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"
33 #include "qemu-coroutine.h"
34 #include "qmp-commands.h"
35 #include "qemu-timer.h"
38 #include <sys/types.h>
40 #include <sys/ioctl.h>
41 #include <sys/queue.h>
51 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
54 BDRV_REQ_COPY_ON_READ
= 0x1,
55 BDRV_REQ_ZERO_WRITE
= 0x2,
58 static void bdrv_dev_change_media_cb(BlockDriverState
*bs
, bool load
);
59 static BlockDriverAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
60 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
61 BlockDriverCompletionFunc
*cb
, void *opaque
);
62 static BlockDriverAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
63 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
64 BlockDriverCompletionFunc
*cb
, void *opaque
);
65 static int coroutine_fn
bdrv_co_readv_em(BlockDriverState
*bs
,
66 int64_t sector_num
, int nb_sectors
,
68 static int coroutine_fn
bdrv_co_writev_em(BlockDriverState
*bs
,
69 int64_t sector_num
, int nb_sectors
,
71 static int coroutine_fn
bdrv_co_do_readv(BlockDriverState
*bs
,
72 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
73 BdrvRequestFlags flags
);
74 static int coroutine_fn
bdrv_co_do_writev(BlockDriverState
*bs
,
75 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
76 BdrvRequestFlags flags
);
77 static BlockDriverAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
81 BlockDriverCompletionFunc
*cb
,
84 static void coroutine_fn
bdrv_co_do_rw(void *opaque
);
85 static int coroutine_fn
bdrv_co_do_write_zeroes(BlockDriverState
*bs
,
86 int64_t sector_num
, int nb_sectors
);
88 static bool bdrv_exceed_bps_limits(BlockDriverState
*bs
, int nb_sectors
,
89 bool is_write
, double elapsed_time
, uint64_t *wait
);
90 static bool bdrv_exceed_iops_limits(BlockDriverState
*bs
, bool is_write
,
91 double elapsed_time
, uint64_t *wait
);
92 static bool bdrv_exceed_io_limits(BlockDriverState
*bs
, int nb_sectors
,
93 bool is_write
, int64_t *wait
);
95 static QTAILQ_HEAD(, BlockDriverState
) bdrv_states
=
96 QTAILQ_HEAD_INITIALIZER(bdrv_states
);
98 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
99 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
101 /* The device to use for VM snapshots */
102 static BlockDriverState
*bs_snapshots
;
104 /* If non-zero, use only whitelisted block drivers */
105 static int use_bdrv_whitelist
;
108 static int is_windows_drive_prefix(const char *filename
)
110 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
111 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
115 int is_windows_drive(const char *filename
)
117 if (is_windows_drive_prefix(filename
) &&
120 if (strstart(filename
, "\\\\.\\", NULL
) ||
121 strstart(filename
, "//./", NULL
))
127 /* throttling disk I/O limits */
128 void bdrv_io_limits_disable(BlockDriverState
*bs
)
130 bs
->io_limits_enabled
= false;
132 while (qemu_co_queue_next(&bs
->throttled_reqs
));
134 if (bs
->block_timer
) {
135 qemu_del_timer(bs
->block_timer
);
136 qemu_free_timer(bs
->block_timer
);
137 bs
->block_timer
= NULL
;
143 memset(&bs
->io_base
, 0, sizeof(bs
->io_base
));
146 static void bdrv_block_timer(void *opaque
)
148 BlockDriverState
*bs
= opaque
;
150 qemu_co_queue_next(&bs
->throttled_reqs
);
153 void bdrv_io_limits_enable(BlockDriverState
*bs
)
155 qemu_co_queue_init(&bs
->throttled_reqs
);
156 bs
->block_timer
= qemu_new_timer_ns(vm_clock
, bdrv_block_timer
, bs
);
157 bs
->slice_time
= 5 * BLOCK_IO_SLICE_TIME
;
158 bs
->slice_start
= qemu_get_clock_ns(vm_clock
);
159 bs
->slice_end
= bs
->slice_start
+ bs
->slice_time
;
160 memset(&bs
->io_base
, 0, sizeof(bs
->io_base
));
161 bs
->io_limits_enabled
= true;
164 bool bdrv_io_limits_enabled(BlockDriverState
*bs
)
166 BlockIOLimit
*io_limits
= &bs
->io_limits
;
167 return io_limits
->bps
[BLOCK_IO_LIMIT_READ
]
168 || io_limits
->bps
[BLOCK_IO_LIMIT_WRITE
]
169 || io_limits
->bps
[BLOCK_IO_LIMIT_TOTAL
]
170 || io_limits
->iops
[BLOCK_IO_LIMIT_READ
]
171 || io_limits
->iops
[BLOCK_IO_LIMIT_WRITE
]
172 || io_limits
->iops
[BLOCK_IO_LIMIT_TOTAL
];
175 static void bdrv_io_limits_intercept(BlockDriverState
*bs
,
176 bool is_write
, int nb_sectors
)
178 int64_t wait_time
= -1;
180 if (!qemu_co_queue_empty(&bs
->throttled_reqs
)) {
181 qemu_co_queue_wait(&bs
->throttled_reqs
);
184 /* In fact, we hope to keep each request's timing, in FIFO mode. The next
185 * throttled requests will not be dequeued until the current request is
186 * allowed to be serviced. So if the current request still exceeds the
187 * limits, it will be inserted to the head. All requests followed it will
188 * be still in throttled_reqs queue.
191 while (bdrv_exceed_io_limits(bs
, nb_sectors
, is_write
, &wait_time
)) {
192 qemu_mod_timer(bs
->block_timer
,
193 wait_time
+ qemu_get_clock_ns(vm_clock
));
194 qemu_co_queue_wait_insert_head(&bs
->throttled_reqs
);
197 qemu_co_queue_next(&bs
->throttled_reqs
);
200 /* check if the path starts with "<protocol>:" */
201 static int path_has_protocol(const char *path
)
206 if (is_windows_drive(path
) ||
207 is_windows_drive_prefix(path
)) {
210 p
= path
+ strcspn(path
, ":/\\");
212 p
= path
+ strcspn(path
, ":/");
218 int path_is_absolute(const char *path
)
221 /* specific case for names like: "\\.\d:" */
222 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
225 return (*path
== '/' || *path
== '\\');
227 return (*path
== '/');
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_get_full_backing_filename(BlockDriverState
*bs
, char *dest
, size_t sz
)
277 if (bs
->backing_file
[0] == '\0' || path_has_protocol(bs
->backing_file
)) {
278 pstrcpy(dest
, sz
, bs
->backing_file
);
280 path_combine(dest
, sz
, bs
->filename
, bs
->backing_file
);
284 void bdrv_register(BlockDriver
*bdrv
)
286 /* Block drivers without coroutine functions need emulation */
287 if (!bdrv
->bdrv_co_readv
) {
288 bdrv
->bdrv_co_readv
= bdrv_co_readv_em
;
289 bdrv
->bdrv_co_writev
= bdrv_co_writev_em
;
291 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
292 * the block driver lacks aio we need to emulate that too.
294 if (!bdrv
->bdrv_aio_readv
) {
295 /* add AIO emulation layer */
296 bdrv
->bdrv_aio_readv
= bdrv_aio_readv_em
;
297 bdrv
->bdrv_aio_writev
= bdrv_aio_writev_em
;
301 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
304 /* create a new block device (by default it is empty) */
305 BlockDriverState
*bdrv_new(const char *device_name
)
307 BlockDriverState
*bs
;
309 bs
= g_malloc0(sizeof(BlockDriverState
));
310 pstrcpy(bs
->device_name
, sizeof(bs
->device_name
), device_name
);
311 if (device_name
[0] != '\0') {
312 QTAILQ_INSERT_TAIL(&bdrv_states
, bs
, list
);
314 bdrv_iostatus_disable(bs
);
318 BlockDriver
*bdrv_find_format(const char *format_name
)
321 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
322 if (!strcmp(drv1
->format_name
, format_name
)) {
329 static int bdrv_is_whitelisted(BlockDriver
*drv
)
331 static const char *whitelist
[] = {
332 CONFIG_BDRV_WHITELIST
337 return 1; /* no whitelist, anything goes */
339 for (p
= whitelist
; *p
; p
++) {
340 if (!strcmp(drv
->format_name
, *p
)) {
347 BlockDriver
*bdrv_find_whitelisted_format(const char *format_name
)
349 BlockDriver
*drv
= bdrv_find_format(format_name
);
350 return drv
&& bdrv_is_whitelisted(drv
) ? drv
: NULL
;
353 typedef struct CreateCo
{
356 QEMUOptionParameter
*options
;
360 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
362 CreateCo
*cco
= opaque
;
365 cco
->ret
= cco
->drv
->bdrv_create(cco
->filename
, cco
->options
);
368 int bdrv_create(BlockDriver
*drv
, const char* filename
,
369 QEMUOptionParameter
*options
)
376 .filename
= g_strdup(filename
),
381 if (!drv
->bdrv_create
) {
385 if (qemu_in_coroutine()) {
386 /* Fast-path if already in coroutine context */
387 bdrv_create_co_entry(&cco
);
389 co
= qemu_coroutine_create(bdrv_create_co_entry
);
390 qemu_coroutine_enter(co
, &cco
);
391 while (cco
.ret
== NOT_DONE
) {
397 g_free(cco
.filename
);
402 int bdrv_create_file(const char* filename
, QEMUOptionParameter
*options
)
406 drv
= bdrv_find_protocol(filename
);
411 return bdrv_create(drv
, filename
, options
);
415 * Create a uniquely-named empty temporary file.
416 * Return 0 upon success, otherwise a negative errno value.
418 int get_tmp_filename(char *filename
, int size
)
421 char temp_dir
[MAX_PATH
];
422 /* GetTempFileName requires that its output buffer (4th param)
423 have length MAX_PATH or greater. */
424 assert(size
>= MAX_PATH
);
425 return (GetTempPath(MAX_PATH
, temp_dir
)
426 && GetTempFileName(temp_dir
, "qem", 0, filename
)
427 ? 0 : -GetLastError());
431 tmpdir
= getenv("TMPDIR");
434 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
437 fd
= mkstemp(filename
);
441 if (close(fd
) != 0) {
450 * Detect host devices. By convention, /dev/cdrom[N] is always
451 * recognized as a host CDROM.
453 static BlockDriver
*find_hdev_driver(const char *filename
)
455 int score_max
= 0, score
;
456 BlockDriver
*drv
= NULL
, *d
;
458 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
459 if (d
->bdrv_probe_device
) {
460 score
= d
->bdrv_probe_device(filename
);
461 if (score
> score_max
) {
471 BlockDriver
*bdrv_find_protocol(const char *filename
)
478 /* TODO Drivers without bdrv_file_open must be specified explicitly */
481 * XXX(hch): we really should not let host device detection
482 * override an explicit protocol specification, but moving this
483 * later breaks access to device names with colons in them.
484 * Thanks to the brain-dead persistent naming schemes on udev-
485 * based Linux systems those actually are quite common.
487 drv1
= find_hdev_driver(filename
);
492 if (!path_has_protocol(filename
)) {
493 return bdrv_find_format("file");
495 p
= strchr(filename
, ':');
498 if (len
> sizeof(protocol
) - 1)
499 len
= sizeof(protocol
) - 1;
500 memcpy(protocol
, filename
, len
);
501 protocol
[len
] = '\0';
502 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
503 if (drv1
->protocol_name
&&
504 !strcmp(drv1
->protocol_name
, protocol
)) {
511 static int find_image_format(const char *filename
, BlockDriver
**pdrv
)
513 int ret
, score
, score_max
;
514 BlockDriver
*drv1
, *drv
;
516 BlockDriverState
*bs
;
518 ret
= bdrv_file_open(&bs
, filename
, 0);
524 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
525 if (bs
->sg
|| !bdrv_is_inserted(bs
)) {
527 drv
= bdrv_find_format("raw");
535 ret
= bdrv_pread(bs
, 0, buf
, sizeof(buf
));
544 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
545 if (drv1
->bdrv_probe
) {
546 score
= drv1
->bdrv_probe(buf
, ret
, filename
);
547 if (score
> score_max
) {
561 * Set the current 'total_sectors' value
563 static int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
565 BlockDriver
*drv
= bs
->drv
;
567 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
571 /* query actual device if possible, otherwise just trust the hint */
572 if (drv
->bdrv_getlength
) {
573 int64_t length
= drv
->bdrv_getlength(bs
);
577 hint
= length
>> BDRV_SECTOR_BITS
;
580 bs
->total_sectors
= hint
;
585 * Set open flags for a given cache mode
587 * Return 0 on success, -1 if the cache mode was invalid.
589 int bdrv_parse_cache_flags(const char *mode
, int *flags
)
591 *flags
&= ~BDRV_O_CACHE_MASK
;
593 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
594 *flags
|= BDRV_O_NOCACHE
| BDRV_O_CACHE_WB
;
595 } else if (!strcmp(mode
, "directsync")) {
596 *flags
|= BDRV_O_NOCACHE
;
597 } else if (!strcmp(mode
, "writeback")) {
598 *flags
|= BDRV_O_CACHE_WB
;
599 } else if (!strcmp(mode
, "unsafe")) {
600 *flags
|= BDRV_O_CACHE_WB
;
601 *flags
|= BDRV_O_NO_FLUSH
;
602 } else if (!strcmp(mode
, "writethrough")) {
603 /* this is the default */
612 * The copy-on-read flag is actually a reference count so multiple users may
613 * use the feature without worrying about clobbering its previous state.
614 * Copy-on-read stays enabled until all users have called to disable it.
616 void bdrv_enable_copy_on_read(BlockDriverState
*bs
)
621 void bdrv_disable_copy_on_read(BlockDriverState
*bs
)
623 assert(bs
->copy_on_read
> 0);
628 * Common part for opening disk images and files
630 static int bdrv_open_common(BlockDriverState
*bs
, const char *filename
,
631 int flags
, BlockDriver
*drv
)
636 assert(bs
->file
== NULL
);
638 trace_bdrv_open_common(bs
, filename
, flags
, drv
->format_name
);
640 bs
->open_flags
= flags
;
641 bs
->buffer_alignment
= 512;
643 assert(bs
->copy_on_read
== 0); /* bdrv_new() and bdrv_close() make it so */
644 if ((flags
& BDRV_O_RDWR
) && (flags
& BDRV_O_COPY_ON_READ
)) {
645 bdrv_enable_copy_on_read(bs
);
648 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
650 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
)) {
655 bs
->opaque
= g_malloc0(drv
->instance_size
);
657 bs
->enable_write_cache
= !!(flags
& BDRV_O_CACHE_WB
);
658 open_flags
= flags
| BDRV_O_CACHE_WB
;
661 * Clear flags that are internal to the block layer before opening the
664 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
667 * Snapshots should be writable.
669 if (bs
->is_temporary
) {
670 open_flags
|= BDRV_O_RDWR
;
673 bs
->read_only
= !(open_flags
& BDRV_O_RDWR
);
675 /* Open the image, either directly or using a protocol */
676 if (drv
->bdrv_file_open
) {
677 ret
= drv
->bdrv_file_open(bs
, filename
, open_flags
);
679 ret
= bdrv_file_open(&bs
->file
, filename
, open_flags
);
681 ret
= drv
->bdrv_open(bs
, open_flags
);
689 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
695 if (bs
->is_temporary
) {
703 bdrv_delete(bs
->file
);
713 * Opens a file using a protocol (file, host_device, nbd, ...)
715 int bdrv_file_open(BlockDriverState
**pbs
, const char *filename
, int flags
)
717 BlockDriverState
*bs
;
721 drv
= bdrv_find_protocol(filename
);
727 ret
= bdrv_open_common(bs
, filename
, flags
, drv
);
738 * Opens a disk image (raw, qcow2, vmdk, ...)
740 int bdrv_open(BlockDriverState
*bs
, const char *filename
, int flags
,
744 char tmp_filename
[PATH_MAX
];
746 if (flags
& BDRV_O_SNAPSHOT
) {
747 BlockDriverState
*bs1
;
750 BlockDriver
*bdrv_qcow2
;
751 QEMUOptionParameter
*options
;
752 char backing_filename
[PATH_MAX
];
754 /* if snapshot, we create a temporary backing file and open it
755 instead of opening 'filename' directly */
757 /* if there is a backing file, use it */
759 ret
= bdrv_open(bs1
, filename
, 0, drv
);
764 total_size
= bdrv_getlength(bs1
) & BDRV_SECTOR_MASK
;
766 if (bs1
->drv
&& bs1
->drv
->protocol_name
)
771 ret
= get_tmp_filename(tmp_filename
, sizeof(tmp_filename
));
776 /* Real path is meaningless for protocols */
778 snprintf(backing_filename
, sizeof(backing_filename
),
780 else if (!realpath(filename
, backing_filename
))
783 bdrv_qcow2
= bdrv_find_format("qcow2");
784 options
= parse_option_parameters("", bdrv_qcow2
->create_options
, NULL
);
786 set_option_parameter_int(options
, BLOCK_OPT_SIZE
, total_size
);
787 set_option_parameter(options
, BLOCK_OPT_BACKING_FILE
, backing_filename
);
789 set_option_parameter(options
, BLOCK_OPT_BACKING_FMT
,
793 ret
= bdrv_create(bdrv_qcow2
, tmp_filename
, options
);
794 free_option_parameters(options
);
799 filename
= tmp_filename
;
801 bs
->is_temporary
= 1;
804 /* Find the right image format driver */
806 ret
= find_image_format(filename
, &drv
);
810 goto unlink_and_fail
;
813 if (flags
& BDRV_O_RDWR
) {
814 flags
|= BDRV_O_ALLOW_RDWR
;
818 ret
= bdrv_open_common(bs
, filename
, flags
, drv
);
820 goto unlink_and_fail
;
823 /* If there is a backing file, use it */
824 if ((flags
& BDRV_O_NO_BACKING
) == 0 && bs
->backing_file
[0] != '\0') {
825 char backing_filename
[PATH_MAX
];
827 BlockDriver
*back_drv
= NULL
;
829 bs
->backing_hd
= bdrv_new("");
830 bdrv_get_full_backing_filename(bs
, backing_filename
,
831 sizeof(backing_filename
));
833 if (bs
->backing_format
[0] != '\0') {
834 back_drv
= bdrv_find_format(bs
->backing_format
);
837 /* backing files always opened read-only */
839 flags
& ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
841 ret
= bdrv_open(bs
->backing_hd
, backing_filename
, back_flags
, back_drv
);
848 if (!bdrv_key_required(bs
)) {
849 bdrv_dev_change_media_cb(bs
, true);
852 /* throttling disk I/O limits */
853 if (bs
->io_limits_enabled
) {
854 bdrv_io_limits_enable(bs
);
860 if (bs
->is_temporary
) {
866 typedef struct BlockReopenQueueEntry
{
868 BDRVReopenState state
;
869 QSIMPLEQ_ENTRY(BlockReopenQueueEntry
) entry
;
870 } BlockReopenQueueEntry
;
873 * Adds a BlockDriverState to a simple queue for an atomic, transactional
874 * reopen of multiple devices.
876 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
877 * already performed, or alternatively may be NULL a new BlockReopenQueue will
878 * be created and initialized. This newly created BlockReopenQueue should be
879 * passed back in for subsequent calls that are intended to be of the same
882 * bs is the BlockDriverState to add to the reopen queue.
884 * flags contains the open flags for the associated bs
886 * returns a pointer to bs_queue, which is either the newly allocated
887 * bs_queue, or the existing bs_queue being used.
890 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
891 BlockDriverState
*bs
, int flags
)
895 BlockReopenQueueEntry
*bs_entry
;
896 if (bs_queue
== NULL
) {
897 bs_queue
= g_new0(BlockReopenQueue
, 1);
898 QSIMPLEQ_INIT(bs_queue
);
902 bdrv_reopen_queue(bs_queue
, bs
->file
, flags
);
905 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
906 QSIMPLEQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
908 bs_entry
->state
.bs
= bs
;
909 bs_entry
->state
.flags
= flags
;
915 * Reopen multiple BlockDriverStates atomically & transactionally.
917 * The queue passed in (bs_queue) must have been built up previous
918 * via bdrv_reopen_queue().
920 * Reopens all BDS specified in the queue, with the appropriate
921 * flags. All devices are prepared for reopen, and failure of any
922 * device will cause all device changes to be abandonded, and intermediate
925 * If all devices prepare successfully, then the changes are committed
929 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
932 BlockReopenQueueEntry
*bs_entry
, *next
;
933 Error
*local_err
= NULL
;
935 assert(bs_queue
!= NULL
);
939 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
940 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, &local_err
)) {
941 error_propagate(errp
, local_err
);
944 bs_entry
->prepared
= true;
947 /* If we reach this point, we have success and just need to apply the
950 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
951 bdrv_reopen_commit(&bs_entry
->state
);
957 QSIMPLEQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
958 if (ret
&& bs_entry
->prepared
) {
959 bdrv_reopen_abort(&bs_entry
->state
);
968 /* Reopen a single BlockDriverState with the specified flags. */
969 int bdrv_reopen(BlockDriverState
*bs
, int bdrv_flags
, Error
**errp
)
972 Error
*local_err
= NULL
;
973 BlockReopenQueue
*queue
= bdrv_reopen_queue(NULL
, bs
, bdrv_flags
);
975 ret
= bdrv_reopen_multiple(queue
, &local_err
);
976 if (local_err
!= NULL
) {
977 error_propagate(errp
, local_err
);
984 * Prepares a BlockDriverState for reopen. All changes are staged in the
985 * 'opaque' field of the BDRVReopenState, which is used and allocated by
986 * the block driver layer .bdrv_reopen_prepare()
988 * bs is the BlockDriverState to reopen
989 * flags are the new open flags
990 * queue is the reopen queue
992 * Returns 0 on success, non-zero on error. On error errp will be set
995 * On failure, bdrv_reopen_abort() will be called to clean up any data.
996 * It is the responsibility of the caller to then call the abort() or
997 * commit() for any other BDS that have been left in a prepare() state
1000 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
1004 Error
*local_err
= NULL
;
1007 assert(reopen_state
!= NULL
);
1008 assert(reopen_state
->bs
->drv
!= NULL
);
1009 drv
= reopen_state
->bs
->drv
;
1011 /* if we are to stay read-only, do not allow permission change
1013 if (!(reopen_state
->bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
1014 reopen_state
->flags
& BDRV_O_RDWR
) {
1015 error_set(errp
, QERR_DEVICE_IS_READ_ONLY
,
1016 reopen_state
->bs
->device_name
);
1021 ret
= bdrv_flush(reopen_state
->bs
);
1023 error_set(errp
, ERROR_CLASS_GENERIC_ERROR
, "Error (%s) flushing drive",
1028 if (drv
->bdrv_reopen_prepare
) {
1029 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
1031 if (local_err
!= NULL
) {
1032 error_propagate(errp
, local_err
);
1034 error_set(errp
, QERR_OPEN_FILE_FAILED
,
1035 reopen_state
->bs
->filename
);
1040 /* It is currently mandatory to have a bdrv_reopen_prepare()
1041 * handler for each supported drv. */
1042 error_set(errp
, QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED
,
1043 drv
->format_name
, reopen_state
->bs
->device_name
,
1044 "reopening of file");
1056 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
1057 * makes them final by swapping the staging BlockDriverState contents into
1058 * the active BlockDriverState contents.
1060 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
1064 assert(reopen_state
!= NULL
);
1065 drv
= reopen_state
->bs
->drv
;
1066 assert(drv
!= NULL
);
1068 /* If there are any driver level actions to take */
1069 if (drv
->bdrv_reopen_commit
) {
1070 drv
->bdrv_reopen_commit(reopen_state
);
1073 /* set BDS specific flags now */
1074 reopen_state
->bs
->open_flags
= reopen_state
->flags
;
1075 reopen_state
->bs
->enable_write_cache
= !!(reopen_state
->flags
&
1077 reopen_state
->bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
1081 * Abort the reopen, and delete and free the staged changes in
1084 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
1088 assert(reopen_state
!= NULL
);
1089 drv
= reopen_state
->bs
->drv
;
1090 assert(drv
!= NULL
);
1092 if (drv
->bdrv_reopen_abort
) {
1093 drv
->bdrv_reopen_abort(reopen_state
);
1098 void bdrv_close(BlockDriverState
*bs
)
1103 block_job_cancel_sync(bs
->job
);
1107 if (bs
== bs_snapshots
) {
1108 bs_snapshots
= NULL
;
1110 if (bs
->backing_hd
) {
1111 bdrv_delete(bs
->backing_hd
);
1112 bs
->backing_hd
= NULL
;
1114 bs
->drv
->bdrv_close(bs
);
1117 if (bs
->is_temporary
) {
1118 unlink(bs
->filename
);
1123 bs
->copy_on_read
= 0;
1124 bs
->backing_file
[0] = '\0';
1125 bs
->backing_format
[0] = '\0';
1126 bs
->total_sectors
= 0;
1132 if (bs
->file
!= NULL
) {
1133 bdrv_delete(bs
->file
);
1138 bdrv_dev_change_media_cb(bs
, false);
1140 /*throttling disk I/O limits*/
1141 if (bs
->io_limits_enabled
) {
1142 bdrv_io_limits_disable(bs
);
1146 void bdrv_close_all(void)
1148 BlockDriverState
*bs
;
1150 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1156 * Wait for pending requests to complete across all BlockDriverStates
1158 * This function does not flush data to disk, use bdrv_flush_all() for that
1159 * after calling this function.
1161 * Note that completion of an asynchronous I/O operation can trigger any
1162 * number of other I/O operations on other devices---for example a coroutine
1163 * can be arbitrarily complex and a constant flow of I/O can come until the
1164 * coroutine is complete. Because of this, it is not possible to have a
1165 * function to drain a single device's I/O queue.
1167 void bdrv_drain_all(void)
1169 BlockDriverState
*bs
;
1173 busy
= qemu_aio_wait();
1175 /* FIXME: We do not have timer support here, so this is effectively
1178 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1179 if (!qemu_co_queue_empty(&bs
->throttled_reqs
)) {
1180 qemu_co_queue_restart_all(&bs
->throttled_reqs
);
1186 /* If requests are still pending there is a bug somewhere */
1187 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1188 assert(QLIST_EMPTY(&bs
->tracked_requests
));
1189 assert(qemu_co_queue_empty(&bs
->throttled_reqs
));
1193 /* make a BlockDriverState anonymous by removing from bdrv_state list.
1194 Also, NULL terminate the device_name to prevent double remove */
1195 void bdrv_make_anon(BlockDriverState
*bs
)
1197 if (bs
->device_name
[0] != '\0') {
1198 QTAILQ_REMOVE(&bdrv_states
, bs
, list
);
1200 bs
->device_name
[0] = '\0';
1203 static void bdrv_rebind(BlockDriverState
*bs
)
1205 if (bs
->drv
&& bs
->drv
->bdrv_rebind
) {
1206 bs
->drv
->bdrv_rebind(bs
);
1210 static void bdrv_move_feature_fields(BlockDriverState
*bs_dest
,
1211 BlockDriverState
*bs_src
)
1213 /* move some fields that need to stay attached to the device */
1214 bs_dest
->open_flags
= bs_src
->open_flags
;
1217 bs_dest
->dev_ops
= bs_src
->dev_ops
;
1218 bs_dest
->dev_opaque
= bs_src
->dev_opaque
;
1219 bs_dest
->dev
= bs_src
->dev
;
1220 bs_dest
->buffer_alignment
= bs_src
->buffer_alignment
;
1221 bs_dest
->copy_on_read
= bs_src
->copy_on_read
;
1223 bs_dest
->enable_write_cache
= bs_src
->enable_write_cache
;
1225 /* i/o timing parameters */
1226 bs_dest
->slice_time
= bs_src
->slice_time
;
1227 bs_dest
->slice_start
= bs_src
->slice_start
;
1228 bs_dest
->slice_end
= bs_src
->slice_end
;
1229 bs_dest
->io_limits
= bs_src
->io_limits
;
1230 bs_dest
->io_base
= bs_src
->io_base
;
1231 bs_dest
->throttled_reqs
= bs_src
->throttled_reqs
;
1232 bs_dest
->block_timer
= bs_src
->block_timer
;
1233 bs_dest
->io_limits_enabled
= bs_src
->io_limits_enabled
;
1236 bs_dest
->on_read_error
= bs_src
->on_read_error
;
1237 bs_dest
->on_write_error
= bs_src
->on_write_error
;
1240 bs_dest
->iostatus_enabled
= bs_src
->iostatus_enabled
;
1241 bs_dest
->iostatus
= bs_src
->iostatus
;
1244 bs_dest
->dirty_count
= bs_src
->dirty_count
;
1245 bs_dest
->dirty_bitmap
= bs_src
->dirty_bitmap
;
1248 bs_dest
->in_use
= bs_src
->in_use
;
1249 bs_dest
->job
= bs_src
->job
;
1251 /* keep the same entry in bdrv_states */
1252 pstrcpy(bs_dest
->device_name
, sizeof(bs_dest
->device_name
),
1253 bs_src
->device_name
);
1254 bs_dest
->list
= bs_src
->list
;
1258 * Swap bs contents for two image chains while they are live,
1259 * while keeping required fields on the BlockDriverState that is
1260 * actually attached to a device.
1262 * This will modify the BlockDriverState fields, and swap contents
1263 * between bs_new and bs_old. Both bs_new and bs_old are modified.
1265 * bs_new is required to be anonymous.
1267 * This function does not create any image files.
1269 void bdrv_swap(BlockDriverState
*bs_new
, BlockDriverState
*bs_old
)
1271 BlockDriverState tmp
;
1273 /* bs_new must be anonymous and shouldn't have anything fancy enabled */
1274 assert(bs_new
->device_name
[0] == '\0');
1275 assert(bs_new
->dirty_bitmap
== NULL
);
1276 assert(bs_new
->job
== NULL
);
1277 assert(bs_new
->dev
== NULL
);
1278 assert(bs_new
->in_use
== 0);
1279 assert(bs_new
->io_limits_enabled
== false);
1280 assert(bs_new
->block_timer
== NULL
);
1286 /* there are some fields that should not be swapped, move them back */
1287 bdrv_move_feature_fields(&tmp
, bs_old
);
1288 bdrv_move_feature_fields(bs_old
, bs_new
);
1289 bdrv_move_feature_fields(bs_new
, &tmp
);
1291 /* bs_new shouldn't be in bdrv_states even after the swap! */
1292 assert(bs_new
->device_name
[0] == '\0');
1294 /* Check a few fields that should remain attached to the device */
1295 assert(bs_new
->dev
== NULL
);
1296 assert(bs_new
->job
== NULL
);
1297 assert(bs_new
->in_use
== 0);
1298 assert(bs_new
->io_limits_enabled
== false);
1299 assert(bs_new
->block_timer
== NULL
);
1301 bdrv_rebind(bs_new
);
1302 bdrv_rebind(bs_old
);
1306 * Add new bs contents at the top of an image chain while the chain is
1307 * live, while keeping required fields on the top layer.
1309 * This will modify the BlockDriverState fields, and swap contents
1310 * between bs_new and bs_top. Both bs_new and bs_top are modified.
1312 * bs_new is required to be anonymous.
1314 * This function does not create any image files.
1316 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
)
1318 bdrv_swap(bs_new
, bs_top
);
1320 /* The contents of 'tmp' will become bs_top, as we are
1321 * swapping bs_new and bs_top contents. */
1322 bs_top
->backing_hd
= bs_new
;
1323 bs_top
->open_flags
&= ~BDRV_O_NO_BACKING
;
1324 pstrcpy(bs_top
->backing_file
, sizeof(bs_top
->backing_file
),
1326 pstrcpy(bs_top
->backing_format
, sizeof(bs_top
->backing_format
),
1327 bs_new
->drv
? bs_new
->drv
->format_name
: "");
1330 void bdrv_delete(BlockDriverState
*bs
)
1334 assert(!bs
->in_use
);
1336 /* remove from list, if necessary */
1341 assert(bs
!= bs_snapshots
);
1345 int bdrv_attach_dev(BlockDriverState
*bs
, void *dev
)
1346 /* TODO change to DeviceState *dev when all users are qdevified */
1352 bdrv_iostatus_reset(bs
);
1356 /* TODO qdevified devices don't use this, remove when devices are qdevified */
1357 void bdrv_attach_dev_nofail(BlockDriverState
*bs
, void *dev
)
1359 if (bdrv_attach_dev(bs
, dev
) < 0) {
1364 void bdrv_detach_dev(BlockDriverState
*bs
, void *dev
)
1365 /* TODO change to DeviceState *dev when all users are qdevified */
1367 assert(bs
->dev
== dev
);
1370 bs
->dev_opaque
= NULL
;
1371 bs
->buffer_alignment
= 512;
1374 /* TODO change to return DeviceState * when all users are qdevified */
1375 void *bdrv_get_attached_dev(BlockDriverState
*bs
)
1380 void bdrv_set_dev_ops(BlockDriverState
*bs
, const BlockDevOps
*ops
,
1384 bs
->dev_opaque
= opaque
;
1385 if (bdrv_dev_has_removable_media(bs
) && bs
== bs_snapshots
) {
1386 bs_snapshots
= NULL
;
1390 void bdrv_emit_qmp_error_event(const BlockDriverState
*bdrv
,
1391 enum MonitorEvent ev
,
1392 BlockErrorAction action
, bool is_read
)
1395 const char *action_str
;
1398 case BDRV_ACTION_REPORT
:
1399 action_str
= "report";
1401 case BDRV_ACTION_IGNORE
:
1402 action_str
= "ignore";
1404 case BDRV_ACTION_STOP
:
1405 action_str
= "stop";
1411 data
= qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1414 is_read
? "read" : "write");
1415 monitor_protocol_event(ev
, data
);
1417 qobject_decref(data
);
1420 static void bdrv_emit_qmp_eject_event(BlockDriverState
*bs
, bool ejected
)
1424 data
= qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }",
1425 bdrv_get_device_name(bs
), ejected
);
1426 monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED
, data
);
1428 qobject_decref(data
);
1431 static void bdrv_dev_change_media_cb(BlockDriverState
*bs
, bool load
)
1433 if (bs
->dev_ops
&& bs
->dev_ops
->change_media_cb
) {
1434 bool tray_was_closed
= !bdrv_dev_is_tray_open(bs
);
1435 bs
->dev_ops
->change_media_cb(bs
->dev_opaque
, load
);
1436 if (tray_was_closed
) {
1438 bdrv_emit_qmp_eject_event(bs
, true);
1442 bdrv_emit_qmp_eject_event(bs
, false);
1447 bool bdrv_dev_has_removable_media(BlockDriverState
*bs
)
1449 return !bs
->dev
|| (bs
->dev_ops
&& bs
->dev_ops
->change_media_cb
);
1452 void bdrv_dev_eject_request(BlockDriverState
*bs
, bool force
)
1454 if (bs
->dev_ops
&& bs
->dev_ops
->eject_request_cb
) {
1455 bs
->dev_ops
->eject_request_cb(bs
->dev_opaque
, force
);
1459 bool bdrv_dev_is_tray_open(BlockDriverState
*bs
)
1461 if (bs
->dev_ops
&& bs
->dev_ops
->is_tray_open
) {
1462 return bs
->dev_ops
->is_tray_open(bs
->dev_opaque
);
1467 static void bdrv_dev_resize_cb(BlockDriverState
*bs
)
1469 if (bs
->dev_ops
&& bs
->dev_ops
->resize_cb
) {
1470 bs
->dev_ops
->resize_cb(bs
->dev_opaque
);
1474 bool bdrv_dev_is_medium_locked(BlockDriverState
*bs
)
1476 if (bs
->dev_ops
&& bs
->dev_ops
->is_medium_locked
) {
1477 return bs
->dev_ops
->is_medium_locked(bs
->dev_opaque
);
1483 * Run consistency checks on an image
1485 * Returns 0 if the check could be completed (it doesn't mean that the image is
1486 * free of errors) or -errno when an internal error occurred. The results of the
1487 * check are stored in res.
1489 int bdrv_check(BlockDriverState
*bs
, BdrvCheckResult
*res
, BdrvCheckMode fix
)
1491 if (bs
->drv
->bdrv_check
== NULL
) {
1495 memset(res
, 0, sizeof(*res
));
1496 return bs
->drv
->bdrv_check(bs
, res
, fix
);
1499 #define COMMIT_BUF_SECTORS 2048
1501 /* commit COW file into the raw image */
1502 int bdrv_commit(BlockDriverState
*bs
)
1504 BlockDriver
*drv
= bs
->drv
;
1505 int64_t sector
, total_sectors
;
1506 int n
, ro
, open_flags
;
1509 char filename
[1024];
1514 if (!bs
->backing_hd
) {
1518 if (bdrv_in_use(bs
) || bdrv_in_use(bs
->backing_hd
)) {
1522 ro
= bs
->backing_hd
->read_only
;
1523 strncpy(filename
, bs
->backing_hd
->filename
, sizeof(filename
));
1524 open_flags
= bs
->backing_hd
->open_flags
;
1527 if (bdrv_reopen(bs
->backing_hd
, open_flags
| BDRV_O_RDWR
, NULL
)) {
1532 total_sectors
= bdrv_getlength(bs
) >> BDRV_SECTOR_BITS
;
1533 buf
= g_malloc(COMMIT_BUF_SECTORS
* BDRV_SECTOR_SIZE
);
1535 for (sector
= 0; sector
< total_sectors
; sector
+= n
) {
1536 if (bdrv_is_allocated(bs
, sector
, COMMIT_BUF_SECTORS
, &n
)) {
1538 if (bdrv_read(bs
, sector
, buf
, n
) != 0) {
1543 if (bdrv_write(bs
->backing_hd
, sector
, buf
, n
) != 0) {
1550 if (drv
->bdrv_make_empty
) {
1551 ret
= drv
->bdrv_make_empty(bs
);
1556 * Make sure all data we wrote to the backing device is actually
1560 bdrv_flush(bs
->backing_hd
);
1566 /* ignoring error return here */
1567 bdrv_reopen(bs
->backing_hd
, open_flags
& ~BDRV_O_RDWR
, NULL
);
1573 int bdrv_commit_all(void)
1575 BlockDriverState
*bs
;
1577 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1578 int ret
= bdrv_commit(bs
);
1586 struct BdrvTrackedRequest
{
1587 BlockDriverState
*bs
;
1591 QLIST_ENTRY(BdrvTrackedRequest
) list
;
1592 Coroutine
*co
; /* owner, used for deadlock detection */
1593 CoQueue wait_queue
; /* coroutines blocked on this request */
1597 * Remove an active request from the tracked requests list
1599 * This function should be called when a tracked request is completing.
1601 static void tracked_request_end(BdrvTrackedRequest
*req
)
1603 QLIST_REMOVE(req
, list
);
1604 qemu_co_queue_restart_all(&req
->wait_queue
);
1608 * Add an active request to the tracked requests list
1610 static void tracked_request_begin(BdrvTrackedRequest
*req
,
1611 BlockDriverState
*bs
,
1613 int nb_sectors
, bool is_write
)
1615 *req
= (BdrvTrackedRequest
){
1617 .sector_num
= sector_num
,
1618 .nb_sectors
= nb_sectors
,
1619 .is_write
= is_write
,
1620 .co
= qemu_coroutine_self(),
1623 qemu_co_queue_init(&req
->wait_queue
);
1625 QLIST_INSERT_HEAD(&bs
->tracked_requests
, req
, list
);
1629 * Round a region to cluster boundaries
1631 static void round_to_clusters(BlockDriverState
*bs
,
1632 int64_t sector_num
, int nb_sectors
,
1633 int64_t *cluster_sector_num
,
1634 int *cluster_nb_sectors
)
1636 BlockDriverInfo bdi
;
1638 if (bdrv_get_info(bs
, &bdi
) < 0 || bdi
.cluster_size
== 0) {
1639 *cluster_sector_num
= sector_num
;
1640 *cluster_nb_sectors
= nb_sectors
;
1642 int64_t c
= bdi
.cluster_size
/ BDRV_SECTOR_SIZE
;
1643 *cluster_sector_num
= QEMU_ALIGN_DOWN(sector_num
, c
);
1644 *cluster_nb_sectors
= QEMU_ALIGN_UP(sector_num
- *cluster_sector_num
+
1649 static bool tracked_request_overlaps(BdrvTrackedRequest
*req
,
1650 int64_t sector_num
, int nb_sectors
) {
1652 if (sector_num
>= req
->sector_num
+ req
->nb_sectors
) {
1656 if (req
->sector_num
>= sector_num
+ nb_sectors
) {
1662 static void coroutine_fn
wait_for_overlapping_requests(BlockDriverState
*bs
,
1663 int64_t sector_num
, int nb_sectors
)
1665 BdrvTrackedRequest
*req
;
1666 int64_t cluster_sector_num
;
1667 int cluster_nb_sectors
;
1670 /* If we touch the same cluster it counts as an overlap. This guarantees
1671 * that allocating writes will be serialized and not race with each other
1672 * for the same cluster. For example, in copy-on-read it ensures that the
1673 * CoR read and write operations are atomic and guest writes cannot
1674 * interleave between them.
1676 round_to_clusters(bs
, sector_num
, nb_sectors
,
1677 &cluster_sector_num
, &cluster_nb_sectors
);
1681 QLIST_FOREACH(req
, &bs
->tracked_requests
, list
) {
1682 if (tracked_request_overlaps(req
, cluster_sector_num
,
1683 cluster_nb_sectors
)) {
1684 /* Hitting this means there was a reentrant request, for
1685 * example, a block driver issuing nested requests. This must
1686 * never happen since it means deadlock.
1688 assert(qemu_coroutine_self() != req
->co
);
1690 qemu_co_queue_wait(&req
->wait_queue
);
1701 * -EINVAL - backing format specified, but no file
1702 * -ENOSPC - can't update the backing file because no space is left in the
1704 * -ENOTSUP - format driver doesn't support changing the backing file
1706 int bdrv_change_backing_file(BlockDriverState
*bs
,
1707 const char *backing_file
, const char *backing_fmt
)
1709 BlockDriver
*drv
= bs
->drv
;
1712 /* Backing file format doesn't make sense without a backing file */
1713 if (backing_fmt
&& !backing_file
) {
1717 if (drv
->bdrv_change_backing_file
!= NULL
) {
1718 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
1724 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
1725 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
1731 * Finds the image layer in the chain that has 'bs' as its backing file.
1733 * active is the current topmost image.
1735 * Returns NULL if bs is not found in active's image chain,
1736 * or if active == bs.
1738 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
1739 BlockDriverState
*bs
)
1741 BlockDriverState
*overlay
= NULL
;
1742 BlockDriverState
*intermediate
;
1744 assert(active
!= NULL
);
1747 /* if bs is the same as active, then by definition it has no overlay
1753 intermediate
= active
;
1754 while (intermediate
->backing_hd
) {
1755 if (intermediate
->backing_hd
== bs
) {
1756 overlay
= intermediate
;
1759 intermediate
= intermediate
->backing_hd
;
1765 typedef struct BlkIntermediateStates
{
1766 BlockDriverState
*bs
;
1767 QSIMPLEQ_ENTRY(BlkIntermediateStates
) entry
;
1768 } BlkIntermediateStates
;
1772 * Drops images above 'base' up to and including 'top', and sets the image
1773 * above 'top' to have base as its backing file.
1775 * Requires that the overlay to 'top' is opened r/w, so that the backing file
1776 * information in 'bs' can be properly updated.
1778 * E.g., this will convert the following chain:
1779 * bottom <- base <- intermediate <- top <- active
1783 * bottom <- base <- active
1785 * It is allowed for bottom==base, in which case it converts:
1787 * base <- intermediate <- top <- active
1794 * if active == top, that is considered an error
1797 int bdrv_drop_intermediate(BlockDriverState
*active
, BlockDriverState
*top
,
1798 BlockDriverState
*base
)
1800 BlockDriverState
*intermediate
;
1801 BlockDriverState
*base_bs
= NULL
;
1802 BlockDriverState
*new_top_bs
= NULL
;
1803 BlkIntermediateStates
*intermediate_state
, *next
;
1806 QSIMPLEQ_HEAD(states_to_delete
, BlkIntermediateStates
) states_to_delete
;
1807 QSIMPLEQ_INIT(&states_to_delete
);
1809 if (!top
->drv
|| !base
->drv
) {
1813 new_top_bs
= bdrv_find_overlay(active
, top
);
1815 if (new_top_bs
== NULL
) {
1816 /* we could not find the image above 'top', this is an error */
1820 /* special case of new_top_bs->backing_hd already pointing to base - nothing
1821 * to do, no intermediate images */
1822 if (new_top_bs
->backing_hd
== base
) {
1829 /* now we will go down through the list, and add each BDS we find
1830 * into our deletion queue, until we hit the 'base'
1832 while (intermediate
) {
1833 intermediate_state
= g_malloc0(sizeof(BlkIntermediateStates
));
1834 intermediate_state
->bs
= intermediate
;
1835 QSIMPLEQ_INSERT_TAIL(&states_to_delete
, intermediate_state
, entry
);
1837 if (intermediate
->backing_hd
== base
) {
1838 base_bs
= intermediate
->backing_hd
;
1841 intermediate
= intermediate
->backing_hd
;
1843 if (base_bs
== NULL
) {
1844 /* something went wrong, we did not end at the base. safely
1845 * unravel everything, and exit with error */
1849 /* success - we can delete the intermediate states, and link top->base */
1850 ret
= bdrv_change_backing_file(new_top_bs
, base_bs
->filename
,
1851 base_bs
->drv
? base_bs
->drv
->format_name
: "");
1855 new_top_bs
->backing_hd
= base_bs
;
1858 QSIMPLEQ_FOREACH_SAFE(intermediate_state
, &states_to_delete
, entry
, next
) {
1859 /* so that bdrv_close() does not recursively close the chain */
1860 intermediate_state
->bs
->backing_hd
= NULL
;
1861 bdrv_delete(intermediate_state
->bs
);
1866 QSIMPLEQ_FOREACH_SAFE(intermediate_state
, &states_to_delete
, entry
, next
) {
1867 g_free(intermediate_state
);
1873 static int bdrv_check_byte_request(BlockDriverState
*bs
, int64_t offset
,
1878 if (!bdrv_is_inserted(bs
))
1884 len
= bdrv_getlength(bs
);
1889 if ((offset
> len
) || (len
- offset
< size
))
1895 static int bdrv_check_request(BlockDriverState
*bs
, int64_t sector_num
,
1898 return bdrv_check_byte_request(bs
, sector_num
* BDRV_SECTOR_SIZE
,
1899 nb_sectors
* BDRV_SECTOR_SIZE
);
1902 typedef struct RwCo
{
1903 BlockDriverState
*bs
;
1911 static void coroutine_fn
bdrv_rw_co_entry(void *opaque
)
1913 RwCo
*rwco
= opaque
;
1915 if (!rwco
->is_write
) {
1916 rwco
->ret
= bdrv_co_do_readv(rwco
->bs
, rwco
->sector_num
,
1917 rwco
->nb_sectors
, rwco
->qiov
, 0);
1919 rwco
->ret
= bdrv_co_do_writev(rwco
->bs
, rwco
->sector_num
,
1920 rwco
->nb_sectors
, rwco
->qiov
, 0);
1925 * Process a synchronous request using coroutines
1927 static int bdrv_rw_co(BlockDriverState
*bs
, int64_t sector_num
, uint8_t *buf
,
1928 int nb_sectors
, bool is_write
)
1931 struct iovec iov
= {
1932 .iov_base
= (void *)buf
,
1933 .iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
,
1938 .sector_num
= sector_num
,
1939 .nb_sectors
= nb_sectors
,
1941 .is_write
= is_write
,
1945 qemu_iovec_init_external(&qiov
, &iov
, 1);
1948 * In sync call context, when the vcpu is blocked, this throttling timer
1949 * will not fire; so the I/O throttling function has to be disabled here
1950 * if it has been enabled.
1952 if (bs
->io_limits_enabled
) {
1953 fprintf(stderr
, "Disabling I/O throttling on '%s' due "
1954 "to synchronous I/O.\n", bdrv_get_device_name(bs
));
1955 bdrv_io_limits_disable(bs
);
1958 if (qemu_in_coroutine()) {
1959 /* Fast-path if already in coroutine context */
1960 bdrv_rw_co_entry(&rwco
);
1962 co
= qemu_coroutine_create(bdrv_rw_co_entry
);
1963 qemu_coroutine_enter(co
, &rwco
);
1964 while (rwco
.ret
== NOT_DONE
) {
1971 /* return < 0 if error. See bdrv_write() for the return codes */
1972 int bdrv_read(BlockDriverState
*bs
, int64_t sector_num
,
1973 uint8_t *buf
, int nb_sectors
)
1975 return bdrv_rw_co(bs
, sector_num
, buf
, nb_sectors
, false);
1978 /* Just like bdrv_read(), but with I/O throttling temporarily disabled */
1979 int bdrv_read_unthrottled(BlockDriverState
*bs
, int64_t sector_num
,
1980 uint8_t *buf
, int nb_sectors
)
1985 enabled
= bs
->io_limits_enabled
;
1986 bs
->io_limits_enabled
= false;
1987 ret
= bdrv_read(bs
, 0, buf
, 1);
1988 bs
->io_limits_enabled
= enabled
;
1992 #define BITS_PER_LONG (sizeof(unsigned long) * 8)
1994 static void set_dirty_bitmap(BlockDriverState
*bs
, int64_t sector_num
,
1995 int nb_sectors
, int dirty
)
1998 unsigned long val
, idx
, bit
;
2000 start
= sector_num
/ BDRV_SECTORS_PER_DIRTY_CHUNK
;
2001 end
= (sector_num
+ nb_sectors
- 1) / BDRV_SECTORS_PER_DIRTY_CHUNK
;
2003 for (; start
<= end
; start
++) {
2004 idx
= start
/ BITS_PER_LONG
;
2005 bit
= start
% BITS_PER_LONG
;
2006 val
= bs
->dirty_bitmap
[idx
];
2008 if (!(val
& (1UL << bit
))) {
2013 if (val
& (1UL << bit
)) {
2015 val
&= ~(1UL << bit
);
2018 bs
->dirty_bitmap
[idx
] = val
;
2022 /* Return < 0 if error. Important errors are:
2023 -EIO generic I/O error (may happen for all errors)
2024 -ENOMEDIUM No media inserted.
2025 -EINVAL Invalid sector number or nb_sectors
2026 -EACCES Trying to write a read-only device
2028 int bdrv_write(BlockDriverState
*bs
, int64_t sector_num
,
2029 const uint8_t *buf
, int nb_sectors
)
2031 return bdrv_rw_co(bs
, sector_num
, (uint8_t *)buf
, nb_sectors
, true);
2034 int bdrv_pread(BlockDriverState
*bs
, int64_t offset
,
2035 void *buf
, int count1
)
2037 uint8_t tmp_buf
[BDRV_SECTOR_SIZE
];
2038 int len
, nb_sectors
, count
;
2043 /* first read to align to sector start */
2044 len
= (BDRV_SECTOR_SIZE
- offset
) & (BDRV_SECTOR_SIZE
- 1);
2047 sector_num
= offset
>> BDRV_SECTOR_BITS
;
2049 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
2051 memcpy(buf
, tmp_buf
+ (offset
& (BDRV_SECTOR_SIZE
- 1)), len
);
2059 /* read the sectors "in place" */
2060 nb_sectors
= count
>> BDRV_SECTOR_BITS
;
2061 if (nb_sectors
> 0) {
2062 if ((ret
= bdrv_read(bs
, sector_num
, buf
, nb_sectors
)) < 0)
2064 sector_num
+= nb_sectors
;
2065 len
= nb_sectors
<< BDRV_SECTOR_BITS
;
2070 /* add data from the last sector */
2072 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
2074 memcpy(buf
, tmp_buf
, count
);
2079 int bdrv_pwrite(BlockDriverState
*bs
, int64_t offset
,
2080 const void *buf
, int count1
)
2082 uint8_t tmp_buf
[BDRV_SECTOR_SIZE
];
2083 int len
, nb_sectors
, count
;
2088 /* first write to align to sector start */
2089 len
= (BDRV_SECTOR_SIZE
- offset
) & (BDRV_SECTOR_SIZE
- 1);
2092 sector_num
= offset
>> BDRV_SECTOR_BITS
;
2094 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
2096 memcpy(tmp_buf
+ (offset
& (BDRV_SECTOR_SIZE
- 1)), buf
, len
);
2097 if ((ret
= bdrv_write(bs
, sector_num
, tmp_buf
, 1)) < 0)
2106 /* write the sectors "in place" */
2107 nb_sectors
= count
>> BDRV_SECTOR_BITS
;
2108 if (nb_sectors
> 0) {
2109 if ((ret
= bdrv_write(bs
, sector_num
, buf
, nb_sectors
)) < 0)
2111 sector_num
+= nb_sectors
;
2112 len
= nb_sectors
<< BDRV_SECTOR_BITS
;
2117 /* add data from the last sector */
2119 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
2121 memcpy(tmp_buf
, buf
, count
);
2122 if ((ret
= bdrv_write(bs
, sector_num
, tmp_buf
, 1)) < 0)
2129 * Writes to the file and ensures that no writes are reordered across this
2130 * request (acts as a barrier)
2132 * Returns 0 on success, -errno in error cases.
2134 int bdrv_pwrite_sync(BlockDriverState
*bs
, int64_t offset
,
2135 const void *buf
, int count
)
2139 ret
= bdrv_pwrite(bs
, offset
, buf
, count
);
2144 /* No flush needed for cache modes that already do it */
2145 if (bs
->enable_write_cache
) {
2152 static int coroutine_fn
bdrv_co_do_copy_on_readv(BlockDriverState
*bs
,
2153 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
2155 /* Perform I/O through a temporary buffer so that users who scribble over
2156 * their read buffer while the operation is in progress do not end up
2157 * modifying the image file. This is critical for zero-copy guest I/O
2158 * where anything might happen inside guest memory.
2160 void *bounce_buffer
;
2162 BlockDriver
*drv
= bs
->drv
;
2164 QEMUIOVector bounce_qiov
;
2165 int64_t cluster_sector_num
;
2166 int cluster_nb_sectors
;
2170 /* Cover entire cluster so no additional backing file I/O is required when
2171 * allocating cluster in the image file.
2173 round_to_clusters(bs
, sector_num
, nb_sectors
,
2174 &cluster_sector_num
, &cluster_nb_sectors
);
2176 trace_bdrv_co_do_copy_on_readv(bs
, sector_num
, nb_sectors
,
2177 cluster_sector_num
, cluster_nb_sectors
);
2179 iov
.iov_len
= cluster_nb_sectors
* BDRV_SECTOR_SIZE
;
2180 iov
.iov_base
= bounce_buffer
= qemu_blockalign(bs
, iov
.iov_len
);
2181 qemu_iovec_init_external(&bounce_qiov
, &iov
, 1);
2183 ret
= drv
->bdrv_co_readv(bs
, cluster_sector_num
, cluster_nb_sectors
,
2189 if (drv
->bdrv_co_write_zeroes
&&
2190 buffer_is_zero(bounce_buffer
, iov
.iov_len
)) {
2191 ret
= bdrv_co_do_write_zeroes(bs
, cluster_sector_num
,
2192 cluster_nb_sectors
);
2194 /* This does not change the data on the disk, it is not necessary
2195 * to flush even in cache=writethrough mode.
2197 ret
= drv
->bdrv_co_writev(bs
, cluster_sector_num
, cluster_nb_sectors
,
2202 /* It might be okay to ignore write errors for guest requests. If this
2203 * is a deliberate copy-on-read then we don't want to ignore the error.
2204 * Simply report it in all cases.
2209 skip_bytes
= (sector_num
- cluster_sector_num
) * BDRV_SECTOR_SIZE
;
2210 qemu_iovec_from_buf(qiov
, 0, bounce_buffer
+ skip_bytes
,
2211 nb_sectors
* BDRV_SECTOR_SIZE
);
2214 qemu_vfree(bounce_buffer
);
2219 * Handle a read request in coroutine context
2221 static int coroutine_fn
bdrv_co_do_readv(BlockDriverState
*bs
,
2222 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
2223 BdrvRequestFlags flags
)
2225 BlockDriver
*drv
= bs
->drv
;
2226 BdrvTrackedRequest req
;
2232 if (bdrv_check_request(bs
, sector_num
, nb_sectors
)) {
2236 /* throttling disk read I/O */
2237 if (bs
->io_limits_enabled
) {
2238 bdrv_io_limits_intercept(bs
, false, nb_sectors
);
2241 if (bs
->copy_on_read
) {
2242 flags
|= BDRV_REQ_COPY_ON_READ
;
2244 if (flags
& BDRV_REQ_COPY_ON_READ
) {
2245 bs
->copy_on_read_in_flight
++;
2248 if (bs
->copy_on_read_in_flight
) {
2249 wait_for_overlapping_requests(bs
, sector_num
, nb_sectors
);
2252 tracked_request_begin(&req
, bs
, sector_num
, nb_sectors
, false);
2254 if (flags
& BDRV_REQ_COPY_ON_READ
) {
2257 ret
= bdrv_co_is_allocated(bs
, sector_num
, nb_sectors
, &pnum
);
2262 if (!ret
|| pnum
!= nb_sectors
) {
2263 ret
= bdrv_co_do_copy_on_readv(bs
, sector_num
, nb_sectors
, qiov
);
2268 ret
= drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
2271 tracked_request_end(&req
);
2273 if (flags
& BDRV_REQ_COPY_ON_READ
) {
2274 bs
->copy_on_read_in_flight
--;
2280 int coroutine_fn
bdrv_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
2281 int nb_sectors
, QEMUIOVector
*qiov
)
2283 trace_bdrv_co_readv(bs
, sector_num
, nb_sectors
);
2285 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
, 0);
2288 int coroutine_fn
bdrv_co_copy_on_readv(BlockDriverState
*bs
,
2289 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
2291 trace_bdrv_co_copy_on_readv(bs
, sector_num
, nb_sectors
);
2293 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
,
2294 BDRV_REQ_COPY_ON_READ
);
2297 static int coroutine_fn
bdrv_co_do_write_zeroes(BlockDriverState
*bs
,
2298 int64_t sector_num
, int nb_sectors
)
2300 BlockDriver
*drv
= bs
->drv
;
2305 /* TODO Emulate only part of misaligned requests instead of letting block
2306 * drivers return -ENOTSUP and emulate everything */
2308 /* First try the efficient write zeroes operation */
2309 if (drv
->bdrv_co_write_zeroes
) {
2310 ret
= drv
->bdrv_co_write_zeroes(bs
, sector_num
, nb_sectors
);
2311 if (ret
!= -ENOTSUP
) {
2316 /* Fall back to bounce buffer if write zeroes is unsupported */
2317 iov
.iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
;
2318 iov
.iov_base
= qemu_blockalign(bs
, iov
.iov_len
);
2319 memset(iov
.iov_base
, 0, iov
.iov_len
);
2320 qemu_iovec_init_external(&qiov
, &iov
, 1);
2322 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, &qiov
);
2324 qemu_vfree(iov
.iov_base
);
2329 * Handle a write request in coroutine context
2331 static int coroutine_fn
bdrv_co_do_writev(BlockDriverState
*bs
,
2332 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
2333 BdrvRequestFlags flags
)
2335 BlockDriver
*drv
= bs
->drv
;
2336 BdrvTrackedRequest req
;
2342 if (bs
->read_only
) {
2345 if (bdrv_check_request(bs
, sector_num
, nb_sectors
)) {
2349 /* throttling disk write I/O */
2350 if (bs
->io_limits_enabled
) {
2351 bdrv_io_limits_intercept(bs
, true, nb_sectors
);
2354 if (bs
->copy_on_read_in_flight
) {
2355 wait_for_overlapping_requests(bs
, sector_num
, nb_sectors
);
2358 tracked_request_begin(&req
, bs
, sector_num
, nb_sectors
, true);
2360 if (flags
& BDRV_REQ_ZERO_WRITE
) {
2361 ret
= bdrv_co_do_write_zeroes(bs
, sector_num
, nb_sectors
);
2363 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, qiov
);
2366 if (ret
== 0 && !bs
->enable_write_cache
) {
2367 ret
= bdrv_co_flush(bs
);
2370 if (bs
->dirty_bitmap
) {
2371 set_dirty_bitmap(bs
, sector_num
, nb_sectors
, 1);
2374 if (bs
->wr_highest_sector
< sector_num
+ nb_sectors
- 1) {
2375 bs
->wr_highest_sector
= sector_num
+ nb_sectors
- 1;
2378 tracked_request_end(&req
);
2383 int coroutine_fn
bdrv_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
2384 int nb_sectors
, QEMUIOVector
*qiov
)
2386 trace_bdrv_co_writev(bs
, sector_num
, nb_sectors
);
2388 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, qiov
, 0);
2391 int coroutine_fn
bdrv_co_write_zeroes(BlockDriverState
*bs
,
2392 int64_t sector_num
, int nb_sectors
)
2394 trace_bdrv_co_write_zeroes(bs
, sector_num
, nb_sectors
);
2396 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, NULL
,
2397 BDRV_REQ_ZERO_WRITE
);
2401 * Truncate file to 'offset' bytes (needed only for file protocols)
2403 int bdrv_truncate(BlockDriverState
*bs
, int64_t offset
)
2405 BlockDriver
*drv
= bs
->drv
;
2409 if (!drv
->bdrv_truncate
)
2413 if (bdrv_in_use(bs
))
2415 ret
= drv
->bdrv_truncate(bs
, offset
);
2417 ret
= refresh_total_sectors(bs
, offset
>> BDRV_SECTOR_BITS
);
2418 bdrv_dev_resize_cb(bs
);
2424 * Length of a allocated file in bytes. Sparse files are counted by actual
2425 * allocated space. Return < 0 if error or unknown.
2427 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
2429 BlockDriver
*drv
= bs
->drv
;
2433 if (drv
->bdrv_get_allocated_file_size
) {
2434 return drv
->bdrv_get_allocated_file_size(bs
);
2437 return bdrv_get_allocated_file_size(bs
->file
);
2443 * Length of a file in bytes. Return < 0 if error or unknown.
2445 int64_t bdrv_getlength(BlockDriverState
*bs
)
2447 BlockDriver
*drv
= bs
->drv
;
2451 if (bs
->growable
|| bdrv_dev_has_removable_media(bs
)) {
2452 if (drv
->bdrv_getlength
) {
2453 return drv
->bdrv_getlength(bs
);
2456 return bs
->total_sectors
* BDRV_SECTOR_SIZE
;
2459 /* return 0 as number of sectors if no device present or error */
2460 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
2463 length
= bdrv_getlength(bs
);
2467 length
= length
>> BDRV_SECTOR_BITS
;
2468 *nb_sectors_ptr
= length
;
2471 /* throttling disk io limits */
2472 void bdrv_set_io_limits(BlockDriverState
*bs
,
2473 BlockIOLimit
*io_limits
)
2475 bs
->io_limits
= *io_limits
;
2476 bs
->io_limits_enabled
= bdrv_io_limits_enabled(bs
);
2479 void bdrv_set_on_error(BlockDriverState
*bs
, BlockdevOnError on_read_error
,
2480 BlockdevOnError on_write_error
)
2482 bs
->on_read_error
= on_read_error
;
2483 bs
->on_write_error
= on_write_error
;
2486 BlockdevOnError
bdrv_get_on_error(BlockDriverState
*bs
, bool is_read
)
2488 return is_read
? bs
->on_read_error
: bs
->on_write_error
;
2491 BlockErrorAction
bdrv_get_error_action(BlockDriverState
*bs
, bool is_read
, int error
)
2493 BlockdevOnError on_err
= is_read
? bs
->on_read_error
: bs
->on_write_error
;
2496 case BLOCKDEV_ON_ERROR_ENOSPC
:
2497 return (error
== ENOSPC
) ? BDRV_ACTION_STOP
: BDRV_ACTION_REPORT
;
2498 case BLOCKDEV_ON_ERROR_STOP
:
2499 return BDRV_ACTION_STOP
;
2500 case BLOCKDEV_ON_ERROR_REPORT
:
2501 return BDRV_ACTION_REPORT
;
2502 case BLOCKDEV_ON_ERROR_IGNORE
:
2503 return BDRV_ACTION_IGNORE
;
2509 /* This is done by device models because, while the block layer knows
2510 * about the error, it does not know whether an operation comes from
2511 * the device or the block layer (from a job, for example).
2513 void bdrv_error_action(BlockDriverState
*bs
, BlockErrorAction action
,
2514 bool is_read
, int error
)
2517 bdrv_emit_qmp_error_event(bs
, QEVENT_BLOCK_IO_ERROR
, action
, is_read
);
2518 if (action
== BDRV_ACTION_STOP
) {
2519 vm_stop(RUN_STATE_IO_ERROR
);
2520 bdrv_iostatus_set_err(bs
, error
);
2524 int bdrv_is_read_only(BlockDriverState
*bs
)
2526 return bs
->read_only
;
2529 int bdrv_is_sg(BlockDriverState
*bs
)
2534 int bdrv_enable_write_cache(BlockDriverState
*bs
)
2536 return bs
->enable_write_cache
;
2539 void bdrv_set_enable_write_cache(BlockDriverState
*bs
, bool wce
)
2541 bs
->enable_write_cache
= wce
;
2543 /* so a reopen() will preserve wce */
2545 bs
->open_flags
|= BDRV_O_CACHE_WB
;
2547 bs
->open_flags
&= ~BDRV_O_CACHE_WB
;
2551 int bdrv_is_encrypted(BlockDriverState
*bs
)
2553 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
)
2555 return bs
->encrypted
;
2558 int bdrv_key_required(BlockDriverState
*bs
)
2560 BlockDriverState
*backing_hd
= bs
->backing_hd
;
2562 if (backing_hd
&& backing_hd
->encrypted
&& !backing_hd
->valid_key
)
2564 return (bs
->encrypted
&& !bs
->valid_key
);
2567 int bdrv_set_key(BlockDriverState
*bs
, const char *key
)
2570 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
) {
2571 ret
= bdrv_set_key(bs
->backing_hd
, key
);
2577 if (!bs
->encrypted
) {
2579 } else if (!bs
->drv
|| !bs
->drv
->bdrv_set_key
) {
2582 ret
= bs
->drv
->bdrv_set_key(bs
, key
);
2585 } else if (!bs
->valid_key
) {
2587 /* call the change callback now, we skipped it on open */
2588 bdrv_dev_change_media_cb(bs
, true);
2593 const char *bdrv_get_format_name(BlockDriverState
*bs
)
2595 return bs
->drv
? bs
->drv
->format_name
: NULL
;
2598 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
2603 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
2604 it(opaque
, drv
->format_name
);
2608 BlockDriverState
*bdrv_find(const char *name
)
2610 BlockDriverState
*bs
;
2612 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2613 if (!strcmp(name
, bs
->device_name
)) {
2620 BlockDriverState
*bdrv_next(BlockDriverState
*bs
)
2623 return QTAILQ_FIRST(&bdrv_states
);
2625 return QTAILQ_NEXT(bs
, list
);
2628 void bdrv_iterate(void (*it
)(void *opaque
, BlockDriverState
*bs
), void *opaque
)
2630 BlockDriverState
*bs
;
2632 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2637 const char *bdrv_get_device_name(BlockDriverState
*bs
)
2639 return bs
->device_name
;
2642 int bdrv_get_flags(BlockDriverState
*bs
)
2644 return bs
->open_flags
;
2647 void bdrv_flush_all(void)
2649 BlockDriverState
*bs
;
2651 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2656 int bdrv_has_zero_init(BlockDriverState
*bs
)
2660 if (bs
->drv
->bdrv_has_zero_init
) {
2661 return bs
->drv
->bdrv_has_zero_init(bs
);
2667 typedef struct BdrvCoIsAllocatedData
{
2668 BlockDriverState
*bs
;
2674 } BdrvCoIsAllocatedData
;
2677 * Returns true iff the specified sector is present in the disk image. Drivers
2678 * not implementing the functionality are assumed to not support backing files,
2679 * hence all their sectors are reported as allocated.
2681 * If 'sector_num' is beyond the end of the disk image the return value is 0
2682 * and 'pnum' is set to 0.
2684 * 'pnum' is set to the number of sectors (including and immediately following
2685 * the specified sector) that are known to be in the same
2686 * allocated/unallocated state.
2688 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
2689 * beyond the end of the disk image it will be clamped.
2691 int coroutine_fn
bdrv_co_is_allocated(BlockDriverState
*bs
, int64_t sector_num
,
2692 int nb_sectors
, int *pnum
)
2696 if (sector_num
>= bs
->total_sectors
) {
2701 n
= bs
->total_sectors
- sector_num
;
2702 if (n
< nb_sectors
) {
2706 if (!bs
->drv
->bdrv_co_is_allocated
) {
2711 return bs
->drv
->bdrv_co_is_allocated(bs
, sector_num
, nb_sectors
, pnum
);
2714 /* Coroutine wrapper for bdrv_is_allocated() */
2715 static void coroutine_fn
bdrv_is_allocated_co_entry(void *opaque
)
2717 BdrvCoIsAllocatedData
*data
= opaque
;
2718 BlockDriverState
*bs
= data
->bs
;
2720 data
->ret
= bdrv_co_is_allocated(bs
, data
->sector_num
, data
->nb_sectors
,
2726 * Synchronous wrapper around bdrv_co_is_allocated().
2728 * See bdrv_co_is_allocated() for details.
2730 int bdrv_is_allocated(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
,
2734 BdrvCoIsAllocatedData data
= {
2736 .sector_num
= sector_num
,
2737 .nb_sectors
= nb_sectors
,
2742 co
= qemu_coroutine_create(bdrv_is_allocated_co_entry
);
2743 qemu_coroutine_enter(co
, &data
);
2744 while (!data
.done
) {
2751 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
2753 * Return true if the given sector is allocated in any image between
2754 * BASE and TOP (inclusive). BASE can be NULL to check if the given
2755 * sector is allocated in any image of the chain. Return false otherwise.
2757 * 'pnum' is set to the number of sectors (including and immediately following
2758 * the specified sector) that are known to be in the same
2759 * allocated/unallocated state.
2762 int coroutine_fn
bdrv_co_is_allocated_above(BlockDriverState
*top
,
2763 BlockDriverState
*base
,
2765 int nb_sectors
, int *pnum
)
2767 BlockDriverState
*intermediate
;
2768 int ret
, n
= nb_sectors
;
2771 while (intermediate
&& intermediate
!= base
) {
2773 ret
= bdrv_co_is_allocated(intermediate
, sector_num
, nb_sectors
,
2783 * [sector_num, nb_sectors] is unallocated on top but intermediate
2786 * [sector_num+x, nr_sectors] allocated.
2788 if (n
> pnum_inter
) {
2792 intermediate
= intermediate
->backing_hd
;
2799 BlockInfoList
*qmp_query_block(Error
**errp
)
2801 BlockInfoList
*head
= NULL
, *cur_item
= NULL
;
2802 BlockDriverState
*bs
;
2804 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2805 BlockInfoList
*info
= g_malloc0(sizeof(*info
));
2807 info
->value
= g_malloc0(sizeof(*info
->value
));
2808 info
->value
->device
= g_strdup(bs
->device_name
);
2809 info
->value
->type
= g_strdup("unknown");
2810 info
->value
->locked
= bdrv_dev_is_medium_locked(bs
);
2811 info
->value
->removable
= bdrv_dev_has_removable_media(bs
);
2813 if (bdrv_dev_has_removable_media(bs
)) {
2814 info
->value
->has_tray_open
= true;
2815 info
->value
->tray_open
= bdrv_dev_is_tray_open(bs
);
2818 if (bdrv_iostatus_is_enabled(bs
)) {
2819 info
->value
->has_io_status
= true;
2820 info
->value
->io_status
= bs
->iostatus
;
2824 info
->value
->has_inserted
= true;
2825 info
->value
->inserted
= g_malloc0(sizeof(*info
->value
->inserted
));
2826 info
->value
->inserted
->file
= g_strdup(bs
->filename
);
2827 info
->value
->inserted
->ro
= bs
->read_only
;
2828 info
->value
->inserted
->drv
= g_strdup(bs
->drv
->format_name
);
2829 info
->value
->inserted
->encrypted
= bs
->encrypted
;
2830 info
->value
->inserted
->encryption_key_missing
= bdrv_key_required(bs
);
2831 if (bs
->backing_file
[0]) {
2832 info
->value
->inserted
->has_backing_file
= true;
2833 info
->value
->inserted
->backing_file
= g_strdup(bs
->backing_file
);
2836 info
->value
->inserted
->backing_file_depth
=
2837 bdrv_get_backing_file_depth(bs
);
2839 if (bs
->io_limits_enabled
) {
2840 info
->value
->inserted
->bps
=
2841 bs
->io_limits
.bps
[BLOCK_IO_LIMIT_TOTAL
];
2842 info
->value
->inserted
->bps_rd
=
2843 bs
->io_limits
.bps
[BLOCK_IO_LIMIT_READ
];
2844 info
->value
->inserted
->bps_wr
=
2845 bs
->io_limits
.bps
[BLOCK_IO_LIMIT_WRITE
];
2846 info
->value
->inserted
->iops
=
2847 bs
->io_limits
.iops
[BLOCK_IO_LIMIT_TOTAL
];
2848 info
->value
->inserted
->iops_rd
=
2849 bs
->io_limits
.iops
[BLOCK_IO_LIMIT_READ
];
2850 info
->value
->inserted
->iops_wr
=
2851 bs
->io_limits
.iops
[BLOCK_IO_LIMIT_WRITE
];
2855 /* XXX: waiting for the qapi to support GSList */
2857 head
= cur_item
= info
;
2859 cur_item
->next
= info
;
2867 /* Consider exposing this as a full fledged QMP command */
2868 static BlockStats
*qmp_query_blockstat(const BlockDriverState
*bs
, Error
**errp
)
2872 s
= g_malloc0(sizeof(*s
));
2874 if (bs
->device_name
[0]) {
2875 s
->has_device
= true;
2876 s
->device
= g_strdup(bs
->device_name
);
2879 s
->stats
= g_malloc0(sizeof(*s
->stats
));
2880 s
->stats
->rd_bytes
= bs
->nr_bytes
[BDRV_ACCT_READ
];
2881 s
->stats
->wr_bytes
= bs
->nr_bytes
[BDRV_ACCT_WRITE
];
2882 s
->stats
->rd_operations
= bs
->nr_ops
[BDRV_ACCT_READ
];
2883 s
->stats
->wr_operations
= bs
->nr_ops
[BDRV_ACCT_WRITE
];
2884 s
->stats
->wr_highest_offset
= bs
->wr_highest_sector
* BDRV_SECTOR_SIZE
;
2885 s
->stats
->flush_operations
= bs
->nr_ops
[BDRV_ACCT_FLUSH
];
2886 s
->stats
->wr_total_time_ns
= bs
->total_time_ns
[BDRV_ACCT_WRITE
];
2887 s
->stats
->rd_total_time_ns
= bs
->total_time_ns
[BDRV_ACCT_READ
];
2888 s
->stats
->flush_total_time_ns
= bs
->total_time_ns
[BDRV_ACCT_FLUSH
];
2891 s
->has_parent
= true;
2892 s
->parent
= qmp_query_blockstat(bs
->file
, NULL
);
2898 BlockStatsList
*qmp_query_blockstats(Error
**errp
)
2900 BlockStatsList
*head
= NULL
, *cur_item
= NULL
;
2901 BlockDriverState
*bs
;
2903 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2904 BlockStatsList
*info
= g_malloc0(sizeof(*info
));
2905 info
->value
= qmp_query_blockstat(bs
, NULL
);
2907 /* XXX: waiting for the qapi to support GSList */
2909 head
= cur_item
= info
;
2911 cur_item
->next
= info
;
2919 const char *bdrv_get_encrypted_filename(BlockDriverState
*bs
)
2921 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
)
2922 return bs
->backing_file
;
2923 else if (bs
->encrypted
)
2924 return bs
->filename
;
2929 void bdrv_get_backing_filename(BlockDriverState
*bs
,
2930 char *filename
, int filename_size
)
2932 pstrcpy(filename
, filename_size
, bs
->backing_file
);
2935 int bdrv_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
2936 const uint8_t *buf
, int nb_sectors
)
2938 BlockDriver
*drv
= bs
->drv
;
2941 if (!drv
->bdrv_write_compressed
)
2943 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
2946 if (bs
->dirty_bitmap
) {
2947 set_dirty_bitmap(bs
, sector_num
, nb_sectors
, 1);
2950 return drv
->bdrv_write_compressed(bs
, sector_num
, buf
, nb_sectors
);
2953 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2955 BlockDriver
*drv
= bs
->drv
;
2958 if (!drv
->bdrv_get_info
)
2960 memset(bdi
, 0, sizeof(*bdi
));
2961 return drv
->bdrv_get_info(bs
, bdi
);
2964 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
2965 int64_t pos
, int size
)
2967 BlockDriver
*drv
= bs
->drv
;
2970 if (drv
->bdrv_save_vmstate
)
2971 return drv
->bdrv_save_vmstate(bs
, buf
, pos
, size
);
2973 return bdrv_save_vmstate(bs
->file
, buf
, pos
, size
);
2977 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
2978 int64_t pos
, int size
)
2980 BlockDriver
*drv
= bs
->drv
;
2983 if (drv
->bdrv_load_vmstate
)
2984 return drv
->bdrv_load_vmstate(bs
, buf
, pos
, size
);
2986 return bdrv_load_vmstate(bs
->file
, buf
, pos
, size
);
2990 void bdrv_debug_event(BlockDriverState
*bs
, BlkDebugEvent event
)
2992 BlockDriver
*drv
= bs
->drv
;
2994 if (!drv
|| !drv
->bdrv_debug_event
) {
2998 drv
->bdrv_debug_event(bs
, event
);
3002 /**************************************************************/
3003 /* handling of snapshots */
3005 int bdrv_can_snapshot(BlockDriverState
*bs
)
3007 BlockDriver
*drv
= bs
->drv
;
3008 if (!drv
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
)) {
3012 if (!drv
->bdrv_snapshot_create
) {
3013 if (bs
->file
!= NULL
) {
3014 return bdrv_can_snapshot(bs
->file
);
3022 int bdrv_is_snapshot(BlockDriverState
*bs
)
3024 return !!(bs
->open_flags
& BDRV_O_SNAPSHOT
);
3027 BlockDriverState
*bdrv_snapshots(void)
3029 BlockDriverState
*bs
;
3032 return bs_snapshots
;
3036 while ((bs
= bdrv_next(bs
))) {
3037 if (bdrv_can_snapshot(bs
)) {
3045 int bdrv_snapshot_create(BlockDriverState
*bs
,
3046 QEMUSnapshotInfo
*sn_info
)
3048 BlockDriver
*drv
= bs
->drv
;
3051 if (drv
->bdrv_snapshot_create
)
3052 return drv
->bdrv_snapshot_create(bs
, sn_info
);
3054 return bdrv_snapshot_create(bs
->file
, sn_info
);
3058 int bdrv_snapshot_goto(BlockDriverState
*bs
,
3059 const char *snapshot_id
)
3061 BlockDriver
*drv
= bs
->drv
;
3066 if (drv
->bdrv_snapshot_goto
)
3067 return drv
->bdrv_snapshot_goto(bs
, snapshot_id
);
3070 drv
->bdrv_close(bs
);
3071 ret
= bdrv_snapshot_goto(bs
->file
, snapshot_id
);
3072 open_ret
= drv
->bdrv_open(bs
, bs
->open_flags
);
3074 bdrv_delete(bs
->file
);
3084 int bdrv_snapshot_delete(BlockDriverState
*bs
, const char *snapshot_id
)
3086 BlockDriver
*drv
= bs
->drv
;
3089 if (drv
->bdrv_snapshot_delete
)
3090 return drv
->bdrv_snapshot_delete(bs
, snapshot_id
);
3092 return bdrv_snapshot_delete(bs
->file
, snapshot_id
);
3096 int bdrv_snapshot_list(BlockDriverState
*bs
,
3097 QEMUSnapshotInfo
**psn_info
)
3099 BlockDriver
*drv
= bs
->drv
;
3102 if (drv
->bdrv_snapshot_list
)
3103 return drv
->bdrv_snapshot_list(bs
, psn_info
);
3105 return bdrv_snapshot_list(bs
->file
, psn_info
);
3109 int bdrv_snapshot_load_tmp(BlockDriverState
*bs
,
3110 const char *snapshot_name
)
3112 BlockDriver
*drv
= bs
->drv
;
3116 if (!bs
->read_only
) {
3119 if (drv
->bdrv_snapshot_load_tmp
) {
3120 return drv
->bdrv_snapshot_load_tmp(bs
, snapshot_name
);
3125 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
3126 const char *backing_file
)
3132 if (bs
->backing_hd
) {
3133 if (strcmp(bs
->backing_file
, backing_file
) == 0) {
3134 return bs
->backing_hd
;
3136 return bdrv_find_backing_image(bs
->backing_hd
, backing_file
);
3143 int bdrv_get_backing_file_depth(BlockDriverState
*bs
)
3149 if (!bs
->backing_hd
) {
3153 return 1 + bdrv_get_backing_file_depth(bs
->backing_hd
);
3156 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
3158 BlockDriverState
*curr_bs
= NULL
;
3166 while (curr_bs
->backing_hd
) {
3167 curr_bs
= curr_bs
->backing_hd
;
3172 #define NB_SUFFIXES 4
3174 char *get_human_readable_size(char *buf
, int buf_size
, int64_t size
)
3176 static const char suffixes
[NB_SUFFIXES
] = "KMGT";
3181 snprintf(buf
, buf_size
, "%" PRId64
, size
);
3184 for(i
= 0; i
< NB_SUFFIXES
; i
++) {
3185 if (size
< (10 * base
)) {
3186 snprintf(buf
, buf_size
, "%0.1f%c",
3187 (double)size
/ base
,
3190 } else if (size
< (1000 * base
) || i
== (NB_SUFFIXES
- 1)) {
3191 snprintf(buf
, buf_size
, "%" PRId64
"%c",
3192 ((size
+ (base
>> 1)) / base
),
3202 char *bdrv_snapshot_dump(char *buf
, int buf_size
, QEMUSnapshotInfo
*sn
)
3204 char buf1
[128], date_buf
[128], clock_buf
[128];
3214 snprintf(buf
, buf_size
,
3215 "%-10s%-20s%7s%20s%15s",
3216 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
3220 ptm
= localtime(&ti
);
3221 strftime(date_buf
, sizeof(date_buf
),
3222 "%Y-%m-%d %H:%M:%S", ptm
);
3224 localtime_r(&ti
, &tm
);
3225 strftime(date_buf
, sizeof(date_buf
),
3226 "%Y-%m-%d %H:%M:%S", &tm
);
3228 secs
= sn
->vm_clock_nsec
/ 1000000000;
3229 snprintf(clock_buf
, sizeof(clock_buf
),
3230 "%02d:%02d:%02d.%03d",
3232 (int)((secs
/ 60) % 60),
3234 (int)((sn
->vm_clock_nsec
/ 1000000) % 1000));
3235 snprintf(buf
, buf_size
,
3236 "%-10s%-20s%7s%20s%15s",
3237 sn
->id_str
, sn
->name
,
3238 get_human_readable_size(buf1
, sizeof(buf1
), sn
->vm_state_size
),
3245 /**************************************************************/
3248 BlockDriverAIOCB
*bdrv_aio_readv(BlockDriverState
*bs
, int64_t sector_num
,
3249 QEMUIOVector
*qiov
, int nb_sectors
,
3250 BlockDriverCompletionFunc
*cb
, void *opaque
)
3252 trace_bdrv_aio_readv(bs
, sector_num
, nb_sectors
, opaque
);
3254 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
,
3258 BlockDriverAIOCB
*bdrv_aio_writev(BlockDriverState
*bs
, int64_t sector_num
,
3259 QEMUIOVector
*qiov
, int nb_sectors
,
3260 BlockDriverCompletionFunc
*cb
, void *opaque
)
3262 trace_bdrv_aio_writev(bs
, sector_num
, nb_sectors
, opaque
);
3264 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
,
3269 typedef struct MultiwriteCB
{
3274 BlockDriverCompletionFunc
*cb
;
3276 QEMUIOVector
*free_qiov
;
3280 static void multiwrite_user_cb(MultiwriteCB
*mcb
)
3284 for (i
= 0; i
< mcb
->num_callbacks
; i
++) {
3285 mcb
->callbacks
[i
].cb(mcb
->callbacks
[i
].opaque
, mcb
->error
);
3286 if (mcb
->callbacks
[i
].free_qiov
) {
3287 qemu_iovec_destroy(mcb
->callbacks
[i
].free_qiov
);
3289 g_free(mcb
->callbacks
[i
].free_qiov
);
3293 static void multiwrite_cb(void *opaque
, int ret
)
3295 MultiwriteCB
*mcb
= opaque
;
3297 trace_multiwrite_cb(mcb
, ret
);
3299 if (ret
< 0 && !mcb
->error
) {
3303 mcb
->num_requests
--;
3304 if (mcb
->num_requests
== 0) {
3305 multiwrite_user_cb(mcb
);
3310 static int multiwrite_req_compare(const void *a
, const void *b
)
3312 const BlockRequest
*req1
= a
, *req2
= b
;
3315 * Note that we can't simply subtract req2->sector from req1->sector
3316 * here as that could overflow the return value.
3318 if (req1
->sector
> req2
->sector
) {
3320 } else if (req1
->sector
< req2
->sector
) {
3328 * Takes a bunch of requests and tries to merge them. Returns the number of
3329 * requests that remain after merging.
3331 static int multiwrite_merge(BlockDriverState
*bs
, BlockRequest
*reqs
,
3332 int num_reqs
, MultiwriteCB
*mcb
)
3336 // Sort requests by start sector
3337 qsort(reqs
, num_reqs
, sizeof(*reqs
), &multiwrite_req_compare
);
3339 // Check if adjacent requests touch the same clusters. If so, combine them,
3340 // filling up gaps with zero sectors.
3342 for (i
= 1; i
< num_reqs
; i
++) {
3344 int64_t oldreq_last
= reqs
[outidx
].sector
+ reqs
[outidx
].nb_sectors
;
3346 // Handle exactly sequential writes and overlapping writes.
3347 if (reqs
[i
].sector
<= oldreq_last
) {
3351 if (reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1 > IOV_MAX
) {
3357 QEMUIOVector
*qiov
= g_malloc0(sizeof(*qiov
));
3358 qemu_iovec_init(qiov
,
3359 reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1);
3361 // Add the first request to the merged one. If the requests are
3362 // overlapping, drop the last sectors of the first request.
3363 size
= (reqs
[i
].sector
- reqs
[outidx
].sector
) << 9;
3364 qemu_iovec_concat(qiov
, reqs
[outidx
].qiov
, 0, size
);
3366 // We should need to add any zeros between the two requests
3367 assert (reqs
[i
].sector
<= oldreq_last
);
3369 // Add the second request
3370 qemu_iovec_concat(qiov
, reqs
[i
].qiov
, 0, reqs
[i
].qiov
->size
);
3372 reqs
[outidx
].nb_sectors
= qiov
->size
>> 9;
3373 reqs
[outidx
].qiov
= qiov
;
3375 mcb
->callbacks
[i
].free_qiov
= reqs
[outidx
].qiov
;
3378 reqs
[outidx
].sector
= reqs
[i
].sector
;
3379 reqs
[outidx
].nb_sectors
= reqs
[i
].nb_sectors
;
3380 reqs
[outidx
].qiov
= reqs
[i
].qiov
;
3388 * Submit multiple AIO write requests at once.
3390 * On success, the function returns 0 and all requests in the reqs array have
3391 * been submitted. In error case this function returns -1, and any of the
3392 * requests may or may not be submitted yet. In particular, this means that the
3393 * callback will be called for some of the requests, for others it won't. The
3394 * caller must check the error field of the BlockRequest to wait for the right
3395 * callbacks (if error != 0, no callback will be called).
3397 * The implementation may modify the contents of the reqs array, e.g. to merge
3398 * requests. However, the fields opaque and error are left unmodified as they
3399 * are used to signal failure for a single request to the caller.
3401 int bdrv_aio_multiwrite(BlockDriverState
*bs
, BlockRequest
*reqs
, int num_reqs
)
3406 /* don't submit writes if we don't have a medium */
3407 if (bs
->drv
== NULL
) {
3408 for (i
= 0; i
< num_reqs
; i
++) {
3409 reqs
[i
].error
= -ENOMEDIUM
;
3414 if (num_reqs
== 0) {
3418 // Create MultiwriteCB structure
3419 mcb
= g_malloc0(sizeof(*mcb
) + num_reqs
* sizeof(*mcb
->callbacks
));
3420 mcb
->num_requests
= 0;
3421 mcb
->num_callbacks
= num_reqs
;
3423 for (i
= 0; i
< num_reqs
; i
++) {
3424 mcb
->callbacks
[i
].cb
= reqs
[i
].cb
;
3425 mcb
->callbacks
[i
].opaque
= reqs
[i
].opaque
;
3428 // Check for mergable requests
3429 num_reqs
= multiwrite_merge(bs
, reqs
, num_reqs
, mcb
);
3431 trace_bdrv_aio_multiwrite(mcb
, mcb
->num_callbacks
, num_reqs
);
3433 /* Run the aio requests. */
3434 mcb
->num_requests
= num_reqs
;
3435 for (i
= 0; i
< num_reqs
; i
++) {
3436 bdrv_aio_writev(bs
, reqs
[i
].sector
, reqs
[i
].qiov
,
3437 reqs
[i
].nb_sectors
, multiwrite_cb
, mcb
);
3443 void bdrv_aio_cancel(BlockDriverAIOCB
*acb
)
3445 acb
->pool
->cancel(acb
);
3448 /* block I/O throttling */
3449 static bool bdrv_exceed_bps_limits(BlockDriverState
*bs
, int nb_sectors
,
3450 bool is_write
, double elapsed_time
, uint64_t *wait
)
3452 uint64_t bps_limit
= 0;
3453 double bytes_limit
, bytes_base
, bytes_res
;
3454 double slice_time
, wait_time
;
3456 if (bs
->io_limits
.bps
[BLOCK_IO_LIMIT_TOTAL
]) {
3457 bps_limit
= bs
->io_limits
.bps
[BLOCK_IO_LIMIT_TOTAL
];
3458 } else if (bs
->io_limits
.bps
[is_write
]) {
3459 bps_limit
= bs
->io_limits
.bps
[is_write
];
3468 slice_time
= bs
->slice_end
- bs
->slice_start
;
3469 slice_time
/= (NANOSECONDS_PER_SECOND
);
3470 bytes_limit
= bps_limit
* slice_time
;
3471 bytes_base
= bs
->nr_bytes
[is_write
] - bs
->io_base
.bytes
[is_write
];
3472 if (bs
->io_limits
.bps
[BLOCK_IO_LIMIT_TOTAL
]) {
3473 bytes_base
+= bs
->nr_bytes
[!is_write
] - bs
->io_base
.bytes
[!is_write
];
3476 /* bytes_base: the bytes of data which have been read/written; and
3477 * it is obtained from the history statistic info.
3478 * bytes_res: the remaining bytes of data which need to be read/written.
3479 * (bytes_base + bytes_res) / bps_limit: used to calcuate
3480 * the total time for completing reading/writting all data.
3482 bytes_res
= (unsigned) nb_sectors
* BDRV_SECTOR_SIZE
;
3484 if (bytes_base
+ bytes_res
<= bytes_limit
) {
3492 /* Calc approx time to dispatch */
3493 wait_time
= (bytes_base
+ bytes_res
) / bps_limit
- elapsed_time
;
3495 /* When the I/O rate at runtime exceeds the limits,
3496 * bs->slice_end need to be extended in order that the current statistic
3497 * info can be kept until the timer fire, so it is increased and tuned
3498 * based on the result of experiment.
3500 bs
->slice_time
= wait_time
* BLOCK_IO_SLICE_TIME
* 10;
3501 bs
->slice_end
+= bs
->slice_time
- 3 * BLOCK_IO_SLICE_TIME
;
3503 *wait
= wait_time
* BLOCK_IO_SLICE_TIME
* 10;
3509 static bool bdrv_exceed_iops_limits(BlockDriverState
*bs
, bool is_write
,
3510 double elapsed_time
, uint64_t *wait
)
3512 uint64_t iops_limit
= 0;
3513 double ios_limit
, ios_base
;
3514 double slice_time
, wait_time
;
3516 if (bs
->io_limits
.iops
[BLOCK_IO_LIMIT_TOTAL
]) {
3517 iops_limit
= bs
->io_limits
.iops
[BLOCK_IO_LIMIT_TOTAL
];
3518 } else if (bs
->io_limits
.iops
[is_write
]) {
3519 iops_limit
= bs
->io_limits
.iops
[is_write
];
3528 slice_time
= bs
->slice_end
- bs
->slice_start
;
3529 slice_time
/= (NANOSECONDS_PER_SECOND
);
3530 ios_limit
= iops_limit
* slice_time
;
3531 ios_base
= bs
->nr_ops
[is_write
] - bs
->io_base
.ios
[is_write
];
3532 if (bs
->io_limits
.iops
[BLOCK_IO_LIMIT_TOTAL
]) {
3533 ios_base
+= bs
->nr_ops
[!is_write
] - bs
->io_base
.ios
[!is_write
];
3536 if (ios_base
+ 1 <= ios_limit
) {
3544 /* Calc approx time to dispatch */
3545 wait_time
= (ios_base
+ 1) / iops_limit
;
3546 if (wait_time
> elapsed_time
) {
3547 wait_time
= wait_time
- elapsed_time
;
3552 bs
->slice_time
= wait_time
* BLOCK_IO_SLICE_TIME
* 10;
3553 bs
->slice_end
+= bs
->slice_time
- 3 * BLOCK_IO_SLICE_TIME
;
3555 *wait
= wait_time
* BLOCK_IO_SLICE_TIME
* 10;
3561 static bool bdrv_exceed_io_limits(BlockDriverState
*bs
, int nb_sectors
,
3562 bool is_write
, int64_t *wait
)
3564 int64_t now
, max_wait
;
3565 uint64_t bps_wait
= 0, iops_wait
= 0;
3566 double elapsed_time
;
3567 int bps_ret
, iops_ret
;
3569 now
= qemu_get_clock_ns(vm_clock
);
3570 if ((bs
->slice_start
< now
)
3571 && (bs
->slice_end
> now
)) {
3572 bs
->slice_end
= now
+ bs
->slice_time
;
3574 bs
->slice_time
= 5 * BLOCK_IO_SLICE_TIME
;
3575 bs
->slice_start
= now
;
3576 bs
->slice_end
= now
+ bs
->slice_time
;
3578 bs
->io_base
.bytes
[is_write
] = bs
->nr_bytes
[is_write
];
3579 bs
->io_base
.bytes
[!is_write
] = bs
->nr_bytes
[!is_write
];
3581 bs
->io_base
.ios
[is_write
] = bs
->nr_ops
[is_write
];
3582 bs
->io_base
.ios
[!is_write
] = bs
->nr_ops
[!is_write
];
3585 elapsed_time
= now
- bs
->slice_start
;
3586 elapsed_time
/= (NANOSECONDS_PER_SECOND
);
3588 bps_ret
= bdrv_exceed_bps_limits(bs
, nb_sectors
,
3589 is_write
, elapsed_time
, &bps_wait
);
3590 iops_ret
= bdrv_exceed_iops_limits(bs
, is_write
,
3591 elapsed_time
, &iops_wait
);
3592 if (bps_ret
|| iops_ret
) {
3593 max_wait
= bps_wait
> iops_wait
? bps_wait
: iops_wait
;
3598 now
= qemu_get_clock_ns(vm_clock
);
3599 if (bs
->slice_end
< now
+ max_wait
) {
3600 bs
->slice_end
= now
+ max_wait
;
3613 /**************************************************************/
3614 /* async block device emulation */
3616 typedef struct BlockDriverAIOCBSync
{
3617 BlockDriverAIOCB common
;
3620 /* vector translation state */
3624 } BlockDriverAIOCBSync
;
3626 static void bdrv_aio_cancel_em(BlockDriverAIOCB
*blockacb
)
3628 BlockDriverAIOCBSync
*acb
=
3629 container_of(blockacb
, BlockDriverAIOCBSync
, common
);
3630 qemu_bh_delete(acb
->bh
);
3632 qemu_aio_release(acb
);
3635 static AIOPool bdrv_em_aio_pool
= {
3636 .aiocb_size
= sizeof(BlockDriverAIOCBSync
),
3637 .cancel
= bdrv_aio_cancel_em
,
3640 static void bdrv_aio_bh_cb(void *opaque
)
3642 BlockDriverAIOCBSync
*acb
= opaque
;
3645 qemu_iovec_from_buf(acb
->qiov
, 0, acb
->bounce
, acb
->qiov
->size
);
3646 qemu_vfree(acb
->bounce
);
3647 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
3648 qemu_bh_delete(acb
->bh
);
3650 qemu_aio_release(acb
);
3653 static BlockDriverAIOCB
*bdrv_aio_rw_vector(BlockDriverState
*bs
,
3657 BlockDriverCompletionFunc
*cb
,
3662 BlockDriverAIOCBSync
*acb
;
3664 acb
= qemu_aio_get(&bdrv_em_aio_pool
, bs
, cb
, opaque
);
3665 acb
->is_write
= is_write
;
3667 acb
->bounce
= qemu_blockalign(bs
, qiov
->size
);
3668 acb
->bh
= qemu_bh_new(bdrv_aio_bh_cb
, acb
);
3671 qemu_iovec_to_buf(acb
->qiov
, 0, acb
->bounce
, qiov
->size
);
3672 acb
->ret
= bs
->drv
->bdrv_write(bs
, sector_num
, acb
->bounce
, nb_sectors
);
3674 acb
->ret
= bs
->drv
->bdrv_read(bs
, sector_num
, acb
->bounce
, nb_sectors
);
3677 qemu_bh_schedule(acb
->bh
);
3679 return &acb
->common
;
3682 static BlockDriverAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
3683 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
3684 BlockDriverCompletionFunc
*cb
, void *opaque
)
3686 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 0);
3689 static BlockDriverAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
3690 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
3691 BlockDriverCompletionFunc
*cb
, void *opaque
)
3693 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 1);
3697 typedef struct BlockDriverAIOCBCoroutine
{
3698 BlockDriverAIOCB common
;
3702 } BlockDriverAIOCBCoroutine
;
3704 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB
*blockacb
)
3709 static AIOPool bdrv_em_co_aio_pool
= {
3710 .aiocb_size
= sizeof(BlockDriverAIOCBCoroutine
),
3711 .cancel
= bdrv_aio_co_cancel_em
,
3714 static void bdrv_co_em_bh(void *opaque
)
3716 BlockDriverAIOCBCoroutine
*acb
= opaque
;
3718 acb
->common
.cb(acb
->common
.opaque
, acb
->req
.error
);
3719 qemu_bh_delete(acb
->bh
);
3720 qemu_aio_release(acb
);
3723 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
3724 static void coroutine_fn
bdrv_co_do_rw(void *opaque
)
3726 BlockDriverAIOCBCoroutine
*acb
= opaque
;
3727 BlockDriverState
*bs
= acb
->common
.bs
;
3729 if (!acb
->is_write
) {
3730 acb
->req
.error
= bdrv_co_do_readv(bs
, acb
->req
.sector
,
3731 acb
->req
.nb_sectors
, acb
->req
.qiov
, 0);
3733 acb
->req
.error
= bdrv_co_do_writev(bs
, acb
->req
.sector
,
3734 acb
->req
.nb_sectors
, acb
->req
.qiov
, 0);
3737 acb
->bh
= qemu_bh_new(bdrv_co_em_bh
, acb
);
3738 qemu_bh_schedule(acb
->bh
);
3741 static BlockDriverAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
3745 BlockDriverCompletionFunc
*cb
,
3750 BlockDriverAIOCBCoroutine
*acb
;
3752 acb
= qemu_aio_get(&bdrv_em_co_aio_pool
, bs
, cb
, opaque
);
3753 acb
->req
.sector
= sector_num
;
3754 acb
->req
.nb_sectors
= nb_sectors
;
3755 acb
->req
.qiov
= qiov
;
3756 acb
->is_write
= is_write
;
3758 co
= qemu_coroutine_create(bdrv_co_do_rw
);
3759 qemu_coroutine_enter(co
, acb
);
3761 return &acb
->common
;
3764 static void coroutine_fn
bdrv_aio_flush_co_entry(void *opaque
)
3766 BlockDriverAIOCBCoroutine
*acb
= opaque
;
3767 BlockDriverState
*bs
= acb
->common
.bs
;
3769 acb
->req
.error
= bdrv_co_flush(bs
);
3770 acb
->bh
= qemu_bh_new(bdrv_co_em_bh
, acb
);
3771 qemu_bh_schedule(acb
->bh
);
3774 BlockDriverAIOCB
*bdrv_aio_flush(BlockDriverState
*bs
,
3775 BlockDriverCompletionFunc
*cb
, void *opaque
)
3777 trace_bdrv_aio_flush(bs
, opaque
);
3780 BlockDriverAIOCBCoroutine
*acb
;
3782 acb
= qemu_aio_get(&bdrv_em_co_aio_pool
, bs
, cb
, opaque
);
3783 co
= qemu_coroutine_create(bdrv_aio_flush_co_entry
);
3784 qemu_coroutine_enter(co
, acb
);
3786 return &acb
->common
;
3789 static void coroutine_fn
bdrv_aio_discard_co_entry(void *opaque
)
3791 BlockDriverAIOCBCoroutine
*acb
= opaque
;
3792 BlockDriverState
*bs
= acb
->common
.bs
;
3794 acb
->req
.error
= bdrv_co_discard(bs
, acb
->req
.sector
, acb
->req
.nb_sectors
);
3795 acb
->bh
= qemu_bh_new(bdrv_co_em_bh
, acb
);
3796 qemu_bh_schedule(acb
->bh
);
3799 BlockDriverAIOCB
*bdrv_aio_discard(BlockDriverState
*bs
,
3800 int64_t sector_num
, int nb_sectors
,
3801 BlockDriverCompletionFunc
*cb
, void *opaque
)
3804 BlockDriverAIOCBCoroutine
*acb
;
3806 trace_bdrv_aio_discard(bs
, sector_num
, nb_sectors
, opaque
);
3808 acb
= qemu_aio_get(&bdrv_em_co_aio_pool
, bs
, cb
, opaque
);
3809 acb
->req
.sector
= sector_num
;
3810 acb
->req
.nb_sectors
= nb_sectors
;
3811 co
= qemu_coroutine_create(bdrv_aio_discard_co_entry
);
3812 qemu_coroutine_enter(co
, acb
);
3814 return &acb
->common
;
3817 void bdrv_init(void)
3819 module_call_init(MODULE_INIT_BLOCK
);
3822 void bdrv_init_with_whitelist(void)
3824 use_bdrv_whitelist
= 1;
3828 void *qemu_aio_get(AIOPool
*pool
, BlockDriverState
*bs
,
3829 BlockDriverCompletionFunc
*cb
, void *opaque
)
3831 BlockDriverAIOCB
*acb
;
3833 if (pool
->free_aiocb
) {
3834 acb
= pool
->free_aiocb
;
3835 pool
->free_aiocb
= acb
->next
;
3837 acb
= g_malloc0(pool
->aiocb_size
);
3842 acb
->opaque
= opaque
;
3846 void qemu_aio_release(void *p
)
3848 BlockDriverAIOCB
*acb
= (BlockDriverAIOCB
*)p
;
3849 AIOPool
*pool
= acb
->pool
;
3850 acb
->next
= pool
->free_aiocb
;
3851 pool
->free_aiocb
= acb
;
3854 /**************************************************************/
3855 /* Coroutine block device emulation */
3857 typedef struct CoroutineIOCompletion
{
3858 Coroutine
*coroutine
;
3860 } CoroutineIOCompletion
;
3862 static void bdrv_co_io_em_complete(void *opaque
, int ret
)
3864 CoroutineIOCompletion
*co
= opaque
;
3867 qemu_coroutine_enter(co
->coroutine
, NULL
);
3870 static int coroutine_fn
bdrv_co_io_em(BlockDriverState
*bs
, int64_t sector_num
,
3871 int nb_sectors
, QEMUIOVector
*iov
,
3874 CoroutineIOCompletion co
= {
3875 .coroutine
= qemu_coroutine_self(),
3877 BlockDriverAIOCB
*acb
;
3880 acb
= bs
->drv
->bdrv_aio_writev(bs
, sector_num
, iov
, nb_sectors
,
3881 bdrv_co_io_em_complete
, &co
);
3883 acb
= bs
->drv
->bdrv_aio_readv(bs
, sector_num
, iov
, nb_sectors
,
3884 bdrv_co_io_em_complete
, &co
);
3887 trace_bdrv_co_io_em(bs
, sector_num
, nb_sectors
, is_write
, acb
);
3891 qemu_coroutine_yield();
3896 static int coroutine_fn
bdrv_co_readv_em(BlockDriverState
*bs
,
3897 int64_t sector_num
, int nb_sectors
,
3900 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, false);
3903 static int coroutine_fn
bdrv_co_writev_em(BlockDriverState
*bs
,
3904 int64_t sector_num
, int nb_sectors
,
3907 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, true);
3910 static void coroutine_fn
bdrv_flush_co_entry(void *opaque
)
3912 RwCo
*rwco
= opaque
;
3914 rwco
->ret
= bdrv_co_flush(rwco
->bs
);
3917 int coroutine_fn
bdrv_co_flush(BlockDriverState
*bs
)
3921 if (!bs
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
)) {
3925 /* Write back cached data to the OS even with cache=unsafe */
3926 if (bs
->drv
->bdrv_co_flush_to_os
) {
3927 ret
= bs
->drv
->bdrv_co_flush_to_os(bs
);
3933 /* But don't actually force it to the disk with cache=unsafe */
3934 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
3938 if (bs
->drv
->bdrv_co_flush_to_disk
) {
3939 ret
= bs
->drv
->bdrv_co_flush_to_disk(bs
);
3940 } else if (bs
->drv
->bdrv_aio_flush
) {
3941 BlockDriverAIOCB
*acb
;
3942 CoroutineIOCompletion co
= {
3943 .coroutine
= qemu_coroutine_self(),
3946 acb
= bs
->drv
->bdrv_aio_flush(bs
, bdrv_co_io_em_complete
, &co
);
3950 qemu_coroutine_yield();
3955 * Some block drivers always operate in either writethrough or unsafe
3956 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
3957 * know how the server works (because the behaviour is hardcoded or
3958 * depends on server-side configuration), so we can't ensure that
3959 * everything is safe on disk. Returning an error doesn't work because
3960 * that would break guests even if the server operates in writethrough
3963 * Let's hope the user knows what he's doing.
3971 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
3972 * in the case of cache=unsafe, so there are no useless flushes.
3975 return bdrv_co_flush(bs
->file
);
3978 void bdrv_invalidate_cache(BlockDriverState
*bs
)
3980 if (bs
->drv
&& bs
->drv
->bdrv_invalidate_cache
) {
3981 bs
->drv
->bdrv_invalidate_cache(bs
);
3985 void bdrv_invalidate_cache_all(void)
3987 BlockDriverState
*bs
;
3989 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
3990 bdrv_invalidate_cache(bs
);
3994 void bdrv_clear_incoming_migration_all(void)
3996 BlockDriverState
*bs
;
3998 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
3999 bs
->open_flags
= bs
->open_flags
& ~(BDRV_O_INCOMING
);
4003 int bdrv_flush(BlockDriverState
*bs
)
4011 if (qemu_in_coroutine()) {
4012 /* Fast-path if already in coroutine context */
4013 bdrv_flush_co_entry(&rwco
);
4015 co
= qemu_coroutine_create(bdrv_flush_co_entry
);
4016 qemu_coroutine_enter(co
, &rwco
);
4017 while (rwco
.ret
== NOT_DONE
) {
4025 static void coroutine_fn
bdrv_discard_co_entry(void *opaque
)
4027 RwCo
*rwco
= opaque
;
4029 rwco
->ret
= bdrv_co_discard(rwco
->bs
, rwco
->sector_num
, rwco
->nb_sectors
);
4032 int coroutine_fn
bdrv_co_discard(BlockDriverState
*bs
, int64_t sector_num
,
4037 } else if (bdrv_check_request(bs
, sector_num
, nb_sectors
)) {
4039 } else if (bs
->read_only
) {
4041 } else if (bs
->drv
->bdrv_co_discard
) {
4042 return bs
->drv
->bdrv_co_discard(bs
, sector_num
, nb_sectors
);
4043 } else if (bs
->drv
->bdrv_aio_discard
) {
4044 BlockDriverAIOCB
*acb
;
4045 CoroutineIOCompletion co
= {
4046 .coroutine
= qemu_coroutine_self(),
4049 acb
= bs
->drv
->bdrv_aio_discard(bs
, sector_num
, nb_sectors
,
4050 bdrv_co_io_em_complete
, &co
);
4054 qemu_coroutine_yield();
4062 int bdrv_discard(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
)
4067 .sector_num
= sector_num
,
4068 .nb_sectors
= nb_sectors
,
4072 if (qemu_in_coroutine()) {
4073 /* Fast-path if already in coroutine context */
4074 bdrv_discard_co_entry(&rwco
);
4076 co
= qemu_coroutine_create(bdrv_discard_co_entry
);
4077 qemu_coroutine_enter(co
, &rwco
);
4078 while (rwco
.ret
== NOT_DONE
) {
4086 /**************************************************************/
4087 /* removable device support */
4090 * Return TRUE if the media is present
4092 int bdrv_is_inserted(BlockDriverState
*bs
)
4094 BlockDriver
*drv
= bs
->drv
;
4098 if (!drv
->bdrv_is_inserted
)
4100 return drv
->bdrv_is_inserted(bs
);
4104 * Return whether the media changed since the last call to this
4105 * function, or -ENOTSUP if we don't know. Most drivers don't know.
4107 int bdrv_media_changed(BlockDriverState
*bs
)
4109 BlockDriver
*drv
= bs
->drv
;
4111 if (drv
&& drv
->bdrv_media_changed
) {
4112 return drv
->bdrv_media_changed(bs
);
4118 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4120 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
4122 BlockDriver
*drv
= bs
->drv
;
4124 if (drv
&& drv
->bdrv_eject
) {
4125 drv
->bdrv_eject(bs
, eject_flag
);
4128 if (bs
->device_name
[0] != '\0') {
4129 bdrv_emit_qmp_eject_event(bs
, eject_flag
);
4134 * Lock or unlock the media (if it is locked, the user won't be able
4135 * to eject it manually).
4137 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
4139 BlockDriver
*drv
= bs
->drv
;
4141 trace_bdrv_lock_medium(bs
, locked
);
4143 if (drv
&& drv
->bdrv_lock_medium
) {
4144 drv
->bdrv_lock_medium(bs
, locked
);
4148 /* needed for generic scsi interface */
4150 int bdrv_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
4152 BlockDriver
*drv
= bs
->drv
;
4154 if (drv
&& drv
->bdrv_ioctl
)
4155 return drv
->bdrv_ioctl(bs
, req
, buf
);
4159 BlockDriverAIOCB
*bdrv_aio_ioctl(BlockDriverState
*bs
,
4160 unsigned long int req
, void *buf
,
4161 BlockDriverCompletionFunc
*cb
, void *opaque
)
4163 BlockDriver
*drv
= bs
->drv
;
4165 if (drv
&& drv
->bdrv_aio_ioctl
)
4166 return drv
->bdrv_aio_ioctl(bs
, req
, buf
, cb
, opaque
);
4170 void bdrv_set_buffer_alignment(BlockDriverState
*bs
, int align
)
4172 bs
->buffer_alignment
= align
;
4175 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
4177 return qemu_memalign((bs
&& bs
->buffer_alignment
) ? bs
->buffer_alignment
: 512, size
);
4180 void bdrv_set_dirty_tracking(BlockDriverState
*bs
, int enable
)
4182 int64_t bitmap_size
;
4184 bs
->dirty_count
= 0;
4186 if (!bs
->dirty_bitmap
) {
4187 bitmap_size
= (bdrv_getlength(bs
) >> BDRV_SECTOR_BITS
) +
4188 BDRV_SECTORS_PER_DIRTY_CHUNK
* BITS_PER_LONG
- 1;
4189 bitmap_size
/= BDRV_SECTORS_PER_DIRTY_CHUNK
* BITS_PER_LONG
;
4191 bs
->dirty_bitmap
= g_new0(unsigned long, bitmap_size
);
4194 if (bs
->dirty_bitmap
) {
4195 g_free(bs
->dirty_bitmap
);
4196 bs
->dirty_bitmap
= NULL
;
4201 int bdrv_get_dirty(BlockDriverState
*bs
, int64_t sector
)
4203 int64_t chunk
= sector
/ (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK
;
4205 if (bs
->dirty_bitmap
&&
4206 (sector
<< BDRV_SECTOR_BITS
) < bdrv_getlength(bs
)) {
4207 return !!(bs
->dirty_bitmap
[chunk
/ (sizeof(unsigned long) * 8)] &
4208 (1UL << (chunk
% (sizeof(unsigned long) * 8))));
4214 void bdrv_reset_dirty(BlockDriverState
*bs
, int64_t cur_sector
,
4217 set_dirty_bitmap(bs
, cur_sector
, nr_sectors
, 0);
4220 int64_t bdrv_get_dirty_count(BlockDriverState
*bs
)
4222 return bs
->dirty_count
;
4225 void bdrv_set_in_use(BlockDriverState
*bs
, int in_use
)
4227 assert(bs
->in_use
!= in_use
);
4228 bs
->in_use
= in_use
;
4231 int bdrv_in_use(BlockDriverState
*bs
)
4236 void bdrv_iostatus_enable(BlockDriverState
*bs
)
4238 bs
->iostatus_enabled
= true;
4239 bs
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
4242 /* The I/O status is only enabled if the drive explicitly
4243 * enables it _and_ the VM is configured to stop on errors */
4244 bool bdrv_iostatus_is_enabled(const BlockDriverState
*bs
)
4246 return (bs
->iostatus_enabled
&&
4247 (bs
->on_write_error
== BLOCKDEV_ON_ERROR_ENOSPC
||
4248 bs
->on_write_error
== BLOCKDEV_ON_ERROR_STOP
||
4249 bs
->on_read_error
== BLOCKDEV_ON_ERROR_STOP
));
4252 void bdrv_iostatus_disable(BlockDriverState
*bs
)
4254 bs
->iostatus_enabled
= false;
4257 void bdrv_iostatus_reset(BlockDriverState
*bs
)
4259 if (bdrv_iostatus_is_enabled(bs
)) {
4260 bs
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
4264 void bdrv_iostatus_set_err(BlockDriverState
*bs
, int error
)
4266 assert(bdrv_iostatus_is_enabled(bs
));
4267 if (bs
->iostatus
== BLOCK_DEVICE_IO_STATUS_OK
) {
4268 bs
->iostatus
= error
== ENOSPC
? BLOCK_DEVICE_IO_STATUS_NOSPACE
:
4269 BLOCK_DEVICE_IO_STATUS_FAILED
;
4274 bdrv_acct_start(BlockDriverState
*bs
, BlockAcctCookie
*cookie
, int64_t bytes
,
4275 enum BlockAcctType type
)
4277 assert(type
< BDRV_MAX_IOTYPE
);
4279 cookie
->bytes
= bytes
;
4280 cookie
->start_time_ns
= get_clock();
4281 cookie
->type
= type
;
4285 bdrv_acct_done(BlockDriverState
*bs
, BlockAcctCookie
*cookie
)
4287 assert(cookie
->type
< BDRV_MAX_IOTYPE
);
4289 bs
->nr_bytes
[cookie
->type
] += cookie
->bytes
;
4290 bs
->nr_ops
[cookie
->type
]++;
4291 bs
->total_time_ns
[cookie
->type
] += get_clock() - cookie
->start_time_ns
;
4294 int bdrv_img_create(const char *filename
, const char *fmt
,
4295 const char *base_filename
, const char *base_fmt
,
4296 char *options
, uint64_t img_size
, int flags
)
4298 QEMUOptionParameter
*param
= NULL
, *create_options
= NULL
;
4299 QEMUOptionParameter
*backing_fmt
, *backing_file
, *size
;
4300 BlockDriverState
*bs
= NULL
;
4301 BlockDriver
*drv
, *proto_drv
;
4302 BlockDriver
*backing_drv
= NULL
;
4305 /* Find driver and parse its options */
4306 drv
= bdrv_find_format(fmt
);
4308 error_report("Unknown file format '%s'", fmt
);
4313 proto_drv
= bdrv_find_protocol(filename
);
4315 error_report("Unknown protocol '%s'", filename
);
4320 create_options
= append_option_parameters(create_options
,
4321 drv
->create_options
);
4322 create_options
= append_option_parameters(create_options
,
4323 proto_drv
->create_options
);
4325 /* Create parameter list with default values */
4326 param
= parse_option_parameters("", create_options
, param
);
4328 set_option_parameter_int(param
, BLOCK_OPT_SIZE
, img_size
);
4330 /* Parse -o options */
4332 param
= parse_option_parameters(options
, create_options
, param
);
4333 if (param
== NULL
) {
4334 error_report("Invalid options for file format '%s'.", fmt
);
4340 if (base_filename
) {
4341 if (set_option_parameter(param
, BLOCK_OPT_BACKING_FILE
,
4343 error_report("Backing file not supported for file format '%s'",
4351 if (set_option_parameter(param
, BLOCK_OPT_BACKING_FMT
, base_fmt
)) {
4352 error_report("Backing file format not supported for file "
4353 "format '%s'", fmt
);
4359 backing_file
= get_option_parameter(param
, BLOCK_OPT_BACKING_FILE
);
4360 if (backing_file
&& backing_file
->value
.s
) {
4361 if (!strcmp(filename
, backing_file
->value
.s
)) {
4362 error_report("Error: Trying to create an image with the "
4363 "same filename as the backing file");
4369 backing_fmt
= get_option_parameter(param
, BLOCK_OPT_BACKING_FMT
);
4370 if (backing_fmt
&& backing_fmt
->value
.s
) {
4371 backing_drv
= bdrv_find_format(backing_fmt
->value
.s
);
4373 error_report("Unknown backing file format '%s'",
4374 backing_fmt
->value
.s
);
4380 // The size for the image must always be specified, with one exception:
4381 // If we are using a backing file, we can obtain the size from there
4382 size
= get_option_parameter(param
, BLOCK_OPT_SIZE
);
4383 if (size
&& size
->value
.n
== -1) {
4384 if (backing_file
&& backing_file
->value
.s
) {
4389 /* backing files always opened read-only */
4391 flags
& ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
4395 ret
= bdrv_open(bs
, backing_file
->value
.s
, back_flags
, backing_drv
);
4397 error_report("Could not open '%s'", backing_file
->value
.s
);
4400 bdrv_get_geometry(bs
, &size
);
4403 snprintf(buf
, sizeof(buf
), "%" PRId64
, size
);
4404 set_option_parameter(param
, BLOCK_OPT_SIZE
, buf
);
4406 error_report("Image creation needs a size parameter");
4412 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
4413 print_option_parameters(param
);
4416 ret
= bdrv_create(drv
, filename
, param
);
4419 if (ret
== -ENOTSUP
) {
4420 error_report("Formatting or formatting option not supported for "
4421 "file format '%s'", fmt
);
4422 } else if (ret
== -EFBIG
) {
4423 error_report("The image size is too large for file format '%s'",
4426 error_report("%s: error while creating %s: %s", filename
, fmt
,
4432 free_option_parameters(create_options
);
4433 free_option_parameters(param
);