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);
654 * Clear flags that are internal to the block layer before opening the
657 open_flags = flags & ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
660 * Snapshots should be writable.
662 if (bs->is_temporary) {
663 open_flags |= BDRV_O_RDWR;
666 bs->keep_read_only = bs->read_only = !(open_flags & BDRV_O_RDWR);
668 /* Open the image, either directly or using a protocol */
669 if (drv->bdrv_file_open) {
670 ret = drv->bdrv_file_open(bs, filename, open_flags);
672 ret = bdrv_file_open(&bs->file, filename, open_flags);
674 ret = drv->bdrv_open(bs, open_flags);
682 ret = refresh_total_sectors(bs, bs->total_sectors);
688 if (bs->is_temporary) {
696 bdrv_delete(bs->file);
706 * Opens a file using a protocol (file, host_device, nbd, ...)
708 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
710 BlockDriverState *bs;
714 drv = bdrv_find_protocol(filename);
720 ret = bdrv_open_common(bs, filename, flags, drv);
731 * Opens a disk image (raw, qcow2, vmdk, ...)
733 int bdrv_open(BlockDriverState *bs, const char *filename, int flags,
737 char tmp_filename[PATH_MAX];
739 if (flags & BDRV_O_SNAPSHOT) {
740 BlockDriverState *bs1;
743 BlockDriver *bdrv_qcow2;
744 QEMUOptionParameter *options;
745 char backing_filename[PATH_MAX];
747 /* if snapshot, we create a temporary backing file and open it
748 instead of opening 'filename' directly */
750 /* if there is a backing file, use it */
752 ret = bdrv_open(bs1, filename, 0, drv);
757 total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
759 if (bs1->drv && bs1->drv->protocol_name)
764 ret = get_tmp_filename(tmp_filename, sizeof(tmp_filename));
769 /* Real path is meaningless for protocols */
771 snprintf(backing_filename, sizeof(backing_filename),
773 else if (!realpath(filename, backing_filename))
776 bdrv_qcow2 = bdrv_find_format("qcow2");
777 options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
779 set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size);
780 set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
782 set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
786 ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
787 free_option_parameters(options);
792 filename = tmp_filename;
794 bs->is_temporary = 1;
797 /* Find the right image format driver */
799 ret = find_image_format(filename, &drv);
803 goto unlink_and_fail;
807 ret = bdrv_open_common(bs, filename, flags, drv);
809 goto unlink_and_fail;
812 /* If there is a backing file, use it */
813 if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') {
814 char backing_filename[PATH_MAX];
816 BlockDriver *back_drv = NULL;
818 bs->backing_hd = bdrv_new("");
819 bdrv_get_full_backing_filename(bs, backing_filename,
820 sizeof(backing_filename));
822 if (bs->backing_format[0] != '\0') {
823 back_drv = bdrv_find_format(bs->backing_format);
826 /* backing files always opened read-only */
828 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
830 ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv);
835 if (bs->is_temporary) {
836 bs->backing_hd->keep_read_only = !(flags & BDRV_O_RDWR);
838 /* base image inherits from "parent" */
839 bs->backing_hd->keep_read_only = bs->keep_read_only;
843 if (!bdrv_key_required(bs)) {
844 bdrv_dev_change_media_cb(bs, true);
847 /* throttling disk I/O limits */
848 if (bs->io_limits_enabled) {
849 bdrv_io_limits_enable(bs);
855 if (bs->is_temporary) {
861 void bdrv_close(BlockDriverState *bs)
866 block_job_cancel_sync(bs->job);
870 if (bs == bs_snapshots) {
873 if (bs->backing_hd) {
874 bdrv_delete(bs->backing_hd);
875 bs->backing_hd = NULL;
877 bs->drv->bdrv_close(bs);
880 if (bs->is_temporary) {
881 unlink(bs->filename);
886 bs->copy_on_read = 0;
887 bs->backing_file[0] = '\0';
888 bs->backing_format[0] = '\0';
889 bs->total_sectors = 0;
895 if (bs->file != NULL) {
896 bdrv_delete(bs->file);
900 bdrv_dev_change_media_cb(bs, false);
903 /*throttling disk I/O limits*/
904 if (bs->io_limits_enabled) {
905 bdrv_io_limits_disable(bs);
909 void bdrv_close_all(void)
911 BlockDriverState *bs;
913 QTAILQ_FOREACH(bs, &bdrv_states, list) {
919 * Wait for pending requests to complete across all BlockDriverStates
921 * This function does not flush data to disk, use bdrv_flush_all() for that
922 * after calling this function.
924 * Note that completion of an asynchronous I/O operation can trigger any
925 * number of other I/O operations on other devices---for example a coroutine
926 * can be arbitrarily complex and a constant flow of I/O can come until the
927 * coroutine is complete. Because of this, it is not possible to have a
928 * function to drain a single device's I/O queue.
930 void bdrv_drain_all(void)
932 BlockDriverState *bs;
936 busy = qemu_aio_wait();
938 /* FIXME: We do not have timer support here, so this is effectively
941 QTAILQ_FOREACH(bs, &bdrv_states, list) {
942 if (!qemu_co_queue_empty(&bs->throttled_reqs)) {
943 qemu_co_queue_restart_all(&bs->throttled_reqs);
949 /* If requests are still pending there is a bug somewhere */
950 QTAILQ_FOREACH(bs, &bdrv_states, list) {
951 assert(QLIST_EMPTY(&bs->tracked_requests));
952 assert(qemu_co_queue_empty(&bs->throttled_reqs));
956 /* make a BlockDriverState anonymous by removing from bdrv_state list.
957 Also, NULL terminate the device_name to prevent double remove */
958 void bdrv_make_anon(BlockDriverState *bs)
960 if (bs->device_name[0] != '\0') {
961 QTAILQ_REMOVE(&bdrv_states, bs, list);
963 bs->device_name[0] = '\0';
966 static void bdrv_rebind(BlockDriverState *bs)
968 if (bs->drv && bs->drv->bdrv_rebind) {
969 bs->drv->bdrv_rebind(bs);
974 * Add new bs contents at the top of an image chain while the chain is
975 * live, while keeping required fields on the top layer.
977 * This will modify the BlockDriverState fields, and swap contents
978 * between bs_new and bs_top. Both bs_new and bs_top are modified.
980 * bs_new is required to be anonymous.
982 * This function does not create any image files.
984 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
986 BlockDriverState tmp;
988 /* bs_new must be anonymous */
989 assert(bs_new->device_name[0] == '\0');
993 /* there are some fields that need to stay on the top layer: */
994 tmp.open_flags = bs_top->open_flags;
997 tmp.dev_ops = bs_top->dev_ops;
998 tmp.dev_opaque = bs_top->dev_opaque;
999 tmp.dev = bs_top->dev;
1000 tmp.buffer_alignment = bs_top->buffer_alignment;
1001 tmp.copy_on_read = bs_top->copy_on_read;
1003 /* i/o timing parameters */
1004 tmp.slice_time = bs_top->slice_time;
1005 tmp.slice_start = bs_top->slice_start;
1006 tmp.slice_end = bs_top->slice_end;
1007 tmp.io_limits = bs_top->io_limits;
1008 tmp.io_base = bs_top->io_base;
1009 tmp.throttled_reqs = bs_top->throttled_reqs;
1010 tmp.block_timer = bs_top->block_timer;
1011 tmp.io_limits_enabled = bs_top->io_limits_enabled;
1014 tmp.cyls = bs_top->cyls;
1015 tmp.heads = bs_top->heads;
1016 tmp.secs = bs_top->secs;
1017 tmp.translation = bs_top->translation;
1020 tmp.on_read_error = bs_top->on_read_error;
1021 tmp.on_write_error = bs_top->on_write_error;
1024 tmp.iostatus_enabled = bs_top->iostatus_enabled;
1025 tmp.iostatus = bs_top->iostatus;
1027 /* keep the same entry in bdrv_states */
1028 pstrcpy(tmp.device_name, sizeof(tmp.device_name), bs_top->device_name);
1029 tmp.list = bs_top->list;
1031 /* The contents of 'tmp' will become bs_top, as we are
1032 * swapping bs_new and bs_top contents. */
1033 tmp.backing_hd = bs_new;
1034 pstrcpy(tmp.backing_file, sizeof(tmp.backing_file), bs_top->filename);
1035 bdrv_get_format(bs_top, tmp.backing_format, sizeof(tmp.backing_format));
1037 /* swap contents of the fixed new bs and the current top */
1041 /* device_name[] was carried over from the old bs_top. bs_new
1042 * shouldn't be in bdrv_states, so we need to make device_name[]
1043 * reflect the anonymity of bs_new
1045 bs_new->device_name[0] = '\0';
1047 /* clear the copied fields in the new backing file */
1048 bdrv_detach_dev(bs_new, bs_new->dev);
1050 qemu_co_queue_init(&bs_new->throttled_reqs);
1051 memset(&bs_new->io_base, 0, sizeof(bs_new->io_base));
1052 memset(&bs_new->io_limits, 0, sizeof(bs_new->io_limits));
1053 bdrv_iostatus_disable(bs_new);
1055 /* we don't use bdrv_io_limits_disable() for this, because we don't want
1056 * to affect or delete the block_timer, as it has been moved to bs_top */
1057 bs_new->io_limits_enabled = false;
1058 bs_new->block_timer = NULL;
1059 bs_new->slice_time = 0;
1060 bs_new->slice_start = 0;
1061 bs_new->slice_end = 0;
1063 bdrv_rebind(bs_new);
1064 bdrv_rebind(bs_top);
1067 void bdrv_delete(BlockDriverState *bs)
1071 assert(!bs->in_use);
1073 /* remove from list, if necessary */
1078 assert(bs != bs_snapshots);
1082 int bdrv_attach_dev(BlockDriverState *bs, void *dev)
1083 /* TODO change to DeviceState *dev when all users are qdevified */
1089 bdrv_iostatus_reset(bs);
1093 /* TODO qdevified devices don't use this, remove when devices are qdevified */
1094 void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev)
1096 if (bdrv_attach_dev(bs, dev) < 0) {
1101 void bdrv_detach_dev(BlockDriverState *bs, void *dev)
1102 /* TODO change to DeviceState *dev when all users are qdevified */
1104 assert(bs->dev == dev);
1107 bs->dev_opaque = NULL;
1108 bs->buffer_alignment = 512;
1111 /* TODO change to return DeviceState * when all users are qdevified */
1112 void *bdrv_get_attached_dev(BlockDriverState *bs)
1117 void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops,
1121 bs->dev_opaque = opaque;
1122 if (bdrv_dev_has_removable_media(bs) && bs == bs_snapshots) {
1123 bs_snapshots = NULL;
1127 void bdrv_emit_qmp_error_event(const BlockDriverState *bdrv,
1128 BlockQMPEventAction action, int is_read)
1131 const char *action_str;
1134 case BDRV_ACTION_REPORT:
1135 action_str = "report";
1137 case BDRV_ACTION_IGNORE:
1138 action_str = "ignore";
1140 case BDRV_ACTION_STOP:
1141 action_str = "stop";
1147 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1150 is_read ? "read" : "write");
1151 monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
1153 qobject_decref(data);
1156 static void bdrv_emit_qmp_eject_event(BlockDriverState *bs, bool ejected)
1160 data = qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }",
1161 bdrv_get_device_name(bs), ejected);
1162 monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED, data);
1164 qobject_decref(data);
1167 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load)
1169 if (bs->dev_ops && bs->dev_ops->change_media_cb) {
1170 bool tray_was_closed = !bdrv_dev_is_tray_open(bs);
1171 bs->dev_ops->change_media_cb(bs->dev_opaque, load);
1172 if (tray_was_closed) {
1174 bdrv_emit_qmp_eject_event(bs, true);
1178 bdrv_emit_qmp_eject_event(bs, false);
1183 bool bdrv_dev_has_removable_media(BlockDriverState *bs)
1185 return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb);
1188 void bdrv_dev_eject_request(BlockDriverState *bs, bool force)
1190 if (bs->dev_ops && bs->dev_ops->eject_request_cb) {
1191 bs->dev_ops->eject_request_cb(bs->dev_opaque, force);
1195 bool bdrv_dev_is_tray_open(BlockDriverState *bs)
1197 if (bs->dev_ops && bs->dev_ops->is_tray_open) {
1198 return bs->dev_ops->is_tray_open(bs->dev_opaque);
1203 static void bdrv_dev_resize_cb(BlockDriverState *bs)
1205 if (bs->dev_ops && bs->dev_ops->resize_cb) {
1206 bs->dev_ops->resize_cb(bs->dev_opaque);
1210 bool bdrv_dev_is_medium_locked(BlockDriverState *bs)
1212 if (bs->dev_ops && bs->dev_ops->is_medium_locked) {
1213 return bs->dev_ops->is_medium_locked(bs->dev_opaque);
1219 * Run consistency checks on an image
1221 * Returns 0 if the check could be completed (it doesn't mean that the image is
1222 * free of errors) or -errno when an internal error occurred. The results of the
1223 * check are stored in res.
1225 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix)
1227 if (bs->drv->bdrv_check == NULL) {
1231 memset(res, 0, sizeof(*res));
1232 return bs->drv->bdrv_check(bs, res, fix);
1235 #define COMMIT_BUF_SECTORS 2048
1237 /* commit COW file into the raw image */
1238 int bdrv_commit(BlockDriverState *bs)
1240 BlockDriver *drv = bs->drv;
1241 BlockDriver *backing_drv;
1242 int64_t sector, total_sectors;
1243 int n, ro, open_flags;
1244 int ret = 0, rw_ret = 0;
1246 char filename[1024];
1247 BlockDriverState *bs_rw, *bs_ro;
1252 if (!bs->backing_hd) {
1256 if (bs->backing_hd->keep_read_only) {
1260 if (bdrv_in_use(bs) || bdrv_in_use(bs->backing_hd)) {
1264 backing_drv = bs->backing_hd->drv;
1265 ro = bs->backing_hd->read_only;
1266 strncpy(filename, bs->backing_hd->filename, sizeof(filename));
1267 open_flags = bs->backing_hd->open_flags;
1271 bdrv_delete(bs->backing_hd);
1272 bs->backing_hd = NULL;
1273 bs_rw = bdrv_new("");
1274 rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR,
1278 /* try to re-open read-only */
1279 bs_ro = bdrv_new("");
1280 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
1284 /* drive not functional anymore */
1288 bs->backing_hd = bs_ro;
1291 bs->backing_hd = bs_rw;
1294 total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
1295 buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
1297 for (sector = 0; sector < total_sectors; sector += n) {
1298 if (bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n)) {
1300 if (bdrv_read(bs, sector, buf, n) != 0) {
1305 if (bdrv_write(bs->backing_hd, sector, buf, n) != 0) {
1312 if (drv->bdrv_make_empty) {
1313 ret = drv->bdrv_make_empty(bs);
1318 * Make sure all data we wrote to the backing device is actually
1322 bdrv_flush(bs->backing_hd);
1329 bdrv_delete(bs->backing_hd);
1330 bs->backing_hd = NULL;
1331 bs_ro = bdrv_new("");
1332 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
1336 /* drive not functional anymore */
1340 bs->backing_hd = bs_ro;
1341 bs->backing_hd->keep_read_only = 0;
1347 int bdrv_commit_all(void)
1349 BlockDriverState *bs;
1351 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1352 int ret = bdrv_commit(bs);
1360 struct BdrvTrackedRequest {
1361 BlockDriverState *bs;
1365 QLIST_ENTRY(BdrvTrackedRequest) list;
1366 Coroutine *co; /* owner, used for deadlock detection */
1367 CoQueue wait_queue; /* coroutines blocked on this request */
1371 * Remove an active request from the tracked requests list
1373 * This function should be called when a tracked request is completing.
1375 static void tracked_request_end(BdrvTrackedRequest *req)
1377 QLIST_REMOVE(req, list);
1378 qemu_co_queue_restart_all(&req->wait_queue);
1382 * Add an active request to the tracked requests list
1384 static void tracked_request_begin(BdrvTrackedRequest *req,
1385 BlockDriverState *bs,
1387 int nb_sectors, bool is_write)
1389 *req = (BdrvTrackedRequest){
1391 .sector_num = sector_num,
1392 .nb_sectors = nb_sectors,
1393 .is_write = is_write,
1394 .co = qemu_coroutine_self(),
1397 qemu_co_queue_init(&req->wait_queue);
1399 QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
1403 * Round a region to cluster boundaries
1405 static void round_to_clusters(BlockDriverState *bs,
1406 int64_t sector_num, int nb_sectors,
1407 int64_t *cluster_sector_num,
1408 int *cluster_nb_sectors)
1410 BlockDriverInfo bdi;
1412 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
1413 *cluster_sector_num = sector_num;
1414 *cluster_nb_sectors = nb_sectors;
1416 int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE;
1417 *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c);
1418 *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num +
1423 static bool tracked_request_overlaps(BdrvTrackedRequest *req,
1424 int64_t sector_num, int nb_sectors) {
1426 if (sector_num >= req->sector_num + req->nb_sectors) {
1430 if (req->sector_num >= sector_num + nb_sectors) {
1436 static void coroutine_fn wait_for_overlapping_requests(BlockDriverState *bs,
1437 int64_t sector_num, int nb_sectors)
1439 BdrvTrackedRequest *req;
1440 int64_t cluster_sector_num;
1441 int cluster_nb_sectors;
1444 /* If we touch the same cluster it counts as an overlap. This guarantees
1445 * that allocating writes will be serialized and not race with each other
1446 * for the same cluster. For example, in copy-on-read it ensures that the
1447 * CoR read and write operations are atomic and guest writes cannot
1448 * interleave between them.
1450 round_to_clusters(bs, sector_num, nb_sectors,
1451 &cluster_sector_num, &cluster_nb_sectors);
1455 QLIST_FOREACH(req, &bs->tracked_requests, list) {
1456 if (tracked_request_overlaps(req, cluster_sector_num,
1457 cluster_nb_sectors)) {
1458 /* Hitting this means there was a reentrant request, for
1459 * example, a block driver issuing nested requests. This must
1460 * never happen since it means deadlock.
1462 assert(qemu_coroutine_self() != req->co);
1464 qemu_co_queue_wait(&req->wait_queue);
1475 * -EINVAL - backing format specified, but no file
1476 * -ENOSPC - can't update the backing file because no space is left in the
1478 * -ENOTSUP - format driver doesn't support changing the backing file
1480 int bdrv_change_backing_file(BlockDriverState *bs,
1481 const char *backing_file, const char *backing_fmt)
1483 BlockDriver *drv = bs->drv;
1486 /* Backing file format doesn't make sense without a backing file */
1487 if (backing_fmt && !backing_file) {
1491 if (drv->bdrv_change_backing_file != NULL) {
1492 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
1498 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
1499 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
1504 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
1509 if (!bdrv_is_inserted(bs))
1515 len = bdrv_getlength(bs);
1520 if ((offset > len) || (len - offset < size))
1526 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
1529 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
1530 nb_sectors * BDRV_SECTOR_SIZE);
1533 typedef struct RwCo {
1534 BlockDriverState *bs;
1542 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
1544 RwCo *rwco = opaque;
1546 if (!rwco->is_write) {
1547 rwco->ret = bdrv_co_do_readv(rwco->bs, rwco->sector_num,
1548 rwco->nb_sectors, rwco->qiov, 0);
1550 rwco->ret = bdrv_co_do_writev(rwco->bs, rwco->sector_num,
1551 rwco->nb_sectors, rwco->qiov, 0);
1556 * Process a synchronous request using coroutines
1558 static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf,
1559 int nb_sectors, bool is_write)
1562 struct iovec iov = {
1563 .iov_base = (void *)buf,
1564 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
1569 .sector_num = sector_num,
1570 .nb_sectors = nb_sectors,
1572 .is_write = is_write,
1576 qemu_iovec_init_external(&qiov, &iov, 1);
1579 * In sync call context, when the vcpu is blocked, this throttling timer
1580 * will not fire; so the I/O throttling function has to be disabled here
1581 * if it has been enabled.
1583 if (bs->io_limits_enabled) {
1584 fprintf(stderr, "Disabling I/O throttling on '%s' due "
1585 "to synchronous I/O.\n", bdrv_get_device_name(bs));
1586 bdrv_io_limits_disable(bs);
1589 if (qemu_in_coroutine()) {
1590 /* Fast-path if already in coroutine context */
1591 bdrv_rw_co_entry(&rwco);
1593 co = qemu_coroutine_create(bdrv_rw_co_entry);
1594 qemu_coroutine_enter(co, &rwco);
1595 while (rwco.ret == NOT_DONE) {
1602 /* return < 0 if error. See bdrv_write() for the return codes */
1603 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
1604 uint8_t *buf, int nb_sectors)
1606 return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false);
1609 #define BITS_PER_LONG (sizeof(unsigned long) * 8)
1611 static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
1612 int nb_sectors, int dirty)
1615 unsigned long val, idx, bit;
1617 start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
1618 end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
1620 for (; start <= end; start++) {
1621 idx = start / BITS_PER_LONG;
1622 bit = start % BITS_PER_LONG;
1623 val = bs->dirty_bitmap[idx];
1625 if (!(val & (1UL << bit))) {
1630 if (val & (1UL << bit)) {
1632 val &= ~(1UL << bit);
1635 bs->dirty_bitmap[idx] = val;
1639 /* Return < 0 if error. Important errors are:
1640 -EIO generic I/O error (may happen for all errors)
1641 -ENOMEDIUM No media inserted.
1642 -EINVAL Invalid sector number or nb_sectors
1643 -EACCES Trying to write a read-only device
1645 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
1646 const uint8_t *buf, int nb_sectors)
1648 return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true);
1651 int bdrv_pread(BlockDriverState *bs, int64_t offset,
1652 void *buf, int count1)
1654 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1655 int len, nb_sectors, count;
1660 /* first read to align to sector start */
1661 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1664 sector_num = offset >> BDRV_SECTOR_BITS;
1666 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1668 memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
1676 /* read the sectors "in place" */
1677 nb_sectors = count >> BDRV_SECTOR_BITS;
1678 if (nb_sectors > 0) {
1679 if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
1681 sector_num += nb_sectors;
1682 len = nb_sectors << BDRV_SECTOR_BITS;
1687 /* add data from the last sector */
1689 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1691 memcpy(buf, tmp_buf, count);
1696 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
1697 const void *buf, int count1)
1699 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1700 int len, nb_sectors, count;
1705 /* first write to align to sector start */
1706 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1709 sector_num = offset >> BDRV_SECTOR_BITS;
1711 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1713 memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
1714 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1723 /* write the sectors "in place" */
1724 nb_sectors = count >> BDRV_SECTOR_BITS;
1725 if (nb_sectors > 0) {
1726 if ((ret = bdrv_write(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(tmp_buf, buf, count);
1739 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1746 * Writes to the file and ensures that no writes are reordered across this
1747 * request (acts as a barrier)
1749 * Returns 0 on success, -errno in error cases.
1751 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
1752 const void *buf, int count)
1756 ret = bdrv_pwrite(bs, offset, buf, count);
1761 /* No flush needed for cache modes that use O_DSYNC */
1762 if ((bs->open_flags & BDRV_O_CACHE_WB) != 0) {
1769 static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs,
1770 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1772 /* Perform I/O through a temporary buffer so that users who scribble over
1773 * their read buffer while the operation is in progress do not end up
1774 * modifying the image file. This is critical for zero-copy guest I/O
1775 * where anything might happen inside guest memory.
1777 void *bounce_buffer;
1779 BlockDriver *drv = bs->drv;
1781 QEMUIOVector bounce_qiov;
1782 int64_t cluster_sector_num;
1783 int cluster_nb_sectors;
1787 /* Cover entire cluster so no additional backing file I/O is required when
1788 * allocating cluster in the image file.
1790 round_to_clusters(bs, sector_num, nb_sectors,
1791 &cluster_sector_num, &cluster_nb_sectors);
1793 trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors,
1794 cluster_sector_num, cluster_nb_sectors);
1796 iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE;
1797 iov.iov_base = bounce_buffer = qemu_blockalign(bs, iov.iov_len);
1798 qemu_iovec_init_external(&bounce_qiov, &iov, 1);
1800 ret = drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors,
1806 if (drv->bdrv_co_write_zeroes &&
1807 buffer_is_zero(bounce_buffer, iov.iov_len)) {
1808 ret = bdrv_co_do_write_zeroes(bs, cluster_sector_num,
1809 cluster_nb_sectors);
1811 ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors,
1816 /* It might be okay to ignore write errors for guest requests. If this
1817 * is a deliberate copy-on-read then we don't want to ignore the error.
1818 * Simply report it in all cases.
1823 skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE;
1824 qemu_iovec_from_buffer(qiov, bounce_buffer + skip_bytes,
1825 nb_sectors * BDRV_SECTOR_SIZE);
1828 qemu_vfree(bounce_buffer);
1833 * Handle a read request in coroutine context
1835 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
1836 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1837 BdrvRequestFlags flags)
1839 BlockDriver *drv = bs->drv;
1840 BdrvTrackedRequest req;
1846 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
1850 /* throttling disk read I/O */
1851 if (bs->io_limits_enabled) {
1852 bdrv_io_limits_intercept(bs, false, nb_sectors);
1855 if (bs->copy_on_read) {
1856 flags |= BDRV_REQ_COPY_ON_READ;
1858 if (flags & BDRV_REQ_COPY_ON_READ) {
1859 bs->copy_on_read_in_flight++;
1862 if (bs->copy_on_read_in_flight) {
1863 wait_for_overlapping_requests(bs, sector_num, nb_sectors);
1866 tracked_request_begin(&req, bs, sector_num, nb_sectors, false);
1868 if (flags & BDRV_REQ_COPY_ON_READ) {
1871 ret = bdrv_co_is_allocated(bs, sector_num, nb_sectors, &pnum);
1876 if (!ret || pnum != nb_sectors) {
1877 ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov);
1882 ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
1885 tracked_request_end(&req);
1887 if (flags & BDRV_REQ_COPY_ON_READ) {
1888 bs->copy_on_read_in_flight--;
1894 int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
1895 int nb_sectors, QEMUIOVector *qiov)
1897 trace_bdrv_co_readv(bs, sector_num, nb_sectors);
1899 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0);
1902 int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
1903 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1905 trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors);
1907 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
1908 BDRV_REQ_COPY_ON_READ);
1911 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
1912 int64_t sector_num, int nb_sectors)
1914 BlockDriver *drv = bs->drv;
1919 /* TODO Emulate only part of misaligned requests instead of letting block
1920 * drivers return -ENOTSUP and emulate everything */
1922 /* First try the efficient write zeroes operation */
1923 if (drv->bdrv_co_write_zeroes) {
1924 ret = drv->bdrv_co_write_zeroes(bs, sector_num, nb_sectors);
1925 if (ret != -ENOTSUP) {
1930 /* Fall back to bounce buffer if write zeroes is unsupported */
1931 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
1932 iov.iov_base = qemu_blockalign(bs, iov.iov_len);
1933 memset(iov.iov_base, 0, iov.iov_len);
1934 qemu_iovec_init_external(&qiov, &iov, 1);
1936 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, &qiov);
1938 qemu_vfree(iov.iov_base);
1943 * Handle a write request in coroutine context
1945 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
1946 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1947 BdrvRequestFlags flags)
1949 BlockDriver *drv = bs->drv;
1950 BdrvTrackedRequest req;
1956 if (bs->read_only) {
1959 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
1963 /* throttling disk write I/O */
1964 if (bs->io_limits_enabled) {
1965 bdrv_io_limits_intercept(bs, true, nb_sectors);
1968 if (bs->copy_on_read_in_flight) {
1969 wait_for_overlapping_requests(bs, sector_num, nb_sectors);
1972 tracked_request_begin(&req, bs, sector_num, nb_sectors, true);
1974 if (flags & BDRV_REQ_ZERO_WRITE) {
1975 ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors);
1977 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
1980 if (bs->dirty_bitmap) {
1981 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1984 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
1985 bs->wr_highest_sector = sector_num + nb_sectors - 1;
1988 tracked_request_end(&req);
1993 int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
1994 int nb_sectors, QEMUIOVector *qiov)
1996 trace_bdrv_co_writev(bs, sector_num, nb_sectors);
1998 return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0);
2001 int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs,
2002 int64_t sector_num, int nb_sectors)
2004 trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors);
2006 return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL,
2007 BDRV_REQ_ZERO_WRITE);
2011 * Truncate file to 'offset' bytes (needed only for file protocols)
2013 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
2015 BlockDriver *drv = bs->drv;
2019 if (!drv->bdrv_truncate)
2023 if (bdrv_in_use(bs))
2025 ret = drv->bdrv_truncate(bs, offset);
2027 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
2028 bdrv_dev_resize_cb(bs);
2034 * Length of a allocated file in bytes. Sparse files are counted by actual
2035 * allocated space. Return < 0 if error or unknown.
2037 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
2039 BlockDriver *drv = bs->drv;
2043 if (drv->bdrv_get_allocated_file_size) {
2044 return drv->bdrv_get_allocated_file_size(bs);
2047 return bdrv_get_allocated_file_size(bs->file);
2053 * Length of a file in bytes. Return < 0 if error or unknown.
2055 int64_t bdrv_getlength(BlockDriverState *bs)
2057 BlockDriver *drv = bs->drv;
2061 if (bs->growable || bdrv_dev_has_removable_media(bs)) {
2062 if (drv->bdrv_getlength) {
2063 return drv->bdrv_getlength(bs);
2066 return bs->total_sectors * BDRV_SECTOR_SIZE;
2069 /* return 0 as number of sectors if no device present or error */
2070 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
2073 length = bdrv_getlength(bs);
2077 length = length >> BDRV_SECTOR_BITS;
2078 *nb_sectors_ptr = length;
2082 uint8_t boot_ind; /* 0x80 - active */
2083 uint8_t head; /* starting head */
2084 uint8_t sector; /* starting sector */
2085 uint8_t cyl; /* starting cylinder */
2086 uint8_t sys_ind; /* What partition type */
2087 uint8_t end_head; /* end head */
2088 uint8_t end_sector; /* end sector */
2089 uint8_t end_cyl; /* end cylinder */
2090 uint32_t start_sect; /* starting sector counting from 0 */
2091 uint32_t nr_sects; /* nr of sectors in partition */
2094 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
2095 static int guess_disk_lchs(BlockDriverState *bs,
2096 int *pcylinders, int *pheads, int *psectors)
2098 uint8_t buf[BDRV_SECTOR_SIZE];
2099 int ret, i, heads, sectors, cylinders;
2100 struct partition *p;
2102 uint64_t nb_sectors;
2105 bdrv_get_geometry(bs, &nb_sectors);
2108 * The function will be invoked during startup not only in sync I/O mode,
2109 * but also in async I/O mode. So the I/O throttling function has to
2110 * be disabled temporarily here, not permanently.
2112 enabled = bs->io_limits_enabled;
2113 bs->io_limits_enabled = false;
2114 ret = bdrv_read(bs, 0, buf, 1);
2115 bs->io_limits_enabled = enabled;
2118 /* test msdos magic */
2119 if (buf[510] != 0x55 || buf[511] != 0xaa)
2121 for(i = 0; i < 4; i++) {
2122 p = ((struct partition *)(buf + 0x1be)) + i;
2123 nr_sects = le32_to_cpu(p->nr_sects);
2124 if (nr_sects && p->end_head) {
2125 /* We make the assumption that the partition terminates on
2126 a cylinder boundary */
2127 heads = p->end_head + 1;
2128 sectors = p->end_sector & 63;
2131 cylinders = nb_sectors / (heads * sectors);
2132 if (cylinders < 1 || cylinders > 16383)
2135 *psectors = sectors;
2136 *pcylinders = cylinders;
2138 printf("guessed geometry: LCHS=%d %d %d\n",
2139 cylinders, heads, sectors);
2147 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
2149 int translation, lba_detected = 0;
2150 int cylinders, heads, secs;
2151 uint64_t nb_sectors;
2153 /* if a geometry hint is available, use it */
2154 bdrv_get_geometry(bs, &nb_sectors);
2155 bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
2156 translation = bdrv_get_translation_hint(bs);
2157 if (cylinders != 0) {
2162 if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
2164 /* if heads > 16, it means that a BIOS LBA
2165 translation was active, so the default
2166 hardware geometry is OK */
2168 goto default_geometry;
2173 /* disable any translation to be in sync with
2174 the logical geometry */
2175 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
2176 bdrv_set_translation_hint(bs,
2177 BIOS_ATA_TRANSLATION_NONE);
2182 /* if no geometry, use a standard physical disk geometry */
2183 cylinders = nb_sectors / (16 * 63);
2185 if (cylinders > 16383)
2187 else if (cylinders < 2)
2192 if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
2193 if ((*pcyls * *pheads) <= 131072) {
2194 bdrv_set_translation_hint(bs,
2195 BIOS_ATA_TRANSLATION_LARGE);
2197 bdrv_set_translation_hint(bs,
2198 BIOS_ATA_TRANSLATION_LBA);
2202 bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
2206 void bdrv_set_geometry_hint(BlockDriverState *bs,
2207 int cyls, int heads, int secs)
2214 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
2216 bs->translation = translation;
2219 void bdrv_get_geometry_hint(BlockDriverState *bs,
2220 int *pcyls, int *pheads, int *psecs)
2223 *pheads = bs->heads;
2227 /* throttling disk io limits */
2228 void bdrv_set_io_limits(BlockDriverState *bs,
2229 BlockIOLimit *io_limits)
2231 bs->io_limits = *io_limits;
2232 bs->io_limits_enabled = bdrv_io_limits_enabled(bs);
2235 /* Recognize floppy formats */
2236 typedef struct FDFormat {
2244 static const FDFormat fd_formats[] = {
2245 /* First entry is default format */
2246 /* 1.44 MB 3"1/2 floppy disks */
2247 { FDRIVE_DRV_144, 18, 80, 1, FDRIVE_RATE_500K, },
2248 { FDRIVE_DRV_144, 20, 80, 1, FDRIVE_RATE_500K, },
2249 { FDRIVE_DRV_144, 21, 80, 1, FDRIVE_RATE_500K, },
2250 { FDRIVE_DRV_144, 21, 82, 1, FDRIVE_RATE_500K, },
2251 { FDRIVE_DRV_144, 21, 83, 1, FDRIVE_RATE_500K, },
2252 { FDRIVE_DRV_144, 22, 80, 1, FDRIVE_RATE_500K, },
2253 { FDRIVE_DRV_144, 23, 80, 1, FDRIVE_RATE_500K, },
2254 { FDRIVE_DRV_144, 24, 80, 1, FDRIVE_RATE_500K, },
2255 /* 2.88 MB 3"1/2 floppy disks */
2256 { FDRIVE_DRV_288, 36, 80, 1, FDRIVE_RATE_1M, },
2257 { FDRIVE_DRV_288, 39, 80, 1, FDRIVE_RATE_1M, },
2258 { FDRIVE_DRV_288, 40, 80, 1, FDRIVE_RATE_1M, },
2259 { FDRIVE_DRV_288, 44, 80, 1, FDRIVE_RATE_1M, },
2260 { FDRIVE_DRV_288, 48, 80, 1, FDRIVE_RATE_1M, },
2261 /* 720 kB 3"1/2 floppy disks */
2262 { FDRIVE_DRV_144, 9, 80, 1, FDRIVE_RATE_250K, },
2263 { FDRIVE_DRV_144, 10, 80, 1, FDRIVE_RATE_250K, },
2264 { FDRIVE_DRV_144, 10, 82, 1, FDRIVE_RATE_250K, },
2265 { FDRIVE_DRV_144, 10, 83, 1, FDRIVE_RATE_250K, },
2266 { FDRIVE_DRV_144, 13, 80, 1, FDRIVE_RATE_250K, },
2267 { FDRIVE_DRV_144, 14, 80, 1, FDRIVE_RATE_250K, },
2268 /* 1.2 MB 5"1/4 floppy disks */
2269 { FDRIVE_DRV_120, 15, 80, 1, FDRIVE_RATE_500K, },
2270 { FDRIVE_DRV_120, 18, 80, 1, FDRIVE_RATE_500K, },
2271 { FDRIVE_DRV_120, 18, 82, 1, FDRIVE_RATE_500K, },
2272 { FDRIVE_DRV_120, 18, 83, 1, FDRIVE_RATE_500K, },
2273 { FDRIVE_DRV_120, 20, 80, 1, FDRIVE_RATE_500K, },
2274 /* 720 kB 5"1/4 floppy disks */
2275 { FDRIVE_DRV_120, 9, 80, 1, FDRIVE_RATE_250K, },
2276 { FDRIVE_DRV_120, 11, 80, 1, FDRIVE_RATE_250K, },
2277 /* 360 kB 5"1/4 floppy disks */
2278 { FDRIVE_DRV_120, 9, 40, 1, FDRIVE_RATE_300K, },
2279 { FDRIVE_DRV_120, 9, 40, 0, FDRIVE_RATE_300K, },
2280 { FDRIVE_DRV_120, 10, 41, 1, FDRIVE_RATE_300K, },
2281 { FDRIVE_DRV_120, 10, 42, 1, FDRIVE_RATE_300K, },
2282 /* 320 kB 5"1/4 floppy disks */
2283 { FDRIVE_DRV_120, 8, 40, 1, FDRIVE_RATE_250K, },
2284 { FDRIVE_DRV_120, 8, 40, 0, FDRIVE_RATE_250K, },
2285 /* 360 kB must match 5"1/4 better than 3"1/2... */
2286 { FDRIVE_DRV_144, 9, 80, 0, FDRIVE_RATE_250K, },
2288 { FDRIVE_DRV_NONE, -1, -1, 0, 0, },
2291 void bdrv_get_floppy_geometry_hint(BlockDriverState *bs, int *nb_heads,
2292 int *max_track, int *last_sect,
2293 FDriveType drive_in, FDriveType *drive,
2296 const FDFormat *parse;
2297 uint64_t nb_sectors, size;
2298 int i, first_match, match;
2300 bdrv_get_geometry_hint(bs, nb_heads, max_track, last_sect);
2301 if (*nb_heads != 0 && *max_track != 0 && *last_sect != 0) {
2302 /* User defined disk */
2303 *rate = FDRIVE_RATE_500K;
2305 bdrv_get_geometry(bs, &nb_sectors);
2308 for (i = 0; ; i++) {
2309 parse = &fd_formats[i];
2310 if (parse->drive == FDRIVE_DRV_NONE) {
2313 if (drive_in == parse->drive ||
2314 drive_in == FDRIVE_DRV_NONE) {
2315 size = (parse->max_head + 1) * parse->max_track *
2317 if (nb_sectors == size) {
2321 if (first_match == -1) {
2327 if (first_match == -1) {
2330 match = first_match;
2332 parse = &fd_formats[match];
2334 *nb_heads = parse->max_head + 1;
2335 *max_track = parse->max_track;
2336 *last_sect = parse->last_sect;
2337 *drive = parse->drive;
2338 *rate = parse->rate;
2342 int bdrv_get_translation_hint(BlockDriverState *bs)
2344 return bs->translation;
2347 void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
2348 BlockErrorAction on_write_error)
2350 bs->on_read_error = on_read_error;
2351 bs->on_write_error = on_write_error;
2354 BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
2356 return is_read ? bs->on_read_error : bs->on_write_error;
2359 int bdrv_is_read_only(BlockDriverState *bs)
2361 return bs->read_only;
2364 int bdrv_is_sg(BlockDriverState *bs)
2369 int bdrv_enable_write_cache(BlockDriverState *bs)
2371 return bs->enable_write_cache;
2374 int bdrv_is_encrypted(BlockDriverState *bs)
2376 if (bs->backing_hd && bs->backing_hd->encrypted)
2378 return bs->encrypted;
2381 int bdrv_key_required(BlockDriverState *bs)
2383 BlockDriverState *backing_hd = bs->backing_hd;
2385 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
2387 return (bs->encrypted && !bs->valid_key);
2390 int bdrv_set_key(BlockDriverState *bs, const char *key)
2393 if (bs->backing_hd && bs->backing_hd->encrypted) {
2394 ret = bdrv_set_key(bs->backing_hd, key);
2400 if (!bs->encrypted) {
2402 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
2405 ret = bs->drv->bdrv_set_key(bs, key);
2408 } else if (!bs->valid_key) {
2410 /* call the change callback now, we skipped it on open */
2411 bdrv_dev_change_media_cb(bs, true);
2416 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
2421 pstrcpy(buf, buf_size, bs->drv->format_name);
2425 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
2430 QLIST_FOREACH(drv, &bdrv_drivers, list) {
2431 it(opaque, drv->format_name);
2435 BlockDriverState *bdrv_find(const char *name)
2437 BlockDriverState *bs;
2439 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2440 if (!strcmp(name, bs->device_name)) {
2447 BlockDriverState *bdrv_next(BlockDriverState *bs)
2450 return QTAILQ_FIRST(&bdrv_states);
2452 return QTAILQ_NEXT(bs, list);
2455 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
2457 BlockDriverState *bs;
2459 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2464 const char *bdrv_get_device_name(BlockDriverState *bs)
2466 return bs->device_name;
2469 void bdrv_flush_all(void)
2471 BlockDriverState *bs;
2473 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2478 int bdrv_has_zero_init(BlockDriverState *bs)
2482 if (bs->drv->bdrv_has_zero_init) {
2483 return bs->drv->bdrv_has_zero_init(bs);
2489 typedef struct BdrvCoIsAllocatedData {
2490 BlockDriverState *bs;
2496 } BdrvCoIsAllocatedData;
2499 * Returns true iff the specified sector is present in the disk image. Drivers
2500 * not implementing the functionality are assumed to not support backing files,
2501 * hence all their sectors are reported as allocated.
2503 * If 'sector_num' is beyond the end of the disk image the return value is 0
2504 * and 'pnum' is set to 0.
2506 * 'pnum' is set to the number of sectors (including and immediately following
2507 * the specified sector) that are known to be in the same
2508 * allocated/unallocated state.
2510 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
2511 * beyond the end of the disk image it will be clamped.
2513 int coroutine_fn bdrv_co_is_allocated(BlockDriverState *bs, int64_t sector_num,
2514 int nb_sectors, int *pnum)
2518 if (sector_num >= bs->total_sectors) {
2523 n = bs->total_sectors - sector_num;
2524 if (n < nb_sectors) {
2528 if (!bs->drv->bdrv_co_is_allocated) {
2533 return bs->drv->bdrv_co_is_allocated(bs, sector_num, nb_sectors, pnum);
2536 /* Coroutine wrapper for bdrv_is_allocated() */
2537 static void coroutine_fn bdrv_is_allocated_co_entry(void *opaque)
2539 BdrvCoIsAllocatedData *data = opaque;
2540 BlockDriverState *bs = data->bs;
2542 data->ret = bdrv_co_is_allocated(bs, data->sector_num, data->nb_sectors,
2548 * Synchronous wrapper around bdrv_co_is_allocated().
2550 * See bdrv_co_is_allocated() for details.
2552 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
2556 BdrvCoIsAllocatedData data = {
2558 .sector_num = sector_num,
2559 .nb_sectors = nb_sectors,
2564 co = qemu_coroutine_create(bdrv_is_allocated_co_entry);
2565 qemu_coroutine_enter(co, &data);
2566 while (!data.done) {
2573 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
2575 * Return true if the given sector is allocated in any image between
2576 * BASE and TOP (inclusive). BASE can be NULL to check if the given
2577 * sector is allocated in any image of the chain. Return false otherwise.
2579 * 'pnum' is set to the number of sectors (including and immediately following
2580 * the specified sector) that are known to be in the same
2581 * allocated/unallocated state.
2584 int coroutine_fn bdrv_co_is_allocated_above(BlockDriverState *top,
2585 BlockDriverState *base,
2587 int nb_sectors, int *pnum)
2589 BlockDriverState *intermediate;
2590 int ret, n = nb_sectors;
2593 while (intermediate && intermediate != base) {
2595 ret = bdrv_co_is_allocated(intermediate, sector_num, nb_sectors,
2605 * [sector_num, nb_sectors] is unallocated on top but intermediate
2608 * [sector_num+x, nr_sectors] allocated.
2610 if (n > pnum_inter) {
2614 intermediate = intermediate->backing_hd;
2621 BlockInfoList *qmp_query_block(Error **errp)
2623 BlockInfoList *head = NULL, *cur_item = NULL;
2624 BlockDriverState *bs;
2626 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2627 BlockInfoList *info = g_malloc0(sizeof(*info));
2629 info->value = g_malloc0(sizeof(*info->value));
2630 info->value->device = g_strdup(bs->device_name);
2631 info->value->type = g_strdup("unknown");
2632 info->value->locked = bdrv_dev_is_medium_locked(bs);
2633 info->value->removable = bdrv_dev_has_removable_media(bs);
2635 if (bdrv_dev_has_removable_media(bs)) {
2636 info->value->has_tray_open = true;
2637 info->value->tray_open = bdrv_dev_is_tray_open(bs);
2640 if (bdrv_iostatus_is_enabled(bs)) {
2641 info->value->has_io_status = true;
2642 info->value->io_status = bs->iostatus;
2646 info->value->has_inserted = true;
2647 info->value->inserted = g_malloc0(sizeof(*info->value->inserted));
2648 info->value->inserted->file = g_strdup(bs->filename);
2649 info->value->inserted->ro = bs->read_only;
2650 info->value->inserted->drv = g_strdup(bs->drv->format_name);
2651 info->value->inserted->encrypted = bs->encrypted;
2652 if (bs->backing_file[0]) {
2653 info->value->inserted->has_backing_file = true;
2654 info->value->inserted->backing_file = g_strdup(bs->backing_file);
2657 if (bs->io_limits_enabled) {
2658 info->value->inserted->bps =
2659 bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL];
2660 info->value->inserted->bps_rd =
2661 bs->io_limits.bps[BLOCK_IO_LIMIT_READ];
2662 info->value->inserted->bps_wr =
2663 bs->io_limits.bps[BLOCK_IO_LIMIT_WRITE];
2664 info->value->inserted->iops =
2665 bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL];
2666 info->value->inserted->iops_rd =
2667 bs->io_limits.iops[BLOCK_IO_LIMIT_READ];
2668 info->value->inserted->iops_wr =
2669 bs->io_limits.iops[BLOCK_IO_LIMIT_WRITE];
2673 /* XXX: waiting for the qapi to support GSList */
2675 head = cur_item = info;
2677 cur_item->next = info;
2685 /* Consider exposing this as a full fledged QMP command */
2686 static BlockStats *qmp_query_blockstat(const BlockDriverState *bs, Error **errp)
2690 s = g_malloc0(sizeof(*s));
2692 if (bs->device_name[0]) {
2693 s->has_device = true;
2694 s->device = g_strdup(bs->device_name);
2697 s->stats = g_malloc0(sizeof(*s->stats));
2698 s->stats->rd_bytes = bs->nr_bytes[BDRV_ACCT_READ];
2699 s->stats->wr_bytes = bs->nr_bytes[BDRV_ACCT_WRITE];
2700 s->stats->rd_operations = bs->nr_ops[BDRV_ACCT_READ];
2701 s->stats->wr_operations = bs->nr_ops[BDRV_ACCT_WRITE];
2702 s->stats->wr_highest_offset = bs->wr_highest_sector * BDRV_SECTOR_SIZE;
2703 s->stats->flush_operations = bs->nr_ops[BDRV_ACCT_FLUSH];
2704 s->stats->wr_total_time_ns = bs->total_time_ns[BDRV_ACCT_WRITE];
2705 s->stats->rd_total_time_ns = bs->total_time_ns[BDRV_ACCT_READ];
2706 s->stats->flush_total_time_ns = bs->total_time_ns[BDRV_ACCT_FLUSH];
2709 s->has_parent = true;
2710 s->parent = qmp_query_blockstat(bs->file, NULL);
2716 BlockStatsList *qmp_query_blockstats(Error **errp)
2718 BlockStatsList *head = NULL, *cur_item = NULL;
2719 BlockDriverState *bs;
2721 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2722 BlockStatsList *info = g_malloc0(sizeof(*info));
2723 info->value = qmp_query_blockstat(bs, NULL);
2725 /* XXX: waiting for the qapi to support GSList */
2727 head = cur_item = info;
2729 cur_item->next = info;
2737 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
2739 if (bs->backing_hd && bs->backing_hd->encrypted)
2740 return bs->backing_file;
2741 else if (bs->encrypted)
2742 return bs->filename;
2747 void bdrv_get_backing_filename(BlockDriverState *bs,
2748 char *filename, int filename_size)
2750 pstrcpy(filename, filename_size, bs->backing_file);
2753 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
2754 const uint8_t *buf, int nb_sectors)
2756 BlockDriver *drv = bs->drv;
2759 if (!drv->bdrv_write_compressed)
2761 if (bdrv_check_request(bs, sector_num, nb_sectors))
2764 if (bs->dirty_bitmap) {
2765 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
2768 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
2771 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2773 BlockDriver *drv = bs->drv;
2776 if (!drv->bdrv_get_info)
2778 memset(bdi, 0, sizeof(*bdi));
2779 return drv->bdrv_get_info(bs, bdi);
2782 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
2783 int64_t pos, int size)
2785 BlockDriver *drv = bs->drv;
2788 if (drv->bdrv_save_vmstate)
2789 return drv->bdrv_save_vmstate(bs, buf, pos, size);
2791 return bdrv_save_vmstate(bs->file, buf, pos, size);
2795 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
2796 int64_t pos, int size)
2798 BlockDriver *drv = bs->drv;
2801 if (drv->bdrv_load_vmstate)
2802 return drv->bdrv_load_vmstate(bs, buf, pos, size);
2804 return bdrv_load_vmstate(bs->file, buf, pos, size);
2808 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
2810 BlockDriver *drv = bs->drv;
2812 if (!drv || !drv->bdrv_debug_event) {
2816 return drv->bdrv_debug_event(bs, event);
2820 /**************************************************************/
2821 /* handling of snapshots */
2823 int bdrv_can_snapshot(BlockDriverState *bs)
2825 BlockDriver *drv = bs->drv;
2826 if (!drv || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
2830 if (!drv->bdrv_snapshot_create) {
2831 if (bs->file != NULL) {
2832 return bdrv_can_snapshot(bs->file);
2840 int bdrv_is_snapshot(BlockDriverState *bs)
2842 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
2845 BlockDriverState *bdrv_snapshots(void)
2847 BlockDriverState *bs;
2850 return bs_snapshots;
2854 while ((bs = bdrv_next(bs))) {
2855 if (bdrv_can_snapshot(bs)) {
2863 int bdrv_snapshot_create(BlockDriverState *bs,
2864 QEMUSnapshotInfo *sn_info)
2866 BlockDriver *drv = bs->drv;
2869 if (drv->bdrv_snapshot_create)
2870 return drv->bdrv_snapshot_create(bs, sn_info);
2872 return bdrv_snapshot_create(bs->file, sn_info);
2876 int bdrv_snapshot_goto(BlockDriverState *bs,
2877 const char *snapshot_id)
2879 BlockDriver *drv = bs->drv;
2884 if (drv->bdrv_snapshot_goto)
2885 return drv->bdrv_snapshot_goto(bs, snapshot_id);
2888 drv->bdrv_close(bs);
2889 ret = bdrv_snapshot_goto(bs->file, snapshot_id);
2890 open_ret = drv->bdrv_open(bs, bs->open_flags);
2892 bdrv_delete(bs->file);
2902 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2904 BlockDriver *drv = bs->drv;
2907 if (drv->bdrv_snapshot_delete)
2908 return drv->bdrv_snapshot_delete(bs, snapshot_id);
2910 return bdrv_snapshot_delete(bs->file, snapshot_id);
2914 int bdrv_snapshot_list(BlockDriverState *bs,
2915 QEMUSnapshotInfo **psn_info)
2917 BlockDriver *drv = bs->drv;
2920 if (drv->bdrv_snapshot_list)
2921 return drv->bdrv_snapshot_list(bs, psn_info);
2923 return bdrv_snapshot_list(bs->file, psn_info);
2927 int bdrv_snapshot_load_tmp(BlockDriverState *bs,
2928 const char *snapshot_name)
2930 BlockDriver *drv = bs->drv;
2934 if (!bs->read_only) {
2937 if (drv->bdrv_snapshot_load_tmp) {
2938 return drv->bdrv_snapshot_load_tmp(bs, snapshot_name);
2943 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
2944 const char *backing_file)
2950 if (bs->backing_hd) {
2951 if (strcmp(bs->backing_file, backing_file) == 0) {
2952 return bs->backing_hd;
2954 return bdrv_find_backing_image(bs->backing_hd, backing_file);
2961 #define NB_SUFFIXES 4
2963 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
2965 static const char suffixes[NB_SUFFIXES] = "KMGT";
2970 snprintf(buf, buf_size, "%" PRId64, size);
2973 for(i = 0; i < NB_SUFFIXES; i++) {
2974 if (size < (10 * base)) {
2975 snprintf(buf, buf_size, "%0.1f%c",
2976 (double)size / base,
2979 } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
2980 snprintf(buf, buf_size, "%" PRId64 "%c",
2981 ((size + (base >> 1)) / base),
2991 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
2993 char buf1[128], date_buf[128], clock_buf[128];
3003 snprintf(buf, buf_size,
3004 "%-10s%-20s%7s%20s%15s",
3005 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
3009 ptm = localtime(&ti);
3010 strftime(date_buf, sizeof(date_buf),
3011 "%Y-%m-%d %H:%M:%S", ptm);
3013 localtime_r(&ti, &tm);
3014 strftime(date_buf, sizeof(date_buf),
3015 "%Y-%m-%d %H:%M:%S", &tm);
3017 secs = sn->vm_clock_nsec / 1000000000;
3018 snprintf(clock_buf, sizeof(clock_buf),
3019 "%02d:%02d:%02d.%03d",
3021 (int)((secs / 60) % 60),
3023 (int)((sn->vm_clock_nsec / 1000000) % 1000));
3024 snprintf(buf, buf_size,
3025 "%-10s%-20s%7s%20s%15s",
3026 sn->id_str, sn->name,
3027 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
3034 /**************************************************************/
3037 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
3038 QEMUIOVector *qiov, int nb_sectors,
3039 BlockDriverCompletionFunc *cb, void *opaque)
3041 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
3043 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
3047 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
3048 QEMUIOVector *qiov, int nb_sectors,
3049 BlockDriverCompletionFunc *cb, void *opaque)
3051 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
3053 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
3058 typedef struct MultiwriteCB {
3063 BlockDriverCompletionFunc *cb;
3065 QEMUIOVector *free_qiov;
3069 static void multiwrite_user_cb(MultiwriteCB *mcb)
3073 for (i = 0; i < mcb->num_callbacks; i++) {
3074 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
3075 if (mcb->callbacks[i].free_qiov) {
3076 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
3078 g_free(mcb->callbacks[i].free_qiov);
3082 static void multiwrite_cb(void *opaque, int ret)
3084 MultiwriteCB *mcb = opaque;
3086 trace_multiwrite_cb(mcb, ret);
3088 if (ret < 0 && !mcb->error) {
3092 mcb->num_requests--;
3093 if (mcb->num_requests == 0) {
3094 multiwrite_user_cb(mcb);
3099 static int multiwrite_req_compare(const void *a, const void *b)
3101 const BlockRequest *req1 = a, *req2 = b;
3104 * Note that we can't simply subtract req2->sector from req1->sector
3105 * here as that could overflow the return value.
3107 if (req1->sector > req2->sector) {
3109 } else if (req1->sector < req2->sector) {
3117 * Takes a bunch of requests and tries to merge them. Returns the number of
3118 * requests that remain after merging.
3120 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
3121 int num_reqs, MultiwriteCB *mcb)
3125 // Sort requests by start sector
3126 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
3128 // Check if adjacent requests touch the same clusters. If so, combine them,
3129 // filling up gaps with zero sectors.
3131 for (i = 1; i < num_reqs; i++) {
3133 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
3135 // Handle exactly sequential writes and overlapping writes.
3136 if (reqs[i].sector <= oldreq_last) {
3140 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
3146 QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
3147 qemu_iovec_init(qiov,
3148 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
3150 // Add the first request to the merged one. If the requests are
3151 // overlapping, drop the last sectors of the first request.
3152 size = (reqs[i].sector - reqs[outidx].sector) << 9;
3153 qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
3155 // We should need to add any zeros between the two requests
3156 assert (reqs[i].sector <= oldreq_last);
3158 // Add the second request
3159 qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
3161 reqs[outidx].nb_sectors = qiov->size >> 9;
3162 reqs[outidx].qiov = qiov;
3164 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
3167 reqs[outidx].sector = reqs[i].sector;
3168 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
3169 reqs[outidx].qiov = reqs[i].qiov;
3177 * Submit multiple AIO write requests at once.
3179 * On success, the function returns 0 and all requests in the reqs array have
3180 * been submitted. In error case this function returns -1, and any of the
3181 * requests may or may not be submitted yet. In particular, this means that the
3182 * callback will be called for some of the requests, for others it won't. The
3183 * caller must check the error field of the BlockRequest to wait for the right
3184 * callbacks (if error != 0, no callback will be called).
3186 * The implementation may modify the contents of the reqs array, e.g. to merge
3187 * requests. However, the fields opaque and error are left unmodified as they
3188 * are used to signal failure for a single request to the caller.
3190 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
3195 /* don't submit writes if we don't have a medium */
3196 if (bs->drv == NULL) {
3197 for (i = 0; i < num_reqs; i++) {
3198 reqs[i].error = -ENOMEDIUM;
3203 if (num_reqs == 0) {
3207 // Create MultiwriteCB structure
3208 mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
3209 mcb->num_requests = 0;
3210 mcb->num_callbacks = num_reqs;
3212 for (i = 0; i < num_reqs; i++) {
3213 mcb->callbacks[i].cb = reqs[i].cb;
3214 mcb->callbacks[i].opaque = reqs[i].opaque;
3217 // Check for mergable requests
3218 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
3220 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
3222 /* Run the aio requests. */
3223 mcb->num_requests = num_reqs;
3224 for (i = 0; i < num_reqs; i++) {
3225 bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
3226 reqs[i].nb_sectors, multiwrite_cb, mcb);
3232 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
3234 acb->pool->cancel(acb);
3237 /* block I/O throttling */
3238 static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors,
3239 bool is_write, double elapsed_time, uint64_t *wait)
3241 uint64_t bps_limit = 0;
3242 double bytes_limit, bytes_base, bytes_res;
3243 double slice_time, wait_time;
3245 if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
3246 bps_limit = bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL];
3247 } else if (bs->io_limits.bps[is_write]) {
3248 bps_limit = bs->io_limits.bps[is_write];
3257 slice_time = bs->slice_end - bs->slice_start;
3258 slice_time /= (NANOSECONDS_PER_SECOND);
3259 bytes_limit = bps_limit * slice_time;
3260 bytes_base = bs->nr_bytes[is_write] - bs->io_base.bytes[is_write];
3261 if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
3262 bytes_base += bs->nr_bytes[!is_write] - bs->io_base.bytes[!is_write];
3265 /* bytes_base: the bytes of data which have been read/written; and
3266 * it is obtained from the history statistic info.
3267 * bytes_res: the remaining bytes of data which need to be read/written.
3268 * (bytes_base + bytes_res) / bps_limit: used to calcuate
3269 * the total time for completing reading/writting all data.
3271 bytes_res = (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
3273 if (bytes_base + bytes_res <= bytes_limit) {
3281 /* Calc approx time to dispatch */
3282 wait_time = (bytes_base + bytes_res) / bps_limit - elapsed_time;
3284 /* When the I/O rate at runtime exceeds the limits,
3285 * bs->slice_end need to be extended in order that the current statistic
3286 * info can be kept until the timer fire, so it is increased and tuned
3287 * based on the result of experiment.
3289 bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10;
3290 bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME;
3292 *wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
3298 static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write,
3299 double elapsed_time, uint64_t *wait)
3301 uint64_t iops_limit = 0;
3302 double ios_limit, ios_base;
3303 double slice_time, wait_time;
3305 if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
3306 iops_limit = bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL];
3307 } else if (bs->io_limits.iops[is_write]) {
3308 iops_limit = bs->io_limits.iops[is_write];
3317 slice_time = bs->slice_end - bs->slice_start;
3318 slice_time /= (NANOSECONDS_PER_SECOND);
3319 ios_limit = iops_limit * slice_time;
3320 ios_base = bs->nr_ops[is_write] - bs->io_base.ios[is_write];
3321 if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
3322 ios_base += bs->nr_ops[!is_write] - bs->io_base.ios[!is_write];
3325 if (ios_base + 1 <= ios_limit) {
3333 /* Calc approx time to dispatch */
3334 wait_time = (ios_base + 1) / iops_limit;
3335 if (wait_time > elapsed_time) {
3336 wait_time = wait_time - elapsed_time;
3341 bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10;
3342 bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME;
3344 *wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
3350 static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors,
3351 bool is_write, int64_t *wait)
3353 int64_t now, max_wait;
3354 uint64_t bps_wait = 0, iops_wait = 0;
3355 double elapsed_time;
3356 int bps_ret, iops_ret;
3358 now = qemu_get_clock_ns(vm_clock);
3359 if ((bs->slice_start < now)
3360 && (bs->slice_end > now)) {
3361 bs->slice_end = now + bs->slice_time;
3363 bs->slice_time = 5 * BLOCK_IO_SLICE_TIME;
3364 bs->slice_start = now;
3365 bs->slice_end = now + bs->slice_time;
3367 bs->io_base.bytes[is_write] = bs->nr_bytes[is_write];
3368 bs->io_base.bytes[!is_write] = bs->nr_bytes[!is_write];
3370 bs->io_base.ios[is_write] = bs->nr_ops[is_write];
3371 bs->io_base.ios[!is_write] = bs->nr_ops[!is_write];
3374 elapsed_time = now - bs->slice_start;
3375 elapsed_time /= (NANOSECONDS_PER_SECOND);
3377 bps_ret = bdrv_exceed_bps_limits(bs, nb_sectors,
3378 is_write, elapsed_time, &bps_wait);
3379 iops_ret = bdrv_exceed_iops_limits(bs, is_write,
3380 elapsed_time, &iops_wait);
3381 if (bps_ret || iops_ret) {
3382 max_wait = bps_wait > iops_wait ? bps_wait : iops_wait;
3387 now = qemu_get_clock_ns(vm_clock);
3388 if (bs->slice_end < now + max_wait) {
3389 bs->slice_end = now + max_wait;
3402 /**************************************************************/
3403 /* async block device emulation */
3405 typedef struct BlockDriverAIOCBSync {
3406 BlockDriverAIOCB common;
3409 /* vector translation state */
3413 } BlockDriverAIOCBSync;
3415 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
3417 BlockDriverAIOCBSync *acb =
3418 container_of(blockacb, BlockDriverAIOCBSync, common);
3419 qemu_bh_delete(acb->bh);
3421 qemu_aio_release(acb);
3424 static AIOPool bdrv_em_aio_pool = {
3425 .aiocb_size = sizeof(BlockDriverAIOCBSync),
3426 .cancel = bdrv_aio_cancel_em,
3429 static void bdrv_aio_bh_cb(void *opaque)
3431 BlockDriverAIOCBSync *acb = opaque;
3434 qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
3435 qemu_vfree(acb->bounce);
3436 acb->common.cb(acb->common.opaque, acb->ret);
3437 qemu_bh_delete(acb->bh);
3439 qemu_aio_release(acb);
3442 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
3446 BlockDriverCompletionFunc *cb,
3451 BlockDriverAIOCBSync *acb;
3453 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
3454 acb->is_write = is_write;
3456 acb->bounce = qemu_blockalign(bs, qiov->size);
3457 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
3460 qemu_iovec_to_buffer(acb->qiov, acb->bounce);
3461 acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
3463 acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
3466 qemu_bh_schedule(acb->bh);
3468 return &acb->common;
3471 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
3472 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
3473 BlockDriverCompletionFunc *cb, void *opaque)
3475 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
3478 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
3479 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
3480 BlockDriverCompletionFunc *cb, void *opaque)
3482 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
3486 typedef struct BlockDriverAIOCBCoroutine {
3487 BlockDriverAIOCB common;
3491 } BlockDriverAIOCBCoroutine;
3493 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb)
3498 static AIOPool bdrv_em_co_aio_pool = {
3499 .aiocb_size = sizeof(BlockDriverAIOCBCoroutine),
3500 .cancel = bdrv_aio_co_cancel_em,
3503 static void bdrv_co_em_bh(void *opaque)
3505 BlockDriverAIOCBCoroutine *acb = opaque;
3507 acb->common.cb(acb->common.opaque, acb->req.error);
3508 qemu_bh_delete(acb->bh);
3509 qemu_aio_release(acb);
3512 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
3513 static void coroutine_fn bdrv_co_do_rw(void *opaque)
3515 BlockDriverAIOCBCoroutine *acb = opaque;
3516 BlockDriverState *bs = acb->common.bs;
3518 if (!acb->is_write) {
3519 acb->req.error = bdrv_co_do_readv(bs, acb->req.sector,
3520 acb->req.nb_sectors, acb->req.qiov, 0);
3522 acb->req.error = bdrv_co_do_writev(bs, acb->req.sector,
3523 acb->req.nb_sectors, acb->req.qiov, 0);
3526 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3527 qemu_bh_schedule(acb->bh);
3530 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
3534 BlockDriverCompletionFunc *cb,
3539 BlockDriverAIOCBCoroutine *acb;
3541 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3542 acb->req.sector = sector_num;
3543 acb->req.nb_sectors = nb_sectors;
3544 acb->req.qiov = qiov;
3545 acb->is_write = is_write;
3547 co = qemu_coroutine_create(bdrv_co_do_rw);
3548 qemu_coroutine_enter(co, acb);
3550 return &acb->common;
3553 static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
3555 BlockDriverAIOCBCoroutine *acb = opaque;
3556 BlockDriverState *bs = acb->common.bs;
3558 acb->req.error = bdrv_co_flush(bs);
3559 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3560 qemu_bh_schedule(acb->bh);
3563 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
3564 BlockDriverCompletionFunc *cb, void *opaque)
3566 trace_bdrv_aio_flush(bs, opaque);
3569 BlockDriverAIOCBCoroutine *acb;
3571 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3572 co = qemu_coroutine_create(bdrv_aio_flush_co_entry);
3573 qemu_coroutine_enter(co, acb);
3575 return &acb->common;
3578 static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque)
3580 BlockDriverAIOCBCoroutine *acb = opaque;
3581 BlockDriverState *bs = acb->common.bs;
3583 acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors);
3584 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3585 qemu_bh_schedule(acb->bh);
3588 BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs,
3589 int64_t sector_num, int nb_sectors,
3590 BlockDriverCompletionFunc *cb, void *opaque)
3593 BlockDriverAIOCBCoroutine *acb;
3595 trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque);
3597 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3598 acb->req.sector = sector_num;
3599 acb->req.nb_sectors = nb_sectors;
3600 co = qemu_coroutine_create(bdrv_aio_discard_co_entry);
3601 qemu_coroutine_enter(co, acb);
3603 return &acb->common;
3606 void bdrv_init(void)
3608 module_call_init(MODULE_INIT_BLOCK);
3611 void bdrv_init_with_whitelist(void)
3613 use_bdrv_whitelist = 1;
3617 void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
3618 BlockDriverCompletionFunc *cb, void *opaque)
3620 BlockDriverAIOCB *acb;
3622 if (pool->free_aiocb) {
3623 acb = pool->free_aiocb;
3624 pool->free_aiocb = acb->next;
3626 acb = g_malloc0(pool->aiocb_size);
3631 acb->opaque = opaque;
3635 void qemu_aio_release(void *p)
3637 BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
3638 AIOPool *pool = acb->pool;
3639 acb->next = pool->free_aiocb;
3640 pool->free_aiocb = acb;
3643 /**************************************************************/
3644 /* Coroutine block device emulation */
3646 typedef struct CoroutineIOCompletion {
3647 Coroutine *coroutine;
3649 } CoroutineIOCompletion;
3651 static void bdrv_co_io_em_complete(void *opaque, int ret)
3653 CoroutineIOCompletion *co = opaque;
3656 qemu_coroutine_enter(co->coroutine, NULL);
3659 static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num,
3660 int nb_sectors, QEMUIOVector *iov,
3663 CoroutineIOCompletion co = {
3664 .coroutine = qemu_coroutine_self(),
3666 BlockDriverAIOCB *acb;
3669 acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors,
3670 bdrv_co_io_em_complete, &co);
3672 acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors,
3673 bdrv_co_io_em_complete, &co);
3676 trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb);
3680 qemu_coroutine_yield();
3685 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
3686 int64_t sector_num, int nb_sectors,
3689 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false);
3692 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
3693 int64_t sector_num, int nb_sectors,
3696 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true);
3699 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
3701 RwCo *rwco = opaque;
3703 rwco->ret = bdrv_co_flush(rwco->bs);
3706 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
3710 if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
3714 /* Write back cached data to the OS even with cache=unsafe */
3715 if (bs->drv->bdrv_co_flush_to_os) {
3716 ret = bs->drv->bdrv_co_flush_to_os(bs);
3722 /* But don't actually force it to the disk with cache=unsafe */
3723 if (bs->open_flags & BDRV_O_NO_FLUSH) {
3727 if (bs->drv->bdrv_co_flush_to_disk) {
3728 ret = bs->drv->bdrv_co_flush_to_disk(bs);
3729 } else if (bs->drv->bdrv_aio_flush) {
3730 BlockDriverAIOCB *acb;
3731 CoroutineIOCompletion co = {
3732 .coroutine = qemu_coroutine_self(),
3735 acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
3739 qemu_coroutine_yield();
3744 * Some block drivers always operate in either writethrough or unsafe
3745 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
3746 * know how the server works (because the behaviour is hardcoded or
3747 * depends on server-side configuration), so we can't ensure that
3748 * everything is safe on disk. Returning an error doesn't work because
3749 * that would break guests even if the server operates in writethrough
3752 * Let's hope the user knows what he's doing.
3760 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
3761 * in the case of cache=unsafe, so there are no useless flushes.
3763 return bdrv_co_flush(bs->file);
3766 void bdrv_invalidate_cache(BlockDriverState *bs)
3768 if (bs->drv && bs->drv->bdrv_invalidate_cache) {
3769 bs->drv->bdrv_invalidate_cache(bs);
3773 void bdrv_invalidate_cache_all(void)
3775 BlockDriverState *bs;
3777 QTAILQ_FOREACH(bs, &bdrv_states, list) {
3778 bdrv_invalidate_cache(bs);
3782 void bdrv_clear_incoming_migration_all(void)
3784 BlockDriverState *bs;
3786 QTAILQ_FOREACH(bs, &bdrv_states, list) {
3787 bs->open_flags = bs->open_flags & ~(BDRV_O_INCOMING);
3791 int bdrv_flush(BlockDriverState *bs)
3799 if (qemu_in_coroutine()) {
3800 /* Fast-path if already in coroutine context */
3801 bdrv_flush_co_entry(&rwco);
3803 co = qemu_coroutine_create(bdrv_flush_co_entry);
3804 qemu_coroutine_enter(co, &rwco);
3805 while (rwco.ret == NOT_DONE) {
3813 static void coroutine_fn bdrv_discard_co_entry(void *opaque)
3815 RwCo *rwco = opaque;
3817 rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors);
3820 int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
3825 } else if (bdrv_check_request(bs, sector_num, nb_sectors)) {
3827 } else if (bs->read_only) {
3829 } else if (bs->drv->bdrv_co_discard) {
3830 return bs->drv->bdrv_co_discard(bs, sector_num, nb_sectors);
3831 } else if (bs->drv->bdrv_aio_discard) {
3832 BlockDriverAIOCB *acb;
3833 CoroutineIOCompletion co = {
3834 .coroutine = qemu_coroutine_self(),
3837 acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors,
3838 bdrv_co_io_em_complete, &co);
3842 qemu_coroutine_yield();
3850 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
3855 .sector_num = sector_num,
3856 .nb_sectors = nb_sectors,
3860 if (qemu_in_coroutine()) {
3861 /* Fast-path if already in coroutine context */
3862 bdrv_discard_co_entry(&rwco);
3864 co = qemu_coroutine_create(bdrv_discard_co_entry);
3865 qemu_coroutine_enter(co, &rwco);
3866 while (rwco.ret == NOT_DONE) {
3874 /**************************************************************/
3875 /* removable device support */
3878 * Return TRUE if the media is present
3880 int bdrv_is_inserted(BlockDriverState *bs)
3882 BlockDriver *drv = bs->drv;
3886 if (!drv->bdrv_is_inserted)
3888 return drv->bdrv_is_inserted(bs);
3892 * Return whether the media changed since the last call to this
3893 * function, or -ENOTSUP if we don't know. Most drivers don't know.
3895 int bdrv_media_changed(BlockDriverState *bs)
3897 BlockDriver *drv = bs->drv;
3899 if (drv && drv->bdrv_media_changed) {
3900 return drv->bdrv_media_changed(bs);
3906 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
3908 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
3910 BlockDriver *drv = bs->drv;
3912 if (drv && drv->bdrv_eject) {
3913 drv->bdrv_eject(bs, eject_flag);
3916 if (bs->device_name[0] != '\0') {
3917 bdrv_emit_qmp_eject_event(bs, eject_flag);
3922 * Lock or unlock the media (if it is locked, the user won't be able
3923 * to eject it manually).
3925 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
3927 BlockDriver *drv = bs->drv;
3929 trace_bdrv_lock_medium(bs, locked);
3931 if (drv && drv->bdrv_lock_medium) {
3932 drv->bdrv_lock_medium(bs, locked);
3936 /* needed for generic scsi interface */
3938 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
3940 BlockDriver *drv = bs->drv;
3942 if (drv && drv->bdrv_ioctl)
3943 return drv->bdrv_ioctl(bs, req, buf);
3947 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
3948 unsigned long int req, void *buf,
3949 BlockDriverCompletionFunc *cb, void *opaque)
3951 BlockDriver *drv = bs->drv;
3953 if (drv && drv->bdrv_aio_ioctl)
3954 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
3958 void bdrv_set_buffer_alignment(BlockDriverState *bs, int align)
3960 bs->buffer_alignment = align;
3963 void *qemu_blockalign(BlockDriverState *bs, size_t size)
3965 return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
3968 void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
3970 int64_t bitmap_size;
3972 bs->dirty_count = 0;
3974 if (!bs->dirty_bitmap) {
3975 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
3976 BDRV_SECTORS_PER_DIRTY_CHUNK * BITS_PER_LONG - 1;
3977 bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * BITS_PER_LONG;
3979 bs->dirty_bitmap = g_new0(unsigned long, bitmap_size);
3982 if (bs->dirty_bitmap) {
3983 g_free(bs->dirty_bitmap);
3984 bs->dirty_bitmap = NULL;
3989 int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
3991 int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
3993 if (bs->dirty_bitmap &&
3994 (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
3995 return !!(bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
3996 (1UL << (chunk % (sizeof(unsigned long) * 8))));
4002 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
4005 set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
4008 int64_t bdrv_get_dirty_count(BlockDriverState *bs)
4010 return bs->dirty_count;
4013 void bdrv_set_in_use(BlockDriverState *bs, int in_use)
4015 assert(bs->in_use != in_use);
4016 bs->in_use = in_use;
4019 int bdrv_in_use(BlockDriverState *bs)
4024 void bdrv_iostatus_enable(BlockDriverState *bs)
4026 bs->iostatus_enabled = true;
4027 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
4030 /* The I/O status is only enabled if the drive explicitly
4031 * enables it _and_ the VM is configured to stop on errors */
4032 bool bdrv_iostatus_is_enabled(const BlockDriverState *bs)
4034 return (bs->iostatus_enabled &&
4035 (bs->on_write_error == BLOCK_ERR_STOP_ENOSPC ||
4036 bs->on_write_error == BLOCK_ERR_STOP_ANY ||
4037 bs->on_read_error == BLOCK_ERR_STOP_ANY));
4040 void bdrv_iostatus_disable(BlockDriverState *bs)
4042 bs->iostatus_enabled = false;
4045 void bdrv_iostatus_reset(BlockDriverState *bs)
4047 if (bdrv_iostatus_is_enabled(bs)) {
4048 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
4052 /* XXX: Today this is set by device models because it makes the implementation
4053 quite simple. However, the block layer knows about the error, so it's
4054 possible to implement this without device models being involved */
4055 void bdrv_iostatus_set_err(BlockDriverState *bs, int error)
4057 if (bdrv_iostatus_is_enabled(bs) &&
4058 bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
4060 bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
4061 BLOCK_DEVICE_IO_STATUS_FAILED;
4066 bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes,
4067 enum BlockAcctType type)
4069 assert(type < BDRV_MAX_IOTYPE);
4071 cookie->bytes = bytes;
4072 cookie->start_time_ns = get_clock();
4073 cookie->type = type;
4077 bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie)
4079 assert(cookie->type < BDRV_MAX_IOTYPE);
4081 bs->nr_bytes[cookie->type] += cookie->bytes;
4082 bs->nr_ops[cookie->type]++;
4083 bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns;
4086 int bdrv_img_create(const char *filename, const char *fmt,
4087 const char *base_filename, const char *base_fmt,
4088 char *options, uint64_t img_size, int flags)
4090 QEMUOptionParameter *param = NULL, *create_options = NULL;
4091 QEMUOptionParameter *backing_fmt, *backing_file, *size;
4092 BlockDriverState *bs = NULL;
4093 BlockDriver *drv, *proto_drv;
4094 BlockDriver *backing_drv = NULL;
4097 /* Find driver and parse its options */
4098 drv = bdrv_find_format(fmt);
4100 error_report("Unknown file format '%s'", fmt);
4105 proto_drv = bdrv_find_protocol(filename);
4107 error_report("Unknown protocol '%s'", filename);
4112 create_options = append_option_parameters(create_options,
4113 drv->create_options);
4114 create_options = append_option_parameters(create_options,
4115 proto_drv->create_options);
4117 /* Create parameter list with default values */
4118 param = parse_option_parameters("", create_options, param);
4120 set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size);
4122 /* Parse -o options */
4124 param = parse_option_parameters(options, create_options, param);
4125 if (param == NULL) {
4126 error_report("Invalid options for file format '%s'.", fmt);
4132 if (base_filename) {
4133 if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE,
4135 error_report("Backing file not supported for file format '%s'",
4143 if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) {
4144 error_report("Backing file format not supported for file "
4145 "format '%s'", fmt);
4151 backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE);
4152 if (backing_file && backing_file->value.s) {
4153 if (!strcmp(filename, backing_file->value.s)) {
4154 error_report("Error: Trying to create an image with the "
4155 "same filename as the backing file");
4161 backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT);
4162 if (backing_fmt && backing_fmt->value.s) {
4163 backing_drv = bdrv_find_format(backing_fmt->value.s);
4165 error_report("Unknown backing file format '%s'",
4166 backing_fmt->value.s);
4172 // The size for the image must always be specified, with one exception:
4173 // If we are using a backing file, we can obtain the size from there
4174 size = get_option_parameter(param, BLOCK_OPT_SIZE);
4175 if (size && size->value.n == -1) {
4176 if (backing_file && backing_file->value.s) {
4181 /* backing files always opened read-only */
4183 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
4187 ret = bdrv_open(bs, backing_file->value.s, back_flags, backing_drv);
4189 error_report("Could not open '%s'", backing_file->value.s);
4192 bdrv_get_geometry(bs, &size);
4195 snprintf(buf, sizeof(buf), "%" PRId64, size);
4196 set_option_parameter(param, BLOCK_OPT_SIZE, buf);
4198 error_report("Image creation needs a size parameter");
4204 printf("Formatting '%s', fmt=%s ", filename, fmt);
4205 print_option_parameters(param);
4208 ret = bdrv_create(drv, filename, param);
4211 if (ret == -ENOTSUP) {
4212 error_report("Formatting or formatting option not supported for "
4213 "file format '%s'", fmt);
4214 } else if (ret == -EFBIG) {
4215 error_report("The image size is too large for file format '%s'",
4218 error_report("%s: error while creating %s: %s", filename, fmt,
4224 free_option_parameters(create_options);
4225 free_option_parameters(param);
4234 void *block_job_create(const BlockJobType *job_type, BlockDriverState *bs,
4235 int64_t speed, BlockDriverCompletionFunc *cb,
4236 void *opaque, Error **errp)
4240 if (bs->job || bdrv_in_use(bs)) {
4241 error_set(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
4244 bdrv_set_in_use(bs, 1);
4246 job = g_malloc0(job_type->instance_size);
4247 job->job_type = job_type;
4250 job->opaque = opaque;
4254 /* Only set speed when necessary to avoid NotSupported error */
4256 Error *local_err = NULL;
4258 block_job_set_speed(job, speed, &local_err);
4259 if (error_is_set(&local_err)) {
4262 bdrv_set_in_use(bs, 0);
4263 error_propagate(errp, local_err);
4270 void block_job_complete(BlockJob *job, int ret)
4272 BlockDriverState *bs = job->bs;
4274 assert(bs->job == job);
4275 job->cb(job->opaque, ret);
4278 bdrv_set_in_use(bs, 0);
4281 void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
4283 Error *local_err = NULL;
4285 if (!job->job_type->set_speed) {
4286 error_set(errp, QERR_NOT_SUPPORTED);
4289 job->job_type->set_speed(job, speed, &local_err);
4290 if (error_is_set(&local_err)) {
4291 error_propagate(errp, local_err);
4298 void block_job_cancel(BlockJob *job)
4300 job->cancelled = true;
4301 if (job->co && !job->busy) {
4302 qemu_coroutine_enter(job->co, NULL);
4306 bool block_job_is_cancelled(BlockJob *job)
4308 return job->cancelled;
4311 struct BlockCancelData {
4313 BlockDriverCompletionFunc *cb;
4319 static void block_job_cancel_cb(void *opaque, int ret)
4321 struct BlockCancelData *data = opaque;
4323 data->cancelled = block_job_is_cancelled(data->job);
4325 data->cb(data->opaque, ret);
4328 int block_job_cancel_sync(BlockJob *job)
4330 struct BlockCancelData data;
4331 BlockDriverState *bs = job->bs;
4333 assert(bs->job == job);
4335 /* Set up our own callback to store the result and chain to
4336 * the original callback.
4340 data.opaque = job->opaque;
4341 data.ret = -EINPROGRESS;
4342 job->cb = block_job_cancel_cb;
4343 job->opaque = &data;
4344 block_job_cancel(job);
4345 while (data.ret == -EINPROGRESS) {
4348 return (data.cancelled && data.ret == 0) ? -ECANCELED : data.ret;
4351 void block_job_sleep_ns(BlockJob *job, QEMUClock *clock, int64_t ns)
4353 /* Check cancellation *before* setting busy = false, too! */
4354 if (!block_job_is_cancelled(job)) {
4356 co_sleep_ns(clock, ns);