4 * Copyright (C) 2014-2016 Red Hat, Inc.
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.
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"
23 /* Number of coroutines to reserve per attached device model */
24 #define COROUTINE_POOL_RESERVATION 64
26 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
28 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb);
34 DriveInfo *legacy_dinfo; /* null unless created by drive_new() */
35 QTAILQ_ENTRY(BlockBackend) link; /* for block_backends */
36 QTAILQ_ENTRY(BlockBackend) monitor_link; /* for monitor_block_backends */
37 BlockBackendPublic public;
39 void *dev; /* attached device model, if any */
40 /* TODO change to DeviceState when all users are qdevified */
41 const BlockDevOps *dev_ops;
44 /* the block size for which the guest device expects atomicity */
47 /* If the BDS tree is removed, some of its options are stored here (which
48 * can be used to restore those options in the new BDS on insert) */
49 BlockBackendRootState root_state;
51 bool enable_write_cache;
53 /* I/O stats (display with "info blockstats"). */
56 BlockdevOnError on_read_error, on_write_error;
57 bool iostatus_enabled;
58 BlockDeviceIoStatus iostatus;
60 bool allow_write_beyond_eof;
62 NotifierList remove_bs_notifiers, insert_bs_notifiers;
65 typedef struct BlockBackendAIOCB {
72 static const AIOCBInfo block_backend_aiocb_info = {
73 .get_aio_context = blk_aiocb_get_aio_context,
74 .aiocb_size = sizeof(BlockBackendAIOCB),
77 static void drive_info_del(DriveInfo *dinfo);
78 static BlockBackend *bdrv_first_blk(BlockDriverState *bs);
80 /* All BlockBackends */
81 static QTAILQ_HEAD(, BlockBackend) block_backends =
82 QTAILQ_HEAD_INITIALIZER(block_backends);
84 /* All BlockBackends referenced by the monitor and which are iterated through by
86 static QTAILQ_HEAD(, BlockBackend) monitor_block_backends =
87 QTAILQ_HEAD_INITIALIZER(monitor_block_backends);
89 static void blk_root_inherit_options(int *child_flags, QDict *child_options,
90 int parent_flags, QDict *parent_options)
92 /* We're not supposed to call this function for root nodes */
95 static void blk_root_drained_begin(BdrvChild *child);
96 static void blk_root_drained_end(BdrvChild *child);
98 static void blk_root_change_media(BdrvChild *child, bool load);
99 static void blk_root_resize(BdrvChild *child);
101 static const char *blk_root_get_name(BdrvChild *child)
103 return blk_name(child->opaque);
106 static const BdrvChildRole child_root = {
107 .inherit_options = blk_root_inherit_options,
109 .change_media = blk_root_change_media,
110 .resize = blk_root_resize,
111 .get_name = blk_root_get_name,
113 .drained_begin = blk_root_drained_begin,
114 .drained_end = blk_root_drained_end,
118 * Create a new BlockBackend with a reference count of one.
119 * Store an error through @errp on failure, unless it's null.
120 * Return the new BlockBackend on success, null on failure.
122 BlockBackend *blk_new(Error **errp)
126 blk = g_new0(BlockBackend, 1);
128 qemu_co_queue_init(&blk->public.throttled_reqs[0]);
129 qemu_co_queue_init(&blk->public.throttled_reqs[1]);
131 notifier_list_init(&blk->remove_bs_notifiers);
132 notifier_list_init(&blk->insert_bs_notifiers);
134 QTAILQ_INSERT_TAIL(&block_backends, blk, link);
139 * Create a new BlockBackend with a new BlockDriverState attached.
140 * Otherwise just like blk_new(), which see.
142 BlockBackend *blk_new_with_bs(Error **errp)
145 BlockDriverState *bs;
152 bs = bdrv_new_root();
153 blk->root = bdrv_root_attach_child(bs, "root", &child_root);
154 blk->root->opaque = blk;
160 * Calls blk_new_with_bs() and then calls bdrv_open() on the BlockDriverState.
162 * Just as with bdrv_open(), after having called this function the reference to
163 * @options belongs to the block layer (even on failure).
165 * TODO: Remove @filename and @flags; it should be possible to specify a whole
166 * BDS tree just by specifying the @options QDict (or @reference,
167 * alternatively). At the time of adding this function, this is not possible,
168 * though, so callers of this function have to be able to specify @filename and
171 BlockBackend *blk_new_open(const char *filename, const char *reference,
172 QDict *options, int flags, Error **errp)
177 blk = blk_new_with_bs(errp);
183 ret = bdrv_open(&blk->root->bs, filename, reference, options, flags, errp);
189 blk_set_enable_write_cache(blk, true);
194 static void blk_delete(BlockBackend *blk)
196 assert(!blk->refcnt);
202 assert(QLIST_EMPTY(&blk->remove_bs_notifiers.notifiers));
203 assert(QLIST_EMPTY(&blk->insert_bs_notifiers.notifiers));
204 QTAILQ_REMOVE(&block_backends, blk, link);
205 drive_info_del(blk->legacy_dinfo);
206 block_acct_cleanup(&blk->stats);
210 static void drive_info_del(DriveInfo *dinfo)
215 qemu_opts_del(dinfo->opts);
216 g_free(dinfo->serial);
220 int blk_get_refcnt(BlockBackend *blk)
222 return blk ? blk->refcnt : 0;
226 * Increment @blk's reference count.
227 * @blk must not be null.
229 void blk_ref(BlockBackend *blk)
235 * Decrement @blk's reference count.
236 * If this drops it to zero, destroy @blk.
237 * For convenience, do nothing if @blk is null.
239 void blk_unref(BlockBackend *blk)
242 assert(blk->refcnt > 0);
243 if (!--blk->refcnt) {
250 * Behaves similarly to blk_next() but iterates over all BlockBackends, even the
251 * ones which are hidden (i.e. are not referenced by the monitor).
253 static BlockBackend *blk_all_next(BlockBackend *blk)
255 return blk ? QTAILQ_NEXT(blk, link)
256 : QTAILQ_FIRST(&block_backends);
259 void blk_remove_all_bs(void)
261 BlockBackend *blk = NULL;
263 while ((blk = blk_all_next(blk)) != NULL) {
264 AioContext *ctx = blk_get_aio_context(blk);
266 aio_context_acquire(ctx);
270 aio_context_release(ctx);
275 * Return the monitor-owned BlockBackend after @blk.
276 * If @blk is null, return the first one.
277 * Else, return @blk's next sibling, which may be null.
279 * To iterate over all BlockBackends, do
280 * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
284 BlockBackend *blk_next(BlockBackend *blk)
286 return blk ? QTAILQ_NEXT(blk, monitor_link)
287 : QTAILQ_FIRST(&monitor_block_backends);
290 struct BdrvNextIterator {
292 BDRV_NEXT_BACKEND_ROOTS,
293 BDRV_NEXT_MONITOR_OWNED,
296 BlockDriverState *bs;
299 /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by
300 * the monitor or attached to a BlockBackend */
301 BdrvNextIterator *bdrv_next(BdrvNextIterator *it, BlockDriverState **bs)
304 it = g_new(BdrvNextIterator, 1);
305 *it = (BdrvNextIterator) {
306 .phase = BDRV_NEXT_BACKEND_ROOTS,
310 /* First, return all root nodes of BlockBackends. In order to avoid
311 * returning a BDS twice when multiple BBs refer to it, we only return it
312 * if the BB is the first one in the parent list of the BDS. */
313 if (it->phase == BDRV_NEXT_BACKEND_ROOTS) {
315 it->blk = blk_all_next(it->blk);
316 *bs = it->blk ? blk_bs(it->blk) : NULL;
317 } while (it->blk && (*bs == NULL || bdrv_first_blk(*bs) != it->blk));
322 it->phase = BDRV_NEXT_MONITOR_OWNED;
325 /* Then return the monitor-owned BDSes without a BB attached. Ignore all
326 * BDSes that are attached to a BlockBackend here; they have been handled
327 * by the above block already */
329 it->bs = bdrv_next_monitor_owned(it->bs);
331 } while (*bs && bdrv_has_blk(*bs));
333 return *bs ? it : NULL;
337 * Add a BlockBackend into the list of backends referenced by the monitor, with
338 * the given @name acting as the handle for the monitor.
339 * Strictly for use by blockdev.c.
341 * @name must not be null or empty.
343 * Returns true on success and false on failure. In the latter case, an Error
344 * object is returned through @errp.
346 bool monitor_add_blk(BlockBackend *blk, const char *name, Error **errp)
349 assert(name && name[0]);
351 if (!id_wellformed(name)) {
352 error_setg(errp, "Invalid device name");
355 if (blk_by_name(name)) {
356 error_setg(errp, "Device with id '%s' already exists", name);
359 if (bdrv_find_node(name)) {
361 "Device name '%s' conflicts with an existing node name",
366 blk->name = g_strdup(name);
367 QTAILQ_INSERT_TAIL(&monitor_block_backends, blk, monitor_link);
372 * Remove a BlockBackend from the list of backends referenced by the monitor.
373 * Strictly for use by blockdev.c.
375 void monitor_remove_blk(BlockBackend *blk)
381 QTAILQ_REMOVE(&monitor_block_backends, blk, monitor_link);
387 * Return @blk's name, a non-null string.
388 * Returns an empty string iff @blk is not referenced by the monitor.
390 const char *blk_name(BlockBackend *blk)
392 return blk->name ?: "";
396 * Return the BlockBackend with name @name if it exists, else null.
397 * @name must not be null.
399 BlockBackend *blk_by_name(const char *name)
401 BlockBackend *blk = NULL;
404 while ((blk = blk_next(blk)) != NULL) {
405 if (!strcmp(name, blk->name)) {
413 * Return the BlockDriverState attached to @blk if any, else null.
415 BlockDriverState *blk_bs(BlockBackend *blk)
417 return blk->root ? blk->root->bs : NULL;
420 static BlockBackend *bdrv_first_blk(BlockDriverState *bs)
423 QLIST_FOREACH(child, &bs->parents, next_parent) {
424 if (child->role == &child_root) {
426 return child->opaque;
435 * Returns true if @bs has an associated BlockBackend.
437 bool bdrv_has_blk(BlockDriverState *bs)
439 return bdrv_first_blk(bs) != NULL;
443 * Return @blk's DriveInfo if any, else null.
445 DriveInfo *blk_legacy_dinfo(BlockBackend *blk)
447 return blk->legacy_dinfo;
451 * Set @blk's DriveInfo to @dinfo, and return it.
452 * @blk must not have a DriveInfo set already.
453 * No other BlockBackend may have the same DriveInfo set.
455 DriveInfo *blk_set_legacy_dinfo(BlockBackend *blk, DriveInfo *dinfo)
457 assert(!blk->legacy_dinfo);
458 return blk->legacy_dinfo = dinfo;
462 * Return the BlockBackend with DriveInfo @dinfo.
465 BlockBackend *blk_by_legacy_dinfo(DriveInfo *dinfo)
467 BlockBackend *blk = NULL;
469 while ((blk = blk_next(blk)) != NULL) {
470 if (blk->legacy_dinfo == dinfo) {
478 * Returns a pointer to the publicly accessible fields of @blk.
480 BlockBackendPublic *blk_get_public(BlockBackend *blk)
486 * Returns a BlockBackend given the associated @public fields.
488 BlockBackend *blk_by_public(BlockBackendPublic *public)
490 return container_of(public, BlockBackend, public);
494 * Disassociates the currently associated BlockDriverState from @blk.
496 void blk_remove_bs(BlockBackend *blk)
498 assert(blk->root->bs->blk == blk);
500 notifier_list_notify(&blk->remove_bs_notifiers, blk);
501 if (blk->public.throttle_state) {
502 throttle_timers_detach_aio_context(&blk->public.throttle_timers);
505 blk_update_root_state(blk);
507 blk->root->bs->blk = NULL;
508 bdrv_root_unref_child(blk->root);
513 * Associates a new BlockDriverState with @blk.
515 void blk_insert_bs(BlockBackend *blk, BlockDriverState *bs)
517 assert(!blk->root && !bs->blk);
519 blk->root = bdrv_root_attach_child(bs, "root", &child_root);
520 blk->root->opaque = blk;
523 notifier_list_notify(&blk->insert_bs_notifiers, blk);
524 if (blk->public.throttle_state) {
525 throttle_timers_attach_aio_context(
526 &blk->public.throttle_timers, bdrv_get_aio_context(bs));
531 * Attach device model @dev to @blk.
532 * Return 0 on success, -EBUSY when a device model is attached already.
534 int blk_attach_dev(BlockBackend *blk, void *dev)
535 /* TODO change to DeviceState *dev when all users are qdevified */
542 blk_iostatus_reset(blk);
547 * Attach device model @dev to @blk.
548 * @blk must not have a device model attached already.
549 * TODO qdevified devices don't use this, remove when devices are qdevified
551 void blk_attach_dev_nofail(BlockBackend *blk, void *dev)
553 if (blk_attach_dev(blk, dev) < 0) {
559 * Detach device model @dev from @blk.
560 * @dev must be currently attached to @blk.
562 void blk_detach_dev(BlockBackend *blk, void *dev)
563 /* TODO change to DeviceState *dev when all users are qdevified */
565 assert(blk->dev == dev);
568 blk->dev_opaque = NULL;
569 blk->guest_block_size = 512;
574 * Return the device model attached to @blk if any, else null.
576 void *blk_get_attached_dev(BlockBackend *blk)
577 /* TODO change to return DeviceState * when all users are qdevified */
583 * Set @blk's device model callbacks to @ops.
584 * @opaque is the opaque argument to pass to the callbacks.
585 * This is for use by device models.
587 void blk_set_dev_ops(BlockBackend *blk, const BlockDevOps *ops,
591 blk->dev_opaque = opaque;
595 * Notify @blk's attached device model of media change.
596 * If @load is true, notify of media load.
597 * Else, notify of media eject.
598 * Also send DEVICE_TRAY_MOVED events as appropriate.
600 void blk_dev_change_media_cb(BlockBackend *blk, bool load)
602 if (blk->dev_ops && blk->dev_ops->change_media_cb) {
603 bool tray_was_open, tray_is_open;
605 tray_was_open = blk_dev_is_tray_open(blk);
606 blk->dev_ops->change_media_cb(blk->dev_opaque, load);
607 tray_is_open = blk_dev_is_tray_open(blk);
609 if (tray_was_open != tray_is_open) {
610 qapi_event_send_device_tray_moved(blk_name(blk), tray_is_open,
616 static void blk_root_change_media(BdrvChild *child, bool load)
618 blk_dev_change_media_cb(child->opaque, load);
622 * Does @blk's attached device model have removable media?
623 * %true if no device model is attached.
625 bool blk_dev_has_removable_media(BlockBackend *blk)
627 return !blk->dev || (blk->dev_ops && blk->dev_ops->change_media_cb);
631 * Does @blk's attached device model have a tray?
633 bool blk_dev_has_tray(BlockBackend *blk)
635 return blk->dev_ops && blk->dev_ops->is_tray_open;
639 * Notify @blk's attached device model of a media eject request.
640 * If @force is true, the medium is about to be yanked out forcefully.
642 void blk_dev_eject_request(BlockBackend *blk, bool force)
644 if (blk->dev_ops && blk->dev_ops->eject_request_cb) {
645 blk->dev_ops->eject_request_cb(blk->dev_opaque, force);
650 * Does @blk's attached device model have a tray, and is it open?
652 bool blk_dev_is_tray_open(BlockBackend *blk)
654 if (blk_dev_has_tray(blk)) {
655 return blk->dev_ops->is_tray_open(blk->dev_opaque);
661 * Does @blk's attached device model have the medium locked?
662 * %false if the device model has no such lock.
664 bool blk_dev_is_medium_locked(BlockBackend *blk)
666 if (blk->dev_ops && blk->dev_ops->is_medium_locked) {
667 return blk->dev_ops->is_medium_locked(blk->dev_opaque);
673 * Notify @blk's attached device model of a backend size change.
675 static void blk_root_resize(BdrvChild *child)
677 BlockBackend *blk = child->opaque;
679 if (blk->dev_ops && blk->dev_ops->resize_cb) {
680 blk->dev_ops->resize_cb(blk->dev_opaque);
684 void blk_iostatus_enable(BlockBackend *blk)
686 blk->iostatus_enabled = true;
687 blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
690 /* The I/O status is only enabled if the drive explicitly
691 * enables it _and_ the VM is configured to stop on errors */
692 bool blk_iostatus_is_enabled(const BlockBackend *blk)
694 return (blk->iostatus_enabled &&
695 (blk->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
696 blk->on_write_error == BLOCKDEV_ON_ERROR_STOP ||
697 blk->on_read_error == BLOCKDEV_ON_ERROR_STOP));
700 BlockDeviceIoStatus blk_iostatus(const BlockBackend *blk)
702 return blk->iostatus;
705 void blk_iostatus_disable(BlockBackend *blk)
707 blk->iostatus_enabled = false;
710 void blk_iostatus_reset(BlockBackend *blk)
712 if (blk_iostatus_is_enabled(blk)) {
713 BlockDriverState *bs = blk_bs(blk);
714 blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
716 block_job_iostatus_reset(bs->job);
721 void blk_iostatus_set_err(BlockBackend *blk, int error)
723 assert(blk_iostatus_is_enabled(blk));
724 if (blk->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
725 blk->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
726 BLOCK_DEVICE_IO_STATUS_FAILED;
730 void blk_set_allow_write_beyond_eof(BlockBackend *blk, bool allow)
732 blk->allow_write_beyond_eof = allow;
735 static int blk_check_byte_request(BlockBackend *blk, int64_t offset,
740 if (size > INT_MAX) {
744 if (!blk_is_available(blk)) {
752 if (!blk->allow_write_beyond_eof) {
753 len = blk_getlength(blk);
758 if (offset > len || len - offset < size) {
766 static int blk_check_request(BlockBackend *blk, int64_t sector_num,
769 if (sector_num < 0 || sector_num > INT64_MAX / BDRV_SECTOR_SIZE) {
773 if (nb_sectors < 0 || nb_sectors > INT_MAX / BDRV_SECTOR_SIZE) {
777 return blk_check_byte_request(blk, sector_num * BDRV_SECTOR_SIZE,
778 nb_sectors * BDRV_SECTOR_SIZE);
781 static int coroutine_fn blk_co_preadv(BlockBackend *blk, int64_t offset,
782 unsigned int bytes, QEMUIOVector *qiov,
783 BdrvRequestFlags flags)
785 int ret = blk_check_byte_request(blk, offset, bytes);
790 /* throttling disk I/O */
791 if (blk->public.throttle_state) {
792 throttle_group_co_io_limits_intercept(blk, bytes, false);
795 return bdrv_co_preadv(blk_bs(blk), offset, bytes, qiov, flags);
798 static int coroutine_fn blk_co_pwritev(BlockBackend *blk, int64_t offset,
799 unsigned int bytes, QEMUIOVector *qiov,
800 BdrvRequestFlags flags)
804 ret = blk_check_byte_request(blk, offset, bytes);
809 /* throttling disk I/O */
810 if (blk->public.throttle_state) {
811 throttle_group_co_io_limits_intercept(blk, bytes, true);
814 if (!blk->enable_write_cache) {
815 flags |= BDRV_REQ_FUA;
818 return bdrv_co_pwritev(blk_bs(blk), offset, bytes, qiov, flags);
821 typedef struct BlkRwCo {
826 BdrvRequestFlags flags;
829 static void blk_read_entry(void *opaque)
831 BlkRwCo *rwco = opaque;
833 rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, rwco->qiov->size,
834 rwco->qiov, rwco->flags);
837 static void blk_write_entry(void *opaque)
839 BlkRwCo *rwco = opaque;
841 rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, rwco->qiov->size,
842 rwco->qiov, rwco->flags);
845 static int blk_prw(BlockBackend *blk, int64_t offset, uint8_t *buf,
846 int64_t bytes, CoroutineEntry co_entry,
847 BdrvRequestFlags flags)
849 AioContext *aio_context;
855 iov = (struct iovec) {
859 qemu_iovec_init_external(&qiov, &iov, 1);
869 co = qemu_coroutine_create(co_entry);
870 qemu_coroutine_enter(co, &rwco);
872 aio_context = blk_get_aio_context(blk);
873 while (rwco.ret == NOT_DONE) {
874 aio_poll(aio_context, true);
880 int blk_pread_unthrottled(BlockBackend *blk, int64_t offset, uint8_t *buf,
885 ret = blk_check_byte_request(blk, offset, count);
890 blk_root_drained_begin(blk->root);
891 ret = blk_pread(blk, offset, buf, count);
892 blk_root_drained_end(blk->root);
896 int blk_write_zeroes(BlockBackend *blk, int64_t offset,
897 int count, BdrvRequestFlags flags)
899 return blk_prw(blk, offset, NULL, count, blk_write_entry,
900 flags | BDRV_REQ_ZERO_WRITE);
903 static void error_callback_bh(void *opaque)
905 struct BlockBackendAIOCB *acb = opaque;
906 qemu_bh_delete(acb->bh);
907 acb->common.cb(acb->common.opaque, acb->ret);
911 BlockAIOCB *blk_abort_aio_request(BlockBackend *blk,
912 BlockCompletionFunc *cb,
913 void *opaque, int ret)
915 struct BlockBackendAIOCB *acb;
918 acb = blk_aio_get(&block_backend_aiocb_info, blk, cb, opaque);
922 bh = aio_bh_new(blk_get_aio_context(blk), error_callback_bh, acb);
924 qemu_bh_schedule(bh);
929 typedef struct BlkAioEmAIOCB {
937 static const AIOCBInfo blk_aio_em_aiocb_info = {
938 .aiocb_size = sizeof(BlkAioEmAIOCB),
941 static void blk_aio_complete(BlkAioEmAIOCB *acb)
944 assert(acb->has_returned);
945 qemu_bh_delete(acb->bh);
947 if (acb->has_returned) {
948 acb->common.cb(acb->common.opaque, acb->rwco.ret);
953 static void blk_aio_complete_bh(void *opaque)
955 blk_aio_complete(opaque);
958 static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset, int bytes,
959 QEMUIOVector *qiov, CoroutineEntry co_entry,
960 BdrvRequestFlags flags,
961 BlockCompletionFunc *cb, void *opaque)
966 acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque);
967 acb->rwco = (BlkRwCo) {
976 acb->has_returned = false;
978 co = qemu_coroutine_create(co_entry);
979 qemu_coroutine_enter(co, acb);
981 acb->has_returned = true;
982 if (acb->rwco.ret != NOT_DONE) {
983 acb->bh = aio_bh_new(blk_get_aio_context(blk), blk_aio_complete_bh, acb);
984 qemu_bh_schedule(acb->bh);
990 static void blk_aio_read_entry(void *opaque)
992 BlkAioEmAIOCB *acb = opaque;
993 BlkRwCo *rwco = &acb->rwco;
995 assert(rwco->qiov->size == acb->bytes);
996 rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, acb->bytes,
997 rwco->qiov, rwco->flags);
998 blk_aio_complete(acb);
1001 static void blk_aio_write_entry(void *opaque)
1003 BlkAioEmAIOCB *acb = opaque;
1004 BlkRwCo *rwco = &acb->rwco;
1006 assert(!rwco->qiov || rwco->qiov->size == acb->bytes);
1007 rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, acb->bytes,
1008 rwco->qiov, rwco->flags);
1009 blk_aio_complete(acb);
1012 BlockAIOCB *blk_aio_write_zeroes(BlockBackend *blk, int64_t offset,
1013 int count, BdrvRequestFlags flags,
1014 BlockCompletionFunc *cb, void *opaque)
1016 return blk_aio_prwv(blk, offset, count, NULL, blk_aio_write_entry,
1017 flags | BDRV_REQ_ZERO_WRITE, cb, opaque);
1020 int blk_pread(BlockBackend *blk, int64_t offset, void *buf, int count)
1022 int ret = blk_prw(blk, offset, buf, count, blk_read_entry, 0);
1029 int blk_pwrite(BlockBackend *blk, int64_t offset, const void *buf, int count,
1030 BdrvRequestFlags flags)
1032 int ret = blk_prw(blk, offset, (void *) buf, count, blk_write_entry,
1040 int64_t blk_getlength(BlockBackend *blk)
1042 if (!blk_is_available(blk)) {
1046 return bdrv_getlength(blk_bs(blk));
1049 void blk_get_geometry(BlockBackend *blk, uint64_t *nb_sectors_ptr)
1052 *nb_sectors_ptr = 0;
1054 bdrv_get_geometry(blk_bs(blk), nb_sectors_ptr);
1058 int64_t blk_nb_sectors(BlockBackend *blk)
1060 if (!blk_is_available(blk)) {
1064 return bdrv_nb_sectors(blk_bs(blk));
1067 BlockAIOCB *blk_aio_preadv(BlockBackend *blk, int64_t offset,
1068 QEMUIOVector *qiov, BdrvRequestFlags flags,
1069 BlockCompletionFunc *cb, void *opaque)
1071 return blk_aio_prwv(blk, offset, qiov->size, qiov,
1072 blk_aio_read_entry, flags, cb, opaque);
1075 BlockAIOCB *blk_aio_pwritev(BlockBackend *blk, int64_t offset,
1076 QEMUIOVector *qiov, BdrvRequestFlags flags,
1077 BlockCompletionFunc *cb, void *opaque)
1079 return blk_aio_prwv(blk, offset, qiov->size, qiov,
1080 blk_aio_write_entry, flags, cb, opaque);
1083 BlockAIOCB *blk_aio_flush(BlockBackend *blk,
1084 BlockCompletionFunc *cb, void *opaque)
1086 if (!blk_is_available(blk)) {
1087 return blk_abort_aio_request(blk, cb, opaque, -ENOMEDIUM);
1090 return bdrv_aio_flush(blk_bs(blk), cb, opaque);
1093 BlockAIOCB *blk_aio_discard(BlockBackend *blk,
1094 int64_t sector_num, int nb_sectors,
1095 BlockCompletionFunc *cb, void *opaque)
1097 int ret = blk_check_request(blk, sector_num, nb_sectors);
1099 return blk_abort_aio_request(blk, cb, opaque, ret);
1102 return bdrv_aio_discard(blk_bs(blk), sector_num, nb_sectors, cb, opaque);
1105 void blk_aio_cancel(BlockAIOCB *acb)
1107 bdrv_aio_cancel(acb);
1110 void blk_aio_cancel_async(BlockAIOCB *acb)
1112 bdrv_aio_cancel_async(acb);
1115 int blk_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
1117 if (!blk_is_available(blk)) {
1121 return bdrv_ioctl(blk_bs(blk), req, buf);
1124 BlockAIOCB *blk_aio_ioctl(BlockBackend *blk, unsigned long int req, void *buf,
1125 BlockCompletionFunc *cb, void *opaque)
1127 if (!blk_is_available(blk)) {
1128 return blk_abort_aio_request(blk, cb, opaque, -ENOMEDIUM);
1131 return bdrv_aio_ioctl(blk_bs(blk), req, buf, cb, opaque);
1134 int blk_co_discard(BlockBackend *blk, int64_t sector_num, int nb_sectors)
1136 int ret = blk_check_request(blk, sector_num, nb_sectors);
1141 return bdrv_co_discard(blk_bs(blk), sector_num, nb_sectors);
1144 int blk_co_flush(BlockBackend *blk)
1146 if (!blk_is_available(blk)) {
1150 return bdrv_co_flush(blk_bs(blk));
1153 int blk_flush(BlockBackend *blk)
1155 if (!blk_is_available(blk)) {
1159 return bdrv_flush(blk_bs(blk));
1162 void blk_drain(BlockBackend *blk)
1165 bdrv_drain(blk_bs(blk));
1169 void blk_drain_all(void)
1174 void blk_set_on_error(BlockBackend *blk, BlockdevOnError on_read_error,
1175 BlockdevOnError on_write_error)
1177 blk->on_read_error = on_read_error;
1178 blk->on_write_error = on_write_error;
1181 BlockdevOnError blk_get_on_error(BlockBackend *blk, bool is_read)
1183 return is_read ? blk->on_read_error : blk->on_write_error;
1186 BlockErrorAction blk_get_error_action(BlockBackend *blk, bool is_read,
1189 BlockdevOnError on_err = blk_get_on_error(blk, is_read);
1192 case BLOCKDEV_ON_ERROR_ENOSPC:
1193 return (error == ENOSPC) ?
1194 BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
1195 case BLOCKDEV_ON_ERROR_STOP:
1196 return BLOCK_ERROR_ACTION_STOP;
1197 case BLOCKDEV_ON_ERROR_REPORT:
1198 return BLOCK_ERROR_ACTION_REPORT;
1199 case BLOCKDEV_ON_ERROR_IGNORE:
1200 return BLOCK_ERROR_ACTION_IGNORE;
1206 static void send_qmp_error_event(BlockBackend *blk,
1207 BlockErrorAction action,
1208 bool is_read, int error)
1210 IoOperationType optype;
1212 optype = is_read ? IO_OPERATION_TYPE_READ : IO_OPERATION_TYPE_WRITE;
1213 qapi_event_send_block_io_error(blk_name(blk), optype, action,
1214 blk_iostatus_is_enabled(blk),
1215 error == ENOSPC, strerror(error),
1219 /* This is done by device models because, while the block layer knows
1220 * about the error, it does not know whether an operation comes from
1221 * the device or the block layer (from a job, for example).
1223 void blk_error_action(BlockBackend *blk, BlockErrorAction action,
1224 bool is_read, int error)
1228 if (action == BLOCK_ERROR_ACTION_STOP) {
1229 /* First set the iostatus, so that "info block" returns an iostatus
1230 * that matches the events raised so far (an additional error iostatus
1231 * is fine, but not a lost one).
1233 blk_iostatus_set_err(blk, error);
1235 /* Then raise the request to stop the VM and the event.
1236 * qemu_system_vmstop_request_prepare has two effects. First,
1237 * it ensures that the STOP event always comes after the
1238 * BLOCK_IO_ERROR event. Second, it ensures that even if management
1239 * can observe the STOP event and do a "cont" before the STOP
1240 * event is issued, the VM will not stop. In this case, vm_start()
1241 * also ensures that the STOP/RESUME pair of events is emitted.
1243 qemu_system_vmstop_request_prepare();
1244 send_qmp_error_event(blk, action, is_read, error);
1245 qemu_system_vmstop_request(RUN_STATE_IO_ERROR);
1247 send_qmp_error_event(blk, action, is_read, error);
1251 int blk_is_read_only(BlockBackend *blk)
1253 BlockDriverState *bs = blk_bs(blk);
1256 return bdrv_is_read_only(bs);
1258 return blk->root_state.read_only;
1262 int blk_is_sg(BlockBackend *blk)
1264 BlockDriverState *bs = blk_bs(blk);
1270 return bdrv_is_sg(bs);
1273 int blk_enable_write_cache(BlockBackend *blk)
1275 return blk->enable_write_cache;
1278 void blk_set_enable_write_cache(BlockBackend *blk, bool wce)
1280 blk->enable_write_cache = wce;
1283 void blk_invalidate_cache(BlockBackend *blk, Error **errp)
1285 BlockDriverState *bs = blk_bs(blk);
1288 error_setg(errp, "Device '%s' has no medium", blk->name);
1292 bdrv_invalidate_cache(bs, errp);
1295 bool blk_is_inserted(BlockBackend *blk)
1297 BlockDriverState *bs = blk_bs(blk);
1299 return bs && bdrv_is_inserted(bs);
1302 bool blk_is_available(BlockBackend *blk)
1304 return blk_is_inserted(blk) && !blk_dev_is_tray_open(blk);
1307 void blk_lock_medium(BlockBackend *blk, bool locked)
1309 BlockDriverState *bs = blk_bs(blk);
1312 bdrv_lock_medium(bs, locked);
1316 void blk_eject(BlockBackend *blk, bool eject_flag)
1318 BlockDriverState *bs = blk_bs(blk);
1321 bdrv_eject(bs, eject_flag);
1325 int blk_get_flags(BlockBackend *blk)
1327 BlockDriverState *bs = blk_bs(blk);
1330 return bdrv_get_flags(bs);
1332 return blk->root_state.open_flags;
1336 int blk_get_max_transfer_length(BlockBackend *blk)
1338 BlockDriverState *bs = blk_bs(blk);
1341 return bs->bl.max_transfer_length;
1347 int blk_get_max_iov(BlockBackend *blk)
1349 return blk->root->bs->bl.max_iov;
1352 void blk_set_guest_block_size(BlockBackend *blk, int align)
1354 blk->guest_block_size = align;
1357 void *blk_try_blockalign(BlockBackend *blk, size_t size)
1359 return qemu_try_blockalign(blk ? blk_bs(blk) : NULL, size);
1362 void *blk_blockalign(BlockBackend *blk, size_t size)
1364 return qemu_blockalign(blk ? blk_bs(blk) : NULL, size);
1367 bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp)
1369 BlockDriverState *bs = blk_bs(blk);
1375 return bdrv_op_is_blocked(bs, op, errp);
1378 void blk_op_unblock(BlockBackend *blk, BlockOpType op, Error *reason)
1380 BlockDriverState *bs = blk_bs(blk);
1383 bdrv_op_unblock(bs, op, reason);
1387 void blk_op_block_all(BlockBackend *blk, Error *reason)
1389 BlockDriverState *bs = blk_bs(blk);
1392 bdrv_op_block_all(bs, reason);
1396 void blk_op_unblock_all(BlockBackend *blk, Error *reason)
1398 BlockDriverState *bs = blk_bs(blk);
1401 bdrv_op_unblock_all(bs, reason);
1405 AioContext *blk_get_aio_context(BlockBackend *blk)
1407 BlockDriverState *bs = blk_bs(blk);
1410 return bdrv_get_aio_context(bs);
1412 return qemu_get_aio_context();
1416 static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb)
1418 BlockBackendAIOCB *blk_acb = DO_UPCAST(BlockBackendAIOCB, common, acb);
1419 return blk_get_aio_context(blk_acb->blk);
1422 void blk_set_aio_context(BlockBackend *blk, AioContext *new_context)
1424 BlockDriverState *bs = blk_bs(blk);
1427 if (blk->public.throttle_state) {
1428 throttle_timers_detach_aio_context(&blk->public.throttle_timers);
1430 bdrv_set_aio_context(bs, new_context);
1431 if (blk->public.throttle_state) {
1432 throttle_timers_attach_aio_context(&blk->public.throttle_timers,
1438 void blk_add_aio_context_notifier(BlockBackend *blk,
1439 void (*attached_aio_context)(AioContext *new_context, void *opaque),
1440 void (*detach_aio_context)(void *opaque), void *opaque)
1442 BlockDriverState *bs = blk_bs(blk);
1445 bdrv_add_aio_context_notifier(bs, attached_aio_context,
1446 detach_aio_context, opaque);
1450 void blk_remove_aio_context_notifier(BlockBackend *blk,
1451 void (*attached_aio_context)(AioContext *,
1453 void (*detach_aio_context)(void *),
1456 BlockDriverState *bs = blk_bs(blk);
1459 bdrv_remove_aio_context_notifier(bs, attached_aio_context,
1460 detach_aio_context, opaque);
1464 void blk_add_remove_bs_notifier(BlockBackend *blk, Notifier *notify)
1466 notifier_list_add(&blk->remove_bs_notifiers, notify);
1469 void blk_add_insert_bs_notifier(BlockBackend *blk, Notifier *notify)
1471 notifier_list_add(&blk->insert_bs_notifiers, notify);
1474 void blk_io_plug(BlockBackend *blk)
1476 BlockDriverState *bs = blk_bs(blk);
1483 void blk_io_unplug(BlockBackend *blk)
1485 BlockDriverState *bs = blk_bs(blk);
1492 BlockAcctStats *blk_get_stats(BlockBackend *blk)
1497 void *blk_aio_get(const AIOCBInfo *aiocb_info, BlockBackend *blk,
1498 BlockCompletionFunc *cb, void *opaque)
1500 return qemu_aio_get(aiocb_info, blk_bs(blk), cb, opaque);
1503 int coroutine_fn blk_co_write_zeroes(BlockBackend *blk, int64_t offset,
1504 int count, BdrvRequestFlags flags)
1506 return blk_co_pwritev(blk, offset, count, NULL,
1507 flags | BDRV_REQ_ZERO_WRITE);
1510 int blk_write_compressed(BlockBackend *blk, int64_t sector_num,
1511 const uint8_t *buf, int nb_sectors)
1513 int ret = blk_check_request(blk, sector_num, nb_sectors);
1518 return bdrv_write_compressed(blk_bs(blk), sector_num, buf, nb_sectors);
1521 int blk_truncate(BlockBackend *blk, int64_t offset)
1523 if (!blk_is_available(blk)) {
1527 return bdrv_truncate(blk_bs(blk), offset);
1530 int blk_discard(BlockBackend *blk, int64_t sector_num, int nb_sectors)
1532 int ret = blk_check_request(blk, sector_num, nb_sectors);
1537 return bdrv_discard(blk_bs(blk), sector_num, nb_sectors);
1540 int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
1541 int64_t pos, int size)
1545 if (!blk_is_available(blk)) {
1549 ret = bdrv_save_vmstate(blk_bs(blk), buf, pos, size);
1554 if (ret == size && !blk->enable_write_cache) {
1555 ret = bdrv_flush(blk_bs(blk));
1558 return ret < 0 ? ret : size;
1561 int blk_load_vmstate(BlockBackend *blk, uint8_t *buf, int64_t pos, int size)
1563 if (!blk_is_available(blk)) {
1567 return bdrv_load_vmstate(blk_bs(blk), buf, pos, size);
1570 int blk_probe_blocksizes(BlockBackend *blk, BlockSizes *bsz)
1572 if (!blk_is_available(blk)) {
1576 return bdrv_probe_blocksizes(blk_bs(blk), bsz);
1579 int blk_probe_geometry(BlockBackend *blk, HDGeometry *geo)
1581 if (!blk_is_available(blk)) {
1585 return bdrv_probe_geometry(blk_bs(blk), geo);
1589 * Updates the BlockBackendRootState object with data from the currently
1590 * attached BlockDriverState.
1592 void blk_update_root_state(BlockBackend *blk)
1596 blk->root_state.open_flags = blk->root->bs->open_flags;
1597 blk->root_state.read_only = blk->root->bs->read_only;
1598 blk->root_state.detect_zeroes = blk->root->bs->detect_zeroes;
1602 * Applies the information in the root state to the given BlockDriverState. This
1603 * does not include the flags which have to be specified for bdrv_open(), use
1604 * blk_get_open_flags_from_root_state() to inquire them.
1606 void blk_apply_root_state(BlockBackend *blk, BlockDriverState *bs)
1608 bs->detect_zeroes = blk->root_state.detect_zeroes;
1612 * Returns the flags to be used for bdrv_open() of a BlockDriverState which is
1613 * supposed to inherit the root state.
1615 int blk_get_open_flags_from_root_state(BlockBackend *blk)
1619 bs_flags = blk->root_state.read_only ? 0 : BDRV_O_RDWR;
1620 bs_flags |= blk->root_state.open_flags & ~BDRV_O_RDWR;
1625 BlockBackendRootState *blk_get_root_state(BlockBackend *blk)
1627 return &blk->root_state;
1630 int blk_commit_all(void)
1632 BlockBackend *blk = NULL;
1634 while ((blk = blk_all_next(blk)) != NULL) {
1635 AioContext *aio_context = blk_get_aio_context(blk);
1637 aio_context_acquire(aio_context);
1638 if (blk_is_inserted(blk) && blk->root->bs->backing) {
1639 int ret = bdrv_commit(blk->root->bs);
1641 aio_context_release(aio_context);
1645 aio_context_release(aio_context);
1650 int blk_flush_all(void)
1652 BlockBackend *blk = NULL;
1655 while ((blk = blk_all_next(blk)) != NULL) {
1656 AioContext *aio_context = blk_get_aio_context(blk);
1659 aio_context_acquire(aio_context);
1660 if (blk_is_inserted(blk)) {
1661 ret = blk_flush(blk);
1662 if (ret < 0 && !result) {
1666 aio_context_release(aio_context);
1673 /* throttling disk I/O limits */
1674 void blk_set_io_limits(BlockBackend *blk, ThrottleConfig *cfg)
1676 throttle_group_config(blk, cfg);
1679 void blk_io_limits_disable(BlockBackend *blk)
1681 assert(blk->public.throttle_state);
1682 bdrv_drained_begin(blk_bs(blk));
1683 throttle_group_unregister_blk(blk);
1684 bdrv_drained_end(blk_bs(blk));
1687 /* should be called before blk_set_io_limits if a limit is set */
1688 void blk_io_limits_enable(BlockBackend *blk, const char *group)
1690 assert(!blk->public.throttle_state);
1691 throttle_group_register_blk(blk, group);
1694 void blk_io_limits_update_group(BlockBackend *blk, const char *group)
1696 /* this BB is not part of any group */
1697 if (!blk->public.throttle_state) {
1701 /* this BB is a part of the same group than the one we want */
1702 if (!g_strcmp0(throttle_group_get_name(blk), group)) {
1706 /* need to change the group this bs belong to */
1707 blk_io_limits_disable(blk);
1708 blk_io_limits_enable(blk, group);
1711 static void blk_root_drained_begin(BdrvChild *child)
1713 BlockBackend *blk = child->opaque;
1715 if (blk->public.io_limits_disabled++ == 0) {
1716 throttle_group_restart_blk(blk);
1720 static void blk_root_drained_end(BdrvChild *child)
1722 BlockBackend *blk = child->opaque;
1724 assert(blk->public.io_limits_disabled);
1725 --blk->public.io_limits_disabled;