2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24 #include "config-host.h"
25 #include "qemu-common.h"
28 #include "block_int.h"
31 #include "qemu-coroutine.h"
32 #include "qmp-commands.h"
33 #include "qemu-timer.h"
36 #include <sys/types.h>
38 #include <sys/ioctl.h>
39 #include <sys/queue.h>
49 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
52 BDRV_REQ_COPY_ON_READ = 0x1,
53 BDRV_REQ_ZERO_WRITE = 0x2,
56 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load);
57 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
58 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
59 BlockDriverCompletionFunc *cb, void *opaque);
60 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
61 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
62 BlockDriverCompletionFunc *cb, void *opaque);
63 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
64 int64_t sector_num, int nb_sectors,
66 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
67 int64_t sector_num, int nb_sectors,
69 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
70 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
71 BdrvRequestFlags flags);
72 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
73 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
74 BdrvRequestFlags flags);
75 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
79 BlockDriverCompletionFunc *cb,
82 static void coroutine_fn bdrv_co_do_rw(void *opaque);
83 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
84 int64_t sector_num, int nb_sectors);
86 static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors,
87 bool is_write, double elapsed_time, uint64_t *wait);
88 static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write,
89 double elapsed_time, uint64_t *wait);
90 static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors,
91 bool is_write, int64_t *wait);
93 static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
94 QTAILQ_HEAD_INITIALIZER(bdrv_states);
96 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
97 QLIST_HEAD_INITIALIZER(bdrv_drivers);
99 /* The device to use for VM snapshots */
100 static BlockDriverState *bs_snapshots;
102 /* If non-zero, use only whitelisted block drivers */
103 static int use_bdrv_whitelist;
106 static int is_windows_drive_prefix(const char *filename)
108 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
109 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
113 int is_windows_drive(const char *filename)
115 if (is_windows_drive_prefix(filename) &&
118 if (strstart(filename, "\\\\.\\", NULL) ||
119 strstart(filename, "//./", NULL))
125 /* throttling disk I/O limits */
126 void bdrv_io_limits_disable(BlockDriverState *bs)
128 bs->io_limits_enabled = false;
130 while (qemu_co_queue_next(&bs->throttled_reqs));
132 if (bs->block_timer) {
133 qemu_del_timer(bs->block_timer);
134 qemu_free_timer(bs->block_timer);
135 bs->block_timer = NULL;
141 memset(&bs->io_base, 0, sizeof(bs->io_base));
144 static void bdrv_block_timer(void *opaque)
146 BlockDriverState *bs = opaque;
148 qemu_co_queue_next(&bs->throttled_reqs);
151 void bdrv_io_limits_enable(BlockDriverState *bs)
153 qemu_co_queue_init(&bs->throttled_reqs);
154 bs->block_timer = qemu_new_timer_ns(vm_clock, bdrv_block_timer, bs);
155 bs->slice_time = 5 * BLOCK_IO_SLICE_TIME;
156 bs->slice_start = qemu_get_clock_ns(vm_clock);
157 bs->slice_end = bs->slice_start + bs->slice_time;
158 memset(&bs->io_base, 0, sizeof(bs->io_base));
159 bs->io_limits_enabled = true;
162 bool bdrv_io_limits_enabled(BlockDriverState *bs)
164 BlockIOLimit *io_limits = &bs->io_limits;
165 return io_limits->bps[BLOCK_IO_LIMIT_READ]
166 || io_limits->bps[BLOCK_IO_LIMIT_WRITE]
167 || io_limits->bps[BLOCK_IO_LIMIT_TOTAL]
168 || io_limits->iops[BLOCK_IO_LIMIT_READ]
169 || io_limits->iops[BLOCK_IO_LIMIT_WRITE]
170 || io_limits->iops[BLOCK_IO_LIMIT_TOTAL];
173 static void bdrv_io_limits_intercept(BlockDriverState *bs,
174 bool is_write, int nb_sectors)
176 int64_t wait_time = -1;
178 if (!qemu_co_queue_empty(&bs->throttled_reqs)) {
179 qemu_co_queue_wait(&bs->throttled_reqs);
182 /* In fact, we hope to keep each request's timing, in FIFO mode. The next
183 * throttled requests will not be dequeued until the current request is
184 * allowed to be serviced. So if the current request still exceeds the
185 * limits, it will be inserted to the head. All requests followed it will
186 * be still in throttled_reqs queue.
189 while (bdrv_exceed_io_limits(bs, nb_sectors, is_write, &wait_time)) {
190 qemu_mod_timer(bs->block_timer,
191 wait_time + qemu_get_clock_ns(vm_clock));
192 qemu_co_queue_wait_insert_head(&bs->throttled_reqs);
195 qemu_co_queue_next(&bs->throttled_reqs);
198 /* check if the path starts with "<protocol>:" */
199 static int path_has_protocol(const char *path)
202 if (is_windows_drive(path) ||
203 is_windows_drive_prefix(path)) {
208 return strchr(path, ':') != NULL;
211 int path_is_absolute(const char *path)
215 /* specific case for names like: "\\.\d:" */
216 if (*path == '/' || *path == '\\')
219 p = strchr(path, ':');
225 return (*p == '/' || *p == '\\');
231 /* if filename is absolute, just copy it to dest. Otherwise, build a
232 path to it by considering it is relative to base_path. URL are
234 void path_combine(char *dest, int dest_size,
235 const char *base_path,
236 const char *filename)
243 if (path_is_absolute(filename)) {
244 pstrcpy(dest, dest_size, filename);
246 p = strchr(base_path, ':');
251 p1 = strrchr(base_path, '/');
255 p2 = strrchr(base_path, '\\');
267 if (len > dest_size - 1)
269 memcpy(dest, base_path, len);
271 pstrcat(dest, dest_size, filename);
275 void bdrv_register(BlockDriver *bdrv)
277 /* Block drivers without coroutine functions need emulation */
278 if (!bdrv->bdrv_co_readv) {
279 bdrv->bdrv_co_readv = bdrv_co_readv_em;
280 bdrv->bdrv_co_writev = bdrv_co_writev_em;
282 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
283 * the block driver lacks aio we need to emulate that too.
285 if (!bdrv->bdrv_aio_readv) {
286 /* add AIO emulation layer */
287 bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
288 bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
292 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
295 /* create a new block device (by default it is empty) */
296 BlockDriverState *bdrv_new(const char *device_name)
298 BlockDriverState *bs;
300 bs = g_malloc0(sizeof(BlockDriverState));
301 pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
302 if (device_name[0] != '\0') {
303 QTAILQ_INSERT_TAIL(&bdrv_states, bs, list);
305 bdrv_iostatus_disable(bs);
309 BlockDriver *bdrv_find_format(const char *format_name)
312 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
313 if (!strcmp(drv1->format_name, format_name)) {
320 static int bdrv_is_whitelisted(BlockDriver *drv)
322 static const char *whitelist[] = {
323 CONFIG_BDRV_WHITELIST
328 return 1; /* no whitelist, anything goes */
330 for (p = whitelist; *p; p++) {
331 if (!strcmp(drv->format_name, *p)) {
338 BlockDriver *bdrv_find_whitelisted_format(const char *format_name)
340 BlockDriver *drv = bdrv_find_format(format_name);
341 return drv && bdrv_is_whitelisted(drv) ? drv : NULL;
344 typedef struct CreateCo {
347 QEMUOptionParameter *options;
351 static void coroutine_fn bdrv_create_co_entry(void *opaque)
353 CreateCo *cco = opaque;
356 cco->ret = cco->drv->bdrv_create(cco->filename, cco->options);
359 int bdrv_create(BlockDriver *drv, const char* filename,
360 QEMUOptionParameter *options)
367 .filename = g_strdup(filename),
372 if (!drv->bdrv_create) {
376 if (qemu_in_coroutine()) {
377 /* Fast-path if already in coroutine context */
378 bdrv_create_co_entry(&cco);
380 co = qemu_coroutine_create(bdrv_create_co_entry);
381 qemu_coroutine_enter(co, &cco);
382 while (cco.ret == NOT_DONE) {
388 g_free(cco.filename);
393 int bdrv_create_file(const char* filename, QEMUOptionParameter *options)
397 drv = bdrv_find_protocol(filename);
402 return bdrv_create(drv, filename, options);
406 void get_tmp_filename(char *filename, int size)
408 char temp_dir[MAX_PATH];
410 GetTempPath(MAX_PATH, temp_dir);
411 GetTempFileName(temp_dir, "qem", 0, filename);
414 void get_tmp_filename(char *filename, int size)
418 /* XXX: race condition possible */
419 tmpdir = getenv("TMPDIR");
422 snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
423 fd = mkstemp(filename);
429 * Detect host devices. By convention, /dev/cdrom[N] is always
430 * recognized as a host CDROM.
432 static BlockDriver *find_hdev_driver(const char *filename)
434 int score_max = 0, score;
435 BlockDriver *drv = NULL, *d;
437 QLIST_FOREACH(d, &bdrv_drivers, list) {
438 if (d->bdrv_probe_device) {
439 score = d->bdrv_probe_device(filename);
440 if (score > score_max) {
450 BlockDriver *bdrv_find_protocol(const char *filename)
457 /* TODO Drivers without bdrv_file_open must be specified explicitly */
460 * XXX(hch): we really should not let host device detection
461 * override an explicit protocol specification, but moving this
462 * later breaks access to device names with colons in them.
463 * Thanks to the brain-dead persistent naming schemes on udev-
464 * based Linux systems those actually are quite common.
466 drv1 = find_hdev_driver(filename);
471 if (!path_has_protocol(filename)) {
472 return bdrv_find_format("file");
474 p = strchr(filename, ':');
477 if (len > sizeof(protocol) - 1)
478 len = sizeof(protocol) - 1;
479 memcpy(protocol, filename, len);
480 protocol[len] = '\0';
481 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
482 if (drv1->protocol_name &&
483 !strcmp(drv1->protocol_name, protocol)) {
490 static int find_image_format(const char *filename, BlockDriver **pdrv)
492 int ret, score, score_max;
493 BlockDriver *drv1, *drv;
495 BlockDriverState *bs;
497 ret = bdrv_file_open(&bs, filename, 0);
503 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
504 if (bs->sg || !bdrv_is_inserted(bs)) {
506 drv = bdrv_find_format("raw");
514 ret = bdrv_pread(bs, 0, buf, sizeof(buf));
523 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
524 if (drv1->bdrv_probe) {
525 score = drv1->bdrv_probe(buf, ret, filename);
526 if (score > score_max) {
540 * Set the current 'total_sectors' value
542 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
544 BlockDriver *drv = bs->drv;
546 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
550 /* query actual device if possible, otherwise just trust the hint */
551 if (drv->bdrv_getlength) {
552 int64_t length = drv->bdrv_getlength(bs);
556 hint = length >> BDRV_SECTOR_BITS;
559 bs->total_sectors = hint;
564 * Set open flags for a given cache mode
566 * Return 0 on success, -1 if the cache mode was invalid.
568 int bdrv_parse_cache_flags(const char *mode, int *flags)
570 *flags &= ~BDRV_O_CACHE_MASK;
572 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
573 *flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB;
574 } else if (!strcmp(mode, "directsync")) {
575 *flags |= BDRV_O_NOCACHE;
576 } else if (!strcmp(mode, "writeback")) {
577 *flags |= BDRV_O_CACHE_WB;
578 } else if (!strcmp(mode, "unsafe")) {
579 *flags |= BDRV_O_CACHE_WB;
580 *flags |= BDRV_O_NO_FLUSH;
581 } else if (!strcmp(mode, "writethrough")) {
582 /* this is the default */
591 * The copy-on-read flag is actually a reference count so multiple users may
592 * use the feature without worrying about clobbering its previous state.
593 * Copy-on-read stays enabled until all users have called to disable it.
595 void bdrv_enable_copy_on_read(BlockDriverState *bs)
600 void bdrv_disable_copy_on_read(BlockDriverState *bs)
602 assert(bs->copy_on_read > 0);
607 * Common part for opening disk images and files
609 static int bdrv_open_common(BlockDriverState *bs, const char *filename,
610 int flags, BlockDriver *drv)
616 trace_bdrv_open_common(bs, filename, flags, drv->format_name);
619 bs->total_sectors = 0;
623 bs->open_flags = flags;
625 bs->buffer_alignment = 512;
627 assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
628 if ((flags & BDRV_O_RDWR) && (flags & BDRV_O_COPY_ON_READ)) {
629 bdrv_enable_copy_on_read(bs);
632 pstrcpy(bs->filename, sizeof(bs->filename), filename);
633 bs->backing_file[0] = '\0';
635 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv)) {
640 bs->opaque = g_malloc0(drv->instance_size);
642 bs->enable_write_cache = !!(flags & BDRV_O_CACHE_WB);
645 * Clear flags that are internal to the block layer before opening the
648 open_flags = flags & ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
651 * Snapshots should be writable.
653 if (bs->is_temporary) {
654 open_flags |= BDRV_O_RDWR;
657 bs->keep_read_only = bs->read_only = !(open_flags & BDRV_O_RDWR);
659 /* Open the image, either directly or using a protocol */
660 if (drv->bdrv_file_open) {
661 ret = drv->bdrv_file_open(bs, filename, open_flags);
663 ret = bdrv_file_open(&bs->file, filename, open_flags);
665 ret = drv->bdrv_open(bs, open_flags);
673 ret = refresh_total_sectors(bs, bs->total_sectors);
679 if (bs->is_temporary) {
687 bdrv_delete(bs->file);
697 * Opens a file using a protocol (file, host_device, nbd, ...)
699 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
701 BlockDriverState *bs;
705 drv = bdrv_find_protocol(filename);
711 ret = bdrv_open_common(bs, filename, flags, drv);
722 * Opens a disk image (raw, qcow2, vmdk, ...)
724 int bdrv_open(BlockDriverState *bs, const char *filename, int flags,
728 char tmp_filename[PATH_MAX];
730 if (flags & BDRV_O_SNAPSHOT) {
731 BlockDriverState *bs1;
734 BlockDriver *bdrv_qcow2;
735 QEMUOptionParameter *options;
736 char backing_filename[PATH_MAX];
738 /* if snapshot, we create a temporary backing file and open it
739 instead of opening 'filename' directly */
741 /* if there is a backing file, use it */
743 ret = bdrv_open(bs1, filename, 0, drv);
748 total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
750 if (bs1->drv && bs1->drv->protocol_name)
755 get_tmp_filename(tmp_filename, sizeof(tmp_filename));
757 /* Real path is meaningless for protocols */
759 snprintf(backing_filename, sizeof(backing_filename),
761 else if (!realpath(filename, backing_filename))
764 bdrv_qcow2 = bdrv_find_format("qcow2");
765 options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
767 set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size);
768 set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
770 set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
774 ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
775 free_option_parameters(options);
780 filename = tmp_filename;
782 bs->is_temporary = 1;
785 /* Find the right image format driver */
787 ret = find_image_format(filename, &drv);
791 goto unlink_and_fail;
795 ret = bdrv_open_common(bs, filename, flags, drv);
797 goto unlink_and_fail;
800 /* If there is a backing file, use it */
801 if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') {
802 char backing_filename[PATH_MAX];
804 BlockDriver *back_drv = NULL;
806 bs->backing_hd = bdrv_new("");
808 if (path_has_protocol(bs->backing_file)) {
809 pstrcpy(backing_filename, sizeof(backing_filename),
812 path_combine(backing_filename, sizeof(backing_filename),
813 filename, bs->backing_file);
816 if (bs->backing_format[0] != '\0') {
817 back_drv = bdrv_find_format(bs->backing_format);
820 /* backing files always opened read-only */
822 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
824 ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv);
829 if (bs->is_temporary) {
830 bs->backing_hd->keep_read_only = !(flags & BDRV_O_RDWR);
832 /* base image inherits from "parent" */
833 bs->backing_hd->keep_read_only = bs->keep_read_only;
837 if (!bdrv_key_required(bs)) {
838 bdrv_dev_change_media_cb(bs, true);
841 /* throttling disk I/O limits */
842 if (bs->io_limits_enabled) {
843 bdrv_io_limits_enable(bs);
849 if (bs->is_temporary) {
855 void bdrv_close(BlockDriverState *bs)
860 block_job_cancel_sync(bs->job);
864 if (bs == bs_snapshots) {
867 if (bs->backing_hd) {
868 bdrv_delete(bs->backing_hd);
869 bs->backing_hd = NULL;
871 bs->drv->bdrv_close(bs);
874 if (bs->is_temporary) {
875 unlink(bs->filename);
880 bs->copy_on_read = 0;
882 if (bs->file != NULL) {
883 bdrv_close(bs->file);
886 bdrv_dev_change_media_cb(bs, false);
889 /*throttling disk I/O limits*/
890 if (bs->io_limits_enabled) {
891 bdrv_io_limits_disable(bs);
895 void bdrv_close_all(void)
897 BlockDriverState *bs;
899 QTAILQ_FOREACH(bs, &bdrv_states, list) {
905 * Wait for pending requests to complete across all BlockDriverStates
907 * This function does not flush data to disk, use bdrv_flush_all() for that
908 * after calling this function.
910 * Note that completion of an asynchronous I/O operation can trigger any
911 * number of other I/O operations on other devices---for example a coroutine
912 * can be arbitrarily complex and a constant flow of I/O can come until the
913 * coroutine is complete. Because of this, it is not possible to have a
914 * function to drain a single device's I/O queue.
916 void bdrv_drain_all(void)
918 BlockDriverState *bs;
922 busy = qemu_aio_wait();
924 /* FIXME: We do not have timer support here, so this is effectively
927 QTAILQ_FOREACH(bs, &bdrv_states, list) {
928 if (!qemu_co_queue_empty(&bs->throttled_reqs)) {
929 qemu_co_queue_restart_all(&bs->throttled_reqs);
935 /* If requests are still pending there is a bug somewhere */
936 QTAILQ_FOREACH(bs, &bdrv_states, list) {
937 assert(QLIST_EMPTY(&bs->tracked_requests));
938 assert(qemu_co_queue_empty(&bs->throttled_reqs));
942 /* make a BlockDriverState anonymous by removing from bdrv_state list.
943 Also, NULL terminate the device_name to prevent double remove */
944 void bdrv_make_anon(BlockDriverState *bs)
946 if (bs->device_name[0] != '\0') {
947 QTAILQ_REMOVE(&bdrv_states, bs, list);
949 bs->device_name[0] = '\0';
952 static void bdrv_rebind(BlockDriverState *bs)
954 if (bs->drv && bs->drv->bdrv_rebind) {
955 bs->drv->bdrv_rebind(bs);
960 * Add new bs contents at the top of an image chain while the chain is
961 * live, while keeping required fields on the top layer.
963 * This will modify the BlockDriverState fields, and swap contents
964 * between bs_new and bs_top. Both bs_new and bs_top are modified.
966 * bs_new is required to be anonymous.
968 * This function does not create any image files.
970 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
972 BlockDriverState tmp;
974 /* bs_new must be anonymous */
975 assert(bs_new->device_name[0] == '\0');
979 /* there are some fields that need to stay on the top layer: */
982 tmp.dev_ops = bs_top->dev_ops;
983 tmp.dev_opaque = bs_top->dev_opaque;
984 tmp.dev = bs_top->dev;
985 tmp.buffer_alignment = bs_top->buffer_alignment;
986 tmp.copy_on_read = bs_top->copy_on_read;
988 /* i/o timing parameters */
989 tmp.slice_time = bs_top->slice_time;
990 tmp.slice_start = bs_top->slice_start;
991 tmp.slice_end = bs_top->slice_end;
992 tmp.io_limits = bs_top->io_limits;
993 tmp.io_base = bs_top->io_base;
994 tmp.throttled_reqs = bs_top->throttled_reqs;
995 tmp.block_timer = bs_top->block_timer;
996 tmp.io_limits_enabled = bs_top->io_limits_enabled;
999 tmp.cyls = bs_top->cyls;
1000 tmp.heads = bs_top->heads;
1001 tmp.secs = bs_top->secs;
1002 tmp.translation = bs_top->translation;
1005 tmp.on_read_error = bs_top->on_read_error;
1006 tmp.on_write_error = bs_top->on_write_error;
1009 tmp.iostatus_enabled = bs_top->iostatus_enabled;
1010 tmp.iostatus = bs_top->iostatus;
1012 /* keep the same entry in bdrv_states */
1013 pstrcpy(tmp.device_name, sizeof(tmp.device_name), bs_top->device_name);
1014 tmp.list = bs_top->list;
1016 /* The contents of 'tmp' will become bs_top, as we are
1017 * swapping bs_new and bs_top contents. */
1018 tmp.backing_hd = bs_new;
1019 pstrcpy(tmp.backing_file, sizeof(tmp.backing_file), bs_top->filename);
1020 bdrv_get_format(bs_top, tmp.backing_format, sizeof(tmp.backing_format));
1022 /* swap contents of the fixed new bs and the current top */
1026 /* device_name[] was carried over from the old bs_top. bs_new
1027 * shouldn't be in bdrv_states, so we need to make device_name[]
1028 * reflect the anonymity of bs_new
1030 bs_new->device_name[0] = '\0';
1032 /* clear the copied fields in the new backing file */
1033 bdrv_detach_dev(bs_new, bs_new->dev);
1035 qemu_co_queue_init(&bs_new->throttled_reqs);
1036 memset(&bs_new->io_base, 0, sizeof(bs_new->io_base));
1037 memset(&bs_new->io_limits, 0, sizeof(bs_new->io_limits));
1038 bdrv_iostatus_disable(bs_new);
1040 /* we don't use bdrv_io_limits_disable() for this, because we don't want
1041 * to affect or delete the block_timer, as it has been moved to bs_top */
1042 bs_new->io_limits_enabled = false;
1043 bs_new->block_timer = NULL;
1044 bs_new->slice_time = 0;
1045 bs_new->slice_start = 0;
1046 bs_new->slice_end = 0;
1048 bdrv_rebind(bs_new);
1049 bdrv_rebind(bs_top);
1052 void bdrv_delete(BlockDriverState *bs)
1056 assert(!bs->in_use);
1058 /* remove from list, if necessary */
1062 if (bs->file != NULL) {
1063 bdrv_delete(bs->file);
1066 assert(bs != bs_snapshots);
1070 int bdrv_attach_dev(BlockDriverState *bs, void *dev)
1071 /* TODO change to DeviceState *dev when all users are qdevified */
1077 bdrv_iostatus_reset(bs);
1081 /* TODO qdevified devices don't use this, remove when devices are qdevified */
1082 void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev)
1084 if (bdrv_attach_dev(bs, dev) < 0) {
1089 void bdrv_detach_dev(BlockDriverState *bs, void *dev)
1090 /* TODO change to DeviceState *dev when all users are qdevified */
1092 assert(bs->dev == dev);
1095 bs->dev_opaque = NULL;
1096 bs->buffer_alignment = 512;
1099 /* TODO change to return DeviceState * when all users are qdevified */
1100 void *bdrv_get_attached_dev(BlockDriverState *bs)
1105 void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops,
1109 bs->dev_opaque = opaque;
1110 if (bdrv_dev_has_removable_media(bs) && bs == bs_snapshots) {
1111 bs_snapshots = NULL;
1115 void bdrv_emit_qmp_error_event(const BlockDriverState *bdrv,
1116 BlockQMPEventAction action, int is_read)
1119 const char *action_str;
1122 case BDRV_ACTION_REPORT:
1123 action_str = "report";
1125 case BDRV_ACTION_IGNORE:
1126 action_str = "ignore";
1128 case BDRV_ACTION_STOP:
1129 action_str = "stop";
1135 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1138 is_read ? "read" : "write");
1139 monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
1141 qobject_decref(data);
1144 static void bdrv_emit_qmp_eject_event(BlockDriverState *bs, bool ejected)
1148 data = qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }",
1149 bdrv_get_device_name(bs), ejected);
1150 monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED, data);
1152 qobject_decref(data);
1155 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load)
1157 if (bs->dev_ops && bs->dev_ops->change_media_cb) {
1158 bool tray_was_closed = !bdrv_dev_is_tray_open(bs);
1159 bs->dev_ops->change_media_cb(bs->dev_opaque, load);
1160 if (tray_was_closed) {
1162 bdrv_emit_qmp_eject_event(bs, true);
1166 bdrv_emit_qmp_eject_event(bs, false);
1171 bool bdrv_dev_has_removable_media(BlockDriverState *bs)
1173 return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb);
1176 void bdrv_dev_eject_request(BlockDriverState *bs, bool force)
1178 if (bs->dev_ops && bs->dev_ops->eject_request_cb) {
1179 bs->dev_ops->eject_request_cb(bs->dev_opaque, force);
1183 bool bdrv_dev_is_tray_open(BlockDriverState *bs)
1185 if (bs->dev_ops && bs->dev_ops->is_tray_open) {
1186 return bs->dev_ops->is_tray_open(bs->dev_opaque);
1191 static void bdrv_dev_resize_cb(BlockDriverState *bs)
1193 if (bs->dev_ops && bs->dev_ops->resize_cb) {
1194 bs->dev_ops->resize_cb(bs->dev_opaque);
1198 bool bdrv_dev_is_medium_locked(BlockDriverState *bs)
1200 if (bs->dev_ops && bs->dev_ops->is_medium_locked) {
1201 return bs->dev_ops->is_medium_locked(bs->dev_opaque);
1207 * Run consistency checks on an image
1209 * Returns 0 if the check could be completed (it doesn't mean that the image is
1210 * free of errors) or -errno when an internal error occurred. The results of the
1211 * check are stored in res.
1213 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res)
1215 if (bs->drv->bdrv_check == NULL) {
1219 memset(res, 0, sizeof(*res));
1220 return bs->drv->bdrv_check(bs, res);
1223 #define COMMIT_BUF_SECTORS 2048
1225 /* commit COW file into the raw image */
1226 int bdrv_commit(BlockDriverState *bs)
1228 BlockDriver *drv = bs->drv;
1229 BlockDriver *backing_drv;
1230 int64_t sector, total_sectors;
1231 int n, ro, open_flags;
1232 int ret = 0, rw_ret = 0;
1234 char filename[1024];
1235 BlockDriverState *bs_rw, *bs_ro;
1240 if (!bs->backing_hd) {
1244 if (bs->backing_hd->keep_read_only) {
1248 if (bdrv_in_use(bs) || bdrv_in_use(bs->backing_hd)) {
1252 backing_drv = bs->backing_hd->drv;
1253 ro = bs->backing_hd->read_only;
1254 strncpy(filename, bs->backing_hd->filename, sizeof(filename));
1255 open_flags = bs->backing_hd->open_flags;
1259 bdrv_delete(bs->backing_hd);
1260 bs->backing_hd = NULL;
1261 bs_rw = bdrv_new("");
1262 rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR,
1266 /* try to re-open read-only */
1267 bs_ro = bdrv_new("");
1268 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
1272 /* drive not functional anymore */
1276 bs->backing_hd = bs_ro;
1279 bs->backing_hd = bs_rw;
1282 total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
1283 buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
1285 for (sector = 0; sector < total_sectors; sector += n) {
1286 if (bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n)) {
1288 if (bdrv_read(bs, sector, buf, n) != 0) {
1293 if (bdrv_write(bs->backing_hd, sector, buf, n) != 0) {
1300 if (drv->bdrv_make_empty) {
1301 ret = drv->bdrv_make_empty(bs);
1306 * Make sure all data we wrote to the backing device is actually
1310 bdrv_flush(bs->backing_hd);
1317 bdrv_delete(bs->backing_hd);
1318 bs->backing_hd = NULL;
1319 bs_ro = bdrv_new("");
1320 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
1324 /* drive not functional anymore */
1328 bs->backing_hd = bs_ro;
1329 bs->backing_hd->keep_read_only = 0;
1335 int bdrv_commit_all(void)
1337 BlockDriverState *bs;
1339 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1340 int ret = bdrv_commit(bs);
1348 struct BdrvTrackedRequest {
1349 BlockDriverState *bs;
1353 QLIST_ENTRY(BdrvTrackedRequest) list;
1354 Coroutine *co; /* owner, used for deadlock detection */
1355 CoQueue wait_queue; /* coroutines blocked on this request */
1359 * Remove an active request from the tracked requests list
1361 * This function should be called when a tracked request is completing.
1363 static void tracked_request_end(BdrvTrackedRequest *req)
1365 QLIST_REMOVE(req, list);
1366 qemu_co_queue_restart_all(&req->wait_queue);
1370 * Add an active request to the tracked requests list
1372 static void tracked_request_begin(BdrvTrackedRequest *req,
1373 BlockDriverState *bs,
1375 int nb_sectors, bool is_write)
1377 *req = (BdrvTrackedRequest){
1379 .sector_num = sector_num,
1380 .nb_sectors = nb_sectors,
1381 .is_write = is_write,
1382 .co = qemu_coroutine_self(),
1385 qemu_co_queue_init(&req->wait_queue);
1387 QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
1391 * Round a region to cluster boundaries
1393 static void round_to_clusters(BlockDriverState *bs,
1394 int64_t sector_num, int nb_sectors,
1395 int64_t *cluster_sector_num,
1396 int *cluster_nb_sectors)
1398 BlockDriverInfo bdi;
1400 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
1401 *cluster_sector_num = sector_num;
1402 *cluster_nb_sectors = nb_sectors;
1404 int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE;
1405 *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c);
1406 *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num +
1411 static bool tracked_request_overlaps(BdrvTrackedRequest *req,
1412 int64_t sector_num, int nb_sectors) {
1414 if (sector_num >= req->sector_num + req->nb_sectors) {
1418 if (req->sector_num >= sector_num + nb_sectors) {
1424 static void coroutine_fn wait_for_overlapping_requests(BlockDriverState *bs,
1425 int64_t sector_num, int nb_sectors)
1427 BdrvTrackedRequest *req;
1428 int64_t cluster_sector_num;
1429 int cluster_nb_sectors;
1432 /* If we touch the same cluster it counts as an overlap. This guarantees
1433 * that allocating writes will be serialized and not race with each other
1434 * for the same cluster. For example, in copy-on-read it ensures that the
1435 * CoR read and write operations are atomic and guest writes cannot
1436 * interleave between them.
1438 round_to_clusters(bs, sector_num, nb_sectors,
1439 &cluster_sector_num, &cluster_nb_sectors);
1443 QLIST_FOREACH(req, &bs->tracked_requests, list) {
1444 if (tracked_request_overlaps(req, cluster_sector_num,
1445 cluster_nb_sectors)) {
1446 /* Hitting this means there was a reentrant request, for
1447 * example, a block driver issuing nested requests. This must
1448 * never happen since it means deadlock.
1450 assert(qemu_coroutine_self() != req->co);
1452 qemu_co_queue_wait(&req->wait_queue);
1463 * -EINVAL - backing format specified, but no file
1464 * -ENOSPC - can't update the backing file because no space is left in the
1466 * -ENOTSUP - format driver doesn't support changing the backing file
1468 int bdrv_change_backing_file(BlockDriverState *bs,
1469 const char *backing_file, const char *backing_fmt)
1471 BlockDriver *drv = bs->drv;
1474 /* Backing file format doesn't make sense without a backing file */
1475 if (backing_fmt && !backing_file) {
1479 if (drv->bdrv_change_backing_file != NULL) {
1480 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
1486 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
1487 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
1492 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
1497 if (!bdrv_is_inserted(bs))
1503 len = bdrv_getlength(bs);
1508 if ((offset > len) || (len - offset < size))
1514 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
1517 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
1518 nb_sectors * BDRV_SECTOR_SIZE);
1521 typedef struct RwCo {
1522 BlockDriverState *bs;
1530 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
1532 RwCo *rwco = opaque;
1534 if (!rwco->is_write) {
1535 rwco->ret = bdrv_co_do_readv(rwco->bs, rwco->sector_num,
1536 rwco->nb_sectors, rwco->qiov, 0);
1538 rwco->ret = bdrv_co_do_writev(rwco->bs, rwco->sector_num,
1539 rwco->nb_sectors, rwco->qiov, 0);
1544 * Process a synchronous request using coroutines
1546 static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf,
1547 int nb_sectors, bool is_write)
1550 struct iovec iov = {
1551 .iov_base = (void *)buf,
1552 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
1557 .sector_num = sector_num,
1558 .nb_sectors = nb_sectors,
1560 .is_write = is_write,
1564 qemu_iovec_init_external(&qiov, &iov, 1);
1567 * In sync call context, when the vcpu is blocked, this throttling timer
1568 * will not fire; so the I/O throttling function has to be disabled here
1569 * if it has been enabled.
1571 if (bs->io_limits_enabled) {
1572 fprintf(stderr, "Disabling I/O throttling on '%s' due "
1573 "to synchronous I/O.\n", bdrv_get_device_name(bs));
1574 bdrv_io_limits_disable(bs);
1577 if (qemu_in_coroutine()) {
1578 /* Fast-path if already in coroutine context */
1579 bdrv_rw_co_entry(&rwco);
1581 co = qemu_coroutine_create(bdrv_rw_co_entry);
1582 qemu_coroutine_enter(co, &rwco);
1583 while (rwco.ret == NOT_DONE) {
1590 /* return < 0 if error. See bdrv_write() for the return codes */
1591 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
1592 uint8_t *buf, int nb_sectors)
1594 return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false);
1597 #define BITS_PER_LONG (sizeof(unsigned long) * 8)
1599 static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
1600 int nb_sectors, int dirty)
1603 unsigned long val, idx, bit;
1605 start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
1606 end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
1608 for (; start <= end; start++) {
1609 idx = start / BITS_PER_LONG;
1610 bit = start % BITS_PER_LONG;
1611 val = bs->dirty_bitmap[idx];
1613 if (!(val & (1UL << bit))) {
1618 if (val & (1UL << bit)) {
1620 val &= ~(1UL << bit);
1623 bs->dirty_bitmap[idx] = val;
1627 /* Return < 0 if error. Important errors are:
1628 -EIO generic I/O error (may happen for all errors)
1629 -ENOMEDIUM No media inserted.
1630 -EINVAL Invalid sector number or nb_sectors
1631 -EACCES Trying to write a read-only device
1633 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
1634 const uint8_t *buf, int nb_sectors)
1636 return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true);
1639 int bdrv_pread(BlockDriverState *bs, int64_t offset,
1640 void *buf, int count1)
1642 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1643 int len, nb_sectors, count;
1648 /* first read to align to sector start */
1649 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1652 sector_num = offset >> BDRV_SECTOR_BITS;
1654 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1656 memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
1664 /* read the sectors "in place" */
1665 nb_sectors = count >> BDRV_SECTOR_BITS;
1666 if (nb_sectors > 0) {
1667 if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
1669 sector_num += nb_sectors;
1670 len = nb_sectors << BDRV_SECTOR_BITS;
1675 /* add data from the last sector */
1677 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1679 memcpy(buf, tmp_buf, count);
1684 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
1685 const void *buf, int count1)
1687 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1688 int len, nb_sectors, count;
1693 /* first write to align to sector start */
1694 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1697 sector_num = offset >> BDRV_SECTOR_BITS;
1699 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1701 memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
1702 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1711 /* write the sectors "in place" */
1712 nb_sectors = count >> BDRV_SECTOR_BITS;
1713 if (nb_sectors > 0) {
1714 if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0)
1716 sector_num += nb_sectors;
1717 len = nb_sectors << BDRV_SECTOR_BITS;
1722 /* add data from the last sector */
1724 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1726 memcpy(tmp_buf, buf, count);
1727 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1734 * Writes to the file and ensures that no writes are reordered across this
1735 * request (acts as a barrier)
1737 * Returns 0 on success, -errno in error cases.
1739 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
1740 const void *buf, int count)
1744 ret = bdrv_pwrite(bs, offset, buf, count);
1749 /* No flush needed for cache modes that use O_DSYNC */
1750 if ((bs->open_flags & BDRV_O_CACHE_WB) != 0) {
1757 static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs,
1758 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1760 /* Perform I/O through a temporary buffer so that users who scribble over
1761 * their read buffer while the operation is in progress do not end up
1762 * modifying the image file. This is critical for zero-copy guest I/O
1763 * where anything might happen inside guest memory.
1765 void *bounce_buffer;
1767 BlockDriver *drv = bs->drv;
1769 QEMUIOVector bounce_qiov;
1770 int64_t cluster_sector_num;
1771 int cluster_nb_sectors;
1775 /* Cover entire cluster so no additional backing file I/O is required when
1776 * allocating cluster in the image file.
1778 round_to_clusters(bs, sector_num, nb_sectors,
1779 &cluster_sector_num, &cluster_nb_sectors);
1781 trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors,
1782 cluster_sector_num, cluster_nb_sectors);
1784 iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE;
1785 iov.iov_base = bounce_buffer = qemu_blockalign(bs, iov.iov_len);
1786 qemu_iovec_init_external(&bounce_qiov, &iov, 1);
1788 ret = drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors,
1794 if (drv->bdrv_co_write_zeroes &&
1795 buffer_is_zero(bounce_buffer, iov.iov_len)) {
1796 ret = bdrv_co_do_write_zeroes(bs, cluster_sector_num,
1797 cluster_nb_sectors);
1799 ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors,
1804 /* It might be okay to ignore write errors for guest requests. If this
1805 * is a deliberate copy-on-read then we don't want to ignore the error.
1806 * Simply report it in all cases.
1811 skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE;
1812 qemu_iovec_from_buffer(qiov, bounce_buffer + skip_bytes,
1813 nb_sectors * BDRV_SECTOR_SIZE);
1816 qemu_vfree(bounce_buffer);
1821 * Handle a read request in coroutine context
1823 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
1824 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1825 BdrvRequestFlags flags)
1827 BlockDriver *drv = bs->drv;
1828 BdrvTrackedRequest req;
1834 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
1838 /* throttling disk read I/O */
1839 if (bs->io_limits_enabled) {
1840 bdrv_io_limits_intercept(bs, false, nb_sectors);
1843 if (bs->copy_on_read) {
1844 flags |= BDRV_REQ_COPY_ON_READ;
1846 if (flags & BDRV_REQ_COPY_ON_READ) {
1847 bs->copy_on_read_in_flight++;
1850 if (bs->copy_on_read_in_flight) {
1851 wait_for_overlapping_requests(bs, sector_num, nb_sectors);
1854 tracked_request_begin(&req, bs, sector_num, nb_sectors, false);
1856 if (flags & BDRV_REQ_COPY_ON_READ) {
1859 ret = bdrv_co_is_allocated(bs, sector_num, nb_sectors, &pnum);
1864 if (!ret || pnum != nb_sectors) {
1865 ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov);
1870 ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
1873 tracked_request_end(&req);
1875 if (flags & BDRV_REQ_COPY_ON_READ) {
1876 bs->copy_on_read_in_flight--;
1882 int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
1883 int nb_sectors, QEMUIOVector *qiov)
1885 trace_bdrv_co_readv(bs, sector_num, nb_sectors);
1887 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0);
1890 int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
1891 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1893 trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors);
1895 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
1896 BDRV_REQ_COPY_ON_READ);
1899 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
1900 int64_t sector_num, int nb_sectors)
1902 BlockDriver *drv = bs->drv;
1907 /* TODO Emulate only part of misaligned requests instead of letting block
1908 * drivers return -ENOTSUP and emulate everything */
1910 /* First try the efficient write zeroes operation */
1911 if (drv->bdrv_co_write_zeroes) {
1912 ret = drv->bdrv_co_write_zeroes(bs, sector_num, nb_sectors);
1913 if (ret != -ENOTSUP) {
1918 /* Fall back to bounce buffer if write zeroes is unsupported */
1919 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
1920 iov.iov_base = qemu_blockalign(bs, iov.iov_len);
1921 memset(iov.iov_base, 0, iov.iov_len);
1922 qemu_iovec_init_external(&qiov, &iov, 1);
1924 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, &qiov);
1926 qemu_vfree(iov.iov_base);
1931 * Handle a write request in coroutine context
1933 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
1934 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1935 BdrvRequestFlags flags)
1937 BlockDriver *drv = bs->drv;
1938 BdrvTrackedRequest req;
1944 if (bs->read_only) {
1947 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
1951 /* throttling disk write I/O */
1952 if (bs->io_limits_enabled) {
1953 bdrv_io_limits_intercept(bs, true, nb_sectors);
1956 if (bs->copy_on_read_in_flight) {
1957 wait_for_overlapping_requests(bs, sector_num, nb_sectors);
1960 tracked_request_begin(&req, bs, sector_num, nb_sectors, true);
1962 if (flags & BDRV_REQ_ZERO_WRITE) {
1963 ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors);
1965 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
1968 if (bs->dirty_bitmap) {
1969 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1972 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
1973 bs->wr_highest_sector = sector_num + nb_sectors - 1;
1976 tracked_request_end(&req);
1981 int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
1982 int nb_sectors, QEMUIOVector *qiov)
1984 trace_bdrv_co_writev(bs, sector_num, nb_sectors);
1986 return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0);
1989 int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs,
1990 int64_t sector_num, int nb_sectors)
1992 trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors);
1994 return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL,
1995 BDRV_REQ_ZERO_WRITE);
1999 * Truncate file to 'offset' bytes (needed only for file protocols)
2001 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
2003 BlockDriver *drv = bs->drv;
2007 if (!drv->bdrv_truncate)
2011 if (bdrv_in_use(bs))
2013 ret = drv->bdrv_truncate(bs, offset);
2015 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
2016 bdrv_dev_resize_cb(bs);
2022 * Length of a allocated file in bytes. Sparse files are counted by actual
2023 * allocated space. Return < 0 if error or unknown.
2025 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
2027 BlockDriver *drv = bs->drv;
2031 if (drv->bdrv_get_allocated_file_size) {
2032 return drv->bdrv_get_allocated_file_size(bs);
2035 return bdrv_get_allocated_file_size(bs->file);
2041 * Length of a file in bytes. Return < 0 if error or unknown.
2043 int64_t bdrv_getlength(BlockDriverState *bs)
2045 BlockDriver *drv = bs->drv;
2049 if (bs->growable || bdrv_dev_has_removable_media(bs)) {
2050 if (drv->bdrv_getlength) {
2051 return drv->bdrv_getlength(bs);
2054 return bs->total_sectors * BDRV_SECTOR_SIZE;
2057 /* return 0 as number of sectors if no device present or error */
2058 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
2061 length = bdrv_getlength(bs);
2065 length = length >> BDRV_SECTOR_BITS;
2066 *nb_sectors_ptr = length;
2070 uint8_t boot_ind; /* 0x80 - active */
2071 uint8_t head; /* starting head */
2072 uint8_t sector; /* starting sector */
2073 uint8_t cyl; /* starting cylinder */
2074 uint8_t sys_ind; /* What partition type */
2075 uint8_t end_head; /* end head */
2076 uint8_t end_sector; /* end sector */
2077 uint8_t end_cyl; /* end cylinder */
2078 uint32_t start_sect; /* starting sector counting from 0 */
2079 uint32_t nr_sects; /* nr of sectors in partition */
2082 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
2083 static int guess_disk_lchs(BlockDriverState *bs,
2084 int *pcylinders, int *pheads, int *psectors)
2086 uint8_t buf[BDRV_SECTOR_SIZE];
2087 int ret, i, heads, sectors, cylinders;
2088 struct partition *p;
2090 uint64_t nb_sectors;
2093 bdrv_get_geometry(bs, &nb_sectors);
2096 * The function will be invoked during startup not only in sync I/O mode,
2097 * but also in async I/O mode. So the I/O throttling function has to
2098 * be disabled temporarily here, not permanently.
2100 enabled = bs->io_limits_enabled;
2101 bs->io_limits_enabled = false;
2102 ret = bdrv_read(bs, 0, buf, 1);
2103 bs->io_limits_enabled = enabled;
2106 /* test msdos magic */
2107 if (buf[510] != 0x55 || buf[511] != 0xaa)
2109 for(i = 0; i < 4; i++) {
2110 p = ((struct partition *)(buf + 0x1be)) + i;
2111 nr_sects = le32_to_cpu(p->nr_sects);
2112 if (nr_sects && p->end_head) {
2113 /* We make the assumption that the partition terminates on
2114 a cylinder boundary */
2115 heads = p->end_head + 1;
2116 sectors = p->end_sector & 63;
2119 cylinders = nb_sectors / (heads * sectors);
2120 if (cylinders < 1 || cylinders > 16383)
2123 *psectors = sectors;
2124 *pcylinders = cylinders;
2126 printf("guessed geometry: LCHS=%d %d %d\n",
2127 cylinders, heads, sectors);
2135 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
2137 int translation, lba_detected = 0;
2138 int cylinders, heads, secs;
2139 uint64_t nb_sectors;
2141 /* if a geometry hint is available, use it */
2142 bdrv_get_geometry(bs, &nb_sectors);
2143 bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
2144 translation = bdrv_get_translation_hint(bs);
2145 if (cylinders != 0) {
2150 if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
2152 /* if heads > 16, it means that a BIOS LBA
2153 translation was active, so the default
2154 hardware geometry is OK */
2156 goto default_geometry;
2161 /* disable any translation to be in sync with
2162 the logical geometry */
2163 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
2164 bdrv_set_translation_hint(bs,
2165 BIOS_ATA_TRANSLATION_NONE);
2170 /* if no geometry, use a standard physical disk geometry */
2171 cylinders = nb_sectors / (16 * 63);
2173 if (cylinders > 16383)
2175 else if (cylinders < 2)
2180 if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
2181 if ((*pcyls * *pheads) <= 131072) {
2182 bdrv_set_translation_hint(bs,
2183 BIOS_ATA_TRANSLATION_LARGE);
2185 bdrv_set_translation_hint(bs,
2186 BIOS_ATA_TRANSLATION_LBA);
2190 bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
2194 void bdrv_set_geometry_hint(BlockDriverState *bs,
2195 int cyls, int heads, int secs)
2202 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
2204 bs->translation = translation;
2207 void bdrv_get_geometry_hint(BlockDriverState *bs,
2208 int *pcyls, int *pheads, int *psecs)
2211 *pheads = bs->heads;
2215 /* throttling disk io limits */
2216 void bdrv_set_io_limits(BlockDriverState *bs,
2217 BlockIOLimit *io_limits)
2219 bs->io_limits = *io_limits;
2220 bs->io_limits_enabled = bdrv_io_limits_enabled(bs);
2223 /* Recognize floppy formats */
2224 typedef struct FDFormat {
2232 static const FDFormat fd_formats[] = {
2233 /* First entry is default format */
2234 /* 1.44 MB 3"1/2 floppy disks */
2235 { FDRIVE_DRV_144, 18, 80, 1, FDRIVE_RATE_500K, },
2236 { FDRIVE_DRV_144, 20, 80, 1, FDRIVE_RATE_500K, },
2237 { FDRIVE_DRV_144, 21, 80, 1, FDRIVE_RATE_500K, },
2238 { FDRIVE_DRV_144, 21, 82, 1, FDRIVE_RATE_500K, },
2239 { FDRIVE_DRV_144, 21, 83, 1, FDRIVE_RATE_500K, },
2240 { FDRIVE_DRV_144, 22, 80, 1, FDRIVE_RATE_500K, },
2241 { FDRIVE_DRV_144, 23, 80, 1, FDRIVE_RATE_500K, },
2242 { FDRIVE_DRV_144, 24, 80, 1, FDRIVE_RATE_500K, },
2243 /* 2.88 MB 3"1/2 floppy disks */
2244 { FDRIVE_DRV_288, 36, 80, 1, FDRIVE_RATE_1M, },
2245 { FDRIVE_DRV_288, 39, 80, 1, FDRIVE_RATE_1M, },
2246 { FDRIVE_DRV_288, 40, 80, 1, FDRIVE_RATE_1M, },
2247 { FDRIVE_DRV_288, 44, 80, 1, FDRIVE_RATE_1M, },
2248 { FDRIVE_DRV_288, 48, 80, 1, FDRIVE_RATE_1M, },
2249 /* 720 kB 3"1/2 floppy disks */
2250 { FDRIVE_DRV_144, 9, 80, 1, FDRIVE_RATE_250K, },
2251 { FDRIVE_DRV_144, 10, 80, 1, FDRIVE_RATE_250K, },
2252 { FDRIVE_DRV_144, 10, 82, 1, FDRIVE_RATE_250K, },
2253 { FDRIVE_DRV_144, 10, 83, 1, FDRIVE_RATE_250K, },
2254 { FDRIVE_DRV_144, 13, 80, 1, FDRIVE_RATE_250K, },
2255 { FDRIVE_DRV_144, 14, 80, 1, FDRIVE_RATE_250K, },
2256 /* 1.2 MB 5"1/4 floppy disks */
2257 { FDRIVE_DRV_120, 15, 80, 1, FDRIVE_RATE_500K, },
2258 { FDRIVE_DRV_120, 18, 80, 1, FDRIVE_RATE_500K, },
2259 { FDRIVE_DRV_120, 18, 82, 1, FDRIVE_RATE_500K, },
2260 { FDRIVE_DRV_120, 18, 83, 1, FDRIVE_RATE_500K, },
2261 { FDRIVE_DRV_120, 20, 80, 1, FDRIVE_RATE_500K, },
2262 /* 720 kB 5"1/4 floppy disks */
2263 { FDRIVE_DRV_120, 9, 80, 1, FDRIVE_RATE_250K, },
2264 { FDRIVE_DRV_120, 11, 80, 1, FDRIVE_RATE_250K, },
2265 /* 360 kB 5"1/4 floppy disks */
2266 { FDRIVE_DRV_120, 9, 40, 1, FDRIVE_RATE_300K, },
2267 { FDRIVE_DRV_120, 9, 40, 0, FDRIVE_RATE_300K, },
2268 { FDRIVE_DRV_120, 10, 41, 1, FDRIVE_RATE_300K, },
2269 { FDRIVE_DRV_120, 10, 42, 1, FDRIVE_RATE_300K, },
2270 /* 320 kB 5"1/4 floppy disks */
2271 { FDRIVE_DRV_120, 8, 40, 1, FDRIVE_RATE_250K, },
2272 { FDRIVE_DRV_120, 8, 40, 0, FDRIVE_RATE_250K, },
2273 /* 360 kB must match 5"1/4 better than 3"1/2... */
2274 { FDRIVE_DRV_144, 9, 80, 0, FDRIVE_RATE_250K, },
2276 { FDRIVE_DRV_NONE, -1, -1, 0, 0, },
2279 void bdrv_get_floppy_geometry_hint(BlockDriverState *bs, int *nb_heads,
2280 int *max_track, int *last_sect,
2281 FDriveType drive_in, FDriveType *drive,
2284 const FDFormat *parse;
2285 uint64_t nb_sectors, size;
2286 int i, first_match, match;
2288 bdrv_get_geometry_hint(bs, nb_heads, max_track, last_sect);
2289 if (*nb_heads != 0 && *max_track != 0 && *last_sect != 0) {
2290 /* User defined disk */
2291 *rate = FDRIVE_RATE_500K;
2293 bdrv_get_geometry(bs, &nb_sectors);
2296 for (i = 0; ; i++) {
2297 parse = &fd_formats[i];
2298 if (parse->drive == FDRIVE_DRV_NONE) {
2301 if (drive_in == parse->drive ||
2302 drive_in == FDRIVE_DRV_NONE) {
2303 size = (parse->max_head + 1) * parse->max_track *
2305 if (nb_sectors == size) {
2309 if (first_match == -1) {
2315 if (first_match == -1) {
2318 match = first_match;
2320 parse = &fd_formats[match];
2322 *nb_heads = parse->max_head + 1;
2323 *max_track = parse->max_track;
2324 *last_sect = parse->last_sect;
2325 *drive = parse->drive;
2326 *rate = parse->rate;
2330 int bdrv_get_translation_hint(BlockDriverState *bs)
2332 return bs->translation;
2335 void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
2336 BlockErrorAction on_write_error)
2338 bs->on_read_error = on_read_error;
2339 bs->on_write_error = on_write_error;
2342 BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
2344 return is_read ? bs->on_read_error : bs->on_write_error;
2347 int bdrv_is_read_only(BlockDriverState *bs)
2349 return bs->read_only;
2352 int bdrv_is_sg(BlockDriverState *bs)
2357 int bdrv_enable_write_cache(BlockDriverState *bs)
2359 return bs->enable_write_cache;
2362 int bdrv_is_encrypted(BlockDriverState *bs)
2364 if (bs->backing_hd && bs->backing_hd->encrypted)
2366 return bs->encrypted;
2369 int bdrv_key_required(BlockDriverState *bs)
2371 BlockDriverState *backing_hd = bs->backing_hd;
2373 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
2375 return (bs->encrypted && !bs->valid_key);
2378 int bdrv_set_key(BlockDriverState *bs, const char *key)
2381 if (bs->backing_hd && bs->backing_hd->encrypted) {
2382 ret = bdrv_set_key(bs->backing_hd, key);
2388 if (!bs->encrypted) {
2390 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
2393 ret = bs->drv->bdrv_set_key(bs, key);
2396 } else if (!bs->valid_key) {
2398 /* call the change callback now, we skipped it on open */
2399 bdrv_dev_change_media_cb(bs, true);
2404 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
2409 pstrcpy(buf, buf_size, bs->drv->format_name);
2413 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
2418 QLIST_FOREACH(drv, &bdrv_drivers, list) {
2419 it(opaque, drv->format_name);
2423 BlockDriverState *bdrv_find(const char *name)
2425 BlockDriverState *bs;
2427 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2428 if (!strcmp(name, bs->device_name)) {
2435 BlockDriverState *bdrv_next(BlockDriverState *bs)
2438 return QTAILQ_FIRST(&bdrv_states);
2440 return QTAILQ_NEXT(bs, list);
2443 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
2445 BlockDriverState *bs;
2447 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2452 const char *bdrv_get_device_name(BlockDriverState *bs)
2454 return bs->device_name;
2457 void bdrv_flush_all(void)
2459 BlockDriverState *bs;
2461 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2466 int bdrv_has_zero_init(BlockDriverState *bs)
2470 if (bs->drv->bdrv_has_zero_init) {
2471 return bs->drv->bdrv_has_zero_init(bs);
2477 typedef struct BdrvCoIsAllocatedData {
2478 BlockDriverState *bs;
2484 } BdrvCoIsAllocatedData;
2487 * Returns true iff the specified sector is present in the disk image. Drivers
2488 * not implementing the functionality are assumed to not support backing files,
2489 * hence all their sectors are reported as allocated.
2491 * If 'sector_num' is beyond the end of the disk image the return value is 0
2492 * and 'pnum' is set to 0.
2494 * 'pnum' is set to the number of sectors (including and immediately following
2495 * the specified sector) that are known to be in the same
2496 * allocated/unallocated state.
2498 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
2499 * beyond the end of the disk image it will be clamped.
2501 int coroutine_fn bdrv_co_is_allocated(BlockDriverState *bs, int64_t sector_num,
2502 int nb_sectors, int *pnum)
2506 if (sector_num >= bs->total_sectors) {
2511 n = bs->total_sectors - sector_num;
2512 if (n < nb_sectors) {
2516 if (!bs->drv->bdrv_co_is_allocated) {
2521 return bs->drv->bdrv_co_is_allocated(bs, sector_num, nb_sectors, pnum);
2524 /* Coroutine wrapper for bdrv_is_allocated() */
2525 static void coroutine_fn bdrv_is_allocated_co_entry(void *opaque)
2527 BdrvCoIsAllocatedData *data = opaque;
2528 BlockDriverState *bs = data->bs;
2530 data->ret = bdrv_co_is_allocated(bs, data->sector_num, data->nb_sectors,
2536 * Synchronous wrapper around bdrv_co_is_allocated().
2538 * See bdrv_co_is_allocated() for details.
2540 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
2544 BdrvCoIsAllocatedData data = {
2546 .sector_num = sector_num,
2547 .nb_sectors = nb_sectors,
2552 co = qemu_coroutine_create(bdrv_is_allocated_co_entry);
2553 qemu_coroutine_enter(co, &data);
2554 while (!data.done) {
2560 BlockInfoList *qmp_query_block(Error **errp)
2562 BlockInfoList *head = NULL, *cur_item = NULL;
2563 BlockDriverState *bs;
2565 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2566 BlockInfoList *info = g_malloc0(sizeof(*info));
2568 info->value = g_malloc0(sizeof(*info->value));
2569 info->value->device = g_strdup(bs->device_name);
2570 info->value->type = g_strdup("unknown");
2571 info->value->locked = bdrv_dev_is_medium_locked(bs);
2572 info->value->removable = bdrv_dev_has_removable_media(bs);
2574 if (bdrv_dev_has_removable_media(bs)) {
2575 info->value->has_tray_open = true;
2576 info->value->tray_open = bdrv_dev_is_tray_open(bs);
2579 if (bdrv_iostatus_is_enabled(bs)) {
2580 info->value->has_io_status = true;
2581 info->value->io_status = bs->iostatus;
2585 info->value->has_inserted = true;
2586 info->value->inserted = g_malloc0(sizeof(*info->value->inserted));
2587 info->value->inserted->file = g_strdup(bs->filename);
2588 info->value->inserted->ro = bs->read_only;
2589 info->value->inserted->drv = g_strdup(bs->drv->format_name);
2590 info->value->inserted->encrypted = bs->encrypted;
2591 if (bs->backing_file[0]) {
2592 info->value->inserted->has_backing_file = true;
2593 info->value->inserted->backing_file = g_strdup(bs->backing_file);
2596 if (bs->io_limits_enabled) {
2597 info->value->inserted->bps =
2598 bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL];
2599 info->value->inserted->bps_rd =
2600 bs->io_limits.bps[BLOCK_IO_LIMIT_READ];
2601 info->value->inserted->bps_wr =
2602 bs->io_limits.bps[BLOCK_IO_LIMIT_WRITE];
2603 info->value->inserted->iops =
2604 bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL];
2605 info->value->inserted->iops_rd =
2606 bs->io_limits.iops[BLOCK_IO_LIMIT_READ];
2607 info->value->inserted->iops_wr =
2608 bs->io_limits.iops[BLOCK_IO_LIMIT_WRITE];
2612 /* XXX: waiting for the qapi to support GSList */
2614 head = cur_item = info;
2616 cur_item->next = info;
2624 /* Consider exposing this as a full fledged QMP command */
2625 static BlockStats *qmp_query_blockstat(const BlockDriverState *bs, Error **errp)
2629 s = g_malloc0(sizeof(*s));
2631 if (bs->device_name[0]) {
2632 s->has_device = true;
2633 s->device = g_strdup(bs->device_name);
2636 s->stats = g_malloc0(sizeof(*s->stats));
2637 s->stats->rd_bytes = bs->nr_bytes[BDRV_ACCT_READ];
2638 s->stats->wr_bytes = bs->nr_bytes[BDRV_ACCT_WRITE];
2639 s->stats->rd_operations = bs->nr_ops[BDRV_ACCT_READ];
2640 s->stats->wr_operations = bs->nr_ops[BDRV_ACCT_WRITE];
2641 s->stats->wr_highest_offset = bs->wr_highest_sector * BDRV_SECTOR_SIZE;
2642 s->stats->flush_operations = bs->nr_ops[BDRV_ACCT_FLUSH];
2643 s->stats->wr_total_time_ns = bs->total_time_ns[BDRV_ACCT_WRITE];
2644 s->stats->rd_total_time_ns = bs->total_time_ns[BDRV_ACCT_READ];
2645 s->stats->flush_total_time_ns = bs->total_time_ns[BDRV_ACCT_FLUSH];
2648 s->has_parent = true;
2649 s->parent = qmp_query_blockstat(bs->file, NULL);
2655 BlockStatsList *qmp_query_blockstats(Error **errp)
2657 BlockStatsList *head = NULL, *cur_item = NULL;
2658 BlockDriverState *bs;
2660 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2661 BlockStatsList *info = g_malloc0(sizeof(*info));
2662 info->value = qmp_query_blockstat(bs, NULL);
2664 /* XXX: waiting for the qapi to support GSList */
2666 head = cur_item = info;
2668 cur_item->next = info;
2676 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
2678 if (bs->backing_hd && bs->backing_hd->encrypted)
2679 return bs->backing_file;
2680 else if (bs->encrypted)
2681 return bs->filename;
2686 void bdrv_get_backing_filename(BlockDriverState *bs,
2687 char *filename, int filename_size)
2689 pstrcpy(filename, filename_size, bs->backing_file);
2692 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
2693 const uint8_t *buf, int nb_sectors)
2695 BlockDriver *drv = bs->drv;
2698 if (!drv->bdrv_write_compressed)
2700 if (bdrv_check_request(bs, sector_num, nb_sectors))
2703 if (bs->dirty_bitmap) {
2704 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
2707 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
2710 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2712 BlockDriver *drv = bs->drv;
2715 if (!drv->bdrv_get_info)
2717 memset(bdi, 0, sizeof(*bdi));
2718 return drv->bdrv_get_info(bs, bdi);
2721 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
2722 int64_t pos, int size)
2724 BlockDriver *drv = bs->drv;
2727 if (drv->bdrv_save_vmstate)
2728 return drv->bdrv_save_vmstate(bs, buf, pos, size);
2730 return bdrv_save_vmstate(bs->file, buf, pos, size);
2734 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
2735 int64_t pos, int size)
2737 BlockDriver *drv = bs->drv;
2740 if (drv->bdrv_load_vmstate)
2741 return drv->bdrv_load_vmstate(bs, buf, pos, size);
2743 return bdrv_load_vmstate(bs->file, buf, pos, size);
2747 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
2749 BlockDriver *drv = bs->drv;
2751 if (!drv || !drv->bdrv_debug_event) {
2755 return drv->bdrv_debug_event(bs, event);
2759 /**************************************************************/
2760 /* handling of snapshots */
2762 int bdrv_can_snapshot(BlockDriverState *bs)
2764 BlockDriver *drv = bs->drv;
2765 if (!drv || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
2769 if (!drv->bdrv_snapshot_create) {
2770 if (bs->file != NULL) {
2771 return bdrv_can_snapshot(bs->file);
2779 int bdrv_is_snapshot(BlockDriverState *bs)
2781 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
2784 BlockDriverState *bdrv_snapshots(void)
2786 BlockDriverState *bs;
2789 return bs_snapshots;
2793 while ((bs = bdrv_next(bs))) {
2794 if (bdrv_can_snapshot(bs)) {
2802 int bdrv_snapshot_create(BlockDriverState *bs,
2803 QEMUSnapshotInfo *sn_info)
2805 BlockDriver *drv = bs->drv;
2808 if (drv->bdrv_snapshot_create)
2809 return drv->bdrv_snapshot_create(bs, sn_info);
2811 return bdrv_snapshot_create(bs->file, sn_info);
2815 int bdrv_snapshot_goto(BlockDriverState *bs,
2816 const char *snapshot_id)
2818 BlockDriver *drv = bs->drv;
2823 if (drv->bdrv_snapshot_goto)
2824 return drv->bdrv_snapshot_goto(bs, snapshot_id);
2827 drv->bdrv_close(bs);
2828 ret = bdrv_snapshot_goto(bs->file, snapshot_id);
2829 open_ret = drv->bdrv_open(bs, bs->open_flags);
2831 bdrv_delete(bs->file);
2841 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2843 BlockDriver *drv = bs->drv;
2846 if (drv->bdrv_snapshot_delete)
2847 return drv->bdrv_snapshot_delete(bs, snapshot_id);
2849 return bdrv_snapshot_delete(bs->file, snapshot_id);
2853 int bdrv_snapshot_list(BlockDriverState *bs,
2854 QEMUSnapshotInfo **psn_info)
2856 BlockDriver *drv = bs->drv;
2859 if (drv->bdrv_snapshot_list)
2860 return drv->bdrv_snapshot_list(bs, psn_info);
2862 return bdrv_snapshot_list(bs->file, psn_info);
2866 int bdrv_snapshot_load_tmp(BlockDriverState *bs,
2867 const char *snapshot_name)
2869 BlockDriver *drv = bs->drv;
2873 if (!bs->read_only) {
2876 if (drv->bdrv_snapshot_load_tmp) {
2877 return drv->bdrv_snapshot_load_tmp(bs, snapshot_name);
2882 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
2883 const char *backing_file)
2889 if (bs->backing_hd) {
2890 if (strcmp(bs->backing_file, backing_file) == 0) {
2891 return bs->backing_hd;
2893 return bdrv_find_backing_image(bs->backing_hd, backing_file);
2900 #define NB_SUFFIXES 4
2902 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
2904 static const char suffixes[NB_SUFFIXES] = "KMGT";
2909 snprintf(buf, buf_size, "%" PRId64, size);
2912 for(i = 0; i < NB_SUFFIXES; i++) {
2913 if (size < (10 * base)) {
2914 snprintf(buf, buf_size, "%0.1f%c",
2915 (double)size / base,
2918 } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
2919 snprintf(buf, buf_size, "%" PRId64 "%c",
2920 ((size + (base >> 1)) / base),
2930 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
2932 char buf1[128], date_buf[128], clock_buf[128];
2942 snprintf(buf, buf_size,
2943 "%-10s%-20s%7s%20s%15s",
2944 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
2948 ptm = localtime(&ti);
2949 strftime(date_buf, sizeof(date_buf),
2950 "%Y-%m-%d %H:%M:%S", ptm);
2952 localtime_r(&ti, &tm);
2953 strftime(date_buf, sizeof(date_buf),
2954 "%Y-%m-%d %H:%M:%S", &tm);
2956 secs = sn->vm_clock_nsec / 1000000000;
2957 snprintf(clock_buf, sizeof(clock_buf),
2958 "%02d:%02d:%02d.%03d",
2960 (int)((secs / 60) % 60),
2962 (int)((sn->vm_clock_nsec / 1000000) % 1000));
2963 snprintf(buf, buf_size,
2964 "%-10s%-20s%7s%20s%15s",
2965 sn->id_str, sn->name,
2966 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
2973 /**************************************************************/
2976 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
2977 QEMUIOVector *qiov, int nb_sectors,
2978 BlockDriverCompletionFunc *cb, void *opaque)
2980 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
2982 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
2986 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
2987 QEMUIOVector *qiov, int nb_sectors,
2988 BlockDriverCompletionFunc *cb, void *opaque)
2990 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
2992 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
2997 typedef struct MultiwriteCB {
3002 BlockDriverCompletionFunc *cb;
3004 QEMUIOVector *free_qiov;
3008 static void multiwrite_user_cb(MultiwriteCB *mcb)
3012 for (i = 0; i < mcb->num_callbacks; i++) {
3013 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
3014 if (mcb->callbacks[i].free_qiov) {
3015 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
3017 g_free(mcb->callbacks[i].free_qiov);
3021 static void multiwrite_cb(void *opaque, int ret)
3023 MultiwriteCB *mcb = opaque;
3025 trace_multiwrite_cb(mcb, ret);
3027 if (ret < 0 && !mcb->error) {
3031 mcb->num_requests--;
3032 if (mcb->num_requests == 0) {
3033 multiwrite_user_cb(mcb);
3038 static int multiwrite_req_compare(const void *a, const void *b)
3040 const BlockRequest *req1 = a, *req2 = b;
3043 * Note that we can't simply subtract req2->sector from req1->sector
3044 * here as that could overflow the return value.
3046 if (req1->sector > req2->sector) {
3048 } else if (req1->sector < req2->sector) {
3056 * Takes a bunch of requests and tries to merge them. Returns the number of
3057 * requests that remain after merging.
3059 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
3060 int num_reqs, MultiwriteCB *mcb)
3064 // Sort requests by start sector
3065 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
3067 // Check if adjacent requests touch the same clusters. If so, combine them,
3068 // filling up gaps with zero sectors.
3070 for (i = 1; i < num_reqs; i++) {
3072 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
3074 // Handle exactly sequential writes and overlapping writes.
3075 if (reqs[i].sector <= oldreq_last) {
3079 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
3085 QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
3086 qemu_iovec_init(qiov,
3087 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
3089 // Add the first request to the merged one. If the requests are
3090 // overlapping, drop the last sectors of the first request.
3091 size = (reqs[i].sector - reqs[outidx].sector) << 9;
3092 qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
3094 // We should need to add any zeros between the two requests
3095 assert (reqs[i].sector <= oldreq_last);
3097 // Add the second request
3098 qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
3100 reqs[outidx].nb_sectors = qiov->size >> 9;
3101 reqs[outidx].qiov = qiov;
3103 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
3106 reqs[outidx].sector = reqs[i].sector;
3107 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
3108 reqs[outidx].qiov = reqs[i].qiov;
3116 * Submit multiple AIO write requests at once.
3118 * On success, the function returns 0 and all requests in the reqs array have
3119 * been submitted. In error case this function returns -1, and any of the
3120 * requests may or may not be submitted yet. In particular, this means that the
3121 * callback will be called for some of the requests, for others it won't. The
3122 * caller must check the error field of the BlockRequest to wait for the right
3123 * callbacks (if error != 0, no callback will be called).
3125 * The implementation may modify the contents of the reqs array, e.g. to merge
3126 * requests. However, the fields opaque and error are left unmodified as they
3127 * are used to signal failure for a single request to the caller.
3129 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
3134 /* don't submit writes if we don't have a medium */
3135 if (bs->drv == NULL) {
3136 for (i = 0; i < num_reqs; i++) {
3137 reqs[i].error = -ENOMEDIUM;
3142 if (num_reqs == 0) {
3146 // Create MultiwriteCB structure
3147 mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
3148 mcb->num_requests = 0;
3149 mcb->num_callbacks = num_reqs;
3151 for (i = 0; i < num_reqs; i++) {
3152 mcb->callbacks[i].cb = reqs[i].cb;
3153 mcb->callbacks[i].opaque = reqs[i].opaque;
3156 // Check for mergable requests
3157 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
3159 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
3161 /* Run the aio requests. */
3162 mcb->num_requests = num_reqs;
3163 for (i = 0; i < num_reqs; i++) {
3164 bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
3165 reqs[i].nb_sectors, multiwrite_cb, mcb);
3171 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
3173 acb->pool->cancel(acb);
3176 /* block I/O throttling */
3177 static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors,
3178 bool is_write, double elapsed_time, uint64_t *wait)
3180 uint64_t bps_limit = 0;
3181 double bytes_limit, bytes_base, bytes_res;
3182 double slice_time, wait_time;
3184 if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
3185 bps_limit = bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL];
3186 } else if (bs->io_limits.bps[is_write]) {
3187 bps_limit = bs->io_limits.bps[is_write];
3196 slice_time = bs->slice_end - bs->slice_start;
3197 slice_time /= (NANOSECONDS_PER_SECOND);
3198 bytes_limit = bps_limit * slice_time;
3199 bytes_base = bs->nr_bytes[is_write] - bs->io_base.bytes[is_write];
3200 if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
3201 bytes_base += bs->nr_bytes[!is_write] - bs->io_base.bytes[!is_write];
3204 /* bytes_base: the bytes of data which have been read/written; and
3205 * it is obtained from the history statistic info.
3206 * bytes_res: the remaining bytes of data which need to be read/written.
3207 * (bytes_base + bytes_res) / bps_limit: used to calcuate
3208 * the total time for completing reading/writting all data.
3210 bytes_res = (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
3212 if (bytes_base + bytes_res <= bytes_limit) {
3220 /* Calc approx time to dispatch */
3221 wait_time = (bytes_base + bytes_res) / bps_limit - elapsed_time;
3223 /* When the I/O rate at runtime exceeds the limits,
3224 * bs->slice_end need to be extended in order that the current statistic
3225 * info can be kept until the timer fire, so it is increased and tuned
3226 * based on the result of experiment.
3228 bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10;
3229 bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME;
3231 *wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
3237 static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write,
3238 double elapsed_time, uint64_t *wait)
3240 uint64_t iops_limit = 0;
3241 double ios_limit, ios_base;
3242 double slice_time, wait_time;
3244 if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
3245 iops_limit = bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL];
3246 } else if (bs->io_limits.iops[is_write]) {
3247 iops_limit = bs->io_limits.iops[is_write];
3256 slice_time = bs->slice_end - bs->slice_start;
3257 slice_time /= (NANOSECONDS_PER_SECOND);
3258 ios_limit = iops_limit * slice_time;
3259 ios_base = bs->nr_ops[is_write] - bs->io_base.ios[is_write];
3260 if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
3261 ios_base += bs->nr_ops[!is_write] - bs->io_base.ios[!is_write];
3264 if (ios_base + 1 <= ios_limit) {
3272 /* Calc approx time to dispatch */
3273 wait_time = (ios_base + 1) / iops_limit;
3274 if (wait_time > elapsed_time) {
3275 wait_time = wait_time - elapsed_time;
3280 bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10;
3281 bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME;
3283 *wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
3289 static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors,
3290 bool is_write, int64_t *wait)
3292 int64_t now, max_wait;
3293 uint64_t bps_wait = 0, iops_wait = 0;
3294 double elapsed_time;
3295 int bps_ret, iops_ret;
3297 now = qemu_get_clock_ns(vm_clock);
3298 if ((bs->slice_start < now)
3299 && (bs->slice_end > now)) {
3300 bs->slice_end = now + bs->slice_time;
3302 bs->slice_time = 5 * BLOCK_IO_SLICE_TIME;
3303 bs->slice_start = now;
3304 bs->slice_end = now + bs->slice_time;
3306 bs->io_base.bytes[is_write] = bs->nr_bytes[is_write];
3307 bs->io_base.bytes[!is_write] = bs->nr_bytes[!is_write];
3309 bs->io_base.ios[is_write] = bs->nr_ops[is_write];
3310 bs->io_base.ios[!is_write] = bs->nr_ops[!is_write];
3313 elapsed_time = now - bs->slice_start;
3314 elapsed_time /= (NANOSECONDS_PER_SECOND);
3316 bps_ret = bdrv_exceed_bps_limits(bs, nb_sectors,
3317 is_write, elapsed_time, &bps_wait);
3318 iops_ret = bdrv_exceed_iops_limits(bs, is_write,
3319 elapsed_time, &iops_wait);
3320 if (bps_ret || iops_ret) {
3321 max_wait = bps_wait > iops_wait ? bps_wait : iops_wait;
3326 now = qemu_get_clock_ns(vm_clock);
3327 if (bs->slice_end < now + max_wait) {
3328 bs->slice_end = now + max_wait;
3341 /**************************************************************/
3342 /* async block device emulation */
3344 typedef struct BlockDriverAIOCBSync {
3345 BlockDriverAIOCB common;
3348 /* vector translation state */
3352 } BlockDriverAIOCBSync;
3354 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
3356 BlockDriverAIOCBSync *acb =
3357 container_of(blockacb, BlockDriverAIOCBSync, common);
3358 qemu_bh_delete(acb->bh);
3360 qemu_aio_release(acb);
3363 static AIOPool bdrv_em_aio_pool = {
3364 .aiocb_size = sizeof(BlockDriverAIOCBSync),
3365 .cancel = bdrv_aio_cancel_em,
3368 static void bdrv_aio_bh_cb(void *opaque)
3370 BlockDriverAIOCBSync *acb = opaque;
3373 qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
3374 qemu_vfree(acb->bounce);
3375 acb->common.cb(acb->common.opaque, acb->ret);
3376 qemu_bh_delete(acb->bh);
3378 qemu_aio_release(acb);
3381 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
3385 BlockDriverCompletionFunc *cb,
3390 BlockDriverAIOCBSync *acb;
3392 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
3393 acb->is_write = is_write;
3395 acb->bounce = qemu_blockalign(bs, qiov->size);
3396 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
3399 qemu_iovec_to_buffer(acb->qiov, acb->bounce);
3400 acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
3402 acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
3405 qemu_bh_schedule(acb->bh);
3407 return &acb->common;
3410 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
3411 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
3412 BlockDriverCompletionFunc *cb, void *opaque)
3414 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
3417 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
3418 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
3419 BlockDriverCompletionFunc *cb, void *opaque)
3421 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
3425 typedef struct BlockDriverAIOCBCoroutine {
3426 BlockDriverAIOCB common;
3430 } BlockDriverAIOCBCoroutine;
3432 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb)
3437 static AIOPool bdrv_em_co_aio_pool = {
3438 .aiocb_size = sizeof(BlockDriverAIOCBCoroutine),
3439 .cancel = bdrv_aio_co_cancel_em,
3442 static void bdrv_co_em_bh(void *opaque)
3444 BlockDriverAIOCBCoroutine *acb = opaque;
3446 acb->common.cb(acb->common.opaque, acb->req.error);
3447 qemu_bh_delete(acb->bh);
3448 qemu_aio_release(acb);
3451 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
3452 static void coroutine_fn bdrv_co_do_rw(void *opaque)
3454 BlockDriverAIOCBCoroutine *acb = opaque;
3455 BlockDriverState *bs = acb->common.bs;
3457 if (!acb->is_write) {
3458 acb->req.error = bdrv_co_do_readv(bs, acb->req.sector,
3459 acb->req.nb_sectors, acb->req.qiov, 0);
3461 acb->req.error = bdrv_co_do_writev(bs, acb->req.sector,
3462 acb->req.nb_sectors, acb->req.qiov, 0);
3465 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3466 qemu_bh_schedule(acb->bh);
3469 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
3473 BlockDriverCompletionFunc *cb,
3478 BlockDriverAIOCBCoroutine *acb;
3480 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3481 acb->req.sector = sector_num;
3482 acb->req.nb_sectors = nb_sectors;
3483 acb->req.qiov = qiov;
3484 acb->is_write = is_write;
3486 co = qemu_coroutine_create(bdrv_co_do_rw);
3487 qemu_coroutine_enter(co, acb);
3489 return &acb->common;
3492 static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
3494 BlockDriverAIOCBCoroutine *acb = opaque;
3495 BlockDriverState *bs = acb->common.bs;
3497 acb->req.error = bdrv_co_flush(bs);
3498 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3499 qemu_bh_schedule(acb->bh);
3502 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
3503 BlockDriverCompletionFunc *cb, void *opaque)
3505 trace_bdrv_aio_flush(bs, opaque);
3508 BlockDriverAIOCBCoroutine *acb;
3510 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3511 co = qemu_coroutine_create(bdrv_aio_flush_co_entry);
3512 qemu_coroutine_enter(co, acb);
3514 return &acb->common;
3517 static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque)
3519 BlockDriverAIOCBCoroutine *acb = opaque;
3520 BlockDriverState *bs = acb->common.bs;
3522 acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors);
3523 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3524 qemu_bh_schedule(acb->bh);
3527 BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs,
3528 int64_t sector_num, int nb_sectors,
3529 BlockDriverCompletionFunc *cb, void *opaque)
3532 BlockDriverAIOCBCoroutine *acb;
3534 trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque);
3536 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3537 acb->req.sector = sector_num;
3538 acb->req.nb_sectors = nb_sectors;
3539 co = qemu_coroutine_create(bdrv_aio_discard_co_entry);
3540 qemu_coroutine_enter(co, acb);
3542 return &acb->common;
3545 void bdrv_init(void)
3547 module_call_init(MODULE_INIT_BLOCK);
3550 void bdrv_init_with_whitelist(void)
3552 use_bdrv_whitelist = 1;
3556 void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
3557 BlockDriverCompletionFunc *cb, void *opaque)
3559 BlockDriverAIOCB *acb;
3561 if (pool->free_aiocb) {
3562 acb = pool->free_aiocb;
3563 pool->free_aiocb = acb->next;
3565 acb = g_malloc0(pool->aiocb_size);
3570 acb->opaque = opaque;
3574 void qemu_aio_release(void *p)
3576 BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
3577 AIOPool *pool = acb->pool;
3578 acb->next = pool->free_aiocb;
3579 pool->free_aiocb = acb;
3582 /**************************************************************/
3583 /* Coroutine block device emulation */
3585 typedef struct CoroutineIOCompletion {
3586 Coroutine *coroutine;
3588 } CoroutineIOCompletion;
3590 static void bdrv_co_io_em_complete(void *opaque, int ret)
3592 CoroutineIOCompletion *co = opaque;
3595 qemu_coroutine_enter(co->coroutine, NULL);
3598 static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num,
3599 int nb_sectors, QEMUIOVector *iov,
3602 CoroutineIOCompletion co = {
3603 .coroutine = qemu_coroutine_self(),
3605 BlockDriverAIOCB *acb;
3608 acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors,
3609 bdrv_co_io_em_complete, &co);
3611 acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors,
3612 bdrv_co_io_em_complete, &co);
3615 trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb);
3619 qemu_coroutine_yield();
3624 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
3625 int64_t sector_num, int nb_sectors,
3628 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false);
3631 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
3632 int64_t sector_num, int nb_sectors,
3635 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true);
3638 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
3640 RwCo *rwco = opaque;
3642 rwco->ret = bdrv_co_flush(rwco->bs);
3645 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
3649 if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
3653 /* Write back cached data to the OS even with cache=unsafe */
3654 if (bs->drv->bdrv_co_flush_to_os) {
3655 ret = bs->drv->bdrv_co_flush_to_os(bs);
3661 /* But don't actually force it to the disk with cache=unsafe */
3662 if (bs->open_flags & BDRV_O_NO_FLUSH) {
3666 if (bs->drv->bdrv_co_flush_to_disk) {
3667 ret = bs->drv->bdrv_co_flush_to_disk(bs);
3668 } else if (bs->drv->bdrv_aio_flush) {
3669 BlockDriverAIOCB *acb;
3670 CoroutineIOCompletion co = {
3671 .coroutine = qemu_coroutine_self(),
3674 acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
3678 qemu_coroutine_yield();
3683 * Some block drivers always operate in either writethrough or unsafe
3684 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
3685 * know how the server works (because the behaviour is hardcoded or
3686 * depends on server-side configuration), so we can't ensure that
3687 * everything is safe on disk. Returning an error doesn't work because
3688 * that would break guests even if the server operates in writethrough
3691 * Let's hope the user knows what he's doing.
3699 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
3700 * in the case of cache=unsafe, so there are no useless flushes.
3702 return bdrv_co_flush(bs->file);
3705 void bdrv_invalidate_cache(BlockDriverState *bs)
3707 if (bs->drv && bs->drv->bdrv_invalidate_cache) {
3708 bs->drv->bdrv_invalidate_cache(bs);
3712 void bdrv_invalidate_cache_all(void)
3714 BlockDriverState *bs;
3716 QTAILQ_FOREACH(bs, &bdrv_states, list) {
3717 bdrv_invalidate_cache(bs);
3721 void bdrv_clear_incoming_migration_all(void)
3723 BlockDriverState *bs;
3725 QTAILQ_FOREACH(bs, &bdrv_states, list) {
3726 bs->open_flags = bs->open_flags & ~(BDRV_O_INCOMING);
3730 int bdrv_flush(BlockDriverState *bs)
3738 if (qemu_in_coroutine()) {
3739 /* Fast-path if already in coroutine context */
3740 bdrv_flush_co_entry(&rwco);
3742 co = qemu_coroutine_create(bdrv_flush_co_entry);
3743 qemu_coroutine_enter(co, &rwco);
3744 while (rwco.ret == NOT_DONE) {
3752 static void coroutine_fn bdrv_discard_co_entry(void *opaque)
3754 RwCo *rwco = opaque;
3756 rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors);
3759 int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
3764 } else if (bdrv_check_request(bs, sector_num, nb_sectors)) {
3766 } else if (bs->read_only) {
3768 } else if (bs->drv->bdrv_co_discard) {
3769 return bs->drv->bdrv_co_discard(bs, sector_num, nb_sectors);
3770 } else if (bs->drv->bdrv_aio_discard) {
3771 BlockDriverAIOCB *acb;
3772 CoroutineIOCompletion co = {
3773 .coroutine = qemu_coroutine_self(),
3776 acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors,
3777 bdrv_co_io_em_complete, &co);
3781 qemu_coroutine_yield();
3789 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
3794 .sector_num = sector_num,
3795 .nb_sectors = nb_sectors,
3799 if (qemu_in_coroutine()) {
3800 /* Fast-path if already in coroutine context */
3801 bdrv_discard_co_entry(&rwco);
3803 co = qemu_coroutine_create(bdrv_discard_co_entry);
3804 qemu_coroutine_enter(co, &rwco);
3805 while (rwco.ret == NOT_DONE) {
3813 /**************************************************************/
3814 /* removable device support */
3817 * Return TRUE if the media is present
3819 int bdrv_is_inserted(BlockDriverState *bs)
3821 BlockDriver *drv = bs->drv;
3825 if (!drv->bdrv_is_inserted)
3827 return drv->bdrv_is_inserted(bs);
3831 * Return whether the media changed since the last call to this
3832 * function, or -ENOTSUP if we don't know. Most drivers don't know.
3834 int bdrv_media_changed(BlockDriverState *bs)
3836 BlockDriver *drv = bs->drv;
3838 if (drv && drv->bdrv_media_changed) {
3839 return drv->bdrv_media_changed(bs);
3845 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
3847 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
3849 BlockDriver *drv = bs->drv;
3851 if (drv && drv->bdrv_eject) {
3852 drv->bdrv_eject(bs, eject_flag);
3855 if (bs->device_name[0] != '\0') {
3856 bdrv_emit_qmp_eject_event(bs, eject_flag);
3861 * Lock or unlock the media (if it is locked, the user won't be able
3862 * to eject it manually).
3864 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
3866 BlockDriver *drv = bs->drv;
3868 trace_bdrv_lock_medium(bs, locked);
3870 if (drv && drv->bdrv_lock_medium) {
3871 drv->bdrv_lock_medium(bs, locked);
3875 /* needed for generic scsi interface */
3877 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
3879 BlockDriver *drv = bs->drv;
3881 if (drv && drv->bdrv_ioctl)
3882 return drv->bdrv_ioctl(bs, req, buf);
3886 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
3887 unsigned long int req, void *buf,
3888 BlockDriverCompletionFunc *cb, void *opaque)
3890 BlockDriver *drv = bs->drv;
3892 if (drv && drv->bdrv_aio_ioctl)
3893 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
3897 void bdrv_set_buffer_alignment(BlockDriverState *bs, int align)
3899 bs->buffer_alignment = align;
3902 void *qemu_blockalign(BlockDriverState *bs, size_t size)
3904 return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
3907 void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
3909 int64_t bitmap_size;
3911 bs->dirty_count = 0;
3913 if (!bs->dirty_bitmap) {
3914 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
3915 BDRV_SECTORS_PER_DIRTY_CHUNK * BITS_PER_LONG - 1;
3916 bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * BITS_PER_LONG;
3918 bs->dirty_bitmap = g_new0(unsigned long, bitmap_size);
3921 if (bs->dirty_bitmap) {
3922 g_free(bs->dirty_bitmap);
3923 bs->dirty_bitmap = NULL;
3928 int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
3930 int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
3932 if (bs->dirty_bitmap &&
3933 (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
3934 return !!(bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
3935 (1UL << (chunk % (sizeof(unsigned long) * 8))));
3941 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
3944 set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
3947 int64_t bdrv_get_dirty_count(BlockDriverState *bs)
3949 return bs->dirty_count;
3952 void bdrv_set_in_use(BlockDriverState *bs, int in_use)
3954 assert(bs->in_use != in_use);
3955 bs->in_use = in_use;
3958 int bdrv_in_use(BlockDriverState *bs)
3963 void bdrv_iostatus_enable(BlockDriverState *bs)
3965 bs->iostatus_enabled = true;
3966 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
3969 /* The I/O status is only enabled if the drive explicitly
3970 * enables it _and_ the VM is configured to stop on errors */
3971 bool bdrv_iostatus_is_enabled(const BlockDriverState *bs)
3973 return (bs->iostatus_enabled &&
3974 (bs->on_write_error == BLOCK_ERR_STOP_ENOSPC ||
3975 bs->on_write_error == BLOCK_ERR_STOP_ANY ||
3976 bs->on_read_error == BLOCK_ERR_STOP_ANY));
3979 void bdrv_iostatus_disable(BlockDriverState *bs)
3981 bs->iostatus_enabled = false;
3984 void bdrv_iostatus_reset(BlockDriverState *bs)
3986 if (bdrv_iostatus_is_enabled(bs)) {
3987 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
3991 /* XXX: Today this is set by device models because it makes the implementation
3992 quite simple. However, the block layer knows about the error, so it's
3993 possible to implement this without device models being involved */
3994 void bdrv_iostatus_set_err(BlockDriverState *bs, int error)
3996 if (bdrv_iostatus_is_enabled(bs) &&
3997 bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
3999 bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
4000 BLOCK_DEVICE_IO_STATUS_FAILED;
4005 bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes,
4006 enum BlockAcctType type)
4008 assert(type < BDRV_MAX_IOTYPE);
4010 cookie->bytes = bytes;
4011 cookie->start_time_ns = get_clock();
4012 cookie->type = type;
4016 bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie)
4018 assert(cookie->type < BDRV_MAX_IOTYPE);
4020 bs->nr_bytes[cookie->type] += cookie->bytes;
4021 bs->nr_ops[cookie->type]++;
4022 bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns;
4025 int bdrv_img_create(const char *filename, const char *fmt,
4026 const char *base_filename, const char *base_fmt,
4027 char *options, uint64_t img_size, int flags)
4029 QEMUOptionParameter *param = NULL, *create_options = NULL;
4030 QEMUOptionParameter *backing_fmt, *backing_file, *size;
4031 BlockDriverState *bs = NULL;
4032 BlockDriver *drv, *proto_drv;
4033 BlockDriver *backing_drv = NULL;
4036 /* Find driver and parse its options */
4037 drv = bdrv_find_format(fmt);
4039 error_report("Unknown file format '%s'", fmt);
4044 proto_drv = bdrv_find_protocol(filename);
4046 error_report("Unknown protocol '%s'", filename);
4051 create_options = append_option_parameters(create_options,
4052 drv->create_options);
4053 create_options = append_option_parameters(create_options,
4054 proto_drv->create_options);
4056 /* Create parameter list with default values */
4057 param = parse_option_parameters("", create_options, param);
4059 set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size);
4061 /* Parse -o options */
4063 param = parse_option_parameters(options, create_options, param);
4064 if (param == NULL) {
4065 error_report("Invalid options for file format '%s'.", fmt);
4071 if (base_filename) {
4072 if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE,
4074 error_report("Backing file not supported for file format '%s'",
4082 if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) {
4083 error_report("Backing file format not supported for file "
4084 "format '%s'", fmt);
4090 backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE);
4091 if (backing_file && backing_file->value.s) {
4092 if (!strcmp(filename, backing_file->value.s)) {
4093 error_report("Error: Trying to create an image with the "
4094 "same filename as the backing file");
4100 backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT);
4101 if (backing_fmt && backing_fmt->value.s) {
4102 backing_drv = bdrv_find_format(backing_fmt->value.s);
4104 error_report("Unknown backing file format '%s'",
4105 backing_fmt->value.s);
4111 // The size for the image must always be specified, with one exception:
4112 // If we are using a backing file, we can obtain the size from there
4113 size = get_option_parameter(param, BLOCK_OPT_SIZE);
4114 if (size && size->value.n == -1) {
4115 if (backing_file && backing_file->value.s) {
4120 /* backing files always opened read-only */
4122 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
4126 ret = bdrv_open(bs, backing_file->value.s, back_flags, backing_drv);
4128 error_report("Could not open '%s'", backing_file->value.s);
4131 bdrv_get_geometry(bs, &size);
4134 snprintf(buf, sizeof(buf), "%" PRId64, size);
4135 set_option_parameter(param, BLOCK_OPT_SIZE, buf);
4137 error_report("Image creation needs a size parameter");
4143 printf("Formatting '%s', fmt=%s ", filename, fmt);
4144 print_option_parameters(param);
4147 ret = bdrv_create(drv, filename, param);
4150 if (ret == -ENOTSUP) {
4151 error_report("Formatting or formatting option not supported for "
4152 "file format '%s'", fmt);
4153 } else if (ret == -EFBIG) {
4154 error_report("The image size is too large for file format '%s'",
4157 error_report("%s: error while creating %s: %s", filename, fmt,
4163 free_option_parameters(create_options);
4164 free_option_parameters(param);
4173 void *block_job_create(const BlockJobType *job_type, BlockDriverState *bs,
4174 int64_t speed, BlockDriverCompletionFunc *cb,
4175 void *opaque, Error **errp)
4179 if (bs->job || bdrv_in_use(bs)) {
4180 error_set(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
4183 bdrv_set_in_use(bs, 1);
4185 job = g_malloc0(job_type->instance_size);
4186 job->job_type = job_type;
4189 job->opaque = opaque;
4192 /* Only set speed when necessary to avoid NotSupported error */
4194 Error *local_err = NULL;
4196 block_job_set_speed(job, speed, &local_err);
4197 if (error_is_set(&local_err)) {
4200 bdrv_set_in_use(bs, 0);
4201 error_propagate(errp, local_err);
4208 void block_job_complete(BlockJob *job, int ret)
4210 BlockDriverState *bs = job->bs;
4212 assert(bs->job == job);
4213 job->cb(job->opaque, ret);
4216 bdrv_set_in_use(bs, 0);
4219 void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
4221 Error *local_err = NULL;
4223 if (!job->job_type->set_speed) {
4224 error_set(errp, QERR_NOT_SUPPORTED);
4227 job->job_type->set_speed(job, speed, &local_err);
4228 if (error_is_set(&local_err)) {
4229 error_propagate(errp, local_err);
4236 void block_job_cancel(BlockJob *job)
4238 job->cancelled = true;
4241 bool block_job_is_cancelled(BlockJob *job)
4243 return job->cancelled;
4246 void block_job_cancel_sync(BlockJob *job)
4248 BlockDriverState *bs = job->bs;
4250 assert(bs->job == job);
4251 block_job_cancel(job);
4252 while (bs->job != NULL && bs->job->busy) {