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 QemuMutex queued_requests_lock
; /* protects queued_requests */
85 CoQueue queued_requests
;
86 bool disable_request_queuing
; /* atomic */
88 VMChangeStateEntry
*vmsh
;
89 bool force_allow_inactivate
;
91 /* Number of in-flight aio requests. BlockDriverState also counts
92 * in-flight requests but aio requests can exist even when blk->root is
93 * NULL, so we cannot rely on its counter for that case.
94 * Accessed with atomic ops.
96 unsigned int in_flight
;
99 typedef struct BlockBackendAIOCB
{
105 static const AIOCBInfo block_backend_aiocb_info
= {
106 .get_aio_context
= blk_aiocb_get_aio_context
,
107 .aiocb_size
= sizeof(BlockBackendAIOCB
),
110 static void drive_info_del(DriveInfo
*dinfo
);
111 static BlockBackend
*bdrv_first_blk(BlockDriverState
*bs
);
113 /* All BlockBackends. Protected by BQL. */
114 static QTAILQ_HEAD(, BlockBackend
) block_backends
=
115 QTAILQ_HEAD_INITIALIZER(block_backends
);
118 * All BlockBackends referenced by the monitor and which are iterated through by
119 * blk_next(). Protected by BQL.
121 static QTAILQ_HEAD(, BlockBackend
) monitor_block_backends
=
122 QTAILQ_HEAD_INITIALIZER(monitor_block_backends
);
124 static void blk_root_inherit_options(BdrvChildRole role
, bool parent_is_format
,
125 int *child_flags
, QDict
*child_options
,
126 int parent_flags
, QDict
*parent_options
)
128 /* We're not supposed to call this function for root nodes */
131 static void blk_root_drained_begin(BdrvChild
*child
);
132 static bool blk_root_drained_poll(BdrvChild
*child
);
133 static void blk_root_drained_end(BdrvChild
*child
);
135 static void blk_root_change_media(BdrvChild
*child
, bool load
);
136 static void blk_root_resize(BdrvChild
*child
);
138 static bool blk_root_change_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
139 GHashTable
*visited
, Transaction
*tran
,
142 static char *blk_root_get_parent_desc(BdrvChild
*child
)
144 BlockBackend
*blk
= child
->opaque
;
145 g_autofree
char *dev_id
= NULL
;
148 return g_strdup_printf("block device '%s'", blk
->name
);
151 dev_id
= blk_get_attached_dev_id(blk
);
153 return g_strdup_printf("block device '%s'", dev_id
);
155 /* TODO Callback into the BB owner for something more detailed */
156 return g_strdup("an unnamed 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
, bool 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
;
193 uint64_t saved_shared_perm
;
195 if (!blk
->disable_perm
) {
199 blk
->disable_perm
= false;
202 * blk->shared_perm contains the permissions we want to share once
203 * migration is really completely done. For now, we need to share
204 * all; but we also need to retain blk->shared_perm, which is
205 * overwritten by a successful blk_set_perm() call. Save it and
208 saved_shared_perm
= blk
->shared_perm
;
210 blk_set_perm(blk
, blk
->perm
, BLK_PERM_ALL
, &local_err
);
212 error_propagate(errp
, local_err
);
213 blk
->disable_perm
= true;
216 blk
->shared_perm
= saved_shared_perm
;
218 if (runstate_check(RUN_STATE_INMIGRATE
)) {
219 /* Activation can happen when migration process is still active, for
220 * example when nbd_server_add is called during non-shared storage
221 * migration. Defer the shared_perm update to migration completion. */
223 blk
->vmsh
= qemu_add_vm_change_state_handler(blk_vm_state_changed
,
229 blk_set_perm(blk
, blk
->perm
, blk
->shared_perm
, &local_err
);
231 error_propagate(errp
, local_err
);
232 blk
->disable_perm
= true;
237 void blk_set_force_allow_inactivate(BlockBackend
*blk
)
240 blk
->force_allow_inactivate
= true;
243 static bool blk_can_inactivate(BlockBackend
*blk
)
245 /* If it is a guest device, inactivate is ok. */
246 if (blk
->dev
|| blk_name(blk
)[0]) {
250 /* Inactivating means no more writes to the image can be done,
251 * even if those writes would be changes invisible to the
252 * guest. For block job BBs that satisfy this, we can just allow
253 * it. This is the case for mirror job source, which is required
254 * by libvirt non-shared block migration. */
255 if (!(blk
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
))) {
259 return blk
->force_allow_inactivate
;
262 static int blk_root_inactivate(BdrvChild
*child
)
264 BlockBackend
*blk
= child
->opaque
;
266 if (blk
->disable_perm
) {
270 if (!blk_can_inactivate(blk
)) {
274 blk
->disable_perm
= true;
276 bdrv_child_try_set_perm(blk
->root
, 0, BLK_PERM_ALL
, &error_abort
);
282 static void blk_root_attach(BdrvChild
*child
)
284 BlockBackend
*blk
= child
->opaque
;
285 BlockBackendAioNotifier
*notifier
;
287 trace_blk_root_attach(child
, blk
, child
->bs
);
289 QLIST_FOREACH(notifier
, &blk
->aio_notifiers
, list
) {
290 bdrv_add_aio_context_notifier(child
->bs
,
291 notifier
->attached_aio_context
,
292 notifier
->detach_aio_context
,
297 static void blk_root_detach(BdrvChild
*child
)
299 BlockBackend
*blk
= child
->opaque
;
300 BlockBackendAioNotifier
*notifier
;
302 trace_blk_root_detach(child
, blk
, child
->bs
);
304 QLIST_FOREACH(notifier
, &blk
->aio_notifiers
, list
) {
305 bdrv_remove_aio_context_notifier(child
->bs
,
306 notifier
->attached_aio_context
,
307 notifier
->detach_aio_context
,
312 static AioContext
*blk_root_get_parent_aio_context(BdrvChild
*c
)
314 BlockBackend
*blk
= c
->opaque
;
317 return blk_get_aio_context(blk
);
320 static const BdrvChildClass child_root
= {
321 .inherit_options
= blk_root_inherit_options
,
323 .change_media
= blk_root_change_media
,
324 .resize
= blk_root_resize
,
325 .get_name
= blk_root_get_name
,
326 .get_parent_desc
= blk_root_get_parent_desc
,
328 .drained_begin
= blk_root_drained_begin
,
329 .drained_poll
= blk_root_drained_poll
,
330 .drained_end
= blk_root_drained_end
,
332 .activate
= blk_root_activate
,
333 .inactivate
= blk_root_inactivate
,
335 .attach
= blk_root_attach
,
336 .detach
= blk_root_detach
,
338 .change_aio_ctx
= blk_root_change_aio_ctx
,
340 .get_parent_aio_context
= blk_root_get_parent_aio_context
,
344 * Create a new BlockBackend with a reference count of one.
346 * @perm is a bitmasks of BLK_PERM_* constants which describes the permissions
347 * to request for a block driver node that is attached to this BlockBackend.
348 * @shared_perm is a bitmask which describes which permissions may be granted
349 * to other users of the attached node.
350 * Both sets of permissions can be changed later using blk_set_perm().
352 * Return the new BlockBackend on success, null on failure.
354 BlockBackend
*blk_new(AioContext
*ctx
, uint64_t perm
, uint64_t shared_perm
)
360 blk
= g_new0(BlockBackend
, 1);
364 blk
->shared_perm
= shared_perm
;
365 blk_set_enable_write_cache(blk
, true);
367 blk
->on_read_error
= BLOCKDEV_ON_ERROR_REPORT
;
368 blk
->on_write_error
= BLOCKDEV_ON_ERROR_ENOSPC
;
370 block_acct_init(&blk
->stats
);
372 qemu_mutex_init(&blk
->queued_requests_lock
);
373 qemu_co_queue_init(&blk
->queued_requests
);
374 notifier_list_init(&blk
->remove_bs_notifiers
);
375 notifier_list_init(&blk
->insert_bs_notifiers
);
376 QLIST_INIT(&blk
->aio_notifiers
);
378 QTAILQ_INSERT_TAIL(&block_backends
, blk
, link
);
383 * Create a new BlockBackend connected to an existing BlockDriverState.
385 * @perm is a bitmasks of BLK_PERM_* constants which describes the
386 * permissions to request for @bs that is attached to this
387 * BlockBackend. @shared_perm is a bitmask which describes which
388 * permissions may be granted to other users of the attached node.
389 * Both sets of permissions can be changed later using blk_set_perm().
391 * Return the new BlockBackend on success, null on failure.
393 * Callers must hold the AioContext lock of @bs.
395 BlockBackend
*blk_new_with_bs(BlockDriverState
*bs
, uint64_t perm
,
396 uint64_t shared_perm
, Error
**errp
)
398 BlockBackend
*blk
= blk_new(bdrv_get_aio_context(bs
), perm
, shared_perm
);
402 if (blk_insert_bs(blk
, bs
, errp
) < 0) {
410 * Creates a new BlockBackend, opens a new BlockDriverState, and connects both.
411 * By default, the new BlockBackend is in the main AioContext, but if the
412 * parameters connect it with any existing node in a different AioContext, it
413 * may end up there instead.
415 * Just as with bdrv_open(), after having called this function the reference to
416 * @options belongs to the block layer (even on failure).
418 * Called without holding an AioContext lock.
420 * TODO: Remove @filename and @flags; it should be possible to specify a whole
421 * BDS tree just by specifying the @options QDict (or @reference,
422 * alternatively). At the time of adding this function, this is not possible,
423 * though, so callers of this function have to be able to specify @filename and
426 BlockBackend
*blk_new_open(const char *filename
, const char *reference
,
427 QDict
*options
, int flags
, Error
**errp
)
430 BlockDriverState
*bs
;
433 uint64_t shared
= BLK_PERM_ALL
;
438 * blk_new_open() is mainly used in .bdrv_create implementations and the
439 * tools where sharing isn't a major concern because the BDS stays private
440 * and the file is generally not supposed to be used by a second process,
441 * so we just request permission according to the flags.
443 * The exceptions are xen_disk and blockdev_init(); in these cases, the
444 * caller of blk_new_open() doesn't make use of the permissions, but they
445 * shouldn't hurt either. We can still share everything here because the
446 * guest devices will add their own blockers if they can't share.
448 if ((flags
& BDRV_O_NO_IO
) == 0) {
449 perm
|= BLK_PERM_CONSISTENT_READ
;
450 if (flags
& BDRV_O_RDWR
) {
451 perm
|= BLK_PERM_WRITE
;
454 if (flags
& BDRV_O_RESIZE
) {
455 perm
|= BLK_PERM_RESIZE
;
457 if (flags
& BDRV_O_NO_SHARE
) {
458 shared
= BLK_PERM_CONSISTENT_READ
| BLK_PERM_WRITE_UNCHANGED
;
461 aio_context_acquire(qemu_get_aio_context());
462 bs
= bdrv_open(filename
, reference
, options
, flags
, errp
);
463 aio_context_release(qemu_get_aio_context());
468 /* bdrv_open() could have moved bs to a different AioContext */
469 ctx
= bdrv_get_aio_context(bs
);
470 blk
= blk_new(bdrv_get_aio_context(bs
), perm
, shared
);
472 blk
->shared_perm
= shared
;
474 aio_context_acquire(ctx
);
475 blk_insert_bs(blk
, bs
, errp
);
477 aio_context_release(ctx
);
487 static void blk_delete(BlockBackend
*blk
)
489 assert(!blk
->refcnt
);
492 if (blk
->public.throttle_group_member
.throttle_state
) {
493 blk_io_limits_disable(blk
);
499 qemu_del_vm_change_state_handler(blk
->vmsh
);
502 assert(QLIST_EMPTY(&blk
->remove_bs_notifiers
.notifiers
));
503 assert(QLIST_EMPTY(&blk
->insert_bs_notifiers
.notifiers
));
504 assert(QLIST_EMPTY(&blk
->aio_notifiers
));
505 assert(qemu_co_queue_empty(&blk
->queued_requests
));
506 qemu_mutex_destroy(&blk
->queued_requests_lock
);
507 QTAILQ_REMOVE(&block_backends
, blk
, link
);
508 drive_info_del(blk
->legacy_dinfo
);
509 block_acct_cleanup(&blk
->stats
);
513 static void drive_info_del(DriveInfo
*dinfo
)
518 qemu_opts_del(dinfo
->opts
);
522 int blk_get_refcnt(BlockBackend
*blk
)
525 return blk
? blk
->refcnt
: 0;
529 * Increment @blk's reference count.
530 * @blk must not be null.
532 void blk_ref(BlockBackend
*blk
)
534 assert(blk
->refcnt
> 0);
540 * Decrement @blk's reference count.
541 * If this drops it to zero, destroy @blk.
542 * For convenience, do nothing if @blk is null.
544 void blk_unref(BlockBackend
*blk
)
548 assert(blk
->refcnt
> 0);
549 if (blk
->refcnt
> 1) {
553 /* blk_drain() cannot resurrect blk, nobody held a reference */
554 assert(blk
->refcnt
== 1);
562 * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
563 * ones which are hidden (i.e. are not referenced by the monitor).
565 BlockBackend
*blk_all_next(BlockBackend
*blk
)
568 return blk
? QTAILQ_NEXT(blk
, link
)
569 : QTAILQ_FIRST(&block_backends
);
572 void blk_remove_all_bs(void)
574 BlockBackend
*blk
= NULL
;
578 while ((blk
= blk_all_next(blk
)) != NULL
) {
579 AioContext
*ctx
= blk_get_aio_context(blk
);
581 aio_context_acquire(ctx
);
585 aio_context_release(ctx
);
590 * Return the monitor-owned BlockBackend after @blk.
591 * If @blk is null, return the first one.
592 * Else, return @blk's next sibling, which may be null.
594 * To iterate over all BlockBackends, do
595 * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
599 BlockBackend
*blk_next(BlockBackend
*blk
)
602 return blk
? QTAILQ_NEXT(blk
, monitor_link
)
603 : QTAILQ_FIRST(&monitor_block_backends
);
606 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
607 * the monitor or attached to a BlockBackend */
608 BlockDriverState
*bdrv_next(BdrvNextIterator
*it
)
610 BlockDriverState
*bs
, *old_bs
;
612 /* Must be called from the main loop */
613 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
615 /* First, return all root nodes of BlockBackends. In order to avoid
616 * returning a BDS twice when multiple BBs refer to it, we only return it
617 * if the BB is the first one in the parent list of the BDS. */
618 if (it
->phase
== BDRV_NEXT_BACKEND_ROOTS
) {
619 BlockBackend
*old_blk
= it
->blk
;
621 old_bs
= old_blk
? blk_bs(old_blk
) : NULL
;
624 it
->blk
= blk_all_next(it
->blk
);
625 bs
= it
->blk
? blk_bs(it
->blk
) : NULL
;
626 } while (it
->blk
&& (bs
== NULL
|| bdrv_first_blk(bs
) != it
->blk
));
638 it
->phase
= BDRV_NEXT_MONITOR_OWNED
;
643 /* Then return the monitor-owned BDSes without a BB attached. Ignore all
644 * BDSes that are attached to a BlockBackend here; they have been handled
645 * by the above block already */
647 it
->bs
= bdrv_next_monitor_owned(it
->bs
);
649 } while (bs
&& bdrv_has_blk(bs
));
659 static void bdrv_next_reset(BdrvNextIterator
*it
)
661 *it
= (BdrvNextIterator
) {
662 .phase
= BDRV_NEXT_BACKEND_ROOTS
,
666 BlockDriverState
*bdrv_first(BdrvNextIterator
*it
)
670 return bdrv_next(it
);
673 /* Must be called when aborting a bdrv_next() iteration before
674 * bdrv_next() returns NULL */
675 void bdrv_next_cleanup(BdrvNextIterator
*it
)
677 /* Must be called from the main loop */
678 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
680 if (it
->phase
== BDRV_NEXT_BACKEND_ROOTS
) {
682 bdrv_unref(blk_bs(it
->blk
));
693 * Add a BlockBackend into the list of backends referenced by the monitor, with
694 * the given @name acting as the handle for the monitor.
695 * Strictly for use by blockdev.c.
697 * @name must not be null or empty.
699 * Returns true on success and false on failure. In the latter case, an Error
700 * object is returned through @errp.
702 bool monitor_add_blk(BlockBackend
*blk
, const char *name
, Error
**errp
)
705 assert(name
&& name
[0]);
708 if (!id_wellformed(name
)) {
709 error_setg(errp
, "Invalid device name");
712 if (blk_by_name(name
)) {
713 error_setg(errp
, "Device with id '%s' already exists", name
);
716 if (bdrv_find_node(name
)) {
718 "Device name '%s' conflicts with an existing node name",
723 blk
->name
= g_strdup(name
);
724 QTAILQ_INSERT_TAIL(&monitor_block_backends
, blk
, monitor_link
);
729 * Remove a BlockBackend from the list of backends referenced by the monitor.
730 * Strictly for use by blockdev.c.
732 void monitor_remove_blk(BlockBackend
*blk
)
740 QTAILQ_REMOVE(&monitor_block_backends
, blk
, monitor_link
);
746 * Return @blk's name, a non-null string.
747 * Returns an empty string iff @blk is not referenced by the monitor.
749 const char *blk_name(const BlockBackend
*blk
)
752 return blk
->name
?: "";
756 * Return the BlockBackend with name @name if it exists, else null.
757 * @name must not be null.
759 BlockBackend
*blk_by_name(const char *name
)
761 BlockBackend
*blk
= NULL
;
765 while ((blk
= blk_next(blk
)) != NULL
) {
766 if (!strcmp(name
, blk
->name
)) {
774 * Return the BlockDriverState attached to @blk if any, else null.
776 BlockDriverState
*blk_bs(BlockBackend
*blk
)
779 return blk
->root
? blk
->root
->bs
: NULL
;
782 static BlockBackend
*bdrv_first_blk(BlockDriverState
*bs
)
788 QLIST_FOREACH(child
, &bs
->parents
, next_parent
) {
789 if (child
->klass
== &child_root
) {
790 return child
->opaque
;
798 * Returns true if @bs has an associated BlockBackend.
800 bool bdrv_has_blk(BlockDriverState
*bs
)
803 return bdrv_first_blk(bs
) != NULL
;
807 * Returns true if @bs has only BlockBackends as parents.
809 bool bdrv_is_root_node(BlockDriverState
*bs
)
814 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
815 if (c
->klass
!= &child_root
) {
824 * Return @blk's DriveInfo if any, else null.
826 DriveInfo
*blk_legacy_dinfo(BlockBackend
*blk
)
829 return blk
->legacy_dinfo
;
833 * Set @blk's DriveInfo to @dinfo, and return it.
834 * @blk must not have a DriveInfo set already.
835 * No other BlockBackend may have the same DriveInfo set.
837 DriveInfo
*blk_set_legacy_dinfo(BlockBackend
*blk
, DriveInfo
*dinfo
)
839 assert(!blk
->legacy_dinfo
);
841 return blk
->legacy_dinfo
= dinfo
;
845 * Return the BlockBackend with DriveInfo @dinfo.
848 BlockBackend
*blk_by_legacy_dinfo(DriveInfo
*dinfo
)
850 BlockBackend
*blk
= NULL
;
853 while ((blk
= blk_next(blk
)) != NULL
) {
854 if (blk
->legacy_dinfo
== dinfo
) {
862 * Returns a pointer to the publicly accessible fields of @blk.
864 BlockBackendPublic
*blk_get_public(BlockBackend
*blk
)
871 * Returns a BlockBackend given the associated @public fields.
873 BlockBackend
*blk_by_public(BlockBackendPublic
*public)
876 return container_of(public, BlockBackend
, public);
880 * Disassociates the currently associated BlockDriverState from @blk.
882 void blk_remove_bs(BlockBackend
*blk
)
884 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
889 notifier_list_notify(&blk
->remove_bs_notifiers
, blk
);
890 if (tgm
->throttle_state
) {
891 BlockDriverState
*bs
= blk_bs(blk
);
894 * Take a ref in case blk_bs() changes across bdrv_drained_begin(), for
895 * example, if a temporary filter node is removed by a blockjob.
898 bdrv_drained_begin(bs
);
899 throttle_group_detach_aio_context(tgm
);
900 throttle_group_attach_aio_context(tgm
, qemu_get_aio_context());
901 bdrv_drained_end(bs
);
905 blk_update_root_state(blk
);
907 /* bdrv_root_unref_child() will cause blk->root to become stale and may
908 * switch to a completion coroutine later on. Let's drain all I/O here
909 * to avoid that and a potential QEMU crash.
914 bdrv_root_unref_child(root
);
918 * Associates a new BlockDriverState with @blk.
920 * Callers must hold the AioContext lock of @bs.
922 int blk_insert_bs(BlockBackend
*blk
, BlockDriverState
*bs
, Error
**errp
)
924 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
927 blk
->root
= bdrv_root_attach_child(bs
, "root", &child_root
,
928 BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
,
929 blk
->perm
, blk
->shared_perm
,
931 if (blk
->root
== NULL
) {
935 notifier_list_notify(&blk
->insert_bs_notifiers
, blk
);
936 if (tgm
->throttle_state
) {
937 throttle_group_detach_aio_context(tgm
);
938 throttle_group_attach_aio_context(tgm
, bdrv_get_aio_context(bs
));
945 * Change BlockDriverState associated with @blk.
947 int blk_replace_bs(BlockBackend
*blk
, BlockDriverState
*new_bs
, Error
**errp
)
950 return bdrv_replace_child_bs(blk
->root
, new_bs
, errp
);
954 * Sets the permission bitmasks that the user of the BlockBackend needs.
956 int blk_set_perm(BlockBackend
*blk
, uint64_t perm
, uint64_t shared_perm
,
962 if (blk
->root
&& !blk
->disable_perm
) {
963 ret
= bdrv_child_try_set_perm(blk
->root
, perm
, shared_perm
, errp
);
970 blk
->shared_perm
= shared_perm
;
975 void blk_get_perm(BlockBackend
*blk
, uint64_t *perm
, uint64_t *shared_perm
)
979 *shared_perm
= blk
->shared_perm
;
983 * Attach device model @dev to @blk.
984 * Return 0 on success, -EBUSY when a device model is attached already.
986 int blk_attach_dev(BlockBackend
*blk
, DeviceState
*dev
)
993 /* While migration is still incoming, we don't need to apply the
994 * permissions of guest device BlockBackends. We might still have a block
995 * job or NBD server writing to the image for storage migration. */
996 if (runstate_check(RUN_STATE_INMIGRATE
)) {
997 blk
->disable_perm
= true;
1002 blk_iostatus_reset(blk
);
1008 * Detach device model @dev from @blk.
1009 * @dev must be currently attached to @blk.
1011 void blk_detach_dev(BlockBackend
*blk
, DeviceState
*dev
)
1013 assert(blk
->dev
== dev
);
1014 GLOBAL_STATE_CODE();
1016 blk
->dev_ops
= NULL
;
1017 blk
->dev_opaque
= NULL
;
1018 blk_set_perm(blk
, 0, BLK_PERM_ALL
, &error_abort
);
1023 * Return the device model attached to @blk if any, else null.
1025 DeviceState
*blk_get_attached_dev(BlockBackend
*blk
)
1027 GLOBAL_STATE_CODE();
1031 /* Return the qdev ID, or if no ID is assigned the QOM path, of the block
1032 * device attached to the BlockBackend. */
1033 char *blk_get_attached_dev_id(BlockBackend
*blk
)
1035 DeviceState
*dev
= blk
->dev
;
1039 return g_strdup("");
1040 } else if (dev
->id
) {
1041 return g_strdup(dev
->id
);
1044 return object_get_canonical_path(OBJECT(dev
)) ?: g_strdup("");
1048 * Return the BlockBackend which has the device model @dev attached if it
1049 * exists, else null.
1051 * @dev must not be null.
1053 BlockBackend
*blk_by_dev(void *dev
)
1055 BlockBackend
*blk
= NULL
;
1057 GLOBAL_STATE_CODE();
1059 assert(dev
!= NULL
);
1060 while ((blk
= blk_all_next(blk
)) != NULL
) {
1061 if (blk
->dev
== dev
) {
1069 * Set @blk's device model callbacks to @ops.
1070 * @opaque is the opaque argument to pass to the callbacks.
1071 * This is for use by device models.
1073 void blk_set_dev_ops(BlockBackend
*blk
, const BlockDevOps
*ops
,
1076 GLOBAL_STATE_CODE();
1078 blk
->dev_opaque
= opaque
;
1080 /* Are we currently quiesced? Should we enforce this right now? */
1081 if (qatomic_read(&blk
->quiesce_counter
) && ops
&& ops
->drained_begin
) {
1082 ops
->drained_begin(opaque
);
1087 * Notify @blk's attached device model of media change.
1089 * If @load is true, notify of media load. This action can fail, meaning that
1090 * the medium cannot be loaded. @errp is set then.
1092 * If @load is false, notify of media eject. This can never fail.
1094 * Also send DEVICE_TRAY_MOVED events as appropriate.
1096 void blk_dev_change_media_cb(BlockBackend
*blk
, bool load
, Error
**errp
)
1098 GLOBAL_STATE_CODE();
1099 if (blk
->dev_ops
&& blk
->dev_ops
->change_media_cb
) {
1100 bool tray_was_open
, tray_is_open
;
1101 Error
*local_err
= NULL
;
1103 tray_was_open
= blk_dev_is_tray_open(blk
);
1104 blk
->dev_ops
->change_media_cb(blk
->dev_opaque
, load
, &local_err
);
1106 assert(load
== true);
1107 error_propagate(errp
, local_err
);
1110 tray_is_open
= blk_dev_is_tray_open(blk
);
1112 if (tray_was_open
!= tray_is_open
) {
1113 char *id
= blk_get_attached_dev_id(blk
);
1114 qapi_event_send_device_tray_moved(blk_name(blk
), id
, tray_is_open
);
1120 static void blk_root_change_media(BdrvChild
*child
, bool load
)
1122 blk_dev_change_media_cb(child
->opaque
, load
, NULL
);
1126 * Does @blk's attached device model have removable media?
1127 * %true if no device model is attached.
1129 bool blk_dev_has_removable_media(BlockBackend
*blk
)
1131 GLOBAL_STATE_CODE();
1132 return !blk
->dev
|| (blk
->dev_ops
&& blk
->dev_ops
->change_media_cb
);
1136 * Does @blk's attached device model have a tray?
1138 bool blk_dev_has_tray(BlockBackend
*blk
)
1141 return blk
->dev_ops
&& blk
->dev_ops
->is_tray_open
;
1145 * Notify @blk's attached device model of a media eject request.
1146 * If @force is true, the medium is about to be yanked out forcefully.
1148 void blk_dev_eject_request(BlockBackend
*blk
, bool force
)
1150 GLOBAL_STATE_CODE();
1151 if (blk
->dev_ops
&& blk
->dev_ops
->eject_request_cb
) {
1152 blk
->dev_ops
->eject_request_cb(blk
->dev_opaque
, force
);
1157 * Does @blk's attached device model have a tray, and is it open?
1159 bool blk_dev_is_tray_open(BlockBackend
*blk
)
1162 if (blk_dev_has_tray(blk
)) {
1163 return blk
->dev_ops
->is_tray_open(blk
->dev_opaque
);
1169 * Does @blk's attached device model have the medium locked?
1170 * %false if the device model has no such lock.
1172 bool blk_dev_is_medium_locked(BlockBackend
*blk
)
1174 GLOBAL_STATE_CODE();
1175 if (blk
->dev_ops
&& blk
->dev_ops
->is_medium_locked
) {
1176 return blk
->dev_ops
->is_medium_locked(blk
->dev_opaque
);
1182 * Notify @blk's attached device model of a backend size change.
1184 static void blk_root_resize(BdrvChild
*child
)
1186 BlockBackend
*blk
= child
->opaque
;
1188 if (blk
->dev_ops
&& blk
->dev_ops
->resize_cb
) {
1189 blk
->dev_ops
->resize_cb(blk
->dev_opaque
);
1193 void blk_iostatus_enable(BlockBackend
*blk
)
1195 GLOBAL_STATE_CODE();
1196 blk
->iostatus_enabled
= true;
1197 blk
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
1200 /* The I/O status is only enabled if the drive explicitly
1201 * enables it _and_ the VM is configured to stop on errors */
1202 bool blk_iostatus_is_enabled(const BlockBackend
*blk
)
1205 return (blk
->iostatus_enabled
&&
1206 (blk
->on_write_error
== BLOCKDEV_ON_ERROR_ENOSPC
||
1207 blk
->on_write_error
== BLOCKDEV_ON_ERROR_STOP
||
1208 blk
->on_read_error
== BLOCKDEV_ON_ERROR_STOP
));
1211 BlockDeviceIoStatus
blk_iostatus(const BlockBackend
*blk
)
1213 GLOBAL_STATE_CODE();
1214 return blk
->iostatus
;
1217 void blk_iostatus_disable(BlockBackend
*blk
)
1219 GLOBAL_STATE_CODE();
1220 blk
->iostatus_enabled
= false;
1223 void blk_iostatus_reset(BlockBackend
*blk
)
1225 GLOBAL_STATE_CODE();
1226 if (blk_iostatus_is_enabled(blk
)) {
1227 blk
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
1231 void blk_iostatus_set_err(BlockBackend
*blk
, int error
)
1234 assert(blk_iostatus_is_enabled(blk
));
1235 if (blk
->iostatus
== BLOCK_DEVICE_IO_STATUS_OK
) {
1236 blk
->iostatus
= error
== ENOSPC
? BLOCK_DEVICE_IO_STATUS_NOSPACE
:
1237 BLOCK_DEVICE_IO_STATUS_FAILED
;
1241 void blk_set_allow_write_beyond_eof(BlockBackend
*blk
, bool allow
)
1244 blk
->allow_write_beyond_eof
= allow
;
1247 void blk_set_allow_aio_context_change(BlockBackend
*blk
, bool allow
)
1250 blk
->allow_aio_context_change
= allow
;
1253 void blk_set_disable_request_queuing(BlockBackend
*blk
, bool disable
)
1256 qatomic_set(&blk
->disable_request_queuing
, disable
);
1259 static int coroutine_fn GRAPH_RDLOCK
1260 blk_check_byte_request(BlockBackend
*blk
, int64_t offset
, int64_t bytes
)
1268 if (!blk_co_is_available(blk
)) {
1276 if (!blk
->allow_write_beyond_eof
) {
1277 len
= bdrv_co_getlength(blk_bs(blk
));
1282 if (offset
> len
|| len
- offset
< bytes
) {
1290 /* Are we currently in a drained section? */
1291 bool blk_in_drain(BlockBackend
*blk
)
1293 GLOBAL_STATE_CODE(); /* change to IO_OR_GS_CODE(), if necessary */
1294 return qatomic_read(&blk
->quiesce_counter
);
1297 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1298 static void coroutine_fn
blk_wait_while_drained(BlockBackend
*blk
)
1300 assert(blk
->in_flight
> 0);
1302 if (qatomic_read(&blk
->quiesce_counter
) &&
1303 !qatomic_read(&blk
->disable_request_queuing
)) {
1305 * Take lock before decrementing in flight counter so main loop thread
1306 * waits for us to enqueue ourselves before it can leave the drained
1309 qemu_mutex_lock(&blk
->queued_requests_lock
);
1310 blk_dec_in_flight(blk
);
1311 qemu_co_queue_wait(&blk
->queued_requests
, &blk
->queued_requests_lock
);
1312 blk_inc_in_flight(blk
);
1313 qemu_mutex_unlock(&blk
->queued_requests_lock
);
1317 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1318 static int coroutine_fn
1319 blk_co_do_preadv_part(BlockBackend
*blk
, int64_t offset
, int64_t bytes
,
1320 QEMUIOVector
*qiov
, size_t qiov_offset
,
1321 BdrvRequestFlags flags
)
1324 BlockDriverState
*bs
;
1327 blk_wait_while_drained(blk
);
1328 GRAPH_RDLOCK_GUARD();
1330 /* Call blk_bs() only after waiting, the graph may have changed */
1332 trace_blk_co_preadv(blk
, bs
, offset
, bytes
, flags
);
1334 ret
= blk_check_byte_request(blk
, offset
, bytes
);
1339 bdrv_inc_in_flight(bs
);
1341 /* throttling disk I/O */
1342 if (blk
->public.throttle_group_member
.throttle_state
) {
1343 throttle_group_co_io_limits_intercept(&blk
->public.throttle_group_member
,
1347 ret
= bdrv_co_preadv_part(blk
->root
, offset
, bytes
, qiov
, qiov_offset
,
1349 bdrv_dec_in_flight(bs
);
1353 int coroutine_fn
blk_co_pread(BlockBackend
*blk
, int64_t offset
, int64_t bytes
,
1354 void *buf
, BdrvRequestFlags flags
)
1356 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, buf
, bytes
);
1359 assert(bytes
<= SIZE_MAX
);
1361 return blk_co_preadv(blk
, offset
, bytes
, &qiov
, flags
);
1364 int coroutine_fn
blk_co_preadv(BlockBackend
*blk
, int64_t offset
,
1365 int64_t bytes
, QEMUIOVector
*qiov
,
1366 BdrvRequestFlags flags
)
1371 blk_inc_in_flight(blk
);
1372 ret
= blk_co_do_preadv_part(blk
, offset
, bytes
, qiov
, 0, flags
);
1373 blk_dec_in_flight(blk
);
1378 int coroutine_fn
blk_co_preadv_part(BlockBackend
*blk
, int64_t offset
,
1379 int64_t bytes
, QEMUIOVector
*qiov
,
1380 size_t qiov_offset
, BdrvRequestFlags flags
)
1385 blk_inc_in_flight(blk
);
1386 ret
= blk_co_do_preadv_part(blk
, offset
, bytes
, qiov
, qiov_offset
, flags
);
1387 blk_dec_in_flight(blk
);
1392 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1393 static int coroutine_fn
1394 blk_co_do_pwritev_part(BlockBackend
*blk
, int64_t offset
, int64_t bytes
,
1395 QEMUIOVector
*qiov
, size_t qiov_offset
,
1396 BdrvRequestFlags flags
)
1399 BlockDriverState
*bs
;
1402 blk_wait_while_drained(blk
);
1403 GRAPH_RDLOCK_GUARD();
1405 /* Call blk_bs() only after waiting, the graph may have changed */
1407 trace_blk_co_pwritev(blk
, bs
, offset
, bytes
, flags
);
1409 ret
= blk_check_byte_request(blk
, offset
, bytes
);
1414 bdrv_inc_in_flight(bs
);
1415 /* throttling disk I/O */
1416 if (blk
->public.throttle_group_member
.throttle_state
) {
1417 throttle_group_co_io_limits_intercept(&blk
->public.throttle_group_member
,
1421 if (!blk
->enable_write_cache
) {
1422 flags
|= BDRV_REQ_FUA
;
1425 ret
= bdrv_co_pwritev_part(blk
->root
, offset
, bytes
, qiov
, qiov_offset
,
1427 bdrv_dec_in_flight(bs
);
1431 int coroutine_fn
blk_co_pwritev_part(BlockBackend
*blk
, int64_t offset
,
1433 QEMUIOVector
*qiov
, size_t qiov_offset
,
1434 BdrvRequestFlags flags
)
1439 blk_inc_in_flight(blk
);
1440 ret
= blk_co_do_pwritev_part(blk
, offset
, bytes
, qiov
, qiov_offset
, flags
);
1441 blk_dec_in_flight(blk
);
1446 int coroutine_fn
blk_co_pwrite(BlockBackend
*blk
, int64_t offset
, int64_t bytes
,
1447 const void *buf
, BdrvRequestFlags flags
)
1449 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, buf
, bytes
);
1452 assert(bytes
<= SIZE_MAX
);
1454 return blk_co_pwritev(blk
, offset
, bytes
, &qiov
, flags
);
1457 int coroutine_fn
blk_co_pwritev(BlockBackend
*blk
, int64_t offset
,
1458 int64_t bytes
, QEMUIOVector
*qiov
,
1459 BdrvRequestFlags flags
)
1462 return blk_co_pwritev_part(blk
, offset
, bytes
, qiov
, 0, flags
);
1465 int coroutine_fn
blk_co_block_status_above(BlockBackend
*blk
,
1466 BlockDriverState
*base
,
1467 int64_t offset
, int64_t bytes
,
1468 int64_t *pnum
, int64_t *map
,
1469 BlockDriverState
**file
)
1472 GRAPH_RDLOCK_GUARD();
1473 return bdrv_co_block_status_above(blk_bs(blk
), base
, offset
, bytes
, pnum
,
1477 int coroutine_fn
blk_co_is_allocated_above(BlockBackend
*blk
,
1478 BlockDriverState
*base
,
1479 bool include_base
, int64_t offset
,
1480 int64_t bytes
, int64_t *pnum
)
1483 GRAPH_RDLOCK_GUARD();
1484 return bdrv_co_is_allocated_above(blk_bs(blk
), base
, include_base
, offset
,
1488 typedef struct BlkRwCo
{
1493 BdrvRequestFlags flags
;
1496 int blk_make_zero(BlockBackend
*blk
, BdrvRequestFlags flags
)
1498 GLOBAL_STATE_CODE();
1499 return bdrv_make_zero(blk
->root
, flags
);
1502 void blk_inc_in_flight(BlockBackend
*blk
)
1505 qatomic_inc(&blk
->in_flight
);
1508 void blk_dec_in_flight(BlockBackend
*blk
)
1511 qatomic_dec(&blk
->in_flight
);
1515 static void error_callback_bh(void *opaque
)
1517 struct BlockBackendAIOCB
*acb
= opaque
;
1519 blk_dec_in_flight(acb
->blk
);
1520 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
1521 qemu_aio_unref(acb
);
1524 BlockAIOCB
*blk_abort_aio_request(BlockBackend
*blk
,
1525 BlockCompletionFunc
*cb
,
1526 void *opaque
, int ret
)
1528 struct BlockBackendAIOCB
*acb
;
1531 blk_inc_in_flight(blk
);
1532 acb
= blk_aio_get(&block_backend_aiocb_info
, blk
, cb
, opaque
);
1536 replay_bh_schedule_oneshot_event(blk_get_aio_context(blk
),
1537 error_callback_bh
, acb
);
1538 return &acb
->common
;
1541 typedef struct BlkAioEmAIOCB
{
1548 static AioContext
*blk_aio_em_aiocb_get_aio_context(BlockAIOCB
*acb_
)
1550 BlkAioEmAIOCB
*acb
= container_of(acb_
, BlkAioEmAIOCB
, common
);
1552 return blk_get_aio_context(acb
->rwco
.blk
);
1555 static const AIOCBInfo blk_aio_em_aiocb_info
= {
1556 .aiocb_size
= sizeof(BlkAioEmAIOCB
),
1557 .get_aio_context
= blk_aio_em_aiocb_get_aio_context
,
1560 static void blk_aio_complete(BlkAioEmAIOCB
*acb
)
1562 if (acb
->has_returned
) {
1563 acb
->common
.cb(acb
->common
.opaque
, acb
->rwco
.ret
);
1564 blk_dec_in_flight(acb
->rwco
.blk
);
1565 qemu_aio_unref(acb
);
1569 static void blk_aio_complete_bh(void *opaque
)
1571 BlkAioEmAIOCB
*acb
= opaque
;
1572 assert(acb
->has_returned
);
1573 blk_aio_complete(acb
);
1576 static BlockAIOCB
*blk_aio_prwv(BlockBackend
*blk
, int64_t offset
,
1578 void *iobuf
, CoroutineEntry co_entry
,
1579 BdrvRequestFlags flags
,
1580 BlockCompletionFunc
*cb
, void *opaque
)
1585 blk_inc_in_flight(blk
);
1586 acb
= blk_aio_get(&blk_aio_em_aiocb_info
, blk
, cb
, opaque
);
1587 acb
->rwco
= (BlkRwCo
) {
1595 acb
->has_returned
= false;
1597 co
= qemu_coroutine_create(co_entry
, acb
);
1598 aio_co_enter(blk_get_aio_context(blk
), co
);
1600 acb
->has_returned
= true;
1601 if (acb
->rwco
.ret
!= NOT_DONE
) {
1602 replay_bh_schedule_oneshot_event(blk_get_aio_context(blk
),
1603 blk_aio_complete_bh
, acb
);
1606 return &acb
->common
;
1609 static void coroutine_fn
blk_aio_read_entry(void *opaque
)
1611 BlkAioEmAIOCB
*acb
= opaque
;
1612 BlkRwCo
*rwco
= &acb
->rwco
;
1613 QEMUIOVector
*qiov
= rwco
->iobuf
;
1615 assert(qiov
->size
== acb
->bytes
);
1616 rwco
->ret
= blk_co_do_preadv_part(rwco
->blk
, rwco
->offset
, acb
->bytes
, qiov
,
1618 blk_aio_complete(acb
);
1621 static void coroutine_fn
blk_aio_write_entry(void *opaque
)
1623 BlkAioEmAIOCB
*acb
= opaque
;
1624 BlkRwCo
*rwco
= &acb
->rwco
;
1625 QEMUIOVector
*qiov
= rwco
->iobuf
;
1627 assert(!qiov
|| qiov
->size
== acb
->bytes
);
1628 rwco
->ret
= blk_co_do_pwritev_part(rwco
->blk
, rwco
->offset
, acb
->bytes
,
1629 qiov
, 0, rwco
->flags
);
1630 blk_aio_complete(acb
);
1633 BlockAIOCB
*blk_aio_pwrite_zeroes(BlockBackend
*blk
, int64_t offset
,
1634 int64_t bytes
, BdrvRequestFlags flags
,
1635 BlockCompletionFunc
*cb
, void *opaque
)
1638 return blk_aio_prwv(blk
, offset
, bytes
, NULL
, blk_aio_write_entry
,
1639 flags
| BDRV_REQ_ZERO_WRITE
, cb
, opaque
);
1642 int64_t coroutine_fn
blk_co_getlength(BlockBackend
*blk
)
1645 GRAPH_RDLOCK_GUARD();
1647 if (!blk_co_is_available(blk
)) {
1651 return bdrv_co_getlength(blk_bs(blk
));
1654 int64_t coroutine_fn
blk_co_nb_sectors(BlockBackend
*blk
)
1656 BlockDriverState
*bs
= blk_bs(blk
);
1659 GRAPH_RDLOCK_GUARD();
1664 return bdrv_co_nb_sectors(bs
);
1669 * This wrapper is written by hand because this function is in the hot I/O path,
1670 * via blk_get_geometry.
1672 int64_t coroutine_mixed_fn
blk_nb_sectors(BlockBackend
*blk
)
1674 BlockDriverState
*bs
= blk_bs(blk
);
1681 return bdrv_nb_sectors(bs
);
1685 /* return 0 as number of sectors if no device present or error */
1686 void coroutine_fn
blk_co_get_geometry(BlockBackend
*blk
,
1687 uint64_t *nb_sectors_ptr
)
1689 int64_t ret
= blk_co_nb_sectors(blk
);
1690 *nb_sectors_ptr
= ret
< 0 ? 0 : ret
;
1694 * This wrapper is written by hand because this function is in the hot I/O path.
1696 void coroutine_mixed_fn
blk_get_geometry(BlockBackend
*blk
,
1697 uint64_t *nb_sectors_ptr
)
1699 int64_t ret
= blk_nb_sectors(blk
);
1700 *nb_sectors_ptr
= ret
< 0 ? 0 : ret
;
1703 BlockAIOCB
*blk_aio_preadv(BlockBackend
*blk
, int64_t offset
,
1704 QEMUIOVector
*qiov
, BdrvRequestFlags flags
,
1705 BlockCompletionFunc
*cb
, void *opaque
)
1708 assert((uint64_t)qiov
->size
<= INT64_MAX
);
1709 return blk_aio_prwv(blk
, offset
, qiov
->size
, qiov
,
1710 blk_aio_read_entry
, flags
, cb
, opaque
);
1713 BlockAIOCB
*blk_aio_pwritev(BlockBackend
*blk
, int64_t offset
,
1714 QEMUIOVector
*qiov
, BdrvRequestFlags flags
,
1715 BlockCompletionFunc
*cb
, void *opaque
)
1718 assert((uint64_t)qiov
->size
<= INT64_MAX
);
1719 return blk_aio_prwv(blk
, offset
, qiov
->size
, qiov
,
1720 blk_aio_write_entry
, flags
, cb
, opaque
);
1723 void blk_aio_cancel(BlockAIOCB
*acb
)
1725 GLOBAL_STATE_CODE();
1726 bdrv_aio_cancel(acb
);
1729 void blk_aio_cancel_async(BlockAIOCB
*acb
)
1732 bdrv_aio_cancel_async(acb
);
1735 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1736 static int coroutine_fn
1737 blk_co_do_ioctl(BlockBackend
*blk
, unsigned long int req
, void *buf
)
1741 blk_wait_while_drained(blk
);
1742 GRAPH_RDLOCK_GUARD();
1744 if (!blk_co_is_available(blk
)) {
1748 return bdrv_co_ioctl(blk_bs(blk
), req
, buf
);
1751 int coroutine_fn
blk_co_ioctl(BlockBackend
*blk
, unsigned long int req
,
1757 blk_inc_in_flight(blk
);
1758 ret
= blk_co_do_ioctl(blk
, req
, buf
);
1759 blk_dec_in_flight(blk
);
1764 static void coroutine_fn
blk_aio_ioctl_entry(void *opaque
)
1766 BlkAioEmAIOCB
*acb
= opaque
;
1767 BlkRwCo
*rwco
= &acb
->rwco
;
1769 rwco
->ret
= blk_co_do_ioctl(rwco
->blk
, rwco
->offset
, rwco
->iobuf
);
1771 blk_aio_complete(acb
);
1774 BlockAIOCB
*blk_aio_ioctl(BlockBackend
*blk
, unsigned long int req
, void *buf
,
1775 BlockCompletionFunc
*cb
, void *opaque
)
1778 return blk_aio_prwv(blk
, req
, 0, buf
, blk_aio_ioctl_entry
, 0, cb
, opaque
);
1781 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1782 static int coroutine_fn
1783 blk_co_do_pdiscard(BlockBackend
*blk
, int64_t offset
, int64_t bytes
)
1788 blk_wait_while_drained(blk
);
1789 GRAPH_RDLOCK_GUARD();
1791 ret
= blk_check_byte_request(blk
, offset
, bytes
);
1796 return bdrv_co_pdiscard(blk
->root
, offset
, bytes
);
1799 static void coroutine_fn
blk_aio_pdiscard_entry(void *opaque
)
1801 BlkAioEmAIOCB
*acb
= opaque
;
1802 BlkRwCo
*rwco
= &acb
->rwco
;
1804 rwco
->ret
= blk_co_do_pdiscard(rwco
->blk
, rwco
->offset
, acb
->bytes
);
1805 blk_aio_complete(acb
);
1808 BlockAIOCB
*blk_aio_pdiscard(BlockBackend
*blk
,
1809 int64_t offset
, int64_t bytes
,
1810 BlockCompletionFunc
*cb
, void *opaque
)
1813 return blk_aio_prwv(blk
, offset
, bytes
, NULL
, blk_aio_pdiscard_entry
, 0,
1817 int coroutine_fn
blk_co_pdiscard(BlockBackend
*blk
, int64_t offset
,
1823 blk_inc_in_flight(blk
);
1824 ret
= blk_co_do_pdiscard(blk
, offset
, bytes
);
1825 blk_dec_in_flight(blk
);
1830 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1831 static int coroutine_fn
blk_co_do_flush(BlockBackend
*blk
)
1834 blk_wait_while_drained(blk
);
1835 GRAPH_RDLOCK_GUARD();
1837 if (!blk_co_is_available(blk
)) {
1841 return bdrv_co_flush(blk_bs(blk
));
1844 static void coroutine_fn
blk_aio_flush_entry(void *opaque
)
1846 BlkAioEmAIOCB
*acb
= opaque
;
1847 BlkRwCo
*rwco
= &acb
->rwco
;
1849 rwco
->ret
= blk_co_do_flush(rwco
->blk
);
1850 blk_aio_complete(acb
);
1853 BlockAIOCB
*blk_aio_flush(BlockBackend
*blk
,
1854 BlockCompletionFunc
*cb
, void *opaque
)
1857 return blk_aio_prwv(blk
, 0, 0, NULL
, blk_aio_flush_entry
, 0, cb
, opaque
);
1860 int coroutine_fn
blk_co_flush(BlockBackend
*blk
)
1865 blk_inc_in_flight(blk
);
1866 ret
= blk_co_do_flush(blk
);
1867 blk_dec_in_flight(blk
);
1872 static void coroutine_fn
blk_aio_zone_report_entry(void *opaque
)
1874 BlkAioEmAIOCB
*acb
= opaque
;
1875 BlkRwCo
*rwco
= &acb
->rwco
;
1877 rwco
->ret
= blk_co_zone_report(rwco
->blk
, rwco
->offset
,
1878 (unsigned int*)(uintptr_t)acb
->bytes
,
1880 blk_aio_complete(acb
);
1883 BlockAIOCB
*blk_aio_zone_report(BlockBackend
*blk
, int64_t offset
,
1884 unsigned int *nr_zones
,
1885 BlockZoneDescriptor
*zones
,
1886 BlockCompletionFunc
*cb
, void *opaque
)
1892 blk_inc_in_flight(blk
);
1893 acb
= blk_aio_get(&blk_aio_em_aiocb_info
, blk
, cb
, opaque
);
1894 acb
->rwco
= (BlkRwCo
) {
1900 acb
->bytes
= (int64_t)(uintptr_t)nr_zones
,
1901 acb
->has_returned
= false;
1903 co
= qemu_coroutine_create(blk_aio_zone_report_entry
, acb
);
1904 aio_co_enter(blk_get_aio_context(blk
), co
);
1906 acb
->has_returned
= true;
1907 if (acb
->rwco
.ret
!= NOT_DONE
) {
1908 replay_bh_schedule_oneshot_event(blk_get_aio_context(blk
),
1909 blk_aio_complete_bh
, acb
);
1912 return &acb
->common
;
1915 static void coroutine_fn
blk_aio_zone_mgmt_entry(void *opaque
)
1917 BlkAioEmAIOCB
*acb
= opaque
;
1918 BlkRwCo
*rwco
= &acb
->rwco
;
1920 rwco
->ret
= blk_co_zone_mgmt(rwco
->blk
,
1921 (BlockZoneOp
)(uintptr_t)rwco
->iobuf
,
1922 rwco
->offset
, acb
->bytes
);
1923 blk_aio_complete(acb
);
1926 BlockAIOCB
*blk_aio_zone_mgmt(BlockBackend
*blk
, BlockZoneOp op
,
1927 int64_t offset
, int64_t len
,
1928 BlockCompletionFunc
*cb
, void *opaque
) {
1933 blk_inc_in_flight(blk
);
1934 acb
= blk_aio_get(&blk_aio_em_aiocb_info
, blk
, cb
, opaque
);
1935 acb
->rwco
= (BlkRwCo
) {
1938 .iobuf
= (void *)(uintptr_t)op
,
1942 acb
->has_returned
= false;
1944 co
= qemu_coroutine_create(blk_aio_zone_mgmt_entry
, acb
);
1945 aio_co_enter(blk_get_aio_context(blk
), co
);
1947 acb
->has_returned
= true;
1948 if (acb
->rwco
.ret
!= NOT_DONE
) {
1949 replay_bh_schedule_oneshot_event(blk_get_aio_context(blk
),
1950 blk_aio_complete_bh
, acb
);
1953 return &acb
->common
;
1956 static void coroutine_fn
blk_aio_zone_append_entry(void *opaque
)
1958 BlkAioEmAIOCB
*acb
= opaque
;
1959 BlkRwCo
*rwco
= &acb
->rwco
;
1961 rwco
->ret
= blk_co_zone_append(rwco
->blk
, (int64_t *)(uintptr_t)acb
->bytes
,
1962 rwco
->iobuf
, rwco
->flags
);
1963 blk_aio_complete(acb
);
1966 BlockAIOCB
*blk_aio_zone_append(BlockBackend
*blk
, int64_t *offset
,
1967 QEMUIOVector
*qiov
, BdrvRequestFlags flags
,
1968 BlockCompletionFunc
*cb
, void *opaque
) {
1973 blk_inc_in_flight(blk
);
1974 acb
= blk_aio_get(&blk_aio_em_aiocb_info
, blk
, cb
, opaque
);
1975 acb
->rwco
= (BlkRwCo
) {
1981 acb
->bytes
= (int64_t)(uintptr_t)offset
;
1982 acb
->has_returned
= false;
1984 co
= qemu_coroutine_create(blk_aio_zone_append_entry
, acb
);
1985 aio_co_enter(blk_get_aio_context(blk
), co
);
1986 acb
->has_returned
= true;
1987 if (acb
->rwco
.ret
!= NOT_DONE
) {
1988 replay_bh_schedule_oneshot_event(blk_get_aio_context(blk
),
1989 blk_aio_complete_bh
, acb
);
1992 return &acb
->common
;
1996 * Send a zone_report command.
1997 * offset is a byte offset from the start of the device. No alignment
1998 * required for offset.
1999 * nr_zones represents IN maximum and OUT actual.
2001 int coroutine_fn
blk_co_zone_report(BlockBackend
*blk
, int64_t offset
,
2002 unsigned int *nr_zones
,
2003 BlockZoneDescriptor
*zones
)
2008 blk_inc_in_flight(blk
); /* increase before waiting */
2009 blk_wait_while_drained(blk
);
2010 GRAPH_RDLOCK_GUARD();
2011 if (!blk_is_available(blk
)) {
2012 blk_dec_in_flight(blk
);
2015 ret
= bdrv_co_zone_report(blk_bs(blk
), offset
, nr_zones
, zones
);
2016 blk_dec_in_flight(blk
);
2021 * Send a zone_management command.
2022 * op is the zone operation;
2023 * offset is the byte offset from the start of the zoned device;
2024 * len is the maximum number of bytes the command should operate on. It
2025 * should be aligned with the device zone size.
2027 int coroutine_fn
blk_co_zone_mgmt(BlockBackend
*blk
, BlockZoneOp op
,
2028 int64_t offset
, int64_t len
)
2033 blk_inc_in_flight(blk
);
2034 blk_wait_while_drained(blk
);
2035 GRAPH_RDLOCK_GUARD();
2037 ret
= blk_check_byte_request(blk
, offset
, len
);
2039 blk_dec_in_flight(blk
);
2043 ret
= bdrv_co_zone_mgmt(blk_bs(blk
), op
, offset
, len
);
2044 blk_dec_in_flight(blk
);
2049 * Send a zone_append command.
2051 int coroutine_fn
blk_co_zone_append(BlockBackend
*blk
, int64_t *offset
,
2052 QEMUIOVector
*qiov
, BdrvRequestFlags flags
)
2057 blk_inc_in_flight(blk
);
2058 blk_wait_while_drained(blk
);
2059 GRAPH_RDLOCK_GUARD();
2060 if (!blk_is_available(blk
)) {
2061 blk_dec_in_flight(blk
);
2065 ret
= bdrv_co_zone_append(blk_bs(blk
), offset
, qiov
, flags
);
2066 blk_dec_in_flight(blk
);
2070 void blk_drain(BlockBackend
*blk
)
2072 BlockDriverState
*bs
= blk_bs(blk
);
2073 GLOBAL_STATE_CODE();
2077 bdrv_drained_begin(bs
);
2080 /* We may have -ENOMEDIUM completions in flight */
2081 AIO_WAIT_WHILE(blk_get_aio_context(blk
),
2082 qatomic_read(&blk
->in_flight
) > 0);
2085 bdrv_drained_end(bs
);
2090 void blk_drain_all(void)
2092 BlockBackend
*blk
= NULL
;
2094 GLOBAL_STATE_CODE();
2096 bdrv_drain_all_begin();
2098 while ((blk
= blk_all_next(blk
)) != NULL
) {
2099 /* We may have -ENOMEDIUM completions in flight */
2100 AIO_WAIT_WHILE_UNLOCKED(NULL
, qatomic_read(&blk
->in_flight
) > 0);
2103 bdrv_drain_all_end();
2106 void blk_set_on_error(BlockBackend
*blk
, BlockdevOnError on_read_error
,
2107 BlockdevOnError on_write_error
)
2109 GLOBAL_STATE_CODE();
2110 blk
->on_read_error
= on_read_error
;
2111 blk
->on_write_error
= on_write_error
;
2114 BlockdevOnError
blk_get_on_error(BlockBackend
*blk
, bool is_read
)
2117 return is_read
? blk
->on_read_error
: blk
->on_write_error
;
2120 BlockErrorAction
blk_get_error_action(BlockBackend
*blk
, bool is_read
,
2123 BlockdevOnError on_err
= blk_get_on_error(blk
, is_read
);
2127 case BLOCKDEV_ON_ERROR_ENOSPC
:
2128 return (error
== ENOSPC
) ?
2129 BLOCK_ERROR_ACTION_STOP
: BLOCK_ERROR_ACTION_REPORT
;
2130 case BLOCKDEV_ON_ERROR_STOP
:
2131 return BLOCK_ERROR_ACTION_STOP
;
2132 case BLOCKDEV_ON_ERROR_REPORT
:
2133 return BLOCK_ERROR_ACTION_REPORT
;
2134 case BLOCKDEV_ON_ERROR_IGNORE
:
2135 return BLOCK_ERROR_ACTION_IGNORE
;
2136 case BLOCKDEV_ON_ERROR_AUTO
:
2142 static void send_qmp_error_event(BlockBackend
*blk
,
2143 BlockErrorAction action
,
2144 bool is_read
, int error
)
2146 IoOperationType optype
;
2147 BlockDriverState
*bs
= blk_bs(blk
);
2149 optype
= is_read
? IO_OPERATION_TYPE_READ
: IO_OPERATION_TYPE_WRITE
;
2150 qapi_event_send_block_io_error(blk_name(blk
),
2151 bs
? bdrv_get_node_name(bs
) : NULL
, optype
,
2152 action
, blk_iostatus_is_enabled(blk
),
2153 error
== ENOSPC
, strerror(error
));
2156 /* This is done by device models because, while the block layer knows
2157 * about the error, it does not know whether an operation comes from
2158 * the device or the block layer (from a job, for example).
2160 void blk_error_action(BlockBackend
*blk
, BlockErrorAction action
,
2161 bool is_read
, int error
)
2166 if (action
== BLOCK_ERROR_ACTION_STOP
) {
2167 /* First set the iostatus, so that "info block" returns an iostatus
2168 * that matches the events raised so far (an additional error iostatus
2169 * is fine, but not a lost one).
2171 blk_iostatus_set_err(blk
, error
);
2173 /* Then raise the request to stop the VM and the event.
2174 * qemu_system_vmstop_request_prepare has two effects. First,
2175 * it ensures that the STOP event always comes after the
2176 * BLOCK_IO_ERROR event. Second, it ensures that even if management
2177 * can observe the STOP event and do a "cont" before the STOP
2178 * event is issued, the VM will not stop. In this case, vm_start()
2179 * also ensures that the STOP/RESUME pair of events is emitted.
2181 qemu_system_vmstop_request_prepare();
2182 send_qmp_error_event(blk
, action
, is_read
, error
);
2183 qemu_system_vmstop_request(RUN_STATE_IO_ERROR
);
2185 send_qmp_error_event(blk
, action
, is_read
, error
);
2190 * Returns true if the BlockBackend can support taking write permissions
2191 * (because its root node is not read-only).
2193 bool blk_supports_write_perm(BlockBackend
*blk
)
2195 BlockDriverState
*bs
= blk_bs(blk
);
2196 GLOBAL_STATE_CODE();
2199 return !bdrv_is_read_only(bs
);
2201 return blk
->root_state
.open_flags
& BDRV_O_RDWR
;
2206 * Returns true if the BlockBackend can be written to in its current
2207 * configuration (i.e. if write permission have been requested)
2209 bool blk_is_writable(BlockBackend
*blk
)
2212 return blk
->perm
& BLK_PERM_WRITE
;
2215 bool blk_is_sg(BlockBackend
*blk
)
2217 BlockDriverState
*bs
= blk_bs(blk
);
2218 GLOBAL_STATE_CODE();
2224 return bdrv_is_sg(bs
);
2227 bool blk_enable_write_cache(BlockBackend
*blk
)
2230 return blk
->enable_write_cache
;
2233 void blk_set_enable_write_cache(BlockBackend
*blk
, bool wce
)
2236 blk
->enable_write_cache
= wce
;
2239 void blk_activate(BlockBackend
*blk
, Error
**errp
)
2241 BlockDriverState
*bs
= blk_bs(blk
);
2242 GLOBAL_STATE_CODE();
2245 error_setg(errp
, "Device '%s' has no medium", blk
->name
);
2250 * Migration code can call this function in coroutine context, so leave
2251 * coroutine context if necessary.
2253 if (qemu_in_coroutine()) {
2254 bdrv_co_activate(bs
, errp
);
2256 bdrv_activate(bs
, errp
);
2260 bool coroutine_fn
blk_co_is_inserted(BlockBackend
*blk
)
2262 BlockDriverState
*bs
= blk_bs(blk
);
2264 assert_bdrv_graph_readable();
2266 return bs
&& bdrv_co_is_inserted(bs
);
2269 bool coroutine_fn
blk_co_is_available(BlockBackend
*blk
)
2272 return blk_co_is_inserted(blk
) && !blk_dev_is_tray_open(blk
);
2275 void coroutine_fn
blk_co_lock_medium(BlockBackend
*blk
, bool locked
)
2277 BlockDriverState
*bs
= blk_bs(blk
);
2279 GRAPH_RDLOCK_GUARD();
2282 bdrv_co_lock_medium(bs
, locked
);
2286 void coroutine_fn
blk_co_eject(BlockBackend
*blk
, bool eject_flag
)
2288 BlockDriverState
*bs
= blk_bs(blk
);
2291 GRAPH_RDLOCK_GUARD();
2294 bdrv_co_eject(bs
, eject_flag
);
2297 /* Whether or not we ejected on the backend,
2298 * the frontend experienced a tray event. */
2299 id
= blk_get_attached_dev_id(blk
);
2300 qapi_event_send_device_tray_moved(blk_name(blk
), id
,
2305 int blk_get_flags(BlockBackend
*blk
)
2307 BlockDriverState
*bs
= blk_bs(blk
);
2308 GLOBAL_STATE_CODE();
2311 return bdrv_get_flags(bs
);
2313 return blk
->root_state
.open_flags
;
2317 /* Returns the minimum request alignment, in bytes; guaranteed nonzero */
2318 uint32_t blk_get_request_alignment(BlockBackend
*blk
)
2320 BlockDriverState
*bs
= blk_bs(blk
);
2322 return bs
? bs
->bl
.request_alignment
: BDRV_SECTOR_SIZE
;
2325 /* Returns the maximum hardware transfer length, in bytes; guaranteed nonzero */
2326 uint64_t blk_get_max_hw_transfer(BlockBackend
*blk
)
2328 BlockDriverState
*bs
= blk_bs(blk
);
2329 uint64_t max
= INT_MAX
;
2333 max
= MIN_NON_ZERO(max
, bs
->bl
.max_hw_transfer
);
2334 max
= MIN_NON_ZERO(max
, bs
->bl
.max_transfer
);
2336 return ROUND_DOWN(max
, blk_get_request_alignment(blk
));
2339 /* Returns the maximum transfer length, in bytes; guaranteed nonzero */
2340 uint32_t blk_get_max_transfer(BlockBackend
*blk
)
2342 BlockDriverState
*bs
= blk_bs(blk
);
2343 uint32_t max
= INT_MAX
;
2347 max
= MIN_NON_ZERO(max
, bs
->bl
.max_transfer
);
2349 return ROUND_DOWN(max
, blk_get_request_alignment(blk
));
2352 int blk_get_max_hw_iov(BlockBackend
*blk
)
2355 return MIN_NON_ZERO(blk
->root
->bs
->bl
.max_hw_iov
,
2356 blk
->root
->bs
->bl
.max_iov
);
2359 int blk_get_max_iov(BlockBackend
*blk
)
2362 return blk
->root
->bs
->bl
.max_iov
;
2365 void *blk_try_blockalign(BlockBackend
*blk
, size_t size
)
2368 return qemu_try_blockalign(blk
? blk_bs(blk
) : NULL
, size
);
2371 void *blk_blockalign(BlockBackend
*blk
, size_t size
)
2374 return qemu_blockalign(blk
? blk_bs(blk
) : NULL
, size
);
2377 bool blk_op_is_blocked(BlockBackend
*blk
, BlockOpType op
, Error
**errp
)
2379 BlockDriverState
*bs
= blk_bs(blk
);
2380 GLOBAL_STATE_CODE();
2386 return bdrv_op_is_blocked(bs
, op
, errp
);
2389 void blk_op_unblock(BlockBackend
*blk
, BlockOpType op
, Error
*reason
)
2391 BlockDriverState
*bs
= blk_bs(blk
);
2392 GLOBAL_STATE_CODE();
2395 bdrv_op_unblock(bs
, op
, reason
);
2399 void blk_op_block_all(BlockBackend
*blk
, Error
*reason
)
2401 BlockDriverState
*bs
= blk_bs(blk
);
2402 GLOBAL_STATE_CODE();
2405 bdrv_op_block_all(bs
, reason
);
2409 void blk_op_unblock_all(BlockBackend
*blk
, Error
*reason
)
2411 BlockDriverState
*bs
= blk_bs(blk
);
2412 GLOBAL_STATE_CODE();
2415 bdrv_op_unblock_all(bs
, reason
);
2419 AioContext
*blk_get_aio_context(BlockBackend
*blk
)
2421 BlockDriverState
*bs
;
2425 return qemu_get_aio_context();
2430 AioContext
*ctx
= bdrv_get_aio_context(blk_bs(blk
));
2431 assert(ctx
== blk
->ctx
);
2437 static AioContext
*blk_aiocb_get_aio_context(BlockAIOCB
*acb
)
2439 BlockBackendAIOCB
*blk_acb
= DO_UPCAST(BlockBackendAIOCB
, common
, acb
);
2440 return blk_get_aio_context(blk_acb
->blk
);
2443 int blk_set_aio_context(BlockBackend
*blk
, AioContext
*new_context
,
2446 bool old_allow_change
;
2447 BlockDriverState
*bs
= blk_bs(blk
);
2450 GLOBAL_STATE_CODE();
2453 blk
->ctx
= new_context
;
2459 old_allow_change
= blk
->allow_aio_context_change
;
2460 blk
->allow_aio_context_change
= true;
2462 ret
= bdrv_try_change_aio_context(bs
, new_context
, NULL
, errp
);
2464 blk
->allow_aio_context_change
= old_allow_change
;
2470 typedef struct BdrvStateBlkRootContext
{
2471 AioContext
*new_ctx
;
2473 } BdrvStateBlkRootContext
;
2475 static void blk_root_set_aio_ctx_commit(void *opaque
)
2477 BdrvStateBlkRootContext
*s
= opaque
;
2478 BlockBackend
*blk
= s
->blk
;
2479 AioContext
*new_context
= s
->new_ctx
;
2480 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
2482 blk
->ctx
= new_context
;
2483 if (tgm
->throttle_state
) {
2484 throttle_group_detach_aio_context(tgm
);
2485 throttle_group_attach_aio_context(tgm
, new_context
);
2489 static TransactionActionDrv set_blk_root_context
= {
2490 .commit
= blk_root_set_aio_ctx_commit
,
2494 static bool blk_root_change_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
2495 GHashTable
*visited
, Transaction
*tran
,
2498 BlockBackend
*blk
= child
->opaque
;
2499 BdrvStateBlkRootContext
*s
;
2501 if (!blk
->allow_aio_context_change
) {
2503 * Manually created BlockBackends (those with a name) that are not
2504 * attached to anything can change their AioContext without updating
2505 * their user; return an error for others.
2507 if (!blk
->name
|| blk
->dev
) {
2508 /* TODO Add BB name/QOM path */
2509 error_setg(errp
, "Cannot change iothread of active block backend");
2514 s
= g_new(BdrvStateBlkRootContext
, 1);
2515 *s
= (BdrvStateBlkRootContext
) {
2520 tran_add(tran
, &set_blk_root_context
, s
);
2524 void blk_add_aio_context_notifier(BlockBackend
*blk
,
2525 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
2526 void (*detach_aio_context
)(void *opaque
), void *opaque
)
2528 BlockBackendAioNotifier
*notifier
;
2529 BlockDriverState
*bs
= blk_bs(blk
);
2530 GLOBAL_STATE_CODE();
2532 notifier
= g_new(BlockBackendAioNotifier
, 1);
2533 notifier
->attached_aio_context
= attached_aio_context
;
2534 notifier
->detach_aio_context
= detach_aio_context
;
2535 notifier
->opaque
= opaque
;
2536 QLIST_INSERT_HEAD(&blk
->aio_notifiers
, notifier
, list
);
2539 bdrv_add_aio_context_notifier(bs
, attached_aio_context
,
2540 detach_aio_context
, opaque
);
2544 void blk_remove_aio_context_notifier(BlockBackend
*blk
,
2545 void (*attached_aio_context
)(AioContext
*,
2547 void (*detach_aio_context
)(void *),
2550 BlockBackendAioNotifier
*notifier
;
2551 BlockDriverState
*bs
= blk_bs(blk
);
2553 GLOBAL_STATE_CODE();
2556 bdrv_remove_aio_context_notifier(bs
, attached_aio_context
,
2557 detach_aio_context
, opaque
);
2560 QLIST_FOREACH(notifier
, &blk
->aio_notifiers
, list
) {
2561 if (notifier
->attached_aio_context
== attached_aio_context
&&
2562 notifier
->detach_aio_context
== detach_aio_context
&&
2563 notifier
->opaque
== opaque
) {
2564 QLIST_REMOVE(notifier
, list
);
2573 void blk_add_remove_bs_notifier(BlockBackend
*blk
, Notifier
*notify
)
2575 GLOBAL_STATE_CODE();
2576 notifier_list_add(&blk
->remove_bs_notifiers
, notify
);
2579 void blk_add_insert_bs_notifier(BlockBackend
*blk
, Notifier
*notify
)
2581 GLOBAL_STATE_CODE();
2582 notifier_list_add(&blk
->insert_bs_notifiers
, notify
);
2585 BlockAcctStats
*blk_get_stats(BlockBackend
*blk
)
2591 void *blk_aio_get(const AIOCBInfo
*aiocb_info
, BlockBackend
*blk
,
2592 BlockCompletionFunc
*cb
, void *opaque
)
2595 return qemu_aio_get(aiocb_info
, blk_bs(blk
), cb
, opaque
);
2598 int coroutine_fn
blk_co_pwrite_zeroes(BlockBackend
*blk
, int64_t offset
,
2599 int64_t bytes
, BdrvRequestFlags flags
)
2602 return blk_co_pwritev(blk
, offset
, bytes
, NULL
,
2603 flags
| BDRV_REQ_ZERO_WRITE
);
2606 int coroutine_fn
blk_co_pwrite_compressed(BlockBackend
*blk
, int64_t offset
,
2607 int64_t bytes
, const void *buf
)
2609 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, buf
, bytes
);
2611 return blk_co_pwritev_part(blk
, offset
, bytes
, &qiov
, 0,
2612 BDRV_REQ_WRITE_COMPRESSED
);
2615 int coroutine_fn
blk_co_truncate(BlockBackend
*blk
, int64_t offset
, bool exact
,
2616 PreallocMode prealloc
, BdrvRequestFlags flags
,
2620 GRAPH_RDLOCK_GUARD();
2621 if (!blk_co_is_available(blk
)) {
2622 error_setg(errp
, "No medium inserted");
2626 return bdrv_co_truncate(blk
->root
, offset
, exact
, prealloc
, flags
, errp
);
2629 int blk_save_vmstate(BlockBackend
*blk
, const uint8_t *buf
,
2630 int64_t pos
, int size
)
2633 GLOBAL_STATE_CODE();
2635 if (!blk_is_available(blk
)) {
2639 ret
= bdrv_save_vmstate(blk_bs(blk
), buf
, pos
, size
);
2644 if (ret
== size
&& !blk
->enable_write_cache
) {
2645 ret
= bdrv_flush(blk_bs(blk
));
2648 return ret
< 0 ? ret
: size
;
2651 int blk_load_vmstate(BlockBackend
*blk
, uint8_t *buf
, int64_t pos
, int size
)
2653 GLOBAL_STATE_CODE();
2654 if (!blk_is_available(blk
)) {
2658 return bdrv_load_vmstate(blk_bs(blk
), buf
, pos
, size
);
2661 int blk_probe_blocksizes(BlockBackend
*blk
, BlockSizes
*bsz
)
2663 GLOBAL_STATE_CODE();
2664 if (!blk_is_available(blk
)) {
2668 return bdrv_probe_blocksizes(blk_bs(blk
), bsz
);
2671 int blk_probe_geometry(BlockBackend
*blk
, HDGeometry
*geo
)
2673 GLOBAL_STATE_CODE();
2674 if (!blk_is_available(blk
)) {
2678 return bdrv_probe_geometry(blk_bs(blk
), geo
);
2682 * Updates the BlockBackendRootState object with data from the currently
2683 * attached BlockDriverState.
2685 void blk_update_root_state(BlockBackend
*blk
)
2687 GLOBAL_STATE_CODE();
2690 blk
->root_state
.open_flags
= blk
->root
->bs
->open_flags
;
2691 blk
->root_state
.detect_zeroes
= blk
->root
->bs
->detect_zeroes
;
2695 * Returns the detect-zeroes setting to be used for bdrv_open() of a
2696 * BlockDriverState which is supposed to inherit the root state.
2698 bool blk_get_detect_zeroes_from_root_state(BlockBackend
*blk
)
2700 GLOBAL_STATE_CODE();
2701 return blk
->root_state
.detect_zeroes
;
2705 * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
2706 * supposed to inherit the root state.
2708 int blk_get_open_flags_from_root_state(BlockBackend
*blk
)
2710 GLOBAL_STATE_CODE();
2711 return blk
->root_state
.open_flags
;
2714 BlockBackendRootState
*blk_get_root_state(BlockBackend
*blk
)
2716 GLOBAL_STATE_CODE();
2717 return &blk
->root_state
;
2720 int blk_commit_all(void)
2722 BlockBackend
*blk
= NULL
;
2723 GLOBAL_STATE_CODE();
2725 while ((blk
= blk_all_next(blk
)) != NULL
) {
2726 AioContext
*aio_context
= blk_get_aio_context(blk
);
2727 BlockDriverState
*unfiltered_bs
= bdrv_skip_filters(blk_bs(blk
));
2729 aio_context_acquire(aio_context
);
2730 if (blk_is_inserted(blk
) && bdrv_cow_child(unfiltered_bs
)) {
2733 ret
= bdrv_commit(unfiltered_bs
);
2735 aio_context_release(aio_context
);
2739 aio_context_release(aio_context
);
2745 /* throttling disk I/O limits */
2746 void blk_set_io_limits(BlockBackend
*blk
, ThrottleConfig
*cfg
)
2748 GLOBAL_STATE_CODE();
2749 throttle_group_config(&blk
->public.throttle_group_member
, cfg
);
2752 void blk_io_limits_disable(BlockBackend
*blk
)
2754 BlockDriverState
*bs
= blk_bs(blk
);
2755 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
2756 assert(tgm
->throttle_state
);
2757 GLOBAL_STATE_CODE();
2760 bdrv_drained_begin(bs
);
2762 throttle_group_unregister_tgm(tgm
);
2764 bdrv_drained_end(bs
);
2769 /* should be called before blk_set_io_limits if a limit is set */
2770 void blk_io_limits_enable(BlockBackend
*blk
, const char *group
)
2772 assert(!blk
->public.throttle_group_member
.throttle_state
);
2773 GLOBAL_STATE_CODE();
2774 throttle_group_register_tgm(&blk
->public.throttle_group_member
,
2775 group
, blk_get_aio_context(blk
));
2778 void blk_io_limits_update_group(BlockBackend
*blk
, const char *group
)
2780 GLOBAL_STATE_CODE();
2781 /* this BB is not part of any group */
2782 if (!blk
->public.throttle_group_member
.throttle_state
) {
2786 /* this BB is a part of the same group than the one we want */
2787 if (!g_strcmp0(throttle_group_get_name(&blk
->public.throttle_group_member
),
2792 /* need to change the group this bs belong to */
2793 blk_io_limits_disable(blk
);
2794 blk_io_limits_enable(blk
, group
);
2797 static void blk_root_drained_begin(BdrvChild
*child
)
2799 BlockBackend
*blk
= child
->opaque
;
2800 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
2802 if (qatomic_fetch_inc(&blk
->quiesce_counter
) == 0) {
2803 if (blk
->dev_ops
&& blk
->dev_ops
->drained_begin
) {
2804 blk
->dev_ops
->drained_begin(blk
->dev_opaque
);
2808 /* Note that blk->root may not be accessible here yet if we are just
2809 * attaching to a BlockDriverState that is drained. Use child instead. */
2811 if (qatomic_fetch_inc(&tgm
->io_limits_disabled
) == 0) {
2812 throttle_group_restart_tgm(tgm
);
2816 static bool blk_root_drained_poll(BdrvChild
*child
)
2818 BlockBackend
*blk
= child
->opaque
;
2820 assert(qatomic_read(&blk
->quiesce_counter
));
2822 if (blk
->dev_ops
&& blk
->dev_ops
->drained_poll
) {
2823 busy
= blk
->dev_ops
->drained_poll(blk
->dev_opaque
);
2825 return busy
|| !!blk
->in_flight
;
2828 static void blk_root_drained_end(BdrvChild
*child
)
2830 BlockBackend
*blk
= child
->opaque
;
2831 assert(qatomic_read(&blk
->quiesce_counter
));
2833 assert(blk
->public.throttle_group_member
.io_limits_disabled
);
2834 qatomic_dec(&blk
->public.throttle_group_member
.io_limits_disabled
);
2836 if (qatomic_fetch_dec(&blk
->quiesce_counter
) == 1) {
2837 if (blk
->dev_ops
&& blk
->dev_ops
->drained_end
) {
2838 blk
->dev_ops
->drained_end(blk
->dev_opaque
);
2840 qemu_mutex_lock(&blk
->queued_requests_lock
);
2841 while (qemu_co_enter_next(&blk
->queued_requests
,
2842 &blk
->queued_requests_lock
)) {
2843 /* Resume all queued requests */
2845 qemu_mutex_unlock(&blk
->queued_requests_lock
);
2849 bool blk_register_buf(BlockBackend
*blk
, void *host
, size_t size
, Error
**errp
)
2851 BlockDriverState
*bs
= blk_bs(blk
);
2853 GLOBAL_STATE_CODE();
2856 return bdrv_register_buf(bs
, host
, size
, errp
);
2861 void blk_unregister_buf(BlockBackend
*blk
, void *host
, size_t size
)
2863 BlockDriverState
*bs
= blk_bs(blk
);
2865 GLOBAL_STATE_CODE();
2868 bdrv_unregister_buf(bs
, host
, size
);
2872 int coroutine_fn
blk_co_copy_range(BlockBackend
*blk_in
, int64_t off_in
,
2873 BlockBackend
*blk_out
, int64_t off_out
,
2874 int64_t bytes
, BdrvRequestFlags read_flags
,
2875 BdrvRequestFlags write_flags
)
2879 GRAPH_RDLOCK_GUARD();
2881 r
= blk_check_byte_request(blk_in
, off_in
, bytes
);
2885 r
= blk_check_byte_request(blk_out
, off_out
, bytes
);
2890 return bdrv_co_copy_range(blk_in
->root
, off_in
,
2891 blk_out
->root
, off_out
,
2892 bytes
, read_flags
, write_flags
);
2895 const BdrvChild
*blk_root(BlockBackend
*blk
)
2897 GLOBAL_STATE_CODE();
2901 int blk_make_empty(BlockBackend
*blk
, Error
**errp
)
2903 GLOBAL_STATE_CODE();
2904 if (!blk_is_available(blk
)) {
2905 error_setg(errp
, "No medium inserted");
2909 return bdrv_make_empty(blk
->root
, errp
);