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);
84 static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors,
85 bool is_write, double elapsed_time, uint64_t *wait);
86 static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write,
87 double elapsed_time, uint64_t *wait);
88 static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors,
89 bool is_write, int64_t *wait);
91 static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
92 QTAILQ_HEAD_INITIALIZER(bdrv_states);
94 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
95 QLIST_HEAD_INITIALIZER(bdrv_drivers);
97 /* The device to use for VM snapshots */
98 static BlockDriverState *bs_snapshots;
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_io_limits_disable(BlockDriverState *bs)
126 bs->io_limits_enabled = false;
128 while (qemu_co_queue_next(&bs->throttled_reqs));
130 if (bs->block_timer) {
131 qemu_del_timer(bs->block_timer);
132 qemu_free_timer(bs->block_timer);
133 bs->block_timer = NULL;
139 memset(&bs->io_base, 0, sizeof(bs->io_base));
142 static void bdrv_block_timer(void *opaque)
144 BlockDriverState *bs = opaque;
146 qemu_co_queue_next(&bs->throttled_reqs);
149 void bdrv_io_limits_enable(BlockDriverState *bs)
151 qemu_co_queue_init(&bs->throttled_reqs);
152 bs->block_timer = qemu_new_timer_ns(vm_clock, bdrv_block_timer, bs);
153 bs->slice_time = 5 * BLOCK_IO_SLICE_TIME;
154 bs->slice_start = qemu_get_clock_ns(vm_clock);
155 bs->slice_end = bs->slice_start + bs->slice_time;
156 memset(&bs->io_base, 0, sizeof(bs->io_base));
157 bs->io_limits_enabled = true;
160 bool bdrv_io_limits_enabled(BlockDriverState *bs)
162 BlockIOLimit *io_limits = &bs->io_limits;
163 return io_limits->bps[BLOCK_IO_LIMIT_READ]
164 || io_limits->bps[BLOCK_IO_LIMIT_WRITE]
165 || io_limits->bps[BLOCK_IO_LIMIT_TOTAL]
166 || io_limits->iops[BLOCK_IO_LIMIT_READ]
167 || io_limits->iops[BLOCK_IO_LIMIT_WRITE]
168 || io_limits->iops[BLOCK_IO_LIMIT_TOTAL];
171 static void bdrv_io_limits_intercept(BlockDriverState *bs,
172 bool is_write, int nb_sectors)
174 int64_t wait_time = -1;
176 if (!qemu_co_queue_empty(&bs->throttled_reqs)) {
177 qemu_co_queue_wait(&bs->throttled_reqs);
180 /* In fact, we hope to keep each request's timing, in FIFO mode. The next
181 * throttled requests will not be dequeued until the current request is
182 * allowed to be serviced. So if the current request still exceeds the
183 * limits, it will be inserted to the head. All requests followed it will
184 * be still in throttled_reqs queue.
187 while (bdrv_exceed_io_limits(bs, nb_sectors, is_write, &wait_time)) {
188 qemu_mod_timer(bs->block_timer,
189 wait_time + qemu_get_clock_ns(vm_clock));
190 qemu_co_queue_wait_insert_head(&bs->throttled_reqs);
193 qemu_co_queue_next(&bs->throttled_reqs);
196 /* check if the path starts with "<protocol>:" */
197 static int path_has_protocol(const char *path)
200 if (is_windows_drive(path) ||
201 is_windows_drive_prefix(path)) {
206 return strchr(path, ':') != NULL;
209 int path_is_absolute(const char *path)
213 /* specific case for names like: "\\.\d:" */
214 if (*path == '/' || *path == '\\')
217 p = strchr(path, ':');
223 return (*p == '/' || *p == '\\');
229 /* if filename is absolute, just copy it to dest. Otherwise, build a
230 path to it by considering it is relative to base_path. URL are
232 void path_combine(char *dest, int dest_size,
233 const char *base_path,
234 const char *filename)
241 if (path_is_absolute(filename)) {
242 pstrcpy(dest, dest_size, filename);
244 p = strchr(base_path, ':');
249 p1 = strrchr(base_path, '/');
253 p2 = strrchr(base_path, '\\');
265 if (len > dest_size - 1)
267 memcpy(dest, base_path, len);
269 pstrcat(dest, dest_size, filename);
273 void bdrv_register(BlockDriver *bdrv)
275 /* Block drivers without coroutine functions need emulation */
276 if (!bdrv->bdrv_co_readv) {
277 bdrv->bdrv_co_readv = bdrv_co_readv_em;
278 bdrv->bdrv_co_writev = bdrv_co_writev_em;
280 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
281 * the block driver lacks aio we need to emulate that too.
283 if (!bdrv->bdrv_aio_readv) {
284 /* add AIO emulation layer */
285 bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
286 bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
290 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
293 /* create a new block device (by default it is empty) */
294 BlockDriverState *bdrv_new(const char *device_name)
296 BlockDriverState *bs;
298 bs = g_malloc0(sizeof(BlockDriverState));
299 pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
300 if (device_name[0] != '\0') {
301 QTAILQ_INSERT_TAIL(&bdrv_states, bs, list);
303 bdrv_iostatus_disable(bs);
307 BlockDriver *bdrv_find_format(const char *format_name)
310 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
311 if (!strcmp(drv1->format_name, format_name)) {
318 static int bdrv_is_whitelisted(BlockDriver *drv)
320 static const char *whitelist[] = {
321 CONFIG_BDRV_WHITELIST
326 return 1; /* no whitelist, anything goes */
328 for (p = whitelist; *p; p++) {
329 if (!strcmp(drv->format_name, *p)) {
336 BlockDriver *bdrv_find_whitelisted_format(const char *format_name)
338 BlockDriver *drv = bdrv_find_format(format_name);
339 return drv && bdrv_is_whitelisted(drv) ? drv : NULL;
342 int bdrv_create(BlockDriver *drv, const char* filename,
343 QEMUOptionParameter *options)
345 if (!drv->bdrv_create)
348 return drv->bdrv_create(filename, options);
351 int bdrv_create_file(const char* filename, QEMUOptionParameter *options)
355 drv = bdrv_find_protocol(filename);
360 return bdrv_create(drv, filename, options);
364 void get_tmp_filename(char *filename, int size)
366 char temp_dir[MAX_PATH];
368 GetTempPath(MAX_PATH, temp_dir);
369 GetTempFileName(temp_dir, "qem", 0, filename);
372 void get_tmp_filename(char *filename, int size)
376 /* XXX: race condition possible */
377 tmpdir = getenv("TMPDIR");
380 snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
381 fd = mkstemp(filename);
387 * Detect host devices. By convention, /dev/cdrom[N] is always
388 * recognized as a host CDROM.
390 static BlockDriver *find_hdev_driver(const char *filename)
392 int score_max = 0, score;
393 BlockDriver *drv = NULL, *d;
395 QLIST_FOREACH(d, &bdrv_drivers, list) {
396 if (d->bdrv_probe_device) {
397 score = d->bdrv_probe_device(filename);
398 if (score > score_max) {
408 BlockDriver *bdrv_find_protocol(const char *filename)
415 /* TODO Drivers without bdrv_file_open must be specified explicitly */
418 * XXX(hch): we really should not let host device detection
419 * override an explicit protocol specification, but moving this
420 * later breaks access to device names with colons in them.
421 * Thanks to the brain-dead persistent naming schemes on udev-
422 * based Linux systems those actually are quite common.
424 drv1 = find_hdev_driver(filename);
429 if (!path_has_protocol(filename)) {
430 return bdrv_find_format("file");
432 p = strchr(filename, ':');
435 if (len > sizeof(protocol) - 1)
436 len = sizeof(protocol) - 1;
437 memcpy(protocol, filename, len);
438 protocol[len] = '\0';
439 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
440 if (drv1->protocol_name &&
441 !strcmp(drv1->protocol_name, protocol)) {
448 static int find_image_format(const char *filename, BlockDriver **pdrv)
450 int ret, score, score_max;
451 BlockDriver *drv1, *drv;
453 BlockDriverState *bs;
455 ret = bdrv_file_open(&bs, filename, 0);
461 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
462 if (bs->sg || !bdrv_is_inserted(bs)) {
464 drv = bdrv_find_format("raw");
472 ret = bdrv_pread(bs, 0, buf, sizeof(buf));
481 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
482 if (drv1->bdrv_probe) {
483 score = drv1->bdrv_probe(buf, ret, filename);
484 if (score > score_max) {
498 * Set the current 'total_sectors' value
500 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
502 BlockDriver *drv = bs->drv;
504 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
508 /* query actual device if possible, otherwise just trust the hint */
509 if (drv->bdrv_getlength) {
510 int64_t length = drv->bdrv_getlength(bs);
514 hint = length >> BDRV_SECTOR_BITS;
517 bs->total_sectors = hint;
522 * Set open flags for a given cache mode
524 * Return 0 on success, -1 if the cache mode was invalid.
526 int bdrv_parse_cache_flags(const char *mode, int *flags)
528 *flags &= ~BDRV_O_CACHE_MASK;
530 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
531 *flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB;
532 } else if (!strcmp(mode, "directsync")) {
533 *flags |= BDRV_O_NOCACHE;
534 } else if (!strcmp(mode, "writeback")) {
535 *flags |= BDRV_O_CACHE_WB;
536 } else if (!strcmp(mode, "unsafe")) {
537 *flags |= BDRV_O_CACHE_WB;
538 *flags |= BDRV_O_NO_FLUSH;
539 } else if (!strcmp(mode, "writethrough")) {
540 /* this is the default */
549 * The copy-on-read flag is actually a reference count so multiple users may
550 * use the feature without worrying about clobbering its previous state.
551 * Copy-on-read stays enabled until all users have called to disable it.
553 void bdrv_enable_copy_on_read(BlockDriverState *bs)
558 void bdrv_disable_copy_on_read(BlockDriverState *bs)
560 assert(bs->copy_on_read > 0);
565 * Common part for opening disk images and files
567 static int bdrv_open_common(BlockDriverState *bs, const char *filename,
568 int flags, BlockDriver *drv)
574 trace_bdrv_open_common(bs, filename, flags, drv->format_name);
577 bs->total_sectors = 0;
581 bs->open_flags = flags;
583 bs->buffer_alignment = 512;
585 assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
586 if ((flags & BDRV_O_RDWR) && (flags & BDRV_O_COPY_ON_READ)) {
587 bdrv_enable_copy_on_read(bs);
590 pstrcpy(bs->filename, sizeof(bs->filename), filename);
591 bs->backing_file[0] = '\0';
593 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv)) {
598 bs->opaque = g_malloc0(drv->instance_size);
600 bs->enable_write_cache = !!(flags & BDRV_O_CACHE_WB);
603 * Clear flags that are internal to the block layer before opening the
606 open_flags = flags & ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
609 * Snapshots should be writable.
611 if (bs->is_temporary) {
612 open_flags |= BDRV_O_RDWR;
615 bs->keep_read_only = bs->read_only = !(open_flags & BDRV_O_RDWR);
617 /* Open the image, either directly or using a protocol */
618 if (drv->bdrv_file_open) {
619 ret = drv->bdrv_file_open(bs, filename, open_flags);
621 ret = bdrv_file_open(&bs->file, filename, open_flags);
623 ret = drv->bdrv_open(bs, open_flags);
631 ret = refresh_total_sectors(bs, bs->total_sectors);
637 if (bs->is_temporary) {
645 bdrv_delete(bs->file);
655 * Opens a file using a protocol (file, host_device, nbd, ...)
657 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
659 BlockDriverState *bs;
663 drv = bdrv_find_protocol(filename);
669 ret = bdrv_open_common(bs, filename, flags, drv);
680 * Opens a disk image (raw, qcow2, vmdk, ...)
682 int bdrv_open(BlockDriverState *bs, const char *filename, int flags,
686 char tmp_filename[PATH_MAX];
688 if (flags & BDRV_O_SNAPSHOT) {
689 BlockDriverState *bs1;
692 BlockDriver *bdrv_qcow2;
693 QEMUOptionParameter *options;
694 char backing_filename[PATH_MAX];
696 /* if snapshot, we create a temporary backing file and open it
697 instead of opening 'filename' directly */
699 /* if there is a backing file, use it */
701 ret = bdrv_open(bs1, filename, 0, drv);
706 total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
708 if (bs1->drv && bs1->drv->protocol_name)
713 get_tmp_filename(tmp_filename, sizeof(tmp_filename));
715 /* Real path is meaningless for protocols */
717 snprintf(backing_filename, sizeof(backing_filename),
719 else if (!realpath(filename, backing_filename))
722 bdrv_qcow2 = bdrv_find_format("qcow2");
723 options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
725 set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size);
726 set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
728 set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
732 ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
733 free_option_parameters(options);
738 filename = tmp_filename;
740 bs->is_temporary = 1;
743 /* Find the right image format driver */
745 ret = find_image_format(filename, &drv);
749 goto unlink_and_fail;
753 ret = bdrv_open_common(bs, filename, flags, drv);
755 goto unlink_and_fail;
758 /* If there is a backing file, use it */
759 if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') {
760 char backing_filename[PATH_MAX];
762 BlockDriver *back_drv = NULL;
764 bs->backing_hd = bdrv_new("");
766 if (path_has_protocol(bs->backing_file)) {
767 pstrcpy(backing_filename, sizeof(backing_filename),
770 path_combine(backing_filename, sizeof(backing_filename),
771 filename, bs->backing_file);
774 if (bs->backing_format[0] != '\0') {
775 back_drv = bdrv_find_format(bs->backing_format);
778 /* backing files always opened read-only */
780 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
782 ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv);
787 if (bs->is_temporary) {
788 bs->backing_hd->keep_read_only = !(flags & BDRV_O_RDWR);
790 /* base image inherits from "parent" */
791 bs->backing_hd->keep_read_only = bs->keep_read_only;
795 if (!bdrv_key_required(bs)) {
796 bdrv_dev_change_media_cb(bs, true);
799 /* throttling disk I/O limits */
800 if (bs->io_limits_enabled) {
801 bdrv_io_limits_enable(bs);
807 if (bs->is_temporary) {
813 void bdrv_close(BlockDriverState *bs)
818 block_job_cancel_sync(bs->job);
822 if (bs == bs_snapshots) {
825 if (bs->backing_hd) {
826 bdrv_delete(bs->backing_hd);
827 bs->backing_hd = NULL;
829 bs->drv->bdrv_close(bs);
832 if (bs->is_temporary) {
833 unlink(bs->filename);
838 bs->copy_on_read = 0;
840 if (bs->file != NULL) {
841 bdrv_close(bs->file);
844 bdrv_dev_change_media_cb(bs, false);
847 /*throttling disk I/O limits*/
848 if (bs->io_limits_enabled) {
849 bdrv_io_limits_disable(bs);
853 void bdrv_close_all(void)
855 BlockDriverState *bs;
857 QTAILQ_FOREACH(bs, &bdrv_states, list) {
863 * Wait for pending requests to complete across all BlockDriverStates
865 * This function does not flush data to disk, use bdrv_flush_all() for that
866 * after calling this function.
868 void bdrv_drain_all(void)
870 BlockDriverState *bs;
874 /* If requests are still pending there is a bug somewhere */
875 QTAILQ_FOREACH(bs, &bdrv_states, list) {
876 assert(QLIST_EMPTY(&bs->tracked_requests));
877 assert(qemu_co_queue_empty(&bs->throttled_reqs));
881 /* make a BlockDriverState anonymous by removing from bdrv_state list.
882 Also, NULL terminate the device_name to prevent double remove */
883 void bdrv_make_anon(BlockDriverState *bs)
885 if (bs->device_name[0] != '\0') {
886 QTAILQ_REMOVE(&bdrv_states, bs, list);
888 bs->device_name[0] = '\0';
892 * Add new bs contents at the top of an image chain while the chain is
893 * live, while keeping required fields on the top layer.
895 * This will modify the BlockDriverState fields, and swap contents
896 * between bs_new and bs_top. Both bs_new and bs_top are modified.
898 * bs_new is required to be anonymous.
900 * This function does not create any image files.
902 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
904 BlockDriverState tmp;
906 /* bs_new must be anonymous */
907 assert(bs_new->device_name[0] == '\0');
911 /* there are some fields that need to stay on the top layer: */
914 tmp.dev_ops = bs_top->dev_ops;
915 tmp.dev_opaque = bs_top->dev_opaque;
916 tmp.dev = bs_top->dev;
917 tmp.buffer_alignment = bs_top->buffer_alignment;
918 tmp.copy_on_read = bs_top->copy_on_read;
920 /* i/o timing parameters */
921 tmp.slice_time = bs_top->slice_time;
922 tmp.slice_start = bs_top->slice_start;
923 tmp.slice_end = bs_top->slice_end;
924 tmp.io_limits = bs_top->io_limits;
925 tmp.io_base = bs_top->io_base;
926 tmp.throttled_reqs = bs_top->throttled_reqs;
927 tmp.block_timer = bs_top->block_timer;
928 tmp.io_limits_enabled = bs_top->io_limits_enabled;
931 tmp.cyls = bs_top->cyls;
932 tmp.heads = bs_top->heads;
933 tmp.secs = bs_top->secs;
934 tmp.translation = bs_top->translation;
937 tmp.on_read_error = bs_top->on_read_error;
938 tmp.on_write_error = bs_top->on_write_error;
941 tmp.iostatus_enabled = bs_top->iostatus_enabled;
942 tmp.iostatus = bs_top->iostatus;
944 /* keep the same entry in bdrv_states */
945 pstrcpy(tmp.device_name, sizeof(tmp.device_name), bs_top->device_name);
946 tmp.list = bs_top->list;
948 /* The contents of 'tmp' will become bs_top, as we are
949 * swapping bs_new and bs_top contents. */
950 tmp.backing_hd = bs_new;
951 pstrcpy(tmp.backing_file, sizeof(tmp.backing_file), bs_top->filename);
952 bdrv_get_format(bs_top, tmp.backing_format, sizeof(tmp.backing_format));
954 /* swap contents of the fixed new bs and the current top */
958 /* device_name[] was carried over from the old bs_top. bs_new
959 * shouldn't be in bdrv_states, so we need to make device_name[]
960 * reflect the anonymity of bs_new
962 bs_new->device_name[0] = '\0';
964 /* clear the copied fields in the new backing file */
965 bdrv_detach_dev(bs_new, bs_new->dev);
967 qemu_co_queue_init(&bs_new->throttled_reqs);
968 memset(&bs_new->io_base, 0, sizeof(bs_new->io_base));
969 memset(&bs_new->io_limits, 0, sizeof(bs_new->io_limits));
970 bdrv_iostatus_disable(bs_new);
972 /* we don't use bdrv_io_limits_disable() for this, because we don't want
973 * to affect or delete the block_timer, as it has been moved to bs_top */
974 bs_new->io_limits_enabled = false;
975 bs_new->block_timer = NULL;
976 bs_new->slice_time = 0;
977 bs_new->slice_start = 0;
978 bs_new->slice_end = 0;
981 void bdrv_delete(BlockDriverState *bs)
987 /* remove from list, if necessary */
991 if (bs->file != NULL) {
992 bdrv_delete(bs->file);
995 assert(bs != bs_snapshots);
999 int bdrv_attach_dev(BlockDriverState *bs, void *dev)
1000 /* TODO change to DeviceState *dev when all users are qdevified */
1006 bdrv_iostatus_reset(bs);
1010 /* TODO qdevified devices don't use this, remove when devices are qdevified */
1011 void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev)
1013 if (bdrv_attach_dev(bs, dev) < 0) {
1018 void bdrv_detach_dev(BlockDriverState *bs, void *dev)
1019 /* TODO change to DeviceState *dev when all users are qdevified */
1021 assert(bs->dev == dev);
1024 bs->dev_opaque = NULL;
1025 bs->buffer_alignment = 512;
1028 /* TODO change to return DeviceState * when all users are qdevified */
1029 void *bdrv_get_attached_dev(BlockDriverState *bs)
1034 void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops,
1038 bs->dev_opaque = opaque;
1039 if (bdrv_dev_has_removable_media(bs) && bs == bs_snapshots) {
1040 bs_snapshots = NULL;
1044 void bdrv_emit_qmp_error_event(const BlockDriverState *bdrv,
1045 BlockQMPEventAction action, int is_read)
1048 const char *action_str;
1051 case BDRV_ACTION_REPORT:
1052 action_str = "report";
1054 case BDRV_ACTION_IGNORE:
1055 action_str = "ignore";
1057 case BDRV_ACTION_STOP:
1058 action_str = "stop";
1064 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1067 is_read ? "read" : "write");
1068 monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
1070 qobject_decref(data);
1073 static void bdrv_emit_qmp_eject_event(BlockDriverState *bs, bool ejected)
1077 data = qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }",
1078 bdrv_get_device_name(bs), ejected);
1079 monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED, data);
1081 qobject_decref(data);
1084 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load)
1086 if (bs->dev_ops && bs->dev_ops->change_media_cb) {
1087 bool tray_was_closed = !bdrv_dev_is_tray_open(bs);
1088 bs->dev_ops->change_media_cb(bs->dev_opaque, load);
1089 if (tray_was_closed) {
1091 bdrv_emit_qmp_eject_event(bs, true);
1095 bdrv_emit_qmp_eject_event(bs, false);
1100 bool bdrv_dev_has_removable_media(BlockDriverState *bs)
1102 return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb);
1105 void bdrv_dev_eject_request(BlockDriverState *bs, bool force)
1107 if (bs->dev_ops && bs->dev_ops->eject_request_cb) {
1108 bs->dev_ops->eject_request_cb(bs->dev_opaque, force);
1112 bool bdrv_dev_is_tray_open(BlockDriverState *bs)
1114 if (bs->dev_ops && bs->dev_ops->is_tray_open) {
1115 return bs->dev_ops->is_tray_open(bs->dev_opaque);
1120 static void bdrv_dev_resize_cb(BlockDriverState *bs)
1122 if (bs->dev_ops && bs->dev_ops->resize_cb) {
1123 bs->dev_ops->resize_cb(bs->dev_opaque);
1127 bool bdrv_dev_is_medium_locked(BlockDriverState *bs)
1129 if (bs->dev_ops && bs->dev_ops->is_medium_locked) {
1130 return bs->dev_ops->is_medium_locked(bs->dev_opaque);
1136 * Run consistency checks on an image
1138 * Returns 0 if the check could be completed (it doesn't mean that the image is
1139 * free of errors) or -errno when an internal error occurred. The results of the
1140 * check are stored in res.
1142 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res)
1144 if (bs->drv->bdrv_check == NULL) {
1148 memset(res, 0, sizeof(*res));
1149 return bs->drv->bdrv_check(bs, res);
1152 #define COMMIT_BUF_SECTORS 2048
1154 /* commit COW file into the raw image */
1155 int bdrv_commit(BlockDriverState *bs)
1157 BlockDriver *drv = bs->drv;
1158 BlockDriver *backing_drv;
1159 int64_t sector, total_sectors;
1160 int n, ro, open_flags;
1161 int ret = 0, rw_ret = 0;
1163 char filename[1024];
1164 BlockDriverState *bs_rw, *bs_ro;
1169 if (!bs->backing_hd) {
1173 if (bs->backing_hd->keep_read_only) {
1177 if (bdrv_in_use(bs) || bdrv_in_use(bs->backing_hd)) {
1181 backing_drv = bs->backing_hd->drv;
1182 ro = bs->backing_hd->read_only;
1183 strncpy(filename, bs->backing_hd->filename, sizeof(filename));
1184 open_flags = bs->backing_hd->open_flags;
1188 bdrv_delete(bs->backing_hd);
1189 bs->backing_hd = NULL;
1190 bs_rw = bdrv_new("");
1191 rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR,
1195 /* try to re-open read-only */
1196 bs_ro = bdrv_new("");
1197 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
1201 /* drive not functional anymore */
1205 bs->backing_hd = bs_ro;
1208 bs->backing_hd = bs_rw;
1211 total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
1212 buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
1214 for (sector = 0; sector < total_sectors; sector += n) {
1215 if (bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n)) {
1217 if (bdrv_read(bs, sector, buf, n) != 0) {
1222 if (bdrv_write(bs->backing_hd, sector, buf, n) != 0) {
1229 if (drv->bdrv_make_empty) {
1230 ret = drv->bdrv_make_empty(bs);
1235 * Make sure all data we wrote to the backing device is actually
1239 bdrv_flush(bs->backing_hd);
1246 bdrv_delete(bs->backing_hd);
1247 bs->backing_hd = NULL;
1248 bs_ro = bdrv_new("");
1249 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
1253 /* drive not functional anymore */
1257 bs->backing_hd = bs_ro;
1258 bs->backing_hd->keep_read_only = 0;
1264 int bdrv_commit_all(void)
1266 BlockDriverState *bs;
1268 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1269 int ret = bdrv_commit(bs);
1277 struct BdrvTrackedRequest {
1278 BlockDriverState *bs;
1282 QLIST_ENTRY(BdrvTrackedRequest) list;
1283 Coroutine *co; /* owner, used for deadlock detection */
1284 CoQueue wait_queue; /* coroutines blocked on this request */
1288 * Remove an active request from the tracked requests list
1290 * This function should be called when a tracked request is completing.
1292 static void tracked_request_end(BdrvTrackedRequest *req)
1294 QLIST_REMOVE(req, list);
1295 qemu_co_queue_restart_all(&req->wait_queue);
1299 * Add an active request to the tracked requests list
1301 static void tracked_request_begin(BdrvTrackedRequest *req,
1302 BlockDriverState *bs,
1304 int nb_sectors, bool is_write)
1306 *req = (BdrvTrackedRequest){
1308 .sector_num = sector_num,
1309 .nb_sectors = nb_sectors,
1310 .is_write = is_write,
1311 .co = qemu_coroutine_self(),
1314 qemu_co_queue_init(&req->wait_queue);
1316 QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
1320 * Round a region to cluster boundaries
1322 static void round_to_clusters(BlockDriverState *bs,
1323 int64_t sector_num, int nb_sectors,
1324 int64_t *cluster_sector_num,
1325 int *cluster_nb_sectors)
1327 BlockDriverInfo bdi;
1329 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
1330 *cluster_sector_num = sector_num;
1331 *cluster_nb_sectors = nb_sectors;
1333 int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE;
1334 *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c);
1335 *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num +
1340 static bool tracked_request_overlaps(BdrvTrackedRequest *req,
1341 int64_t sector_num, int nb_sectors) {
1343 if (sector_num >= req->sector_num + req->nb_sectors) {
1347 if (req->sector_num >= sector_num + nb_sectors) {
1353 static void coroutine_fn wait_for_overlapping_requests(BlockDriverState *bs,
1354 int64_t sector_num, int nb_sectors)
1356 BdrvTrackedRequest *req;
1357 int64_t cluster_sector_num;
1358 int cluster_nb_sectors;
1361 /* If we touch the same cluster it counts as an overlap. This guarantees
1362 * that allocating writes will be serialized and not race with each other
1363 * for the same cluster. For example, in copy-on-read it ensures that the
1364 * CoR read and write operations are atomic and guest writes cannot
1365 * interleave between them.
1367 round_to_clusters(bs, sector_num, nb_sectors,
1368 &cluster_sector_num, &cluster_nb_sectors);
1372 QLIST_FOREACH(req, &bs->tracked_requests, list) {
1373 if (tracked_request_overlaps(req, cluster_sector_num,
1374 cluster_nb_sectors)) {
1375 /* Hitting this means there was a reentrant request, for
1376 * example, a block driver issuing nested requests. This must
1377 * never happen since it means deadlock.
1379 assert(qemu_coroutine_self() != req->co);
1381 qemu_co_queue_wait(&req->wait_queue);
1392 * -EINVAL - backing format specified, but no file
1393 * -ENOSPC - can't update the backing file because no space is left in the
1395 * -ENOTSUP - format driver doesn't support changing the backing file
1397 int bdrv_change_backing_file(BlockDriverState *bs,
1398 const char *backing_file, const char *backing_fmt)
1400 BlockDriver *drv = bs->drv;
1402 if (drv->bdrv_change_backing_file != NULL) {
1403 return drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
1409 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
1414 if (!bdrv_is_inserted(bs))
1420 len = bdrv_getlength(bs);
1425 if ((offset > len) || (len - offset < size))
1431 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
1434 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
1435 nb_sectors * BDRV_SECTOR_SIZE);
1438 typedef struct RwCo {
1439 BlockDriverState *bs;
1447 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
1449 RwCo *rwco = opaque;
1451 if (!rwco->is_write) {
1452 rwco->ret = bdrv_co_do_readv(rwco->bs, rwco->sector_num,
1453 rwco->nb_sectors, rwco->qiov, 0);
1455 rwco->ret = bdrv_co_do_writev(rwco->bs, rwco->sector_num,
1456 rwco->nb_sectors, rwco->qiov, 0);
1461 * Process a synchronous request using coroutines
1463 static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf,
1464 int nb_sectors, bool is_write)
1467 struct iovec iov = {
1468 .iov_base = (void *)buf,
1469 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
1474 .sector_num = sector_num,
1475 .nb_sectors = nb_sectors,
1477 .is_write = is_write,
1481 qemu_iovec_init_external(&qiov, &iov, 1);
1484 * In sync call context, when the vcpu is blocked, this throttling timer
1485 * will not fire; so the I/O throttling function has to be disabled here
1486 * if it has been enabled.
1488 if (bs->io_limits_enabled) {
1489 fprintf(stderr, "Disabling I/O throttling on '%s' due "
1490 "to synchronous I/O.\n", bdrv_get_device_name(bs));
1491 bdrv_io_limits_disable(bs);
1494 if (qemu_in_coroutine()) {
1495 /* Fast-path if already in coroutine context */
1496 bdrv_rw_co_entry(&rwco);
1498 co = qemu_coroutine_create(bdrv_rw_co_entry);
1499 qemu_coroutine_enter(co, &rwco);
1500 while (rwco.ret == NOT_DONE) {
1507 /* return < 0 if error. See bdrv_write() for the return codes */
1508 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
1509 uint8_t *buf, int nb_sectors)
1511 return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false);
1514 static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
1515 int nb_sectors, int dirty)
1518 unsigned long val, idx, bit;
1520 start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
1521 end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
1523 for (; start <= end; start++) {
1524 idx = start / (sizeof(unsigned long) * 8);
1525 bit = start % (sizeof(unsigned long) * 8);
1526 val = bs->dirty_bitmap[idx];
1528 if (!(val & (1UL << bit))) {
1533 if (val & (1UL << bit)) {
1535 val &= ~(1UL << bit);
1538 bs->dirty_bitmap[idx] = val;
1542 /* Return < 0 if error. Important errors are:
1543 -EIO generic I/O error (may happen for all errors)
1544 -ENOMEDIUM No media inserted.
1545 -EINVAL Invalid sector number or nb_sectors
1546 -EACCES Trying to write a read-only device
1548 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
1549 const uint8_t *buf, int nb_sectors)
1551 return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true);
1554 int bdrv_pread(BlockDriverState *bs, int64_t offset,
1555 void *buf, int count1)
1557 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1558 int len, nb_sectors, count;
1563 /* first read to align to sector start */
1564 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1567 sector_num = offset >> BDRV_SECTOR_BITS;
1569 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1571 memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
1579 /* read the sectors "in place" */
1580 nb_sectors = count >> BDRV_SECTOR_BITS;
1581 if (nb_sectors > 0) {
1582 if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
1584 sector_num += nb_sectors;
1585 len = nb_sectors << BDRV_SECTOR_BITS;
1590 /* add data from the last sector */
1592 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1594 memcpy(buf, tmp_buf, count);
1599 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
1600 const void *buf, int count1)
1602 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1603 int len, nb_sectors, count;
1608 /* first write to align to sector start */
1609 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1612 sector_num = offset >> BDRV_SECTOR_BITS;
1614 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1616 memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
1617 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1626 /* write the sectors "in place" */
1627 nb_sectors = count >> BDRV_SECTOR_BITS;
1628 if (nb_sectors > 0) {
1629 if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0)
1631 sector_num += nb_sectors;
1632 len = nb_sectors << BDRV_SECTOR_BITS;
1637 /* add data from the last sector */
1639 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1641 memcpy(tmp_buf, buf, count);
1642 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1649 * Writes to the file and ensures that no writes are reordered across this
1650 * request (acts as a barrier)
1652 * Returns 0 on success, -errno in error cases.
1654 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
1655 const void *buf, int count)
1659 ret = bdrv_pwrite(bs, offset, buf, count);
1664 /* No flush needed for cache modes that use O_DSYNC */
1665 if ((bs->open_flags & BDRV_O_CACHE_WB) != 0) {
1672 static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs,
1673 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1675 /* Perform I/O through a temporary buffer so that users who scribble over
1676 * their read buffer while the operation is in progress do not end up
1677 * modifying the image file. This is critical for zero-copy guest I/O
1678 * where anything might happen inside guest memory.
1680 void *bounce_buffer;
1682 BlockDriver *drv = bs->drv;
1684 QEMUIOVector bounce_qiov;
1685 int64_t cluster_sector_num;
1686 int cluster_nb_sectors;
1690 /* Cover entire cluster so no additional backing file I/O is required when
1691 * allocating cluster in the image file.
1693 round_to_clusters(bs, sector_num, nb_sectors,
1694 &cluster_sector_num, &cluster_nb_sectors);
1696 trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors,
1697 cluster_sector_num, cluster_nb_sectors);
1699 iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE;
1700 iov.iov_base = bounce_buffer = qemu_blockalign(bs, iov.iov_len);
1701 qemu_iovec_init_external(&bounce_qiov, &iov, 1);
1703 ret = drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors,
1709 if (drv->bdrv_co_write_zeroes &&
1710 buffer_is_zero(bounce_buffer, iov.iov_len)) {
1711 ret = drv->bdrv_co_write_zeroes(bs, cluster_sector_num,
1712 cluster_nb_sectors);
1714 ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors,
1719 /* It might be okay to ignore write errors for guest requests. If this
1720 * is a deliberate copy-on-read then we don't want to ignore the error.
1721 * Simply report it in all cases.
1726 skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE;
1727 qemu_iovec_from_buffer(qiov, bounce_buffer + skip_bytes,
1728 nb_sectors * BDRV_SECTOR_SIZE);
1731 qemu_vfree(bounce_buffer);
1736 * Handle a read request in coroutine context
1738 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
1739 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1740 BdrvRequestFlags flags)
1742 BlockDriver *drv = bs->drv;
1743 BdrvTrackedRequest req;
1749 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
1753 /* throttling disk read I/O */
1754 if (bs->io_limits_enabled) {
1755 bdrv_io_limits_intercept(bs, false, nb_sectors);
1758 if (bs->copy_on_read) {
1759 flags |= BDRV_REQ_COPY_ON_READ;
1761 if (flags & BDRV_REQ_COPY_ON_READ) {
1762 bs->copy_on_read_in_flight++;
1765 if (bs->copy_on_read_in_flight) {
1766 wait_for_overlapping_requests(bs, sector_num, nb_sectors);
1769 tracked_request_begin(&req, bs, sector_num, nb_sectors, false);
1771 if (flags & BDRV_REQ_COPY_ON_READ) {
1774 ret = bdrv_co_is_allocated(bs, sector_num, nb_sectors, &pnum);
1779 if (!ret || pnum != nb_sectors) {
1780 ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov);
1785 ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
1788 tracked_request_end(&req);
1790 if (flags & BDRV_REQ_COPY_ON_READ) {
1791 bs->copy_on_read_in_flight--;
1797 int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
1798 int nb_sectors, QEMUIOVector *qiov)
1800 trace_bdrv_co_readv(bs, sector_num, nb_sectors);
1802 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0);
1805 int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
1806 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1808 trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors);
1810 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
1811 BDRV_REQ_COPY_ON_READ);
1814 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
1815 int64_t sector_num, int nb_sectors)
1817 BlockDriver *drv = bs->drv;
1822 /* First try the efficient write zeroes operation */
1823 if (drv->bdrv_co_write_zeroes) {
1824 return drv->bdrv_co_write_zeroes(bs, sector_num, nb_sectors);
1827 /* Fall back to bounce buffer if write zeroes is unsupported */
1828 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
1829 iov.iov_base = qemu_blockalign(bs, iov.iov_len);
1830 memset(iov.iov_base, 0, iov.iov_len);
1831 qemu_iovec_init_external(&qiov, &iov, 1);
1833 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, &qiov);
1835 qemu_vfree(iov.iov_base);
1840 * Handle a write request in coroutine context
1842 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
1843 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1844 BdrvRequestFlags flags)
1846 BlockDriver *drv = bs->drv;
1847 BdrvTrackedRequest req;
1853 if (bs->read_only) {
1856 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
1860 /* throttling disk write I/O */
1861 if (bs->io_limits_enabled) {
1862 bdrv_io_limits_intercept(bs, true, nb_sectors);
1865 if (bs->copy_on_read_in_flight) {
1866 wait_for_overlapping_requests(bs, sector_num, nb_sectors);
1869 tracked_request_begin(&req, bs, sector_num, nb_sectors, true);
1871 if (flags & BDRV_REQ_ZERO_WRITE) {
1872 ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors);
1874 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
1877 if (bs->dirty_bitmap) {
1878 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1881 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
1882 bs->wr_highest_sector = sector_num + nb_sectors - 1;
1885 tracked_request_end(&req);
1890 int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
1891 int nb_sectors, QEMUIOVector *qiov)
1893 trace_bdrv_co_writev(bs, sector_num, nb_sectors);
1895 return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0);
1898 int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs,
1899 int64_t sector_num, int nb_sectors)
1901 trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors);
1903 return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL,
1904 BDRV_REQ_ZERO_WRITE);
1908 * Truncate file to 'offset' bytes (needed only for file protocols)
1910 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
1912 BlockDriver *drv = bs->drv;
1916 if (!drv->bdrv_truncate)
1920 if (bdrv_in_use(bs))
1922 ret = drv->bdrv_truncate(bs, offset);
1924 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
1925 bdrv_dev_resize_cb(bs);
1931 * Length of a allocated file in bytes. Sparse files are counted by actual
1932 * allocated space. Return < 0 if error or unknown.
1934 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
1936 BlockDriver *drv = bs->drv;
1940 if (drv->bdrv_get_allocated_file_size) {
1941 return drv->bdrv_get_allocated_file_size(bs);
1944 return bdrv_get_allocated_file_size(bs->file);
1950 * Length of a file in bytes. Return < 0 if error or unknown.
1952 int64_t bdrv_getlength(BlockDriverState *bs)
1954 BlockDriver *drv = bs->drv;
1958 if (bs->growable || bdrv_dev_has_removable_media(bs)) {
1959 if (drv->bdrv_getlength) {
1960 return drv->bdrv_getlength(bs);
1963 return bs->total_sectors * BDRV_SECTOR_SIZE;
1966 /* return 0 as number of sectors if no device present or error */
1967 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
1970 length = bdrv_getlength(bs);
1974 length = length >> BDRV_SECTOR_BITS;
1975 *nb_sectors_ptr = length;
1979 uint8_t boot_ind; /* 0x80 - active */
1980 uint8_t head; /* starting head */
1981 uint8_t sector; /* starting sector */
1982 uint8_t cyl; /* starting cylinder */
1983 uint8_t sys_ind; /* What partition type */
1984 uint8_t end_head; /* end head */
1985 uint8_t end_sector; /* end sector */
1986 uint8_t end_cyl; /* end cylinder */
1987 uint32_t start_sect; /* starting sector counting from 0 */
1988 uint32_t nr_sects; /* nr of sectors in partition */
1991 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
1992 static int guess_disk_lchs(BlockDriverState *bs,
1993 int *pcylinders, int *pheads, int *psectors)
1995 uint8_t buf[BDRV_SECTOR_SIZE];
1996 int ret, i, heads, sectors, cylinders;
1997 struct partition *p;
1999 uint64_t nb_sectors;
2002 bdrv_get_geometry(bs, &nb_sectors);
2005 * The function will be invoked during startup not only in sync I/O mode,
2006 * but also in async I/O mode. So the I/O throttling function has to
2007 * be disabled temporarily here, not permanently.
2009 enabled = bs->io_limits_enabled;
2010 bs->io_limits_enabled = false;
2011 ret = bdrv_read(bs, 0, buf, 1);
2012 bs->io_limits_enabled = enabled;
2015 /* test msdos magic */
2016 if (buf[510] != 0x55 || buf[511] != 0xaa)
2018 for(i = 0; i < 4; i++) {
2019 p = ((struct partition *)(buf + 0x1be)) + i;
2020 nr_sects = le32_to_cpu(p->nr_sects);
2021 if (nr_sects && p->end_head) {
2022 /* We make the assumption that the partition terminates on
2023 a cylinder boundary */
2024 heads = p->end_head + 1;
2025 sectors = p->end_sector & 63;
2028 cylinders = nb_sectors / (heads * sectors);
2029 if (cylinders < 1 || cylinders > 16383)
2032 *psectors = sectors;
2033 *pcylinders = cylinders;
2035 printf("guessed geometry: LCHS=%d %d %d\n",
2036 cylinders, heads, sectors);
2044 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
2046 int translation, lba_detected = 0;
2047 int cylinders, heads, secs;
2048 uint64_t nb_sectors;
2050 /* if a geometry hint is available, use it */
2051 bdrv_get_geometry(bs, &nb_sectors);
2052 bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
2053 translation = bdrv_get_translation_hint(bs);
2054 if (cylinders != 0) {
2059 if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
2061 /* if heads > 16, it means that a BIOS LBA
2062 translation was active, so the default
2063 hardware geometry is OK */
2065 goto default_geometry;
2070 /* disable any translation to be in sync with
2071 the logical geometry */
2072 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
2073 bdrv_set_translation_hint(bs,
2074 BIOS_ATA_TRANSLATION_NONE);
2079 /* if no geometry, use a standard physical disk geometry */
2080 cylinders = nb_sectors / (16 * 63);
2082 if (cylinders > 16383)
2084 else if (cylinders < 2)
2089 if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
2090 if ((*pcyls * *pheads) <= 131072) {
2091 bdrv_set_translation_hint(bs,
2092 BIOS_ATA_TRANSLATION_LARGE);
2094 bdrv_set_translation_hint(bs,
2095 BIOS_ATA_TRANSLATION_LBA);
2099 bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
2103 void bdrv_set_geometry_hint(BlockDriverState *bs,
2104 int cyls, int heads, int secs)
2111 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
2113 bs->translation = translation;
2116 void bdrv_get_geometry_hint(BlockDriverState *bs,
2117 int *pcyls, int *pheads, int *psecs)
2120 *pheads = bs->heads;
2124 /* throttling disk io limits */
2125 void bdrv_set_io_limits(BlockDriverState *bs,
2126 BlockIOLimit *io_limits)
2128 bs->io_limits = *io_limits;
2129 bs->io_limits_enabled = bdrv_io_limits_enabled(bs);
2132 /* Recognize floppy formats */
2133 typedef struct FDFormat {
2141 static const FDFormat fd_formats[] = {
2142 /* First entry is default format */
2143 /* 1.44 MB 3"1/2 floppy disks */
2144 { FDRIVE_DRV_144, 18, 80, 1, FDRIVE_RATE_500K, },
2145 { FDRIVE_DRV_144, 20, 80, 1, FDRIVE_RATE_500K, },
2146 { FDRIVE_DRV_144, 21, 80, 1, FDRIVE_RATE_500K, },
2147 { FDRIVE_DRV_144, 21, 82, 1, FDRIVE_RATE_500K, },
2148 { FDRIVE_DRV_144, 21, 83, 1, FDRIVE_RATE_500K, },
2149 { FDRIVE_DRV_144, 22, 80, 1, FDRIVE_RATE_500K, },
2150 { FDRIVE_DRV_144, 23, 80, 1, FDRIVE_RATE_500K, },
2151 { FDRIVE_DRV_144, 24, 80, 1, FDRIVE_RATE_500K, },
2152 /* 2.88 MB 3"1/2 floppy disks */
2153 { FDRIVE_DRV_288, 36, 80, 1, FDRIVE_RATE_1M, },
2154 { FDRIVE_DRV_288, 39, 80, 1, FDRIVE_RATE_1M, },
2155 { FDRIVE_DRV_288, 40, 80, 1, FDRIVE_RATE_1M, },
2156 { FDRIVE_DRV_288, 44, 80, 1, FDRIVE_RATE_1M, },
2157 { FDRIVE_DRV_288, 48, 80, 1, FDRIVE_RATE_1M, },
2158 /* 720 kB 3"1/2 floppy disks */
2159 { FDRIVE_DRV_144, 9, 80, 1, FDRIVE_RATE_250K, },
2160 { FDRIVE_DRV_144, 10, 80, 1, FDRIVE_RATE_250K, },
2161 { FDRIVE_DRV_144, 10, 82, 1, FDRIVE_RATE_250K, },
2162 { FDRIVE_DRV_144, 10, 83, 1, FDRIVE_RATE_250K, },
2163 { FDRIVE_DRV_144, 13, 80, 1, FDRIVE_RATE_250K, },
2164 { FDRIVE_DRV_144, 14, 80, 1, FDRIVE_RATE_250K, },
2165 /* 1.2 MB 5"1/4 floppy disks */
2166 { FDRIVE_DRV_120, 15, 80, 1, FDRIVE_RATE_500K, },
2167 { FDRIVE_DRV_120, 18, 80, 1, FDRIVE_RATE_500K, },
2168 { FDRIVE_DRV_120, 18, 82, 1, FDRIVE_RATE_500K, },
2169 { FDRIVE_DRV_120, 18, 83, 1, FDRIVE_RATE_500K, },
2170 { FDRIVE_DRV_120, 20, 80, 1, FDRIVE_RATE_500K, },
2171 /* 720 kB 5"1/4 floppy disks */
2172 { FDRIVE_DRV_120, 9, 80, 1, FDRIVE_RATE_250K, },
2173 { FDRIVE_DRV_120, 11, 80, 1, FDRIVE_RATE_250K, },
2174 /* 360 kB 5"1/4 floppy disks */
2175 { FDRIVE_DRV_120, 9, 40, 1, FDRIVE_RATE_300K, },
2176 { FDRIVE_DRV_120, 9, 40, 0, FDRIVE_RATE_300K, },
2177 { FDRIVE_DRV_120, 10, 41, 1, FDRIVE_RATE_300K, },
2178 { FDRIVE_DRV_120, 10, 42, 1, FDRIVE_RATE_300K, },
2179 /* 320 kB 5"1/4 floppy disks */
2180 { FDRIVE_DRV_120, 8, 40, 1, FDRIVE_RATE_250K, },
2181 { FDRIVE_DRV_120, 8, 40, 0, FDRIVE_RATE_250K, },
2182 /* 360 kB must match 5"1/4 better than 3"1/2... */
2183 { FDRIVE_DRV_144, 9, 80, 0, FDRIVE_RATE_250K, },
2185 { FDRIVE_DRV_NONE, -1, -1, 0, 0, },
2188 void bdrv_get_floppy_geometry_hint(BlockDriverState *bs, int *nb_heads,
2189 int *max_track, int *last_sect,
2190 FDriveType drive_in, FDriveType *drive,
2193 const FDFormat *parse;
2194 uint64_t nb_sectors, size;
2195 int i, first_match, match;
2197 bdrv_get_geometry_hint(bs, nb_heads, max_track, last_sect);
2198 if (*nb_heads != 0 && *max_track != 0 && *last_sect != 0) {
2199 /* User defined disk */
2200 *rate = FDRIVE_RATE_500K;
2202 bdrv_get_geometry(bs, &nb_sectors);
2205 for (i = 0; ; i++) {
2206 parse = &fd_formats[i];
2207 if (parse->drive == FDRIVE_DRV_NONE) {
2210 if (drive_in == parse->drive ||
2211 drive_in == FDRIVE_DRV_NONE) {
2212 size = (parse->max_head + 1) * parse->max_track *
2214 if (nb_sectors == size) {
2218 if (first_match == -1) {
2224 if (first_match == -1) {
2227 match = first_match;
2229 parse = &fd_formats[match];
2231 *nb_heads = parse->max_head + 1;
2232 *max_track = parse->max_track;
2233 *last_sect = parse->last_sect;
2234 *drive = parse->drive;
2235 *rate = parse->rate;
2239 int bdrv_get_translation_hint(BlockDriverState *bs)
2241 return bs->translation;
2244 void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
2245 BlockErrorAction on_write_error)
2247 bs->on_read_error = on_read_error;
2248 bs->on_write_error = on_write_error;
2251 BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
2253 return is_read ? bs->on_read_error : bs->on_write_error;
2256 int bdrv_is_read_only(BlockDriverState *bs)
2258 return bs->read_only;
2261 int bdrv_is_sg(BlockDriverState *bs)
2266 int bdrv_enable_write_cache(BlockDriverState *bs)
2268 return bs->enable_write_cache;
2271 int bdrv_is_encrypted(BlockDriverState *bs)
2273 if (bs->backing_hd && bs->backing_hd->encrypted)
2275 return bs->encrypted;
2278 int bdrv_key_required(BlockDriverState *bs)
2280 BlockDriverState *backing_hd = bs->backing_hd;
2282 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
2284 return (bs->encrypted && !bs->valid_key);
2287 int bdrv_set_key(BlockDriverState *bs, const char *key)
2290 if (bs->backing_hd && bs->backing_hd->encrypted) {
2291 ret = bdrv_set_key(bs->backing_hd, key);
2297 if (!bs->encrypted) {
2299 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
2302 ret = bs->drv->bdrv_set_key(bs, key);
2305 } else if (!bs->valid_key) {
2307 /* call the change callback now, we skipped it on open */
2308 bdrv_dev_change_media_cb(bs, true);
2313 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
2318 pstrcpy(buf, buf_size, bs->drv->format_name);
2322 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
2327 QLIST_FOREACH(drv, &bdrv_drivers, list) {
2328 it(opaque, drv->format_name);
2332 BlockDriverState *bdrv_find(const char *name)
2334 BlockDriverState *bs;
2336 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2337 if (!strcmp(name, bs->device_name)) {
2344 BlockDriverState *bdrv_next(BlockDriverState *bs)
2347 return QTAILQ_FIRST(&bdrv_states);
2349 return QTAILQ_NEXT(bs, list);
2352 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
2354 BlockDriverState *bs;
2356 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2361 const char *bdrv_get_device_name(BlockDriverState *bs)
2363 return bs->device_name;
2366 void bdrv_flush_all(void)
2368 BlockDriverState *bs;
2370 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2375 int bdrv_has_zero_init(BlockDriverState *bs)
2379 if (bs->drv->bdrv_has_zero_init) {
2380 return bs->drv->bdrv_has_zero_init(bs);
2386 typedef struct BdrvCoIsAllocatedData {
2387 BlockDriverState *bs;
2393 } BdrvCoIsAllocatedData;
2396 * Returns true iff the specified sector is present in the disk image. Drivers
2397 * not implementing the functionality are assumed to not support backing files,
2398 * hence all their sectors are reported as allocated.
2400 * If 'sector_num' is beyond the end of the disk image the return value is 0
2401 * and 'pnum' is set to 0.
2403 * 'pnum' is set to the number of sectors (including and immediately following
2404 * the specified sector) that are known to be in the same
2405 * allocated/unallocated state.
2407 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
2408 * beyond the end of the disk image it will be clamped.
2410 int coroutine_fn bdrv_co_is_allocated(BlockDriverState *bs, int64_t sector_num,
2411 int nb_sectors, int *pnum)
2415 if (sector_num >= bs->total_sectors) {
2420 n = bs->total_sectors - sector_num;
2421 if (n < nb_sectors) {
2425 if (!bs->drv->bdrv_co_is_allocated) {
2430 return bs->drv->bdrv_co_is_allocated(bs, sector_num, nb_sectors, pnum);
2433 /* Coroutine wrapper for bdrv_is_allocated() */
2434 static void coroutine_fn bdrv_is_allocated_co_entry(void *opaque)
2436 BdrvCoIsAllocatedData *data = opaque;
2437 BlockDriverState *bs = data->bs;
2439 data->ret = bdrv_co_is_allocated(bs, data->sector_num, data->nb_sectors,
2445 * Synchronous wrapper around bdrv_co_is_allocated().
2447 * See bdrv_co_is_allocated() for details.
2449 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
2453 BdrvCoIsAllocatedData data = {
2455 .sector_num = sector_num,
2456 .nb_sectors = nb_sectors,
2461 co = qemu_coroutine_create(bdrv_is_allocated_co_entry);
2462 qemu_coroutine_enter(co, &data);
2463 while (!data.done) {
2469 BlockInfoList *qmp_query_block(Error **errp)
2471 BlockInfoList *head = NULL, *cur_item = NULL;
2472 BlockDriverState *bs;
2474 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2475 BlockInfoList *info = g_malloc0(sizeof(*info));
2477 info->value = g_malloc0(sizeof(*info->value));
2478 info->value->device = g_strdup(bs->device_name);
2479 info->value->type = g_strdup("unknown");
2480 info->value->locked = bdrv_dev_is_medium_locked(bs);
2481 info->value->removable = bdrv_dev_has_removable_media(bs);
2483 if (bdrv_dev_has_removable_media(bs)) {
2484 info->value->has_tray_open = true;
2485 info->value->tray_open = bdrv_dev_is_tray_open(bs);
2488 if (bdrv_iostatus_is_enabled(bs)) {
2489 info->value->has_io_status = true;
2490 info->value->io_status = bs->iostatus;
2494 info->value->has_inserted = true;
2495 info->value->inserted = g_malloc0(sizeof(*info->value->inserted));
2496 info->value->inserted->file = g_strdup(bs->filename);
2497 info->value->inserted->ro = bs->read_only;
2498 info->value->inserted->drv = g_strdup(bs->drv->format_name);
2499 info->value->inserted->encrypted = bs->encrypted;
2500 if (bs->backing_file[0]) {
2501 info->value->inserted->has_backing_file = true;
2502 info->value->inserted->backing_file = g_strdup(bs->backing_file);
2505 if (bs->io_limits_enabled) {
2506 info->value->inserted->bps =
2507 bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL];
2508 info->value->inserted->bps_rd =
2509 bs->io_limits.bps[BLOCK_IO_LIMIT_READ];
2510 info->value->inserted->bps_wr =
2511 bs->io_limits.bps[BLOCK_IO_LIMIT_WRITE];
2512 info->value->inserted->iops =
2513 bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL];
2514 info->value->inserted->iops_rd =
2515 bs->io_limits.iops[BLOCK_IO_LIMIT_READ];
2516 info->value->inserted->iops_wr =
2517 bs->io_limits.iops[BLOCK_IO_LIMIT_WRITE];
2521 /* XXX: waiting for the qapi to support GSList */
2523 head = cur_item = info;
2525 cur_item->next = info;
2533 /* Consider exposing this as a full fledged QMP command */
2534 static BlockStats *qmp_query_blockstat(const BlockDriverState *bs, Error **errp)
2538 s = g_malloc0(sizeof(*s));
2540 if (bs->device_name[0]) {
2541 s->has_device = true;
2542 s->device = g_strdup(bs->device_name);
2545 s->stats = g_malloc0(sizeof(*s->stats));
2546 s->stats->rd_bytes = bs->nr_bytes[BDRV_ACCT_READ];
2547 s->stats->wr_bytes = bs->nr_bytes[BDRV_ACCT_WRITE];
2548 s->stats->rd_operations = bs->nr_ops[BDRV_ACCT_READ];
2549 s->stats->wr_operations = bs->nr_ops[BDRV_ACCT_WRITE];
2550 s->stats->wr_highest_offset = bs->wr_highest_sector * BDRV_SECTOR_SIZE;
2551 s->stats->flush_operations = bs->nr_ops[BDRV_ACCT_FLUSH];
2552 s->stats->wr_total_time_ns = bs->total_time_ns[BDRV_ACCT_WRITE];
2553 s->stats->rd_total_time_ns = bs->total_time_ns[BDRV_ACCT_READ];
2554 s->stats->flush_total_time_ns = bs->total_time_ns[BDRV_ACCT_FLUSH];
2557 s->has_parent = true;
2558 s->parent = qmp_query_blockstat(bs->file, NULL);
2564 BlockStatsList *qmp_query_blockstats(Error **errp)
2566 BlockStatsList *head = NULL, *cur_item = NULL;
2567 BlockDriverState *bs;
2569 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2570 BlockStatsList *info = g_malloc0(sizeof(*info));
2571 info->value = qmp_query_blockstat(bs, NULL);
2573 /* XXX: waiting for the qapi to support GSList */
2575 head = cur_item = info;
2577 cur_item->next = info;
2585 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
2587 if (bs->backing_hd && bs->backing_hd->encrypted)
2588 return bs->backing_file;
2589 else if (bs->encrypted)
2590 return bs->filename;
2595 void bdrv_get_backing_filename(BlockDriverState *bs,
2596 char *filename, int filename_size)
2598 pstrcpy(filename, filename_size, bs->backing_file);
2601 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
2602 const uint8_t *buf, int nb_sectors)
2604 BlockDriver *drv = bs->drv;
2607 if (!drv->bdrv_write_compressed)
2609 if (bdrv_check_request(bs, sector_num, nb_sectors))
2612 if (bs->dirty_bitmap) {
2613 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
2616 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
2619 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2621 BlockDriver *drv = bs->drv;
2624 if (!drv->bdrv_get_info)
2626 memset(bdi, 0, sizeof(*bdi));
2627 return drv->bdrv_get_info(bs, bdi);
2630 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
2631 int64_t pos, int size)
2633 BlockDriver *drv = bs->drv;
2636 if (drv->bdrv_save_vmstate)
2637 return drv->bdrv_save_vmstate(bs, buf, pos, size);
2639 return bdrv_save_vmstate(bs->file, buf, pos, size);
2643 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
2644 int64_t pos, int size)
2646 BlockDriver *drv = bs->drv;
2649 if (drv->bdrv_load_vmstate)
2650 return drv->bdrv_load_vmstate(bs, buf, pos, size);
2652 return bdrv_load_vmstate(bs->file, buf, pos, size);
2656 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
2658 BlockDriver *drv = bs->drv;
2660 if (!drv || !drv->bdrv_debug_event) {
2664 return drv->bdrv_debug_event(bs, event);
2668 /**************************************************************/
2669 /* handling of snapshots */
2671 int bdrv_can_snapshot(BlockDriverState *bs)
2673 BlockDriver *drv = bs->drv;
2674 if (!drv || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
2678 if (!drv->bdrv_snapshot_create) {
2679 if (bs->file != NULL) {
2680 return bdrv_can_snapshot(bs->file);
2688 int bdrv_is_snapshot(BlockDriverState *bs)
2690 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
2693 BlockDriverState *bdrv_snapshots(void)
2695 BlockDriverState *bs;
2698 return bs_snapshots;
2702 while ((bs = bdrv_next(bs))) {
2703 if (bdrv_can_snapshot(bs)) {
2711 int bdrv_snapshot_create(BlockDriverState *bs,
2712 QEMUSnapshotInfo *sn_info)
2714 BlockDriver *drv = bs->drv;
2717 if (drv->bdrv_snapshot_create)
2718 return drv->bdrv_snapshot_create(bs, sn_info);
2720 return bdrv_snapshot_create(bs->file, sn_info);
2724 int bdrv_snapshot_goto(BlockDriverState *bs,
2725 const char *snapshot_id)
2727 BlockDriver *drv = bs->drv;
2732 if (drv->bdrv_snapshot_goto)
2733 return drv->bdrv_snapshot_goto(bs, snapshot_id);
2736 drv->bdrv_close(bs);
2737 ret = bdrv_snapshot_goto(bs->file, snapshot_id);
2738 open_ret = drv->bdrv_open(bs, bs->open_flags);
2740 bdrv_delete(bs->file);
2750 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2752 BlockDriver *drv = bs->drv;
2755 if (drv->bdrv_snapshot_delete)
2756 return drv->bdrv_snapshot_delete(bs, snapshot_id);
2758 return bdrv_snapshot_delete(bs->file, snapshot_id);
2762 int bdrv_snapshot_list(BlockDriverState *bs,
2763 QEMUSnapshotInfo **psn_info)
2765 BlockDriver *drv = bs->drv;
2768 if (drv->bdrv_snapshot_list)
2769 return drv->bdrv_snapshot_list(bs, psn_info);
2771 return bdrv_snapshot_list(bs->file, psn_info);
2775 int bdrv_snapshot_load_tmp(BlockDriverState *bs,
2776 const char *snapshot_name)
2778 BlockDriver *drv = bs->drv;
2782 if (!bs->read_only) {
2785 if (drv->bdrv_snapshot_load_tmp) {
2786 return drv->bdrv_snapshot_load_tmp(bs, snapshot_name);
2791 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
2792 const char *backing_file)
2798 if (bs->backing_hd) {
2799 if (strcmp(bs->backing_file, backing_file) == 0) {
2800 return bs->backing_hd;
2802 return bdrv_find_backing_image(bs->backing_hd, backing_file);
2809 #define NB_SUFFIXES 4
2811 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
2813 static const char suffixes[NB_SUFFIXES] = "KMGT";
2818 snprintf(buf, buf_size, "%" PRId64, size);
2821 for(i = 0; i < NB_SUFFIXES; i++) {
2822 if (size < (10 * base)) {
2823 snprintf(buf, buf_size, "%0.1f%c",
2824 (double)size / base,
2827 } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
2828 snprintf(buf, buf_size, "%" PRId64 "%c",
2829 ((size + (base >> 1)) / base),
2839 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
2841 char buf1[128], date_buf[128], clock_buf[128];
2851 snprintf(buf, buf_size,
2852 "%-10s%-20s%7s%20s%15s",
2853 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
2857 ptm = localtime(&ti);
2858 strftime(date_buf, sizeof(date_buf),
2859 "%Y-%m-%d %H:%M:%S", ptm);
2861 localtime_r(&ti, &tm);
2862 strftime(date_buf, sizeof(date_buf),
2863 "%Y-%m-%d %H:%M:%S", &tm);
2865 secs = sn->vm_clock_nsec / 1000000000;
2866 snprintf(clock_buf, sizeof(clock_buf),
2867 "%02d:%02d:%02d.%03d",
2869 (int)((secs / 60) % 60),
2871 (int)((sn->vm_clock_nsec / 1000000) % 1000));
2872 snprintf(buf, buf_size,
2873 "%-10s%-20s%7s%20s%15s",
2874 sn->id_str, sn->name,
2875 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
2882 /**************************************************************/
2885 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
2886 QEMUIOVector *qiov, int nb_sectors,
2887 BlockDriverCompletionFunc *cb, void *opaque)
2889 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
2891 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
2895 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
2896 QEMUIOVector *qiov, int nb_sectors,
2897 BlockDriverCompletionFunc *cb, void *opaque)
2899 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
2901 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
2906 typedef struct MultiwriteCB {
2911 BlockDriverCompletionFunc *cb;
2913 QEMUIOVector *free_qiov;
2917 static void multiwrite_user_cb(MultiwriteCB *mcb)
2921 for (i = 0; i < mcb->num_callbacks; i++) {
2922 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
2923 if (mcb->callbacks[i].free_qiov) {
2924 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
2926 g_free(mcb->callbacks[i].free_qiov);
2930 static void multiwrite_cb(void *opaque, int ret)
2932 MultiwriteCB *mcb = opaque;
2934 trace_multiwrite_cb(mcb, ret);
2936 if (ret < 0 && !mcb->error) {
2940 mcb->num_requests--;
2941 if (mcb->num_requests == 0) {
2942 multiwrite_user_cb(mcb);
2947 static int multiwrite_req_compare(const void *a, const void *b)
2949 const BlockRequest *req1 = a, *req2 = b;
2952 * Note that we can't simply subtract req2->sector from req1->sector
2953 * here as that could overflow the return value.
2955 if (req1->sector > req2->sector) {
2957 } else if (req1->sector < req2->sector) {
2965 * Takes a bunch of requests and tries to merge them. Returns the number of
2966 * requests that remain after merging.
2968 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
2969 int num_reqs, MultiwriteCB *mcb)
2973 // Sort requests by start sector
2974 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
2976 // Check if adjacent requests touch the same clusters. If so, combine them,
2977 // filling up gaps with zero sectors.
2979 for (i = 1; i < num_reqs; i++) {
2981 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
2983 // Handle exactly sequential writes and overlapping writes.
2984 if (reqs[i].sector <= oldreq_last) {
2988 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
2994 QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
2995 qemu_iovec_init(qiov,
2996 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
2998 // Add the first request to the merged one. If the requests are
2999 // overlapping, drop the last sectors of the first request.
3000 size = (reqs[i].sector - reqs[outidx].sector) << 9;
3001 qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
3003 // We should need to add any zeros between the two requests
3004 assert (reqs[i].sector <= oldreq_last);
3006 // Add the second request
3007 qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
3009 reqs[outidx].nb_sectors = qiov->size >> 9;
3010 reqs[outidx].qiov = qiov;
3012 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
3015 reqs[outidx].sector = reqs[i].sector;
3016 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
3017 reqs[outidx].qiov = reqs[i].qiov;
3025 * Submit multiple AIO write requests at once.
3027 * On success, the function returns 0 and all requests in the reqs array have
3028 * been submitted. In error case this function returns -1, and any of the
3029 * requests may or may not be submitted yet. In particular, this means that the
3030 * callback will be called for some of the requests, for others it won't. The
3031 * caller must check the error field of the BlockRequest to wait for the right
3032 * callbacks (if error != 0, no callback will be called).
3034 * The implementation may modify the contents of the reqs array, e.g. to merge
3035 * requests. However, the fields opaque and error are left unmodified as they
3036 * are used to signal failure for a single request to the caller.
3038 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
3043 /* don't submit writes if we don't have a medium */
3044 if (bs->drv == NULL) {
3045 for (i = 0; i < num_reqs; i++) {
3046 reqs[i].error = -ENOMEDIUM;
3051 if (num_reqs == 0) {
3055 // Create MultiwriteCB structure
3056 mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
3057 mcb->num_requests = 0;
3058 mcb->num_callbacks = num_reqs;
3060 for (i = 0; i < num_reqs; i++) {
3061 mcb->callbacks[i].cb = reqs[i].cb;
3062 mcb->callbacks[i].opaque = reqs[i].opaque;
3065 // Check for mergable requests
3066 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
3068 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
3070 /* Run the aio requests. */
3071 mcb->num_requests = num_reqs;
3072 for (i = 0; i < num_reqs; i++) {
3073 bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
3074 reqs[i].nb_sectors, multiwrite_cb, mcb);
3080 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
3082 acb->pool->cancel(acb);
3085 /* block I/O throttling */
3086 static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors,
3087 bool is_write, double elapsed_time, uint64_t *wait)
3089 uint64_t bps_limit = 0;
3090 double bytes_limit, bytes_base, bytes_res;
3091 double slice_time, wait_time;
3093 if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
3094 bps_limit = bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL];
3095 } else if (bs->io_limits.bps[is_write]) {
3096 bps_limit = bs->io_limits.bps[is_write];
3105 slice_time = bs->slice_end - bs->slice_start;
3106 slice_time /= (NANOSECONDS_PER_SECOND);
3107 bytes_limit = bps_limit * slice_time;
3108 bytes_base = bs->nr_bytes[is_write] - bs->io_base.bytes[is_write];
3109 if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
3110 bytes_base += bs->nr_bytes[!is_write] - bs->io_base.bytes[!is_write];
3113 /* bytes_base: the bytes of data which have been read/written; and
3114 * it is obtained from the history statistic info.
3115 * bytes_res: the remaining bytes of data which need to be read/written.
3116 * (bytes_base + bytes_res) / bps_limit: used to calcuate
3117 * the total time for completing reading/writting all data.
3119 bytes_res = (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
3121 if (bytes_base + bytes_res <= bytes_limit) {
3129 /* Calc approx time to dispatch */
3130 wait_time = (bytes_base + bytes_res) / bps_limit - elapsed_time;
3132 /* When the I/O rate at runtime exceeds the limits,
3133 * bs->slice_end need to be extended in order that the current statistic
3134 * info can be kept until the timer fire, so it is increased and tuned
3135 * based on the result of experiment.
3137 bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10;
3138 bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME;
3140 *wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
3146 static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write,
3147 double elapsed_time, uint64_t *wait)
3149 uint64_t iops_limit = 0;
3150 double ios_limit, ios_base;
3151 double slice_time, wait_time;
3153 if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
3154 iops_limit = bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL];
3155 } else if (bs->io_limits.iops[is_write]) {
3156 iops_limit = bs->io_limits.iops[is_write];
3165 slice_time = bs->slice_end - bs->slice_start;
3166 slice_time /= (NANOSECONDS_PER_SECOND);
3167 ios_limit = iops_limit * slice_time;
3168 ios_base = bs->nr_ops[is_write] - bs->io_base.ios[is_write];
3169 if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
3170 ios_base += bs->nr_ops[!is_write] - bs->io_base.ios[!is_write];
3173 if (ios_base + 1 <= ios_limit) {
3181 /* Calc approx time to dispatch */
3182 wait_time = (ios_base + 1) / iops_limit;
3183 if (wait_time > elapsed_time) {
3184 wait_time = wait_time - elapsed_time;
3189 bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10;
3190 bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME;
3192 *wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
3198 static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors,
3199 bool is_write, int64_t *wait)
3201 int64_t now, max_wait;
3202 uint64_t bps_wait = 0, iops_wait = 0;
3203 double elapsed_time;
3204 int bps_ret, iops_ret;
3206 now = qemu_get_clock_ns(vm_clock);
3207 if ((bs->slice_start < now)
3208 && (bs->slice_end > now)) {
3209 bs->slice_end = now + bs->slice_time;
3211 bs->slice_time = 5 * BLOCK_IO_SLICE_TIME;
3212 bs->slice_start = now;
3213 bs->slice_end = now + bs->slice_time;
3215 bs->io_base.bytes[is_write] = bs->nr_bytes[is_write];
3216 bs->io_base.bytes[!is_write] = bs->nr_bytes[!is_write];
3218 bs->io_base.ios[is_write] = bs->nr_ops[is_write];
3219 bs->io_base.ios[!is_write] = bs->nr_ops[!is_write];
3222 elapsed_time = now - bs->slice_start;
3223 elapsed_time /= (NANOSECONDS_PER_SECOND);
3225 bps_ret = bdrv_exceed_bps_limits(bs, nb_sectors,
3226 is_write, elapsed_time, &bps_wait);
3227 iops_ret = bdrv_exceed_iops_limits(bs, is_write,
3228 elapsed_time, &iops_wait);
3229 if (bps_ret || iops_ret) {
3230 max_wait = bps_wait > iops_wait ? bps_wait : iops_wait;
3235 now = qemu_get_clock_ns(vm_clock);
3236 if (bs->slice_end < now + max_wait) {
3237 bs->slice_end = now + max_wait;
3250 /**************************************************************/
3251 /* async block device emulation */
3253 typedef struct BlockDriverAIOCBSync {
3254 BlockDriverAIOCB common;
3257 /* vector translation state */
3261 } BlockDriverAIOCBSync;
3263 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
3265 BlockDriverAIOCBSync *acb =
3266 container_of(blockacb, BlockDriverAIOCBSync, common);
3267 qemu_bh_delete(acb->bh);
3269 qemu_aio_release(acb);
3272 static AIOPool bdrv_em_aio_pool = {
3273 .aiocb_size = sizeof(BlockDriverAIOCBSync),
3274 .cancel = bdrv_aio_cancel_em,
3277 static void bdrv_aio_bh_cb(void *opaque)
3279 BlockDriverAIOCBSync *acb = opaque;
3282 qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
3283 qemu_vfree(acb->bounce);
3284 acb->common.cb(acb->common.opaque, acb->ret);
3285 qemu_bh_delete(acb->bh);
3287 qemu_aio_release(acb);
3290 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
3294 BlockDriverCompletionFunc *cb,
3299 BlockDriverAIOCBSync *acb;
3301 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
3302 acb->is_write = is_write;
3304 acb->bounce = qemu_blockalign(bs, qiov->size);
3305 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
3308 qemu_iovec_to_buffer(acb->qiov, acb->bounce);
3309 acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
3311 acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
3314 qemu_bh_schedule(acb->bh);
3316 return &acb->common;
3319 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
3320 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
3321 BlockDriverCompletionFunc *cb, void *opaque)
3323 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
3326 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
3327 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
3328 BlockDriverCompletionFunc *cb, void *opaque)
3330 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
3334 typedef struct BlockDriverAIOCBCoroutine {
3335 BlockDriverAIOCB common;
3339 } BlockDriverAIOCBCoroutine;
3341 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb)
3346 static AIOPool bdrv_em_co_aio_pool = {
3347 .aiocb_size = sizeof(BlockDriverAIOCBCoroutine),
3348 .cancel = bdrv_aio_co_cancel_em,
3351 static void bdrv_co_em_bh(void *opaque)
3353 BlockDriverAIOCBCoroutine *acb = opaque;
3355 acb->common.cb(acb->common.opaque, acb->req.error);
3356 qemu_bh_delete(acb->bh);
3357 qemu_aio_release(acb);
3360 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
3361 static void coroutine_fn bdrv_co_do_rw(void *opaque)
3363 BlockDriverAIOCBCoroutine *acb = opaque;
3364 BlockDriverState *bs = acb->common.bs;
3366 if (!acb->is_write) {
3367 acb->req.error = bdrv_co_do_readv(bs, acb->req.sector,
3368 acb->req.nb_sectors, acb->req.qiov, 0);
3370 acb->req.error = bdrv_co_do_writev(bs, acb->req.sector,
3371 acb->req.nb_sectors, acb->req.qiov, 0);
3374 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3375 qemu_bh_schedule(acb->bh);
3378 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
3382 BlockDriverCompletionFunc *cb,
3387 BlockDriverAIOCBCoroutine *acb;
3389 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3390 acb->req.sector = sector_num;
3391 acb->req.nb_sectors = nb_sectors;
3392 acb->req.qiov = qiov;
3393 acb->is_write = is_write;
3395 co = qemu_coroutine_create(bdrv_co_do_rw);
3396 qemu_coroutine_enter(co, acb);
3398 return &acb->common;
3401 static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
3403 BlockDriverAIOCBCoroutine *acb = opaque;
3404 BlockDriverState *bs = acb->common.bs;
3406 acb->req.error = bdrv_co_flush(bs);
3407 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3408 qemu_bh_schedule(acb->bh);
3411 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
3412 BlockDriverCompletionFunc *cb, void *opaque)
3414 trace_bdrv_aio_flush(bs, opaque);
3417 BlockDriverAIOCBCoroutine *acb;
3419 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3420 co = qemu_coroutine_create(bdrv_aio_flush_co_entry);
3421 qemu_coroutine_enter(co, acb);
3423 return &acb->common;
3426 static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque)
3428 BlockDriverAIOCBCoroutine *acb = opaque;
3429 BlockDriverState *bs = acb->common.bs;
3431 acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors);
3432 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3433 qemu_bh_schedule(acb->bh);
3436 BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs,
3437 int64_t sector_num, int nb_sectors,
3438 BlockDriverCompletionFunc *cb, void *opaque)
3441 BlockDriverAIOCBCoroutine *acb;
3443 trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque);
3445 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3446 acb->req.sector = sector_num;
3447 acb->req.nb_sectors = nb_sectors;
3448 co = qemu_coroutine_create(bdrv_aio_discard_co_entry);
3449 qemu_coroutine_enter(co, acb);
3451 return &acb->common;
3454 void bdrv_init(void)
3456 module_call_init(MODULE_INIT_BLOCK);
3459 void bdrv_init_with_whitelist(void)
3461 use_bdrv_whitelist = 1;
3465 void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
3466 BlockDriverCompletionFunc *cb, void *opaque)
3468 BlockDriverAIOCB *acb;
3470 if (pool->free_aiocb) {
3471 acb = pool->free_aiocb;
3472 pool->free_aiocb = acb->next;
3474 acb = g_malloc0(pool->aiocb_size);
3479 acb->opaque = opaque;
3483 void qemu_aio_release(void *p)
3485 BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
3486 AIOPool *pool = acb->pool;
3487 acb->next = pool->free_aiocb;
3488 pool->free_aiocb = acb;
3491 /**************************************************************/
3492 /* Coroutine block device emulation */
3494 typedef struct CoroutineIOCompletion {
3495 Coroutine *coroutine;
3497 } CoroutineIOCompletion;
3499 static void bdrv_co_io_em_complete(void *opaque, int ret)
3501 CoroutineIOCompletion *co = opaque;
3504 qemu_coroutine_enter(co->coroutine, NULL);
3507 static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num,
3508 int nb_sectors, QEMUIOVector *iov,
3511 CoroutineIOCompletion co = {
3512 .coroutine = qemu_coroutine_self(),
3514 BlockDriverAIOCB *acb;
3517 acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors,
3518 bdrv_co_io_em_complete, &co);
3520 acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors,
3521 bdrv_co_io_em_complete, &co);
3524 trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb);
3528 qemu_coroutine_yield();
3533 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
3534 int64_t sector_num, int nb_sectors,
3537 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false);
3540 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
3541 int64_t sector_num, int nb_sectors,
3544 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true);
3547 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
3549 RwCo *rwco = opaque;
3551 rwco->ret = bdrv_co_flush(rwco->bs);
3554 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
3558 if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
3562 /* Write back cached data to the OS even with cache=unsafe */
3563 if (bs->drv->bdrv_co_flush_to_os) {
3564 ret = bs->drv->bdrv_co_flush_to_os(bs);
3570 /* But don't actually force it to the disk with cache=unsafe */
3571 if (bs->open_flags & BDRV_O_NO_FLUSH) {
3575 if (bs->drv->bdrv_co_flush_to_disk) {
3576 ret = bs->drv->bdrv_co_flush_to_disk(bs);
3577 } else if (bs->drv->bdrv_aio_flush) {
3578 BlockDriverAIOCB *acb;
3579 CoroutineIOCompletion co = {
3580 .coroutine = qemu_coroutine_self(),
3583 acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
3587 qemu_coroutine_yield();
3592 * Some block drivers always operate in either writethrough or unsafe
3593 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
3594 * know how the server works (because the behaviour is hardcoded or
3595 * depends on server-side configuration), so we can't ensure that
3596 * everything is safe on disk. Returning an error doesn't work because
3597 * that would break guests even if the server operates in writethrough
3600 * Let's hope the user knows what he's doing.
3608 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
3609 * in the case of cache=unsafe, so there are no useless flushes.
3611 return bdrv_co_flush(bs->file);
3614 void bdrv_invalidate_cache(BlockDriverState *bs)
3616 if (bs->drv && bs->drv->bdrv_invalidate_cache) {
3617 bs->drv->bdrv_invalidate_cache(bs);
3621 void bdrv_invalidate_cache_all(void)
3623 BlockDriverState *bs;
3625 QTAILQ_FOREACH(bs, &bdrv_states, list) {
3626 bdrv_invalidate_cache(bs);
3630 void bdrv_clear_incoming_migration_all(void)
3632 BlockDriverState *bs;
3634 QTAILQ_FOREACH(bs, &bdrv_states, list) {
3635 bs->open_flags = bs->open_flags & ~(BDRV_O_INCOMING);
3639 int bdrv_flush(BlockDriverState *bs)
3647 if (qemu_in_coroutine()) {
3648 /* Fast-path if already in coroutine context */
3649 bdrv_flush_co_entry(&rwco);
3651 co = qemu_coroutine_create(bdrv_flush_co_entry);
3652 qemu_coroutine_enter(co, &rwco);
3653 while (rwco.ret == NOT_DONE) {
3661 static void coroutine_fn bdrv_discard_co_entry(void *opaque)
3663 RwCo *rwco = opaque;
3665 rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors);
3668 int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
3673 } else if (bdrv_check_request(bs, sector_num, nb_sectors)) {
3675 } else if (bs->read_only) {
3677 } else if (bs->drv->bdrv_co_discard) {
3678 return bs->drv->bdrv_co_discard(bs, sector_num, nb_sectors);
3679 } else if (bs->drv->bdrv_aio_discard) {
3680 BlockDriverAIOCB *acb;
3681 CoroutineIOCompletion co = {
3682 .coroutine = qemu_coroutine_self(),
3685 acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors,
3686 bdrv_co_io_em_complete, &co);
3690 qemu_coroutine_yield();
3698 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
3703 .sector_num = sector_num,
3704 .nb_sectors = nb_sectors,
3708 if (qemu_in_coroutine()) {
3709 /* Fast-path if already in coroutine context */
3710 bdrv_discard_co_entry(&rwco);
3712 co = qemu_coroutine_create(bdrv_discard_co_entry);
3713 qemu_coroutine_enter(co, &rwco);
3714 while (rwco.ret == NOT_DONE) {
3722 /**************************************************************/
3723 /* removable device support */
3726 * Return TRUE if the media is present
3728 int bdrv_is_inserted(BlockDriverState *bs)
3730 BlockDriver *drv = bs->drv;
3734 if (!drv->bdrv_is_inserted)
3736 return drv->bdrv_is_inserted(bs);
3740 * Return whether the media changed since the last call to this
3741 * function, or -ENOTSUP if we don't know. Most drivers don't know.
3743 int bdrv_media_changed(BlockDriverState *bs)
3745 BlockDriver *drv = bs->drv;
3747 if (drv && drv->bdrv_media_changed) {
3748 return drv->bdrv_media_changed(bs);
3754 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
3756 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
3758 BlockDriver *drv = bs->drv;
3760 if (drv && drv->bdrv_eject) {
3761 drv->bdrv_eject(bs, eject_flag);
3764 if (bs->device_name[0] != '\0') {
3765 bdrv_emit_qmp_eject_event(bs, eject_flag);
3770 * Lock or unlock the media (if it is locked, the user won't be able
3771 * to eject it manually).
3773 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
3775 BlockDriver *drv = bs->drv;
3777 trace_bdrv_lock_medium(bs, locked);
3779 if (drv && drv->bdrv_lock_medium) {
3780 drv->bdrv_lock_medium(bs, locked);
3784 /* needed for generic scsi interface */
3786 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
3788 BlockDriver *drv = bs->drv;
3790 if (drv && drv->bdrv_ioctl)
3791 return drv->bdrv_ioctl(bs, req, buf);
3795 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
3796 unsigned long int req, void *buf,
3797 BlockDriverCompletionFunc *cb, void *opaque)
3799 BlockDriver *drv = bs->drv;
3801 if (drv && drv->bdrv_aio_ioctl)
3802 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
3806 void bdrv_set_buffer_alignment(BlockDriverState *bs, int align)
3808 bs->buffer_alignment = align;
3811 void *qemu_blockalign(BlockDriverState *bs, size_t size)
3813 return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
3816 void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
3818 int64_t bitmap_size;
3820 bs->dirty_count = 0;
3822 if (!bs->dirty_bitmap) {
3823 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
3824 BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
3825 bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
3827 bs->dirty_bitmap = g_malloc0(bitmap_size);
3830 if (bs->dirty_bitmap) {
3831 g_free(bs->dirty_bitmap);
3832 bs->dirty_bitmap = NULL;
3837 int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
3839 int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
3841 if (bs->dirty_bitmap &&
3842 (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
3843 return !!(bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
3844 (1UL << (chunk % (sizeof(unsigned long) * 8))));
3850 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
3853 set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
3856 int64_t bdrv_get_dirty_count(BlockDriverState *bs)
3858 return bs->dirty_count;
3861 void bdrv_set_in_use(BlockDriverState *bs, int in_use)
3863 assert(bs->in_use != in_use);
3864 bs->in_use = in_use;
3867 int bdrv_in_use(BlockDriverState *bs)
3872 void bdrv_iostatus_enable(BlockDriverState *bs)
3874 bs->iostatus_enabled = true;
3875 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
3878 /* The I/O status is only enabled if the drive explicitly
3879 * enables it _and_ the VM is configured to stop on errors */
3880 bool bdrv_iostatus_is_enabled(const BlockDriverState *bs)
3882 return (bs->iostatus_enabled &&
3883 (bs->on_write_error == BLOCK_ERR_STOP_ENOSPC ||
3884 bs->on_write_error == BLOCK_ERR_STOP_ANY ||
3885 bs->on_read_error == BLOCK_ERR_STOP_ANY));
3888 void bdrv_iostatus_disable(BlockDriverState *bs)
3890 bs->iostatus_enabled = false;
3893 void bdrv_iostatus_reset(BlockDriverState *bs)
3895 if (bdrv_iostatus_is_enabled(bs)) {
3896 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
3900 /* XXX: Today this is set by device models because it makes the implementation
3901 quite simple. However, the block layer knows about the error, so it's
3902 possible to implement this without device models being involved */
3903 void bdrv_iostatus_set_err(BlockDriverState *bs, int error)
3905 if (bdrv_iostatus_is_enabled(bs) &&
3906 bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
3908 bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
3909 BLOCK_DEVICE_IO_STATUS_FAILED;
3914 bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes,
3915 enum BlockAcctType type)
3917 assert(type < BDRV_MAX_IOTYPE);
3919 cookie->bytes = bytes;
3920 cookie->start_time_ns = get_clock();
3921 cookie->type = type;
3925 bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie)
3927 assert(cookie->type < BDRV_MAX_IOTYPE);
3929 bs->nr_bytes[cookie->type] += cookie->bytes;
3930 bs->nr_ops[cookie->type]++;
3931 bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns;
3934 int bdrv_img_create(const char *filename, const char *fmt,
3935 const char *base_filename, const char *base_fmt,
3936 char *options, uint64_t img_size, int flags)
3938 QEMUOptionParameter *param = NULL, *create_options = NULL;
3939 QEMUOptionParameter *backing_fmt, *backing_file, *size;
3940 BlockDriverState *bs = NULL;
3941 BlockDriver *drv, *proto_drv;
3942 BlockDriver *backing_drv = NULL;
3945 /* Find driver and parse its options */
3946 drv = bdrv_find_format(fmt);
3948 error_report("Unknown file format '%s'", fmt);
3953 proto_drv = bdrv_find_protocol(filename);
3955 error_report("Unknown protocol '%s'", filename);
3960 create_options = append_option_parameters(create_options,
3961 drv->create_options);
3962 create_options = append_option_parameters(create_options,
3963 proto_drv->create_options);
3965 /* Create parameter list with default values */
3966 param = parse_option_parameters("", create_options, param);
3968 set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size);
3970 /* Parse -o options */
3972 param = parse_option_parameters(options, create_options, param);
3973 if (param == NULL) {
3974 error_report("Invalid options for file format '%s'.", fmt);
3980 if (base_filename) {
3981 if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE,
3983 error_report("Backing file not supported for file format '%s'",
3991 if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) {
3992 error_report("Backing file format not supported for file "
3993 "format '%s'", fmt);
3999 backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE);
4000 if (backing_file && backing_file->value.s) {
4001 if (!strcmp(filename, backing_file->value.s)) {
4002 error_report("Error: Trying to create an image with the "
4003 "same filename as the backing file");
4009 backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT);
4010 if (backing_fmt && backing_fmt->value.s) {
4011 backing_drv = bdrv_find_format(backing_fmt->value.s);
4013 error_report("Unknown backing file format '%s'",
4014 backing_fmt->value.s);
4020 // The size for the image must always be specified, with one exception:
4021 // If we are using a backing file, we can obtain the size from there
4022 size = get_option_parameter(param, BLOCK_OPT_SIZE);
4023 if (size && size->value.n == -1) {
4024 if (backing_file && backing_file->value.s) {
4030 ret = bdrv_open(bs, backing_file->value.s, flags, backing_drv);
4032 error_report("Could not open '%s'", backing_file->value.s);
4035 bdrv_get_geometry(bs, &size);
4038 snprintf(buf, sizeof(buf), "%" PRId64, size);
4039 set_option_parameter(param, BLOCK_OPT_SIZE, buf);
4041 error_report("Image creation needs a size parameter");
4047 printf("Formatting '%s', fmt=%s ", filename, fmt);
4048 print_option_parameters(param);
4051 ret = bdrv_create(drv, filename, param);
4054 if (ret == -ENOTSUP) {
4055 error_report("Formatting or formatting option not supported for "
4056 "file format '%s'", fmt);
4057 } else if (ret == -EFBIG) {
4058 error_report("The image size is too large for file format '%s'",
4061 error_report("%s: error while creating %s: %s", filename, fmt,
4067 free_option_parameters(create_options);
4068 free_option_parameters(param);
4077 void *block_job_create(const BlockJobType *job_type, BlockDriverState *bs,
4078 BlockDriverCompletionFunc *cb, void *opaque)
4082 if (bs->job || bdrv_in_use(bs)) {
4085 bdrv_set_in_use(bs, 1);
4087 job = g_malloc0(job_type->instance_size);
4088 job->job_type = job_type;
4091 job->opaque = opaque;
4096 void block_job_complete(BlockJob *job, int ret)
4098 BlockDriverState *bs = job->bs;
4100 assert(bs->job == job);
4101 job->cb(job->opaque, ret);
4104 bdrv_set_in_use(bs, 0);
4107 int block_job_set_speed(BlockJob *job, int64_t value)
4111 if (!job->job_type->set_speed) {
4114 rc = job->job_type->set_speed(job, value);
4121 void block_job_cancel(BlockJob *job)
4123 job->cancelled = true;
4126 bool block_job_is_cancelled(BlockJob *job)
4128 return job->cancelled;
4131 void block_job_cancel_sync(BlockJob *job)
4133 BlockDriverState *bs = job->bs;
4135 assert(bs->job == job);
4136 block_job_cancel(job);
4137 while (bs->job != NULL && bs->job->busy) {