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 QDict
*options
, int flags
, BlockDriver
*drv
)
673 const char *filename
;
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
);
702 filename
= file
->filename
;
704 filename
= qdict_get_try_str(options
, "filename");
707 if (filename
!= NULL
) {
708 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
710 bs
->filename
[0] = '\0';
714 bs
->opaque
= g_malloc0(drv
->instance_size
);
716 bs
->enable_write_cache
= !!(flags
& BDRV_O_CACHE_WB
);
717 open_flags
= bdrv_open_flags(bs
, flags
);
719 bs
->read_only
= !(open_flags
& BDRV_O_RDWR
);
721 /* Open the image, either directly or using a protocol */
722 if (drv
->bdrv_file_open
) {
723 assert(file
== NULL
);
724 assert(drv
->bdrv_parse_filename
|| filename
!= NULL
);
725 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
);
728 qerror_report(ERROR_CLASS_GENERIC_ERROR
, "Can't use '%s' as a "
729 "block driver for the protocol level",
734 assert(file
!= NULL
);
736 ret
= drv
->bdrv_open(bs
, options
, open_flags
);
743 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
749 if (bs
->is_temporary
) {
750 assert(filename
!= NULL
);
765 * Opens a file using a protocol (file, host_device, nbd, ...)
767 * options is a QDict of options to pass to the block drivers, or NULL for an
768 * empty set of options. The reference to the QDict belongs to the block layer
769 * after the call (even on failure), so if the caller intends to reuse the
770 * dictionary, it needs to use QINCREF() before calling bdrv_file_open.
772 int bdrv_file_open(BlockDriverState
**pbs
, const char *filename
,
773 QDict
*options
, int flags
)
775 BlockDriverState
*bs
;
780 /* NULL means an empty set of options */
781 if (options
== NULL
) {
782 options
= qdict_new();
786 bs
->options
= options
;
787 options
= qdict_clone_shallow(options
);
789 /* Fetch the file name from the options QDict if necessary */
791 filename
= qdict_get_try_str(options
, "filename");
792 } else if (filename
&& !qdict_haskey(options
, "filename")) {
793 qdict_put(options
, "filename", qstring_from_str(filename
));
795 qerror_report(ERROR_CLASS_GENERIC_ERROR
, "Can't specify 'file' and "
796 "'filename' options at the same time");
801 /* Find the right block driver */
802 drvname
= qdict_get_try_str(options
, "driver");
804 drv
= bdrv_find_whitelisted_format(drvname
);
805 qdict_del(options
, "driver");
806 } else if (filename
) {
807 drv
= bdrv_find_protocol(filename
);
809 qerror_report(ERROR_CLASS_GENERIC_ERROR
,
810 "Must specify either driver or file");
819 /* Parse the filename and open it */
820 if (drv
->bdrv_parse_filename
&& filename
) {
821 Error
*local_err
= NULL
;
822 drv
->bdrv_parse_filename(filename
, options
, &local_err
);
823 if (error_is_set(&local_err
)) {
824 qerror_report_err(local_err
);
825 error_free(local_err
);
829 qdict_del(options
, "filename");
830 } else if (!drv
->bdrv_parse_filename
&& !filename
) {
831 qerror_report(ERROR_CLASS_GENERIC_ERROR
,
832 "The '%s' block driver requires a file name",
838 ret
= bdrv_open_common(bs
, NULL
, options
, flags
, drv
);
843 /* Check if any unknown options were used */
844 if (qdict_size(options
) != 0) {
845 const QDictEntry
*entry
= qdict_first(options
);
846 qerror_report(ERROR_CLASS_GENERIC_ERROR
, "Block protocol '%s' doesn't "
847 "support the option '%s'",
848 drv
->format_name
, entry
->key
);
861 QDECREF(bs
->options
);
868 * Opens the backing file for a BlockDriverState if not yet open
870 * options is a QDict of options to pass to the block drivers, or NULL for an
871 * empty set of options. The reference to the QDict is transferred to this
872 * function (even on failure), so if the caller intends to reuse the dictionary,
873 * it needs to use QINCREF() before calling bdrv_file_open.
875 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*options
)
877 char backing_filename
[PATH_MAX
];
879 BlockDriver
*back_drv
= NULL
;
881 if (bs
->backing_hd
!= NULL
) {
886 /* NULL means an empty set of options */
887 if (options
== NULL
) {
888 options
= qdict_new();
891 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
892 if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
897 bs
->backing_hd
= bdrv_new("");
898 bdrv_get_full_backing_filename(bs
, backing_filename
,
899 sizeof(backing_filename
));
901 if (bs
->backing_format
[0] != '\0') {
902 back_drv
= bdrv_find_format(bs
->backing_format
);
905 /* backing files always opened read-only */
906 back_flags
= bs
->open_flags
& ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
);
908 ret
= bdrv_open(bs
->backing_hd
,
909 *backing_filename
? backing_filename
: NULL
, options
,
910 back_flags
, back_drv
);
912 bdrv_delete(bs
->backing_hd
);
913 bs
->backing_hd
= NULL
;
914 bs
->open_flags
|= BDRV_O_NO_BACKING
;
920 static void extract_subqdict(QDict
*src
, QDict
**dst
, const char *start
)
922 const QDictEntry
*entry
, *next
;
926 entry
= qdict_first(src
);
928 while (entry
!= NULL
) {
929 next
= qdict_next(src
, entry
);
930 if (strstart(entry
->key
, start
, &p
)) {
931 qobject_incref(entry
->value
);
932 qdict_put_obj(*dst
, p
, entry
->value
);
933 qdict_del(src
, entry
->key
);
940 * Opens a disk image (raw, qcow2, vmdk, ...)
942 * options is a QDict of options to pass to the block drivers, or NULL for an
943 * empty set of options. The reference to the QDict belongs to the block layer
944 * after the call (even on failure), so if the caller intends to reuse the
945 * dictionary, it needs to use QINCREF() before calling bdrv_open.
947 int bdrv_open(BlockDriverState
*bs
, const char *filename
, QDict
*options
,
948 int flags
, BlockDriver
*drv
)
951 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
952 char tmp_filename
[PATH_MAX
+ 1];
953 BlockDriverState
*file
= NULL
;
954 QDict
*file_options
= NULL
;
956 /* NULL means an empty set of options */
957 if (options
== NULL
) {
958 options
= qdict_new();
961 bs
->options
= options
;
962 options
= qdict_clone_shallow(options
);
964 /* For snapshot=on, create a temporary qcow2 overlay */
965 if (flags
& BDRV_O_SNAPSHOT
) {
966 BlockDriverState
*bs1
;
968 BlockDriver
*bdrv_qcow2
;
969 QEMUOptionParameter
*create_options
;
970 char backing_filename
[PATH_MAX
];
972 if (qdict_size(options
) != 0) {
973 error_report("Can't use snapshot=on with driver-specific options");
977 assert(filename
!= NULL
);
979 /* if snapshot, we create a temporary backing file and open it
980 instead of opening 'filename' directly */
982 /* if there is a backing file, use it */
984 ret
= bdrv_open(bs1
, filename
, NULL
, 0, drv
);
989 total_size
= bdrv_getlength(bs1
) & BDRV_SECTOR_MASK
;
993 ret
= get_tmp_filename(tmp_filename
, sizeof(tmp_filename
));
998 /* Real path is meaningless for protocols */
999 if (path_has_protocol(filename
)) {
1000 snprintf(backing_filename
, sizeof(backing_filename
),
1002 } else if (!realpath(filename
, backing_filename
)) {
1007 bdrv_qcow2
= bdrv_find_format("qcow2");
1008 create_options
= parse_option_parameters("", bdrv_qcow2
->create_options
,
1011 set_option_parameter_int(create_options
, BLOCK_OPT_SIZE
, total_size
);
1012 set_option_parameter(create_options
, BLOCK_OPT_BACKING_FILE
,
1015 set_option_parameter(create_options
, BLOCK_OPT_BACKING_FMT
,
1019 ret
= bdrv_create(bdrv_qcow2
, tmp_filename
, create_options
);
1020 free_option_parameters(create_options
);
1025 filename
= tmp_filename
;
1027 bs
->is_temporary
= 1;
1030 /* Open image file without format layer */
1031 if (flags
& BDRV_O_RDWR
) {
1032 flags
|= BDRV_O_ALLOW_RDWR
;
1035 extract_subqdict(options
, &file_options
, "file.");
1037 ret
= bdrv_file_open(&file
, filename
, file_options
,
1038 bdrv_open_flags(bs
, flags
));
1043 /* Find the right image format driver */
1045 ret
= find_image_format(file
, filename
, &drv
);
1049 goto unlink_and_fail
;
1052 /* Open the image */
1053 ret
= bdrv_open_common(bs
, file
, options
, flags
, drv
);
1055 goto unlink_and_fail
;
1058 if (bs
->file
!= file
) {
1063 /* If there is a backing file, use it */
1064 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
1065 QDict
*backing_options
;
1067 extract_subqdict(options
, &backing_options
, "backing.");
1068 ret
= bdrv_open_backing_file(bs
, backing_options
);
1070 goto close_and_fail
;
1074 /* Check if any unknown options were used */
1075 if (qdict_size(options
) != 0) {
1076 const QDictEntry
*entry
= qdict_first(options
);
1077 qerror_report(ERROR_CLASS_GENERIC_ERROR
, "Block format '%s' used by "
1078 "device '%s' doesn't support the option '%s'",
1079 drv
->format_name
, bs
->device_name
, entry
->key
);
1082 goto close_and_fail
;
1086 if (!bdrv_key_required(bs
)) {
1087 bdrv_dev_change_media_cb(bs
, true);
1090 /* throttling disk I/O limits */
1091 if (bs
->io_limits_enabled
) {
1092 bdrv_io_limits_enable(bs
);
1101 if (bs
->is_temporary
) {
1105 QDECREF(bs
->options
);
1116 typedef struct BlockReopenQueueEntry
{
1118 BDRVReopenState state
;
1119 QSIMPLEQ_ENTRY(BlockReopenQueueEntry
) entry
;
1120 } BlockReopenQueueEntry
;
1123 * Adds a BlockDriverState to a simple queue for an atomic, transactional
1124 * reopen of multiple devices.
1126 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
1127 * already performed, or alternatively may be NULL a new BlockReopenQueue will
1128 * be created and initialized. This newly created BlockReopenQueue should be
1129 * passed back in for subsequent calls that are intended to be of the same
1132 * bs is the BlockDriverState to add to the reopen queue.
1134 * flags contains the open flags for the associated bs
1136 * returns a pointer to bs_queue, which is either the newly allocated
1137 * bs_queue, or the existing bs_queue being used.
1140 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
1141 BlockDriverState
*bs
, int flags
)
1145 BlockReopenQueueEntry
*bs_entry
;
1146 if (bs_queue
== NULL
) {
1147 bs_queue
= g_new0(BlockReopenQueue
, 1);
1148 QSIMPLEQ_INIT(bs_queue
);
1152 bdrv_reopen_queue(bs_queue
, bs
->file
, flags
);
1155 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
1156 QSIMPLEQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
1158 bs_entry
->state
.bs
= bs
;
1159 bs_entry
->state
.flags
= flags
;
1165 * Reopen multiple BlockDriverStates atomically & transactionally.
1167 * The queue passed in (bs_queue) must have been built up previous
1168 * via bdrv_reopen_queue().
1170 * Reopens all BDS specified in the queue, with the appropriate
1171 * flags. All devices are prepared for reopen, and failure of any
1172 * device will cause all device changes to be abandonded, and intermediate
1175 * If all devices prepare successfully, then the changes are committed
1179 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
1182 BlockReopenQueueEntry
*bs_entry
, *next
;
1183 Error
*local_err
= NULL
;
1185 assert(bs_queue
!= NULL
);
1189 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
1190 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, &local_err
)) {
1191 error_propagate(errp
, local_err
);
1194 bs_entry
->prepared
= true;
1197 /* If we reach this point, we have success and just need to apply the
1200 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
1201 bdrv_reopen_commit(&bs_entry
->state
);
1207 QSIMPLEQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
1208 if (ret
&& bs_entry
->prepared
) {
1209 bdrv_reopen_abort(&bs_entry
->state
);
1218 /* Reopen a single BlockDriverState with the specified flags. */
1219 int bdrv_reopen(BlockDriverState
*bs
, int bdrv_flags
, Error
**errp
)
1222 Error
*local_err
= NULL
;
1223 BlockReopenQueue
*queue
= bdrv_reopen_queue(NULL
, bs
, bdrv_flags
);
1225 ret
= bdrv_reopen_multiple(queue
, &local_err
);
1226 if (local_err
!= NULL
) {
1227 error_propagate(errp
, local_err
);
1234 * Prepares a BlockDriverState for reopen. All changes are staged in the
1235 * 'opaque' field of the BDRVReopenState, which is used and allocated by
1236 * the block driver layer .bdrv_reopen_prepare()
1238 * bs is the BlockDriverState to reopen
1239 * flags are the new open flags
1240 * queue is the reopen queue
1242 * Returns 0 on success, non-zero on error. On error errp will be set
1245 * On failure, bdrv_reopen_abort() will be called to clean up any data.
1246 * It is the responsibility of the caller to then call the abort() or
1247 * commit() for any other BDS that have been left in a prepare() state
1250 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
1254 Error
*local_err
= NULL
;
1257 assert(reopen_state
!= NULL
);
1258 assert(reopen_state
->bs
->drv
!= NULL
);
1259 drv
= reopen_state
->bs
->drv
;
1261 /* if we are to stay read-only, do not allow permission change
1263 if (!(reopen_state
->bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
1264 reopen_state
->flags
& BDRV_O_RDWR
) {
1265 error_set(errp
, QERR_DEVICE_IS_READ_ONLY
,
1266 reopen_state
->bs
->device_name
);
1271 ret
= bdrv_flush(reopen_state
->bs
);
1273 error_set(errp
, ERROR_CLASS_GENERIC_ERROR
, "Error (%s) flushing drive",
1278 if (drv
->bdrv_reopen_prepare
) {
1279 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
1281 if (local_err
!= NULL
) {
1282 error_propagate(errp
, local_err
);
1284 error_set(errp
, QERR_OPEN_FILE_FAILED
,
1285 reopen_state
->bs
->filename
);
1290 /* It is currently mandatory to have a bdrv_reopen_prepare()
1291 * handler for each supported drv. */
1292 error_set(errp
, QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED
,
1293 drv
->format_name
, reopen_state
->bs
->device_name
,
1294 "reopening of file");
1306 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
1307 * makes them final by swapping the staging BlockDriverState contents into
1308 * the active BlockDriverState contents.
1310 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
1314 assert(reopen_state
!= NULL
);
1315 drv
= reopen_state
->bs
->drv
;
1316 assert(drv
!= NULL
);
1318 /* If there are any driver level actions to take */
1319 if (drv
->bdrv_reopen_commit
) {
1320 drv
->bdrv_reopen_commit(reopen_state
);
1323 /* set BDS specific flags now */
1324 reopen_state
->bs
->open_flags
= reopen_state
->flags
;
1325 reopen_state
->bs
->enable_write_cache
= !!(reopen_state
->flags
&
1327 reopen_state
->bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
1331 * Abort the reopen, and delete and free the staged changes in
1334 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
1338 assert(reopen_state
!= NULL
);
1339 drv
= reopen_state
->bs
->drv
;
1340 assert(drv
!= NULL
);
1342 if (drv
->bdrv_reopen_abort
) {
1343 drv
->bdrv_reopen_abort(reopen_state
);
1348 void bdrv_close(BlockDriverState
*bs
)
1352 block_job_cancel_sync(bs
->job
);
1355 notifier_list_notify(&bs
->close_notifiers
, bs
);
1358 if (bs
== bs_snapshots
) {
1359 bs_snapshots
= NULL
;
1361 if (bs
->backing_hd
) {
1362 bdrv_delete(bs
->backing_hd
);
1363 bs
->backing_hd
= NULL
;
1365 bs
->drv
->bdrv_close(bs
);
1368 if (bs
->is_temporary
) {
1369 unlink(bs
->filename
);
1374 bs
->copy_on_read
= 0;
1375 bs
->backing_file
[0] = '\0';
1376 bs
->backing_format
[0] = '\0';
1377 bs
->total_sectors
= 0;
1382 QDECREF(bs
->options
);
1385 if (bs
->file
!= NULL
) {
1386 bdrv_delete(bs
->file
);
1391 bdrv_dev_change_media_cb(bs
, false);
1393 /*throttling disk I/O limits*/
1394 if (bs
->io_limits_enabled
) {
1395 bdrv_io_limits_disable(bs
);
1399 void bdrv_close_all(void)
1401 BlockDriverState
*bs
;
1403 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1409 * Wait for pending requests to complete across all BlockDriverStates
1411 * This function does not flush data to disk, use bdrv_flush_all() for that
1412 * after calling this function.
1414 * Note that completion of an asynchronous I/O operation can trigger any
1415 * number of other I/O operations on other devices---for example a coroutine
1416 * can be arbitrarily complex and a constant flow of I/O can come until the
1417 * coroutine is complete. Because of this, it is not possible to have a
1418 * function to drain a single device's I/O queue.
1420 void bdrv_drain_all(void)
1422 BlockDriverState
*bs
;
1426 busy
= qemu_aio_wait();
1428 /* FIXME: We do not have timer support here, so this is effectively
1431 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1432 if (!qemu_co_queue_empty(&bs
->throttled_reqs
)) {
1433 qemu_co_queue_restart_all(&bs
->throttled_reqs
);
1439 /* If requests are still pending there is a bug somewhere */
1440 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1441 assert(QLIST_EMPTY(&bs
->tracked_requests
));
1442 assert(qemu_co_queue_empty(&bs
->throttled_reqs
));
1446 /* make a BlockDriverState anonymous by removing from bdrv_state list.
1447 Also, NULL terminate the device_name to prevent double remove */
1448 void bdrv_make_anon(BlockDriverState
*bs
)
1450 if (bs
->device_name
[0] != '\0') {
1451 QTAILQ_REMOVE(&bdrv_states
, bs
, list
);
1453 bs
->device_name
[0] = '\0';
1456 static void bdrv_rebind(BlockDriverState
*bs
)
1458 if (bs
->drv
&& bs
->drv
->bdrv_rebind
) {
1459 bs
->drv
->bdrv_rebind(bs
);
1463 static void bdrv_move_feature_fields(BlockDriverState
*bs_dest
,
1464 BlockDriverState
*bs_src
)
1466 /* move some fields that need to stay attached to the device */
1467 bs_dest
->open_flags
= bs_src
->open_flags
;
1470 bs_dest
->dev_ops
= bs_src
->dev_ops
;
1471 bs_dest
->dev_opaque
= bs_src
->dev_opaque
;
1472 bs_dest
->dev
= bs_src
->dev
;
1473 bs_dest
->buffer_alignment
= bs_src
->buffer_alignment
;
1474 bs_dest
->copy_on_read
= bs_src
->copy_on_read
;
1476 bs_dest
->enable_write_cache
= bs_src
->enable_write_cache
;
1478 /* i/o timing parameters */
1479 bs_dest
->slice_start
= bs_src
->slice_start
;
1480 bs_dest
->slice_end
= bs_src
->slice_end
;
1481 bs_dest
->slice_submitted
= bs_src
->slice_submitted
;
1482 bs_dest
->io_limits
= bs_src
->io_limits
;
1483 bs_dest
->throttled_reqs
= bs_src
->throttled_reqs
;
1484 bs_dest
->block_timer
= bs_src
->block_timer
;
1485 bs_dest
->io_limits_enabled
= bs_src
->io_limits_enabled
;
1488 bs_dest
->on_read_error
= bs_src
->on_read_error
;
1489 bs_dest
->on_write_error
= bs_src
->on_write_error
;
1492 bs_dest
->iostatus_enabled
= bs_src
->iostatus_enabled
;
1493 bs_dest
->iostatus
= bs_src
->iostatus
;
1496 bs_dest
->dirty_bitmap
= bs_src
->dirty_bitmap
;
1499 bs_dest
->in_use
= bs_src
->in_use
;
1500 bs_dest
->job
= bs_src
->job
;
1502 /* keep the same entry in bdrv_states */
1503 pstrcpy(bs_dest
->device_name
, sizeof(bs_dest
->device_name
),
1504 bs_src
->device_name
);
1505 bs_dest
->list
= bs_src
->list
;
1509 * Swap bs contents for two image chains while they are live,
1510 * while keeping required fields on the BlockDriverState that is
1511 * actually attached to a device.
1513 * This will modify the BlockDriverState fields, and swap contents
1514 * between bs_new and bs_old. Both bs_new and bs_old are modified.
1516 * bs_new is required to be anonymous.
1518 * This function does not create any image files.
1520 void bdrv_swap(BlockDriverState
*bs_new
, BlockDriverState
*bs_old
)
1522 BlockDriverState tmp
;
1524 /* bs_new must be anonymous and shouldn't have anything fancy enabled */
1525 assert(bs_new
->device_name
[0] == '\0');
1526 assert(bs_new
->dirty_bitmap
== NULL
);
1527 assert(bs_new
->job
== NULL
);
1528 assert(bs_new
->dev
== NULL
);
1529 assert(bs_new
->in_use
== 0);
1530 assert(bs_new
->io_limits_enabled
== false);
1531 assert(bs_new
->block_timer
== NULL
);
1537 /* there are some fields that should not be swapped, move them back */
1538 bdrv_move_feature_fields(&tmp
, bs_old
);
1539 bdrv_move_feature_fields(bs_old
, bs_new
);
1540 bdrv_move_feature_fields(bs_new
, &tmp
);
1542 /* bs_new shouldn't be in bdrv_states even after the swap! */
1543 assert(bs_new
->device_name
[0] == '\0');
1545 /* Check a few fields that should remain attached to the device */
1546 assert(bs_new
->dev
== NULL
);
1547 assert(bs_new
->job
== NULL
);
1548 assert(bs_new
->in_use
== 0);
1549 assert(bs_new
->io_limits_enabled
== false);
1550 assert(bs_new
->block_timer
== NULL
);
1552 bdrv_rebind(bs_new
);
1553 bdrv_rebind(bs_old
);
1557 * Add new bs contents at the top of an image chain while the chain is
1558 * live, while keeping required fields on the top layer.
1560 * This will modify the BlockDriverState fields, and swap contents
1561 * between bs_new and bs_top. Both bs_new and bs_top are modified.
1563 * bs_new is required to be anonymous.
1565 * This function does not create any image files.
1567 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
)
1569 bdrv_swap(bs_new
, bs_top
);
1571 /* The contents of 'tmp' will become bs_top, as we are
1572 * swapping bs_new and bs_top contents. */
1573 bs_top
->backing_hd
= bs_new
;
1574 bs_top
->open_flags
&= ~BDRV_O_NO_BACKING
;
1575 pstrcpy(bs_top
->backing_file
, sizeof(bs_top
->backing_file
),
1577 pstrcpy(bs_top
->backing_format
, sizeof(bs_top
->backing_format
),
1578 bs_new
->drv
? bs_new
->drv
->format_name
: "");
1581 void bdrv_delete(BlockDriverState
*bs
)
1585 assert(!bs
->in_use
);
1587 /* remove from list, if necessary */
1592 assert(bs
!= bs_snapshots
);
1596 int bdrv_attach_dev(BlockDriverState
*bs
, void *dev
)
1597 /* TODO change to DeviceState *dev when all users are qdevified */
1603 bdrv_iostatus_reset(bs
);
1607 /* TODO qdevified devices don't use this, remove when devices are qdevified */
1608 void bdrv_attach_dev_nofail(BlockDriverState
*bs
, void *dev
)
1610 if (bdrv_attach_dev(bs
, dev
) < 0) {
1615 void bdrv_detach_dev(BlockDriverState
*bs
, void *dev
)
1616 /* TODO change to DeviceState *dev when all users are qdevified */
1618 assert(bs
->dev
== dev
);
1621 bs
->dev_opaque
= NULL
;
1622 bs
->buffer_alignment
= 512;
1625 /* TODO change to return DeviceState * when all users are qdevified */
1626 void *bdrv_get_attached_dev(BlockDriverState
*bs
)
1631 void bdrv_set_dev_ops(BlockDriverState
*bs
, const BlockDevOps
*ops
,
1635 bs
->dev_opaque
= opaque
;
1636 if (bdrv_dev_has_removable_media(bs
) && bs
== bs_snapshots
) {
1637 bs_snapshots
= NULL
;
1641 void bdrv_emit_qmp_error_event(const BlockDriverState
*bdrv
,
1642 enum MonitorEvent ev
,
1643 BlockErrorAction action
, bool is_read
)
1646 const char *action_str
;
1649 case BDRV_ACTION_REPORT
:
1650 action_str
= "report";
1652 case BDRV_ACTION_IGNORE
:
1653 action_str
= "ignore";
1655 case BDRV_ACTION_STOP
:
1656 action_str
= "stop";
1662 data
= qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1665 is_read
? "read" : "write");
1666 monitor_protocol_event(ev
, data
);
1668 qobject_decref(data
);
1671 static void bdrv_emit_qmp_eject_event(BlockDriverState
*bs
, bool ejected
)
1675 data
= qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }",
1676 bdrv_get_device_name(bs
), ejected
);
1677 monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED
, data
);
1679 qobject_decref(data
);
1682 static void bdrv_dev_change_media_cb(BlockDriverState
*bs
, bool load
)
1684 if (bs
->dev_ops
&& bs
->dev_ops
->change_media_cb
) {
1685 bool tray_was_closed
= !bdrv_dev_is_tray_open(bs
);
1686 bs
->dev_ops
->change_media_cb(bs
->dev_opaque
, load
);
1687 if (tray_was_closed
) {
1689 bdrv_emit_qmp_eject_event(bs
, true);
1693 bdrv_emit_qmp_eject_event(bs
, false);
1698 bool bdrv_dev_has_removable_media(BlockDriverState
*bs
)
1700 return !bs
->dev
|| (bs
->dev_ops
&& bs
->dev_ops
->change_media_cb
);
1703 void bdrv_dev_eject_request(BlockDriverState
*bs
, bool force
)
1705 if (bs
->dev_ops
&& bs
->dev_ops
->eject_request_cb
) {
1706 bs
->dev_ops
->eject_request_cb(bs
->dev_opaque
, force
);
1710 bool bdrv_dev_is_tray_open(BlockDriverState
*bs
)
1712 if (bs
->dev_ops
&& bs
->dev_ops
->is_tray_open
) {
1713 return bs
->dev_ops
->is_tray_open(bs
->dev_opaque
);
1718 static void bdrv_dev_resize_cb(BlockDriverState
*bs
)
1720 if (bs
->dev_ops
&& bs
->dev_ops
->resize_cb
) {
1721 bs
->dev_ops
->resize_cb(bs
->dev_opaque
);
1725 bool bdrv_dev_is_medium_locked(BlockDriverState
*bs
)
1727 if (bs
->dev_ops
&& bs
->dev_ops
->is_medium_locked
) {
1728 return bs
->dev_ops
->is_medium_locked(bs
->dev_opaque
);
1734 * Run consistency checks on an image
1736 * Returns 0 if the check could be completed (it doesn't mean that the image is
1737 * free of errors) or -errno when an internal error occurred. The results of the
1738 * check are stored in res.
1740 int bdrv_check(BlockDriverState
*bs
, BdrvCheckResult
*res
, BdrvCheckMode fix
)
1742 if (bs
->drv
->bdrv_check
== NULL
) {
1746 memset(res
, 0, sizeof(*res
));
1747 return bs
->drv
->bdrv_check(bs
, res
, fix
);
1750 #define COMMIT_BUF_SECTORS 2048
1752 /* commit COW file into the raw image */
1753 int bdrv_commit(BlockDriverState
*bs
)
1755 BlockDriver
*drv
= bs
->drv
;
1756 int64_t sector
, total_sectors
;
1757 int n
, ro
, open_flags
;
1760 char filename
[PATH_MAX
];
1765 if (!bs
->backing_hd
) {
1769 if (bdrv_in_use(bs
) || bdrv_in_use(bs
->backing_hd
)) {
1773 ro
= bs
->backing_hd
->read_only
;
1774 /* Use pstrcpy (not strncpy): filename must be NUL-terminated. */
1775 pstrcpy(filename
, sizeof(filename
), bs
->backing_hd
->filename
);
1776 open_flags
= bs
->backing_hd
->open_flags
;
1779 if (bdrv_reopen(bs
->backing_hd
, open_flags
| BDRV_O_RDWR
, NULL
)) {
1784 total_sectors
= bdrv_getlength(bs
) >> BDRV_SECTOR_BITS
;
1785 buf
= g_malloc(COMMIT_BUF_SECTORS
* BDRV_SECTOR_SIZE
);
1787 for (sector
= 0; sector
< total_sectors
; sector
+= n
) {
1788 if (bdrv_is_allocated(bs
, sector
, COMMIT_BUF_SECTORS
, &n
)) {
1790 if (bdrv_read(bs
, sector
, buf
, n
) != 0) {
1795 if (bdrv_write(bs
->backing_hd
, sector
, buf
, n
) != 0) {
1802 if (drv
->bdrv_make_empty
) {
1803 ret
= drv
->bdrv_make_empty(bs
);
1808 * Make sure all data we wrote to the backing device is actually
1812 bdrv_flush(bs
->backing_hd
);
1818 /* ignoring error return here */
1819 bdrv_reopen(bs
->backing_hd
, open_flags
& ~BDRV_O_RDWR
, NULL
);
1825 int bdrv_commit_all(void)
1827 BlockDriverState
*bs
;
1829 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1830 if (bs
->drv
&& bs
->backing_hd
) {
1831 int ret
= bdrv_commit(bs
);
1840 struct BdrvTrackedRequest
{
1841 BlockDriverState
*bs
;
1845 QLIST_ENTRY(BdrvTrackedRequest
) list
;
1846 Coroutine
*co
; /* owner, used for deadlock detection */
1847 CoQueue wait_queue
; /* coroutines blocked on this request */
1851 * Remove an active request from the tracked requests list
1853 * This function should be called when a tracked request is completing.
1855 static void tracked_request_end(BdrvTrackedRequest
*req
)
1857 QLIST_REMOVE(req
, list
);
1858 qemu_co_queue_restart_all(&req
->wait_queue
);
1862 * Add an active request to the tracked requests list
1864 static void tracked_request_begin(BdrvTrackedRequest
*req
,
1865 BlockDriverState
*bs
,
1867 int nb_sectors
, bool is_write
)
1869 *req
= (BdrvTrackedRequest
){
1871 .sector_num
= sector_num
,
1872 .nb_sectors
= nb_sectors
,
1873 .is_write
= is_write
,
1874 .co
= qemu_coroutine_self(),
1877 qemu_co_queue_init(&req
->wait_queue
);
1879 QLIST_INSERT_HEAD(&bs
->tracked_requests
, req
, list
);
1883 * Round a region to cluster boundaries
1885 void bdrv_round_to_clusters(BlockDriverState
*bs
,
1886 int64_t sector_num
, int nb_sectors
,
1887 int64_t *cluster_sector_num
,
1888 int *cluster_nb_sectors
)
1890 BlockDriverInfo bdi
;
1892 if (bdrv_get_info(bs
, &bdi
) < 0 || bdi
.cluster_size
== 0) {
1893 *cluster_sector_num
= sector_num
;
1894 *cluster_nb_sectors
= nb_sectors
;
1896 int64_t c
= bdi
.cluster_size
/ BDRV_SECTOR_SIZE
;
1897 *cluster_sector_num
= QEMU_ALIGN_DOWN(sector_num
, c
);
1898 *cluster_nb_sectors
= QEMU_ALIGN_UP(sector_num
- *cluster_sector_num
+
1903 static bool tracked_request_overlaps(BdrvTrackedRequest
*req
,
1904 int64_t sector_num
, int nb_sectors
) {
1906 if (sector_num
>= req
->sector_num
+ req
->nb_sectors
) {
1910 if (req
->sector_num
>= sector_num
+ nb_sectors
) {
1916 static void coroutine_fn
wait_for_overlapping_requests(BlockDriverState
*bs
,
1917 int64_t sector_num
, int nb_sectors
)
1919 BdrvTrackedRequest
*req
;
1920 int64_t cluster_sector_num
;
1921 int cluster_nb_sectors
;
1924 /* If we touch the same cluster it counts as an overlap. This guarantees
1925 * that allocating writes will be serialized and not race with each other
1926 * for the same cluster. For example, in copy-on-read it ensures that the
1927 * CoR read and write operations are atomic and guest writes cannot
1928 * interleave between them.
1930 bdrv_round_to_clusters(bs
, sector_num
, nb_sectors
,
1931 &cluster_sector_num
, &cluster_nb_sectors
);
1935 QLIST_FOREACH(req
, &bs
->tracked_requests
, list
) {
1936 if (tracked_request_overlaps(req
, cluster_sector_num
,
1937 cluster_nb_sectors
)) {
1938 /* Hitting this means there was a reentrant request, for
1939 * example, a block driver issuing nested requests. This must
1940 * never happen since it means deadlock.
1942 assert(qemu_coroutine_self() != req
->co
);
1944 qemu_co_queue_wait(&req
->wait_queue
);
1955 * -EINVAL - backing format specified, but no file
1956 * -ENOSPC - can't update the backing file because no space is left in the
1958 * -ENOTSUP - format driver doesn't support changing the backing file
1960 int bdrv_change_backing_file(BlockDriverState
*bs
,
1961 const char *backing_file
, const char *backing_fmt
)
1963 BlockDriver
*drv
= bs
->drv
;
1966 /* Backing file format doesn't make sense without a backing file */
1967 if (backing_fmt
&& !backing_file
) {
1971 if (drv
->bdrv_change_backing_file
!= NULL
) {
1972 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
1978 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
1979 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
1985 * Finds the image layer in the chain that has 'bs' as its backing file.
1987 * active is the current topmost image.
1989 * Returns NULL if bs is not found in active's image chain,
1990 * or if active == bs.
1992 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
1993 BlockDriverState
*bs
)
1995 BlockDriverState
*overlay
= NULL
;
1996 BlockDriverState
*intermediate
;
1998 assert(active
!= NULL
);
2001 /* if bs is the same as active, then by definition it has no overlay
2007 intermediate
= active
;
2008 while (intermediate
->backing_hd
) {
2009 if (intermediate
->backing_hd
== bs
) {
2010 overlay
= intermediate
;
2013 intermediate
= intermediate
->backing_hd
;
2019 typedef struct BlkIntermediateStates
{
2020 BlockDriverState
*bs
;
2021 QSIMPLEQ_ENTRY(BlkIntermediateStates
) entry
;
2022 } BlkIntermediateStates
;
2026 * Drops images above 'base' up to and including 'top', and sets the image
2027 * above 'top' to have base as its backing file.
2029 * Requires that the overlay to 'top' is opened r/w, so that the backing file
2030 * information in 'bs' can be properly updated.
2032 * E.g., this will convert the following chain:
2033 * bottom <- base <- intermediate <- top <- active
2037 * bottom <- base <- active
2039 * It is allowed for bottom==base, in which case it converts:
2041 * base <- intermediate <- top <- active
2048 * if active == top, that is considered an error
2051 int bdrv_drop_intermediate(BlockDriverState
*active
, BlockDriverState
*top
,
2052 BlockDriverState
*base
)
2054 BlockDriverState
*intermediate
;
2055 BlockDriverState
*base_bs
= NULL
;
2056 BlockDriverState
*new_top_bs
= NULL
;
2057 BlkIntermediateStates
*intermediate_state
, *next
;
2060 QSIMPLEQ_HEAD(states_to_delete
, BlkIntermediateStates
) states_to_delete
;
2061 QSIMPLEQ_INIT(&states_to_delete
);
2063 if (!top
->drv
|| !base
->drv
) {
2067 new_top_bs
= bdrv_find_overlay(active
, top
);
2069 if (new_top_bs
== NULL
) {
2070 /* we could not find the image above 'top', this is an error */
2074 /* special case of new_top_bs->backing_hd already pointing to base - nothing
2075 * to do, no intermediate images */
2076 if (new_top_bs
->backing_hd
== base
) {
2083 /* now we will go down through the list, and add each BDS we find
2084 * into our deletion queue, until we hit the 'base'
2086 while (intermediate
) {
2087 intermediate_state
= g_malloc0(sizeof(BlkIntermediateStates
));
2088 intermediate_state
->bs
= intermediate
;
2089 QSIMPLEQ_INSERT_TAIL(&states_to_delete
, intermediate_state
, entry
);
2091 if (intermediate
->backing_hd
== base
) {
2092 base_bs
= intermediate
->backing_hd
;
2095 intermediate
= intermediate
->backing_hd
;
2097 if (base_bs
== NULL
) {
2098 /* something went wrong, we did not end at the base. safely
2099 * unravel everything, and exit with error */
2103 /* success - we can delete the intermediate states, and link top->base */
2104 ret
= bdrv_change_backing_file(new_top_bs
, base_bs
->filename
,
2105 base_bs
->drv
? base_bs
->drv
->format_name
: "");
2109 new_top_bs
->backing_hd
= base_bs
;
2112 QSIMPLEQ_FOREACH_SAFE(intermediate_state
, &states_to_delete
, entry
, next
) {
2113 /* so that bdrv_close() does not recursively close the chain */
2114 intermediate_state
->bs
->backing_hd
= NULL
;
2115 bdrv_delete(intermediate_state
->bs
);
2120 QSIMPLEQ_FOREACH_SAFE(intermediate_state
, &states_to_delete
, entry
, next
) {
2121 g_free(intermediate_state
);
2127 static int bdrv_check_byte_request(BlockDriverState
*bs
, int64_t offset
,
2132 if (!bdrv_is_inserted(bs
))
2138 len
= bdrv_getlength(bs
);
2143 if ((offset
> len
) || (len
- offset
< size
))
2149 static int bdrv_check_request(BlockDriverState
*bs
, int64_t sector_num
,
2152 return bdrv_check_byte_request(bs
, sector_num
* BDRV_SECTOR_SIZE
,
2153 nb_sectors
* BDRV_SECTOR_SIZE
);
2156 typedef struct RwCo
{
2157 BlockDriverState
*bs
;
2165 static void coroutine_fn
bdrv_rw_co_entry(void *opaque
)
2167 RwCo
*rwco
= opaque
;
2169 if (!rwco
->is_write
) {
2170 rwco
->ret
= bdrv_co_do_readv(rwco
->bs
, rwco
->sector_num
,
2171 rwco
->nb_sectors
, rwco
->qiov
, 0);
2173 rwco
->ret
= bdrv_co_do_writev(rwco
->bs
, rwco
->sector_num
,
2174 rwco
->nb_sectors
, rwco
->qiov
, 0);
2179 * Process a vectored synchronous request using coroutines
2181 static int bdrv_rwv_co(BlockDriverState
*bs
, int64_t sector_num
,
2182 QEMUIOVector
*qiov
, bool is_write
)
2187 .sector_num
= sector_num
,
2188 .nb_sectors
= qiov
->size
>> BDRV_SECTOR_BITS
,
2190 .is_write
= is_write
,
2193 assert((qiov
->size
& (BDRV_SECTOR_SIZE
- 1)) == 0);
2196 * In sync call context, when the vcpu is blocked, this throttling timer
2197 * will not fire; so the I/O throttling function has to be disabled here
2198 * if it has been enabled.
2200 if (bs
->io_limits_enabled
) {
2201 fprintf(stderr
, "Disabling I/O throttling on '%s' due "
2202 "to synchronous I/O.\n", bdrv_get_device_name(bs
));
2203 bdrv_io_limits_disable(bs
);
2206 if (qemu_in_coroutine()) {
2207 /* Fast-path if already in coroutine context */
2208 bdrv_rw_co_entry(&rwco
);
2210 co
= qemu_coroutine_create(bdrv_rw_co_entry
);
2211 qemu_coroutine_enter(co
, &rwco
);
2212 while (rwco
.ret
== NOT_DONE
) {
2220 * Process a synchronous request using coroutines
2222 static int bdrv_rw_co(BlockDriverState
*bs
, int64_t sector_num
, uint8_t *buf
,
2223 int nb_sectors
, bool is_write
)
2226 struct iovec iov
= {
2227 .iov_base
= (void *)buf
,
2228 .iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
,
2231 qemu_iovec_init_external(&qiov
, &iov
, 1);
2232 return bdrv_rwv_co(bs
, sector_num
, &qiov
, is_write
);
2235 /* return < 0 if error. See bdrv_write() for the return codes */
2236 int bdrv_read(BlockDriverState
*bs
, int64_t sector_num
,
2237 uint8_t *buf
, int nb_sectors
)
2239 return bdrv_rw_co(bs
, sector_num
, buf
, nb_sectors
, false);
2242 /* Just like bdrv_read(), but with I/O throttling temporarily disabled */
2243 int bdrv_read_unthrottled(BlockDriverState
*bs
, int64_t sector_num
,
2244 uint8_t *buf
, int nb_sectors
)
2249 enabled
= bs
->io_limits_enabled
;
2250 bs
->io_limits_enabled
= false;
2251 ret
= bdrv_read(bs
, 0, buf
, 1);
2252 bs
->io_limits_enabled
= enabled
;
2256 /* Return < 0 if error. Important errors are:
2257 -EIO generic I/O error (may happen for all errors)
2258 -ENOMEDIUM No media inserted.
2259 -EINVAL Invalid sector number or nb_sectors
2260 -EACCES Trying to write a read-only device
2262 int bdrv_write(BlockDriverState
*bs
, int64_t sector_num
,
2263 const uint8_t *buf
, int nb_sectors
)
2265 return bdrv_rw_co(bs
, sector_num
, (uint8_t *)buf
, nb_sectors
, true);
2268 int bdrv_writev(BlockDriverState
*bs
, int64_t sector_num
, QEMUIOVector
*qiov
)
2270 return bdrv_rwv_co(bs
, sector_num
, qiov
, true);
2273 int bdrv_pread(BlockDriverState
*bs
, int64_t offset
,
2274 void *buf
, int count1
)
2276 uint8_t tmp_buf
[BDRV_SECTOR_SIZE
];
2277 int len
, nb_sectors
, count
;
2282 /* first read to align to sector start */
2283 len
= (BDRV_SECTOR_SIZE
- offset
) & (BDRV_SECTOR_SIZE
- 1);
2286 sector_num
= offset
>> BDRV_SECTOR_BITS
;
2288 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
2290 memcpy(buf
, tmp_buf
+ (offset
& (BDRV_SECTOR_SIZE
- 1)), len
);
2298 /* read the sectors "in place" */
2299 nb_sectors
= count
>> BDRV_SECTOR_BITS
;
2300 if (nb_sectors
> 0) {
2301 if ((ret
= bdrv_read(bs
, sector_num
, buf
, nb_sectors
)) < 0)
2303 sector_num
+= nb_sectors
;
2304 len
= nb_sectors
<< BDRV_SECTOR_BITS
;
2309 /* add data from the last sector */
2311 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
2313 memcpy(buf
, tmp_buf
, count
);
2318 int bdrv_pwritev(BlockDriverState
*bs
, int64_t offset
, QEMUIOVector
*qiov
)
2320 uint8_t tmp_buf
[BDRV_SECTOR_SIZE
];
2321 int len
, nb_sectors
, count
;
2327 /* first write to align to sector start */
2328 len
= (BDRV_SECTOR_SIZE
- offset
) & (BDRV_SECTOR_SIZE
- 1);
2331 sector_num
= offset
>> BDRV_SECTOR_BITS
;
2333 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
2335 qemu_iovec_to_buf(qiov
, 0, tmp_buf
+ (offset
& (BDRV_SECTOR_SIZE
- 1)),
2337 if ((ret
= bdrv_write(bs
, sector_num
, tmp_buf
, 1)) < 0)
2345 /* write the sectors "in place" */
2346 nb_sectors
= count
>> BDRV_SECTOR_BITS
;
2347 if (nb_sectors
> 0) {
2348 QEMUIOVector qiov_inplace
;
2350 qemu_iovec_init(&qiov_inplace
, qiov
->niov
);
2351 qemu_iovec_concat(&qiov_inplace
, qiov
, len
,
2352 nb_sectors
<< BDRV_SECTOR_BITS
);
2353 ret
= bdrv_writev(bs
, sector_num
, &qiov_inplace
);
2354 qemu_iovec_destroy(&qiov_inplace
);
2359 sector_num
+= nb_sectors
;
2360 len
= nb_sectors
<< BDRV_SECTOR_BITS
;
2364 /* add data from the last sector */
2366 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
2368 qemu_iovec_to_buf(qiov
, qiov
->size
- count
, tmp_buf
, count
);
2369 if ((ret
= bdrv_write(bs
, sector_num
, tmp_buf
, 1)) < 0)
2375 int bdrv_pwrite(BlockDriverState
*bs
, int64_t offset
,
2376 const void *buf
, int count1
)
2379 struct iovec iov
= {
2380 .iov_base
= (void *) buf
,
2384 qemu_iovec_init_external(&qiov
, &iov
, 1);
2385 return bdrv_pwritev(bs
, offset
, &qiov
);
2389 * Writes to the file and ensures that no writes are reordered across this
2390 * request (acts as a barrier)
2392 * Returns 0 on success, -errno in error cases.
2394 int bdrv_pwrite_sync(BlockDriverState
*bs
, int64_t offset
,
2395 const void *buf
, int count
)
2399 ret
= bdrv_pwrite(bs
, offset
, buf
, count
);
2404 /* No flush needed for cache modes that already do it */
2405 if (bs
->enable_write_cache
) {
2412 static int coroutine_fn
bdrv_co_do_copy_on_readv(BlockDriverState
*bs
,
2413 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
2415 /* Perform I/O through a temporary buffer so that users who scribble over
2416 * their read buffer while the operation is in progress do not end up
2417 * modifying the image file. This is critical for zero-copy guest I/O
2418 * where anything might happen inside guest memory.
2420 void *bounce_buffer
;
2422 BlockDriver
*drv
= bs
->drv
;
2424 QEMUIOVector bounce_qiov
;
2425 int64_t cluster_sector_num
;
2426 int cluster_nb_sectors
;
2430 /* Cover entire cluster so no additional backing file I/O is required when
2431 * allocating cluster in the image file.
2433 bdrv_round_to_clusters(bs
, sector_num
, nb_sectors
,
2434 &cluster_sector_num
, &cluster_nb_sectors
);
2436 trace_bdrv_co_do_copy_on_readv(bs
, sector_num
, nb_sectors
,
2437 cluster_sector_num
, cluster_nb_sectors
);
2439 iov
.iov_len
= cluster_nb_sectors
* BDRV_SECTOR_SIZE
;
2440 iov
.iov_base
= bounce_buffer
= qemu_blockalign(bs
, iov
.iov_len
);
2441 qemu_iovec_init_external(&bounce_qiov
, &iov
, 1);
2443 ret
= drv
->bdrv_co_readv(bs
, cluster_sector_num
, cluster_nb_sectors
,
2449 if (drv
->bdrv_co_write_zeroes
&&
2450 buffer_is_zero(bounce_buffer
, iov
.iov_len
)) {
2451 ret
= bdrv_co_do_write_zeroes(bs
, cluster_sector_num
,
2452 cluster_nb_sectors
);
2454 /* This does not change the data on the disk, it is not necessary
2455 * to flush even in cache=writethrough mode.
2457 ret
= drv
->bdrv_co_writev(bs
, cluster_sector_num
, cluster_nb_sectors
,
2462 /* It might be okay to ignore write errors for guest requests. If this
2463 * is a deliberate copy-on-read then we don't want to ignore the error.
2464 * Simply report it in all cases.
2469 skip_bytes
= (sector_num
- cluster_sector_num
) * BDRV_SECTOR_SIZE
;
2470 qemu_iovec_from_buf(qiov
, 0, bounce_buffer
+ skip_bytes
,
2471 nb_sectors
* BDRV_SECTOR_SIZE
);
2474 qemu_vfree(bounce_buffer
);
2479 * Handle a read request in coroutine context
2481 static int coroutine_fn
bdrv_co_do_readv(BlockDriverState
*bs
,
2482 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
2483 BdrvRequestFlags flags
)
2485 BlockDriver
*drv
= bs
->drv
;
2486 BdrvTrackedRequest req
;
2492 if (bdrv_check_request(bs
, sector_num
, nb_sectors
)) {
2496 /* throttling disk read I/O */
2497 if (bs
->io_limits_enabled
) {
2498 bdrv_io_limits_intercept(bs
, false, nb_sectors
);
2501 if (bs
->copy_on_read
) {
2502 flags
|= BDRV_REQ_COPY_ON_READ
;
2504 if (flags
& BDRV_REQ_COPY_ON_READ
) {
2505 bs
->copy_on_read_in_flight
++;
2508 if (bs
->copy_on_read_in_flight
) {
2509 wait_for_overlapping_requests(bs
, sector_num
, nb_sectors
);
2512 tracked_request_begin(&req
, bs
, sector_num
, nb_sectors
, false);
2514 if (flags
& BDRV_REQ_COPY_ON_READ
) {
2517 ret
= bdrv_co_is_allocated(bs
, sector_num
, nb_sectors
, &pnum
);
2522 if (!ret
|| pnum
!= nb_sectors
) {
2523 ret
= bdrv_co_do_copy_on_readv(bs
, sector_num
, nb_sectors
, qiov
);
2528 ret
= drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
2531 tracked_request_end(&req
);
2533 if (flags
& BDRV_REQ_COPY_ON_READ
) {
2534 bs
->copy_on_read_in_flight
--;
2540 int coroutine_fn
bdrv_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
2541 int nb_sectors
, QEMUIOVector
*qiov
)
2543 trace_bdrv_co_readv(bs
, sector_num
, nb_sectors
);
2545 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
, 0);
2548 int coroutine_fn
bdrv_co_copy_on_readv(BlockDriverState
*bs
,
2549 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
2551 trace_bdrv_co_copy_on_readv(bs
, sector_num
, nb_sectors
);
2553 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
,
2554 BDRV_REQ_COPY_ON_READ
);
2557 static int coroutine_fn
bdrv_co_do_write_zeroes(BlockDriverState
*bs
,
2558 int64_t sector_num
, int nb_sectors
)
2560 BlockDriver
*drv
= bs
->drv
;
2565 /* TODO Emulate only part of misaligned requests instead of letting block
2566 * drivers return -ENOTSUP and emulate everything */
2568 /* First try the efficient write zeroes operation */
2569 if (drv
->bdrv_co_write_zeroes
) {
2570 ret
= drv
->bdrv_co_write_zeroes(bs
, sector_num
, nb_sectors
);
2571 if (ret
!= -ENOTSUP
) {
2576 /* Fall back to bounce buffer if write zeroes is unsupported */
2577 iov
.iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
;
2578 iov
.iov_base
= qemu_blockalign(bs
, iov
.iov_len
);
2579 memset(iov
.iov_base
, 0, iov
.iov_len
);
2580 qemu_iovec_init_external(&qiov
, &iov
, 1);
2582 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, &qiov
);
2584 qemu_vfree(iov
.iov_base
);
2589 * Handle a write request in coroutine context
2591 static int coroutine_fn
bdrv_co_do_writev(BlockDriverState
*bs
,
2592 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
2593 BdrvRequestFlags flags
)
2595 BlockDriver
*drv
= bs
->drv
;
2596 BdrvTrackedRequest req
;
2602 if (bs
->read_only
) {
2605 if (bdrv_check_request(bs
, sector_num
, nb_sectors
)) {
2609 /* throttling disk write I/O */
2610 if (bs
->io_limits_enabled
) {
2611 bdrv_io_limits_intercept(bs
, true, nb_sectors
);
2614 if (bs
->copy_on_read_in_flight
) {
2615 wait_for_overlapping_requests(bs
, sector_num
, nb_sectors
);
2618 tracked_request_begin(&req
, bs
, sector_num
, nb_sectors
, true);
2620 if (flags
& BDRV_REQ_ZERO_WRITE
) {
2621 ret
= bdrv_co_do_write_zeroes(bs
, sector_num
, nb_sectors
);
2623 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, qiov
);
2626 if (ret
== 0 && !bs
->enable_write_cache
) {
2627 ret
= bdrv_co_flush(bs
);
2630 if (bs
->dirty_bitmap
) {
2631 bdrv_set_dirty(bs
, sector_num
, nb_sectors
);
2634 if (bs
->wr_highest_sector
< sector_num
+ nb_sectors
- 1) {
2635 bs
->wr_highest_sector
= sector_num
+ nb_sectors
- 1;
2638 tracked_request_end(&req
);
2643 int coroutine_fn
bdrv_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
2644 int nb_sectors
, QEMUIOVector
*qiov
)
2646 trace_bdrv_co_writev(bs
, sector_num
, nb_sectors
);
2648 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, qiov
, 0);
2651 int coroutine_fn
bdrv_co_write_zeroes(BlockDriverState
*bs
,
2652 int64_t sector_num
, int nb_sectors
)
2654 trace_bdrv_co_write_zeroes(bs
, sector_num
, nb_sectors
);
2656 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, NULL
,
2657 BDRV_REQ_ZERO_WRITE
);
2661 * Truncate file to 'offset' bytes (needed only for file protocols)
2663 int bdrv_truncate(BlockDriverState
*bs
, int64_t offset
)
2665 BlockDriver
*drv
= bs
->drv
;
2669 if (!drv
->bdrv_truncate
)
2673 if (bdrv_in_use(bs
))
2675 ret
= drv
->bdrv_truncate(bs
, offset
);
2677 ret
= refresh_total_sectors(bs
, offset
>> BDRV_SECTOR_BITS
);
2678 bdrv_dev_resize_cb(bs
);
2684 * Length of a allocated file in bytes. Sparse files are counted by actual
2685 * allocated space. Return < 0 if error or unknown.
2687 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
2689 BlockDriver
*drv
= bs
->drv
;
2693 if (drv
->bdrv_get_allocated_file_size
) {
2694 return drv
->bdrv_get_allocated_file_size(bs
);
2697 return bdrv_get_allocated_file_size(bs
->file
);
2703 * Length of a file in bytes. Return < 0 if error or unknown.
2705 int64_t bdrv_getlength(BlockDriverState
*bs
)
2707 BlockDriver
*drv
= bs
->drv
;
2711 if (bs
->growable
|| bdrv_dev_has_removable_media(bs
)) {
2712 if (drv
->bdrv_getlength
) {
2713 return drv
->bdrv_getlength(bs
);
2716 return bs
->total_sectors
* BDRV_SECTOR_SIZE
;
2719 /* return 0 as number of sectors if no device present or error */
2720 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
2723 length
= bdrv_getlength(bs
);
2727 length
= length
>> BDRV_SECTOR_BITS
;
2728 *nb_sectors_ptr
= length
;
2731 /* throttling disk io limits */
2732 void bdrv_set_io_limits(BlockDriverState
*bs
,
2733 BlockIOLimit
*io_limits
)
2735 bs
->io_limits
= *io_limits
;
2736 bs
->io_limits_enabled
= bdrv_io_limits_enabled(bs
);
2739 void bdrv_set_on_error(BlockDriverState
*bs
, BlockdevOnError on_read_error
,
2740 BlockdevOnError on_write_error
)
2742 bs
->on_read_error
= on_read_error
;
2743 bs
->on_write_error
= on_write_error
;
2746 BlockdevOnError
bdrv_get_on_error(BlockDriverState
*bs
, bool is_read
)
2748 return is_read
? bs
->on_read_error
: bs
->on_write_error
;
2751 BlockErrorAction
bdrv_get_error_action(BlockDriverState
*bs
, bool is_read
, int error
)
2753 BlockdevOnError on_err
= is_read
? bs
->on_read_error
: bs
->on_write_error
;
2756 case BLOCKDEV_ON_ERROR_ENOSPC
:
2757 return (error
== ENOSPC
) ? BDRV_ACTION_STOP
: BDRV_ACTION_REPORT
;
2758 case BLOCKDEV_ON_ERROR_STOP
:
2759 return BDRV_ACTION_STOP
;
2760 case BLOCKDEV_ON_ERROR_REPORT
:
2761 return BDRV_ACTION_REPORT
;
2762 case BLOCKDEV_ON_ERROR_IGNORE
:
2763 return BDRV_ACTION_IGNORE
;
2769 /* This is done by device models because, while the block layer knows
2770 * about the error, it does not know whether an operation comes from
2771 * the device or the block layer (from a job, for example).
2773 void bdrv_error_action(BlockDriverState
*bs
, BlockErrorAction action
,
2774 bool is_read
, int error
)
2777 bdrv_emit_qmp_error_event(bs
, QEVENT_BLOCK_IO_ERROR
, action
, is_read
);
2778 if (action
== BDRV_ACTION_STOP
) {
2779 vm_stop(RUN_STATE_IO_ERROR
);
2780 bdrv_iostatus_set_err(bs
, error
);
2784 int bdrv_is_read_only(BlockDriverState
*bs
)
2786 return bs
->read_only
;
2789 int bdrv_is_sg(BlockDriverState
*bs
)
2794 int bdrv_enable_write_cache(BlockDriverState
*bs
)
2796 return bs
->enable_write_cache
;
2799 void bdrv_set_enable_write_cache(BlockDriverState
*bs
, bool wce
)
2801 bs
->enable_write_cache
= wce
;
2803 /* so a reopen() will preserve wce */
2805 bs
->open_flags
|= BDRV_O_CACHE_WB
;
2807 bs
->open_flags
&= ~BDRV_O_CACHE_WB
;
2811 int bdrv_is_encrypted(BlockDriverState
*bs
)
2813 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
)
2815 return bs
->encrypted
;
2818 int bdrv_key_required(BlockDriverState
*bs
)
2820 BlockDriverState
*backing_hd
= bs
->backing_hd
;
2822 if (backing_hd
&& backing_hd
->encrypted
&& !backing_hd
->valid_key
)
2824 return (bs
->encrypted
&& !bs
->valid_key
);
2827 int bdrv_set_key(BlockDriverState
*bs
, const char *key
)
2830 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
) {
2831 ret
= bdrv_set_key(bs
->backing_hd
, key
);
2837 if (!bs
->encrypted
) {
2839 } else if (!bs
->drv
|| !bs
->drv
->bdrv_set_key
) {
2842 ret
= bs
->drv
->bdrv_set_key(bs
, key
);
2845 } else if (!bs
->valid_key
) {
2847 /* call the change callback now, we skipped it on open */
2848 bdrv_dev_change_media_cb(bs
, true);
2853 const char *bdrv_get_format_name(BlockDriverState
*bs
)
2855 return bs
->drv
? bs
->drv
->format_name
: NULL
;
2858 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
2863 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
2864 it(opaque
, drv
->format_name
);
2868 BlockDriverState
*bdrv_find(const char *name
)
2870 BlockDriverState
*bs
;
2872 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2873 if (!strcmp(name
, bs
->device_name
)) {
2880 BlockDriverState
*bdrv_next(BlockDriverState
*bs
)
2883 return QTAILQ_FIRST(&bdrv_states
);
2885 return QTAILQ_NEXT(bs
, list
);
2888 void bdrv_iterate(void (*it
)(void *opaque
, BlockDriverState
*bs
), void *opaque
)
2890 BlockDriverState
*bs
;
2892 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2897 const char *bdrv_get_device_name(BlockDriverState
*bs
)
2899 return bs
->device_name
;
2902 int bdrv_get_flags(BlockDriverState
*bs
)
2904 return bs
->open_flags
;
2907 void bdrv_flush_all(void)
2909 BlockDriverState
*bs
;
2911 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2916 int bdrv_has_zero_init(BlockDriverState
*bs
)
2920 if (bs
->drv
->bdrv_has_zero_init
) {
2921 return bs
->drv
->bdrv_has_zero_init(bs
);
2927 typedef struct BdrvCoIsAllocatedData
{
2928 BlockDriverState
*bs
;
2929 BlockDriverState
*base
;
2935 } BdrvCoIsAllocatedData
;
2938 * Returns true iff the specified sector is present in the disk image. Drivers
2939 * not implementing the functionality are assumed to not support backing files,
2940 * hence all their sectors are reported as allocated.
2942 * If 'sector_num' is beyond the end of the disk image the return value is 0
2943 * and 'pnum' is set to 0.
2945 * 'pnum' is set to the number of sectors (including and immediately following
2946 * the specified sector) that are known to be in the same
2947 * allocated/unallocated state.
2949 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
2950 * beyond the end of the disk image it will be clamped.
2952 int coroutine_fn
bdrv_co_is_allocated(BlockDriverState
*bs
, int64_t sector_num
,
2953 int nb_sectors
, int *pnum
)
2957 if (sector_num
>= bs
->total_sectors
) {
2962 n
= bs
->total_sectors
- sector_num
;
2963 if (n
< nb_sectors
) {
2967 if (!bs
->drv
->bdrv_co_is_allocated
) {
2972 return bs
->drv
->bdrv_co_is_allocated(bs
, sector_num
, nb_sectors
, pnum
);
2975 /* Coroutine wrapper for bdrv_is_allocated() */
2976 static void coroutine_fn
bdrv_is_allocated_co_entry(void *opaque
)
2978 BdrvCoIsAllocatedData
*data
= opaque
;
2979 BlockDriverState
*bs
= data
->bs
;
2981 data
->ret
= bdrv_co_is_allocated(bs
, data
->sector_num
, data
->nb_sectors
,
2987 * Synchronous wrapper around bdrv_co_is_allocated().
2989 * See bdrv_co_is_allocated() for details.
2991 int bdrv_is_allocated(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
,
2995 BdrvCoIsAllocatedData data
= {
2997 .sector_num
= sector_num
,
2998 .nb_sectors
= nb_sectors
,
3003 co
= qemu_coroutine_create(bdrv_is_allocated_co_entry
);
3004 qemu_coroutine_enter(co
, &data
);
3005 while (!data
.done
) {
3012 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
3014 * Return true if the given sector is allocated in any image between
3015 * BASE and TOP (inclusive). BASE can be NULL to check if the given
3016 * sector is allocated in any image of the chain. Return false otherwise.
3018 * 'pnum' is set to the number of sectors (including and immediately following
3019 * the specified sector) that are known to be in the same
3020 * allocated/unallocated state.
3023 int coroutine_fn
bdrv_co_is_allocated_above(BlockDriverState
*top
,
3024 BlockDriverState
*base
,
3026 int nb_sectors
, int *pnum
)
3028 BlockDriverState
*intermediate
;
3029 int ret
, n
= nb_sectors
;
3032 while (intermediate
&& intermediate
!= base
) {
3034 ret
= bdrv_co_is_allocated(intermediate
, sector_num
, nb_sectors
,
3044 * [sector_num, nb_sectors] is unallocated on top but intermediate
3047 * [sector_num+x, nr_sectors] allocated.
3049 if (n
> pnum_inter
&&
3050 (intermediate
== top
||
3051 sector_num
+ pnum_inter
< intermediate
->total_sectors
)) {
3055 intermediate
= intermediate
->backing_hd
;
3062 /* Coroutine wrapper for bdrv_is_allocated_above() */
3063 static void coroutine_fn
bdrv_is_allocated_above_co_entry(void *opaque
)
3065 BdrvCoIsAllocatedData
*data
= opaque
;
3066 BlockDriverState
*top
= data
->bs
;
3067 BlockDriverState
*base
= data
->base
;
3069 data
->ret
= bdrv_co_is_allocated_above(top
, base
, data
->sector_num
,
3070 data
->nb_sectors
, data
->pnum
);
3075 * Synchronous wrapper around bdrv_co_is_allocated_above().
3077 * See bdrv_co_is_allocated_above() for details.
3079 int bdrv_is_allocated_above(BlockDriverState
*top
, BlockDriverState
*base
,
3080 int64_t sector_num
, int nb_sectors
, int *pnum
)
3083 BdrvCoIsAllocatedData data
= {
3086 .sector_num
= sector_num
,
3087 .nb_sectors
= nb_sectors
,
3092 co
= qemu_coroutine_create(bdrv_is_allocated_above_co_entry
);
3093 qemu_coroutine_enter(co
, &data
);
3094 while (!data
.done
) {
3100 BlockInfo
*bdrv_query_info(BlockDriverState
*bs
)
3102 BlockInfo
*info
= g_malloc0(sizeof(*info
));
3103 info
->device
= g_strdup(bs
->device_name
);
3104 info
->type
= g_strdup("unknown");
3105 info
->locked
= bdrv_dev_is_medium_locked(bs
);
3106 info
->removable
= bdrv_dev_has_removable_media(bs
);
3108 if (bdrv_dev_has_removable_media(bs
)) {
3109 info
->has_tray_open
= true;
3110 info
->tray_open
= bdrv_dev_is_tray_open(bs
);
3113 if (bdrv_iostatus_is_enabled(bs
)) {
3114 info
->has_io_status
= true;
3115 info
->io_status
= bs
->iostatus
;
3118 if (bs
->dirty_bitmap
) {
3119 info
->has_dirty
= true;
3120 info
->dirty
= g_malloc0(sizeof(*info
->dirty
));
3121 info
->dirty
->count
= bdrv_get_dirty_count(bs
) * BDRV_SECTOR_SIZE
;
3122 info
->dirty
->granularity
=
3123 ((int64_t) BDRV_SECTOR_SIZE
<< hbitmap_granularity(bs
->dirty_bitmap
));
3127 info
->has_inserted
= true;
3128 info
->inserted
= g_malloc0(sizeof(*info
->inserted
));
3129 info
->inserted
->file
= g_strdup(bs
->filename
);
3130 info
->inserted
->ro
= bs
->read_only
;
3131 info
->inserted
->drv
= g_strdup(bs
->drv
->format_name
);
3132 info
->inserted
->encrypted
= bs
->encrypted
;
3133 info
->inserted
->encryption_key_missing
= bdrv_key_required(bs
);
3135 if (bs
->backing_file
[0]) {
3136 info
->inserted
->has_backing_file
= true;
3137 info
->inserted
->backing_file
= g_strdup(bs
->backing_file
);
3140 info
->inserted
->backing_file_depth
= bdrv_get_backing_file_depth(bs
);
3142 if (bs
->io_limits_enabled
) {
3143 info
->inserted
->bps
=
3144 bs
->io_limits
.bps
[BLOCK_IO_LIMIT_TOTAL
];
3145 info
->inserted
->bps_rd
=
3146 bs
->io_limits
.bps
[BLOCK_IO_LIMIT_READ
];
3147 info
->inserted
->bps_wr
=
3148 bs
->io_limits
.bps
[BLOCK_IO_LIMIT_WRITE
];
3149 info
->inserted
->iops
=
3150 bs
->io_limits
.iops
[BLOCK_IO_LIMIT_TOTAL
];
3151 info
->inserted
->iops_rd
=
3152 bs
->io_limits
.iops
[BLOCK_IO_LIMIT_READ
];
3153 info
->inserted
->iops_wr
=
3154 bs
->io_limits
.iops
[BLOCK_IO_LIMIT_WRITE
];
3160 BlockInfoList
*qmp_query_block(Error
**errp
)
3162 BlockInfoList
*head
= NULL
, **p_next
= &head
;
3163 BlockDriverState
*bs
;
3165 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
3166 BlockInfoList
*info
= g_malloc0(sizeof(*info
));
3167 info
->value
= bdrv_query_info(bs
);
3170 p_next
= &info
->next
;
3176 BlockStats
*bdrv_query_stats(const BlockDriverState
*bs
)
3180 s
= g_malloc0(sizeof(*s
));
3182 if (bs
->device_name
[0]) {
3183 s
->has_device
= true;
3184 s
->device
= g_strdup(bs
->device_name
);
3187 s
->stats
= g_malloc0(sizeof(*s
->stats
));
3188 s
->stats
->rd_bytes
= bs
->nr_bytes
[BDRV_ACCT_READ
];
3189 s
->stats
->wr_bytes
= bs
->nr_bytes
[BDRV_ACCT_WRITE
];
3190 s
->stats
->rd_operations
= bs
->nr_ops
[BDRV_ACCT_READ
];
3191 s
->stats
->wr_operations
= bs
->nr_ops
[BDRV_ACCT_WRITE
];
3192 s
->stats
->wr_highest_offset
= bs
->wr_highest_sector
* BDRV_SECTOR_SIZE
;
3193 s
->stats
->flush_operations
= bs
->nr_ops
[BDRV_ACCT_FLUSH
];
3194 s
->stats
->wr_total_time_ns
= bs
->total_time_ns
[BDRV_ACCT_WRITE
];
3195 s
->stats
->rd_total_time_ns
= bs
->total_time_ns
[BDRV_ACCT_READ
];
3196 s
->stats
->flush_total_time_ns
= bs
->total_time_ns
[BDRV_ACCT_FLUSH
];
3199 s
->has_parent
= true;
3200 s
->parent
= bdrv_query_stats(bs
->file
);
3206 BlockStatsList
*qmp_query_blockstats(Error
**errp
)
3208 BlockStatsList
*head
= NULL
, **p_next
= &head
;
3209 BlockDriverState
*bs
;
3211 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
3212 BlockStatsList
*info
= g_malloc0(sizeof(*info
));
3213 info
->value
= bdrv_query_stats(bs
);
3216 p_next
= &info
->next
;
3222 const char *bdrv_get_encrypted_filename(BlockDriverState
*bs
)
3224 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
)
3225 return bs
->backing_file
;
3226 else if (bs
->encrypted
)
3227 return bs
->filename
;
3232 void bdrv_get_backing_filename(BlockDriverState
*bs
,
3233 char *filename
, int filename_size
)
3235 pstrcpy(filename
, filename_size
, bs
->backing_file
);
3238 int bdrv_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
3239 const uint8_t *buf
, int nb_sectors
)
3241 BlockDriver
*drv
= bs
->drv
;
3244 if (!drv
->bdrv_write_compressed
)
3246 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
3249 assert(!bs
->dirty_bitmap
);
3251 return drv
->bdrv_write_compressed(bs
, sector_num
, buf
, nb_sectors
);
3254 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
3256 BlockDriver
*drv
= bs
->drv
;
3259 if (!drv
->bdrv_get_info
)
3261 memset(bdi
, 0, sizeof(*bdi
));
3262 return drv
->bdrv_get_info(bs
, bdi
);
3265 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
3266 int64_t pos
, int size
)
3269 struct iovec iov
= {
3270 .iov_base
= (void *) buf
,
3274 qemu_iovec_init_external(&qiov
, &iov
, 1);
3275 return bdrv_writev_vmstate(bs
, &qiov
, pos
);
3278 int bdrv_writev_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
3280 BlockDriver
*drv
= bs
->drv
;
3284 } else if (drv
->bdrv_save_vmstate
) {
3285 return drv
->bdrv_save_vmstate(bs
, qiov
, pos
);
3286 } else if (bs
->file
) {
3287 return bdrv_writev_vmstate(bs
->file
, qiov
, pos
);
3293 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
3294 int64_t pos
, int size
)
3296 BlockDriver
*drv
= bs
->drv
;
3299 if (drv
->bdrv_load_vmstate
)
3300 return drv
->bdrv_load_vmstate(bs
, buf
, pos
, size
);
3302 return bdrv_load_vmstate(bs
->file
, buf
, pos
, size
);
3306 void bdrv_debug_event(BlockDriverState
*bs
, BlkDebugEvent event
)
3308 BlockDriver
*drv
= bs
->drv
;
3310 if (!drv
|| !drv
->bdrv_debug_event
) {
3314 drv
->bdrv_debug_event(bs
, event
);
3317 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
3320 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
3324 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
3325 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
3331 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
3333 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_resume
) {
3337 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
3338 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
3344 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
3346 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
3350 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
3351 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
3357 /**************************************************************/
3358 /* handling of snapshots */
3360 int bdrv_can_snapshot(BlockDriverState
*bs
)
3362 BlockDriver
*drv
= bs
->drv
;
3363 if (!drv
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
)) {
3367 if (!drv
->bdrv_snapshot_create
) {
3368 if (bs
->file
!= NULL
) {
3369 return bdrv_can_snapshot(bs
->file
);
3377 int bdrv_is_snapshot(BlockDriverState
*bs
)
3379 return !!(bs
->open_flags
& BDRV_O_SNAPSHOT
);
3382 BlockDriverState
*bdrv_snapshots(void)
3384 BlockDriverState
*bs
;
3387 return bs_snapshots
;
3391 while ((bs
= bdrv_next(bs
))) {
3392 if (bdrv_can_snapshot(bs
)) {
3400 int bdrv_snapshot_create(BlockDriverState
*bs
,
3401 QEMUSnapshotInfo
*sn_info
)
3403 BlockDriver
*drv
= bs
->drv
;
3406 if (drv
->bdrv_snapshot_create
)
3407 return drv
->bdrv_snapshot_create(bs
, sn_info
);
3409 return bdrv_snapshot_create(bs
->file
, sn_info
);
3413 int bdrv_snapshot_goto(BlockDriverState
*bs
,
3414 const char *snapshot_id
)
3416 BlockDriver
*drv
= bs
->drv
;
3421 if (drv
->bdrv_snapshot_goto
)
3422 return drv
->bdrv_snapshot_goto(bs
, snapshot_id
);
3425 drv
->bdrv_close(bs
);
3426 ret
= bdrv_snapshot_goto(bs
->file
, snapshot_id
);
3427 open_ret
= drv
->bdrv_open(bs
, NULL
, bs
->open_flags
);
3429 bdrv_delete(bs
->file
);
3439 int bdrv_snapshot_delete(BlockDriverState
*bs
, const char *snapshot_id
)
3441 BlockDriver
*drv
= bs
->drv
;
3444 if (drv
->bdrv_snapshot_delete
)
3445 return drv
->bdrv_snapshot_delete(bs
, snapshot_id
);
3447 return bdrv_snapshot_delete(bs
->file
, snapshot_id
);
3451 int bdrv_snapshot_list(BlockDriverState
*bs
,
3452 QEMUSnapshotInfo
**psn_info
)
3454 BlockDriver
*drv
= bs
->drv
;
3457 if (drv
->bdrv_snapshot_list
)
3458 return drv
->bdrv_snapshot_list(bs
, psn_info
);
3460 return bdrv_snapshot_list(bs
->file
, psn_info
);
3464 int bdrv_snapshot_load_tmp(BlockDriverState
*bs
,
3465 const char *snapshot_name
)
3467 BlockDriver
*drv
= bs
->drv
;
3471 if (!bs
->read_only
) {
3474 if (drv
->bdrv_snapshot_load_tmp
) {
3475 return drv
->bdrv_snapshot_load_tmp(bs
, snapshot_name
);
3480 /* backing_file can either be relative, or absolute, or a protocol. If it is
3481 * relative, it must be relative to the chain. So, passing in bs->filename
3482 * from a BDS as backing_file should not be done, as that may be relative to
3483 * the CWD rather than the chain. */
3484 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
3485 const char *backing_file
)
3487 char *filename_full
= NULL
;
3488 char *backing_file_full
= NULL
;
3489 char *filename_tmp
= NULL
;
3490 int is_protocol
= 0;
3491 BlockDriverState
*curr_bs
= NULL
;
3492 BlockDriverState
*retval
= NULL
;
3494 if (!bs
|| !bs
->drv
|| !backing_file
) {
3498 filename_full
= g_malloc(PATH_MAX
);
3499 backing_file_full
= g_malloc(PATH_MAX
);
3500 filename_tmp
= g_malloc(PATH_MAX
);
3502 is_protocol
= path_has_protocol(backing_file
);
3504 for (curr_bs
= bs
; curr_bs
->backing_hd
; curr_bs
= curr_bs
->backing_hd
) {
3506 /* If either of the filename paths is actually a protocol, then
3507 * compare unmodified paths; otherwise make paths relative */
3508 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
3509 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
3510 retval
= curr_bs
->backing_hd
;
3514 /* If not an absolute filename path, make it relative to the current
3515 * image's filename path */
3516 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
3519 /* We are going to compare absolute pathnames */
3520 if (!realpath(filename_tmp
, filename_full
)) {
3524 /* We need to make sure the backing filename we are comparing against
3525 * is relative to the current image filename (or absolute) */
3526 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
3527 curr_bs
->backing_file
);
3529 if (!realpath(filename_tmp
, backing_file_full
)) {
3533 if (strcmp(backing_file_full
, filename_full
) == 0) {
3534 retval
= curr_bs
->backing_hd
;
3540 g_free(filename_full
);
3541 g_free(backing_file_full
);
3542 g_free(filename_tmp
);
3546 int bdrv_get_backing_file_depth(BlockDriverState
*bs
)
3552 if (!bs
->backing_hd
) {
3556 return 1 + bdrv_get_backing_file_depth(bs
->backing_hd
);
3559 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
3561 BlockDriverState
*curr_bs
= NULL
;
3569 while (curr_bs
->backing_hd
) {
3570 curr_bs
= curr_bs
->backing_hd
;
3575 #define NB_SUFFIXES 4
3577 char *get_human_readable_size(char *buf
, int buf_size
, int64_t size
)
3579 static const char suffixes
[NB_SUFFIXES
] = "KMGT";
3584 snprintf(buf
, buf_size
, "%" PRId64
, size
);
3587 for(i
= 0; i
< NB_SUFFIXES
; i
++) {
3588 if (size
< (10 * base
)) {
3589 snprintf(buf
, buf_size
, "%0.1f%c",
3590 (double)size
/ base
,
3593 } else if (size
< (1000 * base
) || i
== (NB_SUFFIXES
- 1)) {
3594 snprintf(buf
, buf_size
, "%" PRId64
"%c",
3595 ((size
+ (base
>> 1)) / base
),
3605 char *bdrv_snapshot_dump(char *buf
, int buf_size
, QEMUSnapshotInfo
*sn
)
3607 char buf1
[128], date_buf
[128], clock_buf
[128];
3613 snprintf(buf
, buf_size
,
3614 "%-10s%-20s%7s%20s%15s",
3615 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
3618 localtime_r(&ti
, &tm
);
3619 strftime(date_buf
, sizeof(date_buf
),
3620 "%Y-%m-%d %H:%M:%S", &tm
);
3621 secs
= sn
->vm_clock_nsec
/ 1000000000;
3622 snprintf(clock_buf
, sizeof(clock_buf
),
3623 "%02d:%02d:%02d.%03d",
3625 (int)((secs
/ 60) % 60),
3627 (int)((sn
->vm_clock_nsec
/ 1000000) % 1000));
3628 snprintf(buf
, buf_size
,
3629 "%-10s%-20s%7s%20s%15s",
3630 sn
->id_str
, sn
->name
,
3631 get_human_readable_size(buf1
, sizeof(buf1
), sn
->vm_state_size
),
3638 /**************************************************************/
3641 BlockDriverAIOCB
*bdrv_aio_readv(BlockDriverState
*bs
, int64_t sector_num
,
3642 QEMUIOVector
*qiov
, int nb_sectors
,
3643 BlockDriverCompletionFunc
*cb
, void *opaque
)
3645 trace_bdrv_aio_readv(bs
, sector_num
, nb_sectors
, opaque
);
3647 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
,
3651 BlockDriverAIOCB
*bdrv_aio_writev(BlockDriverState
*bs
, int64_t sector_num
,
3652 QEMUIOVector
*qiov
, int nb_sectors
,
3653 BlockDriverCompletionFunc
*cb
, void *opaque
)
3655 trace_bdrv_aio_writev(bs
, sector_num
, nb_sectors
, opaque
);
3657 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
,
3662 typedef struct MultiwriteCB
{
3667 BlockDriverCompletionFunc
*cb
;
3669 QEMUIOVector
*free_qiov
;
3673 static void multiwrite_user_cb(MultiwriteCB
*mcb
)
3677 for (i
= 0; i
< mcb
->num_callbacks
; i
++) {
3678 mcb
->callbacks
[i
].cb(mcb
->callbacks
[i
].opaque
, mcb
->error
);
3679 if (mcb
->callbacks
[i
].free_qiov
) {
3680 qemu_iovec_destroy(mcb
->callbacks
[i
].free_qiov
);
3682 g_free(mcb
->callbacks
[i
].free_qiov
);
3686 static void multiwrite_cb(void *opaque
, int ret
)
3688 MultiwriteCB
*mcb
= opaque
;
3690 trace_multiwrite_cb(mcb
, ret
);
3692 if (ret
< 0 && !mcb
->error
) {
3696 mcb
->num_requests
--;
3697 if (mcb
->num_requests
== 0) {
3698 multiwrite_user_cb(mcb
);
3703 static int multiwrite_req_compare(const void *a
, const void *b
)
3705 const BlockRequest
*req1
= a
, *req2
= b
;
3708 * Note that we can't simply subtract req2->sector from req1->sector
3709 * here as that could overflow the return value.
3711 if (req1
->sector
> req2
->sector
) {
3713 } else if (req1
->sector
< req2
->sector
) {
3721 * Takes a bunch of requests and tries to merge them. Returns the number of
3722 * requests that remain after merging.
3724 static int multiwrite_merge(BlockDriverState
*bs
, BlockRequest
*reqs
,
3725 int num_reqs
, MultiwriteCB
*mcb
)
3729 // Sort requests by start sector
3730 qsort(reqs
, num_reqs
, sizeof(*reqs
), &multiwrite_req_compare
);
3732 // Check if adjacent requests touch the same clusters. If so, combine them,
3733 // filling up gaps with zero sectors.
3735 for (i
= 1; i
< num_reqs
; i
++) {
3737 int64_t oldreq_last
= reqs
[outidx
].sector
+ reqs
[outidx
].nb_sectors
;
3739 // Handle exactly sequential writes and overlapping writes.
3740 if (reqs
[i
].sector
<= oldreq_last
) {
3744 if (reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1 > IOV_MAX
) {
3750 QEMUIOVector
*qiov
= g_malloc0(sizeof(*qiov
));
3751 qemu_iovec_init(qiov
,
3752 reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1);
3754 // Add the first request to the merged one. If the requests are
3755 // overlapping, drop the last sectors of the first request.
3756 size
= (reqs
[i
].sector
- reqs
[outidx
].sector
) << 9;
3757 qemu_iovec_concat(qiov
, reqs
[outidx
].qiov
, 0, size
);
3759 // We should need to add any zeros between the two requests
3760 assert (reqs
[i
].sector
<= oldreq_last
);
3762 // Add the second request
3763 qemu_iovec_concat(qiov
, reqs
[i
].qiov
, 0, reqs
[i
].qiov
->size
);
3765 reqs
[outidx
].nb_sectors
= qiov
->size
>> 9;
3766 reqs
[outidx
].qiov
= qiov
;
3768 mcb
->callbacks
[i
].free_qiov
= reqs
[outidx
].qiov
;
3771 reqs
[outidx
].sector
= reqs
[i
].sector
;
3772 reqs
[outidx
].nb_sectors
= reqs
[i
].nb_sectors
;
3773 reqs
[outidx
].qiov
= reqs
[i
].qiov
;
3781 * Submit multiple AIO write requests at once.
3783 * On success, the function returns 0 and all requests in the reqs array have
3784 * been submitted. In error case this function returns -1, and any of the
3785 * requests may or may not be submitted yet. In particular, this means that the
3786 * callback will be called for some of the requests, for others it won't. The
3787 * caller must check the error field of the BlockRequest to wait for the right
3788 * callbacks (if error != 0, no callback will be called).
3790 * The implementation may modify the contents of the reqs array, e.g. to merge
3791 * requests. However, the fields opaque and error are left unmodified as they
3792 * are used to signal failure for a single request to the caller.
3794 int bdrv_aio_multiwrite(BlockDriverState
*bs
, BlockRequest
*reqs
, int num_reqs
)
3799 /* don't submit writes if we don't have a medium */
3800 if (bs
->drv
== NULL
) {
3801 for (i
= 0; i
< num_reqs
; i
++) {
3802 reqs
[i
].error
= -ENOMEDIUM
;
3807 if (num_reqs
== 0) {
3811 // Create MultiwriteCB structure
3812 mcb
= g_malloc0(sizeof(*mcb
) + num_reqs
* sizeof(*mcb
->callbacks
));
3813 mcb
->num_requests
= 0;
3814 mcb
->num_callbacks
= num_reqs
;
3816 for (i
= 0; i
< num_reqs
; i
++) {
3817 mcb
->callbacks
[i
].cb
= reqs
[i
].cb
;
3818 mcb
->callbacks
[i
].opaque
= reqs
[i
].opaque
;
3821 // Check for mergable requests
3822 num_reqs
= multiwrite_merge(bs
, reqs
, num_reqs
, mcb
);
3824 trace_bdrv_aio_multiwrite(mcb
, mcb
->num_callbacks
, num_reqs
);
3826 /* Run the aio requests. */
3827 mcb
->num_requests
= num_reqs
;
3828 for (i
= 0; i
< num_reqs
; i
++) {
3829 bdrv_aio_writev(bs
, reqs
[i
].sector
, reqs
[i
].qiov
,
3830 reqs
[i
].nb_sectors
, multiwrite_cb
, mcb
);
3836 void bdrv_aio_cancel(BlockDriverAIOCB
*acb
)
3838 acb
->aiocb_info
->cancel(acb
);
3841 /* block I/O throttling */
3842 static bool bdrv_exceed_bps_limits(BlockDriverState
*bs
, int nb_sectors
,
3843 bool is_write
, double elapsed_time
, uint64_t *wait
)
3845 uint64_t bps_limit
= 0;
3847 double bytes_limit
, bytes_base
, bytes_res
;
3848 double slice_time
, wait_time
;
3850 if (bs
->io_limits
.bps
[BLOCK_IO_LIMIT_TOTAL
]) {
3851 bps_limit
= bs
->io_limits
.bps
[BLOCK_IO_LIMIT_TOTAL
];
3852 } else if (bs
->io_limits
.bps
[is_write
]) {
3853 bps_limit
= bs
->io_limits
.bps
[is_write
];
3862 slice_time
= bs
->slice_end
- bs
->slice_start
;
3863 slice_time
/= (NANOSECONDS_PER_SECOND
);
3864 bytes_limit
= bps_limit
* slice_time
;
3865 bytes_base
= bs
->slice_submitted
.bytes
[is_write
];
3866 if (bs
->io_limits
.bps
[BLOCK_IO_LIMIT_TOTAL
]) {
3867 bytes_base
+= bs
->slice_submitted
.bytes
[!is_write
];
3870 /* bytes_base: the bytes of data which have been read/written; and
3871 * it is obtained from the history statistic info.
3872 * bytes_res: the remaining bytes of data which need to be read/written.
3873 * (bytes_base + bytes_res) / bps_limit: used to calcuate
3874 * the total time for completing reading/writting all data.
3876 bytes_res
= (unsigned) nb_sectors
* BDRV_SECTOR_SIZE
;
3878 if (bytes_base
+ bytes_res
<= bytes_limit
) {
3886 /* Calc approx time to dispatch */
3887 wait_time
= (bytes_base
+ bytes_res
) / bps_limit
- elapsed_time
;
3889 /* When the I/O rate at runtime exceeds the limits,
3890 * bs->slice_end need to be extended in order that the current statistic
3891 * info can be kept until the timer fire, so it is increased and tuned
3892 * based on the result of experiment.
3894 extension
= wait_time
* NANOSECONDS_PER_SECOND
;
3895 extension
= DIV_ROUND_UP(extension
, BLOCK_IO_SLICE_TIME
) *
3896 BLOCK_IO_SLICE_TIME
;
3897 bs
->slice_end
+= extension
;
3899 *wait
= wait_time
* NANOSECONDS_PER_SECOND
;
3905 static bool bdrv_exceed_iops_limits(BlockDriverState
*bs
, bool is_write
,
3906 double elapsed_time
, uint64_t *wait
)
3908 uint64_t iops_limit
= 0;
3909 double ios_limit
, ios_base
;
3910 double slice_time
, wait_time
;
3912 if (bs
->io_limits
.iops
[BLOCK_IO_LIMIT_TOTAL
]) {
3913 iops_limit
= bs
->io_limits
.iops
[BLOCK_IO_LIMIT_TOTAL
];
3914 } else if (bs
->io_limits
.iops
[is_write
]) {
3915 iops_limit
= bs
->io_limits
.iops
[is_write
];
3924 slice_time
= bs
->slice_end
- bs
->slice_start
;
3925 slice_time
/= (NANOSECONDS_PER_SECOND
);
3926 ios_limit
= iops_limit
* slice_time
;
3927 ios_base
= bs
->slice_submitted
.ios
[is_write
];
3928 if (bs
->io_limits
.iops
[BLOCK_IO_LIMIT_TOTAL
]) {
3929 ios_base
+= bs
->slice_submitted
.ios
[!is_write
];
3932 if (ios_base
+ 1 <= ios_limit
) {
3940 /* Calc approx time to dispatch, in seconds */
3941 wait_time
= (ios_base
+ 1) / iops_limit
;
3942 if (wait_time
> elapsed_time
) {
3943 wait_time
= wait_time
- elapsed_time
;
3948 /* Exceeded current slice, extend it by another slice time */
3949 bs
->slice_end
+= BLOCK_IO_SLICE_TIME
;
3951 *wait
= wait_time
* NANOSECONDS_PER_SECOND
;
3957 static bool bdrv_exceed_io_limits(BlockDriverState
*bs
, int nb_sectors
,
3958 bool is_write
, int64_t *wait
)
3960 int64_t now
, max_wait
;
3961 uint64_t bps_wait
= 0, iops_wait
= 0;
3962 double elapsed_time
;
3963 int bps_ret
, iops_ret
;
3965 now
= qemu_get_clock_ns(vm_clock
);
3966 if (now
> bs
->slice_end
) {
3967 bs
->slice_start
= now
;
3968 bs
->slice_end
= now
+ BLOCK_IO_SLICE_TIME
;
3969 memset(&bs
->slice_submitted
, 0, sizeof(bs
->slice_submitted
));
3972 elapsed_time
= now
- bs
->slice_start
;
3973 elapsed_time
/= (NANOSECONDS_PER_SECOND
);
3975 bps_ret
= bdrv_exceed_bps_limits(bs
, nb_sectors
,
3976 is_write
, elapsed_time
, &bps_wait
);
3977 iops_ret
= bdrv_exceed_iops_limits(bs
, is_write
,
3978 elapsed_time
, &iops_wait
);
3979 if (bps_ret
|| iops_ret
) {
3980 max_wait
= bps_wait
> iops_wait
? bps_wait
: iops_wait
;
3985 now
= qemu_get_clock_ns(vm_clock
);
3986 if (bs
->slice_end
< now
+ max_wait
) {
3987 bs
->slice_end
= now
+ max_wait
;
3997 bs
->slice_submitted
.bytes
[is_write
] += (int64_t)nb_sectors
*
3999 bs
->slice_submitted
.ios
[is_write
]++;
4004 /**************************************************************/
4005 /* async block device emulation */
4007 typedef struct BlockDriverAIOCBSync
{
4008 BlockDriverAIOCB common
;
4011 /* vector translation state */
4015 } BlockDriverAIOCBSync
;
4017 static void bdrv_aio_cancel_em(BlockDriverAIOCB
*blockacb
)
4019 BlockDriverAIOCBSync
*acb
=
4020 container_of(blockacb
, BlockDriverAIOCBSync
, common
);
4021 qemu_bh_delete(acb
->bh
);
4023 qemu_aio_release(acb
);
4026 static const AIOCBInfo bdrv_em_aiocb_info
= {
4027 .aiocb_size
= sizeof(BlockDriverAIOCBSync
),
4028 .cancel
= bdrv_aio_cancel_em
,
4031 static void bdrv_aio_bh_cb(void *opaque
)
4033 BlockDriverAIOCBSync
*acb
= opaque
;
4036 qemu_iovec_from_buf(acb
->qiov
, 0, acb
->bounce
, acb
->qiov
->size
);
4037 qemu_vfree(acb
->bounce
);
4038 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
4039 qemu_bh_delete(acb
->bh
);
4041 qemu_aio_release(acb
);
4044 static BlockDriverAIOCB
*bdrv_aio_rw_vector(BlockDriverState
*bs
,
4048 BlockDriverCompletionFunc
*cb
,
4053 BlockDriverAIOCBSync
*acb
;
4055 acb
= qemu_aio_get(&bdrv_em_aiocb_info
, bs
, cb
, opaque
);
4056 acb
->is_write
= is_write
;
4058 acb
->bounce
= qemu_blockalign(bs
, qiov
->size
);
4059 acb
->bh
= qemu_bh_new(bdrv_aio_bh_cb
, acb
);
4062 qemu_iovec_to_buf(acb
->qiov
, 0, acb
->bounce
, qiov
->size
);
4063 acb
->ret
= bs
->drv
->bdrv_write(bs
, sector_num
, acb
->bounce
, nb_sectors
);
4065 acb
->ret
= bs
->drv
->bdrv_read(bs
, sector_num
, acb
->bounce
, nb_sectors
);
4068 qemu_bh_schedule(acb
->bh
);
4070 return &acb
->common
;
4073 static BlockDriverAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
4074 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
4075 BlockDriverCompletionFunc
*cb
, void *opaque
)
4077 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 0);
4080 static BlockDriverAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
4081 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
4082 BlockDriverCompletionFunc
*cb
, void *opaque
)
4084 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 1);
4088 typedef struct BlockDriverAIOCBCoroutine
{
4089 BlockDriverAIOCB common
;
4094 } BlockDriverAIOCBCoroutine
;
4096 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB
*blockacb
)
4098 BlockDriverAIOCBCoroutine
*acb
=
4099 container_of(blockacb
, BlockDriverAIOCBCoroutine
, common
);
4108 static const AIOCBInfo bdrv_em_co_aiocb_info
= {
4109 .aiocb_size
= sizeof(BlockDriverAIOCBCoroutine
),
4110 .cancel
= bdrv_aio_co_cancel_em
,
4113 static void bdrv_co_em_bh(void *opaque
)
4115 BlockDriverAIOCBCoroutine
*acb
= opaque
;
4117 acb
->common
.cb(acb
->common
.opaque
, acb
->req
.error
);
4123 qemu_bh_delete(acb
->bh
);
4124 qemu_aio_release(acb
);
4127 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
4128 static void coroutine_fn
bdrv_co_do_rw(void *opaque
)
4130 BlockDriverAIOCBCoroutine
*acb
= opaque
;
4131 BlockDriverState
*bs
= acb
->common
.bs
;
4133 if (!acb
->is_write
) {
4134 acb
->req
.error
= bdrv_co_do_readv(bs
, acb
->req
.sector
,
4135 acb
->req
.nb_sectors
, acb
->req
.qiov
, 0);
4137 acb
->req
.error
= bdrv_co_do_writev(bs
, acb
->req
.sector
,
4138 acb
->req
.nb_sectors
, acb
->req
.qiov
, 0);
4141 acb
->bh
= qemu_bh_new(bdrv_co_em_bh
, acb
);
4142 qemu_bh_schedule(acb
->bh
);
4145 static BlockDriverAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
4149 BlockDriverCompletionFunc
*cb
,
4154 BlockDriverAIOCBCoroutine
*acb
;
4156 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
4157 acb
->req
.sector
= sector_num
;
4158 acb
->req
.nb_sectors
= nb_sectors
;
4159 acb
->req
.qiov
= qiov
;
4160 acb
->is_write
= is_write
;
4163 co
= qemu_coroutine_create(bdrv_co_do_rw
);
4164 qemu_coroutine_enter(co
, acb
);
4166 return &acb
->common
;
4169 static void coroutine_fn
bdrv_aio_flush_co_entry(void *opaque
)
4171 BlockDriverAIOCBCoroutine
*acb
= opaque
;
4172 BlockDriverState
*bs
= acb
->common
.bs
;
4174 acb
->req
.error
= bdrv_co_flush(bs
);
4175 acb
->bh
= qemu_bh_new(bdrv_co_em_bh
, acb
);
4176 qemu_bh_schedule(acb
->bh
);
4179 BlockDriverAIOCB
*bdrv_aio_flush(BlockDriverState
*bs
,
4180 BlockDriverCompletionFunc
*cb
, void *opaque
)
4182 trace_bdrv_aio_flush(bs
, opaque
);
4185 BlockDriverAIOCBCoroutine
*acb
;
4187 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
4190 co
= qemu_coroutine_create(bdrv_aio_flush_co_entry
);
4191 qemu_coroutine_enter(co
, acb
);
4193 return &acb
->common
;
4196 static void coroutine_fn
bdrv_aio_discard_co_entry(void *opaque
)
4198 BlockDriverAIOCBCoroutine
*acb
= opaque
;
4199 BlockDriverState
*bs
= acb
->common
.bs
;
4201 acb
->req
.error
= bdrv_co_discard(bs
, acb
->req
.sector
, acb
->req
.nb_sectors
);
4202 acb
->bh
= qemu_bh_new(bdrv_co_em_bh
, acb
);
4203 qemu_bh_schedule(acb
->bh
);
4206 BlockDriverAIOCB
*bdrv_aio_discard(BlockDriverState
*bs
,
4207 int64_t sector_num
, int nb_sectors
,
4208 BlockDriverCompletionFunc
*cb
, void *opaque
)
4211 BlockDriverAIOCBCoroutine
*acb
;
4213 trace_bdrv_aio_discard(bs
, sector_num
, nb_sectors
, opaque
);
4215 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
4216 acb
->req
.sector
= sector_num
;
4217 acb
->req
.nb_sectors
= nb_sectors
;
4219 co
= qemu_coroutine_create(bdrv_aio_discard_co_entry
);
4220 qemu_coroutine_enter(co
, acb
);
4222 return &acb
->common
;
4225 void bdrv_init(void)
4227 module_call_init(MODULE_INIT_BLOCK
);
4230 void bdrv_init_with_whitelist(void)
4232 use_bdrv_whitelist
= 1;
4236 void *qemu_aio_get(const AIOCBInfo
*aiocb_info
, BlockDriverState
*bs
,
4237 BlockDriverCompletionFunc
*cb
, void *opaque
)
4239 BlockDriverAIOCB
*acb
;
4241 acb
= g_slice_alloc(aiocb_info
->aiocb_size
);
4242 acb
->aiocb_info
= aiocb_info
;
4245 acb
->opaque
= opaque
;
4249 void qemu_aio_release(void *p
)
4251 BlockDriverAIOCB
*acb
= p
;
4252 g_slice_free1(acb
->aiocb_info
->aiocb_size
, acb
);
4255 /**************************************************************/
4256 /* Coroutine block device emulation */
4258 typedef struct CoroutineIOCompletion
{
4259 Coroutine
*coroutine
;
4261 } CoroutineIOCompletion
;
4263 static void bdrv_co_io_em_complete(void *opaque
, int ret
)
4265 CoroutineIOCompletion
*co
= opaque
;
4268 qemu_coroutine_enter(co
->coroutine
, NULL
);
4271 static int coroutine_fn
bdrv_co_io_em(BlockDriverState
*bs
, int64_t sector_num
,
4272 int nb_sectors
, QEMUIOVector
*iov
,
4275 CoroutineIOCompletion co
= {
4276 .coroutine
= qemu_coroutine_self(),
4278 BlockDriverAIOCB
*acb
;
4281 acb
= bs
->drv
->bdrv_aio_writev(bs
, sector_num
, iov
, nb_sectors
,
4282 bdrv_co_io_em_complete
, &co
);
4284 acb
= bs
->drv
->bdrv_aio_readv(bs
, sector_num
, iov
, nb_sectors
,
4285 bdrv_co_io_em_complete
, &co
);
4288 trace_bdrv_co_io_em(bs
, sector_num
, nb_sectors
, is_write
, acb
);
4292 qemu_coroutine_yield();
4297 static int coroutine_fn
bdrv_co_readv_em(BlockDriverState
*bs
,
4298 int64_t sector_num
, int nb_sectors
,
4301 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, false);
4304 static int coroutine_fn
bdrv_co_writev_em(BlockDriverState
*bs
,
4305 int64_t sector_num
, int nb_sectors
,
4308 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, true);
4311 static void coroutine_fn
bdrv_flush_co_entry(void *opaque
)
4313 RwCo
*rwco
= opaque
;
4315 rwco
->ret
= bdrv_co_flush(rwco
->bs
);
4318 int coroutine_fn
bdrv_co_flush(BlockDriverState
*bs
)
4322 if (!bs
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
)) {
4326 /* Write back cached data to the OS even with cache=unsafe */
4327 if (bs
->drv
->bdrv_co_flush_to_os
) {
4328 ret
= bs
->drv
->bdrv_co_flush_to_os(bs
);
4334 /* But don't actually force it to the disk with cache=unsafe */
4335 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
4339 if (bs
->drv
->bdrv_co_flush_to_disk
) {
4340 ret
= bs
->drv
->bdrv_co_flush_to_disk(bs
);
4341 } else if (bs
->drv
->bdrv_aio_flush
) {
4342 BlockDriverAIOCB
*acb
;
4343 CoroutineIOCompletion co
= {
4344 .coroutine
= qemu_coroutine_self(),
4347 acb
= bs
->drv
->bdrv_aio_flush(bs
, bdrv_co_io_em_complete
, &co
);
4351 qemu_coroutine_yield();
4356 * Some block drivers always operate in either writethrough or unsafe
4357 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
4358 * know how the server works (because the behaviour is hardcoded or
4359 * depends on server-side configuration), so we can't ensure that
4360 * everything is safe on disk. Returning an error doesn't work because
4361 * that would break guests even if the server operates in writethrough
4364 * Let's hope the user knows what he's doing.
4372 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
4373 * in the case of cache=unsafe, so there are no useless flushes.
4376 return bdrv_co_flush(bs
->file
);
4379 void bdrv_invalidate_cache(BlockDriverState
*bs
)
4381 if (bs
->drv
&& bs
->drv
->bdrv_invalidate_cache
) {
4382 bs
->drv
->bdrv_invalidate_cache(bs
);
4386 void bdrv_invalidate_cache_all(void)
4388 BlockDriverState
*bs
;
4390 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
4391 bdrv_invalidate_cache(bs
);
4395 void bdrv_clear_incoming_migration_all(void)
4397 BlockDriverState
*bs
;
4399 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
4400 bs
->open_flags
= bs
->open_flags
& ~(BDRV_O_INCOMING
);
4404 int bdrv_flush(BlockDriverState
*bs
)
4412 if (qemu_in_coroutine()) {
4413 /* Fast-path if already in coroutine context */
4414 bdrv_flush_co_entry(&rwco
);
4416 co
= qemu_coroutine_create(bdrv_flush_co_entry
);
4417 qemu_coroutine_enter(co
, &rwco
);
4418 while (rwco
.ret
== NOT_DONE
) {
4426 static void coroutine_fn
bdrv_discard_co_entry(void *opaque
)
4428 RwCo
*rwco
= opaque
;
4430 rwco
->ret
= bdrv_co_discard(rwco
->bs
, rwco
->sector_num
, rwco
->nb_sectors
);
4433 int coroutine_fn
bdrv_co_discard(BlockDriverState
*bs
, int64_t sector_num
,
4438 } else if (bdrv_check_request(bs
, sector_num
, nb_sectors
)) {
4440 } else if (bs
->read_only
) {
4444 if (bs
->dirty_bitmap
) {
4445 bdrv_reset_dirty(bs
, sector_num
, nb_sectors
);
4448 /* Do nothing if disabled. */
4449 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
4453 if (bs
->drv
->bdrv_co_discard
) {
4454 return bs
->drv
->bdrv_co_discard(bs
, sector_num
, nb_sectors
);
4455 } else if (bs
->drv
->bdrv_aio_discard
) {
4456 BlockDriverAIOCB
*acb
;
4457 CoroutineIOCompletion co
= {
4458 .coroutine
= qemu_coroutine_self(),
4461 acb
= bs
->drv
->bdrv_aio_discard(bs
, sector_num
, nb_sectors
,
4462 bdrv_co_io_em_complete
, &co
);
4466 qemu_coroutine_yield();
4474 int bdrv_discard(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
)
4479 .sector_num
= sector_num
,
4480 .nb_sectors
= nb_sectors
,
4484 if (qemu_in_coroutine()) {
4485 /* Fast-path if already in coroutine context */
4486 bdrv_discard_co_entry(&rwco
);
4488 co
= qemu_coroutine_create(bdrv_discard_co_entry
);
4489 qemu_coroutine_enter(co
, &rwco
);
4490 while (rwco
.ret
== NOT_DONE
) {
4498 /**************************************************************/
4499 /* removable device support */
4502 * Return TRUE if the media is present
4504 int bdrv_is_inserted(BlockDriverState
*bs
)
4506 BlockDriver
*drv
= bs
->drv
;
4510 if (!drv
->bdrv_is_inserted
)
4512 return drv
->bdrv_is_inserted(bs
);
4516 * Return whether the media changed since the last call to this
4517 * function, or -ENOTSUP if we don't know. Most drivers don't know.
4519 int bdrv_media_changed(BlockDriverState
*bs
)
4521 BlockDriver
*drv
= bs
->drv
;
4523 if (drv
&& drv
->bdrv_media_changed
) {
4524 return drv
->bdrv_media_changed(bs
);
4530 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4532 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
4534 BlockDriver
*drv
= bs
->drv
;
4536 if (drv
&& drv
->bdrv_eject
) {
4537 drv
->bdrv_eject(bs
, eject_flag
);
4540 if (bs
->device_name
[0] != '\0') {
4541 bdrv_emit_qmp_eject_event(bs
, eject_flag
);
4546 * Lock or unlock the media (if it is locked, the user won't be able
4547 * to eject it manually).
4549 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
4551 BlockDriver
*drv
= bs
->drv
;
4553 trace_bdrv_lock_medium(bs
, locked
);
4555 if (drv
&& drv
->bdrv_lock_medium
) {
4556 drv
->bdrv_lock_medium(bs
, locked
);
4560 /* needed for generic scsi interface */
4562 int bdrv_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
4564 BlockDriver
*drv
= bs
->drv
;
4566 if (drv
&& drv
->bdrv_ioctl
)
4567 return drv
->bdrv_ioctl(bs
, req
, buf
);
4571 BlockDriverAIOCB
*bdrv_aio_ioctl(BlockDriverState
*bs
,
4572 unsigned long int req
, void *buf
,
4573 BlockDriverCompletionFunc
*cb
, void *opaque
)
4575 BlockDriver
*drv
= bs
->drv
;
4577 if (drv
&& drv
->bdrv_aio_ioctl
)
4578 return drv
->bdrv_aio_ioctl(bs
, req
, buf
, cb
, opaque
);
4582 void bdrv_set_buffer_alignment(BlockDriverState
*bs
, int align
)
4584 bs
->buffer_alignment
= align
;
4587 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
4589 return qemu_memalign((bs
&& bs
->buffer_alignment
) ? bs
->buffer_alignment
: 512, size
);
4593 * Check if all memory in this vector is sector aligned.
4595 bool bdrv_qiov_is_aligned(BlockDriverState
*bs
, QEMUIOVector
*qiov
)
4599 for (i
= 0; i
< qiov
->niov
; i
++) {
4600 if ((uintptr_t) qiov
->iov
[i
].iov_base
% bs
->buffer_alignment
) {
4608 void bdrv_set_dirty_tracking(BlockDriverState
*bs
, int granularity
)
4610 int64_t bitmap_size
;
4612 assert((granularity
& (granularity
- 1)) == 0);
4615 granularity
>>= BDRV_SECTOR_BITS
;
4616 assert(!bs
->dirty_bitmap
);
4617 bitmap_size
= (bdrv_getlength(bs
) >> BDRV_SECTOR_BITS
);
4618 bs
->dirty_bitmap
= hbitmap_alloc(bitmap_size
, ffs(granularity
) - 1);
4620 if (bs
->dirty_bitmap
) {
4621 hbitmap_free(bs
->dirty_bitmap
);
4622 bs
->dirty_bitmap
= NULL
;
4627 int bdrv_get_dirty(BlockDriverState
*bs
, int64_t sector
)
4629 if (bs
->dirty_bitmap
) {
4630 return hbitmap_get(bs
->dirty_bitmap
, sector
);
4636 void bdrv_dirty_iter_init(BlockDriverState
*bs
, HBitmapIter
*hbi
)
4638 hbitmap_iter_init(hbi
, bs
->dirty_bitmap
, 0);
4641 void bdrv_set_dirty(BlockDriverState
*bs
, int64_t cur_sector
,
4644 hbitmap_set(bs
->dirty_bitmap
, cur_sector
, nr_sectors
);
4647 void bdrv_reset_dirty(BlockDriverState
*bs
, int64_t cur_sector
,
4650 hbitmap_reset(bs
->dirty_bitmap
, cur_sector
, nr_sectors
);
4653 int64_t bdrv_get_dirty_count(BlockDriverState
*bs
)
4655 if (bs
->dirty_bitmap
) {
4656 return hbitmap_count(bs
->dirty_bitmap
);
4662 void bdrv_set_in_use(BlockDriverState
*bs
, int in_use
)
4664 assert(bs
->in_use
!= in_use
);
4665 bs
->in_use
= in_use
;
4668 int bdrv_in_use(BlockDriverState
*bs
)
4673 void bdrv_iostatus_enable(BlockDriverState
*bs
)
4675 bs
->iostatus_enabled
= true;
4676 bs
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
4679 /* The I/O status is only enabled if the drive explicitly
4680 * enables it _and_ the VM is configured to stop on errors */
4681 bool bdrv_iostatus_is_enabled(const BlockDriverState
*bs
)
4683 return (bs
->iostatus_enabled
&&
4684 (bs
->on_write_error
== BLOCKDEV_ON_ERROR_ENOSPC
||
4685 bs
->on_write_error
== BLOCKDEV_ON_ERROR_STOP
||
4686 bs
->on_read_error
== BLOCKDEV_ON_ERROR_STOP
));
4689 void bdrv_iostatus_disable(BlockDriverState
*bs
)
4691 bs
->iostatus_enabled
= false;
4694 void bdrv_iostatus_reset(BlockDriverState
*bs
)
4696 if (bdrv_iostatus_is_enabled(bs
)) {
4697 bs
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
4699 block_job_iostatus_reset(bs
->job
);
4704 void bdrv_iostatus_set_err(BlockDriverState
*bs
, int error
)
4706 assert(bdrv_iostatus_is_enabled(bs
));
4707 if (bs
->iostatus
== BLOCK_DEVICE_IO_STATUS_OK
) {
4708 bs
->iostatus
= error
== ENOSPC
? BLOCK_DEVICE_IO_STATUS_NOSPACE
:
4709 BLOCK_DEVICE_IO_STATUS_FAILED
;
4714 bdrv_acct_start(BlockDriverState
*bs
, BlockAcctCookie
*cookie
, int64_t bytes
,
4715 enum BlockAcctType type
)
4717 assert(type
< BDRV_MAX_IOTYPE
);
4719 cookie
->bytes
= bytes
;
4720 cookie
->start_time_ns
= get_clock();
4721 cookie
->type
= type
;
4725 bdrv_acct_done(BlockDriverState
*bs
, BlockAcctCookie
*cookie
)
4727 assert(cookie
->type
< BDRV_MAX_IOTYPE
);
4729 bs
->nr_bytes
[cookie
->type
] += cookie
->bytes
;
4730 bs
->nr_ops
[cookie
->type
]++;
4731 bs
->total_time_ns
[cookie
->type
] += get_clock() - cookie
->start_time_ns
;
4734 void bdrv_img_create(const char *filename
, const char *fmt
,
4735 const char *base_filename
, const char *base_fmt
,
4736 char *options
, uint64_t img_size
, int flags
,
4737 Error
**errp
, bool quiet
)
4739 QEMUOptionParameter
*param
= NULL
, *create_options
= NULL
;
4740 QEMUOptionParameter
*backing_fmt
, *backing_file
, *size
;
4741 BlockDriverState
*bs
= NULL
;
4742 BlockDriver
*drv
, *proto_drv
;
4743 BlockDriver
*backing_drv
= NULL
;
4746 /* Find driver and parse its options */
4747 drv
= bdrv_find_format(fmt
);
4749 error_setg(errp
, "Unknown file format '%s'", fmt
);
4753 proto_drv
= bdrv_find_protocol(filename
);
4755 error_setg(errp
, "Unknown protocol '%s'", filename
);
4759 create_options
= append_option_parameters(create_options
,
4760 drv
->create_options
);
4761 create_options
= append_option_parameters(create_options
,
4762 proto_drv
->create_options
);
4764 /* Create parameter list with default values */
4765 param
= parse_option_parameters("", create_options
, param
);
4767 set_option_parameter_int(param
, BLOCK_OPT_SIZE
, img_size
);
4769 /* Parse -o options */
4771 param
= parse_option_parameters(options
, create_options
, param
);
4772 if (param
== NULL
) {
4773 error_setg(errp
, "Invalid options for file format '%s'.", fmt
);
4778 if (base_filename
) {
4779 if (set_option_parameter(param
, BLOCK_OPT_BACKING_FILE
,
4781 error_setg(errp
, "Backing file not supported for file format '%s'",
4788 if (set_option_parameter(param
, BLOCK_OPT_BACKING_FMT
, base_fmt
)) {
4789 error_setg(errp
, "Backing file format not supported for file "
4790 "format '%s'", fmt
);
4795 backing_file
= get_option_parameter(param
, BLOCK_OPT_BACKING_FILE
);
4796 if (backing_file
&& backing_file
->value
.s
) {
4797 if (!strcmp(filename
, backing_file
->value
.s
)) {
4798 error_setg(errp
, "Error: Trying to create an image with the "
4799 "same filename as the backing file");
4804 backing_fmt
= get_option_parameter(param
, BLOCK_OPT_BACKING_FMT
);
4805 if (backing_fmt
&& backing_fmt
->value
.s
) {
4806 backing_drv
= bdrv_find_format(backing_fmt
->value
.s
);
4808 error_setg(errp
, "Unknown backing file format '%s'",
4809 backing_fmt
->value
.s
);
4814 // The size for the image must always be specified, with one exception:
4815 // If we are using a backing file, we can obtain the size from there
4816 size
= get_option_parameter(param
, BLOCK_OPT_SIZE
);
4817 if (size
&& size
->value
.n
== -1) {
4818 if (backing_file
&& backing_file
->value
.s
) {
4823 /* backing files always opened read-only */
4825 flags
& ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
4829 ret
= bdrv_open(bs
, backing_file
->value
.s
, NULL
, back_flags
,
4832 error_setg_errno(errp
, -ret
, "Could not open '%s'",
4833 backing_file
->value
.s
);
4836 bdrv_get_geometry(bs
, &size
);
4839 snprintf(buf
, sizeof(buf
), "%" PRId64
, size
);
4840 set_option_parameter(param
, BLOCK_OPT_SIZE
, buf
);
4842 error_setg(errp
, "Image creation needs a size parameter");
4848 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
4849 print_option_parameters(param
);
4852 ret
= bdrv_create(drv
, filename
, param
);
4854 if (ret
== -ENOTSUP
) {
4855 error_setg(errp
,"Formatting or formatting option not supported for "
4856 "file format '%s'", fmt
);
4857 } else if (ret
== -EFBIG
) {
4858 error_setg(errp
, "The image size is too large for file format '%s'",
4861 error_setg(errp
, "%s: error while creating %s: %s", filename
, fmt
,
4867 free_option_parameters(create_options
);
4868 free_option_parameters(param
);
4875 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
4877 /* Currently BlockDriverState always uses the main loop AioContext */
4878 return qemu_get_aio_context();