spapr: Workaround for broken radix guests
[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 uint64_t perm;
63 uint64_t shared_perm;
64 bool disable_perm;
65
66 bool allow_write_beyond_eof;
67
68 NotifierList remove_bs_notifiers, insert_bs_notifiers;
69
70 int quiesce_counter;
71 };
72
73 typedef struct BlockBackendAIOCB {
74 BlockAIOCB common;
75 BlockBackend *blk;
76 int ret;
77 } BlockBackendAIOCB;
78
79 static const AIOCBInfo block_backend_aiocb_info = {
80 .get_aio_context = blk_aiocb_get_aio_context,
81 .aiocb_size = sizeof(BlockBackendAIOCB),
82 };
83
84 static void drive_info_del(DriveInfo *dinfo);
85 static BlockBackend *bdrv_first_blk(BlockDriverState *bs);
86 static char *blk_get_attached_dev_id(BlockBackend *blk);
87
88 /* All BlockBackends */
89 static QTAILQ_HEAD(, BlockBackend) block_backends =
90 QTAILQ_HEAD_INITIALIZER(block_backends);
91
92 /* All BlockBackends referenced by the monitor and which are iterated through by
93 * blk_next() */
94 static QTAILQ_HEAD(, BlockBackend) monitor_block_backends =
95 QTAILQ_HEAD_INITIALIZER(monitor_block_backends);
96
97 static void blk_root_inherit_options(int *child_flags, QDict *child_options,
98 int parent_flags, QDict *parent_options)
99 {
100 /* We're not supposed to call this function for root nodes */
101 abort();
102 }
103 static void blk_root_drained_begin(BdrvChild *child);
104 static void blk_root_drained_end(BdrvChild *child);
105
106 static void blk_root_change_media(BdrvChild *child, bool load);
107 static void blk_root_resize(BdrvChild *child);
108
109 static char *blk_root_get_parent_desc(BdrvChild *child)
110 {
111 BlockBackend *blk = child->opaque;
112 char *dev_id;
113
114 if (blk->name) {
115 return g_strdup(blk->name);
116 }
117
118 dev_id = blk_get_attached_dev_id(blk);
119 if (*dev_id) {
120 return dev_id;
121 } else {
122 /* TODO Callback into the BB owner for something more detailed */
123 g_free(dev_id);
124 return g_strdup("a block device");
125 }
126 }
127
128 static const char *blk_root_get_name(BdrvChild *child)
129 {
130 return blk_name(child->opaque);
131 }
132
133 static const BdrvChildRole child_root = {
134 .inherit_options = blk_root_inherit_options,
135
136 .change_media = blk_root_change_media,
137 .resize = blk_root_resize,
138 .get_name = blk_root_get_name,
139 .get_parent_desc = blk_root_get_parent_desc,
140
141 .drained_begin = blk_root_drained_begin,
142 .drained_end = blk_root_drained_end,
143 };
144
145 /*
146 * Create a new BlockBackend with a reference count of one.
147 *
148 * @perm is a bitmasks of BLK_PERM_* constants which describes the permissions
149 * to request for a block driver node that is attached to this BlockBackend.
150 * @shared_perm is a bitmask which describes which permissions may be granted
151 * to other users of the attached node.
152 * Both sets of permissions can be changed later using blk_set_perm().
153 *
154 * Return the new BlockBackend on success, null on failure.
155 */
156 BlockBackend *blk_new(uint64_t perm, uint64_t shared_perm)
157 {
158 BlockBackend *blk;
159
160 blk = g_new0(BlockBackend, 1);
161 blk->refcnt = 1;
162 blk->perm = perm;
163 blk->shared_perm = shared_perm;
164 blk_set_enable_write_cache(blk, true);
165
166 qemu_co_queue_init(&blk->public.throttled_reqs[0]);
167 qemu_co_queue_init(&blk->public.throttled_reqs[1]);
168
169 notifier_list_init(&blk->remove_bs_notifiers);
170 notifier_list_init(&blk->insert_bs_notifiers);
171
172 QTAILQ_INSERT_TAIL(&block_backends, blk, link);
173 return blk;
174 }
175
176 /*
177 * Creates a new BlockBackend, opens a new BlockDriverState, and connects both.
178 *
179 * Just as with bdrv_open(), after having called this function the reference to
180 * @options belongs to the block layer (even on failure).
181 *
182 * TODO: Remove @filename and @flags; it should be possible to specify a whole
183 * BDS tree just by specifying the @options QDict (or @reference,
184 * alternatively). At the time of adding this function, this is not possible,
185 * though, so callers of this function have to be able to specify @filename and
186 * @flags.
187 */
188 BlockBackend *blk_new_open(const char *filename, const char *reference,
189 QDict *options, int flags, Error **errp)
190 {
191 BlockBackend *blk;
192 BlockDriverState *bs;
193 uint64_t perm;
194
195 /* blk_new_open() is mainly used in .bdrv_create implementations and the
196 * tools where sharing isn't a concern because the BDS stays private, so we
197 * just request permission according to the flags.
198 *
199 * The exceptions are xen_disk and blockdev_init(); in these cases, the
200 * caller of blk_new_open() doesn't make use of the permissions, but they
201 * shouldn't hurt either. We can still share everything here because the
202 * guest devices will add their own blockers if they can't share. */
203 perm = BLK_PERM_CONSISTENT_READ;
204 if (flags & BDRV_O_RDWR) {
205 perm |= BLK_PERM_WRITE;
206 }
207 if (flags & BDRV_O_RESIZE) {
208 perm |= BLK_PERM_RESIZE;
209 }
210
211 blk = blk_new(perm, BLK_PERM_ALL);
212 bs = bdrv_open(filename, reference, options, flags, errp);
213 if (!bs) {
214 blk_unref(blk);
215 return NULL;
216 }
217
218 blk->root = bdrv_root_attach_child(bs, "root", &child_root,
219 perm, BLK_PERM_ALL, blk, errp);
220 if (!blk->root) {
221 bdrv_unref(bs);
222 blk_unref(blk);
223 return NULL;
224 }
225
226 return blk;
227 }
228
229 static void blk_delete(BlockBackend *blk)
230 {
231 assert(!blk->refcnt);
232 assert(!blk->name);
233 assert(!blk->dev);
234 if (blk->public.throttle_state) {
235 blk_io_limits_disable(blk);
236 }
237 if (blk->root) {
238 blk_remove_bs(blk);
239 }
240 assert(QLIST_EMPTY(&blk->remove_bs_notifiers.notifiers));
241 assert(QLIST_EMPTY(&blk->insert_bs_notifiers.notifiers));
242 QTAILQ_REMOVE(&block_backends, blk, link);
243 drive_info_del(blk->legacy_dinfo);
244 block_acct_cleanup(&blk->stats);
245 g_free(blk);
246 }
247
248 static void drive_info_del(DriveInfo *dinfo)
249 {
250 if (!dinfo) {
251 return;
252 }
253 qemu_opts_del(dinfo->opts);
254 g_free(dinfo->serial);
255 g_free(dinfo);
256 }
257
258 int blk_get_refcnt(BlockBackend *blk)
259 {
260 return blk ? blk->refcnt : 0;
261 }
262
263 /*
264 * Increment @blk's reference count.
265 * @blk must not be null.
266 */
267 void blk_ref(BlockBackend *blk)
268 {
269 blk->refcnt++;
270 }
271
272 /*
273 * Decrement @blk's reference count.
274 * If this drops it to zero, destroy @blk.
275 * For convenience, do nothing if @blk is null.
276 */
277 void blk_unref(BlockBackend *blk)
278 {
279 if (blk) {
280 assert(blk->refcnt > 0);
281 if (!--blk->refcnt) {
282 blk_delete(blk);
283 }
284 }
285 }
286
287 /*
288 * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
289 * ones which are hidden (i.e. are not referenced by the monitor).
290 */
291 static BlockBackend *blk_all_next(BlockBackend *blk)
292 {
293 return blk ? QTAILQ_NEXT(blk, link)
294 : QTAILQ_FIRST(&block_backends);
295 }
296
297 void blk_remove_all_bs(void)
298 {
299 BlockBackend *blk = NULL;
300
301 while ((blk = blk_all_next(blk)) != NULL) {
302 AioContext *ctx = blk_get_aio_context(blk);
303
304 aio_context_acquire(ctx);
305 if (blk->root) {
306 blk_remove_bs(blk);
307 }
308 aio_context_release(ctx);
309 }
310 }
311
312 /*
313 * Return the monitor-owned BlockBackend after @blk.
314 * If @blk is null, return the first one.
315 * Else, return @blk's next sibling, which may be null.
316 *
317 * To iterate over all BlockBackends, do
318 * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
319 * ...
320 * }
321 */
322 BlockBackend *blk_next(BlockBackend *blk)
323 {
324 return blk ? QTAILQ_NEXT(blk, monitor_link)
325 : QTAILQ_FIRST(&monitor_block_backends);
326 }
327
328 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
329 * the monitor or attached to a BlockBackend */
330 BlockDriverState *bdrv_next(BdrvNextIterator *it)
331 {
332 BlockDriverState *bs;
333
334 /* First, return all root nodes of BlockBackends. In order to avoid
335 * returning a BDS twice when multiple BBs refer to it, we only return it
336 * if the BB is the first one in the parent list of the BDS. */
337 if (it->phase == BDRV_NEXT_BACKEND_ROOTS) {
338 do {
339 it->blk = blk_all_next(it->blk);
340 bs = it->blk ? blk_bs(it->blk) : NULL;
341 } while (it->blk && (bs == NULL || bdrv_first_blk(bs) != it->blk));
342
343 if (bs) {
344 return bs;
345 }
346 it->phase = BDRV_NEXT_MONITOR_OWNED;
347 }
348
349 /* Then return the monitor-owned BDSes without a BB attached. Ignore all
350 * BDSes that are attached to a BlockBackend here; they have been handled
351 * by the above block already */
352 do {
353 it->bs = bdrv_next_monitor_owned(it->bs);
354 bs = it->bs;
355 } while (bs && bdrv_has_blk(bs));
356
357 return bs;
358 }
359
360 BlockDriverState *bdrv_first(BdrvNextIterator *it)
361 {
362 *it = (BdrvNextIterator) {
363 .phase = BDRV_NEXT_BACKEND_ROOTS,
364 };
365
366 return bdrv_next(it);
367 }
368
369 /*
370 * Add a BlockBackend into the list of backends referenced by the monitor, with
371 * the given @name acting as the handle for the monitor.
372 * Strictly for use by blockdev.c.
373 *
374 * @name must not be null or empty.
375 *
376 * Returns true on success and false on failure. In the latter case, an Error
377 * object is returned through @errp.
378 */
379 bool monitor_add_blk(BlockBackend *blk, const char *name, Error **errp)
380 {
381 assert(!blk->name);
382 assert(name && name[0]);
383
384 if (!id_wellformed(name)) {
385 error_setg(errp, "Invalid device name");
386 return false;
387 }
388 if (blk_by_name(name)) {
389 error_setg(errp, "Device with id '%s' already exists", name);
390 return false;
391 }
392 if (bdrv_find_node(name)) {
393 error_setg(errp,
394 "Device name '%s' conflicts with an existing node name",
395 name);
396 return false;
397 }
398
399 blk->name = g_strdup(name);
400 QTAILQ_INSERT_TAIL(&monitor_block_backends, blk, monitor_link);
401 return true;
402 }
403
404 /*
405 * Remove a BlockBackend from the list of backends referenced by the monitor.
406 * Strictly for use by blockdev.c.
407 */
408 void monitor_remove_blk(BlockBackend *blk)
409 {
410 if (!blk->name) {
411 return;
412 }
413
414 QTAILQ_REMOVE(&monitor_block_backends, blk, monitor_link);
415 g_free(blk->name);
416 blk->name = NULL;
417 }
418
419 /*
420 * Return @blk's name, a non-null string.
421 * Returns an empty string iff @blk is not referenced by the monitor.
422 */
423 const char *blk_name(BlockBackend *blk)
424 {
425 return blk->name ?: "";
426 }
427
428 /*
429 * Return the BlockBackend with name @name if it exists, else null.
430 * @name must not be null.
431 */
432 BlockBackend *blk_by_name(const char *name)
433 {
434 BlockBackend *blk = NULL;
435
436 assert(name);
437 while ((blk = blk_next(blk)) != NULL) {
438 if (!strcmp(name, blk->name)) {
439 return blk;
440 }
441 }
442 return NULL;
443 }
444
445 /*
446 * Return the BlockDriverState attached to @blk if any, else null.
447 */
448 BlockDriverState *blk_bs(BlockBackend *blk)
449 {
450 return blk->root ? blk->root->bs : NULL;
451 }
452
453 static BlockBackend *bdrv_first_blk(BlockDriverState *bs)
454 {
455 BdrvChild *child;
456 QLIST_FOREACH(child, &bs->parents, next_parent) {
457 if (child->role == &child_root) {
458 return child->opaque;
459 }
460 }
461
462 return NULL;
463 }
464
465 /*
466 * Returns true if @bs has an associated BlockBackend.
467 */
468 bool bdrv_has_blk(BlockDriverState *bs)
469 {
470 return bdrv_first_blk(bs) != NULL;
471 }
472
473 /*
474 * Returns true if @bs has only BlockBackends as parents.
475 */
476 bool bdrv_is_root_node(BlockDriverState *bs)
477 {
478 BdrvChild *c;
479
480 QLIST_FOREACH(c, &bs->parents, next_parent) {
481 if (c->role != &child_root) {
482 return false;
483 }
484 }
485
486 return true;
487 }
488
489 /*
490 * Return @blk's DriveInfo if any, else null.
491 */
492 DriveInfo *blk_legacy_dinfo(BlockBackend *blk)
493 {
494 return blk->legacy_dinfo;
495 }
496
497 /*
498 * Set @blk's DriveInfo to @dinfo, and return it.
499 * @blk must not have a DriveInfo set already.
500 * No other BlockBackend may have the same DriveInfo set.
501 */
502 DriveInfo *blk_set_legacy_dinfo(BlockBackend *blk, DriveInfo *dinfo)
503 {
504 assert(!blk->legacy_dinfo);
505 return blk->legacy_dinfo = dinfo;
506 }
507
508 /*
509 * Return the BlockBackend with DriveInfo @dinfo.
510 * It must exist.
511 */
512 BlockBackend *blk_by_legacy_dinfo(DriveInfo *dinfo)
513 {
514 BlockBackend *blk = NULL;
515
516 while ((blk = blk_next(blk)) != NULL) {
517 if (blk->legacy_dinfo == dinfo) {
518 return blk;
519 }
520 }
521 abort();
522 }
523
524 /*
525 * Returns a pointer to the publicly accessible fields of @blk.
526 */
527 BlockBackendPublic *blk_get_public(BlockBackend *blk)
528 {
529 return &blk->public;
530 }
531
532 /*
533 * Returns a BlockBackend given the associated @public fields.
534 */
535 BlockBackend *blk_by_public(BlockBackendPublic *public)
536 {
537 return container_of(public, BlockBackend, public);
538 }
539
540 /*
541 * Disassociates the currently associated BlockDriverState from @blk.
542 */
543 void blk_remove_bs(BlockBackend *blk)
544 {
545 notifier_list_notify(&blk->remove_bs_notifiers, blk);
546 if (blk->public.throttle_state) {
547 throttle_timers_detach_aio_context(&blk->public.throttle_timers);
548 }
549
550 blk_update_root_state(blk);
551
552 bdrv_root_unref_child(blk->root);
553 blk->root = NULL;
554 }
555
556 /*
557 * Associates a new BlockDriverState with @blk.
558 */
559 int blk_insert_bs(BlockBackend *blk, BlockDriverState *bs, Error **errp)
560 {
561 blk->root = bdrv_root_attach_child(bs, "root", &child_root,
562 blk->perm, blk->shared_perm, blk, errp);
563 if (blk->root == NULL) {
564 return -EPERM;
565 }
566 bdrv_ref(bs);
567
568 notifier_list_notify(&blk->insert_bs_notifiers, blk);
569 if (blk->public.throttle_state) {
570 throttle_timers_attach_aio_context(
571 &blk->public.throttle_timers, bdrv_get_aio_context(bs));
572 }
573
574 return 0;
575 }
576
577 /*
578 * Sets the permission bitmasks that the user of the BlockBackend needs.
579 */
580 int blk_set_perm(BlockBackend *blk, uint64_t perm, uint64_t shared_perm,
581 Error **errp)
582 {
583 int ret;
584
585 if (blk->root && !blk->disable_perm) {
586 ret = bdrv_child_try_set_perm(blk->root, perm, shared_perm, errp);
587 if (ret < 0) {
588 return ret;
589 }
590 }
591
592 blk->perm = perm;
593 blk->shared_perm = shared_perm;
594
595 return 0;
596 }
597
598 void blk_get_perm(BlockBackend *blk, uint64_t *perm, uint64_t *shared_perm)
599 {
600 *perm = blk->perm;
601 *shared_perm = blk->shared_perm;
602 }
603
604 /*
605 * Notifies the user of all BlockBackends that migration has completed. qdev
606 * devices can tighten their permissions in response (specifically revoke
607 * shared write permissions that we needed for storage migration).
608 *
609 * If an error is returned, the VM cannot be allowed to be resumed.
610 */
611 void blk_resume_after_migration(Error **errp)
612 {
613 BlockBackend *blk;
614 Error *local_err = NULL;
615
616 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
617 if (!blk->disable_perm) {
618 continue;
619 }
620
621 blk->disable_perm = false;
622
623 blk_set_perm(blk, blk->perm, blk->shared_perm, &local_err);
624 if (local_err) {
625 error_propagate(errp, local_err);
626 blk->disable_perm = true;
627 return;
628 }
629 }
630 }
631
632 static int blk_do_attach_dev(BlockBackend *blk, void *dev)
633 {
634 if (blk->dev) {
635 return -EBUSY;
636 }
637
638 /* While migration is still incoming, we don't need to apply the
639 * permissions of guest device BlockBackends. We might still have a block
640 * job or NBD server writing to the image for storage migration. */
641 if (runstate_check(RUN_STATE_INMIGRATE)) {
642 blk->disable_perm = true;
643 }
644
645 blk_ref(blk);
646 blk->dev = dev;
647 blk->legacy_dev = false;
648 blk_iostatus_reset(blk);
649
650 return 0;
651 }
652
653 /*
654 * Attach device model @dev to @blk.
655 * Return 0 on success, -EBUSY when a device model is attached already.
656 */
657 int blk_attach_dev(BlockBackend *blk, DeviceState *dev)
658 {
659 return blk_do_attach_dev(blk, dev);
660 }
661
662 /*
663 * Attach device model @dev to @blk.
664 * @blk must not have a device model attached already.
665 * TODO qdevified devices don't use this, remove when devices are qdevified
666 */
667 void blk_attach_dev_legacy(BlockBackend *blk, void *dev)
668 {
669 if (blk_do_attach_dev(blk, dev) < 0) {
670 abort();
671 }
672 blk->legacy_dev = true;
673 }
674
675 /*
676 * Detach device model @dev from @blk.
677 * @dev must be currently attached to @blk.
678 */
679 void blk_detach_dev(BlockBackend *blk, void *dev)
680 /* TODO change to DeviceState *dev when all users are qdevified */
681 {
682 assert(blk->dev == dev);
683 blk->dev = NULL;
684 blk->dev_ops = NULL;
685 blk->dev_opaque = NULL;
686 blk->guest_block_size = 512;
687 blk_set_perm(blk, 0, BLK_PERM_ALL, &error_abort);
688 blk_unref(blk);
689 }
690
691 /*
692 * Return the device model attached to @blk if any, else null.
693 */
694 void *blk_get_attached_dev(BlockBackend *blk)
695 /* TODO change to return DeviceState * when all users are qdevified */
696 {
697 return blk->dev;
698 }
699
700 /* Return the qdev ID, or if no ID is assigned the QOM path, of the block
701 * device attached to the BlockBackend. */
702 static char *blk_get_attached_dev_id(BlockBackend *blk)
703 {
704 DeviceState *dev;
705
706 assert(!blk->legacy_dev);
707 dev = blk->dev;
708
709 if (!dev) {
710 return g_strdup("");
711 } else if (dev->id) {
712 return g_strdup(dev->id);
713 }
714 return object_get_canonical_path(OBJECT(dev));
715 }
716
717 /*
718 * Return the BlockBackend which has the device model @dev attached if it
719 * exists, else null.
720 *
721 * @dev must not be null.
722 */
723 BlockBackend *blk_by_dev(void *dev)
724 {
725 BlockBackend *blk = NULL;
726
727 assert(dev != NULL);
728 while ((blk = blk_all_next(blk)) != NULL) {
729 if (blk->dev == dev) {
730 return blk;
731 }
732 }
733 return NULL;
734 }
735
736 /*
737 * Set @blk's device model callbacks to @ops.
738 * @opaque is the opaque argument to pass to the callbacks.
739 * This is for use by device models.
740 */
741 void blk_set_dev_ops(BlockBackend *blk, const BlockDevOps *ops,
742 void *opaque)
743 {
744 /* All drivers that use blk_set_dev_ops() are qdevified and we want to keep
745 * it that way, so we can assume blk->dev, if present, is a DeviceState if
746 * blk->dev_ops is set. Non-device users may use dev_ops without device. */
747 assert(!blk->legacy_dev);
748
749 blk->dev_ops = ops;
750 blk->dev_opaque = opaque;
751
752 /* Are we currently quiesced? Should we enforce this right now? */
753 if (blk->quiesce_counter && ops->drained_begin) {
754 ops->drained_begin(opaque);
755 }
756 }
757
758 /*
759 * Notify @blk's attached device model of media change.
760 *
761 * If @load is true, notify of media load. This action can fail, meaning that
762 * the medium cannot be loaded. @errp is set then.
763 *
764 * If @load is false, notify of media eject. This can never fail.
765 *
766 * Also send DEVICE_TRAY_MOVED events as appropriate.
767 */
768 void blk_dev_change_media_cb(BlockBackend *blk, bool load, Error **errp)
769 {
770 if (blk->dev_ops && blk->dev_ops->change_media_cb) {
771 bool tray_was_open, tray_is_open;
772 Error *local_err = NULL;
773
774 assert(!blk->legacy_dev);
775
776 tray_was_open = blk_dev_is_tray_open(blk);
777 blk->dev_ops->change_media_cb(blk->dev_opaque, load, &local_err);
778 if (local_err) {
779 assert(load == true);
780 error_propagate(errp, local_err);
781 return;
782 }
783 tray_is_open = blk_dev_is_tray_open(blk);
784
785 if (tray_was_open != tray_is_open) {
786 char *id = blk_get_attached_dev_id(blk);
787 qapi_event_send_device_tray_moved(blk_name(blk), id, tray_is_open,
788 &error_abort);
789 g_free(id);
790 }
791 }
792 }
793
794 static void blk_root_change_media(BdrvChild *child, bool load)
795 {
796 blk_dev_change_media_cb(child->opaque, load, NULL);
797 }
798
799 /*
800 * Does @blk's attached device model have removable media?
801 * %true if no device model is attached.
802 */
803 bool blk_dev_has_removable_media(BlockBackend *blk)
804 {
805 return !blk->dev || (blk->dev_ops && blk->dev_ops->change_media_cb);
806 }
807
808 /*
809 * Does @blk's attached device model have a tray?
810 */
811 bool blk_dev_has_tray(BlockBackend *blk)
812 {
813 return blk->dev_ops && blk->dev_ops->is_tray_open;
814 }
815
816 /*
817 * Notify @blk's attached device model of a media eject request.
818 * If @force is true, the medium is about to be yanked out forcefully.
819 */
820 void blk_dev_eject_request(BlockBackend *blk, bool force)
821 {
822 if (blk->dev_ops && blk->dev_ops->eject_request_cb) {
823 blk->dev_ops->eject_request_cb(blk->dev_opaque, force);
824 }
825 }
826
827 /*
828 * Does @blk's attached device model have a tray, and is it open?
829 */
830 bool blk_dev_is_tray_open(BlockBackend *blk)
831 {
832 if (blk_dev_has_tray(blk)) {
833 return blk->dev_ops->is_tray_open(blk->dev_opaque);
834 }
835 return false;
836 }
837
838 /*
839 * Does @blk's attached device model have the medium locked?
840 * %false if the device model has no such lock.
841 */
842 bool blk_dev_is_medium_locked(BlockBackend *blk)
843 {
844 if (blk->dev_ops && blk->dev_ops->is_medium_locked) {
845 return blk->dev_ops->is_medium_locked(blk->dev_opaque);
846 }
847 return false;
848 }
849
850 /*
851 * Notify @blk's attached device model of a backend size change.
852 */
853 static void blk_root_resize(BdrvChild *child)
854 {
855 BlockBackend *blk = child->opaque;
856
857 if (blk->dev_ops && blk->dev_ops->resize_cb) {
858 blk->dev_ops->resize_cb(blk->dev_opaque);
859 }
860 }
861
862 void blk_iostatus_enable(BlockBackend *blk)
863 {
864 blk->iostatus_enabled = true;
865 blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
866 }
867
868 /* The I/O status is only enabled if the drive explicitly
869 * enables it _and_ the VM is configured to stop on errors */
870 bool blk_iostatus_is_enabled(const BlockBackend *blk)
871 {
872 return (blk->iostatus_enabled &&
873 (blk->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
874 blk->on_write_error == BLOCKDEV_ON_ERROR_STOP ||
875 blk->on_read_error == BLOCKDEV_ON_ERROR_STOP));
876 }
877
878 BlockDeviceIoStatus blk_iostatus(const BlockBackend *blk)
879 {
880 return blk->iostatus;
881 }
882
883 void blk_iostatus_disable(BlockBackend *blk)
884 {
885 blk->iostatus_enabled = false;
886 }
887
888 void blk_iostatus_reset(BlockBackend *blk)
889 {
890 if (blk_iostatus_is_enabled(blk)) {
891 BlockDriverState *bs = blk_bs(blk);
892 blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
893 if (bs && bs->job) {
894 block_job_iostatus_reset(bs->job);
895 }
896 }
897 }
898
899 void blk_iostatus_set_err(BlockBackend *blk, int error)
900 {
901 assert(blk_iostatus_is_enabled(blk));
902 if (blk->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
903 blk->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
904 BLOCK_DEVICE_IO_STATUS_FAILED;
905 }
906 }
907
908 void blk_set_allow_write_beyond_eof(BlockBackend *blk, bool allow)
909 {
910 blk->allow_write_beyond_eof = allow;
911 }
912
913 static int blk_check_byte_request(BlockBackend *blk, int64_t offset,
914 size_t size)
915 {
916 int64_t len;
917
918 if (size > INT_MAX) {
919 return -EIO;
920 }
921
922 if (!blk_is_available(blk)) {
923 return -ENOMEDIUM;
924 }
925
926 if (offset < 0) {
927 return -EIO;
928 }
929
930 if (!blk->allow_write_beyond_eof) {
931 len = blk_getlength(blk);
932 if (len < 0) {
933 return len;
934 }
935
936 if (offset > len || len - offset < size) {
937 return -EIO;
938 }
939 }
940
941 return 0;
942 }
943
944 int coroutine_fn blk_co_preadv(BlockBackend *blk, int64_t offset,
945 unsigned int bytes, QEMUIOVector *qiov,
946 BdrvRequestFlags flags)
947 {
948 int ret;
949 BlockDriverState *bs = blk_bs(blk);
950
951 trace_blk_co_preadv(blk, bs, offset, bytes, flags);
952
953 ret = blk_check_byte_request(blk, offset, bytes);
954 if (ret < 0) {
955 return ret;
956 }
957
958 bdrv_inc_in_flight(bs);
959
960 /* throttling disk I/O */
961 if (blk->public.throttle_state) {
962 throttle_group_co_io_limits_intercept(blk, bytes, false);
963 }
964
965 ret = bdrv_co_preadv(blk->root, offset, bytes, qiov, flags);
966 bdrv_dec_in_flight(bs);
967 return ret;
968 }
969
970 int coroutine_fn blk_co_pwritev(BlockBackend *blk, int64_t offset,
971 unsigned int bytes, QEMUIOVector *qiov,
972 BdrvRequestFlags flags)
973 {
974 int ret;
975 BlockDriverState *bs = blk_bs(blk);
976
977 trace_blk_co_pwritev(blk, bs, offset, bytes, flags);
978
979 ret = blk_check_byte_request(blk, offset, bytes);
980 if (ret < 0) {
981 return ret;
982 }
983
984 bdrv_inc_in_flight(bs);
985
986 /* throttling disk I/O */
987 if (blk->public.throttle_state) {
988 throttle_group_co_io_limits_intercept(blk, bytes, true);
989 }
990
991 if (!blk->enable_write_cache) {
992 flags |= BDRV_REQ_FUA;
993 }
994
995 ret = bdrv_co_pwritev(blk->root, offset, bytes, qiov, flags);
996 bdrv_dec_in_flight(bs);
997 return ret;
998 }
999
1000 typedef struct BlkRwCo {
1001 BlockBackend *blk;
1002 int64_t offset;
1003 QEMUIOVector *qiov;
1004 int ret;
1005 BdrvRequestFlags flags;
1006 } BlkRwCo;
1007
1008 static void blk_read_entry(void *opaque)
1009 {
1010 BlkRwCo *rwco = opaque;
1011
1012 rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, rwco->qiov->size,
1013 rwco->qiov, rwco->flags);
1014 }
1015
1016 static void blk_write_entry(void *opaque)
1017 {
1018 BlkRwCo *rwco = opaque;
1019
1020 rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, rwco->qiov->size,
1021 rwco->qiov, rwco->flags);
1022 }
1023
1024 static int blk_prw(BlockBackend *blk, int64_t offset, uint8_t *buf,
1025 int64_t bytes, CoroutineEntry co_entry,
1026 BdrvRequestFlags flags)
1027 {
1028 QEMUIOVector qiov;
1029 struct iovec iov;
1030 BlkRwCo rwco;
1031
1032 iov = (struct iovec) {
1033 .iov_base = buf,
1034 .iov_len = bytes,
1035 };
1036 qemu_iovec_init_external(&qiov, &iov, 1);
1037
1038 rwco = (BlkRwCo) {
1039 .blk = blk,
1040 .offset = offset,
1041 .qiov = &qiov,
1042 .flags = flags,
1043 .ret = NOT_DONE,
1044 };
1045
1046 if (qemu_in_coroutine()) {
1047 /* Fast-path if already in coroutine context */
1048 co_entry(&rwco);
1049 } else {
1050 Coroutine *co = qemu_coroutine_create(co_entry, &rwco);
1051 bdrv_coroutine_enter(blk_bs(blk), co);
1052 BDRV_POLL_WHILE(blk_bs(blk), rwco.ret == NOT_DONE);
1053 }
1054
1055 return rwco.ret;
1056 }
1057
1058 int blk_pread_unthrottled(BlockBackend *blk, int64_t offset, uint8_t *buf,
1059 int count)
1060 {
1061 int ret;
1062
1063 ret = blk_check_byte_request(blk, offset, count);
1064 if (ret < 0) {
1065 return ret;
1066 }
1067
1068 blk_root_drained_begin(blk->root);
1069 ret = blk_pread(blk, offset, buf, count);
1070 blk_root_drained_end(blk->root);
1071 return ret;
1072 }
1073
1074 int blk_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1075 int count, BdrvRequestFlags flags)
1076 {
1077 return blk_prw(blk, offset, NULL, count, blk_write_entry,
1078 flags | BDRV_REQ_ZERO_WRITE);
1079 }
1080
1081 int blk_make_zero(BlockBackend *blk, BdrvRequestFlags flags)
1082 {
1083 return bdrv_make_zero(blk->root, flags);
1084 }
1085
1086 static void error_callback_bh(void *opaque)
1087 {
1088 struct BlockBackendAIOCB *acb = opaque;
1089
1090 bdrv_dec_in_flight(acb->common.bs);
1091 acb->common.cb(acb->common.opaque, acb->ret);
1092 qemu_aio_unref(acb);
1093 }
1094
1095 BlockAIOCB *blk_abort_aio_request(BlockBackend *blk,
1096 BlockCompletionFunc *cb,
1097 void *opaque, int ret)
1098 {
1099 struct BlockBackendAIOCB *acb;
1100
1101 bdrv_inc_in_flight(blk_bs(blk));
1102 acb = blk_aio_get(&block_backend_aiocb_info, blk, cb, opaque);
1103 acb->blk = blk;
1104 acb->ret = ret;
1105
1106 aio_bh_schedule_oneshot(blk_get_aio_context(blk), error_callback_bh, acb);
1107 return &acb->common;
1108 }
1109
1110 typedef struct BlkAioEmAIOCB {
1111 BlockAIOCB common;
1112 BlkRwCo rwco;
1113 int bytes;
1114 bool has_returned;
1115 } BlkAioEmAIOCB;
1116
1117 static const AIOCBInfo blk_aio_em_aiocb_info = {
1118 .aiocb_size = sizeof(BlkAioEmAIOCB),
1119 };
1120
1121 static void blk_aio_complete(BlkAioEmAIOCB *acb)
1122 {
1123 if (acb->has_returned) {
1124 bdrv_dec_in_flight(acb->common.bs);
1125 acb->common.cb(acb->common.opaque, acb->rwco.ret);
1126 qemu_aio_unref(acb);
1127 }
1128 }
1129
1130 static void blk_aio_complete_bh(void *opaque)
1131 {
1132 BlkAioEmAIOCB *acb = opaque;
1133 assert(acb->has_returned);
1134 blk_aio_complete(acb);
1135 }
1136
1137 static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset, int bytes,
1138 QEMUIOVector *qiov, CoroutineEntry co_entry,
1139 BdrvRequestFlags flags,
1140 BlockCompletionFunc *cb, void *opaque)
1141 {
1142 BlkAioEmAIOCB *acb;
1143 Coroutine *co;
1144
1145 bdrv_inc_in_flight(blk_bs(blk));
1146 acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
1147 acb->rwco = (BlkRwCo) {
1148 .blk = blk,
1149 .offset = offset,
1150 .qiov = qiov,
1151 .flags = flags,
1152 .ret = NOT_DONE,
1153 };
1154 acb->bytes = bytes;
1155 acb->has_returned = false;
1156
1157 co = qemu_coroutine_create(co_entry, acb);
1158 bdrv_coroutine_enter(blk_bs(blk), co);
1159
1160 acb->has_returned = true;
1161 if (acb->rwco.ret != NOT_DONE) {
1162 aio_bh_schedule_oneshot(blk_get_aio_context(blk),
1163 blk_aio_complete_bh, acb);
1164 }
1165
1166 return &acb->common;
1167 }
1168
1169 static void blk_aio_read_entry(void *opaque)
1170 {
1171 BlkAioEmAIOCB *acb = opaque;
1172 BlkRwCo *rwco = &acb->rwco;
1173
1174 assert(rwco->qiov->size == acb->bytes);
1175 rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, acb->bytes,
1176 rwco->qiov, rwco->flags);
1177 blk_aio_complete(acb);
1178 }
1179
1180 static void blk_aio_write_entry(void *opaque)
1181 {
1182 BlkAioEmAIOCB *acb = opaque;
1183 BlkRwCo *rwco = &acb->rwco;
1184
1185 assert(!rwco->qiov || rwco->qiov->size == acb->bytes);
1186 rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, acb->bytes,
1187 rwco->qiov, rwco->flags);
1188 blk_aio_complete(acb);
1189 }
1190
1191 BlockAIOCB *blk_aio_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1192 int count, BdrvRequestFlags flags,
1193 BlockCompletionFunc *cb, void *opaque)
1194 {
1195 return blk_aio_prwv(blk, offset, count, NULL, blk_aio_write_entry,
1196 flags | BDRV_REQ_ZERO_WRITE, cb, opaque);
1197 }
1198
1199 int blk_pread(BlockBackend *blk, int64_t offset, void *buf, int count)
1200 {
1201 int ret = blk_prw(blk, offset, buf, count, blk_read_entry, 0);
1202 if (ret < 0) {
1203 return ret;
1204 }
1205 return count;
1206 }
1207
1208 int blk_pwrite(BlockBackend *blk, int64_t offset, const void *buf, int count,
1209 BdrvRequestFlags flags)
1210 {
1211 int ret = blk_prw(blk, offset, (void *) buf, count, blk_write_entry,
1212 flags);
1213 if (ret < 0) {
1214 return ret;
1215 }
1216 return count;
1217 }
1218
1219 int64_t blk_getlength(BlockBackend *blk)
1220 {
1221 if (!blk_is_available(blk)) {
1222 return -ENOMEDIUM;
1223 }
1224
1225 return bdrv_getlength(blk_bs(blk));
1226 }
1227
1228 void blk_get_geometry(BlockBackend *blk, uint64_t *nb_sectors_ptr)
1229 {
1230 if (!blk_bs(blk)) {
1231 *nb_sectors_ptr = 0;
1232 } else {
1233 bdrv_get_geometry(blk_bs(blk), nb_sectors_ptr);
1234 }
1235 }
1236
1237 int64_t blk_nb_sectors(BlockBackend *blk)
1238 {
1239 if (!blk_is_available(blk)) {
1240 return -ENOMEDIUM;
1241 }
1242
1243 return bdrv_nb_sectors(blk_bs(blk));
1244 }
1245
1246 BlockAIOCB *blk_aio_preadv(BlockBackend *blk, int64_t offset,
1247 QEMUIOVector *qiov, BdrvRequestFlags flags,
1248 BlockCompletionFunc *cb, void *opaque)
1249 {
1250 return blk_aio_prwv(blk, offset, qiov->size, qiov,
1251 blk_aio_read_entry, flags, cb, opaque);
1252 }
1253
1254 BlockAIOCB *blk_aio_pwritev(BlockBackend *blk, int64_t offset,
1255 QEMUIOVector *qiov, BdrvRequestFlags flags,
1256 BlockCompletionFunc *cb, void *opaque)
1257 {
1258 return blk_aio_prwv(blk, offset, qiov->size, qiov,
1259 blk_aio_write_entry, flags, cb, opaque);
1260 }
1261
1262 static void blk_aio_flush_entry(void *opaque)
1263 {
1264 BlkAioEmAIOCB *acb = opaque;
1265 BlkRwCo *rwco = &acb->rwco;
1266
1267 rwco->ret = blk_co_flush(rwco->blk);
1268 blk_aio_complete(acb);
1269 }
1270
1271 BlockAIOCB *blk_aio_flush(BlockBackend *blk,
1272 BlockCompletionFunc *cb, void *opaque)
1273 {
1274 return blk_aio_prwv(blk, 0, 0, NULL, blk_aio_flush_entry, 0, cb, opaque);
1275 }
1276
1277 static void blk_aio_pdiscard_entry(void *opaque)
1278 {
1279 BlkAioEmAIOCB *acb = opaque;
1280 BlkRwCo *rwco = &acb->rwco;
1281
1282 rwco->ret = blk_co_pdiscard(rwco->blk, rwco->offset, acb->bytes);
1283 blk_aio_complete(acb);
1284 }
1285
1286 BlockAIOCB *blk_aio_pdiscard(BlockBackend *blk,
1287 int64_t offset, int count,
1288 BlockCompletionFunc *cb, void *opaque)
1289 {
1290 return blk_aio_prwv(blk, offset, count, NULL, blk_aio_pdiscard_entry, 0,
1291 cb, opaque);
1292 }
1293
1294 void blk_aio_cancel(BlockAIOCB *acb)
1295 {
1296 bdrv_aio_cancel(acb);
1297 }
1298
1299 void blk_aio_cancel_async(BlockAIOCB *acb)
1300 {
1301 bdrv_aio_cancel_async(acb);
1302 }
1303
1304 int blk_co_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1305 {
1306 if (!blk_is_available(blk)) {
1307 return -ENOMEDIUM;
1308 }
1309
1310 return bdrv_co_ioctl(blk_bs(blk), req, buf);
1311 }
1312
1313 static void blk_ioctl_entry(void *opaque)
1314 {
1315 BlkRwCo *rwco = opaque;
1316 rwco->ret = blk_co_ioctl(rwco->blk, rwco->offset,
1317 rwco->qiov->iov[0].iov_base);
1318 }
1319
1320 int blk_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1321 {
1322 return blk_prw(blk, req, buf, 0, blk_ioctl_entry, 0);
1323 }
1324
1325 static void blk_aio_ioctl_entry(void *opaque)
1326 {
1327 BlkAioEmAIOCB *acb = opaque;
1328 BlkRwCo *rwco = &acb->rwco;
1329
1330 rwco->ret = blk_co_ioctl(rwco->blk, rwco->offset,
1331 rwco->qiov->iov[0].iov_base);
1332 blk_aio_complete(acb);
1333 }
1334
1335 BlockAIOCB *blk_aio_ioctl(BlockBackend *blk, unsigned long int req, void *buf,
1336 BlockCompletionFunc *cb, void *opaque)
1337 {
1338 QEMUIOVector qiov;
1339 struct iovec iov;
1340
1341 iov = (struct iovec) {
1342 .iov_base = buf,
1343 .iov_len = 0,
1344 };
1345 qemu_iovec_init_external(&qiov, &iov, 1);
1346
1347 return blk_aio_prwv(blk, req, 0, &qiov, blk_aio_ioctl_entry, 0, cb, opaque);
1348 }
1349
1350 int blk_co_pdiscard(BlockBackend *blk, int64_t offset, int count)
1351 {
1352 int ret = blk_check_byte_request(blk, offset, count);
1353 if (ret < 0) {
1354 return ret;
1355 }
1356
1357 return bdrv_co_pdiscard(blk_bs(blk), offset, count);
1358 }
1359
1360 int blk_co_flush(BlockBackend *blk)
1361 {
1362 if (!blk_is_available(blk)) {
1363 return -ENOMEDIUM;
1364 }
1365
1366 return bdrv_co_flush(blk_bs(blk));
1367 }
1368
1369 static void blk_flush_entry(void *opaque)
1370 {
1371 BlkRwCo *rwco = opaque;
1372 rwco->ret = blk_co_flush(rwco->blk);
1373 }
1374
1375 int blk_flush(BlockBackend *blk)
1376 {
1377 return blk_prw(blk, 0, NULL, 0, blk_flush_entry, 0);
1378 }
1379
1380 void blk_drain(BlockBackend *blk)
1381 {
1382 if (blk_bs(blk)) {
1383 bdrv_drain(blk_bs(blk));
1384 }
1385 }
1386
1387 void blk_drain_all(void)
1388 {
1389 bdrv_drain_all();
1390 }
1391
1392 void blk_set_on_error(BlockBackend *blk, BlockdevOnError on_read_error,
1393 BlockdevOnError on_write_error)
1394 {
1395 blk->on_read_error = on_read_error;
1396 blk->on_write_error = on_write_error;
1397 }
1398
1399 BlockdevOnError blk_get_on_error(BlockBackend *blk, bool is_read)
1400 {
1401 return is_read ? blk->on_read_error : blk->on_write_error;
1402 }
1403
1404 BlockErrorAction blk_get_error_action(BlockBackend *blk, bool is_read,
1405 int error)
1406 {
1407 BlockdevOnError on_err = blk_get_on_error(blk, is_read);
1408
1409 switch (on_err) {
1410 case BLOCKDEV_ON_ERROR_ENOSPC:
1411 return (error == ENOSPC) ?
1412 BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
1413 case BLOCKDEV_ON_ERROR_STOP:
1414 return BLOCK_ERROR_ACTION_STOP;
1415 case BLOCKDEV_ON_ERROR_REPORT:
1416 return BLOCK_ERROR_ACTION_REPORT;
1417 case BLOCKDEV_ON_ERROR_IGNORE:
1418 return BLOCK_ERROR_ACTION_IGNORE;
1419 case BLOCKDEV_ON_ERROR_AUTO:
1420 default:
1421 abort();
1422 }
1423 }
1424
1425 static void send_qmp_error_event(BlockBackend *blk,
1426 BlockErrorAction action,
1427 bool is_read, int error)
1428 {
1429 IoOperationType optype;
1430
1431 optype = is_read ? IO_OPERATION_TYPE_READ : IO_OPERATION_TYPE_WRITE;
1432 qapi_event_send_block_io_error(blk_name(blk),
1433 bdrv_get_node_name(blk_bs(blk)), optype,
1434 action, blk_iostatus_is_enabled(blk),
1435 error == ENOSPC, strerror(error),
1436 &error_abort);
1437 }
1438
1439 /* This is done by device models because, while the block layer knows
1440 * about the error, it does not know whether an operation comes from
1441 * the device or the block layer (from a job, for example).
1442 */
1443 void blk_error_action(BlockBackend *blk, BlockErrorAction action,
1444 bool is_read, int error)
1445 {
1446 assert(error >= 0);
1447
1448 if (action == BLOCK_ERROR_ACTION_STOP) {
1449 /* First set the iostatus, so that "info block" returns an iostatus
1450 * that matches the events raised so far (an additional error iostatus
1451 * is fine, but not a lost one).
1452 */
1453 blk_iostatus_set_err(blk, error);
1454
1455 /* Then raise the request to stop the VM and the event.
1456 * qemu_system_vmstop_request_prepare has two effects. First,
1457 * it ensures that the STOP event always comes after the
1458 * BLOCK_IO_ERROR event. Second, it ensures that even if management
1459 * can observe the STOP event and do a "cont" before the STOP
1460 * event is issued, the VM will not stop. In this case, vm_start()
1461 * also ensures that the STOP/RESUME pair of events is emitted.
1462 */
1463 qemu_system_vmstop_request_prepare();
1464 send_qmp_error_event(blk, action, is_read, error);
1465 qemu_system_vmstop_request(RUN_STATE_IO_ERROR);
1466 } else {
1467 send_qmp_error_event(blk, action, is_read, error);
1468 }
1469 }
1470
1471 int blk_is_read_only(BlockBackend *blk)
1472 {
1473 BlockDriverState *bs = blk_bs(blk);
1474
1475 if (bs) {
1476 return bdrv_is_read_only(bs);
1477 } else {
1478 return blk->root_state.read_only;
1479 }
1480 }
1481
1482 int blk_is_sg(BlockBackend *blk)
1483 {
1484 BlockDriverState *bs = blk_bs(blk);
1485
1486 if (!bs) {
1487 return 0;
1488 }
1489
1490 return bdrv_is_sg(bs);
1491 }
1492
1493 int blk_enable_write_cache(BlockBackend *blk)
1494 {
1495 return blk->enable_write_cache;
1496 }
1497
1498 void blk_set_enable_write_cache(BlockBackend *blk, bool wce)
1499 {
1500 blk->enable_write_cache = wce;
1501 }
1502
1503 void blk_invalidate_cache(BlockBackend *blk, Error **errp)
1504 {
1505 BlockDriverState *bs = blk_bs(blk);
1506
1507 if (!bs) {
1508 error_setg(errp, "Device '%s' has no medium", blk->name);
1509 return;
1510 }
1511
1512 bdrv_invalidate_cache(bs, errp);
1513 }
1514
1515 bool blk_is_inserted(BlockBackend *blk)
1516 {
1517 BlockDriverState *bs = blk_bs(blk);
1518
1519 return bs && bdrv_is_inserted(bs);
1520 }
1521
1522 bool blk_is_available(BlockBackend *blk)
1523 {
1524 return blk_is_inserted(blk) && !blk_dev_is_tray_open(blk);
1525 }
1526
1527 void blk_lock_medium(BlockBackend *blk, bool locked)
1528 {
1529 BlockDriverState *bs = blk_bs(blk);
1530
1531 if (bs) {
1532 bdrv_lock_medium(bs, locked);
1533 }
1534 }
1535
1536 void blk_eject(BlockBackend *blk, bool eject_flag)
1537 {
1538 BlockDriverState *bs = blk_bs(blk);
1539 char *id;
1540
1541 /* blk_eject is only called by qdevified devices */
1542 assert(!blk->legacy_dev);
1543
1544 if (bs) {
1545 bdrv_eject(bs, eject_flag);
1546 }
1547
1548 /* Whether or not we ejected on the backend,
1549 * the frontend experienced a tray event. */
1550 id = blk_get_attached_dev_id(blk);
1551 qapi_event_send_device_tray_moved(blk_name(blk), id,
1552 eject_flag, &error_abort);
1553 g_free(id);
1554 }
1555
1556 int blk_get_flags(BlockBackend *blk)
1557 {
1558 BlockDriverState *bs = blk_bs(blk);
1559
1560 if (bs) {
1561 return bdrv_get_flags(bs);
1562 } else {
1563 return blk->root_state.open_flags;
1564 }
1565 }
1566
1567 /* Returns the maximum transfer length, in bytes; guaranteed nonzero */
1568 uint32_t blk_get_max_transfer(BlockBackend *blk)
1569 {
1570 BlockDriverState *bs = blk_bs(blk);
1571 uint32_t max = 0;
1572
1573 if (bs) {
1574 max = bs->bl.max_transfer;
1575 }
1576 return MIN_NON_ZERO(max, INT_MAX);
1577 }
1578
1579 int blk_get_max_iov(BlockBackend *blk)
1580 {
1581 return blk->root->bs->bl.max_iov;
1582 }
1583
1584 void blk_set_guest_block_size(BlockBackend *blk, int align)
1585 {
1586 blk->guest_block_size = align;
1587 }
1588
1589 void *blk_try_blockalign(BlockBackend *blk, size_t size)
1590 {
1591 return qemu_try_blockalign(blk ? blk_bs(blk) : NULL, size);
1592 }
1593
1594 void *blk_blockalign(BlockBackend *blk, size_t size)
1595 {
1596 return qemu_blockalign(blk ? blk_bs(blk) : NULL, size);
1597 }
1598
1599 bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp)
1600 {
1601 BlockDriverState *bs = blk_bs(blk);
1602
1603 if (!bs) {
1604 return false;
1605 }
1606
1607 return bdrv_op_is_blocked(bs, op, errp);
1608 }
1609
1610 void blk_op_unblock(BlockBackend *blk, BlockOpType op, Error *reason)
1611 {
1612 BlockDriverState *bs = blk_bs(blk);
1613
1614 if (bs) {
1615 bdrv_op_unblock(bs, op, reason);
1616 }
1617 }
1618
1619 void blk_op_block_all(BlockBackend *blk, Error *reason)
1620 {
1621 BlockDriverState *bs = blk_bs(blk);
1622
1623 if (bs) {
1624 bdrv_op_block_all(bs, reason);
1625 }
1626 }
1627
1628 void blk_op_unblock_all(BlockBackend *blk, Error *reason)
1629 {
1630 BlockDriverState *bs = blk_bs(blk);
1631
1632 if (bs) {
1633 bdrv_op_unblock_all(bs, reason);
1634 }
1635 }
1636
1637 AioContext *blk_get_aio_context(BlockBackend *blk)
1638 {
1639 BlockDriverState *bs = blk_bs(blk);
1640
1641 if (bs) {
1642 return bdrv_get_aio_context(bs);
1643 } else {
1644 return qemu_get_aio_context();
1645 }
1646 }
1647
1648 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb)
1649 {
1650 BlockBackendAIOCB *blk_acb = DO_UPCAST(BlockBackendAIOCB, common, acb);
1651 return blk_get_aio_context(blk_acb->blk);
1652 }
1653
1654 void blk_set_aio_context(BlockBackend *blk, AioContext *new_context)
1655 {
1656 BlockDriverState *bs = blk_bs(blk);
1657
1658 if (bs) {
1659 if (blk->public.throttle_state) {
1660 throttle_timers_detach_aio_context(&blk->public.throttle_timers);
1661 }
1662 bdrv_set_aio_context(bs, new_context);
1663 if (blk->public.throttle_state) {
1664 throttle_timers_attach_aio_context(&blk->public.throttle_timers,
1665 new_context);
1666 }
1667 }
1668 }
1669
1670 void blk_add_aio_context_notifier(BlockBackend *blk,
1671 void (*attached_aio_context)(AioContext *new_context, void *opaque),
1672 void (*detach_aio_context)(void *opaque), void *opaque)
1673 {
1674 BlockDriverState *bs = blk_bs(blk);
1675
1676 if (bs) {
1677 bdrv_add_aio_context_notifier(bs, attached_aio_context,
1678 detach_aio_context, opaque);
1679 }
1680 }
1681
1682 void blk_remove_aio_context_notifier(BlockBackend *blk,
1683 void (*attached_aio_context)(AioContext *,
1684 void *),
1685 void (*detach_aio_context)(void *),
1686 void *opaque)
1687 {
1688 BlockDriverState *bs = blk_bs(blk);
1689
1690 if (bs) {
1691 bdrv_remove_aio_context_notifier(bs, attached_aio_context,
1692 detach_aio_context, opaque);
1693 }
1694 }
1695
1696 void blk_add_remove_bs_notifier(BlockBackend *blk, Notifier *notify)
1697 {
1698 notifier_list_add(&blk->remove_bs_notifiers, notify);
1699 }
1700
1701 void blk_add_insert_bs_notifier(BlockBackend *blk, Notifier *notify)
1702 {
1703 notifier_list_add(&blk->insert_bs_notifiers, notify);
1704 }
1705
1706 void blk_io_plug(BlockBackend *blk)
1707 {
1708 BlockDriverState *bs = blk_bs(blk);
1709
1710 if (bs) {
1711 bdrv_io_plug(bs);
1712 }
1713 }
1714
1715 void blk_io_unplug(BlockBackend *blk)
1716 {
1717 BlockDriverState *bs = blk_bs(blk);
1718
1719 if (bs) {
1720 bdrv_io_unplug(bs);
1721 }
1722 }
1723
1724 BlockAcctStats *blk_get_stats(BlockBackend *blk)
1725 {
1726 return &blk->stats;
1727 }
1728
1729 void *blk_aio_get(const AIOCBInfo *aiocb_info, BlockBackend *blk,
1730 BlockCompletionFunc *cb, void *opaque)
1731 {
1732 return qemu_aio_get(aiocb_info, blk_bs(blk), cb, opaque);
1733 }
1734
1735 int coroutine_fn blk_co_pwrite_zeroes(BlockBackend *blk, int64_t offset,
1736 int count, BdrvRequestFlags flags)
1737 {
1738 return blk_co_pwritev(blk, offset, count, NULL,
1739 flags | BDRV_REQ_ZERO_WRITE);
1740 }
1741
1742 int blk_pwrite_compressed(BlockBackend *blk, int64_t offset, const void *buf,
1743 int count)
1744 {
1745 return blk_prw(blk, offset, (void *) buf, count, blk_write_entry,
1746 BDRV_REQ_WRITE_COMPRESSED);
1747 }
1748
1749 int blk_truncate(BlockBackend *blk, int64_t offset)
1750 {
1751 if (!blk_is_available(blk)) {
1752 return -ENOMEDIUM;
1753 }
1754
1755 return bdrv_truncate(blk->root, offset);
1756 }
1757
1758 static void blk_pdiscard_entry(void *opaque)
1759 {
1760 BlkRwCo *rwco = opaque;
1761 rwco->ret = blk_co_pdiscard(rwco->blk, rwco->offset, rwco->qiov->size);
1762 }
1763
1764 int blk_pdiscard(BlockBackend *blk, int64_t offset, int count)
1765 {
1766 return blk_prw(blk, offset, NULL, count, blk_pdiscard_entry, 0);
1767 }
1768
1769 int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
1770 int64_t pos, int size)
1771 {
1772 int ret;
1773
1774 if (!blk_is_available(blk)) {
1775 return -ENOMEDIUM;
1776 }
1777
1778 ret = bdrv_save_vmstate(blk_bs(blk), buf, pos, size);
1779 if (ret < 0) {
1780 return ret;
1781 }
1782
1783 if (ret == size && !blk->enable_write_cache) {
1784 ret = bdrv_flush(blk_bs(blk));
1785 }
1786
1787 return ret < 0 ? ret : size;
1788 }
1789
1790 int blk_load_vmstate(BlockBackend *blk, uint8_t *buf, int64_t pos, int size)
1791 {
1792 if (!blk_is_available(blk)) {
1793 return -ENOMEDIUM;
1794 }
1795
1796 return bdrv_load_vmstate(blk_bs(blk), buf, pos, size);
1797 }
1798
1799 int blk_probe_blocksizes(BlockBackend *blk, BlockSizes *bsz)
1800 {
1801 if (!blk_is_available(blk)) {
1802 return -ENOMEDIUM;
1803 }
1804
1805 return bdrv_probe_blocksizes(blk_bs(blk), bsz);
1806 }
1807
1808 int blk_probe_geometry(BlockBackend *blk, HDGeometry *geo)
1809 {
1810 if (!blk_is_available(blk)) {
1811 return -ENOMEDIUM;
1812 }
1813
1814 return bdrv_probe_geometry(blk_bs(blk), geo);
1815 }
1816
1817 /*
1818 * Updates the BlockBackendRootState object with data from the currently
1819 * attached BlockDriverState.
1820 */
1821 void blk_update_root_state(BlockBackend *blk)
1822 {
1823 assert(blk->root);
1824
1825 blk->root_state.open_flags = blk->root->bs->open_flags;
1826 blk->root_state.read_only = blk->root->bs->read_only;
1827 blk->root_state.detect_zeroes = blk->root->bs->detect_zeroes;
1828 }
1829
1830 /*
1831 * Returns the detect-zeroes setting to be used for bdrv_open() of a
1832 * BlockDriverState which is supposed to inherit the root state.
1833 */
1834 bool blk_get_detect_zeroes_from_root_state(BlockBackend *blk)
1835 {
1836 return blk->root_state.detect_zeroes;
1837 }
1838
1839 /*
1840 * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
1841 * supposed to inherit the root state.
1842 */
1843 int blk_get_open_flags_from_root_state(BlockBackend *blk)
1844 {
1845 int bs_flags;
1846
1847 bs_flags = blk->root_state.read_only ? 0 : BDRV_O_RDWR;
1848 bs_flags |= blk->root_state.open_flags & ~BDRV_O_RDWR;
1849
1850 return bs_flags;
1851 }
1852
1853 BlockBackendRootState *blk_get_root_state(BlockBackend *blk)
1854 {
1855 return &blk->root_state;
1856 }
1857
1858 int blk_commit_all(void)
1859 {
1860 BlockBackend *blk = NULL;
1861
1862 while ((blk = blk_all_next(blk)) != NULL) {
1863 AioContext *aio_context = blk_get_aio_context(blk);
1864
1865 aio_context_acquire(aio_context);
1866 if (blk_is_inserted(blk) && blk->root->bs->backing) {
1867 int ret = bdrv_commit(blk->root->bs);
1868 if (ret < 0) {
1869 aio_context_release(aio_context);
1870 return ret;
1871 }
1872 }
1873 aio_context_release(aio_context);
1874 }
1875 return 0;
1876 }
1877
1878
1879 /* throttling disk I/O limits */
1880 void blk_set_io_limits(BlockBackend *blk, ThrottleConfig *cfg)
1881 {
1882 throttle_group_config(blk, cfg);
1883 }
1884
1885 void blk_io_limits_disable(BlockBackend *blk)
1886 {
1887 assert(blk->public.throttle_state);
1888 bdrv_drained_begin(blk_bs(blk));
1889 throttle_group_unregister_blk(blk);
1890 bdrv_drained_end(blk_bs(blk));
1891 }
1892
1893 /* should be called before blk_set_io_limits if a limit is set */
1894 void blk_io_limits_enable(BlockBackend *blk, const char *group)
1895 {
1896 assert(!blk->public.throttle_state);
1897 throttle_group_register_blk(blk, group);
1898 }
1899
1900 void blk_io_limits_update_group(BlockBackend *blk, const char *group)
1901 {
1902 /* this BB is not part of any group */
1903 if (!blk->public.throttle_state) {
1904 return;
1905 }
1906
1907 /* this BB is a part of the same group than the one we want */
1908 if (!g_strcmp0(throttle_group_get_name(blk), group)) {
1909 return;
1910 }
1911
1912 /* need to change the group this bs belong to */
1913 blk_io_limits_disable(blk);
1914 blk_io_limits_enable(blk, group);
1915 }
1916
1917 static void blk_root_drained_begin(BdrvChild *child)
1918 {
1919 BlockBackend *blk = child->opaque;
1920
1921 if (++blk->quiesce_counter == 1) {
1922 if (blk->dev_ops && blk->dev_ops->drained_begin) {
1923 blk->dev_ops->drained_begin(blk->dev_opaque);
1924 }
1925 }
1926
1927 /* Note that blk->root may not be accessible here yet if we are just
1928 * attaching to a BlockDriverState that is drained. Use child instead. */
1929
1930 if (blk->public.io_limits_disabled++ == 0) {
1931 throttle_group_restart_blk(blk);
1932 }
1933 }
1934
1935 static void blk_root_drained_end(BdrvChild *child)
1936 {
1937 BlockBackend *blk = child->opaque;
1938 assert(blk->quiesce_counter);
1939
1940 assert(blk->public.io_limits_disabled);
1941 --blk->public.io_limits_disabled;
1942
1943 if (--blk->quiesce_counter == 0) {
1944 if (blk->dev_ops && blk->dev_ops->drained_end) {
1945 blk->dev_ops->drained_end(blk->dev_opaque);
1946 }
1947 }
1948 }