2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24 #include "config-host.h"
25 #include "qemu-common.h"
27 #include "monitor/monitor.h"
28 #include "block/block_int.h"
29 #include "block/blockjob.h"
30 #include "qemu/module.h"
31 #include "qapi/qmp/qjson.h"
32 #include "sysemu/sysemu.h"
33 #include "qemu/notify.h"
34 #include "block/coroutine.h"
35 #include "block/qapi.h"
36 #include "qmp-commands.h"
37 #include "qemu/timer.h"
40 #include <sys/types.h>
42 #include <sys/ioctl.h>
43 #include <sys/queue.h>
53 struct BdrvDirtyBitmap {
55 QLIST_ENTRY(BdrvDirtyBitmap) list;
58 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
60 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load);
61 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
62 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
63 BlockDriverCompletionFunc *cb, void *opaque);
64 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
65 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
66 BlockDriverCompletionFunc *cb, void *opaque);
67 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
68 int64_t sector_num, int nb_sectors,
70 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
71 int64_t sector_num, int nb_sectors,
73 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
74 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
75 BdrvRequestFlags flags);
76 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
77 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
78 BdrvRequestFlags flags);
79 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
83 BdrvRequestFlags flags,
84 BlockDriverCompletionFunc *cb,
87 static void coroutine_fn bdrv_co_do_rw(void *opaque);
88 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
89 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags);
91 static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
92 QTAILQ_HEAD_INITIALIZER(bdrv_states);
94 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
95 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
97 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
98 QLIST_HEAD_INITIALIZER(bdrv_drivers);
100 /* If non-zero, use only whitelisted block drivers */
101 static int use_bdrv_whitelist;
104 static int is_windows_drive_prefix(const char *filename)
106 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
107 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
111 int is_windows_drive(const char *filename)
113 if (is_windows_drive_prefix(filename) &&
116 if (strstart(filename, "\\\\.\\", NULL) ||
117 strstart(filename, "//./", NULL))
123 /* throttling disk I/O limits */
124 void bdrv_set_io_limits(BlockDriverState *bs,
129 throttle_config(&bs->throttle_state, cfg);
131 for (i = 0; i < 2; i++) {
132 qemu_co_enter_next(&bs->throttled_reqs[i]);
136 /* this function drain all the throttled IOs */
137 static bool bdrv_start_throttled_reqs(BlockDriverState *bs)
139 bool drained = false;
140 bool enabled = bs->io_limits_enabled;
143 bs->io_limits_enabled = false;
145 for (i = 0; i < 2; i++) {
146 while (qemu_co_enter_next(&bs->throttled_reqs[i])) {
151 bs->io_limits_enabled = enabled;
156 void bdrv_io_limits_disable(BlockDriverState *bs)
158 bs->io_limits_enabled = false;
160 bdrv_start_throttled_reqs(bs);
162 throttle_destroy(&bs->throttle_state);
165 static void bdrv_throttle_read_timer_cb(void *opaque)
167 BlockDriverState *bs = opaque;
168 qemu_co_enter_next(&bs->throttled_reqs[0]);
171 static void bdrv_throttle_write_timer_cb(void *opaque)
173 BlockDriverState *bs = opaque;
174 qemu_co_enter_next(&bs->throttled_reqs[1]);
177 /* should be called before bdrv_set_io_limits if a limit is set */
178 void bdrv_io_limits_enable(BlockDriverState *bs)
180 assert(!bs->io_limits_enabled);
181 throttle_init(&bs->throttle_state,
183 bdrv_throttle_read_timer_cb,
184 bdrv_throttle_write_timer_cb,
186 bs->io_limits_enabled = true;
189 /* This function makes an IO wait if needed
191 * @nb_sectors: the number of sectors of the IO
192 * @is_write: is the IO a write
194 static void bdrv_io_limits_intercept(BlockDriverState *bs,
198 /* does this io must wait */
199 bool must_wait = throttle_schedule_timer(&bs->throttle_state, is_write);
201 /* if must wait or any request of this type throttled queue the IO */
203 !qemu_co_queue_empty(&bs->throttled_reqs[is_write])) {
204 qemu_co_queue_wait(&bs->throttled_reqs[is_write]);
207 /* the IO will be executed, do the accounting */
208 throttle_account(&bs->throttle_state,
210 nb_sectors * BDRV_SECTOR_SIZE);
212 /* if the next request must wait -> do nothing */
213 if (throttle_schedule_timer(&bs->throttle_state, is_write)) {
217 /* else queue next request for execution */
218 qemu_co_queue_next(&bs->throttled_reqs[is_write]);
221 size_t bdrv_opt_mem_align(BlockDriverState *bs)
223 if (!bs || !bs->drv) {
224 /* 4k should be on the safe side */
228 return bs->bl.opt_mem_alignment;
231 /* check if the path starts with "<protocol>:" */
232 static int path_has_protocol(const char *path)
237 if (is_windows_drive(path) ||
238 is_windows_drive_prefix(path)) {
241 p = path + strcspn(path, ":/\\");
243 p = path + strcspn(path, ":/");
249 int path_is_absolute(const char *path)
252 /* specific case for names like: "\\.\d:" */
253 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
256 return (*path == '/' || *path == '\\');
258 return (*path == '/');
262 /* if filename is absolute, just copy it to dest. Otherwise, build a
263 path to it by considering it is relative to base_path. URL are
265 void path_combine(char *dest, int dest_size,
266 const char *base_path,
267 const char *filename)
274 if (path_is_absolute(filename)) {
275 pstrcpy(dest, dest_size, filename);
277 p = strchr(base_path, ':');
282 p1 = strrchr(base_path, '/');
286 p2 = strrchr(base_path, '\\');
298 if (len > dest_size - 1)
300 memcpy(dest, base_path, len);
302 pstrcat(dest, dest_size, filename);
306 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz)
308 if (bs->backing_file[0] == '\0' || path_has_protocol(bs->backing_file)) {
309 pstrcpy(dest, sz, bs->backing_file);
311 path_combine(dest, sz, bs->filename, bs->backing_file);
315 void bdrv_register(BlockDriver *bdrv)
317 /* Block drivers without coroutine functions need emulation */
318 if (!bdrv->bdrv_co_readv) {
319 bdrv->bdrv_co_readv = bdrv_co_readv_em;
320 bdrv->bdrv_co_writev = bdrv_co_writev_em;
322 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
323 * the block driver lacks aio we need to emulate that too.
325 if (!bdrv->bdrv_aio_readv) {
326 /* add AIO emulation layer */
327 bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
328 bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
332 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
335 /* create a new block device (by default it is empty) */
336 BlockDriverState *bdrv_new(const char *device_name)
338 BlockDriverState *bs;
340 bs = g_malloc0(sizeof(BlockDriverState));
341 QLIST_INIT(&bs->dirty_bitmaps);
342 pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
343 if (device_name[0] != '\0') {
344 QTAILQ_INSERT_TAIL(&bdrv_states, bs, device_list);
346 bdrv_iostatus_disable(bs);
347 notifier_list_init(&bs->close_notifiers);
348 notifier_with_return_list_init(&bs->before_write_notifiers);
349 qemu_co_queue_init(&bs->throttled_reqs[0]);
350 qemu_co_queue_init(&bs->throttled_reqs[1]);
356 void bdrv_add_close_notifier(BlockDriverState *bs, Notifier *notify)
358 notifier_list_add(&bs->close_notifiers, notify);
361 BlockDriver *bdrv_find_format(const char *format_name)
364 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
365 if (!strcmp(drv1->format_name, format_name)) {
372 static int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
374 static const char *whitelist_rw[] = {
375 CONFIG_BDRV_RW_WHITELIST
377 static const char *whitelist_ro[] = {
378 CONFIG_BDRV_RO_WHITELIST
382 if (!whitelist_rw[0] && !whitelist_ro[0]) {
383 return 1; /* no whitelist, anything goes */
386 for (p = whitelist_rw; *p; p++) {
387 if (!strcmp(drv->format_name, *p)) {
392 for (p = whitelist_ro; *p; p++) {
393 if (!strcmp(drv->format_name, *p)) {
401 BlockDriver *bdrv_find_whitelisted_format(const char *format_name,
404 BlockDriver *drv = bdrv_find_format(format_name);
405 return drv && bdrv_is_whitelisted(drv, read_only) ? drv : NULL;
408 typedef struct CreateCo {
411 QEMUOptionParameter *options;
416 static void coroutine_fn bdrv_create_co_entry(void *opaque)
418 Error *local_err = NULL;
421 CreateCo *cco = opaque;
424 ret = cco->drv->bdrv_create(cco->filename, cco->options, &local_err);
425 if (error_is_set(&local_err)) {
426 error_propagate(&cco->err, local_err);
431 int bdrv_create(BlockDriver *drv, const char* filename,
432 QEMUOptionParameter *options, Error **errp)
439 .filename = g_strdup(filename),
445 if (!drv->bdrv_create) {
446 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
451 if (qemu_in_coroutine()) {
452 /* Fast-path if already in coroutine context */
453 bdrv_create_co_entry(&cco);
455 co = qemu_coroutine_create(bdrv_create_co_entry);
456 qemu_coroutine_enter(co, &cco);
457 while (cco.ret == NOT_DONE) {
464 if (error_is_set(&cco.err)) {
465 error_propagate(errp, cco.err);
467 error_setg_errno(errp, -ret, "Could not create image");
472 g_free(cco.filename);
476 int bdrv_create_file(const char* filename, QEMUOptionParameter *options,
480 Error *local_err = NULL;
483 drv = bdrv_find_protocol(filename, true);
485 error_setg(errp, "Could not find protocol for file '%s'", filename);
489 ret = bdrv_create(drv, filename, options, &local_err);
490 if (error_is_set(&local_err)) {
491 error_propagate(errp, local_err);
496 int bdrv_refresh_limits(BlockDriverState *bs)
498 BlockDriver *drv = bs->drv;
500 memset(&bs->bl, 0, sizeof(bs->bl));
506 /* Take some limits from the children as a default */
508 bdrv_refresh_limits(bs->file);
509 bs->bl.opt_transfer_length = bs->file->bl.opt_transfer_length;
510 bs->bl.opt_mem_alignment = bs->file->bl.opt_mem_alignment;
512 bs->bl.opt_mem_alignment = 512;
515 if (bs->backing_hd) {
516 bdrv_refresh_limits(bs->backing_hd);
517 bs->bl.opt_transfer_length =
518 MAX(bs->bl.opt_transfer_length,
519 bs->backing_hd->bl.opt_transfer_length);
520 bs->bl.opt_mem_alignment =
521 MAX(bs->bl.opt_mem_alignment,
522 bs->backing_hd->bl.opt_mem_alignment);
525 /* Then let the driver override it */
526 if (drv->bdrv_refresh_limits) {
527 return drv->bdrv_refresh_limits(bs);
534 * Create a uniquely-named empty temporary file.
535 * Return 0 upon success, otherwise a negative errno value.
537 int get_tmp_filename(char *filename, int size)
540 char temp_dir[MAX_PATH];
541 /* GetTempFileName requires that its output buffer (4th param)
542 have length MAX_PATH or greater. */
543 assert(size >= MAX_PATH);
544 return (GetTempPath(MAX_PATH, temp_dir)
545 && GetTempFileName(temp_dir, "qem", 0, filename)
546 ? 0 : -GetLastError());
550 tmpdir = getenv("TMPDIR");
553 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
556 fd = mkstemp(filename);
560 if (close(fd) != 0) {
569 * Detect host devices. By convention, /dev/cdrom[N] is always
570 * recognized as a host CDROM.
572 static BlockDriver *find_hdev_driver(const char *filename)
574 int score_max = 0, score;
575 BlockDriver *drv = NULL, *d;
577 QLIST_FOREACH(d, &bdrv_drivers, list) {
578 if (d->bdrv_probe_device) {
579 score = d->bdrv_probe_device(filename);
580 if (score > score_max) {
590 BlockDriver *bdrv_find_protocol(const char *filename,
591 bool allow_protocol_prefix)
598 /* TODO Drivers without bdrv_file_open must be specified explicitly */
601 * XXX(hch): we really should not let host device detection
602 * override an explicit protocol specification, but moving this
603 * later breaks access to device names with colons in them.
604 * Thanks to the brain-dead persistent naming schemes on udev-
605 * based Linux systems those actually are quite common.
607 drv1 = find_hdev_driver(filename);
612 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
613 return bdrv_find_format("file");
616 p = strchr(filename, ':');
619 if (len > sizeof(protocol) - 1)
620 len = sizeof(protocol) - 1;
621 memcpy(protocol, filename, len);
622 protocol[len] = '\0';
623 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
624 if (drv1->protocol_name &&
625 !strcmp(drv1->protocol_name, protocol)) {
632 static int find_image_format(BlockDriverState *bs, const char *filename,
633 BlockDriver **pdrv, Error **errp)
635 int score, score_max;
636 BlockDriver *drv1, *drv;
640 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
641 if (bs->sg || !bdrv_is_inserted(bs) || bdrv_getlength(bs) == 0) {
642 drv = bdrv_find_format("raw");
644 error_setg(errp, "Could not find raw image format");
651 ret = bdrv_pread(bs, 0, buf, sizeof(buf));
653 error_setg_errno(errp, -ret, "Could not read image for determining its "
661 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
662 if (drv1->bdrv_probe) {
663 score = drv1->bdrv_probe(buf, ret, filename);
664 if (score > score_max) {
671 error_setg(errp, "Could not determine image format: No compatible "
680 * Set the current 'total_sectors' value
682 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
684 BlockDriver *drv = bs->drv;
686 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
690 /* query actual device if possible, otherwise just trust the hint */
691 if (drv->bdrv_getlength) {
692 int64_t length = drv->bdrv_getlength(bs);
696 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
699 bs->total_sectors = hint;
704 * Set open flags for a given discard mode
706 * Return 0 on success, -1 if the discard mode was invalid.
708 int bdrv_parse_discard_flags(const char *mode, int *flags)
710 *flags &= ~BDRV_O_UNMAP;
712 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
714 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
715 *flags |= BDRV_O_UNMAP;
724 * Set open flags for a given cache mode
726 * Return 0 on success, -1 if the cache mode was invalid.
728 int bdrv_parse_cache_flags(const char *mode, int *flags)
730 *flags &= ~BDRV_O_CACHE_MASK;
732 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
733 *flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB;
734 } else if (!strcmp(mode, "directsync")) {
735 *flags |= BDRV_O_NOCACHE;
736 } else if (!strcmp(mode, "writeback")) {
737 *flags |= BDRV_O_CACHE_WB;
738 } else if (!strcmp(mode, "unsafe")) {
739 *flags |= BDRV_O_CACHE_WB;
740 *flags |= BDRV_O_NO_FLUSH;
741 } else if (!strcmp(mode, "writethrough")) {
742 /* this is the default */
751 * The copy-on-read flag is actually a reference count so multiple users may
752 * use the feature without worrying about clobbering its previous state.
753 * Copy-on-read stays enabled until all users have called to disable it.
755 void bdrv_enable_copy_on_read(BlockDriverState *bs)
760 void bdrv_disable_copy_on_read(BlockDriverState *bs)
762 assert(bs->copy_on_read > 0);
766 static int bdrv_open_flags(BlockDriverState *bs, int flags)
768 int open_flags = flags | BDRV_O_CACHE_WB;
771 * Clear flags that are internal to the block layer before opening the
774 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
777 * Snapshots should be writable.
779 if (bs->is_temporary) {
780 open_flags |= BDRV_O_RDWR;
786 static int bdrv_assign_node_name(BlockDriverState *bs,
787 const char *node_name,
794 /* empty string node name is invalid */
795 if (node_name[0] == '\0') {
796 error_setg(errp, "Empty node name");
800 /* takes care of avoiding duplicates node names */
801 if (bdrv_find_node(node_name)) {
802 error_setg(errp, "Duplicate node name");
806 /* copy node name into the bs and insert it into the graph list */
807 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
808 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
814 * Common part for opening disk images and files
816 * Removes all processed options from *options.
818 static int bdrv_open_common(BlockDriverState *bs, BlockDriverState *file,
819 QDict *options, int flags, BlockDriver *drv, Error **errp)
822 const char *filename;
823 const char *node_name = NULL;
824 Error *local_err = NULL;
827 assert(bs->file == NULL);
828 assert(options != NULL && bs->options != options);
831 filename = file->filename;
833 filename = qdict_get_try_str(options, "filename");
836 trace_bdrv_open_common(bs, filename ?: "", flags, drv->format_name);
838 node_name = qdict_get_try_str(options, "node-name");
839 ret = bdrv_assign_node_name(bs, node_name, errp);
843 qdict_del(options, "node-name");
845 /* bdrv_open() with directly using a protocol as drv. This layer is already
846 * opened, so assign it to bs (while file becomes a closed BlockDriverState)
847 * and return immediately. */
848 if (file != NULL && drv->bdrv_file_open) {
853 bs->open_flags = flags;
854 bs->guest_block_size = 512;
855 bs->request_alignment = 512;
856 bs->zero_beyond_eof = true;
857 open_flags = bdrv_open_flags(bs, flags);
858 bs->read_only = !(open_flags & BDRV_O_RDWR);
860 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
862 !bs->read_only && bdrv_is_whitelisted(drv, true)
863 ? "Driver '%s' can only be used for read-only devices"
864 : "Driver '%s' is not whitelisted",
869 assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
870 if (flags & BDRV_O_COPY_ON_READ) {
871 if (!bs->read_only) {
872 bdrv_enable_copy_on_read(bs);
874 error_setg(errp, "Can't use copy-on-read on read-only device");
879 if (filename != NULL) {
880 pstrcpy(bs->filename, sizeof(bs->filename), filename);
882 bs->filename[0] = '\0';
886 bs->opaque = g_malloc0(drv->instance_size);
888 bs->enable_write_cache = !!(flags & BDRV_O_CACHE_WB);
890 /* Open the image, either directly or using a protocol */
891 if (drv->bdrv_file_open) {
892 assert(file == NULL);
893 assert(!drv->bdrv_needs_filename || filename != NULL);
894 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
897 error_setg(errp, "Can't use '%s' as a block driver for the "
898 "protocol level", drv->format_name);
903 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
907 if (error_is_set(&local_err)) {
908 error_propagate(errp, local_err);
909 } else if (bs->filename[0]) {
910 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
912 error_setg_errno(errp, -ret, "Could not open image");
917 ret = refresh_total_sectors(bs, bs->total_sectors);
919 error_setg_errno(errp, -ret, "Could not refresh total sector count");
923 bdrv_refresh_limits(bs);
924 assert(bdrv_opt_mem_align(bs) != 0);
925 assert(bs->request_alignment != 0);
928 if (bs->is_temporary) {
929 assert(bs->filename[0] != '\0');
930 unlink(bs->filename);
944 * Opens a file using a protocol (file, host_device, nbd, ...)
946 * options is a QDict of options to pass to the block drivers, or NULL for an
947 * empty set of options. The reference to the QDict belongs to the block layer
948 * after the call (even on failure), so if the caller intends to reuse the
949 * dictionary, it needs to use QINCREF() before calling bdrv_file_open.
951 int bdrv_file_open(BlockDriverState **pbs, const char *filename,
952 const char *reference, QDict *options, int flags,
955 BlockDriverState *bs = NULL;
958 bool allow_protocol_prefix = false;
959 Error *local_err = NULL;
962 /* NULL means an empty set of options */
963 if (options == NULL) {
964 options = qdict_new();
968 if (filename || qdict_size(options)) {
969 error_setg(errp, "Cannot reference an existing block device with "
970 "additional options or a new filename");
975 bs = bdrv_find(reference);
977 error_setg(errp, "Cannot find block device '%s'", reference);
986 bs->options = options;
987 options = qdict_clone_shallow(options);
989 /* Fetch the file name from the options QDict if necessary */
991 filename = qdict_get_try_str(options, "filename");
992 } else if (filename && !qdict_haskey(options, "filename")) {
993 qdict_put(options, "filename", qstring_from_str(filename));
994 allow_protocol_prefix = true;
996 error_setg(errp, "Can't specify 'file' and 'filename' options at the "
1002 /* Find the right block driver */
1003 drvname = qdict_get_try_str(options, "driver");
1005 drv = bdrv_find_format(drvname);
1007 error_setg(errp, "Unknown driver '%s'", drvname);
1009 qdict_del(options, "driver");
1010 } else if (filename) {
1011 drv = bdrv_find_protocol(filename, allow_protocol_prefix);
1013 error_setg(errp, "Unknown protocol");
1016 error_setg(errp, "Must specify either driver or file");
1021 /* errp has been set already */
1026 /* Parse the filename and open it */
1027 if (drv->bdrv_parse_filename && filename) {
1028 drv->bdrv_parse_filename(filename, options, &local_err);
1029 if (error_is_set(&local_err)) {
1030 error_propagate(errp, local_err);
1034 qdict_del(options, "filename");
1035 } else if (drv->bdrv_needs_filename && !filename) {
1036 error_setg(errp, "The '%s' block driver requires a file name",
1042 if (!drv->bdrv_file_open) {
1043 ret = bdrv_open(bs, filename, options, flags, drv, &local_err);
1046 ret = bdrv_open_common(bs, NULL, options, flags, drv, &local_err);
1049 error_propagate(errp, local_err);
1053 /* Check if any unknown options were used */
1054 if (options && (qdict_size(options) != 0)) {
1055 const QDictEntry *entry = qdict_first(options);
1056 error_setg(errp, "Block protocol '%s' doesn't support the option '%s'",
1057 drv->format_name, entry->key);
1070 QDECREF(bs->options);
1077 * Opens the backing file for a BlockDriverState if not yet open
1079 * options is a QDict of options to pass to the block drivers, or NULL for an
1080 * empty set of options. The reference to the QDict is transferred to this
1081 * function (even on failure), so if the caller intends to reuse the dictionary,
1082 * it needs to use QINCREF() before calling bdrv_file_open.
1084 int bdrv_open_backing_file(BlockDriverState *bs, QDict *options, Error **errp)
1086 char backing_filename[PATH_MAX];
1087 int back_flags, ret;
1088 BlockDriver *back_drv = NULL;
1089 Error *local_err = NULL;
1091 if (bs->backing_hd != NULL) {
1096 /* NULL means an empty set of options */
1097 if (options == NULL) {
1098 options = qdict_new();
1101 bs->open_flags &= ~BDRV_O_NO_BACKING;
1102 if (qdict_haskey(options, "file.filename")) {
1103 backing_filename[0] = '\0';
1104 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
1108 bdrv_get_full_backing_filename(bs, backing_filename,
1109 sizeof(backing_filename));
1112 bs->backing_hd = bdrv_new("");
1114 if (bs->backing_format[0] != '\0') {
1115 back_drv = bdrv_find_format(bs->backing_format);
1118 /* backing files always opened read-only */
1119 back_flags = bs->open_flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT |
1120 BDRV_O_COPY_ON_READ);
1122 ret = bdrv_open(bs->backing_hd,
1123 *backing_filename ? backing_filename : NULL, options,
1124 back_flags, back_drv, &local_err);
1126 bdrv_unref(bs->backing_hd);
1127 bs->backing_hd = NULL;
1128 bs->open_flags |= BDRV_O_NO_BACKING;
1129 error_setg(errp, "Could not open backing file: %s",
1130 error_get_pretty(local_err));
1131 error_free(local_err);
1135 if (bs->backing_hd->file) {
1136 pstrcpy(bs->backing_file, sizeof(bs->backing_file),
1137 bs->backing_hd->file->filename);
1140 /* Recalculate the BlockLimits with the backing file */
1141 bdrv_refresh_limits(bs);
1147 * Opens a disk image whose options are given as BlockdevRef in another block
1150 * If force_raw is true, bdrv_file_open() will be used, thereby preventing any
1151 * image format auto-detection. If it is false and a filename is given,
1152 * bdrv_open() will be used for auto-detection.
1154 * If allow_none is true, no image will be opened if filename is false and no
1155 * BlockdevRef is given. *pbs will remain unchanged and 0 will be returned.
1157 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
1158 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
1159 * itself, all options starting with "${bdref_key}." are considered part of the
1162 * The BlockdevRef will be removed from the options QDict.
1164 int bdrv_open_image(BlockDriverState **pbs, const char *filename,
1165 QDict *options, const char *bdref_key, int flags,
1166 bool force_raw, bool allow_none, Error **errp)
1168 QDict *image_options;
1170 char *bdref_key_dot;
1171 const char *reference;
1173 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
1174 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
1175 g_free(bdref_key_dot);
1177 reference = qdict_get_try_str(options, bdref_key);
1178 if (!filename && !reference && !qdict_size(image_options)) {
1182 error_setg(errp, "A block device must be specified for \"%s\"",
1189 if (filename && !force_raw) {
1190 /* If a filename is given and the block driver should be detected
1191 automatically (instead of using none), use bdrv_open() in order to do
1192 that auto-detection. */
1193 BlockDriverState *bs;
1196 error_setg(errp, "Cannot reference an existing block device while "
1197 "giving a filename");
1203 ret = bdrv_open(bs, filename, image_options, flags, NULL, errp);
1210 ret = bdrv_file_open(pbs, filename, reference, image_options, flags,
1215 qdict_del(options, bdref_key);
1220 * Opens a disk image (raw, qcow2, vmdk, ...)
1222 * options is a QDict of options to pass to the block drivers, or NULL for an
1223 * empty set of options. The reference to the QDict belongs to the block layer
1224 * after the call (even on failure), so if the caller intends to reuse the
1225 * dictionary, it needs to use QINCREF() before calling bdrv_open.
1227 int bdrv_open(BlockDriverState *bs, const char *filename, QDict *options,
1228 int flags, BlockDriver *drv, Error **errp)
1231 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
1232 char tmp_filename[PATH_MAX + 1];
1233 BlockDriverState *file = NULL;
1234 const char *drvname;
1235 Error *local_err = NULL;
1237 /* NULL means an empty set of options */
1238 if (options == NULL) {
1239 options = qdict_new();
1242 bs->options = options;
1243 options = qdict_clone_shallow(options);
1245 /* For snapshot=on, create a temporary qcow2 overlay */
1246 if (flags & BDRV_O_SNAPSHOT) {
1247 BlockDriverState *bs1;
1249 BlockDriver *bdrv_qcow2;
1250 QEMUOptionParameter *create_options;
1251 QDict *snapshot_options;
1253 /* if snapshot, we create a temporary backing file and open it
1254 instead of opening 'filename' directly */
1256 /* Get the required size from the image */
1259 ret = bdrv_open(bs1, filename, options, BDRV_O_NO_BACKING,
1265 total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
1269 /* Create the temporary image */
1270 ret = get_tmp_filename(tmp_filename, sizeof(tmp_filename));
1272 error_setg_errno(errp, -ret, "Could not get temporary filename");
1276 bdrv_qcow2 = bdrv_find_format("qcow2");
1277 create_options = parse_option_parameters("", bdrv_qcow2->create_options,
1280 set_option_parameter_int(create_options, BLOCK_OPT_SIZE, total_size);
1282 ret = bdrv_create(bdrv_qcow2, tmp_filename, create_options, &local_err);
1283 free_option_parameters(create_options);
1285 error_setg_errno(errp, -ret, "Could not create temporary overlay "
1286 "'%s': %s", tmp_filename,
1287 error_get_pretty(local_err));
1288 error_free(local_err);
1293 /* Prepare a new options QDict for the temporary file, where user
1294 * options refer to the backing file */
1296 qdict_put(options, "file.filename", qstring_from_str(filename));
1299 qdict_put(options, "driver", qstring_from_str(drv->format_name));
1302 snapshot_options = qdict_new();
1303 qdict_put(snapshot_options, "backing", options);
1304 qdict_flatten(snapshot_options);
1306 bs->options = snapshot_options;
1307 options = qdict_clone_shallow(bs->options);
1309 filename = tmp_filename;
1311 bs->is_temporary = 1;
1314 /* Open image file without format layer */
1315 if (flags & BDRV_O_RDWR) {
1316 flags |= BDRV_O_ALLOW_RDWR;
1319 ret = bdrv_open_image(&file, filename, options, "file",
1320 bdrv_open_flags(bs, flags | BDRV_O_UNMAP), true, true,
1326 /* Find the right image format driver */
1327 drvname = qdict_get_try_str(options, "driver");
1329 drv = bdrv_find_format(drvname);
1330 qdict_del(options, "driver");
1332 error_setg(errp, "Invalid driver: '%s'", drvname);
1334 goto unlink_and_fail;
1340 ret = find_image_format(file, filename, &drv, &local_err);
1342 error_setg(errp, "Must specify either driver or file");
1344 goto unlink_and_fail;
1349 goto unlink_and_fail;
1352 /* Open the image */
1353 ret = bdrv_open_common(bs, file, options, flags, drv, &local_err);
1355 goto unlink_and_fail;
1358 if (file && (bs->file != file)) {
1363 /* If there is a backing file, use it */
1364 if ((flags & BDRV_O_NO_BACKING) == 0) {
1365 QDict *backing_options;
1367 qdict_extract_subqdict(options, &backing_options, "backing.");
1368 ret = bdrv_open_backing_file(bs, backing_options, &local_err);
1370 goto close_and_fail;
1374 /* Check if any unknown options were used */
1375 if (qdict_size(options) != 0) {
1376 const QDictEntry *entry = qdict_first(options);
1377 error_setg(errp, "Block format '%s' used by device '%s' doesn't "
1378 "support the option '%s'", drv->format_name, bs->device_name,
1382 goto close_and_fail;
1386 if (!bdrv_key_required(bs)) {
1387 bdrv_dev_change_media_cb(bs, true);
1396 if (bs->is_temporary) {
1400 QDECREF(bs->options);
1403 if (error_is_set(&local_err)) {
1404 error_propagate(errp, local_err);
1411 if (error_is_set(&local_err)) {
1412 error_propagate(errp, local_err);
1417 typedef struct BlockReopenQueueEntry {
1419 BDRVReopenState state;
1420 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
1421 } BlockReopenQueueEntry;
1424 * Adds a BlockDriverState to a simple queue for an atomic, transactional
1425 * reopen of multiple devices.
1427 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
1428 * already performed, or alternatively may be NULL a new BlockReopenQueue will
1429 * be created and initialized. This newly created BlockReopenQueue should be
1430 * passed back in for subsequent calls that are intended to be of the same
1433 * bs is the BlockDriverState to add to the reopen queue.
1435 * flags contains the open flags for the associated bs
1437 * returns a pointer to bs_queue, which is either the newly allocated
1438 * bs_queue, or the existing bs_queue being used.
1441 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
1442 BlockDriverState *bs, int flags)
1446 BlockReopenQueueEntry *bs_entry;
1447 if (bs_queue == NULL) {
1448 bs_queue = g_new0(BlockReopenQueue, 1);
1449 QSIMPLEQ_INIT(bs_queue);
1453 bdrv_reopen_queue(bs_queue, bs->file, flags);
1456 bs_entry = g_new0(BlockReopenQueueEntry, 1);
1457 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
1459 bs_entry->state.bs = bs;
1460 bs_entry->state.flags = flags;
1466 * Reopen multiple BlockDriverStates atomically & transactionally.
1468 * The queue passed in (bs_queue) must have been built up previous
1469 * via bdrv_reopen_queue().
1471 * Reopens all BDS specified in the queue, with the appropriate
1472 * flags. All devices are prepared for reopen, and failure of any
1473 * device will cause all device changes to be abandonded, and intermediate
1476 * If all devices prepare successfully, then the changes are committed
1480 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
1483 BlockReopenQueueEntry *bs_entry, *next;
1484 Error *local_err = NULL;
1486 assert(bs_queue != NULL);
1490 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
1491 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
1492 error_propagate(errp, local_err);
1495 bs_entry->prepared = true;
1498 /* If we reach this point, we have success and just need to apply the
1501 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
1502 bdrv_reopen_commit(&bs_entry->state);
1508 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
1509 if (ret && bs_entry->prepared) {
1510 bdrv_reopen_abort(&bs_entry->state);
1519 /* Reopen a single BlockDriverState with the specified flags. */
1520 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
1523 Error *local_err = NULL;
1524 BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, bdrv_flags);
1526 ret = bdrv_reopen_multiple(queue, &local_err);
1527 if (local_err != NULL) {
1528 error_propagate(errp, local_err);
1535 * Prepares a BlockDriverState for reopen. All changes are staged in the
1536 * 'opaque' field of the BDRVReopenState, which is used and allocated by
1537 * the block driver layer .bdrv_reopen_prepare()
1539 * bs is the BlockDriverState to reopen
1540 * flags are the new open flags
1541 * queue is the reopen queue
1543 * Returns 0 on success, non-zero on error. On error errp will be set
1546 * On failure, bdrv_reopen_abort() will be called to clean up any data.
1547 * It is the responsibility of the caller to then call the abort() or
1548 * commit() for any other BDS that have been left in a prepare() state
1551 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
1555 Error *local_err = NULL;
1558 assert(reopen_state != NULL);
1559 assert(reopen_state->bs->drv != NULL);
1560 drv = reopen_state->bs->drv;
1562 /* if we are to stay read-only, do not allow permission change
1564 if (!(reopen_state->bs->open_flags & BDRV_O_ALLOW_RDWR) &&
1565 reopen_state->flags & BDRV_O_RDWR) {
1566 error_set(errp, QERR_DEVICE_IS_READ_ONLY,
1567 reopen_state->bs->device_name);
1572 ret = bdrv_flush(reopen_state->bs);
1574 error_set(errp, ERROR_CLASS_GENERIC_ERROR, "Error (%s) flushing drive",
1579 if (drv->bdrv_reopen_prepare) {
1580 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
1582 if (local_err != NULL) {
1583 error_propagate(errp, local_err);
1585 error_setg(errp, "failed while preparing to reopen image '%s'",
1586 reopen_state->bs->filename);
1591 /* It is currently mandatory to have a bdrv_reopen_prepare()
1592 * handler for each supported drv. */
1593 error_set(errp, QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
1594 drv->format_name, reopen_state->bs->device_name,
1595 "reopening of file");
1607 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
1608 * makes them final by swapping the staging BlockDriverState contents into
1609 * the active BlockDriverState contents.
1611 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
1615 assert(reopen_state != NULL);
1616 drv = reopen_state->bs->drv;
1617 assert(drv != NULL);
1619 /* If there are any driver level actions to take */
1620 if (drv->bdrv_reopen_commit) {
1621 drv->bdrv_reopen_commit(reopen_state);
1624 /* set BDS specific flags now */
1625 reopen_state->bs->open_flags = reopen_state->flags;
1626 reopen_state->bs->enable_write_cache = !!(reopen_state->flags &
1628 reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
1630 bdrv_refresh_limits(reopen_state->bs);
1634 * Abort the reopen, and delete and free the staged changes in
1637 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
1641 assert(reopen_state != NULL);
1642 drv = reopen_state->bs->drv;
1643 assert(drv != NULL);
1645 if (drv->bdrv_reopen_abort) {
1646 drv->bdrv_reopen_abort(reopen_state);
1651 void bdrv_close(BlockDriverState *bs)
1654 block_job_cancel_sync(bs->job);
1656 bdrv_drain_all(); /* complete I/O */
1658 bdrv_drain_all(); /* in case flush left pending I/O */
1659 notifier_list_notify(&bs->close_notifiers, bs);
1662 if (bs->backing_hd) {
1663 bdrv_unref(bs->backing_hd);
1664 bs->backing_hd = NULL;
1666 bs->drv->bdrv_close(bs);
1669 if (bs->is_temporary) {
1670 unlink(bs->filename);
1675 bs->copy_on_read = 0;
1676 bs->backing_file[0] = '\0';
1677 bs->backing_format[0] = '\0';
1678 bs->total_sectors = 0;
1683 bs->zero_beyond_eof = false;
1684 QDECREF(bs->options);
1687 if (bs->file != NULL) {
1688 bdrv_unref(bs->file);
1693 bdrv_dev_change_media_cb(bs, false);
1695 /*throttling disk I/O limits*/
1696 if (bs->io_limits_enabled) {
1697 bdrv_io_limits_disable(bs);
1701 void bdrv_close_all(void)
1703 BlockDriverState *bs;
1705 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
1710 /* Check if any requests are in-flight (including throttled requests) */
1711 static bool bdrv_requests_pending(BlockDriverState *bs)
1713 if (!QLIST_EMPTY(&bs->tracked_requests)) {
1716 if (!qemu_co_queue_empty(&bs->throttled_reqs[0])) {
1719 if (!qemu_co_queue_empty(&bs->throttled_reqs[1])) {
1722 if (bs->file && bdrv_requests_pending(bs->file)) {
1725 if (bs->backing_hd && bdrv_requests_pending(bs->backing_hd)) {
1731 static bool bdrv_requests_pending_all(void)
1733 BlockDriverState *bs;
1734 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
1735 if (bdrv_requests_pending(bs)) {
1743 * Wait for pending requests to complete across all BlockDriverStates
1745 * This function does not flush data to disk, use bdrv_flush_all() for that
1746 * after calling this function.
1748 * Note that completion of an asynchronous I/O operation can trigger any
1749 * number of other I/O operations on other devices---for example a coroutine
1750 * can be arbitrarily complex and a constant flow of I/O can come until the
1751 * coroutine is complete. Because of this, it is not possible to have a
1752 * function to drain a single device's I/O queue.
1754 void bdrv_drain_all(void)
1756 /* Always run first iteration so any pending completion BHs run */
1758 BlockDriverState *bs;
1761 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
1762 bdrv_start_throttled_reqs(bs);
1765 busy = bdrv_requests_pending_all();
1766 busy |= aio_poll(qemu_get_aio_context(), busy);
1770 /* make a BlockDriverState anonymous by removing from bdrv_state and
1771 * graph_bdrv_state list.
1772 Also, NULL terminate the device_name to prevent double remove */
1773 void bdrv_make_anon(BlockDriverState *bs)
1775 if (bs->device_name[0] != '\0') {
1776 QTAILQ_REMOVE(&bdrv_states, bs, device_list);
1778 bs->device_name[0] = '\0';
1779 if (bs->node_name[0] != '\0') {
1780 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
1782 bs->node_name[0] = '\0';
1785 static void bdrv_rebind(BlockDriverState *bs)
1787 if (bs->drv && bs->drv->bdrv_rebind) {
1788 bs->drv->bdrv_rebind(bs);
1792 static void bdrv_move_feature_fields(BlockDriverState *bs_dest,
1793 BlockDriverState *bs_src)
1795 /* move some fields that need to stay attached to the device */
1796 bs_dest->open_flags = bs_src->open_flags;
1799 bs_dest->dev_ops = bs_src->dev_ops;
1800 bs_dest->dev_opaque = bs_src->dev_opaque;
1801 bs_dest->dev = bs_src->dev;
1802 bs_dest->guest_block_size = bs_src->guest_block_size;
1803 bs_dest->copy_on_read = bs_src->copy_on_read;
1805 bs_dest->enable_write_cache = bs_src->enable_write_cache;
1807 /* i/o throttled req */
1808 memcpy(&bs_dest->throttle_state,
1809 &bs_src->throttle_state,
1810 sizeof(ThrottleState));
1811 bs_dest->throttled_reqs[0] = bs_src->throttled_reqs[0];
1812 bs_dest->throttled_reqs[1] = bs_src->throttled_reqs[1];
1813 bs_dest->io_limits_enabled = bs_src->io_limits_enabled;
1816 bs_dest->on_read_error = bs_src->on_read_error;
1817 bs_dest->on_write_error = bs_src->on_write_error;
1820 bs_dest->iostatus_enabled = bs_src->iostatus_enabled;
1821 bs_dest->iostatus = bs_src->iostatus;
1824 bs_dest->dirty_bitmaps = bs_src->dirty_bitmaps;
1826 /* reference count */
1827 bs_dest->refcnt = bs_src->refcnt;
1830 bs_dest->in_use = bs_src->in_use;
1831 bs_dest->job = bs_src->job;
1833 /* keep the same entry in bdrv_states */
1834 pstrcpy(bs_dest->device_name, sizeof(bs_dest->device_name),
1835 bs_src->device_name);
1836 bs_dest->device_list = bs_src->device_list;
1838 /* keep the same entry in graph_bdrv_states
1839 * We do want to swap name but don't want to swap linked list entries
1841 bs_dest->node_list = bs_src->node_list;
1845 * Swap bs contents for two image chains while they are live,
1846 * while keeping required fields on the BlockDriverState that is
1847 * actually attached to a device.
1849 * This will modify the BlockDriverState fields, and swap contents
1850 * between bs_new and bs_old. Both bs_new and bs_old are modified.
1852 * bs_new is required to be anonymous.
1854 * This function does not create any image files.
1856 void bdrv_swap(BlockDriverState *bs_new, BlockDriverState *bs_old)
1858 BlockDriverState tmp;
1860 /* bs_new must be anonymous and shouldn't have anything fancy enabled */
1861 assert(bs_new->device_name[0] == '\0');
1862 assert(QLIST_EMPTY(&bs_new->dirty_bitmaps));
1863 assert(bs_new->job == NULL);
1864 assert(bs_new->dev == NULL);
1865 assert(bs_new->in_use == 0);
1866 assert(bs_new->io_limits_enabled == false);
1867 assert(!throttle_have_timer(&bs_new->throttle_state));
1873 /* there are some fields that should not be swapped, move them back */
1874 bdrv_move_feature_fields(&tmp, bs_old);
1875 bdrv_move_feature_fields(bs_old, bs_new);
1876 bdrv_move_feature_fields(bs_new, &tmp);
1878 /* bs_new shouldn't be in bdrv_states even after the swap! */
1879 assert(bs_new->device_name[0] == '\0');
1881 /* Check a few fields that should remain attached to the device */
1882 assert(bs_new->dev == NULL);
1883 assert(bs_new->job == NULL);
1884 assert(bs_new->in_use == 0);
1885 assert(bs_new->io_limits_enabled == false);
1886 assert(!throttle_have_timer(&bs_new->throttle_state));
1888 bdrv_rebind(bs_new);
1889 bdrv_rebind(bs_old);
1893 * Add new bs contents at the top of an image chain while the chain is
1894 * live, while keeping required fields on the top layer.
1896 * This will modify the BlockDriverState fields, and swap contents
1897 * between bs_new and bs_top. Both bs_new and bs_top are modified.
1899 * bs_new is required to be anonymous.
1901 * This function does not create any image files.
1903 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
1905 bdrv_swap(bs_new, bs_top);
1907 /* The contents of 'tmp' will become bs_top, as we are
1908 * swapping bs_new and bs_top contents. */
1909 bs_top->backing_hd = bs_new;
1910 bs_top->open_flags &= ~BDRV_O_NO_BACKING;
1911 pstrcpy(bs_top->backing_file, sizeof(bs_top->backing_file),
1913 pstrcpy(bs_top->backing_format, sizeof(bs_top->backing_format),
1914 bs_new->drv ? bs_new->drv->format_name : "");
1917 static void bdrv_delete(BlockDriverState *bs)
1921 assert(!bs->in_use);
1922 assert(!bs->refcnt);
1923 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
1927 /* remove from list, if necessary */
1933 int bdrv_attach_dev(BlockDriverState *bs, void *dev)
1934 /* TODO change to DeviceState *dev when all users are qdevified */
1940 bdrv_iostatus_reset(bs);
1944 /* TODO qdevified devices don't use this, remove when devices are qdevified */
1945 void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev)
1947 if (bdrv_attach_dev(bs, dev) < 0) {
1952 void bdrv_detach_dev(BlockDriverState *bs, void *dev)
1953 /* TODO change to DeviceState *dev when all users are qdevified */
1955 assert(bs->dev == dev);
1958 bs->dev_opaque = NULL;
1959 bs->guest_block_size = 512;
1962 /* TODO change to return DeviceState * when all users are qdevified */
1963 void *bdrv_get_attached_dev(BlockDriverState *bs)
1968 void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops,
1972 bs->dev_opaque = opaque;
1975 void bdrv_emit_qmp_error_event(const BlockDriverState *bdrv,
1976 enum MonitorEvent ev,
1977 BlockErrorAction action, bool is_read)
1980 const char *action_str;
1983 case BDRV_ACTION_REPORT:
1984 action_str = "report";
1986 case BDRV_ACTION_IGNORE:
1987 action_str = "ignore";
1989 case BDRV_ACTION_STOP:
1990 action_str = "stop";
1996 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1999 is_read ? "read" : "write");
2000 monitor_protocol_event(ev, data);
2002 qobject_decref(data);
2005 static void bdrv_emit_qmp_eject_event(BlockDriverState *bs, bool ejected)
2009 data = qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }",
2010 bdrv_get_device_name(bs), ejected);
2011 monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED, data);
2013 qobject_decref(data);
2016 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load)
2018 if (bs->dev_ops && bs->dev_ops->change_media_cb) {
2019 bool tray_was_closed = !bdrv_dev_is_tray_open(bs);
2020 bs->dev_ops->change_media_cb(bs->dev_opaque, load);
2021 if (tray_was_closed) {
2023 bdrv_emit_qmp_eject_event(bs, true);
2027 bdrv_emit_qmp_eject_event(bs, false);
2032 bool bdrv_dev_has_removable_media(BlockDriverState *bs)
2034 return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb);
2037 void bdrv_dev_eject_request(BlockDriverState *bs, bool force)
2039 if (bs->dev_ops && bs->dev_ops->eject_request_cb) {
2040 bs->dev_ops->eject_request_cb(bs->dev_opaque, force);
2044 bool bdrv_dev_is_tray_open(BlockDriverState *bs)
2046 if (bs->dev_ops && bs->dev_ops->is_tray_open) {
2047 return bs->dev_ops->is_tray_open(bs->dev_opaque);
2052 static void bdrv_dev_resize_cb(BlockDriverState *bs)
2054 if (bs->dev_ops && bs->dev_ops->resize_cb) {
2055 bs->dev_ops->resize_cb(bs->dev_opaque);
2059 bool bdrv_dev_is_medium_locked(BlockDriverState *bs)
2061 if (bs->dev_ops && bs->dev_ops->is_medium_locked) {
2062 return bs->dev_ops->is_medium_locked(bs->dev_opaque);
2068 * Run consistency checks on an image
2070 * Returns 0 if the check could be completed (it doesn't mean that the image is
2071 * free of errors) or -errno when an internal error occurred. The results of the
2072 * check are stored in res.
2074 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix)
2076 if (bs->drv->bdrv_check == NULL) {
2080 memset(res, 0, sizeof(*res));
2081 return bs->drv->bdrv_check(bs, res, fix);
2084 #define COMMIT_BUF_SECTORS 2048
2086 /* commit COW file into the raw image */
2087 int bdrv_commit(BlockDriverState *bs)
2089 BlockDriver *drv = bs->drv;
2090 int64_t sector, total_sectors, length, backing_length;
2091 int n, ro, open_flags;
2093 uint8_t *buf = NULL;
2094 char filename[PATH_MAX];
2099 if (!bs->backing_hd) {
2103 if (bdrv_in_use(bs) || bdrv_in_use(bs->backing_hd)) {
2107 ro = bs->backing_hd->read_only;
2108 /* Use pstrcpy (not strncpy): filename must be NUL-terminated. */
2109 pstrcpy(filename, sizeof(filename), bs->backing_hd->filename);
2110 open_flags = bs->backing_hd->open_flags;
2113 if (bdrv_reopen(bs->backing_hd, open_flags | BDRV_O_RDWR, NULL)) {
2118 length = bdrv_getlength(bs);
2124 backing_length = bdrv_getlength(bs->backing_hd);
2125 if (backing_length < 0) {
2126 ret = backing_length;
2130 /* If our top snapshot is larger than the backing file image,
2131 * grow the backing file image if possible. If not possible,
2132 * we must return an error */
2133 if (length > backing_length) {
2134 ret = bdrv_truncate(bs->backing_hd, length);
2140 total_sectors = length >> BDRV_SECTOR_BITS;
2141 buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
2143 for (sector = 0; sector < total_sectors; sector += n) {
2144 ret = bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n);
2149 ret = bdrv_read(bs, sector, buf, n);
2154 ret = bdrv_write(bs->backing_hd, sector, buf, n);
2161 if (drv->bdrv_make_empty) {
2162 ret = drv->bdrv_make_empty(bs);
2170 * Make sure all data we wrote to the backing device is actually
2173 if (bs->backing_hd) {
2174 bdrv_flush(bs->backing_hd);
2182 /* ignoring error return here */
2183 bdrv_reopen(bs->backing_hd, open_flags & ~BDRV_O_RDWR, NULL);
2189 int bdrv_commit_all(void)
2191 BlockDriverState *bs;
2193 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
2194 if (bs->drv && bs->backing_hd) {
2195 int ret = bdrv_commit(bs);
2205 * Remove an active request from the tracked requests list
2207 * This function should be called when a tracked request is completing.
2209 static void tracked_request_end(BdrvTrackedRequest *req)
2211 QLIST_REMOVE(req, list);
2212 qemu_co_queue_restart_all(&req->wait_queue);
2216 * Add an active request to the tracked requests list
2218 static void tracked_request_begin(BdrvTrackedRequest *req,
2219 BlockDriverState *bs,
2221 int nb_sectors, bool is_write)
2223 *req = (BdrvTrackedRequest){
2225 .sector_num = sector_num,
2226 .nb_sectors = nb_sectors,
2227 .is_write = is_write,
2228 .co = qemu_coroutine_self(),
2231 qemu_co_queue_init(&req->wait_queue);
2233 QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
2237 * Round a region to cluster boundaries
2239 void bdrv_round_to_clusters(BlockDriverState *bs,
2240 int64_t sector_num, int nb_sectors,
2241 int64_t *cluster_sector_num,
2242 int *cluster_nb_sectors)
2244 BlockDriverInfo bdi;
2246 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
2247 *cluster_sector_num = sector_num;
2248 *cluster_nb_sectors = nb_sectors;
2250 int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE;
2251 *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c);
2252 *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num +
2257 static bool tracked_request_overlaps(BdrvTrackedRequest *req,
2258 int64_t sector_num, int nb_sectors) {
2260 if (sector_num >= req->sector_num + req->nb_sectors) {
2264 if (req->sector_num >= sector_num + nb_sectors) {
2270 static void coroutine_fn wait_for_overlapping_requests(BlockDriverState *bs,
2271 int64_t sector_num, int nb_sectors)
2273 BdrvTrackedRequest *req;
2274 int64_t cluster_sector_num;
2275 int cluster_nb_sectors;
2278 /* If we touch the same cluster it counts as an overlap. This guarantees
2279 * that allocating writes will be serialized and not race with each other
2280 * for the same cluster. For example, in copy-on-read it ensures that the
2281 * CoR read and write operations are atomic and guest writes cannot
2282 * interleave between them.
2284 bdrv_round_to_clusters(bs, sector_num, nb_sectors,
2285 &cluster_sector_num, &cluster_nb_sectors);
2289 QLIST_FOREACH(req, &bs->tracked_requests, list) {
2290 if (tracked_request_overlaps(req, cluster_sector_num,
2291 cluster_nb_sectors)) {
2292 /* Hitting this means there was a reentrant request, for
2293 * example, a block driver issuing nested requests. This must
2294 * never happen since it means deadlock.
2296 assert(qemu_coroutine_self() != req->co);
2298 qemu_co_queue_wait(&req->wait_queue);
2309 * -EINVAL - backing format specified, but no file
2310 * -ENOSPC - can't update the backing file because no space is left in the
2312 * -ENOTSUP - format driver doesn't support changing the backing file
2314 int bdrv_change_backing_file(BlockDriverState *bs,
2315 const char *backing_file, const char *backing_fmt)
2317 BlockDriver *drv = bs->drv;
2320 /* Backing file format doesn't make sense without a backing file */
2321 if (backing_fmt && !backing_file) {
2325 if (drv->bdrv_change_backing_file != NULL) {
2326 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
2332 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
2333 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
2339 * Finds the image layer in the chain that has 'bs' as its backing file.
2341 * active is the current topmost image.
2343 * Returns NULL if bs is not found in active's image chain,
2344 * or if active == bs.
2346 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
2347 BlockDriverState *bs)
2349 BlockDriverState *overlay = NULL;
2350 BlockDriverState *intermediate;
2352 assert(active != NULL);
2355 /* if bs is the same as active, then by definition it has no overlay
2361 intermediate = active;
2362 while (intermediate->backing_hd) {
2363 if (intermediate->backing_hd == bs) {
2364 overlay = intermediate;
2367 intermediate = intermediate->backing_hd;
2373 typedef struct BlkIntermediateStates {
2374 BlockDriverState *bs;
2375 QSIMPLEQ_ENTRY(BlkIntermediateStates) entry;
2376 } BlkIntermediateStates;
2380 * Drops images above 'base' up to and including 'top', and sets the image
2381 * above 'top' to have base as its backing file.
2383 * Requires that the overlay to 'top' is opened r/w, so that the backing file
2384 * information in 'bs' can be properly updated.
2386 * E.g., this will convert the following chain:
2387 * bottom <- base <- intermediate <- top <- active
2391 * bottom <- base <- active
2393 * It is allowed for bottom==base, in which case it converts:
2395 * base <- intermediate <- top <- active
2402 * if active == top, that is considered an error
2405 int bdrv_drop_intermediate(BlockDriverState *active, BlockDriverState *top,
2406 BlockDriverState *base)
2408 BlockDriverState *intermediate;
2409 BlockDriverState *base_bs = NULL;
2410 BlockDriverState *new_top_bs = NULL;
2411 BlkIntermediateStates *intermediate_state, *next;
2414 QSIMPLEQ_HEAD(states_to_delete, BlkIntermediateStates) states_to_delete;
2415 QSIMPLEQ_INIT(&states_to_delete);
2417 if (!top->drv || !base->drv) {
2421 new_top_bs = bdrv_find_overlay(active, top);
2423 if (new_top_bs == NULL) {
2424 /* we could not find the image above 'top', this is an error */
2428 /* special case of new_top_bs->backing_hd already pointing to base - nothing
2429 * to do, no intermediate images */
2430 if (new_top_bs->backing_hd == base) {
2437 /* now we will go down through the list, and add each BDS we find
2438 * into our deletion queue, until we hit the 'base'
2440 while (intermediate) {
2441 intermediate_state = g_malloc0(sizeof(BlkIntermediateStates));
2442 intermediate_state->bs = intermediate;
2443 QSIMPLEQ_INSERT_TAIL(&states_to_delete, intermediate_state, entry);
2445 if (intermediate->backing_hd == base) {
2446 base_bs = intermediate->backing_hd;
2449 intermediate = intermediate->backing_hd;
2451 if (base_bs == NULL) {
2452 /* something went wrong, we did not end at the base. safely
2453 * unravel everything, and exit with error */
2457 /* success - we can delete the intermediate states, and link top->base */
2458 ret = bdrv_change_backing_file(new_top_bs, base_bs->filename,
2459 base_bs->drv ? base_bs->drv->format_name : "");
2463 new_top_bs->backing_hd = base_bs;
2465 bdrv_refresh_limits(new_top_bs);
2467 QSIMPLEQ_FOREACH_SAFE(intermediate_state, &states_to_delete, entry, next) {
2468 /* so that bdrv_close() does not recursively close the chain */
2469 intermediate_state->bs->backing_hd = NULL;
2470 bdrv_unref(intermediate_state->bs);
2475 QSIMPLEQ_FOREACH_SAFE(intermediate_state, &states_to_delete, entry, next) {
2476 g_free(intermediate_state);
2482 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
2487 if (!bdrv_is_inserted(bs))
2493 len = bdrv_getlength(bs);
2498 if ((offset > len) || (len - offset < size))
2504 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
2507 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
2508 nb_sectors * BDRV_SECTOR_SIZE);
2511 typedef struct RwCo {
2512 BlockDriverState *bs;
2518 BdrvRequestFlags flags;
2521 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
2523 RwCo *rwco = opaque;
2525 if (!rwco->is_write) {
2526 rwco->ret = bdrv_co_do_readv(rwco->bs, rwco->sector_num,
2527 rwco->nb_sectors, rwco->qiov,
2530 rwco->ret = bdrv_co_do_writev(rwco->bs, rwco->sector_num,
2531 rwco->nb_sectors, rwco->qiov,
2537 * Process a vectored synchronous request using coroutines
2539 static int bdrv_rwv_co(BlockDriverState *bs, int64_t sector_num,
2540 QEMUIOVector *qiov, bool is_write,
2541 BdrvRequestFlags flags)
2546 .sector_num = sector_num,
2547 .nb_sectors = qiov->size >> BDRV_SECTOR_BITS,
2549 .is_write = is_write,
2553 assert((qiov->size & (BDRV_SECTOR_SIZE - 1)) == 0);
2556 * In sync call context, when the vcpu is blocked, this throttling timer
2557 * will not fire; so the I/O throttling function has to be disabled here
2558 * if it has been enabled.
2560 if (bs->io_limits_enabled) {
2561 fprintf(stderr, "Disabling I/O throttling on '%s' due "
2562 "to synchronous I/O.\n", bdrv_get_device_name(bs));
2563 bdrv_io_limits_disable(bs);
2566 if (qemu_in_coroutine()) {
2567 /* Fast-path if already in coroutine context */
2568 bdrv_rw_co_entry(&rwco);
2570 co = qemu_coroutine_create(bdrv_rw_co_entry);
2571 qemu_coroutine_enter(co, &rwco);
2572 while (rwco.ret == NOT_DONE) {
2580 * Process a synchronous request using coroutines
2582 static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf,
2583 int nb_sectors, bool is_write, BdrvRequestFlags flags)
2586 struct iovec iov = {
2587 .iov_base = (void *)buf,
2588 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
2591 qemu_iovec_init_external(&qiov, &iov, 1);
2592 return bdrv_rwv_co(bs, sector_num, &qiov, is_write, flags);
2595 /* return < 0 if error. See bdrv_write() for the return codes */
2596 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
2597 uint8_t *buf, int nb_sectors)
2599 return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false, 0);
2602 /* Just like bdrv_read(), but with I/O throttling temporarily disabled */
2603 int bdrv_read_unthrottled(BlockDriverState *bs, int64_t sector_num,
2604 uint8_t *buf, int nb_sectors)
2609 enabled = bs->io_limits_enabled;
2610 bs->io_limits_enabled = false;
2611 ret = bdrv_read(bs, sector_num, buf, nb_sectors);
2612 bs->io_limits_enabled = enabled;
2616 /* Return < 0 if error. Important errors are:
2617 -EIO generic I/O error (may happen for all errors)
2618 -ENOMEDIUM No media inserted.
2619 -EINVAL Invalid sector number or nb_sectors
2620 -EACCES Trying to write a read-only device
2622 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
2623 const uint8_t *buf, int nb_sectors)
2625 return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true, 0);
2628 int bdrv_writev(BlockDriverState *bs, int64_t sector_num, QEMUIOVector *qiov)
2630 return bdrv_rwv_co(bs, sector_num, qiov, true, 0);
2633 int bdrv_write_zeroes(BlockDriverState *bs, int64_t sector_num,
2634 int nb_sectors, BdrvRequestFlags flags)
2636 return bdrv_rw_co(bs, sector_num, NULL, nb_sectors, true,
2637 BDRV_REQ_ZERO_WRITE | flags);
2641 * Completely zero out a block device with the help of bdrv_write_zeroes.
2642 * The operation is sped up by checking the block status and only writing
2643 * zeroes to the device if they currently do not return zeroes. Optional
2644 * flags are passed through to bdrv_write_zeroes (e.g. BDRV_REQ_MAY_UNMAP).
2646 * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
2648 int bdrv_make_zero(BlockDriverState *bs, BdrvRequestFlags flags)
2650 int64_t target_size = bdrv_getlength(bs) / BDRV_SECTOR_SIZE;
2651 int64_t ret, nb_sectors, sector_num = 0;
2655 nb_sectors = target_size - sector_num;
2656 if (nb_sectors <= 0) {
2659 if (nb_sectors > INT_MAX) {
2660 nb_sectors = INT_MAX;
2662 ret = bdrv_get_block_status(bs, sector_num, nb_sectors, &n);
2664 error_report("error getting block status at sector %" PRId64 ": %s",
2665 sector_num, strerror(-ret));
2668 if (ret & BDRV_BLOCK_ZERO) {
2672 ret = bdrv_write_zeroes(bs, sector_num, n, flags);
2674 error_report("error writing zeroes at sector %" PRId64 ": %s",
2675 sector_num, strerror(-ret));
2682 int bdrv_pread(BlockDriverState *bs, int64_t offset,
2683 void *buf, int count1)
2685 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
2686 int len, nb_sectors, count;
2691 /* first read to align to sector start */
2692 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
2695 sector_num = offset >> BDRV_SECTOR_BITS;
2697 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
2699 memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
2707 /* read the sectors "in place" */
2708 nb_sectors = count >> BDRV_SECTOR_BITS;
2709 if (nb_sectors > 0) {
2710 if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
2712 sector_num += nb_sectors;
2713 len = nb_sectors << BDRV_SECTOR_BITS;
2718 /* add data from the last sector */
2720 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
2722 memcpy(buf, tmp_buf, count);
2727 int bdrv_pwritev(BlockDriverState *bs, int64_t offset, QEMUIOVector *qiov)
2729 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
2730 int len, nb_sectors, count;
2736 /* first write to align to sector start */
2737 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
2740 sector_num = offset >> BDRV_SECTOR_BITS;
2742 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
2744 qemu_iovec_to_buf(qiov, 0, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)),
2746 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
2754 /* write the sectors "in place" */
2755 nb_sectors = count >> BDRV_SECTOR_BITS;
2756 if (nb_sectors > 0) {
2757 QEMUIOVector qiov_inplace;
2759 qemu_iovec_init(&qiov_inplace, qiov->niov);
2760 qemu_iovec_concat(&qiov_inplace, qiov, len,
2761 nb_sectors << BDRV_SECTOR_BITS);
2762 ret = bdrv_writev(bs, sector_num, &qiov_inplace);
2763 qemu_iovec_destroy(&qiov_inplace);
2768 sector_num += nb_sectors;
2769 len = nb_sectors << BDRV_SECTOR_BITS;
2773 /* add data from the last sector */
2775 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
2777 qemu_iovec_to_buf(qiov, qiov->size - count, tmp_buf, count);
2778 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
2784 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
2785 const void *buf, int count1)
2788 struct iovec iov = {
2789 .iov_base = (void *) buf,
2793 qemu_iovec_init_external(&qiov, &iov, 1);
2794 return bdrv_pwritev(bs, offset, &qiov);
2798 * Writes to the file and ensures that no writes are reordered across this
2799 * request (acts as a barrier)
2801 * Returns 0 on success, -errno in error cases.
2803 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
2804 const void *buf, int count)
2808 ret = bdrv_pwrite(bs, offset, buf, count);
2813 /* No flush needed for cache modes that already do it */
2814 if (bs->enable_write_cache) {
2821 static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs,
2822 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
2824 /* Perform I/O through a temporary buffer so that users who scribble over
2825 * their read buffer while the operation is in progress do not end up
2826 * modifying the image file. This is critical for zero-copy guest I/O
2827 * where anything might happen inside guest memory.
2829 void *bounce_buffer;
2831 BlockDriver *drv = bs->drv;
2833 QEMUIOVector bounce_qiov;
2834 int64_t cluster_sector_num;
2835 int cluster_nb_sectors;
2839 /* Cover entire cluster so no additional backing file I/O is required when
2840 * allocating cluster in the image file.
2842 bdrv_round_to_clusters(bs, sector_num, nb_sectors,
2843 &cluster_sector_num, &cluster_nb_sectors);
2845 trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors,
2846 cluster_sector_num, cluster_nb_sectors);
2848 iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE;
2849 iov.iov_base = bounce_buffer = qemu_blockalign(bs, iov.iov_len);
2850 qemu_iovec_init_external(&bounce_qiov, &iov, 1);
2852 ret = drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors,
2858 if (drv->bdrv_co_write_zeroes &&
2859 buffer_is_zero(bounce_buffer, iov.iov_len)) {
2860 ret = bdrv_co_do_write_zeroes(bs, cluster_sector_num,
2861 cluster_nb_sectors, 0);
2863 /* This does not change the data on the disk, it is not necessary
2864 * to flush even in cache=writethrough mode.
2866 ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors,
2871 /* It might be okay to ignore write errors for guest requests. If this
2872 * is a deliberate copy-on-read then we don't want to ignore the error.
2873 * Simply report it in all cases.
2878 skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE;
2879 qemu_iovec_from_buf(qiov, 0, bounce_buffer + skip_bytes,
2880 nb_sectors * BDRV_SECTOR_SIZE);
2883 qemu_vfree(bounce_buffer);
2888 * Forwards an already correctly aligned request to the BlockDriver. This
2889 * handles copy on read and zeroing after EOF; any other features must be
2890 * implemented by the caller.
2892 static int coroutine_fn bdrv_aligned_preadv(BlockDriverState *bs,
2893 int64_t offset, unsigned int bytes, QEMUIOVector *qiov, int flags)
2895 BlockDriver *drv = bs->drv;
2896 BdrvTrackedRequest req;
2899 int64_t sector_num = offset >> BDRV_SECTOR_BITS;
2900 unsigned int nb_sectors = bytes >> BDRV_SECTOR_BITS;
2902 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
2903 assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
2905 /* Handle Copy on Read and associated serialisation */
2906 if (flags & BDRV_REQ_COPY_ON_READ) {
2907 bs->copy_on_read_in_flight++;
2910 if (bs->copy_on_read_in_flight) {
2911 wait_for_overlapping_requests(bs, sector_num, nb_sectors);
2914 tracked_request_begin(&req, bs, sector_num, nb_sectors, false);
2916 if (flags & BDRV_REQ_COPY_ON_READ) {
2919 ret = bdrv_is_allocated(bs, sector_num, nb_sectors, &pnum);
2924 if (!ret || pnum != nb_sectors) {
2925 ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov);
2930 /* Forward the request to the BlockDriver */
2931 if (!(bs->zero_beyond_eof && bs->growable)) {
2932 ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
2934 /* Read zeros after EOF of growable BDSes */
2935 int64_t len, total_sectors, max_nb_sectors;
2937 len = bdrv_getlength(bs);
2943 total_sectors = DIV_ROUND_UP(len, BDRV_SECTOR_SIZE);
2944 max_nb_sectors = MAX(0, total_sectors - sector_num);
2945 if (max_nb_sectors > 0) {
2946 ret = drv->bdrv_co_readv(bs, sector_num,
2947 MIN(nb_sectors, max_nb_sectors), qiov);
2952 /* Reading beyond end of file is supposed to produce zeroes */
2953 if (ret == 0 && total_sectors < sector_num + nb_sectors) {
2954 uint64_t offset = MAX(0, total_sectors - sector_num);
2955 uint64_t bytes = (sector_num + nb_sectors - offset) *
2957 qemu_iovec_memset(qiov, offset * BDRV_SECTOR_SIZE, 0, bytes);
2962 tracked_request_end(&req);
2964 if (flags & BDRV_REQ_COPY_ON_READ) {
2965 bs->copy_on_read_in_flight--;
2972 * Handle a read request in coroutine context
2974 static int coroutine_fn bdrv_co_do_preadv(BlockDriverState *bs,
2975 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
2976 BdrvRequestFlags flags)
2978 BlockDriver *drv = bs->drv;
2979 /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
2980 uint64_t align = MAX(BDRV_SECTOR_SIZE, bs->request_alignment);
2981 uint8_t *head_buf = NULL;
2982 uint8_t *tail_buf = NULL;
2983 QEMUIOVector local_qiov;
2984 bool use_local_qiov = false;
2990 if (bdrv_check_byte_request(bs, offset, bytes)) {
2994 if (bs->copy_on_read) {
2995 flags |= BDRV_REQ_COPY_ON_READ;
2998 /* throttling disk I/O */
2999 if (bs->io_limits_enabled) {
3000 /* TODO Switch to byte granularity */
3001 bdrv_io_limits_intercept(bs, bytes >> BDRV_SECTOR_BITS, false);
3004 /* Align read if necessary by padding qiov */
3005 if (offset & (align - 1)) {
3006 head_buf = qemu_blockalign(bs, align);
3007 qemu_iovec_init(&local_qiov, qiov->niov + 2);
3008 qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
3009 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
3010 use_local_qiov = true;
3012 bytes += offset & (align - 1);
3013 offset = offset & ~(align - 1);
3016 if ((offset + bytes) & (align - 1)) {
3017 if (!use_local_qiov) {
3018 qemu_iovec_init(&local_qiov, qiov->niov + 1);
3019 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
3020 use_local_qiov = true;
3022 tail_buf = qemu_blockalign(bs, align);
3023 qemu_iovec_add(&local_qiov, tail_buf,
3024 align - ((offset + bytes) & (align - 1)));
3026 bytes = ROUND_UP(bytes, align);
3029 ret = bdrv_aligned_preadv(bs, offset, bytes,
3030 use_local_qiov ? &local_qiov : qiov,
3033 if (use_local_qiov) {
3034 qemu_iovec_destroy(&local_qiov);
3035 qemu_vfree(head_buf);
3036 qemu_vfree(tail_buf);
3042 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
3043 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
3044 BdrvRequestFlags flags)
3046 if (nb_sectors < 0 || nb_sectors > (UINT_MAX >> BDRV_SECTOR_BITS)) {
3050 return bdrv_co_do_preadv(bs, sector_num << BDRV_SECTOR_BITS,
3051 nb_sectors << BDRV_SECTOR_BITS, qiov, flags);
3054 int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
3055 int nb_sectors, QEMUIOVector *qiov)
3057 trace_bdrv_co_readv(bs, sector_num, nb_sectors);
3059 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0);
3062 int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
3063 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
3065 trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors);
3067 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
3068 BDRV_REQ_COPY_ON_READ);
3071 /* if no limit is specified in the BlockLimits use a default
3072 * of 32768 512-byte sectors (16 MiB) per request.
3074 #define MAX_WRITE_ZEROES_DEFAULT 32768
3076 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
3077 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
3079 BlockDriver *drv = bs->drv;
3081 struct iovec iov = {0};
3084 int max_write_zeroes = bs->bl.max_write_zeroes ?
3085 bs->bl.max_write_zeroes : MAX_WRITE_ZEROES_DEFAULT;
3087 while (nb_sectors > 0 && !ret) {
3088 int num = nb_sectors;
3090 /* Align request. Block drivers can expect the "bulk" of the request
3093 if (bs->bl.write_zeroes_alignment
3094 && num > bs->bl.write_zeroes_alignment) {
3095 if (sector_num % bs->bl.write_zeroes_alignment != 0) {
3096 /* Make a small request up to the first aligned sector. */
3097 num = bs->bl.write_zeroes_alignment;
3098 num -= sector_num % bs->bl.write_zeroes_alignment;
3099 } else if ((sector_num + num) % bs->bl.write_zeroes_alignment != 0) {
3100 /* Shorten the request to the last aligned sector. num cannot
3101 * underflow because num > bs->bl.write_zeroes_alignment.
3103 num -= (sector_num + num) % bs->bl.write_zeroes_alignment;
3107 /* limit request size */
3108 if (num > max_write_zeroes) {
3109 num = max_write_zeroes;
3113 /* First try the efficient write zeroes operation */
3114 if (drv->bdrv_co_write_zeroes) {
3115 ret = drv->bdrv_co_write_zeroes(bs, sector_num, num, flags);
3118 if (ret == -ENOTSUP) {
3119 /* Fall back to bounce buffer if write zeroes is unsupported */
3120 iov.iov_len = num * BDRV_SECTOR_SIZE;
3121 if (iov.iov_base == NULL) {
3122 iov.iov_base = qemu_blockalign(bs, num * BDRV_SECTOR_SIZE);
3123 memset(iov.iov_base, 0, num * BDRV_SECTOR_SIZE);
3125 qemu_iovec_init_external(&qiov, &iov, 1);
3127 ret = drv->bdrv_co_writev(bs, sector_num, num, &qiov);
3129 /* Keep bounce buffer around if it is big enough for all
3130 * all future requests.
3132 if (num < max_write_zeroes) {
3133 qemu_vfree(iov.iov_base);
3134 iov.iov_base = NULL;
3142 qemu_vfree(iov.iov_base);
3147 * Forwards an already correctly aligned write request to the BlockDriver.
3149 static int coroutine_fn bdrv_aligned_pwritev(BlockDriverState *bs,
3150 int64_t offset, unsigned int bytes, QEMUIOVector *qiov, int flags)
3152 BlockDriver *drv = bs->drv;
3153 BdrvTrackedRequest req;
3156 int64_t sector_num = offset >> BDRV_SECTOR_BITS;
3157 unsigned int nb_sectors = bytes >> BDRV_SECTOR_BITS;
3159 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
3160 assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
3162 if (bs->copy_on_read_in_flight) {
3163 wait_for_overlapping_requests(bs, sector_num, nb_sectors);
3166 tracked_request_begin(&req, bs, sector_num, nb_sectors, true);
3168 ret = notifier_with_return_list_notify(&bs->before_write_notifiers, &req);
3171 /* Do nothing, write notifier decided to fail this request */
3172 } else if (flags & BDRV_REQ_ZERO_WRITE) {
3173 ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors, flags);
3175 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
3178 if (ret == 0 && !bs->enable_write_cache) {
3179 ret = bdrv_co_flush(bs);
3182 bdrv_set_dirty(bs, sector_num, nb_sectors);
3184 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
3185 bs->wr_highest_sector = sector_num + nb_sectors - 1;
3187 if (bs->growable && ret >= 0) {
3188 bs->total_sectors = MAX(bs->total_sectors, sector_num + nb_sectors);
3191 tracked_request_end(&req);
3197 * Handle a write request in coroutine context
3199 static int coroutine_fn bdrv_co_do_pwritev(BlockDriverState *bs,
3200 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
3201 BdrvRequestFlags flags)
3208 if (bs->read_only) {
3211 if (bdrv_check_byte_request(bs, offset, bytes)) {
3215 /* throttling disk I/O */
3216 if (bs->io_limits_enabled) {
3217 /* TODO Switch to byte granularity */
3218 bdrv_io_limits_intercept(bs, bytes >> BDRV_SECTOR_BITS, true);
3221 ret = bdrv_aligned_pwritev(bs, offset, bytes, qiov, flags);
3226 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
3227 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
3228 BdrvRequestFlags flags)
3230 if (nb_sectors < 0 || nb_sectors > (INT_MAX >> BDRV_SECTOR_BITS)) {
3234 return bdrv_co_do_pwritev(bs, sector_num << BDRV_SECTOR_BITS,
3235 nb_sectors << BDRV_SECTOR_BITS, qiov, flags);
3238 int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
3239 int nb_sectors, QEMUIOVector *qiov)
3241 trace_bdrv_co_writev(bs, sector_num, nb_sectors);
3243 return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0);
3246 int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs,
3247 int64_t sector_num, int nb_sectors,
3248 BdrvRequestFlags flags)
3250 trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors, flags);
3252 if (!(bs->open_flags & BDRV_O_UNMAP)) {
3253 flags &= ~BDRV_REQ_MAY_UNMAP;
3256 return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL,
3257 BDRV_REQ_ZERO_WRITE | flags);
3261 * Truncate file to 'offset' bytes (needed only for file protocols)
3263 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
3265 BlockDriver *drv = bs->drv;
3269 if (!drv->bdrv_truncate)
3273 if (bdrv_in_use(bs))
3275 ret = drv->bdrv_truncate(bs, offset);
3277 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
3278 bdrv_dev_resize_cb(bs);
3284 * Length of a allocated file in bytes. Sparse files are counted by actual
3285 * allocated space. Return < 0 if error or unknown.
3287 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
3289 BlockDriver *drv = bs->drv;
3293 if (drv->bdrv_get_allocated_file_size) {
3294 return drv->bdrv_get_allocated_file_size(bs);
3297 return bdrv_get_allocated_file_size(bs->file);
3303 * Length of a file in bytes. Return < 0 if error or unknown.
3305 int64_t bdrv_getlength(BlockDriverState *bs)
3307 BlockDriver *drv = bs->drv;
3311 if (drv->has_variable_length) {
3312 int ret = refresh_total_sectors(bs, bs->total_sectors);
3317 return bs->total_sectors * BDRV_SECTOR_SIZE;
3320 /* return 0 as number of sectors if no device present or error */
3321 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
3324 length = bdrv_getlength(bs);
3328 length = length >> BDRV_SECTOR_BITS;
3329 *nb_sectors_ptr = length;
3332 void bdrv_set_on_error(BlockDriverState *bs, BlockdevOnError on_read_error,
3333 BlockdevOnError on_write_error)
3335 bs->on_read_error = on_read_error;
3336 bs->on_write_error = on_write_error;
3339 BlockdevOnError bdrv_get_on_error(BlockDriverState *bs, bool is_read)
3341 return is_read ? bs->on_read_error : bs->on_write_error;
3344 BlockErrorAction bdrv_get_error_action(BlockDriverState *bs, bool is_read, int error)
3346 BlockdevOnError on_err = is_read ? bs->on_read_error : bs->on_write_error;
3349 case BLOCKDEV_ON_ERROR_ENOSPC:
3350 return (error == ENOSPC) ? BDRV_ACTION_STOP : BDRV_ACTION_REPORT;
3351 case BLOCKDEV_ON_ERROR_STOP:
3352 return BDRV_ACTION_STOP;
3353 case BLOCKDEV_ON_ERROR_REPORT:
3354 return BDRV_ACTION_REPORT;
3355 case BLOCKDEV_ON_ERROR_IGNORE:
3356 return BDRV_ACTION_IGNORE;
3362 /* This is done by device models because, while the block layer knows
3363 * about the error, it does not know whether an operation comes from
3364 * the device or the block layer (from a job, for example).
3366 void bdrv_error_action(BlockDriverState *bs, BlockErrorAction action,
3367 bool is_read, int error)
3370 bdrv_emit_qmp_error_event(bs, QEVENT_BLOCK_IO_ERROR, action, is_read);
3371 if (action == BDRV_ACTION_STOP) {
3372 vm_stop(RUN_STATE_IO_ERROR);
3373 bdrv_iostatus_set_err(bs, error);
3377 int bdrv_is_read_only(BlockDriverState *bs)
3379 return bs->read_only;
3382 int bdrv_is_sg(BlockDriverState *bs)
3387 int bdrv_enable_write_cache(BlockDriverState *bs)
3389 return bs->enable_write_cache;
3392 void bdrv_set_enable_write_cache(BlockDriverState *bs, bool wce)
3394 bs->enable_write_cache = wce;
3396 /* so a reopen() will preserve wce */
3398 bs->open_flags |= BDRV_O_CACHE_WB;
3400 bs->open_flags &= ~BDRV_O_CACHE_WB;
3404 int bdrv_is_encrypted(BlockDriverState *bs)
3406 if (bs->backing_hd && bs->backing_hd->encrypted)
3408 return bs->encrypted;
3411 int bdrv_key_required(BlockDriverState *bs)
3413 BlockDriverState *backing_hd = bs->backing_hd;
3415 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
3417 return (bs->encrypted && !bs->valid_key);
3420 int bdrv_set_key(BlockDriverState *bs, const char *key)
3423 if (bs->backing_hd && bs->backing_hd->encrypted) {
3424 ret = bdrv_set_key(bs->backing_hd, key);
3430 if (!bs->encrypted) {
3432 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
3435 ret = bs->drv->bdrv_set_key(bs, key);
3438 } else if (!bs->valid_key) {
3440 /* call the change callback now, we skipped it on open */
3441 bdrv_dev_change_media_cb(bs, true);
3446 const char *bdrv_get_format_name(BlockDriverState *bs)
3448 return bs->drv ? bs->drv->format_name : NULL;
3451 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
3456 QLIST_FOREACH(drv, &bdrv_drivers, list) {
3457 it(opaque, drv->format_name);
3461 /* This function is to find block backend bs */
3462 BlockDriverState *bdrv_find(const char *name)
3464 BlockDriverState *bs;
3466 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
3467 if (!strcmp(name, bs->device_name)) {
3474 /* This function is to find a node in the bs graph */
3475 BlockDriverState *bdrv_find_node(const char *node_name)
3477 BlockDriverState *bs;
3481 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3482 if (!strcmp(node_name, bs->node_name)) {
3489 /* Put this QMP function here so it can access the static graph_bdrv_states. */
3490 BlockDeviceInfoList *bdrv_named_nodes_list(void)
3492 BlockDeviceInfoList *list, *entry;
3493 BlockDriverState *bs;
3496 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3497 entry = g_malloc0(sizeof(*entry));
3498 entry->value = bdrv_block_device_info(bs);
3506 BlockDriverState *bdrv_lookup_bs(const char *device,
3507 const char *node_name,
3510 BlockDriverState *bs = NULL;
3512 if ((!device && !node_name) || (device && node_name)) {
3513 error_setg(errp, "Use either device or node-name but not both");
3518 bs = bdrv_find(device);
3521 error_set(errp, QERR_DEVICE_NOT_FOUND, device);
3528 bs = bdrv_find_node(node_name);
3531 error_set(errp, QERR_DEVICE_NOT_FOUND, node_name);
3538 BlockDriverState *bdrv_next(BlockDriverState *bs)
3541 return QTAILQ_FIRST(&bdrv_states);
3543 return QTAILQ_NEXT(bs, device_list);
3546 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
3548 BlockDriverState *bs;
3550 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
3555 const char *bdrv_get_device_name(BlockDriverState *bs)
3557 return bs->device_name;
3560 int bdrv_get_flags(BlockDriverState *bs)
3562 return bs->open_flags;
3565 int bdrv_flush_all(void)
3567 BlockDriverState *bs;
3570 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
3571 int ret = bdrv_flush(bs);
3572 if (ret < 0 && !result) {
3580 int bdrv_has_zero_init_1(BlockDriverState *bs)
3585 int bdrv_has_zero_init(BlockDriverState *bs)
3589 /* If BS is a copy on write image, it is initialized to
3590 the contents of the base image, which may not be zeroes. */
3591 if (bs->backing_hd) {
3594 if (bs->drv->bdrv_has_zero_init) {
3595 return bs->drv->bdrv_has_zero_init(bs);
3602 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
3604 BlockDriverInfo bdi;
3606 if (bs->backing_hd) {
3610 if (bdrv_get_info(bs, &bdi) == 0) {
3611 return bdi.unallocated_blocks_are_zero;
3617 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
3619 BlockDriverInfo bdi;
3621 if (bs->backing_hd || !(bs->open_flags & BDRV_O_UNMAP)) {
3625 if (bdrv_get_info(bs, &bdi) == 0) {
3626 return bdi.can_write_zeroes_with_unmap;
3632 typedef struct BdrvCoGetBlockStatusData {
3633 BlockDriverState *bs;
3634 BlockDriverState *base;
3640 } BdrvCoGetBlockStatusData;
3643 * Returns true iff the specified sector is present in the disk image. Drivers
3644 * not implementing the functionality are assumed to not support backing files,
3645 * hence all their sectors are reported as allocated.
3647 * If 'sector_num' is beyond the end of the disk image the return value is 0
3648 * and 'pnum' is set to 0.
3650 * 'pnum' is set to the number of sectors (including and immediately following
3651 * the specified sector) that are known to be in the same
3652 * allocated/unallocated state.
3654 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
3655 * beyond the end of the disk image it will be clamped.
3657 static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs,
3659 int nb_sectors, int *pnum)
3665 length = bdrv_getlength(bs);
3670 if (sector_num >= (length >> BDRV_SECTOR_BITS)) {
3675 n = bs->total_sectors - sector_num;
3676 if (n < nb_sectors) {
3680 if (!bs->drv->bdrv_co_get_block_status) {
3682 ret = BDRV_BLOCK_DATA;
3683 if (bs->drv->protocol_name) {
3684 ret |= BDRV_BLOCK_OFFSET_VALID | (sector_num * BDRV_SECTOR_SIZE);
3689 ret = bs->drv->bdrv_co_get_block_status(bs, sector_num, nb_sectors, pnum);
3695 if (ret & BDRV_BLOCK_RAW) {
3696 assert(ret & BDRV_BLOCK_OFFSET_VALID);
3697 return bdrv_get_block_status(bs->file, ret >> BDRV_SECTOR_BITS,
3701 if (!(ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO)) {
3702 if (bdrv_unallocated_blocks_are_zero(bs)) {
3703 ret |= BDRV_BLOCK_ZERO;
3704 } else if (bs->backing_hd) {
3705 BlockDriverState *bs2 = bs->backing_hd;
3706 int64_t length2 = bdrv_getlength(bs2);
3707 if (length2 >= 0 && sector_num >= (length2 >> BDRV_SECTOR_BITS)) {
3708 ret |= BDRV_BLOCK_ZERO;
3714 (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) &&
3715 (ret & BDRV_BLOCK_OFFSET_VALID)) {
3716 ret2 = bdrv_co_get_block_status(bs->file, ret >> BDRV_SECTOR_BITS,
3719 /* Ignore errors. This is just providing extra information, it
3720 * is useful but not necessary.
3722 ret |= (ret2 & BDRV_BLOCK_ZERO);
3729 /* Coroutine wrapper for bdrv_get_block_status() */
3730 static void coroutine_fn bdrv_get_block_status_co_entry(void *opaque)
3732 BdrvCoGetBlockStatusData *data = opaque;
3733 BlockDriverState *bs = data->bs;
3735 data->ret = bdrv_co_get_block_status(bs, data->sector_num, data->nb_sectors,
3741 * Synchronous wrapper around bdrv_co_get_block_status().
3743 * See bdrv_co_get_block_status() for details.
3745 int64_t bdrv_get_block_status(BlockDriverState *bs, int64_t sector_num,
3746 int nb_sectors, int *pnum)
3749 BdrvCoGetBlockStatusData data = {
3751 .sector_num = sector_num,
3752 .nb_sectors = nb_sectors,
3757 if (qemu_in_coroutine()) {
3758 /* Fast-path if already in coroutine context */
3759 bdrv_get_block_status_co_entry(&data);
3761 co = qemu_coroutine_create(bdrv_get_block_status_co_entry);
3762 qemu_coroutine_enter(co, &data);
3763 while (!data.done) {
3770 int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num,
3771 int nb_sectors, int *pnum)
3773 int64_t ret = bdrv_get_block_status(bs, sector_num, nb_sectors, pnum);
3778 (ret & BDRV_BLOCK_DATA) ||
3779 ((ret & BDRV_BLOCK_ZERO) && !bdrv_has_zero_init(bs));
3783 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
3785 * Return true if the given sector is allocated in any image between
3786 * BASE and TOP (inclusive). BASE can be NULL to check if the given
3787 * sector is allocated in any image of the chain. Return false otherwise.
3789 * 'pnum' is set to the number of sectors (including and immediately following
3790 * the specified sector) that are known to be in the same
3791 * allocated/unallocated state.
3794 int bdrv_is_allocated_above(BlockDriverState *top,
3795 BlockDriverState *base,
3797 int nb_sectors, int *pnum)
3799 BlockDriverState *intermediate;
3800 int ret, n = nb_sectors;
3803 while (intermediate && intermediate != base) {
3805 ret = bdrv_is_allocated(intermediate, sector_num, nb_sectors,
3815 * [sector_num, nb_sectors] is unallocated on top but intermediate
3818 * [sector_num+x, nr_sectors] allocated.
3820 if (n > pnum_inter &&
3821 (intermediate == top ||
3822 sector_num + pnum_inter < intermediate->total_sectors)) {
3826 intermediate = intermediate->backing_hd;
3833 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
3835 if (bs->backing_hd && bs->backing_hd->encrypted)
3836 return bs->backing_file;
3837 else if (bs->encrypted)
3838 return bs->filename;
3843 void bdrv_get_backing_filename(BlockDriverState *bs,
3844 char *filename, int filename_size)
3846 pstrcpy(filename, filename_size, bs->backing_file);
3849 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
3850 const uint8_t *buf, int nb_sectors)
3852 BlockDriver *drv = bs->drv;
3855 if (!drv->bdrv_write_compressed)
3857 if (bdrv_check_request(bs, sector_num, nb_sectors))
3860 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
3862 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
3865 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
3867 BlockDriver *drv = bs->drv;
3870 if (!drv->bdrv_get_info)
3872 memset(bdi, 0, sizeof(*bdi));
3873 return drv->bdrv_get_info(bs, bdi);
3876 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs)
3878 BlockDriver *drv = bs->drv;
3879 if (drv && drv->bdrv_get_specific_info) {
3880 return drv->bdrv_get_specific_info(bs);
3885 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
3886 int64_t pos, int size)
3889 struct iovec iov = {
3890 .iov_base = (void *) buf,
3894 qemu_iovec_init_external(&qiov, &iov, 1);
3895 return bdrv_writev_vmstate(bs, &qiov, pos);
3898 int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
3900 BlockDriver *drv = bs->drv;
3904 } else if (drv->bdrv_save_vmstate) {
3905 return drv->bdrv_save_vmstate(bs, qiov, pos);
3906 } else if (bs->file) {
3907 return bdrv_writev_vmstate(bs->file, qiov, pos);
3913 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
3914 int64_t pos, int size)
3916 BlockDriver *drv = bs->drv;
3919 if (drv->bdrv_load_vmstate)
3920 return drv->bdrv_load_vmstate(bs, buf, pos, size);
3922 return bdrv_load_vmstate(bs->file, buf, pos, size);
3926 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
3928 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
3932 bs->drv->bdrv_debug_event(bs, event);
3935 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
3938 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
3942 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
3943 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
3949 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
3951 while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
3955 if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
3956 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
3962 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
3964 while (bs && bs->drv && !bs->drv->bdrv_debug_resume) {
3968 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
3969 return bs->drv->bdrv_debug_resume(bs, tag);
3975 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
3977 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
3981 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
3982 return bs->drv->bdrv_debug_is_suspended(bs, tag);
3988 int bdrv_is_snapshot(BlockDriverState *bs)
3990 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
3993 /* backing_file can either be relative, or absolute, or a protocol. If it is
3994 * relative, it must be relative to the chain. So, passing in bs->filename
3995 * from a BDS as backing_file should not be done, as that may be relative to
3996 * the CWD rather than the chain. */
3997 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
3998 const char *backing_file)
4000 char *filename_full = NULL;
4001 char *backing_file_full = NULL;
4002 char *filename_tmp = NULL;
4003 int is_protocol = 0;
4004 BlockDriverState *curr_bs = NULL;
4005 BlockDriverState *retval = NULL;
4007 if (!bs || !bs->drv || !backing_file) {
4011 filename_full = g_malloc(PATH_MAX);
4012 backing_file_full = g_malloc(PATH_MAX);
4013 filename_tmp = g_malloc(PATH_MAX);
4015 is_protocol = path_has_protocol(backing_file);
4017 for (curr_bs = bs; curr_bs->backing_hd; curr_bs = curr_bs->backing_hd) {
4019 /* If either of the filename paths is actually a protocol, then
4020 * compare unmodified paths; otherwise make paths relative */
4021 if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
4022 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
4023 retval = curr_bs->backing_hd;
4027 /* If not an absolute filename path, make it relative to the current
4028 * image's filename path */
4029 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4032 /* We are going to compare absolute pathnames */
4033 if (!realpath(filename_tmp, filename_full)) {
4037 /* We need to make sure the backing filename we are comparing against
4038 * is relative to the current image filename (or absolute) */
4039 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4040 curr_bs->backing_file);
4042 if (!realpath(filename_tmp, backing_file_full)) {
4046 if (strcmp(backing_file_full, filename_full) == 0) {
4047 retval = curr_bs->backing_hd;
4053 g_free(filename_full);
4054 g_free(backing_file_full);
4055 g_free(filename_tmp);
4059 int bdrv_get_backing_file_depth(BlockDriverState *bs)
4065 if (!bs->backing_hd) {
4069 return 1 + bdrv_get_backing_file_depth(bs->backing_hd);
4072 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
4074 BlockDriverState *curr_bs = NULL;
4082 while (curr_bs->backing_hd) {
4083 curr_bs = curr_bs->backing_hd;
4088 /**************************************************************/
4091 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
4092 QEMUIOVector *qiov, int nb_sectors,
4093 BlockDriverCompletionFunc *cb, void *opaque)
4095 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
4097 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0,
4101 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
4102 QEMUIOVector *qiov, int nb_sectors,
4103 BlockDriverCompletionFunc *cb, void *opaque)
4105 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
4107 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0,
4111 BlockDriverAIOCB *bdrv_aio_write_zeroes(BlockDriverState *bs,
4112 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags,
4113 BlockDriverCompletionFunc *cb, void *opaque)
4115 trace_bdrv_aio_write_zeroes(bs, sector_num, nb_sectors, flags, opaque);
4117 return bdrv_co_aio_rw_vector(bs, sector_num, NULL, nb_sectors,
4118 BDRV_REQ_ZERO_WRITE | flags,
4123 typedef struct MultiwriteCB {
4128 BlockDriverCompletionFunc *cb;
4130 QEMUIOVector *free_qiov;
4134 static void multiwrite_user_cb(MultiwriteCB *mcb)
4138 for (i = 0; i < mcb->num_callbacks; i++) {
4139 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
4140 if (mcb->callbacks[i].free_qiov) {
4141 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
4143 g_free(mcb->callbacks[i].free_qiov);
4147 static void multiwrite_cb(void *opaque, int ret)
4149 MultiwriteCB *mcb = opaque;
4151 trace_multiwrite_cb(mcb, ret);
4153 if (ret < 0 && !mcb->error) {
4157 mcb->num_requests--;
4158 if (mcb->num_requests == 0) {
4159 multiwrite_user_cb(mcb);
4164 static int multiwrite_req_compare(const void *a, const void *b)
4166 const BlockRequest *req1 = a, *req2 = b;
4169 * Note that we can't simply subtract req2->sector from req1->sector
4170 * here as that could overflow the return value.
4172 if (req1->sector > req2->sector) {
4174 } else if (req1->sector < req2->sector) {
4182 * Takes a bunch of requests and tries to merge them. Returns the number of
4183 * requests that remain after merging.
4185 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
4186 int num_reqs, MultiwriteCB *mcb)
4190 // Sort requests by start sector
4191 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
4193 // Check if adjacent requests touch the same clusters. If so, combine them,
4194 // filling up gaps with zero sectors.
4196 for (i = 1; i < num_reqs; i++) {
4198 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
4200 // Handle exactly sequential writes and overlapping writes.
4201 if (reqs[i].sector <= oldreq_last) {
4205 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
4211 QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
4212 qemu_iovec_init(qiov,
4213 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
4215 // Add the first request to the merged one. If the requests are
4216 // overlapping, drop the last sectors of the first request.
4217 size = (reqs[i].sector - reqs[outidx].sector) << 9;
4218 qemu_iovec_concat(qiov, reqs[outidx].qiov, 0, size);
4220 // We should need to add any zeros between the two requests
4221 assert (reqs[i].sector <= oldreq_last);
4223 // Add the second request
4224 qemu_iovec_concat(qiov, reqs[i].qiov, 0, reqs[i].qiov->size);
4226 reqs[outidx].nb_sectors = qiov->size >> 9;
4227 reqs[outidx].qiov = qiov;
4229 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
4232 reqs[outidx].sector = reqs[i].sector;
4233 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
4234 reqs[outidx].qiov = reqs[i].qiov;
4242 * Submit multiple AIO write requests at once.
4244 * On success, the function returns 0 and all requests in the reqs array have
4245 * been submitted. In error case this function returns -1, and any of the
4246 * requests may or may not be submitted yet. In particular, this means that the
4247 * callback will be called for some of the requests, for others it won't. The
4248 * caller must check the error field of the BlockRequest to wait for the right
4249 * callbacks (if error != 0, no callback will be called).
4251 * The implementation may modify the contents of the reqs array, e.g. to merge
4252 * requests. However, the fields opaque and error are left unmodified as they
4253 * are used to signal failure for a single request to the caller.
4255 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
4260 /* don't submit writes if we don't have a medium */
4261 if (bs->drv == NULL) {
4262 for (i = 0; i < num_reqs; i++) {
4263 reqs[i].error = -ENOMEDIUM;
4268 if (num_reqs == 0) {
4272 // Create MultiwriteCB structure
4273 mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
4274 mcb->num_requests = 0;
4275 mcb->num_callbacks = num_reqs;
4277 for (i = 0; i < num_reqs; i++) {
4278 mcb->callbacks[i].cb = reqs[i].cb;
4279 mcb->callbacks[i].opaque = reqs[i].opaque;
4282 // Check for mergable requests
4283 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
4285 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
4287 /* Run the aio requests. */
4288 mcb->num_requests = num_reqs;
4289 for (i = 0; i < num_reqs; i++) {
4290 bdrv_co_aio_rw_vector(bs, reqs[i].sector, reqs[i].qiov,
4291 reqs[i].nb_sectors, reqs[i].flags,
4299 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
4301 acb->aiocb_info->cancel(acb);
4304 /**************************************************************/
4305 /* async block device emulation */
4307 typedef struct BlockDriverAIOCBSync {
4308 BlockDriverAIOCB common;
4311 /* vector translation state */
4315 } BlockDriverAIOCBSync;
4317 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
4319 BlockDriverAIOCBSync *acb =
4320 container_of(blockacb, BlockDriverAIOCBSync, common);
4321 qemu_bh_delete(acb->bh);
4323 qemu_aio_release(acb);
4326 static const AIOCBInfo bdrv_em_aiocb_info = {
4327 .aiocb_size = sizeof(BlockDriverAIOCBSync),
4328 .cancel = bdrv_aio_cancel_em,
4331 static void bdrv_aio_bh_cb(void *opaque)
4333 BlockDriverAIOCBSync *acb = opaque;
4336 qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
4337 qemu_vfree(acb->bounce);
4338 acb->common.cb(acb->common.opaque, acb->ret);
4339 qemu_bh_delete(acb->bh);
4341 qemu_aio_release(acb);
4344 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
4348 BlockDriverCompletionFunc *cb,
4353 BlockDriverAIOCBSync *acb;
4355 acb = qemu_aio_get(&bdrv_em_aiocb_info, bs, cb, opaque);
4356 acb->is_write = is_write;
4358 acb->bounce = qemu_blockalign(bs, qiov->size);
4359 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
4362 qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
4363 acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
4365 acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
4368 qemu_bh_schedule(acb->bh);
4370 return &acb->common;
4373 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
4374 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
4375 BlockDriverCompletionFunc *cb, void *opaque)
4377 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
4380 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
4381 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
4382 BlockDriverCompletionFunc *cb, void *opaque)
4384 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
4388 typedef struct BlockDriverAIOCBCoroutine {
4389 BlockDriverAIOCB common;
4394 } BlockDriverAIOCBCoroutine;
4396 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb)
4398 BlockDriverAIOCBCoroutine *acb =
4399 container_of(blockacb, BlockDriverAIOCBCoroutine, common);
4408 static const AIOCBInfo bdrv_em_co_aiocb_info = {
4409 .aiocb_size = sizeof(BlockDriverAIOCBCoroutine),
4410 .cancel = bdrv_aio_co_cancel_em,
4413 static void bdrv_co_em_bh(void *opaque)
4415 BlockDriverAIOCBCoroutine *acb = opaque;
4417 acb->common.cb(acb->common.opaque, acb->req.error);
4423 qemu_bh_delete(acb->bh);
4424 qemu_aio_release(acb);
4427 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
4428 static void coroutine_fn bdrv_co_do_rw(void *opaque)
4430 BlockDriverAIOCBCoroutine *acb = opaque;
4431 BlockDriverState *bs = acb->common.bs;
4433 if (!acb->is_write) {
4434 acb->req.error = bdrv_co_do_readv(bs, acb->req.sector,
4435 acb->req.nb_sectors, acb->req.qiov, acb->req.flags);
4437 acb->req.error = bdrv_co_do_writev(bs, acb->req.sector,
4438 acb->req.nb_sectors, acb->req.qiov, acb->req.flags);
4441 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
4442 qemu_bh_schedule(acb->bh);
4445 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
4449 BdrvRequestFlags flags,
4450 BlockDriverCompletionFunc *cb,
4455 BlockDriverAIOCBCoroutine *acb;
4457 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
4458 acb->req.sector = sector_num;
4459 acb->req.nb_sectors = nb_sectors;
4460 acb->req.qiov = qiov;
4461 acb->req.flags = flags;
4462 acb->is_write = is_write;
4465 co = qemu_coroutine_create(bdrv_co_do_rw);
4466 qemu_coroutine_enter(co, acb);
4468 return &acb->common;
4471 static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
4473 BlockDriverAIOCBCoroutine *acb = opaque;
4474 BlockDriverState *bs = acb->common.bs;
4476 acb->req.error = bdrv_co_flush(bs);
4477 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
4478 qemu_bh_schedule(acb->bh);
4481 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
4482 BlockDriverCompletionFunc *cb, void *opaque)
4484 trace_bdrv_aio_flush(bs, opaque);
4487 BlockDriverAIOCBCoroutine *acb;
4489 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
4492 co = qemu_coroutine_create(bdrv_aio_flush_co_entry);
4493 qemu_coroutine_enter(co, acb);
4495 return &acb->common;
4498 static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque)
4500 BlockDriverAIOCBCoroutine *acb = opaque;
4501 BlockDriverState *bs = acb->common.bs;
4503 acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors);
4504 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
4505 qemu_bh_schedule(acb->bh);
4508 BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs,
4509 int64_t sector_num, int nb_sectors,
4510 BlockDriverCompletionFunc *cb, void *opaque)
4513 BlockDriverAIOCBCoroutine *acb;
4515 trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque);
4517 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
4518 acb->req.sector = sector_num;
4519 acb->req.nb_sectors = nb_sectors;
4521 co = qemu_coroutine_create(bdrv_aio_discard_co_entry);
4522 qemu_coroutine_enter(co, acb);
4524 return &acb->common;
4527 void bdrv_init(void)
4529 module_call_init(MODULE_INIT_BLOCK);
4532 void bdrv_init_with_whitelist(void)
4534 use_bdrv_whitelist = 1;
4538 void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs,
4539 BlockDriverCompletionFunc *cb, void *opaque)
4541 BlockDriverAIOCB *acb;
4543 acb = g_slice_alloc(aiocb_info->aiocb_size);
4544 acb->aiocb_info = aiocb_info;
4547 acb->opaque = opaque;
4551 void qemu_aio_release(void *p)
4553 BlockDriverAIOCB *acb = p;
4554 g_slice_free1(acb->aiocb_info->aiocb_size, acb);
4557 /**************************************************************/
4558 /* Coroutine block device emulation */
4560 typedef struct CoroutineIOCompletion {
4561 Coroutine *coroutine;
4563 } CoroutineIOCompletion;
4565 static void bdrv_co_io_em_complete(void *opaque, int ret)
4567 CoroutineIOCompletion *co = opaque;
4570 qemu_coroutine_enter(co->coroutine, NULL);
4573 static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num,
4574 int nb_sectors, QEMUIOVector *iov,
4577 CoroutineIOCompletion co = {
4578 .coroutine = qemu_coroutine_self(),
4580 BlockDriverAIOCB *acb;
4583 acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors,
4584 bdrv_co_io_em_complete, &co);
4586 acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors,
4587 bdrv_co_io_em_complete, &co);
4590 trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb);
4594 qemu_coroutine_yield();
4599 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
4600 int64_t sector_num, int nb_sectors,
4603 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false);
4606 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
4607 int64_t sector_num, int nb_sectors,
4610 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true);
4613 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
4615 RwCo *rwco = opaque;
4617 rwco->ret = bdrv_co_flush(rwco->bs);
4620 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
4624 if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
4628 /* Write back cached data to the OS even with cache=unsafe */
4629 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS);
4630 if (bs->drv->bdrv_co_flush_to_os) {
4631 ret = bs->drv->bdrv_co_flush_to_os(bs);
4637 /* But don't actually force it to the disk with cache=unsafe */
4638 if (bs->open_flags & BDRV_O_NO_FLUSH) {
4642 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK);
4643 if (bs->drv->bdrv_co_flush_to_disk) {
4644 ret = bs->drv->bdrv_co_flush_to_disk(bs);
4645 } else if (bs->drv->bdrv_aio_flush) {
4646 BlockDriverAIOCB *acb;
4647 CoroutineIOCompletion co = {
4648 .coroutine = qemu_coroutine_self(),
4651 acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
4655 qemu_coroutine_yield();
4660 * Some block drivers always operate in either writethrough or unsafe
4661 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
4662 * know how the server works (because the behaviour is hardcoded or
4663 * depends on server-side configuration), so we can't ensure that
4664 * everything is safe on disk. Returning an error doesn't work because
4665 * that would break guests even if the server operates in writethrough
4668 * Let's hope the user knows what he's doing.
4676 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
4677 * in the case of cache=unsafe, so there are no useless flushes.
4680 return bdrv_co_flush(bs->file);
4683 void bdrv_invalidate_cache(BlockDriverState *bs)
4685 if (bs->drv && bs->drv->bdrv_invalidate_cache) {
4686 bs->drv->bdrv_invalidate_cache(bs);
4690 void bdrv_invalidate_cache_all(void)
4692 BlockDriverState *bs;
4694 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
4695 bdrv_invalidate_cache(bs);
4699 void bdrv_clear_incoming_migration_all(void)
4701 BlockDriverState *bs;
4703 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
4704 bs->open_flags = bs->open_flags & ~(BDRV_O_INCOMING);
4708 int bdrv_flush(BlockDriverState *bs)
4716 if (qemu_in_coroutine()) {
4717 /* Fast-path if already in coroutine context */
4718 bdrv_flush_co_entry(&rwco);
4720 co = qemu_coroutine_create(bdrv_flush_co_entry);
4721 qemu_coroutine_enter(co, &rwco);
4722 while (rwco.ret == NOT_DONE) {
4730 static void coroutine_fn bdrv_discard_co_entry(void *opaque)
4732 RwCo *rwco = opaque;
4734 rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors);
4737 /* if no limit is specified in the BlockLimits use a default
4738 * of 32768 512-byte sectors (16 MiB) per request.
4740 #define MAX_DISCARD_DEFAULT 32768
4742 int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
4749 } else if (bdrv_check_request(bs, sector_num, nb_sectors)) {
4751 } else if (bs->read_only) {
4755 bdrv_reset_dirty(bs, sector_num, nb_sectors);
4757 /* Do nothing if disabled. */
4758 if (!(bs->open_flags & BDRV_O_UNMAP)) {
4762 if (!bs->drv->bdrv_co_discard && !bs->drv->bdrv_aio_discard) {
4766 max_discard = bs->bl.max_discard ? bs->bl.max_discard : MAX_DISCARD_DEFAULT;
4767 while (nb_sectors > 0) {
4769 int num = nb_sectors;
4772 if (bs->bl.discard_alignment &&
4773 num >= bs->bl.discard_alignment &&
4774 sector_num % bs->bl.discard_alignment) {
4775 if (num > bs->bl.discard_alignment) {
4776 num = bs->bl.discard_alignment;
4778 num -= sector_num % bs->bl.discard_alignment;
4781 /* limit request size */
4782 if (num > max_discard) {
4786 if (bs->drv->bdrv_co_discard) {
4787 ret = bs->drv->bdrv_co_discard(bs, sector_num, num);
4789 BlockDriverAIOCB *acb;
4790 CoroutineIOCompletion co = {
4791 .coroutine = qemu_coroutine_self(),
4794 acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors,
4795 bdrv_co_io_em_complete, &co);
4799 qemu_coroutine_yield();
4803 if (ret && ret != -ENOTSUP) {
4813 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
4818 .sector_num = sector_num,
4819 .nb_sectors = nb_sectors,
4823 if (qemu_in_coroutine()) {
4824 /* Fast-path if already in coroutine context */
4825 bdrv_discard_co_entry(&rwco);
4827 co = qemu_coroutine_create(bdrv_discard_co_entry);
4828 qemu_coroutine_enter(co, &rwco);
4829 while (rwco.ret == NOT_DONE) {
4837 /**************************************************************/
4838 /* removable device support */
4841 * Return TRUE if the media is present
4843 int bdrv_is_inserted(BlockDriverState *bs)
4845 BlockDriver *drv = bs->drv;
4849 if (!drv->bdrv_is_inserted)
4851 return drv->bdrv_is_inserted(bs);
4855 * Return whether the media changed since the last call to this
4856 * function, or -ENOTSUP if we don't know. Most drivers don't know.
4858 int bdrv_media_changed(BlockDriverState *bs)
4860 BlockDriver *drv = bs->drv;
4862 if (drv && drv->bdrv_media_changed) {
4863 return drv->bdrv_media_changed(bs);
4869 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4871 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
4873 BlockDriver *drv = bs->drv;
4875 if (drv && drv->bdrv_eject) {
4876 drv->bdrv_eject(bs, eject_flag);
4879 if (bs->device_name[0] != '\0') {
4880 bdrv_emit_qmp_eject_event(bs, eject_flag);
4885 * Lock or unlock the media (if it is locked, the user won't be able
4886 * to eject it manually).
4888 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
4890 BlockDriver *drv = bs->drv;
4892 trace_bdrv_lock_medium(bs, locked);
4894 if (drv && drv->bdrv_lock_medium) {
4895 drv->bdrv_lock_medium(bs, locked);
4899 /* needed for generic scsi interface */
4901 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
4903 BlockDriver *drv = bs->drv;
4905 if (drv && drv->bdrv_ioctl)
4906 return drv->bdrv_ioctl(bs, req, buf);
4910 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
4911 unsigned long int req, void *buf,
4912 BlockDriverCompletionFunc *cb, void *opaque)
4914 BlockDriver *drv = bs->drv;
4916 if (drv && drv->bdrv_aio_ioctl)
4917 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
4921 void bdrv_set_guest_block_size(BlockDriverState *bs, int align)
4923 bs->guest_block_size = align;
4926 void *qemu_blockalign(BlockDriverState *bs, size_t size)
4928 return qemu_memalign(bdrv_opt_mem_align(bs), size);
4932 * Check if all memory in this vector is sector aligned.
4934 bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov)
4937 size_t alignment = bdrv_opt_mem_align(bs);
4939 for (i = 0; i < qiov->niov; i++) {
4940 if ((uintptr_t) qiov->iov[i].iov_base % alignment) {
4943 if (qiov->iov[i].iov_len % alignment) {
4951 BdrvDirtyBitmap *bdrv_create_dirty_bitmap(BlockDriverState *bs, int granularity)
4953 int64_t bitmap_size;
4954 BdrvDirtyBitmap *bitmap;
4956 assert((granularity & (granularity - 1)) == 0);
4958 granularity >>= BDRV_SECTOR_BITS;
4959 assert(granularity);
4960 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS);
4961 bitmap = g_malloc0(sizeof(BdrvDirtyBitmap));
4962 bitmap->bitmap = hbitmap_alloc(bitmap_size, ffs(granularity) - 1);
4963 QLIST_INSERT_HEAD(&bs->dirty_bitmaps, bitmap, list);
4967 void bdrv_release_dirty_bitmap(BlockDriverState *bs, BdrvDirtyBitmap *bitmap)
4969 BdrvDirtyBitmap *bm, *next;
4970 QLIST_FOREACH_SAFE(bm, &bs->dirty_bitmaps, list, next) {
4972 QLIST_REMOVE(bitmap, list);
4973 hbitmap_free(bitmap->bitmap);
4980 BlockDirtyInfoList *bdrv_query_dirty_bitmaps(BlockDriverState *bs)
4982 BdrvDirtyBitmap *bm;
4983 BlockDirtyInfoList *list = NULL;
4984 BlockDirtyInfoList **plist = &list;
4986 QLIST_FOREACH(bm, &bs->dirty_bitmaps, list) {
4987 BlockDirtyInfo *info = g_malloc0(sizeof(BlockDirtyInfo));
4988 BlockDirtyInfoList *entry = g_malloc0(sizeof(BlockDirtyInfoList));
4989 info->count = bdrv_get_dirty_count(bs, bm);
4991 ((int64_t) BDRV_SECTOR_SIZE << hbitmap_granularity(bm->bitmap));
4992 entry->value = info;
4994 plist = &entry->next;
5000 int bdrv_get_dirty(BlockDriverState *bs, BdrvDirtyBitmap *bitmap, int64_t sector)
5003 return hbitmap_get(bitmap->bitmap, sector);
5009 void bdrv_dirty_iter_init(BlockDriverState *bs,
5010 BdrvDirtyBitmap *bitmap, HBitmapIter *hbi)
5012 hbitmap_iter_init(hbi, bitmap->bitmap, 0);
5015 void bdrv_set_dirty(BlockDriverState *bs, int64_t cur_sector,
5018 BdrvDirtyBitmap *bitmap;
5019 QLIST_FOREACH(bitmap, &bs->dirty_bitmaps, list) {
5020 hbitmap_set(bitmap->bitmap, cur_sector, nr_sectors);
5024 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector, int nr_sectors)
5026 BdrvDirtyBitmap *bitmap;
5027 QLIST_FOREACH(bitmap, &bs->dirty_bitmaps, list) {
5028 hbitmap_reset(bitmap->bitmap, cur_sector, nr_sectors);
5032 int64_t bdrv_get_dirty_count(BlockDriverState *bs, BdrvDirtyBitmap *bitmap)
5034 return hbitmap_count(bitmap->bitmap);
5037 /* Get a reference to bs */
5038 void bdrv_ref(BlockDriverState *bs)
5043 /* Release a previously grabbed reference to bs.
5044 * If after releasing, reference count is zero, the BlockDriverState is
5046 void bdrv_unref(BlockDriverState *bs)
5048 assert(bs->refcnt > 0);
5049 if (--bs->refcnt == 0) {
5054 void bdrv_set_in_use(BlockDriverState *bs, int in_use)
5056 assert(bs->in_use != in_use);
5057 bs->in_use = in_use;
5060 int bdrv_in_use(BlockDriverState *bs)
5065 void bdrv_iostatus_enable(BlockDriverState *bs)
5067 bs->iostatus_enabled = true;
5068 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
5071 /* The I/O status is only enabled if the drive explicitly
5072 * enables it _and_ the VM is configured to stop on errors */
5073 bool bdrv_iostatus_is_enabled(const BlockDriverState *bs)
5075 return (bs->iostatus_enabled &&
5076 (bs->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
5077 bs->on_write_error == BLOCKDEV_ON_ERROR_STOP ||
5078 bs->on_read_error == BLOCKDEV_ON_ERROR_STOP));
5081 void bdrv_iostatus_disable(BlockDriverState *bs)
5083 bs->iostatus_enabled = false;
5086 void bdrv_iostatus_reset(BlockDriverState *bs)
5088 if (bdrv_iostatus_is_enabled(bs)) {
5089 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
5091 block_job_iostatus_reset(bs->job);
5096 void bdrv_iostatus_set_err(BlockDriverState *bs, int error)
5098 assert(bdrv_iostatus_is_enabled(bs));
5099 if (bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
5100 bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
5101 BLOCK_DEVICE_IO_STATUS_FAILED;
5106 bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes,
5107 enum BlockAcctType type)
5109 assert(type < BDRV_MAX_IOTYPE);
5111 cookie->bytes = bytes;
5112 cookie->start_time_ns = get_clock();
5113 cookie->type = type;
5117 bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie)
5119 assert(cookie->type < BDRV_MAX_IOTYPE);
5121 bs->nr_bytes[cookie->type] += cookie->bytes;
5122 bs->nr_ops[cookie->type]++;
5123 bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns;
5126 void bdrv_img_create(const char *filename, const char *fmt,
5127 const char *base_filename, const char *base_fmt,
5128 char *options, uint64_t img_size, int flags,
5129 Error **errp, bool quiet)
5131 QEMUOptionParameter *param = NULL, *create_options = NULL;
5132 QEMUOptionParameter *backing_fmt, *backing_file, *size;
5133 BlockDriver *drv, *proto_drv;
5134 BlockDriver *backing_drv = NULL;
5135 Error *local_err = NULL;
5138 /* Find driver and parse its options */
5139 drv = bdrv_find_format(fmt);
5141 error_setg(errp, "Unknown file format '%s'", fmt);
5145 proto_drv = bdrv_find_protocol(filename, true);
5147 error_setg(errp, "Unknown protocol '%s'", filename);
5151 create_options = append_option_parameters(create_options,
5152 drv->create_options);
5153 create_options = append_option_parameters(create_options,
5154 proto_drv->create_options);
5156 /* Create parameter list with default values */
5157 param = parse_option_parameters("", create_options, param);
5159 set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size);
5161 /* Parse -o options */
5163 param = parse_option_parameters(options, create_options, param);
5164 if (param == NULL) {
5165 error_setg(errp, "Invalid options for file format '%s'.", fmt);
5170 if (base_filename) {
5171 if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE,
5173 error_setg(errp, "Backing file not supported for file format '%s'",
5180 if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) {
5181 error_setg(errp, "Backing file format not supported for file "
5182 "format '%s'", fmt);
5187 backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE);
5188 if (backing_file && backing_file->value.s) {
5189 if (!strcmp(filename, backing_file->value.s)) {
5190 error_setg(errp, "Error: Trying to create an image with the "
5191 "same filename as the backing file");
5196 backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT);
5197 if (backing_fmt && backing_fmt->value.s) {
5198 backing_drv = bdrv_find_format(backing_fmt->value.s);
5200 error_setg(errp, "Unknown backing file format '%s'",
5201 backing_fmt->value.s);
5206 // The size for the image must always be specified, with one exception:
5207 // If we are using a backing file, we can obtain the size from there
5208 size = get_option_parameter(param, BLOCK_OPT_SIZE);
5209 if (size && size->value.n == -1) {
5210 if (backing_file && backing_file->value.s) {
5211 BlockDriverState *bs;
5216 /* backing files always opened read-only */
5218 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
5222 ret = bdrv_open(bs, backing_file->value.s, NULL, back_flags,
5223 backing_drv, &local_err);
5225 error_setg_errno(errp, -ret, "Could not open '%s': %s",
5226 backing_file->value.s,
5227 error_get_pretty(local_err));
5228 error_free(local_err);
5233 bdrv_get_geometry(bs, &size);
5236 snprintf(buf, sizeof(buf), "%" PRId64, size);
5237 set_option_parameter(param, BLOCK_OPT_SIZE, buf);
5241 error_setg(errp, "Image creation needs a size parameter");
5247 printf("Formatting '%s', fmt=%s ", filename, fmt);
5248 print_option_parameters(param);
5251 ret = bdrv_create(drv, filename, param, &local_err);
5252 if (ret == -EFBIG) {
5253 /* This is generally a better message than whatever the driver would
5254 * deliver (especially because of the cluster_size_hint), since that
5255 * is most probably not much different from "image too large". */
5256 const char *cluster_size_hint = "";
5257 if (get_option_parameter(create_options, BLOCK_OPT_CLUSTER_SIZE)) {
5258 cluster_size_hint = " (try using a larger cluster size)";
5260 error_setg(errp, "The image size is too large for file format '%s'"
5261 "%s", fmt, cluster_size_hint);
5262 error_free(local_err);
5267 free_option_parameters(create_options);
5268 free_option_parameters(param);
5270 if (error_is_set(&local_err)) {
5271 error_propagate(errp, local_err);
5275 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
5277 /* Currently BlockDriverState always uses the main loop AioContext */
5278 return qemu_get_aio_context();
5281 void bdrv_add_before_write_notifier(BlockDriverState *bs,
5282 NotifierWithReturn *notifier)
5284 notifier_with_return_list_add(&bs->before_write_notifiers, notifier);
5287 int bdrv_amend_options(BlockDriverState *bs, QEMUOptionParameter *options)
5289 if (bs->drv->bdrv_amend_options == NULL) {
5292 return bs->drv->bdrv_amend_options(bs, options);
5295 /* Used to recurse on single child block filters.
5296 * Single child block filter will store their child in bs->file.
5298 bool bdrv_generic_is_first_non_filter(BlockDriverState *bs,
5299 BlockDriverState *candidate)
5305 if (!bs->drv->authorizations[BS_IS_A_FILTER]) {
5306 if (bs == candidate) {
5313 if (!bs->drv->authorizations[BS_FILTER_PASS_DOWN]) {
5321 return bdrv_recurse_is_first_non_filter(bs->file, candidate);
5324 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
5325 BlockDriverState *candidate)
5327 if (bs->drv && bs->drv->bdrv_recurse_is_first_non_filter) {
5328 return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate);
5331 return bdrv_generic_is_first_non_filter(bs, candidate);
5334 /* This function checks if the candidate is the first non filter bs down it's
5335 * bs chain. Since we don't have pointers to parents it explore all bs chains
5336 * from the top. Some filters can choose not to pass down the recursion.
5338 bool bdrv_is_first_non_filter(BlockDriverState *candidate)
5340 BlockDriverState *bs;
5342 /* walk down the bs forest recursively */
5343 QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
5350 perm = bdrv_recurse_is_first_non_filter(bs->file, candidate);
5352 /* candidate is the first non filter */