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 "sysemu/blockdev.h"
19 #include "sysemu/sysemu.h"
20 #include "qapi-event.h"
23 #include "migration/misc.h"
25 /* Number of coroutines to reserve per attached device model */
26 #define COROUTINE_POOL_RESERVATION 64
28 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
30 static AioContext
*blk_aiocb_get_aio_context(BlockAIOCB
*acb
);
36 DriveInfo
*legacy_dinfo
; /* null unless created by drive_new() */
37 QTAILQ_ENTRY(BlockBackend
) link
; /* for block_backends */
38 QTAILQ_ENTRY(BlockBackend
) monitor_link
; /* for monitor_block_backends */
39 BlockBackendPublic
public;
41 void *dev
; /* attached device model, if any */
42 bool legacy_dev
; /* true if dev is not a DeviceState */
43 /* TODO change to DeviceState when all users are qdevified */
44 const BlockDevOps
*dev_ops
;
47 /* the block size for which the guest device expects atomicity */
50 /* If the BDS tree is removed, some of its options are stored here (which
51 * can be used to restore those options in the new BDS on insert) */
52 BlockBackendRootState root_state
;
54 bool enable_write_cache
;
56 /* I/O stats (display with "info blockstats"). */
59 BlockdevOnError on_read_error
, on_write_error
;
60 bool iostatus_enabled
;
61 BlockDeviceIoStatus iostatus
;
67 bool allow_write_beyond_eof
;
69 NotifierList remove_bs_notifiers
, insert_bs_notifiers
;
72 VMChangeStateEntry
*vmsh
;
73 bool force_allow_inactivate
;
76 typedef struct BlockBackendAIOCB
{
82 static const AIOCBInfo block_backend_aiocb_info
= {
83 .get_aio_context
= blk_aiocb_get_aio_context
,
84 .aiocb_size
= sizeof(BlockBackendAIOCB
),
87 static void drive_info_del(DriveInfo
*dinfo
);
88 static BlockBackend
*bdrv_first_blk(BlockDriverState
*bs
);
90 /* All BlockBackends */
91 static QTAILQ_HEAD(, BlockBackend
) block_backends
=
92 QTAILQ_HEAD_INITIALIZER(block_backends
);
94 /* All BlockBackends referenced by the monitor and which are iterated through by
96 static QTAILQ_HEAD(, BlockBackend
) monitor_block_backends
=
97 QTAILQ_HEAD_INITIALIZER(monitor_block_backends
);
99 static void blk_root_inherit_options(int *child_flags
, QDict
*child_options
,
100 int parent_flags
, QDict
*parent_options
)
102 /* We're not supposed to call this function for root nodes */
105 static void blk_root_drained_begin(BdrvChild
*child
);
106 static void blk_root_drained_end(BdrvChild
*child
);
108 static void blk_root_change_media(BdrvChild
*child
, bool load
);
109 static void blk_root_resize(BdrvChild
*child
);
111 static char *blk_root_get_parent_desc(BdrvChild
*child
)
113 BlockBackend
*blk
= child
->opaque
;
117 return g_strdup(blk
->name
);
120 dev_id
= blk_get_attached_dev_id(blk
);
124 /* TODO Callback into the BB owner for something more detailed */
126 return g_strdup("a block device");
130 static const char *blk_root_get_name(BdrvChild
*child
)
132 return blk_name(child
->opaque
);
135 static void blk_vm_state_changed(void *opaque
, int running
, RunState state
)
137 Error
*local_err
= NULL
;
138 BlockBackend
*blk
= opaque
;
140 if (state
== RUN_STATE_INMIGRATE
) {
144 qemu_del_vm_change_state_handler(blk
->vmsh
);
146 blk_set_perm(blk
, blk
->perm
, blk
->shared_perm
, &local_err
);
148 error_report_err(local_err
);
153 * Notifies the user of the BlockBackend that migration has completed. qdev
154 * devices can tighten their permissions in response (specifically revoke
155 * shared write permissions that we needed for storage migration).
157 * If an error is returned, the VM cannot be allowed to be resumed.
159 static void blk_root_activate(BdrvChild
*child
, Error
**errp
)
161 BlockBackend
*blk
= child
->opaque
;
162 Error
*local_err
= NULL
;
164 if (!blk
->disable_perm
) {
168 blk
->disable_perm
= false;
170 blk_set_perm(blk
, blk
->perm
, BLK_PERM_ALL
, &local_err
);
172 error_propagate(errp
, local_err
);
173 blk
->disable_perm
= true;
177 if (runstate_check(RUN_STATE_INMIGRATE
)) {
178 /* Activation can happen when migration process is still active, for
179 * example when nbd_server_add is called during non-shared storage
180 * migration. Defer the shared_perm update to migration completion. */
182 blk
->vmsh
= qemu_add_vm_change_state_handler(blk_vm_state_changed
,
188 blk_set_perm(blk
, blk
->perm
, blk
->shared_perm
, &local_err
);
190 error_propagate(errp
, local_err
);
191 blk
->disable_perm
= true;
196 void blk_set_force_allow_inactivate(BlockBackend
*blk
)
198 blk
->force_allow_inactivate
= true;
201 static bool blk_can_inactivate(BlockBackend
*blk
)
203 /* If it is a guest device, inactivate is ok. */
204 if (blk
->dev
|| blk_name(blk
)[0]) {
208 /* Inactivating means no more writes to the image can be done,
209 * even if those writes would be changes invisible to the
210 * guest. For block job BBs that satisfy this, we can just allow
211 * it. This is the case for mirror job source, which is required
212 * by libvirt non-shared block migration. */
213 if (!(blk
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
))) {
217 return blk
->force_allow_inactivate
;
220 static int blk_root_inactivate(BdrvChild
*child
)
222 BlockBackend
*blk
= child
->opaque
;
224 if (blk
->disable_perm
) {
228 if (!blk_can_inactivate(blk
)) {
232 blk
->disable_perm
= true;
234 bdrv_child_try_set_perm(blk
->root
, 0, BLK_PERM_ALL
, &error_abort
);
240 static const BdrvChildRole child_root
= {
241 .inherit_options
= blk_root_inherit_options
,
243 .change_media
= blk_root_change_media
,
244 .resize
= blk_root_resize
,
245 .get_name
= blk_root_get_name
,
246 .get_parent_desc
= blk_root_get_parent_desc
,
248 .drained_begin
= blk_root_drained_begin
,
249 .drained_end
= blk_root_drained_end
,
251 .activate
= blk_root_activate
,
252 .inactivate
= blk_root_inactivate
,
256 * Create a new BlockBackend with a reference count of one.
258 * @perm is a bitmasks of BLK_PERM_* constants which describes the permissions
259 * to request for a block driver node that is attached to this BlockBackend.
260 * @shared_perm is a bitmask which describes which permissions may be granted
261 * to other users of the attached node.
262 * Both sets of permissions can be changed later using blk_set_perm().
264 * Return the new BlockBackend on success, null on failure.
266 BlockBackend
*blk_new(uint64_t perm
, uint64_t shared_perm
)
270 blk
= g_new0(BlockBackend
, 1);
273 blk
->shared_perm
= shared_perm
;
274 blk_set_enable_write_cache(blk
, true);
276 block_acct_init(&blk
->stats
);
278 notifier_list_init(&blk
->remove_bs_notifiers
);
279 notifier_list_init(&blk
->insert_bs_notifiers
);
281 QTAILQ_INSERT_TAIL(&block_backends
, blk
, link
);
286 * Creates a new BlockBackend, opens a new BlockDriverState, and connects both.
288 * Just as with bdrv_open(), after having called this function the reference to
289 * @options belongs to the block layer (even on failure).
291 * TODO: Remove @filename and @flags; it should be possible to specify a whole
292 * BDS tree just by specifying the @options QDict (or @reference,
293 * alternatively). At the time of adding this function, this is not possible,
294 * though, so callers of this function have to be able to specify @filename and
297 BlockBackend
*blk_new_open(const char *filename
, const char *reference
,
298 QDict
*options
, int flags
, Error
**errp
)
301 BlockDriverState
*bs
;
304 /* blk_new_open() is mainly used in .bdrv_create implementations and the
305 * tools where sharing isn't a concern because the BDS stays private, so we
306 * just request permission according to the flags.
308 * The exceptions are xen_disk and blockdev_init(); in these cases, the
309 * caller of blk_new_open() doesn't make use of the permissions, but they
310 * shouldn't hurt either. We can still share everything here because the
311 * guest devices will add their own blockers if they can't share. */
312 perm
= BLK_PERM_CONSISTENT_READ
;
313 if (flags
& BDRV_O_RDWR
) {
314 perm
|= BLK_PERM_WRITE
;
316 if (flags
& BDRV_O_RESIZE
) {
317 perm
|= BLK_PERM_RESIZE
;
320 blk
= blk_new(perm
, BLK_PERM_ALL
);
321 bs
= bdrv_open(filename
, reference
, options
, flags
, errp
);
327 blk
->root
= bdrv_root_attach_child(bs
, "root", &child_root
,
328 perm
, BLK_PERM_ALL
, blk
, errp
);
338 static void blk_delete(BlockBackend
*blk
)
340 assert(!blk
->refcnt
);
343 if (blk
->public.throttle_group_member
.throttle_state
) {
344 blk_io_limits_disable(blk
);
350 qemu_del_vm_change_state_handler(blk
->vmsh
);
353 assert(QLIST_EMPTY(&blk
->remove_bs_notifiers
.notifiers
));
354 assert(QLIST_EMPTY(&blk
->insert_bs_notifiers
.notifiers
));
355 QTAILQ_REMOVE(&block_backends
, blk
, link
);
356 drive_info_del(blk
->legacy_dinfo
);
357 block_acct_cleanup(&blk
->stats
);
361 static void drive_info_del(DriveInfo
*dinfo
)
366 qemu_opts_del(dinfo
->opts
);
367 g_free(dinfo
->serial
);
371 int blk_get_refcnt(BlockBackend
*blk
)
373 return blk
? blk
->refcnt
: 0;
377 * Increment @blk's reference count.
378 * @blk must not be null.
380 void blk_ref(BlockBackend
*blk
)
386 * Decrement @blk's reference count.
387 * If this drops it to zero, destroy @blk.
388 * For convenience, do nothing if @blk is null.
390 void blk_unref(BlockBackend
*blk
)
393 assert(blk
->refcnt
> 0);
394 if (!--blk
->refcnt
) {
401 * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
402 * ones which are hidden (i.e. are not referenced by the monitor).
404 BlockBackend
*blk_all_next(BlockBackend
*blk
)
406 return blk
? QTAILQ_NEXT(blk
, link
)
407 : QTAILQ_FIRST(&block_backends
);
410 void blk_remove_all_bs(void)
412 BlockBackend
*blk
= NULL
;
414 while ((blk
= blk_all_next(blk
)) != NULL
) {
415 AioContext
*ctx
= blk_get_aio_context(blk
);
417 aio_context_acquire(ctx
);
421 aio_context_release(ctx
);
426 * Return the monitor-owned BlockBackend after @blk.
427 * If @blk is null, return the first one.
428 * Else, return @blk's next sibling, which may be null.
430 * To iterate over all BlockBackends, do
431 * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
435 BlockBackend
*blk_next(BlockBackend
*blk
)
437 return blk
? QTAILQ_NEXT(blk
, monitor_link
)
438 : QTAILQ_FIRST(&monitor_block_backends
);
441 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
442 * the monitor or attached to a BlockBackend */
443 BlockDriverState
*bdrv_next(BdrvNextIterator
*it
)
445 BlockDriverState
*bs
, *old_bs
;
447 /* Must be called from the main loop */
448 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
450 /* First, return all root nodes of BlockBackends. In order to avoid
451 * returning a BDS twice when multiple BBs refer to it, we only return it
452 * if the BB is the first one in the parent list of the BDS. */
453 if (it
->phase
== BDRV_NEXT_BACKEND_ROOTS
) {
454 BlockBackend
*old_blk
= it
->blk
;
456 old_bs
= old_blk
? blk_bs(old_blk
) : NULL
;
459 it
->blk
= blk_all_next(it
->blk
);
460 bs
= it
->blk
? blk_bs(it
->blk
) : NULL
;
461 } while (it
->blk
&& (bs
== NULL
|| bdrv_first_blk(bs
) != it
->blk
));
473 it
->phase
= BDRV_NEXT_MONITOR_OWNED
;
478 /* Then return the monitor-owned BDSes without a BB attached. Ignore all
479 * BDSes that are attached to a BlockBackend here; they have been handled
480 * by the above block already */
482 it
->bs
= bdrv_next_monitor_owned(it
->bs
);
484 } while (bs
&& bdrv_has_blk(bs
));
494 static void bdrv_next_reset(BdrvNextIterator
*it
)
496 *it
= (BdrvNextIterator
) {
497 .phase
= BDRV_NEXT_BACKEND_ROOTS
,
501 BlockDriverState
*bdrv_first(BdrvNextIterator
*it
)
504 return bdrv_next(it
);
507 /* Must be called when aborting a bdrv_next() iteration before
508 * bdrv_next() returns NULL */
509 void bdrv_next_cleanup(BdrvNextIterator
*it
)
511 /* Must be called from the main loop */
512 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
514 if (it
->phase
== BDRV_NEXT_BACKEND_ROOTS
) {
516 bdrv_unref(blk_bs(it
->blk
));
527 * Add a BlockBackend into the list of backends referenced by the monitor, with
528 * the given @name acting as the handle for the monitor.
529 * Strictly for use by blockdev.c.
531 * @name must not be null or empty.
533 * Returns true on success and false on failure. In the latter case, an Error
534 * object is returned through @errp.
536 bool monitor_add_blk(BlockBackend
*blk
, const char *name
, Error
**errp
)
539 assert(name
&& name
[0]);
541 if (!id_wellformed(name
)) {
542 error_setg(errp
, "Invalid device name");
545 if (blk_by_name(name
)) {
546 error_setg(errp
, "Device with id '%s' already exists", name
);
549 if (bdrv_find_node(name
)) {
551 "Device name '%s' conflicts with an existing node name",
556 blk
->name
= g_strdup(name
);
557 QTAILQ_INSERT_TAIL(&monitor_block_backends
, blk
, monitor_link
);
562 * Remove a BlockBackend from the list of backends referenced by the monitor.
563 * Strictly for use by blockdev.c.
565 void monitor_remove_blk(BlockBackend
*blk
)
571 QTAILQ_REMOVE(&monitor_block_backends
, blk
, monitor_link
);
577 * Return @blk's name, a non-null string.
578 * Returns an empty string iff @blk is not referenced by the monitor.
580 const char *blk_name(const BlockBackend
*blk
)
582 return blk
->name
?: "";
586 * Return the BlockBackend with name @name if it exists, else null.
587 * @name must not be null.
589 BlockBackend
*blk_by_name(const char *name
)
591 BlockBackend
*blk
= NULL
;
594 while ((blk
= blk_next(blk
)) != NULL
) {
595 if (!strcmp(name
, blk
->name
)) {
603 * Return the BlockDriverState attached to @blk if any, else null.
605 BlockDriverState
*blk_bs(BlockBackend
*blk
)
607 return blk
->root
? blk
->root
->bs
: NULL
;
610 static BlockBackend
*bdrv_first_blk(BlockDriverState
*bs
)
613 QLIST_FOREACH(child
, &bs
->parents
, next_parent
) {
614 if (child
->role
== &child_root
) {
615 return child
->opaque
;
623 * Returns true if @bs has an associated BlockBackend.
625 bool bdrv_has_blk(BlockDriverState
*bs
)
627 return bdrv_first_blk(bs
) != NULL
;
631 * Returns true if @bs has only BlockBackends as parents.
633 bool bdrv_is_root_node(BlockDriverState
*bs
)
637 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
638 if (c
->role
!= &child_root
) {
647 * Return @blk's DriveInfo if any, else null.
649 DriveInfo
*blk_legacy_dinfo(BlockBackend
*blk
)
651 return blk
->legacy_dinfo
;
655 * Set @blk's DriveInfo to @dinfo, and return it.
656 * @blk must not have a DriveInfo set already.
657 * No other BlockBackend may have the same DriveInfo set.
659 DriveInfo
*blk_set_legacy_dinfo(BlockBackend
*blk
, DriveInfo
*dinfo
)
661 assert(!blk
->legacy_dinfo
);
662 return blk
->legacy_dinfo
= dinfo
;
666 * Return the BlockBackend with DriveInfo @dinfo.
669 BlockBackend
*blk_by_legacy_dinfo(DriveInfo
*dinfo
)
671 BlockBackend
*blk
= NULL
;
673 while ((blk
= blk_next(blk
)) != NULL
) {
674 if (blk
->legacy_dinfo
== dinfo
) {
682 * Returns a pointer to the publicly accessible fields of @blk.
684 BlockBackendPublic
*blk_get_public(BlockBackend
*blk
)
690 * Returns a BlockBackend given the associated @public fields.
692 BlockBackend
*blk_by_public(BlockBackendPublic
*public)
694 return container_of(public, BlockBackend
, public);
698 * Disassociates the currently associated BlockDriverState from @blk.
700 void blk_remove_bs(BlockBackend
*blk
)
702 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
703 BlockDriverState
*bs
;
705 notifier_list_notify(&blk
->remove_bs_notifiers
, blk
);
706 if (tgm
->throttle_state
) {
708 bdrv_drained_begin(bs
);
709 throttle_group_detach_aio_context(tgm
);
710 throttle_group_attach_aio_context(tgm
, qemu_get_aio_context());
711 bdrv_drained_end(bs
);
714 blk_update_root_state(blk
);
716 bdrv_root_unref_child(blk
->root
);
721 * Associates a new BlockDriverState with @blk.
723 int blk_insert_bs(BlockBackend
*blk
, BlockDriverState
*bs
, Error
**errp
)
725 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
726 blk
->root
= bdrv_root_attach_child(bs
, "root", &child_root
,
727 blk
->perm
, blk
->shared_perm
, blk
, errp
);
728 if (blk
->root
== NULL
) {
733 notifier_list_notify(&blk
->insert_bs_notifiers
, blk
);
734 if (tgm
->throttle_state
) {
735 throttle_group_detach_aio_context(tgm
);
736 throttle_group_attach_aio_context(tgm
, bdrv_get_aio_context(bs
));
743 * Sets the permission bitmasks that the user of the BlockBackend needs.
745 int blk_set_perm(BlockBackend
*blk
, uint64_t perm
, uint64_t shared_perm
,
750 if (blk
->root
&& !blk
->disable_perm
) {
751 ret
= bdrv_child_try_set_perm(blk
->root
, perm
, shared_perm
, errp
);
758 blk
->shared_perm
= shared_perm
;
763 void blk_get_perm(BlockBackend
*blk
, uint64_t *perm
, uint64_t *shared_perm
)
766 *shared_perm
= blk
->shared_perm
;
769 static int blk_do_attach_dev(BlockBackend
*blk
, void *dev
)
775 /* While migration is still incoming, we don't need to apply the
776 * permissions of guest device BlockBackends. We might still have a block
777 * job or NBD server writing to the image for storage migration. */
778 if (runstate_check(RUN_STATE_INMIGRATE
)) {
779 blk
->disable_perm
= true;
784 blk
->legacy_dev
= false;
785 blk_iostatus_reset(blk
);
791 * Attach device model @dev to @blk.
792 * Return 0 on success, -EBUSY when a device model is attached already.
794 int blk_attach_dev(BlockBackend
*blk
, DeviceState
*dev
)
796 return blk_do_attach_dev(blk
, dev
);
800 * Attach device model @dev to @blk.
801 * @blk must not have a device model attached already.
802 * TODO qdevified devices don't use this, remove when devices are qdevified
804 void blk_attach_dev_legacy(BlockBackend
*blk
, void *dev
)
806 if (blk_do_attach_dev(blk
, dev
) < 0) {
809 blk
->legacy_dev
= true;
813 * Detach device model @dev from @blk.
814 * @dev must be currently attached to @blk.
816 void blk_detach_dev(BlockBackend
*blk
, void *dev
)
817 /* TODO change to DeviceState *dev when all users are qdevified */
819 assert(blk
->dev
== dev
);
822 blk
->dev_opaque
= NULL
;
823 blk
->guest_block_size
= 512;
824 blk_set_perm(blk
, 0, BLK_PERM_ALL
, &error_abort
);
829 * Return the device model attached to @blk if any, else null.
831 void *blk_get_attached_dev(BlockBackend
*blk
)
832 /* TODO change to return DeviceState * when all users are qdevified */
837 /* Return the qdev ID, or if no ID is assigned the QOM path, of the block
838 * device attached to the BlockBackend. */
839 char *blk_get_attached_dev_id(BlockBackend
*blk
)
843 assert(!blk
->legacy_dev
);
848 } else if (dev
->id
) {
849 return g_strdup(dev
->id
);
851 return object_get_canonical_path(OBJECT(dev
));
855 * Return the BlockBackend which has the device model @dev attached if it
858 * @dev must not be null.
860 BlockBackend
*blk_by_dev(void *dev
)
862 BlockBackend
*blk
= NULL
;
865 while ((blk
= blk_all_next(blk
)) != NULL
) {
866 if (blk
->dev
== dev
) {
874 * Set @blk's device model callbacks to @ops.
875 * @opaque is the opaque argument to pass to the callbacks.
876 * This is for use by device models.
878 void blk_set_dev_ops(BlockBackend
*blk
, const BlockDevOps
*ops
,
881 /* All drivers that use blk_set_dev_ops() are qdevified and we want to keep
882 * it that way, so we can assume blk->dev, if present, is a DeviceState if
883 * blk->dev_ops is set. Non-device users may use dev_ops without device. */
884 assert(!blk
->legacy_dev
);
887 blk
->dev_opaque
= opaque
;
889 /* Are we currently quiesced? Should we enforce this right now? */
890 if (blk
->quiesce_counter
&& ops
->drained_begin
) {
891 ops
->drained_begin(opaque
);
896 * Notify @blk's attached device model of media change.
898 * If @load is true, notify of media load. This action can fail, meaning that
899 * the medium cannot be loaded. @errp is set then.
901 * If @load is false, notify of media eject. This can never fail.
903 * Also send DEVICE_TRAY_MOVED events as appropriate.
905 void blk_dev_change_media_cb(BlockBackend
*blk
, bool load
, Error
**errp
)
907 if (blk
->dev_ops
&& blk
->dev_ops
->change_media_cb
) {
908 bool tray_was_open
, tray_is_open
;
909 Error
*local_err
= NULL
;
911 assert(!blk
->legacy_dev
);
913 tray_was_open
= blk_dev_is_tray_open(blk
);
914 blk
->dev_ops
->change_media_cb(blk
->dev_opaque
, load
, &local_err
);
916 assert(load
== true);
917 error_propagate(errp
, local_err
);
920 tray_is_open
= blk_dev_is_tray_open(blk
);
922 if (tray_was_open
!= tray_is_open
) {
923 char *id
= blk_get_attached_dev_id(blk
);
924 qapi_event_send_device_tray_moved(blk_name(blk
), id
, tray_is_open
,
931 static void blk_root_change_media(BdrvChild
*child
, bool load
)
933 blk_dev_change_media_cb(child
->opaque
, load
, NULL
);
937 * Does @blk's attached device model have removable media?
938 * %true if no device model is attached.
940 bool blk_dev_has_removable_media(BlockBackend
*blk
)
942 return !blk
->dev
|| (blk
->dev_ops
&& blk
->dev_ops
->change_media_cb
);
946 * Does @blk's attached device model have a tray?
948 bool blk_dev_has_tray(BlockBackend
*blk
)
950 return blk
->dev_ops
&& blk
->dev_ops
->is_tray_open
;
954 * Notify @blk's attached device model of a media eject request.
955 * If @force is true, the medium is about to be yanked out forcefully.
957 void blk_dev_eject_request(BlockBackend
*blk
, bool force
)
959 if (blk
->dev_ops
&& blk
->dev_ops
->eject_request_cb
) {
960 blk
->dev_ops
->eject_request_cb(blk
->dev_opaque
, force
);
965 * Does @blk's attached device model have a tray, and is it open?
967 bool blk_dev_is_tray_open(BlockBackend
*blk
)
969 if (blk_dev_has_tray(blk
)) {
970 return blk
->dev_ops
->is_tray_open(blk
->dev_opaque
);
976 * Does @blk's attached device model have the medium locked?
977 * %false if the device model has no such lock.
979 bool blk_dev_is_medium_locked(BlockBackend
*blk
)
981 if (blk
->dev_ops
&& blk
->dev_ops
->is_medium_locked
) {
982 return blk
->dev_ops
->is_medium_locked(blk
->dev_opaque
);
988 * Notify @blk's attached device model of a backend size change.
990 static void blk_root_resize(BdrvChild
*child
)
992 BlockBackend
*blk
= child
->opaque
;
994 if (blk
->dev_ops
&& blk
->dev_ops
->resize_cb
) {
995 blk
->dev_ops
->resize_cb(blk
->dev_opaque
);
999 void blk_iostatus_enable(BlockBackend
*blk
)
1001 blk
->iostatus_enabled
= true;
1002 blk
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
1005 /* The I/O status is only enabled if the drive explicitly
1006 * enables it _and_ the VM is configured to stop on errors */
1007 bool blk_iostatus_is_enabled(const BlockBackend
*blk
)
1009 return (blk
->iostatus_enabled
&&
1010 (blk
->on_write_error
== BLOCKDEV_ON_ERROR_ENOSPC
||
1011 blk
->on_write_error
== BLOCKDEV_ON_ERROR_STOP
||
1012 blk
->on_read_error
== BLOCKDEV_ON_ERROR_STOP
));
1015 BlockDeviceIoStatus
blk_iostatus(const BlockBackend
*blk
)
1017 return blk
->iostatus
;
1020 void blk_iostatus_disable(BlockBackend
*blk
)
1022 blk
->iostatus_enabled
= false;
1025 void blk_iostatus_reset(BlockBackend
*blk
)
1027 if (blk_iostatus_is_enabled(blk
)) {
1028 BlockDriverState
*bs
= blk_bs(blk
);
1029 blk
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
1030 if (bs
&& bs
->job
) {
1031 block_job_iostatus_reset(bs
->job
);
1036 void blk_iostatus_set_err(BlockBackend
*blk
, int error
)
1038 assert(blk_iostatus_is_enabled(blk
));
1039 if (blk
->iostatus
== BLOCK_DEVICE_IO_STATUS_OK
) {
1040 blk
->iostatus
= error
== ENOSPC
? BLOCK_DEVICE_IO_STATUS_NOSPACE
:
1041 BLOCK_DEVICE_IO_STATUS_FAILED
;
1045 void blk_set_allow_write_beyond_eof(BlockBackend
*blk
, bool allow
)
1047 blk
->allow_write_beyond_eof
= allow
;
1050 static int blk_check_byte_request(BlockBackend
*blk
, int64_t offset
,
1055 if (size
> INT_MAX
) {
1059 if (!blk_is_available(blk
)) {
1067 if (!blk
->allow_write_beyond_eof
) {
1068 len
= blk_getlength(blk
);
1073 if (offset
> len
|| len
- offset
< size
) {
1081 int coroutine_fn
blk_co_preadv(BlockBackend
*blk
, int64_t offset
,
1082 unsigned int bytes
, QEMUIOVector
*qiov
,
1083 BdrvRequestFlags flags
)
1086 BlockDriverState
*bs
= blk_bs(blk
);
1088 trace_blk_co_preadv(blk
, bs
, offset
, bytes
, flags
);
1090 ret
= blk_check_byte_request(blk
, offset
, bytes
);
1095 bdrv_inc_in_flight(bs
);
1097 /* throttling disk I/O */
1098 if (blk
->public.throttle_group_member
.throttle_state
) {
1099 throttle_group_co_io_limits_intercept(&blk
->public.throttle_group_member
,
1103 ret
= bdrv_co_preadv(blk
->root
, offset
, bytes
, qiov
, flags
);
1104 bdrv_dec_in_flight(bs
);
1108 int coroutine_fn
blk_co_pwritev(BlockBackend
*blk
, int64_t offset
,
1109 unsigned int bytes
, QEMUIOVector
*qiov
,
1110 BdrvRequestFlags flags
)
1113 BlockDriverState
*bs
= blk_bs(blk
);
1115 trace_blk_co_pwritev(blk
, bs
, offset
, bytes
, flags
);
1117 ret
= blk_check_byte_request(blk
, offset
, bytes
);
1122 bdrv_inc_in_flight(bs
);
1123 /* throttling disk I/O */
1124 if (blk
->public.throttle_group_member
.throttle_state
) {
1125 throttle_group_co_io_limits_intercept(&blk
->public.throttle_group_member
,
1129 if (!blk
->enable_write_cache
) {
1130 flags
|= BDRV_REQ_FUA
;
1133 ret
= bdrv_co_pwritev(blk
->root
, offset
, bytes
, qiov
, flags
);
1134 bdrv_dec_in_flight(bs
);
1138 typedef struct BlkRwCo
{
1143 BdrvRequestFlags flags
;
1146 static void blk_read_entry(void *opaque
)
1148 BlkRwCo
*rwco
= opaque
;
1150 rwco
->ret
= blk_co_preadv(rwco
->blk
, rwco
->offset
, rwco
->qiov
->size
,
1151 rwco
->qiov
, rwco
->flags
);
1154 static void blk_write_entry(void *opaque
)
1156 BlkRwCo
*rwco
= opaque
;
1158 rwco
->ret
= blk_co_pwritev(rwco
->blk
, rwco
->offset
, rwco
->qiov
->size
,
1159 rwco
->qiov
, rwco
->flags
);
1162 static int blk_prw(BlockBackend
*blk
, int64_t offset
, uint8_t *buf
,
1163 int64_t bytes
, CoroutineEntry co_entry
,
1164 BdrvRequestFlags flags
)
1170 iov
= (struct iovec
) {
1174 qemu_iovec_init_external(&qiov
, &iov
, 1);
1184 if (qemu_in_coroutine()) {
1185 /* Fast-path if already in coroutine context */
1188 Coroutine
*co
= qemu_coroutine_create(co_entry
, &rwco
);
1189 bdrv_coroutine_enter(blk_bs(blk
), co
);
1190 BDRV_POLL_WHILE(blk_bs(blk
), rwco
.ret
== NOT_DONE
);
1196 int blk_pread_unthrottled(BlockBackend
*blk
, int64_t offset
, uint8_t *buf
,
1201 ret
= blk_check_byte_request(blk
, offset
, count
);
1206 blk_root_drained_begin(blk
->root
);
1207 ret
= blk_pread(blk
, offset
, buf
, count
);
1208 blk_root_drained_end(blk
->root
);
1212 int blk_pwrite_zeroes(BlockBackend
*blk
, int64_t offset
,
1213 int bytes
, BdrvRequestFlags flags
)
1215 return blk_prw(blk
, offset
, NULL
, bytes
, blk_write_entry
,
1216 flags
| BDRV_REQ_ZERO_WRITE
);
1219 int blk_make_zero(BlockBackend
*blk
, BdrvRequestFlags flags
)
1221 return bdrv_make_zero(blk
->root
, flags
);
1224 static void error_callback_bh(void *opaque
)
1226 struct BlockBackendAIOCB
*acb
= opaque
;
1228 bdrv_dec_in_flight(acb
->common
.bs
);
1229 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
1230 qemu_aio_unref(acb
);
1233 BlockAIOCB
*blk_abort_aio_request(BlockBackend
*blk
,
1234 BlockCompletionFunc
*cb
,
1235 void *opaque
, int ret
)
1237 struct BlockBackendAIOCB
*acb
;
1239 bdrv_inc_in_flight(blk_bs(blk
));
1240 acb
= blk_aio_get(&block_backend_aiocb_info
, blk
, cb
, opaque
);
1244 aio_bh_schedule_oneshot(blk_get_aio_context(blk
), error_callback_bh
, acb
);
1245 return &acb
->common
;
1248 typedef struct BlkAioEmAIOCB
{
1255 static const AIOCBInfo blk_aio_em_aiocb_info
= {
1256 .aiocb_size
= sizeof(BlkAioEmAIOCB
),
1259 static void blk_aio_complete(BlkAioEmAIOCB
*acb
)
1261 if (acb
->has_returned
) {
1262 bdrv_dec_in_flight(acb
->common
.bs
);
1263 acb
->common
.cb(acb
->common
.opaque
, acb
->rwco
.ret
);
1264 qemu_aio_unref(acb
);
1268 static void blk_aio_complete_bh(void *opaque
)
1270 BlkAioEmAIOCB
*acb
= opaque
;
1271 assert(acb
->has_returned
);
1272 blk_aio_complete(acb
);
1275 static BlockAIOCB
*blk_aio_prwv(BlockBackend
*blk
, int64_t offset
, int bytes
,
1276 QEMUIOVector
*qiov
, CoroutineEntry co_entry
,
1277 BdrvRequestFlags flags
,
1278 BlockCompletionFunc
*cb
, void *opaque
)
1283 bdrv_inc_in_flight(blk_bs(blk
));
1284 acb
= blk_aio_get(&blk_aio_em_aiocb_info
, blk
, cb
, opaque
);
1285 acb
->rwco
= (BlkRwCo
) {
1293 acb
->has_returned
= false;
1295 co
= qemu_coroutine_create(co_entry
, acb
);
1296 bdrv_coroutine_enter(blk_bs(blk
), co
);
1298 acb
->has_returned
= true;
1299 if (acb
->rwco
.ret
!= NOT_DONE
) {
1300 aio_bh_schedule_oneshot(blk_get_aio_context(blk
),
1301 blk_aio_complete_bh
, acb
);
1304 return &acb
->common
;
1307 static void blk_aio_read_entry(void *opaque
)
1309 BlkAioEmAIOCB
*acb
= opaque
;
1310 BlkRwCo
*rwco
= &acb
->rwco
;
1312 assert(rwco
->qiov
->size
== acb
->bytes
);
1313 rwco
->ret
= blk_co_preadv(rwco
->blk
, rwco
->offset
, acb
->bytes
,
1314 rwco
->qiov
, rwco
->flags
);
1315 blk_aio_complete(acb
);
1318 static void blk_aio_write_entry(void *opaque
)
1320 BlkAioEmAIOCB
*acb
= opaque
;
1321 BlkRwCo
*rwco
= &acb
->rwco
;
1323 assert(!rwco
->qiov
|| rwco
->qiov
->size
== acb
->bytes
);
1324 rwco
->ret
= blk_co_pwritev(rwco
->blk
, rwco
->offset
, acb
->bytes
,
1325 rwco
->qiov
, rwco
->flags
);
1326 blk_aio_complete(acb
);
1329 BlockAIOCB
*blk_aio_pwrite_zeroes(BlockBackend
*blk
, int64_t offset
,
1330 int count
, BdrvRequestFlags flags
,
1331 BlockCompletionFunc
*cb
, void *opaque
)
1333 return blk_aio_prwv(blk
, offset
, count
, NULL
, blk_aio_write_entry
,
1334 flags
| BDRV_REQ_ZERO_WRITE
, cb
, opaque
);
1337 int blk_pread(BlockBackend
*blk
, int64_t offset
, void *buf
, int count
)
1339 int ret
= blk_prw(blk
, offset
, buf
, count
, blk_read_entry
, 0);
1346 int blk_pwrite(BlockBackend
*blk
, int64_t offset
, const void *buf
, int count
,
1347 BdrvRequestFlags flags
)
1349 int ret
= blk_prw(blk
, offset
, (void *) buf
, count
, blk_write_entry
,
1357 int64_t blk_getlength(BlockBackend
*blk
)
1359 if (!blk_is_available(blk
)) {
1363 return bdrv_getlength(blk_bs(blk
));
1366 void blk_get_geometry(BlockBackend
*blk
, uint64_t *nb_sectors_ptr
)
1369 *nb_sectors_ptr
= 0;
1371 bdrv_get_geometry(blk_bs(blk
), nb_sectors_ptr
);
1375 int64_t blk_nb_sectors(BlockBackend
*blk
)
1377 if (!blk_is_available(blk
)) {
1381 return bdrv_nb_sectors(blk_bs(blk
));
1384 BlockAIOCB
*blk_aio_preadv(BlockBackend
*blk
, int64_t offset
,
1385 QEMUIOVector
*qiov
, BdrvRequestFlags flags
,
1386 BlockCompletionFunc
*cb
, void *opaque
)
1388 return blk_aio_prwv(blk
, offset
, qiov
->size
, qiov
,
1389 blk_aio_read_entry
, flags
, cb
, opaque
);
1392 BlockAIOCB
*blk_aio_pwritev(BlockBackend
*blk
, int64_t offset
,
1393 QEMUIOVector
*qiov
, BdrvRequestFlags flags
,
1394 BlockCompletionFunc
*cb
, void *opaque
)
1396 return blk_aio_prwv(blk
, offset
, qiov
->size
, qiov
,
1397 blk_aio_write_entry
, flags
, cb
, opaque
);
1400 static void blk_aio_flush_entry(void *opaque
)
1402 BlkAioEmAIOCB
*acb
= opaque
;
1403 BlkRwCo
*rwco
= &acb
->rwco
;
1405 rwco
->ret
= blk_co_flush(rwco
->blk
);
1406 blk_aio_complete(acb
);
1409 BlockAIOCB
*blk_aio_flush(BlockBackend
*blk
,
1410 BlockCompletionFunc
*cb
, void *opaque
)
1412 return blk_aio_prwv(blk
, 0, 0, NULL
, blk_aio_flush_entry
, 0, cb
, opaque
);
1415 static void blk_aio_pdiscard_entry(void *opaque
)
1417 BlkAioEmAIOCB
*acb
= opaque
;
1418 BlkRwCo
*rwco
= &acb
->rwco
;
1420 rwco
->ret
= blk_co_pdiscard(rwco
->blk
, rwco
->offset
, acb
->bytes
);
1421 blk_aio_complete(acb
);
1424 BlockAIOCB
*blk_aio_pdiscard(BlockBackend
*blk
,
1425 int64_t offset
, int bytes
,
1426 BlockCompletionFunc
*cb
, void *opaque
)
1428 return blk_aio_prwv(blk
, offset
, bytes
, NULL
, blk_aio_pdiscard_entry
, 0,
1432 void blk_aio_cancel(BlockAIOCB
*acb
)
1434 bdrv_aio_cancel(acb
);
1437 void blk_aio_cancel_async(BlockAIOCB
*acb
)
1439 bdrv_aio_cancel_async(acb
);
1442 int blk_co_ioctl(BlockBackend
*blk
, unsigned long int req
, void *buf
)
1444 if (!blk_is_available(blk
)) {
1448 return bdrv_co_ioctl(blk_bs(blk
), req
, buf
);
1451 static void blk_ioctl_entry(void *opaque
)
1453 BlkRwCo
*rwco
= opaque
;
1454 rwco
->ret
= blk_co_ioctl(rwco
->blk
, rwco
->offset
,
1455 rwco
->qiov
->iov
[0].iov_base
);
1458 int blk_ioctl(BlockBackend
*blk
, unsigned long int req
, void *buf
)
1460 return blk_prw(blk
, req
, buf
, 0, blk_ioctl_entry
, 0);
1463 static void blk_aio_ioctl_entry(void *opaque
)
1465 BlkAioEmAIOCB
*acb
= opaque
;
1466 BlkRwCo
*rwco
= &acb
->rwco
;
1468 rwco
->ret
= blk_co_ioctl(rwco
->blk
, rwco
->offset
,
1469 rwco
->qiov
->iov
[0].iov_base
);
1470 blk_aio_complete(acb
);
1473 BlockAIOCB
*blk_aio_ioctl(BlockBackend
*blk
, unsigned long int req
, void *buf
,
1474 BlockCompletionFunc
*cb
, void *opaque
)
1479 iov
= (struct iovec
) {
1483 qemu_iovec_init_external(&qiov
, &iov
, 1);
1485 return blk_aio_prwv(blk
, req
, 0, &qiov
, blk_aio_ioctl_entry
, 0, cb
, opaque
);
1488 int blk_co_pdiscard(BlockBackend
*blk
, int64_t offset
, int bytes
)
1490 int ret
= blk_check_byte_request(blk
, offset
, bytes
);
1495 return bdrv_co_pdiscard(blk_bs(blk
), offset
, bytes
);
1498 int blk_co_flush(BlockBackend
*blk
)
1500 if (!blk_is_available(blk
)) {
1504 return bdrv_co_flush(blk_bs(blk
));
1507 static void blk_flush_entry(void *opaque
)
1509 BlkRwCo
*rwco
= opaque
;
1510 rwco
->ret
= blk_co_flush(rwco
->blk
);
1513 int blk_flush(BlockBackend
*blk
)
1515 return blk_prw(blk
, 0, NULL
, 0, blk_flush_entry
, 0);
1518 void blk_drain(BlockBackend
*blk
)
1521 bdrv_drain(blk_bs(blk
));
1525 void blk_drain_all(void)
1530 void blk_set_on_error(BlockBackend
*blk
, BlockdevOnError on_read_error
,
1531 BlockdevOnError on_write_error
)
1533 blk
->on_read_error
= on_read_error
;
1534 blk
->on_write_error
= on_write_error
;
1537 BlockdevOnError
blk_get_on_error(BlockBackend
*blk
, bool is_read
)
1539 return is_read
? blk
->on_read_error
: blk
->on_write_error
;
1542 BlockErrorAction
blk_get_error_action(BlockBackend
*blk
, bool is_read
,
1545 BlockdevOnError on_err
= blk_get_on_error(blk
, is_read
);
1548 case BLOCKDEV_ON_ERROR_ENOSPC
:
1549 return (error
== ENOSPC
) ?
1550 BLOCK_ERROR_ACTION_STOP
: BLOCK_ERROR_ACTION_REPORT
;
1551 case BLOCKDEV_ON_ERROR_STOP
:
1552 return BLOCK_ERROR_ACTION_STOP
;
1553 case BLOCKDEV_ON_ERROR_REPORT
:
1554 return BLOCK_ERROR_ACTION_REPORT
;
1555 case BLOCKDEV_ON_ERROR_IGNORE
:
1556 return BLOCK_ERROR_ACTION_IGNORE
;
1557 case BLOCKDEV_ON_ERROR_AUTO
:
1563 static void send_qmp_error_event(BlockBackend
*blk
,
1564 BlockErrorAction action
,
1565 bool is_read
, int error
)
1567 IoOperationType optype
;
1569 optype
= is_read
? IO_OPERATION_TYPE_READ
: IO_OPERATION_TYPE_WRITE
;
1570 qapi_event_send_block_io_error(blk_name(blk
),
1571 bdrv_get_node_name(blk_bs(blk
)), optype
,
1572 action
, blk_iostatus_is_enabled(blk
),
1573 error
== ENOSPC
, strerror(error
),
1577 /* This is done by device models because, while the block layer knows
1578 * about the error, it does not know whether an operation comes from
1579 * the device or the block layer (from a job, for example).
1581 void blk_error_action(BlockBackend
*blk
, BlockErrorAction action
,
1582 bool is_read
, int error
)
1586 if (action
== BLOCK_ERROR_ACTION_STOP
) {
1587 /* First set the iostatus, so that "info block" returns an iostatus
1588 * that matches the events raised so far (an additional error iostatus
1589 * is fine, but not a lost one).
1591 blk_iostatus_set_err(blk
, error
);
1593 /* Then raise the request to stop the VM and the event.
1594 * qemu_system_vmstop_request_prepare has two effects. First,
1595 * it ensures that the STOP event always comes after the
1596 * BLOCK_IO_ERROR event. Second, it ensures that even if management
1597 * can observe the STOP event and do a "cont" before the STOP
1598 * event is issued, the VM will not stop. In this case, vm_start()
1599 * also ensures that the STOP/RESUME pair of events is emitted.
1601 qemu_system_vmstop_request_prepare();
1602 send_qmp_error_event(blk
, action
, is_read
, error
);
1603 qemu_system_vmstop_request(RUN_STATE_IO_ERROR
);
1605 send_qmp_error_event(blk
, action
, is_read
, error
);
1609 int blk_is_read_only(BlockBackend
*blk
)
1611 BlockDriverState
*bs
= blk_bs(blk
);
1614 return bdrv_is_read_only(bs
);
1616 return blk
->root_state
.read_only
;
1620 int blk_is_sg(BlockBackend
*blk
)
1622 BlockDriverState
*bs
= blk_bs(blk
);
1628 return bdrv_is_sg(bs
);
1631 int blk_enable_write_cache(BlockBackend
*blk
)
1633 return blk
->enable_write_cache
;
1636 void blk_set_enable_write_cache(BlockBackend
*blk
, bool wce
)
1638 blk
->enable_write_cache
= wce
;
1641 void blk_invalidate_cache(BlockBackend
*blk
, Error
**errp
)
1643 BlockDriverState
*bs
= blk_bs(blk
);
1646 error_setg(errp
, "Device '%s' has no medium", blk
->name
);
1650 bdrv_invalidate_cache(bs
, errp
);
1653 bool blk_is_inserted(BlockBackend
*blk
)
1655 BlockDriverState
*bs
= blk_bs(blk
);
1657 return bs
&& bdrv_is_inserted(bs
);
1660 bool blk_is_available(BlockBackend
*blk
)
1662 return blk_is_inserted(blk
) && !blk_dev_is_tray_open(blk
);
1665 void blk_lock_medium(BlockBackend
*blk
, bool locked
)
1667 BlockDriverState
*bs
= blk_bs(blk
);
1670 bdrv_lock_medium(bs
, locked
);
1674 void blk_eject(BlockBackend
*blk
, bool eject_flag
)
1676 BlockDriverState
*bs
= blk_bs(blk
);
1679 /* blk_eject is only called by qdevified devices */
1680 assert(!blk
->legacy_dev
);
1683 bdrv_eject(bs
, eject_flag
);
1686 /* Whether or not we ejected on the backend,
1687 * the frontend experienced a tray event. */
1688 id
= blk_get_attached_dev_id(blk
);
1689 qapi_event_send_device_tray_moved(blk_name(blk
), id
,
1690 eject_flag
, &error_abort
);
1694 int blk_get_flags(BlockBackend
*blk
)
1696 BlockDriverState
*bs
= blk_bs(blk
);
1699 return bdrv_get_flags(bs
);
1701 return blk
->root_state
.open_flags
;
1705 /* Returns the maximum transfer length, in bytes; guaranteed nonzero */
1706 uint32_t blk_get_max_transfer(BlockBackend
*blk
)
1708 BlockDriverState
*bs
= blk_bs(blk
);
1712 max
= bs
->bl
.max_transfer
;
1714 return MIN_NON_ZERO(max
, INT_MAX
);
1717 int blk_get_max_iov(BlockBackend
*blk
)
1719 return blk
->root
->bs
->bl
.max_iov
;
1722 void blk_set_guest_block_size(BlockBackend
*blk
, int align
)
1724 blk
->guest_block_size
= align
;
1727 void *blk_try_blockalign(BlockBackend
*blk
, size_t size
)
1729 return qemu_try_blockalign(blk
? blk_bs(blk
) : NULL
, size
);
1732 void *blk_blockalign(BlockBackend
*blk
, size_t size
)
1734 return qemu_blockalign(blk
? blk_bs(blk
) : NULL
, size
);
1737 bool blk_op_is_blocked(BlockBackend
*blk
, BlockOpType op
, Error
**errp
)
1739 BlockDriverState
*bs
= blk_bs(blk
);
1745 return bdrv_op_is_blocked(bs
, op
, errp
);
1748 void blk_op_unblock(BlockBackend
*blk
, BlockOpType op
, Error
*reason
)
1750 BlockDriverState
*bs
= blk_bs(blk
);
1753 bdrv_op_unblock(bs
, op
, reason
);
1757 void blk_op_block_all(BlockBackend
*blk
, Error
*reason
)
1759 BlockDriverState
*bs
= blk_bs(blk
);
1762 bdrv_op_block_all(bs
, reason
);
1766 void blk_op_unblock_all(BlockBackend
*blk
, Error
*reason
)
1768 BlockDriverState
*bs
= blk_bs(blk
);
1771 bdrv_op_unblock_all(bs
, reason
);
1775 AioContext
*blk_get_aio_context(BlockBackend
*blk
)
1777 BlockDriverState
*bs
= blk_bs(blk
);
1780 return bdrv_get_aio_context(bs
);
1782 return qemu_get_aio_context();
1786 static AioContext
*blk_aiocb_get_aio_context(BlockAIOCB
*acb
)
1788 BlockBackendAIOCB
*blk_acb
= DO_UPCAST(BlockBackendAIOCB
, common
, acb
);
1789 return blk_get_aio_context(blk_acb
->blk
);
1792 void blk_set_aio_context(BlockBackend
*blk
, AioContext
*new_context
)
1794 BlockDriverState
*bs
= blk_bs(blk
);
1795 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
1798 if (tgm
->throttle_state
) {
1799 bdrv_drained_begin(bs
);
1800 throttle_group_detach_aio_context(tgm
);
1801 throttle_group_attach_aio_context(tgm
, new_context
);
1802 bdrv_drained_end(bs
);
1804 bdrv_set_aio_context(bs
, new_context
);
1808 void blk_add_aio_context_notifier(BlockBackend
*blk
,
1809 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
1810 void (*detach_aio_context
)(void *opaque
), void *opaque
)
1812 BlockDriverState
*bs
= blk_bs(blk
);
1815 bdrv_add_aio_context_notifier(bs
, attached_aio_context
,
1816 detach_aio_context
, opaque
);
1820 void blk_remove_aio_context_notifier(BlockBackend
*blk
,
1821 void (*attached_aio_context
)(AioContext
*,
1823 void (*detach_aio_context
)(void *),
1826 BlockDriverState
*bs
= blk_bs(blk
);
1829 bdrv_remove_aio_context_notifier(bs
, attached_aio_context
,
1830 detach_aio_context
, opaque
);
1834 void blk_add_remove_bs_notifier(BlockBackend
*blk
, Notifier
*notify
)
1836 notifier_list_add(&blk
->remove_bs_notifiers
, notify
);
1839 void blk_add_insert_bs_notifier(BlockBackend
*blk
, Notifier
*notify
)
1841 notifier_list_add(&blk
->insert_bs_notifiers
, notify
);
1844 void blk_io_plug(BlockBackend
*blk
)
1846 BlockDriverState
*bs
= blk_bs(blk
);
1853 void blk_io_unplug(BlockBackend
*blk
)
1855 BlockDriverState
*bs
= blk_bs(blk
);
1862 BlockAcctStats
*blk_get_stats(BlockBackend
*blk
)
1867 void *blk_aio_get(const AIOCBInfo
*aiocb_info
, BlockBackend
*blk
,
1868 BlockCompletionFunc
*cb
, void *opaque
)
1870 return qemu_aio_get(aiocb_info
, blk_bs(blk
), cb
, opaque
);
1873 int coroutine_fn
blk_co_pwrite_zeroes(BlockBackend
*blk
, int64_t offset
,
1874 int bytes
, BdrvRequestFlags flags
)
1876 return blk_co_pwritev(blk
, offset
, bytes
, NULL
,
1877 flags
| BDRV_REQ_ZERO_WRITE
);
1880 int blk_pwrite_compressed(BlockBackend
*blk
, int64_t offset
, const void *buf
,
1883 return blk_prw(blk
, offset
, (void *) buf
, count
, blk_write_entry
,
1884 BDRV_REQ_WRITE_COMPRESSED
);
1887 int blk_truncate(BlockBackend
*blk
, int64_t offset
, PreallocMode prealloc
,
1890 if (!blk_is_available(blk
)) {
1891 error_setg(errp
, "No medium inserted");
1895 return bdrv_truncate(blk
->root
, offset
, prealloc
, errp
);
1898 static void blk_pdiscard_entry(void *opaque
)
1900 BlkRwCo
*rwco
= opaque
;
1901 rwco
->ret
= blk_co_pdiscard(rwco
->blk
, rwco
->offset
, rwco
->qiov
->size
);
1904 int blk_pdiscard(BlockBackend
*blk
, int64_t offset
, int bytes
)
1906 return blk_prw(blk
, offset
, NULL
, bytes
, blk_pdiscard_entry
, 0);
1909 int blk_save_vmstate(BlockBackend
*blk
, const uint8_t *buf
,
1910 int64_t pos
, int size
)
1914 if (!blk_is_available(blk
)) {
1918 ret
= bdrv_save_vmstate(blk_bs(blk
), buf
, pos
, size
);
1923 if (ret
== size
&& !blk
->enable_write_cache
) {
1924 ret
= bdrv_flush(blk_bs(blk
));
1927 return ret
< 0 ? ret
: size
;
1930 int blk_load_vmstate(BlockBackend
*blk
, uint8_t *buf
, int64_t pos
, int size
)
1932 if (!blk_is_available(blk
)) {
1936 return bdrv_load_vmstate(blk_bs(blk
), buf
, pos
, size
);
1939 int blk_probe_blocksizes(BlockBackend
*blk
, BlockSizes
*bsz
)
1941 if (!blk_is_available(blk
)) {
1945 return bdrv_probe_blocksizes(blk_bs(blk
), bsz
);
1948 int blk_probe_geometry(BlockBackend
*blk
, HDGeometry
*geo
)
1950 if (!blk_is_available(blk
)) {
1954 return bdrv_probe_geometry(blk_bs(blk
), geo
);
1958 * Updates the BlockBackendRootState object with data from the currently
1959 * attached BlockDriverState.
1961 void blk_update_root_state(BlockBackend
*blk
)
1965 blk
->root_state
.open_flags
= blk
->root
->bs
->open_flags
;
1966 blk
->root_state
.read_only
= blk
->root
->bs
->read_only
;
1967 blk
->root_state
.detect_zeroes
= blk
->root
->bs
->detect_zeroes
;
1971 * Returns the detect-zeroes setting to be used for bdrv_open() of a
1972 * BlockDriverState which is supposed to inherit the root state.
1974 bool blk_get_detect_zeroes_from_root_state(BlockBackend
*blk
)
1976 return blk
->root_state
.detect_zeroes
;
1980 * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
1981 * supposed to inherit the root state.
1983 int blk_get_open_flags_from_root_state(BlockBackend
*blk
)
1987 bs_flags
= blk
->root_state
.read_only
? 0 : BDRV_O_RDWR
;
1988 bs_flags
|= blk
->root_state
.open_flags
& ~BDRV_O_RDWR
;
1993 BlockBackendRootState
*blk_get_root_state(BlockBackend
*blk
)
1995 return &blk
->root_state
;
1998 int blk_commit_all(void)
2000 BlockBackend
*blk
= NULL
;
2002 while ((blk
= blk_all_next(blk
)) != NULL
) {
2003 AioContext
*aio_context
= blk_get_aio_context(blk
);
2005 aio_context_acquire(aio_context
);
2006 if (blk_is_inserted(blk
) && blk
->root
->bs
->backing
) {
2007 int ret
= bdrv_commit(blk
->root
->bs
);
2009 aio_context_release(aio_context
);
2013 aio_context_release(aio_context
);
2019 /* throttling disk I/O limits */
2020 void blk_set_io_limits(BlockBackend
*blk
, ThrottleConfig
*cfg
)
2022 throttle_group_config(&blk
->public.throttle_group_member
, cfg
);
2025 void blk_io_limits_disable(BlockBackend
*blk
)
2027 BlockDriverState
*bs
= blk_bs(blk
);
2028 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
2029 assert(tgm
->throttle_state
);
2031 bdrv_drained_begin(bs
);
2033 throttle_group_unregister_tgm(tgm
);
2035 bdrv_drained_end(bs
);
2039 /* should be called before blk_set_io_limits if a limit is set */
2040 void blk_io_limits_enable(BlockBackend
*blk
, const char *group
)
2042 assert(!blk
->public.throttle_group_member
.throttle_state
);
2043 throttle_group_register_tgm(&blk
->public.throttle_group_member
,
2044 group
, blk_get_aio_context(blk
));
2047 void blk_io_limits_update_group(BlockBackend
*blk
, const char *group
)
2049 /* this BB is not part of any group */
2050 if (!blk
->public.throttle_group_member
.throttle_state
) {
2054 /* this BB is a part of the same group than the one we want */
2055 if (!g_strcmp0(throttle_group_get_name(&blk
->public.throttle_group_member
),
2060 /* need to change the group this bs belong to */
2061 blk_io_limits_disable(blk
);
2062 blk_io_limits_enable(blk
, group
);
2065 static void blk_root_drained_begin(BdrvChild
*child
)
2067 BlockBackend
*blk
= child
->opaque
;
2069 if (++blk
->quiesce_counter
== 1) {
2070 if (blk
->dev_ops
&& blk
->dev_ops
->drained_begin
) {
2071 blk
->dev_ops
->drained_begin(blk
->dev_opaque
);
2075 /* Note that blk->root may not be accessible here yet if we are just
2076 * attaching to a BlockDriverState that is drained. Use child instead. */
2078 if (atomic_fetch_inc(&blk
->public.throttle_group_member
.io_limits_disabled
) == 0) {
2079 throttle_group_restart_tgm(&blk
->public.throttle_group_member
);
2083 static void blk_root_drained_end(BdrvChild
*child
)
2085 BlockBackend
*blk
= child
->opaque
;
2086 assert(blk
->quiesce_counter
);
2088 assert(blk
->public.throttle_group_member
.io_limits_disabled
);
2089 atomic_dec(&blk
->public.throttle_group_member
.io_limits_disabled
);
2091 if (--blk
->quiesce_counter
== 0) {
2092 if (blk
->dev_ops
&& blk
->dev_ops
->drained_end
) {
2093 blk
->dev_ops
->drained_end(blk
->dev_opaque
);