4
* Copyright (C) 2014-2016 Red Hat, Inc.
7
* Markus Armbruster <armbru@redhat.com>,
9
* This work is licensed under the terms of the GNU LGPL, version 2.1
10
* or later. See the COPYING.LIB file in the top-level directory.
13
#include "qemu/osdep.h"
14
#include "sysemu/block-backend.h"
15
#include "block/block_int.h"
16
#include "block/blockjob.h"
17
#include "block/coroutines.h"
18
#include "block/throttle-groups.h"
19
#include "hw/qdev-core.h"
20
#include "sysemu/blockdev.h"
21
#include "sysemu/runstate.h"
22
#include "sysemu/replay.h"
23
#include "qapi/error.h"
24
#include "qapi/qapi-events-block.h"
26
#include "qemu/main-loop.h"
27
#include "qemu/option.h"
29
#include "migration/misc.h"
31
/* Number of coroutines to reserve per attached device model */
32
#define COROUTINE_POOL_RESERVATION 64
34
#define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
36
typedef struct BlockBackendAioNotifier {37
void (*attached_aio_context)(AioContext *new_context, void *opaque);
38
void (*detach_aio_context)(void *opaque);
40
QLIST_ENTRY(BlockBackendAioNotifier) list;
41
} BlockBackendAioNotifier;
47
AioContext *ctx; /* access with atomic operations only */
48
DriveInfo *legacy_dinfo; /* null unless created by drive_new() */
49
QTAILQ_ENTRY(BlockBackend) link; /* for block_backends */
50
QTAILQ_ENTRY(BlockBackend) monitor_link; /* for monitor_block_backends */
51
BlockBackendPublic public;
53
DeviceState *dev; /* attached device model, if any */
54
const BlockDevOps *dev_ops;
57
/* If the BDS tree is removed, some of its options are stored here (which
58
* can be used to restore those options in the new BDS on insert) */
59
BlockBackendRootState root_state;
61
bool enable_write_cache;
63
/* I/O stats (display with "info blockstats"). */
66
BlockdevOnError on_read_error, on_write_error;
67
bool iostatus_enabled;
68
BlockDeviceIoStatus iostatus;
74
bool allow_aio_context_change;
75
bool allow_write_beyond_eof;
77
/* Protected by BQL */
78
NotifierList remove_bs_notifiers, insert_bs_notifiers;
79
QLIST_HEAD(, BlockBackendAioNotifier) aio_notifiers;
81
int quiesce_counter; /* atomic: written under BQL, read by other threads */
82
QemuMutex queued_requests_lock; /* protects queued_requests */
83
CoQueue queued_requests;
84
bool disable_request_queuing; /* atomic */
86
VMChangeStateEntry *vmsh;
87
bool force_allow_inactivate;
89
/* Number of in-flight aio requests. BlockDriverState also counts
90
* in-flight requests but aio requests can exist even when blk->root is
91
* NULL, so we cannot rely on its counter for that case.
92
* Accessed with atomic ops.
94
unsigned int in_flight;
97
typedef struct BlockBackendAIOCB {103
static const AIOCBInfo block_backend_aiocb_info = {104
.aiocb_size = sizeof(BlockBackendAIOCB),
107
static void drive_info_del(DriveInfo *dinfo);
108
static BlockBackend *bdrv_first_blk(BlockDriverState *bs);
110
/* All BlockBackends. Protected by BQL. */
111
static QTAILQ_HEAD(, BlockBackend) block_backends =
112
QTAILQ_HEAD_INITIALIZER(block_backends);
115
* All BlockBackends referenced by the monitor and which are iterated through by
116
* blk_next(). Protected by BQL.
118
static QTAILQ_HEAD(, BlockBackend) monitor_block_backends =
119
QTAILQ_HEAD_INITIALIZER(monitor_block_backends);
121
static int coroutine_mixed_fn GRAPH_RDLOCK
122
blk_set_perm_locked(BlockBackend *blk, uint64_t perm, uint64_t shared_perm,
125
static void blk_root_inherit_options(BdrvChildRole role, bool parent_is_format,
126
int *child_flags, QDict *child_options,
127
int parent_flags, QDict *parent_options)
129
/* We're not supposed to call this function for root nodes */
132
static void blk_root_drained_begin(BdrvChild *child);
133
static bool blk_root_drained_poll(BdrvChild *child);
134
static void blk_root_drained_end(BdrvChild *child);
136
static void blk_root_change_media(BdrvChild *child, bool load);
137
static void blk_root_resize(BdrvChild *child);
139
static bool blk_root_change_aio_ctx(BdrvChild *child, AioContext *ctx,
140
GHashTable *visited, Transaction *tran,
143
static char *blk_root_get_parent_desc(BdrvChild *child)
145
BlockBackend *blk = child->opaque;
146
g_autofree char *dev_id = NULL;
149
return g_strdup_printf("block device '%s'", blk->name);152
dev_id = blk_get_attached_dev_id(blk);
154
return g_strdup_printf("block device '%s'", dev_id);156
/* TODO Callback into the BB owner for something more detailed */
157
return g_strdup("an unnamed block device");161
static const char *blk_root_get_name(BdrvChild *child)
163
return blk_name(child->opaque);
166
static void blk_vm_state_changed(void *opaque, bool running, RunState state)
168
Error *local_err = NULL;
169
BlockBackend *blk = opaque;
171
if (state == RUN_STATE_INMIGRATE) {175
qemu_del_vm_change_state_handler(blk->vmsh);
177
blk_set_perm(blk, blk->perm, blk->shared_perm, &local_err);
179
error_report_err(local_err);
184
* Notifies the user of the BlockBackend that migration has completed. qdev
185
* devices can tighten their permissions in response (specifically revoke
186
* shared write permissions that we needed for storage migration).
188
* If an error is returned, the VM cannot be allowed to be resumed.
190
static void GRAPH_RDLOCK blk_root_activate(BdrvChild *child, Error **errp)
192
BlockBackend *blk = child->opaque;
193
Error *local_err = NULL;
194
uint64_t saved_shared_perm;
196
if (!blk->disable_perm) {200
blk->disable_perm = false;
203
* blk->shared_perm contains the permissions we want to share once
204
* migration is really completely done. For now, we need to share
205
* all; but we also need to retain blk->shared_perm, which is
206
* overwritten by a successful blk_set_perm() call. Save it and
209
saved_shared_perm = blk->shared_perm;
211
blk_set_perm_locked(blk, blk->perm, BLK_PERM_ALL, &local_err);
213
error_propagate(errp, local_err);
214
blk->disable_perm = true;
217
blk->shared_perm = saved_shared_perm;
219
if (runstate_check(RUN_STATE_INMIGRATE)) {220
/* Activation can happen when migration process is still active, for
221
* example when nbd_server_add is called during non-shared storage
222
* migration. Defer the shared_perm update to migration completion. */
224
blk->vmsh = qemu_add_vm_change_state_handler(blk_vm_state_changed,
230
blk_set_perm_locked(blk, blk->perm, blk->shared_perm, &local_err);
232
error_propagate(errp, local_err);
233
blk->disable_perm = true;
238
void blk_set_force_allow_inactivate(BlockBackend *blk)
241
blk->force_allow_inactivate = true;
244
static bool blk_can_inactivate(BlockBackend *blk)
246
/* If it is a guest device, inactivate is ok. */
247
if (blk->dev || blk_name(blk)[0]) {251
/* Inactivating means no more writes to the image can be done,
252
* even if those writes would be changes invisible to the
253
* guest. For block job BBs that satisfy this, we can just allow
254
* it. This is the case for mirror job source, which is required
255
* by libvirt non-shared block migration. */
256
if (!(blk->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED))) {260
return blk->force_allow_inactivate;
263
static int GRAPH_RDLOCK blk_root_inactivate(BdrvChild *child)
265
BlockBackend *blk = child->opaque;
267
if (blk->disable_perm) {271
if (!blk_can_inactivate(blk)) {275
blk->disable_perm = true;
277
bdrv_child_try_set_perm(blk->root, 0, BLK_PERM_ALL, &error_abort);
283
static void blk_root_attach(BdrvChild *child)
285
BlockBackend *blk = child->opaque;
286
BlockBackendAioNotifier *notifier;
288
trace_blk_root_attach(child, blk, child->bs);
290
QLIST_FOREACH(notifier, &blk->aio_notifiers, list) {291
bdrv_add_aio_context_notifier(child->bs,
292
notifier->attached_aio_context,
293
notifier->detach_aio_context,
298
static void blk_root_detach(BdrvChild *child)
300
BlockBackend *blk = child->opaque;
301
BlockBackendAioNotifier *notifier;
303
trace_blk_root_detach(child, blk, child->bs);
305
QLIST_FOREACH(notifier, &blk->aio_notifiers, list) {306
bdrv_remove_aio_context_notifier(child->bs,
307
notifier->attached_aio_context,
308
notifier->detach_aio_context,
313
static AioContext *blk_root_get_parent_aio_context(BdrvChild *c)
315
BlockBackend *blk = c->opaque;
318
return blk_get_aio_context(blk);
321
static const BdrvChildClass child_root = {322
.inherit_options = blk_root_inherit_options,
324
.change_media = blk_root_change_media,
325
.resize = blk_root_resize,
326
.get_name = blk_root_get_name,
327
.get_parent_desc = blk_root_get_parent_desc,
329
.drained_begin = blk_root_drained_begin,
330
.drained_poll = blk_root_drained_poll,
331
.drained_end = blk_root_drained_end,
333
.activate = blk_root_activate,
334
.inactivate = blk_root_inactivate,
336
.attach = blk_root_attach,
337
.detach = blk_root_detach,
339
.change_aio_ctx = blk_root_change_aio_ctx,
341
.get_parent_aio_context = blk_root_get_parent_aio_context,
345
* Create a new BlockBackend with a reference count of one.
347
* @perm is a bitmasks of BLK_PERM_* constants which describes the permissions
348
* to request for a block driver node that is attached to this BlockBackend.
349
* @shared_perm is a bitmask which describes which permissions may be granted
350
* to other users of the attached node.
351
* Both sets of permissions can be changed later using blk_set_perm().
353
* Return the new BlockBackend on success, null on failure.
355
BlockBackend *blk_new(AioContext *ctx, uint64_t perm, uint64_t shared_perm)
361
blk = g_new0(BlockBackend, 1);
365
blk->shared_perm = shared_perm;
366
blk_set_enable_write_cache(blk, true);
368
blk->on_read_error = BLOCKDEV_ON_ERROR_REPORT;
369
blk->on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
371
block_acct_init(&blk->stats);
373
qemu_mutex_init(&blk->queued_requests_lock);
374
qemu_co_queue_init(&blk->queued_requests);
375
notifier_list_init(&blk->remove_bs_notifiers);
376
notifier_list_init(&blk->insert_bs_notifiers);
377
QLIST_INIT(&blk->aio_notifiers);
379
QTAILQ_INSERT_TAIL(&block_backends, blk, link);
384
* Create a new BlockBackend connected to an existing BlockDriverState.
386
* @perm is a bitmasks of BLK_PERM_* constants which describes the
387
* permissions to request for @bs that is attached to this
388
* BlockBackend. @shared_perm is a bitmask which describes which
389
* permissions may be granted to other users of the attached node.
390
* Both sets of permissions can be changed later using blk_set_perm().
392
* Return the new BlockBackend on success, null on failure.
394
BlockBackend *blk_new_with_bs(BlockDriverState *bs, uint64_t perm,
395
uint64_t shared_perm, Error **errp)
397
BlockBackend *blk = blk_new(bdrv_get_aio_context(bs), perm, shared_perm);
401
if (blk_insert_bs(blk, bs, errp) < 0) {409
* Creates a new BlockBackend, opens a new BlockDriverState, and connects both.
410
* By default, the new BlockBackend is in the main AioContext, but if the
411
* parameters connect it with any existing node in a different AioContext, it
412
* may end up there instead.
414
* Just as with bdrv_open(), after having called this function the reference to
415
* @options belongs to the block layer (even on failure).
417
* TODO: Remove @filename and @flags; it should be possible to specify a whole
418
* BDS tree just by specifying the @options QDict (or @reference,
419
* alternatively). At the time of adding this function, this is not possible,
420
* though, so callers of this function have to be able to specify @filename and
423
BlockBackend *blk_new_open(const char *filename, const char *reference,
424
QDict *options, int flags, Error **errp)
427
BlockDriverState *bs;
429
uint64_t shared = BLK_PERM_ALL;
434
* blk_new_open() is mainly used in .bdrv_create implementations and the
435
* tools where sharing isn't a major concern because the BDS stays private
436
* and the file is generally not supposed to be used by a second process,
437
* so we just request permission according to the flags.
439
* The exceptions are xen_disk and blockdev_init(); in these cases, the
440
* caller of blk_new_open() doesn't make use of the permissions, but they
441
* shouldn't hurt either. We can still share everything here because the
442
* guest devices will add their own blockers if they can't share.
444
if ((flags & BDRV_O_NO_IO) == 0) {445
perm |= BLK_PERM_CONSISTENT_READ;
446
if (flags & BDRV_O_RDWR) {447
perm |= BLK_PERM_WRITE;
450
if (flags & BDRV_O_RESIZE) {451
perm |= BLK_PERM_RESIZE;
453
if (flags & BDRV_O_NO_SHARE) {454
shared = BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
457
bs = bdrv_open(filename, reference, options, flags, errp);
462
/* bdrv_open() could have moved bs to a different AioContext */
463
blk = blk_new(bdrv_get_aio_context(bs), perm, shared);
465
blk->shared_perm = shared;
467
blk_insert_bs(blk, bs, errp);
478
static void blk_delete(BlockBackend *blk)
480
assert(!blk->refcnt);
483
if (blk->public.throttle_group_member.throttle_state) {484
blk_io_limits_disable(blk);
490
qemu_del_vm_change_state_handler(blk->vmsh);
493
assert(QLIST_EMPTY(&blk->remove_bs_notifiers.notifiers));
494
assert(QLIST_EMPTY(&blk->insert_bs_notifiers.notifiers));
495
assert(QLIST_EMPTY(&blk->aio_notifiers));
496
assert(qemu_co_queue_empty(&blk->queued_requests));
497
qemu_mutex_destroy(&blk->queued_requests_lock);
498
QTAILQ_REMOVE(&block_backends, blk, link);
499
drive_info_del(blk->legacy_dinfo);
500
block_acct_cleanup(&blk->stats);
504
static void drive_info_del(DriveInfo *dinfo)
509
qemu_opts_del(dinfo->opts);
513
int blk_get_refcnt(BlockBackend *blk)
516
return blk ? blk->refcnt : 0;
520
* Increment @blk's reference count.
521
* @blk must not be null.
523
void blk_ref(BlockBackend *blk)
525
assert(blk->refcnt > 0);
531
* Decrement @blk's reference count.
532
* If this drops it to zero, destroy @blk.
533
* For convenience, do nothing if @blk is null.
535
void blk_unref(BlockBackend *blk)
539
assert(blk->refcnt > 0);
540
if (blk->refcnt > 1) {544
/* blk_drain() cannot resurrect blk, nobody held a reference */
545
assert(blk->refcnt == 1);
553
* Behaves similarly to blk_next() but iterates over all BlockBackends, even the
554
* ones which are hidden (i.e. are not referenced by the monitor).
556
BlockBackend *blk_all_next(BlockBackend *blk)
559
return blk ? QTAILQ_NEXT(blk, link)
560
: QTAILQ_FIRST(&block_backends);
563
void blk_remove_all_bs(void)
565
BlockBackend *blk = NULL;
569
while ((blk = blk_all_next(blk)) != NULL) {577
* Return the monitor-owned BlockBackend after @blk.
578
* If @blk is null, return the first one.
579
* Else, return @blk's next sibling, which may be null.
581
* To iterate over all BlockBackends, do
582
* for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {586
BlockBackend *blk_next(BlockBackend *blk)
589
return blk ? QTAILQ_NEXT(blk, monitor_link)
590
: QTAILQ_FIRST(&monitor_block_backends);
593
/* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
594
* the monitor or attached to a BlockBackend */
595
BlockDriverState *bdrv_next(BdrvNextIterator *it)
597
BlockDriverState *bs, *old_bs;
599
/* Must be called from the main loop */
600
assert(qemu_get_current_aio_context() == qemu_get_aio_context());
604
/* First, return all root nodes of BlockBackends. In order to avoid
605
* returning a BDS twice when multiple BBs refer to it, we only return it
606
* if the BB is the first one in the parent list of the BDS. */
607
if (it->phase == BDRV_NEXT_BACKEND_ROOTS) {608
BlockBackend *old_blk = it->blk;
611
it->blk = blk_all_next(it->blk);
612
bs = it->blk ? blk_bs(it->blk) : NULL;
613
} while (it->blk && (bs == NULL || bdrv_first_blk(bs) != it->blk));
626
it->phase = BDRV_NEXT_MONITOR_OWNED;
629
/* Then return the monitor-owned BDSes without a BB attached. Ignore all
630
* BDSes that are attached to a BlockBackend here; they have been handled
631
* by the above block already */
633
it->bs = bdrv_next_monitor_owned(it->bs);
635
} while (bs && bdrv_has_blk(bs));
645
static void bdrv_next_reset(BdrvNextIterator *it)
647
*it = (BdrvNextIterator) {648
.phase = BDRV_NEXT_BACKEND_ROOTS,
652
BlockDriverState *bdrv_first(BdrvNextIterator *it)
656
return bdrv_next(it);
659
/* Must be called when aborting a bdrv_next() iteration before
660
* bdrv_next() returns NULL */
661
void bdrv_next_cleanup(BdrvNextIterator *it)
663
/* Must be called from the main loop */
664
assert(qemu_get_current_aio_context() == qemu_get_aio_context());
668
if (it->phase == BDRV_NEXT_BACKEND_ROOTS && it->blk) {676
* Add a BlockBackend into the list of backends referenced by the monitor, with
677
* the given @name acting as the handle for the monitor.
678
* Strictly for use by blockdev.c.
680
* @name must not be null or empty.
682
* Returns true on success and false on failure. In the latter case, an Error
683
* object is returned through @errp.
685
bool monitor_add_blk(BlockBackend *blk, const char *name, Error **errp)
688
assert(name && name[0]);
691
if (!id_wellformed(name)) {692
error_setg(errp, "Invalid device name");
695
if (blk_by_name(name)) {696
error_setg(errp, "Device with id '%s' already exists", name);
699
if (bdrv_find_node(name)) {701
"Device name '%s' conflicts with an existing node name",
706
blk->name = g_strdup(name);
707
QTAILQ_INSERT_TAIL(&monitor_block_backends, blk, monitor_link);
712
* Remove a BlockBackend from the list of backends referenced by the monitor.
713
* Strictly for use by blockdev.c.
715
void monitor_remove_blk(BlockBackend *blk)
723
QTAILQ_REMOVE(&monitor_block_backends, blk, monitor_link);
729
* Return @blk's name, a non-null string.
730
* Returns an empty string iff @blk is not referenced by the monitor.
732
const char *blk_name(const BlockBackend *blk)
735
return blk->name ?: "";
739
* Return the BlockBackend with name @name if it exists, else null.
740
* @name must not be null.
742
BlockBackend *blk_by_name(const char *name)
744
BlockBackend *blk = NULL;
748
while ((blk = blk_next(blk)) != NULL) {749
if (!strcmp(name, blk->name)) {757
* Return the BlockDriverState attached to @blk if any, else null.
759
BlockDriverState *blk_bs(BlockBackend *blk)
762
return blk->root ? blk->root->bs : NULL;
765
static BlockBackend * GRAPH_RDLOCK bdrv_first_blk(BlockDriverState *bs)
770
assert_bdrv_graph_readable();
772
QLIST_FOREACH(child, &bs->parents, next_parent) {773
if (child->klass == &child_root) {774
return child->opaque;
782
* Returns true if @bs has an associated BlockBackend.
784
bool bdrv_has_blk(BlockDriverState *bs)
787
return bdrv_first_blk(bs) != NULL;
791
* Returns true if @bs has only BlockBackends as parents.
793
bool bdrv_is_root_node(BlockDriverState *bs)
798
assert_bdrv_graph_readable();
800
QLIST_FOREACH(c, &bs->parents, next_parent) {801
if (c->klass != &child_root) {810
* Return @blk's DriveInfo if any, else null.
812
DriveInfo *blk_legacy_dinfo(BlockBackend *blk)
815
return blk->legacy_dinfo;
819
* Set @blk's DriveInfo to @dinfo, and return it.
820
* @blk must not have a DriveInfo set already.
821
* No other BlockBackend may have the same DriveInfo set.
823
DriveInfo *blk_set_legacy_dinfo(BlockBackend *blk, DriveInfo *dinfo)
825
assert(!blk->legacy_dinfo);
827
return blk->legacy_dinfo = dinfo;
831
* Return the BlockBackend with DriveInfo @dinfo.
834
BlockBackend *blk_by_legacy_dinfo(DriveInfo *dinfo)
836
BlockBackend *blk = NULL;
839
while ((blk = blk_next(blk)) != NULL) {840
if (blk->legacy_dinfo == dinfo) {848
* Returns a pointer to the publicly accessible fields of @blk.
850
BlockBackendPublic *blk_get_public(BlockBackend *blk)
857
* Returns a BlockBackend given the associated @public fields.
859
BlockBackend *blk_by_public(BlockBackendPublic *public)
862
return container_of(public, BlockBackend, public);
866
* Disassociates the currently associated BlockDriverState from @blk.
868
void blk_remove_bs(BlockBackend *blk)
870
ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
875
notifier_list_notify(&blk->remove_bs_notifiers, blk);
876
if (tgm->throttle_state) {877
BlockDriverState *bs = blk_bs(blk);
880
* Take a ref in case blk_bs() changes across bdrv_drained_begin(), for
881
* example, if a temporary filter node is removed by a blockjob.
884
bdrv_drained_begin(bs);
885
throttle_group_detach_aio_context(tgm);
886
throttle_group_attach_aio_context(tgm, qemu_get_aio_context());
887
bdrv_drained_end(bs);
891
blk_update_root_state(blk);
893
/* bdrv_root_unref_child() will cause blk->root to become stale and may
894
* switch to a completion coroutine later on. Let's drain all I/O here
895
* to avoid that and a potential QEMU crash.
902
bdrv_root_unref_child(root);
903
bdrv_graph_wrunlock();
907
* Associates a new BlockDriverState with @blk.
909
int blk_insert_bs(BlockBackend *blk, BlockDriverState *bs, Error **errp)
911
ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
916
blk->root = bdrv_root_attach_child(bs, "root", &child_root,
917
BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY,
918
blk->perm, blk->shared_perm,
920
bdrv_graph_wrunlock();
921
if (blk->root == NULL) {925
notifier_list_notify(&blk->insert_bs_notifiers, blk);
926
if (tgm->throttle_state) {927
throttle_group_detach_aio_context(tgm);
928
throttle_group_attach_aio_context(tgm, bdrv_get_aio_context(bs));
935
* Change BlockDriverState associated with @blk.
937
int blk_replace_bs(BlockBackend *blk, BlockDriverState *new_bs, Error **errp)
940
return bdrv_replace_child_bs(blk->root, new_bs, errp);
944
* Sets the permission bitmasks that the user of the BlockBackend needs.
946
static int coroutine_mixed_fn GRAPH_RDLOCK
947
blk_set_perm_locked(BlockBackend *blk, uint64_t perm, uint64_t shared_perm,
953
if (blk->root && !blk->disable_perm) {954
ret = bdrv_child_try_set_perm(blk->root, perm, shared_perm, errp);
961
blk->shared_perm = shared_perm;
966
int blk_set_perm(BlockBackend *blk, uint64_t perm, uint64_t shared_perm,
970
GRAPH_RDLOCK_GUARD_MAINLOOP();
972
return blk_set_perm_locked(blk, perm, shared_perm, errp);
975
void blk_get_perm(BlockBackend *blk, uint64_t *perm, uint64_t *shared_perm)
979
*shared_perm = blk->shared_perm;
983
* Attach device model @dev to @blk.
984
* Return 0 on success, -EBUSY when a device model is attached already.
986
int blk_attach_dev(BlockBackend *blk, DeviceState *dev)
993
/* While migration is still incoming, we don't need to apply the
994
* permissions of guest device BlockBackends. We might still have a block
995
* job or NBD server writing to the image for storage migration. */
996
if (runstate_check(RUN_STATE_INMIGRATE)) {997
blk->disable_perm = true;
1002
blk_iostatus_reset(blk);
1008
* Detach device model @dev from @blk.
1009
* @dev must be currently attached to @blk.
1011
void blk_detach_dev(BlockBackend *blk, DeviceState *dev)
1013
assert(blk->dev == dev);
1014
GLOBAL_STATE_CODE();
1016
blk->dev_ops = NULL;
1017
blk->dev_opaque = NULL;
1018
blk_set_perm(blk, 0, BLK_PERM_ALL, &error_abort);
1023
* Return the device model attached to @blk if any, else null.
1025
DeviceState *blk_get_attached_dev(BlockBackend *blk)
1027
GLOBAL_STATE_CODE();
1031
/* Return the qdev ID, or if no ID is assigned the QOM path, of the block
1032
* device attached to the BlockBackend. */
1033
char *blk_get_attached_dev_id(BlockBackend *blk)
1035
DeviceState *dev = blk->dev;
1039
return g_strdup("");1040
} else if (dev->id) {1041
return g_strdup(dev->id);
1044
return object_get_canonical_path(OBJECT(dev)) ?: g_strdup("");1048
* Return the BlockBackend which has the device model @dev attached if it
1049
* exists, else null.
1051
* @dev must not be null.
1053
BlockBackend *blk_by_dev(void *dev)
1055
BlockBackend *blk = NULL;
1057
GLOBAL_STATE_CODE();
1059
assert(dev != NULL);
1060
while ((blk = blk_all_next(blk)) != NULL) {1061
if (blk->dev == dev) {1069
* Set @blk's device model callbacks to @ops.
1070
* @opaque is the opaque argument to pass to the callbacks.
1071
* This is for use by device models.
1073
void blk_set_dev_ops(BlockBackend *blk, const BlockDevOps *ops,
1076
GLOBAL_STATE_CODE();
1078
blk->dev_opaque = opaque;
1080
/* Are we currently quiesced? Should we enforce this right now? */
1081
if (qatomic_read(&blk->quiesce_counter) && ops && ops->drained_begin) {1082
ops->drained_begin(opaque);
1087
* Notify @blk's attached device model of media change.
1089
* If @load is true, notify of media load. This action can fail, meaning that
1090
* the medium cannot be loaded. @errp is set then.
1092
* If @load is false, notify of media eject. This can never fail.
1094
* Also send DEVICE_TRAY_MOVED events as appropriate.
1096
void blk_dev_change_media_cb(BlockBackend *blk, bool load, Error **errp)
1098
GLOBAL_STATE_CODE();
1099
if (blk->dev_ops && blk->dev_ops->change_media_cb) {1100
bool tray_was_open, tray_is_open;
1101
Error *local_err = NULL;
1103
tray_was_open = blk_dev_is_tray_open(blk);
1104
blk->dev_ops->change_media_cb(blk->dev_opaque, load, &local_err);
1106
assert(load == true);
1107
error_propagate(errp, local_err);
1110
tray_is_open = blk_dev_is_tray_open(blk);
1112
if (tray_was_open != tray_is_open) {1113
char *id = blk_get_attached_dev_id(blk);
1114
qapi_event_send_device_tray_moved(blk_name(blk), id, tray_is_open);
1120
static void blk_root_change_media(BdrvChild *child, bool load)
1122
blk_dev_change_media_cb(child->opaque, load, NULL);
1126
* Does @blk's attached device model have removable media?
1127
* %true if no device model is attached.
1129
bool blk_dev_has_removable_media(BlockBackend *blk)
1131
GLOBAL_STATE_CODE();
1132
return !blk->dev || (blk->dev_ops && blk->dev_ops->change_media_cb);
1136
* Does @blk's attached device model have a tray?
1138
bool blk_dev_has_tray(BlockBackend *blk)
1141
return blk->dev_ops && blk->dev_ops->is_tray_open;
1145
* Notify @blk's attached device model of a media eject request.
1146
* If @force is true, the medium is about to be yanked out forcefully.
1148
void blk_dev_eject_request(BlockBackend *blk, bool force)
1150
GLOBAL_STATE_CODE();
1151
if (blk->dev_ops && blk->dev_ops->eject_request_cb) {1152
blk->dev_ops->eject_request_cb(blk->dev_opaque, force);
1157
* Does @blk's attached device model have a tray, and is it open?
1159
bool blk_dev_is_tray_open(BlockBackend *blk)
1162
if (blk_dev_has_tray(blk)) {1163
return blk->dev_ops->is_tray_open(blk->dev_opaque);
1169
* Does @blk's attached device model have the medium locked?
1170
* %false if the device model has no such lock.
1172
bool blk_dev_is_medium_locked(BlockBackend *blk)
1174
GLOBAL_STATE_CODE();
1175
if (blk->dev_ops && blk->dev_ops->is_medium_locked) {1176
return blk->dev_ops->is_medium_locked(blk->dev_opaque);
1182
* Notify @blk's attached device model of a backend size change.
1184
static void blk_root_resize(BdrvChild *child)
1186
BlockBackend *blk = child->opaque;
1188
if (blk->dev_ops && blk->dev_ops->resize_cb) {1189
blk->dev_ops->resize_cb(blk->dev_opaque);
1193
void blk_iostatus_enable(BlockBackend *blk)
1195
GLOBAL_STATE_CODE();
1196
blk->iostatus_enabled = true;
1197
blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
1200
/* The I/O status is only enabled if the drive explicitly
1201
* enables it _and_ the VM is configured to stop on errors */
1202
bool blk_iostatus_is_enabled(const BlockBackend *blk)
1205
return (blk->iostatus_enabled &&
1206
(blk->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
1207
blk->on_write_error == BLOCKDEV_ON_ERROR_STOP ||
1208
blk->on_read_error == BLOCKDEV_ON_ERROR_STOP));
1211
BlockDeviceIoStatus blk_iostatus(const BlockBackend *blk)
1213
GLOBAL_STATE_CODE();
1214
return blk->iostatus;
1217
void blk_iostatus_disable(BlockBackend *blk)
1219
GLOBAL_STATE_CODE();
1220
blk->iostatus_enabled = false;
1223
void blk_iostatus_reset(BlockBackend *blk)
1225
GLOBAL_STATE_CODE();
1226
if (blk_iostatus_is_enabled(blk)) {1227
blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
1231
void blk_iostatus_set_err(BlockBackend *blk, int error)
1234
assert(blk_iostatus_is_enabled(blk));
1235
if (blk->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {1236
blk->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
1237
BLOCK_DEVICE_IO_STATUS_FAILED;
1241
void blk_set_allow_write_beyond_eof(BlockBackend *blk, bool allow)
1244
blk->allow_write_beyond_eof = allow;
1247
void blk_set_allow_aio_context_change(BlockBackend *blk, bool allow)
1250
blk->allow_aio_context_change = allow;
1253
void blk_set_disable_request_queuing(BlockBackend *blk, bool disable)
1256
qatomic_set(&blk->disable_request_queuing, disable);
1259
static int coroutine_fn GRAPH_RDLOCK
1260
blk_check_byte_request(BlockBackend *blk, int64_t offset, int64_t bytes)
1268
if (!blk_co_is_available(blk)) {1276
if (!blk->allow_write_beyond_eof) {1277
len = bdrv_co_getlength(blk_bs(blk));
1282
if (offset > len || len - offset < bytes) {1290
/* Are we currently in a drained section? */
1291
bool blk_in_drain(BlockBackend *blk)
1293
GLOBAL_STATE_CODE(); /* change to IO_OR_GS_CODE(), if necessary */
1294
return qatomic_read(&blk->quiesce_counter);
1297
/* To be called between exactly one pair of blk_inc/dec_in_flight() */
1298
static void coroutine_fn blk_wait_while_drained(BlockBackend *blk)
1300
assert(blk->in_flight > 0);
1302
if (qatomic_read(&blk->quiesce_counter) &&
1303
!qatomic_read(&blk->disable_request_queuing)) {1305
* Take lock before decrementing in flight counter so main loop thread
1306
* waits for us to enqueue ourselves before it can leave the drained
1309
qemu_mutex_lock(&blk->queued_requests_lock);
1310
blk_dec_in_flight(blk);
1311
qemu_co_queue_wait(&blk->queued_requests, &blk->queued_requests_lock);
1312
blk_inc_in_flight(blk);
1313
qemu_mutex_unlock(&blk->queued_requests_lock);
1317
/* To be called between exactly one pair of blk_inc/dec_in_flight() */
1318
static int coroutine_fn
1319
blk_co_do_preadv_part(BlockBackend *blk, int64_t offset, int64_t bytes,
1320
QEMUIOVector *qiov, size_t qiov_offset,
1321
BdrvRequestFlags flags)
1324
BlockDriverState *bs;
1327
blk_wait_while_drained(blk);
1328
GRAPH_RDLOCK_GUARD();
1330
/* Call blk_bs() only after waiting, the graph may have changed */
1332
trace_blk_co_preadv(blk, bs, offset, bytes, flags);
1334
ret = blk_check_byte_request(blk, offset, bytes);
1339
bdrv_inc_in_flight(bs);
1341
/* throttling disk I/O */
1342
if (blk->public.throttle_group_member.throttle_state) {1343
throttle_group_co_io_limits_intercept(&blk->public.throttle_group_member,
1344
bytes, THROTTLE_READ);
1347
ret = bdrv_co_preadv_part(blk->root, offset, bytes, qiov, qiov_offset,
1349
bdrv_dec_in_flight(bs);
1353
int coroutine_fn blk_co_pread(BlockBackend *blk, int64_t offset, int64_t bytes,
1354
void *buf, BdrvRequestFlags flags)
1356
QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
1359
assert(bytes <= SIZE_MAX);
1361
return blk_co_preadv(blk, offset, bytes, &qiov, flags);
1364
int coroutine_fn blk_co_preadv(BlockBackend *blk, int64_t offset,
1365
int64_t bytes, QEMUIOVector *qiov,
1366
BdrvRequestFlags flags)
1371
blk_inc_in_flight(blk);
1372
ret = blk_co_do_preadv_part(blk, offset, bytes, qiov, 0, flags);
1373
blk_dec_in_flight(blk);
1378
int coroutine_fn blk_co_preadv_part(BlockBackend *blk, int64_t offset,
1379
int64_t bytes, QEMUIOVector *qiov,
1380
size_t qiov_offset, BdrvRequestFlags flags)
1385
blk_inc_in_flight(blk);
1386
ret = blk_co_do_preadv_part(blk, offset, bytes, qiov, qiov_offset, flags);
1387
blk_dec_in_flight(blk);
1392
/* To be called between exactly one pair of blk_inc/dec_in_flight() */
1393
static int coroutine_fn
1394
blk_co_do_pwritev_part(BlockBackend *blk, int64_t offset, int64_t bytes,
1395
QEMUIOVector *qiov, size_t qiov_offset,
1396
BdrvRequestFlags flags)
1399
BlockDriverState *bs;
1402
blk_wait_while_drained(blk);
1403
GRAPH_RDLOCK_GUARD();
1405
/* Call blk_bs() only after waiting, the graph may have changed */
1407
trace_blk_co_pwritev(blk, bs, offset, bytes, flags);
1409
ret = blk_check_byte_request(blk, offset, bytes);
1414
bdrv_inc_in_flight(bs);
1415
/* throttling disk I/O */
1416
if (blk->public.throttle_group_member.throttle_state) {1417
throttle_group_co_io_limits_intercept(&blk->public.throttle_group_member,
1418
bytes, THROTTLE_WRITE);
1421
if (!blk->enable_write_cache) {1422
flags |= BDRV_REQ_FUA;
1425
ret = bdrv_co_pwritev_part(blk->root, offset, bytes, qiov, qiov_offset,
1427
bdrv_dec_in_flight(bs);
1431
int coroutine_fn blk_co_pwritev_part(BlockBackend *blk, int64_t offset,
1433
QEMUIOVector *qiov, size_t qiov_offset,
1434
BdrvRequestFlags flags)
1439
blk_inc_in_flight(blk);
1440
ret = blk_co_do_pwritev_part(blk, offset, bytes, qiov, qiov_offset, flags);
1441
blk_dec_in_flight(blk);
1446
int coroutine_fn blk_co_pwrite(BlockBackend *blk, int64_t offset, int64_t bytes,
1447
const void *buf, BdrvRequestFlags flags)
1449
QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
1452
assert(bytes <= SIZE_MAX);
1454
return blk_co_pwritev(blk, offset, bytes, &qiov, flags);
1457
int coroutine_fn blk_co_pwritev(BlockBackend *blk, int64_t offset,
1458
int64_t bytes, QEMUIOVector *qiov,
1459
BdrvRequestFlags flags)
1462
return blk_co_pwritev_part(blk, offset, bytes, qiov, 0, flags);
1465
int coroutine_fn blk_co_block_status_above(BlockBackend *blk,
1466
BlockDriverState *base,
1467
int64_t offset, int64_t bytes,
1468
int64_t *pnum, int64_t *map,
1469
BlockDriverState **file)
1472
GRAPH_RDLOCK_GUARD();
1473
return bdrv_co_block_status_above(blk_bs(blk), base, offset, bytes, pnum,
1477
int coroutine_fn blk_co_is_allocated_above(BlockBackend *blk,
1478
BlockDriverState *base,
1479
bool include_base, int64_t offset,
1480
int64_t bytes, int64_t *pnum)
1483
GRAPH_RDLOCK_GUARD();
1484
return bdrv_co_is_allocated_above(blk_bs(blk), base, include_base, offset,
1488
typedef struct BlkRwCo {1493
BdrvRequestFlags flags;
1496
int blk_make_zero(BlockBackend *blk, BdrvRequestFlags flags)
1498
GLOBAL_STATE_CODE();
1499
return bdrv_make_zero(blk->root, flags);
1502
void blk_inc_in_flight(BlockBackend *blk)
1505
qatomic_inc(&blk->in_flight);
1508
void blk_dec_in_flight(BlockBackend *blk)
1511
qatomic_dec(&blk->in_flight);
1515
static void error_callback_bh(void *opaque)
1517
struct BlockBackendAIOCB *acb = opaque;
1519
blk_dec_in_flight(acb->blk);
1520
acb->common.cb(acb->common.opaque, acb->ret);
1521
qemu_aio_unref(acb);
1524
BlockAIOCB *blk_abort_aio_request(BlockBackend *blk,
1525
BlockCompletionFunc *cb,
1526
void *opaque, int ret)
1528
struct BlockBackendAIOCB *acb;
1531
blk_inc_in_flight(blk);
1532
acb = blk_aio_get(&block_backend_aiocb_info, blk, cb, opaque);
1536
replay_bh_schedule_oneshot_event(qemu_get_current_aio_context(),
1537
error_callback_bh, acb);
1538
return &acb->common;
1541
typedef struct BlkAioEmAIOCB {1548
static const AIOCBInfo blk_aio_em_aiocb_info = {1549
.aiocb_size = sizeof(BlkAioEmAIOCB),
1552
static void blk_aio_complete(BlkAioEmAIOCB *acb)
1554
if (acb->has_returned) {1555
acb->common.cb(acb->common.opaque, acb->rwco.ret);
1556
blk_dec_in_flight(acb->rwco.blk);
1557
qemu_aio_unref(acb);
1561
static void blk_aio_complete_bh(void *opaque)
1563
BlkAioEmAIOCB *acb = opaque;
1564
assert(acb->has_returned);
1565
blk_aio_complete(acb);
1568
static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset,
1570
void *iobuf, CoroutineEntry co_entry,
1571
BdrvRequestFlags flags,
1572
BlockCompletionFunc *cb, void *opaque)
1577
blk_inc_in_flight(blk);
1578
acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1579
acb->rwco = (BlkRwCo) {1587
acb->has_returned = false;
1589
co = qemu_coroutine_create(co_entry, acb);
1590
aio_co_enter(qemu_get_current_aio_context(), co);
1592
acb->has_returned = true;
1593
if (acb->rwco.ret != NOT_DONE) {1594
replay_bh_schedule_oneshot_event(qemu_get_current_aio_context(),
1595
blk_aio_complete_bh, acb);
1598
return &acb->common;
1601
static void coroutine_fn blk_aio_read_entry(void *opaque)
1603
BlkAioEmAIOCB *acb = opaque;
1604
BlkRwCo *rwco = &acb->rwco;
1605
QEMUIOVector *qiov = rwco->iobuf;
1607
assert(qiov->size == acb->bytes);
1608
rwco->ret = blk_co_do_preadv_part(rwco->blk, rwco->offset, acb->bytes, qiov,
1610
blk_aio_complete(acb);
1613
static void coroutine_fn blk_aio_write_entry(void *opaque)
1615
BlkAioEmAIOCB *acb = opaque;
1616
BlkRwCo *rwco = &acb->rwco;
1617
QEMUIOVector *qiov = rwco->iobuf;
1619
assert(!qiov || qiov->size == acb->bytes);
1620
rwco->ret = blk_co_do_pwritev_part(rwco->blk, rwco->offset, acb->bytes,
1621
qiov, 0, rwco->flags);
1622
blk_aio_complete(acb);
1625
BlockAIOCB *blk_aio_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1626
int64_t bytes, BdrvRequestFlags flags,
1627
BlockCompletionFunc *cb, void *opaque)
1630
return blk_aio_prwv(blk, offset, bytes, NULL, blk_aio_write_entry,
1631
flags | BDRV_REQ_ZERO_WRITE, cb, opaque);
1634
int64_t coroutine_fn blk_co_getlength(BlockBackend *blk)
1637
GRAPH_RDLOCK_GUARD();
1639
if (!blk_co_is_available(blk)) {1643
return bdrv_co_getlength(blk_bs(blk));
1646
int64_t coroutine_fn blk_co_nb_sectors(BlockBackend *blk)
1648
BlockDriverState *bs = blk_bs(blk);
1651
GRAPH_RDLOCK_GUARD();
1656
return bdrv_co_nb_sectors(bs);
1661
* This wrapper is written by hand because this function is in the hot I/O path,
1662
* via blk_get_geometry.
1664
int64_t coroutine_mixed_fn blk_nb_sectors(BlockBackend *blk)
1666
BlockDriverState *bs = blk_bs(blk);
1673
return bdrv_nb_sectors(bs);
1677
/* return 0 as number of sectors if no device present or error */
1678
void coroutine_fn blk_co_get_geometry(BlockBackend *blk,
1679
uint64_t *nb_sectors_ptr)
1681
int64_t ret = blk_co_nb_sectors(blk);
1682
*nb_sectors_ptr = ret < 0 ? 0 : ret;
1686
* This wrapper is written by hand because this function is in the hot I/O path.
1688
void coroutine_mixed_fn blk_get_geometry(BlockBackend *blk,
1689
uint64_t *nb_sectors_ptr)
1691
int64_t ret = blk_nb_sectors(blk);
1692
*nb_sectors_ptr = ret < 0 ? 0 : ret;
1695
BlockAIOCB *blk_aio_preadv(BlockBackend *blk, int64_t offset,
1696
QEMUIOVector *qiov, BdrvRequestFlags flags,
1697
BlockCompletionFunc *cb, void *opaque)
1700
assert((uint64_t)qiov->size <= INT64_MAX);
1701
return blk_aio_prwv(blk, offset, qiov->size, qiov,
1702
blk_aio_read_entry, flags, cb, opaque);
1705
BlockAIOCB *blk_aio_pwritev(BlockBackend *blk, int64_t offset,
1706
QEMUIOVector *qiov, BdrvRequestFlags flags,
1707
BlockCompletionFunc *cb, void *opaque)
1710
assert((uint64_t)qiov->size <= INT64_MAX);
1711
return blk_aio_prwv(blk, offset, qiov->size, qiov,
1712
blk_aio_write_entry, flags, cb, opaque);
1715
void blk_aio_cancel(BlockAIOCB *acb)
1717
GLOBAL_STATE_CODE();
1718
bdrv_aio_cancel(acb);
1721
void blk_aio_cancel_async(BlockAIOCB *acb)
1724
bdrv_aio_cancel_async(acb);
1727
/* To be called between exactly one pair of blk_inc/dec_in_flight() */
1728
static int coroutine_fn
1729
blk_co_do_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1733
blk_wait_while_drained(blk);
1734
GRAPH_RDLOCK_GUARD();
1736
if (!blk_co_is_available(blk)) {1740
return bdrv_co_ioctl(blk_bs(blk), req, buf);
1743
int coroutine_fn blk_co_ioctl(BlockBackend *blk, unsigned long int req,
1749
blk_inc_in_flight(blk);
1750
ret = blk_co_do_ioctl(blk, req, buf);
1751
blk_dec_in_flight(blk);
1756
static void coroutine_fn blk_aio_ioctl_entry(void *opaque)
1758
BlkAioEmAIOCB *acb = opaque;
1759
BlkRwCo *rwco = &acb->rwco;
1761
rwco->ret = blk_co_do_ioctl(rwco->blk, rwco->offset, rwco->iobuf);
1763
blk_aio_complete(acb);
1766
BlockAIOCB *blk_aio_ioctl(BlockBackend *blk, unsigned long int req, void *buf,
1767
BlockCompletionFunc *cb, void *opaque)
1770
return blk_aio_prwv(blk, req, 0, buf, blk_aio_ioctl_entry, 0, cb, opaque);
1773
/* To be called between exactly one pair of blk_inc/dec_in_flight() */
1774
static int coroutine_fn
1775
blk_co_do_pdiscard(BlockBackend *blk, int64_t offset, int64_t bytes)
1780
blk_wait_while_drained(blk);
1781
GRAPH_RDLOCK_GUARD();
1783
ret = blk_check_byte_request(blk, offset, bytes);
1788
return bdrv_co_pdiscard(blk->root, offset, bytes);
1791
static void coroutine_fn blk_aio_pdiscard_entry(void *opaque)
1793
BlkAioEmAIOCB *acb = opaque;
1794
BlkRwCo *rwco = &acb->rwco;
1796
rwco->ret = blk_co_do_pdiscard(rwco->blk, rwco->offset, acb->bytes);
1797
blk_aio_complete(acb);
1800
BlockAIOCB *blk_aio_pdiscard(BlockBackend *blk,
1801
int64_t offset, int64_t bytes,
1802
BlockCompletionFunc *cb, void *opaque)
1805
return blk_aio_prwv(blk, offset, bytes, NULL, blk_aio_pdiscard_entry, 0,
1809
int coroutine_fn blk_co_pdiscard(BlockBackend *blk, int64_t offset,
1815
blk_inc_in_flight(blk);
1816
ret = blk_co_do_pdiscard(blk, offset, bytes);
1817
blk_dec_in_flight(blk);
1822
/* To be called between exactly one pair of blk_inc/dec_in_flight() */
1823
static int coroutine_fn blk_co_do_flush(BlockBackend *blk)
1826
blk_wait_while_drained(blk);
1827
GRAPH_RDLOCK_GUARD();
1829
if (!blk_co_is_available(blk)) {1833
return bdrv_co_flush(blk_bs(blk));
1836
static void coroutine_fn blk_aio_flush_entry(void *opaque)
1838
BlkAioEmAIOCB *acb = opaque;
1839
BlkRwCo *rwco = &acb->rwco;
1841
rwco->ret = blk_co_do_flush(rwco->blk);
1842
blk_aio_complete(acb);
1845
BlockAIOCB *blk_aio_flush(BlockBackend *blk,
1846
BlockCompletionFunc *cb, void *opaque)
1849
return blk_aio_prwv(blk, 0, 0, NULL, blk_aio_flush_entry, 0, cb, opaque);
1852
int coroutine_fn blk_co_flush(BlockBackend *blk)
1857
blk_inc_in_flight(blk);
1858
ret = blk_co_do_flush(blk);
1859
blk_dec_in_flight(blk);
1864
static void coroutine_fn blk_aio_zone_report_entry(void *opaque)
1866
BlkAioEmAIOCB *acb = opaque;
1867
BlkRwCo *rwco = &acb->rwco;
1869
rwco->ret = blk_co_zone_report(rwco->blk, rwco->offset,
1870
(unsigned int*)(uintptr_t)acb->bytes,
1872
blk_aio_complete(acb);
1875
BlockAIOCB *blk_aio_zone_report(BlockBackend *blk, int64_t offset,
1876
unsigned int *nr_zones,
1877
BlockZoneDescriptor *zones,
1878
BlockCompletionFunc *cb, void *opaque)
1884
blk_inc_in_flight(blk);
1885
acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1886
acb->rwco = (BlkRwCo) {1892
acb->bytes = (int64_t)(uintptr_t)nr_zones,
1893
acb->has_returned = false;
1895
co = qemu_coroutine_create(blk_aio_zone_report_entry, acb);
1896
aio_co_enter(qemu_get_current_aio_context(), co);
1898
acb->has_returned = true;
1899
if (acb->rwco.ret != NOT_DONE) {1900
replay_bh_schedule_oneshot_event(qemu_get_current_aio_context(),
1901
blk_aio_complete_bh, acb);
1904
return &acb->common;
1907
static void coroutine_fn blk_aio_zone_mgmt_entry(void *opaque)
1909
BlkAioEmAIOCB *acb = opaque;
1910
BlkRwCo *rwco = &acb->rwco;
1912
rwco->ret = blk_co_zone_mgmt(rwco->blk,
1913
(BlockZoneOp)(uintptr_t)rwco->iobuf,
1914
rwco->offset, acb->bytes);
1915
blk_aio_complete(acb);
1918
BlockAIOCB *blk_aio_zone_mgmt(BlockBackend *blk, BlockZoneOp op,
1919
int64_t offset, int64_t len,
1920
BlockCompletionFunc *cb, void *opaque) {1925
blk_inc_in_flight(blk);
1926
acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1927
acb->rwco = (BlkRwCo) {1930
.iobuf = (void *)(uintptr_t)op,
1934
acb->has_returned = false;
1936
co = qemu_coroutine_create(blk_aio_zone_mgmt_entry, acb);
1937
aio_co_enter(qemu_get_current_aio_context(), co);
1939
acb->has_returned = true;
1940
if (acb->rwco.ret != NOT_DONE) {1941
replay_bh_schedule_oneshot_event(qemu_get_current_aio_context(),
1942
blk_aio_complete_bh, acb);
1945
return &acb->common;
1948
static void coroutine_fn blk_aio_zone_append_entry(void *opaque)
1950
BlkAioEmAIOCB *acb = opaque;
1951
BlkRwCo *rwco = &acb->rwco;
1953
rwco->ret = blk_co_zone_append(rwco->blk, (int64_t *)(uintptr_t)acb->bytes,
1954
rwco->iobuf, rwco->flags);
1955
blk_aio_complete(acb);
1958
BlockAIOCB *blk_aio_zone_append(BlockBackend *blk, int64_t *offset,
1959
QEMUIOVector *qiov, BdrvRequestFlags flags,
1960
BlockCompletionFunc *cb, void *opaque) {1965
blk_inc_in_flight(blk);
1966
acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1967
acb->rwco = (BlkRwCo) {1973
acb->bytes = (int64_t)(uintptr_t)offset;
1974
acb->has_returned = false;
1976
co = qemu_coroutine_create(blk_aio_zone_append_entry, acb);
1977
aio_co_enter(qemu_get_current_aio_context(), co);
1978
acb->has_returned = true;
1979
if (acb->rwco.ret != NOT_DONE) {1980
replay_bh_schedule_oneshot_event(qemu_get_current_aio_context(),
1981
blk_aio_complete_bh, acb);
1984
return &acb->common;
1988
* Send a zone_report command.
1989
* offset is a byte offset from the start of the device. No alignment
1990
* required for offset.
1991
* nr_zones represents IN maximum and OUT actual.
1993
int coroutine_fn blk_co_zone_report(BlockBackend *blk, int64_t offset,
1994
unsigned int *nr_zones,
1995
BlockZoneDescriptor *zones)
2000
blk_inc_in_flight(blk); /* increase before waiting */
2001
blk_wait_while_drained(blk);
2002
GRAPH_RDLOCK_GUARD();
2003
if (!blk_is_available(blk)) {2004
blk_dec_in_flight(blk);
2007
ret = bdrv_co_zone_report(blk_bs(blk), offset, nr_zones, zones);
2008
blk_dec_in_flight(blk);
2013
* Send a zone_management command.
2014
* op is the zone operation;
2015
* offset is the byte offset from the start of the zoned device;
2016
* len is the maximum number of bytes the command should operate on. It
2017
* should be aligned with the device zone size.
2019
int coroutine_fn blk_co_zone_mgmt(BlockBackend *blk, BlockZoneOp op,
2020
int64_t offset, int64_t len)
2025
blk_inc_in_flight(blk);
2026
blk_wait_while_drained(blk);
2027
GRAPH_RDLOCK_GUARD();
2029
ret = blk_check_byte_request(blk, offset, len);
2031
blk_dec_in_flight(blk);
2035
ret = bdrv_co_zone_mgmt(blk_bs(blk), op, offset, len);
2036
blk_dec_in_flight(blk);
2041
* Send a zone_append command.
2043
int coroutine_fn blk_co_zone_append(BlockBackend *blk, int64_t *offset,
2044
QEMUIOVector *qiov, BdrvRequestFlags flags)
2049
blk_inc_in_flight(blk);
2050
blk_wait_while_drained(blk);
2051
GRAPH_RDLOCK_GUARD();
2052
if (!blk_is_available(blk)) {2053
blk_dec_in_flight(blk);
2057
ret = bdrv_co_zone_append(blk_bs(blk), offset, qiov, flags);
2058
blk_dec_in_flight(blk);
2062
void blk_drain(BlockBackend *blk)
2064
BlockDriverState *bs = blk_bs(blk);
2065
GLOBAL_STATE_CODE();
2069
bdrv_drained_begin(bs);
2072
/* We may have -ENOMEDIUM completions in flight */
2073
AIO_WAIT_WHILE(blk_get_aio_context(blk),
2074
qatomic_read(&blk->in_flight) > 0);
2077
bdrv_drained_end(bs);
2082
void blk_drain_all(void)
2084
BlockBackend *blk = NULL;
2086
GLOBAL_STATE_CODE();
2088
bdrv_drain_all_begin();
2090
while ((blk = blk_all_next(blk)) != NULL) {2091
/* We may have -ENOMEDIUM completions in flight */
2092
AIO_WAIT_WHILE_UNLOCKED(NULL, qatomic_read(&blk->in_flight) > 0);
2095
bdrv_drain_all_end();
2098
void blk_set_on_error(BlockBackend *blk, BlockdevOnError on_read_error,
2099
BlockdevOnError on_write_error)
2101
GLOBAL_STATE_CODE();
2102
blk->on_read_error = on_read_error;
2103
blk->on_write_error = on_write_error;
2106
BlockdevOnError blk_get_on_error(BlockBackend *blk, bool is_read)
2109
return is_read ? blk->on_read_error : blk->on_write_error;
2112
BlockErrorAction blk_get_error_action(BlockBackend *blk, bool is_read,
2115
BlockdevOnError on_err = blk_get_on_error(blk, is_read);
2119
case BLOCKDEV_ON_ERROR_ENOSPC:
2120
return (error == ENOSPC) ?
2121
BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
2122
case BLOCKDEV_ON_ERROR_STOP:
2123
return BLOCK_ERROR_ACTION_STOP;
2124
case BLOCKDEV_ON_ERROR_REPORT:
2125
return BLOCK_ERROR_ACTION_REPORT;
2126
case BLOCKDEV_ON_ERROR_IGNORE:
2127
return BLOCK_ERROR_ACTION_IGNORE;
2128
case BLOCKDEV_ON_ERROR_AUTO:
2134
static void send_qmp_error_event(BlockBackend *blk,
2135
BlockErrorAction action,
2136
bool is_read, int error)
2138
IoOperationType optype;
2139
BlockDriverState *bs = blk_bs(blk);
2141
optype = is_read ? IO_OPERATION_TYPE_READ : IO_OPERATION_TYPE_WRITE;
2142
qapi_event_send_block_io_error(blk_name(blk),
2143
bs ? bdrv_get_node_name(bs) : NULL, optype,
2144
action, blk_iostatus_is_enabled(blk),
2145
error == ENOSPC, strerror(error));
2148
/* This is done by device models because, while the block layer knows
2149
* about the error, it does not know whether an operation comes from
2150
* the device or the block layer (from a job, for example).
2152
void blk_error_action(BlockBackend *blk, BlockErrorAction action,
2153
bool is_read, int error)
2158
if (action == BLOCK_ERROR_ACTION_STOP) {2159
/* First set the iostatus, so that "info block" returns an iostatus
2160
* that matches the events raised so far (an additional error iostatus
2161
* is fine, but not a lost one).
2163
blk_iostatus_set_err(blk, error);
2165
/* Then raise the request to stop the VM and the event.
2166
* qemu_system_vmstop_request_prepare has two effects. First,
2167
* it ensures that the STOP event always comes after the
2168
* BLOCK_IO_ERROR event. Second, it ensures that even if management
2169
* can observe the STOP event and do a "cont" before the STOP
2170
* event is issued, the VM will not stop. In this case, vm_start()
2171
* also ensures that the STOP/RESUME pair of events is emitted.
2173
qemu_system_vmstop_request_prepare();
2174
send_qmp_error_event(blk, action, is_read, error);
2175
qemu_system_vmstop_request(RUN_STATE_IO_ERROR);
2177
send_qmp_error_event(blk, action, is_read, error);
2182
* Returns true if the BlockBackend can support taking write permissions
2183
* (because its root node is not read-only).
2185
bool blk_supports_write_perm(BlockBackend *blk)
2187
BlockDriverState *bs = blk_bs(blk);
2188
GLOBAL_STATE_CODE();
2191
return !bdrv_is_read_only(bs);
2193
return blk->root_state.open_flags & BDRV_O_RDWR;
2198
* Returns true if the BlockBackend can be written to in its current
2199
* configuration (i.e. if write permission have been requested)
2201
bool blk_is_writable(BlockBackend *blk)
2204
return blk->perm & BLK_PERM_WRITE;
2207
bool blk_is_sg(BlockBackend *blk)
2209
BlockDriverState *bs = blk_bs(blk);
2210
GLOBAL_STATE_CODE();
2216
return bdrv_is_sg(bs);
2219
bool blk_enable_write_cache(BlockBackend *blk)
2222
return blk->enable_write_cache;
2225
void blk_set_enable_write_cache(BlockBackend *blk, bool wce)
2228
blk->enable_write_cache = wce;
2231
void blk_activate(BlockBackend *blk, Error **errp)
2233
BlockDriverState *bs = blk_bs(blk);
2234
GLOBAL_STATE_CODE();
2237
error_setg(errp, "Device '%s' has no medium", blk->name);
2242
* Migration code can call this function in coroutine context, so leave
2243
* coroutine context if necessary.
2245
if (qemu_in_coroutine()) {2246
bdrv_co_activate(bs, errp);
2248
GRAPH_RDLOCK_GUARD_MAINLOOP();
2249
bdrv_activate(bs, errp);
2253
bool coroutine_fn blk_co_is_inserted(BlockBackend *blk)
2255
BlockDriverState *bs = blk_bs(blk);
2257
assert_bdrv_graph_readable();
2259
return bs && bdrv_co_is_inserted(bs);
2262
bool coroutine_fn blk_co_is_available(BlockBackend *blk)
2265
return blk_co_is_inserted(blk) && !blk_dev_is_tray_open(blk);
2268
void coroutine_fn blk_co_lock_medium(BlockBackend *blk, bool locked)
2270
BlockDriverState *bs = blk_bs(blk);
2272
GRAPH_RDLOCK_GUARD();
2275
bdrv_co_lock_medium(bs, locked);
2279
void coroutine_fn blk_co_eject(BlockBackend *blk, bool eject_flag)
2281
BlockDriverState *bs = blk_bs(blk);
2284
GRAPH_RDLOCK_GUARD();
2287
bdrv_co_eject(bs, eject_flag);
2290
/* Whether or not we ejected on the backend,
2291
* the frontend experienced a tray event. */
2292
id = blk_get_attached_dev_id(blk);
2293
qapi_event_send_device_tray_moved(blk_name(blk), id,
2298
int blk_get_flags(BlockBackend *blk)
2300
BlockDriverState *bs = blk_bs(blk);
2301
GLOBAL_STATE_CODE();
2304
return bdrv_get_flags(bs);
2306
return blk->root_state.open_flags;
2310
/* Returns the minimum request alignment, in bytes; guaranteed nonzero */
2311
uint32_t blk_get_request_alignment(BlockBackend *blk)
2313
BlockDriverState *bs = blk_bs(blk);
2315
return bs ? bs->bl.request_alignment : BDRV_SECTOR_SIZE;
2318
/* Returns the maximum hardware transfer length, in bytes; guaranteed nonzero */
2319
uint64_t blk_get_max_hw_transfer(BlockBackend *blk)
2321
BlockDriverState *bs = blk_bs(blk);
2322
uint64_t max = INT_MAX;
2326
max = MIN_NON_ZERO(max, bs->bl.max_hw_transfer);
2327
max = MIN_NON_ZERO(max, bs->bl.max_transfer);
2329
return ROUND_DOWN(max, blk_get_request_alignment(blk));
2332
/* Returns the maximum transfer length, in bytes; guaranteed nonzero */
2333
uint32_t blk_get_max_transfer(BlockBackend *blk)
2335
BlockDriverState *bs = blk_bs(blk);
2336
uint32_t max = INT_MAX;
2340
max = MIN_NON_ZERO(max, bs->bl.max_transfer);
2342
return ROUND_DOWN(max, blk_get_request_alignment(blk));
2345
int blk_get_max_hw_iov(BlockBackend *blk)
2348
return MIN_NON_ZERO(blk->root->bs->bl.max_hw_iov,
2349
blk->root->bs->bl.max_iov);
2352
int blk_get_max_iov(BlockBackend *blk)
2355
return blk->root->bs->bl.max_iov;
2358
void *blk_try_blockalign(BlockBackend *blk, size_t size)
2361
return qemu_try_blockalign(blk ? blk_bs(blk) : NULL, size);
2364
void *blk_blockalign(BlockBackend *blk, size_t size)
2367
return qemu_blockalign(blk ? blk_bs(blk) : NULL, size);
2370
bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp)
2372
BlockDriverState *bs = blk_bs(blk);
2373
GLOBAL_STATE_CODE();
2374
GRAPH_RDLOCK_GUARD_MAINLOOP();
2380
return bdrv_op_is_blocked(bs, op, errp);
2383
void blk_op_unblock(BlockBackend *blk, BlockOpType op, Error *reason)
2385
BlockDriverState *bs = blk_bs(blk);
2386
GLOBAL_STATE_CODE();
2389
bdrv_op_unblock(bs, op, reason);
2393
void blk_op_block_all(BlockBackend *blk, Error *reason)
2395
BlockDriverState *bs = blk_bs(blk);
2396
GLOBAL_STATE_CODE();
2399
bdrv_op_block_all(bs, reason);
2403
void blk_op_unblock_all(BlockBackend *blk, Error *reason)
2405
BlockDriverState *bs = blk_bs(blk);
2406
GLOBAL_STATE_CODE();
2409
bdrv_op_unblock_all(bs, reason);
2414
* Return BB's current AioContext. Note that this context may change
2415
* concurrently at any time, with one exception: If the BB has a root node
2416
* attached, its context will only change through bdrv_try_change_aio_context(),
2417
* which creates a drained section. Therefore, incrementing such a BB's
2418
* in-flight counter will prevent its context from changing.
2420
AioContext *blk_get_aio_context(BlockBackend *blk)
2425
return qemu_get_aio_context();
2428
return qatomic_read(&blk->ctx);
2431
int blk_set_aio_context(BlockBackend *blk, AioContext *new_context,
2434
bool old_allow_change;
2435
BlockDriverState *bs = blk_bs(blk);
2438
GLOBAL_STATE_CODE();
2441
qatomic_set(&blk->ctx, new_context);
2447
old_allow_change = blk->allow_aio_context_change;
2448
blk->allow_aio_context_change = true;
2450
ret = bdrv_try_change_aio_context(bs, new_context, NULL, errp);
2452
blk->allow_aio_context_change = old_allow_change;
2458
typedef struct BdrvStateBlkRootContext {2459
AioContext *new_ctx;
2461
} BdrvStateBlkRootContext;
2463
static void blk_root_set_aio_ctx_commit(void *opaque)
2465
BdrvStateBlkRootContext *s = opaque;
2466
BlockBackend *blk = s->blk;
2467
AioContext *new_context = s->new_ctx;
2468
ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
2470
qatomic_set(&blk->ctx, new_context);
2471
if (tgm->throttle_state) {2472
throttle_group_detach_aio_context(tgm);
2473
throttle_group_attach_aio_context(tgm, new_context);
2477
static TransactionActionDrv set_blk_root_context = {2478
.commit = blk_root_set_aio_ctx_commit,
2482
static bool blk_root_change_aio_ctx(BdrvChild *child, AioContext *ctx,
2483
GHashTable *visited, Transaction *tran,
2486
BlockBackend *blk = child->opaque;
2487
BdrvStateBlkRootContext *s;
2489
if (!blk->allow_aio_context_change) {2491
* Manually created BlockBackends (those with a name) that are not
2492
* attached to anything can change their AioContext without updating
2493
* their user; return an error for others.
2495
if (!blk->name || blk->dev) {2496
/* TODO Add BB name/QOM path */
2497
error_setg(errp, "Cannot change iothread of active block backend");
2502
s = g_new(BdrvStateBlkRootContext, 1);
2503
*s = (BdrvStateBlkRootContext) {2508
tran_add(tran, &set_blk_root_context, s);
2512
void blk_add_aio_context_notifier(BlockBackend *blk,
2513
void (*attached_aio_context)(AioContext *new_context, void *opaque),
2514
void (*detach_aio_context)(void *opaque), void *opaque)
2516
BlockBackendAioNotifier *notifier;
2517
BlockDriverState *bs = blk_bs(blk);
2518
GLOBAL_STATE_CODE();
2520
notifier = g_new(BlockBackendAioNotifier, 1);
2521
notifier->attached_aio_context = attached_aio_context;
2522
notifier->detach_aio_context = detach_aio_context;
2523
notifier->opaque = opaque;
2524
QLIST_INSERT_HEAD(&blk->aio_notifiers, notifier, list);
2527
bdrv_add_aio_context_notifier(bs, attached_aio_context,
2528
detach_aio_context, opaque);
2532
void blk_remove_aio_context_notifier(BlockBackend *blk,
2533
void (*attached_aio_context)(AioContext *,
2535
void (*detach_aio_context)(void *),
2538
BlockBackendAioNotifier *notifier;
2539
BlockDriverState *bs = blk_bs(blk);
2541
GLOBAL_STATE_CODE();
2544
bdrv_remove_aio_context_notifier(bs, attached_aio_context,
2545
detach_aio_context, opaque);
2548
QLIST_FOREACH(notifier, &blk->aio_notifiers, list) {2549
if (notifier->attached_aio_context == attached_aio_context &&
2550
notifier->detach_aio_context == detach_aio_context &&
2551
notifier->opaque == opaque) {2552
QLIST_REMOVE(notifier, list);
2561
void blk_add_remove_bs_notifier(BlockBackend *blk, Notifier *notify)
2563
GLOBAL_STATE_CODE();
2564
notifier_list_add(&blk->remove_bs_notifiers, notify);
2567
void blk_add_insert_bs_notifier(BlockBackend *blk, Notifier *notify)
2569
GLOBAL_STATE_CODE();
2570
notifier_list_add(&blk->insert_bs_notifiers, notify);
2573
BlockAcctStats *blk_get_stats(BlockBackend *blk)
2579
void *blk_aio_get(const AIOCBInfo *aiocb_info, BlockBackend *blk,
2580
BlockCompletionFunc *cb, void *opaque)
2583
return qemu_aio_get(aiocb_info, blk_bs(blk), cb, opaque);
2586
int coroutine_fn blk_co_pwrite_zeroes(BlockBackend *blk, int64_t offset,
2587
int64_t bytes, BdrvRequestFlags flags)
2590
return blk_co_pwritev(blk, offset, bytes, NULL,
2591
flags | BDRV_REQ_ZERO_WRITE);
2594
int coroutine_fn blk_co_pwrite_compressed(BlockBackend *blk, int64_t offset,
2595
int64_t bytes, const void *buf)
2597
QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
2599
return blk_co_pwritev_part(blk, offset, bytes, &qiov, 0,
2600
BDRV_REQ_WRITE_COMPRESSED);
2603
int coroutine_fn blk_co_truncate(BlockBackend *blk, int64_t offset, bool exact,
2604
PreallocMode prealloc, BdrvRequestFlags flags,
2608
GRAPH_RDLOCK_GUARD();
2609
if (!blk_co_is_available(blk)) {2610
error_setg(errp, "No medium inserted");
2614
return bdrv_co_truncate(blk->root, offset, exact, prealloc, flags, errp);
2617
int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
2618
int64_t pos, int size)
2621
GLOBAL_STATE_CODE();
2623
if (!blk_is_available(blk)) {2627
ret = bdrv_save_vmstate(blk_bs(blk), buf, pos, size);
2632
if (ret == size && !blk->enable_write_cache) {2633
ret = bdrv_flush(blk_bs(blk));
2636
return ret < 0 ? ret : size;
2639
int blk_load_vmstate(BlockBackend *blk, uint8_t *buf, int64_t pos, int size)
2641
GLOBAL_STATE_CODE();
2642
if (!blk_is_available(blk)) {2646
return bdrv_load_vmstate(blk_bs(blk), buf, pos, size);
2649
int blk_probe_blocksizes(BlockBackend *blk, BlockSizes *bsz)
2651
GLOBAL_STATE_CODE();
2652
GRAPH_RDLOCK_GUARD_MAINLOOP();
2654
if (!blk_is_available(blk)) {2658
return bdrv_probe_blocksizes(blk_bs(blk), bsz);
2661
int blk_probe_geometry(BlockBackend *blk, HDGeometry *geo)
2663
GLOBAL_STATE_CODE();
2664
if (!blk_is_available(blk)) {2668
return bdrv_probe_geometry(blk_bs(blk), geo);
2672
* Updates the BlockBackendRootState object with data from the currently
2673
* attached BlockDriverState.
2675
void blk_update_root_state(BlockBackend *blk)
2677
GLOBAL_STATE_CODE();
2680
blk->root_state.open_flags = blk->root->bs->open_flags;
2681
blk->root_state.detect_zeroes = blk->root->bs->detect_zeroes;
2685
* Returns the detect-zeroes setting to be used for bdrv_open() of a
2686
* BlockDriverState which is supposed to inherit the root state.
2688
bool blk_get_detect_zeroes_from_root_state(BlockBackend *blk)
2690
GLOBAL_STATE_CODE();
2691
return blk->root_state.detect_zeroes;
2695
* Returns the flags to be used for bdrv_open() of a BlockDriverState which is
2696
* supposed to inherit the root state.
2698
int blk_get_open_flags_from_root_state(BlockBackend *blk)
2700
GLOBAL_STATE_CODE();
2701
return blk->root_state.open_flags;
2704
BlockBackendRootState *blk_get_root_state(BlockBackend *blk)
2706
GLOBAL_STATE_CODE();
2707
return &blk->root_state;
2710
int blk_commit_all(void)
2712
BlockBackend *blk = NULL;
2713
GLOBAL_STATE_CODE();
2714
GRAPH_RDLOCK_GUARD_MAINLOOP();
2716
while ((blk = blk_all_next(blk)) != NULL) {2717
BlockDriverState *unfiltered_bs = bdrv_skip_filters(blk_bs(blk));
2719
if (blk_is_inserted(blk) && bdrv_cow_child(unfiltered_bs)) {2722
ret = bdrv_commit(unfiltered_bs);
2732
/* throttling disk I/O limits */
2733
void blk_set_io_limits(BlockBackend *blk, ThrottleConfig *cfg)
2735
GLOBAL_STATE_CODE();
2736
throttle_group_config(&blk->public.throttle_group_member, cfg);
2739
void blk_io_limits_disable(BlockBackend *blk)
2741
BlockDriverState *bs = blk_bs(blk);
2742
ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
2743
assert(tgm->throttle_state);
2744
GLOBAL_STATE_CODE();
2747
bdrv_drained_begin(bs);
2749
throttle_group_unregister_tgm(tgm);
2751
bdrv_drained_end(bs);
2756
/* should be called before blk_set_io_limits if a limit is set */
2757
void blk_io_limits_enable(BlockBackend *blk, const char *group)
2759
assert(!blk->public.throttle_group_member.throttle_state);
2760
GLOBAL_STATE_CODE();
2761
throttle_group_register_tgm(&blk->public.throttle_group_member,
2762
group, blk_get_aio_context(blk));
2765
void blk_io_limits_update_group(BlockBackend *blk, const char *group)
2767
GLOBAL_STATE_CODE();
2768
/* this BB is not part of any group */
2769
if (!blk->public.throttle_group_member.throttle_state) {2773
/* this BB is a part of the same group than the one we want */
2774
if (!g_strcmp0(throttle_group_get_name(&blk->public.throttle_group_member),
2779
/* need to change the group this bs belong to */
2780
blk_io_limits_disable(blk);
2781
blk_io_limits_enable(blk, group);
2784
static void blk_root_drained_begin(BdrvChild *child)
2786
BlockBackend *blk = child->opaque;
2787
ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
2789
if (qatomic_fetch_inc(&blk->quiesce_counter) == 0) {2790
if (blk->dev_ops && blk->dev_ops->drained_begin) {2791
blk->dev_ops->drained_begin(blk->dev_opaque);
2795
/* Note that blk->root may not be accessible here yet if we are just
2796
* attaching to a BlockDriverState that is drained. Use child instead. */
2798
if (qatomic_fetch_inc(&tgm->io_limits_disabled) == 0) {2799
throttle_group_restart_tgm(tgm);
2803
static bool blk_root_drained_poll(BdrvChild *child)
2805
BlockBackend *blk = child->opaque;
2807
assert(qatomic_read(&blk->quiesce_counter));
2809
if (blk->dev_ops && blk->dev_ops->drained_poll) {2810
busy = blk->dev_ops->drained_poll(blk->dev_opaque);
2812
return busy || !!blk->in_flight;
2815
static void blk_root_drained_end(BdrvChild *child)
2817
BlockBackend *blk = child->opaque;
2818
assert(qatomic_read(&blk->quiesce_counter));
2820
assert(blk->public.throttle_group_member.io_limits_disabled);
2821
qatomic_dec(&blk->public.throttle_group_member.io_limits_disabled);
2823
if (qatomic_fetch_dec(&blk->quiesce_counter) == 1) {2824
if (blk->dev_ops && blk->dev_ops->drained_end) {2825
blk->dev_ops->drained_end(blk->dev_opaque);
2827
qemu_mutex_lock(&blk->queued_requests_lock);
2828
while (qemu_co_enter_next(&blk->queued_requests,
2829
&blk->queued_requests_lock)) {2830
/* Resume all queued requests */
2832
qemu_mutex_unlock(&blk->queued_requests_lock);
2836
bool blk_register_buf(BlockBackend *blk, void *host, size_t size, Error **errp)
2838
BlockDriverState *bs = blk_bs(blk);
2840
GLOBAL_STATE_CODE();
2843
return bdrv_register_buf(bs, host, size, errp);
2848
void blk_unregister_buf(BlockBackend *blk, void *host, size_t size)
2850
BlockDriverState *bs = blk_bs(blk);
2852
GLOBAL_STATE_CODE();
2855
bdrv_unregister_buf(bs, host, size);
2859
int coroutine_fn blk_co_copy_range(BlockBackend *blk_in, int64_t off_in,
2860
BlockBackend *blk_out, int64_t off_out,
2861
int64_t bytes, BdrvRequestFlags read_flags,
2862
BdrvRequestFlags write_flags)
2866
GRAPH_RDLOCK_GUARD();
2868
r = blk_check_byte_request(blk_in, off_in, bytes);
2872
r = blk_check_byte_request(blk_out, off_out, bytes);
2877
return bdrv_co_copy_range(blk_in->root, off_in,
2878
blk_out->root, off_out,
2879
bytes, read_flags, write_flags);
2882
const BdrvChild *blk_root(BlockBackend *blk)
2884
GLOBAL_STATE_CODE();
2888
int blk_make_empty(BlockBackend *blk, Error **errp)
2890
GLOBAL_STATE_CODE();
2891
GRAPH_RDLOCK_GUARD_MAINLOOP();
2893
if (!blk_is_available(blk)) {2894
error_setg(errp, "No medium inserted");
2898
return bdrv_make_empty(blk->root, errp);