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/coroutines.h"
18 #include "block/throttle-groups.h"
19 #include "hw/qdev-core.h"
20 #include "sysemu/blockdev.h"
21 #include "sysemu/runstate.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 /* If the BDS tree is removed, some of its options are stored here (which
60 * can be used to restore those options in the new BDS on insert) */
61 BlockBackendRootState root_state
;
63 bool enable_write_cache
;
65 /* I/O stats (display with "info blockstats"). */
68 BlockdevOnError on_read_error
, on_write_error
;
69 bool iostatus_enabled
;
70 BlockDeviceIoStatus iostatus
;
76 bool allow_aio_context_change
;
77 bool allow_write_beyond_eof
;
79 /* Protected by BQL */
80 NotifierList remove_bs_notifiers
, insert_bs_notifiers
;
81 QLIST_HEAD(, BlockBackendAioNotifier
) aio_notifiers
;
83 int quiesce_counter
; /* atomic: written under BQL, read by other threads */
84 CoQueue queued_requests
;
85 bool disable_request_queuing
; /* atomic */
87 VMChangeStateEntry
*vmsh
;
88 bool force_allow_inactivate
;
90 /* Number of in-flight aio requests. BlockDriverState also counts
91 * in-flight requests but aio requests can exist even when blk->root is
92 * NULL, so we cannot rely on its counter for that case.
93 * Accessed with atomic ops.
95 unsigned int in_flight
;
98 typedef struct BlockBackendAIOCB
{
104 static const AIOCBInfo block_backend_aiocb_info
= {
105 .get_aio_context
= blk_aiocb_get_aio_context
,
106 .aiocb_size
= sizeof(BlockBackendAIOCB
),
109 static void drive_info_del(DriveInfo
*dinfo
);
110 static BlockBackend
*bdrv_first_blk(BlockDriverState
*bs
);
112 /* All BlockBackends. Protected by BQL. */
113 static QTAILQ_HEAD(, BlockBackend
) block_backends
=
114 QTAILQ_HEAD_INITIALIZER(block_backends
);
117 * All BlockBackends referenced by the monitor and which are iterated through by
118 * blk_next(). Protected by BQL.
120 static QTAILQ_HEAD(, BlockBackend
) monitor_block_backends
=
121 QTAILQ_HEAD_INITIALIZER(monitor_block_backends
);
123 static void blk_root_inherit_options(BdrvChildRole role
, bool parent_is_format
,
124 int *child_flags
, QDict
*child_options
,
125 int parent_flags
, QDict
*parent_options
)
127 /* We're not supposed to call this function for root nodes */
130 static void blk_root_drained_begin(BdrvChild
*child
);
131 static bool blk_root_drained_poll(BdrvChild
*child
);
132 static void blk_root_drained_end(BdrvChild
*child
);
134 static void blk_root_change_media(BdrvChild
*child
, bool load
);
135 static void blk_root_resize(BdrvChild
*child
);
137 static bool blk_root_change_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
138 GHashTable
*visited
, Transaction
*tran
,
141 static char *blk_root_get_parent_desc(BdrvChild
*child
)
143 BlockBackend
*blk
= child
->opaque
;
144 g_autofree
char *dev_id
= NULL
;
147 return g_strdup_printf("block device '%s'", blk
->name
);
150 dev_id
= blk_get_attached_dev_id(blk
);
152 return g_strdup_printf("block device '%s'", dev_id
);
154 /* TODO Callback into the BB owner for something more detailed */
155 return g_strdup("an unnamed block device");
159 static const char *blk_root_get_name(BdrvChild
*child
)
161 return blk_name(child
->opaque
);
164 static void blk_vm_state_changed(void *opaque
, bool running
, RunState state
)
166 Error
*local_err
= NULL
;
167 BlockBackend
*blk
= opaque
;
169 if (state
== RUN_STATE_INMIGRATE
) {
173 qemu_del_vm_change_state_handler(blk
->vmsh
);
175 blk_set_perm(blk
, blk
->perm
, blk
->shared_perm
, &local_err
);
177 error_report_err(local_err
);
182 * Notifies the user of the BlockBackend that migration has completed. qdev
183 * devices can tighten their permissions in response (specifically revoke
184 * shared write permissions that we needed for storage migration).
186 * If an error is returned, the VM cannot be allowed to be resumed.
188 static void blk_root_activate(BdrvChild
*child
, Error
**errp
)
190 BlockBackend
*blk
= child
->opaque
;
191 Error
*local_err
= NULL
;
192 uint64_t saved_shared_perm
;
194 if (!blk
->disable_perm
) {
198 blk
->disable_perm
= false;
201 * blk->shared_perm contains the permissions we want to share once
202 * migration is really completely done. For now, we need to share
203 * all; but we also need to retain blk->shared_perm, which is
204 * overwritten by a successful blk_set_perm() call. Save it and
207 saved_shared_perm
= blk
->shared_perm
;
209 blk_set_perm(blk
, blk
->perm
, BLK_PERM_ALL
, &local_err
);
211 error_propagate(errp
, local_err
);
212 blk
->disable_perm
= true;
215 blk
->shared_perm
= saved_shared_perm
;
217 if (runstate_check(RUN_STATE_INMIGRATE
)) {
218 /* Activation can happen when migration process is still active, for
219 * example when nbd_server_add is called during non-shared storage
220 * migration. Defer the shared_perm update to migration completion. */
222 blk
->vmsh
= qemu_add_vm_change_state_handler(blk_vm_state_changed
,
228 blk_set_perm(blk
, blk
->perm
, blk
->shared_perm
, &local_err
);
230 error_propagate(errp
, local_err
);
231 blk
->disable_perm
= true;
236 void blk_set_force_allow_inactivate(BlockBackend
*blk
)
239 blk
->force_allow_inactivate
= true;
242 static bool blk_can_inactivate(BlockBackend
*blk
)
244 /* If it is a guest device, inactivate is ok. */
245 if (blk
->dev
|| blk_name(blk
)[0]) {
249 /* Inactivating means no more writes to the image can be done,
250 * even if those writes would be changes invisible to the
251 * guest. For block job BBs that satisfy this, we can just allow
252 * it. This is the case for mirror job source, which is required
253 * by libvirt non-shared block migration. */
254 if (!(blk
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
))) {
258 return blk
->force_allow_inactivate
;
261 static int blk_root_inactivate(BdrvChild
*child
)
263 BlockBackend
*blk
= child
->opaque
;
265 if (blk
->disable_perm
) {
269 if (!blk_can_inactivate(blk
)) {
273 blk
->disable_perm
= true;
275 bdrv_child_try_set_perm(blk
->root
, 0, BLK_PERM_ALL
, &error_abort
);
281 static void blk_root_attach(BdrvChild
*child
)
283 BlockBackend
*blk
= child
->opaque
;
284 BlockBackendAioNotifier
*notifier
;
286 trace_blk_root_attach(child
, blk
, child
->bs
);
288 QLIST_FOREACH(notifier
, &blk
->aio_notifiers
, list
) {
289 bdrv_add_aio_context_notifier(child
->bs
,
290 notifier
->attached_aio_context
,
291 notifier
->detach_aio_context
,
296 static void blk_root_detach(BdrvChild
*child
)
298 BlockBackend
*blk
= child
->opaque
;
299 BlockBackendAioNotifier
*notifier
;
301 trace_blk_root_detach(child
, blk
, child
->bs
);
303 QLIST_FOREACH(notifier
, &blk
->aio_notifiers
, list
) {
304 bdrv_remove_aio_context_notifier(child
->bs
,
305 notifier
->attached_aio_context
,
306 notifier
->detach_aio_context
,
311 static AioContext
*blk_root_get_parent_aio_context(BdrvChild
*c
)
313 BlockBackend
*blk
= c
->opaque
;
316 return blk_get_aio_context(blk
);
319 static const BdrvChildClass child_root
= {
320 .inherit_options
= blk_root_inherit_options
,
322 .change_media
= blk_root_change_media
,
323 .resize
= blk_root_resize
,
324 .get_name
= blk_root_get_name
,
325 .get_parent_desc
= blk_root_get_parent_desc
,
327 .drained_begin
= blk_root_drained_begin
,
328 .drained_poll
= blk_root_drained_poll
,
329 .drained_end
= blk_root_drained_end
,
331 .activate
= blk_root_activate
,
332 .inactivate
= blk_root_inactivate
,
334 .attach
= blk_root_attach
,
335 .detach
= blk_root_detach
,
337 .change_aio_ctx
= blk_root_change_aio_ctx
,
339 .get_parent_aio_context
= blk_root_get_parent_aio_context
,
343 * Create a new BlockBackend with a reference count of one.
345 * @perm is a bitmasks of BLK_PERM_* constants which describes the permissions
346 * to request for a block driver node that is attached to this BlockBackend.
347 * @shared_perm is a bitmask which describes which permissions may be granted
348 * to other users of the attached node.
349 * Both sets of permissions can be changed later using blk_set_perm().
351 * Return the new BlockBackend on success, null on failure.
353 BlockBackend
*blk_new(AioContext
*ctx
, uint64_t perm
, uint64_t shared_perm
)
359 blk
= g_new0(BlockBackend
, 1);
363 blk
->shared_perm
= shared_perm
;
364 blk_set_enable_write_cache(blk
, true);
366 blk
->on_read_error
= BLOCKDEV_ON_ERROR_REPORT
;
367 blk
->on_write_error
= BLOCKDEV_ON_ERROR_ENOSPC
;
369 block_acct_init(&blk
->stats
);
371 qemu_co_queue_init(&blk
->queued_requests
);
372 notifier_list_init(&blk
->remove_bs_notifiers
);
373 notifier_list_init(&blk
->insert_bs_notifiers
);
374 QLIST_INIT(&blk
->aio_notifiers
);
376 QTAILQ_INSERT_TAIL(&block_backends
, blk
, link
);
381 * Create a new BlockBackend connected to an existing BlockDriverState.
383 * @perm is a bitmasks of BLK_PERM_* constants which describes the
384 * permissions to request for @bs that is attached to this
385 * BlockBackend. @shared_perm is a bitmask which describes which
386 * permissions may be granted to other users of the attached node.
387 * Both sets of permissions can be changed later using blk_set_perm().
389 * Return the new BlockBackend on success, null on failure.
391 BlockBackend
*blk_new_with_bs(BlockDriverState
*bs
, uint64_t perm
,
392 uint64_t shared_perm
, Error
**errp
)
394 BlockBackend
*blk
= blk_new(bdrv_get_aio_context(bs
), perm
, shared_perm
);
398 if (blk_insert_bs(blk
, bs
, errp
) < 0) {
406 * Creates a new BlockBackend, opens a new BlockDriverState, and connects both.
407 * The new BlockBackend is in the main AioContext.
409 * Just as with bdrv_open(), after having called this function the reference to
410 * @options belongs to the block layer (even on failure).
412 * TODO: Remove @filename and @flags; it should be possible to specify a whole
413 * BDS tree just by specifying the @options QDict (or @reference,
414 * alternatively). At the time of adding this function, this is not possible,
415 * though, so callers of this function have to be able to specify @filename and
418 BlockBackend
*blk_new_open(const char *filename
, const char *reference
,
419 QDict
*options
, int flags
, Error
**errp
)
422 BlockDriverState
*bs
;
424 uint64_t shared
= BLK_PERM_ALL
;
429 * blk_new_open() is mainly used in .bdrv_create implementations and the
430 * tools where sharing isn't a major concern because the BDS stays private
431 * and the file is generally not supposed to be used by a second process,
432 * so we just request permission according to the flags.
434 * The exceptions are xen_disk and blockdev_init(); in these cases, the
435 * caller of blk_new_open() doesn't make use of the permissions, but they
436 * shouldn't hurt either. We can still share everything here because the
437 * guest devices will add their own blockers if they can't share.
439 if ((flags
& BDRV_O_NO_IO
) == 0) {
440 perm
|= BLK_PERM_CONSISTENT_READ
;
441 if (flags
& BDRV_O_RDWR
) {
442 perm
|= BLK_PERM_WRITE
;
445 if (flags
& BDRV_O_RESIZE
) {
446 perm
|= BLK_PERM_RESIZE
;
448 if (flags
& BDRV_O_NO_SHARE
) {
449 shared
= BLK_PERM_CONSISTENT_READ
| BLK_PERM_WRITE_UNCHANGED
;
452 blk
= blk_new(qemu_get_aio_context(), perm
, shared
);
453 bs
= bdrv_open(filename
, reference
, options
, flags
, errp
);
459 blk
->root
= bdrv_root_attach_child(bs
, "root", &child_root
,
460 BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
,
461 perm
, shared
, blk
, errp
);
470 static void blk_delete(BlockBackend
*blk
)
472 assert(!blk
->refcnt
);
475 if (blk
->public.throttle_group_member
.throttle_state
) {
476 blk_io_limits_disable(blk
);
482 qemu_del_vm_change_state_handler(blk
->vmsh
);
485 assert(QLIST_EMPTY(&blk
->remove_bs_notifiers
.notifiers
));
486 assert(QLIST_EMPTY(&blk
->insert_bs_notifiers
.notifiers
));
487 assert(QLIST_EMPTY(&blk
->aio_notifiers
));
488 QTAILQ_REMOVE(&block_backends
, blk
, link
);
489 drive_info_del(blk
->legacy_dinfo
);
490 block_acct_cleanup(&blk
->stats
);
494 static void drive_info_del(DriveInfo
*dinfo
)
499 qemu_opts_del(dinfo
->opts
);
503 int blk_get_refcnt(BlockBackend
*blk
)
506 return blk
? blk
->refcnt
: 0;
510 * Increment @blk's reference count.
511 * @blk must not be null.
513 void blk_ref(BlockBackend
*blk
)
515 assert(blk
->refcnt
> 0);
521 * Decrement @blk's reference count.
522 * If this drops it to zero, destroy @blk.
523 * For convenience, do nothing if @blk is null.
525 void blk_unref(BlockBackend
*blk
)
529 assert(blk
->refcnt
> 0);
530 if (blk
->refcnt
> 1) {
534 /* blk_drain() cannot resurrect blk, nobody held a reference */
535 assert(blk
->refcnt
== 1);
543 * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
544 * ones which are hidden (i.e. are not referenced by the monitor).
546 BlockBackend
*blk_all_next(BlockBackend
*blk
)
549 return blk
? QTAILQ_NEXT(blk
, link
)
550 : QTAILQ_FIRST(&block_backends
);
553 void blk_remove_all_bs(void)
555 BlockBackend
*blk
= NULL
;
559 while ((blk
= blk_all_next(blk
)) != NULL
) {
560 AioContext
*ctx
= blk_get_aio_context(blk
);
562 aio_context_acquire(ctx
);
566 aio_context_release(ctx
);
571 * Return the monitor-owned BlockBackend after @blk.
572 * If @blk is null, return the first one.
573 * Else, return @blk's next sibling, which may be null.
575 * To iterate over all BlockBackends, do
576 * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
580 BlockBackend
*blk_next(BlockBackend
*blk
)
583 return blk
? QTAILQ_NEXT(blk
, monitor_link
)
584 : QTAILQ_FIRST(&monitor_block_backends
);
587 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
588 * the monitor or attached to a BlockBackend */
589 BlockDriverState
*bdrv_next(BdrvNextIterator
*it
)
591 BlockDriverState
*bs
, *old_bs
;
593 /* Must be called from the main loop */
594 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
596 /* First, return all root nodes of BlockBackends. In order to avoid
597 * returning a BDS twice when multiple BBs refer to it, we only return it
598 * if the BB is the first one in the parent list of the BDS. */
599 if (it
->phase
== BDRV_NEXT_BACKEND_ROOTS
) {
600 BlockBackend
*old_blk
= it
->blk
;
602 old_bs
= old_blk
? blk_bs(old_blk
) : NULL
;
605 it
->blk
= blk_all_next(it
->blk
);
606 bs
= it
->blk
? blk_bs(it
->blk
) : NULL
;
607 } while (it
->blk
&& (bs
== NULL
|| bdrv_first_blk(bs
) != it
->blk
));
619 it
->phase
= BDRV_NEXT_MONITOR_OWNED
;
624 /* Then return the monitor-owned BDSes without a BB attached. Ignore all
625 * BDSes that are attached to a BlockBackend here; they have been handled
626 * by the above block already */
628 it
->bs
= bdrv_next_monitor_owned(it
->bs
);
630 } while (bs
&& bdrv_has_blk(bs
));
640 static void bdrv_next_reset(BdrvNextIterator
*it
)
642 *it
= (BdrvNextIterator
) {
643 .phase
= BDRV_NEXT_BACKEND_ROOTS
,
647 BlockDriverState
*bdrv_first(BdrvNextIterator
*it
)
651 return bdrv_next(it
);
654 /* Must be called when aborting a bdrv_next() iteration before
655 * bdrv_next() returns NULL */
656 void bdrv_next_cleanup(BdrvNextIterator
*it
)
658 /* Must be called from the main loop */
659 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
661 if (it
->phase
== BDRV_NEXT_BACKEND_ROOTS
) {
663 bdrv_unref(blk_bs(it
->blk
));
674 * Add a BlockBackend into the list of backends referenced by the monitor, with
675 * the given @name acting as the handle for the monitor.
676 * Strictly for use by blockdev.c.
678 * @name must not be null or empty.
680 * Returns true on success and false on failure. In the latter case, an Error
681 * object is returned through @errp.
683 bool monitor_add_blk(BlockBackend
*blk
, const char *name
, Error
**errp
)
686 assert(name
&& name
[0]);
689 if (!id_wellformed(name
)) {
690 error_setg(errp
, "Invalid device name");
693 if (blk_by_name(name
)) {
694 error_setg(errp
, "Device with id '%s' already exists", name
);
697 if (bdrv_find_node(name
)) {
699 "Device name '%s' conflicts with an existing node name",
704 blk
->name
= g_strdup(name
);
705 QTAILQ_INSERT_TAIL(&monitor_block_backends
, blk
, monitor_link
);
710 * Remove a BlockBackend from the list of backends referenced by the monitor.
711 * Strictly for use by blockdev.c.
713 void monitor_remove_blk(BlockBackend
*blk
)
721 QTAILQ_REMOVE(&monitor_block_backends
, blk
, monitor_link
);
727 * Return @blk's name, a non-null string.
728 * Returns an empty string iff @blk is not referenced by the monitor.
730 const char *blk_name(const BlockBackend
*blk
)
733 return blk
->name
?: "";
737 * Return the BlockBackend with name @name if it exists, else null.
738 * @name must not be null.
740 BlockBackend
*blk_by_name(const char *name
)
742 BlockBackend
*blk
= NULL
;
746 while ((blk
= blk_next(blk
)) != NULL
) {
747 if (!strcmp(name
, blk
->name
)) {
755 * Return the BlockDriverState attached to @blk if any, else null.
757 BlockDriverState
*blk_bs(BlockBackend
*blk
)
760 return blk
->root
? blk
->root
->bs
: NULL
;
763 static BlockBackend
*bdrv_first_blk(BlockDriverState
*bs
)
769 QLIST_FOREACH(child
, &bs
->parents
, next_parent
) {
770 if (child
->klass
== &child_root
) {
771 return child
->opaque
;
779 * Returns true if @bs has an associated BlockBackend.
781 bool bdrv_has_blk(BlockDriverState
*bs
)
784 return bdrv_first_blk(bs
) != NULL
;
788 * Returns true if @bs has only BlockBackends as parents.
790 bool bdrv_is_root_node(BlockDriverState
*bs
)
795 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
796 if (c
->klass
!= &child_root
) {
805 * Return @blk's DriveInfo if any, else null.
807 DriveInfo
*blk_legacy_dinfo(BlockBackend
*blk
)
810 return blk
->legacy_dinfo
;
814 * Set @blk's DriveInfo to @dinfo, and return it.
815 * @blk must not have a DriveInfo set already.
816 * No other BlockBackend may have the same DriveInfo set.
818 DriveInfo
*blk_set_legacy_dinfo(BlockBackend
*blk
, DriveInfo
*dinfo
)
820 assert(!blk
->legacy_dinfo
);
822 return blk
->legacy_dinfo
= dinfo
;
826 * Return the BlockBackend with DriveInfo @dinfo.
829 BlockBackend
*blk_by_legacy_dinfo(DriveInfo
*dinfo
)
831 BlockBackend
*blk
= NULL
;
834 while ((blk
= blk_next(blk
)) != NULL
) {
835 if (blk
->legacy_dinfo
== dinfo
) {
843 * Returns a pointer to the publicly accessible fields of @blk.
845 BlockBackendPublic
*blk_get_public(BlockBackend
*blk
)
852 * Returns a BlockBackend given the associated @public fields.
854 BlockBackend
*blk_by_public(BlockBackendPublic
*public)
857 return container_of(public, BlockBackend
, public);
861 * Disassociates the currently associated BlockDriverState from @blk.
863 void blk_remove_bs(BlockBackend
*blk
)
865 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
870 notifier_list_notify(&blk
->remove_bs_notifiers
, blk
);
871 if (tgm
->throttle_state
) {
872 BlockDriverState
*bs
= blk_bs(blk
);
875 * Take a ref in case blk_bs() changes across bdrv_drained_begin(), for
876 * example, if a temporary filter node is removed by a blockjob.
879 bdrv_drained_begin(bs
);
880 throttle_group_detach_aio_context(tgm
);
881 throttle_group_attach_aio_context(tgm
, qemu_get_aio_context());
882 bdrv_drained_end(bs
);
886 blk_update_root_state(blk
);
888 /* bdrv_root_unref_child() will cause blk->root to become stale and may
889 * switch to a completion coroutine later on. Let's drain all I/O here
890 * to avoid that and a potential QEMU crash.
895 bdrv_root_unref_child(root
);
899 * Associates a new BlockDriverState with @blk.
901 int blk_insert_bs(BlockBackend
*blk
, BlockDriverState
*bs
, Error
**errp
)
903 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
906 blk
->root
= bdrv_root_attach_child(bs
, "root", &child_root
,
907 BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
,
908 blk
->perm
, blk
->shared_perm
,
910 if (blk
->root
== NULL
) {
914 notifier_list_notify(&blk
->insert_bs_notifiers
, blk
);
915 if (tgm
->throttle_state
) {
916 throttle_group_detach_aio_context(tgm
);
917 throttle_group_attach_aio_context(tgm
, bdrv_get_aio_context(bs
));
924 * Change BlockDriverState associated with @blk.
926 int blk_replace_bs(BlockBackend
*blk
, BlockDriverState
*new_bs
, Error
**errp
)
929 return bdrv_replace_child_bs(blk
->root
, new_bs
, errp
);
933 * Sets the permission bitmasks that the user of the BlockBackend needs.
935 int blk_set_perm(BlockBackend
*blk
, uint64_t perm
, uint64_t shared_perm
,
941 if (blk
->root
&& !blk
->disable_perm
) {
942 ret
= bdrv_child_try_set_perm(blk
->root
, perm
, shared_perm
, errp
);
949 blk
->shared_perm
= shared_perm
;
954 void blk_get_perm(BlockBackend
*blk
, uint64_t *perm
, uint64_t *shared_perm
)
958 *shared_perm
= blk
->shared_perm
;
962 * Attach device model @dev to @blk.
963 * Return 0 on success, -EBUSY when a device model is attached already.
965 int blk_attach_dev(BlockBackend
*blk
, DeviceState
*dev
)
972 /* While migration is still incoming, we don't need to apply the
973 * permissions of guest device BlockBackends. We might still have a block
974 * job or NBD server writing to the image for storage migration. */
975 if (runstate_check(RUN_STATE_INMIGRATE
)) {
976 blk
->disable_perm
= true;
981 blk_iostatus_reset(blk
);
987 * Detach device model @dev from @blk.
988 * @dev must be currently attached to @blk.
990 void blk_detach_dev(BlockBackend
*blk
, DeviceState
*dev
)
992 assert(blk
->dev
== dev
);
996 blk
->dev_opaque
= NULL
;
997 blk_set_perm(blk
, 0, BLK_PERM_ALL
, &error_abort
);
1002 * Return the device model attached to @blk if any, else null.
1004 DeviceState
*blk_get_attached_dev(BlockBackend
*blk
)
1006 GLOBAL_STATE_CODE();
1010 /* Return the qdev ID, or if no ID is assigned the QOM path, of the block
1011 * device attached to the BlockBackend. */
1012 char *blk_get_attached_dev_id(BlockBackend
*blk
)
1014 DeviceState
*dev
= blk
->dev
;
1018 return g_strdup("");
1019 } else if (dev
->id
) {
1020 return g_strdup(dev
->id
);
1023 return object_get_canonical_path(OBJECT(dev
)) ?: g_strdup("");
1027 * Return the BlockBackend which has the device model @dev attached if it
1028 * exists, else null.
1030 * @dev must not be null.
1032 BlockBackend
*blk_by_dev(void *dev
)
1034 BlockBackend
*blk
= NULL
;
1036 GLOBAL_STATE_CODE();
1038 assert(dev
!= NULL
);
1039 while ((blk
= blk_all_next(blk
)) != NULL
) {
1040 if (blk
->dev
== dev
) {
1048 * Set @blk's device model callbacks to @ops.
1049 * @opaque is the opaque argument to pass to the callbacks.
1050 * This is for use by device models.
1052 void blk_set_dev_ops(BlockBackend
*blk
, const BlockDevOps
*ops
,
1055 GLOBAL_STATE_CODE();
1057 blk
->dev_opaque
= opaque
;
1059 /* Are we currently quiesced? Should we enforce this right now? */
1060 if (qatomic_read(&blk
->quiesce_counter
) && ops
&& ops
->drained_begin
) {
1061 ops
->drained_begin(opaque
);
1066 * Notify @blk's attached device model of media change.
1068 * If @load is true, notify of media load. This action can fail, meaning that
1069 * the medium cannot be loaded. @errp is set then.
1071 * If @load is false, notify of media eject. This can never fail.
1073 * Also send DEVICE_TRAY_MOVED events as appropriate.
1075 void blk_dev_change_media_cb(BlockBackend
*blk
, bool load
, Error
**errp
)
1077 GLOBAL_STATE_CODE();
1078 if (blk
->dev_ops
&& blk
->dev_ops
->change_media_cb
) {
1079 bool tray_was_open
, tray_is_open
;
1080 Error
*local_err
= NULL
;
1082 tray_was_open
= blk_dev_is_tray_open(blk
);
1083 blk
->dev_ops
->change_media_cb(blk
->dev_opaque
, load
, &local_err
);
1085 assert(load
== true);
1086 error_propagate(errp
, local_err
);
1089 tray_is_open
= blk_dev_is_tray_open(blk
);
1091 if (tray_was_open
!= tray_is_open
) {
1092 char *id
= blk_get_attached_dev_id(blk
);
1093 qapi_event_send_device_tray_moved(blk_name(blk
), id
, tray_is_open
);
1099 static void blk_root_change_media(BdrvChild
*child
, bool load
)
1101 blk_dev_change_media_cb(child
->opaque
, load
, NULL
);
1105 * Does @blk's attached device model have removable media?
1106 * %true if no device model is attached.
1108 bool blk_dev_has_removable_media(BlockBackend
*blk
)
1110 GLOBAL_STATE_CODE();
1111 return !blk
->dev
|| (blk
->dev_ops
&& blk
->dev_ops
->change_media_cb
);
1115 * Does @blk's attached device model have a tray?
1117 bool blk_dev_has_tray(BlockBackend
*blk
)
1120 return blk
->dev_ops
&& blk
->dev_ops
->is_tray_open
;
1124 * Notify @blk's attached device model of a media eject request.
1125 * If @force is true, the medium is about to be yanked out forcefully.
1127 void blk_dev_eject_request(BlockBackend
*blk
, bool force
)
1129 GLOBAL_STATE_CODE();
1130 if (blk
->dev_ops
&& blk
->dev_ops
->eject_request_cb
) {
1131 blk
->dev_ops
->eject_request_cb(blk
->dev_opaque
, force
);
1136 * Does @blk's attached device model have a tray, and is it open?
1138 bool blk_dev_is_tray_open(BlockBackend
*blk
)
1141 if (blk_dev_has_tray(blk
)) {
1142 return blk
->dev_ops
->is_tray_open(blk
->dev_opaque
);
1148 * Does @blk's attached device model have the medium locked?
1149 * %false if the device model has no such lock.
1151 bool blk_dev_is_medium_locked(BlockBackend
*blk
)
1153 GLOBAL_STATE_CODE();
1154 if (blk
->dev_ops
&& blk
->dev_ops
->is_medium_locked
) {
1155 return blk
->dev_ops
->is_medium_locked(blk
->dev_opaque
);
1161 * Notify @blk's attached device model of a backend size change.
1163 static void blk_root_resize(BdrvChild
*child
)
1165 BlockBackend
*blk
= child
->opaque
;
1167 if (blk
->dev_ops
&& blk
->dev_ops
->resize_cb
) {
1168 blk
->dev_ops
->resize_cb(blk
->dev_opaque
);
1172 void blk_iostatus_enable(BlockBackend
*blk
)
1174 GLOBAL_STATE_CODE();
1175 blk
->iostatus_enabled
= true;
1176 blk
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
1179 /* The I/O status is only enabled if the drive explicitly
1180 * enables it _and_ the VM is configured to stop on errors */
1181 bool blk_iostatus_is_enabled(const BlockBackend
*blk
)
1184 return (blk
->iostatus_enabled
&&
1185 (blk
->on_write_error
== BLOCKDEV_ON_ERROR_ENOSPC
||
1186 blk
->on_write_error
== BLOCKDEV_ON_ERROR_STOP
||
1187 blk
->on_read_error
== BLOCKDEV_ON_ERROR_STOP
));
1190 BlockDeviceIoStatus
blk_iostatus(const BlockBackend
*blk
)
1192 GLOBAL_STATE_CODE();
1193 return blk
->iostatus
;
1196 void blk_iostatus_disable(BlockBackend
*blk
)
1198 GLOBAL_STATE_CODE();
1199 blk
->iostatus_enabled
= false;
1202 void blk_iostatus_reset(BlockBackend
*blk
)
1204 GLOBAL_STATE_CODE();
1205 if (blk_iostatus_is_enabled(blk
)) {
1206 blk
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
1210 void blk_iostatus_set_err(BlockBackend
*blk
, int error
)
1213 assert(blk_iostatus_is_enabled(blk
));
1214 if (blk
->iostatus
== BLOCK_DEVICE_IO_STATUS_OK
) {
1215 blk
->iostatus
= error
== ENOSPC
? BLOCK_DEVICE_IO_STATUS_NOSPACE
:
1216 BLOCK_DEVICE_IO_STATUS_FAILED
;
1220 void blk_set_allow_write_beyond_eof(BlockBackend
*blk
, bool allow
)
1223 blk
->allow_write_beyond_eof
= allow
;
1226 void blk_set_allow_aio_context_change(BlockBackend
*blk
, bool allow
)
1229 blk
->allow_aio_context_change
= allow
;
1232 void blk_set_disable_request_queuing(BlockBackend
*blk
, bool disable
)
1235 qatomic_set(&blk
->disable_request_queuing
, disable
);
1238 static int coroutine_fn GRAPH_RDLOCK
1239 blk_check_byte_request(BlockBackend
*blk
, int64_t offset
, int64_t bytes
)
1247 if (!blk_co_is_available(blk
)) {
1255 if (!blk
->allow_write_beyond_eof
) {
1256 len
= bdrv_co_getlength(blk_bs(blk
));
1261 if (offset
> len
|| len
- offset
< bytes
) {
1269 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1270 static void coroutine_fn
blk_wait_while_drained(BlockBackend
*blk
)
1272 assert(blk
->in_flight
> 0);
1274 if (qatomic_read(&blk
->quiesce_counter
) &&
1275 !qatomic_read(&blk
->disable_request_queuing
)) {
1276 blk_dec_in_flight(blk
);
1277 qemu_co_queue_wait(&blk
->queued_requests
, NULL
);
1278 blk_inc_in_flight(blk
);
1282 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1283 static int coroutine_fn
1284 blk_co_do_preadv_part(BlockBackend
*blk
, int64_t offset
, int64_t bytes
,
1285 QEMUIOVector
*qiov
, size_t qiov_offset
,
1286 BdrvRequestFlags flags
)
1289 BlockDriverState
*bs
;
1292 blk_wait_while_drained(blk
);
1293 GRAPH_RDLOCK_GUARD();
1295 /* Call blk_bs() only after waiting, the graph may have changed */
1297 trace_blk_co_preadv(blk
, bs
, offset
, bytes
, flags
);
1299 ret
= blk_check_byte_request(blk
, offset
, bytes
);
1304 bdrv_inc_in_flight(bs
);
1306 /* throttling disk I/O */
1307 if (blk
->public.throttle_group_member
.throttle_state
) {
1308 throttle_group_co_io_limits_intercept(&blk
->public.throttle_group_member
,
1312 ret
= bdrv_co_preadv_part(blk
->root
, offset
, bytes
, qiov
, qiov_offset
,
1314 bdrv_dec_in_flight(bs
);
1318 int coroutine_fn
blk_co_pread(BlockBackend
*blk
, int64_t offset
, int64_t bytes
,
1319 void *buf
, BdrvRequestFlags flags
)
1321 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, buf
, bytes
);
1324 assert(bytes
<= SIZE_MAX
);
1326 return blk_co_preadv(blk
, offset
, bytes
, &qiov
, flags
);
1329 int coroutine_fn
blk_co_preadv(BlockBackend
*blk
, int64_t offset
,
1330 int64_t bytes
, QEMUIOVector
*qiov
,
1331 BdrvRequestFlags flags
)
1336 blk_inc_in_flight(blk
);
1337 ret
= blk_co_do_preadv_part(blk
, offset
, bytes
, qiov
, 0, flags
);
1338 blk_dec_in_flight(blk
);
1343 int coroutine_fn
blk_co_preadv_part(BlockBackend
*blk
, int64_t offset
,
1344 int64_t bytes
, QEMUIOVector
*qiov
,
1345 size_t qiov_offset
, BdrvRequestFlags flags
)
1350 blk_inc_in_flight(blk
);
1351 ret
= blk_co_do_preadv_part(blk
, offset
, bytes
, qiov
, qiov_offset
, flags
);
1352 blk_dec_in_flight(blk
);
1357 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1358 static int coroutine_fn
1359 blk_co_do_pwritev_part(BlockBackend
*blk
, int64_t offset
, int64_t bytes
,
1360 QEMUIOVector
*qiov
, size_t qiov_offset
,
1361 BdrvRequestFlags flags
)
1364 BlockDriverState
*bs
;
1367 blk_wait_while_drained(blk
);
1368 GRAPH_RDLOCK_GUARD();
1370 /* Call blk_bs() only after waiting, the graph may have changed */
1372 trace_blk_co_pwritev(blk
, bs
, offset
, bytes
, flags
);
1374 ret
= blk_check_byte_request(blk
, offset
, bytes
);
1379 bdrv_inc_in_flight(bs
);
1380 /* throttling disk I/O */
1381 if (blk
->public.throttle_group_member
.throttle_state
) {
1382 throttle_group_co_io_limits_intercept(&blk
->public.throttle_group_member
,
1386 if (!blk
->enable_write_cache
) {
1387 flags
|= BDRV_REQ_FUA
;
1390 ret
= bdrv_co_pwritev_part(blk
->root
, offset
, bytes
, qiov
, qiov_offset
,
1392 bdrv_dec_in_flight(bs
);
1396 int coroutine_fn
blk_co_pwritev_part(BlockBackend
*blk
, int64_t offset
,
1398 QEMUIOVector
*qiov
, size_t qiov_offset
,
1399 BdrvRequestFlags flags
)
1404 blk_inc_in_flight(blk
);
1405 ret
= blk_co_do_pwritev_part(blk
, offset
, bytes
, qiov
, qiov_offset
, flags
);
1406 blk_dec_in_flight(blk
);
1411 int coroutine_fn
blk_co_pwrite(BlockBackend
*blk
, int64_t offset
, int64_t bytes
,
1412 const void *buf
, BdrvRequestFlags flags
)
1414 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, buf
, bytes
);
1417 assert(bytes
<= SIZE_MAX
);
1419 return blk_co_pwritev(blk
, offset
, bytes
, &qiov
, flags
);
1422 int coroutine_fn
blk_co_pwritev(BlockBackend
*blk
, int64_t offset
,
1423 int64_t bytes
, QEMUIOVector
*qiov
,
1424 BdrvRequestFlags flags
)
1427 return blk_co_pwritev_part(blk
, offset
, bytes
, qiov
, 0, flags
);
1430 int coroutine_fn
blk_co_block_status_above(BlockBackend
*blk
,
1431 BlockDriverState
*base
,
1432 int64_t offset
, int64_t bytes
,
1433 int64_t *pnum
, int64_t *map
,
1434 BlockDriverState
**file
)
1437 GRAPH_RDLOCK_GUARD();
1438 return bdrv_co_block_status_above(blk_bs(blk
), base
, offset
, bytes
, pnum
,
1442 int coroutine_fn
blk_co_is_allocated_above(BlockBackend
*blk
,
1443 BlockDriverState
*base
,
1444 bool include_base
, int64_t offset
,
1445 int64_t bytes
, int64_t *pnum
)
1448 GRAPH_RDLOCK_GUARD();
1449 return bdrv_co_is_allocated_above(blk_bs(blk
), base
, include_base
, offset
,
1453 typedef struct BlkRwCo
{
1458 BdrvRequestFlags flags
;
1461 int blk_make_zero(BlockBackend
*blk
, BdrvRequestFlags flags
)
1463 GLOBAL_STATE_CODE();
1464 return bdrv_make_zero(blk
->root
, flags
);
1467 void blk_inc_in_flight(BlockBackend
*blk
)
1470 qatomic_inc(&blk
->in_flight
);
1473 void blk_dec_in_flight(BlockBackend
*blk
)
1476 qatomic_dec(&blk
->in_flight
);
1480 static void error_callback_bh(void *opaque
)
1482 struct BlockBackendAIOCB
*acb
= opaque
;
1484 blk_dec_in_flight(acb
->blk
);
1485 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
1486 qemu_aio_unref(acb
);
1489 BlockAIOCB
*blk_abort_aio_request(BlockBackend
*blk
,
1490 BlockCompletionFunc
*cb
,
1491 void *opaque
, int ret
)
1493 struct BlockBackendAIOCB
*acb
;
1496 blk_inc_in_flight(blk
);
1497 acb
= blk_aio_get(&block_backend_aiocb_info
, blk
, cb
, opaque
);
1501 replay_bh_schedule_oneshot_event(blk_get_aio_context(blk
),
1502 error_callback_bh
, acb
);
1503 return &acb
->common
;
1506 typedef struct BlkAioEmAIOCB
{
1513 static AioContext
*blk_aio_em_aiocb_get_aio_context(BlockAIOCB
*acb_
)
1515 BlkAioEmAIOCB
*acb
= container_of(acb_
, BlkAioEmAIOCB
, common
);
1517 return blk_get_aio_context(acb
->rwco
.blk
);
1520 static const AIOCBInfo blk_aio_em_aiocb_info
= {
1521 .aiocb_size
= sizeof(BlkAioEmAIOCB
),
1522 .get_aio_context
= blk_aio_em_aiocb_get_aio_context
,
1525 static void blk_aio_complete(BlkAioEmAIOCB
*acb
)
1527 if (acb
->has_returned
) {
1528 acb
->common
.cb(acb
->common
.opaque
, acb
->rwco
.ret
);
1529 blk_dec_in_flight(acb
->rwco
.blk
);
1530 qemu_aio_unref(acb
);
1534 static void blk_aio_complete_bh(void *opaque
)
1536 BlkAioEmAIOCB
*acb
= opaque
;
1537 assert(acb
->has_returned
);
1538 blk_aio_complete(acb
);
1541 static BlockAIOCB
*blk_aio_prwv(BlockBackend
*blk
, int64_t offset
,
1543 void *iobuf
, CoroutineEntry co_entry
,
1544 BdrvRequestFlags flags
,
1545 BlockCompletionFunc
*cb
, void *opaque
)
1550 blk_inc_in_flight(blk
);
1551 acb
= blk_aio_get(&blk_aio_em_aiocb_info
, blk
, cb
, opaque
);
1552 acb
->rwco
= (BlkRwCo
) {
1560 acb
->has_returned
= false;
1562 co
= qemu_coroutine_create(co_entry
, acb
);
1563 aio_co_enter(blk_get_aio_context(blk
), co
);
1565 acb
->has_returned
= true;
1566 if (acb
->rwco
.ret
!= NOT_DONE
) {
1567 replay_bh_schedule_oneshot_event(blk_get_aio_context(blk
),
1568 blk_aio_complete_bh
, acb
);
1571 return &acb
->common
;
1574 static void coroutine_fn
blk_aio_read_entry(void *opaque
)
1576 BlkAioEmAIOCB
*acb
= opaque
;
1577 BlkRwCo
*rwco
= &acb
->rwco
;
1578 QEMUIOVector
*qiov
= rwco
->iobuf
;
1580 assert(qiov
->size
== acb
->bytes
);
1581 rwco
->ret
= blk_co_do_preadv_part(rwco
->blk
, rwco
->offset
, acb
->bytes
, qiov
,
1583 blk_aio_complete(acb
);
1586 static void coroutine_fn
blk_aio_write_entry(void *opaque
)
1588 BlkAioEmAIOCB
*acb
= opaque
;
1589 BlkRwCo
*rwco
= &acb
->rwco
;
1590 QEMUIOVector
*qiov
= rwco
->iobuf
;
1592 assert(!qiov
|| qiov
->size
== acb
->bytes
);
1593 rwco
->ret
= blk_co_do_pwritev_part(rwco
->blk
, rwco
->offset
, acb
->bytes
,
1594 qiov
, 0, rwco
->flags
);
1595 blk_aio_complete(acb
);
1598 BlockAIOCB
*blk_aio_pwrite_zeroes(BlockBackend
*blk
, int64_t offset
,
1599 int64_t bytes
, BdrvRequestFlags flags
,
1600 BlockCompletionFunc
*cb
, void *opaque
)
1603 return blk_aio_prwv(blk
, offset
, bytes
, NULL
, blk_aio_write_entry
,
1604 flags
| BDRV_REQ_ZERO_WRITE
, cb
, opaque
);
1607 int64_t coroutine_fn
blk_co_getlength(BlockBackend
*blk
)
1610 GRAPH_RDLOCK_GUARD();
1612 if (!blk_co_is_available(blk
)) {
1616 return bdrv_co_getlength(blk_bs(blk
));
1619 int64_t coroutine_fn
blk_co_nb_sectors(BlockBackend
*blk
)
1621 BlockDriverState
*bs
= blk_bs(blk
);
1624 GRAPH_RDLOCK_GUARD();
1629 return bdrv_co_nb_sectors(bs
);
1634 * This wrapper is written by hand because this function is in the hot I/O path,
1635 * via blk_get_geometry.
1637 int64_t coroutine_mixed_fn
blk_nb_sectors(BlockBackend
*blk
)
1639 BlockDriverState
*bs
= blk_bs(blk
);
1646 return bdrv_nb_sectors(bs
);
1650 /* return 0 as number of sectors if no device present or error */
1651 void coroutine_fn
blk_co_get_geometry(BlockBackend
*blk
,
1652 uint64_t *nb_sectors_ptr
)
1654 int64_t ret
= blk_co_nb_sectors(blk
);
1655 *nb_sectors_ptr
= ret
< 0 ? 0 : ret
;
1659 * This wrapper is written by hand because this function is in the hot I/O path.
1661 void coroutine_mixed_fn
blk_get_geometry(BlockBackend
*blk
,
1662 uint64_t *nb_sectors_ptr
)
1664 int64_t ret
= blk_nb_sectors(blk
);
1665 *nb_sectors_ptr
= ret
< 0 ? 0 : ret
;
1668 BlockAIOCB
*blk_aio_preadv(BlockBackend
*blk
, int64_t offset
,
1669 QEMUIOVector
*qiov
, BdrvRequestFlags flags
,
1670 BlockCompletionFunc
*cb
, void *opaque
)
1673 assert((uint64_t)qiov
->size
<= INT64_MAX
);
1674 return blk_aio_prwv(blk
, offset
, qiov
->size
, qiov
,
1675 blk_aio_read_entry
, flags
, cb
, opaque
);
1678 BlockAIOCB
*blk_aio_pwritev(BlockBackend
*blk
, int64_t offset
,
1679 QEMUIOVector
*qiov
, BdrvRequestFlags flags
,
1680 BlockCompletionFunc
*cb
, void *opaque
)
1683 assert((uint64_t)qiov
->size
<= INT64_MAX
);
1684 return blk_aio_prwv(blk
, offset
, qiov
->size
, qiov
,
1685 blk_aio_write_entry
, flags
, cb
, opaque
);
1688 void blk_aio_cancel(BlockAIOCB
*acb
)
1690 GLOBAL_STATE_CODE();
1691 bdrv_aio_cancel(acb
);
1694 void blk_aio_cancel_async(BlockAIOCB
*acb
)
1697 bdrv_aio_cancel_async(acb
);
1700 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1701 static int coroutine_fn
1702 blk_co_do_ioctl(BlockBackend
*blk
, unsigned long int req
, void *buf
)
1706 blk_wait_while_drained(blk
);
1707 GRAPH_RDLOCK_GUARD();
1709 if (!blk_co_is_available(blk
)) {
1713 return bdrv_co_ioctl(blk_bs(blk
), req
, buf
);
1716 int coroutine_fn
blk_co_ioctl(BlockBackend
*blk
, unsigned long int req
,
1722 blk_inc_in_flight(blk
);
1723 ret
= blk_co_do_ioctl(blk
, req
, buf
);
1724 blk_dec_in_flight(blk
);
1729 static void coroutine_fn
blk_aio_ioctl_entry(void *opaque
)
1731 BlkAioEmAIOCB
*acb
= opaque
;
1732 BlkRwCo
*rwco
= &acb
->rwco
;
1734 rwco
->ret
= blk_co_do_ioctl(rwco
->blk
, rwco
->offset
, rwco
->iobuf
);
1736 blk_aio_complete(acb
);
1739 BlockAIOCB
*blk_aio_ioctl(BlockBackend
*blk
, unsigned long int req
, void *buf
,
1740 BlockCompletionFunc
*cb
, void *opaque
)
1743 return blk_aio_prwv(blk
, req
, 0, buf
, blk_aio_ioctl_entry
, 0, cb
, opaque
);
1746 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1747 static int coroutine_fn
1748 blk_co_do_pdiscard(BlockBackend
*blk
, int64_t offset
, int64_t bytes
)
1753 blk_wait_while_drained(blk
);
1754 GRAPH_RDLOCK_GUARD();
1756 ret
= blk_check_byte_request(blk
, offset
, bytes
);
1761 return bdrv_co_pdiscard(blk
->root
, offset
, bytes
);
1764 static void coroutine_fn
blk_aio_pdiscard_entry(void *opaque
)
1766 BlkAioEmAIOCB
*acb
= opaque
;
1767 BlkRwCo
*rwco
= &acb
->rwco
;
1769 rwco
->ret
= blk_co_do_pdiscard(rwco
->blk
, rwco
->offset
, acb
->bytes
);
1770 blk_aio_complete(acb
);
1773 BlockAIOCB
*blk_aio_pdiscard(BlockBackend
*blk
,
1774 int64_t offset
, int64_t bytes
,
1775 BlockCompletionFunc
*cb
, void *opaque
)
1778 return blk_aio_prwv(blk
, offset
, bytes
, NULL
, blk_aio_pdiscard_entry
, 0,
1782 int coroutine_fn
blk_co_pdiscard(BlockBackend
*blk
, int64_t offset
,
1788 blk_inc_in_flight(blk
);
1789 ret
= blk_co_do_pdiscard(blk
, offset
, bytes
);
1790 blk_dec_in_flight(blk
);
1795 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1796 static int coroutine_fn
blk_co_do_flush(BlockBackend
*blk
)
1799 blk_wait_while_drained(blk
);
1800 GRAPH_RDLOCK_GUARD();
1802 if (!blk_co_is_available(blk
)) {
1806 return bdrv_co_flush(blk_bs(blk
));
1809 static void coroutine_fn
blk_aio_flush_entry(void *opaque
)
1811 BlkAioEmAIOCB
*acb
= opaque
;
1812 BlkRwCo
*rwco
= &acb
->rwco
;
1814 rwco
->ret
= blk_co_do_flush(rwco
->blk
);
1815 blk_aio_complete(acb
);
1818 BlockAIOCB
*blk_aio_flush(BlockBackend
*blk
,
1819 BlockCompletionFunc
*cb
, void *opaque
)
1822 return blk_aio_prwv(blk
, 0, 0, NULL
, blk_aio_flush_entry
, 0, cb
, opaque
);
1825 int coroutine_fn
blk_co_flush(BlockBackend
*blk
)
1830 blk_inc_in_flight(blk
);
1831 ret
= blk_co_do_flush(blk
);
1832 blk_dec_in_flight(blk
);
1837 void blk_drain(BlockBackend
*blk
)
1839 BlockDriverState
*bs
= blk_bs(blk
);
1840 GLOBAL_STATE_CODE();
1844 bdrv_drained_begin(bs
);
1847 /* We may have -ENOMEDIUM completions in flight */
1848 AIO_WAIT_WHILE(blk_get_aio_context(blk
),
1849 qatomic_read(&blk
->in_flight
) > 0);
1852 bdrv_drained_end(bs
);
1857 void blk_drain_all(void)
1859 BlockBackend
*blk
= NULL
;
1861 GLOBAL_STATE_CODE();
1863 bdrv_drain_all_begin();
1865 while ((blk
= blk_all_next(blk
)) != NULL
) {
1866 AioContext
*ctx
= blk_get_aio_context(blk
);
1868 aio_context_acquire(ctx
);
1870 /* We may have -ENOMEDIUM completions in flight */
1871 AIO_WAIT_WHILE(ctx
, qatomic_read(&blk
->in_flight
) > 0);
1873 aio_context_release(ctx
);
1876 bdrv_drain_all_end();
1879 void blk_set_on_error(BlockBackend
*blk
, BlockdevOnError on_read_error
,
1880 BlockdevOnError on_write_error
)
1882 GLOBAL_STATE_CODE();
1883 blk
->on_read_error
= on_read_error
;
1884 blk
->on_write_error
= on_write_error
;
1887 BlockdevOnError
blk_get_on_error(BlockBackend
*blk
, bool is_read
)
1890 return is_read
? blk
->on_read_error
: blk
->on_write_error
;
1893 BlockErrorAction
blk_get_error_action(BlockBackend
*blk
, bool is_read
,
1896 BlockdevOnError on_err
= blk_get_on_error(blk
, is_read
);
1900 case BLOCKDEV_ON_ERROR_ENOSPC
:
1901 return (error
== ENOSPC
) ?
1902 BLOCK_ERROR_ACTION_STOP
: BLOCK_ERROR_ACTION_REPORT
;
1903 case BLOCKDEV_ON_ERROR_STOP
:
1904 return BLOCK_ERROR_ACTION_STOP
;
1905 case BLOCKDEV_ON_ERROR_REPORT
:
1906 return BLOCK_ERROR_ACTION_REPORT
;
1907 case BLOCKDEV_ON_ERROR_IGNORE
:
1908 return BLOCK_ERROR_ACTION_IGNORE
;
1909 case BLOCKDEV_ON_ERROR_AUTO
:
1915 static void send_qmp_error_event(BlockBackend
*blk
,
1916 BlockErrorAction action
,
1917 bool is_read
, int error
)
1919 IoOperationType optype
;
1920 BlockDriverState
*bs
= blk_bs(blk
);
1922 optype
= is_read
? IO_OPERATION_TYPE_READ
: IO_OPERATION_TYPE_WRITE
;
1923 qapi_event_send_block_io_error(blk_name(blk
),
1924 bs
? bdrv_get_node_name(bs
) : NULL
, optype
,
1925 action
, blk_iostatus_is_enabled(blk
),
1926 error
== ENOSPC
, strerror(error
));
1929 /* This is done by device models because, while the block layer knows
1930 * about the error, it does not know whether an operation comes from
1931 * the device or the block layer (from a job, for example).
1933 void blk_error_action(BlockBackend
*blk
, BlockErrorAction action
,
1934 bool is_read
, int error
)
1939 if (action
== BLOCK_ERROR_ACTION_STOP
) {
1940 /* First set the iostatus, so that "info block" returns an iostatus
1941 * that matches the events raised so far (an additional error iostatus
1942 * is fine, but not a lost one).
1944 blk_iostatus_set_err(blk
, error
);
1946 /* Then raise the request to stop the VM and the event.
1947 * qemu_system_vmstop_request_prepare has two effects. First,
1948 * it ensures that the STOP event always comes after the
1949 * BLOCK_IO_ERROR event. Second, it ensures that even if management
1950 * can observe the STOP event and do a "cont" before the STOP
1951 * event is issued, the VM will not stop. In this case, vm_start()
1952 * also ensures that the STOP/RESUME pair of events is emitted.
1954 qemu_system_vmstop_request_prepare();
1955 send_qmp_error_event(blk
, action
, is_read
, error
);
1956 qemu_system_vmstop_request(RUN_STATE_IO_ERROR
);
1958 send_qmp_error_event(blk
, action
, is_read
, error
);
1963 * Returns true if the BlockBackend can support taking write permissions
1964 * (because its root node is not read-only).
1966 bool blk_supports_write_perm(BlockBackend
*blk
)
1968 BlockDriverState
*bs
= blk_bs(blk
);
1969 GLOBAL_STATE_CODE();
1972 return !bdrv_is_read_only(bs
);
1974 return blk
->root_state
.open_flags
& BDRV_O_RDWR
;
1979 * Returns true if the BlockBackend can be written to in its current
1980 * configuration (i.e. if write permission have been requested)
1982 bool blk_is_writable(BlockBackend
*blk
)
1985 return blk
->perm
& BLK_PERM_WRITE
;
1988 bool blk_is_sg(BlockBackend
*blk
)
1990 BlockDriverState
*bs
= blk_bs(blk
);
1991 GLOBAL_STATE_CODE();
1997 return bdrv_is_sg(bs
);
2000 bool blk_enable_write_cache(BlockBackend
*blk
)
2003 return blk
->enable_write_cache
;
2006 void blk_set_enable_write_cache(BlockBackend
*blk
, bool wce
)
2009 blk
->enable_write_cache
= wce
;
2012 void blk_activate(BlockBackend
*blk
, Error
**errp
)
2014 BlockDriverState
*bs
= blk_bs(blk
);
2015 GLOBAL_STATE_CODE();
2018 error_setg(errp
, "Device '%s' has no medium", blk
->name
);
2022 bdrv_activate(bs
, errp
);
2025 bool coroutine_fn
blk_co_is_inserted(BlockBackend
*blk
)
2027 BlockDriverState
*bs
= blk_bs(blk
);
2029 assert_bdrv_graph_readable();
2031 return bs
&& bdrv_co_is_inserted(bs
);
2034 bool coroutine_fn
blk_co_is_available(BlockBackend
*blk
)
2037 return blk_co_is_inserted(blk
) && !blk_dev_is_tray_open(blk
);
2040 void coroutine_fn
blk_co_lock_medium(BlockBackend
*blk
, bool locked
)
2042 BlockDriverState
*bs
= blk_bs(blk
);
2044 GRAPH_RDLOCK_GUARD();
2047 bdrv_co_lock_medium(bs
, locked
);
2051 void coroutine_fn
blk_co_eject(BlockBackend
*blk
, bool eject_flag
)
2053 BlockDriverState
*bs
= blk_bs(blk
);
2056 GRAPH_RDLOCK_GUARD();
2059 bdrv_co_eject(bs
, eject_flag
);
2062 /* Whether or not we ejected on the backend,
2063 * the frontend experienced a tray event. */
2064 id
= blk_get_attached_dev_id(blk
);
2065 qapi_event_send_device_tray_moved(blk_name(blk
), id
,
2070 int blk_get_flags(BlockBackend
*blk
)
2072 BlockDriverState
*bs
= blk_bs(blk
);
2073 GLOBAL_STATE_CODE();
2076 return bdrv_get_flags(bs
);
2078 return blk
->root_state
.open_flags
;
2082 /* Returns the minimum request alignment, in bytes; guaranteed nonzero */
2083 uint32_t blk_get_request_alignment(BlockBackend
*blk
)
2085 BlockDriverState
*bs
= blk_bs(blk
);
2087 return bs
? bs
->bl
.request_alignment
: BDRV_SECTOR_SIZE
;
2090 /* Returns the maximum hardware transfer length, in bytes; guaranteed nonzero */
2091 uint64_t blk_get_max_hw_transfer(BlockBackend
*blk
)
2093 BlockDriverState
*bs
= blk_bs(blk
);
2094 uint64_t max
= INT_MAX
;
2098 max
= MIN_NON_ZERO(max
, bs
->bl
.max_hw_transfer
);
2099 max
= MIN_NON_ZERO(max
, bs
->bl
.max_transfer
);
2101 return ROUND_DOWN(max
, blk_get_request_alignment(blk
));
2104 /* Returns the maximum transfer length, in bytes; guaranteed nonzero */
2105 uint32_t blk_get_max_transfer(BlockBackend
*blk
)
2107 BlockDriverState
*bs
= blk_bs(blk
);
2108 uint32_t max
= INT_MAX
;
2112 max
= MIN_NON_ZERO(max
, bs
->bl
.max_transfer
);
2114 return ROUND_DOWN(max
, blk_get_request_alignment(blk
));
2117 int blk_get_max_hw_iov(BlockBackend
*blk
)
2120 return MIN_NON_ZERO(blk
->root
->bs
->bl
.max_hw_iov
,
2121 blk
->root
->bs
->bl
.max_iov
);
2124 int blk_get_max_iov(BlockBackend
*blk
)
2127 return blk
->root
->bs
->bl
.max_iov
;
2130 void *blk_try_blockalign(BlockBackend
*blk
, size_t size
)
2133 return qemu_try_blockalign(blk
? blk_bs(blk
) : NULL
, size
);
2136 void *blk_blockalign(BlockBackend
*blk
, size_t size
)
2139 return qemu_blockalign(blk
? blk_bs(blk
) : NULL
, size
);
2142 bool blk_op_is_blocked(BlockBackend
*blk
, BlockOpType op
, Error
**errp
)
2144 BlockDriverState
*bs
= blk_bs(blk
);
2145 GLOBAL_STATE_CODE();
2151 return bdrv_op_is_blocked(bs
, op
, errp
);
2154 void blk_op_unblock(BlockBackend
*blk
, BlockOpType op
, Error
*reason
)
2156 BlockDriverState
*bs
= blk_bs(blk
);
2157 GLOBAL_STATE_CODE();
2160 bdrv_op_unblock(bs
, op
, reason
);
2164 void blk_op_block_all(BlockBackend
*blk
, Error
*reason
)
2166 BlockDriverState
*bs
= blk_bs(blk
);
2167 GLOBAL_STATE_CODE();
2170 bdrv_op_block_all(bs
, reason
);
2174 void blk_op_unblock_all(BlockBackend
*blk
, Error
*reason
)
2176 BlockDriverState
*bs
= blk_bs(blk
);
2177 GLOBAL_STATE_CODE();
2180 bdrv_op_unblock_all(bs
, reason
);
2184 AioContext
*blk_get_aio_context(BlockBackend
*blk
)
2186 BlockDriverState
*bs
= blk_bs(blk
);
2190 AioContext
*ctx
= bdrv_get_aio_context(blk_bs(blk
));
2191 assert(ctx
== blk
->ctx
);
2197 static AioContext
*blk_aiocb_get_aio_context(BlockAIOCB
*acb
)
2199 BlockBackendAIOCB
*blk_acb
= DO_UPCAST(BlockBackendAIOCB
, common
, acb
);
2200 return blk_get_aio_context(blk_acb
->blk
);
2203 static int blk_do_set_aio_context(BlockBackend
*blk
, AioContext
*new_context
,
2204 bool update_root_node
, Error
**errp
)
2206 BlockDriverState
*bs
= blk_bs(blk
);
2207 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
2213 if (update_root_node
) {
2215 * update_root_node MUST be false for blk_root_set_aio_ctx_commit(),
2216 * as we are already in the commit function of a transaction.
2218 ret
= bdrv_try_change_aio_context(bs
, new_context
, blk
->root
, errp
);
2225 * Make blk->ctx consistent with the root node before we invoke any
2226 * other operations like drain that might inquire blk->ctx
2228 blk
->ctx
= new_context
;
2229 if (tgm
->throttle_state
) {
2230 bdrv_drained_begin(bs
);
2231 throttle_group_detach_aio_context(tgm
);
2232 throttle_group_attach_aio_context(tgm
, new_context
);
2233 bdrv_drained_end(bs
);
2238 blk
->ctx
= new_context
;
2244 int blk_set_aio_context(BlockBackend
*blk
, AioContext
*new_context
,
2247 GLOBAL_STATE_CODE();
2248 return blk_do_set_aio_context(blk
, new_context
, true, errp
);
2251 typedef struct BdrvStateBlkRootContext
{
2252 AioContext
*new_ctx
;
2254 } BdrvStateBlkRootContext
;
2256 static void blk_root_set_aio_ctx_commit(void *opaque
)
2258 BdrvStateBlkRootContext
*s
= opaque
;
2259 BlockBackend
*blk
= s
->blk
;
2261 blk_do_set_aio_context(blk
, s
->new_ctx
, false, &error_abort
);
2264 static TransactionActionDrv set_blk_root_context
= {
2265 .commit
= blk_root_set_aio_ctx_commit
,
2269 static bool blk_root_change_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
2270 GHashTable
*visited
, Transaction
*tran
,
2273 BlockBackend
*blk
= child
->opaque
;
2274 BdrvStateBlkRootContext
*s
;
2276 if (!blk
->allow_aio_context_change
) {
2278 * Manually created BlockBackends (those with a name) that are not
2279 * attached to anything can change their AioContext without updating
2280 * their user; return an error for others.
2282 if (!blk
->name
|| blk
->dev
) {
2283 /* TODO Add BB name/QOM path */
2284 error_setg(errp
, "Cannot change iothread of active block backend");
2289 s
= g_new(BdrvStateBlkRootContext
, 1);
2290 *s
= (BdrvStateBlkRootContext
) {
2295 tran_add(tran
, &set_blk_root_context
, s
);
2299 void blk_add_aio_context_notifier(BlockBackend
*blk
,
2300 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
2301 void (*detach_aio_context
)(void *opaque
), void *opaque
)
2303 BlockBackendAioNotifier
*notifier
;
2304 BlockDriverState
*bs
= blk_bs(blk
);
2305 GLOBAL_STATE_CODE();
2307 notifier
= g_new(BlockBackendAioNotifier
, 1);
2308 notifier
->attached_aio_context
= attached_aio_context
;
2309 notifier
->detach_aio_context
= detach_aio_context
;
2310 notifier
->opaque
= opaque
;
2311 QLIST_INSERT_HEAD(&blk
->aio_notifiers
, notifier
, list
);
2314 bdrv_add_aio_context_notifier(bs
, attached_aio_context
,
2315 detach_aio_context
, opaque
);
2319 void blk_remove_aio_context_notifier(BlockBackend
*blk
,
2320 void (*attached_aio_context
)(AioContext
*,
2322 void (*detach_aio_context
)(void *),
2325 BlockBackendAioNotifier
*notifier
;
2326 BlockDriverState
*bs
= blk_bs(blk
);
2328 GLOBAL_STATE_CODE();
2331 bdrv_remove_aio_context_notifier(bs
, attached_aio_context
,
2332 detach_aio_context
, opaque
);
2335 QLIST_FOREACH(notifier
, &blk
->aio_notifiers
, list
) {
2336 if (notifier
->attached_aio_context
== attached_aio_context
&&
2337 notifier
->detach_aio_context
== detach_aio_context
&&
2338 notifier
->opaque
== opaque
) {
2339 QLIST_REMOVE(notifier
, list
);
2348 void blk_add_remove_bs_notifier(BlockBackend
*blk
, Notifier
*notify
)
2350 GLOBAL_STATE_CODE();
2351 notifier_list_add(&blk
->remove_bs_notifiers
, notify
);
2354 void blk_add_insert_bs_notifier(BlockBackend
*blk
, Notifier
*notify
)
2356 GLOBAL_STATE_CODE();
2357 notifier_list_add(&blk
->insert_bs_notifiers
, notify
);
2360 void coroutine_fn
blk_co_io_plug(BlockBackend
*blk
)
2362 BlockDriverState
*bs
= blk_bs(blk
);
2364 GRAPH_RDLOCK_GUARD();
2367 bdrv_co_io_plug(bs
);
2371 void coroutine_fn
blk_co_io_unplug(BlockBackend
*blk
)
2373 BlockDriverState
*bs
= blk_bs(blk
);
2375 GRAPH_RDLOCK_GUARD();
2378 bdrv_co_io_unplug(bs
);
2382 BlockAcctStats
*blk_get_stats(BlockBackend
*blk
)
2388 void *blk_aio_get(const AIOCBInfo
*aiocb_info
, BlockBackend
*blk
,
2389 BlockCompletionFunc
*cb
, void *opaque
)
2392 return qemu_aio_get(aiocb_info
, blk_bs(blk
), cb
, opaque
);
2395 int coroutine_fn
blk_co_pwrite_zeroes(BlockBackend
*blk
, int64_t offset
,
2396 int64_t bytes
, BdrvRequestFlags flags
)
2399 return blk_co_pwritev(blk
, offset
, bytes
, NULL
,
2400 flags
| BDRV_REQ_ZERO_WRITE
);
2403 int coroutine_fn
blk_co_pwrite_compressed(BlockBackend
*blk
, int64_t offset
,
2404 int64_t bytes
, const void *buf
)
2406 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, buf
, bytes
);
2408 return blk_co_pwritev_part(blk
, offset
, bytes
, &qiov
, 0,
2409 BDRV_REQ_WRITE_COMPRESSED
);
2412 int coroutine_fn
blk_co_truncate(BlockBackend
*blk
, int64_t offset
, bool exact
,
2413 PreallocMode prealloc
, BdrvRequestFlags flags
,
2417 GRAPH_RDLOCK_GUARD();
2418 if (!blk_co_is_available(blk
)) {
2419 error_setg(errp
, "No medium inserted");
2423 return bdrv_co_truncate(blk
->root
, offset
, exact
, prealloc
, flags
, errp
);
2426 int blk_save_vmstate(BlockBackend
*blk
, const uint8_t *buf
,
2427 int64_t pos
, int size
)
2430 GLOBAL_STATE_CODE();
2432 if (!blk_is_available(blk
)) {
2436 ret
= bdrv_save_vmstate(blk_bs(blk
), buf
, pos
, size
);
2441 if (ret
== size
&& !blk
->enable_write_cache
) {
2442 ret
= bdrv_flush(blk_bs(blk
));
2445 return ret
< 0 ? ret
: size
;
2448 int blk_load_vmstate(BlockBackend
*blk
, uint8_t *buf
, int64_t pos
, int size
)
2450 GLOBAL_STATE_CODE();
2451 if (!blk_is_available(blk
)) {
2455 return bdrv_load_vmstate(blk_bs(blk
), buf
, pos
, size
);
2458 int blk_probe_blocksizes(BlockBackend
*blk
, BlockSizes
*bsz
)
2460 GLOBAL_STATE_CODE();
2461 if (!blk_is_available(blk
)) {
2465 return bdrv_probe_blocksizes(blk_bs(blk
), bsz
);
2468 int blk_probe_geometry(BlockBackend
*blk
, HDGeometry
*geo
)
2470 GLOBAL_STATE_CODE();
2471 if (!blk_is_available(blk
)) {
2475 return bdrv_probe_geometry(blk_bs(blk
), geo
);
2479 * Updates the BlockBackendRootState object with data from the currently
2480 * attached BlockDriverState.
2482 void blk_update_root_state(BlockBackend
*blk
)
2484 GLOBAL_STATE_CODE();
2487 blk
->root_state
.open_flags
= blk
->root
->bs
->open_flags
;
2488 blk
->root_state
.detect_zeroes
= blk
->root
->bs
->detect_zeroes
;
2492 * Returns the detect-zeroes setting to be used for bdrv_open() of a
2493 * BlockDriverState which is supposed to inherit the root state.
2495 bool blk_get_detect_zeroes_from_root_state(BlockBackend
*blk
)
2497 GLOBAL_STATE_CODE();
2498 return blk
->root_state
.detect_zeroes
;
2502 * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
2503 * supposed to inherit the root state.
2505 int blk_get_open_flags_from_root_state(BlockBackend
*blk
)
2507 GLOBAL_STATE_CODE();
2508 return blk
->root_state
.open_flags
;
2511 BlockBackendRootState
*blk_get_root_state(BlockBackend
*blk
)
2513 GLOBAL_STATE_CODE();
2514 return &blk
->root_state
;
2517 int blk_commit_all(void)
2519 BlockBackend
*blk
= NULL
;
2520 GLOBAL_STATE_CODE();
2522 while ((blk
= blk_all_next(blk
)) != NULL
) {
2523 AioContext
*aio_context
= blk_get_aio_context(blk
);
2524 BlockDriverState
*unfiltered_bs
= bdrv_skip_filters(blk_bs(blk
));
2526 aio_context_acquire(aio_context
);
2527 if (blk_is_inserted(blk
) && bdrv_cow_child(unfiltered_bs
)) {
2530 ret
= bdrv_commit(unfiltered_bs
);
2532 aio_context_release(aio_context
);
2536 aio_context_release(aio_context
);
2542 /* throttling disk I/O limits */
2543 void blk_set_io_limits(BlockBackend
*blk
, ThrottleConfig
*cfg
)
2545 GLOBAL_STATE_CODE();
2546 throttle_group_config(&blk
->public.throttle_group_member
, cfg
);
2549 void blk_io_limits_disable(BlockBackend
*blk
)
2551 BlockDriverState
*bs
= blk_bs(blk
);
2552 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
2553 assert(tgm
->throttle_state
);
2554 GLOBAL_STATE_CODE();
2557 bdrv_drained_begin(bs
);
2559 throttle_group_unregister_tgm(tgm
);
2561 bdrv_drained_end(bs
);
2566 /* should be called before blk_set_io_limits if a limit is set */
2567 void blk_io_limits_enable(BlockBackend
*blk
, const char *group
)
2569 assert(!blk
->public.throttle_group_member
.throttle_state
);
2570 GLOBAL_STATE_CODE();
2571 throttle_group_register_tgm(&blk
->public.throttle_group_member
,
2572 group
, blk_get_aio_context(blk
));
2575 void blk_io_limits_update_group(BlockBackend
*blk
, const char *group
)
2577 GLOBAL_STATE_CODE();
2578 /* this BB is not part of any group */
2579 if (!blk
->public.throttle_group_member
.throttle_state
) {
2583 /* this BB is a part of the same group than the one we want */
2584 if (!g_strcmp0(throttle_group_get_name(&blk
->public.throttle_group_member
),
2589 /* need to change the group this bs belong to */
2590 blk_io_limits_disable(blk
);
2591 blk_io_limits_enable(blk
, group
);
2594 static void blk_root_drained_begin(BdrvChild
*child
)
2596 BlockBackend
*blk
= child
->opaque
;
2597 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
2599 if (qatomic_fetch_inc(&blk
->quiesce_counter
) == 0) {
2600 if (blk
->dev_ops
&& blk
->dev_ops
->drained_begin
) {
2601 blk
->dev_ops
->drained_begin(blk
->dev_opaque
);
2605 /* Note that blk->root may not be accessible here yet if we are just
2606 * attaching to a BlockDriverState that is drained. Use child instead. */
2608 if (qatomic_fetch_inc(&tgm
->io_limits_disabled
) == 0) {
2609 throttle_group_restart_tgm(tgm
);
2613 static bool blk_root_drained_poll(BdrvChild
*child
)
2615 BlockBackend
*blk
= child
->opaque
;
2617 assert(qatomic_read(&blk
->quiesce_counter
));
2619 if (blk
->dev_ops
&& blk
->dev_ops
->drained_poll
) {
2620 busy
= blk
->dev_ops
->drained_poll(blk
->dev_opaque
);
2622 return busy
|| !!blk
->in_flight
;
2625 static void blk_root_drained_end(BdrvChild
*child
)
2627 BlockBackend
*blk
= child
->opaque
;
2628 assert(qatomic_read(&blk
->quiesce_counter
));
2630 assert(blk
->public.throttle_group_member
.io_limits_disabled
);
2631 qatomic_dec(&blk
->public.throttle_group_member
.io_limits_disabled
);
2633 if (qatomic_fetch_dec(&blk
->quiesce_counter
) == 1) {
2634 if (blk
->dev_ops
&& blk
->dev_ops
->drained_end
) {
2635 blk
->dev_ops
->drained_end(blk
->dev_opaque
);
2637 while (qemu_co_enter_next(&blk
->queued_requests
, NULL
)) {
2638 /* Resume all queued requests */
2643 bool blk_register_buf(BlockBackend
*blk
, void *host
, size_t size
, Error
**errp
)
2645 BlockDriverState
*bs
= blk_bs(blk
);
2647 GLOBAL_STATE_CODE();
2650 return bdrv_register_buf(bs
, host
, size
, errp
);
2655 void blk_unregister_buf(BlockBackend
*blk
, void *host
, size_t size
)
2657 BlockDriverState
*bs
= blk_bs(blk
);
2659 GLOBAL_STATE_CODE();
2662 bdrv_unregister_buf(bs
, host
, size
);
2666 int coroutine_fn
blk_co_copy_range(BlockBackend
*blk_in
, int64_t off_in
,
2667 BlockBackend
*blk_out
, int64_t off_out
,
2668 int64_t bytes
, BdrvRequestFlags read_flags
,
2669 BdrvRequestFlags write_flags
)
2673 GRAPH_RDLOCK_GUARD();
2675 r
= blk_check_byte_request(blk_in
, off_in
, bytes
);
2679 r
= blk_check_byte_request(blk_out
, off_out
, bytes
);
2684 return bdrv_co_copy_range(blk_in
->root
, off_in
,
2685 blk_out
->root
, off_out
,
2686 bytes
, read_flags
, write_flags
);
2689 const BdrvChild
*blk_root(BlockBackend
*blk
)
2691 GLOBAL_STATE_CODE();
2695 int blk_make_empty(BlockBackend
*blk
, Error
**errp
)
2697 GLOBAL_STATE_CODE();
2698 if (!blk_is_available(blk
)) {
2699 error_setg(errp
, "No medium inserted");
2703 return bdrv_make_empty(blk
->root
, errp
);