tests: virtio-9p: rename PCI configuration test
[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 #include "trace.h"
23
24 /* Number of coroutines to reserve per attached device model */
25 #define COROUTINE_POOL_RESERVATION 64
26
27 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
28
29 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb);
30
31 struct BlockBackend {
32 char *name;
33 int refcnt;
34 BdrvChild *root;
35 DriveInfo *legacy_dinfo; /* null unless created by drive_new() */
36 QTAILQ_ENTRY(BlockBackend) link; /* for block_backends */
37 QTAILQ_ENTRY(BlockBackend) monitor_link; /* for monitor_block_backends */
38 BlockBackendPublic public;
39
40 void *dev; /* attached device model, if any */
41 bool legacy_dev; /* true if dev is not a DeviceState */
42 /* TODO change to DeviceState when all users are qdevified */
43 const BlockDevOps *dev_ops;
44 void *dev_opaque;
45
46 /* the block size for which the guest device expects atomicity */
47 int guest_block_size;
48
49 /* If the BDS tree is removed, some of its options are stored here (which
50 * can be used to restore those options in the new BDS on insert) */
51 BlockBackendRootState root_state;
52
53 bool enable_write_cache;
54
55 /* I/O stats (display with "info blockstats"). */
56 BlockAcctStats stats;
57
58 BlockdevOnError on_read_error, on_write_error;
59 bool iostatus_enabled;
60 BlockDeviceIoStatus iostatus;
61
62 bool allow_write_beyond_eof;
63
64 NotifierList remove_bs_notifiers, insert_bs_notifiers;
65 };
66
67 typedef struct BlockBackendAIOCB {
68 BlockAIOCB common;
69 BlockBackend *blk;
70 int ret;
71 } BlockBackendAIOCB;
72
73 static const AIOCBInfo block_backend_aiocb_info = {
74 .get_aio_context = blk_aiocb_get_aio_context,
75 .aiocb_size = sizeof(BlockBackendAIOCB),
76 };
77
78 static void drive_info_del(DriveInfo *dinfo);
79 static BlockBackend *bdrv_first_blk(BlockDriverState *bs);
80
81 /* All BlockBackends */
82 static QTAILQ_HEAD(, BlockBackend) block_backends =
83 QTAILQ_HEAD_INITIALIZER(block_backends);
84
85 /* All BlockBackends referenced by the monitor and which are iterated through by
86 * blk_next() */
87 static QTAILQ_HEAD(, BlockBackend) monitor_block_backends =
88 QTAILQ_HEAD_INITIALIZER(monitor_block_backends);
89
90 static void blk_root_inherit_options(int *child_flags, QDict *child_options,
91 int parent_flags, QDict *parent_options)
92 {
93 /* We're not supposed to call this function for root nodes */
94 abort();
95 }
96 static void blk_root_drained_begin(BdrvChild *child);
97 static void blk_root_drained_end(BdrvChild *child);
98
99 static void blk_root_change_media(BdrvChild *child, bool load);
100 static void blk_root_resize(BdrvChild *child);
101
102 static const char *blk_root_get_name(BdrvChild *child)
103 {
104 return blk_name(child->opaque);
105 }
106
107 static const BdrvChildRole child_root = {
108 .inherit_options = blk_root_inherit_options,
109
110 .change_media = blk_root_change_media,
111 .resize = blk_root_resize,
112 .get_name = blk_root_get_name,
113
114 .drained_begin = blk_root_drained_begin,
115 .drained_end = blk_root_drained_end,
116 };
117
118 /*
119 * Create a new BlockBackend with a reference count of one.
120 * Store an error through @errp on failure, unless it's null.
121 * Return the new BlockBackend on success, null on failure.
122 */
123 BlockBackend *blk_new(void)
124 {
125 BlockBackend *blk;
126
127 blk = g_new0(BlockBackend, 1);
128 blk->refcnt = 1;
129 blk_set_enable_write_cache(blk, true);
130
131 qemu_co_queue_init(&blk->public.throttled_reqs[0]);
132 qemu_co_queue_init(&blk->public.throttled_reqs[1]);
133
134 notifier_list_init(&blk->remove_bs_notifiers);
135 notifier_list_init(&blk->insert_bs_notifiers);
136
137 QTAILQ_INSERT_TAIL(&block_backends, blk, link);
138 return blk;
139 }
140
141 /*
142 * Creates a new BlockBackend, opens a new BlockDriverState, and connects both.
143 *
144 * Just as with bdrv_open(), after having called this function the reference to
145 * @options belongs to the block layer (even on failure).
146 *
147 * TODO: Remove @filename and @flags; it should be possible to specify a whole
148 * BDS tree just by specifying the @options QDict (or @reference,
149 * alternatively). At the time of adding this function, this is not possible,
150 * though, so callers of this function have to be able to specify @filename and
151 * @flags.
152 */
153 BlockBackend *blk_new_open(const char *filename, const char *reference,
154 QDict *options, int flags, Error **errp)
155 {
156 BlockBackend *blk;
157 BlockDriverState *bs;
158
159 blk = blk_new();
160 bs = bdrv_open(filename, reference, options, flags, errp);
161 if (!bs) {
162 blk_unref(blk);
163 return NULL;
164 }
165
166 blk->root = bdrv_root_attach_child(bs, "root", &child_root, blk);
167
168 return blk;
169 }
170
171 static void blk_delete(BlockBackend *blk)
172 {
173 assert(!blk->refcnt);
174 assert(!blk->name);
175 assert(!blk->dev);
176 if (blk->root) {
177 blk_remove_bs(blk);
178 }
179 assert(QLIST_EMPTY(&blk->remove_bs_notifiers.notifiers));
180 assert(QLIST_EMPTY(&blk->insert_bs_notifiers.notifiers));
181 QTAILQ_REMOVE(&block_backends, blk, link);
182 drive_info_del(blk->legacy_dinfo);
183 block_acct_cleanup(&blk->stats);
184 g_free(blk);
185 }
186
187 static void drive_info_del(DriveInfo *dinfo)
188 {
189 if (!dinfo) {
190 return;
191 }
192 qemu_opts_del(dinfo->opts);
193 g_free(dinfo->serial);
194 g_free(dinfo);
195 }
196
197 int blk_get_refcnt(BlockBackend *blk)
198 {
199 return blk ? blk->refcnt : 0;
200 }
201
202 /*
203 * Increment @blk's reference count.
204 * @blk must not be null.
205 */
206 void blk_ref(BlockBackend *blk)
207 {
208 blk->refcnt++;
209 }
210
211 /*
212 * Decrement @blk's reference count.
213 * If this drops it to zero, destroy @blk.
214 * For convenience, do nothing if @blk is null.
215 */
216 void blk_unref(BlockBackend *blk)
217 {
218 if (blk) {
219 assert(blk->refcnt > 0);
220 if (!--blk->refcnt) {
221 blk_delete(blk);
222 }
223 }
224 }
225
226 /*
227 * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
228 * ones which are hidden (i.e. are not referenced by the monitor).
229 */
230 static BlockBackend *blk_all_next(BlockBackend *blk)
231 {
232 return blk ? QTAILQ_NEXT(blk, link)
233 : QTAILQ_FIRST(&block_backends);
234 }
235
236 void blk_remove_all_bs(void)
237 {
238 BlockBackend *blk = NULL;
239
240 while ((blk = blk_all_next(blk)) != NULL) {
241 AioContext *ctx = blk_get_aio_context(blk);
242
243 aio_context_acquire(ctx);
244 if (blk->root) {
245 blk_remove_bs(blk);
246 }
247 aio_context_release(ctx);
248 }
249 }
250
251 /*
252 * Return the monitor-owned BlockBackend after @blk.
253 * If @blk is null, return the first one.
254 * Else, return @blk's next sibling, which may be null.
255 *
256 * To iterate over all BlockBackends, do
257 * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
258 * ...
259 * }
260 */
261 BlockBackend *blk_next(BlockBackend *blk)
262 {
263 return blk ? QTAILQ_NEXT(blk, monitor_link)
264 : QTAILQ_FIRST(&monitor_block_backends);
265 }
266
267 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
268 * the monitor or attached to a BlockBackend */
269 BlockDriverState *bdrv_next(BdrvNextIterator *it)
270 {
271 BlockDriverState *bs;
272
273 /* First, return all root nodes of BlockBackends. In order to avoid
274 * returning a BDS twice when multiple BBs refer to it, we only return it
275 * if the BB is the first one in the parent list of the BDS. */
276 if (it->phase == BDRV_NEXT_BACKEND_ROOTS) {
277 do {
278 it->blk = blk_all_next(it->blk);
279 bs = it->blk ? blk_bs(it->blk) : NULL;
280 } while (it->blk && (bs == NULL || bdrv_first_blk(bs) != it->blk));
281
282 if (bs) {
283 return bs;
284 }
285 it->phase = BDRV_NEXT_MONITOR_OWNED;
286 }
287
288 /* Then return the monitor-owned BDSes without a BB attached. Ignore all
289 * BDSes that are attached to a BlockBackend here; they have been handled
290 * by the above block already */
291 do {
292 it->bs = bdrv_next_monitor_owned(it->bs);
293 bs = it->bs;
294 } while (bs && bdrv_has_blk(bs));
295
296 return bs;
297 }
298
299 BlockDriverState *bdrv_first(BdrvNextIterator *it)
300 {
301 *it = (BdrvNextIterator) {
302 .phase = BDRV_NEXT_BACKEND_ROOTS,
303 };
304
305 return bdrv_next(it);
306 }
307
308 /*
309 * Add a BlockBackend into the list of backends referenced by the monitor, with
310 * the given @name acting as the handle for the monitor.
311 * Strictly for use by blockdev.c.
312 *
313 * @name must not be null or empty.
314 *
315 * Returns true on success and false on failure. In the latter case, an Error
316 * object is returned through @errp.
317 */
318 bool monitor_add_blk(BlockBackend *blk, const char *name, Error **errp)
319 {
320 assert(!blk->name);
321 assert(name && name[0]);
322
323 if (!id_wellformed(name)) {
324 error_setg(errp, "Invalid device name");
325 return false;
326 }
327 if (blk_by_name(name)) {
328 error_setg(errp, "Device with id '%s' already exists", name);
329 return false;
330 }
331 if (bdrv_find_node(name)) {
332 error_setg(errp,
333 "Device name '%s' conflicts with an existing node name",
334 name);
335 return false;
336 }
337
338 blk->name = g_strdup(name);
339 QTAILQ_INSERT_TAIL(&monitor_block_backends, blk, monitor_link);
340 return true;
341 }
342
343 /*
344 * Remove a BlockBackend from the list of backends referenced by the monitor.
345 * Strictly for use by blockdev.c.
346 */
347 void monitor_remove_blk(BlockBackend *blk)
348 {
349 if (!blk->name) {
350 return;
351 }
352
353 QTAILQ_REMOVE(&monitor_block_backends, blk, monitor_link);
354 g_free(blk->name);
355 blk->name = NULL;
356 }
357
358 /*
359 * Return @blk's name, a non-null string.
360 * Returns an empty string iff @blk is not referenced by the monitor.
361 */
362 const char *blk_name(BlockBackend *blk)
363 {
364 return blk->name ?: "";
365 }
366
367 /*
368 * Return the BlockBackend with name @name if it exists, else null.
369 * @name must not be null.
370 */
371 BlockBackend *blk_by_name(const char *name)
372 {
373 BlockBackend *blk = NULL;
374
375 assert(name);
376 while ((blk = blk_next(blk)) != NULL) {
377 if (!strcmp(name, blk->name)) {
378 return blk;
379 }
380 }
381 return NULL;
382 }
383
384 /*
385 * Return the BlockDriverState attached to @blk if any, else null.
386 */
387 BlockDriverState *blk_bs(BlockBackend *blk)
388 {
389 return blk->root ? blk->root->bs : NULL;
390 }
391
392 static BlockBackend *bdrv_first_blk(BlockDriverState *bs)
393 {
394 BdrvChild *child;
395 QLIST_FOREACH(child, &bs->parents, next_parent) {
396 if (child->role == &child_root) {
397 return child->opaque;
398 }
399 }
400
401 return NULL;
402 }
403
404 /*
405 * Returns true if @bs has an associated BlockBackend.
406 */
407 bool bdrv_has_blk(BlockDriverState *bs)
408 {
409 return bdrv_first_blk(bs) != NULL;
410 }
411
412 /*
413 * Returns true if @bs has only BlockBackends as parents.
414 */
415 bool bdrv_is_root_node(BlockDriverState *bs)
416 {
417 BdrvChild *c;
418
419 QLIST_FOREACH(c, &bs->parents, next_parent) {
420 if (c->role != &child_root) {
421 return false;
422 }
423 }
424
425 return true;
426 }
427
428 /*
429 * Return @blk's DriveInfo if any, else null.
430 */
431 DriveInfo *blk_legacy_dinfo(BlockBackend *blk)
432 {
433 return blk->legacy_dinfo;
434 }
435
436 /*
437 * Set @blk's DriveInfo to @dinfo, and return it.
438 * @blk must not have a DriveInfo set already.
439 * No other BlockBackend may have the same DriveInfo set.
440 */
441 DriveInfo *blk_set_legacy_dinfo(BlockBackend *blk, DriveInfo *dinfo)
442 {
443 assert(!blk->legacy_dinfo);
444 return blk->legacy_dinfo = dinfo;
445 }
446
447 /*
448 * Return the BlockBackend with DriveInfo @dinfo.
449 * It must exist.
450 */
451 BlockBackend *blk_by_legacy_dinfo(DriveInfo *dinfo)
452 {
453 BlockBackend *blk = NULL;
454
455 while ((blk = blk_next(blk)) != NULL) {
456 if (blk->legacy_dinfo == dinfo) {
457 return blk;
458 }
459 }
460 abort();
461 }
462
463 /*
464 * Returns a pointer to the publicly accessible fields of @blk.
465 */
466 BlockBackendPublic *blk_get_public(BlockBackend *blk)
467 {
468 return &blk->public;
469 }
470
471 /*
472 * Returns a BlockBackend given the associated @public fields.
473 */
474 BlockBackend *blk_by_public(BlockBackendPublic *public)
475 {
476 return container_of(public, BlockBackend, public);
477 }
478
479 /*
480 * Disassociates the currently associated BlockDriverState from @blk.
481 */
482 void blk_remove_bs(BlockBackend *blk)
483 {
484 notifier_list_notify(&blk->remove_bs_notifiers, blk);
485 if (blk->public.throttle_state) {
486 throttle_timers_detach_aio_context(&blk->public.throttle_timers);
487 }
488
489 blk_update_root_state(blk);
490
491 bdrv_root_unref_child(blk->root);
492 blk->root = NULL;
493 }
494
495 /*
496 * Associates a new BlockDriverState with @blk.
497 */
498 void blk_insert_bs(BlockBackend *blk, BlockDriverState *bs)
499 {
500 bdrv_ref(bs);
501 blk->root = bdrv_root_attach_child(bs, "root", &child_root, blk);
502
503 notifier_list_notify(&blk->insert_bs_notifiers, blk);
504 if (blk->public.throttle_state) {
505 throttle_timers_attach_aio_context(
506 &blk->public.throttle_timers, bdrv_get_aio_context(bs));
507 }
508 }
509
510 static int blk_do_attach_dev(BlockBackend *blk, void *dev)
511 {
512 if (blk->dev) {
513 return -EBUSY;
514 }
515 blk_ref(blk);
516 blk->dev = dev;
517 blk->legacy_dev = false;
518 blk_iostatus_reset(blk);
519 return 0;
520 }
521
522 /*
523 * Attach device model @dev to @blk.
524 * Return 0 on success, -EBUSY when a device model is attached already.
525 */
526 int blk_attach_dev(BlockBackend *blk, DeviceState *dev)
527 {
528 return blk_do_attach_dev(blk, dev);
529 }
530
531 /*
532 * Attach device model @dev to @blk.
533 * @blk must not have a device model attached already.
534 * TODO qdevified devices don't use this, remove when devices are qdevified
535 */
536 void blk_attach_dev_legacy(BlockBackend *blk, void *dev)
537 {
538 if (blk_do_attach_dev(blk, dev) < 0) {
539 abort();
540 }
541 blk->legacy_dev = true;
542 }
543
544 /*
545 * Detach device model @dev from @blk.
546 * @dev must be currently attached to @blk.
547 */
548 void blk_detach_dev(BlockBackend *blk, void *dev)
549 /* TODO change to DeviceState *dev when all users are qdevified */
550 {
551 assert(blk->dev == dev);
552 blk->dev = NULL;
553 blk->dev_ops = NULL;
554 blk->dev_opaque = NULL;
555 blk->guest_block_size = 512;
556 blk_unref(blk);
557 }
558
559 /*
560 * Return the device model attached to @blk if any, else null.
561 */
562 void *blk_get_attached_dev(BlockBackend *blk)
563 /* TODO change to return DeviceState * when all users are qdevified */
564 {
565 return blk->dev;
566 }
567
568 /* Return the qdev ID, or if no ID is assigned the QOM path, of the block
569 * device attached to the BlockBackend. */
570 static char *blk_get_attached_dev_id(BlockBackend *blk)
571 {
572 DeviceState *dev;
573
574 assert(!blk->legacy_dev);
575 dev = blk->dev;
576
577 if (!dev) {
578 return g_strdup("");
579 } else if (dev->id) {
580 return g_strdup(dev->id);
581 }
582 return object_get_canonical_path(OBJECT(dev));
583 }
584
585 /*
586 * Return the BlockBackend which has the device model @dev attached if it
587 * exists, else null.
588 *
589 * @dev must not be null.
590 */
591 BlockBackend *blk_by_dev(void *dev)
592 {
593 BlockBackend *blk = NULL;
594
595 assert(dev != NULL);
596 while ((blk = blk_all_next(blk)) != NULL) {
597 if (blk->dev == dev) {
598 return blk;
599 }
600 }
601 return NULL;
602 }
603
604 /*
605 * Set @blk's device model callbacks to @ops.
606 * @opaque is the opaque argument to pass to the callbacks.
607 * This is for use by device models.
608 */
609 void blk_set_dev_ops(BlockBackend *blk, const BlockDevOps *ops,
610 void *opaque)
611 {
612 /* All drivers that use blk_set_dev_ops() are qdevified and we want to keep
613 * it that way, so we can assume blk->dev is a DeviceState if blk->dev_ops
614 * is set. */
615 assert(!blk->legacy_dev);
616
617 blk->dev_ops = ops;
618 blk->dev_opaque = opaque;
619 }
620
621 /*
622 * Notify @blk's attached device model of media change.
623 * If @load is true, notify of media load.
624 * Else, notify of media eject.
625 * Also send DEVICE_TRAY_MOVED events as appropriate.
626 */
627 void blk_dev_change_media_cb(BlockBackend *blk, bool load)
628 {
629 if (blk->dev_ops && blk->dev_ops->change_media_cb) {
630 bool tray_was_open, tray_is_open;
631
632 assert(!blk->legacy_dev);
633
634 tray_was_open = blk_dev_is_tray_open(blk);
635 blk->dev_ops->change_media_cb(blk->dev_opaque, load);
636 tray_is_open = blk_dev_is_tray_open(blk);
637
638 if (tray_was_open != tray_is_open) {
639 char *id = blk_get_attached_dev_id(blk);
640 qapi_event_send_device_tray_moved(blk_name(blk), id, tray_is_open,
641 &error_abort);
642 g_free(id);
643 }
644 }
645 }
646
647 static void blk_root_change_media(BdrvChild *child, bool load)
648 {
649 blk_dev_change_media_cb(child->opaque, load);
650 }
651
652 /*
653 * Does @blk's attached device model have removable media?
654 * %true if no device model is attached.
655 */
656 bool blk_dev_has_removable_media(BlockBackend *blk)
657 {
658 return !blk->dev || (blk->dev_ops && blk->dev_ops->change_media_cb);
659 }
660
661 /*
662 * Does @blk's attached device model have a tray?
663 */
664 bool blk_dev_has_tray(BlockBackend *blk)
665 {
666 return blk->dev_ops && blk->dev_ops->is_tray_open;
667 }
668
669 /*
670 * Notify @blk's attached device model of a media eject request.
671 * If @force is true, the medium is about to be yanked out forcefully.
672 */
673 void blk_dev_eject_request(BlockBackend *blk, bool force)
674 {
675 if (blk->dev_ops && blk->dev_ops->eject_request_cb) {
676 blk->dev_ops->eject_request_cb(blk->dev_opaque, force);
677 }
678 }
679
680 /*
681 * Does @blk's attached device model have a tray, and is it open?
682 */
683 bool blk_dev_is_tray_open(BlockBackend *blk)
684 {
685 if (blk_dev_has_tray(blk)) {
686 return blk->dev_ops->is_tray_open(blk->dev_opaque);
687 }
688 return false;
689 }
690
691 /*
692 * Does @blk's attached device model have the medium locked?
693 * %false if the device model has no such lock.
694 */
695 bool blk_dev_is_medium_locked(BlockBackend *blk)
696 {
697 if (blk->dev_ops && blk->dev_ops->is_medium_locked) {
698 return blk->dev_ops->is_medium_locked(blk->dev_opaque);
699 }
700 return false;
701 }
702
703 /*
704 * Notify @blk's attached device model of a backend size change.
705 */
706 static void blk_root_resize(BdrvChild *child)
707 {
708 BlockBackend *blk = child->opaque;
709
710 if (blk->dev_ops && blk->dev_ops->resize_cb) {
711 blk->dev_ops->resize_cb(blk->dev_opaque);
712 }
713 }
714
715 void blk_iostatus_enable(BlockBackend *blk)
716 {
717 blk->iostatus_enabled = true;
718 blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
719 }
720
721 /* The I/O status is only enabled if the drive explicitly
722 * enables it _and_ the VM is configured to stop on errors */
723 bool blk_iostatus_is_enabled(const BlockBackend *blk)
724 {
725 return (blk->iostatus_enabled &&
726 (blk->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
727 blk->on_write_error == BLOCKDEV_ON_ERROR_STOP ||
728 blk->on_read_error == BLOCKDEV_ON_ERROR_STOP));
729 }
730
731 BlockDeviceIoStatus blk_iostatus(const BlockBackend *blk)
732 {
733 return blk->iostatus;
734 }
735
736 void blk_iostatus_disable(BlockBackend *blk)
737 {
738 blk->iostatus_enabled = false;
739 }
740
741 void blk_iostatus_reset(BlockBackend *blk)
742 {
743 if (blk_iostatus_is_enabled(blk)) {
744 BlockDriverState *bs = blk_bs(blk);
745 blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
746 if (bs && bs->job) {
747 block_job_iostatus_reset(bs->job);
748 }
749 }
750 }
751
752 void blk_iostatus_set_err(BlockBackend *blk, int error)
753 {
754 assert(blk_iostatus_is_enabled(blk));
755 if (blk->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
756 blk->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
757 BLOCK_DEVICE_IO_STATUS_FAILED;
758 }
759 }
760
761 void blk_set_allow_write_beyond_eof(BlockBackend *blk, bool allow)
762 {
763 blk->allow_write_beyond_eof = allow;
764 }
765
766 static int blk_check_byte_request(BlockBackend *blk, int64_t offset,
767 size_t size)
768 {
769 int64_t len;
770
771 if (size > INT_MAX) {
772 return -EIO;
773 }
774
775 if (!blk_is_available(blk)) {
776 return -ENOMEDIUM;
777 }
778
779 if (offset < 0) {
780 return -EIO;
781 }
782
783 if (!blk->allow_write_beyond_eof) {
784 len = blk_getlength(blk);
785 if (len < 0) {
786 return len;
787 }
788
789 if (offset > len || len - offset < size) {
790 return -EIO;
791 }
792 }
793
794 return 0;
795 }
796
797 int coroutine_fn blk_co_preadv(BlockBackend *blk, int64_t offset,
798 unsigned int bytes, QEMUIOVector *qiov,
799 BdrvRequestFlags flags)
800 {
801 int ret;
802 BlockDriverState *bs = blk_bs(blk);
803
804 trace_blk_co_preadv(blk, bs, offset, bytes, flags);
805
806 ret = blk_check_byte_request(blk, offset, bytes);
807 if (ret < 0) {
808 return ret;
809 }
810
811 bdrv_inc_in_flight(bs);
812
813 /* throttling disk I/O */
814 if (blk->public.throttle_state) {
815 throttle_group_co_io_limits_intercept(blk, bytes, false);
816 }
817
818 ret = bdrv_co_preadv(blk->root, offset, bytes, qiov, flags);
819 bdrv_dec_in_flight(bs);
820 return ret;
821 }
822
823 int coroutine_fn blk_co_pwritev(BlockBackend *blk, int64_t offset,
824 unsigned int bytes, QEMUIOVector *qiov,
825 BdrvRequestFlags flags)
826 {
827 int ret;
828 BlockDriverState *bs = blk_bs(blk);
829
830 trace_blk_co_pwritev(blk, bs, offset, bytes, flags);
831
832 ret = blk_check_byte_request(blk, offset, bytes);
833 if (ret < 0) {
834 return ret;
835 }
836
837 bdrv_inc_in_flight(bs);
838
839 /* throttling disk I/O */
840 if (blk->public.throttle_state) {
841 throttle_group_co_io_limits_intercept(blk, bytes, true);
842 }
843
844 if (!blk->enable_write_cache) {
845 flags |= BDRV_REQ_FUA;
846 }
847
848 ret = bdrv_co_pwritev(blk->root, offset, bytes, qiov, flags);
849 bdrv_dec_in_flight(bs);
850 return ret;
851 }
852
853 typedef struct BlkRwCo {
854 BlockBackend *blk;
855 int64_t offset;
856 QEMUIOVector *qiov;
857 int ret;
858 BdrvRequestFlags flags;
859 } BlkRwCo;
860
861 static void blk_read_entry(void *opaque)
862 {
863 BlkRwCo *rwco = opaque;
864
865 rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, rwco->qiov->size,
866 rwco->qiov, rwco->flags);
867 }
868
869 static void blk_write_entry(void *opaque)
870 {
871 BlkRwCo *rwco = opaque;
872
873 rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, rwco->qiov->size,
874 rwco->qiov, rwco->flags);
875 }
876
877 static int blk_prw(BlockBackend *blk, int64_t offset, uint8_t *buf,
878 int64_t bytes, CoroutineEntry co_entry,
879 BdrvRequestFlags flags)
880 {
881 QEMUIOVector qiov;
882 struct iovec iov;
883 Coroutine *co;
884 BlkRwCo rwco;
885
886 iov = (struct iovec) {
887 .iov_base = buf,
888 .iov_len = bytes,
889 };
890 qemu_iovec_init_external(&qiov, &iov, 1);
891
892 rwco = (BlkRwCo) {
893 .blk = blk,
894 .offset = offset,
895 .qiov = &qiov,
896 .flags = flags,
897 .ret = NOT_DONE,
898 };
899
900 co = qemu_coroutine_create(co_entry, &rwco);
901 qemu_coroutine_enter(co);
902 BDRV_POLL_WHILE(blk_bs(blk), rwco.ret == NOT_DONE);
903
904 return rwco.ret;
905 }
906
907 int blk_pread_unthrottled(BlockBackend *blk, int64_t offset, uint8_t *buf,
908 int count)
909 {
910 int ret;
911
912 ret = blk_check_byte_request(blk, offset, count);
913 if (ret < 0) {
914 return ret;
915 }
916
917 blk_root_drained_begin(blk->root);
918 ret = blk_pread(blk, offset, buf, count);
919 blk_root_drained_end(blk->root);
920 return ret;
921 }
922
923 int blk_pwrite_zeroes(BlockBackend *blk, int64_t offset,
924 int count, BdrvRequestFlags flags)
925 {
926 return blk_prw(blk, offset, NULL, count, blk_write_entry,
927 flags | BDRV_REQ_ZERO_WRITE);
928 }
929
930 int blk_make_zero(BlockBackend *blk, BdrvRequestFlags flags)
931 {
932 return bdrv_make_zero(blk->root, flags);
933 }
934
935 static void error_callback_bh(void *opaque)
936 {
937 struct BlockBackendAIOCB *acb = opaque;
938
939 bdrv_dec_in_flight(acb->common.bs);
940 acb->common.cb(acb->common.opaque, acb->ret);
941 qemu_aio_unref(acb);
942 }
943
944 BlockAIOCB *blk_abort_aio_request(BlockBackend *blk,
945 BlockCompletionFunc *cb,
946 void *opaque, int ret)
947 {
948 struct BlockBackendAIOCB *acb;
949
950 bdrv_inc_in_flight(blk_bs(blk));
951 acb = blk_aio_get(&block_backend_aiocb_info, blk, cb, opaque);
952 acb->blk = blk;
953 acb->ret = ret;
954
955 aio_bh_schedule_oneshot(blk_get_aio_context(blk), error_callback_bh, acb);
956 return &acb->common;
957 }
958
959 typedef struct BlkAioEmAIOCB {
960 BlockAIOCB common;
961 BlkRwCo rwco;
962 int bytes;
963 bool has_returned;
964 } BlkAioEmAIOCB;
965
966 static const AIOCBInfo blk_aio_em_aiocb_info = {
967 .aiocb_size = sizeof(BlkAioEmAIOCB),
968 };
969
970 static void blk_aio_complete(BlkAioEmAIOCB *acb)
971 {
972 if (acb->has_returned) {
973 bdrv_dec_in_flight(acb->common.bs);
974 acb->common.cb(acb->common.opaque, acb->rwco.ret);
975 qemu_aio_unref(acb);
976 }
977 }
978
979 static void blk_aio_complete_bh(void *opaque)
980 {
981 BlkAioEmAIOCB *acb = opaque;
982
983 assert(acb->has_returned);
984 blk_aio_complete(acb);
985 }
986
987 static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset, int bytes,
988 QEMUIOVector *qiov, CoroutineEntry co_entry,
989 BdrvRequestFlags flags,
990 BlockCompletionFunc *cb, void *opaque)
991 {
992 BlkAioEmAIOCB *acb;
993 Coroutine *co;
994
995 bdrv_inc_in_flight(blk_bs(blk));
996 acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
997 acb->rwco = (BlkRwCo) {
998 .blk = blk,
999 .offset = offset,
1000 .qiov = qiov,
1001 .flags = flags,
1002 .ret = NOT_DONE,
1003 };
1004 acb->bytes = bytes;
1005 acb->has_returned = false;
1006
1007 co = qemu_coroutine_create(co_entry, acb);
1008 qemu_coroutine_enter(co);
1009
1010 acb->has_returned = true;
1011 if (acb->rwco.ret != NOT_DONE) {
1012 aio_bh_schedule_oneshot(blk_get_aio_context(blk),
1013 blk_aio_complete_bh, acb);
1014 }
1015
1016 return &acb->common;
1017 }
1018
1019 static void blk_aio_read_entry(void *opaque)
1020 {
1021 BlkAioEmAIOCB *acb = opaque;
1022 BlkRwCo *rwco = &acb->rwco;
1023
1024 assert(rwco->qiov->size == acb->bytes);
1025 rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, acb->bytes,
1026 rwco->qiov, rwco->flags);
1027 blk_aio_complete(acb);
1028 }
1029
1030 static void blk_aio_write_entry(void *opaque)
1031 {
1032 BlkAioEmAIOCB *acb = opaque;
1033 BlkRwCo *rwco = &acb->rwco;
1034
1035 assert(!rwco->qiov || rwco->qiov->size == acb->bytes);
1036 rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, acb->bytes,
1037 rwco->qiov, rwco->flags);
1038 blk_aio_complete(acb);
1039 }
1040
1041 BlockAIOCB *blk_aio_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1042 int count, BdrvRequestFlags flags,
1043 BlockCompletionFunc *cb, void *opaque)
1044 {
1045 return blk_aio_prwv(blk, offset, count, NULL, blk_aio_write_entry,
1046 flags | BDRV_REQ_ZERO_WRITE, cb, opaque);
1047 }
1048
1049 int blk_pread(BlockBackend *blk, int64_t offset, void *buf, int count)
1050 {
1051 int ret = blk_prw(blk, offset, buf, count, blk_read_entry, 0);
1052 if (ret < 0) {
1053 return ret;
1054 }
1055 return count;
1056 }
1057
1058 int blk_pwrite(BlockBackend *blk, int64_t offset, const void *buf, int count,
1059 BdrvRequestFlags flags)
1060 {
1061 int ret = blk_prw(blk, offset, (void *) buf, count, blk_write_entry,
1062 flags);
1063 if (ret < 0) {
1064 return ret;
1065 }
1066 return count;
1067 }
1068
1069 int64_t blk_getlength(BlockBackend *blk)
1070 {
1071 if (!blk_is_available(blk)) {
1072 return -ENOMEDIUM;
1073 }
1074
1075 return bdrv_getlength(blk_bs(blk));
1076 }
1077
1078 void blk_get_geometry(BlockBackend *blk, uint64_t *nb_sectors_ptr)
1079 {
1080 if (!blk_bs(blk)) {
1081 *nb_sectors_ptr = 0;
1082 } else {
1083 bdrv_get_geometry(blk_bs(blk), nb_sectors_ptr);
1084 }
1085 }
1086
1087 int64_t blk_nb_sectors(BlockBackend *blk)
1088 {
1089 if (!blk_is_available(blk)) {
1090 return -ENOMEDIUM;
1091 }
1092
1093 return bdrv_nb_sectors(blk_bs(blk));
1094 }
1095
1096 BlockAIOCB *blk_aio_preadv(BlockBackend *blk, int64_t offset,
1097 QEMUIOVector *qiov, BdrvRequestFlags flags,
1098 BlockCompletionFunc *cb, void *opaque)
1099 {
1100 return blk_aio_prwv(blk, offset, qiov->size, qiov,
1101 blk_aio_read_entry, flags, cb, opaque);
1102 }
1103
1104 BlockAIOCB *blk_aio_pwritev(BlockBackend *blk, int64_t offset,
1105 QEMUIOVector *qiov, BdrvRequestFlags flags,
1106 BlockCompletionFunc *cb, void *opaque)
1107 {
1108 return blk_aio_prwv(blk, offset, qiov->size, qiov,
1109 blk_aio_write_entry, flags, cb, opaque);
1110 }
1111
1112 static void blk_aio_flush_entry(void *opaque)
1113 {
1114 BlkAioEmAIOCB *acb = opaque;
1115 BlkRwCo *rwco = &acb->rwco;
1116
1117 rwco->ret = blk_co_flush(rwco->blk);
1118 blk_aio_complete(acb);
1119 }
1120
1121 BlockAIOCB *blk_aio_flush(BlockBackend *blk,
1122 BlockCompletionFunc *cb, void *opaque)
1123 {
1124 return blk_aio_prwv(blk, 0, 0, NULL, blk_aio_flush_entry, 0, cb, opaque);
1125 }
1126
1127 static void blk_aio_pdiscard_entry(void *opaque)
1128 {
1129 BlkAioEmAIOCB *acb = opaque;
1130 BlkRwCo *rwco = &acb->rwco;
1131
1132 rwco->ret = blk_co_pdiscard(rwco->blk, rwco->offset, acb->bytes);
1133 blk_aio_complete(acb);
1134 }
1135
1136 BlockAIOCB *blk_aio_pdiscard(BlockBackend *blk,
1137 int64_t offset, int count,
1138 BlockCompletionFunc *cb, void *opaque)
1139 {
1140 return blk_aio_prwv(blk, offset, count, NULL, blk_aio_pdiscard_entry, 0,
1141 cb, opaque);
1142 }
1143
1144 void blk_aio_cancel(BlockAIOCB *acb)
1145 {
1146 bdrv_aio_cancel(acb);
1147 }
1148
1149 void blk_aio_cancel_async(BlockAIOCB *acb)
1150 {
1151 bdrv_aio_cancel_async(acb);
1152 }
1153
1154 int blk_co_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1155 {
1156 if (!blk_is_available(blk)) {
1157 return -ENOMEDIUM;
1158 }
1159
1160 return bdrv_co_ioctl(blk_bs(blk), req, buf);
1161 }
1162
1163 static void blk_ioctl_entry(void *opaque)
1164 {
1165 BlkRwCo *rwco = opaque;
1166 rwco->ret = blk_co_ioctl(rwco->blk, rwco->offset,
1167 rwco->qiov->iov[0].iov_base);
1168 }
1169
1170 int blk_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1171 {
1172 return blk_prw(blk, req, buf, 0, blk_ioctl_entry, 0);
1173 }
1174
1175 static void blk_aio_ioctl_entry(void *opaque)
1176 {
1177 BlkAioEmAIOCB *acb = opaque;
1178 BlkRwCo *rwco = &acb->rwco;
1179
1180 rwco->ret = blk_co_ioctl(rwco->blk, rwco->offset,
1181 rwco->qiov->iov[0].iov_base);
1182 blk_aio_complete(acb);
1183 }
1184
1185 BlockAIOCB *blk_aio_ioctl(BlockBackend *blk, unsigned long int req, void *buf,
1186 BlockCompletionFunc *cb, void *opaque)
1187 {
1188 QEMUIOVector qiov;
1189 struct iovec iov;
1190
1191 iov = (struct iovec) {
1192 .iov_base = buf,
1193 .iov_len = 0,
1194 };
1195 qemu_iovec_init_external(&qiov, &iov, 1);
1196
1197 return blk_aio_prwv(blk, req, 0, &qiov, blk_aio_ioctl_entry, 0, cb, opaque);
1198 }
1199
1200 int blk_co_pdiscard(BlockBackend *blk, int64_t offset, int count)
1201 {
1202 int ret = blk_check_byte_request(blk, offset, count);
1203 if (ret < 0) {
1204 return ret;
1205 }
1206
1207 return bdrv_co_pdiscard(blk_bs(blk), offset, count);
1208 }
1209
1210 int blk_co_flush(BlockBackend *blk)
1211 {
1212 if (!blk_is_available(blk)) {
1213 return -ENOMEDIUM;
1214 }
1215
1216 return bdrv_co_flush(blk_bs(blk));
1217 }
1218
1219 static void blk_flush_entry(void *opaque)
1220 {
1221 BlkRwCo *rwco = opaque;
1222 rwco->ret = blk_co_flush(rwco->blk);
1223 }
1224
1225 int blk_flush(BlockBackend *blk)
1226 {
1227 return blk_prw(blk, 0, NULL, 0, blk_flush_entry, 0);
1228 }
1229
1230 void blk_drain(BlockBackend *blk)
1231 {
1232 if (blk_bs(blk)) {
1233 bdrv_drain(blk_bs(blk));
1234 }
1235 }
1236
1237 void blk_drain_all(void)
1238 {
1239 bdrv_drain_all();
1240 }
1241
1242 void blk_set_on_error(BlockBackend *blk, BlockdevOnError on_read_error,
1243 BlockdevOnError on_write_error)
1244 {
1245 blk->on_read_error = on_read_error;
1246 blk->on_write_error = on_write_error;
1247 }
1248
1249 BlockdevOnError blk_get_on_error(BlockBackend *blk, bool is_read)
1250 {
1251 return is_read ? blk->on_read_error : blk->on_write_error;
1252 }
1253
1254 BlockErrorAction blk_get_error_action(BlockBackend *blk, bool is_read,
1255 int error)
1256 {
1257 BlockdevOnError on_err = blk_get_on_error(blk, is_read);
1258
1259 switch (on_err) {
1260 case BLOCKDEV_ON_ERROR_ENOSPC:
1261 return (error == ENOSPC) ?
1262 BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
1263 case BLOCKDEV_ON_ERROR_STOP:
1264 return BLOCK_ERROR_ACTION_STOP;
1265 case BLOCKDEV_ON_ERROR_REPORT:
1266 return BLOCK_ERROR_ACTION_REPORT;
1267 case BLOCKDEV_ON_ERROR_IGNORE:
1268 return BLOCK_ERROR_ACTION_IGNORE;
1269 case BLOCKDEV_ON_ERROR_AUTO:
1270 default:
1271 abort();
1272 }
1273 }
1274
1275 static void send_qmp_error_event(BlockBackend *blk,
1276 BlockErrorAction action,
1277 bool is_read, int error)
1278 {
1279 IoOperationType optype;
1280
1281 optype = is_read ? IO_OPERATION_TYPE_READ : IO_OPERATION_TYPE_WRITE;
1282 qapi_event_send_block_io_error(blk_name(blk),
1283 bdrv_get_node_name(blk_bs(blk)), optype,
1284 action, blk_iostatus_is_enabled(blk),
1285 error == ENOSPC, strerror(error),
1286 &error_abort);
1287 }
1288
1289 /* This is done by device models because, while the block layer knows
1290 * about the error, it does not know whether an operation comes from
1291 * the device or the block layer (from a job, for example).
1292 */
1293 void blk_error_action(BlockBackend *blk, BlockErrorAction action,
1294 bool is_read, int error)
1295 {
1296 assert(error >= 0);
1297
1298 if (action == BLOCK_ERROR_ACTION_STOP) {
1299 /* First set the iostatus, so that "info block" returns an iostatus
1300 * that matches the events raised so far (an additional error iostatus
1301 * is fine, but not a lost one).
1302 */
1303 blk_iostatus_set_err(blk, error);
1304
1305 /* Then raise the request to stop the VM and the event.
1306 * qemu_system_vmstop_request_prepare has two effects. First,
1307 * it ensures that the STOP event always comes after the
1308 * BLOCK_IO_ERROR event. Second, it ensures that even if management
1309 * can observe the STOP event and do a "cont" before the STOP
1310 * event is issued, the VM will not stop. In this case, vm_start()
1311 * also ensures that the STOP/RESUME pair of events is emitted.
1312 */
1313 qemu_system_vmstop_request_prepare();
1314 send_qmp_error_event(blk, action, is_read, error);
1315 qemu_system_vmstop_request(RUN_STATE_IO_ERROR);
1316 } else {
1317 send_qmp_error_event(blk, action, is_read, error);
1318 }
1319 }
1320
1321 int blk_is_read_only(BlockBackend *blk)
1322 {
1323 BlockDriverState *bs = blk_bs(blk);
1324
1325 if (bs) {
1326 return bdrv_is_read_only(bs);
1327 } else {
1328 return blk->root_state.read_only;
1329 }
1330 }
1331
1332 int blk_is_sg(BlockBackend *blk)
1333 {
1334 BlockDriverState *bs = blk_bs(blk);
1335
1336 if (!bs) {
1337 return 0;
1338 }
1339
1340 return bdrv_is_sg(bs);
1341 }
1342
1343 int blk_enable_write_cache(BlockBackend *blk)
1344 {
1345 return blk->enable_write_cache;
1346 }
1347
1348 void blk_set_enable_write_cache(BlockBackend *blk, bool wce)
1349 {
1350 blk->enable_write_cache = wce;
1351 }
1352
1353 void blk_invalidate_cache(BlockBackend *blk, Error **errp)
1354 {
1355 BlockDriverState *bs = blk_bs(blk);
1356
1357 if (!bs) {
1358 error_setg(errp, "Device '%s' has no medium", blk->name);
1359 return;
1360 }
1361
1362 bdrv_invalidate_cache(bs, errp);
1363 }
1364
1365 bool blk_is_inserted(BlockBackend *blk)
1366 {
1367 BlockDriverState *bs = blk_bs(blk);
1368
1369 return bs && bdrv_is_inserted(bs);
1370 }
1371
1372 bool blk_is_available(BlockBackend *blk)
1373 {
1374 return blk_is_inserted(blk) && !blk_dev_is_tray_open(blk);
1375 }
1376
1377 void blk_lock_medium(BlockBackend *blk, bool locked)
1378 {
1379 BlockDriverState *bs = blk_bs(blk);
1380
1381 if (bs) {
1382 bdrv_lock_medium(bs, locked);
1383 }
1384 }
1385
1386 void blk_eject(BlockBackend *blk, bool eject_flag)
1387 {
1388 BlockDriverState *bs = blk_bs(blk);
1389 char *id;
1390
1391 /* blk_eject is only called by qdevified devices */
1392 assert(!blk->legacy_dev);
1393
1394 if (bs) {
1395 bdrv_eject(bs, eject_flag);
1396 }
1397
1398 /* Whether or not we ejected on the backend,
1399 * the frontend experienced a tray event. */
1400 id = blk_get_attached_dev_id(blk);
1401 qapi_event_send_device_tray_moved(blk_name(blk), id,
1402 eject_flag, &error_abort);
1403 g_free(id);
1404 }
1405
1406 int blk_get_flags(BlockBackend *blk)
1407 {
1408 BlockDriverState *bs = blk_bs(blk);
1409
1410 if (bs) {
1411 return bdrv_get_flags(bs);
1412 } else {
1413 return blk->root_state.open_flags;
1414 }
1415 }
1416
1417 /* Returns the maximum transfer length, in bytes; guaranteed nonzero */
1418 uint32_t blk_get_max_transfer(BlockBackend *blk)
1419 {
1420 BlockDriverState *bs = blk_bs(blk);
1421 uint32_t max = 0;
1422
1423 if (bs) {
1424 max = bs->bl.max_transfer;
1425 }
1426 return MIN_NON_ZERO(max, INT_MAX);
1427 }
1428
1429 int blk_get_max_iov(BlockBackend *blk)
1430 {
1431 return blk->root->bs->bl.max_iov;
1432 }
1433
1434 void blk_set_guest_block_size(BlockBackend *blk, int align)
1435 {
1436 blk->guest_block_size = align;
1437 }
1438
1439 void *blk_try_blockalign(BlockBackend *blk, size_t size)
1440 {
1441 return qemu_try_blockalign(blk ? blk_bs(blk) : NULL, size);
1442 }
1443
1444 void *blk_blockalign(BlockBackend *blk, size_t size)
1445 {
1446 return qemu_blockalign(blk ? blk_bs(blk) : NULL, size);
1447 }
1448
1449 bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp)
1450 {
1451 BlockDriverState *bs = blk_bs(blk);
1452
1453 if (!bs) {
1454 return false;
1455 }
1456
1457 return bdrv_op_is_blocked(bs, op, errp);
1458 }
1459
1460 void blk_op_unblock(BlockBackend *blk, BlockOpType op, Error *reason)
1461 {
1462 BlockDriverState *bs = blk_bs(blk);
1463
1464 if (bs) {
1465 bdrv_op_unblock(bs, op, reason);
1466 }
1467 }
1468
1469 void blk_op_block_all(BlockBackend *blk, Error *reason)
1470 {
1471 BlockDriverState *bs = blk_bs(blk);
1472
1473 if (bs) {
1474 bdrv_op_block_all(bs, reason);
1475 }
1476 }
1477
1478 void blk_op_unblock_all(BlockBackend *blk, Error *reason)
1479 {
1480 BlockDriverState *bs = blk_bs(blk);
1481
1482 if (bs) {
1483 bdrv_op_unblock_all(bs, reason);
1484 }
1485 }
1486
1487 AioContext *blk_get_aio_context(BlockBackend *blk)
1488 {
1489 BlockDriverState *bs = blk_bs(blk);
1490
1491 if (bs) {
1492 return bdrv_get_aio_context(bs);
1493 } else {
1494 return qemu_get_aio_context();
1495 }
1496 }
1497
1498 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb)
1499 {
1500 BlockBackendAIOCB *blk_acb = DO_UPCAST(BlockBackendAIOCB, common, acb);
1501 return blk_get_aio_context(blk_acb->blk);
1502 }
1503
1504 void blk_set_aio_context(BlockBackend *blk, AioContext *new_context)
1505 {
1506 BlockDriverState *bs = blk_bs(blk);
1507
1508 if (bs) {
1509 if (blk->public.throttle_state) {
1510 throttle_timers_detach_aio_context(&blk->public.throttle_timers);
1511 }
1512 bdrv_set_aio_context(bs, new_context);
1513 if (blk->public.throttle_state) {
1514 throttle_timers_attach_aio_context(&blk->public.throttle_timers,
1515 new_context);
1516 }
1517 }
1518 }
1519
1520 void blk_add_aio_context_notifier(BlockBackend *blk,
1521 void (*attached_aio_context)(AioContext *new_context, void *opaque),
1522 void (*detach_aio_context)(void *opaque), void *opaque)
1523 {
1524 BlockDriverState *bs = blk_bs(blk);
1525
1526 if (bs) {
1527 bdrv_add_aio_context_notifier(bs, attached_aio_context,
1528 detach_aio_context, opaque);
1529 }
1530 }
1531
1532 void blk_remove_aio_context_notifier(BlockBackend *blk,
1533 void (*attached_aio_context)(AioContext *,
1534 void *),
1535 void (*detach_aio_context)(void *),
1536 void *opaque)
1537 {
1538 BlockDriverState *bs = blk_bs(blk);
1539
1540 if (bs) {
1541 bdrv_remove_aio_context_notifier(bs, attached_aio_context,
1542 detach_aio_context, opaque);
1543 }
1544 }
1545
1546 void blk_add_remove_bs_notifier(BlockBackend *blk, Notifier *notify)
1547 {
1548 notifier_list_add(&blk->remove_bs_notifiers, notify);
1549 }
1550
1551 void blk_add_insert_bs_notifier(BlockBackend *blk, Notifier *notify)
1552 {
1553 notifier_list_add(&blk->insert_bs_notifiers, notify);
1554 }
1555
1556 void blk_io_plug(BlockBackend *blk)
1557 {
1558 BlockDriverState *bs = blk_bs(blk);
1559
1560 if (bs) {
1561 bdrv_io_plug(bs);
1562 }
1563 }
1564
1565 void blk_io_unplug(BlockBackend *blk)
1566 {
1567 BlockDriverState *bs = blk_bs(blk);
1568
1569 if (bs) {
1570 bdrv_io_unplug(bs);
1571 }
1572 }
1573
1574 BlockAcctStats *blk_get_stats(BlockBackend *blk)
1575 {
1576 return &blk->stats;
1577 }
1578
1579 void *blk_aio_get(const AIOCBInfo *aiocb_info, BlockBackend *blk,
1580 BlockCompletionFunc *cb, void *opaque)
1581 {
1582 return qemu_aio_get(aiocb_info, blk_bs(blk), cb, opaque);
1583 }
1584
1585 int coroutine_fn blk_co_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1586 int count, BdrvRequestFlags flags)
1587 {
1588 return blk_co_pwritev(blk, offset, count, NULL,
1589 flags | BDRV_REQ_ZERO_WRITE);
1590 }
1591
1592 int blk_pwrite_compressed(BlockBackend *blk, int64_t offset, const void *buf,
1593 int count)
1594 {
1595 return blk_prw(blk, offset, (void *) buf, count, blk_write_entry,
1596 BDRV_REQ_WRITE_COMPRESSED);
1597 }
1598
1599 int blk_truncate(BlockBackend *blk, int64_t offset)
1600 {
1601 if (!blk_is_available(blk)) {
1602 return -ENOMEDIUM;
1603 }
1604
1605 return bdrv_truncate(blk_bs(blk), offset);
1606 }
1607
1608 static void blk_pdiscard_entry(void *opaque)
1609 {
1610 BlkRwCo *rwco = opaque;
1611 rwco->ret = blk_co_pdiscard(rwco->blk, rwco->offset, rwco->qiov->size);
1612 }
1613
1614 int blk_pdiscard(BlockBackend *blk, int64_t offset, int count)
1615 {
1616 return blk_prw(blk, offset, NULL, count, blk_pdiscard_entry, 0);
1617 }
1618
1619 int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
1620 int64_t pos, int size)
1621 {
1622 int ret;
1623
1624 if (!blk_is_available(blk)) {
1625 return -ENOMEDIUM;
1626 }
1627
1628 ret = bdrv_save_vmstate(blk_bs(blk), buf, pos, size);
1629 if (ret < 0) {
1630 return ret;
1631 }
1632
1633 if (ret == size && !blk->enable_write_cache) {
1634 ret = bdrv_flush(blk_bs(blk));
1635 }
1636
1637 return ret < 0 ? ret : size;
1638 }
1639
1640 int blk_load_vmstate(BlockBackend *blk, uint8_t *buf, int64_t pos, int size)
1641 {
1642 if (!blk_is_available(blk)) {
1643 return -ENOMEDIUM;
1644 }
1645
1646 return bdrv_load_vmstate(blk_bs(blk), buf, pos, size);
1647 }
1648
1649 int blk_probe_blocksizes(BlockBackend *blk, BlockSizes *bsz)
1650 {
1651 if (!blk_is_available(blk)) {
1652 return -ENOMEDIUM;
1653 }
1654
1655 return bdrv_probe_blocksizes(blk_bs(blk), bsz);
1656 }
1657
1658 int blk_probe_geometry(BlockBackend *blk, HDGeometry *geo)
1659 {
1660 if (!blk_is_available(blk)) {
1661 return -ENOMEDIUM;
1662 }
1663
1664 return bdrv_probe_geometry(blk_bs(blk), geo);
1665 }
1666
1667 /*
1668 * Updates the BlockBackendRootState object with data from the currently
1669 * attached BlockDriverState.
1670 */
1671 void blk_update_root_state(BlockBackend *blk)
1672 {
1673 assert(blk->root);
1674
1675 blk->root_state.open_flags = blk->root->bs->open_flags;
1676 blk->root_state.read_only = blk->root->bs->read_only;
1677 blk->root_state.detect_zeroes = blk->root->bs->detect_zeroes;
1678 }
1679
1680 /*
1681 * Returns the detect-zeroes setting to be used for bdrv_open() of a
1682 * BlockDriverState which is supposed to inherit the root state.
1683 */
1684 bool blk_get_detect_zeroes_from_root_state(BlockBackend *blk)
1685 {
1686 return blk->root_state.detect_zeroes;
1687 }
1688
1689 /*
1690 * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
1691 * supposed to inherit the root state.
1692 */
1693 int blk_get_open_flags_from_root_state(BlockBackend *blk)
1694 {
1695 int bs_flags;
1696
1697 bs_flags = blk->root_state.read_only ? 0 : BDRV_O_RDWR;
1698 bs_flags |= blk->root_state.open_flags & ~BDRV_O_RDWR;
1699
1700 return bs_flags;
1701 }
1702
1703 BlockBackendRootState *blk_get_root_state(BlockBackend *blk)
1704 {
1705 return &blk->root_state;
1706 }
1707
1708 int blk_commit_all(void)
1709 {
1710 BlockBackend *blk = NULL;
1711
1712 while ((blk = blk_all_next(blk)) != NULL) {
1713 AioContext *aio_context = blk_get_aio_context(blk);
1714
1715 aio_context_acquire(aio_context);
1716 if (blk_is_inserted(blk) && blk->root->bs->backing) {
1717 int ret = bdrv_commit(blk->root->bs);
1718 if (ret < 0) {
1719 aio_context_release(aio_context);
1720 return ret;
1721 }
1722 }
1723 aio_context_release(aio_context);
1724 }
1725 return 0;
1726 }
1727
1728
1729 /* throttling disk I/O limits */
1730 void blk_set_io_limits(BlockBackend *blk, ThrottleConfig *cfg)
1731 {
1732 throttle_group_config(blk, cfg);
1733 }
1734
1735 void blk_io_limits_disable(BlockBackend *blk)
1736 {
1737 assert(blk->public.throttle_state);
1738 bdrv_drained_begin(blk_bs(blk));
1739 throttle_group_unregister_blk(blk);
1740 bdrv_drained_end(blk_bs(blk));
1741 }
1742
1743 /* should be called before blk_set_io_limits if a limit is set */
1744 void blk_io_limits_enable(BlockBackend *blk, const char *group)
1745 {
1746 assert(!blk->public.throttle_state);
1747 throttle_group_register_blk(blk, group);
1748 }
1749
1750 void blk_io_limits_update_group(BlockBackend *blk, const char *group)
1751 {
1752 /* this BB is not part of any group */
1753 if (!blk->public.throttle_state) {
1754 return;
1755 }
1756
1757 /* this BB is a part of the same group than the one we want */
1758 if (!g_strcmp0(throttle_group_get_name(blk), group)) {
1759 return;
1760 }
1761
1762 /* need to change the group this bs belong to */
1763 blk_io_limits_disable(blk);
1764 blk_io_limits_enable(blk, group);
1765 }
1766
1767 static void blk_root_drained_begin(BdrvChild *child)
1768 {
1769 BlockBackend *blk = child->opaque;
1770
1771 /* Note that blk->root may not be accessible here yet if we are just
1772 * attaching to a BlockDriverState that is drained. Use child instead. */
1773
1774 if (blk->public.io_limits_disabled++ == 0) {
1775 throttle_group_restart_blk(blk);
1776 }
1777 }
1778
1779 static void blk_root_drained_end(BdrvChild *child)
1780 {
1781 BlockBackend *blk = child->opaque;
1782
1783 assert(blk->public.io_limits_disabled);
1784 --blk->public.io_limits_disabled;
1785 }