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/throttle-groups.h"
  18 #include "sysemu/blockdev.h"
  19 #include "sysemu/sysemu.h"
  20 #include "qapi-event.h"
  21 #include "qemu/id.h"
  22 #include "trace.h"
  23
  24 /* Number of coroutines to reserve per attached device model */
  25 #define COROUTINE_POOL_RESERVATION 64
  26
  27 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
  28
  29 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb);
  30
  31 struct BlockBackend {
  32     char *name;
  33     int refcnt;
  34     BdrvChild *root;
  35     DriveInfo *legacy_dinfo;    /* null unless created by drive_new() */
  36     QTAILQ_ENTRY(BlockBackend) link;         /* for block_backends */
  37     QTAILQ_ENTRY(BlockBackend) monitor_link; /* for monitor_block_backends */
  38     BlockBackendPublic public;
  39
  40     void *dev;                  /* attached device model, if any */
  41     bool legacy_dev;            /* true if dev is not a DeviceState */
  42     /* TODO change to DeviceState when all users are qdevified */
  43     const BlockDevOps *dev_ops;
  44     void *dev_opaque;
  45
  46     /* the block size for which the guest device expects atomicity */
  47     int guest_block_size;
  48
  49     /* If the BDS tree is removed, some of its options are stored here (which
  50      * can be used to restore those options in the new BDS on insert) */
  51     BlockBackendRootState root_state;
  52
  53     bool enable_write_cache;
  54
  55     /* I/O stats (display with "info blockstats"). */
  56     BlockAcctStats stats;
  57
  58     BlockdevOnError on_read_error, on_write_error;
  59     bool iostatus_enabled;
  60     BlockDeviceIoStatus iostatus;
  61
  62     uint64_t perm;
  63     uint64_t shared_perm;
  64
  65     bool allow_write_beyond_eof;
  66
  67     NotifierList remove_bs_notifiers, insert_bs_notifiers;
  68 };
  69
  70 typedef struct BlockBackendAIOCB {
  71     BlockAIOCB common;
  72     BlockBackend *blk;
  73     int ret;
  74 } BlockBackendAIOCB;
  75
  76 static const AIOCBInfo block_backend_aiocb_info = {
  77     .get_aio_context = blk_aiocb_get_aio_context,
  78     .aiocb_size = sizeof(BlockBackendAIOCB),
  79 };
  80
  81 static void drive_info_del(DriveInfo *dinfo);
  82 static BlockBackend *bdrv_first_blk(BlockDriverState *bs);
  83
  84 /* All BlockBackends */
  85 static QTAILQ_HEAD(, BlockBackend) block_backends =
  86     QTAILQ_HEAD_INITIALIZER(block_backends);
  87
  88 /* All BlockBackends referenced by the monitor and which are iterated through by
  89  * blk_next() */
  90 static QTAILQ_HEAD(, BlockBackend) monitor_block_backends =
  91     QTAILQ_HEAD_INITIALIZER(monitor_block_backends);
  92
  93 static void blk_root_inherit_options(int *child_flags, QDict *child_options,
  94                                      int parent_flags, QDict *parent_options)
  95 {
  96     /* We're not supposed to call this function for root nodes */
  97     abort();
  98 }
  99 static void blk_root_drained_begin(BdrvChild *child);
 100 static void blk_root_drained_end(BdrvChild *child);
 101
 102 static void blk_root_change_media(BdrvChild *child, bool load);
 103 static void blk_root_resize(BdrvChild *child);
 104
 105 static const char *blk_root_get_name(BdrvChild *child)
 106 {
 107     return blk_name(child->opaque);
 108 }
 109
 110 static const BdrvChildRole child_root = {
 111     .inherit_options    = blk_root_inherit_options,
 112
 113     .change_media       = blk_root_change_media,
 114     .resize             = blk_root_resize,
 115     .get_name           = blk_root_get_name,
 116
 117     .drained_begin      = blk_root_drained_begin,
 118     .drained_end        = blk_root_drained_end,
 119 };
 120
 121 /*
 122  * Create a new BlockBackend with a reference count of one.
 123  *
 124  * @perm is a bitmasks of BLK_PERM_* constants which describes the permissions
 125  * to request for a block driver node that is attached to this BlockBackend.
 126  * @shared_perm is a bitmask which describes which permissions may be granted
 127  * to other users of the attached node.
 128  * Both sets of permissions can be changed later using blk_set_perm().
 129  *
 130  * Return the new BlockBackend on success, null on failure.
 131  */
 132 BlockBackend *blk_new(uint64_t perm, uint64_t shared_perm)
 133 {
 134     BlockBackend *blk;
 135
 136     blk = g_new0(BlockBackend, 1);
 137     blk->refcnt = 1;
 138     blk->perm = perm;
 139     blk->shared_perm = shared_perm;
 140     blk_set_enable_write_cache(blk, true);
 141
 142     qemu_co_queue_init(&blk->public.throttled_reqs[0]);
 143     qemu_co_queue_init(&blk->public.throttled_reqs[1]);
 144
 145     notifier_list_init(&blk->remove_bs_notifiers);
 146     notifier_list_init(&blk->insert_bs_notifiers);
 147
 148     QTAILQ_INSERT_TAIL(&block_backends, blk, link);
 149     return blk;
 150 }
 151
 152 /*
 153  * Creates a new BlockBackend, opens a new BlockDriverState, and connects both.
 154  *
 155  * Just as with bdrv_open(), after having called this function the reference to
 156  * @options belongs to the block layer (even on failure).
 157  *
 158  * TODO: Remove @filename and @flags; it should be possible to specify a whole
 159  * BDS tree just by specifying the @options QDict (or @reference,
 160  * alternatively). At the time of adding this function, this is not possible,
 161  * though, so callers of this function have to be able to specify @filename and
 162  * @flags.
 163  */
 164 BlockBackend *blk_new_open(const char *filename, const char *reference,
 165                            QDict *options, int flags, Error **errp)
 166 {
 167     BlockBackend *blk;
 168     BlockDriverState *bs;
 169
 170     blk = blk_new(0, BLK_PERM_ALL);
 171     bs = bdrv_open(filename, reference, options, flags, errp);
 172     if (!bs) {
 173         blk_unref(blk);
 174         return NULL;
 175     }
 176
 177     /* FIXME Use real permissions */
 178     blk->root = bdrv_root_attach_child(bs, "root", &child_root,
 179                                        0, BLK_PERM_ALL, blk, &error_abort);
 180
 181     return blk;
 182 }
 183
 184 static void blk_delete(BlockBackend *blk)
 185 {
 186     assert(!blk->refcnt);
 187     assert(!blk->name);
 188     assert(!blk->dev);
 189     if (blk->root) {
 190         blk_remove_bs(blk);
 191     }
 192     assert(QLIST_EMPTY(&blk->remove_bs_notifiers.notifiers));
 193     assert(QLIST_EMPTY(&blk->insert_bs_notifiers.notifiers));
 194     QTAILQ_REMOVE(&block_backends, blk, link);
 195     drive_info_del(blk->legacy_dinfo);
 196     block_acct_cleanup(&blk->stats);
 197     g_free(blk);
 198 }
 199
 200 static void drive_info_del(DriveInfo *dinfo)
 201 {
 202     if (!dinfo) {
 203         return;
 204     }
 205     qemu_opts_del(dinfo->opts);
 206     g_free(dinfo->serial);
 207     g_free(dinfo);
 208 }
 209
 210 int blk_get_refcnt(BlockBackend *blk)
 211 {
 212     return blk ? blk->refcnt : 0;
 213 }
 214
 215 /*
 216  * Increment @blk's reference count.
 217  * @blk must not be null.
 218  */
 219 void blk_ref(BlockBackend *blk)
 220 {
 221     blk->refcnt++;
 222 }
 223
 224 /*
 225  * Decrement @blk's reference count.
 226  * If this drops it to zero, destroy @blk.
 227  * For convenience, do nothing if @blk is null.
 228  */
 229 void blk_unref(BlockBackend *blk)
 230 {
 231     if (blk) {
 232         assert(blk->refcnt > 0);
 233         if (!--blk->refcnt) {
 234             blk_delete(blk);
 235         }
 236     }
 237 }
 238
 239 /*
 240  * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
 241  * ones which are hidden (i.e. are not referenced by the monitor).
 242  */
 243 static BlockBackend *blk_all_next(BlockBackend *blk)
 244 {
 245     return blk ? QTAILQ_NEXT(blk, link)
 246                : QTAILQ_FIRST(&block_backends);
 247 }
 248
 249 void blk_remove_all_bs(void)
 250 {
 251     BlockBackend *blk = NULL;
 252
 253     while ((blk = blk_all_next(blk)) != NULL) {
 254         AioContext *ctx = blk_get_aio_context(blk);
 255
 256         aio_context_acquire(ctx);
 257         if (blk->root) {
 258             blk_remove_bs(blk);
 259         }
 260         aio_context_release(ctx);
 261     }
 262 }
 263
 264 /*
 265  * Return the monitor-owned BlockBackend after @blk.
 266  * If @blk is null, return the first one.
 267  * Else, return @blk's next sibling, which may be null.
 268  *
 269  * To iterate over all BlockBackends, do
 270  * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
 271  *     ...
 272  * }
 273  */
 274 BlockBackend *blk_next(BlockBackend *blk)
 275 {
 276     return blk ? QTAILQ_NEXT(blk, monitor_link)
 277                : QTAILQ_FIRST(&monitor_block_backends);
 278 }
 279
 280 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
 281  * the monitor or attached to a BlockBackend */
 282 BlockDriverState *bdrv_next(BdrvNextIterator *it)
 283 {
 284     BlockDriverState *bs;
 285
 286     /* First, return all root nodes of BlockBackends. In order to avoid
 287      * returning a BDS twice when multiple BBs refer to it, we only return it
 288      * if the BB is the first one in the parent list of the BDS. */
 289     if (it->phase == BDRV_NEXT_BACKEND_ROOTS) {
 290         do {
 291             it->blk = blk_all_next(it->blk);
 292             bs = it->blk ? blk_bs(it->blk) : NULL;
 293         } while (it->blk && (bs == NULL || bdrv_first_blk(bs) != it->blk));
 294
 295         if (bs) {
 296             return bs;
 297         }
 298         it->phase = BDRV_NEXT_MONITOR_OWNED;
 299     }
 300
 301     /* Then return the monitor-owned BDSes without a BB attached. Ignore all
 302      * BDSes that are attached to a BlockBackend here; they have been handled
 303      * by the above block already */
 304     do {
 305         it->bs = bdrv_next_monitor_owned(it->bs);
 306         bs = it->bs;
 307     } while (bs && bdrv_has_blk(bs));
 308
 309     return bs;
 310 }
 311
 312 BlockDriverState *bdrv_first(BdrvNextIterator *it)
 313 {
 314     *it = (BdrvNextIterator) {
 315         .phase = BDRV_NEXT_BACKEND_ROOTS,
 316     };
 317
 318     return bdrv_next(it);
 319 }
 320
 321 /*
 322  * Add a BlockBackend into the list of backends referenced by the monitor, with
 323  * the given @name acting as the handle for the monitor.
 324  * Strictly for use by blockdev.c.
 325  *
 326  * @name must not be null or empty.
 327  *
 328  * Returns true on success and false on failure. In the latter case, an Error
 329  * object is returned through @errp.
 330  */
 331 bool monitor_add_blk(BlockBackend *blk, const char *name, Error **errp)
 332 {
 333     assert(!blk->name);
 334     assert(name && name[0]);
 335
 336     if (!id_wellformed(name)) {
 337         error_setg(errp, "Invalid device name");
 338         return false;
 339     }
 340     if (blk_by_name(name)) {
 341         error_setg(errp, "Device with id '%s' already exists", name);
 342         return false;
 343     }
 344     if (bdrv_find_node(name)) {
 345         error_setg(errp,
 346                    "Device name '%s' conflicts with an existing node name",
 347                    name);
 348         return false;
 349     }
 350
 351     blk->name = g_strdup(name);
 352     QTAILQ_INSERT_TAIL(&monitor_block_backends, blk, monitor_link);
 353     return true;
 354 }
 355
 356 /*
 357  * Remove a BlockBackend from the list of backends referenced by the monitor.
 358  * Strictly for use by blockdev.c.
 359  */
 360 void monitor_remove_blk(BlockBackend *blk)
 361 {
 362     if (!blk->name) {
 363         return;
 364     }
 365
 366     QTAILQ_REMOVE(&monitor_block_backends, blk, monitor_link);
 367     g_free(blk->name);
 368     blk->name = NULL;
 369 }
 370
 371 /*
 372  * Return @blk's name, a non-null string.
 373  * Returns an empty string iff @blk is not referenced by the monitor.
 374  */
 375 const char *blk_name(BlockBackend *blk)
 376 {
 377     return blk->name ?: "";
 378 }
 379
 380 /*
 381  * Return the BlockBackend with name @name if it exists, else null.
 382  * @name must not be null.
 383  */
 384 BlockBackend *blk_by_name(const char *name)
 385 {
 386     BlockBackend *blk = NULL;
 387
 388     assert(name);
 389     while ((blk = blk_next(blk)) != NULL) {
 390         if (!strcmp(name, blk->name)) {
 391             return blk;
 392         }
 393     }
 394     return NULL;
 395 }
 396
 397 /*
 398  * Return the BlockDriverState attached to @blk if any, else null.
 399  */
 400 BlockDriverState *blk_bs(BlockBackend *blk)
 401 {
 402     return blk->root ? blk->root->bs : NULL;
 403 }
 404
 405 static BlockBackend *bdrv_first_blk(BlockDriverState *bs)
 406 {
 407     BdrvChild *child;
 408     QLIST_FOREACH(child, &bs->parents, next_parent) {
 409         if (child->role == &child_root) {
 410             return child->opaque;
 411         }
 412     }
 413
 414     return NULL;
 415 }
 416
 417 /*
 418  * Returns true if @bs has an associated BlockBackend.
 419  */
 420 bool bdrv_has_blk(BlockDriverState *bs)
 421 {
 422     return bdrv_first_blk(bs) != NULL;
 423 }
 424
 425 /*
 426  * Returns true if @bs has only BlockBackends as parents.
 427  */
 428 bool bdrv_is_root_node(BlockDriverState *bs)
 429 {
 430     BdrvChild *c;
 431
 432     QLIST_FOREACH(c, &bs->parents, next_parent) {
 433         if (c->role != &child_root) {
 434             return false;
 435         }
 436     }
 437
 438     return true;
 439 }
 440
 441 /*
 442  * Return @blk's DriveInfo if any, else null.
 443  */
 444 DriveInfo *blk_legacy_dinfo(BlockBackend *blk)
 445 {
 446     return blk->legacy_dinfo;
 447 }
 448
 449 /*
 450  * Set @blk's DriveInfo to @dinfo, and return it.
 451  * @blk must not have a DriveInfo set already.
 452  * No other BlockBackend may have the same DriveInfo set.
 453  */
 454 DriveInfo *blk_set_legacy_dinfo(BlockBackend *blk, DriveInfo *dinfo)
 455 {
 456     assert(!blk->legacy_dinfo);
 457     return blk->legacy_dinfo = dinfo;
 458 }
 459
 460 /*
 461  * Return the BlockBackend with DriveInfo @dinfo.
 462  * It must exist.
 463  */
 464 BlockBackend *blk_by_legacy_dinfo(DriveInfo *dinfo)
 465 {
 466     BlockBackend *blk = NULL;
 467
 468     while ((blk = blk_next(blk)) != NULL) {
 469         if (blk->legacy_dinfo == dinfo) {
 470             return blk;
 471         }
 472     }
 473     abort();
 474 }
 475
 476 /*
 477  * Returns a pointer to the publicly accessible fields of @blk.
 478  */
 479 BlockBackendPublic *blk_get_public(BlockBackend *blk)
 480 {
 481     return &blk->public;
 482 }
 483
 484 /*
 485  * Returns a BlockBackend given the associated @public fields.
 486  */
 487 BlockBackend *blk_by_public(BlockBackendPublic *public)
 488 {
 489     return container_of(public, BlockBackend, public);
 490 }
 491
 492 /*
 493  * Disassociates the currently associated BlockDriverState from @blk.
 494  */
 495 void blk_remove_bs(BlockBackend *blk)
 496 {
 497     notifier_list_notify(&blk->remove_bs_notifiers, blk);
 498     if (blk->public.throttle_state) {
 499         throttle_timers_detach_aio_context(&blk->public.throttle_timers);
 500     }
 501
 502     blk_update_root_state(blk);
 503
 504     bdrv_root_unref_child(blk->root);
 505     blk->root = NULL;
 506 }
 507
 508 /*
 509  * Associates a new BlockDriverState with @blk.
 510  */
 511 int blk_insert_bs(BlockBackend *blk, BlockDriverState *bs, Error **errp)
 512 {
 513     blk->root = bdrv_root_attach_child(bs, "root", &child_root,
 514                                        blk->perm, blk->shared_perm, blk, errp);
 515     if (blk->root == NULL) {
 516         return -EPERM;
 517     }
 518     bdrv_ref(bs);
 519
 520     notifier_list_notify(&blk->insert_bs_notifiers, blk);
 521     if (blk->public.throttle_state) {
 522         throttle_timers_attach_aio_context(
 523             &blk->public.throttle_timers, bdrv_get_aio_context(bs));
 524     }
 525
 526     return 0;
 527 }
 528
 529 /*
 530  * Sets the permission bitmasks that the user of the BlockBackend needs.
 531  */
 532 int blk_set_perm(BlockBackend *blk, uint64_t perm, uint64_t shared_perm,
 533                  Error **errp)
 534 {
 535     int ret;
 536
 537     if (blk->root) {
 538         ret = bdrv_child_try_set_perm(blk->root, perm, shared_perm, errp);
 539         if (ret < 0) {
 540             return ret;
 541         }
 542     }
 543
 544     blk->perm = perm;
 545     blk->shared_perm = shared_perm;
 546
 547     return 0;
 548 }
 549
 550 static int blk_do_attach_dev(BlockBackend *blk, void *dev)
 551 {
 552     if (blk->dev) {
 553         return -EBUSY;
 554     }
 555     blk_ref(blk);
 556     blk->dev = dev;
 557     blk->legacy_dev = false;
 558     blk_iostatus_reset(blk);
 559     return 0;
 560 }
 561
 562 /*
 563  * Attach device model @dev to @blk.
 564  * Return 0 on success, -EBUSY when a device model is attached already.
 565  */
 566 int blk_attach_dev(BlockBackend *blk, DeviceState *dev)
 567 {
 568     return blk_do_attach_dev(blk, dev);
 569 }
 570
 571 /*
 572  * Attach device model @dev to @blk.
 573  * @blk must not have a device model attached already.
 574  * TODO qdevified devices don't use this, remove when devices are qdevified
 575  */
 576 void blk_attach_dev_legacy(BlockBackend *blk, void *dev)
 577 {
 578     if (blk_do_attach_dev(blk, dev) < 0) {
 579         abort();
 580     }
 581     blk->legacy_dev = true;
 582 }
 583
 584 /*
 585  * Detach device model @dev from @blk.
 586  * @dev must be currently attached to @blk.
 587  */
 588 void blk_detach_dev(BlockBackend *blk, void *dev)
 589 /* TODO change to DeviceState *dev when all users are qdevified */
 590 {
 591     assert(blk->dev == dev);
 592     blk->dev = NULL;
 593     blk->dev_ops = NULL;
 594     blk->dev_opaque = NULL;
 595     blk->guest_block_size = 512;
 596     blk_set_perm(blk, 0, BLK_PERM_ALL, &error_abort);
 597     blk_unref(blk);
 598 }
 599
 600 /*
 601  * Return the device model attached to @blk if any, else null.
 602  */
 603 void *blk_get_attached_dev(BlockBackend *blk)
 604 /* TODO change to return DeviceState * when all users are qdevified */
 605 {
 606     return blk->dev;
 607 }
 608
 609 /* Return the qdev ID, or if no ID is assigned the QOM path, of the block
 610  * device attached to the BlockBackend. */
 611 static char *blk_get_attached_dev_id(BlockBackend *blk)
 612 {
 613     DeviceState *dev;
 614
 615     assert(!blk->legacy_dev);
 616     dev = blk->dev;
 617
 618     if (!dev) {
 619         return g_strdup("");
 620     } else if (dev->id) {
 621         return g_strdup(dev->id);
 622     }
 623     return object_get_canonical_path(OBJECT(dev));
 624 }
 625
 626 /*
 627  * Return the BlockBackend which has the device model @dev attached if it
 628  * exists, else null.
 629  *
 630  * @dev must not be null.
 631  */
 632 BlockBackend *blk_by_dev(void *dev)
 633 {
 634     BlockBackend *blk = NULL;
 635
 636     assert(dev != NULL);
 637     while ((blk = blk_all_next(blk)) != NULL) {
 638         if (blk->dev == dev) {
 639             return blk;
 640         }
 641     }
 642     return NULL;
 643 }
 644
 645 /*
 646  * Set @blk's device model callbacks to @ops.
 647  * @opaque is the opaque argument to pass to the callbacks.
 648  * This is for use by device models.
 649  */
 650 void blk_set_dev_ops(BlockBackend *blk, const BlockDevOps *ops,
 651                      void *opaque)
 652 {
 653     /* All drivers that use blk_set_dev_ops() are qdevified and we want to keep
 654      * it that way, so we can assume blk->dev is a DeviceState if blk->dev_ops
 655      * is set. */
 656     assert(!blk->legacy_dev);
 657
 658     blk->dev_ops = ops;
 659     blk->dev_opaque = opaque;
 660 }
 661
 662 /*
 663  * Notify @blk's attached device model of media change.
 664  * If @load is true, notify of media load.
 665  * Else, notify of media eject.
 666  * Also send DEVICE_TRAY_MOVED events as appropriate.
 667  */
 668 void blk_dev_change_media_cb(BlockBackend *blk, bool load)
 669 {
 670     if (blk->dev_ops && blk->dev_ops->change_media_cb) {
 671         bool tray_was_open, tray_is_open;
 672
 673         assert(!blk->legacy_dev);
 674
 675         tray_was_open = blk_dev_is_tray_open(blk);
 676         blk->dev_ops->change_media_cb(blk->dev_opaque, load);
 677         tray_is_open = blk_dev_is_tray_open(blk);
 678
 679         if (tray_was_open != tray_is_open) {
 680             char *id = blk_get_attached_dev_id(blk);
 681             qapi_event_send_device_tray_moved(blk_name(blk), id, tray_is_open,
 682                                               &error_abort);
 683             g_free(id);
 684         }
 685     }
 686 }
 687
 688 static void blk_root_change_media(BdrvChild *child, bool load)
 689 {
 690     blk_dev_change_media_cb(child->opaque, load);
 691 }
 692
 693 /*
 694  * Does @blk's attached device model have removable media?
 695  * %true if no device model is attached.
 696  */
 697 bool blk_dev_has_removable_media(BlockBackend *blk)
 698 {
 699     return !blk->dev || (blk->dev_ops && blk->dev_ops->change_media_cb);
 700 }
 701
 702 /*
 703  * Does @blk's attached device model have a tray?
 704  */
 705 bool blk_dev_has_tray(BlockBackend *blk)
 706 {
 707     return blk->dev_ops && blk->dev_ops->is_tray_open;
 708 }
 709
 710 /*
 711  * Notify @blk's attached device model of a media eject request.
 712  * If @force is true, the medium is about to be yanked out forcefully.
 713  */
 714 void blk_dev_eject_request(BlockBackend *blk, bool force)
 715 {
 716     if (blk->dev_ops && blk->dev_ops->eject_request_cb) {
 717         blk->dev_ops->eject_request_cb(blk->dev_opaque, force);
 718     }
 719 }
 720
 721 /*
 722  * Does @blk's attached device model have a tray, and is it open?
 723  */
 724 bool blk_dev_is_tray_open(BlockBackend *blk)
 725 {
 726     if (blk_dev_has_tray(blk)) {
 727         return blk->dev_ops->is_tray_open(blk->dev_opaque);
 728     }
 729     return false;
 730 }
 731
 732 /*
 733  * Does @blk's attached device model have the medium locked?
 734  * %false if the device model has no such lock.
 735  */
 736 bool blk_dev_is_medium_locked(BlockBackend *blk)
 737 {
 738     if (blk->dev_ops && blk->dev_ops->is_medium_locked) {
 739         return blk->dev_ops->is_medium_locked(blk->dev_opaque);
 740     }
 741     return false;
 742 }
 743
 744 /*
 745  * Notify @blk's attached device model of a backend size change.
 746  */
 747 static void blk_root_resize(BdrvChild *child)
 748 {
 749     BlockBackend *blk = child->opaque;
 750
 751     if (blk->dev_ops && blk->dev_ops->resize_cb) {
 752         blk->dev_ops->resize_cb(blk->dev_opaque);
 753     }
 754 }
 755
 756 void blk_iostatus_enable(BlockBackend *blk)
 757 {
 758     blk->iostatus_enabled = true;
 759     blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 760 }
 761
 762 /* The I/O status is only enabled if the drive explicitly
 763  * enables it _and_ the VM is configured to stop on errors */
 764 bool blk_iostatus_is_enabled(const BlockBackend *blk)
 765 {
 766     return (blk->iostatus_enabled &&
 767            (blk->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
 768             blk->on_write_error == BLOCKDEV_ON_ERROR_STOP   ||
 769             blk->on_read_error == BLOCKDEV_ON_ERROR_STOP));
 770 }
 771
 772 BlockDeviceIoStatus blk_iostatus(const BlockBackend *blk)
 773 {
 774     return blk->iostatus;
 775 }
 776
 777 void blk_iostatus_disable(BlockBackend *blk)
 778 {
 779     blk->iostatus_enabled = false;
 780 }
 781
 782 void blk_iostatus_reset(BlockBackend *blk)
 783 {
 784     if (blk_iostatus_is_enabled(blk)) {
 785         BlockDriverState *bs = blk_bs(blk);
 786         blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 787         if (bs && bs->job) {
 788             block_job_iostatus_reset(bs->job);
 789         }
 790     }
 791 }
 792
 793 void blk_iostatus_set_err(BlockBackend *blk, int error)
 794 {
 795     assert(blk_iostatus_is_enabled(blk));
 796     if (blk->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 797         blk->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 798                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 799     }
 800 }
 801
 802 void blk_set_allow_write_beyond_eof(BlockBackend *blk, bool allow)
 803 {
 804     blk->allow_write_beyond_eof = allow;
 805 }
 806
 807 static int blk_check_byte_request(BlockBackend *blk, int64_t offset,
 808                                   size_t size)
 809 {
 810     int64_t len;
 811
 812     if (size > INT_MAX) {
 813         return -EIO;
 814     }
 815
 816     if (!blk_is_available(blk)) {
 817         return -ENOMEDIUM;
 818     }
 819
 820     if (offset < 0) {
 821         return -EIO;
 822     }
 823
 824     if (!blk->allow_write_beyond_eof) {
 825         len = blk_getlength(blk);
 826         if (len < 0) {
 827             return len;
 828         }
 829
 830         if (offset > len || len - offset < size) {
 831             return -EIO;
 832         }
 833     }
 834
 835     return 0;
 836 }
 837
 838 int coroutine_fn blk_co_preadv(BlockBackend *blk, int64_t offset,
 839                                unsigned int bytes, QEMUIOVector *qiov,
 840                                BdrvRequestFlags flags)
 841 {
 842     int ret;
 843     BlockDriverState *bs = blk_bs(blk);
 844
 845     trace_blk_co_preadv(blk, bs, offset, bytes, flags);
 846
 847     ret = blk_check_byte_request(blk, offset, bytes);
 848     if (ret < 0) {
 849         return ret;
 850     }
 851
 852     bdrv_inc_in_flight(bs);
 853
 854     /* throttling disk I/O */
 855     if (blk->public.throttle_state) {
 856         throttle_group_co_io_limits_intercept(blk, bytes, false);
 857     }
 858
 859     ret = bdrv_co_preadv(blk->root, offset, bytes, qiov, flags);
 860     bdrv_dec_in_flight(bs);
 861     return ret;
 862 }
 863
 864 int coroutine_fn blk_co_pwritev(BlockBackend *blk, int64_t offset,
 865                                 unsigned int bytes, QEMUIOVector *qiov,
 866                                 BdrvRequestFlags flags)
 867 {
 868     int ret;
 869     BlockDriverState *bs = blk_bs(blk);
 870
 871     trace_blk_co_pwritev(blk, bs, offset, bytes, flags);
 872
 873     ret = blk_check_byte_request(blk, offset, bytes);
 874     if (ret < 0) {
 875         return ret;
 876     }
 877
 878     bdrv_inc_in_flight(bs);
 879
 880     /* throttling disk I/O */
 881     if (blk->public.throttle_state) {
 882         throttle_group_co_io_limits_intercept(blk, bytes, true);
 883     }
 884
 885     if (!blk->enable_write_cache) {
 886         flags |= BDRV_REQ_FUA;
 887     }
 888
 889     ret = bdrv_co_pwritev(blk->root, offset, bytes, qiov, flags);
 890     bdrv_dec_in_flight(bs);
 891     return ret;
 892 }
 893
 894 typedef struct BlkRwCo {
 895     BlockBackend *blk;
 896     int64_t offset;
 897     QEMUIOVector *qiov;
 898     int ret;
 899     BdrvRequestFlags flags;
 900 } BlkRwCo;
 901
 902 static void blk_read_entry(void *opaque)
 903 {
 904     BlkRwCo *rwco = opaque;
 905
 906     rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, rwco->qiov->size,
 907                               rwco->qiov, rwco->flags);
 908 }
 909
 910 static void blk_write_entry(void *opaque)
 911 {
 912     BlkRwCo *rwco = opaque;
 913
 914     rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, rwco->qiov->size,
 915                                rwco->qiov, rwco->flags);
 916 }
 917
 918 static int blk_prw(BlockBackend *blk, int64_t offset, uint8_t *buf,
 919                    int64_t bytes, CoroutineEntry co_entry,
 920                    BdrvRequestFlags flags)
 921 {
 922     QEMUIOVector qiov;
 923     struct iovec iov;
 924     BlkRwCo rwco;
 925
 926     iov = (struct iovec) {
 927         .iov_base = buf,
 928         .iov_len = bytes,
 929     };
 930     qemu_iovec_init_external(&qiov, &iov, 1);
 931
 932     rwco = (BlkRwCo) {
 933         .blk    = blk,
 934         .offset = offset,
 935         .qiov   = &qiov,
 936         .flags  = flags,
 937         .ret    = NOT_DONE,
 938     };
 939
 940     if (qemu_in_coroutine()) {
 941         /* Fast-path if already in coroutine context */
 942         co_entry(&rwco);
 943     } else {
 944         Coroutine *co = qemu_coroutine_create(co_entry, &rwco);
 945         qemu_coroutine_enter(co);
 946         BDRV_POLL_WHILE(blk_bs(blk), rwco.ret == NOT_DONE);
 947     }
 948
 949     return rwco.ret;
 950 }
 951
 952 int blk_pread_unthrottled(BlockBackend *blk, int64_t offset, uint8_t *buf,
 953                           int count)
 954 {
 955     int ret;
 956
 957     ret = blk_check_byte_request(blk, offset, count);
 958     if (ret < 0) {
 959         return ret;
 960     }
 961
 962     blk_root_drained_begin(blk->root);
 963     ret = blk_pread(blk, offset, buf, count);
 964     blk_root_drained_end(blk->root);
 965     return ret;
 966 }
 967
 968 int blk_pwrite_zeroes(BlockBackend *blk, int64_t offset,
 969                       int count, BdrvRequestFlags flags)
 970 {
 971     return blk_prw(blk, offset, NULL, count, blk_write_entry,
 972                    flags | BDRV_REQ_ZERO_WRITE);
 973 }
 974
 975 int blk_make_zero(BlockBackend *blk, BdrvRequestFlags flags)
 976 {
 977     return bdrv_make_zero(blk->root, flags);
 978 }
 979
 980 static void error_callback_bh(void *opaque)
 981 {
 982     struct BlockBackendAIOCB *acb = opaque;
 983
 984     bdrv_dec_in_flight(acb->common.bs);
 985     acb->common.cb(acb->common.opaque, acb->ret);
 986     qemu_aio_unref(acb);
 987 }
 988
 989 BlockAIOCB *blk_abort_aio_request(BlockBackend *blk,
 990                                   BlockCompletionFunc *cb,
 991                                   void *opaque, int ret)
 992 {
 993     struct BlockBackendAIOCB *acb;
 994
 995     bdrv_inc_in_flight(blk_bs(blk));
 996     acb = blk_aio_get(&block_backend_aiocb_info, blk, cb, opaque);
 997     acb->blk = blk;
 998     acb->ret = ret;
 999
1000     aio_bh_schedule_oneshot(blk_get_aio_context(blk), error_callback_bh, acb);
1001     return &acb->common;
1002 }
1003
1004 typedef struct BlkAioEmAIOCB {
1005     BlockAIOCB common;
1006     BlkRwCo rwco;
1007     int bytes;
1008     bool has_returned;
1009 } BlkAioEmAIOCB;
1010
1011 static const AIOCBInfo blk_aio_em_aiocb_info = {
1012     .aiocb_size         = sizeof(BlkAioEmAIOCB),
1013 };
1014
1015 static void blk_aio_complete(BlkAioEmAIOCB *acb)
1016 {
1017     if (acb->has_returned) {
1018         bdrv_dec_in_flight(acb->common.bs);
1019         acb->common.cb(acb->common.opaque, acb->rwco.ret);
1020         qemu_aio_unref(acb);
1021     }
1022 }
1023
1024 static void blk_aio_complete_bh(void *opaque)
1025 {
1026     BlkAioEmAIOCB *acb = opaque;
1027     assert(acb->has_returned);
1028     blk_aio_complete(acb);
1029 }
1030
1031 static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset, int bytes,
1032                                 QEMUIOVector *qiov, CoroutineEntry co_entry,
1033                                 BdrvRequestFlags flags,
1034                                 BlockCompletionFunc *cb, void *opaque)
1035 {
1036     BlkAioEmAIOCB *acb;
1037     Coroutine *co;
1038
1039     bdrv_inc_in_flight(blk_bs(blk));
1040     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1041     acb->rwco = (BlkRwCo) {
1042         .blk    = blk,
1043         .offset = offset,
1044         .qiov   = qiov,
1045         .flags  = flags,
1046         .ret    = NOT_DONE,
1047     };
1048     acb->bytes = bytes;
1049     acb->has_returned = false;
1050
1051     co = qemu_coroutine_create(co_entry, acb);
1052     qemu_coroutine_enter(co);
1053
1054     acb->has_returned = true;
1055     if (acb->rwco.ret != NOT_DONE) {
1056         aio_bh_schedule_oneshot(blk_get_aio_context(blk),
1057                                 blk_aio_complete_bh, acb);
1058     }
1059
1060     return &acb->common;
1061 }
1062
1063 static void blk_aio_read_entry(void *opaque)
1064 {
1065     BlkAioEmAIOCB *acb = opaque;
1066     BlkRwCo *rwco = &acb->rwco;
1067
1068     assert(rwco->qiov->size == acb->bytes);
1069     rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, acb->bytes,
1070                               rwco->qiov, rwco->flags);
1071     blk_aio_complete(acb);
1072 }
1073
1074 static void blk_aio_write_entry(void *opaque)
1075 {
1076     BlkAioEmAIOCB *acb = opaque;
1077     BlkRwCo *rwco = &acb->rwco;
1078
1079     assert(!rwco->qiov || rwco->qiov->size == acb->bytes);
1080     rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, acb->bytes,
1081                                rwco->qiov, rwco->flags);
1082     blk_aio_complete(acb);
1083 }
1084
1085 BlockAIOCB *blk_aio_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1086                                   int count, BdrvRequestFlags flags,
1087                                   BlockCompletionFunc *cb, void *opaque)
1088 {
1089     return blk_aio_prwv(blk, offset, count, NULL, blk_aio_write_entry,
1090                         flags | BDRV_REQ_ZERO_WRITE, cb, opaque);
1091 }
1092
1093 int blk_pread(BlockBackend *blk, int64_t offset, void *buf, int count)
1094 {
1095     int ret = blk_prw(blk, offset, buf, count, blk_read_entry, 0);
1096     if (ret < 0) {
1097         return ret;
1098     }
1099     return count;
1100 }
1101
1102 int blk_pwrite(BlockBackend *blk, int64_t offset, const void *buf, int count,
1103                BdrvRequestFlags flags)
1104 {
1105     int ret = blk_prw(blk, offset, (void *) buf, count, blk_write_entry,
1106                       flags);
1107     if (ret < 0) {
1108         return ret;
1109     }
1110     return count;
1111 }
1112
1113 int64_t blk_getlength(BlockBackend *blk)
1114 {
1115     if (!blk_is_available(blk)) {
1116         return -ENOMEDIUM;
1117     }
1118
1119     return bdrv_getlength(blk_bs(blk));
1120 }
1121
1122 void blk_get_geometry(BlockBackend *blk, uint64_t *nb_sectors_ptr)
1123 {
1124     if (!blk_bs(blk)) {
1125         *nb_sectors_ptr = 0;
1126     } else {
1127         bdrv_get_geometry(blk_bs(blk), nb_sectors_ptr);
1128     }
1129 }
1130
1131 int64_t blk_nb_sectors(BlockBackend *blk)
1132 {
1133     if (!blk_is_available(blk)) {
1134         return -ENOMEDIUM;
1135     }
1136
1137     return bdrv_nb_sectors(blk_bs(blk));
1138 }
1139
1140 BlockAIOCB *blk_aio_preadv(BlockBackend *blk, int64_t offset,
1141                            QEMUIOVector *qiov, BdrvRequestFlags flags,
1142                            BlockCompletionFunc *cb, void *opaque)
1143 {
1144     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1145                         blk_aio_read_entry, flags, cb, opaque);
1146 }
1147
1148 BlockAIOCB *blk_aio_pwritev(BlockBackend *blk, int64_t offset,
1149                             QEMUIOVector *qiov, BdrvRequestFlags flags,
1150                             BlockCompletionFunc *cb, void *opaque)
1151 {
1152     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1153                         blk_aio_write_entry, flags, cb, opaque);
1154 }
1155
1156 static void blk_aio_flush_entry(void *opaque)
1157 {
1158     BlkAioEmAIOCB *acb = opaque;
1159     BlkRwCo *rwco = &acb->rwco;
1160
1161     rwco->ret = blk_co_flush(rwco->blk);
1162     blk_aio_complete(acb);
1163 }
1164
1165 BlockAIOCB *blk_aio_flush(BlockBackend *blk,
1166                           BlockCompletionFunc *cb, void *opaque)
1167 {
1168     return blk_aio_prwv(blk, 0, 0, NULL, blk_aio_flush_entry, 0, cb, opaque);
1169 }
1170
1171 static void blk_aio_pdiscard_entry(void *opaque)
1172 {
1173     BlkAioEmAIOCB *acb = opaque;
1174     BlkRwCo *rwco = &acb->rwco;
1175
1176     rwco->ret = blk_co_pdiscard(rwco->blk, rwco->offset, acb->bytes);
1177     blk_aio_complete(acb);
1178 }
1179
1180 BlockAIOCB *blk_aio_pdiscard(BlockBackend *blk,
1181                              int64_t offset, int count,
1182                              BlockCompletionFunc *cb, void *opaque)
1183 {
1184     return blk_aio_prwv(blk, offset, count, NULL, blk_aio_pdiscard_entry, 0,
1185                         cb, opaque);
1186 }
1187
1188 void blk_aio_cancel(BlockAIOCB *acb)
1189 {
1190     bdrv_aio_cancel(acb);
1191 }
1192
1193 void blk_aio_cancel_async(BlockAIOCB *acb)
1194 {
1195     bdrv_aio_cancel_async(acb);
1196 }
1197
1198 int blk_co_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1199 {
1200     if (!blk_is_available(blk)) {
1201         return -ENOMEDIUM;
1202     }
1203
1204     return bdrv_co_ioctl(blk_bs(blk), req, buf);
1205 }
1206
1207 static void blk_ioctl_entry(void *opaque)
1208 {
1209     BlkRwCo *rwco = opaque;
1210     rwco->ret = blk_co_ioctl(rwco->blk, rwco->offset,
1211                              rwco->qiov->iov[0].iov_base);
1212 }
1213
1214 int blk_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1215 {
1216     return blk_prw(blk, req, buf, 0, blk_ioctl_entry, 0);
1217 }
1218
1219 static void blk_aio_ioctl_entry(void *opaque)
1220 {
1221     BlkAioEmAIOCB *acb = opaque;
1222     BlkRwCo *rwco = &acb->rwco;
1223
1224     rwco->ret = blk_co_ioctl(rwco->blk, rwco->offset,
1225                              rwco->qiov->iov[0].iov_base);
1226     blk_aio_complete(acb);
1227 }
1228
1229 BlockAIOCB *blk_aio_ioctl(BlockBackend *blk, unsigned long int req, void *buf,
1230                           BlockCompletionFunc *cb, void *opaque)
1231 {
1232     QEMUIOVector qiov;
1233     struct iovec iov;
1234
1235     iov = (struct iovec) {
1236         .iov_base = buf,
1237         .iov_len = 0,
1238     };
1239     qemu_iovec_init_external(&qiov, &iov, 1);
1240
1241     return blk_aio_prwv(blk, req, 0, &qiov, blk_aio_ioctl_entry, 0, cb, opaque);
1242 }
1243
1244 int blk_co_pdiscard(BlockBackend *blk, int64_t offset, int count)
1245 {
1246     int ret = blk_check_byte_request(blk, offset, count);
1247     if (ret < 0) {
1248         return ret;
1249     }
1250
1251     return bdrv_co_pdiscard(blk_bs(blk), offset, count);
1252 }
1253
1254 int blk_co_flush(BlockBackend *blk)
1255 {
1256     if (!blk_is_available(blk)) {
1257         return -ENOMEDIUM;
1258     }
1259
1260     return bdrv_co_flush(blk_bs(blk));
1261 }
1262
1263 static void blk_flush_entry(void *opaque)
1264 {
1265     BlkRwCo *rwco = opaque;
1266     rwco->ret = blk_co_flush(rwco->blk);
1267 }
1268
1269 int blk_flush(BlockBackend *blk)
1270 {
1271     return blk_prw(blk, 0, NULL, 0, blk_flush_entry, 0);
1272 }
1273
1274 void blk_drain(BlockBackend *blk)
1275 {
1276     if (blk_bs(blk)) {
1277         bdrv_drain(blk_bs(blk));
1278     }
1279 }
1280
1281 void blk_drain_all(void)
1282 {
1283     bdrv_drain_all();
1284 }
1285
1286 void blk_set_on_error(BlockBackend *blk, BlockdevOnError on_read_error,
1287                       BlockdevOnError on_write_error)
1288 {
1289     blk->on_read_error = on_read_error;
1290     blk->on_write_error = on_write_error;
1291 }
1292
1293 BlockdevOnError blk_get_on_error(BlockBackend *blk, bool is_read)
1294 {
1295     return is_read ? blk->on_read_error : blk->on_write_error;
1296 }
1297
1298 BlockErrorAction blk_get_error_action(BlockBackend *blk, bool is_read,
1299                                       int error)
1300 {
1301     BlockdevOnError on_err = blk_get_on_error(blk, is_read);
1302
1303     switch (on_err) {
1304     case BLOCKDEV_ON_ERROR_ENOSPC:
1305         return (error == ENOSPC) ?
1306                BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
1307     case BLOCKDEV_ON_ERROR_STOP:
1308         return BLOCK_ERROR_ACTION_STOP;
1309     case BLOCKDEV_ON_ERROR_REPORT:
1310         return BLOCK_ERROR_ACTION_REPORT;
1311     case BLOCKDEV_ON_ERROR_IGNORE:
1312         return BLOCK_ERROR_ACTION_IGNORE;
1313     case BLOCKDEV_ON_ERROR_AUTO:
1314     default:
1315         abort();
1316     }
1317 }
1318
1319 static void send_qmp_error_event(BlockBackend *blk,
1320                                  BlockErrorAction action,
1321                                  bool is_read, int error)
1322 {
1323     IoOperationType optype;
1324
1325     optype = is_read ? IO_OPERATION_TYPE_READ : IO_OPERATION_TYPE_WRITE;
1326     qapi_event_send_block_io_error(blk_name(blk),
1327                                    bdrv_get_node_name(blk_bs(blk)), optype,
1328                                    action, blk_iostatus_is_enabled(blk),
1329                                    error == ENOSPC, strerror(error),
1330                                    &error_abort);
1331 }
1332
1333 /* This is done by device models because, while the block layer knows
1334  * about the error, it does not know whether an operation comes from
1335  * the device or the block layer (from a job, for example).
1336  */
1337 void blk_error_action(BlockBackend *blk, BlockErrorAction action,
1338                       bool is_read, int error)
1339 {
1340     assert(error >= 0);
1341
1342     if (action == BLOCK_ERROR_ACTION_STOP) {
1343         /* First set the iostatus, so that "info block" returns an iostatus
1344          * that matches the events raised so far (an additional error iostatus
1345          * is fine, but not a lost one).
1346          */
1347         blk_iostatus_set_err(blk, error);
1348
1349         /* Then raise the request to stop the VM and the event.
1350          * qemu_system_vmstop_request_prepare has two effects.  First,
1351          * it ensures that the STOP event always comes after the
1352          * BLOCK_IO_ERROR event.  Second, it ensures that even if management
1353          * can observe the STOP event and do a "cont" before the STOP
1354          * event is issued, the VM will not stop.  In this case, vm_start()
1355          * also ensures that the STOP/RESUME pair of events is emitted.
1356          */
1357         qemu_system_vmstop_request_prepare();
1358         send_qmp_error_event(blk, action, is_read, error);
1359         qemu_system_vmstop_request(RUN_STATE_IO_ERROR);
1360     } else {
1361         send_qmp_error_event(blk, action, is_read, error);
1362     }
1363 }
1364
1365 int blk_is_read_only(BlockBackend *blk)
1366 {
1367     BlockDriverState *bs = blk_bs(blk);
1368
1369     if (bs) {
1370         return bdrv_is_read_only(bs);
1371     } else {
1372         return blk->root_state.read_only;
1373     }
1374 }
1375
1376 int blk_is_sg(BlockBackend *blk)
1377 {
1378     BlockDriverState *bs = blk_bs(blk);
1379
1380     if (!bs) {
1381         return 0;
1382     }
1383
1384     return bdrv_is_sg(bs);
1385 }
1386
1387 int blk_enable_write_cache(BlockBackend *blk)
1388 {
1389     return blk->enable_write_cache;
1390 }
1391
1392 void blk_set_enable_write_cache(BlockBackend *blk, bool wce)
1393 {
1394     blk->enable_write_cache = wce;
1395 }
1396
1397 void blk_invalidate_cache(BlockBackend *blk, Error **errp)
1398 {
1399     BlockDriverState *bs = blk_bs(blk);
1400
1401     if (!bs) {
1402         error_setg(errp, "Device '%s' has no medium", blk->name);
1403         return;
1404     }
1405
1406     bdrv_invalidate_cache(bs, errp);
1407 }
1408
1409 bool blk_is_inserted(BlockBackend *blk)
1410 {
1411     BlockDriverState *bs = blk_bs(blk);
1412
1413     return bs && bdrv_is_inserted(bs);
1414 }
1415
1416 bool blk_is_available(BlockBackend *blk)
1417 {
1418     return blk_is_inserted(blk) && !blk_dev_is_tray_open(blk);
1419 }
1420
1421 void blk_lock_medium(BlockBackend *blk, bool locked)
1422 {
1423     BlockDriverState *bs = blk_bs(blk);
1424
1425     if (bs) {
1426         bdrv_lock_medium(bs, locked);
1427     }
1428 }
1429
1430 void blk_eject(BlockBackend *blk, bool eject_flag)
1431 {
1432     BlockDriverState *bs = blk_bs(blk);
1433     char *id;
1434
1435     /* blk_eject is only called by qdevified devices */
1436     assert(!blk->legacy_dev);
1437
1438     if (bs) {
1439         bdrv_eject(bs, eject_flag);
1440     }
1441
1442     /* Whether or not we ejected on the backend,
1443      * the frontend experienced a tray event. */
1444     id = blk_get_attached_dev_id(blk);
1445     qapi_event_send_device_tray_moved(blk_name(blk), id,
1446                                       eject_flag, &error_abort);
1447     g_free(id);
1448 }
1449
1450 int blk_get_flags(BlockBackend *blk)
1451 {
1452     BlockDriverState *bs = blk_bs(blk);
1453
1454     if (bs) {
1455         return bdrv_get_flags(bs);
1456     } else {
1457         return blk->root_state.open_flags;
1458     }
1459 }
1460
1461 /* Returns the maximum transfer length, in bytes; guaranteed nonzero */
1462 uint32_t blk_get_max_transfer(BlockBackend *blk)
1463 {
1464     BlockDriverState *bs = blk_bs(blk);
1465     uint32_t max = 0;
1466
1467     if (bs) {
1468         max = bs->bl.max_transfer;
1469     }
1470     return MIN_NON_ZERO(max, INT_MAX);
1471 }
1472
1473 int blk_get_max_iov(BlockBackend *blk)
1474 {
1475     return blk->root->bs->bl.max_iov;
1476 }
1477
1478 void blk_set_guest_block_size(BlockBackend *blk, int align)
1479 {
1480     blk->guest_block_size = align;
1481 }
1482
1483 void *blk_try_blockalign(BlockBackend *blk, size_t size)
1484 {
1485     return qemu_try_blockalign(blk ? blk_bs(blk) : NULL, size);
1486 }
1487
1488 void *blk_blockalign(BlockBackend *blk, size_t size)
1489 {
1490     return qemu_blockalign(blk ? blk_bs(blk) : NULL, size);
1491 }
1492
1493 bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp)
1494 {
1495     BlockDriverState *bs = blk_bs(blk);
1496
1497     if (!bs) {
1498         return false;
1499     }
1500
1501     return bdrv_op_is_blocked(bs, op, errp);
1502 }
1503
1504 void blk_op_unblock(BlockBackend *blk, BlockOpType op, Error *reason)
1505 {
1506     BlockDriverState *bs = blk_bs(blk);
1507
1508     if (bs) {
1509         bdrv_op_unblock(bs, op, reason);
1510     }
1511 }
1512
1513 void blk_op_block_all(BlockBackend *blk, Error *reason)
1514 {
1515     BlockDriverState *bs = blk_bs(blk);
1516
1517     if (bs) {
1518         bdrv_op_block_all(bs, reason);
1519     }
1520 }
1521
1522 void blk_op_unblock_all(BlockBackend *blk, Error *reason)
1523 {
1524     BlockDriverState *bs = blk_bs(blk);
1525
1526     if (bs) {
1527         bdrv_op_unblock_all(bs, reason);
1528     }
1529 }
1530
1531 AioContext *blk_get_aio_context(BlockBackend *blk)
1532 {
1533     BlockDriverState *bs = blk_bs(blk);
1534
1535     if (bs) {
1536         return bdrv_get_aio_context(bs);
1537     } else {
1538         return qemu_get_aio_context();
1539     }
1540 }
1541
1542 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb)
1543 {
1544     BlockBackendAIOCB *blk_acb = DO_UPCAST(BlockBackendAIOCB, common, acb);
1545     return blk_get_aio_context(blk_acb->blk);
1546 }
1547
1548 void blk_set_aio_context(BlockBackend *blk, AioContext *new_context)
1549 {
1550     BlockDriverState *bs = blk_bs(blk);
1551
1552     if (bs) {
1553         if (blk->public.throttle_state) {
1554             throttle_timers_detach_aio_context(&blk->public.throttle_timers);
1555         }
1556         bdrv_set_aio_context(bs, new_context);
1557         if (blk->public.throttle_state) {
1558             throttle_timers_attach_aio_context(&blk->public.throttle_timers,
1559                                                new_context);
1560         }
1561     }
1562 }
1563
1564 void blk_add_aio_context_notifier(BlockBackend *blk,
1565         void (*attached_aio_context)(AioContext *new_context, void *opaque),
1566         void (*detach_aio_context)(void *opaque), void *opaque)
1567 {
1568     BlockDriverState *bs = blk_bs(blk);
1569
1570     if (bs) {
1571         bdrv_add_aio_context_notifier(bs, attached_aio_context,
1572                                       detach_aio_context, opaque);
1573     }
1574 }
1575
1576 void blk_remove_aio_context_notifier(BlockBackend *blk,
1577                                      void (*attached_aio_context)(AioContext *,
1578                                                                   void *),
1579                                      void (*detach_aio_context)(void *),
1580                                      void *opaque)
1581 {
1582     BlockDriverState *bs = blk_bs(blk);
1583
1584     if (bs) {
1585         bdrv_remove_aio_context_notifier(bs, attached_aio_context,
1586                                          detach_aio_context, opaque);
1587     }
1588 }
1589
1590 void blk_add_remove_bs_notifier(BlockBackend *blk, Notifier *notify)
1591 {
1592     notifier_list_add(&blk->remove_bs_notifiers, notify);
1593 }
1594
1595 void blk_add_insert_bs_notifier(BlockBackend *blk, Notifier *notify)
1596 {
1597     notifier_list_add(&blk->insert_bs_notifiers, notify);
1598 }
1599
1600 void blk_io_plug(BlockBackend *blk)
1601 {
1602     BlockDriverState *bs = blk_bs(blk);
1603
1604     if (bs) {
1605         bdrv_io_plug(bs);
1606     }
1607 }
1608
1609 void blk_io_unplug(BlockBackend *blk)
1610 {
1611     BlockDriverState *bs = blk_bs(blk);
1612
1613     if (bs) {
1614         bdrv_io_unplug(bs);
1615     }
1616 }
1617
1618 BlockAcctStats *blk_get_stats(BlockBackend *blk)
1619 {
1620     return &blk->stats;
1621 }
1622
1623 void *blk_aio_get(const AIOCBInfo *aiocb_info, BlockBackend *blk,
1624                   BlockCompletionFunc *cb, void *opaque)
1625 {
1626     return qemu_aio_get(aiocb_info, blk_bs(blk), cb, opaque);
1627 }
1628
1629 int coroutine_fn blk_co_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1630                                       int count, BdrvRequestFlags flags)
1631 {
1632     return blk_co_pwritev(blk, offset, count, NULL,
1633                           flags | BDRV_REQ_ZERO_WRITE);
1634 }
1635
1636 int blk_pwrite_compressed(BlockBackend *blk, int64_t offset, const void *buf,
1637                           int count)
1638 {
1639     return blk_prw(blk, offset, (void *) buf, count, blk_write_entry,
1640                    BDRV_REQ_WRITE_COMPRESSED);
1641 }
1642
1643 int blk_truncate(BlockBackend *blk, int64_t offset)
1644 {
1645     if (!blk_is_available(blk)) {
1646         return -ENOMEDIUM;
1647     }
1648
1649     return bdrv_truncate(blk->root, offset);
1650 }
1651
1652 static void blk_pdiscard_entry(void *opaque)
1653 {
1654     BlkRwCo *rwco = opaque;
1655     rwco->ret = blk_co_pdiscard(rwco->blk, rwco->offset, rwco->qiov->size);
1656 }
1657
1658 int blk_pdiscard(BlockBackend *blk, int64_t offset, int count)
1659 {
1660     return blk_prw(blk, offset, NULL, count, blk_pdiscard_entry, 0);
1661 }
1662
1663 int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
1664                      int64_t pos, int size)
1665 {
1666     int ret;
1667
1668     if (!blk_is_available(blk)) {
1669         return -ENOMEDIUM;
1670     }
1671
1672     ret = bdrv_save_vmstate(blk_bs(blk), buf, pos, size);
1673     if (ret < 0) {
1674         return ret;
1675     }
1676
1677     if (ret == size && !blk->enable_write_cache) {
1678         ret = bdrv_flush(blk_bs(blk));
1679     }
1680
1681     return ret < 0 ? ret : size;
1682 }
1683
1684 int blk_load_vmstate(BlockBackend *blk, uint8_t *buf, int64_t pos, int size)
1685 {
1686     if (!blk_is_available(blk)) {
1687         return -ENOMEDIUM;
1688     }
1689
1690     return bdrv_load_vmstate(blk_bs(blk), buf, pos, size);
1691 }
1692
1693 int blk_probe_blocksizes(BlockBackend *blk, BlockSizes *bsz)
1694 {
1695     if (!blk_is_available(blk)) {
1696         return -ENOMEDIUM;
1697     }
1698
1699     return bdrv_probe_blocksizes(blk_bs(blk), bsz);
1700 }
1701
1702 int blk_probe_geometry(BlockBackend *blk, HDGeometry *geo)
1703 {
1704     if (!blk_is_available(blk)) {
1705         return -ENOMEDIUM;
1706     }
1707
1708     return bdrv_probe_geometry(blk_bs(blk), geo);
1709 }
1710
1711 /*
1712  * Updates the BlockBackendRootState object with data from the currently
1713  * attached BlockDriverState.
1714  */
1715 void blk_update_root_state(BlockBackend *blk)
1716 {
1717     assert(blk->root);
1718
1719     blk->root_state.open_flags    = blk->root->bs->open_flags;
1720     blk->root_state.read_only     = blk->root->bs->read_only;
1721     blk->root_state.detect_zeroes = blk->root->bs->detect_zeroes;
1722 }
1723
1724 /*
1725  * Returns the detect-zeroes setting to be used for bdrv_open() of a
1726  * BlockDriverState which is supposed to inherit the root state.
1727  */
1728 bool blk_get_detect_zeroes_from_root_state(BlockBackend *blk)
1729 {
1730     return blk->root_state.detect_zeroes;
1731 }
1732
1733 /*
1734  * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
1735  * supposed to inherit the root state.
1736  */
1737 int blk_get_open_flags_from_root_state(BlockBackend *blk)
1738 {
1739     int bs_flags;
1740
1741     bs_flags = blk->root_state.read_only ? 0 : BDRV_O_RDWR;
1742     bs_flags |= blk->root_state.open_flags & ~BDRV_O_RDWR;
1743
1744     return bs_flags;
1745 }
1746
1747 BlockBackendRootState *blk_get_root_state(BlockBackend *blk)
1748 {
1749     return &blk->root_state;
1750 }
1751
1752 int blk_commit_all(void)
1753 {
1754     BlockBackend *blk = NULL;
1755
1756     while ((blk = blk_all_next(blk)) != NULL) {
1757         AioContext *aio_context = blk_get_aio_context(blk);
1758
1759         aio_context_acquire(aio_context);
1760         if (blk_is_inserted(blk) && blk->root->bs->backing) {
1761             int ret = bdrv_commit(blk->root->bs);
1762             if (ret < 0) {
1763                 aio_context_release(aio_context);
1764                 return ret;
1765             }
1766         }
1767         aio_context_release(aio_context);
1768     }
1769     return 0;
1770 }
1771
1772
1773 /* throttling disk I/O limits */
1774 void blk_set_io_limits(BlockBackend *blk, ThrottleConfig *cfg)
1775 {
1776     throttle_group_config(blk, cfg);
1777 }
1778
1779 void blk_io_limits_disable(BlockBackend *blk)
1780 {
1781     assert(blk->public.throttle_state);
1782     bdrv_drained_begin(blk_bs(blk));
1783     throttle_group_unregister_blk(blk);
1784     bdrv_drained_end(blk_bs(blk));
1785 }
1786
1787 /* should be called before blk_set_io_limits if a limit is set */
1788 void blk_io_limits_enable(BlockBackend *blk, const char *group)
1789 {
1790     assert(!blk->public.throttle_state);
1791     throttle_group_register_blk(blk, group);
1792 }
1793
1794 void blk_io_limits_update_group(BlockBackend *blk, const char *group)
1795 {
1796     /* this BB is not part of any group */
1797     if (!blk->public.throttle_state) {
1798         return;
1799     }
1800
1801     /* this BB is a part of the same group than the one we want */
1802     if (!g_strcmp0(throttle_group_get_name(blk), group)) {
1803         return;
1804     }
1805
1806     /* need to change the group this bs belong to */
1807     blk_io_limits_disable(blk);
1808     blk_io_limits_enable(blk, group);
1809 }
1810
1811 static void blk_root_drained_begin(BdrvChild *child)
1812 {
1813     BlockBackend *blk = child->opaque;
1814
1815     /* Note that blk->root may not be accessible here yet if we are just
1816      * attaching to a BlockDriverState that is drained. Use child instead. */
1817
1818     if (blk->public.io_limits_disabled++ == 0) {
1819         throttle_group_restart_blk(blk);
1820     }
1821 }
1822
1823 static void blk_root_drained_end(BdrvChild *child)
1824 {
1825     BlockBackend *blk = child->opaque;
1826
1827     assert(blk->public.io_limits_disabled);
1828     --blk->public.io_limits_disabled;
1829 }