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)
204 if (is_windows_drive(path) ||
205 is_windows_drive_prefix(path)) {
208 p = path + strcspn(path, ":/\\");
210 p = path + strcspn(path, ":/");
216 int path_is_absolute(const char *path)
219 /* specific case for names like: "\\.\d:" */
220 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
223 return (*path == '/' || *path == '\\');
225 return (*path == '/');
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_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz)
275 if (bs->backing_file[0] == '\0' || path_has_protocol(bs->backing_file)) {
276 pstrcpy(dest, sz, bs->backing_file);
278 path_combine(dest, sz, bs->filename, bs->backing_file);
282 void bdrv_register(BlockDriver *bdrv)
284 /* Block drivers without coroutine functions need emulation */
285 if (!bdrv->bdrv_co_readv) {
286 bdrv->bdrv_co_readv = bdrv_co_readv_em;
287 bdrv->bdrv_co_writev = bdrv_co_writev_em;
289 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
290 * the block driver lacks aio we need to emulate that too.
292 if (!bdrv->bdrv_aio_readv) {
293 /* add AIO emulation layer */
294 bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
295 bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
299 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
302 /* create a new block device (by default it is empty) */
303 BlockDriverState *bdrv_new(const char *device_name)
305 BlockDriverState *bs;
307 bs = g_malloc0(sizeof(BlockDriverState));
308 pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
309 if (device_name[0] != '\0') {
310 QTAILQ_INSERT_TAIL(&bdrv_states, bs, list);
312 bdrv_iostatus_disable(bs);
316 BlockDriver *bdrv_find_format(const char *format_name)
319 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
320 if (!strcmp(drv1->format_name, format_name)) {
327 static int bdrv_is_whitelisted(BlockDriver *drv)
329 static const char *whitelist[] = {
330 CONFIG_BDRV_WHITELIST
335 return 1; /* no whitelist, anything goes */
337 for (p = whitelist; *p; p++) {
338 if (!strcmp(drv->format_name, *p)) {
345 BlockDriver *bdrv_find_whitelisted_format(const char *format_name)
347 BlockDriver *drv = bdrv_find_format(format_name);
348 return drv && bdrv_is_whitelisted(drv) ? drv : NULL;
351 typedef struct CreateCo {
354 QEMUOptionParameter *options;
358 static void coroutine_fn bdrv_create_co_entry(void *opaque)
360 CreateCo *cco = opaque;
363 cco->ret = cco->drv->bdrv_create(cco->filename, cco->options);
366 int bdrv_create(BlockDriver *drv, const char* filename,
367 QEMUOptionParameter *options)
374 .filename = g_strdup(filename),
379 if (!drv->bdrv_create) {
383 if (qemu_in_coroutine()) {
384 /* Fast-path if already in coroutine context */
385 bdrv_create_co_entry(&cco);
387 co = qemu_coroutine_create(bdrv_create_co_entry);
388 qemu_coroutine_enter(co, &cco);
389 while (cco.ret == NOT_DONE) {
395 g_free(cco.filename);
400 int bdrv_create_file(const char* filename, QEMUOptionParameter *options)
404 drv = bdrv_find_protocol(filename);
409 return bdrv_create(drv, filename, options);
413 * Create a uniquely-named empty temporary file.
414 * Return 0 upon success, otherwise a negative errno value.
416 int get_tmp_filename(char *filename, int size)
419 char temp_dir[MAX_PATH];
420 /* GetTempFileName requires that its output buffer (4th param)
421 have length MAX_PATH or greater. */
422 assert(size >= MAX_PATH);
423 return (GetTempPath(MAX_PATH, temp_dir)
424 && GetTempFileName(temp_dir, "qem", 0, filename)
425 ? 0 : -GetLastError());
429 tmpdir = getenv("TMPDIR");
432 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
435 fd = mkstemp(filename);
436 if (fd < 0 || close(fd)) {
444 * Detect host devices. By convention, /dev/cdrom[N] is always
445 * recognized as a host CDROM.
447 static BlockDriver *find_hdev_driver(const char *filename)
449 int score_max = 0, score;
450 BlockDriver *drv = NULL, *d;
452 QLIST_FOREACH(d, &bdrv_drivers, list) {
453 if (d->bdrv_probe_device) {
454 score = d->bdrv_probe_device(filename);
455 if (score > score_max) {
465 BlockDriver *bdrv_find_protocol(const char *filename)
472 /* TODO Drivers without bdrv_file_open must be specified explicitly */
475 * XXX(hch): we really should not let host device detection
476 * override an explicit protocol specification, but moving this
477 * later breaks access to device names with colons in them.
478 * Thanks to the brain-dead persistent naming schemes on udev-
479 * based Linux systems those actually are quite common.
481 drv1 = find_hdev_driver(filename);
486 if (!path_has_protocol(filename)) {
487 return bdrv_find_format("file");
489 p = strchr(filename, ':');
492 if (len > sizeof(protocol) - 1)
493 len = sizeof(protocol) - 1;
494 memcpy(protocol, filename, len);
495 protocol[len] = '\0';
496 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
497 if (drv1->protocol_name &&
498 !strcmp(drv1->protocol_name, protocol)) {
505 static int find_image_format(const char *filename, BlockDriver **pdrv)
507 int ret, score, score_max;
508 BlockDriver *drv1, *drv;
510 BlockDriverState *bs;
512 ret = bdrv_file_open(&bs, filename, 0);
518 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
519 if (bs->sg || !bdrv_is_inserted(bs)) {
521 drv = bdrv_find_format("raw");
529 ret = bdrv_pread(bs, 0, buf, sizeof(buf));
538 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
539 if (drv1->bdrv_probe) {
540 score = drv1->bdrv_probe(buf, ret, filename);
541 if (score > score_max) {
555 * Set the current 'total_sectors' value
557 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
559 BlockDriver *drv = bs->drv;
561 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
565 /* query actual device if possible, otherwise just trust the hint */
566 if (drv->bdrv_getlength) {
567 int64_t length = drv->bdrv_getlength(bs);
571 hint = length >> BDRV_SECTOR_BITS;
574 bs->total_sectors = hint;
579 * Set open flags for a given cache mode
581 * Return 0 on success, -1 if the cache mode was invalid.
583 int bdrv_parse_cache_flags(const char *mode, int *flags)
585 *flags &= ~BDRV_O_CACHE_MASK;
587 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
588 *flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB;
589 } else if (!strcmp(mode, "directsync")) {
590 *flags |= BDRV_O_NOCACHE;
591 } else if (!strcmp(mode, "writeback")) {
592 *flags |= BDRV_O_CACHE_WB;
593 } else if (!strcmp(mode, "unsafe")) {
594 *flags |= BDRV_O_CACHE_WB;
595 *flags |= BDRV_O_NO_FLUSH;
596 } else if (!strcmp(mode, "writethrough")) {
597 /* this is the default */
606 * The copy-on-read flag is actually a reference count so multiple users may
607 * use the feature without worrying about clobbering its previous state.
608 * Copy-on-read stays enabled until all users have called to disable it.
610 void bdrv_enable_copy_on_read(BlockDriverState *bs)
615 void bdrv_disable_copy_on_read(BlockDriverState *bs)
617 assert(bs->copy_on_read > 0);
622 * Common part for opening disk images and files
624 static int bdrv_open_common(BlockDriverState *bs, const char *filename,
625 int flags, BlockDriver *drv)
630 assert(bs->file == NULL);
632 trace_bdrv_open_common(bs, filename, flags, drv->format_name);
634 bs->open_flags = flags;
635 bs->buffer_alignment = 512;
637 assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
638 if ((flags & BDRV_O_RDWR) && (flags & BDRV_O_COPY_ON_READ)) {
639 bdrv_enable_copy_on_read(bs);
642 pstrcpy(bs->filename, sizeof(bs->filename), filename);
644 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv)) {
649 bs->opaque = g_malloc0(drv->instance_size);
651 bs->enable_write_cache = !!(flags & BDRV_O_CACHE_WB);
652 open_flags = flags | BDRV_O_CACHE_WB;
655 * Clear flags that are internal to the block layer before opening the
658 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
661 * Snapshots should be writable.
663 if (bs->is_temporary) {
664 open_flags |= BDRV_O_RDWR;
667 bs->keep_read_only = bs->read_only = !(open_flags & BDRV_O_RDWR);
669 /* Open the image, either directly or using a protocol */
670 if (drv->bdrv_file_open) {
671 ret = drv->bdrv_file_open(bs, filename, open_flags);
673 ret = bdrv_file_open(&bs->file, filename, open_flags);
675 ret = drv->bdrv_open(bs, open_flags);
683 ret = refresh_total_sectors(bs, bs->total_sectors);
689 if (bs->is_temporary) {
697 bdrv_delete(bs->file);
707 * Opens a file using a protocol (file, host_device, nbd, ...)
709 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
711 BlockDriverState *bs;
715 drv = bdrv_find_protocol(filename);
721 ret = bdrv_open_common(bs, filename, flags, drv);
732 * Opens a disk image (raw, qcow2, vmdk, ...)
734 int bdrv_open(BlockDriverState *bs, const char *filename, int flags,
738 char tmp_filename[PATH_MAX];
740 if (flags & BDRV_O_SNAPSHOT) {
741 BlockDriverState *bs1;
744 BlockDriver *bdrv_qcow2;
745 QEMUOptionParameter *options;
746 char backing_filename[PATH_MAX];
748 /* if snapshot, we create a temporary backing file and open it
749 instead of opening 'filename' directly */
751 /* if there is a backing file, use it */
753 ret = bdrv_open(bs1, filename, 0, drv);
758 total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
760 if (bs1->drv && bs1->drv->protocol_name)
765 ret = get_tmp_filename(tmp_filename, sizeof(tmp_filename));
770 /* Real path is meaningless for protocols */
772 snprintf(backing_filename, sizeof(backing_filename),
774 else if (!realpath(filename, backing_filename))
777 bdrv_qcow2 = bdrv_find_format("qcow2");
778 options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
780 set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size);
781 set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
783 set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
787 ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
788 free_option_parameters(options);
793 filename = tmp_filename;
795 bs->is_temporary = 1;
798 /* Find the right image format driver */
800 ret = find_image_format(filename, &drv);
804 goto unlink_and_fail;
808 ret = bdrv_open_common(bs, filename, flags, drv);
810 goto unlink_and_fail;
813 /* If there is a backing file, use it */
814 if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') {
815 char backing_filename[PATH_MAX];
817 BlockDriver *back_drv = NULL;
819 bs->backing_hd = bdrv_new("");
820 bdrv_get_full_backing_filename(bs, backing_filename,
821 sizeof(backing_filename));
823 if (bs->backing_format[0] != '\0') {
824 back_drv = bdrv_find_format(bs->backing_format);
827 /* backing files always opened read-only */
829 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
831 ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv);
836 if (bs->is_temporary) {
837 bs->backing_hd->keep_read_only = !(flags & BDRV_O_RDWR);
839 /* base image inherits from "parent" */
840 bs->backing_hd->keep_read_only = bs->keep_read_only;
844 if (!bdrv_key_required(bs)) {
845 bdrv_dev_change_media_cb(bs, true);
848 /* throttling disk I/O limits */
849 if (bs->io_limits_enabled) {
850 bdrv_io_limits_enable(bs);
856 if (bs->is_temporary) {
862 void bdrv_close(BlockDriverState *bs)
867 block_job_cancel_sync(bs->job);
871 if (bs == bs_snapshots) {
874 if (bs->backing_hd) {
875 bdrv_delete(bs->backing_hd);
876 bs->backing_hd = NULL;
878 bs->drv->bdrv_close(bs);
881 if (bs->is_temporary) {
882 unlink(bs->filename);
887 bs->copy_on_read = 0;
888 bs->backing_file[0] = '\0';
889 bs->backing_format[0] = '\0';
890 bs->total_sectors = 0;
896 if (bs->file != NULL) {
897 bdrv_delete(bs->file);
901 bdrv_dev_change_media_cb(bs, false);
904 /*throttling disk I/O limits*/
905 if (bs->io_limits_enabled) {
906 bdrv_io_limits_disable(bs);
910 void bdrv_close_all(void)
912 BlockDriverState *bs;
914 QTAILQ_FOREACH(bs, &bdrv_states, list) {
920 * Wait for pending requests to complete across all BlockDriverStates
922 * This function does not flush data to disk, use bdrv_flush_all() for that
923 * after calling this function.
925 * Note that completion of an asynchronous I/O operation can trigger any
926 * number of other I/O operations on other devices---for example a coroutine
927 * can be arbitrarily complex and a constant flow of I/O can come until the
928 * coroutine is complete. Because of this, it is not possible to have a
929 * function to drain a single device's I/O queue.
931 void bdrv_drain_all(void)
933 BlockDriverState *bs;
937 busy = qemu_aio_wait();
939 /* FIXME: We do not have timer support here, so this is effectively
942 QTAILQ_FOREACH(bs, &bdrv_states, list) {
943 if (!qemu_co_queue_empty(&bs->throttled_reqs)) {
944 qemu_co_queue_restart_all(&bs->throttled_reqs);
950 /* If requests are still pending there is a bug somewhere */
951 QTAILQ_FOREACH(bs, &bdrv_states, list) {
952 assert(QLIST_EMPTY(&bs->tracked_requests));
953 assert(qemu_co_queue_empty(&bs->throttled_reqs));
957 /* make a BlockDriverState anonymous by removing from bdrv_state list.
958 Also, NULL terminate the device_name to prevent double remove */
959 void bdrv_make_anon(BlockDriverState *bs)
961 if (bs->device_name[0] != '\0') {
962 QTAILQ_REMOVE(&bdrv_states, bs, list);
964 bs->device_name[0] = '\0';
967 static void bdrv_rebind(BlockDriverState *bs)
969 if (bs->drv && bs->drv->bdrv_rebind) {
970 bs->drv->bdrv_rebind(bs);
974 static void bdrv_move_feature_fields(BlockDriverState *bs_dest,
975 BlockDriverState *bs_src)
977 /* move some fields that need to stay attached to the device */
978 bs_dest->open_flags = bs_src->open_flags;
981 bs_dest->dev_ops = bs_src->dev_ops;
982 bs_dest->dev_opaque = bs_src->dev_opaque;
983 bs_dest->dev = bs_src->dev;
984 bs_dest->buffer_alignment = bs_src->buffer_alignment;
985 bs_dest->copy_on_read = bs_src->copy_on_read;
987 bs_dest->enable_write_cache = bs_src->enable_write_cache;
989 /* i/o timing parameters */
990 bs_dest->slice_time = bs_src->slice_time;
991 bs_dest->slice_start = bs_src->slice_start;
992 bs_dest->slice_end = bs_src->slice_end;
993 bs_dest->io_limits = bs_src->io_limits;
994 bs_dest->io_base = bs_src->io_base;
995 bs_dest->throttled_reqs = bs_src->throttled_reqs;
996 bs_dest->block_timer = bs_src->block_timer;
997 bs_dest->io_limits_enabled = bs_src->io_limits_enabled;
1000 bs_dest->cyls = bs_src->cyls;
1001 bs_dest->heads = bs_src->heads;
1002 bs_dest->secs = bs_src->secs;
1003 bs_dest->translation = bs_src->translation;
1006 bs_dest->on_read_error = bs_src->on_read_error;
1007 bs_dest->on_write_error = bs_src->on_write_error;
1010 bs_dest->iostatus_enabled = bs_src->iostatus_enabled;
1011 bs_dest->iostatus = bs_src->iostatus;
1014 bs_dest->dirty_count = bs_src->dirty_count;
1015 bs_dest->dirty_bitmap = bs_src->dirty_bitmap;
1018 bs_dest->in_use = bs_src->in_use;
1019 bs_dest->job = bs_src->job;
1021 /* keep the same entry in bdrv_states */
1022 pstrcpy(bs_dest->device_name, sizeof(bs_dest->device_name),
1023 bs_src->device_name);
1024 bs_dest->list = bs_src->list;
1028 * Swap bs contents for two image chains while they are live,
1029 * while keeping required fields on the BlockDriverState that is
1030 * actually attached to a device.
1032 * This will modify the BlockDriverState fields, and swap contents
1033 * between bs_new and bs_old. Both bs_new and bs_old are modified.
1035 * bs_new is required to be anonymous.
1037 * This function does not create any image files.
1039 void bdrv_swap(BlockDriverState *bs_new, BlockDriverState *bs_old)
1041 BlockDriverState tmp;
1043 /* bs_new must be anonymous and shouldn't have anything fancy enabled */
1044 assert(bs_new->device_name[0] == '\0');
1045 assert(bs_new->dirty_bitmap == NULL);
1046 assert(bs_new->job == NULL);
1047 assert(bs_new->dev == NULL);
1048 assert(bs_new->in_use == 0);
1049 assert(bs_new->io_limits_enabled == false);
1050 assert(bs_new->block_timer == NULL);
1056 /* there are some fields that should not be swapped, move them back */
1057 bdrv_move_feature_fields(&tmp, bs_old);
1058 bdrv_move_feature_fields(bs_old, bs_new);
1059 bdrv_move_feature_fields(bs_new, &tmp);
1061 /* bs_new shouldn't be in bdrv_states even after the swap! */
1062 assert(bs_new->device_name[0] == '\0');
1064 /* Check a few fields that should remain attached to the device */
1065 assert(bs_new->dev == NULL);
1066 assert(bs_new->job == NULL);
1067 assert(bs_new->in_use == 0);
1068 assert(bs_new->io_limits_enabled == false);
1069 assert(bs_new->block_timer == NULL);
1071 bdrv_rebind(bs_new);
1072 bdrv_rebind(bs_old);
1076 * Add new bs contents at the top of an image chain while the chain is
1077 * live, while keeping required fields on the top layer.
1079 * This will modify the BlockDriverState fields, and swap contents
1080 * between bs_new and bs_top. Both bs_new and bs_top are modified.
1082 * bs_new is required to be anonymous.
1084 * This function does not create any image files.
1086 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
1088 bdrv_swap(bs_new, bs_top);
1090 /* The contents of 'tmp' will become bs_top, as we are
1091 * swapping bs_new and bs_top contents. */
1092 bs_top->backing_hd = bs_new;
1093 bs_top->open_flags &= ~BDRV_O_NO_BACKING;
1094 pstrcpy(bs_top->backing_file, sizeof(bs_top->backing_file),
1096 pstrcpy(bs_top->backing_format, sizeof(bs_top->backing_format),
1097 bs_new->drv ? bs_new->drv->format_name : "");
1100 void bdrv_delete(BlockDriverState *bs)
1104 assert(!bs->in_use);
1106 /* remove from list, if necessary */
1111 assert(bs != bs_snapshots);
1115 int bdrv_attach_dev(BlockDriverState *bs, void *dev)
1116 /* TODO change to DeviceState *dev when all users are qdevified */
1122 bdrv_iostatus_reset(bs);
1126 /* TODO qdevified devices don't use this, remove when devices are qdevified */
1127 void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev)
1129 if (bdrv_attach_dev(bs, dev) < 0) {
1134 void bdrv_detach_dev(BlockDriverState *bs, void *dev)
1135 /* TODO change to DeviceState *dev when all users are qdevified */
1137 assert(bs->dev == dev);
1140 bs->dev_opaque = NULL;
1141 bs->buffer_alignment = 512;
1144 /* TODO change to return DeviceState * when all users are qdevified */
1145 void *bdrv_get_attached_dev(BlockDriverState *bs)
1150 void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops,
1154 bs->dev_opaque = opaque;
1155 if (bdrv_dev_has_removable_media(bs) && bs == bs_snapshots) {
1156 bs_snapshots = NULL;
1160 void bdrv_emit_qmp_error_event(const BlockDriverState *bdrv,
1161 BlockQMPEventAction action, int is_read)
1164 const char *action_str;
1167 case BDRV_ACTION_REPORT:
1168 action_str = "report";
1170 case BDRV_ACTION_IGNORE:
1171 action_str = "ignore";
1173 case BDRV_ACTION_STOP:
1174 action_str = "stop";
1180 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1183 is_read ? "read" : "write");
1184 monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
1186 qobject_decref(data);
1189 static void bdrv_emit_qmp_eject_event(BlockDriverState *bs, bool ejected)
1193 data = qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }",
1194 bdrv_get_device_name(bs), ejected);
1195 monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED, data);
1197 qobject_decref(data);
1200 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load)
1202 if (bs->dev_ops && bs->dev_ops->change_media_cb) {
1203 bool tray_was_closed = !bdrv_dev_is_tray_open(bs);
1204 bs->dev_ops->change_media_cb(bs->dev_opaque, load);
1205 if (tray_was_closed) {
1207 bdrv_emit_qmp_eject_event(bs, true);
1211 bdrv_emit_qmp_eject_event(bs, false);
1216 bool bdrv_dev_has_removable_media(BlockDriverState *bs)
1218 return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb);
1221 void bdrv_dev_eject_request(BlockDriverState *bs, bool force)
1223 if (bs->dev_ops && bs->dev_ops->eject_request_cb) {
1224 bs->dev_ops->eject_request_cb(bs->dev_opaque, force);
1228 bool bdrv_dev_is_tray_open(BlockDriverState *bs)
1230 if (bs->dev_ops && bs->dev_ops->is_tray_open) {
1231 return bs->dev_ops->is_tray_open(bs->dev_opaque);
1236 static void bdrv_dev_resize_cb(BlockDriverState *bs)
1238 if (bs->dev_ops && bs->dev_ops->resize_cb) {
1239 bs->dev_ops->resize_cb(bs->dev_opaque);
1243 bool bdrv_dev_is_medium_locked(BlockDriverState *bs)
1245 if (bs->dev_ops && bs->dev_ops->is_medium_locked) {
1246 return bs->dev_ops->is_medium_locked(bs->dev_opaque);
1252 * Run consistency checks on an image
1254 * Returns 0 if the check could be completed (it doesn't mean that the image is
1255 * free of errors) or -errno when an internal error occurred. The results of the
1256 * check are stored in res.
1258 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix)
1260 if (bs->drv->bdrv_check == NULL) {
1264 memset(res, 0, sizeof(*res));
1265 return bs->drv->bdrv_check(bs, res, fix);
1268 #define COMMIT_BUF_SECTORS 2048
1270 /* commit COW file into the raw image */
1271 int bdrv_commit(BlockDriverState *bs)
1273 BlockDriver *drv = bs->drv;
1274 BlockDriver *backing_drv;
1275 int64_t sector, total_sectors;
1276 int n, ro, open_flags;
1277 int ret = 0, rw_ret = 0;
1279 char filename[1024];
1280 BlockDriverState *bs_rw, *bs_ro;
1285 if (!bs->backing_hd) {
1289 if (bs->backing_hd->keep_read_only) {
1293 if (bdrv_in_use(bs) || bdrv_in_use(bs->backing_hd)) {
1297 backing_drv = bs->backing_hd->drv;
1298 ro = bs->backing_hd->read_only;
1299 strncpy(filename, bs->backing_hd->filename, sizeof(filename));
1300 open_flags = bs->backing_hd->open_flags;
1304 bdrv_delete(bs->backing_hd);
1305 bs->backing_hd = NULL;
1306 bs_rw = bdrv_new("");
1307 rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR,
1311 /* try to re-open read-only */
1312 bs_ro = bdrv_new("");
1313 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
1317 /* drive not functional anymore */
1321 bs->backing_hd = bs_ro;
1324 bs->backing_hd = bs_rw;
1327 total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
1328 buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
1330 for (sector = 0; sector < total_sectors; sector += n) {
1331 if (bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n)) {
1333 if (bdrv_read(bs, sector, buf, n) != 0) {
1338 if (bdrv_write(bs->backing_hd, sector, buf, n) != 0) {
1345 if (drv->bdrv_make_empty) {
1346 ret = drv->bdrv_make_empty(bs);
1351 * Make sure all data we wrote to the backing device is actually
1355 bdrv_flush(bs->backing_hd);
1362 bdrv_delete(bs->backing_hd);
1363 bs->backing_hd = NULL;
1364 bs_ro = bdrv_new("");
1365 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
1369 /* drive not functional anymore */
1373 bs->backing_hd = bs_ro;
1374 bs->backing_hd->keep_read_only = 0;
1380 int bdrv_commit_all(void)
1382 BlockDriverState *bs;
1384 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1385 int ret = bdrv_commit(bs);
1393 struct BdrvTrackedRequest {
1394 BlockDriverState *bs;
1398 QLIST_ENTRY(BdrvTrackedRequest) list;
1399 Coroutine *co; /* owner, used for deadlock detection */
1400 CoQueue wait_queue; /* coroutines blocked on this request */
1404 * Remove an active request from the tracked requests list
1406 * This function should be called when a tracked request is completing.
1408 static void tracked_request_end(BdrvTrackedRequest *req)
1410 QLIST_REMOVE(req, list);
1411 qemu_co_queue_restart_all(&req->wait_queue);
1415 * Add an active request to the tracked requests list
1417 static void tracked_request_begin(BdrvTrackedRequest *req,
1418 BlockDriverState *bs,
1420 int nb_sectors, bool is_write)
1422 *req = (BdrvTrackedRequest){
1424 .sector_num = sector_num,
1425 .nb_sectors = nb_sectors,
1426 .is_write = is_write,
1427 .co = qemu_coroutine_self(),
1430 qemu_co_queue_init(&req->wait_queue);
1432 QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
1436 * Round a region to cluster boundaries
1438 static void round_to_clusters(BlockDriverState *bs,
1439 int64_t sector_num, int nb_sectors,
1440 int64_t *cluster_sector_num,
1441 int *cluster_nb_sectors)
1443 BlockDriverInfo bdi;
1445 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
1446 *cluster_sector_num = sector_num;
1447 *cluster_nb_sectors = nb_sectors;
1449 int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE;
1450 *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c);
1451 *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num +
1456 static bool tracked_request_overlaps(BdrvTrackedRequest *req,
1457 int64_t sector_num, int nb_sectors) {
1459 if (sector_num >= req->sector_num + req->nb_sectors) {
1463 if (req->sector_num >= sector_num + nb_sectors) {
1469 static void coroutine_fn wait_for_overlapping_requests(BlockDriverState *bs,
1470 int64_t sector_num, int nb_sectors)
1472 BdrvTrackedRequest *req;
1473 int64_t cluster_sector_num;
1474 int cluster_nb_sectors;
1477 /* If we touch the same cluster it counts as an overlap. This guarantees
1478 * that allocating writes will be serialized and not race with each other
1479 * for the same cluster. For example, in copy-on-read it ensures that the
1480 * CoR read and write operations are atomic and guest writes cannot
1481 * interleave between them.
1483 round_to_clusters(bs, sector_num, nb_sectors,
1484 &cluster_sector_num, &cluster_nb_sectors);
1488 QLIST_FOREACH(req, &bs->tracked_requests, list) {
1489 if (tracked_request_overlaps(req, cluster_sector_num,
1490 cluster_nb_sectors)) {
1491 /* Hitting this means there was a reentrant request, for
1492 * example, a block driver issuing nested requests. This must
1493 * never happen since it means deadlock.
1495 assert(qemu_coroutine_self() != req->co);
1497 qemu_co_queue_wait(&req->wait_queue);
1508 * -EINVAL - backing format specified, but no file
1509 * -ENOSPC - can't update the backing file because no space is left in the
1511 * -ENOTSUP - format driver doesn't support changing the backing file
1513 int bdrv_change_backing_file(BlockDriverState *bs,
1514 const char *backing_file, const char *backing_fmt)
1516 BlockDriver *drv = bs->drv;
1519 /* Backing file format doesn't make sense without a backing file */
1520 if (backing_fmt && !backing_file) {
1524 if (drv->bdrv_change_backing_file != NULL) {
1525 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
1531 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
1532 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
1537 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
1542 if (!bdrv_is_inserted(bs))
1548 len = bdrv_getlength(bs);
1553 if ((offset > len) || (len - offset < size))
1559 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
1562 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
1563 nb_sectors * BDRV_SECTOR_SIZE);
1566 typedef struct RwCo {
1567 BlockDriverState *bs;
1575 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
1577 RwCo *rwco = opaque;
1579 if (!rwco->is_write) {
1580 rwco->ret = bdrv_co_do_readv(rwco->bs, rwco->sector_num,
1581 rwco->nb_sectors, rwco->qiov, 0);
1583 rwco->ret = bdrv_co_do_writev(rwco->bs, rwco->sector_num,
1584 rwco->nb_sectors, rwco->qiov, 0);
1589 * Process a synchronous request using coroutines
1591 static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf,
1592 int nb_sectors, bool is_write)
1595 struct iovec iov = {
1596 .iov_base = (void *)buf,
1597 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
1602 .sector_num = sector_num,
1603 .nb_sectors = nb_sectors,
1605 .is_write = is_write,
1609 qemu_iovec_init_external(&qiov, &iov, 1);
1612 * In sync call context, when the vcpu is blocked, this throttling timer
1613 * will not fire; so the I/O throttling function has to be disabled here
1614 * if it has been enabled.
1616 if (bs->io_limits_enabled) {
1617 fprintf(stderr, "Disabling I/O throttling on '%s' due "
1618 "to synchronous I/O.\n", bdrv_get_device_name(bs));
1619 bdrv_io_limits_disable(bs);
1622 if (qemu_in_coroutine()) {
1623 /* Fast-path if already in coroutine context */
1624 bdrv_rw_co_entry(&rwco);
1626 co = qemu_coroutine_create(bdrv_rw_co_entry);
1627 qemu_coroutine_enter(co, &rwco);
1628 while (rwco.ret == NOT_DONE) {
1635 /* return < 0 if error. See bdrv_write() for the return codes */
1636 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
1637 uint8_t *buf, int nb_sectors)
1639 return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false);
1642 /* Just like bdrv_read(), but with I/O throttling temporarily disabled */
1643 int bdrv_read_unthrottled(BlockDriverState *bs, int64_t sector_num,
1644 uint8_t *buf, int nb_sectors)
1649 enabled = bs->io_limits_enabled;
1650 bs->io_limits_enabled = false;
1651 ret = bdrv_read(bs, 0, buf, 1);
1652 bs->io_limits_enabled = enabled;
1656 #define BITS_PER_LONG (sizeof(unsigned long) * 8)
1658 static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
1659 int nb_sectors, int dirty)
1662 unsigned long val, idx, bit;
1664 start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
1665 end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
1667 for (; start <= end; start++) {
1668 idx = start / BITS_PER_LONG;
1669 bit = start % BITS_PER_LONG;
1670 val = bs->dirty_bitmap[idx];
1672 if (!(val & (1UL << bit))) {
1677 if (val & (1UL << bit)) {
1679 val &= ~(1UL << bit);
1682 bs->dirty_bitmap[idx] = val;
1686 /* Return < 0 if error. Important errors are:
1687 -EIO generic I/O error (may happen for all errors)
1688 -ENOMEDIUM No media inserted.
1689 -EINVAL Invalid sector number or nb_sectors
1690 -EACCES Trying to write a read-only device
1692 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
1693 const uint8_t *buf, int nb_sectors)
1695 return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true);
1698 int bdrv_pread(BlockDriverState *bs, int64_t offset,
1699 void *buf, int count1)
1701 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1702 int len, nb_sectors, count;
1707 /* first read to align to sector start */
1708 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1711 sector_num = offset >> BDRV_SECTOR_BITS;
1713 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1715 memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
1723 /* read the sectors "in place" */
1724 nb_sectors = count >> BDRV_SECTOR_BITS;
1725 if (nb_sectors > 0) {
1726 if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
1728 sector_num += nb_sectors;
1729 len = nb_sectors << BDRV_SECTOR_BITS;
1734 /* add data from the last sector */
1736 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1738 memcpy(buf, tmp_buf, count);
1743 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
1744 const void *buf, int count1)
1746 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1747 int len, nb_sectors, count;
1752 /* first write to align to sector start */
1753 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1756 sector_num = offset >> BDRV_SECTOR_BITS;
1758 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1760 memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
1761 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1770 /* write the sectors "in place" */
1771 nb_sectors = count >> BDRV_SECTOR_BITS;
1772 if (nb_sectors > 0) {
1773 if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0)
1775 sector_num += nb_sectors;
1776 len = nb_sectors << BDRV_SECTOR_BITS;
1781 /* add data from the last sector */
1783 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1785 memcpy(tmp_buf, buf, count);
1786 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1793 * Writes to the file and ensures that no writes are reordered across this
1794 * request (acts as a barrier)
1796 * Returns 0 on success, -errno in error cases.
1798 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
1799 const void *buf, int count)
1803 ret = bdrv_pwrite(bs, offset, buf, count);
1808 /* No flush needed for cache modes that already do it */
1809 if (bs->enable_write_cache) {
1816 static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs,
1817 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1819 /* Perform I/O through a temporary buffer so that users who scribble over
1820 * their read buffer while the operation is in progress do not end up
1821 * modifying the image file. This is critical for zero-copy guest I/O
1822 * where anything might happen inside guest memory.
1824 void *bounce_buffer;
1826 BlockDriver *drv = bs->drv;
1828 QEMUIOVector bounce_qiov;
1829 int64_t cluster_sector_num;
1830 int cluster_nb_sectors;
1834 /* Cover entire cluster so no additional backing file I/O is required when
1835 * allocating cluster in the image file.
1837 round_to_clusters(bs, sector_num, nb_sectors,
1838 &cluster_sector_num, &cluster_nb_sectors);
1840 trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors,
1841 cluster_sector_num, cluster_nb_sectors);
1843 iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE;
1844 iov.iov_base = bounce_buffer = qemu_blockalign(bs, iov.iov_len);
1845 qemu_iovec_init_external(&bounce_qiov, &iov, 1);
1847 ret = drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors,
1853 if (drv->bdrv_co_write_zeroes &&
1854 buffer_is_zero(bounce_buffer, iov.iov_len)) {
1855 ret = bdrv_co_do_write_zeroes(bs, cluster_sector_num,
1856 cluster_nb_sectors);
1858 /* This does not change the data on the disk, it is not necessary
1859 * to flush even in cache=writethrough mode.
1861 ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors,
1866 /* It might be okay to ignore write errors for guest requests. If this
1867 * is a deliberate copy-on-read then we don't want to ignore the error.
1868 * Simply report it in all cases.
1873 skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE;
1874 qemu_iovec_from_buffer(qiov, bounce_buffer + skip_bytes,
1875 nb_sectors * BDRV_SECTOR_SIZE);
1878 qemu_vfree(bounce_buffer);
1883 * Handle a read request in coroutine context
1885 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
1886 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1887 BdrvRequestFlags flags)
1889 BlockDriver *drv = bs->drv;
1890 BdrvTrackedRequest req;
1896 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
1900 /* throttling disk read I/O */
1901 if (bs->io_limits_enabled) {
1902 bdrv_io_limits_intercept(bs, false, nb_sectors);
1905 if (bs->copy_on_read) {
1906 flags |= BDRV_REQ_COPY_ON_READ;
1908 if (flags & BDRV_REQ_COPY_ON_READ) {
1909 bs->copy_on_read_in_flight++;
1912 if (bs->copy_on_read_in_flight) {
1913 wait_for_overlapping_requests(bs, sector_num, nb_sectors);
1916 tracked_request_begin(&req, bs, sector_num, nb_sectors, false);
1918 if (flags & BDRV_REQ_COPY_ON_READ) {
1921 ret = bdrv_co_is_allocated(bs, sector_num, nb_sectors, &pnum);
1926 if (!ret || pnum != nb_sectors) {
1927 ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov);
1932 ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
1935 tracked_request_end(&req);
1937 if (flags & BDRV_REQ_COPY_ON_READ) {
1938 bs->copy_on_read_in_flight--;
1944 int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
1945 int nb_sectors, QEMUIOVector *qiov)
1947 trace_bdrv_co_readv(bs, sector_num, nb_sectors);
1949 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0);
1952 int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
1953 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1955 trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors);
1957 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
1958 BDRV_REQ_COPY_ON_READ);
1961 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
1962 int64_t sector_num, int nb_sectors)
1964 BlockDriver *drv = bs->drv;
1969 /* TODO Emulate only part of misaligned requests instead of letting block
1970 * drivers return -ENOTSUP and emulate everything */
1972 /* First try the efficient write zeroes operation */
1973 if (drv->bdrv_co_write_zeroes) {
1974 ret = drv->bdrv_co_write_zeroes(bs, sector_num, nb_sectors);
1975 if (ret != -ENOTSUP) {
1980 /* Fall back to bounce buffer if write zeroes is unsupported */
1981 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
1982 iov.iov_base = qemu_blockalign(bs, iov.iov_len);
1983 memset(iov.iov_base, 0, iov.iov_len);
1984 qemu_iovec_init_external(&qiov, &iov, 1);
1986 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, &qiov);
1988 qemu_vfree(iov.iov_base);
1993 * Handle a write request in coroutine context
1995 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
1996 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1997 BdrvRequestFlags flags)
1999 BlockDriver *drv = bs->drv;
2000 BdrvTrackedRequest req;
2006 if (bs->read_only) {
2009 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
2013 /* throttling disk write I/O */
2014 if (bs->io_limits_enabled) {
2015 bdrv_io_limits_intercept(bs, true, nb_sectors);
2018 if (bs->copy_on_read_in_flight) {
2019 wait_for_overlapping_requests(bs, sector_num, nb_sectors);
2022 tracked_request_begin(&req, bs, sector_num, nb_sectors, true);
2024 if (flags & BDRV_REQ_ZERO_WRITE) {
2025 ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors);
2027 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
2030 if (ret == 0 && !bs->enable_write_cache) {
2031 ret = bdrv_co_flush(bs);
2034 if (bs->dirty_bitmap) {
2035 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
2038 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
2039 bs->wr_highest_sector = sector_num + nb_sectors - 1;
2042 tracked_request_end(&req);
2047 int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
2048 int nb_sectors, QEMUIOVector *qiov)
2050 trace_bdrv_co_writev(bs, sector_num, nb_sectors);
2052 return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0);
2055 int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs,
2056 int64_t sector_num, int nb_sectors)
2058 trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors);
2060 return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL,
2061 BDRV_REQ_ZERO_WRITE);
2065 * Truncate file to 'offset' bytes (needed only for file protocols)
2067 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
2069 BlockDriver *drv = bs->drv;
2073 if (!drv->bdrv_truncate)
2077 if (bdrv_in_use(bs))
2079 ret = drv->bdrv_truncate(bs, offset);
2081 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
2082 bdrv_dev_resize_cb(bs);
2088 * Length of a allocated file in bytes. Sparse files are counted by actual
2089 * allocated space. Return < 0 if error or unknown.
2091 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
2093 BlockDriver *drv = bs->drv;
2097 if (drv->bdrv_get_allocated_file_size) {
2098 return drv->bdrv_get_allocated_file_size(bs);
2101 return bdrv_get_allocated_file_size(bs->file);
2107 * Length of a file in bytes. Return < 0 if error or unknown.
2109 int64_t bdrv_getlength(BlockDriverState *bs)
2111 BlockDriver *drv = bs->drv;
2115 if (bs->growable || bdrv_dev_has_removable_media(bs)) {
2116 if (drv->bdrv_getlength) {
2117 return drv->bdrv_getlength(bs);
2120 return bs->total_sectors * BDRV_SECTOR_SIZE;
2123 /* return 0 as number of sectors if no device present or error */
2124 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
2127 length = bdrv_getlength(bs);
2131 length = length >> BDRV_SECTOR_BITS;
2132 *nb_sectors_ptr = length;
2136 uint8_t boot_ind; /* 0x80 - active */
2137 uint8_t head; /* starting head */
2138 uint8_t sector; /* starting sector */
2139 uint8_t cyl; /* starting cylinder */
2140 uint8_t sys_ind; /* What partition type */
2141 uint8_t end_head; /* end head */
2142 uint8_t end_sector; /* end sector */
2143 uint8_t end_cyl; /* end cylinder */
2144 uint32_t start_sect; /* starting sector counting from 0 */
2145 uint32_t nr_sects; /* nr of sectors in partition */
2148 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
2149 static int guess_disk_lchs(BlockDriverState *bs,
2150 int *pcylinders, int *pheads, int *psectors)
2152 uint8_t buf[BDRV_SECTOR_SIZE];
2153 int i, heads, sectors, cylinders;
2154 struct partition *p;
2156 uint64_t nb_sectors;
2158 bdrv_get_geometry(bs, &nb_sectors);
2161 * The function will be invoked during startup not only in sync I/O mode,
2162 * but also in async I/O mode. So the I/O throttling function has to
2163 * be disabled temporarily here, not permanently.
2165 if (bdrv_read_unthrottled(bs, 0, buf, 1) < 0) {
2168 /* test msdos magic */
2169 if (buf[510] != 0x55 || buf[511] != 0xaa)
2171 for(i = 0; i < 4; i++) {
2172 p = ((struct partition *)(buf + 0x1be)) + i;
2173 nr_sects = le32_to_cpu(p->nr_sects);
2174 if (nr_sects && p->end_head) {
2175 /* We make the assumption that the partition terminates on
2176 a cylinder boundary */
2177 heads = p->end_head + 1;
2178 sectors = p->end_sector & 63;
2181 cylinders = nb_sectors / (heads * sectors);
2182 if (cylinders < 1 || cylinders > 16383)
2185 *psectors = sectors;
2186 *pcylinders = cylinders;
2188 printf("guessed geometry: LCHS=%d %d %d\n",
2189 cylinders, heads, sectors);
2197 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
2199 int translation, lba_detected = 0;
2200 int cylinders, heads, secs;
2201 uint64_t nb_sectors;
2203 /* if a geometry hint is available, use it */
2204 bdrv_get_geometry(bs, &nb_sectors);
2205 bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
2206 translation = bdrv_get_translation_hint(bs);
2207 if (cylinders != 0) {
2212 if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
2214 /* if heads > 16, it means that a BIOS LBA
2215 translation was active, so the default
2216 hardware geometry is OK */
2218 goto default_geometry;
2223 /* disable any translation to be in sync with
2224 the logical geometry */
2225 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
2226 bdrv_set_translation_hint(bs,
2227 BIOS_ATA_TRANSLATION_NONE);
2232 /* if no geometry, use a standard physical disk geometry */
2233 cylinders = nb_sectors / (16 * 63);
2235 if (cylinders > 16383)
2237 else if (cylinders < 2)
2242 if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
2243 if ((*pcyls * *pheads) <= 131072) {
2244 bdrv_set_translation_hint(bs,
2245 BIOS_ATA_TRANSLATION_LARGE);
2247 bdrv_set_translation_hint(bs,
2248 BIOS_ATA_TRANSLATION_LBA);
2252 bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
2256 void bdrv_set_geometry_hint(BlockDriverState *bs,
2257 int cyls, int heads, int secs)
2264 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
2266 bs->translation = translation;
2269 void bdrv_get_geometry_hint(BlockDriverState *bs,
2270 int *pcyls, int *pheads, int *psecs)
2273 *pheads = bs->heads;
2277 /* throttling disk io limits */
2278 void bdrv_set_io_limits(BlockDriverState *bs,
2279 BlockIOLimit *io_limits)
2281 bs->io_limits = *io_limits;
2282 bs->io_limits_enabled = bdrv_io_limits_enabled(bs);
2285 /* Recognize floppy formats */
2286 typedef struct FDFormat {
2294 static const FDFormat fd_formats[] = {
2295 /* First entry is default format */
2296 /* 1.44 MB 3"1/2 floppy disks */
2297 { FDRIVE_DRV_144, 18, 80, 1, FDRIVE_RATE_500K, },
2298 { FDRIVE_DRV_144, 20, 80, 1, FDRIVE_RATE_500K, },
2299 { FDRIVE_DRV_144, 21, 80, 1, FDRIVE_RATE_500K, },
2300 { FDRIVE_DRV_144, 21, 82, 1, FDRIVE_RATE_500K, },
2301 { FDRIVE_DRV_144, 21, 83, 1, FDRIVE_RATE_500K, },
2302 { FDRIVE_DRV_144, 22, 80, 1, FDRIVE_RATE_500K, },
2303 { FDRIVE_DRV_144, 23, 80, 1, FDRIVE_RATE_500K, },
2304 { FDRIVE_DRV_144, 24, 80, 1, FDRIVE_RATE_500K, },
2305 /* 2.88 MB 3"1/2 floppy disks */
2306 { FDRIVE_DRV_288, 36, 80, 1, FDRIVE_RATE_1M, },
2307 { FDRIVE_DRV_288, 39, 80, 1, FDRIVE_RATE_1M, },
2308 { FDRIVE_DRV_288, 40, 80, 1, FDRIVE_RATE_1M, },
2309 { FDRIVE_DRV_288, 44, 80, 1, FDRIVE_RATE_1M, },
2310 { FDRIVE_DRV_288, 48, 80, 1, FDRIVE_RATE_1M, },
2311 /* 720 kB 3"1/2 floppy disks */
2312 { FDRIVE_DRV_144, 9, 80, 1, FDRIVE_RATE_250K, },
2313 { FDRIVE_DRV_144, 10, 80, 1, FDRIVE_RATE_250K, },
2314 { FDRIVE_DRV_144, 10, 82, 1, FDRIVE_RATE_250K, },
2315 { FDRIVE_DRV_144, 10, 83, 1, FDRIVE_RATE_250K, },
2316 { FDRIVE_DRV_144, 13, 80, 1, FDRIVE_RATE_250K, },
2317 { FDRIVE_DRV_144, 14, 80, 1, FDRIVE_RATE_250K, },
2318 /* 1.2 MB 5"1/4 floppy disks */
2319 { FDRIVE_DRV_120, 15, 80, 1, FDRIVE_RATE_500K, },
2320 { FDRIVE_DRV_120, 18, 80, 1, FDRIVE_RATE_500K, },
2321 { FDRIVE_DRV_120, 18, 82, 1, FDRIVE_RATE_500K, },
2322 { FDRIVE_DRV_120, 18, 83, 1, FDRIVE_RATE_500K, },
2323 { FDRIVE_DRV_120, 20, 80, 1, FDRIVE_RATE_500K, },
2324 /* 720 kB 5"1/4 floppy disks */
2325 { FDRIVE_DRV_120, 9, 80, 1, FDRIVE_RATE_250K, },
2326 { FDRIVE_DRV_120, 11, 80, 1, FDRIVE_RATE_250K, },
2327 /* 360 kB 5"1/4 floppy disks */
2328 { FDRIVE_DRV_120, 9, 40, 1, FDRIVE_RATE_300K, },
2329 { FDRIVE_DRV_120, 9, 40, 0, FDRIVE_RATE_300K, },
2330 { FDRIVE_DRV_120, 10, 41, 1, FDRIVE_RATE_300K, },
2331 { FDRIVE_DRV_120, 10, 42, 1, FDRIVE_RATE_300K, },
2332 /* 320 kB 5"1/4 floppy disks */
2333 { FDRIVE_DRV_120, 8, 40, 1, FDRIVE_RATE_250K, },
2334 { FDRIVE_DRV_120, 8, 40, 0, FDRIVE_RATE_250K, },
2335 /* 360 kB must match 5"1/4 better than 3"1/2... */
2336 { FDRIVE_DRV_144, 9, 80, 0, FDRIVE_RATE_250K, },
2338 { FDRIVE_DRV_NONE, -1, -1, 0, 0, },
2341 void bdrv_get_floppy_geometry_hint(BlockDriverState *bs, int *nb_heads,
2342 int *max_track, int *last_sect,
2343 FDriveType drive_in, FDriveType *drive,
2346 const FDFormat *parse;
2347 uint64_t nb_sectors, size;
2348 int i, first_match, match;
2350 bdrv_get_geometry(bs, &nb_sectors);
2353 for (i = 0; ; i++) {
2354 parse = &fd_formats[i];
2355 if (parse->drive == FDRIVE_DRV_NONE) {
2358 if (drive_in == parse->drive ||
2359 drive_in == FDRIVE_DRV_NONE) {
2360 size = (parse->max_head + 1) * parse->max_track *
2362 if (nb_sectors == size) {
2366 if (first_match == -1) {
2372 if (first_match == -1) {
2375 match = first_match;
2377 parse = &fd_formats[match];
2379 *nb_heads = parse->max_head + 1;
2380 *max_track = parse->max_track;
2381 *last_sect = parse->last_sect;
2382 *drive = parse->drive;
2383 *rate = parse->rate;
2386 int bdrv_get_translation_hint(BlockDriverState *bs)
2388 return bs->translation;
2391 void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
2392 BlockErrorAction on_write_error)
2394 bs->on_read_error = on_read_error;
2395 bs->on_write_error = on_write_error;
2398 BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
2400 return is_read ? bs->on_read_error : bs->on_write_error;
2403 int bdrv_is_read_only(BlockDriverState *bs)
2405 return bs->read_only;
2408 int bdrv_is_sg(BlockDriverState *bs)
2413 int bdrv_enable_write_cache(BlockDriverState *bs)
2415 return bs->enable_write_cache;
2418 void bdrv_set_enable_write_cache(BlockDriverState *bs, bool wce)
2420 bs->enable_write_cache = wce;
2423 int bdrv_is_encrypted(BlockDriverState *bs)
2425 if (bs->backing_hd && bs->backing_hd->encrypted)
2427 return bs->encrypted;
2430 int bdrv_key_required(BlockDriverState *bs)
2432 BlockDriverState *backing_hd = bs->backing_hd;
2434 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
2436 return (bs->encrypted && !bs->valid_key);
2439 int bdrv_set_key(BlockDriverState *bs, const char *key)
2442 if (bs->backing_hd && bs->backing_hd->encrypted) {
2443 ret = bdrv_set_key(bs->backing_hd, key);
2449 if (!bs->encrypted) {
2451 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
2454 ret = bs->drv->bdrv_set_key(bs, key);
2457 } else if (!bs->valid_key) {
2459 /* call the change callback now, we skipped it on open */
2460 bdrv_dev_change_media_cb(bs, true);
2465 const char *bdrv_get_format_name(BlockDriverState *bs)
2467 return bs->drv ? bs->drv->format_name : NULL;
2470 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
2475 QLIST_FOREACH(drv, &bdrv_drivers, list) {
2476 it(opaque, drv->format_name);
2480 BlockDriverState *bdrv_find(const char *name)
2482 BlockDriverState *bs;
2484 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2485 if (!strcmp(name, bs->device_name)) {
2492 BlockDriverState *bdrv_next(BlockDriverState *bs)
2495 return QTAILQ_FIRST(&bdrv_states);
2497 return QTAILQ_NEXT(bs, list);
2500 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
2502 BlockDriverState *bs;
2504 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2509 const char *bdrv_get_device_name(BlockDriverState *bs)
2511 return bs->device_name;
2514 int bdrv_get_flags(BlockDriverState *bs)
2516 return bs->open_flags;
2519 void bdrv_flush_all(void)
2521 BlockDriverState *bs;
2523 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2528 int bdrv_has_zero_init(BlockDriverState *bs)
2532 if (bs->drv->bdrv_has_zero_init) {
2533 return bs->drv->bdrv_has_zero_init(bs);
2539 typedef struct BdrvCoIsAllocatedData {
2540 BlockDriverState *bs;
2546 } BdrvCoIsAllocatedData;
2549 * Returns true iff the specified sector is present in the disk image. Drivers
2550 * not implementing the functionality are assumed to not support backing files,
2551 * hence all their sectors are reported as allocated.
2553 * If 'sector_num' is beyond the end of the disk image the return value is 0
2554 * and 'pnum' is set to 0.
2556 * 'pnum' is set to the number of sectors (including and immediately following
2557 * the specified sector) that are known to be in the same
2558 * allocated/unallocated state.
2560 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
2561 * beyond the end of the disk image it will be clamped.
2563 int coroutine_fn bdrv_co_is_allocated(BlockDriverState *bs, int64_t sector_num,
2564 int nb_sectors, int *pnum)
2568 if (sector_num >= bs->total_sectors) {
2573 n = bs->total_sectors - sector_num;
2574 if (n < nb_sectors) {
2578 if (!bs->drv->bdrv_co_is_allocated) {
2583 return bs->drv->bdrv_co_is_allocated(bs, sector_num, nb_sectors, pnum);
2586 /* Coroutine wrapper for bdrv_is_allocated() */
2587 static void coroutine_fn bdrv_is_allocated_co_entry(void *opaque)
2589 BdrvCoIsAllocatedData *data = opaque;
2590 BlockDriverState *bs = data->bs;
2592 data->ret = bdrv_co_is_allocated(bs, data->sector_num, data->nb_sectors,
2598 * Synchronous wrapper around bdrv_co_is_allocated().
2600 * See bdrv_co_is_allocated() for details.
2602 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
2606 BdrvCoIsAllocatedData data = {
2608 .sector_num = sector_num,
2609 .nb_sectors = nb_sectors,
2614 co = qemu_coroutine_create(bdrv_is_allocated_co_entry);
2615 qemu_coroutine_enter(co, &data);
2616 while (!data.done) {
2623 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
2625 * Return true if the given sector is allocated in any image between
2626 * BASE and TOP (inclusive). BASE can be NULL to check if the given
2627 * sector is allocated in any image of the chain. Return false otherwise.
2629 * 'pnum' is set to the number of sectors (including and immediately following
2630 * the specified sector) that are known to be in the same
2631 * allocated/unallocated state.
2634 int coroutine_fn bdrv_co_is_allocated_above(BlockDriverState *top,
2635 BlockDriverState *base,
2637 int nb_sectors, int *pnum)
2639 BlockDriverState *intermediate;
2640 int ret, n = nb_sectors;
2643 while (intermediate && intermediate != base) {
2645 ret = bdrv_co_is_allocated(intermediate, sector_num, nb_sectors,
2655 * [sector_num, nb_sectors] is unallocated on top but intermediate
2658 * [sector_num+x, nr_sectors] allocated.
2660 if (n > pnum_inter) {
2664 intermediate = intermediate->backing_hd;
2671 BlockInfoList *qmp_query_block(Error **errp)
2673 BlockInfoList *head = NULL, *cur_item = NULL;
2674 BlockDriverState *bs;
2676 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2677 BlockInfoList *info = g_malloc0(sizeof(*info));
2679 info->value = g_malloc0(sizeof(*info->value));
2680 info->value->device = g_strdup(bs->device_name);
2681 info->value->type = g_strdup("unknown");
2682 info->value->locked = bdrv_dev_is_medium_locked(bs);
2683 info->value->removable = bdrv_dev_has_removable_media(bs);
2685 if (bdrv_dev_has_removable_media(bs)) {
2686 info->value->has_tray_open = true;
2687 info->value->tray_open = bdrv_dev_is_tray_open(bs);
2690 if (bdrv_iostatus_is_enabled(bs)) {
2691 info->value->has_io_status = true;
2692 info->value->io_status = bs->iostatus;
2696 info->value->has_inserted = true;
2697 info->value->inserted = g_malloc0(sizeof(*info->value->inserted));
2698 info->value->inserted->file = g_strdup(bs->filename);
2699 info->value->inserted->ro = bs->read_only;
2700 info->value->inserted->drv = g_strdup(bs->drv->format_name);
2701 info->value->inserted->encrypted = bs->encrypted;
2702 if (bs->backing_file[0]) {
2703 info->value->inserted->has_backing_file = true;
2704 info->value->inserted->backing_file = g_strdup(bs->backing_file);
2707 if (bs->io_limits_enabled) {
2708 info->value->inserted->bps =
2709 bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL];
2710 info->value->inserted->bps_rd =
2711 bs->io_limits.bps[BLOCK_IO_LIMIT_READ];
2712 info->value->inserted->bps_wr =
2713 bs->io_limits.bps[BLOCK_IO_LIMIT_WRITE];
2714 info->value->inserted->iops =
2715 bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL];
2716 info->value->inserted->iops_rd =
2717 bs->io_limits.iops[BLOCK_IO_LIMIT_READ];
2718 info->value->inserted->iops_wr =
2719 bs->io_limits.iops[BLOCK_IO_LIMIT_WRITE];
2723 /* XXX: waiting for the qapi to support GSList */
2725 head = cur_item = info;
2727 cur_item->next = info;
2735 /* Consider exposing this as a full fledged QMP command */
2736 static BlockStats *qmp_query_blockstat(const BlockDriverState *bs, Error **errp)
2740 s = g_malloc0(sizeof(*s));
2742 if (bs->device_name[0]) {
2743 s->has_device = true;
2744 s->device = g_strdup(bs->device_name);
2747 s->stats = g_malloc0(sizeof(*s->stats));
2748 s->stats->rd_bytes = bs->nr_bytes[BDRV_ACCT_READ];
2749 s->stats->wr_bytes = bs->nr_bytes[BDRV_ACCT_WRITE];
2750 s->stats->rd_operations = bs->nr_ops[BDRV_ACCT_READ];
2751 s->stats->wr_operations = bs->nr_ops[BDRV_ACCT_WRITE];
2752 s->stats->wr_highest_offset = bs->wr_highest_sector * BDRV_SECTOR_SIZE;
2753 s->stats->flush_operations = bs->nr_ops[BDRV_ACCT_FLUSH];
2754 s->stats->wr_total_time_ns = bs->total_time_ns[BDRV_ACCT_WRITE];
2755 s->stats->rd_total_time_ns = bs->total_time_ns[BDRV_ACCT_READ];
2756 s->stats->flush_total_time_ns = bs->total_time_ns[BDRV_ACCT_FLUSH];
2759 s->has_parent = true;
2760 s->parent = qmp_query_blockstat(bs->file, NULL);
2766 BlockStatsList *qmp_query_blockstats(Error **errp)
2768 BlockStatsList *head = NULL, *cur_item = NULL;
2769 BlockDriverState *bs;
2771 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2772 BlockStatsList *info = g_malloc0(sizeof(*info));
2773 info->value = qmp_query_blockstat(bs, NULL);
2775 /* XXX: waiting for the qapi to support GSList */
2777 head = cur_item = info;
2779 cur_item->next = info;
2787 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
2789 if (bs->backing_hd && bs->backing_hd->encrypted)
2790 return bs->backing_file;
2791 else if (bs->encrypted)
2792 return bs->filename;
2797 void bdrv_get_backing_filename(BlockDriverState *bs,
2798 char *filename, int filename_size)
2800 pstrcpy(filename, filename_size, bs->backing_file);
2803 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
2804 const uint8_t *buf, int nb_sectors)
2806 BlockDriver *drv = bs->drv;
2809 if (!drv->bdrv_write_compressed)
2811 if (bdrv_check_request(bs, sector_num, nb_sectors))
2814 if (bs->dirty_bitmap) {
2815 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
2818 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
2821 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2823 BlockDriver *drv = bs->drv;
2826 if (!drv->bdrv_get_info)
2828 memset(bdi, 0, sizeof(*bdi));
2829 return drv->bdrv_get_info(bs, bdi);
2832 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
2833 int64_t pos, int size)
2835 BlockDriver *drv = bs->drv;
2838 if (drv->bdrv_save_vmstate)
2839 return drv->bdrv_save_vmstate(bs, buf, pos, size);
2841 return bdrv_save_vmstate(bs->file, buf, pos, size);
2845 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
2846 int64_t pos, int size)
2848 BlockDriver *drv = bs->drv;
2851 if (drv->bdrv_load_vmstate)
2852 return drv->bdrv_load_vmstate(bs, buf, pos, size);
2854 return bdrv_load_vmstate(bs->file, buf, pos, size);
2858 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
2860 BlockDriver *drv = bs->drv;
2862 if (!drv || !drv->bdrv_debug_event) {
2866 return drv->bdrv_debug_event(bs, event);
2870 /**************************************************************/
2871 /* handling of snapshots */
2873 int bdrv_can_snapshot(BlockDriverState *bs)
2875 BlockDriver *drv = bs->drv;
2876 if (!drv || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
2880 if (!drv->bdrv_snapshot_create) {
2881 if (bs->file != NULL) {
2882 return bdrv_can_snapshot(bs->file);
2890 int bdrv_is_snapshot(BlockDriverState *bs)
2892 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
2895 BlockDriverState *bdrv_snapshots(void)
2897 BlockDriverState *bs;
2900 return bs_snapshots;
2904 while ((bs = bdrv_next(bs))) {
2905 if (bdrv_can_snapshot(bs)) {
2913 int bdrv_snapshot_create(BlockDriverState *bs,
2914 QEMUSnapshotInfo *sn_info)
2916 BlockDriver *drv = bs->drv;
2919 if (drv->bdrv_snapshot_create)
2920 return drv->bdrv_snapshot_create(bs, sn_info);
2922 return bdrv_snapshot_create(bs->file, sn_info);
2926 int bdrv_snapshot_goto(BlockDriverState *bs,
2927 const char *snapshot_id)
2929 BlockDriver *drv = bs->drv;
2934 if (drv->bdrv_snapshot_goto)
2935 return drv->bdrv_snapshot_goto(bs, snapshot_id);
2938 drv->bdrv_close(bs);
2939 ret = bdrv_snapshot_goto(bs->file, snapshot_id);
2940 open_ret = drv->bdrv_open(bs, bs->open_flags);
2942 bdrv_delete(bs->file);
2952 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2954 BlockDriver *drv = bs->drv;
2957 if (drv->bdrv_snapshot_delete)
2958 return drv->bdrv_snapshot_delete(bs, snapshot_id);
2960 return bdrv_snapshot_delete(bs->file, snapshot_id);
2964 int bdrv_snapshot_list(BlockDriverState *bs,
2965 QEMUSnapshotInfo **psn_info)
2967 BlockDriver *drv = bs->drv;
2970 if (drv->bdrv_snapshot_list)
2971 return drv->bdrv_snapshot_list(bs, psn_info);
2973 return bdrv_snapshot_list(bs->file, psn_info);
2977 int bdrv_snapshot_load_tmp(BlockDriverState *bs,
2978 const char *snapshot_name)
2980 BlockDriver *drv = bs->drv;
2984 if (!bs->read_only) {
2987 if (drv->bdrv_snapshot_load_tmp) {
2988 return drv->bdrv_snapshot_load_tmp(bs, snapshot_name);
2993 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
2994 const char *backing_file)
3000 if (bs->backing_hd) {
3001 if (strcmp(bs->backing_file, backing_file) == 0) {
3002 return bs->backing_hd;
3004 return bdrv_find_backing_image(bs->backing_hd, backing_file);
3011 #define NB_SUFFIXES 4
3013 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
3015 static const char suffixes[NB_SUFFIXES] = "KMGT";
3020 snprintf(buf, buf_size, "%" PRId64, size);
3023 for(i = 0; i < NB_SUFFIXES; i++) {
3024 if (size < (10 * base)) {
3025 snprintf(buf, buf_size, "%0.1f%c",
3026 (double)size / base,
3029 } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
3030 snprintf(buf, buf_size, "%" PRId64 "%c",
3031 ((size + (base >> 1)) / base),
3041 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
3043 char buf1[128], date_buf[128], clock_buf[128];
3053 snprintf(buf, buf_size,
3054 "%-10s%-20s%7s%20s%15s",
3055 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
3059 ptm = localtime(&ti);
3060 strftime(date_buf, sizeof(date_buf),
3061 "%Y-%m-%d %H:%M:%S", ptm);
3063 localtime_r(&ti, &tm);
3064 strftime(date_buf, sizeof(date_buf),
3065 "%Y-%m-%d %H:%M:%S", &tm);
3067 secs = sn->vm_clock_nsec / 1000000000;
3068 snprintf(clock_buf, sizeof(clock_buf),
3069 "%02d:%02d:%02d.%03d",
3071 (int)((secs / 60) % 60),
3073 (int)((sn->vm_clock_nsec / 1000000) % 1000));
3074 snprintf(buf, buf_size,
3075 "%-10s%-20s%7s%20s%15s",
3076 sn->id_str, sn->name,
3077 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
3084 /**************************************************************/
3087 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
3088 QEMUIOVector *qiov, int nb_sectors,
3089 BlockDriverCompletionFunc *cb, void *opaque)
3091 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
3093 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
3097 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
3098 QEMUIOVector *qiov, int nb_sectors,
3099 BlockDriverCompletionFunc *cb, void *opaque)
3101 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
3103 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
3108 typedef struct MultiwriteCB {
3113 BlockDriverCompletionFunc *cb;
3115 QEMUIOVector *free_qiov;
3119 static void multiwrite_user_cb(MultiwriteCB *mcb)
3123 for (i = 0; i < mcb->num_callbacks; i++) {
3124 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
3125 if (mcb->callbacks[i].free_qiov) {
3126 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
3128 g_free(mcb->callbacks[i].free_qiov);
3132 static void multiwrite_cb(void *opaque, int ret)
3134 MultiwriteCB *mcb = opaque;
3136 trace_multiwrite_cb(mcb, ret);
3138 if (ret < 0 && !mcb->error) {
3142 mcb->num_requests--;
3143 if (mcb->num_requests == 0) {
3144 multiwrite_user_cb(mcb);
3149 static int multiwrite_req_compare(const void *a, const void *b)
3151 const BlockRequest *req1 = a, *req2 = b;
3154 * Note that we can't simply subtract req2->sector from req1->sector
3155 * here as that could overflow the return value.
3157 if (req1->sector > req2->sector) {
3159 } else if (req1->sector < req2->sector) {
3167 * Takes a bunch of requests and tries to merge them. Returns the number of
3168 * requests that remain after merging.
3170 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
3171 int num_reqs, MultiwriteCB *mcb)
3175 // Sort requests by start sector
3176 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
3178 // Check if adjacent requests touch the same clusters. If so, combine them,
3179 // filling up gaps with zero sectors.
3181 for (i = 1; i < num_reqs; i++) {
3183 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
3185 // Handle exactly sequential writes and overlapping writes.
3186 if (reqs[i].sector <= oldreq_last) {
3190 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
3196 QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
3197 qemu_iovec_init(qiov,
3198 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
3200 // Add the first request to the merged one. If the requests are
3201 // overlapping, drop the last sectors of the first request.
3202 size = (reqs[i].sector - reqs[outidx].sector) << 9;
3203 qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
3205 // We should need to add any zeros between the two requests
3206 assert (reqs[i].sector <= oldreq_last);
3208 // Add the second request
3209 qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
3211 reqs[outidx].nb_sectors = qiov->size >> 9;
3212 reqs[outidx].qiov = qiov;
3214 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
3217 reqs[outidx].sector = reqs[i].sector;
3218 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
3219 reqs[outidx].qiov = reqs[i].qiov;
3227 * Submit multiple AIO write requests at once.
3229 * On success, the function returns 0 and all requests in the reqs array have
3230 * been submitted. In error case this function returns -1, and any of the
3231 * requests may or may not be submitted yet. In particular, this means that the
3232 * callback will be called for some of the requests, for others it won't. The
3233 * caller must check the error field of the BlockRequest to wait for the right
3234 * callbacks (if error != 0, no callback will be called).
3236 * The implementation may modify the contents of the reqs array, e.g. to merge
3237 * requests. However, the fields opaque and error are left unmodified as they
3238 * are used to signal failure for a single request to the caller.
3240 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
3245 /* don't submit writes if we don't have a medium */
3246 if (bs->drv == NULL) {
3247 for (i = 0; i < num_reqs; i++) {
3248 reqs[i].error = -ENOMEDIUM;
3253 if (num_reqs == 0) {
3257 // Create MultiwriteCB structure
3258 mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
3259 mcb->num_requests = 0;
3260 mcb->num_callbacks = num_reqs;
3262 for (i = 0; i < num_reqs; i++) {
3263 mcb->callbacks[i].cb = reqs[i].cb;
3264 mcb->callbacks[i].opaque = reqs[i].opaque;
3267 // Check for mergable requests
3268 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
3270 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
3272 /* Run the aio requests. */
3273 mcb->num_requests = num_reqs;
3274 for (i = 0; i < num_reqs; i++) {
3275 bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
3276 reqs[i].nb_sectors, multiwrite_cb, mcb);
3282 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
3284 acb->pool->cancel(acb);
3287 /* block I/O throttling */
3288 static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors,
3289 bool is_write, double elapsed_time, uint64_t *wait)
3291 uint64_t bps_limit = 0;
3292 double bytes_limit, bytes_base, bytes_res;
3293 double slice_time, wait_time;
3295 if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
3296 bps_limit = bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL];
3297 } else if (bs->io_limits.bps[is_write]) {
3298 bps_limit = bs->io_limits.bps[is_write];
3307 slice_time = bs->slice_end - bs->slice_start;
3308 slice_time /= (NANOSECONDS_PER_SECOND);
3309 bytes_limit = bps_limit * slice_time;
3310 bytes_base = bs->nr_bytes[is_write] - bs->io_base.bytes[is_write];
3311 if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
3312 bytes_base += bs->nr_bytes[!is_write] - bs->io_base.bytes[!is_write];
3315 /* bytes_base: the bytes of data which have been read/written; and
3316 * it is obtained from the history statistic info.
3317 * bytes_res: the remaining bytes of data which need to be read/written.
3318 * (bytes_base + bytes_res) / bps_limit: used to calcuate
3319 * the total time for completing reading/writting all data.
3321 bytes_res = (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
3323 if (bytes_base + bytes_res <= bytes_limit) {
3331 /* Calc approx time to dispatch */
3332 wait_time = (bytes_base + bytes_res) / bps_limit - elapsed_time;
3334 /* When the I/O rate at runtime exceeds the limits,
3335 * bs->slice_end need to be extended in order that the current statistic
3336 * info can be kept until the timer fire, so it is increased and tuned
3337 * based on the result of experiment.
3339 bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10;
3340 bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME;
3342 *wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
3348 static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write,
3349 double elapsed_time, uint64_t *wait)
3351 uint64_t iops_limit = 0;
3352 double ios_limit, ios_base;
3353 double slice_time, wait_time;
3355 if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
3356 iops_limit = bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL];
3357 } else if (bs->io_limits.iops[is_write]) {
3358 iops_limit = bs->io_limits.iops[is_write];
3367 slice_time = bs->slice_end - bs->slice_start;
3368 slice_time /= (NANOSECONDS_PER_SECOND);
3369 ios_limit = iops_limit * slice_time;
3370 ios_base = bs->nr_ops[is_write] - bs->io_base.ios[is_write];
3371 if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
3372 ios_base += bs->nr_ops[!is_write] - bs->io_base.ios[!is_write];
3375 if (ios_base + 1 <= ios_limit) {
3383 /* Calc approx time to dispatch */
3384 wait_time = (ios_base + 1) / iops_limit;
3385 if (wait_time > elapsed_time) {
3386 wait_time = wait_time - elapsed_time;
3391 bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10;
3392 bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME;
3394 *wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
3400 static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors,
3401 bool is_write, int64_t *wait)
3403 int64_t now, max_wait;
3404 uint64_t bps_wait = 0, iops_wait = 0;
3405 double elapsed_time;
3406 int bps_ret, iops_ret;
3408 now = qemu_get_clock_ns(vm_clock);
3409 if ((bs->slice_start < now)
3410 && (bs->slice_end > now)) {
3411 bs->slice_end = now + bs->slice_time;
3413 bs->slice_time = 5 * BLOCK_IO_SLICE_TIME;
3414 bs->slice_start = now;
3415 bs->slice_end = now + bs->slice_time;
3417 bs->io_base.bytes[is_write] = bs->nr_bytes[is_write];
3418 bs->io_base.bytes[!is_write] = bs->nr_bytes[!is_write];
3420 bs->io_base.ios[is_write] = bs->nr_ops[is_write];
3421 bs->io_base.ios[!is_write] = bs->nr_ops[!is_write];
3424 elapsed_time = now - bs->slice_start;
3425 elapsed_time /= (NANOSECONDS_PER_SECOND);
3427 bps_ret = bdrv_exceed_bps_limits(bs, nb_sectors,
3428 is_write, elapsed_time, &bps_wait);
3429 iops_ret = bdrv_exceed_iops_limits(bs, is_write,
3430 elapsed_time, &iops_wait);
3431 if (bps_ret || iops_ret) {
3432 max_wait = bps_wait > iops_wait ? bps_wait : iops_wait;
3437 now = qemu_get_clock_ns(vm_clock);
3438 if (bs->slice_end < now + max_wait) {
3439 bs->slice_end = now + max_wait;
3452 /**************************************************************/
3453 /* async block device emulation */
3455 typedef struct BlockDriverAIOCBSync {
3456 BlockDriverAIOCB common;
3459 /* vector translation state */
3463 } BlockDriverAIOCBSync;
3465 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
3467 BlockDriverAIOCBSync *acb =
3468 container_of(blockacb, BlockDriverAIOCBSync, common);
3469 qemu_bh_delete(acb->bh);
3471 qemu_aio_release(acb);
3474 static AIOPool bdrv_em_aio_pool = {
3475 .aiocb_size = sizeof(BlockDriverAIOCBSync),
3476 .cancel = bdrv_aio_cancel_em,
3479 static void bdrv_aio_bh_cb(void *opaque)
3481 BlockDriverAIOCBSync *acb = opaque;
3484 qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
3485 qemu_vfree(acb->bounce);
3486 acb->common.cb(acb->common.opaque, acb->ret);
3487 qemu_bh_delete(acb->bh);
3489 qemu_aio_release(acb);
3492 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
3496 BlockDriverCompletionFunc *cb,
3501 BlockDriverAIOCBSync *acb;
3503 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
3504 acb->is_write = is_write;
3506 acb->bounce = qemu_blockalign(bs, qiov->size);
3507 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
3510 qemu_iovec_to_buffer(acb->qiov, acb->bounce);
3511 acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
3513 acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
3516 qemu_bh_schedule(acb->bh);
3518 return &acb->common;
3521 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
3522 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
3523 BlockDriverCompletionFunc *cb, void *opaque)
3525 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
3528 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
3529 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
3530 BlockDriverCompletionFunc *cb, void *opaque)
3532 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
3536 typedef struct BlockDriverAIOCBCoroutine {
3537 BlockDriverAIOCB common;
3541 } BlockDriverAIOCBCoroutine;
3543 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb)
3548 static AIOPool bdrv_em_co_aio_pool = {
3549 .aiocb_size = sizeof(BlockDriverAIOCBCoroutine),
3550 .cancel = bdrv_aio_co_cancel_em,
3553 static void bdrv_co_em_bh(void *opaque)
3555 BlockDriverAIOCBCoroutine *acb = opaque;
3557 acb->common.cb(acb->common.opaque, acb->req.error);
3558 qemu_bh_delete(acb->bh);
3559 qemu_aio_release(acb);
3562 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
3563 static void coroutine_fn bdrv_co_do_rw(void *opaque)
3565 BlockDriverAIOCBCoroutine *acb = opaque;
3566 BlockDriverState *bs = acb->common.bs;
3568 if (!acb->is_write) {
3569 acb->req.error = bdrv_co_do_readv(bs, acb->req.sector,
3570 acb->req.nb_sectors, acb->req.qiov, 0);
3572 acb->req.error = bdrv_co_do_writev(bs, acb->req.sector,
3573 acb->req.nb_sectors, acb->req.qiov, 0);
3576 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3577 qemu_bh_schedule(acb->bh);
3580 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
3584 BlockDriverCompletionFunc *cb,
3589 BlockDriverAIOCBCoroutine *acb;
3591 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3592 acb->req.sector = sector_num;
3593 acb->req.nb_sectors = nb_sectors;
3594 acb->req.qiov = qiov;
3595 acb->is_write = is_write;
3597 co = qemu_coroutine_create(bdrv_co_do_rw);
3598 qemu_coroutine_enter(co, acb);
3600 return &acb->common;
3603 static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
3605 BlockDriverAIOCBCoroutine *acb = opaque;
3606 BlockDriverState *bs = acb->common.bs;
3608 acb->req.error = bdrv_co_flush(bs);
3609 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3610 qemu_bh_schedule(acb->bh);
3613 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
3614 BlockDriverCompletionFunc *cb, void *opaque)
3616 trace_bdrv_aio_flush(bs, opaque);
3619 BlockDriverAIOCBCoroutine *acb;
3621 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3622 co = qemu_coroutine_create(bdrv_aio_flush_co_entry);
3623 qemu_coroutine_enter(co, acb);
3625 return &acb->common;
3628 static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque)
3630 BlockDriverAIOCBCoroutine *acb = opaque;
3631 BlockDriverState *bs = acb->common.bs;
3633 acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors);
3634 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3635 qemu_bh_schedule(acb->bh);
3638 BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs,
3639 int64_t sector_num, int nb_sectors,
3640 BlockDriverCompletionFunc *cb, void *opaque)
3643 BlockDriverAIOCBCoroutine *acb;
3645 trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque);
3647 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3648 acb->req.sector = sector_num;
3649 acb->req.nb_sectors = nb_sectors;
3650 co = qemu_coroutine_create(bdrv_aio_discard_co_entry);
3651 qemu_coroutine_enter(co, acb);
3653 return &acb->common;
3656 void bdrv_init(void)
3658 module_call_init(MODULE_INIT_BLOCK);
3661 void bdrv_init_with_whitelist(void)
3663 use_bdrv_whitelist = 1;
3667 void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
3668 BlockDriverCompletionFunc *cb, void *opaque)
3670 BlockDriverAIOCB *acb;
3672 if (pool->free_aiocb) {
3673 acb = pool->free_aiocb;
3674 pool->free_aiocb = acb->next;
3676 acb = g_malloc0(pool->aiocb_size);
3681 acb->opaque = opaque;
3685 void qemu_aio_release(void *p)
3687 BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
3688 AIOPool *pool = acb->pool;
3689 acb->next = pool->free_aiocb;
3690 pool->free_aiocb = acb;
3693 /**************************************************************/
3694 /* Coroutine block device emulation */
3696 typedef struct CoroutineIOCompletion {
3697 Coroutine *coroutine;
3699 } CoroutineIOCompletion;
3701 static void bdrv_co_io_em_complete(void *opaque, int ret)
3703 CoroutineIOCompletion *co = opaque;
3706 qemu_coroutine_enter(co->coroutine, NULL);
3709 static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num,
3710 int nb_sectors, QEMUIOVector *iov,
3713 CoroutineIOCompletion co = {
3714 .coroutine = qemu_coroutine_self(),
3716 BlockDriverAIOCB *acb;
3719 acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors,
3720 bdrv_co_io_em_complete, &co);
3722 acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors,
3723 bdrv_co_io_em_complete, &co);
3726 trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb);
3730 qemu_coroutine_yield();
3735 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
3736 int64_t sector_num, int nb_sectors,
3739 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false);
3742 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
3743 int64_t sector_num, int nb_sectors,
3746 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true);
3749 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
3751 RwCo *rwco = opaque;
3753 rwco->ret = bdrv_co_flush(rwco->bs);
3756 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
3760 if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
3764 /* Write back cached data to the OS even with cache=unsafe */
3765 if (bs->drv->bdrv_co_flush_to_os) {
3766 ret = bs->drv->bdrv_co_flush_to_os(bs);
3772 /* But don't actually force it to the disk with cache=unsafe */
3773 if (bs->open_flags & BDRV_O_NO_FLUSH) {
3777 if (bs->drv->bdrv_co_flush_to_disk) {
3778 ret = bs->drv->bdrv_co_flush_to_disk(bs);
3779 } else if (bs->drv->bdrv_aio_flush) {
3780 BlockDriverAIOCB *acb;
3781 CoroutineIOCompletion co = {
3782 .coroutine = qemu_coroutine_self(),
3785 acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
3789 qemu_coroutine_yield();
3794 * Some block drivers always operate in either writethrough or unsafe
3795 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
3796 * know how the server works (because the behaviour is hardcoded or
3797 * depends on server-side configuration), so we can't ensure that
3798 * everything is safe on disk. Returning an error doesn't work because
3799 * that would break guests even if the server operates in writethrough
3802 * Let's hope the user knows what he's doing.
3810 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
3811 * in the case of cache=unsafe, so there are no useless flushes.
3813 return bdrv_co_flush(bs->file);
3816 void bdrv_invalidate_cache(BlockDriverState *bs)
3818 if (bs->drv && bs->drv->bdrv_invalidate_cache) {
3819 bs->drv->bdrv_invalidate_cache(bs);
3823 void bdrv_invalidate_cache_all(void)
3825 BlockDriverState *bs;
3827 QTAILQ_FOREACH(bs, &bdrv_states, list) {
3828 bdrv_invalidate_cache(bs);
3832 void bdrv_clear_incoming_migration_all(void)
3834 BlockDriverState *bs;
3836 QTAILQ_FOREACH(bs, &bdrv_states, list) {
3837 bs->open_flags = bs->open_flags & ~(BDRV_O_INCOMING);
3841 int bdrv_flush(BlockDriverState *bs)
3849 if (qemu_in_coroutine()) {
3850 /* Fast-path if already in coroutine context */
3851 bdrv_flush_co_entry(&rwco);
3853 co = qemu_coroutine_create(bdrv_flush_co_entry);
3854 qemu_coroutine_enter(co, &rwco);
3855 while (rwco.ret == NOT_DONE) {
3863 static void coroutine_fn bdrv_discard_co_entry(void *opaque)
3865 RwCo *rwco = opaque;
3867 rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors);
3870 int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
3875 } else if (bdrv_check_request(bs, sector_num, nb_sectors)) {
3877 } else if (bs->read_only) {
3879 } else if (bs->drv->bdrv_co_discard) {
3880 return bs->drv->bdrv_co_discard(bs, sector_num, nb_sectors);
3881 } else if (bs->drv->bdrv_aio_discard) {
3882 BlockDriverAIOCB *acb;
3883 CoroutineIOCompletion co = {
3884 .coroutine = qemu_coroutine_self(),
3887 acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors,
3888 bdrv_co_io_em_complete, &co);
3892 qemu_coroutine_yield();
3900 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
3905 .sector_num = sector_num,
3906 .nb_sectors = nb_sectors,
3910 if (qemu_in_coroutine()) {
3911 /* Fast-path if already in coroutine context */
3912 bdrv_discard_co_entry(&rwco);
3914 co = qemu_coroutine_create(bdrv_discard_co_entry);
3915 qemu_coroutine_enter(co, &rwco);
3916 while (rwco.ret == NOT_DONE) {
3924 /**************************************************************/
3925 /* removable device support */
3928 * Return TRUE if the media is present
3930 int bdrv_is_inserted(BlockDriverState *bs)
3932 BlockDriver *drv = bs->drv;
3936 if (!drv->bdrv_is_inserted)
3938 return drv->bdrv_is_inserted(bs);
3942 * Return whether the media changed since the last call to this
3943 * function, or -ENOTSUP if we don't know. Most drivers don't know.
3945 int bdrv_media_changed(BlockDriverState *bs)
3947 BlockDriver *drv = bs->drv;
3949 if (drv && drv->bdrv_media_changed) {
3950 return drv->bdrv_media_changed(bs);
3956 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
3958 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
3960 BlockDriver *drv = bs->drv;
3962 if (drv && drv->bdrv_eject) {
3963 drv->bdrv_eject(bs, eject_flag);
3966 if (bs->device_name[0] != '\0') {
3967 bdrv_emit_qmp_eject_event(bs, eject_flag);
3972 * Lock or unlock the media (if it is locked, the user won't be able
3973 * to eject it manually).
3975 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
3977 BlockDriver *drv = bs->drv;
3979 trace_bdrv_lock_medium(bs, locked);
3981 if (drv && drv->bdrv_lock_medium) {
3982 drv->bdrv_lock_medium(bs, locked);
3986 /* needed for generic scsi interface */
3988 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
3990 BlockDriver *drv = bs->drv;
3992 if (drv && drv->bdrv_ioctl)
3993 return drv->bdrv_ioctl(bs, req, buf);
3997 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
3998 unsigned long int req, void *buf,
3999 BlockDriverCompletionFunc *cb, void *opaque)
4001 BlockDriver *drv = bs->drv;
4003 if (drv && drv->bdrv_aio_ioctl)
4004 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
4008 void bdrv_set_buffer_alignment(BlockDriverState *bs, int align)
4010 bs->buffer_alignment = align;
4013 void *qemu_blockalign(BlockDriverState *bs, size_t size)
4015 return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
4018 void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
4020 int64_t bitmap_size;
4022 bs->dirty_count = 0;
4024 if (!bs->dirty_bitmap) {
4025 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
4026 BDRV_SECTORS_PER_DIRTY_CHUNK * BITS_PER_LONG - 1;
4027 bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * BITS_PER_LONG;
4029 bs->dirty_bitmap = g_new0(unsigned long, bitmap_size);
4032 if (bs->dirty_bitmap) {
4033 g_free(bs->dirty_bitmap);
4034 bs->dirty_bitmap = NULL;
4039 int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
4041 int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
4043 if (bs->dirty_bitmap &&
4044 (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
4045 return !!(bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
4046 (1UL << (chunk % (sizeof(unsigned long) * 8))));
4052 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
4055 set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
4058 int64_t bdrv_get_dirty_count(BlockDriverState *bs)
4060 return bs->dirty_count;
4063 void bdrv_set_in_use(BlockDriverState *bs, int in_use)
4065 assert(bs->in_use != in_use);
4066 bs->in_use = in_use;
4069 int bdrv_in_use(BlockDriverState *bs)
4074 void bdrv_iostatus_enable(BlockDriverState *bs)
4076 bs->iostatus_enabled = true;
4077 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
4080 /* The I/O status is only enabled if the drive explicitly
4081 * enables it _and_ the VM is configured to stop on errors */
4082 bool bdrv_iostatus_is_enabled(const BlockDriverState *bs)
4084 return (bs->iostatus_enabled &&
4085 (bs->on_write_error == BLOCK_ERR_STOP_ENOSPC ||
4086 bs->on_write_error == BLOCK_ERR_STOP_ANY ||
4087 bs->on_read_error == BLOCK_ERR_STOP_ANY));
4090 void bdrv_iostatus_disable(BlockDriverState *bs)
4092 bs->iostatus_enabled = false;
4095 void bdrv_iostatus_reset(BlockDriverState *bs)
4097 if (bdrv_iostatus_is_enabled(bs)) {
4098 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
4102 /* XXX: Today this is set by device models because it makes the implementation
4103 quite simple. However, the block layer knows about the error, so it's
4104 possible to implement this without device models being involved */
4105 void bdrv_iostatus_set_err(BlockDriverState *bs, int error)
4107 if (bdrv_iostatus_is_enabled(bs) &&
4108 bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
4110 bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
4111 BLOCK_DEVICE_IO_STATUS_FAILED;
4116 bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes,
4117 enum BlockAcctType type)
4119 assert(type < BDRV_MAX_IOTYPE);
4121 cookie->bytes = bytes;
4122 cookie->start_time_ns = get_clock();
4123 cookie->type = type;
4127 bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie)
4129 assert(cookie->type < BDRV_MAX_IOTYPE);
4131 bs->nr_bytes[cookie->type] += cookie->bytes;
4132 bs->nr_ops[cookie->type]++;
4133 bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns;
4136 int bdrv_img_create(const char *filename, const char *fmt,
4137 const char *base_filename, const char *base_fmt,
4138 char *options, uint64_t img_size, int flags)
4140 QEMUOptionParameter *param = NULL, *create_options = NULL;
4141 QEMUOptionParameter *backing_fmt, *backing_file, *size;
4142 BlockDriverState *bs = NULL;
4143 BlockDriver *drv, *proto_drv;
4144 BlockDriver *backing_drv = NULL;
4147 /* Find driver and parse its options */
4148 drv = bdrv_find_format(fmt);
4150 error_report("Unknown file format '%s'", fmt);
4155 proto_drv = bdrv_find_protocol(filename);
4157 error_report("Unknown protocol '%s'", filename);
4162 create_options = append_option_parameters(create_options,
4163 drv->create_options);
4164 create_options = append_option_parameters(create_options,
4165 proto_drv->create_options);
4167 /* Create parameter list with default values */
4168 param = parse_option_parameters("", create_options, param);
4170 set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size);
4172 /* Parse -o options */
4174 param = parse_option_parameters(options, create_options, param);
4175 if (param == NULL) {
4176 error_report("Invalid options for file format '%s'.", fmt);
4182 if (base_filename) {
4183 if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE,
4185 error_report("Backing file not supported for file format '%s'",
4193 if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) {
4194 error_report("Backing file format not supported for file "
4195 "format '%s'", fmt);
4201 backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE);
4202 if (backing_file && backing_file->value.s) {
4203 if (!strcmp(filename, backing_file->value.s)) {
4204 error_report("Error: Trying to create an image with the "
4205 "same filename as the backing file");
4211 backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT);
4212 if (backing_fmt && backing_fmt->value.s) {
4213 backing_drv = bdrv_find_format(backing_fmt->value.s);
4215 error_report("Unknown backing file format '%s'",
4216 backing_fmt->value.s);
4222 // The size for the image must always be specified, with one exception:
4223 // If we are using a backing file, we can obtain the size from there
4224 size = get_option_parameter(param, BLOCK_OPT_SIZE);
4225 if (size && size->value.n == -1) {
4226 if (backing_file && backing_file->value.s) {
4231 /* backing files always opened read-only */
4233 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
4237 ret = bdrv_open(bs, backing_file->value.s, back_flags, backing_drv);
4239 error_report("Could not open '%s'", backing_file->value.s);
4242 bdrv_get_geometry(bs, &size);
4245 snprintf(buf, sizeof(buf), "%" PRId64, size);
4246 set_option_parameter(param, BLOCK_OPT_SIZE, buf);
4248 error_report("Image creation needs a size parameter");
4254 printf("Formatting '%s', fmt=%s ", filename, fmt);
4255 print_option_parameters(param);
4258 ret = bdrv_create(drv, filename, param);
4261 if (ret == -ENOTSUP) {
4262 error_report("Formatting or formatting option not supported for "
4263 "file format '%s'", fmt);
4264 } else if (ret == -EFBIG) {
4265 error_report("The image size is too large for file format '%s'",
4268 error_report("%s: error while creating %s: %s", filename, fmt,
4274 free_option_parameters(create_options);
4275 free_option_parameters(param);
4284 void *block_job_create(const BlockJobType *job_type, BlockDriverState *bs,
4285 int64_t speed, BlockDriverCompletionFunc *cb,
4286 void *opaque, Error **errp)
4290 if (bs->job || bdrv_in_use(bs)) {
4291 error_set(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
4294 bdrv_set_in_use(bs, 1);
4296 job = g_malloc0(job_type->instance_size);
4297 job->job_type = job_type;
4300 job->opaque = opaque;
4304 /* Only set speed when necessary to avoid NotSupported error */
4306 Error *local_err = NULL;
4308 block_job_set_speed(job, speed, &local_err);
4309 if (error_is_set(&local_err)) {
4312 bdrv_set_in_use(bs, 0);
4313 error_propagate(errp, local_err);
4320 void block_job_complete(BlockJob *job, int ret)
4322 BlockDriverState *bs = job->bs;
4324 assert(bs->job == job);
4325 job->cb(job->opaque, ret);
4328 bdrv_set_in_use(bs, 0);
4331 void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
4333 Error *local_err = NULL;
4335 if (!job->job_type->set_speed) {
4336 error_set(errp, QERR_NOT_SUPPORTED);
4339 job->job_type->set_speed(job, speed, &local_err);
4340 if (error_is_set(&local_err)) {
4341 error_propagate(errp, local_err);
4348 void block_job_cancel(BlockJob *job)
4350 job->cancelled = true;
4351 if (job->co && !job->busy) {
4352 qemu_coroutine_enter(job->co, NULL);
4356 bool block_job_is_cancelled(BlockJob *job)
4358 return job->cancelled;
4361 struct BlockCancelData {
4363 BlockDriverCompletionFunc *cb;
4369 static void block_job_cancel_cb(void *opaque, int ret)
4371 struct BlockCancelData *data = opaque;
4373 data->cancelled = block_job_is_cancelled(data->job);
4375 data->cb(data->opaque, ret);
4378 int block_job_cancel_sync(BlockJob *job)
4380 struct BlockCancelData data;
4381 BlockDriverState *bs = job->bs;
4383 assert(bs->job == job);
4385 /* Set up our own callback to store the result and chain to
4386 * the original callback.
4390 data.opaque = job->opaque;
4391 data.ret = -EINPROGRESS;
4392 job->cb = block_job_cancel_cb;
4393 job->opaque = &data;
4394 block_job_cancel(job);
4395 while (data.ret == -EINPROGRESS) {
4398 return (data.cancelled && data.ret == 0) ? -ECANCELED : data.ret;
4401 void block_job_sleep_ns(BlockJob *job, QEMUClock *clock, int64_t ns)
4403 /* Check cancellation *before* setting busy = false, too! */
4404 if (!block_job_is_cancelled(job)) {
4406 co_sleep_ns(clock, ns);