2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24 #include "config-host.h"
25 #include "qemu-common.h"
28 #include "block_int.h"
31 #include "qemu-coroutine.h"
32 #include "qmp-commands.h"
33 #include "qemu-timer.h"
36 #include <sys/types.h>
38 #include <sys/ioctl.h>
39 #include <sys/queue.h>
49 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
52 BDRV_REQ_COPY_ON_READ = 0x1,
53 BDRV_REQ_ZERO_WRITE = 0x2,
56 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load);
57 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
58 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
59 BlockDriverCompletionFunc *cb, void *opaque);
60 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
61 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
62 BlockDriverCompletionFunc *cb, void *opaque);
63 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
64 int64_t sector_num, int nb_sectors,
66 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
67 int64_t sector_num, int nb_sectors,
69 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
70 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
71 BdrvRequestFlags flags);
72 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
73 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
74 BdrvRequestFlags flags);
75 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
79 BlockDriverCompletionFunc *cb,
82 static void coroutine_fn bdrv_co_do_rw(void *opaque);
83 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
84 int64_t sector_num, int nb_sectors);
86 static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors,
87 bool is_write, double elapsed_time, uint64_t *wait);
88 static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write,
89 double elapsed_time, uint64_t *wait);
90 static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors,
91 bool is_write, int64_t *wait);
93 static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
94 QTAILQ_HEAD_INITIALIZER(bdrv_states);
96 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
97 QLIST_HEAD_INITIALIZER(bdrv_drivers);
99 /* The device to use for VM snapshots */
100 static BlockDriverState *bs_snapshots;
102 /* If non-zero, use only whitelisted block drivers */
103 static int use_bdrv_whitelist;
106 static int is_windows_drive_prefix(const char *filename)
108 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
109 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
113 int is_windows_drive(const char *filename)
115 if (is_windows_drive_prefix(filename) &&
118 if (strstart(filename, "\\\\.\\", NULL) ||
119 strstart(filename, "//./", NULL))
125 /* throttling disk I/O limits */
126 void bdrv_io_limits_disable(BlockDriverState *bs)
128 bs->io_limits_enabled = false;
130 while (qemu_co_queue_next(&bs->throttled_reqs));
132 if (bs->block_timer) {
133 qemu_del_timer(bs->block_timer);
134 qemu_free_timer(bs->block_timer);
135 bs->block_timer = NULL;
141 memset(&bs->io_base, 0, sizeof(bs->io_base));
144 static void bdrv_block_timer(void *opaque)
146 BlockDriverState *bs = opaque;
148 qemu_co_queue_next(&bs->throttled_reqs);
151 void bdrv_io_limits_enable(BlockDriverState *bs)
153 qemu_co_queue_init(&bs->throttled_reqs);
154 bs->block_timer = qemu_new_timer_ns(vm_clock, bdrv_block_timer, bs);
155 bs->slice_time = 5 * BLOCK_IO_SLICE_TIME;
156 bs->slice_start = qemu_get_clock_ns(vm_clock);
157 bs->slice_end = bs->slice_start + bs->slice_time;
158 memset(&bs->io_base, 0, sizeof(bs->io_base));
159 bs->io_limits_enabled = true;
162 bool bdrv_io_limits_enabled(BlockDriverState *bs)
164 BlockIOLimit *io_limits = &bs->io_limits;
165 return io_limits->bps[BLOCK_IO_LIMIT_READ]
166 || io_limits->bps[BLOCK_IO_LIMIT_WRITE]
167 || io_limits->bps[BLOCK_IO_LIMIT_TOTAL]
168 || io_limits->iops[BLOCK_IO_LIMIT_READ]
169 || io_limits->iops[BLOCK_IO_LIMIT_WRITE]
170 || io_limits->iops[BLOCK_IO_LIMIT_TOTAL];
173 static void bdrv_io_limits_intercept(BlockDriverState *bs,
174 bool is_write, int nb_sectors)
176 int64_t wait_time = -1;
178 if (!qemu_co_queue_empty(&bs->throttled_reqs)) {
179 qemu_co_queue_wait(&bs->throttled_reqs);
182 /* In fact, we hope to keep each request's timing, in FIFO mode. The next
183 * throttled requests will not be dequeued until the current request is
184 * allowed to be serviced. So if the current request still exceeds the
185 * limits, it will be inserted to the head. All requests followed it will
186 * be still in throttled_reqs queue.
189 while (bdrv_exceed_io_limits(bs, nb_sectors, is_write, &wait_time)) {
190 qemu_mod_timer(bs->block_timer,
191 wait_time + qemu_get_clock_ns(vm_clock));
192 qemu_co_queue_wait_insert_head(&bs->throttled_reqs);
195 qemu_co_queue_next(&bs->throttled_reqs);
198 /* check if the path starts with "<protocol>:" */
199 static int path_has_protocol(const char *path)
204 if (is_windows_drive(path) ||
205 is_windows_drive_prefix(path)) {
208 p = path + strcspn(path, ":/\\");
210 p = path + strcspn(path, ":/");
216 int path_is_absolute(const char *path)
219 /* specific case for names like: "\\.\d:" */
220 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
223 return (*path == '/' || *path == '\\');
225 return (*path == '/');
229 /* if filename is absolute, just copy it to dest. Otherwise, build a
230 path to it by considering it is relative to base_path. URL are
232 void path_combine(char *dest, int dest_size,
233 const char *base_path,
234 const char *filename)
241 if (path_is_absolute(filename)) {
242 pstrcpy(dest, dest_size, filename);
244 p = strchr(base_path, ':');
249 p1 = strrchr(base_path, '/');
253 p2 = strrchr(base_path, '\\');
265 if (len > dest_size - 1)
267 memcpy(dest, base_path, len);
269 pstrcat(dest, dest_size, filename);
273 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz)
275 if (bs->backing_file[0] == '\0' || path_has_protocol(bs->backing_file)) {
276 pstrcpy(dest, sz, bs->backing_file);
278 path_combine(dest, sz, bs->filename, bs->backing_file);
282 void bdrv_register(BlockDriver *bdrv)
284 /* Block drivers without coroutine functions need emulation */
285 if (!bdrv->bdrv_co_readv) {
286 bdrv->bdrv_co_readv = bdrv_co_readv_em;
287 bdrv->bdrv_co_writev = bdrv_co_writev_em;
289 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
290 * the block driver lacks aio we need to emulate that too.
292 if (!bdrv->bdrv_aio_readv) {
293 /* add AIO emulation layer */
294 bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
295 bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
299 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
302 /* create a new block device (by default it is empty) */
303 BlockDriverState *bdrv_new(const char *device_name)
305 BlockDriverState *bs;
307 bs = g_malloc0(sizeof(BlockDriverState));
308 pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
309 if (device_name[0] != '\0') {
310 QTAILQ_INSERT_TAIL(&bdrv_states, bs, list);
312 bdrv_iostatus_disable(bs);
316 BlockDriver *bdrv_find_format(const char *format_name)
319 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
320 if (!strcmp(drv1->format_name, format_name)) {
327 static int bdrv_is_whitelisted(BlockDriver *drv)
329 static const char *whitelist[] = {
330 CONFIG_BDRV_WHITELIST
335 return 1; /* no whitelist, anything goes */
337 for (p = whitelist; *p; p++) {
338 if (!strcmp(drv->format_name, *p)) {
345 BlockDriver *bdrv_find_whitelisted_format(const char *format_name)
347 BlockDriver *drv = bdrv_find_format(format_name);
348 return drv && bdrv_is_whitelisted(drv) ? drv : NULL;
351 typedef struct CreateCo {
354 QEMUOptionParameter *options;
358 static void coroutine_fn bdrv_create_co_entry(void *opaque)
360 CreateCo *cco = opaque;
363 cco->ret = cco->drv->bdrv_create(cco->filename, cco->options);
366 int bdrv_create(BlockDriver *drv, const char* filename,
367 QEMUOptionParameter *options)
374 .filename = g_strdup(filename),
379 if (!drv->bdrv_create) {
383 if (qemu_in_coroutine()) {
384 /* Fast-path if already in coroutine context */
385 bdrv_create_co_entry(&cco);
387 co = qemu_coroutine_create(bdrv_create_co_entry);
388 qemu_coroutine_enter(co, &cco);
389 while (cco.ret == NOT_DONE) {
395 g_free(cco.filename);
400 int bdrv_create_file(const char* filename, QEMUOptionParameter *options)
404 drv = bdrv_find_protocol(filename);
409 return bdrv_create(drv, filename, options);
413 void get_tmp_filename(char *filename, int size)
415 char temp_dir[MAX_PATH];
417 GetTempPath(MAX_PATH, temp_dir);
418 GetTempFileName(temp_dir, "qem", 0, filename);
421 void get_tmp_filename(char *filename, int size)
425 /* XXX: race condition possible */
426 tmpdir = getenv("TMPDIR");
429 snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
430 fd = mkstemp(filename);
436 * Detect host devices. By convention, /dev/cdrom[N] is always
437 * recognized as a host CDROM.
439 static BlockDriver *find_hdev_driver(const char *filename)
441 int score_max = 0, score;
442 BlockDriver *drv = NULL, *d;
444 QLIST_FOREACH(d, &bdrv_drivers, list) {
445 if (d->bdrv_probe_device) {
446 score = d->bdrv_probe_device(filename);
447 if (score > score_max) {
457 BlockDriver *bdrv_find_protocol(const char *filename)
464 /* TODO Drivers without bdrv_file_open must be specified explicitly */
467 * XXX(hch): we really should not let host device detection
468 * override an explicit protocol specification, but moving this
469 * later breaks access to device names with colons in them.
470 * Thanks to the brain-dead persistent naming schemes on udev-
471 * based Linux systems those actually are quite common.
473 drv1 = find_hdev_driver(filename);
478 if (!path_has_protocol(filename)) {
479 return bdrv_find_format("file");
481 p = strchr(filename, ':');
484 if (len > sizeof(protocol) - 1)
485 len = sizeof(protocol) - 1;
486 memcpy(protocol, filename, len);
487 protocol[len] = '\0';
488 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
489 if (drv1->protocol_name &&
490 !strcmp(drv1->protocol_name, protocol)) {
497 static int find_image_format(const char *filename, BlockDriver **pdrv)
499 int ret, score, score_max;
500 BlockDriver *drv1, *drv;
502 BlockDriverState *bs;
504 ret = bdrv_file_open(&bs, filename, 0);
510 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
511 if (bs->sg || !bdrv_is_inserted(bs)) {
513 drv = bdrv_find_format("raw");
521 ret = bdrv_pread(bs, 0, buf, sizeof(buf));
530 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
531 if (drv1->bdrv_probe) {
532 score = drv1->bdrv_probe(buf, ret, filename);
533 if (score > score_max) {
547 * Set the current 'total_sectors' value
549 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
551 BlockDriver *drv = bs->drv;
553 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
557 /* query actual device if possible, otherwise just trust the hint */
558 if (drv->bdrv_getlength) {
559 int64_t length = drv->bdrv_getlength(bs);
563 hint = length >> BDRV_SECTOR_BITS;
566 bs->total_sectors = hint;
571 * Set open flags for a given cache mode
573 * Return 0 on success, -1 if the cache mode was invalid.
575 int bdrv_parse_cache_flags(const char *mode, int *flags)
577 *flags &= ~BDRV_O_CACHE_MASK;
579 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
580 *flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB;
581 } else if (!strcmp(mode, "directsync")) {
582 *flags |= BDRV_O_NOCACHE;
583 } else if (!strcmp(mode, "writeback")) {
584 *flags |= BDRV_O_CACHE_WB;
585 } else if (!strcmp(mode, "unsafe")) {
586 *flags |= BDRV_O_CACHE_WB;
587 *flags |= BDRV_O_NO_FLUSH;
588 } else if (!strcmp(mode, "writethrough")) {
589 /* this is the default */
598 * The copy-on-read flag is actually a reference count so multiple users may
599 * use the feature without worrying about clobbering its previous state.
600 * Copy-on-read stays enabled until all users have called to disable it.
602 void bdrv_enable_copy_on_read(BlockDriverState *bs)
607 void bdrv_disable_copy_on_read(BlockDriverState *bs)
609 assert(bs->copy_on_read > 0);
614 * Common part for opening disk images and files
616 static int bdrv_open_common(BlockDriverState *bs, const char *filename,
617 int flags, BlockDriver *drv)
622 assert(bs->file == NULL);
624 trace_bdrv_open_common(bs, filename, flags, drv->format_name);
626 bs->open_flags = flags;
627 bs->buffer_alignment = 512;
629 assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
630 if ((flags & BDRV_O_RDWR) && (flags & BDRV_O_COPY_ON_READ)) {
631 bdrv_enable_copy_on_read(bs);
634 pstrcpy(bs->filename, sizeof(bs->filename), filename);
636 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv)) {
641 bs->opaque = g_malloc0(drv->instance_size);
643 bs->enable_write_cache = !!(flags & BDRV_O_CACHE_WB);
646 * Clear flags that are internal to the block layer before opening the
649 open_flags = flags & ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
652 * Snapshots should be writable.
654 if (bs->is_temporary) {
655 open_flags |= BDRV_O_RDWR;
658 bs->keep_read_only = bs->read_only = !(open_flags & BDRV_O_RDWR);
660 /* Open the image, either directly or using a protocol */
661 if (drv->bdrv_file_open) {
662 ret = drv->bdrv_file_open(bs, filename, open_flags);
664 ret = bdrv_file_open(&bs->file, filename, open_flags);
666 ret = drv->bdrv_open(bs, open_flags);
674 ret = refresh_total_sectors(bs, bs->total_sectors);
680 if (bs->is_temporary) {
688 bdrv_delete(bs->file);
698 * Opens a file using a protocol (file, host_device, nbd, ...)
700 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
702 BlockDriverState *bs;
706 drv = bdrv_find_protocol(filename);
712 ret = bdrv_open_common(bs, filename, flags, drv);
723 * Opens a disk image (raw, qcow2, vmdk, ...)
725 int bdrv_open(BlockDriverState *bs, const char *filename, int flags,
729 char tmp_filename[PATH_MAX];
731 if (flags & BDRV_O_SNAPSHOT) {
732 BlockDriverState *bs1;
735 BlockDriver *bdrv_qcow2;
736 QEMUOptionParameter *options;
737 char backing_filename[PATH_MAX];
739 /* if snapshot, we create a temporary backing file and open it
740 instead of opening 'filename' directly */
742 /* if there is a backing file, use it */
744 ret = bdrv_open(bs1, filename, 0, drv);
749 total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
751 if (bs1->drv && bs1->drv->protocol_name)
756 get_tmp_filename(tmp_filename, sizeof(tmp_filename));
758 /* Real path is meaningless for protocols */
760 snprintf(backing_filename, sizeof(backing_filename),
762 else if (!realpath(filename, backing_filename))
765 bdrv_qcow2 = bdrv_find_format("qcow2");
766 options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
768 set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size);
769 set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
771 set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
775 ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
776 free_option_parameters(options);
781 filename = tmp_filename;
783 bs->is_temporary = 1;
786 /* Find the right image format driver */
788 ret = find_image_format(filename, &drv);
792 goto unlink_and_fail;
796 ret = bdrv_open_common(bs, filename, flags, drv);
798 goto unlink_and_fail;
801 /* If there is a backing file, use it */
802 if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') {
803 char backing_filename[PATH_MAX];
805 BlockDriver *back_drv = NULL;
807 bs->backing_hd = bdrv_new("");
808 bdrv_get_full_backing_filename(bs, backing_filename,
809 sizeof(backing_filename));
811 if (bs->backing_format[0] != '\0') {
812 back_drv = bdrv_find_format(bs->backing_format);
815 /* backing files always opened read-only */
817 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
819 ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv);
824 if (bs->is_temporary) {
825 bs->backing_hd->keep_read_only = !(flags & BDRV_O_RDWR);
827 /* base image inherits from "parent" */
828 bs->backing_hd->keep_read_only = bs->keep_read_only;
832 if (!bdrv_key_required(bs)) {
833 bdrv_dev_change_media_cb(bs, true);
836 /* throttling disk I/O limits */
837 if (bs->io_limits_enabled) {
838 bdrv_io_limits_enable(bs);
844 if (bs->is_temporary) {
850 void bdrv_close(BlockDriverState *bs)
855 block_job_cancel_sync(bs->job);
859 if (bs == bs_snapshots) {
862 if (bs->backing_hd) {
863 bdrv_delete(bs->backing_hd);
864 bs->backing_hd = NULL;
866 bs->drv->bdrv_close(bs);
869 if (bs->is_temporary) {
870 unlink(bs->filename);
875 bs->copy_on_read = 0;
876 bs->backing_file[0] = '\0';
877 bs->backing_format[0] = '\0';
878 bs->total_sectors = 0;
884 if (bs->file != NULL) {
885 bdrv_delete(bs->file);
889 bdrv_dev_change_media_cb(bs, false);
892 /*throttling disk I/O limits*/
893 if (bs->io_limits_enabled) {
894 bdrv_io_limits_disable(bs);
898 void bdrv_close_all(void)
900 BlockDriverState *bs;
902 QTAILQ_FOREACH(bs, &bdrv_states, list) {
908 * Wait for pending requests to complete across all BlockDriverStates
910 * This function does not flush data to disk, use bdrv_flush_all() for that
911 * after calling this function.
913 * Note that completion of an asynchronous I/O operation can trigger any
914 * number of other I/O operations on other devices---for example a coroutine
915 * can be arbitrarily complex and a constant flow of I/O can come until the
916 * coroutine is complete. Because of this, it is not possible to have a
917 * function to drain a single device's I/O queue.
919 void bdrv_drain_all(void)
921 BlockDriverState *bs;
925 busy = qemu_aio_wait();
927 /* FIXME: We do not have timer support here, so this is effectively
930 QTAILQ_FOREACH(bs, &bdrv_states, list) {
931 if (!qemu_co_queue_empty(&bs->throttled_reqs)) {
932 qemu_co_queue_restart_all(&bs->throttled_reqs);
938 /* If requests are still pending there is a bug somewhere */
939 QTAILQ_FOREACH(bs, &bdrv_states, list) {
940 assert(QLIST_EMPTY(&bs->tracked_requests));
941 assert(qemu_co_queue_empty(&bs->throttled_reqs));
945 /* make a BlockDriverState anonymous by removing from bdrv_state list.
946 Also, NULL terminate the device_name to prevent double remove */
947 void bdrv_make_anon(BlockDriverState *bs)
949 if (bs->device_name[0] != '\0') {
950 QTAILQ_REMOVE(&bdrv_states, bs, list);
952 bs->device_name[0] = '\0';
955 static void bdrv_rebind(BlockDriverState *bs)
957 if (bs->drv && bs->drv->bdrv_rebind) {
958 bs->drv->bdrv_rebind(bs);
963 * Add new bs contents at the top of an image chain while the chain is
964 * live, while keeping required fields on the top layer.
966 * This will modify the BlockDriverState fields, and swap contents
967 * between bs_new and bs_top. Both bs_new and bs_top are modified.
969 * bs_new is required to be anonymous.
971 * This function does not create any image files.
973 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
975 BlockDriverState tmp;
977 /* bs_new must be anonymous */
978 assert(bs_new->device_name[0] == '\0');
982 /* there are some fields that need to stay on the top layer: */
983 tmp.open_flags = bs_top->open_flags;
986 tmp.dev_ops = bs_top->dev_ops;
987 tmp.dev_opaque = bs_top->dev_opaque;
988 tmp.dev = bs_top->dev;
989 tmp.buffer_alignment = bs_top->buffer_alignment;
990 tmp.copy_on_read = bs_top->copy_on_read;
992 /* i/o timing parameters */
993 tmp.slice_time = bs_top->slice_time;
994 tmp.slice_start = bs_top->slice_start;
995 tmp.slice_end = bs_top->slice_end;
996 tmp.io_limits = bs_top->io_limits;
997 tmp.io_base = bs_top->io_base;
998 tmp.throttled_reqs = bs_top->throttled_reqs;
999 tmp.block_timer = bs_top->block_timer;
1000 tmp.io_limits_enabled = bs_top->io_limits_enabled;
1003 tmp.cyls = bs_top->cyls;
1004 tmp.heads = bs_top->heads;
1005 tmp.secs = bs_top->secs;
1006 tmp.translation = bs_top->translation;
1009 tmp.on_read_error = bs_top->on_read_error;
1010 tmp.on_write_error = bs_top->on_write_error;
1013 tmp.iostatus_enabled = bs_top->iostatus_enabled;
1014 tmp.iostatus = bs_top->iostatus;
1016 /* keep the same entry in bdrv_states */
1017 pstrcpy(tmp.device_name, sizeof(tmp.device_name), bs_top->device_name);
1018 tmp.list = bs_top->list;
1020 /* The contents of 'tmp' will become bs_top, as we are
1021 * swapping bs_new and bs_top contents. */
1022 tmp.backing_hd = bs_new;
1023 pstrcpy(tmp.backing_file, sizeof(tmp.backing_file), bs_top->filename);
1024 bdrv_get_format(bs_top, tmp.backing_format, sizeof(tmp.backing_format));
1026 /* swap contents of the fixed new bs and the current top */
1030 /* device_name[] was carried over from the old bs_top. bs_new
1031 * shouldn't be in bdrv_states, so we need to make device_name[]
1032 * reflect the anonymity of bs_new
1034 bs_new->device_name[0] = '\0';
1036 /* clear the copied fields in the new backing file */
1037 bdrv_detach_dev(bs_new, bs_new->dev);
1039 qemu_co_queue_init(&bs_new->throttled_reqs);
1040 memset(&bs_new->io_base, 0, sizeof(bs_new->io_base));
1041 memset(&bs_new->io_limits, 0, sizeof(bs_new->io_limits));
1042 bdrv_iostatus_disable(bs_new);
1044 /* we don't use bdrv_io_limits_disable() for this, because we don't want
1045 * to affect or delete the block_timer, as it has been moved to bs_top */
1046 bs_new->io_limits_enabled = false;
1047 bs_new->block_timer = NULL;
1048 bs_new->slice_time = 0;
1049 bs_new->slice_start = 0;
1050 bs_new->slice_end = 0;
1052 bdrv_rebind(bs_new);
1053 bdrv_rebind(bs_top);
1056 void bdrv_delete(BlockDriverState *bs)
1060 assert(!bs->in_use);
1062 /* remove from list, if necessary */
1067 assert(bs != bs_snapshots);
1071 int bdrv_attach_dev(BlockDriverState *bs, void *dev)
1072 /* TODO change to DeviceState *dev when all users are qdevified */
1078 bdrv_iostatus_reset(bs);
1082 /* TODO qdevified devices don't use this, remove when devices are qdevified */
1083 void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev)
1085 if (bdrv_attach_dev(bs, dev) < 0) {
1090 void bdrv_detach_dev(BlockDriverState *bs, void *dev)
1091 /* TODO change to DeviceState *dev when all users are qdevified */
1093 assert(bs->dev == dev);
1096 bs->dev_opaque = NULL;
1097 bs->buffer_alignment = 512;
1100 /* TODO change to return DeviceState * when all users are qdevified */
1101 void *bdrv_get_attached_dev(BlockDriverState *bs)
1106 void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops,
1110 bs->dev_opaque = opaque;
1111 if (bdrv_dev_has_removable_media(bs) && bs == bs_snapshots) {
1112 bs_snapshots = NULL;
1116 void bdrv_emit_qmp_error_event(const BlockDriverState *bdrv,
1117 BlockQMPEventAction action, int is_read)
1120 const char *action_str;
1123 case BDRV_ACTION_REPORT:
1124 action_str = "report";
1126 case BDRV_ACTION_IGNORE:
1127 action_str = "ignore";
1129 case BDRV_ACTION_STOP:
1130 action_str = "stop";
1136 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1139 is_read ? "read" : "write");
1140 monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
1142 qobject_decref(data);
1145 static void bdrv_emit_qmp_eject_event(BlockDriverState *bs, bool ejected)
1149 data = qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }",
1150 bdrv_get_device_name(bs), ejected);
1151 monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED, data);
1153 qobject_decref(data);
1156 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load)
1158 if (bs->dev_ops && bs->dev_ops->change_media_cb) {
1159 bool tray_was_closed = !bdrv_dev_is_tray_open(bs);
1160 bs->dev_ops->change_media_cb(bs->dev_opaque, load);
1161 if (tray_was_closed) {
1163 bdrv_emit_qmp_eject_event(bs, true);
1167 bdrv_emit_qmp_eject_event(bs, false);
1172 bool bdrv_dev_has_removable_media(BlockDriverState *bs)
1174 return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb);
1177 void bdrv_dev_eject_request(BlockDriverState *bs, bool force)
1179 if (bs->dev_ops && bs->dev_ops->eject_request_cb) {
1180 bs->dev_ops->eject_request_cb(bs->dev_opaque, force);
1184 bool bdrv_dev_is_tray_open(BlockDriverState *bs)
1186 if (bs->dev_ops && bs->dev_ops->is_tray_open) {
1187 return bs->dev_ops->is_tray_open(bs->dev_opaque);
1192 static void bdrv_dev_resize_cb(BlockDriverState *bs)
1194 if (bs->dev_ops && bs->dev_ops->resize_cb) {
1195 bs->dev_ops->resize_cb(bs->dev_opaque);
1199 bool bdrv_dev_is_medium_locked(BlockDriverState *bs)
1201 if (bs->dev_ops && bs->dev_ops->is_medium_locked) {
1202 return bs->dev_ops->is_medium_locked(bs->dev_opaque);
1208 * Run consistency checks on an image
1210 * Returns 0 if the check could be completed (it doesn't mean that the image is
1211 * free of errors) or -errno when an internal error occurred. The results of the
1212 * check are stored in res.
1214 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res)
1216 if (bs->drv->bdrv_check == NULL) {
1220 memset(res, 0, sizeof(*res));
1221 return bs->drv->bdrv_check(bs, res);
1224 #define COMMIT_BUF_SECTORS 2048
1226 /* commit COW file into the raw image */
1227 int bdrv_commit(BlockDriverState *bs)
1229 BlockDriver *drv = bs->drv;
1230 BlockDriver *backing_drv;
1231 int64_t sector, total_sectors;
1232 int n, ro, open_flags;
1233 int ret = 0, rw_ret = 0;
1235 char filename[1024];
1236 BlockDriverState *bs_rw, *bs_ro;
1241 if (!bs->backing_hd) {
1245 if (bs->backing_hd->keep_read_only) {
1249 if (bdrv_in_use(bs) || bdrv_in_use(bs->backing_hd)) {
1253 backing_drv = bs->backing_hd->drv;
1254 ro = bs->backing_hd->read_only;
1255 strncpy(filename, bs->backing_hd->filename, sizeof(filename));
1256 open_flags = bs->backing_hd->open_flags;
1260 bdrv_delete(bs->backing_hd);
1261 bs->backing_hd = NULL;
1262 bs_rw = bdrv_new("");
1263 rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR,
1267 /* try to re-open read-only */
1268 bs_ro = bdrv_new("");
1269 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
1273 /* drive not functional anymore */
1277 bs->backing_hd = bs_ro;
1280 bs->backing_hd = bs_rw;
1283 total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
1284 buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
1286 for (sector = 0; sector < total_sectors; sector += n) {
1287 if (bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n)) {
1289 if (bdrv_read(bs, sector, buf, n) != 0) {
1294 if (bdrv_write(bs->backing_hd, sector, buf, n) != 0) {
1301 if (drv->bdrv_make_empty) {
1302 ret = drv->bdrv_make_empty(bs);
1307 * Make sure all data we wrote to the backing device is actually
1311 bdrv_flush(bs->backing_hd);
1318 bdrv_delete(bs->backing_hd);
1319 bs->backing_hd = NULL;
1320 bs_ro = bdrv_new("");
1321 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
1325 /* drive not functional anymore */
1329 bs->backing_hd = bs_ro;
1330 bs->backing_hd->keep_read_only = 0;
1336 int bdrv_commit_all(void)
1338 BlockDriverState *bs;
1340 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1341 int ret = bdrv_commit(bs);
1349 struct BdrvTrackedRequest {
1350 BlockDriverState *bs;
1354 QLIST_ENTRY(BdrvTrackedRequest) list;
1355 Coroutine *co; /* owner, used for deadlock detection */
1356 CoQueue wait_queue; /* coroutines blocked on this request */
1360 * Remove an active request from the tracked requests list
1362 * This function should be called when a tracked request is completing.
1364 static void tracked_request_end(BdrvTrackedRequest *req)
1366 QLIST_REMOVE(req, list);
1367 qemu_co_queue_restart_all(&req->wait_queue);
1371 * Add an active request to the tracked requests list
1373 static void tracked_request_begin(BdrvTrackedRequest *req,
1374 BlockDriverState *bs,
1376 int nb_sectors, bool is_write)
1378 *req = (BdrvTrackedRequest){
1380 .sector_num = sector_num,
1381 .nb_sectors = nb_sectors,
1382 .is_write = is_write,
1383 .co = qemu_coroutine_self(),
1386 qemu_co_queue_init(&req->wait_queue);
1388 QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
1392 * Round a region to cluster boundaries
1394 static void round_to_clusters(BlockDriverState *bs,
1395 int64_t sector_num, int nb_sectors,
1396 int64_t *cluster_sector_num,
1397 int *cluster_nb_sectors)
1399 BlockDriverInfo bdi;
1401 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
1402 *cluster_sector_num = sector_num;
1403 *cluster_nb_sectors = nb_sectors;
1405 int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE;
1406 *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c);
1407 *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num +
1412 static bool tracked_request_overlaps(BdrvTrackedRequest *req,
1413 int64_t sector_num, int nb_sectors) {
1415 if (sector_num >= req->sector_num + req->nb_sectors) {
1419 if (req->sector_num >= sector_num + nb_sectors) {
1425 static void coroutine_fn wait_for_overlapping_requests(BlockDriverState *bs,
1426 int64_t sector_num, int nb_sectors)
1428 BdrvTrackedRequest *req;
1429 int64_t cluster_sector_num;
1430 int cluster_nb_sectors;
1433 /* If we touch the same cluster it counts as an overlap. This guarantees
1434 * that allocating writes will be serialized and not race with each other
1435 * for the same cluster. For example, in copy-on-read it ensures that the
1436 * CoR read and write operations are atomic and guest writes cannot
1437 * interleave between them.
1439 round_to_clusters(bs, sector_num, nb_sectors,
1440 &cluster_sector_num, &cluster_nb_sectors);
1444 QLIST_FOREACH(req, &bs->tracked_requests, list) {
1445 if (tracked_request_overlaps(req, cluster_sector_num,
1446 cluster_nb_sectors)) {
1447 /* Hitting this means there was a reentrant request, for
1448 * example, a block driver issuing nested requests. This must
1449 * never happen since it means deadlock.
1451 assert(qemu_coroutine_self() != req->co);
1453 qemu_co_queue_wait(&req->wait_queue);
1464 * -EINVAL - backing format specified, but no file
1465 * -ENOSPC - can't update the backing file because no space is left in the
1467 * -ENOTSUP - format driver doesn't support changing the backing file
1469 int bdrv_change_backing_file(BlockDriverState *bs,
1470 const char *backing_file, const char *backing_fmt)
1472 BlockDriver *drv = bs->drv;
1475 /* Backing file format doesn't make sense without a backing file */
1476 if (backing_fmt && !backing_file) {
1480 if (drv->bdrv_change_backing_file != NULL) {
1481 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
1487 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
1488 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
1493 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
1498 if (!bdrv_is_inserted(bs))
1504 len = bdrv_getlength(bs);
1509 if ((offset > len) || (len - offset < size))
1515 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
1518 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
1519 nb_sectors * BDRV_SECTOR_SIZE);
1522 typedef struct RwCo {
1523 BlockDriverState *bs;
1531 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
1533 RwCo *rwco = opaque;
1535 if (!rwco->is_write) {
1536 rwco->ret = bdrv_co_do_readv(rwco->bs, rwco->sector_num,
1537 rwco->nb_sectors, rwco->qiov, 0);
1539 rwco->ret = bdrv_co_do_writev(rwco->bs, rwco->sector_num,
1540 rwco->nb_sectors, rwco->qiov, 0);
1545 * Process a synchronous request using coroutines
1547 static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf,
1548 int nb_sectors, bool is_write)
1551 struct iovec iov = {
1552 .iov_base = (void *)buf,
1553 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
1558 .sector_num = sector_num,
1559 .nb_sectors = nb_sectors,
1561 .is_write = is_write,
1565 qemu_iovec_init_external(&qiov, &iov, 1);
1568 * In sync call context, when the vcpu is blocked, this throttling timer
1569 * will not fire; so the I/O throttling function has to be disabled here
1570 * if it has been enabled.
1572 if (bs->io_limits_enabled) {
1573 fprintf(stderr, "Disabling I/O throttling on '%s' due "
1574 "to synchronous I/O.\n", bdrv_get_device_name(bs));
1575 bdrv_io_limits_disable(bs);
1578 if (qemu_in_coroutine()) {
1579 /* Fast-path if already in coroutine context */
1580 bdrv_rw_co_entry(&rwco);
1582 co = qemu_coroutine_create(bdrv_rw_co_entry);
1583 qemu_coroutine_enter(co, &rwco);
1584 while (rwco.ret == NOT_DONE) {
1591 /* return < 0 if error. See bdrv_write() for the return codes */
1592 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
1593 uint8_t *buf, int nb_sectors)
1595 return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false);
1598 #define BITS_PER_LONG (sizeof(unsigned long) * 8)
1600 static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
1601 int nb_sectors, int dirty)
1604 unsigned long val, idx, bit;
1606 start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
1607 end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
1609 for (; start <= end; start++) {
1610 idx = start / BITS_PER_LONG;
1611 bit = start % BITS_PER_LONG;
1612 val = bs->dirty_bitmap[idx];
1614 if (!(val & (1UL << bit))) {
1619 if (val & (1UL << bit)) {
1621 val &= ~(1UL << bit);
1624 bs->dirty_bitmap[idx] = val;
1628 /* Return < 0 if error. Important errors are:
1629 -EIO generic I/O error (may happen for all errors)
1630 -ENOMEDIUM No media inserted.
1631 -EINVAL Invalid sector number or nb_sectors
1632 -EACCES Trying to write a read-only device
1634 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
1635 const uint8_t *buf, int nb_sectors)
1637 return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true);
1640 int bdrv_pread(BlockDriverState *bs, int64_t offset,
1641 void *buf, int count1)
1643 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1644 int len, nb_sectors, count;
1649 /* first read to align to sector start */
1650 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1653 sector_num = offset >> BDRV_SECTOR_BITS;
1655 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1657 memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
1665 /* read the sectors "in place" */
1666 nb_sectors = count >> BDRV_SECTOR_BITS;
1667 if (nb_sectors > 0) {
1668 if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
1670 sector_num += nb_sectors;
1671 len = nb_sectors << BDRV_SECTOR_BITS;
1676 /* add data from the last sector */
1678 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1680 memcpy(buf, tmp_buf, count);
1685 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
1686 const void *buf, int count1)
1688 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1689 int len, nb_sectors, count;
1694 /* first write to align to sector start */
1695 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1698 sector_num = offset >> BDRV_SECTOR_BITS;
1700 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1702 memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
1703 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1712 /* write the sectors "in place" */
1713 nb_sectors = count >> BDRV_SECTOR_BITS;
1714 if (nb_sectors > 0) {
1715 if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0)
1717 sector_num += nb_sectors;
1718 len = nb_sectors << BDRV_SECTOR_BITS;
1723 /* add data from the last sector */
1725 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1727 memcpy(tmp_buf, buf, count);
1728 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1735 * Writes to the file and ensures that no writes are reordered across this
1736 * request (acts as a barrier)
1738 * Returns 0 on success, -errno in error cases.
1740 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
1741 const void *buf, int count)
1745 ret = bdrv_pwrite(bs, offset, buf, count);
1750 /* No flush needed for cache modes that use O_DSYNC */
1751 if ((bs->open_flags & BDRV_O_CACHE_WB) != 0) {
1758 static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs,
1759 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1761 /* Perform I/O through a temporary buffer so that users who scribble over
1762 * their read buffer while the operation is in progress do not end up
1763 * modifying the image file. This is critical for zero-copy guest I/O
1764 * where anything might happen inside guest memory.
1766 void *bounce_buffer;
1768 BlockDriver *drv = bs->drv;
1770 QEMUIOVector bounce_qiov;
1771 int64_t cluster_sector_num;
1772 int cluster_nb_sectors;
1776 /* Cover entire cluster so no additional backing file I/O is required when
1777 * allocating cluster in the image file.
1779 round_to_clusters(bs, sector_num, nb_sectors,
1780 &cluster_sector_num, &cluster_nb_sectors);
1782 trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors,
1783 cluster_sector_num, cluster_nb_sectors);
1785 iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE;
1786 iov.iov_base = bounce_buffer = qemu_blockalign(bs, iov.iov_len);
1787 qemu_iovec_init_external(&bounce_qiov, &iov, 1);
1789 ret = drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors,
1795 if (drv->bdrv_co_write_zeroes &&
1796 buffer_is_zero(bounce_buffer, iov.iov_len)) {
1797 ret = bdrv_co_do_write_zeroes(bs, cluster_sector_num,
1798 cluster_nb_sectors);
1800 ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors,
1805 /* It might be okay to ignore write errors for guest requests. If this
1806 * is a deliberate copy-on-read then we don't want to ignore the error.
1807 * Simply report it in all cases.
1812 skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE;
1813 qemu_iovec_from_buffer(qiov, bounce_buffer + skip_bytes,
1814 nb_sectors * BDRV_SECTOR_SIZE);
1817 qemu_vfree(bounce_buffer);
1822 * Handle a read request in coroutine context
1824 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
1825 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1826 BdrvRequestFlags flags)
1828 BlockDriver *drv = bs->drv;
1829 BdrvTrackedRequest req;
1835 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
1839 /* throttling disk read I/O */
1840 if (bs->io_limits_enabled) {
1841 bdrv_io_limits_intercept(bs, false, nb_sectors);
1844 if (bs->copy_on_read) {
1845 flags |= BDRV_REQ_COPY_ON_READ;
1847 if (flags & BDRV_REQ_COPY_ON_READ) {
1848 bs->copy_on_read_in_flight++;
1851 if (bs->copy_on_read_in_flight) {
1852 wait_for_overlapping_requests(bs, sector_num, nb_sectors);
1855 tracked_request_begin(&req, bs, sector_num, nb_sectors, false);
1857 if (flags & BDRV_REQ_COPY_ON_READ) {
1860 ret = bdrv_co_is_allocated(bs, sector_num, nb_sectors, &pnum);
1865 if (!ret || pnum != nb_sectors) {
1866 ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov);
1871 ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
1874 tracked_request_end(&req);
1876 if (flags & BDRV_REQ_COPY_ON_READ) {
1877 bs->copy_on_read_in_flight--;
1883 int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
1884 int nb_sectors, QEMUIOVector *qiov)
1886 trace_bdrv_co_readv(bs, sector_num, nb_sectors);
1888 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0);
1891 int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
1892 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1894 trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors);
1896 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
1897 BDRV_REQ_COPY_ON_READ);
1900 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
1901 int64_t sector_num, int nb_sectors)
1903 BlockDriver *drv = bs->drv;
1908 /* TODO Emulate only part of misaligned requests instead of letting block
1909 * drivers return -ENOTSUP and emulate everything */
1911 /* First try the efficient write zeroes operation */
1912 if (drv->bdrv_co_write_zeroes) {
1913 ret = drv->bdrv_co_write_zeroes(bs, sector_num, nb_sectors);
1914 if (ret != -ENOTSUP) {
1919 /* Fall back to bounce buffer if write zeroes is unsupported */
1920 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
1921 iov.iov_base = qemu_blockalign(bs, iov.iov_len);
1922 memset(iov.iov_base, 0, iov.iov_len);
1923 qemu_iovec_init_external(&qiov, &iov, 1);
1925 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, &qiov);
1927 qemu_vfree(iov.iov_base);
1932 * Handle a write request in coroutine context
1934 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
1935 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1936 BdrvRequestFlags flags)
1938 BlockDriver *drv = bs->drv;
1939 BdrvTrackedRequest req;
1945 if (bs->read_only) {
1948 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
1952 /* throttling disk write I/O */
1953 if (bs->io_limits_enabled) {
1954 bdrv_io_limits_intercept(bs, true, nb_sectors);
1957 if (bs->copy_on_read_in_flight) {
1958 wait_for_overlapping_requests(bs, sector_num, nb_sectors);
1961 tracked_request_begin(&req, bs, sector_num, nb_sectors, true);
1963 if (flags & BDRV_REQ_ZERO_WRITE) {
1964 ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors);
1966 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
1969 if (bs->dirty_bitmap) {
1970 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1973 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
1974 bs->wr_highest_sector = sector_num + nb_sectors - 1;
1977 tracked_request_end(&req);
1982 int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
1983 int nb_sectors, QEMUIOVector *qiov)
1985 trace_bdrv_co_writev(bs, sector_num, nb_sectors);
1987 return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0);
1990 int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs,
1991 int64_t sector_num, int nb_sectors)
1993 trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors);
1995 return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL,
1996 BDRV_REQ_ZERO_WRITE);
2000 * Truncate file to 'offset' bytes (needed only for file protocols)
2002 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
2004 BlockDriver *drv = bs->drv;
2008 if (!drv->bdrv_truncate)
2012 if (bdrv_in_use(bs))
2014 ret = drv->bdrv_truncate(bs, offset);
2016 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
2017 bdrv_dev_resize_cb(bs);
2023 * Length of a allocated file in bytes. Sparse files are counted by actual
2024 * allocated space. Return < 0 if error or unknown.
2026 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
2028 BlockDriver *drv = bs->drv;
2032 if (drv->bdrv_get_allocated_file_size) {
2033 return drv->bdrv_get_allocated_file_size(bs);
2036 return bdrv_get_allocated_file_size(bs->file);
2042 * Length of a file in bytes. Return < 0 if error or unknown.
2044 int64_t bdrv_getlength(BlockDriverState *bs)
2046 BlockDriver *drv = bs->drv;
2050 if (bs->growable || bdrv_dev_has_removable_media(bs)) {
2051 if (drv->bdrv_getlength) {
2052 return drv->bdrv_getlength(bs);
2055 return bs->total_sectors * BDRV_SECTOR_SIZE;
2058 /* return 0 as number of sectors if no device present or error */
2059 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
2062 length = bdrv_getlength(bs);
2066 length = length >> BDRV_SECTOR_BITS;
2067 *nb_sectors_ptr = length;
2071 uint8_t boot_ind; /* 0x80 - active */
2072 uint8_t head; /* starting head */
2073 uint8_t sector; /* starting sector */
2074 uint8_t cyl; /* starting cylinder */
2075 uint8_t sys_ind; /* What partition type */
2076 uint8_t end_head; /* end head */
2077 uint8_t end_sector; /* end sector */
2078 uint8_t end_cyl; /* end cylinder */
2079 uint32_t start_sect; /* starting sector counting from 0 */
2080 uint32_t nr_sects; /* nr of sectors in partition */
2083 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
2084 static int guess_disk_lchs(BlockDriverState *bs,
2085 int *pcylinders, int *pheads, int *psectors)
2087 uint8_t buf[BDRV_SECTOR_SIZE];
2088 int ret, i, heads, sectors, cylinders;
2089 struct partition *p;
2091 uint64_t nb_sectors;
2094 bdrv_get_geometry(bs, &nb_sectors);
2097 * The function will be invoked during startup not only in sync I/O mode,
2098 * but also in async I/O mode. So the I/O throttling function has to
2099 * be disabled temporarily here, not permanently.
2101 enabled = bs->io_limits_enabled;
2102 bs->io_limits_enabled = false;
2103 ret = bdrv_read(bs, 0, buf, 1);
2104 bs->io_limits_enabled = enabled;
2107 /* test msdos magic */
2108 if (buf[510] != 0x55 || buf[511] != 0xaa)
2110 for(i = 0; i < 4; i++) {
2111 p = ((struct partition *)(buf + 0x1be)) + i;
2112 nr_sects = le32_to_cpu(p->nr_sects);
2113 if (nr_sects && p->end_head) {
2114 /* We make the assumption that the partition terminates on
2115 a cylinder boundary */
2116 heads = p->end_head + 1;
2117 sectors = p->end_sector & 63;
2120 cylinders = nb_sectors / (heads * sectors);
2121 if (cylinders < 1 || cylinders > 16383)
2124 *psectors = sectors;
2125 *pcylinders = cylinders;
2127 printf("guessed geometry: LCHS=%d %d %d\n",
2128 cylinders, heads, sectors);
2136 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
2138 int translation, lba_detected = 0;
2139 int cylinders, heads, secs;
2140 uint64_t nb_sectors;
2142 /* if a geometry hint is available, use it */
2143 bdrv_get_geometry(bs, &nb_sectors);
2144 bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
2145 translation = bdrv_get_translation_hint(bs);
2146 if (cylinders != 0) {
2151 if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
2153 /* if heads > 16, it means that a BIOS LBA
2154 translation was active, so the default
2155 hardware geometry is OK */
2157 goto default_geometry;
2162 /* disable any translation to be in sync with
2163 the logical geometry */
2164 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
2165 bdrv_set_translation_hint(bs,
2166 BIOS_ATA_TRANSLATION_NONE);
2171 /* if no geometry, use a standard physical disk geometry */
2172 cylinders = nb_sectors / (16 * 63);
2174 if (cylinders > 16383)
2176 else if (cylinders < 2)
2181 if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
2182 if ((*pcyls * *pheads) <= 131072) {
2183 bdrv_set_translation_hint(bs,
2184 BIOS_ATA_TRANSLATION_LARGE);
2186 bdrv_set_translation_hint(bs,
2187 BIOS_ATA_TRANSLATION_LBA);
2191 bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
2195 void bdrv_set_geometry_hint(BlockDriverState *bs,
2196 int cyls, int heads, int secs)
2203 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
2205 bs->translation = translation;
2208 void bdrv_get_geometry_hint(BlockDriverState *bs,
2209 int *pcyls, int *pheads, int *psecs)
2212 *pheads = bs->heads;
2216 /* throttling disk io limits */
2217 void bdrv_set_io_limits(BlockDriverState *bs,
2218 BlockIOLimit *io_limits)
2220 bs->io_limits = *io_limits;
2221 bs->io_limits_enabled = bdrv_io_limits_enabled(bs);
2224 /* Recognize floppy formats */
2225 typedef struct FDFormat {
2233 static const FDFormat fd_formats[] = {
2234 /* First entry is default format */
2235 /* 1.44 MB 3"1/2 floppy disks */
2236 { FDRIVE_DRV_144, 18, 80, 1, FDRIVE_RATE_500K, },
2237 { FDRIVE_DRV_144, 20, 80, 1, FDRIVE_RATE_500K, },
2238 { FDRIVE_DRV_144, 21, 80, 1, FDRIVE_RATE_500K, },
2239 { FDRIVE_DRV_144, 21, 82, 1, FDRIVE_RATE_500K, },
2240 { FDRIVE_DRV_144, 21, 83, 1, FDRIVE_RATE_500K, },
2241 { FDRIVE_DRV_144, 22, 80, 1, FDRIVE_RATE_500K, },
2242 { FDRIVE_DRV_144, 23, 80, 1, FDRIVE_RATE_500K, },
2243 { FDRIVE_DRV_144, 24, 80, 1, FDRIVE_RATE_500K, },
2244 /* 2.88 MB 3"1/2 floppy disks */
2245 { FDRIVE_DRV_288, 36, 80, 1, FDRIVE_RATE_1M, },
2246 { FDRIVE_DRV_288, 39, 80, 1, FDRIVE_RATE_1M, },
2247 { FDRIVE_DRV_288, 40, 80, 1, FDRIVE_RATE_1M, },
2248 { FDRIVE_DRV_288, 44, 80, 1, FDRIVE_RATE_1M, },
2249 { FDRIVE_DRV_288, 48, 80, 1, FDRIVE_RATE_1M, },
2250 /* 720 kB 3"1/2 floppy disks */
2251 { FDRIVE_DRV_144, 9, 80, 1, FDRIVE_RATE_250K, },
2252 { FDRIVE_DRV_144, 10, 80, 1, FDRIVE_RATE_250K, },
2253 { FDRIVE_DRV_144, 10, 82, 1, FDRIVE_RATE_250K, },
2254 { FDRIVE_DRV_144, 10, 83, 1, FDRIVE_RATE_250K, },
2255 { FDRIVE_DRV_144, 13, 80, 1, FDRIVE_RATE_250K, },
2256 { FDRIVE_DRV_144, 14, 80, 1, FDRIVE_RATE_250K, },
2257 /* 1.2 MB 5"1/4 floppy disks */
2258 { FDRIVE_DRV_120, 15, 80, 1, FDRIVE_RATE_500K, },
2259 { FDRIVE_DRV_120, 18, 80, 1, FDRIVE_RATE_500K, },
2260 { FDRIVE_DRV_120, 18, 82, 1, FDRIVE_RATE_500K, },
2261 { FDRIVE_DRV_120, 18, 83, 1, FDRIVE_RATE_500K, },
2262 { FDRIVE_DRV_120, 20, 80, 1, FDRIVE_RATE_500K, },
2263 /* 720 kB 5"1/4 floppy disks */
2264 { FDRIVE_DRV_120, 9, 80, 1, FDRIVE_RATE_250K, },
2265 { FDRIVE_DRV_120, 11, 80, 1, FDRIVE_RATE_250K, },
2266 /* 360 kB 5"1/4 floppy disks */
2267 { FDRIVE_DRV_120, 9, 40, 1, FDRIVE_RATE_300K, },
2268 { FDRIVE_DRV_120, 9, 40, 0, FDRIVE_RATE_300K, },
2269 { FDRIVE_DRV_120, 10, 41, 1, FDRIVE_RATE_300K, },
2270 { FDRIVE_DRV_120, 10, 42, 1, FDRIVE_RATE_300K, },
2271 /* 320 kB 5"1/4 floppy disks */
2272 { FDRIVE_DRV_120, 8, 40, 1, FDRIVE_RATE_250K, },
2273 { FDRIVE_DRV_120, 8, 40, 0, FDRIVE_RATE_250K, },
2274 /* 360 kB must match 5"1/4 better than 3"1/2... */
2275 { FDRIVE_DRV_144, 9, 80, 0, FDRIVE_RATE_250K, },
2277 { FDRIVE_DRV_NONE, -1, -1, 0, 0, },
2280 void bdrv_get_floppy_geometry_hint(BlockDriverState *bs, int *nb_heads,
2281 int *max_track, int *last_sect,
2282 FDriveType drive_in, FDriveType *drive,
2285 const FDFormat *parse;
2286 uint64_t nb_sectors, size;
2287 int i, first_match, match;
2289 bdrv_get_geometry_hint(bs, nb_heads, max_track, last_sect);
2290 if (*nb_heads != 0 && *max_track != 0 && *last_sect != 0) {
2291 /* User defined disk */
2292 *rate = FDRIVE_RATE_500K;
2294 bdrv_get_geometry(bs, &nb_sectors);
2297 for (i = 0; ; i++) {
2298 parse = &fd_formats[i];
2299 if (parse->drive == FDRIVE_DRV_NONE) {
2302 if (drive_in == parse->drive ||
2303 drive_in == FDRIVE_DRV_NONE) {
2304 size = (parse->max_head + 1) * parse->max_track *
2306 if (nb_sectors == size) {
2310 if (first_match == -1) {
2316 if (first_match == -1) {
2319 match = first_match;
2321 parse = &fd_formats[match];
2323 *nb_heads = parse->max_head + 1;
2324 *max_track = parse->max_track;
2325 *last_sect = parse->last_sect;
2326 *drive = parse->drive;
2327 *rate = parse->rate;
2331 int bdrv_get_translation_hint(BlockDriverState *bs)
2333 return bs->translation;
2336 void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
2337 BlockErrorAction on_write_error)
2339 bs->on_read_error = on_read_error;
2340 bs->on_write_error = on_write_error;
2343 BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
2345 return is_read ? bs->on_read_error : bs->on_write_error;
2348 int bdrv_is_read_only(BlockDriverState *bs)
2350 return bs->read_only;
2353 int bdrv_is_sg(BlockDriverState *bs)
2358 int bdrv_enable_write_cache(BlockDriverState *bs)
2360 return bs->enable_write_cache;
2363 int bdrv_is_encrypted(BlockDriverState *bs)
2365 if (bs->backing_hd && bs->backing_hd->encrypted)
2367 return bs->encrypted;
2370 int bdrv_key_required(BlockDriverState *bs)
2372 BlockDriverState *backing_hd = bs->backing_hd;
2374 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
2376 return (bs->encrypted && !bs->valid_key);
2379 int bdrv_set_key(BlockDriverState *bs, const char *key)
2382 if (bs->backing_hd && bs->backing_hd->encrypted) {
2383 ret = bdrv_set_key(bs->backing_hd, key);
2389 if (!bs->encrypted) {
2391 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
2394 ret = bs->drv->bdrv_set_key(bs, key);
2397 } else if (!bs->valid_key) {
2399 /* call the change callback now, we skipped it on open */
2400 bdrv_dev_change_media_cb(bs, true);
2405 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
2410 pstrcpy(buf, buf_size, bs->drv->format_name);
2414 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
2419 QLIST_FOREACH(drv, &bdrv_drivers, list) {
2420 it(opaque, drv->format_name);
2424 BlockDriverState *bdrv_find(const char *name)
2426 BlockDriverState *bs;
2428 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2429 if (!strcmp(name, bs->device_name)) {
2436 BlockDriverState *bdrv_next(BlockDriverState *bs)
2439 return QTAILQ_FIRST(&bdrv_states);
2441 return QTAILQ_NEXT(bs, list);
2444 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
2446 BlockDriverState *bs;
2448 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2453 const char *bdrv_get_device_name(BlockDriverState *bs)
2455 return bs->device_name;
2458 void bdrv_flush_all(void)
2460 BlockDriverState *bs;
2462 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2467 int bdrv_has_zero_init(BlockDriverState *bs)
2471 if (bs->drv->bdrv_has_zero_init) {
2472 return bs->drv->bdrv_has_zero_init(bs);
2478 typedef struct BdrvCoIsAllocatedData {
2479 BlockDriverState *bs;
2485 } BdrvCoIsAllocatedData;
2488 * Returns true iff the specified sector is present in the disk image. Drivers
2489 * not implementing the functionality are assumed to not support backing files,
2490 * hence all their sectors are reported as allocated.
2492 * If 'sector_num' is beyond the end of the disk image the return value is 0
2493 * and 'pnum' is set to 0.
2495 * 'pnum' is set to the number of sectors (including and immediately following
2496 * the specified sector) that are known to be in the same
2497 * allocated/unallocated state.
2499 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
2500 * beyond the end of the disk image it will be clamped.
2502 int coroutine_fn bdrv_co_is_allocated(BlockDriverState *bs, int64_t sector_num,
2503 int nb_sectors, int *pnum)
2507 if (sector_num >= bs->total_sectors) {
2512 n = bs->total_sectors - sector_num;
2513 if (n < nb_sectors) {
2517 if (!bs->drv->bdrv_co_is_allocated) {
2522 return bs->drv->bdrv_co_is_allocated(bs, sector_num, nb_sectors, pnum);
2525 /* Coroutine wrapper for bdrv_is_allocated() */
2526 static void coroutine_fn bdrv_is_allocated_co_entry(void *opaque)
2528 BdrvCoIsAllocatedData *data = opaque;
2529 BlockDriverState *bs = data->bs;
2531 data->ret = bdrv_co_is_allocated(bs, data->sector_num, data->nb_sectors,
2537 * Synchronous wrapper around bdrv_co_is_allocated().
2539 * See bdrv_co_is_allocated() for details.
2541 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
2545 BdrvCoIsAllocatedData data = {
2547 .sector_num = sector_num,
2548 .nb_sectors = nb_sectors,
2553 co = qemu_coroutine_create(bdrv_is_allocated_co_entry);
2554 qemu_coroutine_enter(co, &data);
2555 while (!data.done) {
2561 BlockInfoList *qmp_query_block(Error **errp)
2563 BlockInfoList *head = NULL, *cur_item = NULL;
2564 BlockDriverState *bs;
2566 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2567 BlockInfoList *info = g_malloc0(sizeof(*info));
2569 info->value = g_malloc0(sizeof(*info->value));
2570 info->value->device = g_strdup(bs->device_name);
2571 info->value->type = g_strdup("unknown");
2572 info->value->locked = bdrv_dev_is_medium_locked(bs);
2573 info->value->removable = bdrv_dev_has_removable_media(bs);
2575 if (bdrv_dev_has_removable_media(bs)) {
2576 info->value->has_tray_open = true;
2577 info->value->tray_open = bdrv_dev_is_tray_open(bs);
2580 if (bdrv_iostatus_is_enabled(bs)) {
2581 info->value->has_io_status = true;
2582 info->value->io_status = bs->iostatus;
2586 info->value->has_inserted = true;
2587 info->value->inserted = g_malloc0(sizeof(*info->value->inserted));
2588 info->value->inserted->file = g_strdup(bs->filename);
2589 info->value->inserted->ro = bs->read_only;
2590 info->value->inserted->drv = g_strdup(bs->drv->format_name);
2591 info->value->inserted->encrypted = bs->encrypted;
2592 if (bs->backing_file[0]) {
2593 info->value->inserted->has_backing_file = true;
2594 info->value->inserted->backing_file = g_strdup(bs->backing_file);
2597 if (bs->io_limits_enabled) {
2598 info->value->inserted->bps =
2599 bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL];
2600 info->value->inserted->bps_rd =
2601 bs->io_limits.bps[BLOCK_IO_LIMIT_READ];
2602 info->value->inserted->bps_wr =
2603 bs->io_limits.bps[BLOCK_IO_LIMIT_WRITE];
2604 info->value->inserted->iops =
2605 bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL];
2606 info->value->inserted->iops_rd =
2607 bs->io_limits.iops[BLOCK_IO_LIMIT_READ];
2608 info->value->inserted->iops_wr =
2609 bs->io_limits.iops[BLOCK_IO_LIMIT_WRITE];
2613 /* XXX: waiting for the qapi to support GSList */
2615 head = cur_item = info;
2617 cur_item->next = info;
2625 /* Consider exposing this as a full fledged QMP command */
2626 static BlockStats *qmp_query_blockstat(const BlockDriverState *bs, Error **errp)
2630 s = g_malloc0(sizeof(*s));
2632 if (bs->device_name[0]) {
2633 s->has_device = true;
2634 s->device = g_strdup(bs->device_name);
2637 s->stats = g_malloc0(sizeof(*s->stats));
2638 s->stats->rd_bytes = bs->nr_bytes[BDRV_ACCT_READ];
2639 s->stats->wr_bytes = bs->nr_bytes[BDRV_ACCT_WRITE];
2640 s->stats->rd_operations = bs->nr_ops[BDRV_ACCT_READ];
2641 s->stats->wr_operations = bs->nr_ops[BDRV_ACCT_WRITE];
2642 s->stats->wr_highest_offset = bs->wr_highest_sector * BDRV_SECTOR_SIZE;
2643 s->stats->flush_operations = bs->nr_ops[BDRV_ACCT_FLUSH];
2644 s->stats->wr_total_time_ns = bs->total_time_ns[BDRV_ACCT_WRITE];
2645 s->stats->rd_total_time_ns = bs->total_time_ns[BDRV_ACCT_READ];
2646 s->stats->flush_total_time_ns = bs->total_time_ns[BDRV_ACCT_FLUSH];
2649 s->has_parent = true;
2650 s->parent = qmp_query_blockstat(bs->file, NULL);
2656 BlockStatsList *qmp_query_blockstats(Error **errp)
2658 BlockStatsList *head = NULL, *cur_item = NULL;
2659 BlockDriverState *bs;
2661 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2662 BlockStatsList *info = g_malloc0(sizeof(*info));
2663 info->value = qmp_query_blockstat(bs, NULL);
2665 /* XXX: waiting for the qapi to support GSList */
2667 head = cur_item = info;
2669 cur_item->next = info;
2677 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
2679 if (bs->backing_hd && bs->backing_hd->encrypted)
2680 return bs->backing_file;
2681 else if (bs->encrypted)
2682 return bs->filename;
2687 void bdrv_get_backing_filename(BlockDriverState *bs,
2688 char *filename, int filename_size)
2690 pstrcpy(filename, filename_size, bs->backing_file);
2693 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
2694 const uint8_t *buf, int nb_sectors)
2696 BlockDriver *drv = bs->drv;
2699 if (!drv->bdrv_write_compressed)
2701 if (bdrv_check_request(bs, sector_num, nb_sectors))
2704 if (bs->dirty_bitmap) {
2705 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
2708 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
2711 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2713 BlockDriver *drv = bs->drv;
2716 if (!drv->bdrv_get_info)
2718 memset(bdi, 0, sizeof(*bdi));
2719 return drv->bdrv_get_info(bs, bdi);
2722 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
2723 int64_t pos, int size)
2725 BlockDriver *drv = bs->drv;
2728 if (drv->bdrv_save_vmstate)
2729 return drv->bdrv_save_vmstate(bs, buf, pos, size);
2731 return bdrv_save_vmstate(bs->file, buf, pos, size);
2735 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
2736 int64_t pos, int size)
2738 BlockDriver *drv = bs->drv;
2741 if (drv->bdrv_load_vmstate)
2742 return drv->bdrv_load_vmstate(bs, buf, pos, size);
2744 return bdrv_load_vmstate(bs->file, buf, pos, size);
2748 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
2750 BlockDriver *drv = bs->drv;
2752 if (!drv || !drv->bdrv_debug_event) {
2756 return drv->bdrv_debug_event(bs, event);
2760 /**************************************************************/
2761 /* handling of snapshots */
2763 int bdrv_can_snapshot(BlockDriverState *bs)
2765 BlockDriver *drv = bs->drv;
2766 if (!drv || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
2770 if (!drv->bdrv_snapshot_create) {
2771 if (bs->file != NULL) {
2772 return bdrv_can_snapshot(bs->file);
2780 int bdrv_is_snapshot(BlockDriverState *bs)
2782 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
2785 BlockDriverState *bdrv_snapshots(void)
2787 BlockDriverState *bs;
2790 return bs_snapshots;
2794 while ((bs = bdrv_next(bs))) {
2795 if (bdrv_can_snapshot(bs)) {
2803 int bdrv_snapshot_create(BlockDriverState *bs,
2804 QEMUSnapshotInfo *sn_info)
2806 BlockDriver *drv = bs->drv;
2809 if (drv->bdrv_snapshot_create)
2810 return drv->bdrv_snapshot_create(bs, sn_info);
2812 return bdrv_snapshot_create(bs->file, sn_info);
2816 int bdrv_snapshot_goto(BlockDriverState *bs,
2817 const char *snapshot_id)
2819 BlockDriver *drv = bs->drv;
2824 if (drv->bdrv_snapshot_goto)
2825 return drv->bdrv_snapshot_goto(bs, snapshot_id);
2828 drv->bdrv_close(bs);
2829 ret = bdrv_snapshot_goto(bs->file, snapshot_id);
2830 open_ret = drv->bdrv_open(bs, bs->open_flags);
2832 bdrv_delete(bs->file);
2842 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2844 BlockDriver *drv = bs->drv;
2847 if (drv->bdrv_snapshot_delete)
2848 return drv->bdrv_snapshot_delete(bs, snapshot_id);
2850 return bdrv_snapshot_delete(bs->file, snapshot_id);
2854 int bdrv_snapshot_list(BlockDriverState *bs,
2855 QEMUSnapshotInfo **psn_info)
2857 BlockDriver *drv = bs->drv;
2860 if (drv->bdrv_snapshot_list)
2861 return drv->bdrv_snapshot_list(bs, psn_info);
2863 return bdrv_snapshot_list(bs->file, psn_info);
2867 int bdrv_snapshot_load_tmp(BlockDriverState *bs,
2868 const char *snapshot_name)
2870 BlockDriver *drv = bs->drv;
2874 if (!bs->read_only) {
2877 if (drv->bdrv_snapshot_load_tmp) {
2878 return drv->bdrv_snapshot_load_tmp(bs, snapshot_name);
2883 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
2884 const char *backing_file)
2890 if (bs->backing_hd) {
2891 if (strcmp(bs->backing_file, backing_file) == 0) {
2892 return bs->backing_hd;
2894 return bdrv_find_backing_image(bs->backing_hd, backing_file);
2901 #define NB_SUFFIXES 4
2903 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
2905 static const char suffixes[NB_SUFFIXES] = "KMGT";
2910 snprintf(buf, buf_size, "%" PRId64, size);
2913 for(i = 0; i < NB_SUFFIXES; i++) {
2914 if (size < (10 * base)) {
2915 snprintf(buf, buf_size, "%0.1f%c",
2916 (double)size / base,
2919 } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
2920 snprintf(buf, buf_size, "%" PRId64 "%c",
2921 ((size + (base >> 1)) / base),
2931 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
2933 char buf1[128], date_buf[128], clock_buf[128];
2943 snprintf(buf, buf_size,
2944 "%-10s%-20s%7s%20s%15s",
2945 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
2949 ptm = localtime(&ti);
2950 strftime(date_buf, sizeof(date_buf),
2951 "%Y-%m-%d %H:%M:%S", ptm);
2953 localtime_r(&ti, &tm);
2954 strftime(date_buf, sizeof(date_buf),
2955 "%Y-%m-%d %H:%M:%S", &tm);
2957 secs = sn->vm_clock_nsec / 1000000000;
2958 snprintf(clock_buf, sizeof(clock_buf),
2959 "%02d:%02d:%02d.%03d",
2961 (int)((secs / 60) % 60),
2963 (int)((sn->vm_clock_nsec / 1000000) % 1000));
2964 snprintf(buf, buf_size,
2965 "%-10s%-20s%7s%20s%15s",
2966 sn->id_str, sn->name,
2967 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
2974 /**************************************************************/
2977 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
2978 QEMUIOVector *qiov, int nb_sectors,
2979 BlockDriverCompletionFunc *cb, void *opaque)
2981 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
2983 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
2987 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
2988 QEMUIOVector *qiov, int nb_sectors,
2989 BlockDriverCompletionFunc *cb, void *opaque)
2991 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
2993 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
2998 typedef struct MultiwriteCB {
3003 BlockDriverCompletionFunc *cb;
3005 QEMUIOVector *free_qiov;
3009 static void multiwrite_user_cb(MultiwriteCB *mcb)
3013 for (i = 0; i < mcb->num_callbacks; i++) {
3014 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
3015 if (mcb->callbacks[i].free_qiov) {
3016 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
3018 g_free(mcb->callbacks[i].free_qiov);
3022 static void multiwrite_cb(void *opaque, int ret)
3024 MultiwriteCB *mcb = opaque;
3026 trace_multiwrite_cb(mcb, ret);
3028 if (ret < 0 && !mcb->error) {
3032 mcb->num_requests--;
3033 if (mcb->num_requests == 0) {
3034 multiwrite_user_cb(mcb);
3039 static int multiwrite_req_compare(const void *a, const void *b)
3041 const BlockRequest *req1 = a, *req2 = b;
3044 * Note that we can't simply subtract req2->sector from req1->sector
3045 * here as that could overflow the return value.
3047 if (req1->sector > req2->sector) {
3049 } else if (req1->sector < req2->sector) {
3057 * Takes a bunch of requests and tries to merge them. Returns the number of
3058 * requests that remain after merging.
3060 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
3061 int num_reqs, MultiwriteCB *mcb)
3065 // Sort requests by start sector
3066 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
3068 // Check if adjacent requests touch the same clusters. If so, combine them,
3069 // filling up gaps with zero sectors.
3071 for (i = 1; i < num_reqs; i++) {
3073 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
3075 // Handle exactly sequential writes and overlapping writes.
3076 if (reqs[i].sector <= oldreq_last) {
3080 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
3086 QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
3087 qemu_iovec_init(qiov,
3088 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
3090 // Add the first request to the merged one. If the requests are
3091 // overlapping, drop the last sectors of the first request.
3092 size = (reqs[i].sector - reqs[outidx].sector) << 9;
3093 qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
3095 // We should need to add any zeros between the two requests
3096 assert (reqs[i].sector <= oldreq_last);
3098 // Add the second request
3099 qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
3101 reqs[outidx].nb_sectors = qiov->size >> 9;
3102 reqs[outidx].qiov = qiov;
3104 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
3107 reqs[outidx].sector = reqs[i].sector;
3108 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
3109 reqs[outidx].qiov = reqs[i].qiov;
3117 * Submit multiple AIO write requests at once.
3119 * On success, the function returns 0 and all requests in the reqs array have
3120 * been submitted. In error case this function returns -1, and any of the
3121 * requests may or may not be submitted yet. In particular, this means that the
3122 * callback will be called for some of the requests, for others it won't. The
3123 * caller must check the error field of the BlockRequest to wait for the right
3124 * callbacks (if error != 0, no callback will be called).
3126 * The implementation may modify the contents of the reqs array, e.g. to merge
3127 * requests. However, the fields opaque and error are left unmodified as they
3128 * are used to signal failure for a single request to the caller.
3130 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
3135 /* don't submit writes if we don't have a medium */
3136 if (bs->drv == NULL) {
3137 for (i = 0; i < num_reqs; i++) {
3138 reqs[i].error = -ENOMEDIUM;
3143 if (num_reqs == 0) {
3147 // Create MultiwriteCB structure
3148 mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
3149 mcb->num_requests = 0;
3150 mcb->num_callbacks = num_reqs;
3152 for (i = 0; i < num_reqs; i++) {
3153 mcb->callbacks[i].cb = reqs[i].cb;
3154 mcb->callbacks[i].opaque = reqs[i].opaque;
3157 // Check for mergable requests
3158 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
3160 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
3162 /* Run the aio requests. */
3163 mcb->num_requests = num_reqs;
3164 for (i = 0; i < num_reqs; i++) {
3165 bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
3166 reqs[i].nb_sectors, multiwrite_cb, mcb);
3172 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
3174 acb->pool->cancel(acb);
3177 /* block I/O throttling */
3178 static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors,
3179 bool is_write, double elapsed_time, uint64_t *wait)
3181 uint64_t bps_limit = 0;
3182 double bytes_limit, bytes_base, bytes_res;
3183 double slice_time, wait_time;
3185 if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
3186 bps_limit = bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL];
3187 } else if (bs->io_limits.bps[is_write]) {
3188 bps_limit = bs->io_limits.bps[is_write];
3197 slice_time = bs->slice_end - bs->slice_start;
3198 slice_time /= (NANOSECONDS_PER_SECOND);
3199 bytes_limit = bps_limit * slice_time;
3200 bytes_base = bs->nr_bytes[is_write] - bs->io_base.bytes[is_write];
3201 if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
3202 bytes_base += bs->nr_bytes[!is_write] - bs->io_base.bytes[!is_write];
3205 /* bytes_base: the bytes of data which have been read/written; and
3206 * it is obtained from the history statistic info.
3207 * bytes_res: the remaining bytes of data which need to be read/written.
3208 * (bytes_base + bytes_res) / bps_limit: used to calcuate
3209 * the total time for completing reading/writting all data.
3211 bytes_res = (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
3213 if (bytes_base + bytes_res <= bytes_limit) {
3221 /* Calc approx time to dispatch */
3222 wait_time = (bytes_base + bytes_res) / bps_limit - elapsed_time;
3224 /* When the I/O rate at runtime exceeds the limits,
3225 * bs->slice_end need to be extended in order that the current statistic
3226 * info can be kept until the timer fire, so it is increased and tuned
3227 * based on the result of experiment.
3229 bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10;
3230 bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME;
3232 *wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
3238 static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write,
3239 double elapsed_time, uint64_t *wait)
3241 uint64_t iops_limit = 0;
3242 double ios_limit, ios_base;
3243 double slice_time, wait_time;
3245 if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
3246 iops_limit = bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL];
3247 } else if (bs->io_limits.iops[is_write]) {
3248 iops_limit = bs->io_limits.iops[is_write];
3257 slice_time = bs->slice_end - bs->slice_start;
3258 slice_time /= (NANOSECONDS_PER_SECOND);
3259 ios_limit = iops_limit * slice_time;
3260 ios_base = bs->nr_ops[is_write] - bs->io_base.ios[is_write];
3261 if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
3262 ios_base += bs->nr_ops[!is_write] - bs->io_base.ios[!is_write];
3265 if (ios_base + 1 <= ios_limit) {
3273 /* Calc approx time to dispatch */
3274 wait_time = (ios_base + 1) / iops_limit;
3275 if (wait_time > elapsed_time) {
3276 wait_time = wait_time - elapsed_time;
3281 bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10;
3282 bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME;
3284 *wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
3290 static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors,
3291 bool is_write, int64_t *wait)
3293 int64_t now, max_wait;
3294 uint64_t bps_wait = 0, iops_wait = 0;
3295 double elapsed_time;
3296 int bps_ret, iops_ret;
3298 now = qemu_get_clock_ns(vm_clock);
3299 if ((bs->slice_start < now)
3300 && (bs->slice_end > now)) {
3301 bs->slice_end = now + bs->slice_time;
3303 bs->slice_time = 5 * BLOCK_IO_SLICE_TIME;
3304 bs->slice_start = now;
3305 bs->slice_end = now + bs->slice_time;
3307 bs->io_base.bytes[is_write] = bs->nr_bytes[is_write];
3308 bs->io_base.bytes[!is_write] = bs->nr_bytes[!is_write];
3310 bs->io_base.ios[is_write] = bs->nr_ops[is_write];
3311 bs->io_base.ios[!is_write] = bs->nr_ops[!is_write];
3314 elapsed_time = now - bs->slice_start;
3315 elapsed_time /= (NANOSECONDS_PER_SECOND);
3317 bps_ret = bdrv_exceed_bps_limits(bs, nb_sectors,
3318 is_write, elapsed_time, &bps_wait);
3319 iops_ret = bdrv_exceed_iops_limits(bs, is_write,
3320 elapsed_time, &iops_wait);
3321 if (bps_ret || iops_ret) {
3322 max_wait = bps_wait > iops_wait ? bps_wait : iops_wait;
3327 now = qemu_get_clock_ns(vm_clock);
3328 if (bs->slice_end < now + max_wait) {
3329 bs->slice_end = now + max_wait;
3342 /**************************************************************/
3343 /* async block device emulation */
3345 typedef struct BlockDriverAIOCBSync {
3346 BlockDriverAIOCB common;
3349 /* vector translation state */
3353 } BlockDriverAIOCBSync;
3355 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
3357 BlockDriverAIOCBSync *acb =
3358 container_of(blockacb, BlockDriverAIOCBSync, common);
3359 qemu_bh_delete(acb->bh);
3361 qemu_aio_release(acb);
3364 static AIOPool bdrv_em_aio_pool = {
3365 .aiocb_size = sizeof(BlockDriverAIOCBSync),
3366 .cancel = bdrv_aio_cancel_em,
3369 static void bdrv_aio_bh_cb(void *opaque)
3371 BlockDriverAIOCBSync *acb = opaque;
3374 qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
3375 qemu_vfree(acb->bounce);
3376 acb->common.cb(acb->common.opaque, acb->ret);
3377 qemu_bh_delete(acb->bh);
3379 qemu_aio_release(acb);
3382 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
3386 BlockDriverCompletionFunc *cb,
3391 BlockDriverAIOCBSync *acb;
3393 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
3394 acb->is_write = is_write;
3396 acb->bounce = qemu_blockalign(bs, qiov->size);
3397 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
3400 qemu_iovec_to_buffer(acb->qiov, acb->bounce);
3401 acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
3403 acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
3406 qemu_bh_schedule(acb->bh);
3408 return &acb->common;
3411 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
3412 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
3413 BlockDriverCompletionFunc *cb, void *opaque)
3415 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
3418 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
3419 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
3420 BlockDriverCompletionFunc *cb, void *opaque)
3422 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
3426 typedef struct BlockDriverAIOCBCoroutine {
3427 BlockDriverAIOCB common;
3431 } BlockDriverAIOCBCoroutine;
3433 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb)
3438 static AIOPool bdrv_em_co_aio_pool = {
3439 .aiocb_size = sizeof(BlockDriverAIOCBCoroutine),
3440 .cancel = bdrv_aio_co_cancel_em,
3443 static void bdrv_co_em_bh(void *opaque)
3445 BlockDriverAIOCBCoroutine *acb = opaque;
3447 acb->common.cb(acb->common.opaque, acb->req.error);
3448 qemu_bh_delete(acb->bh);
3449 qemu_aio_release(acb);
3452 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
3453 static void coroutine_fn bdrv_co_do_rw(void *opaque)
3455 BlockDriverAIOCBCoroutine *acb = opaque;
3456 BlockDriverState *bs = acb->common.bs;
3458 if (!acb->is_write) {
3459 acb->req.error = bdrv_co_do_readv(bs, acb->req.sector,
3460 acb->req.nb_sectors, acb->req.qiov, 0);
3462 acb->req.error = bdrv_co_do_writev(bs, acb->req.sector,
3463 acb->req.nb_sectors, acb->req.qiov, 0);
3466 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3467 qemu_bh_schedule(acb->bh);
3470 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
3474 BlockDriverCompletionFunc *cb,
3479 BlockDriverAIOCBCoroutine *acb;
3481 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3482 acb->req.sector = sector_num;
3483 acb->req.nb_sectors = nb_sectors;
3484 acb->req.qiov = qiov;
3485 acb->is_write = is_write;
3487 co = qemu_coroutine_create(bdrv_co_do_rw);
3488 qemu_coroutine_enter(co, acb);
3490 return &acb->common;
3493 static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
3495 BlockDriverAIOCBCoroutine *acb = opaque;
3496 BlockDriverState *bs = acb->common.bs;
3498 acb->req.error = bdrv_co_flush(bs);
3499 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3500 qemu_bh_schedule(acb->bh);
3503 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
3504 BlockDriverCompletionFunc *cb, void *opaque)
3506 trace_bdrv_aio_flush(bs, opaque);
3509 BlockDriverAIOCBCoroutine *acb;
3511 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3512 co = qemu_coroutine_create(bdrv_aio_flush_co_entry);
3513 qemu_coroutine_enter(co, acb);
3515 return &acb->common;
3518 static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque)
3520 BlockDriverAIOCBCoroutine *acb = opaque;
3521 BlockDriverState *bs = acb->common.bs;
3523 acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors);
3524 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3525 qemu_bh_schedule(acb->bh);
3528 BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs,
3529 int64_t sector_num, int nb_sectors,
3530 BlockDriverCompletionFunc *cb, void *opaque)
3533 BlockDriverAIOCBCoroutine *acb;
3535 trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque);
3537 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3538 acb->req.sector = sector_num;
3539 acb->req.nb_sectors = nb_sectors;
3540 co = qemu_coroutine_create(bdrv_aio_discard_co_entry);
3541 qemu_coroutine_enter(co, acb);
3543 return &acb->common;
3546 void bdrv_init(void)
3548 module_call_init(MODULE_INIT_BLOCK);
3551 void bdrv_init_with_whitelist(void)
3553 use_bdrv_whitelist = 1;
3557 void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
3558 BlockDriverCompletionFunc *cb, void *opaque)
3560 BlockDriverAIOCB *acb;
3562 if (pool->free_aiocb) {
3563 acb = pool->free_aiocb;
3564 pool->free_aiocb = acb->next;
3566 acb = g_malloc0(pool->aiocb_size);
3571 acb->opaque = opaque;
3575 void qemu_aio_release(void *p)
3577 BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
3578 AIOPool *pool = acb->pool;
3579 acb->next = pool->free_aiocb;
3580 pool->free_aiocb = acb;
3583 /**************************************************************/
3584 /* Coroutine block device emulation */
3586 typedef struct CoroutineIOCompletion {
3587 Coroutine *coroutine;
3589 } CoroutineIOCompletion;
3591 static void bdrv_co_io_em_complete(void *opaque, int ret)
3593 CoroutineIOCompletion *co = opaque;
3596 qemu_coroutine_enter(co->coroutine, NULL);
3599 static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num,
3600 int nb_sectors, QEMUIOVector *iov,
3603 CoroutineIOCompletion co = {
3604 .coroutine = qemu_coroutine_self(),
3606 BlockDriverAIOCB *acb;
3609 acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors,
3610 bdrv_co_io_em_complete, &co);
3612 acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors,
3613 bdrv_co_io_em_complete, &co);
3616 trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb);
3620 qemu_coroutine_yield();
3625 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
3626 int64_t sector_num, int nb_sectors,
3629 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false);
3632 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
3633 int64_t sector_num, int nb_sectors,
3636 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true);
3639 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
3641 RwCo *rwco = opaque;
3643 rwco->ret = bdrv_co_flush(rwco->bs);
3646 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
3650 if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
3654 /* Write back cached data to the OS even with cache=unsafe */
3655 if (bs->drv->bdrv_co_flush_to_os) {
3656 ret = bs->drv->bdrv_co_flush_to_os(bs);
3662 /* But don't actually force it to the disk with cache=unsafe */
3663 if (bs->open_flags & BDRV_O_NO_FLUSH) {
3667 if (bs->drv->bdrv_co_flush_to_disk) {
3668 ret = bs->drv->bdrv_co_flush_to_disk(bs);
3669 } else if (bs->drv->bdrv_aio_flush) {
3670 BlockDriverAIOCB *acb;
3671 CoroutineIOCompletion co = {
3672 .coroutine = qemu_coroutine_self(),
3675 acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
3679 qemu_coroutine_yield();
3684 * Some block drivers always operate in either writethrough or unsafe
3685 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
3686 * know how the server works (because the behaviour is hardcoded or
3687 * depends on server-side configuration), so we can't ensure that
3688 * everything is safe on disk. Returning an error doesn't work because
3689 * that would break guests even if the server operates in writethrough
3692 * Let's hope the user knows what he's doing.
3700 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
3701 * in the case of cache=unsafe, so there are no useless flushes.
3703 return bdrv_co_flush(bs->file);
3706 void bdrv_invalidate_cache(BlockDriverState *bs)
3708 if (bs->drv && bs->drv->bdrv_invalidate_cache) {
3709 bs->drv->bdrv_invalidate_cache(bs);
3713 void bdrv_invalidate_cache_all(void)
3715 BlockDriverState *bs;
3717 QTAILQ_FOREACH(bs, &bdrv_states, list) {
3718 bdrv_invalidate_cache(bs);
3722 void bdrv_clear_incoming_migration_all(void)
3724 BlockDriverState *bs;
3726 QTAILQ_FOREACH(bs, &bdrv_states, list) {
3727 bs->open_flags = bs->open_flags & ~(BDRV_O_INCOMING);
3731 int bdrv_flush(BlockDriverState *bs)
3739 if (qemu_in_coroutine()) {
3740 /* Fast-path if already in coroutine context */
3741 bdrv_flush_co_entry(&rwco);
3743 co = qemu_coroutine_create(bdrv_flush_co_entry);
3744 qemu_coroutine_enter(co, &rwco);
3745 while (rwco.ret == NOT_DONE) {
3753 static void coroutine_fn bdrv_discard_co_entry(void *opaque)
3755 RwCo *rwco = opaque;
3757 rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors);
3760 int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
3765 } else if (bdrv_check_request(bs, sector_num, nb_sectors)) {
3767 } else if (bs->read_only) {
3769 } else if (bs->drv->bdrv_co_discard) {
3770 return bs->drv->bdrv_co_discard(bs, sector_num, nb_sectors);
3771 } else if (bs->drv->bdrv_aio_discard) {
3772 BlockDriverAIOCB *acb;
3773 CoroutineIOCompletion co = {
3774 .coroutine = qemu_coroutine_self(),
3777 acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors,
3778 bdrv_co_io_em_complete, &co);
3782 qemu_coroutine_yield();
3790 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
3795 .sector_num = sector_num,
3796 .nb_sectors = nb_sectors,
3800 if (qemu_in_coroutine()) {
3801 /* Fast-path if already in coroutine context */
3802 bdrv_discard_co_entry(&rwco);
3804 co = qemu_coroutine_create(bdrv_discard_co_entry);
3805 qemu_coroutine_enter(co, &rwco);
3806 while (rwco.ret == NOT_DONE) {
3814 /**************************************************************/
3815 /* removable device support */
3818 * Return TRUE if the media is present
3820 int bdrv_is_inserted(BlockDriverState *bs)
3822 BlockDriver *drv = bs->drv;
3826 if (!drv->bdrv_is_inserted)
3828 return drv->bdrv_is_inserted(bs);
3832 * Return whether the media changed since the last call to this
3833 * function, or -ENOTSUP if we don't know. Most drivers don't know.
3835 int bdrv_media_changed(BlockDriverState *bs)
3837 BlockDriver *drv = bs->drv;
3839 if (drv && drv->bdrv_media_changed) {
3840 return drv->bdrv_media_changed(bs);
3846 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
3848 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
3850 BlockDriver *drv = bs->drv;
3852 if (drv && drv->bdrv_eject) {
3853 drv->bdrv_eject(bs, eject_flag);
3856 if (bs->device_name[0] != '\0') {
3857 bdrv_emit_qmp_eject_event(bs, eject_flag);
3862 * Lock or unlock the media (if it is locked, the user won't be able
3863 * to eject it manually).
3865 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
3867 BlockDriver *drv = bs->drv;
3869 trace_bdrv_lock_medium(bs, locked);
3871 if (drv && drv->bdrv_lock_medium) {
3872 drv->bdrv_lock_medium(bs, locked);
3876 /* needed for generic scsi interface */
3878 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
3880 BlockDriver *drv = bs->drv;
3882 if (drv && drv->bdrv_ioctl)
3883 return drv->bdrv_ioctl(bs, req, buf);
3887 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
3888 unsigned long int req, void *buf,
3889 BlockDriverCompletionFunc *cb, void *opaque)
3891 BlockDriver *drv = bs->drv;
3893 if (drv && drv->bdrv_aio_ioctl)
3894 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
3898 void bdrv_set_buffer_alignment(BlockDriverState *bs, int align)
3900 bs->buffer_alignment = align;
3903 void *qemu_blockalign(BlockDriverState *bs, size_t size)
3905 return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
3908 void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
3910 int64_t bitmap_size;
3912 bs->dirty_count = 0;
3914 if (!bs->dirty_bitmap) {
3915 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
3916 BDRV_SECTORS_PER_DIRTY_CHUNK * BITS_PER_LONG - 1;
3917 bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * BITS_PER_LONG;
3919 bs->dirty_bitmap = g_new0(unsigned long, bitmap_size);
3922 if (bs->dirty_bitmap) {
3923 g_free(bs->dirty_bitmap);
3924 bs->dirty_bitmap = NULL;
3929 int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
3931 int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
3933 if (bs->dirty_bitmap &&
3934 (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
3935 return !!(bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
3936 (1UL << (chunk % (sizeof(unsigned long) * 8))));
3942 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
3945 set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
3948 int64_t bdrv_get_dirty_count(BlockDriverState *bs)
3950 return bs->dirty_count;
3953 void bdrv_set_in_use(BlockDriverState *bs, int in_use)
3955 assert(bs->in_use != in_use);
3956 bs->in_use = in_use;
3959 int bdrv_in_use(BlockDriverState *bs)
3964 void bdrv_iostatus_enable(BlockDriverState *bs)
3966 bs->iostatus_enabled = true;
3967 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
3970 /* The I/O status is only enabled if the drive explicitly
3971 * enables it _and_ the VM is configured to stop on errors */
3972 bool bdrv_iostatus_is_enabled(const BlockDriverState *bs)
3974 return (bs->iostatus_enabled &&
3975 (bs->on_write_error == BLOCK_ERR_STOP_ENOSPC ||
3976 bs->on_write_error == BLOCK_ERR_STOP_ANY ||
3977 bs->on_read_error == BLOCK_ERR_STOP_ANY));
3980 void bdrv_iostatus_disable(BlockDriverState *bs)
3982 bs->iostatus_enabled = false;
3985 void bdrv_iostatus_reset(BlockDriverState *bs)
3987 if (bdrv_iostatus_is_enabled(bs)) {
3988 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
3992 /* XXX: Today this is set by device models because it makes the implementation
3993 quite simple. However, the block layer knows about the error, so it's
3994 possible to implement this without device models being involved */
3995 void bdrv_iostatus_set_err(BlockDriverState *bs, int error)
3997 if (bdrv_iostatus_is_enabled(bs) &&
3998 bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
4000 bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
4001 BLOCK_DEVICE_IO_STATUS_FAILED;
4006 bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes,
4007 enum BlockAcctType type)
4009 assert(type < BDRV_MAX_IOTYPE);
4011 cookie->bytes = bytes;
4012 cookie->start_time_ns = get_clock();
4013 cookie->type = type;
4017 bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie)
4019 assert(cookie->type < BDRV_MAX_IOTYPE);
4021 bs->nr_bytes[cookie->type] += cookie->bytes;
4022 bs->nr_ops[cookie->type]++;
4023 bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns;
4026 int bdrv_img_create(const char *filename, const char *fmt,
4027 const char *base_filename, const char *base_fmt,
4028 char *options, uint64_t img_size, int flags)
4030 QEMUOptionParameter *param = NULL, *create_options = NULL;
4031 QEMUOptionParameter *backing_fmt, *backing_file, *size;
4032 BlockDriverState *bs = NULL;
4033 BlockDriver *drv, *proto_drv;
4034 BlockDriver *backing_drv = NULL;
4037 /* Find driver and parse its options */
4038 drv = bdrv_find_format(fmt);
4040 error_report("Unknown file format '%s'", fmt);
4045 proto_drv = bdrv_find_protocol(filename);
4047 error_report("Unknown protocol '%s'", filename);
4052 create_options = append_option_parameters(create_options,
4053 drv->create_options);
4054 create_options = append_option_parameters(create_options,
4055 proto_drv->create_options);
4057 /* Create parameter list with default values */
4058 param = parse_option_parameters("", create_options, param);
4060 set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size);
4062 /* Parse -o options */
4064 param = parse_option_parameters(options, create_options, param);
4065 if (param == NULL) {
4066 error_report("Invalid options for file format '%s'.", fmt);
4072 if (base_filename) {
4073 if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE,
4075 error_report("Backing file not supported for file format '%s'",
4083 if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) {
4084 error_report("Backing file format not supported for file "
4085 "format '%s'", fmt);
4091 backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE);
4092 if (backing_file && backing_file->value.s) {
4093 if (!strcmp(filename, backing_file->value.s)) {
4094 error_report("Error: Trying to create an image with the "
4095 "same filename as the backing file");
4101 backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT);
4102 if (backing_fmt && backing_fmt->value.s) {
4103 backing_drv = bdrv_find_format(backing_fmt->value.s);
4105 error_report("Unknown backing file format '%s'",
4106 backing_fmt->value.s);
4112 // The size for the image must always be specified, with one exception:
4113 // If we are using a backing file, we can obtain the size from there
4114 size = get_option_parameter(param, BLOCK_OPT_SIZE);
4115 if (size && size->value.n == -1) {
4116 if (backing_file && backing_file->value.s) {
4121 /* backing files always opened read-only */
4123 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
4127 ret = bdrv_open(bs, backing_file->value.s, back_flags, backing_drv);
4129 error_report("Could not open '%s'", backing_file->value.s);
4132 bdrv_get_geometry(bs, &size);
4135 snprintf(buf, sizeof(buf), "%" PRId64, size);
4136 set_option_parameter(param, BLOCK_OPT_SIZE, buf);
4138 error_report("Image creation needs a size parameter");
4144 printf("Formatting '%s', fmt=%s ", filename, fmt);
4145 print_option_parameters(param);
4148 ret = bdrv_create(drv, filename, param);
4151 if (ret == -ENOTSUP) {
4152 error_report("Formatting or formatting option not supported for "
4153 "file format '%s'", fmt);
4154 } else if (ret == -EFBIG) {
4155 error_report("The image size is too large for file format '%s'",
4158 error_report("%s: error while creating %s: %s", filename, fmt,
4164 free_option_parameters(create_options);
4165 free_option_parameters(param);
4174 void *block_job_create(const BlockJobType *job_type, BlockDriverState *bs,
4175 int64_t speed, BlockDriverCompletionFunc *cb,
4176 void *opaque, Error **errp)
4180 if (bs->job || bdrv_in_use(bs)) {
4181 error_set(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
4184 bdrv_set_in_use(bs, 1);
4186 job = g_malloc0(job_type->instance_size);
4187 job->job_type = job_type;
4190 job->opaque = opaque;
4194 /* Only set speed when necessary to avoid NotSupported error */
4196 Error *local_err = NULL;
4198 block_job_set_speed(job, speed, &local_err);
4199 if (error_is_set(&local_err)) {
4202 bdrv_set_in_use(bs, 0);
4203 error_propagate(errp, local_err);
4210 void block_job_complete(BlockJob *job, int ret)
4212 BlockDriverState *bs = job->bs;
4214 assert(bs->job == job);
4215 job->cb(job->opaque, ret);
4218 bdrv_set_in_use(bs, 0);
4221 void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
4223 Error *local_err = NULL;
4225 if (!job->job_type->set_speed) {
4226 error_set(errp, QERR_NOT_SUPPORTED);
4229 job->job_type->set_speed(job, speed, &local_err);
4230 if (error_is_set(&local_err)) {
4231 error_propagate(errp, local_err);
4238 void block_job_cancel(BlockJob *job)
4240 job->cancelled = true;
4241 if (job->co && !job->busy) {
4242 qemu_coroutine_enter(job->co, NULL);
4246 bool block_job_is_cancelled(BlockJob *job)
4248 return job->cancelled;
4251 struct BlockCancelData {
4253 BlockDriverCompletionFunc *cb;
4259 static void block_job_cancel_cb(void *opaque, int ret)
4261 struct BlockCancelData *data = opaque;
4263 data->cancelled = block_job_is_cancelled(data->job);
4265 data->cb(data->opaque, ret);
4268 int block_job_cancel_sync(BlockJob *job)
4270 struct BlockCancelData data;
4271 BlockDriverState *bs = job->bs;
4273 assert(bs->job == job);
4275 /* Set up our own callback to store the result and chain to
4276 * the original callback.
4280 data.opaque = job->opaque;
4281 data.ret = -EINPROGRESS;
4282 job->cb = block_job_cancel_cb;
4283 job->opaque = &data;
4284 block_job_cancel(job);
4285 while (data.ret == -EINPROGRESS) {
4288 return (data.cancelled && data.ret == 0) ? -ECANCELED : data.ret;
4291 void block_job_sleep_ns(BlockJob *job, QEMUClock *clock, int64_t ns)
4293 /* Check cancellation *before* setting busy = false, too! */
4294 if (!block_job_is_cancelled(job)) {
4296 co_sleep_ns(clock, ns);