2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24 #include "config-host.h"
25 #include "qemu-common.h"
27 #include "block/block_int.h"
28 #include "block/blockjob.h"
29 #include "qemu/module.h"
30 #include "qapi/qmp/qjson.h"
31 #include "sysemu/sysemu.h"
32 #include "qemu/notify.h"
33 #include "block/coroutine.h"
34 #include "block/qapi.h"
35 #include "qmp-commands.h"
36 #include "qemu/timer.h"
37 #include "qapi-event.h"
40 #include <sys/types.h>
42 #include <sys/ioctl.h>
43 #include <sys/queue.h>
53 struct BdrvDirtyBitmap {
55 QLIST_ENTRY(BdrvDirtyBitmap) list;
58 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
60 #define COROUTINE_POOL_RESERVATION 64 /* number of coroutines to reserve */
62 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load);
63 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
64 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
65 BlockDriverCompletionFunc *cb, void *opaque);
66 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
67 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
68 BlockDriverCompletionFunc *cb, void *opaque);
69 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
70 int64_t sector_num, int nb_sectors,
72 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
73 int64_t sector_num, int nb_sectors,
75 static int coroutine_fn bdrv_co_do_preadv(BlockDriverState *bs,
76 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
77 BdrvRequestFlags flags);
78 static int coroutine_fn bdrv_co_do_pwritev(BlockDriverState *bs,
79 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
80 BdrvRequestFlags flags);
81 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
85 BdrvRequestFlags flags,
86 BlockDriverCompletionFunc *cb,
89 static void coroutine_fn bdrv_co_do_rw(void *opaque);
90 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
91 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags);
93 static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
94 QTAILQ_HEAD_INITIALIZER(bdrv_states);
96 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
97 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
99 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
100 QLIST_HEAD_INITIALIZER(bdrv_drivers);
102 /* If non-zero, use only whitelisted block drivers */
103 static int use_bdrv_whitelist;
106 static int is_windows_drive_prefix(const char *filename)
108 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
109 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
113 int is_windows_drive(const char *filename)
115 if (is_windows_drive_prefix(filename) &&
118 if (strstart(filename, "\\\\.\\", NULL) ||
119 strstart(filename, "//./", NULL))
125 /* throttling disk I/O limits */
126 void bdrv_set_io_limits(BlockDriverState *bs,
131 throttle_config(&bs->throttle_state, cfg);
133 for (i = 0; i < 2; i++) {
134 qemu_co_enter_next(&bs->throttled_reqs[i]);
138 /* this function drain all the throttled IOs */
139 static bool bdrv_start_throttled_reqs(BlockDriverState *bs)
141 bool drained = false;
142 bool enabled = bs->io_limits_enabled;
145 bs->io_limits_enabled = false;
147 for (i = 0; i < 2; i++) {
148 while (qemu_co_enter_next(&bs->throttled_reqs[i])) {
153 bs->io_limits_enabled = enabled;
158 void bdrv_io_limits_disable(BlockDriverState *bs)
160 bs->io_limits_enabled = false;
162 bdrv_start_throttled_reqs(bs);
164 throttle_destroy(&bs->throttle_state);
167 static void bdrv_throttle_read_timer_cb(void *opaque)
169 BlockDriverState *bs = opaque;
170 qemu_co_enter_next(&bs->throttled_reqs[0]);
173 static void bdrv_throttle_write_timer_cb(void *opaque)
175 BlockDriverState *bs = opaque;
176 qemu_co_enter_next(&bs->throttled_reqs[1]);
179 /* should be called before bdrv_set_io_limits if a limit is set */
180 void bdrv_io_limits_enable(BlockDriverState *bs)
182 assert(!bs->io_limits_enabled);
183 throttle_init(&bs->throttle_state,
184 bdrv_get_aio_context(bs),
186 bdrv_throttle_read_timer_cb,
187 bdrv_throttle_write_timer_cb,
189 bs->io_limits_enabled = true;
192 /* This function makes an IO wait if needed
194 * @nb_sectors: the number of sectors of the IO
195 * @is_write: is the IO a write
197 static void bdrv_io_limits_intercept(BlockDriverState *bs,
201 /* does this io must wait */
202 bool must_wait = throttle_schedule_timer(&bs->throttle_state, is_write);
204 /* if must wait or any request of this type throttled queue the IO */
206 !qemu_co_queue_empty(&bs->throttled_reqs[is_write])) {
207 qemu_co_queue_wait(&bs->throttled_reqs[is_write]);
210 /* the IO will be executed, do the accounting */
211 throttle_account(&bs->throttle_state, is_write, bytes);
214 /* if the next request must wait -> do nothing */
215 if (throttle_schedule_timer(&bs->throttle_state, is_write)) {
219 /* else queue next request for execution */
220 qemu_co_queue_next(&bs->throttled_reqs[is_write]);
223 size_t bdrv_opt_mem_align(BlockDriverState *bs)
225 if (!bs || !bs->drv) {
226 /* 4k should be on the safe side */
230 return bs->bl.opt_mem_alignment;
233 /* check if the path starts with "<protocol>:" */
234 static int path_has_protocol(const char *path)
239 if (is_windows_drive(path) ||
240 is_windows_drive_prefix(path)) {
243 p = path + strcspn(path, ":/\\");
245 p = path + strcspn(path, ":/");
251 int path_is_absolute(const char *path)
254 /* specific case for names like: "\\.\d:" */
255 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
258 return (*path == '/' || *path == '\\');
260 return (*path == '/');
264 /* if filename is absolute, just copy it to dest. Otherwise, build a
265 path to it by considering it is relative to base_path. URL are
267 void path_combine(char *dest, int dest_size,
268 const char *base_path,
269 const char *filename)
276 if (path_is_absolute(filename)) {
277 pstrcpy(dest, dest_size, filename);
279 p = strchr(base_path, ':');
284 p1 = strrchr(base_path, '/');
288 p2 = strrchr(base_path, '\\');
300 if (len > dest_size - 1)
302 memcpy(dest, base_path, len);
304 pstrcat(dest, dest_size, filename);
308 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz)
310 if (bs->backing_file[0] == '\0' || path_has_protocol(bs->backing_file)) {
311 pstrcpy(dest, sz, bs->backing_file);
313 path_combine(dest, sz, bs->filename, bs->backing_file);
317 void bdrv_register(BlockDriver *bdrv)
319 /* Block drivers without coroutine functions need emulation */
320 if (!bdrv->bdrv_co_readv) {
321 bdrv->bdrv_co_readv = bdrv_co_readv_em;
322 bdrv->bdrv_co_writev = bdrv_co_writev_em;
324 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
325 * the block driver lacks aio we need to emulate that too.
327 if (!bdrv->bdrv_aio_readv) {
328 /* add AIO emulation layer */
329 bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
330 bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
334 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
337 /* create a new block device (by default it is empty) */
338 BlockDriverState *bdrv_new(const char *device_name, Error **errp)
340 BlockDriverState *bs;
343 if (bdrv_find(device_name)) {
344 error_setg(errp, "Device with id '%s' already exists",
348 if (bdrv_find_node(device_name)) {
349 error_setg(errp, "Device with node-name '%s' already exists",
354 bs = g_malloc0(sizeof(BlockDriverState));
355 QLIST_INIT(&bs->dirty_bitmaps);
356 pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
357 if (device_name[0] != '\0') {
358 QTAILQ_INSERT_TAIL(&bdrv_states, bs, device_list);
360 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
361 QLIST_INIT(&bs->op_blockers[i]);
363 bdrv_iostatus_disable(bs);
364 notifier_list_init(&bs->close_notifiers);
365 notifier_with_return_list_init(&bs->before_write_notifiers);
366 qemu_co_queue_init(&bs->throttled_reqs[0]);
367 qemu_co_queue_init(&bs->throttled_reqs[1]);
369 bs->aio_context = qemu_get_aio_context();
374 void bdrv_add_close_notifier(BlockDriverState *bs, Notifier *notify)
376 notifier_list_add(&bs->close_notifiers, notify);
379 BlockDriver *bdrv_find_format(const char *format_name)
382 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
383 if (!strcmp(drv1->format_name, format_name)) {
390 static int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
392 static const char *whitelist_rw[] = {
393 CONFIG_BDRV_RW_WHITELIST
395 static const char *whitelist_ro[] = {
396 CONFIG_BDRV_RO_WHITELIST
400 if (!whitelist_rw[0] && !whitelist_ro[0]) {
401 return 1; /* no whitelist, anything goes */
404 for (p = whitelist_rw; *p; p++) {
405 if (!strcmp(drv->format_name, *p)) {
410 for (p = whitelist_ro; *p; p++) {
411 if (!strcmp(drv->format_name, *p)) {
419 BlockDriver *bdrv_find_whitelisted_format(const char *format_name,
422 BlockDriver *drv = bdrv_find_format(format_name);
423 return drv && bdrv_is_whitelisted(drv, read_only) ? drv : NULL;
426 typedef struct CreateCo {
434 static void coroutine_fn bdrv_create_co_entry(void *opaque)
436 Error *local_err = NULL;
439 CreateCo *cco = opaque;
442 ret = cco->drv->bdrv_create(cco->filename, cco->opts, &local_err);
444 error_propagate(&cco->err, local_err);
449 int bdrv_create(BlockDriver *drv, const char* filename,
450 QemuOpts *opts, Error **errp)
457 .filename = g_strdup(filename),
463 if (!drv->bdrv_create) {
464 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
469 if (qemu_in_coroutine()) {
470 /* Fast-path if already in coroutine context */
471 bdrv_create_co_entry(&cco);
473 co = qemu_coroutine_create(bdrv_create_co_entry);
474 qemu_coroutine_enter(co, &cco);
475 while (cco.ret == NOT_DONE) {
476 aio_poll(qemu_get_aio_context(), true);
483 error_propagate(errp, cco.err);
485 error_setg_errno(errp, -ret, "Could not create image");
490 g_free(cco.filename);
494 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
497 Error *local_err = NULL;
500 drv = bdrv_find_protocol(filename, true);
502 error_setg(errp, "Could not find protocol for file '%s'", filename);
506 ret = bdrv_create(drv, filename, opts, &local_err);
508 error_propagate(errp, local_err);
513 void bdrv_refresh_limits(BlockDriverState *bs, Error **errp)
515 BlockDriver *drv = bs->drv;
516 Error *local_err = NULL;
518 memset(&bs->bl, 0, sizeof(bs->bl));
524 /* Take some limits from the children as a default */
526 bdrv_refresh_limits(bs->file, &local_err);
528 error_propagate(errp, local_err);
531 bs->bl.opt_transfer_length = bs->file->bl.opt_transfer_length;
532 bs->bl.opt_mem_alignment = bs->file->bl.opt_mem_alignment;
534 bs->bl.opt_mem_alignment = 512;
537 if (bs->backing_hd) {
538 bdrv_refresh_limits(bs->backing_hd, &local_err);
540 error_propagate(errp, local_err);
543 bs->bl.opt_transfer_length =
544 MAX(bs->bl.opt_transfer_length,
545 bs->backing_hd->bl.opt_transfer_length);
546 bs->bl.opt_mem_alignment =
547 MAX(bs->bl.opt_mem_alignment,
548 bs->backing_hd->bl.opt_mem_alignment);
551 /* Then let the driver override it */
552 if (drv->bdrv_refresh_limits) {
553 drv->bdrv_refresh_limits(bs, errp);
558 * Create a uniquely-named empty temporary file.
559 * Return 0 upon success, otherwise a negative errno value.
561 int get_tmp_filename(char *filename, int size)
564 char temp_dir[MAX_PATH];
565 /* GetTempFileName requires that its output buffer (4th param)
566 have length MAX_PATH or greater. */
567 assert(size >= MAX_PATH);
568 return (GetTempPath(MAX_PATH, temp_dir)
569 && GetTempFileName(temp_dir, "qem", 0, filename)
570 ? 0 : -GetLastError());
574 tmpdir = getenv("TMPDIR");
578 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
581 fd = mkstemp(filename);
585 if (close(fd) != 0) {
594 * Detect host devices. By convention, /dev/cdrom[N] is always
595 * recognized as a host CDROM.
597 static BlockDriver *find_hdev_driver(const char *filename)
599 int score_max = 0, score;
600 BlockDriver *drv = NULL, *d;
602 QLIST_FOREACH(d, &bdrv_drivers, list) {
603 if (d->bdrv_probe_device) {
604 score = d->bdrv_probe_device(filename);
605 if (score > score_max) {
615 BlockDriver *bdrv_find_protocol(const char *filename,
616 bool allow_protocol_prefix)
623 /* TODO Drivers without bdrv_file_open must be specified explicitly */
626 * XXX(hch): we really should not let host device detection
627 * override an explicit protocol specification, but moving this
628 * later breaks access to device names with colons in them.
629 * Thanks to the brain-dead persistent naming schemes on udev-
630 * based Linux systems those actually are quite common.
632 drv1 = find_hdev_driver(filename);
637 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
638 return bdrv_find_format("file");
641 p = strchr(filename, ':');
644 if (len > sizeof(protocol) - 1)
645 len = sizeof(protocol) - 1;
646 memcpy(protocol, filename, len);
647 protocol[len] = '\0';
648 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
649 if (drv1->protocol_name &&
650 !strcmp(drv1->protocol_name, protocol)) {
657 static int find_image_format(BlockDriverState *bs, const char *filename,
658 BlockDriver **pdrv, Error **errp)
660 int score, score_max;
661 BlockDriver *drv1, *drv;
665 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
666 if (bs->sg || !bdrv_is_inserted(bs) || bdrv_getlength(bs) == 0) {
667 drv = bdrv_find_format("raw");
669 error_setg(errp, "Could not find raw image format");
676 ret = bdrv_pread(bs, 0, buf, sizeof(buf));
678 error_setg_errno(errp, -ret, "Could not read image for determining its "
686 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
687 if (drv1->bdrv_probe) {
688 score = drv1->bdrv_probe(buf, ret, filename);
689 if (score > score_max) {
696 error_setg(errp, "Could not determine image format: No compatible "
705 * Set the current 'total_sectors' value
706 * Return 0 on success, -errno on error.
708 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
710 BlockDriver *drv = bs->drv;
712 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
716 /* query actual device if possible, otherwise just trust the hint */
717 if (drv->bdrv_getlength) {
718 int64_t length = drv->bdrv_getlength(bs);
722 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
725 bs->total_sectors = hint;
730 * Set open flags for a given discard mode
732 * Return 0 on success, -1 if the discard mode was invalid.
734 int bdrv_parse_discard_flags(const char *mode, int *flags)
736 *flags &= ~BDRV_O_UNMAP;
738 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
740 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
741 *flags |= BDRV_O_UNMAP;
750 * Set open flags for a given cache mode
752 * Return 0 on success, -1 if the cache mode was invalid.
754 int bdrv_parse_cache_flags(const char *mode, int *flags)
756 *flags &= ~BDRV_O_CACHE_MASK;
758 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
759 *flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB;
760 } else if (!strcmp(mode, "directsync")) {
761 *flags |= BDRV_O_NOCACHE;
762 } else if (!strcmp(mode, "writeback")) {
763 *flags |= BDRV_O_CACHE_WB;
764 } else if (!strcmp(mode, "unsafe")) {
765 *flags |= BDRV_O_CACHE_WB;
766 *flags |= BDRV_O_NO_FLUSH;
767 } else if (!strcmp(mode, "writethrough")) {
768 /* this is the default */
777 * The copy-on-read flag is actually a reference count so multiple users may
778 * use the feature without worrying about clobbering its previous state.
779 * Copy-on-read stays enabled until all users have called to disable it.
781 void bdrv_enable_copy_on_read(BlockDriverState *bs)
786 void bdrv_disable_copy_on_read(BlockDriverState *bs)
788 assert(bs->copy_on_read > 0);
793 * Returns the flags that a temporary snapshot should get, based on the
794 * originally requested flags (the originally requested image will have flags
795 * like a backing file)
797 static int bdrv_temp_snapshot_flags(int flags)
799 return (flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
803 * Returns the flags that bs->file should get, based on the given flags for
806 static int bdrv_inherited_flags(int flags)
808 /* Enable protocol handling, disable format probing for bs->file */
809 flags |= BDRV_O_PROTOCOL;
811 /* Our block drivers take care to send flushes and respect unmap policy,
812 * so we can enable both unconditionally on lower layers. */
813 flags |= BDRV_O_CACHE_WB | BDRV_O_UNMAP;
815 /* Clear flags that only apply to the top layer */
816 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
822 * Returns the flags that bs->backing_hd should get, based on the given flags
825 static int bdrv_backing_flags(int flags)
827 /* backing files always opened read-only */
828 flags &= ~(BDRV_O_RDWR | BDRV_O_COPY_ON_READ);
830 /* snapshot=on is handled on the top layer */
831 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
836 static int bdrv_open_flags(BlockDriverState *bs, int flags)
838 int open_flags = flags | BDRV_O_CACHE_WB;
841 * Clear flags that are internal to the block layer before opening the
844 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
847 * Snapshots should be writable.
849 if (flags & BDRV_O_TEMPORARY) {
850 open_flags |= BDRV_O_RDWR;
856 static void bdrv_assign_node_name(BlockDriverState *bs,
857 const char *node_name,
864 /* empty string node name is invalid */
865 if (node_name[0] == '\0') {
866 error_setg(errp, "Empty node name");
870 /* takes care of avoiding namespaces collisions */
871 if (bdrv_find(node_name)) {
872 error_setg(errp, "node-name=%s is conflicting with a device id",
877 /* takes care of avoiding duplicates node names */
878 if (bdrv_find_node(node_name)) {
879 error_setg(errp, "Duplicate node name");
883 /* copy node name into the bs and insert it into the graph list */
884 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
885 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
889 * Common part for opening disk images and files
891 * Removes all processed options from *options.
893 static int bdrv_open_common(BlockDriverState *bs, BlockDriverState *file,
894 QDict *options, int flags, BlockDriver *drv, Error **errp)
897 const char *filename;
898 const char *node_name = NULL;
899 Error *local_err = NULL;
902 assert(bs->file == NULL);
903 assert(options != NULL && bs->options != options);
906 filename = file->filename;
908 filename = qdict_get_try_str(options, "filename");
911 if (drv->bdrv_needs_filename && !filename) {
912 error_setg(errp, "The '%s' block driver requires a file name",
917 trace_bdrv_open_common(bs, filename ?: "", flags, drv->format_name);
919 node_name = qdict_get_try_str(options, "node-name");
920 bdrv_assign_node_name(bs, node_name, &local_err);
922 error_propagate(errp, local_err);
925 qdict_del(options, "node-name");
927 /* bdrv_open() with directly using a protocol as drv. This layer is already
928 * opened, so assign it to bs (while file becomes a closed BlockDriverState)
929 * and return immediately. */
930 if (file != NULL && drv->bdrv_file_open) {
935 bs->open_flags = flags;
936 bs->guest_block_size = 512;
937 bs->request_alignment = 512;
938 bs->zero_beyond_eof = true;
939 open_flags = bdrv_open_flags(bs, flags);
940 bs->read_only = !(open_flags & BDRV_O_RDWR);
941 bs->growable = !!(flags & BDRV_O_PROTOCOL);
943 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
945 !bs->read_only && bdrv_is_whitelisted(drv, true)
946 ? "Driver '%s' can only be used for read-only devices"
947 : "Driver '%s' is not whitelisted",
952 assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
953 if (flags & BDRV_O_COPY_ON_READ) {
954 if (!bs->read_only) {
955 bdrv_enable_copy_on_read(bs);
957 error_setg(errp, "Can't use copy-on-read on read-only device");
962 if (filename != NULL) {
963 pstrcpy(bs->filename, sizeof(bs->filename), filename);
965 bs->filename[0] = '\0';
969 bs->opaque = g_malloc0(drv->instance_size);
971 bs->enable_write_cache = !!(flags & BDRV_O_CACHE_WB);
973 /* Open the image, either directly or using a protocol */
974 if (drv->bdrv_file_open) {
975 assert(file == NULL);
976 assert(!drv->bdrv_needs_filename || filename != NULL);
977 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
980 error_setg(errp, "Can't use '%s' as a block driver for the "
981 "protocol level", drv->format_name);
986 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
991 error_propagate(errp, local_err);
992 } else if (bs->filename[0]) {
993 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
995 error_setg_errno(errp, -ret, "Could not open image");
1000 ret = refresh_total_sectors(bs, bs->total_sectors);
1002 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1006 bdrv_refresh_limits(bs, &local_err);
1008 error_propagate(errp, local_err);
1013 assert(bdrv_opt_mem_align(bs) != 0);
1014 assert((bs->request_alignment != 0) || bs->sg);
1025 static QDict *parse_json_filename(const char *filename, Error **errp)
1027 QObject *options_obj;
1031 ret = strstart(filename, "json:", &filename);
1034 options_obj = qobject_from_json(filename);
1036 error_setg(errp, "Could not parse the JSON options");
1040 if (qobject_type(options_obj) != QTYPE_QDICT) {
1041 qobject_decref(options_obj);
1042 error_setg(errp, "Invalid JSON object given");
1046 options = qobject_to_qdict(options_obj);
1047 qdict_flatten(options);
1053 * Fills in default options for opening images and converts the legacy
1054 * filename/flags pair to option QDict entries.
1056 static int bdrv_fill_options(QDict **options, const char **pfilename, int flags,
1057 BlockDriver *drv, Error **errp)
1059 const char *filename = *pfilename;
1060 const char *drvname;
1061 bool protocol = flags & BDRV_O_PROTOCOL;
1062 bool parse_filename = false;
1063 Error *local_err = NULL;
1065 /* Parse json: pseudo-protocol */
1066 if (filename && g_str_has_prefix(filename, "json:")) {
1067 QDict *json_options = parse_json_filename(filename, &local_err);
1069 error_propagate(errp, local_err);
1073 /* Options given in the filename have lower priority than options
1074 * specified directly */
1075 qdict_join(*options, json_options, false);
1076 QDECREF(json_options);
1077 *pfilename = filename = NULL;
1080 /* Fetch the file name from the options QDict if necessary */
1081 if (protocol && filename) {
1082 if (!qdict_haskey(*options, "filename")) {
1083 qdict_put(*options, "filename", qstring_from_str(filename));
1084 parse_filename = true;
1086 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1092 /* Find the right block driver */
1093 filename = qdict_get_try_str(*options, "filename");
1094 drvname = qdict_get_try_str(*options, "driver");
1098 error_setg(errp, "Driver specified twice");
1101 drvname = drv->format_name;
1102 qdict_put(*options, "driver", qstring_from_str(drvname));
1104 if (!drvname && protocol) {
1106 drv = bdrv_find_protocol(filename, parse_filename);
1108 error_setg(errp, "Unknown protocol");
1112 drvname = drv->format_name;
1113 qdict_put(*options, "driver", qstring_from_str(drvname));
1115 error_setg(errp, "Must specify either driver or file");
1118 } else if (drvname) {
1119 drv = bdrv_find_format(drvname);
1121 error_setg(errp, "Unknown driver '%s'", drvname);
1127 assert(drv || !protocol);
1129 /* Driver-specific filename parsing */
1130 if (drv && drv->bdrv_parse_filename && parse_filename) {
1131 drv->bdrv_parse_filename(filename, *options, &local_err);
1133 error_propagate(errp, local_err);
1137 if (!drv->bdrv_needs_filename) {
1138 qdict_del(*options, "filename");
1145 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd)
1148 if (bs->backing_hd) {
1149 assert(bs->backing_blocker);
1150 bdrv_op_unblock_all(bs->backing_hd, bs->backing_blocker);
1151 } else if (backing_hd) {
1152 error_setg(&bs->backing_blocker,
1153 "device is used as backing hd of '%s'",
1157 bs->backing_hd = backing_hd;
1159 error_free(bs->backing_blocker);
1160 bs->backing_blocker = NULL;
1163 bs->open_flags &= ~BDRV_O_NO_BACKING;
1164 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_hd->filename);
1165 pstrcpy(bs->backing_format, sizeof(bs->backing_format),
1166 backing_hd->drv ? backing_hd->drv->format_name : "");
1168 bdrv_op_block_all(bs->backing_hd, bs->backing_blocker);
1169 /* Otherwise we won't be able to commit due to check in bdrv_commit */
1170 bdrv_op_unblock(bs->backing_hd, BLOCK_OP_TYPE_COMMIT,
1171 bs->backing_blocker);
1173 bdrv_refresh_limits(bs, NULL);
1177 * Opens the backing file for a BlockDriverState if not yet open
1179 * options is a QDict of options to pass to the block drivers, or NULL for an
1180 * empty set of options. The reference to the QDict is transferred to this
1181 * function (even on failure), so if the caller intends to reuse the dictionary,
1182 * it needs to use QINCREF() before calling bdrv_file_open.
1184 int bdrv_open_backing_file(BlockDriverState *bs, QDict *options, Error **errp)
1186 char *backing_filename = g_malloc0(PATH_MAX);
1188 BlockDriver *back_drv = NULL;
1189 BlockDriverState *backing_hd;
1190 Error *local_err = NULL;
1192 if (bs->backing_hd != NULL) {
1197 /* NULL means an empty set of options */
1198 if (options == NULL) {
1199 options = qdict_new();
1202 bs->open_flags &= ~BDRV_O_NO_BACKING;
1203 if (qdict_haskey(options, "file.filename")) {
1204 backing_filename[0] = '\0';
1205 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
1209 bdrv_get_full_backing_filename(bs, backing_filename, PATH_MAX);
1212 if (!bs->drv || !bs->drv->supports_backing) {
1214 error_setg(errp, "Driver doesn't support backing files");
1219 backing_hd = bdrv_new("", errp);
1221 if (bs->backing_format[0] != '\0') {
1222 back_drv = bdrv_find_format(bs->backing_format);
1225 assert(bs->backing_hd == NULL);
1226 ret = bdrv_open(&backing_hd,
1227 *backing_filename ? backing_filename : NULL, NULL, options,
1228 bdrv_backing_flags(bs->open_flags), back_drv, &local_err);
1230 bdrv_unref(backing_hd);
1232 bs->open_flags |= BDRV_O_NO_BACKING;
1233 error_setg(errp, "Could not open backing file: %s",
1234 error_get_pretty(local_err));
1235 error_free(local_err);
1238 bdrv_set_backing_hd(bs, backing_hd);
1241 g_free(backing_filename);
1246 * Opens a disk image whose options are given as BlockdevRef in another block
1249 * If allow_none is true, no image will be opened if filename is false and no
1250 * BlockdevRef is given. *pbs will remain unchanged and 0 will be returned.
1252 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
1253 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
1254 * itself, all options starting with "${bdref_key}." are considered part of the
1257 * The BlockdevRef will be removed from the options QDict.
1259 * To conform with the behavior of bdrv_open(), *pbs has to be NULL.
1261 int bdrv_open_image(BlockDriverState **pbs, const char *filename,
1262 QDict *options, const char *bdref_key, int flags,
1263 bool allow_none, Error **errp)
1265 QDict *image_options;
1267 char *bdref_key_dot;
1268 const char *reference;
1271 assert(*pbs == NULL);
1273 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
1274 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
1275 g_free(bdref_key_dot);
1277 reference = qdict_get_try_str(options, bdref_key);
1278 if (!filename && !reference && !qdict_size(image_options)) {
1282 error_setg(errp, "A block device must be specified for \"%s\"",
1286 QDECREF(image_options);
1290 ret = bdrv_open(pbs, filename, reference, image_options, flags, NULL, errp);
1293 qdict_del(options, bdref_key);
1297 int bdrv_append_temp_snapshot(BlockDriverState *bs, int flags, Error **errp)
1299 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
1300 char *tmp_filename = g_malloc0(PATH_MAX + 1);
1302 BlockDriver *bdrv_qcow2;
1303 QemuOpts *opts = NULL;
1304 QDict *snapshot_options;
1305 BlockDriverState *bs_snapshot;
1309 /* if snapshot, we create a temporary backing file and open it
1310 instead of opening 'filename' directly */
1312 /* Get the required size from the image */
1313 total_size = bdrv_getlength(bs);
1314 if (total_size < 0) {
1316 error_setg_errno(errp, -total_size, "Could not get image size");
1320 /* Create the temporary image */
1321 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
1323 error_setg_errno(errp, -ret, "Could not get temporary filename");
1327 bdrv_qcow2 = bdrv_find_format("qcow2");
1328 opts = qemu_opts_create(bdrv_qcow2->create_opts, NULL, 0,
1330 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size);
1331 ret = bdrv_create(bdrv_qcow2, tmp_filename, opts, &local_err);
1332 qemu_opts_del(opts);
1334 error_setg_errno(errp, -ret, "Could not create temporary overlay "
1335 "'%s': %s", tmp_filename,
1336 error_get_pretty(local_err));
1337 error_free(local_err);
1341 /* Prepare a new options QDict for the temporary file */
1342 snapshot_options = qdict_new();
1343 qdict_put(snapshot_options, "file.driver",
1344 qstring_from_str("file"));
1345 qdict_put(snapshot_options, "file.filename",
1346 qstring_from_str(tmp_filename));
1348 bs_snapshot = bdrv_new("", &error_abort);
1350 ret = bdrv_open(&bs_snapshot, NULL, NULL, snapshot_options,
1351 flags, bdrv_qcow2, &local_err);
1353 error_propagate(errp, local_err);
1357 bdrv_append(bs_snapshot, bs);
1360 g_free(tmp_filename);
1365 * Opens a disk image (raw, qcow2, vmdk, ...)
1367 * options is a QDict of options to pass to the block drivers, or NULL for an
1368 * empty set of options. The reference to the QDict belongs to the block layer
1369 * after the call (even on failure), so if the caller intends to reuse the
1370 * dictionary, it needs to use QINCREF() before calling bdrv_open.
1372 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
1373 * If it is not NULL, the referenced BDS will be reused.
1375 * The reference parameter may be used to specify an existing block device which
1376 * should be opened. If specified, neither options nor a filename may be given,
1377 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
1379 int bdrv_open(BlockDriverState **pbs, const char *filename,
1380 const char *reference, QDict *options, int flags,
1381 BlockDriver *drv, Error **errp)
1384 BlockDriverState *file = NULL, *bs;
1385 const char *drvname;
1386 Error *local_err = NULL;
1387 int snapshot_flags = 0;
1392 bool options_non_empty = options ? qdict_size(options) : false;
1396 error_setg(errp, "Cannot reuse an existing BDS when referencing "
1397 "another block device");
1401 if (filename || options_non_empty) {
1402 error_setg(errp, "Cannot reference an existing block device with "
1403 "additional options or a new filename");
1407 bs = bdrv_lookup_bs(reference, reference, errp);
1419 bs = bdrv_new("", &error_abort);
1422 /* NULL means an empty set of options */
1423 if (options == NULL) {
1424 options = qdict_new();
1427 ret = bdrv_fill_options(&options, &filename, flags, drv, &local_err);
1432 /* Find the right image format driver */
1434 drvname = qdict_get_try_str(options, "driver");
1436 drv = bdrv_find_format(drvname);
1437 qdict_del(options, "driver");
1439 error_setg(errp, "Unknown driver: '%s'", drvname);
1445 assert(drvname || !(flags & BDRV_O_PROTOCOL));
1446 if (drv && !drv->bdrv_file_open) {
1447 /* If the user explicitly wants a format driver here, we'll need to add
1448 * another layer for the protocol in bs->file */
1449 flags &= ~BDRV_O_PROTOCOL;
1452 bs->options = options;
1453 options = qdict_clone_shallow(options);
1455 /* Open image file without format layer */
1456 if ((flags & BDRV_O_PROTOCOL) == 0) {
1457 if (flags & BDRV_O_RDWR) {
1458 flags |= BDRV_O_ALLOW_RDWR;
1460 if (flags & BDRV_O_SNAPSHOT) {
1461 snapshot_flags = bdrv_temp_snapshot_flags(flags);
1462 flags = bdrv_backing_flags(flags);
1465 assert(file == NULL);
1466 ret = bdrv_open_image(&file, filename, options, "file",
1467 bdrv_inherited_flags(flags),
1474 /* Image format probing */
1476 ret = find_image_format(file, filename, &drv, &local_err);
1481 error_setg(errp, "Must specify either driver or file");
1486 /* Open the image */
1487 ret = bdrv_open_common(bs, file, options, flags, drv, &local_err);
1492 if (file && (bs->file != file)) {
1497 /* If there is a backing file, use it */
1498 if ((flags & BDRV_O_NO_BACKING) == 0) {
1499 QDict *backing_options;
1501 qdict_extract_subqdict(options, &backing_options, "backing.");
1502 ret = bdrv_open_backing_file(bs, backing_options, &local_err);
1504 goto close_and_fail;
1508 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
1509 * temporary snapshot afterwards. */
1510 if (snapshot_flags) {
1511 ret = bdrv_append_temp_snapshot(bs, snapshot_flags, &local_err);
1513 goto close_and_fail;
1517 /* Check if any unknown options were used */
1518 if (options && (qdict_size(options) != 0)) {
1519 const QDictEntry *entry = qdict_first(options);
1520 if (flags & BDRV_O_PROTOCOL) {
1521 error_setg(errp, "Block protocol '%s' doesn't support the option "
1522 "'%s'", drv->format_name, entry->key);
1524 error_setg(errp, "Block format '%s' used by device '%s' doesn't "
1525 "support the option '%s'", drv->format_name,
1526 bs->device_name, entry->key);
1530 goto close_and_fail;
1533 if (!bdrv_key_required(bs)) {
1534 bdrv_dev_change_media_cb(bs, true);
1535 } else if (!runstate_check(RUN_STATE_PRELAUNCH)
1536 && !runstate_check(RUN_STATE_INMIGRATE)
1537 && !runstate_check(RUN_STATE_PAUSED)) { /* HACK */
1539 "Guest must be stopped for opening of encrypted image");
1541 goto close_and_fail;
1552 QDECREF(bs->options);
1556 /* If *pbs is NULL, a new BDS has been created in this function and
1557 needs to be freed now. Otherwise, it does not need to be closed,
1558 since it has not really been opened yet. */
1562 error_propagate(errp, local_err);
1567 /* See fail path, but now the BDS has to be always closed */
1575 error_propagate(errp, local_err);
1580 typedef struct BlockReopenQueueEntry {
1582 BDRVReopenState state;
1583 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
1584 } BlockReopenQueueEntry;
1587 * Adds a BlockDriverState to a simple queue for an atomic, transactional
1588 * reopen of multiple devices.
1590 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
1591 * already performed, or alternatively may be NULL a new BlockReopenQueue will
1592 * be created and initialized. This newly created BlockReopenQueue should be
1593 * passed back in for subsequent calls that are intended to be of the same
1596 * bs is the BlockDriverState to add to the reopen queue.
1598 * flags contains the open flags for the associated bs
1600 * returns a pointer to bs_queue, which is either the newly allocated
1601 * bs_queue, or the existing bs_queue being used.
1604 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
1605 BlockDriverState *bs, int flags)
1609 BlockReopenQueueEntry *bs_entry;
1610 if (bs_queue == NULL) {
1611 bs_queue = g_new0(BlockReopenQueue, 1);
1612 QSIMPLEQ_INIT(bs_queue);
1615 /* bdrv_open() masks this flag out */
1616 flags &= ~BDRV_O_PROTOCOL;
1619 bdrv_reopen_queue(bs_queue, bs->file, bdrv_inherited_flags(flags));
1622 bs_entry = g_new0(BlockReopenQueueEntry, 1);
1623 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
1625 bs_entry->state.bs = bs;
1626 bs_entry->state.flags = flags;
1632 * Reopen multiple BlockDriverStates atomically & transactionally.
1634 * The queue passed in (bs_queue) must have been built up previous
1635 * via bdrv_reopen_queue().
1637 * Reopens all BDS specified in the queue, with the appropriate
1638 * flags. All devices are prepared for reopen, and failure of any
1639 * device will cause all device changes to be abandonded, and intermediate
1642 * If all devices prepare successfully, then the changes are committed
1646 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
1649 BlockReopenQueueEntry *bs_entry, *next;
1650 Error *local_err = NULL;
1652 assert(bs_queue != NULL);
1656 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
1657 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
1658 error_propagate(errp, local_err);
1661 bs_entry->prepared = true;
1664 /* If we reach this point, we have success and just need to apply the
1667 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
1668 bdrv_reopen_commit(&bs_entry->state);
1674 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
1675 if (ret && bs_entry->prepared) {
1676 bdrv_reopen_abort(&bs_entry->state);
1685 /* Reopen a single BlockDriverState with the specified flags. */
1686 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
1689 Error *local_err = NULL;
1690 BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, bdrv_flags);
1692 ret = bdrv_reopen_multiple(queue, &local_err);
1693 if (local_err != NULL) {
1694 error_propagate(errp, local_err);
1701 * Prepares a BlockDriverState for reopen. All changes are staged in the
1702 * 'opaque' field of the BDRVReopenState, which is used and allocated by
1703 * the block driver layer .bdrv_reopen_prepare()
1705 * bs is the BlockDriverState to reopen
1706 * flags are the new open flags
1707 * queue is the reopen queue
1709 * Returns 0 on success, non-zero on error. On error errp will be set
1712 * On failure, bdrv_reopen_abort() will be called to clean up any data.
1713 * It is the responsibility of the caller to then call the abort() or
1714 * commit() for any other BDS that have been left in a prepare() state
1717 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
1721 Error *local_err = NULL;
1724 assert(reopen_state != NULL);
1725 assert(reopen_state->bs->drv != NULL);
1726 drv = reopen_state->bs->drv;
1728 /* if we are to stay read-only, do not allow permission change
1730 if (!(reopen_state->bs->open_flags & BDRV_O_ALLOW_RDWR) &&
1731 reopen_state->flags & BDRV_O_RDWR) {
1732 error_set(errp, QERR_DEVICE_IS_READ_ONLY,
1733 reopen_state->bs->device_name);
1738 ret = bdrv_flush(reopen_state->bs);
1740 error_set(errp, ERROR_CLASS_GENERIC_ERROR, "Error (%s) flushing drive",
1745 if (drv->bdrv_reopen_prepare) {
1746 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
1748 if (local_err != NULL) {
1749 error_propagate(errp, local_err);
1751 error_setg(errp, "failed while preparing to reopen image '%s'",
1752 reopen_state->bs->filename);
1757 /* It is currently mandatory to have a bdrv_reopen_prepare()
1758 * handler for each supported drv. */
1759 error_set(errp, QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
1760 drv->format_name, reopen_state->bs->device_name,
1761 "reopening of file");
1773 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
1774 * makes them final by swapping the staging BlockDriverState contents into
1775 * the active BlockDriverState contents.
1777 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
1781 assert(reopen_state != NULL);
1782 drv = reopen_state->bs->drv;
1783 assert(drv != NULL);
1785 /* If there are any driver level actions to take */
1786 if (drv->bdrv_reopen_commit) {
1787 drv->bdrv_reopen_commit(reopen_state);
1790 /* set BDS specific flags now */
1791 reopen_state->bs->open_flags = reopen_state->flags;
1792 reopen_state->bs->enable_write_cache = !!(reopen_state->flags &
1794 reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
1796 bdrv_refresh_limits(reopen_state->bs, NULL);
1800 * Abort the reopen, and delete and free the staged changes in
1803 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
1807 assert(reopen_state != NULL);
1808 drv = reopen_state->bs->drv;
1809 assert(drv != NULL);
1811 if (drv->bdrv_reopen_abort) {
1812 drv->bdrv_reopen_abort(reopen_state);
1817 void bdrv_close(BlockDriverState *bs)
1820 block_job_cancel_sync(bs->job);
1822 bdrv_drain_all(); /* complete I/O */
1824 bdrv_drain_all(); /* in case flush left pending I/O */
1825 notifier_list_notify(&bs->close_notifiers, bs);
1828 if (bs->backing_hd) {
1829 BlockDriverState *backing_hd = bs->backing_hd;
1830 bdrv_set_backing_hd(bs, NULL);
1831 bdrv_unref(backing_hd);
1833 bs->drv->bdrv_close(bs);
1837 bs->copy_on_read = 0;
1838 bs->backing_file[0] = '\0';
1839 bs->backing_format[0] = '\0';
1840 bs->total_sectors = 0;
1845 bs->zero_beyond_eof = false;
1846 QDECREF(bs->options);
1849 if (bs->file != NULL) {
1850 bdrv_unref(bs->file);
1855 bdrv_dev_change_media_cb(bs, false);
1857 /*throttling disk I/O limits*/
1858 if (bs->io_limits_enabled) {
1859 bdrv_io_limits_disable(bs);
1863 void bdrv_close_all(void)
1865 BlockDriverState *bs;
1867 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
1868 AioContext *aio_context = bdrv_get_aio_context(bs);
1870 aio_context_acquire(aio_context);
1872 aio_context_release(aio_context);
1876 /* Check if any requests are in-flight (including throttled requests) */
1877 static bool bdrv_requests_pending(BlockDriverState *bs)
1879 if (!QLIST_EMPTY(&bs->tracked_requests)) {
1882 if (!qemu_co_queue_empty(&bs->throttled_reqs[0])) {
1885 if (!qemu_co_queue_empty(&bs->throttled_reqs[1])) {
1888 if (bs->file && bdrv_requests_pending(bs->file)) {
1891 if (bs->backing_hd && bdrv_requests_pending(bs->backing_hd)) {
1898 * Wait for pending requests to complete across all BlockDriverStates
1900 * This function does not flush data to disk, use bdrv_flush_all() for that
1901 * after calling this function.
1903 * Note that completion of an asynchronous I/O operation can trigger any
1904 * number of other I/O operations on other devices---for example a coroutine
1905 * can be arbitrarily complex and a constant flow of I/O can come until the
1906 * coroutine is complete. Because of this, it is not possible to have a
1907 * function to drain a single device's I/O queue.
1909 void bdrv_drain_all(void)
1911 /* Always run first iteration so any pending completion BHs run */
1913 BlockDriverState *bs;
1918 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
1919 AioContext *aio_context = bdrv_get_aio_context(bs);
1922 aio_context_acquire(aio_context);
1923 bdrv_flush_io_queue(bs);
1924 bdrv_start_throttled_reqs(bs);
1925 bs_busy = bdrv_requests_pending(bs);
1926 bs_busy |= aio_poll(aio_context, bs_busy);
1927 aio_context_release(aio_context);
1934 /* make a BlockDriverState anonymous by removing from bdrv_state and
1935 * graph_bdrv_state list.
1936 Also, NULL terminate the device_name to prevent double remove */
1937 void bdrv_make_anon(BlockDriverState *bs)
1939 if (bs->device_name[0] != '\0') {
1940 QTAILQ_REMOVE(&bdrv_states, bs, device_list);
1942 bs->device_name[0] = '\0';
1943 if (bs->node_name[0] != '\0') {
1944 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
1946 bs->node_name[0] = '\0';
1949 static void bdrv_rebind(BlockDriverState *bs)
1951 if (bs->drv && bs->drv->bdrv_rebind) {
1952 bs->drv->bdrv_rebind(bs);
1956 static void bdrv_move_feature_fields(BlockDriverState *bs_dest,
1957 BlockDriverState *bs_src)
1959 /* move some fields that need to stay attached to the device */
1962 bs_dest->dev_ops = bs_src->dev_ops;
1963 bs_dest->dev_opaque = bs_src->dev_opaque;
1964 bs_dest->dev = bs_src->dev;
1965 bs_dest->guest_block_size = bs_src->guest_block_size;
1966 bs_dest->copy_on_read = bs_src->copy_on_read;
1968 bs_dest->enable_write_cache = bs_src->enable_write_cache;
1970 /* i/o throttled req */
1971 memcpy(&bs_dest->throttle_state,
1972 &bs_src->throttle_state,
1973 sizeof(ThrottleState));
1974 bs_dest->throttled_reqs[0] = bs_src->throttled_reqs[0];
1975 bs_dest->throttled_reqs[1] = bs_src->throttled_reqs[1];
1976 bs_dest->io_limits_enabled = bs_src->io_limits_enabled;
1979 bs_dest->on_read_error = bs_src->on_read_error;
1980 bs_dest->on_write_error = bs_src->on_write_error;
1983 bs_dest->iostatus_enabled = bs_src->iostatus_enabled;
1984 bs_dest->iostatus = bs_src->iostatus;
1987 bs_dest->dirty_bitmaps = bs_src->dirty_bitmaps;
1989 /* reference count */
1990 bs_dest->refcnt = bs_src->refcnt;
1993 bs_dest->job = bs_src->job;
1995 /* keep the same entry in bdrv_states */
1996 pstrcpy(bs_dest->device_name, sizeof(bs_dest->device_name),
1997 bs_src->device_name);
1998 bs_dest->device_list = bs_src->device_list;
1999 memcpy(bs_dest->op_blockers, bs_src->op_blockers,
2000 sizeof(bs_dest->op_blockers));
2004 * Swap bs contents for two image chains while they are live,
2005 * while keeping required fields on the BlockDriverState that is
2006 * actually attached to a device.
2008 * This will modify the BlockDriverState fields, and swap contents
2009 * between bs_new and bs_old. Both bs_new and bs_old are modified.
2011 * bs_new is required to be anonymous.
2013 * This function does not create any image files.
2015 void bdrv_swap(BlockDriverState *bs_new, BlockDriverState *bs_old)
2017 BlockDriverState tmp;
2019 /* The code needs to swap the node_name but simply swapping node_list won't
2020 * work so first remove the nodes from the graph list, do the swap then
2021 * insert them back if needed.
2023 if (bs_new->node_name[0] != '\0') {
2024 QTAILQ_REMOVE(&graph_bdrv_states, bs_new, node_list);
2026 if (bs_old->node_name[0] != '\0') {
2027 QTAILQ_REMOVE(&graph_bdrv_states, bs_old, node_list);
2030 /* bs_new must be anonymous and shouldn't have anything fancy enabled */
2031 assert(bs_new->device_name[0] == '\0');
2032 assert(QLIST_EMPTY(&bs_new->dirty_bitmaps));
2033 assert(bs_new->job == NULL);
2034 assert(bs_new->dev == NULL);
2035 assert(bs_new->io_limits_enabled == false);
2036 assert(!throttle_have_timer(&bs_new->throttle_state));
2042 /* there are some fields that should not be swapped, move them back */
2043 bdrv_move_feature_fields(&tmp, bs_old);
2044 bdrv_move_feature_fields(bs_old, bs_new);
2045 bdrv_move_feature_fields(bs_new, &tmp);
2047 /* bs_new shouldn't be in bdrv_states even after the swap! */
2048 assert(bs_new->device_name[0] == '\0');
2050 /* Check a few fields that should remain attached to the device */
2051 assert(bs_new->dev == NULL);
2052 assert(bs_new->job == NULL);
2053 assert(bs_new->io_limits_enabled == false);
2054 assert(!throttle_have_timer(&bs_new->throttle_state));
2056 /* insert the nodes back into the graph node list if needed */
2057 if (bs_new->node_name[0] != '\0') {
2058 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs_new, node_list);
2060 if (bs_old->node_name[0] != '\0') {
2061 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs_old, node_list);
2064 bdrv_rebind(bs_new);
2065 bdrv_rebind(bs_old);
2069 * Add new bs contents at the top of an image chain while the chain is
2070 * live, while keeping required fields on the top layer.
2072 * This will modify the BlockDriverState fields, and swap contents
2073 * between bs_new and bs_top. Both bs_new and bs_top are modified.
2075 * bs_new is required to be anonymous.
2077 * This function does not create any image files.
2079 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
2081 bdrv_swap(bs_new, bs_top);
2083 /* The contents of 'tmp' will become bs_top, as we are
2084 * swapping bs_new and bs_top contents. */
2085 bdrv_set_backing_hd(bs_top, bs_new);
2088 static void bdrv_delete(BlockDriverState *bs)
2092 assert(bdrv_op_blocker_is_empty(bs));
2093 assert(!bs->refcnt);
2094 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
2098 /* remove from list, if necessary */
2104 int bdrv_attach_dev(BlockDriverState *bs, void *dev)
2105 /* TODO change to DeviceState *dev when all users are qdevified */
2111 bdrv_iostatus_reset(bs);
2113 /* We're expecting I/O from the device so bump up coroutine pool size */
2114 qemu_coroutine_adjust_pool_size(COROUTINE_POOL_RESERVATION);
2118 /* TODO qdevified devices don't use this, remove when devices are qdevified */
2119 void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev)
2121 if (bdrv_attach_dev(bs, dev) < 0) {
2126 void bdrv_detach_dev(BlockDriverState *bs, void *dev)
2127 /* TODO change to DeviceState *dev when all users are qdevified */
2129 assert(bs->dev == dev);
2132 bs->dev_opaque = NULL;
2133 bs->guest_block_size = 512;
2134 qemu_coroutine_adjust_pool_size(-COROUTINE_POOL_RESERVATION);
2137 /* TODO change to return DeviceState * when all users are qdevified */
2138 void *bdrv_get_attached_dev(BlockDriverState *bs)
2143 void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops,
2147 bs->dev_opaque = opaque;
2150 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load)
2152 if (bs->dev_ops && bs->dev_ops->change_media_cb) {
2153 bool tray_was_closed = !bdrv_dev_is_tray_open(bs);
2154 bs->dev_ops->change_media_cb(bs->dev_opaque, load);
2155 if (tray_was_closed) {
2157 qapi_event_send_device_tray_moved(bdrv_get_device_name(bs),
2158 true, &error_abort);
2162 qapi_event_send_device_tray_moved(bdrv_get_device_name(bs),
2163 false, &error_abort);
2168 bool bdrv_dev_has_removable_media(BlockDriverState *bs)
2170 return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb);
2173 void bdrv_dev_eject_request(BlockDriverState *bs, bool force)
2175 if (bs->dev_ops && bs->dev_ops->eject_request_cb) {
2176 bs->dev_ops->eject_request_cb(bs->dev_opaque, force);
2180 bool bdrv_dev_is_tray_open(BlockDriverState *bs)
2182 if (bs->dev_ops && bs->dev_ops->is_tray_open) {
2183 return bs->dev_ops->is_tray_open(bs->dev_opaque);
2188 static void bdrv_dev_resize_cb(BlockDriverState *bs)
2190 if (bs->dev_ops && bs->dev_ops->resize_cb) {
2191 bs->dev_ops->resize_cb(bs->dev_opaque);
2195 bool bdrv_dev_is_medium_locked(BlockDriverState *bs)
2197 if (bs->dev_ops && bs->dev_ops->is_medium_locked) {
2198 return bs->dev_ops->is_medium_locked(bs->dev_opaque);
2204 * Run consistency checks on an image
2206 * Returns 0 if the check could be completed (it doesn't mean that the image is
2207 * free of errors) or -errno when an internal error occurred. The results of the
2208 * check are stored in res.
2210 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix)
2212 if (bs->drv->bdrv_check == NULL) {
2216 memset(res, 0, sizeof(*res));
2217 return bs->drv->bdrv_check(bs, res, fix);
2220 #define COMMIT_BUF_SECTORS 2048
2222 /* commit COW file into the raw image */
2223 int bdrv_commit(BlockDriverState *bs)
2225 BlockDriver *drv = bs->drv;
2226 int64_t sector, total_sectors, length, backing_length;
2227 int n, ro, open_flags;
2229 uint8_t *buf = NULL;
2230 char filename[PATH_MAX];
2235 if (!bs->backing_hd) {
2239 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT, NULL) ||
2240 bdrv_op_is_blocked(bs->backing_hd, BLOCK_OP_TYPE_COMMIT, NULL)) {
2244 ro = bs->backing_hd->read_only;
2245 /* Use pstrcpy (not strncpy): filename must be NUL-terminated. */
2246 pstrcpy(filename, sizeof(filename), bs->backing_hd->filename);
2247 open_flags = bs->backing_hd->open_flags;
2250 if (bdrv_reopen(bs->backing_hd, open_flags | BDRV_O_RDWR, NULL)) {
2255 length = bdrv_getlength(bs);
2261 backing_length = bdrv_getlength(bs->backing_hd);
2262 if (backing_length < 0) {
2263 ret = backing_length;
2267 /* If our top snapshot is larger than the backing file image,
2268 * grow the backing file image if possible. If not possible,
2269 * we must return an error */
2270 if (length > backing_length) {
2271 ret = bdrv_truncate(bs->backing_hd, length);
2277 total_sectors = length >> BDRV_SECTOR_BITS;
2279 /* qemu_try_blockalign() for bs will choose an alignment that works for
2280 * bs->backing_hd as well, so no need to compare the alignment manually. */
2281 buf = qemu_try_blockalign(bs, COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
2287 for (sector = 0; sector < total_sectors; sector += n) {
2288 ret = bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n);
2293 ret = bdrv_read(bs, sector, buf, n);
2298 ret = bdrv_write(bs->backing_hd, sector, buf, n);
2305 if (drv->bdrv_make_empty) {
2306 ret = drv->bdrv_make_empty(bs);
2314 * Make sure all data we wrote to the backing device is actually
2317 if (bs->backing_hd) {
2318 bdrv_flush(bs->backing_hd);
2326 /* ignoring error return here */
2327 bdrv_reopen(bs->backing_hd, open_flags & ~BDRV_O_RDWR, NULL);
2333 int bdrv_commit_all(void)
2335 BlockDriverState *bs;
2337 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
2338 AioContext *aio_context = bdrv_get_aio_context(bs);
2340 aio_context_acquire(aio_context);
2341 if (bs->drv && bs->backing_hd) {
2342 int ret = bdrv_commit(bs);
2344 aio_context_release(aio_context);
2348 aio_context_release(aio_context);
2354 * Remove an active request from the tracked requests list
2356 * This function should be called when a tracked request is completing.
2358 static void tracked_request_end(BdrvTrackedRequest *req)
2360 if (req->serialising) {
2361 req->bs->serialising_in_flight--;
2364 QLIST_REMOVE(req, list);
2365 qemu_co_queue_restart_all(&req->wait_queue);
2369 * Add an active request to the tracked requests list
2371 static void tracked_request_begin(BdrvTrackedRequest *req,
2372 BlockDriverState *bs,
2374 unsigned int bytes, bool is_write)
2376 *req = (BdrvTrackedRequest){
2380 .is_write = is_write,
2381 .co = qemu_coroutine_self(),
2382 .serialising = false,
2383 .overlap_offset = offset,
2384 .overlap_bytes = bytes,
2387 qemu_co_queue_init(&req->wait_queue);
2389 QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
2392 static void mark_request_serialising(BdrvTrackedRequest *req, uint64_t align)
2394 int64_t overlap_offset = req->offset & ~(align - 1);
2395 unsigned int overlap_bytes = ROUND_UP(req->offset + req->bytes, align)
2398 if (!req->serialising) {
2399 req->bs->serialising_in_flight++;
2400 req->serialising = true;
2403 req->overlap_offset = MIN(req->overlap_offset, overlap_offset);
2404 req->overlap_bytes = MAX(req->overlap_bytes, overlap_bytes);
2408 * Round a region to cluster boundaries
2410 void bdrv_round_to_clusters(BlockDriverState *bs,
2411 int64_t sector_num, int nb_sectors,
2412 int64_t *cluster_sector_num,
2413 int *cluster_nb_sectors)
2415 BlockDriverInfo bdi;
2417 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
2418 *cluster_sector_num = sector_num;
2419 *cluster_nb_sectors = nb_sectors;
2421 int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE;
2422 *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c);
2423 *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num +
2428 static int bdrv_get_cluster_size(BlockDriverState *bs)
2430 BlockDriverInfo bdi;
2433 ret = bdrv_get_info(bs, &bdi);
2434 if (ret < 0 || bdi.cluster_size == 0) {
2435 return bs->request_alignment;
2437 return bdi.cluster_size;
2441 static bool tracked_request_overlaps(BdrvTrackedRequest *req,
2442 int64_t offset, unsigned int bytes)
2445 if (offset >= req->overlap_offset + req->overlap_bytes) {
2449 if (req->overlap_offset >= offset + bytes) {
2455 static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self)
2457 BlockDriverState *bs = self->bs;
2458 BdrvTrackedRequest *req;
2460 bool waited = false;
2462 if (!bs->serialising_in_flight) {
2468 QLIST_FOREACH(req, &bs->tracked_requests, list) {
2469 if (req == self || (!req->serialising && !self->serialising)) {
2472 if (tracked_request_overlaps(req, self->overlap_offset,
2473 self->overlap_bytes))
2475 /* Hitting this means there was a reentrant request, for
2476 * example, a block driver issuing nested requests. This must
2477 * never happen since it means deadlock.
2479 assert(qemu_coroutine_self() != req->co);
2481 /* If the request is already (indirectly) waiting for us, or
2482 * will wait for us as soon as it wakes up, then just go on
2483 * (instead of producing a deadlock in the former case). */
2484 if (!req->waiting_for) {
2485 self->waiting_for = req;
2486 qemu_co_queue_wait(&req->wait_queue);
2487 self->waiting_for = NULL;
2502 * -EINVAL - backing format specified, but no file
2503 * -ENOSPC - can't update the backing file because no space is left in the
2505 * -ENOTSUP - format driver doesn't support changing the backing file
2507 int bdrv_change_backing_file(BlockDriverState *bs,
2508 const char *backing_file, const char *backing_fmt)
2510 BlockDriver *drv = bs->drv;
2513 /* Backing file format doesn't make sense without a backing file */
2514 if (backing_fmt && !backing_file) {
2518 if (drv->bdrv_change_backing_file != NULL) {
2519 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
2525 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
2526 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
2532 * Finds the image layer in the chain that has 'bs' as its backing file.
2534 * active is the current topmost image.
2536 * Returns NULL if bs is not found in active's image chain,
2537 * or if active == bs.
2539 * Returns the bottommost base image if bs == NULL.
2541 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
2542 BlockDriverState *bs)
2544 while (active && bs != active->backing_hd) {
2545 active = active->backing_hd;
2551 /* Given a BDS, searches for the base layer. */
2552 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
2554 return bdrv_find_overlay(bs, NULL);
2557 typedef struct BlkIntermediateStates {
2558 BlockDriverState *bs;
2559 QSIMPLEQ_ENTRY(BlkIntermediateStates) entry;
2560 } BlkIntermediateStates;
2564 * Drops images above 'base' up to and including 'top', and sets the image
2565 * above 'top' to have base as its backing file.
2567 * Requires that the overlay to 'top' is opened r/w, so that the backing file
2568 * information in 'bs' can be properly updated.
2570 * E.g., this will convert the following chain:
2571 * bottom <- base <- intermediate <- top <- active
2575 * bottom <- base <- active
2577 * It is allowed for bottom==base, in which case it converts:
2579 * base <- intermediate <- top <- active
2585 * If backing_file_str is non-NULL, it will be used when modifying top's
2586 * overlay image metadata.
2589 * if active == top, that is considered an error
2592 int bdrv_drop_intermediate(BlockDriverState *active, BlockDriverState *top,
2593 BlockDriverState *base, const char *backing_file_str)
2595 BlockDriverState *intermediate;
2596 BlockDriverState *base_bs = NULL;
2597 BlockDriverState *new_top_bs = NULL;
2598 BlkIntermediateStates *intermediate_state, *next;
2601 QSIMPLEQ_HEAD(states_to_delete, BlkIntermediateStates) states_to_delete;
2602 QSIMPLEQ_INIT(&states_to_delete);
2604 if (!top->drv || !base->drv) {
2608 new_top_bs = bdrv_find_overlay(active, top);
2610 if (new_top_bs == NULL) {
2611 /* we could not find the image above 'top', this is an error */
2615 /* special case of new_top_bs->backing_hd already pointing to base - nothing
2616 * to do, no intermediate images */
2617 if (new_top_bs->backing_hd == base) {
2624 /* now we will go down through the list, and add each BDS we find
2625 * into our deletion queue, until we hit the 'base'
2627 while (intermediate) {
2628 intermediate_state = g_malloc0(sizeof(BlkIntermediateStates));
2629 intermediate_state->bs = intermediate;
2630 QSIMPLEQ_INSERT_TAIL(&states_to_delete, intermediate_state, entry);
2632 if (intermediate->backing_hd == base) {
2633 base_bs = intermediate->backing_hd;
2636 intermediate = intermediate->backing_hd;
2638 if (base_bs == NULL) {
2639 /* something went wrong, we did not end at the base. safely
2640 * unravel everything, and exit with error */
2644 /* success - we can delete the intermediate states, and link top->base */
2645 backing_file_str = backing_file_str ? backing_file_str : base_bs->filename;
2646 ret = bdrv_change_backing_file(new_top_bs, backing_file_str,
2647 base_bs->drv ? base_bs->drv->format_name : "");
2651 bdrv_set_backing_hd(new_top_bs, base_bs);
2653 QSIMPLEQ_FOREACH_SAFE(intermediate_state, &states_to_delete, entry, next) {
2654 /* so that bdrv_close() does not recursively close the chain */
2655 bdrv_set_backing_hd(intermediate_state->bs, NULL);
2656 bdrv_unref(intermediate_state->bs);
2661 QSIMPLEQ_FOREACH_SAFE(intermediate_state, &states_to_delete, entry, next) {
2662 g_free(intermediate_state);
2668 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
2673 if (size > INT_MAX) {
2677 if (!bdrv_is_inserted(bs))
2683 len = bdrv_getlength(bs);
2688 if ((offset > len) || (len - offset < size))
2694 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
2697 if (nb_sectors < 0 || nb_sectors > INT_MAX / BDRV_SECTOR_SIZE) {
2701 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
2702 nb_sectors * BDRV_SECTOR_SIZE);
2705 typedef struct RwCo {
2706 BlockDriverState *bs;
2711 BdrvRequestFlags flags;
2714 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
2716 RwCo *rwco = opaque;
2718 if (!rwco->is_write) {
2719 rwco->ret = bdrv_co_do_preadv(rwco->bs, rwco->offset,
2720 rwco->qiov->size, rwco->qiov,
2723 rwco->ret = bdrv_co_do_pwritev(rwco->bs, rwco->offset,
2724 rwco->qiov->size, rwco->qiov,
2730 * Process a vectored synchronous request using coroutines
2732 static int bdrv_prwv_co(BlockDriverState *bs, int64_t offset,
2733 QEMUIOVector *qiov, bool is_write,
2734 BdrvRequestFlags flags)
2741 .is_write = is_write,
2747 * In sync call context, when the vcpu is blocked, this throttling timer
2748 * will not fire; so the I/O throttling function has to be disabled here
2749 * if it has been enabled.
2751 if (bs->io_limits_enabled) {
2752 fprintf(stderr, "Disabling I/O throttling on '%s' due "
2753 "to synchronous I/O.\n", bdrv_get_device_name(bs));
2754 bdrv_io_limits_disable(bs);
2757 if (qemu_in_coroutine()) {
2758 /* Fast-path if already in coroutine context */
2759 bdrv_rw_co_entry(&rwco);
2761 AioContext *aio_context = bdrv_get_aio_context(bs);
2763 co = qemu_coroutine_create(bdrv_rw_co_entry);
2764 qemu_coroutine_enter(co, &rwco);
2765 while (rwco.ret == NOT_DONE) {
2766 aio_poll(aio_context, true);
2773 * Process a synchronous request using coroutines
2775 static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf,
2776 int nb_sectors, bool is_write, BdrvRequestFlags flags)
2779 struct iovec iov = {
2780 .iov_base = (void *)buf,
2781 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
2784 if (nb_sectors < 0 || nb_sectors > INT_MAX / BDRV_SECTOR_SIZE) {
2788 qemu_iovec_init_external(&qiov, &iov, 1);
2789 return bdrv_prwv_co(bs, sector_num << BDRV_SECTOR_BITS,
2790 &qiov, is_write, flags);
2793 /* return < 0 if error. See bdrv_write() for the return codes */
2794 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
2795 uint8_t *buf, int nb_sectors)
2797 return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false, 0);
2800 /* Just like bdrv_read(), but with I/O throttling temporarily disabled */
2801 int bdrv_read_unthrottled(BlockDriverState *bs, int64_t sector_num,
2802 uint8_t *buf, int nb_sectors)
2807 enabled = bs->io_limits_enabled;
2808 bs->io_limits_enabled = false;
2809 ret = bdrv_read(bs, sector_num, buf, nb_sectors);
2810 bs->io_limits_enabled = enabled;
2814 /* Return < 0 if error. Important errors are:
2815 -EIO generic I/O error (may happen for all errors)
2816 -ENOMEDIUM No media inserted.
2817 -EINVAL Invalid sector number or nb_sectors
2818 -EACCES Trying to write a read-only device
2820 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
2821 const uint8_t *buf, int nb_sectors)
2823 return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true, 0);
2826 int bdrv_write_zeroes(BlockDriverState *bs, int64_t sector_num,
2827 int nb_sectors, BdrvRequestFlags flags)
2829 return bdrv_rw_co(bs, sector_num, NULL, nb_sectors, true,
2830 BDRV_REQ_ZERO_WRITE | flags);
2834 * Completely zero out a block device with the help of bdrv_write_zeroes.
2835 * The operation is sped up by checking the block status and only writing
2836 * zeroes to the device if they currently do not return zeroes. Optional
2837 * flags are passed through to bdrv_write_zeroes (e.g. BDRV_REQ_MAY_UNMAP).
2839 * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
2841 int bdrv_make_zero(BlockDriverState *bs, BdrvRequestFlags flags)
2843 int64_t target_sectors, ret, nb_sectors, sector_num = 0;
2846 target_sectors = bdrv_nb_sectors(bs);
2847 if (target_sectors < 0) {
2848 return target_sectors;
2852 nb_sectors = target_sectors - sector_num;
2853 if (nb_sectors <= 0) {
2856 if (nb_sectors > INT_MAX) {
2857 nb_sectors = INT_MAX;
2859 ret = bdrv_get_block_status(bs, sector_num, nb_sectors, &n);
2861 error_report("error getting block status at sector %" PRId64 ": %s",
2862 sector_num, strerror(-ret));
2865 if (ret & BDRV_BLOCK_ZERO) {
2869 ret = bdrv_write_zeroes(bs, sector_num, n, flags);
2871 error_report("error writing zeroes at sector %" PRId64 ": %s",
2872 sector_num, strerror(-ret));
2879 int bdrv_pread(BlockDriverState *bs, int64_t offset, void *buf, int bytes)
2882 struct iovec iov = {
2883 .iov_base = (void *)buf,
2892 qemu_iovec_init_external(&qiov, &iov, 1);
2893 ret = bdrv_prwv_co(bs, offset, &qiov, false, 0);
2901 int bdrv_pwritev(BlockDriverState *bs, int64_t offset, QEMUIOVector *qiov)
2905 ret = bdrv_prwv_co(bs, offset, qiov, true, 0);
2913 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
2914 const void *buf, int bytes)
2917 struct iovec iov = {
2918 .iov_base = (void *) buf,
2926 qemu_iovec_init_external(&qiov, &iov, 1);
2927 return bdrv_pwritev(bs, offset, &qiov);
2931 * Writes to the file and ensures that no writes are reordered across this
2932 * request (acts as a barrier)
2934 * Returns 0 on success, -errno in error cases.
2936 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
2937 const void *buf, int count)
2941 ret = bdrv_pwrite(bs, offset, buf, count);
2946 /* No flush needed for cache modes that already do it */
2947 if (bs->enable_write_cache) {
2954 static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs,
2955 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
2957 /* Perform I/O through a temporary buffer so that users who scribble over
2958 * their read buffer while the operation is in progress do not end up
2959 * modifying the image file. This is critical for zero-copy guest I/O
2960 * where anything might happen inside guest memory.
2962 void *bounce_buffer;
2964 BlockDriver *drv = bs->drv;
2966 QEMUIOVector bounce_qiov;
2967 int64_t cluster_sector_num;
2968 int cluster_nb_sectors;
2972 /* Cover entire cluster so no additional backing file I/O is required when
2973 * allocating cluster in the image file.
2975 bdrv_round_to_clusters(bs, sector_num, nb_sectors,
2976 &cluster_sector_num, &cluster_nb_sectors);
2978 trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors,
2979 cluster_sector_num, cluster_nb_sectors);
2981 iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE;
2982 iov.iov_base = bounce_buffer = qemu_try_blockalign(bs, iov.iov_len);
2983 if (bounce_buffer == NULL) {
2988 qemu_iovec_init_external(&bounce_qiov, &iov, 1);
2990 ret = drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors,
2996 if (drv->bdrv_co_write_zeroes &&
2997 buffer_is_zero(bounce_buffer, iov.iov_len)) {
2998 ret = bdrv_co_do_write_zeroes(bs, cluster_sector_num,
2999 cluster_nb_sectors, 0);
3001 /* This does not change the data on the disk, it is not necessary
3002 * to flush even in cache=writethrough mode.
3004 ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors,
3009 /* It might be okay to ignore write errors for guest requests. If this
3010 * is a deliberate copy-on-read then we don't want to ignore the error.
3011 * Simply report it in all cases.
3016 skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE;
3017 qemu_iovec_from_buf(qiov, 0, bounce_buffer + skip_bytes,
3018 nb_sectors * BDRV_SECTOR_SIZE);
3021 qemu_vfree(bounce_buffer);
3026 * Forwards an already correctly aligned request to the BlockDriver. This
3027 * handles copy on read and zeroing after EOF; any other features must be
3028 * implemented by the caller.
3030 static int coroutine_fn bdrv_aligned_preadv(BlockDriverState *bs,
3031 BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
3032 int64_t align, QEMUIOVector *qiov, int flags)
3034 BlockDriver *drv = bs->drv;
3037 int64_t sector_num = offset >> BDRV_SECTOR_BITS;
3038 unsigned int nb_sectors = bytes >> BDRV_SECTOR_BITS;
3040 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
3041 assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
3042 assert(!qiov || bytes == qiov->size);
3044 /* Handle Copy on Read and associated serialisation */
3045 if (flags & BDRV_REQ_COPY_ON_READ) {
3046 /* If we touch the same cluster it counts as an overlap. This
3047 * guarantees that allocating writes will be serialized and not race
3048 * with each other for the same cluster. For example, in copy-on-read
3049 * it ensures that the CoR read and write operations are atomic and
3050 * guest writes cannot interleave between them. */
3051 mark_request_serialising(req, bdrv_get_cluster_size(bs));
3054 wait_serialising_requests(req);
3056 if (flags & BDRV_REQ_COPY_ON_READ) {
3059 ret = bdrv_is_allocated(bs, sector_num, nb_sectors, &pnum);
3064 if (!ret || pnum != nb_sectors) {
3065 ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov);
3070 /* Forward the request to the BlockDriver */
3071 if (!(bs->zero_beyond_eof && bs->growable)) {
3072 ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
3074 /* Read zeros after EOF of growable BDSes */
3075 int64_t total_sectors, max_nb_sectors;
3077 total_sectors = bdrv_nb_sectors(bs);
3078 if (total_sectors < 0) {
3079 ret = total_sectors;
3083 max_nb_sectors = ROUND_UP(MAX(0, total_sectors - sector_num),
3084 align >> BDRV_SECTOR_BITS);
3085 if (max_nb_sectors > 0) {
3086 QEMUIOVector local_qiov;
3087 size_t local_sectors;
3089 max_nb_sectors = MIN(max_nb_sectors, SIZE_MAX / BDRV_SECTOR_BITS);
3090 local_sectors = MIN(max_nb_sectors, nb_sectors);
3092 qemu_iovec_init(&local_qiov, qiov->niov);
3093 qemu_iovec_concat(&local_qiov, qiov, 0,
3094 local_sectors * BDRV_SECTOR_SIZE);
3096 ret = drv->bdrv_co_readv(bs, sector_num, local_sectors,
3099 qemu_iovec_destroy(&local_qiov);
3104 /* Reading beyond end of file is supposed to produce zeroes */
3105 if (ret == 0 && total_sectors < sector_num + nb_sectors) {
3106 uint64_t offset = MAX(0, total_sectors - sector_num);
3107 uint64_t bytes = (sector_num + nb_sectors - offset) *
3109 qemu_iovec_memset(qiov, offset * BDRV_SECTOR_SIZE, 0, bytes);
3118 * Handle a read request in coroutine context
3120 static int coroutine_fn bdrv_co_do_preadv(BlockDriverState *bs,
3121 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
3122 BdrvRequestFlags flags)
3124 BlockDriver *drv = bs->drv;
3125 BdrvTrackedRequest req;
3127 /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
3128 uint64_t align = MAX(BDRV_SECTOR_SIZE, bs->request_alignment);
3129 uint8_t *head_buf = NULL;
3130 uint8_t *tail_buf = NULL;
3131 QEMUIOVector local_qiov;
3132 bool use_local_qiov = false;
3138 if (bdrv_check_byte_request(bs, offset, bytes)) {
3142 if (bs->copy_on_read) {
3143 flags |= BDRV_REQ_COPY_ON_READ;
3146 /* throttling disk I/O */
3147 if (bs->io_limits_enabled) {
3148 bdrv_io_limits_intercept(bs, bytes, false);
3151 /* Align read if necessary by padding qiov */
3152 if (offset & (align - 1)) {
3153 head_buf = qemu_blockalign(bs, align);
3154 qemu_iovec_init(&local_qiov, qiov->niov + 2);
3155 qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
3156 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
3157 use_local_qiov = true;
3159 bytes += offset & (align - 1);
3160 offset = offset & ~(align - 1);
3163 if ((offset + bytes) & (align - 1)) {
3164 if (!use_local_qiov) {
3165 qemu_iovec_init(&local_qiov, qiov->niov + 1);
3166 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
3167 use_local_qiov = true;
3169 tail_buf = qemu_blockalign(bs, align);
3170 qemu_iovec_add(&local_qiov, tail_buf,
3171 align - ((offset + bytes) & (align - 1)));
3173 bytes = ROUND_UP(bytes, align);
3176 tracked_request_begin(&req, bs, offset, bytes, false);
3177 ret = bdrv_aligned_preadv(bs, &req, offset, bytes, align,
3178 use_local_qiov ? &local_qiov : qiov,
3180 tracked_request_end(&req);
3182 if (use_local_qiov) {
3183 qemu_iovec_destroy(&local_qiov);
3184 qemu_vfree(head_buf);
3185 qemu_vfree(tail_buf);
3191 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
3192 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
3193 BdrvRequestFlags flags)
3195 if (nb_sectors < 0 || nb_sectors > (UINT_MAX >> BDRV_SECTOR_BITS)) {
3199 return bdrv_co_do_preadv(bs, sector_num << BDRV_SECTOR_BITS,
3200 nb_sectors << BDRV_SECTOR_BITS, qiov, flags);
3203 int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
3204 int nb_sectors, QEMUIOVector *qiov)
3206 trace_bdrv_co_readv(bs, sector_num, nb_sectors);
3208 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0);
3211 int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
3212 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
3214 trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors);
3216 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
3217 BDRV_REQ_COPY_ON_READ);
3220 /* if no limit is specified in the BlockLimits use a default
3221 * of 32768 512-byte sectors (16 MiB) per request.
3223 #define MAX_WRITE_ZEROES_DEFAULT 32768
3225 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
3226 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
3228 BlockDriver *drv = bs->drv;
3230 struct iovec iov = {0};
3233 int max_write_zeroes = bs->bl.max_write_zeroes ?
3234 bs->bl.max_write_zeroes : MAX_WRITE_ZEROES_DEFAULT;
3236 while (nb_sectors > 0 && !ret) {
3237 int num = nb_sectors;
3239 /* Align request. Block drivers can expect the "bulk" of the request
3242 if (bs->bl.write_zeroes_alignment
3243 && num > bs->bl.write_zeroes_alignment) {
3244 if (sector_num % bs->bl.write_zeroes_alignment != 0) {
3245 /* Make a small request up to the first aligned sector. */
3246 num = bs->bl.write_zeroes_alignment;
3247 num -= sector_num % bs->bl.write_zeroes_alignment;
3248 } else if ((sector_num + num) % bs->bl.write_zeroes_alignment != 0) {
3249 /* Shorten the request to the last aligned sector. num cannot
3250 * underflow because num > bs->bl.write_zeroes_alignment.
3252 num -= (sector_num + num) % bs->bl.write_zeroes_alignment;
3256 /* limit request size */
3257 if (num > max_write_zeroes) {
3258 num = max_write_zeroes;
3262 /* First try the efficient write zeroes operation */
3263 if (drv->bdrv_co_write_zeroes) {
3264 ret = drv->bdrv_co_write_zeroes(bs, sector_num, num, flags);
3267 if (ret == -ENOTSUP) {
3268 /* Fall back to bounce buffer if write zeroes is unsupported */
3269 iov.iov_len = num * BDRV_SECTOR_SIZE;
3270 if (iov.iov_base == NULL) {
3271 iov.iov_base = qemu_try_blockalign(bs, num * BDRV_SECTOR_SIZE);
3272 if (iov.iov_base == NULL) {
3276 memset(iov.iov_base, 0, num * BDRV_SECTOR_SIZE);
3278 qemu_iovec_init_external(&qiov, &iov, 1);
3280 ret = drv->bdrv_co_writev(bs, sector_num, num, &qiov);
3282 /* Keep bounce buffer around if it is big enough for all
3283 * all future requests.
3285 if (num < max_write_zeroes) {
3286 qemu_vfree(iov.iov_base);
3287 iov.iov_base = NULL;
3296 qemu_vfree(iov.iov_base);
3301 * Forwards an already correctly aligned write request to the BlockDriver.
3303 static int coroutine_fn bdrv_aligned_pwritev(BlockDriverState *bs,
3304 BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
3305 QEMUIOVector *qiov, int flags)
3307 BlockDriver *drv = bs->drv;
3311 int64_t sector_num = offset >> BDRV_SECTOR_BITS;
3312 unsigned int nb_sectors = bytes >> BDRV_SECTOR_BITS;
3314 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
3315 assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
3316 assert(!qiov || bytes == qiov->size);
3318 waited = wait_serialising_requests(req);
3319 assert(!waited || !req->serialising);
3320 assert(req->overlap_offset <= offset);
3321 assert(offset + bytes <= req->overlap_offset + req->overlap_bytes);
3323 ret = notifier_with_return_list_notify(&bs->before_write_notifiers, req);
3325 if (!ret && bs->detect_zeroes != BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF &&
3326 !(flags & BDRV_REQ_ZERO_WRITE) && drv->bdrv_co_write_zeroes &&
3327 qemu_iovec_is_zero(qiov)) {
3328 flags |= BDRV_REQ_ZERO_WRITE;
3329 if (bs->detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP) {
3330 flags |= BDRV_REQ_MAY_UNMAP;
3335 /* Do nothing, write notifier decided to fail this request */
3336 } else if (flags & BDRV_REQ_ZERO_WRITE) {
3337 BLKDBG_EVENT(bs, BLKDBG_PWRITEV_ZERO);
3338 ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors, flags);
3340 BLKDBG_EVENT(bs, BLKDBG_PWRITEV);
3341 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
3343 BLKDBG_EVENT(bs, BLKDBG_PWRITEV_DONE);
3345 if (ret == 0 && !bs->enable_write_cache) {
3346 ret = bdrv_co_flush(bs);
3349 bdrv_set_dirty(bs, sector_num, nb_sectors);
3351 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
3352 bs->wr_highest_sector = sector_num + nb_sectors - 1;
3354 if (bs->growable && ret >= 0) {
3355 bs->total_sectors = MAX(bs->total_sectors, sector_num + nb_sectors);
3362 * Handle a write request in coroutine context
3364 static int coroutine_fn bdrv_co_do_pwritev(BlockDriverState *bs,
3365 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
3366 BdrvRequestFlags flags)
3368 BdrvTrackedRequest req;
3369 /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
3370 uint64_t align = MAX(BDRV_SECTOR_SIZE, bs->request_alignment);
3371 uint8_t *head_buf = NULL;
3372 uint8_t *tail_buf = NULL;
3373 QEMUIOVector local_qiov;
3374 bool use_local_qiov = false;
3380 if (bs->read_only) {
3383 if (bdrv_check_byte_request(bs, offset, bytes)) {
3387 /* throttling disk I/O */
3388 if (bs->io_limits_enabled) {
3389 bdrv_io_limits_intercept(bs, bytes, true);
3393 * Align write if necessary by performing a read-modify-write cycle.
3394 * Pad qiov with the read parts and be sure to have a tracked request not
3395 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
3397 tracked_request_begin(&req, bs, offset, bytes, true);
3399 if (offset & (align - 1)) {
3400 QEMUIOVector head_qiov;
3401 struct iovec head_iov;
3403 mark_request_serialising(&req, align);
3404 wait_serialising_requests(&req);
3406 head_buf = qemu_blockalign(bs, align);
3407 head_iov = (struct iovec) {
3408 .iov_base = head_buf,
3411 qemu_iovec_init_external(&head_qiov, &head_iov, 1);
3413 BLKDBG_EVENT(bs, BLKDBG_PWRITEV_RMW_HEAD);
3414 ret = bdrv_aligned_preadv(bs, &req, offset & ~(align - 1), align,
3415 align, &head_qiov, 0);
3419 BLKDBG_EVENT(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD);
3421 qemu_iovec_init(&local_qiov, qiov->niov + 2);
3422 qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
3423 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
3424 use_local_qiov = true;
3426 bytes += offset & (align - 1);
3427 offset = offset & ~(align - 1);
3430 if ((offset + bytes) & (align - 1)) {
3431 QEMUIOVector tail_qiov;
3432 struct iovec tail_iov;
3436 mark_request_serialising(&req, align);
3437 waited = wait_serialising_requests(&req);
3438 assert(!waited || !use_local_qiov);
3440 tail_buf = qemu_blockalign(bs, align);
3441 tail_iov = (struct iovec) {
3442 .iov_base = tail_buf,
3445 qemu_iovec_init_external(&tail_qiov, &tail_iov, 1);
3447 BLKDBG_EVENT(bs, BLKDBG_PWRITEV_RMW_TAIL);
3448 ret = bdrv_aligned_preadv(bs, &req, (offset + bytes) & ~(align - 1), align,
3449 align, &tail_qiov, 0);
3453 BLKDBG_EVENT(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL);
3455 if (!use_local_qiov) {
3456 qemu_iovec_init(&local_qiov, qiov->niov + 1);
3457 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
3458 use_local_qiov = true;
3461 tail_bytes = (offset + bytes) & (align - 1);
3462 qemu_iovec_add(&local_qiov, tail_buf + tail_bytes, align - tail_bytes);
3464 bytes = ROUND_UP(bytes, align);
3467 ret = bdrv_aligned_pwritev(bs, &req, offset, bytes,
3468 use_local_qiov ? &local_qiov : qiov,
3472 tracked_request_end(&req);
3474 if (use_local_qiov) {
3475 qemu_iovec_destroy(&local_qiov);
3477 qemu_vfree(head_buf);
3478 qemu_vfree(tail_buf);
3483 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
3484 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
3485 BdrvRequestFlags flags)
3487 if (nb_sectors < 0 || nb_sectors > (INT_MAX >> BDRV_SECTOR_BITS)) {
3491 return bdrv_co_do_pwritev(bs, sector_num << BDRV_SECTOR_BITS,
3492 nb_sectors << BDRV_SECTOR_BITS, qiov, flags);
3495 int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
3496 int nb_sectors, QEMUIOVector *qiov)
3498 trace_bdrv_co_writev(bs, sector_num, nb_sectors);
3500 return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0);
3503 int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs,
3504 int64_t sector_num, int nb_sectors,
3505 BdrvRequestFlags flags)
3507 trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors, flags);
3509 if (!(bs->open_flags & BDRV_O_UNMAP)) {
3510 flags &= ~BDRV_REQ_MAY_UNMAP;
3513 return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL,
3514 BDRV_REQ_ZERO_WRITE | flags);
3518 * Truncate file to 'offset' bytes (needed only for file protocols)
3520 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
3522 BlockDriver *drv = bs->drv;
3526 if (!drv->bdrv_truncate)
3531 ret = drv->bdrv_truncate(bs, offset);
3533 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
3534 bdrv_dev_resize_cb(bs);
3540 * Length of a allocated file in bytes. Sparse files are counted by actual
3541 * allocated space. Return < 0 if error or unknown.
3543 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
3545 BlockDriver *drv = bs->drv;
3549 if (drv->bdrv_get_allocated_file_size) {
3550 return drv->bdrv_get_allocated_file_size(bs);
3553 return bdrv_get_allocated_file_size(bs->file);
3559 * Return number of sectors on success, -errno on error.
3561 int64_t bdrv_nb_sectors(BlockDriverState *bs)
3563 BlockDriver *drv = bs->drv;
3568 if (drv->has_variable_length) {
3569 int ret = refresh_total_sectors(bs, bs->total_sectors);
3574 return bs->total_sectors;
3578 * Return length in bytes on success, -errno on error.
3579 * The length is always a multiple of BDRV_SECTOR_SIZE.
3581 int64_t bdrv_getlength(BlockDriverState *bs)
3583 int64_t ret = bdrv_nb_sectors(bs);
3585 return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
3588 /* return 0 as number of sectors if no device present or error */
3589 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
3591 int64_t nb_sectors = bdrv_nb_sectors(bs);
3593 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
3596 void bdrv_set_on_error(BlockDriverState *bs, BlockdevOnError on_read_error,
3597 BlockdevOnError on_write_error)
3599 bs->on_read_error = on_read_error;
3600 bs->on_write_error = on_write_error;
3603 BlockdevOnError bdrv_get_on_error(BlockDriverState *bs, bool is_read)
3605 return is_read ? bs->on_read_error : bs->on_write_error;
3608 BlockErrorAction bdrv_get_error_action(BlockDriverState *bs, bool is_read, int error)
3610 BlockdevOnError on_err = is_read ? bs->on_read_error : bs->on_write_error;
3613 case BLOCKDEV_ON_ERROR_ENOSPC:
3614 return (error == ENOSPC) ?
3615 BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
3616 case BLOCKDEV_ON_ERROR_STOP:
3617 return BLOCK_ERROR_ACTION_STOP;
3618 case BLOCKDEV_ON_ERROR_REPORT:
3619 return BLOCK_ERROR_ACTION_REPORT;
3620 case BLOCKDEV_ON_ERROR_IGNORE:
3621 return BLOCK_ERROR_ACTION_IGNORE;
3627 /* This is done by device models because, while the block layer knows
3628 * about the error, it does not know whether an operation comes from
3629 * the device or the block layer (from a job, for example).
3631 void bdrv_error_action(BlockDriverState *bs, BlockErrorAction action,
3632 bool is_read, int error)
3636 if (action == BLOCK_ERROR_ACTION_STOP) {
3637 /* First set the iostatus, so that "info block" returns an iostatus
3638 * that matches the events raised so far (an additional error iostatus
3639 * is fine, but not a lost one).
3641 bdrv_iostatus_set_err(bs, error);
3643 /* Then raise the request to stop the VM and the event.
3644 * qemu_system_vmstop_request_prepare has two effects. First,
3645 * it ensures that the STOP event always comes after the
3646 * BLOCK_IO_ERROR event. Second, it ensures that even if management
3647 * can observe the STOP event and do a "cont" before the STOP
3648 * event is issued, the VM will not stop. In this case, vm_start()
3649 * also ensures that the STOP/RESUME pair of events is emitted.
3651 qemu_system_vmstop_request_prepare();
3652 qapi_event_send_block_io_error(bdrv_get_device_name(bs),
3653 is_read ? IO_OPERATION_TYPE_READ :
3654 IO_OPERATION_TYPE_WRITE,
3655 action, &error_abort);
3656 qemu_system_vmstop_request(RUN_STATE_IO_ERROR);
3658 qapi_event_send_block_io_error(bdrv_get_device_name(bs),
3659 is_read ? IO_OPERATION_TYPE_READ :
3660 IO_OPERATION_TYPE_WRITE,
3661 action, &error_abort);
3665 int bdrv_is_read_only(BlockDriverState *bs)
3667 return bs->read_only;
3670 int bdrv_is_sg(BlockDriverState *bs)
3675 int bdrv_enable_write_cache(BlockDriverState *bs)
3677 return bs->enable_write_cache;
3680 void bdrv_set_enable_write_cache(BlockDriverState *bs, bool wce)
3682 bs->enable_write_cache = wce;
3684 /* so a reopen() will preserve wce */
3686 bs->open_flags |= BDRV_O_CACHE_WB;
3688 bs->open_flags &= ~BDRV_O_CACHE_WB;
3692 int bdrv_is_encrypted(BlockDriverState *bs)
3694 if (bs->backing_hd && bs->backing_hd->encrypted)
3696 return bs->encrypted;
3699 int bdrv_key_required(BlockDriverState *bs)
3701 BlockDriverState *backing_hd = bs->backing_hd;
3703 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
3705 return (bs->encrypted && !bs->valid_key);
3708 int bdrv_set_key(BlockDriverState *bs, const char *key)
3711 if (bs->backing_hd && bs->backing_hd->encrypted) {
3712 ret = bdrv_set_key(bs->backing_hd, key);
3718 if (!bs->encrypted) {
3720 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
3723 ret = bs->drv->bdrv_set_key(bs, key);
3726 } else if (!bs->valid_key) {
3728 /* call the change callback now, we skipped it on open */
3729 bdrv_dev_change_media_cb(bs, true);
3734 const char *bdrv_get_format_name(BlockDriverState *bs)
3736 return bs->drv ? bs->drv->format_name : NULL;
3739 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
3744 const char **formats = NULL;
3746 QLIST_FOREACH(drv, &bdrv_drivers, list) {
3747 if (drv->format_name) {
3750 while (formats && i && !found) {
3751 found = !strcmp(formats[--i], drv->format_name);
3755 formats = g_realloc(formats, (count + 1) * sizeof(char *));
3756 formats[count++] = drv->format_name;
3757 it(opaque, drv->format_name);
3764 /* This function is to find block backend bs */
3765 BlockDriverState *bdrv_find(const char *name)
3767 BlockDriverState *bs;
3769 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
3770 if (!strcmp(name, bs->device_name)) {
3777 /* This function is to find a node in the bs graph */
3778 BlockDriverState *bdrv_find_node(const char *node_name)
3780 BlockDriverState *bs;
3784 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3785 if (!strcmp(node_name, bs->node_name)) {
3792 /* Put this QMP function here so it can access the static graph_bdrv_states. */
3793 BlockDeviceInfoList *bdrv_named_nodes_list(void)
3795 BlockDeviceInfoList *list, *entry;
3796 BlockDriverState *bs;
3799 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3800 entry = g_malloc0(sizeof(*entry));
3801 entry->value = bdrv_block_device_info(bs);
3809 BlockDriverState *bdrv_lookup_bs(const char *device,
3810 const char *node_name,
3813 BlockDriverState *bs = NULL;
3816 bs = bdrv_find(device);
3824 bs = bdrv_find_node(node_name);
3831 error_setg(errp, "Cannot find device=%s nor node_name=%s",
3832 device ? device : "",
3833 node_name ? node_name : "");
3837 /* If 'base' is in the same chain as 'top', return true. Otherwise,
3838 * return false. If either argument is NULL, return false. */
3839 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
3841 while (top && top != base) {
3842 top = top->backing_hd;
3848 BlockDriverState *bdrv_next(BlockDriverState *bs)
3851 return QTAILQ_FIRST(&bdrv_states);
3853 return QTAILQ_NEXT(bs, device_list);
3856 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
3858 BlockDriverState *bs;
3860 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
3865 const char *bdrv_get_device_name(BlockDriverState *bs)
3867 return bs->device_name;
3870 int bdrv_get_flags(BlockDriverState *bs)
3872 return bs->open_flags;
3875 int bdrv_flush_all(void)
3877 BlockDriverState *bs;
3880 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
3881 AioContext *aio_context = bdrv_get_aio_context(bs);
3884 aio_context_acquire(aio_context);
3885 ret = bdrv_flush(bs);
3886 if (ret < 0 && !result) {
3889 aio_context_release(aio_context);
3895 int bdrv_has_zero_init_1(BlockDriverState *bs)
3900 int bdrv_has_zero_init(BlockDriverState *bs)
3904 /* If BS is a copy on write image, it is initialized to
3905 the contents of the base image, which may not be zeroes. */
3906 if (bs->backing_hd) {
3909 if (bs->drv->bdrv_has_zero_init) {
3910 return bs->drv->bdrv_has_zero_init(bs);
3917 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
3919 BlockDriverInfo bdi;
3921 if (bs->backing_hd) {
3925 if (bdrv_get_info(bs, &bdi) == 0) {
3926 return bdi.unallocated_blocks_are_zero;
3932 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
3934 BlockDriverInfo bdi;
3936 if (bs->backing_hd || !(bs->open_flags & BDRV_O_UNMAP)) {
3940 if (bdrv_get_info(bs, &bdi) == 0) {
3941 return bdi.can_write_zeroes_with_unmap;
3947 typedef struct BdrvCoGetBlockStatusData {
3948 BlockDriverState *bs;
3949 BlockDriverState *base;
3955 } BdrvCoGetBlockStatusData;
3958 * Returns true iff the specified sector is present in the disk image. Drivers
3959 * not implementing the functionality are assumed to not support backing files,
3960 * hence all their sectors are reported as allocated.
3962 * If 'sector_num' is beyond the end of the disk image the return value is 0
3963 * and 'pnum' is set to 0.
3965 * 'pnum' is set to the number of sectors (including and immediately following
3966 * the specified sector) that are known to be in the same
3967 * allocated/unallocated state.
3969 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
3970 * beyond the end of the disk image it will be clamped.
3972 static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs,
3974 int nb_sectors, int *pnum)
3976 int64_t total_sectors;
3980 total_sectors = bdrv_nb_sectors(bs);
3981 if (total_sectors < 0) {
3982 return total_sectors;
3985 if (sector_num >= total_sectors) {
3990 n = total_sectors - sector_num;
3991 if (n < nb_sectors) {
3995 if (!bs->drv->bdrv_co_get_block_status) {
3997 ret = BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED;
3998 if (bs->drv->protocol_name) {
3999 ret |= BDRV_BLOCK_OFFSET_VALID | (sector_num * BDRV_SECTOR_SIZE);
4004 ret = bs->drv->bdrv_co_get_block_status(bs, sector_num, nb_sectors, pnum);
4010 if (ret & BDRV_BLOCK_RAW) {
4011 assert(ret & BDRV_BLOCK_OFFSET_VALID);
4012 return bdrv_get_block_status(bs->file, ret >> BDRV_SECTOR_BITS,
4016 if (ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ZERO)) {
4017 ret |= BDRV_BLOCK_ALLOCATED;
4020 if (!(ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO)) {
4021 if (bdrv_unallocated_blocks_are_zero(bs)) {
4022 ret |= BDRV_BLOCK_ZERO;
4023 } else if (bs->backing_hd) {
4024 BlockDriverState *bs2 = bs->backing_hd;
4025 int64_t nb_sectors2 = bdrv_nb_sectors(bs2);
4026 if (nb_sectors2 >= 0 && sector_num >= nb_sectors2) {
4027 ret |= BDRV_BLOCK_ZERO;
4033 (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) &&
4034 (ret & BDRV_BLOCK_OFFSET_VALID)) {
4035 ret2 = bdrv_co_get_block_status(bs->file, ret >> BDRV_SECTOR_BITS,
4038 /* Ignore errors. This is just providing extra information, it
4039 * is useful but not necessary.
4041 ret |= (ret2 & BDRV_BLOCK_ZERO);
4048 /* Coroutine wrapper for bdrv_get_block_status() */
4049 static void coroutine_fn bdrv_get_block_status_co_entry(void *opaque)
4051 BdrvCoGetBlockStatusData *data = opaque;
4052 BlockDriverState *bs = data->bs;
4054 data->ret = bdrv_co_get_block_status(bs, data->sector_num, data->nb_sectors,
4060 * Synchronous wrapper around bdrv_co_get_block_status().
4062 * See bdrv_co_get_block_status() for details.
4064 int64_t bdrv_get_block_status(BlockDriverState *bs, int64_t sector_num,
4065 int nb_sectors, int *pnum)
4068 BdrvCoGetBlockStatusData data = {
4070 .sector_num = sector_num,
4071 .nb_sectors = nb_sectors,
4076 if (qemu_in_coroutine()) {
4077 /* Fast-path if already in coroutine context */
4078 bdrv_get_block_status_co_entry(&data);
4080 AioContext *aio_context = bdrv_get_aio_context(bs);
4082 co = qemu_coroutine_create(bdrv_get_block_status_co_entry);
4083 qemu_coroutine_enter(co, &data);
4084 while (!data.done) {
4085 aio_poll(aio_context, true);
4091 int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num,
4092 int nb_sectors, int *pnum)
4094 int64_t ret = bdrv_get_block_status(bs, sector_num, nb_sectors, pnum);
4098 return !!(ret & BDRV_BLOCK_ALLOCATED);
4102 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
4104 * Return true if the given sector is allocated in any image between
4105 * BASE and TOP (inclusive). BASE can be NULL to check if the given
4106 * sector is allocated in any image of the chain. Return false otherwise.
4108 * 'pnum' is set to the number of sectors (including and immediately following
4109 * the specified sector) that are known to be in the same
4110 * allocated/unallocated state.
4113 int bdrv_is_allocated_above(BlockDriverState *top,
4114 BlockDriverState *base,
4116 int nb_sectors, int *pnum)
4118 BlockDriverState *intermediate;
4119 int ret, n = nb_sectors;
4122 while (intermediate && intermediate != base) {
4124 ret = bdrv_is_allocated(intermediate, sector_num, nb_sectors,
4134 * [sector_num, nb_sectors] is unallocated on top but intermediate
4137 * [sector_num+x, nr_sectors] allocated.
4139 if (n > pnum_inter &&
4140 (intermediate == top ||
4141 sector_num + pnum_inter < intermediate->total_sectors)) {
4145 intermediate = intermediate->backing_hd;
4152 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
4154 if (bs->backing_hd && bs->backing_hd->encrypted)
4155 return bs->backing_file;
4156 else if (bs->encrypted)
4157 return bs->filename;
4162 void bdrv_get_backing_filename(BlockDriverState *bs,
4163 char *filename, int filename_size)
4165 pstrcpy(filename, filename_size, bs->backing_file);
4168 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
4169 const uint8_t *buf, int nb_sectors)
4171 BlockDriver *drv = bs->drv;
4174 if (!drv->bdrv_write_compressed)
4176 if (bdrv_check_request(bs, sector_num, nb_sectors))
4179 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
4181 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
4184 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
4186 BlockDriver *drv = bs->drv;
4189 if (!drv->bdrv_get_info)
4191 memset(bdi, 0, sizeof(*bdi));
4192 return drv->bdrv_get_info(bs, bdi);
4195 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs)
4197 BlockDriver *drv = bs->drv;
4198 if (drv && drv->bdrv_get_specific_info) {
4199 return drv->bdrv_get_specific_info(bs);
4204 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
4205 int64_t pos, int size)
4208 struct iovec iov = {
4209 .iov_base = (void *) buf,
4213 qemu_iovec_init_external(&qiov, &iov, 1);
4214 return bdrv_writev_vmstate(bs, &qiov, pos);
4217 int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
4219 BlockDriver *drv = bs->drv;
4223 } else if (drv->bdrv_save_vmstate) {
4224 return drv->bdrv_save_vmstate(bs, qiov, pos);
4225 } else if (bs->file) {
4226 return bdrv_writev_vmstate(bs->file, qiov, pos);
4232 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
4233 int64_t pos, int size)
4235 BlockDriver *drv = bs->drv;
4238 if (drv->bdrv_load_vmstate)
4239 return drv->bdrv_load_vmstate(bs, buf, pos, size);
4241 return bdrv_load_vmstate(bs->file, buf, pos, size);
4245 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
4247 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
4251 bs->drv->bdrv_debug_event(bs, event);
4254 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
4257 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
4261 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
4262 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
4268 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
4270 while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
4274 if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
4275 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
4281 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
4283 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
4287 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
4288 return bs->drv->bdrv_debug_resume(bs, tag);
4294 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
4296 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
4300 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
4301 return bs->drv->bdrv_debug_is_suspended(bs, tag);
4307 int bdrv_is_snapshot(BlockDriverState *bs)
4309 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
4312 /* backing_file can either be relative, or absolute, or a protocol. If it is
4313 * relative, it must be relative to the chain. So, passing in bs->filename
4314 * from a BDS as backing_file should not be done, as that may be relative to
4315 * the CWD rather than the chain. */
4316 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
4317 const char *backing_file)
4319 char *filename_full = NULL;
4320 char *backing_file_full = NULL;
4321 char *filename_tmp = NULL;
4322 int is_protocol = 0;
4323 BlockDriverState *curr_bs = NULL;
4324 BlockDriverState *retval = NULL;
4326 if (!bs || !bs->drv || !backing_file) {
4330 filename_full = g_malloc(PATH_MAX);
4331 backing_file_full = g_malloc(PATH_MAX);
4332 filename_tmp = g_malloc(PATH_MAX);
4334 is_protocol = path_has_protocol(backing_file);
4336 for (curr_bs = bs; curr_bs->backing_hd; curr_bs = curr_bs->backing_hd) {
4338 /* If either of the filename paths is actually a protocol, then
4339 * compare unmodified paths; otherwise make paths relative */
4340 if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
4341 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
4342 retval = curr_bs->backing_hd;
4346 /* If not an absolute filename path, make it relative to the current
4347 * image's filename path */
4348 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4351 /* We are going to compare absolute pathnames */
4352 if (!realpath(filename_tmp, filename_full)) {
4356 /* We need to make sure the backing filename we are comparing against
4357 * is relative to the current image filename (or absolute) */
4358 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4359 curr_bs->backing_file);
4361 if (!realpath(filename_tmp, backing_file_full)) {
4365 if (strcmp(backing_file_full, filename_full) == 0) {
4366 retval = curr_bs->backing_hd;
4372 g_free(filename_full);
4373 g_free(backing_file_full);
4374 g_free(filename_tmp);
4378 int bdrv_get_backing_file_depth(BlockDriverState *bs)
4384 if (!bs->backing_hd) {
4388 return 1 + bdrv_get_backing_file_depth(bs->backing_hd);
4391 /**************************************************************/
4394 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
4395 QEMUIOVector *qiov, int nb_sectors,
4396 BlockDriverCompletionFunc *cb, void *opaque)
4398 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
4400 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0,
4404 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
4405 QEMUIOVector *qiov, int nb_sectors,
4406 BlockDriverCompletionFunc *cb, void *opaque)
4408 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
4410 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0,
4414 BlockDriverAIOCB *bdrv_aio_write_zeroes(BlockDriverState *bs,
4415 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags,
4416 BlockDriverCompletionFunc *cb, void *opaque)
4418 trace_bdrv_aio_write_zeroes(bs, sector_num, nb_sectors, flags, opaque);
4420 return bdrv_co_aio_rw_vector(bs, sector_num, NULL, nb_sectors,
4421 BDRV_REQ_ZERO_WRITE | flags,
4426 typedef struct MultiwriteCB {
4431 BlockDriverCompletionFunc *cb;
4433 QEMUIOVector *free_qiov;
4437 static void multiwrite_user_cb(MultiwriteCB *mcb)
4441 for (i = 0; i < mcb->num_callbacks; i++) {
4442 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
4443 if (mcb->callbacks[i].free_qiov) {
4444 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
4446 g_free(mcb->callbacks[i].free_qiov);
4450 static void multiwrite_cb(void *opaque, int ret)
4452 MultiwriteCB *mcb = opaque;
4454 trace_multiwrite_cb(mcb, ret);
4456 if (ret < 0 && !mcb->error) {
4460 mcb->num_requests--;
4461 if (mcb->num_requests == 0) {
4462 multiwrite_user_cb(mcb);
4467 static int multiwrite_req_compare(const void *a, const void *b)
4469 const BlockRequest *req1 = a, *req2 = b;
4472 * Note that we can't simply subtract req2->sector from req1->sector
4473 * here as that could overflow the return value.
4475 if (req1->sector > req2->sector) {
4477 } else if (req1->sector < req2->sector) {
4485 * Takes a bunch of requests and tries to merge them. Returns the number of
4486 * requests that remain after merging.
4488 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
4489 int num_reqs, MultiwriteCB *mcb)
4493 // Sort requests by start sector
4494 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
4496 // Check if adjacent requests touch the same clusters. If so, combine them,
4497 // filling up gaps with zero sectors.
4499 for (i = 1; i < num_reqs; i++) {
4501 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
4503 // Handle exactly sequential writes and overlapping writes.
4504 if (reqs[i].sector <= oldreq_last) {
4508 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
4514 QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
4515 qemu_iovec_init(qiov,
4516 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
4518 // Add the first request to the merged one. If the requests are
4519 // overlapping, drop the last sectors of the first request.
4520 size = (reqs[i].sector - reqs[outidx].sector) << 9;
4521 qemu_iovec_concat(qiov, reqs[outidx].qiov, 0, size);
4523 // We should need to add any zeros between the two requests
4524 assert (reqs[i].sector <= oldreq_last);
4526 // Add the second request
4527 qemu_iovec_concat(qiov, reqs[i].qiov, 0, reqs[i].qiov->size);
4529 reqs[outidx].nb_sectors = qiov->size >> 9;
4530 reqs[outidx].qiov = qiov;
4532 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
4535 reqs[outidx].sector = reqs[i].sector;
4536 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
4537 reqs[outidx].qiov = reqs[i].qiov;
4545 * Submit multiple AIO write requests at once.
4547 * On success, the function returns 0 and all requests in the reqs array have
4548 * been submitted. In error case this function returns -1, and any of the
4549 * requests may or may not be submitted yet. In particular, this means that the
4550 * callback will be called for some of the requests, for others it won't. The
4551 * caller must check the error field of the BlockRequest to wait for the right
4552 * callbacks (if error != 0, no callback will be called).
4554 * The implementation may modify the contents of the reqs array, e.g. to merge
4555 * requests. However, the fields opaque and error are left unmodified as they
4556 * are used to signal failure for a single request to the caller.
4558 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
4563 /* don't submit writes if we don't have a medium */
4564 if (bs->drv == NULL) {
4565 for (i = 0; i < num_reqs; i++) {
4566 reqs[i].error = -ENOMEDIUM;
4571 if (num_reqs == 0) {
4575 // Create MultiwriteCB structure
4576 mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
4577 mcb->num_requests = 0;
4578 mcb->num_callbacks = num_reqs;
4580 for (i = 0; i < num_reqs; i++) {
4581 mcb->callbacks[i].cb = reqs[i].cb;
4582 mcb->callbacks[i].opaque = reqs[i].opaque;
4585 // Check for mergable requests
4586 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
4588 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
4590 /* Run the aio requests. */
4591 mcb->num_requests = num_reqs;
4592 for (i = 0; i < num_reqs; i++) {
4593 bdrv_co_aio_rw_vector(bs, reqs[i].sector, reqs[i].qiov,
4594 reqs[i].nb_sectors, reqs[i].flags,
4602 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
4604 acb->aiocb_info->cancel(acb);
4607 /**************************************************************/
4608 /* async block device emulation */
4610 typedef struct BlockDriverAIOCBSync {
4611 BlockDriverAIOCB common;
4614 /* vector translation state */
4618 } BlockDriverAIOCBSync;
4620 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
4622 BlockDriverAIOCBSync *acb =
4623 container_of(blockacb, BlockDriverAIOCBSync, common);
4624 qemu_bh_delete(acb->bh);
4626 qemu_aio_release(acb);
4629 static const AIOCBInfo bdrv_em_aiocb_info = {
4630 .aiocb_size = sizeof(BlockDriverAIOCBSync),
4631 .cancel = bdrv_aio_cancel_em,
4634 static void bdrv_aio_bh_cb(void *opaque)
4636 BlockDriverAIOCBSync *acb = opaque;
4638 if (!acb->is_write && acb->ret >= 0) {
4639 qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
4641 qemu_vfree(acb->bounce);
4642 acb->common.cb(acb->common.opaque, acb->ret);
4643 qemu_bh_delete(acb->bh);
4645 qemu_aio_release(acb);
4648 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
4652 BlockDriverCompletionFunc *cb,
4657 BlockDriverAIOCBSync *acb;
4659 acb = qemu_aio_get(&bdrv_em_aiocb_info, bs, cb, opaque);
4660 acb->is_write = is_write;
4662 acb->bounce = qemu_try_blockalign(bs, qiov->size);
4663 acb->bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_aio_bh_cb, acb);
4665 if (acb->bounce == NULL) {
4667 } else if (is_write) {
4668 qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
4669 acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
4671 acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
4674 qemu_bh_schedule(acb->bh);
4676 return &acb->common;
4679 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
4680 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
4681 BlockDriverCompletionFunc *cb, void *opaque)
4683 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
4686 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
4687 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
4688 BlockDriverCompletionFunc *cb, void *opaque)
4690 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
4694 typedef struct BlockDriverAIOCBCoroutine {
4695 BlockDriverAIOCB common;
4700 } BlockDriverAIOCBCoroutine;
4702 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb)
4704 AioContext *aio_context = bdrv_get_aio_context(blockacb->bs);
4705 BlockDriverAIOCBCoroutine *acb =
4706 container_of(blockacb, BlockDriverAIOCBCoroutine, common);
4711 aio_poll(aio_context, true);
4715 static const AIOCBInfo bdrv_em_co_aiocb_info = {
4716 .aiocb_size = sizeof(BlockDriverAIOCBCoroutine),
4717 .cancel = bdrv_aio_co_cancel_em,
4720 static void bdrv_co_em_bh(void *opaque)
4722 BlockDriverAIOCBCoroutine *acb = opaque;
4724 acb->common.cb(acb->common.opaque, acb->req.error);
4730 qemu_bh_delete(acb->bh);
4731 qemu_aio_release(acb);
4734 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
4735 static void coroutine_fn bdrv_co_do_rw(void *opaque)
4737 BlockDriverAIOCBCoroutine *acb = opaque;
4738 BlockDriverState *bs = acb->common.bs;
4740 if (!acb->is_write) {
4741 acb->req.error = bdrv_co_do_readv(bs, acb->req.sector,
4742 acb->req.nb_sectors, acb->req.qiov, acb->req.flags);
4744 acb->req.error = bdrv_co_do_writev(bs, acb->req.sector,
4745 acb->req.nb_sectors, acb->req.qiov, acb->req.flags);
4748 acb->bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_co_em_bh, acb);
4749 qemu_bh_schedule(acb->bh);
4752 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
4756 BdrvRequestFlags flags,
4757 BlockDriverCompletionFunc *cb,
4762 BlockDriverAIOCBCoroutine *acb;
4764 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
4765 acb->req.sector = sector_num;
4766 acb->req.nb_sectors = nb_sectors;
4767 acb->req.qiov = qiov;
4768 acb->req.flags = flags;
4769 acb->is_write = is_write;
4772 co = qemu_coroutine_create(bdrv_co_do_rw);
4773 qemu_coroutine_enter(co, acb);
4775 return &acb->common;
4778 static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
4780 BlockDriverAIOCBCoroutine *acb = opaque;
4781 BlockDriverState *bs = acb->common.bs;
4783 acb->req.error = bdrv_co_flush(bs);
4784 acb->bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_co_em_bh, acb);
4785 qemu_bh_schedule(acb->bh);
4788 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
4789 BlockDriverCompletionFunc *cb, void *opaque)
4791 trace_bdrv_aio_flush(bs, opaque);
4794 BlockDriverAIOCBCoroutine *acb;
4796 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
4799 co = qemu_coroutine_create(bdrv_aio_flush_co_entry);
4800 qemu_coroutine_enter(co, acb);
4802 return &acb->common;
4805 static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque)
4807 BlockDriverAIOCBCoroutine *acb = opaque;
4808 BlockDriverState *bs = acb->common.bs;
4810 acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors);
4811 acb->bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_co_em_bh, acb);
4812 qemu_bh_schedule(acb->bh);
4815 BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs,
4816 int64_t sector_num, int nb_sectors,
4817 BlockDriverCompletionFunc *cb, void *opaque)
4820 BlockDriverAIOCBCoroutine *acb;
4822 trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque);
4824 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
4825 acb->req.sector = sector_num;
4826 acb->req.nb_sectors = nb_sectors;
4828 co = qemu_coroutine_create(bdrv_aio_discard_co_entry);
4829 qemu_coroutine_enter(co, acb);
4831 return &acb->common;
4834 void bdrv_init(void)
4836 module_call_init(MODULE_INIT_BLOCK);
4839 void bdrv_init_with_whitelist(void)
4841 use_bdrv_whitelist = 1;
4845 void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs,
4846 BlockDriverCompletionFunc *cb, void *opaque)
4848 BlockDriverAIOCB *acb;
4850 acb = g_slice_alloc(aiocb_info->aiocb_size);
4851 acb->aiocb_info = aiocb_info;
4854 acb->opaque = opaque;
4858 void qemu_aio_release(void *p)
4860 BlockDriverAIOCB *acb = p;
4861 g_slice_free1(acb->aiocb_info->aiocb_size, acb);
4864 /**************************************************************/
4865 /* Coroutine block device emulation */
4867 typedef struct CoroutineIOCompletion {
4868 Coroutine *coroutine;
4870 } CoroutineIOCompletion;
4872 static void bdrv_co_io_em_complete(void *opaque, int ret)
4874 CoroutineIOCompletion *co = opaque;
4877 qemu_coroutine_enter(co->coroutine, NULL);
4880 static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num,
4881 int nb_sectors, QEMUIOVector *iov,
4884 CoroutineIOCompletion co = {
4885 .coroutine = qemu_coroutine_self(),
4887 BlockDriverAIOCB *acb;
4890 acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors,
4891 bdrv_co_io_em_complete, &co);
4893 acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors,
4894 bdrv_co_io_em_complete, &co);
4897 trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb);
4901 qemu_coroutine_yield();
4906 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
4907 int64_t sector_num, int nb_sectors,
4910 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false);
4913 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
4914 int64_t sector_num, int nb_sectors,
4917 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true);
4920 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
4922 RwCo *rwco = opaque;
4924 rwco->ret = bdrv_co_flush(rwco->bs);
4927 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
4931 if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
4935 /* Write back cached data to the OS even with cache=unsafe */
4936 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS);
4937 if (bs->drv->bdrv_co_flush_to_os) {
4938 ret = bs->drv->bdrv_co_flush_to_os(bs);
4944 /* But don't actually force it to the disk with cache=unsafe */
4945 if (bs->open_flags & BDRV_O_NO_FLUSH) {
4949 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK);
4950 if (bs->drv->bdrv_co_flush_to_disk) {
4951 ret = bs->drv->bdrv_co_flush_to_disk(bs);
4952 } else if (bs->drv->bdrv_aio_flush) {
4953 BlockDriverAIOCB *acb;
4954 CoroutineIOCompletion co = {
4955 .coroutine = qemu_coroutine_self(),
4958 acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
4962 qemu_coroutine_yield();
4967 * Some block drivers always operate in either writethrough or unsafe
4968 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
4969 * know how the server works (because the behaviour is hardcoded or
4970 * depends on server-side configuration), so we can't ensure that
4971 * everything is safe on disk. Returning an error doesn't work because
4972 * that would break guests even if the server operates in writethrough
4975 * Let's hope the user knows what he's doing.
4983 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
4984 * in the case of cache=unsafe, so there are no useless flushes.
4987 return bdrv_co_flush(bs->file);
4990 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
4992 Error *local_err = NULL;
4999 if (bs->drv->bdrv_invalidate_cache) {
5000 bs->drv->bdrv_invalidate_cache(bs, &local_err);
5001 } else if (bs->file) {
5002 bdrv_invalidate_cache(bs->file, &local_err);
5005 error_propagate(errp, local_err);
5009 ret = refresh_total_sectors(bs, bs->total_sectors);
5011 error_setg_errno(errp, -ret, "Could not refresh total sector count");
5016 void bdrv_invalidate_cache_all(Error **errp)
5018 BlockDriverState *bs;
5019 Error *local_err = NULL;
5021 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
5022 AioContext *aio_context = bdrv_get_aio_context(bs);
5024 aio_context_acquire(aio_context);
5025 bdrv_invalidate_cache(bs, &local_err);
5026 aio_context_release(aio_context);
5028 error_propagate(errp, local_err);
5034 void bdrv_clear_incoming_migration_all(void)
5036 BlockDriverState *bs;
5038 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
5039 AioContext *aio_context = bdrv_get_aio_context(bs);
5041 aio_context_acquire(aio_context);
5042 bs->open_flags = bs->open_flags & ~(BDRV_O_INCOMING);
5043 aio_context_release(aio_context);
5047 int bdrv_flush(BlockDriverState *bs)
5055 if (qemu_in_coroutine()) {
5056 /* Fast-path if already in coroutine context */
5057 bdrv_flush_co_entry(&rwco);
5059 AioContext *aio_context = bdrv_get_aio_context(bs);
5061 co = qemu_coroutine_create(bdrv_flush_co_entry);
5062 qemu_coroutine_enter(co, &rwco);
5063 while (rwco.ret == NOT_DONE) {
5064 aio_poll(aio_context, true);
5071 typedef struct DiscardCo {
5072 BlockDriverState *bs;
5077 static void coroutine_fn bdrv_discard_co_entry(void *opaque)
5079 DiscardCo *rwco = opaque;
5081 rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors);
5084 /* if no limit is specified in the BlockLimits use a default
5085 * of 32768 512-byte sectors (16 MiB) per request.
5087 #define MAX_DISCARD_DEFAULT 32768
5089 int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
5096 } else if (bdrv_check_request(bs, sector_num, nb_sectors)) {
5098 } else if (bs->read_only) {
5102 bdrv_reset_dirty(bs, sector_num, nb_sectors);
5104 /* Do nothing if disabled. */
5105 if (!(bs->open_flags & BDRV_O_UNMAP)) {
5109 if (!bs->drv->bdrv_co_discard && !bs->drv->bdrv_aio_discard) {
5113 max_discard = bs->bl.max_discard ? bs->bl.max_discard : MAX_DISCARD_DEFAULT;
5114 while (nb_sectors > 0) {
5116 int num = nb_sectors;
5119 if (bs->bl.discard_alignment &&
5120 num >= bs->bl.discard_alignment &&
5121 sector_num % bs->bl.discard_alignment) {
5122 if (num > bs->bl.discard_alignment) {
5123 num = bs->bl.discard_alignment;
5125 num -= sector_num % bs->bl.discard_alignment;
5128 /* limit request size */
5129 if (num > max_discard) {
5133 if (bs->drv->bdrv_co_discard) {
5134 ret = bs->drv->bdrv_co_discard(bs, sector_num, num);
5136 BlockDriverAIOCB *acb;
5137 CoroutineIOCompletion co = {
5138 .coroutine = qemu_coroutine_self(),
5141 acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors,
5142 bdrv_co_io_em_complete, &co);
5146 qemu_coroutine_yield();
5150 if (ret && ret != -ENOTSUP) {
5160 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
5165 .sector_num = sector_num,
5166 .nb_sectors = nb_sectors,
5170 if (qemu_in_coroutine()) {
5171 /* Fast-path if already in coroutine context */
5172 bdrv_discard_co_entry(&rwco);
5174 AioContext *aio_context = bdrv_get_aio_context(bs);
5176 co = qemu_coroutine_create(bdrv_discard_co_entry);
5177 qemu_coroutine_enter(co, &rwco);
5178 while (rwco.ret == NOT_DONE) {
5179 aio_poll(aio_context, true);
5186 /**************************************************************/
5187 /* removable device support */
5190 * Return TRUE if the media is present
5192 int bdrv_is_inserted(BlockDriverState *bs)
5194 BlockDriver *drv = bs->drv;
5198 if (!drv->bdrv_is_inserted)
5200 return drv->bdrv_is_inserted(bs);
5204 * Return whether the media changed since the last call to this
5205 * function, or -ENOTSUP if we don't know. Most drivers don't know.
5207 int bdrv_media_changed(BlockDriverState *bs)
5209 BlockDriver *drv = bs->drv;
5211 if (drv && drv->bdrv_media_changed) {
5212 return drv->bdrv_media_changed(bs);
5218 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
5220 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
5222 BlockDriver *drv = bs->drv;
5224 if (drv && drv->bdrv_eject) {
5225 drv->bdrv_eject(bs, eject_flag);
5228 if (bs->device_name[0] != '\0') {
5229 qapi_event_send_device_tray_moved(bdrv_get_device_name(bs),
5230 eject_flag, &error_abort);
5235 * Lock or unlock the media (if it is locked, the user won't be able
5236 * to eject it manually).
5238 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
5240 BlockDriver *drv = bs->drv;
5242 trace_bdrv_lock_medium(bs, locked);
5244 if (drv && drv->bdrv_lock_medium) {
5245 drv->bdrv_lock_medium(bs, locked);
5249 /* needed for generic scsi interface */
5251 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
5253 BlockDriver *drv = bs->drv;
5255 if (drv && drv->bdrv_ioctl)
5256 return drv->bdrv_ioctl(bs, req, buf);
5260 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
5261 unsigned long int req, void *buf,
5262 BlockDriverCompletionFunc *cb, void *opaque)
5264 BlockDriver *drv = bs->drv;
5266 if (drv && drv->bdrv_aio_ioctl)
5267 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
5271 void bdrv_set_guest_block_size(BlockDriverState *bs, int align)
5273 bs->guest_block_size = align;
5276 void *qemu_blockalign(BlockDriverState *bs, size_t size)
5278 return qemu_memalign(bdrv_opt_mem_align(bs), size);
5281 void *qemu_try_blockalign(BlockDriverState *bs, size_t size)
5283 size_t align = bdrv_opt_mem_align(bs);
5285 /* Ensure that NULL is never returned on success */
5291 return qemu_try_memalign(align, size);
5295 * Check if all memory in this vector is sector aligned.
5297 bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov)
5300 size_t alignment = bdrv_opt_mem_align(bs);
5302 for (i = 0; i < qiov->niov; i++) {
5303 if ((uintptr_t) qiov->iov[i].iov_base % alignment) {
5306 if (qiov->iov[i].iov_len % alignment) {
5314 BdrvDirtyBitmap *bdrv_create_dirty_bitmap(BlockDriverState *bs, int granularity,
5317 int64_t bitmap_size;
5318 BdrvDirtyBitmap *bitmap;
5320 assert((granularity & (granularity - 1)) == 0);
5322 granularity >>= BDRV_SECTOR_BITS;
5323 assert(granularity);
5324 bitmap_size = bdrv_nb_sectors(bs);
5325 if (bitmap_size < 0) {
5326 error_setg_errno(errp, -bitmap_size, "could not get length of device");
5327 errno = -bitmap_size;
5330 bitmap = g_malloc0(sizeof(BdrvDirtyBitmap));
5331 bitmap->bitmap = hbitmap_alloc(bitmap_size, ffs(granularity) - 1);
5332 QLIST_INSERT_HEAD(&bs->dirty_bitmaps, bitmap, list);
5336 void bdrv_release_dirty_bitmap(BlockDriverState *bs, BdrvDirtyBitmap *bitmap)
5338 BdrvDirtyBitmap *bm, *next;
5339 QLIST_FOREACH_SAFE(bm, &bs->dirty_bitmaps, list, next) {
5341 QLIST_REMOVE(bitmap, list);
5342 hbitmap_free(bitmap->bitmap);
5349 BlockDirtyInfoList *bdrv_query_dirty_bitmaps(BlockDriverState *bs)
5351 BdrvDirtyBitmap *bm;
5352 BlockDirtyInfoList *list = NULL;
5353 BlockDirtyInfoList **plist = &list;
5355 QLIST_FOREACH(bm, &bs->dirty_bitmaps, list) {
5356 BlockDirtyInfo *info = g_malloc0(sizeof(BlockDirtyInfo));
5357 BlockDirtyInfoList *entry = g_malloc0(sizeof(BlockDirtyInfoList));
5358 info->count = bdrv_get_dirty_count(bs, bm);
5360 ((int64_t) BDRV_SECTOR_SIZE << hbitmap_granularity(bm->bitmap));
5361 entry->value = info;
5363 plist = &entry->next;
5369 int bdrv_get_dirty(BlockDriverState *bs, BdrvDirtyBitmap *bitmap, int64_t sector)
5372 return hbitmap_get(bitmap->bitmap, sector);
5378 void bdrv_dirty_iter_init(BlockDriverState *bs,
5379 BdrvDirtyBitmap *bitmap, HBitmapIter *hbi)
5381 hbitmap_iter_init(hbi, bitmap->bitmap, 0);
5384 void bdrv_set_dirty(BlockDriverState *bs, int64_t cur_sector,
5387 BdrvDirtyBitmap *bitmap;
5388 QLIST_FOREACH(bitmap, &bs->dirty_bitmaps, list) {
5389 hbitmap_set(bitmap->bitmap, cur_sector, nr_sectors);
5393 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector, int nr_sectors)
5395 BdrvDirtyBitmap *bitmap;
5396 QLIST_FOREACH(bitmap, &bs->dirty_bitmaps, list) {
5397 hbitmap_reset(bitmap->bitmap, cur_sector, nr_sectors);
5401 int64_t bdrv_get_dirty_count(BlockDriverState *bs, BdrvDirtyBitmap *bitmap)
5403 return hbitmap_count(bitmap->bitmap);
5406 /* Get a reference to bs */
5407 void bdrv_ref(BlockDriverState *bs)
5412 /* Release a previously grabbed reference to bs.
5413 * If after releasing, reference count is zero, the BlockDriverState is
5415 void bdrv_unref(BlockDriverState *bs)
5420 assert(bs->refcnt > 0);
5421 if (--bs->refcnt == 0) {
5426 struct BdrvOpBlocker {
5428 QLIST_ENTRY(BdrvOpBlocker) list;
5431 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
5433 BdrvOpBlocker *blocker;
5434 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
5435 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
5436 blocker = QLIST_FIRST(&bs->op_blockers[op]);
5438 error_setg(errp, "Device '%s' is busy: %s",
5439 bs->device_name, error_get_pretty(blocker->reason));
5446 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
5448 BdrvOpBlocker *blocker;
5449 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
5451 blocker = g_malloc0(sizeof(BdrvOpBlocker));
5452 blocker->reason = reason;
5453 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
5456 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
5458 BdrvOpBlocker *blocker, *next;
5459 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
5460 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
5461 if (blocker->reason == reason) {
5462 QLIST_REMOVE(blocker, list);
5468 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
5471 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5472 bdrv_op_block(bs, i, reason);
5476 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
5479 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5480 bdrv_op_unblock(bs, i, reason);
5484 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
5488 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5489 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
5496 void bdrv_iostatus_enable(BlockDriverState *bs)
5498 bs->iostatus_enabled = true;
5499 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
5502 /* The I/O status is only enabled if the drive explicitly
5503 * enables it _and_ the VM is configured to stop on errors */
5504 bool bdrv_iostatus_is_enabled(const BlockDriverState *bs)
5506 return (bs->iostatus_enabled &&
5507 (bs->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
5508 bs->on_write_error == BLOCKDEV_ON_ERROR_STOP ||
5509 bs->on_read_error == BLOCKDEV_ON_ERROR_STOP));
5512 void bdrv_iostatus_disable(BlockDriverState *bs)
5514 bs->iostatus_enabled = false;
5517 void bdrv_iostatus_reset(BlockDriverState *bs)
5519 if (bdrv_iostatus_is_enabled(bs)) {
5520 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
5522 block_job_iostatus_reset(bs->job);
5527 void bdrv_iostatus_set_err(BlockDriverState *bs, int error)
5529 assert(bdrv_iostatus_is_enabled(bs));
5530 if (bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
5531 bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
5532 BLOCK_DEVICE_IO_STATUS_FAILED;
5537 bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes,
5538 enum BlockAcctType type)
5540 assert(type < BDRV_MAX_IOTYPE);
5542 cookie->bytes = bytes;
5543 cookie->start_time_ns = get_clock();
5544 cookie->type = type;
5548 bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie)
5550 assert(cookie->type < BDRV_MAX_IOTYPE);
5552 bs->nr_bytes[cookie->type] += cookie->bytes;
5553 bs->nr_ops[cookie->type]++;
5554 bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns;
5557 void bdrv_img_create(const char *filename, const char *fmt,
5558 const char *base_filename, const char *base_fmt,
5559 char *options, uint64_t img_size, int flags,
5560 Error **errp, bool quiet)
5562 QemuOptsList *create_opts = NULL;
5563 QemuOpts *opts = NULL;
5564 const char *backing_fmt, *backing_file;
5566 BlockDriver *drv, *proto_drv;
5567 BlockDriver *backing_drv = NULL;
5568 Error *local_err = NULL;
5571 /* Find driver and parse its options */
5572 drv = bdrv_find_format(fmt);
5574 error_setg(errp, "Unknown file format '%s'", fmt);
5578 proto_drv = bdrv_find_protocol(filename, true);
5580 error_setg(errp, "Unknown protocol '%s'", filename);
5584 create_opts = qemu_opts_append(create_opts, drv->create_opts);
5585 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
5587 /* Create parameter list with default values */
5588 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
5589 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size);
5591 /* Parse -o options */
5593 if (qemu_opts_do_parse(opts, options, NULL) != 0) {
5594 error_setg(errp, "Invalid options for file format '%s'", fmt);
5599 if (base_filename) {
5600 if (qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename)) {
5601 error_setg(errp, "Backing file not supported for file format '%s'",
5608 if (qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt)) {
5609 error_setg(errp, "Backing file format not supported for file "
5610 "format '%s'", fmt);
5615 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
5617 if (!strcmp(filename, backing_file)) {
5618 error_setg(errp, "Error: Trying to create an image with the "
5619 "same filename as the backing file");
5624 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
5626 backing_drv = bdrv_find_format(backing_fmt);
5628 error_setg(errp, "Unknown backing file format '%s'",
5634 // The size for the image must always be specified, with one exception:
5635 // If we are using a backing file, we can obtain the size from there
5636 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
5639 BlockDriverState *bs;
5643 /* backing files always opened read-only */
5645 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
5648 ret = bdrv_open(&bs, backing_file, NULL, NULL, back_flags,
5649 backing_drv, &local_err);
5651 error_setg_errno(errp, -ret, "Could not open '%s': %s",
5653 error_get_pretty(local_err));
5654 error_free(local_err);
5658 size = bdrv_getlength(bs);
5660 error_setg_errno(errp, -size, "Could not get size of '%s'",
5666 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size);
5670 error_setg(errp, "Image creation needs a size parameter");
5676 printf("Formatting '%s', fmt=%s ", filename, fmt);
5677 qemu_opts_print(opts);
5681 ret = bdrv_create(drv, filename, opts, &local_err);
5683 if (ret == -EFBIG) {
5684 /* This is generally a better message than whatever the driver would
5685 * deliver (especially because of the cluster_size_hint), since that
5686 * is most probably not much different from "image too large". */
5687 const char *cluster_size_hint = "";
5688 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
5689 cluster_size_hint = " (try using a larger cluster size)";
5691 error_setg(errp, "The image size is too large for file format '%s'"
5692 "%s", fmt, cluster_size_hint);
5693 error_free(local_err);
5698 qemu_opts_del(opts);
5699 qemu_opts_free(create_opts);
5701 error_propagate(errp, local_err);
5705 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
5707 return bs->aio_context;
5710 void bdrv_detach_aio_context(BlockDriverState *bs)
5716 if (bs->io_limits_enabled) {
5717 throttle_detach_aio_context(&bs->throttle_state);
5719 if (bs->drv->bdrv_detach_aio_context) {
5720 bs->drv->bdrv_detach_aio_context(bs);
5723 bdrv_detach_aio_context(bs->file);
5725 if (bs->backing_hd) {
5726 bdrv_detach_aio_context(bs->backing_hd);
5729 bs->aio_context = NULL;
5732 void bdrv_attach_aio_context(BlockDriverState *bs,
5733 AioContext *new_context)
5739 bs->aio_context = new_context;
5741 if (bs->backing_hd) {
5742 bdrv_attach_aio_context(bs->backing_hd, new_context);
5745 bdrv_attach_aio_context(bs->file, new_context);
5747 if (bs->drv->bdrv_attach_aio_context) {
5748 bs->drv->bdrv_attach_aio_context(bs, new_context);
5750 if (bs->io_limits_enabled) {
5751 throttle_attach_aio_context(&bs->throttle_state, new_context);
5755 void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context)
5757 bdrv_drain_all(); /* ensure there are no in-flight requests */
5759 bdrv_detach_aio_context(bs);
5761 /* This function executes in the old AioContext so acquire the new one in
5762 * case it runs in a different thread.
5764 aio_context_acquire(new_context);
5765 bdrv_attach_aio_context(bs, new_context);
5766 aio_context_release(new_context);
5769 void bdrv_add_before_write_notifier(BlockDriverState *bs,
5770 NotifierWithReturn *notifier)
5772 notifier_with_return_list_add(&bs->before_write_notifiers, notifier);
5775 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts)
5777 if (!bs->drv->bdrv_amend_options) {
5780 return bs->drv->bdrv_amend_options(bs, opts);
5783 /* This function will be called by the bdrv_recurse_is_first_non_filter method
5784 * of block filter and by bdrv_is_first_non_filter.
5785 * It is used to test if the given bs is the candidate or recurse more in the
5788 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
5789 BlockDriverState *candidate)
5791 /* return false if basic checks fails */
5792 if (!bs || !bs->drv) {
5796 /* the code reached a non block filter driver -> check if the bs is
5797 * the same as the candidate. It's the recursion termination condition.
5799 if (!bs->drv->is_filter) {
5800 return bs == candidate;
5802 /* Down this path the driver is a block filter driver */
5804 /* If the block filter recursion method is defined use it to recurse down
5807 if (bs->drv->bdrv_recurse_is_first_non_filter) {
5808 return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate);
5811 /* the driver is a block filter but don't allow to recurse -> return false
5816 /* This function checks if the candidate is the first non filter bs down it's
5817 * bs chain. Since we don't have pointers to parents it explore all bs chains
5818 * from the top. Some filters can choose not to pass down the recursion.
5820 bool bdrv_is_first_non_filter(BlockDriverState *candidate)
5822 BlockDriverState *bs;
5824 /* walk down the bs forest recursively */
5825 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
5828 /* try to recurse in this top level bs */
5829 perm = bdrv_recurse_is_first_non_filter(bs, candidate);
5831 /* candidate is the first non filter */
5840 BlockDriverState *check_to_replace_node(const char *node_name, Error **errp)
5842 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
5843 if (!to_replace_bs) {
5844 error_setg(errp, "Node name '%s' not found", node_name);
5848 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
5852 /* We don't want arbitrary node of the BDS chain to be replaced only the top
5853 * most non filter in order to prevent data corruption.
5854 * Another benefit is that this tests exclude backing files which are
5855 * blocked by the backing blockers.
5857 if (!bdrv_is_first_non_filter(to_replace_bs)) {
5858 error_setg(errp, "Only top most non filter can be replaced");
5862 return to_replace_bs;
5865 void bdrv_io_plug(BlockDriverState *bs)
5867 BlockDriver *drv = bs->drv;
5868 if (drv && drv->bdrv_io_plug) {
5869 drv->bdrv_io_plug(bs);
5870 } else if (bs->file) {
5871 bdrv_io_plug(bs->file);
5875 void bdrv_io_unplug(BlockDriverState *bs)
5877 BlockDriver *drv = bs->drv;
5878 if (drv && drv->bdrv_io_unplug) {
5879 drv->bdrv_io_unplug(bs);
5880 } else if (bs->file) {
5881 bdrv_io_unplug(bs->file);
5885 void bdrv_flush_io_queue(BlockDriverState *bs)
5887 BlockDriver *drv = bs->drv;
5888 if (drv && drv->bdrv_flush_io_queue) {
5889 drv->bdrv_flush_io_queue(bs);
5890 } else if (bs->file) {
5891 bdrv_flush_io_queue(bs->file);