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 static char *blk_get_attached_dev_id(BlockBackend *blk);
  84
  85 /* All BlockBackends */
  86 static QTAILQ_HEAD(, BlockBackend) block_backends =
  87     QTAILQ_HEAD_INITIALIZER(block_backends);
  88
  89 /* All BlockBackends referenced by the monitor and which are iterated through by
  90  * blk_next() */
  91 static QTAILQ_HEAD(, BlockBackend) monitor_block_backends =
  92     QTAILQ_HEAD_INITIALIZER(monitor_block_backends);
  93
  94 static void blk_root_inherit_options(int *child_flags, QDict *child_options,
  95                                      int parent_flags, QDict *parent_options)
  96 {
  97     /* We're not supposed to call this function for root nodes */
  98     abort();
  99 }
 100 static void blk_root_drained_begin(BdrvChild *child);
 101 static void blk_root_drained_end(BdrvChild *child);
 102
 103 static void blk_root_change_media(BdrvChild *child, bool load);
 104 static void blk_root_resize(BdrvChild *child);
 105
 106 static char *blk_root_get_parent_desc(BdrvChild *child)
 107 {
 108     BlockBackend *blk = child->opaque;
 109     char *dev_id;
 110
 111     if (blk->name) {
 112         return g_strdup(blk->name);
 113     }
 114
 115     dev_id = blk_get_attached_dev_id(blk);
 116     if (*dev_id) {
 117         return dev_id;
 118     } else {
 119         /* TODO Callback into the BB owner for something more detailed */
 120         g_free(dev_id);
 121         return g_strdup("a block device");
 122     }
 123 }
 124
 125 static const char *blk_root_get_name(BdrvChild *child)
 126 {
 127     return blk_name(child->opaque);
 128 }
 129
 130 static const BdrvChildRole child_root = {
 131     .inherit_options    = blk_root_inherit_options,
 132
 133     .change_media       = blk_root_change_media,
 134     .resize             = blk_root_resize,
 135     .get_name           = blk_root_get_name,
 136     .get_parent_desc    = blk_root_get_parent_desc,
 137
 138     .drained_begin      = blk_root_drained_begin,
 139     .drained_end        = blk_root_drained_end,
 140 };
 141
 142 /*
 143  * Create a new BlockBackend with a reference count of one.
 144  *
 145  * @perm is a bitmasks of BLK_PERM_* constants which describes the permissions
 146  * to request for a block driver node that is attached to this BlockBackend.
 147  * @shared_perm is a bitmask which describes which permissions may be granted
 148  * to other users of the attached node.
 149  * Both sets of permissions can be changed later using blk_set_perm().
 150  *
 151  * Return the new BlockBackend on success, null on failure.
 152  */
 153 BlockBackend *blk_new(uint64_t perm, uint64_t shared_perm)
 154 {
 155     BlockBackend *blk;
 156
 157     blk = g_new0(BlockBackend, 1);
 158     blk->refcnt = 1;
 159     blk->perm = perm;
 160     blk->shared_perm = shared_perm;
 161     blk_set_enable_write_cache(blk, true);
 162
 163     qemu_co_queue_init(&blk->public.throttled_reqs[0]);
 164     qemu_co_queue_init(&blk->public.throttled_reqs[1]);
 165
 166     notifier_list_init(&blk->remove_bs_notifiers);
 167     notifier_list_init(&blk->insert_bs_notifiers);
 168
 169     QTAILQ_INSERT_TAIL(&block_backends, blk, link);
 170     return blk;
 171 }
 172
 173 /*
 174  * Creates a new BlockBackend, opens a new BlockDriverState, and connects both.
 175  *
 176  * Just as with bdrv_open(), after having called this function the reference to
 177  * @options belongs to the block layer (even on failure).
 178  *
 179  * TODO: Remove @filename and @flags; it should be possible to specify a whole
 180  * BDS tree just by specifying the @options QDict (or @reference,
 181  * alternatively). At the time of adding this function, this is not possible,
 182  * though, so callers of this function have to be able to specify @filename and
 183  * @flags.
 184  */
 185 BlockBackend *blk_new_open(const char *filename, const char *reference,
 186                            QDict *options, int flags, Error **errp)
 187 {
 188     BlockBackend *blk;
 189     BlockDriverState *bs;
 190     uint64_t perm;
 191
 192     /* blk_new_open() is mainly used in .bdrv_create implementations and the
 193      * tools where sharing isn't a concern because the BDS stays private, so we
 194      * just request permission according to the flags.
 195      *
 196      * The exceptions are xen_disk and blockdev_init(); in these cases, the
 197      * caller of blk_new_open() doesn't make use of the permissions, but they
 198      * shouldn't hurt either. We can still share everything here because the
 199      * guest devices will add their own blockers if they can't share. */
 200     perm = BLK_PERM_CONSISTENT_READ;
 201     if (flags & BDRV_O_RDWR) {
 202         perm |= BLK_PERM_WRITE;
 203     }
 204     if (flags & BDRV_O_RESIZE) {
 205         perm |= BLK_PERM_RESIZE;
 206     }
 207
 208     blk = blk_new(perm, BLK_PERM_ALL);
 209     bs = bdrv_open(filename, reference, options, flags, errp);
 210     if (!bs) {
 211         blk_unref(blk);
 212         return NULL;
 213     }
 214
 215     blk->root = bdrv_root_attach_child(bs, "root", &child_root,
 216                                        perm, BLK_PERM_ALL, blk, &error_abort);
 217
 218     return blk;
 219 }
 220
 221 static void blk_delete(BlockBackend *blk)
 222 {
 223     assert(!blk->refcnt);
 224     assert(!blk->name);
 225     assert(!blk->dev);
 226     if (blk->root) {
 227         blk_remove_bs(blk);
 228     }
 229     assert(QLIST_EMPTY(&blk->remove_bs_notifiers.notifiers));
 230     assert(QLIST_EMPTY(&blk->insert_bs_notifiers.notifiers));
 231     QTAILQ_REMOVE(&block_backends, blk, link);
 232     drive_info_del(blk->legacy_dinfo);
 233     block_acct_cleanup(&blk->stats);
 234     g_free(blk);
 235 }
 236
 237 static void drive_info_del(DriveInfo *dinfo)
 238 {
 239     if (!dinfo) {
 240         return;
 241     }
 242     qemu_opts_del(dinfo->opts);
 243     g_free(dinfo->serial);
 244     g_free(dinfo);
 245 }
 246
 247 int blk_get_refcnt(BlockBackend *blk)
 248 {
 249     return blk ? blk->refcnt : 0;
 250 }
 251
 252 /*
 253  * Increment @blk's reference count.
 254  * @blk must not be null.
 255  */
 256 void blk_ref(BlockBackend *blk)
 257 {
 258     blk->refcnt++;
 259 }
 260
 261 /*
 262  * Decrement @blk's reference count.
 263  * If this drops it to zero, destroy @blk.
 264  * For convenience, do nothing if @blk is null.
 265  */
 266 void blk_unref(BlockBackend *blk)
 267 {
 268     if (blk) {
 269         assert(blk->refcnt > 0);
 270         if (!--blk->refcnt) {
 271             blk_delete(blk);
 272         }
 273     }
 274 }
 275
 276 /*
 277  * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
 278  * ones which are hidden (i.e. are not referenced by the monitor).
 279  */
 280 static BlockBackend *blk_all_next(BlockBackend *blk)
 281 {
 282     return blk ? QTAILQ_NEXT(blk, link)
 283                : QTAILQ_FIRST(&block_backends);
 284 }
 285
 286 void blk_remove_all_bs(void)
 287 {
 288     BlockBackend *blk = NULL;
 289
 290     while ((blk = blk_all_next(blk)) != NULL) {
 291         AioContext *ctx = blk_get_aio_context(blk);
 292
 293         aio_context_acquire(ctx);
 294         if (blk->root) {
 295             blk_remove_bs(blk);
 296         }
 297         aio_context_release(ctx);
 298     }
 299 }
 300
 301 /*
 302  * Return the monitor-owned BlockBackend after @blk.
 303  * If @blk is null, return the first one.
 304  * Else, return @blk's next sibling, which may be null.
 305  *
 306  * To iterate over all BlockBackends, do
 307  * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
 308  *     ...
 309  * }
 310  */
 311 BlockBackend *blk_next(BlockBackend *blk)
 312 {
 313     return blk ? QTAILQ_NEXT(blk, monitor_link)
 314                : QTAILQ_FIRST(&monitor_block_backends);
 315 }
 316
 317 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
 318  * the monitor or attached to a BlockBackend */
 319 BlockDriverState *bdrv_next(BdrvNextIterator *it)
 320 {
 321     BlockDriverState *bs;
 322
 323     /* First, return all root nodes of BlockBackends. In order to avoid
 324      * returning a BDS twice when multiple BBs refer to it, we only return it
 325      * if the BB is the first one in the parent list of the BDS. */
 326     if (it->phase == BDRV_NEXT_BACKEND_ROOTS) {
 327         do {
 328             it->blk = blk_all_next(it->blk);
 329             bs = it->blk ? blk_bs(it->blk) : NULL;
 330         } while (it->blk && (bs == NULL || bdrv_first_blk(bs) != it->blk));
 331
 332         if (bs) {
 333             return bs;
 334         }
 335         it->phase = BDRV_NEXT_MONITOR_OWNED;
 336     }
 337
 338     /* Then return the monitor-owned BDSes without a BB attached. Ignore all
 339      * BDSes that are attached to a BlockBackend here; they have been handled
 340      * by the above block already */
 341     do {
 342         it->bs = bdrv_next_monitor_owned(it->bs);
 343         bs = it->bs;
 344     } while (bs && bdrv_has_blk(bs));
 345
 346     return bs;
 347 }
 348
 349 BlockDriverState *bdrv_first(BdrvNextIterator *it)
 350 {
 351     *it = (BdrvNextIterator) {
 352         .phase = BDRV_NEXT_BACKEND_ROOTS,
 353     };
 354
 355     return bdrv_next(it);
 356 }
 357
 358 /*
 359  * Add a BlockBackend into the list of backends referenced by the monitor, with
 360  * the given @name acting as the handle for the monitor.
 361  * Strictly for use by blockdev.c.
 362  *
 363  * @name must not be null or empty.
 364  *
 365  * Returns true on success and false on failure. In the latter case, an Error
 366  * object is returned through @errp.
 367  */
 368 bool monitor_add_blk(BlockBackend *blk, const char *name, Error **errp)
 369 {
 370     assert(!blk->name);
 371     assert(name && name[0]);
 372
 373     if (!id_wellformed(name)) {
 374         error_setg(errp, "Invalid device name");
 375         return false;
 376     }
 377     if (blk_by_name(name)) {
 378         error_setg(errp, "Device with id '%s' already exists", name);
 379         return false;
 380     }
 381     if (bdrv_find_node(name)) {
 382         error_setg(errp,
 383                    "Device name '%s' conflicts with an existing node name",
 384                    name);
 385         return false;
 386     }
 387
 388     blk->name = g_strdup(name);
 389     QTAILQ_INSERT_TAIL(&monitor_block_backends, blk, monitor_link);
 390     return true;
 391 }
 392
 393 /*
 394  * Remove a BlockBackend from the list of backends referenced by the monitor.
 395  * Strictly for use by blockdev.c.
 396  */
 397 void monitor_remove_blk(BlockBackend *blk)
 398 {
 399     if (!blk->name) {
 400         return;
 401     }
 402
 403     QTAILQ_REMOVE(&monitor_block_backends, blk, monitor_link);
 404     g_free(blk->name);
 405     blk->name = NULL;
 406 }
 407
 408 /*
 409  * Return @blk's name, a non-null string.
 410  * Returns an empty string iff @blk is not referenced by the monitor.
 411  */
 412 const char *blk_name(BlockBackend *blk)
 413 {
 414     return blk->name ?: "";
 415 }
 416
 417 /*
 418  * Return the BlockBackend with name @name if it exists, else null.
 419  * @name must not be null.
 420  */
 421 BlockBackend *blk_by_name(const char *name)
 422 {
 423     BlockBackend *blk = NULL;
 424
 425     assert(name);
 426     while ((blk = blk_next(blk)) != NULL) {
 427         if (!strcmp(name, blk->name)) {
 428             return blk;
 429         }
 430     }
 431     return NULL;
 432 }
 433
 434 /*
 435  * Return the BlockDriverState attached to @blk if any, else null.
 436  */
 437 BlockDriverState *blk_bs(BlockBackend *blk)
 438 {
 439     return blk->root ? blk->root->bs : NULL;
 440 }
 441
 442 static BlockBackend *bdrv_first_blk(BlockDriverState *bs)
 443 {
 444     BdrvChild *child;
 445     QLIST_FOREACH(child, &bs->parents, next_parent) {
 446         if (child->role == &child_root) {
 447             return child->opaque;
 448         }
 449     }
 450
 451     return NULL;
 452 }
 453
 454 /*
 455  * Returns true if @bs has an associated BlockBackend.
 456  */
 457 bool bdrv_has_blk(BlockDriverState *bs)
 458 {
 459     return bdrv_first_blk(bs) != NULL;
 460 }
 461
 462 /*
 463  * Returns true if @bs has only BlockBackends as parents.
 464  */
 465 bool bdrv_is_root_node(BlockDriverState *bs)
 466 {
 467     BdrvChild *c;
 468
 469     QLIST_FOREACH(c, &bs->parents, next_parent) {
 470         if (c->role != &child_root) {
 471             return false;
 472         }
 473     }
 474
 475     return true;
 476 }
 477
 478 /*
 479  * Return @blk's DriveInfo if any, else null.
 480  */
 481 DriveInfo *blk_legacy_dinfo(BlockBackend *blk)
 482 {
 483     return blk->legacy_dinfo;
 484 }
 485
 486 /*
 487  * Set @blk's DriveInfo to @dinfo, and return it.
 488  * @blk must not have a DriveInfo set already.
 489  * No other BlockBackend may have the same DriveInfo set.
 490  */
 491 DriveInfo *blk_set_legacy_dinfo(BlockBackend *blk, DriveInfo *dinfo)
 492 {
 493     assert(!blk->legacy_dinfo);
 494     return blk->legacy_dinfo = dinfo;
 495 }
 496
 497 /*
 498  * Return the BlockBackend with DriveInfo @dinfo.
 499  * It must exist.
 500  */
 501 BlockBackend *blk_by_legacy_dinfo(DriveInfo *dinfo)
 502 {
 503     BlockBackend *blk = NULL;
 504
 505     while ((blk = blk_next(blk)) != NULL) {
 506         if (blk->legacy_dinfo == dinfo) {
 507             return blk;
 508         }
 509     }
 510     abort();
 511 }
 512
 513 /*
 514  * Returns a pointer to the publicly accessible fields of @blk.
 515  */
 516 BlockBackendPublic *blk_get_public(BlockBackend *blk)
 517 {
 518     return &blk->public;
 519 }
 520
 521 /*
 522  * Returns a BlockBackend given the associated @public fields.
 523  */
 524 BlockBackend *blk_by_public(BlockBackendPublic *public)
 525 {
 526     return container_of(public, BlockBackend, public);
 527 }
 528
 529 /*
 530  * Disassociates the currently associated BlockDriverState from @blk.
 531  */
 532 void blk_remove_bs(BlockBackend *blk)
 533 {
 534     notifier_list_notify(&blk->remove_bs_notifiers, blk);
 535     if (blk->public.throttle_state) {
 536         throttle_timers_detach_aio_context(&blk->public.throttle_timers);
 537     }
 538
 539     blk_update_root_state(blk);
 540
 541     bdrv_root_unref_child(blk->root);
 542     blk->root = NULL;
 543 }
 544
 545 /*
 546  * Associates a new BlockDriverState with @blk.
 547  */
 548 int blk_insert_bs(BlockBackend *blk, BlockDriverState *bs, Error **errp)
 549 {
 550     blk->root = bdrv_root_attach_child(bs, "root", &child_root,
 551                                        blk->perm, blk->shared_perm, blk, errp);
 552     if (blk->root == NULL) {
 553         return -EPERM;
 554     }
 555     bdrv_ref(bs);
 556
 557     notifier_list_notify(&blk->insert_bs_notifiers, blk);
 558     if (blk->public.throttle_state) {
 559         throttle_timers_attach_aio_context(
 560             &blk->public.throttle_timers, bdrv_get_aio_context(bs));
 561     }
 562
 563     return 0;
 564 }
 565
 566 /*
 567  * Sets the permission bitmasks that the user of the BlockBackend needs.
 568  */
 569 int blk_set_perm(BlockBackend *blk, uint64_t perm, uint64_t shared_perm,
 570                  Error **errp)
 571 {
 572     int ret;
 573
 574     if (blk->root) {
 575         ret = bdrv_child_try_set_perm(blk->root, perm, shared_perm, errp);
 576         if (ret < 0) {
 577             return ret;
 578         }
 579     }
 580
 581     blk->perm = perm;
 582     blk->shared_perm = shared_perm;
 583
 584     return 0;
 585 }
 586
 587 void blk_get_perm(BlockBackend *blk, uint64_t *perm, uint64_t *shared_perm)
 588 {
 589     *perm = blk->perm;
 590     *shared_perm = blk->shared_perm;
 591 }
 592
 593 static int blk_do_attach_dev(BlockBackend *blk, void *dev)
 594 {
 595     if (blk->dev) {
 596         return -EBUSY;
 597     }
 598     blk_ref(blk);
 599     blk->dev = dev;
 600     blk->legacy_dev = false;
 601     blk_iostatus_reset(blk);
 602     return 0;
 603 }
 604
 605 /*
 606  * Attach device model @dev to @blk.
 607  * Return 0 on success, -EBUSY when a device model is attached already.
 608  */
 609 int blk_attach_dev(BlockBackend *blk, DeviceState *dev)
 610 {
 611     return blk_do_attach_dev(blk, dev);
 612 }
 613
 614 /*
 615  * Attach device model @dev to @blk.
 616  * @blk must not have a device model attached already.
 617  * TODO qdevified devices don't use this, remove when devices are qdevified
 618  */
 619 void blk_attach_dev_legacy(BlockBackend *blk, void *dev)
 620 {
 621     if (blk_do_attach_dev(blk, dev) < 0) {
 622         abort();
 623     }
 624     blk->legacy_dev = true;
 625 }
 626
 627 /*
 628  * Detach device model @dev from @blk.
 629  * @dev must be currently attached to @blk.
 630  */
 631 void blk_detach_dev(BlockBackend *blk, void *dev)
 632 /* TODO change to DeviceState *dev when all users are qdevified */
 633 {
 634     assert(blk->dev == dev);
 635     blk->dev = NULL;
 636     blk->dev_ops = NULL;
 637     blk->dev_opaque = NULL;
 638     blk->guest_block_size = 512;
 639     blk_set_perm(blk, 0, BLK_PERM_ALL, &error_abort);
 640     blk_unref(blk);
 641 }
 642
 643 /*
 644  * Return the device model attached to @blk if any, else null.
 645  */
 646 void *blk_get_attached_dev(BlockBackend *blk)
 647 /* TODO change to return DeviceState * when all users are qdevified */
 648 {
 649     return blk->dev;
 650 }
 651
 652 /* Return the qdev ID, or if no ID is assigned the QOM path, of the block
 653  * device attached to the BlockBackend. */
 654 static char *blk_get_attached_dev_id(BlockBackend *blk)
 655 {
 656     DeviceState *dev;
 657
 658     assert(!blk->legacy_dev);
 659     dev = blk->dev;
 660
 661     if (!dev) {
 662         return g_strdup("");
 663     } else if (dev->id) {
 664         return g_strdup(dev->id);
 665     }
 666     return object_get_canonical_path(OBJECT(dev));
 667 }
 668
 669 /*
 670  * Return the BlockBackend which has the device model @dev attached if it
 671  * exists, else null.
 672  *
 673  * @dev must not be null.
 674  */
 675 BlockBackend *blk_by_dev(void *dev)
 676 {
 677     BlockBackend *blk = NULL;
 678
 679     assert(dev != NULL);
 680     while ((blk = blk_all_next(blk)) != NULL) {
 681         if (blk->dev == dev) {
 682             return blk;
 683         }
 684     }
 685     return NULL;
 686 }
 687
 688 /*
 689  * Set @blk's device model callbacks to @ops.
 690  * @opaque is the opaque argument to pass to the callbacks.
 691  * This is for use by device models.
 692  */
 693 void blk_set_dev_ops(BlockBackend *blk, const BlockDevOps *ops,
 694                      void *opaque)
 695 {
 696     /* All drivers that use blk_set_dev_ops() are qdevified and we want to keep
 697      * it that way, so we can assume blk->dev is a DeviceState if blk->dev_ops
 698      * is set. */
 699     assert(!blk->legacy_dev);
 700
 701     blk->dev_ops = ops;
 702     blk->dev_opaque = opaque;
 703 }
 704
 705 /*
 706  * Notify @blk's attached device model of media change.
 707  *
 708  * If @load is true, notify of media load. This action can fail, meaning that
 709  * the medium cannot be loaded. @errp is set then.
 710  *
 711  * If @load is false, notify of media eject. This can never fail.
 712  *
 713  * Also send DEVICE_TRAY_MOVED events as appropriate.
 714  */
 715 void blk_dev_change_media_cb(BlockBackend *blk, bool load, Error **errp)
 716 {
 717     if (blk->dev_ops && blk->dev_ops->change_media_cb) {
 718         bool tray_was_open, tray_is_open;
 719         Error *local_err = NULL;
 720
 721         assert(!blk->legacy_dev);
 722
 723         tray_was_open = blk_dev_is_tray_open(blk);
 724         blk->dev_ops->change_media_cb(blk->dev_opaque, load, &local_err);
 725         if (local_err) {
 726             assert(load == true);
 727             error_propagate(errp, local_err);
 728             return;
 729         }
 730         tray_is_open = blk_dev_is_tray_open(blk);
 731
 732         if (tray_was_open != tray_is_open) {
 733             char *id = blk_get_attached_dev_id(blk);
 734             qapi_event_send_device_tray_moved(blk_name(blk), id, tray_is_open,
 735                                               &error_abort);
 736             g_free(id);
 737         }
 738     }
 739 }
 740
 741 static void blk_root_change_media(BdrvChild *child, bool load)
 742 {
 743     blk_dev_change_media_cb(child->opaque, load, NULL);
 744 }
 745
 746 /*
 747  * Does @blk's attached device model have removable media?
 748  * %true if no device model is attached.
 749  */
 750 bool blk_dev_has_removable_media(BlockBackend *blk)
 751 {
 752     return !blk->dev || (blk->dev_ops && blk->dev_ops->change_media_cb);
 753 }
 754
 755 /*
 756  * Does @blk's attached device model have a tray?
 757  */
 758 bool blk_dev_has_tray(BlockBackend *blk)
 759 {
 760     return blk->dev_ops && blk->dev_ops->is_tray_open;
 761 }
 762
 763 /*
 764  * Notify @blk's attached device model of a media eject request.
 765  * If @force is true, the medium is about to be yanked out forcefully.
 766  */
 767 void blk_dev_eject_request(BlockBackend *blk, bool force)
 768 {
 769     if (blk->dev_ops && blk->dev_ops->eject_request_cb) {
 770         blk->dev_ops->eject_request_cb(blk->dev_opaque, force);
 771     }
 772 }
 773
 774 /*
 775  * Does @blk's attached device model have a tray, and is it open?
 776  */
 777 bool blk_dev_is_tray_open(BlockBackend *blk)
 778 {
 779     if (blk_dev_has_tray(blk)) {
 780         return blk->dev_ops->is_tray_open(blk->dev_opaque);
 781     }
 782     return false;
 783 }
 784
 785 /*
 786  * Does @blk's attached device model have the medium locked?
 787  * %false if the device model has no such lock.
 788  */
 789 bool blk_dev_is_medium_locked(BlockBackend *blk)
 790 {
 791     if (blk->dev_ops && blk->dev_ops->is_medium_locked) {
 792         return blk->dev_ops->is_medium_locked(blk->dev_opaque);
 793     }
 794     return false;
 795 }
 796
 797 /*
 798  * Notify @blk's attached device model of a backend size change.
 799  */
 800 static void blk_root_resize(BdrvChild *child)
 801 {
 802     BlockBackend *blk = child->opaque;
 803
 804     if (blk->dev_ops && blk->dev_ops->resize_cb) {
 805         blk->dev_ops->resize_cb(blk->dev_opaque);
 806     }
 807 }
 808
 809 void blk_iostatus_enable(BlockBackend *blk)
 810 {
 811     blk->iostatus_enabled = true;
 812     blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 813 }
 814
 815 /* The I/O status is only enabled if the drive explicitly
 816  * enables it _and_ the VM is configured to stop on errors */
 817 bool blk_iostatus_is_enabled(const BlockBackend *blk)
 818 {
 819     return (blk->iostatus_enabled &&
 820            (blk->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
 821             blk->on_write_error == BLOCKDEV_ON_ERROR_STOP   ||
 822             blk->on_read_error == BLOCKDEV_ON_ERROR_STOP));
 823 }
 824
 825 BlockDeviceIoStatus blk_iostatus(const BlockBackend *blk)
 826 {
 827     return blk->iostatus;
 828 }
 829
 830 void blk_iostatus_disable(BlockBackend *blk)
 831 {
 832     blk->iostatus_enabled = false;
 833 }
 834
 835 void blk_iostatus_reset(BlockBackend *blk)
 836 {
 837     if (blk_iostatus_is_enabled(blk)) {
 838         BlockDriverState *bs = blk_bs(blk);
 839         blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 840         if (bs && bs->job) {
 841             block_job_iostatus_reset(bs->job);
 842         }
 843     }
 844 }
 845
 846 void blk_iostatus_set_err(BlockBackend *blk, int error)
 847 {
 848     assert(blk_iostatus_is_enabled(blk));
 849     if (blk->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 850         blk->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 851                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 852     }
 853 }
 854
 855 void blk_set_allow_write_beyond_eof(BlockBackend *blk, bool allow)
 856 {
 857     blk->allow_write_beyond_eof = allow;
 858 }
 859
 860 static int blk_check_byte_request(BlockBackend *blk, int64_t offset,
 861                                   size_t size)
 862 {
 863     int64_t len;
 864
 865     if (size > INT_MAX) {
 866         return -EIO;
 867     }
 868
 869     if (!blk_is_available(blk)) {
 870         return -ENOMEDIUM;
 871     }
 872
 873     if (offset < 0) {
 874         return -EIO;
 875     }
 876
 877     if (!blk->allow_write_beyond_eof) {
 878         len = blk_getlength(blk);
 879         if (len < 0) {
 880             return len;
 881         }
 882
 883         if (offset > len || len - offset < size) {
 884             return -EIO;
 885         }
 886     }
 887
 888     return 0;
 889 }
 890
 891 int coroutine_fn blk_co_preadv(BlockBackend *blk, int64_t offset,
 892                                unsigned int bytes, QEMUIOVector *qiov,
 893                                BdrvRequestFlags flags)
 894 {
 895     int ret;
 896     BlockDriverState *bs = blk_bs(blk);
 897
 898     trace_blk_co_preadv(blk, bs, offset, bytes, flags);
 899
 900     ret = blk_check_byte_request(blk, offset, bytes);
 901     if (ret < 0) {
 902         return ret;
 903     }
 904
 905     bdrv_inc_in_flight(bs);
 906
 907     /* throttling disk I/O */
 908     if (blk->public.throttle_state) {
 909         throttle_group_co_io_limits_intercept(blk, bytes, false);
 910     }
 911
 912     ret = bdrv_co_preadv(blk->root, offset, bytes, qiov, flags);
 913     bdrv_dec_in_flight(bs);
 914     return ret;
 915 }
 916
 917 int coroutine_fn blk_co_pwritev(BlockBackend *blk, int64_t offset,
 918                                 unsigned int bytes, QEMUIOVector *qiov,
 919                                 BdrvRequestFlags flags)
 920 {
 921     int ret;
 922     BlockDriverState *bs = blk_bs(blk);
 923
 924     trace_blk_co_pwritev(blk, bs, offset, bytes, flags);
 925
 926     ret = blk_check_byte_request(blk, offset, bytes);
 927     if (ret < 0) {
 928         return ret;
 929     }
 930
 931     bdrv_inc_in_flight(bs);
 932
 933     /* throttling disk I/O */
 934     if (blk->public.throttle_state) {
 935         throttle_group_co_io_limits_intercept(blk, bytes, true);
 936     }
 937
 938     if (!blk->enable_write_cache) {
 939         flags |= BDRV_REQ_FUA;
 940     }
 941
 942     ret = bdrv_co_pwritev(blk->root, offset, bytes, qiov, flags);
 943     bdrv_dec_in_flight(bs);
 944     return ret;
 945 }
 946
 947 typedef struct BlkRwCo {
 948     BlockBackend *blk;
 949     int64_t offset;
 950     QEMUIOVector *qiov;
 951     int ret;
 952     BdrvRequestFlags flags;
 953 } BlkRwCo;
 954
 955 static void blk_read_entry(void *opaque)
 956 {
 957     BlkRwCo *rwco = opaque;
 958
 959     rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, rwco->qiov->size,
 960                               rwco->qiov, rwco->flags);
 961 }
 962
 963 static void blk_write_entry(void *opaque)
 964 {
 965     BlkRwCo *rwco = opaque;
 966
 967     rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, rwco->qiov->size,
 968                                rwco->qiov, rwco->flags);
 969 }
 970
 971 static int blk_prw(BlockBackend *blk, int64_t offset, uint8_t *buf,
 972                    int64_t bytes, CoroutineEntry co_entry,
 973                    BdrvRequestFlags flags)
 974 {
 975     QEMUIOVector qiov;
 976     struct iovec iov;
 977     BlkRwCo rwco;
 978
 979     iov = (struct iovec) {
 980         .iov_base = buf,
 981         .iov_len = bytes,
 982     };
 983     qemu_iovec_init_external(&qiov, &iov, 1);
 984
 985     rwco = (BlkRwCo) {
 986         .blk    = blk,
 987         .offset = offset,
 988         .qiov   = &qiov,
 989         .flags  = flags,
 990         .ret    = NOT_DONE,
 991     };
 992
 993     if (qemu_in_coroutine()) {
 994         /* Fast-path if already in coroutine context */
 995         co_entry(&rwco);
 996     } else {
 997         Coroutine *co = qemu_coroutine_create(co_entry, &rwco);
 998         qemu_coroutine_enter(co);
 999         BDRV_POLL_WHILE(blk_bs(blk), rwco.ret == NOT_DONE);
1000     }
1001
1002     return rwco.ret;
1003 }
1004
1005 int blk_pread_unthrottled(BlockBackend *blk, int64_t offset, uint8_t *buf,
1006                           int count)
1007 {
1008     int ret;
1009
1010     ret = blk_check_byte_request(blk, offset, count);
1011     if (ret < 0) {
1012         return ret;
1013     }
1014
1015     blk_root_drained_begin(blk->root);
1016     ret = blk_pread(blk, offset, buf, count);
1017     blk_root_drained_end(blk->root);
1018     return ret;
1019 }
1020
1021 int blk_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1022                       int count, BdrvRequestFlags flags)
1023 {
1024     return blk_prw(blk, offset, NULL, count, blk_write_entry,
1025                    flags | BDRV_REQ_ZERO_WRITE);
1026 }
1027
1028 int blk_make_zero(BlockBackend *blk, BdrvRequestFlags flags)
1029 {
1030     return bdrv_make_zero(blk->root, flags);
1031 }
1032
1033 static void error_callback_bh(void *opaque)
1034 {
1035     struct BlockBackendAIOCB *acb = opaque;
1036
1037     bdrv_dec_in_flight(acb->common.bs);
1038     acb->common.cb(acb->common.opaque, acb->ret);
1039     qemu_aio_unref(acb);
1040 }
1041
1042 BlockAIOCB *blk_abort_aio_request(BlockBackend *blk,
1043                                   BlockCompletionFunc *cb,
1044                                   void *opaque, int ret)
1045 {
1046     struct BlockBackendAIOCB *acb;
1047
1048     bdrv_inc_in_flight(blk_bs(blk));
1049     acb = blk_aio_get(&block_backend_aiocb_info, blk, cb, opaque);
1050     acb->blk = blk;
1051     acb->ret = ret;
1052
1053     aio_bh_schedule_oneshot(blk_get_aio_context(blk), error_callback_bh, acb);
1054     return &acb->common;
1055 }
1056
1057 typedef struct BlkAioEmAIOCB {
1058     BlockAIOCB common;
1059     BlkRwCo rwco;
1060     int bytes;
1061     bool has_returned;
1062 } BlkAioEmAIOCB;
1063
1064 static const AIOCBInfo blk_aio_em_aiocb_info = {
1065     .aiocb_size         = sizeof(BlkAioEmAIOCB),
1066 };
1067
1068 static void blk_aio_complete(BlkAioEmAIOCB *acb)
1069 {
1070     if (acb->has_returned) {
1071         bdrv_dec_in_flight(acb->common.bs);
1072         acb->common.cb(acb->common.opaque, acb->rwco.ret);
1073         qemu_aio_unref(acb);
1074     }
1075 }
1076
1077 static void blk_aio_complete_bh(void *opaque)
1078 {
1079     BlkAioEmAIOCB *acb = opaque;
1080     assert(acb->has_returned);
1081     blk_aio_complete(acb);
1082 }
1083
1084 static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset, int bytes,
1085                                 QEMUIOVector *qiov, CoroutineEntry co_entry,
1086                                 BdrvRequestFlags flags,
1087                                 BlockCompletionFunc *cb, void *opaque)
1088 {
1089     BlkAioEmAIOCB *acb;
1090     Coroutine *co;
1091
1092     bdrv_inc_in_flight(blk_bs(blk));
1093     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1094     acb->rwco = (BlkRwCo) {
1095         .blk    = blk,
1096         .offset = offset,
1097         .qiov   = qiov,
1098         .flags  = flags,
1099         .ret    = NOT_DONE,
1100     };
1101     acb->bytes = bytes;
1102     acb->has_returned = false;
1103
1104     co = qemu_coroutine_create(co_entry, acb);
1105     qemu_coroutine_enter(co);
1106
1107     acb->has_returned = true;
1108     if (acb->rwco.ret != NOT_DONE) {
1109         aio_bh_schedule_oneshot(blk_get_aio_context(blk),
1110                                 blk_aio_complete_bh, acb);
1111     }
1112
1113     return &acb->common;
1114 }
1115
1116 static void blk_aio_read_entry(void *opaque)
1117 {
1118     BlkAioEmAIOCB *acb = opaque;
1119     BlkRwCo *rwco = &acb->rwco;
1120
1121     assert(rwco->qiov->size == acb->bytes);
1122     rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, acb->bytes,
1123                               rwco->qiov, rwco->flags);
1124     blk_aio_complete(acb);
1125 }
1126
1127 static void blk_aio_write_entry(void *opaque)
1128 {
1129     BlkAioEmAIOCB *acb = opaque;
1130     BlkRwCo *rwco = &acb->rwco;
1131
1132     assert(!rwco->qiov || rwco->qiov->size == acb->bytes);
1133     rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, acb->bytes,
1134                                rwco->qiov, rwco->flags);
1135     blk_aio_complete(acb);
1136 }
1137
1138 BlockAIOCB *blk_aio_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1139                                   int count, BdrvRequestFlags flags,
1140                                   BlockCompletionFunc *cb, void *opaque)
1141 {
1142     return blk_aio_prwv(blk, offset, count, NULL, blk_aio_write_entry,
1143                         flags | BDRV_REQ_ZERO_WRITE, cb, opaque);
1144 }
1145
1146 int blk_pread(BlockBackend *blk, int64_t offset, void *buf, int count)
1147 {
1148     int ret = blk_prw(blk, offset, buf, count, blk_read_entry, 0);
1149     if (ret < 0) {
1150         return ret;
1151     }
1152     return count;
1153 }
1154
1155 int blk_pwrite(BlockBackend *blk, int64_t offset, const void *buf, int count,
1156                BdrvRequestFlags flags)
1157 {
1158     int ret = blk_prw(blk, offset, (void *) buf, count, blk_write_entry,
1159                       flags);
1160     if (ret < 0) {
1161         return ret;
1162     }
1163     return count;
1164 }
1165
1166 int64_t blk_getlength(BlockBackend *blk)
1167 {
1168     if (!blk_is_available(blk)) {
1169         return -ENOMEDIUM;
1170     }
1171
1172     return bdrv_getlength(blk_bs(blk));
1173 }
1174
1175 void blk_get_geometry(BlockBackend *blk, uint64_t *nb_sectors_ptr)
1176 {
1177     if (!blk_bs(blk)) {
1178         *nb_sectors_ptr = 0;
1179     } else {
1180         bdrv_get_geometry(blk_bs(blk), nb_sectors_ptr);
1181     }
1182 }
1183
1184 int64_t blk_nb_sectors(BlockBackend *blk)
1185 {
1186     if (!blk_is_available(blk)) {
1187         return -ENOMEDIUM;
1188     }
1189
1190     return bdrv_nb_sectors(blk_bs(blk));
1191 }
1192
1193 BlockAIOCB *blk_aio_preadv(BlockBackend *blk, int64_t offset,
1194                            QEMUIOVector *qiov, BdrvRequestFlags flags,
1195                            BlockCompletionFunc *cb, void *opaque)
1196 {
1197     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1198                         blk_aio_read_entry, flags, cb, opaque);
1199 }
1200
1201 BlockAIOCB *blk_aio_pwritev(BlockBackend *blk, int64_t offset,
1202                             QEMUIOVector *qiov, BdrvRequestFlags flags,
1203                             BlockCompletionFunc *cb, void *opaque)
1204 {
1205     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1206                         blk_aio_write_entry, flags, cb, opaque);
1207 }
1208
1209 static void blk_aio_flush_entry(void *opaque)
1210 {
1211     BlkAioEmAIOCB *acb = opaque;
1212     BlkRwCo *rwco = &acb->rwco;
1213
1214     rwco->ret = blk_co_flush(rwco->blk);
1215     blk_aio_complete(acb);
1216 }
1217
1218 BlockAIOCB *blk_aio_flush(BlockBackend *blk,
1219                           BlockCompletionFunc *cb, void *opaque)
1220 {
1221     return blk_aio_prwv(blk, 0, 0, NULL, blk_aio_flush_entry, 0, cb, opaque);
1222 }
1223
1224 static void blk_aio_pdiscard_entry(void *opaque)
1225 {
1226     BlkAioEmAIOCB *acb = opaque;
1227     BlkRwCo *rwco = &acb->rwco;
1228
1229     rwco->ret = blk_co_pdiscard(rwco->blk, rwco->offset, acb->bytes);
1230     blk_aio_complete(acb);
1231 }
1232
1233 BlockAIOCB *blk_aio_pdiscard(BlockBackend *blk,
1234                              int64_t offset, int count,
1235                              BlockCompletionFunc *cb, void *opaque)
1236 {
1237     return blk_aio_prwv(blk, offset, count, NULL, blk_aio_pdiscard_entry, 0,
1238                         cb, opaque);
1239 }
1240
1241 void blk_aio_cancel(BlockAIOCB *acb)
1242 {
1243     bdrv_aio_cancel(acb);
1244 }
1245
1246 void blk_aio_cancel_async(BlockAIOCB *acb)
1247 {
1248     bdrv_aio_cancel_async(acb);
1249 }
1250
1251 int blk_co_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1252 {
1253     if (!blk_is_available(blk)) {
1254         return -ENOMEDIUM;
1255     }
1256
1257     return bdrv_co_ioctl(blk_bs(blk), req, buf);
1258 }
1259
1260 static void blk_ioctl_entry(void *opaque)
1261 {
1262     BlkRwCo *rwco = opaque;
1263     rwco->ret = blk_co_ioctl(rwco->blk, rwco->offset,
1264                              rwco->qiov->iov[0].iov_base);
1265 }
1266
1267 int blk_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1268 {
1269     return blk_prw(blk, req, buf, 0, blk_ioctl_entry, 0);
1270 }
1271
1272 static void blk_aio_ioctl_entry(void *opaque)
1273 {
1274     BlkAioEmAIOCB *acb = opaque;
1275     BlkRwCo *rwco = &acb->rwco;
1276
1277     rwco->ret = blk_co_ioctl(rwco->blk, rwco->offset,
1278                              rwco->qiov->iov[0].iov_base);
1279     blk_aio_complete(acb);
1280 }
1281
1282 BlockAIOCB *blk_aio_ioctl(BlockBackend *blk, unsigned long int req, void *buf,
1283                           BlockCompletionFunc *cb, void *opaque)
1284 {
1285     QEMUIOVector qiov;
1286     struct iovec iov;
1287
1288     iov = (struct iovec) {
1289         .iov_base = buf,
1290         .iov_len = 0,
1291     };
1292     qemu_iovec_init_external(&qiov, &iov, 1);
1293
1294     return blk_aio_prwv(blk, req, 0, &qiov, blk_aio_ioctl_entry, 0, cb, opaque);
1295 }
1296
1297 int blk_co_pdiscard(BlockBackend *blk, int64_t offset, int count)
1298 {
1299     int ret = blk_check_byte_request(blk, offset, count);
1300     if (ret < 0) {
1301         return ret;
1302     }
1303
1304     return bdrv_co_pdiscard(blk_bs(blk), offset, count);
1305 }
1306
1307 int blk_co_flush(BlockBackend *blk)
1308 {
1309     if (!blk_is_available(blk)) {
1310         return -ENOMEDIUM;
1311     }
1312
1313     return bdrv_co_flush(blk_bs(blk));
1314 }
1315
1316 static void blk_flush_entry(void *opaque)
1317 {
1318     BlkRwCo *rwco = opaque;
1319     rwco->ret = blk_co_flush(rwco->blk);
1320 }
1321
1322 int blk_flush(BlockBackend *blk)
1323 {
1324     return blk_prw(blk, 0, NULL, 0, blk_flush_entry, 0);
1325 }
1326
1327 void blk_drain(BlockBackend *blk)
1328 {
1329     if (blk_bs(blk)) {
1330         bdrv_drain(blk_bs(blk));
1331     }
1332 }
1333
1334 void blk_drain_all(void)
1335 {
1336     bdrv_drain_all();
1337 }
1338
1339 void blk_set_on_error(BlockBackend *blk, BlockdevOnError on_read_error,
1340                       BlockdevOnError on_write_error)
1341 {
1342     blk->on_read_error = on_read_error;
1343     blk->on_write_error = on_write_error;
1344 }
1345
1346 BlockdevOnError blk_get_on_error(BlockBackend *blk, bool is_read)
1347 {
1348     return is_read ? blk->on_read_error : blk->on_write_error;
1349 }
1350
1351 BlockErrorAction blk_get_error_action(BlockBackend *blk, bool is_read,
1352                                       int error)
1353 {
1354     BlockdevOnError on_err = blk_get_on_error(blk, is_read);
1355
1356     switch (on_err) {
1357     case BLOCKDEV_ON_ERROR_ENOSPC:
1358         return (error == ENOSPC) ?
1359                BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
1360     case BLOCKDEV_ON_ERROR_STOP:
1361         return BLOCK_ERROR_ACTION_STOP;
1362     case BLOCKDEV_ON_ERROR_REPORT:
1363         return BLOCK_ERROR_ACTION_REPORT;
1364     case BLOCKDEV_ON_ERROR_IGNORE:
1365         return BLOCK_ERROR_ACTION_IGNORE;
1366     case BLOCKDEV_ON_ERROR_AUTO:
1367     default:
1368         abort();
1369     }
1370 }
1371
1372 static void send_qmp_error_event(BlockBackend *blk,
1373                                  BlockErrorAction action,
1374                                  bool is_read, int error)
1375 {
1376     IoOperationType optype;
1377
1378     optype = is_read ? IO_OPERATION_TYPE_READ : IO_OPERATION_TYPE_WRITE;
1379     qapi_event_send_block_io_error(blk_name(blk),
1380                                    bdrv_get_node_name(blk_bs(blk)), optype,
1381                                    action, blk_iostatus_is_enabled(blk),
1382                                    error == ENOSPC, strerror(error),
1383                                    &error_abort);
1384 }
1385
1386 /* This is done by device models because, while the block layer knows
1387  * about the error, it does not know whether an operation comes from
1388  * the device or the block layer (from a job, for example).
1389  */
1390 void blk_error_action(BlockBackend *blk, BlockErrorAction action,
1391                       bool is_read, int error)
1392 {
1393     assert(error >= 0);
1394
1395     if (action == BLOCK_ERROR_ACTION_STOP) {
1396         /* First set the iostatus, so that "info block" returns an iostatus
1397          * that matches the events raised so far (an additional error iostatus
1398          * is fine, but not a lost one).
1399          */
1400         blk_iostatus_set_err(blk, error);
1401
1402         /* Then raise the request to stop the VM and the event.
1403          * qemu_system_vmstop_request_prepare has two effects.  First,
1404          * it ensures that the STOP event always comes after the
1405          * BLOCK_IO_ERROR event.  Second, it ensures that even if management
1406          * can observe the STOP event and do a "cont" before the STOP
1407          * event is issued, the VM will not stop.  In this case, vm_start()
1408          * also ensures that the STOP/RESUME pair of events is emitted.
1409          */
1410         qemu_system_vmstop_request_prepare();
1411         send_qmp_error_event(blk, action, is_read, error);
1412         qemu_system_vmstop_request(RUN_STATE_IO_ERROR);
1413     } else {
1414         send_qmp_error_event(blk, action, is_read, error);
1415     }
1416 }
1417
1418 int blk_is_read_only(BlockBackend *blk)
1419 {
1420     BlockDriverState *bs = blk_bs(blk);
1421
1422     if (bs) {
1423         return bdrv_is_read_only(bs);
1424     } else {
1425         return blk->root_state.read_only;
1426     }
1427 }
1428
1429 int blk_is_sg(BlockBackend *blk)
1430 {
1431     BlockDriverState *bs = blk_bs(blk);
1432
1433     if (!bs) {
1434         return 0;
1435     }
1436
1437     return bdrv_is_sg(bs);
1438 }
1439
1440 int blk_enable_write_cache(BlockBackend *blk)
1441 {
1442     return blk->enable_write_cache;
1443 }
1444
1445 void blk_set_enable_write_cache(BlockBackend *blk, bool wce)
1446 {
1447     blk->enable_write_cache = wce;
1448 }
1449
1450 void blk_invalidate_cache(BlockBackend *blk, Error **errp)
1451 {
1452     BlockDriverState *bs = blk_bs(blk);
1453
1454     if (!bs) {
1455         error_setg(errp, "Device '%s' has no medium", blk->name);
1456         return;
1457     }
1458
1459     bdrv_invalidate_cache(bs, errp);
1460 }
1461
1462 bool blk_is_inserted(BlockBackend *blk)
1463 {
1464     BlockDriverState *bs = blk_bs(blk);
1465
1466     return bs && bdrv_is_inserted(bs);
1467 }
1468
1469 bool blk_is_available(BlockBackend *blk)
1470 {
1471     return blk_is_inserted(blk) && !blk_dev_is_tray_open(blk);
1472 }
1473
1474 void blk_lock_medium(BlockBackend *blk, bool locked)
1475 {
1476     BlockDriverState *bs = blk_bs(blk);
1477
1478     if (bs) {
1479         bdrv_lock_medium(bs, locked);
1480     }
1481 }
1482
1483 void blk_eject(BlockBackend *blk, bool eject_flag)
1484 {
1485     BlockDriverState *bs = blk_bs(blk);
1486     char *id;
1487
1488     /* blk_eject is only called by qdevified devices */
1489     assert(!blk->legacy_dev);
1490
1491     if (bs) {
1492         bdrv_eject(bs, eject_flag);
1493     }
1494
1495     /* Whether or not we ejected on the backend,
1496      * the frontend experienced a tray event. */
1497     id = blk_get_attached_dev_id(blk);
1498     qapi_event_send_device_tray_moved(blk_name(blk), id,
1499                                       eject_flag, &error_abort);
1500     g_free(id);
1501 }
1502
1503 int blk_get_flags(BlockBackend *blk)
1504 {
1505     BlockDriverState *bs = blk_bs(blk);
1506
1507     if (bs) {
1508         return bdrv_get_flags(bs);
1509     } else {
1510         return blk->root_state.open_flags;
1511     }
1512 }
1513
1514 /* Returns the maximum transfer length, in bytes; guaranteed nonzero */
1515 uint32_t blk_get_max_transfer(BlockBackend *blk)
1516 {
1517     BlockDriverState *bs = blk_bs(blk);
1518     uint32_t max = 0;
1519
1520     if (bs) {
1521         max = bs->bl.max_transfer;
1522     }
1523     return MIN_NON_ZERO(max, INT_MAX);
1524 }
1525
1526 int blk_get_max_iov(BlockBackend *blk)
1527 {
1528     return blk->root->bs->bl.max_iov;
1529 }
1530
1531 void blk_set_guest_block_size(BlockBackend *blk, int align)
1532 {
1533     blk->guest_block_size = align;
1534 }
1535
1536 void *blk_try_blockalign(BlockBackend *blk, size_t size)
1537 {
1538     return qemu_try_blockalign(blk ? blk_bs(blk) : NULL, size);
1539 }
1540
1541 void *blk_blockalign(BlockBackend *blk, size_t size)
1542 {
1543     return qemu_blockalign(blk ? blk_bs(blk) : NULL, size);
1544 }
1545
1546 bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp)
1547 {
1548     BlockDriverState *bs = blk_bs(blk);
1549
1550     if (!bs) {
1551         return false;
1552     }
1553
1554     return bdrv_op_is_blocked(bs, op, errp);
1555 }
1556
1557 void blk_op_unblock(BlockBackend *blk, BlockOpType op, Error *reason)
1558 {
1559     BlockDriverState *bs = blk_bs(blk);
1560
1561     if (bs) {
1562         bdrv_op_unblock(bs, op, reason);
1563     }
1564 }
1565
1566 void blk_op_block_all(BlockBackend *blk, Error *reason)
1567 {
1568     BlockDriverState *bs = blk_bs(blk);
1569
1570     if (bs) {
1571         bdrv_op_block_all(bs, reason);
1572     }
1573 }
1574
1575 void blk_op_unblock_all(BlockBackend *blk, Error *reason)
1576 {
1577     BlockDriverState *bs = blk_bs(blk);
1578
1579     if (bs) {
1580         bdrv_op_unblock_all(bs, reason);
1581     }
1582 }
1583
1584 AioContext *blk_get_aio_context(BlockBackend *blk)
1585 {
1586     BlockDriverState *bs = blk_bs(blk);
1587
1588     if (bs) {
1589         return bdrv_get_aio_context(bs);
1590     } else {
1591         return qemu_get_aio_context();
1592     }
1593 }
1594
1595 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb)
1596 {
1597     BlockBackendAIOCB *blk_acb = DO_UPCAST(BlockBackendAIOCB, common, acb);
1598     return blk_get_aio_context(blk_acb->blk);
1599 }
1600
1601 void blk_set_aio_context(BlockBackend *blk, AioContext *new_context)
1602 {
1603     BlockDriverState *bs = blk_bs(blk);
1604
1605     if (bs) {
1606         if (blk->public.throttle_state) {
1607             throttle_timers_detach_aio_context(&blk->public.throttle_timers);
1608         }
1609         bdrv_set_aio_context(bs, new_context);
1610         if (blk->public.throttle_state) {
1611             throttle_timers_attach_aio_context(&blk->public.throttle_timers,
1612                                                new_context);
1613         }
1614     }
1615 }
1616
1617 void blk_add_aio_context_notifier(BlockBackend *blk,
1618         void (*attached_aio_context)(AioContext *new_context, void *opaque),
1619         void (*detach_aio_context)(void *opaque), void *opaque)
1620 {
1621     BlockDriverState *bs = blk_bs(blk);
1622
1623     if (bs) {
1624         bdrv_add_aio_context_notifier(bs, attached_aio_context,
1625                                       detach_aio_context, opaque);
1626     }
1627 }
1628
1629 void blk_remove_aio_context_notifier(BlockBackend *blk,
1630                                      void (*attached_aio_context)(AioContext *,
1631                                                                   void *),
1632                                      void (*detach_aio_context)(void *),
1633                                      void *opaque)
1634 {
1635     BlockDriverState *bs = blk_bs(blk);
1636
1637     if (bs) {
1638         bdrv_remove_aio_context_notifier(bs, attached_aio_context,
1639                                          detach_aio_context, opaque);
1640     }
1641 }
1642
1643 void blk_add_remove_bs_notifier(BlockBackend *blk, Notifier *notify)
1644 {
1645     notifier_list_add(&blk->remove_bs_notifiers, notify);
1646 }
1647
1648 void blk_add_insert_bs_notifier(BlockBackend *blk, Notifier *notify)
1649 {
1650     notifier_list_add(&blk->insert_bs_notifiers, notify);
1651 }
1652
1653 void blk_io_plug(BlockBackend *blk)
1654 {
1655     BlockDriverState *bs = blk_bs(blk);
1656
1657     if (bs) {
1658         bdrv_io_plug(bs);
1659     }
1660 }
1661
1662 void blk_io_unplug(BlockBackend *blk)
1663 {
1664     BlockDriverState *bs = blk_bs(blk);
1665
1666     if (bs) {
1667         bdrv_io_unplug(bs);
1668     }
1669 }
1670
1671 BlockAcctStats *blk_get_stats(BlockBackend *blk)
1672 {
1673     return &blk->stats;
1674 }
1675
1676 void *blk_aio_get(const AIOCBInfo *aiocb_info, BlockBackend *blk,
1677                   BlockCompletionFunc *cb, void *opaque)
1678 {
1679     return qemu_aio_get(aiocb_info, blk_bs(blk), cb, opaque);
1680 }
1681
1682 int coroutine_fn blk_co_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1683                                       int count, BdrvRequestFlags flags)
1684 {
1685     return blk_co_pwritev(blk, offset, count, NULL,
1686                           flags | BDRV_REQ_ZERO_WRITE);
1687 }
1688
1689 int blk_pwrite_compressed(BlockBackend *blk, int64_t offset, const void *buf,
1690                           int count)
1691 {
1692     return blk_prw(blk, offset, (void *) buf, count, blk_write_entry,
1693                    BDRV_REQ_WRITE_COMPRESSED);
1694 }
1695
1696 int blk_truncate(BlockBackend *blk, int64_t offset)
1697 {
1698     if (!blk_is_available(blk)) {
1699         return -ENOMEDIUM;
1700     }
1701
1702     return bdrv_truncate(blk->root, offset);
1703 }
1704
1705 static void blk_pdiscard_entry(void *opaque)
1706 {
1707     BlkRwCo *rwco = opaque;
1708     rwco->ret = blk_co_pdiscard(rwco->blk, rwco->offset, rwco->qiov->size);
1709 }
1710
1711 int blk_pdiscard(BlockBackend *blk, int64_t offset, int count)
1712 {
1713     return blk_prw(blk, offset, NULL, count, blk_pdiscard_entry, 0);
1714 }
1715
1716 int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
1717                      int64_t pos, int size)
1718 {
1719     int ret;
1720
1721     if (!blk_is_available(blk)) {
1722         return -ENOMEDIUM;
1723     }
1724
1725     ret = bdrv_save_vmstate(blk_bs(blk), buf, pos, size);
1726     if (ret < 0) {
1727         return ret;
1728     }
1729
1730     if (ret == size && !blk->enable_write_cache) {
1731         ret = bdrv_flush(blk_bs(blk));
1732     }
1733
1734     return ret < 0 ? ret : size;
1735 }
1736
1737 int blk_load_vmstate(BlockBackend *blk, uint8_t *buf, int64_t pos, int size)
1738 {
1739     if (!blk_is_available(blk)) {
1740         return -ENOMEDIUM;
1741     }
1742
1743     return bdrv_load_vmstate(blk_bs(blk), buf, pos, size);
1744 }
1745
1746 int blk_probe_blocksizes(BlockBackend *blk, BlockSizes *bsz)
1747 {
1748     if (!blk_is_available(blk)) {
1749         return -ENOMEDIUM;
1750     }
1751
1752     return bdrv_probe_blocksizes(blk_bs(blk), bsz);
1753 }
1754
1755 int blk_probe_geometry(BlockBackend *blk, HDGeometry *geo)
1756 {
1757     if (!blk_is_available(blk)) {
1758         return -ENOMEDIUM;
1759     }
1760
1761     return bdrv_probe_geometry(blk_bs(blk), geo);
1762 }
1763
1764 /*
1765  * Updates the BlockBackendRootState object with data from the currently
1766  * attached BlockDriverState.
1767  */
1768 void blk_update_root_state(BlockBackend *blk)
1769 {
1770     assert(blk->root);
1771
1772     blk->root_state.open_flags    = blk->root->bs->open_flags;
1773     blk->root_state.read_only     = blk->root->bs->read_only;
1774     blk->root_state.detect_zeroes = blk->root->bs->detect_zeroes;
1775 }
1776
1777 /*
1778  * Returns the detect-zeroes setting to be used for bdrv_open() of a
1779  * BlockDriverState which is supposed to inherit the root state.
1780  */
1781 bool blk_get_detect_zeroes_from_root_state(BlockBackend *blk)
1782 {
1783     return blk->root_state.detect_zeroes;
1784 }
1785
1786 /*
1787  * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
1788  * supposed to inherit the root state.
1789  */
1790 int blk_get_open_flags_from_root_state(BlockBackend *blk)
1791 {
1792     int bs_flags;
1793
1794     bs_flags = blk->root_state.read_only ? 0 : BDRV_O_RDWR;
1795     bs_flags |= blk->root_state.open_flags & ~BDRV_O_RDWR;
1796
1797     return bs_flags;
1798 }
1799
1800 BlockBackendRootState *blk_get_root_state(BlockBackend *blk)
1801 {
1802     return &blk->root_state;
1803 }
1804
1805 int blk_commit_all(void)
1806 {
1807     BlockBackend *blk = NULL;
1808
1809     while ((blk = blk_all_next(blk)) != NULL) {
1810         AioContext *aio_context = blk_get_aio_context(blk);
1811
1812         aio_context_acquire(aio_context);
1813         if (blk_is_inserted(blk) && blk->root->bs->backing) {
1814             int ret = bdrv_commit(blk->root->bs);
1815             if (ret < 0) {
1816                 aio_context_release(aio_context);
1817                 return ret;
1818             }
1819         }
1820         aio_context_release(aio_context);
1821     }
1822     return 0;
1823 }
1824
1825
1826 /* throttling disk I/O limits */
1827 void blk_set_io_limits(BlockBackend *blk, ThrottleConfig *cfg)
1828 {
1829     throttle_group_config(blk, cfg);
1830 }
1831
1832 void blk_io_limits_disable(BlockBackend *blk)
1833 {
1834     assert(blk->public.throttle_state);
1835     bdrv_drained_begin(blk_bs(blk));
1836     throttle_group_unregister_blk(blk);
1837     bdrv_drained_end(blk_bs(blk));
1838 }
1839
1840 /* should be called before blk_set_io_limits if a limit is set */
1841 void blk_io_limits_enable(BlockBackend *blk, const char *group)
1842 {
1843     assert(!blk->public.throttle_state);
1844     throttle_group_register_blk(blk, group);
1845 }
1846
1847 void blk_io_limits_update_group(BlockBackend *blk, const char *group)
1848 {
1849     /* this BB is not part of any group */
1850     if (!blk->public.throttle_state) {
1851         return;
1852     }
1853
1854     /* this BB is a part of the same group than the one we want */
1855     if (!g_strcmp0(throttle_group_get_name(blk), group)) {
1856         return;
1857     }
1858
1859     /* need to change the group this bs belong to */
1860     blk_io_limits_disable(blk);
1861     blk_io_limits_enable(blk, group);
1862 }
1863
1864 static void blk_root_drained_begin(BdrvChild *child)
1865 {
1866     BlockBackend *blk = child->opaque;
1867
1868     /* Note that blk->root may not be accessible here yet if we are just
1869      * attaching to a BlockDriverState that is drained. Use child instead. */
1870
1871     if (blk->public.io_limits_disabled++ == 0) {
1872         throttle_group_restart_blk(blk);
1873     }
1874 }
1875
1876 static void blk_root_drained_end(BdrvChild *child)
1877 {
1878     BlockBackend *blk = child->opaque;
1879
1880     assert(blk->public.io_limits_disabled);
1881     --blk->public.io_limits_disabled;
1882 }