4 * Copyright (C) 2014-2016 Red Hat, Inc.
7 * Markus Armbruster <armbru@redhat.com>,
9 * This work is licensed under the terms of the GNU LGPL, version 2.1
10 * or later. See the COPYING.LIB file in the top-level directory.
13 #include "qemu/osdep.h"
14 #include "sysemu/block-backend.h"
15 #include "block/block_int.h"
16 #include "block/blockjob.h"
17 #include "block/throttle-groups.h"
18 #include "hw/qdev-core.h"
19 #include "sysemu/blockdev.h"
20 #include "sysemu/sysemu.h"
21 #include "qapi/error.h"
22 #include "qapi/qapi-events-block.h"
24 #include "qemu/option.h"
26 #include "migration/misc.h"
28 /* Number of coroutines to reserve per attached device model */
29 #define COROUTINE_POOL_RESERVATION 64
31 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
33 static AioContext
*blk_aiocb_get_aio_context(BlockAIOCB
*acb
);
35 typedef struct BlockBackendAioNotifier
{
36 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
);
37 void (*detach_aio_context
)(void *opaque
);
39 QLIST_ENTRY(BlockBackendAioNotifier
) list
;
40 } BlockBackendAioNotifier
;
47 DriveInfo
*legacy_dinfo
; /* null unless created by drive_new() */
48 QTAILQ_ENTRY(BlockBackend
) link
; /* for block_backends */
49 QTAILQ_ENTRY(BlockBackend
) monitor_link
; /* for monitor_block_backends */
50 BlockBackendPublic
public;
52 DeviceState
*dev
; /* attached device model, if any */
53 const BlockDevOps
*dev_ops
;
56 /* the block size for which the guest device expects atomicity */
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 NotifierList remove_bs_notifiers
, insert_bs_notifiers
;
80 QLIST_HEAD(, BlockBackendAioNotifier
) aio_notifiers
;
83 VMChangeStateEntry
*vmsh
;
84 bool force_allow_inactivate
;
86 /* Number of in-flight aio requests. BlockDriverState also counts
87 * in-flight requests but aio requests can exist even when blk->root is
88 * NULL, so we cannot rely on its counter for that case.
89 * Accessed with atomic ops.
91 unsigned int in_flight
;
94 typedef struct BlockBackendAIOCB
{
100 static const AIOCBInfo block_backend_aiocb_info
= {
101 .get_aio_context
= blk_aiocb_get_aio_context
,
102 .aiocb_size
= sizeof(BlockBackendAIOCB
),
105 static void drive_info_del(DriveInfo
*dinfo
);
106 static BlockBackend
*bdrv_first_blk(BlockDriverState
*bs
);
108 /* All BlockBackends */
109 static QTAILQ_HEAD(, BlockBackend
) block_backends
=
110 QTAILQ_HEAD_INITIALIZER(block_backends
);
112 /* All BlockBackends referenced by the monitor and which are iterated through by
114 static QTAILQ_HEAD(, BlockBackend
) monitor_block_backends
=
115 QTAILQ_HEAD_INITIALIZER(monitor_block_backends
);
117 static void blk_root_inherit_options(int *child_flags
, QDict
*child_options
,
118 int parent_flags
, QDict
*parent_options
)
120 /* We're not supposed to call this function for root nodes */
123 static void blk_root_drained_begin(BdrvChild
*child
);
124 static bool blk_root_drained_poll(BdrvChild
*child
);
125 static void blk_root_drained_end(BdrvChild
*child
, int *drained_end_counter
);
127 static void blk_root_change_media(BdrvChild
*child
, bool load
);
128 static void blk_root_resize(BdrvChild
*child
);
130 static bool blk_root_can_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
131 GSList
**ignore
, Error
**errp
);
132 static void blk_root_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
135 static char *blk_root_get_parent_desc(BdrvChild
*child
)
137 BlockBackend
*blk
= child
->opaque
;
141 return g_strdup(blk
->name
);
144 dev_id
= blk_get_attached_dev_id(blk
);
148 /* TODO Callback into the BB owner for something more detailed */
150 return g_strdup("a block device");
154 static const char *blk_root_get_name(BdrvChild
*child
)
156 return blk_name(child
->opaque
);
159 static void blk_vm_state_changed(void *opaque
, int running
, RunState state
)
161 Error
*local_err
= NULL
;
162 BlockBackend
*blk
= opaque
;
164 if (state
== RUN_STATE_INMIGRATE
) {
168 qemu_del_vm_change_state_handler(blk
->vmsh
);
170 blk_set_perm(blk
, blk
->perm
, blk
->shared_perm
, &local_err
);
172 error_report_err(local_err
);
177 * Notifies the user of the BlockBackend that migration has completed. qdev
178 * devices can tighten their permissions in response (specifically revoke
179 * shared write permissions that we needed for storage migration).
181 * If an error is returned, the VM cannot be allowed to be resumed.
183 static void blk_root_activate(BdrvChild
*child
, Error
**errp
)
185 BlockBackend
*blk
= child
->opaque
;
186 Error
*local_err
= NULL
;
188 if (!blk
->disable_perm
) {
192 blk
->disable_perm
= false;
194 blk_set_perm(blk
, blk
->perm
, BLK_PERM_ALL
, &local_err
);
196 error_propagate(errp
, local_err
);
197 blk
->disable_perm
= true;
201 if (runstate_check(RUN_STATE_INMIGRATE
)) {
202 /* Activation can happen when migration process is still active, for
203 * example when nbd_server_add is called during non-shared storage
204 * migration. Defer the shared_perm update to migration completion. */
206 blk
->vmsh
= qemu_add_vm_change_state_handler(blk_vm_state_changed
,
212 blk_set_perm(blk
, blk
->perm
, blk
->shared_perm
, &local_err
);
214 error_propagate(errp
, local_err
);
215 blk
->disable_perm
= true;
220 void blk_set_force_allow_inactivate(BlockBackend
*blk
)
222 blk
->force_allow_inactivate
= true;
225 static bool blk_can_inactivate(BlockBackend
*blk
)
227 /* If it is a guest device, inactivate is ok. */
228 if (blk
->dev
|| blk_name(blk
)[0]) {
232 /* Inactivating means no more writes to the image can be done,
233 * even if those writes would be changes invisible to the
234 * guest. For block job BBs that satisfy this, we can just allow
235 * it. This is the case for mirror job source, which is required
236 * by libvirt non-shared block migration. */
237 if (!(blk
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
))) {
241 return blk
->force_allow_inactivate
;
244 static int blk_root_inactivate(BdrvChild
*child
)
246 BlockBackend
*blk
= child
->opaque
;
248 if (blk
->disable_perm
) {
252 if (!blk_can_inactivate(blk
)) {
256 blk
->disable_perm
= true;
258 bdrv_child_try_set_perm(blk
->root
, 0, BLK_PERM_ALL
, &error_abort
);
264 static void blk_root_attach(BdrvChild
*child
)
266 BlockBackend
*blk
= child
->opaque
;
267 BlockBackendAioNotifier
*notifier
;
269 trace_blk_root_attach(child
, blk
, child
->bs
);
271 QLIST_FOREACH(notifier
, &blk
->aio_notifiers
, list
) {
272 bdrv_add_aio_context_notifier(child
->bs
,
273 notifier
->attached_aio_context
,
274 notifier
->detach_aio_context
,
279 static void blk_root_detach(BdrvChild
*child
)
281 BlockBackend
*blk
= child
->opaque
;
282 BlockBackendAioNotifier
*notifier
;
284 trace_blk_root_detach(child
, blk
, child
->bs
);
286 QLIST_FOREACH(notifier
, &blk
->aio_notifiers
, list
) {
287 bdrv_remove_aio_context_notifier(child
->bs
,
288 notifier
->attached_aio_context
,
289 notifier
->detach_aio_context
,
294 static const BdrvChildRole child_root
= {
295 .inherit_options
= blk_root_inherit_options
,
297 .change_media
= blk_root_change_media
,
298 .resize
= blk_root_resize
,
299 .get_name
= blk_root_get_name
,
300 .get_parent_desc
= blk_root_get_parent_desc
,
302 .drained_begin
= blk_root_drained_begin
,
303 .drained_poll
= blk_root_drained_poll
,
304 .drained_end
= blk_root_drained_end
,
306 .activate
= blk_root_activate
,
307 .inactivate
= blk_root_inactivate
,
309 .attach
= blk_root_attach
,
310 .detach
= blk_root_detach
,
312 .can_set_aio_ctx
= blk_root_can_set_aio_ctx
,
313 .set_aio_ctx
= blk_root_set_aio_ctx
,
317 * Create a new BlockBackend with a reference count of one.
319 * @perm is a bitmasks of BLK_PERM_* constants which describes the permissions
320 * to request for a block driver node that is attached to this BlockBackend.
321 * @shared_perm is a bitmask which describes which permissions may be granted
322 * to other users of the attached node.
323 * Both sets of permissions can be changed later using blk_set_perm().
325 * Return the new BlockBackend on success, null on failure.
327 BlockBackend
*blk_new(AioContext
*ctx
, uint64_t perm
, uint64_t shared_perm
)
331 blk
= g_new0(BlockBackend
, 1);
335 blk
->shared_perm
= shared_perm
;
336 blk_set_enable_write_cache(blk
, true);
338 blk
->on_read_error
= BLOCKDEV_ON_ERROR_REPORT
;
339 blk
->on_write_error
= BLOCKDEV_ON_ERROR_ENOSPC
;
341 block_acct_init(&blk
->stats
);
343 notifier_list_init(&blk
->remove_bs_notifiers
);
344 notifier_list_init(&blk
->insert_bs_notifiers
);
345 QLIST_INIT(&blk
->aio_notifiers
);
347 QTAILQ_INSERT_TAIL(&block_backends
, blk
, link
);
352 * Creates a new BlockBackend, opens a new BlockDriverState, and connects both.
353 * The new BlockBackend is in the main AioContext.
355 * Just as with bdrv_open(), after having called this function the reference to
356 * @options belongs to the block layer (even on failure).
358 * TODO: Remove @filename and @flags; it should be possible to specify a whole
359 * BDS tree just by specifying the @options QDict (or @reference,
360 * alternatively). At the time of adding this function, this is not possible,
361 * though, so callers of this function have to be able to specify @filename and
364 BlockBackend
*blk_new_open(const char *filename
, const char *reference
,
365 QDict
*options
, int flags
, Error
**errp
)
368 BlockDriverState
*bs
;
371 /* blk_new_open() is mainly used in .bdrv_create implementations and the
372 * tools where sharing isn't a concern because the BDS stays private, so we
373 * just request permission according to the flags.
375 * The exceptions are xen_disk and blockdev_init(); in these cases, the
376 * caller of blk_new_open() doesn't make use of the permissions, but they
377 * shouldn't hurt either. We can still share everything here because the
378 * guest devices will add their own blockers if they can't share. */
379 if ((flags
& BDRV_O_NO_IO
) == 0) {
380 perm
|= BLK_PERM_CONSISTENT_READ
;
381 if (flags
& BDRV_O_RDWR
) {
382 perm
|= BLK_PERM_WRITE
;
385 if (flags
& BDRV_O_RESIZE
) {
386 perm
|= BLK_PERM_RESIZE
;
389 blk
= blk_new(qemu_get_aio_context(), perm
, BLK_PERM_ALL
);
390 bs
= bdrv_open(filename
, reference
, options
, flags
, errp
);
396 blk
->root
= bdrv_root_attach_child(bs
, "root", &child_root
, blk
->ctx
,
397 perm
, BLK_PERM_ALL
, blk
, errp
);
406 static void blk_delete(BlockBackend
*blk
)
408 assert(!blk
->refcnt
);
411 if (blk
->public.throttle_group_member
.throttle_state
) {
412 blk_io_limits_disable(blk
);
418 qemu_del_vm_change_state_handler(blk
->vmsh
);
421 assert(QLIST_EMPTY(&blk
->remove_bs_notifiers
.notifiers
));
422 assert(QLIST_EMPTY(&blk
->insert_bs_notifiers
.notifiers
));
423 assert(QLIST_EMPTY(&blk
->aio_notifiers
));
424 QTAILQ_REMOVE(&block_backends
, blk
, link
);
425 drive_info_del(blk
->legacy_dinfo
);
426 block_acct_cleanup(&blk
->stats
);
430 static void drive_info_del(DriveInfo
*dinfo
)
435 qemu_opts_del(dinfo
->opts
);
439 int blk_get_refcnt(BlockBackend
*blk
)
441 return blk
? blk
->refcnt
: 0;
445 * Increment @blk's reference count.
446 * @blk must not be null.
448 void blk_ref(BlockBackend
*blk
)
450 assert(blk
->refcnt
> 0);
455 * Decrement @blk's reference count.
456 * If this drops it to zero, destroy @blk.
457 * For convenience, do nothing if @blk is null.
459 void blk_unref(BlockBackend
*blk
)
462 assert(blk
->refcnt
> 0);
463 if (blk
->refcnt
> 1) {
467 /* blk_drain() cannot resurrect blk, nobody held a reference */
468 assert(blk
->refcnt
== 1);
476 * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
477 * ones which are hidden (i.e. are not referenced by the monitor).
479 BlockBackend
*blk_all_next(BlockBackend
*blk
)
481 return blk
? QTAILQ_NEXT(blk
, link
)
482 : QTAILQ_FIRST(&block_backends
);
485 void blk_remove_all_bs(void)
487 BlockBackend
*blk
= NULL
;
489 while ((blk
= blk_all_next(blk
)) != NULL
) {
490 AioContext
*ctx
= blk_get_aio_context(blk
);
492 aio_context_acquire(ctx
);
496 aio_context_release(ctx
);
501 * Return the monitor-owned BlockBackend after @blk.
502 * If @blk is null, return the first one.
503 * Else, return @blk's next sibling, which may be null.
505 * To iterate over all BlockBackends, do
506 * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
510 BlockBackend
*blk_next(BlockBackend
*blk
)
512 return blk
? QTAILQ_NEXT(blk
, monitor_link
)
513 : QTAILQ_FIRST(&monitor_block_backends
);
516 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
517 * the monitor or attached to a BlockBackend */
518 BlockDriverState
*bdrv_next(BdrvNextIterator
*it
)
520 BlockDriverState
*bs
, *old_bs
;
522 /* Must be called from the main loop */
523 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
525 /* First, return all root nodes of BlockBackends. In order to avoid
526 * returning a BDS twice when multiple BBs refer to it, we only return it
527 * if the BB is the first one in the parent list of the BDS. */
528 if (it
->phase
== BDRV_NEXT_BACKEND_ROOTS
) {
529 BlockBackend
*old_blk
= it
->blk
;
531 old_bs
= old_blk
? blk_bs(old_blk
) : NULL
;
534 it
->blk
= blk_all_next(it
->blk
);
535 bs
= it
->blk
? blk_bs(it
->blk
) : NULL
;
536 } while (it
->blk
&& (bs
== NULL
|| bdrv_first_blk(bs
) != it
->blk
));
548 it
->phase
= BDRV_NEXT_MONITOR_OWNED
;
553 /* Then return the monitor-owned BDSes without a BB attached. Ignore all
554 * BDSes that are attached to a BlockBackend here; they have been handled
555 * by the above block already */
557 it
->bs
= bdrv_next_monitor_owned(it
->bs
);
559 } while (bs
&& bdrv_has_blk(bs
));
569 static void bdrv_next_reset(BdrvNextIterator
*it
)
571 *it
= (BdrvNextIterator
) {
572 .phase
= BDRV_NEXT_BACKEND_ROOTS
,
576 BlockDriverState
*bdrv_first(BdrvNextIterator
*it
)
579 return bdrv_next(it
);
582 /* Must be called when aborting a bdrv_next() iteration before
583 * bdrv_next() returns NULL */
584 void bdrv_next_cleanup(BdrvNextIterator
*it
)
586 /* Must be called from the main loop */
587 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
589 if (it
->phase
== BDRV_NEXT_BACKEND_ROOTS
) {
591 bdrv_unref(blk_bs(it
->blk
));
602 * Add a BlockBackend into the list of backends referenced by the monitor, with
603 * the given @name acting as the handle for the monitor.
604 * Strictly for use by blockdev.c.
606 * @name must not be null or empty.
608 * Returns true on success and false on failure. In the latter case, an Error
609 * object is returned through @errp.
611 bool monitor_add_blk(BlockBackend
*blk
, const char *name
, Error
**errp
)
614 assert(name
&& name
[0]);
616 if (!id_wellformed(name
)) {
617 error_setg(errp
, "Invalid device name");
620 if (blk_by_name(name
)) {
621 error_setg(errp
, "Device with id '%s' already exists", name
);
624 if (bdrv_find_node(name
)) {
626 "Device name '%s' conflicts with an existing node name",
631 blk
->name
= g_strdup(name
);
632 QTAILQ_INSERT_TAIL(&monitor_block_backends
, blk
, monitor_link
);
637 * Remove a BlockBackend from the list of backends referenced by the monitor.
638 * Strictly for use by blockdev.c.
640 void monitor_remove_blk(BlockBackend
*blk
)
646 QTAILQ_REMOVE(&monitor_block_backends
, blk
, monitor_link
);
652 * Return @blk's name, a non-null string.
653 * Returns an empty string iff @blk is not referenced by the monitor.
655 const char *blk_name(const BlockBackend
*blk
)
657 return blk
->name
?: "";
661 * Return the BlockBackend with name @name if it exists, else null.
662 * @name must not be null.
664 BlockBackend
*blk_by_name(const char *name
)
666 BlockBackend
*blk
= NULL
;
669 while ((blk
= blk_next(blk
)) != NULL
) {
670 if (!strcmp(name
, blk
->name
)) {
678 * Return the BlockDriverState attached to @blk if any, else null.
680 BlockDriverState
*blk_bs(BlockBackend
*blk
)
682 return blk
->root
? blk
->root
->bs
: NULL
;
685 static BlockBackend
*bdrv_first_blk(BlockDriverState
*bs
)
688 QLIST_FOREACH(child
, &bs
->parents
, next_parent
) {
689 if (child
->role
== &child_root
) {
690 return child
->opaque
;
698 * Returns true if @bs has an associated BlockBackend.
700 bool bdrv_has_blk(BlockDriverState
*bs
)
702 return bdrv_first_blk(bs
) != NULL
;
706 * Returns true if @bs has only BlockBackends as parents.
708 bool bdrv_is_root_node(BlockDriverState
*bs
)
712 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
713 if (c
->role
!= &child_root
) {
722 * Return @blk's DriveInfo if any, else null.
724 DriveInfo
*blk_legacy_dinfo(BlockBackend
*blk
)
726 return blk
->legacy_dinfo
;
730 * Set @blk's DriveInfo to @dinfo, and return it.
731 * @blk must not have a DriveInfo set already.
732 * No other BlockBackend may have the same DriveInfo set.
734 DriveInfo
*blk_set_legacy_dinfo(BlockBackend
*blk
, DriveInfo
*dinfo
)
736 assert(!blk
->legacy_dinfo
);
737 return blk
->legacy_dinfo
= dinfo
;
741 * Return the BlockBackend with DriveInfo @dinfo.
744 BlockBackend
*blk_by_legacy_dinfo(DriveInfo
*dinfo
)
746 BlockBackend
*blk
= NULL
;
748 while ((blk
= blk_next(blk
)) != NULL
) {
749 if (blk
->legacy_dinfo
== dinfo
) {
757 * Returns a pointer to the publicly accessible fields of @blk.
759 BlockBackendPublic
*blk_get_public(BlockBackend
*blk
)
765 * Returns a BlockBackend given the associated @public fields.
767 BlockBackend
*blk_by_public(BlockBackendPublic
*public)
769 return container_of(public, BlockBackend
, public);
773 * Disassociates the currently associated BlockDriverState from @blk.
775 void blk_remove_bs(BlockBackend
*blk
)
777 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
778 BlockDriverState
*bs
;
780 notifier_list_notify(&blk
->remove_bs_notifiers
, blk
);
781 if (tgm
->throttle_state
) {
783 bdrv_drained_begin(bs
);
784 throttle_group_detach_aio_context(tgm
);
785 throttle_group_attach_aio_context(tgm
, qemu_get_aio_context());
786 bdrv_drained_end(bs
);
789 blk_update_root_state(blk
);
791 /* bdrv_root_unref_child() will cause blk->root to become stale and may
792 * switch to a completion coroutine later on. Let's drain all I/O here
793 * to avoid that and a potential QEMU crash.
796 bdrv_root_unref_child(blk
->root
);
801 * Associates a new BlockDriverState with @blk.
803 int blk_insert_bs(BlockBackend
*blk
, BlockDriverState
*bs
, Error
**errp
)
805 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
807 blk
->root
= bdrv_root_attach_child(bs
, "root", &child_root
, blk
->ctx
,
808 blk
->perm
, blk
->shared_perm
, blk
, errp
);
809 if (blk
->root
== NULL
) {
813 notifier_list_notify(&blk
->insert_bs_notifiers
, blk
);
814 if (tgm
->throttle_state
) {
815 throttle_group_detach_aio_context(tgm
);
816 throttle_group_attach_aio_context(tgm
, bdrv_get_aio_context(bs
));
823 * Sets the permission bitmasks that the user of the BlockBackend needs.
825 int blk_set_perm(BlockBackend
*blk
, uint64_t perm
, uint64_t shared_perm
,
830 if (blk
->root
&& !blk
->disable_perm
) {
831 ret
= bdrv_child_try_set_perm(blk
->root
, perm
, shared_perm
, errp
);
838 blk
->shared_perm
= shared_perm
;
843 void blk_get_perm(BlockBackend
*blk
, uint64_t *perm
, uint64_t *shared_perm
)
846 *shared_perm
= blk
->shared_perm
;
850 * Attach device model @dev to @blk.
851 * Return 0 on success, -EBUSY when a device model is attached already.
853 int blk_attach_dev(BlockBackend
*blk
, DeviceState
*dev
)
859 /* While migration is still incoming, we don't need to apply the
860 * permissions of guest device BlockBackends. We might still have a block
861 * job or NBD server writing to the image for storage migration. */
862 if (runstate_check(RUN_STATE_INMIGRATE
)) {
863 blk
->disable_perm
= true;
868 blk_iostatus_reset(blk
);
874 * Detach device model @dev from @blk.
875 * @dev must be currently attached to @blk.
877 void blk_detach_dev(BlockBackend
*blk
, DeviceState
*dev
)
879 assert(blk
->dev
== dev
);
882 blk
->dev_opaque
= NULL
;
883 blk
->guest_block_size
= 512;
884 blk_set_perm(blk
, 0, BLK_PERM_ALL
, &error_abort
);
889 * Return the device model attached to @blk if any, else null.
891 DeviceState
*blk_get_attached_dev(BlockBackend
*blk
)
896 /* Return the qdev ID, or if no ID is assigned the QOM path, of the block
897 * device attached to the BlockBackend. */
898 char *blk_get_attached_dev_id(BlockBackend
*blk
)
900 DeviceState
*dev
= blk
->dev
;
904 } else if (dev
->id
) {
905 return g_strdup(dev
->id
);
908 return object_get_canonical_path(OBJECT(dev
)) ?: g_strdup("");
912 * Return the BlockBackend which has the device model @dev attached if it
915 * @dev must not be null.
917 BlockBackend
*blk_by_dev(void *dev
)
919 BlockBackend
*blk
= NULL
;
922 while ((blk
= blk_all_next(blk
)) != NULL
) {
923 if (blk
->dev
== dev
) {
931 * Set @blk's device model callbacks to @ops.
932 * @opaque is the opaque argument to pass to the callbacks.
933 * This is for use by device models.
935 void blk_set_dev_ops(BlockBackend
*blk
, const BlockDevOps
*ops
,
939 blk
->dev_opaque
= opaque
;
941 /* Are we currently quiesced? Should we enforce this right now? */
942 if (blk
->quiesce_counter
&& ops
->drained_begin
) {
943 ops
->drained_begin(opaque
);
948 * Notify @blk's attached device model of media change.
950 * If @load is true, notify of media load. This action can fail, meaning that
951 * the medium cannot be loaded. @errp is set then.
953 * If @load is false, notify of media eject. This can never fail.
955 * Also send DEVICE_TRAY_MOVED events as appropriate.
957 void blk_dev_change_media_cb(BlockBackend
*blk
, bool load
, Error
**errp
)
959 if (blk
->dev_ops
&& blk
->dev_ops
->change_media_cb
) {
960 bool tray_was_open
, tray_is_open
;
961 Error
*local_err
= NULL
;
963 tray_was_open
= blk_dev_is_tray_open(blk
);
964 blk
->dev_ops
->change_media_cb(blk
->dev_opaque
, load
, &local_err
);
966 assert(load
== true);
967 error_propagate(errp
, local_err
);
970 tray_is_open
= blk_dev_is_tray_open(blk
);
972 if (tray_was_open
!= tray_is_open
) {
973 char *id
= blk_get_attached_dev_id(blk
);
974 qapi_event_send_device_tray_moved(blk_name(blk
), id
, tray_is_open
);
980 static void blk_root_change_media(BdrvChild
*child
, bool load
)
982 blk_dev_change_media_cb(child
->opaque
, load
, NULL
);
986 * Does @blk's attached device model have removable media?
987 * %true if no device model is attached.
989 bool blk_dev_has_removable_media(BlockBackend
*blk
)
991 return !blk
->dev
|| (blk
->dev_ops
&& blk
->dev_ops
->change_media_cb
);
995 * Does @blk's attached device model have a tray?
997 bool blk_dev_has_tray(BlockBackend
*blk
)
999 return blk
->dev_ops
&& blk
->dev_ops
->is_tray_open
;
1003 * Notify @blk's attached device model of a media eject request.
1004 * If @force is true, the medium is about to be yanked out forcefully.
1006 void blk_dev_eject_request(BlockBackend
*blk
, bool force
)
1008 if (blk
->dev_ops
&& blk
->dev_ops
->eject_request_cb
) {
1009 blk
->dev_ops
->eject_request_cb(blk
->dev_opaque
, force
);
1014 * Does @blk's attached device model have a tray, and is it open?
1016 bool blk_dev_is_tray_open(BlockBackend
*blk
)
1018 if (blk_dev_has_tray(blk
)) {
1019 return blk
->dev_ops
->is_tray_open(blk
->dev_opaque
);
1025 * Does @blk's attached device model have the medium locked?
1026 * %false if the device model has no such lock.
1028 bool blk_dev_is_medium_locked(BlockBackend
*blk
)
1030 if (blk
->dev_ops
&& blk
->dev_ops
->is_medium_locked
) {
1031 return blk
->dev_ops
->is_medium_locked(blk
->dev_opaque
);
1037 * Notify @blk's attached device model of a backend size change.
1039 static void blk_root_resize(BdrvChild
*child
)
1041 BlockBackend
*blk
= child
->opaque
;
1043 if (blk
->dev_ops
&& blk
->dev_ops
->resize_cb
) {
1044 blk
->dev_ops
->resize_cb(blk
->dev_opaque
);
1048 void blk_iostatus_enable(BlockBackend
*blk
)
1050 blk
->iostatus_enabled
= true;
1051 blk
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
1054 /* The I/O status is only enabled if the drive explicitly
1055 * enables it _and_ the VM is configured to stop on errors */
1056 bool blk_iostatus_is_enabled(const BlockBackend
*blk
)
1058 return (blk
->iostatus_enabled
&&
1059 (blk
->on_write_error
== BLOCKDEV_ON_ERROR_ENOSPC
||
1060 blk
->on_write_error
== BLOCKDEV_ON_ERROR_STOP
||
1061 blk
->on_read_error
== BLOCKDEV_ON_ERROR_STOP
));
1064 BlockDeviceIoStatus
blk_iostatus(const BlockBackend
*blk
)
1066 return blk
->iostatus
;
1069 void blk_iostatus_disable(BlockBackend
*blk
)
1071 blk
->iostatus_enabled
= false;
1074 void blk_iostatus_reset(BlockBackend
*blk
)
1076 if (blk_iostatus_is_enabled(blk
)) {
1077 blk
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
1081 void blk_iostatus_set_err(BlockBackend
*blk
, int error
)
1083 assert(blk_iostatus_is_enabled(blk
));
1084 if (blk
->iostatus
== BLOCK_DEVICE_IO_STATUS_OK
) {
1085 blk
->iostatus
= error
== ENOSPC
? BLOCK_DEVICE_IO_STATUS_NOSPACE
:
1086 BLOCK_DEVICE_IO_STATUS_FAILED
;
1090 void blk_set_allow_write_beyond_eof(BlockBackend
*blk
, bool allow
)
1092 blk
->allow_write_beyond_eof
= allow
;
1095 void blk_set_allow_aio_context_change(BlockBackend
*blk
, bool allow
)
1097 blk
->allow_aio_context_change
= allow
;
1100 static int blk_check_byte_request(BlockBackend
*blk
, int64_t offset
,
1105 if (size
> INT_MAX
) {
1109 if (!blk_is_available(blk
)) {
1117 if (!blk
->allow_write_beyond_eof
) {
1118 len
= blk_getlength(blk
);
1123 if (offset
> len
|| len
- offset
< size
) {
1131 int coroutine_fn
blk_co_preadv(BlockBackend
*blk
, int64_t offset
,
1132 unsigned int bytes
, QEMUIOVector
*qiov
,
1133 BdrvRequestFlags flags
)
1136 BlockDriverState
*bs
= blk_bs(blk
);
1138 trace_blk_co_preadv(blk
, bs
, offset
, bytes
, flags
);
1140 ret
= blk_check_byte_request(blk
, offset
, bytes
);
1145 bdrv_inc_in_flight(bs
);
1147 /* throttling disk I/O */
1148 if (blk
->public.throttle_group_member
.throttle_state
) {
1149 throttle_group_co_io_limits_intercept(&blk
->public.throttle_group_member
,
1153 ret
= bdrv_co_preadv(blk
->root
, offset
, bytes
, qiov
, flags
);
1154 bdrv_dec_in_flight(bs
);
1158 int coroutine_fn
blk_co_pwritev(BlockBackend
*blk
, int64_t offset
,
1159 unsigned int bytes
, QEMUIOVector
*qiov
,
1160 BdrvRequestFlags flags
)
1163 BlockDriverState
*bs
= blk_bs(blk
);
1165 trace_blk_co_pwritev(blk
, bs
, offset
, bytes
, flags
);
1167 ret
= blk_check_byte_request(blk
, offset
, bytes
);
1172 bdrv_inc_in_flight(bs
);
1173 /* throttling disk I/O */
1174 if (blk
->public.throttle_group_member
.throttle_state
) {
1175 throttle_group_co_io_limits_intercept(&blk
->public.throttle_group_member
,
1179 if (!blk
->enable_write_cache
) {
1180 flags
|= BDRV_REQ_FUA
;
1183 ret
= bdrv_co_pwritev(blk
->root
, offset
, bytes
, qiov
, flags
);
1184 bdrv_dec_in_flight(bs
);
1188 typedef struct BlkRwCo
{
1193 BdrvRequestFlags flags
;
1196 static void blk_read_entry(void *opaque
)
1198 BlkRwCo
*rwco
= opaque
;
1199 QEMUIOVector
*qiov
= rwco
->iobuf
;
1201 rwco
->ret
= blk_co_preadv(rwco
->blk
, rwco
->offset
, qiov
->size
,
1206 static void blk_write_entry(void *opaque
)
1208 BlkRwCo
*rwco
= opaque
;
1209 QEMUIOVector
*qiov
= rwco
->iobuf
;
1211 rwco
->ret
= blk_co_pwritev(rwco
->blk
, rwco
->offset
, qiov
->size
,
1216 static int blk_prw(BlockBackend
*blk
, int64_t offset
, uint8_t *buf
,
1217 int64_t bytes
, CoroutineEntry co_entry
,
1218 BdrvRequestFlags flags
)
1220 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, buf
, bytes
);
1229 if (qemu_in_coroutine()) {
1230 /* Fast-path if already in coroutine context */
1233 Coroutine
*co
= qemu_coroutine_create(co_entry
, &rwco
);
1234 bdrv_coroutine_enter(blk_bs(blk
), co
);
1235 BDRV_POLL_WHILE(blk_bs(blk
), rwco
.ret
== NOT_DONE
);
1241 int blk_pread_unthrottled(BlockBackend
*blk
, int64_t offset
, uint8_t *buf
,
1246 ret
= blk_check_byte_request(blk
, offset
, count
);
1251 blk_root_drained_begin(blk
->root
);
1252 ret
= blk_pread(blk
, offset
, buf
, count
);
1253 blk_root_drained_end(blk
->root
, NULL
);
1257 int blk_pwrite_zeroes(BlockBackend
*blk
, int64_t offset
,
1258 int bytes
, BdrvRequestFlags flags
)
1260 return blk_prw(blk
, offset
, NULL
, bytes
, blk_write_entry
,
1261 flags
| BDRV_REQ_ZERO_WRITE
);
1264 int blk_make_zero(BlockBackend
*blk
, BdrvRequestFlags flags
)
1266 return bdrv_make_zero(blk
->root
, flags
);
1269 void blk_inc_in_flight(BlockBackend
*blk
)
1271 atomic_inc(&blk
->in_flight
);
1274 void blk_dec_in_flight(BlockBackend
*blk
)
1276 atomic_dec(&blk
->in_flight
);
1280 static void error_callback_bh(void *opaque
)
1282 struct BlockBackendAIOCB
*acb
= opaque
;
1284 blk_dec_in_flight(acb
->blk
);
1285 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
1286 qemu_aio_unref(acb
);
1289 BlockAIOCB
*blk_abort_aio_request(BlockBackend
*blk
,
1290 BlockCompletionFunc
*cb
,
1291 void *opaque
, int ret
)
1293 struct BlockBackendAIOCB
*acb
;
1295 blk_inc_in_flight(blk
);
1296 acb
= blk_aio_get(&block_backend_aiocb_info
, blk
, cb
, opaque
);
1300 aio_bh_schedule_oneshot(blk_get_aio_context(blk
), error_callback_bh
, acb
);
1301 return &acb
->common
;
1304 typedef struct BlkAioEmAIOCB
{
1311 static const AIOCBInfo blk_aio_em_aiocb_info
= {
1312 .aiocb_size
= sizeof(BlkAioEmAIOCB
),
1315 static void blk_aio_complete(BlkAioEmAIOCB
*acb
)
1317 if (acb
->has_returned
) {
1318 acb
->common
.cb(acb
->common
.opaque
, acb
->rwco
.ret
);
1319 blk_dec_in_flight(acb
->rwco
.blk
);
1320 qemu_aio_unref(acb
);
1324 static void blk_aio_complete_bh(void *opaque
)
1326 BlkAioEmAIOCB
*acb
= opaque
;
1327 assert(acb
->has_returned
);
1328 blk_aio_complete(acb
);
1331 static BlockAIOCB
*blk_aio_prwv(BlockBackend
*blk
, int64_t offset
, int bytes
,
1332 void *iobuf
, CoroutineEntry co_entry
,
1333 BdrvRequestFlags flags
,
1334 BlockCompletionFunc
*cb
, void *opaque
)
1339 blk_inc_in_flight(blk
);
1340 acb
= blk_aio_get(&blk_aio_em_aiocb_info
, blk
, cb
, opaque
);
1341 acb
->rwco
= (BlkRwCo
) {
1349 acb
->has_returned
= false;
1351 co
= qemu_coroutine_create(co_entry
, acb
);
1352 bdrv_coroutine_enter(blk_bs(blk
), co
);
1354 acb
->has_returned
= true;
1355 if (acb
->rwco
.ret
!= NOT_DONE
) {
1356 aio_bh_schedule_oneshot(blk_get_aio_context(blk
),
1357 blk_aio_complete_bh
, acb
);
1360 return &acb
->common
;
1363 static void blk_aio_read_entry(void *opaque
)
1365 BlkAioEmAIOCB
*acb
= opaque
;
1366 BlkRwCo
*rwco
= &acb
->rwco
;
1367 QEMUIOVector
*qiov
= rwco
->iobuf
;
1369 assert(qiov
->size
== acb
->bytes
);
1370 rwco
->ret
= blk_co_preadv(rwco
->blk
, rwco
->offset
, acb
->bytes
,
1372 blk_aio_complete(acb
);
1375 static void blk_aio_write_entry(void *opaque
)
1377 BlkAioEmAIOCB
*acb
= opaque
;
1378 BlkRwCo
*rwco
= &acb
->rwco
;
1379 QEMUIOVector
*qiov
= rwco
->iobuf
;
1381 assert(!qiov
|| qiov
->size
== acb
->bytes
);
1382 rwco
->ret
= blk_co_pwritev(rwco
->blk
, rwco
->offset
, acb
->bytes
,
1384 blk_aio_complete(acb
);
1387 BlockAIOCB
*blk_aio_pwrite_zeroes(BlockBackend
*blk
, int64_t offset
,
1388 int count
, BdrvRequestFlags flags
,
1389 BlockCompletionFunc
*cb
, void *opaque
)
1391 return blk_aio_prwv(blk
, offset
, count
, NULL
, blk_aio_write_entry
,
1392 flags
| BDRV_REQ_ZERO_WRITE
, cb
, opaque
);
1395 int blk_pread(BlockBackend
*blk
, int64_t offset
, void *buf
, int count
)
1397 int ret
= blk_prw(blk
, offset
, buf
, count
, blk_read_entry
, 0);
1404 int blk_pwrite(BlockBackend
*blk
, int64_t offset
, const void *buf
, int count
,
1405 BdrvRequestFlags flags
)
1407 int ret
= blk_prw(blk
, offset
, (void *) buf
, count
, blk_write_entry
,
1415 int64_t blk_getlength(BlockBackend
*blk
)
1417 if (!blk_is_available(blk
)) {
1421 return bdrv_getlength(blk_bs(blk
));
1424 void blk_get_geometry(BlockBackend
*blk
, uint64_t *nb_sectors_ptr
)
1427 *nb_sectors_ptr
= 0;
1429 bdrv_get_geometry(blk_bs(blk
), nb_sectors_ptr
);
1433 int64_t blk_nb_sectors(BlockBackend
*blk
)
1435 if (!blk_is_available(blk
)) {
1439 return bdrv_nb_sectors(blk_bs(blk
));
1442 BlockAIOCB
*blk_aio_preadv(BlockBackend
*blk
, int64_t offset
,
1443 QEMUIOVector
*qiov
, BdrvRequestFlags flags
,
1444 BlockCompletionFunc
*cb
, void *opaque
)
1446 return blk_aio_prwv(blk
, offset
, qiov
->size
, qiov
,
1447 blk_aio_read_entry
, flags
, cb
, opaque
);
1450 BlockAIOCB
*blk_aio_pwritev(BlockBackend
*blk
, int64_t offset
,
1451 QEMUIOVector
*qiov
, BdrvRequestFlags flags
,
1452 BlockCompletionFunc
*cb
, void *opaque
)
1454 return blk_aio_prwv(blk
, offset
, qiov
->size
, qiov
,
1455 blk_aio_write_entry
, flags
, cb
, opaque
);
1458 static void blk_aio_flush_entry(void *opaque
)
1460 BlkAioEmAIOCB
*acb
= opaque
;
1461 BlkRwCo
*rwco
= &acb
->rwco
;
1463 rwco
->ret
= blk_co_flush(rwco
->blk
);
1464 blk_aio_complete(acb
);
1467 BlockAIOCB
*blk_aio_flush(BlockBackend
*blk
,
1468 BlockCompletionFunc
*cb
, void *opaque
)
1470 return blk_aio_prwv(blk
, 0, 0, NULL
, blk_aio_flush_entry
, 0, cb
, opaque
);
1473 static void blk_aio_pdiscard_entry(void *opaque
)
1475 BlkAioEmAIOCB
*acb
= opaque
;
1476 BlkRwCo
*rwco
= &acb
->rwco
;
1478 rwco
->ret
= blk_co_pdiscard(rwco
->blk
, rwco
->offset
, acb
->bytes
);
1479 blk_aio_complete(acb
);
1482 BlockAIOCB
*blk_aio_pdiscard(BlockBackend
*blk
,
1483 int64_t offset
, int bytes
,
1484 BlockCompletionFunc
*cb
, void *opaque
)
1486 return blk_aio_prwv(blk
, offset
, bytes
, NULL
, blk_aio_pdiscard_entry
, 0,
1490 void blk_aio_cancel(BlockAIOCB
*acb
)
1492 bdrv_aio_cancel(acb
);
1495 void blk_aio_cancel_async(BlockAIOCB
*acb
)
1497 bdrv_aio_cancel_async(acb
);
1500 int blk_co_ioctl(BlockBackend
*blk
, unsigned long int req
, void *buf
)
1502 if (!blk_is_available(blk
)) {
1506 return bdrv_co_ioctl(blk_bs(blk
), req
, buf
);
1509 static void blk_ioctl_entry(void *opaque
)
1511 BlkRwCo
*rwco
= opaque
;
1512 QEMUIOVector
*qiov
= rwco
->iobuf
;
1514 rwco
->ret
= blk_co_ioctl(rwco
->blk
, rwco
->offset
,
1515 qiov
->iov
[0].iov_base
);
1519 int blk_ioctl(BlockBackend
*blk
, unsigned long int req
, void *buf
)
1521 return blk_prw(blk
, req
, buf
, 0, blk_ioctl_entry
, 0);
1524 static void blk_aio_ioctl_entry(void *opaque
)
1526 BlkAioEmAIOCB
*acb
= opaque
;
1527 BlkRwCo
*rwco
= &acb
->rwco
;
1529 rwco
->ret
= blk_co_ioctl(rwco
->blk
, rwco
->offset
, rwco
->iobuf
);
1531 blk_aio_complete(acb
);
1534 BlockAIOCB
*blk_aio_ioctl(BlockBackend
*blk
, unsigned long int req
, void *buf
,
1535 BlockCompletionFunc
*cb
, void *opaque
)
1537 return blk_aio_prwv(blk
, req
, 0, buf
, blk_aio_ioctl_entry
, 0, cb
, opaque
);
1540 int blk_co_pdiscard(BlockBackend
*blk
, int64_t offset
, int bytes
)
1542 int ret
= blk_check_byte_request(blk
, offset
, bytes
);
1547 return bdrv_co_pdiscard(blk
->root
, offset
, bytes
);
1550 int blk_co_flush(BlockBackend
*blk
)
1552 if (!blk_is_available(blk
)) {
1556 return bdrv_co_flush(blk_bs(blk
));
1559 static void blk_flush_entry(void *opaque
)
1561 BlkRwCo
*rwco
= opaque
;
1562 rwco
->ret
= blk_co_flush(rwco
->blk
);
1566 int blk_flush(BlockBackend
*blk
)
1568 return blk_prw(blk
, 0, NULL
, 0, blk_flush_entry
, 0);
1571 void blk_drain(BlockBackend
*blk
)
1573 BlockDriverState
*bs
= blk_bs(blk
);
1576 bdrv_drained_begin(bs
);
1579 /* We may have -ENOMEDIUM completions in flight */
1580 AIO_WAIT_WHILE(blk_get_aio_context(blk
),
1581 atomic_mb_read(&blk
->in_flight
) > 0);
1584 bdrv_drained_end(bs
);
1588 void blk_drain_all(void)
1590 BlockBackend
*blk
= NULL
;
1592 bdrv_drain_all_begin();
1594 while ((blk
= blk_all_next(blk
)) != NULL
) {
1595 AioContext
*ctx
= blk_get_aio_context(blk
);
1597 aio_context_acquire(ctx
);
1599 /* We may have -ENOMEDIUM completions in flight */
1600 AIO_WAIT_WHILE(ctx
, atomic_mb_read(&blk
->in_flight
) > 0);
1602 aio_context_release(ctx
);
1605 bdrv_drain_all_end();
1608 void blk_set_on_error(BlockBackend
*blk
, BlockdevOnError on_read_error
,
1609 BlockdevOnError on_write_error
)
1611 blk
->on_read_error
= on_read_error
;
1612 blk
->on_write_error
= on_write_error
;
1615 BlockdevOnError
blk_get_on_error(BlockBackend
*blk
, bool is_read
)
1617 return is_read
? blk
->on_read_error
: blk
->on_write_error
;
1620 BlockErrorAction
blk_get_error_action(BlockBackend
*blk
, bool is_read
,
1623 BlockdevOnError on_err
= blk_get_on_error(blk
, is_read
);
1626 case BLOCKDEV_ON_ERROR_ENOSPC
:
1627 return (error
== ENOSPC
) ?
1628 BLOCK_ERROR_ACTION_STOP
: BLOCK_ERROR_ACTION_REPORT
;
1629 case BLOCKDEV_ON_ERROR_STOP
:
1630 return BLOCK_ERROR_ACTION_STOP
;
1631 case BLOCKDEV_ON_ERROR_REPORT
:
1632 return BLOCK_ERROR_ACTION_REPORT
;
1633 case BLOCKDEV_ON_ERROR_IGNORE
:
1634 return BLOCK_ERROR_ACTION_IGNORE
;
1635 case BLOCKDEV_ON_ERROR_AUTO
:
1641 static void send_qmp_error_event(BlockBackend
*blk
,
1642 BlockErrorAction action
,
1643 bool is_read
, int error
)
1645 IoOperationType optype
;
1646 BlockDriverState
*bs
= blk_bs(blk
);
1648 optype
= is_read
? IO_OPERATION_TYPE_READ
: IO_OPERATION_TYPE_WRITE
;
1649 qapi_event_send_block_io_error(blk_name(blk
), !!bs
,
1650 bs
? bdrv_get_node_name(bs
) : NULL
, optype
,
1651 action
, blk_iostatus_is_enabled(blk
),
1652 error
== ENOSPC
, strerror(error
));
1655 /* This is done by device models because, while the block layer knows
1656 * about the error, it does not know whether an operation comes from
1657 * the device or the block layer (from a job, for example).
1659 void blk_error_action(BlockBackend
*blk
, BlockErrorAction action
,
1660 bool is_read
, int error
)
1664 if (action
== BLOCK_ERROR_ACTION_STOP
) {
1665 /* First set the iostatus, so that "info block" returns an iostatus
1666 * that matches the events raised so far (an additional error iostatus
1667 * is fine, but not a lost one).
1669 blk_iostatus_set_err(blk
, error
);
1671 /* Then raise the request to stop the VM and the event.
1672 * qemu_system_vmstop_request_prepare has two effects. First,
1673 * it ensures that the STOP event always comes after the
1674 * BLOCK_IO_ERROR event. Second, it ensures that even if management
1675 * can observe the STOP event and do a "cont" before the STOP
1676 * event is issued, the VM will not stop. In this case, vm_start()
1677 * also ensures that the STOP/RESUME pair of events is emitted.
1679 qemu_system_vmstop_request_prepare();
1680 send_qmp_error_event(blk
, action
, is_read
, error
);
1681 qemu_system_vmstop_request(RUN_STATE_IO_ERROR
);
1683 send_qmp_error_event(blk
, action
, is_read
, error
);
1687 bool blk_is_read_only(BlockBackend
*blk
)
1689 BlockDriverState
*bs
= blk_bs(blk
);
1692 return bdrv_is_read_only(bs
);
1694 return blk
->root_state
.read_only
;
1698 bool blk_is_sg(BlockBackend
*blk
)
1700 BlockDriverState
*bs
= blk_bs(blk
);
1706 return bdrv_is_sg(bs
);
1709 bool blk_enable_write_cache(BlockBackend
*blk
)
1711 return blk
->enable_write_cache
;
1714 void blk_set_enable_write_cache(BlockBackend
*blk
, bool wce
)
1716 blk
->enable_write_cache
= wce
;
1719 void blk_invalidate_cache(BlockBackend
*blk
, Error
**errp
)
1721 BlockDriverState
*bs
= blk_bs(blk
);
1724 error_setg(errp
, "Device '%s' has no medium", blk
->name
);
1728 bdrv_invalidate_cache(bs
, errp
);
1731 bool blk_is_inserted(BlockBackend
*blk
)
1733 BlockDriverState
*bs
= blk_bs(blk
);
1735 return bs
&& bdrv_is_inserted(bs
);
1738 bool blk_is_available(BlockBackend
*blk
)
1740 return blk_is_inserted(blk
) && !blk_dev_is_tray_open(blk
);
1743 void blk_lock_medium(BlockBackend
*blk
, bool locked
)
1745 BlockDriverState
*bs
= blk_bs(blk
);
1748 bdrv_lock_medium(bs
, locked
);
1752 void blk_eject(BlockBackend
*blk
, bool eject_flag
)
1754 BlockDriverState
*bs
= blk_bs(blk
);
1758 bdrv_eject(bs
, eject_flag
);
1761 /* Whether or not we ejected on the backend,
1762 * the frontend experienced a tray event. */
1763 id
= blk_get_attached_dev_id(blk
);
1764 qapi_event_send_device_tray_moved(blk_name(blk
), id
,
1769 int blk_get_flags(BlockBackend
*blk
)
1771 BlockDriverState
*bs
= blk_bs(blk
);
1774 return bdrv_get_flags(bs
);
1776 return blk
->root_state
.open_flags
;
1780 /* Returns the minimum request alignment, in bytes; guaranteed nonzero */
1781 uint32_t blk_get_request_alignment(BlockBackend
*blk
)
1783 BlockDriverState
*bs
= blk_bs(blk
);
1784 return bs
? bs
->bl
.request_alignment
: BDRV_SECTOR_SIZE
;
1787 /* Returns the maximum transfer length, in bytes; guaranteed nonzero */
1788 uint32_t blk_get_max_transfer(BlockBackend
*blk
)
1790 BlockDriverState
*bs
= blk_bs(blk
);
1794 max
= bs
->bl
.max_transfer
;
1796 return MIN_NON_ZERO(max
, INT_MAX
);
1799 int blk_get_max_iov(BlockBackend
*blk
)
1801 return blk
->root
->bs
->bl
.max_iov
;
1804 void blk_set_guest_block_size(BlockBackend
*blk
, int align
)
1806 blk
->guest_block_size
= align
;
1809 void *blk_try_blockalign(BlockBackend
*blk
, size_t size
)
1811 return qemu_try_blockalign(blk
? blk_bs(blk
) : NULL
, size
);
1814 void *blk_blockalign(BlockBackend
*blk
, size_t size
)
1816 return qemu_blockalign(blk
? blk_bs(blk
) : NULL
, size
);
1819 bool blk_op_is_blocked(BlockBackend
*blk
, BlockOpType op
, Error
**errp
)
1821 BlockDriverState
*bs
= blk_bs(blk
);
1827 return bdrv_op_is_blocked(bs
, op
, errp
);
1830 void blk_op_unblock(BlockBackend
*blk
, BlockOpType op
, Error
*reason
)
1832 BlockDriverState
*bs
= blk_bs(blk
);
1835 bdrv_op_unblock(bs
, op
, reason
);
1839 void blk_op_block_all(BlockBackend
*blk
, Error
*reason
)
1841 BlockDriverState
*bs
= blk_bs(blk
);
1844 bdrv_op_block_all(bs
, reason
);
1848 void blk_op_unblock_all(BlockBackend
*blk
, Error
*reason
)
1850 BlockDriverState
*bs
= blk_bs(blk
);
1853 bdrv_op_unblock_all(bs
, reason
);
1857 AioContext
*blk_get_aio_context(BlockBackend
*blk
)
1859 BlockDriverState
*bs
= blk_bs(blk
);
1862 AioContext
*ctx
= bdrv_get_aio_context(blk_bs(blk
));
1863 assert(ctx
== blk
->ctx
);
1869 static AioContext
*blk_aiocb_get_aio_context(BlockAIOCB
*acb
)
1871 BlockBackendAIOCB
*blk_acb
= DO_UPCAST(BlockBackendAIOCB
, common
, acb
);
1872 return blk_get_aio_context(blk_acb
->blk
);
1875 static int blk_do_set_aio_context(BlockBackend
*blk
, AioContext
*new_context
,
1876 bool update_root_node
, Error
**errp
)
1878 BlockDriverState
*bs
= blk_bs(blk
);
1879 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
1883 if (update_root_node
) {
1884 ret
= bdrv_child_try_set_aio_context(bs
, new_context
, blk
->root
,
1890 if (tgm
->throttle_state
) {
1891 bdrv_drained_begin(bs
);
1892 throttle_group_detach_aio_context(tgm
);
1893 throttle_group_attach_aio_context(tgm
, new_context
);
1894 bdrv_drained_end(bs
);
1898 blk
->ctx
= new_context
;
1902 int blk_set_aio_context(BlockBackend
*blk
, AioContext
*new_context
,
1905 return blk_do_set_aio_context(blk
, new_context
, true, errp
);
1908 static bool blk_root_can_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1909 GSList
**ignore
, Error
**errp
)
1911 BlockBackend
*blk
= child
->opaque
;
1913 if (blk
->allow_aio_context_change
) {
1917 /* Only manually created BlockBackends that are not attached to anything
1918 * can change their AioContext without updating their user. */
1919 if (!blk
->name
|| blk
->dev
) {
1920 /* TODO Add BB name/QOM path */
1921 error_setg(errp
, "Cannot change iothread of active block backend");
1928 static void blk_root_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1931 BlockBackend
*blk
= child
->opaque
;
1932 blk_do_set_aio_context(blk
, ctx
, false, &error_abort
);
1935 void blk_add_aio_context_notifier(BlockBackend
*blk
,
1936 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
1937 void (*detach_aio_context
)(void *opaque
), void *opaque
)
1939 BlockBackendAioNotifier
*notifier
;
1940 BlockDriverState
*bs
= blk_bs(blk
);
1942 notifier
= g_new(BlockBackendAioNotifier
, 1);
1943 notifier
->attached_aio_context
= attached_aio_context
;
1944 notifier
->detach_aio_context
= detach_aio_context
;
1945 notifier
->opaque
= opaque
;
1946 QLIST_INSERT_HEAD(&blk
->aio_notifiers
, notifier
, list
);
1949 bdrv_add_aio_context_notifier(bs
, attached_aio_context
,
1950 detach_aio_context
, opaque
);
1954 void blk_remove_aio_context_notifier(BlockBackend
*blk
,
1955 void (*attached_aio_context
)(AioContext
*,
1957 void (*detach_aio_context
)(void *),
1960 BlockBackendAioNotifier
*notifier
;
1961 BlockDriverState
*bs
= blk_bs(blk
);
1964 bdrv_remove_aio_context_notifier(bs
, attached_aio_context
,
1965 detach_aio_context
, opaque
);
1968 QLIST_FOREACH(notifier
, &blk
->aio_notifiers
, list
) {
1969 if (notifier
->attached_aio_context
== attached_aio_context
&&
1970 notifier
->detach_aio_context
== detach_aio_context
&&
1971 notifier
->opaque
== opaque
) {
1972 QLIST_REMOVE(notifier
, list
);
1981 void blk_add_remove_bs_notifier(BlockBackend
*blk
, Notifier
*notify
)
1983 notifier_list_add(&blk
->remove_bs_notifiers
, notify
);
1986 void blk_add_insert_bs_notifier(BlockBackend
*blk
, Notifier
*notify
)
1988 notifier_list_add(&blk
->insert_bs_notifiers
, notify
);
1991 void blk_io_plug(BlockBackend
*blk
)
1993 BlockDriverState
*bs
= blk_bs(blk
);
2000 void blk_io_unplug(BlockBackend
*blk
)
2002 BlockDriverState
*bs
= blk_bs(blk
);
2009 BlockAcctStats
*blk_get_stats(BlockBackend
*blk
)
2014 void *blk_aio_get(const AIOCBInfo
*aiocb_info
, BlockBackend
*blk
,
2015 BlockCompletionFunc
*cb
, void *opaque
)
2017 return qemu_aio_get(aiocb_info
, blk_bs(blk
), cb
, opaque
);
2020 int coroutine_fn
blk_co_pwrite_zeroes(BlockBackend
*blk
, int64_t offset
,
2021 int bytes
, BdrvRequestFlags flags
)
2023 return blk_co_pwritev(blk
, offset
, bytes
, NULL
,
2024 flags
| BDRV_REQ_ZERO_WRITE
);
2027 int blk_pwrite_compressed(BlockBackend
*blk
, int64_t offset
, const void *buf
,
2030 return blk_prw(blk
, offset
, (void *) buf
, count
, blk_write_entry
,
2031 BDRV_REQ_WRITE_COMPRESSED
);
2034 int blk_truncate(BlockBackend
*blk
, int64_t offset
, PreallocMode prealloc
,
2037 if (!blk_is_available(blk
)) {
2038 error_setg(errp
, "No medium inserted");
2042 return bdrv_truncate(blk
->root
, offset
, prealloc
, errp
);
2045 static void blk_pdiscard_entry(void *opaque
)
2047 BlkRwCo
*rwco
= opaque
;
2048 QEMUIOVector
*qiov
= rwco
->iobuf
;
2050 rwco
->ret
= blk_co_pdiscard(rwco
->blk
, rwco
->offset
, qiov
->size
);
2054 int blk_pdiscard(BlockBackend
*blk
, int64_t offset
, int bytes
)
2056 return blk_prw(blk
, offset
, NULL
, bytes
, blk_pdiscard_entry
, 0);
2059 int blk_save_vmstate(BlockBackend
*blk
, const uint8_t *buf
,
2060 int64_t pos
, int size
)
2064 if (!blk_is_available(blk
)) {
2068 ret
= bdrv_save_vmstate(blk_bs(blk
), buf
, pos
, size
);
2073 if (ret
== size
&& !blk
->enable_write_cache
) {
2074 ret
= bdrv_flush(blk_bs(blk
));
2077 return ret
< 0 ? ret
: size
;
2080 int blk_load_vmstate(BlockBackend
*blk
, uint8_t *buf
, int64_t pos
, int size
)
2082 if (!blk_is_available(blk
)) {
2086 return bdrv_load_vmstate(blk_bs(blk
), buf
, pos
, size
);
2089 int blk_probe_blocksizes(BlockBackend
*blk
, BlockSizes
*bsz
)
2091 if (!blk_is_available(blk
)) {
2095 return bdrv_probe_blocksizes(blk_bs(blk
), bsz
);
2098 int blk_probe_geometry(BlockBackend
*blk
, HDGeometry
*geo
)
2100 if (!blk_is_available(blk
)) {
2104 return bdrv_probe_geometry(blk_bs(blk
), geo
);
2108 * Updates the BlockBackendRootState object with data from the currently
2109 * attached BlockDriverState.
2111 void blk_update_root_state(BlockBackend
*blk
)
2115 blk
->root_state
.open_flags
= blk
->root
->bs
->open_flags
;
2116 blk
->root_state
.read_only
= blk
->root
->bs
->read_only
;
2117 blk
->root_state
.detect_zeroes
= blk
->root
->bs
->detect_zeroes
;
2121 * Returns the detect-zeroes setting to be used for bdrv_open() of a
2122 * BlockDriverState which is supposed to inherit the root state.
2124 bool blk_get_detect_zeroes_from_root_state(BlockBackend
*blk
)
2126 return blk
->root_state
.detect_zeroes
;
2130 * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
2131 * supposed to inherit the root state.
2133 int blk_get_open_flags_from_root_state(BlockBackend
*blk
)
2137 bs_flags
= blk
->root_state
.read_only
? 0 : BDRV_O_RDWR
;
2138 bs_flags
|= blk
->root_state
.open_flags
& ~BDRV_O_RDWR
;
2143 BlockBackendRootState
*blk_get_root_state(BlockBackend
*blk
)
2145 return &blk
->root_state
;
2148 int blk_commit_all(void)
2150 BlockBackend
*blk
= NULL
;
2152 while ((blk
= blk_all_next(blk
)) != NULL
) {
2153 AioContext
*aio_context
= blk_get_aio_context(blk
);
2155 aio_context_acquire(aio_context
);
2156 if (blk_is_inserted(blk
) && blk
->root
->bs
->backing
) {
2157 int ret
= bdrv_commit(blk
->root
->bs
);
2159 aio_context_release(aio_context
);
2163 aio_context_release(aio_context
);
2169 /* throttling disk I/O limits */
2170 void blk_set_io_limits(BlockBackend
*blk
, ThrottleConfig
*cfg
)
2172 throttle_group_config(&blk
->public.throttle_group_member
, cfg
);
2175 void blk_io_limits_disable(BlockBackend
*blk
)
2177 BlockDriverState
*bs
= blk_bs(blk
);
2178 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
2179 assert(tgm
->throttle_state
);
2181 bdrv_drained_begin(bs
);
2183 throttle_group_unregister_tgm(tgm
);
2185 bdrv_drained_end(bs
);
2189 /* should be called before blk_set_io_limits if a limit is set */
2190 void blk_io_limits_enable(BlockBackend
*blk
, const char *group
)
2192 assert(!blk
->public.throttle_group_member
.throttle_state
);
2193 throttle_group_register_tgm(&blk
->public.throttle_group_member
,
2194 group
, blk_get_aio_context(blk
));
2197 void blk_io_limits_update_group(BlockBackend
*blk
, const char *group
)
2199 /* this BB is not part of any group */
2200 if (!blk
->public.throttle_group_member
.throttle_state
) {
2204 /* this BB is a part of the same group than the one we want */
2205 if (!g_strcmp0(throttle_group_get_name(&blk
->public.throttle_group_member
),
2210 /* need to change the group this bs belong to */
2211 blk_io_limits_disable(blk
);
2212 blk_io_limits_enable(blk
, group
);
2215 static void blk_root_drained_begin(BdrvChild
*child
)
2217 BlockBackend
*blk
= child
->opaque
;
2219 if (++blk
->quiesce_counter
== 1) {
2220 if (blk
->dev_ops
&& blk
->dev_ops
->drained_begin
) {
2221 blk
->dev_ops
->drained_begin(blk
->dev_opaque
);
2225 /* Note that blk->root may not be accessible here yet if we are just
2226 * attaching to a BlockDriverState that is drained. Use child instead. */
2228 if (atomic_fetch_inc(&blk
->public.throttle_group_member
.io_limits_disabled
) == 0) {
2229 throttle_group_restart_tgm(&blk
->public.throttle_group_member
);
2233 static bool blk_root_drained_poll(BdrvChild
*child
)
2235 BlockBackend
*blk
= child
->opaque
;
2236 assert(blk
->quiesce_counter
);
2237 return !!blk
->in_flight
;
2240 static void blk_root_drained_end(BdrvChild
*child
, int *drained_end_counter
)
2242 BlockBackend
*blk
= child
->opaque
;
2243 assert(blk
->quiesce_counter
);
2245 assert(blk
->public.throttle_group_member
.io_limits_disabled
);
2246 atomic_dec(&blk
->public.throttle_group_member
.io_limits_disabled
);
2248 if (--blk
->quiesce_counter
== 0) {
2249 if (blk
->dev_ops
&& blk
->dev_ops
->drained_end
) {
2250 blk
->dev_ops
->drained_end(blk
->dev_opaque
);
2255 void blk_register_buf(BlockBackend
*blk
, void *host
, size_t size
)
2257 bdrv_register_buf(blk_bs(blk
), host
, size
);
2260 void blk_unregister_buf(BlockBackend
*blk
, void *host
)
2262 bdrv_unregister_buf(blk_bs(blk
), host
);
2265 int coroutine_fn
blk_co_copy_range(BlockBackend
*blk_in
, int64_t off_in
,
2266 BlockBackend
*blk_out
, int64_t off_out
,
2267 int bytes
, BdrvRequestFlags read_flags
,
2268 BdrvRequestFlags write_flags
)
2271 r
= blk_check_byte_request(blk_in
, off_in
, bytes
);
2275 r
= blk_check_byte_request(blk_out
, off_out
, bytes
);
2279 return bdrv_co_copy_range(blk_in
->root
, off_in
,
2280 blk_out
->root
, off_out
,
2281 bytes
, read_flags
, write_flags
);
2284 const BdrvChild
*blk_root(BlockBackend
*blk
)