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/error.h"
21 #include "qapi/qapi-events-block.h"
23 #include "qemu/option.h"
25 #include "migration/misc.h"
27 /* Number of coroutines to reserve per attached device model */
28 #define COROUTINE_POOL_RESERVATION 64
30 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
32 static AioContext
*blk_aiocb_get_aio_context(BlockAIOCB
*acb
);
34 typedef struct BlockBackendAioNotifier
{
35 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
);
36 void (*detach_aio_context
)(void *opaque
);
38 QLIST_ENTRY(BlockBackendAioNotifier
) list
;
39 } BlockBackendAioNotifier
;
45 DriveInfo
*legacy_dinfo
; /* null unless created by drive_new() */
46 QTAILQ_ENTRY(BlockBackend
) link
; /* for block_backends */
47 QTAILQ_ENTRY(BlockBackend
) monitor_link
; /* for monitor_block_backends */
48 BlockBackendPublic
public;
50 void *dev
; /* attached device model, if any */
51 bool legacy_dev
; /* true if dev is not a DeviceState */
52 /* TODO change to DeviceState when all users are qdevified */
53 const BlockDevOps
*dev_ops
;
56 /* the block size for which the guest device expects atomicity */
59 /* If the BDS tree is removed, some of its options are stored here (which
60 * can be used to restore those options in the new BDS on insert) */
61 BlockBackendRootState root_state
;
63 bool enable_write_cache
;
65 /* I/O stats (display with "info blockstats"). */
68 BlockdevOnError on_read_error
, on_write_error
;
69 bool iostatus_enabled
;
70 BlockDeviceIoStatus iostatus
;
76 bool allow_write_beyond_eof
;
78 NotifierList remove_bs_notifiers
, insert_bs_notifiers
;
79 QLIST_HEAD(, BlockBackendAioNotifier
) aio_notifiers
;
82 VMChangeStateEntry
*vmsh
;
83 bool force_allow_inactivate
;
85 /* Number of in-flight aio requests. BlockDriverState also counts
86 * in-flight requests but aio requests can exist even when blk->root is
87 * NULL, so we cannot rely on its counter for that case.
88 * Accessed with atomic ops.
90 unsigned int in_flight
;
93 typedef struct BlockBackendAIOCB
{
99 static const AIOCBInfo block_backend_aiocb_info
= {
100 .get_aio_context
= blk_aiocb_get_aio_context
,
101 .aiocb_size
= sizeof(BlockBackendAIOCB
),
104 static void drive_info_del(DriveInfo
*dinfo
);
105 static BlockBackend
*bdrv_first_blk(BlockDriverState
*bs
);
107 /* All BlockBackends */
108 static QTAILQ_HEAD(, BlockBackend
) block_backends
=
109 QTAILQ_HEAD_INITIALIZER(block_backends
);
111 /* All BlockBackends referenced by the monitor and which are iterated through by
113 static QTAILQ_HEAD(, BlockBackend
) monitor_block_backends
=
114 QTAILQ_HEAD_INITIALIZER(monitor_block_backends
);
116 static void blk_root_inherit_options(int *child_flags
, QDict
*child_options
,
117 int parent_flags
, QDict
*parent_options
)
119 /* We're not supposed to call this function for root nodes */
122 static void blk_root_drained_begin(BdrvChild
*child
);
123 static bool blk_root_drained_poll(BdrvChild
*child
);
124 static void blk_root_drained_end(BdrvChild
*child
);
126 static void blk_root_change_media(BdrvChild
*child
, bool load
);
127 static void blk_root_resize(BdrvChild
*child
);
129 static char *blk_root_get_parent_desc(BdrvChild
*child
)
131 BlockBackend
*blk
= child
->opaque
;
135 return g_strdup(blk
->name
);
138 dev_id
= blk_get_attached_dev_id(blk
);
142 /* TODO Callback into the BB owner for something more detailed */
144 return g_strdup("a block device");
148 static const char *blk_root_get_name(BdrvChild
*child
)
150 return blk_name(child
->opaque
);
153 static void blk_vm_state_changed(void *opaque
, int running
, RunState state
)
155 Error
*local_err
= NULL
;
156 BlockBackend
*blk
= opaque
;
158 if (state
== RUN_STATE_INMIGRATE
) {
162 qemu_del_vm_change_state_handler(blk
->vmsh
);
164 blk_set_perm(blk
, blk
->perm
, blk
->shared_perm
, &local_err
);
166 error_report_err(local_err
);
171 * Notifies the user of the BlockBackend that migration has completed. qdev
172 * devices can tighten their permissions in response (specifically revoke
173 * shared write permissions that we needed for storage migration).
175 * If an error is returned, the VM cannot be allowed to be resumed.
177 static void blk_root_activate(BdrvChild
*child
, Error
**errp
)
179 BlockBackend
*blk
= child
->opaque
;
180 Error
*local_err
= NULL
;
182 if (!blk
->disable_perm
) {
186 blk
->disable_perm
= false;
188 blk_set_perm(blk
, blk
->perm
, BLK_PERM_ALL
, &local_err
);
190 error_propagate(errp
, local_err
);
191 blk
->disable_perm
= true;
195 if (runstate_check(RUN_STATE_INMIGRATE
)) {
196 /* Activation can happen when migration process is still active, for
197 * example when nbd_server_add is called during non-shared storage
198 * migration. Defer the shared_perm update to migration completion. */
200 blk
->vmsh
= qemu_add_vm_change_state_handler(blk_vm_state_changed
,
206 blk_set_perm(blk
, blk
->perm
, blk
->shared_perm
, &local_err
);
208 error_propagate(errp
, local_err
);
209 blk
->disable_perm
= true;
214 void blk_set_force_allow_inactivate(BlockBackend
*blk
)
216 blk
->force_allow_inactivate
= true;
219 static bool blk_can_inactivate(BlockBackend
*blk
)
221 /* If it is a guest device, inactivate is ok. */
222 if (blk
->dev
|| blk_name(blk
)[0]) {
226 /* Inactivating means no more writes to the image can be done,
227 * even if those writes would be changes invisible to the
228 * guest. For block job BBs that satisfy this, we can just allow
229 * it. This is the case for mirror job source, which is required
230 * by libvirt non-shared block migration. */
231 if (!(blk
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
))) {
235 return blk
->force_allow_inactivate
;
238 static int blk_root_inactivate(BdrvChild
*child
)
240 BlockBackend
*blk
= child
->opaque
;
242 if (blk
->disable_perm
) {
246 if (!blk_can_inactivate(blk
)) {
250 blk
->disable_perm
= true;
252 bdrv_child_try_set_perm(blk
->root
, 0, BLK_PERM_ALL
, &error_abort
);
258 static void blk_root_attach(BdrvChild
*child
)
260 BlockBackend
*blk
= child
->opaque
;
261 BlockBackendAioNotifier
*notifier
;
263 trace_blk_root_attach(child
, blk
, child
->bs
);
265 QLIST_FOREACH(notifier
, &blk
->aio_notifiers
, list
) {
266 bdrv_add_aio_context_notifier(child
->bs
,
267 notifier
->attached_aio_context
,
268 notifier
->detach_aio_context
,
273 static void blk_root_detach(BdrvChild
*child
)
275 BlockBackend
*blk
= child
->opaque
;
276 BlockBackendAioNotifier
*notifier
;
278 trace_blk_root_detach(child
, blk
, child
->bs
);
280 QLIST_FOREACH(notifier
, &blk
->aio_notifiers
, list
) {
281 bdrv_remove_aio_context_notifier(child
->bs
,
282 notifier
->attached_aio_context
,
283 notifier
->detach_aio_context
,
288 static const BdrvChildRole child_root
= {
289 .inherit_options
= blk_root_inherit_options
,
291 .change_media
= blk_root_change_media
,
292 .resize
= blk_root_resize
,
293 .get_name
= blk_root_get_name
,
294 .get_parent_desc
= blk_root_get_parent_desc
,
296 .drained_begin
= blk_root_drained_begin
,
297 .drained_poll
= blk_root_drained_poll
,
298 .drained_end
= blk_root_drained_end
,
300 .activate
= blk_root_activate
,
301 .inactivate
= blk_root_inactivate
,
303 .attach
= blk_root_attach
,
304 .detach
= blk_root_detach
,
308 * Create a new BlockBackend with a reference count of one.
310 * @perm is a bitmasks of BLK_PERM_* constants which describes the permissions
311 * to request for a block driver node that is attached to this BlockBackend.
312 * @shared_perm is a bitmask which describes which permissions may be granted
313 * to other users of the attached node.
314 * Both sets of permissions can be changed later using blk_set_perm().
316 * Return the new BlockBackend on success, null on failure.
318 BlockBackend
*blk_new(uint64_t perm
, uint64_t shared_perm
)
322 blk
= g_new0(BlockBackend
, 1);
325 blk
->shared_perm
= shared_perm
;
326 blk_set_enable_write_cache(blk
, true);
328 block_acct_init(&blk
->stats
);
330 notifier_list_init(&blk
->remove_bs_notifiers
);
331 notifier_list_init(&blk
->insert_bs_notifiers
);
332 QLIST_INIT(&blk
->aio_notifiers
);
334 QTAILQ_INSERT_TAIL(&block_backends
, blk
, link
);
339 * Creates a new BlockBackend, opens a new BlockDriverState, and connects both.
341 * Just as with bdrv_open(), after having called this function the reference to
342 * @options belongs to the block layer (even on failure).
344 * TODO: Remove @filename and @flags; it should be possible to specify a whole
345 * BDS tree just by specifying the @options QDict (or @reference,
346 * alternatively). At the time of adding this function, this is not possible,
347 * though, so callers of this function have to be able to specify @filename and
350 BlockBackend
*blk_new_open(const char *filename
, const char *reference
,
351 QDict
*options
, int flags
, Error
**errp
)
354 BlockDriverState
*bs
;
357 /* blk_new_open() is mainly used in .bdrv_create implementations and the
358 * tools where sharing isn't a concern because the BDS stays private, so we
359 * just request permission according to the flags.
361 * The exceptions are xen_disk and blockdev_init(); in these cases, the
362 * caller of blk_new_open() doesn't make use of the permissions, but they
363 * shouldn't hurt either. We can still share everything here because the
364 * guest devices will add their own blockers if they can't share. */
365 if ((flags
& BDRV_O_NO_IO
) == 0) {
366 perm
|= BLK_PERM_CONSISTENT_READ
;
367 if (flags
& BDRV_O_RDWR
) {
368 perm
|= BLK_PERM_WRITE
;
371 if (flags
& BDRV_O_RESIZE
) {
372 perm
|= BLK_PERM_RESIZE
;
375 blk
= blk_new(perm
, BLK_PERM_ALL
);
376 bs
= bdrv_open(filename
, reference
, options
, flags
, errp
);
382 blk
->root
= bdrv_root_attach_child(bs
, "root", &child_root
,
383 perm
, BLK_PERM_ALL
, blk
, errp
);
393 static void blk_delete(BlockBackend
*blk
)
395 assert(!blk
->refcnt
);
398 if (blk
->public.throttle_group_member
.throttle_state
) {
399 blk_io_limits_disable(blk
);
405 qemu_del_vm_change_state_handler(blk
->vmsh
);
408 assert(QLIST_EMPTY(&blk
->remove_bs_notifiers
.notifiers
));
409 assert(QLIST_EMPTY(&blk
->insert_bs_notifiers
.notifiers
));
410 assert(QLIST_EMPTY(&blk
->aio_notifiers
));
411 QTAILQ_REMOVE(&block_backends
, blk
, link
);
412 drive_info_del(blk
->legacy_dinfo
);
413 block_acct_cleanup(&blk
->stats
);
417 static void drive_info_del(DriveInfo
*dinfo
)
422 qemu_opts_del(dinfo
->opts
);
426 int blk_get_refcnt(BlockBackend
*blk
)
428 return blk
? blk
->refcnt
: 0;
432 * Increment @blk's reference count.
433 * @blk must not be null.
435 void blk_ref(BlockBackend
*blk
)
437 assert(blk
->refcnt
> 0);
442 * Decrement @blk's reference count.
443 * If this drops it to zero, destroy @blk.
444 * For convenience, do nothing if @blk is null.
446 void blk_unref(BlockBackend
*blk
)
449 assert(blk
->refcnt
> 0);
450 if (blk
->refcnt
> 1) {
454 /* blk_drain() cannot resurrect blk, nobody held a reference */
455 assert(blk
->refcnt
== 1);
463 * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
464 * ones which are hidden (i.e. are not referenced by the monitor).
466 BlockBackend
*blk_all_next(BlockBackend
*blk
)
468 return blk
? QTAILQ_NEXT(blk
, link
)
469 : QTAILQ_FIRST(&block_backends
);
472 void blk_remove_all_bs(void)
474 BlockBackend
*blk
= NULL
;
476 while ((blk
= blk_all_next(blk
)) != NULL
) {
477 AioContext
*ctx
= blk_get_aio_context(blk
);
479 aio_context_acquire(ctx
);
483 aio_context_release(ctx
);
488 * Return the monitor-owned BlockBackend after @blk.
489 * If @blk is null, return the first one.
490 * Else, return @blk's next sibling, which may be null.
492 * To iterate over all BlockBackends, do
493 * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
497 BlockBackend
*blk_next(BlockBackend
*blk
)
499 return blk
? QTAILQ_NEXT(blk
, monitor_link
)
500 : QTAILQ_FIRST(&monitor_block_backends
);
503 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
504 * the monitor or attached to a BlockBackend */
505 BlockDriverState
*bdrv_next(BdrvNextIterator
*it
)
507 BlockDriverState
*bs
, *old_bs
;
509 /* Must be called from the main loop */
510 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
512 /* First, return all root nodes of BlockBackends. In order to avoid
513 * returning a BDS twice when multiple BBs refer to it, we only return it
514 * if the BB is the first one in the parent list of the BDS. */
515 if (it
->phase
== BDRV_NEXT_BACKEND_ROOTS
) {
516 BlockBackend
*old_blk
= it
->blk
;
518 old_bs
= old_blk
? blk_bs(old_blk
) : NULL
;
521 it
->blk
= blk_all_next(it
->blk
);
522 bs
= it
->blk
? blk_bs(it
->blk
) : NULL
;
523 } while (it
->blk
&& (bs
== NULL
|| bdrv_first_blk(bs
) != it
->blk
));
535 it
->phase
= BDRV_NEXT_MONITOR_OWNED
;
540 /* Then return the monitor-owned BDSes without a BB attached. Ignore all
541 * BDSes that are attached to a BlockBackend here; they have been handled
542 * by the above block already */
544 it
->bs
= bdrv_next_monitor_owned(it
->bs
);
546 } while (bs
&& bdrv_has_blk(bs
));
556 static void bdrv_next_reset(BdrvNextIterator
*it
)
558 *it
= (BdrvNextIterator
) {
559 .phase
= BDRV_NEXT_BACKEND_ROOTS
,
563 BlockDriverState
*bdrv_first(BdrvNextIterator
*it
)
566 return bdrv_next(it
);
569 /* Must be called when aborting a bdrv_next() iteration before
570 * bdrv_next() returns NULL */
571 void bdrv_next_cleanup(BdrvNextIterator
*it
)
573 /* Must be called from the main loop */
574 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
576 if (it
->phase
== BDRV_NEXT_BACKEND_ROOTS
) {
578 bdrv_unref(blk_bs(it
->blk
));
589 * Add a BlockBackend into the list of backends referenced by the monitor, with
590 * the given @name acting as the handle for the monitor.
591 * Strictly for use by blockdev.c.
593 * @name must not be null or empty.
595 * Returns true on success and false on failure. In the latter case, an Error
596 * object is returned through @errp.
598 bool monitor_add_blk(BlockBackend
*blk
, const char *name
, Error
**errp
)
601 assert(name
&& name
[0]);
603 if (!id_wellformed(name
)) {
604 error_setg(errp
, "Invalid device name");
607 if (blk_by_name(name
)) {
608 error_setg(errp
, "Device with id '%s' already exists", name
);
611 if (bdrv_find_node(name
)) {
613 "Device name '%s' conflicts with an existing node name",
618 blk
->name
= g_strdup(name
);
619 QTAILQ_INSERT_TAIL(&monitor_block_backends
, blk
, monitor_link
);
624 * Remove a BlockBackend from the list of backends referenced by the monitor.
625 * Strictly for use by blockdev.c.
627 void monitor_remove_blk(BlockBackend
*blk
)
633 QTAILQ_REMOVE(&monitor_block_backends
, blk
, monitor_link
);
639 * Return @blk's name, a non-null string.
640 * Returns an empty string iff @blk is not referenced by the monitor.
642 const char *blk_name(const BlockBackend
*blk
)
644 return blk
->name
?: "";
648 * Return the BlockBackend with name @name if it exists, else null.
649 * @name must not be null.
651 BlockBackend
*blk_by_name(const char *name
)
653 BlockBackend
*blk
= NULL
;
656 while ((blk
= blk_next(blk
)) != NULL
) {
657 if (!strcmp(name
, blk
->name
)) {
665 * Return the BlockDriverState attached to @blk if any, else null.
667 BlockDriverState
*blk_bs(BlockBackend
*blk
)
669 return blk
->root
? blk
->root
->bs
: NULL
;
672 static BlockBackend
*bdrv_first_blk(BlockDriverState
*bs
)
675 QLIST_FOREACH(child
, &bs
->parents
, next_parent
) {
676 if (child
->role
== &child_root
) {
677 return child
->opaque
;
685 * Returns true if @bs has an associated BlockBackend.
687 bool bdrv_has_blk(BlockDriverState
*bs
)
689 return bdrv_first_blk(bs
) != NULL
;
693 * Returns true if @bs has only BlockBackends as parents.
695 bool bdrv_is_root_node(BlockDriverState
*bs
)
699 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
700 if (c
->role
!= &child_root
) {
709 * Return @blk's DriveInfo if any, else null.
711 DriveInfo
*blk_legacy_dinfo(BlockBackend
*blk
)
713 return blk
->legacy_dinfo
;
717 * Set @blk's DriveInfo to @dinfo, and return it.
718 * @blk must not have a DriveInfo set already.
719 * No other BlockBackend may have the same DriveInfo set.
721 DriveInfo
*blk_set_legacy_dinfo(BlockBackend
*blk
, DriveInfo
*dinfo
)
723 assert(!blk
->legacy_dinfo
);
724 return blk
->legacy_dinfo
= dinfo
;
728 * Return the BlockBackend with DriveInfo @dinfo.
731 BlockBackend
*blk_by_legacy_dinfo(DriveInfo
*dinfo
)
733 BlockBackend
*blk
= NULL
;
735 while ((blk
= blk_next(blk
)) != NULL
) {
736 if (blk
->legacy_dinfo
== dinfo
) {
744 * Returns a pointer to the publicly accessible fields of @blk.
746 BlockBackendPublic
*blk_get_public(BlockBackend
*blk
)
752 * Returns a BlockBackend given the associated @public fields.
754 BlockBackend
*blk_by_public(BlockBackendPublic
*public)
756 return container_of(public, BlockBackend
, public);
760 * Disassociates the currently associated BlockDriverState from @blk.
762 void blk_remove_bs(BlockBackend
*blk
)
764 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
765 BlockDriverState
*bs
;
767 notifier_list_notify(&blk
->remove_bs_notifiers
, blk
);
768 if (tgm
->throttle_state
) {
770 bdrv_drained_begin(bs
);
771 throttle_group_detach_aio_context(tgm
);
772 throttle_group_attach_aio_context(tgm
, qemu_get_aio_context());
773 bdrv_drained_end(bs
);
776 blk_update_root_state(blk
);
778 /* bdrv_root_unref_child() will cause blk->root to become stale and may
779 * switch to a completion coroutine later on. Let's drain all I/O here
780 * to avoid that and a potential QEMU crash.
783 bdrv_root_unref_child(blk
->root
);
788 * Associates a new BlockDriverState with @blk.
790 int blk_insert_bs(BlockBackend
*blk
, BlockDriverState
*bs
, Error
**errp
)
792 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
793 blk
->root
= bdrv_root_attach_child(bs
, "root", &child_root
,
794 blk
->perm
, blk
->shared_perm
, blk
, errp
);
795 if (blk
->root
== NULL
) {
800 notifier_list_notify(&blk
->insert_bs_notifiers
, blk
);
801 if (tgm
->throttle_state
) {
802 throttle_group_detach_aio_context(tgm
);
803 throttle_group_attach_aio_context(tgm
, bdrv_get_aio_context(bs
));
810 * Sets the permission bitmasks that the user of the BlockBackend needs.
812 int blk_set_perm(BlockBackend
*blk
, uint64_t perm
, uint64_t shared_perm
,
817 if (blk
->root
&& !blk
->disable_perm
) {
818 ret
= bdrv_child_try_set_perm(blk
->root
, perm
, shared_perm
, errp
);
825 blk
->shared_perm
= shared_perm
;
830 void blk_get_perm(BlockBackend
*blk
, uint64_t *perm
, uint64_t *shared_perm
)
833 *shared_perm
= blk
->shared_perm
;
836 static int blk_do_attach_dev(BlockBackend
*blk
, void *dev
)
842 /* While migration is still incoming, we don't need to apply the
843 * permissions of guest device BlockBackends. We might still have a block
844 * job or NBD server writing to the image for storage migration. */
845 if (runstate_check(RUN_STATE_INMIGRATE
)) {
846 blk
->disable_perm
= true;
851 blk
->legacy_dev
= false;
852 blk_iostatus_reset(blk
);
858 * Attach device model @dev to @blk.
859 * Return 0 on success, -EBUSY when a device model is attached already.
861 int blk_attach_dev(BlockBackend
*blk
, DeviceState
*dev
)
863 return blk_do_attach_dev(blk
, dev
);
867 * Attach device model @dev to @blk.
868 * @blk must not have a device model attached already.
869 * TODO qdevified devices don't use this, remove when devices are qdevified
871 void blk_attach_dev_legacy(BlockBackend
*blk
, void *dev
)
873 if (blk_do_attach_dev(blk
, dev
) < 0) {
876 blk
->legacy_dev
= true;
880 * Detach device model @dev from @blk.
881 * @dev must be currently attached to @blk.
883 void blk_detach_dev(BlockBackend
*blk
, void *dev
)
884 /* TODO change to DeviceState *dev when all users are qdevified */
886 assert(blk
->dev
== dev
);
889 blk
->dev_opaque
= NULL
;
890 blk
->guest_block_size
= 512;
891 blk_set_perm(blk
, 0, BLK_PERM_ALL
, &error_abort
);
896 * Return the device model attached to @blk if any, else null.
898 void *blk_get_attached_dev(BlockBackend
*blk
)
899 /* TODO change to return DeviceState * when all users are qdevified */
904 /* Return the qdev ID, or if no ID is assigned the QOM path, of the block
905 * device attached to the BlockBackend. */
906 char *blk_get_attached_dev_id(BlockBackend
*blk
)
910 assert(!blk
->legacy_dev
);
915 } else if (dev
->id
) {
916 return g_strdup(dev
->id
);
918 return object_get_canonical_path(OBJECT(dev
));
922 * Return the BlockBackend which has the device model @dev attached if it
925 * @dev must not be null.
927 BlockBackend
*blk_by_dev(void *dev
)
929 BlockBackend
*blk
= NULL
;
932 while ((blk
= blk_all_next(blk
)) != NULL
) {
933 if (blk
->dev
== dev
) {
941 * Set @blk's device model callbacks to @ops.
942 * @opaque is the opaque argument to pass to the callbacks.
943 * This is for use by device models.
945 void blk_set_dev_ops(BlockBackend
*blk
, const BlockDevOps
*ops
,
948 /* All drivers that use blk_set_dev_ops() are qdevified and we want to keep
949 * it that way, so we can assume blk->dev, if present, is a DeviceState if
950 * blk->dev_ops is set. Non-device users may use dev_ops without device. */
951 assert(!blk
->legacy_dev
);
954 blk
->dev_opaque
= opaque
;
956 /* Are we currently quiesced? Should we enforce this right now? */
957 if (blk
->quiesce_counter
&& ops
->drained_begin
) {
958 ops
->drained_begin(opaque
);
963 * Notify @blk's attached device model of media change.
965 * If @load is true, notify of media load. This action can fail, meaning that
966 * the medium cannot be loaded. @errp is set then.
968 * If @load is false, notify of media eject. This can never fail.
970 * Also send DEVICE_TRAY_MOVED events as appropriate.
972 void blk_dev_change_media_cb(BlockBackend
*blk
, bool load
, Error
**errp
)
974 if (blk
->dev_ops
&& blk
->dev_ops
->change_media_cb
) {
975 bool tray_was_open
, tray_is_open
;
976 Error
*local_err
= NULL
;
978 assert(!blk
->legacy_dev
);
980 tray_was_open
= blk_dev_is_tray_open(blk
);
981 blk
->dev_ops
->change_media_cb(blk
->dev_opaque
, load
, &local_err
);
983 assert(load
== true);
984 error_propagate(errp
, local_err
);
987 tray_is_open
= blk_dev_is_tray_open(blk
);
989 if (tray_was_open
!= tray_is_open
) {
990 char *id
= blk_get_attached_dev_id(blk
);
991 qapi_event_send_device_tray_moved(blk_name(blk
), id
, tray_is_open
);
997 static void blk_root_change_media(BdrvChild
*child
, bool load
)
999 blk_dev_change_media_cb(child
->opaque
, load
, NULL
);
1003 * Does @blk's attached device model have removable media?
1004 * %true if no device model is attached.
1006 bool blk_dev_has_removable_media(BlockBackend
*blk
)
1008 return !blk
->dev
|| (blk
->dev_ops
&& blk
->dev_ops
->change_media_cb
);
1012 * Does @blk's attached device model have a tray?
1014 bool blk_dev_has_tray(BlockBackend
*blk
)
1016 return blk
->dev_ops
&& blk
->dev_ops
->is_tray_open
;
1020 * Notify @blk's attached device model of a media eject request.
1021 * If @force is true, the medium is about to be yanked out forcefully.
1023 void blk_dev_eject_request(BlockBackend
*blk
, bool force
)
1025 if (blk
->dev_ops
&& blk
->dev_ops
->eject_request_cb
) {
1026 blk
->dev_ops
->eject_request_cb(blk
->dev_opaque
, force
);
1031 * Does @blk's attached device model have a tray, and is it open?
1033 bool blk_dev_is_tray_open(BlockBackend
*blk
)
1035 if (blk_dev_has_tray(blk
)) {
1036 return blk
->dev_ops
->is_tray_open(blk
->dev_opaque
);
1042 * Does @blk's attached device model have the medium locked?
1043 * %false if the device model has no such lock.
1045 bool blk_dev_is_medium_locked(BlockBackend
*blk
)
1047 if (blk
->dev_ops
&& blk
->dev_ops
->is_medium_locked
) {
1048 return blk
->dev_ops
->is_medium_locked(blk
->dev_opaque
);
1054 * Notify @blk's attached device model of a backend size change.
1056 static void blk_root_resize(BdrvChild
*child
)
1058 BlockBackend
*blk
= child
->opaque
;
1060 if (blk
->dev_ops
&& blk
->dev_ops
->resize_cb
) {
1061 blk
->dev_ops
->resize_cb(blk
->dev_opaque
);
1065 void blk_iostatus_enable(BlockBackend
*blk
)
1067 blk
->iostatus_enabled
= true;
1068 blk
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
1071 /* The I/O status is only enabled if the drive explicitly
1072 * enables it _and_ the VM is configured to stop on errors */
1073 bool blk_iostatus_is_enabled(const BlockBackend
*blk
)
1075 return (blk
->iostatus_enabled
&&
1076 (blk
->on_write_error
== BLOCKDEV_ON_ERROR_ENOSPC
||
1077 blk
->on_write_error
== BLOCKDEV_ON_ERROR_STOP
||
1078 blk
->on_read_error
== BLOCKDEV_ON_ERROR_STOP
));
1081 BlockDeviceIoStatus
blk_iostatus(const BlockBackend
*blk
)
1083 return blk
->iostatus
;
1086 void blk_iostatus_disable(BlockBackend
*blk
)
1088 blk
->iostatus_enabled
= false;
1091 void blk_iostatus_reset(BlockBackend
*blk
)
1093 if (blk_iostatus_is_enabled(blk
)) {
1094 BlockDriverState
*bs
= blk_bs(blk
);
1095 blk
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
1096 if (bs
&& bs
->job
) {
1097 block_job_iostatus_reset(bs
->job
);
1102 void blk_iostatus_set_err(BlockBackend
*blk
, int error
)
1104 assert(blk_iostatus_is_enabled(blk
));
1105 if (blk
->iostatus
== BLOCK_DEVICE_IO_STATUS_OK
) {
1106 blk
->iostatus
= error
== ENOSPC
? BLOCK_DEVICE_IO_STATUS_NOSPACE
:
1107 BLOCK_DEVICE_IO_STATUS_FAILED
;
1111 void blk_set_allow_write_beyond_eof(BlockBackend
*blk
, bool allow
)
1113 blk
->allow_write_beyond_eof
= allow
;
1116 static int blk_check_byte_request(BlockBackend
*blk
, int64_t offset
,
1121 if (size
> INT_MAX
) {
1125 if (!blk_is_available(blk
)) {
1133 if (!blk
->allow_write_beyond_eof
) {
1134 len
= blk_getlength(blk
);
1139 if (offset
> len
|| len
- offset
< size
) {
1147 int coroutine_fn
blk_co_preadv(BlockBackend
*blk
, int64_t offset
,
1148 unsigned int bytes
, QEMUIOVector
*qiov
,
1149 BdrvRequestFlags flags
)
1152 BlockDriverState
*bs
= blk_bs(blk
);
1154 trace_blk_co_preadv(blk
, bs
, offset
, bytes
, flags
);
1156 ret
= blk_check_byte_request(blk
, offset
, bytes
);
1161 bdrv_inc_in_flight(bs
);
1163 /* throttling disk I/O */
1164 if (blk
->public.throttle_group_member
.throttle_state
) {
1165 throttle_group_co_io_limits_intercept(&blk
->public.throttle_group_member
,
1169 ret
= bdrv_co_preadv(blk
->root
, offset
, bytes
, qiov
, flags
);
1170 bdrv_dec_in_flight(bs
);
1174 int coroutine_fn
blk_co_pwritev(BlockBackend
*blk
, int64_t offset
,
1175 unsigned int bytes
, QEMUIOVector
*qiov
,
1176 BdrvRequestFlags flags
)
1179 BlockDriverState
*bs
= blk_bs(blk
);
1181 trace_blk_co_pwritev(blk
, bs
, offset
, bytes
, flags
);
1183 ret
= blk_check_byte_request(blk
, offset
, bytes
);
1188 bdrv_inc_in_flight(bs
);
1189 /* throttling disk I/O */
1190 if (blk
->public.throttle_group_member
.throttle_state
) {
1191 throttle_group_co_io_limits_intercept(&blk
->public.throttle_group_member
,
1195 if (!blk
->enable_write_cache
) {
1196 flags
|= BDRV_REQ_FUA
;
1199 ret
= bdrv_co_pwritev(blk
->root
, offset
, bytes
, qiov
, flags
);
1200 bdrv_dec_in_flight(bs
);
1204 typedef struct BlkRwCo
{
1209 BdrvRequestFlags flags
;
1212 static void blk_read_entry(void *opaque
)
1214 BlkRwCo
*rwco
= opaque
;
1215 QEMUIOVector
*qiov
= rwco
->iobuf
;
1217 rwco
->ret
= blk_co_preadv(rwco
->blk
, rwco
->offset
, qiov
->size
,
1221 static void blk_write_entry(void *opaque
)
1223 BlkRwCo
*rwco
= opaque
;
1224 QEMUIOVector
*qiov
= rwco
->iobuf
;
1226 rwco
->ret
= blk_co_pwritev(rwco
->blk
, rwco
->offset
, qiov
->size
,
1230 static int blk_prw(BlockBackend
*blk
, int64_t offset
, uint8_t *buf
,
1231 int64_t bytes
, CoroutineEntry co_entry
,
1232 BdrvRequestFlags flags
)
1238 iov
= (struct iovec
) {
1242 qemu_iovec_init_external(&qiov
, &iov
, 1);
1252 if (qemu_in_coroutine()) {
1253 /* Fast-path if already in coroutine context */
1256 Coroutine
*co
= qemu_coroutine_create(co_entry
, &rwco
);
1257 bdrv_coroutine_enter(blk_bs(blk
), co
);
1258 BDRV_POLL_WHILE(blk_bs(blk
), rwco
.ret
== NOT_DONE
);
1264 int blk_pread_unthrottled(BlockBackend
*blk
, int64_t offset
, uint8_t *buf
,
1269 ret
= blk_check_byte_request(blk
, offset
, count
);
1274 blk_root_drained_begin(blk
->root
);
1275 ret
= blk_pread(blk
, offset
, buf
, count
);
1276 blk_root_drained_end(blk
->root
);
1280 int blk_pwrite_zeroes(BlockBackend
*blk
, int64_t offset
,
1281 int bytes
, BdrvRequestFlags flags
)
1283 return blk_prw(blk
, offset
, NULL
, bytes
, blk_write_entry
,
1284 flags
| BDRV_REQ_ZERO_WRITE
);
1287 int blk_make_zero(BlockBackend
*blk
, BdrvRequestFlags flags
)
1289 return bdrv_make_zero(blk
->root
, flags
);
1292 static void blk_inc_in_flight(BlockBackend
*blk
)
1294 atomic_inc(&blk
->in_flight
);
1297 static void blk_dec_in_flight(BlockBackend
*blk
)
1299 atomic_dec(&blk
->in_flight
);
1303 static void error_callback_bh(void *opaque
)
1305 struct BlockBackendAIOCB
*acb
= opaque
;
1307 blk_dec_in_flight(acb
->blk
);
1308 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
1309 qemu_aio_unref(acb
);
1312 BlockAIOCB
*blk_abort_aio_request(BlockBackend
*blk
,
1313 BlockCompletionFunc
*cb
,
1314 void *opaque
, int ret
)
1316 struct BlockBackendAIOCB
*acb
;
1318 blk_inc_in_flight(blk
);
1319 acb
= blk_aio_get(&block_backend_aiocb_info
, blk
, cb
, opaque
);
1323 aio_bh_schedule_oneshot(blk_get_aio_context(blk
), error_callback_bh
, acb
);
1324 return &acb
->common
;
1327 typedef struct BlkAioEmAIOCB
{
1334 static const AIOCBInfo blk_aio_em_aiocb_info
= {
1335 .aiocb_size
= sizeof(BlkAioEmAIOCB
),
1338 static void blk_aio_complete(BlkAioEmAIOCB
*acb
)
1340 if (acb
->has_returned
) {
1341 acb
->common
.cb(acb
->common
.opaque
, acb
->rwco
.ret
);
1342 blk_dec_in_flight(acb
->rwco
.blk
);
1343 qemu_aio_unref(acb
);
1347 static void blk_aio_complete_bh(void *opaque
)
1349 BlkAioEmAIOCB
*acb
= opaque
;
1350 assert(acb
->has_returned
);
1351 blk_aio_complete(acb
);
1354 static BlockAIOCB
*blk_aio_prwv(BlockBackend
*blk
, int64_t offset
, int bytes
,
1355 void *iobuf
, CoroutineEntry co_entry
,
1356 BdrvRequestFlags flags
,
1357 BlockCompletionFunc
*cb
, void *opaque
)
1362 blk_inc_in_flight(blk
);
1363 acb
= blk_aio_get(&blk_aio_em_aiocb_info
, blk
, cb
, opaque
);
1364 acb
->rwco
= (BlkRwCo
) {
1372 acb
->has_returned
= false;
1374 co
= qemu_coroutine_create(co_entry
, acb
);
1375 bdrv_coroutine_enter(blk_bs(blk
), co
);
1377 acb
->has_returned
= true;
1378 if (acb
->rwco
.ret
!= NOT_DONE
) {
1379 aio_bh_schedule_oneshot(blk_get_aio_context(blk
),
1380 blk_aio_complete_bh
, acb
);
1383 return &acb
->common
;
1386 static void blk_aio_read_entry(void *opaque
)
1388 BlkAioEmAIOCB
*acb
= opaque
;
1389 BlkRwCo
*rwco
= &acb
->rwco
;
1390 QEMUIOVector
*qiov
= rwco
->iobuf
;
1392 assert(qiov
->size
== acb
->bytes
);
1393 rwco
->ret
= blk_co_preadv(rwco
->blk
, rwco
->offset
, acb
->bytes
,
1395 blk_aio_complete(acb
);
1398 static void blk_aio_write_entry(void *opaque
)
1400 BlkAioEmAIOCB
*acb
= opaque
;
1401 BlkRwCo
*rwco
= &acb
->rwco
;
1402 QEMUIOVector
*qiov
= rwco
->iobuf
;
1404 assert(!qiov
|| qiov
->size
== acb
->bytes
);
1405 rwco
->ret
= blk_co_pwritev(rwco
->blk
, rwco
->offset
, acb
->bytes
,
1407 blk_aio_complete(acb
);
1410 BlockAIOCB
*blk_aio_pwrite_zeroes(BlockBackend
*blk
, int64_t offset
,
1411 int count
, BdrvRequestFlags flags
,
1412 BlockCompletionFunc
*cb
, void *opaque
)
1414 return blk_aio_prwv(blk
, offset
, count
, NULL
, blk_aio_write_entry
,
1415 flags
| BDRV_REQ_ZERO_WRITE
, cb
, opaque
);
1418 int blk_pread(BlockBackend
*blk
, int64_t offset
, void *buf
, int count
)
1420 int ret
= blk_prw(blk
, offset
, buf
, count
, blk_read_entry
, 0);
1427 int blk_pwrite(BlockBackend
*blk
, int64_t offset
, const void *buf
, int count
,
1428 BdrvRequestFlags flags
)
1430 int ret
= blk_prw(blk
, offset
, (void *) buf
, count
, blk_write_entry
,
1438 int64_t blk_getlength(BlockBackend
*blk
)
1440 if (!blk_is_available(blk
)) {
1444 return bdrv_getlength(blk_bs(blk
));
1447 void blk_get_geometry(BlockBackend
*blk
, uint64_t *nb_sectors_ptr
)
1450 *nb_sectors_ptr
= 0;
1452 bdrv_get_geometry(blk_bs(blk
), nb_sectors_ptr
);
1456 int64_t blk_nb_sectors(BlockBackend
*blk
)
1458 if (!blk_is_available(blk
)) {
1462 return bdrv_nb_sectors(blk_bs(blk
));
1465 BlockAIOCB
*blk_aio_preadv(BlockBackend
*blk
, int64_t offset
,
1466 QEMUIOVector
*qiov
, BdrvRequestFlags flags
,
1467 BlockCompletionFunc
*cb
, void *opaque
)
1469 return blk_aio_prwv(blk
, offset
, qiov
->size
, qiov
,
1470 blk_aio_read_entry
, flags
, cb
, opaque
);
1473 BlockAIOCB
*blk_aio_pwritev(BlockBackend
*blk
, int64_t offset
,
1474 QEMUIOVector
*qiov
, BdrvRequestFlags flags
,
1475 BlockCompletionFunc
*cb
, void *opaque
)
1477 return blk_aio_prwv(blk
, offset
, qiov
->size
, qiov
,
1478 blk_aio_write_entry
, flags
, cb
, opaque
);
1481 static void blk_aio_flush_entry(void *opaque
)
1483 BlkAioEmAIOCB
*acb
= opaque
;
1484 BlkRwCo
*rwco
= &acb
->rwco
;
1486 rwco
->ret
= blk_co_flush(rwco
->blk
);
1487 blk_aio_complete(acb
);
1490 BlockAIOCB
*blk_aio_flush(BlockBackend
*blk
,
1491 BlockCompletionFunc
*cb
, void *opaque
)
1493 return blk_aio_prwv(blk
, 0, 0, NULL
, blk_aio_flush_entry
, 0, cb
, opaque
);
1496 static void blk_aio_pdiscard_entry(void *opaque
)
1498 BlkAioEmAIOCB
*acb
= opaque
;
1499 BlkRwCo
*rwco
= &acb
->rwco
;
1501 rwco
->ret
= blk_co_pdiscard(rwco
->blk
, rwco
->offset
, acb
->bytes
);
1502 blk_aio_complete(acb
);
1505 BlockAIOCB
*blk_aio_pdiscard(BlockBackend
*blk
,
1506 int64_t offset
, int bytes
,
1507 BlockCompletionFunc
*cb
, void *opaque
)
1509 return blk_aio_prwv(blk
, offset
, bytes
, NULL
, blk_aio_pdiscard_entry
, 0,
1513 void blk_aio_cancel(BlockAIOCB
*acb
)
1515 bdrv_aio_cancel(acb
);
1518 void blk_aio_cancel_async(BlockAIOCB
*acb
)
1520 bdrv_aio_cancel_async(acb
);
1523 int blk_co_ioctl(BlockBackend
*blk
, unsigned long int req
, void *buf
)
1525 if (!blk_is_available(blk
)) {
1529 return bdrv_co_ioctl(blk_bs(blk
), req
, buf
);
1532 static void blk_ioctl_entry(void *opaque
)
1534 BlkRwCo
*rwco
= opaque
;
1535 QEMUIOVector
*qiov
= rwco
->iobuf
;
1537 rwco
->ret
= blk_co_ioctl(rwco
->blk
, rwco
->offset
,
1538 qiov
->iov
[0].iov_base
);
1541 int blk_ioctl(BlockBackend
*blk
, unsigned long int req
, void *buf
)
1543 return blk_prw(blk
, req
, buf
, 0, blk_ioctl_entry
, 0);
1546 static void blk_aio_ioctl_entry(void *opaque
)
1548 BlkAioEmAIOCB
*acb
= opaque
;
1549 BlkRwCo
*rwco
= &acb
->rwco
;
1551 rwco
->ret
= blk_co_ioctl(rwco
->blk
, rwco
->offset
, rwco
->iobuf
);
1553 blk_aio_complete(acb
);
1556 BlockAIOCB
*blk_aio_ioctl(BlockBackend
*blk
, unsigned long int req
, void *buf
,
1557 BlockCompletionFunc
*cb
, void *opaque
)
1559 return blk_aio_prwv(blk
, req
, 0, buf
, blk_aio_ioctl_entry
, 0, cb
, opaque
);
1562 int blk_co_pdiscard(BlockBackend
*blk
, int64_t offset
, int bytes
)
1564 int ret
= blk_check_byte_request(blk
, offset
, bytes
);
1569 return bdrv_co_pdiscard(blk
->root
, offset
, bytes
);
1572 int blk_co_flush(BlockBackend
*blk
)
1574 if (!blk_is_available(blk
)) {
1578 return bdrv_co_flush(blk_bs(blk
));
1581 static void blk_flush_entry(void *opaque
)
1583 BlkRwCo
*rwco
= opaque
;
1584 rwco
->ret
= blk_co_flush(rwco
->blk
);
1587 int blk_flush(BlockBackend
*blk
)
1589 return blk_prw(blk
, 0, NULL
, 0, blk_flush_entry
, 0);
1592 void blk_drain(BlockBackend
*blk
)
1594 BlockDriverState
*bs
= blk_bs(blk
);
1597 bdrv_drained_begin(bs
);
1600 /* We may have -ENOMEDIUM completions in flight */
1601 AIO_WAIT_WHILE(blk_get_aio_context(blk
),
1602 atomic_mb_read(&blk
->in_flight
) > 0);
1605 bdrv_drained_end(bs
);
1609 void blk_drain_all(void)
1611 BlockBackend
*blk
= NULL
;
1613 bdrv_drain_all_begin();
1615 while ((blk
= blk_all_next(blk
)) != NULL
) {
1616 AioContext
*ctx
= blk_get_aio_context(blk
);
1618 aio_context_acquire(ctx
);
1620 /* We may have -ENOMEDIUM completions in flight */
1621 AIO_WAIT_WHILE(ctx
, atomic_mb_read(&blk
->in_flight
) > 0);
1623 aio_context_release(ctx
);
1626 bdrv_drain_all_end();
1629 void blk_set_on_error(BlockBackend
*blk
, BlockdevOnError on_read_error
,
1630 BlockdevOnError on_write_error
)
1632 blk
->on_read_error
= on_read_error
;
1633 blk
->on_write_error
= on_write_error
;
1636 BlockdevOnError
blk_get_on_error(BlockBackend
*blk
, bool is_read
)
1638 return is_read
? blk
->on_read_error
: blk
->on_write_error
;
1641 BlockErrorAction
blk_get_error_action(BlockBackend
*blk
, bool is_read
,
1644 BlockdevOnError on_err
= blk_get_on_error(blk
, is_read
);
1647 case BLOCKDEV_ON_ERROR_ENOSPC
:
1648 return (error
== ENOSPC
) ?
1649 BLOCK_ERROR_ACTION_STOP
: BLOCK_ERROR_ACTION_REPORT
;
1650 case BLOCKDEV_ON_ERROR_STOP
:
1651 return BLOCK_ERROR_ACTION_STOP
;
1652 case BLOCKDEV_ON_ERROR_REPORT
:
1653 return BLOCK_ERROR_ACTION_REPORT
;
1654 case BLOCKDEV_ON_ERROR_IGNORE
:
1655 return BLOCK_ERROR_ACTION_IGNORE
;
1656 case BLOCKDEV_ON_ERROR_AUTO
:
1662 static void send_qmp_error_event(BlockBackend
*blk
,
1663 BlockErrorAction action
,
1664 bool is_read
, int error
)
1666 IoOperationType optype
;
1667 BlockDriverState
*bs
= blk_bs(blk
);
1669 optype
= is_read
? IO_OPERATION_TYPE_READ
: IO_OPERATION_TYPE_WRITE
;
1670 qapi_event_send_block_io_error(blk_name(blk
), !!bs
,
1671 bs
? bdrv_get_node_name(bs
) : NULL
, optype
,
1672 action
, blk_iostatus_is_enabled(blk
),
1673 error
== ENOSPC
, strerror(error
));
1676 /* This is done by device models because, while the block layer knows
1677 * about the error, it does not know whether an operation comes from
1678 * the device or the block layer (from a job, for example).
1680 void blk_error_action(BlockBackend
*blk
, BlockErrorAction action
,
1681 bool is_read
, int error
)
1685 if (action
== BLOCK_ERROR_ACTION_STOP
) {
1686 /* First set the iostatus, so that "info block" returns an iostatus
1687 * that matches the events raised so far (an additional error iostatus
1688 * is fine, but not a lost one).
1690 blk_iostatus_set_err(blk
, error
);
1692 /* Then raise the request to stop the VM and the event.
1693 * qemu_system_vmstop_request_prepare has two effects. First,
1694 * it ensures that the STOP event always comes after the
1695 * BLOCK_IO_ERROR event. Second, it ensures that even if management
1696 * can observe the STOP event and do a "cont" before the STOP
1697 * event is issued, the VM will not stop. In this case, vm_start()
1698 * also ensures that the STOP/RESUME pair of events is emitted.
1700 qemu_system_vmstop_request_prepare();
1701 send_qmp_error_event(blk
, action
, is_read
, error
);
1702 qemu_system_vmstop_request(RUN_STATE_IO_ERROR
);
1704 send_qmp_error_event(blk
, action
, is_read
, error
);
1708 int blk_is_read_only(BlockBackend
*blk
)
1710 BlockDriverState
*bs
= blk_bs(blk
);
1713 return bdrv_is_read_only(bs
);
1715 return blk
->root_state
.read_only
;
1719 int blk_is_sg(BlockBackend
*blk
)
1721 BlockDriverState
*bs
= blk_bs(blk
);
1727 return bdrv_is_sg(bs
);
1730 int blk_enable_write_cache(BlockBackend
*blk
)
1732 return blk
->enable_write_cache
;
1735 void blk_set_enable_write_cache(BlockBackend
*blk
, bool wce
)
1737 blk
->enable_write_cache
= wce
;
1740 void blk_invalidate_cache(BlockBackend
*blk
, Error
**errp
)
1742 BlockDriverState
*bs
= blk_bs(blk
);
1745 error_setg(errp
, "Device '%s' has no medium", blk
->name
);
1749 bdrv_invalidate_cache(bs
, errp
);
1752 bool blk_is_inserted(BlockBackend
*blk
)
1754 BlockDriverState
*bs
= blk_bs(blk
);
1756 return bs
&& bdrv_is_inserted(bs
);
1759 bool blk_is_available(BlockBackend
*blk
)
1761 return blk_is_inserted(blk
) && !blk_dev_is_tray_open(blk
);
1764 void blk_lock_medium(BlockBackend
*blk
, bool locked
)
1766 BlockDriverState
*bs
= blk_bs(blk
);
1769 bdrv_lock_medium(bs
, locked
);
1773 void blk_eject(BlockBackend
*blk
, bool eject_flag
)
1775 BlockDriverState
*bs
= blk_bs(blk
);
1778 /* blk_eject is only called by qdevified devices */
1779 assert(!blk
->legacy_dev
);
1782 bdrv_eject(bs
, eject_flag
);
1785 /* Whether or not we ejected on the backend,
1786 * the frontend experienced a tray event. */
1787 id
= blk_get_attached_dev_id(blk
);
1788 qapi_event_send_device_tray_moved(blk_name(blk
), id
,
1793 int blk_get_flags(BlockBackend
*blk
)
1795 BlockDriverState
*bs
= blk_bs(blk
);
1798 return bdrv_get_flags(bs
);
1800 return blk
->root_state
.open_flags
;
1804 /* Returns the maximum transfer length, in bytes; guaranteed nonzero */
1805 uint32_t blk_get_max_transfer(BlockBackend
*blk
)
1807 BlockDriverState
*bs
= blk_bs(blk
);
1811 max
= bs
->bl
.max_transfer
;
1813 return MIN_NON_ZERO(max
, INT_MAX
);
1816 int blk_get_max_iov(BlockBackend
*blk
)
1818 return blk
->root
->bs
->bl
.max_iov
;
1821 void blk_set_guest_block_size(BlockBackend
*blk
, int align
)
1823 blk
->guest_block_size
= align
;
1826 void *blk_try_blockalign(BlockBackend
*blk
, size_t size
)
1828 return qemu_try_blockalign(blk
? blk_bs(blk
) : NULL
, size
);
1831 void *blk_blockalign(BlockBackend
*blk
, size_t size
)
1833 return qemu_blockalign(blk
? blk_bs(blk
) : NULL
, size
);
1836 bool blk_op_is_blocked(BlockBackend
*blk
, BlockOpType op
, Error
**errp
)
1838 BlockDriverState
*bs
= blk_bs(blk
);
1844 return bdrv_op_is_blocked(bs
, op
, errp
);
1847 void blk_op_unblock(BlockBackend
*blk
, BlockOpType op
, Error
*reason
)
1849 BlockDriverState
*bs
= blk_bs(blk
);
1852 bdrv_op_unblock(bs
, op
, reason
);
1856 void blk_op_block_all(BlockBackend
*blk
, Error
*reason
)
1858 BlockDriverState
*bs
= blk_bs(blk
);
1861 bdrv_op_block_all(bs
, reason
);
1865 void blk_op_unblock_all(BlockBackend
*blk
, Error
*reason
)
1867 BlockDriverState
*bs
= blk_bs(blk
);
1870 bdrv_op_unblock_all(bs
, reason
);
1874 AioContext
*blk_get_aio_context(BlockBackend
*blk
)
1876 return bdrv_get_aio_context(blk_bs(blk
));
1879 static AioContext
*blk_aiocb_get_aio_context(BlockAIOCB
*acb
)
1881 BlockBackendAIOCB
*blk_acb
= DO_UPCAST(BlockBackendAIOCB
, common
, acb
);
1882 return blk_get_aio_context(blk_acb
->blk
);
1885 void blk_set_aio_context(BlockBackend
*blk
, AioContext
*new_context
)
1887 BlockDriverState
*bs
= blk_bs(blk
);
1888 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
1891 if (tgm
->throttle_state
) {
1892 bdrv_drained_begin(bs
);
1893 throttle_group_detach_aio_context(tgm
);
1894 throttle_group_attach_aio_context(tgm
, new_context
);
1895 bdrv_drained_end(bs
);
1897 bdrv_set_aio_context(bs
, new_context
);
1901 void blk_add_aio_context_notifier(BlockBackend
*blk
,
1902 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
1903 void (*detach_aio_context
)(void *opaque
), void *opaque
)
1905 BlockBackendAioNotifier
*notifier
;
1906 BlockDriverState
*bs
= blk_bs(blk
);
1908 notifier
= g_new(BlockBackendAioNotifier
, 1);
1909 notifier
->attached_aio_context
= attached_aio_context
;
1910 notifier
->detach_aio_context
= detach_aio_context
;
1911 notifier
->opaque
= opaque
;
1912 QLIST_INSERT_HEAD(&blk
->aio_notifiers
, notifier
, list
);
1915 bdrv_add_aio_context_notifier(bs
, attached_aio_context
,
1916 detach_aio_context
, opaque
);
1920 void blk_remove_aio_context_notifier(BlockBackend
*blk
,
1921 void (*attached_aio_context
)(AioContext
*,
1923 void (*detach_aio_context
)(void *),
1926 BlockBackendAioNotifier
*notifier
;
1927 BlockDriverState
*bs
= blk_bs(blk
);
1930 bdrv_remove_aio_context_notifier(bs
, attached_aio_context
,
1931 detach_aio_context
, opaque
);
1934 QLIST_FOREACH(notifier
, &blk
->aio_notifiers
, list
) {
1935 if (notifier
->attached_aio_context
== attached_aio_context
&&
1936 notifier
->detach_aio_context
== detach_aio_context
&&
1937 notifier
->opaque
== opaque
) {
1938 QLIST_REMOVE(notifier
, list
);
1947 void blk_add_remove_bs_notifier(BlockBackend
*blk
, Notifier
*notify
)
1949 notifier_list_add(&blk
->remove_bs_notifiers
, notify
);
1952 void blk_add_insert_bs_notifier(BlockBackend
*blk
, Notifier
*notify
)
1954 notifier_list_add(&blk
->insert_bs_notifiers
, notify
);
1957 void blk_io_plug(BlockBackend
*blk
)
1959 BlockDriverState
*bs
= blk_bs(blk
);
1966 void blk_io_unplug(BlockBackend
*blk
)
1968 BlockDriverState
*bs
= blk_bs(blk
);
1975 BlockAcctStats
*blk_get_stats(BlockBackend
*blk
)
1980 void *blk_aio_get(const AIOCBInfo
*aiocb_info
, BlockBackend
*blk
,
1981 BlockCompletionFunc
*cb
, void *opaque
)
1983 return qemu_aio_get(aiocb_info
, blk_bs(blk
), cb
, opaque
);
1986 int coroutine_fn
blk_co_pwrite_zeroes(BlockBackend
*blk
, int64_t offset
,
1987 int bytes
, BdrvRequestFlags flags
)
1989 return blk_co_pwritev(blk
, offset
, bytes
, NULL
,
1990 flags
| BDRV_REQ_ZERO_WRITE
);
1993 int blk_pwrite_compressed(BlockBackend
*blk
, int64_t offset
, const void *buf
,
1996 return blk_prw(blk
, offset
, (void *) buf
, count
, blk_write_entry
,
1997 BDRV_REQ_WRITE_COMPRESSED
);
2000 int blk_truncate(BlockBackend
*blk
, int64_t offset
, PreallocMode prealloc
,
2003 if (!blk_is_available(blk
)) {
2004 error_setg(errp
, "No medium inserted");
2008 return bdrv_truncate(blk
->root
, offset
, prealloc
, errp
);
2011 static void blk_pdiscard_entry(void *opaque
)
2013 BlkRwCo
*rwco
= opaque
;
2014 QEMUIOVector
*qiov
= rwco
->iobuf
;
2016 rwco
->ret
= blk_co_pdiscard(rwco
->blk
, rwco
->offset
, qiov
->size
);
2019 int blk_pdiscard(BlockBackend
*blk
, int64_t offset
, int bytes
)
2021 return blk_prw(blk
, offset
, NULL
, bytes
, blk_pdiscard_entry
, 0);
2024 int blk_save_vmstate(BlockBackend
*blk
, const uint8_t *buf
,
2025 int64_t pos
, int size
)
2029 if (!blk_is_available(blk
)) {
2033 ret
= bdrv_save_vmstate(blk_bs(blk
), buf
, pos
, size
);
2038 if (ret
== size
&& !blk
->enable_write_cache
) {
2039 ret
= bdrv_flush(blk_bs(blk
));
2042 return ret
< 0 ? ret
: size
;
2045 int blk_load_vmstate(BlockBackend
*blk
, uint8_t *buf
, int64_t pos
, int size
)
2047 if (!blk_is_available(blk
)) {
2051 return bdrv_load_vmstate(blk_bs(blk
), buf
, pos
, size
);
2054 int blk_probe_blocksizes(BlockBackend
*blk
, BlockSizes
*bsz
)
2056 if (!blk_is_available(blk
)) {
2060 return bdrv_probe_blocksizes(blk_bs(blk
), bsz
);
2063 int blk_probe_geometry(BlockBackend
*blk
, HDGeometry
*geo
)
2065 if (!blk_is_available(blk
)) {
2069 return bdrv_probe_geometry(blk_bs(blk
), geo
);
2073 * Updates the BlockBackendRootState object with data from the currently
2074 * attached BlockDriverState.
2076 void blk_update_root_state(BlockBackend
*blk
)
2080 blk
->root_state
.open_flags
= blk
->root
->bs
->open_flags
;
2081 blk
->root_state
.read_only
= blk
->root
->bs
->read_only
;
2082 blk
->root_state
.detect_zeroes
= blk
->root
->bs
->detect_zeroes
;
2086 * Returns the detect-zeroes setting to be used for bdrv_open() of a
2087 * BlockDriverState which is supposed to inherit the root state.
2089 bool blk_get_detect_zeroes_from_root_state(BlockBackend
*blk
)
2091 return blk
->root_state
.detect_zeroes
;
2095 * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
2096 * supposed to inherit the root state.
2098 int blk_get_open_flags_from_root_state(BlockBackend
*blk
)
2102 bs_flags
= blk
->root_state
.read_only
? 0 : BDRV_O_RDWR
;
2103 bs_flags
|= blk
->root_state
.open_flags
& ~BDRV_O_RDWR
;
2108 BlockBackendRootState
*blk_get_root_state(BlockBackend
*blk
)
2110 return &blk
->root_state
;
2113 int blk_commit_all(void)
2115 BlockBackend
*blk
= NULL
;
2117 while ((blk
= blk_all_next(blk
)) != NULL
) {
2118 AioContext
*aio_context
= blk_get_aio_context(blk
);
2120 aio_context_acquire(aio_context
);
2121 if (blk_is_inserted(blk
) && blk
->root
->bs
->backing
) {
2122 int ret
= bdrv_commit(blk
->root
->bs
);
2124 aio_context_release(aio_context
);
2128 aio_context_release(aio_context
);
2134 /* throttling disk I/O limits */
2135 void blk_set_io_limits(BlockBackend
*blk
, ThrottleConfig
*cfg
)
2137 throttle_group_config(&blk
->public.throttle_group_member
, cfg
);
2140 void blk_io_limits_disable(BlockBackend
*blk
)
2142 BlockDriverState
*bs
= blk_bs(blk
);
2143 ThrottleGroupMember
*tgm
= &blk
->public.throttle_group_member
;
2144 assert(tgm
->throttle_state
);
2146 bdrv_drained_begin(bs
);
2148 throttle_group_unregister_tgm(tgm
);
2150 bdrv_drained_end(bs
);
2154 /* should be called before blk_set_io_limits if a limit is set */
2155 void blk_io_limits_enable(BlockBackend
*blk
, const char *group
)
2157 assert(!blk
->public.throttle_group_member
.throttle_state
);
2158 throttle_group_register_tgm(&blk
->public.throttle_group_member
,
2159 group
, blk_get_aio_context(blk
));
2162 void blk_io_limits_update_group(BlockBackend
*blk
, const char *group
)
2164 /* this BB is not part of any group */
2165 if (!blk
->public.throttle_group_member
.throttle_state
) {
2169 /* this BB is a part of the same group than the one we want */
2170 if (!g_strcmp0(throttle_group_get_name(&blk
->public.throttle_group_member
),
2175 /* need to change the group this bs belong to */
2176 blk_io_limits_disable(blk
);
2177 blk_io_limits_enable(blk
, group
);
2180 static void blk_root_drained_begin(BdrvChild
*child
)
2182 BlockBackend
*blk
= child
->opaque
;
2184 if (++blk
->quiesce_counter
== 1) {
2185 if (blk
->dev_ops
&& blk
->dev_ops
->drained_begin
) {
2186 blk
->dev_ops
->drained_begin(blk
->dev_opaque
);
2190 /* Note that blk->root may not be accessible here yet if we are just
2191 * attaching to a BlockDriverState that is drained. Use child instead. */
2193 if (atomic_fetch_inc(&blk
->public.throttle_group_member
.io_limits_disabled
) == 0) {
2194 throttle_group_restart_tgm(&blk
->public.throttle_group_member
);
2198 static bool blk_root_drained_poll(BdrvChild
*child
)
2200 BlockBackend
*blk
= child
->opaque
;
2201 assert(blk
->quiesce_counter
);
2202 return !!blk
->in_flight
;
2205 static void blk_root_drained_end(BdrvChild
*child
)
2207 BlockBackend
*blk
= child
->opaque
;
2208 assert(blk
->quiesce_counter
);
2210 assert(blk
->public.throttle_group_member
.io_limits_disabled
);
2211 atomic_dec(&blk
->public.throttle_group_member
.io_limits_disabled
);
2213 if (--blk
->quiesce_counter
== 0) {
2214 if (blk
->dev_ops
&& blk
->dev_ops
->drained_end
) {
2215 blk
->dev_ops
->drained_end(blk
->dev_opaque
);
2220 void blk_register_buf(BlockBackend
*blk
, void *host
, size_t size
)
2222 bdrv_register_buf(blk_bs(blk
), host
, size
);
2225 void blk_unregister_buf(BlockBackend
*blk
, void *host
)
2227 bdrv_unregister_buf(blk_bs(blk
), host
);
2230 int coroutine_fn
blk_co_copy_range(BlockBackend
*blk_in
, int64_t off_in
,
2231 BlockBackend
*blk_out
, int64_t off_out
,
2232 int bytes
, BdrvRequestFlags read_flags
,
2233 BdrvRequestFlags write_flags
)
2236 r
= blk_check_byte_request(blk_in
, off_in
, bytes
);
2240 r
= blk_check_byte_request(blk_out
, off_out
, bytes
);
2244 return bdrv_co_copy_range(blk_in
->root
, off_in
,
2245 blk_out
->root
, off_out
,
2246 bytes
, read_flags
, write_flags
);