block: User BdrvChild callback for device name
[qemu.git] / block / block-backend.c
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/throttle-groups.h"
18 #include "sysemu/blockdev.h"
19 #include "sysemu/sysemu.h"
20 #include "qapi-event.h"
21 #include "qemu/id.h"
22
23 /* Number of coroutines to reserve per attached device model */
24 #define COROUTINE_POOL_RESERVATION 64
25
26 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
27
28 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb);
29
30 struct BlockBackend {
31 char *name;
32 int refcnt;
33 BdrvChild *root;
34 DriveInfo *legacy_dinfo; /* null unless created by drive_new() */
35 QTAILQ_ENTRY(BlockBackend) link; /* for block_backends */
36 QTAILQ_ENTRY(BlockBackend) monitor_link; /* for monitor_block_backends */
37 BlockBackendPublic public;
38
39 void *dev; /* attached device model, if any */
40 /* TODO change to DeviceState when all users are qdevified */
41 const BlockDevOps *dev_ops;
42 void *dev_opaque;
43
44 /* the block size for which the guest device expects atomicity */
45 int guest_block_size;
46
47 /* If the BDS tree is removed, some of its options are stored here (which
48 * can be used to restore those options in the new BDS on insert) */
49 BlockBackendRootState root_state;
50
51 bool enable_write_cache;
52
53 /* I/O stats (display with "info blockstats"). */
54 BlockAcctStats stats;
55
56 BlockdevOnError on_read_error, on_write_error;
57 bool iostatus_enabled;
58 BlockDeviceIoStatus iostatus;
59
60 bool allow_write_beyond_eof;
61
62 NotifierList remove_bs_notifiers, insert_bs_notifiers;
63 };
64
65 typedef struct BlockBackendAIOCB {
66 BlockAIOCB common;
67 QEMUBH *bh;
68 BlockBackend *blk;
69 int ret;
70 } BlockBackendAIOCB;
71
72 static const AIOCBInfo block_backend_aiocb_info = {
73 .get_aio_context = blk_aiocb_get_aio_context,
74 .aiocb_size = sizeof(BlockBackendAIOCB),
75 };
76
77 static void drive_info_del(DriveInfo *dinfo);
78
79 /* All BlockBackends */
80 static QTAILQ_HEAD(, BlockBackend) block_backends =
81 QTAILQ_HEAD_INITIALIZER(block_backends);
82
83 /* All BlockBackends referenced by the monitor and which are iterated through by
84 * blk_next() */
85 static QTAILQ_HEAD(, BlockBackend) monitor_block_backends =
86 QTAILQ_HEAD_INITIALIZER(monitor_block_backends);
87
88 static void blk_root_inherit_options(int *child_flags, QDict *child_options,
89 int parent_flags, QDict *parent_options)
90 {
91 /* We're not supposed to call this function for root nodes */
92 abort();
93 }
94 static void blk_root_drained_begin(BdrvChild *child);
95 static void blk_root_drained_end(BdrvChild *child);
96
97 static void blk_root_change_media(BdrvChild *child, bool load);
98 static void blk_root_resize(BdrvChild *child);
99
100 static const char *blk_root_get_name(BdrvChild *child)
101 {
102 return blk_name(child->opaque);
103 }
104
105 static const BdrvChildRole child_root = {
106 .inherit_options = blk_root_inherit_options,
107
108 .change_media = blk_root_change_media,
109 .resize = blk_root_resize,
110 .get_name = blk_root_get_name,
111
112 .drained_begin = blk_root_drained_begin,
113 .drained_end = blk_root_drained_end,
114 };
115
116 /*
117 * Create a new BlockBackend with a reference count of one.
118 * Store an error through @errp on failure, unless it's null.
119 * Return the new BlockBackend on success, null on failure.
120 */
121 BlockBackend *blk_new(Error **errp)
122 {
123 BlockBackend *blk;
124
125 blk = g_new0(BlockBackend, 1);
126 blk->refcnt = 1;
127 qemu_co_queue_init(&blk->public.throttled_reqs[0]);
128 qemu_co_queue_init(&blk->public.throttled_reqs[1]);
129
130 notifier_list_init(&blk->remove_bs_notifiers);
131 notifier_list_init(&blk->insert_bs_notifiers);
132
133 QTAILQ_INSERT_TAIL(&block_backends, blk, link);
134 return blk;
135 }
136
137 /*
138 * Create a new BlockBackend with a new BlockDriverState attached.
139 * Otherwise just like blk_new(), which see.
140 */
141 BlockBackend *blk_new_with_bs(Error **errp)
142 {
143 BlockBackend *blk;
144 BlockDriverState *bs;
145
146 blk = blk_new(errp);
147 if (!blk) {
148 return NULL;
149 }
150
151 bs = bdrv_new_root();
152 blk->root = bdrv_root_attach_child(bs, "root", &child_root);
153 blk->root->opaque = blk;
154 bs->blk = blk;
155 return blk;
156 }
157
158 /*
159 * Calls blk_new_with_bs() and then calls bdrv_open() on the BlockDriverState.
160 *
161 * Just as with bdrv_open(), after having called this function the reference to
162 * @options belongs to the block layer (even on failure).
163 *
164 * TODO: Remove @filename and @flags; it should be possible to specify a whole
165 * BDS tree just by specifying the @options QDict (or @reference,
166 * alternatively). At the time of adding this function, this is not possible,
167 * though, so callers of this function have to be able to specify @filename and
168 * @flags.
169 */
170 BlockBackend *blk_new_open(const char *filename, const char *reference,
171 QDict *options, int flags, Error **errp)
172 {
173 BlockBackend *blk;
174 int ret;
175
176 blk = blk_new_with_bs(errp);
177 if (!blk) {
178 QDECREF(options);
179 return NULL;
180 }
181
182 ret = bdrv_open(&blk->root->bs, filename, reference, options, flags, errp);
183 if (ret < 0) {
184 blk_unref(blk);
185 return NULL;
186 }
187
188 blk_set_enable_write_cache(blk, true);
189
190 return blk;
191 }
192
193 static void blk_delete(BlockBackend *blk)
194 {
195 assert(!blk->refcnt);
196 assert(!blk->name);
197 assert(!blk->dev);
198 if (blk->root) {
199 blk_remove_bs(blk);
200 }
201 assert(QLIST_EMPTY(&blk->remove_bs_notifiers.notifiers));
202 assert(QLIST_EMPTY(&blk->insert_bs_notifiers.notifiers));
203 QTAILQ_REMOVE(&block_backends, blk, link);
204 drive_info_del(blk->legacy_dinfo);
205 block_acct_cleanup(&blk->stats);
206 g_free(blk);
207 }
208
209 static void drive_info_del(DriveInfo *dinfo)
210 {
211 if (!dinfo) {
212 return;
213 }
214 qemu_opts_del(dinfo->opts);
215 g_free(dinfo->serial);
216 g_free(dinfo);
217 }
218
219 int blk_get_refcnt(BlockBackend *blk)
220 {
221 return blk ? blk->refcnt : 0;
222 }
223
224 /*
225 * Increment @blk's reference count.
226 * @blk must not be null.
227 */
228 void blk_ref(BlockBackend *blk)
229 {
230 blk->refcnt++;
231 }
232
233 /*
234 * Decrement @blk's reference count.
235 * If this drops it to zero, destroy @blk.
236 * For convenience, do nothing if @blk is null.
237 */
238 void blk_unref(BlockBackend *blk)
239 {
240 if (blk) {
241 assert(blk->refcnt > 0);
242 if (!--blk->refcnt) {
243 blk_delete(blk);
244 }
245 }
246 }
247
248 /*
249 * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
250 * ones which are hidden (i.e. are not referenced by the monitor).
251 */
252 static BlockBackend *blk_all_next(BlockBackend *blk)
253 {
254 return blk ? QTAILQ_NEXT(blk, link)
255 : QTAILQ_FIRST(&block_backends);
256 }
257
258 void blk_remove_all_bs(void)
259 {
260 BlockBackend *blk = NULL;
261
262 while ((blk = blk_all_next(blk)) != NULL) {
263 AioContext *ctx = blk_get_aio_context(blk);
264
265 aio_context_acquire(ctx);
266 if (blk->root) {
267 blk_remove_bs(blk);
268 }
269 aio_context_release(ctx);
270 }
271 }
272
273 /*
274 * Return the monitor-owned BlockBackend after @blk.
275 * If @blk is null, return the first one.
276 * Else, return @blk's next sibling, which may be null.
277 *
278 * To iterate over all BlockBackends, do
279 * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
280 * ...
281 * }
282 */
283 BlockBackend *blk_next(BlockBackend *blk)
284 {
285 return blk ? QTAILQ_NEXT(blk, monitor_link)
286 : QTAILQ_FIRST(&monitor_block_backends);
287 }
288
289 /*
290 * Iterates over all BlockDriverStates which are attached to a BlockBackend.
291 * This function is for use by bdrv_next().
292 *
293 * @bs must be NULL or a BDS that is attached to a BB.
294 */
295 BlockDriverState *blk_next_root_bs(BlockDriverState *bs)
296 {
297 BlockBackend *blk;
298
299 if (bs) {
300 assert(bs->blk);
301 blk = bs->blk;
302 } else {
303 blk = NULL;
304 }
305
306 do {
307 blk = blk_all_next(blk);
308 } while (blk && !blk->root);
309
310 return blk ? blk->root->bs : NULL;
311 }
312
313 /*
314 * Add a BlockBackend into the list of backends referenced by the monitor, with
315 * the given @name acting as the handle for the monitor.
316 * Strictly for use by blockdev.c.
317 *
318 * @name must not be null or empty.
319 *
320 * Returns true on success and false on failure. In the latter case, an Error
321 * object is returned through @errp.
322 */
323 bool monitor_add_blk(BlockBackend *blk, const char *name, Error **errp)
324 {
325 assert(!blk->name);
326 assert(name && name[0]);
327
328 if (!id_wellformed(name)) {
329 error_setg(errp, "Invalid device name");
330 return false;
331 }
332 if (blk_by_name(name)) {
333 error_setg(errp, "Device with id '%s' already exists", name);
334 return false;
335 }
336 if (bdrv_find_node(name)) {
337 error_setg(errp,
338 "Device name '%s' conflicts with an existing node name",
339 name);
340 return false;
341 }
342
343 blk->name = g_strdup(name);
344 QTAILQ_INSERT_TAIL(&monitor_block_backends, blk, monitor_link);
345 return true;
346 }
347
348 /*
349 * Remove a BlockBackend from the list of backends referenced by the monitor.
350 * Strictly for use by blockdev.c.
351 */
352 void monitor_remove_blk(BlockBackend *blk)
353 {
354 if (!blk->name) {
355 return;
356 }
357
358 QTAILQ_REMOVE(&monitor_block_backends, blk, monitor_link);
359 g_free(blk->name);
360 blk->name = NULL;
361 }
362
363 /*
364 * Return @blk's name, a non-null string.
365 * Returns an empty string iff @blk is not referenced by the monitor.
366 */
367 const char *blk_name(BlockBackend *blk)
368 {
369 return blk->name ?: "";
370 }
371
372 /*
373 * Return the BlockBackend with name @name if it exists, else null.
374 * @name must not be null.
375 */
376 BlockBackend *blk_by_name(const char *name)
377 {
378 BlockBackend *blk = NULL;
379
380 assert(name);
381 while ((blk = blk_next(blk)) != NULL) {
382 if (!strcmp(name, blk->name)) {
383 return blk;
384 }
385 }
386 return NULL;
387 }
388
389 /*
390 * Return the BlockDriverState attached to @blk if any, else null.
391 */
392 BlockDriverState *blk_bs(BlockBackend *blk)
393 {
394 return blk->root ? blk->root->bs : NULL;
395 }
396
397 /*
398 * Return @blk's DriveInfo if any, else null.
399 */
400 DriveInfo *blk_legacy_dinfo(BlockBackend *blk)
401 {
402 return blk->legacy_dinfo;
403 }
404
405 /*
406 * Set @blk's DriveInfo to @dinfo, and return it.
407 * @blk must not have a DriveInfo set already.
408 * No other BlockBackend may have the same DriveInfo set.
409 */
410 DriveInfo *blk_set_legacy_dinfo(BlockBackend *blk, DriveInfo *dinfo)
411 {
412 assert(!blk->legacy_dinfo);
413 return blk->legacy_dinfo = dinfo;
414 }
415
416 /*
417 * Return the BlockBackend with DriveInfo @dinfo.
418 * It must exist.
419 */
420 BlockBackend *blk_by_legacy_dinfo(DriveInfo *dinfo)
421 {
422 BlockBackend *blk = NULL;
423
424 while ((blk = blk_next(blk)) != NULL) {
425 if (blk->legacy_dinfo == dinfo) {
426 return blk;
427 }
428 }
429 abort();
430 }
431
432 /*
433 * Returns a pointer to the publicly accessible fields of @blk.
434 */
435 BlockBackendPublic *blk_get_public(BlockBackend *blk)
436 {
437 return &blk->public;
438 }
439
440 /*
441 * Returns a BlockBackend given the associated @public fields.
442 */
443 BlockBackend *blk_by_public(BlockBackendPublic *public)
444 {
445 return container_of(public, BlockBackend, public);
446 }
447
448 /*
449 * Disassociates the currently associated BlockDriverState from @blk.
450 */
451 void blk_remove_bs(BlockBackend *blk)
452 {
453 assert(blk->root->bs->blk == blk);
454
455 notifier_list_notify(&blk->remove_bs_notifiers, blk);
456 if (blk->public.throttle_state) {
457 throttle_timers_detach_aio_context(&blk->public.throttle_timers);
458 }
459
460 blk_update_root_state(blk);
461
462 blk->root->bs->blk = NULL;
463 bdrv_root_unref_child(blk->root);
464 blk->root = NULL;
465 }
466
467 /*
468 * Associates a new BlockDriverState with @blk.
469 */
470 void blk_insert_bs(BlockBackend *blk, BlockDriverState *bs)
471 {
472 assert(!blk->root && !bs->blk);
473 bdrv_ref(bs);
474 blk->root = bdrv_root_attach_child(bs, "root", &child_root);
475 blk->root->opaque = blk;
476 bs->blk = blk;
477
478 notifier_list_notify(&blk->insert_bs_notifiers, blk);
479 if (blk->public.throttle_state) {
480 throttle_timers_attach_aio_context(
481 &blk->public.throttle_timers, bdrv_get_aio_context(bs));
482 }
483 }
484
485 /*
486 * Attach device model @dev to @blk.
487 * Return 0 on success, -EBUSY when a device model is attached already.
488 */
489 int blk_attach_dev(BlockBackend *blk, void *dev)
490 /* TODO change to DeviceState *dev when all users are qdevified */
491 {
492 if (blk->dev) {
493 return -EBUSY;
494 }
495 blk_ref(blk);
496 blk->dev = dev;
497 blk_iostatus_reset(blk);
498 return 0;
499 }
500
501 /*
502 * Attach device model @dev to @blk.
503 * @blk must not have a device model attached already.
504 * TODO qdevified devices don't use this, remove when devices are qdevified
505 */
506 void blk_attach_dev_nofail(BlockBackend *blk, void *dev)
507 {
508 if (blk_attach_dev(blk, dev) < 0) {
509 abort();
510 }
511 }
512
513 /*
514 * Detach device model @dev from @blk.
515 * @dev must be currently attached to @blk.
516 */
517 void blk_detach_dev(BlockBackend *blk, void *dev)
518 /* TODO change to DeviceState *dev when all users are qdevified */
519 {
520 assert(blk->dev == dev);
521 blk->dev = NULL;
522 blk->dev_ops = NULL;
523 blk->dev_opaque = NULL;
524 blk->guest_block_size = 512;
525 blk_unref(blk);
526 }
527
528 /*
529 * Return the device model attached to @blk if any, else null.
530 */
531 void *blk_get_attached_dev(BlockBackend *blk)
532 /* TODO change to return DeviceState * when all users are qdevified */
533 {
534 return blk->dev;
535 }
536
537 /*
538 * Set @blk's device model callbacks to @ops.
539 * @opaque is the opaque argument to pass to the callbacks.
540 * This is for use by device models.
541 */
542 void blk_set_dev_ops(BlockBackend *blk, const BlockDevOps *ops,
543 void *opaque)
544 {
545 blk->dev_ops = ops;
546 blk->dev_opaque = opaque;
547 }
548
549 /*
550 * Notify @blk's attached device model of media change.
551 * If @load is true, notify of media load.
552 * Else, notify of media eject.
553 * Also send DEVICE_TRAY_MOVED events as appropriate.
554 */
555 void blk_dev_change_media_cb(BlockBackend *blk, bool load)
556 {
557 if (blk->dev_ops && blk->dev_ops->change_media_cb) {
558 bool tray_was_open, tray_is_open;
559
560 tray_was_open = blk_dev_is_tray_open(blk);
561 blk->dev_ops->change_media_cb(blk->dev_opaque, load);
562 tray_is_open = blk_dev_is_tray_open(blk);
563
564 if (tray_was_open != tray_is_open) {
565 qapi_event_send_device_tray_moved(blk_name(blk), tray_is_open,
566 &error_abort);
567 }
568 }
569 }
570
571 static void blk_root_change_media(BdrvChild *child, bool load)
572 {
573 blk_dev_change_media_cb(child->opaque, load);
574 }
575
576 /*
577 * Does @blk's attached device model have removable media?
578 * %true if no device model is attached.
579 */
580 bool blk_dev_has_removable_media(BlockBackend *blk)
581 {
582 return !blk->dev || (blk->dev_ops && blk->dev_ops->change_media_cb);
583 }
584
585 /*
586 * Does @blk's attached device model have a tray?
587 */
588 bool blk_dev_has_tray(BlockBackend *blk)
589 {
590 return blk->dev_ops && blk->dev_ops->is_tray_open;
591 }
592
593 /*
594 * Notify @blk's attached device model of a media eject request.
595 * If @force is true, the medium is about to be yanked out forcefully.
596 */
597 void blk_dev_eject_request(BlockBackend *blk, bool force)
598 {
599 if (blk->dev_ops && blk->dev_ops->eject_request_cb) {
600 blk->dev_ops->eject_request_cb(blk->dev_opaque, force);
601 }
602 }
603
604 /*
605 * Does @blk's attached device model have a tray, and is it open?
606 */
607 bool blk_dev_is_tray_open(BlockBackend *blk)
608 {
609 if (blk_dev_has_tray(blk)) {
610 return blk->dev_ops->is_tray_open(blk->dev_opaque);
611 }
612 return false;
613 }
614
615 /*
616 * Does @blk's attached device model have the medium locked?
617 * %false if the device model has no such lock.
618 */
619 bool blk_dev_is_medium_locked(BlockBackend *blk)
620 {
621 if (blk->dev_ops && blk->dev_ops->is_medium_locked) {
622 return blk->dev_ops->is_medium_locked(blk->dev_opaque);
623 }
624 return false;
625 }
626
627 /*
628 * Notify @blk's attached device model of a backend size change.
629 */
630 static void blk_root_resize(BdrvChild *child)
631 {
632 BlockBackend *blk = child->opaque;
633
634 if (blk->dev_ops && blk->dev_ops->resize_cb) {
635 blk->dev_ops->resize_cb(blk->dev_opaque);
636 }
637 }
638
639 void blk_iostatus_enable(BlockBackend *blk)
640 {
641 blk->iostatus_enabled = true;
642 blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
643 }
644
645 /* The I/O status is only enabled if the drive explicitly
646 * enables it _and_ the VM is configured to stop on errors */
647 bool blk_iostatus_is_enabled(const BlockBackend *blk)
648 {
649 return (blk->iostatus_enabled &&
650 (blk->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
651 blk->on_write_error == BLOCKDEV_ON_ERROR_STOP ||
652 blk->on_read_error == BLOCKDEV_ON_ERROR_STOP));
653 }
654
655 BlockDeviceIoStatus blk_iostatus(const BlockBackend *blk)
656 {
657 return blk->iostatus;
658 }
659
660 void blk_iostatus_disable(BlockBackend *blk)
661 {
662 blk->iostatus_enabled = false;
663 }
664
665 void blk_iostatus_reset(BlockBackend *blk)
666 {
667 if (blk_iostatus_is_enabled(blk)) {
668 BlockDriverState *bs = blk_bs(blk);
669 blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
670 if (bs && bs->job) {
671 block_job_iostatus_reset(bs->job);
672 }
673 }
674 }
675
676 void blk_iostatus_set_err(BlockBackend *blk, int error)
677 {
678 assert(blk_iostatus_is_enabled(blk));
679 if (blk->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
680 blk->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
681 BLOCK_DEVICE_IO_STATUS_FAILED;
682 }
683 }
684
685 void blk_set_allow_write_beyond_eof(BlockBackend *blk, bool allow)
686 {
687 blk->allow_write_beyond_eof = allow;
688 }
689
690 static int blk_check_byte_request(BlockBackend *blk, int64_t offset,
691 size_t size)
692 {
693 int64_t len;
694
695 if (size > INT_MAX) {
696 return -EIO;
697 }
698
699 if (!blk_is_available(blk)) {
700 return -ENOMEDIUM;
701 }
702
703 if (offset < 0) {
704 return -EIO;
705 }
706
707 if (!blk->allow_write_beyond_eof) {
708 len = blk_getlength(blk);
709 if (len < 0) {
710 return len;
711 }
712
713 if (offset > len || len - offset < size) {
714 return -EIO;
715 }
716 }
717
718 return 0;
719 }
720
721 static int blk_check_request(BlockBackend *blk, int64_t sector_num,
722 int nb_sectors)
723 {
724 if (sector_num < 0 || sector_num > INT64_MAX / BDRV_SECTOR_SIZE) {
725 return -EIO;
726 }
727
728 if (nb_sectors < 0 || nb_sectors > INT_MAX / BDRV_SECTOR_SIZE) {
729 return -EIO;
730 }
731
732 return blk_check_byte_request(blk, sector_num * BDRV_SECTOR_SIZE,
733 nb_sectors * BDRV_SECTOR_SIZE);
734 }
735
736 static int coroutine_fn blk_co_preadv(BlockBackend *blk, int64_t offset,
737 unsigned int bytes, QEMUIOVector *qiov,
738 BdrvRequestFlags flags)
739 {
740 int ret = blk_check_byte_request(blk, offset, bytes);
741 if (ret < 0) {
742 return ret;
743 }
744
745 /* throttling disk I/O */
746 if (blk->public.throttle_state) {
747 throttle_group_co_io_limits_intercept(blk, bytes, false);
748 }
749
750 return bdrv_co_preadv(blk_bs(blk), offset, bytes, qiov, flags);
751 }
752
753 static int coroutine_fn blk_co_pwritev(BlockBackend *blk, int64_t offset,
754 unsigned int bytes, QEMUIOVector *qiov,
755 BdrvRequestFlags flags)
756 {
757 int ret;
758
759 ret = blk_check_byte_request(blk, offset, bytes);
760 if (ret < 0) {
761 return ret;
762 }
763
764 /* throttling disk I/O */
765 if (blk->public.throttle_state) {
766 throttle_group_co_io_limits_intercept(blk, bytes, true);
767 }
768
769 if (!blk->enable_write_cache) {
770 flags |= BDRV_REQ_FUA;
771 }
772
773 return bdrv_co_pwritev(blk_bs(blk), offset, bytes, qiov, flags);
774 }
775
776 typedef struct BlkRwCo {
777 BlockBackend *blk;
778 int64_t offset;
779 QEMUIOVector *qiov;
780 int ret;
781 BdrvRequestFlags flags;
782 } BlkRwCo;
783
784 static void blk_read_entry(void *opaque)
785 {
786 BlkRwCo *rwco = opaque;
787
788 rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, rwco->qiov->size,
789 rwco->qiov, rwco->flags);
790 }
791
792 static void blk_write_entry(void *opaque)
793 {
794 BlkRwCo *rwco = opaque;
795
796 rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, rwco->qiov->size,
797 rwco->qiov, rwco->flags);
798 }
799
800 static int blk_prw(BlockBackend *blk, int64_t offset, uint8_t *buf,
801 int64_t bytes, CoroutineEntry co_entry,
802 BdrvRequestFlags flags)
803 {
804 AioContext *aio_context;
805 QEMUIOVector qiov;
806 struct iovec iov;
807 Coroutine *co;
808 BlkRwCo rwco;
809
810 iov = (struct iovec) {
811 .iov_base = buf,
812 .iov_len = bytes,
813 };
814 qemu_iovec_init_external(&qiov, &iov, 1);
815
816 rwco = (BlkRwCo) {
817 .blk = blk,
818 .offset = offset,
819 .qiov = &qiov,
820 .flags = flags,
821 .ret = NOT_DONE,
822 };
823
824 co = qemu_coroutine_create(co_entry);
825 qemu_coroutine_enter(co, &rwco);
826
827 aio_context = blk_get_aio_context(blk);
828 while (rwco.ret == NOT_DONE) {
829 aio_poll(aio_context, true);
830 }
831
832 return rwco.ret;
833 }
834
835 int blk_pread_unthrottled(BlockBackend *blk, int64_t offset, uint8_t *buf,
836 int count)
837 {
838 int ret;
839
840 ret = blk_check_byte_request(blk, offset, count);
841 if (ret < 0) {
842 return ret;
843 }
844
845 blk_root_drained_begin(blk->root);
846 ret = blk_pread(blk, offset, buf, count);
847 blk_root_drained_end(blk->root);
848 return ret;
849 }
850
851 int blk_write_zeroes(BlockBackend *blk, int64_t offset,
852 int count, BdrvRequestFlags flags)
853 {
854 return blk_prw(blk, offset, NULL, count, blk_write_entry,
855 flags | BDRV_REQ_ZERO_WRITE);
856 }
857
858 static void error_callback_bh(void *opaque)
859 {
860 struct BlockBackendAIOCB *acb = opaque;
861 qemu_bh_delete(acb->bh);
862 acb->common.cb(acb->common.opaque, acb->ret);
863 qemu_aio_unref(acb);
864 }
865
866 BlockAIOCB *blk_abort_aio_request(BlockBackend *blk,
867 BlockCompletionFunc *cb,
868 void *opaque, int ret)
869 {
870 struct BlockBackendAIOCB *acb;
871 QEMUBH *bh;
872
873 acb = blk_aio_get(&block_backend_aiocb_info, blk, cb, opaque);
874 acb->blk = blk;
875 acb->ret = ret;
876
877 bh = aio_bh_new(blk_get_aio_context(blk), error_callback_bh, acb);
878 acb->bh = bh;
879 qemu_bh_schedule(bh);
880
881 return &acb->common;
882 }
883
884 typedef struct BlkAioEmAIOCB {
885 BlockAIOCB common;
886 BlkRwCo rwco;
887 int bytes;
888 bool has_returned;
889 QEMUBH* bh;
890 } BlkAioEmAIOCB;
891
892 static const AIOCBInfo blk_aio_em_aiocb_info = {
893 .aiocb_size = sizeof(BlkAioEmAIOCB),
894 };
895
896 static void blk_aio_complete(BlkAioEmAIOCB *acb)
897 {
898 if (acb->bh) {
899 assert(acb->has_returned);
900 qemu_bh_delete(acb->bh);
901 }
902 if (acb->has_returned) {
903 acb->common.cb(acb->common.opaque, acb->rwco.ret);
904 qemu_aio_unref(acb);
905 }
906 }
907
908 static void blk_aio_complete_bh(void *opaque)
909 {
910 blk_aio_complete(opaque);
911 }
912
913 static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset, int bytes,
914 QEMUIOVector *qiov, CoroutineEntry co_entry,
915 BdrvRequestFlags flags,
916 BlockCompletionFunc *cb, void *opaque)
917 {
918 BlkAioEmAIOCB *acb;
919 Coroutine *co;
920
921 acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
922 acb->rwco = (BlkRwCo) {
923 .blk = blk,
924 .offset = offset,
925 .qiov = qiov,
926 .flags = flags,
927 .ret = NOT_DONE,
928 };
929 acb->bytes = bytes;
930 acb->bh = NULL;
931 acb->has_returned = false;
932
933 co = qemu_coroutine_create(co_entry);
934 qemu_coroutine_enter(co, acb);
935
936 acb->has_returned = true;
937 if (acb->rwco.ret != NOT_DONE) {
938 acb->bh = aio_bh_new(blk_get_aio_context(blk), blk_aio_complete_bh, acb);
939 qemu_bh_schedule(acb->bh);
940 }
941
942 return &acb->common;
943 }
944
945 static void blk_aio_read_entry(void *opaque)
946 {
947 BlkAioEmAIOCB *acb = opaque;
948 BlkRwCo *rwco = &acb->rwco;
949
950 assert(rwco->qiov->size == acb->bytes);
951 rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, acb->bytes,
952 rwco->qiov, rwco->flags);
953 blk_aio_complete(acb);
954 }
955
956 static void blk_aio_write_entry(void *opaque)
957 {
958 BlkAioEmAIOCB *acb = opaque;
959 BlkRwCo *rwco = &acb->rwco;
960
961 assert(!rwco->qiov || rwco->qiov->size == acb->bytes);
962 rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, acb->bytes,
963 rwco->qiov, rwco->flags);
964 blk_aio_complete(acb);
965 }
966
967 BlockAIOCB *blk_aio_write_zeroes(BlockBackend *blk, int64_t offset,
968 int count, BdrvRequestFlags flags,
969 BlockCompletionFunc *cb, void *opaque)
970 {
971 return blk_aio_prwv(blk, offset, count, NULL, blk_aio_write_entry,
972 flags | BDRV_REQ_ZERO_WRITE, cb, opaque);
973 }
974
975 int blk_pread(BlockBackend *blk, int64_t offset, void *buf, int count)
976 {
977 int ret = blk_prw(blk, offset, buf, count, blk_read_entry, 0);
978 if (ret < 0) {
979 return ret;
980 }
981 return count;
982 }
983
984 int blk_pwrite(BlockBackend *blk, int64_t offset, const void *buf, int count,
985 BdrvRequestFlags flags)
986 {
987 int ret = blk_prw(blk, offset, (void *) buf, count, blk_write_entry,
988 flags);
989 if (ret < 0) {
990 return ret;
991 }
992 return count;
993 }
994
995 int64_t blk_getlength(BlockBackend *blk)
996 {
997 if (!blk_is_available(blk)) {
998 return -ENOMEDIUM;
999 }
1000
1001 return bdrv_getlength(blk_bs(blk));
1002 }
1003
1004 void blk_get_geometry(BlockBackend *blk, uint64_t *nb_sectors_ptr)
1005 {
1006 if (!blk_bs(blk)) {
1007 *nb_sectors_ptr = 0;
1008 } else {
1009 bdrv_get_geometry(blk_bs(blk), nb_sectors_ptr);
1010 }
1011 }
1012
1013 int64_t blk_nb_sectors(BlockBackend *blk)
1014 {
1015 if (!blk_is_available(blk)) {
1016 return -ENOMEDIUM;
1017 }
1018
1019 return bdrv_nb_sectors(blk_bs(blk));
1020 }
1021
1022 BlockAIOCB *blk_aio_preadv(BlockBackend *blk, int64_t offset,
1023 QEMUIOVector *qiov, BdrvRequestFlags flags,
1024 BlockCompletionFunc *cb, void *opaque)
1025 {
1026 return blk_aio_prwv(blk, offset, qiov->size, qiov,
1027 blk_aio_read_entry, flags, cb, opaque);
1028 }
1029
1030 BlockAIOCB *blk_aio_pwritev(BlockBackend *blk, int64_t offset,
1031 QEMUIOVector *qiov, BdrvRequestFlags flags,
1032 BlockCompletionFunc *cb, void *opaque)
1033 {
1034 return blk_aio_prwv(blk, offset, qiov->size, qiov,
1035 blk_aio_write_entry, flags, cb, opaque);
1036 }
1037
1038 BlockAIOCB *blk_aio_flush(BlockBackend *blk,
1039 BlockCompletionFunc *cb, void *opaque)
1040 {
1041 if (!blk_is_available(blk)) {
1042 return blk_abort_aio_request(blk, cb, opaque, -ENOMEDIUM);
1043 }
1044
1045 return bdrv_aio_flush(blk_bs(blk), cb, opaque);
1046 }
1047
1048 BlockAIOCB *blk_aio_discard(BlockBackend *blk,
1049 int64_t sector_num, int nb_sectors,
1050 BlockCompletionFunc *cb, void *opaque)
1051 {
1052 int ret = blk_check_request(blk, sector_num, nb_sectors);
1053 if (ret < 0) {
1054 return blk_abort_aio_request(blk, cb, opaque, ret);
1055 }
1056
1057 return bdrv_aio_discard(blk_bs(blk), sector_num, nb_sectors, cb, opaque);
1058 }
1059
1060 void blk_aio_cancel(BlockAIOCB *acb)
1061 {
1062 bdrv_aio_cancel(acb);
1063 }
1064
1065 void blk_aio_cancel_async(BlockAIOCB *acb)
1066 {
1067 bdrv_aio_cancel_async(acb);
1068 }
1069
1070 int blk_aio_multiwrite(BlockBackend *blk, BlockRequest *reqs, int num_reqs)
1071 {
1072 int i, ret;
1073
1074 for (i = 0; i < num_reqs; i++) {
1075 ret = blk_check_request(blk, reqs[i].sector, reqs[i].nb_sectors);
1076 if (ret < 0) {
1077 return ret;
1078 }
1079 }
1080
1081 return bdrv_aio_multiwrite(blk_bs(blk), reqs, num_reqs);
1082 }
1083
1084 int blk_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1085 {
1086 if (!blk_is_available(blk)) {
1087 return -ENOMEDIUM;
1088 }
1089
1090 return bdrv_ioctl(blk_bs(blk), req, buf);
1091 }
1092
1093 BlockAIOCB *blk_aio_ioctl(BlockBackend *blk, unsigned long int req, void *buf,
1094 BlockCompletionFunc *cb, void *opaque)
1095 {
1096 if (!blk_is_available(blk)) {
1097 return blk_abort_aio_request(blk, cb, opaque, -ENOMEDIUM);
1098 }
1099
1100 return bdrv_aio_ioctl(blk_bs(blk), req, buf, cb, opaque);
1101 }
1102
1103 int blk_co_discard(BlockBackend *blk, int64_t sector_num, int nb_sectors)
1104 {
1105 int ret = blk_check_request(blk, sector_num, nb_sectors);
1106 if (ret < 0) {
1107 return ret;
1108 }
1109
1110 return bdrv_co_discard(blk_bs(blk), sector_num, nb_sectors);
1111 }
1112
1113 int blk_co_flush(BlockBackend *blk)
1114 {
1115 if (!blk_is_available(blk)) {
1116 return -ENOMEDIUM;
1117 }
1118
1119 return bdrv_co_flush(blk_bs(blk));
1120 }
1121
1122 int blk_flush(BlockBackend *blk)
1123 {
1124 if (!blk_is_available(blk)) {
1125 return -ENOMEDIUM;
1126 }
1127
1128 return bdrv_flush(blk_bs(blk));
1129 }
1130
1131 void blk_drain(BlockBackend *blk)
1132 {
1133 if (blk_bs(blk)) {
1134 bdrv_drain(blk_bs(blk));
1135 }
1136 }
1137
1138 void blk_drain_all(void)
1139 {
1140 bdrv_drain_all();
1141 }
1142
1143 void blk_set_on_error(BlockBackend *blk, BlockdevOnError on_read_error,
1144 BlockdevOnError on_write_error)
1145 {
1146 blk->on_read_error = on_read_error;
1147 blk->on_write_error = on_write_error;
1148 }
1149
1150 BlockdevOnError blk_get_on_error(BlockBackend *blk, bool is_read)
1151 {
1152 return is_read ? blk->on_read_error : blk->on_write_error;
1153 }
1154
1155 BlockErrorAction blk_get_error_action(BlockBackend *blk, bool is_read,
1156 int error)
1157 {
1158 BlockdevOnError on_err = blk_get_on_error(blk, is_read);
1159
1160 switch (on_err) {
1161 case BLOCKDEV_ON_ERROR_ENOSPC:
1162 return (error == ENOSPC) ?
1163 BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
1164 case BLOCKDEV_ON_ERROR_STOP:
1165 return BLOCK_ERROR_ACTION_STOP;
1166 case BLOCKDEV_ON_ERROR_REPORT:
1167 return BLOCK_ERROR_ACTION_REPORT;
1168 case BLOCKDEV_ON_ERROR_IGNORE:
1169 return BLOCK_ERROR_ACTION_IGNORE;
1170 default:
1171 abort();
1172 }
1173 }
1174
1175 static void send_qmp_error_event(BlockBackend *blk,
1176 BlockErrorAction action,
1177 bool is_read, int error)
1178 {
1179 IoOperationType optype;
1180
1181 optype = is_read ? IO_OPERATION_TYPE_READ : IO_OPERATION_TYPE_WRITE;
1182 qapi_event_send_block_io_error(blk_name(blk), optype, action,
1183 blk_iostatus_is_enabled(blk),
1184 error == ENOSPC, strerror(error),
1185 &error_abort);
1186 }
1187
1188 /* This is done by device models because, while the block layer knows
1189 * about the error, it does not know whether an operation comes from
1190 * the device or the block layer (from a job, for example).
1191 */
1192 void blk_error_action(BlockBackend *blk, BlockErrorAction action,
1193 bool is_read, int error)
1194 {
1195 assert(error >= 0);
1196
1197 if (action == BLOCK_ERROR_ACTION_STOP) {
1198 /* First set the iostatus, so that "info block" returns an iostatus
1199 * that matches the events raised so far (an additional error iostatus
1200 * is fine, but not a lost one).
1201 */
1202 blk_iostatus_set_err(blk, error);
1203
1204 /* Then raise the request to stop the VM and the event.
1205 * qemu_system_vmstop_request_prepare has two effects. First,
1206 * it ensures that the STOP event always comes after the
1207 * BLOCK_IO_ERROR event. Second, it ensures that even if management
1208 * can observe the STOP event and do a "cont" before the STOP
1209 * event is issued, the VM will not stop. In this case, vm_start()
1210 * also ensures that the STOP/RESUME pair of events is emitted.
1211 */
1212 qemu_system_vmstop_request_prepare();
1213 send_qmp_error_event(blk, action, is_read, error);
1214 qemu_system_vmstop_request(RUN_STATE_IO_ERROR);
1215 } else {
1216 send_qmp_error_event(blk, action, is_read, error);
1217 }
1218 }
1219
1220 int blk_is_read_only(BlockBackend *blk)
1221 {
1222 BlockDriverState *bs = blk_bs(blk);
1223
1224 if (bs) {
1225 return bdrv_is_read_only(bs);
1226 } else {
1227 return blk->root_state.read_only;
1228 }
1229 }
1230
1231 int blk_is_sg(BlockBackend *blk)
1232 {
1233 BlockDriverState *bs = blk_bs(blk);
1234
1235 if (!bs) {
1236 return 0;
1237 }
1238
1239 return bdrv_is_sg(bs);
1240 }
1241
1242 int blk_enable_write_cache(BlockBackend *blk)
1243 {
1244 return blk->enable_write_cache;
1245 }
1246
1247 void blk_set_enable_write_cache(BlockBackend *blk, bool wce)
1248 {
1249 blk->enable_write_cache = wce;
1250 }
1251
1252 void blk_invalidate_cache(BlockBackend *blk, Error **errp)
1253 {
1254 BlockDriverState *bs = blk_bs(blk);
1255
1256 if (!bs) {
1257 error_setg(errp, "Device '%s' has no medium", blk->name);
1258 return;
1259 }
1260
1261 bdrv_invalidate_cache(bs, errp);
1262 }
1263
1264 bool blk_is_inserted(BlockBackend *blk)
1265 {
1266 BlockDriverState *bs = blk_bs(blk);
1267
1268 return bs && bdrv_is_inserted(bs);
1269 }
1270
1271 bool blk_is_available(BlockBackend *blk)
1272 {
1273 return blk_is_inserted(blk) && !blk_dev_is_tray_open(blk);
1274 }
1275
1276 void blk_lock_medium(BlockBackend *blk, bool locked)
1277 {
1278 BlockDriverState *bs = blk_bs(blk);
1279
1280 if (bs) {
1281 bdrv_lock_medium(bs, locked);
1282 }
1283 }
1284
1285 void blk_eject(BlockBackend *blk, bool eject_flag)
1286 {
1287 BlockDriverState *bs = blk_bs(blk);
1288
1289 if (bs) {
1290 bdrv_eject(bs, eject_flag);
1291 }
1292 }
1293
1294 int blk_get_flags(BlockBackend *blk)
1295 {
1296 BlockDriverState *bs = blk_bs(blk);
1297
1298 if (bs) {
1299 return bdrv_get_flags(bs);
1300 } else {
1301 return blk->root_state.open_flags;
1302 }
1303 }
1304
1305 int blk_get_max_transfer_length(BlockBackend *blk)
1306 {
1307 BlockDriverState *bs = blk_bs(blk);
1308
1309 if (bs) {
1310 return bs->bl.max_transfer_length;
1311 } else {
1312 return 0;
1313 }
1314 }
1315
1316 int blk_get_max_iov(BlockBackend *blk)
1317 {
1318 return blk->root->bs->bl.max_iov;
1319 }
1320
1321 void blk_set_guest_block_size(BlockBackend *blk, int align)
1322 {
1323 blk->guest_block_size = align;
1324 }
1325
1326 void *blk_try_blockalign(BlockBackend *blk, size_t size)
1327 {
1328 return qemu_try_blockalign(blk ? blk_bs(blk) : NULL, size);
1329 }
1330
1331 void *blk_blockalign(BlockBackend *blk, size_t size)
1332 {
1333 return qemu_blockalign(blk ? blk_bs(blk) : NULL, size);
1334 }
1335
1336 bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp)
1337 {
1338 BlockDriverState *bs = blk_bs(blk);
1339
1340 if (!bs) {
1341 return false;
1342 }
1343
1344 return bdrv_op_is_blocked(bs, op, errp);
1345 }
1346
1347 void blk_op_unblock(BlockBackend *blk, BlockOpType op, Error *reason)
1348 {
1349 BlockDriverState *bs = blk_bs(blk);
1350
1351 if (bs) {
1352 bdrv_op_unblock(bs, op, reason);
1353 }
1354 }
1355
1356 void blk_op_block_all(BlockBackend *blk, Error *reason)
1357 {
1358 BlockDriverState *bs = blk_bs(blk);
1359
1360 if (bs) {
1361 bdrv_op_block_all(bs, reason);
1362 }
1363 }
1364
1365 void blk_op_unblock_all(BlockBackend *blk, Error *reason)
1366 {
1367 BlockDriverState *bs = blk_bs(blk);
1368
1369 if (bs) {
1370 bdrv_op_unblock_all(bs, reason);
1371 }
1372 }
1373
1374 AioContext *blk_get_aio_context(BlockBackend *blk)
1375 {
1376 BlockDriverState *bs = blk_bs(blk);
1377
1378 if (bs) {
1379 return bdrv_get_aio_context(bs);
1380 } else {
1381 return qemu_get_aio_context();
1382 }
1383 }
1384
1385 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb)
1386 {
1387 BlockBackendAIOCB *blk_acb = DO_UPCAST(BlockBackendAIOCB, common, acb);
1388 return blk_get_aio_context(blk_acb->blk);
1389 }
1390
1391 void blk_set_aio_context(BlockBackend *blk, AioContext *new_context)
1392 {
1393 BlockDriverState *bs = blk_bs(blk);
1394
1395 if (bs) {
1396 if (blk->public.throttle_state) {
1397 throttle_timers_detach_aio_context(&blk->public.throttle_timers);
1398 }
1399 bdrv_set_aio_context(bs, new_context);
1400 if (blk->public.throttle_state) {
1401 throttle_timers_attach_aio_context(&blk->public.throttle_timers,
1402 new_context);
1403 }
1404 }
1405 }
1406
1407 void blk_add_aio_context_notifier(BlockBackend *blk,
1408 void (*attached_aio_context)(AioContext *new_context, void *opaque),
1409 void (*detach_aio_context)(void *opaque), void *opaque)
1410 {
1411 BlockDriverState *bs = blk_bs(blk);
1412
1413 if (bs) {
1414 bdrv_add_aio_context_notifier(bs, attached_aio_context,
1415 detach_aio_context, opaque);
1416 }
1417 }
1418
1419 void blk_remove_aio_context_notifier(BlockBackend *blk,
1420 void (*attached_aio_context)(AioContext *,
1421 void *),
1422 void (*detach_aio_context)(void *),
1423 void *opaque)
1424 {
1425 BlockDriverState *bs = blk_bs(blk);
1426
1427 if (bs) {
1428 bdrv_remove_aio_context_notifier(bs, attached_aio_context,
1429 detach_aio_context, opaque);
1430 }
1431 }
1432
1433 void blk_add_remove_bs_notifier(BlockBackend *blk, Notifier *notify)
1434 {
1435 notifier_list_add(&blk->remove_bs_notifiers, notify);
1436 }
1437
1438 void blk_add_insert_bs_notifier(BlockBackend *blk, Notifier *notify)
1439 {
1440 notifier_list_add(&blk->insert_bs_notifiers, notify);
1441 }
1442
1443 void blk_io_plug(BlockBackend *blk)
1444 {
1445 BlockDriverState *bs = blk_bs(blk);
1446
1447 if (bs) {
1448 bdrv_io_plug(bs);
1449 }
1450 }
1451
1452 void blk_io_unplug(BlockBackend *blk)
1453 {
1454 BlockDriverState *bs = blk_bs(blk);
1455
1456 if (bs) {
1457 bdrv_io_unplug(bs);
1458 }
1459 }
1460
1461 BlockAcctStats *blk_get_stats(BlockBackend *blk)
1462 {
1463 return &blk->stats;
1464 }
1465
1466 void *blk_aio_get(const AIOCBInfo *aiocb_info, BlockBackend *blk,
1467 BlockCompletionFunc *cb, void *opaque)
1468 {
1469 return qemu_aio_get(aiocb_info, blk_bs(blk), cb, opaque);
1470 }
1471
1472 int coroutine_fn blk_co_write_zeroes(BlockBackend *blk, int64_t offset,
1473 int count, BdrvRequestFlags flags)
1474 {
1475 return blk_co_pwritev(blk, offset, count, NULL,
1476 flags | BDRV_REQ_ZERO_WRITE);
1477 }
1478
1479 int blk_write_compressed(BlockBackend *blk, int64_t sector_num,
1480 const uint8_t *buf, int nb_sectors)
1481 {
1482 int ret = blk_check_request(blk, sector_num, nb_sectors);
1483 if (ret < 0) {
1484 return ret;
1485 }
1486
1487 return bdrv_write_compressed(blk_bs(blk), sector_num, buf, nb_sectors);
1488 }
1489
1490 int blk_truncate(BlockBackend *blk, int64_t offset)
1491 {
1492 if (!blk_is_available(blk)) {
1493 return -ENOMEDIUM;
1494 }
1495
1496 return bdrv_truncate(blk_bs(blk), offset);
1497 }
1498
1499 int blk_discard(BlockBackend *blk, int64_t sector_num, int nb_sectors)
1500 {
1501 int ret = blk_check_request(blk, sector_num, nb_sectors);
1502 if (ret < 0) {
1503 return ret;
1504 }
1505
1506 return bdrv_discard(blk_bs(blk), sector_num, nb_sectors);
1507 }
1508
1509 int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
1510 int64_t pos, int size)
1511 {
1512 int ret;
1513
1514 if (!blk_is_available(blk)) {
1515 return -ENOMEDIUM;
1516 }
1517
1518 ret = bdrv_save_vmstate(blk_bs(blk), buf, pos, size);
1519 if (ret < 0) {
1520 return ret;
1521 }
1522
1523 if (ret == size && !blk->enable_write_cache) {
1524 ret = bdrv_flush(blk_bs(blk));
1525 }
1526
1527 return ret < 0 ? ret : size;
1528 }
1529
1530 int blk_load_vmstate(BlockBackend *blk, uint8_t *buf, int64_t pos, int size)
1531 {
1532 if (!blk_is_available(blk)) {
1533 return -ENOMEDIUM;
1534 }
1535
1536 return bdrv_load_vmstate(blk_bs(blk), buf, pos, size);
1537 }
1538
1539 int blk_probe_blocksizes(BlockBackend *blk, BlockSizes *bsz)
1540 {
1541 if (!blk_is_available(blk)) {
1542 return -ENOMEDIUM;
1543 }
1544
1545 return bdrv_probe_blocksizes(blk_bs(blk), bsz);
1546 }
1547
1548 int blk_probe_geometry(BlockBackend *blk, HDGeometry *geo)
1549 {
1550 if (!blk_is_available(blk)) {
1551 return -ENOMEDIUM;
1552 }
1553
1554 return bdrv_probe_geometry(blk_bs(blk), geo);
1555 }
1556
1557 /*
1558 * Updates the BlockBackendRootState object with data from the currently
1559 * attached BlockDriverState.
1560 */
1561 void blk_update_root_state(BlockBackend *blk)
1562 {
1563 assert(blk->root);
1564
1565 blk->root_state.open_flags = blk->root->bs->open_flags;
1566 blk->root_state.read_only = blk->root->bs->read_only;
1567 blk->root_state.detect_zeroes = blk->root->bs->detect_zeroes;
1568 }
1569
1570 /*
1571 * Applies the information in the root state to the given BlockDriverState. This
1572 * does not include the flags which have to be specified for bdrv_open(), use
1573 * blk_get_open_flags_from_root_state() to inquire them.
1574 */
1575 void blk_apply_root_state(BlockBackend *blk, BlockDriverState *bs)
1576 {
1577 bs->detect_zeroes = blk->root_state.detect_zeroes;
1578 }
1579
1580 /*
1581 * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
1582 * supposed to inherit the root state.
1583 */
1584 int blk_get_open_flags_from_root_state(BlockBackend *blk)
1585 {
1586 int bs_flags;
1587
1588 bs_flags = blk->root_state.read_only ? 0 : BDRV_O_RDWR;
1589 bs_flags |= blk->root_state.open_flags & ~BDRV_O_RDWR;
1590
1591 return bs_flags;
1592 }
1593
1594 BlockBackendRootState *blk_get_root_state(BlockBackend *blk)
1595 {
1596 return &blk->root_state;
1597 }
1598
1599 int blk_commit_all(void)
1600 {
1601 BlockBackend *blk = NULL;
1602
1603 while ((blk = blk_all_next(blk)) != NULL) {
1604 AioContext *aio_context = blk_get_aio_context(blk);
1605
1606 aio_context_acquire(aio_context);
1607 if (blk_is_inserted(blk) && blk->root->bs->backing) {
1608 int ret = bdrv_commit(blk->root->bs);
1609 if (ret < 0) {
1610 aio_context_release(aio_context);
1611 return ret;
1612 }
1613 }
1614 aio_context_release(aio_context);
1615 }
1616 return 0;
1617 }
1618
1619 int blk_flush_all(void)
1620 {
1621 BlockBackend *blk = NULL;
1622 int result = 0;
1623
1624 while ((blk = blk_all_next(blk)) != NULL) {
1625 AioContext *aio_context = blk_get_aio_context(blk);
1626 int ret;
1627
1628 aio_context_acquire(aio_context);
1629 if (blk_is_inserted(blk)) {
1630 ret = blk_flush(blk);
1631 if (ret < 0 && !result) {
1632 result = ret;
1633 }
1634 }
1635 aio_context_release(aio_context);
1636 }
1637
1638 return result;
1639 }
1640
1641
1642 /* throttling disk I/O limits */
1643 void blk_set_io_limits(BlockBackend *blk, ThrottleConfig *cfg)
1644 {
1645 throttle_group_config(blk, cfg);
1646 }
1647
1648 void blk_io_limits_disable(BlockBackend *blk)
1649 {
1650 assert(blk->public.throttle_state);
1651 bdrv_drained_begin(blk_bs(blk));
1652 throttle_group_unregister_blk(blk);
1653 bdrv_drained_end(blk_bs(blk));
1654 }
1655
1656 /* should be called before blk_set_io_limits if a limit is set */
1657 void blk_io_limits_enable(BlockBackend *blk, const char *group)
1658 {
1659 assert(!blk->public.throttle_state);
1660 throttle_group_register_blk(blk, group);
1661 }
1662
1663 void blk_io_limits_update_group(BlockBackend *blk, const char *group)
1664 {
1665 /* this BB is not part of any group */
1666 if (!blk->public.throttle_state) {
1667 return;
1668 }
1669
1670 /* this BB is a part of the same group than the one we want */
1671 if (!g_strcmp0(throttle_group_get_name(blk), group)) {
1672 return;
1673 }
1674
1675 /* need to change the group this bs belong to */
1676 blk_io_limits_disable(blk);
1677 blk_io_limits_enable(blk, group);
1678 }
1679
1680 static void blk_root_drained_begin(BdrvChild *child)
1681 {
1682 BlockBackend *blk = child->opaque;
1683
1684 if (blk->public.io_limits_disabled++ == 0) {
1685 throttle_group_restart_blk(blk);
1686 }
1687 }
1688
1689 static void blk_root_drained_end(BdrvChild *child)
1690 {
1691 BlockBackend *blk = child->opaque;
1692
1693 assert(blk->public.io_limits_disabled);
1694 --blk->public.io_limits_disabled;
1695 }