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
;
84 CoQueue queued_requests
;
85 bool disable_request_queuing
;
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
, int *drained_end_counter
);
134 static void blk_root_change_media(BdrvChild
*child
, bool load
);
135 static void blk_root_resize(BdrvChild
*child
);
137 static bool blk_root_can_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
138 GSList
**ignore
, Error
**errp
);
139 static void blk_root_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
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
;
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 .can_set_aio_ctx
= blk_root_can_set_aio_ctx
,
338 .set_aio_ctx
= blk_root_set_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_co_queue_init(&blk
->queued_requests
);
373 notifier_list_init(&blk
->remove_bs_notifiers
);
374 notifier_list_init(&blk
->insert_bs_notifiers
);
375 QLIST_INIT(&blk
->aio_notifiers
);
377 QTAILQ_INSERT_TAIL(&block_backends
, blk
, link
);
382 * Create a new BlockBackend connected to an existing BlockDriverState.
384 * @perm is a bitmasks of BLK_PERM_* constants which describes the
385 * permissions to request for @bs that is attached to this
386 * BlockBackend. @shared_perm is a bitmask which describes which
387 * permissions may be granted to other users of the attached node.
388 * Both sets of permissions can be changed later using blk_set_perm().
390 * Return the new BlockBackend on success, null on failure.
392 BlockBackend
*blk_new_with_bs(BlockDriverState
*bs
, uint64_t perm
,
393 uint64_t shared_perm
, Error
**errp
)
395 BlockBackend
*blk
= blk_new(bdrv_get_aio_context(bs
), perm
, shared_perm
);
399 if (blk_insert_bs(blk
, bs
, errp
) < 0) {
407 * Creates a new BlockBackend, opens a new BlockDriverState, and connects both.
408 * The new BlockBackend is in the main AioContext.
410 * Just as with bdrv_open(), after having called this function the reference to
411 * @options belongs to the block layer (even on failure).
413 * TODO: Remove @filename and @flags; it should be possible to specify a whole
414 * BDS tree just by specifying the @options QDict (or @reference,
415 * alternatively). At the time of adding this function, this is not possible,
416 * though, so callers of this function have to be able to specify @filename and
419 BlockBackend
*blk_new_open(const char *filename
, const char *reference
,
420 QDict
*options
, int flags
, Error
**errp
)
423 BlockDriverState
*bs
;
425 uint64_t shared
= BLK_PERM_ALL
;
430 * blk_new_open() is mainly used in .bdrv_create implementations and the
431 * tools where sharing isn't a major concern because the BDS stays private
432 * and the file is generally not supposed to be used by a second process,
433 * so we just request permission according to the flags.
435 * The exceptions are xen_disk and blockdev_init(); in these cases, the
436 * caller of blk_new_open() doesn't make use of the permissions, but they
437 * shouldn't hurt either. We can still share everything here because the
438 * guest devices will add their own blockers if they can't share.
440 if ((flags
& BDRV_O_NO_IO
) == 0) {
441 perm
|= BLK_PERM_CONSISTENT_READ
;
442 if (flags
& BDRV_O_RDWR
) {
443 perm
|= BLK_PERM_WRITE
;
446 if (flags
& BDRV_O_RESIZE
) {
447 perm
|= BLK_PERM_RESIZE
;
449 if (flags
& BDRV_O_NO_SHARE
) {
450 shared
= BLK_PERM_CONSISTENT_READ
| BLK_PERM_WRITE_UNCHANGED
;
453 blk
= blk_new(qemu_get_aio_context(), perm
, shared
);
454 bs
= bdrv_open(filename
, reference
, options
, flags
, errp
);
460 blk
->root
= bdrv_root_attach_child(bs
, "root", &child_root
,
461 BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
,
462 perm
, shared
, blk
, errp
);
471 static void blk_delete(BlockBackend
*blk
)
473 assert(!blk
->refcnt
);
476 if (blk
->public.throttle_group_member
.throttle_state
) {
477 blk_io_limits_disable(blk
);
483 qemu_del_vm_change_state_handler(blk
->vmsh
);
486 assert(QLIST_EMPTY(&blk
->remove_bs_notifiers
.notifiers
));
487 assert(QLIST_EMPTY(&blk
->insert_bs_notifiers
.notifiers
));
488 assert(QLIST_EMPTY(&blk
->aio_notifiers
));
489 QTAILQ_REMOVE(&block_backends
, blk
, link
);
490 drive_info_del(blk
->legacy_dinfo
);
491 block_acct_cleanup(&blk
->stats
);
495 static void drive_info_del(DriveInfo
*dinfo
)
500 qemu_opts_del(dinfo
->opts
);
504 int blk_get_refcnt(BlockBackend
*blk
)
507 return blk
? blk
->refcnt
: 0;
511 * Increment @blk's reference count.
512 * @blk must not be null.
514 void blk_ref(BlockBackend
*blk
)
516 assert(blk
->refcnt
> 0);
522 * Decrement @blk's reference count.
523 * If this drops it to zero, destroy @blk.
524 * For convenience, do nothing if @blk is null.
526 void blk_unref(BlockBackend
*blk
)
530 assert(blk
->refcnt
> 0);
531 if (blk
->refcnt
> 1) {
535 /* blk_drain() cannot resurrect blk, nobody held a reference */
536 assert(blk
->refcnt
== 1);
544 * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
545 * ones which are hidden (i.e. are not referenced by the monitor).
547 BlockBackend
*blk_all_next(BlockBackend
*blk
)
550 return blk
? QTAILQ_NEXT(blk
, link
)
551 : QTAILQ_FIRST(&block_backends
);
554 void blk_remove_all_bs(void)
556 BlockBackend
*blk
= NULL
;
560 while ((blk
= blk_all_next(blk
)) != NULL
) {
561 AioContext
*ctx
= blk_get_aio_context(blk
);
563 aio_context_acquire(ctx
);
567 aio_context_release(ctx
);
572 * Return the monitor-owned BlockBackend after @blk.
573 * If @blk is null, return the first one.
574 * Else, return @blk's next sibling, which may be null.
576 * To iterate over all BlockBackends, do
577 * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
581 BlockBackend
*blk_next(BlockBackend
*blk
)
584 return blk
? QTAILQ_NEXT(blk
, monitor_link
)
585 : QTAILQ_FIRST(&monitor_block_backends
);
588 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
589 * the monitor or attached to a BlockBackend */
590 BlockDriverState
*bdrv_next(BdrvNextIterator
*it
)
592 BlockDriverState
*bs
, *old_bs
;
594 /* Must be called from the main loop */
595 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
597 /* First, return all root nodes of BlockBackends. In order to avoid
598 * returning a BDS twice when multiple BBs refer to it, we only return it
599 * if the BB is the first one in the parent list of the BDS. */
600 if (it
->phase
== BDRV_NEXT_BACKEND_ROOTS
) {
601 BlockBackend
*old_blk
= it
->blk
;
603 old_bs
= old_blk
? blk_bs(old_blk
) : NULL
;
606 it
->blk
= blk_all_next(it
->blk
);
607 bs
= it
->blk
? blk_bs(it
->blk
) : NULL
;
608 } while (it
->blk
&& (bs
== NULL
|| bdrv_first_blk(bs
) != it
->blk
));
620 it
->phase
= BDRV_NEXT_MONITOR_OWNED
;
625 /* Then return the monitor-owned BDSes without a BB attached. Ignore all
626 * BDSes that are attached to a BlockBackend here; they have been handled
627 * by the above block already */
629 it
->bs
= bdrv_next_monitor_owned(it
->bs
);
631 } while (bs
&& bdrv_has_blk(bs
));
641 static void bdrv_next_reset(BdrvNextIterator
*it
)
643 *it
= (BdrvNextIterator
) {
644 .phase
= BDRV_NEXT_BACKEND_ROOTS
,
648 BlockDriverState
*bdrv_first(BdrvNextIterator
*it
)
652 return bdrv_next(it
);
655 /* Must be called when aborting a bdrv_next() iteration before
656 * bdrv_next() returns NULL */
657 void bdrv_next_cleanup(BdrvNextIterator
*it
)
659 /* Must be called from the main loop */
660 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
662 if (it
->phase
== BDRV_NEXT_BACKEND_ROOTS
) {
664 bdrv_unref(blk_bs(it
->blk
));
675 * Add a BlockBackend into the list of backends referenced by the monitor, with
676 * the given @name acting as the handle for the monitor.
677 * Strictly for use by blockdev.c.
679 * @name must not be null or empty.
681 * Returns true on success and false on failure. In the latter case, an Error
682 * object is returned through @errp.
684 bool monitor_add_blk(BlockBackend
*blk
, const char *name
, Error
**errp
)
687 assert(name
&& name
[0]);
690 if (!id_wellformed(name
)) {
691 error_setg(errp
, "Invalid device name");
694 if (blk_by_name(name
)) {
695 error_setg(errp
, "Device with id '%s' already exists", name
);
698 if (bdrv_find_node(name
)) {
700 "Device name '%s' conflicts with an existing node name",
705 blk
->name
= g_strdup(name
);
706 QTAILQ_INSERT_TAIL(&monitor_block_backends
, blk
, monitor_link
);
711 * Remove a BlockBackend from the list of backends referenced by the monitor.
712 * Strictly for use by blockdev.c.
714 void monitor_remove_blk(BlockBackend
*blk
)
722 QTAILQ_REMOVE(&monitor_block_backends
, blk
, monitor_link
);
728 * Return @blk's name, a non-null string.
729 * Returns an empty string iff @blk is not referenced by the monitor.
731 const char *blk_name(const BlockBackend
*blk
)
734 return blk
->name
?: "";
738 * Return the BlockBackend with name @name if it exists, else null.
739 * @name must not be null.
741 BlockBackend
*blk_by_name(const char *name
)
743 BlockBackend
*blk
= NULL
;
747 while ((blk
= blk_next(blk
)) != NULL
) {
748 if (!strcmp(name
, blk
->name
)) {
756 * Return the BlockDriverState attached to @blk if any, else null.
758 BlockDriverState
*blk_bs(BlockBackend
*blk
)
761 return blk
->root
? blk
->root
->bs
: NULL
;
764 static BlockBackend
*bdrv_first_blk(BlockDriverState
*bs
)
770 QLIST_FOREACH(child
, &bs
->parents
, next_parent
) {
771 if (child
->klass
== &child_root
) {
772 return child
->opaque
;
780 * Returns true if @bs has an associated BlockBackend.
782 bool bdrv_has_blk(BlockDriverState
*bs
)
785 return bdrv_first_blk(bs
) != NULL
;
789 * Returns true if @bs has only BlockBackends as parents.
791 bool bdrv_is_root_node(BlockDriverState
*bs
)
796 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
797 if (c
->klass
!= &child_root
) {
806 * Return @blk's DriveInfo if any, else null.
808 DriveInfo
*blk_legacy_dinfo(BlockBackend
*blk
)
811 return blk
->legacy_dinfo
;
815 * Set @blk's DriveInfo to @dinfo, and return it.
816 * @blk must not have a DriveInfo set already.
817 * No other BlockBackend may have the same DriveInfo set.
819 DriveInfo
*blk_set_legacy_dinfo(BlockBackend
*blk
, DriveInfo
*dinfo
)
821 assert(!blk
->legacy_dinfo
);
823 return blk
->legacy_dinfo
= dinfo
;
827 * Return the BlockBackend with DriveInfo @dinfo.
830 BlockBackend
*blk_by_legacy_dinfo(DriveInfo
*dinfo
)
832 BlockBackend
*blk
= NULL
;
835 while ((blk
= blk_next(blk
)) != NULL
) {
836 if (blk
->legacy_dinfo
== dinfo
) {
844 * Returns a pointer to the publicly accessible fields of @blk.
846 BlockBackendPublic
*blk_get_public(BlockBackend
*blk
)
853 * Returns a BlockBackend given the associated @public fields.
855 BlockBackend
*blk_by_public(BlockBackendPublic
*public)
858 return container_of(public, BlockBackend
, public);
862 * Disassociates the currently associated BlockDriverState from @blk.
864 void blk_remove_bs(BlockBackend
*blk
)
866 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
871 notifier_list_notify(&blk
->remove_bs_notifiers
, blk
);
872 if (tgm
->throttle_state
) {
873 BlockDriverState
*bs
= blk_bs(blk
);
876 * Take a ref in case blk_bs() changes across bdrv_drained_begin(), for
877 * example, if a temporary filter node is removed by a blockjob.
880 bdrv_drained_begin(bs
);
881 throttle_group_detach_aio_context(tgm
);
882 throttle_group_attach_aio_context(tgm
, qemu_get_aio_context());
883 bdrv_drained_end(bs
);
887 blk_update_root_state(blk
);
889 /* bdrv_root_unref_child() will cause blk->root to become stale and may
890 * switch to a completion coroutine later on. Let's drain all I/O here
891 * to avoid that and a potential QEMU crash.
896 bdrv_root_unref_child(root
);
900 * Associates a new BlockDriverState with @blk.
902 int blk_insert_bs(BlockBackend
*blk
, BlockDriverState
*bs
, Error
**errp
)
904 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
907 blk
->root
= bdrv_root_attach_child(bs
, "root", &child_root
,
908 BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
,
909 blk
->perm
, blk
->shared_perm
,
911 if (blk
->root
== NULL
) {
915 notifier_list_notify(&blk
->insert_bs_notifiers
, blk
);
916 if (tgm
->throttle_state
) {
917 throttle_group_detach_aio_context(tgm
);
918 throttle_group_attach_aio_context(tgm
, bdrv_get_aio_context(bs
));
925 * Change BlockDriverState associated with @blk.
927 int blk_replace_bs(BlockBackend
*blk
, BlockDriverState
*new_bs
, Error
**errp
)
930 return bdrv_replace_child_bs(blk
->root
, new_bs
, errp
);
934 * Sets the permission bitmasks that the user of the BlockBackend needs.
936 int blk_set_perm(BlockBackend
*blk
, uint64_t perm
, uint64_t shared_perm
,
942 if (blk
->root
&& !blk
->disable_perm
) {
943 ret
= bdrv_child_try_set_perm(blk
->root
, perm
, shared_perm
, errp
);
950 blk
->shared_perm
= shared_perm
;
955 void blk_get_perm(BlockBackend
*blk
, uint64_t *perm
, uint64_t *shared_perm
)
959 *shared_perm
= blk
->shared_perm
;
963 * Attach device model @dev to @blk.
964 * Return 0 on success, -EBUSY when a device model is attached already.
966 int blk_attach_dev(BlockBackend
*blk
, DeviceState
*dev
)
973 /* While migration is still incoming, we don't need to apply the
974 * permissions of guest device BlockBackends. We might still have a block
975 * job or NBD server writing to the image for storage migration. */
976 if (runstate_check(RUN_STATE_INMIGRATE
)) {
977 blk
->disable_perm
= true;
982 blk_iostatus_reset(blk
);
988 * Detach device model @dev from @blk.
989 * @dev must be currently attached to @blk.
991 void blk_detach_dev(BlockBackend
*blk
, DeviceState
*dev
)
993 assert(blk
->dev
== dev
);
997 blk
->dev_opaque
= NULL
;
998 blk_set_perm(blk
, 0, BLK_PERM_ALL
, &error_abort
);
1003 * Return the device model attached to @blk if any, else null.
1005 DeviceState
*blk_get_attached_dev(BlockBackend
*blk
)
1007 GLOBAL_STATE_CODE();
1011 /* Return the qdev ID, or if no ID is assigned the QOM path, of the block
1012 * device attached to the BlockBackend. */
1013 char *blk_get_attached_dev_id(BlockBackend
*blk
)
1015 DeviceState
*dev
= blk
->dev
;
1019 return g_strdup("");
1020 } else if (dev
->id
) {
1021 return g_strdup(dev
->id
);
1024 return object_get_canonical_path(OBJECT(dev
)) ?: g_strdup("");
1028 * Return the BlockBackend which has the device model @dev attached if it
1029 * exists, else null.
1031 * @dev must not be null.
1033 BlockBackend
*blk_by_dev(void *dev
)
1035 BlockBackend
*blk
= NULL
;
1037 GLOBAL_STATE_CODE();
1039 assert(dev
!= NULL
);
1040 while ((blk
= blk_all_next(blk
)) != NULL
) {
1041 if (blk
->dev
== dev
) {
1049 * Set @blk's device model callbacks to @ops.
1050 * @opaque is the opaque argument to pass to the callbacks.
1051 * This is for use by device models.
1053 void blk_set_dev_ops(BlockBackend
*blk
, const BlockDevOps
*ops
,
1056 GLOBAL_STATE_CODE();
1058 blk
->dev_opaque
= opaque
;
1060 /* Are we currently quiesced? Should we enforce this right now? */
1061 if (blk
->quiesce_counter
&& ops
&& ops
->drained_begin
) {
1062 ops
->drained_begin(opaque
);
1067 * Notify @blk's attached device model of media change.
1069 * If @load is true, notify of media load. This action can fail, meaning that
1070 * the medium cannot be loaded. @errp is set then.
1072 * If @load is false, notify of media eject. This can never fail.
1074 * Also send DEVICE_TRAY_MOVED events as appropriate.
1076 void blk_dev_change_media_cb(BlockBackend
*blk
, bool load
, Error
**errp
)
1078 GLOBAL_STATE_CODE();
1079 if (blk
->dev_ops
&& blk
->dev_ops
->change_media_cb
) {
1080 bool tray_was_open
, tray_is_open
;
1081 Error
*local_err
= NULL
;
1083 tray_was_open
= blk_dev_is_tray_open(blk
);
1084 blk
->dev_ops
->change_media_cb(blk
->dev_opaque
, load
, &local_err
);
1086 assert(load
== true);
1087 error_propagate(errp
, local_err
);
1090 tray_is_open
= blk_dev_is_tray_open(blk
);
1092 if (tray_was_open
!= tray_is_open
) {
1093 char *id
= blk_get_attached_dev_id(blk
);
1094 qapi_event_send_device_tray_moved(blk_name(blk
), id
, tray_is_open
);
1100 static void blk_root_change_media(BdrvChild
*child
, bool load
)
1102 blk_dev_change_media_cb(child
->opaque
, load
, NULL
);
1106 * Does @blk's attached device model have removable media?
1107 * %true if no device model is attached.
1109 bool blk_dev_has_removable_media(BlockBackend
*blk
)
1111 GLOBAL_STATE_CODE();
1112 return !blk
->dev
|| (blk
->dev_ops
&& blk
->dev_ops
->change_media_cb
);
1116 * Does @blk's attached device model have a tray?
1118 bool blk_dev_has_tray(BlockBackend
*blk
)
1121 return blk
->dev_ops
&& blk
->dev_ops
->is_tray_open
;
1125 * Notify @blk's attached device model of a media eject request.
1126 * If @force is true, the medium is about to be yanked out forcefully.
1128 void blk_dev_eject_request(BlockBackend
*blk
, bool force
)
1130 GLOBAL_STATE_CODE();
1131 if (blk
->dev_ops
&& blk
->dev_ops
->eject_request_cb
) {
1132 blk
->dev_ops
->eject_request_cb(blk
->dev_opaque
, force
);
1137 * Does @blk's attached device model have a tray, and is it open?
1139 bool blk_dev_is_tray_open(BlockBackend
*blk
)
1142 if (blk_dev_has_tray(blk
)) {
1143 return blk
->dev_ops
->is_tray_open(blk
->dev_opaque
);
1149 * Does @blk's attached device model have the medium locked?
1150 * %false if the device model has no such lock.
1152 bool blk_dev_is_medium_locked(BlockBackend
*blk
)
1154 GLOBAL_STATE_CODE();
1155 if (blk
->dev_ops
&& blk
->dev_ops
->is_medium_locked
) {
1156 return blk
->dev_ops
->is_medium_locked(blk
->dev_opaque
);
1162 * Notify @blk's attached device model of a backend size change.
1164 static void blk_root_resize(BdrvChild
*child
)
1166 BlockBackend
*blk
= child
->opaque
;
1168 if (blk
->dev_ops
&& blk
->dev_ops
->resize_cb
) {
1169 blk
->dev_ops
->resize_cb(blk
->dev_opaque
);
1173 void blk_iostatus_enable(BlockBackend
*blk
)
1175 GLOBAL_STATE_CODE();
1176 blk
->iostatus_enabled
= true;
1177 blk
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
1180 /* The I/O status is only enabled if the drive explicitly
1181 * enables it _and_ the VM is configured to stop on errors */
1182 bool blk_iostatus_is_enabled(const BlockBackend
*blk
)
1185 return (blk
->iostatus_enabled
&&
1186 (blk
->on_write_error
== BLOCKDEV_ON_ERROR_ENOSPC
||
1187 blk
->on_write_error
== BLOCKDEV_ON_ERROR_STOP
||
1188 blk
->on_read_error
== BLOCKDEV_ON_ERROR_STOP
));
1191 BlockDeviceIoStatus
blk_iostatus(const BlockBackend
*blk
)
1193 GLOBAL_STATE_CODE();
1194 return blk
->iostatus
;
1197 void blk_iostatus_disable(BlockBackend
*blk
)
1199 GLOBAL_STATE_CODE();
1200 blk
->iostatus_enabled
= false;
1203 void blk_iostatus_reset(BlockBackend
*blk
)
1205 GLOBAL_STATE_CODE();
1206 if (blk_iostatus_is_enabled(blk
)) {
1207 blk
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
1211 void blk_iostatus_set_err(BlockBackend
*blk
, int error
)
1214 assert(blk_iostatus_is_enabled(blk
));
1215 if (blk
->iostatus
== BLOCK_DEVICE_IO_STATUS_OK
) {
1216 blk
->iostatus
= error
== ENOSPC
? BLOCK_DEVICE_IO_STATUS_NOSPACE
:
1217 BLOCK_DEVICE_IO_STATUS_FAILED
;
1221 void blk_set_allow_write_beyond_eof(BlockBackend
*blk
, bool allow
)
1224 blk
->allow_write_beyond_eof
= allow
;
1227 void blk_set_allow_aio_context_change(BlockBackend
*blk
, bool allow
)
1230 blk
->allow_aio_context_change
= allow
;
1233 void blk_set_disable_request_queuing(BlockBackend
*blk
, bool disable
)
1236 blk
->disable_request_queuing
= disable
;
1239 static int blk_check_byte_request(BlockBackend
*blk
, int64_t offset
,
1248 if (!blk_is_available(blk
)) {
1256 if (!blk
->allow_write_beyond_eof
) {
1257 len
= blk_getlength(blk
);
1262 if (offset
> len
|| len
- offset
< bytes
) {
1270 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1271 static void coroutine_fn
blk_wait_while_drained(BlockBackend
*blk
)
1273 assert(blk
->in_flight
> 0);
1275 if (blk
->quiesce_counter
&& !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
);
1294 /* Call blk_bs() only after waiting, the graph may have changed */
1296 trace_blk_co_preadv(blk
, bs
, offset
, bytes
, flags
);
1298 ret
= blk_check_byte_request(blk
, offset
, bytes
);
1303 bdrv_inc_in_flight(bs
);
1305 /* throttling disk I/O */
1306 if (blk
->public.throttle_group_member
.throttle_state
) {
1307 throttle_group_co_io_limits_intercept(&blk
->public.throttle_group_member
,
1311 ret
= bdrv_co_preadv_part(blk
->root
, offset
, bytes
, qiov
, qiov_offset
,
1313 bdrv_dec_in_flight(bs
);
1317 int coroutine_fn
blk_co_preadv(BlockBackend
*blk
, int64_t offset
,
1318 int64_t bytes
, QEMUIOVector
*qiov
,
1319 BdrvRequestFlags flags
)
1324 blk_inc_in_flight(blk
);
1325 ret
= blk_co_do_preadv_part(blk
, offset
, bytes
, qiov
, 0, flags
);
1326 blk_dec_in_flight(blk
);
1331 int coroutine_fn
blk_co_preadv_part(BlockBackend
*blk
, int64_t offset
,
1332 int64_t bytes
, QEMUIOVector
*qiov
,
1333 size_t qiov_offset
, BdrvRequestFlags flags
)
1338 blk_inc_in_flight(blk
);
1339 ret
= blk_co_do_preadv_part(blk
, offset
, bytes
, qiov
, qiov_offset
, flags
);
1340 blk_dec_in_flight(blk
);
1345 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1347 blk_co_do_pwritev_part(BlockBackend
*blk
, int64_t offset
, int64_t bytes
,
1348 QEMUIOVector
*qiov
, size_t qiov_offset
,
1349 BdrvRequestFlags flags
)
1352 BlockDriverState
*bs
;
1355 blk_wait_while_drained(blk
);
1357 /* Call blk_bs() only after waiting, the graph may have changed */
1359 trace_blk_co_pwritev(blk
, bs
, offset
, bytes
, flags
);
1361 ret
= blk_check_byte_request(blk
, offset
, bytes
);
1366 bdrv_inc_in_flight(bs
);
1367 /* throttling disk I/O */
1368 if (blk
->public.throttle_group_member
.throttle_state
) {
1369 throttle_group_co_io_limits_intercept(&blk
->public.throttle_group_member
,
1373 if (!blk
->enable_write_cache
) {
1374 flags
|= BDRV_REQ_FUA
;
1377 ret
= bdrv_co_pwritev_part(blk
->root
, offset
, bytes
, qiov
, qiov_offset
,
1379 bdrv_dec_in_flight(bs
);
1383 int coroutine_fn
blk_co_pwritev_part(BlockBackend
*blk
, int64_t offset
,
1385 QEMUIOVector
*qiov
, size_t qiov_offset
,
1386 BdrvRequestFlags flags
)
1391 blk_inc_in_flight(blk
);
1392 ret
= blk_co_do_pwritev_part(blk
, offset
, bytes
, qiov
, qiov_offset
, flags
);
1393 blk_dec_in_flight(blk
);
1398 int coroutine_fn
blk_co_pwritev(BlockBackend
*blk
, int64_t offset
,
1399 int64_t bytes
, QEMUIOVector
*qiov
,
1400 BdrvRequestFlags flags
)
1403 return blk_co_pwritev_part(blk
, offset
, bytes
, qiov
, 0, flags
);
1406 typedef struct BlkRwCo
{
1411 BdrvRequestFlags flags
;
1414 int blk_make_zero(BlockBackend
*blk
, BdrvRequestFlags flags
)
1416 GLOBAL_STATE_CODE();
1417 return bdrv_make_zero(blk
->root
, flags
);
1420 void blk_inc_in_flight(BlockBackend
*blk
)
1423 qatomic_inc(&blk
->in_flight
);
1426 void blk_dec_in_flight(BlockBackend
*blk
)
1429 qatomic_dec(&blk
->in_flight
);
1433 static void error_callback_bh(void *opaque
)
1435 struct BlockBackendAIOCB
*acb
= opaque
;
1437 blk_dec_in_flight(acb
->blk
);
1438 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
1439 qemu_aio_unref(acb
);
1442 BlockAIOCB
*blk_abort_aio_request(BlockBackend
*blk
,
1443 BlockCompletionFunc
*cb
,
1444 void *opaque
, int ret
)
1446 struct BlockBackendAIOCB
*acb
;
1449 blk_inc_in_flight(blk
);
1450 acb
= blk_aio_get(&block_backend_aiocb_info
, blk
, cb
, opaque
);
1454 replay_bh_schedule_oneshot_event(blk_get_aio_context(blk
),
1455 error_callback_bh
, acb
);
1456 return &acb
->common
;
1459 typedef struct BlkAioEmAIOCB
{
1466 static AioContext
*blk_aio_em_aiocb_get_aio_context(BlockAIOCB
*acb_
)
1468 BlkAioEmAIOCB
*acb
= container_of(acb_
, BlkAioEmAIOCB
, common
);
1470 return blk_get_aio_context(acb
->rwco
.blk
);
1473 static const AIOCBInfo blk_aio_em_aiocb_info
= {
1474 .aiocb_size
= sizeof(BlkAioEmAIOCB
),
1475 .get_aio_context
= blk_aio_em_aiocb_get_aio_context
,
1478 static void blk_aio_complete(BlkAioEmAIOCB
*acb
)
1480 if (acb
->has_returned
) {
1481 acb
->common
.cb(acb
->common
.opaque
, acb
->rwco
.ret
);
1482 blk_dec_in_flight(acb
->rwco
.blk
);
1483 qemu_aio_unref(acb
);
1487 static void blk_aio_complete_bh(void *opaque
)
1489 BlkAioEmAIOCB
*acb
= opaque
;
1490 assert(acb
->has_returned
);
1491 blk_aio_complete(acb
);
1494 static BlockAIOCB
*blk_aio_prwv(BlockBackend
*blk
, int64_t offset
,
1496 void *iobuf
, CoroutineEntry co_entry
,
1497 BdrvRequestFlags flags
,
1498 BlockCompletionFunc
*cb
, void *opaque
)
1503 blk_inc_in_flight(blk
);
1504 acb
= blk_aio_get(&blk_aio_em_aiocb_info
, blk
, cb
, opaque
);
1505 acb
->rwco
= (BlkRwCo
) {
1513 acb
->has_returned
= false;
1515 co
= qemu_coroutine_create(co_entry
, acb
);
1516 bdrv_coroutine_enter(blk_bs(blk
), co
);
1518 acb
->has_returned
= true;
1519 if (acb
->rwco
.ret
!= NOT_DONE
) {
1520 replay_bh_schedule_oneshot_event(blk_get_aio_context(blk
),
1521 blk_aio_complete_bh
, acb
);
1524 return &acb
->common
;
1527 static void blk_aio_read_entry(void *opaque
)
1529 BlkAioEmAIOCB
*acb
= opaque
;
1530 BlkRwCo
*rwco
= &acb
->rwco
;
1531 QEMUIOVector
*qiov
= rwco
->iobuf
;
1533 assert(qiov
->size
== acb
->bytes
);
1534 rwco
->ret
= blk_co_do_preadv_part(rwco
->blk
, rwco
->offset
, acb
->bytes
, qiov
,
1536 blk_aio_complete(acb
);
1539 static void blk_aio_write_entry(void *opaque
)
1541 BlkAioEmAIOCB
*acb
= opaque
;
1542 BlkRwCo
*rwco
= &acb
->rwco
;
1543 QEMUIOVector
*qiov
= rwco
->iobuf
;
1545 assert(!qiov
|| qiov
->size
== acb
->bytes
);
1546 rwco
->ret
= blk_co_do_pwritev_part(rwco
->blk
, rwco
->offset
, acb
->bytes
,
1547 qiov
, 0, rwco
->flags
);
1548 blk_aio_complete(acb
);
1551 BlockAIOCB
*blk_aio_pwrite_zeroes(BlockBackend
*blk
, int64_t offset
,
1552 int64_t bytes
, BdrvRequestFlags flags
,
1553 BlockCompletionFunc
*cb
, void *opaque
)
1556 return blk_aio_prwv(blk
, offset
, bytes
, NULL
, blk_aio_write_entry
,
1557 flags
| BDRV_REQ_ZERO_WRITE
, cb
, opaque
);
1560 int64_t blk_getlength(BlockBackend
*blk
)
1563 if (!blk_is_available(blk
)) {
1567 return bdrv_getlength(blk_bs(blk
));
1570 void blk_get_geometry(BlockBackend
*blk
, uint64_t *nb_sectors_ptr
)
1574 *nb_sectors_ptr
= 0;
1576 bdrv_get_geometry(blk_bs(blk
), nb_sectors_ptr
);
1580 int64_t blk_nb_sectors(BlockBackend
*blk
)
1583 if (!blk_is_available(blk
)) {
1587 return bdrv_nb_sectors(blk_bs(blk
));
1590 BlockAIOCB
*blk_aio_preadv(BlockBackend
*blk
, int64_t offset
,
1591 QEMUIOVector
*qiov
, BdrvRequestFlags flags
,
1592 BlockCompletionFunc
*cb
, void *opaque
)
1595 assert((uint64_t)qiov
->size
<= INT64_MAX
);
1596 return blk_aio_prwv(blk
, offset
, qiov
->size
, qiov
,
1597 blk_aio_read_entry
, flags
, cb
, opaque
);
1600 BlockAIOCB
*blk_aio_pwritev(BlockBackend
*blk
, int64_t offset
,
1601 QEMUIOVector
*qiov
, BdrvRequestFlags flags
,
1602 BlockCompletionFunc
*cb
, void *opaque
)
1605 assert((uint64_t)qiov
->size
<= INT64_MAX
);
1606 return blk_aio_prwv(blk
, offset
, qiov
->size
, qiov
,
1607 blk_aio_write_entry
, flags
, cb
, opaque
);
1610 void blk_aio_cancel(BlockAIOCB
*acb
)
1612 GLOBAL_STATE_CODE();
1613 bdrv_aio_cancel(acb
);
1616 void blk_aio_cancel_async(BlockAIOCB
*acb
)
1619 bdrv_aio_cancel_async(acb
);
1622 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1624 blk_co_do_ioctl(BlockBackend
*blk
, unsigned long int req
, void *buf
)
1628 blk_wait_while_drained(blk
);
1630 if (!blk_is_available(blk
)) {
1634 return bdrv_co_ioctl(blk_bs(blk
), req
, buf
);
1637 int blk_ioctl(BlockBackend
*blk
, unsigned long int req
, void *buf
)
1642 blk_inc_in_flight(blk
);
1643 ret
= blk_do_ioctl(blk
, req
, buf
);
1644 blk_dec_in_flight(blk
);
1649 static void blk_aio_ioctl_entry(void *opaque
)
1651 BlkAioEmAIOCB
*acb
= opaque
;
1652 BlkRwCo
*rwco
= &acb
->rwco
;
1654 rwco
->ret
= blk_co_do_ioctl(rwco
->blk
, rwco
->offset
, rwco
->iobuf
);
1656 blk_aio_complete(acb
);
1659 BlockAIOCB
*blk_aio_ioctl(BlockBackend
*blk
, unsigned long int req
, void *buf
,
1660 BlockCompletionFunc
*cb
, void *opaque
)
1663 return blk_aio_prwv(blk
, req
, 0, buf
, blk_aio_ioctl_entry
, 0, cb
, opaque
);
1666 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1668 blk_co_do_pdiscard(BlockBackend
*blk
, int64_t offset
, int64_t bytes
)
1673 blk_wait_while_drained(blk
);
1675 ret
= blk_check_byte_request(blk
, offset
, bytes
);
1680 return bdrv_co_pdiscard(blk
->root
, offset
, bytes
);
1683 static void blk_aio_pdiscard_entry(void *opaque
)
1685 BlkAioEmAIOCB
*acb
= opaque
;
1686 BlkRwCo
*rwco
= &acb
->rwco
;
1688 rwco
->ret
= blk_co_do_pdiscard(rwco
->blk
, rwco
->offset
, acb
->bytes
);
1689 blk_aio_complete(acb
);
1692 BlockAIOCB
*blk_aio_pdiscard(BlockBackend
*blk
,
1693 int64_t offset
, int64_t bytes
,
1694 BlockCompletionFunc
*cb
, void *opaque
)
1697 return blk_aio_prwv(blk
, offset
, bytes
, NULL
, blk_aio_pdiscard_entry
, 0,
1701 int coroutine_fn
blk_co_pdiscard(BlockBackend
*blk
, int64_t offset
,
1707 blk_inc_in_flight(blk
);
1708 ret
= blk_co_do_pdiscard(blk
, offset
, bytes
);
1709 blk_dec_in_flight(blk
);
1714 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1715 int coroutine_fn
blk_co_do_flush(BlockBackend
*blk
)
1717 blk_wait_while_drained(blk
);
1720 if (!blk_is_available(blk
)) {
1724 return bdrv_co_flush(blk_bs(blk
));
1727 static void blk_aio_flush_entry(void *opaque
)
1729 BlkAioEmAIOCB
*acb
= opaque
;
1730 BlkRwCo
*rwco
= &acb
->rwco
;
1732 rwco
->ret
= blk_co_do_flush(rwco
->blk
);
1733 blk_aio_complete(acb
);
1736 BlockAIOCB
*blk_aio_flush(BlockBackend
*blk
,
1737 BlockCompletionFunc
*cb
, void *opaque
)
1740 return blk_aio_prwv(blk
, 0, 0, NULL
, blk_aio_flush_entry
, 0, cb
, opaque
);
1743 int coroutine_fn
blk_co_flush(BlockBackend
*blk
)
1748 blk_inc_in_flight(blk
);
1749 ret
= blk_co_do_flush(blk
);
1750 blk_dec_in_flight(blk
);
1755 void blk_drain(BlockBackend
*blk
)
1757 BlockDriverState
*bs
= blk_bs(blk
);
1758 GLOBAL_STATE_CODE();
1762 bdrv_drained_begin(bs
);
1765 /* We may have -ENOMEDIUM completions in flight */
1766 AIO_WAIT_WHILE(blk_get_aio_context(blk
),
1767 qatomic_mb_read(&blk
->in_flight
) > 0);
1770 bdrv_drained_end(bs
);
1775 void blk_drain_all(void)
1777 BlockBackend
*blk
= NULL
;
1779 GLOBAL_STATE_CODE();
1781 bdrv_drain_all_begin();
1783 while ((blk
= blk_all_next(blk
)) != NULL
) {
1784 AioContext
*ctx
= blk_get_aio_context(blk
);
1786 aio_context_acquire(ctx
);
1788 /* We may have -ENOMEDIUM completions in flight */
1789 AIO_WAIT_WHILE(ctx
, qatomic_mb_read(&blk
->in_flight
) > 0);
1791 aio_context_release(ctx
);
1794 bdrv_drain_all_end();
1797 void blk_set_on_error(BlockBackend
*blk
, BlockdevOnError on_read_error
,
1798 BlockdevOnError on_write_error
)
1800 GLOBAL_STATE_CODE();
1801 blk
->on_read_error
= on_read_error
;
1802 blk
->on_write_error
= on_write_error
;
1805 BlockdevOnError
blk_get_on_error(BlockBackend
*blk
, bool is_read
)
1808 return is_read
? blk
->on_read_error
: blk
->on_write_error
;
1811 BlockErrorAction
blk_get_error_action(BlockBackend
*blk
, bool is_read
,
1814 BlockdevOnError on_err
= blk_get_on_error(blk
, is_read
);
1818 case BLOCKDEV_ON_ERROR_ENOSPC
:
1819 return (error
== ENOSPC
) ?
1820 BLOCK_ERROR_ACTION_STOP
: BLOCK_ERROR_ACTION_REPORT
;
1821 case BLOCKDEV_ON_ERROR_STOP
:
1822 return BLOCK_ERROR_ACTION_STOP
;
1823 case BLOCKDEV_ON_ERROR_REPORT
:
1824 return BLOCK_ERROR_ACTION_REPORT
;
1825 case BLOCKDEV_ON_ERROR_IGNORE
:
1826 return BLOCK_ERROR_ACTION_IGNORE
;
1827 case BLOCKDEV_ON_ERROR_AUTO
:
1833 static void send_qmp_error_event(BlockBackend
*blk
,
1834 BlockErrorAction action
,
1835 bool is_read
, int error
)
1837 IoOperationType optype
;
1838 BlockDriverState
*bs
= blk_bs(blk
);
1840 optype
= is_read
? IO_OPERATION_TYPE_READ
: IO_OPERATION_TYPE_WRITE
;
1841 qapi_event_send_block_io_error(blk_name(blk
), !!bs
,
1842 bs
? bdrv_get_node_name(bs
) : NULL
, optype
,
1843 action
, blk_iostatus_is_enabled(blk
),
1844 error
== ENOSPC
, strerror(error
));
1847 /* This is done by device models because, while the block layer knows
1848 * about the error, it does not know whether an operation comes from
1849 * the device or the block layer (from a job, for example).
1851 void blk_error_action(BlockBackend
*blk
, BlockErrorAction action
,
1852 bool is_read
, int error
)
1857 if (action
== BLOCK_ERROR_ACTION_STOP
) {
1858 /* First set the iostatus, so that "info block" returns an iostatus
1859 * that matches the events raised so far (an additional error iostatus
1860 * is fine, but not a lost one).
1862 blk_iostatus_set_err(blk
, error
);
1864 /* Then raise the request to stop the VM and the event.
1865 * qemu_system_vmstop_request_prepare has two effects. First,
1866 * it ensures that the STOP event always comes after the
1867 * BLOCK_IO_ERROR event. Second, it ensures that even if management
1868 * can observe the STOP event and do a "cont" before the STOP
1869 * event is issued, the VM will not stop. In this case, vm_start()
1870 * also ensures that the STOP/RESUME pair of events is emitted.
1872 qemu_system_vmstop_request_prepare();
1873 send_qmp_error_event(blk
, action
, is_read
, error
);
1874 qemu_system_vmstop_request(RUN_STATE_IO_ERROR
);
1876 send_qmp_error_event(blk
, action
, is_read
, error
);
1881 * Returns true if the BlockBackend can support taking write permissions
1882 * (because its root node is not read-only).
1884 bool blk_supports_write_perm(BlockBackend
*blk
)
1886 BlockDriverState
*bs
= blk_bs(blk
);
1887 GLOBAL_STATE_CODE();
1890 return !bdrv_is_read_only(bs
);
1892 return blk
->root_state
.open_flags
& BDRV_O_RDWR
;
1897 * Returns true if the BlockBackend can be written to in its current
1898 * configuration (i.e. if write permission have been requested)
1900 bool blk_is_writable(BlockBackend
*blk
)
1903 return blk
->perm
& BLK_PERM_WRITE
;
1906 bool blk_is_sg(BlockBackend
*blk
)
1908 BlockDriverState
*bs
= blk_bs(blk
);
1909 GLOBAL_STATE_CODE();
1915 return bdrv_is_sg(bs
);
1918 bool blk_enable_write_cache(BlockBackend
*blk
)
1921 return blk
->enable_write_cache
;
1924 void blk_set_enable_write_cache(BlockBackend
*blk
, bool wce
)
1926 GLOBAL_STATE_CODE();
1927 blk
->enable_write_cache
= wce
;
1930 void blk_activate(BlockBackend
*blk
, Error
**errp
)
1932 BlockDriverState
*bs
= blk_bs(blk
);
1933 GLOBAL_STATE_CODE();
1936 error_setg(errp
, "Device '%s' has no medium", blk
->name
);
1940 bdrv_activate(bs
, errp
);
1943 bool blk_is_inserted(BlockBackend
*blk
)
1945 BlockDriverState
*bs
= blk_bs(blk
);
1948 return bs
&& bdrv_is_inserted(bs
);
1951 bool blk_is_available(BlockBackend
*blk
)
1954 return blk_is_inserted(blk
) && !blk_dev_is_tray_open(blk
);
1957 void blk_lock_medium(BlockBackend
*blk
, bool locked
)
1959 BlockDriverState
*bs
= blk_bs(blk
);
1963 bdrv_lock_medium(bs
, locked
);
1967 void blk_eject(BlockBackend
*blk
, bool eject_flag
)
1969 BlockDriverState
*bs
= blk_bs(blk
);
1974 bdrv_eject(bs
, eject_flag
);
1977 /* Whether or not we ejected on the backend,
1978 * the frontend experienced a tray event. */
1979 id
= blk_get_attached_dev_id(blk
);
1980 qapi_event_send_device_tray_moved(blk_name(blk
), id
,
1985 int blk_get_flags(BlockBackend
*blk
)
1987 BlockDriverState
*bs
= blk_bs(blk
);
1988 GLOBAL_STATE_CODE();
1991 return bdrv_get_flags(bs
);
1993 return blk
->root_state
.open_flags
;
1997 /* Returns the minimum request alignment, in bytes; guaranteed nonzero */
1998 uint32_t blk_get_request_alignment(BlockBackend
*blk
)
2000 BlockDriverState
*bs
= blk_bs(blk
);
2002 return bs
? bs
->bl
.request_alignment
: BDRV_SECTOR_SIZE
;
2005 /* Returns the maximum hardware transfer length, in bytes; guaranteed nonzero */
2006 uint64_t blk_get_max_hw_transfer(BlockBackend
*blk
)
2008 BlockDriverState
*bs
= blk_bs(blk
);
2009 uint64_t max
= INT_MAX
;
2013 max
= MIN_NON_ZERO(max
, bs
->bl
.max_hw_transfer
);
2014 max
= MIN_NON_ZERO(max
, bs
->bl
.max_transfer
);
2016 return ROUND_DOWN(max
, blk_get_request_alignment(blk
));
2019 /* Returns the maximum transfer length, in bytes; guaranteed nonzero */
2020 uint32_t blk_get_max_transfer(BlockBackend
*blk
)
2022 BlockDriverState
*bs
= blk_bs(blk
);
2023 uint32_t max
= INT_MAX
;
2027 max
= MIN_NON_ZERO(max
, bs
->bl
.max_transfer
);
2029 return ROUND_DOWN(max
, blk_get_request_alignment(blk
));
2032 int blk_get_max_hw_iov(BlockBackend
*blk
)
2035 return MIN_NON_ZERO(blk
->root
->bs
->bl
.max_hw_iov
,
2036 blk
->root
->bs
->bl
.max_iov
);
2039 int blk_get_max_iov(BlockBackend
*blk
)
2042 return blk
->root
->bs
->bl
.max_iov
;
2045 void *blk_try_blockalign(BlockBackend
*blk
, size_t size
)
2048 return qemu_try_blockalign(blk
? blk_bs(blk
) : NULL
, size
);
2051 void *blk_blockalign(BlockBackend
*blk
, size_t size
)
2054 return qemu_blockalign(blk
? blk_bs(blk
) : NULL
, size
);
2057 bool blk_op_is_blocked(BlockBackend
*blk
, BlockOpType op
, Error
**errp
)
2059 BlockDriverState
*bs
= blk_bs(blk
);
2060 GLOBAL_STATE_CODE();
2066 return bdrv_op_is_blocked(bs
, op
, errp
);
2069 void blk_op_unblock(BlockBackend
*blk
, BlockOpType op
, Error
*reason
)
2071 BlockDriverState
*bs
= blk_bs(blk
);
2072 GLOBAL_STATE_CODE();
2075 bdrv_op_unblock(bs
, op
, reason
);
2079 void blk_op_block_all(BlockBackend
*blk
, Error
*reason
)
2081 BlockDriverState
*bs
= blk_bs(blk
);
2082 GLOBAL_STATE_CODE();
2085 bdrv_op_block_all(bs
, reason
);
2089 void blk_op_unblock_all(BlockBackend
*blk
, Error
*reason
)
2091 BlockDriverState
*bs
= blk_bs(blk
);
2092 GLOBAL_STATE_CODE();
2095 bdrv_op_unblock_all(bs
, reason
);
2099 AioContext
*blk_get_aio_context(BlockBackend
*blk
)
2101 BlockDriverState
*bs
= blk_bs(blk
);
2105 AioContext
*ctx
= bdrv_get_aio_context(blk_bs(blk
));
2106 assert(ctx
== blk
->ctx
);
2112 static AioContext
*blk_aiocb_get_aio_context(BlockAIOCB
*acb
)
2114 BlockBackendAIOCB
*blk_acb
= DO_UPCAST(BlockBackendAIOCB
, common
, acb
);
2115 return blk_get_aio_context(blk_acb
->blk
);
2118 static int blk_do_set_aio_context(BlockBackend
*blk
, AioContext
*new_context
,
2119 bool update_root_node
, Error
**errp
)
2121 BlockDriverState
*bs
= blk_bs(blk
);
2122 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
2128 if (update_root_node
) {
2129 ret
= bdrv_child_try_set_aio_context(bs
, new_context
, blk
->root
,
2136 if (tgm
->throttle_state
) {
2137 bdrv_drained_begin(bs
);
2138 throttle_group_detach_aio_context(tgm
);
2139 throttle_group_attach_aio_context(tgm
, new_context
);
2140 bdrv_drained_end(bs
);
2146 blk
->ctx
= new_context
;
2150 int blk_set_aio_context(BlockBackend
*blk
, AioContext
*new_context
,
2153 GLOBAL_STATE_CODE();
2154 return blk_do_set_aio_context(blk
, new_context
, true, errp
);
2157 static bool blk_root_can_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
2158 GSList
**ignore
, Error
**errp
)
2160 BlockBackend
*blk
= child
->opaque
;
2162 if (blk
->allow_aio_context_change
) {
2166 /* Only manually created BlockBackends that are not attached to anything
2167 * can change their AioContext without updating their user. */
2168 if (!blk
->name
|| blk
->dev
) {
2169 /* TODO Add BB name/QOM path */
2170 error_setg(errp
, "Cannot change iothread of active block backend");
2177 static void blk_root_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
2180 BlockBackend
*blk
= child
->opaque
;
2181 blk_do_set_aio_context(blk
, ctx
, false, &error_abort
);
2184 void blk_add_aio_context_notifier(BlockBackend
*blk
,
2185 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
2186 void (*detach_aio_context
)(void *opaque
), void *opaque
)
2188 BlockBackendAioNotifier
*notifier
;
2189 BlockDriverState
*bs
= blk_bs(blk
);
2190 GLOBAL_STATE_CODE();
2192 notifier
= g_new(BlockBackendAioNotifier
, 1);
2193 notifier
->attached_aio_context
= attached_aio_context
;
2194 notifier
->detach_aio_context
= detach_aio_context
;
2195 notifier
->opaque
= opaque
;
2196 QLIST_INSERT_HEAD(&blk
->aio_notifiers
, notifier
, list
);
2199 bdrv_add_aio_context_notifier(bs
, attached_aio_context
,
2200 detach_aio_context
, opaque
);
2204 void blk_remove_aio_context_notifier(BlockBackend
*blk
,
2205 void (*attached_aio_context
)(AioContext
*,
2207 void (*detach_aio_context
)(void *),
2210 BlockBackendAioNotifier
*notifier
;
2211 BlockDriverState
*bs
= blk_bs(blk
);
2213 GLOBAL_STATE_CODE();
2216 bdrv_remove_aio_context_notifier(bs
, attached_aio_context
,
2217 detach_aio_context
, opaque
);
2220 QLIST_FOREACH(notifier
, &blk
->aio_notifiers
, list
) {
2221 if (notifier
->attached_aio_context
== attached_aio_context
&&
2222 notifier
->detach_aio_context
== detach_aio_context
&&
2223 notifier
->opaque
== opaque
) {
2224 QLIST_REMOVE(notifier
, list
);
2233 void blk_add_remove_bs_notifier(BlockBackend
*blk
, Notifier
*notify
)
2235 GLOBAL_STATE_CODE();
2236 notifier_list_add(&blk
->remove_bs_notifiers
, notify
);
2239 void blk_add_insert_bs_notifier(BlockBackend
*blk
, Notifier
*notify
)
2241 GLOBAL_STATE_CODE();
2242 notifier_list_add(&blk
->insert_bs_notifiers
, notify
);
2245 void blk_io_plug(BlockBackend
*blk
)
2247 BlockDriverState
*bs
= blk_bs(blk
);
2255 void blk_io_unplug(BlockBackend
*blk
)
2257 BlockDriverState
*bs
= blk_bs(blk
);
2265 BlockAcctStats
*blk_get_stats(BlockBackend
*blk
)
2271 void *blk_aio_get(const AIOCBInfo
*aiocb_info
, BlockBackend
*blk
,
2272 BlockCompletionFunc
*cb
, void *opaque
)
2275 return qemu_aio_get(aiocb_info
, blk_bs(blk
), cb
, opaque
);
2278 int coroutine_fn
blk_co_pwrite_zeroes(BlockBackend
*blk
, int64_t offset
,
2279 int64_t bytes
, BdrvRequestFlags flags
)
2282 return blk_co_pwritev(blk
, offset
, bytes
, NULL
,
2283 flags
| BDRV_REQ_ZERO_WRITE
);
2286 int coroutine_fn
blk_co_pwrite_compressed(BlockBackend
*blk
, int64_t offset
,
2287 int64_t bytes
, const void *buf
)
2289 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, buf
, bytes
);
2291 return blk_co_pwritev_part(blk
, offset
, bytes
, &qiov
, 0,
2292 BDRV_REQ_WRITE_COMPRESSED
);
2295 int blk_truncate(BlockBackend
*blk
, int64_t offset
, bool exact
,
2296 PreallocMode prealloc
, BdrvRequestFlags flags
, Error
**errp
)
2299 if (!blk_is_available(blk
)) {
2300 error_setg(errp
, "No medium inserted");
2304 return bdrv_truncate(blk
->root
, offset
, exact
, prealloc
, flags
, errp
);
2307 int blk_save_vmstate(BlockBackend
*blk
, const uint8_t *buf
,
2308 int64_t pos
, int size
)
2311 GLOBAL_STATE_CODE();
2313 if (!blk_is_available(blk
)) {
2317 ret
= bdrv_save_vmstate(blk_bs(blk
), buf
, pos
, size
);
2322 if (ret
== size
&& !blk
->enable_write_cache
) {
2323 ret
= bdrv_flush(blk_bs(blk
));
2326 return ret
< 0 ? ret
: size
;
2329 int blk_load_vmstate(BlockBackend
*blk
, uint8_t *buf
, int64_t pos
, int size
)
2331 GLOBAL_STATE_CODE();
2332 if (!blk_is_available(blk
)) {
2336 return bdrv_load_vmstate(blk_bs(blk
), buf
, pos
, size
);
2339 int blk_probe_blocksizes(BlockBackend
*blk
, BlockSizes
*bsz
)
2341 GLOBAL_STATE_CODE();
2342 if (!blk_is_available(blk
)) {
2346 return bdrv_probe_blocksizes(blk_bs(blk
), bsz
);
2349 int blk_probe_geometry(BlockBackend
*blk
, HDGeometry
*geo
)
2351 GLOBAL_STATE_CODE();
2352 if (!blk_is_available(blk
)) {
2356 return bdrv_probe_geometry(blk_bs(blk
), geo
);
2360 * Updates the BlockBackendRootState object with data from the currently
2361 * attached BlockDriverState.
2363 void blk_update_root_state(BlockBackend
*blk
)
2365 GLOBAL_STATE_CODE();
2368 blk
->root_state
.open_flags
= blk
->root
->bs
->open_flags
;
2369 blk
->root_state
.detect_zeroes
= blk
->root
->bs
->detect_zeroes
;
2373 * Returns the detect-zeroes setting to be used for bdrv_open() of a
2374 * BlockDriverState which is supposed to inherit the root state.
2376 bool blk_get_detect_zeroes_from_root_state(BlockBackend
*blk
)
2378 GLOBAL_STATE_CODE();
2379 return blk
->root_state
.detect_zeroes
;
2383 * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
2384 * supposed to inherit the root state.
2386 int blk_get_open_flags_from_root_state(BlockBackend
*blk
)
2388 GLOBAL_STATE_CODE();
2389 return blk
->root_state
.open_flags
;
2392 BlockBackendRootState
*blk_get_root_state(BlockBackend
*blk
)
2394 GLOBAL_STATE_CODE();
2395 return &blk
->root_state
;
2398 int blk_commit_all(void)
2400 BlockBackend
*blk
= NULL
;
2401 GLOBAL_STATE_CODE();
2403 while ((blk
= blk_all_next(blk
)) != NULL
) {
2404 AioContext
*aio_context
= blk_get_aio_context(blk
);
2405 BlockDriverState
*unfiltered_bs
= bdrv_skip_filters(blk_bs(blk
));
2407 aio_context_acquire(aio_context
);
2408 if (blk_is_inserted(blk
) && bdrv_cow_child(unfiltered_bs
)) {
2411 ret
= bdrv_commit(unfiltered_bs
);
2413 aio_context_release(aio_context
);
2417 aio_context_release(aio_context
);
2423 /* throttling disk I/O limits */
2424 void blk_set_io_limits(BlockBackend
*blk
, ThrottleConfig
*cfg
)
2426 GLOBAL_STATE_CODE();
2427 throttle_group_config(&blk
->public.throttle_group_member
, cfg
);
2430 void blk_io_limits_disable(BlockBackend
*blk
)
2432 BlockDriverState
*bs
= blk_bs(blk
);
2433 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
2434 assert(tgm
->throttle_state
);
2435 GLOBAL_STATE_CODE();
2438 bdrv_drained_begin(bs
);
2440 throttle_group_unregister_tgm(tgm
);
2442 bdrv_drained_end(bs
);
2447 /* should be called before blk_set_io_limits if a limit is set */
2448 void blk_io_limits_enable(BlockBackend
*blk
, const char *group
)
2450 assert(!blk
->public.throttle_group_member
.throttle_state
);
2451 GLOBAL_STATE_CODE();
2452 throttle_group_register_tgm(&blk
->public.throttle_group_member
,
2453 group
, blk_get_aio_context(blk
));
2456 void blk_io_limits_update_group(BlockBackend
*blk
, const char *group
)
2458 GLOBAL_STATE_CODE();
2459 /* this BB is not part of any group */
2460 if (!blk
->public.throttle_group_member
.throttle_state
) {
2464 /* this BB is a part of the same group than the one we want */
2465 if (!g_strcmp0(throttle_group_get_name(&blk
->public.throttle_group_member
),
2470 /* need to change the group this bs belong to */
2471 blk_io_limits_disable(blk
);
2472 blk_io_limits_enable(blk
, group
);
2475 static void blk_root_drained_begin(BdrvChild
*child
)
2477 BlockBackend
*blk
= child
->opaque
;
2478 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
2480 if (++blk
->quiesce_counter
== 1) {
2481 if (blk
->dev_ops
&& blk
->dev_ops
->drained_begin
) {
2482 blk
->dev_ops
->drained_begin(blk
->dev_opaque
);
2486 /* Note that blk->root may not be accessible here yet if we are just
2487 * attaching to a BlockDriverState that is drained. Use child instead. */
2489 if (qatomic_fetch_inc(&tgm
->io_limits_disabled
) == 0) {
2490 throttle_group_restart_tgm(tgm
);
2494 static bool blk_root_drained_poll(BdrvChild
*child
)
2496 BlockBackend
*blk
= child
->opaque
;
2498 assert(blk
->quiesce_counter
);
2500 if (blk
->dev_ops
&& blk
->dev_ops
->drained_poll
) {
2501 busy
= blk
->dev_ops
->drained_poll(blk
->dev_opaque
);
2503 return busy
|| !!blk
->in_flight
;
2506 static void blk_root_drained_end(BdrvChild
*child
, int *drained_end_counter
)
2508 BlockBackend
*blk
= child
->opaque
;
2509 assert(blk
->quiesce_counter
);
2511 assert(blk
->public.throttle_group_member
.io_limits_disabled
);
2512 qatomic_dec(&blk
->public.throttle_group_member
.io_limits_disabled
);
2514 if (--blk
->quiesce_counter
== 0) {
2515 if (blk
->dev_ops
&& blk
->dev_ops
->drained_end
) {
2516 blk
->dev_ops
->drained_end(blk
->dev_opaque
);
2518 while (qemu_co_enter_next(&blk
->queued_requests
, NULL
)) {
2519 /* Resume all queued requests */
2524 void blk_register_buf(BlockBackend
*blk
, void *host
, size_t size
)
2526 GLOBAL_STATE_CODE();
2527 bdrv_register_buf(blk_bs(blk
), host
, size
);
2530 void blk_unregister_buf(BlockBackend
*blk
, void *host
)
2532 GLOBAL_STATE_CODE();
2533 bdrv_unregister_buf(blk_bs(blk
), host
);
2536 int coroutine_fn
blk_co_copy_range(BlockBackend
*blk_in
, int64_t off_in
,
2537 BlockBackend
*blk_out
, int64_t off_out
,
2538 int64_t bytes
, BdrvRequestFlags read_flags
,
2539 BdrvRequestFlags write_flags
)
2544 r
= blk_check_byte_request(blk_in
, off_in
, bytes
);
2548 r
= blk_check_byte_request(blk_out
, off_out
, bytes
);
2552 return bdrv_co_copy_range(blk_in
->root
, off_in
,
2553 blk_out
->root
, off_out
,
2554 bytes
, read_flags
, write_flags
);
2557 const BdrvChild
*blk_root(BlockBackend
*blk
)
2559 GLOBAL_STATE_CODE();
2563 int blk_make_empty(BlockBackend
*blk
, Error
**errp
)
2565 GLOBAL_STATE_CODE();
2566 if (!blk_is_available(blk
)) {
2567 error_setg(errp
, "No medium inserted");
2571 return bdrv_make_empty(blk
->root
, errp
);