2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24 #include "config-host.h"
25 #include "qemu-common.h"
28 #include "block_int.h"
31 #include "qemu-coroutine.h"
32 #include "qmp-commands.h"
33 #include "qemu-timer.h"
36 #include <sys/types.h>
38 #include <sys/ioctl.h>
39 #include <sys/queue.h>
49 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
52 BDRV_REQ_COPY_ON_READ = 0x1,
53 BDRV_REQ_ZERO_WRITE = 0x2,
56 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load);
57 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
58 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
59 BlockDriverCompletionFunc *cb, void *opaque);
60 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
61 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
62 BlockDriverCompletionFunc *cb, void *opaque);
63 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
64 int64_t sector_num, int nb_sectors,
66 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
67 int64_t sector_num, int nb_sectors,
69 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
70 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
71 BdrvRequestFlags flags);
72 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
73 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
74 BdrvRequestFlags flags);
75 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
79 BlockDriverCompletionFunc *cb,
82 static void coroutine_fn bdrv_co_do_rw(void *opaque);
83 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
84 int64_t sector_num, int nb_sectors);
86 static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors,
87 bool is_write, double elapsed_time, uint64_t *wait);
88 static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write,
89 double elapsed_time, uint64_t *wait);
90 static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors,
91 bool is_write, int64_t *wait);
93 static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
94 QTAILQ_HEAD_INITIALIZER(bdrv_states);
96 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
97 QLIST_HEAD_INITIALIZER(bdrv_drivers);
99 /* The device to use for VM snapshots */
100 static BlockDriverState *bs_snapshots;
102 /* If non-zero, use only whitelisted block drivers */
103 static int use_bdrv_whitelist;
106 static int is_windows_drive_prefix(const char *filename)
108 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
109 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
113 int is_windows_drive(const char *filename)
115 if (is_windows_drive_prefix(filename) &&
118 if (strstart(filename, "\\\\.\\", NULL) ||
119 strstart(filename, "//./", NULL))
125 /* throttling disk I/O limits */
126 void bdrv_io_limits_disable(BlockDriverState *bs)
128 bs->io_limits_enabled = false;
130 while (qemu_co_queue_next(&bs->throttled_reqs));
132 if (bs->block_timer) {
133 qemu_del_timer(bs->block_timer);
134 qemu_free_timer(bs->block_timer);
135 bs->block_timer = NULL;
141 memset(&bs->io_base, 0, sizeof(bs->io_base));
144 static void bdrv_block_timer(void *opaque)
146 BlockDriverState *bs = opaque;
148 qemu_co_queue_next(&bs->throttled_reqs);
151 void bdrv_io_limits_enable(BlockDriverState *bs)
153 qemu_co_queue_init(&bs->throttled_reqs);
154 bs->block_timer = qemu_new_timer_ns(vm_clock, bdrv_block_timer, bs);
155 bs->slice_time = 5 * BLOCK_IO_SLICE_TIME;
156 bs->slice_start = qemu_get_clock_ns(vm_clock);
157 bs->slice_end = bs->slice_start + bs->slice_time;
158 memset(&bs->io_base, 0, sizeof(bs->io_base));
159 bs->io_limits_enabled = true;
162 bool bdrv_io_limits_enabled(BlockDriverState *bs)
164 BlockIOLimit *io_limits = &bs->io_limits;
165 return io_limits->bps[BLOCK_IO_LIMIT_READ]
166 || io_limits->bps[BLOCK_IO_LIMIT_WRITE]
167 || io_limits->bps[BLOCK_IO_LIMIT_TOTAL]
168 || io_limits->iops[BLOCK_IO_LIMIT_READ]
169 || io_limits->iops[BLOCK_IO_LIMIT_WRITE]
170 || io_limits->iops[BLOCK_IO_LIMIT_TOTAL];
173 static void bdrv_io_limits_intercept(BlockDriverState *bs,
174 bool is_write, int nb_sectors)
176 int64_t wait_time = -1;
178 if (!qemu_co_queue_empty(&bs->throttled_reqs)) {
179 qemu_co_queue_wait(&bs->throttled_reqs);
182 /* In fact, we hope to keep each request's timing, in FIFO mode. The next
183 * throttled requests will not be dequeued until the current request is
184 * allowed to be serviced. So if the current request still exceeds the
185 * limits, it will be inserted to the head. All requests followed it will
186 * be still in throttled_reqs queue.
189 while (bdrv_exceed_io_limits(bs, nb_sectors, is_write, &wait_time)) {
190 qemu_mod_timer(bs->block_timer,
191 wait_time + qemu_get_clock_ns(vm_clock));
192 qemu_co_queue_wait_insert_head(&bs->throttled_reqs);
195 qemu_co_queue_next(&bs->throttled_reqs);
198 /* check if the path starts with "<protocol>:" */
199 static int path_has_protocol(const char *path)
202 if (is_windows_drive(path) ||
203 is_windows_drive_prefix(path)) {
208 return strchr(path, ':') != NULL;
211 int path_is_absolute(const char *path)
215 /* specific case for names like: "\\.\d:" */
216 if (*path == '/' || *path == '\\')
219 p = strchr(path, ':');
225 return (*p == '/' || *p == '\\');
231 /* if filename is absolute, just copy it to dest. Otherwise, build a
232 path to it by considering it is relative to base_path. URL are
234 void path_combine(char *dest, int dest_size,
235 const char *base_path,
236 const char *filename)
243 if (path_is_absolute(filename)) {
244 pstrcpy(dest, dest_size, filename);
246 p = strchr(base_path, ':');
251 p1 = strrchr(base_path, '/');
255 p2 = strrchr(base_path, '\\');
267 if (len > dest_size - 1)
269 memcpy(dest, base_path, len);
271 pstrcat(dest, dest_size, filename);
275 void bdrv_register(BlockDriver *bdrv)
277 /* Block drivers without coroutine functions need emulation */
278 if (!bdrv->bdrv_co_readv) {
279 bdrv->bdrv_co_readv = bdrv_co_readv_em;
280 bdrv->bdrv_co_writev = bdrv_co_writev_em;
282 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
283 * the block driver lacks aio we need to emulate that too.
285 if (!bdrv->bdrv_aio_readv) {
286 /* add AIO emulation layer */
287 bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
288 bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
292 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
295 /* create a new block device (by default it is empty) */
296 BlockDriverState *bdrv_new(const char *device_name)
298 BlockDriverState *bs;
300 bs = g_malloc0(sizeof(BlockDriverState));
301 pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
302 if (device_name[0] != '\0') {
303 QTAILQ_INSERT_TAIL(&bdrv_states, bs, list);
305 bdrv_iostatus_disable(bs);
309 BlockDriver *bdrv_find_format(const char *format_name)
312 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
313 if (!strcmp(drv1->format_name, format_name)) {
320 static int bdrv_is_whitelisted(BlockDriver *drv)
322 static const char *whitelist[] = {
323 CONFIG_BDRV_WHITELIST
328 return 1; /* no whitelist, anything goes */
330 for (p = whitelist; *p; p++) {
331 if (!strcmp(drv->format_name, *p)) {
338 BlockDriver *bdrv_find_whitelisted_format(const char *format_name)
340 BlockDriver *drv = bdrv_find_format(format_name);
341 return drv && bdrv_is_whitelisted(drv) ? drv : NULL;
344 typedef struct CreateCo {
347 QEMUOptionParameter *options;
351 static void coroutine_fn bdrv_create_co_entry(void *opaque)
353 CreateCo *cco = opaque;
356 cco->ret = cco->drv->bdrv_create(cco->filename, cco->options);
359 int bdrv_create(BlockDriver *drv, const char* filename,
360 QEMUOptionParameter *options)
367 .filename = g_strdup(filename),
372 if (!drv->bdrv_create) {
376 if (qemu_in_coroutine()) {
377 /* Fast-path if already in coroutine context */
378 bdrv_create_co_entry(&cco);
380 co = qemu_coroutine_create(bdrv_create_co_entry);
381 qemu_coroutine_enter(co, &cco);
382 while (cco.ret == NOT_DONE) {
388 g_free(cco.filename);
393 int bdrv_create_file(const char* filename, QEMUOptionParameter *options)
397 drv = bdrv_find_protocol(filename);
402 return bdrv_create(drv, filename, options);
406 void get_tmp_filename(char *filename, int size)
408 char temp_dir[MAX_PATH];
410 GetTempPath(MAX_PATH, temp_dir);
411 GetTempFileName(temp_dir, "qem", 0, filename);
414 void get_tmp_filename(char *filename, int size)
418 /* XXX: race condition possible */
419 tmpdir = getenv("TMPDIR");
422 snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
423 fd = mkstemp(filename);
429 * Detect host devices. By convention, /dev/cdrom[N] is always
430 * recognized as a host CDROM.
432 static BlockDriver *find_hdev_driver(const char *filename)
434 int score_max = 0, score;
435 BlockDriver *drv = NULL, *d;
437 QLIST_FOREACH(d, &bdrv_drivers, list) {
438 if (d->bdrv_probe_device) {
439 score = d->bdrv_probe_device(filename);
440 if (score > score_max) {
450 BlockDriver *bdrv_find_protocol(const char *filename)
457 /* TODO Drivers without bdrv_file_open must be specified explicitly */
460 * XXX(hch): we really should not let host device detection
461 * override an explicit protocol specification, but moving this
462 * later breaks access to device names with colons in them.
463 * Thanks to the brain-dead persistent naming schemes on udev-
464 * based Linux systems those actually are quite common.
466 drv1 = find_hdev_driver(filename);
471 if (!path_has_protocol(filename)) {
472 return bdrv_find_format("file");
474 p = strchr(filename, ':');
477 if (len > sizeof(protocol) - 1)
478 len = sizeof(protocol) - 1;
479 memcpy(protocol, filename, len);
480 protocol[len] = '\0';
481 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
482 if (drv1->protocol_name &&
483 !strcmp(drv1->protocol_name, protocol)) {
490 static int find_image_format(const char *filename, BlockDriver **pdrv)
492 int ret, score, score_max;
493 BlockDriver *drv1, *drv;
495 BlockDriverState *bs;
497 ret = bdrv_file_open(&bs, filename, 0);
503 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
504 if (bs->sg || !bdrv_is_inserted(bs)) {
506 drv = bdrv_find_format("raw");
514 ret = bdrv_pread(bs, 0, buf, sizeof(buf));
523 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
524 if (drv1->bdrv_probe) {
525 score = drv1->bdrv_probe(buf, ret, filename);
526 if (score > score_max) {
540 * Set the current 'total_sectors' value
542 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
544 BlockDriver *drv = bs->drv;
546 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
550 /* query actual device if possible, otherwise just trust the hint */
551 if (drv->bdrv_getlength) {
552 int64_t length = drv->bdrv_getlength(bs);
556 hint = length >> BDRV_SECTOR_BITS;
559 bs->total_sectors = hint;
564 * Set open flags for a given cache mode
566 * Return 0 on success, -1 if the cache mode was invalid.
568 int bdrv_parse_cache_flags(const char *mode, int *flags)
570 *flags &= ~BDRV_O_CACHE_MASK;
572 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
573 *flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB;
574 } else if (!strcmp(mode, "directsync")) {
575 *flags |= BDRV_O_NOCACHE;
576 } else if (!strcmp(mode, "writeback")) {
577 *flags |= BDRV_O_CACHE_WB;
578 } else if (!strcmp(mode, "unsafe")) {
579 *flags |= BDRV_O_CACHE_WB;
580 *flags |= BDRV_O_NO_FLUSH;
581 } else if (!strcmp(mode, "writethrough")) {
582 /* this is the default */
591 * The copy-on-read flag is actually a reference count so multiple users may
592 * use the feature without worrying about clobbering its previous state.
593 * Copy-on-read stays enabled until all users have called to disable it.
595 void bdrv_enable_copy_on_read(BlockDriverState *bs)
600 void bdrv_disable_copy_on_read(BlockDriverState *bs)
602 assert(bs->copy_on_read > 0);
607 * Common part for opening disk images and files
609 static int bdrv_open_common(BlockDriverState *bs, const char *filename,
610 int flags, BlockDriver *drv)
616 trace_bdrv_open_common(bs, filename, flags, drv->format_name);
619 bs->total_sectors = 0;
623 bs->open_flags = flags;
625 bs->buffer_alignment = 512;
627 assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
628 if ((flags & BDRV_O_RDWR) && (flags & BDRV_O_COPY_ON_READ)) {
629 bdrv_enable_copy_on_read(bs);
632 pstrcpy(bs->filename, sizeof(bs->filename), filename);
633 bs->backing_file[0] = '\0';
635 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv)) {
640 bs->opaque = g_malloc0(drv->instance_size);
642 bs->enable_write_cache = !!(flags & BDRV_O_CACHE_WB);
645 * Clear flags that are internal to the block layer before opening the
648 open_flags = flags & ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
651 * Snapshots should be writable.
653 if (bs->is_temporary) {
654 open_flags |= BDRV_O_RDWR;
657 bs->keep_read_only = bs->read_only = !(open_flags & BDRV_O_RDWR);
659 /* Open the image, either directly or using a protocol */
660 if (drv->bdrv_file_open) {
661 ret = drv->bdrv_file_open(bs, filename, open_flags);
663 ret = bdrv_file_open(&bs->file, filename, open_flags);
665 ret = drv->bdrv_open(bs, open_flags);
673 ret = refresh_total_sectors(bs, bs->total_sectors);
679 if (bs->is_temporary) {
687 bdrv_delete(bs->file);
697 * Opens a file using a protocol (file, host_device, nbd, ...)
699 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
701 BlockDriverState *bs;
705 drv = bdrv_find_protocol(filename);
711 ret = bdrv_open_common(bs, filename, flags, drv);
722 * Opens a disk image (raw, qcow2, vmdk, ...)
724 int bdrv_open(BlockDriverState *bs, const char *filename, int flags,
728 char tmp_filename[PATH_MAX];
730 if (flags & BDRV_O_SNAPSHOT) {
731 BlockDriverState *bs1;
734 BlockDriver *bdrv_qcow2;
735 QEMUOptionParameter *options;
736 char backing_filename[PATH_MAX];
738 /* if snapshot, we create a temporary backing file and open it
739 instead of opening 'filename' directly */
741 /* if there is a backing file, use it */
743 ret = bdrv_open(bs1, filename, 0, drv);
748 total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
750 if (bs1->drv && bs1->drv->protocol_name)
755 get_tmp_filename(tmp_filename, sizeof(tmp_filename));
757 /* Real path is meaningless for protocols */
759 snprintf(backing_filename, sizeof(backing_filename),
761 else if (!realpath(filename, backing_filename))
764 bdrv_qcow2 = bdrv_find_format("qcow2");
765 options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
767 set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size);
768 set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
770 set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
774 ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
775 free_option_parameters(options);
780 filename = tmp_filename;
782 bs->is_temporary = 1;
785 /* Find the right image format driver */
787 ret = find_image_format(filename, &drv);
791 goto unlink_and_fail;
795 ret = bdrv_open_common(bs, filename, flags, drv);
797 goto unlink_and_fail;
800 /* If there is a backing file, use it */
801 if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') {
802 char backing_filename[PATH_MAX];
804 BlockDriver *back_drv = NULL;
806 bs->backing_hd = bdrv_new("");
808 if (path_has_protocol(bs->backing_file)) {
809 pstrcpy(backing_filename, sizeof(backing_filename),
812 path_combine(backing_filename, sizeof(backing_filename),
813 filename, bs->backing_file);
816 if (bs->backing_format[0] != '\0') {
817 back_drv = bdrv_find_format(bs->backing_format);
820 /* backing files always opened read-only */
822 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
824 ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv);
829 if (bs->is_temporary) {
830 bs->backing_hd->keep_read_only = !(flags & BDRV_O_RDWR);
832 /* base image inherits from "parent" */
833 bs->backing_hd->keep_read_only = bs->keep_read_only;
837 if (!bdrv_key_required(bs)) {
838 bdrv_dev_change_media_cb(bs, true);
841 /* throttling disk I/O limits */
842 if (bs->io_limits_enabled) {
843 bdrv_io_limits_enable(bs);
849 if (bs->is_temporary) {
855 void bdrv_close(BlockDriverState *bs)
860 block_job_cancel_sync(bs->job);
864 if (bs == bs_snapshots) {
867 if (bs->backing_hd) {
868 bdrv_delete(bs->backing_hd);
869 bs->backing_hd = NULL;
871 bs->drv->bdrv_close(bs);
874 if (bs->is_temporary) {
875 unlink(bs->filename);
880 bs->copy_on_read = 0;
882 if (bs->file != NULL) {
883 bdrv_close(bs->file);
886 bdrv_dev_change_media_cb(bs, false);
889 /*throttling disk I/O limits*/
890 if (bs->io_limits_enabled) {
891 bdrv_io_limits_disable(bs);
895 void bdrv_close_all(void)
897 BlockDriverState *bs;
899 QTAILQ_FOREACH(bs, &bdrv_states, list) {
905 * Wait for pending requests to complete across all BlockDriverStates
907 * This function does not flush data to disk, use bdrv_flush_all() for that
908 * after calling this function.
910 * Note that completion of an asynchronous I/O operation can trigger any
911 * number of other I/O operations on other devices---for example a coroutine
912 * can be arbitrarily complex and a constant flow of I/O can come until the
913 * coroutine is complete. Because of this, it is not possible to have a
914 * function to drain a single device's I/O queue.
916 void bdrv_drain_all(void)
918 BlockDriverState *bs;
922 busy = qemu_aio_wait();
924 /* FIXME: We do not have timer support here, so this is effectively
927 QTAILQ_FOREACH(bs, &bdrv_states, list) {
928 if (!qemu_co_queue_empty(&bs->throttled_reqs)) {
929 qemu_co_queue_restart_all(&bs->throttled_reqs);
935 /* If requests are still pending there is a bug somewhere */
936 QTAILQ_FOREACH(bs, &bdrv_states, list) {
937 assert(QLIST_EMPTY(&bs->tracked_requests));
938 assert(qemu_co_queue_empty(&bs->throttled_reqs));
942 /* make a BlockDriverState anonymous by removing from bdrv_state list.
943 Also, NULL terminate the device_name to prevent double remove */
944 void bdrv_make_anon(BlockDriverState *bs)
946 if (bs->device_name[0] != '\0') {
947 QTAILQ_REMOVE(&bdrv_states, bs, list);
949 bs->device_name[0] = '\0';
953 * Add new bs contents at the top of an image chain while the chain is
954 * live, while keeping required fields on the top layer.
956 * This will modify the BlockDriverState fields, and swap contents
957 * between bs_new and bs_top. Both bs_new and bs_top are modified.
959 * bs_new is required to be anonymous.
961 * This function does not create any image files.
963 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
965 BlockDriverState tmp;
967 /* bs_new must be anonymous */
968 assert(bs_new->device_name[0] == '\0');
972 /* there are some fields that need to stay on the top layer: */
975 tmp.dev_ops = bs_top->dev_ops;
976 tmp.dev_opaque = bs_top->dev_opaque;
977 tmp.dev = bs_top->dev;
978 tmp.buffer_alignment = bs_top->buffer_alignment;
979 tmp.copy_on_read = bs_top->copy_on_read;
981 /* i/o timing parameters */
982 tmp.slice_time = bs_top->slice_time;
983 tmp.slice_start = bs_top->slice_start;
984 tmp.slice_end = bs_top->slice_end;
985 tmp.io_limits = bs_top->io_limits;
986 tmp.io_base = bs_top->io_base;
987 tmp.throttled_reqs = bs_top->throttled_reqs;
988 tmp.block_timer = bs_top->block_timer;
989 tmp.io_limits_enabled = bs_top->io_limits_enabled;
992 tmp.cyls = bs_top->cyls;
993 tmp.heads = bs_top->heads;
994 tmp.secs = bs_top->secs;
995 tmp.translation = bs_top->translation;
998 tmp.on_read_error = bs_top->on_read_error;
999 tmp.on_write_error = bs_top->on_write_error;
1002 tmp.iostatus_enabled = bs_top->iostatus_enabled;
1003 tmp.iostatus = bs_top->iostatus;
1005 /* keep the same entry in bdrv_states */
1006 pstrcpy(tmp.device_name, sizeof(tmp.device_name), bs_top->device_name);
1007 tmp.list = bs_top->list;
1009 /* The contents of 'tmp' will become bs_top, as we are
1010 * swapping bs_new and bs_top contents. */
1011 tmp.backing_hd = bs_new;
1012 pstrcpy(tmp.backing_file, sizeof(tmp.backing_file), bs_top->filename);
1013 bdrv_get_format(bs_top, tmp.backing_format, sizeof(tmp.backing_format));
1015 /* swap contents of the fixed new bs and the current top */
1019 /* device_name[] was carried over from the old bs_top. bs_new
1020 * shouldn't be in bdrv_states, so we need to make device_name[]
1021 * reflect the anonymity of bs_new
1023 bs_new->device_name[0] = '\0';
1025 /* clear the copied fields in the new backing file */
1026 bdrv_detach_dev(bs_new, bs_new->dev);
1028 qemu_co_queue_init(&bs_new->throttled_reqs);
1029 memset(&bs_new->io_base, 0, sizeof(bs_new->io_base));
1030 memset(&bs_new->io_limits, 0, sizeof(bs_new->io_limits));
1031 bdrv_iostatus_disable(bs_new);
1033 /* we don't use bdrv_io_limits_disable() for this, because we don't want
1034 * to affect or delete the block_timer, as it has been moved to bs_top */
1035 bs_new->io_limits_enabled = false;
1036 bs_new->block_timer = NULL;
1037 bs_new->slice_time = 0;
1038 bs_new->slice_start = 0;
1039 bs_new->slice_end = 0;
1042 void bdrv_delete(BlockDriverState *bs)
1046 assert(!bs->in_use);
1048 /* remove from list, if necessary */
1052 if (bs->file != NULL) {
1053 bdrv_delete(bs->file);
1056 assert(bs != bs_snapshots);
1060 int bdrv_attach_dev(BlockDriverState *bs, void *dev)
1061 /* TODO change to DeviceState *dev when all users are qdevified */
1067 bdrv_iostatus_reset(bs);
1071 /* TODO qdevified devices don't use this, remove when devices are qdevified */
1072 void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev)
1074 if (bdrv_attach_dev(bs, dev) < 0) {
1079 void bdrv_detach_dev(BlockDriverState *bs, void *dev)
1080 /* TODO change to DeviceState *dev when all users are qdevified */
1082 assert(bs->dev == dev);
1085 bs->dev_opaque = NULL;
1086 bs->buffer_alignment = 512;
1089 /* TODO change to return DeviceState * when all users are qdevified */
1090 void *bdrv_get_attached_dev(BlockDriverState *bs)
1095 void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops,
1099 bs->dev_opaque = opaque;
1100 if (bdrv_dev_has_removable_media(bs) && bs == bs_snapshots) {
1101 bs_snapshots = NULL;
1105 void bdrv_emit_qmp_error_event(const BlockDriverState *bdrv,
1106 BlockQMPEventAction action, int is_read)
1109 const char *action_str;
1112 case BDRV_ACTION_REPORT:
1113 action_str = "report";
1115 case BDRV_ACTION_IGNORE:
1116 action_str = "ignore";
1118 case BDRV_ACTION_STOP:
1119 action_str = "stop";
1125 data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1128 is_read ? "read" : "write");
1129 monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
1131 qobject_decref(data);
1134 static void bdrv_emit_qmp_eject_event(BlockDriverState *bs, bool ejected)
1138 data = qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }",
1139 bdrv_get_device_name(bs), ejected);
1140 monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED, data);
1142 qobject_decref(data);
1145 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load)
1147 if (bs->dev_ops && bs->dev_ops->change_media_cb) {
1148 bool tray_was_closed = !bdrv_dev_is_tray_open(bs);
1149 bs->dev_ops->change_media_cb(bs->dev_opaque, load);
1150 if (tray_was_closed) {
1152 bdrv_emit_qmp_eject_event(bs, true);
1156 bdrv_emit_qmp_eject_event(bs, false);
1161 bool bdrv_dev_has_removable_media(BlockDriverState *bs)
1163 return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb);
1166 void bdrv_dev_eject_request(BlockDriverState *bs, bool force)
1168 if (bs->dev_ops && bs->dev_ops->eject_request_cb) {
1169 bs->dev_ops->eject_request_cb(bs->dev_opaque, force);
1173 bool bdrv_dev_is_tray_open(BlockDriverState *bs)
1175 if (bs->dev_ops && bs->dev_ops->is_tray_open) {
1176 return bs->dev_ops->is_tray_open(bs->dev_opaque);
1181 static void bdrv_dev_resize_cb(BlockDriverState *bs)
1183 if (bs->dev_ops && bs->dev_ops->resize_cb) {
1184 bs->dev_ops->resize_cb(bs->dev_opaque);
1188 bool bdrv_dev_is_medium_locked(BlockDriverState *bs)
1190 if (bs->dev_ops && bs->dev_ops->is_medium_locked) {
1191 return bs->dev_ops->is_medium_locked(bs->dev_opaque);
1197 * Run consistency checks on an image
1199 * Returns 0 if the check could be completed (it doesn't mean that the image is
1200 * free of errors) or -errno when an internal error occurred. The results of the
1201 * check are stored in res.
1203 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res)
1205 if (bs->drv->bdrv_check == NULL) {
1209 memset(res, 0, sizeof(*res));
1210 return bs->drv->bdrv_check(bs, res);
1213 #define COMMIT_BUF_SECTORS 2048
1215 /* commit COW file into the raw image */
1216 int bdrv_commit(BlockDriverState *bs)
1218 BlockDriver *drv = bs->drv;
1219 BlockDriver *backing_drv;
1220 int64_t sector, total_sectors;
1221 int n, ro, open_flags;
1222 int ret = 0, rw_ret = 0;
1224 char filename[1024];
1225 BlockDriverState *bs_rw, *bs_ro;
1230 if (!bs->backing_hd) {
1234 if (bs->backing_hd->keep_read_only) {
1238 if (bdrv_in_use(bs) || bdrv_in_use(bs->backing_hd)) {
1242 backing_drv = bs->backing_hd->drv;
1243 ro = bs->backing_hd->read_only;
1244 strncpy(filename, bs->backing_hd->filename, sizeof(filename));
1245 open_flags = bs->backing_hd->open_flags;
1249 bdrv_delete(bs->backing_hd);
1250 bs->backing_hd = NULL;
1251 bs_rw = bdrv_new("");
1252 rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR,
1256 /* try to re-open read-only */
1257 bs_ro = bdrv_new("");
1258 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
1262 /* drive not functional anymore */
1266 bs->backing_hd = bs_ro;
1269 bs->backing_hd = bs_rw;
1272 total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
1273 buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
1275 for (sector = 0; sector < total_sectors; sector += n) {
1276 if (bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n)) {
1278 if (bdrv_read(bs, sector, buf, n) != 0) {
1283 if (bdrv_write(bs->backing_hd, sector, buf, n) != 0) {
1290 if (drv->bdrv_make_empty) {
1291 ret = drv->bdrv_make_empty(bs);
1296 * Make sure all data we wrote to the backing device is actually
1300 bdrv_flush(bs->backing_hd);
1307 bdrv_delete(bs->backing_hd);
1308 bs->backing_hd = NULL;
1309 bs_ro = bdrv_new("");
1310 ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
1314 /* drive not functional anymore */
1318 bs->backing_hd = bs_ro;
1319 bs->backing_hd->keep_read_only = 0;
1325 int bdrv_commit_all(void)
1327 BlockDriverState *bs;
1329 QTAILQ_FOREACH(bs, &bdrv_states, list) {
1330 int ret = bdrv_commit(bs);
1338 struct BdrvTrackedRequest {
1339 BlockDriverState *bs;
1343 QLIST_ENTRY(BdrvTrackedRequest) list;
1344 Coroutine *co; /* owner, used for deadlock detection */
1345 CoQueue wait_queue; /* coroutines blocked on this request */
1349 * Remove an active request from the tracked requests list
1351 * This function should be called when a tracked request is completing.
1353 static void tracked_request_end(BdrvTrackedRequest *req)
1355 QLIST_REMOVE(req, list);
1356 qemu_co_queue_restart_all(&req->wait_queue);
1360 * Add an active request to the tracked requests list
1362 static void tracked_request_begin(BdrvTrackedRequest *req,
1363 BlockDriverState *bs,
1365 int nb_sectors, bool is_write)
1367 *req = (BdrvTrackedRequest){
1369 .sector_num = sector_num,
1370 .nb_sectors = nb_sectors,
1371 .is_write = is_write,
1372 .co = qemu_coroutine_self(),
1375 qemu_co_queue_init(&req->wait_queue);
1377 QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
1381 * Round a region to cluster boundaries
1383 static void round_to_clusters(BlockDriverState *bs,
1384 int64_t sector_num, int nb_sectors,
1385 int64_t *cluster_sector_num,
1386 int *cluster_nb_sectors)
1388 BlockDriverInfo bdi;
1390 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
1391 *cluster_sector_num = sector_num;
1392 *cluster_nb_sectors = nb_sectors;
1394 int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE;
1395 *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c);
1396 *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num +
1401 static bool tracked_request_overlaps(BdrvTrackedRequest *req,
1402 int64_t sector_num, int nb_sectors) {
1404 if (sector_num >= req->sector_num + req->nb_sectors) {
1408 if (req->sector_num >= sector_num + nb_sectors) {
1414 static void coroutine_fn wait_for_overlapping_requests(BlockDriverState *bs,
1415 int64_t sector_num, int nb_sectors)
1417 BdrvTrackedRequest *req;
1418 int64_t cluster_sector_num;
1419 int cluster_nb_sectors;
1422 /* If we touch the same cluster it counts as an overlap. This guarantees
1423 * that allocating writes will be serialized and not race with each other
1424 * for the same cluster. For example, in copy-on-read it ensures that the
1425 * CoR read and write operations are atomic and guest writes cannot
1426 * interleave between them.
1428 round_to_clusters(bs, sector_num, nb_sectors,
1429 &cluster_sector_num, &cluster_nb_sectors);
1433 QLIST_FOREACH(req, &bs->tracked_requests, list) {
1434 if (tracked_request_overlaps(req, cluster_sector_num,
1435 cluster_nb_sectors)) {
1436 /* Hitting this means there was a reentrant request, for
1437 * example, a block driver issuing nested requests. This must
1438 * never happen since it means deadlock.
1440 assert(qemu_coroutine_self() != req->co);
1442 qemu_co_queue_wait(&req->wait_queue);
1453 * -EINVAL - backing format specified, but no file
1454 * -ENOSPC - can't update the backing file because no space is left in the
1456 * -ENOTSUP - format driver doesn't support changing the backing file
1458 int bdrv_change_backing_file(BlockDriverState *bs,
1459 const char *backing_file, const char *backing_fmt)
1461 BlockDriver *drv = bs->drv;
1464 /* Backing file format doesn't make sense without a backing file */
1465 if (backing_fmt && !backing_file) {
1469 if (drv->bdrv_change_backing_file != NULL) {
1470 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
1476 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
1477 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
1482 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
1487 if (!bdrv_is_inserted(bs))
1493 len = bdrv_getlength(bs);
1498 if ((offset > len) || (len - offset < size))
1504 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
1507 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
1508 nb_sectors * BDRV_SECTOR_SIZE);
1511 typedef struct RwCo {
1512 BlockDriverState *bs;
1520 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
1522 RwCo *rwco = opaque;
1524 if (!rwco->is_write) {
1525 rwco->ret = bdrv_co_do_readv(rwco->bs, rwco->sector_num,
1526 rwco->nb_sectors, rwco->qiov, 0);
1528 rwco->ret = bdrv_co_do_writev(rwco->bs, rwco->sector_num,
1529 rwco->nb_sectors, rwco->qiov, 0);
1534 * Process a synchronous request using coroutines
1536 static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf,
1537 int nb_sectors, bool is_write)
1540 struct iovec iov = {
1541 .iov_base = (void *)buf,
1542 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
1547 .sector_num = sector_num,
1548 .nb_sectors = nb_sectors,
1550 .is_write = is_write,
1554 qemu_iovec_init_external(&qiov, &iov, 1);
1557 * In sync call context, when the vcpu is blocked, this throttling timer
1558 * will not fire; so the I/O throttling function has to be disabled here
1559 * if it has been enabled.
1561 if (bs->io_limits_enabled) {
1562 fprintf(stderr, "Disabling I/O throttling on '%s' due "
1563 "to synchronous I/O.\n", bdrv_get_device_name(bs));
1564 bdrv_io_limits_disable(bs);
1567 if (qemu_in_coroutine()) {
1568 /* Fast-path if already in coroutine context */
1569 bdrv_rw_co_entry(&rwco);
1571 co = qemu_coroutine_create(bdrv_rw_co_entry);
1572 qemu_coroutine_enter(co, &rwco);
1573 while (rwco.ret == NOT_DONE) {
1580 /* return < 0 if error. See bdrv_write() for the return codes */
1581 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
1582 uint8_t *buf, int nb_sectors)
1584 return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false);
1587 static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
1588 int nb_sectors, int dirty)
1591 unsigned long val, idx, bit;
1593 start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
1594 end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
1596 for (; start <= end; start++) {
1597 idx = start / (sizeof(unsigned long) * 8);
1598 bit = start % (sizeof(unsigned long) * 8);
1599 val = bs->dirty_bitmap[idx];
1601 if (!(val & (1UL << bit))) {
1606 if (val & (1UL << bit)) {
1608 val &= ~(1UL << bit);
1611 bs->dirty_bitmap[idx] = val;
1615 /* Return < 0 if error. Important errors are:
1616 -EIO generic I/O error (may happen for all errors)
1617 -ENOMEDIUM No media inserted.
1618 -EINVAL Invalid sector number or nb_sectors
1619 -EACCES Trying to write a read-only device
1621 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
1622 const uint8_t *buf, int nb_sectors)
1624 return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true);
1627 int bdrv_pread(BlockDriverState *bs, int64_t offset,
1628 void *buf, int count1)
1630 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1631 int len, nb_sectors, count;
1636 /* first read to align to sector start */
1637 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1640 sector_num = offset >> BDRV_SECTOR_BITS;
1642 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1644 memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
1652 /* read the sectors "in place" */
1653 nb_sectors = count >> BDRV_SECTOR_BITS;
1654 if (nb_sectors > 0) {
1655 if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
1657 sector_num += nb_sectors;
1658 len = nb_sectors << BDRV_SECTOR_BITS;
1663 /* add data from the last sector */
1665 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1667 memcpy(buf, tmp_buf, count);
1672 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
1673 const void *buf, int count1)
1675 uint8_t tmp_buf[BDRV_SECTOR_SIZE];
1676 int len, nb_sectors, count;
1681 /* first write to align to sector start */
1682 len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
1685 sector_num = offset >> BDRV_SECTOR_BITS;
1687 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1689 memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
1690 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1699 /* write the sectors "in place" */
1700 nb_sectors = count >> BDRV_SECTOR_BITS;
1701 if (nb_sectors > 0) {
1702 if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0)
1704 sector_num += nb_sectors;
1705 len = nb_sectors << BDRV_SECTOR_BITS;
1710 /* add data from the last sector */
1712 if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1714 memcpy(tmp_buf, buf, count);
1715 if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1722 * Writes to the file and ensures that no writes are reordered across this
1723 * request (acts as a barrier)
1725 * Returns 0 on success, -errno in error cases.
1727 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
1728 const void *buf, int count)
1732 ret = bdrv_pwrite(bs, offset, buf, count);
1737 /* No flush needed for cache modes that use O_DSYNC */
1738 if ((bs->open_flags & BDRV_O_CACHE_WB) != 0) {
1745 static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs,
1746 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1748 /* Perform I/O through a temporary buffer so that users who scribble over
1749 * their read buffer while the operation is in progress do not end up
1750 * modifying the image file. This is critical for zero-copy guest I/O
1751 * where anything might happen inside guest memory.
1753 void *bounce_buffer;
1755 BlockDriver *drv = bs->drv;
1757 QEMUIOVector bounce_qiov;
1758 int64_t cluster_sector_num;
1759 int cluster_nb_sectors;
1763 /* Cover entire cluster so no additional backing file I/O is required when
1764 * allocating cluster in the image file.
1766 round_to_clusters(bs, sector_num, nb_sectors,
1767 &cluster_sector_num, &cluster_nb_sectors);
1769 trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors,
1770 cluster_sector_num, cluster_nb_sectors);
1772 iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE;
1773 iov.iov_base = bounce_buffer = qemu_blockalign(bs, iov.iov_len);
1774 qemu_iovec_init_external(&bounce_qiov, &iov, 1);
1776 ret = drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors,
1782 if (drv->bdrv_co_write_zeroes &&
1783 buffer_is_zero(bounce_buffer, iov.iov_len)) {
1784 ret = bdrv_co_do_write_zeroes(bs, cluster_sector_num,
1785 cluster_nb_sectors);
1787 ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors,
1792 /* It might be okay to ignore write errors for guest requests. If this
1793 * is a deliberate copy-on-read then we don't want to ignore the error.
1794 * Simply report it in all cases.
1799 skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE;
1800 qemu_iovec_from_buffer(qiov, bounce_buffer + skip_bytes,
1801 nb_sectors * BDRV_SECTOR_SIZE);
1804 qemu_vfree(bounce_buffer);
1809 * Handle a read request in coroutine context
1811 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
1812 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1813 BdrvRequestFlags flags)
1815 BlockDriver *drv = bs->drv;
1816 BdrvTrackedRequest req;
1822 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
1826 /* throttling disk read I/O */
1827 if (bs->io_limits_enabled) {
1828 bdrv_io_limits_intercept(bs, false, nb_sectors);
1831 if (bs->copy_on_read) {
1832 flags |= BDRV_REQ_COPY_ON_READ;
1834 if (flags & BDRV_REQ_COPY_ON_READ) {
1835 bs->copy_on_read_in_flight++;
1838 if (bs->copy_on_read_in_flight) {
1839 wait_for_overlapping_requests(bs, sector_num, nb_sectors);
1842 tracked_request_begin(&req, bs, sector_num, nb_sectors, false);
1844 if (flags & BDRV_REQ_COPY_ON_READ) {
1847 ret = bdrv_co_is_allocated(bs, sector_num, nb_sectors, &pnum);
1852 if (!ret || pnum != nb_sectors) {
1853 ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov);
1858 ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
1861 tracked_request_end(&req);
1863 if (flags & BDRV_REQ_COPY_ON_READ) {
1864 bs->copy_on_read_in_flight--;
1870 int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
1871 int nb_sectors, QEMUIOVector *qiov)
1873 trace_bdrv_co_readv(bs, sector_num, nb_sectors);
1875 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0);
1878 int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
1879 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1881 trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors);
1883 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
1884 BDRV_REQ_COPY_ON_READ);
1887 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
1888 int64_t sector_num, int nb_sectors)
1890 BlockDriver *drv = bs->drv;
1895 /* TODO Emulate only part of misaligned requests instead of letting block
1896 * drivers return -ENOTSUP and emulate everything */
1898 /* First try the efficient write zeroes operation */
1899 if (drv->bdrv_co_write_zeroes) {
1900 ret = drv->bdrv_co_write_zeroes(bs, sector_num, nb_sectors);
1901 if (ret != -ENOTSUP) {
1906 /* Fall back to bounce buffer if write zeroes is unsupported */
1907 iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
1908 iov.iov_base = qemu_blockalign(bs, iov.iov_len);
1909 memset(iov.iov_base, 0, iov.iov_len);
1910 qemu_iovec_init_external(&qiov, &iov, 1);
1912 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, &qiov);
1914 qemu_vfree(iov.iov_base);
1919 * Handle a write request in coroutine context
1921 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
1922 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1923 BdrvRequestFlags flags)
1925 BlockDriver *drv = bs->drv;
1926 BdrvTrackedRequest req;
1932 if (bs->read_only) {
1935 if (bdrv_check_request(bs, sector_num, nb_sectors)) {
1939 /* throttling disk write I/O */
1940 if (bs->io_limits_enabled) {
1941 bdrv_io_limits_intercept(bs, true, nb_sectors);
1944 if (bs->copy_on_read_in_flight) {
1945 wait_for_overlapping_requests(bs, sector_num, nb_sectors);
1948 tracked_request_begin(&req, bs, sector_num, nb_sectors, true);
1950 if (flags & BDRV_REQ_ZERO_WRITE) {
1951 ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors);
1953 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
1956 if (bs->dirty_bitmap) {
1957 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1960 if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
1961 bs->wr_highest_sector = sector_num + nb_sectors - 1;
1964 tracked_request_end(&req);
1969 int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
1970 int nb_sectors, QEMUIOVector *qiov)
1972 trace_bdrv_co_writev(bs, sector_num, nb_sectors);
1974 return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0);
1977 int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs,
1978 int64_t sector_num, int nb_sectors)
1980 trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors);
1982 return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL,
1983 BDRV_REQ_ZERO_WRITE);
1987 * Truncate file to 'offset' bytes (needed only for file protocols)
1989 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
1991 BlockDriver *drv = bs->drv;
1995 if (!drv->bdrv_truncate)
1999 if (bdrv_in_use(bs))
2001 ret = drv->bdrv_truncate(bs, offset);
2003 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
2004 bdrv_dev_resize_cb(bs);
2010 * Length of a allocated file in bytes. Sparse files are counted by actual
2011 * allocated space. Return < 0 if error or unknown.
2013 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
2015 BlockDriver *drv = bs->drv;
2019 if (drv->bdrv_get_allocated_file_size) {
2020 return drv->bdrv_get_allocated_file_size(bs);
2023 return bdrv_get_allocated_file_size(bs->file);
2029 * Length of a file in bytes. Return < 0 if error or unknown.
2031 int64_t bdrv_getlength(BlockDriverState *bs)
2033 BlockDriver *drv = bs->drv;
2037 if (bs->growable || bdrv_dev_has_removable_media(bs)) {
2038 if (drv->bdrv_getlength) {
2039 return drv->bdrv_getlength(bs);
2042 return bs->total_sectors * BDRV_SECTOR_SIZE;
2045 /* return 0 as number of sectors if no device present or error */
2046 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
2049 length = bdrv_getlength(bs);
2053 length = length >> BDRV_SECTOR_BITS;
2054 *nb_sectors_ptr = length;
2058 uint8_t boot_ind; /* 0x80 - active */
2059 uint8_t head; /* starting head */
2060 uint8_t sector; /* starting sector */
2061 uint8_t cyl; /* starting cylinder */
2062 uint8_t sys_ind; /* What partition type */
2063 uint8_t end_head; /* end head */
2064 uint8_t end_sector; /* end sector */
2065 uint8_t end_cyl; /* end cylinder */
2066 uint32_t start_sect; /* starting sector counting from 0 */
2067 uint32_t nr_sects; /* nr of sectors in partition */
2070 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
2071 static int guess_disk_lchs(BlockDriverState *bs,
2072 int *pcylinders, int *pheads, int *psectors)
2074 uint8_t buf[BDRV_SECTOR_SIZE];
2075 int ret, i, heads, sectors, cylinders;
2076 struct partition *p;
2078 uint64_t nb_sectors;
2081 bdrv_get_geometry(bs, &nb_sectors);
2084 * The function will be invoked during startup not only in sync I/O mode,
2085 * but also in async I/O mode. So the I/O throttling function has to
2086 * be disabled temporarily here, not permanently.
2088 enabled = bs->io_limits_enabled;
2089 bs->io_limits_enabled = false;
2090 ret = bdrv_read(bs, 0, buf, 1);
2091 bs->io_limits_enabled = enabled;
2094 /* test msdos magic */
2095 if (buf[510] != 0x55 || buf[511] != 0xaa)
2097 for(i = 0; i < 4; i++) {
2098 p = ((struct partition *)(buf + 0x1be)) + i;
2099 nr_sects = le32_to_cpu(p->nr_sects);
2100 if (nr_sects && p->end_head) {
2101 /* We make the assumption that the partition terminates on
2102 a cylinder boundary */
2103 heads = p->end_head + 1;
2104 sectors = p->end_sector & 63;
2107 cylinders = nb_sectors / (heads * sectors);
2108 if (cylinders < 1 || cylinders > 16383)
2111 *psectors = sectors;
2112 *pcylinders = cylinders;
2114 printf("guessed geometry: LCHS=%d %d %d\n",
2115 cylinders, heads, sectors);
2123 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
2125 int translation, lba_detected = 0;
2126 int cylinders, heads, secs;
2127 uint64_t nb_sectors;
2129 /* if a geometry hint is available, use it */
2130 bdrv_get_geometry(bs, &nb_sectors);
2131 bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
2132 translation = bdrv_get_translation_hint(bs);
2133 if (cylinders != 0) {
2138 if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
2140 /* if heads > 16, it means that a BIOS LBA
2141 translation was active, so the default
2142 hardware geometry is OK */
2144 goto default_geometry;
2149 /* disable any translation to be in sync with
2150 the logical geometry */
2151 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
2152 bdrv_set_translation_hint(bs,
2153 BIOS_ATA_TRANSLATION_NONE);
2158 /* if no geometry, use a standard physical disk geometry */
2159 cylinders = nb_sectors / (16 * 63);
2161 if (cylinders > 16383)
2163 else if (cylinders < 2)
2168 if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
2169 if ((*pcyls * *pheads) <= 131072) {
2170 bdrv_set_translation_hint(bs,
2171 BIOS_ATA_TRANSLATION_LARGE);
2173 bdrv_set_translation_hint(bs,
2174 BIOS_ATA_TRANSLATION_LBA);
2178 bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
2182 void bdrv_set_geometry_hint(BlockDriverState *bs,
2183 int cyls, int heads, int secs)
2190 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
2192 bs->translation = translation;
2195 void bdrv_get_geometry_hint(BlockDriverState *bs,
2196 int *pcyls, int *pheads, int *psecs)
2199 *pheads = bs->heads;
2203 /* throttling disk io limits */
2204 void bdrv_set_io_limits(BlockDriverState *bs,
2205 BlockIOLimit *io_limits)
2207 bs->io_limits = *io_limits;
2208 bs->io_limits_enabled = bdrv_io_limits_enabled(bs);
2211 /* Recognize floppy formats */
2212 typedef struct FDFormat {
2220 static const FDFormat fd_formats[] = {
2221 /* First entry is default format */
2222 /* 1.44 MB 3"1/2 floppy disks */
2223 { FDRIVE_DRV_144, 18, 80, 1, FDRIVE_RATE_500K, },
2224 { FDRIVE_DRV_144, 20, 80, 1, FDRIVE_RATE_500K, },
2225 { FDRIVE_DRV_144, 21, 80, 1, FDRIVE_RATE_500K, },
2226 { FDRIVE_DRV_144, 21, 82, 1, FDRIVE_RATE_500K, },
2227 { FDRIVE_DRV_144, 21, 83, 1, FDRIVE_RATE_500K, },
2228 { FDRIVE_DRV_144, 22, 80, 1, FDRIVE_RATE_500K, },
2229 { FDRIVE_DRV_144, 23, 80, 1, FDRIVE_RATE_500K, },
2230 { FDRIVE_DRV_144, 24, 80, 1, FDRIVE_RATE_500K, },
2231 /* 2.88 MB 3"1/2 floppy disks */
2232 { FDRIVE_DRV_288, 36, 80, 1, FDRIVE_RATE_1M, },
2233 { FDRIVE_DRV_288, 39, 80, 1, FDRIVE_RATE_1M, },
2234 { FDRIVE_DRV_288, 40, 80, 1, FDRIVE_RATE_1M, },
2235 { FDRIVE_DRV_288, 44, 80, 1, FDRIVE_RATE_1M, },
2236 { FDRIVE_DRV_288, 48, 80, 1, FDRIVE_RATE_1M, },
2237 /* 720 kB 3"1/2 floppy disks */
2238 { FDRIVE_DRV_144, 9, 80, 1, FDRIVE_RATE_250K, },
2239 { FDRIVE_DRV_144, 10, 80, 1, FDRIVE_RATE_250K, },
2240 { FDRIVE_DRV_144, 10, 82, 1, FDRIVE_RATE_250K, },
2241 { FDRIVE_DRV_144, 10, 83, 1, FDRIVE_RATE_250K, },
2242 { FDRIVE_DRV_144, 13, 80, 1, FDRIVE_RATE_250K, },
2243 { FDRIVE_DRV_144, 14, 80, 1, FDRIVE_RATE_250K, },
2244 /* 1.2 MB 5"1/4 floppy disks */
2245 { FDRIVE_DRV_120, 15, 80, 1, FDRIVE_RATE_500K, },
2246 { FDRIVE_DRV_120, 18, 80, 1, FDRIVE_RATE_500K, },
2247 { FDRIVE_DRV_120, 18, 82, 1, FDRIVE_RATE_500K, },
2248 { FDRIVE_DRV_120, 18, 83, 1, FDRIVE_RATE_500K, },
2249 { FDRIVE_DRV_120, 20, 80, 1, FDRIVE_RATE_500K, },
2250 /* 720 kB 5"1/4 floppy disks */
2251 { FDRIVE_DRV_120, 9, 80, 1, FDRIVE_RATE_250K, },
2252 { FDRIVE_DRV_120, 11, 80, 1, FDRIVE_RATE_250K, },
2253 /* 360 kB 5"1/4 floppy disks */
2254 { FDRIVE_DRV_120, 9, 40, 1, FDRIVE_RATE_300K, },
2255 { FDRIVE_DRV_120, 9, 40, 0, FDRIVE_RATE_300K, },
2256 { FDRIVE_DRV_120, 10, 41, 1, FDRIVE_RATE_300K, },
2257 { FDRIVE_DRV_120, 10, 42, 1, FDRIVE_RATE_300K, },
2258 /* 320 kB 5"1/4 floppy disks */
2259 { FDRIVE_DRV_120, 8, 40, 1, FDRIVE_RATE_250K, },
2260 { FDRIVE_DRV_120, 8, 40, 0, FDRIVE_RATE_250K, },
2261 /* 360 kB must match 5"1/4 better than 3"1/2... */
2262 { FDRIVE_DRV_144, 9, 80, 0, FDRIVE_RATE_250K, },
2264 { FDRIVE_DRV_NONE, -1, -1, 0, 0, },
2267 void bdrv_get_floppy_geometry_hint(BlockDriverState *bs, int *nb_heads,
2268 int *max_track, int *last_sect,
2269 FDriveType drive_in, FDriveType *drive,
2272 const FDFormat *parse;
2273 uint64_t nb_sectors, size;
2274 int i, first_match, match;
2276 bdrv_get_geometry_hint(bs, nb_heads, max_track, last_sect);
2277 if (*nb_heads != 0 && *max_track != 0 && *last_sect != 0) {
2278 /* User defined disk */
2279 *rate = FDRIVE_RATE_500K;
2281 bdrv_get_geometry(bs, &nb_sectors);
2284 for (i = 0; ; i++) {
2285 parse = &fd_formats[i];
2286 if (parse->drive == FDRIVE_DRV_NONE) {
2289 if (drive_in == parse->drive ||
2290 drive_in == FDRIVE_DRV_NONE) {
2291 size = (parse->max_head + 1) * parse->max_track *
2293 if (nb_sectors == size) {
2297 if (first_match == -1) {
2303 if (first_match == -1) {
2306 match = first_match;
2308 parse = &fd_formats[match];
2310 *nb_heads = parse->max_head + 1;
2311 *max_track = parse->max_track;
2312 *last_sect = parse->last_sect;
2313 *drive = parse->drive;
2314 *rate = parse->rate;
2318 int bdrv_get_translation_hint(BlockDriverState *bs)
2320 return bs->translation;
2323 void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
2324 BlockErrorAction on_write_error)
2326 bs->on_read_error = on_read_error;
2327 bs->on_write_error = on_write_error;
2330 BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
2332 return is_read ? bs->on_read_error : bs->on_write_error;
2335 int bdrv_is_read_only(BlockDriverState *bs)
2337 return bs->read_only;
2340 int bdrv_is_sg(BlockDriverState *bs)
2345 int bdrv_enable_write_cache(BlockDriverState *bs)
2347 return bs->enable_write_cache;
2350 int bdrv_is_encrypted(BlockDriverState *bs)
2352 if (bs->backing_hd && bs->backing_hd->encrypted)
2354 return bs->encrypted;
2357 int bdrv_key_required(BlockDriverState *bs)
2359 BlockDriverState *backing_hd = bs->backing_hd;
2361 if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
2363 return (bs->encrypted && !bs->valid_key);
2366 int bdrv_set_key(BlockDriverState *bs, const char *key)
2369 if (bs->backing_hd && bs->backing_hd->encrypted) {
2370 ret = bdrv_set_key(bs->backing_hd, key);
2376 if (!bs->encrypted) {
2378 } else if (!bs->drv || !bs->drv->bdrv_set_key) {
2381 ret = bs->drv->bdrv_set_key(bs, key);
2384 } else if (!bs->valid_key) {
2386 /* call the change callback now, we skipped it on open */
2387 bdrv_dev_change_media_cb(bs, true);
2392 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
2397 pstrcpy(buf, buf_size, bs->drv->format_name);
2401 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
2406 QLIST_FOREACH(drv, &bdrv_drivers, list) {
2407 it(opaque, drv->format_name);
2411 BlockDriverState *bdrv_find(const char *name)
2413 BlockDriverState *bs;
2415 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2416 if (!strcmp(name, bs->device_name)) {
2423 BlockDriverState *bdrv_next(BlockDriverState *bs)
2426 return QTAILQ_FIRST(&bdrv_states);
2428 return QTAILQ_NEXT(bs, list);
2431 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
2433 BlockDriverState *bs;
2435 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2440 const char *bdrv_get_device_name(BlockDriverState *bs)
2442 return bs->device_name;
2445 void bdrv_flush_all(void)
2447 BlockDriverState *bs;
2449 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2454 int bdrv_has_zero_init(BlockDriverState *bs)
2458 if (bs->drv->bdrv_has_zero_init) {
2459 return bs->drv->bdrv_has_zero_init(bs);
2465 typedef struct BdrvCoIsAllocatedData {
2466 BlockDriverState *bs;
2472 } BdrvCoIsAllocatedData;
2475 * Returns true iff the specified sector is present in the disk image. Drivers
2476 * not implementing the functionality are assumed to not support backing files,
2477 * hence all their sectors are reported as allocated.
2479 * If 'sector_num' is beyond the end of the disk image the return value is 0
2480 * and 'pnum' is set to 0.
2482 * 'pnum' is set to the number of sectors (including and immediately following
2483 * the specified sector) that are known to be in the same
2484 * allocated/unallocated state.
2486 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
2487 * beyond the end of the disk image it will be clamped.
2489 int coroutine_fn bdrv_co_is_allocated(BlockDriverState *bs, int64_t sector_num,
2490 int nb_sectors, int *pnum)
2494 if (sector_num >= bs->total_sectors) {
2499 n = bs->total_sectors - sector_num;
2500 if (n < nb_sectors) {
2504 if (!bs->drv->bdrv_co_is_allocated) {
2509 return bs->drv->bdrv_co_is_allocated(bs, sector_num, nb_sectors, pnum);
2512 /* Coroutine wrapper for bdrv_is_allocated() */
2513 static void coroutine_fn bdrv_is_allocated_co_entry(void *opaque)
2515 BdrvCoIsAllocatedData *data = opaque;
2516 BlockDriverState *bs = data->bs;
2518 data->ret = bdrv_co_is_allocated(bs, data->sector_num, data->nb_sectors,
2524 * Synchronous wrapper around bdrv_co_is_allocated().
2526 * See bdrv_co_is_allocated() for details.
2528 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
2532 BdrvCoIsAllocatedData data = {
2534 .sector_num = sector_num,
2535 .nb_sectors = nb_sectors,
2540 co = qemu_coroutine_create(bdrv_is_allocated_co_entry);
2541 qemu_coroutine_enter(co, &data);
2542 while (!data.done) {
2548 BlockInfoList *qmp_query_block(Error **errp)
2550 BlockInfoList *head = NULL, *cur_item = NULL;
2551 BlockDriverState *bs;
2553 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2554 BlockInfoList *info = g_malloc0(sizeof(*info));
2556 info->value = g_malloc0(sizeof(*info->value));
2557 info->value->device = g_strdup(bs->device_name);
2558 info->value->type = g_strdup("unknown");
2559 info->value->locked = bdrv_dev_is_medium_locked(bs);
2560 info->value->removable = bdrv_dev_has_removable_media(bs);
2562 if (bdrv_dev_has_removable_media(bs)) {
2563 info->value->has_tray_open = true;
2564 info->value->tray_open = bdrv_dev_is_tray_open(bs);
2567 if (bdrv_iostatus_is_enabled(bs)) {
2568 info->value->has_io_status = true;
2569 info->value->io_status = bs->iostatus;
2573 info->value->has_inserted = true;
2574 info->value->inserted = g_malloc0(sizeof(*info->value->inserted));
2575 info->value->inserted->file = g_strdup(bs->filename);
2576 info->value->inserted->ro = bs->read_only;
2577 info->value->inserted->drv = g_strdup(bs->drv->format_name);
2578 info->value->inserted->encrypted = bs->encrypted;
2579 if (bs->backing_file[0]) {
2580 info->value->inserted->has_backing_file = true;
2581 info->value->inserted->backing_file = g_strdup(bs->backing_file);
2584 if (bs->io_limits_enabled) {
2585 info->value->inserted->bps =
2586 bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL];
2587 info->value->inserted->bps_rd =
2588 bs->io_limits.bps[BLOCK_IO_LIMIT_READ];
2589 info->value->inserted->bps_wr =
2590 bs->io_limits.bps[BLOCK_IO_LIMIT_WRITE];
2591 info->value->inserted->iops =
2592 bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL];
2593 info->value->inserted->iops_rd =
2594 bs->io_limits.iops[BLOCK_IO_LIMIT_READ];
2595 info->value->inserted->iops_wr =
2596 bs->io_limits.iops[BLOCK_IO_LIMIT_WRITE];
2600 /* XXX: waiting for the qapi to support GSList */
2602 head = cur_item = info;
2604 cur_item->next = info;
2612 /* Consider exposing this as a full fledged QMP command */
2613 static BlockStats *qmp_query_blockstat(const BlockDriverState *bs, Error **errp)
2617 s = g_malloc0(sizeof(*s));
2619 if (bs->device_name[0]) {
2620 s->has_device = true;
2621 s->device = g_strdup(bs->device_name);
2624 s->stats = g_malloc0(sizeof(*s->stats));
2625 s->stats->rd_bytes = bs->nr_bytes[BDRV_ACCT_READ];
2626 s->stats->wr_bytes = bs->nr_bytes[BDRV_ACCT_WRITE];
2627 s->stats->rd_operations = bs->nr_ops[BDRV_ACCT_READ];
2628 s->stats->wr_operations = bs->nr_ops[BDRV_ACCT_WRITE];
2629 s->stats->wr_highest_offset = bs->wr_highest_sector * BDRV_SECTOR_SIZE;
2630 s->stats->flush_operations = bs->nr_ops[BDRV_ACCT_FLUSH];
2631 s->stats->wr_total_time_ns = bs->total_time_ns[BDRV_ACCT_WRITE];
2632 s->stats->rd_total_time_ns = bs->total_time_ns[BDRV_ACCT_READ];
2633 s->stats->flush_total_time_ns = bs->total_time_ns[BDRV_ACCT_FLUSH];
2636 s->has_parent = true;
2637 s->parent = qmp_query_blockstat(bs->file, NULL);
2643 BlockStatsList *qmp_query_blockstats(Error **errp)
2645 BlockStatsList *head = NULL, *cur_item = NULL;
2646 BlockDriverState *bs;
2648 QTAILQ_FOREACH(bs, &bdrv_states, list) {
2649 BlockStatsList *info = g_malloc0(sizeof(*info));
2650 info->value = qmp_query_blockstat(bs, NULL);
2652 /* XXX: waiting for the qapi to support GSList */
2654 head = cur_item = info;
2656 cur_item->next = info;
2664 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
2666 if (bs->backing_hd && bs->backing_hd->encrypted)
2667 return bs->backing_file;
2668 else if (bs->encrypted)
2669 return bs->filename;
2674 void bdrv_get_backing_filename(BlockDriverState *bs,
2675 char *filename, int filename_size)
2677 pstrcpy(filename, filename_size, bs->backing_file);
2680 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
2681 const uint8_t *buf, int nb_sectors)
2683 BlockDriver *drv = bs->drv;
2686 if (!drv->bdrv_write_compressed)
2688 if (bdrv_check_request(bs, sector_num, nb_sectors))
2691 if (bs->dirty_bitmap) {
2692 set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
2695 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
2698 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2700 BlockDriver *drv = bs->drv;
2703 if (!drv->bdrv_get_info)
2705 memset(bdi, 0, sizeof(*bdi));
2706 return drv->bdrv_get_info(bs, bdi);
2709 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
2710 int64_t pos, int size)
2712 BlockDriver *drv = bs->drv;
2715 if (drv->bdrv_save_vmstate)
2716 return drv->bdrv_save_vmstate(bs, buf, pos, size);
2718 return bdrv_save_vmstate(bs->file, buf, pos, size);
2722 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
2723 int64_t pos, int size)
2725 BlockDriver *drv = bs->drv;
2728 if (drv->bdrv_load_vmstate)
2729 return drv->bdrv_load_vmstate(bs, buf, pos, size);
2731 return bdrv_load_vmstate(bs->file, buf, pos, size);
2735 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
2737 BlockDriver *drv = bs->drv;
2739 if (!drv || !drv->bdrv_debug_event) {
2743 return drv->bdrv_debug_event(bs, event);
2747 /**************************************************************/
2748 /* handling of snapshots */
2750 int bdrv_can_snapshot(BlockDriverState *bs)
2752 BlockDriver *drv = bs->drv;
2753 if (!drv || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
2757 if (!drv->bdrv_snapshot_create) {
2758 if (bs->file != NULL) {
2759 return bdrv_can_snapshot(bs->file);
2767 int bdrv_is_snapshot(BlockDriverState *bs)
2769 return !!(bs->open_flags & BDRV_O_SNAPSHOT);
2772 BlockDriverState *bdrv_snapshots(void)
2774 BlockDriverState *bs;
2777 return bs_snapshots;
2781 while ((bs = bdrv_next(bs))) {
2782 if (bdrv_can_snapshot(bs)) {
2790 int bdrv_snapshot_create(BlockDriverState *bs,
2791 QEMUSnapshotInfo *sn_info)
2793 BlockDriver *drv = bs->drv;
2796 if (drv->bdrv_snapshot_create)
2797 return drv->bdrv_snapshot_create(bs, sn_info);
2799 return bdrv_snapshot_create(bs->file, sn_info);
2803 int bdrv_snapshot_goto(BlockDriverState *bs,
2804 const char *snapshot_id)
2806 BlockDriver *drv = bs->drv;
2811 if (drv->bdrv_snapshot_goto)
2812 return drv->bdrv_snapshot_goto(bs, snapshot_id);
2815 drv->bdrv_close(bs);
2816 ret = bdrv_snapshot_goto(bs->file, snapshot_id);
2817 open_ret = drv->bdrv_open(bs, bs->open_flags);
2819 bdrv_delete(bs->file);
2829 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2831 BlockDriver *drv = bs->drv;
2834 if (drv->bdrv_snapshot_delete)
2835 return drv->bdrv_snapshot_delete(bs, snapshot_id);
2837 return bdrv_snapshot_delete(bs->file, snapshot_id);
2841 int bdrv_snapshot_list(BlockDriverState *bs,
2842 QEMUSnapshotInfo **psn_info)
2844 BlockDriver *drv = bs->drv;
2847 if (drv->bdrv_snapshot_list)
2848 return drv->bdrv_snapshot_list(bs, psn_info);
2850 return bdrv_snapshot_list(bs->file, psn_info);
2854 int bdrv_snapshot_load_tmp(BlockDriverState *bs,
2855 const char *snapshot_name)
2857 BlockDriver *drv = bs->drv;
2861 if (!bs->read_only) {
2864 if (drv->bdrv_snapshot_load_tmp) {
2865 return drv->bdrv_snapshot_load_tmp(bs, snapshot_name);
2870 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
2871 const char *backing_file)
2877 if (bs->backing_hd) {
2878 if (strcmp(bs->backing_file, backing_file) == 0) {
2879 return bs->backing_hd;
2881 return bdrv_find_backing_image(bs->backing_hd, backing_file);
2888 #define NB_SUFFIXES 4
2890 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
2892 static const char suffixes[NB_SUFFIXES] = "KMGT";
2897 snprintf(buf, buf_size, "%" PRId64, size);
2900 for(i = 0; i < NB_SUFFIXES; i++) {
2901 if (size < (10 * base)) {
2902 snprintf(buf, buf_size, "%0.1f%c",
2903 (double)size / base,
2906 } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
2907 snprintf(buf, buf_size, "%" PRId64 "%c",
2908 ((size + (base >> 1)) / base),
2918 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
2920 char buf1[128], date_buf[128], clock_buf[128];
2930 snprintf(buf, buf_size,
2931 "%-10s%-20s%7s%20s%15s",
2932 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
2936 ptm = localtime(&ti);
2937 strftime(date_buf, sizeof(date_buf),
2938 "%Y-%m-%d %H:%M:%S", ptm);
2940 localtime_r(&ti, &tm);
2941 strftime(date_buf, sizeof(date_buf),
2942 "%Y-%m-%d %H:%M:%S", &tm);
2944 secs = sn->vm_clock_nsec / 1000000000;
2945 snprintf(clock_buf, sizeof(clock_buf),
2946 "%02d:%02d:%02d.%03d",
2948 (int)((secs / 60) % 60),
2950 (int)((sn->vm_clock_nsec / 1000000) % 1000));
2951 snprintf(buf, buf_size,
2952 "%-10s%-20s%7s%20s%15s",
2953 sn->id_str, sn->name,
2954 get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
2961 /**************************************************************/
2964 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
2965 QEMUIOVector *qiov, int nb_sectors,
2966 BlockDriverCompletionFunc *cb, void *opaque)
2968 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
2970 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
2974 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
2975 QEMUIOVector *qiov, int nb_sectors,
2976 BlockDriverCompletionFunc *cb, void *opaque)
2978 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
2980 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
2985 typedef struct MultiwriteCB {
2990 BlockDriverCompletionFunc *cb;
2992 QEMUIOVector *free_qiov;
2996 static void multiwrite_user_cb(MultiwriteCB *mcb)
3000 for (i = 0; i < mcb->num_callbacks; i++) {
3001 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
3002 if (mcb->callbacks[i].free_qiov) {
3003 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
3005 g_free(mcb->callbacks[i].free_qiov);
3009 static void multiwrite_cb(void *opaque, int ret)
3011 MultiwriteCB *mcb = opaque;
3013 trace_multiwrite_cb(mcb, ret);
3015 if (ret < 0 && !mcb->error) {
3019 mcb->num_requests--;
3020 if (mcb->num_requests == 0) {
3021 multiwrite_user_cb(mcb);
3026 static int multiwrite_req_compare(const void *a, const void *b)
3028 const BlockRequest *req1 = a, *req2 = b;
3031 * Note that we can't simply subtract req2->sector from req1->sector
3032 * here as that could overflow the return value.
3034 if (req1->sector > req2->sector) {
3036 } else if (req1->sector < req2->sector) {
3044 * Takes a bunch of requests and tries to merge them. Returns the number of
3045 * requests that remain after merging.
3047 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
3048 int num_reqs, MultiwriteCB *mcb)
3052 // Sort requests by start sector
3053 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
3055 // Check if adjacent requests touch the same clusters. If so, combine them,
3056 // filling up gaps with zero sectors.
3058 for (i = 1; i < num_reqs; i++) {
3060 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
3062 // Handle exactly sequential writes and overlapping writes.
3063 if (reqs[i].sector <= oldreq_last) {
3067 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
3073 QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
3074 qemu_iovec_init(qiov,
3075 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
3077 // Add the first request to the merged one. If the requests are
3078 // overlapping, drop the last sectors of the first request.
3079 size = (reqs[i].sector - reqs[outidx].sector) << 9;
3080 qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
3082 // We should need to add any zeros between the two requests
3083 assert (reqs[i].sector <= oldreq_last);
3085 // Add the second request
3086 qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
3088 reqs[outidx].nb_sectors = qiov->size >> 9;
3089 reqs[outidx].qiov = qiov;
3091 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
3094 reqs[outidx].sector = reqs[i].sector;
3095 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
3096 reqs[outidx].qiov = reqs[i].qiov;
3104 * Submit multiple AIO write requests at once.
3106 * On success, the function returns 0 and all requests in the reqs array have
3107 * been submitted. In error case this function returns -1, and any of the
3108 * requests may or may not be submitted yet. In particular, this means that the
3109 * callback will be called for some of the requests, for others it won't. The
3110 * caller must check the error field of the BlockRequest to wait for the right
3111 * callbacks (if error != 0, no callback will be called).
3113 * The implementation may modify the contents of the reqs array, e.g. to merge
3114 * requests. However, the fields opaque and error are left unmodified as they
3115 * are used to signal failure for a single request to the caller.
3117 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
3122 /* don't submit writes if we don't have a medium */
3123 if (bs->drv == NULL) {
3124 for (i = 0; i < num_reqs; i++) {
3125 reqs[i].error = -ENOMEDIUM;
3130 if (num_reqs == 0) {
3134 // Create MultiwriteCB structure
3135 mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
3136 mcb->num_requests = 0;
3137 mcb->num_callbacks = num_reqs;
3139 for (i = 0; i < num_reqs; i++) {
3140 mcb->callbacks[i].cb = reqs[i].cb;
3141 mcb->callbacks[i].opaque = reqs[i].opaque;
3144 // Check for mergable requests
3145 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
3147 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
3149 /* Run the aio requests. */
3150 mcb->num_requests = num_reqs;
3151 for (i = 0; i < num_reqs; i++) {
3152 bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
3153 reqs[i].nb_sectors, multiwrite_cb, mcb);
3159 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
3161 acb->pool->cancel(acb);
3164 /* block I/O throttling */
3165 static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors,
3166 bool is_write, double elapsed_time, uint64_t *wait)
3168 uint64_t bps_limit = 0;
3169 double bytes_limit, bytes_base, bytes_res;
3170 double slice_time, wait_time;
3172 if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
3173 bps_limit = bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL];
3174 } else if (bs->io_limits.bps[is_write]) {
3175 bps_limit = bs->io_limits.bps[is_write];
3184 slice_time = bs->slice_end - bs->slice_start;
3185 slice_time /= (NANOSECONDS_PER_SECOND);
3186 bytes_limit = bps_limit * slice_time;
3187 bytes_base = bs->nr_bytes[is_write] - bs->io_base.bytes[is_write];
3188 if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
3189 bytes_base += bs->nr_bytes[!is_write] - bs->io_base.bytes[!is_write];
3192 /* bytes_base: the bytes of data which have been read/written; and
3193 * it is obtained from the history statistic info.
3194 * bytes_res: the remaining bytes of data which need to be read/written.
3195 * (bytes_base + bytes_res) / bps_limit: used to calcuate
3196 * the total time for completing reading/writting all data.
3198 bytes_res = (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
3200 if (bytes_base + bytes_res <= bytes_limit) {
3208 /* Calc approx time to dispatch */
3209 wait_time = (bytes_base + bytes_res) / bps_limit - elapsed_time;
3211 /* When the I/O rate at runtime exceeds the limits,
3212 * bs->slice_end need to be extended in order that the current statistic
3213 * info can be kept until the timer fire, so it is increased and tuned
3214 * based on the result of experiment.
3216 bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10;
3217 bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME;
3219 *wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
3225 static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write,
3226 double elapsed_time, uint64_t *wait)
3228 uint64_t iops_limit = 0;
3229 double ios_limit, ios_base;
3230 double slice_time, wait_time;
3232 if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
3233 iops_limit = bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL];
3234 } else if (bs->io_limits.iops[is_write]) {
3235 iops_limit = bs->io_limits.iops[is_write];
3244 slice_time = bs->slice_end - bs->slice_start;
3245 slice_time /= (NANOSECONDS_PER_SECOND);
3246 ios_limit = iops_limit * slice_time;
3247 ios_base = bs->nr_ops[is_write] - bs->io_base.ios[is_write];
3248 if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
3249 ios_base += bs->nr_ops[!is_write] - bs->io_base.ios[!is_write];
3252 if (ios_base + 1 <= ios_limit) {
3260 /* Calc approx time to dispatch */
3261 wait_time = (ios_base + 1) / iops_limit;
3262 if (wait_time > elapsed_time) {
3263 wait_time = wait_time - elapsed_time;
3268 bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10;
3269 bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME;
3271 *wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
3277 static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors,
3278 bool is_write, int64_t *wait)
3280 int64_t now, max_wait;
3281 uint64_t bps_wait = 0, iops_wait = 0;
3282 double elapsed_time;
3283 int bps_ret, iops_ret;
3285 now = qemu_get_clock_ns(vm_clock);
3286 if ((bs->slice_start < now)
3287 && (bs->slice_end > now)) {
3288 bs->slice_end = now + bs->slice_time;
3290 bs->slice_time = 5 * BLOCK_IO_SLICE_TIME;
3291 bs->slice_start = now;
3292 bs->slice_end = now + bs->slice_time;
3294 bs->io_base.bytes[is_write] = bs->nr_bytes[is_write];
3295 bs->io_base.bytes[!is_write] = bs->nr_bytes[!is_write];
3297 bs->io_base.ios[is_write] = bs->nr_ops[is_write];
3298 bs->io_base.ios[!is_write] = bs->nr_ops[!is_write];
3301 elapsed_time = now - bs->slice_start;
3302 elapsed_time /= (NANOSECONDS_PER_SECOND);
3304 bps_ret = bdrv_exceed_bps_limits(bs, nb_sectors,
3305 is_write, elapsed_time, &bps_wait);
3306 iops_ret = bdrv_exceed_iops_limits(bs, is_write,
3307 elapsed_time, &iops_wait);
3308 if (bps_ret || iops_ret) {
3309 max_wait = bps_wait > iops_wait ? bps_wait : iops_wait;
3314 now = qemu_get_clock_ns(vm_clock);
3315 if (bs->slice_end < now + max_wait) {
3316 bs->slice_end = now + max_wait;
3329 /**************************************************************/
3330 /* async block device emulation */
3332 typedef struct BlockDriverAIOCBSync {
3333 BlockDriverAIOCB common;
3336 /* vector translation state */
3340 } BlockDriverAIOCBSync;
3342 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
3344 BlockDriverAIOCBSync *acb =
3345 container_of(blockacb, BlockDriverAIOCBSync, common);
3346 qemu_bh_delete(acb->bh);
3348 qemu_aio_release(acb);
3351 static AIOPool bdrv_em_aio_pool = {
3352 .aiocb_size = sizeof(BlockDriverAIOCBSync),
3353 .cancel = bdrv_aio_cancel_em,
3356 static void bdrv_aio_bh_cb(void *opaque)
3358 BlockDriverAIOCBSync *acb = opaque;
3361 qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
3362 qemu_vfree(acb->bounce);
3363 acb->common.cb(acb->common.opaque, acb->ret);
3364 qemu_bh_delete(acb->bh);
3366 qemu_aio_release(acb);
3369 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
3373 BlockDriverCompletionFunc *cb,
3378 BlockDriverAIOCBSync *acb;
3380 acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
3381 acb->is_write = is_write;
3383 acb->bounce = qemu_blockalign(bs, qiov->size);
3384 acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
3387 qemu_iovec_to_buffer(acb->qiov, acb->bounce);
3388 acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
3390 acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
3393 qemu_bh_schedule(acb->bh);
3395 return &acb->common;
3398 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
3399 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
3400 BlockDriverCompletionFunc *cb, void *opaque)
3402 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
3405 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
3406 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
3407 BlockDriverCompletionFunc *cb, void *opaque)
3409 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
3413 typedef struct BlockDriverAIOCBCoroutine {
3414 BlockDriverAIOCB common;
3418 } BlockDriverAIOCBCoroutine;
3420 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb)
3425 static AIOPool bdrv_em_co_aio_pool = {
3426 .aiocb_size = sizeof(BlockDriverAIOCBCoroutine),
3427 .cancel = bdrv_aio_co_cancel_em,
3430 static void bdrv_co_em_bh(void *opaque)
3432 BlockDriverAIOCBCoroutine *acb = opaque;
3434 acb->common.cb(acb->common.opaque, acb->req.error);
3435 qemu_bh_delete(acb->bh);
3436 qemu_aio_release(acb);
3439 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
3440 static void coroutine_fn bdrv_co_do_rw(void *opaque)
3442 BlockDriverAIOCBCoroutine *acb = opaque;
3443 BlockDriverState *bs = acb->common.bs;
3445 if (!acb->is_write) {
3446 acb->req.error = bdrv_co_do_readv(bs, acb->req.sector,
3447 acb->req.nb_sectors, acb->req.qiov, 0);
3449 acb->req.error = bdrv_co_do_writev(bs, acb->req.sector,
3450 acb->req.nb_sectors, acb->req.qiov, 0);
3453 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3454 qemu_bh_schedule(acb->bh);
3457 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
3461 BlockDriverCompletionFunc *cb,
3466 BlockDriverAIOCBCoroutine *acb;
3468 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3469 acb->req.sector = sector_num;
3470 acb->req.nb_sectors = nb_sectors;
3471 acb->req.qiov = qiov;
3472 acb->is_write = is_write;
3474 co = qemu_coroutine_create(bdrv_co_do_rw);
3475 qemu_coroutine_enter(co, acb);
3477 return &acb->common;
3480 static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
3482 BlockDriverAIOCBCoroutine *acb = opaque;
3483 BlockDriverState *bs = acb->common.bs;
3485 acb->req.error = bdrv_co_flush(bs);
3486 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3487 qemu_bh_schedule(acb->bh);
3490 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
3491 BlockDriverCompletionFunc *cb, void *opaque)
3493 trace_bdrv_aio_flush(bs, opaque);
3496 BlockDriverAIOCBCoroutine *acb;
3498 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3499 co = qemu_coroutine_create(bdrv_aio_flush_co_entry);
3500 qemu_coroutine_enter(co, acb);
3502 return &acb->common;
3505 static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque)
3507 BlockDriverAIOCBCoroutine *acb = opaque;
3508 BlockDriverState *bs = acb->common.bs;
3510 acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors);
3511 acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
3512 qemu_bh_schedule(acb->bh);
3515 BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs,
3516 int64_t sector_num, int nb_sectors,
3517 BlockDriverCompletionFunc *cb, void *opaque)
3520 BlockDriverAIOCBCoroutine *acb;
3522 trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque);
3524 acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
3525 acb->req.sector = sector_num;
3526 acb->req.nb_sectors = nb_sectors;
3527 co = qemu_coroutine_create(bdrv_aio_discard_co_entry);
3528 qemu_coroutine_enter(co, acb);
3530 return &acb->common;
3533 void bdrv_init(void)
3535 module_call_init(MODULE_INIT_BLOCK);
3538 void bdrv_init_with_whitelist(void)
3540 use_bdrv_whitelist = 1;
3544 void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
3545 BlockDriverCompletionFunc *cb, void *opaque)
3547 BlockDriverAIOCB *acb;
3549 if (pool->free_aiocb) {
3550 acb = pool->free_aiocb;
3551 pool->free_aiocb = acb->next;
3553 acb = g_malloc0(pool->aiocb_size);
3558 acb->opaque = opaque;
3562 void qemu_aio_release(void *p)
3564 BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
3565 AIOPool *pool = acb->pool;
3566 acb->next = pool->free_aiocb;
3567 pool->free_aiocb = acb;
3570 /**************************************************************/
3571 /* Coroutine block device emulation */
3573 typedef struct CoroutineIOCompletion {
3574 Coroutine *coroutine;
3576 } CoroutineIOCompletion;
3578 static void bdrv_co_io_em_complete(void *opaque, int ret)
3580 CoroutineIOCompletion *co = opaque;
3583 qemu_coroutine_enter(co->coroutine, NULL);
3586 static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num,
3587 int nb_sectors, QEMUIOVector *iov,
3590 CoroutineIOCompletion co = {
3591 .coroutine = qemu_coroutine_self(),
3593 BlockDriverAIOCB *acb;
3596 acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors,
3597 bdrv_co_io_em_complete, &co);
3599 acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors,
3600 bdrv_co_io_em_complete, &co);
3603 trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb);
3607 qemu_coroutine_yield();
3612 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
3613 int64_t sector_num, int nb_sectors,
3616 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false);
3619 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
3620 int64_t sector_num, int nb_sectors,
3623 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true);
3626 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
3628 RwCo *rwco = opaque;
3630 rwco->ret = bdrv_co_flush(rwco->bs);
3633 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
3637 if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
3641 /* Write back cached data to the OS even with cache=unsafe */
3642 if (bs->drv->bdrv_co_flush_to_os) {
3643 ret = bs->drv->bdrv_co_flush_to_os(bs);
3649 /* But don't actually force it to the disk with cache=unsafe */
3650 if (bs->open_flags & BDRV_O_NO_FLUSH) {
3654 if (bs->drv->bdrv_co_flush_to_disk) {
3655 ret = bs->drv->bdrv_co_flush_to_disk(bs);
3656 } else if (bs->drv->bdrv_aio_flush) {
3657 BlockDriverAIOCB *acb;
3658 CoroutineIOCompletion co = {
3659 .coroutine = qemu_coroutine_self(),
3662 acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
3666 qemu_coroutine_yield();
3671 * Some block drivers always operate in either writethrough or unsafe
3672 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
3673 * know how the server works (because the behaviour is hardcoded or
3674 * depends on server-side configuration), so we can't ensure that
3675 * everything is safe on disk. Returning an error doesn't work because
3676 * that would break guests even if the server operates in writethrough
3679 * Let's hope the user knows what he's doing.
3687 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
3688 * in the case of cache=unsafe, so there are no useless flushes.
3690 return bdrv_co_flush(bs->file);
3693 void bdrv_invalidate_cache(BlockDriverState *bs)
3695 if (bs->drv && bs->drv->bdrv_invalidate_cache) {
3696 bs->drv->bdrv_invalidate_cache(bs);
3700 void bdrv_invalidate_cache_all(void)
3702 BlockDriverState *bs;
3704 QTAILQ_FOREACH(bs, &bdrv_states, list) {
3705 bdrv_invalidate_cache(bs);
3709 void bdrv_clear_incoming_migration_all(void)
3711 BlockDriverState *bs;
3713 QTAILQ_FOREACH(bs, &bdrv_states, list) {
3714 bs->open_flags = bs->open_flags & ~(BDRV_O_INCOMING);
3718 int bdrv_flush(BlockDriverState *bs)
3726 if (qemu_in_coroutine()) {
3727 /* Fast-path if already in coroutine context */
3728 bdrv_flush_co_entry(&rwco);
3730 co = qemu_coroutine_create(bdrv_flush_co_entry);
3731 qemu_coroutine_enter(co, &rwco);
3732 while (rwco.ret == NOT_DONE) {
3740 static void coroutine_fn bdrv_discard_co_entry(void *opaque)
3742 RwCo *rwco = opaque;
3744 rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors);
3747 int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
3752 } else if (bdrv_check_request(bs, sector_num, nb_sectors)) {
3754 } else if (bs->read_only) {
3756 } else if (bs->drv->bdrv_co_discard) {
3757 return bs->drv->bdrv_co_discard(bs, sector_num, nb_sectors);
3758 } else if (bs->drv->bdrv_aio_discard) {
3759 BlockDriverAIOCB *acb;
3760 CoroutineIOCompletion co = {
3761 .coroutine = qemu_coroutine_self(),
3764 acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors,
3765 bdrv_co_io_em_complete, &co);
3769 qemu_coroutine_yield();
3777 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
3782 .sector_num = sector_num,
3783 .nb_sectors = nb_sectors,
3787 if (qemu_in_coroutine()) {
3788 /* Fast-path if already in coroutine context */
3789 bdrv_discard_co_entry(&rwco);
3791 co = qemu_coroutine_create(bdrv_discard_co_entry);
3792 qemu_coroutine_enter(co, &rwco);
3793 while (rwco.ret == NOT_DONE) {
3801 /**************************************************************/
3802 /* removable device support */
3805 * Return TRUE if the media is present
3807 int bdrv_is_inserted(BlockDriverState *bs)
3809 BlockDriver *drv = bs->drv;
3813 if (!drv->bdrv_is_inserted)
3815 return drv->bdrv_is_inserted(bs);
3819 * Return whether the media changed since the last call to this
3820 * function, or -ENOTSUP if we don't know. Most drivers don't know.
3822 int bdrv_media_changed(BlockDriverState *bs)
3824 BlockDriver *drv = bs->drv;
3826 if (drv && drv->bdrv_media_changed) {
3827 return drv->bdrv_media_changed(bs);
3833 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
3835 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
3837 BlockDriver *drv = bs->drv;
3839 if (drv && drv->bdrv_eject) {
3840 drv->bdrv_eject(bs, eject_flag);
3843 if (bs->device_name[0] != '\0') {
3844 bdrv_emit_qmp_eject_event(bs, eject_flag);
3849 * Lock or unlock the media (if it is locked, the user won't be able
3850 * to eject it manually).
3852 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
3854 BlockDriver *drv = bs->drv;
3856 trace_bdrv_lock_medium(bs, locked);
3858 if (drv && drv->bdrv_lock_medium) {
3859 drv->bdrv_lock_medium(bs, locked);
3863 /* needed for generic scsi interface */
3865 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
3867 BlockDriver *drv = bs->drv;
3869 if (drv && drv->bdrv_ioctl)
3870 return drv->bdrv_ioctl(bs, req, buf);
3874 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
3875 unsigned long int req, void *buf,
3876 BlockDriverCompletionFunc *cb, void *opaque)
3878 BlockDriver *drv = bs->drv;
3880 if (drv && drv->bdrv_aio_ioctl)
3881 return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
3885 void bdrv_set_buffer_alignment(BlockDriverState *bs, int align)
3887 bs->buffer_alignment = align;
3890 void *qemu_blockalign(BlockDriverState *bs, size_t size)
3892 return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
3895 void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
3897 int64_t bitmap_size;
3899 bs->dirty_count = 0;
3901 if (!bs->dirty_bitmap) {
3902 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
3903 BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
3904 bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
3906 bs->dirty_bitmap = g_malloc0(bitmap_size);
3909 if (bs->dirty_bitmap) {
3910 g_free(bs->dirty_bitmap);
3911 bs->dirty_bitmap = NULL;
3916 int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
3918 int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
3920 if (bs->dirty_bitmap &&
3921 (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
3922 return !!(bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
3923 (1UL << (chunk % (sizeof(unsigned long) * 8))));
3929 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
3932 set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
3935 int64_t bdrv_get_dirty_count(BlockDriverState *bs)
3937 return bs->dirty_count;
3940 void bdrv_set_in_use(BlockDriverState *bs, int in_use)
3942 assert(bs->in_use != in_use);
3943 bs->in_use = in_use;
3946 int bdrv_in_use(BlockDriverState *bs)
3951 void bdrv_iostatus_enable(BlockDriverState *bs)
3953 bs->iostatus_enabled = true;
3954 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
3957 /* The I/O status is only enabled if the drive explicitly
3958 * enables it _and_ the VM is configured to stop on errors */
3959 bool bdrv_iostatus_is_enabled(const BlockDriverState *bs)
3961 return (bs->iostatus_enabled &&
3962 (bs->on_write_error == BLOCK_ERR_STOP_ENOSPC ||
3963 bs->on_write_error == BLOCK_ERR_STOP_ANY ||
3964 bs->on_read_error == BLOCK_ERR_STOP_ANY));
3967 void bdrv_iostatus_disable(BlockDriverState *bs)
3969 bs->iostatus_enabled = false;
3972 void bdrv_iostatus_reset(BlockDriverState *bs)
3974 if (bdrv_iostatus_is_enabled(bs)) {
3975 bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
3979 /* XXX: Today this is set by device models because it makes the implementation
3980 quite simple. However, the block layer knows about the error, so it's
3981 possible to implement this without device models being involved */
3982 void bdrv_iostatus_set_err(BlockDriverState *bs, int error)
3984 if (bdrv_iostatus_is_enabled(bs) &&
3985 bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
3987 bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
3988 BLOCK_DEVICE_IO_STATUS_FAILED;
3993 bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes,
3994 enum BlockAcctType type)
3996 assert(type < BDRV_MAX_IOTYPE);
3998 cookie->bytes = bytes;
3999 cookie->start_time_ns = get_clock();
4000 cookie->type = type;
4004 bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie)
4006 assert(cookie->type < BDRV_MAX_IOTYPE);
4008 bs->nr_bytes[cookie->type] += cookie->bytes;
4009 bs->nr_ops[cookie->type]++;
4010 bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns;
4013 int bdrv_img_create(const char *filename, const char *fmt,
4014 const char *base_filename, const char *base_fmt,
4015 char *options, uint64_t img_size, int flags)
4017 QEMUOptionParameter *param = NULL, *create_options = NULL;
4018 QEMUOptionParameter *backing_fmt, *backing_file, *size;
4019 BlockDriverState *bs = NULL;
4020 BlockDriver *drv, *proto_drv;
4021 BlockDriver *backing_drv = NULL;
4024 /* Find driver and parse its options */
4025 drv = bdrv_find_format(fmt);
4027 error_report("Unknown file format '%s'", fmt);
4032 proto_drv = bdrv_find_protocol(filename);
4034 error_report("Unknown protocol '%s'", filename);
4039 create_options = append_option_parameters(create_options,
4040 drv->create_options);
4041 create_options = append_option_parameters(create_options,
4042 proto_drv->create_options);
4044 /* Create parameter list with default values */
4045 param = parse_option_parameters("", create_options, param);
4047 set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size);
4049 /* Parse -o options */
4051 param = parse_option_parameters(options, create_options, param);
4052 if (param == NULL) {
4053 error_report("Invalid options for file format '%s'.", fmt);
4059 if (base_filename) {
4060 if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE,
4062 error_report("Backing file not supported for file format '%s'",
4070 if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) {
4071 error_report("Backing file format not supported for file "
4072 "format '%s'", fmt);
4078 backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE);
4079 if (backing_file && backing_file->value.s) {
4080 if (!strcmp(filename, backing_file->value.s)) {
4081 error_report("Error: Trying to create an image with the "
4082 "same filename as the backing file");
4088 backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT);
4089 if (backing_fmt && backing_fmt->value.s) {
4090 backing_drv = bdrv_find_format(backing_fmt->value.s);
4092 error_report("Unknown backing file format '%s'",
4093 backing_fmt->value.s);
4099 // The size for the image must always be specified, with one exception:
4100 // If we are using a backing file, we can obtain the size from there
4101 size = get_option_parameter(param, BLOCK_OPT_SIZE);
4102 if (size && size->value.n == -1) {
4103 if (backing_file && backing_file->value.s) {
4109 ret = bdrv_open(bs, backing_file->value.s, flags, backing_drv);
4111 error_report("Could not open '%s'", backing_file->value.s);
4114 bdrv_get_geometry(bs, &size);
4117 snprintf(buf, sizeof(buf), "%" PRId64, size);
4118 set_option_parameter(param, BLOCK_OPT_SIZE, buf);
4120 error_report("Image creation needs a size parameter");
4126 printf("Formatting '%s', fmt=%s ", filename, fmt);
4127 print_option_parameters(param);
4130 ret = bdrv_create(drv, filename, param);
4133 if (ret == -ENOTSUP) {
4134 error_report("Formatting or formatting option not supported for "
4135 "file format '%s'", fmt);
4136 } else if (ret == -EFBIG) {
4137 error_report("The image size is too large for file format '%s'",
4140 error_report("%s: error while creating %s: %s", filename, fmt,
4146 free_option_parameters(create_options);
4147 free_option_parameters(param);
4156 void *block_job_create(const BlockJobType *job_type, BlockDriverState *bs,
4157 int64_t speed, BlockDriverCompletionFunc *cb,
4158 void *opaque, Error **errp)
4162 if (bs->job || bdrv_in_use(bs)) {
4163 error_set(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
4166 bdrv_set_in_use(bs, 1);
4168 job = g_malloc0(job_type->instance_size);
4169 job->job_type = job_type;
4172 job->opaque = opaque;
4175 /* Only set speed when necessary to avoid NotSupported error */
4177 Error *local_err = NULL;
4179 block_job_set_speed(job, speed, &local_err);
4180 if (error_is_set(&local_err)) {
4183 bdrv_set_in_use(bs, 0);
4184 error_propagate(errp, local_err);
4191 void block_job_complete(BlockJob *job, int ret)
4193 BlockDriverState *bs = job->bs;
4195 assert(bs->job == job);
4196 job->cb(job->opaque, ret);
4199 bdrv_set_in_use(bs, 0);
4202 void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
4204 Error *local_err = NULL;
4206 if (!job->job_type->set_speed) {
4207 error_set(errp, QERR_NOT_SUPPORTED);
4210 job->job_type->set_speed(job, speed, &local_err);
4211 if (error_is_set(&local_err)) {
4212 error_propagate(errp, local_err);
4219 void block_job_cancel(BlockJob *job)
4221 job->cancelled = true;
4224 bool block_job_is_cancelled(BlockJob *job)
4226 return job->cancelled;
4229 void block_job_cancel_sync(BlockJob *job)
4231 BlockDriverState *bs = job->bs;
4233 assert(bs->job == job);
4234 block_job_cancel(job);
4235 while (bs->job != NULL && bs->job->busy) {