xref: /openbmc/qemu/block/block-backend.c (revision 646b5378)
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; /* access with atomic operations only */
48     DriveInfo *legacy_dinfo;    /* null unless created by drive_new() */
49     QTAILQ_ENTRY(BlockBackend) link;         /* for block_backends */
50     QTAILQ_ENTRY(BlockBackend) monitor_link; /* for monitor_block_backends */
51     BlockBackendPublic public;
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 BlockBackend *blk_new_with_bs(BlockDriverState *bs, uint64_t perm,
395                               uint64_t shared_perm, Error **errp)
396 {
397     BlockBackend *blk = blk_new(bdrv_get_aio_context(bs), perm, shared_perm);
398 
399     GLOBAL_STATE_CODE();
400 
401     if (blk_insert_bs(blk, bs, errp) < 0) {
402         blk_unref(blk);
403         return NULL;
404     }
405     return blk;
406 }
407 
408 /*
409  * Creates a new BlockBackend, opens a new BlockDriverState, and connects both.
410  * By default, the new BlockBackend is in the main AioContext, but if the
411  * parameters connect it with any existing node in a different AioContext, it
412  * may end up there instead.
413  *
414  * Just as with bdrv_open(), after having called this function the reference to
415  * @options belongs to the block layer (even on failure).
416  *
417  * TODO: Remove @filename and @flags; it should be possible to specify a whole
418  * BDS tree just by specifying the @options QDict (or @reference,
419  * alternatively). At the time of adding this function, this is not possible,
420  * though, so callers of this function have to be able to specify @filename and
421  * @flags.
422  */
423 BlockBackend *blk_new_open(const char *filename, const char *reference,
424                            QDict *options, int flags, Error **errp)
425 {
426     BlockBackend *blk;
427     BlockDriverState *bs;
428     uint64_t perm = 0;
429     uint64_t shared = BLK_PERM_ALL;
430 
431     GLOBAL_STATE_CODE();
432 
433     /*
434      * blk_new_open() is mainly used in .bdrv_create implementations and the
435      * tools where sharing isn't a major concern because the BDS stays private
436      * and the file is generally not supposed to be used by a second process,
437      * so we just request permission according to the flags.
438      *
439      * The exceptions are xen_disk and blockdev_init(); in these cases, the
440      * caller of blk_new_open() doesn't make use of the permissions, but they
441      * shouldn't hurt either. We can still share everything here because the
442      * guest devices will add their own blockers if they can't share.
443      */
444     if ((flags & BDRV_O_NO_IO) == 0) {
445         perm |= BLK_PERM_CONSISTENT_READ;
446         if (flags & BDRV_O_RDWR) {
447             perm |= BLK_PERM_WRITE;
448         }
449     }
450     if (flags & BDRV_O_RESIZE) {
451         perm |= BLK_PERM_RESIZE;
452     }
453     if (flags & BDRV_O_NO_SHARE) {
454         shared = BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
455     }
456 
457     bs = bdrv_open(filename, reference, options, flags, errp);
458     if (!bs) {
459         return NULL;
460     }
461 
462     /* bdrv_open() could have moved bs to a different AioContext */
463     blk = blk_new(bdrv_get_aio_context(bs), perm, shared);
464     blk->perm = perm;
465     blk->shared_perm = shared;
466 
467     blk_insert_bs(blk, bs, errp);
468     bdrv_unref(bs);
469 
470     if (!blk->root) {
471         blk_unref(blk);
472         return NULL;
473     }
474 
475     return blk;
476 }
477 
478 static void blk_delete(BlockBackend *blk)
479 {
480     assert(!blk->refcnt);
481     assert(!blk->name);
482     assert(!blk->dev);
483     if (blk->public.throttle_group_member.throttle_state) {
484         blk_io_limits_disable(blk);
485     }
486     if (blk->root) {
487         blk_remove_bs(blk);
488     }
489     if (blk->vmsh) {
490         qemu_del_vm_change_state_handler(blk->vmsh);
491         blk->vmsh = NULL;
492     }
493     assert(QLIST_EMPTY(&blk->remove_bs_notifiers.notifiers));
494     assert(QLIST_EMPTY(&blk->insert_bs_notifiers.notifiers));
495     assert(QLIST_EMPTY(&blk->aio_notifiers));
496     assert(qemu_co_queue_empty(&blk->queued_requests));
497     qemu_mutex_destroy(&blk->queued_requests_lock);
498     QTAILQ_REMOVE(&block_backends, blk, link);
499     drive_info_del(blk->legacy_dinfo);
500     block_acct_cleanup(&blk->stats);
501     g_free(blk);
502 }
503 
504 static void drive_info_del(DriveInfo *dinfo)
505 {
506     if (!dinfo) {
507         return;
508     }
509     qemu_opts_del(dinfo->opts);
510     g_free(dinfo);
511 }
512 
513 int blk_get_refcnt(BlockBackend *blk)
514 {
515     GLOBAL_STATE_CODE();
516     return blk ? blk->refcnt : 0;
517 }
518 
519 /*
520  * Increment @blk's reference count.
521  * @blk must not be null.
522  */
523 void blk_ref(BlockBackend *blk)
524 {
525     assert(blk->refcnt > 0);
526     GLOBAL_STATE_CODE();
527     blk->refcnt++;
528 }
529 
530 /*
531  * Decrement @blk's reference count.
532  * If this drops it to zero, destroy @blk.
533  * For convenience, do nothing if @blk is null.
534  */
535 void blk_unref(BlockBackend *blk)
536 {
537     GLOBAL_STATE_CODE();
538     if (blk) {
539         assert(blk->refcnt > 0);
540         if (blk->refcnt > 1) {
541             blk->refcnt--;
542         } else {
543             blk_drain(blk);
544             /* blk_drain() cannot resurrect blk, nobody held a reference */
545             assert(blk->refcnt == 1);
546             blk->refcnt = 0;
547             blk_delete(blk);
548         }
549     }
550 }
551 
552 /*
553  * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
554  * ones which are hidden (i.e. are not referenced by the monitor).
555  */
556 BlockBackend *blk_all_next(BlockBackend *blk)
557 {
558     GLOBAL_STATE_CODE();
559     return blk ? QTAILQ_NEXT(blk, link)
560                : QTAILQ_FIRST(&block_backends);
561 }
562 
563 void blk_remove_all_bs(void)
564 {
565     BlockBackend *blk = NULL;
566 
567     GLOBAL_STATE_CODE();
568 
569     while ((blk = blk_all_next(blk)) != NULL) {
570         if (blk->root) {
571             blk_remove_bs(blk);
572         }
573     }
574 }
575 
576 /*
577  * Return the monitor-owned BlockBackend after @blk.
578  * If @blk is null, return the first one.
579  * Else, return @blk's next sibling, which may be null.
580  *
581  * To iterate over all BlockBackends, do
582  * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
583  *     ...
584  * }
585  */
586 BlockBackend *blk_next(BlockBackend *blk)
587 {
588     GLOBAL_STATE_CODE();
589     return blk ? QTAILQ_NEXT(blk, monitor_link)
590                : QTAILQ_FIRST(&monitor_block_backends);
591 }
592 
593 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
594  * the monitor or attached to a BlockBackend */
595 BlockDriverState *bdrv_next(BdrvNextIterator *it)
596 {
597     BlockDriverState *bs, *old_bs;
598 
599     /* Must be called from the main loop */
600     assert(qemu_get_current_aio_context() == qemu_get_aio_context());
601 
602     old_bs = it->bs;
603 
604     /* First, return all root nodes of BlockBackends. In order to avoid
605      * returning a BDS twice when multiple BBs refer to it, we only return it
606      * if the BB is the first one in the parent list of the BDS. */
607     if (it->phase == BDRV_NEXT_BACKEND_ROOTS) {
608         BlockBackend *old_blk = it->blk;
609 
610         do {
611             it->blk = blk_all_next(it->blk);
612             bs = it->blk ? blk_bs(it->blk) : NULL;
613         } while (it->blk && (bs == NULL || bdrv_first_blk(bs) != it->blk));
614 
615         if (it->blk) {
616             blk_ref(it->blk);
617         }
618         blk_unref(old_blk);
619 
620         if (bs) {
621             bdrv_ref(bs);
622             bdrv_unref(old_bs);
623             it->bs = bs;
624             return bs;
625         }
626         it->phase = BDRV_NEXT_MONITOR_OWNED;
627     }
628 
629     /* Then return the monitor-owned BDSes without a BB attached. Ignore all
630      * BDSes that are attached to a BlockBackend here; they have been handled
631      * by the above block already */
632     do {
633         it->bs = bdrv_next_monitor_owned(it->bs);
634         bs = it->bs;
635     } while (bs && bdrv_has_blk(bs));
636 
637     if (bs) {
638         bdrv_ref(bs);
639     }
640     bdrv_unref(old_bs);
641 
642     return bs;
643 }
644 
645 static void bdrv_next_reset(BdrvNextIterator *it)
646 {
647     *it = (BdrvNextIterator) {
648         .phase = BDRV_NEXT_BACKEND_ROOTS,
649     };
650 }
651 
652 BlockDriverState *bdrv_first(BdrvNextIterator *it)
653 {
654     GLOBAL_STATE_CODE();
655     bdrv_next_reset(it);
656     return bdrv_next(it);
657 }
658 
659 /* Must be called when aborting a bdrv_next() iteration before
660  * bdrv_next() returns NULL */
661 void bdrv_next_cleanup(BdrvNextIterator *it)
662 {
663     /* Must be called from the main loop */
664     assert(qemu_get_current_aio_context() == qemu_get_aio_context());
665 
666     bdrv_unref(it->bs);
667 
668     if (it->phase == BDRV_NEXT_BACKEND_ROOTS && it->blk) {
669         blk_unref(it->blk);
670     }
671 
672     bdrv_next_reset(it);
673 }
674 
675 /*
676  * Add a BlockBackend into the list of backends referenced by the monitor, with
677  * the given @name acting as the handle for the monitor.
678  * Strictly for use by blockdev.c.
679  *
680  * @name must not be null or empty.
681  *
682  * Returns true on success and false on failure. In the latter case, an Error
683  * object is returned through @errp.
684  */
685 bool monitor_add_blk(BlockBackend *blk, const char *name, Error **errp)
686 {
687     assert(!blk->name);
688     assert(name && name[0]);
689     GLOBAL_STATE_CODE();
690 
691     if (!id_wellformed(name)) {
692         error_setg(errp, "Invalid device name");
693         return false;
694     }
695     if (blk_by_name(name)) {
696         error_setg(errp, "Device with id '%s' already exists", name);
697         return false;
698     }
699     if (bdrv_find_node(name)) {
700         error_setg(errp,
701                    "Device name '%s' conflicts with an existing node name",
702                    name);
703         return false;
704     }
705 
706     blk->name = g_strdup(name);
707     QTAILQ_INSERT_TAIL(&monitor_block_backends, blk, monitor_link);
708     return true;
709 }
710 
711 /*
712  * Remove a BlockBackend from the list of backends referenced by the monitor.
713  * Strictly for use by blockdev.c.
714  */
715 void monitor_remove_blk(BlockBackend *blk)
716 {
717     GLOBAL_STATE_CODE();
718 
719     if (!blk->name) {
720         return;
721     }
722 
723     QTAILQ_REMOVE(&monitor_block_backends, blk, monitor_link);
724     g_free(blk->name);
725     blk->name = NULL;
726 }
727 
728 /*
729  * Return @blk's name, a non-null string.
730  * Returns an empty string iff @blk is not referenced by the monitor.
731  */
732 const char *blk_name(const BlockBackend *blk)
733 {
734     IO_CODE();
735     return blk->name ?: "";
736 }
737 
738 /*
739  * Return the BlockBackend with name @name if it exists, else null.
740  * @name must not be null.
741  */
742 BlockBackend *blk_by_name(const char *name)
743 {
744     BlockBackend *blk = NULL;
745 
746     GLOBAL_STATE_CODE();
747     assert(name);
748     while ((blk = blk_next(blk)) != NULL) {
749         if (!strcmp(name, blk->name)) {
750             return blk;
751         }
752     }
753     return NULL;
754 }
755 
756 /*
757  * Return the BlockDriverState attached to @blk if any, else null.
758  */
759 BlockDriverState *blk_bs(BlockBackend *blk)
760 {
761     IO_CODE();
762     return blk->root ? blk->root->bs : NULL;
763 }
764 
765 static BlockBackend * GRAPH_RDLOCK bdrv_first_blk(BlockDriverState *bs)
766 {
767     BdrvChild *child;
768 
769     GLOBAL_STATE_CODE();
770     assert_bdrv_graph_readable();
771 
772     QLIST_FOREACH(child, &bs->parents, next_parent) {
773         if (child->klass == &child_root) {
774             return child->opaque;
775         }
776     }
777 
778     return NULL;
779 }
780 
781 /*
782  * Returns true if @bs has an associated BlockBackend.
783  */
784 bool bdrv_has_blk(BlockDriverState *bs)
785 {
786     GLOBAL_STATE_CODE();
787     return bdrv_first_blk(bs) != NULL;
788 }
789 
790 /*
791  * Returns true if @bs has only BlockBackends as parents.
792  */
793 bool bdrv_is_root_node(BlockDriverState *bs)
794 {
795     BdrvChild *c;
796 
797     GLOBAL_STATE_CODE();
798     assert_bdrv_graph_readable();
799 
800     QLIST_FOREACH(c, &bs->parents, next_parent) {
801         if (c->klass != &child_root) {
802             return false;
803         }
804     }
805 
806     return true;
807 }
808 
809 /*
810  * Return @blk's DriveInfo if any, else null.
811  */
812 DriveInfo *blk_legacy_dinfo(BlockBackend *blk)
813 {
814     GLOBAL_STATE_CODE();
815     return blk->legacy_dinfo;
816 }
817 
818 /*
819  * Set @blk's DriveInfo to @dinfo, and return it.
820  * @blk must not have a DriveInfo set already.
821  * No other BlockBackend may have the same DriveInfo set.
822  */
823 DriveInfo *blk_set_legacy_dinfo(BlockBackend *blk, DriveInfo *dinfo)
824 {
825     assert(!blk->legacy_dinfo);
826     GLOBAL_STATE_CODE();
827     return blk->legacy_dinfo = dinfo;
828 }
829 
830 /*
831  * Return the BlockBackend with DriveInfo @dinfo.
832  * It must exist.
833  */
834 BlockBackend *blk_by_legacy_dinfo(DriveInfo *dinfo)
835 {
836     BlockBackend *blk = NULL;
837     GLOBAL_STATE_CODE();
838 
839     while ((blk = blk_next(blk)) != NULL) {
840         if (blk->legacy_dinfo == dinfo) {
841             return blk;
842         }
843     }
844     abort();
845 }
846 
847 /*
848  * Returns a pointer to the publicly accessible fields of @blk.
849  */
850 BlockBackendPublic *blk_get_public(BlockBackend *blk)
851 {
852     GLOBAL_STATE_CODE();
853     return &blk->public;
854 }
855 
856 /*
857  * Disassociates the currently associated BlockDriverState from @blk.
858  */
859 void blk_remove_bs(BlockBackend *blk)
860 {
861     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
862     BdrvChild *root;
863 
864     GLOBAL_STATE_CODE();
865 
866     notifier_list_notify(&blk->remove_bs_notifiers, blk);
867     if (tgm->throttle_state) {
868         BlockDriverState *bs = blk_bs(blk);
869 
870         /*
871          * Take a ref in case blk_bs() changes across bdrv_drained_begin(), for
872          * example, if a temporary filter node is removed by a blockjob.
873          */
874         bdrv_ref(bs);
875         bdrv_drained_begin(bs);
876         throttle_group_detach_aio_context(tgm);
877         throttle_group_attach_aio_context(tgm, qemu_get_aio_context());
878         bdrv_drained_end(bs);
879         bdrv_unref(bs);
880     }
881 
882     blk_update_root_state(blk);
883 
884     /* bdrv_root_unref_child() will cause blk->root to become stale and may
885      * switch to a completion coroutine later on. Let's drain all I/O here
886      * to avoid that and a potential QEMU crash.
887      */
888     blk_drain(blk);
889     root = blk->root;
890     blk->root = NULL;
891 
892     bdrv_graph_wrlock();
893     bdrv_root_unref_child(root);
894     bdrv_graph_wrunlock();
895 }
896 
897 /*
898  * Associates a new BlockDriverState with @blk.
899  */
900 int blk_insert_bs(BlockBackend *blk, BlockDriverState *bs, Error **errp)
901 {
902     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
903 
904     GLOBAL_STATE_CODE();
905     bdrv_ref(bs);
906     bdrv_graph_wrlock();
907     blk->root = bdrv_root_attach_child(bs, "root", &child_root,
908                                        BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY,
909                                        blk->perm, blk->shared_perm,
910                                        blk, errp);
911     bdrv_graph_wrunlock();
912     if (blk->root == NULL) {
913         return -EPERM;
914     }
915 
916     notifier_list_notify(&blk->insert_bs_notifiers, blk);
917     if (tgm->throttle_state) {
918         throttle_group_detach_aio_context(tgm);
919         throttle_group_attach_aio_context(tgm, bdrv_get_aio_context(bs));
920     }
921 
922     return 0;
923 }
924 
925 /*
926  * Change BlockDriverState associated with @blk.
927  */
928 int blk_replace_bs(BlockBackend *blk, BlockDriverState *new_bs, Error **errp)
929 {
930     GLOBAL_STATE_CODE();
931     return bdrv_replace_child_bs(blk->root, new_bs, errp);
932 }
933 
934 /*
935  * Sets the permission bitmasks that the user of the BlockBackend needs.
936  */
937 static int coroutine_mixed_fn GRAPH_RDLOCK
938 blk_set_perm_locked(BlockBackend *blk, uint64_t perm, uint64_t shared_perm,
939                     Error **errp)
940 {
941     int ret;
942     GLOBAL_STATE_CODE();
943 
944     if (blk->root && !blk->disable_perm) {
945         ret = bdrv_child_try_set_perm(blk->root, perm, shared_perm, errp);
946         if (ret < 0) {
947             return ret;
948         }
949     }
950 
951     blk->perm = perm;
952     blk->shared_perm = shared_perm;
953 
954     return 0;
955 }
956 
957 int blk_set_perm(BlockBackend *blk, uint64_t perm, uint64_t shared_perm,
958                  Error **errp)
959 {
960     GLOBAL_STATE_CODE();
961     GRAPH_RDLOCK_GUARD_MAINLOOP();
962 
963     return blk_set_perm_locked(blk, perm, shared_perm, errp);
964 }
965 
966 void blk_get_perm(BlockBackend *blk, uint64_t *perm, uint64_t *shared_perm)
967 {
968     GLOBAL_STATE_CODE();
969     *perm = blk->perm;
970     *shared_perm = blk->shared_perm;
971 }
972 
973 /*
974  * Attach device model @dev to @blk.
975  * Return 0 on success, -EBUSY when a device model is attached already.
976  */
977 int blk_attach_dev(BlockBackend *blk, DeviceState *dev)
978 {
979     GLOBAL_STATE_CODE();
980     if (blk->dev) {
981         return -EBUSY;
982     }
983 
984     /* While migration is still incoming, we don't need to apply the
985      * permissions of guest device BlockBackends. We might still have a block
986      * job or NBD server writing to the image for storage migration. */
987     if (runstate_check(RUN_STATE_INMIGRATE)) {
988         blk->disable_perm = true;
989     }
990 
991     blk_ref(blk);
992     blk->dev = dev;
993     blk_iostatus_reset(blk);
994 
995     return 0;
996 }
997 
998 /*
999  * Detach device model @dev from @blk.
1000  * @dev must be currently attached to @blk.
1001  */
1002 void blk_detach_dev(BlockBackend *blk, DeviceState *dev)
1003 {
1004     assert(blk->dev == dev);
1005     GLOBAL_STATE_CODE();
1006     blk->dev = NULL;
1007     blk->dev_ops = NULL;
1008     blk->dev_opaque = NULL;
1009     blk_set_perm(blk, 0, BLK_PERM_ALL, &error_abort);
1010     blk_unref(blk);
1011 }
1012 
1013 /*
1014  * Return the device model attached to @blk if any, else null.
1015  */
1016 DeviceState *blk_get_attached_dev(BlockBackend *blk)
1017 {
1018     GLOBAL_STATE_CODE();
1019     return blk->dev;
1020 }
1021 
1022 static char *blk_get_attached_dev_id_or_path(BlockBackend *blk, bool want_id)
1023 {
1024     DeviceState *dev = blk->dev;
1025     IO_CODE();
1026 
1027     if (!dev) {
1028         return g_strdup("");
1029     } else if (want_id && dev->id) {
1030         return g_strdup(dev->id);
1031     }
1032 
1033     return object_get_canonical_path(OBJECT(dev)) ?: g_strdup("");
1034 }
1035 
1036 /*
1037  * Return the qdev ID, or if no ID is assigned the QOM path, of the block
1038  * device attached to the BlockBackend.
1039  */
1040 char *blk_get_attached_dev_id(BlockBackend *blk)
1041 {
1042     return blk_get_attached_dev_id_or_path(blk, true);
1043 }
1044 
1045 static char *blk_get_attached_dev_path(BlockBackend *blk)
1046 {
1047     return blk_get_attached_dev_id_or_path(blk, false);
1048 }
1049 
1050 /*
1051  * Return the BlockBackend which has the device model @dev attached if it
1052  * exists, else null.
1053  *
1054  * @dev must not be null.
1055  */
1056 BlockBackend *blk_by_dev(void *dev)
1057 {
1058     BlockBackend *blk = NULL;
1059 
1060     GLOBAL_STATE_CODE();
1061 
1062     assert(dev != NULL);
1063     while ((blk = blk_all_next(blk)) != NULL) {
1064         if (blk->dev == dev) {
1065             return blk;
1066         }
1067     }
1068     return NULL;
1069 }
1070 
1071 /*
1072  * Set @blk's device model callbacks to @ops.
1073  * @opaque is the opaque argument to pass to the callbacks.
1074  * This is for use by device models.
1075  */
1076 void blk_set_dev_ops(BlockBackend *blk, const BlockDevOps *ops,
1077                      void *opaque)
1078 {
1079     GLOBAL_STATE_CODE();
1080     blk->dev_ops = ops;
1081     blk->dev_opaque = opaque;
1082 
1083     /* Are we currently quiesced? Should we enforce this right now? */
1084     if (qatomic_read(&blk->quiesce_counter) && ops && ops->drained_begin) {
1085         ops->drained_begin(opaque);
1086     }
1087 }
1088 
1089 /*
1090  * Notify @blk's attached device model of media change.
1091  *
1092  * If @load is true, notify of media load. This action can fail, meaning that
1093  * the medium cannot be loaded. @errp is set then.
1094  *
1095  * If @load is false, notify of media eject. This can never fail.
1096  *
1097  * Also send DEVICE_TRAY_MOVED events as appropriate.
1098  */
1099 void blk_dev_change_media_cb(BlockBackend *blk, bool load, Error **errp)
1100 {
1101     GLOBAL_STATE_CODE();
1102     if (blk->dev_ops && blk->dev_ops->change_media_cb) {
1103         bool tray_was_open, tray_is_open;
1104         Error *local_err = NULL;
1105 
1106         tray_was_open = blk_dev_is_tray_open(blk);
1107         blk->dev_ops->change_media_cb(blk->dev_opaque, load, &local_err);
1108         if (local_err) {
1109             assert(load == true);
1110             error_propagate(errp, local_err);
1111             return;
1112         }
1113         tray_is_open = blk_dev_is_tray_open(blk);
1114 
1115         if (tray_was_open != tray_is_open) {
1116             char *id = blk_get_attached_dev_id(blk);
1117             qapi_event_send_device_tray_moved(blk_name(blk), id, tray_is_open);
1118             g_free(id);
1119         }
1120     }
1121 }
1122 
1123 static void blk_root_change_media(BdrvChild *child, bool load)
1124 {
1125     blk_dev_change_media_cb(child->opaque, load, NULL);
1126 }
1127 
1128 /*
1129  * Does @blk's attached device model have removable media?
1130  * %true if no device model is attached.
1131  */
1132 bool blk_dev_has_removable_media(BlockBackend *blk)
1133 {
1134     GLOBAL_STATE_CODE();
1135     return !blk->dev || (blk->dev_ops && blk->dev_ops->change_media_cb);
1136 }
1137 
1138 /*
1139  * Does @blk's attached device model have a tray?
1140  */
1141 bool blk_dev_has_tray(BlockBackend *blk)
1142 {
1143     IO_CODE();
1144     return blk->dev_ops && blk->dev_ops->is_tray_open;
1145 }
1146 
1147 /*
1148  * Notify @blk's attached device model of a media eject request.
1149  * If @force is true, the medium is about to be yanked out forcefully.
1150  */
1151 void blk_dev_eject_request(BlockBackend *blk, bool force)
1152 {
1153     GLOBAL_STATE_CODE();
1154     if (blk->dev_ops && blk->dev_ops->eject_request_cb) {
1155         blk->dev_ops->eject_request_cb(blk->dev_opaque, force);
1156     }
1157 }
1158 
1159 /*
1160  * Does @blk's attached device model have a tray, and is it open?
1161  */
1162 bool blk_dev_is_tray_open(BlockBackend *blk)
1163 {
1164     IO_CODE();
1165     if (blk_dev_has_tray(blk)) {
1166         return blk->dev_ops->is_tray_open(blk->dev_opaque);
1167     }
1168     return false;
1169 }
1170 
1171 /*
1172  * Does @blk's attached device model have the medium locked?
1173  * %false if the device model has no such lock.
1174  */
1175 bool blk_dev_is_medium_locked(BlockBackend *blk)
1176 {
1177     GLOBAL_STATE_CODE();
1178     if (blk->dev_ops && blk->dev_ops->is_medium_locked) {
1179         return blk->dev_ops->is_medium_locked(blk->dev_opaque);
1180     }
1181     return false;
1182 }
1183 
1184 /*
1185  * Notify @blk's attached device model of a backend size change.
1186  */
1187 static void blk_root_resize(BdrvChild *child)
1188 {
1189     BlockBackend *blk = child->opaque;
1190 
1191     if (blk->dev_ops && blk->dev_ops->resize_cb) {
1192         blk->dev_ops->resize_cb(blk->dev_opaque);
1193     }
1194 }
1195 
1196 void blk_iostatus_enable(BlockBackend *blk)
1197 {
1198     GLOBAL_STATE_CODE();
1199     blk->iostatus_enabled = true;
1200     blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
1201 }
1202 
1203 /* The I/O status is only enabled if the drive explicitly
1204  * enables it _and_ the VM is configured to stop on errors */
1205 bool blk_iostatus_is_enabled(const BlockBackend *blk)
1206 {
1207     IO_CODE();
1208     return (blk->iostatus_enabled &&
1209            (blk->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
1210             blk->on_write_error == BLOCKDEV_ON_ERROR_STOP   ||
1211             blk->on_read_error == BLOCKDEV_ON_ERROR_STOP));
1212 }
1213 
1214 BlockDeviceIoStatus blk_iostatus(const BlockBackend *blk)
1215 {
1216     GLOBAL_STATE_CODE();
1217     return blk->iostatus;
1218 }
1219 
1220 void blk_iostatus_reset(BlockBackend *blk)
1221 {
1222     GLOBAL_STATE_CODE();
1223     if (blk_iostatus_is_enabled(blk)) {
1224         blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
1225     }
1226 }
1227 
1228 void blk_iostatus_set_err(BlockBackend *blk, int error)
1229 {
1230     IO_CODE();
1231     assert(blk_iostatus_is_enabled(blk));
1232     if (blk->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
1233         blk->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
1234                                           BLOCK_DEVICE_IO_STATUS_FAILED;
1235     }
1236 }
1237 
1238 void blk_set_allow_write_beyond_eof(BlockBackend *blk, bool allow)
1239 {
1240     IO_CODE();
1241     blk->allow_write_beyond_eof = allow;
1242 }
1243 
1244 void blk_set_allow_aio_context_change(BlockBackend *blk, bool allow)
1245 {
1246     IO_CODE();
1247     blk->allow_aio_context_change = allow;
1248 }
1249 
1250 void blk_set_disable_request_queuing(BlockBackend *blk, bool disable)
1251 {
1252     IO_CODE();
1253     qatomic_set(&blk->disable_request_queuing, disable);
1254 }
1255 
1256 static int coroutine_fn GRAPH_RDLOCK
1257 blk_check_byte_request(BlockBackend *blk, int64_t offset, int64_t bytes)
1258 {
1259     int64_t len;
1260 
1261     if (bytes < 0) {
1262         return -EIO;
1263     }
1264 
1265     if (!blk_co_is_available(blk)) {
1266         return -ENOMEDIUM;
1267     }
1268 
1269     if (offset < 0) {
1270         return -EIO;
1271     }
1272 
1273     if (!blk->allow_write_beyond_eof) {
1274         len = bdrv_co_getlength(blk_bs(blk));
1275         if (len < 0) {
1276             return len;
1277         }
1278 
1279         if (offset > len || len - offset < bytes) {
1280             return -EIO;
1281         }
1282     }
1283 
1284     return 0;
1285 }
1286 
1287 /* Are we currently in a drained section? */
1288 bool blk_in_drain(BlockBackend *blk)
1289 {
1290     GLOBAL_STATE_CODE(); /* change to IO_OR_GS_CODE(), if necessary */
1291     return qatomic_read(&blk->quiesce_counter);
1292 }
1293 
1294 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1295 static void coroutine_fn blk_wait_while_drained(BlockBackend *blk)
1296 {
1297     assert(blk->in_flight > 0);
1298 
1299     if (qatomic_read(&blk->quiesce_counter) &&
1300         !qatomic_read(&blk->disable_request_queuing)) {
1301         /*
1302          * Take lock before decrementing in flight counter so main loop thread
1303          * waits for us to enqueue ourselves before it can leave the drained
1304          * section.
1305          */
1306         qemu_mutex_lock(&blk->queued_requests_lock);
1307         blk_dec_in_flight(blk);
1308         qemu_co_queue_wait(&blk->queued_requests, &blk->queued_requests_lock);
1309         blk_inc_in_flight(blk);
1310         qemu_mutex_unlock(&blk->queued_requests_lock);
1311     }
1312 }
1313 
1314 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1315 static int coroutine_fn
1316 blk_co_do_preadv_part(BlockBackend *blk, int64_t offset, int64_t bytes,
1317                       QEMUIOVector *qiov, size_t qiov_offset,
1318                       BdrvRequestFlags flags)
1319 {
1320     int ret;
1321     BlockDriverState *bs;
1322     IO_CODE();
1323 
1324     blk_wait_while_drained(blk);
1325     GRAPH_RDLOCK_GUARD();
1326 
1327     /* Call blk_bs() only after waiting, the graph may have changed */
1328     bs = blk_bs(blk);
1329     trace_blk_co_preadv(blk, bs, offset, bytes, flags);
1330 
1331     ret = blk_check_byte_request(blk, offset, bytes);
1332     if (ret < 0) {
1333         return ret;
1334     }
1335 
1336     bdrv_inc_in_flight(bs);
1337 
1338     /* throttling disk I/O */
1339     if (blk->public.throttle_group_member.throttle_state) {
1340         throttle_group_co_io_limits_intercept(&blk->public.throttle_group_member,
1341                 bytes, THROTTLE_READ);
1342     }
1343 
1344     ret = bdrv_co_preadv_part(blk->root, offset, bytes, qiov, qiov_offset,
1345                               flags);
1346     bdrv_dec_in_flight(bs);
1347     return ret;
1348 }
1349 
1350 int coroutine_fn blk_co_pread(BlockBackend *blk, int64_t offset, int64_t bytes,
1351                               void *buf, BdrvRequestFlags flags)
1352 {
1353     QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
1354     IO_OR_GS_CODE();
1355 
1356     assert(bytes <= SIZE_MAX);
1357 
1358     return blk_co_preadv(blk, offset, bytes, &qiov, flags);
1359 }
1360 
1361 int coroutine_fn blk_co_preadv(BlockBackend *blk, int64_t offset,
1362                                int64_t bytes, QEMUIOVector *qiov,
1363                                BdrvRequestFlags flags)
1364 {
1365     int ret;
1366     IO_OR_GS_CODE();
1367 
1368     blk_inc_in_flight(blk);
1369     ret = blk_co_do_preadv_part(blk, offset, bytes, qiov, 0, flags);
1370     blk_dec_in_flight(blk);
1371 
1372     return ret;
1373 }
1374 
1375 int coroutine_fn blk_co_preadv_part(BlockBackend *blk, int64_t offset,
1376                                     int64_t bytes, QEMUIOVector *qiov,
1377                                     size_t qiov_offset, BdrvRequestFlags flags)
1378 {
1379     int ret;
1380     IO_OR_GS_CODE();
1381 
1382     blk_inc_in_flight(blk);
1383     ret = blk_co_do_preadv_part(blk, offset, bytes, qiov, qiov_offset, flags);
1384     blk_dec_in_flight(blk);
1385 
1386     return ret;
1387 }
1388 
1389 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1390 static int coroutine_fn
1391 blk_co_do_pwritev_part(BlockBackend *blk, int64_t offset, int64_t bytes,
1392                        QEMUIOVector *qiov, size_t qiov_offset,
1393                        BdrvRequestFlags flags)
1394 {
1395     int ret;
1396     BlockDriverState *bs;
1397     IO_CODE();
1398 
1399     blk_wait_while_drained(blk);
1400     GRAPH_RDLOCK_GUARD();
1401 
1402     /* Call blk_bs() only after waiting, the graph may have changed */
1403     bs = blk_bs(blk);
1404     trace_blk_co_pwritev(blk, bs, offset, bytes, flags);
1405 
1406     ret = blk_check_byte_request(blk, offset, bytes);
1407     if (ret < 0) {
1408         return ret;
1409     }
1410 
1411     bdrv_inc_in_flight(bs);
1412     /* throttling disk I/O */
1413     if (blk->public.throttle_group_member.throttle_state) {
1414         throttle_group_co_io_limits_intercept(&blk->public.throttle_group_member,
1415                 bytes, THROTTLE_WRITE);
1416     }
1417 
1418     if (!blk->enable_write_cache) {
1419         flags |= BDRV_REQ_FUA;
1420     }
1421 
1422     ret = bdrv_co_pwritev_part(blk->root, offset, bytes, qiov, qiov_offset,
1423                                flags);
1424     bdrv_dec_in_flight(bs);
1425     return ret;
1426 }
1427 
1428 int coroutine_fn blk_co_pwritev_part(BlockBackend *blk, int64_t offset,
1429                                      int64_t bytes,
1430                                      QEMUIOVector *qiov, size_t qiov_offset,
1431                                      BdrvRequestFlags flags)
1432 {
1433     int ret;
1434     IO_OR_GS_CODE();
1435 
1436     blk_inc_in_flight(blk);
1437     ret = blk_co_do_pwritev_part(blk, offset, bytes, qiov, qiov_offset, flags);
1438     blk_dec_in_flight(blk);
1439 
1440     return ret;
1441 }
1442 
1443 int coroutine_fn blk_co_pwrite(BlockBackend *blk, int64_t offset, int64_t bytes,
1444                                const void *buf, BdrvRequestFlags flags)
1445 {
1446     QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
1447     IO_OR_GS_CODE();
1448 
1449     assert(bytes <= SIZE_MAX);
1450 
1451     return blk_co_pwritev(blk, offset, bytes, &qiov, flags);
1452 }
1453 
1454 int coroutine_fn blk_co_pwritev(BlockBackend *blk, int64_t offset,
1455                                 int64_t bytes, QEMUIOVector *qiov,
1456                                 BdrvRequestFlags flags)
1457 {
1458     IO_OR_GS_CODE();
1459     return blk_co_pwritev_part(blk, offset, bytes, qiov, 0, flags);
1460 }
1461 
1462 int coroutine_fn blk_co_block_status_above(BlockBackend *blk,
1463                                            BlockDriverState *base,
1464                                            int64_t offset, int64_t bytes,
1465                                            int64_t *pnum, int64_t *map,
1466                                            BlockDriverState **file)
1467 {
1468     IO_CODE();
1469     GRAPH_RDLOCK_GUARD();
1470     return bdrv_co_block_status_above(blk_bs(blk), base, offset, bytes, pnum,
1471                                       map, file);
1472 }
1473 
1474 int coroutine_fn blk_co_is_allocated_above(BlockBackend *blk,
1475                                            BlockDriverState *base,
1476                                            bool include_base, int64_t offset,
1477                                            int64_t bytes, int64_t *pnum)
1478 {
1479     IO_CODE();
1480     GRAPH_RDLOCK_GUARD();
1481     return bdrv_co_is_allocated_above(blk_bs(blk), base, include_base, offset,
1482                                       bytes, pnum);
1483 }
1484 
1485 typedef struct BlkRwCo {
1486     BlockBackend *blk;
1487     int64_t offset;
1488     void *iobuf;
1489     int ret;
1490     BdrvRequestFlags flags;
1491 } BlkRwCo;
1492 
1493 int blk_make_zero(BlockBackend *blk, BdrvRequestFlags flags)
1494 {
1495     GLOBAL_STATE_CODE();
1496     return bdrv_make_zero(blk->root, flags);
1497 }
1498 
1499 void blk_inc_in_flight(BlockBackend *blk)
1500 {
1501     IO_CODE();
1502     qatomic_inc(&blk->in_flight);
1503 }
1504 
1505 void blk_dec_in_flight(BlockBackend *blk)
1506 {
1507     IO_CODE();
1508     qatomic_dec(&blk->in_flight);
1509     aio_wait_kick();
1510 }
1511 
1512 static void error_callback_bh(void *opaque)
1513 {
1514     struct BlockBackendAIOCB *acb = opaque;
1515 
1516     blk_dec_in_flight(acb->blk);
1517     acb->common.cb(acb->common.opaque, acb->ret);
1518     qemu_aio_unref(acb);
1519 }
1520 
1521 BlockAIOCB *blk_abort_aio_request(BlockBackend *blk,
1522                                   BlockCompletionFunc *cb,
1523                                   void *opaque, int ret)
1524 {
1525     struct BlockBackendAIOCB *acb;
1526     IO_CODE();
1527 
1528     blk_inc_in_flight(blk);
1529     acb = blk_aio_get(&block_backend_aiocb_info, blk, cb, opaque);
1530     acb->blk = blk;
1531     acb->ret = ret;
1532 
1533     replay_bh_schedule_oneshot_event(qemu_get_current_aio_context(),
1534                                      error_callback_bh, acb);
1535     return &acb->common;
1536 }
1537 
1538 typedef struct BlkAioEmAIOCB {
1539     BlockAIOCB common;
1540     BlkRwCo rwco;
1541     int64_t bytes;
1542     bool has_returned;
1543 } BlkAioEmAIOCB;
1544 
1545 static const AIOCBInfo blk_aio_em_aiocb_info = {
1546     .aiocb_size         = sizeof(BlkAioEmAIOCB),
1547 };
1548 
1549 static void blk_aio_complete(BlkAioEmAIOCB *acb)
1550 {
1551     if (acb->has_returned) {
1552         acb->common.cb(acb->common.opaque, acb->rwco.ret);
1553         blk_dec_in_flight(acb->rwco.blk);
1554         qemu_aio_unref(acb);
1555     }
1556 }
1557 
1558 static void blk_aio_complete_bh(void *opaque)
1559 {
1560     BlkAioEmAIOCB *acb = opaque;
1561     assert(acb->has_returned);
1562     blk_aio_complete(acb);
1563 }
1564 
1565 static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset,
1566                                 int64_t bytes,
1567                                 void *iobuf, CoroutineEntry co_entry,
1568                                 BdrvRequestFlags flags,
1569                                 BlockCompletionFunc *cb, void *opaque)
1570 {
1571     BlkAioEmAIOCB *acb;
1572     Coroutine *co;
1573 
1574     blk_inc_in_flight(blk);
1575     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1576     acb->rwco = (BlkRwCo) {
1577         .blk    = blk,
1578         .offset = offset,
1579         .iobuf  = iobuf,
1580         .flags  = flags,
1581         .ret    = NOT_DONE,
1582     };
1583     acb->bytes = bytes;
1584     acb->has_returned = false;
1585 
1586     co = qemu_coroutine_create(co_entry, acb);
1587     aio_co_enter(qemu_get_current_aio_context(), co);
1588 
1589     acb->has_returned = true;
1590     if (acb->rwco.ret != NOT_DONE) {
1591         replay_bh_schedule_oneshot_event(qemu_get_current_aio_context(),
1592                                          blk_aio_complete_bh, acb);
1593     }
1594 
1595     return &acb->common;
1596 }
1597 
1598 static void coroutine_fn blk_aio_read_entry(void *opaque)
1599 {
1600     BlkAioEmAIOCB *acb = opaque;
1601     BlkRwCo *rwco = &acb->rwco;
1602     QEMUIOVector *qiov = rwco->iobuf;
1603 
1604     assert(qiov->size == acb->bytes);
1605     rwco->ret = blk_co_do_preadv_part(rwco->blk, rwco->offset, acb->bytes, qiov,
1606                                       0, rwco->flags);
1607     blk_aio_complete(acb);
1608 }
1609 
1610 static void coroutine_fn blk_aio_write_entry(void *opaque)
1611 {
1612     BlkAioEmAIOCB *acb = opaque;
1613     BlkRwCo *rwco = &acb->rwco;
1614     QEMUIOVector *qiov = rwco->iobuf;
1615 
1616     assert(!qiov || qiov->size == acb->bytes);
1617     rwco->ret = blk_co_do_pwritev_part(rwco->blk, rwco->offset, acb->bytes,
1618                                        qiov, 0, rwco->flags);
1619     blk_aio_complete(acb);
1620 }
1621 
1622 BlockAIOCB *blk_aio_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1623                                   int64_t bytes, BdrvRequestFlags flags,
1624                                   BlockCompletionFunc *cb, void *opaque)
1625 {
1626     IO_CODE();
1627     return blk_aio_prwv(blk, offset, bytes, NULL, blk_aio_write_entry,
1628                         flags | BDRV_REQ_ZERO_WRITE, cb, opaque);
1629 }
1630 
1631 int64_t coroutine_fn blk_co_getlength(BlockBackend *blk)
1632 {
1633     IO_CODE();
1634     GRAPH_RDLOCK_GUARD();
1635 
1636     if (!blk_co_is_available(blk)) {
1637         return -ENOMEDIUM;
1638     }
1639 
1640     return bdrv_co_getlength(blk_bs(blk));
1641 }
1642 
1643 int64_t coroutine_fn blk_co_nb_sectors(BlockBackend *blk)
1644 {
1645     BlockDriverState *bs = blk_bs(blk);
1646 
1647     IO_CODE();
1648     GRAPH_RDLOCK_GUARD();
1649 
1650     if (!bs) {
1651         return -ENOMEDIUM;
1652     } else {
1653         return bdrv_co_nb_sectors(bs);
1654     }
1655 }
1656 
1657 /*
1658  * This wrapper is written by hand because this function is in the hot I/O path,
1659  * via blk_get_geometry.
1660  */
1661 int64_t coroutine_mixed_fn blk_nb_sectors(BlockBackend *blk)
1662 {
1663     BlockDriverState *bs = blk_bs(blk);
1664 
1665     IO_CODE();
1666 
1667     if (!bs) {
1668         return -ENOMEDIUM;
1669     } else {
1670         return bdrv_nb_sectors(bs);
1671     }
1672 }
1673 
1674 /* return 0 as number of sectors if no device present or error */
1675 void coroutine_fn blk_co_get_geometry(BlockBackend *blk,
1676                                       uint64_t *nb_sectors_ptr)
1677 {
1678     int64_t ret = blk_co_nb_sectors(blk);
1679     *nb_sectors_ptr = ret < 0 ? 0 : ret;
1680 }
1681 
1682 /*
1683  * This wrapper is written by hand because this function is in the hot I/O path.
1684  */
1685 void coroutine_mixed_fn blk_get_geometry(BlockBackend *blk,
1686                                          uint64_t *nb_sectors_ptr)
1687 {
1688     int64_t ret = blk_nb_sectors(blk);
1689     *nb_sectors_ptr = ret < 0 ? 0 : ret;
1690 }
1691 
1692 BlockAIOCB *blk_aio_preadv(BlockBackend *blk, int64_t offset,
1693                            QEMUIOVector *qiov, BdrvRequestFlags flags,
1694                            BlockCompletionFunc *cb, void *opaque)
1695 {
1696     IO_CODE();
1697     assert((uint64_t)qiov->size <= INT64_MAX);
1698     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1699                         blk_aio_read_entry, flags, cb, opaque);
1700 }
1701 
1702 BlockAIOCB *blk_aio_pwritev(BlockBackend *blk, int64_t offset,
1703                             QEMUIOVector *qiov, BdrvRequestFlags flags,
1704                             BlockCompletionFunc *cb, void *opaque)
1705 {
1706     IO_CODE();
1707     assert((uint64_t)qiov->size <= INT64_MAX);
1708     return blk_aio_prwv(blk, offset, qiov->size, qiov,
1709                         blk_aio_write_entry, flags, cb, opaque);
1710 }
1711 
1712 void blk_aio_cancel(BlockAIOCB *acb)
1713 {
1714     GLOBAL_STATE_CODE();
1715     bdrv_aio_cancel(acb);
1716 }
1717 
1718 void blk_aio_cancel_async(BlockAIOCB *acb)
1719 {
1720     IO_CODE();
1721     bdrv_aio_cancel_async(acb);
1722 }
1723 
1724 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1725 static int coroutine_fn
1726 blk_co_do_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1727 {
1728     IO_CODE();
1729 
1730     blk_wait_while_drained(blk);
1731     GRAPH_RDLOCK_GUARD();
1732 
1733     if (!blk_co_is_available(blk)) {
1734         return -ENOMEDIUM;
1735     }
1736 
1737     return bdrv_co_ioctl(blk_bs(blk), req, buf);
1738 }
1739 
1740 int coroutine_fn blk_co_ioctl(BlockBackend *blk, unsigned long int req,
1741                               void *buf)
1742 {
1743     int ret;
1744     IO_OR_GS_CODE();
1745 
1746     blk_inc_in_flight(blk);
1747     ret = blk_co_do_ioctl(blk, req, buf);
1748     blk_dec_in_flight(blk);
1749 
1750     return ret;
1751 }
1752 
1753 static void coroutine_fn blk_aio_ioctl_entry(void *opaque)
1754 {
1755     BlkAioEmAIOCB *acb = opaque;
1756     BlkRwCo *rwco = &acb->rwco;
1757 
1758     rwco->ret = blk_co_do_ioctl(rwco->blk, rwco->offset, rwco->iobuf);
1759 
1760     blk_aio_complete(acb);
1761 }
1762 
1763 BlockAIOCB *blk_aio_ioctl(BlockBackend *blk, unsigned long int req, void *buf,
1764                           BlockCompletionFunc *cb, void *opaque)
1765 {
1766     IO_CODE();
1767     return blk_aio_prwv(blk, req, 0, buf, blk_aio_ioctl_entry, 0, cb, opaque);
1768 }
1769 
1770 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1771 static int coroutine_fn
1772 blk_co_do_pdiscard(BlockBackend *blk, int64_t offset, int64_t bytes)
1773 {
1774     int ret;
1775     IO_CODE();
1776 
1777     blk_wait_while_drained(blk);
1778     GRAPH_RDLOCK_GUARD();
1779 
1780     ret = blk_check_byte_request(blk, offset, bytes);
1781     if (ret < 0) {
1782         return ret;
1783     }
1784 
1785     return bdrv_co_pdiscard(blk->root, offset, bytes);
1786 }
1787 
1788 static void coroutine_fn blk_aio_pdiscard_entry(void *opaque)
1789 {
1790     BlkAioEmAIOCB *acb = opaque;
1791     BlkRwCo *rwco = &acb->rwco;
1792 
1793     rwco->ret = blk_co_do_pdiscard(rwco->blk, rwco->offset, acb->bytes);
1794     blk_aio_complete(acb);
1795 }
1796 
1797 BlockAIOCB *blk_aio_pdiscard(BlockBackend *blk,
1798                              int64_t offset, int64_t bytes,
1799                              BlockCompletionFunc *cb, void *opaque)
1800 {
1801     IO_CODE();
1802     return blk_aio_prwv(blk, offset, bytes, NULL, blk_aio_pdiscard_entry, 0,
1803                         cb, opaque);
1804 }
1805 
1806 int coroutine_fn blk_co_pdiscard(BlockBackend *blk, int64_t offset,
1807                                  int64_t bytes)
1808 {
1809     int ret;
1810     IO_OR_GS_CODE();
1811 
1812     blk_inc_in_flight(blk);
1813     ret = blk_co_do_pdiscard(blk, offset, bytes);
1814     blk_dec_in_flight(blk);
1815 
1816     return ret;
1817 }
1818 
1819 /* To be called between exactly one pair of blk_inc/dec_in_flight() */
1820 static int coroutine_fn blk_co_do_flush(BlockBackend *blk)
1821 {
1822     IO_CODE();
1823     blk_wait_while_drained(blk);
1824     GRAPH_RDLOCK_GUARD();
1825 
1826     if (!blk_co_is_available(blk)) {
1827         return -ENOMEDIUM;
1828     }
1829 
1830     return bdrv_co_flush(blk_bs(blk));
1831 }
1832 
1833 static void coroutine_fn blk_aio_flush_entry(void *opaque)
1834 {
1835     BlkAioEmAIOCB *acb = opaque;
1836     BlkRwCo *rwco = &acb->rwco;
1837 
1838     rwco->ret = blk_co_do_flush(rwco->blk);
1839     blk_aio_complete(acb);
1840 }
1841 
1842 BlockAIOCB *blk_aio_flush(BlockBackend *blk,
1843                           BlockCompletionFunc *cb, void *opaque)
1844 {
1845     IO_CODE();
1846     return blk_aio_prwv(blk, 0, 0, NULL, blk_aio_flush_entry, 0, cb, opaque);
1847 }
1848 
1849 int coroutine_fn blk_co_flush(BlockBackend *blk)
1850 {
1851     int ret;
1852     IO_OR_GS_CODE();
1853 
1854     blk_inc_in_flight(blk);
1855     ret = blk_co_do_flush(blk);
1856     blk_dec_in_flight(blk);
1857 
1858     return ret;
1859 }
1860 
1861 static void coroutine_fn blk_aio_zone_report_entry(void *opaque)
1862 {
1863     BlkAioEmAIOCB *acb = opaque;
1864     BlkRwCo *rwco = &acb->rwco;
1865 
1866     rwco->ret = blk_co_zone_report(rwco->blk, rwco->offset,
1867                                    (unsigned int*)(uintptr_t)acb->bytes,
1868                                    rwco->iobuf);
1869     blk_aio_complete(acb);
1870 }
1871 
1872 BlockAIOCB *blk_aio_zone_report(BlockBackend *blk, int64_t offset,
1873                                 unsigned int *nr_zones,
1874                                 BlockZoneDescriptor  *zones,
1875                                 BlockCompletionFunc *cb, void *opaque)
1876 {
1877     BlkAioEmAIOCB *acb;
1878     Coroutine *co;
1879     IO_CODE();
1880 
1881     blk_inc_in_flight(blk);
1882     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1883     acb->rwco = (BlkRwCo) {
1884         .blk    = blk,
1885         .offset = offset,
1886         .iobuf  = zones,
1887         .ret    = NOT_DONE,
1888     };
1889     acb->bytes = (int64_t)(uintptr_t)nr_zones,
1890     acb->has_returned = false;
1891 
1892     co = qemu_coroutine_create(blk_aio_zone_report_entry, acb);
1893     aio_co_enter(qemu_get_current_aio_context(), co);
1894 
1895     acb->has_returned = true;
1896     if (acb->rwco.ret != NOT_DONE) {
1897         replay_bh_schedule_oneshot_event(qemu_get_current_aio_context(),
1898                                          blk_aio_complete_bh, acb);
1899     }
1900 
1901     return &acb->common;
1902 }
1903 
1904 static void coroutine_fn blk_aio_zone_mgmt_entry(void *opaque)
1905 {
1906     BlkAioEmAIOCB *acb = opaque;
1907     BlkRwCo *rwco = &acb->rwco;
1908 
1909     rwco->ret = blk_co_zone_mgmt(rwco->blk,
1910                                  (BlockZoneOp)(uintptr_t)rwco->iobuf,
1911                                  rwco->offset, acb->bytes);
1912     blk_aio_complete(acb);
1913 }
1914 
1915 BlockAIOCB *blk_aio_zone_mgmt(BlockBackend *blk, BlockZoneOp op,
1916                               int64_t offset, int64_t len,
1917                               BlockCompletionFunc *cb, void *opaque) {
1918     BlkAioEmAIOCB *acb;
1919     Coroutine *co;
1920     IO_CODE();
1921 
1922     blk_inc_in_flight(blk);
1923     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1924     acb->rwco = (BlkRwCo) {
1925         .blk    = blk,
1926         .offset = offset,
1927         .iobuf  = (void *)(uintptr_t)op,
1928         .ret    = NOT_DONE,
1929     };
1930     acb->bytes = len;
1931     acb->has_returned = false;
1932 
1933     co = qemu_coroutine_create(blk_aio_zone_mgmt_entry, acb);
1934     aio_co_enter(qemu_get_current_aio_context(), co);
1935 
1936     acb->has_returned = true;
1937     if (acb->rwco.ret != NOT_DONE) {
1938         replay_bh_schedule_oneshot_event(qemu_get_current_aio_context(),
1939                                          blk_aio_complete_bh, acb);
1940     }
1941 
1942     return &acb->common;
1943 }
1944 
1945 static void coroutine_fn blk_aio_zone_append_entry(void *opaque)
1946 {
1947     BlkAioEmAIOCB *acb = opaque;
1948     BlkRwCo *rwco = &acb->rwco;
1949 
1950     rwco->ret = blk_co_zone_append(rwco->blk, (int64_t *)(uintptr_t)acb->bytes,
1951                                    rwco->iobuf, rwco->flags);
1952     blk_aio_complete(acb);
1953 }
1954 
1955 BlockAIOCB *blk_aio_zone_append(BlockBackend *blk, int64_t *offset,
1956                                 QEMUIOVector *qiov, BdrvRequestFlags flags,
1957                                 BlockCompletionFunc *cb, void *opaque) {
1958     BlkAioEmAIOCB *acb;
1959     Coroutine *co;
1960     IO_CODE();
1961 
1962     blk_inc_in_flight(blk);
1963     acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1964     acb->rwco = (BlkRwCo) {
1965         .blk    = blk,
1966         .ret    = NOT_DONE,
1967         .flags  = flags,
1968         .iobuf  = qiov,
1969     };
1970     acb->bytes = (int64_t)(uintptr_t)offset;
1971     acb->has_returned = false;
1972 
1973     co = qemu_coroutine_create(blk_aio_zone_append_entry, acb);
1974     aio_co_enter(qemu_get_current_aio_context(), co);
1975     acb->has_returned = true;
1976     if (acb->rwco.ret != NOT_DONE) {
1977         replay_bh_schedule_oneshot_event(qemu_get_current_aio_context(),
1978                                          blk_aio_complete_bh, acb);
1979     }
1980 
1981     return &acb->common;
1982 }
1983 
1984 /*
1985  * Send a zone_report command.
1986  * offset is a byte offset from the start of the device. No alignment
1987  * required for offset.
1988  * nr_zones represents IN maximum and OUT actual.
1989  */
1990 int coroutine_fn blk_co_zone_report(BlockBackend *blk, int64_t offset,
1991                                     unsigned int *nr_zones,
1992                                     BlockZoneDescriptor *zones)
1993 {
1994     int ret;
1995     IO_CODE();
1996 
1997     blk_inc_in_flight(blk); /* increase before waiting */
1998     blk_wait_while_drained(blk);
1999     GRAPH_RDLOCK_GUARD();
2000     if (!blk_is_available(blk)) {
2001         blk_dec_in_flight(blk);
2002         return -ENOMEDIUM;
2003     }
2004     ret = bdrv_co_zone_report(blk_bs(blk), offset, nr_zones, zones);
2005     blk_dec_in_flight(blk);
2006     return ret;
2007 }
2008 
2009 /*
2010  * Send a zone_management command.
2011  * op is the zone operation;
2012  * offset is the byte offset from the start of the zoned device;
2013  * len is the maximum number of bytes the command should operate on. It
2014  * should be aligned with the device zone size.
2015  */
2016 int coroutine_fn blk_co_zone_mgmt(BlockBackend *blk, BlockZoneOp op,
2017         int64_t offset, int64_t len)
2018 {
2019     int ret;
2020     IO_CODE();
2021 
2022     blk_inc_in_flight(blk);
2023     blk_wait_while_drained(blk);
2024     GRAPH_RDLOCK_GUARD();
2025 
2026     ret = blk_check_byte_request(blk, offset, len);
2027     if (ret < 0) {
2028         blk_dec_in_flight(blk);
2029         return ret;
2030     }
2031 
2032     ret = bdrv_co_zone_mgmt(blk_bs(blk), op, offset, len);
2033     blk_dec_in_flight(blk);
2034     return ret;
2035 }
2036 
2037 /*
2038  * Send a zone_append command.
2039  */
2040 int coroutine_fn blk_co_zone_append(BlockBackend *blk, int64_t *offset,
2041         QEMUIOVector *qiov, BdrvRequestFlags flags)
2042 {
2043     int ret;
2044     IO_CODE();
2045 
2046     blk_inc_in_flight(blk);
2047     blk_wait_while_drained(blk);
2048     GRAPH_RDLOCK_GUARD();
2049     if (!blk_is_available(blk)) {
2050         blk_dec_in_flight(blk);
2051         return -ENOMEDIUM;
2052     }
2053 
2054     ret = bdrv_co_zone_append(blk_bs(blk), offset, qiov, flags);
2055     blk_dec_in_flight(blk);
2056     return ret;
2057 }
2058 
2059 void blk_drain(BlockBackend *blk)
2060 {
2061     BlockDriverState *bs = blk_bs(blk);
2062     GLOBAL_STATE_CODE();
2063 
2064     if (bs) {
2065         bdrv_ref(bs);
2066         bdrv_drained_begin(bs);
2067     }
2068 
2069     /* We may have -ENOMEDIUM completions in flight */
2070     AIO_WAIT_WHILE(blk_get_aio_context(blk),
2071                    qatomic_read(&blk->in_flight) > 0);
2072 
2073     if (bs) {
2074         bdrv_drained_end(bs);
2075         bdrv_unref(bs);
2076     }
2077 }
2078 
2079 void blk_drain_all(void)
2080 {
2081     BlockBackend *blk = NULL;
2082 
2083     GLOBAL_STATE_CODE();
2084 
2085     bdrv_drain_all_begin();
2086 
2087     while ((blk = blk_all_next(blk)) != NULL) {
2088         /* We may have -ENOMEDIUM completions in flight */
2089         AIO_WAIT_WHILE_UNLOCKED(NULL, qatomic_read(&blk->in_flight) > 0);
2090     }
2091 
2092     bdrv_drain_all_end();
2093 }
2094 
2095 void blk_set_on_error(BlockBackend *blk, BlockdevOnError on_read_error,
2096                       BlockdevOnError on_write_error)
2097 {
2098     GLOBAL_STATE_CODE();
2099     blk->on_read_error = on_read_error;
2100     blk->on_write_error = on_write_error;
2101 }
2102 
2103 BlockdevOnError blk_get_on_error(BlockBackend *blk, bool is_read)
2104 {
2105     IO_CODE();
2106     return is_read ? blk->on_read_error : blk->on_write_error;
2107 }
2108 
2109 BlockErrorAction blk_get_error_action(BlockBackend *blk, bool is_read,
2110                                       int error)
2111 {
2112     BlockdevOnError on_err = blk_get_on_error(blk, is_read);
2113     IO_CODE();
2114 
2115     switch (on_err) {
2116     case BLOCKDEV_ON_ERROR_ENOSPC:
2117         return (error == ENOSPC) ?
2118                BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
2119     case BLOCKDEV_ON_ERROR_STOP:
2120         return BLOCK_ERROR_ACTION_STOP;
2121     case BLOCKDEV_ON_ERROR_REPORT:
2122         return BLOCK_ERROR_ACTION_REPORT;
2123     case BLOCKDEV_ON_ERROR_IGNORE:
2124         return BLOCK_ERROR_ACTION_IGNORE;
2125     case BLOCKDEV_ON_ERROR_AUTO:
2126     default:
2127         abort();
2128     }
2129 }
2130 
2131 static void send_qmp_error_event(BlockBackend *blk,
2132                                  BlockErrorAction action,
2133                                  bool is_read, int error)
2134 {
2135     IoOperationType optype;
2136     BlockDriverState *bs = blk_bs(blk);
2137 
2138     optype = is_read ? IO_OPERATION_TYPE_READ : IO_OPERATION_TYPE_WRITE;
2139     qapi_event_send_block_io_error(blk_name(blk),
2140                                    blk_get_attached_dev_path(blk),
2141                                    bs ? bdrv_get_node_name(bs) : NULL, optype,
2142                                    action, blk_iostatus_is_enabled(blk),
2143                                    error == ENOSPC, strerror(error));
2144 }
2145 
2146 /* This is done by device models because, while the block layer knows
2147  * about the error, it does not know whether an operation comes from
2148  * the device or the block layer (from a job, for example).
2149  */
2150 void blk_error_action(BlockBackend *blk, BlockErrorAction action,
2151                       bool is_read, int error)
2152 {
2153     assert(error >= 0);
2154     IO_CODE();
2155 
2156     if (action == BLOCK_ERROR_ACTION_STOP) {
2157         /* First set the iostatus, so that "info block" returns an iostatus
2158          * that matches the events raised so far (an additional error iostatus
2159          * is fine, but not a lost one).
2160          */
2161         blk_iostatus_set_err(blk, error);
2162 
2163         /* Then raise the request to stop the VM and the event.
2164          * qemu_system_vmstop_request_prepare has two effects.  First,
2165          * it ensures that the STOP event always comes after the
2166          * BLOCK_IO_ERROR event.  Second, it ensures that even if management
2167          * can observe the STOP event and do a "cont" before the STOP
2168          * event is issued, the VM will not stop.  In this case, vm_start()
2169          * also ensures that the STOP/RESUME pair of events is emitted.
2170          */
2171         qemu_system_vmstop_request_prepare();
2172         send_qmp_error_event(blk, action, is_read, error);
2173         qemu_system_vmstop_request(RUN_STATE_IO_ERROR);
2174     } else {
2175         send_qmp_error_event(blk, action, is_read, error);
2176     }
2177 }
2178 
2179 /*
2180  * Returns true if the BlockBackend can support taking write permissions
2181  * (because its root node is not read-only).
2182  */
2183 bool blk_supports_write_perm(BlockBackend *blk)
2184 {
2185     BlockDriverState *bs = blk_bs(blk);
2186     GLOBAL_STATE_CODE();
2187 
2188     if (bs) {
2189         return !bdrv_is_read_only(bs);
2190     } else {
2191         return blk->root_state.open_flags & BDRV_O_RDWR;
2192     }
2193 }
2194 
2195 /*
2196  * Returns true if the BlockBackend can be written to in its current
2197  * configuration (i.e. if write permission have been requested)
2198  */
2199 bool blk_is_writable(BlockBackend *blk)
2200 {
2201     IO_CODE();
2202     return blk->perm & BLK_PERM_WRITE;
2203 }
2204 
2205 bool blk_is_sg(BlockBackend *blk)
2206 {
2207     BlockDriverState *bs = blk_bs(blk);
2208     GLOBAL_STATE_CODE();
2209 
2210     if (!bs) {
2211         return false;
2212     }
2213 
2214     return bdrv_is_sg(bs);
2215 }
2216 
2217 bool blk_enable_write_cache(BlockBackend *blk)
2218 {
2219     IO_CODE();
2220     return blk->enable_write_cache;
2221 }
2222 
2223 void blk_set_enable_write_cache(BlockBackend *blk, bool wce)
2224 {
2225     IO_CODE();
2226     blk->enable_write_cache = wce;
2227 }
2228 
2229 bool coroutine_fn blk_co_is_inserted(BlockBackend *blk)
2230 {
2231     BlockDriverState *bs = blk_bs(blk);
2232     IO_CODE();
2233     assert_bdrv_graph_readable();
2234 
2235     return bs && bdrv_co_is_inserted(bs);
2236 }
2237 
2238 bool coroutine_fn blk_co_is_available(BlockBackend *blk)
2239 {
2240     IO_CODE();
2241     return blk_co_is_inserted(blk) && !blk_dev_is_tray_open(blk);
2242 }
2243 
2244 void coroutine_fn blk_co_lock_medium(BlockBackend *blk, bool locked)
2245 {
2246     BlockDriverState *bs = blk_bs(blk);
2247     IO_CODE();
2248     GRAPH_RDLOCK_GUARD();
2249 
2250     if (bs) {
2251         bdrv_co_lock_medium(bs, locked);
2252     }
2253 }
2254 
2255 void coroutine_fn blk_co_eject(BlockBackend *blk, bool eject_flag)
2256 {
2257     BlockDriverState *bs = blk_bs(blk);
2258     char *id;
2259     IO_CODE();
2260     GRAPH_RDLOCK_GUARD();
2261 
2262     if (bs) {
2263         bdrv_co_eject(bs, eject_flag);
2264     }
2265 
2266     /* Whether or not we ejected on the backend,
2267      * the frontend experienced a tray event. */
2268     id = blk_get_attached_dev_id(blk);
2269     qapi_event_send_device_tray_moved(blk_name(blk), id,
2270                                       eject_flag);
2271     g_free(id);
2272 }
2273 
2274 int blk_get_flags(BlockBackend *blk)
2275 {
2276     BlockDriverState *bs = blk_bs(blk);
2277     GLOBAL_STATE_CODE();
2278 
2279     if (bs) {
2280         return bdrv_get_flags(bs);
2281     } else {
2282         return blk->root_state.open_flags;
2283     }
2284 }
2285 
2286 /* Returns the minimum request alignment, in bytes; guaranteed nonzero */
2287 uint32_t blk_get_request_alignment(BlockBackend *blk)
2288 {
2289     BlockDriverState *bs = blk_bs(blk);
2290     IO_CODE();
2291     return bs ? bs->bl.request_alignment : BDRV_SECTOR_SIZE;
2292 }
2293 
2294 /* Returns the maximum hardware transfer length, in bytes; guaranteed nonzero */
2295 uint64_t blk_get_max_hw_transfer(BlockBackend *blk)
2296 {
2297     BlockDriverState *bs = blk_bs(blk);
2298     uint64_t max = INT_MAX;
2299     IO_CODE();
2300 
2301     if (bs) {
2302         max = MIN_NON_ZERO(max, bs->bl.max_hw_transfer);
2303         max = MIN_NON_ZERO(max, bs->bl.max_transfer);
2304     }
2305     return ROUND_DOWN(max, blk_get_request_alignment(blk));
2306 }
2307 
2308 /* Returns the maximum transfer length, in bytes; guaranteed nonzero */
2309 uint32_t blk_get_max_transfer(BlockBackend *blk)
2310 {
2311     BlockDriverState *bs = blk_bs(blk);
2312     uint32_t max = INT_MAX;
2313     IO_CODE();
2314 
2315     if (bs) {
2316         max = MIN_NON_ZERO(max, bs->bl.max_transfer);
2317     }
2318     return ROUND_DOWN(max, blk_get_request_alignment(blk));
2319 }
2320 
2321 int blk_get_max_hw_iov(BlockBackend *blk)
2322 {
2323     IO_CODE();
2324     return MIN_NON_ZERO(blk->root->bs->bl.max_hw_iov,
2325                         blk->root->bs->bl.max_iov);
2326 }
2327 
2328 int blk_get_max_iov(BlockBackend *blk)
2329 {
2330     IO_CODE();
2331     return blk->root->bs->bl.max_iov;
2332 }
2333 
2334 void *blk_try_blockalign(BlockBackend *blk, size_t size)
2335 {
2336     IO_CODE();
2337     return qemu_try_blockalign(blk ? blk_bs(blk) : NULL, size);
2338 }
2339 
2340 void *blk_blockalign(BlockBackend *blk, size_t size)
2341 {
2342     IO_CODE();
2343     return qemu_blockalign(blk ? blk_bs(blk) : NULL, size);
2344 }
2345 
2346 bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp)
2347 {
2348     BlockDriverState *bs = blk_bs(blk);
2349     GLOBAL_STATE_CODE();
2350     GRAPH_RDLOCK_GUARD_MAINLOOP();
2351 
2352     if (!bs) {
2353         return false;
2354     }
2355 
2356     return bdrv_op_is_blocked(bs, op, errp);
2357 }
2358 
2359 /**
2360  * Return BB's current AioContext.  Note that this context may change
2361  * concurrently at any time, with one exception: If the BB has a root node
2362  * attached, its context will only change through bdrv_try_change_aio_context(),
2363  * which creates a drained section.  Therefore, incrementing such a BB's
2364  * in-flight counter will prevent its context from changing.
2365  */
2366 AioContext *blk_get_aio_context(BlockBackend *blk)
2367 {
2368     IO_CODE();
2369 
2370     if (!blk) {
2371         return qemu_get_aio_context();
2372     }
2373 
2374     return qatomic_read(&blk->ctx);
2375 }
2376 
2377 int blk_set_aio_context(BlockBackend *blk, AioContext *new_context,
2378                         Error **errp)
2379 {
2380     bool old_allow_change;
2381     BlockDriverState *bs = blk_bs(blk);
2382     int ret;
2383 
2384     GLOBAL_STATE_CODE();
2385 
2386     if (!bs) {
2387         qatomic_set(&blk->ctx, new_context);
2388         return 0;
2389     }
2390 
2391     bdrv_ref(bs);
2392 
2393     old_allow_change = blk->allow_aio_context_change;
2394     blk->allow_aio_context_change = true;
2395 
2396     ret = bdrv_try_change_aio_context(bs, new_context, NULL, errp);
2397 
2398     blk->allow_aio_context_change = old_allow_change;
2399 
2400     bdrv_unref(bs);
2401     return ret;
2402 }
2403 
2404 typedef struct BdrvStateBlkRootContext {
2405     AioContext *new_ctx;
2406     BlockBackend *blk;
2407 } BdrvStateBlkRootContext;
2408 
2409 static void blk_root_set_aio_ctx_commit(void *opaque)
2410 {
2411     BdrvStateBlkRootContext *s = opaque;
2412     BlockBackend *blk = s->blk;
2413     AioContext *new_context = s->new_ctx;
2414     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
2415 
2416     qatomic_set(&blk->ctx, new_context);
2417     if (tgm->throttle_state) {
2418         throttle_group_detach_aio_context(tgm);
2419         throttle_group_attach_aio_context(tgm, new_context);
2420     }
2421 }
2422 
2423 static TransactionActionDrv set_blk_root_context = {
2424     .commit = blk_root_set_aio_ctx_commit,
2425     .clean = g_free,
2426 };
2427 
2428 static bool blk_root_change_aio_ctx(BdrvChild *child, AioContext *ctx,
2429                                     GHashTable *visited, Transaction *tran,
2430                                     Error **errp)
2431 {
2432     BlockBackend *blk = child->opaque;
2433     BdrvStateBlkRootContext *s;
2434 
2435     if (!blk->allow_aio_context_change) {
2436         /*
2437          * Manually created BlockBackends (those with a name) that are not
2438          * attached to anything can change their AioContext without updating
2439          * their user; return an error for others.
2440          */
2441         if (!blk->name || blk->dev) {
2442             /* TODO Add BB name/QOM path */
2443             error_setg(errp, "Cannot change iothread of active block backend");
2444             return false;
2445         }
2446     }
2447 
2448     s = g_new(BdrvStateBlkRootContext, 1);
2449     *s = (BdrvStateBlkRootContext) {
2450         .new_ctx = ctx,
2451         .blk = blk,
2452     };
2453 
2454     tran_add(tran, &set_blk_root_context, s);
2455     return true;
2456 }
2457 
2458 void blk_add_aio_context_notifier(BlockBackend *blk,
2459         void (*attached_aio_context)(AioContext *new_context, void *opaque),
2460         void (*detach_aio_context)(void *opaque), void *opaque)
2461 {
2462     BlockBackendAioNotifier *notifier;
2463     BlockDriverState *bs = blk_bs(blk);
2464     GLOBAL_STATE_CODE();
2465 
2466     notifier = g_new(BlockBackendAioNotifier, 1);
2467     notifier->attached_aio_context = attached_aio_context;
2468     notifier->detach_aio_context = detach_aio_context;
2469     notifier->opaque = opaque;
2470     QLIST_INSERT_HEAD(&blk->aio_notifiers, notifier, list);
2471 
2472     if (bs) {
2473         bdrv_add_aio_context_notifier(bs, attached_aio_context,
2474                                       detach_aio_context, opaque);
2475     }
2476 }
2477 
2478 void blk_remove_aio_context_notifier(BlockBackend *blk,
2479                                      void (*attached_aio_context)(AioContext *,
2480                                                                   void *),
2481                                      void (*detach_aio_context)(void *),
2482                                      void *opaque)
2483 {
2484     BlockBackendAioNotifier *notifier;
2485     BlockDriverState *bs = blk_bs(blk);
2486 
2487     GLOBAL_STATE_CODE();
2488 
2489     if (bs) {
2490         bdrv_remove_aio_context_notifier(bs, attached_aio_context,
2491                                          detach_aio_context, opaque);
2492     }
2493 
2494     QLIST_FOREACH(notifier, &blk->aio_notifiers, list) {
2495         if (notifier->attached_aio_context == attached_aio_context &&
2496             notifier->detach_aio_context == detach_aio_context &&
2497             notifier->opaque == opaque) {
2498             QLIST_REMOVE(notifier, list);
2499             g_free(notifier);
2500             return;
2501         }
2502     }
2503 
2504     abort();
2505 }
2506 
2507 void blk_add_remove_bs_notifier(BlockBackend *blk, Notifier *notify)
2508 {
2509     GLOBAL_STATE_CODE();
2510     notifier_list_add(&blk->remove_bs_notifiers, notify);
2511 }
2512 
2513 BlockAcctStats *blk_get_stats(BlockBackend *blk)
2514 {
2515     IO_CODE();
2516     return &blk->stats;
2517 }
2518 
2519 void *blk_aio_get(const AIOCBInfo *aiocb_info, BlockBackend *blk,
2520                   BlockCompletionFunc *cb, void *opaque)
2521 {
2522     IO_CODE();
2523     return qemu_aio_get(aiocb_info, blk_bs(blk), cb, opaque);
2524 }
2525 
2526 int coroutine_fn blk_co_pwrite_zeroes(BlockBackend *blk, int64_t offset,
2527                                       int64_t bytes, BdrvRequestFlags flags)
2528 {
2529     IO_OR_GS_CODE();
2530     return blk_co_pwritev(blk, offset, bytes, NULL,
2531                           flags | BDRV_REQ_ZERO_WRITE);
2532 }
2533 
2534 int coroutine_fn blk_co_pwrite_compressed(BlockBackend *blk, int64_t offset,
2535                                           int64_t bytes, const void *buf)
2536 {
2537     QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
2538     IO_OR_GS_CODE();
2539     return blk_co_pwritev_part(blk, offset, bytes, &qiov, 0,
2540                                BDRV_REQ_WRITE_COMPRESSED);
2541 }
2542 
2543 int coroutine_fn blk_co_truncate(BlockBackend *blk, int64_t offset, bool exact,
2544                                  PreallocMode prealloc, BdrvRequestFlags flags,
2545                                  Error **errp)
2546 {
2547     IO_OR_GS_CODE();
2548     GRAPH_RDLOCK_GUARD();
2549     if (!blk_co_is_available(blk)) {
2550         error_setg(errp, "No medium inserted");
2551         return -ENOMEDIUM;
2552     }
2553 
2554     return bdrv_co_truncate(blk->root, offset, exact, prealloc, flags, errp);
2555 }
2556 
2557 int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
2558                      int64_t pos, int size)
2559 {
2560     int ret;
2561     GLOBAL_STATE_CODE();
2562 
2563     if (!blk_is_available(blk)) {
2564         return -ENOMEDIUM;
2565     }
2566 
2567     ret = bdrv_save_vmstate(blk_bs(blk), buf, pos, size);
2568     if (ret < 0) {
2569         return ret;
2570     }
2571 
2572     if (ret == size && !blk->enable_write_cache) {
2573         ret = bdrv_flush(blk_bs(blk));
2574     }
2575 
2576     return ret < 0 ? ret : size;
2577 }
2578 
2579 int blk_load_vmstate(BlockBackend *blk, uint8_t *buf, int64_t pos, int size)
2580 {
2581     GLOBAL_STATE_CODE();
2582     if (!blk_is_available(blk)) {
2583         return -ENOMEDIUM;
2584     }
2585 
2586     return bdrv_load_vmstate(blk_bs(blk), buf, pos, size);
2587 }
2588 
2589 int blk_probe_blocksizes(BlockBackend *blk, BlockSizes *bsz)
2590 {
2591     GLOBAL_STATE_CODE();
2592     GRAPH_RDLOCK_GUARD_MAINLOOP();
2593 
2594     if (!blk_is_available(blk)) {
2595         return -ENOMEDIUM;
2596     }
2597 
2598     return bdrv_probe_blocksizes(blk_bs(blk), bsz);
2599 }
2600 
2601 int blk_probe_geometry(BlockBackend *blk, HDGeometry *geo)
2602 {
2603     GLOBAL_STATE_CODE();
2604     if (!blk_is_available(blk)) {
2605         return -ENOMEDIUM;
2606     }
2607 
2608     return bdrv_probe_geometry(blk_bs(blk), geo);
2609 }
2610 
2611 /*
2612  * Updates the BlockBackendRootState object with data from the currently
2613  * attached BlockDriverState.
2614  */
2615 void blk_update_root_state(BlockBackend *blk)
2616 {
2617     GLOBAL_STATE_CODE();
2618     assert(blk->root);
2619 
2620     blk->root_state.open_flags    = blk->root->bs->open_flags;
2621     blk->root_state.detect_zeroes = blk->root->bs->detect_zeroes;
2622 }
2623 
2624 /*
2625  * Returns the detect-zeroes setting to be used for bdrv_open() of a
2626  * BlockDriverState which is supposed to inherit the root state.
2627  */
2628 bool blk_get_detect_zeroes_from_root_state(BlockBackend *blk)
2629 {
2630     GLOBAL_STATE_CODE();
2631     return blk->root_state.detect_zeroes;
2632 }
2633 
2634 /*
2635  * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
2636  * supposed to inherit the root state.
2637  */
2638 int blk_get_open_flags_from_root_state(BlockBackend *blk)
2639 {
2640     GLOBAL_STATE_CODE();
2641     return blk->root_state.open_flags;
2642 }
2643 
2644 BlockBackendRootState *blk_get_root_state(BlockBackend *blk)
2645 {
2646     GLOBAL_STATE_CODE();
2647     return &blk->root_state;
2648 }
2649 
2650 int blk_commit_all(void)
2651 {
2652     BlockBackend *blk = NULL;
2653     GLOBAL_STATE_CODE();
2654     GRAPH_RDLOCK_GUARD_MAINLOOP();
2655 
2656     while ((blk = blk_all_next(blk)) != NULL) {
2657         BlockDriverState *unfiltered_bs = bdrv_skip_filters(blk_bs(blk));
2658 
2659         if (blk_is_inserted(blk) && bdrv_cow_child(unfiltered_bs)) {
2660             int ret;
2661 
2662             ret = bdrv_commit(unfiltered_bs);
2663             if (ret < 0) {
2664                 return ret;
2665             }
2666         }
2667     }
2668     return 0;
2669 }
2670 
2671 
2672 /* throttling disk I/O limits */
2673 void blk_set_io_limits(BlockBackend *blk, ThrottleConfig *cfg)
2674 {
2675     GLOBAL_STATE_CODE();
2676     throttle_group_config(&blk->public.throttle_group_member, cfg);
2677 }
2678 
2679 void blk_io_limits_disable(BlockBackend *blk)
2680 {
2681     BlockDriverState *bs = blk_bs(blk);
2682     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
2683     assert(tgm->throttle_state);
2684     GLOBAL_STATE_CODE();
2685     if (bs) {
2686         bdrv_ref(bs);
2687         bdrv_drained_begin(bs);
2688     }
2689     throttle_group_unregister_tgm(tgm);
2690     if (bs) {
2691         bdrv_drained_end(bs);
2692         bdrv_unref(bs);
2693     }
2694 }
2695 
2696 /* should be called before blk_set_io_limits if a limit is set */
2697 void blk_io_limits_enable(BlockBackend *blk, const char *group)
2698 {
2699     assert(!blk->public.throttle_group_member.throttle_state);
2700     GLOBAL_STATE_CODE();
2701     throttle_group_register_tgm(&blk->public.throttle_group_member,
2702                                 group, blk_get_aio_context(blk));
2703 }
2704 
2705 void blk_io_limits_update_group(BlockBackend *blk, const char *group)
2706 {
2707     GLOBAL_STATE_CODE();
2708     /* this BB is not part of any group */
2709     if (!blk->public.throttle_group_member.throttle_state) {
2710         return;
2711     }
2712 
2713     /* this BB is a part of the same group than the one we want */
2714     if (!g_strcmp0(throttle_group_get_name(&blk->public.throttle_group_member),
2715                 group)) {
2716         return;
2717     }
2718 
2719     /* need to change the group this bs belong to */
2720     blk_io_limits_disable(blk);
2721     blk_io_limits_enable(blk, group);
2722 }
2723 
2724 static void blk_root_drained_begin(BdrvChild *child)
2725 {
2726     BlockBackend *blk = child->opaque;
2727     ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
2728 
2729     if (qatomic_fetch_inc(&blk->quiesce_counter) == 0) {
2730         if (blk->dev_ops && blk->dev_ops->drained_begin) {
2731             blk->dev_ops->drained_begin(blk->dev_opaque);
2732         }
2733     }
2734 
2735     /* Note that blk->root may not be accessible here yet if we are just
2736      * attaching to a BlockDriverState that is drained. Use child instead. */
2737 
2738     if (qatomic_fetch_inc(&tgm->io_limits_disabled) == 0) {
2739         throttle_group_restart_tgm(tgm);
2740     }
2741 }
2742 
2743 static bool blk_root_drained_poll(BdrvChild *child)
2744 {
2745     BlockBackend *blk = child->opaque;
2746     bool busy = false;
2747     assert(qatomic_read(&blk->quiesce_counter));
2748 
2749     if (blk->dev_ops && blk->dev_ops->drained_poll) {
2750         busy = blk->dev_ops->drained_poll(blk->dev_opaque);
2751     }
2752     return busy || !!blk->in_flight;
2753 }
2754 
2755 static void blk_root_drained_end(BdrvChild *child)
2756 {
2757     BlockBackend *blk = child->opaque;
2758     assert(qatomic_read(&blk->quiesce_counter));
2759 
2760     assert(blk->public.throttle_group_member.io_limits_disabled);
2761     qatomic_dec(&blk->public.throttle_group_member.io_limits_disabled);
2762 
2763     if (qatomic_fetch_dec(&blk->quiesce_counter) == 1) {
2764         if (blk->dev_ops && blk->dev_ops->drained_end) {
2765             blk->dev_ops->drained_end(blk->dev_opaque);
2766         }
2767         qemu_mutex_lock(&blk->queued_requests_lock);
2768         while (qemu_co_enter_next(&blk->queued_requests,
2769                                   &blk->queued_requests_lock)) {
2770             /* Resume all queued requests */
2771         }
2772         qemu_mutex_unlock(&blk->queued_requests_lock);
2773     }
2774 }
2775 
2776 bool blk_register_buf(BlockBackend *blk, void *host, size_t size, Error **errp)
2777 {
2778     BlockDriverState *bs = blk_bs(blk);
2779 
2780     GLOBAL_STATE_CODE();
2781 
2782     if (bs) {
2783         return bdrv_register_buf(bs, host, size, errp);
2784     }
2785     return true;
2786 }
2787 
2788 void blk_unregister_buf(BlockBackend *blk, void *host, size_t size)
2789 {
2790     BlockDriverState *bs = blk_bs(blk);
2791 
2792     GLOBAL_STATE_CODE();
2793 
2794     if (bs) {
2795         bdrv_unregister_buf(bs, host, size);
2796     }
2797 }
2798 
2799 int coroutine_fn blk_co_copy_range(BlockBackend *blk_in, int64_t off_in,
2800                                    BlockBackend *blk_out, int64_t off_out,
2801                                    int64_t bytes, BdrvRequestFlags read_flags,
2802                                    BdrvRequestFlags write_flags)
2803 {
2804     int r;
2805     IO_CODE();
2806     GRAPH_RDLOCK_GUARD();
2807 
2808     r = blk_check_byte_request(blk_in, off_in, bytes);
2809     if (r) {
2810         return r;
2811     }
2812     r = blk_check_byte_request(blk_out, off_out, bytes);
2813     if (r) {
2814         return r;
2815     }
2816 
2817     return bdrv_co_copy_range(blk_in->root, off_in,
2818                               blk_out->root, off_out,
2819                               bytes, read_flags, write_flags);
2820 }
2821 
2822 const BdrvChild *blk_root(BlockBackend *blk)
2823 {
2824     GLOBAL_STATE_CODE();
2825     return blk->root;
2826 }
2827 
2828 int blk_make_empty(BlockBackend *blk, Error **errp)
2829 {
2830     GLOBAL_STATE_CODE();
2831     GRAPH_RDLOCK_GUARD_MAINLOOP();
2832 
2833     if (!blk_is_available(blk)) {
2834         error_setg(errp, "No medium inserted");
2835         return -ENOMEDIUM;
2836     }
2837 
2838     return bdrv_make_empty(blk->root, errp);
2839 }
2840