_inc_in_flight(BlockBackend *blk)
1441 {
1442     IO_CODE();
1443     qatomic_inc(&blk->in_flight);
1444 }
1445
1446 void blk_dec_in_flight(BlockBackend *blk)
1447 {
1448     IO_CODE();
1449     qatomic_dec(&blk->in_flight);
1450     aio_wait_kick();
1451 }
1452
1453 static void error_callback_bh(void *opaque)
1454 {
1455     struct BlockBackendAIOCB *acb = opaque;
1456
1457     blk_dec_in_flight(acb->blk);
1458     acb->common.cb(acb->common.opaque, acb->ret);
1459     qemu_aio_unref(acb);
1460 }
1461
1462 BlockAIOCB *blk_abort_aio_request(BlockBackend *blk,
1463                                   BlockCompletionFunc *cb,
1464                                   void *opaque, int ret)
1465 {
1466     struct BlockBackendAIOCB *acb;
1467     IO_CODE();
1468
1469     blk_inc_in_flight(blk);
1470     acb = blk_aio_get(&block_backend_aiocb_info, blk, cb, opaque);
1471     acb->blk = blk;
1472     acb->ret = ret;
1473
1474     replay_bh_schedule_oneshot_event(blk_get_aio_context(blk),
1475                                      error_callback_bh, acb);
1476     return &acb->common;
1477 }
1478
1479 typedef struct BlkAioEmAIOCB {
1480     BlockAIOCB common;
1481     BlkRwCo rwco;
1482     int64_t bytes;
1483     bool has_returned;
1484 } BlkAioEmAIOCB;
1485
1486 static AioContext *blk_aio_em_aiocb_get_aio_context(BlockAIOCB *acb_)
1487 {
1488     BlkAioEmAIOCB *acb = container_of(acb_, BlkAioEmAIOCB, common);
1489
1490     return blk_get_aio_context(acb->rwco.blk);
1491 }
1492
1493 static const AIOCBInfo blk_aio_em_aiocb_info = {
1494     .aiocb_size         = sizeof(BlkAioEmAIOCB),
1495     .get_aio_context    = blk_aio_em_aiocb_get_aio_context,
1496 };
1497
1498 static void blk_aio_complete(BlkAioEmAIOCB *acb)
1499 {
1500     if (acb->has_returned) {
1501         acb->common.cb(acb->common.opaque, acb->rwco.ret);
1502         blk_dec_in_flight(acb->rwco.blk);
1503         qemu_aio_unref(acb);
1504     }
1505 }
1506
1507 static void blk_aio_complete_bh(void *opaque)
1508 {
1509     BlkAioEmAIOCB *acb = opaque;
1510     assert(acb->has_returned);
1511     blk_aio_complete(acb);
1512 }
1513
1514 static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset,
1515                                 int64_t bytes,
1516                                 void *iobuf, CoroutineEntry co_entry,
1517                                 BdrvRequestFlags flags,
1518                                 BlockCompletionFunc *cb, void *opaque)
1519 {
1520     BlkAioEmAIOCB *acb;
1521     Coroutine *co;
1522
1523     blk_inc_in_flight(blk);
1524     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1525     acb->rwco = (BlkRwCo) {
1526         .blk    = blk,
1527         .offset = offset,
1528         .iobuf  = iobuf,
1529         .flags  = flags,
1530         .ret    = NOT_DONE,
1531     };
1532     acb->bytes = bytes;
1533     acb->has_returned = false;
1534
1535     co = qemu_coroutine_create(co_entry, acb);
1536     bdrv_coroutine_enter(blk_bs(blk), co);
1537
1538     acb->has_returned = true;
1539     if (acb->rwco.ret != NOT_DONE) {
1540         replay_bh_schedule_oneshot_event(blk_get_aio_context(blk),
1541                                          blk_aio_complete_bh, acb);
1542     }
1543
1544     return &acb->common;
1545 }
1546
1547 static void coroutine_fn blk_aio_read_entry(void *opaque)
1548 {
1549     BlkAioEmAIOCB *acb = opaque;
1550     BlkRwCo *rwco = &acb->rwco;
1551     QEMUIOVector *qiov = rwco->iobuf;
1552
1553     assert(qiov->size == acb->bytes);
1554     rwco->ret = blk_co_do_preadv_part(rwco->blk, rwco->offset, acb->bytes, qiov,
1555                                       0, rwco->flags);
1556     blk_aio_complete(acb);
1557 }
1558
1559 static void coroutine_fn blk_aio_write_entry(void *opaque)
1560 {
1561     BlkAioEmAIOCB *acb = opaque;
1562     BlkRwCo *rwco = &acb->rwco;
1563     QEMUIOVector *qiov = rwco->iobuf;
1564
1565     assert(!qiov || qiov->size == acb->bytes);
1566     rwco->ret = blk_co_do_pwritev_part(rwco->blk, rwco->offset, acb->bytes,
1567                                        qiov, 0, rwco->flags);
1568     blk_aio_complete(acb);
1569 }
1570
1571 BlockAIOCB *blk_aio_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1572                                   int64_t bytes, BdrvRequestFlags flags,
1573                                   BlockCompletionFunc *cb, void *opaque)
1574 {
1575     IO_CODE();
1576     return blk_aio_prwv(blk, offset, bytes, NULL, blk_aio_write_entry,
1577                         flags | BDRV_REQ_ZERO_WRITE, cb, opaque);
1578 }
1579
1580 int64_t blk_getlength(BlockBackend *blk)
1581 {
1582     IO_CODE();
1583     if (!blk_is_available(blk)) {
1584         return -ENOMEDIUM;
1585     }
1586
1587     return bdrv_getlength(blk_bs(blk));
1588 }
1589
1590 void blk_get_geometry(BlockBackend *blk, uint64_t *nb_sectors_ptr)
1591 {
1592     IO_CODE();
1593     if (!blk_bs(blk)) {
1594         *nb_sectors_ptr = 0;
1595     } else {
1596         bdrv_get_geometry(blk_bs(blk), nb_sectors_ptr);
1597     }
1598 }
1599
1600 int64_t blk_nb_sectors(BlockBackend *blk)
1601 {
1602     IO_CODE();
1603     if (!blk_is_available(blk)) {
1604         return -ENOMEDIUM;
1605     }
1606
1607     return bdrv_nb_sectors(blk_bs(blk));
1608 }
1609
1610 BlockAIOCB *blk_aio_preadv(BlockBackend *blk, int64_t offset,
1611                            QEMUIOVector *qiov, BdrvRequestFlags flags,
1612                            BlockCompletionFunc *cb, void *opaque)
1613 {
1614     IO_CODE();
1615     assert((uint64_t)qiov->size <= INT64_MAX);
1616     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1617                         blk_aio_read_entry, flags, cb, opaque);
1618 }
1619
1620 BlockAIOCB *blk_aio_pwritev(BlockBackend *blk, int64_t offset,
1621                             QEMUIOVector *qiov, BdrvRequestFlags flags,
1622                             BlockCompletionFunc *cb, void *opaque)
1623 {
1624     IO_CODE();
1625     assert((uint64_t)qiov->size <= INT64_MAX);
1626     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1627                         blk_aio_write_entry, flags, cb, opaque);
1628 }
1629
1630 void blk_aio_cancel(BlockAIOCB *acb)
1631 {
1632     GLOBAL_STATE_CODE();
1633     bdrv_aio_cancel(acb);
1634 }
1635
1636 void blk_aio_cancel_async(BlockAIOCB *acb)
1637 {
1638     IO_CODE();
1639     bdrv_aio_cancel_async(acb);
1640 }
1641
1642 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1643 static int coroutine_fn
1644 blk_co_do_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1645 {
1646     IO_CODE();
1647
1648     blk_wait_while_drained(blk);
1649
1650     if (!blk_is_available(blk)) {
1651         return -ENOMEDIUM;
1652     }
1653
1654     return bdrv_co_ioctl(blk_bs(blk), req, buf);
1655 }
1656
1657 int coroutine_fn blk_co_ioctl(BlockBackend *blk, unsigned long int req,
1658                               void *buf)
1659 {
1660     int ret;
1661     IO_OR_GS_CODE();
1662
1663     blk_inc_in_flight(blk);
1664     ret = blk_co_do_ioctl(blk, req, buf);
1665     blk_dec_in_flight(blk);
1666
1667     return ret;
1668 }
1669
1670 static void coroutine_fn blk_aio_ioctl_entry(void *opaque)
1671 {
1672     BlkAioEmAIOCB *acb = opaque;
1673     BlkRwCo *rwco = &acb->rwco;
1674
1675     rwco->ret = blk_co_do_ioctl(rwco->blk, rwco->offset, rwco->iobuf);
1676
1677     blk_aio_complete(acb);
1678 }
1679
1680 BlockAIOCB *blk_aio_ioctl(BlockBackend *blk, unsigned long int req, void *buf,
1681                           BlockCompletionFunc *cb, void *opaque)
1682 {
1683     IO_CODE();
1684     return blk_aio_prwv(blk, req, 0, buf, blk_aio_ioctl_entry, 0, cb, opaque);
1685 }
1686
1687 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1688 static int coroutine_fn
1689 blk_co_do_pdiscard(BlockBackend *blk, int64_t offset, int64_t bytes)
1690 {
1691     int ret;
1692     IO_CODE();
1693
1694     blk_wait_while_drained(blk);
1695
1696     ret = blk_check_byte_request(blk, offset, bytes);
1697     if (ret < 0) {
1698         return ret;
1699     }
1700
1701     return bdrv_co_pdiscard(blk->root, offset, bytes);
1702 }
1703
1704 static void coroutine_fn blk_aio_pdiscard_entry(void *opaque)
1705 {
1706     BlkAioEmAIOCB *acb = opaque;
1707     BlkRwCo *rwco = &acb->rwco;
1708
1709     rwco->ret = blk_co_do_pdiscard(rwco->blk, rwco->offset, acb->bytes);
1710     blk_aio_complete(acb);
1711 }
1712
1713 BlockAIOCB *blk_aio_pdiscard(BlockBackend *blk,
1714                              int64_t offset, int64_t bytes,
1715                              BlockCompletionFunc *cb, void *opaque)
1716 {
1717     IO_CODE();
1718     return blk_aio_prwv(blk, offset, bytes, NULL, blk_aio_pdiscard_entry, 0,
1719                         cb, opaque);
1720 }
1721
1722 int coroutine_fn blk_co_pdiscard(BlockBackend *blk, int64_t offset,
1723                                  int64_t bytes)
1724 {
1725     int ret;
1726     IO_OR_GS_CODE();
1727
1728     blk_inc_in_flight(blk);
1729     ret = blk_co_do_pdiscard(blk, offset, bytes);
1730     blk_dec_in_flight(blk);
1731
1732     return ret;
1733 }
1734
1735 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1736 static int coroutine_fn blk_co_do_flush(BlockBackend *blk)
1737 {
1738     blk_wait_while_drained(blk);
1739     IO_CODE();
1740
1741     if (!blk_is_available(blk)) {
1742         return -ENOMEDIUM;
1743     }
1744
1745     return bdrv_co_flush(blk_bs(blk));
1746 }
1747
1748 static void coroutine_fn blk_aio_flush_entry(void *opaque)
1749 {
1750     BlkAioEmAIOCB *acb = opaque;
1751     BlkRwCo *rwco = &acb->rwco;
1752
1753     rwco->ret = blk_co_do_flush(rwco->blk);
1754     blk_aio_complete(acb);
1755 }
1756
1757 BlockAIOCB *blk_aio_flush(BlockBackend *blk,
1758                           BlockCompletionFunc *cb, void *opaque)
1759 {
1760     IO_CODE();
1761     return blk_aio_prwv(blk, 0, 0, NULL, blk_aio_flush_entry, 0, cb, opaque);
1762 }
1763
1764 int coroutine_fn blk_co_flush(BlockBackend *blk)
1765 {
1766     int ret;
1767     IO_OR_GS_CODE();
1768
1769     blk_inc_in_flight(blk);
1770     ret = blk_co_do_flush(blk);
1771     blk_dec_in_flight(blk);
1772
1773     return ret;
1774 }
1775
1776 void blk_drain(BlockBackend *blk)
1777 {
1778     BlockDriverState *bs = blk_bs(blk);
1779     GLOBAL_STATE_CODE();
1780
1781     if (bs) {
1782         bdrv_ref(bs);
1783         bdrv_drained_begin(bs);
1784     }
1785
1786     /* We may have -ENOMEDIUM completions in flight */
1787     AIO_WAIT_WHILE(blk_get_aio_context(blk),
1788                    qatomic_mb_read(&blk->in_flight) > 0);
1789
1790     if (bs) {
1791         bdrv_drained_end(bs);
1792         bdrv_unref(bs);
1793     }
1794 }
1795
1796 void blk_drain_all(void)
1797 {
1798     BlockBackend *blk = NULL;
1799
1800     GLOBAL_STATE_CODE();
1801
1802     bdrv_drain_all_begin();
1803
1804     while ((blk = blk_all_next(blk)) != NULL) {
1805         AioContext *ctx = blk_get_aio_context(blk);
1806
1807         aio_context_acquire(ctx);
1808
1809         /* We may have -ENOMEDIUM completions in flight */
1810         AIO_WAIT_WHILE(ctx, qatomic_mb_read(&blk->in_flight) > 0);
1811
1812         aio_context_release(ctx);
1813     }
1814
1815     bdrv_drain_all_end();
1816 }
1817
1818 void blk_set_on_error(BlockBackend *blk, BlockdevOnError on_read_error,
1819                       BlockdevOnError on_write_error)
1820 {
1821     GLOBAL_STATE_CODE();
1822     blk->on_read_error = on_read_error;
1823     blk->on_write_error = on_write_error;
1824 }
1825
1826 BlockdevOnError blk_get_on_error(BlockBackend *blk, bool is_read)
1827 {
1828     IO_CODE();
1829     return is_read ? blk->on_read_error : blk->on_write_error;
1830 }
1831
1832 BlockErrorAction blk_get_error_action(BlockBackend *blk, bool is_read,
1833                                       int error)
1834 {
1835     BlockdevOnError on_err = blk_get_on_error(blk, is_read);
1836     IO_CODE();
1837
1838     switch (on_err) {
1839     case BLOCKDEV_ON_ERROR_ENOSPC:
1840         return (error == ENOSPC) ?
1841                BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
1842     case BLOCKDEV_ON_ERROR_STOP:
1843         return BLOCK_ERROR_ACTION_STOP;
1844     case BLOCKDEV_ON_ERROR_REPORT:
1845         return BLOCK_ERROR_ACTION_REPORT;
1846     case BLOCKDEV_ON_ERROR_IGNORE:
1847         return BLOCK_ERROR_ACTION_IGNORE;
1848     case BLOCKDEV_ON_ERROR_AUTO:
1849     default:
1850         abort();
1851     }
1852 }
1853
1854 static void send_qmp_error_event(BlockBackend *blk,
1855                                  BlockErrorAction action,
1856                                  bool is_read, int error)
1857 {
1858     IoOperationType optype;
1859     BlockDriverState *bs = blk_bs(blk);
1860
1861     optype = is_read ? IO_OPERATION_TYPE_READ : IO_OPERATION_TYPE_WRITE;
1862     qapi_event_send_block_io_error(blk_name(blk), !!bs,
1863                                    bs ? bdrv_get_node_name(bs) : NULL, optype,
1864                                    action, blk_iostatus_is_enabled(blk),
1865                                    error == ENOSPC, strerror(error));
1866 }
1867
1868 /* This is done by device models because, while the block layer knows
1869  * about the error, it does not know whether an operation comes from
1870  * the device or the block layer (from a job, for example).
1871  */
1872 void blk_error_action(BlockBackend *blk, BlockErrorAction action,
1873                       bool is_read, int error)
1874 {
1875     assert(error >= 0);
1876     IO_CODE();
1877
1878     if (action == BLOCK_ERROR_ACTION_STOP) {
1879         /* First set the iostatus, so that "info block" returns an iostatus
1880          * that matches the events raised so far (an additional error iostatus
1881          * is fine, but not a lost one).
1882          */
1883         blk_iostatus_set_err(blk, error);
1884
1885         /* Then raise the request to stop the VM and the event.
1886          * qemu_system_vmstop_request_prepare has two effects.  First,
1887          * it ensures that the STOP event always comes after the
1888          * BLOCK_IO_ERROR event.  Second, it ensures that even if management
1889          * can observe the STOP event and do a "cont" before the STOP
1890          * event is issued, the VM will not stop.  In this case, vm_start()
1891          * also ensures that the STOP/RESUME pair of events is emitted.
1892          */
1893         qemu_system_vmstop_request_prepare();
1894         send_qmp_error_event(blk, action, is_read, error);
1895         qemu_system_vmstop_request(RUN_STATE_IO_ERROR);
1896     } else {
1897         send_qmp_error_event(blk, action, is_read, error);
1898     }
1899 }
1900
1901 /*
1902  * Returns true if the BlockBackend can support taking write permissions
1903  * (because its root node is not read-only).
1904  */
1905 bool blk_supports_write_perm(BlockBackend *blk)
1906 {
1907     BlockDriverState *bs = blk_bs(blk);
1908     GLOBAL_STATE_CODE();
1909
1910     if (bs) {
1911         return !bdrv_is_read_only(bs);
1912     } else {
1913         return blk->root_state.open_flags & BDRV_O_RDWR;
1914     }
1915 }
1916
1917 /*
1918  * Returns true if the BlockBackend can be written to in its current
1919  * configuration (i.e. if write permission have been requested)
1920  */
1921 bool blk_is_writable(BlockBackend *blk)
1922 {
1923     IO_CODE();
1924     return blk->perm & BLK_PERM_WRITE;
1925 }
1926
1927 bool blk_is_sg(BlockBackend *blk)
1928 {
1929     BlockDriverState *bs = blk_bs(blk);
1930     GLOBAL_STATE_CODE();
1931
1932     if (!bs) {
1933         return false;
1934     }
1935
1936     return bdrv_is_sg(bs);
1937 }
1938
1939 bool blk_enable_write_cache(BlockBackend *blk)
1940 {
1941     IO_CODE();
1942     return blk->enable_write_cache;
1943 }
1944
1945 void blk_set_enable_write_cache(BlockBackend *blk, bool wce)
1946 {
1947     IO_CODE();
1948     blk->enable_write_cache = wce;
1949 }
1950
1951 void blk_activate(BlockBackend *blk, Error **errp)
1952 {
1953     BlockDriverState *bs = blk_bs(blk);
1954     GLOBAL_STATE_CODE();
1955
1956     if (!bs) {
1957         error_setg(errp, "Device '%s' has no medium", blk->name);
1958         return;
1959     }
1960
1961     bdrv_activate(bs, errp);
1962 }
1963
1964 bool blk_is_inserted(BlockBackend *blk)
1965 {
1966     BlockDriverState *bs = blk_bs(blk);
1967     IO_CODE();
1968
1969     return bs && bdrv_is_inserted(bs);
1970 }
1971
1972 bool blk_is_available(BlockBackend *blk)
1973 {
1974     IO_CODE();
1975     return blk_is_inserted(blk) && !blk_dev_is_tray_open(blk);
1976 }
1977
1978 void blk_lock_medium(BlockBackend *blk, bool locked)
1979 {
1980     BlockDriverState *bs = blk_bs(blk);
1981     IO_CODE();
1982
1983     if (bs) {
1984         bdrv_lock_medium(bs, locked);
1985     }
1986 }
1987
1988 void blk_eject(BlockBackend *blk, bool eject_flag)
1989 {
1990     BlockDriverState *bs = blk_bs(blk);
1991     char *id;
1992     IO_CODE();
1993
1994     if (bs) {
1995         bdrv_eject(bs, eject_flag);
1996     }
1997
1998     /* Whether or not we ejected on the backend,
1999      * the frontend experienced a tray event. */
2000     id = blk_get_attached_dev_id(blk);
2001     qapi_event_send_device_tray_moved(blk_name(blk), id,
2002                                       eject_flag);
2003     g_free(id);
2004 }
2005
2006 int blk_get_flags(BlockBackend *blk)
2007 {
2008     BlockDriverState *bs = blk_bs(blk);
2009     GLOBAL_STATE_CODE();
2010
2011     if (bs) {
2012         return bdrv_get_flags(bs);
2013     } else {
2014         return blk->root_state.open_flags;
2015     }
2016 }
2017
2018 /* Returns the minimum request alignment, in bytes; guaranteed nonzero */
2019 uint32_t blk_get_request_alignment(BlockBackend *blk)
2020 {
2021     BlockDriverState *bs = blk_bs(blk);
2022     IO_CODE();
2023     return bs ? bs->bl.request_alignment : BDRV_SECTOR_SIZE;
2024 }
2025
2026 /* Returns the maximum hardware transfer length, in bytes; guaranteed nonzero */
2027 uint64_t blk_get_max_hw_transfer(BlockBackend *blk)
2028 {
2029     BlockDriverState *bs = blk_bs(blk);
2030     uint64_t max = INT_MAX;
2031     IO_CODE();
2032
2033     if (bs) {
2034         max = MIN_NON_ZERO(max, bs->bl.max_hw_transfer);
2035         max = MIN_NON_ZERO(max, bs->bl.max_transfer);
2036     }
2037     return ROUND_DOWN(max, blk_get_request_alignment(blk));
2038 }
2039
2040 /* Returns the maximum transfer length, in bytes; guaranteed nonzero */
2041 uint32_t blk_get_max_transfer(BlockBackend *blk)
2042 {
2043     BlockDriverState *bs = blk_bs(blk);
2044     uint32_t max = INT_MAX;
2045     IO_CODE();
2046
2047     if (bs) {
2048         max = MIN_NON_ZERO(max, bs->bl.max_transfer);
2049     }
2050     return ROUND_DOWN(max, blk_get_request_alignment(blk));
2051 }
2052
2053 int blk_get_max_hw_iov(BlockBackend *blk)
2054 {
2055     IO_CODE();
2056     return MIN_NON_ZERO(blk->root->bs->bl.max_hw_iov,
2057                         blk->root->bs->bl.max_iov);
2058 }
2059
2060 int blk_get_max_iov(BlockBackend *blk)
2061 {
2062     IO_CODE();
2063     return blk->root->bs->bl.max_iov;
2064 }
2065
2066 void *blk_try_blockalign(BlockBackend *blk, size_t size)
2067 {
2068     IO_CODE();
2069     return qemu_try_blockalign(blk ? blk_bs(blk) : NULL, size);
2070 }
2071
2072 void *blk_blockalign(BlockBackend *blk, size_t size)
2073 {
2074     IO_CODE();
2075     return qemu_blockalign(blk ? blk_bs(blk) : NULL, size);
2076 }
2077
2078 bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp)
2079 {
2080     BlockDriverState *bs = blk_bs(blk);
2081     GLOBAL_STATE_CODE();
2082
2083     if (!bs) {
2084         return false;
2085     }
2086
2087     return bdrv_op_is_blocked(bs, op, errp);
2088 }
2089
2090 void blk_op_unblock(BlockBackend *blk, BlockOpType op, Error *reason)
2091 {
2092     BlockDriverState *bs = blk_bs(blk);
2093     GLOBAL_STATE_CODE();
2094
2095     if (bs) {
2096         bdrv_op_unblock(bs, op, reason);
2097     }
2098 }
2099
2100 void blk_op_block_all(BlockBackend *blk, Error *reason)
2101 {
2102     BlockDriverState *bs = blk_bs(blk);
2103     GLOBAL_STATE_CODE();
2104
2105     if (bs) {
2106         bdrv_op_block_all(bs, reason);
2107     }
2108 }
2109
2110 void blk_op_unblock_all(BlockBackend *blk, Error *reason)
2111 {
2112     BlockDriverState *bs = blk_bs(blk);
2113     GLOBAL_STATE_CODE();
2114
2115     if (bs) {
2116         bdrv_op_unblock_all(bs, reason);
2117     }
2118 }
2119
2120 AioContext *blk_get_aio_context(BlockBackend *blk)
2121 {
2122     BlockDriverState *bs = blk_bs(blk);
2123     IO_CODE();
2124
2125     if (bs) {
2126         AioContext *ctx = bdrv_get_aio_context(blk_bs(blk));
2127         assert(ctx == blk->ctx);
2128     }
2129
2130     return blk->ctx;
2131 }
2132
2133 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb)
2134 {
2135     BlockBackendAIOCB *blk_acb = DO_UPCAST(BlockBackendAIOCB, common, acb);
2136     return blk_get_aio_context(blk_acb->blk);
2137 }
2138
2139 static int blk_do_set_aio_context(BlockBackend *blk, AioContext *new_context,
2140                                   bool update_root_node, Error **errp)
2141 {
2142     BlockDriverState *bs = blk_bs(blk);
2143     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
2144     int ret;
2145
2146     if (bs) {
2147         bdrv_ref(bs);
2148
2149         if (update_root_node) {
2150             /*
2151              * update_root_node MUST be false for blk_root_set_aio_ctx_commit(),
2152              * as we are already in the commit function of a transaction.
2153              */
2154             ret = bdrv_try_change_aio_context(bs, new_context, blk->root, errp);
2155             if (ret < 0) {
2156                 bdrv_unref(bs);
2157                 return ret;
2158             }
2159         }
2160         if (tgm->throttle_state) {
2161             bdrv_drained_begin(bs);
2162             throttle_group_detach_aio_context(tgm);
2163             throttle_group_attach_aio_context(tgm, new_context);
2164             bdrv_drained_end(bs);
2165         }
2166
2167         bdrv_unref(bs);
2168     }
2169
2170     blk->ctx = new_context;
2171     return 0;
2172 }
2173
2174 int blk_set_aio_context(BlockBackend *blk, AioContext *new_context,
2175                         Error **errp)
2176 {
2177     GLOBAL_STATE_CODE();
2178     return blk_do_set_aio_context(blk, new_context, true, errp);
2179 }
2180
2181 typedef struct BdrvStateBlkRootContext {
2182     AioContext *new_ctx;
2183     BlockBackend *blk;
2184 } BdrvStateBlkRootContext;
2185
2186 static void blk_root_set_aio_ctx_commit(void *opaque)
2187 {
2188     BdrvStateBlkRootContext *s = opaque;
2189     BlockBackend *blk = s->blk;
2190
2191     blk_do_set_aio_context(blk, s->new_ctx, false, &error_abort);
2192 }
2193
2194 static TransactionActionDrv set_blk_root_context = {
2195     .commit = blk_root_set_aio_ctx_commit,
2196     .clean = g_free,
2197 };
2198
2199 static bool blk_root_change_aio_ctx(BdrvChild *child, AioContext *ctx,
2200                                     GHashTable *visited, Transaction *tran,
2201                                     Error **errp)
2202 {
2203     BlockBackend *blk = child->opaque;
2204     BdrvStateBlkRootContext *s;
2205
2206     if (!blk->allow_aio_context_change) {
2207         /*
2208          * Manually created BlockBackends (those with a name) that are not
2209          * attached to anything can change their AioContext without updating
2210          * their user; return an error for others.
2211          */
2212         if (!blk->name || blk->dev) {
2213             /* TODO Add BB name/QOM path */
2214             error_setg(errp, "Cannot change iothread of active block backend");
2215             return false;
2216         }
2217     }
2218
2219     s = g_new(BdrvStateBlkRootContext, 1);
2220     *s = (BdrvStateBlkRootContext) {
2221         .new_ctx = ctx,
2222         .blk = blk,
2223     };
2224
2225     tran_add(tran, &set_blk_root_context, s);
2226     return true;
2227 }
2228
2229 void blk_add_aio_context_notifier(BlockBackend *blk,
2230         void (*attached_aio_context)(AioContext *new_context, void *opaque),
2231         void (*detach_aio_context)(void *opaque), void *opaque)
2232 {
2233     BlockBackendAioNotifier *notifier;
2234     BlockDriverState *bs = blk_bs(blk);
2235     GLOBAL_STATE_CODE();
2236
2237     notifier = g_new(BlockBackendAioNotifier, 1);
2238     notifier->attached_aio_context = attached_aio_context;
2239     notifier->detach_aio_context = detach_aio_context;
2240     notifier->opaque = opaque;
2241     QLIST_INSERT_HEAD(&blk->aio_notifiers, notifier, list);
2242
2243     if (bs) {
2244         bdrv_add_aio_context_notifier(bs, attached_aio_context,
2245                                       detach_aio_context, opaque);
2246     }
2247 }
2248
2249 void blk_remove_aio_context_notifier(BlockBackend *blk,
2250                                      void (*attached_aio_context)(AioContext *,
2251                                                                   void *),
2252                                      void (*detach_aio_context)(void *),
2253                                      void *opaque)
2254 {
2255     BlockBackendAioNotifier *notifier;
2256     BlockDriverState *bs = blk_bs(blk);
2257
2258     GLOBAL_STATE_CODE();
2259
2260     if (bs) {
2261         bdrv_remove_aio_context_notifier(bs, attached_aio_context,
2262                                          detach_aio_context, opaque);
2263     }
2264
2265     QLIST_FOREACH(notifier, &blk->aio_notifiers, list) {
2266         if (notifier->attached_aio_context == attached_aio_context &&
2267             notifier->detach_aio_context == detach_aio_context &&
2268             notifier->opaque == opaque) {
2269             QLIST_REMOVE(notifier, list);
2270             g_free(notifier);
2271             return;
2272         }
2273     }
2274
2275     abort();
2276 }
2277
2278 void blk_add_remove_bs_notifier(BlockBackend *blk, Notifier *notify)
2279 {
2280     GLOBAL_STATE_CODE();
2281     notifier_list_add(&blk->remove_bs_notifiers, notify);
2282 }
2283
2284 void blk_add_insert_bs_notifier(BlockBackend *blk, Notifier *notify)
2285 {
2286     GLOBAL_STATE_CODE();
2287     notifier_list_add(&blk->insert_bs_notifiers, notify);
2288 }
2289
2290 void blk_io_plug(BlockBackend *blk)
2291 {
2292     BlockDriverState *bs = blk_bs(blk);
2293     IO_CODE();
2294
2295     if (bs) {
2296         bdrv_io_plug(bs);
2297     }
2298 }
2299
2300 void blk_io_unplug(BlockBackend *blk)
2301 {
2302     BlockDriverState *bs = blk_bs(blk);
2303     IO_CODE();
2304
2305     if (bs) {
2306         bdrv_io_unplug(bs);
2307     }
2308 }
2309
2310 BlockAcctStats *blk_get_stats(BlockBackend *blk)
2311 {
2312     IO_CODE();
2313     return &blk->stats;
2314 }
2315
2316 void *blk_aio_get(const AIOCBInfo *aiocb_info, BlockBackend *blk,
2317                   BlockCompletionFunc *cb, void *opaque)
2318 {
2319     IO_CODE();
2320     return qemu_aio_get(aiocb_info, blk_bs(blk), cb, opaque);
2321 }
2322
2323 int coroutine_fn blk_co_pwrite_zeroes(BlockBackend *blk, int64_t offset,
2324                                       int64_t bytes, BdrvRequestFlags flags)
2325 {
2326     IO_OR_GS_CODE();
2327     return blk_co_pwritev(blk, offset, bytes, NULL,
2328                           flags | BDRV_REQ_ZERO_WRITE);
2329 }
2330
2331 int coroutine_fn blk_co_pwrite_compressed(BlockBackend *blk, int64_t offset,
2332                                           int64_t bytes, const void *buf)
2333 {
2334     QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
2335     IO_OR_GS_CODE();
2336     return blk_co_pwritev_part(blk, offset, bytes, &qiov, 0,
2337                                BDRV_REQ_WRITE_COMPRESSED);
2338 }
2339
2340 int coroutine_fn blk_co_truncate(BlockBackend *blk, int64_t offset, bool exact,
2341                                  PreallocMode prealloc, BdrvRequestFlags flags,
2342                                  Error **errp)
2343 {
2344     IO_OR_GS_CODE();
2345     if (!blk_is_available(blk)) {
2346         error_setg(errp, "No medium inserted");
2347         return -ENOMEDIUM;
2348     }
2349
2350     return bdrv_co_truncate(blk->root, offset, exact, prealloc, flags, errp);
2351 }
2352
2353 int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
2354                      int64_t pos, int size)
2355 {
2356     int ret;
2357     GLOBAL_STATE_CODE();
2358
2359     if (!blk_is_available(blk)) {
2360         return -ENOMEDIUM;
2361     }
2362
2363     ret = bdrv_save_vmstate(blk_bs(blk), buf, pos, size);
2364     if (ret < 0) {
2365         return ret;
2366     }
2367
2368     if (ret == size && !blk->enable_write_cache) {
2369         ret = bdrv_flush(blk_bs(blk));
2370     }
2371
2372     return ret < 0 ? ret : size;
2373 }
2374
2375 int blk_load_vmstate(BlockBackend *blk, uint8_t *buf, int64_t pos, int size)
2376 {
2377     GLOBAL_STATE_CODE();
2378     if (!blk_is_available(blk)) {
2379         return -ENOMEDIUM;
2380     }
2381
2382     return bdrv_load_vmstate(blk_bs(blk), buf, pos, size);
2383 }
2384
2385 int blk_probe_blocksizes(BlockBackend *blk, BlockSizes *bsz)
2386 {
2387     GLOBAL_STATE_CODE();
2388     if (!blk_is_available(blk)) {
2389         return -ENOMEDIUM;
2390     }
2391
2392     return bdrv_probe_blocksizes(blk_bs(blk), bsz);
2393 }
2394
2395 int blk_probe_geometry(BlockBackend *blk, HDGeometry *geo)
2396 {
2397     GLOBAL_STATE_CODE();
2398     if (!blk_is_available(blk)) {
2399         return -ENOMEDIUM;
2400     }
2401
2402     return bdrv_probe_geometry(blk_bs(blk), geo);
2403 }
2404
2405 /*
2406  * Updates the BlockBackendRootState object with data from the currently
2407  * attached BlockDriverState.
2408  */
2409 void blk_update_root_state(BlockBackend *blk)
2410 {
2411     GLOBAL_STATE_CODE();
2412     assert(blk->root);
2413
2414     blk->root_state.open_flags    = blk->root->bs->open_flags;
2415     blk->root_state.detect_zeroes = blk->root->bs->detect_zeroes;
2416 }
2417
2418 /*
2419  * Returns the detect-zeroes setting to be used for bdrv_open() of a
2420  * BlockDriverState which is supposed to inherit the root state.
2421  */
2422 bool blk_get_detect_zeroes_from_root_state(BlockBackend *blk)
2423 {
2424     GLOBAL_STATE_CODE();
2425     return blk->root_state.detect_zeroes;
2426 }
2427
2428 /*
2429  * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
2430  * supposed to inherit the root state.
2431  */
2432 int blk_get_open_flags_from_root_state(BlockBackend *blk)
2433 {
2434     GLOBAL_STATE_CODE();
2435     return blk->root_state.open_flags;
2436 }
2437
2438 BlockBackendRootState *blk_get_root_state(BlockBackend *blk)
2439 {
2440     GLOBAL_STATE_CODE();
2441     return &blk->root_state;
2442 }
2443
2444 int blk_commit_all(void)
2445 {
2446     BlockBackend *blk = NULL;
2447     GLOBAL_STATE_CODE();
2448
2449     while ((blk = blk_all_next(blk)) != NULL) {
2450         AioContext *aio_context = blk_get_aio_context(blk);
2451         BlockDriverState *unfiltered_bs = bdrv_skip_filters(blk_bs(blk));
2452
2453         aio_context_acquire(aio_context);
2454         if (blk_is_inserted(blk) && bdrv_cow_child(unfiltered_bs)) {
2455             int ret;
2456
2457             ret = bdrv_commit(unfiltered_bs);
2458             if (ret < 0) {
2459                 aio_context_release(aio_context);
2460                 return ret;
2461             }
2462         }
2463         aio_context_release(aio_context);
2464     }
2465     return 0;
2466 }
2467
2468
2469 /* throttling disk I/O limits */
2470 void blk_set_io_limits(BlockBackend *blk, ThrottleConfig *cfg)
2471 {
2472     GLOBAL_STATE_CODE();
2473     throttle_group_config(&blk->public.throttle_group_member, cfg);
2474 }
2475
2476 void blk_io_limits_disable(BlockBackend *blk)
2477 {
2478     BlockDriverState *bs = blk_bs(blk);
2479     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
2480     assert(tgm->throttle_state);
2481     GLOBAL_STATE_CODE();
2482     if (bs) {
2483         bdrv_ref(bs);
2484         bdrv_drained_begin(bs);
2485     }
2486     throttle_group_unregister_tgm(tgm);
2487     if (bs) {
2488         bdrv_drained_end(bs);
2489         bdrv_unref(bs);
2490     }
2491 }
2492
2493 /* should be called before blk_set_io_limits if a limit is set */
2494 void blk_io_limits_enable(BlockBackend *blk, const char *group)
2495 {
2496     assert(!blk->public.throttle_group_member.throttle_state);
2497     GLOBAL_STATE_CODE();
2498     throttle_group_register_tgm(&blk->public.throttle_group_member,
2499                                 group, blk_get_aio_context(blk));
2500 }
2501
2502 void blk_io_limits_update_group(BlockBackend *blk, const char *group)
2503 {
2504     GLOBAL_STATE_CODE();
2505     /* this BB is not part of any group */
2506     if (!blk->public.throttle_group_member.throttle_state) {
2507         return;
2508     }
2509
2510     /* this BB is a part of the same group than the one we want */
2511     if (!g_strcmp0(throttle_group_get_name(&blk->public.throttle_group_member),
2512                 group)) {
2513         return;
2514     }
2515
2516     /* need to change the group this bs belong to */
2517     blk_io_limits_disable(blk);
2518     blk_io_limits_enable(blk, group);
2519 }
2520
2521 static void blk_root_drained_begin(BdrvChild *child)
2522 {
2523     BlockBackend *blk = child->opaque;
2524     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
2525
2526     if (++blk->quiesce_counter == 1) {
2527         if (blk->dev_ops && blk->dev_ops->drained_begin) {
2528             blk->dev_ops->drained_begin(blk->dev_opaque);
2529         }
2530     }
2531
2532     /* Note that blk->root may not be accessible here yet if we are just
2533      * attaching to a BlockDriverState that is drained. Use child instead. */
2534
2535     if (qatomic_fetch_inc(&tgm->io_limits_disabled) == 0) {
2536         throttle_group_restart_tgm(tgm);
2537     }
2538 }
2539
2540 static bool blk_root_drained_poll(BdrvChild *child)
2541 {
2542     BlockBackend *blk = child->opaque;
2543     bool busy = false;
2544     assert(blk->quiesce_counter);
2545
2546     if (blk->dev_ops && blk->dev_ops->drained_poll) {
2547         busy = blk->dev_ops->drained_poll(blk->dev_opaque);
2548     }
2549     return busy || !!blk->in_flight;
2550 }
2551
2552 static void blk_root_drained_end(BdrvChild *child, int *drained_end_counter)
2553 {
2554     BlockBackend *blk = child->opaque;
2555     assert(blk->quiesce_counter);
2556
2557     assert(blk->public.throttle_group_member.io_limits_disabled);
2558     qatomic_dec(&blk->public.throttle_group_member.io_limits_disabled);
2559
2560     if (--blk->quiesce_counter == 0) {
2561         if (blk->dev_ops && blk->dev_ops->drained_end) {
2562             blk->dev_ops->drained_end(blk->dev_opaque);
2563         }
2564         while (qemu_co_enter_next(&blk->queued_requests, NULL)) {
2565             /* Resume all queued requests */
2566         }
2567     }
2568 }
2569
2570 bool blk_register_buf(BlockBackend *blk, void *host, size_t size, Error **errp)
2571 {
2572     GLOBAL_STATE_CODE();
2573     return bdrv_register_buf(blk_bs(blk), host, size, errp);
2574 }
2575
2576 void blk_unregister_buf(BlockBackend *blk, void *host, size_t size)
2577 {
2578     GLOBAL_STATE_CODE();
2579     bdrv_unregister_buf(blk_bs(blk), host, size);
2580 }
2581
2582 int coroutine_fn blk_co_copy_range(BlockBackend *blk_in, int64_t off_in,
2583                                    BlockBackend *blk_out, int64_t off_out,
2584                                    int64_t bytes, BdrvRequestFlags read_flags,
2585                                    BdrvRequestFlags write_flags)
2586 {
2587     int r;
2588     IO_CODE();
2589
2590     r = blk_check_byte_request(blk_in, off_in, bytes);
2591     if (r) {
2592         return r;
2593     }
2594     r = blk_check_byte_request(blk_out, off_out, bytes);
2595     if (r) {
2596         return r;
2597     }
2598     return bdrv_co_copy_range(blk_in->root, off_in,
2599                               blk_out->root, off_out,
2600                               bytes, read_flags, write_flags);
2601 }
2602
2603 const BdrvChild *blk_root(BlockBackend *blk)
2604 {
2605     GLOBAL_STATE_CODE();
2606     return blk->root;
2607 }
2608
2609 int blk_make_empty(BlockBackend *blk, Error **errp)
2610 {
2611     GLOBAL_STATE_CODE();
2612     if (!blk_is_available(blk)) {
2613         error_setg(errp, "No medium inserted");
2614         return -ENOMEDIUM;
2615     }
2616
2617     return bdrv_make_empty(blk->root, errp);
2618 }