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 /* the block size for which the guest device expects atomicity */
62 /* If the BDS tree is removed, some of its options are stored here (which
63 * can be used to restore those options in the new BDS on insert) */
64 BlockBackendRootState root_state
;
66 bool enable_write_cache
;
68 /* I/O stats (display with "info blockstats"). */
71 BlockdevOnError on_read_error
, on_write_error
;
72 bool iostatus_enabled
;
73 BlockDeviceIoStatus iostatus
;
79 bool allow_aio_context_change
;
80 bool allow_write_beyond_eof
;
82 NotifierList remove_bs_notifiers
, insert_bs_notifiers
;
83 QLIST_HEAD(, BlockBackendAioNotifier
) aio_notifiers
;
86 CoQueue queued_requests
;
87 bool disable_request_queuing
;
89 VMChangeStateEntry
*vmsh
;
90 bool force_allow_inactivate
;
92 /* Number of in-flight aio requests. BlockDriverState also counts
93 * in-flight requests but aio requests can exist even when blk->root is
94 * NULL, so we cannot rely on its counter for that case.
95 * Accessed with atomic ops.
97 unsigned int in_flight
;
100 typedef struct BlockBackendAIOCB
{
106 static const AIOCBInfo block_backend_aiocb_info
= {
107 .get_aio_context
= blk_aiocb_get_aio_context
,
108 .aiocb_size
= sizeof(BlockBackendAIOCB
),
111 static void drive_info_del(DriveInfo
*dinfo
);
112 static BlockBackend
*bdrv_first_blk(BlockDriverState
*bs
);
114 /* All BlockBackends */
115 static QTAILQ_HEAD(, BlockBackend
) block_backends
=
116 QTAILQ_HEAD_INITIALIZER(block_backends
);
118 /* All BlockBackends referenced by the monitor and which are iterated through by
120 static QTAILQ_HEAD(, BlockBackend
) monitor_block_backends
=
121 QTAILQ_HEAD_INITIALIZER(monitor_block_backends
);
123 static void blk_root_inherit_options(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
;
194 if (!blk
->disable_perm
) {
198 blk
->disable_perm
= false;
200 blk_set_perm(blk
, blk
->perm
, BLK_PERM_ALL
, &local_err
);
202 error_propagate(errp
, local_err
);
203 blk
->disable_perm
= true;
207 if (runstate_check(RUN_STATE_INMIGRATE
)) {
208 /* Activation can happen when migration process is still active, for
209 * example when nbd_server_add is called during non-shared storage
210 * migration. Defer the shared_perm update to migration completion. */
212 blk
->vmsh
= qemu_add_vm_change_state_handler(blk_vm_state_changed
,
218 blk_set_perm(blk
, blk
->perm
, blk
->shared_perm
, &local_err
);
220 error_propagate(errp
, local_err
);
221 blk
->disable_perm
= true;
226 void blk_set_force_allow_inactivate(BlockBackend
*blk
)
228 blk
->force_allow_inactivate
= true;
231 static bool blk_can_inactivate(BlockBackend
*blk
)
233 /* If it is a guest device, inactivate is ok. */
234 if (blk
->dev
|| blk_name(blk
)[0]) {
238 /* Inactivating means no more writes to the image can be done,
239 * even if those writes would be changes invisible to the
240 * guest. For block job BBs that satisfy this, we can just allow
241 * it. This is the case for mirror job source, which is required
242 * by libvirt non-shared block migration. */
243 if (!(blk
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
))) {
247 return blk
->force_allow_inactivate
;
250 static int blk_root_inactivate(BdrvChild
*child
)
252 BlockBackend
*blk
= child
->opaque
;
254 if (blk
->disable_perm
) {
258 if (!blk_can_inactivate(blk
)) {
262 blk
->disable_perm
= true;
264 bdrv_child_try_set_perm(blk
->root
, 0, BLK_PERM_ALL
, &error_abort
);
270 static void blk_root_attach(BdrvChild
*child
)
272 BlockBackend
*blk
= child
->opaque
;
273 BlockBackendAioNotifier
*notifier
;
275 trace_blk_root_attach(child
, blk
, child
->bs
);
277 QLIST_FOREACH(notifier
, &blk
->aio_notifiers
, list
) {
278 bdrv_add_aio_context_notifier(child
->bs
,
279 notifier
->attached_aio_context
,
280 notifier
->detach_aio_context
,
285 static void blk_root_detach(BdrvChild
*child
)
287 BlockBackend
*blk
= child
->opaque
;
288 BlockBackendAioNotifier
*notifier
;
290 trace_blk_root_detach(child
, blk
, child
->bs
);
292 QLIST_FOREACH(notifier
, &blk
->aio_notifiers
, list
) {
293 bdrv_remove_aio_context_notifier(child
->bs
,
294 notifier
->attached_aio_context
,
295 notifier
->detach_aio_context
,
300 static AioContext
*blk_root_get_parent_aio_context(BdrvChild
*c
)
302 BlockBackend
*blk
= c
->opaque
;
304 return blk_get_aio_context(blk
);
307 static const BdrvChildClass child_root
= {
308 .inherit_options
= blk_root_inherit_options
,
310 .change_media
= blk_root_change_media
,
311 .resize
= blk_root_resize
,
312 .get_name
= blk_root_get_name
,
313 .get_parent_desc
= blk_root_get_parent_desc
,
315 .drained_begin
= blk_root_drained_begin
,
316 .drained_poll
= blk_root_drained_poll
,
317 .drained_end
= blk_root_drained_end
,
319 .activate
= blk_root_activate
,
320 .inactivate
= blk_root_inactivate
,
322 .attach
= blk_root_attach
,
323 .detach
= blk_root_detach
,
325 .can_set_aio_ctx
= blk_root_can_set_aio_ctx
,
326 .set_aio_ctx
= blk_root_set_aio_ctx
,
328 .get_parent_aio_context
= blk_root_get_parent_aio_context
,
332 * Create a new BlockBackend with a reference count of one.
334 * @perm is a bitmasks of BLK_PERM_* constants which describes the permissions
335 * to request for a block driver node that is attached to this BlockBackend.
336 * @shared_perm is a bitmask which describes which permissions may be granted
337 * to other users of the attached node.
338 * Both sets of permissions can be changed later using blk_set_perm().
340 * Return the new BlockBackend on success, null on failure.
342 BlockBackend
*blk_new(AioContext
*ctx
, uint64_t perm
, uint64_t shared_perm
)
346 blk
= g_new0(BlockBackend
, 1);
350 blk
->shared_perm
= shared_perm
;
351 blk_set_enable_write_cache(blk
, true);
353 blk
->on_read_error
= BLOCKDEV_ON_ERROR_REPORT
;
354 blk
->on_write_error
= BLOCKDEV_ON_ERROR_ENOSPC
;
356 block_acct_init(&blk
->stats
);
358 qemu_co_queue_init(&blk
->queued_requests
);
359 notifier_list_init(&blk
->remove_bs_notifiers
);
360 notifier_list_init(&blk
->insert_bs_notifiers
);
361 QLIST_INIT(&blk
->aio_notifiers
);
363 QTAILQ_INSERT_TAIL(&block_backends
, blk
, link
);
368 * Create a new BlockBackend connected to an existing BlockDriverState.
370 * @perm is a bitmasks of BLK_PERM_* constants which describes the
371 * permissions to request for @bs that is attached to this
372 * BlockBackend. @shared_perm is a bitmask which describes which
373 * permissions may be granted to other users of the attached node.
374 * Both sets of permissions can be changed later using blk_set_perm().
376 * Return the new BlockBackend on success, null on failure.
378 BlockBackend
*blk_new_with_bs(BlockDriverState
*bs
, uint64_t perm
,
379 uint64_t shared_perm
, Error
**errp
)
381 BlockBackend
*blk
= blk_new(bdrv_get_aio_context(bs
), perm
, shared_perm
);
383 if (blk_insert_bs(blk
, bs
, errp
) < 0) {
391 * Creates a new BlockBackend, opens a new BlockDriverState, and connects both.
392 * The new BlockBackend is in the main AioContext.
394 * Just as with bdrv_open(), after having called this function the reference to
395 * @options belongs to the block layer (even on failure).
397 * TODO: Remove @filename and @flags; it should be possible to specify a whole
398 * BDS tree just by specifying the @options QDict (or @reference,
399 * alternatively). At the time of adding this function, this is not possible,
400 * though, so callers of this function have to be able to specify @filename and
403 BlockBackend
*blk_new_open(const char *filename
, const char *reference
,
404 QDict
*options
, int flags
, Error
**errp
)
407 BlockDriverState
*bs
;
409 uint64_t shared
= BLK_PERM_ALL
;
412 * blk_new_open() is mainly used in .bdrv_create implementations and the
413 * tools where sharing isn't a major concern because the BDS stays private
414 * and the file is generally not supposed to be used by a second process,
415 * so we just request permission according to the flags.
417 * The exceptions are xen_disk and blockdev_init(); in these cases, the
418 * caller of blk_new_open() doesn't make use of the permissions, but they
419 * shouldn't hurt either. We can still share everything here because the
420 * guest devices will add their own blockers if they can't share.
422 if ((flags
& BDRV_O_NO_IO
) == 0) {
423 perm
|= BLK_PERM_CONSISTENT_READ
;
424 if (flags
& BDRV_O_RDWR
) {
425 perm
|= BLK_PERM_WRITE
;
428 if (flags
& BDRV_O_RESIZE
) {
429 perm
|= BLK_PERM_RESIZE
;
431 if (flags
& BDRV_O_NO_SHARE
) {
432 shared
= BLK_PERM_CONSISTENT_READ
| BLK_PERM_WRITE_UNCHANGED
;
435 blk
= blk_new(qemu_get_aio_context(), perm
, shared
);
436 bs
= bdrv_open(filename
, reference
, options
, flags
, errp
);
442 blk
->root
= bdrv_root_attach_child(bs
, "root", &child_root
,
443 BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
,
444 perm
, shared
, blk
, errp
);
453 static void blk_delete(BlockBackend
*blk
)
455 assert(!blk
->refcnt
);
458 if (blk
->public.throttle_group_member
.throttle_state
) {
459 blk_io_limits_disable(blk
);
465 qemu_del_vm_change_state_handler(blk
->vmsh
);
468 assert(QLIST_EMPTY(&blk
->remove_bs_notifiers
.notifiers
));
469 assert(QLIST_EMPTY(&blk
->insert_bs_notifiers
.notifiers
));
470 assert(QLIST_EMPTY(&blk
->aio_notifiers
));
471 QTAILQ_REMOVE(&block_backends
, blk
, link
);
472 drive_info_del(blk
->legacy_dinfo
);
473 block_acct_cleanup(&blk
->stats
);
477 static void drive_info_del(DriveInfo
*dinfo
)
482 qemu_opts_del(dinfo
->opts
);
486 int blk_get_refcnt(BlockBackend
*blk
)
488 return blk
? blk
->refcnt
: 0;
492 * Increment @blk's reference count.
493 * @blk must not be null.
495 void blk_ref(BlockBackend
*blk
)
497 assert(blk
->refcnt
> 0);
502 * Decrement @blk's reference count.
503 * If this drops it to zero, destroy @blk.
504 * For convenience, do nothing if @blk is null.
506 void blk_unref(BlockBackend
*blk
)
509 assert(blk
->refcnt
> 0);
510 if (blk
->refcnt
> 1) {
514 /* blk_drain() cannot resurrect blk, nobody held a reference */
515 assert(blk
->refcnt
== 1);
523 * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
524 * ones which are hidden (i.e. are not referenced by the monitor).
526 BlockBackend
*blk_all_next(BlockBackend
*blk
)
528 return blk
? QTAILQ_NEXT(blk
, link
)
529 : QTAILQ_FIRST(&block_backends
);
532 void blk_remove_all_bs(void)
534 BlockBackend
*blk
= NULL
;
536 while ((blk
= blk_all_next(blk
)) != NULL
) {
537 AioContext
*ctx
= blk_get_aio_context(blk
);
539 aio_context_acquire(ctx
);
543 aio_context_release(ctx
);
548 * Return the monitor-owned BlockBackend after @blk.
549 * If @blk is null, return the first one.
550 * Else, return @blk's next sibling, which may be null.
552 * To iterate over all BlockBackends, do
553 * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
557 BlockBackend
*blk_next(BlockBackend
*blk
)
559 return blk
? QTAILQ_NEXT(blk
, monitor_link
)
560 : QTAILQ_FIRST(&monitor_block_backends
);
563 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
564 * the monitor or attached to a BlockBackend */
565 BlockDriverState
*bdrv_next(BdrvNextIterator
*it
)
567 BlockDriverState
*bs
, *old_bs
;
569 /* Must be called from the main loop */
570 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
572 /* First, return all root nodes of BlockBackends. In order to avoid
573 * returning a BDS twice when multiple BBs refer to it, we only return it
574 * if the BB is the first one in the parent list of the BDS. */
575 if (it
->phase
== BDRV_NEXT_BACKEND_ROOTS
) {
576 BlockBackend
*old_blk
= it
->blk
;
578 old_bs
= old_blk
? blk_bs(old_blk
) : NULL
;
581 it
->blk
= blk_all_next(it
->blk
);
582 bs
= it
->blk
? blk_bs(it
->blk
) : NULL
;
583 } while (it
->blk
&& (bs
== NULL
|| bdrv_first_blk(bs
) != it
->blk
));
595 it
->phase
= BDRV_NEXT_MONITOR_OWNED
;
600 /* Then return the monitor-owned BDSes without a BB attached. Ignore all
601 * BDSes that are attached to a BlockBackend here; they have been handled
602 * by the above block already */
604 it
->bs
= bdrv_next_monitor_owned(it
->bs
);
606 } while (bs
&& bdrv_has_blk(bs
));
616 static void bdrv_next_reset(BdrvNextIterator
*it
)
618 *it
= (BdrvNextIterator
) {
619 .phase
= BDRV_NEXT_BACKEND_ROOTS
,
623 BlockDriverState
*bdrv_first(BdrvNextIterator
*it
)
626 return bdrv_next(it
);
629 /* Must be called when aborting a bdrv_next() iteration before
630 * bdrv_next() returns NULL */
631 void bdrv_next_cleanup(BdrvNextIterator
*it
)
633 /* Must be called from the main loop */
634 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
636 if (it
->phase
== BDRV_NEXT_BACKEND_ROOTS
) {
638 bdrv_unref(blk_bs(it
->blk
));
649 * Add a BlockBackend into the list of backends referenced by the monitor, with
650 * the given @name acting as the handle for the monitor.
651 * Strictly for use by blockdev.c.
653 * @name must not be null or empty.
655 * Returns true on success and false on failure. In the latter case, an Error
656 * object is returned through @errp.
658 bool monitor_add_blk(BlockBackend
*blk
, const char *name
, Error
**errp
)
661 assert(name
&& name
[0]);
663 if (!id_wellformed(name
)) {
664 error_setg(errp
, "Invalid device name");
667 if (blk_by_name(name
)) {
668 error_setg(errp
, "Device with id '%s' already exists", name
);
671 if (bdrv_find_node(name
)) {
673 "Device name '%s' conflicts with an existing node name",
678 blk
->name
= g_strdup(name
);
679 QTAILQ_INSERT_TAIL(&monitor_block_backends
, blk
, monitor_link
);
684 * Remove a BlockBackend from the list of backends referenced by the monitor.
685 * Strictly for use by blockdev.c.
687 void monitor_remove_blk(BlockBackend
*blk
)
693 QTAILQ_REMOVE(&monitor_block_backends
, blk
, monitor_link
);
699 * Return @blk's name, a non-null string.
700 * Returns an empty string iff @blk is not referenced by the monitor.
702 const char *blk_name(const BlockBackend
*blk
)
704 return blk
->name
?: "";
708 * Return the BlockBackend with name @name if it exists, else null.
709 * @name must not be null.
711 BlockBackend
*blk_by_name(const char *name
)
713 BlockBackend
*blk
= NULL
;
716 while ((blk
= blk_next(blk
)) != NULL
) {
717 if (!strcmp(name
, blk
->name
)) {
725 * Return the BlockDriverState attached to @blk if any, else null.
727 BlockDriverState
*blk_bs(BlockBackend
*blk
)
729 return blk
->root
? blk
->root
->bs
: NULL
;
732 static BlockBackend
*bdrv_first_blk(BlockDriverState
*bs
)
735 QLIST_FOREACH(child
, &bs
->parents
, next_parent
) {
736 if (child
->klass
== &child_root
) {
737 return child
->opaque
;
745 * Returns true if @bs has an associated BlockBackend.
747 bool bdrv_has_blk(BlockDriverState
*bs
)
749 return bdrv_first_blk(bs
) != NULL
;
753 * Returns true if @bs has only BlockBackends as parents.
755 bool bdrv_is_root_node(BlockDriverState
*bs
)
759 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
760 if (c
->klass
!= &child_root
) {
769 * Return @blk's DriveInfo if any, else null.
771 DriveInfo
*blk_legacy_dinfo(BlockBackend
*blk
)
773 return blk
->legacy_dinfo
;
777 * Set @blk's DriveInfo to @dinfo, and return it.
778 * @blk must not have a DriveInfo set already.
779 * No other BlockBackend may have the same DriveInfo set.
781 DriveInfo
*blk_set_legacy_dinfo(BlockBackend
*blk
, DriveInfo
*dinfo
)
783 assert(!blk
->legacy_dinfo
);
784 return blk
->legacy_dinfo
= dinfo
;
788 * Return the BlockBackend with DriveInfo @dinfo.
791 BlockBackend
*blk_by_legacy_dinfo(DriveInfo
*dinfo
)
793 BlockBackend
*blk
= NULL
;
795 while ((blk
= blk_next(blk
)) != NULL
) {
796 if (blk
->legacy_dinfo
== dinfo
) {
804 * Returns a pointer to the publicly accessible fields of @blk.
806 BlockBackendPublic
*blk_get_public(BlockBackend
*blk
)
812 * Returns a BlockBackend given the associated @public fields.
814 BlockBackend
*blk_by_public(BlockBackendPublic
*public)
816 return container_of(public, BlockBackend
, public);
820 * Disassociates the currently associated BlockDriverState from @blk.
822 void blk_remove_bs(BlockBackend
*blk
)
824 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
825 BlockDriverState
*bs
;
828 notifier_list_notify(&blk
->remove_bs_notifiers
, blk
);
829 if (tgm
->throttle_state
) {
831 bdrv_drained_begin(bs
);
832 throttle_group_detach_aio_context(tgm
);
833 throttle_group_attach_aio_context(tgm
, qemu_get_aio_context());
834 bdrv_drained_end(bs
);
837 blk_update_root_state(blk
);
839 /* bdrv_root_unref_child() will cause blk->root to become stale and may
840 * switch to a completion coroutine later on. Let's drain all I/O here
841 * to avoid that and a potential QEMU crash.
846 bdrv_root_unref_child(root
);
850 * Associates a new BlockDriverState with @blk.
852 int blk_insert_bs(BlockBackend
*blk
, BlockDriverState
*bs
, Error
**errp
)
854 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
856 blk
->root
= bdrv_root_attach_child(bs
, "root", &child_root
,
857 BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
,
858 blk
->perm
, blk
->shared_perm
,
860 if (blk
->root
== NULL
) {
864 notifier_list_notify(&blk
->insert_bs_notifiers
, blk
);
865 if (tgm
->throttle_state
) {
866 throttle_group_detach_aio_context(tgm
);
867 throttle_group_attach_aio_context(tgm
, bdrv_get_aio_context(bs
));
874 * Change BlockDriverState associated with @blk.
876 int blk_replace_bs(BlockBackend
*blk
, BlockDriverState
*new_bs
, Error
**errp
)
878 return bdrv_replace_child_bs(blk
->root
, new_bs
, errp
);
882 * Sets the permission bitmasks that the user of the BlockBackend needs.
884 int blk_set_perm(BlockBackend
*blk
, uint64_t perm
, uint64_t shared_perm
,
889 if (blk
->root
&& !blk
->disable_perm
) {
890 ret
= bdrv_child_try_set_perm(blk
->root
, perm
, shared_perm
, errp
);
897 blk
->shared_perm
= shared_perm
;
902 void blk_get_perm(BlockBackend
*blk
, uint64_t *perm
, uint64_t *shared_perm
)
905 *shared_perm
= blk
->shared_perm
;
909 * Attach device model @dev to @blk.
910 * Return 0 on success, -EBUSY when a device model is attached already.
912 int blk_attach_dev(BlockBackend
*blk
, DeviceState
*dev
)
918 /* While migration is still incoming, we don't need to apply the
919 * permissions of guest device BlockBackends. We might still have a block
920 * job or NBD server writing to the image for storage migration. */
921 if (runstate_check(RUN_STATE_INMIGRATE
)) {
922 blk
->disable_perm
= true;
927 blk_iostatus_reset(blk
);
933 * Detach device model @dev from @blk.
934 * @dev must be currently attached to @blk.
936 void blk_detach_dev(BlockBackend
*blk
, DeviceState
*dev
)
938 assert(blk
->dev
== dev
);
941 blk
->dev_opaque
= NULL
;
942 blk
->guest_block_size
= 512;
943 blk_set_perm(blk
, 0, BLK_PERM_ALL
, &error_abort
);
948 * Return the device model attached to @blk if any, else null.
950 DeviceState
*blk_get_attached_dev(BlockBackend
*blk
)
955 /* Return the qdev ID, or if no ID is assigned the QOM path, of the block
956 * device attached to the BlockBackend. */
957 char *blk_get_attached_dev_id(BlockBackend
*blk
)
959 DeviceState
*dev
= blk
->dev
;
963 } else if (dev
->id
) {
964 return g_strdup(dev
->id
);
967 return object_get_canonical_path(OBJECT(dev
)) ?: g_strdup("");
971 * Return the BlockBackend which has the device model @dev attached if it
974 * @dev must not be null.
976 BlockBackend
*blk_by_dev(void *dev
)
978 BlockBackend
*blk
= NULL
;
981 while ((blk
= blk_all_next(blk
)) != NULL
) {
982 if (blk
->dev
== dev
) {
990 * Set @blk's device model callbacks to @ops.
991 * @opaque is the opaque argument to pass to the callbacks.
992 * This is for use by device models.
994 void blk_set_dev_ops(BlockBackend
*blk
, const BlockDevOps
*ops
,
998 blk
->dev_opaque
= opaque
;
1000 /* Are we currently quiesced? Should we enforce this right now? */
1001 if (blk
->quiesce_counter
&& ops
->drained_begin
) {
1002 ops
->drained_begin(opaque
);
1007 * Notify @blk's attached device model of media change.
1009 * If @load is true, notify of media load. This action can fail, meaning that
1010 * the medium cannot be loaded. @errp is set then.
1012 * If @load is false, notify of media eject. This can never fail.
1014 * Also send DEVICE_TRAY_MOVED events as appropriate.
1016 void blk_dev_change_media_cb(BlockBackend
*blk
, bool load
, Error
**errp
)
1018 if (blk
->dev_ops
&& blk
->dev_ops
->change_media_cb
) {
1019 bool tray_was_open
, tray_is_open
;
1020 Error
*local_err
= NULL
;
1022 tray_was_open
= blk_dev_is_tray_open(blk
);
1023 blk
->dev_ops
->change_media_cb(blk
->dev_opaque
, load
, &local_err
);
1025 assert(load
== true);
1026 error_propagate(errp
, local_err
);
1029 tray_is_open
= blk_dev_is_tray_open(blk
);
1031 if (tray_was_open
!= tray_is_open
) {
1032 char *id
= blk_get_attached_dev_id(blk
);
1033 qapi_event_send_device_tray_moved(blk_name(blk
), id
, tray_is_open
);
1039 static void blk_root_change_media(BdrvChild
*child
, bool load
)
1041 blk_dev_change_media_cb(child
->opaque
, load
, NULL
);
1045 * Does @blk's attached device model have removable media?
1046 * %true if no device model is attached.
1048 bool blk_dev_has_removable_media(BlockBackend
*blk
)
1050 return !blk
->dev
|| (blk
->dev_ops
&& blk
->dev_ops
->change_media_cb
);
1054 * Does @blk's attached device model have a tray?
1056 bool blk_dev_has_tray(BlockBackend
*blk
)
1058 return blk
->dev_ops
&& blk
->dev_ops
->is_tray_open
;
1062 * Notify @blk's attached device model of a media eject request.
1063 * If @force is true, the medium is about to be yanked out forcefully.
1065 void blk_dev_eject_request(BlockBackend
*blk
, bool force
)
1067 if (blk
->dev_ops
&& blk
->dev_ops
->eject_request_cb
) {
1068 blk
->dev_ops
->eject_request_cb(blk
->dev_opaque
, force
);
1073 * Does @blk's attached device model have a tray, and is it open?
1075 bool blk_dev_is_tray_open(BlockBackend
*blk
)
1077 if (blk_dev_has_tray(blk
)) {
1078 return blk
->dev_ops
->is_tray_open(blk
->dev_opaque
);
1084 * Does @blk's attached device model have the medium locked?
1085 * %false if the device model has no such lock.
1087 bool blk_dev_is_medium_locked(BlockBackend
*blk
)
1089 if (blk
->dev_ops
&& blk
->dev_ops
->is_medium_locked
) {
1090 return blk
->dev_ops
->is_medium_locked(blk
->dev_opaque
);
1096 * Notify @blk's attached device model of a backend size change.
1098 static void blk_root_resize(BdrvChild
*child
)
1100 BlockBackend
*blk
= child
->opaque
;
1102 if (blk
->dev_ops
&& blk
->dev_ops
->resize_cb
) {
1103 blk
->dev_ops
->resize_cb(blk
->dev_opaque
);
1107 void blk_iostatus_enable(BlockBackend
*blk
)
1109 blk
->iostatus_enabled
= true;
1110 blk
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
1113 /* The I/O status is only enabled if the drive explicitly
1114 * enables it _and_ the VM is configured to stop on errors */
1115 bool blk_iostatus_is_enabled(const BlockBackend
*blk
)
1117 return (blk
->iostatus_enabled
&&
1118 (blk
->on_write_error
== BLOCKDEV_ON_ERROR_ENOSPC
||
1119 blk
->on_write_error
== BLOCKDEV_ON_ERROR_STOP
||
1120 blk
->on_read_error
== BLOCKDEV_ON_ERROR_STOP
));
1123 BlockDeviceIoStatus
blk_iostatus(const BlockBackend
*blk
)
1125 return blk
->iostatus
;
1128 void blk_iostatus_disable(BlockBackend
*blk
)
1130 blk
->iostatus_enabled
= false;
1133 void blk_iostatus_reset(BlockBackend
*blk
)
1135 if (blk_iostatus_is_enabled(blk
)) {
1136 blk
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
1140 void blk_iostatus_set_err(BlockBackend
*blk
, int error
)
1142 assert(blk_iostatus_is_enabled(blk
));
1143 if (blk
->iostatus
== BLOCK_DEVICE_IO_STATUS_OK
) {
1144 blk
->iostatus
= error
== ENOSPC
? BLOCK_DEVICE_IO_STATUS_NOSPACE
:
1145 BLOCK_DEVICE_IO_STATUS_FAILED
;
1149 void blk_set_allow_write_beyond_eof(BlockBackend
*blk
, bool allow
)
1151 blk
->allow_write_beyond_eof
= allow
;
1154 void blk_set_allow_aio_context_change(BlockBackend
*blk
, bool allow
)
1156 blk
->allow_aio_context_change
= allow
;
1159 void blk_set_disable_request_queuing(BlockBackend
*blk
, bool disable
)
1161 blk
->disable_request_queuing
= disable
;
1164 static int blk_check_byte_request(BlockBackend
*blk
, int64_t offset
,
1173 if (!blk_is_available(blk
)) {
1181 if (!blk
->allow_write_beyond_eof
) {
1182 len
= blk_getlength(blk
);
1187 if (offset
> len
|| len
- offset
< bytes
) {
1195 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1196 static void coroutine_fn
blk_wait_while_drained(BlockBackend
*blk
)
1198 assert(blk
->in_flight
> 0);
1200 if (blk
->quiesce_counter
&& !blk
->disable_request_queuing
) {
1201 blk_dec_in_flight(blk
);
1202 qemu_co_queue_wait(&blk
->queued_requests
, NULL
);
1203 blk_inc_in_flight(blk
);
1207 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1209 blk_co_do_preadv(BlockBackend
*blk
, int64_t offset
, int64_t bytes
,
1210 QEMUIOVector
*qiov
, BdrvRequestFlags flags
)
1213 BlockDriverState
*bs
;
1215 blk_wait_while_drained(blk
);
1217 /* Call blk_bs() only after waiting, the graph may have changed */
1219 trace_blk_co_preadv(blk
, bs
, offset
, bytes
, flags
);
1221 ret
= blk_check_byte_request(blk
, offset
, bytes
);
1226 bdrv_inc_in_flight(bs
);
1228 /* throttling disk I/O */
1229 if (blk
->public.throttle_group_member
.throttle_state
) {
1230 throttle_group_co_io_limits_intercept(&blk
->public.throttle_group_member
,
1234 ret
= bdrv_co_preadv(blk
->root
, offset
, bytes
, qiov
, flags
);
1235 bdrv_dec_in_flight(bs
);
1239 int coroutine_fn
blk_co_preadv(BlockBackend
*blk
, int64_t offset
,
1240 int64_t bytes
, QEMUIOVector
*qiov
,
1241 BdrvRequestFlags flags
)
1245 blk_inc_in_flight(blk
);
1246 ret
= blk_co_do_preadv(blk
, offset
, bytes
, qiov
, flags
);
1247 blk_dec_in_flight(blk
);
1252 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1254 blk_co_do_pwritev_part(BlockBackend
*blk
, int64_t offset
, int64_t bytes
,
1255 QEMUIOVector
*qiov
, size_t qiov_offset
,
1256 BdrvRequestFlags flags
)
1259 BlockDriverState
*bs
;
1261 blk_wait_while_drained(blk
);
1263 /* Call blk_bs() only after waiting, the graph may have changed */
1265 trace_blk_co_pwritev(blk
, bs
, offset
, bytes
, flags
);
1267 ret
= blk_check_byte_request(blk
, offset
, bytes
);
1272 bdrv_inc_in_flight(bs
);
1273 /* throttling disk I/O */
1274 if (blk
->public.throttle_group_member
.throttle_state
) {
1275 throttle_group_co_io_limits_intercept(&blk
->public.throttle_group_member
,
1279 if (!blk
->enable_write_cache
) {
1280 flags
|= BDRV_REQ_FUA
;
1283 ret
= bdrv_co_pwritev_part(blk
->root
, offset
, bytes
, qiov
, qiov_offset
,
1285 bdrv_dec_in_flight(bs
);
1289 int coroutine_fn
blk_co_pwritev_part(BlockBackend
*blk
, int64_t offset
,
1291 QEMUIOVector
*qiov
, size_t qiov_offset
,
1292 BdrvRequestFlags flags
)
1296 blk_inc_in_flight(blk
);
1297 ret
= blk_co_do_pwritev_part(blk
, offset
, bytes
, qiov
, qiov_offset
, flags
);
1298 blk_dec_in_flight(blk
);
1303 int coroutine_fn
blk_co_pwritev(BlockBackend
*blk
, int64_t offset
,
1304 int64_t bytes
, QEMUIOVector
*qiov
,
1305 BdrvRequestFlags flags
)
1307 return blk_co_pwritev_part(blk
, offset
, bytes
, qiov
, 0, flags
);
1310 static int coroutine_fn
blk_pwritev_part(BlockBackend
*blk
, int64_t offset
,
1312 QEMUIOVector
*qiov
, size_t qiov_offset
,
1313 BdrvRequestFlags flags
)
1317 blk_inc_in_flight(blk
);
1318 ret
= blk_do_pwritev_part(blk
, offset
, bytes
, qiov
, qiov_offset
, flags
);
1319 blk_dec_in_flight(blk
);
1324 typedef struct BlkRwCo
{
1329 BdrvRequestFlags flags
;
1332 int blk_pwrite_zeroes(BlockBackend
*blk
, int64_t offset
,
1333 int64_t bytes
, BdrvRequestFlags flags
)
1335 return blk_pwritev_part(blk
, offset
, bytes
, NULL
, 0,
1336 flags
| BDRV_REQ_ZERO_WRITE
);
1339 int blk_make_zero(BlockBackend
*blk
, BdrvRequestFlags flags
)
1341 return bdrv_make_zero(blk
->root
, flags
);
1344 void blk_inc_in_flight(BlockBackend
*blk
)
1346 qatomic_inc(&blk
->in_flight
);
1349 void blk_dec_in_flight(BlockBackend
*blk
)
1351 qatomic_dec(&blk
->in_flight
);
1355 static void error_callback_bh(void *opaque
)
1357 struct BlockBackendAIOCB
*acb
= opaque
;
1359 blk_dec_in_flight(acb
->blk
);
1360 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
1361 qemu_aio_unref(acb
);
1364 BlockAIOCB
*blk_abort_aio_request(BlockBackend
*blk
,
1365 BlockCompletionFunc
*cb
,
1366 void *opaque
, int ret
)
1368 struct BlockBackendAIOCB
*acb
;
1370 blk_inc_in_flight(blk
);
1371 acb
= blk_aio_get(&block_backend_aiocb_info
, blk
, cb
, opaque
);
1375 replay_bh_schedule_oneshot_event(blk_get_aio_context(blk
),
1376 error_callback_bh
, acb
);
1377 return &acb
->common
;
1380 typedef struct BlkAioEmAIOCB
{
1387 static AioContext
*blk_aio_em_aiocb_get_aio_context(BlockAIOCB
*acb_
)
1389 BlkAioEmAIOCB
*acb
= container_of(acb_
, BlkAioEmAIOCB
, common
);
1391 return blk_get_aio_context(acb
->rwco
.blk
);
1394 static const AIOCBInfo blk_aio_em_aiocb_info
= {
1395 .aiocb_size
= sizeof(BlkAioEmAIOCB
),
1396 .get_aio_context
= blk_aio_em_aiocb_get_aio_context
,
1399 static void blk_aio_complete(BlkAioEmAIOCB
*acb
)
1401 if (acb
->has_returned
) {
1402 acb
->common
.cb(acb
->common
.opaque
, acb
->rwco
.ret
);
1403 blk_dec_in_flight(acb
->rwco
.blk
);
1404 qemu_aio_unref(acb
);
1408 static void blk_aio_complete_bh(void *opaque
)
1410 BlkAioEmAIOCB
*acb
= opaque
;
1411 assert(acb
->has_returned
);
1412 blk_aio_complete(acb
);
1415 static BlockAIOCB
*blk_aio_prwv(BlockBackend
*blk
, int64_t offset
,
1417 void *iobuf
, CoroutineEntry co_entry
,
1418 BdrvRequestFlags flags
,
1419 BlockCompletionFunc
*cb
, void *opaque
)
1424 blk_inc_in_flight(blk
);
1425 acb
= blk_aio_get(&blk_aio_em_aiocb_info
, blk
, cb
, opaque
);
1426 acb
->rwco
= (BlkRwCo
) {
1434 acb
->has_returned
= false;
1436 co
= qemu_coroutine_create(co_entry
, acb
);
1437 bdrv_coroutine_enter(blk_bs(blk
), co
);
1439 acb
->has_returned
= true;
1440 if (acb
->rwco
.ret
!= NOT_DONE
) {
1441 replay_bh_schedule_oneshot_event(blk_get_aio_context(blk
),
1442 blk_aio_complete_bh
, acb
);
1445 return &acb
->common
;
1448 static void blk_aio_read_entry(void *opaque
)
1450 BlkAioEmAIOCB
*acb
= opaque
;
1451 BlkRwCo
*rwco
= &acb
->rwco
;
1452 QEMUIOVector
*qiov
= rwco
->iobuf
;
1454 assert(qiov
->size
== acb
->bytes
);
1455 rwco
->ret
= blk_co_do_preadv(rwco
->blk
, rwco
->offset
, acb
->bytes
,
1457 blk_aio_complete(acb
);
1460 static void blk_aio_write_entry(void *opaque
)
1462 BlkAioEmAIOCB
*acb
= opaque
;
1463 BlkRwCo
*rwco
= &acb
->rwco
;
1464 QEMUIOVector
*qiov
= rwco
->iobuf
;
1466 assert(!qiov
|| qiov
->size
== acb
->bytes
);
1467 rwco
->ret
= blk_co_do_pwritev_part(rwco
->blk
, rwco
->offset
, acb
->bytes
,
1468 qiov
, 0, rwco
->flags
);
1469 blk_aio_complete(acb
);
1472 BlockAIOCB
*blk_aio_pwrite_zeroes(BlockBackend
*blk
, int64_t offset
,
1473 int64_t bytes
, BdrvRequestFlags flags
,
1474 BlockCompletionFunc
*cb
, void *opaque
)
1476 return blk_aio_prwv(blk
, offset
, bytes
, NULL
, blk_aio_write_entry
,
1477 flags
| BDRV_REQ_ZERO_WRITE
, cb
, opaque
);
1480 int blk_pread(BlockBackend
*blk
, int64_t offset
, void *buf
, int bytes
)
1483 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, buf
, bytes
);
1485 blk_inc_in_flight(blk
);
1486 ret
= blk_do_preadv(blk
, offset
, bytes
, &qiov
, 0);
1487 blk_dec_in_flight(blk
);
1489 return ret
< 0 ? ret
: bytes
;
1492 int blk_pwrite(BlockBackend
*blk
, int64_t offset
, const void *buf
, int bytes
,
1493 BdrvRequestFlags flags
)
1496 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, buf
, bytes
);
1498 ret
= blk_pwritev_part(blk
, offset
, bytes
, &qiov
, 0, flags
);
1500 return ret
< 0 ? ret
: bytes
;
1503 int64_t blk_getlength(BlockBackend
*blk
)
1505 if (!blk_is_available(blk
)) {
1509 return bdrv_getlength(blk_bs(blk
));
1512 void blk_get_geometry(BlockBackend
*blk
, uint64_t *nb_sectors_ptr
)
1515 *nb_sectors_ptr
= 0;
1517 bdrv_get_geometry(blk_bs(blk
), nb_sectors_ptr
);
1521 int64_t blk_nb_sectors(BlockBackend
*blk
)
1523 if (!blk_is_available(blk
)) {
1527 return bdrv_nb_sectors(blk_bs(blk
));
1530 BlockAIOCB
*blk_aio_preadv(BlockBackend
*blk
, int64_t offset
,
1531 QEMUIOVector
*qiov
, BdrvRequestFlags flags
,
1532 BlockCompletionFunc
*cb
, void *opaque
)
1534 assert((uint64_t)qiov
->size
<= INT64_MAX
);
1535 return blk_aio_prwv(blk
, offset
, qiov
->size
, qiov
,
1536 blk_aio_read_entry
, flags
, cb
, opaque
);
1539 BlockAIOCB
*blk_aio_pwritev(BlockBackend
*blk
, int64_t offset
,
1540 QEMUIOVector
*qiov
, BdrvRequestFlags flags
,
1541 BlockCompletionFunc
*cb
, void *opaque
)
1543 assert((uint64_t)qiov
->size
<= INT64_MAX
);
1544 return blk_aio_prwv(blk
, offset
, qiov
->size
, qiov
,
1545 blk_aio_write_entry
, flags
, cb
, opaque
);
1548 void blk_aio_cancel(BlockAIOCB
*acb
)
1550 bdrv_aio_cancel(acb
);
1553 void blk_aio_cancel_async(BlockAIOCB
*acb
)
1555 bdrv_aio_cancel_async(acb
);
1558 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1560 blk_co_do_ioctl(BlockBackend
*blk
, unsigned long int req
, void *buf
)
1562 blk_wait_while_drained(blk
);
1564 if (!blk_is_available(blk
)) {
1568 return bdrv_co_ioctl(blk_bs(blk
), req
, buf
);
1571 int blk_ioctl(BlockBackend
*blk
, unsigned long int req
, void *buf
)
1575 blk_inc_in_flight(blk
);
1576 ret
= blk_do_ioctl(blk
, req
, buf
);
1577 blk_dec_in_flight(blk
);
1582 static void blk_aio_ioctl_entry(void *opaque
)
1584 BlkAioEmAIOCB
*acb
= opaque
;
1585 BlkRwCo
*rwco
= &acb
->rwco
;
1587 rwco
->ret
= blk_co_do_ioctl(rwco
->blk
, rwco
->offset
, rwco
->iobuf
);
1589 blk_aio_complete(acb
);
1592 BlockAIOCB
*blk_aio_ioctl(BlockBackend
*blk
, unsigned long int req
, void *buf
,
1593 BlockCompletionFunc
*cb
, void *opaque
)
1595 return blk_aio_prwv(blk
, req
, 0, buf
, blk_aio_ioctl_entry
, 0, cb
, opaque
);
1598 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1600 blk_co_do_pdiscard(BlockBackend
*blk
, int64_t offset
, int64_t bytes
)
1604 blk_wait_while_drained(blk
);
1606 ret
= blk_check_byte_request(blk
, offset
, bytes
);
1611 return bdrv_co_pdiscard(blk
->root
, offset
, bytes
);
1614 static void blk_aio_pdiscard_entry(void *opaque
)
1616 BlkAioEmAIOCB
*acb
= opaque
;
1617 BlkRwCo
*rwco
= &acb
->rwco
;
1619 rwco
->ret
= blk_co_do_pdiscard(rwco
->blk
, rwco
->offset
, acb
->bytes
);
1620 blk_aio_complete(acb
);
1623 BlockAIOCB
*blk_aio_pdiscard(BlockBackend
*blk
,
1624 int64_t offset
, int64_t bytes
,
1625 BlockCompletionFunc
*cb
, void *opaque
)
1627 return blk_aio_prwv(blk
, offset
, bytes
, NULL
, blk_aio_pdiscard_entry
, 0,
1631 int coroutine_fn
blk_co_pdiscard(BlockBackend
*blk
, int64_t offset
,
1636 blk_inc_in_flight(blk
);
1637 ret
= blk_co_do_pdiscard(blk
, offset
, bytes
);
1638 blk_dec_in_flight(blk
);
1643 int blk_pdiscard(BlockBackend
*blk
, int64_t offset
, int64_t bytes
)
1647 blk_inc_in_flight(blk
);
1648 ret
= blk_do_pdiscard(blk
, offset
, bytes
);
1649 blk_dec_in_flight(blk
);
1654 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1655 int coroutine_fn
blk_co_do_flush(BlockBackend
*blk
)
1657 blk_wait_while_drained(blk
);
1659 if (!blk_is_available(blk
)) {
1663 return bdrv_co_flush(blk_bs(blk
));
1666 static void blk_aio_flush_entry(void *opaque
)
1668 BlkAioEmAIOCB
*acb
= opaque
;
1669 BlkRwCo
*rwco
= &acb
->rwco
;
1671 rwco
->ret
= blk_co_do_flush(rwco
->blk
);
1672 blk_aio_complete(acb
);
1675 BlockAIOCB
*blk_aio_flush(BlockBackend
*blk
,
1676 BlockCompletionFunc
*cb
, void *opaque
)
1678 return blk_aio_prwv(blk
, 0, 0, NULL
, blk_aio_flush_entry
, 0, cb
, opaque
);
1681 int coroutine_fn
blk_co_flush(BlockBackend
*blk
)
1685 blk_inc_in_flight(blk
);
1686 ret
= blk_co_do_flush(blk
);
1687 blk_dec_in_flight(blk
);
1692 int blk_flush(BlockBackend
*blk
)
1696 blk_inc_in_flight(blk
);
1697 ret
= blk_do_flush(blk
);
1698 blk_dec_in_flight(blk
);
1703 void blk_drain(BlockBackend
*blk
)
1705 BlockDriverState
*bs
= blk_bs(blk
);
1708 bdrv_drained_begin(bs
);
1711 /* We may have -ENOMEDIUM completions in flight */
1712 AIO_WAIT_WHILE(blk_get_aio_context(blk
),
1713 qatomic_mb_read(&blk
->in_flight
) > 0);
1716 bdrv_drained_end(bs
);
1720 void blk_drain_all(void)
1722 BlockBackend
*blk
= NULL
;
1724 bdrv_drain_all_begin();
1726 while ((blk
= blk_all_next(blk
)) != NULL
) {
1727 AioContext
*ctx
= blk_get_aio_context(blk
);
1729 aio_context_acquire(ctx
);
1731 /* We may have -ENOMEDIUM completions in flight */
1732 AIO_WAIT_WHILE(ctx
, qatomic_mb_read(&blk
->in_flight
) > 0);
1734 aio_context_release(ctx
);
1737 bdrv_drain_all_end();
1740 void blk_set_on_error(BlockBackend
*blk
, BlockdevOnError on_read_error
,
1741 BlockdevOnError on_write_error
)
1743 blk
->on_read_error
= on_read_error
;
1744 blk
->on_write_error
= on_write_error
;
1747 BlockdevOnError
blk_get_on_error(BlockBackend
*blk
, bool is_read
)
1749 return is_read
? blk
->on_read_error
: blk
->on_write_error
;
1752 BlockErrorAction
blk_get_error_action(BlockBackend
*blk
, bool is_read
,
1755 BlockdevOnError on_err
= blk_get_on_error(blk
, is_read
);
1758 case BLOCKDEV_ON_ERROR_ENOSPC
:
1759 return (error
== ENOSPC
) ?
1760 BLOCK_ERROR_ACTION_STOP
: BLOCK_ERROR_ACTION_REPORT
;
1761 case BLOCKDEV_ON_ERROR_STOP
:
1762 return BLOCK_ERROR_ACTION_STOP
;
1763 case BLOCKDEV_ON_ERROR_REPORT
:
1764 return BLOCK_ERROR_ACTION_REPORT
;
1765 case BLOCKDEV_ON_ERROR_IGNORE
:
1766 return BLOCK_ERROR_ACTION_IGNORE
;
1767 case BLOCKDEV_ON_ERROR_AUTO
:
1773 static void send_qmp_error_event(BlockBackend
*blk
,
1774 BlockErrorAction action
,
1775 bool is_read
, int error
)
1777 IoOperationType optype
;
1778 BlockDriverState
*bs
= blk_bs(blk
);
1780 optype
= is_read
? IO_OPERATION_TYPE_READ
: IO_OPERATION_TYPE_WRITE
;
1781 qapi_event_send_block_io_error(blk_name(blk
), !!bs
,
1782 bs
? bdrv_get_node_name(bs
) : NULL
, optype
,
1783 action
, blk_iostatus_is_enabled(blk
),
1784 error
== ENOSPC
, strerror(error
));
1787 /* This is done by device models because, while the block layer knows
1788 * about the error, it does not know whether an operation comes from
1789 * the device or the block layer (from a job, for example).
1791 void blk_error_action(BlockBackend
*blk
, BlockErrorAction action
,
1792 bool is_read
, int error
)
1796 if (action
== BLOCK_ERROR_ACTION_STOP
) {
1797 /* First set the iostatus, so that "info block" returns an iostatus
1798 * that matches the events raised so far (an additional error iostatus
1799 * is fine, but not a lost one).
1801 blk_iostatus_set_err(blk
, error
);
1803 /* Then raise the request to stop the VM and the event.
1804 * qemu_system_vmstop_request_prepare has two effects. First,
1805 * it ensures that the STOP event always comes after the
1806 * BLOCK_IO_ERROR event. Second, it ensures that even if management
1807 * can observe the STOP event and do a "cont" before the STOP
1808 * event is issued, the VM will not stop. In this case, vm_start()
1809 * also ensures that the STOP/RESUME pair of events is emitted.
1811 qemu_system_vmstop_request_prepare();
1812 send_qmp_error_event(blk
, action
, is_read
, error
);
1813 qemu_system_vmstop_request(RUN_STATE_IO_ERROR
);
1815 send_qmp_error_event(blk
, action
, is_read
, error
);
1820 * Returns true if the BlockBackend can support taking write permissions
1821 * (because its root node is not read-only).
1823 bool blk_supports_write_perm(BlockBackend
*blk
)
1825 BlockDriverState
*bs
= blk_bs(blk
);
1828 return !bdrv_is_read_only(bs
);
1830 return blk
->root_state
.open_flags
& BDRV_O_RDWR
;
1835 * Returns true if the BlockBackend can be written to in its current
1836 * configuration (i.e. if write permission have been requested)
1838 bool blk_is_writable(BlockBackend
*blk
)
1840 return blk
->perm
& BLK_PERM_WRITE
;
1843 bool blk_is_sg(BlockBackend
*blk
)
1845 BlockDriverState
*bs
= blk_bs(blk
);
1851 return bdrv_is_sg(bs
);
1854 bool blk_enable_write_cache(BlockBackend
*blk
)
1856 return blk
->enable_write_cache
;
1859 void blk_set_enable_write_cache(BlockBackend
*blk
, bool wce
)
1861 blk
->enable_write_cache
= wce
;
1864 void blk_invalidate_cache(BlockBackend
*blk
, Error
**errp
)
1866 BlockDriverState
*bs
= blk_bs(blk
);
1869 error_setg(errp
, "Device '%s' has no medium", blk
->name
);
1873 bdrv_invalidate_cache(bs
, errp
);
1876 bool blk_is_inserted(BlockBackend
*blk
)
1878 BlockDriverState
*bs
= blk_bs(blk
);
1880 return bs
&& bdrv_is_inserted(bs
);
1883 bool blk_is_available(BlockBackend
*blk
)
1885 return blk_is_inserted(blk
) && !blk_dev_is_tray_open(blk
);
1888 void blk_lock_medium(BlockBackend
*blk
, bool locked
)
1890 BlockDriverState
*bs
= blk_bs(blk
);
1893 bdrv_lock_medium(bs
, locked
);
1897 void blk_eject(BlockBackend
*blk
, bool eject_flag
)
1899 BlockDriverState
*bs
= blk_bs(blk
);
1903 bdrv_eject(bs
, eject_flag
);
1906 /* Whether or not we ejected on the backend,
1907 * the frontend experienced a tray event. */
1908 id
= blk_get_attached_dev_id(blk
);
1909 qapi_event_send_device_tray_moved(blk_name(blk
), id
,
1914 int blk_get_flags(BlockBackend
*blk
)
1916 BlockDriverState
*bs
= blk_bs(blk
);
1919 return bdrv_get_flags(bs
);
1921 return blk
->root_state
.open_flags
;
1925 /* Returns the minimum request alignment, in bytes; guaranteed nonzero */
1926 uint32_t blk_get_request_alignment(BlockBackend
*blk
)
1928 BlockDriverState
*bs
= blk_bs(blk
);
1929 return bs
? bs
->bl
.request_alignment
: BDRV_SECTOR_SIZE
;
1932 /* Returns the maximum hardware transfer length, in bytes; guaranteed nonzero */
1933 uint64_t blk_get_max_hw_transfer(BlockBackend
*blk
)
1935 BlockDriverState
*bs
= blk_bs(blk
);
1936 uint64_t max
= INT_MAX
;
1939 max
= MIN_NON_ZERO(max
, bs
->bl
.max_hw_transfer
);
1940 max
= MIN_NON_ZERO(max
, bs
->bl
.max_transfer
);
1942 return ROUND_DOWN(max
, blk_get_request_alignment(blk
));
1945 /* Returns the maximum transfer length, in bytes; guaranteed nonzero */
1946 uint32_t blk_get_max_transfer(BlockBackend
*blk
)
1948 BlockDriverState
*bs
= blk_bs(blk
);
1949 uint32_t max
= INT_MAX
;
1952 max
= MIN_NON_ZERO(max
, bs
->bl
.max_transfer
);
1954 return ROUND_DOWN(max
, blk_get_request_alignment(blk
));
1957 int blk_get_max_hw_iov(BlockBackend
*blk
)
1959 return MIN_NON_ZERO(blk
->root
->bs
->bl
.max_hw_iov
,
1960 blk
->root
->bs
->bl
.max_iov
);
1963 int blk_get_max_iov(BlockBackend
*blk
)
1965 return blk
->root
->bs
->bl
.max_iov
;
1968 void blk_set_guest_block_size(BlockBackend
*blk
, int align
)
1970 blk
->guest_block_size
= align
;
1973 void *blk_try_blockalign(BlockBackend
*blk
, size_t size
)
1975 return qemu_try_blockalign(blk
? blk_bs(blk
) : NULL
, size
);
1978 void *blk_blockalign(BlockBackend
*blk
, size_t size
)
1980 return qemu_blockalign(blk
? blk_bs(blk
) : NULL
, size
);
1983 bool blk_op_is_blocked(BlockBackend
*blk
, BlockOpType op
, Error
**errp
)
1985 BlockDriverState
*bs
= blk_bs(blk
);
1991 return bdrv_op_is_blocked(bs
, op
, errp
);
1994 void blk_op_unblock(BlockBackend
*blk
, BlockOpType op
, Error
*reason
)
1996 BlockDriverState
*bs
= blk_bs(blk
);
1999 bdrv_op_unblock(bs
, op
, reason
);
2003 void blk_op_block_all(BlockBackend
*blk
, Error
*reason
)
2005 BlockDriverState
*bs
= blk_bs(blk
);
2008 bdrv_op_block_all(bs
, reason
);
2012 void blk_op_unblock_all(BlockBackend
*blk
, Error
*reason
)
2014 BlockDriverState
*bs
= blk_bs(blk
);
2017 bdrv_op_unblock_all(bs
, reason
);
2021 AioContext
*blk_get_aio_context(BlockBackend
*blk
)
2023 BlockDriverState
*bs
= blk_bs(blk
);
2026 AioContext
*ctx
= bdrv_get_aio_context(blk_bs(blk
));
2027 assert(ctx
== blk
->ctx
);
2033 static AioContext
*blk_aiocb_get_aio_context(BlockAIOCB
*acb
)
2035 BlockBackendAIOCB
*blk_acb
= DO_UPCAST(BlockBackendAIOCB
, common
, acb
);
2036 return blk_get_aio_context(blk_acb
->blk
);
2039 static int blk_do_set_aio_context(BlockBackend
*blk
, AioContext
*new_context
,
2040 bool update_root_node
, Error
**errp
)
2042 BlockDriverState
*bs
= blk_bs(blk
);
2043 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
2047 if (update_root_node
) {
2048 ret
= bdrv_child_try_set_aio_context(bs
, new_context
, blk
->root
,
2054 if (tgm
->throttle_state
) {
2055 bdrv_drained_begin(bs
);
2056 throttle_group_detach_aio_context(tgm
);
2057 throttle_group_attach_aio_context(tgm
, new_context
);
2058 bdrv_drained_end(bs
);
2062 blk
->ctx
= new_context
;
2066 int blk_set_aio_context(BlockBackend
*blk
, AioContext
*new_context
,
2069 return blk_do_set_aio_context(blk
, new_context
, true, errp
);
2072 static bool blk_root_can_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
2073 GSList
**ignore
, Error
**errp
)
2075 BlockBackend
*blk
= child
->opaque
;
2077 if (blk
->allow_aio_context_change
) {
2081 /* Only manually created BlockBackends that are not attached to anything
2082 * can change their AioContext without updating their user. */
2083 if (!blk
->name
|| blk
->dev
) {
2084 /* TODO Add BB name/QOM path */
2085 error_setg(errp
, "Cannot change iothread of active block backend");
2092 static void blk_root_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
2095 BlockBackend
*blk
= child
->opaque
;
2096 blk_do_set_aio_context(blk
, ctx
, false, &error_abort
);
2099 void blk_add_aio_context_notifier(BlockBackend
*blk
,
2100 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
2101 void (*detach_aio_context
)(void *opaque
), void *opaque
)
2103 BlockBackendAioNotifier
*notifier
;
2104 BlockDriverState
*bs
= blk_bs(blk
);
2106 notifier
= g_new(BlockBackendAioNotifier
, 1);
2107 notifier
->attached_aio_context
= attached_aio_context
;
2108 notifier
->detach_aio_context
= detach_aio_context
;
2109 notifier
->opaque
= opaque
;
2110 QLIST_INSERT_HEAD(&blk
->aio_notifiers
, notifier
, list
);
2113 bdrv_add_aio_context_notifier(bs
, attached_aio_context
,
2114 detach_aio_context
, opaque
);
2118 void blk_remove_aio_context_notifier(BlockBackend
*blk
,
2119 void (*attached_aio_context
)(AioContext
*,
2121 void (*detach_aio_context
)(void *),
2124 BlockBackendAioNotifier
*notifier
;
2125 BlockDriverState
*bs
= blk_bs(blk
);
2128 bdrv_remove_aio_context_notifier(bs
, attached_aio_context
,
2129 detach_aio_context
, opaque
);
2132 QLIST_FOREACH(notifier
, &blk
->aio_notifiers
, list
) {
2133 if (notifier
->attached_aio_context
== attached_aio_context
&&
2134 notifier
->detach_aio_context
== detach_aio_context
&&
2135 notifier
->opaque
== opaque
) {
2136 QLIST_REMOVE(notifier
, list
);
2145 void blk_add_remove_bs_notifier(BlockBackend
*blk
, Notifier
*notify
)
2147 notifier_list_add(&blk
->remove_bs_notifiers
, notify
);
2150 void blk_add_insert_bs_notifier(BlockBackend
*blk
, Notifier
*notify
)
2152 notifier_list_add(&blk
->insert_bs_notifiers
, notify
);
2155 void blk_io_plug(BlockBackend
*blk
)
2157 BlockDriverState
*bs
= blk_bs(blk
);
2164 void blk_io_unplug(BlockBackend
*blk
)
2166 BlockDriverState
*bs
= blk_bs(blk
);
2173 BlockAcctStats
*blk_get_stats(BlockBackend
*blk
)
2178 void *blk_aio_get(const AIOCBInfo
*aiocb_info
, BlockBackend
*blk
,
2179 BlockCompletionFunc
*cb
, void *opaque
)
2181 return qemu_aio_get(aiocb_info
, blk_bs(blk
), cb
, opaque
);
2184 int coroutine_fn
blk_co_pwrite_zeroes(BlockBackend
*blk
, int64_t offset
,
2185 int64_t bytes
, BdrvRequestFlags flags
)
2187 return blk_co_pwritev(blk
, offset
, bytes
, NULL
,
2188 flags
| BDRV_REQ_ZERO_WRITE
);
2191 int blk_pwrite_compressed(BlockBackend
*blk
, int64_t offset
, const void *buf
,
2194 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, buf
, bytes
);
2195 return blk_pwritev_part(blk
, offset
, bytes
, &qiov
, 0,
2196 BDRV_REQ_WRITE_COMPRESSED
);
2199 int blk_truncate(BlockBackend
*blk
, int64_t offset
, bool exact
,
2200 PreallocMode prealloc
, BdrvRequestFlags flags
, Error
**errp
)
2202 if (!blk_is_available(blk
)) {
2203 error_setg(errp
, "No medium inserted");
2207 return bdrv_truncate(blk
->root
, offset
, exact
, prealloc
, flags
, errp
);
2210 int blk_save_vmstate(BlockBackend
*blk
, const uint8_t *buf
,
2211 int64_t pos
, int size
)
2215 if (!blk_is_available(blk
)) {
2219 ret
= bdrv_save_vmstate(blk_bs(blk
), buf
, pos
, size
);
2224 if (ret
== size
&& !blk
->enable_write_cache
) {
2225 ret
= bdrv_flush(blk_bs(blk
));
2228 return ret
< 0 ? ret
: size
;
2231 int blk_load_vmstate(BlockBackend
*blk
, uint8_t *buf
, int64_t pos
, int size
)
2233 if (!blk_is_available(blk
)) {
2237 return bdrv_load_vmstate(blk_bs(blk
), buf
, pos
, size
);
2240 int blk_probe_blocksizes(BlockBackend
*blk
, BlockSizes
*bsz
)
2242 if (!blk_is_available(blk
)) {
2246 return bdrv_probe_blocksizes(blk_bs(blk
), bsz
);
2249 int blk_probe_geometry(BlockBackend
*blk
, HDGeometry
*geo
)
2251 if (!blk_is_available(blk
)) {
2255 return bdrv_probe_geometry(blk_bs(blk
), geo
);
2259 * Updates the BlockBackendRootState object with data from the currently
2260 * attached BlockDriverState.
2262 void blk_update_root_state(BlockBackend
*blk
)
2266 blk
->root_state
.open_flags
= blk
->root
->bs
->open_flags
;
2267 blk
->root_state
.detect_zeroes
= blk
->root
->bs
->detect_zeroes
;
2271 * Returns the detect-zeroes setting to be used for bdrv_open() of a
2272 * BlockDriverState which is supposed to inherit the root state.
2274 bool blk_get_detect_zeroes_from_root_state(BlockBackend
*blk
)
2276 return blk
->root_state
.detect_zeroes
;
2280 * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
2281 * supposed to inherit the root state.
2283 int blk_get_open_flags_from_root_state(BlockBackend
*blk
)
2285 return blk
->root_state
.open_flags
;
2288 BlockBackendRootState
*blk_get_root_state(BlockBackend
*blk
)
2290 return &blk
->root_state
;
2293 int blk_commit_all(void)
2295 BlockBackend
*blk
= NULL
;
2297 while ((blk
= blk_all_next(blk
)) != NULL
) {
2298 AioContext
*aio_context
= blk_get_aio_context(blk
);
2299 BlockDriverState
*unfiltered_bs
= bdrv_skip_filters(blk_bs(blk
));
2301 aio_context_acquire(aio_context
);
2302 if (blk_is_inserted(blk
) && bdrv_cow_child(unfiltered_bs
)) {
2305 ret
= bdrv_commit(unfiltered_bs
);
2307 aio_context_release(aio_context
);
2311 aio_context_release(aio_context
);
2317 /* throttling disk I/O limits */
2318 void blk_set_io_limits(BlockBackend
*blk
, ThrottleConfig
*cfg
)
2320 throttle_group_config(&blk
->public.throttle_group_member
, cfg
);
2323 void blk_io_limits_disable(BlockBackend
*blk
)
2325 BlockDriverState
*bs
= blk_bs(blk
);
2326 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
2327 assert(tgm
->throttle_state
);
2329 bdrv_drained_begin(bs
);
2331 throttle_group_unregister_tgm(tgm
);
2333 bdrv_drained_end(bs
);
2337 /* should be called before blk_set_io_limits if a limit is set */
2338 void blk_io_limits_enable(BlockBackend
*blk
, const char *group
)
2340 assert(!blk
->public.throttle_group_member
.throttle_state
);
2341 throttle_group_register_tgm(&blk
->public.throttle_group_member
,
2342 group
, blk_get_aio_context(blk
));
2345 void blk_io_limits_update_group(BlockBackend
*blk
, const char *group
)
2347 /* this BB is not part of any group */
2348 if (!blk
->public.throttle_group_member
.throttle_state
) {
2352 /* this BB is a part of the same group than the one we want */
2353 if (!g_strcmp0(throttle_group_get_name(&blk
->public.throttle_group_member
),
2358 /* need to change the group this bs belong to */
2359 blk_io_limits_disable(blk
);
2360 blk_io_limits_enable(blk
, group
);
2363 static void blk_root_drained_begin(BdrvChild
*child
)
2365 BlockBackend
*blk
= child
->opaque
;
2366 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
2368 if (++blk
->quiesce_counter
== 1) {
2369 if (blk
->dev_ops
&& blk
->dev_ops
->drained_begin
) {
2370 blk
->dev_ops
->drained_begin(blk
->dev_opaque
);
2374 /* Note that blk->root may not be accessible here yet if we are just
2375 * attaching to a BlockDriverState that is drained. Use child instead. */
2377 if (qatomic_fetch_inc(&tgm
->io_limits_disabled
) == 0) {
2378 throttle_group_restart_tgm(tgm
);
2382 static bool blk_root_drained_poll(BdrvChild
*child
)
2384 BlockBackend
*blk
= child
->opaque
;
2386 assert(blk
->quiesce_counter
);
2388 if (blk
->dev_ops
&& blk
->dev_ops
->drained_poll
) {
2389 busy
= blk
->dev_ops
->drained_poll(blk
->dev_opaque
);
2391 return busy
|| !!blk
->in_flight
;
2394 static void blk_root_drained_end(BdrvChild
*child
, int *drained_end_counter
)
2396 BlockBackend
*blk
= child
->opaque
;
2397 assert(blk
->quiesce_counter
);
2399 assert(blk
->public.throttle_group_member
.io_limits_disabled
);
2400 qatomic_dec(&blk
->public.throttle_group_member
.io_limits_disabled
);
2402 if (--blk
->quiesce_counter
== 0) {
2403 if (blk
->dev_ops
&& blk
->dev_ops
->drained_end
) {
2404 blk
->dev_ops
->drained_end(blk
->dev_opaque
);
2406 while (qemu_co_enter_next(&blk
->queued_requests
, NULL
)) {
2407 /* Resume all queued requests */
2412 void blk_register_buf(BlockBackend
*blk
, void *host
, size_t size
)
2414 bdrv_register_buf(blk_bs(blk
), host
, size
);
2417 void blk_unregister_buf(BlockBackend
*blk
, void *host
)
2419 bdrv_unregister_buf(blk_bs(blk
), host
);
2422 int coroutine_fn
blk_co_copy_range(BlockBackend
*blk_in
, int64_t off_in
,
2423 BlockBackend
*blk_out
, int64_t off_out
,
2424 int64_t bytes
, BdrvRequestFlags read_flags
,
2425 BdrvRequestFlags write_flags
)
2428 r
= blk_check_byte_request(blk_in
, off_in
, bytes
);
2432 r
= blk_check_byte_request(blk_out
, off_out
, bytes
);
2436 return bdrv_co_copy_range(blk_in
->root
, off_in
,
2437 blk_out
->root
, off_out
,
2438 bytes
, read_flags
, write_flags
);
2441 const BdrvChild
*blk_root(BlockBackend
*blk
)
2446 int blk_make_empty(BlockBackend
*blk
, Error
**errp
)
2448 if (!blk_is_available(blk
)) {
2449 error_setg(errp
, "No medium inserted");
2453 return bdrv_make_empty(blk
->root
, errp
);