block/block-backend.c

   1 /*
   2  * QEMU Block backends
   3  *
   4  * Copyright (C) 2014-2016 Red Hat, Inc.
   5  *
   6  * Authors:
   7  *  Markus Armbruster <armbru@redhat.com>,
   8  *
   9  * This work is licensed under the terms of the GNU LGPL, version 2.1
  10  * or later.  See the COPYING.LIB file in the top-level directory.
  11  */
  12
  13 #include "qemu/osdep.h"
  14 #include "sysemu/block-backend.h"
  15 #include "block/block_int.h"
  16 #include "block/blockjob.h"
  17 #include "block/throttle-groups.h"
  18 #include "sysemu/blockdev.h"
  19 #include "sysemu/sysemu.h"
  20 #include "qapi-event.h"
  21 #include "qemu/id.h"
  22 #include "trace.h"
  23
  24 /* Number of coroutines to reserve per attached device model */
  25 #define COROUTINE_POOL_RESERVATION 64
  26
  27 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
  28
  29 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb);
  30
  31 struct BlockBackend {
  32     char *name;
  33     int refcnt;
  34     BdrvChild *root;
  35     DriveInfo *legacy_dinfo;    /* null unless created by drive_new() */
  36     QTAILQ_ENTRY(BlockBackend) link;         /* for block_backends */
  37     QTAILQ_ENTRY(BlockBackend) monitor_link; /* for monitor_block_backends */
  38     BlockBackendPublic public;
  39
  40     void *dev;                  /* attached device model, if any */
  41     bool legacy_dev;            /* true if dev is not a DeviceState */
  42     /* TODO change to DeviceState when all users are qdevified */
  43     const BlockDevOps *dev_ops;
  44     void *dev_opaque;
  45
  46     /* the block size for which the guest device expects atomicity */
  47     int guest_block_size;
  48
  49     /* If the BDS tree is removed, some of its options are stored here (which
  50      * can be used to restore those options in the new BDS on insert) */
  51     BlockBackendRootState root_state;
  52
  53     bool enable_write_cache;
  54
  55     /* I/O stats (display with "info blockstats"). */
  56     BlockAcctStats stats;
  57
  58     BlockdevOnError on_read_error, on_write_error;
  59     bool iostatus_enabled;
  60     BlockDeviceIoStatus iostatus;
  61
  62     uint64_t perm;
  63     uint64_t shared_perm;
  64
  65     bool allow_write_beyond_eof;
  66
  67     NotifierList remove_bs_notifiers, insert_bs_notifiers;
  68 };
  69
  70 typedef struct BlockBackendAIOCB {
  71     BlockAIOCB common;
  72     BlockBackend *blk;
  73     int ret;
  74 } BlockBackendAIOCB;
  75
  76 static const AIOCBInfo block_backend_aiocb_info = {
  77     .get_aio_context = blk_aiocb_get_aio_context,
  78     .aiocb_size = sizeof(BlockBackendAIOCB),
  79 };
  80
  81 static void drive_info_del(DriveInfo *dinfo);
  82 static BlockBackend *bdrv_first_blk(BlockDriverState *bs);
  83
  84 /* All BlockBackends */
  85 static QTAILQ_HEAD(, BlockBackend) block_backends =
  86     QTAILQ_HEAD_INITIALIZER(block_backends);
  87
  88 /* All BlockBackends referenced by the monitor and which are iterated through by
  89  * blk_next() */
  90 static QTAILQ_HEAD(, BlockBackend) monitor_block_backends =
  91     QTAILQ_HEAD_INITIALIZER(monitor_block_backends);
  92
  93 static void blk_root_inherit_options(int *child_flags, QDict *child_options,
  94                                      int parent_flags, QDict *parent_options)
  95 {
  96     /* We're not supposed to call this function for root nodes */
  97     abort();
  98 }
  99 static void blk_root_drained_begin(BdrvChild *child);
 100 static void blk_root_drained_end(BdrvChild *child);
 101
 102 static void blk_root_change_media(BdrvChild *child, bool load);
 103 static void blk_root_resize(BdrvChild *child);
 104
 105 static const char *blk_root_get_name(BdrvChild *child)
 106 {
 107     return blk_name(child->opaque);
 108 }
 109
 110 static const BdrvChildRole child_root = {
 111     .inherit_options    = blk_root_inherit_options,
 112
 113     .change_media       = blk_root_change_media,
 114     .resize             = blk_root_resize,
 115     .get_name           = blk_root_get_name,
 116
 117     .drained_begin      = blk_root_drained_begin,
 118     .drained_end        = blk_root_drained_end,
 119 };
 120
 121 /*
 122  * Create a new BlockBackend with a reference count of one.
 123  * Store an error through @errp on failure, unless it's null.
 124  * Return the new BlockBackend on success, null on failure.
 125  */
 126 BlockBackend *blk_new(void)
 127 {
 128     BlockBackend *blk;
 129
 130     blk = g_new0(BlockBackend, 1);
 131     blk->refcnt = 1;
 132     blk->perm = 0;
 133     blk->shared_perm = BLK_PERM_ALL;
 134     blk_set_enable_write_cache(blk, true);
 135
 136     qemu_co_queue_init(&blk->public.throttled_reqs[0]);
 137     qemu_co_queue_init(&blk->public.throttled_reqs[1]);
 138
 139     notifier_list_init(&blk->remove_bs_notifiers);
 140     notifier_list_init(&blk->insert_bs_notifiers);
 141
 142     QTAILQ_INSERT_TAIL(&block_backends, blk, link);
 143     return blk;
 144 }
 145
 146 /*
 147  * Creates a new BlockBackend, opens a new BlockDriverState, and connects both.
 148  *
 149  * Just as with bdrv_open(), after having called this function the reference to
 150  * @options belongs to the block layer (even on failure).
 151  *
 152  * TODO: Remove @filename and @flags; it should be possible to specify a whole
 153  * BDS tree just by specifying the @options QDict (or @reference,
 154  * alternatively). At the time of adding this function, this is not possible,
 155  * though, so callers of this function have to be able to specify @filename and
 156  * @flags.
 157  */
 158 BlockBackend *blk_new_open(const char *filename, const char *reference,
 159                            QDict *options, int flags, Error **errp)
 160 {
 161     BlockBackend *blk;
 162     BlockDriverState *bs;
 163
 164     blk = blk_new();
 165     bs = bdrv_open(filename, reference, options, flags, errp);
 166     if (!bs) {
 167         blk_unref(blk);
 168         return NULL;
 169     }
 170
 171     /* FIXME Use real permissions */
 172     blk->root = bdrv_root_attach_child(bs, "root", &child_root,
 173                                        0, BLK_PERM_ALL, blk, &error_abort);
 174
 175     return blk;
 176 }
 177
 178 static void blk_delete(BlockBackend *blk)
 179 {
 180     assert(!blk->refcnt);
 181     assert(!blk->name);
 182     assert(!blk->dev);
 183     if (blk->root) {
 184         blk_remove_bs(blk);
 185     }
 186     assert(QLIST_EMPTY(&blk->remove_bs_notifiers.notifiers));
 187     assert(QLIST_EMPTY(&blk->insert_bs_notifiers.notifiers));
 188     QTAILQ_REMOVE(&block_backends, blk, link);
 189     drive_info_del(blk->legacy_dinfo);
 190     block_acct_cleanup(&blk->stats);
 191     g_free(blk);
 192 }
 193
 194 static void drive_info_del(DriveInfo *dinfo)
 195 {
 196     if (!dinfo) {
 197         return;
 198     }
 199     qemu_opts_del(dinfo->opts);
 200     g_free(dinfo->serial);
 201     g_free(dinfo);
 202 }
 203
 204 int blk_get_refcnt(BlockBackend *blk)
 205 {
 206     return blk ? blk->refcnt : 0;
 207 }
 208
 209 /*
 210  * Increment @blk's reference count.
 211  * @blk must not be null.
 212  */
 213 void blk_ref(BlockBackend *blk)
 214 {
 215     blk->refcnt++;
 216 }
 217
 218 /*
 219  * Decrement @blk's reference count.
 220  * If this drops it to zero, destroy @blk.
 221  * For convenience, do nothing if @blk is null.
 222  */
 223 void blk_unref(BlockBackend *blk)
 224 {
 225     if (blk) {
 226         assert(blk->refcnt > 0);
 227         if (!--blk->refcnt) {
 228             blk_delete(blk);
 229         }
 230     }
 231 }
 232
 233 /*
 234  * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
 235  * ones which are hidden (i.e. are not referenced by the monitor).
 236  */
 237 static BlockBackend *blk_all_next(BlockBackend *blk)
 238 {
 239     return blk ? QTAILQ_NEXT(blk, link)
 240                : QTAILQ_FIRST(&block_backends);
 241 }
 242
 243 void blk_remove_all_bs(void)
 244 {
 245     BlockBackend *blk = NULL;
 246
 247     while ((blk = blk_all_next(blk)) != NULL) {
 248         AioContext *ctx = blk_get_aio_context(blk);
 249
 250         aio_context_acquire(ctx);
 251         if (blk->root) {
 252             blk_remove_bs(blk);
 253         }
 254         aio_context_release(ctx);
 255     }
 256 }
 257
 258 /*
 259  * Return the monitor-owned BlockBackend after @blk.
 260  * If @blk is null, return the first one.
 261  * Else, return @blk's next sibling, which may be null.
 262  *
 263  * To iterate over all BlockBackends, do
 264  * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
 265  *     ...
 266  * }
 267  */
 268 BlockBackend *blk_next(BlockBackend *blk)
 269 {
 270     return blk ? QTAILQ_NEXT(blk, monitor_link)
 271                : QTAILQ_FIRST(&monitor_block_backends);
 272 }
 273
 274 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
 275  * the monitor or attached to a BlockBackend */
 276 BlockDriverState *bdrv_next(BdrvNextIterator *it)
 277 {
 278     BlockDriverState *bs;
 279
 280     /* First, return all root nodes of BlockBackends. In order to avoid
 281      * returning a BDS twice when multiple BBs refer to it, we only return it
 282      * if the BB is the first one in the parent list of the BDS. */
 283     if (it->phase == BDRV_NEXT_BACKEND_ROOTS) {
 284         do {
 285             it->blk = blk_all_next(it->blk);
 286             bs = it->blk ? blk_bs(it->blk) : NULL;
 287         } while (it->blk && (bs == NULL || bdrv_first_blk(bs) != it->blk));
 288
 289         if (bs) {
 290             return bs;
 291         }
 292         it->phase = BDRV_NEXT_MONITOR_OWNED;
 293     }
 294
 295     /* Then return the monitor-owned BDSes without a BB attached. Ignore all
 296      * BDSes that are attached to a BlockBackend here; they have been handled
 297      * by the above block already */
 298     do {
 299         it->bs = bdrv_next_monitor_owned(it->bs);
 300         bs = it->bs;
 301     } while (bs && bdrv_has_blk(bs));
 302
 303     return bs;
 304 }
 305
 306 BlockDriverState *bdrv_first(BdrvNextIterator *it)
 307 {
 308     *it = (BdrvNextIterator) {
 309         .phase = BDRV_NEXT_BACKEND_ROOTS,
 310     };
 311
 312     return bdrv_next(it);
 313 }
 314
 315 /*
 316  * Add a BlockBackend into the list of backends referenced by the monitor, with
 317  * the given @name acting as the handle for the monitor.
 318  * Strictly for use by blockdev.c.
 319  *
 320  * @name must not be null or empty.
 321  *
 322  * Returns true on success and false on failure. In the latter case, an Error
 323  * object is returned through @errp.
 324  */
 325 bool monitor_add_blk(BlockBackend *blk, const char *name, Error **errp)
 326 {
 327     assert(!blk->name);
 328     assert(name && name[0]);
 329
 330     if (!id_wellformed(name)) {
 331         error_setg(errp, "Invalid device name");
 332         return false;
 333     }
 334     if (blk_by_name(name)) {
 335         error_setg(errp, "Device with id '%s' already exists", name);
 336         return false;
 337     }
 338     if (bdrv_find_node(name)) {
 339         error_setg(errp,
 340                    "Device name '%s' conflicts with an existing node name",
 341                    name);
 342         return false;
 343     }
 344
 345     blk->name = g_strdup(name);
 346     QTAILQ_INSERT_TAIL(&monitor_block_backends, blk, monitor_link);
 347     return true;
 348 }
 349
 350 /*
 351  * Remove a BlockBackend from the list of backends referenced by the monitor.
 352  * Strictly for use by blockdev.c.
 353  */
 354 void monitor_remove_blk(BlockBackend *blk)
 355 {
 356     if (!blk->name) {
 357         return;
 358     }
 359
 360     QTAILQ_REMOVE(&monitor_block_backends, blk, monitor_link);
 361     g_free(blk->name);
 362     blk->name = NULL;
 363 }
 364
 365 /*
 366  * Return @blk's name, a non-null string.
 367  * Returns an empty string iff @blk is not referenced by the monitor.
 368  */
 369 const char *blk_name(BlockBackend *blk)
 370 {
 371     return blk->name ?: "";
 372 }
 373
 374 /*
 375  * Return the BlockBackend with name @name if it exists, else null.
 376  * @name must not be null.
 377  */
 378 BlockBackend *blk_by_name(const char *name)
 379 {
 380     BlockBackend *blk = NULL;
 381
 382     assert(name);
 383     while ((blk = blk_next(blk)) != NULL) {
 384         if (!strcmp(name, blk->name)) {
 385             return blk;
 386         }
 387     }
 388     return NULL;
 389 }
 390
 391 /*
 392  * Return the BlockDriverState attached to @blk if any, else null.
 393  */
 394 BlockDriverState *blk_bs(BlockBackend *blk)
 395 {
 396     return blk->root ? blk->root->bs : NULL;
 397 }
 398
 399 static BlockBackend *bdrv_first_blk(BlockDriverState *bs)
 400 {
 401     BdrvChild *child;
 402     QLIST_FOREACH(child, &bs->parents, next_parent) {
 403         if (child->role == &child_root) {
 404             return child->opaque;
 405         }
 406     }
 407
 408     return NULL;
 409 }
 410
 411 /*
 412  * Returns true if @bs has an associated BlockBackend.
 413  */
 414 bool bdrv_has_blk(BlockDriverState *bs)
 415 {
 416     return bdrv_first_blk(bs) != NULL;
 417 }
 418
 419 /*
 420  * Returns true if @bs has only BlockBackends as parents.
 421  */
 422 bool bdrv_is_root_node(BlockDriverState *bs)
 423 {
 424     BdrvChild *c;
 425
 426     QLIST_FOREACH(c, &bs->parents, next_parent) {
 427         if (c->role != &child_root) {
 428             return false;
 429         }
 430     }
 431
 432     return true;
 433 }
 434
 435 /*
 436  * Return @blk's DriveInfo if any, else null.
 437  */
 438 DriveInfo *blk_legacy_dinfo(BlockBackend *blk)
 439 {
 440     return blk->legacy_dinfo;
 441 }
 442
 443 /*
 444  * Set @blk's DriveInfo to @dinfo, and return it.
 445  * @blk must not have a DriveInfo set already.
 446  * No other BlockBackend may have the same DriveInfo set.
 447  */
 448 DriveInfo *blk_set_legacy_dinfo(BlockBackend *blk, DriveInfo *dinfo)
 449 {
 450     assert(!blk->legacy_dinfo);
 451     return blk->legacy_dinfo = dinfo;
 452 }
 453
 454 /*
 455  * Return the BlockBackend with DriveInfo @dinfo.
 456  * It must exist.
 457  */
 458 BlockBackend *blk_by_legacy_dinfo(DriveInfo *dinfo)
 459 {
 460     BlockBackend *blk = NULL;
 461
 462     while ((blk = blk_next(blk)) != NULL) {
 463         if (blk->legacy_dinfo == dinfo) {
 464             return blk;
 465         }
 466     }
 467     abort();
 468 }
 469
 470 /*
 471  * Returns a pointer to the publicly accessible fields of @blk.
 472  */
 473 BlockBackendPublic *blk_get_public(BlockBackend *blk)
 474 {
 475     return &blk->public;
 476 }
 477
 478 /*
 479  * Returns a BlockBackend given the associated @public fields.
 480  */
 481 BlockBackend *blk_by_public(BlockBackendPublic *public)
 482 {
 483     return container_of(public, BlockBackend, public);
 484 }
 485
 486 /*
 487  * Disassociates the currently associated BlockDriverState from @blk.
 488  */
 489 void blk_remove_bs(BlockBackend *blk)
 490 {
 491     notifier_list_notify(&blk->remove_bs_notifiers, blk);
 492     if (blk->public.throttle_state) {
 493         throttle_timers_detach_aio_context(&blk->public.throttle_timers);
 494     }
 495
 496     blk_update_root_state(blk);
 497
 498     bdrv_root_unref_child(blk->root);
 499     blk->root = NULL;
 500 }
 501
 502 /*
 503  * Associates a new BlockDriverState with @blk.
 504  */
 505 void blk_insert_bs(BlockBackend *blk, BlockDriverState *bs)
 506 {
 507     bdrv_ref(bs);
 508     /* FIXME Use real permissions */
 509     blk->root = bdrv_root_attach_child(bs, "root", &child_root,
 510                                        0, BLK_PERM_ALL, blk, &error_abort);
 511
 512     notifier_list_notify(&blk->insert_bs_notifiers, blk);
 513     if (blk->public.throttle_state) {
 514         throttle_timers_attach_aio_context(
 515             &blk->public.throttle_timers, bdrv_get_aio_context(bs));
 516     }
 517 }
 518
 519 /*
 520  * Sets the permission bitmasks that the user of the BlockBackend needs.
 521  */
 522 int blk_set_perm(BlockBackend *blk, uint64_t perm, uint64_t shared_perm,
 523                  Error **errp)
 524 {
 525     int ret;
 526
 527     if (blk->root) {
 528         ret = bdrv_child_try_set_perm(blk->root, perm, shared_perm, errp);
 529         if (ret < 0) {
 530             return ret;
 531         }
 532     }
 533
 534     blk->perm = perm;
 535     blk->shared_perm = shared_perm;
 536
 537     return 0;
 538 }
 539
 540 static int blk_do_attach_dev(BlockBackend *blk, void *dev)
 541 {
 542     if (blk->dev) {
 543         return -EBUSY;
 544     }
 545     blk_ref(blk);
 546     blk->dev = dev;
 547     blk->legacy_dev = false;
 548     blk_iostatus_reset(blk);
 549     return 0;
 550 }
 551
 552 /*
 553  * Attach device model @dev to @blk.
 554  * Return 0 on success, -EBUSY when a device model is attached already.
 555  */
 556 int blk_attach_dev(BlockBackend *blk, DeviceState *dev)
 557 {
 558     return blk_do_attach_dev(blk, dev);
 559 }
 560
 561 /*
 562  * Attach device model @dev to @blk.
 563  * @blk must not have a device model attached already.
 564  * TODO qdevified devices don't use this, remove when devices are qdevified
 565  */
 566 void blk_attach_dev_legacy(BlockBackend *blk, void *dev)
 567 {
 568     if (blk_do_attach_dev(blk, dev) < 0) {
 569         abort();
 570     }
 571     blk->legacy_dev = true;
 572 }
 573
 574 /*
 575  * Detach device model @dev from @blk.
 576  * @dev must be currently attached to @blk.
 577  */
 578 void blk_detach_dev(BlockBackend *blk, void *dev)
 579 /* TODO change to DeviceState *dev when all users are qdevified */
 580 {
 581     assert(blk->dev == dev);
 582     blk->dev = NULL;
 583     blk->dev_ops = NULL;
 584     blk->dev_opaque = NULL;
 585     blk->guest_block_size = 512;
 586     blk_set_perm(blk, 0, BLK_PERM_ALL, &error_abort);
 587     blk_unref(blk);
 588 }
 589
 590 /*
 591  * Return the device model attached to @blk if any, else null.
 592  */
 593 void *blk_get_attached_dev(BlockBackend *blk)
 594 /* TODO change to return DeviceState * when all users are qdevified */
 595 {
 596     return blk->dev;
 597 }
 598
 599 /* Return the qdev ID, or if no ID is assigned the QOM path, of the block
 600  * device attached to the BlockBackend. */
 601 static char *blk_get_attached_dev_id(BlockBackend *blk)
 602 {
 603     DeviceState *dev;
 604
 605     assert(!blk->legacy_dev);
 606     dev = blk->dev;
 607
 608     if (!dev) {
 609         return g_strdup("");
 610     } else if (dev->id) {
 611         return g_strdup(dev->id);
 612     }
 613     return object_get_canonical_path(OBJECT(dev));
 614 }
 615
 616 /*
 617  * Return the BlockBackend which has the device model @dev attached if it
 618  * exists, else null.
 619  *
 620  * @dev must not be null.
 621  */
 622 BlockBackend *blk_by_dev(void *dev)
 623 {
 624     BlockBackend *blk = NULL;
 625
 626     assert(dev != NULL);
 627     while ((blk = blk_all_next(blk)) != NULL) {
 628         if (blk->dev == dev) {
 629             return blk;
 630         }
 631     }
 632     return NULL;
 633 }
 634
 635 /*
 636  * Set @blk's device model callbacks to @ops.
 637  * @opaque is the opaque argument to pass to the callbacks.
 638  * This is for use by device models.
 639  */
 640 void blk_set_dev_ops(BlockBackend *blk, const BlockDevOps *ops,
 641                      void *opaque)
 642 {
 643     /* All drivers that use blk_set_dev_ops() are qdevified and we want to keep
 644      * it that way, so we can assume blk->dev is a DeviceState if blk->dev_ops
 645      * is set. */
 646     assert(!blk->legacy_dev);
 647
 648     blk->dev_ops = ops;
 649     blk->dev_opaque = opaque;
 650 }
 651
 652 /*
 653  * Notify @blk's attached device model of media change.
 654  * If @load is true, notify of media load.
 655  * Else, notify of media eject.
 656  * Also send DEVICE_TRAY_MOVED events as appropriate.
 657  */
 658 void blk_dev_change_media_cb(BlockBackend *blk, bool load)
 659 {
 660     if (blk->dev_ops && blk->dev_ops->change_media_cb) {
 661         bool tray_was_open, tray_is_open;
 662
 663         assert(!blk->legacy_dev);
 664
 665         tray_was_open = blk_dev_is_tray_open(blk);
 666         blk->dev_ops->change_media_cb(blk->dev_opaque, load);
 667         tray_is_open = blk_dev_is_tray_open(blk);
 668
 669         if (tray_was_open != tray_is_open) {
 670             char *id = blk_get_attached_dev_id(blk);
 671             qapi_event_send_device_tray_moved(blk_name(blk), id, tray_is_open,
 672                                               &error_abort);
 673             g_free(id);
 674         }
 675     }
 676 }
 677
 678 static void blk_root_change_media(BdrvChild *child, bool load)
 679 {
 680     blk_dev_change_media_cb(child->opaque, load);
 681 }
 682
 683 /*
 684  * Does @blk's attached device model have removable media?
 685  * %true if no device model is attached.
 686  */
 687 bool blk_dev_has_removable_media(BlockBackend *blk)
 688 {
 689     return !blk->dev || (blk->dev_ops && blk->dev_ops->change_media_cb);
 690 }
 691
 692 /*
 693  * Does @blk's attached device model have a tray?
 694  */
 695 bool blk_dev_has_tray(BlockBackend *blk)
 696 {
 697     return blk->dev_ops && blk->dev_ops->is_tray_open;
 698 }
 699
 700 /*
 701  * Notify @blk's attached device model of a media eject request.
 702  * If @force is true, the medium is about to be yanked out forcefully.
 703  */
 704 void blk_dev_eject_request(BlockBackend *blk, bool force)
 705 {
 706     if (blk->dev_ops && blk->dev_ops->eject_request_cb) {
 707         blk->dev_ops->eject_request_cb(blk->dev_opaque, force);
 708     }
 709 }
 710
 711 /*
 712  * Does @blk's attached device model have a tray, and is it open?
 713  */
 714 bool blk_dev_is_tray_open(BlockBackend *blk)
 715 {
 716     if (blk_dev_has_tray(blk)) {
 717         return blk->dev_ops->is_tray_open(blk->dev_opaque);
 718     }
 719     return false;
 720 }
 721
 722 /*
 723  * Does @blk's attached device model have the medium locked?
 724  * %false if the device model has no such lock.
 725  */
 726 bool blk_dev_is_medium_locked(BlockBackend *blk)
 727 {
 728     if (blk->dev_ops && blk->dev_ops->is_medium_locked) {
 729         return blk->dev_ops->is_medium_locked(blk->dev_opaque);
 730     }
 731     return false;
 732 }
 733
 734 /*
 735  * Notify @blk's attached device model of a backend size change.
 736  */
 737 static void blk_root_resize(BdrvChild *child)
 738 {
 739     BlockBackend *blk = child->opaque;
 740
 741     if (blk->dev_ops && blk->dev_ops->resize_cb) {
 742         blk->dev_ops->resize_cb(blk->dev_opaque);
 743     }
 744 }
 745
 746 void blk_iostatus_enable(BlockBackend *blk)
 747 {
 748     blk->iostatus_enabled = true;
 749     blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 750 }
 751
 752 /* The I/O status is only enabled if the drive explicitly
 753  * enables it _and_ the VM is configured to stop on errors */
 754 bool blk_iostatus_is_enabled(const BlockBackend *blk)
 755 {
 756     return (blk->iostatus_enabled &&
 757            (blk->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
 758             blk->on_write_error == BLOCKDEV_ON_ERROR_STOP   ||
 759             blk->on_read_error == BLOCKDEV_ON_ERROR_STOP));
 760 }
 761
 762 BlockDeviceIoStatus blk_iostatus(const BlockBackend *blk)
 763 {
 764     return blk->iostatus;
 765 }
 766
 767 void blk_iostatus_disable(BlockBackend *blk)
 768 {
 769     blk->iostatus_enabled = false;
 770 }
 771
 772 void blk_iostatus_reset(BlockBackend *blk)
 773 {
 774     if (blk_iostatus_is_enabled(blk)) {
 775         BlockDriverState *bs = blk_bs(blk);
 776         blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 777         if (bs && bs->job) {
 778             block_job_iostatus_reset(bs->job);
 779         }
 780     }
 781 }
 782
 783 void blk_iostatus_set_err(BlockBackend *blk, int error)
 784 {
 785     assert(blk_iostatus_is_enabled(blk));
 786     if (blk->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 787         blk->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 788                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 789     }
 790 }
 791
 792 void blk_set_allow_write_beyond_eof(BlockBackend *blk, bool allow)
 793 {
 794     blk->allow_write_beyond_eof = allow;
 795 }
 796
 797 static int blk_check_byte_request(BlockBackend *blk, int64_t offset,
 798                                   size_t size)
 799 {
 800     int64_t len;
 801
 802     if (size > INT_MAX) {
 803         return -EIO;
 804     }
 805
 806     if (!blk_is_available(blk)) {
 807         return -ENOMEDIUM;
 808     }
 809
 810     if (offset < 0) {
 811         return -EIO;
 812     }
 813
 814     if (!blk->allow_write_beyond_eof) {
 815         len = blk_getlength(blk);
 816         if (len < 0) {
 817             return len;
 818         }
 819
 820         if (offset > len || len - offset < size) {
 821             return -EIO;
 822         }
 823     }
 824
 825     return 0;
 826 }
 827
 828 int coroutine_fn blk_co_preadv(BlockBackend *blk, int64_t offset,
 829                                unsigned int bytes, QEMUIOVector *qiov,
 830                                BdrvRequestFlags flags)
 831 {
 832     int ret;
 833     BlockDriverState *bs = blk_bs(blk);
 834
 835     trace_blk_co_preadv(blk, bs, offset, bytes, flags);
 836
 837     ret = blk_check_byte_request(blk, offset, bytes);
 838     if (ret < 0) {
 839         return ret;
 840     }
 841
 842     bdrv_inc_in_flight(bs);
 843
 844     /* throttling disk I/O */
 845     if (blk->public.throttle_state) {
 846         throttle_group_co_io_limits_intercept(blk, bytes, false);
 847     }
 848
 849     ret = bdrv_co_preadv(blk->root, offset, bytes, qiov, flags);
 850     bdrv_dec_in_flight(bs);
 851     return ret;
 852 }
 853
 854 int coroutine_fn blk_co_pwritev(BlockBackend *blk, int64_t offset,
 855                                 unsigned int bytes, QEMUIOVector *qiov,
 856                                 BdrvRequestFlags flags)
 857 {
 858     int ret;
 859     BlockDriverState *bs = blk_bs(blk);
 860
 861     trace_blk_co_pwritev(blk, bs, offset, bytes, flags);
 862
 863     ret = blk_check_byte_request(blk, offset, bytes);
 864     if (ret < 0) {
 865         return ret;
 866     }
 867
 868     bdrv_inc_in_flight(bs);
 869
 870     /* throttling disk I/O */
 871     if (blk->public.throttle_state) {
 872         throttle_group_co_io_limits_intercept(blk, bytes, true);
 873     }
 874
 875     if (!blk->enable_write_cache) {
 876         flags |= BDRV_REQ_FUA;
 877     }
 878
 879     ret = bdrv_co_pwritev(blk->root, offset, bytes, qiov, flags);
 880     bdrv_dec_in_flight(bs);
 881     return ret;
 882 }
 883
 884 typedef struct BlkRwCo {
 885     BlockBackend *blk;
 886     int64_t offset;
 887     QEMUIOVector *qiov;
 888     int ret;
 889     BdrvRequestFlags flags;
 890 } BlkRwCo;
 891
 892 static void blk_read_entry(void *opaque)
 893 {
 894     BlkRwCo *rwco = opaque;
 895
 896     rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, rwco->qiov->size,
 897                               rwco->qiov, rwco->flags);
 898 }
 899
 900 static void blk_write_entry(void *opaque)
 901 {
 902     BlkRwCo *rwco = opaque;
 903
 904     rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, rwco->qiov->size,
 905                                rwco->qiov, rwco->flags);
 906 }
 907
 908 static int blk_prw(BlockBackend *blk, int64_t offset, uint8_t *buf,
 909                    int64_t bytes, CoroutineEntry co_entry,
 910                    BdrvRequestFlags flags)
 911 {
 912     QEMUIOVector qiov;
 913     struct iovec iov;
 914     BlkRwCo rwco;
 915
 916     iov = (struct iovec) {
 917         .iov_base = buf,
 918         .iov_len = bytes,
 919     };
 920     qemu_iovec_init_external(&qiov, &iov, 1);
 921
 922     rwco = (BlkRwCo) {
 923         .blk    = blk,
 924         .offset = offset,
 925         .qiov   = &qiov,
 926         .flags  = flags,
 927         .ret    = NOT_DONE,
 928     };
 929
 930     if (qemu_in_coroutine()) {
 931         /* Fast-path if already in coroutine context */
 932         co_entry(&rwco);
 933     } else {
 934         Coroutine *co = qemu_coroutine_create(co_entry, &rwco);
 935         qemu_coroutine_enter(co);
 936         BDRV_POLL_WHILE(blk_bs(blk), rwco.ret == NOT_DONE);
 937     }
 938
 939     return rwco.ret;
 940 }
 941
 942 int blk_pread_unthrottled(BlockBackend *blk, int64_t offset, uint8_t *buf,
 943                           int count)
 944 {
 945     int ret;
 946
 947     ret = blk_check_byte_request(blk, offset, count);
 948     if (ret < 0) {
 949         return ret;
 950     }
 951
 952     blk_root_drained_begin(blk->root);
 953     ret = blk_pread(blk, offset, buf, count);
 954     blk_root_drained_end(blk->root);
 955     return ret;
 956 }
 957
 958 int blk_pwrite_zeroes(BlockBackend *blk, int64_t offset,
 959                       int count, BdrvRequestFlags flags)
 960 {
 961     return blk_prw(blk, offset, NULL, count, blk_write_entry,
 962                    flags | BDRV_REQ_ZERO_WRITE);
 963 }
 964
 965 int blk_make_zero(BlockBackend *blk, BdrvRequestFlags flags)
 966 {
 967     return bdrv_make_zero(blk->root, flags);
 968 }
 969
 970 static void error_callback_bh(void *opaque)
 971 {
 972     struct BlockBackendAIOCB *acb = opaque;
 973
 974     bdrv_dec_in_flight(acb->common.bs);
 975     acb->common.cb(acb->common.opaque, acb->ret);
 976     qemu_aio_unref(acb);
 977 }
 978
 979 BlockAIOCB *blk_abort_aio_request(BlockBackend *blk,
 980                                   BlockCompletionFunc *cb,
 981                                   void *opaque, int ret)
 982 {
 983     struct BlockBackendAIOCB *acb;
 984
 985     bdrv_inc_in_flight(blk_bs(blk));
 986     acb = blk_aio_get(&block_backend_aiocb_info, blk, cb, opaque);
 987     acb->blk = blk;
 988     acb->ret = ret;
 989
 990     aio_bh_schedule_oneshot(blk_get_aio_context(blk), error_callback_bh, acb);
 991     return &acb->common;
 992 }
 993
 994 typedef struct BlkAioEmAIOCB {
 995     BlockAIOCB common;
 996     BlkRwCo rwco;
 997     int bytes;
 998     bool has_returned;
 999 } BlkAioEmAIOCB;
1000
1001 static const AIOCBInfo blk_aio_em_aiocb_info = {
1002     .aiocb_size         = sizeof(BlkAioEmAIOCB),
1003 };
1004
1005 static void blk_aio_complete(BlkAioEmAIOCB *acb)
1006 {
1007     if (acb->has_returned) {
1008         bdrv_dec_in_flight(acb->common.bs);
1009         acb->common.cb(acb->common.opaque, acb->rwco.ret);
1010         qemu_aio_unref(acb);
1011     }
1012 }
1013
1014 static void blk_aio_complete_bh(void *opaque)
1015 {
1016     BlkAioEmAIOCB *acb = opaque;
1017     assert(acb->has_returned);
1018     blk_aio_complete(acb);
1019 }
1020
1021 static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset, int bytes,
1022                                 QEMUIOVector *qiov, CoroutineEntry co_entry,
1023                                 BdrvRequestFlags flags,
1024                                 BlockCompletionFunc *cb, void *opaque)
1025 {
1026     BlkAioEmAIOCB *acb;
1027     Coroutine *co;
1028
1029     bdrv_inc_in_flight(blk_bs(blk));
1030     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1031     acb->rwco = (BlkRwCo) {
1032         .blk    = blk,
1033         .offset = offset,
1034         .qiov   = qiov,
1035         .flags  = flags,
1036         .ret    = NOT_DONE,
1037     };
1038     acb->bytes = bytes;
1039     acb->has_returned = false;
1040
1041     co = qemu_coroutine_create(co_entry, acb);
1042     qemu_coroutine_enter(co);
1043
1044     acb->has_returned = true;
1045     if (acb->rwco.ret != NOT_DONE) {
1046         aio_bh_schedule_oneshot(blk_get_aio_context(blk),
1047                                 blk_aio_complete_bh, acb);
1048     }
1049
1050     return &acb->common;
1051 }
1052
1053 static void blk_aio_read_entry(void *opaque)
1054 {
1055     BlkAioEmAIOCB *acb = opaque;
1056     BlkRwCo *rwco = &acb->rwco;
1057
1058     assert(rwco->qiov->size == acb->bytes);
1059     rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, acb->bytes,
1060                               rwco->qiov, rwco->flags);
1061     blk_aio_complete(acb);
1062 }
1063
1064 static void blk_aio_write_entry(void *opaque)
1065 {
1066     BlkAioEmAIOCB *acb = opaque;
1067     BlkRwCo *rwco = &acb->rwco;
1068
1069     assert(!rwco->qiov || rwco->qiov->size == acb->bytes);
1070     rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, acb->bytes,
1071                                rwco->qiov, rwco->flags);
1072     blk_aio_complete(acb);
1073 }
1074
1075 BlockAIOCB *blk_aio_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1076                                   int count, BdrvRequestFlags flags,
1077                                   BlockCompletionFunc *cb, void *opaque)
1078 {
1079     return blk_aio_prwv(blk, offset, count, NULL, blk_aio_write_entry,
1080                         flags | BDRV_REQ_ZERO_WRITE, cb, opaque);
1081 }
1082
1083 int blk_pread(BlockBackend *blk, int64_t offset, void *buf, int count)
1084 {
1085     int ret = blk_prw(blk, offset, buf, count, blk_read_entry, 0);
1086     if (ret < 0) {
1087         return ret;
1088     }
1089     return count;
1090 }
1091
1092 int blk_pwrite(BlockBackend *blk, int64_t offset, const void *buf, int count,
1093                BdrvRequestFlags flags)
1094 {
1095     int ret = blk_prw(blk, offset, (void *) buf, count, blk_write_entry,
1096                       flags);
1097     if (ret < 0) {
1098         return ret;
1099     }
1100     return count;
1101 }
1102
1103 int64_t blk_getlength(BlockBackend *blk)
1104 {
1105     if (!blk_is_available(blk)) {
1106         return -ENOMEDIUM;
1107     }
1108
1109     return bdrv_getlength(blk_bs(blk));
1110 }
1111
1112 void blk_get_geometry(BlockBackend *blk, uint64_t *nb_sectors_ptr)
1113 {
1114     if (!blk_bs(blk)) {
1115         *nb_sectors_ptr = 0;
1116     } else {
1117         bdrv_get_geometry(blk_bs(blk), nb_sectors_ptr);
1118     }
1119 }
1120
1121 int64_t blk_nb_sectors(BlockBackend *blk)
1122 {
1123     if (!blk_is_available(blk)) {
1124         return -ENOMEDIUM;
1125     }
1126
1127     return bdrv_nb_sectors(blk_bs(blk));
1128 }
1129
1130 BlockAIOCB *blk_aio_preadv(BlockBackend *blk, int64_t offset,
1131                            QEMUIOVector *qiov, BdrvRequestFlags flags,
1132                            BlockCompletionFunc *cb, void *opaque)
1133 {
1134     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1135                         blk_aio_read_entry, flags, cb, opaque);
1136 }
1137
1138 BlockAIOCB *blk_aio_pwritev(BlockBackend *blk, int64_t offset,
1139                             QEMUIOVector *qiov, BdrvRequestFlags flags,
1140                             BlockCompletionFunc *cb, void *opaque)
1141 {
1142     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1143                         blk_aio_write_entry, flags, cb, opaque);
1144 }
1145
1146 static void blk_aio_flush_entry(void *opaque)
1147 {
1148     BlkAioEmAIOCB *acb = opaque;
1149     BlkRwCo *rwco = &acb->rwco;
1150
1151     rwco->ret = blk_co_flush(rwco->blk);
1152     blk_aio_complete(acb);
1153 }
1154
1155 BlockAIOCB *blk_aio_flush(BlockBackend *blk,
1156                           BlockCompletionFunc *cb, void *opaque)
1157 {
1158     return blk_aio_prwv(blk, 0, 0, NULL, blk_aio_flush_entry, 0, cb, opaque);
1159 }
1160
1161 static void blk_aio_pdiscard_entry(void *opaque)
1162 {
1163     BlkAioEmAIOCB *acb = opaque;
1164     BlkRwCo *rwco = &acb->rwco;
1165
1166     rwco->ret = blk_co_pdiscard(rwco->blk, rwco->offset, acb->bytes);
1167     blk_aio_complete(acb);
1168 }
1169
1170 BlockAIOCB *blk_aio_pdiscard(BlockBackend *blk,
1171                              int64_t offset, int count,
1172                              BlockCompletionFunc *cb, void *opaque)
1173 {
1174     return blk_aio_prwv(blk, offset, count, NULL, blk_aio_pdiscard_entry, 0,
1175                         cb, opaque);
1176 }
1177
1178 void blk_aio_cancel(BlockAIOCB *acb)
1179 {
1180     bdrv_aio_cancel(acb);
1181 }
1182
1183 void blk_aio_cancel_async(BlockAIOCB *acb)
1184 {
1185     bdrv_aio_cancel_async(acb);
1186 }
1187
1188 int blk_co_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1189 {
1190     if (!blk_is_available(blk)) {
1191         return -ENOMEDIUM;
1192     }
1193
1194     return bdrv_co_ioctl(blk_bs(blk), req, buf);
1195 }
1196
1197 static void blk_ioctl_entry(void *opaque)
1198 {
1199     BlkRwCo *rwco = opaque;
1200     rwco->ret = blk_co_ioctl(rwco->blk, rwco->offset,
1201                              rwco->qiov->iov[0].iov_base);
1202 }
1203
1204 int blk_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1205 {
1206     return blk_prw(blk, req, buf, 0, blk_ioctl_entry, 0);
1207 }
1208
1209 static void blk_aio_ioctl_entry(void *opaque)
1210 {
1211     BlkAioEmAIOCB *acb = opaque;
1212     BlkRwCo *rwco = &acb->rwco;
1213
1214     rwco->ret = blk_co_ioctl(rwco->blk, rwco->offset,
1215                              rwco->qiov->iov[0].iov_base);
1216     blk_aio_complete(acb);
1217 }
1218
1219 BlockAIOCB *blk_aio_ioctl(BlockBackend *blk, unsigned long int req, void *buf,
1220                           BlockCompletionFunc *cb, void *opaque)
1221 {
1222     QEMUIOVector qiov;
1223     struct iovec iov;
1224
1225     iov = (struct iovec) {
1226         .iov_base = buf,
1227         .iov_len = 0,
1228     };
1229     qemu_iovec_init_external(&qiov, &iov, 1);
1230
1231     return blk_aio_prwv(blk, req, 0, &qiov, blk_aio_ioctl_entry, 0, cb, opaque);
1232 }
1233
1234 int blk_co_pdiscard(BlockBackend *blk, int64_t offset, int count)
1235 {
1236     int ret = blk_check_byte_request(blk, offset, count);
1237     if (ret < 0) {
1238         return ret;
1239     }
1240
1241     return bdrv_co_pdiscard(blk_bs(blk), offset, count);
1242 }
1243
1244 int blk_co_flush(BlockBackend *blk)
1245 {
1246     if (!blk_is_available(blk)) {
1247         return -ENOMEDIUM;
1248     }
1249
1250     return bdrv_co_flush(blk_bs(blk));
1251 }
1252
1253 static void blk_flush_entry(void *opaque)
1254 {
1255     BlkRwCo *rwco = opaque;
1256     rwco->ret = blk_co_flush(rwco->blk);
1257 }
1258
1259 int blk_flush(BlockBackend *blk)
1260 {
1261     return blk_prw(blk, 0, NULL, 0, blk_flush_entry, 0);
1262 }
1263
1264 void blk_drain(BlockBackend *blk)
1265 {
1266     if (blk_bs(blk)) {
1267         bdrv_drain(blk_bs(blk));
1268     }
1269 }
1270
1271 void blk_drain_all(void)
1272 {
1273     bdrv_drain_all();
1274 }
1275
1276 void blk_set_on_error(BlockBackend *blk, BlockdevOnError on_read_error,
1277                       BlockdevOnError on_write_error)
1278 {
1279     blk->on_read_error = on_read_error;
1280     blk->on_write_error = on_write_error;
1281 }
1282
1283 BlockdevOnError blk_get_on_error(BlockBackend *blk, bool is_read)
1284 {
1285     return is_read ? blk->on_read_error : blk->on_write_error;
1286 }
1287
1288 BlockErrorAction blk_get_error_action(BlockBackend *blk, bool is_read,
1289                                       int error)
1290 {
1291     BlockdevOnError on_err = blk_get_on_error(blk, is_read);
1292
1293     switch (on_err) {
1294     case BLOCKDEV_ON_ERROR_ENOSPC:
1295         return (error == ENOSPC) ?
1296                BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
1297     case BLOCKDEV_ON_ERROR_STOP:
1298         return BLOCK_ERROR_ACTION_STOP;
1299     case BLOCKDEV_ON_ERROR_REPORT:
1300         return BLOCK_ERROR_ACTION_REPORT;
1301     case BLOCKDEV_ON_ERROR_IGNORE:
1302         return BLOCK_ERROR_ACTION_IGNORE;
1303     case BLOCKDEV_ON_ERROR_AUTO:
1304     default:
1305         abort();
1306     }
1307 }
1308
1309 static void send_qmp_error_event(BlockBackend *blk,
1310                                  BlockErrorAction action,
1311                                  bool is_read, int error)
1312 {
1313     IoOperationType optype;
1314
1315     optype = is_read ? IO_OPERATION_TYPE_READ : IO_OPERATION_TYPE_WRITE;
1316     qapi_event_send_block_io_error(blk_name(blk),
1317                                    bdrv_get_node_name(blk_bs(blk)), optype,
1318                                    action, blk_iostatus_is_enabled(blk),
1319                                    error == ENOSPC, strerror(error),
1320                                    &error_abort);
1321 }
1322
1323 /* This is done by device models because, while the block layer knows
1324  * about the error, it does not know whether an operation comes from
1325  * the device or the block layer (from a job, for example).
1326  */
1327 void blk_error_action(BlockBackend *blk, BlockErrorAction action,
1328                       bool is_read, int error)
1329 {
1330     assert(error >= 0);
1331
1332     if (action == BLOCK_ERROR_ACTION_STOP) {
1333         /* First set the iostatus, so that "info block" returns an iostatus
1334          * that matches the events raised so far (an additional error iostatus
1335          * is fine, but not a lost one).
1336          */
1337         blk_iostatus_set_err(blk, error);
1338
1339         /* Then raise the request to stop the VM and the event.
1340          * qemu_system_vmstop_request_prepare has two effects.  First,
1341          * it ensures that the STOP event always comes after the
1342          * BLOCK_IO_ERROR event.  Second, it ensures that even if management
1343          * can observe the STOP event and do a "cont" before the STOP
1344          * event is issued, the VM will not stop.  In this case, vm_start()
1345          * also ensures that the STOP/RESUME pair of events is emitted.
1346          */
1347         qemu_system_vmstop_request_prepare();
1348         send_qmp_error_event(blk, action, is_read, error);
1349         qemu_system_vmstop_request(RUN_STATE_IO_ERROR);
1350     } else {
1351         send_qmp_error_event(blk, action, is_read, error);
1352     }
1353 }
1354
1355 int blk_is_read_only(BlockBackend *blk)
1356 {
1357     BlockDriverState *bs = blk_bs(blk);
1358
1359     if (bs) {
1360         return bdrv_is_read_only(bs);
1361     } else {
1362         return blk->root_state.read_only;
1363     }
1364 }
1365
1366 int blk_is_sg(BlockBackend *blk)
1367 {
1368     BlockDriverState *bs = blk_bs(blk);
1369
1370     if (!bs) {
1371         return 0;
1372     }
1373
1374     return bdrv_is_sg(bs);
1375 }
1376
1377 int blk_enable_write_cache(BlockBackend *blk)
1378 {
1379     return blk->enable_write_cache;
1380 }
1381
1382 void blk_set_enable_write_cache(BlockBackend *blk, bool wce)
1383 {
1384     blk->enable_write_cache = wce;
1385 }
1386
1387 void blk_invalidate_cache(BlockBackend *blk, Error **errp)
1388 {
1389     BlockDriverState *bs = blk_bs(blk);
1390
1391     if (!bs) {
1392         error_setg(errp, "Device '%s' has no medium", blk->name);
1393         return;
1394     }
1395
1396     bdrv_invalidate_cache(bs, errp);
1397 }
1398
1399 bool blk_is_inserted(BlockBackend *blk)
1400 {
1401     BlockDriverState *bs = blk_bs(blk);
1402
1403     return bs && bdrv_is_inserted(bs);
1404 }
1405
1406 bool blk_is_available(BlockBackend *blk)
1407 {
1408     return blk_is_inserted(blk) && !blk_dev_is_tray_open(blk);
1409 }
1410
1411 void blk_lock_medium(BlockBackend *blk, bool locked)
1412 {
1413     BlockDriverState *bs = blk_bs(blk);
1414
1415     if (bs) {
1416         bdrv_lock_medium(bs, locked);
1417     }
1418 }
1419
1420 void blk_eject(BlockBackend *blk, bool eject_flag)
1421 {
1422     BlockDriverState *bs = blk_bs(blk);
1423     char *id;
1424
1425     /* blk_eject is only called by qdevified devices */
1426     assert(!blk->legacy_dev);
1427
1428     if (bs) {
1429         bdrv_eject(bs, eject_flag);
1430     }
1431
1432     /* Whether or not we ejected on the backend,
1433      * the frontend experienced a tray event. */
1434     id = blk_get_attached_dev_id(blk);
1435     qapi_event_send_device_tray_moved(blk_name(blk), id,
1436                                       eject_flag, &error_abort);
1437     g_free(id);
1438 }
1439
1440 int blk_get_flags(BlockBackend *blk)
1441 {
1442     BlockDriverState *bs = blk_bs(blk);
1443
1444     if (bs) {
1445         return bdrv_get_flags(bs);
1446     } else {
1447         return blk->root_state.open_flags;
1448     }
1449 }
1450
1451 /* Returns the maximum transfer length, in bytes; guaranteed nonzero */
1452 uint32_t blk_get_max_transfer(BlockBackend *blk)
1453 {
1454     BlockDriverState *bs = blk_bs(blk);
1455     uint32_t max = 0;
1456
1457     if (bs) {
1458         max = bs->bl.max_transfer;
1459     }
1460     return MIN_NON_ZERO(max, INT_MAX);
1461 }
1462
1463 int blk_get_max_iov(BlockBackend *blk)
1464 {
1465     return blk->root->bs->bl.max_iov;
1466 }
1467
1468 void blk_set_guest_block_size(BlockBackend *blk, int align)
1469 {
1470     blk->guest_block_size = align;
1471 }
1472
1473 void *blk_try_blockalign(BlockBackend *blk, size_t size)
1474 {
1475     return qemu_try_blockalign(blk ? blk_bs(blk) : NULL, size);
1476 }
1477
1478 void *blk_blockalign(BlockBackend *blk, size_t size)
1479 {
1480     return qemu_blockalign(blk ? blk_bs(blk) : NULL, size);
1481 }
1482
1483 bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp)
1484 {
1485     BlockDriverState *bs = blk_bs(blk);
1486
1487     if (!bs) {
1488         return false;
1489     }
1490
1491     return bdrv_op_is_blocked(bs, op, errp);
1492 }
1493
1494 void blk_op_unblock(BlockBackend *blk, BlockOpType op, Error *reason)
1495 {
1496     BlockDriverState *bs = blk_bs(blk);
1497
1498     if (bs) {
1499         bdrv_op_unblock(bs, op, reason);
1500     }
1501 }
1502
1503 void blk_op_block_all(BlockBackend *blk, Error *reason)
1504 {
1505     BlockDriverState *bs = blk_bs(blk);
1506
1507     if (bs) {
1508         bdrv_op_block_all(bs, reason);
1509     }
1510 }
1511
1512 void blk_op_unblock_all(BlockBackend *blk, Error *reason)
1513 {
1514     BlockDriverState *bs = blk_bs(blk);
1515
1516     if (bs) {
1517         bdrv_op_unblock_all(bs, reason);
1518     }
1519 }
1520
1521 AioContext *blk_get_aio_context(BlockBackend *blk)
1522 {
1523     BlockDriverState *bs = blk_bs(blk);
1524
1525     if (bs) {
1526         return bdrv_get_aio_context(bs);
1527     } else {
1528         return qemu_get_aio_context();
1529     }
1530 }
1531
1532 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb)
1533 {
1534     BlockBackendAIOCB *blk_acb = DO_UPCAST(BlockBackendAIOCB, common, acb);
1535     return blk_get_aio_context(blk_acb->blk);
1536 }
1537
1538 void blk_set_aio_context(BlockBackend *blk, AioContext *new_context)
1539 {
1540     BlockDriverState *bs = blk_bs(blk);
1541
1542     if (bs) {
1543         if (blk->public.throttle_state) {
1544             throttle_timers_detach_aio_context(&blk->public.throttle_timers);
1545         }
1546         bdrv_set_aio_context(bs, new_context);
1547         if (blk->public.throttle_state) {
1548             throttle_timers_attach_aio_context(&blk->public.throttle_timers,
1549                                                new_context);
1550         }
1551     }
1552 }
1553
1554 void blk_add_aio_context_notifier(BlockBackend *blk,
1555         void (*attached_aio_context)(AioContext *new_context, void *opaque),
1556         void (*detach_aio_context)(void *opaque), void *opaque)
1557 {
1558     BlockDriverState *bs = blk_bs(blk);
1559
1560     if (bs) {
1561         bdrv_add_aio_context_notifier(bs, attached_aio_context,
1562                                       detach_aio_context, opaque);
1563     }
1564 }
1565
1566 void blk_remove_aio_context_notifier(BlockBackend *blk,
1567                                      void (*attached_aio_context)(AioContext *,
1568                                                                   void *),
1569                                      void (*detach_aio_context)(void *),
1570                                      void *opaque)
1571 {
1572     BlockDriverState *bs = blk_bs(blk);
1573
1574     if (bs) {
1575         bdrv_remove_aio_context_notifier(bs, attached_aio_context,
1576                                          detach_aio_context, opaque);
1577     }
1578 }
1579
1580 void blk_add_remove_bs_notifier(BlockBackend *blk, Notifier *notify)
1581 {
1582     notifier_list_add(&blk->remove_bs_notifiers, notify);
1583 }
1584
1585 void blk_add_insert_bs_notifier(BlockBackend *blk, Notifier *notify)
1586 {
1587     notifier_list_add(&blk->insert_bs_notifiers, notify);
1588 }
1589
1590 void blk_io_plug(BlockBackend *blk)
1591 {
1592     BlockDriverState *bs = blk_bs(blk);
1593
1594     if (bs) {
1595         bdrv_io_plug(bs);
1596     }
1597 }
1598
1599 void blk_io_unplug(BlockBackend *blk)
1600 {
1601     BlockDriverState *bs = blk_bs(blk);
1602
1603     if (bs) {
1604         bdrv_io_unplug(bs);
1605     }
1606 }
1607
1608 BlockAcctStats *blk_get_stats(BlockBackend *blk)
1609 {
1610     return &blk->stats;
1611 }
1612
1613 void *blk_aio_get(const AIOCBInfo *aiocb_info, BlockBackend *blk,
1614                   BlockCompletionFunc *cb, void *opaque)
1615 {
1616     return qemu_aio_get(aiocb_info, blk_bs(blk), cb, opaque);
1617 }
1618
1619 int coroutine_fn blk_co_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1620                                       int count, BdrvRequestFlags flags)
1621 {
1622     return blk_co_pwritev(blk, offset, count, NULL,
1623                           flags | BDRV_REQ_ZERO_WRITE);
1624 }
1625
1626 int blk_pwrite_compressed(BlockBackend *blk, int64_t offset, const void *buf,
1627                           int count)
1628 {
1629     return blk_prw(blk, offset, (void *) buf, count, blk_write_entry,
1630                    BDRV_REQ_WRITE_COMPRESSED);
1631 }
1632
1633 int blk_truncate(BlockBackend *blk, int64_t offset)
1634 {
1635     if (!blk_is_available(blk)) {
1636         return -ENOMEDIUM;
1637     }
1638
1639     return bdrv_truncate(blk->root, offset);
1640 }
1641
1642 static void blk_pdiscard_entry(void *opaque)
1643 {
1644     BlkRwCo *rwco = opaque;
1645     rwco->ret = blk_co_pdiscard(rwco->blk, rwco->offset, rwco->qiov->size);
1646 }
1647
1648 int blk_pdiscard(BlockBackend *blk, int64_t offset, int count)
1649 {
1650     return blk_prw(blk, offset, NULL, count, blk_pdiscard_entry, 0);
1651 }
1652
1653 int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
1654                      int64_t pos, int size)
1655 {
1656     int ret;
1657
1658     if (!blk_is_available(blk)) {
1659         return -ENOMEDIUM;
1660     }
1661
1662     ret = bdrv_save_vmstate(blk_bs(blk), buf, pos, size);
1663     if (ret < 0) {
1664         return ret;
1665     }
1666
1667     if (ret == size && !blk->enable_write_cache) {
1668         ret = bdrv_flush(blk_bs(blk));
1669     }
1670
1671     return ret < 0 ? ret : size;
1672 }
1673
1674 int blk_load_vmstate(BlockBackend *blk, uint8_t *buf, int64_t pos, int size)
1675 {
1676     if (!blk_is_available(blk)) {
1677         return -ENOMEDIUM;
1678     }
1679
1680     return bdrv_load_vmstate(blk_bs(blk), buf, pos, size);
1681 }
1682
1683 int blk_probe_blocksizes(BlockBackend *blk, BlockSizes *bsz)
1684 {
1685     if (!blk_is_available(blk)) {
1686         return -ENOMEDIUM;
1687     }
1688
1689     return bdrv_probe_blocksizes(blk_bs(blk), bsz);
1690 }
1691
1692 int blk_probe_geometry(BlockBackend *blk, HDGeometry *geo)
1693 {
1694     if (!blk_is_available(blk)) {
1695         return -ENOMEDIUM;
1696     }
1697
1698     return bdrv_probe_geometry(blk_bs(blk), geo);
1699 }
1700
1701 /*
1702  * Updates the BlockBackendRootState object with data from the currently
1703  * attached BlockDriverState.
1704  */
1705 void blk_update_root_state(BlockBackend *blk)
1706 {
1707     assert(blk->root);
1708
1709     blk->root_state.open_flags    = blk->root->bs->open_flags;
1710     blk->root_state.read_only     = blk->root->bs->read_only;
1711     blk->root_state.detect_zeroes = blk->root->bs->detect_zeroes;
1712 }
1713
1714 /*
1715  * Returns the detect-zeroes setting to be used for bdrv_open() of a
1716  * BlockDriverState which is supposed to inherit the root state.
1717  */
1718 bool blk_get_detect_zeroes_from_root_state(BlockBackend *blk)
1719 {
1720     return blk->root_state.detect_zeroes;
1721 }
1722
1723 /*
1724  * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
1725  * supposed to inherit the root state.
1726  */
1727 int blk_get_open_flags_from_root_state(BlockBackend *blk)
1728 {
1729     int bs_flags;
1730
1731     bs_flags = blk->root_state.read_only ? 0 : BDRV_O_RDWR;
1732     bs_flags |= blk->root_state.open_flags & ~BDRV_O_RDWR;
1733
1734     return bs_flags;
1735 }
1736
1737 BlockBackendRootState *blk_get_root_state(BlockBackend *blk)
1738 {
1739     return &blk->root_state;
1740 }
1741
1742 int blk_commit_all(void)
1743 {
1744     BlockBackend *blk = NULL;
1745
1746     while ((blk = blk_all_next(blk)) != NULL) {
1747         AioContext *aio_context = blk_get_aio_context(blk);
1748
1749         aio_context_acquire(aio_context);
1750         if (blk_is_inserted(blk) && blk->root->bs->backing) {
1751             int ret = bdrv_commit(blk->root->bs);
1752             if (ret < 0) {
1753                 aio_context_release(aio_context);
1754                 return ret;
1755             }
1756         }
1757         aio_context_release(aio_context);
1758     }
1759     return 0;
1760 }
1761
1762
1763 /* throttling disk I/O limits */
1764 void blk_set_io_limits(BlockBackend *blk, ThrottleConfig *cfg)
1765 {
1766     throttle_group_config(blk, cfg);
1767 }
1768
1769 void blk_io_limits_disable(BlockBackend *blk)
1770 {
1771     assert(blk->public.throttle_state);
1772     bdrv_drained_begin(blk_bs(blk));
1773     throttle_group_unregister_blk(blk);
1774     bdrv_drained_end(blk_bs(blk));
1775 }
1776
1777 /* should be called before blk_set_io_limits if a limit is set */
1778 void blk_io_limits_enable(BlockBackend *blk, const char *group)
1779 {
1780     assert(!blk->public.throttle_state);
1781     throttle_group_register_blk(blk, group);
1782 }
1783
1784 void blk_io_limits_update_group(BlockBackend *blk, const char *group)
1785 {
1786     /* this BB is not part of any group */
1787     if (!blk->public.throttle_state) {
1788         return;
1789     }
1790
1791     /* this BB is a part of the same group than the one we want */
1792     if (!g_strcmp0(throttle_group_get_name(blk), group)) {
1793         return;
1794     }
1795
1796     /* need to change the group this bs belong to */
1797     blk_io_limits_disable(blk);
1798     blk_io_limits_enable(blk, group);
1799 }
1800
1801 static void blk_root_drained_begin(BdrvChild *child)
1802 {
1803     BlockBackend *blk = child->opaque;
1804
1805     /* Note that blk->root may not be accessible here yet if we are just
1806      * attaching to a BlockDriverState that is drained. Use child instead. */
1807
1808     if (blk->public.io_limits_disabled++ == 0) {
1809         throttle_group_restart_blk(blk);
1810     }
1811 }
1812
1813 static void blk_root_drained_end(BdrvChild *child)
1814 {
1815     BlockBackend *blk = child->opaque;
1816
1817     assert(blk->public.io_limits_disabled);
1818     --blk->public.io_limits_disabled;
1819 }