xref: /openbmc/qemu/block/block-backend.c (revision 6bc0bcc8)
1 /*
2  * QEMU Block backends
3  *
4  * Copyright (C) 2014-2016 Red Hat, Inc.
5  *
6  * Authors:
7  *  Markus Armbruster <armbru@redhat.com>,
8  *
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.
11  */
12 
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"
25 #include "qemu/id.h"
26 #include "qemu/main-loop.h"
27 #include "qemu/option.h"
28 #include "trace.h"
29 #include "migration/misc.h"
30 
31 /* Number of coroutines to reserve per attached device model */
32 #define COROUTINE_POOL_RESERVATION 64
33 
34 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
35 
36 typedef struct BlockBackendAioNotifier {
37     void (*attached_aio_context)(AioContext *new_context, void *opaque);
38     void (*detach_aio_context)(void *opaque);
39     void *opaque;
40     QLIST_ENTRY(BlockBackendAioNotifier) list;
41 } BlockBackendAioNotifier;
42 
43 struct BlockBackend {
44     char *name;
45     int refcnt;
46     BdrvChild *root;
47     AioContext *ctx;
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;
52 
53     DeviceState *dev;           /* attached device model, if any */
54     const BlockDevOps *dev_ops;
55     void *dev_opaque;
56 
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;
60 
61     bool enable_write_cache;
62 
63     /* I/O stats (display with "info blockstats"). */
64     BlockAcctStats stats;
65 
66     BlockdevOnError on_read_error, on_write_error;
67     bool iostatus_enabled;
68     BlockDeviceIoStatus iostatus;
69 
70     uint64_t perm;
71     uint64_t shared_perm;
72     bool disable_perm;
73 
74     bool allow_aio_context_change;
75     bool allow_write_beyond_eof;
76 
77     /* Protected by BQL */
78     NotifierList remove_bs_notifiers, insert_bs_notifiers;
79     QLIST_HEAD(, BlockBackendAioNotifier) aio_notifiers;
80 
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 */
85 
86     VMChangeStateEntry *vmsh;
87     bool force_allow_inactivate;
88 
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.
93      */
94     unsigned int in_flight;
95 };
96 
97 typedef struct BlockBackendAIOCB {
98     BlockAIOCB common;
99     BlockBackend *blk;
100     int ret;
101 } BlockBackendAIOCB;
102 
103 static const AIOCBInfo block_backend_aiocb_info = {
104     .aiocb_size = sizeof(BlockBackendAIOCB),
105 };
106 
107 static void drive_info_del(DriveInfo *dinfo);
108 static BlockBackend *bdrv_first_blk(BlockDriverState *bs);
109 
110 /* All BlockBackends. Protected by BQL. */
111 static QTAILQ_HEAD(, BlockBackend) block_backends =
112     QTAILQ_HEAD_INITIALIZER(block_backends);
113 
114 /*
115  * All BlockBackends referenced by the monitor and which are iterated through by
116  * blk_next(). Protected by BQL.
117  */
118 static QTAILQ_HEAD(, BlockBackend) monitor_block_backends =
119     QTAILQ_HEAD_INITIALIZER(monitor_block_backends);
120 
121 static int coroutine_mixed_fn GRAPH_RDLOCK
122 blk_set_perm_locked(BlockBackend *blk, uint64_t perm, uint64_t shared_perm,
123                     Error **errp);
124 
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)
128 {
129     /* We're not supposed to call this function for root nodes */
130     abort();
131 }
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);
135 
136 static void blk_root_change_media(BdrvChild *child, bool load);
137 static void blk_root_resize(BdrvChild *child);
138 
139 static bool blk_root_change_aio_ctx(BdrvChild *child, AioContext *ctx,
140                                     GHashTable *visited, Transaction *tran,
141                                     Error **errp);
142 
143 static char *blk_root_get_parent_desc(BdrvChild *child)
144 {
145     BlockBackend *blk = child->opaque;
146     g_autofree char *dev_id = NULL;
147 
148     if (blk->name) {
149         return g_strdup_printf("block device '%s'", blk->name);
150     }
151 
152     dev_id = blk_get_attached_dev_id(blk);
153     if (*dev_id) {
154         return g_strdup_printf("block device '%s'", dev_id);
155     } else {
156         /* TODO Callback into the BB owner for something more detailed */
157         return g_strdup("an unnamed block device");
158     }
159 }
160 
161 static const char *blk_root_get_name(BdrvChild *child)
162 {
163     return blk_name(child->opaque);
164 }
165 
166 static void blk_vm_state_changed(void *opaque, bool running, RunState state)
167 {
168     Error *local_err = NULL;
169     BlockBackend *blk = opaque;
170 
171     if (state == RUN_STATE_INMIGRATE) {
172         return;
173     }
174 
175     qemu_del_vm_change_state_handler(blk->vmsh);
176     blk->vmsh = NULL;
177     blk_set_perm(blk, blk->perm, blk->shared_perm, &local_err);
178     if (local_err) {
179         error_report_err(local_err);
180     }
181 }
182 
183 /*
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).
187  *
188  * If an error is returned, the VM cannot be allowed to be resumed.
189  */
190 static void GRAPH_RDLOCK blk_root_activate(BdrvChild *child, Error **errp)
191 {
192     BlockBackend *blk = child->opaque;
193     Error *local_err = NULL;
194     uint64_t saved_shared_perm;
195 
196     if (!blk->disable_perm) {
197         return;
198     }
199 
200     blk->disable_perm = false;
201 
202     /*
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
207      * restore it below.
208      */
209     saved_shared_perm = blk->shared_perm;
210 
211     blk_set_perm_locked(blk, blk->perm, BLK_PERM_ALL, &local_err);
212     if (local_err) {
213         error_propagate(errp, local_err);
214         blk->disable_perm = true;
215         return;
216     }
217     blk->shared_perm = saved_shared_perm;
218 
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. */
223         if (!blk->vmsh) {
224             blk->vmsh = qemu_add_vm_change_state_handler(blk_vm_state_changed,
225                                                          blk);
226         }
227         return;
228     }
229 
230     blk_set_perm_locked(blk, blk->perm, blk->shared_perm, &local_err);
231     if (local_err) {
232         error_propagate(errp, local_err);
233         blk->disable_perm = true;
234         return;
235     }
236 }
237 
238 void blk_set_force_allow_inactivate(BlockBackend *blk)
239 {
240     GLOBAL_STATE_CODE();
241     blk->force_allow_inactivate = true;
242 }
243 
244 static bool blk_can_inactivate(BlockBackend *blk)
245 {
246     /* If it is a guest device, inactivate is ok. */
247     if (blk->dev || blk_name(blk)[0]) {
248         return true;
249     }
250 
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))) {
257         return true;
258     }
259 
260     return blk->force_allow_inactivate;
261 }
262 
263 static int GRAPH_RDLOCK blk_root_inactivate(BdrvChild *child)
264 {
265     BlockBackend *blk = child->opaque;
266 
267     if (blk->disable_perm) {
268         return 0;
269     }
270 
271     if (!blk_can_inactivate(blk)) {
272         return -EPERM;
273     }
274 
275     blk->disable_perm = true;
276     if (blk->root) {
277         bdrv_child_try_set_perm(blk->root, 0, BLK_PERM_ALL, &error_abort);
278     }
279 
280     return 0;
281 }
282 
283 static void blk_root_attach(BdrvChild *child)
284 {
285     BlockBackend *blk = child->opaque;
286     BlockBackendAioNotifier *notifier;
287 
288     trace_blk_root_attach(child, blk, child->bs);
289 
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,
294                 notifier->opaque);
295     }
296 }
297 
298 static void blk_root_detach(BdrvChild *child)
299 {
300     BlockBackend *blk = child->opaque;
301     BlockBackendAioNotifier *notifier;
302 
303     trace_blk_root_detach(child, blk, child->bs);
304 
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,
309                 notifier->opaque);
310     }
311 }
312 
313 static AioContext *blk_root_get_parent_aio_context(BdrvChild *c)
314 {
315     BlockBackend *blk = c->opaque;
316     IO_CODE();
317 
318     return blk_get_aio_context(blk);
319 }
320 
321 static const BdrvChildClass child_root = {
322     .inherit_options    = blk_root_inherit_options,
323 
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,
328 
329     .drained_begin      = blk_root_drained_begin,
330     .drained_poll       = blk_root_drained_poll,
331     .drained_end        = blk_root_drained_end,
332 
333     .activate           = blk_root_activate,
334     .inactivate         = blk_root_inactivate,
335 
336     .attach             = blk_root_attach,
337     .detach             = blk_root_detach,
338 
339     .change_aio_ctx     = blk_root_change_aio_ctx,
340 
341     .get_parent_aio_context = blk_root_get_parent_aio_context,
342 };
343 
344 /*
345  * Create a new BlockBackend with a reference count of one.
346  *
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().
352  *
353  * Return the new BlockBackend on success, null on failure.
354  */
355 BlockBackend *blk_new(AioContext *ctx, uint64_t perm, uint64_t shared_perm)
356 {
357     BlockBackend *blk;
358 
359     GLOBAL_STATE_CODE();
360 
361     blk = g_new0(BlockBackend, 1);
362     blk->refcnt = 1;
363     blk->ctx = ctx;
364     blk->perm = perm;
365     blk->shared_perm = shared_perm;
366     blk_set_enable_write_cache(blk, true);
367 
368     blk->on_read_error = BLOCKDEV_ON_ERROR_REPORT;
369     blk->on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
370 
371     block_acct_init(&blk->stats);
372 
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);
378 
379     QTAILQ_INSERT_TAIL(&block_backends, blk, link);
380     return blk;
381 }
382 
383 /*
384  * Create a new BlockBackend connected to an existing BlockDriverState.
385  *
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().
391  *
392  * Return the new BlockBackend on success, null on failure.
393  *
394  * Callers must hold the AioContext lock of @bs.
395  */
396 BlockBackend *blk_new_with_bs(BlockDriverState *bs, uint64_t perm,
397                               uint64_t shared_perm, Error **errp)
398 {
399     BlockBackend *blk = blk_new(bdrv_get_aio_context(bs), perm, shared_perm);
400 
401     GLOBAL_STATE_CODE();
402 
403     if (blk_insert_bs(blk, bs, errp) < 0) {
404         blk_unref(blk);
405         return NULL;
406     }
407     return blk;
408 }
409 
410 /*
411  * Creates a new BlockBackend, opens a new BlockDriverState, and connects both.
412  * By default, the new BlockBackend is in the main AioContext, but if the
413  * parameters connect it with any existing node in a different AioContext, it
414  * may end up there instead.
415  *
416  * Just as with bdrv_open(), after having called this function the reference to
417  * @options belongs to the block layer (even on failure).
418  *
419  * Called without holding an AioContext lock.
420  *
421  * TODO: Remove @filename and @flags; it should be possible to specify a whole
422  * BDS tree just by specifying the @options QDict (or @reference,
423  * alternatively). At the time of adding this function, this is not possible,
424  * though, so callers of this function have to be able to specify @filename and
425  * @flags.
426  */
427 BlockBackend *blk_new_open(const char *filename, const char *reference,
428                            QDict *options, int flags, Error **errp)
429 {
430     BlockBackend *blk;
431     BlockDriverState *bs;
432     AioContext *ctx;
433     uint64_t perm = 0;
434     uint64_t shared = BLK_PERM_ALL;
435 
436     GLOBAL_STATE_CODE();
437 
438     /*
439      * blk_new_open() is mainly used in .bdrv_create implementations and the
440      * tools where sharing isn't a major concern because the BDS stays private
441      * and the file is generally not supposed to be used by a second process,
442      * so we just request permission according to the flags.
443      *
444      * The exceptions are xen_disk and blockdev_init(); in these cases, the
445      * caller of blk_new_open() doesn't make use of the permissions, but they
446      * shouldn't hurt either. We can still share everything here because the
447      * guest devices will add their own blockers if they can't share.
448      */
449     if ((flags & BDRV_O_NO_IO) == 0) {
450         perm |= BLK_PERM_CONSISTENT_READ;
451         if (flags & BDRV_O_RDWR) {
452             perm |= BLK_PERM_WRITE;
453         }
454     }
455     if (flags & BDRV_O_RESIZE) {
456         perm |= BLK_PERM_RESIZE;
457     }
458     if (flags & BDRV_O_NO_SHARE) {
459         shared = BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
460     }
461 
462     aio_context_acquire(qemu_get_aio_context());
463     bs = bdrv_open(filename, reference, options, flags, errp);
464     aio_context_release(qemu_get_aio_context());
465     if (!bs) {
466         return NULL;
467     }
468 
469     /* bdrv_open() could have moved bs to a different AioContext */
470     ctx = bdrv_get_aio_context(bs);
471     blk = blk_new(bdrv_get_aio_context(bs), perm, shared);
472     blk->perm = perm;
473     blk->shared_perm = shared;
474 
475     aio_context_acquire(ctx);
476     blk_insert_bs(blk, bs, errp);
477     bdrv_unref(bs);
478     aio_context_release(ctx);
479 
480     if (!blk->root) {
481         blk_unref(blk);
482         return NULL;
483     }
484 
485     return blk;
486 }
487 
488 static void blk_delete(BlockBackend *blk)
489 {
490     assert(!blk->refcnt);
491     assert(!blk->name);
492     assert(!blk->dev);
493     if (blk->public.throttle_group_member.throttle_state) {
494         blk_io_limits_disable(blk);
495     }
496     if (blk->root) {
497         blk_remove_bs(blk);
498     }
499     if (blk->vmsh) {
500         qemu_del_vm_change_state_handler(blk->vmsh);
501         blk->vmsh = NULL;
502     }
503     assert(QLIST_EMPTY(&blk->remove_bs_notifiers.notifiers));
504     assert(QLIST_EMPTY(&blk->insert_bs_notifiers.notifiers));
505     assert(QLIST_EMPTY(&blk->aio_notifiers));
506     assert(qemu_co_queue_empty(&blk->queued_requests));
507     qemu_mutex_destroy(&blk->queued_requests_lock);
508     QTAILQ_REMOVE(&block_backends, blk, link);
509     drive_info_del(blk->legacy_dinfo);
510     block_acct_cleanup(&blk->stats);
511     g_free(blk);
512 }
513 
514 static void drive_info_del(DriveInfo *dinfo)
515 {
516     if (!dinfo) {
517         return;
518     }
519     qemu_opts_del(dinfo->opts);
520     g_free(dinfo);
521 }
522 
523 int blk_get_refcnt(BlockBackend *blk)
524 {
525     GLOBAL_STATE_CODE();
526     return blk ? blk->refcnt : 0;
527 }
528 
529 /*
530  * Increment @blk's reference count.
531  * @blk must not be null.
532  */
533 void blk_ref(BlockBackend *blk)
534 {
535     assert(blk->refcnt > 0);
536     GLOBAL_STATE_CODE();
537     blk->refcnt++;
538 }
539 
540 /*
541  * Decrement @blk's reference count.
542  * If this drops it to zero, destroy @blk.
543  * For convenience, do nothing if @blk is null.
544  */
545 void blk_unref(BlockBackend *blk)
546 {
547     GLOBAL_STATE_CODE();
548     if (blk) {
549         assert(blk->refcnt > 0);
550         if (blk->refcnt > 1) {
551             blk->refcnt--;
552         } else {
553             blk_drain(blk);
554             /* blk_drain() cannot resurrect blk, nobody held a reference */
555             assert(blk->refcnt == 1);
556             blk->refcnt = 0;
557             blk_delete(blk);
558         }
559     }
560 }
561 
562 /*
563  * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
564  * ones which are hidden (i.e. are not referenced by the monitor).
565  */
566 BlockBackend *blk_all_next(BlockBackend *blk)
567 {
568     GLOBAL_STATE_CODE();
569     return blk ? QTAILQ_NEXT(blk, link)
570                : QTAILQ_FIRST(&block_backends);
571 }
572 
573 void blk_remove_all_bs(void)
574 {
575     BlockBackend *blk = NULL;
576 
577     GLOBAL_STATE_CODE();
578 
579     while ((blk = blk_all_next(blk)) != NULL) {
580         AioContext *ctx = blk_get_aio_context(blk);
581 
582         aio_context_acquire(ctx);
583         if (blk->root) {
584             blk_remove_bs(blk);
585         }
586         aio_context_release(ctx);
587     }
588 }
589 
590 /*
591  * Return the monitor-owned BlockBackend after @blk.
592  * If @blk is null, return the first one.
593  * Else, return @blk's next sibling, which may be null.
594  *
595  * To iterate over all BlockBackends, do
596  * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
597  *     ...
598  * }
599  */
600 BlockBackend *blk_next(BlockBackend *blk)
601 {
602     GLOBAL_STATE_CODE();
603     return blk ? QTAILQ_NEXT(blk, monitor_link)
604                : QTAILQ_FIRST(&monitor_block_backends);
605 }
606 
607 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
608  * the monitor or attached to a BlockBackend */
609 BlockDriverState *bdrv_next(BdrvNextIterator *it)
610 {
611     BlockDriverState *bs, *old_bs;
612 
613     /* Must be called from the main loop */
614     assert(qemu_get_current_aio_context() == qemu_get_aio_context());
615 
616     /* First, return all root nodes of BlockBackends. In order to avoid
617      * returning a BDS twice when multiple BBs refer to it, we only return it
618      * if the BB is the first one in the parent list of the BDS. */
619     if (it->phase == BDRV_NEXT_BACKEND_ROOTS) {
620         BlockBackend *old_blk = it->blk;
621 
622         old_bs = old_blk ? blk_bs(old_blk) : NULL;
623 
624         do {
625             it->blk = blk_all_next(it->blk);
626             bs = it->blk ? blk_bs(it->blk) : NULL;
627         } while (it->blk && (bs == NULL || bdrv_first_blk(bs) != it->blk));
628 
629         if (it->blk) {
630             blk_ref(it->blk);
631         }
632         blk_unref(old_blk);
633 
634         if (bs) {
635             bdrv_ref(bs);
636             bdrv_unref(old_bs);
637             return bs;
638         }
639         it->phase = BDRV_NEXT_MONITOR_OWNED;
640     } else {
641         old_bs = it->bs;
642     }
643 
644     /* Then return the monitor-owned BDSes without a BB attached. Ignore all
645      * BDSes that are attached to a BlockBackend here; they have been handled
646      * by the above block already */
647     do {
648         it->bs = bdrv_next_monitor_owned(it->bs);
649         bs = it->bs;
650     } while (bs && bdrv_has_blk(bs));
651 
652     if (bs) {
653         bdrv_ref(bs);
654     }
655     bdrv_unref(old_bs);
656 
657     return bs;
658 }
659 
660 static void bdrv_next_reset(BdrvNextIterator *it)
661 {
662     *it = (BdrvNextIterator) {
663         .phase = BDRV_NEXT_BACKEND_ROOTS,
664     };
665 }
666 
667 BlockDriverState *bdrv_first(BdrvNextIterator *it)
668 {
669     GLOBAL_STATE_CODE();
670     bdrv_next_reset(it);
671     return bdrv_next(it);
672 }
673 
674 /* Must be called when aborting a bdrv_next() iteration before
675  * bdrv_next() returns NULL */
676 void bdrv_next_cleanup(BdrvNextIterator *it)
677 {
678     /* Must be called from the main loop */
679     assert(qemu_get_current_aio_context() == qemu_get_aio_context());
680 
681     if (it->phase == BDRV_NEXT_BACKEND_ROOTS) {
682         if (it->blk) {
683             bdrv_unref(blk_bs(it->blk));
684             blk_unref(it->blk);
685         }
686     } else {
687         bdrv_unref(it->bs);
688     }
689 
690     bdrv_next_reset(it);
691 }
692 
693 /*
694  * Add a BlockBackend into the list of backends referenced by the monitor, with
695  * the given @name acting as the handle for the monitor.
696  * Strictly for use by blockdev.c.
697  *
698  * @name must not be null or empty.
699  *
700  * Returns true on success and false on failure. In the latter case, an Error
701  * object is returned through @errp.
702  */
703 bool monitor_add_blk(BlockBackend *blk, const char *name, Error **errp)
704 {
705     assert(!blk->name);
706     assert(name && name[0]);
707     GLOBAL_STATE_CODE();
708 
709     if (!id_wellformed(name)) {
710         error_setg(errp, "Invalid device name");
711         return false;
712     }
713     if (blk_by_name(name)) {
714         error_setg(errp, "Device with id '%s' already exists", name);
715         return false;
716     }
717     if (bdrv_find_node(name)) {
718         error_setg(errp,
719                    "Device name '%s' conflicts with an existing node name",
720                    name);
721         return false;
722     }
723 
724     blk->name = g_strdup(name);
725     QTAILQ_INSERT_TAIL(&monitor_block_backends, blk, monitor_link);
726     return true;
727 }
728 
729 /*
730  * Remove a BlockBackend from the list of backends referenced by the monitor.
731  * Strictly for use by blockdev.c.
732  */
733 void monitor_remove_blk(BlockBackend *blk)
734 {
735     GLOBAL_STATE_CODE();
736 
737     if (!blk->name) {
738         return;
739     }
740 
741     QTAILQ_REMOVE(&monitor_block_backends, blk, monitor_link);
742     g_free(blk->name);
743     blk->name = NULL;
744 }
745 
746 /*
747  * Return @blk's name, a non-null string.
748  * Returns an empty string iff @blk is not referenced by the monitor.
749  */
750 const char *blk_name(const BlockBackend *blk)
751 {
752     IO_CODE();
753     return blk->name ?: "";
754 }
755 
756 /*
757  * Return the BlockBackend with name @name if it exists, else null.
758  * @name must not be null.
759  */
760 BlockBackend *blk_by_name(const char *name)
761 {
762     BlockBackend *blk = NULL;
763 
764     GLOBAL_STATE_CODE();
765     assert(name);
766     while ((blk = blk_next(blk)) != NULL) {
767         if (!strcmp(name, blk->name)) {
768             return blk;
769         }
770     }
771     return NULL;
772 }
773 
774 /*
775  * Return the BlockDriverState attached to @blk if any, else null.
776  */
777 BlockDriverState *blk_bs(BlockBackend *blk)
778 {
779     IO_CODE();
780     return blk->root ? blk->root->bs : NULL;
781 }
782 
783 static BlockBackend * GRAPH_RDLOCK bdrv_first_blk(BlockDriverState *bs)
784 {
785     BdrvChild *child;
786 
787     GLOBAL_STATE_CODE();
788     assert_bdrv_graph_readable();
789 
790     QLIST_FOREACH(child, &bs->parents, next_parent) {
791         if (child->klass == &child_root) {
792             return child->opaque;
793         }
794     }
795 
796     return NULL;
797 }
798 
799 /*
800  * Returns true if @bs has an associated BlockBackend.
801  */
802 bool bdrv_has_blk(BlockDriverState *bs)
803 {
804     GLOBAL_STATE_CODE();
805     return bdrv_first_blk(bs) != NULL;
806 }
807 
808 /*
809  * Returns true if @bs has only BlockBackends as parents.
810  */
811 bool bdrv_is_root_node(BlockDriverState *bs)
812 {
813     BdrvChild *c;
814 
815     GLOBAL_STATE_CODE();
816     assert_bdrv_graph_readable();
817 
818     QLIST_FOREACH(c, &bs->parents, next_parent) {
819         if (c->klass != &child_root) {
820             return false;
821         }
822     }
823 
824     return true;
825 }
826 
827 /*
828  * Return @blk's DriveInfo if any, else null.
829  */
830 DriveInfo *blk_legacy_dinfo(BlockBackend *blk)
831 {
832     GLOBAL_STATE_CODE();
833     return blk->legacy_dinfo;
834 }
835 
836 /*
837  * Set @blk's DriveInfo to @dinfo, and return it.
838  * @blk must not have a DriveInfo set already.
839  * No other BlockBackend may have the same DriveInfo set.
840  */
841 DriveInfo *blk_set_legacy_dinfo(BlockBackend *blk, DriveInfo *dinfo)
842 {
843     assert(!blk->legacy_dinfo);
844     GLOBAL_STATE_CODE();
845     return blk->legacy_dinfo = dinfo;
846 }
847 
848 /*
849  * Return the BlockBackend with DriveInfo @dinfo.
850  * It must exist.
851  */
852 BlockBackend *blk_by_legacy_dinfo(DriveInfo *dinfo)
853 {
854     BlockBackend *blk = NULL;
855     GLOBAL_STATE_CODE();
856 
857     while ((blk = blk_next(blk)) != NULL) {
858         if (blk->legacy_dinfo == dinfo) {
859             return blk;
860         }
861     }
862     abort();
863 }
864 
865 /*
866  * Returns a pointer to the publicly accessible fields of @blk.
867  */
868 BlockBackendPublic *blk_get_public(BlockBackend *blk)
869 {
870     GLOBAL_STATE_CODE();
871     return &blk->public;
872 }
873 
874 /*
875  * Returns a BlockBackend given the associated @public fields.
876  */
877 BlockBackend *blk_by_public(BlockBackendPublic *public)
878 {
879     GLOBAL_STATE_CODE();
880     return container_of(public, BlockBackend, public);
881 }
882 
883 /*
884  * Disassociates the currently associated BlockDriverState from @blk.
885  *
886  * The caller must hold the AioContext lock for the BlockBackend.
887  */
888 void blk_remove_bs(BlockBackend *blk)
889 {
890     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
891     BdrvChild *root;
892     AioContext *ctx;
893 
894     GLOBAL_STATE_CODE();
895 
896     notifier_list_notify(&blk->remove_bs_notifiers, blk);
897     if (tgm->throttle_state) {
898         BlockDriverState *bs = blk_bs(blk);
899 
900         /*
901          * Take a ref in case blk_bs() changes across bdrv_drained_begin(), for
902          * example, if a temporary filter node is removed by a blockjob.
903          */
904         bdrv_ref(bs);
905         bdrv_drained_begin(bs);
906         throttle_group_detach_aio_context(tgm);
907         throttle_group_attach_aio_context(tgm, qemu_get_aio_context());
908         bdrv_drained_end(bs);
909         bdrv_unref(bs);
910     }
911 
912     blk_update_root_state(blk);
913 
914     /* bdrv_root_unref_child() will cause blk->root to become stale and may
915      * switch to a completion coroutine later on. Let's drain all I/O here
916      * to avoid that and a potential QEMU crash.
917      */
918     blk_drain(blk);
919     root = blk->root;
920     blk->root = NULL;
921 
922     ctx = bdrv_get_aio_context(root->bs);
923     bdrv_graph_wrlock(root->bs);
924     bdrv_root_unref_child(root);
925     bdrv_graph_wrunlock_ctx(ctx);
926 }
927 
928 /*
929  * Associates a new BlockDriverState with @blk.
930  *
931  * Callers must hold the AioContext lock of @bs.
932  */
933 int blk_insert_bs(BlockBackend *blk, BlockDriverState *bs, Error **errp)
934 {
935     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
936     AioContext *ctx = bdrv_get_aio_context(bs);
937 
938     GLOBAL_STATE_CODE();
939     bdrv_ref(bs);
940     bdrv_graph_wrlock(bs);
941     blk->root = bdrv_root_attach_child(bs, "root", &child_root,
942                                        BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY,
943                                        blk->perm, blk->shared_perm,
944                                        blk, errp);
945     bdrv_graph_wrunlock_ctx(ctx);
946     if (blk->root == NULL) {
947         return -EPERM;
948     }
949 
950     notifier_list_notify(&blk->insert_bs_notifiers, blk);
951     if (tgm->throttle_state) {
952         throttle_group_detach_aio_context(tgm);
953         throttle_group_attach_aio_context(tgm, bdrv_get_aio_context(bs));
954     }
955 
956     return 0;
957 }
958 
959 /*
960  * Change BlockDriverState associated with @blk.
961  */
962 int blk_replace_bs(BlockBackend *blk, BlockDriverState *new_bs, Error **errp)
963 {
964     GLOBAL_STATE_CODE();
965     return bdrv_replace_child_bs(blk->root, new_bs, errp);
966 }
967 
968 /*
969  * Sets the permission bitmasks that the user of the BlockBackend needs.
970  */
971 static int coroutine_mixed_fn GRAPH_RDLOCK
972 blk_set_perm_locked(BlockBackend *blk, uint64_t perm, uint64_t shared_perm,
973                     Error **errp)
974 {
975     int ret;
976     GLOBAL_STATE_CODE();
977 
978     if (blk->root && !blk->disable_perm) {
979         ret = bdrv_child_try_set_perm(blk->root, perm, shared_perm, errp);
980         if (ret < 0) {
981             return ret;
982         }
983     }
984 
985     blk->perm = perm;
986     blk->shared_perm = shared_perm;
987 
988     return 0;
989 }
990 
991 int blk_set_perm(BlockBackend *blk, uint64_t perm, uint64_t shared_perm,
992                  Error **errp)
993 {
994     GLOBAL_STATE_CODE();
995     GRAPH_RDLOCK_GUARD_MAINLOOP();
996 
997     return blk_set_perm_locked(blk, perm, shared_perm, errp);
998 }
999 
1000 void blk_get_perm(BlockBackend *blk, uint64_t *perm, uint64_t *shared_perm)
1001 {
1002     GLOBAL_STATE_CODE();
1003     *perm = blk->perm;
1004     *shared_perm = blk->shared_perm;
1005 }
1006 
1007 /*
1008  * Attach device model @dev to @blk.
1009  * Return 0 on success, -EBUSY when a device model is attached already.
1010  */
1011 int blk_attach_dev(BlockBackend *blk, DeviceState *dev)
1012 {
1013     GLOBAL_STATE_CODE();
1014     if (blk->dev) {
1015         return -EBUSY;
1016     }
1017 
1018     /* While migration is still incoming, we don't need to apply the
1019      * permissions of guest device BlockBackends. We might still have a block
1020      * job or NBD server writing to the image for storage migration. */
1021     if (runstate_check(RUN_STATE_INMIGRATE)) {
1022         blk->disable_perm = true;
1023     }
1024 
1025     blk_ref(blk);
1026     blk->dev = dev;
1027     blk_iostatus_reset(blk);
1028 
1029     return 0;
1030 }
1031 
1032 /*
1033  * Detach device model @dev from @blk.
1034  * @dev must be currently attached to @blk.
1035  */
1036 void blk_detach_dev(BlockBackend *blk, DeviceState *dev)
1037 {
1038     assert(blk->dev == dev);
1039     GLOBAL_STATE_CODE();
1040     blk->dev = NULL;
1041     blk->dev_ops = NULL;
1042     blk->dev_opaque = NULL;
1043     blk_set_perm(blk, 0, BLK_PERM_ALL, &error_abort);
1044     blk_unref(blk);
1045 }
1046 
1047 /*
1048  * Return the device model attached to @blk if any, else null.
1049  */
1050 DeviceState *blk_get_attached_dev(BlockBackend *blk)
1051 {
1052     GLOBAL_STATE_CODE();
1053     return blk->dev;
1054 }
1055 
1056 /* Return the qdev ID, or if no ID is assigned the QOM path, of the block
1057  * device attached to the BlockBackend. */
1058 char *blk_get_attached_dev_id(BlockBackend *blk)
1059 {
1060     DeviceState *dev = blk->dev;
1061     IO_CODE();
1062 
1063     if (!dev) {
1064         return g_strdup("");
1065     } else if (dev->id) {
1066         return g_strdup(dev->id);
1067     }
1068 
1069     return object_get_canonical_path(OBJECT(dev)) ?: g_strdup("");
1070 }
1071 
1072 /*
1073  * Return the BlockBackend which has the device model @dev attached if it
1074  * exists, else null.
1075  *
1076  * @dev must not be null.
1077  */
1078 BlockBackend *blk_by_dev(void *dev)
1079 {
1080     BlockBackend *blk = NULL;
1081 
1082     GLOBAL_STATE_CODE();
1083 
1084     assert(dev != NULL);
1085     while ((blk = blk_all_next(blk)) != NULL) {
1086         if (blk->dev == dev) {
1087             return blk;
1088         }
1089     }
1090     return NULL;
1091 }
1092 
1093 /*
1094  * Set @blk's device model callbacks to @ops.
1095  * @opaque is the opaque argument to pass to the callbacks.
1096  * This is for use by device models.
1097  */
1098 void blk_set_dev_ops(BlockBackend *blk, const BlockDevOps *ops,
1099                      void *opaque)
1100 {
1101     GLOBAL_STATE_CODE();
1102     blk->dev_ops = ops;
1103     blk->dev_opaque = opaque;
1104 
1105     /* Are we currently quiesced? Should we enforce this right now? */
1106     if (qatomic_read(&blk->quiesce_counter) && ops && ops->drained_begin) {
1107         ops->drained_begin(opaque);
1108     }
1109 }
1110 
1111 /*
1112  * Notify @blk's attached device model of media change.
1113  *
1114  * If @load is true, notify of media load. This action can fail, meaning that
1115  * the medium cannot be loaded. @errp is set then.
1116  *
1117  * If @load is false, notify of media eject. This can never fail.
1118  *
1119  * Also send DEVICE_TRAY_MOVED events as appropriate.
1120  */
1121 void blk_dev_change_media_cb(BlockBackend *blk, bool load, Error **errp)
1122 {
1123     GLOBAL_STATE_CODE();
1124     if (blk->dev_ops && blk->dev_ops->change_media_cb) {
1125         bool tray_was_open, tray_is_open;
1126         Error *local_err = NULL;
1127 
1128         tray_was_open = blk_dev_is_tray_open(blk);
1129         blk->dev_ops->change_media_cb(blk->dev_opaque, load, &local_err);
1130         if (local_err) {
1131             assert(load == true);
1132             error_propagate(errp, local_err);
1133             return;
1134         }
1135         tray_is_open = blk_dev_is_tray_open(blk);
1136 
1137         if (tray_was_open != tray_is_open) {
1138             char *id = blk_get_attached_dev_id(blk);
1139             qapi_event_send_device_tray_moved(blk_name(blk), id, tray_is_open);
1140             g_free(id);
1141         }
1142     }
1143 }
1144 
1145 static void blk_root_change_media(BdrvChild *child, bool load)
1146 {
1147     blk_dev_change_media_cb(child->opaque, load, NULL);
1148 }
1149 
1150 /*
1151  * Does @blk's attached device model have removable media?
1152  * %true if no device model is attached.
1153  */
1154 bool blk_dev_has_removable_media(BlockBackend *blk)
1155 {
1156     GLOBAL_STATE_CODE();
1157     return !blk->dev || (blk->dev_ops && blk->dev_ops->change_media_cb);
1158 }
1159 
1160 /*
1161  * Does @blk's attached device model have a tray?
1162  */
1163 bool blk_dev_has_tray(BlockBackend *blk)
1164 {
1165     IO_CODE();
1166     return blk->dev_ops && blk->dev_ops->is_tray_open;
1167 }
1168 
1169 /*
1170  * Notify @blk's attached device model of a media eject request.
1171  * If @force is true, the medium is about to be yanked out forcefully.
1172  */
1173 void blk_dev_eject_request(BlockBackend *blk, bool force)
1174 {
1175     GLOBAL_STATE_CODE();
1176     if (blk->dev_ops && blk->dev_ops->eject_request_cb) {
1177         blk->dev_ops->eject_request_cb(blk->dev_opaque, force);
1178     }
1179 }
1180 
1181 /*
1182  * Does @blk's attached device model have a tray, and is it open?
1183  */
1184 bool blk_dev_is_tray_open(BlockBackend *blk)
1185 {
1186     IO_CODE();
1187     if (blk_dev_has_tray(blk)) {
1188         return blk->dev_ops->is_tray_open(blk->dev_opaque);
1189     }
1190     return false;
1191 }
1192 
1193 /*
1194  * Does @blk's attached device model have the medium locked?
1195  * %false if the device model has no such lock.
1196  */
1197 bool blk_dev_is_medium_locked(BlockBackend *blk)
1198 {
1199     GLOBAL_STATE_CODE();
1200     if (blk->dev_ops && blk->dev_ops->is_medium_locked) {
1201         return blk->dev_ops->is_medium_locked(blk->dev_opaque);
1202     }
1203     return false;
1204 }
1205 
1206 /*
1207  * Notify @blk's attached device model of a backend size change.
1208  */
1209 static void blk_root_resize(BdrvChild *child)
1210 {
1211     BlockBackend *blk = child->opaque;
1212 
1213     if (blk->dev_ops && blk->dev_ops->resize_cb) {
1214         blk->dev_ops->resize_cb(blk->dev_opaque);
1215     }
1216 }
1217 
1218 void blk_iostatus_enable(BlockBackend *blk)
1219 {
1220     GLOBAL_STATE_CODE();
1221     blk->iostatus_enabled = true;
1222     blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
1223 }
1224 
1225 /* The I/O status is only enabled if the drive explicitly
1226  * enables it _and_ the VM is configured to stop on errors */
1227 bool blk_iostatus_is_enabled(const BlockBackend *blk)
1228 {
1229     IO_CODE();
1230     return (blk->iostatus_enabled &&
1231            (blk->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
1232             blk->on_write_error == BLOCKDEV_ON_ERROR_STOP   ||
1233             blk->on_read_error == BLOCKDEV_ON_ERROR_STOP));
1234 }
1235 
1236 BlockDeviceIoStatus blk_iostatus(const BlockBackend *blk)
1237 {
1238     GLOBAL_STATE_CODE();
1239     return blk->iostatus;
1240 }
1241 
1242 void blk_iostatus_disable(BlockBackend *blk)
1243 {
1244     GLOBAL_STATE_CODE();
1245     blk->iostatus_enabled = false;
1246 }
1247 
1248 void blk_iostatus_reset(BlockBackend *blk)
1249 {
1250     GLOBAL_STATE_CODE();
1251     if (blk_iostatus_is_enabled(blk)) {
1252         blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
1253     }
1254 }
1255 
1256 void blk_iostatus_set_err(BlockBackend *blk, int error)
1257 {
1258     IO_CODE();
1259     assert(blk_iostatus_is_enabled(blk));
1260     if (blk->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
1261         blk->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
1262                                           BLOCK_DEVICE_IO_STATUS_FAILED;
1263     }
1264 }
1265 
1266 void blk_set_allow_write_beyond_eof(BlockBackend *blk, bool allow)
1267 {
1268     IO_CODE();
1269     blk->allow_write_beyond_eof = allow;
1270 }
1271 
1272 void blk_set_allow_aio_context_change(BlockBackend *blk, bool allow)
1273 {
1274     IO_CODE();
1275     blk->allow_aio_context_change = allow;
1276 }
1277 
1278 void blk_set_disable_request_queuing(BlockBackend *blk, bool disable)
1279 {
1280     IO_CODE();
1281     qatomic_set(&blk->disable_request_queuing, disable);
1282 }
1283 
1284 static int coroutine_fn GRAPH_RDLOCK
1285 blk_check_byte_request(BlockBackend *blk, int64_t offset, int64_t bytes)
1286 {
1287     int64_t len;
1288 
1289     if (bytes < 0) {
1290         return -EIO;
1291     }
1292 
1293     if (!blk_co_is_available(blk)) {
1294         return -ENOMEDIUM;
1295     }
1296 
1297     if (offset < 0) {
1298         return -EIO;
1299     }
1300 
1301     if (!blk->allow_write_beyond_eof) {
1302         len = bdrv_co_getlength(blk_bs(blk));
1303         if (len < 0) {
1304             return len;
1305         }
1306 
1307         if (offset > len || len - offset < bytes) {
1308             return -EIO;
1309         }
1310     }
1311 
1312     return 0;
1313 }
1314 
1315 /* Are we currently in a drained section? */
1316 bool blk_in_drain(BlockBackend *blk)
1317 {
1318     GLOBAL_STATE_CODE(); /* change to IO_OR_GS_CODE(), if necessary */
1319     return qatomic_read(&blk->quiesce_counter);
1320 }
1321 
1322 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1323 static void coroutine_fn blk_wait_while_drained(BlockBackend *blk)
1324 {
1325     assert(blk->in_flight > 0);
1326 
1327     if (qatomic_read(&blk->quiesce_counter) &&
1328         !qatomic_read(&blk->disable_request_queuing)) {
1329         /*
1330          * Take lock before decrementing in flight counter so main loop thread
1331          * waits for us to enqueue ourselves before it can leave the drained
1332          * section.
1333          */
1334         qemu_mutex_lock(&blk->queued_requests_lock);
1335         blk_dec_in_flight(blk);
1336         qemu_co_queue_wait(&blk->queued_requests, &blk->queued_requests_lock);
1337         blk_inc_in_flight(blk);
1338         qemu_mutex_unlock(&blk->queued_requests_lock);
1339     }
1340 }
1341 
1342 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1343 static int coroutine_fn
1344 blk_co_do_preadv_part(BlockBackend *blk, int64_t offset, int64_t bytes,
1345                       QEMUIOVector *qiov, size_t qiov_offset,
1346                       BdrvRequestFlags flags)
1347 {
1348     int ret;
1349     BlockDriverState *bs;
1350     IO_CODE();
1351 
1352     blk_wait_while_drained(blk);
1353     GRAPH_RDLOCK_GUARD();
1354 
1355     /* Call blk_bs() only after waiting, the graph may have changed */
1356     bs = blk_bs(blk);
1357     trace_blk_co_preadv(blk, bs, offset, bytes, flags);
1358 
1359     ret = blk_check_byte_request(blk, offset, bytes);
1360     if (ret < 0) {
1361         return ret;
1362     }
1363 
1364     bdrv_inc_in_flight(bs);
1365 
1366     /* throttling disk I/O */
1367     if (blk->public.throttle_group_member.throttle_state) {
1368         throttle_group_co_io_limits_intercept(&blk->public.throttle_group_member,
1369                 bytes, THROTTLE_READ);
1370     }
1371 
1372     ret = bdrv_co_preadv_part(blk->root, offset, bytes, qiov, qiov_offset,
1373                               flags);
1374     bdrv_dec_in_flight(bs);
1375     return ret;
1376 }
1377 
1378 int coroutine_fn blk_co_pread(BlockBackend *blk, int64_t offset, int64_t bytes,
1379                               void *buf, BdrvRequestFlags flags)
1380 {
1381     QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
1382     IO_OR_GS_CODE();
1383 
1384     assert(bytes <= SIZE_MAX);
1385 
1386     return blk_co_preadv(blk, offset, bytes, &qiov, flags);
1387 }
1388 
1389 int coroutine_fn blk_co_preadv(BlockBackend *blk, int64_t offset,
1390                                int64_t bytes, QEMUIOVector *qiov,
1391                                BdrvRequestFlags flags)
1392 {
1393     int ret;
1394     IO_OR_GS_CODE();
1395 
1396     blk_inc_in_flight(blk);
1397     ret = blk_co_do_preadv_part(blk, offset, bytes, qiov, 0, flags);
1398     blk_dec_in_flight(blk);
1399 
1400     return ret;
1401 }
1402 
1403 int coroutine_fn blk_co_preadv_part(BlockBackend *blk, int64_t offset,
1404                                     int64_t bytes, QEMUIOVector *qiov,
1405                                     size_t qiov_offset, BdrvRequestFlags flags)
1406 {
1407     int ret;
1408     IO_OR_GS_CODE();
1409 
1410     blk_inc_in_flight(blk);
1411     ret = blk_co_do_preadv_part(blk, offset, bytes, qiov, qiov_offset, flags);
1412     blk_dec_in_flight(blk);
1413 
1414     return ret;
1415 }
1416 
1417 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1418 static int coroutine_fn
1419 blk_co_do_pwritev_part(BlockBackend *blk, int64_t offset, int64_t bytes,
1420                        QEMUIOVector *qiov, size_t qiov_offset,
1421                        BdrvRequestFlags flags)
1422 {
1423     int ret;
1424     BlockDriverState *bs;
1425     IO_CODE();
1426 
1427     blk_wait_while_drained(blk);
1428     GRAPH_RDLOCK_GUARD();
1429 
1430     /* Call blk_bs() only after waiting, the graph may have changed */
1431     bs = blk_bs(blk);
1432     trace_blk_co_pwritev(blk, bs, offset, bytes, flags);
1433 
1434     ret = blk_check_byte_request(blk, offset, bytes);
1435     if (ret < 0) {
1436         return ret;
1437     }
1438 
1439     bdrv_inc_in_flight(bs);
1440     /* throttling disk I/O */
1441     if (blk->public.throttle_group_member.throttle_state) {
1442         throttle_group_co_io_limits_intercept(&blk->public.throttle_group_member,
1443                 bytes, THROTTLE_WRITE);
1444     }
1445 
1446     if (!blk->enable_write_cache) {
1447         flags |= BDRV_REQ_FUA;
1448     }
1449 
1450     ret = bdrv_co_pwritev_part(blk->root, offset, bytes, qiov, qiov_offset,
1451                                flags);
1452     bdrv_dec_in_flight(bs);
1453     return ret;
1454 }
1455 
1456 int coroutine_fn blk_co_pwritev_part(BlockBackend *blk, int64_t offset,
1457                                      int64_t bytes,
1458                                      QEMUIOVector *qiov, size_t qiov_offset,
1459                                      BdrvRequestFlags flags)
1460 {
1461     int ret;
1462     IO_OR_GS_CODE();
1463 
1464     blk_inc_in_flight(blk);
1465     ret = blk_co_do_pwritev_part(blk, offset, bytes, qiov, qiov_offset, flags);
1466     blk_dec_in_flight(blk);
1467 
1468     return ret;
1469 }
1470 
1471 int coroutine_fn blk_co_pwrite(BlockBackend *blk, int64_t offset, int64_t bytes,
1472                                const void *buf, BdrvRequestFlags flags)
1473 {
1474     QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
1475     IO_OR_GS_CODE();
1476 
1477     assert(bytes <= SIZE_MAX);
1478 
1479     return blk_co_pwritev(blk, offset, bytes, &qiov, flags);
1480 }
1481 
1482 int coroutine_fn blk_co_pwritev(BlockBackend *blk, int64_t offset,
1483                                 int64_t bytes, QEMUIOVector *qiov,
1484                                 BdrvRequestFlags flags)
1485 {
1486     IO_OR_GS_CODE();
1487     return blk_co_pwritev_part(blk, offset, bytes, qiov, 0, flags);
1488 }
1489 
1490 int coroutine_fn blk_co_block_status_above(BlockBackend *blk,
1491                                            BlockDriverState *base,
1492                                            int64_t offset, int64_t bytes,
1493                                            int64_t *pnum, int64_t *map,
1494                                            BlockDriverState **file)
1495 {
1496     IO_CODE();
1497     GRAPH_RDLOCK_GUARD();
1498     return bdrv_co_block_status_above(blk_bs(blk), base, offset, bytes, pnum,
1499                                       map, file);
1500 }
1501 
1502 int coroutine_fn blk_co_is_allocated_above(BlockBackend *blk,
1503                                            BlockDriverState *base,
1504                                            bool include_base, int64_t offset,
1505                                            int64_t bytes, int64_t *pnum)
1506 {
1507     IO_CODE();
1508     GRAPH_RDLOCK_GUARD();
1509     return bdrv_co_is_allocated_above(blk_bs(blk), base, include_base, offset,
1510                                       bytes, pnum);
1511 }
1512 
1513 typedef struct BlkRwCo {
1514     BlockBackend *blk;
1515     int64_t offset;
1516     void *iobuf;
1517     int ret;
1518     BdrvRequestFlags flags;
1519 } BlkRwCo;
1520 
1521 int blk_make_zero(BlockBackend *blk, BdrvRequestFlags flags)
1522 {
1523     GLOBAL_STATE_CODE();
1524     return bdrv_make_zero(blk->root, flags);
1525 }
1526 
1527 void blk_inc_in_flight(BlockBackend *blk)
1528 {
1529     IO_CODE();
1530     qatomic_inc(&blk->in_flight);
1531 }
1532 
1533 void blk_dec_in_flight(BlockBackend *blk)
1534 {
1535     IO_CODE();
1536     qatomic_dec(&blk->in_flight);
1537     aio_wait_kick();
1538 }
1539 
1540 static void error_callback_bh(void *opaque)
1541 {
1542     struct BlockBackendAIOCB *acb = opaque;
1543 
1544     blk_dec_in_flight(acb->blk);
1545     acb->common.cb(acb->common.opaque, acb->ret);
1546     qemu_aio_unref(acb);
1547 }
1548 
1549 BlockAIOCB *blk_abort_aio_request(BlockBackend *blk,
1550                                   BlockCompletionFunc *cb,
1551                                   void *opaque, int ret)
1552 {
1553     struct BlockBackendAIOCB *acb;
1554     IO_CODE();
1555 
1556     blk_inc_in_flight(blk);
1557     acb = blk_aio_get(&block_backend_aiocb_info, blk, cb, opaque);
1558     acb->blk = blk;
1559     acb->ret = ret;
1560 
1561     replay_bh_schedule_oneshot_event(qemu_get_current_aio_context(),
1562                                      error_callback_bh, acb);
1563     return &acb->common;
1564 }
1565 
1566 typedef struct BlkAioEmAIOCB {
1567     BlockAIOCB common;
1568     BlkRwCo rwco;
1569     int64_t bytes;
1570     bool has_returned;
1571 } BlkAioEmAIOCB;
1572 
1573 static const AIOCBInfo blk_aio_em_aiocb_info = {
1574     .aiocb_size         = sizeof(BlkAioEmAIOCB),
1575 };
1576 
1577 static void blk_aio_complete(BlkAioEmAIOCB *acb)
1578 {
1579     if (acb->has_returned) {
1580         acb->common.cb(acb->common.opaque, acb->rwco.ret);
1581         blk_dec_in_flight(acb->rwco.blk);
1582         qemu_aio_unref(acb);
1583     }
1584 }
1585 
1586 static void blk_aio_complete_bh(void *opaque)
1587 {
1588     BlkAioEmAIOCB *acb = opaque;
1589     assert(acb->has_returned);
1590     blk_aio_complete(acb);
1591 }
1592 
1593 static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset,
1594                                 int64_t bytes,
1595                                 void *iobuf, CoroutineEntry co_entry,
1596                                 BdrvRequestFlags flags,
1597                                 BlockCompletionFunc *cb, void *opaque)
1598 {
1599     BlkAioEmAIOCB *acb;
1600     Coroutine *co;
1601 
1602     blk_inc_in_flight(blk);
1603     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1604     acb->rwco = (BlkRwCo) {
1605         .blk    = blk,
1606         .offset = offset,
1607         .iobuf  = iobuf,
1608         .flags  = flags,
1609         .ret    = NOT_DONE,
1610     };
1611     acb->bytes = bytes;
1612     acb->has_returned = false;
1613 
1614     co = qemu_coroutine_create(co_entry, acb);
1615     aio_co_enter(qemu_get_current_aio_context(), co);
1616 
1617     acb->has_returned = true;
1618     if (acb->rwco.ret != NOT_DONE) {
1619         replay_bh_schedule_oneshot_event(qemu_get_current_aio_context(),
1620                                          blk_aio_complete_bh, acb);
1621     }
1622 
1623     return &acb->common;
1624 }
1625 
1626 static void coroutine_fn blk_aio_read_entry(void *opaque)
1627 {
1628     BlkAioEmAIOCB *acb = opaque;
1629     BlkRwCo *rwco = &acb->rwco;
1630     QEMUIOVector *qiov = rwco->iobuf;
1631 
1632     assert(qiov->size == acb->bytes);
1633     rwco->ret = blk_co_do_preadv_part(rwco->blk, rwco->offset, acb->bytes, qiov,
1634                                       0, rwco->flags);
1635     blk_aio_complete(acb);
1636 }
1637 
1638 static void coroutine_fn blk_aio_write_entry(void *opaque)
1639 {
1640     BlkAioEmAIOCB *acb = opaque;
1641     BlkRwCo *rwco = &acb->rwco;
1642     QEMUIOVector *qiov = rwco->iobuf;
1643 
1644     assert(!qiov || qiov->size == acb->bytes);
1645     rwco->ret = blk_co_do_pwritev_part(rwco->blk, rwco->offset, acb->bytes,
1646                                        qiov, 0, rwco->flags);
1647     blk_aio_complete(acb);
1648 }
1649 
1650 BlockAIOCB *blk_aio_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1651                                   int64_t bytes, BdrvRequestFlags flags,
1652                                   BlockCompletionFunc *cb, void *opaque)
1653 {
1654     IO_CODE();
1655     return blk_aio_prwv(blk, offset, bytes, NULL, blk_aio_write_entry,
1656                         flags | BDRV_REQ_ZERO_WRITE, cb, opaque);
1657 }
1658 
1659 int64_t coroutine_fn blk_co_getlength(BlockBackend *blk)
1660 {
1661     IO_CODE();
1662     GRAPH_RDLOCK_GUARD();
1663 
1664     if (!blk_co_is_available(blk)) {
1665         return -ENOMEDIUM;
1666     }
1667 
1668     return bdrv_co_getlength(blk_bs(blk));
1669 }
1670 
1671 int64_t coroutine_fn blk_co_nb_sectors(BlockBackend *blk)
1672 {
1673     BlockDriverState *bs = blk_bs(blk);
1674 
1675     IO_CODE();
1676     GRAPH_RDLOCK_GUARD();
1677 
1678     if (!bs) {
1679         return -ENOMEDIUM;
1680     } else {
1681         return bdrv_co_nb_sectors(bs);
1682     }
1683 }
1684 
1685 /*
1686  * This wrapper is written by hand because this function is in the hot I/O path,
1687  * via blk_get_geometry.
1688  */
1689 int64_t coroutine_mixed_fn blk_nb_sectors(BlockBackend *blk)
1690 {
1691     BlockDriverState *bs = blk_bs(blk);
1692 
1693     IO_CODE();
1694 
1695     if (!bs) {
1696         return -ENOMEDIUM;
1697     } else {
1698         return bdrv_nb_sectors(bs);
1699     }
1700 }
1701 
1702 /* return 0 as number of sectors if no device present or error */
1703 void coroutine_fn blk_co_get_geometry(BlockBackend *blk,
1704                                       uint64_t *nb_sectors_ptr)
1705 {
1706     int64_t ret = blk_co_nb_sectors(blk);
1707     *nb_sectors_ptr = ret < 0 ? 0 : ret;
1708 }
1709 
1710 /*
1711  * This wrapper is written by hand because this function is in the hot I/O path.
1712  */
1713 void coroutine_mixed_fn blk_get_geometry(BlockBackend *blk,
1714                                          uint64_t *nb_sectors_ptr)
1715 {
1716     int64_t ret = blk_nb_sectors(blk);
1717     *nb_sectors_ptr = ret < 0 ? 0 : ret;
1718 }
1719 
1720 BlockAIOCB *blk_aio_preadv(BlockBackend *blk, int64_t offset,
1721                            QEMUIOVector *qiov, BdrvRequestFlags flags,
1722                            BlockCompletionFunc *cb, void *opaque)
1723 {
1724     IO_CODE();
1725     assert((uint64_t)qiov->size <= INT64_MAX);
1726     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1727                         blk_aio_read_entry, flags, cb, opaque);
1728 }
1729 
1730 BlockAIOCB *blk_aio_pwritev(BlockBackend *blk, int64_t offset,
1731                             QEMUIOVector *qiov, BdrvRequestFlags flags,
1732                             BlockCompletionFunc *cb, void *opaque)
1733 {
1734     IO_CODE();
1735     assert((uint64_t)qiov->size <= INT64_MAX);
1736     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1737                         blk_aio_write_entry, flags, cb, opaque);
1738 }
1739 
1740 void blk_aio_cancel(BlockAIOCB *acb)
1741 {
1742     GLOBAL_STATE_CODE();
1743     bdrv_aio_cancel(acb);
1744 }
1745 
1746 void blk_aio_cancel_async(BlockAIOCB *acb)
1747 {
1748     IO_CODE();
1749     bdrv_aio_cancel_async(acb);
1750 }
1751 
1752 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1753 static int coroutine_fn
1754 blk_co_do_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1755 {
1756     IO_CODE();
1757 
1758     blk_wait_while_drained(blk);
1759     GRAPH_RDLOCK_GUARD();
1760 
1761     if (!blk_co_is_available(blk)) {
1762         return -ENOMEDIUM;
1763     }
1764 
1765     return bdrv_co_ioctl(blk_bs(blk), req, buf);
1766 }
1767 
1768 int coroutine_fn blk_co_ioctl(BlockBackend *blk, unsigned long int req,
1769                               void *buf)
1770 {
1771     int ret;
1772     IO_OR_GS_CODE();
1773 
1774     blk_inc_in_flight(blk);
1775     ret = blk_co_do_ioctl(blk, req, buf);
1776     blk_dec_in_flight(blk);
1777 
1778     return ret;
1779 }
1780 
1781 static void coroutine_fn blk_aio_ioctl_entry(void *opaque)
1782 {
1783     BlkAioEmAIOCB *acb = opaque;
1784     BlkRwCo *rwco = &acb->rwco;
1785 
1786     rwco->ret = blk_co_do_ioctl(rwco->blk, rwco->offset, rwco->iobuf);
1787 
1788     blk_aio_complete(acb);
1789 }
1790 
1791 BlockAIOCB *blk_aio_ioctl(BlockBackend *blk, unsigned long int req, void *buf,
1792                           BlockCompletionFunc *cb, void *opaque)
1793 {
1794     IO_CODE();
1795     return blk_aio_prwv(blk, req, 0, buf, blk_aio_ioctl_entry, 0, cb, opaque);
1796 }
1797 
1798 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1799 static int coroutine_fn
1800 blk_co_do_pdiscard(BlockBackend *blk, int64_t offset, int64_t bytes)
1801 {
1802     int ret;
1803     IO_CODE();
1804 
1805     blk_wait_while_drained(blk);
1806     GRAPH_RDLOCK_GUARD();
1807 
1808     ret = blk_check_byte_request(blk, offset, bytes);
1809     if (ret < 0) {
1810         return ret;
1811     }
1812 
1813     return bdrv_co_pdiscard(blk->root, offset, bytes);
1814 }
1815 
1816 static void coroutine_fn blk_aio_pdiscard_entry(void *opaque)
1817 {
1818     BlkAioEmAIOCB *acb = opaque;
1819     BlkRwCo *rwco = &acb->rwco;
1820 
1821     rwco->ret = blk_co_do_pdiscard(rwco->blk, rwco->offset, acb->bytes);
1822     blk_aio_complete(acb);
1823 }
1824 
1825 BlockAIOCB *blk_aio_pdiscard(BlockBackend *blk,
1826                              int64_t offset, int64_t bytes,
1827                              BlockCompletionFunc *cb, void *opaque)
1828 {
1829     IO_CODE();
1830     return blk_aio_prwv(blk, offset, bytes, NULL, blk_aio_pdiscard_entry, 0,
1831                         cb, opaque);
1832 }
1833 
1834 int coroutine_fn blk_co_pdiscard(BlockBackend *blk, int64_t offset,
1835                                  int64_t bytes)
1836 {
1837     int ret;
1838     IO_OR_GS_CODE();
1839 
1840     blk_inc_in_flight(blk);
1841     ret = blk_co_do_pdiscard(blk, offset, bytes);
1842     blk_dec_in_flight(blk);
1843 
1844     return ret;
1845 }
1846 
1847 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1848 static int coroutine_fn blk_co_do_flush(BlockBackend *blk)
1849 {
1850     IO_CODE();
1851     blk_wait_while_drained(blk);
1852     GRAPH_RDLOCK_GUARD();
1853 
1854     if (!blk_co_is_available(blk)) {
1855         return -ENOMEDIUM;
1856     }
1857 
1858     return bdrv_co_flush(blk_bs(blk));
1859 }
1860 
1861 static void coroutine_fn blk_aio_flush_entry(void *opaque)
1862 {
1863     BlkAioEmAIOCB *acb = opaque;
1864     BlkRwCo *rwco = &acb->rwco;
1865 
1866     rwco->ret = blk_co_do_flush(rwco->blk);
1867     blk_aio_complete(acb);
1868 }
1869 
1870 BlockAIOCB *blk_aio_flush(BlockBackend *blk,
1871                           BlockCompletionFunc *cb, void *opaque)
1872 {
1873     IO_CODE();
1874     return blk_aio_prwv(blk, 0, 0, NULL, blk_aio_flush_entry, 0, cb, opaque);
1875 }
1876 
1877 int coroutine_fn blk_co_flush(BlockBackend *blk)
1878 {
1879     int ret;
1880     IO_OR_GS_CODE();
1881 
1882     blk_inc_in_flight(blk);
1883     ret = blk_co_do_flush(blk);
1884     blk_dec_in_flight(blk);
1885 
1886     return ret;
1887 }
1888 
1889 static void coroutine_fn blk_aio_zone_report_entry(void *opaque)
1890 {
1891     BlkAioEmAIOCB *acb = opaque;
1892     BlkRwCo *rwco = &acb->rwco;
1893 
1894     rwco->ret = blk_co_zone_report(rwco->blk, rwco->offset,
1895                                    (unsigned int*)(uintptr_t)acb->bytes,
1896                                    rwco->iobuf);
1897     blk_aio_complete(acb);
1898 }
1899 
1900 BlockAIOCB *blk_aio_zone_report(BlockBackend *blk, int64_t offset,
1901                                 unsigned int *nr_zones,
1902                                 BlockZoneDescriptor  *zones,
1903                                 BlockCompletionFunc *cb, void *opaque)
1904 {
1905     BlkAioEmAIOCB *acb;
1906     Coroutine *co;
1907     IO_CODE();
1908 
1909     blk_inc_in_flight(blk);
1910     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1911     acb->rwco = (BlkRwCo) {
1912         .blk    = blk,
1913         .offset = offset,
1914         .iobuf  = zones,
1915         .ret    = NOT_DONE,
1916     };
1917     acb->bytes = (int64_t)(uintptr_t)nr_zones,
1918     acb->has_returned = false;
1919 
1920     co = qemu_coroutine_create(blk_aio_zone_report_entry, acb);
1921     aio_co_enter(qemu_get_current_aio_context(), co);
1922 
1923     acb->has_returned = true;
1924     if (acb->rwco.ret != NOT_DONE) {
1925         replay_bh_schedule_oneshot_event(qemu_get_current_aio_context(),
1926                                          blk_aio_complete_bh, acb);
1927     }
1928 
1929     return &acb->common;
1930 }
1931 
1932 static void coroutine_fn blk_aio_zone_mgmt_entry(void *opaque)
1933 {
1934     BlkAioEmAIOCB *acb = opaque;
1935     BlkRwCo *rwco = &acb->rwco;
1936 
1937     rwco->ret = blk_co_zone_mgmt(rwco->blk,
1938                                  (BlockZoneOp)(uintptr_t)rwco->iobuf,
1939                                  rwco->offset, acb->bytes);
1940     blk_aio_complete(acb);
1941 }
1942 
1943 BlockAIOCB *blk_aio_zone_mgmt(BlockBackend *blk, BlockZoneOp op,
1944                               int64_t offset, int64_t len,
1945                               BlockCompletionFunc *cb, void *opaque) {
1946     BlkAioEmAIOCB *acb;
1947     Coroutine *co;
1948     IO_CODE();
1949 
1950     blk_inc_in_flight(blk);
1951     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1952     acb->rwco = (BlkRwCo) {
1953         .blk    = blk,
1954         .offset = offset,
1955         .iobuf  = (void *)(uintptr_t)op,
1956         .ret    = NOT_DONE,
1957     };
1958     acb->bytes = len;
1959     acb->has_returned = false;
1960 
1961     co = qemu_coroutine_create(blk_aio_zone_mgmt_entry, acb);
1962     aio_co_enter(qemu_get_current_aio_context(), co);
1963 
1964     acb->has_returned = true;
1965     if (acb->rwco.ret != NOT_DONE) {
1966         replay_bh_schedule_oneshot_event(qemu_get_current_aio_context(),
1967                                          blk_aio_complete_bh, acb);
1968     }
1969 
1970     return &acb->common;
1971 }
1972 
1973 static void coroutine_fn blk_aio_zone_append_entry(void *opaque)
1974 {
1975     BlkAioEmAIOCB *acb = opaque;
1976     BlkRwCo *rwco = &acb->rwco;
1977 
1978     rwco->ret = blk_co_zone_append(rwco->blk, (int64_t *)(uintptr_t)acb->bytes,
1979                                    rwco->iobuf, rwco->flags);
1980     blk_aio_complete(acb);
1981 }
1982 
1983 BlockAIOCB *blk_aio_zone_append(BlockBackend *blk, int64_t *offset,
1984                                 QEMUIOVector *qiov, BdrvRequestFlags flags,
1985                                 BlockCompletionFunc *cb, void *opaque) {
1986     BlkAioEmAIOCB *acb;
1987     Coroutine *co;
1988     IO_CODE();
1989 
1990     blk_inc_in_flight(blk);
1991     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1992     acb->rwco = (BlkRwCo) {
1993         .blk    = blk,
1994         .ret    = NOT_DONE,
1995         .flags  = flags,
1996         .iobuf  = qiov,
1997     };
1998     acb->bytes = (int64_t)(uintptr_t)offset;
1999     acb->has_returned = false;
2000 
2001     co = qemu_coroutine_create(blk_aio_zone_append_entry, acb);
2002     aio_co_enter(qemu_get_current_aio_context(), co);
2003     acb->has_returned = true;
2004     if (acb->rwco.ret != NOT_DONE) {
2005         replay_bh_schedule_oneshot_event(qemu_get_current_aio_context(),
2006                                          blk_aio_complete_bh, acb);
2007     }
2008 
2009     return &acb->common;
2010 }
2011 
2012 /*
2013  * Send a zone_report command.
2014  * offset is a byte offset from the start of the device. No alignment
2015  * required for offset.
2016  * nr_zones represents IN maximum and OUT actual.
2017  */
2018 int coroutine_fn blk_co_zone_report(BlockBackend *blk, int64_t offset,
2019                                     unsigned int *nr_zones,
2020                                     BlockZoneDescriptor *zones)
2021 {
2022     int ret;
2023     IO_CODE();
2024 
2025     blk_inc_in_flight(blk); /* increase before waiting */
2026     blk_wait_while_drained(blk);
2027     GRAPH_RDLOCK_GUARD();
2028     if (!blk_is_available(blk)) {
2029         blk_dec_in_flight(blk);
2030         return -ENOMEDIUM;
2031     }
2032     ret = bdrv_co_zone_report(blk_bs(blk), offset, nr_zones, zones);
2033     blk_dec_in_flight(blk);
2034     return ret;
2035 }
2036 
2037 /*
2038  * Send a zone_management command.
2039  * op is the zone operation;
2040  * offset is the byte offset from the start of the zoned device;
2041  * len is the maximum number of bytes the command should operate on. It
2042  * should be aligned with the device zone size.
2043  */
2044 int coroutine_fn blk_co_zone_mgmt(BlockBackend *blk, BlockZoneOp op,
2045         int64_t offset, int64_t len)
2046 {
2047     int ret;
2048     IO_CODE();
2049 
2050     blk_inc_in_flight(blk);
2051     blk_wait_while_drained(blk);
2052     GRAPH_RDLOCK_GUARD();
2053 
2054     ret = blk_check_byte_request(blk, offset, len);
2055     if (ret < 0) {
2056         blk_dec_in_flight(blk);
2057         return ret;
2058     }
2059 
2060     ret = bdrv_co_zone_mgmt(blk_bs(blk), op, offset, len);
2061     blk_dec_in_flight(blk);
2062     return ret;
2063 }
2064 
2065 /*
2066  * Send a zone_append command.
2067  */
2068 int coroutine_fn blk_co_zone_append(BlockBackend *blk, int64_t *offset,
2069         QEMUIOVector *qiov, BdrvRequestFlags flags)
2070 {
2071     int ret;
2072     IO_CODE();
2073 
2074     blk_inc_in_flight(blk);
2075     blk_wait_while_drained(blk);
2076     GRAPH_RDLOCK_GUARD();
2077     if (!blk_is_available(blk)) {
2078         blk_dec_in_flight(blk);
2079         return -ENOMEDIUM;
2080     }
2081 
2082     ret = bdrv_co_zone_append(blk_bs(blk), offset, qiov, flags);
2083     blk_dec_in_flight(blk);
2084     return ret;
2085 }
2086 
2087 void blk_drain(BlockBackend *blk)
2088 {
2089     BlockDriverState *bs = blk_bs(blk);
2090     GLOBAL_STATE_CODE();
2091 
2092     if (bs) {
2093         bdrv_ref(bs);
2094         bdrv_drained_begin(bs);
2095     }
2096 
2097     /* We may have -ENOMEDIUM completions in flight */
2098     AIO_WAIT_WHILE(blk_get_aio_context(blk),
2099                    qatomic_read(&blk->in_flight) > 0);
2100 
2101     if (bs) {
2102         bdrv_drained_end(bs);
2103         bdrv_unref(bs);
2104     }
2105 }
2106 
2107 void blk_drain_all(void)
2108 {
2109     BlockBackend *blk = NULL;
2110 
2111     GLOBAL_STATE_CODE();
2112 
2113     bdrv_drain_all_begin();
2114 
2115     while ((blk = blk_all_next(blk)) != NULL) {
2116         /* We may have -ENOMEDIUM completions in flight */
2117         AIO_WAIT_WHILE_UNLOCKED(NULL, qatomic_read(&blk->in_flight) > 0);
2118     }
2119 
2120     bdrv_drain_all_end();
2121 }
2122 
2123 void blk_set_on_error(BlockBackend *blk, BlockdevOnError on_read_error,
2124                       BlockdevOnError on_write_error)
2125 {
2126     GLOBAL_STATE_CODE();
2127     blk->on_read_error = on_read_error;
2128     blk->on_write_error = on_write_error;
2129 }
2130 
2131 BlockdevOnError blk_get_on_error(BlockBackend *blk, bool is_read)
2132 {
2133     IO_CODE();
2134     return is_read ? blk->on_read_error : blk->on_write_error;
2135 }
2136 
2137 BlockErrorAction blk_get_error_action(BlockBackend *blk, bool is_read,
2138                                       int error)
2139 {
2140     BlockdevOnError on_err = blk_get_on_error(blk, is_read);
2141     IO_CODE();
2142 
2143     switch (on_err) {
2144     case BLOCKDEV_ON_ERROR_ENOSPC:
2145         return (error == ENOSPC) ?
2146                BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
2147     case BLOCKDEV_ON_ERROR_STOP:
2148         return BLOCK_ERROR_ACTION_STOP;
2149     case BLOCKDEV_ON_ERROR_REPORT:
2150         return BLOCK_ERROR_ACTION_REPORT;
2151     case BLOCKDEV_ON_ERROR_IGNORE:
2152         return BLOCK_ERROR_ACTION_IGNORE;
2153     case BLOCKDEV_ON_ERROR_AUTO:
2154     default:
2155         abort();
2156     }
2157 }
2158 
2159 static void send_qmp_error_event(BlockBackend *blk,
2160                                  BlockErrorAction action,
2161                                  bool is_read, int error)
2162 {
2163     IoOperationType optype;
2164     BlockDriverState *bs = blk_bs(blk);
2165 
2166     optype = is_read ? IO_OPERATION_TYPE_READ : IO_OPERATION_TYPE_WRITE;
2167     qapi_event_send_block_io_error(blk_name(blk),
2168                                    bs ? bdrv_get_node_name(bs) : NULL, optype,
2169                                    action, blk_iostatus_is_enabled(blk),
2170                                    error == ENOSPC, strerror(error));
2171 }
2172 
2173 /* This is done by device models because, while the block layer knows
2174  * about the error, it does not know whether an operation comes from
2175  * the device or the block layer (from a job, for example).
2176  */
2177 void blk_error_action(BlockBackend *blk, BlockErrorAction action,
2178                       bool is_read, int error)
2179 {
2180     assert(error >= 0);
2181     IO_CODE();
2182 
2183     if (action == BLOCK_ERROR_ACTION_STOP) {
2184         /* First set the iostatus, so that "info block" returns an iostatus
2185          * that matches the events raised so far (an additional error iostatus
2186          * is fine, but not a lost one).
2187          */
2188         blk_iostatus_set_err(blk, error);
2189 
2190         /* Then raise the request to stop the VM and the event.
2191          * qemu_system_vmstop_request_prepare has two effects.  First,
2192          * it ensures that the STOP event always comes after the
2193          * BLOCK_IO_ERROR event.  Second, it ensures that even if management
2194          * can observe the STOP event and do a "cont" before the STOP
2195          * event is issued, the VM will not stop.  In this case, vm_start()
2196          * also ensures that the STOP/RESUME pair of events is emitted.
2197          */
2198         qemu_system_vmstop_request_prepare();
2199         send_qmp_error_event(blk, action, is_read, error);
2200         qemu_system_vmstop_request(RUN_STATE_IO_ERROR);
2201     } else {
2202         send_qmp_error_event(blk, action, is_read, error);
2203     }
2204 }
2205 
2206 /*
2207  * Returns true if the BlockBackend can support taking write permissions
2208  * (because its root node is not read-only).
2209  */
2210 bool blk_supports_write_perm(BlockBackend *blk)
2211 {
2212     BlockDriverState *bs = blk_bs(blk);
2213     GLOBAL_STATE_CODE();
2214 
2215     if (bs) {
2216         return !bdrv_is_read_only(bs);
2217     } else {
2218         return blk->root_state.open_flags & BDRV_O_RDWR;
2219     }
2220 }
2221 
2222 /*
2223  * Returns true if the BlockBackend can be written to in its current
2224  * configuration (i.e. if write permission have been requested)
2225  */
2226 bool blk_is_writable(BlockBackend *blk)
2227 {
2228     IO_CODE();
2229     return blk->perm & BLK_PERM_WRITE;
2230 }
2231 
2232 bool blk_is_sg(BlockBackend *blk)
2233 {
2234     BlockDriverState *bs = blk_bs(blk);
2235     GLOBAL_STATE_CODE();
2236 
2237     if (!bs) {
2238         return false;
2239     }
2240 
2241     return bdrv_is_sg(bs);
2242 }
2243 
2244 bool blk_enable_write_cache(BlockBackend *blk)
2245 {
2246     IO_CODE();
2247     return blk->enable_write_cache;
2248 }
2249 
2250 void blk_set_enable_write_cache(BlockBackend *blk, bool wce)
2251 {
2252     IO_CODE();
2253     blk->enable_write_cache = wce;
2254 }
2255 
2256 void blk_activate(BlockBackend *blk, Error **errp)
2257 {
2258     BlockDriverState *bs = blk_bs(blk);
2259     GLOBAL_STATE_CODE();
2260 
2261     if (!bs) {
2262         error_setg(errp, "Device '%s' has no medium", blk->name);
2263         return;
2264     }
2265 
2266     /*
2267      * Migration code can call this function in coroutine context, so leave
2268      * coroutine context if necessary.
2269      */
2270     if (qemu_in_coroutine()) {
2271         bdrv_co_activate(bs, errp);
2272     } else {
2273         GRAPH_RDLOCK_GUARD_MAINLOOP();
2274         bdrv_activate(bs, errp);
2275     }
2276 }
2277 
2278 bool coroutine_fn blk_co_is_inserted(BlockBackend *blk)
2279 {
2280     BlockDriverState *bs = blk_bs(blk);
2281     IO_CODE();
2282     assert_bdrv_graph_readable();
2283 
2284     return bs && bdrv_co_is_inserted(bs);
2285 }
2286 
2287 bool coroutine_fn blk_co_is_available(BlockBackend *blk)
2288 {
2289     IO_CODE();
2290     return blk_co_is_inserted(blk) && !blk_dev_is_tray_open(blk);
2291 }
2292 
2293 void coroutine_fn blk_co_lock_medium(BlockBackend *blk, bool locked)
2294 {
2295     BlockDriverState *bs = blk_bs(blk);
2296     IO_CODE();
2297     GRAPH_RDLOCK_GUARD();
2298 
2299     if (bs) {
2300         bdrv_co_lock_medium(bs, locked);
2301     }
2302 }
2303 
2304 void coroutine_fn blk_co_eject(BlockBackend *blk, bool eject_flag)
2305 {
2306     BlockDriverState *bs = blk_bs(blk);
2307     char *id;
2308     IO_CODE();
2309     GRAPH_RDLOCK_GUARD();
2310 
2311     if (bs) {
2312         bdrv_co_eject(bs, eject_flag);
2313     }
2314 
2315     /* Whether or not we ejected on the backend,
2316      * the frontend experienced a tray event. */
2317     id = blk_get_attached_dev_id(blk);
2318     qapi_event_send_device_tray_moved(blk_name(blk), id,
2319                                       eject_flag);
2320     g_free(id);
2321 }
2322 
2323 int blk_get_flags(BlockBackend *blk)
2324 {
2325     BlockDriverState *bs = blk_bs(blk);
2326     GLOBAL_STATE_CODE();
2327 
2328     if (bs) {
2329         return bdrv_get_flags(bs);
2330     } else {
2331         return blk->root_state.open_flags;
2332     }
2333 }
2334 
2335 /* Returns the minimum request alignment, in bytes; guaranteed nonzero */
2336 uint32_t blk_get_request_alignment(BlockBackend *blk)
2337 {
2338     BlockDriverState *bs = blk_bs(blk);
2339     IO_CODE();
2340     return bs ? bs->bl.request_alignment : BDRV_SECTOR_SIZE;
2341 }
2342 
2343 /* Returns the maximum hardware transfer length, in bytes; guaranteed nonzero */
2344 uint64_t blk_get_max_hw_transfer(BlockBackend *blk)
2345 {
2346     BlockDriverState *bs = blk_bs(blk);
2347     uint64_t max = INT_MAX;
2348     IO_CODE();
2349 
2350     if (bs) {
2351         max = MIN_NON_ZERO(max, bs->bl.max_hw_transfer);
2352         max = MIN_NON_ZERO(max, bs->bl.max_transfer);
2353     }
2354     return ROUND_DOWN(max, blk_get_request_alignment(blk));
2355 }
2356 
2357 /* Returns the maximum transfer length, in bytes; guaranteed nonzero */
2358 uint32_t blk_get_max_transfer(BlockBackend *blk)
2359 {
2360     BlockDriverState *bs = blk_bs(blk);
2361     uint32_t max = INT_MAX;
2362     IO_CODE();
2363 
2364     if (bs) {
2365         max = MIN_NON_ZERO(max, bs->bl.max_transfer);
2366     }
2367     return ROUND_DOWN(max, blk_get_request_alignment(blk));
2368 }
2369 
2370 int blk_get_max_hw_iov(BlockBackend *blk)
2371 {
2372     IO_CODE();
2373     return MIN_NON_ZERO(blk->root->bs->bl.max_hw_iov,
2374                         blk->root->bs->bl.max_iov);
2375 }
2376 
2377 int blk_get_max_iov(BlockBackend *blk)
2378 {
2379     IO_CODE();
2380     return blk->root->bs->bl.max_iov;
2381 }
2382 
2383 void *blk_try_blockalign(BlockBackend *blk, size_t size)
2384 {
2385     IO_CODE();
2386     return qemu_try_blockalign(blk ? blk_bs(blk) : NULL, size);
2387 }
2388 
2389 void *blk_blockalign(BlockBackend *blk, size_t size)
2390 {
2391     IO_CODE();
2392     return qemu_blockalign(blk ? blk_bs(blk) : NULL, size);
2393 }
2394 
2395 bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp)
2396 {
2397     BlockDriverState *bs = blk_bs(blk);
2398     GLOBAL_STATE_CODE();
2399     GRAPH_RDLOCK_GUARD_MAINLOOP();
2400 
2401     if (!bs) {
2402         return false;
2403     }
2404 
2405     return bdrv_op_is_blocked(bs, op, errp);
2406 }
2407 
2408 void blk_op_unblock(BlockBackend *blk, BlockOpType op, Error *reason)
2409 {
2410     BlockDriverState *bs = blk_bs(blk);
2411     GLOBAL_STATE_CODE();
2412 
2413     if (bs) {
2414         bdrv_op_unblock(bs, op, reason);
2415     }
2416 }
2417 
2418 void blk_op_block_all(BlockBackend *blk, Error *reason)
2419 {
2420     BlockDriverState *bs = blk_bs(blk);
2421     GLOBAL_STATE_CODE();
2422 
2423     if (bs) {
2424         bdrv_op_block_all(bs, reason);
2425     }
2426 }
2427 
2428 void blk_op_unblock_all(BlockBackend *blk, Error *reason)
2429 {
2430     BlockDriverState *bs = blk_bs(blk);
2431     GLOBAL_STATE_CODE();
2432 
2433     if (bs) {
2434         bdrv_op_unblock_all(bs, reason);
2435     }
2436 }
2437 
2438 AioContext *blk_get_aio_context(BlockBackend *blk)
2439 {
2440     BlockDriverState *bs;
2441     IO_CODE();
2442 
2443     if (!blk) {
2444         return qemu_get_aio_context();
2445     }
2446 
2447     bs = blk_bs(blk);
2448     if (bs) {
2449         AioContext *ctx = bdrv_get_aio_context(blk_bs(blk));
2450         assert(ctx == blk->ctx);
2451     }
2452 
2453     return blk->ctx;
2454 }
2455 
2456 int blk_set_aio_context(BlockBackend *blk, AioContext *new_context,
2457                         Error **errp)
2458 {
2459     bool old_allow_change;
2460     BlockDriverState *bs = blk_bs(blk);
2461     int ret;
2462 
2463     GLOBAL_STATE_CODE();
2464 
2465     if (!bs) {
2466         blk->ctx = new_context;
2467         return 0;
2468     }
2469 
2470     bdrv_ref(bs);
2471 
2472     old_allow_change = blk->allow_aio_context_change;
2473     blk->allow_aio_context_change = true;
2474 
2475     ret = bdrv_try_change_aio_context(bs, new_context, NULL, errp);
2476 
2477     blk->allow_aio_context_change = old_allow_change;
2478 
2479     bdrv_unref(bs);
2480     return ret;
2481 }
2482 
2483 typedef struct BdrvStateBlkRootContext {
2484     AioContext *new_ctx;
2485     BlockBackend *blk;
2486 } BdrvStateBlkRootContext;
2487 
2488 static void blk_root_set_aio_ctx_commit(void *opaque)
2489 {
2490     BdrvStateBlkRootContext *s = opaque;
2491     BlockBackend *blk = s->blk;
2492     AioContext *new_context = s->new_ctx;
2493     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
2494 
2495     blk->ctx = new_context;
2496     if (tgm->throttle_state) {
2497         throttle_group_detach_aio_context(tgm);
2498         throttle_group_attach_aio_context(tgm, new_context);
2499     }
2500 }
2501 
2502 static TransactionActionDrv set_blk_root_context = {
2503     .commit = blk_root_set_aio_ctx_commit,
2504     .clean = g_free,
2505 };
2506 
2507 static bool blk_root_change_aio_ctx(BdrvChild *child, AioContext *ctx,
2508                                     GHashTable *visited, Transaction *tran,
2509                                     Error **errp)
2510 {
2511     BlockBackend *blk = child->opaque;
2512     BdrvStateBlkRootContext *s;
2513 
2514     if (!blk->allow_aio_context_change) {
2515         /*
2516          * Manually created BlockBackends (those with a name) that are not
2517          * attached to anything can change their AioContext without updating
2518          * their user; return an error for others.
2519          */
2520         if (!blk->name || blk->dev) {
2521             /* TODO Add BB name/QOM path */
2522             error_setg(errp, "Cannot change iothread of active block backend");
2523             return false;
2524         }
2525     }
2526 
2527     s = g_new(BdrvStateBlkRootContext, 1);
2528     *s = (BdrvStateBlkRootContext) {
2529         .new_ctx = ctx,
2530         .blk = blk,
2531     };
2532 
2533     tran_add(tran, &set_blk_root_context, s);
2534     return true;
2535 }
2536 
2537 void blk_add_aio_context_notifier(BlockBackend *blk,
2538         void (*attached_aio_context)(AioContext *new_context, void *opaque),
2539         void (*detach_aio_context)(void *opaque), void *opaque)
2540 {
2541     BlockBackendAioNotifier *notifier;
2542     BlockDriverState *bs = blk_bs(blk);
2543     GLOBAL_STATE_CODE();
2544 
2545     notifier = g_new(BlockBackendAioNotifier, 1);
2546     notifier->attached_aio_context = attached_aio_context;
2547     notifier->detach_aio_context = detach_aio_context;
2548     notifier->opaque = opaque;
2549     QLIST_INSERT_HEAD(&blk->aio_notifiers, notifier, list);
2550 
2551     if (bs) {
2552         bdrv_add_aio_context_notifier(bs, attached_aio_context,
2553                                       detach_aio_context, opaque);
2554     }
2555 }
2556 
2557 void blk_remove_aio_context_notifier(BlockBackend *blk,
2558                                      void (*attached_aio_context)(AioContext *,
2559                                                                   void *),
2560                                      void (*detach_aio_context)(void *),
2561                                      void *opaque)
2562 {
2563     BlockBackendAioNotifier *notifier;
2564     BlockDriverState *bs = blk_bs(blk);
2565 
2566     GLOBAL_STATE_CODE();
2567 
2568     if (bs) {
2569         bdrv_remove_aio_context_notifier(bs, attached_aio_context,
2570                                          detach_aio_context, opaque);
2571     }
2572 
2573     QLIST_FOREACH(notifier, &blk->aio_notifiers, list) {
2574         if (notifier->attached_aio_context == attached_aio_context &&
2575             notifier->detach_aio_context == detach_aio_context &&
2576             notifier->opaque == opaque) {
2577             QLIST_REMOVE(notifier, list);
2578             g_free(notifier);
2579             return;
2580         }
2581     }
2582 
2583     abort();
2584 }
2585 
2586 void blk_add_remove_bs_notifier(BlockBackend *blk, Notifier *notify)
2587 {
2588     GLOBAL_STATE_CODE();
2589     notifier_list_add(&blk->remove_bs_notifiers, notify);
2590 }
2591 
2592 void blk_add_insert_bs_notifier(BlockBackend *blk, Notifier *notify)
2593 {
2594     GLOBAL_STATE_CODE();
2595     notifier_list_add(&blk->insert_bs_notifiers, notify);
2596 }
2597 
2598 BlockAcctStats *blk_get_stats(BlockBackend *blk)
2599 {
2600     IO_CODE();
2601     return &blk->stats;
2602 }
2603 
2604 void *blk_aio_get(const AIOCBInfo *aiocb_info, BlockBackend *blk,
2605                   BlockCompletionFunc *cb, void *opaque)
2606 {
2607     IO_CODE();
2608     return qemu_aio_get(aiocb_info, blk_bs(blk), cb, opaque);
2609 }
2610 
2611 int coroutine_fn blk_co_pwrite_zeroes(BlockBackend *blk, int64_t offset,
2612                                       int64_t bytes, BdrvRequestFlags flags)
2613 {
2614     IO_OR_GS_CODE();
2615     return blk_co_pwritev(blk, offset, bytes, NULL,
2616                           flags | BDRV_REQ_ZERO_WRITE);
2617 }
2618 
2619 int coroutine_fn blk_co_pwrite_compressed(BlockBackend *blk, int64_t offset,
2620                                           int64_t bytes, const void *buf)
2621 {
2622     QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
2623     IO_OR_GS_CODE();
2624     return blk_co_pwritev_part(blk, offset, bytes, &qiov, 0,
2625                                BDRV_REQ_WRITE_COMPRESSED);
2626 }
2627 
2628 int coroutine_fn blk_co_truncate(BlockBackend *blk, int64_t offset, bool exact,
2629                                  PreallocMode prealloc, BdrvRequestFlags flags,
2630                                  Error **errp)
2631 {
2632     IO_OR_GS_CODE();
2633     GRAPH_RDLOCK_GUARD();
2634     if (!blk_co_is_available(blk)) {
2635         error_setg(errp, "No medium inserted");
2636         return -ENOMEDIUM;
2637     }
2638 
2639     return bdrv_co_truncate(blk->root, offset, exact, prealloc, flags, errp);
2640 }
2641 
2642 int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
2643                      int64_t pos, int size)
2644 {
2645     int ret;
2646     GLOBAL_STATE_CODE();
2647 
2648     if (!blk_is_available(blk)) {
2649         return -ENOMEDIUM;
2650     }
2651 
2652     ret = bdrv_save_vmstate(blk_bs(blk), buf, pos, size);
2653     if (ret < 0) {
2654         return ret;
2655     }
2656 
2657     if (ret == size && !blk->enable_write_cache) {
2658         ret = bdrv_flush(blk_bs(blk));
2659     }
2660 
2661     return ret < 0 ? ret : size;
2662 }
2663 
2664 int blk_load_vmstate(BlockBackend *blk, uint8_t *buf, int64_t pos, int size)
2665 {
2666     GLOBAL_STATE_CODE();
2667     if (!blk_is_available(blk)) {
2668         return -ENOMEDIUM;
2669     }
2670 
2671     return bdrv_load_vmstate(blk_bs(blk), buf, pos, size);
2672 }
2673 
2674 int blk_probe_blocksizes(BlockBackend *blk, BlockSizes *bsz)
2675 {
2676     GLOBAL_STATE_CODE();
2677     GRAPH_RDLOCK_GUARD_MAINLOOP();
2678 
2679     if (!blk_is_available(blk)) {
2680         return -ENOMEDIUM;
2681     }
2682 
2683     return bdrv_probe_blocksizes(blk_bs(blk), bsz);
2684 }
2685 
2686 int blk_probe_geometry(BlockBackend *blk, HDGeometry *geo)
2687 {
2688     GLOBAL_STATE_CODE();
2689     if (!blk_is_available(blk)) {
2690         return -ENOMEDIUM;
2691     }
2692 
2693     return bdrv_probe_geometry(blk_bs(blk), geo);
2694 }
2695 
2696 /*
2697  * Updates the BlockBackendRootState object with data from the currently
2698  * attached BlockDriverState.
2699  */
2700 void blk_update_root_state(BlockBackend *blk)
2701 {
2702     GLOBAL_STATE_CODE();
2703     assert(blk->root);
2704 
2705     blk->root_state.open_flags    = blk->root->bs->open_flags;
2706     blk->root_state.detect_zeroes = blk->root->bs->detect_zeroes;
2707 }
2708 
2709 /*
2710  * Returns the detect-zeroes setting to be used for bdrv_open() of a
2711  * BlockDriverState which is supposed to inherit the root state.
2712  */
2713 bool blk_get_detect_zeroes_from_root_state(BlockBackend *blk)
2714 {
2715     GLOBAL_STATE_CODE();
2716     return blk->root_state.detect_zeroes;
2717 }
2718 
2719 /*
2720  * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
2721  * supposed to inherit the root state.
2722  */
2723 int blk_get_open_flags_from_root_state(BlockBackend *blk)
2724 {
2725     GLOBAL_STATE_CODE();
2726     return blk->root_state.open_flags;
2727 }
2728 
2729 BlockBackendRootState *blk_get_root_state(BlockBackend *blk)
2730 {
2731     GLOBAL_STATE_CODE();
2732     return &blk->root_state;
2733 }
2734 
2735 int blk_commit_all(void)
2736 {
2737     BlockBackend *blk = NULL;
2738     GLOBAL_STATE_CODE();
2739     GRAPH_RDLOCK_GUARD_MAINLOOP();
2740 
2741     while ((blk = blk_all_next(blk)) != NULL) {
2742         AioContext *aio_context = blk_get_aio_context(blk);
2743         BlockDriverState *unfiltered_bs = bdrv_skip_filters(blk_bs(blk));
2744 
2745         aio_context_acquire(aio_context);
2746         if (blk_is_inserted(blk) && bdrv_cow_child(unfiltered_bs)) {
2747             int ret;
2748 
2749             ret = bdrv_commit(unfiltered_bs);
2750             if (ret < 0) {
2751                 aio_context_release(aio_context);
2752                 return ret;
2753             }
2754         }
2755         aio_context_release(aio_context);
2756     }
2757     return 0;
2758 }
2759 
2760 
2761 /* throttling disk I/O limits */
2762 void blk_set_io_limits(BlockBackend *blk, ThrottleConfig *cfg)
2763 {
2764     GLOBAL_STATE_CODE();
2765     throttle_group_config(&blk->public.throttle_group_member, cfg);
2766 }
2767 
2768 void blk_io_limits_disable(BlockBackend *blk)
2769 {
2770     BlockDriverState *bs = blk_bs(blk);
2771     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
2772     assert(tgm->throttle_state);
2773     GLOBAL_STATE_CODE();
2774     if (bs) {
2775         bdrv_ref(bs);
2776         bdrv_drained_begin(bs);
2777     }
2778     throttle_group_unregister_tgm(tgm);
2779     if (bs) {
2780         bdrv_drained_end(bs);
2781         bdrv_unref(bs);
2782     }
2783 }
2784 
2785 /* should be called before blk_set_io_limits if a limit is set */
2786 void blk_io_limits_enable(BlockBackend *blk, const char *group)
2787 {
2788     assert(!blk->public.throttle_group_member.throttle_state);
2789     GLOBAL_STATE_CODE();
2790     throttle_group_register_tgm(&blk->public.throttle_group_member,
2791                                 group, blk_get_aio_context(blk));
2792 }
2793 
2794 void blk_io_limits_update_group(BlockBackend *blk, const char *group)
2795 {
2796     GLOBAL_STATE_CODE();
2797     /* this BB is not part of any group */
2798     if (!blk->public.throttle_group_member.throttle_state) {
2799         return;
2800     }
2801 
2802     /* this BB is a part of the same group than the one we want */
2803     if (!g_strcmp0(throttle_group_get_name(&blk->public.throttle_group_member),
2804                 group)) {
2805         return;
2806     }
2807 
2808     /* need to change the group this bs belong to */
2809     blk_io_limits_disable(blk);
2810     blk_io_limits_enable(blk, group);
2811 }
2812 
2813 static void blk_root_drained_begin(BdrvChild *child)
2814 {
2815     BlockBackend *blk = child->opaque;
2816     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
2817 
2818     if (qatomic_fetch_inc(&blk->quiesce_counter) == 0) {
2819         if (blk->dev_ops && blk->dev_ops->drained_begin) {
2820             blk->dev_ops->drained_begin(blk->dev_opaque);
2821         }
2822     }
2823 
2824     /* Note that blk->root may not be accessible here yet if we are just
2825      * attaching to a BlockDriverState that is drained. Use child instead. */
2826 
2827     if (qatomic_fetch_inc(&tgm->io_limits_disabled) == 0) {
2828         throttle_group_restart_tgm(tgm);
2829     }
2830 }
2831 
2832 static bool blk_root_drained_poll(BdrvChild *child)
2833 {
2834     BlockBackend *blk = child->opaque;
2835     bool busy = false;
2836     assert(qatomic_read(&blk->quiesce_counter));
2837 
2838     if (blk->dev_ops && blk->dev_ops->drained_poll) {
2839         busy = blk->dev_ops->drained_poll(blk->dev_opaque);
2840     }
2841     return busy || !!blk->in_flight;
2842 }
2843 
2844 static void blk_root_drained_end(BdrvChild *child)
2845 {
2846     BlockBackend *blk = child->opaque;
2847     assert(qatomic_read(&blk->quiesce_counter));
2848 
2849     assert(blk->public.throttle_group_member.io_limits_disabled);
2850     qatomic_dec(&blk->public.throttle_group_member.io_limits_disabled);
2851 
2852     if (qatomic_fetch_dec(&blk->quiesce_counter) == 1) {
2853         if (blk->dev_ops && blk->dev_ops->drained_end) {
2854             blk->dev_ops->drained_end(blk->dev_opaque);
2855         }
2856         qemu_mutex_lock(&blk->queued_requests_lock);
2857         while (qemu_co_enter_next(&blk->queued_requests,
2858                                   &blk->queued_requests_lock)) {
2859             /* Resume all queued requests */
2860         }
2861         qemu_mutex_unlock(&blk->queued_requests_lock);
2862     }
2863 }
2864 
2865 bool blk_register_buf(BlockBackend *blk, void *host, size_t size, Error **errp)
2866 {
2867     BlockDriverState *bs = blk_bs(blk);
2868 
2869     GLOBAL_STATE_CODE();
2870 
2871     if (bs) {
2872         return bdrv_register_buf(bs, host, size, errp);
2873     }
2874     return true;
2875 }
2876 
2877 void blk_unregister_buf(BlockBackend *blk, void *host, size_t size)
2878 {
2879     BlockDriverState *bs = blk_bs(blk);
2880 
2881     GLOBAL_STATE_CODE();
2882 
2883     if (bs) {
2884         bdrv_unregister_buf(bs, host, size);
2885     }
2886 }
2887 
2888 int coroutine_fn blk_co_copy_range(BlockBackend *blk_in, int64_t off_in,
2889                                    BlockBackend *blk_out, int64_t off_out,
2890                                    int64_t bytes, BdrvRequestFlags read_flags,
2891                                    BdrvRequestFlags write_flags)
2892 {
2893     int r;
2894     IO_CODE();
2895     GRAPH_RDLOCK_GUARD();
2896 
2897     r = blk_check_byte_request(blk_in, off_in, bytes);
2898     if (r) {
2899         return r;
2900     }
2901     r = blk_check_byte_request(blk_out, off_out, bytes);
2902     if (r) {
2903         return r;
2904     }
2905 
2906     return bdrv_co_copy_range(blk_in->root, off_in,
2907                               blk_out->root, off_out,
2908                               bytes, read_flags, write_flags);
2909 }
2910 
2911 const BdrvChild *blk_root(BlockBackend *blk)
2912 {
2913     GLOBAL_STATE_CODE();
2914     return blk->root;
2915 }
2916 
2917 int blk_make_empty(BlockBackend *blk, Error **errp)
2918 {
2919     GLOBAL_STATE_CODE();
2920     GRAPH_RDLOCK_GUARD_MAINLOOP();
2921 
2922     if (!blk_is_available(blk)) {
2923         error_setg(errp, "No medium inserted");
2924         return -ENOMEDIUM;
2925     }
2926 
2927     return bdrv_make_empty(blk->root, errp);
2928 }
2929