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"
28 #include "block_int.h"
31 #include "qemu-coroutine.h"
32 #include "qmp-commands.h"
33 #include "qemu-timer.h"
36 #include <sys/types.h>
38 #include <sys/ioctl.h>
39 #include <sys/queue.h>
49 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
52 BDRV_REQ_COPY_ON_READ = 0x1,
53 BDRV_REQ_ZERO_WRITE = 0x2,
56 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load);
57 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
58 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
59 BlockDriverCompletionFunc *cb, void *opaque);
60 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
61 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
62 BlockDriverCompletionFunc *cb, void *opaque);
63 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
64 int64_t sector_num, int nb_sectors,
66 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
67 int64_t sector_num, int nb_sectors,
69 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
70 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
71 BdrvRequestFlags flags);
72 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
73 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
74 BdrvRequestFlags flags);
75 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
79 BlockDriverCompletionFunc *cb,
82 static void coroutine_fn bdrv_co_do_rw(void *opaque);
83 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
84 int64_t sector_num, int nb_sectors);
86 static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors,
87 bool is_write, double elapsed_time, uint64_t *wait);
88 static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write,
89 double elapsed_time, uint64_t *wait);
90 static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors,
91 bool is_write, int64_t *wait);
93 static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
94 QTAILQ_HEAD_INITIALIZER(bdrv_states);
96 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
97 QLIST_HEAD_INITIALIZER(bdrv_drivers);
99 /* The device to use for VM snapshots */
100 static BlockDriverState *bs_snapshots;
102 /* If non-zero, use only whitelisted block drivers */
103 static int use_bdrv_whitelist;
106 static int is_windows_drive_prefix(const char *filename)
108 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
109 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
113 int is_windows_drive(const char *filename)
115 if (is_windows_drive_prefix(filename) &&
118 if (strstart(filename, "\\\\.\\", NULL) ||
119 strstart(filename, "//./", NULL))
125 /* throttling disk I/O limits */
126 void bdrv_io_limits_disable(BlockDriverState *bs)
128 bs->io_limits_enabled = false;
130 while (qemu_co_queue_next(&bs->throttled_reqs));
132 if (bs->block_timer) {
133 qemu_del_timer(bs->block_timer);
134 qemu_free_timer(bs->block_timer);
135 bs->block_timer = NULL;
141 memset(&bs->io_base, 0, sizeof(bs->io_base));
144 static void bdrv_block_timer(void *opaque)
146 BlockDriverState *bs = opaque;
148 qemu_co_queue_next(&bs->throttled_reqs);
151 void bdrv_io_limits_enable(BlockDriverState *bs)
153 qemu_co_queue_init(&bs->throttled_reqs);
154 bs->block_timer = qemu_new_timer_ns(vm_clock, bdrv_block_timer, bs);
155 bs->slice_time = 5 * BLOCK_IO_SLICE_TIME;
156 bs->slice_start = qemu_get_clock_ns(vm_clock);
157 bs->slice_end = bs->slice_start + bs->slice_time;
158 memset(&bs->io_base, 0, sizeof(bs->io_base));
159 bs->io_limits_enabled = true;
162 bool bdrv_io_limits_enabled(BlockDriverState *bs)
164 BlockIOLimit *io_limits = &bs->io_limits;
165 return io_limits->bps[BLOCK_IO_LIMIT_READ]
166 || io_limits->bps[BLOCK_IO_LIMIT_WRITE]
167 || io_limits->bps[BLOCK_IO_LIMIT_TOTAL]
168 || io_limits->iops[BLOCK_IO_LIMIT_READ]
169 || io_limits->iops[BLOCK_IO_LIMIT_WRITE]
170 || io_limits->iops[BLOCK_IO_LIMIT_TOTAL];
173 static void bdrv_io_limits_intercept(BlockDriverState *bs,
174 bool is_write, int nb_sectors)
176 int64_t wait_time = -1;
178 if (!qemu_co_queue_empty(&bs->throttled_reqs)) {
179 qemu_co_queue_wait(&bs->throttled_reqs);
182 /* In fact, we hope to keep each request's timing, in FIFO mode. The next
183 * throttled requests will not be dequeued until the current request is
184 * allowed to be serviced. So if the current request still exceeds the
185 * limits, it will be inserted to the head. All requests followed it will
186 * be still in throttled_reqs queue.
189 while (bdrv_exceed_io_limits(bs, nb_sectors, is_write, &wait_time)) {
190 qemu_mod_timer(bs->block_timer,
191 wait_time + qemu_get_clock_ns(vm_clock));
192 qemu_co_queue_wait_insert_head(&bs->throttled_reqs);
195 qemu_co_queue_next(&bs->throttled_reqs);
198 /* check if the path starts with "<protocol>:" */
199 static int path_has_protocol(const char *path)
202 if (is_windows_drive(path) ||
203 is_windows_drive_prefix(path)) {
208 return strchr(path, ':') != NULL;
211 int path_is_absolute(const char *path)
215 /* specific case for names like: "\\.\d:" */
216 if (*path == '/' || *path == '\\')
219 p = strchr(path, ':');
225 return (*p == '/' || *p == '\\');
231 /* if filename is absolute, just copy it to dest. Otherwise, build a
232 path to it by considering it is relative to base_path. URL are
234 void path_combine(char *dest, int dest_size,
235 const char *base_path,
236 const char *filename)
243 if (path_is_absolute(filename)) {
244 pstrcpy(dest, dest_size, filename);
246 p = strchr(base_path, ':');
251 p1 = strrchr(base_path, '/');
255 p2 = strrchr(base_path, '\\');
267 if (len > dest_size - 1)
269 memcpy(dest, base_path, len);
271 pstrcat(dest, dest_size, filename);
275 void bdrv_register(BlockDriver *bdrv)
277 /* Block drivers without coroutine functions need emulation */
278 if (!bdrv->bdrv_co_readv) {
279 bdrv->bdrv_co_readv = bdrv_co_readv_em;
280 bdrv->bdrv_co_writev = bdrv_co_writev_em;
282 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
283 * the block driver lacks aio we need to emulate that too.
285 if (!bdrv->bdrv_aio_readv) {
286 /* add AIO emulation layer */
287 bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
288 bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
292 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
295 /* create a new block device (by default it is empty) */
296 BlockDriverState *bdrv_new(const char *device_name)
298 BlockDriverState *bs;
300 bs = g_malloc0(sizeof(BlockDriverState));
301 pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
302 if (device_name[0] != '\0') {
303 QTAILQ_INSERT_TAIL(&bdrv_states, bs, list);
305 bdrv_iostatus_disable(bs);
309 BlockDriver *bdrv_find_format(const char *format_name)
312 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
313 if (!strcmp(drv1->format_name, format_name)) {
320 static int bdrv_is_whitelisted(BlockDriver *drv)
322 static const char *whitelist[] = {
323 CONFIG_BDRV_WHITELIST
328 return 1; /* no whitelist, anything goes */
330 for (p = whitelist; *p; p++) {
331 if (!strcmp(drv->format_name, *p)) {
338 BlockDriver *bdrv_find_whitelisted_format(const char *format_name)
340 BlockDriver *drv = bdrv_find_format(format_name);
341 return drv && bdrv_is_whitelisted(drv) ? drv : NULL;
344 int bdrv_create(BlockDriver *drv, const char* filename,
345 QEMUOptionParameter *options)
347 if (!drv->bdrv_create)
350 return drv->bdrv_create(filename, options);
353 int bdrv_create_file(const char* filename, QEMUOptionParameter *options)
357 drv = bdrv_find_protocol(filename);
362 return bdrv_create(drv, filename, options);
366 void get_tmp_filename(char *filename, int size)
368 char temp_dir[MAX_PATH];
370 GetTempPath(MAX_PATH, temp_dir);
371 GetTempFileName(temp_dir, "qem", 0, filename);
374 void get_tmp_filename(char *filename, int size)
378 /* XXX: race condition possible */
379 tmpdir = getenv("TMPDIR");
382 snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
383 fd = mkstemp(filename);
389 * Detect host devices. By convention, /dev/cdrom[N] is always
390 * recognized as a host CDROM.
392 static BlockDriver *find_hdev_driver(const char *filename)
394 int score_max = 0, score;
395 BlockDriver *drv = NULL, *d;
397 QLIST_FOREACH(d, &bdrv_drivers, list) {
398 if (d->bdrv_probe_device) {
399 score = d->bdrv_probe_device(filename);
400 if (score > score_max) {
410 BlockDriver *bdrv_find_protocol(const char *filename)
417 /* TODO Drivers without bdrv_file_open must be specified explicitly */
420 * XXX(hch): we really should not let host device detection
421 * override an explicit protocol specification, but moving this
422 * later breaks access to device names with colons in them.
423 * Thanks to the brain-dead persistent naming schemes on udev-
424 * based Linux systems those actually are quite common.
426 drv1 = find_hdev_driver(filename);
431 if (!path_has_protocol(filename)) {
432 return bdrv_find_format("file");
434 p = strchr(filename, ':');
437 if (len > sizeof(protocol) - 1)
438 len = sizeof(protocol) - 1;
439 memcpy(protocol, filename, len);
440 protocol[len] = '\0';
441 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
442 if (drv1->protocol_name &&
443 !strcmp(drv1->protocol_name, protocol)) {
450 static int find_image_format(const char *filename, BlockDriver **pdrv)
452 int ret, score, score_max;
453 BlockDriver *drv1, *drv;
455 BlockDriverState *bs;
457 ret = bdrv_file_open(&bs, filename, 0);
463 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
464 if (bs->sg || !bdrv_is_inserted(bs)) {
466 drv = bdrv_find_format("raw");
474 ret = bdrv_pread(bs, 0, buf, sizeof(buf));
483 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
484 if (drv1->bdrv_probe) {
485 score = drv1->bdrv_probe(buf, ret, filename);
486 if (score > score_max) {
500 * Set the current 'total_sectors' value
502 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
504 BlockDriver *drv = bs->drv;
506 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
510 /* query actual device if possible, otherwise just trust the hint */
511 if (drv->bdrv_getlength) {
512 int64_t length = drv->bdrv_getlength(bs);
516 hint = length >> BDRV_SECTOR_BITS;
519 bs->total_sectors = hint;
524 * Set open flags for a given cache mode
526 * Return 0 on success, -1 if the cache mode was invalid.
528 int bdrv_parse_cache_flags(const char *mode, int *flags)
530 *flags &= ~BDRV_O_CACHE_MASK;
532 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
533 *flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB;
534 } else if (!strcmp(mode, "directsync")) {
535 *flags |= BDRV_O_NOCACHE;
536 } else if (!strcmp(mode, "writeback")) {
537 *flags |= BDRV_O_CACHE_WB;
538 } else if (!strcmp(mode, "unsafe")) {
539 *flags |= BDRV_O_CACHE_WB;
540 *flags |= BDRV_O_NO_FLUSH;
541 } else if (!strcmp(mode, "writethrough")) {
542 /* this is the default */
551 * The copy-on-read flag is actually a reference count so multiple users may
552 * use the feature without worrying about clobbering its previous state.
553 * Copy-on-read stays enabled until all users have called to disable it.
555 void bdrv_enable_copy_on_read(BlockDriverState *bs)
560 void bdrv_disable_copy_on_read(BlockDriverState *bs)
562 assert(bs->copy_on_read > 0);
567 * Common part for opening disk images and files
569 static int bdrv_open_common(BlockDriverState *bs, const char *filename,
570 int flags, BlockDriver *drv)
576 trace_bdrv_open_common(bs, filename, flags, drv->format_name);
579 bs->total_sectors = 0;
583 bs->open_flags = flags;
585 bs->buffer_alignment = 512;
587 assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
588 if ((flags & BDRV_O_RDWR) && (flags & BDRV_O_COPY_ON_READ)) {
589 bdrv_enable_copy_on_read(bs);
592 pstrcpy(bs->filename, sizeof(bs->filename), filename);
593 bs->backing_file[0] = '\0';
595 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv)) {
600 bs->opaque = g_malloc0(drv->instance_size);
602 bs->enable_write_cache = !!(flags & BDRV_O_CACHE_WB);
605 * Clear flags that are internal to the block layer before opening the
608 open_flags = flags & ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
611 * Snapshots should be writable.
613 if (bs->is_temporary) {
614 open_flags |= BDRV_O_RDWR;
617 bs->keep_read_only = bs->read_only = !(open_flags & BDRV_O_RDWR);
619 /* Open the image, either directly or using a protocol */
620 if (drv->bdrv_file_open) {
621 ret = drv->bdrv_file_open(bs, filename, open_flags);
623 ret = bdrv_file_open(&bs->file, filename, open_flags);
625 ret = drv->bdrv_open(bs, open_flags);
633 ret = refresh_total_sectors(bs, bs->total_sectors);
639 if (bs->is_temporary) {
647 bdrv_delete(bs->file);
657 * Opens a file using a protocol (file, host_device, nbd, ...)
659 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
661 BlockDriverState *bs;
665 drv = bdrv_find_protocol(filename);
671 ret = bdrv_open_common(bs, filename, flags, drv);
682 * Opens a disk image (raw, qcow2, vmdk, ...)
684 int bdrv_open(BlockDriverState *bs, const char *filename, int flags,
688 char tmp_filename[PATH_MAX];
690 if (flags & BDRV_O_SNAPSHOT) {
691 BlockDriverState *bs1;
694 BlockDriver *bdrv_qcow2;
695 QEMUOptionParameter *options;
696 char backing_filename[PATH_MAX];
698 /* if snapshot, we create a temporary backing file and open it
699 instead of opening 'filename' directly */
701 /* if there is a backing file, use it */
703 ret = bdrv_open(bs1, filename, 0, drv);
708 total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
710 if (bs1->drv && bs1->drv->protocol_name)
715 get_tmp_filename(tmp_filename, sizeof(tmp_filename));
717 /* Real path is meaningless for protocols */
719 snprintf(backing_filename, sizeof(backing_filename),
721 else if (!realpath(filename, backing_filename))
724 bdrv_qcow2 = bdrv_find_format("qcow2");
725 options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
727 set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size);
728 set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
730 set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
734 ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
735 free_option_parameters(options);
740 filename = tmp_filename;
742 bs->is_temporary = 1;
745 /* Find the right image format driver */
747 ret = find_image_format(filename, &drv);
751 goto unlink_and_fail;
755 ret = bdrv_open_common(bs, filename, flags, drv);
757 goto unlink_and_fail;
760 /* If there is a backing file, use it */
761 if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') {
762 char backing_filename[PATH_MAX];
764 BlockDriver *back_drv = NULL;
766 bs->backing_hd = bdrv_new("");
768 if (path_has_protocol(bs->backing_file)) {
769 pstrcpy(backing_filename, sizeof(backing_filename),
772 path_combine(backing_filename, sizeof(backing_filename),
773 filename, bs->backing_file);
776 if (bs->backing_format[0] != '\0') {
777 back_drv = bdrv_find_format(bs->backing_format);
780 /* backing files always opened read-only */
782 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
784 ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv);
789 if (bs->is_temporary) {
790 bs->backing_hd->keep_read_only = !(flags & BDRV_O_RDWR);
792 /* base image inherits from "parent" */
793 bs->backing_hd->keep_read_only = bs->keep_read_only;
797 if (!bdrv_key_required(bs)) {
798 bdrv_dev_change_media_cb(bs, true);
801 /* throttling disk I/O limits */
802 if (bs->io_limits_enabled) {
803 bdrv_io_limits_enable(bs);
809 if (bs->is_temporary) {
815 void bdrv_close(BlockDriverState *bs)
820 block_job_cancel_sync(bs->job);
824 if (bs == bs_snapshots) {
827 if (bs->backing_hd) {
828 bdrv_delete(bs->backing_hd);
829 bs->backing_hd = NULL;
831 bs->drv->bdrv_close(bs);
834 if (bs->is_temporary) {
835 unlink(bs->filename);
840 bs->copy_on_read = 0;
842 if (bs->file != NULL) {
843 bdrv_close(bs->file);
846 bdrv_dev_change_media_cb(bs, false);
849 /*throttling disk I/O limits*/
850 if (bs->io_limits_enabled) {
851 bdrv_io_limits_disable(bs);
855 void bdrv_close_all(void)
857 BlockDriverState *bs;
859 QTAILQ_FOREACH(bs, &bdrv_states, list) {
865 * Wait for pending requests to complete across all BlockDriverStates
867 * This function does not flush data to disk, use bdrv_flush_all() for that
868 * after calling this function.
870 void bdrv_drain_all(void)
872 BlockDriverState *bs;
876 /* If requests are still pending there is a bug somewhere */
877 QTAILQ_FOREACH(bs, &bdrv_states, list) {
878 assert(QLIST_EMPTY(&bs->tracked_requests));
879 assert(qemu_co_queue_empty(&bs->throttled_reqs));
883 /* make a BlockDriverState anonymous by removing from bdrv_state list.
884 Also, NULL terminate the device_name to prevent double remove */
885 void bdrv_make_anon(BlockDriverState *bs)
887 if (bs->device_name[0] != '\0') {
888 QTAILQ_REMOVE(&bdrv_states, bs, list);
890 bs->device_name[0] = '\0';
894 * Add new bs contents at the top of an image chain while the chain is
895 * live, while keeping required fields on the top layer.
897 * This will modify the BlockDriverState fields, and swap contents
898 * between bs_new and bs_top. Both bs_new and bs_top are modified.
900 * bs_new is required to be anonymous.
902 * This function does not create any image files.
904 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
906 BlockDriverState tmp;
908 /* bs_new must be anonymous */
909 assert(bs_new->device_name[0] == '\0');
913 /* there are some fields that need to stay on the top layer: */
916 tmp.dev_ops = bs_top->dev_ops;
917 tmp.dev_opaque = bs_top->dev_opaque;
918 tmp.dev = bs_top->dev;
919 tmp.buffer_alignment = bs_top->buffer_alignment;
920 tmp.copy_on_read = bs_top->copy_on_read;
922 /* i/o timing parameters */
923 tmp.slice_time = bs_top->slice_time;
924 tmp.slice_start = bs_top->slice_start;
925 tmp.slice_end = bs_top->slice_end;
926 tmp.io_limits = bs_top->io_limits;
927 tmp.io_base = bs_top->io_base;
928 tmp.throttled_reqs = bs_top->throttled_reqs;
929 tmp.block_timer = bs_top->block_timer;
930 tmp.io_limits_enabled = bs_top->io_limits_enabled;
933 tmp.cyls = bs_top->cyls;
934 tmp.heads = bs_top->heads;
935 tmp.secs = bs_top->secs;
936 tmp.translation = bs_top->translation;
939 tmp.on_read_error = bs_top->on_read_error;
940 tmp.on_write_error = bs_top->on_write_error;
943 tmp.iostatus_enabled = bs_top->iostatus_enabled;
944 tmp.iostatus = bs_top->iostatus;
946 /* keep the same entry in bdrv_states */
947 pstrcpy(tmp.device_name, sizeof(tmp.device_name), bs_top->device_name);
948 tmp.list = bs_top->list;
950 /* The contents of 'tmp' will become bs_top, as we are
951 * swapping bs_new and bs_top contents. */
952 tmp.backing_hd = bs_new;
953 pstrcpy(tmp.backing_file, sizeof(tmp.backing_file), bs_top->filename);
954 bdrv_get_format(bs_top, tmp.backing_format, sizeof(tmp.backing_format));
956 /* swap contents of the fixed new bs and the current top */
960 /* device_name[] was carried over from the old bs_top. bs_new
961 * shouldn't be in bdrv_states, so we need to make device_name[]
962 * reflect the anonymity of bs_new
964 bs_new->device_name[0] = '\0';
966 /* clear the copied fields in the new backing file */
967 bdrv_detach_dev(bs_new, bs_new->dev);
969 qemu_co_queue_init(&bs_new->throttled_reqs);
970 memset(&bs_new->io_base, 0, sizeof(bs_new->io_base));
971 memset(&bs_new->io_limits, 0, sizeof(bs_new->io_limits));
972 bdrv_iostatus_disable(bs_new);
974 /* we don't use bdrv_io_limits_disable() for this, because we don't want
975 * to affect or delete the block_timer, as it has been moved to bs_top */
976 bs_new->io_limits_enabled = false;
977 bs_new->block_timer = NULL;
978 bs_new->slice_time = 0;
979 bs_new->slice_start = 0;
980 bs_new->slice_end = 0;
983 void bdrv_delete(BlockDriverState *bs)
989 /* remove from list, if necessary */
993 if (bs->file != NULL) {
994 bdrv_delete(bs->file);
997 assert(bs != bs_snapshots);
1001 int bdrv_attach_dev(BlockDriverState *bs, void *dev)
1002 /* TODO change to DeviceState *dev when all users are qdevified */
1008 bdrv_iostatus_reset(bs);
1012 /* TODO qdevified devices don't use this, remove when devices are qdevified */
1013 void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev)
1015 if (bdrv_attach_dev(bs, dev) < 0) {
1020 void bdrv_detach_dev(BlockDriverState *bs, void *dev)
1021 /* TODO change to DeviceState *dev when all users are qdevified */
1023 assert(bs->dev == dev);
1026 bs->dev_opaque = NULL;
1027 bs->buffer_alignment = 512;
1030 /* TODO change to return DeviceState * when all users are qdevified */
1031 void *bdrv_get_attached_dev(BlockDriverState *bs)
1036 void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops,
1040 bs->dev_opaque = opaque;
1041 if (bdrv_dev_has_removable_media(bs) && bs == bs_snapshots) {
1042 bs_snapshots = NULL;
1046 void bdrv_emit_qmp_error_event(const BlockDriverState *bdrv,
1047 BlockQMPEventAction action, int is_read)
1050 const char *action_str;
1053 case BDRV_ACTION_REPORT:
1054 action_str = "report";
1056 case BDRV_ACTION_IGNORE:
1057 action_str = "ignore";
1059 case BDRV_ACTION_STOP:
1060 action_str = "stop";
1066 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1069 is_read ? "read" : "write");
1070 monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
1072 qobject_decref(data);
1075 static void bdrv_emit_qmp_eject_event(BlockDriverState *bs, bool ejected)
1079 data = qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }",
1080 bdrv_get_device_name(bs), ejected);
1081 monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED, data);
1083 qobject_decref(data);
1086 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load)
1088 if (bs->dev_ops && bs->dev_ops->change_media_cb) {
1089 bool tray_was_closed = !bdrv_dev_is_tray_open(bs);
1090 bs->dev_ops->change_media_cb(bs->dev_opaque, load);
1091 if (tray_was_closed) {
1093 bdrv_emit_qmp_eject_event(bs, true);
1097 bdrv_emit_qmp_eject_event(bs, false);
1102 bool bdrv_dev_has_removable_media(BlockDriverState *bs)
1104 return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb);
1107 void bdrv_dev_eject_request(BlockDriverState *bs, bool force)
1109 if (bs->dev_ops && bs->dev_ops->eject_request_cb) {
1110 bs->dev_ops->eject_request_cb(bs->dev_opaque, force);
1114 bool bdrv_dev_is_tray_open(BlockDriverState *bs)
1116 if (bs->dev_ops && bs->dev_ops->is_tray_open) {
1117 return bs->dev_ops->is_tray_open(bs->dev_opaque);
1122 static void bdrv_dev_resize_cb(BlockDriverState *bs)
1124 if (bs->dev_ops && bs->dev_ops->resize_cb) {
1125 bs->dev_ops->resize_cb(bs->dev_opaque);
1129 bool bdrv_dev_is_medium_locked(BlockDriverState *bs)
1131 if (bs->dev_ops && bs->dev_ops->is_medium_locked) {
1132 return bs->dev_ops->is_medium_locked(bs->dev_opaque);
1138 * Run consistency checks on an image
1140 * Returns 0 if the check could be completed (it doesn't mean that the image is
1141 * free of errors) or -errno when an internal error occurred. The results of the
1142 * check are stored in res.
1144 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res)
1146 if (bs->drv->bdrv_check == NULL) {
1150 memset(res, 0, sizeof(*res));
1151 return bs->drv->bdrv_check(bs, res);
1154 #define COMMIT_BUF_SECTORS 2048
1156 /* commit COW file into the raw image */
1157 int bdrv_commit(BlockDriverState *bs)
1159 BlockDriver *drv = bs->drv;
1160 BlockDriver *backing_drv;
1161 int64_t sector, total_sectors;
1162 int n, ro, open_flags;
1163 int ret = 0, rw_ret = 0;
1165 char filename[1024];
1166 BlockDriverState *bs_rw, *bs_ro;
1171 if (!bs->backing_hd) {
1175 if (bs->backing_hd->keep_read_only) {
1179 if (bdrv_in_use(bs) || bdrv_in_use(bs->backing_hd)) {
1183 backing_drv = bs->backing_hd->drv;
1184 ro = bs->backing_hd->read_only;
1185 strncpy(filename, bs->backing_hd->filename, sizeof(filename));
1186 open_flags = bs->backing_hd->open_flags;
1190 bdrv_delete(bs->backing_hd);
1191 bs->backing_hd = NULL;
1192 bs_rw = bdrv_new("");
1193 rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR,
1197 /* try to re-open read-only */
1198 bs_ro = bdrv_new("");
1199 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
1203 /* drive not functional anymore */
1207 bs->backing_hd = bs_ro;
1210 bs->backing_hd = bs_rw;
1213 total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
1214 buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
1216 for (sector = 0; sector < total_sectors; sector += n) {
1217 if (bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n)) {
1219 if (bdrv_read(bs, sector, buf, n) != 0) {
1224 if (bdrv_write(bs->backing_hd, sector, buf, n) != 0) {
1231 if (drv->bdrv_make_empty) {
1232 ret = drv->bdrv_make_empty(bs);
1237 * Make sure all data we wrote to the backing device is actually
1241 bdrv_flush(bs->backing_hd);
1248 bdrv_delete(bs->backing_hd);
1249 bs->backing_hd = NULL;
1250 bs_ro = bdrv_new("");
1251 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
1255 /* drive not functional anymore */
1259 bs->backing_hd = bs_ro;
1260 bs->backing_hd->keep_read_only = 0;
1266 int bdrv_commit_all(void)
1268 BlockDriverState *bs;
1270 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1271 int ret = bdrv_commit(bs);
1279 struct BdrvTrackedRequest {
1280 BlockDriverState *bs;
1284 QLIST_ENTRY(BdrvTrackedRequest) list;
1285 Coroutine *co; /* owner, used for deadlock detection */
1286 CoQueue wait_queue; /* coroutines blocked on this request */
1290 * Remove an active request from the tracked requests list
1292 * This function should be called when a tracked request is completing.
1294 static void tracked_request_end(BdrvTrackedRequest *req)
1296 QLIST_REMOVE(req, list);
1297 qemu_co_queue_restart_all(&req->wait_queue);
1301 * Add an active request to the tracked requests list
1303 static void tracked_request_begin(BdrvTrackedRequest *req,
1304 BlockDriverState *bs,
1306 int nb_sectors, bool is_write)
1308 *req = (BdrvTrackedRequest){
1310 .sector_num = sector_num,
1311 .nb_sectors = nb_sectors,
1312 .is_write = is_write,
1313 .co = qemu_coroutine_self(),
1316 qemu_co_queue_init(&req->wait_queue);
1318 QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
1322 * Round a region to cluster boundaries
1324 static void round_to_clusters(BlockDriverState *bs,
1325 int64_t sector_num, int nb_sectors,
1326 int64_t *cluster_sector_num,
1327 int *cluster_nb_sectors)
1329 BlockDriverInfo bdi;
1331 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
1332 *cluster_sector_num = sector_num;
1333 *cluster_nb_sectors = nb_sectors;
1335 int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE;
1336 *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c);
1337 *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num +
1342 static bool tracked_request_overlaps(BdrvTrackedRequest *req,
1343 int64_t sector_num, int nb_sectors) {
1345 if (sector_num >= req->sector_num + req->nb_sectors) {
1349 if (req->sector_num >= sector_num + nb_sectors) {
1355 static void coroutine_fn wait_for_overlapping_requests(BlockDriverState *bs,
1356 int64_t sector_num, int nb_sectors)
1358 BdrvTrackedRequest *req;
1359 int64_t cluster_sector_num;
1360 int cluster_nb_sectors;
1363 /* If we touch the same cluster it counts as an overlap. This guarantees
1364 * that allocating writes will be serialized and not race with each other
1365 * for the same cluster. For example, in copy-on-read it ensures that the
1366 * CoR read and write operations are atomic and guest writes cannot
1367 * interleave between them.
1369 round_to_clusters(bs, sector_num, nb_sectors,
1370 &cluster_sector_num, &cluster_nb_sectors);
1374 QLIST_FOREACH(req, &bs->tracked_requests, list) {
1375 if (tracked_request_overlaps(req, cluster_sector_num,
1376 cluster_nb_sectors)) {
1377 /* Hitting this means there was a reentrant request, for
1378 * example, a block driver issuing nested requests. This must
1379 * never happen since it means deadlock.
1381 assert(qemu_coroutine_self() != req->co);
1383 qemu_co_queue_wait(&req->wait_queue);
1394 * -EINVAL - backing format specified, but no file
1395 * -ENOSPC - can't update the backing file because no space is left in the
1397 * -ENOTSUP - format driver doesn't support changing the backing file
1399 int bdrv_change_backing_file(BlockDriverState *bs,
1400 const char *backing_file, const char *backing_fmt)
1402 BlockDriver *drv = bs->drv;
1404 if (drv->bdrv_change_backing_file != NULL) {
1405 return drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
1411 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
1416 if (!bdrv_is_inserted(bs))
1422 len = bdrv_getlength(bs);
1427 if ((offset > len) || (len - offset < size))
1433 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
1436 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
1437 nb_sectors * BDRV_SECTOR_SIZE);
1440 typedef struct RwCo {
1441 BlockDriverState *bs;
1449 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
1451 RwCo *rwco = opaque;
1453 if (!rwco->is_write) {
1454 rwco->ret = bdrv_co_do_readv(rwco->bs, rwco->sector_num,
1455 rwco->nb_sectors, rwco->qiov, 0);
1457 rwco->ret = bdrv_co_do_writev(rwco->bs, rwco->sector_num,
1458 rwco->nb_sectors, rwco->qiov, 0);
1463 * Process a synchronous request using coroutines
1465 static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf,
1466 int nb_sectors, bool is_write)
1469 struct iovec iov = {
1470 .iov_base = (void *)buf,
1471 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
1476 .sector_num = sector_num,
1477 .nb_sectors = nb_sectors,
1479 .is_write = is_write,
1483 qemu_iovec_init_external(&qiov, &iov, 1);
1486 * In sync call context, when the vcpu is blocked, this throttling timer
1487 * will not fire; so the I/O throttling function has to be disabled here
1488 * if it has been enabled.
1490 if (bs->io_limits_enabled) {
1491 fprintf(stderr, "Disabling I/O throttling on '%s' due "
1492 "to synchronous I/O.\n", bdrv_get_device_name(bs));
1493 bdrv_io_limits_disable(bs);
1496 if (qemu_in_coroutine()) {
1497 /* Fast-path if already in coroutine context */
1498 bdrv_rw_co_entry(&rwco);
1500 co = qemu_coroutine_create(bdrv_rw_co_entry);
1501 qemu_coroutine_enter(co, &rwco);
1502 while (rwco.ret == NOT_DONE) {
1509 /* return < 0 if error. See bdrv_write() for the return codes */
1510 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
1511 uint8_t *buf, int nb_sectors)
1513 return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false);
1516 static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
1517 int nb_sectors, int dirty)
1520 unsigned long val, idx, bit;
1522 start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
1523 end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
1525 for (; start <= end; start++) {
1526 idx = start / (sizeof(unsigned long) * 8);
1527 bit = start % (sizeof(unsigned long) * 8);
1528 val = bs->dirty_bitmap[idx];
1530 if (!(val & (1UL << bit))) {
1535 if (val & (1UL << bit)) {
1537 val &= ~(1UL << bit);
1540 bs->dirty_bitmap[idx] = val;
1544 /* Return < 0 if error. Important errors are:
1545 -EIO generic I/O error (may happen for all errors)
1546 -ENOMEDIUM No media inserted.
1547 -EINVAL Invalid sector number or nb_sectors
1548 -EACCES Trying to write a read-only device
1550 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
1551 const uint8_t *buf, int nb_sectors)
1553 return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true);
1556 int bdrv_pread(BlockDriverState *bs, int64_t offset,
1557 void *buf, int count1)
1559 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1560 int len, nb_sectors, count;
1565 /* first read to align to sector start */
1566 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1569 sector_num = offset >> BDRV_SECTOR_BITS;
1571 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1573 memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
1581 /* read the sectors "in place" */
1582 nb_sectors = count >> BDRV_SECTOR_BITS;
1583 if (nb_sectors > 0) {
1584 if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
1586 sector_num += nb_sectors;
1587 len = nb_sectors << BDRV_SECTOR_BITS;
1592 /* add data from the last sector */
1594 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1596 memcpy(buf, tmp_buf, count);
1601 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
1602 const void *buf, int count1)
1604 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1605 int len, nb_sectors, count;
1610 /* first write to align to sector start */
1611 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1614 sector_num = offset >> BDRV_SECTOR_BITS;
1616 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1618 memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
1619 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1628 /* write the sectors "in place" */
1629 nb_sectors = count >> BDRV_SECTOR_BITS;
1630 if (nb_sectors > 0) {
1631 if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0)
1633 sector_num += nb_sectors;
1634 len = nb_sectors << BDRV_SECTOR_BITS;
1639 /* add data from the last sector */
1641 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1643 memcpy(tmp_buf, buf, count);
1644 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1651 * Writes to the file and ensures that no writes are reordered across this
1652 * request (acts as a barrier)
1654 * Returns 0 on success, -errno in error cases.
1656 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
1657 const void *buf, int count)
1661 ret = bdrv_pwrite(bs, offset, buf, count);
1666 /* No flush needed for cache modes that use O_DSYNC */
1667 if ((bs->open_flags & BDRV_O_CACHE_WB) != 0) {
1674 static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs,
1675 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1677 /* Perform I/O through a temporary buffer so that users who scribble over
1678 * their read buffer while the operation is in progress do not end up
1679 * modifying the image file. This is critical for zero-copy guest I/O
1680 * where anything might happen inside guest memory.
1682 void *bounce_buffer;
1684 BlockDriver *drv = bs->drv;
1686 QEMUIOVector bounce_qiov;
1687 int64_t cluster_sector_num;
1688 int cluster_nb_sectors;
1692 /* Cover entire cluster so no additional backing file I/O is required when
1693 * allocating cluster in the image file.
1695 round_to_clusters(bs, sector_num, nb_sectors,
1696 &cluster_sector_num, &cluster_nb_sectors);
1698 trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors,
1699 cluster_sector_num, cluster_nb_sectors);
1701 iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE;
1702 iov.iov_base = bounce_buffer = qemu_blockalign(bs, iov.iov_len);
1703 qemu_iovec_init_external(&bounce_qiov, &iov, 1);
1705 ret = drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors,
1711 if (drv->bdrv_co_write_zeroes &&
1712 buffer_is_zero(bounce_buffer, iov.iov_len)) {
1713 ret = bdrv_co_do_write_zeroes(bs, cluster_sector_num,
1714 cluster_nb_sectors);
1716 ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors,
1721 /* It might be okay to ignore write errors for guest requests. If this
1722 * is a deliberate copy-on-read then we don't want to ignore the error.
1723 * Simply report it in all cases.
1728 skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE;
1729 qemu_iovec_from_buffer(qiov, bounce_buffer + skip_bytes,
1730 nb_sectors * BDRV_SECTOR_SIZE);
1733 qemu_vfree(bounce_buffer);
1738 * Handle a read request in coroutine context
1740 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
1741 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1742 BdrvRequestFlags flags)
1744 BlockDriver *drv = bs->drv;
1745 BdrvTrackedRequest req;
1751 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
1755 /* throttling disk read I/O */
1756 if (bs->io_limits_enabled) {
1757 bdrv_io_limits_intercept(bs, false, nb_sectors);
1760 if (bs->copy_on_read) {
1761 flags |= BDRV_REQ_COPY_ON_READ;
1763 if (flags & BDRV_REQ_COPY_ON_READ) {
1764 bs->copy_on_read_in_flight++;
1767 if (bs->copy_on_read_in_flight) {
1768 wait_for_overlapping_requests(bs, sector_num, nb_sectors);
1771 tracked_request_begin(&req, bs, sector_num, nb_sectors, false);
1773 if (flags & BDRV_REQ_COPY_ON_READ) {
1776 ret = bdrv_co_is_allocated(bs, sector_num, nb_sectors, &pnum);
1781 if (!ret || pnum != nb_sectors) {
1782 ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov);
1787 ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
1790 tracked_request_end(&req);
1792 if (flags & BDRV_REQ_COPY_ON_READ) {
1793 bs->copy_on_read_in_flight--;
1799 int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
1800 int nb_sectors, QEMUIOVector *qiov)
1802 trace_bdrv_co_readv(bs, sector_num, nb_sectors);
1804 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0);
1807 int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
1808 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1810 trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors);
1812 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
1813 BDRV_REQ_COPY_ON_READ);
1816 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
1817 int64_t sector_num, int nb_sectors)
1819 BlockDriver *drv = bs->drv;
1824 /* TODO Emulate only part of misaligned requests instead of letting block
1825 * drivers return -ENOTSUP and emulate everything */
1827 /* First try the efficient write zeroes operation */
1828 if (drv->bdrv_co_write_zeroes) {
1829 ret = drv->bdrv_co_write_zeroes(bs, sector_num, nb_sectors);
1830 if (ret != -ENOTSUP) {
1835 /* Fall back to bounce buffer if write zeroes is unsupported */
1836 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
1837 iov.iov_base = qemu_blockalign(bs, iov.iov_len);
1838 memset(iov.iov_base, 0, iov.iov_len);
1839 qemu_iovec_init_external(&qiov, &iov, 1);
1841 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, &qiov);
1843 qemu_vfree(iov.iov_base);
1848 * Handle a write request in coroutine context
1850 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
1851 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1852 BdrvRequestFlags flags)
1854 BlockDriver *drv = bs->drv;
1855 BdrvTrackedRequest req;
1861 if (bs->read_only) {
1864 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
1868 /* throttling disk write I/O */
1869 if (bs->io_limits_enabled) {
1870 bdrv_io_limits_intercept(bs, true, nb_sectors);
1873 if (bs->copy_on_read_in_flight) {
1874 wait_for_overlapping_requests(bs, sector_num, nb_sectors);
1877 tracked_request_begin(&req, bs, sector_num, nb_sectors, true);
1879 if (flags & BDRV_REQ_ZERO_WRITE) {
1880 ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors);
1882 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
1885 if (bs->dirty_bitmap) {
1886 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1889 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
1890 bs->wr_highest_sector = sector_num + nb_sectors - 1;
1893 tracked_request_end(&req);
1898 int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
1899 int nb_sectors, QEMUIOVector *qiov)
1901 trace_bdrv_co_writev(bs, sector_num, nb_sectors);
1903 return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0);
1906 int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs,
1907 int64_t sector_num, int nb_sectors)
1909 trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors);
1911 return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL,
1912 BDRV_REQ_ZERO_WRITE);
1916 * Truncate file to 'offset' bytes (needed only for file protocols)
1918 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
1920 BlockDriver *drv = bs->drv;
1924 if (!drv->bdrv_truncate)
1928 if (bdrv_in_use(bs))
1930 ret = drv->bdrv_truncate(bs, offset);
1932 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
1933 bdrv_dev_resize_cb(bs);
1939 * Length of a allocated file in bytes. Sparse files are counted by actual
1940 * allocated space. Return < 0 if error or unknown.
1942 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
1944 BlockDriver *drv = bs->drv;
1948 if (drv->bdrv_get_allocated_file_size) {
1949 return drv->bdrv_get_allocated_file_size(bs);
1952 return bdrv_get_allocated_file_size(bs->file);
1958 * Length of a file in bytes. Return < 0 if error or unknown.
1960 int64_t bdrv_getlength(BlockDriverState *bs)
1962 BlockDriver *drv = bs->drv;
1966 if (bs->growable || bdrv_dev_has_removable_media(bs)) {
1967 if (drv->bdrv_getlength) {
1968 return drv->bdrv_getlength(bs);
1971 return bs->total_sectors * BDRV_SECTOR_SIZE;
1974 /* return 0 as number of sectors if no device present or error */
1975 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
1978 length = bdrv_getlength(bs);
1982 length = length >> BDRV_SECTOR_BITS;
1983 *nb_sectors_ptr = length;
1987 uint8_t boot_ind; /* 0x80 - active */
1988 uint8_t head; /* starting head */
1989 uint8_t sector; /* starting sector */
1990 uint8_t cyl; /* starting cylinder */
1991 uint8_t sys_ind; /* What partition type */
1992 uint8_t end_head; /* end head */
1993 uint8_t end_sector; /* end sector */
1994 uint8_t end_cyl; /* end cylinder */
1995 uint32_t start_sect; /* starting sector counting from 0 */
1996 uint32_t nr_sects; /* nr of sectors in partition */
1999 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
2000 static int guess_disk_lchs(BlockDriverState *bs,
2001 int *pcylinders, int *pheads, int *psectors)
2003 uint8_t buf[BDRV_SECTOR_SIZE];
2004 int ret, i, heads, sectors, cylinders;
2005 struct partition *p;
2007 uint64_t nb_sectors;
2010 bdrv_get_geometry(bs, &nb_sectors);
2013 * The function will be invoked during startup not only in sync I/O mode,
2014 * but also in async I/O mode. So the I/O throttling function has to
2015 * be disabled temporarily here, not permanently.
2017 enabled = bs->io_limits_enabled;
2018 bs->io_limits_enabled = false;
2019 ret = bdrv_read(bs, 0, buf, 1);
2020 bs->io_limits_enabled = enabled;
2023 /* test msdos magic */
2024 if (buf[510] != 0x55 || buf[511] != 0xaa)
2026 for(i = 0; i < 4; i++) {
2027 p = ((struct partition *)(buf + 0x1be)) + i;
2028 nr_sects = le32_to_cpu(p->nr_sects);
2029 if (nr_sects && p->end_head) {
2030 /* We make the assumption that the partition terminates on
2031 a cylinder boundary */
2032 heads = p->end_head + 1;
2033 sectors = p->end_sector & 63;
2036 cylinders = nb_sectors / (heads * sectors);
2037 if (cylinders < 1 || cylinders > 16383)
2040 *psectors = sectors;
2041 *pcylinders = cylinders;
2043 printf("guessed geometry: LCHS=%d %d %d\n",
2044 cylinders, heads, sectors);
2052 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
2054 int translation, lba_detected = 0;
2055 int cylinders, heads, secs;
2056 uint64_t nb_sectors;
2058 /* if a geometry hint is available, use it */
2059 bdrv_get_geometry(bs, &nb_sectors);
2060 bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
2061 translation = bdrv_get_translation_hint(bs);
2062 if (cylinders != 0) {
2067 if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
2069 /* if heads > 16, it means that a BIOS LBA
2070 translation was active, so the default
2071 hardware geometry is OK */
2073 goto default_geometry;
2078 /* disable any translation to be in sync with
2079 the logical geometry */
2080 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
2081 bdrv_set_translation_hint(bs,
2082 BIOS_ATA_TRANSLATION_NONE);
2087 /* if no geometry, use a standard physical disk geometry */
2088 cylinders = nb_sectors / (16 * 63);
2090 if (cylinders > 16383)
2092 else if (cylinders < 2)
2097 if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
2098 if ((*pcyls * *pheads) <= 131072) {
2099 bdrv_set_translation_hint(bs,
2100 BIOS_ATA_TRANSLATION_LARGE);
2102 bdrv_set_translation_hint(bs,
2103 BIOS_ATA_TRANSLATION_LBA);
2107 bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
2111 void bdrv_set_geometry_hint(BlockDriverState *bs,
2112 int cyls, int heads, int secs)
2119 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
2121 bs->translation = translation;
2124 void bdrv_get_geometry_hint(BlockDriverState *bs,
2125 int *pcyls, int *pheads, int *psecs)
2128 *pheads = bs->heads;
2132 /* throttling disk io limits */
2133 void bdrv_set_io_limits(BlockDriverState *bs,
2134 BlockIOLimit *io_limits)
2136 bs->io_limits = *io_limits;
2137 bs->io_limits_enabled = bdrv_io_limits_enabled(bs);
2140 /* Recognize floppy formats */
2141 typedef struct FDFormat {
2149 static const FDFormat fd_formats[] = {
2150 /* First entry is default format */
2151 /* 1.44 MB 3"1/2 floppy disks */
2152 { FDRIVE_DRV_144, 18, 80, 1, FDRIVE_RATE_500K, },
2153 { FDRIVE_DRV_144, 20, 80, 1, FDRIVE_RATE_500K, },
2154 { FDRIVE_DRV_144, 21, 80, 1, FDRIVE_RATE_500K, },
2155 { FDRIVE_DRV_144, 21, 82, 1, FDRIVE_RATE_500K, },
2156 { FDRIVE_DRV_144, 21, 83, 1, FDRIVE_RATE_500K, },
2157 { FDRIVE_DRV_144, 22, 80, 1, FDRIVE_RATE_500K, },
2158 { FDRIVE_DRV_144, 23, 80, 1, FDRIVE_RATE_500K, },
2159 { FDRIVE_DRV_144, 24, 80, 1, FDRIVE_RATE_500K, },
2160 /* 2.88 MB 3"1/2 floppy disks */
2161 { FDRIVE_DRV_288, 36, 80, 1, FDRIVE_RATE_1M, },
2162 { FDRIVE_DRV_288, 39, 80, 1, FDRIVE_RATE_1M, },
2163 { FDRIVE_DRV_288, 40, 80, 1, FDRIVE_RATE_1M, },
2164 { FDRIVE_DRV_288, 44, 80, 1, FDRIVE_RATE_1M, },
2165 { FDRIVE_DRV_288, 48, 80, 1, FDRIVE_RATE_1M, },
2166 /* 720 kB 3"1/2 floppy disks */
2167 { FDRIVE_DRV_144, 9, 80, 1, FDRIVE_RATE_250K, },
2168 { FDRIVE_DRV_144, 10, 80, 1, FDRIVE_RATE_250K, },
2169 { FDRIVE_DRV_144, 10, 82, 1, FDRIVE_RATE_250K, },
2170 { FDRIVE_DRV_144, 10, 83, 1, FDRIVE_RATE_250K, },
2171 { FDRIVE_DRV_144, 13, 80, 1, FDRIVE_RATE_250K, },
2172 { FDRIVE_DRV_144, 14, 80, 1, FDRIVE_RATE_250K, },
2173 /* 1.2 MB 5"1/4 floppy disks */
2174 { FDRIVE_DRV_120, 15, 80, 1, FDRIVE_RATE_500K, },
2175 { FDRIVE_DRV_120, 18, 80, 1, FDRIVE_RATE_500K, },
2176 { FDRIVE_DRV_120, 18, 82, 1, FDRIVE_RATE_500K, },
2177 { FDRIVE_DRV_120, 18, 83, 1, FDRIVE_RATE_500K, },
2178 { FDRIVE_DRV_120, 20, 80, 1, FDRIVE_RATE_500K, },
2179 /* 720 kB 5"1/4 floppy disks */
2180 { FDRIVE_DRV_120, 9, 80, 1, FDRIVE_RATE_250K, },
2181 { FDRIVE_DRV_120, 11, 80, 1, FDRIVE_RATE_250K, },
2182 /* 360 kB 5"1/4 floppy disks */
2183 { FDRIVE_DRV_120, 9, 40, 1, FDRIVE_RATE_300K, },
2184 { FDRIVE_DRV_120, 9, 40, 0, FDRIVE_RATE_300K, },
2185 { FDRIVE_DRV_120, 10, 41, 1, FDRIVE_RATE_300K, },
2186 { FDRIVE_DRV_120, 10, 42, 1, FDRIVE_RATE_300K, },
2187 /* 320 kB 5"1/4 floppy disks */
2188 { FDRIVE_DRV_120, 8, 40, 1, FDRIVE_RATE_250K, },
2189 { FDRIVE_DRV_120, 8, 40, 0, FDRIVE_RATE_250K, },
2190 /* 360 kB must match 5"1/4 better than 3"1/2... */
2191 { FDRIVE_DRV_144, 9, 80, 0, FDRIVE_RATE_250K, },
2193 { FDRIVE_DRV_NONE, -1, -1, 0, 0, },
2196 void bdrv_get_floppy_geometry_hint(BlockDriverState *bs, int *nb_heads,
2197 int *max_track, int *last_sect,
2198 FDriveType drive_in, FDriveType *drive,
2201 const FDFormat *parse;
2202 uint64_t nb_sectors, size;
2203 int i, first_match, match;
2205 bdrv_get_geometry_hint(bs, nb_heads, max_track, last_sect);
2206 if (*nb_heads != 0 && *max_track != 0 && *last_sect != 0) {
2207 /* User defined disk */
2208 *rate = FDRIVE_RATE_500K;
2210 bdrv_get_geometry(bs, &nb_sectors);
2213 for (i = 0; ; i++) {
2214 parse = &fd_formats[i];
2215 if (parse->drive == FDRIVE_DRV_NONE) {
2218 if (drive_in == parse->drive ||
2219 drive_in == FDRIVE_DRV_NONE) {
2220 size = (parse->max_head + 1) * parse->max_track *
2222 if (nb_sectors == size) {
2226 if (first_match == -1) {
2232 if (first_match == -1) {
2235 match = first_match;
2237 parse = &fd_formats[match];
2239 *nb_heads = parse->max_head + 1;
2240 *max_track = parse->max_track;
2241 *last_sect = parse->last_sect;
2242 *drive = parse->drive;
2243 *rate = parse->rate;
2247 int bdrv_get_translation_hint(BlockDriverState *bs)
2249 return bs->translation;
2252 void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
2253 BlockErrorAction on_write_error)
2255 bs->on_read_error = on_read_error;
2256 bs->on_write_error = on_write_error;
2259 BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
2261 return is_read ? bs->on_read_error : bs->on_write_error;
2264 int bdrv_is_read_only(BlockDriverState *bs)
2266 return bs->read_only;
2269 int bdrv_is_sg(BlockDriverState *bs)
2274 int bdrv_enable_write_cache(BlockDriverState *bs)
2276 return bs->enable_write_cache;
2279 int bdrv_is_encrypted(BlockDriverState *bs)
2281 if (bs->backing_hd && bs->backing_hd->encrypted)
2283 return bs->encrypted;
2286 int bdrv_key_required(BlockDriverState *bs)
2288 BlockDriverState *backing_hd = bs->backing_hd;
2290 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
2292 return (bs->encrypted && !bs->valid_key);
2295 int bdrv_set_key(BlockDriverState *bs, const char *key)
2298 if (bs->backing_hd && bs->backing_hd->encrypted) {
2299 ret = bdrv_set_key(bs->backing_hd, key);
2305 if (!bs->encrypted) {
2307 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
2310 ret = bs->drv->bdrv_set_key(bs, key);
2313 } else if (!bs->valid_key) {
2315 /* call the change callback now, we skipped it on open */
2316 bdrv_dev_change_media_cb(bs, true);
2321 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
2326 pstrcpy(buf, buf_size, bs->drv->format_name);
2330 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
2335 QLIST_FOREACH(drv, &bdrv_drivers, list) {
2336 it(opaque, drv->format_name);
2340 BlockDriverState *bdrv_find(const char *name)
2342 BlockDriverState *bs;
2344 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2345 if (!strcmp(name, bs->device_name)) {
2352 BlockDriverState *bdrv_next(BlockDriverState *bs)
2355 return QTAILQ_FIRST(&bdrv_states);
2357 return QTAILQ_NEXT(bs, list);
2360 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
2362 BlockDriverState *bs;
2364 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2369 const char *bdrv_get_device_name(BlockDriverState *bs)
2371 return bs->device_name;
2374 void bdrv_flush_all(void)
2376 BlockDriverState *bs;
2378 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2383 int bdrv_has_zero_init(BlockDriverState *bs)
2387 if (bs->drv->bdrv_has_zero_init) {
2388 return bs->drv->bdrv_has_zero_init(bs);
2394 typedef struct BdrvCoIsAllocatedData {
2395 BlockDriverState *bs;
2401 } BdrvCoIsAllocatedData;
2404 * Returns true iff the specified sector is present in the disk image. Drivers
2405 * not implementing the functionality are assumed to not support backing files,
2406 * hence all their sectors are reported as allocated.
2408 * If 'sector_num' is beyond the end of the disk image the return value is 0
2409 * and 'pnum' is set to 0.
2411 * 'pnum' is set to the number of sectors (including and immediately following
2412 * the specified sector) that are known to be in the same
2413 * allocated/unallocated state.
2415 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
2416 * beyond the end of the disk image it will be clamped.
2418 int coroutine_fn bdrv_co_is_allocated(BlockDriverState *bs, int64_t sector_num,
2419 int nb_sectors, int *pnum)
2423 if (sector_num >= bs->total_sectors) {
2428 n = bs->total_sectors - sector_num;
2429 if (n < nb_sectors) {
2433 if (!bs->drv->bdrv_co_is_allocated) {
2438 return bs->drv->bdrv_co_is_allocated(bs, sector_num, nb_sectors, pnum);
2441 /* Coroutine wrapper for bdrv_is_allocated() */
2442 static void coroutine_fn bdrv_is_allocated_co_entry(void *opaque)
2444 BdrvCoIsAllocatedData *data = opaque;
2445 BlockDriverState *bs = data->bs;
2447 data->ret = bdrv_co_is_allocated(bs, data->sector_num, data->nb_sectors,
2453 * Synchronous wrapper around bdrv_co_is_allocated().
2455 * See bdrv_co_is_allocated() for details.
2457 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
2461 BdrvCoIsAllocatedData data = {
2463 .sector_num = sector_num,
2464 .nb_sectors = nb_sectors,
2469 co = qemu_coroutine_create(bdrv_is_allocated_co_entry);
2470 qemu_coroutine_enter(co, &data);
2471 while (!data.done) {
2477 BlockInfoList *qmp_query_block(Error **errp)
2479 BlockInfoList *head = NULL, *cur_item = NULL;
2480 BlockDriverState *bs;
2482 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2483 BlockInfoList *info = g_malloc0(sizeof(*info));
2485 info->value = g_malloc0(sizeof(*info->value));
2486 info->value->device = g_strdup(bs->device_name);
2487 info->value->type = g_strdup("unknown");
2488 info->value->locked = bdrv_dev_is_medium_locked(bs);
2489 info->value->removable = bdrv_dev_has_removable_media(bs);
2491 if (bdrv_dev_has_removable_media(bs)) {
2492 info->value->has_tray_open = true;
2493 info->value->tray_open = bdrv_dev_is_tray_open(bs);
2496 if (bdrv_iostatus_is_enabled(bs)) {
2497 info->value->has_io_status = true;
2498 info->value->io_status = bs->iostatus;
2502 info->value->has_inserted = true;
2503 info->value->inserted = g_malloc0(sizeof(*info->value->inserted));
2504 info->value->inserted->file = g_strdup(bs->filename);
2505 info->value->inserted->ro = bs->read_only;
2506 info->value->inserted->drv = g_strdup(bs->drv->format_name);
2507 info->value->inserted->encrypted = bs->encrypted;
2508 if (bs->backing_file[0]) {
2509 info->value->inserted->has_backing_file = true;
2510 info->value->inserted->backing_file = g_strdup(bs->backing_file);
2513 if (bs->io_limits_enabled) {
2514 info->value->inserted->bps =
2515 bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL];
2516 info->value->inserted->bps_rd =
2517 bs->io_limits.bps[BLOCK_IO_LIMIT_READ];
2518 info->value->inserted->bps_wr =
2519 bs->io_limits.bps[BLOCK_IO_LIMIT_WRITE];
2520 info->value->inserted->iops =
2521 bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL];
2522 info->value->inserted->iops_rd =
2523 bs->io_limits.iops[BLOCK_IO_LIMIT_READ];
2524 info->value->inserted->iops_wr =
2525 bs->io_limits.iops[BLOCK_IO_LIMIT_WRITE];
2529 /* XXX: waiting for the qapi to support GSList */
2531 head = cur_item = info;
2533 cur_item->next = info;
2541 /* Consider exposing this as a full fledged QMP command */
2542 static BlockStats *qmp_query_blockstat(const BlockDriverState *bs, Error **errp)
2546 s = g_malloc0(sizeof(*s));
2548 if (bs->device_name[0]) {
2549 s->has_device = true;
2550 s->device = g_strdup(bs->device_name);
2553 s->stats = g_malloc0(sizeof(*s->stats));
2554 s->stats->rd_bytes = bs->nr_bytes[BDRV_ACCT_READ];
2555 s->stats->wr_bytes = bs->nr_bytes[BDRV_ACCT_WRITE];
2556 s->stats->rd_operations = bs->nr_ops[BDRV_ACCT_READ];
2557 s->stats->wr_operations = bs->nr_ops[BDRV_ACCT_WRITE];
2558 s->stats->wr_highest_offset = bs->wr_highest_sector * BDRV_SECTOR_SIZE;
2559 s->stats->flush_operations = bs->nr_ops[BDRV_ACCT_FLUSH];
2560 s->stats->wr_total_time_ns = bs->total_time_ns[BDRV_ACCT_WRITE];
2561 s->stats->rd_total_time_ns = bs->total_time_ns[BDRV_ACCT_READ];
2562 s->stats->flush_total_time_ns = bs->total_time_ns[BDRV_ACCT_FLUSH];
2565 s->has_parent = true;
2566 s->parent = qmp_query_blockstat(bs->file, NULL);
2572 BlockStatsList *qmp_query_blockstats(Error **errp)
2574 BlockStatsList *head = NULL, *cur_item = NULL;
2575 BlockDriverState *bs;
2577 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2578 BlockStatsList *info = g_malloc0(sizeof(*info));
2579 info->value = qmp_query_blockstat(bs, NULL);
2581 /* XXX: waiting for the qapi to support GSList */
2583 head = cur_item = info;
2585 cur_item->next = info;
2593 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
2595 if (bs->backing_hd && bs->backing_hd->encrypted)
2596 return bs->backing_file;
2597 else if (bs->encrypted)
2598 return bs->filename;
2603 void bdrv_get_backing_filename(BlockDriverState *bs,
2604 char *filename, int filename_size)
2606 pstrcpy(filename, filename_size, bs->backing_file);
2609 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
2610 const uint8_t *buf, int nb_sectors)
2612 BlockDriver *drv = bs->drv;
2615 if (!drv->bdrv_write_compressed)
2617 if (bdrv_check_request(bs, sector_num, nb_sectors))
2620 if (bs->dirty_bitmap) {
2621 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
2624 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
2627 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2629 BlockDriver *drv = bs->drv;
2632 if (!drv->bdrv_get_info)
2634 memset(bdi, 0, sizeof(*bdi));
2635 return drv->bdrv_get_info(bs, bdi);
2638 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
2639 int64_t pos, int size)
2641 BlockDriver *drv = bs->drv;
2644 if (drv->bdrv_save_vmstate)
2645 return drv->bdrv_save_vmstate(bs, buf, pos, size);
2647 return bdrv_save_vmstate(bs->file, buf, pos, size);
2651 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
2652 int64_t pos, int size)
2654 BlockDriver *drv = bs->drv;
2657 if (drv->bdrv_load_vmstate)
2658 return drv->bdrv_load_vmstate(bs, buf, pos, size);
2660 return bdrv_load_vmstate(bs->file, buf, pos, size);
2664 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
2666 BlockDriver *drv = bs->drv;
2668 if (!drv || !drv->bdrv_debug_event) {
2672 return drv->bdrv_debug_event(bs, event);
2676 /**************************************************************/
2677 /* handling of snapshots */
2679 int bdrv_can_snapshot(BlockDriverState *bs)
2681 BlockDriver *drv = bs->drv;
2682 if (!drv || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
2686 if (!drv->bdrv_snapshot_create) {
2687 if (bs->file != NULL) {
2688 return bdrv_can_snapshot(bs->file);
2696 int bdrv_is_snapshot(BlockDriverState *bs)
2698 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
2701 BlockDriverState *bdrv_snapshots(void)
2703 BlockDriverState *bs;
2706 return bs_snapshots;
2710 while ((bs = bdrv_next(bs))) {
2711 if (bdrv_can_snapshot(bs)) {
2719 int bdrv_snapshot_create(BlockDriverState *bs,
2720 QEMUSnapshotInfo *sn_info)
2722 BlockDriver *drv = bs->drv;
2725 if (drv->bdrv_snapshot_create)
2726 return drv->bdrv_snapshot_create(bs, sn_info);
2728 return bdrv_snapshot_create(bs->file, sn_info);
2732 int bdrv_snapshot_goto(BlockDriverState *bs,
2733 const char *snapshot_id)
2735 BlockDriver *drv = bs->drv;
2740 if (drv->bdrv_snapshot_goto)
2741 return drv->bdrv_snapshot_goto(bs, snapshot_id);
2744 drv->bdrv_close(bs);
2745 ret = bdrv_snapshot_goto(bs->file, snapshot_id);
2746 open_ret = drv->bdrv_open(bs, bs->open_flags);
2748 bdrv_delete(bs->file);
2758 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2760 BlockDriver *drv = bs->drv;
2763 if (drv->bdrv_snapshot_delete)
2764 return drv->bdrv_snapshot_delete(bs, snapshot_id);
2766 return bdrv_snapshot_delete(bs->file, snapshot_id);
2770 int bdrv_snapshot_list(BlockDriverState *bs,
2771 QEMUSnapshotInfo **psn_info)
2773 BlockDriver *drv = bs->drv;
2776 if (drv->bdrv_snapshot_list)
2777 return drv->bdrv_snapshot_list(bs, psn_info);
2779 return bdrv_snapshot_list(bs->file, psn_info);
2783 int bdrv_snapshot_load_tmp(BlockDriverState *bs,
2784 const char *snapshot_name)
2786 BlockDriver *drv = bs->drv;
2790 if (!bs->read_only) {
2793 if (drv->bdrv_snapshot_load_tmp) {
2794 return drv->bdrv_snapshot_load_tmp(bs, snapshot_name);
2799 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
2800 const char *backing_file)
2806 if (bs->backing_hd) {
2807 if (strcmp(bs->backing_file, backing_file) == 0) {
2808 return bs->backing_hd;
2810 return bdrv_find_backing_image(bs->backing_hd, backing_file);
2817 #define NB_SUFFIXES 4
2819 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
2821 static const char suffixes[NB_SUFFIXES] = "KMGT";
2826 snprintf(buf, buf_size, "%" PRId64, size);
2829 for(i = 0; i < NB_SUFFIXES; i++) {
2830 if (size < (10 * base)) {
2831 snprintf(buf, buf_size, "%0.1f%c",
2832 (double)size / base,
2835 } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
2836 snprintf(buf, buf_size, "%" PRId64 "%c",
2837 ((size + (base >> 1)) / base),
2847 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
2849 char buf1[128], date_buf[128], clock_buf[128];
2859 snprintf(buf, buf_size,
2860 "%-10s%-20s%7s%20s%15s",
2861 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
2865 ptm = localtime(&ti);
2866 strftime(date_buf, sizeof(date_buf),
2867 "%Y-%m-%d %H:%M:%S", ptm);
2869 localtime_r(&ti, &tm);
2870 strftime(date_buf, sizeof(date_buf),
2871 "%Y-%m-%d %H:%M:%S", &tm);
2873 secs = sn->vm_clock_nsec / 1000000000;
2874 snprintf(clock_buf, sizeof(clock_buf),
2875 "%02d:%02d:%02d.%03d",
2877 (int)((secs / 60) % 60),
2879 (int)((sn->vm_clock_nsec / 1000000) % 1000));
2880 snprintf(buf, buf_size,
2881 "%-10s%-20s%7s%20s%15s",
2882 sn->id_str, sn->name,
2883 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
2890 /**************************************************************/
2893 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
2894 QEMUIOVector *qiov, int nb_sectors,
2895 BlockDriverCompletionFunc *cb, void *opaque)
2897 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
2899 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
2903 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
2904 QEMUIOVector *qiov, int nb_sectors,
2905 BlockDriverCompletionFunc *cb, void *opaque)
2907 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
2909 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
2914 typedef struct MultiwriteCB {
2919 BlockDriverCompletionFunc *cb;
2921 QEMUIOVector *free_qiov;
2925 static void multiwrite_user_cb(MultiwriteCB *mcb)
2929 for (i = 0; i < mcb->num_callbacks; i++) {
2930 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
2931 if (mcb->callbacks[i].free_qiov) {
2932 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
2934 g_free(mcb->callbacks[i].free_qiov);
2938 static void multiwrite_cb(void *opaque, int ret)
2940 MultiwriteCB *mcb = opaque;
2942 trace_multiwrite_cb(mcb, ret);
2944 if (ret < 0 && !mcb->error) {
2948 mcb->num_requests--;
2949 if (mcb->num_requests == 0) {
2950 multiwrite_user_cb(mcb);
2955 static int multiwrite_req_compare(const void *a, const void *b)
2957 const BlockRequest *req1 = a, *req2 = b;
2960 * Note that we can't simply subtract req2->sector from req1->sector
2961 * here as that could overflow the return value.
2963 if (req1->sector > req2->sector) {
2965 } else if (req1->sector < req2->sector) {
2973 * Takes a bunch of requests and tries to merge them. Returns the number of
2974 * requests that remain after merging.
2976 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
2977 int num_reqs, MultiwriteCB *mcb)
2981 // Sort requests by start sector
2982 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
2984 // Check if adjacent requests touch the same clusters. If so, combine them,
2985 // filling up gaps with zero sectors.
2987 for (i = 1; i < num_reqs; i++) {
2989 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
2991 // Handle exactly sequential writes and overlapping writes.
2992 if (reqs[i].sector <= oldreq_last) {
2996 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
3002 QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
3003 qemu_iovec_init(qiov,
3004 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
3006 // Add the first request to the merged one. If the requests are
3007 // overlapping, drop the last sectors of the first request.
3008 size = (reqs[i].sector - reqs[outidx].sector) << 9;
3009 qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
3011 // We should need to add any zeros between the two requests
3012 assert (reqs[i].sector <= oldreq_last);
3014 // Add the second request
3015 qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
3017 reqs[outidx].nb_sectors = qiov->size >> 9;
3018 reqs[outidx].qiov = qiov;
3020 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
3023 reqs[outidx].sector = reqs[i].sector;
3024 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
3025 reqs[outidx].qiov = reqs[i].qiov;
3033 * Submit multiple AIO write requests at once.
3035 * On success, the function returns 0 and all requests in the reqs array have
3036 * been submitted. In error case this function returns -1, and any of the
3037 * requests may or may not be submitted yet. In particular, this means that the
3038 * callback will be called for some of the requests, for others it won't. The
3039 * caller must check the error field of the BlockRequest to wait for the right
3040 * callbacks (if error != 0, no callback will be called).
3042 * The implementation may modify the contents of the reqs array, e.g. to merge
3043 * requests. However, the fields opaque and error are left unmodified as they
3044 * are used to signal failure for a single request to the caller.
3046 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
3051 /* don't submit writes if we don't have a medium */
3052 if (bs->drv == NULL) {
3053 for (i = 0; i < num_reqs; i++) {
3054 reqs[i].error = -ENOMEDIUM;
3059 if (num_reqs == 0) {
3063 // Create MultiwriteCB structure
3064 mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
3065 mcb->num_requests = 0;
3066 mcb->num_callbacks = num_reqs;
3068 for (i = 0; i < num_reqs; i++) {
3069 mcb->callbacks[i].cb = reqs[i].cb;
3070 mcb->callbacks[i].opaque = reqs[i].opaque;
3073 // Check for mergable requests
3074 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
3076 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
3078 /* Run the aio requests. */
3079 mcb->num_requests = num_reqs;
3080 for (i = 0; i < num_reqs; i++) {
3081 bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
3082 reqs[i].nb_sectors, multiwrite_cb, mcb);
3088 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
3090 acb->pool->cancel(acb);
3093 /* block I/O throttling */
3094 static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors,
3095 bool is_write, double elapsed_time, uint64_t *wait)
3097 uint64_t bps_limit = 0;
3098 double bytes_limit, bytes_base, bytes_res;
3099 double slice_time, wait_time;
3101 if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
3102 bps_limit = bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL];
3103 } else if (bs->io_limits.bps[is_write]) {
3104 bps_limit = bs->io_limits.bps[is_write];
3113 slice_time = bs->slice_end - bs->slice_start;
3114 slice_time /= (NANOSECONDS_PER_SECOND);
3115 bytes_limit = bps_limit * slice_time;
3116 bytes_base = bs->nr_bytes[is_write] - bs->io_base.bytes[is_write];
3117 if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
3118 bytes_base += bs->nr_bytes[!is_write] - bs->io_base.bytes[!is_write];
3121 /* bytes_base: the bytes of data which have been read/written; and
3122 * it is obtained from the history statistic info.
3123 * bytes_res: the remaining bytes of data which need to be read/written.
3124 * (bytes_base + bytes_res) / bps_limit: used to calcuate
3125 * the total time for completing reading/writting all data.
3127 bytes_res = (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
3129 if (bytes_base + bytes_res <= bytes_limit) {
3137 /* Calc approx time to dispatch */
3138 wait_time = (bytes_base + bytes_res) / bps_limit - elapsed_time;
3140 /* When the I/O rate at runtime exceeds the limits,
3141 * bs->slice_end need to be extended in order that the current statistic
3142 * info can be kept until the timer fire, so it is increased and tuned
3143 * based on the result of experiment.
3145 bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10;
3146 bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME;
3148 *wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
3154 static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write,
3155 double elapsed_time, uint64_t *wait)
3157 uint64_t iops_limit = 0;
3158 double ios_limit, ios_base;
3159 double slice_time, wait_time;
3161 if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
3162 iops_limit = bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL];
3163 } else if (bs->io_limits.iops[is_write]) {
3164 iops_limit = bs->io_limits.iops[is_write];
3173 slice_time = bs->slice_end - bs->slice_start;
3174 slice_time /= (NANOSECONDS_PER_SECOND);
3175 ios_limit = iops_limit * slice_time;
3176 ios_base = bs->nr_ops[is_write] - bs->io_base.ios[is_write];
3177 if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
3178 ios_base += bs->nr_ops[!is_write] - bs->io_base.ios[!is_write];
3181 if (ios_base + 1 <= ios_limit) {
3189 /* Calc approx time to dispatch */
3190 wait_time = (ios_base + 1) / iops_limit;
3191 if (wait_time > elapsed_time) {
3192 wait_time = wait_time - elapsed_time;
3197 bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10;
3198 bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME;
3200 *wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
3206 static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors,
3207 bool is_write, int64_t *wait)
3209 int64_t now, max_wait;
3210 uint64_t bps_wait = 0, iops_wait = 0;
3211 double elapsed_time;
3212 int bps_ret, iops_ret;
3214 now = qemu_get_clock_ns(vm_clock);
3215 if ((bs->slice_start < now)
3216 && (bs->slice_end > now)) {
3217 bs->slice_end = now + bs->slice_time;
3219 bs->slice_time = 5 * BLOCK_IO_SLICE_TIME;
3220 bs->slice_start = now;
3221 bs->slice_end = now + bs->slice_time;
3223 bs->io_base.bytes[is_write] = bs->nr_bytes[is_write];
3224 bs->io_base.bytes[!is_write] = bs->nr_bytes[!is_write];
3226 bs->io_base.ios[is_write] = bs->nr_ops[is_write];
3227 bs->io_base.ios[!is_write] = bs->nr_ops[!is_write];
3230 elapsed_time = now - bs->slice_start;
3231 elapsed_time /= (NANOSECONDS_PER_SECOND);
3233 bps_ret = bdrv_exceed_bps_limits(bs, nb_sectors,
3234 is_write, elapsed_time, &bps_wait);
3235 iops_ret = bdrv_exceed_iops_limits(bs, is_write,
3236 elapsed_time, &iops_wait);
3237 if (bps_ret || iops_ret) {
3238 max_wait = bps_wait > iops_wait ? bps_wait : iops_wait;
3243 now = qemu_get_clock_ns(vm_clock);
3244 if (bs->slice_end < now + max_wait) {
3245 bs->slice_end = now + max_wait;
3258 /**************************************************************/
3259 /* async block device emulation */
3261 typedef struct BlockDriverAIOCBSync {
3262 BlockDriverAIOCB common;
3265 /* vector translation state */
3269 } BlockDriverAIOCBSync;
3271 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
3273 BlockDriverAIOCBSync *acb =
3274 container_of(blockacb, BlockDriverAIOCBSync, common);
3275 qemu_bh_delete(acb->bh);
3277 qemu_aio_release(acb);
3280 static AIOPool bdrv_em_aio_pool = {
3281 .aiocb_size = sizeof(BlockDriverAIOCBSync),
3282 .cancel = bdrv_aio_cancel_em,
3285 static void bdrv_aio_bh_cb(void *opaque)
3287 BlockDriverAIOCBSync *acb = opaque;
3290 qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
3291 qemu_vfree(acb->bounce);
3292 acb->common.cb(acb->common.opaque, acb->ret);
3293 qemu_bh_delete(acb->bh);
3295 qemu_aio_release(acb);
3298 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
3302 BlockDriverCompletionFunc *cb,
3307 BlockDriverAIOCBSync *acb;
3309 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
3310 acb->is_write = is_write;
3312 acb->bounce = qemu_blockalign(bs, qiov->size);
3313 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
3316 qemu_iovec_to_buffer(acb->qiov, acb->bounce);
3317 acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
3319 acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
3322 qemu_bh_schedule(acb->bh);
3324 return &acb->common;
3327 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
3328 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
3329 BlockDriverCompletionFunc *cb, void *opaque)
3331 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
3334 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
3335 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
3336 BlockDriverCompletionFunc *cb, void *opaque)
3338 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
3342 typedef struct BlockDriverAIOCBCoroutine {
3343 BlockDriverAIOCB common;
3347 } BlockDriverAIOCBCoroutine;
3349 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb)
3354 static AIOPool bdrv_em_co_aio_pool = {
3355 .aiocb_size = sizeof(BlockDriverAIOCBCoroutine),
3356 .cancel = bdrv_aio_co_cancel_em,
3359 static void bdrv_co_em_bh(void *opaque)
3361 BlockDriverAIOCBCoroutine *acb = opaque;
3363 acb->common.cb(acb->common.opaque, acb->req.error);
3364 qemu_bh_delete(acb->bh);
3365 qemu_aio_release(acb);
3368 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
3369 static void coroutine_fn bdrv_co_do_rw(void *opaque)
3371 BlockDriverAIOCBCoroutine *acb = opaque;
3372 BlockDriverState *bs = acb->common.bs;
3374 if (!acb->is_write) {
3375 acb->req.error = bdrv_co_do_readv(bs, acb->req.sector,
3376 acb->req.nb_sectors, acb->req.qiov, 0);
3378 acb->req.error = bdrv_co_do_writev(bs, acb->req.sector,
3379 acb->req.nb_sectors, acb->req.qiov, 0);
3382 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3383 qemu_bh_schedule(acb->bh);
3386 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
3390 BlockDriverCompletionFunc *cb,
3395 BlockDriverAIOCBCoroutine *acb;
3397 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3398 acb->req.sector = sector_num;
3399 acb->req.nb_sectors = nb_sectors;
3400 acb->req.qiov = qiov;
3401 acb->is_write = is_write;
3403 co = qemu_coroutine_create(bdrv_co_do_rw);
3404 qemu_coroutine_enter(co, acb);
3406 return &acb->common;
3409 static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
3411 BlockDriverAIOCBCoroutine *acb = opaque;
3412 BlockDriverState *bs = acb->common.bs;
3414 acb->req.error = bdrv_co_flush(bs);
3415 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3416 qemu_bh_schedule(acb->bh);
3419 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
3420 BlockDriverCompletionFunc *cb, void *opaque)
3422 trace_bdrv_aio_flush(bs, opaque);
3425 BlockDriverAIOCBCoroutine *acb;
3427 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3428 co = qemu_coroutine_create(bdrv_aio_flush_co_entry);
3429 qemu_coroutine_enter(co, acb);
3431 return &acb->common;
3434 static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque)
3436 BlockDriverAIOCBCoroutine *acb = opaque;
3437 BlockDriverState *bs = acb->common.bs;
3439 acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors);
3440 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3441 qemu_bh_schedule(acb->bh);
3444 BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs,
3445 int64_t sector_num, int nb_sectors,
3446 BlockDriverCompletionFunc *cb, void *opaque)
3449 BlockDriverAIOCBCoroutine *acb;
3451 trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque);
3453 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3454 acb->req.sector = sector_num;
3455 acb->req.nb_sectors = nb_sectors;
3456 co = qemu_coroutine_create(bdrv_aio_discard_co_entry);
3457 qemu_coroutine_enter(co, acb);
3459 return &acb->common;
3462 void bdrv_init(void)
3464 module_call_init(MODULE_INIT_BLOCK);
3467 void bdrv_init_with_whitelist(void)
3469 use_bdrv_whitelist = 1;
3473 void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
3474 BlockDriverCompletionFunc *cb, void *opaque)
3476 BlockDriverAIOCB *acb;
3478 if (pool->free_aiocb) {
3479 acb = pool->free_aiocb;
3480 pool->free_aiocb = acb->next;
3482 acb = g_malloc0(pool->aiocb_size);
3487 acb->opaque = opaque;
3491 void qemu_aio_release(void *p)
3493 BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
3494 AIOPool *pool = acb->pool;
3495 acb->next = pool->free_aiocb;
3496 pool->free_aiocb = acb;
3499 /**************************************************************/
3500 /* Coroutine block device emulation */
3502 typedef struct CoroutineIOCompletion {
3503 Coroutine *coroutine;
3505 } CoroutineIOCompletion;
3507 static void bdrv_co_io_em_complete(void *opaque, int ret)
3509 CoroutineIOCompletion *co = opaque;
3512 qemu_coroutine_enter(co->coroutine, NULL);
3515 static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num,
3516 int nb_sectors, QEMUIOVector *iov,
3519 CoroutineIOCompletion co = {
3520 .coroutine = qemu_coroutine_self(),
3522 BlockDriverAIOCB *acb;
3525 acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors,
3526 bdrv_co_io_em_complete, &co);
3528 acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors,
3529 bdrv_co_io_em_complete, &co);
3532 trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb);
3536 qemu_coroutine_yield();
3541 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
3542 int64_t sector_num, int nb_sectors,
3545 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false);
3548 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
3549 int64_t sector_num, int nb_sectors,
3552 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true);
3555 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
3557 RwCo *rwco = opaque;
3559 rwco->ret = bdrv_co_flush(rwco->bs);
3562 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
3566 if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
3570 /* Write back cached data to the OS even with cache=unsafe */
3571 if (bs->drv->bdrv_co_flush_to_os) {
3572 ret = bs->drv->bdrv_co_flush_to_os(bs);
3578 /* But don't actually force it to the disk with cache=unsafe */
3579 if (bs->open_flags & BDRV_O_NO_FLUSH) {
3583 if (bs->drv->bdrv_co_flush_to_disk) {
3584 ret = bs->drv->bdrv_co_flush_to_disk(bs);
3585 } else if (bs->drv->bdrv_aio_flush) {
3586 BlockDriverAIOCB *acb;
3587 CoroutineIOCompletion co = {
3588 .coroutine = qemu_coroutine_self(),
3591 acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
3595 qemu_coroutine_yield();
3600 * Some block drivers always operate in either writethrough or unsafe
3601 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
3602 * know how the server works (because the behaviour is hardcoded or
3603 * depends on server-side configuration), so we can't ensure that
3604 * everything is safe on disk. Returning an error doesn't work because
3605 * that would break guests even if the server operates in writethrough
3608 * Let's hope the user knows what he's doing.
3616 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
3617 * in the case of cache=unsafe, so there are no useless flushes.
3619 return bdrv_co_flush(bs->file);
3622 void bdrv_invalidate_cache(BlockDriverState *bs)
3624 if (bs->drv && bs->drv->bdrv_invalidate_cache) {
3625 bs->drv->bdrv_invalidate_cache(bs);
3629 void bdrv_invalidate_cache_all(void)
3631 BlockDriverState *bs;
3633 QTAILQ_FOREACH(bs, &bdrv_states, list) {
3634 bdrv_invalidate_cache(bs);
3638 void bdrv_clear_incoming_migration_all(void)
3640 BlockDriverState *bs;
3642 QTAILQ_FOREACH(bs, &bdrv_states, list) {
3643 bs->open_flags = bs->open_flags & ~(BDRV_O_INCOMING);
3647 int bdrv_flush(BlockDriverState *bs)
3655 if (qemu_in_coroutine()) {
3656 /* Fast-path if already in coroutine context */
3657 bdrv_flush_co_entry(&rwco);
3659 co = qemu_coroutine_create(bdrv_flush_co_entry);
3660 qemu_coroutine_enter(co, &rwco);
3661 while (rwco.ret == NOT_DONE) {
3669 static void coroutine_fn bdrv_discard_co_entry(void *opaque)
3671 RwCo *rwco = opaque;
3673 rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors);
3676 int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
3681 } else if (bdrv_check_request(bs, sector_num, nb_sectors)) {
3683 } else if (bs->read_only) {
3685 } else if (bs->drv->bdrv_co_discard) {
3686 return bs->drv->bdrv_co_discard(bs, sector_num, nb_sectors);
3687 } else if (bs->drv->bdrv_aio_discard) {
3688 BlockDriverAIOCB *acb;
3689 CoroutineIOCompletion co = {
3690 .coroutine = qemu_coroutine_self(),
3693 acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors,
3694 bdrv_co_io_em_complete, &co);
3698 qemu_coroutine_yield();
3706 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
3711 .sector_num = sector_num,
3712 .nb_sectors = nb_sectors,
3716 if (qemu_in_coroutine()) {
3717 /* Fast-path if already in coroutine context */
3718 bdrv_discard_co_entry(&rwco);
3720 co = qemu_coroutine_create(bdrv_discard_co_entry);
3721 qemu_coroutine_enter(co, &rwco);
3722 while (rwco.ret == NOT_DONE) {
3730 /**************************************************************/
3731 /* removable device support */
3734 * Return TRUE if the media is present
3736 int bdrv_is_inserted(BlockDriverState *bs)
3738 BlockDriver *drv = bs->drv;
3742 if (!drv->bdrv_is_inserted)
3744 return drv->bdrv_is_inserted(bs);
3748 * Return whether the media changed since the last call to this
3749 * function, or -ENOTSUP if we don't know. Most drivers don't know.
3751 int bdrv_media_changed(BlockDriverState *bs)
3753 BlockDriver *drv = bs->drv;
3755 if (drv && drv->bdrv_media_changed) {
3756 return drv->bdrv_media_changed(bs);
3762 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
3764 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
3766 BlockDriver *drv = bs->drv;
3768 if (drv && drv->bdrv_eject) {
3769 drv->bdrv_eject(bs, eject_flag);
3772 if (bs->device_name[0] != '\0') {
3773 bdrv_emit_qmp_eject_event(bs, eject_flag);
3778 * Lock or unlock the media (if it is locked, the user won't be able
3779 * to eject it manually).
3781 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
3783 BlockDriver *drv = bs->drv;
3785 trace_bdrv_lock_medium(bs, locked);
3787 if (drv && drv->bdrv_lock_medium) {
3788 drv->bdrv_lock_medium(bs, locked);
3792 /* needed for generic scsi interface */
3794 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
3796 BlockDriver *drv = bs->drv;
3798 if (drv && drv->bdrv_ioctl)
3799 return drv->bdrv_ioctl(bs, req, buf);
3803 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
3804 unsigned long int req, void *buf,
3805 BlockDriverCompletionFunc *cb, void *opaque)
3807 BlockDriver *drv = bs->drv;
3809 if (drv && drv->bdrv_aio_ioctl)
3810 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
3814 void bdrv_set_buffer_alignment(BlockDriverState *bs, int align)
3816 bs->buffer_alignment = align;
3819 void *qemu_blockalign(BlockDriverState *bs, size_t size)
3821 return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
3824 void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
3826 int64_t bitmap_size;
3828 bs->dirty_count = 0;
3830 if (!bs->dirty_bitmap) {
3831 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
3832 BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
3833 bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
3835 bs->dirty_bitmap = g_malloc0(bitmap_size);
3838 if (bs->dirty_bitmap) {
3839 g_free(bs->dirty_bitmap);
3840 bs->dirty_bitmap = NULL;
3845 int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
3847 int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
3849 if (bs->dirty_bitmap &&
3850 (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
3851 return !!(bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
3852 (1UL << (chunk % (sizeof(unsigned long) * 8))));
3858 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
3861 set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
3864 int64_t bdrv_get_dirty_count(BlockDriverState *bs)
3866 return bs->dirty_count;
3869 void bdrv_set_in_use(BlockDriverState *bs, int in_use)
3871 assert(bs->in_use != in_use);
3872 bs->in_use = in_use;
3875 int bdrv_in_use(BlockDriverState *bs)
3880 void bdrv_iostatus_enable(BlockDriverState *bs)
3882 bs->iostatus_enabled = true;
3883 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
3886 /* The I/O status is only enabled if the drive explicitly
3887 * enables it _and_ the VM is configured to stop on errors */
3888 bool bdrv_iostatus_is_enabled(const BlockDriverState *bs)
3890 return (bs->iostatus_enabled &&
3891 (bs->on_write_error == BLOCK_ERR_STOP_ENOSPC ||
3892 bs->on_write_error == BLOCK_ERR_STOP_ANY ||
3893 bs->on_read_error == BLOCK_ERR_STOP_ANY));
3896 void bdrv_iostatus_disable(BlockDriverState *bs)
3898 bs->iostatus_enabled = false;
3901 void bdrv_iostatus_reset(BlockDriverState *bs)
3903 if (bdrv_iostatus_is_enabled(bs)) {
3904 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
3908 /* XXX: Today this is set by device models because it makes the implementation
3909 quite simple. However, the block layer knows about the error, so it's
3910 possible to implement this without device models being involved */
3911 void bdrv_iostatus_set_err(BlockDriverState *bs, int error)
3913 if (bdrv_iostatus_is_enabled(bs) &&
3914 bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
3916 bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
3917 BLOCK_DEVICE_IO_STATUS_FAILED;
3922 bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes,
3923 enum BlockAcctType type)
3925 assert(type < BDRV_MAX_IOTYPE);
3927 cookie->bytes = bytes;
3928 cookie->start_time_ns = get_clock();
3929 cookie->type = type;
3933 bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie)
3935 assert(cookie->type < BDRV_MAX_IOTYPE);
3937 bs->nr_bytes[cookie->type] += cookie->bytes;
3938 bs->nr_ops[cookie->type]++;
3939 bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns;
3942 int bdrv_img_create(const char *filename, const char *fmt,
3943 const char *base_filename, const char *base_fmt,
3944 char *options, uint64_t img_size, int flags)
3946 QEMUOptionParameter *param = NULL, *create_options = NULL;
3947 QEMUOptionParameter *backing_fmt, *backing_file, *size;
3948 BlockDriverState *bs = NULL;
3949 BlockDriver *drv, *proto_drv;
3950 BlockDriver *backing_drv = NULL;
3953 /* Find driver and parse its options */
3954 drv = bdrv_find_format(fmt);
3956 error_report("Unknown file format '%s'", fmt);
3961 proto_drv = bdrv_find_protocol(filename);
3963 error_report("Unknown protocol '%s'", filename);
3968 create_options = append_option_parameters(create_options,
3969 drv->create_options);
3970 create_options = append_option_parameters(create_options,
3971 proto_drv->create_options);
3973 /* Create parameter list with default values */
3974 param = parse_option_parameters("", create_options, param);
3976 set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size);
3978 /* Parse -o options */
3980 param = parse_option_parameters(options, create_options, param);
3981 if (param == NULL) {
3982 error_report("Invalid options for file format '%s'.", fmt);
3988 if (base_filename) {
3989 if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE,
3991 error_report("Backing file not supported for file format '%s'",
3999 if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) {
4000 error_report("Backing file format not supported for file "
4001 "format '%s'", fmt);
4007 backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE);
4008 if (backing_file && backing_file->value.s) {
4009 if (!strcmp(filename, backing_file->value.s)) {
4010 error_report("Error: Trying to create an image with the "
4011 "same filename as the backing file");
4017 backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT);
4018 if (backing_fmt && backing_fmt->value.s) {
4019 backing_drv = bdrv_find_format(backing_fmt->value.s);
4021 error_report("Unknown backing file format '%s'",
4022 backing_fmt->value.s);
4028 // The size for the image must always be specified, with one exception:
4029 // If we are using a backing file, we can obtain the size from there
4030 size = get_option_parameter(param, BLOCK_OPT_SIZE);
4031 if (size && size->value.n == -1) {
4032 if (backing_file && backing_file->value.s) {
4038 ret = bdrv_open(bs, backing_file->value.s, flags, backing_drv);
4040 error_report("Could not open '%s'", backing_file->value.s);
4043 bdrv_get_geometry(bs, &size);
4046 snprintf(buf, sizeof(buf), "%" PRId64, size);
4047 set_option_parameter(param, BLOCK_OPT_SIZE, buf);
4049 error_report("Image creation needs a size parameter");
4055 printf("Formatting '%s', fmt=%s ", filename, fmt);
4056 print_option_parameters(param);
4059 ret = bdrv_create(drv, filename, param);
4062 if (ret == -ENOTSUP) {
4063 error_report("Formatting or formatting option not supported for "
4064 "file format '%s'", fmt);
4065 } else if (ret == -EFBIG) {
4066 error_report("The image size is too large for file format '%s'",
4069 error_report("%s: error while creating %s: %s", filename, fmt,
4075 free_option_parameters(create_options);
4076 free_option_parameters(param);
4085 void *block_job_create(const BlockJobType *job_type, BlockDriverState *bs,
4086 BlockDriverCompletionFunc *cb, void *opaque)
4090 if (bs->job || bdrv_in_use(bs)) {
4093 bdrv_set_in_use(bs, 1);
4095 job = g_malloc0(job_type->instance_size);
4096 job->job_type = job_type;
4099 job->opaque = opaque;
4104 void block_job_complete(BlockJob *job, int ret)
4106 BlockDriverState *bs = job->bs;
4108 assert(bs->job == job);
4109 job->cb(job->opaque, ret);
4112 bdrv_set_in_use(bs, 0);
4115 int block_job_set_speed(BlockJob *job, int64_t value)
4119 if (!job->job_type->set_speed) {
4122 rc = job->job_type->set_speed(job, value);
4129 void block_job_cancel(BlockJob *job)
4131 job->cancelled = true;
4134 bool block_job_is_cancelled(BlockJob *job)
4136 return job->cancelled;
4139 void block_job_cancel_sync(BlockJob *job)
4141 BlockDriverState *bs = job->bs;
4143 assert(bs->job == job);
4144 block_job_cancel(job);
4145 while (bs->job != NULL && bs->job->busy) {