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"
27 #include "monitor/monitor.h"
28 #include "block/block_int.h"
29 #include "block/blockjob.h"
30 #include "qemu/module.h"
31 #include "qapi/qmp/qjson.h"
32 #include "sysemu/sysemu.h"
33 #include "qemu/notify.h"
34 #include "block/coroutine.h"
35 #include "qmp-commands.h"
36 #include "qemu/timer.h"
39 #include <sys/types.h>
41 #include <sys/ioctl.h>
42 #include <sys/queue.h>
52 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
55 BDRV_REQ_COPY_ON_READ
= 0x1,
56 BDRV_REQ_ZERO_WRITE
= 0x2,
59 static void bdrv_dev_change_media_cb(BlockDriverState
*bs
, bool load
);
60 static BlockDriverAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
61 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
62 BlockDriverCompletionFunc
*cb
, void *opaque
);
63 static BlockDriverAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
64 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
65 BlockDriverCompletionFunc
*cb
, void *opaque
);
66 static int coroutine_fn
bdrv_co_readv_em(BlockDriverState
*bs
,
67 int64_t sector_num
, int nb_sectors
,
69 static int coroutine_fn
bdrv_co_writev_em(BlockDriverState
*bs
,
70 int64_t sector_num
, int nb_sectors
,
72 static int coroutine_fn
bdrv_co_do_readv(BlockDriverState
*bs
,
73 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
74 BdrvRequestFlags flags
);
75 static int coroutine_fn
bdrv_co_do_writev(BlockDriverState
*bs
,
76 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
77 BdrvRequestFlags flags
);
78 static BlockDriverAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
82 BlockDriverCompletionFunc
*cb
,
85 static void coroutine_fn
bdrv_co_do_rw(void *opaque
);
86 static int coroutine_fn
bdrv_co_do_write_zeroes(BlockDriverState
*bs
,
87 int64_t sector_num
, int nb_sectors
);
89 static bool bdrv_exceed_bps_limits(BlockDriverState
*bs
, int nb_sectors
,
90 bool is_write
, double elapsed_time
, uint64_t *wait
);
91 static bool bdrv_exceed_iops_limits(BlockDriverState
*bs
, bool is_write
,
92 double elapsed_time
, uint64_t *wait
);
93 static bool bdrv_exceed_io_limits(BlockDriverState
*bs
, int nb_sectors
,
94 bool is_write
, int64_t *wait
);
96 static QTAILQ_HEAD(, BlockDriverState
) bdrv_states
=
97 QTAILQ_HEAD_INITIALIZER(bdrv_states
);
99 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
100 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
102 /* The device to use for VM snapshots */
103 static BlockDriverState
*bs_snapshots
;
105 /* If non-zero, use only whitelisted block drivers */
106 static int use_bdrv_whitelist
;
109 static int is_windows_drive_prefix(const char *filename
)
111 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
112 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
116 int is_windows_drive(const char *filename
)
118 if (is_windows_drive_prefix(filename
) &&
121 if (strstart(filename
, "\\\\.\\", NULL
) ||
122 strstart(filename
, "//./", NULL
))
128 /* throttling disk I/O limits */
129 void bdrv_io_limits_disable(BlockDriverState
*bs
)
131 bs
->io_limits_enabled
= false;
133 while (qemu_co_queue_next(&bs
->throttled_reqs
));
135 if (bs
->block_timer
) {
136 qemu_del_timer(bs
->block_timer
);
137 qemu_free_timer(bs
->block_timer
);
138 bs
->block_timer
= NULL
;
145 static void bdrv_block_timer(void *opaque
)
147 BlockDriverState
*bs
= opaque
;
149 qemu_co_queue_next(&bs
->throttled_reqs
);
152 void bdrv_io_limits_enable(BlockDriverState
*bs
)
154 qemu_co_queue_init(&bs
->throttled_reqs
);
155 bs
->block_timer
= qemu_new_timer_ns(vm_clock
, bdrv_block_timer
, bs
);
156 bs
->io_limits_enabled
= true;
159 bool bdrv_io_limits_enabled(BlockDriverState
*bs
)
161 BlockIOLimit
*io_limits
= &bs
->io_limits
;
162 return io_limits
->bps
[BLOCK_IO_LIMIT_READ
]
163 || io_limits
->bps
[BLOCK_IO_LIMIT_WRITE
]
164 || io_limits
->bps
[BLOCK_IO_LIMIT_TOTAL
]
165 || io_limits
->iops
[BLOCK_IO_LIMIT_READ
]
166 || io_limits
->iops
[BLOCK_IO_LIMIT_WRITE
]
167 || io_limits
->iops
[BLOCK_IO_LIMIT_TOTAL
];
170 static void bdrv_io_limits_intercept(BlockDriverState
*bs
,
171 bool is_write
, int nb_sectors
)
173 int64_t wait_time
= -1;
175 if (!qemu_co_queue_empty(&bs
->throttled_reqs
)) {
176 qemu_co_queue_wait(&bs
->throttled_reqs
);
179 /* In fact, we hope to keep each request's timing, in FIFO mode. The next
180 * throttled requests will not be dequeued until the current request is
181 * allowed to be serviced. So if the current request still exceeds the
182 * limits, it will be inserted to the head. All requests followed it will
183 * be still in throttled_reqs queue.
186 while (bdrv_exceed_io_limits(bs
, nb_sectors
, is_write
, &wait_time
)) {
187 qemu_mod_timer(bs
->block_timer
,
188 wait_time
+ qemu_get_clock_ns(vm_clock
));
189 qemu_co_queue_wait_insert_head(&bs
->throttled_reqs
);
192 qemu_co_queue_next(&bs
->throttled_reqs
);
195 /* check if the path starts with "<protocol>:" */
196 static int path_has_protocol(const char *path
)
201 if (is_windows_drive(path
) ||
202 is_windows_drive_prefix(path
)) {
205 p
= path
+ strcspn(path
, ":/\\");
207 p
= path
+ strcspn(path
, ":/");
213 int path_is_absolute(const char *path
)
216 /* specific case for names like: "\\.\d:" */
217 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
220 return (*path
== '/' || *path
== '\\');
222 return (*path
== '/');
226 /* if filename is absolute, just copy it to dest. Otherwise, build a
227 path to it by considering it is relative to base_path. URL are
229 void path_combine(char *dest
, int dest_size
,
230 const char *base_path
,
231 const char *filename
)
238 if (path_is_absolute(filename
)) {
239 pstrcpy(dest
, dest_size
, filename
);
241 p
= strchr(base_path
, ':');
246 p1
= strrchr(base_path
, '/');
250 p2
= strrchr(base_path
, '\\');
262 if (len
> dest_size
- 1)
264 memcpy(dest
, base_path
, len
);
266 pstrcat(dest
, dest_size
, filename
);
270 void bdrv_get_full_backing_filename(BlockDriverState
*bs
, char *dest
, size_t sz
)
272 if (bs
->backing_file
[0] == '\0' || path_has_protocol(bs
->backing_file
)) {
273 pstrcpy(dest
, sz
, bs
->backing_file
);
275 path_combine(dest
, sz
, bs
->filename
, bs
->backing_file
);
279 void bdrv_register(BlockDriver
*bdrv
)
281 /* Block drivers without coroutine functions need emulation */
282 if (!bdrv
->bdrv_co_readv
) {
283 bdrv
->bdrv_co_readv
= bdrv_co_readv_em
;
284 bdrv
->bdrv_co_writev
= bdrv_co_writev_em
;
286 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
287 * the block driver lacks aio we need to emulate that too.
289 if (!bdrv
->bdrv_aio_readv
) {
290 /* add AIO emulation layer */
291 bdrv
->bdrv_aio_readv
= bdrv_aio_readv_em
;
292 bdrv
->bdrv_aio_writev
= bdrv_aio_writev_em
;
296 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
299 /* create a new block device (by default it is empty) */
300 BlockDriverState
*bdrv_new(const char *device_name
)
302 BlockDriverState
*bs
;
304 bs
= g_malloc0(sizeof(BlockDriverState
));
305 pstrcpy(bs
->device_name
, sizeof(bs
->device_name
), device_name
);
306 if (device_name
[0] != '\0') {
307 QTAILQ_INSERT_TAIL(&bdrv_states
, bs
, list
);
309 bdrv_iostatus_disable(bs
);
310 notifier_list_init(&bs
->close_notifiers
);
315 void bdrv_add_close_notifier(BlockDriverState
*bs
, Notifier
*notify
)
317 notifier_list_add(&bs
->close_notifiers
, notify
);
320 BlockDriver
*bdrv_find_format(const char *format_name
)
323 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
324 if (!strcmp(drv1
->format_name
, format_name
)) {
331 static int bdrv_is_whitelisted(BlockDriver
*drv
)
333 static const char *whitelist
[] = {
334 CONFIG_BDRV_WHITELIST
339 return 1; /* no whitelist, anything goes */
341 for (p
= whitelist
; *p
; p
++) {
342 if (!strcmp(drv
->format_name
, *p
)) {
349 BlockDriver
*bdrv_find_whitelisted_format(const char *format_name
)
351 BlockDriver
*drv
= bdrv_find_format(format_name
);
352 return drv
&& bdrv_is_whitelisted(drv
) ? drv
: NULL
;
355 typedef struct CreateCo
{
358 QEMUOptionParameter
*options
;
362 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
364 CreateCo
*cco
= opaque
;
367 cco
->ret
= cco
->drv
->bdrv_create(cco
->filename
, cco
->options
);
370 int bdrv_create(BlockDriver
*drv
, const char* filename
,
371 QEMUOptionParameter
*options
)
378 .filename
= g_strdup(filename
),
383 if (!drv
->bdrv_create
) {
388 if (qemu_in_coroutine()) {
389 /* Fast-path if already in coroutine context */
390 bdrv_create_co_entry(&cco
);
392 co
= qemu_coroutine_create(bdrv_create_co_entry
);
393 qemu_coroutine_enter(co
, &cco
);
394 while (cco
.ret
== NOT_DONE
) {
402 g_free(cco
.filename
);
406 int bdrv_create_file(const char* filename
, QEMUOptionParameter
*options
)
410 drv
= bdrv_find_protocol(filename
);
415 return bdrv_create(drv
, filename
, options
);
419 * Create a uniquely-named empty temporary file.
420 * Return 0 upon success, otherwise a negative errno value.
422 int get_tmp_filename(char *filename
, int size
)
425 char temp_dir
[MAX_PATH
];
426 /* GetTempFileName requires that its output buffer (4th param)
427 have length MAX_PATH or greater. */
428 assert(size
>= MAX_PATH
);
429 return (GetTempPath(MAX_PATH
, temp_dir
)
430 && GetTempFileName(temp_dir
, "qem", 0, filename
)
431 ? 0 : -GetLastError());
435 tmpdir
= getenv("TMPDIR");
438 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
441 fd
= mkstemp(filename
);
445 if (close(fd
) != 0) {
454 * Detect host devices. By convention, /dev/cdrom[N] is always
455 * recognized as a host CDROM.
457 static BlockDriver
*find_hdev_driver(const char *filename
)
459 int score_max
= 0, score
;
460 BlockDriver
*drv
= NULL
, *d
;
462 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
463 if (d
->bdrv_probe_device
) {
464 score
= d
->bdrv_probe_device(filename
);
465 if (score
> score_max
) {
475 BlockDriver
*bdrv_find_protocol(const char *filename
)
482 /* TODO Drivers without bdrv_file_open must be specified explicitly */
485 * XXX(hch): we really should not let host device detection
486 * override an explicit protocol specification, but moving this
487 * later breaks access to device names with colons in them.
488 * Thanks to the brain-dead persistent naming schemes on udev-
489 * based Linux systems those actually are quite common.
491 drv1
= find_hdev_driver(filename
);
496 if (!path_has_protocol(filename
)) {
497 return bdrv_find_format("file");
499 p
= strchr(filename
, ':');
502 if (len
> sizeof(protocol
) - 1)
503 len
= sizeof(protocol
) - 1;
504 memcpy(protocol
, filename
, len
);
505 protocol
[len
] = '\0';
506 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
507 if (drv1
->protocol_name
&&
508 !strcmp(drv1
->protocol_name
, protocol
)) {
515 static int find_image_format(BlockDriverState
*bs
, const char *filename
,
518 int score
, score_max
;
519 BlockDriver
*drv1
, *drv
;
523 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
524 if (bs
->sg
|| !bdrv_is_inserted(bs
) || bdrv_getlength(bs
) == 0) {
525 drv
= bdrv_find_format("raw");
533 ret
= bdrv_pread(bs
, 0, buf
, sizeof(buf
));
541 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
542 if (drv1
->bdrv_probe
) {
543 score
= drv1
->bdrv_probe(buf
, ret
, filename
);
544 if (score
> score_max
) {
558 * Set the current 'total_sectors' value
560 static int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
562 BlockDriver
*drv
= bs
->drv
;
564 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
568 /* query actual device if possible, otherwise just trust the hint */
569 if (drv
->bdrv_getlength
) {
570 int64_t length
= drv
->bdrv_getlength(bs
);
574 hint
= length
>> BDRV_SECTOR_BITS
;
577 bs
->total_sectors
= hint
;
582 * Set open flags for a given discard mode
584 * Return 0 on success, -1 if the discard mode was invalid.
586 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
588 *flags
&= ~BDRV_O_UNMAP
;
590 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
592 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
593 *flags
|= BDRV_O_UNMAP
;
602 * Set open flags for a given cache mode
604 * Return 0 on success, -1 if the cache mode was invalid.
606 int bdrv_parse_cache_flags(const char *mode
, int *flags
)
608 *flags
&= ~BDRV_O_CACHE_MASK
;
610 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
611 *flags
|= BDRV_O_NOCACHE
| BDRV_O_CACHE_WB
;
612 } else if (!strcmp(mode
, "directsync")) {
613 *flags
|= BDRV_O_NOCACHE
;
614 } else if (!strcmp(mode
, "writeback")) {
615 *flags
|= BDRV_O_CACHE_WB
;
616 } else if (!strcmp(mode
, "unsafe")) {
617 *flags
|= BDRV_O_CACHE_WB
;
618 *flags
|= BDRV_O_NO_FLUSH
;
619 } else if (!strcmp(mode
, "writethrough")) {
620 /* this is the default */
629 * The copy-on-read flag is actually a reference count so multiple users may
630 * use the feature without worrying about clobbering its previous state.
631 * Copy-on-read stays enabled until all users have called to disable it.
633 void bdrv_enable_copy_on_read(BlockDriverState
*bs
)
638 void bdrv_disable_copy_on_read(BlockDriverState
*bs
)
640 assert(bs
->copy_on_read
> 0);
644 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
646 int open_flags
= flags
| BDRV_O_CACHE_WB
;
649 * Clear flags that are internal to the block layer before opening the
652 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
655 * Snapshots should be writable.
657 if (bs
->is_temporary
) {
658 open_flags
|= BDRV_O_RDWR
;
665 * Common part for opening disk images and files
667 * Removes all processed options from *options.
669 static int bdrv_open_common(BlockDriverState
*bs
, BlockDriverState
*file
,
670 const char *filename
, QDict
*options
,
671 int flags
, BlockDriver
*drv
)
676 assert(bs
->file
== NULL
);
677 assert(options
!= NULL
&& bs
->options
!= options
);
679 trace_bdrv_open_common(bs
, filename
, flags
, drv
->format_name
);
681 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
)) {
685 /* bdrv_open() with directly using a protocol as drv. This layer is already
686 * opened, so assign it to bs (while file becomes a closed BlockDriverState)
687 * and return immediately. */
688 if (file
!= NULL
&& drv
->bdrv_file_open
) {
693 bs
->open_flags
= flags
;
694 bs
->buffer_alignment
= 512;
696 assert(bs
->copy_on_read
== 0); /* bdrv_new() and bdrv_close() make it so */
697 if ((flags
& BDRV_O_RDWR
) && (flags
& BDRV_O_COPY_ON_READ
)) {
698 bdrv_enable_copy_on_read(bs
);
701 if (filename
!= NULL
) {
702 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
704 bs
->filename
[0] = '\0';
708 bs
->opaque
= g_malloc0(drv
->instance_size
);
710 bs
->enable_write_cache
= !!(flags
& BDRV_O_CACHE_WB
);
711 open_flags
= bdrv_open_flags(bs
, flags
);
713 bs
->read_only
= !(open_flags
& BDRV_O_RDWR
);
715 /* Open the image, either directly or using a protocol */
716 if (drv
->bdrv_file_open
) {
717 assert(file
== NULL
);
718 assert(drv
->bdrv_parse_filename
|| filename
!= NULL
);
719 ret
= drv
->bdrv_file_open(bs
, filename
, options
, open_flags
);
722 qerror_report(ERROR_CLASS_GENERIC_ERROR
, "Can't use '%s' as a "
723 "block driver for the protocol level",
728 assert(file
!= NULL
);
730 ret
= drv
->bdrv_open(bs
, options
, open_flags
);
737 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
743 if (bs
->is_temporary
) {
744 assert(filename
!= NULL
);
759 * Opens a file using a protocol (file, host_device, nbd, ...)
761 * options is a QDict of options to pass to the block drivers, or NULL for an
762 * empty set of options. The reference to the QDict belongs to the block layer
763 * after the call (even on failure), so if the caller intends to reuse the
764 * dictionary, it needs to use QINCREF() before calling bdrv_file_open.
766 int bdrv_file_open(BlockDriverState
**pbs
, const char *filename
,
767 QDict
*options
, int flags
)
769 BlockDriverState
*bs
;
774 /* NULL means an empty set of options */
775 if (options
== NULL
) {
776 options
= qdict_new();
780 bs
->options
= options
;
781 options
= qdict_clone_shallow(options
);
783 /* Find the right block driver */
784 drvname
= qdict_get_try_str(options
, "driver");
786 drv
= bdrv_find_whitelisted_format(drvname
);
787 qdict_del(options
, "driver");
788 } else if (filename
) {
789 drv
= bdrv_find_protocol(filename
);
791 qerror_report(ERROR_CLASS_GENERIC_ERROR
,
792 "Must specify either driver or file");
801 /* Parse the filename and open it */
802 if (drv
->bdrv_parse_filename
&& filename
) {
803 Error
*local_err
= NULL
;
804 drv
->bdrv_parse_filename(filename
, options
, &local_err
);
805 if (error_is_set(&local_err
)) {
806 qerror_report_err(local_err
);
807 error_free(local_err
);
811 } else if (!drv
->bdrv_parse_filename
&& !filename
) {
812 qerror_report(ERROR_CLASS_GENERIC_ERROR
,
813 "The '%s' block driver requires a file name",
819 ret
= bdrv_open_common(bs
, NULL
, filename
, options
, flags
, drv
);
824 /* Check if any unknown options were used */
825 if (qdict_size(options
) != 0) {
826 const QDictEntry
*entry
= qdict_first(options
);
827 qerror_report(ERROR_CLASS_GENERIC_ERROR
, "Block protocol '%s' doesn't "
828 "support the option '%s'",
829 drv
->format_name
, entry
->key
);
842 QDECREF(bs
->options
);
848 int bdrv_open_backing_file(BlockDriverState
*bs
)
850 char backing_filename
[PATH_MAX
];
852 BlockDriver
*back_drv
= NULL
;
854 if (bs
->backing_hd
!= NULL
) {
858 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
859 if (bs
->backing_file
[0] == '\0') {
863 bs
->backing_hd
= bdrv_new("");
864 bdrv_get_full_backing_filename(bs
, backing_filename
,
865 sizeof(backing_filename
));
867 if (bs
->backing_format
[0] != '\0') {
868 back_drv
= bdrv_find_format(bs
->backing_format
);
871 /* backing files always opened read-only */
872 back_flags
= bs
->open_flags
& ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
);
874 ret
= bdrv_open(bs
->backing_hd
, backing_filename
, NULL
,
875 back_flags
, back_drv
);
877 bdrv_delete(bs
->backing_hd
);
878 bs
->backing_hd
= NULL
;
879 bs
->open_flags
|= BDRV_O_NO_BACKING
;
885 static void extract_subqdict(QDict
*src
, QDict
**dst
, const char *start
)
887 const QDictEntry
*entry
, *next
;
891 entry
= qdict_first(src
);
893 while (entry
!= NULL
) {
894 next
= qdict_next(src
, entry
);
895 if (strstart(entry
->key
, start
, &p
)) {
896 qobject_incref(entry
->value
);
897 qdict_put_obj(*dst
, p
, entry
->value
);
898 qdict_del(src
, entry
->key
);
905 * Opens a disk image (raw, qcow2, vmdk, ...)
907 * options is a QDict of options to pass to the block drivers, or NULL for an
908 * empty set of options. The reference to the QDict belongs to the block layer
909 * after the call (even on failure), so if the caller intends to reuse the
910 * dictionary, it needs to use QINCREF() before calling bdrv_open.
912 int bdrv_open(BlockDriverState
*bs
, const char *filename
, QDict
*options
,
913 int flags
, BlockDriver
*drv
)
916 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
917 char tmp_filename
[PATH_MAX
+ 1];
918 BlockDriverState
*file
= NULL
;
919 QDict
*file_options
= NULL
;
921 /* NULL means an empty set of options */
922 if (options
== NULL
) {
923 options
= qdict_new();
926 bs
->options
= options
;
927 options
= qdict_clone_shallow(options
);
929 /* For snapshot=on, create a temporary qcow2 overlay */
930 if (flags
& BDRV_O_SNAPSHOT
) {
931 BlockDriverState
*bs1
;
933 BlockDriver
*bdrv_qcow2
;
934 QEMUOptionParameter
*create_options
;
935 char backing_filename
[PATH_MAX
];
937 if (qdict_size(options
) != 0) {
938 error_report("Can't use snapshot=on with driver-specific options");
942 assert(filename
!= NULL
);
944 /* if snapshot, we create a temporary backing file and open it
945 instead of opening 'filename' directly */
947 /* if there is a backing file, use it */
949 ret
= bdrv_open(bs1
, filename
, NULL
, 0, drv
);
954 total_size
= bdrv_getlength(bs1
) & BDRV_SECTOR_MASK
;
958 ret
= get_tmp_filename(tmp_filename
, sizeof(tmp_filename
));
963 /* Real path is meaningless for protocols */
964 if (path_has_protocol(filename
)) {
965 snprintf(backing_filename
, sizeof(backing_filename
),
967 } else if (!realpath(filename
, backing_filename
)) {
972 bdrv_qcow2
= bdrv_find_format("qcow2");
973 create_options
= parse_option_parameters("", bdrv_qcow2
->create_options
,
976 set_option_parameter_int(create_options
, BLOCK_OPT_SIZE
, total_size
);
977 set_option_parameter(create_options
, BLOCK_OPT_BACKING_FILE
,
980 set_option_parameter(create_options
, BLOCK_OPT_BACKING_FMT
,
984 ret
= bdrv_create(bdrv_qcow2
, tmp_filename
, create_options
);
985 free_option_parameters(create_options
);
990 filename
= tmp_filename
;
992 bs
->is_temporary
= 1;
995 /* Open image file without format layer */
996 if (flags
& BDRV_O_RDWR
) {
997 flags
|= BDRV_O_ALLOW_RDWR
;
1000 extract_subqdict(options
, &file_options
, "file.");
1002 ret
= bdrv_file_open(&file
, filename
, file_options
,
1003 bdrv_open_flags(bs
, flags
));
1008 /* Find the right image format driver */
1010 ret
= find_image_format(file
, filename
, &drv
);
1014 goto unlink_and_fail
;
1017 /* Open the image */
1018 ret
= bdrv_open_common(bs
, file
, filename
, options
, flags
, drv
);
1020 goto unlink_and_fail
;
1023 if (bs
->file
!= file
) {
1028 /* If there is a backing file, use it */
1029 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
1030 ret
= bdrv_open_backing_file(bs
);
1032 goto close_and_fail
;
1036 /* Check if any unknown options were used */
1037 if (qdict_size(options
) != 0) {
1038 const QDictEntry
*entry
= qdict_first(options
);
1039 qerror_report(ERROR_CLASS_GENERIC_ERROR
, "Block format '%s' used by "
1040 "device '%s' doesn't support the option '%s'",
1041 drv
->format_name
, bs
->device_name
, entry
->key
);
1044 goto close_and_fail
;
1048 if (!bdrv_key_required(bs
)) {
1049 bdrv_dev_change_media_cb(bs
, true);
1052 /* throttling disk I/O limits */
1053 if (bs
->io_limits_enabled
) {
1054 bdrv_io_limits_enable(bs
);
1063 if (bs
->is_temporary
) {
1067 QDECREF(bs
->options
);
1078 typedef struct BlockReopenQueueEntry
{
1080 BDRVReopenState state
;
1081 QSIMPLEQ_ENTRY(BlockReopenQueueEntry
) entry
;
1082 } BlockReopenQueueEntry
;
1085 * Adds a BlockDriverState to a simple queue for an atomic, transactional
1086 * reopen of multiple devices.
1088 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
1089 * already performed, or alternatively may be NULL a new BlockReopenQueue will
1090 * be created and initialized. This newly created BlockReopenQueue should be
1091 * passed back in for subsequent calls that are intended to be of the same
1094 * bs is the BlockDriverState to add to the reopen queue.
1096 * flags contains the open flags for the associated bs
1098 * returns a pointer to bs_queue, which is either the newly allocated
1099 * bs_queue, or the existing bs_queue being used.
1102 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
1103 BlockDriverState
*bs
, int flags
)
1107 BlockReopenQueueEntry
*bs_entry
;
1108 if (bs_queue
== NULL
) {
1109 bs_queue
= g_new0(BlockReopenQueue
, 1);
1110 QSIMPLEQ_INIT(bs_queue
);
1114 bdrv_reopen_queue(bs_queue
, bs
->file
, flags
);
1117 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
1118 QSIMPLEQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
1120 bs_entry
->state
.bs
= bs
;
1121 bs_entry
->state
.flags
= flags
;
1127 * Reopen multiple BlockDriverStates atomically & transactionally.
1129 * The queue passed in (bs_queue) must have been built up previous
1130 * via bdrv_reopen_queue().
1132 * Reopens all BDS specified in the queue, with the appropriate
1133 * flags. All devices are prepared for reopen, and failure of any
1134 * device will cause all device changes to be abandonded, and intermediate
1137 * If all devices prepare successfully, then the changes are committed
1141 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
1144 BlockReopenQueueEntry
*bs_entry
, *next
;
1145 Error
*local_err
= NULL
;
1147 assert(bs_queue
!= NULL
);
1151 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
1152 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, &local_err
)) {
1153 error_propagate(errp
, local_err
);
1156 bs_entry
->prepared
= true;
1159 /* If we reach this point, we have success and just need to apply the
1162 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
1163 bdrv_reopen_commit(&bs_entry
->state
);
1169 QSIMPLEQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
1170 if (ret
&& bs_entry
->prepared
) {
1171 bdrv_reopen_abort(&bs_entry
->state
);
1180 /* Reopen a single BlockDriverState with the specified flags. */
1181 int bdrv_reopen(BlockDriverState
*bs
, int bdrv_flags
, Error
**errp
)
1184 Error
*local_err
= NULL
;
1185 BlockReopenQueue
*queue
= bdrv_reopen_queue(NULL
, bs
, bdrv_flags
);
1187 ret
= bdrv_reopen_multiple(queue
, &local_err
);
1188 if (local_err
!= NULL
) {
1189 error_propagate(errp
, local_err
);
1196 * Prepares a BlockDriverState for reopen. All changes are staged in the
1197 * 'opaque' field of the BDRVReopenState, which is used and allocated by
1198 * the block driver layer .bdrv_reopen_prepare()
1200 * bs is the BlockDriverState to reopen
1201 * flags are the new open flags
1202 * queue is the reopen queue
1204 * Returns 0 on success, non-zero on error. On error errp will be set
1207 * On failure, bdrv_reopen_abort() will be called to clean up any data.
1208 * It is the responsibility of the caller to then call the abort() or
1209 * commit() for any other BDS that have been left in a prepare() state
1212 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
1216 Error
*local_err
= NULL
;
1219 assert(reopen_state
!= NULL
);
1220 assert(reopen_state
->bs
->drv
!= NULL
);
1221 drv
= reopen_state
->bs
->drv
;
1223 /* if we are to stay read-only, do not allow permission change
1225 if (!(reopen_state
->bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
1226 reopen_state
->flags
& BDRV_O_RDWR
) {
1227 error_set(errp
, QERR_DEVICE_IS_READ_ONLY
,
1228 reopen_state
->bs
->device_name
);
1233 ret
= bdrv_flush(reopen_state
->bs
);
1235 error_set(errp
, ERROR_CLASS_GENERIC_ERROR
, "Error (%s) flushing drive",
1240 if (drv
->bdrv_reopen_prepare
) {
1241 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
1243 if (local_err
!= NULL
) {
1244 error_propagate(errp
, local_err
);
1246 error_set(errp
, QERR_OPEN_FILE_FAILED
,
1247 reopen_state
->bs
->filename
);
1252 /* It is currently mandatory to have a bdrv_reopen_prepare()
1253 * handler for each supported drv. */
1254 error_set(errp
, QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED
,
1255 drv
->format_name
, reopen_state
->bs
->device_name
,
1256 "reopening of file");
1268 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
1269 * makes them final by swapping the staging BlockDriverState contents into
1270 * the active BlockDriverState contents.
1272 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
1276 assert(reopen_state
!= NULL
);
1277 drv
= reopen_state
->bs
->drv
;
1278 assert(drv
!= NULL
);
1280 /* If there are any driver level actions to take */
1281 if (drv
->bdrv_reopen_commit
) {
1282 drv
->bdrv_reopen_commit(reopen_state
);
1285 /* set BDS specific flags now */
1286 reopen_state
->bs
->open_flags
= reopen_state
->flags
;
1287 reopen_state
->bs
->enable_write_cache
= !!(reopen_state
->flags
&
1289 reopen_state
->bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
1293 * Abort the reopen, and delete and free the staged changes in
1296 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
1300 assert(reopen_state
!= NULL
);
1301 drv
= reopen_state
->bs
->drv
;
1302 assert(drv
!= NULL
);
1304 if (drv
->bdrv_reopen_abort
) {
1305 drv
->bdrv_reopen_abort(reopen_state
);
1310 void bdrv_close(BlockDriverState
*bs
)
1314 block_job_cancel_sync(bs
->job
);
1317 notifier_list_notify(&bs
->close_notifiers
, bs
);
1320 if (bs
== bs_snapshots
) {
1321 bs_snapshots
= NULL
;
1323 if (bs
->backing_hd
) {
1324 bdrv_delete(bs
->backing_hd
);
1325 bs
->backing_hd
= NULL
;
1327 bs
->drv
->bdrv_close(bs
);
1330 if (bs
->is_temporary
) {
1331 unlink(bs
->filename
);
1336 bs
->copy_on_read
= 0;
1337 bs
->backing_file
[0] = '\0';
1338 bs
->backing_format
[0] = '\0';
1339 bs
->total_sectors
= 0;
1344 QDECREF(bs
->options
);
1347 if (bs
->file
!= NULL
) {
1348 bdrv_delete(bs
->file
);
1353 bdrv_dev_change_media_cb(bs
, false);
1355 /*throttling disk I/O limits*/
1356 if (bs
->io_limits_enabled
) {
1357 bdrv_io_limits_disable(bs
);
1361 void bdrv_close_all(void)
1363 BlockDriverState
*bs
;
1365 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1371 * Wait for pending requests to complete across all BlockDriverStates
1373 * This function does not flush data to disk, use bdrv_flush_all() for that
1374 * after calling this function.
1376 * Note that completion of an asynchronous I/O operation can trigger any
1377 * number of other I/O operations on other devices---for example a coroutine
1378 * can be arbitrarily complex and a constant flow of I/O can come until the
1379 * coroutine is complete. Because of this, it is not possible to have a
1380 * function to drain a single device's I/O queue.
1382 void bdrv_drain_all(void)
1384 BlockDriverState
*bs
;
1388 busy
= qemu_aio_wait();
1390 /* FIXME: We do not have timer support here, so this is effectively
1393 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1394 if (!qemu_co_queue_empty(&bs
->throttled_reqs
)) {
1395 qemu_co_queue_restart_all(&bs
->throttled_reqs
);
1401 /* If requests are still pending there is a bug somewhere */
1402 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1403 assert(QLIST_EMPTY(&bs
->tracked_requests
));
1404 assert(qemu_co_queue_empty(&bs
->throttled_reqs
));
1408 /* make a BlockDriverState anonymous by removing from bdrv_state list.
1409 Also, NULL terminate the device_name to prevent double remove */
1410 void bdrv_make_anon(BlockDriverState
*bs
)
1412 if (bs
->device_name
[0] != '\0') {
1413 QTAILQ_REMOVE(&bdrv_states
, bs
, list
);
1415 bs
->device_name
[0] = '\0';
1418 static void bdrv_rebind(BlockDriverState
*bs
)
1420 if (bs
->drv
&& bs
->drv
->bdrv_rebind
) {
1421 bs
->drv
->bdrv_rebind(bs
);
1425 static void bdrv_move_feature_fields(BlockDriverState
*bs_dest
,
1426 BlockDriverState
*bs_src
)
1428 /* move some fields that need to stay attached to the device */
1429 bs_dest
->open_flags
= bs_src
->open_flags
;
1432 bs_dest
->dev_ops
= bs_src
->dev_ops
;
1433 bs_dest
->dev_opaque
= bs_src
->dev_opaque
;
1434 bs_dest
->dev
= bs_src
->dev
;
1435 bs_dest
->buffer_alignment
= bs_src
->buffer_alignment
;
1436 bs_dest
->copy_on_read
= bs_src
->copy_on_read
;
1438 bs_dest
->enable_write_cache
= bs_src
->enable_write_cache
;
1440 /* i/o timing parameters */
1441 bs_dest
->slice_start
= bs_src
->slice_start
;
1442 bs_dest
->slice_end
= bs_src
->slice_end
;
1443 bs_dest
->slice_submitted
= bs_src
->slice_submitted
;
1444 bs_dest
->io_limits
= bs_src
->io_limits
;
1445 bs_dest
->throttled_reqs
= bs_src
->throttled_reqs
;
1446 bs_dest
->block_timer
= bs_src
->block_timer
;
1447 bs_dest
->io_limits_enabled
= bs_src
->io_limits_enabled
;
1450 bs_dest
->on_read_error
= bs_src
->on_read_error
;
1451 bs_dest
->on_write_error
= bs_src
->on_write_error
;
1454 bs_dest
->iostatus_enabled
= bs_src
->iostatus_enabled
;
1455 bs_dest
->iostatus
= bs_src
->iostatus
;
1458 bs_dest
->dirty_bitmap
= bs_src
->dirty_bitmap
;
1461 bs_dest
->in_use
= bs_src
->in_use
;
1462 bs_dest
->job
= bs_src
->job
;
1464 /* keep the same entry in bdrv_states */
1465 pstrcpy(bs_dest
->device_name
, sizeof(bs_dest
->device_name
),
1466 bs_src
->device_name
);
1467 bs_dest
->list
= bs_src
->list
;
1471 * Swap bs contents for two image chains while they are live,
1472 * while keeping required fields on the BlockDriverState that is
1473 * actually attached to a device.
1475 * This will modify the BlockDriverState fields, and swap contents
1476 * between bs_new and bs_old. Both bs_new and bs_old are modified.
1478 * bs_new is required to be anonymous.
1480 * This function does not create any image files.
1482 void bdrv_swap(BlockDriverState
*bs_new
, BlockDriverState
*bs_old
)
1484 BlockDriverState tmp
;
1486 /* bs_new must be anonymous and shouldn't have anything fancy enabled */
1487 assert(bs_new
->device_name
[0] == '\0');
1488 assert(bs_new
->dirty_bitmap
== NULL
);
1489 assert(bs_new
->job
== NULL
);
1490 assert(bs_new
->dev
== NULL
);
1491 assert(bs_new
->in_use
== 0);
1492 assert(bs_new
->io_limits_enabled
== false);
1493 assert(bs_new
->block_timer
== NULL
);
1499 /* there are some fields that should not be swapped, move them back */
1500 bdrv_move_feature_fields(&tmp
, bs_old
);
1501 bdrv_move_feature_fields(bs_old
, bs_new
);
1502 bdrv_move_feature_fields(bs_new
, &tmp
);
1504 /* bs_new shouldn't be in bdrv_states even after the swap! */
1505 assert(bs_new
->device_name
[0] == '\0');
1507 /* Check a few fields that should remain attached to the device */
1508 assert(bs_new
->dev
== NULL
);
1509 assert(bs_new
->job
== NULL
);
1510 assert(bs_new
->in_use
== 0);
1511 assert(bs_new
->io_limits_enabled
== false);
1512 assert(bs_new
->block_timer
== NULL
);
1514 bdrv_rebind(bs_new
);
1515 bdrv_rebind(bs_old
);
1519 * Add new bs contents at the top of an image chain while the chain is
1520 * live, while keeping required fields on the top layer.
1522 * This will modify the BlockDriverState fields, and swap contents
1523 * between bs_new and bs_top. Both bs_new and bs_top are modified.
1525 * bs_new is required to be anonymous.
1527 * This function does not create any image files.
1529 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
)
1531 bdrv_swap(bs_new
, bs_top
);
1533 /* The contents of 'tmp' will become bs_top, as we are
1534 * swapping bs_new and bs_top contents. */
1535 bs_top
->backing_hd
= bs_new
;
1536 bs_top
->open_flags
&= ~BDRV_O_NO_BACKING
;
1537 pstrcpy(bs_top
->backing_file
, sizeof(bs_top
->backing_file
),
1539 pstrcpy(bs_top
->backing_format
, sizeof(bs_top
->backing_format
),
1540 bs_new
->drv
? bs_new
->drv
->format_name
: "");
1543 void bdrv_delete(BlockDriverState
*bs
)
1547 assert(!bs
->in_use
);
1549 /* remove from list, if necessary */
1554 assert(bs
!= bs_snapshots
);
1558 int bdrv_attach_dev(BlockDriverState
*bs
, void *dev
)
1559 /* TODO change to DeviceState *dev when all users are qdevified */
1565 bdrv_iostatus_reset(bs
);
1569 /* TODO qdevified devices don't use this, remove when devices are qdevified */
1570 void bdrv_attach_dev_nofail(BlockDriverState
*bs
, void *dev
)
1572 if (bdrv_attach_dev(bs
, dev
) < 0) {
1577 void bdrv_detach_dev(BlockDriverState
*bs
, void *dev
)
1578 /* TODO change to DeviceState *dev when all users are qdevified */
1580 assert(bs
->dev
== dev
);
1583 bs
->dev_opaque
= NULL
;
1584 bs
->buffer_alignment
= 512;
1587 /* TODO change to return DeviceState * when all users are qdevified */
1588 void *bdrv_get_attached_dev(BlockDriverState
*bs
)
1593 void bdrv_set_dev_ops(BlockDriverState
*bs
, const BlockDevOps
*ops
,
1597 bs
->dev_opaque
= opaque
;
1598 if (bdrv_dev_has_removable_media(bs
) && bs
== bs_snapshots
) {
1599 bs_snapshots
= NULL
;
1603 void bdrv_emit_qmp_error_event(const BlockDriverState
*bdrv
,
1604 enum MonitorEvent ev
,
1605 BlockErrorAction action
, bool is_read
)
1608 const char *action_str
;
1611 case BDRV_ACTION_REPORT
:
1612 action_str
= "report";
1614 case BDRV_ACTION_IGNORE
:
1615 action_str
= "ignore";
1617 case BDRV_ACTION_STOP
:
1618 action_str
= "stop";
1624 data
= qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1627 is_read
? "read" : "write");
1628 monitor_protocol_event(ev
, data
);
1630 qobject_decref(data
);
1633 static void bdrv_emit_qmp_eject_event(BlockDriverState
*bs
, bool ejected
)
1637 data
= qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }",
1638 bdrv_get_device_name(bs
), ejected
);
1639 monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED
, data
);
1641 qobject_decref(data
);
1644 static void bdrv_dev_change_media_cb(BlockDriverState
*bs
, bool load
)
1646 if (bs
->dev_ops
&& bs
->dev_ops
->change_media_cb
) {
1647 bool tray_was_closed
= !bdrv_dev_is_tray_open(bs
);
1648 bs
->dev_ops
->change_media_cb(bs
->dev_opaque
, load
);
1649 if (tray_was_closed
) {
1651 bdrv_emit_qmp_eject_event(bs
, true);
1655 bdrv_emit_qmp_eject_event(bs
, false);
1660 bool bdrv_dev_has_removable_media(BlockDriverState
*bs
)
1662 return !bs
->dev
|| (bs
->dev_ops
&& bs
->dev_ops
->change_media_cb
);
1665 void bdrv_dev_eject_request(BlockDriverState
*bs
, bool force
)
1667 if (bs
->dev_ops
&& bs
->dev_ops
->eject_request_cb
) {
1668 bs
->dev_ops
->eject_request_cb(bs
->dev_opaque
, force
);
1672 bool bdrv_dev_is_tray_open(BlockDriverState
*bs
)
1674 if (bs
->dev_ops
&& bs
->dev_ops
->is_tray_open
) {
1675 return bs
->dev_ops
->is_tray_open(bs
->dev_opaque
);
1680 static void bdrv_dev_resize_cb(BlockDriverState
*bs
)
1682 if (bs
->dev_ops
&& bs
->dev_ops
->resize_cb
) {
1683 bs
->dev_ops
->resize_cb(bs
->dev_opaque
);
1687 bool bdrv_dev_is_medium_locked(BlockDriverState
*bs
)
1689 if (bs
->dev_ops
&& bs
->dev_ops
->is_medium_locked
) {
1690 return bs
->dev_ops
->is_medium_locked(bs
->dev_opaque
);
1696 * Run consistency checks on an image
1698 * Returns 0 if the check could be completed (it doesn't mean that the image is
1699 * free of errors) or -errno when an internal error occurred. The results of the
1700 * check are stored in res.
1702 int bdrv_check(BlockDriverState
*bs
, BdrvCheckResult
*res
, BdrvCheckMode fix
)
1704 if (bs
->drv
->bdrv_check
== NULL
) {
1708 memset(res
, 0, sizeof(*res
));
1709 return bs
->drv
->bdrv_check(bs
, res
, fix
);
1712 #define COMMIT_BUF_SECTORS 2048
1714 /* commit COW file into the raw image */
1715 int bdrv_commit(BlockDriverState
*bs
)
1717 BlockDriver
*drv
= bs
->drv
;
1718 int64_t sector
, total_sectors
;
1719 int n
, ro
, open_flags
;
1722 char filename
[PATH_MAX
];
1727 if (!bs
->backing_hd
) {
1731 if (bdrv_in_use(bs
) || bdrv_in_use(bs
->backing_hd
)) {
1735 ro
= bs
->backing_hd
->read_only
;
1736 /* Use pstrcpy (not strncpy): filename must be NUL-terminated. */
1737 pstrcpy(filename
, sizeof(filename
), bs
->backing_hd
->filename
);
1738 open_flags
= bs
->backing_hd
->open_flags
;
1741 if (bdrv_reopen(bs
->backing_hd
, open_flags
| BDRV_O_RDWR
, NULL
)) {
1746 total_sectors
= bdrv_getlength(bs
) >> BDRV_SECTOR_BITS
;
1747 buf
= g_malloc(COMMIT_BUF_SECTORS
* BDRV_SECTOR_SIZE
);
1749 for (sector
= 0; sector
< total_sectors
; sector
+= n
) {
1750 if (bdrv_is_allocated(bs
, sector
, COMMIT_BUF_SECTORS
, &n
)) {
1752 if (bdrv_read(bs
, sector
, buf
, n
) != 0) {
1757 if (bdrv_write(bs
->backing_hd
, sector
, buf
, n
) != 0) {
1764 if (drv
->bdrv_make_empty
) {
1765 ret
= drv
->bdrv_make_empty(bs
);
1770 * Make sure all data we wrote to the backing device is actually
1774 bdrv_flush(bs
->backing_hd
);
1780 /* ignoring error return here */
1781 bdrv_reopen(bs
->backing_hd
, open_flags
& ~BDRV_O_RDWR
, NULL
);
1787 int bdrv_commit_all(void)
1789 BlockDriverState
*bs
;
1791 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1792 if (bs
->drv
&& bs
->backing_hd
) {
1793 int ret
= bdrv_commit(bs
);
1802 struct BdrvTrackedRequest
{
1803 BlockDriverState
*bs
;
1807 QLIST_ENTRY(BdrvTrackedRequest
) list
;
1808 Coroutine
*co
; /* owner, used for deadlock detection */
1809 CoQueue wait_queue
; /* coroutines blocked on this request */
1813 * Remove an active request from the tracked requests list
1815 * This function should be called when a tracked request is completing.
1817 static void tracked_request_end(BdrvTrackedRequest
*req
)
1819 QLIST_REMOVE(req
, list
);
1820 qemu_co_queue_restart_all(&req
->wait_queue
);
1824 * Add an active request to the tracked requests list
1826 static void tracked_request_begin(BdrvTrackedRequest
*req
,
1827 BlockDriverState
*bs
,
1829 int nb_sectors
, bool is_write
)
1831 *req
= (BdrvTrackedRequest
){
1833 .sector_num
= sector_num
,
1834 .nb_sectors
= nb_sectors
,
1835 .is_write
= is_write
,
1836 .co
= qemu_coroutine_self(),
1839 qemu_co_queue_init(&req
->wait_queue
);
1841 QLIST_INSERT_HEAD(&bs
->tracked_requests
, req
, list
);
1845 * Round a region to cluster boundaries
1847 void bdrv_round_to_clusters(BlockDriverState
*bs
,
1848 int64_t sector_num
, int nb_sectors
,
1849 int64_t *cluster_sector_num
,
1850 int *cluster_nb_sectors
)
1852 BlockDriverInfo bdi
;
1854 if (bdrv_get_info(bs
, &bdi
) < 0 || bdi
.cluster_size
== 0) {
1855 *cluster_sector_num
= sector_num
;
1856 *cluster_nb_sectors
= nb_sectors
;
1858 int64_t c
= bdi
.cluster_size
/ BDRV_SECTOR_SIZE
;
1859 *cluster_sector_num
= QEMU_ALIGN_DOWN(sector_num
, c
);
1860 *cluster_nb_sectors
= QEMU_ALIGN_UP(sector_num
- *cluster_sector_num
+
1865 static bool tracked_request_overlaps(BdrvTrackedRequest
*req
,
1866 int64_t sector_num
, int nb_sectors
) {
1868 if (sector_num
>= req
->sector_num
+ req
->nb_sectors
) {
1872 if (req
->sector_num
>= sector_num
+ nb_sectors
) {
1878 static void coroutine_fn
wait_for_overlapping_requests(BlockDriverState
*bs
,
1879 int64_t sector_num
, int nb_sectors
)
1881 BdrvTrackedRequest
*req
;
1882 int64_t cluster_sector_num
;
1883 int cluster_nb_sectors
;
1886 /* If we touch the same cluster it counts as an overlap. This guarantees
1887 * that allocating writes will be serialized and not race with each other
1888 * for the same cluster. For example, in copy-on-read it ensures that the
1889 * CoR read and write operations are atomic and guest writes cannot
1890 * interleave between them.
1892 bdrv_round_to_clusters(bs
, sector_num
, nb_sectors
,
1893 &cluster_sector_num
, &cluster_nb_sectors
);
1897 QLIST_FOREACH(req
, &bs
->tracked_requests
, list
) {
1898 if (tracked_request_overlaps(req
, cluster_sector_num
,
1899 cluster_nb_sectors
)) {
1900 /* Hitting this means there was a reentrant request, for
1901 * example, a block driver issuing nested requests. This must
1902 * never happen since it means deadlock.
1904 assert(qemu_coroutine_self() != req
->co
);
1906 qemu_co_queue_wait(&req
->wait_queue
);
1917 * -EINVAL - backing format specified, but no file
1918 * -ENOSPC - can't update the backing file because no space is left in the
1920 * -ENOTSUP - format driver doesn't support changing the backing file
1922 int bdrv_change_backing_file(BlockDriverState
*bs
,
1923 const char *backing_file
, const char *backing_fmt
)
1925 BlockDriver
*drv
= bs
->drv
;
1928 /* Backing file format doesn't make sense without a backing file */
1929 if (backing_fmt
&& !backing_file
) {
1933 if (drv
->bdrv_change_backing_file
!= NULL
) {
1934 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
1940 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
1941 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
1947 * Finds the image layer in the chain that has 'bs' as its backing file.
1949 * active is the current topmost image.
1951 * Returns NULL if bs is not found in active's image chain,
1952 * or if active == bs.
1954 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
1955 BlockDriverState
*bs
)
1957 BlockDriverState
*overlay
= NULL
;
1958 BlockDriverState
*intermediate
;
1960 assert(active
!= NULL
);
1963 /* if bs is the same as active, then by definition it has no overlay
1969 intermediate
= active
;
1970 while (intermediate
->backing_hd
) {
1971 if (intermediate
->backing_hd
== bs
) {
1972 overlay
= intermediate
;
1975 intermediate
= intermediate
->backing_hd
;
1981 typedef struct BlkIntermediateStates
{
1982 BlockDriverState
*bs
;
1983 QSIMPLEQ_ENTRY(BlkIntermediateStates
) entry
;
1984 } BlkIntermediateStates
;
1988 * Drops images above 'base' up to and including 'top', and sets the image
1989 * above 'top' to have base as its backing file.
1991 * Requires that the overlay to 'top' is opened r/w, so that the backing file
1992 * information in 'bs' can be properly updated.
1994 * E.g., this will convert the following chain:
1995 * bottom <- base <- intermediate <- top <- active
1999 * bottom <- base <- active
2001 * It is allowed for bottom==base, in which case it converts:
2003 * base <- intermediate <- top <- active
2010 * if active == top, that is considered an error
2013 int bdrv_drop_intermediate(BlockDriverState
*active
, BlockDriverState
*top
,
2014 BlockDriverState
*base
)
2016 BlockDriverState
*intermediate
;
2017 BlockDriverState
*base_bs
= NULL
;
2018 BlockDriverState
*new_top_bs
= NULL
;
2019 BlkIntermediateStates
*intermediate_state
, *next
;
2022 QSIMPLEQ_HEAD(states_to_delete
, BlkIntermediateStates
) states_to_delete
;
2023 QSIMPLEQ_INIT(&states_to_delete
);
2025 if (!top
->drv
|| !base
->drv
) {
2029 new_top_bs
= bdrv_find_overlay(active
, top
);
2031 if (new_top_bs
== NULL
) {
2032 /* we could not find the image above 'top', this is an error */
2036 /* special case of new_top_bs->backing_hd already pointing to base - nothing
2037 * to do, no intermediate images */
2038 if (new_top_bs
->backing_hd
== base
) {
2045 /* now we will go down through the list, and add each BDS we find
2046 * into our deletion queue, until we hit the 'base'
2048 while (intermediate
) {
2049 intermediate_state
= g_malloc0(sizeof(BlkIntermediateStates
));
2050 intermediate_state
->bs
= intermediate
;
2051 QSIMPLEQ_INSERT_TAIL(&states_to_delete
, intermediate_state
, entry
);
2053 if (intermediate
->backing_hd
== base
) {
2054 base_bs
= intermediate
->backing_hd
;
2057 intermediate
= intermediate
->backing_hd
;
2059 if (base_bs
== NULL
) {
2060 /* something went wrong, we did not end at the base. safely
2061 * unravel everything, and exit with error */
2065 /* success - we can delete the intermediate states, and link top->base */
2066 ret
= bdrv_change_backing_file(new_top_bs
, base_bs
->filename
,
2067 base_bs
->drv
? base_bs
->drv
->format_name
: "");
2071 new_top_bs
->backing_hd
= base_bs
;
2074 QSIMPLEQ_FOREACH_SAFE(intermediate_state
, &states_to_delete
, entry
, next
) {
2075 /* so that bdrv_close() does not recursively close the chain */
2076 intermediate_state
->bs
->backing_hd
= NULL
;
2077 bdrv_delete(intermediate_state
->bs
);
2082 QSIMPLEQ_FOREACH_SAFE(intermediate_state
, &states_to_delete
, entry
, next
) {
2083 g_free(intermediate_state
);
2089 static int bdrv_check_byte_request(BlockDriverState
*bs
, int64_t offset
,
2094 if (!bdrv_is_inserted(bs
))
2100 len
= bdrv_getlength(bs
);
2105 if ((offset
> len
) || (len
- offset
< size
))
2111 static int bdrv_check_request(BlockDriverState
*bs
, int64_t sector_num
,
2114 return bdrv_check_byte_request(bs
, sector_num
* BDRV_SECTOR_SIZE
,
2115 nb_sectors
* BDRV_SECTOR_SIZE
);
2118 typedef struct RwCo
{
2119 BlockDriverState
*bs
;
2127 static void coroutine_fn
bdrv_rw_co_entry(void *opaque
)
2129 RwCo
*rwco
= opaque
;
2131 if (!rwco
->is_write
) {
2132 rwco
->ret
= bdrv_co_do_readv(rwco
->bs
, rwco
->sector_num
,
2133 rwco
->nb_sectors
, rwco
->qiov
, 0);
2135 rwco
->ret
= bdrv_co_do_writev(rwco
->bs
, rwco
->sector_num
,
2136 rwco
->nb_sectors
, rwco
->qiov
, 0);
2141 * Process a vectored synchronous request using coroutines
2143 static int bdrv_rwv_co(BlockDriverState
*bs
, int64_t sector_num
,
2144 QEMUIOVector
*qiov
, bool is_write
)
2149 .sector_num
= sector_num
,
2150 .nb_sectors
= qiov
->size
>> BDRV_SECTOR_BITS
,
2152 .is_write
= is_write
,
2155 assert((qiov
->size
& (BDRV_SECTOR_SIZE
- 1)) == 0);
2158 * In sync call context, when the vcpu is blocked, this throttling timer
2159 * will not fire; so the I/O throttling function has to be disabled here
2160 * if it has been enabled.
2162 if (bs
->io_limits_enabled
) {
2163 fprintf(stderr
, "Disabling I/O throttling on '%s' due "
2164 "to synchronous I/O.\n", bdrv_get_device_name(bs
));
2165 bdrv_io_limits_disable(bs
);
2168 if (qemu_in_coroutine()) {
2169 /* Fast-path if already in coroutine context */
2170 bdrv_rw_co_entry(&rwco
);
2172 co
= qemu_coroutine_create(bdrv_rw_co_entry
);
2173 qemu_coroutine_enter(co
, &rwco
);
2174 while (rwco
.ret
== NOT_DONE
) {
2182 * Process a synchronous request using coroutines
2184 static int bdrv_rw_co(BlockDriverState
*bs
, int64_t sector_num
, uint8_t *buf
,
2185 int nb_sectors
, bool is_write
)
2188 struct iovec iov
= {
2189 .iov_base
= (void *)buf
,
2190 .iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
,
2193 qemu_iovec_init_external(&qiov
, &iov
, 1);
2194 return bdrv_rwv_co(bs
, sector_num
, &qiov
, is_write
);
2197 /* return < 0 if error. See bdrv_write() for the return codes */
2198 int bdrv_read(BlockDriverState
*bs
, int64_t sector_num
,
2199 uint8_t *buf
, int nb_sectors
)
2201 return bdrv_rw_co(bs
, sector_num
, buf
, nb_sectors
, false);
2204 /* Just like bdrv_read(), but with I/O throttling temporarily disabled */
2205 int bdrv_read_unthrottled(BlockDriverState
*bs
, int64_t sector_num
,
2206 uint8_t *buf
, int nb_sectors
)
2211 enabled
= bs
->io_limits_enabled
;
2212 bs
->io_limits_enabled
= false;
2213 ret
= bdrv_read(bs
, 0, buf
, 1);
2214 bs
->io_limits_enabled
= enabled
;
2218 /* Return < 0 if error. Important errors are:
2219 -EIO generic I/O error (may happen for all errors)
2220 -ENOMEDIUM No media inserted.
2221 -EINVAL Invalid sector number or nb_sectors
2222 -EACCES Trying to write a read-only device
2224 int bdrv_write(BlockDriverState
*bs
, int64_t sector_num
,
2225 const uint8_t *buf
, int nb_sectors
)
2227 return bdrv_rw_co(bs
, sector_num
, (uint8_t *)buf
, nb_sectors
, true);
2230 int bdrv_writev(BlockDriverState
*bs
, int64_t sector_num
, QEMUIOVector
*qiov
)
2232 return bdrv_rwv_co(bs
, sector_num
, qiov
, true);
2235 int bdrv_pread(BlockDriverState
*bs
, int64_t offset
,
2236 void *buf
, int count1
)
2238 uint8_t tmp_buf
[BDRV_SECTOR_SIZE
];
2239 int len
, nb_sectors
, count
;
2244 /* first read to align to sector start */
2245 len
= (BDRV_SECTOR_SIZE
- offset
) & (BDRV_SECTOR_SIZE
- 1);
2248 sector_num
= offset
>> BDRV_SECTOR_BITS
;
2250 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
2252 memcpy(buf
, tmp_buf
+ (offset
& (BDRV_SECTOR_SIZE
- 1)), len
);
2260 /* read the sectors "in place" */
2261 nb_sectors
= count
>> BDRV_SECTOR_BITS
;
2262 if (nb_sectors
> 0) {
2263 if ((ret
= bdrv_read(bs
, sector_num
, buf
, nb_sectors
)) < 0)
2265 sector_num
+= nb_sectors
;
2266 len
= nb_sectors
<< BDRV_SECTOR_BITS
;
2271 /* add data from the last sector */
2273 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
2275 memcpy(buf
, tmp_buf
, count
);
2280 int bdrv_pwritev(BlockDriverState
*bs
, int64_t offset
, QEMUIOVector
*qiov
)
2282 uint8_t tmp_buf
[BDRV_SECTOR_SIZE
];
2283 int len
, nb_sectors
, count
;
2289 /* first write to align to sector start */
2290 len
= (BDRV_SECTOR_SIZE
- offset
) & (BDRV_SECTOR_SIZE
- 1);
2293 sector_num
= offset
>> BDRV_SECTOR_BITS
;
2295 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
2297 qemu_iovec_to_buf(qiov
, 0, tmp_buf
+ (offset
& (BDRV_SECTOR_SIZE
- 1)),
2299 if ((ret
= bdrv_write(bs
, sector_num
, tmp_buf
, 1)) < 0)
2307 /* write the sectors "in place" */
2308 nb_sectors
= count
>> BDRV_SECTOR_BITS
;
2309 if (nb_sectors
> 0) {
2310 QEMUIOVector qiov_inplace
;
2312 qemu_iovec_init(&qiov_inplace
, qiov
->niov
);
2313 qemu_iovec_concat(&qiov_inplace
, qiov
, len
,
2314 nb_sectors
<< BDRV_SECTOR_BITS
);
2315 ret
= bdrv_writev(bs
, sector_num
, &qiov_inplace
);
2316 qemu_iovec_destroy(&qiov_inplace
);
2321 sector_num
+= nb_sectors
;
2322 len
= nb_sectors
<< BDRV_SECTOR_BITS
;
2326 /* add data from the last sector */
2328 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
2330 qemu_iovec_to_buf(qiov
, qiov
->size
- count
, tmp_buf
, count
);
2331 if ((ret
= bdrv_write(bs
, sector_num
, tmp_buf
, 1)) < 0)
2337 int bdrv_pwrite(BlockDriverState
*bs
, int64_t offset
,
2338 const void *buf
, int count1
)
2341 struct iovec iov
= {
2342 .iov_base
= (void *) buf
,
2346 qemu_iovec_init_external(&qiov
, &iov
, 1);
2347 return bdrv_pwritev(bs
, offset
, &qiov
);
2351 * Writes to the file and ensures that no writes are reordered across this
2352 * request (acts as a barrier)
2354 * Returns 0 on success, -errno in error cases.
2356 int bdrv_pwrite_sync(BlockDriverState
*bs
, int64_t offset
,
2357 const void *buf
, int count
)
2361 ret
= bdrv_pwrite(bs
, offset
, buf
, count
);
2366 /* No flush needed for cache modes that already do it */
2367 if (bs
->enable_write_cache
) {
2374 static int coroutine_fn
bdrv_co_do_copy_on_readv(BlockDriverState
*bs
,
2375 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
2377 /* Perform I/O through a temporary buffer so that users who scribble over
2378 * their read buffer while the operation is in progress do not end up
2379 * modifying the image file. This is critical for zero-copy guest I/O
2380 * where anything might happen inside guest memory.
2382 void *bounce_buffer
;
2384 BlockDriver
*drv
= bs
->drv
;
2386 QEMUIOVector bounce_qiov
;
2387 int64_t cluster_sector_num
;
2388 int cluster_nb_sectors
;
2392 /* Cover entire cluster so no additional backing file I/O is required when
2393 * allocating cluster in the image file.
2395 bdrv_round_to_clusters(bs
, sector_num
, nb_sectors
,
2396 &cluster_sector_num
, &cluster_nb_sectors
);
2398 trace_bdrv_co_do_copy_on_readv(bs
, sector_num
, nb_sectors
,
2399 cluster_sector_num
, cluster_nb_sectors
);
2401 iov
.iov_len
= cluster_nb_sectors
* BDRV_SECTOR_SIZE
;
2402 iov
.iov_base
= bounce_buffer
= qemu_blockalign(bs
, iov
.iov_len
);
2403 qemu_iovec_init_external(&bounce_qiov
, &iov
, 1);
2405 ret
= drv
->bdrv_co_readv(bs
, cluster_sector_num
, cluster_nb_sectors
,
2411 if (drv
->bdrv_co_write_zeroes
&&
2412 buffer_is_zero(bounce_buffer
, iov
.iov_len
)) {
2413 ret
= bdrv_co_do_write_zeroes(bs
, cluster_sector_num
,
2414 cluster_nb_sectors
);
2416 /* This does not change the data on the disk, it is not necessary
2417 * to flush even in cache=writethrough mode.
2419 ret
= drv
->bdrv_co_writev(bs
, cluster_sector_num
, cluster_nb_sectors
,
2424 /* It might be okay to ignore write errors for guest requests. If this
2425 * is a deliberate copy-on-read then we don't want to ignore the error.
2426 * Simply report it in all cases.
2431 skip_bytes
= (sector_num
- cluster_sector_num
) * BDRV_SECTOR_SIZE
;
2432 qemu_iovec_from_buf(qiov
, 0, bounce_buffer
+ skip_bytes
,
2433 nb_sectors
* BDRV_SECTOR_SIZE
);
2436 qemu_vfree(bounce_buffer
);
2441 * Handle a read request in coroutine context
2443 static int coroutine_fn
bdrv_co_do_readv(BlockDriverState
*bs
,
2444 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
2445 BdrvRequestFlags flags
)
2447 BlockDriver
*drv
= bs
->drv
;
2448 BdrvTrackedRequest req
;
2454 if (bdrv_check_request(bs
, sector_num
, nb_sectors
)) {
2458 /* throttling disk read I/O */
2459 if (bs
->io_limits_enabled
) {
2460 bdrv_io_limits_intercept(bs
, false, nb_sectors
);
2463 if (bs
->copy_on_read
) {
2464 flags
|= BDRV_REQ_COPY_ON_READ
;
2466 if (flags
& BDRV_REQ_COPY_ON_READ
) {
2467 bs
->copy_on_read_in_flight
++;
2470 if (bs
->copy_on_read_in_flight
) {
2471 wait_for_overlapping_requests(bs
, sector_num
, nb_sectors
);
2474 tracked_request_begin(&req
, bs
, sector_num
, nb_sectors
, false);
2476 if (flags
& BDRV_REQ_COPY_ON_READ
) {
2479 ret
= bdrv_co_is_allocated(bs
, sector_num
, nb_sectors
, &pnum
);
2484 if (!ret
|| pnum
!= nb_sectors
) {
2485 ret
= bdrv_co_do_copy_on_readv(bs
, sector_num
, nb_sectors
, qiov
);
2490 ret
= drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
2493 tracked_request_end(&req
);
2495 if (flags
& BDRV_REQ_COPY_ON_READ
) {
2496 bs
->copy_on_read_in_flight
--;
2502 int coroutine_fn
bdrv_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
2503 int nb_sectors
, QEMUIOVector
*qiov
)
2505 trace_bdrv_co_readv(bs
, sector_num
, nb_sectors
);
2507 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
, 0);
2510 int coroutine_fn
bdrv_co_copy_on_readv(BlockDriverState
*bs
,
2511 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
2513 trace_bdrv_co_copy_on_readv(bs
, sector_num
, nb_sectors
);
2515 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
,
2516 BDRV_REQ_COPY_ON_READ
);
2519 static int coroutine_fn
bdrv_co_do_write_zeroes(BlockDriverState
*bs
,
2520 int64_t sector_num
, int nb_sectors
)
2522 BlockDriver
*drv
= bs
->drv
;
2527 /* TODO Emulate only part of misaligned requests instead of letting block
2528 * drivers return -ENOTSUP and emulate everything */
2530 /* First try the efficient write zeroes operation */
2531 if (drv
->bdrv_co_write_zeroes
) {
2532 ret
= drv
->bdrv_co_write_zeroes(bs
, sector_num
, nb_sectors
);
2533 if (ret
!= -ENOTSUP
) {
2538 /* Fall back to bounce buffer if write zeroes is unsupported */
2539 iov
.iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
;
2540 iov
.iov_base
= qemu_blockalign(bs
, iov
.iov_len
);
2541 memset(iov
.iov_base
, 0, iov
.iov_len
);
2542 qemu_iovec_init_external(&qiov
, &iov
, 1);
2544 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, &qiov
);
2546 qemu_vfree(iov
.iov_base
);
2551 * Handle a write request in coroutine context
2553 static int coroutine_fn
bdrv_co_do_writev(BlockDriverState
*bs
,
2554 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
2555 BdrvRequestFlags flags
)
2557 BlockDriver
*drv
= bs
->drv
;
2558 BdrvTrackedRequest req
;
2564 if (bs
->read_only
) {
2567 if (bdrv_check_request(bs
, sector_num
, nb_sectors
)) {
2571 /* throttling disk write I/O */
2572 if (bs
->io_limits_enabled
) {
2573 bdrv_io_limits_intercept(bs
, true, nb_sectors
);
2576 if (bs
->copy_on_read_in_flight
) {
2577 wait_for_overlapping_requests(bs
, sector_num
, nb_sectors
);
2580 tracked_request_begin(&req
, bs
, sector_num
, nb_sectors
, true);
2582 if (flags
& BDRV_REQ_ZERO_WRITE
) {
2583 ret
= bdrv_co_do_write_zeroes(bs
, sector_num
, nb_sectors
);
2585 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, qiov
);
2588 if (ret
== 0 && !bs
->enable_write_cache
) {
2589 ret
= bdrv_co_flush(bs
);
2592 if (bs
->dirty_bitmap
) {
2593 bdrv_set_dirty(bs
, sector_num
, nb_sectors
);
2596 if (bs
->wr_highest_sector
< sector_num
+ nb_sectors
- 1) {
2597 bs
->wr_highest_sector
= sector_num
+ nb_sectors
- 1;
2600 tracked_request_end(&req
);
2605 int coroutine_fn
bdrv_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
2606 int nb_sectors
, QEMUIOVector
*qiov
)
2608 trace_bdrv_co_writev(bs
, sector_num
, nb_sectors
);
2610 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, qiov
, 0);
2613 int coroutine_fn
bdrv_co_write_zeroes(BlockDriverState
*bs
,
2614 int64_t sector_num
, int nb_sectors
)
2616 trace_bdrv_co_write_zeroes(bs
, sector_num
, nb_sectors
);
2618 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, NULL
,
2619 BDRV_REQ_ZERO_WRITE
);
2623 * Truncate file to 'offset' bytes (needed only for file protocols)
2625 int bdrv_truncate(BlockDriverState
*bs
, int64_t offset
)
2627 BlockDriver
*drv
= bs
->drv
;
2631 if (!drv
->bdrv_truncate
)
2635 if (bdrv_in_use(bs
))
2637 ret
= drv
->bdrv_truncate(bs
, offset
);
2639 ret
= refresh_total_sectors(bs
, offset
>> BDRV_SECTOR_BITS
);
2640 bdrv_dev_resize_cb(bs
);
2646 * Length of a allocated file in bytes. Sparse files are counted by actual
2647 * allocated space. Return < 0 if error or unknown.
2649 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
2651 BlockDriver
*drv
= bs
->drv
;
2655 if (drv
->bdrv_get_allocated_file_size
) {
2656 return drv
->bdrv_get_allocated_file_size(bs
);
2659 return bdrv_get_allocated_file_size(bs
->file
);
2665 * Length of a file in bytes. Return < 0 if error or unknown.
2667 int64_t bdrv_getlength(BlockDriverState
*bs
)
2669 BlockDriver
*drv
= bs
->drv
;
2673 if (bs
->growable
|| bdrv_dev_has_removable_media(bs
)) {
2674 if (drv
->bdrv_getlength
) {
2675 return drv
->bdrv_getlength(bs
);
2678 return bs
->total_sectors
* BDRV_SECTOR_SIZE
;
2681 /* return 0 as number of sectors if no device present or error */
2682 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
2685 length
= bdrv_getlength(bs
);
2689 length
= length
>> BDRV_SECTOR_BITS
;
2690 *nb_sectors_ptr
= length
;
2693 /* throttling disk io limits */
2694 void bdrv_set_io_limits(BlockDriverState
*bs
,
2695 BlockIOLimit
*io_limits
)
2697 bs
->io_limits
= *io_limits
;
2698 bs
->io_limits_enabled
= bdrv_io_limits_enabled(bs
);
2701 void bdrv_set_on_error(BlockDriverState
*bs
, BlockdevOnError on_read_error
,
2702 BlockdevOnError on_write_error
)
2704 bs
->on_read_error
= on_read_error
;
2705 bs
->on_write_error
= on_write_error
;
2708 BlockdevOnError
bdrv_get_on_error(BlockDriverState
*bs
, bool is_read
)
2710 return is_read
? bs
->on_read_error
: bs
->on_write_error
;
2713 BlockErrorAction
bdrv_get_error_action(BlockDriverState
*bs
, bool is_read
, int error
)
2715 BlockdevOnError on_err
= is_read
? bs
->on_read_error
: bs
->on_write_error
;
2718 case BLOCKDEV_ON_ERROR_ENOSPC
:
2719 return (error
== ENOSPC
) ? BDRV_ACTION_STOP
: BDRV_ACTION_REPORT
;
2720 case BLOCKDEV_ON_ERROR_STOP
:
2721 return BDRV_ACTION_STOP
;
2722 case BLOCKDEV_ON_ERROR_REPORT
:
2723 return BDRV_ACTION_REPORT
;
2724 case BLOCKDEV_ON_ERROR_IGNORE
:
2725 return BDRV_ACTION_IGNORE
;
2731 /* This is done by device models because, while the block layer knows
2732 * about the error, it does not know whether an operation comes from
2733 * the device or the block layer (from a job, for example).
2735 void bdrv_error_action(BlockDriverState
*bs
, BlockErrorAction action
,
2736 bool is_read
, int error
)
2739 bdrv_emit_qmp_error_event(bs
, QEVENT_BLOCK_IO_ERROR
, action
, is_read
);
2740 if (action
== BDRV_ACTION_STOP
) {
2741 vm_stop(RUN_STATE_IO_ERROR
);
2742 bdrv_iostatus_set_err(bs
, error
);
2746 int bdrv_is_read_only(BlockDriverState
*bs
)
2748 return bs
->read_only
;
2751 int bdrv_is_sg(BlockDriverState
*bs
)
2756 int bdrv_enable_write_cache(BlockDriverState
*bs
)
2758 return bs
->enable_write_cache
;
2761 void bdrv_set_enable_write_cache(BlockDriverState
*bs
, bool wce
)
2763 bs
->enable_write_cache
= wce
;
2765 /* so a reopen() will preserve wce */
2767 bs
->open_flags
|= BDRV_O_CACHE_WB
;
2769 bs
->open_flags
&= ~BDRV_O_CACHE_WB
;
2773 int bdrv_is_encrypted(BlockDriverState
*bs
)
2775 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
)
2777 return bs
->encrypted
;
2780 int bdrv_key_required(BlockDriverState
*bs
)
2782 BlockDriverState
*backing_hd
= bs
->backing_hd
;
2784 if (backing_hd
&& backing_hd
->encrypted
&& !backing_hd
->valid_key
)
2786 return (bs
->encrypted
&& !bs
->valid_key
);
2789 int bdrv_set_key(BlockDriverState
*bs
, const char *key
)
2792 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
) {
2793 ret
= bdrv_set_key(bs
->backing_hd
, key
);
2799 if (!bs
->encrypted
) {
2801 } else if (!bs
->drv
|| !bs
->drv
->bdrv_set_key
) {
2804 ret
= bs
->drv
->bdrv_set_key(bs
, key
);
2807 } else if (!bs
->valid_key
) {
2809 /* call the change callback now, we skipped it on open */
2810 bdrv_dev_change_media_cb(bs
, true);
2815 const char *bdrv_get_format_name(BlockDriverState
*bs
)
2817 return bs
->drv
? bs
->drv
->format_name
: NULL
;
2820 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
2825 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
2826 it(opaque
, drv
->format_name
);
2830 BlockDriverState
*bdrv_find(const char *name
)
2832 BlockDriverState
*bs
;
2834 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2835 if (!strcmp(name
, bs
->device_name
)) {
2842 BlockDriverState
*bdrv_next(BlockDriverState
*bs
)
2845 return QTAILQ_FIRST(&bdrv_states
);
2847 return QTAILQ_NEXT(bs
, list
);
2850 void bdrv_iterate(void (*it
)(void *opaque
, BlockDriverState
*bs
), void *opaque
)
2852 BlockDriverState
*bs
;
2854 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2859 const char *bdrv_get_device_name(BlockDriverState
*bs
)
2861 return bs
->device_name
;
2864 int bdrv_get_flags(BlockDriverState
*bs
)
2866 return bs
->open_flags
;
2869 void bdrv_flush_all(void)
2871 BlockDriverState
*bs
;
2873 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2878 int bdrv_has_zero_init(BlockDriverState
*bs
)
2882 if (bs
->drv
->bdrv_has_zero_init
) {
2883 return bs
->drv
->bdrv_has_zero_init(bs
);
2889 typedef struct BdrvCoIsAllocatedData
{
2890 BlockDriverState
*bs
;
2891 BlockDriverState
*base
;
2897 } BdrvCoIsAllocatedData
;
2900 * Returns true iff the specified sector is present in the disk image. Drivers
2901 * not implementing the functionality are assumed to not support backing files,
2902 * hence all their sectors are reported as allocated.
2904 * If 'sector_num' is beyond the end of the disk image the return value is 0
2905 * and 'pnum' is set to 0.
2907 * 'pnum' is set to the number of sectors (including and immediately following
2908 * the specified sector) that are known to be in the same
2909 * allocated/unallocated state.
2911 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
2912 * beyond the end of the disk image it will be clamped.
2914 int coroutine_fn
bdrv_co_is_allocated(BlockDriverState
*bs
, int64_t sector_num
,
2915 int nb_sectors
, int *pnum
)
2919 if (sector_num
>= bs
->total_sectors
) {
2924 n
= bs
->total_sectors
- sector_num
;
2925 if (n
< nb_sectors
) {
2929 if (!bs
->drv
->bdrv_co_is_allocated
) {
2934 return bs
->drv
->bdrv_co_is_allocated(bs
, sector_num
, nb_sectors
, pnum
);
2937 /* Coroutine wrapper for bdrv_is_allocated() */
2938 static void coroutine_fn
bdrv_is_allocated_co_entry(void *opaque
)
2940 BdrvCoIsAllocatedData
*data
= opaque
;
2941 BlockDriverState
*bs
= data
->bs
;
2943 data
->ret
= bdrv_co_is_allocated(bs
, data
->sector_num
, data
->nb_sectors
,
2949 * Synchronous wrapper around bdrv_co_is_allocated().
2951 * See bdrv_co_is_allocated() for details.
2953 int bdrv_is_allocated(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
,
2957 BdrvCoIsAllocatedData data
= {
2959 .sector_num
= sector_num
,
2960 .nb_sectors
= nb_sectors
,
2965 co
= qemu_coroutine_create(bdrv_is_allocated_co_entry
);
2966 qemu_coroutine_enter(co
, &data
);
2967 while (!data
.done
) {
2974 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
2976 * Return true if the given sector is allocated in any image between
2977 * BASE and TOP (inclusive). BASE can be NULL to check if the given
2978 * sector is allocated in any image of the chain. Return false otherwise.
2980 * 'pnum' is set to the number of sectors (including and immediately following
2981 * the specified sector) that are known to be in the same
2982 * allocated/unallocated state.
2985 int coroutine_fn
bdrv_co_is_allocated_above(BlockDriverState
*top
,
2986 BlockDriverState
*base
,
2988 int nb_sectors
, int *pnum
)
2990 BlockDriverState
*intermediate
;
2991 int ret
, n
= nb_sectors
;
2994 while (intermediate
&& intermediate
!= base
) {
2996 ret
= bdrv_co_is_allocated(intermediate
, sector_num
, nb_sectors
,
3006 * [sector_num, nb_sectors] is unallocated on top but intermediate
3009 * [sector_num+x, nr_sectors] allocated.
3011 if (n
> pnum_inter
&&
3012 (intermediate
== top
||
3013 sector_num
+ pnum_inter
< intermediate
->total_sectors
)) {
3017 intermediate
= intermediate
->backing_hd
;
3024 /* Coroutine wrapper for bdrv_is_allocated_above() */
3025 static void coroutine_fn
bdrv_is_allocated_above_co_entry(void *opaque
)
3027 BdrvCoIsAllocatedData
*data
= opaque
;
3028 BlockDriverState
*top
= data
->bs
;
3029 BlockDriverState
*base
= data
->base
;
3031 data
->ret
= bdrv_co_is_allocated_above(top
, base
, data
->sector_num
,
3032 data
->nb_sectors
, data
->pnum
);
3037 * Synchronous wrapper around bdrv_co_is_allocated_above().
3039 * See bdrv_co_is_allocated_above() for details.
3041 int bdrv_is_allocated_above(BlockDriverState
*top
, BlockDriverState
*base
,
3042 int64_t sector_num
, int nb_sectors
, int *pnum
)
3045 BdrvCoIsAllocatedData data
= {
3048 .sector_num
= sector_num
,
3049 .nb_sectors
= nb_sectors
,
3054 co
= qemu_coroutine_create(bdrv_is_allocated_above_co_entry
);
3055 qemu_coroutine_enter(co
, &data
);
3056 while (!data
.done
) {
3062 BlockInfo
*bdrv_query_info(BlockDriverState
*bs
)
3064 BlockInfo
*info
= g_malloc0(sizeof(*info
));
3065 info
->device
= g_strdup(bs
->device_name
);
3066 info
->type
= g_strdup("unknown");
3067 info
->locked
= bdrv_dev_is_medium_locked(bs
);
3068 info
->removable
= bdrv_dev_has_removable_media(bs
);
3070 if (bdrv_dev_has_removable_media(bs
)) {
3071 info
->has_tray_open
= true;
3072 info
->tray_open
= bdrv_dev_is_tray_open(bs
);
3075 if (bdrv_iostatus_is_enabled(bs
)) {
3076 info
->has_io_status
= true;
3077 info
->io_status
= bs
->iostatus
;
3080 if (bs
->dirty_bitmap
) {
3081 info
->has_dirty
= true;
3082 info
->dirty
= g_malloc0(sizeof(*info
->dirty
));
3083 info
->dirty
->count
= bdrv_get_dirty_count(bs
) * BDRV_SECTOR_SIZE
;
3084 info
->dirty
->granularity
=
3085 ((int64_t) BDRV_SECTOR_SIZE
<< hbitmap_granularity(bs
->dirty_bitmap
));
3089 info
->has_inserted
= true;
3090 info
->inserted
= g_malloc0(sizeof(*info
->inserted
));
3091 info
->inserted
->file
= g_strdup(bs
->filename
);
3092 info
->inserted
->ro
= bs
->read_only
;
3093 info
->inserted
->drv
= g_strdup(bs
->drv
->format_name
);
3094 info
->inserted
->encrypted
= bs
->encrypted
;
3095 info
->inserted
->encryption_key_missing
= bdrv_key_required(bs
);
3097 if (bs
->backing_file
[0]) {
3098 info
->inserted
->has_backing_file
= true;
3099 info
->inserted
->backing_file
= g_strdup(bs
->backing_file
);
3102 info
->inserted
->backing_file_depth
= bdrv_get_backing_file_depth(bs
);
3104 if (bs
->io_limits_enabled
) {
3105 info
->inserted
->bps
=
3106 bs
->io_limits
.bps
[BLOCK_IO_LIMIT_TOTAL
];
3107 info
->inserted
->bps_rd
=
3108 bs
->io_limits
.bps
[BLOCK_IO_LIMIT_READ
];
3109 info
->inserted
->bps_wr
=
3110 bs
->io_limits
.bps
[BLOCK_IO_LIMIT_WRITE
];
3111 info
->inserted
->iops
=
3112 bs
->io_limits
.iops
[BLOCK_IO_LIMIT_TOTAL
];
3113 info
->inserted
->iops_rd
=
3114 bs
->io_limits
.iops
[BLOCK_IO_LIMIT_READ
];
3115 info
->inserted
->iops_wr
=
3116 bs
->io_limits
.iops
[BLOCK_IO_LIMIT_WRITE
];
3122 BlockInfoList
*qmp_query_block(Error
**errp
)
3124 BlockInfoList
*head
= NULL
, **p_next
= &head
;
3125 BlockDriverState
*bs
;
3127 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
3128 BlockInfoList
*info
= g_malloc0(sizeof(*info
));
3129 info
->value
= bdrv_query_info(bs
);
3132 p_next
= &info
->next
;
3138 BlockStats
*bdrv_query_stats(const BlockDriverState
*bs
)
3142 s
= g_malloc0(sizeof(*s
));
3144 if (bs
->device_name
[0]) {
3145 s
->has_device
= true;
3146 s
->device
= g_strdup(bs
->device_name
);
3149 s
->stats
= g_malloc0(sizeof(*s
->stats
));
3150 s
->stats
->rd_bytes
= bs
->nr_bytes
[BDRV_ACCT_READ
];
3151 s
->stats
->wr_bytes
= bs
->nr_bytes
[BDRV_ACCT_WRITE
];
3152 s
->stats
->rd_operations
= bs
->nr_ops
[BDRV_ACCT_READ
];
3153 s
->stats
->wr_operations
= bs
->nr_ops
[BDRV_ACCT_WRITE
];
3154 s
->stats
->wr_highest_offset
= bs
->wr_highest_sector
* BDRV_SECTOR_SIZE
;
3155 s
->stats
->flush_operations
= bs
->nr_ops
[BDRV_ACCT_FLUSH
];
3156 s
->stats
->wr_total_time_ns
= bs
->total_time_ns
[BDRV_ACCT_WRITE
];
3157 s
->stats
->rd_total_time_ns
= bs
->total_time_ns
[BDRV_ACCT_READ
];
3158 s
->stats
->flush_total_time_ns
= bs
->total_time_ns
[BDRV_ACCT_FLUSH
];
3161 s
->has_parent
= true;
3162 s
->parent
= bdrv_query_stats(bs
->file
);
3168 BlockStatsList
*qmp_query_blockstats(Error
**errp
)
3170 BlockStatsList
*head
= NULL
, **p_next
= &head
;
3171 BlockDriverState
*bs
;
3173 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
3174 BlockStatsList
*info
= g_malloc0(sizeof(*info
));
3175 info
->value
= bdrv_query_stats(bs
);
3178 p_next
= &info
->next
;
3184 const char *bdrv_get_encrypted_filename(BlockDriverState
*bs
)
3186 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
)
3187 return bs
->backing_file
;
3188 else if (bs
->encrypted
)
3189 return bs
->filename
;
3194 void bdrv_get_backing_filename(BlockDriverState
*bs
,
3195 char *filename
, int filename_size
)
3197 pstrcpy(filename
, filename_size
, bs
->backing_file
);
3200 int bdrv_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
3201 const uint8_t *buf
, int nb_sectors
)
3203 BlockDriver
*drv
= bs
->drv
;
3206 if (!drv
->bdrv_write_compressed
)
3208 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
3211 assert(!bs
->dirty_bitmap
);
3213 return drv
->bdrv_write_compressed(bs
, sector_num
, buf
, nb_sectors
);
3216 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
3218 BlockDriver
*drv
= bs
->drv
;
3221 if (!drv
->bdrv_get_info
)
3223 memset(bdi
, 0, sizeof(*bdi
));
3224 return drv
->bdrv_get_info(bs
, bdi
);
3227 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
3228 int64_t pos
, int size
)
3231 struct iovec iov
= {
3232 .iov_base
= (void *) buf
,
3236 qemu_iovec_init_external(&qiov
, &iov
, 1);
3237 return bdrv_writev_vmstate(bs
, &qiov
, pos
);
3240 int bdrv_writev_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
3242 BlockDriver
*drv
= bs
->drv
;
3246 } else if (drv
->bdrv_save_vmstate
) {
3247 return drv
->bdrv_save_vmstate(bs
, qiov
, pos
);
3248 } else if (bs
->file
) {
3249 return bdrv_writev_vmstate(bs
->file
, qiov
, pos
);
3255 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
3256 int64_t pos
, int size
)
3258 BlockDriver
*drv
= bs
->drv
;
3261 if (drv
->bdrv_load_vmstate
)
3262 return drv
->bdrv_load_vmstate(bs
, buf
, pos
, size
);
3264 return bdrv_load_vmstate(bs
->file
, buf
, pos
, size
);
3268 void bdrv_debug_event(BlockDriverState
*bs
, BlkDebugEvent event
)
3270 BlockDriver
*drv
= bs
->drv
;
3272 if (!drv
|| !drv
->bdrv_debug_event
) {
3276 drv
->bdrv_debug_event(bs
, event
);
3279 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
3282 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
3286 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
3287 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
3293 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
3295 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_resume
) {
3299 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
3300 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
3306 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
3308 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
3312 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
3313 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
3319 /**************************************************************/
3320 /* handling of snapshots */
3322 int bdrv_can_snapshot(BlockDriverState
*bs
)
3324 BlockDriver
*drv
= bs
->drv
;
3325 if (!drv
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
)) {
3329 if (!drv
->bdrv_snapshot_create
) {
3330 if (bs
->file
!= NULL
) {
3331 return bdrv_can_snapshot(bs
->file
);
3339 int bdrv_is_snapshot(BlockDriverState
*bs
)
3341 return !!(bs
->open_flags
& BDRV_O_SNAPSHOT
);
3344 BlockDriverState
*bdrv_snapshots(void)
3346 BlockDriverState
*bs
;
3349 return bs_snapshots
;
3353 while ((bs
= bdrv_next(bs
))) {
3354 if (bdrv_can_snapshot(bs
)) {
3362 int bdrv_snapshot_create(BlockDriverState
*bs
,
3363 QEMUSnapshotInfo
*sn_info
)
3365 BlockDriver
*drv
= bs
->drv
;
3368 if (drv
->bdrv_snapshot_create
)
3369 return drv
->bdrv_snapshot_create(bs
, sn_info
);
3371 return bdrv_snapshot_create(bs
->file
, sn_info
);
3375 int bdrv_snapshot_goto(BlockDriverState
*bs
,
3376 const char *snapshot_id
)
3378 BlockDriver
*drv
= bs
->drv
;
3383 if (drv
->bdrv_snapshot_goto
)
3384 return drv
->bdrv_snapshot_goto(bs
, snapshot_id
);
3387 drv
->bdrv_close(bs
);
3388 ret
= bdrv_snapshot_goto(bs
->file
, snapshot_id
);
3389 open_ret
= drv
->bdrv_open(bs
, NULL
, bs
->open_flags
);
3391 bdrv_delete(bs
->file
);
3401 int bdrv_snapshot_delete(BlockDriverState
*bs
, const char *snapshot_id
)
3403 BlockDriver
*drv
= bs
->drv
;
3406 if (drv
->bdrv_snapshot_delete
)
3407 return drv
->bdrv_snapshot_delete(bs
, snapshot_id
);
3409 return bdrv_snapshot_delete(bs
->file
, snapshot_id
);
3413 int bdrv_snapshot_list(BlockDriverState
*bs
,
3414 QEMUSnapshotInfo
**psn_info
)
3416 BlockDriver
*drv
= bs
->drv
;
3419 if (drv
->bdrv_snapshot_list
)
3420 return drv
->bdrv_snapshot_list(bs
, psn_info
);
3422 return bdrv_snapshot_list(bs
->file
, psn_info
);
3426 int bdrv_snapshot_load_tmp(BlockDriverState
*bs
,
3427 const char *snapshot_name
)
3429 BlockDriver
*drv
= bs
->drv
;
3433 if (!bs
->read_only
) {
3436 if (drv
->bdrv_snapshot_load_tmp
) {
3437 return drv
->bdrv_snapshot_load_tmp(bs
, snapshot_name
);
3442 /* backing_file can either be relative, or absolute, or a protocol. If it is
3443 * relative, it must be relative to the chain. So, passing in bs->filename
3444 * from a BDS as backing_file should not be done, as that may be relative to
3445 * the CWD rather than the chain. */
3446 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
3447 const char *backing_file
)
3449 char *filename_full
= NULL
;
3450 char *backing_file_full
= NULL
;
3451 char *filename_tmp
= NULL
;
3452 int is_protocol
= 0;
3453 BlockDriverState
*curr_bs
= NULL
;
3454 BlockDriverState
*retval
= NULL
;
3456 if (!bs
|| !bs
->drv
|| !backing_file
) {
3460 filename_full
= g_malloc(PATH_MAX
);
3461 backing_file_full
= g_malloc(PATH_MAX
);
3462 filename_tmp
= g_malloc(PATH_MAX
);
3464 is_protocol
= path_has_protocol(backing_file
);
3466 for (curr_bs
= bs
; curr_bs
->backing_hd
; curr_bs
= curr_bs
->backing_hd
) {
3468 /* If either of the filename paths is actually a protocol, then
3469 * compare unmodified paths; otherwise make paths relative */
3470 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
3471 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
3472 retval
= curr_bs
->backing_hd
;
3476 /* If not an absolute filename path, make it relative to the current
3477 * image's filename path */
3478 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
3481 /* We are going to compare absolute pathnames */
3482 if (!realpath(filename_tmp
, filename_full
)) {
3486 /* We need to make sure the backing filename we are comparing against
3487 * is relative to the current image filename (or absolute) */
3488 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
3489 curr_bs
->backing_file
);
3491 if (!realpath(filename_tmp
, backing_file_full
)) {
3495 if (strcmp(backing_file_full
, filename_full
) == 0) {
3496 retval
= curr_bs
->backing_hd
;
3502 g_free(filename_full
);
3503 g_free(backing_file_full
);
3504 g_free(filename_tmp
);
3508 int bdrv_get_backing_file_depth(BlockDriverState
*bs
)
3514 if (!bs
->backing_hd
) {
3518 return 1 + bdrv_get_backing_file_depth(bs
->backing_hd
);
3521 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
3523 BlockDriverState
*curr_bs
= NULL
;
3531 while (curr_bs
->backing_hd
) {
3532 curr_bs
= curr_bs
->backing_hd
;
3537 #define NB_SUFFIXES 4
3539 char *get_human_readable_size(char *buf
, int buf_size
, int64_t size
)
3541 static const char suffixes
[NB_SUFFIXES
] = "KMGT";
3546 snprintf(buf
, buf_size
, "%" PRId64
, size
);
3549 for(i
= 0; i
< NB_SUFFIXES
; i
++) {
3550 if (size
< (10 * base
)) {
3551 snprintf(buf
, buf_size
, "%0.1f%c",
3552 (double)size
/ base
,
3555 } else if (size
< (1000 * base
) || i
== (NB_SUFFIXES
- 1)) {
3556 snprintf(buf
, buf_size
, "%" PRId64
"%c",
3557 ((size
+ (base
>> 1)) / base
),
3567 char *bdrv_snapshot_dump(char *buf
, int buf_size
, QEMUSnapshotInfo
*sn
)
3569 char buf1
[128], date_buf
[128], clock_buf
[128];
3575 snprintf(buf
, buf_size
,
3576 "%-10s%-20s%7s%20s%15s",
3577 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
3580 localtime_r(&ti
, &tm
);
3581 strftime(date_buf
, sizeof(date_buf
),
3582 "%Y-%m-%d %H:%M:%S", &tm
);
3583 secs
= sn
->vm_clock_nsec
/ 1000000000;
3584 snprintf(clock_buf
, sizeof(clock_buf
),
3585 "%02d:%02d:%02d.%03d",
3587 (int)((secs
/ 60) % 60),
3589 (int)((sn
->vm_clock_nsec
/ 1000000) % 1000));
3590 snprintf(buf
, buf_size
,
3591 "%-10s%-20s%7s%20s%15s",
3592 sn
->id_str
, sn
->name
,
3593 get_human_readable_size(buf1
, sizeof(buf1
), sn
->vm_state_size
),
3600 /**************************************************************/
3603 BlockDriverAIOCB
*bdrv_aio_readv(BlockDriverState
*bs
, int64_t sector_num
,
3604 QEMUIOVector
*qiov
, int nb_sectors
,
3605 BlockDriverCompletionFunc
*cb
, void *opaque
)
3607 trace_bdrv_aio_readv(bs
, sector_num
, nb_sectors
, opaque
);
3609 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
,
3613 BlockDriverAIOCB
*bdrv_aio_writev(BlockDriverState
*bs
, int64_t sector_num
,
3614 QEMUIOVector
*qiov
, int nb_sectors
,
3615 BlockDriverCompletionFunc
*cb
, void *opaque
)
3617 trace_bdrv_aio_writev(bs
, sector_num
, nb_sectors
, opaque
);
3619 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
,
3624 typedef struct MultiwriteCB
{
3629 BlockDriverCompletionFunc
*cb
;
3631 QEMUIOVector
*free_qiov
;
3635 static void multiwrite_user_cb(MultiwriteCB
*mcb
)
3639 for (i
= 0; i
< mcb
->num_callbacks
; i
++) {
3640 mcb
->callbacks
[i
].cb(mcb
->callbacks
[i
].opaque
, mcb
->error
);
3641 if (mcb
->callbacks
[i
].free_qiov
) {
3642 qemu_iovec_destroy(mcb
->callbacks
[i
].free_qiov
);
3644 g_free(mcb
->callbacks
[i
].free_qiov
);
3648 static void multiwrite_cb(void *opaque
, int ret
)
3650 MultiwriteCB
*mcb
= opaque
;
3652 trace_multiwrite_cb(mcb
, ret
);
3654 if (ret
< 0 && !mcb
->error
) {
3658 mcb
->num_requests
--;
3659 if (mcb
->num_requests
== 0) {
3660 multiwrite_user_cb(mcb
);
3665 static int multiwrite_req_compare(const void *a
, const void *b
)
3667 const BlockRequest
*req1
= a
, *req2
= b
;
3670 * Note that we can't simply subtract req2->sector from req1->sector
3671 * here as that could overflow the return value.
3673 if (req1
->sector
> req2
->sector
) {
3675 } else if (req1
->sector
< req2
->sector
) {
3683 * Takes a bunch of requests and tries to merge them. Returns the number of
3684 * requests that remain after merging.
3686 static int multiwrite_merge(BlockDriverState
*bs
, BlockRequest
*reqs
,
3687 int num_reqs
, MultiwriteCB
*mcb
)
3691 // Sort requests by start sector
3692 qsort(reqs
, num_reqs
, sizeof(*reqs
), &multiwrite_req_compare
);
3694 // Check if adjacent requests touch the same clusters. If so, combine them,
3695 // filling up gaps with zero sectors.
3697 for (i
= 1; i
< num_reqs
; i
++) {
3699 int64_t oldreq_last
= reqs
[outidx
].sector
+ reqs
[outidx
].nb_sectors
;
3701 // Handle exactly sequential writes and overlapping writes.
3702 if (reqs
[i
].sector
<= oldreq_last
) {
3706 if (reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1 > IOV_MAX
) {
3712 QEMUIOVector
*qiov
= g_malloc0(sizeof(*qiov
));
3713 qemu_iovec_init(qiov
,
3714 reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1);
3716 // Add the first request to the merged one. If the requests are
3717 // overlapping, drop the last sectors of the first request.
3718 size
= (reqs
[i
].sector
- reqs
[outidx
].sector
) << 9;
3719 qemu_iovec_concat(qiov
, reqs
[outidx
].qiov
, 0, size
);
3721 // We should need to add any zeros between the two requests
3722 assert (reqs
[i
].sector
<= oldreq_last
);
3724 // Add the second request
3725 qemu_iovec_concat(qiov
, reqs
[i
].qiov
, 0, reqs
[i
].qiov
->size
);
3727 reqs
[outidx
].nb_sectors
= qiov
->size
>> 9;
3728 reqs
[outidx
].qiov
= qiov
;
3730 mcb
->callbacks
[i
].free_qiov
= reqs
[outidx
].qiov
;
3733 reqs
[outidx
].sector
= reqs
[i
].sector
;
3734 reqs
[outidx
].nb_sectors
= reqs
[i
].nb_sectors
;
3735 reqs
[outidx
].qiov
= reqs
[i
].qiov
;
3743 * Submit multiple AIO write requests at once.
3745 * On success, the function returns 0 and all requests in the reqs array have
3746 * been submitted. In error case this function returns -1, and any of the
3747 * requests may or may not be submitted yet. In particular, this means that the
3748 * callback will be called for some of the requests, for others it won't. The
3749 * caller must check the error field of the BlockRequest to wait for the right
3750 * callbacks (if error != 0, no callback will be called).
3752 * The implementation may modify the contents of the reqs array, e.g. to merge
3753 * requests. However, the fields opaque and error are left unmodified as they
3754 * are used to signal failure for a single request to the caller.
3756 int bdrv_aio_multiwrite(BlockDriverState
*bs
, BlockRequest
*reqs
, int num_reqs
)
3761 /* don't submit writes if we don't have a medium */
3762 if (bs
->drv
== NULL
) {
3763 for (i
= 0; i
< num_reqs
; i
++) {
3764 reqs
[i
].error
= -ENOMEDIUM
;
3769 if (num_reqs
== 0) {
3773 // Create MultiwriteCB structure
3774 mcb
= g_malloc0(sizeof(*mcb
) + num_reqs
* sizeof(*mcb
->callbacks
));
3775 mcb
->num_requests
= 0;
3776 mcb
->num_callbacks
= num_reqs
;
3778 for (i
= 0; i
< num_reqs
; i
++) {
3779 mcb
->callbacks
[i
].cb
= reqs
[i
].cb
;
3780 mcb
->callbacks
[i
].opaque
= reqs
[i
].opaque
;
3783 // Check for mergable requests
3784 num_reqs
= multiwrite_merge(bs
, reqs
, num_reqs
, mcb
);
3786 trace_bdrv_aio_multiwrite(mcb
, mcb
->num_callbacks
, num_reqs
);
3788 /* Run the aio requests. */
3789 mcb
->num_requests
= num_reqs
;
3790 for (i
= 0; i
< num_reqs
; i
++) {
3791 bdrv_aio_writev(bs
, reqs
[i
].sector
, reqs
[i
].qiov
,
3792 reqs
[i
].nb_sectors
, multiwrite_cb
, mcb
);
3798 void bdrv_aio_cancel(BlockDriverAIOCB
*acb
)
3800 acb
->aiocb_info
->cancel(acb
);
3803 /* block I/O throttling */
3804 static bool bdrv_exceed_bps_limits(BlockDriverState
*bs
, int nb_sectors
,
3805 bool is_write
, double elapsed_time
, uint64_t *wait
)
3807 uint64_t bps_limit
= 0;
3809 double bytes_limit
, bytes_base
, bytes_res
;
3810 double slice_time
, wait_time
;
3812 if (bs
->io_limits
.bps
[BLOCK_IO_LIMIT_TOTAL
]) {
3813 bps_limit
= bs
->io_limits
.bps
[BLOCK_IO_LIMIT_TOTAL
];
3814 } else if (bs
->io_limits
.bps
[is_write
]) {
3815 bps_limit
= bs
->io_limits
.bps
[is_write
];
3824 slice_time
= bs
->slice_end
- bs
->slice_start
;
3825 slice_time
/= (NANOSECONDS_PER_SECOND
);
3826 bytes_limit
= bps_limit
* slice_time
;
3827 bytes_base
= bs
->slice_submitted
.bytes
[is_write
];
3828 if (bs
->io_limits
.bps
[BLOCK_IO_LIMIT_TOTAL
]) {
3829 bytes_base
+= bs
->slice_submitted
.bytes
[!is_write
];
3832 /* bytes_base: the bytes of data which have been read/written; and
3833 * it is obtained from the history statistic info.
3834 * bytes_res: the remaining bytes of data which need to be read/written.
3835 * (bytes_base + bytes_res) / bps_limit: used to calcuate
3836 * the total time for completing reading/writting all data.
3838 bytes_res
= (unsigned) nb_sectors
* BDRV_SECTOR_SIZE
;
3840 if (bytes_base
+ bytes_res
<= bytes_limit
) {
3848 /* Calc approx time to dispatch */
3849 wait_time
= (bytes_base
+ bytes_res
) / bps_limit
- elapsed_time
;
3851 /* When the I/O rate at runtime exceeds the limits,
3852 * bs->slice_end need to be extended in order that the current statistic
3853 * info can be kept until the timer fire, so it is increased and tuned
3854 * based on the result of experiment.
3856 extension
= wait_time
* NANOSECONDS_PER_SECOND
;
3857 extension
= DIV_ROUND_UP(extension
, BLOCK_IO_SLICE_TIME
) *
3858 BLOCK_IO_SLICE_TIME
;
3859 bs
->slice_end
+= extension
;
3861 *wait
= wait_time
* NANOSECONDS_PER_SECOND
;
3867 static bool bdrv_exceed_iops_limits(BlockDriverState
*bs
, bool is_write
,
3868 double elapsed_time
, uint64_t *wait
)
3870 uint64_t iops_limit
= 0;
3871 double ios_limit
, ios_base
;
3872 double slice_time
, wait_time
;
3874 if (bs
->io_limits
.iops
[BLOCK_IO_LIMIT_TOTAL
]) {
3875 iops_limit
= bs
->io_limits
.iops
[BLOCK_IO_LIMIT_TOTAL
];
3876 } else if (bs
->io_limits
.iops
[is_write
]) {
3877 iops_limit
= bs
->io_limits
.iops
[is_write
];
3886 slice_time
= bs
->slice_end
- bs
->slice_start
;
3887 slice_time
/= (NANOSECONDS_PER_SECOND
);
3888 ios_limit
= iops_limit
* slice_time
;
3889 ios_base
= bs
->slice_submitted
.ios
[is_write
];
3890 if (bs
->io_limits
.iops
[BLOCK_IO_LIMIT_TOTAL
]) {
3891 ios_base
+= bs
->slice_submitted
.ios
[!is_write
];
3894 if (ios_base
+ 1 <= ios_limit
) {
3902 /* Calc approx time to dispatch, in seconds */
3903 wait_time
= (ios_base
+ 1) / iops_limit
;
3904 if (wait_time
> elapsed_time
) {
3905 wait_time
= wait_time
- elapsed_time
;
3910 /* Exceeded current slice, extend it by another slice time */
3911 bs
->slice_end
+= BLOCK_IO_SLICE_TIME
;
3913 *wait
= wait_time
* NANOSECONDS_PER_SECOND
;
3919 static bool bdrv_exceed_io_limits(BlockDriverState
*bs
, int nb_sectors
,
3920 bool is_write
, int64_t *wait
)
3922 int64_t now
, max_wait
;
3923 uint64_t bps_wait
= 0, iops_wait
= 0;
3924 double elapsed_time
;
3925 int bps_ret
, iops_ret
;
3927 now
= qemu_get_clock_ns(vm_clock
);
3928 if (now
> bs
->slice_end
) {
3929 bs
->slice_start
= now
;
3930 bs
->slice_end
= now
+ BLOCK_IO_SLICE_TIME
;
3931 memset(&bs
->slice_submitted
, 0, sizeof(bs
->slice_submitted
));
3934 elapsed_time
= now
- bs
->slice_start
;
3935 elapsed_time
/= (NANOSECONDS_PER_SECOND
);
3937 bps_ret
= bdrv_exceed_bps_limits(bs
, nb_sectors
,
3938 is_write
, elapsed_time
, &bps_wait
);
3939 iops_ret
= bdrv_exceed_iops_limits(bs
, is_write
,
3940 elapsed_time
, &iops_wait
);
3941 if (bps_ret
|| iops_ret
) {
3942 max_wait
= bps_wait
> iops_wait
? bps_wait
: iops_wait
;
3947 now
= qemu_get_clock_ns(vm_clock
);
3948 if (bs
->slice_end
< now
+ max_wait
) {
3949 bs
->slice_end
= now
+ max_wait
;
3959 bs
->slice_submitted
.bytes
[is_write
] += (int64_t)nb_sectors
*
3961 bs
->slice_submitted
.ios
[is_write
]++;
3966 /**************************************************************/
3967 /* async block device emulation */
3969 typedef struct BlockDriverAIOCBSync
{
3970 BlockDriverAIOCB common
;
3973 /* vector translation state */
3977 } BlockDriverAIOCBSync
;
3979 static void bdrv_aio_cancel_em(BlockDriverAIOCB
*blockacb
)
3981 BlockDriverAIOCBSync
*acb
=
3982 container_of(blockacb
, BlockDriverAIOCBSync
, common
);
3983 qemu_bh_delete(acb
->bh
);
3985 qemu_aio_release(acb
);
3988 static const AIOCBInfo bdrv_em_aiocb_info
= {
3989 .aiocb_size
= sizeof(BlockDriverAIOCBSync
),
3990 .cancel
= bdrv_aio_cancel_em
,
3993 static void bdrv_aio_bh_cb(void *opaque
)
3995 BlockDriverAIOCBSync
*acb
= opaque
;
3998 qemu_iovec_from_buf(acb
->qiov
, 0, acb
->bounce
, acb
->qiov
->size
);
3999 qemu_vfree(acb
->bounce
);
4000 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
4001 qemu_bh_delete(acb
->bh
);
4003 qemu_aio_release(acb
);
4006 static BlockDriverAIOCB
*bdrv_aio_rw_vector(BlockDriverState
*bs
,
4010 BlockDriverCompletionFunc
*cb
,
4015 BlockDriverAIOCBSync
*acb
;
4017 acb
= qemu_aio_get(&bdrv_em_aiocb_info
, bs
, cb
, opaque
);
4018 acb
->is_write
= is_write
;
4020 acb
->bounce
= qemu_blockalign(bs
, qiov
->size
);
4021 acb
->bh
= qemu_bh_new(bdrv_aio_bh_cb
, acb
);
4024 qemu_iovec_to_buf(acb
->qiov
, 0, acb
->bounce
, qiov
->size
);
4025 acb
->ret
= bs
->drv
->bdrv_write(bs
, sector_num
, acb
->bounce
, nb_sectors
);
4027 acb
->ret
= bs
->drv
->bdrv_read(bs
, sector_num
, acb
->bounce
, nb_sectors
);
4030 qemu_bh_schedule(acb
->bh
);
4032 return &acb
->common
;
4035 static BlockDriverAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
4036 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
4037 BlockDriverCompletionFunc
*cb
, void *opaque
)
4039 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 0);
4042 static BlockDriverAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
4043 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
4044 BlockDriverCompletionFunc
*cb
, void *opaque
)
4046 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 1);
4050 typedef struct BlockDriverAIOCBCoroutine
{
4051 BlockDriverAIOCB common
;
4056 } BlockDriverAIOCBCoroutine
;
4058 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB
*blockacb
)
4060 BlockDriverAIOCBCoroutine
*acb
=
4061 container_of(blockacb
, BlockDriverAIOCBCoroutine
, common
);
4070 static const AIOCBInfo bdrv_em_co_aiocb_info
= {
4071 .aiocb_size
= sizeof(BlockDriverAIOCBCoroutine
),
4072 .cancel
= bdrv_aio_co_cancel_em
,
4075 static void bdrv_co_em_bh(void *opaque
)
4077 BlockDriverAIOCBCoroutine
*acb
= opaque
;
4079 acb
->common
.cb(acb
->common
.opaque
, acb
->req
.error
);
4085 qemu_bh_delete(acb
->bh
);
4086 qemu_aio_release(acb
);
4089 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
4090 static void coroutine_fn
bdrv_co_do_rw(void *opaque
)
4092 BlockDriverAIOCBCoroutine
*acb
= opaque
;
4093 BlockDriverState
*bs
= acb
->common
.bs
;
4095 if (!acb
->is_write
) {
4096 acb
->req
.error
= bdrv_co_do_readv(bs
, acb
->req
.sector
,
4097 acb
->req
.nb_sectors
, acb
->req
.qiov
, 0);
4099 acb
->req
.error
= bdrv_co_do_writev(bs
, acb
->req
.sector
,
4100 acb
->req
.nb_sectors
, acb
->req
.qiov
, 0);
4103 acb
->bh
= qemu_bh_new(bdrv_co_em_bh
, acb
);
4104 qemu_bh_schedule(acb
->bh
);
4107 static BlockDriverAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
4111 BlockDriverCompletionFunc
*cb
,
4116 BlockDriverAIOCBCoroutine
*acb
;
4118 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
4119 acb
->req
.sector
= sector_num
;
4120 acb
->req
.nb_sectors
= nb_sectors
;
4121 acb
->req
.qiov
= qiov
;
4122 acb
->is_write
= is_write
;
4125 co
= qemu_coroutine_create(bdrv_co_do_rw
);
4126 qemu_coroutine_enter(co
, acb
);
4128 return &acb
->common
;
4131 static void coroutine_fn
bdrv_aio_flush_co_entry(void *opaque
)
4133 BlockDriverAIOCBCoroutine
*acb
= opaque
;
4134 BlockDriverState
*bs
= acb
->common
.bs
;
4136 acb
->req
.error
= bdrv_co_flush(bs
);
4137 acb
->bh
= qemu_bh_new(bdrv_co_em_bh
, acb
);
4138 qemu_bh_schedule(acb
->bh
);
4141 BlockDriverAIOCB
*bdrv_aio_flush(BlockDriverState
*bs
,
4142 BlockDriverCompletionFunc
*cb
, void *opaque
)
4144 trace_bdrv_aio_flush(bs
, opaque
);
4147 BlockDriverAIOCBCoroutine
*acb
;
4149 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
4152 co
= qemu_coroutine_create(bdrv_aio_flush_co_entry
);
4153 qemu_coroutine_enter(co
, acb
);
4155 return &acb
->common
;
4158 static void coroutine_fn
bdrv_aio_discard_co_entry(void *opaque
)
4160 BlockDriverAIOCBCoroutine
*acb
= opaque
;
4161 BlockDriverState
*bs
= acb
->common
.bs
;
4163 acb
->req
.error
= bdrv_co_discard(bs
, acb
->req
.sector
, acb
->req
.nb_sectors
);
4164 acb
->bh
= qemu_bh_new(bdrv_co_em_bh
, acb
);
4165 qemu_bh_schedule(acb
->bh
);
4168 BlockDriverAIOCB
*bdrv_aio_discard(BlockDriverState
*bs
,
4169 int64_t sector_num
, int nb_sectors
,
4170 BlockDriverCompletionFunc
*cb
, void *opaque
)
4173 BlockDriverAIOCBCoroutine
*acb
;
4175 trace_bdrv_aio_discard(bs
, sector_num
, nb_sectors
, opaque
);
4177 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
4178 acb
->req
.sector
= sector_num
;
4179 acb
->req
.nb_sectors
= nb_sectors
;
4181 co
= qemu_coroutine_create(bdrv_aio_discard_co_entry
);
4182 qemu_coroutine_enter(co
, acb
);
4184 return &acb
->common
;
4187 void bdrv_init(void)
4189 module_call_init(MODULE_INIT_BLOCK
);
4192 void bdrv_init_with_whitelist(void)
4194 use_bdrv_whitelist
= 1;
4198 void *qemu_aio_get(const AIOCBInfo
*aiocb_info
, BlockDriverState
*bs
,
4199 BlockDriverCompletionFunc
*cb
, void *opaque
)
4201 BlockDriverAIOCB
*acb
;
4203 acb
= g_slice_alloc(aiocb_info
->aiocb_size
);
4204 acb
->aiocb_info
= aiocb_info
;
4207 acb
->opaque
= opaque
;
4211 void qemu_aio_release(void *p
)
4213 BlockDriverAIOCB
*acb
= p
;
4214 g_slice_free1(acb
->aiocb_info
->aiocb_size
, acb
);
4217 /**************************************************************/
4218 /* Coroutine block device emulation */
4220 typedef struct CoroutineIOCompletion
{
4221 Coroutine
*coroutine
;
4223 } CoroutineIOCompletion
;
4225 static void bdrv_co_io_em_complete(void *opaque
, int ret
)
4227 CoroutineIOCompletion
*co
= opaque
;
4230 qemu_coroutine_enter(co
->coroutine
, NULL
);
4233 static int coroutine_fn
bdrv_co_io_em(BlockDriverState
*bs
, int64_t sector_num
,
4234 int nb_sectors
, QEMUIOVector
*iov
,
4237 CoroutineIOCompletion co
= {
4238 .coroutine
= qemu_coroutine_self(),
4240 BlockDriverAIOCB
*acb
;
4243 acb
= bs
->drv
->bdrv_aio_writev(bs
, sector_num
, iov
, nb_sectors
,
4244 bdrv_co_io_em_complete
, &co
);
4246 acb
= bs
->drv
->bdrv_aio_readv(bs
, sector_num
, iov
, nb_sectors
,
4247 bdrv_co_io_em_complete
, &co
);
4250 trace_bdrv_co_io_em(bs
, sector_num
, nb_sectors
, is_write
, acb
);
4254 qemu_coroutine_yield();
4259 static int coroutine_fn
bdrv_co_readv_em(BlockDriverState
*bs
,
4260 int64_t sector_num
, int nb_sectors
,
4263 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, false);
4266 static int coroutine_fn
bdrv_co_writev_em(BlockDriverState
*bs
,
4267 int64_t sector_num
, int nb_sectors
,
4270 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, true);
4273 static void coroutine_fn
bdrv_flush_co_entry(void *opaque
)
4275 RwCo
*rwco
= opaque
;
4277 rwco
->ret
= bdrv_co_flush(rwco
->bs
);
4280 int coroutine_fn
bdrv_co_flush(BlockDriverState
*bs
)
4284 if (!bs
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
)) {
4288 /* Write back cached data to the OS even with cache=unsafe */
4289 if (bs
->drv
->bdrv_co_flush_to_os
) {
4290 ret
= bs
->drv
->bdrv_co_flush_to_os(bs
);
4296 /* But don't actually force it to the disk with cache=unsafe */
4297 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
4301 if (bs
->drv
->bdrv_co_flush_to_disk
) {
4302 ret
= bs
->drv
->bdrv_co_flush_to_disk(bs
);
4303 } else if (bs
->drv
->bdrv_aio_flush
) {
4304 BlockDriverAIOCB
*acb
;
4305 CoroutineIOCompletion co
= {
4306 .coroutine
= qemu_coroutine_self(),
4309 acb
= bs
->drv
->bdrv_aio_flush(bs
, bdrv_co_io_em_complete
, &co
);
4313 qemu_coroutine_yield();
4318 * Some block drivers always operate in either writethrough or unsafe
4319 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
4320 * know how the server works (because the behaviour is hardcoded or
4321 * depends on server-side configuration), so we can't ensure that
4322 * everything is safe on disk. Returning an error doesn't work because
4323 * that would break guests even if the server operates in writethrough
4326 * Let's hope the user knows what he's doing.
4334 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
4335 * in the case of cache=unsafe, so there are no useless flushes.
4338 return bdrv_co_flush(bs
->file
);
4341 void bdrv_invalidate_cache(BlockDriverState
*bs
)
4343 if (bs
->drv
&& bs
->drv
->bdrv_invalidate_cache
) {
4344 bs
->drv
->bdrv_invalidate_cache(bs
);
4348 void bdrv_invalidate_cache_all(void)
4350 BlockDriverState
*bs
;
4352 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
4353 bdrv_invalidate_cache(bs
);
4357 void bdrv_clear_incoming_migration_all(void)
4359 BlockDriverState
*bs
;
4361 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
4362 bs
->open_flags
= bs
->open_flags
& ~(BDRV_O_INCOMING
);
4366 int bdrv_flush(BlockDriverState
*bs
)
4374 if (qemu_in_coroutine()) {
4375 /* Fast-path if already in coroutine context */
4376 bdrv_flush_co_entry(&rwco
);
4378 co
= qemu_coroutine_create(bdrv_flush_co_entry
);
4379 qemu_coroutine_enter(co
, &rwco
);
4380 while (rwco
.ret
== NOT_DONE
) {
4388 static void coroutine_fn
bdrv_discard_co_entry(void *opaque
)
4390 RwCo
*rwco
= opaque
;
4392 rwco
->ret
= bdrv_co_discard(rwco
->bs
, rwco
->sector_num
, rwco
->nb_sectors
);
4395 int coroutine_fn
bdrv_co_discard(BlockDriverState
*bs
, int64_t sector_num
,
4400 } else if (bdrv_check_request(bs
, sector_num
, nb_sectors
)) {
4402 } else if (bs
->read_only
) {
4406 if (bs
->dirty_bitmap
) {
4407 bdrv_reset_dirty(bs
, sector_num
, nb_sectors
);
4410 /* Do nothing if disabled. */
4411 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
4415 if (bs
->drv
->bdrv_co_discard
) {
4416 return bs
->drv
->bdrv_co_discard(bs
, sector_num
, nb_sectors
);
4417 } else if (bs
->drv
->bdrv_aio_discard
) {
4418 BlockDriverAIOCB
*acb
;
4419 CoroutineIOCompletion co
= {
4420 .coroutine
= qemu_coroutine_self(),
4423 acb
= bs
->drv
->bdrv_aio_discard(bs
, sector_num
, nb_sectors
,
4424 bdrv_co_io_em_complete
, &co
);
4428 qemu_coroutine_yield();
4436 int bdrv_discard(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
)
4441 .sector_num
= sector_num
,
4442 .nb_sectors
= nb_sectors
,
4446 if (qemu_in_coroutine()) {
4447 /* Fast-path if already in coroutine context */
4448 bdrv_discard_co_entry(&rwco
);
4450 co
= qemu_coroutine_create(bdrv_discard_co_entry
);
4451 qemu_coroutine_enter(co
, &rwco
);
4452 while (rwco
.ret
== NOT_DONE
) {
4460 /**************************************************************/
4461 /* removable device support */
4464 * Return TRUE if the media is present
4466 int bdrv_is_inserted(BlockDriverState
*bs
)
4468 BlockDriver
*drv
= bs
->drv
;
4472 if (!drv
->bdrv_is_inserted
)
4474 return drv
->bdrv_is_inserted(bs
);
4478 * Return whether the media changed since the last call to this
4479 * function, or -ENOTSUP if we don't know. Most drivers don't know.
4481 int bdrv_media_changed(BlockDriverState
*bs
)
4483 BlockDriver
*drv
= bs
->drv
;
4485 if (drv
&& drv
->bdrv_media_changed
) {
4486 return drv
->bdrv_media_changed(bs
);
4492 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4494 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
4496 BlockDriver
*drv
= bs
->drv
;
4498 if (drv
&& drv
->bdrv_eject
) {
4499 drv
->bdrv_eject(bs
, eject_flag
);
4502 if (bs
->device_name
[0] != '\0') {
4503 bdrv_emit_qmp_eject_event(bs
, eject_flag
);
4508 * Lock or unlock the media (if it is locked, the user won't be able
4509 * to eject it manually).
4511 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
4513 BlockDriver
*drv
= bs
->drv
;
4515 trace_bdrv_lock_medium(bs
, locked
);
4517 if (drv
&& drv
->bdrv_lock_medium
) {
4518 drv
->bdrv_lock_medium(bs
, locked
);
4522 /* needed for generic scsi interface */
4524 int bdrv_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
4526 BlockDriver
*drv
= bs
->drv
;
4528 if (drv
&& drv
->bdrv_ioctl
)
4529 return drv
->bdrv_ioctl(bs
, req
, buf
);
4533 BlockDriverAIOCB
*bdrv_aio_ioctl(BlockDriverState
*bs
,
4534 unsigned long int req
, void *buf
,
4535 BlockDriverCompletionFunc
*cb
, void *opaque
)
4537 BlockDriver
*drv
= bs
->drv
;
4539 if (drv
&& drv
->bdrv_aio_ioctl
)
4540 return drv
->bdrv_aio_ioctl(bs
, req
, buf
, cb
, opaque
);
4544 void bdrv_set_buffer_alignment(BlockDriverState
*bs
, int align
)
4546 bs
->buffer_alignment
= align
;
4549 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
4551 return qemu_memalign((bs
&& bs
->buffer_alignment
) ? bs
->buffer_alignment
: 512, size
);
4555 * Check if all memory in this vector is sector aligned.
4557 bool bdrv_qiov_is_aligned(BlockDriverState
*bs
, QEMUIOVector
*qiov
)
4561 for (i
= 0; i
< qiov
->niov
; i
++) {
4562 if ((uintptr_t) qiov
->iov
[i
].iov_base
% bs
->buffer_alignment
) {
4570 void bdrv_set_dirty_tracking(BlockDriverState
*bs
, int granularity
)
4572 int64_t bitmap_size
;
4574 assert((granularity
& (granularity
- 1)) == 0);
4577 granularity
>>= BDRV_SECTOR_BITS
;
4578 assert(!bs
->dirty_bitmap
);
4579 bitmap_size
= (bdrv_getlength(bs
) >> BDRV_SECTOR_BITS
);
4580 bs
->dirty_bitmap
= hbitmap_alloc(bitmap_size
, ffs(granularity
) - 1);
4582 if (bs
->dirty_bitmap
) {
4583 hbitmap_free(bs
->dirty_bitmap
);
4584 bs
->dirty_bitmap
= NULL
;
4589 int bdrv_get_dirty(BlockDriverState
*bs
, int64_t sector
)
4591 if (bs
->dirty_bitmap
) {
4592 return hbitmap_get(bs
->dirty_bitmap
, sector
);
4598 void bdrv_dirty_iter_init(BlockDriverState
*bs
, HBitmapIter
*hbi
)
4600 hbitmap_iter_init(hbi
, bs
->dirty_bitmap
, 0);
4603 void bdrv_set_dirty(BlockDriverState
*bs
, int64_t cur_sector
,
4606 hbitmap_set(bs
->dirty_bitmap
, cur_sector
, nr_sectors
);
4609 void bdrv_reset_dirty(BlockDriverState
*bs
, int64_t cur_sector
,
4612 hbitmap_reset(bs
->dirty_bitmap
, cur_sector
, nr_sectors
);
4615 int64_t bdrv_get_dirty_count(BlockDriverState
*bs
)
4617 if (bs
->dirty_bitmap
) {
4618 return hbitmap_count(bs
->dirty_bitmap
);
4624 void bdrv_set_in_use(BlockDriverState
*bs
, int in_use
)
4626 assert(bs
->in_use
!= in_use
);
4627 bs
->in_use
= in_use
;
4630 int bdrv_in_use(BlockDriverState
*bs
)
4635 void bdrv_iostatus_enable(BlockDriverState
*bs
)
4637 bs
->iostatus_enabled
= true;
4638 bs
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
4641 /* The I/O status is only enabled if the drive explicitly
4642 * enables it _and_ the VM is configured to stop on errors */
4643 bool bdrv_iostatus_is_enabled(const BlockDriverState
*bs
)
4645 return (bs
->iostatus_enabled
&&
4646 (bs
->on_write_error
== BLOCKDEV_ON_ERROR_ENOSPC
||
4647 bs
->on_write_error
== BLOCKDEV_ON_ERROR_STOP
||
4648 bs
->on_read_error
== BLOCKDEV_ON_ERROR_STOP
));
4651 void bdrv_iostatus_disable(BlockDriverState
*bs
)
4653 bs
->iostatus_enabled
= false;
4656 void bdrv_iostatus_reset(BlockDriverState
*bs
)
4658 if (bdrv_iostatus_is_enabled(bs
)) {
4659 bs
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
4661 block_job_iostatus_reset(bs
->job
);
4666 void bdrv_iostatus_set_err(BlockDriverState
*bs
, int error
)
4668 assert(bdrv_iostatus_is_enabled(bs
));
4669 if (bs
->iostatus
== BLOCK_DEVICE_IO_STATUS_OK
) {
4670 bs
->iostatus
= error
== ENOSPC
? BLOCK_DEVICE_IO_STATUS_NOSPACE
:
4671 BLOCK_DEVICE_IO_STATUS_FAILED
;
4676 bdrv_acct_start(BlockDriverState
*bs
, BlockAcctCookie
*cookie
, int64_t bytes
,
4677 enum BlockAcctType type
)
4679 assert(type
< BDRV_MAX_IOTYPE
);
4681 cookie
->bytes
= bytes
;
4682 cookie
->start_time_ns
= get_clock();
4683 cookie
->type
= type
;
4687 bdrv_acct_done(BlockDriverState
*bs
, BlockAcctCookie
*cookie
)
4689 assert(cookie
->type
< BDRV_MAX_IOTYPE
);
4691 bs
->nr_bytes
[cookie
->type
] += cookie
->bytes
;
4692 bs
->nr_ops
[cookie
->type
]++;
4693 bs
->total_time_ns
[cookie
->type
] += get_clock() - cookie
->start_time_ns
;
4696 void bdrv_img_create(const char *filename
, const char *fmt
,
4697 const char *base_filename
, const char *base_fmt
,
4698 char *options
, uint64_t img_size
, int flags
,
4699 Error
**errp
, bool quiet
)
4701 QEMUOptionParameter
*param
= NULL
, *create_options
= NULL
;
4702 QEMUOptionParameter
*backing_fmt
, *backing_file
, *size
;
4703 BlockDriverState
*bs
= NULL
;
4704 BlockDriver
*drv
, *proto_drv
;
4705 BlockDriver
*backing_drv
= NULL
;
4708 /* Find driver and parse its options */
4709 drv
= bdrv_find_format(fmt
);
4711 error_setg(errp
, "Unknown file format '%s'", fmt
);
4715 proto_drv
= bdrv_find_protocol(filename
);
4717 error_setg(errp
, "Unknown protocol '%s'", filename
);
4721 create_options
= append_option_parameters(create_options
,
4722 drv
->create_options
);
4723 create_options
= append_option_parameters(create_options
,
4724 proto_drv
->create_options
);
4726 /* Create parameter list with default values */
4727 param
= parse_option_parameters("", create_options
, param
);
4729 set_option_parameter_int(param
, BLOCK_OPT_SIZE
, img_size
);
4731 /* Parse -o options */
4733 param
= parse_option_parameters(options
, create_options
, param
);
4734 if (param
== NULL
) {
4735 error_setg(errp
, "Invalid options for file format '%s'.", fmt
);
4740 if (base_filename
) {
4741 if (set_option_parameter(param
, BLOCK_OPT_BACKING_FILE
,
4743 error_setg(errp
, "Backing file not supported for file format '%s'",
4750 if (set_option_parameter(param
, BLOCK_OPT_BACKING_FMT
, base_fmt
)) {
4751 error_setg(errp
, "Backing file format not supported for file "
4752 "format '%s'", fmt
);
4757 backing_file
= get_option_parameter(param
, BLOCK_OPT_BACKING_FILE
);
4758 if (backing_file
&& backing_file
->value
.s
) {
4759 if (!strcmp(filename
, backing_file
->value
.s
)) {
4760 error_setg(errp
, "Error: Trying to create an image with the "
4761 "same filename as the backing file");
4766 backing_fmt
= get_option_parameter(param
, BLOCK_OPT_BACKING_FMT
);
4767 if (backing_fmt
&& backing_fmt
->value
.s
) {
4768 backing_drv
= bdrv_find_format(backing_fmt
->value
.s
);
4770 error_setg(errp
, "Unknown backing file format '%s'",
4771 backing_fmt
->value
.s
);
4776 // The size for the image must always be specified, with one exception:
4777 // If we are using a backing file, we can obtain the size from there
4778 size
= get_option_parameter(param
, BLOCK_OPT_SIZE
);
4779 if (size
&& size
->value
.n
== -1) {
4780 if (backing_file
&& backing_file
->value
.s
) {
4785 /* backing files always opened read-only */
4787 flags
& ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
4791 ret
= bdrv_open(bs
, backing_file
->value
.s
, NULL
, back_flags
,
4794 error_setg_errno(errp
, -ret
, "Could not open '%s'",
4795 backing_file
->value
.s
);
4798 bdrv_get_geometry(bs
, &size
);
4801 snprintf(buf
, sizeof(buf
), "%" PRId64
, size
);
4802 set_option_parameter(param
, BLOCK_OPT_SIZE
, buf
);
4804 error_setg(errp
, "Image creation needs a size parameter");
4810 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
4811 print_option_parameters(param
);
4814 ret
= bdrv_create(drv
, filename
, param
);
4816 if (ret
== -ENOTSUP
) {
4817 error_setg(errp
,"Formatting or formatting option not supported for "
4818 "file format '%s'", fmt
);
4819 } else if (ret
== -EFBIG
) {
4820 error_setg(errp
, "The image size is too large for file format '%s'",
4823 error_setg(errp
, "%s: error while creating %s: %s", filename
, fmt
,
4829 free_option_parameters(create_options
);
4830 free_option_parameters(param
);
4837 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
4839 /* Currently BlockDriverState always uses the main loop AioContext */
4840 return qemu_get_aio_context();