4 * Copyright (C) 2014-2016 Red Hat, Inc.
7 * Markus Armbruster <armbru@redhat.com>,
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.
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 "hw/qdev-core.h"
19 #include "sysemu/blockdev.h"
20 #include "sysemu/runstate.h"
21 #include "sysemu/sysemu.h"
22 #include "sysemu/replay.h"
23 #include "qapi/error.h"
24 #include "qapi/qapi-events-block.h"
26 #include "qemu/main-loop.h"
27 #include "qemu/option.h"
29 #include "migration/misc.h"
31 /* Number of coroutines to reserve per attached device model */
32 #define COROUTINE_POOL_RESERVATION 64
34 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
36 static AioContext
*blk_aiocb_get_aio_context(BlockAIOCB
*acb
);
38 typedef struct BlockBackendAioNotifier
{
39 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
);
40 void (*detach_aio_context
)(void *opaque
);
42 QLIST_ENTRY(BlockBackendAioNotifier
) list
;
43 } BlockBackendAioNotifier
;
50 DriveInfo
*legacy_dinfo
; /* null unless created by drive_new() */
51 QTAILQ_ENTRY(BlockBackend
) link
; /* for block_backends */
52 QTAILQ_ENTRY(BlockBackend
) monitor_link
; /* for monitor_block_backends */
53 BlockBackendPublic
public;
55 DeviceState
*dev
; /* attached device model, if any */
56 const BlockDevOps
*dev_ops
;
59 /* the block size for which the guest device expects atomicity */
62 /* If the BDS tree is removed, some of its options are stored here (which
63 * can be used to restore those options in the new BDS on insert) */
64 BlockBackendRootState root_state
;
66 bool enable_write_cache
;
68 /* I/O stats (display with "info blockstats"). */
71 BlockdevOnError on_read_error
, on_write_error
;
72 bool iostatus_enabled
;
73 BlockDeviceIoStatus iostatus
;
79 bool allow_aio_context_change
;
80 bool allow_write_beyond_eof
;
82 NotifierList remove_bs_notifiers
, insert_bs_notifiers
;
83 QLIST_HEAD(, BlockBackendAioNotifier
) aio_notifiers
;
86 CoQueue queued_requests
;
87 bool disable_request_queuing
;
89 VMChangeStateEntry
*vmsh
;
90 bool force_allow_inactivate
;
92 /* Number of in-flight aio requests. BlockDriverState also counts
93 * in-flight requests but aio requests can exist even when blk->root is
94 * NULL, so we cannot rely on its counter for that case.
95 * Accessed with atomic ops.
97 unsigned int in_flight
;
100 typedef struct BlockBackendAIOCB
{
106 static const AIOCBInfo block_backend_aiocb_info
= {
107 .get_aio_context
= blk_aiocb_get_aio_context
,
108 .aiocb_size
= sizeof(BlockBackendAIOCB
),
111 static void drive_info_del(DriveInfo
*dinfo
);
112 static BlockBackend
*bdrv_first_blk(BlockDriverState
*bs
);
114 /* All BlockBackends */
115 static QTAILQ_HEAD(, BlockBackend
) block_backends
=
116 QTAILQ_HEAD_INITIALIZER(block_backends
);
118 /* All BlockBackends referenced by the monitor and which are iterated through by
120 static QTAILQ_HEAD(, BlockBackend
) monitor_block_backends
=
121 QTAILQ_HEAD_INITIALIZER(monitor_block_backends
);
123 static void blk_root_inherit_options(int *child_flags
, QDict
*child_options
,
124 int parent_flags
, QDict
*parent_options
)
126 /* We're not supposed to call this function for root nodes */
129 static void blk_root_drained_begin(BdrvChild
*child
);
130 static bool blk_root_drained_poll(BdrvChild
*child
);
131 static void blk_root_drained_end(BdrvChild
*child
, int *drained_end_counter
);
133 static void blk_root_change_media(BdrvChild
*child
, bool load
);
134 static void blk_root_resize(BdrvChild
*child
);
136 static bool blk_root_can_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
137 GSList
**ignore
, Error
**errp
);
138 static void blk_root_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
141 static char *blk_root_get_parent_desc(BdrvChild
*child
)
143 BlockBackend
*blk
= child
->opaque
;
147 return g_strdup(blk
->name
);
150 dev_id
= blk_get_attached_dev_id(blk
);
154 /* TODO Callback into the BB owner for something more detailed */
156 return g_strdup("a block device");
160 static const char *blk_root_get_name(BdrvChild
*child
)
162 return blk_name(child
->opaque
);
165 static void blk_vm_state_changed(void *opaque
, int running
, RunState state
)
167 Error
*local_err
= NULL
;
168 BlockBackend
*blk
= opaque
;
170 if (state
== RUN_STATE_INMIGRATE
) {
174 qemu_del_vm_change_state_handler(blk
->vmsh
);
176 blk_set_perm(blk
, blk
->perm
, blk
->shared_perm
, &local_err
);
178 error_report_err(local_err
);
183 * Notifies the user of the BlockBackend that migration has completed. qdev
184 * devices can tighten their permissions in response (specifically revoke
185 * shared write permissions that we needed for storage migration).
187 * If an error is returned, the VM cannot be allowed to be resumed.
189 static void blk_root_activate(BdrvChild
*child
, Error
**errp
)
191 BlockBackend
*blk
= child
->opaque
;
192 Error
*local_err
= NULL
;
194 if (!blk
->disable_perm
) {
198 blk
->disable_perm
= false;
200 blk_set_perm(blk
, blk
->perm
, BLK_PERM_ALL
, &local_err
);
202 error_propagate(errp
, local_err
);
203 blk
->disable_perm
= true;
207 if (runstate_check(RUN_STATE_INMIGRATE
)) {
208 /* Activation can happen when migration process is still active, for
209 * example when nbd_server_add is called during non-shared storage
210 * migration. Defer the shared_perm update to migration completion. */
212 blk
->vmsh
= qemu_add_vm_change_state_handler(blk_vm_state_changed
,
218 blk_set_perm(blk
, blk
->perm
, blk
->shared_perm
, &local_err
);
220 error_propagate(errp
, local_err
);
221 blk
->disable_perm
= true;
226 void blk_set_force_allow_inactivate(BlockBackend
*blk
)
228 blk
->force_allow_inactivate
= true;
231 static bool blk_can_inactivate(BlockBackend
*blk
)
233 /* If it is a guest device, inactivate is ok. */
234 if (blk
->dev
|| blk_name(blk
)[0]) {
238 /* Inactivating means no more writes to the image can be done,
239 * even if those writes would be changes invisible to the
240 * guest. For block job BBs that satisfy this, we can just allow
241 * it. This is the case for mirror job source, which is required
242 * by libvirt non-shared block migration. */
243 if (!(blk
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
))) {
247 return blk
->force_allow_inactivate
;
250 static int blk_root_inactivate(BdrvChild
*child
)
252 BlockBackend
*blk
= child
->opaque
;
254 if (blk
->disable_perm
) {
258 if (!blk_can_inactivate(blk
)) {
262 blk
->disable_perm
= true;
264 bdrv_child_try_set_perm(blk
->root
, 0, BLK_PERM_ALL
, &error_abort
);
270 static void blk_root_attach(BdrvChild
*child
)
272 BlockBackend
*blk
= child
->opaque
;
273 BlockBackendAioNotifier
*notifier
;
275 trace_blk_root_attach(child
, blk
, child
->bs
);
277 QLIST_FOREACH(notifier
, &blk
->aio_notifiers
, list
) {
278 bdrv_add_aio_context_notifier(child
->bs
,
279 notifier
->attached_aio_context
,
280 notifier
->detach_aio_context
,
285 static void blk_root_detach(BdrvChild
*child
)
287 BlockBackend
*blk
= child
->opaque
;
288 BlockBackendAioNotifier
*notifier
;
290 trace_blk_root_detach(child
, blk
, child
->bs
);
292 QLIST_FOREACH(notifier
, &blk
->aio_notifiers
, list
) {
293 bdrv_remove_aio_context_notifier(child
->bs
,
294 notifier
->attached_aio_context
,
295 notifier
->detach_aio_context
,
300 static const BdrvChildRole child_root
= {
301 .inherit_options
= blk_root_inherit_options
,
303 .change_media
= blk_root_change_media
,
304 .resize
= blk_root_resize
,
305 .get_name
= blk_root_get_name
,
306 .get_parent_desc
= blk_root_get_parent_desc
,
308 .drained_begin
= blk_root_drained_begin
,
309 .drained_poll
= blk_root_drained_poll
,
310 .drained_end
= blk_root_drained_end
,
312 .activate
= blk_root_activate
,
313 .inactivate
= blk_root_inactivate
,
315 .attach
= blk_root_attach
,
316 .detach
= blk_root_detach
,
318 .can_set_aio_ctx
= blk_root_can_set_aio_ctx
,
319 .set_aio_ctx
= blk_root_set_aio_ctx
,
323 * Create a new BlockBackend with a reference count of one.
325 * @perm is a bitmasks of BLK_PERM_* constants which describes the permissions
326 * to request for a block driver node that is attached to this BlockBackend.
327 * @shared_perm is a bitmask which describes which permissions may be granted
328 * to other users of the attached node.
329 * Both sets of permissions can be changed later using blk_set_perm().
331 * Return the new BlockBackend on success, null on failure.
333 BlockBackend
*blk_new(AioContext
*ctx
, uint64_t perm
, uint64_t shared_perm
)
337 blk
= g_new0(BlockBackend
, 1);
341 blk
->shared_perm
= shared_perm
;
342 blk_set_enable_write_cache(blk
, true);
344 blk
->on_read_error
= BLOCKDEV_ON_ERROR_REPORT
;
345 blk
->on_write_error
= BLOCKDEV_ON_ERROR_ENOSPC
;
347 block_acct_init(&blk
->stats
);
349 qemu_co_queue_init(&blk
->queued_requests
);
350 notifier_list_init(&blk
->remove_bs_notifiers
);
351 notifier_list_init(&blk
->insert_bs_notifiers
);
352 QLIST_INIT(&blk
->aio_notifiers
);
354 QTAILQ_INSERT_TAIL(&block_backends
, blk
, link
);
359 * Create a new BlockBackend connected to an existing BlockDriverState.
361 * @perm is a bitmasks of BLK_PERM_* constants which describes the
362 * permissions to request for @bs that is attached to this
363 * BlockBackend. @shared_perm is a bitmask which describes which
364 * permissions may be granted to other users of the attached node.
365 * Both sets of permissions can be changed later using blk_set_perm().
367 * Return the new BlockBackend on success, null on failure.
369 BlockBackend
*blk_new_with_bs(BlockDriverState
*bs
, uint64_t perm
,
370 uint64_t shared_perm
, Error
**errp
)
372 BlockBackend
*blk
= blk_new(bdrv_get_aio_context(bs
), perm
, shared_perm
);
374 if (blk_insert_bs(blk
, bs
, errp
) < 0) {
382 * Creates a new BlockBackend, opens a new BlockDriverState, and connects both.
383 * The new BlockBackend is in the main AioContext.
385 * Just as with bdrv_open(), after having called this function the reference to
386 * @options belongs to the block layer (even on failure).
388 * TODO: Remove @filename and @flags; it should be possible to specify a whole
389 * BDS tree just by specifying the @options QDict (or @reference,
390 * alternatively). At the time of adding this function, this is not possible,
391 * though, so callers of this function have to be able to specify @filename and
394 BlockBackend
*blk_new_open(const char *filename
, const char *reference
,
395 QDict
*options
, int flags
, Error
**errp
)
398 BlockDriverState
*bs
;
401 /* blk_new_open() is mainly used in .bdrv_create implementations and the
402 * tools where sharing isn't a concern because the BDS stays private, so we
403 * just request permission according to the flags.
405 * The exceptions are xen_disk and blockdev_init(); in these cases, the
406 * caller of blk_new_open() doesn't make use of the permissions, but they
407 * shouldn't hurt either. We can still share everything here because the
408 * guest devices will add their own blockers if they can't share. */
409 if ((flags
& BDRV_O_NO_IO
) == 0) {
410 perm
|= BLK_PERM_CONSISTENT_READ
;
411 if (flags
& BDRV_O_RDWR
) {
412 perm
|= BLK_PERM_WRITE
;
415 if (flags
& BDRV_O_RESIZE
) {
416 perm
|= BLK_PERM_RESIZE
;
419 blk
= blk_new(qemu_get_aio_context(), perm
, BLK_PERM_ALL
);
420 bs
= bdrv_open(filename
, reference
, options
, flags
, errp
);
426 blk
->root
= bdrv_root_attach_child(bs
, "root", &child_root
, blk
->ctx
,
427 perm
, BLK_PERM_ALL
, blk
, errp
);
436 static void blk_delete(BlockBackend
*blk
)
438 assert(!blk
->refcnt
);
441 if (blk
->public.throttle_group_member
.throttle_state
) {
442 blk_io_limits_disable(blk
);
448 qemu_del_vm_change_state_handler(blk
->vmsh
);
451 assert(QLIST_EMPTY(&blk
->remove_bs_notifiers
.notifiers
));
452 assert(QLIST_EMPTY(&blk
->insert_bs_notifiers
.notifiers
));
453 assert(QLIST_EMPTY(&blk
->aio_notifiers
));
454 QTAILQ_REMOVE(&block_backends
, blk
, link
);
455 drive_info_del(blk
->legacy_dinfo
);
456 block_acct_cleanup(&blk
->stats
);
460 static void drive_info_del(DriveInfo
*dinfo
)
465 qemu_opts_del(dinfo
->opts
);
469 int blk_get_refcnt(BlockBackend
*blk
)
471 return blk
? blk
->refcnt
: 0;
475 * Increment @blk's reference count.
476 * @blk must not be null.
478 void blk_ref(BlockBackend
*blk
)
480 assert(blk
->refcnt
> 0);
485 * Decrement @blk's reference count.
486 * If this drops it to zero, destroy @blk.
487 * For convenience, do nothing if @blk is null.
489 void blk_unref(BlockBackend
*blk
)
492 assert(blk
->refcnt
> 0);
493 if (blk
->refcnt
> 1) {
497 /* blk_drain() cannot resurrect blk, nobody held a reference */
498 assert(blk
->refcnt
== 1);
506 * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
507 * ones which are hidden (i.e. are not referenced by the monitor).
509 BlockBackend
*blk_all_next(BlockBackend
*blk
)
511 return blk
? QTAILQ_NEXT(blk
, link
)
512 : QTAILQ_FIRST(&block_backends
);
515 void blk_remove_all_bs(void)
517 BlockBackend
*blk
= NULL
;
519 while ((blk
= blk_all_next(blk
)) != NULL
) {
520 AioContext
*ctx
= blk_get_aio_context(blk
);
522 aio_context_acquire(ctx
);
526 aio_context_release(ctx
);
531 * Return the monitor-owned BlockBackend after @blk.
532 * If @blk is null, return the first one.
533 * Else, return @blk's next sibling, which may be null.
535 * To iterate over all BlockBackends, do
536 * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
540 BlockBackend
*blk_next(BlockBackend
*blk
)
542 return blk
? QTAILQ_NEXT(blk
, monitor_link
)
543 : QTAILQ_FIRST(&monitor_block_backends
);
546 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
547 * the monitor or attached to a BlockBackend */
548 BlockDriverState
*bdrv_next(BdrvNextIterator
*it
)
550 BlockDriverState
*bs
, *old_bs
;
552 /* Must be called from the main loop */
553 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
555 /* First, return all root nodes of BlockBackends. In order to avoid
556 * returning a BDS twice when multiple BBs refer to it, we only return it
557 * if the BB is the first one in the parent list of the BDS. */
558 if (it
->phase
== BDRV_NEXT_BACKEND_ROOTS
) {
559 BlockBackend
*old_blk
= it
->blk
;
561 old_bs
= old_blk
? blk_bs(old_blk
) : NULL
;
564 it
->blk
= blk_all_next(it
->blk
);
565 bs
= it
->blk
? blk_bs(it
->blk
) : NULL
;
566 } while (it
->blk
&& (bs
== NULL
|| bdrv_first_blk(bs
) != it
->blk
));
578 it
->phase
= BDRV_NEXT_MONITOR_OWNED
;
583 /* Then return the monitor-owned BDSes without a BB attached. Ignore all
584 * BDSes that are attached to a BlockBackend here; they have been handled
585 * by the above block already */
587 it
->bs
= bdrv_next_monitor_owned(it
->bs
);
589 } while (bs
&& bdrv_has_blk(bs
));
599 static void bdrv_next_reset(BdrvNextIterator
*it
)
601 *it
= (BdrvNextIterator
) {
602 .phase
= BDRV_NEXT_BACKEND_ROOTS
,
606 BlockDriverState
*bdrv_first(BdrvNextIterator
*it
)
609 return bdrv_next(it
);
612 /* Must be called when aborting a bdrv_next() iteration before
613 * bdrv_next() returns NULL */
614 void bdrv_next_cleanup(BdrvNextIterator
*it
)
616 /* Must be called from the main loop */
617 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
619 if (it
->phase
== BDRV_NEXT_BACKEND_ROOTS
) {
621 bdrv_unref(blk_bs(it
->blk
));
632 * Add a BlockBackend into the list of backends referenced by the monitor, with
633 * the given @name acting as the handle for the monitor.
634 * Strictly for use by blockdev.c.
636 * @name must not be null or empty.
638 * Returns true on success and false on failure. In the latter case, an Error
639 * object is returned through @errp.
641 bool monitor_add_blk(BlockBackend
*blk
, const char *name
, Error
**errp
)
644 assert(name
&& name
[0]);
646 if (!id_wellformed(name
)) {
647 error_setg(errp
, "Invalid device name");
650 if (blk_by_name(name
)) {
651 error_setg(errp
, "Device with id '%s' already exists", name
);
654 if (bdrv_find_node(name
)) {
656 "Device name '%s' conflicts with an existing node name",
661 blk
->name
= g_strdup(name
);
662 QTAILQ_INSERT_TAIL(&monitor_block_backends
, blk
, monitor_link
);
667 * Remove a BlockBackend from the list of backends referenced by the monitor.
668 * Strictly for use by blockdev.c.
670 void monitor_remove_blk(BlockBackend
*blk
)
676 QTAILQ_REMOVE(&monitor_block_backends
, blk
, monitor_link
);
682 * Return @blk's name, a non-null string.
683 * Returns an empty string iff @blk is not referenced by the monitor.
685 const char *blk_name(const BlockBackend
*blk
)
687 return blk
->name
?: "";
691 * Return the BlockBackend with name @name if it exists, else null.
692 * @name must not be null.
694 BlockBackend
*blk_by_name(const char *name
)
696 BlockBackend
*blk
= NULL
;
699 while ((blk
= blk_next(blk
)) != NULL
) {
700 if (!strcmp(name
, blk
->name
)) {
708 * Return the BlockDriverState attached to @blk if any, else null.
710 BlockDriverState
*blk_bs(BlockBackend
*blk
)
712 return blk
->root
? blk
->root
->bs
: NULL
;
715 static BlockBackend
*bdrv_first_blk(BlockDriverState
*bs
)
718 QLIST_FOREACH(child
, &bs
->parents
, next_parent
) {
719 if (child
->role
== &child_root
) {
720 return child
->opaque
;
728 * Returns true if @bs has an associated BlockBackend.
730 bool bdrv_has_blk(BlockDriverState
*bs
)
732 return bdrv_first_blk(bs
) != NULL
;
736 * Returns true if @bs has only BlockBackends as parents.
738 bool bdrv_is_root_node(BlockDriverState
*bs
)
742 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
743 if (c
->role
!= &child_root
) {
752 * Return @blk's DriveInfo if any, else null.
754 DriveInfo
*blk_legacy_dinfo(BlockBackend
*blk
)
756 return blk
->legacy_dinfo
;
760 * Set @blk's DriveInfo to @dinfo, and return it.
761 * @blk must not have a DriveInfo set already.
762 * No other BlockBackend may have the same DriveInfo set.
764 DriveInfo
*blk_set_legacy_dinfo(BlockBackend
*blk
, DriveInfo
*dinfo
)
766 assert(!blk
->legacy_dinfo
);
767 return blk
->legacy_dinfo
= dinfo
;
771 * Return the BlockBackend with DriveInfo @dinfo.
774 BlockBackend
*blk_by_legacy_dinfo(DriveInfo
*dinfo
)
776 BlockBackend
*blk
= NULL
;
778 while ((blk
= blk_next(blk
)) != NULL
) {
779 if (blk
->legacy_dinfo
== dinfo
) {
787 * Returns a pointer to the publicly accessible fields of @blk.
789 BlockBackendPublic
*blk_get_public(BlockBackend
*blk
)
795 * Returns a BlockBackend given the associated @public fields.
797 BlockBackend
*blk_by_public(BlockBackendPublic
*public)
799 return container_of(public, BlockBackend
, public);
803 * Disassociates the currently associated BlockDriverState from @blk.
805 void blk_remove_bs(BlockBackend
*blk
)
807 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
808 BlockDriverState
*bs
;
810 notifier_list_notify(&blk
->remove_bs_notifiers
, blk
);
811 if (tgm
->throttle_state
) {
813 bdrv_drained_begin(bs
);
814 throttle_group_detach_aio_context(tgm
);
815 throttle_group_attach_aio_context(tgm
, qemu_get_aio_context());
816 bdrv_drained_end(bs
);
819 blk_update_root_state(blk
);
821 /* bdrv_root_unref_child() will cause blk->root to become stale and may
822 * switch to a completion coroutine later on. Let's drain all I/O here
823 * to avoid that and a potential QEMU crash.
826 bdrv_root_unref_child(blk
->root
);
831 * Associates a new BlockDriverState with @blk.
833 int blk_insert_bs(BlockBackend
*blk
, BlockDriverState
*bs
, Error
**errp
)
835 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
837 blk
->root
= bdrv_root_attach_child(bs
, "root", &child_root
, blk
->ctx
,
838 blk
->perm
, blk
->shared_perm
, blk
, errp
);
839 if (blk
->root
== NULL
) {
843 notifier_list_notify(&blk
->insert_bs_notifiers
, blk
);
844 if (tgm
->throttle_state
) {
845 throttle_group_detach_aio_context(tgm
);
846 throttle_group_attach_aio_context(tgm
, bdrv_get_aio_context(bs
));
853 * Sets the permission bitmasks that the user of the BlockBackend needs.
855 int blk_set_perm(BlockBackend
*blk
, uint64_t perm
, uint64_t shared_perm
,
860 if (blk
->root
&& !blk
->disable_perm
) {
861 ret
= bdrv_child_try_set_perm(blk
->root
, perm
, shared_perm
, errp
);
868 blk
->shared_perm
= shared_perm
;
873 void blk_get_perm(BlockBackend
*blk
, uint64_t *perm
, uint64_t *shared_perm
)
876 *shared_perm
= blk
->shared_perm
;
880 * Attach device model @dev to @blk.
881 * Return 0 on success, -EBUSY when a device model is attached already.
883 int blk_attach_dev(BlockBackend
*blk
, DeviceState
*dev
)
889 /* While migration is still incoming, we don't need to apply the
890 * permissions of guest device BlockBackends. We might still have a block
891 * job or NBD server writing to the image for storage migration. */
892 if (runstate_check(RUN_STATE_INMIGRATE
)) {
893 blk
->disable_perm
= true;
898 blk_iostatus_reset(blk
);
904 * Detach device model @dev from @blk.
905 * @dev must be currently attached to @blk.
907 void blk_detach_dev(BlockBackend
*blk
, DeviceState
*dev
)
909 assert(blk
->dev
== dev
);
912 blk
->dev_opaque
= NULL
;
913 blk
->guest_block_size
= 512;
914 blk_set_perm(blk
, 0, BLK_PERM_ALL
, &error_abort
);
919 * Return the device model attached to @blk if any, else null.
921 DeviceState
*blk_get_attached_dev(BlockBackend
*blk
)
926 /* Return the qdev ID, or if no ID is assigned the QOM path, of the block
927 * device attached to the BlockBackend. */
928 char *blk_get_attached_dev_id(BlockBackend
*blk
)
930 DeviceState
*dev
= blk
->dev
;
934 } else if (dev
->id
) {
935 return g_strdup(dev
->id
);
938 return object_get_canonical_path(OBJECT(dev
)) ?: g_strdup("");
942 * Return the BlockBackend which has the device model @dev attached if it
945 * @dev must not be null.
947 BlockBackend
*blk_by_dev(void *dev
)
949 BlockBackend
*blk
= NULL
;
952 while ((blk
= blk_all_next(blk
)) != NULL
) {
953 if (blk
->dev
== dev
) {
961 * Set @blk's device model callbacks to @ops.
962 * @opaque is the opaque argument to pass to the callbacks.
963 * This is for use by device models.
965 void blk_set_dev_ops(BlockBackend
*blk
, const BlockDevOps
*ops
,
969 blk
->dev_opaque
= opaque
;
971 /* Are we currently quiesced? Should we enforce this right now? */
972 if (blk
->quiesce_counter
&& ops
->drained_begin
) {
973 ops
->drained_begin(opaque
);
978 * Notify @blk's attached device model of media change.
980 * If @load is true, notify of media load. This action can fail, meaning that
981 * the medium cannot be loaded. @errp is set then.
983 * If @load is false, notify of media eject. This can never fail.
985 * Also send DEVICE_TRAY_MOVED events as appropriate.
987 void blk_dev_change_media_cb(BlockBackend
*blk
, bool load
, Error
**errp
)
989 if (blk
->dev_ops
&& blk
->dev_ops
->change_media_cb
) {
990 bool tray_was_open
, tray_is_open
;
991 Error
*local_err
= NULL
;
993 tray_was_open
= blk_dev_is_tray_open(blk
);
994 blk
->dev_ops
->change_media_cb(blk
->dev_opaque
, load
, &local_err
);
996 assert(load
== true);
997 error_propagate(errp
, local_err
);
1000 tray_is_open
= blk_dev_is_tray_open(blk
);
1002 if (tray_was_open
!= tray_is_open
) {
1003 char *id
= blk_get_attached_dev_id(blk
);
1004 qapi_event_send_device_tray_moved(blk_name(blk
), id
, tray_is_open
);
1010 static void blk_root_change_media(BdrvChild
*child
, bool load
)
1012 blk_dev_change_media_cb(child
->opaque
, load
, NULL
);
1016 * Does @blk's attached device model have removable media?
1017 * %true if no device model is attached.
1019 bool blk_dev_has_removable_media(BlockBackend
*blk
)
1021 return !blk
->dev
|| (blk
->dev_ops
&& blk
->dev_ops
->change_media_cb
);
1025 * Does @blk's attached device model have a tray?
1027 bool blk_dev_has_tray(BlockBackend
*blk
)
1029 return blk
->dev_ops
&& blk
->dev_ops
->is_tray_open
;
1033 * Notify @blk's attached device model of a media eject request.
1034 * If @force is true, the medium is about to be yanked out forcefully.
1036 void blk_dev_eject_request(BlockBackend
*blk
, bool force
)
1038 if (blk
->dev_ops
&& blk
->dev_ops
->eject_request_cb
) {
1039 blk
->dev_ops
->eject_request_cb(blk
->dev_opaque
, force
);
1044 * Does @blk's attached device model have a tray, and is it open?
1046 bool blk_dev_is_tray_open(BlockBackend
*blk
)
1048 if (blk_dev_has_tray(blk
)) {
1049 return blk
->dev_ops
->is_tray_open(blk
->dev_opaque
);
1055 * Does @blk's attached device model have the medium locked?
1056 * %false if the device model has no such lock.
1058 bool blk_dev_is_medium_locked(BlockBackend
*blk
)
1060 if (blk
->dev_ops
&& blk
->dev_ops
->is_medium_locked
) {
1061 return blk
->dev_ops
->is_medium_locked(blk
->dev_opaque
);
1067 * Notify @blk's attached device model of a backend size change.
1069 static void blk_root_resize(BdrvChild
*child
)
1071 BlockBackend
*blk
= child
->opaque
;
1073 if (blk
->dev_ops
&& blk
->dev_ops
->resize_cb
) {
1074 blk
->dev_ops
->resize_cb(blk
->dev_opaque
);
1078 void blk_iostatus_enable(BlockBackend
*blk
)
1080 blk
->iostatus_enabled
= true;
1081 blk
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
1084 /* The I/O status is only enabled if the drive explicitly
1085 * enables it _and_ the VM is configured to stop on errors */
1086 bool blk_iostatus_is_enabled(const BlockBackend
*blk
)
1088 return (blk
->iostatus_enabled
&&
1089 (blk
->on_write_error
== BLOCKDEV_ON_ERROR_ENOSPC
||
1090 blk
->on_write_error
== BLOCKDEV_ON_ERROR_STOP
||
1091 blk
->on_read_error
== BLOCKDEV_ON_ERROR_STOP
));
1094 BlockDeviceIoStatus
blk_iostatus(const BlockBackend
*blk
)
1096 return blk
->iostatus
;
1099 void blk_iostatus_disable(BlockBackend
*blk
)
1101 blk
->iostatus_enabled
= false;
1104 void blk_iostatus_reset(BlockBackend
*blk
)
1106 if (blk_iostatus_is_enabled(blk
)) {
1107 blk
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
1111 void blk_iostatus_set_err(BlockBackend
*blk
, int error
)
1113 assert(blk_iostatus_is_enabled(blk
));
1114 if (blk
->iostatus
== BLOCK_DEVICE_IO_STATUS_OK
) {
1115 blk
->iostatus
= error
== ENOSPC
? BLOCK_DEVICE_IO_STATUS_NOSPACE
:
1116 BLOCK_DEVICE_IO_STATUS_FAILED
;
1120 void blk_set_allow_write_beyond_eof(BlockBackend
*blk
, bool allow
)
1122 blk
->allow_write_beyond_eof
= allow
;
1125 void blk_set_allow_aio_context_change(BlockBackend
*blk
, bool allow
)
1127 blk
->allow_aio_context_change
= allow
;
1130 void blk_set_disable_request_queuing(BlockBackend
*blk
, bool disable
)
1132 blk
->disable_request_queuing
= disable
;
1135 static int blk_check_byte_request(BlockBackend
*blk
, int64_t offset
,
1140 if (size
> INT_MAX
) {
1144 if (!blk_is_available(blk
)) {
1152 if (!blk
->allow_write_beyond_eof
) {
1153 len
= blk_getlength(blk
);
1158 if (offset
> len
|| len
- offset
< size
) {
1166 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1167 static void coroutine_fn
blk_wait_while_drained(BlockBackend
*blk
)
1169 assert(blk
->in_flight
> 0);
1171 if (blk
->quiesce_counter
&& !blk
->disable_request_queuing
) {
1172 blk_dec_in_flight(blk
);
1173 qemu_co_queue_wait(&blk
->queued_requests
, NULL
);
1174 blk_inc_in_flight(blk
);
1178 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1179 static int coroutine_fn
1180 blk_do_preadv(BlockBackend
*blk
, int64_t offset
, unsigned int bytes
,
1181 QEMUIOVector
*qiov
, BdrvRequestFlags flags
)
1184 BlockDriverState
*bs
;
1186 blk_wait_while_drained(blk
);
1188 /* Call blk_bs() only after waiting, the graph may have changed */
1190 trace_blk_co_preadv(blk
, bs
, offset
, bytes
, flags
);
1192 ret
= blk_check_byte_request(blk
, offset
, bytes
);
1197 bdrv_inc_in_flight(bs
);
1199 /* throttling disk I/O */
1200 if (blk
->public.throttle_group_member
.throttle_state
) {
1201 throttle_group_co_io_limits_intercept(&blk
->public.throttle_group_member
,
1205 ret
= bdrv_co_preadv(blk
->root
, offset
, bytes
, qiov
, flags
);
1206 bdrv_dec_in_flight(bs
);
1210 int coroutine_fn
blk_co_preadv(BlockBackend
*blk
, int64_t offset
,
1211 unsigned int bytes
, QEMUIOVector
*qiov
,
1212 BdrvRequestFlags flags
)
1216 blk_inc_in_flight(blk
);
1217 ret
= blk_do_preadv(blk
, offset
, bytes
, qiov
, flags
);
1218 blk_dec_in_flight(blk
);
1223 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1224 static int coroutine_fn
1225 blk_do_pwritev_part(BlockBackend
*blk
, int64_t offset
, unsigned int bytes
,
1226 QEMUIOVector
*qiov
, size_t qiov_offset
,
1227 BdrvRequestFlags flags
)
1230 BlockDriverState
*bs
;
1232 blk_wait_while_drained(blk
);
1234 /* Call blk_bs() only after waiting, the graph may have changed */
1236 trace_blk_co_pwritev(blk
, bs
, offset
, bytes
, flags
);
1238 ret
= blk_check_byte_request(blk
, offset
, bytes
);
1243 bdrv_inc_in_flight(bs
);
1244 /* throttling disk I/O */
1245 if (blk
->public.throttle_group_member
.throttle_state
) {
1246 throttle_group_co_io_limits_intercept(&blk
->public.throttle_group_member
,
1250 if (!blk
->enable_write_cache
) {
1251 flags
|= BDRV_REQ_FUA
;
1254 ret
= bdrv_co_pwritev_part(blk
->root
, offset
, bytes
, qiov
, qiov_offset
,
1256 bdrv_dec_in_flight(bs
);
1260 int coroutine_fn
blk_co_pwritev_part(BlockBackend
*blk
, int64_t offset
,
1262 QEMUIOVector
*qiov
, size_t qiov_offset
,
1263 BdrvRequestFlags flags
)
1267 blk_inc_in_flight(blk
);
1268 ret
= blk_do_pwritev_part(blk
, offset
, bytes
, qiov
, qiov_offset
, flags
);
1269 blk_dec_in_flight(blk
);
1274 int coroutine_fn
blk_co_pwritev(BlockBackend
*blk
, int64_t offset
,
1275 unsigned int bytes
, QEMUIOVector
*qiov
,
1276 BdrvRequestFlags flags
)
1278 return blk_co_pwritev_part(blk
, offset
, bytes
, qiov
, 0, flags
);
1281 typedef struct BlkRwCo
{
1286 BdrvRequestFlags flags
;
1289 static void blk_read_entry(void *opaque
)
1291 BlkRwCo
*rwco
= opaque
;
1292 QEMUIOVector
*qiov
= rwco
->iobuf
;
1294 rwco
->ret
= blk_do_preadv(rwco
->blk
, rwco
->offset
, qiov
->size
,
1299 static void blk_write_entry(void *opaque
)
1301 BlkRwCo
*rwco
= opaque
;
1302 QEMUIOVector
*qiov
= rwco
->iobuf
;
1304 rwco
->ret
= blk_do_pwritev_part(rwco
->blk
, rwco
->offset
, qiov
->size
,
1305 qiov
, 0, rwco
->flags
);
1309 static int blk_prw(BlockBackend
*blk
, int64_t offset
, uint8_t *buf
,
1310 int64_t bytes
, CoroutineEntry co_entry
,
1311 BdrvRequestFlags flags
)
1313 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, buf
, bytes
);
1322 blk_inc_in_flight(blk
);
1323 if (qemu_in_coroutine()) {
1324 /* Fast-path if already in coroutine context */
1327 Coroutine
*co
= qemu_coroutine_create(co_entry
, &rwco
);
1328 bdrv_coroutine_enter(blk_bs(blk
), co
);
1329 BDRV_POLL_WHILE(blk_bs(blk
), rwco
.ret
== NOT_DONE
);
1331 blk_dec_in_flight(blk
);
1336 int blk_pwrite_zeroes(BlockBackend
*blk
, int64_t offset
,
1337 int bytes
, BdrvRequestFlags flags
)
1339 return blk_prw(blk
, offset
, NULL
, bytes
, blk_write_entry
,
1340 flags
| BDRV_REQ_ZERO_WRITE
);
1343 int blk_make_zero(BlockBackend
*blk
, BdrvRequestFlags flags
)
1345 return bdrv_make_zero(blk
->root
, flags
);
1348 void blk_inc_in_flight(BlockBackend
*blk
)
1350 atomic_inc(&blk
->in_flight
);
1353 void blk_dec_in_flight(BlockBackend
*blk
)
1355 atomic_dec(&blk
->in_flight
);
1359 static void error_callback_bh(void *opaque
)
1361 struct BlockBackendAIOCB
*acb
= opaque
;
1363 blk_dec_in_flight(acb
->blk
);
1364 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
1365 qemu_aio_unref(acb
);
1368 BlockAIOCB
*blk_abort_aio_request(BlockBackend
*blk
,
1369 BlockCompletionFunc
*cb
,
1370 void *opaque
, int ret
)
1372 struct BlockBackendAIOCB
*acb
;
1374 blk_inc_in_flight(blk
);
1375 acb
= blk_aio_get(&block_backend_aiocb_info
, blk
, cb
, opaque
);
1379 replay_bh_schedule_oneshot_event(blk_get_aio_context(blk
),
1380 error_callback_bh
, acb
);
1381 return &acb
->common
;
1384 typedef struct BlkAioEmAIOCB
{
1391 static const AIOCBInfo blk_aio_em_aiocb_info
= {
1392 .aiocb_size
= sizeof(BlkAioEmAIOCB
),
1395 static void blk_aio_complete(BlkAioEmAIOCB
*acb
)
1397 if (acb
->has_returned
) {
1398 acb
->common
.cb(acb
->common
.opaque
, acb
->rwco
.ret
);
1399 blk_dec_in_flight(acb
->rwco
.blk
);
1400 qemu_aio_unref(acb
);
1404 static void blk_aio_complete_bh(void *opaque
)
1406 BlkAioEmAIOCB
*acb
= opaque
;
1407 assert(acb
->has_returned
);
1408 blk_aio_complete(acb
);
1411 static BlockAIOCB
*blk_aio_prwv(BlockBackend
*blk
, int64_t offset
, int bytes
,
1412 void *iobuf
, CoroutineEntry co_entry
,
1413 BdrvRequestFlags flags
,
1414 BlockCompletionFunc
*cb
, void *opaque
)
1419 blk_inc_in_flight(blk
);
1420 acb
= blk_aio_get(&blk_aio_em_aiocb_info
, blk
, cb
, opaque
);
1421 acb
->rwco
= (BlkRwCo
) {
1429 acb
->has_returned
= false;
1431 co
= qemu_coroutine_create(co_entry
, acb
);
1432 bdrv_coroutine_enter(blk_bs(blk
), co
);
1434 acb
->has_returned
= true;
1435 if (acb
->rwco
.ret
!= NOT_DONE
) {
1436 replay_bh_schedule_oneshot_event(blk_get_aio_context(blk
),
1437 blk_aio_complete_bh
, acb
);
1440 return &acb
->common
;
1443 static void blk_aio_read_entry(void *opaque
)
1445 BlkAioEmAIOCB
*acb
= opaque
;
1446 BlkRwCo
*rwco
= &acb
->rwco
;
1447 QEMUIOVector
*qiov
= rwco
->iobuf
;
1449 assert(qiov
->size
== acb
->bytes
);
1450 rwco
->ret
= blk_do_preadv(rwco
->blk
, rwco
->offset
, acb
->bytes
,
1452 blk_aio_complete(acb
);
1455 static void blk_aio_write_entry(void *opaque
)
1457 BlkAioEmAIOCB
*acb
= opaque
;
1458 BlkRwCo
*rwco
= &acb
->rwco
;
1459 QEMUIOVector
*qiov
= rwco
->iobuf
;
1461 assert(!qiov
|| qiov
->size
== acb
->bytes
);
1462 rwco
->ret
= blk_do_pwritev_part(rwco
->blk
, rwco
->offset
, acb
->bytes
,
1463 qiov
, 0, rwco
->flags
);
1464 blk_aio_complete(acb
);
1467 BlockAIOCB
*blk_aio_pwrite_zeroes(BlockBackend
*blk
, int64_t offset
,
1468 int count
, BdrvRequestFlags flags
,
1469 BlockCompletionFunc
*cb
, void *opaque
)
1471 return blk_aio_prwv(blk
, offset
, count
, NULL
, blk_aio_write_entry
,
1472 flags
| BDRV_REQ_ZERO_WRITE
, cb
, opaque
);
1475 int blk_pread(BlockBackend
*blk
, int64_t offset
, void *buf
, int count
)
1477 int ret
= blk_prw(blk
, offset
, buf
, count
, blk_read_entry
, 0);
1484 int blk_pwrite(BlockBackend
*blk
, int64_t offset
, const void *buf
, int count
,
1485 BdrvRequestFlags flags
)
1487 int ret
= blk_prw(blk
, offset
, (void *) buf
, count
, blk_write_entry
,
1495 int64_t blk_getlength(BlockBackend
*blk
)
1497 if (!blk_is_available(blk
)) {
1501 return bdrv_getlength(blk_bs(blk
));
1504 void blk_get_geometry(BlockBackend
*blk
, uint64_t *nb_sectors_ptr
)
1507 *nb_sectors_ptr
= 0;
1509 bdrv_get_geometry(blk_bs(blk
), nb_sectors_ptr
);
1513 int64_t blk_nb_sectors(BlockBackend
*blk
)
1515 if (!blk_is_available(blk
)) {
1519 return bdrv_nb_sectors(blk_bs(blk
));
1522 BlockAIOCB
*blk_aio_preadv(BlockBackend
*blk
, int64_t offset
,
1523 QEMUIOVector
*qiov
, BdrvRequestFlags flags
,
1524 BlockCompletionFunc
*cb
, void *opaque
)
1526 return blk_aio_prwv(blk
, offset
, qiov
->size
, qiov
,
1527 blk_aio_read_entry
, flags
, cb
, opaque
);
1530 BlockAIOCB
*blk_aio_pwritev(BlockBackend
*blk
, int64_t offset
,
1531 QEMUIOVector
*qiov
, BdrvRequestFlags flags
,
1532 BlockCompletionFunc
*cb
, void *opaque
)
1534 return blk_aio_prwv(blk
, offset
, qiov
->size
, qiov
,
1535 blk_aio_write_entry
, flags
, cb
, opaque
);
1538 void blk_aio_cancel(BlockAIOCB
*acb
)
1540 bdrv_aio_cancel(acb
);
1543 void blk_aio_cancel_async(BlockAIOCB
*acb
)
1545 bdrv_aio_cancel_async(acb
);
1548 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1549 static int coroutine_fn
1550 blk_do_ioctl(BlockBackend
*blk
, unsigned long int req
, void *buf
)
1552 blk_wait_while_drained(blk
);
1554 if (!blk_is_available(blk
)) {
1558 return bdrv_co_ioctl(blk_bs(blk
), req
, buf
);
1561 static void blk_ioctl_entry(void *opaque
)
1563 BlkRwCo
*rwco
= opaque
;
1564 QEMUIOVector
*qiov
= rwco
->iobuf
;
1566 rwco
->ret
= blk_do_ioctl(rwco
->blk
, rwco
->offset
, qiov
->iov
[0].iov_base
);
1570 int blk_ioctl(BlockBackend
*blk
, unsigned long int req
, void *buf
)
1572 return blk_prw(blk
, req
, buf
, 0, blk_ioctl_entry
, 0);
1575 static void blk_aio_ioctl_entry(void *opaque
)
1577 BlkAioEmAIOCB
*acb
= opaque
;
1578 BlkRwCo
*rwco
= &acb
->rwco
;
1580 rwco
->ret
= blk_do_ioctl(rwco
->blk
, rwco
->offset
, rwco
->iobuf
);
1582 blk_aio_complete(acb
);
1585 BlockAIOCB
*blk_aio_ioctl(BlockBackend
*blk
, unsigned long int req
, void *buf
,
1586 BlockCompletionFunc
*cb
, void *opaque
)
1588 return blk_aio_prwv(blk
, req
, 0, buf
, blk_aio_ioctl_entry
, 0, cb
, opaque
);
1591 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1592 static int coroutine_fn
1593 blk_do_pdiscard(BlockBackend
*blk
, int64_t offset
, int bytes
)
1597 blk_wait_while_drained(blk
);
1599 ret
= blk_check_byte_request(blk
, offset
, bytes
);
1604 return bdrv_co_pdiscard(blk
->root
, offset
, bytes
);
1607 static void blk_aio_pdiscard_entry(void *opaque
)
1609 BlkAioEmAIOCB
*acb
= opaque
;
1610 BlkRwCo
*rwco
= &acb
->rwco
;
1612 rwco
->ret
= blk_do_pdiscard(rwco
->blk
, rwco
->offset
, acb
->bytes
);
1613 blk_aio_complete(acb
);
1616 BlockAIOCB
*blk_aio_pdiscard(BlockBackend
*blk
,
1617 int64_t offset
, int bytes
,
1618 BlockCompletionFunc
*cb
, void *opaque
)
1620 return blk_aio_prwv(blk
, offset
, bytes
, NULL
, blk_aio_pdiscard_entry
, 0,
1624 int coroutine_fn
blk_co_pdiscard(BlockBackend
*blk
, int64_t offset
, int bytes
)
1628 blk_inc_in_flight(blk
);
1629 ret
= blk_do_pdiscard(blk
, offset
, bytes
);
1630 blk_dec_in_flight(blk
);
1635 static void blk_pdiscard_entry(void *opaque
)
1637 BlkRwCo
*rwco
= opaque
;
1638 QEMUIOVector
*qiov
= rwco
->iobuf
;
1640 rwco
->ret
= blk_do_pdiscard(rwco
->blk
, rwco
->offset
, qiov
->size
);
1644 int blk_pdiscard(BlockBackend
*blk
, int64_t offset
, int bytes
)
1646 return blk_prw(blk
, offset
, NULL
, bytes
, blk_pdiscard_entry
, 0);
1649 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1650 static int coroutine_fn
blk_do_flush(BlockBackend
*blk
)
1652 blk_wait_while_drained(blk
);
1654 if (!blk_is_available(blk
)) {
1658 return bdrv_co_flush(blk_bs(blk
));
1661 static void blk_aio_flush_entry(void *opaque
)
1663 BlkAioEmAIOCB
*acb
= opaque
;
1664 BlkRwCo
*rwco
= &acb
->rwco
;
1666 rwco
->ret
= blk_do_flush(rwco
->blk
);
1667 blk_aio_complete(acb
);
1670 BlockAIOCB
*blk_aio_flush(BlockBackend
*blk
,
1671 BlockCompletionFunc
*cb
, void *opaque
)
1673 return blk_aio_prwv(blk
, 0, 0, NULL
, blk_aio_flush_entry
, 0, cb
, opaque
);
1676 int coroutine_fn
blk_co_flush(BlockBackend
*blk
)
1680 blk_inc_in_flight(blk
);
1681 ret
= blk_do_flush(blk
);
1682 blk_dec_in_flight(blk
);
1687 static void blk_flush_entry(void *opaque
)
1689 BlkRwCo
*rwco
= opaque
;
1690 rwco
->ret
= blk_do_flush(rwco
->blk
);
1694 int blk_flush(BlockBackend
*blk
)
1696 return blk_prw(blk
, 0, NULL
, 0, blk_flush_entry
, 0);
1699 void blk_drain(BlockBackend
*blk
)
1701 BlockDriverState
*bs
= blk_bs(blk
);
1704 bdrv_drained_begin(bs
);
1707 /* We may have -ENOMEDIUM completions in flight */
1708 AIO_WAIT_WHILE(blk_get_aio_context(blk
),
1709 atomic_mb_read(&blk
->in_flight
) > 0);
1712 bdrv_drained_end(bs
);
1716 void blk_drain_all(void)
1718 BlockBackend
*blk
= NULL
;
1720 bdrv_drain_all_begin();
1722 while ((blk
= blk_all_next(blk
)) != NULL
) {
1723 AioContext
*ctx
= blk_get_aio_context(blk
);
1725 aio_context_acquire(ctx
);
1727 /* We may have -ENOMEDIUM completions in flight */
1728 AIO_WAIT_WHILE(ctx
, atomic_mb_read(&blk
->in_flight
) > 0);
1730 aio_context_release(ctx
);
1733 bdrv_drain_all_end();
1736 void blk_set_on_error(BlockBackend
*blk
, BlockdevOnError on_read_error
,
1737 BlockdevOnError on_write_error
)
1739 blk
->on_read_error
= on_read_error
;
1740 blk
->on_write_error
= on_write_error
;
1743 BlockdevOnError
blk_get_on_error(BlockBackend
*blk
, bool is_read
)
1745 return is_read
? blk
->on_read_error
: blk
->on_write_error
;
1748 BlockErrorAction
blk_get_error_action(BlockBackend
*blk
, bool is_read
,
1751 BlockdevOnError on_err
= blk_get_on_error(blk
, is_read
);
1754 case BLOCKDEV_ON_ERROR_ENOSPC
:
1755 return (error
== ENOSPC
) ?
1756 BLOCK_ERROR_ACTION_STOP
: BLOCK_ERROR_ACTION_REPORT
;
1757 case BLOCKDEV_ON_ERROR_STOP
:
1758 return BLOCK_ERROR_ACTION_STOP
;
1759 case BLOCKDEV_ON_ERROR_REPORT
:
1760 return BLOCK_ERROR_ACTION_REPORT
;
1761 case BLOCKDEV_ON_ERROR_IGNORE
:
1762 return BLOCK_ERROR_ACTION_IGNORE
;
1763 case BLOCKDEV_ON_ERROR_AUTO
:
1769 static void send_qmp_error_event(BlockBackend
*blk
,
1770 BlockErrorAction action
,
1771 bool is_read
, int error
)
1773 IoOperationType optype
;
1774 BlockDriverState
*bs
= blk_bs(blk
);
1776 optype
= is_read
? IO_OPERATION_TYPE_READ
: IO_OPERATION_TYPE_WRITE
;
1777 qapi_event_send_block_io_error(blk_name(blk
), !!bs
,
1778 bs
? bdrv_get_node_name(bs
) : NULL
, optype
,
1779 action
, blk_iostatus_is_enabled(blk
),
1780 error
== ENOSPC
, strerror(error
));
1783 /* This is done by device models because, while the block layer knows
1784 * about the error, it does not know whether an operation comes from
1785 * the device or the block layer (from a job, for example).
1787 void blk_error_action(BlockBackend
*blk
, BlockErrorAction action
,
1788 bool is_read
, int error
)
1792 if (action
== BLOCK_ERROR_ACTION_STOP
) {
1793 /* First set the iostatus, so that "info block" returns an iostatus
1794 * that matches the events raised so far (an additional error iostatus
1795 * is fine, but not a lost one).
1797 blk_iostatus_set_err(blk
, error
);
1799 /* Then raise the request to stop the VM and the event.
1800 * qemu_system_vmstop_request_prepare has two effects. First,
1801 * it ensures that the STOP event always comes after the
1802 * BLOCK_IO_ERROR event. Second, it ensures that even if management
1803 * can observe the STOP event and do a "cont" before the STOP
1804 * event is issued, the VM will not stop. In this case, vm_start()
1805 * also ensures that the STOP/RESUME pair of events is emitted.
1807 qemu_system_vmstop_request_prepare();
1808 send_qmp_error_event(blk
, action
, is_read
, error
);
1809 qemu_system_vmstop_request(RUN_STATE_IO_ERROR
);
1811 send_qmp_error_event(blk
, action
, is_read
, error
);
1815 bool blk_is_read_only(BlockBackend
*blk
)
1817 BlockDriverState
*bs
= blk_bs(blk
);
1820 return bdrv_is_read_only(bs
);
1822 return blk
->root_state
.read_only
;
1826 bool blk_is_sg(BlockBackend
*blk
)
1828 BlockDriverState
*bs
= blk_bs(blk
);
1834 return bdrv_is_sg(bs
);
1837 bool blk_enable_write_cache(BlockBackend
*blk
)
1839 return blk
->enable_write_cache
;
1842 void blk_set_enable_write_cache(BlockBackend
*blk
, bool wce
)
1844 blk
->enable_write_cache
= wce
;
1847 void blk_invalidate_cache(BlockBackend
*blk
, Error
**errp
)
1849 BlockDriverState
*bs
= blk_bs(blk
);
1852 error_setg(errp
, "Device '%s' has no medium", blk
->name
);
1856 bdrv_invalidate_cache(bs
, errp
);
1859 bool blk_is_inserted(BlockBackend
*blk
)
1861 BlockDriverState
*bs
= blk_bs(blk
);
1863 return bs
&& bdrv_is_inserted(bs
);
1866 bool blk_is_available(BlockBackend
*blk
)
1868 return blk_is_inserted(blk
) && !blk_dev_is_tray_open(blk
);
1871 void blk_lock_medium(BlockBackend
*blk
, bool locked
)
1873 BlockDriverState
*bs
= blk_bs(blk
);
1876 bdrv_lock_medium(bs
, locked
);
1880 void blk_eject(BlockBackend
*blk
, bool eject_flag
)
1882 BlockDriverState
*bs
= blk_bs(blk
);
1886 bdrv_eject(bs
, eject_flag
);
1889 /* Whether or not we ejected on the backend,
1890 * the frontend experienced a tray event. */
1891 id
= blk_get_attached_dev_id(blk
);
1892 qapi_event_send_device_tray_moved(blk_name(blk
), id
,
1897 int blk_get_flags(BlockBackend
*blk
)
1899 BlockDriverState
*bs
= blk_bs(blk
);
1902 return bdrv_get_flags(bs
);
1904 return blk
->root_state
.open_flags
;
1908 /* Returns the minimum request alignment, in bytes; guaranteed nonzero */
1909 uint32_t blk_get_request_alignment(BlockBackend
*blk
)
1911 BlockDriverState
*bs
= blk_bs(blk
);
1912 return bs
? bs
->bl
.request_alignment
: BDRV_SECTOR_SIZE
;
1915 /* Returns the maximum transfer length, in bytes; guaranteed nonzero */
1916 uint32_t blk_get_max_transfer(BlockBackend
*blk
)
1918 BlockDriverState
*bs
= blk_bs(blk
);
1922 max
= bs
->bl
.max_transfer
;
1924 return MIN_NON_ZERO(max
, INT_MAX
);
1927 int blk_get_max_iov(BlockBackend
*blk
)
1929 return blk
->root
->bs
->bl
.max_iov
;
1932 void blk_set_guest_block_size(BlockBackend
*blk
, int align
)
1934 blk
->guest_block_size
= align
;
1937 void *blk_try_blockalign(BlockBackend
*blk
, size_t size
)
1939 return qemu_try_blockalign(blk
? blk_bs(blk
) : NULL
, size
);
1942 void *blk_blockalign(BlockBackend
*blk
, size_t size
)
1944 return qemu_blockalign(blk
? blk_bs(blk
) : NULL
, size
);
1947 bool blk_op_is_blocked(BlockBackend
*blk
, BlockOpType op
, Error
**errp
)
1949 BlockDriverState
*bs
= blk_bs(blk
);
1955 return bdrv_op_is_blocked(bs
, op
, errp
);
1958 void blk_op_unblock(BlockBackend
*blk
, BlockOpType op
, Error
*reason
)
1960 BlockDriverState
*bs
= blk_bs(blk
);
1963 bdrv_op_unblock(bs
, op
, reason
);
1967 void blk_op_block_all(BlockBackend
*blk
, Error
*reason
)
1969 BlockDriverState
*bs
= blk_bs(blk
);
1972 bdrv_op_block_all(bs
, reason
);
1976 void blk_op_unblock_all(BlockBackend
*blk
, Error
*reason
)
1978 BlockDriverState
*bs
= blk_bs(blk
);
1981 bdrv_op_unblock_all(bs
, reason
);
1985 AioContext
*blk_get_aio_context(BlockBackend
*blk
)
1987 BlockDriverState
*bs
= blk_bs(blk
);
1990 AioContext
*ctx
= bdrv_get_aio_context(blk_bs(blk
));
1991 assert(ctx
== blk
->ctx
);
1997 static AioContext
*blk_aiocb_get_aio_context(BlockAIOCB
*acb
)
1999 BlockBackendAIOCB
*blk_acb
= DO_UPCAST(BlockBackendAIOCB
, common
, acb
);
2000 return blk_get_aio_context(blk_acb
->blk
);
2003 static int blk_do_set_aio_context(BlockBackend
*blk
, AioContext
*new_context
,
2004 bool update_root_node
, Error
**errp
)
2006 BlockDriverState
*bs
= blk_bs(blk
);
2007 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
2011 if (update_root_node
) {
2012 ret
= bdrv_child_try_set_aio_context(bs
, new_context
, blk
->root
,
2018 if (tgm
->throttle_state
) {
2019 bdrv_drained_begin(bs
);
2020 throttle_group_detach_aio_context(tgm
);
2021 throttle_group_attach_aio_context(tgm
, new_context
);
2022 bdrv_drained_end(bs
);
2026 blk
->ctx
= new_context
;
2030 int blk_set_aio_context(BlockBackend
*blk
, AioContext
*new_context
,
2033 return blk_do_set_aio_context(blk
, new_context
, true, errp
);
2036 static bool blk_root_can_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
2037 GSList
**ignore
, Error
**errp
)
2039 BlockBackend
*blk
= child
->opaque
;
2041 if (blk
->allow_aio_context_change
) {
2045 /* Only manually created BlockBackends that are not attached to anything
2046 * can change their AioContext without updating their user. */
2047 if (!blk
->name
|| blk
->dev
) {
2048 /* TODO Add BB name/QOM path */
2049 error_setg(errp
, "Cannot change iothread of active block backend");
2056 static void blk_root_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
2059 BlockBackend
*blk
= child
->opaque
;
2060 blk_do_set_aio_context(blk
, ctx
, false, &error_abort
);
2063 void blk_add_aio_context_notifier(BlockBackend
*blk
,
2064 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
2065 void (*detach_aio_context
)(void *opaque
), void *opaque
)
2067 BlockBackendAioNotifier
*notifier
;
2068 BlockDriverState
*bs
= blk_bs(blk
);
2070 notifier
= g_new(BlockBackendAioNotifier
, 1);
2071 notifier
->attached_aio_context
= attached_aio_context
;
2072 notifier
->detach_aio_context
= detach_aio_context
;
2073 notifier
->opaque
= opaque
;
2074 QLIST_INSERT_HEAD(&blk
->aio_notifiers
, notifier
, list
);
2077 bdrv_add_aio_context_notifier(bs
, attached_aio_context
,
2078 detach_aio_context
, opaque
);
2082 void blk_remove_aio_context_notifier(BlockBackend
*blk
,
2083 void (*attached_aio_context
)(AioContext
*,
2085 void (*detach_aio_context
)(void *),
2088 BlockBackendAioNotifier
*notifier
;
2089 BlockDriverState
*bs
= blk_bs(blk
);
2092 bdrv_remove_aio_context_notifier(bs
, attached_aio_context
,
2093 detach_aio_context
, opaque
);
2096 QLIST_FOREACH(notifier
, &blk
->aio_notifiers
, list
) {
2097 if (notifier
->attached_aio_context
== attached_aio_context
&&
2098 notifier
->detach_aio_context
== detach_aio_context
&&
2099 notifier
->opaque
== opaque
) {
2100 QLIST_REMOVE(notifier
, list
);
2109 void blk_add_remove_bs_notifier(BlockBackend
*blk
, Notifier
*notify
)
2111 notifier_list_add(&blk
->remove_bs_notifiers
, notify
);
2114 void blk_add_insert_bs_notifier(BlockBackend
*blk
, Notifier
*notify
)
2116 notifier_list_add(&blk
->insert_bs_notifiers
, notify
);
2119 void blk_io_plug(BlockBackend
*blk
)
2121 BlockDriverState
*bs
= blk_bs(blk
);
2128 void blk_io_unplug(BlockBackend
*blk
)
2130 BlockDriverState
*bs
= blk_bs(blk
);
2137 BlockAcctStats
*blk_get_stats(BlockBackend
*blk
)
2142 void *blk_aio_get(const AIOCBInfo
*aiocb_info
, BlockBackend
*blk
,
2143 BlockCompletionFunc
*cb
, void *opaque
)
2145 return qemu_aio_get(aiocb_info
, blk_bs(blk
), cb
, opaque
);
2148 int coroutine_fn
blk_co_pwrite_zeroes(BlockBackend
*blk
, int64_t offset
,
2149 int bytes
, BdrvRequestFlags flags
)
2151 return blk_co_pwritev(blk
, offset
, bytes
, NULL
,
2152 flags
| BDRV_REQ_ZERO_WRITE
);
2155 int blk_pwrite_compressed(BlockBackend
*blk
, int64_t offset
, const void *buf
,
2158 return blk_prw(blk
, offset
, (void *) buf
, count
, blk_write_entry
,
2159 BDRV_REQ_WRITE_COMPRESSED
);
2162 int blk_truncate(BlockBackend
*blk
, int64_t offset
, bool exact
,
2163 PreallocMode prealloc
, BdrvRequestFlags flags
, Error
**errp
)
2165 if (!blk_is_available(blk
)) {
2166 error_setg(errp
, "No medium inserted");
2170 return bdrv_truncate(blk
->root
, offset
, exact
, prealloc
, flags
, errp
);
2173 int blk_save_vmstate(BlockBackend
*blk
, const uint8_t *buf
,
2174 int64_t pos
, int size
)
2178 if (!blk_is_available(blk
)) {
2182 ret
= bdrv_save_vmstate(blk_bs(blk
), buf
, pos
, size
);
2187 if (ret
== size
&& !blk
->enable_write_cache
) {
2188 ret
= bdrv_flush(blk_bs(blk
));
2191 return ret
< 0 ? ret
: size
;
2194 int blk_load_vmstate(BlockBackend
*blk
, uint8_t *buf
, int64_t pos
, int size
)
2196 if (!blk_is_available(blk
)) {
2200 return bdrv_load_vmstate(blk_bs(blk
), buf
, pos
, size
);
2203 int blk_probe_blocksizes(BlockBackend
*blk
, BlockSizes
*bsz
)
2205 if (!blk_is_available(blk
)) {
2209 return bdrv_probe_blocksizes(blk_bs(blk
), bsz
);
2212 int blk_probe_geometry(BlockBackend
*blk
, HDGeometry
*geo
)
2214 if (!blk_is_available(blk
)) {
2218 return bdrv_probe_geometry(blk_bs(blk
), geo
);
2222 * Updates the BlockBackendRootState object with data from the currently
2223 * attached BlockDriverState.
2225 void blk_update_root_state(BlockBackend
*blk
)
2229 blk
->root_state
.open_flags
= blk
->root
->bs
->open_flags
;
2230 blk
->root_state
.read_only
= blk
->root
->bs
->read_only
;
2231 blk
->root_state
.detect_zeroes
= blk
->root
->bs
->detect_zeroes
;
2235 * Returns the detect-zeroes setting to be used for bdrv_open() of a
2236 * BlockDriverState which is supposed to inherit the root state.
2238 bool blk_get_detect_zeroes_from_root_state(BlockBackend
*blk
)
2240 return blk
->root_state
.detect_zeroes
;
2244 * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
2245 * supposed to inherit the root state.
2247 int blk_get_open_flags_from_root_state(BlockBackend
*blk
)
2251 bs_flags
= blk
->root_state
.read_only
? 0 : BDRV_O_RDWR
;
2252 bs_flags
|= blk
->root_state
.open_flags
& ~BDRV_O_RDWR
;
2257 BlockBackendRootState
*blk_get_root_state(BlockBackend
*blk
)
2259 return &blk
->root_state
;
2262 int blk_commit_all(void)
2264 BlockBackend
*blk
= NULL
;
2266 while ((blk
= blk_all_next(blk
)) != NULL
) {
2267 AioContext
*aio_context
= blk_get_aio_context(blk
);
2269 aio_context_acquire(aio_context
);
2270 if (blk_is_inserted(blk
) && blk
->root
->bs
->backing
) {
2271 int ret
= bdrv_commit(blk
->root
->bs
);
2273 aio_context_release(aio_context
);
2277 aio_context_release(aio_context
);
2283 /* throttling disk I/O limits */
2284 void blk_set_io_limits(BlockBackend
*blk
, ThrottleConfig
*cfg
)
2286 throttle_group_config(&blk
->public.throttle_group_member
, cfg
);
2289 void blk_io_limits_disable(BlockBackend
*blk
)
2291 BlockDriverState
*bs
= blk_bs(blk
);
2292 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
2293 assert(tgm
->throttle_state
);
2295 bdrv_drained_begin(bs
);
2297 throttle_group_unregister_tgm(tgm
);
2299 bdrv_drained_end(bs
);
2303 /* should be called before blk_set_io_limits if a limit is set */
2304 void blk_io_limits_enable(BlockBackend
*blk
, const char *group
)
2306 assert(!blk
->public.throttle_group_member
.throttle_state
);
2307 throttle_group_register_tgm(&blk
->public.throttle_group_member
,
2308 group
, blk_get_aio_context(blk
));
2311 void blk_io_limits_update_group(BlockBackend
*blk
, const char *group
)
2313 /* this BB is not part of any group */
2314 if (!blk
->public.throttle_group_member
.throttle_state
) {
2318 /* this BB is a part of the same group than the one we want */
2319 if (!g_strcmp0(throttle_group_get_name(&blk
->public.throttle_group_member
),
2324 /* need to change the group this bs belong to */
2325 blk_io_limits_disable(blk
);
2326 blk_io_limits_enable(blk
, group
);
2329 static void blk_root_drained_begin(BdrvChild
*child
)
2331 BlockBackend
*blk
= child
->opaque
;
2333 if (++blk
->quiesce_counter
== 1) {
2334 if (blk
->dev_ops
&& blk
->dev_ops
->drained_begin
) {
2335 blk
->dev_ops
->drained_begin(blk
->dev_opaque
);
2339 /* Note that blk->root may not be accessible here yet if we are just
2340 * attaching to a BlockDriverState that is drained. Use child instead. */
2342 if (atomic_fetch_inc(&blk
->public.throttle_group_member
.io_limits_disabled
) == 0) {
2343 throttle_group_restart_tgm(&blk
->public.throttle_group_member
);
2347 static bool blk_root_drained_poll(BdrvChild
*child
)
2349 BlockBackend
*blk
= child
->opaque
;
2350 assert(blk
->quiesce_counter
);
2351 return !!blk
->in_flight
;
2354 static void blk_root_drained_end(BdrvChild
*child
, int *drained_end_counter
)
2356 BlockBackend
*blk
= child
->opaque
;
2357 assert(blk
->quiesce_counter
);
2359 assert(blk
->public.throttle_group_member
.io_limits_disabled
);
2360 atomic_dec(&blk
->public.throttle_group_member
.io_limits_disabled
);
2362 if (--blk
->quiesce_counter
== 0) {
2363 if (blk
->dev_ops
&& blk
->dev_ops
->drained_end
) {
2364 blk
->dev_ops
->drained_end(blk
->dev_opaque
);
2366 while (qemu_co_enter_next(&blk
->queued_requests
, NULL
)) {
2367 /* Resume all queued requests */
2372 void blk_register_buf(BlockBackend
*blk
, void *host
, size_t size
)
2374 bdrv_register_buf(blk_bs(blk
), host
, size
);
2377 void blk_unregister_buf(BlockBackend
*blk
, void *host
)
2379 bdrv_unregister_buf(blk_bs(blk
), host
);
2382 int coroutine_fn
blk_co_copy_range(BlockBackend
*blk_in
, int64_t off_in
,
2383 BlockBackend
*blk_out
, int64_t off_out
,
2384 int bytes
, BdrvRequestFlags read_flags
,
2385 BdrvRequestFlags write_flags
)
2388 r
= blk_check_byte_request(blk_in
, off_in
, bytes
);
2392 r
= blk_check_byte_request(blk_out
, off_out
, bytes
);
2396 return bdrv_co_copy_range(blk_in
->root
, off_in
,
2397 blk_out
->root
, off_out
,
2398 bytes
, read_flags
, write_flags
);
2401 const BdrvChild
*blk_root(BlockBackend
*blk
)