2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24 #include "config-host.h"
25 #include "qemu-common.h"
27 #include "monitor/monitor.h"
28 #include "block/block_int.h"
29 #include "block/blockjob.h"
30 #include "qemu/module.h"
31 #include "qapi/qmp/qjson.h"
32 #include "sysemu/sysemu.h"
33 #include "qemu/notify.h"
34 #include "block/coroutine.h"
35 #include "block/qapi.h"
36 #include "qmp-commands.h"
37 #include "qemu/timer.h"
40 #include <sys/types.h>
42 #include <sys/ioctl.h>
43 #include <sys/queue.h>
53 struct BdrvDirtyBitmap {
55 QLIST_ENTRY(BdrvDirtyBitmap) list;
58 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
60 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load);
61 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
62 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
63 BlockDriverCompletionFunc *cb, void *opaque);
64 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
65 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
66 BlockDriverCompletionFunc *cb, void *opaque);
67 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
68 int64_t sector_num, int nb_sectors,
70 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
71 int64_t sector_num, int nb_sectors,
73 static int coroutine_fn bdrv_co_do_preadv(BlockDriverState *bs,
74 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
75 BdrvRequestFlags flags);
76 static int coroutine_fn bdrv_co_do_pwritev(BlockDriverState *bs,
77 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
78 BdrvRequestFlags flags);
79 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
83 BdrvRequestFlags flags,
84 BlockDriverCompletionFunc *cb,
87 static void coroutine_fn bdrv_co_do_rw(void *opaque);
88 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
89 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags);
91 static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
92 QTAILQ_HEAD_INITIALIZER(bdrv_states);
94 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
95 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
97 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
98 QLIST_HEAD_INITIALIZER(bdrv_drivers);
100 /* If non-zero, use only whitelisted block drivers */
101 static int use_bdrv_whitelist;
104 static int is_windows_drive_prefix(const char *filename)
106 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
107 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
111 int is_windows_drive(const char *filename)
113 if (is_windows_drive_prefix(filename) &&
116 if (strstart(filename, "\\\\.\\", NULL) ||
117 strstart(filename, "//./", NULL))
123 /* throttling disk I/O limits */
124 void bdrv_set_io_limits(BlockDriverState *bs,
129 throttle_config(&bs->throttle_state, cfg);
131 for (i = 0; i < 2; i++) {
132 qemu_co_enter_next(&bs->throttled_reqs[i]);
136 /* this function drain all the throttled IOs */
137 static bool bdrv_start_throttled_reqs(BlockDriverState *bs)
139 bool drained = false;
140 bool enabled = bs->io_limits_enabled;
143 bs->io_limits_enabled = false;
145 for (i = 0; i < 2; i++) {
146 while (qemu_co_enter_next(&bs->throttled_reqs[i])) {
151 bs->io_limits_enabled = enabled;
156 void bdrv_io_limits_disable(BlockDriverState *bs)
158 bs->io_limits_enabled = false;
160 bdrv_start_throttled_reqs(bs);
162 throttle_destroy(&bs->throttle_state);
165 static void bdrv_throttle_read_timer_cb(void *opaque)
167 BlockDriverState *bs = opaque;
168 qemu_co_enter_next(&bs->throttled_reqs[0]);
171 static void bdrv_throttle_write_timer_cb(void *opaque)
173 BlockDriverState *bs = opaque;
174 qemu_co_enter_next(&bs->throttled_reqs[1]);
177 /* should be called before bdrv_set_io_limits if a limit is set */
178 void bdrv_io_limits_enable(BlockDriverState *bs)
180 assert(!bs->io_limits_enabled);
181 throttle_init(&bs->throttle_state,
183 bdrv_throttle_read_timer_cb,
184 bdrv_throttle_write_timer_cb,
186 bs->io_limits_enabled = true;
189 /* This function makes an IO wait if needed
191 * @nb_sectors: the number of sectors of the IO
192 * @is_write: is the IO a write
194 static void bdrv_io_limits_intercept(BlockDriverState *bs,
198 /* does this io must wait */
199 bool must_wait = throttle_schedule_timer(&bs->throttle_state, is_write);
201 /* if must wait or any request of this type throttled queue the IO */
203 !qemu_co_queue_empty(&bs->throttled_reqs[is_write])) {
204 qemu_co_queue_wait(&bs->throttled_reqs[is_write]);
207 /* the IO will be executed, do the accounting */
208 throttle_account(&bs->throttle_state, is_write, bytes);
211 /* if the next request must wait -> do nothing */
212 if (throttle_schedule_timer(&bs->throttle_state, is_write)) {
216 /* else queue next request for execution */
217 qemu_co_queue_next(&bs->throttled_reqs[is_write]);
220 size_t bdrv_opt_mem_align(BlockDriverState *bs)
222 if (!bs || !bs->drv) {
223 /* 4k should be on the safe side */
227 return bs->bl.opt_mem_alignment;
230 /* check if the path starts with "<protocol>:" */
231 static int path_has_protocol(const char *path)
236 if (is_windows_drive(path) ||
237 is_windows_drive_prefix(path)) {
240 p = path + strcspn(path, ":/\\");
242 p = path + strcspn(path, ":/");
248 int path_is_absolute(const char *path)
251 /* specific case for names like: "\\.\d:" */
252 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
255 return (*path == '/' || *path == '\\');
257 return (*path == '/');
261 /* if filename is absolute, just copy it to dest. Otherwise, build a
262 path to it by considering it is relative to base_path. URL are
264 void path_combine(char *dest, int dest_size,
265 const char *base_path,
266 const char *filename)
273 if (path_is_absolute(filename)) {
274 pstrcpy(dest, dest_size, filename);
276 p = strchr(base_path, ':');
281 p1 = strrchr(base_path, '/');
285 p2 = strrchr(base_path, '\\');
297 if (len > dest_size - 1)
299 memcpy(dest, base_path, len);
301 pstrcat(dest, dest_size, filename);
305 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz)
307 if (bs->backing_file[0] == '\0' || path_has_protocol(bs->backing_file)) {
308 pstrcpy(dest, sz, bs->backing_file);
310 path_combine(dest, sz, bs->filename, bs->backing_file);
314 void bdrv_register(BlockDriver *bdrv)
316 /* Block drivers without coroutine functions need emulation */
317 if (!bdrv->bdrv_co_readv) {
318 bdrv->bdrv_co_readv = bdrv_co_readv_em;
319 bdrv->bdrv_co_writev = bdrv_co_writev_em;
321 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
322 * the block driver lacks aio we need to emulate that too.
324 if (!bdrv->bdrv_aio_readv) {
325 /* add AIO emulation layer */
326 bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
327 bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
331 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
334 /* create a new block device (by default it is empty) */
335 BlockDriverState *bdrv_new(const char *device_name)
337 BlockDriverState *bs;
339 bs = g_malloc0(sizeof(BlockDriverState));
340 QLIST_INIT(&bs->dirty_bitmaps);
341 pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
342 if (device_name[0] != '\0') {
343 QTAILQ_INSERT_TAIL(&bdrv_states, bs, device_list);
345 bdrv_iostatus_disable(bs);
346 notifier_list_init(&bs->close_notifiers);
347 notifier_with_return_list_init(&bs->before_write_notifiers);
348 qemu_co_queue_init(&bs->throttled_reqs[0]);
349 qemu_co_queue_init(&bs->throttled_reqs[1]);
355 void bdrv_add_close_notifier(BlockDriverState *bs, Notifier *notify)
357 notifier_list_add(&bs->close_notifiers, notify);
360 BlockDriver *bdrv_find_format(const char *format_name)
363 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
364 if (!strcmp(drv1->format_name, format_name)) {
371 static int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
373 static const char *whitelist_rw[] = {
374 CONFIG_BDRV_RW_WHITELIST
376 static const char *whitelist_ro[] = {
377 CONFIG_BDRV_RO_WHITELIST
381 if (!whitelist_rw[0] && !whitelist_ro[0]) {
382 return 1; /* no whitelist, anything goes */
385 for (p = whitelist_rw; *p; p++) {
386 if (!strcmp(drv->format_name, *p)) {
391 for (p = whitelist_ro; *p; p++) {
392 if (!strcmp(drv->format_name, *p)) {
400 BlockDriver *bdrv_find_whitelisted_format(const char *format_name,
403 BlockDriver *drv = bdrv_find_format(format_name);
404 return drv && bdrv_is_whitelisted(drv, read_only) ? drv : NULL;
407 typedef struct CreateCo {
410 QEMUOptionParameter *options;
415 static void coroutine_fn bdrv_create_co_entry(void *opaque)
417 Error *local_err = NULL;
420 CreateCo *cco = opaque;
423 ret = cco->drv->bdrv_create(cco->filename, cco->options, &local_err);
425 error_propagate(&cco->err, local_err);
430 int bdrv_create(BlockDriver *drv, const char* filename,
431 QEMUOptionParameter *options, Error **errp)
438 .filename = g_strdup(filename),
444 if (!drv->bdrv_create) {
445 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
450 if (qemu_in_coroutine()) {
451 /* Fast-path if already in coroutine context */
452 bdrv_create_co_entry(&cco);
454 co = qemu_coroutine_create(bdrv_create_co_entry);
455 qemu_coroutine_enter(co, &cco);
456 while (cco.ret == NOT_DONE) {
464 error_propagate(errp, cco.err);
466 error_setg_errno(errp, -ret, "Could not create image");
471 g_free(cco.filename);
475 int bdrv_create_file(const char* filename, QEMUOptionParameter *options,
479 Error *local_err = NULL;
482 drv = bdrv_find_protocol(filename, true);
484 error_setg(errp, "Could not find protocol for file '%s'", filename);
488 ret = bdrv_create(drv, filename, options, &local_err);
490 error_propagate(errp, local_err);
495 int bdrv_refresh_limits(BlockDriverState *bs)
497 BlockDriver *drv = bs->drv;
499 memset(&bs->bl, 0, sizeof(bs->bl));
505 /* Take some limits from the children as a default */
507 bdrv_refresh_limits(bs->file);
508 bs->bl.opt_transfer_length = bs->file->bl.opt_transfer_length;
509 bs->bl.opt_mem_alignment = bs->file->bl.opt_mem_alignment;
511 bs->bl.opt_mem_alignment = 512;
514 if (bs->backing_hd) {
515 bdrv_refresh_limits(bs->backing_hd);
516 bs->bl.opt_transfer_length =
517 MAX(bs->bl.opt_transfer_length,
518 bs->backing_hd->bl.opt_transfer_length);
519 bs->bl.opt_mem_alignment =
520 MAX(bs->bl.opt_mem_alignment,
521 bs->backing_hd->bl.opt_mem_alignment);
524 /* Then let the driver override it */
525 if (drv->bdrv_refresh_limits) {
526 return drv->bdrv_refresh_limits(bs);
533 * Create a uniquely-named empty temporary file.
534 * Return 0 upon success, otherwise a negative errno value.
536 int get_tmp_filename(char *filename, int size)
539 char temp_dir[MAX_PATH];
540 /* GetTempFileName requires that its output buffer (4th param)
541 have length MAX_PATH or greater. */
542 assert(size >= MAX_PATH);
543 return (GetTempPath(MAX_PATH, temp_dir)
544 && GetTempFileName(temp_dir, "qem", 0, filename)
545 ? 0 : -GetLastError());
549 tmpdir = getenv("TMPDIR");
552 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
555 fd = mkstemp(filename);
559 if (close(fd) != 0) {
568 * Detect host devices. By convention, /dev/cdrom[N] is always
569 * recognized as a host CDROM.
571 static BlockDriver *find_hdev_driver(const char *filename)
573 int score_max = 0, score;
574 BlockDriver *drv = NULL, *d;
576 QLIST_FOREACH(d, &bdrv_drivers, list) {
577 if (d->bdrv_probe_device) {
578 score = d->bdrv_probe_device(filename);
579 if (score > score_max) {
589 BlockDriver *bdrv_find_protocol(const char *filename,
590 bool allow_protocol_prefix)
597 /* TODO Drivers without bdrv_file_open must be specified explicitly */
600 * XXX(hch): we really should not let host device detection
601 * override an explicit protocol specification, but moving this
602 * later breaks access to device names with colons in them.
603 * Thanks to the brain-dead persistent naming schemes on udev-
604 * based Linux systems those actually are quite common.
606 drv1 = find_hdev_driver(filename);
611 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
612 return bdrv_find_format("file");
615 p = strchr(filename, ':');
618 if (len > sizeof(protocol) - 1)
619 len = sizeof(protocol) - 1;
620 memcpy(protocol, filename, len);
621 protocol[len] = '\0';
622 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
623 if (drv1->protocol_name &&
624 !strcmp(drv1->protocol_name, protocol)) {
631 static int find_image_format(BlockDriverState *bs, const char *filename,
632 BlockDriver **pdrv, Error **errp)
634 int score, score_max;
635 BlockDriver *drv1, *drv;
639 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
640 if (bs->sg || !bdrv_is_inserted(bs) || bdrv_getlength(bs) == 0) {
641 drv = bdrv_find_format("raw");
643 error_setg(errp, "Could not find raw image format");
650 ret = bdrv_pread(bs, 0, buf, sizeof(buf));
652 error_setg_errno(errp, -ret, "Could not read image for determining its "
660 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
661 if (drv1->bdrv_probe) {
662 score = drv1->bdrv_probe(buf, ret, filename);
663 if (score > score_max) {
670 error_setg(errp, "Could not determine image format: No compatible "
679 * Set the current 'total_sectors' value
681 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
683 BlockDriver *drv = bs->drv;
685 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
689 /* query actual device if possible, otherwise just trust the hint */
690 if (drv->bdrv_getlength) {
691 int64_t length = drv->bdrv_getlength(bs);
695 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
698 bs->total_sectors = hint;
703 * Set open flags for a given discard mode
705 * Return 0 on success, -1 if the discard mode was invalid.
707 int bdrv_parse_discard_flags(const char *mode, int *flags)
709 *flags &= ~BDRV_O_UNMAP;
711 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
713 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
714 *flags |= BDRV_O_UNMAP;
723 * Set open flags for a given cache mode
725 * Return 0 on success, -1 if the cache mode was invalid.
727 int bdrv_parse_cache_flags(const char *mode, int *flags)
729 *flags &= ~BDRV_O_CACHE_MASK;
731 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
732 *flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB;
733 } else if (!strcmp(mode, "directsync")) {
734 *flags |= BDRV_O_NOCACHE;
735 } else if (!strcmp(mode, "writeback")) {
736 *flags |= BDRV_O_CACHE_WB;
737 } else if (!strcmp(mode, "unsafe")) {
738 *flags |= BDRV_O_CACHE_WB;
739 *flags |= BDRV_O_NO_FLUSH;
740 } else if (!strcmp(mode, "writethrough")) {
741 /* this is the default */
750 * The copy-on-read flag is actually a reference count so multiple users may
751 * use the feature without worrying about clobbering its previous state.
752 * Copy-on-read stays enabled until all users have called to disable it.
754 void bdrv_enable_copy_on_read(BlockDriverState *bs)
759 void bdrv_disable_copy_on_read(BlockDriverState *bs)
761 assert(bs->copy_on_read > 0);
765 static int bdrv_open_flags(BlockDriverState *bs, int flags)
767 int open_flags = flags | BDRV_O_CACHE_WB;
770 * Clear flags that are internal to the block layer before opening the
773 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
776 * Snapshots should be writable.
778 if (bs->is_temporary) {
779 open_flags |= BDRV_O_RDWR;
785 static int bdrv_assign_node_name(BlockDriverState *bs,
786 const char *node_name,
793 /* empty string node name is invalid */
794 if (node_name[0] == '\0') {
795 error_setg(errp, "Empty node name");
799 /* takes care of avoiding namespaces collisions */
800 if (bdrv_find(node_name)) {
801 error_setg(errp, "node-name=%s is conflicting with a device id",
806 /* takes care of avoiding duplicates node names */
807 if (bdrv_find_node(node_name)) {
808 error_setg(errp, "Duplicate node name");
812 /* copy node name into the bs and insert it into the graph list */
813 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
814 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
820 * Common part for opening disk images and files
822 * Removes all processed options from *options.
824 static int bdrv_open_common(BlockDriverState *bs, BlockDriverState *file,
825 QDict *options, int flags, BlockDriver *drv, Error **errp)
828 const char *filename;
829 const char *node_name = NULL;
830 Error *local_err = NULL;
833 assert(bs->file == NULL);
834 assert(options != NULL && bs->options != options);
837 filename = file->filename;
839 filename = qdict_get_try_str(options, "filename");
842 if (drv->bdrv_needs_filename && !filename) {
843 error_setg(errp, "The '%s' block driver requires a file name",
848 trace_bdrv_open_common(bs, filename ?: "", flags, drv->format_name);
850 node_name = qdict_get_try_str(options, "node-name");
851 ret = bdrv_assign_node_name(bs, node_name, errp);
855 qdict_del(options, "node-name");
857 /* bdrv_open() with directly using a protocol as drv. This layer is already
858 * opened, so assign it to bs (while file becomes a closed BlockDriverState)
859 * and return immediately. */
860 if (file != NULL && drv->bdrv_file_open) {
865 bs->open_flags = flags;
866 bs->guest_block_size = 512;
867 bs->request_alignment = 512;
868 bs->zero_beyond_eof = true;
869 open_flags = bdrv_open_flags(bs, flags);
870 bs->read_only = !(open_flags & BDRV_O_RDWR);
872 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
874 !bs->read_only && bdrv_is_whitelisted(drv, true)
875 ? "Driver '%s' can only be used for read-only devices"
876 : "Driver '%s' is not whitelisted",
881 assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
882 if (flags & BDRV_O_COPY_ON_READ) {
883 if (!bs->read_only) {
884 bdrv_enable_copy_on_read(bs);
886 error_setg(errp, "Can't use copy-on-read on read-only device");
891 if (filename != NULL) {
892 pstrcpy(bs->filename, sizeof(bs->filename), filename);
894 bs->filename[0] = '\0';
898 bs->opaque = g_malloc0(drv->instance_size);
900 bs->enable_write_cache = !!(flags & BDRV_O_CACHE_WB);
902 /* Open the image, either directly or using a protocol */
903 if (drv->bdrv_file_open) {
904 assert(file == NULL);
905 assert(!drv->bdrv_needs_filename || filename != NULL);
906 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
909 error_setg(errp, "Can't use '%s' as a block driver for the "
910 "protocol level", drv->format_name);
915 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
920 error_propagate(errp, local_err);
921 } else if (bs->filename[0]) {
922 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
924 error_setg_errno(errp, -ret, "Could not open image");
929 ret = refresh_total_sectors(bs, bs->total_sectors);
931 error_setg_errno(errp, -ret, "Could not refresh total sector count");
935 bdrv_refresh_limits(bs);
936 assert(bdrv_opt_mem_align(bs) != 0);
937 assert(bs->request_alignment != 0);
940 if (bs->is_temporary) {
941 assert(bs->filename[0] != '\0');
942 unlink(bs->filename);
956 * Opens a file using a protocol (file, host_device, nbd, ...)
958 * options is a QDict of options to pass to the block drivers, or NULL for an
959 * empty set of options. The reference to the QDict belongs to the block layer
960 * after the call (even on failure), so if the caller intends to reuse the
961 * dictionary, it needs to use QINCREF() before calling bdrv_file_open.
963 int bdrv_file_open(BlockDriverState **pbs, const char *filename,
964 const char *reference, QDict *options, int flags,
967 BlockDriverState *bs = NULL;
970 bool allow_protocol_prefix = false;
971 Error *local_err = NULL;
974 /* NULL means an empty set of options */
975 if (options == NULL) {
976 options = qdict_new();
980 if (filename || qdict_size(options)) {
981 error_setg(errp, "Cannot reference an existing block device with "
982 "additional options or a new filename");
987 bs = bdrv_lookup_bs(reference, reference, errp);
997 bs->options = options;
998 options = qdict_clone_shallow(options);
1000 /* Fetch the file name from the options QDict if necessary */
1002 filename = qdict_get_try_str(options, "filename");
1003 } else if (filename && !qdict_haskey(options, "filename")) {
1004 qdict_put(options, "filename", qstring_from_str(filename));
1005 allow_protocol_prefix = true;
1007 error_setg(errp, "Can't specify 'file' and 'filename' options at the "
1013 /* Find the right block driver */
1014 drvname = qdict_get_try_str(options, "driver");
1016 drv = bdrv_find_format(drvname);
1018 error_setg(errp, "Unknown driver '%s'", drvname);
1020 qdict_del(options, "driver");
1021 } else if (filename) {
1022 drv = bdrv_find_protocol(filename, allow_protocol_prefix);
1024 error_setg(errp, "Unknown protocol");
1027 error_setg(errp, "Must specify either driver or file");
1032 /* errp has been set already */
1037 /* Parse the filename and open it */
1038 if (drv->bdrv_parse_filename && filename) {
1039 drv->bdrv_parse_filename(filename, options, &local_err);
1041 error_propagate(errp, local_err);
1045 qdict_del(options, "filename");
1048 if (!drv->bdrv_file_open) {
1049 ret = bdrv_open(&bs, filename, options, flags, drv, &local_err);
1052 ret = bdrv_open_common(bs, NULL, options, flags, drv, &local_err);
1055 error_propagate(errp, local_err);
1059 /* Check if any unknown options were used */
1060 if (options && (qdict_size(options) != 0)) {
1061 const QDictEntry *entry = qdict_first(options);
1062 error_setg(errp, "Block protocol '%s' doesn't support the option '%s'",
1063 drv->format_name, entry->key);
1076 QDECREF(bs->options);
1083 * Opens the backing file for a BlockDriverState if not yet open
1085 * options is a QDict of options to pass to the block drivers, or NULL for an
1086 * empty set of options. The reference to the QDict is transferred to this
1087 * function (even on failure), so if the caller intends to reuse the dictionary,
1088 * it needs to use QINCREF() before calling bdrv_file_open.
1090 int bdrv_open_backing_file(BlockDriverState *bs, QDict *options, Error **errp)
1092 char backing_filename[PATH_MAX];
1093 int back_flags, ret;
1094 BlockDriver *back_drv = NULL;
1095 Error *local_err = NULL;
1097 if (bs->backing_hd != NULL) {
1102 /* NULL means an empty set of options */
1103 if (options == NULL) {
1104 options = qdict_new();
1107 bs->open_flags &= ~BDRV_O_NO_BACKING;
1108 if (qdict_haskey(options, "file.filename")) {
1109 backing_filename[0] = '\0';
1110 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
1114 bdrv_get_full_backing_filename(bs, backing_filename,
1115 sizeof(backing_filename));
1118 if (bs->backing_format[0] != '\0') {
1119 back_drv = bdrv_find_format(bs->backing_format);
1122 /* backing files always opened read-only */
1123 back_flags = bs->open_flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT |
1124 BDRV_O_COPY_ON_READ);
1126 assert(bs->backing_hd == NULL);
1127 ret = bdrv_open(&bs->backing_hd,
1128 *backing_filename ? backing_filename : NULL, options,
1129 back_flags, back_drv, &local_err);
1131 bs->backing_hd = NULL;
1132 bs->open_flags |= BDRV_O_NO_BACKING;
1133 error_setg(errp, "Could not open backing file: %s",
1134 error_get_pretty(local_err));
1135 error_free(local_err);
1139 if (bs->backing_hd->file) {
1140 pstrcpy(bs->backing_file, sizeof(bs->backing_file),
1141 bs->backing_hd->file->filename);
1144 /* Recalculate the BlockLimits with the backing file */
1145 bdrv_refresh_limits(bs);
1151 * Opens a disk image whose options are given as BlockdevRef in another block
1154 * If force_raw is true, bdrv_file_open() will be used, thereby preventing any
1155 * image format auto-detection. If it is false and a filename is given,
1156 * bdrv_open() will be used for auto-detection.
1158 * If allow_none is true, no image will be opened if filename is false and no
1159 * BlockdevRef is given. *pbs will remain unchanged and 0 will be returned.
1161 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
1162 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
1163 * itself, all options starting with "${bdref_key}." are considered part of the
1166 * The BlockdevRef will be removed from the options QDict.
1168 * To conform with the behavior of bdrv_open(), *pbs has to be NULL.
1170 int bdrv_open_image(BlockDriverState **pbs, const char *filename,
1171 QDict *options, const char *bdref_key, int flags,
1172 bool force_raw, bool allow_none, Error **errp)
1174 QDict *image_options;
1176 char *bdref_key_dot;
1177 const char *reference;
1180 assert(*pbs == NULL);
1182 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
1183 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
1184 g_free(bdref_key_dot);
1186 reference = qdict_get_try_str(options, bdref_key);
1187 if (!filename && !reference && !qdict_size(image_options)) {
1191 error_setg(errp, "A block device must be specified for \"%s\"",
1198 if (filename && !force_raw) {
1199 /* If a filename is given and the block driver should be detected
1200 automatically (instead of using none), use bdrv_open() in order to do
1201 that auto-detection. */
1203 error_setg(errp, "Cannot reference an existing block device while "
1204 "giving a filename");
1209 ret = bdrv_open(pbs, filename, image_options, flags, NULL, errp);
1211 ret = bdrv_file_open(pbs, filename, reference, image_options, flags,
1216 qdict_del(options, bdref_key);
1221 * Opens a disk image (raw, qcow2, vmdk, ...)
1223 * options is a QDict of options to pass to the block drivers, or NULL for an
1224 * empty set of options. The reference to the QDict belongs to the block layer
1225 * after the call (even on failure), so if the caller intends to reuse the
1226 * dictionary, it needs to use QINCREF() before calling bdrv_open.
1228 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
1229 * If it is not NULL, the referenced BDS will be reused.
1231 int bdrv_open(BlockDriverState **pbs, const char *filename, QDict *options,
1232 int flags, BlockDriver *drv, Error **errp)
1235 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
1236 char tmp_filename[PATH_MAX + 1];
1237 BlockDriverState *file = NULL, *bs;
1238 const char *drvname;
1239 Error *local_err = NULL;
1249 /* NULL means an empty set of options */
1250 if (options == NULL) {
1251 options = qdict_new();
1254 bs->options = options;
1255 options = qdict_clone_shallow(options);
1257 /* For snapshot=on, create a temporary qcow2 overlay */
1258 if (flags & BDRV_O_SNAPSHOT) {
1259 BlockDriverState *bs1;
1261 BlockDriver *bdrv_qcow2;
1262 QEMUOptionParameter *create_options;
1263 QDict *snapshot_options;
1265 /* if snapshot, we create a temporary backing file and open it
1266 instead of opening 'filename' directly */
1268 /* Get the required size from the image */
1271 ret = bdrv_open(&bs1, filename, options, BDRV_O_NO_BACKING,
1276 total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
1280 /* Create the temporary image */
1281 ret = get_tmp_filename(tmp_filename, sizeof(tmp_filename));
1283 error_setg_errno(errp, -ret, "Could not get temporary filename");
1287 bdrv_qcow2 = bdrv_find_format("qcow2");
1288 create_options = parse_option_parameters("", bdrv_qcow2->create_options,
1291 set_option_parameter_int(create_options, BLOCK_OPT_SIZE, total_size);
1293 ret = bdrv_create(bdrv_qcow2, tmp_filename, create_options, &local_err);
1294 free_option_parameters(create_options);
1296 error_setg_errno(errp, -ret, "Could not create temporary overlay "
1297 "'%s': %s", tmp_filename,
1298 error_get_pretty(local_err));
1299 error_free(local_err);
1304 /* Prepare a new options QDict for the temporary file, where user
1305 * options refer to the backing file */
1307 qdict_put(options, "file.filename", qstring_from_str(filename));
1310 qdict_put(options, "driver", qstring_from_str(drv->format_name));
1313 snapshot_options = qdict_new();
1314 qdict_put(snapshot_options, "backing", options);
1315 qdict_flatten(snapshot_options);
1317 bs->options = snapshot_options;
1318 options = qdict_clone_shallow(bs->options);
1320 filename = tmp_filename;
1322 bs->is_temporary = 1;
1325 /* Open image file without format layer */
1326 if (flags & BDRV_O_RDWR) {
1327 flags |= BDRV_O_ALLOW_RDWR;
1330 assert(file == NULL);
1331 ret = bdrv_open_image(&file, filename, options, "file",
1332 bdrv_open_flags(bs, flags | BDRV_O_UNMAP), true, true,
1338 /* Find the right image format driver */
1339 drvname = qdict_get_try_str(options, "driver");
1341 drv = bdrv_find_format(drvname);
1342 qdict_del(options, "driver");
1344 error_setg(errp, "Invalid driver: '%s'", drvname);
1346 goto unlink_and_fail;
1352 ret = find_image_format(file, filename, &drv, &local_err);
1354 error_setg(errp, "Must specify either driver or file");
1356 goto unlink_and_fail;
1361 goto unlink_and_fail;
1364 /* Open the image */
1365 ret = bdrv_open_common(bs, file, options, flags, drv, &local_err);
1367 goto unlink_and_fail;
1370 if (file && (bs->file != file)) {
1375 /* If there is a backing file, use it */
1376 if ((flags & BDRV_O_NO_BACKING) == 0) {
1377 QDict *backing_options;
1379 qdict_extract_subqdict(options, &backing_options, "backing.");
1380 ret = bdrv_open_backing_file(bs, backing_options, &local_err);
1382 goto close_and_fail;
1386 /* Check if any unknown options were used */
1387 if (qdict_size(options) != 0) {
1388 const QDictEntry *entry = qdict_first(options);
1389 error_setg(errp, "Block format '%s' used by device '%s' doesn't "
1390 "support the option '%s'", drv->format_name, bs->device_name,
1394 goto close_and_fail;
1398 if (!bdrv_key_required(bs)) {
1399 bdrv_dev_change_media_cb(bs, true);
1409 if (bs->is_temporary) {
1413 QDECREF(bs->options);
1417 /* If *pbs is NULL, a new BDS has been created in this function and
1418 needs to be freed now. Otherwise, it does not need to be closed,
1419 since it has not really been opened yet. */
1423 error_propagate(errp, local_err);
1428 /* See fail path, but now the BDS has to be always closed */
1436 error_propagate(errp, local_err);
1441 typedef struct BlockReopenQueueEntry {
1443 BDRVReopenState state;
1444 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
1445 } BlockReopenQueueEntry;
1448 * Adds a BlockDriverState to a simple queue for an atomic, transactional
1449 * reopen of multiple devices.
1451 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
1452 * already performed, or alternatively may be NULL a new BlockReopenQueue will
1453 * be created and initialized. This newly created BlockReopenQueue should be
1454 * passed back in for subsequent calls that are intended to be of the same
1457 * bs is the BlockDriverState to add to the reopen queue.
1459 * flags contains the open flags for the associated bs
1461 * returns a pointer to bs_queue, which is either the newly allocated
1462 * bs_queue, or the existing bs_queue being used.
1465 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
1466 BlockDriverState *bs, int flags)
1470 BlockReopenQueueEntry *bs_entry;
1471 if (bs_queue == NULL) {
1472 bs_queue = g_new0(BlockReopenQueue, 1);
1473 QSIMPLEQ_INIT(bs_queue);
1477 bdrv_reopen_queue(bs_queue, bs->file, flags);
1480 bs_entry = g_new0(BlockReopenQueueEntry, 1);
1481 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
1483 bs_entry->state.bs = bs;
1484 bs_entry->state.flags = flags;
1490 * Reopen multiple BlockDriverStates atomically & transactionally.
1492 * The queue passed in (bs_queue) must have been built up previous
1493 * via bdrv_reopen_queue().
1495 * Reopens all BDS specified in the queue, with the appropriate
1496 * flags. All devices are prepared for reopen, and failure of any
1497 * device will cause all device changes to be abandonded, and intermediate
1500 * If all devices prepare successfully, then the changes are committed
1504 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
1507 BlockReopenQueueEntry *bs_entry, *next;
1508 Error *local_err = NULL;
1510 assert(bs_queue != NULL);
1514 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
1515 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
1516 error_propagate(errp, local_err);
1519 bs_entry->prepared = true;
1522 /* If we reach this point, we have success and just need to apply the
1525 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
1526 bdrv_reopen_commit(&bs_entry->state);
1532 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
1533 if (ret && bs_entry->prepared) {
1534 bdrv_reopen_abort(&bs_entry->state);
1543 /* Reopen a single BlockDriverState with the specified flags. */
1544 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
1547 Error *local_err = NULL;
1548 BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, bdrv_flags);
1550 ret = bdrv_reopen_multiple(queue, &local_err);
1551 if (local_err != NULL) {
1552 error_propagate(errp, local_err);
1559 * Prepares a BlockDriverState for reopen. All changes are staged in the
1560 * 'opaque' field of the BDRVReopenState, which is used and allocated by
1561 * the block driver layer .bdrv_reopen_prepare()
1563 * bs is the BlockDriverState to reopen
1564 * flags are the new open flags
1565 * queue is the reopen queue
1567 * Returns 0 on success, non-zero on error. On error errp will be set
1570 * On failure, bdrv_reopen_abort() will be called to clean up any data.
1571 * It is the responsibility of the caller to then call the abort() or
1572 * commit() for any other BDS that have been left in a prepare() state
1575 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
1579 Error *local_err = NULL;
1582 assert(reopen_state != NULL);
1583 assert(reopen_state->bs->drv != NULL);
1584 drv = reopen_state->bs->drv;
1586 /* if we are to stay read-only, do not allow permission change
1588 if (!(reopen_state->bs->open_flags & BDRV_O_ALLOW_RDWR) &&
1589 reopen_state->flags & BDRV_O_RDWR) {
1590 error_set(errp, QERR_DEVICE_IS_READ_ONLY,
1591 reopen_state->bs->device_name);
1596 ret = bdrv_flush(reopen_state->bs);
1598 error_set(errp, ERROR_CLASS_GENERIC_ERROR, "Error (%s) flushing drive",
1603 if (drv->bdrv_reopen_prepare) {
1604 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
1606 if (local_err != NULL) {
1607 error_propagate(errp, local_err);
1609 error_setg(errp, "failed while preparing to reopen image '%s'",
1610 reopen_state->bs->filename);
1615 /* It is currently mandatory to have a bdrv_reopen_prepare()
1616 * handler for each supported drv. */
1617 error_set(errp, QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
1618 drv->format_name, reopen_state->bs->device_name,
1619 "reopening of file");
1631 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
1632 * makes them final by swapping the staging BlockDriverState contents into
1633 * the active BlockDriverState contents.
1635 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
1639 assert(reopen_state != NULL);
1640 drv = reopen_state->bs->drv;
1641 assert(drv != NULL);
1643 /* If there are any driver level actions to take */
1644 if (drv->bdrv_reopen_commit) {
1645 drv->bdrv_reopen_commit(reopen_state);
1648 /* set BDS specific flags now */
1649 reopen_state->bs->open_flags = reopen_state->flags;
1650 reopen_state->bs->enable_write_cache = !!(reopen_state->flags &
1652 reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
1654 bdrv_refresh_limits(reopen_state->bs);
1658 * Abort the reopen, and delete and free the staged changes in
1661 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
1665 assert(reopen_state != NULL);
1666 drv = reopen_state->bs->drv;
1667 assert(drv != NULL);
1669 if (drv->bdrv_reopen_abort) {
1670 drv->bdrv_reopen_abort(reopen_state);
1675 void bdrv_close(BlockDriverState *bs)
1678 block_job_cancel_sync(bs->job);
1680 bdrv_drain_all(); /* complete I/O */
1682 bdrv_drain_all(); /* in case flush left pending I/O */
1683 notifier_list_notify(&bs->close_notifiers, bs);
1686 if (bs->backing_hd) {
1687 bdrv_unref(bs->backing_hd);
1688 bs->backing_hd = NULL;
1690 bs->drv->bdrv_close(bs);
1693 if (bs->is_temporary) {
1694 unlink(bs->filename);
1699 bs->copy_on_read = 0;
1700 bs->backing_file[0] = '\0';
1701 bs->backing_format[0] = '\0';
1702 bs->total_sectors = 0;
1707 bs->zero_beyond_eof = false;
1708 QDECREF(bs->options);
1711 if (bs->file != NULL) {
1712 bdrv_unref(bs->file);
1717 bdrv_dev_change_media_cb(bs, false);
1719 /*throttling disk I/O limits*/
1720 if (bs->io_limits_enabled) {
1721 bdrv_io_limits_disable(bs);
1725 void bdrv_close_all(void)
1727 BlockDriverState *bs;
1729 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
1734 /* Check if any requests are in-flight (including throttled requests) */
1735 static bool bdrv_requests_pending(BlockDriverState *bs)
1737 if (!QLIST_EMPTY(&bs->tracked_requests)) {
1740 if (!qemu_co_queue_empty(&bs->throttled_reqs[0])) {
1743 if (!qemu_co_queue_empty(&bs->throttled_reqs[1])) {
1746 if (bs->file && bdrv_requests_pending(bs->file)) {
1749 if (bs->backing_hd && bdrv_requests_pending(bs->backing_hd)) {
1755 static bool bdrv_requests_pending_all(void)
1757 BlockDriverState *bs;
1758 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
1759 if (bdrv_requests_pending(bs)) {
1767 * Wait for pending requests to complete across all BlockDriverStates
1769 * This function does not flush data to disk, use bdrv_flush_all() for that
1770 * after calling this function.
1772 * Note that completion of an asynchronous I/O operation can trigger any
1773 * number of other I/O operations on other devices---for example a coroutine
1774 * can be arbitrarily complex and a constant flow of I/O can come until the
1775 * coroutine is complete. Because of this, it is not possible to have a
1776 * function to drain a single device's I/O queue.
1778 void bdrv_drain_all(void)
1780 /* Always run first iteration so any pending completion BHs run */
1782 BlockDriverState *bs;
1785 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
1786 bdrv_start_throttled_reqs(bs);
1789 busy = bdrv_requests_pending_all();
1790 busy |= aio_poll(qemu_get_aio_context(), busy);
1794 /* make a BlockDriverState anonymous by removing from bdrv_state and
1795 * graph_bdrv_state list.
1796 Also, NULL terminate the device_name to prevent double remove */
1797 void bdrv_make_anon(BlockDriverState *bs)
1799 if (bs->device_name[0] != '\0') {
1800 QTAILQ_REMOVE(&bdrv_states, bs, device_list);
1802 bs->device_name[0] = '\0';
1803 if (bs->node_name[0] != '\0') {
1804 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
1806 bs->node_name[0] = '\0';
1809 static void bdrv_rebind(BlockDriverState *bs)
1811 if (bs->drv && bs->drv->bdrv_rebind) {
1812 bs->drv->bdrv_rebind(bs);
1816 static void bdrv_move_feature_fields(BlockDriverState *bs_dest,
1817 BlockDriverState *bs_src)
1819 /* move some fields that need to stay attached to the device */
1820 bs_dest->open_flags = bs_src->open_flags;
1823 bs_dest->dev_ops = bs_src->dev_ops;
1824 bs_dest->dev_opaque = bs_src->dev_opaque;
1825 bs_dest->dev = bs_src->dev;
1826 bs_dest->guest_block_size = bs_src->guest_block_size;
1827 bs_dest->copy_on_read = bs_src->copy_on_read;
1829 bs_dest->enable_write_cache = bs_src->enable_write_cache;
1831 /* i/o throttled req */
1832 memcpy(&bs_dest->throttle_state,
1833 &bs_src->throttle_state,
1834 sizeof(ThrottleState));
1835 bs_dest->throttled_reqs[0] = bs_src->throttled_reqs[0];
1836 bs_dest->throttled_reqs[1] = bs_src->throttled_reqs[1];
1837 bs_dest->io_limits_enabled = bs_src->io_limits_enabled;
1840 bs_dest->on_read_error = bs_src->on_read_error;
1841 bs_dest->on_write_error = bs_src->on_write_error;
1844 bs_dest->iostatus_enabled = bs_src->iostatus_enabled;
1845 bs_dest->iostatus = bs_src->iostatus;
1848 bs_dest->dirty_bitmaps = bs_src->dirty_bitmaps;
1850 /* reference count */
1851 bs_dest->refcnt = bs_src->refcnt;
1854 bs_dest->in_use = bs_src->in_use;
1855 bs_dest->job = bs_src->job;
1857 /* keep the same entry in bdrv_states */
1858 pstrcpy(bs_dest->device_name, sizeof(bs_dest->device_name),
1859 bs_src->device_name);
1860 bs_dest->device_list = bs_src->device_list;
1862 /* keep the same entry in graph_bdrv_states
1863 * We do want to swap name but don't want to swap linked list entries
1865 bs_dest->node_list = bs_src->node_list;
1869 * Swap bs contents for two image chains while they are live,
1870 * while keeping required fields on the BlockDriverState that is
1871 * actually attached to a device.
1873 * This will modify the BlockDriverState fields, and swap contents
1874 * between bs_new and bs_old. Both bs_new and bs_old are modified.
1876 * bs_new is required to be anonymous.
1878 * This function does not create any image files.
1880 void bdrv_swap(BlockDriverState *bs_new, BlockDriverState *bs_old)
1882 BlockDriverState tmp;
1884 /* bs_new must be anonymous and shouldn't have anything fancy enabled */
1885 assert(bs_new->device_name[0] == '\0');
1886 assert(QLIST_EMPTY(&bs_new->dirty_bitmaps));
1887 assert(bs_new->job == NULL);
1888 assert(bs_new->dev == NULL);
1889 assert(bs_new->in_use == 0);
1890 assert(bs_new->io_limits_enabled == false);
1891 assert(!throttle_have_timer(&bs_new->throttle_state));
1897 /* there are some fields that should not be swapped, move them back */
1898 bdrv_move_feature_fields(&tmp, bs_old);
1899 bdrv_move_feature_fields(bs_old, bs_new);
1900 bdrv_move_feature_fields(bs_new, &tmp);
1902 /* bs_new shouldn't be in bdrv_states even after the swap! */
1903 assert(bs_new->device_name[0] == '\0');
1905 /* Check a few fields that should remain attached to the device */
1906 assert(bs_new->dev == NULL);
1907 assert(bs_new->job == NULL);
1908 assert(bs_new->in_use == 0);
1909 assert(bs_new->io_limits_enabled == false);
1910 assert(!throttle_have_timer(&bs_new->throttle_state));
1912 bdrv_rebind(bs_new);
1913 bdrv_rebind(bs_old);
1917 * Add new bs contents at the top of an image chain while the chain is
1918 * live, while keeping required fields on the top layer.
1920 * This will modify the BlockDriverState fields, and swap contents
1921 * between bs_new and bs_top. Both bs_new and bs_top are modified.
1923 * bs_new is required to be anonymous.
1925 * This function does not create any image files.
1927 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
1929 bdrv_swap(bs_new, bs_top);
1931 /* The contents of 'tmp' will become bs_top, as we are
1932 * swapping bs_new and bs_top contents. */
1933 bs_top->backing_hd = bs_new;
1934 bs_top->open_flags &= ~BDRV_O_NO_BACKING;
1935 pstrcpy(bs_top->backing_file, sizeof(bs_top->backing_file),
1937 pstrcpy(bs_top->backing_format, sizeof(bs_top->backing_format),
1938 bs_new->drv ? bs_new->drv->format_name : "");
1941 static void bdrv_delete(BlockDriverState *bs)
1945 assert(!bs->in_use);
1946 assert(!bs->refcnt);
1947 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
1951 /* remove from list, if necessary */
1957 int bdrv_attach_dev(BlockDriverState *bs, void *dev)
1958 /* TODO change to DeviceState *dev when all users are qdevified */
1964 bdrv_iostatus_reset(bs);
1968 /* TODO qdevified devices don't use this, remove when devices are qdevified */
1969 void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev)
1971 if (bdrv_attach_dev(bs, dev) < 0) {
1976 void bdrv_detach_dev(BlockDriverState *bs, void *dev)
1977 /* TODO change to DeviceState *dev when all users are qdevified */
1979 assert(bs->dev == dev);
1982 bs->dev_opaque = NULL;
1983 bs->guest_block_size = 512;
1986 /* TODO change to return DeviceState * when all users are qdevified */
1987 void *bdrv_get_attached_dev(BlockDriverState *bs)
1992 void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops,
1996 bs->dev_opaque = opaque;
1999 void bdrv_emit_qmp_error_event(const BlockDriverState *bdrv,
2000 enum MonitorEvent ev,
2001 BlockErrorAction action, bool is_read)
2004 const char *action_str;
2007 case BDRV_ACTION_REPORT:
2008 action_str = "report";
2010 case BDRV_ACTION_IGNORE:
2011 action_str = "ignore";
2013 case BDRV_ACTION_STOP:
2014 action_str = "stop";
2020 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
2023 is_read ? "read" : "write");
2024 monitor_protocol_event(ev, data);
2026 qobject_decref(data);
2029 static void bdrv_emit_qmp_eject_event(BlockDriverState *bs, bool ejected)
2033 data = qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }",
2034 bdrv_get_device_name(bs), ejected);
2035 monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED, data);
2037 qobject_decref(data);
2040 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load)
2042 if (bs->dev_ops && bs->dev_ops->change_media_cb) {
2043 bool tray_was_closed = !bdrv_dev_is_tray_open(bs);
2044 bs->dev_ops->change_media_cb(bs->dev_opaque, load);
2045 if (tray_was_closed) {
2047 bdrv_emit_qmp_eject_event(bs, true);
2051 bdrv_emit_qmp_eject_event(bs, false);
2056 bool bdrv_dev_has_removable_media(BlockDriverState *bs)
2058 return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb);
2061 void bdrv_dev_eject_request(BlockDriverState *bs, bool force)
2063 if (bs->dev_ops && bs->dev_ops->eject_request_cb) {
2064 bs->dev_ops->eject_request_cb(bs->dev_opaque, force);
2068 bool bdrv_dev_is_tray_open(BlockDriverState *bs)
2070 if (bs->dev_ops && bs->dev_ops->is_tray_open) {
2071 return bs->dev_ops->is_tray_open(bs->dev_opaque);
2076 static void bdrv_dev_resize_cb(BlockDriverState *bs)
2078 if (bs->dev_ops && bs->dev_ops->resize_cb) {
2079 bs->dev_ops->resize_cb(bs->dev_opaque);
2083 bool bdrv_dev_is_medium_locked(BlockDriverState *bs)
2085 if (bs->dev_ops && bs->dev_ops->is_medium_locked) {
2086 return bs->dev_ops->is_medium_locked(bs->dev_opaque);
2092 * Run consistency checks on an image
2094 * Returns 0 if the check could be completed (it doesn't mean that the image is
2095 * free of errors) or -errno when an internal error occurred. The results of the
2096 * check are stored in res.
2098 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix)
2100 if (bs->drv->bdrv_check == NULL) {
2104 memset(res, 0, sizeof(*res));
2105 return bs->drv->bdrv_check(bs, res, fix);
2108 #define COMMIT_BUF_SECTORS 2048
2110 /* commit COW file into the raw image */
2111 int bdrv_commit(BlockDriverState *bs)
2113 BlockDriver *drv = bs->drv;
2114 int64_t sector, total_sectors, length, backing_length;
2115 int n, ro, open_flags;
2117 uint8_t *buf = NULL;
2118 char filename[PATH_MAX];
2123 if (!bs->backing_hd) {
2127 if (bdrv_in_use(bs) || bdrv_in_use(bs->backing_hd)) {
2131 ro = bs->backing_hd->read_only;
2132 /* Use pstrcpy (not strncpy): filename must be NUL-terminated. */
2133 pstrcpy(filename, sizeof(filename), bs->backing_hd->filename);
2134 open_flags = bs->backing_hd->open_flags;
2137 if (bdrv_reopen(bs->backing_hd, open_flags | BDRV_O_RDWR, NULL)) {
2142 length = bdrv_getlength(bs);
2148 backing_length = bdrv_getlength(bs->backing_hd);
2149 if (backing_length < 0) {
2150 ret = backing_length;
2154 /* If our top snapshot is larger than the backing file image,
2155 * grow the backing file image if possible. If not possible,
2156 * we must return an error */
2157 if (length > backing_length) {
2158 ret = bdrv_truncate(bs->backing_hd, length);
2164 total_sectors = length >> BDRV_SECTOR_BITS;
2165 buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
2167 for (sector = 0; sector < total_sectors; sector += n) {
2168 ret = bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n);
2173 ret = bdrv_read(bs, sector, buf, n);
2178 ret = bdrv_write(bs->backing_hd, sector, buf, n);
2185 if (drv->bdrv_make_empty) {
2186 ret = drv->bdrv_make_empty(bs);
2194 * Make sure all data we wrote to the backing device is actually
2197 if (bs->backing_hd) {
2198 bdrv_flush(bs->backing_hd);
2206 /* ignoring error return here */
2207 bdrv_reopen(bs->backing_hd, open_flags & ~BDRV_O_RDWR, NULL);
2213 int bdrv_commit_all(void)
2215 BlockDriverState *bs;
2217 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
2218 if (bs->drv && bs->backing_hd) {
2219 int ret = bdrv_commit(bs);
2229 * Remove an active request from the tracked requests list
2231 * This function should be called when a tracked request is completing.
2233 static void tracked_request_end(BdrvTrackedRequest *req)
2235 if (req->serialising) {
2236 req->bs->serialising_in_flight--;
2239 QLIST_REMOVE(req, list);
2240 qemu_co_queue_restart_all(&req->wait_queue);
2244 * Add an active request to the tracked requests list
2246 static void tracked_request_begin(BdrvTrackedRequest *req,
2247 BlockDriverState *bs,
2249 unsigned int bytes, bool is_write)
2251 *req = (BdrvTrackedRequest){
2255 .is_write = is_write,
2256 .co = qemu_coroutine_self(),
2257 .serialising = false,
2258 .overlap_offset = offset,
2259 .overlap_bytes = bytes,
2262 qemu_co_queue_init(&req->wait_queue);
2264 QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
2267 static void mark_request_serialising(BdrvTrackedRequest *req, uint64_t align)
2269 int64_t overlap_offset = req->offset & ~(align - 1);
2270 unsigned int overlap_bytes = ROUND_UP(req->offset + req->bytes, align)
2273 if (!req->serialising) {
2274 req->bs->serialising_in_flight++;
2275 req->serialising = true;
2278 req->overlap_offset = MIN(req->overlap_offset, overlap_offset);
2279 req->overlap_bytes = MAX(req->overlap_bytes, overlap_bytes);
2283 * Round a region to cluster boundaries
2285 void bdrv_round_to_clusters(BlockDriverState *bs,
2286 int64_t sector_num, int nb_sectors,
2287 int64_t *cluster_sector_num,
2288 int *cluster_nb_sectors)
2290 BlockDriverInfo bdi;
2292 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
2293 *cluster_sector_num = sector_num;
2294 *cluster_nb_sectors = nb_sectors;
2296 int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE;
2297 *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c);
2298 *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num +
2303 static int bdrv_get_cluster_size(BlockDriverState *bs)
2305 BlockDriverInfo bdi;
2308 ret = bdrv_get_info(bs, &bdi);
2309 if (ret < 0 || bdi.cluster_size == 0) {
2310 return bs->request_alignment;
2312 return bdi.cluster_size;
2316 static bool tracked_request_overlaps(BdrvTrackedRequest *req,
2317 int64_t offset, unsigned int bytes)
2320 if (offset >= req->overlap_offset + req->overlap_bytes) {
2324 if (req->overlap_offset >= offset + bytes) {
2330 static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self)
2332 BlockDriverState *bs = self->bs;
2333 BdrvTrackedRequest *req;
2335 bool waited = false;
2337 if (!bs->serialising_in_flight) {
2343 QLIST_FOREACH(req, &bs->tracked_requests, list) {
2344 if (req == self || (!req->serialising && !self->serialising)) {
2347 if (tracked_request_overlaps(req, self->overlap_offset,
2348 self->overlap_bytes))
2350 /* Hitting this means there was a reentrant request, for
2351 * example, a block driver issuing nested requests. This must
2352 * never happen since it means deadlock.
2354 assert(qemu_coroutine_self() != req->co);
2356 /* If the request is already (indirectly) waiting for us, or
2357 * will wait for us as soon as it wakes up, then just go on
2358 * (instead of producing a deadlock in the former case). */
2359 if (!req->waiting_for) {
2360 self->waiting_for = req;
2361 qemu_co_queue_wait(&req->wait_queue);
2362 self->waiting_for = NULL;
2377 * -EINVAL - backing format specified, but no file
2378 * -ENOSPC - can't update the backing file because no space is left in the
2380 * -ENOTSUP - format driver doesn't support changing the backing file
2382 int bdrv_change_backing_file(BlockDriverState *bs,
2383 const char *backing_file, const char *backing_fmt)
2385 BlockDriver *drv = bs->drv;
2388 /* Backing file format doesn't make sense without a backing file */
2389 if (backing_fmt && !backing_file) {
2393 if (drv->bdrv_change_backing_file != NULL) {
2394 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
2400 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
2401 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
2407 * Finds the image layer in the chain that has 'bs' as its backing file.
2409 * active is the current topmost image.
2411 * Returns NULL if bs is not found in active's image chain,
2412 * or if active == bs.
2414 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
2415 BlockDriverState *bs)
2417 BlockDriverState *overlay = NULL;
2418 BlockDriverState *intermediate;
2420 assert(active != NULL);
2423 /* if bs is the same as active, then by definition it has no overlay
2429 intermediate = active;
2430 while (intermediate->backing_hd) {
2431 if (intermediate->backing_hd == bs) {
2432 overlay = intermediate;
2435 intermediate = intermediate->backing_hd;
2441 typedef struct BlkIntermediateStates {
2442 BlockDriverState *bs;
2443 QSIMPLEQ_ENTRY(BlkIntermediateStates) entry;
2444 } BlkIntermediateStates;
2448 * Drops images above 'base' up to and including 'top', and sets the image
2449 * above 'top' to have base as its backing file.
2451 * Requires that the overlay to 'top' is opened r/w, so that the backing file
2452 * information in 'bs' can be properly updated.
2454 * E.g., this will convert the following chain:
2455 * bottom <- base <- intermediate <- top <- active
2459 * bottom <- base <- active
2461 * It is allowed for bottom==base, in which case it converts:
2463 * base <- intermediate <- top <- active
2470 * if active == top, that is considered an error
2473 int bdrv_drop_intermediate(BlockDriverState *active, BlockDriverState *top,
2474 BlockDriverState *base)
2476 BlockDriverState *intermediate;
2477 BlockDriverState *base_bs = NULL;
2478 BlockDriverState *new_top_bs = NULL;
2479 BlkIntermediateStates *intermediate_state, *next;
2482 QSIMPLEQ_HEAD(states_to_delete, BlkIntermediateStates) states_to_delete;
2483 QSIMPLEQ_INIT(&states_to_delete);
2485 if (!top->drv || !base->drv) {
2489 new_top_bs = bdrv_find_overlay(active, top);
2491 if (new_top_bs == NULL) {
2492 /* we could not find the image above 'top', this is an error */
2496 /* special case of new_top_bs->backing_hd already pointing to base - nothing
2497 * to do, no intermediate images */
2498 if (new_top_bs->backing_hd == base) {
2505 /* now we will go down through the list, and add each BDS we find
2506 * into our deletion queue, until we hit the 'base'
2508 while (intermediate) {
2509 intermediate_state = g_malloc0(sizeof(BlkIntermediateStates));
2510 intermediate_state->bs = intermediate;
2511 QSIMPLEQ_INSERT_TAIL(&states_to_delete, intermediate_state, entry);
2513 if (intermediate->backing_hd == base) {
2514 base_bs = intermediate->backing_hd;
2517 intermediate = intermediate->backing_hd;
2519 if (base_bs == NULL) {
2520 /* something went wrong, we did not end at the base. safely
2521 * unravel everything, and exit with error */
2525 /* success - we can delete the intermediate states, and link top->base */
2526 ret = bdrv_change_backing_file(new_top_bs, base_bs->filename,
2527 base_bs->drv ? base_bs->drv->format_name : "");
2531 new_top_bs->backing_hd = base_bs;
2533 bdrv_refresh_limits(new_top_bs);
2535 QSIMPLEQ_FOREACH_SAFE(intermediate_state, &states_to_delete, entry, next) {
2536 /* so that bdrv_close() does not recursively close the chain */
2537 intermediate_state->bs->backing_hd = NULL;
2538 bdrv_unref(intermediate_state->bs);
2543 QSIMPLEQ_FOREACH_SAFE(intermediate_state, &states_to_delete, entry, next) {
2544 g_free(intermediate_state);
2550 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
2555 if (!bdrv_is_inserted(bs))
2561 len = bdrv_getlength(bs);
2566 if ((offset > len) || (len - offset < size))
2572 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
2575 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
2576 nb_sectors * BDRV_SECTOR_SIZE);
2579 typedef struct RwCo {
2580 BlockDriverState *bs;
2585 BdrvRequestFlags flags;
2588 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
2590 RwCo *rwco = opaque;
2592 if (!rwco->is_write) {
2593 rwco->ret = bdrv_co_do_preadv(rwco->bs, rwco->offset,
2594 rwco->qiov->size, rwco->qiov,
2597 rwco->ret = bdrv_co_do_pwritev(rwco->bs, rwco->offset,
2598 rwco->qiov->size, rwco->qiov,
2604 * Process a vectored synchronous request using coroutines
2606 static int bdrv_prwv_co(BlockDriverState *bs, int64_t offset,
2607 QEMUIOVector *qiov, bool is_write,
2608 BdrvRequestFlags flags)
2615 .is_write = is_write,
2621 * In sync call context, when the vcpu is blocked, this throttling timer
2622 * will not fire; so the I/O throttling function has to be disabled here
2623 * if it has been enabled.
2625 if (bs->io_limits_enabled) {
2626 fprintf(stderr, "Disabling I/O throttling on '%s' due "
2627 "to synchronous I/O.\n", bdrv_get_device_name(bs));
2628 bdrv_io_limits_disable(bs);
2631 if (qemu_in_coroutine()) {
2632 /* Fast-path if already in coroutine context */
2633 bdrv_rw_co_entry(&rwco);
2635 co = qemu_coroutine_create(bdrv_rw_co_entry);
2636 qemu_coroutine_enter(co, &rwco);
2637 while (rwco.ret == NOT_DONE) {
2645 * Process a synchronous request using coroutines
2647 static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf,
2648 int nb_sectors, bool is_write, BdrvRequestFlags flags)
2651 struct iovec iov = {
2652 .iov_base = (void *)buf,
2653 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
2656 qemu_iovec_init_external(&qiov, &iov, 1);
2657 return bdrv_prwv_co(bs, sector_num << BDRV_SECTOR_BITS,
2658 &qiov, is_write, flags);
2661 /* return < 0 if error. See bdrv_write() for the return codes */
2662 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
2663 uint8_t *buf, int nb_sectors)
2665 return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false, 0);
2668 /* Just like bdrv_read(), but with I/O throttling temporarily disabled */
2669 int bdrv_read_unthrottled(BlockDriverState *bs, int64_t sector_num,
2670 uint8_t *buf, int nb_sectors)
2675 enabled = bs->io_limits_enabled;
2676 bs->io_limits_enabled = false;
2677 ret = bdrv_read(bs, sector_num, buf, nb_sectors);
2678 bs->io_limits_enabled = enabled;
2682 /* Return < 0 if error. Important errors are:
2683 -EIO generic I/O error (may happen for all errors)
2684 -ENOMEDIUM No media inserted.
2685 -EINVAL Invalid sector number or nb_sectors
2686 -EACCES Trying to write a read-only device
2688 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
2689 const uint8_t *buf, int nb_sectors)
2691 return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true, 0);
2694 int bdrv_write_zeroes(BlockDriverState *bs, int64_t sector_num,
2695 int nb_sectors, BdrvRequestFlags flags)
2697 return bdrv_rw_co(bs, sector_num, NULL, nb_sectors, true,
2698 BDRV_REQ_ZERO_WRITE | flags);
2702 * Completely zero out a block device with the help of bdrv_write_zeroes.
2703 * The operation is sped up by checking the block status and only writing
2704 * zeroes to the device if they currently do not return zeroes. Optional
2705 * flags are passed through to bdrv_write_zeroes (e.g. BDRV_REQ_MAY_UNMAP).
2707 * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
2709 int bdrv_make_zero(BlockDriverState *bs, BdrvRequestFlags flags)
2711 int64_t target_size = bdrv_getlength(bs) / BDRV_SECTOR_SIZE;
2712 int64_t ret, nb_sectors, sector_num = 0;
2716 nb_sectors = target_size - sector_num;
2717 if (nb_sectors <= 0) {
2720 if (nb_sectors > INT_MAX) {
2721 nb_sectors = INT_MAX;
2723 ret = bdrv_get_block_status(bs, sector_num, nb_sectors, &n);
2725 error_report("error getting block status at sector %" PRId64 ": %s",
2726 sector_num, strerror(-ret));
2729 if (ret & BDRV_BLOCK_ZERO) {
2733 ret = bdrv_write_zeroes(bs, sector_num, n, flags);
2735 error_report("error writing zeroes at sector %" PRId64 ": %s",
2736 sector_num, strerror(-ret));
2743 int bdrv_pread(BlockDriverState *bs, int64_t offset, void *buf, int bytes)
2746 struct iovec iov = {
2747 .iov_base = (void *)buf,
2756 qemu_iovec_init_external(&qiov, &iov, 1);
2757 ret = bdrv_prwv_co(bs, offset, &qiov, false, 0);
2765 int bdrv_pwritev(BlockDriverState *bs, int64_t offset, QEMUIOVector *qiov)
2769 ret = bdrv_prwv_co(bs, offset, qiov, true, 0);
2777 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
2778 const void *buf, int bytes)
2781 struct iovec iov = {
2782 .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 * Forwards an already correctly aligned request to the BlockDriver. This
2886 * handles copy on read and zeroing after EOF; any other features must be
2887 * implemented by the caller.
2889 static int coroutine_fn bdrv_aligned_preadv(BlockDriverState *bs,
2890 BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
2891 int64_t align, QEMUIOVector *qiov, int flags)
2893 BlockDriver *drv = bs->drv;
2896 int64_t sector_num = offset >> BDRV_SECTOR_BITS;
2897 unsigned int nb_sectors = bytes >> BDRV_SECTOR_BITS;
2899 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
2900 assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
2902 /* Handle Copy on Read and associated serialisation */
2903 if (flags & BDRV_REQ_COPY_ON_READ) {
2904 /* If we touch the same cluster it counts as an overlap. This
2905 * guarantees that allocating writes will be serialized and not race
2906 * with each other for the same cluster. For example, in copy-on-read
2907 * it ensures that the CoR read and write operations are atomic and
2908 * guest writes cannot interleave between them. */
2909 mark_request_serialising(req, bdrv_get_cluster_size(bs));
2912 wait_serialising_requests(req);
2914 if (flags & BDRV_REQ_COPY_ON_READ) {
2917 ret = bdrv_is_allocated(bs, sector_num, nb_sectors, &pnum);
2922 if (!ret || pnum != nb_sectors) {
2923 ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov);
2928 /* Forward the request to the BlockDriver */
2929 if (!(bs->zero_beyond_eof && bs->growable)) {
2930 ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
2932 /* Read zeros after EOF of growable BDSes */
2933 int64_t len, total_sectors, max_nb_sectors;
2935 len = bdrv_getlength(bs);
2941 total_sectors = DIV_ROUND_UP(len, BDRV_SECTOR_SIZE);
2942 max_nb_sectors = ROUND_UP(MAX(0, total_sectors - sector_num),
2943 align >> BDRV_SECTOR_BITS);
2944 if (max_nb_sectors > 0) {
2945 ret = drv->bdrv_co_readv(bs, sector_num,
2946 MIN(nb_sectors, max_nb_sectors), qiov);
2951 /* Reading beyond end of file is supposed to produce zeroes */
2952 if (ret == 0 && total_sectors < sector_num + nb_sectors) {
2953 uint64_t offset = MAX(0, total_sectors - sector_num);
2954 uint64_t bytes = (sector_num + nb_sectors - offset) *
2956 qemu_iovec_memset(qiov, offset * BDRV_SECTOR_SIZE, 0, bytes);
2965 * Handle a read request in coroutine context
2967 static int coroutine_fn bdrv_co_do_preadv(BlockDriverState *bs,
2968 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
2969 BdrvRequestFlags flags)
2971 BlockDriver *drv = bs->drv;
2972 BdrvTrackedRequest req;
2974 /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
2975 uint64_t align = MAX(BDRV_SECTOR_SIZE, bs->request_alignment);
2976 uint8_t *head_buf = NULL;
2977 uint8_t *tail_buf = NULL;
2978 QEMUIOVector local_qiov;
2979 bool use_local_qiov = false;
2985 if (bdrv_check_byte_request(bs, offset, bytes)) {
2989 if (bs->copy_on_read) {
2990 flags |= BDRV_REQ_COPY_ON_READ;
2993 /* throttling disk I/O */
2994 if (bs->io_limits_enabled) {
2995 bdrv_io_limits_intercept(bs, bytes, false);
2998 /* Align read if necessary by padding qiov */
2999 if (offset & (align - 1)) {
3000 head_buf = qemu_blockalign(bs, align);
3001 qemu_iovec_init(&local_qiov, qiov->niov + 2);
3002 qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
3003 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
3004 use_local_qiov = true;
3006 bytes += offset & (align - 1);
3007 offset = offset & ~(align - 1);
3010 if ((offset + bytes) & (align - 1)) {
3011 if (!use_local_qiov) {
3012 qemu_iovec_init(&local_qiov, qiov->niov + 1);
3013 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
3014 use_local_qiov = true;
3016 tail_buf = qemu_blockalign(bs, align);
3017 qemu_iovec_add(&local_qiov, tail_buf,
3018 align - ((offset + bytes) & (align - 1)));
3020 bytes = ROUND_UP(bytes, align);
3023 tracked_request_begin(&req, bs, offset, bytes, false);
3024 ret = bdrv_aligned_preadv(bs, &req, offset, bytes, align,
3025 use_local_qiov ? &local_qiov : qiov,
3027 tracked_request_end(&req);
3029 if (use_local_qiov) {
3030 qemu_iovec_destroy(&local_qiov);
3031 qemu_vfree(head_buf);
3032 qemu_vfree(tail_buf);
3038 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
3039 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
3040 BdrvRequestFlags flags)
3042 if (nb_sectors < 0 || nb_sectors > (UINT_MAX >> BDRV_SECTOR_BITS)) {
3046 return bdrv_co_do_preadv(bs, sector_num << BDRV_SECTOR_BITS,
3047 nb_sectors << BDRV_SECTOR_BITS, qiov, flags);
3050 int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
3051 int nb_sectors, QEMUIOVector *qiov)
3053 trace_bdrv_co_readv(bs, sector_num, nb_sectors);
3055 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0);
3058 int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
3059 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
3061 trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors);
3063 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
3064 BDRV_REQ_COPY_ON_READ);
3067 /* if no limit is specified in the BlockLimits use a default
3068 * of 32768 512-byte sectors (16 MiB) per request.
3070 #define MAX_WRITE_ZEROES_DEFAULT 32768
3072 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
3073 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
3075 BlockDriver *drv = bs->drv;
3077 struct iovec iov = {0};
3080 int max_write_zeroes = bs->bl.max_write_zeroes ?
3081 bs->bl.max_write_zeroes : MAX_WRITE_ZEROES_DEFAULT;
3083 while (nb_sectors > 0 && !ret) {
3084 int num = nb_sectors;
3086 /* Align request. Block drivers can expect the "bulk" of the request
3089 if (bs->bl.write_zeroes_alignment
3090 && num > bs->bl.write_zeroes_alignment) {
3091 if (sector_num % bs->bl.write_zeroes_alignment != 0) {
3092 /* Make a small request up to the first aligned sector. */
3093 num = bs->bl.write_zeroes_alignment;
3094 num -= sector_num % bs->bl.write_zeroes_alignment;
3095 } else if ((sector_num + num) % bs->bl.write_zeroes_alignment != 0) {
3096 /* Shorten the request to the last aligned sector. num cannot
3097 * underflow because num > bs->bl.write_zeroes_alignment.
3099 num -= (sector_num + num) % bs->bl.write_zeroes_alignment;
3103 /* limit request size */
3104 if (num > max_write_zeroes) {
3105 num = max_write_zeroes;
3109 /* First try the efficient write zeroes operation */
3110 if (drv->bdrv_co_write_zeroes) {
3111 ret = drv->bdrv_co_write_zeroes(bs, sector_num, num, flags);
3114 if (ret == -ENOTSUP) {
3115 /* Fall back to bounce buffer if write zeroes is unsupported */
3116 iov.iov_len = num * BDRV_SECTOR_SIZE;
3117 if (iov.iov_base == NULL) {
3118 iov.iov_base = qemu_blockalign(bs, num * BDRV_SECTOR_SIZE);
3119 memset(iov.iov_base, 0, num * BDRV_SECTOR_SIZE);
3121 qemu_iovec_init_external(&qiov, &iov, 1);
3123 ret = drv->bdrv_co_writev(bs, sector_num, num, &qiov);
3125 /* Keep bounce buffer around if it is big enough for all
3126 * all future requests.
3128 if (num < max_write_zeroes) {
3129 qemu_vfree(iov.iov_base);
3130 iov.iov_base = NULL;
3138 qemu_vfree(iov.iov_base);
3143 * Forwards an already correctly aligned write request to the BlockDriver.
3145 static int coroutine_fn bdrv_aligned_pwritev(BlockDriverState *bs,
3146 BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
3147 QEMUIOVector *qiov, int flags)
3149 BlockDriver *drv = bs->drv;
3153 int64_t sector_num = offset >> BDRV_SECTOR_BITS;
3154 unsigned int nb_sectors = bytes >> BDRV_SECTOR_BITS;
3156 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
3157 assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
3159 waited = wait_serialising_requests(req);
3160 assert(!waited || !req->serialising);
3161 assert(req->overlap_offset <= offset);
3162 assert(offset + bytes <= req->overlap_offset + req->overlap_bytes);
3164 ret = notifier_with_return_list_notify(&bs->before_write_notifiers, req);
3167 /* Do nothing, write notifier decided to fail this request */
3168 } else if (flags & BDRV_REQ_ZERO_WRITE) {
3169 BLKDBG_EVENT(bs, BLKDBG_PWRITEV_ZERO);
3170 ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors, flags);
3172 BLKDBG_EVENT(bs, BLKDBG_PWRITEV);
3173 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
3175 BLKDBG_EVENT(bs, BLKDBG_PWRITEV_DONE);
3177 if (ret == 0 && !bs->enable_write_cache) {
3178 ret = bdrv_co_flush(bs);
3181 bdrv_set_dirty(bs, sector_num, nb_sectors);
3183 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
3184 bs->wr_highest_sector = sector_num + nb_sectors - 1;
3186 if (bs->growable && ret >= 0) {
3187 bs->total_sectors = MAX(bs->total_sectors, sector_num + nb_sectors);
3194 * Handle a write request in coroutine context
3196 static int coroutine_fn bdrv_co_do_pwritev(BlockDriverState *bs,
3197 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
3198 BdrvRequestFlags flags)
3200 BdrvTrackedRequest req;
3201 /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
3202 uint64_t align = MAX(BDRV_SECTOR_SIZE, bs->request_alignment);
3203 uint8_t *head_buf = NULL;
3204 uint8_t *tail_buf = NULL;
3205 QEMUIOVector local_qiov;
3206 bool use_local_qiov = false;
3212 if (bs->read_only) {
3215 if (bdrv_check_byte_request(bs, offset, bytes)) {
3219 /* throttling disk I/O */
3220 if (bs->io_limits_enabled) {
3221 bdrv_io_limits_intercept(bs, bytes, true);
3225 * Align write if necessary by performing a read-modify-write cycle.
3226 * Pad qiov with the read parts and be sure to have a tracked request not
3227 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
3229 tracked_request_begin(&req, bs, offset, bytes, true);
3231 if (offset & (align - 1)) {
3232 QEMUIOVector head_qiov;
3233 struct iovec head_iov;
3235 mark_request_serialising(&req, align);
3236 wait_serialising_requests(&req);
3238 head_buf = qemu_blockalign(bs, align);
3239 head_iov = (struct iovec) {
3240 .iov_base = head_buf,
3243 qemu_iovec_init_external(&head_qiov, &head_iov, 1);
3245 BLKDBG_EVENT(bs, BLKDBG_PWRITEV_RMW_HEAD);
3246 ret = bdrv_aligned_preadv(bs, &req, offset & ~(align - 1), align,
3247 align, &head_qiov, 0);
3251 BLKDBG_EVENT(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD);
3253 qemu_iovec_init(&local_qiov, qiov->niov + 2);
3254 qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
3255 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
3256 use_local_qiov = true;
3258 bytes += offset & (align - 1);
3259 offset = offset & ~(align - 1);
3262 if ((offset + bytes) & (align - 1)) {
3263 QEMUIOVector tail_qiov;
3264 struct iovec tail_iov;
3268 mark_request_serialising(&req, align);
3269 waited = wait_serialising_requests(&req);
3270 assert(!waited || !use_local_qiov);
3272 tail_buf = qemu_blockalign(bs, align);
3273 tail_iov = (struct iovec) {
3274 .iov_base = tail_buf,
3277 qemu_iovec_init_external(&tail_qiov, &tail_iov, 1);
3279 BLKDBG_EVENT(bs, BLKDBG_PWRITEV_RMW_TAIL);
3280 ret = bdrv_aligned_preadv(bs, &req, (offset + bytes) & ~(align - 1), align,
3281 align, &tail_qiov, 0);
3285 BLKDBG_EVENT(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL);
3287 if (!use_local_qiov) {
3288 qemu_iovec_init(&local_qiov, qiov->niov + 1);
3289 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
3290 use_local_qiov = true;
3293 tail_bytes = (offset + bytes) & (align - 1);
3294 qemu_iovec_add(&local_qiov, tail_buf + tail_bytes, align - tail_bytes);
3296 bytes = ROUND_UP(bytes, align);
3299 ret = bdrv_aligned_pwritev(bs, &req, offset, bytes,
3300 use_local_qiov ? &local_qiov : qiov,
3304 tracked_request_end(&req);
3306 if (use_local_qiov) {
3307 qemu_iovec_destroy(&local_qiov);
3309 qemu_vfree(head_buf);
3310 qemu_vfree(tail_buf);
3315 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
3316 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
3317 BdrvRequestFlags flags)
3319 if (nb_sectors < 0 || nb_sectors > (INT_MAX >> BDRV_SECTOR_BITS)) {
3323 return bdrv_co_do_pwritev(bs, sector_num << BDRV_SECTOR_BITS,
3324 nb_sectors << BDRV_SECTOR_BITS, qiov, flags);
3327 int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
3328 int nb_sectors, QEMUIOVector *qiov)
3330 trace_bdrv_co_writev(bs, sector_num, nb_sectors);
3332 return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0);
3335 int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs,
3336 int64_t sector_num, int nb_sectors,
3337 BdrvRequestFlags flags)
3339 trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors, flags);
3341 if (!(bs->open_flags & BDRV_O_UNMAP)) {
3342 flags &= ~BDRV_REQ_MAY_UNMAP;
3345 return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL,
3346 BDRV_REQ_ZERO_WRITE | flags);
3350 * Truncate file to 'offset' bytes (needed only for file protocols)
3352 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
3354 BlockDriver *drv = bs->drv;
3358 if (!drv->bdrv_truncate)
3362 if (bdrv_in_use(bs))
3364 ret = drv->bdrv_truncate(bs, offset);
3366 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
3367 bdrv_dev_resize_cb(bs);
3373 * Length of a allocated file in bytes. Sparse files are counted by actual
3374 * allocated space. Return < 0 if error or unknown.
3376 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
3378 BlockDriver *drv = bs->drv;
3382 if (drv->bdrv_get_allocated_file_size) {
3383 return drv->bdrv_get_allocated_file_size(bs);
3386 return bdrv_get_allocated_file_size(bs->file);
3392 * Length of a file in bytes. Return < 0 if error or unknown.
3394 int64_t bdrv_getlength(BlockDriverState *bs)
3396 BlockDriver *drv = bs->drv;
3400 if (drv->has_variable_length) {
3401 int ret = refresh_total_sectors(bs, bs->total_sectors);
3406 return bs->total_sectors * BDRV_SECTOR_SIZE;
3409 /* return 0 as number of sectors if no device present or error */
3410 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
3413 length = bdrv_getlength(bs);
3417 length = length >> BDRV_SECTOR_BITS;
3418 *nb_sectors_ptr = length;
3421 void bdrv_set_on_error(BlockDriverState *bs, BlockdevOnError on_read_error,
3422 BlockdevOnError on_write_error)
3424 bs->on_read_error = on_read_error;
3425 bs->on_write_error = on_write_error;
3428 BlockdevOnError bdrv_get_on_error(BlockDriverState *bs, bool is_read)
3430 return is_read ? bs->on_read_error : bs->on_write_error;
3433 BlockErrorAction bdrv_get_error_action(BlockDriverState *bs, bool is_read, int error)
3435 BlockdevOnError on_err = is_read ? bs->on_read_error : bs->on_write_error;
3438 case BLOCKDEV_ON_ERROR_ENOSPC:
3439 return (error == ENOSPC) ? BDRV_ACTION_STOP : BDRV_ACTION_REPORT;
3440 case BLOCKDEV_ON_ERROR_STOP:
3441 return BDRV_ACTION_STOP;
3442 case BLOCKDEV_ON_ERROR_REPORT:
3443 return BDRV_ACTION_REPORT;
3444 case BLOCKDEV_ON_ERROR_IGNORE:
3445 return BDRV_ACTION_IGNORE;
3451 /* This is done by device models because, while the block layer knows
3452 * about the error, it does not know whether an operation comes from
3453 * the device or the block layer (from a job, for example).
3455 void bdrv_error_action(BlockDriverState *bs, BlockErrorAction action,
3456 bool is_read, int error)
3459 bdrv_emit_qmp_error_event(bs, QEVENT_BLOCK_IO_ERROR, action, is_read);
3460 if (action == BDRV_ACTION_STOP) {
3461 vm_stop(RUN_STATE_IO_ERROR);
3462 bdrv_iostatus_set_err(bs, error);
3466 int bdrv_is_read_only(BlockDriverState *bs)
3468 return bs->read_only;
3471 int bdrv_is_sg(BlockDriverState *bs)
3476 int bdrv_enable_write_cache(BlockDriverState *bs)
3478 return bs->enable_write_cache;
3481 void bdrv_set_enable_write_cache(BlockDriverState *bs, bool wce)
3483 bs->enable_write_cache = wce;
3485 /* so a reopen() will preserve wce */
3487 bs->open_flags |= BDRV_O_CACHE_WB;
3489 bs->open_flags &= ~BDRV_O_CACHE_WB;
3493 int bdrv_is_encrypted(BlockDriverState *bs)
3495 if (bs->backing_hd && bs->backing_hd->encrypted)
3497 return bs->encrypted;
3500 int bdrv_key_required(BlockDriverState *bs)
3502 BlockDriverState *backing_hd = bs->backing_hd;
3504 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
3506 return (bs->encrypted && !bs->valid_key);
3509 int bdrv_set_key(BlockDriverState *bs, const char *key)
3512 if (bs->backing_hd && bs->backing_hd->encrypted) {
3513 ret = bdrv_set_key(bs->backing_hd, key);
3519 if (!bs->encrypted) {
3521 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
3524 ret = bs->drv->bdrv_set_key(bs, key);
3527 } else if (!bs->valid_key) {
3529 /* call the change callback now, we skipped it on open */
3530 bdrv_dev_change_media_cb(bs, true);
3535 const char *bdrv_get_format_name(BlockDriverState *bs)
3537 return bs->drv ? bs->drv->format_name : NULL;
3540 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
3545 QLIST_FOREACH(drv, &bdrv_drivers, list) {
3546 it(opaque, drv->format_name);
3550 /* This function is to find block backend bs */
3551 BlockDriverState *bdrv_find(const char *name)
3553 BlockDriverState *bs;
3555 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
3556 if (!strcmp(name, bs->device_name)) {
3563 /* This function is to find a node in the bs graph */
3564 BlockDriverState *bdrv_find_node(const char *node_name)
3566 BlockDriverState *bs;
3570 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3571 if (!strcmp(node_name, bs->node_name)) {
3578 /* Put this QMP function here so it can access the static graph_bdrv_states. */
3579 BlockDeviceInfoList *bdrv_named_nodes_list(void)
3581 BlockDeviceInfoList *list, *entry;
3582 BlockDriverState *bs;
3585 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3586 entry = g_malloc0(sizeof(*entry));
3587 entry->value = bdrv_block_device_info(bs);
3595 BlockDriverState *bdrv_lookup_bs(const char *device,
3596 const char *node_name,
3599 BlockDriverState *bs = NULL;
3602 bs = bdrv_find(device);
3610 bs = bdrv_find_node(node_name);
3617 error_setg(errp, "Cannot find device=%s nor node_name=%s",
3618 device ? device : "",
3619 node_name ? node_name : "");
3623 BlockDriverState *bdrv_next(BlockDriverState *bs)
3626 return QTAILQ_FIRST(&bdrv_states);
3628 return QTAILQ_NEXT(bs, device_list);
3631 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
3633 BlockDriverState *bs;
3635 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
3640 const char *bdrv_get_device_name(BlockDriverState *bs)
3642 return bs->device_name;
3645 int bdrv_get_flags(BlockDriverState *bs)
3647 return bs->open_flags;
3650 int bdrv_flush_all(void)
3652 BlockDriverState *bs;
3655 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
3656 int ret = bdrv_flush(bs);
3657 if (ret < 0 && !result) {
3665 int bdrv_has_zero_init_1(BlockDriverState *bs)
3670 int bdrv_has_zero_init(BlockDriverState *bs)
3674 /* If BS is a copy on write image, it is initialized to
3675 the contents of the base image, which may not be zeroes. */
3676 if (bs->backing_hd) {
3679 if (bs->drv->bdrv_has_zero_init) {
3680 return bs->drv->bdrv_has_zero_init(bs);
3687 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
3689 BlockDriverInfo bdi;
3691 if (bs->backing_hd) {
3695 if (bdrv_get_info(bs, &bdi) == 0) {
3696 return bdi.unallocated_blocks_are_zero;
3702 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
3704 BlockDriverInfo bdi;
3706 if (bs->backing_hd || !(bs->open_flags & BDRV_O_UNMAP)) {
3710 if (bdrv_get_info(bs, &bdi) == 0) {
3711 return bdi.can_write_zeroes_with_unmap;
3717 typedef struct BdrvCoGetBlockStatusData {
3718 BlockDriverState *bs;
3719 BlockDriverState *base;
3725 } BdrvCoGetBlockStatusData;
3728 * Returns true iff the specified sector is present in the disk image. Drivers
3729 * not implementing the functionality are assumed to not support backing files,
3730 * hence all their sectors are reported as allocated.
3732 * If 'sector_num' is beyond the end of the disk image the return value is 0
3733 * and 'pnum' is set to 0.
3735 * 'pnum' is set to the number of sectors (including and immediately following
3736 * the specified sector) that are known to be in the same
3737 * allocated/unallocated state.
3739 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
3740 * beyond the end of the disk image it will be clamped.
3742 static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs,
3744 int nb_sectors, int *pnum)
3750 length = bdrv_getlength(bs);
3755 if (sector_num >= (length >> BDRV_SECTOR_BITS)) {
3760 n = bs->total_sectors - sector_num;
3761 if (n < nb_sectors) {
3765 if (!bs->drv->bdrv_co_get_block_status) {
3767 ret = BDRV_BLOCK_DATA;
3768 if (bs->drv->protocol_name) {
3769 ret |= BDRV_BLOCK_OFFSET_VALID | (sector_num * BDRV_SECTOR_SIZE);
3774 ret = bs->drv->bdrv_co_get_block_status(bs, sector_num, nb_sectors, pnum);
3780 if (ret & BDRV_BLOCK_RAW) {
3781 assert(ret & BDRV_BLOCK_OFFSET_VALID);
3782 return bdrv_get_block_status(bs->file, ret >> BDRV_SECTOR_BITS,
3786 if (!(ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO)) {
3787 if (bdrv_unallocated_blocks_are_zero(bs)) {
3788 ret |= BDRV_BLOCK_ZERO;
3789 } else if (bs->backing_hd) {
3790 BlockDriverState *bs2 = bs->backing_hd;
3791 int64_t length2 = bdrv_getlength(bs2);
3792 if (length2 >= 0 && sector_num >= (length2 >> BDRV_SECTOR_BITS)) {
3793 ret |= BDRV_BLOCK_ZERO;
3799 (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) &&
3800 (ret & BDRV_BLOCK_OFFSET_VALID)) {
3801 ret2 = bdrv_co_get_block_status(bs->file, ret >> BDRV_SECTOR_BITS,
3804 /* Ignore errors. This is just providing extra information, it
3805 * is useful but not necessary.
3807 ret |= (ret2 & BDRV_BLOCK_ZERO);
3814 /* Coroutine wrapper for bdrv_get_block_status() */
3815 static void coroutine_fn bdrv_get_block_status_co_entry(void *opaque)
3817 BdrvCoGetBlockStatusData *data = opaque;
3818 BlockDriverState *bs = data->bs;
3820 data->ret = bdrv_co_get_block_status(bs, data->sector_num, data->nb_sectors,
3826 * Synchronous wrapper around bdrv_co_get_block_status().
3828 * See bdrv_co_get_block_status() for details.
3830 int64_t bdrv_get_block_status(BlockDriverState *bs, int64_t sector_num,
3831 int nb_sectors, int *pnum)
3834 BdrvCoGetBlockStatusData data = {
3836 .sector_num = sector_num,
3837 .nb_sectors = nb_sectors,
3842 if (qemu_in_coroutine()) {
3843 /* Fast-path if already in coroutine context */
3844 bdrv_get_block_status_co_entry(&data);
3846 co = qemu_coroutine_create(bdrv_get_block_status_co_entry);
3847 qemu_coroutine_enter(co, &data);
3848 while (!data.done) {
3855 int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num,
3856 int nb_sectors, int *pnum)
3858 int64_t ret = bdrv_get_block_status(bs, sector_num, nb_sectors, pnum);
3863 (ret & BDRV_BLOCK_DATA) ||
3864 ((ret & BDRV_BLOCK_ZERO) && !bdrv_has_zero_init(bs));
3868 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
3870 * Return true if the given sector is allocated in any image between
3871 * BASE and TOP (inclusive). BASE can be NULL to check if the given
3872 * sector is allocated in any image of the chain. Return false otherwise.
3874 * 'pnum' is set to the number of sectors (including and immediately following
3875 * the specified sector) that are known to be in the same
3876 * allocated/unallocated state.
3879 int bdrv_is_allocated_above(BlockDriverState *top,
3880 BlockDriverState *base,
3882 int nb_sectors, int *pnum)
3884 BlockDriverState *intermediate;
3885 int ret, n = nb_sectors;
3888 while (intermediate && intermediate != base) {
3890 ret = bdrv_is_allocated(intermediate, sector_num, nb_sectors,
3900 * [sector_num, nb_sectors] is unallocated on top but intermediate
3903 * [sector_num+x, nr_sectors] allocated.
3905 if (n > pnum_inter &&
3906 (intermediate == top ||
3907 sector_num + pnum_inter < intermediate->total_sectors)) {
3911 intermediate = intermediate->backing_hd;
3918 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
3920 if (bs->backing_hd && bs->backing_hd->encrypted)
3921 return bs->backing_file;
3922 else if (bs->encrypted)
3923 return bs->filename;
3928 void bdrv_get_backing_filename(BlockDriverState *bs,
3929 char *filename, int filename_size)
3931 pstrcpy(filename, filename_size, bs->backing_file);
3934 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
3935 const uint8_t *buf, int nb_sectors)
3937 BlockDriver *drv = bs->drv;
3940 if (!drv->bdrv_write_compressed)
3942 if (bdrv_check_request(bs, sector_num, nb_sectors))
3945 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
3947 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
3950 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
3952 BlockDriver *drv = bs->drv;
3955 if (!drv->bdrv_get_info)
3957 memset(bdi, 0, sizeof(*bdi));
3958 return drv->bdrv_get_info(bs, bdi);
3961 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs)
3963 BlockDriver *drv = bs->drv;
3964 if (drv && drv->bdrv_get_specific_info) {
3965 return drv->bdrv_get_specific_info(bs);
3970 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
3971 int64_t pos, int size)
3974 struct iovec iov = {
3975 .iov_base = (void *) buf,
3979 qemu_iovec_init_external(&qiov, &iov, 1);
3980 return bdrv_writev_vmstate(bs, &qiov, pos);
3983 int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
3985 BlockDriver *drv = bs->drv;
3989 } else if (drv->bdrv_save_vmstate) {
3990 return drv->bdrv_save_vmstate(bs, qiov, pos);
3991 } else if (bs->file) {
3992 return bdrv_writev_vmstate(bs->file, qiov, pos);
3998 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
3999 int64_t pos, int size)
4001 BlockDriver *drv = bs->drv;
4004 if (drv->bdrv_load_vmstate)
4005 return drv->bdrv_load_vmstate(bs, buf, pos, size);
4007 return bdrv_load_vmstate(bs->file, buf, pos, size);
4011 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
4013 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
4017 bs->drv->bdrv_debug_event(bs, event);
4020 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
4023 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
4027 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
4028 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
4034 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
4036 while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
4040 if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
4041 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
4047 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
4049 while (bs && bs->drv && !bs->drv->bdrv_debug_resume) {
4053 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
4054 return bs->drv->bdrv_debug_resume(bs, tag);
4060 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
4062 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
4066 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
4067 return bs->drv->bdrv_debug_is_suspended(bs, tag);
4073 int bdrv_is_snapshot(BlockDriverState *bs)
4075 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
4078 /* backing_file can either be relative, or absolute, or a protocol. If it is
4079 * relative, it must be relative to the chain. So, passing in bs->filename
4080 * from a BDS as backing_file should not be done, as that may be relative to
4081 * the CWD rather than the chain. */
4082 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
4083 const char *backing_file)
4085 char *filename_full = NULL;
4086 char *backing_file_full = NULL;
4087 char *filename_tmp = NULL;
4088 int is_protocol = 0;
4089 BlockDriverState *curr_bs = NULL;
4090 BlockDriverState *retval = NULL;
4092 if (!bs || !bs->drv || !backing_file) {
4096 filename_full = g_malloc(PATH_MAX);
4097 backing_file_full = g_malloc(PATH_MAX);
4098 filename_tmp = g_malloc(PATH_MAX);
4100 is_protocol = path_has_protocol(backing_file);
4102 for (curr_bs = bs; curr_bs->backing_hd; curr_bs = curr_bs->backing_hd) {
4104 /* If either of the filename paths is actually a protocol, then
4105 * compare unmodified paths; otherwise make paths relative */
4106 if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
4107 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
4108 retval = curr_bs->backing_hd;
4112 /* If not an absolute filename path, make it relative to the current
4113 * image's filename path */
4114 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4117 /* We are going to compare absolute pathnames */
4118 if (!realpath(filename_tmp, filename_full)) {
4122 /* We need to make sure the backing filename we are comparing against
4123 * is relative to the current image filename (or absolute) */
4124 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4125 curr_bs->backing_file);
4127 if (!realpath(filename_tmp, backing_file_full)) {
4131 if (strcmp(backing_file_full, filename_full) == 0) {
4132 retval = curr_bs->backing_hd;
4138 g_free(filename_full);
4139 g_free(backing_file_full);
4140 g_free(filename_tmp);
4144 int bdrv_get_backing_file_depth(BlockDriverState *bs)
4150 if (!bs->backing_hd) {
4154 return 1 + bdrv_get_backing_file_depth(bs->backing_hd);
4157 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
4159 BlockDriverState *curr_bs = NULL;
4167 while (curr_bs->backing_hd) {
4168 curr_bs = curr_bs->backing_hd;
4173 /**************************************************************/
4176 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
4177 QEMUIOVector *qiov, int nb_sectors,
4178 BlockDriverCompletionFunc *cb, void *opaque)
4180 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
4182 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0,
4186 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
4187 QEMUIOVector *qiov, int nb_sectors,
4188 BlockDriverCompletionFunc *cb, void *opaque)
4190 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
4192 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0,
4196 BlockDriverAIOCB *bdrv_aio_write_zeroes(BlockDriverState *bs,
4197 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags,
4198 BlockDriverCompletionFunc *cb, void *opaque)
4200 trace_bdrv_aio_write_zeroes(bs, sector_num, nb_sectors, flags, opaque);
4202 return bdrv_co_aio_rw_vector(bs, sector_num, NULL, nb_sectors,
4203 BDRV_REQ_ZERO_WRITE | flags,
4208 typedef struct MultiwriteCB {
4213 BlockDriverCompletionFunc *cb;
4215 QEMUIOVector *free_qiov;
4219 static void multiwrite_user_cb(MultiwriteCB *mcb)
4223 for (i = 0; i < mcb->num_callbacks; i++) {
4224 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
4225 if (mcb->callbacks[i].free_qiov) {
4226 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
4228 g_free(mcb->callbacks[i].free_qiov);
4232 static void multiwrite_cb(void *opaque, int ret)
4234 MultiwriteCB *mcb = opaque;
4236 trace_multiwrite_cb(mcb, ret);
4238 if (ret < 0 && !mcb->error) {
4242 mcb->num_requests--;
4243 if (mcb->num_requests == 0) {
4244 multiwrite_user_cb(mcb);
4249 static int multiwrite_req_compare(const void *a, const void *b)
4251 const BlockRequest *req1 = a, *req2 = b;
4254 * Note that we can't simply subtract req2->sector from req1->sector
4255 * here as that could overflow the return value.
4257 if (req1->sector > req2->sector) {
4259 } else if (req1->sector < req2->sector) {
4267 * Takes a bunch of requests and tries to merge them. Returns the number of
4268 * requests that remain after merging.
4270 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
4271 int num_reqs, MultiwriteCB *mcb)
4275 // Sort requests by start sector
4276 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
4278 // Check if adjacent requests touch the same clusters. If so, combine them,
4279 // filling up gaps with zero sectors.
4281 for (i = 1; i < num_reqs; i++) {
4283 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
4285 // Handle exactly sequential writes and overlapping writes.
4286 if (reqs[i].sector <= oldreq_last) {
4290 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
4296 QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
4297 qemu_iovec_init(qiov,
4298 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
4300 // Add the first request to the merged one. If the requests are
4301 // overlapping, drop the last sectors of the first request.
4302 size = (reqs[i].sector - reqs[outidx].sector) << 9;
4303 qemu_iovec_concat(qiov, reqs[outidx].qiov, 0, size);
4305 // We should need to add any zeros between the two requests
4306 assert (reqs[i].sector <= oldreq_last);
4308 // Add the second request
4309 qemu_iovec_concat(qiov, reqs[i].qiov, 0, reqs[i].qiov->size);
4311 reqs[outidx].nb_sectors = qiov->size >> 9;
4312 reqs[outidx].qiov = qiov;
4314 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
4317 reqs[outidx].sector = reqs[i].sector;
4318 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
4319 reqs[outidx].qiov = reqs[i].qiov;
4327 * Submit multiple AIO write requests at once.
4329 * On success, the function returns 0 and all requests in the reqs array have
4330 * been submitted. In error case this function returns -1, and any of the
4331 * requests may or may not be submitted yet. In particular, this means that the
4332 * callback will be called for some of the requests, for others it won't. The
4333 * caller must check the error field of the BlockRequest to wait for the right
4334 * callbacks (if error != 0, no callback will be called).
4336 * The implementation may modify the contents of the reqs array, e.g. to merge
4337 * requests. However, the fields opaque and error are left unmodified as they
4338 * are used to signal failure for a single request to the caller.
4340 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
4345 /* don't submit writes if we don't have a medium */
4346 if (bs->drv == NULL) {
4347 for (i = 0; i < num_reqs; i++) {
4348 reqs[i].error = -ENOMEDIUM;
4353 if (num_reqs == 0) {
4357 // Create MultiwriteCB structure
4358 mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
4359 mcb->num_requests = 0;
4360 mcb->num_callbacks = num_reqs;
4362 for (i = 0; i < num_reqs; i++) {
4363 mcb->callbacks[i].cb = reqs[i].cb;
4364 mcb->callbacks[i].opaque = reqs[i].opaque;
4367 // Check for mergable requests
4368 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
4370 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
4372 /* Run the aio requests. */
4373 mcb->num_requests = num_reqs;
4374 for (i = 0; i < num_reqs; i++) {
4375 bdrv_co_aio_rw_vector(bs, reqs[i].sector, reqs[i].qiov,
4376 reqs[i].nb_sectors, reqs[i].flags,
4384 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
4386 acb->aiocb_info->cancel(acb);
4389 /**************************************************************/
4390 /* async block device emulation */
4392 typedef struct BlockDriverAIOCBSync {
4393 BlockDriverAIOCB common;
4396 /* vector translation state */
4400 } BlockDriverAIOCBSync;
4402 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
4404 BlockDriverAIOCBSync *acb =
4405 container_of(blockacb, BlockDriverAIOCBSync, common);
4406 qemu_bh_delete(acb->bh);
4408 qemu_aio_release(acb);
4411 static const AIOCBInfo bdrv_em_aiocb_info = {
4412 .aiocb_size = sizeof(BlockDriverAIOCBSync),
4413 .cancel = bdrv_aio_cancel_em,
4416 static void bdrv_aio_bh_cb(void *opaque)
4418 BlockDriverAIOCBSync *acb = opaque;
4421 qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
4422 qemu_vfree(acb->bounce);
4423 acb->common.cb(acb->common.opaque, acb->ret);
4424 qemu_bh_delete(acb->bh);
4426 qemu_aio_release(acb);
4429 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
4433 BlockDriverCompletionFunc *cb,
4438 BlockDriverAIOCBSync *acb;
4440 acb = qemu_aio_get(&bdrv_em_aiocb_info, bs, cb, opaque);
4441 acb->is_write = is_write;
4443 acb->bounce = qemu_blockalign(bs, qiov->size);
4444 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
4447 qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
4448 acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
4450 acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
4453 qemu_bh_schedule(acb->bh);
4455 return &acb->common;
4458 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
4459 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
4460 BlockDriverCompletionFunc *cb, void *opaque)
4462 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
4465 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
4466 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
4467 BlockDriverCompletionFunc *cb, void *opaque)
4469 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
4473 typedef struct BlockDriverAIOCBCoroutine {
4474 BlockDriverAIOCB common;
4479 } BlockDriverAIOCBCoroutine;
4481 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb)
4483 BlockDriverAIOCBCoroutine *acb =
4484 container_of(blockacb, BlockDriverAIOCBCoroutine, common);
4493 static const AIOCBInfo bdrv_em_co_aiocb_info = {
4494 .aiocb_size = sizeof(BlockDriverAIOCBCoroutine),
4495 .cancel = bdrv_aio_co_cancel_em,
4498 static void bdrv_co_em_bh(void *opaque)
4500 BlockDriverAIOCBCoroutine *acb = opaque;
4502 acb->common.cb(acb->common.opaque, acb->req.error);
4508 qemu_bh_delete(acb->bh);
4509 qemu_aio_release(acb);
4512 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
4513 static void coroutine_fn bdrv_co_do_rw(void *opaque)
4515 BlockDriverAIOCBCoroutine *acb = opaque;
4516 BlockDriverState *bs = acb->common.bs;
4518 if (!acb->is_write) {
4519 acb->req.error = bdrv_co_do_readv(bs, acb->req.sector,
4520 acb->req.nb_sectors, acb->req.qiov, acb->req.flags);
4522 acb->req.error = bdrv_co_do_writev(bs, acb->req.sector,
4523 acb->req.nb_sectors, acb->req.qiov, acb->req.flags);
4526 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
4527 qemu_bh_schedule(acb->bh);
4530 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
4534 BdrvRequestFlags flags,
4535 BlockDriverCompletionFunc *cb,
4540 BlockDriverAIOCBCoroutine *acb;
4542 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
4543 acb->req.sector = sector_num;
4544 acb->req.nb_sectors = nb_sectors;
4545 acb->req.qiov = qiov;
4546 acb->req.flags = flags;
4547 acb->is_write = is_write;
4550 co = qemu_coroutine_create(bdrv_co_do_rw);
4551 qemu_coroutine_enter(co, acb);
4553 return &acb->common;
4556 static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
4558 BlockDriverAIOCBCoroutine *acb = opaque;
4559 BlockDriverState *bs = acb->common.bs;
4561 acb->req.error = bdrv_co_flush(bs);
4562 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
4563 qemu_bh_schedule(acb->bh);
4566 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
4567 BlockDriverCompletionFunc *cb, void *opaque)
4569 trace_bdrv_aio_flush(bs, opaque);
4572 BlockDriverAIOCBCoroutine *acb;
4574 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
4577 co = qemu_coroutine_create(bdrv_aio_flush_co_entry);
4578 qemu_coroutine_enter(co, acb);
4580 return &acb->common;
4583 static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque)
4585 BlockDriverAIOCBCoroutine *acb = opaque;
4586 BlockDriverState *bs = acb->common.bs;
4588 acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors);
4589 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
4590 qemu_bh_schedule(acb->bh);
4593 BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs,
4594 int64_t sector_num, int nb_sectors,
4595 BlockDriverCompletionFunc *cb, void *opaque)
4598 BlockDriverAIOCBCoroutine *acb;
4600 trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque);
4602 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
4603 acb->req.sector = sector_num;
4604 acb->req.nb_sectors = nb_sectors;
4606 co = qemu_coroutine_create(bdrv_aio_discard_co_entry);
4607 qemu_coroutine_enter(co, acb);
4609 return &acb->common;
4612 void bdrv_init(void)
4614 module_call_init(MODULE_INIT_BLOCK);
4617 void bdrv_init_with_whitelist(void)
4619 use_bdrv_whitelist = 1;
4623 void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs,
4624 BlockDriverCompletionFunc *cb, void *opaque)
4626 BlockDriverAIOCB *acb;
4628 acb = g_slice_alloc(aiocb_info->aiocb_size);
4629 acb->aiocb_info = aiocb_info;
4632 acb->opaque = opaque;
4636 void qemu_aio_release(void *p)
4638 BlockDriverAIOCB *acb = p;
4639 g_slice_free1(acb->aiocb_info->aiocb_size, acb);
4642 /**************************************************************/
4643 /* Coroutine block device emulation */
4645 typedef struct CoroutineIOCompletion {
4646 Coroutine *coroutine;
4648 } CoroutineIOCompletion;
4650 static void bdrv_co_io_em_complete(void *opaque, int ret)
4652 CoroutineIOCompletion *co = opaque;
4655 qemu_coroutine_enter(co->coroutine, NULL);
4658 static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num,
4659 int nb_sectors, QEMUIOVector *iov,
4662 CoroutineIOCompletion co = {
4663 .coroutine = qemu_coroutine_self(),
4665 BlockDriverAIOCB *acb;
4668 acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors,
4669 bdrv_co_io_em_complete, &co);
4671 acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors,
4672 bdrv_co_io_em_complete, &co);
4675 trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb);
4679 qemu_coroutine_yield();
4684 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
4685 int64_t sector_num, int nb_sectors,
4688 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false);
4691 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
4692 int64_t sector_num, int nb_sectors,
4695 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true);
4698 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
4700 RwCo *rwco = opaque;
4702 rwco->ret = bdrv_co_flush(rwco->bs);
4705 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
4709 if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
4713 /* Write back cached data to the OS even with cache=unsafe */
4714 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS);
4715 if (bs->drv->bdrv_co_flush_to_os) {
4716 ret = bs->drv->bdrv_co_flush_to_os(bs);
4722 /* But don't actually force it to the disk with cache=unsafe */
4723 if (bs->open_flags & BDRV_O_NO_FLUSH) {
4727 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK);
4728 if (bs->drv->bdrv_co_flush_to_disk) {
4729 ret = bs->drv->bdrv_co_flush_to_disk(bs);
4730 } else if (bs->drv->bdrv_aio_flush) {
4731 BlockDriverAIOCB *acb;
4732 CoroutineIOCompletion co = {
4733 .coroutine = qemu_coroutine_self(),
4736 acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
4740 qemu_coroutine_yield();
4745 * Some block drivers always operate in either writethrough or unsafe
4746 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
4747 * know how the server works (because the behaviour is hardcoded or
4748 * depends on server-side configuration), so we can't ensure that
4749 * everything is safe on disk. Returning an error doesn't work because
4750 * that would break guests even if the server operates in writethrough
4753 * Let's hope the user knows what he's doing.
4761 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
4762 * in the case of cache=unsafe, so there are no useless flushes.
4765 return bdrv_co_flush(bs->file);
4768 void bdrv_invalidate_cache(BlockDriverState *bs)
4770 if (bs->drv && bs->drv->bdrv_invalidate_cache) {
4771 bs->drv->bdrv_invalidate_cache(bs);
4775 void bdrv_invalidate_cache_all(void)
4777 BlockDriverState *bs;
4779 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
4780 bdrv_invalidate_cache(bs);
4784 void bdrv_clear_incoming_migration_all(void)
4786 BlockDriverState *bs;
4788 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
4789 bs->open_flags = bs->open_flags & ~(BDRV_O_INCOMING);
4793 int bdrv_flush(BlockDriverState *bs)
4801 if (qemu_in_coroutine()) {
4802 /* Fast-path if already in coroutine context */
4803 bdrv_flush_co_entry(&rwco);
4805 co = qemu_coroutine_create(bdrv_flush_co_entry);
4806 qemu_coroutine_enter(co, &rwco);
4807 while (rwco.ret == NOT_DONE) {
4815 typedef struct DiscardCo {
4816 BlockDriverState *bs;
4821 static void coroutine_fn bdrv_discard_co_entry(void *opaque)
4823 DiscardCo *rwco = opaque;
4825 rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors);
4828 /* if no limit is specified in the BlockLimits use a default
4829 * of 32768 512-byte sectors (16 MiB) per request.
4831 #define MAX_DISCARD_DEFAULT 32768
4833 int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
4840 } else if (bdrv_check_request(bs, sector_num, nb_sectors)) {
4842 } else if (bs->read_only) {
4846 bdrv_reset_dirty(bs, sector_num, nb_sectors);
4848 /* Do nothing if disabled. */
4849 if (!(bs->open_flags & BDRV_O_UNMAP)) {
4853 if (!bs->drv->bdrv_co_discard && !bs->drv->bdrv_aio_discard) {
4857 max_discard = bs->bl.max_discard ? bs->bl.max_discard : MAX_DISCARD_DEFAULT;
4858 while (nb_sectors > 0) {
4860 int num = nb_sectors;
4863 if (bs->bl.discard_alignment &&
4864 num >= bs->bl.discard_alignment &&
4865 sector_num % bs->bl.discard_alignment) {
4866 if (num > bs->bl.discard_alignment) {
4867 num = bs->bl.discard_alignment;
4869 num -= sector_num % bs->bl.discard_alignment;
4872 /* limit request size */
4873 if (num > max_discard) {
4877 if (bs->drv->bdrv_co_discard) {
4878 ret = bs->drv->bdrv_co_discard(bs, sector_num, num);
4880 BlockDriverAIOCB *acb;
4881 CoroutineIOCompletion co = {
4882 .coroutine = qemu_coroutine_self(),
4885 acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors,
4886 bdrv_co_io_em_complete, &co);
4890 qemu_coroutine_yield();
4894 if (ret && ret != -ENOTSUP) {
4904 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
4909 .sector_num = sector_num,
4910 .nb_sectors = nb_sectors,
4914 if (qemu_in_coroutine()) {
4915 /* Fast-path if already in coroutine context */
4916 bdrv_discard_co_entry(&rwco);
4918 co = qemu_coroutine_create(bdrv_discard_co_entry);
4919 qemu_coroutine_enter(co, &rwco);
4920 while (rwco.ret == NOT_DONE) {
4928 /**************************************************************/
4929 /* removable device support */
4932 * Return TRUE if the media is present
4934 int bdrv_is_inserted(BlockDriverState *bs)
4936 BlockDriver *drv = bs->drv;
4940 if (!drv->bdrv_is_inserted)
4942 return drv->bdrv_is_inserted(bs);
4946 * Return whether the media changed since the last call to this
4947 * function, or -ENOTSUP if we don't know. Most drivers don't know.
4949 int bdrv_media_changed(BlockDriverState *bs)
4951 BlockDriver *drv = bs->drv;
4953 if (drv && drv->bdrv_media_changed) {
4954 return drv->bdrv_media_changed(bs);
4960 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4962 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
4964 BlockDriver *drv = bs->drv;
4966 if (drv && drv->bdrv_eject) {
4967 drv->bdrv_eject(bs, eject_flag);
4970 if (bs->device_name[0] != '\0') {
4971 bdrv_emit_qmp_eject_event(bs, eject_flag);
4976 * Lock or unlock the media (if it is locked, the user won't be able
4977 * to eject it manually).
4979 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
4981 BlockDriver *drv = bs->drv;
4983 trace_bdrv_lock_medium(bs, locked);
4985 if (drv && drv->bdrv_lock_medium) {
4986 drv->bdrv_lock_medium(bs, locked);
4990 /* needed for generic scsi interface */
4992 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
4994 BlockDriver *drv = bs->drv;
4996 if (drv && drv->bdrv_ioctl)
4997 return drv->bdrv_ioctl(bs, req, buf);
5001 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
5002 unsigned long int req, void *buf,
5003 BlockDriverCompletionFunc *cb, void *opaque)
5005 BlockDriver *drv = bs->drv;
5007 if (drv && drv->bdrv_aio_ioctl)
5008 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
5012 void bdrv_set_guest_block_size(BlockDriverState *bs, int align)
5014 bs->guest_block_size = align;
5017 void *qemu_blockalign(BlockDriverState *bs, size_t size)
5019 return qemu_memalign(bdrv_opt_mem_align(bs), size);
5023 * Check if all memory in this vector is sector aligned.
5025 bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov)
5028 size_t alignment = bdrv_opt_mem_align(bs);
5030 for (i = 0; i < qiov->niov; i++) {
5031 if ((uintptr_t) qiov->iov[i].iov_base % alignment) {
5034 if (qiov->iov[i].iov_len % alignment) {
5042 BdrvDirtyBitmap *bdrv_create_dirty_bitmap(BlockDriverState *bs, int granularity)
5044 int64_t bitmap_size;
5045 BdrvDirtyBitmap *bitmap;
5047 assert((granularity & (granularity - 1)) == 0);
5049 granularity >>= BDRV_SECTOR_BITS;
5050 assert(granularity);
5051 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS);
5052 bitmap = g_malloc0(sizeof(BdrvDirtyBitmap));
5053 bitmap->bitmap = hbitmap_alloc(bitmap_size, ffs(granularity) - 1);
5054 QLIST_INSERT_HEAD(&bs->dirty_bitmaps, bitmap, list);
5058 void bdrv_release_dirty_bitmap(BlockDriverState *bs, BdrvDirtyBitmap *bitmap)
5060 BdrvDirtyBitmap *bm, *next;
5061 QLIST_FOREACH_SAFE(bm, &bs->dirty_bitmaps, list, next) {
5063 QLIST_REMOVE(bitmap, list);
5064 hbitmap_free(bitmap->bitmap);
5071 BlockDirtyInfoList *bdrv_query_dirty_bitmaps(BlockDriverState *bs)
5073 BdrvDirtyBitmap *bm;
5074 BlockDirtyInfoList *list = NULL;
5075 BlockDirtyInfoList **plist = &list;
5077 QLIST_FOREACH(bm, &bs->dirty_bitmaps, list) {
5078 BlockDirtyInfo *info = g_malloc0(sizeof(BlockDirtyInfo));
5079 BlockDirtyInfoList *entry = g_malloc0(sizeof(BlockDirtyInfoList));
5080 info->count = bdrv_get_dirty_count(bs, bm);
5082 ((int64_t) BDRV_SECTOR_SIZE << hbitmap_granularity(bm->bitmap));
5083 entry->value = info;
5085 plist = &entry->next;
5091 int bdrv_get_dirty(BlockDriverState *bs, BdrvDirtyBitmap *bitmap, int64_t sector)
5094 return hbitmap_get(bitmap->bitmap, sector);
5100 void bdrv_dirty_iter_init(BlockDriverState *bs,
5101 BdrvDirtyBitmap *bitmap, HBitmapIter *hbi)
5103 hbitmap_iter_init(hbi, bitmap->bitmap, 0);
5106 void bdrv_set_dirty(BlockDriverState *bs, int64_t cur_sector,
5109 BdrvDirtyBitmap *bitmap;
5110 QLIST_FOREACH(bitmap, &bs->dirty_bitmaps, list) {
5111 hbitmap_set(bitmap->bitmap, cur_sector, nr_sectors);
5115 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector, int nr_sectors)
5117 BdrvDirtyBitmap *bitmap;
5118 QLIST_FOREACH(bitmap, &bs->dirty_bitmaps, list) {
5119 hbitmap_reset(bitmap->bitmap, cur_sector, nr_sectors);
5123 int64_t bdrv_get_dirty_count(BlockDriverState *bs, BdrvDirtyBitmap *bitmap)
5125 return hbitmap_count(bitmap->bitmap);
5128 /* Get a reference to bs */
5129 void bdrv_ref(BlockDriverState *bs)
5134 /* Release a previously grabbed reference to bs.
5135 * If after releasing, reference count is zero, the BlockDriverState is
5137 void bdrv_unref(BlockDriverState *bs)
5139 assert(bs->refcnt > 0);
5140 if (--bs->refcnt == 0) {
5145 void bdrv_set_in_use(BlockDriverState *bs, int in_use)
5147 assert(bs->in_use != in_use);
5148 bs->in_use = in_use;
5151 int bdrv_in_use(BlockDriverState *bs)
5156 void bdrv_iostatus_enable(BlockDriverState *bs)
5158 bs->iostatus_enabled = true;
5159 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
5162 /* The I/O status is only enabled if the drive explicitly
5163 * enables it _and_ the VM is configured to stop on errors */
5164 bool bdrv_iostatus_is_enabled(const BlockDriverState *bs)
5166 return (bs->iostatus_enabled &&
5167 (bs->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
5168 bs->on_write_error == BLOCKDEV_ON_ERROR_STOP ||
5169 bs->on_read_error == BLOCKDEV_ON_ERROR_STOP));
5172 void bdrv_iostatus_disable(BlockDriverState *bs)
5174 bs->iostatus_enabled = false;
5177 void bdrv_iostatus_reset(BlockDriverState *bs)
5179 if (bdrv_iostatus_is_enabled(bs)) {
5180 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
5182 block_job_iostatus_reset(bs->job);
5187 void bdrv_iostatus_set_err(BlockDriverState *bs, int error)
5189 assert(bdrv_iostatus_is_enabled(bs));
5190 if (bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
5191 bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
5192 BLOCK_DEVICE_IO_STATUS_FAILED;
5197 bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes,
5198 enum BlockAcctType type)
5200 assert(type < BDRV_MAX_IOTYPE);
5202 cookie->bytes = bytes;
5203 cookie->start_time_ns = get_clock();
5204 cookie->type = type;
5208 bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie)
5210 assert(cookie->type < BDRV_MAX_IOTYPE);
5212 bs->nr_bytes[cookie->type] += cookie->bytes;
5213 bs->nr_ops[cookie->type]++;
5214 bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns;
5217 void bdrv_img_create(const char *filename, const char *fmt,
5218 const char *base_filename, const char *base_fmt,
5219 char *options, uint64_t img_size, int flags,
5220 Error **errp, bool quiet)
5222 QEMUOptionParameter *param = NULL, *create_options = NULL;
5223 QEMUOptionParameter *backing_fmt, *backing_file, *size;
5224 BlockDriver *drv, *proto_drv;
5225 BlockDriver *backing_drv = NULL;
5226 Error *local_err = NULL;
5229 /* Find driver and parse its options */
5230 drv = bdrv_find_format(fmt);
5232 error_setg(errp, "Unknown file format '%s'", fmt);
5236 proto_drv = bdrv_find_protocol(filename, true);
5238 error_setg(errp, "Unknown protocol '%s'", filename);
5242 create_options = append_option_parameters(create_options,
5243 drv->create_options);
5244 create_options = append_option_parameters(create_options,
5245 proto_drv->create_options);
5247 /* Create parameter list with default values */
5248 param = parse_option_parameters("", create_options, param);
5250 set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size);
5252 /* Parse -o options */
5254 param = parse_option_parameters(options, create_options, param);
5255 if (param == NULL) {
5256 error_setg(errp, "Invalid options for file format '%s'.", fmt);
5261 if (base_filename) {
5262 if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE,
5264 error_setg(errp, "Backing file not supported for file format '%s'",
5271 if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) {
5272 error_setg(errp, "Backing file format not supported for file "
5273 "format '%s'", fmt);
5278 backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE);
5279 if (backing_file && backing_file->value.s) {
5280 if (!strcmp(filename, backing_file->value.s)) {
5281 error_setg(errp, "Error: Trying to create an image with the "
5282 "same filename as the backing file");
5287 backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT);
5288 if (backing_fmt && backing_fmt->value.s) {
5289 backing_drv = bdrv_find_format(backing_fmt->value.s);
5291 error_setg(errp, "Unknown backing file format '%s'",
5292 backing_fmt->value.s);
5297 // The size for the image must always be specified, with one exception:
5298 // If we are using a backing file, we can obtain the size from there
5299 size = get_option_parameter(param, BLOCK_OPT_SIZE);
5300 if (size && size->value.n == -1) {
5301 if (backing_file && backing_file->value.s) {
5302 BlockDriverState *bs;
5307 /* backing files always opened read-only */
5309 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
5312 ret = bdrv_open(&bs, backing_file->value.s, NULL, back_flags,
5313 backing_drv, &local_err);
5315 error_setg_errno(errp, -ret, "Could not open '%s': %s",
5316 backing_file->value.s,
5317 error_get_pretty(local_err));
5318 error_free(local_err);
5322 bdrv_get_geometry(bs, &size);
5325 snprintf(buf, sizeof(buf), "%" PRId64, size);
5326 set_option_parameter(param, BLOCK_OPT_SIZE, buf);
5330 error_setg(errp, "Image creation needs a size parameter");
5336 printf("Formatting '%s', fmt=%s ", filename, fmt);
5337 print_option_parameters(param);
5340 ret = bdrv_create(drv, filename, param, &local_err);
5341 if (ret == -EFBIG) {
5342 /* This is generally a better message than whatever the driver would
5343 * deliver (especially because of the cluster_size_hint), since that
5344 * is most probably not much different from "image too large". */
5345 const char *cluster_size_hint = "";
5346 if (get_option_parameter(create_options, BLOCK_OPT_CLUSTER_SIZE)) {
5347 cluster_size_hint = " (try using a larger cluster size)";
5349 error_setg(errp, "The image size is too large for file format '%s'"
5350 "%s", fmt, cluster_size_hint);
5351 error_free(local_err);
5356 free_option_parameters(create_options);
5357 free_option_parameters(param);
5360 error_propagate(errp, local_err);
5364 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
5366 /* Currently BlockDriverState always uses the main loop AioContext */
5367 return qemu_get_aio_context();
5370 void bdrv_add_before_write_notifier(BlockDriverState *bs,
5371 NotifierWithReturn *notifier)
5373 notifier_with_return_list_add(&bs->before_write_notifiers, notifier);
5376 int bdrv_amend_options(BlockDriverState *bs, QEMUOptionParameter *options)
5378 if (bs->drv->bdrv_amend_options == NULL) {
5381 return bs->drv->bdrv_amend_options(bs, options);
5384 /* Used to recurse on single child block filters.
5385 * Single child block filter will store their child in bs->file.
5387 bool bdrv_generic_is_first_non_filter(BlockDriverState *bs,
5388 BlockDriverState *candidate)
5394 if (!bs->drv->authorizations[BS_IS_A_FILTER]) {
5395 if (bs == candidate) {
5402 if (!bs->drv->authorizations[BS_FILTER_PASS_DOWN]) {
5410 return bdrv_recurse_is_first_non_filter(bs->file, candidate);
5413 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
5414 BlockDriverState *candidate)
5416 if (bs->drv && bs->drv->bdrv_recurse_is_first_non_filter) {
5417 return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate);
5420 return bdrv_generic_is_first_non_filter(bs, candidate);
5423 /* This function checks if the candidate is the first non filter bs down it's
5424 * bs chain. Since we don't have pointers to parents it explore all bs chains
5425 * from the top. Some filters can choose not to pass down the recursion.
5427 bool bdrv_is_first_non_filter(BlockDriverState *candidate)
5429 BlockDriverState *bs;
5431 /* walk down the bs forest recursively */
5432 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
5435 perm = bdrv_recurse_is_first_non_filter(bs, candidate);
5437 /* candidate is the first non filter */