2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24 #include "config-host.h"
25 #include "qemu-common.h"
27 #include "monitor/monitor.h"
28 #include "block/block_int.h"
29 #include "block/blockjob.h"
30 #include "qemu/module.h"
31 #include "qapi/qmp/qjson.h"
32 #include "sysemu/sysemu.h"
33 #include "qemu/notify.h"
34 #include "block/coroutine.h"
35 #include "qmp-commands.h"
36 #include "qemu/timer.h"
39 #include <sys/types.h>
41 #include <sys/ioctl.h>
42 #include <sys/queue.h>
52 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
55 BDRV_REQ_COPY_ON_READ
= 0x1,
56 BDRV_REQ_ZERO_WRITE
= 0x2,
59 static void bdrv_dev_change_media_cb(BlockDriverState
*bs
, bool load
);
60 static BlockDriverAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
61 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
62 BlockDriverCompletionFunc
*cb
, void *opaque
);
63 static BlockDriverAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
64 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
65 BlockDriverCompletionFunc
*cb
, void *opaque
);
66 static int coroutine_fn
bdrv_co_readv_em(BlockDriverState
*bs
,
67 int64_t sector_num
, int nb_sectors
,
69 static int coroutine_fn
bdrv_co_writev_em(BlockDriverState
*bs
,
70 int64_t sector_num
, int nb_sectors
,
72 static int coroutine_fn
bdrv_co_do_readv(BlockDriverState
*bs
,
73 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
74 BdrvRequestFlags flags
);
75 static int coroutine_fn
bdrv_co_do_writev(BlockDriverState
*bs
,
76 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
77 BdrvRequestFlags flags
);
78 static BlockDriverAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
82 BlockDriverCompletionFunc
*cb
,
85 static void coroutine_fn
bdrv_co_do_rw(void *opaque
);
86 static int coroutine_fn
bdrv_co_do_write_zeroes(BlockDriverState
*bs
,
87 int64_t sector_num
, int nb_sectors
);
89 static bool bdrv_exceed_bps_limits(BlockDriverState
*bs
, int nb_sectors
,
90 bool is_write
, double elapsed_time
, uint64_t *wait
);
91 static bool bdrv_exceed_iops_limits(BlockDriverState
*bs
, bool is_write
,
92 double elapsed_time
, uint64_t *wait
);
93 static bool bdrv_exceed_io_limits(BlockDriverState
*bs
, int nb_sectors
,
94 bool is_write
, int64_t *wait
);
96 static QTAILQ_HEAD(, BlockDriverState
) bdrv_states
=
97 QTAILQ_HEAD_INITIALIZER(bdrv_states
);
99 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
100 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
102 /* The device to use for VM snapshots */
103 static BlockDriverState
*bs_snapshots
;
105 /* If non-zero, use only whitelisted block drivers */
106 static int use_bdrv_whitelist
;
109 static int is_windows_drive_prefix(const char *filename
)
111 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
112 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
116 int is_windows_drive(const char *filename
)
118 if (is_windows_drive_prefix(filename
) &&
121 if (strstart(filename
, "\\\\.\\", NULL
) ||
122 strstart(filename
, "//./", NULL
))
128 /* throttling disk I/O limits */
129 void bdrv_io_limits_disable(BlockDriverState
*bs
)
131 bs
->io_limits_enabled
= false;
133 while (qemu_co_queue_next(&bs
->throttled_reqs
));
135 if (bs
->block_timer
) {
136 qemu_del_timer(bs
->block_timer
);
137 qemu_free_timer(bs
->block_timer
);
138 bs
->block_timer
= NULL
;
145 static void bdrv_block_timer(void *opaque
)
147 BlockDriverState
*bs
= opaque
;
149 qemu_co_queue_next(&bs
->throttled_reqs
);
152 void bdrv_io_limits_enable(BlockDriverState
*bs
)
154 qemu_co_queue_init(&bs
->throttled_reqs
);
155 bs
->block_timer
= qemu_new_timer_ns(vm_clock
, bdrv_block_timer
, bs
);
156 bs
->io_limits_enabled
= true;
159 bool bdrv_io_limits_enabled(BlockDriverState
*bs
)
161 BlockIOLimit
*io_limits
= &bs
->io_limits
;
162 return io_limits
->bps
[BLOCK_IO_LIMIT_READ
]
163 || io_limits
->bps
[BLOCK_IO_LIMIT_WRITE
]
164 || io_limits
->bps
[BLOCK_IO_LIMIT_TOTAL
]
165 || io_limits
->iops
[BLOCK_IO_LIMIT_READ
]
166 || io_limits
->iops
[BLOCK_IO_LIMIT_WRITE
]
167 || io_limits
->iops
[BLOCK_IO_LIMIT_TOTAL
];
170 static void bdrv_io_limits_intercept(BlockDriverState
*bs
,
171 bool is_write
, int nb_sectors
)
173 int64_t wait_time
= -1;
175 if (!qemu_co_queue_empty(&bs
->throttled_reqs
)) {
176 qemu_co_queue_wait(&bs
->throttled_reqs
);
179 /* In fact, we hope to keep each request's timing, in FIFO mode. The next
180 * throttled requests will not be dequeued until the current request is
181 * allowed to be serviced. So if the current request still exceeds the
182 * limits, it will be inserted to the head. All requests followed it will
183 * be still in throttled_reqs queue.
186 while (bdrv_exceed_io_limits(bs
, nb_sectors
, is_write
, &wait_time
)) {
187 qemu_mod_timer(bs
->block_timer
,
188 wait_time
+ qemu_get_clock_ns(vm_clock
));
189 qemu_co_queue_wait_insert_head(&bs
->throttled_reqs
);
192 qemu_co_queue_next(&bs
->throttled_reqs
);
195 /* check if the path starts with "<protocol>:" */
196 static int path_has_protocol(const char *path
)
201 if (is_windows_drive(path
) ||
202 is_windows_drive_prefix(path
)) {
205 p
= path
+ strcspn(path
, ":/\\");
207 p
= path
+ strcspn(path
, ":/");
213 int path_is_absolute(const char *path
)
216 /* specific case for names like: "\\.\d:" */
217 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
220 return (*path
== '/' || *path
== '\\');
222 return (*path
== '/');
226 /* if filename is absolute, just copy it to dest. Otherwise, build a
227 path to it by considering it is relative to base_path. URL are
229 void path_combine(char *dest
, int dest_size
,
230 const char *base_path
,
231 const char *filename
)
238 if (path_is_absolute(filename
)) {
239 pstrcpy(dest
, dest_size
, filename
);
241 p
= strchr(base_path
, ':');
246 p1
= strrchr(base_path
, '/');
250 p2
= strrchr(base_path
, '\\');
262 if (len
> dest_size
- 1)
264 memcpy(dest
, base_path
, len
);
266 pstrcat(dest
, dest_size
, filename
);
270 void bdrv_get_full_backing_filename(BlockDriverState
*bs
, char *dest
, size_t sz
)
272 if (bs
->backing_file
[0] == '\0' || path_has_protocol(bs
->backing_file
)) {
273 pstrcpy(dest
, sz
, bs
->backing_file
);
275 path_combine(dest
, sz
, bs
->filename
, bs
->backing_file
);
279 void bdrv_register(BlockDriver
*bdrv
)
281 /* Block drivers without coroutine functions need emulation */
282 if (!bdrv
->bdrv_co_readv
) {
283 bdrv
->bdrv_co_readv
= bdrv_co_readv_em
;
284 bdrv
->bdrv_co_writev
= bdrv_co_writev_em
;
286 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
287 * the block driver lacks aio we need to emulate that too.
289 if (!bdrv
->bdrv_aio_readv
) {
290 /* add AIO emulation layer */
291 bdrv
->bdrv_aio_readv
= bdrv_aio_readv_em
;
292 bdrv
->bdrv_aio_writev
= bdrv_aio_writev_em
;
296 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
299 /* create a new block device (by default it is empty) */
300 BlockDriverState
*bdrv_new(const char *device_name
)
302 BlockDriverState
*bs
;
304 bs
= g_malloc0(sizeof(BlockDriverState
));
305 pstrcpy(bs
->device_name
, sizeof(bs
->device_name
), device_name
);
306 if (device_name
[0] != '\0') {
307 QTAILQ_INSERT_TAIL(&bdrv_states
, bs
, list
);
309 bdrv_iostatus_disable(bs
);
310 notifier_list_init(&bs
->close_notifiers
);
315 void bdrv_add_close_notifier(BlockDriverState
*bs
, Notifier
*notify
)
317 notifier_list_add(&bs
->close_notifiers
, notify
);
320 BlockDriver
*bdrv_find_format(const char *format_name
)
323 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
324 if (!strcmp(drv1
->format_name
, format_name
)) {
331 static int bdrv_is_whitelisted(BlockDriver
*drv
)
333 static const char *whitelist
[] = {
334 CONFIG_BDRV_WHITELIST
339 return 1; /* no whitelist, anything goes */
341 for (p
= whitelist
; *p
; p
++) {
342 if (!strcmp(drv
->format_name
, *p
)) {
349 BlockDriver
*bdrv_find_whitelisted_format(const char *format_name
)
351 BlockDriver
*drv
= bdrv_find_format(format_name
);
352 return drv
&& bdrv_is_whitelisted(drv
) ? drv
: NULL
;
355 typedef struct CreateCo
{
358 QEMUOptionParameter
*options
;
362 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
364 CreateCo
*cco
= opaque
;
367 cco
->ret
= cco
->drv
->bdrv_create(cco
->filename
, cco
->options
);
370 int bdrv_create(BlockDriver
*drv
, const char* filename
,
371 QEMUOptionParameter
*options
)
378 .filename
= g_strdup(filename
),
383 if (!drv
->bdrv_create
) {
388 if (qemu_in_coroutine()) {
389 /* Fast-path if already in coroutine context */
390 bdrv_create_co_entry(&cco
);
392 co
= qemu_coroutine_create(bdrv_create_co_entry
);
393 qemu_coroutine_enter(co
, &cco
);
394 while (cco
.ret
== NOT_DONE
) {
402 g_free(cco
.filename
);
406 int bdrv_create_file(const char* filename
, QEMUOptionParameter
*options
)
410 drv
= bdrv_find_protocol(filename
);
415 return bdrv_create(drv
, filename
, options
);
419 * Create a uniquely-named empty temporary file.
420 * Return 0 upon success, otherwise a negative errno value.
422 int get_tmp_filename(char *filename
, int size
)
425 char temp_dir
[MAX_PATH
];
426 /* GetTempFileName requires that its output buffer (4th param)
427 have length MAX_PATH or greater. */
428 assert(size
>= MAX_PATH
);
429 return (GetTempPath(MAX_PATH
, temp_dir
)
430 && GetTempFileName(temp_dir
, "qem", 0, filename
)
431 ? 0 : -GetLastError());
435 tmpdir
= getenv("TMPDIR");
438 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
441 fd
= mkstemp(filename
);
445 if (close(fd
) != 0) {
454 * Detect host devices. By convention, /dev/cdrom[N] is always
455 * recognized as a host CDROM.
457 static BlockDriver
*find_hdev_driver(const char *filename
)
459 int score_max
= 0, score
;
460 BlockDriver
*drv
= NULL
, *d
;
462 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
463 if (d
->bdrv_probe_device
) {
464 score
= d
->bdrv_probe_device(filename
);
465 if (score
> score_max
) {
475 BlockDriver
*bdrv_find_protocol(const char *filename
)
482 /* TODO Drivers without bdrv_file_open must be specified explicitly */
485 * XXX(hch): we really should not let host device detection
486 * override an explicit protocol specification, but moving this
487 * later breaks access to device names with colons in them.
488 * Thanks to the brain-dead persistent naming schemes on udev-
489 * based Linux systems those actually are quite common.
491 drv1
= find_hdev_driver(filename
);
496 if (!path_has_protocol(filename
)) {
497 return bdrv_find_format("file");
499 p
= strchr(filename
, ':');
502 if (len
> sizeof(protocol
) - 1)
503 len
= sizeof(protocol
) - 1;
504 memcpy(protocol
, filename
, len
);
505 protocol
[len
] = '\0';
506 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
507 if (drv1
->protocol_name
&&
508 !strcmp(drv1
->protocol_name
, protocol
)) {
515 static int find_image_format(BlockDriverState
*bs
, const char *filename
,
518 int score
, score_max
;
519 BlockDriver
*drv1
, *drv
;
523 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
524 if (bs
->sg
|| !bdrv_is_inserted(bs
) || bdrv_getlength(bs
) == 0) {
525 drv
= bdrv_find_format("raw");
533 ret
= bdrv_pread(bs
, 0, buf
, sizeof(buf
));
541 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
542 if (drv1
->bdrv_probe
) {
543 score
= drv1
->bdrv_probe(buf
, ret
, filename
);
544 if (score
> score_max
) {
558 * Set the current 'total_sectors' value
560 static int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
562 BlockDriver
*drv
= bs
->drv
;
564 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
568 /* query actual device if possible, otherwise just trust the hint */
569 if (drv
->bdrv_getlength
) {
570 int64_t length
= drv
->bdrv_getlength(bs
);
574 hint
= length
>> BDRV_SECTOR_BITS
;
577 bs
->total_sectors
= hint
;
582 * Set open flags for a given discard mode
584 * Return 0 on success, -1 if the discard mode was invalid.
586 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
588 *flags
&= ~BDRV_O_UNMAP
;
590 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
592 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
593 *flags
|= BDRV_O_UNMAP
;
602 * Set open flags for a given cache mode
604 * Return 0 on success, -1 if the cache mode was invalid.
606 int bdrv_parse_cache_flags(const char *mode
, int *flags
)
608 *flags
&= ~BDRV_O_CACHE_MASK
;
610 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
611 *flags
|= BDRV_O_NOCACHE
| BDRV_O_CACHE_WB
;
612 } else if (!strcmp(mode
, "directsync")) {
613 *flags
|= BDRV_O_NOCACHE
;
614 } else if (!strcmp(mode
, "writeback")) {
615 *flags
|= BDRV_O_CACHE_WB
;
616 } else if (!strcmp(mode
, "unsafe")) {
617 *flags
|= BDRV_O_CACHE_WB
;
618 *flags
|= BDRV_O_NO_FLUSH
;
619 } else if (!strcmp(mode
, "writethrough")) {
620 /* this is the default */
629 * The copy-on-read flag is actually a reference count so multiple users may
630 * use the feature without worrying about clobbering its previous state.
631 * Copy-on-read stays enabled until all users have called to disable it.
633 void bdrv_enable_copy_on_read(BlockDriverState
*bs
)
638 void bdrv_disable_copy_on_read(BlockDriverState
*bs
)
640 assert(bs
->copy_on_read
> 0);
644 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
646 int open_flags
= flags
| BDRV_O_CACHE_WB
;
649 * Clear flags that are internal to the block layer before opening the
652 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
655 * Snapshots should be writable.
657 if (bs
->is_temporary
) {
658 open_flags
|= BDRV_O_RDWR
;
665 * Common part for opening disk images and files
667 * Removes all processed options from *options.
669 static int bdrv_open_common(BlockDriverState
*bs
, BlockDriverState
*file
,
670 const char *filename
, QDict
*options
,
671 int flags
, BlockDriver
*drv
)
676 assert(bs
->file
== NULL
);
677 assert(options
!= NULL
&& bs
->options
!= options
);
679 trace_bdrv_open_common(bs
, filename
, flags
, drv
->format_name
);
681 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
)) {
685 /* bdrv_open() with directly using a protocol as drv. This layer is already
686 * opened, so assign it to bs (while file becomes a closed BlockDriverState)
687 * and return immediately. */
688 if (file
!= NULL
&& drv
->bdrv_file_open
) {
693 bs
->open_flags
= flags
;
694 bs
->buffer_alignment
= 512;
696 assert(bs
->copy_on_read
== 0); /* bdrv_new() and bdrv_close() make it so */
697 if ((flags
& BDRV_O_RDWR
) && (flags
& BDRV_O_COPY_ON_READ
)) {
698 bdrv_enable_copy_on_read(bs
);
701 if (filename
!= NULL
) {
702 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
704 bs
->filename
[0] = '\0';
708 bs
->opaque
= g_malloc0(drv
->instance_size
);
710 bs
->enable_write_cache
= !!(flags
& BDRV_O_CACHE_WB
);
711 open_flags
= bdrv_open_flags(bs
, flags
);
713 bs
->read_only
= !(open_flags
& BDRV_O_RDWR
);
715 /* Open the image, either directly or using a protocol */
716 if (drv
->bdrv_file_open
) {
717 assert(file
== NULL
);
718 assert(drv
->bdrv_parse_filename
|| filename
!= NULL
);
719 ret
= drv
->bdrv_file_open(bs
, filename
, options
, open_flags
);
722 qerror_report(ERROR_CLASS_GENERIC_ERROR
, "Can't use '%s' as a "
723 "block driver for the protocol level",
728 assert(file
!= NULL
);
730 ret
= drv
->bdrv_open(bs
, options
, open_flags
);
737 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
743 if (bs
->is_temporary
) {
744 assert(filename
!= NULL
);
759 * Opens a file using a protocol (file, host_device, nbd, ...)
761 * options is a QDict of options to pass to the block drivers, or NULL for an
762 * empty set of options. The reference to the QDict belongs to the block layer
763 * after the call (even on failure), so if the caller intends to reuse the
764 * dictionary, it needs to use QINCREF() before calling bdrv_file_open.
766 int bdrv_file_open(BlockDriverState
**pbs
, const char *filename
,
767 QDict
*options
, int flags
)
769 BlockDriverState
*bs
;
774 /* NULL means an empty set of options */
775 if (options
== NULL
) {
776 options
= qdict_new();
780 bs
->options
= options
;
781 options
= qdict_clone_shallow(options
);
783 /* Find the right block driver */
784 drvname
= qdict_get_try_str(options
, "driver");
786 drv
= bdrv_find_whitelisted_format(drvname
);
787 qdict_del(options
, "driver");
788 } else if (filename
) {
789 drv
= bdrv_find_protocol(filename
);
791 qerror_report(ERROR_CLASS_GENERIC_ERROR
,
792 "Must specify either driver or file");
801 /* Parse the filename and open it */
802 if (drv
->bdrv_parse_filename
&& filename
) {
803 Error
*local_err
= NULL
;
804 drv
->bdrv_parse_filename(filename
, options
, &local_err
);
805 if (error_is_set(&local_err
)) {
806 qerror_report_err(local_err
);
807 error_free(local_err
);
811 } else if (!drv
->bdrv_parse_filename
&& !filename
) {
812 qerror_report(ERROR_CLASS_GENERIC_ERROR
,
813 "The '%s' block driver requires a file name",
819 ret
= bdrv_open_common(bs
, NULL
, filename
, options
, flags
, drv
);
824 /* Check if any unknown options were used */
825 if (qdict_size(options
) != 0) {
826 const QDictEntry
*entry
= qdict_first(options
);
827 qerror_report(ERROR_CLASS_GENERIC_ERROR
, "Block protocol '%s' doesn't "
828 "support the option '%s'",
829 drv
->format_name
, entry
->key
);
842 QDECREF(bs
->options
);
849 * Opens the backing file for a BlockDriverState if not yet open
851 * options is a QDict of options to pass to the block drivers, or NULL for an
852 * empty set of options. The reference to the QDict is transferred to this
853 * function (even on failure), so if the caller intends to reuse the dictionary,
854 * it needs to use QINCREF() before calling bdrv_file_open.
856 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*options
)
858 char backing_filename
[PATH_MAX
];
860 BlockDriver
*back_drv
= NULL
;
862 if (bs
->backing_hd
!= NULL
) {
867 /* NULL means an empty set of options */
868 if (options
== NULL
) {
869 options
= qdict_new();
872 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
873 if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
878 bs
->backing_hd
= bdrv_new("");
879 bdrv_get_full_backing_filename(bs
, backing_filename
,
880 sizeof(backing_filename
));
882 if (bs
->backing_format
[0] != '\0') {
883 back_drv
= bdrv_find_format(bs
->backing_format
);
886 /* backing files always opened read-only */
887 back_flags
= bs
->open_flags
& ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
);
889 ret
= bdrv_open(bs
->backing_hd
,
890 *backing_filename
? backing_filename
: NULL
, options
,
891 back_flags
, back_drv
);
893 bdrv_delete(bs
->backing_hd
);
894 bs
->backing_hd
= NULL
;
895 bs
->open_flags
|= BDRV_O_NO_BACKING
;
901 static void extract_subqdict(QDict
*src
, QDict
**dst
, const char *start
)
903 const QDictEntry
*entry
, *next
;
907 entry
= qdict_first(src
);
909 while (entry
!= NULL
) {
910 next
= qdict_next(src
, entry
);
911 if (strstart(entry
->key
, start
, &p
)) {
912 qobject_incref(entry
->value
);
913 qdict_put_obj(*dst
, p
, entry
->value
);
914 qdict_del(src
, entry
->key
);
921 * Opens a disk image (raw, qcow2, vmdk, ...)
923 * options is a QDict of options to pass to the block drivers, or NULL for an
924 * empty set of options. The reference to the QDict belongs to the block layer
925 * after the call (even on failure), so if the caller intends to reuse the
926 * dictionary, it needs to use QINCREF() before calling bdrv_open.
928 int bdrv_open(BlockDriverState
*bs
, const char *filename
, QDict
*options
,
929 int flags
, BlockDriver
*drv
)
932 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
933 char tmp_filename
[PATH_MAX
+ 1];
934 BlockDriverState
*file
= NULL
;
935 QDict
*file_options
= NULL
;
937 /* NULL means an empty set of options */
938 if (options
== NULL
) {
939 options
= qdict_new();
942 bs
->options
= options
;
943 options
= qdict_clone_shallow(options
);
945 /* For snapshot=on, create a temporary qcow2 overlay */
946 if (flags
& BDRV_O_SNAPSHOT
) {
947 BlockDriverState
*bs1
;
949 BlockDriver
*bdrv_qcow2
;
950 QEMUOptionParameter
*create_options
;
951 char backing_filename
[PATH_MAX
];
953 if (qdict_size(options
) != 0) {
954 error_report("Can't use snapshot=on with driver-specific options");
958 assert(filename
!= NULL
);
960 /* if snapshot, we create a temporary backing file and open it
961 instead of opening 'filename' directly */
963 /* if there is a backing file, use it */
965 ret
= bdrv_open(bs1
, filename
, NULL
, 0, drv
);
970 total_size
= bdrv_getlength(bs1
) & BDRV_SECTOR_MASK
;
974 ret
= get_tmp_filename(tmp_filename
, sizeof(tmp_filename
));
979 /* Real path is meaningless for protocols */
980 if (path_has_protocol(filename
)) {
981 snprintf(backing_filename
, sizeof(backing_filename
),
983 } else if (!realpath(filename
, backing_filename
)) {
988 bdrv_qcow2
= bdrv_find_format("qcow2");
989 create_options
= parse_option_parameters("", bdrv_qcow2
->create_options
,
992 set_option_parameter_int(create_options
, BLOCK_OPT_SIZE
, total_size
);
993 set_option_parameter(create_options
, BLOCK_OPT_BACKING_FILE
,
996 set_option_parameter(create_options
, BLOCK_OPT_BACKING_FMT
,
1000 ret
= bdrv_create(bdrv_qcow2
, tmp_filename
, create_options
);
1001 free_option_parameters(create_options
);
1006 filename
= tmp_filename
;
1008 bs
->is_temporary
= 1;
1011 /* Open image file without format layer */
1012 if (flags
& BDRV_O_RDWR
) {
1013 flags
|= BDRV_O_ALLOW_RDWR
;
1016 extract_subqdict(options
, &file_options
, "file.");
1018 ret
= bdrv_file_open(&file
, filename
, file_options
,
1019 bdrv_open_flags(bs
, flags
));
1024 /* Find the right image format driver */
1026 ret
= find_image_format(file
, filename
, &drv
);
1030 goto unlink_and_fail
;
1033 /* Open the image */
1034 ret
= bdrv_open_common(bs
, file
, filename
, options
, flags
, drv
);
1036 goto unlink_and_fail
;
1039 if (bs
->file
!= file
) {
1044 /* If there is a backing file, use it */
1045 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
1046 QDict
*backing_options
;
1048 extract_subqdict(options
, &backing_options
, "backing.");
1049 ret
= bdrv_open_backing_file(bs
, backing_options
);
1051 goto close_and_fail
;
1055 /* Check if any unknown options were used */
1056 if (qdict_size(options
) != 0) {
1057 const QDictEntry
*entry
= qdict_first(options
);
1058 qerror_report(ERROR_CLASS_GENERIC_ERROR
, "Block format '%s' used by "
1059 "device '%s' doesn't support the option '%s'",
1060 drv
->format_name
, bs
->device_name
, entry
->key
);
1063 goto close_and_fail
;
1067 if (!bdrv_key_required(bs
)) {
1068 bdrv_dev_change_media_cb(bs
, true);
1071 /* throttling disk I/O limits */
1072 if (bs
->io_limits_enabled
) {
1073 bdrv_io_limits_enable(bs
);
1082 if (bs
->is_temporary
) {
1086 QDECREF(bs
->options
);
1097 typedef struct BlockReopenQueueEntry
{
1099 BDRVReopenState state
;
1100 QSIMPLEQ_ENTRY(BlockReopenQueueEntry
) entry
;
1101 } BlockReopenQueueEntry
;
1104 * Adds a BlockDriverState to a simple queue for an atomic, transactional
1105 * reopen of multiple devices.
1107 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
1108 * already performed, or alternatively may be NULL a new BlockReopenQueue will
1109 * be created and initialized. This newly created BlockReopenQueue should be
1110 * passed back in for subsequent calls that are intended to be of the same
1113 * bs is the BlockDriverState to add to the reopen queue.
1115 * flags contains the open flags for the associated bs
1117 * returns a pointer to bs_queue, which is either the newly allocated
1118 * bs_queue, or the existing bs_queue being used.
1121 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
1122 BlockDriverState
*bs
, int flags
)
1126 BlockReopenQueueEntry
*bs_entry
;
1127 if (bs_queue
== NULL
) {
1128 bs_queue
= g_new0(BlockReopenQueue
, 1);
1129 QSIMPLEQ_INIT(bs_queue
);
1133 bdrv_reopen_queue(bs_queue
, bs
->file
, flags
);
1136 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
1137 QSIMPLEQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
1139 bs_entry
->state
.bs
= bs
;
1140 bs_entry
->state
.flags
= flags
;
1146 * Reopen multiple BlockDriverStates atomically & transactionally.
1148 * The queue passed in (bs_queue) must have been built up previous
1149 * via bdrv_reopen_queue().
1151 * Reopens all BDS specified in the queue, with the appropriate
1152 * flags. All devices are prepared for reopen, and failure of any
1153 * device will cause all device changes to be abandonded, and intermediate
1156 * If all devices prepare successfully, then the changes are committed
1160 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
1163 BlockReopenQueueEntry
*bs_entry
, *next
;
1164 Error
*local_err
= NULL
;
1166 assert(bs_queue
!= NULL
);
1170 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
1171 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, &local_err
)) {
1172 error_propagate(errp
, local_err
);
1175 bs_entry
->prepared
= true;
1178 /* If we reach this point, we have success and just need to apply the
1181 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
1182 bdrv_reopen_commit(&bs_entry
->state
);
1188 QSIMPLEQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
1189 if (ret
&& bs_entry
->prepared
) {
1190 bdrv_reopen_abort(&bs_entry
->state
);
1199 /* Reopen a single BlockDriverState with the specified flags. */
1200 int bdrv_reopen(BlockDriverState
*bs
, int bdrv_flags
, Error
**errp
)
1203 Error
*local_err
= NULL
;
1204 BlockReopenQueue
*queue
= bdrv_reopen_queue(NULL
, bs
, bdrv_flags
);
1206 ret
= bdrv_reopen_multiple(queue
, &local_err
);
1207 if (local_err
!= NULL
) {
1208 error_propagate(errp
, local_err
);
1215 * Prepares a BlockDriverState for reopen. All changes are staged in the
1216 * 'opaque' field of the BDRVReopenState, which is used and allocated by
1217 * the block driver layer .bdrv_reopen_prepare()
1219 * bs is the BlockDriverState to reopen
1220 * flags are the new open flags
1221 * queue is the reopen queue
1223 * Returns 0 on success, non-zero on error. On error errp will be set
1226 * On failure, bdrv_reopen_abort() will be called to clean up any data.
1227 * It is the responsibility of the caller to then call the abort() or
1228 * commit() for any other BDS that have been left in a prepare() state
1231 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
1235 Error
*local_err
= NULL
;
1238 assert(reopen_state
!= NULL
);
1239 assert(reopen_state
->bs
->drv
!= NULL
);
1240 drv
= reopen_state
->bs
->drv
;
1242 /* if we are to stay read-only, do not allow permission change
1244 if (!(reopen_state
->bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
1245 reopen_state
->flags
& BDRV_O_RDWR
) {
1246 error_set(errp
, QERR_DEVICE_IS_READ_ONLY
,
1247 reopen_state
->bs
->device_name
);
1252 ret
= bdrv_flush(reopen_state
->bs
);
1254 error_set(errp
, ERROR_CLASS_GENERIC_ERROR
, "Error (%s) flushing drive",
1259 if (drv
->bdrv_reopen_prepare
) {
1260 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
1262 if (local_err
!= NULL
) {
1263 error_propagate(errp
, local_err
);
1265 error_set(errp
, QERR_OPEN_FILE_FAILED
,
1266 reopen_state
->bs
->filename
);
1271 /* It is currently mandatory to have a bdrv_reopen_prepare()
1272 * handler for each supported drv. */
1273 error_set(errp
, QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED
,
1274 drv
->format_name
, reopen_state
->bs
->device_name
,
1275 "reopening of file");
1287 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
1288 * makes them final by swapping the staging BlockDriverState contents into
1289 * the active BlockDriverState contents.
1291 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
1295 assert(reopen_state
!= NULL
);
1296 drv
= reopen_state
->bs
->drv
;
1297 assert(drv
!= NULL
);
1299 /* If there are any driver level actions to take */
1300 if (drv
->bdrv_reopen_commit
) {
1301 drv
->bdrv_reopen_commit(reopen_state
);
1304 /* set BDS specific flags now */
1305 reopen_state
->bs
->open_flags
= reopen_state
->flags
;
1306 reopen_state
->bs
->enable_write_cache
= !!(reopen_state
->flags
&
1308 reopen_state
->bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
1312 * Abort the reopen, and delete and free the staged changes in
1315 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
1319 assert(reopen_state
!= NULL
);
1320 drv
= reopen_state
->bs
->drv
;
1321 assert(drv
!= NULL
);
1323 if (drv
->bdrv_reopen_abort
) {
1324 drv
->bdrv_reopen_abort(reopen_state
);
1329 void bdrv_close(BlockDriverState
*bs
)
1333 block_job_cancel_sync(bs
->job
);
1336 notifier_list_notify(&bs
->close_notifiers
, bs
);
1339 if (bs
== bs_snapshots
) {
1340 bs_snapshots
= NULL
;
1342 if (bs
->backing_hd
) {
1343 bdrv_delete(bs
->backing_hd
);
1344 bs
->backing_hd
= NULL
;
1346 bs
->drv
->bdrv_close(bs
);
1349 if (bs
->is_temporary
) {
1350 unlink(bs
->filename
);
1355 bs
->copy_on_read
= 0;
1356 bs
->backing_file
[0] = '\0';
1357 bs
->backing_format
[0] = '\0';
1358 bs
->total_sectors
= 0;
1363 QDECREF(bs
->options
);
1366 if (bs
->file
!= NULL
) {
1367 bdrv_delete(bs
->file
);
1372 bdrv_dev_change_media_cb(bs
, false);
1374 /*throttling disk I/O limits*/
1375 if (bs
->io_limits_enabled
) {
1376 bdrv_io_limits_disable(bs
);
1380 void bdrv_close_all(void)
1382 BlockDriverState
*bs
;
1384 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1390 * Wait for pending requests to complete across all BlockDriverStates
1392 * This function does not flush data to disk, use bdrv_flush_all() for that
1393 * after calling this function.
1395 * Note that completion of an asynchronous I/O operation can trigger any
1396 * number of other I/O operations on other devices---for example a coroutine
1397 * can be arbitrarily complex and a constant flow of I/O can come until the
1398 * coroutine is complete. Because of this, it is not possible to have a
1399 * function to drain a single device's I/O queue.
1401 void bdrv_drain_all(void)
1403 BlockDriverState
*bs
;
1407 busy
= qemu_aio_wait();
1409 /* FIXME: We do not have timer support here, so this is effectively
1412 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1413 if (!qemu_co_queue_empty(&bs
->throttled_reqs
)) {
1414 qemu_co_queue_restart_all(&bs
->throttled_reqs
);
1420 /* If requests are still pending there is a bug somewhere */
1421 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1422 assert(QLIST_EMPTY(&bs
->tracked_requests
));
1423 assert(qemu_co_queue_empty(&bs
->throttled_reqs
));
1427 /* make a BlockDriverState anonymous by removing from bdrv_state list.
1428 Also, NULL terminate the device_name to prevent double remove */
1429 void bdrv_make_anon(BlockDriverState
*bs
)
1431 if (bs
->device_name
[0] != '\0') {
1432 QTAILQ_REMOVE(&bdrv_states
, bs
, list
);
1434 bs
->device_name
[0] = '\0';
1437 static void bdrv_rebind(BlockDriverState
*bs
)
1439 if (bs
->drv
&& bs
->drv
->bdrv_rebind
) {
1440 bs
->drv
->bdrv_rebind(bs
);
1444 static void bdrv_move_feature_fields(BlockDriverState
*bs_dest
,
1445 BlockDriverState
*bs_src
)
1447 /* move some fields that need to stay attached to the device */
1448 bs_dest
->open_flags
= bs_src
->open_flags
;
1451 bs_dest
->dev_ops
= bs_src
->dev_ops
;
1452 bs_dest
->dev_opaque
= bs_src
->dev_opaque
;
1453 bs_dest
->dev
= bs_src
->dev
;
1454 bs_dest
->buffer_alignment
= bs_src
->buffer_alignment
;
1455 bs_dest
->copy_on_read
= bs_src
->copy_on_read
;
1457 bs_dest
->enable_write_cache
= bs_src
->enable_write_cache
;
1459 /* i/o timing parameters */
1460 bs_dest
->slice_start
= bs_src
->slice_start
;
1461 bs_dest
->slice_end
= bs_src
->slice_end
;
1462 bs_dest
->slice_submitted
= bs_src
->slice_submitted
;
1463 bs_dest
->io_limits
= bs_src
->io_limits
;
1464 bs_dest
->throttled_reqs
= bs_src
->throttled_reqs
;
1465 bs_dest
->block_timer
= bs_src
->block_timer
;
1466 bs_dest
->io_limits_enabled
= bs_src
->io_limits_enabled
;
1469 bs_dest
->on_read_error
= bs_src
->on_read_error
;
1470 bs_dest
->on_write_error
= bs_src
->on_write_error
;
1473 bs_dest
->iostatus_enabled
= bs_src
->iostatus_enabled
;
1474 bs_dest
->iostatus
= bs_src
->iostatus
;
1477 bs_dest
->dirty_bitmap
= bs_src
->dirty_bitmap
;
1480 bs_dest
->in_use
= bs_src
->in_use
;
1481 bs_dest
->job
= bs_src
->job
;
1483 /* keep the same entry in bdrv_states */
1484 pstrcpy(bs_dest
->device_name
, sizeof(bs_dest
->device_name
),
1485 bs_src
->device_name
);
1486 bs_dest
->list
= bs_src
->list
;
1490 * Swap bs contents for two image chains while they are live,
1491 * while keeping required fields on the BlockDriverState that is
1492 * actually attached to a device.
1494 * This will modify the BlockDriverState fields, and swap contents
1495 * between bs_new and bs_old. Both bs_new and bs_old are modified.
1497 * bs_new is required to be anonymous.
1499 * This function does not create any image files.
1501 void bdrv_swap(BlockDriverState
*bs_new
, BlockDriverState
*bs_old
)
1503 BlockDriverState tmp
;
1505 /* bs_new must be anonymous and shouldn't have anything fancy enabled */
1506 assert(bs_new
->device_name
[0] == '\0');
1507 assert(bs_new
->dirty_bitmap
== NULL
);
1508 assert(bs_new
->job
== NULL
);
1509 assert(bs_new
->dev
== NULL
);
1510 assert(bs_new
->in_use
== 0);
1511 assert(bs_new
->io_limits_enabled
== false);
1512 assert(bs_new
->block_timer
== NULL
);
1518 /* there are some fields that should not be swapped, move them back */
1519 bdrv_move_feature_fields(&tmp
, bs_old
);
1520 bdrv_move_feature_fields(bs_old
, bs_new
);
1521 bdrv_move_feature_fields(bs_new
, &tmp
);
1523 /* bs_new shouldn't be in bdrv_states even after the swap! */
1524 assert(bs_new
->device_name
[0] == '\0');
1526 /* Check a few fields that should remain attached to the device */
1527 assert(bs_new
->dev
== NULL
);
1528 assert(bs_new
->job
== NULL
);
1529 assert(bs_new
->in_use
== 0);
1530 assert(bs_new
->io_limits_enabled
== false);
1531 assert(bs_new
->block_timer
== NULL
);
1533 bdrv_rebind(bs_new
);
1534 bdrv_rebind(bs_old
);
1538 * Add new bs contents at the top of an image chain while the chain is
1539 * live, while keeping required fields on the top layer.
1541 * This will modify the BlockDriverState fields, and swap contents
1542 * between bs_new and bs_top. Both bs_new and bs_top are modified.
1544 * bs_new is required to be anonymous.
1546 * This function does not create any image files.
1548 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
)
1550 bdrv_swap(bs_new
, bs_top
);
1552 /* The contents of 'tmp' will become bs_top, as we are
1553 * swapping bs_new and bs_top contents. */
1554 bs_top
->backing_hd
= bs_new
;
1555 bs_top
->open_flags
&= ~BDRV_O_NO_BACKING
;
1556 pstrcpy(bs_top
->backing_file
, sizeof(bs_top
->backing_file
),
1558 pstrcpy(bs_top
->backing_format
, sizeof(bs_top
->backing_format
),
1559 bs_new
->drv
? bs_new
->drv
->format_name
: "");
1562 void bdrv_delete(BlockDriverState
*bs
)
1566 assert(!bs
->in_use
);
1568 /* remove from list, if necessary */
1573 assert(bs
!= bs_snapshots
);
1577 int bdrv_attach_dev(BlockDriverState
*bs
, void *dev
)
1578 /* TODO change to DeviceState *dev when all users are qdevified */
1584 bdrv_iostatus_reset(bs
);
1588 /* TODO qdevified devices don't use this, remove when devices are qdevified */
1589 void bdrv_attach_dev_nofail(BlockDriverState
*bs
, void *dev
)
1591 if (bdrv_attach_dev(bs
, dev
) < 0) {
1596 void bdrv_detach_dev(BlockDriverState
*bs
, void *dev
)
1597 /* TODO change to DeviceState *dev when all users are qdevified */
1599 assert(bs
->dev
== dev
);
1602 bs
->dev_opaque
= NULL
;
1603 bs
->buffer_alignment
= 512;
1606 /* TODO change to return DeviceState * when all users are qdevified */
1607 void *bdrv_get_attached_dev(BlockDriverState
*bs
)
1612 void bdrv_set_dev_ops(BlockDriverState
*bs
, const BlockDevOps
*ops
,
1616 bs
->dev_opaque
= opaque
;
1617 if (bdrv_dev_has_removable_media(bs
) && bs
== bs_snapshots
) {
1618 bs_snapshots
= NULL
;
1622 void bdrv_emit_qmp_error_event(const BlockDriverState
*bdrv
,
1623 enum MonitorEvent ev
,
1624 BlockErrorAction action
, bool is_read
)
1627 const char *action_str
;
1630 case BDRV_ACTION_REPORT
:
1631 action_str
= "report";
1633 case BDRV_ACTION_IGNORE
:
1634 action_str
= "ignore";
1636 case BDRV_ACTION_STOP
:
1637 action_str
= "stop";
1643 data
= qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1646 is_read
? "read" : "write");
1647 monitor_protocol_event(ev
, data
);
1649 qobject_decref(data
);
1652 static void bdrv_emit_qmp_eject_event(BlockDriverState
*bs
, bool ejected
)
1656 data
= qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }",
1657 bdrv_get_device_name(bs
), ejected
);
1658 monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED
, data
);
1660 qobject_decref(data
);
1663 static void bdrv_dev_change_media_cb(BlockDriverState
*bs
, bool load
)
1665 if (bs
->dev_ops
&& bs
->dev_ops
->change_media_cb
) {
1666 bool tray_was_closed
= !bdrv_dev_is_tray_open(bs
);
1667 bs
->dev_ops
->change_media_cb(bs
->dev_opaque
, load
);
1668 if (tray_was_closed
) {
1670 bdrv_emit_qmp_eject_event(bs
, true);
1674 bdrv_emit_qmp_eject_event(bs
, false);
1679 bool bdrv_dev_has_removable_media(BlockDriverState
*bs
)
1681 return !bs
->dev
|| (bs
->dev_ops
&& bs
->dev_ops
->change_media_cb
);
1684 void bdrv_dev_eject_request(BlockDriverState
*bs
, bool force
)
1686 if (bs
->dev_ops
&& bs
->dev_ops
->eject_request_cb
) {
1687 bs
->dev_ops
->eject_request_cb(bs
->dev_opaque
, force
);
1691 bool bdrv_dev_is_tray_open(BlockDriverState
*bs
)
1693 if (bs
->dev_ops
&& bs
->dev_ops
->is_tray_open
) {
1694 return bs
->dev_ops
->is_tray_open(bs
->dev_opaque
);
1699 static void bdrv_dev_resize_cb(BlockDriverState
*bs
)
1701 if (bs
->dev_ops
&& bs
->dev_ops
->resize_cb
) {
1702 bs
->dev_ops
->resize_cb(bs
->dev_opaque
);
1706 bool bdrv_dev_is_medium_locked(BlockDriverState
*bs
)
1708 if (bs
->dev_ops
&& bs
->dev_ops
->is_medium_locked
) {
1709 return bs
->dev_ops
->is_medium_locked(bs
->dev_opaque
);
1715 * Run consistency checks on an image
1717 * Returns 0 if the check could be completed (it doesn't mean that the image is
1718 * free of errors) or -errno when an internal error occurred. The results of the
1719 * check are stored in res.
1721 int bdrv_check(BlockDriverState
*bs
, BdrvCheckResult
*res
, BdrvCheckMode fix
)
1723 if (bs
->drv
->bdrv_check
== NULL
) {
1727 memset(res
, 0, sizeof(*res
));
1728 return bs
->drv
->bdrv_check(bs
, res
, fix
);
1731 #define COMMIT_BUF_SECTORS 2048
1733 /* commit COW file into the raw image */
1734 int bdrv_commit(BlockDriverState
*bs
)
1736 BlockDriver
*drv
= bs
->drv
;
1737 int64_t sector
, total_sectors
;
1738 int n
, ro
, open_flags
;
1741 char filename
[PATH_MAX
];
1746 if (!bs
->backing_hd
) {
1750 if (bdrv_in_use(bs
) || bdrv_in_use(bs
->backing_hd
)) {
1754 ro
= bs
->backing_hd
->read_only
;
1755 /* Use pstrcpy (not strncpy): filename must be NUL-terminated. */
1756 pstrcpy(filename
, sizeof(filename
), bs
->backing_hd
->filename
);
1757 open_flags
= bs
->backing_hd
->open_flags
;
1760 if (bdrv_reopen(bs
->backing_hd
, open_flags
| BDRV_O_RDWR
, NULL
)) {
1765 total_sectors
= bdrv_getlength(bs
) >> BDRV_SECTOR_BITS
;
1766 buf
= g_malloc(COMMIT_BUF_SECTORS
* BDRV_SECTOR_SIZE
);
1768 for (sector
= 0; sector
< total_sectors
; sector
+= n
) {
1769 if (bdrv_is_allocated(bs
, sector
, COMMIT_BUF_SECTORS
, &n
)) {
1771 if (bdrv_read(bs
, sector
, buf
, n
) != 0) {
1776 if (bdrv_write(bs
->backing_hd
, sector
, buf
, n
) != 0) {
1783 if (drv
->bdrv_make_empty
) {
1784 ret
= drv
->bdrv_make_empty(bs
);
1789 * Make sure all data we wrote to the backing device is actually
1793 bdrv_flush(bs
->backing_hd
);
1799 /* ignoring error return here */
1800 bdrv_reopen(bs
->backing_hd
, open_flags
& ~BDRV_O_RDWR
, NULL
);
1806 int bdrv_commit_all(void)
1808 BlockDriverState
*bs
;
1810 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1811 if (bs
->drv
&& bs
->backing_hd
) {
1812 int ret
= bdrv_commit(bs
);
1821 struct BdrvTrackedRequest
{
1822 BlockDriverState
*bs
;
1826 QLIST_ENTRY(BdrvTrackedRequest
) list
;
1827 Coroutine
*co
; /* owner, used for deadlock detection */
1828 CoQueue wait_queue
; /* coroutines blocked on this request */
1832 * Remove an active request from the tracked requests list
1834 * This function should be called when a tracked request is completing.
1836 static void tracked_request_end(BdrvTrackedRequest
*req
)
1838 QLIST_REMOVE(req
, list
);
1839 qemu_co_queue_restart_all(&req
->wait_queue
);
1843 * Add an active request to the tracked requests list
1845 static void tracked_request_begin(BdrvTrackedRequest
*req
,
1846 BlockDriverState
*bs
,
1848 int nb_sectors
, bool is_write
)
1850 *req
= (BdrvTrackedRequest
){
1852 .sector_num
= sector_num
,
1853 .nb_sectors
= nb_sectors
,
1854 .is_write
= is_write
,
1855 .co
= qemu_coroutine_self(),
1858 qemu_co_queue_init(&req
->wait_queue
);
1860 QLIST_INSERT_HEAD(&bs
->tracked_requests
, req
, list
);
1864 * Round a region to cluster boundaries
1866 void bdrv_round_to_clusters(BlockDriverState
*bs
,
1867 int64_t sector_num
, int nb_sectors
,
1868 int64_t *cluster_sector_num
,
1869 int *cluster_nb_sectors
)
1871 BlockDriverInfo bdi
;
1873 if (bdrv_get_info(bs
, &bdi
) < 0 || bdi
.cluster_size
== 0) {
1874 *cluster_sector_num
= sector_num
;
1875 *cluster_nb_sectors
= nb_sectors
;
1877 int64_t c
= bdi
.cluster_size
/ BDRV_SECTOR_SIZE
;
1878 *cluster_sector_num
= QEMU_ALIGN_DOWN(sector_num
, c
);
1879 *cluster_nb_sectors
= QEMU_ALIGN_UP(sector_num
- *cluster_sector_num
+
1884 static bool tracked_request_overlaps(BdrvTrackedRequest
*req
,
1885 int64_t sector_num
, int nb_sectors
) {
1887 if (sector_num
>= req
->sector_num
+ req
->nb_sectors
) {
1891 if (req
->sector_num
>= sector_num
+ nb_sectors
) {
1897 static void coroutine_fn
wait_for_overlapping_requests(BlockDriverState
*bs
,
1898 int64_t sector_num
, int nb_sectors
)
1900 BdrvTrackedRequest
*req
;
1901 int64_t cluster_sector_num
;
1902 int cluster_nb_sectors
;
1905 /* If we touch the same cluster it counts as an overlap. This guarantees
1906 * that allocating writes will be serialized and not race with each other
1907 * for the same cluster. For example, in copy-on-read it ensures that the
1908 * CoR read and write operations are atomic and guest writes cannot
1909 * interleave between them.
1911 bdrv_round_to_clusters(bs
, sector_num
, nb_sectors
,
1912 &cluster_sector_num
, &cluster_nb_sectors
);
1916 QLIST_FOREACH(req
, &bs
->tracked_requests
, list
) {
1917 if (tracked_request_overlaps(req
, cluster_sector_num
,
1918 cluster_nb_sectors
)) {
1919 /* Hitting this means there was a reentrant request, for
1920 * example, a block driver issuing nested requests. This must
1921 * never happen since it means deadlock.
1923 assert(qemu_coroutine_self() != req
->co
);
1925 qemu_co_queue_wait(&req
->wait_queue
);
1936 * -EINVAL - backing format specified, but no file
1937 * -ENOSPC - can't update the backing file because no space is left in the
1939 * -ENOTSUP - format driver doesn't support changing the backing file
1941 int bdrv_change_backing_file(BlockDriverState
*bs
,
1942 const char *backing_file
, const char *backing_fmt
)
1944 BlockDriver
*drv
= bs
->drv
;
1947 /* Backing file format doesn't make sense without a backing file */
1948 if (backing_fmt
&& !backing_file
) {
1952 if (drv
->bdrv_change_backing_file
!= NULL
) {
1953 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
1959 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
1960 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
1966 * Finds the image layer in the chain that has 'bs' as its backing file.
1968 * active is the current topmost image.
1970 * Returns NULL if bs is not found in active's image chain,
1971 * or if active == bs.
1973 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
1974 BlockDriverState
*bs
)
1976 BlockDriverState
*overlay
= NULL
;
1977 BlockDriverState
*intermediate
;
1979 assert(active
!= NULL
);
1982 /* if bs is the same as active, then by definition it has no overlay
1988 intermediate
= active
;
1989 while (intermediate
->backing_hd
) {
1990 if (intermediate
->backing_hd
== bs
) {
1991 overlay
= intermediate
;
1994 intermediate
= intermediate
->backing_hd
;
2000 typedef struct BlkIntermediateStates
{
2001 BlockDriverState
*bs
;
2002 QSIMPLEQ_ENTRY(BlkIntermediateStates
) entry
;
2003 } BlkIntermediateStates
;
2007 * Drops images above 'base' up to and including 'top', and sets the image
2008 * above 'top' to have base as its backing file.
2010 * Requires that the overlay to 'top' is opened r/w, so that the backing file
2011 * information in 'bs' can be properly updated.
2013 * E.g., this will convert the following chain:
2014 * bottom <- base <- intermediate <- top <- active
2018 * bottom <- base <- active
2020 * It is allowed for bottom==base, in which case it converts:
2022 * base <- intermediate <- top <- active
2029 * if active == top, that is considered an error
2032 int bdrv_drop_intermediate(BlockDriverState
*active
, BlockDriverState
*top
,
2033 BlockDriverState
*base
)
2035 BlockDriverState
*intermediate
;
2036 BlockDriverState
*base_bs
= NULL
;
2037 BlockDriverState
*new_top_bs
= NULL
;
2038 BlkIntermediateStates
*intermediate_state
, *next
;
2041 QSIMPLEQ_HEAD(states_to_delete
, BlkIntermediateStates
) states_to_delete
;
2042 QSIMPLEQ_INIT(&states_to_delete
);
2044 if (!top
->drv
|| !base
->drv
) {
2048 new_top_bs
= bdrv_find_overlay(active
, top
);
2050 if (new_top_bs
== NULL
) {
2051 /* we could not find the image above 'top', this is an error */
2055 /* special case of new_top_bs->backing_hd already pointing to base - nothing
2056 * to do, no intermediate images */
2057 if (new_top_bs
->backing_hd
== base
) {
2064 /* now we will go down through the list, and add each BDS we find
2065 * into our deletion queue, until we hit the 'base'
2067 while (intermediate
) {
2068 intermediate_state
= g_malloc0(sizeof(BlkIntermediateStates
));
2069 intermediate_state
->bs
= intermediate
;
2070 QSIMPLEQ_INSERT_TAIL(&states_to_delete
, intermediate_state
, entry
);
2072 if (intermediate
->backing_hd
== base
) {
2073 base_bs
= intermediate
->backing_hd
;
2076 intermediate
= intermediate
->backing_hd
;
2078 if (base_bs
== NULL
) {
2079 /* something went wrong, we did not end at the base. safely
2080 * unravel everything, and exit with error */
2084 /* success - we can delete the intermediate states, and link top->base */
2085 ret
= bdrv_change_backing_file(new_top_bs
, base_bs
->filename
,
2086 base_bs
->drv
? base_bs
->drv
->format_name
: "");
2090 new_top_bs
->backing_hd
= base_bs
;
2093 QSIMPLEQ_FOREACH_SAFE(intermediate_state
, &states_to_delete
, entry
, next
) {
2094 /* so that bdrv_close() does not recursively close the chain */
2095 intermediate_state
->bs
->backing_hd
= NULL
;
2096 bdrv_delete(intermediate_state
->bs
);
2101 QSIMPLEQ_FOREACH_SAFE(intermediate_state
, &states_to_delete
, entry
, next
) {
2102 g_free(intermediate_state
);
2108 static int bdrv_check_byte_request(BlockDriverState
*bs
, int64_t offset
,
2113 if (!bdrv_is_inserted(bs
))
2119 len
= bdrv_getlength(bs
);
2124 if ((offset
> len
) || (len
- offset
< size
))
2130 static int bdrv_check_request(BlockDriverState
*bs
, int64_t sector_num
,
2133 return bdrv_check_byte_request(bs
, sector_num
* BDRV_SECTOR_SIZE
,
2134 nb_sectors
* BDRV_SECTOR_SIZE
);
2137 typedef struct RwCo
{
2138 BlockDriverState
*bs
;
2146 static void coroutine_fn
bdrv_rw_co_entry(void *opaque
)
2148 RwCo
*rwco
= opaque
;
2150 if (!rwco
->is_write
) {
2151 rwco
->ret
= bdrv_co_do_readv(rwco
->bs
, rwco
->sector_num
,
2152 rwco
->nb_sectors
, rwco
->qiov
, 0);
2154 rwco
->ret
= bdrv_co_do_writev(rwco
->bs
, rwco
->sector_num
,
2155 rwco
->nb_sectors
, rwco
->qiov
, 0);
2160 * Process a vectored synchronous request using coroutines
2162 static int bdrv_rwv_co(BlockDriverState
*bs
, int64_t sector_num
,
2163 QEMUIOVector
*qiov
, bool is_write
)
2168 .sector_num
= sector_num
,
2169 .nb_sectors
= qiov
->size
>> BDRV_SECTOR_BITS
,
2171 .is_write
= is_write
,
2174 assert((qiov
->size
& (BDRV_SECTOR_SIZE
- 1)) == 0);
2177 * In sync call context, when the vcpu is blocked, this throttling timer
2178 * will not fire; so the I/O throttling function has to be disabled here
2179 * if it has been enabled.
2181 if (bs
->io_limits_enabled
) {
2182 fprintf(stderr
, "Disabling I/O throttling on '%s' due "
2183 "to synchronous I/O.\n", bdrv_get_device_name(bs
));
2184 bdrv_io_limits_disable(bs
);
2187 if (qemu_in_coroutine()) {
2188 /* Fast-path if already in coroutine context */
2189 bdrv_rw_co_entry(&rwco
);
2191 co
= qemu_coroutine_create(bdrv_rw_co_entry
);
2192 qemu_coroutine_enter(co
, &rwco
);
2193 while (rwco
.ret
== NOT_DONE
) {
2201 * Process a synchronous request using coroutines
2203 static int bdrv_rw_co(BlockDriverState
*bs
, int64_t sector_num
, uint8_t *buf
,
2204 int nb_sectors
, bool is_write
)
2207 struct iovec iov
= {
2208 .iov_base
= (void *)buf
,
2209 .iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
,
2212 qemu_iovec_init_external(&qiov
, &iov
, 1);
2213 return bdrv_rwv_co(bs
, sector_num
, &qiov
, is_write
);
2216 /* return < 0 if error. See bdrv_write() for the return codes */
2217 int bdrv_read(BlockDriverState
*bs
, int64_t sector_num
,
2218 uint8_t *buf
, int nb_sectors
)
2220 return bdrv_rw_co(bs
, sector_num
, buf
, nb_sectors
, false);
2223 /* Just like bdrv_read(), but with I/O throttling temporarily disabled */
2224 int bdrv_read_unthrottled(BlockDriverState
*bs
, int64_t sector_num
,
2225 uint8_t *buf
, int nb_sectors
)
2230 enabled
= bs
->io_limits_enabled
;
2231 bs
->io_limits_enabled
= false;
2232 ret
= bdrv_read(bs
, 0, buf
, 1);
2233 bs
->io_limits_enabled
= enabled
;
2237 /* Return < 0 if error. Important errors are:
2238 -EIO generic I/O error (may happen for all errors)
2239 -ENOMEDIUM No media inserted.
2240 -EINVAL Invalid sector number or nb_sectors
2241 -EACCES Trying to write a read-only device
2243 int bdrv_write(BlockDriverState
*bs
, int64_t sector_num
,
2244 const uint8_t *buf
, int nb_sectors
)
2246 return bdrv_rw_co(bs
, sector_num
, (uint8_t *)buf
, nb_sectors
, true);
2249 int bdrv_writev(BlockDriverState
*bs
, int64_t sector_num
, QEMUIOVector
*qiov
)
2251 return bdrv_rwv_co(bs
, sector_num
, qiov
, true);
2254 int bdrv_pread(BlockDriverState
*bs
, int64_t offset
,
2255 void *buf
, int count1
)
2257 uint8_t tmp_buf
[BDRV_SECTOR_SIZE
];
2258 int len
, nb_sectors
, count
;
2263 /* first read to align to sector start */
2264 len
= (BDRV_SECTOR_SIZE
- offset
) & (BDRV_SECTOR_SIZE
- 1);
2267 sector_num
= offset
>> BDRV_SECTOR_BITS
;
2269 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
2271 memcpy(buf
, tmp_buf
+ (offset
& (BDRV_SECTOR_SIZE
- 1)), len
);
2279 /* read the sectors "in place" */
2280 nb_sectors
= count
>> BDRV_SECTOR_BITS
;
2281 if (nb_sectors
> 0) {
2282 if ((ret
= bdrv_read(bs
, sector_num
, buf
, nb_sectors
)) < 0)
2284 sector_num
+= nb_sectors
;
2285 len
= nb_sectors
<< BDRV_SECTOR_BITS
;
2290 /* add data from the last sector */
2292 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
2294 memcpy(buf
, tmp_buf
, count
);
2299 int bdrv_pwritev(BlockDriverState
*bs
, int64_t offset
, QEMUIOVector
*qiov
)
2301 uint8_t tmp_buf
[BDRV_SECTOR_SIZE
];
2302 int len
, nb_sectors
, count
;
2308 /* first write to align to sector start */
2309 len
= (BDRV_SECTOR_SIZE
- offset
) & (BDRV_SECTOR_SIZE
- 1);
2312 sector_num
= offset
>> BDRV_SECTOR_BITS
;
2314 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
2316 qemu_iovec_to_buf(qiov
, 0, tmp_buf
+ (offset
& (BDRV_SECTOR_SIZE
- 1)),
2318 if ((ret
= bdrv_write(bs
, sector_num
, tmp_buf
, 1)) < 0)
2326 /* write the sectors "in place" */
2327 nb_sectors
= count
>> BDRV_SECTOR_BITS
;
2328 if (nb_sectors
> 0) {
2329 QEMUIOVector qiov_inplace
;
2331 qemu_iovec_init(&qiov_inplace
, qiov
->niov
);
2332 qemu_iovec_concat(&qiov_inplace
, qiov
, len
,
2333 nb_sectors
<< BDRV_SECTOR_BITS
);
2334 ret
= bdrv_writev(bs
, sector_num
, &qiov_inplace
);
2335 qemu_iovec_destroy(&qiov_inplace
);
2340 sector_num
+= nb_sectors
;
2341 len
= nb_sectors
<< BDRV_SECTOR_BITS
;
2345 /* add data from the last sector */
2347 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
2349 qemu_iovec_to_buf(qiov
, qiov
->size
- count
, tmp_buf
, count
);
2350 if ((ret
= bdrv_write(bs
, sector_num
, tmp_buf
, 1)) < 0)
2356 int bdrv_pwrite(BlockDriverState
*bs
, int64_t offset
,
2357 const void *buf
, int count1
)
2360 struct iovec iov
= {
2361 .iov_base
= (void *) buf
,
2365 qemu_iovec_init_external(&qiov
, &iov
, 1);
2366 return bdrv_pwritev(bs
, offset
, &qiov
);
2370 * Writes to the file and ensures that no writes are reordered across this
2371 * request (acts as a barrier)
2373 * Returns 0 on success, -errno in error cases.
2375 int bdrv_pwrite_sync(BlockDriverState
*bs
, int64_t offset
,
2376 const void *buf
, int count
)
2380 ret
= bdrv_pwrite(bs
, offset
, buf
, count
);
2385 /* No flush needed for cache modes that already do it */
2386 if (bs
->enable_write_cache
) {
2393 static int coroutine_fn
bdrv_co_do_copy_on_readv(BlockDriverState
*bs
,
2394 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
2396 /* Perform I/O through a temporary buffer so that users who scribble over
2397 * their read buffer while the operation is in progress do not end up
2398 * modifying the image file. This is critical for zero-copy guest I/O
2399 * where anything might happen inside guest memory.
2401 void *bounce_buffer
;
2403 BlockDriver
*drv
= bs
->drv
;
2405 QEMUIOVector bounce_qiov
;
2406 int64_t cluster_sector_num
;
2407 int cluster_nb_sectors
;
2411 /* Cover entire cluster so no additional backing file I/O is required when
2412 * allocating cluster in the image file.
2414 bdrv_round_to_clusters(bs
, sector_num
, nb_sectors
,
2415 &cluster_sector_num
, &cluster_nb_sectors
);
2417 trace_bdrv_co_do_copy_on_readv(bs
, sector_num
, nb_sectors
,
2418 cluster_sector_num
, cluster_nb_sectors
);
2420 iov
.iov_len
= cluster_nb_sectors
* BDRV_SECTOR_SIZE
;
2421 iov
.iov_base
= bounce_buffer
= qemu_blockalign(bs
, iov
.iov_len
);
2422 qemu_iovec_init_external(&bounce_qiov
, &iov
, 1);
2424 ret
= drv
->bdrv_co_readv(bs
, cluster_sector_num
, cluster_nb_sectors
,
2430 if (drv
->bdrv_co_write_zeroes
&&
2431 buffer_is_zero(bounce_buffer
, iov
.iov_len
)) {
2432 ret
= bdrv_co_do_write_zeroes(bs
, cluster_sector_num
,
2433 cluster_nb_sectors
);
2435 /* This does not change the data on the disk, it is not necessary
2436 * to flush even in cache=writethrough mode.
2438 ret
= drv
->bdrv_co_writev(bs
, cluster_sector_num
, cluster_nb_sectors
,
2443 /* It might be okay to ignore write errors for guest requests. If this
2444 * is a deliberate copy-on-read then we don't want to ignore the error.
2445 * Simply report it in all cases.
2450 skip_bytes
= (sector_num
- cluster_sector_num
) * BDRV_SECTOR_SIZE
;
2451 qemu_iovec_from_buf(qiov
, 0, bounce_buffer
+ skip_bytes
,
2452 nb_sectors
* BDRV_SECTOR_SIZE
);
2455 qemu_vfree(bounce_buffer
);
2460 * Handle a read request in coroutine context
2462 static int coroutine_fn
bdrv_co_do_readv(BlockDriverState
*bs
,
2463 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
2464 BdrvRequestFlags flags
)
2466 BlockDriver
*drv
= bs
->drv
;
2467 BdrvTrackedRequest req
;
2473 if (bdrv_check_request(bs
, sector_num
, nb_sectors
)) {
2477 /* throttling disk read I/O */
2478 if (bs
->io_limits_enabled
) {
2479 bdrv_io_limits_intercept(bs
, false, nb_sectors
);
2482 if (bs
->copy_on_read
) {
2483 flags
|= BDRV_REQ_COPY_ON_READ
;
2485 if (flags
& BDRV_REQ_COPY_ON_READ
) {
2486 bs
->copy_on_read_in_flight
++;
2489 if (bs
->copy_on_read_in_flight
) {
2490 wait_for_overlapping_requests(bs
, sector_num
, nb_sectors
);
2493 tracked_request_begin(&req
, bs
, sector_num
, nb_sectors
, false);
2495 if (flags
& BDRV_REQ_COPY_ON_READ
) {
2498 ret
= bdrv_co_is_allocated(bs
, sector_num
, nb_sectors
, &pnum
);
2503 if (!ret
|| pnum
!= nb_sectors
) {
2504 ret
= bdrv_co_do_copy_on_readv(bs
, sector_num
, nb_sectors
, qiov
);
2509 ret
= drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
2512 tracked_request_end(&req
);
2514 if (flags
& BDRV_REQ_COPY_ON_READ
) {
2515 bs
->copy_on_read_in_flight
--;
2521 int coroutine_fn
bdrv_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
2522 int nb_sectors
, QEMUIOVector
*qiov
)
2524 trace_bdrv_co_readv(bs
, sector_num
, nb_sectors
);
2526 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
, 0);
2529 int coroutine_fn
bdrv_co_copy_on_readv(BlockDriverState
*bs
,
2530 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
2532 trace_bdrv_co_copy_on_readv(bs
, sector_num
, nb_sectors
);
2534 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
,
2535 BDRV_REQ_COPY_ON_READ
);
2538 static int coroutine_fn
bdrv_co_do_write_zeroes(BlockDriverState
*bs
,
2539 int64_t sector_num
, int nb_sectors
)
2541 BlockDriver
*drv
= bs
->drv
;
2546 /* TODO Emulate only part of misaligned requests instead of letting block
2547 * drivers return -ENOTSUP and emulate everything */
2549 /* First try the efficient write zeroes operation */
2550 if (drv
->bdrv_co_write_zeroes
) {
2551 ret
= drv
->bdrv_co_write_zeroes(bs
, sector_num
, nb_sectors
);
2552 if (ret
!= -ENOTSUP
) {
2557 /* Fall back to bounce buffer if write zeroes is unsupported */
2558 iov
.iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
;
2559 iov
.iov_base
= qemu_blockalign(bs
, iov
.iov_len
);
2560 memset(iov
.iov_base
, 0, iov
.iov_len
);
2561 qemu_iovec_init_external(&qiov
, &iov
, 1);
2563 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, &qiov
);
2565 qemu_vfree(iov
.iov_base
);
2570 * Handle a write request in coroutine context
2572 static int coroutine_fn
bdrv_co_do_writev(BlockDriverState
*bs
,
2573 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
2574 BdrvRequestFlags flags
)
2576 BlockDriver
*drv
= bs
->drv
;
2577 BdrvTrackedRequest req
;
2583 if (bs
->read_only
) {
2586 if (bdrv_check_request(bs
, sector_num
, nb_sectors
)) {
2590 /* throttling disk write I/O */
2591 if (bs
->io_limits_enabled
) {
2592 bdrv_io_limits_intercept(bs
, true, nb_sectors
);
2595 if (bs
->copy_on_read_in_flight
) {
2596 wait_for_overlapping_requests(bs
, sector_num
, nb_sectors
);
2599 tracked_request_begin(&req
, bs
, sector_num
, nb_sectors
, true);
2601 if (flags
& BDRV_REQ_ZERO_WRITE
) {
2602 ret
= bdrv_co_do_write_zeroes(bs
, sector_num
, nb_sectors
);
2604 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, qiov
);
2607 if (ret
== 0 && !bs
->enable_write_cache
) {
2608 ret
= bdrv_co_flush(bs
);
2611 if (bs
->dirty_bitmap
) {
2612 bdrv_set_dirty(bs
, sector_num
, nb_sectors
);
2615 if (bs
->wr_highest_sector
< sector_num
+ nb_sectors
- 1) {
2616 bs
->wr_highest_sector
= sector_num
+ nb_sectors
- 1;
2619 tracked_request_end(&req
);
2624 int coroutine_fn
bdrv_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
2625 int nb_sectors
, QEMUIOVector
*qiov
)
2627 trace_bdrv_co_writev(bs
, sector_num
, nb_sectors
);
2629 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, qiov
, 0);
2632 int coroutine_fn
bdrv_co_write_zeroes(BlockDriverState
*bs
,
2633 int64_t sector_num
, int nb_sectors
)
2635 trace_bdrv_co_write_zeroes(bs
, sector_num
, nb_sectors
);
2637 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, NULL
,
2638 BDRV_REQ_ZERO_WRITE
);
2642 * Truncate file to 'offset' bytes (needed only for file protocols)
2644 int bdrv_truncate(BlockDriverState
*bs
, int64_t offset
)
2646 BlockDriver
*drv
= bs
->drv
;
2650 if (!drv
->bdrv_truncate
)
2654 if (bdrv_in_use(bs
))
2656 ret
= drv
->bdrv_truncate(bs
, offset
);
2658 ret
= refresh_total_sectors(bs
, offset
>> BDRV_SECTOR_BITS
);
2659 bdrv_dev_resize_cb(bs
);
2665 * Length of a allocated file in bytes. Sparse files are counted by actual
2666 * allocated space. Return < 0 if error or unknown.
2668 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
2670 BlockDriver
*drv
= bs
->drv
;
2674 if (drv
->bdrv_get_allocated_file_size
) {
2675 return drv
->bdrv_get_allocated_file_size(bs
);
2678 return bdrv_get_allocated_file_size(bs
->file
);
2684 * Length of a file in bytes. Return < 0 if error or unknown.
2686 int64_t bdrv_getlength(BlockDriverState
*bs
)
2688 BlockDriver
*drv
= bs
->drv
;
2692 if (bs
->growable
|| bdrv_dev_has_removable_media(bs
)) {
2693 if (drv
->bdrv_getlength
) {
2694 return drv
->bdrv_getlength(bs
);
2697 return bs
->total_sectors
* BDRV_SECTOR_SIZE
;
2700 /* return 0 as number of sectors if no device present or error */
2701 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
2704 length
= bdrv_getlength(bs
);
2708 length
= length
>> BDRV_SECTOR_BITS
;
2709 *nb_sectors_ptr
= length
;
2712 /* throttling disk io limits */
2713 void bdrv_set_io_limits(BlockDriverState
*bs
,
2714 BlockIOLimit
*io_limits
)
2716 bs
->io_limits
= *io_limits
;
2717 bs
->io_limits_enabled
= bdrv_io_limits_enabled(bs
);
2720 void bdrv_set_on_error(BlockDriverState
*bs
, BlockdevOnError on_read_error
,
2721 BlockdevOnError on_write_error
)
2723 bs
->on_read_error
= on_read_error
;
2724 bs
->on_write_error
= on_write_error
;
2727 BlockdevOnError
bdrv_get_on_error(BlockDriverState
*bs
, bool is_read
)
2729 return is_read
? bs
->on_read_error
: bs
->on_write_error
;
2732 BlockErrorAction
bdrv_get_error_action(BlockDriverState
*bs
, bool is_read
, int error
)
2734 BlockdevOnError on_err
= is_read
? bs
->on_read_error
: bs
->on_write_error
;
2737 case BLOCKDEV_ON_ERROR_ENOSPC
:
2738 return (error
== ENOSPC
) ? BDRV_ACTION_STOP
: BDRV_ACTION_REPORT
;
2739 case BLOCKDEV_ON_ERROR_STOP
:
2740 return BDRV_ACTION_STOP
;
2741 case BLOCKDEV_ON_ERROR_REPORT
:
2742 return BDRV_ACTION_REPORT
;
2743 case BLOCKDEV_ON_ERROR_IGNORE
:
2744 return BDRV_ACTION_IGNORE
;
2750 /* This is done by device models because, while the block layer knows
2751 * about the error, it does not know whether an operation comes from
2752 * the device or the block layer (from a job, for example).
2754 void bdrv_error_action(BlockDriverState
*bs
, BlockErrorAction action
,
2755 bool is_read
, int error
)
2758 bdrv_emit_qmp_error_event(bs
, QEVENT_BLOCK_IO_ERROR
, action
, is_read
);
2759 if (action
== BDRV_ACTION_STOP
) {
2760 vm_stop(RUN_STATE_IO_ERROR
);
2761 bdrv_iostatus_set_err(bs
, error
);
2765 int bdrv_is_read_only(BlockDriverState
*bs
)
2767 return bs
->read_only
;
2770 int bdrv_is_sg(BlockDriverState
*bs
)
2775 int bdrv_enable_write_cache(BlockDriverState
*bs
)
2777 return bs
->enable_write_cache
;
2780 void bdrv_set_enable_write_cache(BlockDriverState
*bs
, bool wce
)
2782 bs
->enable_write_cache
= wce
;
2784 /* so a reopen() will preserve wce */
2786 bs
->open_flags
|= BDRV_O_CACHE_WB
;
2788 bs
->open_flags
&= ~BDRV_O_CACHE_WB
;
2792 int bdrv_is_encrypted(BlockDriverState
*bs
)
2794 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
)
2796 return bs
->encrypted
;
2799 int bdrv_key_required(BlockDriverState
*bs
)
2801 BlockDriverState
*backing_hd
= bs
->backing_hd
;
2803 if (backing_hd
&& backing_hd
->encrypted
&& !backing_hd
->valid_key
)
2805 return (bs
->encrypted
&& !bs
->valid_key
);
2808 int bdrv_set_key(BlockDriverState
*bs
, const char *key
)
2811 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
) {
2812 ret
= bdrv_set_key(bs
->backing_hd
, key
);
2818 if (!bs
->encrypted
) {
2820 } else if (!bs
->drv
|| !bs
->drv
->bdrv_set_key
) {
2823 ret
= bs
->drv
->bdrv_set_key(bs
, key
);
2826 } else if (!bs
->valid_key
) {
2828 /* call the change callback now, we skipped it on open */
2829 bdrv_dev_change_media_cb(bs
, true);
2834 const char *bdrv_get_format_name(BlockDriverState
*bs
)
2836 return bs
->drv
? bs
->drv
->format_name
: NULL
;
2839 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
2844 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
2845 it(opaque
, drv
->format_name
);
2849 BlockDriverState
*bdrv_find(const char *name
)
2851 BlockDriverState
*bs
;
2853 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2854 if (!strcmp(name
, bs
->device_name
)) {
2861 BlockDriverState
*bdrv_next(BlockDriverState
*bs
)
2864 return QTAILQ_FIRST(&bdrv_states
);
2866 return QTAILQ_NEXT(bs
, list
);
2869 void bdrv_iterate(void (*it
)(void *opaque
, BlockDriverState
*bs
), void *opaque
)
2871 BlockDriverState
*bs
;
2873 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2878 const char *bdrv_get_device_name(BlockDriverState
*bs
)
2880 return bs
->device_name
;
2883 int bdrv_get_flags(BlockDriverState
*bs
)
2885 return bs
->open_flags
;
2888 void bdrv_flush_all(void)
2890 BlockDriverState
*bs
;
2892 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2897 int bdrv_has_zero_init(BlockDriverState
*bs
)
2901 if (bs
->drv
->bdrv_has_zero_init
) {
2902 return bs
->drv
->bdrv_has_zero_init(bs
);
2908 typedef struct BdrvCoIsAllocatedData
{
2909 BlockDriverState
*bs
;
2910 BlockDriverState
*base
;
2916 } BdrvCoIsAllocatedData
;
2919 * Returns true iff the specified sector is present in the disk image. Drivers
2920 * not implementing the functionality are assumed to not support backing files,
2921 * hence all their sectors are reported as allocated.
2923 * If 'sector_num' is beyond the end of the disk image the return value is 0
2924 * and 'pnum' is set to 0.
2926 * 'pnum' is set to the number of sectors (including and immediately following
2927 * the specified sector) that are known to be in the same
2928 * allocated/unallocated state.
2930 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
2931 * beyond the end of the disk image it will be clamped.
2933 int coroutine_fn
bdrv_co_is_allocated(BlockDriverState
*bs
, int64_t sector_num
,
2934 int nb_sectors
, int *pnum
)
2938 if (sector_num
>= bs
->total_sectors
) {
2943 n
= bs
->total_sectors
- sector_num
;
2944 if (n
< nb_sectors
) {
2948 if (!bs
->drv
->bdrv_co_is_allocated
) {
2953 return bs
->drv
->bdrv_co_is_allocated(bs
, sector_num
, nb_sectors
, pnum
);
2956 /* Coroutine wrapper for bdrv_is_allocated() */
2957 static void coroutine_fn
bdrv_is_allocated_co_entry(void *opaque
)
2959 BdrvCoIsAllocatedData
*data
= opaque
;
2960 BlockDriverState
*bs
= data
->bs
;
2962 data
->ret
= bdrv_co_is_allocated(bs
, data
->sector_num
, data
->nb_sectors
,
2968 * Synchronous wrapper around bdrv_co_is_allocated().
2970 * See bdrv_co_is_allocated() for details.
2972 int bdrv_is_allocated(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
,
2976 BdrvCoIsAllocatedData data
= {
2978 .sector_num
= sector_num
,
2979 .nb_sectors
= nb_sectors
,
2984 co
= qemu_coroutine_create(bdrv_is_allocated_co_entry
);
2985 qemu_coroutine_enter(co
, &data
);
2986 while (!data
.done
) {
2993 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
2995 * Return true if the given sector is allocated in any image between
2996 * BASE and TOP (inclusive). BASE can be NULL to check if the given
2997 * sector is allocated in any image of the chain. Return false otherwise.
2999 * 'pnum' is set to the number of sectors (including and immediately following
3000 * the specified sector) that are known to be in the same
3001 * allocated/unallocated state.
3004 int coroutine_fn
bdrv_co_is_allocated_above(BlockDriverState
*top
,
3005 BlockDriverState
*base
,
3007 int nb_sectors
, int *pnum
)
3009 BlockDriverState
*intermediate
;
3010 int ret
, n
= nb_sectors
;
3013 while (intermediate
&& intermediate
!= base
) {
3015 ret
= bdrv_co_is_allocated(intermediate
, sector_num
, nb_sectors
,
3025 * [sector_num, nb_sectors] is unallocated on top but intermediate
3028 * [sector_num+x, nr_sectors] allocated.
3030 if (n
> pnum_inter
&&
3031 (intermediate
== top
||
3032 sector_num
+ pnum_inter
< intermediate
->total_sectors
)) {
3036 intermediate
= intermediate
->backing_hd
;
3043 /* Coroutine wrapper for bdrv_is_allocated_above() */
3044 static void coroutine_fn
bdrv_is_allocated_above_co_entry(void *opaque
)
3046 BdrvCoIsAllocatedData
*data
= opaque
;
3047 BlockDriverState
*top
= data
->bs
;
3048 BlockDriverState
*base
= data
->base
;
3050 data
->ret
= bdrv_co_is_allocated_above(top
, base
, data
->sector_num
,
3051 data
->nb_sectors
, data
->pnum
);
3056 * Synchronous wrapper around bdrv_co_is_allocated_above().
3058 * See bdrv_co_is_allocated_above() for details.
3060 int bdrv_is_allocated_above(BlockDriverState
*top
, BlockDriverState
*base
,
3061 int64_t sector_num
, int nb_sectors
, int *pnum
)
3064 BdrvCoIsAllocatedData data
= {
3067 .sector_num
= sector_num
,
3068 .nb_sectors
= nb_sectors
,
3073 co
= qemu_coroutine_create(bdrv_is_allocated_above_co_entry
);
3074 qemu_coroutine_enter(co
, &data
);
3075 while (!data
.done
) {
3081 BlockInfo
*bdrv_query_info(BlockDriverState
*bs
)
3083 BlockInfo
*info
= g_malloc0(sizeof(*info
));
3084 info
->device
= g_strdup(bs
->device_name
);
3085 info
->type
= g_strdup("unknown");
3086 info
->locked
= bdrv_dev_is_medium_locked(bs
);
3087 info
->removable
= bdrv_dev_has_removable_media(bs
);
3089 if (bdrv_dev_has_removable_media(bs
)) {
3090 info
->has_tray_open
= true;
3091 info
->tray_open
= bdrv_dev_is_tray_open(bs
);
3094 if (bdrv_iostatus_is_enabled(bs
)) {
3095 info
->has_io_status
= true;
3096 info
->io_status
= bs
->iostatus
;
3099 if (bs
->dirty_bitmap
) {
3100 info
->has_dirty
= true;
3101 info
->dirty
= g_malloc0(sizeof(*info
->dirty
));
3102 info
->dirty
->count
= bdrv_get_dirty_count(bs
) * BDRV_SECTOR_SIZE
;
3103 info
->dirty
->granularity
=
3104 ((int64_t) BDRV_SECTOR_SIZE
<< hbitmap_granularity(bs
->dirty_bitmap
));
3108 info
->has_inserted
= true;
3109 info
->inserted
= g_malloc0(sizeof(*info
->inserted
));
3110 info
->inserted
->file
= g_strdup(bs
->filename
);
3111 info
->inserted
->ro
= bs
->read_only
;
3112 info
->inserted
->drv
= g_strdup(bs
->drv
->format_name
);
3113 info
->inserted
->encrypted
= bs
->encrypted
;
3114 info
->inserted
->encryption_key_missing
= bdrv_key_required(bs
);
3116 if (bs
->backing_file
[0]) {
3117 info
->inserted
->has_backing_file
= true;
3118 info
->inserted
->backing_file
= g_strdup(bs
->backing_file
);
3121 info
->inserted
->backing_file_depth
= bdrv_get_backing_file_depth(bs
);
3123 if (bs
->io_limits_enabled
) {
3124 info
->inserted
->bps
=
3125 bs
->io_limits
.bps
[BLOCK_IO_LIMIT_TOTAL
];
3126 info
->inserted
->bps_rd
=
3127 bs
->io_limits
.bps
[BLOCK_IO_LIMIT_READ
];
3128 info
->inserted
->bps_wr
=
3129 bs
->io_limits
.bps
[BLOCK_IO_LIMIT_WRITE
];
3130 info
->inserted
->iops
=
3131 bs
->io_limits
.iops
[BLOCK_IO_LIMIT_TOTAL
];
3132 info
->inserted
->iops_rd
=
3133 bs
->io_limits
.iops
[BLOCK_IO_LIMIT_READ
];
3134 info
->inserted
->iops_wr
=
3135 bs
->io_limits
.iops
[BLOCK_IO_LIMIT_WRITE
];
3141 BlockInfoList
*qmp_query_block(Error
**errp
)
3143 BlockInfoList
*head
= NULL
, **p_next
= &head
;
3144 BlockDriverState
*bs
;
3146 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
3147 BlockInfoList
*info
= g_malloc0(sizeof(*info
));
3148 info
->value
= bdrv_query_info(bs
);
3151 p_next
= &info
->next
;
3157 BlockStats
*bdrv_query_stats(const BlockDriverState
*bs
)
3161 s
= g_malloc0(sizeof(*s
));
3163 if (bs
->device_name
[0]) {
3164 s
->has_device
= true;
3165 s
->device
= g_strdup(bs
->device_name
);
3168 s
->stats
= g_malloc0(sizeof(*s
->stats
));
3169 s
->stats
->rd_bytes
= bs
->nr_bytes
[BDRV_ACCT_READ
];
3170 s
->stats
->wr_bytes
= bs
->nr_bytes
[BDRV_ACCT_WRITE
];
3171 s
->stats
->rd_operations
= bs
->nr_ops
[BDRV_ACCT_READ
];
3172 s
->stats
->wr_operations
= bs
->nr_ops
[BDRV_ACCT_WRITE
];
3173 s
->stats
->wr_highest_offset
= bs
->wr_highest_sector
* BDRV_SECTOR_SIZE
;
3174 s
->stats
->flush_operations
= bs
->nr_ops
[BDRV_ACCT_FLUSH
];
3175 s
->stats
->wr_total_time_ns
= bs
->total_time_ns
[BDRV_ACCT_WRITE
];
3176 s
->stats
->rd_total_time_ns
= bs
->total_time_ns
[BDRV_ACCT_READ
];
3177 s
->stats
->flush_total_time_ns
= bs
->total_time_ns
[BDRV_ACCT_FLUSH
];
3180 s
->has_parent
= true;
3181 s
->parent
= bdrv_query_stats(bs
->file
);
3187 BlockStatsList
*qmp_query_blockstats(Error
**errp
)
3189 BlockStatsList
*head
= NULL
, **p_next
= &head
;
3190 BlockDriverState
*bs
;
3192 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
3193 BlockStatsList
*info
= g_malloc0(sizeof(*info
));
3194 info
->value
= bdrv_query_stats(bs
);
3197 p_next
= &info
->next
;
3203 const char *bdrv_get_encrypted_filename(BlockDriverState
*bs
)
3205 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
)
3206 return bs
->backing_file
;
3207 else if (bs
->encrypted
)
3208 return bs
->filename
;
3213 void bdrv_get_backing_filename(BlockDriverState
*bs
,
3214 char *filename
, int filename_size
)
3216 pstrcpy(filename
, filename_size
, bs
->backing_file
);
3219 int bdrv_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
3220 const uint8_t *buf
, int nb_sectors
)
3222 BlockDriver
*drv
= bs
->drv
;
3225 if (!drv
->bdrv_write_compressed
)
3227 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
3230 assert(!bs
->dirty_bitmap
);
3232 return drv
->bdrv_write_compressed(bs
, sector_num
, buf
, nb_sectors
);
3235 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
3237 BlockDriver
*drv
= bs
->drv
;
3240 if (!drv
->bdrv_get_info
)
3242 memset(bdi
, 0, sizeof(*bdi
));
3243 return drv
->bdrv_get_info(bs
, bdi
);
3246 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
3247 int64_t pos
, int size
)
3250 struct iovec iov
= {
3251 .iov_base
= (void *) buf
,
3255 qemu_iovec_init_external(&qiov
, &iov
, 1);
3256 return bdrv_writev_vmstate(bs
, &qiov
, pos
);
3259 int bdrv_writev_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
3261 BlockDriver
*drv
= bs
->drv
;
3265 } else if (drv
->bdrv_save_vmstate
) {
3266 return drv
->bdrv_save_vmstate(bs
, qiov
, pos
);
3267 } else if (bs
->file
) {
3268 return bdrv_writev_vmstate(bs
->file
, qiov
, pos
);
3274 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
3275 int64_t pos
, int size
)
3277 BlockDriver
*drv
= bs
->drv
;
3280 if (drv
->bdrv_load_vmstate
)
3281 return drv
->bdrv_load_vmstate(bs
, buf
, pos
, size
);
3283 return bdrv_load_vmstate(bs
->file
, buf
, pos
, size
);
3287 void bdrv_debug_event(BlockDriverState
*bs
, BlkDebugEvent event
)
3289 BlockDriver
*drv
= bs
->drv
;
3291 if (!drv
|| !drv
->bdrv_debug_event
) {
3295 drv
->bdrv_debug_event(bs
, event
);
3298 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
3301 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
3305 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
3306 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
3312 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
3314 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_resume
) {
3318 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
3319 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
3325 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
3327 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
3331 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
3332 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
3338 /**************************************************************/
3339 /* handling of snapshots */
3341 int bdrv_can_snapshot(BlockDriverState
*bs
)
3343 BlockDriver
*drv
= bs
->drv
;
3344 if (!drv
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
)) {
3348 if (!drv
->bdrv_snapshot_create
) {
3349 if (bs
->file
!= NULL
) {
3350 return bdrv_can_snapshot(bs
->file
);
3358 int bdrv_is_snapshot(BlockDriverState
*bs
)
3360 return !!(bs
->open_flags
& BDRV_O_SNAPSHOT
);
3363 BlockDriverState
*bdrv_snapshots(void)
3365 BlockDriverState
*bs
;
3368 return bs_snapshots
;
3372 while ((bs
= bdrv_next(bs
))) {
3373 if (bdrv_can_snapshot(bs
)) {
3381 int bdrv_snapshot_create(BlockDriverState
*bs
,
3382 QEMUSnapshotInfo
*sn_info
)
3384 BlockDriver
*drv
= bs
->drv
;
3387 if (drv
->bdrv_snapshot_create
)
3388 return drv
->bdrv_snapshot_create(bs
, sn_info
);
3390 return bdrv_snapshot_create(bs
->file
, sn_info
);
3394 int bdrv_snapshot_goto(BlockDriverState
*bs
,
3395 const char *snapshot_id
)
3397 BlockDriver
*drv
= bs
->drv
;
3402 if (drv
->bdrv_snapshot_goto
)
3403 return drv
->bdrv_snapshot_goto(bs
, snapshot_id
);
3406 drv
->bdrv_close(bs
);
3407 ret
= bdrv_snapshot_goto(bs
->file
, snapshot_id
);
3408 open_ret
= drv
->bdrv_open(bs
, NULL
, bs
->open_flags
);
3410 bdrv_delete(bs
->file
);
3420 int bdrv_snapshot_delete(BlockDriverState
*bs
, const char *snapshot_id
)
3422 BlockDriver
*drv
= bs
->drv
;
3425 if (drv
->bdrv_snapshot_delete
)
3426 return drv
->bdrv_snapshot_delete(bs
, snapshot_id
);
3428 return bdrv_snapshot_delete(bs
->file
, snapshot_id
);
3432 int bdrv_snapshot_list(BlockDriverState
*bs
,
3433 QEMUSnapshotInfo
**psn_info
)
3435 BlockDriver
*drv
= bs
->drv
;
3438 if (drv
->bdrv_snapshot_list
)
3439 return drv
->bdrv_snapshot_list(bs
, psn_info
);
3441 return bdrv_snapshot_list(bs
->file
, psn_info
);
3445 int bdrv_snapshot_load_tmp(BlockDriverState
*bs
,
3446 const char *snapshot_name
)
3448 BlockDriver
*drv
= bs
->drv
;
3452 if (!bs
->read_only
) {
3455 if (drv
->bdrv_snapshot_load_tmp
) {
3456 return drv
->bdrv_snapshot_load_tmp(bs
, snapshot_name
);
3461 /* backing_file can either be relative, or absolute, or a protocol. If it is
3462 * relative, it must be relative to the chain. So, passing in bs->filename
3463 * from a BDS as backing_file should not be done, as that may be relative to
3464 * the CWD rather than the chain. */
3465 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
3466 const char *backing_file
)
3468 char *filename_full
= NULL
;
3469 char *backing_file_full
= NULL
;
3470 char *filename_tmp
= NULL
;
3471 int is_protocol
= 0;
3472 BlockDriverState
*curr_bs
= NULL
;
3473 BlockDriverState
*retval
= NULL
;
3475 if (!bs
|| !bs
->drv
|| !backing_file
) {
3479 filename_full
= g_malloc(PATH_MAX
);
3480 backing_file_full
= g_malloc(PATH_MAX
);
3481 filename_tmp
= g_malloc(PATH_MAX
);
3483 is_protocol
= path_has_protocol(backing_file
);
3485 for (curr_bs
= bs
; curr_bs
->backing_hd
; curr_bs
= curr_bs
->backing_hd
) {
3487 /* If either of the filename paths is actually a protocol, then
3488 * compare unmodified paths; otherwise make paths relative */
3489 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
3490 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
3491 retval
= curr_bs
->backing_hd
;
3495 /* If not an absolute filename path, make it relative to the current
3496 * image's filename path */
3497 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
3500 /* We are going to compare absolute pathnames */
3501 if (!realpath(filename_tmp
, filename_full
)) {
3505 /* We need to make sure the backing filename we are comparing against
3506 * is relative to the current image filename (or absolute) */
3507 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
3508 curr_bs
->backing_file
);
3510 if (!realpath(filename_tmp
, backing_file_full
)) {
3514 if (strcmp(backing_file_full
, filename_full
) == 0) {
3515 retval
= curr_bs
->backing_hd
;
3521 g_free(filename_full
);
3522 g_free(backing_file_full
);
3523 g_free(filename_tmp
);
3527 int bdrv_get_backing_file_depth(BlockDriverState
*bs
)
3533 if (!bs
->backing_hd
) {
3537 return 1 + bdrv_get_backing_file_depth(bs
->backing_hd
);
3540 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
3542 BlockDriverState
*curr_bs
= NULL
;
3550 while (curr_bs
->backing_hd
) {
3551 curr_bs
= curr_bs
->backing_hd
;
3556 #define NB_SUFFIXES 4
3558 char *get_human_readable_size(char *buf
, int buf_size
, int64_t size
)
3560 static const char suffixes
[NB_SUFFIXES
] = "KMGT";
3565 snprintf(buf
, buf_size
, "%" PRId64
, size
);
3568 for(i
= 0; i
< NB_SUFFIXES
; i
++) {
3569 if (size
< (10 * base
)) {
3570 snprintf(buf
, buf_size
, "%0.1f%c",
3571 (double)size
/ base
,
3574 } else if (size
< (1000 * base
) || i
== (NB_SUFFIXES
- 1)) {
3575 snprintf(buf
, buf_size
, "%" PRId64
"%c",
3576 ((size
+ (base
>> 1)) / base
),
3586 char *bdrv_snapshot_dump(char *buf
, int buf_size
, QEMUSnapshotInfo
*sn
)
3588 char buf1
[128], date_buf
[128], clock_buf
[128];
3594 snprintf(buf
, buf_size
,
3595 "%-10s%-20s%7s%20s%15s",
3596 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
3599 localtime_r(&ti
, &tm
);
3600 strftime(date_buf
, sizeof(date_buf
),
3601 "%Y-%m-%d %H:%M:%S", &tm
);
3602 secs
= sn
->vm_clock_nsec
/ 1000000000;
3603 snprintf(clock_buf
, sizeof(clock_buf
),
3604 "%02d:%02d:%02d.%03d",
3606 (int)((secs
/ 60) % 60),
3608 (int)((sn
->vm_clock_nsec
/ 1000000) % 1000));
3609 snprintf(buf
, buf_size
,
3610 "%-10s%-20s%7s%20s%15s",
3611 sn
->id_str
, sn
->name
,
3612 get_human_readable_size(buf1
, sizeof(buf1
), sn
->vm_state_size
),
3619 /**************************************************************/
3622 BlockDriverAIOCB
*bdrv_aio_readv(BlockDriverState
*bs
, int64_t sector_num
,
3623 QEMUIOVector
*qiov
, int nb_sectors
,
3624 BlockDriverCompletionFunc
*cb
, void *opaque
)
3626 trace_bdrv_aio_readv(bs
, sector_num
, nb_sectors
, opaque
);
3628 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
,
3632 BlockDriverAIOCB
*bdrv_aio_writev(BlockDriverState
*bs
, int64_t sector_num
,
3633 QEMUIOVector
*qiov
, int nb_sectors
,
3634 BlockDriverCompletionFunc
*cb
, void *opaque
)
3636 trace_bdrv_aio_writev(bs
, sector_num
, nb_sectors
, opaque
);
3638 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
,
3643 typedef struct MultiwriteCB
{
3648 BlockDriverCompletionFunc
*cb
;
3650 QEMUIOVector
*free_qiov
;
3654 static void multiwrite_user_cb(MultiwriteCB
*mcb
)
3658 for (i
= 0; i
< mcb
->num_callbacks
; i
++) {
3659 mcb
->callbacks
[i
].cb(mcb
->callbacks
[i
].opaque
, mcb
->error
);
3660 if (mcb
->callbacks
[i
].free_qiov
) {
3661 qemu_iovec_destroy(mcb
->callbacks
[i
].free_qiov
);
3663 g_free(mcb
->callbacks
[i
].free_qiov
);
3667 static void multiwrite_cb(void *opaque
, int ret
)
3669 MultiwriteCB
*mcb
= opaque
;
3671 trace_multiwrite_cb(mcb
, ret
);
3673 if (ret
< 0 && !mcb
->error
) {
3677 mcb
->num_requests
--;
3678 if (mcb
->num_requests
== 0) {
3679 multiwrite_user_cb(mcb
);
3684 static int multiwrite_req_compare(const void *a
, const void *b
)
3686 const BlockRequest
*req1
= a
, *req2
= b
;
3689 * Note that we can't simply subtract req2->sector from req1->sector
3690 * here as that could overflow the return value.
3692 if (req1
->sector
> req2
->sector
) {
3694 } else if (req1
->sector
< req2
->sector
) {
3702 * Takes a bunch of requests and tries to merge them. Returns the number of
3703 * requests that remain after merging.
3705 static int multiwrite_merge(BlockDriverState
*bs
, BlockRequest
*reqs
,
3706 int num_reqs
, MultiwriteCB
*mcb
)
3710 // Sort requests by start sector
3711 qsort(reqs
, num_reqs
, sizeof(*reqs
), &multiwrite_req_compare
);
3713 // Check if adjacent requests touch the same clusters. If so, combine them,
3714 // filling up gaps with zero sectors.
3716 for (i
= 1; i
< num_reqs
; i
++) {
3718 int64_t oldreq_last
= reqs
[outidx
].sector
+ reqs
[outidx
].nb_sectors
;
3720 // Handle exactly sequential writes and overlapping writes.
3721 if (reqs
[i
].sector
<= oldreq_last
) {
3725 if (reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1 > IOV_MAX
) {
3731 QEMUIOVector
*qiov
= g_malloc0(sizeof(*qiov
));
3732 qemu_iovec_init(qiov
,
3733 reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1);
3735 // Add the first request to the merged one. If the requests are
3736 // overlapping, drop the last sectors of the first request.
3737 size
= (reqs
[i
].sector
- reqs
[outidx
].sector
) << 9;
3738 qemu_iovec_concat(qiov
, reqs
[outidx
].qiov
, 0, size
);
3740 // We should need to add any zeros between the two requests
3741 assert (reqs
[i
].sector
<= oldreq_last
);
3743 // Add the second request
3744 qemu_iovec_concat(qiov
, reqs
[i
].qiov
, 0, reqs
[i
].qiov
->size
);
3746 reqs
[outidx
].nb_sectors
= qiov
->size
>> 9;
3747 reqs
[outidx
].qiov
= qiov
;
3749 mcb
->callbacks
[i
].free_qiov
= reqs
[outidx
].qiov
;
3752 reqs
[outidx
].sector
= reqs
[i
].sector
;
3753 reqs
[outidx
].nb_sectors
= reqs
[i
].nb_sectors
;
3754 reqs
[outidx
].qiov
= reqs
[i
].qiov
;
3762 * Submit multiple AIO write requests at once.
3764 * On success, the function returns 0 and all requests in the reqs array have
3765 * been submitted. In error case this function returns -1, and any of the
3766 * requests may or may not be submitted yet. In particular, this means that the
3767 * callback will be called for some of the requests, for others it won't. The
3768 * caller must check the error field of the BlockRequest to wait for the right
3769 * callbacks (if error != 0, no callback will be called).
3771 * The implementation may modify the contents of the reqs array, e.g. to merge
3772 * requests. However, the fields opaque and error are left unmodified as they
3773 * are used to signal failure for a single request to the caller.
3775 int bdrv_aio_multiwrite(BlockDriverState
*bs
, BlockRequest
*reqs
, int num_reqs
)
3780 /* don't submit writes if we don't have a medium */
3781 if (bs
->drv
== NULL
) {
3782 for (i
= 0; i
< num_reqs
; i
++) {
3783 reqs
[i
].error
= -ENOMEDIUM
;
3788 if (num_reqs
== 0) {
3792 // Create MultiwriteCB structure
3793 mcb
= g_malloc0(sizeof(*mcb
) + num_reqs
* sizeof(*mcb
->callbacks
));
3794 mcb
->num_requests
= 0;
3795 mcb
->num_callbacks
= num_reqs
;
3797 for (i
= 0; i
< num_reqs
; i
++) {
3798 mcb
->callbacks
[i
].cb
= reqs
[i
].cb
;
3799 mcb
->callbacks
[i
].opaque
= reqs
[i
].opaque
;
3802 // Check for mergable requests
3803 num_reqs
= multiwrite_merge(bs
, reqs
, num_reqs
, mcb
);
3805 trace_bdrv_aio_multiwrite(mcb
, mcb
->num_callbacks
, num_reqs
);
3807 /* Run the aio requests. */
3808 mcb
->num_requests
= num_reqs
;
3809 for (i
= 0; i
< num_reqs
; i
++) {
3810 bdrv_aio_writev(bs
, reqs
[i
].sector
, reqs
[i
].qiov
,
3811 reqs
[i
].nb_sectors
, multiwrite_cb
, mcb
);
3817 void bdrv_aio_cancel(BlockDriverAIOCB
*acb
)
3819 acb
->aiocb_info
->cancel(acb
);
3822 /* block I/O throttling */
3823 static bool bdrv_exceed_bps_limits(BlockDriverState
*bs
, int nb_sectors
,
3824 bool is_write
, double elapsed_time
, uint64_t *wait
)
3826 uint64_t bps_limit
= 0;
3828 double bytes_limit
, bytes_base
, bytes_res
;
3829 double slice_time
, wait_time
;
3831 if (bs
->io_limits
.bps
[BLOCK_IO_LIMIT_TOTAL
]) {
3832 bps_limit
= bs
->io_limits
.bps
[BLOCK_IO_LIMIT_TOTAL
];
3833 } else if (bs
->io_limits
.bps
[is_write
]) {
3834 bps_limit
= bs
->io_limits
.bps
[is_write
];
3843 slice_time
= bs
->slice_end
- bs
->slice_start
;
3844 slice_time
/= (NANOSECONDS_PER_SECOND
);
3845 bytes_limit
= bps_limit
* slice_time
;
3846 bytes_base
= bs
->slice_submitted
.bytes
[is_write
];
3847 if (bs
->io_limits
.bps
[BLOCK_IO_LIMIT_TOTAL
]) {
3848 bytes_base
+= bs
->slice_submitted
.bytes
[!is_write
];
3851 /* bytes_base: the bytes of data which have been read/written; and
3852 * it is obtained from the history statistic info.
3853 * bytes_res: the remaining bytes of data which need to be read/written.
3854 * (bytes_base + bytes_res) / bps_limit: used to calcuate
3855 * the total time for completing reading/writting all data.
3857 bytes_res
= (unsigned) nb_sectors
* BDRV_SECTOR_SIZE
;
3859 if (bytes_base
+ bytes_res
<= bytes_limit
) {
3867 /* Calc approx time to dispatch */
3868 wait_time
= (bytes_base
+ bytes_res
) / bps_limit
- elapsed_time
;
3870 /* When the I/O rate at runtime exceeds the limits,
3871 * bs->slice_end need to be extended in order that the current statistic
3872 * info can be kept until the timer fire, so it is increased and tuned
3873 * based on the result of experiment.
3875 extension
= wait_time
* NANOSECONDS_PER_SECOND
;
3876 extension
= DIV_ROUND_UP(extension
, BLOCK_IO_SLICE_TIME
) *
3877 BLOCK_IO_SLICE_TIME
;
3878 bs
->slice_end
+= extension
;
3880 *wait
= wait_time
* NANOSECONDS_PER_SECOND
;
3886 static bool bdrv_exceed_iops_limits(BlockDriverState
*bs
, bool is_write
,
3887 double elapsed_time
, uint64_t *wait
)
3889 uint64_t iops_limit
= 0;
3890 double ios_limit
, ios_base
;
3891 double slice_time
, wait_time
;
3893 if (bs
->io_limits
.iops
[BLOCK_IO_LIMIT_TOTAL
]) {
3894 iops_limit
= bs
->io_limits
.iops
[BLOCK_IO_LIMIT_TOTAL
];
3895 } else if (bs
->io_limits
.iops
[is_write
]) {
3896 iops_limit
= bs
->io_limits
.iops
[is_write
];
3905 slice_time
= bs
->slice_end
- bs
->slice_start
;
3906 slice_time
/= (NANOSECONDS_PER_SECOND
);
3907 ios_limit
= iops_limit
* slice_time
;
3908 ios_base
= bs
->slice_submitted
.ios
[is_write
];
3909 if (bs
->io_limits
.iops
[BLOCK_IO_LIMIT_TOTAL
]) {
3910 ios_base
+= bs
->slice_submitted
.ios
[!is_write
];
3913 if (ios_base
+ 1 <= ios_limit
) {
3921 /* Calc approx time to dispatch, in seconds */
3922 wait_time
= (ios_base
+ 1) / iops_limit
;
3923 if (wait_time
> elapsed_time
) {
3924 wait_time
= wait_time
- elapsed_time
;
3929 /* Exceeded current slice, extend it by another slice time */
3930 bs
->slice_end
+= BLOCK_IO_SLICE_TIME
;
3932 *wait
= wait_time
* NANOSECONDS_PER_SECOND
;
3938 static bool bdrv_exceed_io_limits(BlockDriverState
*bs
, int nb_sectors
,
3939 bool is_write
, int64_t *wait
)
3941 int64_t now
, max_wait
;
3942 uint64_t bps_wait
= 0, iops_wait
= 0;
3943 double elapsed_time
;
3944 int bps_ret
, iops_ret
;
3946 now
= qemu_get_clock_ns(vm_clock
);
3947 if (now
> bs
->slice_end
) {
3948 bs
->slice_start
= now
;
3949 bs
->slice_end
= now
+ BLOCK_IO_SLICE_TIME
;
3950 memset(&bs
->slice_submitted
, 0, sizeof(bs
->slice_submitted
));
3953 elapsed_time
= now
- bs
->slice_start
;
3954 elapsed_time
/= (NANOSECONDS_PER_SECOND
);
3956 bps_ret
= bdrv_exceed_bps_limits(bs
, nb_sectors
,
3957 is_write
, elapsed_time
, &bps_wait
);
3958 iops_ret
= bdrv_exceed_iops_limits(bs
, is_write
,
3959 elapsed_time
, &iops_wait
);
3960 if (bps_ret
|| iops_ret
) {
3961 max_wait
= bps_wait
> iops_wait
? bps_wait
: iops_wait
;
3966 now
= qemu_get_clock_ns(vm_clock
);
3967 if (bs
->slice_end
< now
+ max_wait
) {
3968 bs
->slice_end
= now
+ max_wait
;
3978 bs
->slice_submitted
.bytes
[is_write
] += (int64_t)nb_sectors
*
3980 bs
->slice_submitted
.ios
[is_write
]++;
3985 /**************************************************************/
3986 /* async block device emulation */
3988 typedef struct BlockDriverAIOCBSync
{
3989 BlockDriverAIOCB common
;
3992 /* vector translation state */
3996 } BlockDriverAIOCBSync
;
3998 static void bdrv_aio_cancel_em(BlockDriverAIOCB
*blockacb
)
4000 BlockDriverAIOCBSync
*acb
=
4001 container_of(blockacb
, BlockDriverAIOCBSync
, common
);
4002 qemu_bh_delete(acb
->bh
);
4004 qemu_aio_release(acb
);
4007 static const AIOCBInfo bdrv_em_aiocb_info
= {
4008 .aiocb_size
= sizeof(BlockDriverAIOCBSync
),
4009 .cancel
= bdrv_aio_cancel_em
,
4012 static void bdrv_aio_bh_cb(void *opaque
)
4014 BlockDriverAIOCBSync
*acb
= opaque
;
4017 qemu_iovec_from_buf(acb
->qiov
, 0, acb
->bounce
, acb
->qiov
->size
);
4018 qemu_vfree(acb
->bounce
);
4019 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
4020 qemu_bh_delete(acb
->bh
);
4022 qemu_aio_release(acb
);
4025 static BlockDriverAIOCB
*bdrv_aio_rw_vector(BlockDriverState
*bs
,
4029 BlockDriverCompletionFunc
*cb
,
4034 BlockDriverAIOCBSync
*acb
;
4036 acb
= qemu_aio_get(&bdrv_em_aiocb_info
, bs
, cb
, opaque
);
4037 acb
->is_write
= is_write
;
4039 acb
->bounce
= qemu_blockalign(bs
, qiov
->size
);
4040 acb
->bh
= qemu_bh_new(bdrv_aio_bh_cb
, acb
);
4043 qemu_iovec_to_buf(acb
->qiov
, 0, acb
->bounce
, qiov
->size
);
4044 acb
->ret
= bs
->drv
->bdrv_write(bs
, sector_num
, acb
->bounce
, nb_sectors
);
4046 acb
->ret
= bs
->drv
->bdrv_read(bs
, sector_num
, acb
->bounce
, nb_sectors
);
4049 qemu_bh_schedule(acb
->bh
);
4051 return &acb
->common
;
4054 static BlockDriverAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
4055 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
4056 BlockDriverCompletionFunc
*cb
, void *opaque
)
4058 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 0);
4061 static BlockDriverAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
4062 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
4063 BlockDriverCompletionFunc
*cb
, void *opaque
)
4065 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 1);
4069 typedef struct BlockDriverAIOCBCoroutine
{
4070 BlockDriverAIOCB common
;
4075 } BlockDriverAIOCBCoroutine
;
4077 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB
*blockacb
)
4079 BlockDriverAIOCBCoroutine
*acb
=
4080 container_of(blockacb
, BlockDriverAIOCBCoroutine
, common
);
4089 static const AIOCBInfo bdrv_em_co_aiocb_info
= {
4090 .aiocb_size
= sizeof(BlockDriverAIOCBCoroutine
),
4091 .cancel
= bdrv_aio_co_cancel_em
,
4094 static void bdrv_co_em_bh(void *opaque
)
4096 BlockDriverAIOCBCoroutine
*acb
= opaque
;
4098 acb
->common
.cb(acb
->common
.opaque
, acb
->req
.error
);
4104 qemu_bh_delete(acb
->bh
);
4105 qemu_aio_release(acb
);
4108 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
4109 static void coroutine_fn
bdrv_co_do_rw(void *opaque
)
4111 BlockDriverAIOCBCoroutine
*acb
= opaque
;
4112 BlockDriverState
*bs
= acb
->common
.bs
;
4114 if (!acb
->is_write
) {
4115 acb
->req
.error
= bdrv_co_do_readv(bs
, acb
->req
.sector
,
4116 acb
->req
.nb_sectors
, acb
->req
.qiov
, 0);
4118 acb
->req
.error
= bdrv_co_do_writev(bs
, acb
->req
.sector
,
4119 acb
->req
.nb_sectors
, acb
->req
.qiov
, 0);
4122 acb
->bh
= qemu_bh_new(bdrv_co_em_bh
, acb
);
4123 qemu_bh_schedule(acb
->bh
);
4126 static BlockDriverAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
4130 BlockDriverCompletionFunc
*cb
,
4135 BlockDriverAIOCBCoroutine
*acb
;
4137 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
4138 acb
->req
.sector
= sector_num
;
4139 acb
->req
.nb_sectors
= nb_sectors
;
4140 acb
->req
.qiov
= qiov
;
4141 acb
->is_write
= is_write
;
4144 co
= qemu_coroutine_create(bdrv_co_do_rw
);
4145 qemu_coroutine_enter(co
, acb
);
4147 return &acb
->common
;
4150 static void coroutine_fn
bdrv_aio_flush_co_entry(void *opaque
)
4152 BlockDriverAIOCBCoroutine
*acb
= opaque
;
4153 BlockDriverState
*bs
= acb
->common
.bs
;
4155 acb
->req
.error
= bdrv_co_flush(bs
);
4156 acb
->bh
= qemu_bh_new(bdrv_co_em_bh
, acb
);
4157 qemu_bh_schedule(acb
->bh
);
4160 BlockDriverAIOCB
*bdrv_aio_flush(BlockDriverState
*bs
,
4161 BlockDriverCompletionFunc
*cb
, void *opaque
)
4163 trace_bdrv_aio_flush(bs
, opaque
);
4166 BlockDriverAIOCBCoroutine
*acb
;
4168 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
4171 co
= qemu_coroutine_create(bdrv_aio_flush_co_entry
);
4172 qemu_coroutine_enter(co
, acb
);
4174 return &acb
->common
;
4177 static void coroutine_fn
bdrv_aio_discard_co_entry(void *opaque
)
4179 BlockDriverAIOCBCoroutine
*acb
= opaque
;
4180 BlockDriverState
*bs
= acb
->common
.bs
;
4182 acb
->req
.error
= bdrv_co_discard(bs
, acb
->req
.sector
, acb
->req
.nb_sectors
);
4183 acb
->bh
= qemu_bh_new(bdrv_co_em_bh
, acb
);
4184 qemu_bh_schedule(acb
->bh
);
4187 BlockDriverAIOCB
*bdrv_aio_discard(BlockDriverState
*bs
,
4188 int64_t sector_num
, int nb_sectors
,
4189 BlockDriverCompletionFunc
*cb
, void *opaque
)
4192 BlockDriverAIOCBCoroutine
*acb
;
4194 trace_bdrv_aio_discard(bs
, sector_num
, nb_sectors
, opaque
);
4196 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
4197 acb
->req
.sector
= sector_num
;
4198 acb
->req
.nb_sectors
= nb_sectors
;
4200 co
= qemu_coroutine_create(bdrv_aio_discard_co_entry
);
4201 qemu_coroutine_enter(co
, acb
);
4203 return &acb
->common
;
4206 void bdrv_init(void)
4208 module_call_init(MODULE_INIT_BLOCK
);
4211 void bdrv_init_with_whitelist(void)
4213 use_bdrv_whitelist
= 1;
4217 void *qemu_aio_get(const AIOCBInfo
*aiocb_info
, BlockDriverState
*bs
,
4218 BlockDriverCompletionFunc
*cb
, void *opaque
)
4220 BlockDriverAIOCB
*acb
;
4222 acb
= g_slice_alloc(aiocb_info
->aiocb_size
);
4223 acb
->aiocb_info
= aiocb_info
;
4226 acb
->opaque
= opaque
;
4230 void qemu_aio_release(void *p
)
4232 BlockDriverAIOCB
*acb
= p
;
4233 g_slice_free1(acb
->aiocb_info
->aiocb_size
, acb
);
4236 /**************************************************************/
4237 /* Coroutine block device emulation */
4239 typedef struct CoroutineIOCompletion
{
4240 Coroutine
*coroutine
;
4242 } CoroutineIOCompletion
;
4244 static void bdrv_co_io_em_complete(void *opaque
, int ret
)
4246 CoroutineIOCompletion
*co
= opaque
;
4249 qemu_coroutine_enter(co
->coroutine
, NULL
);
4252 static int coroutine_fn
bdrv_co_io_em(BlockDriverState
*bs
, int64_t sector_num
,
4253 int nb_sectors
, QEMUIOVector
*iov
,
4256 CoroutineIOCompletion co
= {
4257 .coroutine
= qemu_coroutine_self(),
4259 BlockDriverAIOCB
*acb
;
4262 acb
= bs
->drv
->bdrv_aio_writev(bs
, sector_num
, iov
, nb_sectors
,
4263 bdrv_co_io_em_complete
, &co
);
4265 acb
= bs
->drv
->bdrv_aio_readv(bs
, sector_num
, iov
, nb_sectors
,
4266 bdrv_co_io_em_complete
, &co
);
4269 trace_bdrv_co_io_em(bs
, sector_num
, nb_sectors
, is_write
, acb
);
4273 qemu_coroutine_yield();
4278 static int coroutine_fn
bdrv_co_readv_em(BlockDriverState
*bs
,
4279 int64_t sector_num
, int nb_sectors
,
4282 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, false);
4285 static int coroutine_fn
bdrv_co_writev_em(BlockDriverState
*bs
,
4286 int64_t sector_num
, int nb_sectors
,
4289 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, true);
4292 static void coroutine_fn
bdrv_flush_co_entry(void *opaque
)
4294 RwCo
*rwco
= opaque
;
4296 rwco
->ret
= bdrv_co_flush(rwco
->bs
);
4299 int coroutine_fn
bdrv_co_flush(BlockDriverState
*bs
)
4303 if (!bs
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
)) {
4307 /* Write back cached data to the OS even with cache=unsafe */
4308 if (bs
->drv
->bdrv_co_flush_to_os
) {
4309 ret
= bs
->drv
->bdrv_co_flush_to_os(bs
);
4315 /* But don't actually force it to the disk with cache=unsafe */
4316 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
4320 if (bs
->drv
->bdrv_co_flush_to_disk
) {
4321 ret
= bs
->drv
->bdrv_co_flush_to_disk(bs
);
4322 } else if (bs
->drv
->bdrv_aio_flush
) {
4323 BlockDriverAIOCB
*acb
;
4324 CoroutineIOCompletion co
= {
4325 .coroutine
= qemu_coroutine_self(),
4328 acb
= bs
->drv
->bdrv_aio_flush(bs
, bdrv_co_io_em_complete
, &co
);
4332 qemu_coroutine_yield();
4337 * Some block drivers always operate in either writethrough or unsafe
4338 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
4339 * know how the server works (because the behaviour is hardcoded or
4340 * depends on server-side configuration), so we can't ensure that
4341 * everything is safe on disk. Returning an error doesn't work because
4342 * that would break guests even if the server operates in writethrough
4345 * Let's hope the user knows what he's doing.
4353 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
4354 * in the case of cache=unsafe, so there are no useless flushes.
4357 return bdrv_co_flush(bs
->file
);
4360 void bdrv_invalidate_cache(BlockDriverState
*bs
)
4362 if (bs
->drv
&& bs
->drv
->bdrv_invalidate_cache
) {
4363 bs
->drv
->bdrv_invalidate_cache(bs
);
4367 void bdrv_invalidate_cache_all(void)
4369 BlockDriverState
*bs
;
4371 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
4372 bdrv_invalidate_cache(bs
);
4376 void bdrv_clear_incoming_migration_all(void)
4378 BlockDriverState
*bs
;
4380 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
4381 bs
->open_flags
= bs
->open_flags
& ~(BDRV_O_INCOMING
);
4385 int bdrv_flush(BlockDriverState
*bs
)
4393 if (qemu_in_coroutine()) {
4394 /* Fast-path if already in coroutine context */
4395 bdrv_flush_co_entry(&rwco
);
4397 co
= qemu_coroutine_create(bdrv_flush_co_entry
);
4398 qemu_coroutine_enter(co
, &rwco
);
4399 while (rwco
.ret
== NOT_DONE
) {
4407 static void coroutine_fn
bdrv_discard_co_entry(void *opaque
)
4409 RwCo
*rwco
= opaque
;
4411 rwco
->ret
= bdrv_co_discard(rwco
->bs
, rwco
->sector_num
, rwco
->nb_sectors
);
4414 int coroutine_fn
bdrv_co_discard(BlockDriverState
*bs
, int64_t sector_num
,
4419 } else if (bdrv_check_request(bs
, sector_num
, nb_sectors
)) {
4421 } else if (bs
->read_only
) {
4425 if (bs
->dirty_bitmap
) {
4426 bdrv_reset_dirty(bs
, sector_num
, nb_sectors
);
4429 /* Do nothing if disabled. */
4430 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
4434 if (bs
->drv
->bdrv_co_discard
) {
4435 return bs
->drv
->bdrv_co_discard(bs
, sector_num
, nb_sectors
);
4436 } else if (bs
->drv
->bdrv_aio_discard
) {
4437 BlockDriverAIOCB
*acb
;
4438 CoroutineIOCompletion co
= {
4439 .coroutine
= qemu_coroutine_self(),
4442 acb
= bs
->drv
->bdrv_aio_discard(bs
, sector_num
, nb_sectors
,
4443 bdrv_co_io_em_complete
, &co
);
4447 qemu_coroutine_yield();
4455 int bdrv_discard(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
)
4460 .sector_num
= sector_num
,
4461 .nb_sectors
= nb_sectors
,
4465 if (qemu_in_coroutine()) {
4466 /* Fast-path if already in coroutine context */
4467 bdrv_discard_co_entry(&rwco
);
4469 co
= qemu_coroutine_create(bdrv_discard_co_entry
);
4470 qemu_coroutine_enter(co
, &rwco
);
4471 while (rwco
.ret
== NOT_DONE
) {
4479 /**************************************************************/
4480 /* removable device support */
4483 * Return TRUE if the media is present
4485 int bdrv_is_inserted(BlockDriverState
*bs
)
4487 BlockDriver
*drv
= bs
->drv
;
4491 if (!drv
->bdrv_is_inserted
)
4493 return drv
->bdrv_is_inserted(bs
);
4497 * Return whether the media changed since the last call to this
4498 * function, or -ENOTSUP if we don't know. Most drivers don't know.
4500 int bdrv_media_changed(BlockDriverState
*bs
)
4502 BlockDriver
*drv
= bs
->drv
;
4504 if (drv
&& drv
->bdrv_media_changed
) {
4505 return drv
->bdrv_media_changed(bs
);
4511 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4513 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
4515 BlockDriver
*drv
= bs
->drv
;
4517 if (drv
&& drv
->bdrv_eject
) {
4518 drv
->bdrv_eject(bs
, eject_flag
);
4521 if (bs
->device_name
[0] != '\0') {
4522 bdrv_emit_qmp_eject_event(bs
, eject_flag
);
4527 * Lock or unlock the media (if it is locked, the user won't be able
4528 * to eject it manually).
4530 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
4532 BlockDriver
*drv
= bs
->drv
;
4534 trace_bdrv_lock_medium(bs
, locked
);
4536 if (drv
&& drv
->bdrv_lock_medium
) {
4537 drv
->bdrv_lock_medium(bs
, locked
);
4541 /* needed for generic scsi interface */
4543 int bdrv_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
4545 BlockDriver
*drv
= bs
->drv
;
4547 if (drv
&& drv
->bdrv_ioctl
)
4548 return drv
->bdrv_ioctl(bs
, req
, buf
);
4552 BlockDriverAIOCB
*bdrv_aio_ioctl(BlockDriverState
*bs
,
4553 unsigned long int req
, void *buf
,
4554 BlockDriverCompletionFunc
*cb
, void *opaque
)
4556 BlockDriver
*drv
= bs
->drv
;
4558 if (drv
&& drv
->bdrv_aio_ioctl
)
4559 return drv
->bdrv_aio_ioctl(bs
, req
, buf
, cb
, opaque
);
4563 void bdrv_set_buffer_alignment(BlockDriverState
*bs
, int align
)
4565 bs
->buffer_alignment
= align
;
4568 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
4570 return qemu_memalign((bs
&& bs
->buffer_alignment
) ? bs
->buffer_alignment
: 512, size
);
4574 * Check if all memory in this vector is sector aligned.
4576 bool bdrv_qiov_is_aligned(BlockDriverState
*bs
, QEMUIOVector
*qiov
)
4580 for (i
= 0; i
< qiov
->niov
; i
++) {
4581 if ((uintptr_t) qiov
->iov
[i
].iov_base
% bs
->buffer_alignment
) {
4589 void bdrv_set_dirty_tracking(BlockDriverState
*bs
, int granularity
)
4591 int64_t bitmap_size
;
4593 assert((granularity
& (granularity
- 1)) == 0);
4596 granularity
>>= BDRV_SECTOR_BITS
;
4597 assert(!bs
->dirty_bitmap
);
4598 bitmap_size
= (bdrv_getlength(bs
) >> BDRV_SECTOR_BITS
);
4599 bs
->dirty_bitmap
= hbitmap_alloc(bitmap_size
, ffs(granularity
) - 1);
4601 if (bs
->dirty_bitmap
) {
4602 hbitmap_free(bs
->dirty_bitmap
);
4603 bs
->dirty_bitmap
= NULL
;
4608 int bdrv_get_dirty(BlockDriverState
*bs
, int64_t sector
)
4610 if (bs
->dirty_bitmap
) {
4611 return hbitmap_get(bs
->dirty_bitmap
, sector
);
4617 void bdrv_dirty_iter_init(BlockDriverState
*bs
, HBitmapIter
*hbi
)
4619 hbitmap_iter_init(hbi
, bs
->dirty_bitmap
, 0);
4622 void bdrv_set_dirty(BlockDriverState
*bs
, int64_t cur_sector
,
4625 hbitmap_set(bs
->dirty_bitmap
, cur_sector
, nr_sectors
);
4628 void bdrv_reset_dirty(BlockDriverState
*bs
, int64_t cur_sector
,
4631 hbitmap_reset(bs
->dirty_bitmap
, cur_sector
, nr_sectors
);
4634 int64_t bdrv_get_dirty_count(BlockDriverState
*bs
)
4636 if (bs
->dirty_bitmap
) {
4637 return hbitmap_count(bs
->dirty_bitmap
);
4643 void bdrv_set_in_use(BlockDriverState
*bs
, int in_use
)
4645 assert(bs
->in_use
!= in_use
);
4646 bs
->in_use
= in_use
;
4649 int bdrv_in_use(BlockDriverState
*bs
)
4654 void bdrv_iostatus_enable(BlockDriverState
*bs
)
4656 bs
->iostatus_enabled
= true;
4657 bs
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
4660 /* The I/O status is only enabled if the drive explicitly
4661 * enables it _and_ the VM is configured to stop on errors */
4662 bool bdrv_iostatus_is_enabled(const BlockDriverState
*bs
)
4664 return (bs
->iostatus_enabled
&&
4665 (bs
->on_write_error
== BLOCKDEV_ON_ERROR_ENOSPC
||
4666 bs
->on_write_error
== BLOCKDEV_ON_ERROR_STOP
||
4667 bs
->on_read_error
== BLOCKDEV_ON_ERROR_STOP
));
4670 void bdrv_iostatus_disable(BlockDriverState
*bs
)
4672 bs
->iostatus_enabled
= false;
4675 void bdrv_iostatus_reset(BlockDriverState
*bs
)
4677 if (bdrv_iostatus_is_enabled(bs
)) {
4678 bs
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
4680 block_job_iostatus_reset(bs
->job
);
4685 void bdrv_iostatus_set_err(BlockDriverState
*bs
, int error
)
4687 assert(bdrv_iostatus_is_enabled(bs
));
4688 if (bs
->iostatus
== BLOCK_DEVICE_IO_STATUS_OK
) {
4689 bs
->iostatus
= error
== ENOSPC
? BLOCK_DEVICE_IO_STATUS_NOSPACE
:
4690 BLOCK_DEVICE_IO_STATUS_FAILED
;
4695 bdrv_acct_start(BlockDriverState
*bs
, BlockAcctCookie
*cookie
, int64_t bytes
,
4696 enum BlockAcctType type
)
4698 assert(type
< BDRV_MAX_IOTYPE
);
4700 cookie
->bytes
= bytes
;
4701 cookie
->start_time_ns
= get_clock();
4702 cookie
->type
= type
;
4706 bdrv_acct_done(BlockDriverState
*bs
, BlockAcctCookie
*cookie
)
4708 assert(cookie
->type
< BDRV_MAX_IOTYPE
);
4710 bs
->nr_bytes
[cookie
->type
] += cookie
->bytes
;
4711 bs
->nr_ops
[cookie
->type
]++;
4712 bs
->total_time_ns
[cookie
->type
] += get_clock() - cookie
->start_time_ns
;
4715 void bdrv_img_create(const char *filename
, const char *fmt
,
4716 const char *base_filename
, const char *base_fmt
,
4717 char *options
, uint64_t img_size
, int flags
,
4718 Error
**errp
, bool quiet
)
4720 QEMUOptionParameter
*param
= NULL
, *create_options
= NULL
;
4721 QEMUOptionParameter
*backing_fmt
, *backing_file
, *size
;
4722 BlockDriverState
*bs
= NULL
;
4723 BlockDriver
*drv
, *proto_drv
;
4724 BlockDriver
*backing_drv
= NULL
;
4727 /* Find driver and parse its options */
4728 drv
= bdrv_find_format(fmt
);
4730 error_setg(errp
, "Unknown file format '%s'", fmt
);
4734 proto_drv
= bdrv_find_protocol(filename
);
4736 error_setg(errp
, "Unknown protocol '%s'", filename
);
4740 create_options
= append_option_parameters(create_options
,
4741 drv
->create_options
);
4742 create_options
= append_option_parameters(create_options
,
4743 proto_drv
->create_options
);
4745 /* Create parameter list with default values */
4746 param
= parse_option_parameters("", create_options
, param
);
4748 set_option_parameter_int(param
, BLOCK_OPT_SIZE
, img_size
);
4750 /* Parse -o options */
4752 param
= parse_option_parameters(options
, create_options
, param
);
4753 if (param
== NULL
) {
4754 error_setg(errp
, "Invalid options for file format '%s'.", fmt
);
4759 if (base_filename
) {
4760 if (set_option_parameter(param
, BLOCK_OPT_BACKING_FILE
,
4762 error_setg(errp
, "Backing file not supported for file format '%s'",
4769 if (set_option_parameter(param
, BLOCK_OPT_BACKING_FMT
, base_fmt
)) {
4770 error_setg(errp
, "Backing file format not supported for file "
4771 "format '%s'", fmt
);
4776 backing_file
= get_option_parameter(param
, BLOCK_OPT_BACKING_FILE
);
4777 if (backing_file
&& backing_file
->value
.s
) {
4778 if (!strcmp(filename
, backing_file
->value
.s
)) {
4779 error_setg(errp
, "Error: Trying to create an image with the "
4780 "same filename as the backing file");
4785 backing_fmt
= get_option_parameter(param
, BLOCK_OPT_BACKING_FMT
);
4786 if (backing_fmt
&& backing_fmt
->value
.s
) {
4787 backing_drv
= bdrv_find_format(backing_fmt
->value
.s
);
4789 error_setg(errp
, "Unknown backing file format '%s'",
4790 backing_fmt
->value
.s
);
4795 // The size for the image must always be specified, with one exception:
4796 // If we are using a backing file, we can obtain the size from there
4797 size
= get_option_parameter(param
, BLOCK_OPT_SIZE
);
4798 if (size
&& size
->value
.n
== -1) {
4799 if (backing_file
&& backing_file
->value
.s
) {
4804 /* backing files always opened read-only */
4806 flags
& ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
4810 ret
= bdrv_open(bs
, backing_file
->value
.s
, NULL
, back_flags
,
4813 error_setg_errno(errp
, -ret
, "Could not open '%s'",
4814 backing_file
->value
.s
);
4817 bdrv_get_geometry(bs
, &size
);
4820 snprintf(buf
, sizeof(buf
), "%" PRId64
, size
);
4821 set_option_parameter(param
, BLOCK_OPT_SIZE
, buf
);
4823 error_setg(errp
, "Image creation needs a size parameter");
4829 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
4830 print_option_parameters(param
);
4833 ret
= bdrv_create(drv
, filename
, param
);
4835 if (ret
== -ENOTSUP
) {
4836 error_setg(errp
,"Formatting or formatting option not supported for "
4837 "file format '%s'", fmt
);
4838 } else if (ret
== -EFBIG
) {
4839 error_setg(errp
, "The image size is too large for file format '%s'",
4842 error_setg(errp
, "%s: error while creating %s: %s", filename
, fmt
,
4848 free_option_parameters(create_options
);
4849 free_option_parameters(param
);
4856 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
4858 /* Currently BlockDriverState always uses the main loop AioContext */
4859 return qemu_get_aio_context();