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 "monitor/monitor.h"
28 #include "block/block_int.h"
29 #include "block/blockjob.h"
30 #include "qemu/module.h"
31 #include "qapi/qmp/qjson.h"
32 #include "sysemu/sysemu.h"
33 #include "qemu/notify.h"
34 #include "block/coroutine.h"
35 #include "block/qapi.h"
36 #include "qmp-commands.h"
37 #include "qemu/timer.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 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load);
61 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
62 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
63 BlockDriverCompletionFunc *cb, void *opaque);
64 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
65 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
66 BlockDriverCompletionFunc *cb, void *opaque);
67 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
68 int64_t sector_num, int nb_sectors,
70 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
71 int64_t sector_num, int nb_sectors,
73 static int coroutine_fn bdrv_co_do_preadv(BlockDriverState *bs,
74 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
75 BdrvRequestFlags flags);
76 static int coroutine_fn bdrv_co_do_pwritev(BlockDriverState *bs,
77 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
78 BdrvRequestFlags flags);
79 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
83 BdrvRequestFlags flags,
84 BlockDriverCompletionFunc *cb,
87 static void coroutine_fn bdrv_co_do_rw(void *opaque);
88 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
89 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags);
91 static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
92 QTAILQ_HEAD_INITIALIZER(bdrv_states);
94 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
95 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
97 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
98 QLIST_HEAD_INITIALIZER(bdrv_drivers);
100 /* If non-zero, use only whitelisted block drivers */
101 static int use_bdrv_whitelist;
104 static int is_windows_drive_prefix(const char *filename)
106 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
107 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
111 int is_windows_drive(const char *filename)
113 if (is_windows_drive_prefix(filename) &&
116 if (strstart(filename, "\\\\.\\", NULL) ||
117 strstart(filename, "//./", NULL))
123 /* throttling disk I/O limits */
124 void bdrv_set_io_limits(BlockDriverState *bs,
129 throttle_config(&bs->throttle_state, cfg);
131 for (i = 0; i < 2; i++) {
132 qemu_co_enter_next(&bs->throttled_reqs[i]);
136 /* this function drain all the throttled IOs */
137 static bool bdrv_start_throttled_reqs(BlockDriverState *bs)
139 bool drained = false;
140 bool enabled = bs->io_limits_enabled;
143 bs->io_limits_enabled = false;
145 for (i = 0; i < 2; i++) {
146 while (qemu_co_enter_next(&bs->throttled_reqs[i])) {
151 bs->io_limits_enabled = enabled;
156 void bdrv_io_limits_disable(BlockDriverState *bs)
158 bs->io_limits_enabled = false;
160 bdrv_start_throttled_reqs(bs);
162 throttle_destroy(&bs->throttle_state);
165 static void bdrv_throttle_read_timer_cb(void *opaque)
167 BlockDriverState *bs = opaque;
168 qemu_co_enter_next(&bs->throttled_reqs[0]);
171 static void bdrv_throttle_write_timer_cb(void *opaque)
173 BlockDriverState *bs = opaque;
174 qemu_co_enter_next(&bs->throttled_reqs[1]);
177 /* should be called before bdrv_set_io_limits if a limit is set */
178 void bdrv_io_limits_enable(BlockDriverState *bs)
180 assert(!bs->io_limits_enabled);
181 throttle_init(&bs->throttle_state,
183 bdrv_throttle_read_timer_cb,
184 bdrv_throttle_write_timer_cb,
186 bs->io_limits_enabled = true;
189 /* This function makes an IO wait if needed
191 * @nb_sectors: the number of sectors of the IO
192 * @is_write: is the IO a write
194 static void bdrv_io_limits_intercept(BlockDriverState *bs,
198 /* does this io must wait */
199 bool must_wait = throttle_schedule_timer(&bs->throttle_state, is_write);
201 /* if must wait or any request of this type throttled queue the IO */
203 !qemu_co_queue_empty(&bs->throttled_reqs[is_write])) {
204 qemu_co_queue_wait(&bs->throttled_reqs[is_write]);
207 /* the IO will be executed, do the accounting */
208 throttle_account(&bs->throttle_state, is_write, bytes);
211 /* if the next request must wait -> do nothing */
212 if (throttle_schedule_timer(&bs->throttle_state, is_write)) {
216 /* else queue next request for execution */
217 qemu_co_queue_next(&bs->throttled_reqs[is_write]);
220 size_t bdrv_opt_mem_align(BlockDriverState *bs)
222 if (!bs || !bs->drv) {
223 /* 4k should be on the safe side */
227 return bs->bl.opt_mem_alignment;
230 /* check if the path starts with "<protocol>:" */
231 static int path_has_protocol(const char *path)
236 if (is_windows_drive(path) ||
237 is_windows_drive_prefix(path)) {
240 p = path + strcspn(path, ":/\\");
242 p = path + strcspn(path, ":/");
248 int path_is_absolute(const char *path)
251 /* specific case for names like: "\\.\d:" */
252 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
255 return (*path == '/' || *path == '\\');
257 return (*path == '/');
261 /* if filename is absolute, just copy it to dest. Otherwise, build a
262 path to it by considering it is relative to base_path. URL are
264 void path_combine(char *dest, int dest_size,
265 const char *base_path,
266 const char *filename)
273 if (path_is_absolute(filename)) {
274 pstrcpy(dest, dest_size, filename);
276 p = strchr(base_path, ':');
281 p1 = strrchr(base_path, '/');
285 p2 = strrchr(base_path, '\\');
297 if (len > dest_size - 1)
299 memcpy(dest, base_path, len);
301 pstrcat(dest, dest_size, filename);
305 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz)
307 if (bs->backing_file[0] == '\0' || path_has_protocol(bs->backing_file)) {
308 pstrcpy(dest, sz, bs->backing_file);
310 path_combine(dest, sz, bs->filename, bs->backing_file);
314 void bdrv_register(BlockDriver *bdrv)
316 /* Block drivers without coroutine functions need emulation */
317 if (!bdrv->bdrv_co_readv) {
318 bdrv->bdrv_co_readv = bdrv_co_readv_em;
319 bdrv->bdrv_co_writev = bdrv_co_writev_em;
321 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
322 * the block driver lacks aio we need to emulate that too.
324 if (!bdrv->bdrv_aio_readv) {
325 /* add AIO emulation layer */
326 bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
327 bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
331 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
334 /* create a new block device (by default it is empty) */
335 BlockDriverState *bdrv_new(const char *device_name)
337 BlockDriverState *bs;
339 bs = g_malloc0(sizeof(BlockDriverState));
340 QLIST_INIT(&bs->dirty_bitmaps);
341 pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
342 if (device_name[0] != '\0') {
343 QTAILQ_INSERT_TAIL(&bdrv_states, bs, device_list);
345 bdrv_iostatus_disable(bs);
346 notifier_list_init(&bs->close_notifiers);
347 notifier_with_return_list_init(&bs->before_write_notifiers);
348 qemu_co_queue_init(&bs->throttled_reqs[0]);
349 qemu_co_queue_init(&bs->throttled_reqs[1]);
355 void bdrv_add_close_notifier(BlockDriverState *bs, Notifier *notify)
357 notifier_list_add(&bs->close_notifiers, notify);
360 BlockDriver *bdrv_find_format(const char *format_name)
363 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
364 if (!strcmp(drv1->format_name, format_name)) {
371 static int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
373 static const char *whitelist_rw[] = {
374 CONFIG_BDRV_RW_WHITELIST
376 static const char *whitelist_ro[] = {
377 CONFIG_BDRV_RO_WHITELIST
381 if (!whitelist_rw[0] && !whitelist_ro[0]) {
382 return 1; /* no whitelist, anything goes */
385 for (p = whitelist_rw; *p; p++) {
386 if (!strcmp(drv->format_name, *p)) {
391 for (p = whitelist_ro; *p; p++) {
392 if (!strcmp(drv->format_name, *p)) {
400 BlockDriver *bdrv_find_whitelisted_format(const char *format_name,
403 BlockDriver *drv = bdrv_find_format(format_name);
404 return drv && bdrv_is_whitelisted(drv, read_only) ? drv : NULL;
407 typedef struct CreateCo {
410 QEMUOptionParameter *options;
415 static void coroutine_fn bdrv_create_co_entry(void *opaque)
417 Error *local_err = NULL;
420 CreateCo *cco = opaque;
423 ret = cco->drv->bdrv_create(cco->filename, cco->options, &local_err);
425 error_propagate(&cco->err, local_err);
430 int bdrv_create(BlockDriver *drv, const char* filename,
431 QEMUOptionParameter *options, Error **errp)
438 .filename = g_strdup(filename),
444 if (!drv->bdrv_create) {
445 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
450 if (qemu_in_coroutine()) {
451 /* Fast-path if already in coroutine context */
452 bdrv_create_co_entry(&cco);
454 co = qemu_coroutine_create(bdrv_create_co_entry);
455 qemu_coroutine_enter(co, &cco);
456 while (cco.ret == NOT_DONE) {
464 error_propagate(errp, cco.err);
466 error_setg_errno(errp, -ret, "Could not create image");
471 g_free(cco.filename);
475 int bdrv_create_file(const char* filename, QEMUOptionParameter *options,
479 Error *local_err = NULL;
482 drv = bdrv_find_protocol(filename, true);
484 error_setg(errp, "Could not find protocol for file '%s'", filename);
488 ret = bdrv_create(drv, filename, options, &local_err);
490 error_propagate(errp, local_err);
495 int bdrv_refresh_limits(BlockDriverState *bs)
497 BlockDriver *drv = bs->drv;
499 memset(&bs->bl, 0, sizeof(bs->bl));
505 /* Take some limits from the children as a default */
507 bdrv_refresh_limits(bs->file);
508 bs->bl.opt_transfer_length = bs->file->bl.opt_transfer_length;
509 bs->bl.opt_mem_alignment = bs->file->bl.opt_mem_alignment;
511 bs->bl.opt_mem_alignment = 512;
514 if (bs->backing_hd) {
515 bdrv_refresh_limits(bs->backing_hd);
516 bs->bl.opt_transfer_length =
517 MAX(bs->bl.opt_transfer_length,
518 bs->backing_hd->bl.opt_transfer_length);
519 bs->bl.opt_mem_alignment =
520 MAX(bs->bl.opt_mem_alignment,
521 bs->backing_hd->bl.opt_mem_alignment);
524 /* Then let the driver override it */
525 if (drv->bdrv_refresh_limits) {
526 return drv->bdrv_refresh_limits(bs);
533 * Create a uniquely-named empty temporary file.
534 * Return 0 upon success, otherwise a negative errno value.
536 int get_tmp_filename(char *filename, int size)
539 char temp_dir[MAX_PATH];
540 /* GetTempFileName requires that its output buffer (4th param)
541 have length MAX_PATH or greater. */
542 assert(size >= MAX_PATH);
543 return (GetTempPath(MAX_PATH, temp_dir)
544 && GetTempFileName(temp_dir, "qem", 0, filename)
545 ? 0 : -GetLastError());
549 tmpdir = getenv("TMPDIR");
553 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
556 fd = mkstemp(filename);
560 if (close(fd) != 0) {
569 * Detect host devices. By convention, /dev/cdrom[N] is always
570 * recognized as a host CDROM.
572 static BlockDriver *find_hdev_driver(const char *filename)
574 int score_max = 0, score;
575 BlockDriver *drv = NULL, *d;
577 QLIST_FOREACH(d, &bdrv_drivers, list) {
578 if (d->bdrv_probe_device) {
579 score = d->bdrv_probe_device(filename);
580 if (score > score_max) {
590 BlockDriver *bdrv_find_protocol(const char *filename,
591 bool allow_protocol_prefix)
598 /* TODO Drivers without bdrv_file_open must be specified explicitly */
601 * XXX(hch): we really should not let host device detection
602 * override an explicit protocol specification, but moving this
603 * later breaks access to device names with colons in them.
604 * Thanks to the brain-dead persistent naming schemes on udev-
605 * based Linux systems those actually are quite common.
607 drv1 = find_hdev_driver(filename);
612 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
613 return bdrv_find_format("file");
616 p = strchr(filename, ':');
619 if (len > sizeof(protocol) - 1)
620 len = sizeof(protocol) - 1;
621 memcpy(protocol, filename, len);
622 protocol[len] = '\0';
623 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
624 if (drv1->protocol_name &&
625 !strcmp(drv1->protocol_name, protocol)) {
632 static int find_image_format(BlockDriverState *bs, const char *filename,
633 BlockDriver **pdrv, Error **errp)
635 int score, score_max;
636 BlockDriver *drv1, *drv;
640 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
641 if (bs->sg || !bdrv_is_inserted(bs) || bdrv_getlength(bs) == 0) {
642 drv = bdrv_find_format("raw");
644 error_setg(errp, "Could not find raw image format");
651 ret = bdrv_pread(bs, 0, buf, sizeof(buf));
653 error_setg_errno(errp, -ret, "Could not read image for determining its "
661 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
662 if (drv1->bdrv_probe) {
663 score = drv1->bdrv_probe(buf, ret, filename);
664 if (score > score_max) {
671 error_setg(errp, "Could not determine image format: No compatible "
680 * Set the current 'total_sectors' value
682 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
684 BlockDriver *drv = bs->drv;
686 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
690 /* query actual device if possible, otherwise just trust the hint */
691 if (drv->bdrv_getlength) {
692 int64_t length = drv->bdrv_getlength(bs);
696 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
699 bs->total_sectors = hint;
704 * Set open flags for a given discard mode
706 * Return 0 on success, -1 if the discard mode was invalid.
708 int bdrv_parse_discard_flags(const char *mode, int *flags)
710 *flags &= ~BDRV_O_UNMAP;
712 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
714 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
715 *flags |= BDRV_O_UNMAP;
724 * Set open flags for a given cache mode
726 * Return 0 on success, -1 if the cache mode was invalid.
728 int bdrv_parse_cache_flags(const char *mode, int *flags)
730 *flags &= ~BDRV_O_CACHE_MASK;
732 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
733 *flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB;
734 } else if (!strcmp(mode, "directsync")) {
735 *flags |= BDRV_O_NOCACHE;
736 } else if (!strcmp(mode, "writeback")) {
737 *flags |= BDRV_O_CACHE_WB;
738 } else if (!strcmp(mode, "unsafe")) {
739 *flags |= BDRV_O_CACHE_WB;
740 *flags |= BDRV_O_NO_FLUSH;
741 } else if (!strcmp(mode, "writethrough")) {
742 /* this is the default */
751 * The copy-on-read flag is actually a reference count so multiple users may
752 * use the feature without worrying about clobbering its previous state.
753 * Copy-on-read stays enabled until all users have called to disable it.
755 void bdrv_enable_copy_on_read(BlockDriverState *bs)
760 void bdrv_disable_copy_on_read(BlockDriverState *bs)
762 assert(bs->copy_on_read > 0);
766 static int bdrv_open_flags(BlockDriverState *bs, int flags)
768 int open_flags = flags | BDRV_O_CACHE_WB;
770 /* The backing file of a temporary snapshot is read-only */
771 if (flags & BDRV_O_SNAPSHOT) {
772 open_flags &= ~BDRV_O_RDWR;
776 * Clear flags that are internal to the block layer before opening the
779 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
782 * Snapshots should be writable.
784 if (bs->is_temporary) {
785 open_flags |= BDRV_O_RDWR;
791 static int bdrv_assign_node_name(BlockDriverState *bs,
792 const char *node_name,
799 /* empty string node name is invalid */
800 if (node_name[0] == '\0') {
801 error_setg(errp, "Empty node name");
805 /* takes care of avoiding namespaces collisions */
806 if (bdrv_find(node_name)) {
807 error_setg(errp, "node-name=%s is conflicting with a device id",
812 /* takes care of avoiding duplicates node names */
813 if (bdrv_find_node(node_name)) {
814 error_setg(errp, "Duplicate node name");
818 /* copy node name into the bs and insert it into the graph list */
819 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
820 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
826 * Common part for opening disk images and files
828 * Removes all processed options from *options.
830 static int bdrv_open_common(BlockDriverState *bs, BlockDriverState *file,
831 QDict *options, int flags, BlockDriver *drv, Error **errp)
834 const char *filename;
835 const char *node_name = NULL;
836 Error *local_err = NULL;
839 assert(bs->file == NULL);
840 assert(options != NULL && bs->options != options);
843 filename = file->filename;
845 filename = qdict_get_try_str(options, "filename");
848 if (drv->bdrv_needs_filename && !filename) {
849 error_setg(errp, "The '%s' block driver requires a file name",
854 trace_bdrv_open_common(bs, filename ?: "", flags, drv->format_name);
856 node_name = qdict_get_try_str(options, "node-name");
857 ret = bdrv_assign_node_name(bs, node_name, errp);
861 qdict_del(options, "node-name");
863 /* bdrv_open() with directly using a protocol as drv. This layer is already
864 * opened, so assign it to bs (while file becomes a closed BlockDriverState)
865 * and return immediately. */
866 if (file != NULL && drv->bdrv_file_open) {
871 bs->open_flags = flags;
872 bs->guest_block_size = 512;
873 bs->request_alignment = 512;
874 bs->zero_beyond_eof = true;
875 open_flags = bdrv_open_flags(bs, flags);
876 bs->read_only = !(open_flags & BDRV_O_RDWR);
878 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
880 !bs->read_only && bdrv_is_whitelisted(drv, true)
881 ? "Driver '%s' can only be used for read-only devices"
882 : "Driver '%s' is not whitelisted",
887 assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
888 if (flags & BDRV_O_COPY_ON_READ) {
889 if (!bs->read_only) {
890 bdrv_enable_copy_on_read(bs);
892 error_setg(errp, "Can't use copy-on-read on read-only device");
897 if (filename != NULL) {
898 pstrcpy(bs->filename, sizeof(bs->filename), filename);
900 bs->filename[0] = '\0';
904 bs->opaque = g_malloc0(drv->instance_size);
906 bs->enable_write_cache = !!(flags & BDRV_O_CACHE_WB);
908 /* Open the image, either directly or using a protocol */
909 if (drv->bdrv_file_open) {
910 assert(file == NULL);
911 assert(!drv->bdrv_needs_filename || filename != NULL);
912 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
915 error_setg(errp, "Can't use '%s' as a block driver for the "
916 "protocol level", drv->format_name);
921 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
926 error_propagate(errp, local_err);
927 } else if (bs->filename[0]) {
928 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
930 error_setg_errno(errp, -ret, "Could not open image");
935 ret = refresh_total_sectors(bs, bs->total_sectors);
937 error_setg_errno(errp, -ret, "Could not refresh total sector count");
941 bdrv_refresh_limits(bs);
942 assert(bdrv_opt_mem_align(bs) != 0);
943 assert((bs->request_alignment != 0) || bs->sg);
946 if (bs->is_temporary) {
947 assert(bs->filename[0] != '\0');
948 unlink(bs->filename);
962 * Opens a file using a protocol (file, host_device, nbd, ...)
964 * options is an indirect pointer to a QDict of options to pass to the block
965 * drivers, or pointer to NULL for an empty set of options. If this function
966 * takes ownership of the QDict reference, it will set *options to NULL;
967 * otherwise, it will contain unused/unrecognized options after this function
968 * returns. Then, the caller is responsible for freeing it. If it intends to
969 * reuse the QDict, QINCREF() should be called beforehand.
971 static int bdrv_file_open(BlockDriverState *bs, const char *filename,
972 QDict **options, int flags, Error **errp)
976 bool parse_filename = false;
977 Error *local_err = NULL;
980 /* Fetch the file name from the options QDict if necessary */
982 filename = qdict_get_try_str(*options, "filename");
983 } else if (filename && !qdict_haskey(*options, "filename")) {
984 qdict_put(*options, "filename", qstring_from_str(filename));
985 parse_filename = true;
987 error_setg(errp, "Can't specify 'file' and 'filename' options at the "
993 /* Find the right block driver */
994 drvname = qdict_get_try_str(*options, "driver");
996 drv = bdrv_find_format(drvname);
998 error_setg(errp, "Unknown driver '%s'", drvname);
1000 qdict_del(*options, "driver");
1001 } else if (filename) {
1002 drv = bdrv_find_protocol(filename, parse_filename);
1004 error_setg(errp, "Unknown protocol");
1007 error_setg(errp, "Must specify either driver or file");
1012 /* errp has been set already */
1017 /* Parse the filename and open it */
1018 if (drv->bdrv_parse_filename && parse_filename) {
1019 drv->bdrv_parse_filename(filename, *options, &local_err);
1021 error_propagate(errp, local_err);
1026 if (!drv->bdrv_needs_filename) {
1027 qdict_del(*options, "filename");
1029 filename = qdict_get_str(*options, "filename");
1033 if (!drv->bdrv_file_open) {
1034 ret = bdrv_open(&bs, filename, NULL, *options, flags, drv, &local_err);
1037 ret = bdrv_open_common(bs, NULL, *options, flags, drv, &local_err);
1040 error_propagate(errp, local_err);
1052 * Opens the backing file for a BlockDriverState if not yet open
1054 * options is a QDict of options to pass to the block drivers, or NULL for an
1055 * empty set of options. The reference to the QDict is transferred to this
1056 * function (even on failure), so if the caller intends to reuse the dictionary,
1057 * it needs to use QINCREF() before calling bdrv_file_open.
1059 int bdrv_open_backing_file(BlockDriverState *bs, QDict *options, Error **errp)
1061 char backing_filename[PATH_MAX];
1062 int back_flags, ret;
1063 BlockDriver *back_drv = NULL;
1064 Error *local_err = NULL;
1066 if (bs->backing_hd != NULL) {
1071 /* NULL means an empty set of options */
1072 if (options == NULL) {
1073 options = qdict_new();
1076 bs->open_flags &= ~BDRV_O_NO_BACKING;
1077 if (qdict_haskey(options, "file.filename")) {
1078 backing_filename[0] = '\0';
1079 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
1083 bdrv_get_full_backing_filename(bs, backing_filename,
1084 sizeof(backing_filename));
1087 if (bs->backing_format[0] != '\0') {
1088 back_drv = bdrv_find_format(bs->backing_format);
1091 /* backing files always opened read-only */
1092 back_flags = bs->open_flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT |
1093 BDRV_O_COPY_ON_READ);
1095 assert(bs->backing_hd == NULL);
1096 ret = bdrv_open(&bs->backing_hd,
1097 *backing_filename ? backing_filename : NULL, NULL, options,
1098 back_flags, back_drv, &local_err);
1100 bs->backing_hd = NULL;
1101 bs->open_flags |= BDRV_O_NO_BACKING;
1102 error_setg(errp, "Could not open backing file: %s",
1103 error_get_pretty(local_err));
1104 error_free(local_err);
1108 if (bs->backing_hd->file) {
1109 pstrcpy(bs->backing_file, sizeof(bs->backing_file),
1110 bs->backing_hd->file->filename);
1113 /* Recalculate the BlockLimits with the backing file */
1114 bdrv_refresh_limits(bs);
1120 * Opens a disk image whose options are given as BlockdevRef in another block
1123 * If allow_none is true, no image will be opened if filename is false and no
1124 * BlockdevRef is given. *pbs will remain unchanged and 0 will be returned.
1126 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
1127 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
1128 * itself, all options starting with "${bdref_key}." are considered part of the
1131 * The BlockdevRef will be removed from the options QDict.
1133 * To conform with the behavior of bdrv_open(), *pbs has to be NULL.
1135 int bdrv_open_image(BlockDriverState **pbs, const char *filename,
1136 QDict *options, const char *bdref_key, int flags,
1137 bool allow_none, Error **errp)
1139 QDict *image_options;
1141 char *bdref_key_dot;
1142 const char *reference;
1145 assert(*pbs == NULL);
1147 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
1148 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
1149 g_free(bdref_key_dot);
1151 reference = qdict_get_try_str(options, bdref_key);
1152 if (!filename && !reference && !qdict_size(image_options)) {
1156 error_setg(errp, "A block device must be specified for \"%s\"",
1163 ret = bdrv_open(pbs, filename, reference, image_options, flags, NULL, errp);
1166 qdict_del(options, bdref_key);
1170 void bdrv_append_temp_snapshot(BlockDriverState *bs, Error **errp)
1172 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
1173 char tmp_filename[PATH_MAX + 1];
1176 BlockDriver *bdrv_qcow2;
1177 QEMUOptionParameter *create_options;
1178 QDict *snapshot_options;
1179 BlockDriverState *bs_snapshot;
1183 /* if snapshot, we create a temporary backing file and open it
1184 instead of opening 'filename' directly */
1186 /* Get the required size from the image */
1187 total_size = bdrv_getlength(bs) & BDRV_SECTOR_MASK;
1189 /* Create the temporary image */
1190 ret = get_tmp_filename(tmp_filename, sizeof(tmp_filename));
1192 error_setg_errno(errp, -ret, "Could not get temporary filename");
1196 bdrv_qcow2 = bdrv_find_format("qcow2");
1197 create_options = parse_option_parameters("", bdrv_qcow2->create_options,
1200 set_option_parameter_int(create_options, BLOCK_OPT_SIZE, total_size);
1202 ret = bdrv_create(bdrv_qcow2, tmp_filename, create_options, &local_err);
1203 free_option_parameters(create_options);
1205 error_setg_errno(errp, -ret, "Could not create temporary overlay "
1206 "'%s': %s", tmp_filename,
1207 error_get_pretty(local_err));
1208 error_free(local_err);
1212 /* Prepare a new options QDict for the temporary file */
1213 snapshot_options = qdict_new();
1214 qdict_put(snapshot_options, "file.driver",
1215 qstring_from_str("file"));
1216 qdict_put(snapshot_options, "file.filename",
1217 qstring_from_str(tmp_filename));
1219 bs_snapshot = bdrv_new("");
1220 bs_snapshot->is_temporary = 1;
1222 ret = bdrv_open(&bs_snapshot, NULL, NULL, snapshot_options,
1223 bs->open_flags & ~BDRV_O_SNAPSHOT, bdrv_qcow2, &local_err);
1225 error_propagate(errp, local_err);
1229 bdrv_append(bs_snapshot, bs);
1233 * Opens a disk image (raw, qcow2, vmdk, ...)
1235 * options is a QDict of options to pass to the block drivers, or NULL for an
1236 * empty set of options. The reference to the QDict belongs to the block layer
1237 * after the call (even on failure), so if the caller intends to reuse the
1238 * dictionary, it needs to use QINCREF() before calling bdrv_open.
1240 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
1241 * If it is not NULL, the referenced BDS will be reused.
1243 * The reference parameter may be used to specify an existing block device which
1244 * should be opened. If specified, neither options nor a filename may be given,
1245 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
1247 int bdrv_open(BlockDriverState **pbs, const char *filename,
1248 const char *reference, QDict *options, int flags,
1249 BlockDriver *drv, Error **errp)
1252 BlockDriverState *file = NULL, *bs;
1253 const char *drvname;
1254 Error *local_err = NULL;
1259 bool options_non_empty = options ? qdict_size(options) : false;
1263 error_setg(errp, "Cannot reuse an existing BDS when referencing "
1264 "another block device");
1268 if (filename || options_non_empty) {
1269 error_setg(errp, "Cannot reference an existing block device with "
1270 "additional options or a new filename");
1274 bs = bdrv_lookup_bs(reference, reference, errp);
1289 /* NULL means an empty set of options */
1290 if (options == NULL) {
1291 options = qdict_new();
1294 bs->options = options;
1295 options = qdict_clone_shallow(options);
1297 if (flags & BDRV_O_PROTOCOL) {
1299 ret = bdrv_file_open(bs, filename, &options, flags & ~BDRV_O_PROTOCOL,
1304 } else if (bs->drv) {
1305 goto close_and_fail;
1311 /* Open image file without format layer */
1312 if (flags & BDRV_O_RDWR) {
1313 flags |= BDRV_O_ALLOW_RDWR;
1316 assert(file == NULL);
1317 ret = bdrv_open_image(&file, filename, options, "file",
1318 bdrv_open_flags(bs, flags | BDRV_O_UNMAP) |
1319 BDRV_O_PROTOCOL, true, &local_err);
1321 goto unlink_and_fail;
1324 /* Find the right image format driver */
1325 drvname = qdict_get_try_str(options, "driver");
1327 drv = bdrv_find_format(drvname);
1328 qdict_del(options, "driver");
1330 error_setg(errp, "Invalid driver: '%s'", drvname);
1332 goto unlink_and_fail;
1338 ret = find_image_format(file, filename, &drv, &local_err);
1340 error_setg(errp, "Must specify either driver or file");
1342 goto unlink_and_fail;
1347 goto unlink_and_fail;
1350 /* Open the image */
1351 ret = bdrv_open_common(bs, file, options, flags, drv, &local_err);
1353 goto unlink_and_fail;
1356 if (file && (bs->file != file)) {
1361 /* If there is a backing file, use it */
1362 if ((flags & BDRV_O_NO_BACKING) == 0) {
1363 QDict *backing_options;
1365 qdict_extract_subqdict(options, &backing_options, "backing.");
1366 ret = bdrv_open_backing_file(bs, backing_options, &local_err);
1368 goto close_and_fail;
1372 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
1373 * temporary snapshot afterwards. */
1374 if (flags & BDRV_O_SNAPSHOT) {
1375 bdrv_append_temp_snapshot(bs, &local_err);
1377 error_propagate(errp, local_err);
1378 goto close_and_fail;
1384 /* Check if any unknown options were used */
1385 if (options && (qdict_size(options) != 0)) {
1386 const QDictEntry *entry = qdict_first(options);
1387 if (flags & BDRV_O_PROTOCOL) {
1388 error_setg(errp, "Block protocol '%s' doesn't support the option "
1389 "'%s'", drv->format_name, entry->key);
1391 error_setg(errp, "Block format '%s' used by device '%s' doesn't "
1392 "support the option '%s'", drv->format_name,
1393 bs->device_name, entry->key);
1397 goto close_and_fail;
1400 if (!bdrv_key_required(bs)) {
1401 bdrv_dev_change_media_cb(bs, true);
1402 } else if (!runstate_check(RUN_STATE_PRELAUNCH)
1403 && !runstate_check(RUN_STATE_INMIGRATE)
1404 && !runstate_check(RUN_STATE_PAUSED)) { /* HACK */
1406 "Guest must be stopped for opening of encrypted image");
1408 goto close_and_fail;
1419 if (bs->is_temporary) {
1423 QDECREF(bs->options);
1427 /* If *pbs is NULL, a new BDS has been created in this function and
1428 needs to be freed now. Otherwise, it does not need to be closed,
1429 since it has not really been opened yet. */
1433 error_propagate(errp, local_err);
1438 /* See fail path, but now the BDS has to be always closed */
1446 error_propagate(errp, local_err);
1451 typedef struct BlockReopenQueueEntry {
1453 BDRVReopenState state;
1454 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
1455 } BlockReopenQueueEntry;
1458 * Adds a BlockDriverState to a simple queue for an atomic, transactional
1459 * reopen of multiple devices.
1461 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
1462 * already performed, or alternatively may be NULL a new BlockReopenQueue will
1463 * be created and initialized. This newly created BlockReopenQueue should be
1464 * passed back in for subsequent calls that are intended to be of the same
1467 * bs is the BlockDriverState to add to the reopen queue.
1469 * flags contains the open flags for the associated bs
1471 * returns a pointer to bs_queue, which is either the newly allocated
1472 * bs_queue, or the existing bs_queue being used.
1475 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
1476 BlockDriverState *bs, int flags)
1480 BlockReopenQueueEntry *bs_entry;
1481 if (bs_queue == NULL) {
1482 bs_queue = g_new0(BlockReopenQueue, 1);
1483 QSIMPLEQ_INIT(bs_queue);
1487 bdrv_reopen_queue(bs_queue, bs->file, flags);
1490 bs_entry = g_new0(BlockReopenQueueEntry, 1);
1491 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
1493 bs_entry->state.bs = bs;
1494 bs_entry->state.flags = flags;
1500 * Reopen multiple BlockDriverStates atomically & transactionally.
1502 * The queue passed in (bs_queue) must have been built up previous
1503 * via bdrv_reopen_queue().
1505 * Reopens all BDS specified in the queue, with the appropriate
1506 * flags. All devices are prepared for reopen, and failure of any
1507 * device will cause all device changes to be abandonded, and intermediate
1510 * If all devices prepare successfully, then the changes are committed
1514 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
1517 BlockReopenQueueEntry *bs_entry, *next;
1518 Error *local_err = NULL;
1520 assert(bs_queue != NULL);
1524 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
1525 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
1526 error_propagate(errp, local_err);
1529 bs_entry->prepared = true;
1532 /* If we reach this point, we have success and just need to apply the
1535 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
1536 bdrv_reopen_commit(&bs_entry->state);
1542 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
1543 if (ret && bs_entry->prepared) {
1544 bdrv_reopen_abort(&bs_entry->state);
1553 /* Reopen a single BlockDriverState with the specified flags. */
1554 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
1557 Error *local_err = NULL;
1558 BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, bdrv_flags);
1560 ret = bdrv_reopen_multiple(queue, &local_err);
1561 if (local_err != NULL) {
1562 error_propagate(errp, local_err);
1569 * Prepares a BlockDriverState for reopen. All changes are staged in the
1570 * 'opaque' field of the BDRVReopenState, which is used and allocated by
1571 * the block driver layer .bdrv_reopen_prepare()
1573 * bs is the BlockDriverState to reopen
1574 * flags are the new open flags
1575 * queue is the reopen queue
1577 * Returns 0 on success, non-zero on error. On error errp will be set
1580 * On failure, bdrv_reopen_abort() will be called to clean up any data.
1581 * It is the responsibility of the caller to then call the abort() or
1582 * commit() for any other BDS that have been left in a prepare() state
1585 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
1589 Error *local_err = NULL;
1592 assert(reopen_state != NULL);
1593 assert(reopen_state->bs->drv != NULL);
1594 drv = reopen_state->bs->drv;
1596 /* if we are to stay read-only, do not allow permission change
1598 if (!(reopen_state->bs->open_flags & BDRV_O_ALLOW_RDWR) &&
1599 reopen_state->flags & BDRV_O_RDWR) {
1600 error_set(errp, QERR_DEVICE_IS_READ_ONLY,
1601 reopen_state->bs->device_name);
1606 ret = bdrv_flush(reopen_state->bs);
1608 error_set(errp, ERROR_CLASS_GENERIC_ERROR, "Error (%s) flushing drive",
1613 if (drv->bdrv_reopen_prepare) {
1614 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
1616 if (local_err != NULL) {
1617 error_propagate(errp, local_err);
1619 error_setg(errp, "failed while preparing to reopen image '%s'",
1620 reopen_state->bs->filename);
1625 /* It is currently mandatory to have a bdrv_reopen_prepare()
1626 * handler for each supported drv. */
1627 error_set(errp, QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
1628 drv->format_name, reopen_state->bs->device_name,
1629 "reopening of file");
1641 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
1642 * makes them final by swapping the staging BlockDriverState contents into
1643 * the active BlockDriverState contents.
1645 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
1649 assert(reopen_state != NULL);
1650 drv = reopen_state->bs->drv;
1651 assert(drv != NULL);
1653 /* If there are any driver level actions to take */
1654 if (drv->bdrv_reopen_commit) {
1655 drv->bdrv_reopen_commit(reopen_state);
1658 /* set BDS specific flags now */
1659 reopen_state->bs->open_flags = reopen_state->flags;
1660 reopen_state->bs->enable_write_cache = !!(reopen_state->flags &
1662 reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
1664 bdrv_refresh_limits(reopen_state->bs);
1668 * Abort the reopen, and delete and free the staged changes in
1671 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
1675 assert(reopen_state != NULL);
1676 drv = reopen_state->bs->drv;
1677 assert(drv != NULL);
1679 if (drv->bdrv_reopen_abort) {
1680 drv->bdrv_reopen_abort(reopen_state);
1685 void bdrv_close(BlockDriverState *bs)
1688 block_job_cancel_sync(bs->job);
1690 bdrv_drain_all(); /* complete I/O */
1692 bdrv_drain_all(); /* in case flush left pending I/O */
1693 notifier_list_notify(&bs->close_notifiers, bs);
1696 if (bs->backing_hd) {
1697 bdrv_unref(bs->backing_hd);
1698 bs->backing_hd = NULL;
1700 bs->drv->bdrv_close(bs);
1703 if (bs->is_temporary) {
1704 unlink(bs->filename);
1709 bs->copy_on_read = 0;
1710 bs->backing_file[0] = '\0';
1711 bs->backing_format[0] = '\0';
1712 bs->total_sectors = 0;
1717 bs->zero_beyond_eof = false;
1718 QDECREF(bs->options);
1721 if (bs->file != NULL) {
1722 bdrv_unref(bs->file);
1727 bdrv_dev_change_media_cb(bs, false);
1729 /*throttling disk I/O limits*/
1730 if (bs->io_limits_enabled) {
1731 bdrv_io_limits_disable(bs);
1735 void bdrv_close_all(void)
1737 BlockDriverState *bs;
1739 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
1744 /* Check if any requests are in-flight (including throttled requests) */
1745 static bool bdrv_requests_pending(BlockDriverState *bs)
1747 if (!QLIST_EMPTY(&bs->tracked_requests)) {
1750 if (!qemu_co_queue_empty(&bs->throttled_reqs[0])) {
1753 if (!qemu_co_queue_empty(&bs->throttled_reqs[1])) {
1756 if (bs->file && bdrv_requests_pending(bs->file)) {
1759 if (bs->backing_hd && bdrv_requests_pending(bs->backing_hd)) {
1765 static bool bdrv_requests_pending_all(void)
1767 BlockDriverState *bs;
1768 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
1769 if (bdrv_requests_pending(bs)) {
1777 * Wait for pending requests to complete across all BlockDriverStates
1779 * This function does not flush data to disk, use bdrv_flush_all() for that
1780 * after calling this function.
1782 * Note that completion of an asynchronous I/O operation can trigger any
1783 * number of other I/O operations on other devices---for example a coroutine
1784 * can be arbitrarily complex and a constant flow of I/O can come until the
1785 * coroutine is complete. Because of this, it is not possible to have a
1786 * function to drain a single device's I/O queue.
1788 void bdrv_drain_all(void)
1790 /* Always run first iteration so any pending completion BHs run */
1792 BlockDriverState *bs;
1795 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
1796 bdrv_start_throttled_reqs(bs);
1799 busy = bdrv_requests_pending_all();
1800 busy |= aio_poll(qemu_get_aio_context(), busy);
1804 /* make a BlockDriverState anonymous by removing from bdrv_state and
1805 * graph_bdrv_state list.
1806 Also, NULL terminate the device_name to prevent double remove */
1807 void bdrv_make_anon(BlockDriverState *bs)
1809 if (bs->device_name[0] != '\0') {
1810 QTAILQ_REMOVE(&bdrv_states, bs, device_list);
1812 bs->device_name[0] = '\0';
1813 if (bs->node_name[0] != '\0') {
1814 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
1816 bs->node_name[0] = '\0';
1819 static void bdrv_rebind(BlockDriverState *bs)
1821 if (bs->drv && bs->drv->bdrv_rebind) {
1822 bs->drv->bdrv_rebind(bs);
1826 static void bdrv_move_feature_fields(BlockDriverState *bs_dest,
1827 BlockDriverState *bs_src)
1829 /* move some fields that need to stay attached to the device */
1830 bs_dest->open_flags = bs_src->open_flags;
1833 bs_dest->dev_ops = bs_src->dev_ops;
1834 bs_dest->dev_opaque = bs_src->dev_opaque;
1835 bs_dest->dev = bs_src->dev;
1836 bs_dest->guest_block_size = bs_src->guest_block_size;
1837 bs_dest->copy_on_read = bs_src->copy_on_read;
1839 bs_dest->enable_write_cache = bs_src->enable_write_cache;
1841 /* i/o throttled req */
1842 memcpy(&bs_dest->throttle_state,
1843 &bs_src->throttle_state,
1844 sizeof(ThrottleState));
1845 bs_dest->throttled_reqs[0] = bs_src->throttled_reqs[0];
1846 bs_dest->throttled_reqs[1] = bs_src->throttled_reqs[1];
1847 bs_dest->io_limits_enabled = bs_src->io_limits_enabled;
1850 bs_dest->on_read_error = bs_src->on_read_error;
1851 bs_dest->on_write_error = bs_src->on_write_error;
1854 bs_dest->iostatus_enabled = bs_src->iostatus_enabled;
1855 bs_dest->iostatus = bs_src->iostatus;
1858 bs_dest->dirty_bitmaps = bs_src->dirty_bitmaps;
1860 /* reference count */
1861 bs_dest->refcnt = bs_src->refcnt;
1864 bs_dest->in_use = bs_src->in_use;
1865 bs_dest->job = bs_src->job;
1867 /* keep the same entry in bdrv_states */
1868 pstrcpy(bs_dest->device_name, sizeof(bs_dest->device_name),
1869 bs_src->device_name);
1870 bs_dest->device_list = bs_src->device_list;
1874 * Swap bs contents for two image chains while they are live,
1875 * while keeping required fields on the BlockDriverState that is
1876 * actually attached to a device.
1878 * This will modify the BlockDriverState fields, and swap contents
1879 * between bs_new and bs_old. Both bs_new and bs_old are modified.
1881 * bs_new is required to be anonymous.
1883 * This function does not create any image files.
1885 void bdrv_swap(BlockDriverState *bs_new, BlockDriverState *bs_old)
1887 BlockDriverState tmp;
1889 /* The code needs to swap the node_name but simply swapping node_list won't
1890 * work so first remove the nodes from the graph list, do the swap then
1891 * insert them back if needed.
1893 if (bs_new->node_name[0] != '\0') {
1894 QTAILQ_REMOVE(&graph_bdrv_states, bs_new, node_list);
1896 if (bs_old->node_name[0] != '\0') {
1897 QTAILQ_REMOVE(&graph_bdrv_states, bs_old, node_list);
1900 /* bs_new must be anonymous and shouldn't have anything fancy enabled */
1901 assert(bs_new->device_name[0] == '\0');
1902 assert(QLIST_EMPTY(&bs_new->dirty_bitmaps));
1903 assert(bs_new->job == NULL);
1904 assert(bs_new->dev == NULL);
1905 assert(bs_new->in_use == 0);
1906 assert(bs_new->io_limits_enabled == false);
1907 assert(!throttle_have_timer(&bs_new->throttle_state));
1913 /* there are some fields that should not be swapped, move them back */
1914 bdrv_move_feature_fields(&tmp, bs_old);
1915 bdrv_move_feature_fields(bs_old, bs_new);
1916 bdrv_move_feature_fields(bs_new, &tmp);
1918 /* bs_new shouldn't be in bdrv_states even after the swap! */
1919 assert(bs_new->device_name[0] == '\0');
1921 /* Check a few fields that should remain attached to the device */
1922 assert(bs_new->dev == NULL);
1923 assert(bs_new->job == NULL);
1924 assert(bs_new->in_use == 0);
1925 assert(bs_new->io_limits_enabled == false);
1926 assert(!throttle_have_timer(&bs_new->throttle_state));
1928 /* insert the nodes back into the graph node list if needed */
1929 if (bs_new->node_name[0] != '\0') {
1930 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs_new, node_list);
1932 if (bs_old->node_name[0] != '\0') {
1933 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs_old, node_list);
1936 bdrv_rebind(bs_new);
1937 bdrv_rebind(bs_old);
1941 * Add new bs contents at the top of an image chain while the chain is
1942 * live, while keeping required fields on the top layer.
1944 * This will modify the BlockDriverState fields, and swap contents
1945 * between bs_new and bs_top. Both bs_new and bs_top are modified.
1947 * bs_new is required to be anonymous.
1949 * This function does not create any image files.
1951 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
1953 bdrv_swap(bs_new, bs_top);
1955 /* The contents of 'tmp' will become bs_top, as we are
1956 * swapping bs_new and bs_top contents. */
1957 bs_top->backing_hd = bs_new;
1958 bs_top->open_flags &= ~BDRV_O_NO_BACKING;
1959 pstrcpy(bs_top->backing_file, sizeof(bs_top->backing_file),
1961 pstrcpy(bs_top->backing_format, sizeof(bs_top->backing_format),
1962 bs_new->drv ? bs_new->drv->format_name : "");
1965 static void bdrv_delete(BlockDriverState *bs)
1969 assert(!bs->in_use);
1970 assert(!bs->refcnt);
1971 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
1975 /* remove from list, if necessary */
1981 int bdrv_attach_dev(BlockDriverState *bs, void *dev)
1982 /* TODO change to DeviceState *dev when all users are qdevified */
1988 bdrv_iostatus_reset(bs);
1992 /* TODO qdevified devices don't use this, remove when devices are qdevified */
1993 void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev)
1995 if (bdrv_attach_dev(bs, dev) < 0) {
2000 void bdrv_detach_dev(BlockDriverState *bs, void *dev)
2001 /* TODO change to DeviceState *dev when all users are qdevified */
2003 assert(bs->dev == dev);
2006 bs->dev_opaque = NULL;
2007 bs->guest_block_size = 512;
2010 /* TODO change to return DeviceState * when all users are qdevified */
2011 void *bdrv_get_attached_dev(BlockDriverState *bs)
2016 void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops,
2020 bs->dev_opaque = opaque;
2023 void bdrv_emit_qmp_error_event(const BlockDriverState *bdrv,
2024 enum MonitorEvent ev,
2025 BlockErrorAction action, bool is_read)
2028 const char *action_str;
2031 case BDRV_ACTION_REPORT:
2032 action_str = "report";
2034 case BDRV_ACTION_IGNORE:
2035 action_str = "ignore";
2037 case BDRV_ACTION_STOP:
2038 action_str = "stop";
2044 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
2047 is_read ? "read" : "write");
2048 monitor_protocol_event(ev, data);
2050 qobject_decref(data);
2053 static void bdrv_emit_qmp_eject_event(BlockDriverState *bs, bool ejected)
2057 data = qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }",
2058 bdrv_get_device_name(bs), ejected);
2059 monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED, data);
2061 qobject_decref(data);
2064 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load)
2066 if (bs->dev_ops && bs->dev_ops->change_media_cb) {
2067 bool tray_was_closed = !bdrv_dev_is_tray_open(bs);
2068 bs->dev_ops->change_media_cb(bs->dev_opaque, load);
2069 if (tray_was_closed) {
2071 bdrv_emit_qmp_eject_event(bs, true);
2075 bdrv_emit_qmp_eject_event(bs, false);
2080 bool bdrv_dev_has_removable_media(BlockDriverState *bs)
2082 return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb);
2085 void bdrv_dev_eject_request(BlockDriverState *bs, bool force)
2087 if (bs->dev_ops && bs->dev_ops->eject_request_cb) {
2088 bs->dev_ops->eject_request_cb(bs->dev_opaque, force);
2092 bool bdrv_dev_is_tray_open(BlockDriverState *bs)
2094 if (bs->dev_ops && bs->dev_ops->is_tray_open) {
2095 return bs->dev_ops->is_tray_open(bs->dev_opaque);
2100 static void bdrv_dev_resize_cb(BlockDriverState *bs)
2102 if (bs->dev_ops && bs->dev_ops->resize_cb) {
2103 bs->dev_ops->resize_cb(bs->dev_opaque);
2107 bool bdrv_dev_is_medium_locked(BlockDriverState *bs)
2109 if (bs->dev_ops && bs->dev_ops->is_medium_locked) {
2110 return bs->dev_ops->is_medium_locked(bs->dev_opaque);
2116 * Run consistency checks on an image
2118 * Returns 0 if the check could be completed (it doesn't mean that the image is
2119 * free of errors) or -errno when an internal error occurred. The results of the
2120 * check are stored in res.
2122 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix)
2124 if (bs->drv->bdrv_check == NULL) {
2128 memset(res, 0, sizeof(*res));
2129 return bs->drv->bdrv_check(bs, res, fix);
2132 #define COMMIT_BUF_SECTORS 2048
2134 /* commit COW file into the raw image */
2135 int bdrv_commit(BlockDriverState *bs)
2137 BlockDriver *drv = bs->drv;
2138 int64_t sector, total_sectors, length, backing_length;
2139 int n, ro, open_flags;
2141 uint8_t *buf = NULL;
2142 char filename[PATH_MAX];
2147 if (!bs->backing_hd) {
2151 if (bdrv_in_use(bs) || bdrv_in_use(bs->backing_hd)) {
2155 ro = bs->backing_hd->read_only;
2156 /* Use pstrcpy (not strncpy): filename must be NUL-terminated. */
2157 pstrcpy(filename, sizeof(filename), bs->backing_hd->filename);
2158 open_flags = bs->backing_hd->open_flags;
2161 if (bdrv_reopen(bs->backing_hd, open_flags | BDRV_O_RDWR, NULL)) {
2166 length = bdrv_getlength(bs);
2172 backing_length = bdrv_getlength(bs->backing_hd);
2173 if (backing_length < 0) {
2174 ret = backing_length;
2178 /* If our top snapshot is larger than the backing file image,
2179 * grow the backing file image if possible. If not possible,
2180 * we must return an error */
2181 if (length > backing_length) {
2182 ret = bdrv_truncate(bs->backing_hd, length);
2188 total_sectors = length >> BDRV_SECTOR_BITS;
2189 buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
2191 for (sector = 0; sector < total_sectors; sector += n) {
2192 ret = bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n);
2197 ret = bdrv_read(bs, sector, buf, n);
2202 ret = bdrv_write(bs->backing_hd, sector, buf, n);
2209 if (drv->bdrv_make_empty) {
2210 ret = drv->bdrv_make_empty(bs);
2218 * Make sure all data we wrote to the backing device is actually
2221 if (bs->backing_hd) {
2222 bdrv_flush(bs->backing_hd);
2230 /* ignoring error return here */
2231 bdrv_reopen(bs->backing_hd, open_flags & ~BDRV_O_RDWR, NULL);
2237 int bdrv_commit_all(void)
2239 BlockDriverState *bs;
2241 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
2242 if (bs->drv && bs->backing_hd) {
2243 int ret = bdrv_commit(bs);
2253 * Remove an active request from the tracked requests list
2255 * This function should be called when a tracked request is completing.
2257 static void tracked_request_end(BdrvTrackedRequest *req)
2259 if (req->serialising) {
2260 req->bs->serialising_in_flight--;
2263 QLIST_REMOVE(req, list);
2264 qemu_co_queue_restart_all(&req->wait_queue);
2268 * Add an active request to the tracked requests list
2270 static void tracked_request_begin(BdrvTrackedRequest *req,
2271 BlockDriverState *bs,
2273 unsigned int bytes, bool is_write)
2275 *req = (BdrvTrackedRequest){
2279 .is_write = is_write,
2280 .co = qemu_coroutine_self(),
2281 .serialising = false,
2282 .overlap_offset = offset,
2283 .overlap_bytes = bytes,
2286 qemu_co_queue_init(&req->wait_queue);
2288 QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
2291 static void mark_request_serialising(BdrvTrackedRequest *req, uint64_t align)
2293 int64_t overlap_offset = req->offset & ~(align - 1);
2294 unsigned int overlap_bytes = ROUND_UP(req->offset + req->bytes, align)
2297 if (!req->serialising) {
2298 req->bs->serialising_in_flight++;
2299 req->serialising = true;
2302 req->overlap_offset = MIN(req->overlap_offset, overlap_offset);
2303 req->overlap_bytes = MAX(req->overlap_bytes, overlap_bytes);
2307 * Round a region to cluster boundaries
2309 void bdrv_round_to_clusters(BlockDriverState *bs,
2310 int64_t sector_num, int nb_sectors,
2311 int64_t *cluster_sector_num,
2312 int *cluster_nb_sectors)
2314 BlockDriverInfo bdi;
2316 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
2317 *cluster_sector_num = sector_num;
2318 *cluster_nb_sectors = nb_sectors;
2320 int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE;
2321 *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c);
2322 *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num +
2327 static int bdrv_get_cluster_size(BlockDriverState *bs)
2329 BlockDriverInfo bdi;
2332 ret = bdrv_get_info(bs, &bdi);
2333 if (ret < 0 || bdi.cluster_size == 0) {
2334 return bs->request_alignment;
2336 return bdi.cluster_size;
2340 static bool tracked_request_overlaps(BdrvTrackedRequest *req,
2341 int64_t offset, unsigned int bytes)
2344 if (offset >= req->overlap_offset + req->overlap_bytes) {
2348 if (req->overlap_offset >= offset + bytes) {
2354 static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self)
2356 BlockDriverState *bs = self->bs;
2357 BdrvTrackedRequest *req;
2359 bool waited = false;
2361 if (!bs->serialising_in_flight) {
2367 QLIST_FOREACH(req, &bs->tracked_requests, list) {
2368 if (req == self || (!req->serialising && !self->serialising)) {
2371 if (tracked_request_overlaps(req, self->overlap_offset,
2372 self->overlap_bytes))
2374 /* Hitting this means there was a reentrant request, for
2375 * example, a block driver issuing nested requests. This must
2376 * never happen since it means deadlock.
2378 assert(qemu_coroutine_self() != req->co);
2380 /* If the request is already (indirectly) waiting for us, or
2381 * will wait for us as soon as it wakes up, then just go on
2382 * (instead of producing a deadlock in the former case). */
2383 if (!req->waiting_for) {
2384 self->waiting_for = req;
2385 qemu_co_queue_wait(&req->wait_queue);
2386 self->waiting_for = NULL;
2401 * -EINVAL - backing format specified, but no file
2402 * -ENOSPC - can't update the backing file because no space is left in the
2404 * -ENOTSUP - format driver doesn't support changing the backing file
2406 int bdrv_change_backing_file(BlockDriverState *bs,
2407 const char *backing_file, const char *backing_fmt)
2409 BlockDriver *drv = bs->drv;
2412 /* Backing file format doesn't make sense without a backing file */
2413 if (backing_fmt && !backing_file) {
2417 if (drv->bdrv_change_backing_file != NULL) {
2418 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
2424 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
2425 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
2431 * Finds the image layer in the chain that has 'bs' as its backing file.
2433 * active is the current topmost image.
2435 * Returns NULL if bs is not found in active's image chain,
2436 * or if active == bs.
2438 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
2439 BlockDriverState *bs)
2441 BlockDriverState *overlay = NULL;
2442 BlockDriverState *intermediate;
2444 assert(active != NULL);
2447 /* if bs is the same as active, then by definition it has no overlay
2453 intermediate = active;
2454 while (intermediate->backing_hd) {
2455 if (intermediate->backing_hd == bs) {
2456 overlay = intermediate;
2459 intermediate = intermediate->backing_hd;
2465 typedef struct BlkIntermediateStates {
2466 BlockDriverState *bs;
2467 QSIMPLEQ_ENTRY(BlkIntermediateStates) entry;
2468 } BlkIntermediateStates;
2472 * Drops images above 'base' up to and including 'top', and sets the image
2473 * above 'top' to have base as its backing file.
2475 * Requires that the overlay to 'top' is opened r/w, so that the backing file
2476 * information in 'bs' can be properly updated.
2478 * E.g., this will convert the following chain:
2479 * bottom <- base <- intermediate <- top <- active
2483 * bottom <- base <- active
2485 * It is allowed for bottom==base, in which case it converts:
2487 * base <- intermediate <- top <- active
2494 * if active == top, that is considered an error
2497 int bdrv_drop_intermediate(BlockDriverState *active, BlockDriverState *top,
2498 BlockDriverState *base)
2500 BlockDriverState *intermediate;
2501 BlockDriverState *base_bs = NULL;
2502 BlockDriverState *new_top_bs = NULL;
2503 BlkIntermediateStates *intermediate_state, *next;
2506 QSIMPLEQ_HEAD(states_to_delete, BlkIntermediateStates) states_to_delete;
2507 QSIMPLEQ_INIT(&states_to_delete);
2509 if (!top->drv || !base->drv) {
2513 new_top_bs = bdrv_find_overlay(active, top);
2515 if (new_top_bs == NULL) {
2516 /* we could not find the image above 'top', this is an error */
2520 /* special case of new_top_bs->backing_hd already pointing to base - nothing
2521 * to do, no intermediate images */
2522 if (new_top_bs->backing_hd == base) {
2529 /* now we will go down through the list, and add each BDS we find
2530 * into our deletion queue, until we hit the 'base'
2532 while (intermediate) {
2533 intermediate_state = g_malloc0(sizeof(BlkIntermediateStates));
2534 intermediate_state->bs = intermediate;
2535 QSIMPLEQ_INSERT_TAIL(&states_to_delete, intermediate_state, entry);
2537 if (intermediate->backing_hd == base) {
2538 base_bs = intermediate->backing_hd;
2541 intermediate = intermediate->backing_hd;
2543 if (base_bs == NULL) {
2544 /* something went wrong, we did not end at the base. safely
2545 * unravel everything, and exit with error */
2549 /* success - we can delete the intermediate states, and link top->base */
2550 ret = bdrv_change_backing_file(new_top_bs, base_bs->filename,
2551 base_bs->drv ? base_bs->drv->format_name : "");
2555 new_top_bs->backing_hd = base_bs;
2557 bdrv_refresh_limits(new_top_bs);
2559 QSIMPLEQ_FOREACH_SAFE(intermediate_state, &states_to_delete, entry, next) {
2560 /* so that bdrv_close() does not recursively close the chain */
2561 intermediate_state->bs->backing_hd = NULL;
2562 bdrv_unref(intermediate_state->bs);
2567 QSIMPLEQ_FOREACH_SAFE(intermediate_state, &states_to_delete, entry, next) {
2568 g_free(intermediate_state);
2574 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
2579 if (!bdrv_is_inserted(bs))
2585 len = bdrv_getlength(bs);
2590 if ((offset > len) || (len - offset < size))
2596 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
2599 if (nb_sectors > INT_MAX / BDRV_SECTOR_SIZE) {
2603 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
2604 nb_sectors * BDRV_SECTOR_SIZE);
2607 typedef struct RwCo {
2608 BlockDriverState *bs;
2613 BdrvRequestFlags flags;
2616 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
2618 RwCo *rwco = opaque;
2620 if (!rwco->is_write) {
2621 rwco->ret = bdrv_co_do_preadv(rwco->bs, rwco->offset,
2622 rwco->qiov->size, rwco->qiov,
2625 rwco->ret = bdrv_co_do_pwritev(rwco->bs, rwco->offset,
2626 rwco->qiov->size, rwco->qiov,
2632 * Process a vectored synchronous request using coroutines
2634 static int bdrv_prwv_co(BlockDriverState *bs, int64_t offset,
2635 QEMUIOVector *qiov, bool is_write,
2636 BdrvRequestFlags flags)
2643 .is_write = is_write,
2649 * In sync call context, when the vcpu is blocked, this throttling timer
2650 * will not fire; so the I/O throttling function has to be disabled here
2651 * if it has been enabled.
2653 if (bs->io_limits_enabled) {
2654 fprintf(stderr, "Disabling I/O throttling on '%s' due "
2655 "to synchronous I/O.\n", bdrv_get_device_name(bs));
2656 bdrv_io_limits_disable(bs);
2659 if (qemu_in_coroutine()) {
2660 /* Fast-path if already in coroutine context */
2661 bdrv_rw_co_entry(&rwco);
2663 co = qemu_coroutine_create(bdrv_rw_co_entry);
2664 qemu_coroutine_enter(co, &rwco);
2665 while (rwco.ret == NOT_DONE) {
2673 * Process a synchronous request using coroutines
2675 static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf,
2676 int nb_sectors, bool is_write, BdrvRequestFlags flags)
2679 struct iovec iov = {
2680 .iov_base = (void *)buf,
2681 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
2684 qemu_iovec_init_external(&qiov, &iov, 1);
2685 return bdrv_prwv_co(bs, sector_num << BDRV_SECTOR_BITS,
2686 &qiov, is_write, flags);
2689 /* return < 0 if error. See bdrv_write() for the return codes */
2690 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
2691 uint8_t *buf, int nb_sectors)
2693 return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false, 0);
2696 /* Just like bdrv_read(), but with I/O throttling temporarily disabled */
2697 int bdrv_read_unthrottled(BlockDriverState *bs, int64_t sector_num,
2698 uint8_t *buf, int nb_sectors)
2703 enabled = bs->io_limits_enabled;
2704 bs->io_limits_enabled = false;
2705 ret = bdrv_read(bs, sector_num, buf, nb_sectors);
2706 bs->io_limits_enabled = enabled;
2710 /* Return < 0 if error. Important errors are:
2711 -EIO generic I/O error (may happen for all errors)
2712 -ENOMEDIUM No media inserted.
2713 -EINVAL Invalid sector number or nb_sectors
2714 -EACCES Trying to write a read-only device
2716 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
2717 const uint8_t *buf, int nb_sectors)
2719 return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true, 0);
2722 int bdrv_write_zeroes(BlockDriverState *bs, int64_t sector_num,
2723 int nb_sectors, BdrvRequestFlags flags)
2725 return bdrv_rw_co(bs, sector_num, NULL, nb_sectors, true,
2726 BDRV_REQ_ZERO_WRITE | flags);
2730 * Completely zero out a block device with the help of bdrv_write_zeroes.
2731 * The operation is sped up by checking the block status and only writing
2732 * zeroes to the device if they currently do not return zeroes. Optional
2733 * flags are passed through to bdrv_write_zeroes (e.g. BDRV_REQ_MAY_UNMAP).
2735 * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
2737 int bdrv_make_zero(BlockDriverState *bs, BdrvRequestFlags flags)
2739 int64_t target_size = bdrv_getlength(bs) / BDRV_SECTOR_SIZE;
2740 int64_t ret, nb_sectors, sector_num = 0;
2744 nb_sectors = target_size - sector_num;
2745 if (nb_sectors <= 0) {
2748 if (nb_sectors > INT_MAX) {
2749 nb_sectors = INT_MAX;
2751 ret = bdrv_get_block_status(bs, sector_num, nb_sectors, &n);
2753 error_report("error getting block status at sector %" PRId64 ": %s",
2754 sector_num, strerror(-ret));
2757 if (ret & BDRV_BLOCK_ZERO) {
2761 ret = bdrv_write_zeroes(bs, sector_num, n, flags);
2763 error_report("error writing zeroes at sector %" PRId64 ": %s",
2764 sector_num, strerror(-ret));
2771 int bdrv_pread(BlockDriverState *bs, int64_t offset, void *buf, int bytes)
2774 struct iovec iov = {
2775 .iov_base = (void *)buf,
2784 qemu_iovec_init_external(&qiov, &iov, 1);
2785 ret = bdrv_prwv_co(bs, offset, &qiov, false, 0);
2793 int bdrv_pwritev(BlockDriverState *bs, int64_t offset, QEMUIOVector *qiov)
2797 ret = bdrv_prwv_co(bs, offset, qiov, true, 0);
2805 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
2806 const void *buf, int bytes)
2809 struct iovec iov = {
2810 .iov_base = (void *) buf,
2818 qemu_iovec_init_external(&qiov, &iov, 1);
2819 return bdrv_pwritev(bs, offset, &qiov);
2823 * Writes to the file and ensures that no writes are reordered across this
2824 * request (acts as a barrier)
2826 * Returns 0 on success, -errno in error cases.
2828 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
2829 const void *buf, int count)
2833 ret = bdrv_pwrite(bs, offset, buf, count);
2838 /* No flush needed for cache modes that already do it */
2839 if (bs->enable_write_cache) {
2846 static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs,
2847 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
2849 /* Perform I/O through a temporary buffer so that users who scribble over
2850 * their read buffer while the operation is in progress do not end up
2851 * modifying the image file. This is critical for zero-copy guest I/O
2852 * where anything might happen inside guest memory.
2854 void *bounce_buffer;
2856 BlockDriver *drv = bs->drv;
2858 QEMUIOVector bounce_qiov;
2859 int64_t cluster_sector_num;
2860 int cluster_nb_sectors;
2864 /* Cover entire cluster so no additional backing file I/O is required when
2865 * allocating cluster in the image file.
2867 bdrv_round_to_clusters(bs, sector_num, nb_sectors,
2868 &cluster_sector_num, &cluster_nb_sectors);
2870 trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors,
2871 cluster_sector_num, cluster_nb_sectors);
2873 iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE;
2874 iov.iov_base = bounce_buffer = qemu_blockalign(bs, iov.iov_len);
2875 qemu_iovec_init_external(&bounce_qiov, &iov, 1);
2877 ret = drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors,
2883 if (drv->bdrv_co_write_zeroes &&
2884 buffer_is_zero(bounce_buffer, iov.iov_len)) {
2885 ret = bdrv_co_do_write_zeroes(bs, cluster_sector_num,
2886 cluster_nb_sectors, 0);
2888 /* This does not change the data on the disk, it is not necessary
2889 * to flush even in cache=writethrough mode.
2891 ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors,
2896 /* It might be okay to ignore write errors for guest requests. If this
2897 * is a deliberate copy-on-read then we don't want to ignore the error.
2898 * Simply report it in all cases.
2903 skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE;
2904 qemu_iovec_from_buf(qiov, 0, bounce_buffer + skip_bytes,
2905 nb_sectors * BDRV_SECTOR_SIZE);
2908 qemu_vfree(bounce_buffer);
2913 * Forwards an already correctly aligned request to the BlockDriver. This
2914 * handles copy on read and zeroing after EOF; any other features must be
2915 * implemented by the caller.
2917 static int coroutine_fn bdrv_aligned_preadv(BlockDriverState *bs,
2918 BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
2919 int64_t align, QEMUIOVector *qiov, int flags)
2921 BlockDriver *drv = bs->drv;
2924 int64_t sector_num = offset >> BDRV_SECTOR_BITS;
2925 unsigned int nb_sectors = bytes >> BDRV_SECTOR_BITS;
2927 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
2928 assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
2930 /* Handle Copy on Read and associated serialisation */
2931 if (flags & BDRV_REQ_COPY_ON_READ) {
2932 /* If we touch the same cluster it counts as an overlap. This
2933 * guarantees that allocating writes will be serialized and not race
2934 * with each other for the same cluster. For example, in copy-on-read
2935 * it ensures that the CoR read and write operations are atomic and
2936 * guest writes cannot interleave between them. */
2937 mark_request_serialising(req, bdrv_get_cluster_size(bs));
2940 wait_serialising_requests(req);
2942 if (flags & BDRV_REQ_COPY_ON_READ) {
2945 ret = bdrv_is_allocated(bs, sector_num, nb_sectors, &pnum);
2950 if (!ret || pnum != nb_sectors) {
2951 ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov);
2956 /* Forward the request to the BlockDriver */
2957 if (!(bs->zero_beyond_eof && bs->growable)) {
2958 ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
2960 /* Read zeros after EOF of growable BDSes */
2961 int64_t len, total_sectors, max_nb_sectors;
2963 len = bdrv_getlength(bs);
2969 total_sectors = DIV_ROUND_UP(len, BDRV_SECTOR_SIZE);
2970 max_nb_sectors = ROUND_UP(MAX(0, total_sectors - sector_num),
2971 align >> BDRV_SECTOR_BITS);
2972 if (max_nb_sectors > 0) {
2973 ret = drv->bdrv_co_readv(bs, sector_num,
2974 MIN(nb_sectors, max_nb_sectors), qiov);
2979 /* Reading beyond end of file is supposed to produce zeroes */
2980 if (ret == 0 && total_sectors < sector_num + nb_sectors) {
2981 uint64_t offset = MAX(0, total_sectors - sector_num);
2982 uint64_t bytes = (sector_num + nb_sectors - offset) *
2984 qemu_iovec_memset(qiov, offset * BDRV_SECTOR_SIZE, 0, bytes);
2993 * Handle a read request in coroutine context
2995 static int coroutine_fn bdrv_co_do_preadv(BlockDriverState *bs,
2996 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
2997 BdrvRequestFlags flags)
2999 BlockDriver *drv = bs->drv;
3000 BdrvTrackedRequest req;
3002 /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
3003 uint64_t align = MAX(BDRV_SECTOR_SIZE, bs->request_alignment);
3004 uint8_t *head_buf = NULL;
3005 uint8_t *tail_buf = NULL;
3006 QEMUIOVector local_qiov;
3007 bool use_local_qiov = false;
3013 if (bdrv_check_byte_request(bs, offset, bytes)) {
3017 if (bs->copy_on_read) {
3018 flags |= BDRV_REQ_COPY_ON_READ;
3021 /* throttling disk I/O */
3022 if (bs->io_limits_enabled) {
3023 bdrv_io_limits_intercept(bs, bytes, false);
3026 /* Align read if necessary by padding qiov */
3027 if (offset & (align - 1)) {
3028 head_buf = qemu_blockalign(bs, align);
3029 qemu_iovec_init(&local_qiov, qiov->niov + 2);
3030 qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
3031 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
3032 use_local_qiov = true;
3034 bytes += offset & (align - 1);
3035 offset = offset & ~(align - 1);
3038 if ((offset + bytes) & (align - 1)) {
3039 if (!use_local_qiov) {
3040 qemu_iovec_init(&local_qiov, qiov->niov + 1);
3041 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
3042 use_local_qiov = true;
3044 tail_buf = qemu_blockalign(bs, align);
3045 qemu_iovec_add(&local_qiov, tail_buf,
3046 align - ((offset + bytes) & (align - 1)));
3048 bytes = ROUND_UP(bytes, align);
3051 tracked_request_begin(&req, bs, offset, bytes, false);
3052 ret = bdrv_aligned_preadv(bs, &req, offset, bytes, align,
3053 use_local_qiov ? &local_qiov : qiov,
3055 tracked_request_end(&req);
3057 if (use_local_qiov) {
3058 qemu_iovec_destroy(&local_qiov);
3059 qemu_vfree(head_buf);
3060 qemu_vfree(tail_buf);
3066 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
3067 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
3068 BdrvRequestFlags flags)
3070 if (nb_sectors < 0 || nb_sectors > (UINT_MAX >> BDRV_SECTOR_BITS)) {
3074 return bdrv_co_do_preadv(bs, sector_num << BDRV_SECTOR_BITS,
3075 nb_sectors << BDRV_SECTOR_BITS, qiov, flags);
3078 int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
3079 int nb_sectors, QEMUIOVector *qiov)
3081 trace_bdrv_co_readv(bs, sector_num, nb_sectors);
3083 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0);
3086 int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
3087 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
3089 trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors);
3091 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
3092 BDRV_REQ_COPY_ON_READ);
3095 /* if no limit is specified in the BlockLimits use a default
3096 * of 32768 512-byte sectors (16 MiB) per request.
3098 #define MAX_WRITE_ZEROES_DEFAULT 32768
3100 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
3101 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
3103 BlockDriver *drv = bs->drv;
3105 struct iovec iov = {0};
3108 int max_write_zeroes = bs->bl.max_write_zeroes ?
3109 bs->bl.max_write_zeroes : MAX_WRITE_ZEROES_DEFAULT;
3111 while (nb_sectors > 0 && !ret) {
3112 int num = nb_sectors;
3114 /* Align request. Block drivers can expect the "bulk" of the request
3117 if (bs->bl.write_zeroes_alignment
3118 && num > bs->bl.write_zeroes_alignment) {
3119 if (sector_num % bs->bl.write_zeroes_alignment != 0) {
3120 /* Make a small request up to the first aligned sector. */
3121 num = bs->bl.write_zeroes_alignment;
3122 num -= sector_num % bs->bl.write_zeroes_alignment;
3123 } else if ((sector_num + num) % bs->bl.write_zeroes_alignment != 0) {
3124 /* Shorten the request to the last aligned sector. num cannot
3125 * underflow because num > bs->bl.write_zeroes_alignment.
3127 num -= (sector_num + num) % bs->bl.write_zeroes_alignment;
3131 /* limit request size */
3132 if (num > max_write_zeroes) {
3133 num = max_write_zeroes;
3137 /* First try the efficient write zeroes operation */
3138 if (drv->bdrv_co_write_zeroes) {
3139 ret = drv->bdrv_co_write_zeroes(bs, sector_num, num, flags);
3142 if (ret == -ENOTSUP) {
3143 /* Fall back to bounce buffer if write zeroes is unsupported */
3144 iov.iov_len = num * BDRV_SECTOR_SIZE;
3145 if (iov.iov_base == NULL) {
3146 iov.iov_base = qemu_blockalign(bs, num * BDRV_SECTOR_SIZE);
3147 memset(iov.iov_base, 0, num * BDRV_SECTOR_SIZE);
3149 qemu_iovec_init_external(&qiov, &iov, 1);
3151 ret = drv->bdrv_co_writev(bs, sector_num, num, &qiov);
3153 /* Keep bounce buffer around if it is big enough for all
3154 * all future requests.
3156 if (num < max_write_zeroes) {
3157 qemu_vfree(iov.iov_base);
3158 iov.iov_base = NULL;
3166 qemu_vfree(iov.iov_base);
3171 * Forwards an already correctly aligned write request to the BlockDriver.
3173 static int coroutine_fn bdrv_aligned_pwritev(BlockDriverState *bs,
3174 BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
3175 QEMUIOVector *qiov, int flags)
3177 BlockDriver *drv = bs->drv;
3181 int64_t sector_num = offset >> BDRV_SECTOR_BITS;
3182 unsigned int nb_sectors = bytes >> BDRV_SECTOR_BITS;
3184 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
3185 assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
3187 waited = wait_serialising_requests(req);
3188 assert(!waited || !req->serialising);
3189 assert(req->overlap_offset <= offset);
3190 assert(offset + bytes <= req->overlap_offset + req->overlap_bytes);
3192 ret = notifier_with_return_list_notify(&bs->before_write_notifiers, req);
3195 /* Do nothing, write notifier decided to fail this request */
3196 } else if (flags & BDRV_REQ_ZERO_WRITE) {
3197 BLKDBG_EVENT(bs, BLKDBG_PWRITEV_ZERO);
3198 ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors, flags);
3200 BLKDBG_EVENT(bs, BLKDBG_PWRITEV);
3201 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
3203 BLKDBG_EVENT(bs, BLKDBG_PWRITEV_DONE);
3205 if (ret == 0 && !bs->enable_write_cache) {
3206 ret = bdrv_co_flush(bs);
3209 bdrv_set_dirty(bs, sector_num, nb_sectors);
3211 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
3212 bs->wr_highest_sector = sector_num + nb_sectors - 1;
3214 if (bs->growable && ret >= 0) {
3215 bs->total_sectors = MAX(bs->total_sectors, sector_num + nb_sectors);
3222 * Handle a write request in coroutine context
3224 static int coroutine_fn bdrv_co_do_pwritev(BlockDriverState *bs,
3225 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
3226 BdrvRequestFlags flags)
3228 BdrvTrackedRequest req;
3229 /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
3230 uint64_t align = MAX(BDRV_SECTOR_SIZE, bs->request_alignment);
3231 uint8_t *head_buf = NULL;
3232 uint8_t *tail_buf = NULL;
3233 QEMUIOVector local_qiov;
3234 bool use_local_qiov = false;
3240 if (bs->read_only) {
3243 if (bdrv_check_byte_request(bs, offset, bytes)) {
3247 /* throttling disk I/O */
3248 if (bs->io_limits_enabled) {
3249 bdrv_io_limits_intercept(bs, bytes, true);
3253 * Align write if necessary by performing a read-modify-write cycle.
3254 * Pad qiov with the read parts and be sure to have a tracked request not
3255 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
3257 tracked_request_begin(&req, bs, offset, bytes, true);
3259 if (offset & (align - 1)) {
3260 QEMUIOVector head_qiov;
3261 struct iovec head_iov;
3263 mark_request_serialising(&req, align);
3264 wait_serialising_requests(&req);
3266 head_buf = qemu_blockalign(bs, align);
3267 head_iov = (struct iovec) {
3268 .iov_base = head_buf,
3271 qemu_iovec_init_external(&head_qiov, &head_iov, 1);
3273 BLKDBG_EVENT(bs, BLKDBG_PWRITEV_RMW_HEAD);
3274 ret = bdrv_aligned_preadv(bs, &req, offset & ~(align - 1), align,
3275 align, &head_qiov, 0);
3279 BLKDBG_EVENT(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD);
3281 qemu_iovec_init(&local_qiov, qiov->niov + 2);
3282 qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
3283 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
3284 use_local_qiov = true;
3286 bytes += offset & (align - 1);
3287 offset = offset & ~(align - 1);
3290 if ((offset + bytes) & (align - 1)) {
3291 QEMUIOVector tail_qiov;
3292 struct iovec tail_iov;
3296 mark_request_serialising(&req, align);
3297 waited = wait_serialising_requests(&req);
3298 assert(!waited || !use_local_qiov);
3300 tail_buf = qemu_blockalign(bs, align);
3301 tail_iov = (struct iovec) {
3302 .iov_base = tail_buf,
3305 qemu_iovec_init_external(&tail_qiov, &tail_iov, 1);
3307 BLKDBG_EVENT(bs, BLKDBG_PWRITEV_RMW_TAIL);
3308 ret = bdrv_aligned_preadv(bs, &req, (offset + bytes) & ~(align - 1), align,
3309 align, &tail_qiov, 0);
3313 BLKDBG_EVENT(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL);
3315 if (!use_local_qiov) {
3316 qemu_iovec_init(&local_qiov, qiov->niov + 1);
3317 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
3318 use_local_qiov = true;
3321 tail_bytes = (offset + bytes) & (align - 1);
3322 qemu_iovec_add(&local_qiov, tail_buf + tail_bytes, align - tail_bytes);
3324 bytes = ROUND_UP(bytes, align);
3327 ret = bdrv_aligned_pwritev(bs, &req, offset, bytes,
3328 use_local_qiov ? &local_qiov : qiov,
3332 tracked_request_end(&req);
3334 if (use_local_qiov) {
3335 qemu_iovec_destroy(&local_qiov);
3337 qemu_vfree(head_buf);
3338 qemu_vfree(tail_buf);
3343 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
3344 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
3345 BdrvRequestFlags flags)
3347 if (nb_sectors < 0 || nb_sectors > (INT_MAX >> BDRV_SECTOR_BITS)) {
3351 return bdrv_co_do_pwritev(bs, sector_num << BDRV_SECTOR_BITS,
3352 nb_sectors << BDRV_SECTOR_BITS, qiov, flags);
3355 int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
3356 int nb_sectors, QEMUIOVector *qiov)
3358 trace_bdrv_co_writev(bs, sector_num, nb_sectors);
3360 return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0);
3363 int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs,
3364 int64_t sector_num, int nb_sectors,
3365 BdrvRequestFlags flags)
3367 trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors, flags);
3369 if (!(bs->open_flags & BDRV_O_UNMAP)) {
3370 flags &= ~BDRV_REQ_MAY_UNMAP;
3373 return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL,
3374 BDRV_REQ_ZERO_WRITE | flags);
3378 * Truncate file to 'offset' bytes (needed only for file protocols)
3380 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
3382 BlockDriver *drv = bs->drv;
3386 if (!drv->bdrv_truncate)
3390 if (bdrv_in_use(bs))
3392 ret = drv->bdrv_truncate(bs, offset);
3394 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
3395 bdrv_dev_resize_cb(bs);
3401 * Length of a allocated file in bytes. Sparse files are counted by actual
3402 * allocated space. Return < 0 if error or unknown.
3404 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
3406 BlockDriver *drv = bs->drv;
3410 if (drv->bdrv_get_allocated_file_size) {
3411 return drv->bdrv_get_allocated_file_size(bs);
3414 return bdrv_get_allocated_file_size(bs->file);
3420 * Length of a file in bytes. Return < 0 if error or unknown.
3422 int64_t bdrv_getlength(BlockDriverState *bs)
3424 BlockDriver *drv = bs->drv;
3428 if (drv->has_variable_length) {
3429 int ret = refresh_total_sectors(bs, bs->total_sectors);
3434 return bs->total_sectors * BDRV_SECTOR_SIZE;
3437 /* return 0 as number of sectors if no device present or error */
3438 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
3441 length = bdrv_getlength(bs);
3445 length = length >> BDRV_SECTOR_BITS;
3446 *nb_sectors_ptr = length;
3449 void bdrv_set_on_error(BlockDriverState *bs, BlockdevOnError on_read_error,
3450 BlockdevOnError on_write_error)
3452 bs->on_read_error = on_read_error;
3453 bs->on_write_error = on_write_error;
3456 BlockdevOnError bdrv_get_on_error(BlockDriverState *bs, bool is_read)
3458 return is_read ? bs->on_read_error : bs->on_write_error;
3461 BlockErrorAction bdrv_get_error_action(BlockDriverState *bs, bool is_read, int error)
3463 BlockdevOnError on_err = is_read ? bs->on_read_error : bs->on_write_error;
3466 case BLOCKDEV_ON_ERROR_ENOSPC:
3467 return (error == ENOSPC) ? BDRV_ACTION_STOP : BDRV_ACTION_REPORT;
3468 case BLOCKDEV_ON_ERROR_STOP:
3469 return BDRV_ACTION_STOP;
3470 case BLOCKDEV_ON_ERROR_REPORT:
3471 return BDRV_ACTION_REPORT;
3472 case BLOCKDEV_ON_ERROR_IGNORE:
3473 return BDRV_ACTION_IGNORE;
3479 /* This is done by device models because, while the block layer knows
3480 * about the error, it does not know whether an operation comes from
3481 * the device or the block layer (from a job, for example).
3483 void bdrv_error_action(BlockDriverState *bs, BlockErrorAction action,
3484 bool is_read, int error)
3487 bdrv_emit_qmp_error_event(bs, QEVENT_BLOCK_IO_ERROR, action, is_read);
3488 if (action == BDRV_ACTION_STOP) {
3489 vm_stop(RUN_STATE_IO_ERROR);
3490 bdrv_iostatus_set_err(bs, error);
3494 int bdrv_is_read_only(BlockDriverState *bs)
3496 return bs->read_only;
3499 int bdrv_is_sg(BlockDriverState *bs)
3504 int bdrv_enable_write_cache(BlockDriverState *bs)
3506 return bs->enable_write_cache;
3509 void bdrv_set_enable_write_cache(BlockDriverState *bs, bool wce)
3511 bs->enable_write_cache = wce;
3513 /* so a reopen() will preserve wce */
3515 bs->open_flags |= BDRV_O_CACHE_WB;
3517 bs->open_flags &= ~BDRV_O_CACHE_WB;
3521 int bdrv_is_encrypted(BlockDriverState *bs)
3523 if (bs->backing_hd && bs->backing_hd->encrypted)
3525 return bs->encrypted;
3528 int bdrv_key_required(BlockDriverState *bs)
3530 BlockDriverState *backing_hd = bs->backing_hd;
3532 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
3534 return (bs->encrypted && !bs->valid_key);
3537 int bdrv_set_key(BlockDriverState *bs, const char *key)
3540 if (bs->backing_hd && bs->backing_hd->encrypted) {
3541 ret = bdrv_set_key(bs->backing_hd, key);
3547 if (!bs->encrypted) {
3549 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
3552 ret = bs->drv->bdrv_set_key(bs, key);
3555 } else if (!bs->valid_key) {
3557 /* call the change callback now, we skipped it on open */
3558 bdrv_dev_change_media_cb(bs, true);
3563 const char *bdrv_get_format_name(BlockDriverState *bs)
3565 return bs->drv ? bs->drv->format_name : NULL;
3568 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
3573 QLIST_FOREACH(drv, &bdrv_drivers, list) {
3574 it(opaque, drv->format_name);
3578 /* This function is to find block backend bs */
3579 BlockDriverState *bdrv_find(const char *name)
3581 BlockDriverState *bs;
3583 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
3584 if (!strcmp(name, bs->device_name)) {
3591 /* This function is to find a node in the bs graph */
3592 BlockDriverState *bdrv_find_node(const char *node_name)
3594 BlockDriverState *bs;
3598 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3599 if (!strcmp(node_name, bs->node_name)) {
3606 /* Put this QMP function here so it can access the static graph_bdrv_states. */
3607 BlockDeviceInfoList *bdrv_named_nodes_list(void)
3609 BlockDeviceInfoList *list, *entry;
3610 BlockDriverState *bs;
3613 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3614 entry = g_malloc0(sizeof(*entry));
3615 entry->value = bdrv_block_device_info(bs);
3623 BlockDriverState *bdrv_lookup_bs(const char *device,
3624 const char *node_name,
3627 BlockDriverState *bs = NULL;
3630 bs = bdrv_find(device);
3638 bs = bdrv_find_node(node_name);
3645 error_setg(errp, "Cannot find device=%s nor node_name=%s",
3646 device ? device : "",
3647 node_name ? node_name : "");
3651 BlockDriverState *bdrv_next(BlockDriverState *bs)
3654 return QTAILQ_FIRST(&bdrv_states);
3656 return QTAILQ_NEXT(bs, device_list);
3659 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
3661 BlockDriverState *bs;
3663 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
3668 const char *bdrv_get_device_name(BlockDriverState *bs)
3670 return bs->device_name;
3673 int bdrv_get_flags(BlockDriverState *bs)
3675 return bs->open_flags;
3678 int bdrv_flush_all(void)
3680 BlockDriverState *bs;
3683 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
3684 int ret = bdrv_flush(bs);
3685 if (ret < 0 && !result) {
3693 int bdrv_has_zero_init_1(BlockDriverState *bs)
3698 int bdrv_has_zero_init(BlockDriverState *bs)
3702 /* If BS is a copy on write image, it is initialized to
3703 the contents of the base image, which may not be zeroes. */
3704 if (bs->backing_hd) {
3707 if (bs->drv->bdrv_has_zero_init) {
3708 return bs->drv->bdrv_has_zero_init(bs);
3715 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
3717 BlockDriverInfo bdi;
3719 if (bs->backing_hd) {
3723 if (bdrv_get_info(bs, &bdi) == 0) {
3724 return bdi.unallocated_blocks_are_zero;
3730 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
3732 BlockDriverInfo bdi;
3734 if (bs->backing_hd || !(bs->open_flags & BDRV_O_UNMAP)) {
3738 if (bdrv_get_info(bs, &bdi) == 0) {
3739 return bdi.can_write_zeroes_with_unmap;
3745 typedef struct BdrvCoGetBlockStatusData {
3746 BlockDriverState *bs;
3747 BlockDriverState *base;
3753 } BdrvCoGetBlockStatusData;
3756 * Returns true iff the specified sector is present in the disk image. Drivers
3757 * not implementing the functionality are assumed to not support backing files,
3758 * hence all their sectors are reported as allocated.
3760 * If 'sector_num' is beyond the end of the disk image the return value is 0
3761 * and 'pnum' is set to 0.
3763 * 'pnum' is set to the number of sectors (including and immediately following
3764 * the specified sector) that are known to be in the same
3765 * allocated/unallocated state.
3767 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
3768 * beyond the end of the disk image it will be clamped.
3770 static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs,
3772 int nb_sectors, int *pnum)
3778 length = bdrv_getlength(bs);
3783 if (sector_num >= (length >> BDRV_SECTOR_BITS)) {
3788 n = bs->total_sectors - sector_num;
3789 if (n < nb_sectors) {
3793 if (!bs->drv->bdrv_co_get_block_status) {
3795 ret = BDRV_BLOCK_DATA;
3796 if (bs->drv->protocol_name) {
3797 ret |= BDRV_BLOCK_OFFSET_VALID | (sector_num * BDRV_SECTOR_SIZE);
3802 ret = bs->drv->bdrv_co_get_block_status(bs, sector_num, nb_sectors, pnum);
3808 if (ret & BDRV_BLOCK_RAW) {
3809 assert(ret & BDRV_BLOCK_OFFSET_VALID);
3810 return bdrv_get_block_status(bs->file, ret >> BDRV_SECTOR_BITS,
3814 if (!(ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO)) {
3815 if (bdrv_unallocated_blocks_are_zero(bs)) {
3816 ret |= BDRV_BLOCK_ZERO;
3817 } else if (bs->backing_hd) {
3818 BlockDriverState *bs2 = bs->backing_hd;
3819 int64_t length2 = bdrv_getlength(bs2);
3820 if (length2 >= 0 && sector_num >= (length2 >> BDRV_SECTOR_BITS)) {
3821 ret |= BDRV_BLOCK_ZERO;
3827 (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) &&
3828 (ret & BDRV_BLOCK_OFFSET_VALID)) {
3829 ret2 = bdrv_co_get_block_status(bs->file, ret >> BDRV_SECTOR_BITS,
3832 /* Ignore errors. This is just providing extra information, it
3833 * is useful but not necessary.
3835 ret |= (ret2 & BDRV_BLOCK_ZERO);
3842 /* Coroutine wrapper for bdrv_get_block_status() */
3843 static void coroutine_fn bdrv_get_block_status_co_entry(void *opaque)
3845 BdrvCoGetBlockStatusData *data = opaque;
3846 BlockDriverState *bs = data->bs;
3848 data->ret = bdrv_co_get_block_status(bs, data->sector_num, data->nb_sectors,
3854 * Synchronous wrapper around bdrv_co_get_block_status().
3856 * See bdrv_co_get_block_status() for details.
3858 int64_t bdrv_get_block_status(BlockDriverState *bs, int64_t sector_num,
3859 int nb_sectors, int *pnum)
3862 BdrvCoGetBlockStatusData data = {
3864 .sector_num = sector_num,
3865 .nb_sectors = nb_sectors,
3870 if (qemu_in_coroutine()) {
3871 /* Fast-path if already in coroutine context */
3872 bdrv_get_block_status_co_entry(&data);
3874 co = qemu_coroutine_create(bdrv_get_block_status_co_entry);
3875 qemu_coroutine_enter(co, &data);
3876 while (!data.done) {
3883 int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num,
3884 int nb_sectors, int *pnum)
3886 int64_t ret = bdrv_get_block_status(bs, sector_num, nb_sectors, pnum);
3891 (ret & BDRV_BLOCK_DATA) ||
3892 ((ret & BDRV_BLOCK_ZERO) && !bdrv_has_zero_init(bs));
3896 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
3898 * Return true if the given sector is allocated in any image between
3899 * BASE and TOP (inclusive). BASE can be NULL to check if the given
3900 * sector is allocated in any image of the chain. Return false otherwise.
3902 * 'pnum' is set to the number of sectors (including and immediately following
3903 * the specified sector) that are known to be in the same
3904 * allocated/unallocated state.
3907 int bdrv_is_allocated_above(BlockDriverState *top,
3908 BlockDriverState *base,
3910 int nb_sectors, int *pnum)
3912 BlockDriverState *intermediate;
3913 int ret, n = nb_sectors;
3916 while (intermediate && intermediate != base) {
3918 ret = bdrv_is_allocated(intermediate, sector_num, nb_sectors,
3928 * [sector_num, nb_sectors] is unallocated on top but intermediate
3931 * [sector_num+x, nr_sectors] allocated.
3933 if (n > pnum_inter &&
3934 (intermediate == top ||
3935 sector_num + pnum_inter < intermediate->total_sectors)) {
3939 intermediate = intermediate->backing_hd;
3946 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
3948 if (bs->backing_hd && bs->backing_hd->encrypted)
3949 return bs->backing_file;
3950 else if (bs->encrypted)
3951 return bs->filename;
3956 void bdrv_get_backing_filename(BlockDriverState *bs,
3957 char *filename, int filename_size)
3959 pstrcpy(filename, filename_size, bs->backing_file);
3962 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
3963 const uint8_t *buf, int nb_sectors)
3965 BlockDriver *drv = bs->drv;
3968 if (!drv->bdrv_write_compressed)
3970 if (bdrv_check_request(bs, sector_num, nb_sectors))
3973 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
3975 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
3978 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
3980 BlockDriver *drv = bs->drv;
3983 if (!drv->bdrv_get_info)
3985 memset(bdi, 0, sizeof(*bdi));
3986 return drv->bdrv_get_info(bs, bdi);
3989 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs)
3991 BlockDriver *drv = bs->drv;
3992 if (drv && drv->bdrv_get_specific_info) {
3993 return drv->bdrv_get_specific_info(bs);
3998 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
3999 int64_t pos, int size)
4002 struct iovec iov = {
4003 .iov_base = (void *) buf,
4007 qemu_iovec_init_external(&qiov, &iov, 1);
4008 return bdrv_writev_vmstate(bs, &qiov, pos);
4011 int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
4013 BlockDriver *drv = bs->drv;
4017 } else if (drv->bdrv_save_vmstate) {
4018 return drv->bdrv_save_vmstate(bs, qiov, pos);
4019 } else if (bs->file) {
4020 return bdrv_writev_vmstate(bs->file, qiov, pos);
4026 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
4027 int64_t pos, int size)
4029 BlockDriver *drv = bs->drv;
4032 if (drv->bdrv_load_vmstate)
4033 return drv->bdrv_load_vmstate(bs, buf, pos, size);
4035 return bdrv_load_vmstate(bs->file, buf, pos, size);
4039 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
4041 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
4045 bs->drv->bdrv_debug_event(bs, event);
4048 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
4051 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
4055 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
4056 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
4062 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
4064 while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
4068 if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
4069 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
4075 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
4077 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
4081 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
4082 return bs->drv->bdrv_debug_resume(bs, tag);
4088 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
4090 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
4094 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
4095 return bs->drv->bdrv_debug_is_suspended(bs, tag);
4101 int bdrv_is_snapshot(BlockDriverState *bs)
4103 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
4106 /* backing_file can either be relative, or absolute, or a protocol. If it is
4107 * relative, it must be relative to the chain. So, passing in bs->filename
4108 * from a BDS as backing_file should not be done, as that may be relative to
4109 * the CWD rather than the chain. */
4110 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
4111 const char *backing_file)
4113 char *filename_full = NULL;
4114 char *backing_file_full = NULL;
4115 char *filename_tmp = NULL;
4116 int is_protocol = 0;
4117 BlockDriverState *curr_bs = NULL;
4118 BlockDriverState *retval = NULL;
4120 if (!bs || !bs->drv || !backing_file) {
4124 filename_full = g_malloc(PATH_MAX);
4125 backing_file_full = g_malloc(PATH_MAX);
4126 filename_tmp = g_malloc(PATH_MAX);
4128 is_protocol = path_has_protocol(backing_file);
4130 for (curr_bs = bs; curr_bs->backing_hd; curr_bs = curr_bs->backing_hd) {
4132 /* If either of the filename paths is actually a protocol, then
4133 * compare unmodified paths; otherwise make paths relative */
4134 if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
4135 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
4136 retval = curr_bs->backing_hd;
4140 /* If not an absolute filename path, make it relative to the current
4141 * image's filename path */
4142 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4145 /* We are going to compare absolute pathnames */
4146 if (!realpath(filename_tmp, filename_full)) {
4150 /* We need to make sure the backing filename we are comparing against
4151 * is relative to the current image filename (or absolute) */
4152 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4153 curr_bs->backing_file);
4155 if (!realpath(filename_tmp, backing_file_full)) {
4159 if (strcmp(backing_file_full, filename_full) == 0) {
4160 retval = curr_bs->backing_hd;
4166 g_free(filename_full);
4167 g_free(backing_file_full);
4168 g_free(filename_tmp);
4172 int bdrv_get_backing_file_depth(BlockDriverState *bs)
4178 if (!bs->backing_hd) {
4182 return 1 + bdrv_get_backing_file_depth(bs->backing_hd);
4185 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
4187 BlockDriverState *curr_bs = NULL;
4195 while (curr_bs->backing_hd) {
4196 curr_bs = curr_bs->backing_hd;
4201 /**************************************************************/
4204 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
4205 QEMUIOVector *qiov, int nb_sectors,
4206 BlockDriverCompletionFunc *cb, void *opaque)
4208 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
4210 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0,
4214 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
4215 QEMUIOVector *qiov, int nb_sectors,
4216 BlockDriverCompletionFunc *cb, void *opaque)
4218 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
4220 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0,
4224 BlockDriverAIOCB *bdrv_aio_write_zeroes(BlockDriverState *bs,
4225 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags,
4226 BlockDriverCompletionFunc *cb, void *opaque)
4228 trace_bdrv_aio_write_zeroes(bs, sector_num, nb_sectors, flags, opaque);
4230 return bdrv_co_aio_rw_vector(bs, sector_num, NULL, nb_sectors,
4231 BDRV_REQ_ZERO_WRITE | flags,
4236 typedef struct MultiwriteCB {
4241 BlockDriverCompletionFunc *cb;
4243 QEMUIOVector *free_qiov;
4247 static void multiwrite_user_cb(MultiwriteCB *mcb)
4251 for (i = 0; i < mcb->num_callbacks; i++) {
4252 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
4253 if (mcb->callbacks[i].free_qiov) {
4254 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
4256 g_free(mcb->callbacks[i].free_qiov);
4260 static void multiwrite_cb(void *opaque, int ret)
4262 MultiwriteCB *mcb = opaque;
4264 trace_multiwrite_cb(mcb, ret);
4266 if (ret < 0 && !mcb->error) {
4270 mcb->num_requests--;
4271 if (mcb->num_requests == 0) {
4272 multiwrite_user_cb(mcb);
4277 static int multiwrite_req_compare(const void *a, const void *b)
4279 const BlockRequest *req1 = a, *req2 = b;
4282 * Note that we can't simply subtract req2->sector from req1->sector
4283 * here as that could overflow the return value.
4285 if (req1->sector > req2->sector) {
4287 } else if (req1->sector < req2->sector) {
4295 * Takes a bunch of requests and tries to merge them. Returns the number of
4296 * requests that remain after merging.
4298 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
4299 int num_reqs, MultiwriteCB *mcb)
4303 // Sort requests by start sector
4304 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
4306 // Check if adjacent requests touch the same clusters. If so, combine them,
4307 // filling up gaps with zero sectors.
4309 for (i = 1; i < num_reqs; i++) {
4311 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
4313 // Handle exactly sequential writes and overlapping writes.
4314 if (reqs[i].sector <= oldreq_last) {
4318 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
4324 QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
4325 qemu_iovec_init(qiov,
4326 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
4328 // Add the first request to the merged one. If the requests are
4329 // overlapping, drop the last sectors of the first request.
4330 size = (reqs[i].sector - reqs[outidx].sector) << 9;
4331 qemu_iovec_concat(qiov, reqs[outidx].qiov, 0, size);
4333 // We should need to add any zeros between the two requests
4334 assert (reqs[i].sector <= oldreq_last);
4336 // Add the second request
4337 qemu_iovec_concat(qiov, reqs[i].qiov, 0, reqs[i].qiov->size);
4339 reqs[outidx].nb_sectors = qiov->size >> 9;
4340 reqs[outidx].qiov = qiov;
4342 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
4345 reqs[outidx].sector = reqs[i].sector;
4346 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
4347 reqs[outidx].qiov = reqs[i].qiov;
4355 * Submit multiple AIO write requests at once.
4357 * On success, the function returns 0 and all requests in the reqs array have
4358 * been submitted. In error case this function returns -1, and any of the
4359 * requests may or may not be submitted yet. In particular, this means that the
4360 * callback will be called for some of the requests, for others it won't. The
4361 * caller must check the error field of the BlockRequest to wait for the right
4362 * callbacks (if error != 0, no callback will be called).
4364 * The implementation may modify the contents of the reqs array, e.g. to merge
4365 * requests. However, the fields opaque and error are left unmodified as they
4366 * are used to signal failure for a single request to the caller.
4368 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
4373 /* don't submit writes if we don't have a medium */
4374 if (bs->drv == NULL) {
4375 for (i = 0; i < num_reqs; i++) {
4376 reqs[i].error = -ENOMEDIUM;
4381 if (num_reqs == 0) {
4385 // Create MultiwriteCB structure
4386 mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
4387 mcb->num_requests = 0;
4388 mcb->num_callbacks = num_reqs;
4390 for (i = 0; i < num_reqs; i++) {
4391 mcb->callbacks[i].cb = reqs[i].cb;
4392 mcb->callbacks[i].opaque = reqs[i].opaque;
4395 // Check for mergable requests
4396 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
4398 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
4400 /* Run the aio requests. */
4401 mcb->num_requests = num_reqs;
4402 for (i = 0; i < num_reqs; i++) {
4403 bdrv_co_aio_rw_vector(bs, reqs[i].sector, reqs[i].qiov,
4404 reqs[i].nb_sectors, reqs[i].flags,
4412 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
4414 acb->aiocb_info->cancel(acb);
4417 /**************************************************************/
4418 /* async block device emulation */
4420 typedef struct BlockDriverAIOCBSync {
4421 BlockDriverAIOCB common;
4424 /* vector translation state */
4428 } BlockDriverAIOCBSync;
4430 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
4432 BlockDriverAIOCBSync *acb =
4433 container_of(blockacb, BlockDriverAIOCBSync, common);
4434 qemu_bh_delete(acb->bh);
4436 qemu_aio_release(acb);
4439 static const AIOCBInfo bdrv_em_aiocb_info = {
4440 .aiocb_size = sizeof(BlockDriverAIOCBSync),
4441 .cancel = bdrv_aio_cancel_em,
4444 static void bdrv_aio_bh_cb(void *opaque)
4446 BlockDriverAIOCBSync *acb = opaque;
4449 qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
4450 qemu_vfree(acb->bounce);
4451 acb->common.cb(acb->common.opaque, acb->ret);
4452 qemu_bh_delete(acb->bh);
4454 qemu_aio_release(acb);
4457 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
4461 BlockDriverCompletionFunc *cb,
4466 BlockDriverAIOCBSync *acb;
4468 acb = qemu_aio_get(&bdrv_em_aiocb_info, bs, cb, opaque);
4469 acb->is_write = is_write;
4471 acb->bounce = qemu_blockalign(bs, qiov->size);
4472 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
4475 qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
4476 acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
4478 acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
4481 qemu_bh_schedule(acb->bh);
4483 return &acb->common;
4486 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
4487 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
4488 BlockDriverCompletionFunc *cb, void *opaque)
4490 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
4493 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
4494 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
4495 BlockDriverCompletionFunc *cb, void *opaque)
4497 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
4501 typedef struct BlockDriverAIOCBCoroutine {
4502 BlockDriverAIOCB common;
4507 } BlockDriverAIOCBCoroutine;
4509 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb)
4511 BlockDriverAIOCBCoroutine *acb =
4512 container_of(blockacb, BlockDriverAIOCBCoroutine, common);
4521 static const AIOCBInfo bdrv_em_co_aiocb_info = {
4522 .aiocb_size = sizeof(BlockDriverAIOCBCoroutine),
4523 .cancel = bdrv_aio_co_cancel_em,
4526 static void bdrv_co_em_bh(void *opaque)
4528 BlockDriverAIOCBCoroutine *acb = opaque;
4530 acb->common.cb(acb->common.opaque, acb->req.error);
4536 qemu_bh_delete(acb->bh);
4537 qemu_aio_release(acb);
4540 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
4541 static void coroutine_fn bdrv_co_do_rw(void *opaque)
4543 BlockDriverAIOCBCoroutine *acb = opaque;
4544 BlockDriverState *bs = acb->common.bs;
4546 if (!acb->is_write) {
4547 acb->req.error = bdrv_co_do_readv(bs, acb->req.sector,
4548 acb->req.nb_sectors, acb->req.qiov, acb->req.flags);
4550 acb->req.error = bdrv_co_do_writev(bs, acb->req.sector,
4551 acb->req.nb_sectors, acb->req.qiov, acb->req.flags);
4554 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
4555 qemu_bh_schedule(acb->bh);
4558 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
4562 BdrvRequestFlags flags,
4563 BlockDriverCompletionFunc *cb,
4568 BlockDriverAIOCBCoroutine *acb;
4570 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
4571 acb->req.sector = sector_num;
4572 acb->req.nb_sectors = nb_sectors;
4573 acb->req.qiov = qiov;
4574 acb->req.flags = flags;
4575 acb->is_write = is_write;
4578 co = qemu_coroutine_create(bdrv_co_do_rw);
4579 qemu_coroutine_enter(co, acb);
4581 return &acb->common;
4584 static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
4586 BlockDriverAIOCBCoroutine *acb = opaque;
4587 BlockDriverState *bs = acb->common.bs;
4589 acb->req.error = bdrv_co_flush(bs);
4590 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
4591 qemu_bh_schedule(acb->bh);
4594 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
4595 BlockDriverCompletionFunc *cb, void *opaque)
4597 trace_bdrv_aio_flush(bs, opaque);
4600 BlockDriverAIOCBCoroutine *acb;
4602 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
4605 co = qemu_coroutine_create(bdrv_aio_flush_co_entry);
4606 qemu_coroutine_enter(co, acb);
4608 return &acb->common;
4611 static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque)
4613 BlockDriverAIOCBCoroutine *acb = opaque;
4614 BlockDriverState *bs = acb->common.bs;
4616 acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors);
4617 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
4618 qemu_bh_schedule(acb->bh);
4621 BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs,
4622 int64_t sector_num, int nb_sectors,
4623 BlockDriverCompletionFunc *cb, void *opaque)
4626 BlockDriverAIOCBCoroutine *acb;
4628 trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque);
4630 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
4631 acb->req.sector = sector_num;
4632 acb->req.nb_sectors = nb_sectors;
4634 co = qemu_coroutine_create(bdrv_aio_discard_co_entry);
4635 qemu_coroutine_enter(co, acb);
4637 return &acb->common;
4640 void bdrv_init(void)
4642 module_call_init(MODULE_INIT_BLOCK);
4645 void bdrv_init_with_whitelist(void)
4647 use_bdrv_whitelist = 1;
4651 void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs,
4652 BlockDriverCompletionFunc *cb, void *opaque)
4654 BlockDriverAIOCB *acb;
4656 acb = g_slice_alloc(aiocb_info->aiocb_size);
4657 acb->aiocb_info = aiocb_info;
4660 acb->opaque = opaque;
4664 void qemu_aio_release(void *p)
4666 BlockDriverAIOCB *acb = p;
4667 g_slice_free1(acb->aiocb_info->aiocb_size, acb);
4670 /**************************************************************/
4671 /* Coroutine block device emulation */
4673 typedef struct CoroutineIOCompletion {
4674 Coroutine *coroutine;
4676 } CoroutineIOCompletion;
4678 static void bdrv_co_io_em_complete(void *opaque, int ret)
4680 CoroutineIOCompletion *co = opaque;
4683 qemu_coroutine_enter(co->coroutine, NULL);
4686 static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num,
4687 int nb_sectors, QEMUIOVector *iov,
4690 CoroutineIOCompletion co = {
4691 .coroutine = qemu_coroutine_self(),
4693 BlockDriverAIOCB *acb;
4696 acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors,
4697 bdrv_co_io_em_complete, &co);
4699 acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors,
4700 bdrv_co_io_em_complete, &co);
4703 trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb);
4707 qemu_coroutine_yield();
4712 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
4713 int64_t sector_num, int nb_sectors,
4716 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false);
4719 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
4720 int64_t sector_num, int nb_sectors,
4723 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true);
4726 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
4728 RwCo *rwco = opaque;
4730 rwco->ret = bdrv_co_flush(rwco->bs);
4733 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
4737 if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
4741 /* Write back cached data to the OS even with cache=unsafe */
4742 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS);
4743 if (bs->drv->bdrv_co_flush_to_os) {
4744 ret = bs->drv->bdrv_co_flush_to_os(bs);
4750 /* But don't actually force it to the disk with cache=unsafe */
4751 if (bs->open_flags & BDRV_O_NO_FLUSH) {
4755 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK);
4756 if (bs->drv->bdrv_co_flush_to_disk) {
4757 ret = bs->drv->bdrv_co_flush_to_disk(bs);
4758 } else if (bs->drv->bdrv_aio_flush) {
4759 BlockDriverAIOCB *acb;
4760 CoroutineIOCompletion co = {
4761 .coroutine = qemu_coroutine_self(),
4764 acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
4768 qemu_coroutine_yield();
4773 * Some block drivers always operate in either writethrough or unsafe
4774 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
4775 * know how the server works (because the behaviour is hardcoded or
4776 * depends on server-side configuration), so we can't ensure that
4777 * everything is safe on disk. Returning an error doesn't work because
4778 * that would break guests even if the server operates in writethrough
4781 * Let's hope the user knows what he's doing.
4789 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
4790 * in the case of cache=unsafe, so there are no useless flushes.
4793 return bdrv_co_flush(bs->file);
4796 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
4798 Error *local_err = NULL;
4805 if (bs->drv->bdrv_invalidate_cache) {
4806 bs->drv->bdrv_invalidate_cache(bs, &local_err);
4807 } else if (bs->file) {
4808 bdrv_invalidate_cache(bs->file, &local_err);
4811 error_propagate(errp, local_err);
4815 ret = refresh_total_sectors(bs, bs->total_sectors);
4817 error_setg_errno(errp, -ret, "Could not refresh total sector count");
4822 void bdrv_invalidate_cache_all(Error **errp)
4824 BlockDriverState *bs;
4825 Error *local_err = NULL;
4827 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
4828 bdrv_invalidate_cache(bs, &local_err);
4830 error_propagate(errp, local_err);
4836 void bdrv_clear_incoming_migration_all(void)
4838 BlockDriverState *bs;
4840 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
4841 bs->open_flags = bs->open_flags & ~(BDRV_O_INCOMING);
4845 int bdrv_flush(BlockDriverState *bs)
4853 if (qemu_in_coroutine()) {
4854 /* Fast-path if already in coroutine context */
4855 bdrv_flush_co_entry(&rwco);
4857 co = qemu_coroutine_create(bdrv_flush_co_entry);
4858 qemu_coroutine_enter(co, &rwco);
4859 while (rwco.ret == NOT_DONE) {
4867 typedef struct DiscardCo {
4868 BlockDriverState *bs;
4873 static void coroutine_fn bdrv_discard_co_entry(void *opaque)
4875 DiscardCo *rwco = opaque;
4877 rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors);
4880 /* if no limit is specified in the BlockLimits use a default
4881 * of 32768 512-byte sectors (16 MiB) per request.
4883 #define MAX_DISCARD_DEFAULT 32768
4885 int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
4892 } else if (bdrv_check_request(bs, sector_num, nb_sectors)) {
4894 } else if (bs->read_only) {
4898 bdrv_reset_dirty(bs, sector_num, nb_sectors);
4900 /* Do nothing if disabled. */
4901 if (!(bs->open_flags & BDRV_O_UNMAP)) {
4905 if (!bs->drv->bdrv_co_discard && !bs->drv->bdrv_aio_discard) {
4909 max_discard = bs->bl.max_discard ? bs->bl.max_discard : MAX_DISCARD_DEFAULT;
4910 while (nb_sectors > 0) {
4912 int num = nb_sectors;
4915 if (bs->bl.discard_alignment &&
4916 num >= bs->bl.discard_alignment &&
4917 sector_num % bs->bl.discard_alignment) {
4918 if (num > bs->bl.discard_alignment) {
4919 num = bs->bl.discard_alignment;
4921 num -= sector_num % bs->bl.discard_alignment;
4924 /* limit request size */
4925 if (num > max_discard) {
4929 if (bs->drv->bdrv_co_discard) {
4930 ret = bs->drv->bdrv_co_discard(bs, sector_num, num);
4932 BlockDriverAIOCB *acb;
4933 CoroutineIOCompletion co = {
4934 .coroutine = qemu_coroutine_self(),
4937 acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors,
4938 bdrv_co_io_em_complete, &co);
4942 qemu_coroutine_yield();
4946 if (ret && ret != -ENOTSUP) {
4956 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
4961 .sector_num = sector_num,
4962 .nb_sectors = nb_sectors,
4966 if (qemu_in_coroutine()) {
4967 /* Fast-path if already in coroutine context */
4968 bdrv_discard_co_entry(&rwco);
4970 co = qemu_coroutine_create(bdrv_discard_co_entry);
4971 qemu_coroutine_enter(co, &rwco);
4972 while (rwco.ret == NOT_DONE) {
4980 /**************************************************************/
4981 /* removable device support */
4984 * Return TRUE if the media is present
4986 int bdrv_is_inserted(BlockDriverState *bs)
4988 BlockDriver *drv = bs->drv;
4992 if (!drv->bdrv_is_inserted)
4994 return drv->bdrv_is_inserted(bs);
4998 * Return whether the media changed since the last call to this
4999 * function, or -ENOTSUP if we don't know. Most drivers don't know.
5001 int bdrv_media_changed(BlockDriverState *bs)
5003 BlockDriver *drv = bs->drv;
5005 if (drv && drv->bdrv_media_changed) {
5006 return drv->bdrv_media_changed(bs);
5012 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
5014 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
5016 BlockDriver *drv = bs->drv;
5018 if (drv && drv->bdrv_eject) {
5019 drv->bdrv_eject(bs, eject_flag);
5022 if (bs->device_name[0] != '\0') {
5023 bdrv_emit_qmp_eject_event(bs, eject_flag);
5028 * Lock or unlock the media (if it is locked, the user won't be able
5029 * to eject it manually).
5031 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
5033 BlockDriver *drv = bs->drv;
5035 trace_bdrv_lock_medium(bs, locked);
5037 if (drv && drv->bdrv_lock_medium) {
5038 drv->bdrv_lock_medium(bs, locked);
5042 /* needed for generic scsi interface */
5044 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
5046 BlockDriver *drv = bs->drv;
5048 if (drv && drv->bdrv_ioctl)
5049 return drv->bdrv_ioctl(bs, req, buf);
5053 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
5054 unsigned long int req, void *buf,
5055 BlockDriverCompletionFunc *cb, void *opaque)
5057 BlockDriver *drv = bs->drv;
5059 if (drv && drv->bdrv_aio_ioctl)
5060 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
5064 void bdrv_set_guest_block_size(BlockDriverState *bs, int align)
5066 bs->guest_block_size = align;
5069 void *qemu_blockalign(BlockDriverState *bs, size_t size)
5071 return qemu_memalign(bdrv_opt_mem_align(bs), size);
5075 * Check if all memory in this vector is sector aligned.
5077 bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov)
5080 size_t alignment = bdrv_opt_mem_align(bs);
5082 for (i = 0; i < qiov->niov; i++) {
5083 if ((uintptr_t) qiov->iov[i].iov_base % alignment) {
5086 if (qiov->iov[i].iov_len % alignment) {
5094 BdrvDirtyBitmap *bdrv_create_dirty_bitmap(BlockDriverState *bs, int granularity)
5096 int64_t bitmap_size;
5097 BdrvDirtyBitmap *bitmap;
5099 assert((granularity & (granularity - 1)) == 0);
5101 granularity >>= BDRV_SECTOR_BITS;
5102 assert(granularity);
5103 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS);
5104 bitmap = g_malloc0(sizeof(BdrvDirtyBitmap));
5105 bitmap->bitmap = hbitmap_alloc(bitmap_size, ffs(granularity) - 1);
5106 QLIST_INSERT_HEAD(&bs->dirty_bitmaps, bitmap, list);
5110 void bdrv_release_dirty_bitmap(BlockDriverState *bs, BdrvDirtyBitmap *bitmap)
5112 BdrvDirtyBitmap *bm, *next;
5113 QLIST_FOREACH_SAFE(bm, &bs->dirty_bitmaps, list, next) {
5115 QLIST_REMOVE(bitmap, list);
5116 hbitmap_free(bitmap->bitmap);
5123 BlockDirtyInfoList *bdrv_query_dirty_bitmaps(BlockDriverState *bs)
5125 BdrvDirtyBitmap *bm;
5126 BlockDirtyInfoList *list = NULL;
5127 BlockDirtyInfoList **plist = &list;
5129 QLIST_FOREACH(bm, &bs->dirty_bitmaps, list) {
5130 BlockDirtyInfo *info = g_malloc0(sizeof(BlockDirtyInfo));
5131 BlockDirtyInfoList *entry = g_malloc0(sizeof(BlockDirtyInfoList));
5132 info->count = bdrv_get_dirty_count(bs, bm);
5134 ((int64_t) BDRV_SECTOR_SIZE << hbitmap_granularity(bm->bitmap));
5135 entry->value = info;
5137 plist = &entry->next;
5143 int bdrv_get_dirty(BlockDriverState *bs, BdrvDirtyBitmap *bitmap, int64_t sector)
5146 return hbitmap_get(bitmap->bitmap, sector);
5152 void bdrv_dirty_iter_init(BlockDriverState *bs,
5153 BdrvDirtyBitmap *bitmap, HBitmapIter *hbi)
5155 hbitmap_iter_init(hbi, bitmap->bitmap, 0);
5158 void bdrv_set_dirty(BlockDriverState *bs, int64_t cur_sector,
5161 BdrvDirtyBitmap *bitmap;
5162 QLIST_FOREACH(bitmap, &bs->dirty_bitmaps, list) {
5163 hbitmap_set(bitmap->bitmap, cur_sector, nr_sectors);
5167 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector, int nr_sectors)
5169 BdrvDirtyBitmap *bitmap;
5170 QLIST_FOREACH(bitmap, &bs->dirty_bitmaps, list) {
5171 hbitmap_reset(bitmap->bitmap, cur_sector, nr_sectors);
5175 int64_t bdrv_get_dirty_count(BlockDriverState *bs, BdrvDirtyBitmap *bitmap)
5177 return hbitmap_count(bitmap->bitmap);
5180 /* Get a reference to bs */
5181 void bdrv_ref(BlockDriverState *bs)
5186 /* Release a previously grabbed reference to bs.
5187 * If after releasing, reference count is zero, the BlockDriverState is
5189 void bdrv_unref(BlockDriverState *bs)
5191 assert(bs->refcnt > 0);
5192 if (--bs->refcnt == 0) {
5197 void bdrv_set_in_use(BlockDriverState *bs, int in_use)
5199 assert(bs->in_use != in_use);
5200 bs->in_use = in_use;
5203 int bdrv_in_use(BlockDriverState *bs)
5208 void bdrv_iostatus_enable(BlockDriverState *bs)
5210 bs->iostatus_enabled = true;
5211 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
5214 /* The I/O status is only enabled if the drive explicitly
5215 * enables it _and_ the VM is configured to stop on errors */
5216 bool bdrv_iostatus_is_enabled(const BlockDriverState *bs)
5218 return (bs->iostatus_enabled &&
5219 (bs->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
5220 bs->on_write_error == BLOCKDEV_ON_ERROR_STOP ||
5221 bs->on_read_error == BLOCKDEV_ON_ERROR_STOP));
5224 void bdrv_iostatus_disable(BlockDriverState *bs)
5226 bs->iostatus_enabled = false;
5229 void bdrv_iostatus_reset(BlockDriverState *bs)
5231 if (bdrv_iostatus_is_enabled(bs)) {
5232 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
5234 block_job_iostatus_reset(bs->job);
5239 void bdrv_iostatus_set_err(BlockDriverState *bs, int error)
5241 assert(bdrv_iostatus_is_enabled(bs));
5242 if (bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
5243 bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
5244 BLOCK_DEVICE_IO_STATUS_FAILED;
5249 bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes,
5250 enum BlockAcctType type)
5252 assert(type < BDRV_MAX_IOTYPE);
5254 cookie->bytes = bytes;
5255 cookie->start_time_ns = get_clock();
5256 cookie->type = type;
5260 bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie)
5262 assert(cookie->type < BDRV_MAX_IOTYPE);
5264 bs->nr_bytes[cookie->type] += cookie->bytes;
5265 bs->nr_ops[cookie->type]++;
5266 bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns;
5269 void bdrv_img_create(const char *filename, const char *fmt,
5270 const char *base_filename, const char *base_fmt,
5271 char *options, uint64_t img_size, int flags,
5272 Error **errp, bool quiet)
5274 QEMUOptionParameter *param = NULL, *create_options = NULL;
5275 QEMUOptionParameter *backing_fmt, *backing_file, *size;
5276 BlockDriver *drv, *proto_drv;
5277 BlockDriver *backing_drv = NULL;
5278 Error *local_err = NULL;
5281 /* Find driver and parse its options */
5282 drv = bdrv_find_format(fmt);
5284 error_setg(errp, "Unknown file format '%s'", fmt);
5288 proto_drv = bdrv_find_protocol(filename, true);
5290 error_setg(errp, "Unknown protocol '%s'", filename);
5294 create_options = append_option_parameters(create_options,
5295 drv->create_options);
5296 create_options = append_option_parameters(create_options,
5297 proto_drv->create_options);
5299 /* Create parameter list with default values */
5300 param = parse_option_parameters("", create_options, param);
5302 set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size);
5304 /* Parse -o options */
5306 param = parse_option_parameters(options, create_options, param);
5307 if (param == NULL) {
5308 error_setg(errp, "Invalid options for file format '%s'.", fmt);
5313 if (base_filename) {
5314 if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE,
5316 error_setg(errp, "Backing file not supported for file format '%s'",
5323 if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) {
5324 error_setg(errp, "Backing file format not supported for file "
5325 "format '%s'", fmt);
5330 backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE);
5331 if (backing_file && backing_file->value.s) {
5332 if (!strcmp(filename, backing_file->value.s)) {
5333 error_setg(errp, "Error: Trying to create an image with the "
5334 "same filename as the backing file");
5339 backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT);
5340 if (backing_fmt && backing_fmt->value.s) {
5341 backing_drv = bdrv_find_format(backing_fmt->value.s);
5343 error_setg(errp, "Unknown backing file format '%s'",
5344 backing_fmt->value.s);
5349 // The size for the image must always be specified, with one exception:
5350 // If we are using a backing file, we can obtain the size from there
5351 size = get_option_parameter(param, BLOCK_OPT_SIZE);
5352 if (size && size->value.n == -1) {
5353 if (backing_file && backing_file->value.s) {
5354 BlockDriverState *bs;
5359 /* backing files always opened read-only */
5361 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
5364 ret = bdrv_open(&bs, backing_file->value.s, NULL, NULL, back_flags,
5365 backing_drv, &local_err);
5367 error_setg_errno(errp, -ret, "Could not open '%s': %s",
5368 backing_file->value.s,
5369 error_get_pretty(local_err));
5370 error_free(local_err);
5374 bdrv_get_geometry(bs, &size);
5377 snprintf(buf, sizeof(buf), "%" PRId64, size);
5378 set_option_parameter(param, BLOCK_OPT_SIZE, buf);
5382 error_setg(errp, "Image creation needs a size parameter");
5388 printf("Formatting '%s', fmt=%s ", filename, fmt);
5389 print_option_parameters(param);
5392 ret = bdrv_create(drv, filename, param, &local_err);
5393 if (ret == -EFBIG) {
5394 /* This is generally a better message than whatever the driver would
5395 * deliver (especially because of the cluster_size_hint), since that
5396 * is most probably not much different from "image too large". */
5397 const char *cluster_size_hint = "";
5398 if (get_option_parameter(create_options, BLOCK_OPT_CLUSTER_SIZE)) {
5399 cluster_size_hint = " (try using a larger cluster size)";
5401 error_setg(errp, "The image size is too large for file format '%s'"
5402 "%s", fmt, cluster_size_hint);
5403 error_free(local_err);
5408 free_option_parameters(create_options);
5409 free_option_parameters(param);
5412 error_propagate(errp, local_err);
5416 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
5418 /* Currently BlockDriverState always uses the main loop AioContext */
5419 return qemu_get_aio_context();
5422 void bdrv_add_before_write_notifier(BlockDriverState *bs,
5423 NotifierWithReturn *notifier)
5425 notifier_with_return_list_add(&bs->before_write_notifiers, notifier);
5428 int bdrv_amend_options(BlockDriverState *bs, QEMUOptionParameter *options)
5430 if (bs->drv->bdrv_amend_options == NULL) {
5433 return bs->drv->bdrv_amend_options(bs, options);
5436 /* This function will be called by the bdrv_recurse_is_first_non_filter method
5437 * of block filter and by bdrv_is_first_non_filter.
5438 * It is used to test if the given bs is the candidate or recurse more in the
5441 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
5442 BlockDriverState *candidate)
5444 /* return false if basic checks fails */
5445 if (!bs || !bs->drv) {
5449 /* the code reached a non block filter driver -> check if the bs is
5450 * the same as the candidate. It's the recursion termination condition.
5452 if (!bs->drv->is_filter) {
5453 return bs == candidate;
5455 /* Down this path the driver is a block filter driver */
5457 /* If the block filter recursion method is defined use it to recurse down
5460 if (bs->drv->bdrv_recurse_is_first_non_filter) {
5461 return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate);
5464 /* the driver is a block filter but don't allow to recurse -> return false
5469 /* This function checks if the candidate is the first non filter bs down it's
5470 * bs chain. Since we don't have pointers to parents it explore all bs chains
5471 * from the top. Some filters can choose not to pass down the recursion.
5473 bool bdrv_is_first_non_filter(BlockDriverState *candidate)
5475 BlockDriverState *bs;
5477 /* walk down the bs forest recursively */
5478 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
5481 /* try to recurse in this top level bs */
5482 perm = bdrv_recurse_is_first_non_filter(bs, candidate);
5484 /* candidate is the first non filter */