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     bool disable_perm;
  65
  66     bool allow_write_beyond_eof;
  67
  68     NotifierList remove_bs_notifiers, insert_bs_notifiers;
  69
  70     int quiesce_counter;
  71 };
  72
  73 typedef struct BlockBackendAIOCB {
  74     BlockAIOCB common;
  75     BlockBackend *blk;
  76     int ret;
  77 } BlockBackendAIOCB;
  78
  79 static const AIOCBInfo block_backend_aiocb_info = {
  80     .get_aio_context = blk_aiocb_get_aio_context,
  81     .aiocb_size = sizeof(BlockBackendAIOCB),
  82 };
  83
  84 static void drive_info_del(DriveInfo *dinfo);
  85 static BlockBackend *bdrv_first_blk(BlockDriverState *bs);
  86
  87 /* All BlockBackends */
  88 static QTAILQ_HEAD(, BlockBackend) block_backends =
  89     QTAILQ_HEAD_INITIALIZER(block_backends);
  90
  91 /* All BlockBackends referenced by the monitor and which are iterated through by
  92  * blk_next() */
  93 static QTAILQ_HEAD(, BlockBackend) monitor_block_backends =
  94     QTAILQ_HEAD_INITIALIZER(monitor_block_backends);
  95
  96 static void blk_root_inherit_options(int *child_flags, QDict *child_options,
  97                                      int parent_flags, QDict *parent_options)
  98 {
  99     /* We're not supposed to call this function for root nodes */
 100     abort();
 101 }
 102 static void blk_root_drained_begin(BdrvChild *child);
 103 static void blk_root_drained_end(BdrvChild *child);
 104
 105 static void blk_root_change_media(BdrvChild *child, bool load);
 106 static void blk_root_resize(BdrvChild *child);
 107
 108 static char *blk_root_get_parent_desc(BdrvChild *child)
 109 {
 110     BlockBackend *blk = child->opaque;
 111     char *dev_id;
 112
 113     if (blk->name) {
 114         return g_strdup(blk->name);
 115     }
 116
 117     dev_id = blk_get_attached_dev_id(blk);
 118     if (*dev_id) {
 119         return dev_id;
 120     } else {
 121         /* TODO Callback into the BB owner for something more detailed */
 122         g_free(dev_id);
 123         return g_strdup("a block device");
 124     }
 125 }
 126
 127 static const char *blk_root_get_name(BdrvChild *child)
 128 {
 129     return blk_name(child->opaque);
 130 }
 131
 132 /*
 133  * Notifies the user of the BlockBackend that migration has completed. qdev
 134  * devices can tighten their permissions in response (specifically revoke
 135  * shared write permissions that we needed for storage migration).
 136  *
 137  * If an error is returned, the VM cannot be allowed to be resumed.
 138  */
 139 static void blk_root_activate(BdrvChild *child, Error **errp)
 140 {
 141     BlockBackend *blk = child->opaque;
 142     Error *local_err = NULL;
 143
 144     if (!blk->disable_perm) {
 145         return;
 146     }
 147
 148     blk->disable_perm = false;
 149
 150     blk_set_perm(blk, blk->perm, blk->shared_perm, &local_err);
 151     if (local_err) {
 152         error_propagate(errp, local_err);
 153         blk->disable_perm = true;
 154         return;
 155     }
 156 }
 157
 158 static int blk_root_inactivate(BdrvChild *child)
 159 {
 160     BlockBackend *blk = child->opaque;
 161
 162     if (blk->disable_perm) {
 163         return 0;
 164     }
 165
 166     /* Only inactivate BlockBackends for guest devices (which are inactive at
 167      * this point because the VM is stopped) and unattached monitor-owned
 168      * BlockBackends. If there is still any other user like a block job, then
 169      * we simply can't inactivate the image. */
 170     if (!blk->dev && !blk_name(blk)[0]) {
 171         return -EPERM;
 172     }
 173
 174     blk->disable_perm = true;
 175     if (blk->root) {
 176         bdrv_child_try_set_perm(blk->root, 0, BLK_PERM_ALL, &error_abort);
 177     }
 178
 179     return 0;
 180 }
 181
 182 static const BdrvChildRole child_root = {
 183     .inherit_options    = blk_root_inherit_options,
 184
 185     .change_media       = blk_root_change_media,
 186     .resize             = blk_root_resize,
 187     .get_name           = blk_root_get_name,
 188     .get_parent_desc    = blk_root_get_parent_desc,
 189
 190     .drained_begin      = blk_root_drained_begin,
 191     .drained_end        = blk_root_drained_end,
 192
 193     .activate           = blk_root_activate,
 194     .inactivate         = blk_root_inactivate,
 195 };
 196
 197 /*
 198  * Create a new BlockBackend with a reference count of one.
 199  *
 200  * @perm is a bitmasks of BLK_PERM_* constants which describes the permissions
 201  * to request for a block driver node that is attached to this BlockBackend.
 202  * @shared_perm is a bitmask which describes which permissions may be granted
 203  * to other users of the attached node.
 204  * Both sets of permissions can be changed later using blk_set_perm().
 205  *
 206  * Return the new BlockBackend on success, null on failure.
 207  */
 208 BlockBackend *blk_new(uint64_t perm, uint64_t shared_perm)
 209 {
 210     BlockBackend *blk;
 211
 212     blk = g_new0(BlockBackend, 1);
 213     blk->refcnt = 1;
 214     blk->perm = perm;
 215     blk->shared_perm = shared_perm;
 216     blk_set_enable_write_cache(blk, true);
 217
 218     qemu_co_mutex_init(&blk->public.throttled_reqs_lock);
 219     qemu_co_queue_init(&blk->public.throttled_reqs[0]);
 220     qemu_co_queue_init(&blk->public.throttled_reqs[1]);
 221     block_acct_init(&blk->stats);
 222
 223     notifier_list_init(&blk->remove_bs_notifiers);
 224     notifier_list_init(&blk->insert_bs_notifiers);
 225
 226     QTAILQ_INSERT_TAIL(&block_backends, blk, link);
 227     return blk;
 228 }
 229
 230 /*
 231  * Creates a new BlockBackend, opens a new BlockDriverState, and connects both.
 232  *
 233  * Just as with bdrv_open(), after having called this function the reference to
 234  * @options belongs to the block layer (even on failure).
 235  *
 236  * TODO: Remove @filename and @flags; it should be possible to specify a whole
 237  * BDS tree just by specifying the @options QDict (or @reference,
 238  * alternatively). At the time of adding this function, this is not possible,
 239  * though, so callers of this function have to be able to specify @filename and
 240  * @flags.
 241  */
 242 BlockBackend *blk_new_open(const char *filename, const char *reference,
 243                            QDict *options, int flags, Error **errp)
 244 {
 245     BlockBackend *blk;
 246     BlockDriverState *bs;
 247     uint64_t perm;
 248
 249     /* blk_new_open() is mainly used in .bdrv_create implementations and the
 250      * tools where sharing isn't a concern because the BDS stays private, so we
 251      * just request permission according to the flags.
 252      *
 253      * The exceptions are xen_disk and blockdev_init(); in these cases, the
 254      * caller of blk_new_open() doesn't make use of the permissions, but they
 255      * shouldn't hurt either. We can still share everything here because the
 256      * guest devices will add their own blockers if they can't share. */
 257     perm = BLK_PERM_CONSISTENT_READ;
 258     if (flags & BDRV_O_RDWR) {
 259         perm |= BLK_PERM_WRITE;
 260     }
 261     if (flags & BDRV_O_RESIZE) {
 262         perm |= BLK_PERM_RESIZE;
 263     }
 264
 265     blk = blk_new(perm, BLK_PERM_ALL);
 266     bs = bdrv_open(filename, reference, options, flags, errp);
 267     if (!bs) {
 268         blk_unref(blk);
 269         return NULL;
 270     }
 271
 272     blk->root = bdrv_root_attach_child(bs, "root", &child_root,
 273                                        perm, BLK_PERM_ALL, blk, errp);
 274     if (!blk->root) {
 275         bdrv_unref(bs);
 276         blk_unref(blk);
 277         return NULL;
 278     }
 279
 280     return blk;
 281 }
 282
 283 static void blk_delete(BlockBackend *blk)
 284 {
 285     assert(!blk->refcnt);
 286     assert(!blk->name);
 287     assert(!blk->dev);
 288     if (blk->public.throttle_state) {
 289         blk_io_limits_disable(blk);
 290     }
 291     if (blk->root) {
 292         blk_remove_bs(blk);
 293     }
 294     assert(QLIST_EMPTY(&blk->remove_bs_notifiers.notifiers));
 295     assert(QLIST_EMPTY(&blk->insert_bs_notifiers.notifiers));
 296     QTAILQ_REMOVE(&block_backends, blk, link);
 297     drive_info_del(blk->legacy_dinfo);
 298     block_acct_cleanup(&blk->stats);
 299     g_free(blk);
 300 }
 301
 302 static void drive_info_del(DriveInfo *dinfo)
 303 {
 304     if (!dinfo) {
 305         return;
 306     }
 307     qemu_opts_del(dinfo->opts);
 308     g_free(dinfo->serial);
 309     g_free(dinfo);
 310 }
 311
 312 int blk_get_refcnt(BlockBackend *blk)
 313 {
 314     return blk ? blk->refcnt : 0;
 315 }
 316
 317 /*
 318  * Increment @blk's reference count.
 319  * @blk must not be null.
 320  */
 321 void blk_ref(BlockBackend *blk)
 322 {
 323     blk->refcnt++;
 324 }
 325
 326 /*
 327  * Decrement @blk's reference count.
 328  * If this drops it to zero, destroy @blk.
 329  * For convenience, do nothing if @blk is null.
 330  */
 331 void blk_unref(BlockBackend *blk)
 332 {
 333     if (blk) {
 334         assert(blk->refcnt > 0);
 335         if (!--blk->refcnt) {
 336             blk_delete(blk);
 337         }
 338     }
 339 }
 340
 341 /*
 342  * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
 343  * ones which are hidden (i.e. are not referenced by the monitor).
 344  */
 345 BlockBackend *blk_all_next(BlockBackend *blk)
 346 {
 347     return blk ? QTAILQ_NEXT(blk, link)
 348                : QTAILQ_FIRST(&block_backends);
 349 }
 350
 351 void blk_remove_all_bs(void)
 352 {
 353     BlockBackend *blk = NULL;
 354
 355     while ((blk = blk_all_next(blk)) != NULL) {
 356         AioContext *ctx = blk_get_aio_context(blk);
 357
 358         aio_context_acquire(ctx);
 359         if (blk->root) {
 360             blk_remove_bs(blk);
 361         }
 362         aio_context_release(ctx);
 363     }
 364 }
 365
 366 /*
 367  * Return the monitor-owned BlockBackend after @blk.
 368  * If @blk is null, return the first one.
 369  * Else, return @blk's next sibling, which may be null.
 370  *
 371  * To iterate over all BlockBackends, do
 372  * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
 373  *     ...
 374  * }
 375  */
 376 BlockBackend *blk_next(BlockBackend *blk)
 377 {
 378     return blk ? QTAILQ_NEXT(blk, monitor_link)
 379                : QTAILQ_FIRST(&monitor_block_backends);
 380 }
 381
 382 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
 383  * the monitor or attached to a BlockBackend */
 384 BlockDriverState *bdrv_next(BdrvNextIterator *it)
 385 {
 386     BlockDriverState *bs;
 387
 388     /* First, return all root nodes of BlockBackends. In order to avoid
 389      * returning a BDS twice when multiple BBs refer to it, we only return it
 390      * if the BB is the first one in the parent list of the BDS. */
 391     if (it->phase == BDRV_NEXT_BACKEND_ROOTS) {
 392         do {
 393             it->blk = blk_all_next(it->blk);
 394             bs = it->blk ? blk_bs(it->blk) : NULL;
 395         } while (it->blk && (bs == NULL || bdrv_first_blk(bs) != it->blk));
 396
 397         if (bs) {
 398             return bs;
 399         }
 400         it->phase = BDRV_NEXT_MONITOR_OWNED;
 401     }
 402
 403     /* Then return the monitor-owned BDSes without a BB attached. Ignore all
 404      * BDSes that are attached to a BlockBackend here; they have been handled
 405      * by the above block already */
 406     do {
 407         it->bs = bdrv_next_monitor_owned(it->bs);
 408         bs = it->bs;
 409     } while (bs && bdrv_has_blk(bs));
 410
 411     return bs;
 412 }
 413
 414 BlockDriverState *bdrv_first(BdrvNextIterator *it)
 415 {
 416     *it = (BdrvNextIterator) {
 417         .phase = BDRV_NEXT_BACKEND_ROOTS,
 418     };
 419
 420     return bdrv_next(it);
 421 }
 422
 423 /*
 424  * Add a BlockBackend into the list of backends referenced by the monitor, with
 425  * the given @name acting as the handle for the monitor.
 426  * Strictly for use by blockdev.c.
 427  *
 428  * @name must not be null or empty.
 429  *
 430  * Returns true on success and false on failure. In the latter case, an Error
 431  * object is returned through @errp.
 432  */
 433 bool monitor_add_blk(BlockBackend *blk, const char *name, Error **errp)
 434 {
 435     assert(!blk->name);
 436     assert(name && name[0]);
 437
 438     if (!id_wellformed(name)) {
 439         error_setg(errp, "Invalid device name");
 440         return false;
 441     }
 442     if (blk_by_name(name)) {
 443         error_setg(errp, "Device with id '%s' already exists", name);
 444         return false;
 445     }
 446     if (bdrv_find_node(name)) {
 447         error_setg(errp,
 448                    "Device name '%s' conflicts with an existing node name",
 449                    name);
 450         return false;
 451     }
 452
 453     blk->name = g_strdup(name);
 454     QTAILQ_INSERT_TAIL(&monitor_block_backends, blk, monitor_link);
 455     return true;
 456 }
 457
 458 /*
 459  * Remove a BlockBackend from the list of backends referenced by the monitor.
 460  * Strictly for use by blockdev.c.
 461  */
 462 void monitor_remove_blk(BlockBackend *blk)
 463 {
 464     if (!blk->name) {
 465         return;
 466     }
 467
 468     QTAILQ_REMOVE(&monitor_block_backends, blk, monitor_link);
 469     g_free(blk->name);
 470     blk->name = NULL;
 471 }
 472
 473 /*
 474  * Return @blk's name, a non-null string.
 475  * Returns an empty string iff @blk is not referenced by the monitor.
 476  */
 477 const char *blk_name(const BlockBackend *blk)
 478 {
 479     return blk->name ?: "";
 480 }
 481
 482 /*
 483  * Return the BlockBackend with name @name if it exists, else null.
 484  * @name must not be null.
 485  */
 486 BlockBackend *blk_by_name(const char *name)
 487 {
 488     BlockBackend *blk = NULL;
 489
 490     assert(name);
 491     while ((blk = blk_next(blk)) != NULL) {
 492         if (!strcmp(name, blk->name)) {
 493             return blk;
 494         }
 495     }
 496     return NULL;
 497 }
 498
 499 /*
 500  * Return the BlockDriverState attached to @blk if any, else null.
 501  */
 502 BlockDriverState *blk_bs(BlockBackend *blk)
 503 {
 504     return blk->root ? blk->root->bs : NULL;
 505 }
 506
 507 static BlockBackend *bdrv_first_blk(BlockDriverState *bs)
 508 {
 509     BdrvChild *child;
 510     QLIST_FOREACH(child, &bs->parents, next_parent) {
 511         if (child->role == &child_root) {
 512             return child->opaque;
 513         }
 514     }
 515
 516     return NULL;
 517 }
 518
 519 /*
 520  * Returns true if @bs has an associated BlockBackend.
 521  */
 522 bool bdrv_has_blk(BlockDriverState *bs)
 523 {
 524     return bdrv_first_blk(bs) != NULL;
 525 }
 526
 527 /*
 528  * Returns true if @bs has only BlockBackends as parents.
 529  */
 530 bool bdrv_is_root_node(BlockDriverState *bs)
 531 {
 532     BdrvChild *c;
 533
 534     QLIST_FOREACH(c, &bs->parents, next_parent) {
 535         if (c->role != &child_root) {
 536             return false;
 537         }
 538     }
 539
 540     return true;
 541 }
 542
 543 /*
 544  * Return @blk's DriveInfo if any, else null.
 545  */
 546 DriveInfo *blk_legacy_dinfo(BlockBackend *blk)
 547 {
 548     return blk->legacy_dinfo;
 549 }
 550
 551 /*
 552  * Set @blk's DriveInfo to @dinfo, and return it.
 553  * @blk must not have a DriveInfo set already.
 554  * No other BlockBackend may have the same DriveInfo set.
 555  */
 556 DriveInfo *blk_set_legacy_dinfo(BlockBackend *blk, DriveInfo *dinfo)
 557 {
 558     assert(!blk->legacy_dinfo);
 559     return blk->legacy_dinfo = dinfo;
 560 }
 561
 562 /*
 563  * Return the BlockBackend with DriveInfo @dinfo.
 564  * It must exist.
 565  */
 566 BlockBackend *blk_by_legacy_dinfo(DriveInfo *dinfo)
 567 {
 568     BlockBackend *blk = NULL;
 569
 570     while ((blk = blk_next(blk)) != NULL) {
 571         if (blk->legacy_dinfo == dinfo) {
 572             return blk;
 573         }
 574     }
 575     abort();
 576 }
 577
 578 /*
 579  * Returns a pointer to the publicly accessible fields of @blk.
 580  */
 581 BlockBackendPublic *blk_get_public(BlockBackend *blk)
 582 {
 583     return &blk->public;
 584 }
 585
 586 /*
 587  * Returns a BlockBackend given the associated @public fields.
 588  */
 589 BlockBackend *blk_by_public(BlockBackendPublic *public)
 590 {
 591     return container_of(public, BlockBackend, public);
 592 }
 593
 594 /*
 595  * Disassociates the currently associated BlockDriverState from @blk.
 596  */
 597 void blk_remove_bs(BlockBackend *blk)
 598 {
 599     notifier_list_notify(&blk->remove_bs_notifiers, blk);
 600     if (blk->public.throttle_state) {
 601         throttle_timers_detach_aio_context(&blk->public.throttle_timers);
 602     }
 603
 604     blk_update_root_state(blk);
 605
 606     bdrv_root_unref_child(blk->root);
 607     blk->root = NULL;
 608 }
 609
 610 /*
 611  * Associates a new BlockDriverState with @blk.
 612  */
 613 int blk_insert_bs(BlockBackend *blk, BlockDriverState *bs, Error **errp)
 614 {
 615     blk->root = bdrv_root_attach_child(bs, "root", &child_root,
 616                                        blk->perm, blk->shared_perm, blk, errp);
 617     if (blk->root == NULL) {
 618         return -EPERM;
 619     }
 620     bdrv_ref(bs);
 621
 622     notifier_list_notify(&blk->insert_bs_notifiers, blk);
 623     if (blk->public.throttle_state) {
 624         throttle_timers_attach_aio_context(
 625             &blk->public.throttle_timers, bdrv_get_aio_context(bs));
 626     }
 627
 628     return 0;
 629 }
 630
 631 /*
 632  * Sets the permission bitmasks that the user of the BlockBackend needs.
 633  */
 634 int blk_set_perm(BlockBackend *blk, uint64_t perm, uint64_t shared_perm,
 635                  Error **errp)
 636 {
 637     int ret;
 638
 639     if (blk->root && !blk->disable_perm) {
 640         ret = bdrv_child_try_set_perm(blk->root, perm, shared_perm, errp);
 641         if (ret < 0) {
 642             return ret;
 643         }
 644     }
 645
 646     blk->perm = perm;
 647     blk->shared_perm = shared_perm;
 648
 649     return 0;
 650 }
 651
 652 void blk_get_perm(BlockBackend *blk, uint64_t *perm, uint64_t *shared_perm)
 653 {
 654     *perm = blk->perm;
 655     *shared_perm = blk->shared_perm;
 656 }
 657
 658 static int blk_do_attach_dev(BlockBackend *blk, void *dev)
 659 {
 660     if (blk->dev) {
 661         return -EBUSY;
 662     }
 663
 664     /* While migration is still incoming, we don't need to apply the
 665      * permissions of guest device BlockBackends. We might still have a block
 666      * job or NBD server writing to the image for storage migration. */
 667     if (runstate_check(RUN_STATE_INMIGRATE)) {
 668         blk->disable_perm = true;
 669     }
 670
 671     blk_ref(blk);
 672     blk->dev = dev;
 673     blk->legacy_dev = false;
 674     blk_iostatus_reset(blk);
 675
 676     return 0;
 677 }
 678
 679 /*
 680  * Attach device model @dev to @blk.
 681  * Return 0 on success, -EBUSY when a device model is attached already.
 682  */
 683 int blk_attach_dev(BlockBackend *blk, DeviceState *dev)
 684 {
 685     return blk_do_attach_dev(blk, dev);
 686 }
 687
 688 /*
 689  * Attach device model @dev to @blk.
 690  * @blk must not have a device model attached already.
 691  * TODO qdevified devices don't use this, remove when devices are qdevified
 692  */
 693 void blk_attach_dev_legacy(BlockBackend *blk, void *dev)
 694 {
 695     if (blk_do_attach_dev(blk, dev) < 0) {
 696         abort();
 697     }
 698     blk->legacy_dev = true;
 699 }
 700
 701 /*
 702  * Detach device model @dev from @blk.
 703  * @dev must be currently attached to @blk.
 704  */
 705 void blk_detach_dev(BlockBackend *blk, void *dev)
 706 /* TODO change to DeviceState *dev when all users are qdevified */
 707 {
 708     assert(blk->dev == dev);
 709     blk->dev = NULL;
 710     blk->dev_ops = NULL;
 711     blk->dev_opaque = NULL;
 712     blk->guest_block_size = 512;
 713     blk_set_perm(blk, 0, BLK_PERM_ALL, &error_abort);
 714     blk_unref(blk);
 715 }
 716
 717 /*
 718  * Return the device model attached to @blk if any, else null.
 719  */
 720 void *blk_get_attached_dev(BlockBackend *blk)
 721 /* TODO change to return DeviceState * when all users are qdevified */
 722 {
 723     return blk->dev;
 724 }
 725
 726 /* Return the qdev ID, or if no ID is assigned the QOM path, of the block
 727  * device attached to the BlockBackend. */
 728 char *blk_get_attached_dev_id(BlockBackend *blk)
 729 {
 730     DeviceState *dev;
 731
 732     assert(!blk->legacy_dev);
 733     dev = blk->dev;
 734
 735     if (!dev) {
 736         return g_strdup("");
 737     } else if (dev->id) {
 738         return g_strdup(dev->id);
 739     }
 740     return object_get_canonical_path(OBJECT(dev));
 741 }
 742
 743 /*
 744  * Return the BlockBackend which has the device model @dev attached if it
 745  * exists, else null.
 746  *
 747  * @dev must not be null.
 748  */
 749 BlockBackend *blk_by_dev(void *dev)
 750 {
 751     BlockBackend *blk = NULL;
 752
 753     assert(dev != NULL);
 754     while ((blk = blk_all_next(blk)) != NULL) {
 755         if (blk->dev == dev) {
 756             return blk;
 757         }
 758     }
 759     return NULL;
 760 }
 761
 762 /*
 763  * Set @blk's device model callbacks to @ops.
 764  * @opaque is the opaque argument to pass to the callbacks.
 765  * This is for use by device models.
 766  */
 767 void blk_set_dev_ops(BlockBackend *blk, const BlockDevOps *ops,
 768                      void *opaque)
 769 {
 770     /* All drivers that use blk_set_dev_ops() are qdevified and we want to keep
 771      * it that way, so we can assume blk->dev, if present, is a DeviceState if
 772      * blk->dev_ops is set. Non-device users may use dev_ops without device. */
 773     assert(!blk->legacy_dev);
 774
 775     blk->dev_ops = ops;
 776     blk->dev_opaque = opaque;
 777
 778     /* Are we currently quiesced? Should we enforce this right now? */
 779     if (blk->quiesce_counter && ops->drained_begin) {
 780         ops->drained_begin(opaque);
 781     }
 782 }
 783
 784 /*
 785  * Notify @blk's attached device model of media change.
 786  *
 787  * If @load is true, notify of media load. This action can fail, meaning that
 788  * the medium cannot be loaded. @errp is set then.
 789  *
 790  * If @load is false, notify of media eject. This can never fail.
 791  *
 792  * Also send DEVICE_TRAY_MOVED events as appropriate.
 793  */
 794 void blk_dev_change_media_cb(BlockBackend *blk, bool load, Error **errp)
 795 {
 796     if (blk->dev_ops && blk->dev_ops->change_media_cb) {
 797         bool tray_was_open, tray_is_open;
 798         Error *local_err = NULL;
 799
 800         assert(!blk->legacy_dev);
 801
 802         tray_was_open = blk_dev_is_tray_open(blk);
 803         blk->dev_ops->change_media_cb(blk->dev_opaque, load, &local_err);
 804         if (local_err) {
 805             assert(load == true);
 806             error_propagate(errp, local_err);
 807             return;
 808         }
 809         tray_is_open = blk_dev_is_tray_open(blk);
 810
 811         if (tray_was_open != tray_is_open) {
 812             char *id = blk_get_attached_dev_id(blk);
 813             qapi_event_send_device_tray_moved(blk_name(blk), id, tray_is_open,
 814                                               &error_abort);
 815             g_free(id);
 816         }
 817     }
 818 }
 819
 820 static void blk_root_change_media(BdrvChild *child, bool load)
 821 {
 822     blk_dev_change_media_cb(child->opaque, load, NULL);
 823 }
 824
 825 /*
 826  * Does @blk's attached device model have removable media?
 827  * %true if no device model is attached.
 828  */
 829 bool blk_dev_has_removable_media(BlockBackend *blk)
 830 {
 831     return !blk->dev || (blk->dev_ops && blk->dev_ops->change_media_cb);
 832 }
 833
 834 /*
 835  * Does @blk's attached device model have a tray?
 836  */
 837 bool blk_dev_has_tray(BlockBackend *blk)
 838 {
 839     return blk->dev_ops && blk->dev_ops->is_tray_open;
 840 }
 841
 842 /*
 843  * Notify @blk's attached device model of a media eject request.
 844  * If @force is true, the medium is about to be yanked out forcefully.
 845  */
 846 void blk_dev_eject_request(BlockBackend *blk, bool force)
 847 {
 848     if (blk->dev_ops && blk->dev_ops->eject_request_cb) {
 849         blk->dev_ops->eject_request_cb(blk->dev_opaque, force);
 850     }
 851 }
 852
 853 /*
 854  * Does @blk's attached device model have a tray, and is it open?
 855  */
 856 bool blk_dev_is_tray_open(BlockBackend *blk)
 857 {
 858     if (blk_dev_has_tray(blk)) {
 859         return blk->dev_ops->is_tray_open(blk->dev_opaque);
 860     }
 861     return false;
 862 }
 863
 864 /*
 865  * Does @blk's attached device model have the medium locked?
 866  * %false if the device model has no such lock.
 867  */
 868 bool blk_dev_is_medium_locked(BlockBackend *blk)
 869 {
 870     if (blk->dev_ops && blk->dev_ops->is_medium_locked) {
 871         return blk->dev_ops->is_medium_locked(blk->dev_opaque);
 872     }
 873     return false;
 874 }
 875
 876 /*
 877  * Notify @blk's attached device model of a backend size change.
 878  */
 879 static void blk_root_resize(BdrvChild *child)
 880 {
 881     BlockBackend *blk = child->opaque;
 882
 883     if (blk->dev_ops && blk->dev_ops->resize_cb) {
 884         blk->dev_ops->resize_cb(blk->dev_opaque);
 885     }
 886 }
 887
 888 void blk_iostatus_enable(BlockBackend *blk)
 889 {
 890     blk->iostatus_enabled = true;
 891     blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 892 }
 893
 894 /* The I/O status is only enabled if the drive explicitly
 895  * enables it _and_ the VM is configured to stop on errors */
 896 bool blk_iostatus_is_enabled(const BlockBackend *blk)
 897 {
 898     return (blk->iostatus_enabled &&
 899            (blk->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
 900             blk->on_write_error == BLOCKDEV_ON_ERROR_STOP   ||
 901             blk->on_read_error == BLOCKDEV_ON_ERROR_STOP));
 902 }
 903
 904 BlockDeviceIoStatus blk_iostatus(const BlockBackend *blk)
 905 {
 906     return blk->iostatus;
 907 }
 908
 909 void blk_iostatus_disable(BlockBackend *blk)
 910 {
 911     blk->iostatus_enabled = false;
 912 }
 913
 914 void blk_iostatus_reset(BlockBackend *blk)
 915 {
 916     if (blk_iostatus_is_enabled(blk)) {
 917         BlockDriverState *bs = blk_bs(blk);
 918         blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 919         if (bs && bs->job) {
 920             block_job_iostatus_reset(bs->job);
 921         }
 922     }
 923 }
 924
 925 void blk_iostatus_set_err(BlockBackend *blk, int error)
 926 {
 927     assert(blk_iostatus_is_enabled(blk));
 928     if (blk->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 929         blk->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 930                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 931     }
 932 }
 933
 934 void blk_set_allow_write_beyond_eof(BlockBackend *blk, bool allow)
 935 {
 936     blk->allow_write_beyond_eof = allow;
 937 }
 938
 939 static int blk_check_byte_request(BlockBackend *blk, int64_t offset,
 940                                   size_t size)
 941 {
 942     int64_t len;
 943
 944     if (size > INT_MAX) {
 945         return -EIO;
 946     }
 947
 948     if (!blk_is_available(blk)) {
 949         return -ENOMEDIUM;
 950     }
 951
 952     if (offset < 0) {
 953         return -EIO;
 954     }
 955
 956     if (!blk->allow_write_beyond_eof) {
 957         len = blk_getlength(blk);
 958         if (len < 0) {
 959             return len;
 960         }
 961
 962         if (offset > len || len - offset < size) {
 963             return -EIO;
 964         }
 965     }
 966
 967     return 0;
 968 }
 969
 970 int coroutine_fn blk_co_preadv(BlockBackend *blk, int64_t offset,
 971                                unsigned int bytes, QEMUIOVector *qiov,
 972                                BdrvRequestFlags flags)
 973 {
 974     int ret;
 975     BlockDriverState *bs = blk_bs(blk);
 976
 977     trace_blk_co_preadv(blk, bs, offset, bytes, flags);
 978
 979     ret = blk_check_byte_request(blk, offset, bytes);
 980     if (ret < 0) {
 981         return ret;
 982     }
 983
 984     bdrv_inc_in_flight(bs);
 985
 986     /* throttling disk I/O */
 987     if (blk->public.throttle_state) {
 988         throttle_group_co_io_limits_intercept(blk, bytes, false);
 989     }
 990
 991     ret = bdrv_co_preadv(blk->root, offset, bytes, qiov, flags);
 992     bdrv_dec_in_flight(bs);
 993     return ret;
 994 }
 995
 996 int coroutine_fn blk_co_pwritev(BlockBackend *blk, int64_t offset,
 997                                 unsigned int bytes, QEMUIOVector *qiov,
 998                                 BdrvRequestFlags flags)
 999 {
1000     int ret;
1001     BlockDriverState *bs = blk_bs(blk);
1002
1003     trace_blk_co_pwritev(blk, bs, offset, bytes, flags);
1004
1005     ret = blk_check_byte_request(blk, offset, bytes);
1006     if (ret < 0) {
1007         return ret;
1008     }
1009
1010     bdrv_inc_in_flight(bs);
1011
1012     /* throttling disk I/O */
1013     if (blk->public.throttle_state) {
1014         throttle_group_co_io_limits_intercept(blk, bytes, true);
1015     }
1016
1017     if (!blk->enable_write_cache) {
1018         flags |= BDRV_REQ_FUA;
1019     }
1020
1021     ret = bdrv_co_pwritev(blk->root, offset, bytes, qiov, flags);
1022     bdrv_dec_in_flight(bs);
1023     return ret;
1024 }
1025
1026 typedef struct BlkRwCo {
1027     BlockBackend *blk;
1028     int64_t offset;
1029     QEMUIOVector *qiov;
1030     int ret;
1031     BdrvRequestFlags flags;
1032 } BlkRwCo;
1033
1034 static void blk_read_entry(void *opaque)
1035 {
1036     BlkRwCo *rwco = opaque;
1037
1038     rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, rwco->qiov->size,
1039                               rwco->qiov, rwco->flags);
1040 }
1041
1042 static void blk_write_entry(void *opaque)
1043 {
1044     BlkRwCo *rwco = opaque;
1045
1046     rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, rwco->qiov->size,
1047                                rwco->qiov, rwco->flags);
1048 }
1049
1050 static int blk_prw(BlockBackend *blk, int64_t offset, uint8_t *buf,
1051                    int64_t bytes, CoroutineEntry co_entry,
1052                    BdrvRequestFlags flags)
1053 {
1054     QEMUIOVector qiov;
1055     struct iovec iov;
1056     BlkRwCo rwco;
1057
1058     iov = (struct iovec) {
1059         .iov_base = buf,
1060         .iov_len = bytes,
1061     };
1062     qemu_iovec_init_external(&qiov, &iov, 1);
1063
1064     rwco = (BlkRwCo) {
1065         .blk    = blk,
1066         .offset = offset,
1067         .qiov   = &qiov,
1068         .flags  = flags,
1069         .ret    = NOT_DONE,
1070     };
1071
1072     if (qemu_in_coroutine()) {
1073         /* Fast-path if already in coroutine context */
1074         co_entry(&rwco);
1075     } else {
1076         Coroutine *co = qemu_coroutine_create(co_entry, &rwco);
1077         bdrv_coroutine_enter(blk_bs(blk), co);
1078         BDRV_POLL_WHILE(blk_bs(blk), rwco.ret == NOT_DONE);
1079     }
1080
1081     return rwco.ret;
1082 }
1083
1084 int blk_pread_unthrottled(BlockBackend *blk, int64_t offset, uint8_t *buf,
1085                           int count)
1086 {
1087     int ret;
1088
1089     ret = blk_check_byte_request(blk, offset, count);
1090     if (ret < 0) {
1091         return ret;
1092     }
1093
1094     blk_root_drained_begin(blk->root);
1095     ret = blk_pread(blk, offset, buf, count);
1096     blk_root_drained_end(blk->root);
1097     return ret;
1098 }
1099
1100 int blk_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1101                       int bytes, BdrvRequestFlags flags)
1102 {
1103     return blk_prw(blk, offset, NULL, bytes, blk_write_entry,
1104                    flags | BDRV_REQ_ZERO_WRITE);
1105 }
1106
1107 int blk_make_zero(BlockBackend *blk, BdrvRequestFlags flags)
1108 {
1109     return bdrv_make_zero(blk->root, flags);
1110 }
1111
1112 static void error_callback_bh(void *opaque)
1113 {
1114     struct BlockBackendAIOCB *acb = opaque;
1115
1116     bdrv_dec_in_flight(acb->common.bs);
1117     acb->common.cb(acb->common.opaque, acb->ret);
1118     qemu_aio_unref(acb);
1119 }
1120
1121 BlockAIOCB *blk_abort_aio_request(BlockBackend *blk,
1122                                   BlockCompletionFunc *cb,
1123                                   void *opaque, int ret)
1124 {
1125     struct BlockBackendAIOCB *acb;
1126
1127     bdrv_inc_in_flight(blk_bs(blk));
1128     acb = blk_aio_get(&block_backend_aiocb_info, blk, cb, opaque);
1129     acb->blk = blk;
1130     acb->ret = ret;
1131
1132     aio_bh_schedule_oneshot(blk_get_aio_context(blk), error_callback_bh, acb);
1133     return &acb->common;
1134 }
1135
1136 typedef struct BlkAioEmAIOCB {
1137     BlockAIOCB common;
1138     BlkRwCo rwco;
1139     int bytes;
1140     bool has_returned;
1141 } BlkAioEmAIOCB;
1142
1143 static const AIOCBInfo blk_aio_em_aiocb_info = {
1144     .aiocb_size         = sizeof(BlkAioEmAIOCB),
1145 };
1146
1147 static void blk_aio_complete(BlkAioEmAIOCB *acb)
1148 {
1149     if (acb->has_returned) {
1150         bdrv_dec_in_flight(acb->common.bs);
1151         acb->common.cb(acb->common.opaque, acb->rwco.ret);
1152         qemu_aio_unref(acb);
1153     }
1154 }
1155
1156 static void blk_aio_complete_bh(void *opaque)
1157 {
1158     BlkAioEmAIOCB *acb = opaque;
1159     assert(acb->has_returned);
1160     blk_aio_complete(acb);
1161 }
1162
1163 static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset, int bytes,
1164                                 QEMUIOVector *qiov, CoroutineEntry co_entry,
1165                                 BdrvRequestFlags flags,
1166                                 BlockCompletionFunc *cb, void *opaque)
1167 {
1168     BlkAioEmAIOCB *acb;
1169     Coroutine *co;
1170
1171     bdrv_inc_in_flight(blk_bs(blk));
1172     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1173     acb->rwco = (BlkRwCo) {
1174         .blk    = blk,
1175         .offset = offset,
1176         .qiov   = qiov,
1177         .flags  = flags,
1178         .ret    = NOT_DONE,
1179     };
1180     acb->bytes = bytes;
1181     acb->has_returned = false;
1182
1183     co = qemu_coroutine_create(co_entry, acb);
1184     bdrv_coroutine_enter(blk_bs(blk), co);
1185
1186     acb->has_returned = true;
1187     if (acb->rwco.ret != NOT_DONE) {
1188         aio_bh_schedule_oneshot(blk_get_aio_context(blk),
1189                                 blk_aio_complete_bh, acb);
1190     }
1191
1192     return &acb->common;
1193 }
1194
1195 static void blk_aio_read_entry(void *opaque)
1196 {
1197     BlkAioEmAIOCB *acb = opaque;
1198     BlkRwCo *rwco = &acb->rwco;
1199
1200     assert(rwco->qiov->size == acb->bytes);
1201     rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, acb->bytes,
1202                               rwco->qiov, rwco->flags);
1203     blk_aio_complete(acb);
1204 }
1205
1206 static void blk_aio_write_entry(void *opaque)
1207 {
1208     BlkAioEmAIOCB *acb = opaque;
1209     BlkRwCo *rwco = &acb->rwco;
1210
1211     assert(!rwco->qiov || rwco->qiov->size == acb->bytes);
1212     rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, acb->bytes,
1213                                rwco->qiov, rwco->flags);
1214     blk_aio_complete(acb);
1215 }
1216
1217 BlockAIOCB *blk_aio_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1218                                   int count, BdrvRequestFlags flags,
1219                                   BlockCompletionFunc *cb, void *opaque)
1220 {
1221     return blk_aio_prwv(blk, offset, count, NULL, blk_aio_write_entry,
1222                         flags | BDRV_REQ_ZERO_WRITE, cb, opaque);
1223 }
1224
1225 int blk_pread(BlockBackend *blk, int64_t offset, void *buf, int count)
1226 {
1227     int ret = blk_prw(blk, offset, buf, count, blk_read_entry, 0);
1228     if (ret < 0) {
1229         return ret;
1230     }
1231     return count;
1232 }
1233
1234 int blk_pwrite(BlockBackend *blk, int64_t offset, const void *buf, int count,
1235                BdrvRequestFlags flags)
1236 {
1237     int ret = blk_prw(blk, offset, (void *) buf, count, blk_write_entry,
1238                       flags);
1239     if (ret < 0) {
1240         return ret;
1241     }
1242     return count;
1243 }
1244
1245 int64_t blk_getlength(BlockBackend *blk)
1246 {
1247     if (!blk_is_available(blk)) {
1248         return -ENOMEDIUM;
1249     }
1250
1251     return bdrv_getlength(blk_bs(blk));
1252 }
1253
1254 void blk_get_geometry(BlockBackend *blk, uint64_t *nb_sectors_ptr)
1255 {
1256     if (!blk_bs(blk)) {
1257         *nb_sectors_ptr = 0;
1258     } else {
1259         bdrv_get_geometry(blk_bs(blk), nb_sectors_ptr);
1260     }
1261 }
1262
1263 int64_t blk_nb_sectors(BlockBackend *blk)
1264 {
1265     if (!blk_is_available(blk)) {
1266         return -ENOMEDIUM;
1267     }
1268
1269     return bdrv_nb_sectors(blk_bs(blk));
1270 }
1271
1272 BlockAIOCB *blk_aio_preadv(BlockBackend *blk, int64_t offset,
1273                            QEMUIOVector *qiov, BdrvRequestFlags flags,
1274                            BlockCompletionFunc *cb, void *opaque)
1275 {
1276     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1277                         blk_aio_read_entry, flags, cb, opaque);
1278 }
1279
1280 BlockAIOCB *blk_aio_pwritev(BlockBackend *blk, int64_t offset,
1281                             QEMUIOVector *qiov, BdrvRequestFlags flags,
1282                             BlockCompletionFunc *cb, void *opaque)
1283 {
1284     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1285                         blk_aio_write_entry, flags, cb, opaque);
1286 }
1287
1288 static void blk_aio_flush_entry(void *opaque)
1289 {
1290     BlkAioEmAIOCB *acb = opaque;
1291     BlkRwCo *rwco = &acb->rwco;
1292
1293     rwco->ret = blk_co_flush(rwco->blk);
1294     blk_aio_complete(acb);
1295 }
1296
1297 BlockAIOCB *blk_aio_flush(BlockBackend *blk,
1298                           BlockCompletionFunc *cb, void *opaque)
1299 {
1300     return blk_aio_prwv(blk, 0, 0, NULL, blk_aio_flush_entry, 0, cb, opaque);
1301 }
1302
1303 static void blk_aio_pdiscard_entry(void *opaque)
1304 {
1305     BlkAioEmAIOCB *acb = opaque;
1306     BlkRwCo *rwco = &acb->rwco;
1307
1308     rwco->ret = blk_co_pdiscard(rwco->blk, rwco->offset, acb->bytes);
1309     blk_aio_complete(acb);
1310 }
1311
1312 BlockAIOCB *blk_aio_pdiscard(BlockBackend *blk,
1313                              int64_t offset, int bytes,
1314                              BlockCompletionFunc *cb, void *opaque)
1315 {
1316     return blk_aio_prwv(blk, offset, bytes, NULL, blk_aio_pdiscard_entry, 0,
1317                         cb, opaque);
1318 }
1319
1320 void blk_aio_cancel(BlockAIOCB *acb)
1321 {
1322     bdrv_aio_cancel(acb);
1323 }
1324
1325 void blk_aio_cancel_async(BlockAIOCB *acb)
1326 {
1327     bdrv_aio_cancel_async(acb);
1328 }
1329
1330 int blk_co_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1331 {
1332     if (!blk_is_available(blk)) {
1333         return -ENOMEDIUM;
1334     }
1335
1336     return bdrv_co_ioctl(blk_bs(blk), req, buf);
1337 }
1338
1339 static void blk_ioctl_entry(void *opaque)
1340 {
1341     BlkRwCo *rwco = opaque;
1342     rwco->ret = blk_co_ioctl(rwco->blk, rwco->offset,
1343                              rwco->qiov->iov[0].iov_base);
1344 }
1345
1346 int blk_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1347 {
1348     return blk_prw(blk, req, buf, 0, blk_ioctl_entry, 0);
1349 }
1350
1351 static void blk_aio_ioctl_entry(void *opaque)
1352 {
1353     BlkAioEmAIOCB *acb = opaque;
1354     BlkRwCo *rwco = &acb->rwco;
1355
1356     rwco->ret = blk_co_ioctl(rwco->blk, rwco->offset,
1357                              rwco->qiov->iov[0].iov_base);
1358     blk_aio_complete(acb);
1359 }
1360
1361 BlockAIOCB *blk_aio_ioctl(BlockBackend *blk, unsigned long int req, void *buf,
1362                           BlockCompletionFunc *cb, void *opaque)
1363 {
1364     QEMUIOVector qiov;
1365     struct iovec iov;
1366
1367     iov = (struct iovec) {
1368         .iov_base = buf,
1369         .iov_len = 0,
1370     };
1371     qemu_iovec_init_external(&qiov, &iov, 1);
1372
1373     return blk_aio_prwv(blk, req, 0, &qiov, blk_aio_ioctl_entry, 0, cb, opaque);
1374 }
1375
1376 int blk_co_pdiscard(BlockBackend *blk, int64_t offset, int bytes)
1377 {
1378     int ret = blk_check_byte_request(blk, offset, bytes);
1379     if (ret < 0) {
1380         return ret;
1381     }
1382
1383     return bdrv_co_pdiscard(blk_bs(blk), offset, bytes);
1384 }
1385
1386 int blk_co_flush(BlockBackend *blk)
1387 {
1388     if (!blk_is_available(blk)) {
1389         return -ENOMEDIUM;
1390     }
1391
1392     return bdrv_co_flush(blk_bs(blk));
1393 }
1394
1395 static void blk_flush_entry(void *opaque)
1396 {
1397     BlkRwCo *rwco = opaque;
1398     rwco->ret = blk_co_flush(rwco->blk);
1399 }
1400
1401 int blk_flush(BlockBackend *blk)
1402 {
1403     return blk_prw(blk, 0, NULL, 0, blk_flush_entry, 0);
1404 }
1405
1406 void blk_drain(BlockBackend *blk)
1407 {
1408     if (blk_bs(blk)) {
1409         bdrv_drain(blk_bs(blk));
1410     }
1411 }
1412
1413 void blk_drain_all(void)
1414 {
1415     bdrv_drain_all();
1416 }
1417
1418 void blk_set_on_error(BlockBackend *blk, BlockdevOnError on_read_error,
1419                       BlockdevOnError on_write_error)
1420 {
1421     blk->on_read_error = on_read_error;
1422     blk->on_write_error = on_write_error;
1423 }
1424
1425 BlockdevOnError blk_get_on_error(BlockBackend *blk, bool is_read)
1426 {
1427     return is_read ? blk->on_read_error : blk->on_write_error;
1428 }
1429
1430 BlockErrorAction blk_get_error_action(BlockBackend *blk, bool is_read,
1431                                       int error)
1432 {
1433     BlockdevOnError on_err = blk_get_on_error(blk, is_read);
1434
1435     switch (on_err) {
1436     case BLOCKDEV_ON_ERROR_ENOSPC:
1437         return (error == ENOSPC) ?
1438                BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
1439     case BLOCKDEV_ON_ERROR_STOP:
1440         return BLOCK_ERROR_ACTION_STOP;
1441     case BLOCKDEV_ON_ERROR_REPORT:
1442         return BLOCK_ERROR_ACTION_REPORT;
1443     case BLOCKDEV_ON_ERROR_IGNORE:
1444         return BLOCK_ERROR_ACTION_IGNORE;
1445     case BLOCKDEV_ON_ERROR_AUTO:
1446     default:
1447         abort();
1448     }
1449 }
1450
1451 static void send_qmp_error_event(BlockBackend *blk,
1452                                  BlockErrorAction action,
1453                                  bool is_read, int error)
1454 {
1455     IoOperationType optype;
1456
1457     optype = is_read ? IO_OPERATION_TYPE_READ : IO_OPERATION_TYPE_WRITE;
1458     qapi_event_send_block_io_error(blk_name(blk),
1459                                    bdrv_get_node_name(blk_bs(blk)), optype,
1460                                    action, blk_iostatus_is_enabled(blk),
1461                                    error == ENOSPC, strerror(error),
1462                                    &error_abort);
1463 }
1464
1465 /* This is done by device models because, while the block layer knows
1466  * about the error, it does not know whether an operation comes from
1467  * the device or the block layer (from a job, for example).
1468  */
1469 void blk_error_action(BlockBackend *blk, BlockErrorAction action,
1470                       bool is_read, int error)
1471 {
1472     assert(error >= 0);
1473
1474     if (action == BLOCK_ERROR_ACTION_STOP) {
1475         /* First set the iostatus, so that "info block" returns an iostatus
1476          * that matches the events raised so far (an additional error iostatus
1477          * is fine, but not a lost one).
1478          */
1479         blk_iostatus_set_err(blk, error);
1480
1481         /* Then raise the request to stop the VM and the event.
1482          * qemu_system_vmstop_request_prepare has two effects.  First,
1483          * it ensures that the STOP event always comes after the
1484          * BLOCK_IO_ERROR event.  Second, it ensures that even if management
1485          * can observe the STOP event and do a "cont" before the STOP
1486          * event is issued, the VM will not stop.  In this case, vm_start()
1487          * also ensures that the STOP/RESUME pair of events is emitted.
1488          */
1489         qemu_system_vmstop_request_prepare();
1490         send_qmp_error_event(blk, action, is_read, error);
1491         qemu_system_vmstop_request(RUN_STATE_IO_ERROR);
1492     } else {
1493         send_qmp_error_event(blk, action, is_read, error);
1494     }
1495 }
1496
1497 int blk_is_read_only(BlockBackend *blk)
1498 {
1499     BlockDriverState *bs = blk_bs(blk);
1500
1501     if (bs) {
1502         return bdrv_is_read_only(bs);
1503     } else {
1504         return blk->root_state.read_only;
1505     }
1506 }
1507
1508 int blk_is_sg(BlockBackend *blk)
1509 {
1510     BlockDriverState *bs = blk_bs(blk);
1511
1512     if (!bs) {
1513         return 0;
1514     }
1515
1516     return bdrv_is_sg(bs);
1517 }
1518
1519 int blk_enable_write_cache(BlockBackend *blk)
1520 {
1521     return blk->enable_write_cache;
1522 }
1523
1524 void blk_set_enable_write_cache(BlockBackend *blk, bool wce)
1525 {
1526     blk->enable_write_cache = wce;
1527 }
1528
1529 void blk_invalidate_cache(BlockBackend *blk, Error **errp)
1530 {
1531     BlockDriverState *bs = blk_bs(blk);
1532
1533     if (!bs) {
1534         error_setg(errp, "Device '%s' has no medium", blk->name);
1535         return;
1536     }
1537
1538     bdrv_invalidate_cache(bs, errp);
1539 }
1540
1541 bool blk_is_inserted(BlockBackend *blk)
1542 {
1543     BlockDriverState *bs = blk_bs(blk);
1544
1545     return bs && bdrv_is_inserted(bs);
1546 }
1547
1548 bool blk_is_available(BlockBackend *blk)
1549 {
1550     return blk_is_inserted(blk) && !blk_dev_is_tray_open(blk);
1551 }
1552
1553 void blk_lock_medium(BlockBackend *blk, bool locked)
1554 {
1555     BlockDriverState *bs = blk_bs(blk);
1556
1557     if (bs) {
1558         bdrv_lock_medium(bs, locked);
1559     }
1560 }
1561
1562 void blk_eject(BlockBackend *blk, bool eject_flag)
1563 {
1564     BlockDriverState *bs = blk_bs(blk);
1565     char *id;
1566
1567     /* blk_eject is only called by qdevified devices */
1568     assert(!blk->legacy_dev);
1569
1570     if (bs) {
1571         bdrv_eject(bs, eject_flag);
1572     }
1573
1574     /* Whether or not we ejected on the backend,
1575      * the frontend experienced a tray event. */
1576     id = blk_get_attached_dev_id(blk);
1577     qapi_event_send_device_tray_moved(blk_name(blk), id,
1578                                       eject_flag, &error_abort);
1579     g_free(id);
1580 }
1581
1582 int blk_get_flags(BlockBackend *blk)
1583 {
1584     BlockDriverState *bs = blk_bs(blk);
1585
1586     if (bs) {
1587         return bdrv_get_flags(bs);
1588     } else {
1589         return blk->root_state.open_flags;
1590     }
1591 }
1592
1593 /* Returns the maximum transfer length, in bytes; guaranteed nonzero */
1594 uint32_t blk_get_max_transfer(BlockBackend *blk)
1595 {
1596     BlockDriverState *bs = blk_bs(blk);
1597     uint32_t max = 0;
1598
1599     if (bs) {
1600         max = bs->bl.max_transfer;
1601     }
1602     return MIN_NON_ZERO(max, INT_MAX);
1603 }
1604
1605 int blk_get_max_iov(BlockBackend *blk)
1606 {
1607     return blk->root->bs->bl.max_iov;
1608 }
1609
1610 void blk_set_guest_block_size(BlockBackend *blk, int align)
1611 {
1612     blk->guest_block_size = align;
1613 }
1614
1615 void *blk_try_blockalign(BlockBackend *blk, size_t size)
1616 {
1617     return qemu_try_blockalign(blk ? blk_bs(blk) : NULL, size);
1618 }
1619
1620 void *blk_blockalign(BlockBackend *blk, size_t size)
1621 {
1622     return qemu_blockalign(blk ? blk_bs(blk) : NULL, size);
1623 }
1624
1625 bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp)
1626 {
1627     BlockDriverState *bs = blk_bs(blk);
1628
1629     if (!bs) {
1630         return false;
1631     }
1632
1633     return bdrv_op_is_blocked(bs, op, errp);
1634 }
1635
1636 void blk_op_unblock(BlockBackend *blk, BlockOpType op, Error *reason)
1637 {
1638     BlockDriverState *bs = blk_bs(blk);
1639
1640     if (bs) {
1641         bdrv_op_unblock(bs, op, reason);
1642     }
1643 }
1644
1645 void blk_op_block_all(BlockBackend *blk, Error *reason)
1646 {
1647     BlockDriverState *bs = blk_bs(blk);
1648
1649     if (bs) {
1650         bdrv_op_block_all(bs, reason);
1651     }
1652 }
1653
1654 void blk_op_unblock_all(BlockBackend *blk, Error *reason)
1655 {
1656     BlockDriverState *bs = blk_bs(blk);
1657
1658     if (bs) {
1659         bdrv_op_unblock_all(bs, reason);
1660     }
1661 }
1662
1663 AioContext *blk_get_aio_context(BlockBackend *blk)
1664 {
1665     BlockDriverState *bs = blk_bs(blk);
1666
1667     if (bs) {
1668         return bdrv_get_aio_context(bs);
1669     } else {
1670         return qemu_get_aio_context();
1671     }
1672 }
1673
1674 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb)
1675 {
1676     BlockBackendAIOCB *blk_acb = DO_UPCAST(BlockBackendAIOCB, common, acb);
1677     return blk_get_aio_context(blk_acb->blk);
1678 }
1679
1680 void blk_set_aio_context(BlockBackend *blk, AioContext *new_context)
1681 {
1682     BlockDriverState *bs = blk_bs(blk);
1683
1684     if (bs) {
1685         if (blk->public.throttle_state) {
1686             throttle_timers_detach_aio_context(&blk->public.throttle_timers);
1687         }
1688         bdrv_set_aio_context(bs, new_context);
1689         if (blk->public.throttle_state) {
1690             throttle_timers_attach_aio_context(&blk->public.throttle_timers,
1691                                                new_context);
1692         }
1693     }
1694 }
1695
1696 void blk_add_aio_context_notifier(BlockBackend *blk,
1697         void (*attached_aio_context)(AioContext *new_context, void *opaque),
1698         void (*detach_aio_context)(void *opaque), void *opaque)
1699 {
1700     BlockDriverState *bs = blk_bs(blk);
1701
1702     if (bs) {
1703         bdrv_add_aio_context_notifier(bs, attached_aio_context,
1704                                       detach_aio_context, opaque);
1705     }
1706 }
1707
1708 void blk_remove_aio_context_notifier(BlockBackend *blk,
1709                                      void (*attached_aio_context)(AioContext *,
1710                                                                   void *),
1711                                      void (*detach_aio_context)(void *),
1712                                      void *opaque)
1713 {
1714     BlockDriverState *bs = blk_bs(blk);
1715
1716     if (bs) {
1717         bdrv_remove_aio_context_notifier(bs, attached_aio_context,
1718                                          detach_aio_context, opaque);
1719     }
1720 }
1721
1722 void blk_add_remove_bs_notifier(BlockBackend *blk, Notifier *notify)
1723 {
1724     notifier_list_add(&blk->remove_bs_notifiers, notify);
1725 }
1726
1727 void blk_add_insert_bs_notifier(BlockBackend *blk, Notifier *notify)
1728 {
1729     notifier_list_add(&blk->insert_bs_notifiers, notify);
1730 }
1731
1732 void blk_io_plug(BlockBackend *blk)
1733 {
1734     BlockDriverState *bs = blk_bs(blk);
1735
1736     if (bs) {
1737         bdrv_io_plug(bs);
1738     }
1739 }
1740
1741 void blk_io_unplug(BlockBackend *blk)
1742 {
1743     BlockDriverState *bs = blk_bs(blk);
1744
1745     if (bs) {
1746         bdrv_io_unplug(bs);
1747     }
1748 }
1749
1750 BlockAcctStats *blk_get_stats(BlockBackend *blk)
1751 {
1752     return &blk->stats;
1753 }
1754
1755 void *blk_aio_get(const AIOCBInfo *aiocb_info, BlockBackend *blk,
1756                   BlockCompletionFunc *cb, void *opaque)
1757 {
1758     return qemu_aio_get(aiocb_info, blk_bs(blk), cb, opaque);
1759 }
1760
1761 int coroutine_fn blk_co_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1762                                       int bytes, BdrvRequestFlags flags)
1763 {
1764     return blk_co_pwritev(blk, offset, bytes, NULL,
1765                           flags | BDRV_REQ_ZERO_WRITE);
1766 }
1767
1768 int blk_pwrite_compressed(BlockBackend *blk, int64_t offset, const void *buf,
1769                           int count)
1770 {
1771     return blk_prw(blk, offset, (void *) buf, count, blk_write_entry,
1772                    BDRV_REQ_WRITE_COMPRESSED);
1773 }
1774
1775 int blk_truncate(BlockBackend *blk, int64_t offset, PreallocMode prealloc,
1776                  Error **errp)
1777 {
1778     if (!blk_is_available(blk)) {
1779         error_setg(errp, "No medium inserted");
1780         return -ENOMEDIUM;
1781     }
1782
1783     return bdrv_truncate(blk->root, offset, prealloc, errp);
1784 }
1785
1786 static void blk_pdiscard_entry(void *opaque)
1787 {
1788     BlkRwCo *rwco = opaque;
1789     rwco->ret = blk_co_pdiscard(rwco->blk, rwco->offset, rwco->qiov->size);
1790 }
1791
1792 int blk_pdiscard(BlockBackend *blk, int64_t offset, int bytes)
1793 {
1794     return blk_prw(blk, offset, NULL, bytes, blk_pdiscard_entry, 0);
1795 }
1796
1797 int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
1798                      int64_t pos, int size)
1799 {
1800     int ret;
1801
1802     if (!blk_is_available(blk)) {
1803         return -ENOMEDIUM;
1804     }
1805
1806     ret = bdrv_save_vmstate(blk_bs(blk), buf, pos, size);
1807     if (ret < 0) {
1808         return ret;
1809     }
1810
1811     if (ret == size && !blk->enable_write_cache) {
1812         ret = bdrv_flush(blk_bs(blk));
1813     }
1814
1815     return ret < 0 ? ret : size;
1816 }
1817
1818 int blk_load_vmstate(BlockBackend *blk, uint8_t *buf, int64_t pos, int size)
1819 {
1820     if (!blk_is_available(blk)) {
1821         return -ENOMEDIUM;
1822     }
1823
1824     return bdrv_load_vmstate(blk_bs(blk), buf, pos, size);
1825 }
1826
1827 int blk_probe_blocksizes(BlockBackend *blk, BlockSizes *bsz)
1828 {
1829     if (!blk_is_available(blk)) {
1830         return -ENOMEDIUM;
1831     }
1832
1833     return bdrv_probe_blocksizes(blk_bs(blk), bsz);
1834 }
1835
1836 int blk_probe_geometry(BlockBackend *blk, HDGeometry *geo)
1837 {
1838     if (!blk_is_available(blk)) {
1839         return -ENOMEDIUM;
1840     }
1841
1842     return bdrv_probe_geometry(blk_bs(blk), geo);
1843 }
1844
1845 /*
1846  * Updates the BlockBackendRootState object with data from the currently
1847  * attached BlockDriverState.
1848  */
1849 void blk_update_root_state(BlockBackend *blk)
1850 {
1851     assert(blk->root);
1852
1853     blk->root_state.open_flags    = blk->root->bs->open_flags;
1854     blk->root_state.read_only     = blk->root->bs->read_only;
1855     blk->root_state.detect_zeroes = blk->root->bs->detect_zeroes;
1856 }
1857
1858 /*
1859  * Returns the detect-zeroes setting to be used for bdrv_open() of a
1860  * BlockDriverState which is supposed to inherit the root state.
1861  */
1862 bool blk_get_detect_zeroes_from_root_state(BlockBackend *blk)
1863 {
1864     return blk->root_state.detect_zeroes;
1865 }
1866
1867 /*
1868  * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
1869  * supposed to inherit the root state.
1870  */
1871 int blk_get_open_flags_from_root_state(BlockBackend *blk)
1872 {
1873     int bs_flags;
1874
1875     bs_flags = blk->root_state.read_only ? 0 : BDRV_O_RDWR;
1876     bs_flags |= blk->root_state.open_flags & ~BDRV_O_RDWR;
1877
1878     return bs_flags;
1879 }
1880
1881 BlockBackendRootState *blk_get_root_state(BlockBackend *blk)
1882 {
1883     return &blk->root_state;
1884 }
1885
1886 int blk_commit_all(void)
1887 {
1888     BlockBackend *blk = NULL;
1889
1890     while ((blk = blk_all_next(blk)) != NULL) {
1891         AioContext *aio_context = blk_get_aio_context(blk);
1892
1893         aio_context_acquire(aio_context);
1894         if (blk_is_inserted(blk) && blk->root->bs->backing) {
1895             int ret = bdrv_commit(blk->root->bs);
1896             if (ret < 0) {
1897                 aio_context_release(aio_context);
1898                 return ret;
1899             }
1900         }
1901         aio_context_release(aio_context);
1902     }
1903     return 0;
1904 }
1905
1906
1907 /* throttling disk I/O limits */
1908 void blk_set_io_limits(BlockBackend *blk, ThrottleConfig *cfg)
1909 {
1910     throttle_group_config(blk, cfg);
1911 }
1912
1913 void blk_io_limits_disable(BlockBackend *blk)
1914 {
1915     assert(blk->public.throttle_state);
1916     bdrv_drained_begin(blk_bs(blk));
1917     throttle_group_unregister_blk(blk);
1918     bdrv_drained_end(blk_bs(blk));
1919 }
1920
1921 /* should be called before blk_set_io_limits if a limit is set */
1922 void blk_io_limits_enable(BlockBackend *blk, const char *group)
1923 {
1924     assert(!blk->public.throttle_state);
1925     throttle_group_register_blk(blk, group);
1926 }
1927
1928 void blk_io_limits_update_group(BlockBackend *blk, const char *group)
1929 {
1930     /* this BB is not part of any group */
1931     if (!blk->public.throttle_state) {
1932         return;
1933     }
1934
1935     /* this BB is a part of the same group than the one we want */
1936     if (!g_strcmp0(throttle_group_get_name(blk), group)) {
1937         return;
1938     }
1939
1940     /* need to change the group this bs belong to */
1941     blk_io_limits_disable(blk);
1942     blk_io_limits_enable(blk, group);
1943 }
1944
1945 static void blk_root_drained_begin(BdrvChild *child)
1946 {
1947     BlockBackend *blk = child->opaque;
1948
1949     if (++blk->quiesce_counter == 1) {
1950         if (blk->dev_ops && blk->dev_ops->drained_begin) {
1951             blk->dev_ops->drained_begin(blk->dev_opaque);
1952         }
1953     }
1954
1955     /* Note that blk->root may not be accessible here yet if we are just
1956      * attaching to a BlockDriverState that is drained. Use child instead. */
1957
1958     if (atomic_fetch_inc(&blk->public.io_limits_disabled) == 0) {
1959         throttle_group_restart_blk(blk);
1960     }
1961 }
1962
1963 static void blk_root_drained_end(BdrvChild *child)
1964 {
1965     BlockBackend *blk = child->opaque;
1966     assert(blk->quiesce_counter);
1967
1968     assert(blk->public.io_limits_disabled);
1969     atomic_dec(&blk->public.io_limits_disabled);
1970
1971     if (--blk->quiesce_counter == 0) {
1972         if (blk->dev_ops && blk->dev_ops->drained_end) {
1973             blk->dev_ops->drained_end(blk->dev_opaque);
1974         }
1975     }
1976 }