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     uint64_t perm;
 170
 171     /* blk_new_open() is mainly used in .bdrv_create implementations and the
 172      * tools where sharing isn't a concern because the BDS stays private, so we
 173      * just request permission according to the flags.
 174      *
 175      * The exceptions are xen_disk and blockdev_init(); in these cases, the
 176      * caller of blk_new_open() doesn't make use of the permissions, but they
 177      * shouldn't hurt either. We can still share everything here because the
 178      * guest devices will add their own blockers if they can't share. */
 179     perm = BLK_PERM_CONSISTENT_READ;
 180     if (flags & BDRV_O_RDWR) {
 181         perm |= BLK_PERM_WRITE;
 182     }
 183     if (flags & BDRV_O_RESIZE) {
 184         perm |= BLK_PERM_RESIZE;
 185     }
 186
 187     blk = blk_new(perm, BLK_PERM_ALL);
 188     bs = bdrv_open(filename, reference, options, flags, errp);
 189     if (!bs) {
 190         blk_unref(blk);
 191         return NULL;
 192     }
 193
 194     blk->root = bdrv_root_attach_child(bs, "root", &child_root,
 195                                        perm, BLK_PERM_ALL, blk, &error_abort);
 196
 197     return blk;
 198 }
 199
 200 static void blk_delete(BlockBackend *blk)
 201 {
 202     assert(!blk->refcnt);
 203     assert(!blk->name);
 204     assert(!blk->dev);
 205     if (blk->root) {
 206         blk_remove_bs(blk);
 207     }
 208     assert(QLIST_EMPTY(&blk->remove_bs_notifiers.notifiers));
 209     assert(QLIST_EMPTY(&blk->insert_bs_notifiers.notifiers));
 210     QTAILQ_REMOVE(&block_backends, blk, link);
 211     drive_info_del(blk->legacy_dinfo);
 212     block_acct_cleanup(&blk->stats);
 213     g_free(blk);
 214 }
 215
 216 static void drive_info_del(DriveInfo *dinfo)
 217 {
 218     if (!dinfo) {
 219         return;
 220     }
 221     qemu_opts_del(dinfo->opts);
 222     g_free(dinfo->serial);
 223     g_free(dinfo);
 224 }
 225
 226 int blk_get_refcnt(BlockBackend *blk)
 227 {
 228     return blk ? blk->refcnt : 0;
 229 }
 230
 231 /*
 232  * Increment @blk's reference count.
 233  * @blk must not be null.
 234  */
 235 void blk_ref(BlockBackend *blk)
 236 {
 237     blk->refcnt++;
 238 }
 239
 240 /*
 241  * Decrement @blk's reference count.
 242  * If this drops it to zero, destroy @blk.
 243  * For convenience, do nothing if @blk is null.
 244  */
 245 void blk_unref(BlockBackend *blk)
 246 {
 247     if (blk) {
 248         assert(blk->refcnt > 0);
 249         if (!--blk->refcnt) {
 250             blk_delete(blk);
 251         }
 252     }
 253 }
 254
 255 /*
 256  * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
 257  * ones which are hidden (i.e. are not referenced by the monitor).
 258  */
 259 static BlockBackend *blk_all_next(BlockBackend *blk)
 260 {
 261     return blk ? QTAILQ_NEXT(blk, link)
 262                : QTAILQ_FIRST(&block_backends);
 263 }
 264
 265 void blk_remove_all_bs(void)
 266 {
 267     BlockBackend *blk = NULL;
 268
 269     while ((blk = blk_all_next(blk)) != NULL) {
 270         AioContext *ctx = blk_get_aio_context(blk);
 271
 272         aio_context_acquire(ctx);
 273         if (blk->root) {
 274             blk_remove_bs(blk);
 275         }
 276         aio_context_release(ctx);
 277     }
 278 }
 279
 280 /*
 281  * Return the monitor-owned BlockBackend after @blk.
 282  * If @blk is null, return the first one.
 283  * Else, return @blk's next sibling, which may be null.
 284  *
 285  * To iterate over all BlockBackends, do
 286  * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
 287  *     ...
 288  * }
 289  */
 290 BlockBackend *blk_next(BlockBackend *blk)
 291 {
 292     return blk ? QTAILQ_NEXT(blk, monitor_link)
 293                : QTAILQ_FIRST(&monitor_block_backends);
 294 }
 295
 296 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
 297  * the monitor or attached to a BlockBackend */
 298 BlockDriverState *bdrv_next(BdrvNextIterator *it)
 299 {
 300     BlockDriverState *bs;
 301
 302     /* First, return all root nodes of BlockBackends. In order to avoid
 303      * returning a BDS twice when multiple BBs refer to it, we only return it
 304      * if the BB is the first one in the parent list of the BDS. */
 305     if (it->phase == BDRV_NEXT_BACKEND_ROOTS) {
 306         do {
 307             it->blk = blk_all_next(it->blk);
 308             bs = it->blk ? blk_bs(it->blk) : NULL;
 309         } while (it->blk && (bs == NULL || bdrv_first_blk(bs) != it->blk));
 310
 311         if (bs) {
 312             return bs;
 313         }
 314         it->phase = BDRV_NEXT_MONITOR_OWNED;
 315     }
 316
 317     /* Then return the monitor-owned BDSes without a BB attached. Ignore all
 318      * BDSes that are attached to a BlockBackend here; they have been handled
 319      * by the above block already */
 320     do {
 321         it->bs = bdrv_next_monitor_owned(it->bs);
 322         bs = it->bs;
 323     } while (bs && bdrv_has_blk(bs));
 324
 325     return bs;
 326 }
 327
 328 BlockDriverState *bdrv_first(BdrvNextIterator *it)
 329 {
 330     *it = (BdrvNextIterator) {
 331         .phase = BDRV_NEXT_BACKEND_ROOTS,
 332     };
 333
 334     return bdrv_next(it);
 335 }
 336
 337 /*
 338  * Add a BlockBackend into the list of backends referenced by the monitor, with
 339  * the given @name acting as the handle for the monitor.
 340  * Strictly for use by blockdev.c.
 341  *
 342  * @name must not be null or empty.
 343  *
 344  * Returns true on success and false on failure. In the latter case, an Error
 345  * object is returned through @errp.
 346  */
 347 bool monitor_add_blk(BlockBackend *blk, const char *name, Error **errp)
 348 {
 349     assert(!blk->name);
 350     assert(name && name[0]);
 351
 352     if (!id_wellformed(name)) {
 353         error_setg(errp, "Invalid device name");
 354         return false;
 355     }
 356     if (blk_by_name(name)) {
 357         error_setg(errp, "Device with id '%s' already exists", name);
 358         return false;
 359     }
 360     if (bdrv_find_node(name)) {
 361         error_setg(errp,
 362                    "Device name '%s' conflicts with an existing node name",
 363                    name);
 364         return false;
 365     }
 366
 367     blk->name = g_strdup(name);
 368     QTAILQ_INSERT_TAIL(&monitor_block_backends, blk, monitor_link);
 369     return true;
 370 }
 371
 372 /*
 373  * Remove a BlockBackend from the list of backends referenced by the monitor.
 374  * Strictly for use by blockdev.c.
 375  */
 376 void monitor_remove_blk(BlockBackend *blk)
 377 {
 378     if (!blk->name) {
 379         return;
 380     }
 381
 382     QTAILQ_REMOVE(&monitor_block_backends, blk, monitor_link);
 383     g_free(blk->name);
 384     blk->name = NULL;
 385 }
 386
 387 /*
 388  * Return @blk's name, a non-null string.
 389  * Returns an empty string iff @blk is not referenced by the monitor.
 390  */
 391 const char *blk_name(BlockBackend *blk)
 392 {
 393     return blk->name ?: "";
 394 }
 395
 396 /*
 397  * Return the BlockBackend with name @name if it exists, else null.
 398  * @name must not be null.
 399  */
 400 BlockBackend *blk_by_name(const char *name)
 401 {
 402     BlockBackend *blk = NULL;
 403
 404     assert(name);
 405     while ((blk = blk_next(blk)) != NULL) {
 406         if (!strcmp(name, blk->name)) {
 407             return blk;
 408         }
 409     }
 410     return NULL;
 411 }
 412
 413 /*
 414  * Return the BlockDriverState attached to @blk if any, else null.
 415  */
 416 BlockDriverState *blk_bs(BlockBackend *blk)
 417 {
 418     return blk->root ? blk->root->bs : NULL;
 419 }
 420
 421 static BlockBackend *bdrv_first_blk(BlockDriverState *bs)
 422 {
 423     BdrvChild *child;
 424     QLIST_FOREACH(child, &bs->parents, next_parent) {
 425         if (child->role == &child_root) {
 426             return child->opaque;
 427         }
 428     }
 429
 430     return NULL;
 431 }
 432
 433 /*
 434  * Returns true if @bs has an associated BlockBackend.
 435  */
 436 bool bdrv_has_blk(BlockDriverState *bs)
 437 {
 438     return bdrv_first_blk(bs) != NULL;
 439 }
 440
 441 /*
 442  * Returns true if @bs has only BlockBackends as parents.
 443  */
 444 bool bdrv_is_root_node(BlockDriverState *bs)
 445 {
 446     BdrvChild *c;
 447
 448     QLIST_FOREACH(c, &bs->parents, next_parent) {
 449         if (c->role != &child_root) {
 450             return false;
 451         }
 452     }
 453
 454     return true;
 455 }
 456
 457 /*
 458  * Return @blk's DriveInfo if any, else null.
 459  */
 460 DriveInfo *blk_legacy_dinfo(BlockBackend *blk)
 461 {
 462     return blk->legacy_dinfo;
 463 }
 464
 465 /*
 466  * Set @blk's DriveInfo to @dinfo, and return it.
 467  * @blk must not have a DriveInfo set already.
 468  * No other BlockBackend may have the same DriveInfo set.
 469  */
 470 DriveInfo *blk_set_legacy_dinfo(BlockBackend *blk, DriveInfo *dinfo)
 471 {
 472     assert(!blk->legacy_dinfo);
 473     return blk->legacy_dinfo = dinfo;
 474 }
 475
 476 /*
 477  * Return the BlockBackend with DriveInfo @dinfo.
 478  * It must exist.
 479  */
 480 BlockBackend *blk_by_legacy_dinfo(DriveInfo *dinfo)
 481 {
 482     BlockBackend *blk = NULL;
 483
 484     while ((blk = blk_next(blk)) != NULL) {
 485         if (blk->legacy_dinfo == dinfo) {
 486             return blk;
 487         }
 488     }
 489     abort();
 490 }
 491
 492 /*
 493  * Returns a pointer to the publicly accessible fields of @blk.
 494  */
 495 BlockBackendPublic *blk_get_public(BlockBackend *blk)
 496 {
 497     return &blk->public;
 498 }
 499
 500 /*
 501  * Returns a BlockBackend given the associated @public fields.
 502  */
 503 BlockBackend *blk_by_public(BlockBackendPublic *public)
 504 {
 505     return container_of(public, BlockBackend, public);
 506 }
 507
 508 /*
 509  * Disassociates the currently associated BlockDriverState from @blk.
 510  */
 511 void blk_remove_bs(BlockBackend *blk)
 512 {
 513     notifier_list_notify(&blk->remove_bs_notifiers, blk);
 514     if (blk->public.throttle_state) {
 515         throttle_timers_detach_aio_context(&blk->public.throttle_timers);
 516     }
 517
 518     blk_update_root_state(blk);
 519
 520     bdrv_root_unref_child(blk->root);
 521     blk->root = NULL;
 522 }
 523
 524 /*
 525  * Associates a new BlockDriverState with @blk.
 526  */
 527 int blk_insert_bs(BlockBackend *blk, BlockDriverState *bs, Error **errp)
 528 {
 529     blk->root = bdrv_root_attach_child(bs, "root", &child_root,
 530                                        blk->perm, blk->shared_perm, blk, errp);
 531     if (blk->root == NULL) {
 532         return -EPERM;
 533     }
 534     bdrv_ref(bs);
 535
 536     notifier_list_notify(&blk->insert_bs_notifiers, blk);
 537     if (blk->public.throttle_state) {
 538         throttle_timers_attach_aio_context(
 539             &blk->public.throttle_timers, bdrv_get_aio_context(bs));
 540     }
 541
 542     return 0;
 543 }
 544
 545 /*
 546  * Sets the permission bitmasks that the user of the BlockBackend needs.
 547  */
 548 int blk_set_perm(BlockBackend *blk, uint64_t perm, uint64_t shared_perm,
 549                  Error **errp)
 550 {
 551     int ret;
 552
 553     if (blk->root) {
 554         ret = bdrv_child_try_set_perm(blk->root, perm, shared_perm, errp);
 555         if (ret < 0) {
 556             return ret;
 557         }
 558     }
 559
 560     blk->perm = perm;
 561     blk->shared_perm = shared_perm;
 562
 563     return 0;
 564 }
 565
 566 static int blk_do_attach_dev(BlockBackend *blk, void *dev)
 567 {
 568     if (blk->dev) {
 569         return -EBUSY;
 570     }
 571     blk_ref(blk);
 572     blk->dev = dev;
 573     blk->legacy_dev = false;
 574     blk_iostatus_reset(blk);
 575     return 0;
 576 }
 577
 578 /*
 579  * Attach device model @dev to @blk.
 580  * Return 0 on success, -EBUSY when a device model is attached already.
 581  */
 582 int blk_attach_dev(BlockBackend *blk, DeviceState *dev)
 583 {
 584     return blk_do_attach_dev(blk, dev);
 585 }
 586
 587 /*
 588  * Attach device model @dev to @blk.
 589  * @blk must not have a device model attached already.
 590  * TODO qdevified devices don't use this, remove when devices are qdevified
 591  */
 592 void blk_attach_dev_legacy(BlockBackend *blk, void *dev)
 593 {
 594     if (blk_do_attach_dev(blk, dev) < 0) {
 595         abort();
 596     }
 597     blk->legacy_dev = true;
 598 }
 599
 600 /*
 601  * Detach device model @dev from @blk.
 602  * @dev must be currently attached to @blk.
 603  */
 604 void blk_detach_dev(BlockBackend *blk, void *dev)
 605 /* TODO change to DeviceState *dev when all users are qdevified */
 606 {
 607     assert(blk->dev == dev);
 608     blk->dev = NULL;
 609     blk->dev_ops = NULL;
 610     blk->dev_opaque = NULL;
 611     blk->guest_block_size = 512;
 612     blk_set_perm(blk, 0, BLK_PERM_ALL, &error_abort);
 613     blk_unref(blk);
 614 }
 615
 616 /*
 617  * Return the device model attached to @blk if any, else null.
 618  */
 619 void *blk_get_attached_dev(BlockBackend *blk)
 620 /* TODO change to return DeviceState * when all users are qdevified */
 621 {
 622     return blk->dev;
 623 }
 624
 625 /* Return the qdev ID, or if no ID is assigned the QOM path, of the block
 626  * device attached to the BlockBackend. */
 627 static char *blk_get_attached_dev_id(BlockBackend *blk)
 628 {
 629     DeviceState *dev;
 630
 631     assert(!blk->legacy_dev);
 632     dev = blk->dev;
 633
 634     if (!dev) {
 635         return g_strdup("");
 636     } else if (dev->id) {
 637         return g_strdup(dev->id);
 638     }
 639     return object_get_canonical_path(OBJECT(dev));
 640 }
 641
 642 /*
 643  * Return the BlockBackend which has the device model @dev attached if it
 644  * exists, else null.
 645  *
 646  * @dev must not be null.
 647  */
 648 BlockBackend *blk_by_dev(void *dev)
 649 {
 650     BlockBackend *blk = NULL;
 651
 652     assert(dev != NULL);
 653     while ((blk = blk_all_next(blk)) != NULL) {
 654         if (blk->dev == dev) {
 655             return blk;
 656         }
 657     }
 658     return NULL;
 659 }
 660
 661 /*
 662  * Set @blk's device model callbacks to @ops.
 663  * @opaque is the opaque argument to pass to the callbacks.
 664  * This is for use by device models.
 665  */
 666 void blk_set_dev_ops(BlockBackend *blk, const BlockDevOps *ops,
 667                      void *opaque)
 668 {
 669     /* All drivers that use blk_set_dev_ops() are qdevified and we want to keep
 670      * it that way, so we can assume blk->dev is a DeviceState if blk->dev_ops
 671      * is set. */
 672     assert(!blk->legacy_dev);
 673
 674     blk->dev_ops = ops;
 675     blk->dev_opaque = opaque;
 676 }
 677
 678 /*
 679  * Notify @blk's attached device model of media change.
 680  *
 681  * If @load is true, notify of media load. This action can fail, meaning that
 682  * the medium cannot be loaded. @errp is set then.
 683  *
 684  * If @load is false, notify of media eject. This can never fail.
 685  *
 686  * Also send DEVICE_TRAY_MOVED events as appropriate.
 687  */
 688 void blk_dev_change_media_cb(BlockBackend *blk, bool load, Error **errp)
 689 {
 690     if (blk->dev_ops && blk->dev_ops->change_media_cb) {
 691         bool tray_was_open, tray_is_open;
 692         Error *local_err = NULL;
 693
 694         assert(!blk->legacy_dev);
 695
 696         tray_was_open = blk_dev_is_tray_open(blk);
 697         blk->dev_ops->change_media_cb(blk->dev_opaque, load, &local_err);
 698         if (local_err) {
 699             assert(load == true);
 700             error_propagate(errp, local_err);
 701             return;
 702         }
 703         tray_is_open = blk_dev_is_tray_open(blk);
 704
 705         if (tray_was_open != tray_is_open) {
 706             char *id = blk_get_attached_dev_id(blk);
 707             qapi_event_send_device_tray_moved(blk_name(blk), id, tray_is_open,
 708                                               &error_abort);
 709             g_free(id);
 710         }
 711     }
 712 }
 713
 714 static void blk_root_change_media(BdrvChild *child, bool load)
 715 {
 716     blk_dev_change_media_cb(child->opaque, load, NULL);
 717 }
 718
 719 /*
 720  * Does @blk's attached device model have removable media?
 721  * %true if no device model is attached.
 722  */
 723 bool blk_dev_has_removable_media(BlockBackend *blk)
 724 {
 725     return !blk->dev || (blk->dev_ops && blk->dev_ops->change_media_cb);
 726 }
 727
 728 /*
 729  * Does @blk's attached device model have a tray?
 730  */
 731 bool blk_dev_has_tray(BlockBackend *blk)
 732 {
 733     return blk->dev_ops && blk->dev_ops->is_tray_open;
 734 }
 735
 736 /*
 737  * Notify @blk's attached device model of a media eject request.
 738  * If @force is true, the medium is about to be yanked out forcefully.
 739  */
 740 void blk_dev_eject_request(BlockBackend *blk, bool force)
 741 {
 742     if (blk->dev_ops && blk->dev_ops->eject_request_cb) {
 743         blk->dev_ops->eject_request_cb(blk->dev_opaque, force);
 744     }
 745 }
 746
 747 /*
 748  * Does @blk's attached device model have a tray, and is it open?
 749  */
 750 bool blk_dev_is_tray_open(BlockBackend *blk)
 751 {
 752     if (blk_dev_has_tray(blk)) {
 753         return blk->dev_ops->is_tray_open(blk->dev_opaque);
 754     }
 755     return false;
 756 }
 757
 758 /*
 759  * Does @blk's attached device model have the medium locked?
 760  * %false if the device model has no such lock.
 761  */
 762 bool blk_dev_is_medium_locked(BlockBackend *blk)
 763 {
 764     if (blk->dev_ops && blk->dev_ops->is_medium_locked) {
 765         return blk->dev_ops->is_medium_locked(blk->dev_opaque);
 766     }
 767     return false;
 768 }
 769
 770 /*
 771  * Notify @blk's attached device model of a backend size change.
 772  */
 773 static void blk_root_resize(BdrvChild *child)
 774 {
 775     BlockBackend *blk = child->opaque;
 776
 777     if (blk->dev_ops && blk->dev_ops->resize_cb) {
 778         blk->dev_ops->resize_cb(blk->dev_opaque);
 779     }
 780 }
 781
 782 void blk_iostatus_enable(BlockBackend *blk)
 783 {
 784     blk->iostatus_enabled = true;
 785     blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 786 }
 787
 788 /* The I/O status is only enabled if the drive explicitly
 789  * enables it _and_ the VM is configured to stop on errors */
 790 bool blk_iostatus_is_enabled(const BlockBackend *blk)
 791 {
 792     return (blk->iostatus_enabled &&
 793            (blk->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
 794             blk->on_write_error == BLOCKDEV_ON_ERROR_STOP   ||
 795             blk->on_read_error == BLOCKDEV_ON_ERROR_STOP));
 796 }
 797
 798 BlockDeviceIoStatus blk_iostatus(const BlockBackend *blk)
 799 {
 800     return blk->iostatus;
 801 }
 802
 803 void blk_iostatus_disable(BlockBackend *blk)
 804 {
 805     blk->iostatus_enabled = false;
 806 }
 807
 808 void blk_iostatus_reset(BlockBackend *blk)
 809 {
 810     if (blk_iostatus_is_enabled(blk)) {
 811         BlockDriverState *bs = blk_bs(blk);
 812         blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 813         if (bs && bs->job) {
 814             block_job_iostatus_reset(bs->job);
 815         }
 816     }
 817 }
 818
 819 void blk_iostatus_set_err(BlockBackend *blk, int error)
 820 {
 821     assert(blk_iostatus_is_enabled(blk));
 822     if (blk->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 823         blk->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 824                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 825     }
 826 }
 827
 828 void blk_set_allow_write_beyond_eof(BlockBackend *blk, bool allow)
 829 {
 830     blk->allow_write_beyond_eof = allow;
 831 }
 832
 833 static int blk_check_byte_request(BlockBackend *blk, int64_t offset,
 834                                   size_t size)
 835 {
 836     int64_t len;
 837
 838     if (size > INT_MAX) {
 839         return -EIO;
 840     }
 841
 842     if (!blk_is_available(blk)) {
 843         return -ENOMEDIUM;
 844     }
 845
 846     if (offset < 0) {
 847         return -EIO;
 848     }
 849
 850     if (!blk->allow_write_beyond_eof) {
 851         len = blk_getlength(blk);
 852         if (len < 0) {
 853             return len;
 854         }
 855
 856         if (offset > len || len - offset < size) {
 857             return -EIO;
 858         }
 859     }
 860
 861     return 0;
 862 }
 863
 864 int coroutine_fn blk_co_preadv(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_preadv(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, false);
 883     }
 884
 885     ret = bdrv_co_preadv(blk->root, offset, bytes, qiov, flags);
 886     bdrv_dec_in_flight(bs);
 887     return ret;
 888 }
 889
 890 int coroutine_fn blk_co_pwritev(BlockBackend *blk, int64_t offset,
 891                                 unsigned int bytes, QEMUIOVector *qiov,
 892                                 BdrvRequestFlags flags)
 893 {
 894     int ret;
 895     BlockDriverState *bs = blk_bs(blk);
 896
 897     trace_blk_co_pwritev(blk, bs, offset, bytes, flags);
 898
 899     ret = blk_check_byte_request(blk, offset, bytes);
 900     if (ret < 0) {
 901         return ret;
 902     }
 903
 904     bdrv_inc_in_flight(bs);
 905
 906     /* throttling disk I/O */
 907     if (blk->public.throttle_state) {
 908         throttle_group_co_io_limits_intercept(blk, bytes, true);
 909     }
 910
 911     if (!blk->enable_write_cache) {
 912         flags |= BDRV_REQ_FUA;
 913     }
 914
 915     ret = bdrv_co_pwritev(blk->root, offset, bytes, qiov, flags);
 916     bdrv_dec_in_flight(bs);
 917     return ret;
 918 }
 919
 920 typedef struct BlkRwCo {
 921     BlockBackend *blk;
 922     int64_t offset;
 923     QEMUIOVector *qiov;
 924     int ret;
 925     BdrvRequestFlags flags;
 926 } BlkRwCo;
 927
 928 static void blk_read_entry(void *opaque)
 929 {
 930     BlkRwCo *rwco = opaque;
 931
 932     rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, rwco->qiov->size,
 933                               rwco->qiov, rwco->flags);
 934 }
 935
 936 static void blk_write_entry(void *opaque)
 937 {
 938     BlkRwCo *rwco = opaque;
 939
 940     rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, rwco->qiov->size,
 941                                rwco->qiov, rwco->flags);
 942 }
 943
 944 static int blk_prw(BlockBackend *blk, int64_t offset, uint8_t *buf,
 945                    int64_t bytes, CoroutineEntry co_entry,
 946                    BdrvRequestFlags flags)
 947 {
 948     QEMUIOVector qiov;
 949     struct iovec iov;
 950     BlkRwCo rwco;
 951
 952     iov = (struct iovec) {
 953         .iov_base = buf,
 954         .iov_len = bytes,
 955     };
 956     qemu_iovec_init_external(&qiov, &iov, 1);
 957
 958     rwco = (BlkRwCo) {
 959         .blk    = blk,
 960         .offset = offset,
 961         .qiov   = &qiov,
 962         .flags  = flags,
 963         .ret    = NOT_DONE,
 964     };
 965
 966     if (qemu_in_coroutine()) {
 967         /* Fast-path if already in coroutine context */
 968         co_entry(&rwco);
 969     } else {
 970         Coroutine *co = qemu_coroutine_create(co_entry, &rwco);
 971         qemu_coroutine_enter(co);
 972         BDRV_POLL_WHILE(blk_bs(blk), rwco.ret == NOT_DONE);
 973     }
 974
 975     return rwco.ret;
 976 }
 977
 978 int blk_pread_unthrottled(BlockBackend *blk, int64_t offset, uint8_t *buf,
 979                           int count)
 980 {
 981     int ret;
 982
 983     ret = blk_check_byte_request(blk, offset, count);
 984     if (ret < 0) {
 985         return ret;
 986     }
 987
 988     blk_root_drained_begin(blk->root);
 989     ret = blk_pread(blk, offset, buf, count);
 990     blk_root_drained_end(blk->root);
 991     return ret;
 992 }
 993
 994 int blk_pwrite_zeroes(BlockBackend *blk, int64_t offset,
 995                       int count, BdrvRequestFlags flags)
 996 {
 997     return blk_prw(blk, offset, NULL, count, blk_write_entry,
 998                    flags | BDRV_REQ_ZERO_WRITE);
 999 }
1000
1001 int blk_make_zero(BlockBackend *blk, BdrvRequestFlags flags)
1002 {
1003     return bdrv_make_zero(blk->root, flags);
1004 }
1005
1006 static void error_callback_bh(void *opaque)
1007 {
1008     struct BlockBackendAIOCB *acb = opaque;
1009
1010     bdrv_dec_in_flight(acb->common.bs);
1011     acb->common.cb(acb->common.opaque, acb->ret);
1012     qemu_aio_unref(acb);
1013 }
1014
1015 BlockAIOCB *blk_abort_aio_request(BlockBackend *blk,
1016                                   BlockCompletionFunc *cb,
1017                                   void *opaque, int ret)
1018 {
1019     struct BlockBackendAIOCB *acb;
1020
1021     bdrv_inc_in_flight(blk_bs(blk));
1022     acb = blk_aio_get(&block_backend_aiocb_info, blk, cb, opaque);
1023     acb->blk = blk;
1024     acb->ret = ret;
1025
1026     aio_bh_schedule_oneshot(blk_get_aio_context(blk), error_callback_bh, acb);
1027     return &acb->common;
1028 }
1029
1030 typedef struct BlkAioEmAIOCB {
1031     BlockAIOCB common;
1032     BlkRwCo rwco;
1033     int bytes;
1034     bool has_returned;
1035 } BlkAioEmAIOCB;
1036
1037 static const AIOCBInfo blk_aio_em_aiocb_info = {
1038     .aiocb_size         = sizeof(BlkAioEmAIOCB),
1039 };
1040
1041 static void blk_aio_complete(BlkAioEmAIOCB *acb)
1042 {
1043     if (acb->has_returned) {
1044         bdrv_dec_in_flight(acb->common.bs);
1045         acb->common.cb(acb->common.opaque, acb->rwco.ret);
1046         qemu_aio_unref(acb);
1047     }
1048 }
1049
1050 static void blk_aio_complete_bh(void *opaque)
1051 {
1052     BlkAioEmAIOCB *acb = opaque;
1053     assert(acb->has_returned);
1054     blk_aio_complete(acb);
1055 }
1056
1057 static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset, int bytes,
1058                                 QEMUIOVector *qiov, CoroutineEntry co_entry,
1059                                 BdrvRequestFlags flags,
1060                                 BlockCompletionFunc *cb, void *opaque)
1061 {
1062     BlkAioEmAIOCB *acb;
1063     Coroutine *co;
1064
1065     bdrv_inc_in_flight(blk_bs(blk));
1066     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1067     acb->rwco = (BlkRwCo) {
1068         .blk    = blk,
1069         .offset = offset,
1070         .qiov   = qiov,
1071         .flags  = flags,
1072         .ret    = NOT_DONE,
1073     };
1074     acb->bytes = bytes;
1075     acb->has_returned = false;
1076
1077     co = qemu_coroutine_create(co_entry, acb);
1078     qemu_coroutine_enter(co);
1079
1080     acb->has_returned = true;
1081     if (acb->rwco.ret != NOT_DONE) {
1082         aio_bh_schedule_oneshot(blk_get_aio_context(blk),
1083                                 blk_aio_complete_bh, acb);
1084     }
1085
1086     return &acb->common;
1087 }
1088
1089 static void blk_aio_read_entry(void *opaque)
1090 {
1091     BlkAioEmAIOCB *acb = opaque;
1092     BlkRwCo *rwco = &acb->rwco;
1093
1094     assert(rwco->qiov->size == acb->bytes);
1095     rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, acb->bytes,
1096                               rwco->qiov, rwco->flags);
1097     blk_aio_complete(acb);
1098 }
1099
1100 static void blk_aio_write_entry(void *opaque)
1101 {
1102     BlkAioEmAIOCB *acb = opaque;
1103     BlkRwCo *rwco = &acb->rwco;
1104
1105     assert(!rwco->qiov || rwco->qiov->size == acb->bytes);
1106     rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, acb->bytes,
1107                                rwco->qiov, rwco->flags);
1108     blk_aio_complete(acb);
1109 }
1110
1111 BlockAIOCB *blk_aio_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1112                                   int count, BdrvRequestFlags flags,
1113                                   BlockCompletionFunc *cb, void *opaque)
1114 {
1115     return blk_aio_prwv(blk, offset, count, NULL, blk_aio_write_entry,
1116                         flags | BDRV_REQ_ZERO_WRITE, cb, opaque);
1117 }
1118
1119 int blk_pread(BlockBackend *blk, int64_t offset, void *buf, int count)
1120 {
1121     int ret = blk_prw(blk, offset, buf, count, blk_read_entry, 0);
1122     if (ret < 0) {
1123         return ret;
1124     }
1125     return count;
1126 }
1127
1128 int blk_pwrite(BlockBackend *blk, int64_t offset, const void *buf, int count,
1129                BdrvRequestFlags flags)
1130 {
1131     int ret = blk_prw(blk, offset, (void *) buf, count, blk_write_entry,
1132                       flags);
1133     if (ret < 0) {
1134         return ret;
1135     }
1136     return count;
1137 }
1138
1139 int64_t blk_getlength(BlockBackend *blk)
1140 {
1141     if (!blk_is_available(blk)) {
1142         return -ENOMEDIUM;
1143     }
1144
1145     return bdrv_getlength(blk_bs(blk));
1146 }
1147
1148 void blk_get_geometry(BlockBackend *blk, uint64_t *nb_sectors_ptr)
1149 {
1150     if (!blk_bs(blk)) {
1151         *nb_sectors_ptr = 0;
1152     } else {
1153         bdrv_get_geometry(blk_bs(blk), nb_sectors_ptr);
1154     }
1155 }
1156
1157 int64_t blk_nb_sectors(BlockBackend *blk)
1158 {
1159     if (!blk_is_available(blk)) {
1160         return -ENOMEDIUM;
1161     }
1162
1163     return bdrv_nb_sectors(blk_bs(blk));
1164 }
1165
1166 BlockAIOCB *blk_aio_preadv(BlockBackend *blk, int64_t offset,
1167                            QEMUIOVector *qiov, BdrvRequestFlags flags,
1168                            BlockCompletionFunc *cb, void *opaque)
1169 {
1170     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1171                         blk_aio_read_entry, flags, cb, opaque);
1172 }
1173
1174 BlockAIOCB *blk_aio_pwritev(BlockBackend *blk, int64_t offset,
1175                             QEMUIOVector *qiov, BdrvRequestFlags flags,
1176                             BlockCompletionFunc *cb, void *opaque)
1177 {
1178     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1179                         blk_aio_write_entry, flags, cb, opaque);
1180 }
1181
1182 static void blk_aio_flush_entry(void *opaque)
1183 {
1184     BlkAioEmAIOCB *acb = opaque;
1185     BlkRwCo *rwco = &acb->rwco;
1186
1187     rwco->ret = blk_co_flush(rwco->blk);
1188     blk_aio_complete(acb);
1189 }
1190
1191 BlockAIOCB *blk_aio_flush(BlockBackend *blk,
1192                           BlockCompletionFunc *cb, void *opaque)
1193 {
1194     return blk_aio_prwv(blk, 0, 0, NULL, blk_aio_flush_entry, 0, cb, opaque);
1195 }
1196
1197 static void blk_aio_pdiscard_entry(void *opaque)
1198 {
1199     BlkAioEmAIOCB *acb = opaque;
1200     BlkRwCo *rwco = &acb->rwco;
1201
1202     rwco->ret = blk_co_pdiscard(rwco->blk, rwco->offset, acb->bytes);
1203     blk_aio_complete(acb);
1204 }
1205
1206 BlockAIOCB *blk_aio_pdiscard(BlockBackend *blk,
1207                              int64_t offset, int count,
1208                              BlockCompletionFunc *cb, void *opaque)
1209 {
1210     return blk_aio_prwv(blk, offset, count, NULL, blk_aio_pdiscard_entry, 0,
1211                         cb, opaque);
1212 }
1213
1214 void blk_aio_cancel(BlockAIOCB *acb)
1215 {
1216     bdrv_aio_cancel(acb);
1217 }
1218
1219 void blk_aio_cancel_async(BlockAIOCB *acb)
1220 {
1221     bdrv_aio_cancel_async(acb);
1222 }
1223
1224 int blk_co_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1225 {
1226     if (!blk_is_available(blk)) {
1227         return -ENOMEDIUM;
1228     }
1229
1230     return bdrv_co_ioctl(blk_bs(blk), req, buf);
1231 }
1232
1233 static void blk_ioctl_entry(void *opaque)
1234 {
1235     BlkRwCo *rwco = opaque;
1236     rwco->ret = blk_co_ioctl(rwco->blk, rwco->offset,
1237                              rwco->qiov->iov[0].iov_base);
1238 }
1239
1240 int blk_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1241 {
1242     return blk_prw(blk, req, buf, 0, blk_ioctl_entry, 0);
1243 }
1244
1245 static void blk_aio_ioctl_entry(void *opaque)
1246 {
1247     BlkAioEmAIOCB *acb = opaque;
1248     BlkRwCo *rwco = &acb->rwco;
1249
1250     rwco->ret = blk_co_ioctl(rwco->blk, rwco->offset,
1251                              rwco->qiov->iov[0].iov_base);
1252     blk_aio_complete(acb);
1253 }
1254
1255 BlockAIOCB *blk_aio_ioctl(BlockBackend *blk, unsigned long int req, void *buf,
1256                           BlockCompletionFunc *cb, void *opaque)
1257 {
1258     QEMUIOVector qiov;
1259     struct iovec iov;
1260
1261     iov = (struct iovec) {
1262         .iov_base = buf,
1263         .iov_len = 0,
1264     };
1265     qemu_iovec_init_external(&qiov, &iov, 1);
1266
1267     return blk_aio_prwv(blk, req, 0, &qiov, blk_aio_ioctl_entry, 0, cb, opaque);
1268 }
1269
1270 int blk_co_pdiscard(BlockBackend *blk, int64_t offset, int count)
1271 {
1272     int ret = blk_check_byte_request(blk, offset, count);
1273     if (ret < 0) {
1274         return ret;
1275     }
1276
1277     return bdrv_co_pdiscard(blk_bs(blk), offset, count);
1278 }
1279
1280 int blk_co_flush(BlockBackend *blk)
1281 {
1282     if (!blk_is_available(blk)) {
1283         return -ENOMEDIUM;
1284     }
1285
1286     return bdrv_co_flush(blk_bs(blk));
1287 }
1288
1289 static void blk_flush_entry(void *opaque)
1290 {
1291     BlkRwCo *rwco = opaque;
1292     rwco->ret = blk_co_flush(rwco->blk);
1293 }
1294
1295 int blk_flush(BlockBackend *blk)
1296 {
1297     return blk_prw(blk, 0, NULL, 0, blk_flush_entry, 0);
1298 }
1299
1300 void blk_drain(BlockBackend *blk)
1301 {
1302     if (blk_bs(blk)) {
1303         bdrv_drain(blk_bs(blk));
1304     }
1305 }
1306
1307 void blk_drain_all(void)
1308 {
1309     bdrv_drain_all();
1310 }
1311
1312 void blk_set_on_error(BlockBackend *blk, BlockdevOnError on_read_error,
1313                       BlockdevOnError on_write_error)
1314 {
1315     blk->on_read_error = on_read_error;
1316     blk->on_write_error = on_write_error;
1317 }
1318
1319 BlockdevOnError blk_get_on_error(BlockBackend *blk, bool is_read)
1320 {
1321     return is_read ? blk->on_read_error : blk->on_write_error;
1322 }
1323
1324 BlockErrorAction blk_get_error_action(BlockBackend *blk, bool is_read,
1325                                       int error)
1326 {
1327     BlockdevOnError on_err = blk_get_on_error(blk, is_read);
1328
1329     switch (on_err) {
1330     case BLOCKDEV_ON_ERROR_ENOSPC:
1331         return (error == ENOSPC) ?
1332                BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
1333     case BLOCKDEV_ON_ERROR_STOP:
1334         return BLOCK_ERROR_ACTION_STOP;
1335     case BLOCKDEV_ON_ERROR_REPORT:
1336         return BLOCK_ERROR_ACTION_REPORT;
1337     case BLOCKDEV_ON_ERROR_IGNORE:
1338         return BLOCK_ERROR_ACTION_IGNORE;
1339     case BLOCKDEV_ON_ERROR_AUTO:
1340     default:
1341         abort();
1342     }
1343 }
1344
1345 static void send_qmp_error_event(BlockBackend *blk,
1346                                  BlockErrorAction action,
1347                                  bool is_read, int error)
1348 {
1349     IoOperationType optype;
1350
1351     optype = is_read ? IO_OPERATION_TYPE_READ : IO_OPERATION_TYPE_WRITE;
1352     qapi_event_send_block_io_error(blk_name(blk),
1353                                    bdrv_get_node_name(blk_bs(blk)), optype,
1354                                    action, blk_iostatus_is_enabled(blk),
1355                                    error == ENOSPC, strerror(error),
1356                                    &error_abort);
1357 }
1358
1359 /* This is done by device models because, while the block layer knows
1360  * about the error, it does not know whether an operation comes from
1361  * the device or the block layer (from a job, for example).
1362  */
1363 void blk_error_action(BlockBackend *blk, BlockErrorAction action,
1364                       bool is_read, int error)
1365 {
1366     assert(error >= 0);
1367
1368     if (action == BLOCK_ERROR_ACTION_STOP) {
1369         /* First set the iostatus, so that "info block" returns an iostatus
1370          * that matches the events raised so far (an additional error iostatus
1371          * is fine, but not a lost one).
1372          */
1373         blk_iostatus_set_err(blk, error);
1374
1375         /* Then raise the request to stop the VM and the event.
1376          * qemu_system_vmstop_request_prepare has two effects.  First,
1377          * it ensures that the STOP event always comes after the
1378          * BLOCK_IO_ERROR event.  Second, it ensures that even if management
1379          * can observe the STOP event and do a "cont" before the STOP
1380          * event is issued, the VM will not stop.  In this case, vm_start()
1381          * also ensures that the STOP/RESUME pair of events is emitted.
1382          */
1383         qemu_system_vmstop_request_prepare();
1384         send_qmp_error_event(blk, action, is_read, error);
1385         qemu_system_vmstop_request(RUN_STATE_IO_ERROR);
1386     } else {
1387         send_qmp_error_event(blk, action, is_read, error);
1388     }
1389 }
1390
1391 int blk_is_read_only(BlockBackend *blk)
1392 {
1393     BlockDriverState *bs = blk_bs(blk);
1394
1395     if (bs) {
1396         return bdrv_is_read_only(bs);
1397     } else {
1398         return blk->root_state.read_only;
1399     }
1400 }
1401
1402 int blk_is_sg(BlockBackend *blk)
1403 {
1404     BlockDriverState *bs = blk_bs(blk);
1405
1406     if (!bs) {
1407         return 0;
1408     }
1409
1410     return bdrv_is_sg(bs);
1411 }
1412
1413 int blk_enable_write_cache(BlockBackend *blk)
1414 {
1415     return blk->enable_write_cache;
1416 }
1417
1418 void blk_set_enable_write_cache(BlockBackend *blk, bool wce)
1419 {
1420     blk->enable_write_cache = wce;
1421 }
1422
1423 void blk_invalidate_cache(BlockBackend *blk, Error **errp)
1424 {
1425     BlockDriverState *bs = blk_bs(blk);
1426
1427     if (!bs) {
1428         error_setg(errp, "Device '%s' has no medium", blk->name);
1429         return;
1430     }
1431
1432     bdrv_invalidate_cache(bs, errp);
1433 }
1434
1435 bool blk_is_inserted(BlockBackend *blk)
1436 {
1437     BlockDriverState *bs = blk_bs(blk);
1438
1439     return bs && bdrv_is_inserted(bs);
1440 }
1441
1442 bool blk_is_available(BlockBackend *blk)
1443 {
1444     return blk_is_inserted(blk) && !blk_dev_is_tray_open(blk);
1445 }
1446
1447 void blk_lock_medium(BlockBackend *blk, bool locked)
1448 {
1449     BlockDriverState *bs = blk_bs(blk);
1450
1451     if (bs) {
1452         bdrv_lock_medium(bs, locked);
1453     }
1454 }
1455
1456 void blk_eject(BlockBackend *blk, bool eject_flag)
1457 {
1458     BlockDriverState *bs = blk_bs(blk);
1459     char *id;
1460
1461     /* blk_eject is only called by qdevified devices */
1462     assert(!blk->legacy_dev);
1463
1464     if (bs) {
1465         bdrv_eject(bs, eject_flag);
1466     }
1467
1468     /* Whether or not we ejected on the backend,
1469      * the frontend experienced a tray event. */
1470     id = blk_get_attached_dev_id(blk);
1471     qapi_event_send_device_tray_moved(blk_name(blk), id,
1472                                       eject_flag, &error_abort);
1473     g_free(id);
1474 }
1475
1476 int blk_get_flags(BlockBackend *blk)
1477 {
1478     BlockDriverState *bs = blk_bs(blk);
1479
1480     if (bs) {
1481         return bdrv_get_flags(bs);
1482     } else {
1483         return blk->root_state.open_flags;
1484     }
1485 }
1486
1487 /* Returns the maximum transfer length, in bytes; guaranteed nonzero */
1488 uint32_t blk_get_max_transfer(BlockBackend *blk)
1489 {
1490     BlockDriverState *bs = blk_bs(blk);
1491     uint32_t max = 0;
1492
1493     if (bs) {
1494         max = bs->bl.max_transfer;
1495     }
1496     return MIN_NON_ZERO(max, INT_MAX);
1497 }
1498
1499 int blk_get_max_iov(BlockBackend *blk)
1500 {
1501     return blk->root->bs->bl.max_iov;
1502 }
1503
1504 void blk_set_guest_block_size(BlockBackend *blk, int align)
1505 {
1506     blk->guest_block_size = align;
1507 }
1508
1509 void *blk_try_blockalign(BlockBackend *blk, size_t size)
1510 {
1511     return qemu_try_blockalign(blk ? blk_bs(blk) : NULL, size);
1512 }
1513
1514 void *blk_blockalign(BlockBackend *blk, size_t size)
1515 {
1516     return qemu_blockalign(blk ? blk_bs(blk) : NULL, size);
1517 }
1518
1519 bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp)
1520 {
1521     BlockDriverState *bs = blk_bs(blk);
1522
1523     if (!bs) {
1524         return false;
1525     }
1526
1527     return bdrv_op_is_blocked(bs, op, errp);
1528 }
1529
1530 void blk_op_unblock(BlockBackend *blk, BlockOpType op, Error *reason)
1531 {
1532     BlockDriverState *bs = blk_bs(blk);
1533
1534     if (bs) {
1535         bdrv_op_unblock(bs, op, reason);
1536     }
1537 }
1538
1539 void blk_op_block_all(BlockBackend *blk, Error *reason)
1540 {
1541     BlockDriverState *bs = blk_bs(blk);
1542
1543     if (bs) {
1544         bdrv_op_block_all(bs, reason);
1545     }
1546 }
1547
1548 void blk_op_unblock_all(BlockBackend *blk, Error *reason)
1549 {
1550     BlockDriverState *bs = blk_bs(blk);
1551
1552     if (bs) {
1553         bdrv_op_unblock_all(bs, reason);
1554     }
1555 }
1556
1557 AioContext *blk_get_aio_context(BlockBackend *blk)
1558 {
1559     BlockDriverState *bs = blk_bs(blk);
1560
1561     if (bs) {
1562         return bdrv_get_aio_context(bs);
1563     } else {
1564         return qemu_get_aio_context();
1565     }
1566 }
1567
1568 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb)
1569 {
1570     BlockBackendAIOCB *blk_acb = DO_UPCAST(BlockBackendAIOCB, common, acb);
1571     return blk_get_aio_context(blk_acb->blk);
1572 }
1573
1574 void blk_set_aio_context(BlockBackend *blk, AioContext *new_context)
1575 {
1576     BlockDriverState *bs = blk_bs(blk);
1577
1578     if (bs) {
1579         if (blk->public.throttle_state) {
1580             throttle_timers_detach_aio_context(&blk->public.throttle_timers);
1581         }
1582         bdrv_set_aio_context(bs, new_context);
1583         if (blk->public.throttle_state) {
1584             throttle_timers_attach_aio_context(&blk->public.throttle_timers,
1585                                                new_context);
1586         }
1587     }
1588 }
1589
1590 void blk_add_aio_context_notifier(BlockBackend *blk,
1591         void (*attached_aio_context)(AioContext *new_context, void *opaque),
1592         void (*detach_aio_context)(void *opaque), void *opaque)
1593 {
1594     BlockDriverState *bs = blk_bs(blk);
1595
1596     if (bs) {
1597         bdrv_add_aio_context_notifier(bs, attached_aio_context,
1598                                       detach_aio_context, opaque);
1599     }
1600 }
1601
1602 void blk_remove_aio_context_notifier(BlockBackend *blk,
1603                                      void (*attached_aio_context)(AioContext *,
1604                                                                   void *),
1605                                      void (*detach_aio_context)(void *),
1606                                      void *opaque)
1607 {
1608     BlockDriverState *bs = blk_bs(blk);
1609
1610     if (bs) {
1611         bdrv_remove_aio_context_notifier(bs, attached_aio_context,
1612                                          detach_aio_context, opaque);
1613     }
1614 }
1615
1616 void blk_add_remove_bs_notifier(BlockBackend *blk, Notifier *notify)
1617 {
1618     notifier_list_add(&blk->remove_bs_notifiers, notify);
1619 }
1620
1621 void blk_add_insert_bs_notifier(BlockBackend *blk, Notifier *notify)
1622 {
1623     notifier_list_add(&blk->insert_bs_notifiers, notify);
1624 }
1625
1626 void blk_io_plug(BlockBackend *blk)
1627 {
1628     BlockDriverState *bs = blk_bs(blk);
1629
1630     if (bs) {
1631         bdrv_io_plug(bs);
1632     }
1633 }
1634
1635 void blk_io_unplug(BlockBackend *blk)
1636 {
1637     BlockDriverState *bs = blk_bs(blk);
1638
1639     if (bs) {
1640         bdrv_io_unplug(bs);
1641     }
1642 }
1643
1644 BlockAcctStats *blk_get_stats(BlockBackend *blk)
1645 {
1646     return &blk->stats;
1647 }
1648
1649 void *blk_aio_get(const AIOCBInfo *aiocb_info, BlockBackend *blk,
1650                   BlockCompletionFunc *cb, void *opaque)
1651 {
1652     return qemu_aio_get(aiocb_info, blk_bs(blk), cb, opaque);
1653 }
1654
1655 int coroutine_fn blk_co_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1656                                       int count, BdrvRequestFlags flags)
1657 {
1658     return blk_co_pwritev(blk, offset, count, NULL,
1659                           flags | BDRV_REQ_ZERO_WRITE);
1660 }
1661
1662 int blk_pwrite_compressed(BlockBackend *blk, int64_t offset, const void *buf,
1663                           int count)
1664 {
1665     return blk_prw(blk, offset, (void *) buf, count, blk_write_entry,
1666                    BDRV_REQ_WRITE_COMPRESSED);
1667 }
1668
1669 int blk_truncate(BlockBackend *blk, int64_t offset)
1670 {
1671     if (!blk_is_available(blk)) {
1672         return -ENOMEDIUM;
1673     }
1674
1675     return bdrv_truncate(blk->root, offset);
1676 }
1677
1678 static void blk_pdiscard_entry(void *opaque)
1679 {
1680     BlkRwCo *rwco = opaque;
1681     rwco->ret = blk_co_pdiscard(rwco->blk, rwco->offset, rwco->qiov->size);
1682 }
1683
1684 int blk_pdiscard(BlockBackend *blk, int64_t offset, int count)
1685 {
1686     return blk_prw(blk, offset, NULL, count, blk_pdiscard_entry, 0);
1687 }
1688
1689 int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
1690                      int64_t pos, int size)
1691 {
1692     int ret;
1693
1694     if (!blk_is_available(blk)) {
1695         return -ENOMEDIUM;
1696     }
1697
1698     ret = bdrv_save_vmstate(blk_bs(blk), buf, pos, size);
1699     if (ret < 0) {
1700         return ret;
1701     }
1702
1703     if (ret == size && !blk->enable_write_cache) {
1704         ret = bdrv_flush(blk_bs(blk));
1705     }
1706
1707     return ret < 0 ? ret : size;
1708 }
1709
1710 int blk_load_vmstate(BlockBackend *blk, uint8_t *buf, int64_t pos, int size)
1711 {
1712     if (!blk_is_available(blk)) {
1713         return -ENOMEDIUM;
1714     }
1715
1716     return bdrv_load_vmstate(blk_bs(blk), buf, pos, size);
1717 }
1718
1719 int blk_probe_blocksizes(BlockBackend *blk, BlockSizes *bsz)
1720 {
1721     if (!blk_is_available(blk)) {
1722         return -ENOMEDIUM;
1723     }
1724
1725     return bdrv_probe_blocksizes(blk_bs(blk), bsz);
1726 }
1727
1728 int blk_probe_geometry(BlockBackend *blk, HDGeometry *geo)
1729 {
1730     if (!blk_is_available(blk)) {
1731         return -ENOMEDIUM;
1732     }
1733
1734     return bdrv_probe_geometry(blk_bs(blk), geo);
1735 }
1736
1737 /*
1738  * Updates the BlockBackendRootState object with data from the currently
1739  * attached BlockDriverState.
1740  */
1741 void blk_update_root_state(BlockBackend *blk)
1742 {
1743     assert(blk->root);
1744
1745     blk->root_state.open_flags    = blk->root->bs->open_flags;
1746     blk->root_state.read_only     = blk->root->bs->read_only;
1747     blk->root_state.detect_zeroes = blk->root->bs->detect_zeroes;
1748 }
1749
1750 /*
1751  * Returns the detect-zeroes setting to be used for bdrv_open() of a
1752  * BlockDriverState which is supposed to inherit the root state.
1753  */
1754 bool blk_get_detect_zeroes_from_root_state(BlockBackend *blk)
1755 {
1756     return blk->root_state.detect_zeroes;
1757 }
1758
1759 /*
1760  * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
1761  * supposed to inherit the root state.
1762  */
1763 int blk_get_open_flags_from_root_state(BlockBackend *blk)
1764 {
1765     int bs_flags;
1766
1767     bs_flags = blk->root_state.read_only ? 0 : BDRV_O_RDWR;
1768     bs_flags |= blk->root_state.open_flags & ~BDRV_O_RDWR;
1769
1770     return bs_flags;
1771 }
1772
1773 BlockBackendRootState *blk_get_root_state(BlockBackend *blk)
1774 {
1775     return &blk->root_state;
1776 }
1777
1778 int blk_commit_all(void)
1779 {
1780     BlockBackend *blk = NULL;
1781
1782     while ((blk = blk_all_next(blk)) != NULL) {
1783         AioContext *aio_context = blk_get_aio_context(blk);
1784
1785         aio_context_acquire(aio_context);
1786         if (blk_is_inserted(blk) && blk->root->bs->backing) {
1787             int ret = bdrv_commit(blk->root->bs);
1788             if (ret < 0) {
1789                 aio_context_release(aio_context);
1790                 return ret;
1791             }
1792         }
1793         aio_context_release(aio_context);
1794     }
1795     return 0;
1796 }
1797
1798
1799 /* throttling disk I/O limits */
1800 void blk_set_io_limits(BlockBackend *blk, ThrottleConfig *cfg)
1801 {
1802     throttle_group_config(blk, cfg);
1803 }
1804
1805 void blk_io_limits_disable(BlockBackend *blk)
1806 {
1807     assert(blk->public.throttle_state);
1808     bdrv_drained_begin(blk_bs(blk));
1809     throttle_group_unregister_blk(blk);
1810     bdrv_drained_end(blk_bs(blk));
1811 }
1812
1813 /* should be called before blk_set_io_limits if a limit is set */
1814 void blk_io_limits_enable(BlockBackend *blk, const char *group)
1815 {
1816     assert(!blk->public.throttle_state);
1817     throttle_group_register_blk(blk, group);
1818 }
1819
1820 void blk_io_limits_update_group(BlockBackend *blk, const char *group)
1821 {
1822     /* this BB is not part of any group */
1823     if (!blk->public.throttle_state) {
1824         return;
1825     }
1826
1827     /* this BB is a part of the same group than the one we want */
1828     if (!g_strcmp0(throttle_group_get_name(blk), group)) {
1829         return;
1830     }
1831
1832     /* need to change the group this bs belong to */
1833     blk_io_limits_disable(blk);
1834     blk_io_limits_enable(blk, group);
1835 }
1836
1837 static void blk_root_drained_begin(BdrvChild *child)
1838 {
1839     BlockBackend *blk = child->opaque;
1840
1841     /* Note that blk->root may not be accessible here yet if we are just
1842      * attaching to a BlockDriverState that is drained. Use child instead. */
1843
1844     if (blk->public.io_limits_disabled++ == 0) {
1845         throttle_group_restart_blk(blk);
1846     }
1847 }
1848
1849 static void blk_root_drained_end(BdrvChild *child)
1850 {
1851     BlockBackend *blk = child->opaque;
1852
1853     assert(blk->public.io_limits_disabled);
1854     --blk->public.io_limits_disabled;
1855 }