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_readv(BlockDriverState *bs,
74 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
75 BdrvRequestFlags flags);
76 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
77 int64_t sector_num, int nb_sectors, 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,
210 nb_sectors * BDRV_SECTOR_SIZE);
212 /* if the next request must wait -> do nothing */
213 if (throttle_schedule_timer(&bs->throttle_state, is_write)) {
217 /* else queue next request for execution */
218 qemu_co_queue_next(&bs->throttled_reqs[is_write]);
221 size_t bdrv_opt_mem_align(BlockDriverState *bs)
223 if (!bs || !bs->drv) {
224 /* 4k should be on the safe side */
228 return bs->bl.opt_mem_alignment;
231 /* check if the path starts with "<protocol>:" */
232 static int path_has_protocol(const char *path)
237 if (is_windows_drive(path) ||
238 is_windows_drive_prefix(path)) {
241 p = path + strcspn(path, ":/\\");
243 p = path + strcspn(path, ":/");
249 int path_is_absolute(const char *path)
252 /* specific case for names like: "\\.\d:" */
253 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
256 return (*path == '/' || *path == '\\');
258 return (*path == '/');
262 /* if filename is absolute, just copy it to dest. Otherwise, build a
263 path to it by considering it is relative to base_path. URL are
265 void path_combine(char *dest, int dest_size,
266 const char *base_path,
267 const char *filename)
274 if (path_is_absolute(filename)) {
275 pstrcpy(dest, dest_size, filename);
277 p = strchr(base_path, ':');
282 p1 = strrchr(base_path, '/');
286 p2 = strrchr(base_path, '\\');
298 if (len > dest_size - 1)
300 memcpy(dest, base_path, len);
302 pstrcat(dest, dest_size, filename);
306 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz)
308 if (bs->backing_file[0] == '\0' || path_has_protocol(bs->backing_file)) {
309 pstrcpy(dest, sz, bs->backing_file);
311 path_combine(dest, sz, bs->filename, bs->backing_file);
315 void bdrv_register(BlockDriver *bdrv)
317 /* Block drivers without coroutine functions need emulation */
318 if (!bdrv->bdrv_co_readv) {
319 bdrv->bdrv_co_readv = bdrv_co_readv_em;
320 bdrv->bdrv_co_writev = bdrv_co_writev_em;
322 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
323 * the block driver lacks aio we need to emulate that too.
325 if (!bdrv->bdrv_aio_readv) {
326 /* add AIO emulation layer */
327 bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
328 bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
332 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
335 /* create a new block device (by default it is empty) */
336 BlockDriverState *bdrv_new(const char *device_name)
338 BlockDriverState *bs;
340 bs = g_malloc0(sizeof(BlockDriverState));
341 QLIST_INIT(&bs->dirty_bitmaps);
342 pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
343 if (device_name[0] != '\0') {
344 QTAILQ_INSERT_TAIL(&bdrv_states, bs, device_list);
346 bdrv_iostatus_disable(bs);
347 notifier_list_init(&bs->close_notifiers);
348 notifier_with_return_list_init(&bs->before_write_notifiers);
349 qemu_co_queue_init(&bs->throttled_reqs[0]);
350 qemu_co_queue_init(&bs->throttled_reqs[1]);
356 void bdrv_add_close_notifier(BlockDriverState *bs, Notifier *notify)
358 notifier_list_add(&bs->close_notifiers, notify);
361 BlockDriver *bdrv_find_format(const char *format_name)
364 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
365 if (!strcmp(drv1->format_name, format_name)) {
372 static int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
374 static const char *whitelist_rw[] = {
375 CONFIG_BDRV_RW_WHITELIST
377 static const char *whitelist_ro[] = {
378 CONFIG_BDRV_RO_WHITELIST
382 if (!whitelist_rw[0] && !whitelist_ro[0]) {
383 return 1; /* no whitelist, anything goes */
386 for (p = whitelist_rw; *p; p++) {
387 if (!strcmp(drv->format_name, *p)) {
392 for (p = whitelist_ro; *p; p++) {
393 if (!strcmp(drv->format_name, *p)) {
401 BlockDriver *bdrv_find_whitelisted_format(const char *format_name,
404 BlockDriver *drv = bdrv_find_format(format_name);
405 return drv && bdrv_is_whitelisted(drv, read_only) ? drv : NULL;
408 typedef struct CreateCo {
411 QEMUOptionParameter *options;
416 static void coroutine_fn bdrv_create_co_entry(void *opaque)
418 Error *local_err = NULL;
421 CreateCo *cco = opaque;
424 ret = cco->drv->bdrv_create(cco->filename, cco->options, &local_err);
425 if (error_is_set(&local_err)) {
426 error_propagate(&cco->err, local_err);
431 int bdrv_create(BlockDriver *drv, const char* filename,
432 QEMUOptionParameter *options, Error **errp)
439 .filename = g_strdup(filename),
445 if (!drv->bdrv_create) {
446 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
451 if (qemu_in_coroutine()) {
452 /* Fast-path if already in coroutine context */
453 bdrv_create_co_entry(&cco);
455 co = qemu_coroutine_create(bdrv_create_co_entry);
456 qemu_coroutine_enter(co, &cco);
457 while (cco.ret == NOT_DONE) {
464 if (error_is_set(&cco.err)) {
465 error_propagate(errp, cco.err);
467 error_setg_errno(errp, -ret, "Could not create image");
472 g_free(cco.filename);
476 int bdrv_create_file(const char* filename, QEMUOptionParameter *options,
480 Error *local_err = NULL;
483 drv = bdrv_find_protocol(filename, true);
485 error_setg(errp, "Could not find protocol for file '%s'", filename);
489 ret = bdrv_create(drv, filename, options, &local_err);
490 if (error_is_set(&local_err)) {
491 error_propagate(errp, local_err);
496 int bdrv_refresh_limits(BlockDriverState *bs)
498 BlockDriver *drv = bs->drv;
500 memset(&bs->bl, 0, sizeof(bs->bl));
506 /* Take some limits from the children as a default */
508 bdrv_refresh_limits(bs->file);
509 bs->bl.opt_transfer_length = bs->file->bl.opt_transfer_length;
510 bs->bl.opt_mem_alignment = bs->file->bl.opt_mem_alignment;
512 bs->bl.opt_mem_alignment = 512;
515 if (bs->backing_hd) {
516 bdrv_refresh_limits(bs->backing_hd);
517 bs->bl.opt_transfer_length =
518 MAX(bs->bl.opt_transfer_length,
519 bs->backing_hd->bl.opt_transfer_length);
520 bs->bl.opt_mem_alignment =
521 MAX(bs->bl.opt_mem_alignment,
522 bs->backing_hd->bl.opt_mem_alignment);
525 /* Then let the driver override it */
526 if (drv->bdrv_refresh_limits) {
527 return drv->bdrv_refresh_limits(bs);
534 * Create a uniquely-named empty temporary file.
535 * Return 0 upon success, otherwise a negative errno value.
537 int get_tmp_filename(char *filename, int size)
540 char temp_dir[MAX_PATH];
541 /* GetTempFileName requires that its output buffer (4th param)
542 have length MAX_PATH or greater. */
543 assert(size >= MAX_PATH);
544 return (GetTempPath(MAX_PATH, temp_dir)
545 && GetTempFileName(temp_dir, "qem", 0, filename)
546 ? 0 : -GetLastError());
550 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;
771 * Clear flags that are internal to the block layer before opening the
774 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
777 * Snapshots should be writable.
779 if (bs->is_temporary) {
780 open_flags |= BDRV_O_RDWR;
786 static int bdrv_assign_node_name(BlockDriverState *bs,
787 const char *node_name,
794 /* empty string node name is invalid */
795 if (node_name[0] == '\0') {
796 error_setg(errp, "Empty node name");
800 /* takes care of avoiding duplicates node names */
801 if (bdrv_find_node(node_name)) {
802 error_setg(errp, "Duplicate node name");
806 /* copy node name into the bs and insert it into the graph list */
807 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
808 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
814 * Common part for opening disk images and files
816 * Removes all processed options from *options.
818 static int bdrv_open_common(BlockDriverState *bs, BlockDriverState *file,
819 QDict *options, int flags, BlockDriver *drv, Error **errp)
822 const char *filename;
823 const char *node_name = NULL;
824 Error *local_err = NULL;
827 assert(bs->file == NULL);
828 assert(options != NULL && bs->options != options);
831 filename = file->filename;
833 filename = qdict_get_try_str(options, "filename");
836 trace_bdrv_open_common(bs, filename ?: "", flags, drv->format_name);
838 node_name = qdict_get_try_str(options, "node-name");
839 ret = bdrv_assign_node_name(bs, node_name, errp);
843 qdict_del(options, "node-name");
845 /* bdrv_open() with directly using a protocol as drv. This layer is already
846 * opened, so assign it to bs (while file becomes a closed BlockDriverState)
847 * and return immediately. */
848 if (file != NULL && drv->bdrv_file_open) {
853 bs->open_flags = flags;
854 bs->guest_block_size = 512;
855 bs->zero_beyond_eof = true;
856 open_flags = bdrv_open_flags(bs, flags);
857 bs->read_only = !(open_flags & BDRV_O_RDWR);
859 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
861 !bs->read_only && bdrv_is_whitelisted(drv, true)
862 ? "Driver '%s' can only be used for read-only devices"
863 : "Driver '%s' is not whitelisted",
868 assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
869 if (flags & BDRV_O_COPY_ON_READ) {
870 if (!bs->read_only) {
871 bdrv_enable_copy_on_read(bs);
873 error_setg(errp, "Can't use copy-on-read on read-only device");
878 if (filename != NULL) {
879 pstrcpy(bs->filename, sizeof(bs->filename), filename);
881 bs->filename[0] = '\0';
885 bs->opaque = g_malloc0(drv->instance_size);
887 bs->enable_write_cache = !!(flags & BDRV_O_CACHE_WB);
889 /* Open the image, either directly or using a protocol */
890 if (drv->bdrv_file_open) {
891 assert(file == NULL);
892 assert(!drv->bdrv_needs_filename || filename != NULL);
893 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
896 error_setg(errp, "Can't use '%s' as a block driver for the "
897 "protocol level", drv->format_name);
902 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
906 if (error_is_set(&local_err)) {
907 error_propagate(errp, local_err);
908 } else if (bs->filename[0]) {
909 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
911 error_setg_errno(errp, -ret, "Could not open image");
916 ret = refresh_total_sectors(bs, bs->total_sectors);
918 error_setg_errno(errp, -ret, "Could not refresh total sector count");
922 bdrv_refresh_limits(bs);
925 if (bs->is_temporary) {
926 assert(bs->filename[0] != '\0');
927 unlink(bs->filename);
941 * Opens a file using a protocol (file, host_device, nbd, ...)
943 * options is a QDict of options to pass to the block drivers, or NULL for an
944 * empty set of options. The reference to the QDict belongs to the block layer
945 * after the call (even on failure), so if the caller intends to reuse the
946 * dictionary, it needs to use QINCREF() before calling bdrv_file_open.
948 int bdrv_file_open(BlockDriverState **pbs, const char *filename,
949 const char *reference, QDict *options, int flags,
952 BlockDriverState *bs = NULL;
955 bool allow_protocol_prefix = false;
956 Error *local_err = NULL;
959 /* NULL means an empty set of options */
960 if (options == NULL) {
961 options = qdict_new();
965 if (filename || qdict_size(options)) {
966 error_setg(errp, "Cannot reference an existing block device with "
967 "additional options or a new filename");
972 bs = bdrv_find(reference);
974 error_setg(errp, "Cannot find block device '%s'", reference);
983 bs->options = options;
984 options = qdict_clone_shallow(options);
986 /* Fetch the file name from the options QDict if necessary */
988 filename = qdict_get_try_str(options, "filename");
989 } else if (filename && !qdict_haskey(options, "filename")) {
990 qdict_put(options, "filename", qstring_from_str(filename));
991 allow_protocol_prefix = true;
993 error_setg(errp, "Can't specify 'file' and 'filename' options at the "
999 /* Find the right block driver */
1000 drvname = qdict_get_try_str(options, "driver");
1002 drv = bdrv_find_format(drvname);
1004 error_setg(errp, "Unknown driver '%s'", drvname);
1006 qdict_del(options, "driver");
1007 } else if (filename) {
1008 drv = bdrv_find_protocol(filename, allow_protocol_prefix);
1010 error_setg(errp, "Unknown protocol");
1013 error_setg(errp, "Must specify either driver or file");
1018 /* errp has been set already */
1023 /* Parse the filename and open it */
1024 if (drv->bdrv_parse_filename && filename) {
1025 drv->bdrv_parse_filename(filename, options, &local_err);
1026 if (error_is_set(&local_err)) {
1027 error_propagate(errp, local_err);
1031 qdict_del(options, "filename");
1032 } else if (drv->bdrv_needs_filename && !filename) {
1033 error_setg(errp, "The '%s' block driver requires a file name",
1039 if (!drv->bdrv_file_open) {
1040 ret = bdrv_open(bs, filename, options, flags, drv, &local_err);
1043 ret = bdrv_open_common(bs, NULL, options, flags, drv, &local_err);
1046 error_propagate(errp, local_err);
1050 /* Check if any unknown options were used */
1051 if (options && (qdict_size(options) != 0)) {
1052 const QDictEntry *entry = qdict_first(options);
1053 error_setg(errp, "Block protocol '%s' doesn't support the option '%s'",
1054 drv->format_name, entry->key);
1067 QDECREF(bs->options);
1074 * Opens the backing file for a BlockDriverState if not yet open
1076 * options is a QDict of options to pass to the block drivers, or NULL for an
1077 * empty set of options. The reference to the QDict is transferred to this
1078 * function (even on failure), so if the caller intends to reuse the dictionary,
1079 * it needs to use QINCREF() before calling bdrv_file_open.
1081 int bdrv_open_backing_file(BlockDriverState *bs, QDict *options, Error **errp)
1083 char backing_filename[PATH_MAX];
1084 int back_flags, ret;
1085 BlockDriver *back_drv = NULL;
1086 Error *local_err = NULL;
1088 if (bs->backing_hd != NULL) {
1093 /* NULL means an empty set of options */
1094 if (options == NULL) {
1095 options = qdict_new();
1098 bs->open_flags &= ~BDRV_O_NO_BACKING;
1099 if (qdict_haskey(options, "file.filename")) {
1100 backing_filename[0] = '\0';
1101 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
1105 bdrv_get_full_backing_filename(bs, backing_filename,
1106 sizeof(backing_filename));
1109 bs->backing_hd = bdrv_new("");
1111 if (bs->backing_format[0] != '\0') {
1112 back_drv = bdrv_find_format(bs->backing_format);
1115 /* backing files always opened read-only */
1116 back_flags = bs->open_flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT |
1117 BDRV_O_COPY_ON_READ);
1119 ret = bdrv_open(bs->backing_hd,
1120 *backing_filename ? backing_filename : NULL, options,
1121 back_flags, back_drv, &local_err);
1123 bdrv_unref(bs->backing_hd);
1124 bs->backing_hd = NULL;
1125 bs->open_flags |= BDRV_O_NO_BACKING;
1126 error_setg(errp, "Could not open backing file: %s",
1127 error_get_pretty(local_err));
1128 error_free(local_err);
1132 if (bs->backing_hd->file) {
1133 pstrcpy(bs->backing_file, sizeof(bs->backing_file),
1134 bs->backing_hd->file->filename);
1137 /* Recalculate the BlockLimits with the backing file */
1138 bdrv_refresh_limits(bs);
1144 * Opens a disk image whose options are given as BlockdevRef in another block
1147 * If force_raw is true, bdrv_file_open() will be used, thereby preventing any
1148 * image format auto-detection. If it is false and a filename is given,
1149 * bdrv_open() will be used for auto-detection.
1151 * If allow_none is true, no image will be opened if filename is false and no
1152 * BlockdevRef is given. *pbs will remain unchanged and 0 will be returned.
1154 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
1155 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
1156 * itself, all options starting with "${bdref_key}." are considered part of the
1159 * The BlockdevRef will be removed from the options QDict.
1161 int bdrv_open_image(BlockDriverState **pbs, const char *filename,
1162 QDict *options, const char *bdref_key, int flags,
1163 bool force_raw, bool allow_none, Error **errp)
1165 QDict *image_options;
1167 char *bdref_key_dot;
1168 const char *reference;
1170 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
1171 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
1172 g_free(bdref_key_dot);
1174 reference = qdict_get_try_str(options, bdref_key);
1175 if (!filename && !reference && !qdict_size(image_options)) {
1179 error_setg(errp, "A block device must be specified for \"%s\"",
1186 if (filename && !force_raw) {
1187 /* If a filename is given and the block driver should be detected
1188 automatically (instead of using none), use bdrv_open() in order to do
1189 that auto-detection. */
1190 BlockDriverState *bs;
1193 error_setg(errp, "Cannot reference an existing block device while "
1194 "giving a filename");
1200 ret = bdrv_open(bs, filename, image_options, flags, NULL, errp);
1207 ret = bdrv_file_open(pbs, filename, reference, image_options, flags,
1212 qdict_del(options, bdref_key);
1217 * Opens a disk image (raw, qcow2, vmdk, ...)
1219 * options is a QDict of options to pass to the block drivers, or NULL for an
1220 * empty set of options. The reference to the QDict belongs to the block layer
1221 * after the call (even on failure), so if the caller intends to reuse the
1222 * dictionary, it needs to use QINCREF() before calling bdrv_open.
1224 int bdrv_open(BlockDriverState *bs, const char *filename, QDict *options,
1225 int flags, BlockDriver *drv, Error **errp)
1228 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
1229 char tmp_filename[PATH_MAX + 1];
1230 BlockDriverState *file = NULL;
1231 const char *drvname;
1232 Error *local_err = NULL;
1234 /* NULL means an empty set of options */
1235 if (options == NULL) {
1236 options = qdict_new();
1239 bs->options = options;
1240 options = qdict_clone_shallow(options);
1242 /* For snapshot=on, create a temporary qcow2 overlay */
1243 if (flags & BDRV_O_SNAPSHOT) {
1244 BlockDriverState *bs1;
1246 BlockDriver *bdrv_qcow2;
1247 QEMUOptionParameter *create_options;
1248 QDict *snapshot_options;
1250 /* if snapshot, we create a temporary backing file and open it
1251 instead of opening 'filename' directly */
1253 /* Get the required size from the image */
1256 ret = bdrv_open(bs1, filename, options, BDRV_O_NO_BACKING,
1262 total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
1266 /* Create the temporary image */
1267 ret = get_tmp_filename(tmp_filename, sizeof(tmp_filename));
1269 error_setg_errno(errp, -ret, "Could not get temporary filename");
1273 bdrv_qcow2 = bdrv_find_format("qcow2");
1274 create_options = parse_option_parameters("", bdrv_qcow2->create_options,
1277 set_option_parameter_int(create_options, BLOCK_OPT_SIZE, total_size);
1279 ret = bdrv_create(bdrv_qcow2, tmp_filename, create_options, &local_err);
1280 free_option_parameters(create_options);
1282 error_setg_errno(errp, -ret, "Could not create temporary overlay "
1283 "'%s': %s", tmp_filename,
1284 error_get_pretty(local_err));
1285 error_free(local_err);
1290 /* Prepare a new options QDict for the temporary file, where user
1291 * options refer to the backing file */
1293 qdict_put(options, "file.filename", qstring_from_str(filename));
1296 qdict_put(options, "driver", qstring_from_str(drv->format_name));
1299 snapshot_options = qdict_new();
1300 qdict_put(snapshot_options, "backing", options);
1301 qdict_flatten(snapshot_options);
1303 bs->options = snapshot_options;
1304 options = qdict_clone_shallow(bs->options);
1306 filename = tmp_filename;
1308 bs->is_temporary = 1;
1311 /* Open image file without format layer */
1312 if (flags & BDRV_O_RDWR) {
1313 flags |= BDRV_O_ALLOW_RDWR;
1316 ret = bdrv_open_image(&file, filename, options, "file",
1317 bdrv_open_flags(bs, flags | BDRV_O_UNMAP), true, true,
1323 /* Find the right image format driver */
1324 drvname = qdict_get_try_str(options, "driver");
1326 drv = bdrv_find_format(drvname);
1327 qdict_del(options, "driver");
1329 error_setg(errp, "Invalid driver: '%s'", drvname);
1331 goto unlink_and_fail;
1337 ret = find_image_format(file, filename, &drv, &local_err);
1339 error_setg(errp, "Must specify either driver or file");
1341 goto unlink_and_fail;
1346 goto unlink_and_fail;
1349 /* Open the image */
1350 ret = bdrv_open_common(bs, file, options, flags, drv, &local_err);
1352 goto unlink_and_fail;
1355 if (file && (bs->file != file)) {
1360 /* If there is a backing file, use it */
1361 if ((flags & BDRV_O_NO_BACKING) == 0) {
1362 QDict *backing_options;
1364 qdict_extract_subqdict(options, &backing_options, "backing.");
1365 ret = bdrv_open_backing_file(bs, backing_options, &local_err);
1367 goto close_and_fail;
1371 /* Check if any unknown options were used */
1372 if (qdict_size(options) != 0) {
1373 const QDictEntry *entry = qdict_first(options);
1374 error_setg(errp, "Block format '%s' used by device '%s' doesn't "
1375 "support the option '%s'", drv->format_name, bs->device_name,
1379 goto close_and_fail;
1383 if (!bdrv_key_required(bs)) {
1384 bdrv_dev_change_media_cb(bs, true);
1393 if (bs->is_temporary) {
1397 QDECREF(bs->options);
1400 if (error_is_set(&local_err)) {
1401 error_propagate(errp, local_err);
1408 if (error_is_set(&local_err)) {
1409 error_propagate(errp, local_err);
1414 typedef struct BlockReopenQueueEntry {
1416 BDRVReopenState state;
1417 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
1418 } BlockReopenQueueEntry;
1421 * Adds a BlockDriverState to a simple queue for an atomic, transactional
1422 * reopen of multiple devices.
1424 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
1425 * already performed, or alternatively may be NULL a new BlockReopenQueue will
1426 * be created and initialized. This newly created BlockReopenQueue should be
1427 * passed back in for subsequent calls that are intended to be of the same
1430 * bs is the BlockDriverState to add to the reopen queue.
1432 * flags contains the open flags for the associated bs
1434 * returns a pointer to bs_queue, which is either the newly allocated
1435 * bs_queue, or the existing bs_queue being used.
1438 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
1439 BlockDriverState *bs, int flags)
1443 BlockReopenQueueEntry *bs_entry;
1444 if (bs_queue == NULL) {
1445 bs_queue = g_new0(BlockReopenQueue, 1);
1446 QSIMPLEQ_INIT(bs_queue);
1450 bdrv_reopen_queue(bs_queue, bs->file, flags);
1453 bs_entry = g_new0(BlockReopenQueueEntry, 1);
1454 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
1456 bs_entry->state.bs = bs;
1457 bs_entry->state.flags = flags;
1463 * Reopen multiple BlockDriverStates atomically & transactionally.
1465 * The queue passed in (bs_queue) must have been built up previous
1466 * via bdrv_reopen_queue().
1468 * Reopens all BDS specified in the queue, with the appropriate
1469 * flags. All devices are prepared for reopen, and failure of any
1470 * device will cause all device changes to be abandonded, and intermediate
1473 * If all devices prepare successfully, then the changes are committed
1477 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
1480 BlockReopenQueueEntry *bs_entry, *next;
1481 Error *local_err = NULL;
1483 assert(bs_queue != NULL);
1487 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
1488 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
1489 error_propagate(errp, local_err);
1492 bs_entry->prepared = true;
1495 /* If we reach this point, we have success and just need to apply the
1498 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
1499 bdrv_reopen_commit(&bs_entry->state);
1505 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
1506 if (ret && bs_entry->prepared) {
1507 bdrv_reopen_abort(&bs_entry->state);
1516 /* Reopen a single BlockDriverState with the specified flags. */
1517 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
1520 Error *local_err = NULL;
1521 BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, bdrv_flags);
1523 ret = bdrv_reopen_multiple(queue, &local_err);
1524 if (local_err != NULL) {
1525 error_propagate(errp, local_err);
1532 * Prepares a BlockDriverState for reopen. All changes are staged in the
1533 * 'opaque' field of the BDRVReopenState, which is used and allocated by
1534 * the block driver layer .bdrv_reopen_prepare()
1536 * bs is the BlockDriverState to reopen
1537 * flags are the new open flags
1538 * queue is the reopen queue
1540 * Returns 0 on success, non-zero on error. On error errp will be set
1543 * On failure, bdrv_reopen_abort() will be called to clean up any data.
1544 * It is the responsibility of the caller to then call the abort() or
1545 * commit() for any other BDS that have been left in a prepare() state
1548 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
1552 Error *local_err = NULL;
1555 assert(reopen_state != NULL);
1556 assert(reopen_state->bs->drv != NULL);
1557 drv = reopen_state->bs->drv;
1559 /* if we are to stay read-only, do not allow permission change
1561 if (!(reopen_state->bs->open_flags & BDRV_O_ALLOW_RDWR) &&
1562 reopen_state->flags & BDRV_O_RDWR) {
1563 error_set(errp, QERR_DEVICE_IS_READ_ONLY,
1564 reopen_state->bs->device_name);
1569 ret = bdrv_flush(reopen_state->bs);
1571 error_set(errp, ERROR_CLASS_GENERIC_ERROR, "Error (%s) flushing drive",
1576 if (drv->bdrv_reopen_prepare) {
1577 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
1579 if (local_err != NULL) {
1580 error_propagate(errp, local_err);
1582 error_setg(errp, "failed while preparing to reopen image '%s'",
1583 reopen_state->bs->filename);
1588 /* It is currently mandatory to have a bdrv_reopen_prepare()
1589 * handler for each supported drv. */
1590 error_set(errp, QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
1591 drv->format_name, reopen_state->bs->device_name,
1592 "reopening of file");
1604 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
1605 * makes them final by swapping the staging BlockDriverState contents into
1606 * the active BlockDriverState contents.
1608 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
1612 assert(reopen_state != NULL);
1613 drv = reopen_state->bs->drv;
1614 assert(drv != NULL);
1616 /* If there are any driver level actions to take */
1617 if (drv->bdrv_reopen_commit) {
1618 drv->bdrv_reopen_commit(reopen_state);
1621 /* set BDS specific flags now */
1622 reopen_state->bs->open_flags = reopen_state->flags;
1623 reopen_state->bs->enable_write_cache = !!(reopen_state->flags &
1625 reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
1627 bdrv_refresh_limits(reopen_state->bs);
1631 * Abort the reopen, and delete and free the staged changes in
1634 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
1638 assert(reopen_state != NULL);
1639 drv = reopen_state->bs->drv;
1640 assert(drv != NULL);
1642 if (drv->bdrv_reopen_abort) {
1643 drv->bdrv_reopen_abort(reopen_state);
1648 void bdrv_close(BlockDriverState *bs)
1651 block_job_cancel_sync(bs->job);
1653 bdrv_drain_all(); /* complete I/O */
1655 bdrv_drain_all(); /* in case flush left pending I/O */
1656 notifier_list_notify(&bs->close_notifiers, bs);
1659 if (bs->backing_hd) {
1660 bdrv_unref(bs->backing_hd);
1661 bs->backing_hd = NULL;
1663 bs->drv->bdrv_close(bs);
1666 if (bs->is_temporary) {
1667 unlink(bs->filename);
1672 bs->copy_on_read = 0;
1673 bs->backing_file[0] = '\0';
1674 bs->backing_format[0] = '\0';
1675 bs->total_sectors = 0;
1680 bs->zero_beyond_eof = false;
1681 QDECREF(bs->options);
1684 if (bs->file != NULL) {
1685 bdrv_unref(bs->file);
1690 bdrv_dev_change_media_cb(bs, false);
1692 /*throttling disk I/O limits*/
1693 if (bs->io_limits_enabled) {
1694 bdrv_io_limits_disable(bs);
1698 void bdrv_close_all(void)
1700 BlockDriverState *bs;
1702 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
1707 /* Check if any requests are in-flight (including throttled requests) */
1708 static bool bdrv_requests_pending(BlockDriverState *bs)
1710 if (!QLIST_EMPTY(&bs->tracked_requests)) {
1713 if (!qemu_co_queue_empty(&bs->throttled_reqs[0])) {
1716 if (!qemu_co_queue_empty(&bs->throttled_reqs[1])) {
1719 if (bs->file && bdrv_requests_pending(bs->file)) {
1722 if (bs->backing_hd && bdrv_requests_pending(bs->backing_hd)) {
1728 static bool bdrv_requests_pending_all(void)
1730 BlockDriverState *bs;
1731 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
1732 if (bdrv_requests_pending(bs)) {
1740 * Wait for pending requests to complete across all BlockDriverStates
1742 * This function does not flush data to disk, use bdrv_flush_all() for that
1743 * after calling this function.
1745 * Note that completion of an asynchronous I/O operation can trigger any
1746 * number of other I/O operations on other devices---for example a coroutine
1747 * can be arbitrarily complex and a constant flow of I/O can come until the
1748 * coroutine is complete. Because of this, it is not possible to have a
1749 * function to drain a single device's I/O queue.
1751 void bdrv_drain_all(void)
1753 /* Always run first iteration so any pending completion BHs run */
1755 BlockDriverState *bs;
1758 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
1759 bdrv_start_throttled_reqs(bs);
1762 busy = bdrv_requests_pending_all();
1763 busy |= aio_poll(qemu_get_aio_context(), busy);
1767 /* make a BlockDriverState anonymous by removing from bdrv_state and
1768 * graph_bdrv_state list.
1769 Also, NULL terminate the device_name to prevent double remove */
1770 void bdrv_make_anon(BlockDriverState *bs)
1772 if (bs->device_name[0] != '\0') {
1773 QTAILQ_REMOVE(&bdrv_states, bs, device_list);
1775 bs->device_name[0] = '\0';
1776 if (bs->node_name[0] != '\0') {
1777 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
1779 bs->node_name[0] = '\0';
1782 static void bdrv_rebind(BlockDriverState *bs)
1784 if (bs->drv && bs->drv->bdrv_rebind) {
1785 bs->drv->bdrv_rebind(bs);
1789 static void bdrv_move_feature_fields(BlockDriverState *bs_dest,
1790 BlockDriverState *bs_src)
1792 /* move some fields that need to stay attached to the device */
1793 bs_dest->open_flags = bs_src->open_flags;
1796 bs_dest->dev_ops = bs_src->dev_ops;
1797 bs_dest->dev_opaque = bs_src->dev_opaque;
1798 bs_dest->dev = bs_src->dev;
1799 bs_dest->guest_block_size = bs_src->guest_block_size;
1800 bs_dest->copy_on_read = bs_src->copy_on_read;
1802 bs_dest->enable_write_cache = bs_src->enable_write_cache;
1804 /* i/o throttled req */
1805 memcpy(&bs_dest->throttle_state,
1806 &bs_src->throttle_state,
1807 sizeof(ThrottleState));
1808 bs_dest->throttled_reqs[0] = bs_src->throttled_reqs[0];
1809 bs_dest->throttled_reqs[1] = bs_src->throttled_reqs[1];
1810 bs_dest->io_limits_enabled = bs_src->io_limits_enabled;
1813 bs_dest->on_read_error = bs_src->on_read_error;
1814 bs_dest->on_write_error = bs_src->on_write_error;
1817 bs_dest->iostatus_enabled = bs_src->iostatus_enabled;
1818 bs_dest->iostatus = bs_src->iostatus;
1821 bs_dest->dirty_bitmaps = bs_src->dirty_bitmaps;
1823 /* reference count */
1824 bs_dest->refcnt = bs_src->refcnt;
1827 bs_dest->in_use = bs_src->in_use;
1828 bs_dest->job = bs_src->job;
1830 /* keep the same entry in bdrv_states */
1831 pstrcpy(bs_dest->device_name, sizeof(bs_dest->device_name),
1832 bs_src->device_name);
1833 bs_dest->device_list = bs_src->device_list;
1835 /* keep the same entry in graph_bdrv_states
1836 * We do want to swap name but don't want to swap linked list entries
1838 bs_dest->node_list = bs_src->node_list;
1842 * Swap bs contents for two image chains while they are live,
1843 * while keeping required fields on the BlockDriverState that is
1844 * actually attached to a device.
1846 * This will modify the BlockDriverState fields, and swap contents
1847 * between bs_new and bs_old. Both bs_new and bs_old are modified.
1849 * bs_new is required to be anonymous.
1851 * This function does not create any image files.
1853 void bdrv_swap(BlockDriverState *bs_new, BlockDriverState *bs_old)
1855 BlockDriverState tmp;
1857 /* bs_new must be anonymous and shouldn't have anything fancy enabled */
1858 assert(bs_new->device_name[0] == '\0');
1859 assert(QLIST_EMPTY(&bs_new->dirty_bitmaps));
1860 assert(bs_new->job == NULL);
1861 assert(bs_new->dev == NULL);
1862 assert(bs_new->in_use == 0);
1863 assert(bs_new->io_limits_enabled == false);
1864 assert(!throttle_have_timer(&bs_new->throttle_state));
1870 /* there are some fields that should not be swapped, move them back */
1871 bdrv_move_feature_fields(&tmp, bs_old);
1872 bdrv_move_feature_fields(bs_old, bs_new);
1873 bdrv_move_feature_fields(bs_new, &tmp);
1875 /* bs_new shouldn't be in bdrv_states even after the swap! */
1876 assert(bs_new->device_name[0] == '\0');
1878 /* Check a few fields that should remain attached to the device */
1879 assert(bs_new->dev == NULL);
1880 assert(bs_new->job == NULL);
1881 assert(bs_new->in_use == 0);
1882 assert(bs_new->io_limits_enabled == false);
1883 assert(!throttle_have_timer(&bs_new->throttle_state));
1885 bdrv_rebind(bs_new);
1886 bdrv_rebind(bs_old);
1890 * Add new bs contents at the top of an image chain while the chain is
1891 * live, while keeping required fields on the top layer.
1893 * This will modify the BlockDriverState fields, and swap contents
1894 * between bs_new and bs_top. Both bs_new and bs_top are modified.
1896 * bs_new is required to be anonymous.
1898 * This function does not create any image files.
1900 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
1902 bdrv_swap(bs_new, bs_top);
1904 /* The contents of 'tmp' will become bs_top, as we are
1905 * swapping bs_new and bs_top contents. */
1906 bs_top->backing_hd = bs_new;
1907 bs_top->open_flags &= ~BDRV_O_NO_BACKING;
1908 pstrcpy(bs_top->backing_file, sizeof(bs_top->backing_file),
1910 pstrcpy(bs_top->backing_format, sizeof(bs_top->backing_format),
1911 bs_new->drv ? bs_new->drv->format_name : "");
1914 static void bdrv_delete(BlockDriverState *bs)
1918 assert(!bs->in_use);
1919 assert(!bs->refcnt);
1920 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
1924 /* remove from list, if necessary */
1930 int bdrv_attach_dev(BlockDriverState *bs, void *dev)
1931 /* TODO change to DeviceState *dev when all users are qdevified */
1937 bdrv_iostatus_reset(bs);
1941 /* TODO qdevified devices don't use this, remove when devices are qdevified */
1942 void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev)
1944 if (bdrv_attach_dev(bs, dev) < 0) {
1949 void bdrv_detach_dev(BlockDriverState *bs, void *dev)
1950 /* TODO change to DeviceState *dev when all users are qdevified */
1952 assert(bs->dev == dev);
1955 bs->dev_opaque = NULL;
1956 bs->guest_block_size = 512;
1959 /* TODO change to return DeviceState * when all users are qdevified */
1960 void *bdrv_get_attached_dev(BlockDriverState *bs)
1965 void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops,
1969 bs->dev_opaque = opaque;
1972 void bdrv_emit_qmp_error_event(const BlockDriverState *bdrv,
1973 enum MonitorEvent ev,
1974 BlockErrorAction action, bool is_read)
1977 const char *action_str;
1980 case BDRV_ACTION_REPORT:
1981 action_str = "report";
1983 case BDRV_ACTION_IGNORE:
1984 action_str = "ignore";
1986 case BDRV_ACTION_STOP:
1987 action_str = "stop";
1993 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1996 is_read ? "read" : "write");
1997 monitor_protocol_event(ev, data);
1999 qobject_decref(data);
2002 static void bdrv_emit_qmp_eject_event(BlockDriverState *bs, bool ejected)
2006 data = qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }",
2007 bdrv_get_device_name(bs), ejected);
2008 monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED, data);
2010 qobject_decref(data);
2013 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load)
2015 if (bs->dev_ops && bs->dev_ops->change_media_cb) {
2016 bool tray_was_closed = !bdrv_dev_is_tray_open(bs);
2017 bs->dev_ops->change_media_cb(bs->dev_opaque, load);
2018 if (tray_was_closed) {
2020 bdrv_emit_qmp_eject_event(bs, true);
2024 bdrv_emit_qmp_eject_event(bs, false);
2029 bool bdrv_dev_has_removable_media(BlockDriverState *bs)
2031 return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb);
2034 void bdrv_dev_eject_request(BlockDriverState *bs, bool force)
2036 if (bs->dev_ops && bs->dev_ops->eject_request_cb) {
2037 bs->dev_ops->eject_request_cb(bs->dev_opaque, force);
2041 bool bdrv_dev_is_tray_open(BlockDriverState *bs)
2043 if (bs->dev_ops && bs->dev_ops->is_tray_open) {
2044 return bs->dev_ops->is_tray_open(bs->dev_opaque);
2049 static void bdrv_dev_resize_cb(BlockDriverState *bs)
2051 if (bs->dev_ops && bs->dev_ops->resize_cb) {
2052 bs->dev_ops->resize_cb(bs->dev_opaque);
2056 bool bdrv_dev_is_medium_locked(BlockDriverState *bs)
2058 if (bs->dev_ops && bs->dev_ops->is_medium_locked) {
2059 return bs->dev_ops->is_medium_locked(bs->dev_opaque);
2065 * Run consistency checks on an image
2067 * Returns 0 if the check could be completed (it doesn't mean that the image is
2068 * free of errors) or -errno when an internal error occurred. The results of the
2069 * check are stored in res.
2071 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix)
2073 if (bs->drv->bdrv_check == NULL) {
2077 memset(res, 0, sizeof(*res));
2078 return bs->drv->bdrv_check(bs, res, fix);
2081 #define COMMIT_BUF_SECTORS 2048
2083 /* commit COW file into the raw image */
2084 int bdrv_commit(BlockDriverState *bs)
2086 BlockDriver *drv = bs->drv;
2087 int64_t sector, total_sectors, length, backing_length;
2088 int n, ro, open_flags;
2090 uint8_t *buf = NULL;
2091 char filename[PATH_MAX];
2096 if (!bs->backing_hd) {
2100 if (bdrv_in_use(bs) || bdrv_in_use(bs->backing_hd)) {
2104 ro = bs->backing_hd->read_only;
2105 /* Use pstrcpy (not strncpy): filename must be NUL-terminated. */
2106 pstrcpy(filename, sizeof(filename), bs->backing_hd->filename);
2107 open_flags = bs->backing_hd->open_flags;
2110 if (bdrv_reopen(bs->backing_hd, open_flags | BDRV_O_RDWR, NULL)) {
2115 length = bdrv_getlength(bs);
2121 backing_length = bdrv_getlength(bs->backing_hd);
2122 if (backing_length < 0) {
2123 ret = backing_length;
2127 /* If our top snapshot is larger than the backing file image,
2128 * grow the backing file image if possible. If not possible,
2129 * we must return an error */
2130 if (length > backing_length) {
2131 ret = bdrv_truncate(bs->backing_hd, length);
2137 total_sectors = length >> BDRV_SECTOR_BITS;
2138 buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
2140 for (sector = 0; sector < total_sectors; sector += n) {
2141 ret = bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n);
2146 ret = bdrv_read(bs, sector, buf, n);
2151 ret = bdrv_write(bs->backing_hd, sector, buf, n);
2158 if (drv->bdrv_make_empty) {
2159 ret = drv->bdrv_make_empty(bs);
2167 * Make sure all data we wrote to the backing device is actually
2170 if (bs->backing_hd) {
2171 bdrv_flush(bs->backing_hd);
2179 /* ignoring error return here */
2180 bdrv_reopen(bs->backing_hd, open_flags & ~BDRV_O_RDWR, NULL);
2186 int bdrv_commit_all(void)
2188 BlockDriverState *bs;
2190 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
2191 if (bs->drv && bs->backing_hd) {
2192 int ret = bdrv_commit(bs);
2202 * Remove an active request from the tracked requests list
2204 * This function should be called when a tracked request is completing.
2206 static void tracked_request_end(BdrvTrackedRequest *req)
2208 QLIST_REMOVE(req, list);
2209 qemu_co_queue_restart_all(&req->wait_queue);
2213 * Add an active request to the tracked requests list
2215 static void tracked_request_begin(BdrvTrackedRequest *req,
2216 BlockDriverState *bs,
2218 int nb_sectors, bool is_write)
2220 *req = (BdrvTrackedRequest){
2222 .sector_num = sector_num,
2223 .nb_sectors = nb_sectors,
2224 .is_write = is_write,
2225 .co = qemu_coroutine_self(),
2228 qemu_co_queue_init(&req->wait_queue);
2230 QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
2234 * Round a region to cluster boundaries
2236 void bdrv_round_to_clusters(BlockDriverState *bs,
2237 int64_t sector_num, int nb_sectors,
2238 int64_t *cluster_sector_num,
2239 int *cluster_nb_sectors)
2241 BlockDriverInfo bdi;
2243 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
2244 *cluster_sector_num = sector_num;
2245 *cluster_nb_sectors = nb_sectors;
2247 int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE;
2248 *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c);
2249 *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num +
2254 static bool tracked_request_overlaps(BdrvTrackedRequest *req,
2255 int64_t sector_num, int nb_sectors) {
2257 if (sector_num >= req->sector_num + req->nb_sectors) {
2261 if (req->sector_num >= sector_num + nb_sectors) {
2267 static void coroutine_fn wait_for_overlapping_requests(BlockDriverState *bs,
2268 int64_t sector_num, int nb_sectors)
2270 BdrvTrackedRequest *req;
2271 int64_t cluster_sector_num;
2272 int cluster_nb_sectors;
2275 /* If we touch the same cluster it counts as an overlap. This guarantees
2276 * that allocating writes will be serialized and not race with each other
2277 * for the same cluster. For example, in copy-on-read it ensures that the
2278 * CoR read and write operations are atomic and guest writes cannot
2279 * interleave between them.
2281 bdrv_round_to_clusters(bs, sector_num, nb_sectors,
2282 &cluster_sector_num, &cluster_nb_sectors);
2286 QLIST_FOREACH(req, &bs->tracked_requests, list) {
2287 if (tracked_request_overlaps(req, cluster_sector_num,
2288 cluster_nb_sectors)) {
2289 /* Hitting this means there was a reentrant request, for
2290 * example, a block driver issuing nested requests. This must
2291 * never happen since it means deadlock.
2293 assert(qemu_coroutine_self() != req->co);
2295 qemu_co_queue_wait(&req->wait_queue);
2306 * -EINVAL - backing format specified, but no file
2307 * -ENOSPC - can't update the backing file because no space is left in the
2309 * -ENOTSUP - format driver doesn't support changing the backing file
2311 int bdrv_change_backing_file(BlockDriverState *bs,
2312 const char *backing_file, const char *backing_fmt)
2314 BlockDriver *drv = bs->drv;
2317 /* Backing file format doesn't make sense without a backing file */
2318 if (backing_fmt && !backing_file) {
2322 if (drv->bdrv_change_backing_file != NULL) {
2323 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
2329 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
2330 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
2336 * Finds the image layer in the chain that has 'bs' as its backing file.
2338 * active is the current topmost image.
2340 * Returns NULL if bs is not found in active's image chain,
2341 * or if active == bs.
2343 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
2344 BlockDriverState *bs)
2346 BlockDriverState *overlay = NULL;
2347 BlockDriverState *intermediate;
2349 assert(active != NULL);
2352 /* if bs is the same as active, then by definition it has no overlay
2358 intermediate = active;
2359 while (intermediate->backing_hd) {
2360 if (intermediate->backing_hd == bs) {
2361 overlay = intermediate;
2364 intermediate = intermediate->backing_hd;
2370 typedef struct BlkIntermediateStates {
2371 BlockDriverState *bs;
2372 QSIMPLEQ_ENTRY(BlkIntermediateStates) entry;
2373 } BlkIntermediateStates;
2377 * Drops images above 'base' up to and including 'top', and sets the image
2378 * above 'top' to have base as its backing file.
2380 * Requires that the overlay to 'top' is opened r/w, so that the backing file
2381 * information in 'bs' can be properly updated.
2383 * E.g., this will convert the following chain:
2384 * bottom <- base <- intermediate <- top <- active
2388 * bottom <- base <- active
2390 * It is allowed for bottom==base, in which case it converts:
2392 * base <- intermediate <- top <- active
2399 * if active == top, that is considered an error
2402 int bdrv_drop_intermediate(BlockDriverState *active, BlockDriverState *top,
2403 BlockDriverState *base)
2405 BlockDriverState *intermediate;
2406 BlockDriverState *base_bs = NULL;
2407 BlockDriverState *new_top_bs = NULL;
2408 BlkIntermediateStates *intermediate_state, *next;
2411 QSIMPLEQ_HEAD(states_to_delete, BlkIntermediateStates) states_to_delete;
2412 QSIMPLEQ_INIT(&states_to_delete);
2414 if (!top->drv || !base->drv) {
2418 new_top_bs = bdrv_find_overlay(active, top);
2420 if (new_top_bs == NULL) {
2421 /* we could not find the image above 'top', this is an error */
2425 /* special case of new_top_bs->backing_hd already pointing to base - nothing
2426 * to do, no intermediate images */
2427 if (new_top_bs->backing_hd == base) {
2434 /* now we will go down through the list, and add each BDS we find
2435 * into our deletion queue, until we hit the 'base'
2437 while (intermediate) {
2438 intermediate_state = g_malloc0(sizeof(BlkIntermediateStates));
2439 intermediate_state->bs = intermediate;
2440 QSIMPLEQ_INSERT_TAIL(&states_to_delete, intermediate_state, entry);
2442 if (intermediate->backing_hd == base) {
2443 base_bs = intermediate->backing_hd;
2446 intermediate = intermediate->backing_hd;
2448 if (base_bs == NULL) {
2449 /* something went wrong, we did not end at the base. safely
2450 * unravel everything, and exit with error */
2454 /* success - we can delete the intermediate states, and link top->base */
2455 ret = bdrv_change_backing_file(new_top_bs, base_bs->filename,
2456 base_bs->drv ? base_bs->drv->format_name : "");
2460 new_top_bs->backing_hd = base_bs;
2462 bdrv_refresh_limits(new_top_bs);
2464 QSIMPLEQ_FOREACH_SAFE(intermediate_state, &states_to_delete, entry, next) {
2465 /* so that bdrv_close() does not recursively close the chain */
2466 intermediate_state->bs->backing_hd = NULL;
2467 bdrv_unref(intermediate_state->bs);
2472 QSIMPLEQ_FOREACH_SAFE(intermediate_state, &states_to_delete, entry, next) {
2473 g_free(intermediate_state);
2479 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
2484 if (!bdrv_is_inserted(bs))
2490 len = bdrv_getlength(bs);
2495 if ((offset > len) || (len - offset < size))
2501 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
2504 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
2505 nb_sectors * BDRV_SECTOR_SIZE);
2508 typedef struct RwCo {
2509 BlockDriverState *bs;
2515 BdrvRequestFlags flags;
2518 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
2520 RwCo *rwco = opaque;
2522 if (!rwco->is_write) {
2523 rwco->ret = bdrv_co_do_readv(rwco->bs, rwco->sector_num,
2524 rwco->nb_sectors, rwco->qiov,
2527 rwco->ret = bdrv_co_do_writev(rwco->bs, rwco->sector_num,
2528 rwco->nb_sectors, rwco->qiov,
2534 * Process a vectored synchronous request using coroutines
2536 static int bdrv_rwv_co(BlockDriverState *bs, int64_t sector_num,
2537 QEMUIOVector *qiov, bool is_write,
2538 BdrvRequestFlags flags)
2543 .sector_num = sector_num,
2544 .nb_sectors = qiov->size >> BDRV_SECTOR_BITS,
2546 .is_write = is_write,
2550 assert((qiov->size & (BDRV_SECTOR_SIZE - 1)) == 0);
2553 * In sync call context, when the vcpu is blocked, this throttling timer
2554 * will not fire; so the I/O throttling function has to be disabled here
2555 * if it has been enabled.
2557 if (bs->io_limits_enabled) {
2558 fprintf(stderr, "Disabling I/O throttling on '%s' due "
2559 "to synchronous I/O.\n", bdrv_get_device_name(bs));
2560 bdrv_io_limits_disable(bs);
2563 if (qemu_in_coroutine()) {
2564 /* Fast-path if already in coroutine context */
2565 bdrv_rw_co_entry(&rwco);
2567 co = qemu_coroutine_create(bdrv_rw_co_entry);
2568 qemu_coroutine_enter(co, &rwco);
2569 while (rwco.ret == NOT_DONE) {
2577 * Process a synchronous request using coroutines
2579 static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf,
2580 int nb_sectors, bool is_write, BdrvRequestFlags flags)
2583 struct iovec iov = {
2584 .iov_base = (void *)buf,
2585 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
2588 qemu_iovec_init_external(&qiov, &iov, 1);
2589 return bdrv_rwv_co(bs, sector_num, &qiov, is_write, flags);
2592 /* return < 0 if error. See bdrv_write() for the return codes */
2593 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
2594 uint8_t *buf, int nb_sectors)
2596 return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false, 0);
2599 /* Just like bdrv_read(), but with I/O throttling temporarily disabled */
2600 int bdrv_read_unthrottled(BlockDriverState *bs, int64_t sector_num,
2601 uint8_t *buf, int nb_sectors)
2606 enabled = bs->io_limits_enabled;
2607 bs->io_limits_enabled = false;
2608 ret = bdrv_read(bs, sector_num, buf, nb_sectors);
2609 bs->io_limits_enabled = enabled;
2613 /* Return < 0 if error. Important errors are:
2614 -EIO generic I/O error (may happen for all errors)
2615 -ENOMEDIUM No media inserted.
2616 -EINVAL Invalid sector number or nb_sectors
2617 -EACCES Trying to write a read-only device
2619 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
2620 const uint8_t *buf, int nb_sectors)
2622 return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true, 0);
2625 int bdrv_writev(BlockDriverState *bs, int64_t sector_num, QEMUIOVector *qiov)
2627 return bdrv_rwv_co(bs, sector_num, qiov, true, 0);
2630 int bdrv_write_zeroes(BlockDriverState *bs, int64_t sector_num,
2631 int nb_sectors, BdrvRequestFlags flags)
2633 return bdrv_rw_co(bs, sector_num, NULL, nb_sectors, true,
2634 BDRV_REQ_ZERO_WRITE | flags);
2638 * Completely zero out a block device with the help of bdrv_write_zeroes.
2639 * The operation is sped up by checking the block status and only writing
2640 * zeroes to the device if they currently do not return zeroes. Optional
2641 * flags are passed through to bdrv_write_zeroes (e.g. BDRV_REQ_MAY_UNMAP).
2643 * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
2645 int bdrv_make_zero(BlockDriverState *bs, BdrvRequestFlags flags)
2647 int64_t target_size = bdrv_getlength(bs) / BDRV_SECTOR_SIZE;
2648 int64_t ret, nb_sectors, sector_num = 0;
2652 nb_sectors = target_size - sector_num;
2653 if (nb_sectors <= 0) {
2656 if (nb_sectors > INT_MAX) {
2657 nb_sectors = INT_MAX;
2659 ret = bdrv_get_block_status(bs, sector_num, nb_sectors, &n);
2661 error_report("error getting block status at sector %" PRId64 ": %s",
2662 sector_num, strerror(-ret));
2665 if (ret & BDRV_BLOCK_ZERO) {
2669 ret = bdrv_write_zeroes(bs, sector_num, n, flags);
2671 error_report("error writing zeroes at sector %" PRId64 ": %s",
2672 sector_num, strerror(-ret));
2679 int bdrv_pread(BlockDriverState *bs, int64_t offset,
2680 void *buf, int count1)
2682 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
2683 int len, nb_sectors, count;
2688 /* first read to align to sector start */
2689 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
2692 sector_num = offset >> BDRV_SECTOR_BITS;
2694 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
2696 memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
2704 /* read the sectors "in place" */
2705 nb_sectors = count >> BDRV_SECTOR_BITS;
2706 if (nb_sectors > 0) {
2707 if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
2709 sector_num += nb_sectors;
2710 len = nb_sectors << BDRV_SECTOR_BITS;
2715 /* add data from the last sector */
2717 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
2719 memcpy(buf, tmp_buf, count);
2724 int bdrv_pwritev(BlockDriverState *bs, int64_t offset, QEMUIOVector *qiov)
2726 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
2727 int len, nb_sectors, count;
2733 /* first write to align to sector start */
2734 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
2737 sector_num = offset >> BDRV_SECTOR_BITS;
2739 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
2741 qemu_iovec_to_buf(qiov, 0, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)),
2743 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
2751 /* write the sectors "in place" */
2752 nb_sectors = count >> BDRV_SECTOR_BITS;
2753 if (nb_sectors > 0) {
2754 QEMUIOVector qiov_inplace;
2756 qemu_iovec_init(&qiov_inplace, qiov->niov);
2757 qemu_iovec_concat(&qiov_inplace, qiov, len,
2758 nb_sectors << BDRV_SECTOR_BITS);
2759 ret = bdrv_writev(bs, sector_num, &qiov_inplace);
2760 qemu_iovec_destroy(&qiov_inplace);
2765 sector_num += nb_sectors;
2766 len = nb_sectors << BDRV_SECTOR_BITS;
2770 /* add data from the last sector */
2772 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
2774 qemu_iovec_to_buf(qiov, qiov->size - count, tmp_buf, count);
2775 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
2781 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
2782 const void *buf, int count1)
2785 struct iovec iov = {
2786 .iov_base = (void *) buf,
2790 qemu_iovec_init_external(&qiov, &iov, 1);
2791 return bdrv_pwritev(bs, offset, &qiov);
2795 * Writes to the file and ensures that no writes are reordered across this
2796 * request (acts as a barrier)
2798 * Returns 0 on success, -errno in error cases.
2800 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
2801 const void *buf, int count)
2805 ret = bdrv_pwrite(bs, offset, buf, count);
2810 /* No flush needed for cache modes that already do it */
2811 if (bs->enable_write_cache) {
2818 static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs,
2819 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
2821 /* Perform I/O through a temporary buffer so that users who scribble over
2822 * their read buffer while the operation is in progress do not end up
2823 * modifying the image file. This is critical for zero-copy guest I/O
2824 * where anything might happen inside guest memory.
2826 void *bounce_buffer;
2828 BlockDriver *drv = bs->drv;
2830 QEMUIOVector bounce_qiov;
2831 int64_t cluster_sector_num;
2832 int cluster_nb_sectors;
2836 /* Cover entire cluster so no additional backing file I/O is required when
2837 * allocating cluster in the image file.
2839 bdrv_round_to_clusters(bs, sector_num, nb_sectors,
2840 &cluster_sector_num, &cluster_nb_sectors);
2842 trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors,
2843 cluster_sector_num, cluster_nb_sectors);
2845 iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE;
2846 iov.iov_base = bounce_buffer = qemu_blockalign(bs, iov.iov_len);
2847 qemu_iovec_init_external(&bounce_qiov, &iov, 1);
2849 ret = drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors,
2855 if (drv->bdrv_co_write_zeroes &&
2856 buffer_is_zero(bounce_buffer, iov.iov_len)) {
2857 ret = bdrv_co_do_write_zeroes(bs, cluster_sector_num,
2858 cluster_nb_sectors, 0);
2860 /* This does not change the data on the disk, it is not necessary
2861 * to flush even in cache=writethrough mode.
2863 ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors,
2868 /* It might be okay to ignore write errors for guest requests. If this
2869 * is a deliberate copy-on-read then we don't want to ignore the error.
2870 * Simply report it in all cases.
2875 skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE;
2876 qemu_iovec_from_buf(qiov, 0, bounce_buffer + skip_bytes,
2877 nb_sectors * BDRV_SECTOR_SIZE);
2880 qemu_vfree(bounce_buffer);
2885 * Handle a read request in coroutine context
2887 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
2888 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
2889 BdrvRequestFlags flags)
2891 BlockDriver *drv = bs->drv;
2892 BdrvTrackedRequest req;
2898 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
2902 if (bs->copy_on_read) {
2903 flags |= BDRV_REQ_COPY_ON_READ;
2905 if (flags & BDRV_REQ_COPY_ON_READ) {
2906 bs->copy_on_read_in_flight++;
2909 if (bs->copy_on_read_in_flight) {
2910 wait_for_overlapping_requests(bs, sector_num, nb_sectors);
2913 /* throttling disk I/O */
2914 if (bs->io_limits_enabled) {
2915 bdrv_io_limits_intercept(bs, nb_sectors, false);
2918 tracked_request_begin(&req, bs, sector_num, nb_sectors, false);
2920 if (flags & BDRV_REQ_COPY_ON_READ) {
2923 ret = bdrv_is_allocated(bs, sector_num, nb_sectors, &pnum);
2928 if (!ret || pnum != nb_sectors) {
2929 ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov);
2934 if (!(bs->zero_beyond_eof && bs->growable)) {
2935 ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
2937 /* Read zeros after EOF of growable BDSes */
2938 int64_t len, total_sectors, max_nb_sectors;
2940 len = bdrv_getlength(bs);
2946 total_sectors = DIV_ROUND_UP(len, BDRV_SECTOR_SIZE);
2947 max_nb_sectors = MAX(0, total_sectors - sector_num);
2948 if (max_nb_sectors > 0) {
2949 ret = drv->bdrv_co_readv(bs, sector_num,
2950 MIN(nb_sectors, max_nb_sectors), qiov);
2955 /* Reading beyond end of file is supposed to produce zeroes */
2956 if (ret == 0 && total_sectors < sector_num + nb_sectors) {
2957 uint64_t offset = MAX(0, total_sectors - sector_num);
2958 uint64_t bytes = (sector_num + nb_sectors - offset) *
2960 qemu_iovec_memset(qiov, offset * BDRV_SECTOR_SIZE, 0, bytes);
2965 tracked_request_end(&req);
2967 if (flags & BDRV_REQ_COPY_ON_READ) {
2968 bs->copy_on_read_in_flight--;
2974 int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
2975 int nb_sectors, QEMUIOVector *qiov)
2977 trace_bdrv_co_readv(bs, sector_num, nb_sectors);
2979 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0);
2982 int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
2983 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
2985 trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors);
2987 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
2988 BDRV_REQ_COPY_ON_READ);
2991 /* if no limit is specified in the BlockLimits use a default
2992 * of 32768 512-byte sectors (16 MiB) per request.
2994 #define MAX_WRITE_ZEROES_DEFAULT 32768
2996 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
2997 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
2999 BlockDriver *drv = bs->drv;
3001 struct iovec iov = {0};
3004 int max_write_zeroes = bs->bl.max_write_zeroes ?
3005 bs->bl.max_write_zeroes : MAX_WRITE_ZEROES_DEFAULT;
3007 while (nb_sectors > 0 && !ret) {
3008 int num = nb_sectors;
3010 /* Align request. Block drivers can expect the "bulk" of the request
3013 if (bs->bl.write_zeroes_alignment
3014 && num > bs->bl.write_zeroes_alignment) {
3015 if (sector_num % bs->bl.write_zeroes_alignment != 0) {
3016 /* Make a small request up to the first aligned sector. */
3017 num = bs->bl.write_zeroes_alignment;
3018 num -= sector_num % bs->bl.write_zeroes_alignment;
3019 } else if ((sector_num + num) % bs->bl.write_zeroes_alignment != 0) {
3020 /* Shorten the request to the last aligned sector. num cannot
3021 * underflow because num > bs->bl.write_zeroes_alignment.
3023 num -= (sector_num + num) % bs->bl.write_zeroes_alignment;
3027 /* limit request size */
3028 if (num > max_write_zeroes) {
3029 num = max_write_zeroes;
3033 /* First try the efficient write zeroes operation */
3034 if (drv->bdrv_co_write_zeroes) {
3035 ret = drv->bdrv_co_write_zeroes(bs, sector_num, num, flags);
3038 if (ret == -ENOTSUP) {
3039 /* Fall back to bounce buffer if write zeroes is unsupported */
3040 iov.iov_len = num * BDRV_SECTOR_SIZE;
3041 if (iov.iov_base == NULL) {
3042 iov.iov_base = qemu_blockalign(bs, num * BDRV_SECTOR_SIZE);
3043 memset(iov.iov_base, 0, num * BDRV_SECTOR_SIZE);
3045 qemu_iovec_init_external(&qiov, &iov, 1);
3047 ret = drv->bdrv_co_writev(bs, sector_num, num, &qiov);
3049 /* Keep bounce buffer around if it is big enough for all
3050 * all future requests.
3052 if (num < max_write_zeroes) {
3053 qemu_vfree(iov.iov_base);
3054 iov.iov_base = NULL;
3062 qemu_vfree(iov.iov_base);
3067 * Handle a write request in coroutine context
3069 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
3070 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
3071 BdrvRequestFlags flags)
3073 BlockDriver *drv = bs->drv;
3074 BdrvTrackedRequest req;
3080 if (bs->read_only) {
3083 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
3087 if (bs->copy_on_read_in_flight) {
3088 wait_for_overlapping_requests(bs, sector_num, nb_sectors);
3091 /* throttling disk I/O */
3092 if (bs->io_limits_enabled) {
3093 bdrv_io_limits_intercept(bs, nb_sectors, true);
3096 tracked_request_begin(&req, bs, sector_num, nb_sectors, true);
3098 ret = notifier_with_return_list_notify(&bs->before_write_notifiers, &req);
3101 /* Do nothing, write notifier decided to fail this request */
3102 } else if (flags & BDRV_REQ_ZERO_WRITE) {
3103 ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors, flags);
3105 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
3108 if (ret == 0 && !bs->enable_write_cache) {
3109 ret = bdrv_co_flush(bs);
3112 bdrv_set_dirty(bs, sector_num, nb_sectors);
3114 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
3115 bs->wr_highest_sector = sector_num + nb_sectors - 1;
3117 if (bs->growable && ret >= 0) {
3118 bs->total_sectors = MAX(bs->total_sectors, sector_num + nb_sectors);
3121 tracked_request_end(&req);
3126 int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
3127 int nb_sectors, QEMUIOVector *qiov)
3129 trace_bdrv_co_writev(bs, sector_num, nb_sectors);
3131 return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0);
3134 int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs,
3135 int64_t sector_num, int nb_sectors,
3136 BdrvRequestFlags flags)
3138 trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors, flags);
3140 if (!(bs->open_flags & BDRV_O_UNMAP)) {
3141 flags &= ~BDRV_REQ_MAY_UNMAP;
3144 return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL,
3145 BDRV_REQ_ZERO_WRITE | flags);
3149 * Truncate file to 'offset' bytes (needed only for file protocols)
3151 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
3153 BlockDriver *drv = bs->drv;
3157 if (!drv->bdrv_truncate)
3161 if (bdrv_in_use(bs))
3163 ret = drv->bdrv_truncate(bs, offset);
3165 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
3166 bdrv_dev_resize_cb(bs);
3172 * Length of a allocated file in bytes. Sparse files are counted by actual
3173 * allocated space. Return < 0 if error or unknown.
3175 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
3177 BlockDriver *drv = bs->drv;
3181 if (drv->bdrv_get_allocated_file_size) {
3182 return drv->bdrv_get_allocated_file_size(bs);
3185 return bdrv_get_allocated_file_size(bs->file);
3191 * Length of a file in bytes. Return < 0 if error or unknown.
3193 int64_t bdrv_getlength(BlockDriverState *bs)
3195 BlockDriver *drv = bs->drv;
3199 if (drv->has_variable_length) {
3200 int ret = refresh_total_sectors(bs, bs->total_sectors);
3205 return bs->total_sectors * BDRV_SECTOR_SIZE;
3208 /* return 0 as number of sectors if no device present or error */
3209 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
3212 length = bdrv_getlength(bs);
3216 length = length >> BDRV_SECTOR_BITS;
3217 *nb_sectors_ptr = length;
3220 void bdrv_set_on_error(BlockDriverState *bs, BlockdevOnError on_read_error,
3221 BlockdevOnError on_write_error)
3223 bs->on_read_error = on_read_error;
3224 bs->on_write_error = on_write_error;
3227 BlockdevOnError bdrv_get_on_error(BlockDriverState *bs, bool is_read)
3229 return is_read ? bs->on_read_error : bs->on_write_error;
3232 BlockErrorAction bdrv_get_error_action(BlockDriverState *bs, bool is_read, int error)
3234 BlockdevOnError on_err = is_read ? bs->on_read_error : bs->on_write_error;
3237 case BLOCKDEV_ON_ERROR_ENOSPC:
3238 return (error == ENOSPC) ? BDRV_ACTION_STOP : BDRV_ACTION_REPORT;
3239 case BLOCKDEV_ON_ERROR_STOP:
3240 return BDRV_ACTION_STOP;
3241 case BLOCKDEV_ON_ERROR_REPORT:
3242 return BDRV_ACTION_REPORT;
3243 case BLOCKDEV_ON_ERROR_IGNORE:
3244 return BDRV_ACTION_IGNORE;
3250 /* This is done by device models because, while the block layer knows
3251 * about the error, it does not know whether an operation comes from
3252 * the device or the block layer (from a job, for example).
3254 void bdrv_error_action(BlockDriverState *bs, BlockErrorAction action,
3255 bool is_read, int error)
3258 bdrv_emit_qmp_error_event(bs, QEVENT_BLOCK_IO_ERROR, action, is_read);
3259 if (action == BDRV_ACTION_STOP) {
3260 vm_stop(RUN_STATE_IO_ERROR);
3261 bdrv_iostatus_set_err(bs, error);
3265 int bdrv_is_read_only(BlockDriverState *bs)
3267 return bs->read_only;
3270 int bdrv_is_sg(BlockDriverState *bs)
3275 int bdrv_enable_write_cache(BlockDriverState *bs)
3277 return bs->enable_write_cache;
3280 void bdrv_set_enable_write_cache(BlockDriverState *bs, bool wce)
3282 bs->enable_write_cache = wce;
3284 /* so a reopen() will preserve wce */
3286 bs->open_flags |= BDRV_O_CACHE_WB;
3288 bs->open_flags &= ~BDRV_O_CACHE_WB;
3292 int bdrv_is_encrypted(BlockDriverState *bs)
3294 if (bs->backing_hd && bs->backing_hd->encrypted)
3296 return bs->encrypted;
3299 int bdrv_key_required(BlockDriverState *bs)
3301 BlockDriverState *backing_hd = bs->backing_hd;
3303 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
3305 return (bs->encrypted && !bs->valid_key);
3308 int bdrv_set_key(BlockDriverState *bs, const char *key)
3311 if (bs->backing_hd && bs->backing_hd->encrypted) {
3312 ret = bdrv_set_key(bs->backing_hd, key);
3318 if (!bs->encrypted) {
3320 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
3323 ret = bs->drv->bdrv_set_key(bs, key);
3326 } else if (!bs->valid_key) {
3328 /* call the change callback now, we skipped it on open */
3329 bdrv_dev_change_media_cb(bs, true);
3334 const char *bdrv_get_format_name(BlockDriverState *bs)
3336 return bs->drv ? bs->drv->format_name : NULL;
3339 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
3344 QLIST_FOREACH(drv, &bdrv_drivers, list) {
3345 it(opaque, drv->format_name);
3349 /* This function is to find block backend bs */
3350 BlockDriverState *bdrv_find(const char *name)
3352 BlockDriverState *bs;
3354 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
3355 if (!strcmp(name, bs->device_name)) {
3362 /* This function is to find a node in the bs graph */
3363 BlockDriverState *bdrv_find_node(const char *node_name)
3365 BlockDriverState *bs;
3369 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3370 if (!strcmp(node_name, bs->node_name)) {
3377 /* Put this QMP function here so it can access the static graph_bdrv_states. */
3378 BlockDeviceInfoList *bdrv_named_nodes_list(void)
3380 BlockDeviceInfoList *list, *entry;
3381 BlockDriverState *bs;
3384 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3385 entry = g_malloc0(sizeof(*entry));
3386 entry->value = bdrv_block_device_info(bs);
3394 BlockDriverState *bdrv_lookup_bs(const char *device,
3395 const char *node_name,
3398 BlockDriverState *bs = NULL;
3400 if ((!device && !node_name) || (device && node_name)) {
3401 error_setg(errp, "Use either device or node-name but not both");
3406 bs = bdrv_find(device);
3409 error_set(errp, QERR_DEVICE_NOT_FOUND, device);
3416 bs = bdrv_find_node(node_name);
3419 error_set(errp, QERR_DEVICE_NOT_FOUND, node_name);
3426 BlockDriverState *bdrv_next(BlockDriverState *bs)
3429 return QTAILQ_FIRST(&bdrv_states);
3431 return QTAILQ_NEXT(bs, device_list);
3434 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
3436 BlockDriverState *bs;
3438 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
3443 const char *bdrv_get_device_name(BlockDriverState *bs)
3445 return bs->device_name;
3448 int bdrv_get_flags(BlockDriverState *bs)
3450 return bs->open_flags;
3453 int bdrv_flush_all(void)
3455 BlockDriverState *bs;
3458 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
3459 int ret = bdrv_flush(bs);
3460 if (ret < 0 && !result) {
3468 int bdrv_has_zero_init_1(BlockDriverState *bs)
3473 int bdrv_has_zero_init(BlockDriverState *bs)
3477 /* If BS is a copy on write image, it is initialized to
3478 the contents of the base image, which may not be zeroes. */
3479 if (bs->backing_hd) {
3482 if (bs->drv->bdrv_has_zero_init) {
3483 return bs->drv->bdrv_has_zero_init(bs);
3490 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
3492 BlockDriverInfo bdi;
3494 if (bs->backing_hd) {
3498 if (bdrv_get_info(bs, &bdi) == 0) {
3499 return bdi.unallocated_blocks_are_zero;
3505 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
3507 BlockDriverInfo bdi;
3509 if (bs->backing_hd || !(bs->open_flags & BDRV_O_UNMAP)) {
3513 if (bdrv_get_info(bs, &bdi) == 0) {
3514 return bdi.can_write_zeroes_with_unmap;
3520 typedef struct BdrvCoGetBlockStatusData {
3521 BlockDriverState *bs;
3522 BlockDriverState *base;
3528 } BdrvCoGetBlockStatusData;
3531 * Returns true iff the specified sector is present in the disk image. Drivers
3532 * not implementing the functionality are assumed to not support backing files,
3533 * hence all their sectors are reported as allocated.
3535 * If 'sector_num' is beyond the end of the disk image the return value is 0
3536 * and 'pnum' is set to 0.
3538 * 'pnum' is set to the number of sectors (including and immediately following
3539 * the specified sector) that are known to be in the same
3540 * allocated/unallocated state.
3542 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
3543 * beyond the end of the disk image it will be clamped.
3545 static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs,
3547 int nb_sectors, int *pnum)
3553 length = bdrv_getlength(bs);
3558 if (sector_num >= (length >> BDRV_SECTOR_BITS)) {
3563 n = bs->total_sectors - sector_num;
3564 if (n < nb_sectors) {
3568 if (!bs->drv->bdrv_co_get_block_status) {
3570 ret = BDRV_BLOCK_DATA;
3571 if (bs->drv->protocol_name) {
3572 ret |= BDRV_BLOCK_OFFSET_VALID | (sector_num * BDRV_SECTOR_SIZE);
3577 ret = bs->drv->bdrv_co_get_block_status(bs, sector_num, nb_sectors, pnum);
3583 if (ret & BDRV_BLOCK_RAW) {
3584 assert(ret & BDRV_BLOCK_OFFSET_VALID);
3585 return bdrv_get_block_status(bs->file, ret >> BDRV_SECTOR_BITS,
3589 if (!(ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO)) {
3590 if (bdrv_unallocated_blocks_are_zero(bs)) {
3591 ret |= BDRV_BLOCK_ZERO;
3592 } else if (bs->backing_hd) {
3593 BlockDriverState *bs2 = bs->backing_hd;
3594 int64_t length2 = bdrv_getlength(bs2);
3595 if (length2 >= 0 && sector_num >= (length2 >> BDRV_SECTOR_BITS)) {
3596 ret |= BDRV_BLOCK_ZERO;
3602 (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) &&
3603 (ret & BDRV_BLOCK_OFFSET_VALID)) {
3604 ret2 = bdrv_co_get_block_status(bs->file, ret >> BDRV_SECTOR_BITS,
3607 /* Ignore errors. This is just providing extra information, it
3608 * is useful but not necessary.
3610 ret |= (ret2 & BDRV_BLOCK_ZERO);
3617 /* Coroutine wrapper for bdrv_get_block_status() */
3618 static void coroutine_fn bdrv_get_block_status_co_entry(void *opaque)
3620 BdrvCoGetBlockStatusData *data = opaque;
3621 BlockDriverState *bs = data->bs;
3623 data->ret = bdrv_co_get_block_status(bs, data->sector_num, data->nb_sectors,
3629 * Synchronous wrapper around bdrv_co_get_block_status().
3631 * See bdrv_co_get_block_status() for details.
3633 int64_t bdrv_get_block_status(BlockDriverState *bs, int64_t sector_num,
3634 int nb_sectors, int *pnum)
3637 BdrvCoGetBlockStatusData data = {
3639 .sector_num = sector_num,
3640 .nb_sectors = nb_sectors,
3645 if (qemu_in_coroutine()) {
3646 /* Fast-path if already in coroutine context */
3647 bdrv_get_block_status_co_entry(&data);
3649 co = qemu_coroutine_create(bdrv_get_block_status_co_entry);
3650 qemu_coroutine_enter(co, &data);
3651 while (!data.done) {
3658 int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num,
3659 int nb_sectors, int *pnum)
3661 int64_t ret = bdrv_get_block_status(bs, sector_num, nb_sectors, pnum);
3666 (ret & BDRV_BLOCK_DATA) ||
3667 ((ret & BDRV_BLOCK_ZERO) && !bdrv_has_zero_init(bs));
3671 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
3673 * Return true if the given sector is allocated in any image between
3674 * BASE and TOP (inclusive). BASE can be NULL to check if the given
3675 * sector is allocated in any image of the chain. Return false otherwise.
3677 * 'pnum' is set to the number of sectors (including and immediately following
3678 * the specified sector) that are known to be in the same
3679 * allocated/unallocated state.
3682 int bdrv_is_allocated_above(BlockDriverState *top,
3683 BlockDriverState *base,
3685 int nb_sectors, int *pnum)
3687 BlockDriverState *intermediate;
3688 int ret, n = nb_sectors;
3691 while (intermediate && intermediate != base) {
3693 ret = bdrv_is_allocated(intermediate, sector_num, nb_sectors,
3703 * [sector_num, nb_sectors] is unallocated on top but intermediate
3706 * [sector_num+x, nr_sectors] allocated.
3708 if (n > pnum_inter &&
3709 (intermediate == top ||
3710 sector_num + pnum_inter < intermediate->total_sectors)) {
3714 intermediate = intermediate->backing_hd;
3721 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
3723 if (bs->backing_hd && bs->backing_hd->encrypted)
3724 return bs->backing_file;
3725 else if (bs->encrypted)
3726 return bs->filename;
3731 void bdrv_get_backing_filename(BlockDriverState *bs,
3732 char *filename, int filename_size)
3734 pstrcpy(filename, filename_size, bs->backing_file);
3737 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
3738 const uint8_t *buf, int nb_sectors)
3740 BlockDriver *drv = bs->drv;
3743 if (!drv->bdrv_write_compressed)
3745 if (bdrv_check_request(bs, sector_num, nb_sectors))
3748 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
3750 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
3753 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
3755 BlockDriver *drv = bs->drv;
3758 if (!drv->bdrv_get_info)
3760 memset(bdi, 0, sizeof(*bdi));
3761 return drv->bdrv_get_info(bs, bdi);
3764 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs)
3766 BlockDriver *drv = bs->drv;
3767 if (drv && drv->bdrv_get_specific_info) {
3768 return drv->bdrv_get_specific_info(bs);
3773 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
3774 int64_t pos, int size)
3777 struct iovec iov = {
3778 .iov_base = (void *) buf,
3782 qemu_iovec_init_external(&qiov, &iov, 1);
3783 return bdrv_writev_vmstate(bs, &qiov, pos);
3786 int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
3788 BlockDriver *drv = bs->drv;
3792 } else if (drv->bdrv_save_vmstate) {
3793 return drv->bdrv_save_vmstate(bs, qiov, pos);
3794 } else if (bs->file) {
3795 return bdrv_writev_vmstate(bs->file, qiov, pos);
3801 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
3802 int64_t pos, int size)
3804 BlockDriver *drv = bs->drv;
3807 if (drv->bdrv_load_vmstate)
3808 return drv->bdrv_load_vmstate(bs, buf, pos, size);
3810 return bdrv_load_vmstate(bs->file, buf, pos, size);
3814 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
3816 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
3820 bs->drv->bdrv_debug_event(bs, event);
3823 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
3826 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
3830 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
3831 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
3837 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
3839 while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
3843 if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
3844 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
3850 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
3852 while (bs && bs->drv && !bs->drv->bdrv_debug_resume) {
3856 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
3857 return bs->drv->bdrv_debug_resume(bs, tag);
3863 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
3865 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
3869 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
3870 return bs->drv->bdrv_debug_is_suspended(bs, tag);
3876 int bdrv_is_snapshot(BlockDriverState *bs)
3878 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
3881 /* backing_file can either be relative, or absolute, or a protocol. If it is
3882 * relative, it must be relative to the chain. So, passing in bs->filename
3883 * from a BDS as backing_file should not be done, as that may be relative to
3884 * the CWD rather than the chain. */
3885 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
3886 const char *backing_file)
3888 char *filename_full = NULL;
3889 char *backing_file_full = NULL;
3890 char *filename_tmp = NULL;
3891 int is_protocol = 0;
3892 BlockDriverState *curr_bs = NULL;
3893 BlockDriverState *retval = NULL;
3895 if (!bs || !bs->drv || !backing_file) {
3899 filename_full = g_malloc(PATH_MAX);
3900 backing_file_full = g_malloc(PATH_MAX);
3901 filename_tmp = g_malloc(PATH_MAX);
3903 is_protocol = path_has_protocol(backing_file);
3905 for (curr_bs = bs; curr_bs->backing_hd; curr_bs = curr_bs->backing_hd) {
3907 /* If either of the filename paths is actually a protocol, then
3908 * compare unmodified paths; otherwise make paths relative */
3909 if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
3910 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
3911 retval = curr_bs->backing_hd;
3915 /* If not an absolute filename path, make it relative to the current
3916 * image's filename path */
3917 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
3920 /* We are going to compare absolute pathnames */
3921 if (!realpath(filename_tmp, filename_full)) {
3925 /* We need to make sure the backing filename we are comparing against
3926 * is relative to the current image filename (or absolute) */
3927 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
3928 curr_bs->backing_file);
3930 if (!realpath(filename_tmp, backing_file_full)) {
3934 if (strcmp(backing_file_full, filename_full) == 0) {
3935 retval = curr_bs->backing_hd;
3941 g_free(filename_full);
3942 g_free(backing_file_full);
3943 g_free(filename_tmp);
3947 int bdrv_get_backing_file_depth(BlockDriverState *bs)
3953 if (!bs->backing_hd) {
3957 return 1 + bdrv_get_backing_file_depth(bs->backing_hd);
3960 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
3962 BlockDriverState *curr_bs = NULL;
3970 while (curr_bs->backing_hd) {
3971 curr_bs = curr_bs->backing_hd;
3976 /**************************************************************/
3979 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
3980 QEMUIOVector *qiov, int nb_sectors,
3981 BlockDriverCompletionFunc *cb, void *opaque)
3983 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
3985 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0,
3989 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
3990 QEMUIOVector *qiov, int nb_sectors,
3991 BlockDriverCompletionFunc *cb, void *opaque)
3993 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
3995 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0,
3999 BlockDriverAIOCB *bdrv_aio_write_zeroes(BlockDriverState *bs,
4000 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags,
4001 BlockDriverCompletionFunc *cb, void *opaque)
4003 trace_bdrv_aio_write_zeroes(bs, sector_num, nb_sectors, flags, opaque);
4005 return bdrv_co_aio_rw_vector(bs, sector_num, NULL, nb_sectors,
4006 BDRV_REQ_ZERO_WRITE | flags,
4011 typedef struct MultiwriteCB {
4016 BlockDriverCompletionFunc *cb;
4018 QEMUIOVector *free_qiov;
4022 static void multiwrite_user_cb(MultiwriteCB *mcb)
4026 for (i = 0; i < mcb->num_callbacks; i++) {
4027 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
4028 if (mcb->callbacks[i].free_qiov) {
4029 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
4031 g_free(mcb->callbacks[i].free_qiov);
4035 static void multiwrite_cb(void *opaque, int ret)
4037 MultiwriteCB *mcb = opaque;
4039 trace_multiwrite_cb(mcb, ret);
4041 if (ret < 0 && !mcb->error) {
4045 mcb->num_requests--;
4046 if (mcb->num_requests == 0) {
4047 multiwrite_user_cb(mcb);
4052 static int multiwrite_req_compare(const void *a, const void *b)
4054 const BlockRequest *req1 = a, *req2 = b;
4057 * Note that we can't simply subtract req2->sector from req1->sector
4058 * here as that could overflow the return value.
4060 if (req1->sector > req2->sector) {
4062 } else if (req1->sector < req2->sector) {
4070 * Takes a bunch of requests and tries to merge them. Returns the number of
4071 * requests that remain after merging.
4073 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
4074 int num_reqs, MultiwriteCB *mcb)
4078 // Sort requests by start sector
4079 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
4081 // Check if adjacent requests touch the same clusters. If so, combine them,
4082 // filling up gaps with zero sectors.
4084 for (i = 1; i < num_reqs; i++) {
4086 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
4088 // Handle exactly sequential writes and overlapping writes.
4089 if (reqs[i].sector <= oldreq_last) {
4093 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
4099 QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
4100 qemu_iovec_init(qiov,
4101 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
4103 // Add the first request to the merged one. If the requests are
4104 // overlapping, drop the last sectors of the first request.
4105 size = (reqs[i].sector - reqs[outidx].sector) << 9;
4106 qemu_iovec_concat(qiov, reqs[outidx].qiov, 0, size);
4108 // We should need to add any zeros between the two requests
4109 assert (reqs[i].sector <= oldreq_last);
4111 // Add the second request
4112 qemu_iovec_concat(qiov, reqs[i].qiov, 0, reqs[i].qiov->size);
4114 reqs[outidx].nb_sectors = qiov->size >> 9;
4115 reqs[outidx].qiov = qiov;
4117 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
4120 reqs[outidx].sector = reqs[i].sector;
4121 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
4122 reqs[outidx].qiov = reqs[i].qiov;
4130 * Submit multiple AIO write requests at once.
4132 * On success, the function returns 0 and all requests in the reqs array have
4133 * been submitted. In error case this function returns -1, and any of the
4134 * requests may or may not be submitted yet. In particular, this means that the
4135 * callback will be called for some of the requests, for others it won't. The
4136 * caller must check the error field of the BlockRequest to wait for the right
4137 * callbacks (if error != 0, no callback will be called).
4139 * The implementation may modify the contents of the reqs array, e.g. to merge
4140 * requests. However, the fields opaque and error are left unmodified as they
4141 * are used to signal failure for a single request to the caller.
4143 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
4148 /* don't submit writes if we don't have a medium */
4149 if (bs->drv == NULL) {
4150 for (i = 0; i < num_reqs; i++) {
4151 reqs[i].error = -ENOMEDIUM;
4156 if (num_reqs == 0) {
4160 // Create MultiwriteCB structure
4161 mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
4162 mcb->num_requests = 0;
4163 mcb->num_callbacks = num_reqs;
4165 for (i = 0; i < num_reqs; i++) {
4166 mcb->callbacks[i].cb = reqs[i].cb;
4167 mcb->callbacks[i].opaque = reqs[i].opaque;
4170 // Check for mergable requests
4171 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
4173 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
4175 /* Run the aio requests. */
4176 mcb->num_requests = num_reqs;
4177 for (i = 0; i < num_reqs; i++) {
4178 bdrv_co_aio_rw_vector(bs, reqs[i].sector, reqs[i].qiov,
4179 reqs[i].nb_sectors, reqs[i].flags,
4187 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
4189 acb->aiocb_info->cancel(acb);
4192 /**************************************************************/
4193 /* async block device emulation */
4195 typedef struct BlockDriverAIOCBSync {
4196 BlockDriverAIOCB common;
4199 /* vector translation state */
4203 } BlockDriverAIOCBSync;
4205 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
4207 BlockDriverAIOCBSync *acb =
4208 container_of(blockacb, BlockDriverAIOCBSync, common);
4209 qemu_bh_delete(acb->bh);
4211 qemu_aio_release(acb);
4214 static const AIOCBInfo bdrv_em_aiocb_info = {
4215 .aiocb_size = sizeof(BlockDriverAIOCBSync),
4216 .cancel = bdrv_aio_cancel_em,
4219 static void bdrv_aio_bh_cb(void *opaque)
4221 BlockDriverAIOCBSync *acb = opaque;
4224 qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
4225 qemu_vfree(acb->bounce);
4226 acb->common.cb(acb->common.opaque, acb->ret);
4227 qemu_bh_delete(acb->bh);
4229 qemu_aio_release(acb);
4232 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
4236 BlockDriverCompletionFunc *cb,
4241 BlockDriverAIOCBSync *acb;
4243 acb = qemu_aio_get(&bdrv_em_aiocb_info, bs, cb, opaque);
4244 acb->is_write = is_write;
4246 acb->bounce = qemu_blockalign(bs, qiov->size);
4247 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
4250 qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
4251 acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
4253 acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
4256 qemu_bh_schedule(acb->bh);
4258 return &acb->common;
4261 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
4262 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
4263 BlockDriverCompletionFunc *cb, void *opaque)
4265 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
4268 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
4269 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
4270 BlockDriverCompletionFunc *cb, void *opaque)
4272 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
4276 typedef struct BlockDriverAIOCBCoroutine {
4277 BlockDriverAIOCB common;
4282 } BlockDriverAIOCBCoroutine;
4284 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb)
4286 BlockDriverAIOCBCoroutine *acb =
4287 container_of(blockacb, BlockDriverAIOCBCoroutine, common);
4296 static const AIOCBInfo bdrv_em_co_aiocb_info = {
4297 .aiocb_size = sizeof(BlockDriverAIOCBCoroutine),
4298 .cancel = bdrv_aio_co_cancel_em,
4301 static void bdrv_co_em_bh(void *opaque)
4303 BlockDriverAIOCBCoroutine *acb = opaque;
4305 acb->common.cb(acb->common.opaque, acb->req.error);
4311 qemu_bh_delete(acb->bh);
4312 qemu_aio_release(acb);
4315 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
4316 static void coroutine_fn bdrv_co_do_rw(void *opaque)
4318 BlockDriverAIOCBCoroutine *acb = opaque;
4319 BlockDriverState *bs = acb->common.bs;
4321 if (!acb->is_write) {
4322 acb->req.error = bdrv_co_do_readv(bs, acb->req.sector,
4323 acb->req.nb_sectors, acb->req.qiov, acb->req.flags);
4325 acb->req.error = bdrv_co_do_writev(bs, acb->req.sector,
4326 acb->req.nb_sectors, acb->req.qiov, acb->req.flags);
4329 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
4330 qemu_bh_schedule(acb->bh);
4333 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
4337 BdrvRequestFlags flags,
4338 BlockDriverCompletionFunc *cb,
4343 BlockDriverAIOCBCoroutine *acb;
4345 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
4346 acb->req.sector = sector_num;
4347 acb->req.nb_sectors = nb_sectors;
4348 acb->req.qiov = qiov;
4349 acb->req.flags = flags;
4350 acb->is_write = is_write;
4353 co = qemu_coroutine_create(bdrv_co_do_rw);
4354 qemu_coroutine_enter(co, acb);
4356 return &acb->common;
4359 static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
4361 BlockDriverAIOCBCoroutine *acb = opaque;
4362 BlockDriverState *bs = acb->common.bs;
4364 acb->req.error = bdrv_co_flush(bs);
4365 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
4366 qemu_bh_schedule(acb->bh);
4369 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
4370 BlockDriverCompletionFunc *cb, void *opaque)
4372 trace_bdrv_aio_flush(bs, opaque);
4375 BlockDriverAIOCBCoroutine *acb;
4377 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
4380 co = qemu_coroutine_create(bdrv_aio_flush_co_entry);
4381 qemu_coroutine_enter(co, acb);
4383 return &acb->common;
4386 static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque)
4388 BlockDriverAIOCBCoroutine *acb = opaque;
4389 BlockDriverState *bs = acb->common.bs;
4391 acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors);
4392 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
4393 qemu_bh_schedule(acb->bh);
4396 BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs,
4397 int64_t sector_num, int nb_sectors,
4398 BlockDriverCompletionFunc *cb, void *opaque)
4401 BlockDriverAIOCBCoroutine *acb;
4403 trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque);
4405 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
4406 acb->req.sector = sector_num;
4407 acb->req.nb_sectors = nb_sectors;
4409 co = qemu_coroutine_create(bdrv_aio_discard_co_entry);
4410 qemu_coroutine_enter(co, acb);
4412 return &acb->common;
4415 void bdrv_init(void)
4417 module_call_init(MODULE_INIT_BLOCK);
4420 void bdrv_init_with_whitelist(void)
4422 use_bdrv_whitelist = 1;
4426 void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs,
4427 BlockDriverCompletionFunc *cb, void *opaque)
4429 BlockDriverAIOCB *acb;
4431 acb = g_slice_alloc(aiocb_info->aiocb_size);
4432 acb->aiocb_info = aiocb_info;
4435 acb->opaque = opaque;
4439 void qemu_aio_release(void *p)
4441 BlockDriverAIOCB *acb = p;
4442 g_slice_free1(acb->aiocb_info->aiocb_size, acb);
4445 /**************************************************************/
4446 /* Coroutine block device emulation */
4448 typedef struct CoroutineIOCompletion {
4449 Coroutine *coroutine;
4451 } CoroutineIOCompletion;
4453 static void bdrv_co_io_em_complete(void *opaque, int ret)
4455 CoroutineIOCompletion *co = opaque;
4458 qemu_coroutine_enter(co->coroutine, NULL);
4461 static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num,
4462 int nb_sectors, QEMUIOVector *iov,
4465 CoroutineIOCompletion co = {
4466 .coroutine = qemu_coroutine_self(),
4468 BlockDriverAIOCB *acb;
4471 acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors,
4472 bdrv_co_io_em_complete, &co);
4474 acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors,
4475 bdrv_co_io_em_complete, &co);
4478 trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb);
4482 qemu_coroutine_yield();
4487 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
4488 int64_t sector_num, int nb_sectors,
4491 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false);
4494 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
4495 int64_t sector_num, int nb_sectors,
4498 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true);
4501 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
4503 RwCo *rwco = opaque;
4505 rwco->ret = bdrv_co_flush(rwco->bs);
4508 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
4512 if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
4516 /* Write back cached data to the OS even with cache=unsafe */
4517 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS);
4518 if (bs->drv->bdrv_co_flush_to_os) {
4519 ret = bs->drv->bdrv_co_flush_to_os(bs);
4525 /* But don't actually force it to the disk with cache=unsafe */
4526 if (bs->open_flags & BDRV_O_NO_FLUSH) {
4530 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK);
4531 if (bs->drv->bdrv_co_flush_to_disk) {
4532 ret = bs->drv->bdrv_co_flush_to_disk(bs);
4533 } else if (bs->drv->bdrv_aio_flush) {
4534 BlockDriverAIOCB *acb;
4535 CoroutineIOCompletion co = {
4536 .coroutine = qemu_coroutine_self(),
4539 acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
4543 qemu_coroutine_yield();
4548 * Some block drivers always operate in either writethrough or unsafe
4549 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
4550 * know how the server works (because the behaviour is hardcoded or
4551 * depends on server-side configuration), so we can't ensure that
4552 * everything is safe on disk. Returning an error doesn't work because
4553 * that would break guests even if the server operates in writethrough
4556 * Let's hope the user knows what he's doing.
4564 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
4565 * in the case of cache=unsafe, so there are no useless flushes.
4568 return bdrv_co_flush(bs->file);
4571 void bdrv_invalidate_cache(BlockDriverState *bs)
4573 if (bs->drv && bs->drv->bdrv_invalidate_cache) {
4574 bs->drv->bdrv_invalidate_cache(bs);
4578 void bdrv_invalidate_cache_all(void)
4580 BlockDriverState *bs;
4582 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
4583 bdrv_invalidate_cache(bs);
4587 void bdrv_clear_incoming_migration_all(void)
4589 BlockDriverState *bs;
4591 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
4592 bs->open_flags = bs->open_flags & ~(BDRV_O_INCOMING);
4596 int bdrv_flush(BlockDriverState *bs)
4604 if (qemu_in_coroutine()) {
4605 /* Fast-path if already in coroutine context */
4606 bdrv_flush_co_entry(&rwco);
4608 co = qemu_coroutine_create(bdrv_flush_co_entry);
4609 qemu_coroutine_enter(co, &rwco);
4610 while (rwco.ret == NOT_DONE) {
4618 static void coroutine_fn bdrv_discard_co_entry(void *opaque)
4620 RwCo *rwco = opaque;
4622 rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors);
4625 /* if no limit is specified in the BlockLimits use a default
4626 * of 32768 512-byte sectors (16 MiB) per request.
4628 #define MAX_DISCARD_DEFAULT 32768
4630 int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
4637 } else if (bdrv_check_request(bs, sector_num, nb_sectors)) {
4639 } else if (bs->read_only) {
4643 bdrv_reset_dirty(bs, sector_num, nb_sectors);
4645 /* Do nothing if disabled. */
4646 if (!(bs->open_flags & BDRV_O_UNMAP)) {
4650 if (!bs->drv->bdrv_co_discard && !bs->drv->bdrv_aio_discard) {
4654 max_discard = bs->bl.max_discard ? bs->bl.max_discard : MAX_DISCARD_DEFAULT;
4655 while (nb_sectors > 0) {
4657 int num = nb_sectors;
4660 if (bs->bl.discard_alignment &&
4661 num >= bs->bl.discard_alignment &&
4662 sector_num % bs->bl.discard_alignment) {
4663 if (num > bs->bl.discard_alignment) {
4664 num = bs->bl.discard_alignment;
4666 num -= sector_num % bs->bl.discard_alignment;
4669 /* limit request size */
4670 if (num > max_discard) {
4674 if (bs->drv->bdrv_co_discard) {
4675 ret = bs->drv->bdrv_co_discard(bs, sector_num, num);
4677 BlockDriverAIOCB *acb;
4678 CoroutineIOCompletion co = {
4679 .coroutine = qemu_coroutine_self(),
4682 acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors,
4683 bdrv_co_io_em_complete, &co);
4687 qemu_coroutine_yield();
4691 if (ret && ret != -ENOTSUP) {
4701 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
4706 .sector_num = sector_num,
4707 .nb_sectors = nb_sectors,
4711 if (qemu_in_coroutine()) {
4712 /* Fast-path if already in coroutine context */
4713 bdrv_discard_co_entry(&rwco);
4715 co = qemu_coroutine_create(bdrv_discard_co_entry);
4716 qemu_coroutine_enter(co, &rwco);
4717 while (rwco.ret == NOT_DONE) {
4725 /**************************************************************/
4726 /* removable device support */
4729 * Return TRUE if the media is present
4731 int bdrv_is_inserted(BlockDriverState *bs)
4733 BlockDriver *drv = bs->drv;
4737 if (!drv->bdrv_is_inserted)
4739 return drv->bdrv_is_inserted(bs);
4743 * Return whether the media changed since the last call to this
4744 * function, or -ENOTSUP if we don't know. Most drivers don't know.
4746 int bdrv_media_changed(BlockDriverState *bs)
4748 BlockDriver *drv = bs->drv;
4750 if (drv && drv->bdrv_media_changed) {
4751 return drv->bdrv_media_changed(bs);
4757 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4759 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
4761 BlockDriver *drv = bs->drv;
4763 if (drv && drv->bdrv_eject) {
4764 drv->bdrv_eject(bs, eject_flag);
4767 if (bs->device_name[0] != '\0') {
4768 bdrv_emit_qmp_eject_event(bs, eject_flag);
4773 * Lock or unlock the media (if it is locked, the user won't be able
4774 * to eject it manually).
4776 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
4778 BlockDriver *drv = bs->drv;
4780 trace_bdrv_lock_medium(bs, locked);
4782 if (drv && drv->bdrv_lock_medium) {
4783 drv->bdrv_lock_medium(bs, locked);
4787 /* needed for generic scsi interface */
4789 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
4791 BlockDriver *drv = bs->drv;
4793 if (drv && drv->bdrv_ioctl)
4794 return drv->bdrv_ioctl(bs, req, buf);
4798 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
4799 unsigned long int req, void *buf,
4800 BlockDriverCompletionFunc *cb, void *opaque)
4802 BlockDriver *drv = bs->drv;
4804 if (drv && drv->bdrv_aio_ioctl)
4805 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
4809 void bdrv_set_guest_block_size(BlockDriverState *bs, int align)
4811 bs->guest_block_size = align;
4814 void *qemu_blockalign(BlockDriverState *bs, size_t size)
4816 return qemu_memalign(bdrv_opt_mem_align(bs), size);
4820 * Check if all memory in this vector is sector aligned.
4822 bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov)
4825 size_t alignment = bdrv_opt_mem_align(bs);
4827 for (i = 0; i < qiov->niov; i++) {
4828 if ((uintptr_t) qiov->iov[i].iov_base % alignment) {
4831 if (qiov->iov[i].iov_len % alignment) {
4839 BdrvDirtyBitmap *bdrv_create_dirty_bitmap(BlockDriverState *bs, int granularity)
4841 int64_t bitmap_size;
4842 BdrvDirtyBitmap *bitmap;
4844 assert((granularity & (granularity - 1)) == 0);
4846 granularity >>= BDRV_SECTOR_BITS;
4847 assert(granularity);
4848 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS);
4849 bitmap = g_malloc0(sizeof(BdrvDirtyBitmap));
4850 bitmap->bitmap = hbitmap_alloc(bitmap_size, ffs(granularity) - 1);
4851 QLIST_INSERT_HEAD(&bs->dirty_bitmaps, bitmap, list);
4855 void bdrv_release_dirty_bitmap(BlockDriverState *bs, BdrvDirtyBitmap *bitmap)
4857 BdrvDirtyBitmap *bm, *next;
4858 QLIST_FOREACH_SAFE(bm, &bs->dirty_bitmaps, list, next) {
4860 QLIST_REMOVE(bitmap, list);
4861 hbitmap_free(bitmap->bitmap);
4868 BlockDirtyInfoList *bdrv_query_dirty_bitmaps(BlockDriverState *bs)
4870 BdrvDirtyBitmap *bm;
4871 BlockDirtyInfoList *list = NULL;
4872 BlockDirtyInfoList **plist = &list;
4874 QLIST_FOREACH(bm, &bs->dirty_bitmaps, list) {
4875 BlockDirtyInfo *info = g_malloc0(sizeof(BlockDirtyInfo));
4876 BlockDirtyInfoList *entry = g_malloc0(sizeof(BlockDirtyInfoList));
4877 info->count = bdrv_get_dirty_count(bs, bm);
4879 ((int64_t) BDRV_SECTOR_SIZE << hbitmap_granularity(bm->bitmap));
4880 entry->value = info;
4882 plist = &entry->next;
4888 int bdrv_get_dirty(BlockDriverState *bs, BdrvDirtyBitmap *bitmap, int64_t sector)
4891 return hbitmap_get(bitmap->bitmap, sector);
4897 void bdrv_dirty_iter_init(BlockDriverState *bs,
4898 BdrvDirtyBitmap *bitmap, HBitmapIter *hbi)
4900 hbitmap_iter_init(hbi, bitmap->bitmap, 0);
4903 void bdrv_set_dirty(BlockDriverState *bs, int64_t cur_sector,
4906 BdrvDirtyBitmap *bitmap;
4907 QLIST_FOREACH(bitmap, &bs->dirty_bitmaps, list) {
4908 hbitmap_set(bitmap->bitmap, cur_sector, nr_sectors);
4912 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector, int nr_sectors)
4914 BdrvDirtyBitmap *bitmap;
4915 QLIST_FOREACH(bitmap, &bs->dirty_bitmaps, list) {
4916 hbitmap_reset(bitmap->bitmap, cur_sector, nr_sectors);
4920 int64_t bdrv_get_dirty_count(BlockDriverState *bs, BdrvDirtyBitmap *bitmap)
4922 return hbitmap_count(bitmap->bitmap);
4925 /* Get a reference to bs */
4926 void bdrv_ref(BlockDriverState *bs)
4931 /* Release a previously grabbed reference to bs.
4932 * If after releasing, reference count is zero, the BlockDriverState is
4934 void bdrv_unref(BlockDriverState *bs)
4936 assert(bs->refcnt > 0);
4937 if (--bs->refcnt == 0) {
4942 void bdrv_set_in_use(BlockDriverState *bs, int in_use)
4944 assert(bs->in_use != in_use);
4945 bs->in_use = in_use;
4948 int bdrv_in_use(BlockDriverState *bs)
4953 void bdrv_iostatus_enable(BlockDriverState *bs)
4955 bs->iostatus_enabled = true;
4956 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
4959 /* The I/O status is only enabled if the drive explicitly
4960 * enables it _and_ the VM is configured to stop on errors */
4961 bool bdrv_iostatus_is_enabled(const BlockDriverState *bs)
4963 return (bs->iostatus_enabled &&
4964 (bs->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
4965 bs->on_write_error == BLOCKDEV_ON_ERROR_STOP ||
4966 bs->on_read_error == BLOCKDEV_ON_ERROR_STOP));
4969 void bdrv_iostatus_disable(BlockDriverState *bs)
4971 bs->iostatus_enabled = false;
4974 void bdrv_iostatus_reset(BlockDriverState *bs)
4976 if (bdrv_iostatus_is_enabled(bs)) {
4977 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
4979 block_job_iostatus_reset(bs->job);
4984 void bdrv_iostatus_set_err(BlockDriverState *bs, int error)
4986 assert(bdrv_iostatus_is_enabled(bs));
4987 if (bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
4988 bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
4989 BLOCK_DEVICE_IO_STATUS_FAILED;
4994 bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes,
4995 enum BlockAcctType type)
4997 assert(type < BDRV_MAX_IOTYPE);
4999 cookie->bytes = bytes;
5000 cookie->start_time_ns = get_clock();
5001 cookie->type = type;
5005 bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie)
5007 assert(cookie->type < BDRV_MAX_IOTYPE);
5009 bs->nr_bytes[cookie->type] += cookie->bytes;
5010 bs->nr_ops[cookie->type]++;
5011 bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns;
5014 void bdrv_img_create(const char *filename, const char *fmt,
5015 const char *base_filename, const char *base_fmt,
5016 char *options, uint64_t img_size, int flags,
5017 Error **errp, bool quiet)
5019 QEMUOptionParameter *param = NULL, *create_options = NULL;
5020 QEMUOptionParameter *backing_fmt, *backing_file, *size;
5021 BlockDriver *drv, *proto_drv;
5022 BlockDriver *backing_drv = NULL;
5023 Error *local_err = NULL;
5026 /* Find driver and parse its options */
5027 drv = bdrv_find_format(fmt);
5029 error_setg(errp, "Unknown file format '%s'", fmt);
5033 proto_drv = bdrv_find_protocol(filename, true);
5035 error_setg(errp, "Unknown protocol '%s'", filename);
5039 create_options = append_option_parameters(create_options,
5040 drv->create_options);
5041 create_options = append_option_parameters(create_options,
5042 proto_drv->create_options);
5044 /* Create parameter list with default values */
5045 param = parse_option_parameters("", create_options, param);
5047 set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size);
5049 /* Parse -o options */
5051 param = parse_option_parameters(options, create_options, param);
5052 if (param == NULL) {
5053 error_setg(errp, "Invalid options for file format '%s'.", fmt);
5058 if (base_filename) {
5059 if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE,
5061 error_setg(errp, "Backing file not supported for file format '%s'",
5068 if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) {
5069 error_setg(errp, "Backing file format not supported for file "
5070 "format '%s'", fmt);
5075 backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE);
5076 if (backing_file && backing_file->value.s) {
5077 if (!strcmp(filename, backing_file->value.s)) {
5078 error_setg(errp, "Error: Trying to create an image with the "
5079 "same filename as the backing file");
5084 backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT);
5085 if (backing_fmt && backing_fmt->value.s) {
5086 backing_drv = bdrv_find_format(backing_fmt->value.s);
5088 error_setg(errp, "Unknown backing file format '%s'",
5089 backing_fmt->value.s);
5094 // The size for the image must always be specified, with one exception:
5095 // If we are using a backing file, we can obtain the size from there
5096 size = get_option_parameter(param, BLOCK_OPT_SIZE);
5097 if (size && size->value.n == -1) {
5098 if (backing_file && backing_file->value.s) {
5099 BlockDriverState *bs;
5104 /* backing files always opened read-only */
5106 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
5110 ret = bdrv_open(bs, backing_file->value.s, NULL, back_flags,
5111 backing_drv, &local_err);
5113 error_setg_errno(errp, -ret, "Could not open '%s': %s",
5114 backing_file->value.s,
5115 error_get_pretty(local_err));
5116 error_free(local_err);
5121 bdrv_get_geometry(bs, &size);
5124 snprintf(buf, sizeof(buf), "%" PRId64, size);
5125 set_option_parameter(param, BLOCK_OPT_SIZE, buf);
5129 error_setg(errp, "Image creation needs a size parameter");
5135 printf("Formatting '%s', fmt=%s ", filename, fmt);
5136 print_option_parameters(param);
5139 ret = bdrv_create(drv, filename, param, &local_err);
5140 if (ret == -EFBIG) {
5141 /* This is generally a better message than whatever the driver would
5142 * deliver (especially because of the cluster_size_hint), since that
5143 * is most probably not much different from "image too large". */
5144 const char *cluster_size_hint = "";
5145 if (get_option_parameter(create_options, BLOCK_OPT_CLUSTER_SIZE)) {
5146 cluster_size_hint = " (try using a larger cluster size)";
5148 error_setg(errp, "The image size is too large for file format '%s'"
5149 "%s", fmt, cluster_size_hint);
5150 error_free(local_err);
5155 free_option_parameters(create_options);
5156 free_option_parameters(param);
5158 if (error_is_set(&local_err)) {
5159 error_propagate(errp, local_err);
5163 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
5165 /* Currently BlockDriverState always uses the main loop AioContext */
5166 return qemu_get_aio_context();
5169 void bdrv_add_before_write_notifier(BlockDriverState *bs,
5170 NotifierWithReturn *notifier)
5172 notifier_with_return_list_add(&bs->before_write_notifiers, notifier);
5175 int bdrv_amend_options(BlockDriverState *bs, QEMUOptionParameter *options)
5177 if (bs->drv->bdrv_amend_options == NULL) {
5180 return bs->drv->bdrv_amend_options(bs, options);
5183 /* Used to recurse on single child block filters.
5184 * Single child block filter will store their child in bs->file.
5186 bool bdrv_generic_is_first_non_filter(BlockDriverState *bs,
5187 BlockDriverState *candidate)
5193 if (!bs->drv->authorizations[BS_IS_A_FILTER]) {
5194 if (bs == candidate) {
5201 if (!bs->drv->authorizations[BS_FILTER_PASS_DOWN]) {
5209 return bdrv_recurse_is_first_non_filter(bs->file, candidate);
5212 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
5213 BlockDriverState *candidate)
5215 if (bs->drv && bs->drv->bdrv_recurse_is_first_non_filter) {
5216 return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate);
5219 return bdrv_generic_is_first_non_filter(bs, candidate);
5222 /* This function checks if the candidate is the first non filter bs down it's
5223 * bs chain. Since we don't have pointers to parents it explore all bs chains
5224 * from the top. Some filters can choose not to pass down the recursion.
5226 bool bdrv_is_first_non_filter(BlockDriverState *candidate)
5228 BlockDriverState *bs;
5230 /* walk down the bs forest recursively */
5231 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
5238 perm = bdrv_recurse_is_first_non_filter(bs->file, candidate);
5240 /* candidate is the first non filter */