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
;
144 memset(&bs
->io_base
, 0, sizeof(bs
->io_base
));
147 static void bdrv_block_timer(void *opaque
)
149 BlockDriverState
*bs
= opaque
;
151 qemu_co_queue_next(&bs
->throttled_reqs
);
154 void bdrv_io_limits_enable(BlockDriverState
*bs
)
156 qemu_co_queue_init(&bs
->throttled_reqs
);
157 bs
->block_timer
= qemu_new_timer_ns(vm_clock
, bdrv_block_timer
, bs
);
158 bs
->io_limits_enabled
= true;
161 bool bdrv_io_limits_enabled(BlockDriverState
*bs
)
163 BlockIOLimit
*io_limits
= &bs
->io_limits
;
164 return io_limits
->bps
[BLOCK_IO_LIMIT_READ
]
165 || io_limits
->bps
[BLOCK_IO_LIMIT_WRITE
]
166 || io_limits
->bps
[BLOCK_IO_LIMIT_TOTAL
]
167 || io_limits
->iops
[BLOCK_IO_LIMIT_READ
]
168 || io_limits
->iops
[BLOCK_IO_LIMIT_WRITE
]
169 || io_limits
->iops
[BLOCK_IO_LIMIT_TOTAL
];
172 static void bdrv_io_limits_intercept(BlockDriverState
*bs
,
173 bool is_write
, int nb_sectors
)
175 int64_t wait_time
= -1;
177 if (!qemu_co_queue_empty(&bs
->throttled_reqs
)) {
178 qemu_co_queue_wait(&bs
->throttled_reqs
);
181 /* In fact, we hope to keep each request's timing, in FIFO mode. The next
182 * throttled requests will not be dequeued until the current request is
183 * allowed to be serviced. So if the current request still exceeds the
184 * limits, it will be inserted to the head. All requests followed it will
185 * be still in throttled_reqs queue.
188 while (bdrv_exceed_io_limits(bs
, nb_sectors
, is_write
, &wait_time
)) {
189 qemu_mod_timer(bs
->block_timer
,
190 wait_time
+ qemu_get_clock_ns(vm_clock
));
191 qemu_co_queue_wait_insert_head(&bs
->throttled_reqs
);
194 qemu_co_queue_next(&bs
->throttled_reqs
);
197 /* check if the path starts with "<protocol>:" */
198 static int path_has_protocol(const char *path
)
203 if (is_windows_drive(path
) ||
204 is_windows_drive_prefix(path
)) {
207 p
= path
+ strcspn(path
, ":/\\");
209 p
= path
+ strcspn(path
, ":/");
215 int path_is_absolute(const char *path
)
218 /* specific case for names like: "\\.\d:" */
219 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
222 return (*path
== '/' || *path
== '\\');
224 return (*path
== '/');
228 /* if filename is absolute, just copy it to dest. Otherwise, build a
229 path to it by considering it is relative to base_path. URL are
231 void path_combine(char *dest
, int dest_size
,
232 const char *base_path
,
233 const char *filename
)
240 if (path_is_absolute(filename
)) {
241 pstrcpy(dest
, dest_size
, filename
);
243 p
= strchr(base_path
, ':');
248 p1
= strrchr(base_path
, '/');
252 p2
= strrchr(base_path
, '\\');
264 if (len
> dest_size
- 1)
266 memcpy(dest
, base_path
, len
);
268 pstrcat(dest
, dest_size
, filename
);
272 void bdrv_get_full_backing_filename(BlockDriverState
*bs
, char *dest
, size_t sz
)
274 if (bs
->backing_file
[0] == '\0' || path_has_protocol(bs
->backing_file
)) {
275 pstrcpy(dest
, sz
, bs
->backing_file
);
277 path_combine(dest
, sz
, bs
->filename
, bs
->backing_file
);
281 void bdrv_register(BlockDriver
*bdrv
)
283 /* Block drivers without coroutine functions need emulation */
284 if (!bdrv
->bdrv_co_readv
) {
285 bdrv
->bdrv_co_readv
= bdrv_co_readv_em
;
286 bdrv
->bdrv_co_writev
= bdrv_co_writev_em
;
288 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
289 * the block driver lacks aio we need to emulate that too.
291 if (!bdrv
->bdrv_aio_readv
) {
292 /* add AIO emulation layer */
293 bdrv
->bdrv_aio_readv
= bdrv_aio_readv_em
;
294 bdrv
->bdrv_aio_writev
= bdrv_aio_writev_em
;
298 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
301 /* create a new block device (by default it is empty) */
302 BlockDriverState
*bdrv_new(const char *device_name
)
304 BlockDriverState
*bs
;
306 bs
= g_malloc0(sizeof(BlockDriverState
));
307 pstrcpy(bs
->device_name
, sizeof(bs
->device_name
), device_name
);
308 if (device_name
[0] != '\0') {
309 QTAILQ_INSERT_TAIL(&bdrv_states
, bs
, list
);
311 bdrv_iostatus_disable(bs
);
312 notifier_list_init(&bs
->close_notifiers
);
317 void bdrv_add_close_notifier(BlockDriverState
*bs
, Notifier
*notify
)
319 notifier_list_add(&bs
->close_notifiers
, notify
);
322 BlockDriver
*bdrv_find_format(const char *format_name
)
325 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
326 if (!strcmp(drv1
->format_name
, format_name
)) {
333 static int bdrv_is_whitelisted(BlockDriver
*drv
)
335 static const char *whitelist
[] = {
336 CONFIG_BDRV_WHITELIST
341 return 1; /* no whitelist, anything goes */
343 for (p
= whitelist
; *p
; p
++) {
344 if (!strcmp(drv
->format_name
, *p
)) {
351 BlockDriver
*bdrv_find_whitelisted_format(const char *format_name
)
353 BlockDriver
*drv
= bdrv_find_format(format_name
);
354 return drv
&& bdrv_is_whitelisted(drv
) ? drv
: NULL
;
357 typedef struct CreateCo
{
360 QEMUOptionParameter
*options
;
364 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
366 CreateCo
*cco
= opaque
;
369 cco
->ret
= cco
->drv
->bdrv_create(cco
->filename
, cco
->options
);
372 int bdrv_create(BlockDriver
*drv
, const char* filename
,
373 QEMUOptionParameter
*options
)
380 .filename
= g_strdup(filename
),
385 if (!drv
->bdrv_create
) {
390 if (qemu_in_coroutine()) {
391 /* Fast-path if already in coroutine context */
392 bdrv_create_co_entry(&cco
);
394 co
= qemu_coroutine_create(bdrv_create_co_entry
);
395 qemu_coroutine_enter(co
, &cco
);
396 while (cco
.ret
== NOT_DONE
) {
404 g_free(cco
.filename
);
408 int bdrv_create_file(const char* filename
, QEMUOptionParameter
*options
)
412 drv
= bdrv_find_protocol(filename
);
417 return bdrv_create(drv
, filename
, options
);
421 * Create a uniquely-named empty temporary file.
422 * Return 0 upon success, otherwise a negative errno value.
424 int get_tmp_filename(char *filename
, int size
)
427 char temp_dir
[MAX_PATH
];
428 /* GetTempFileName requires that its output buffer (4th param)
429 have length MAX_PATH or greater. */
430 assert(size
>= MAX_PATH
);
431 return (GetTempPath(MAX_PATH
, temp_dir
)
432 && GetTempFileName(temp_dir
, "qem", 0, filename
)
433 ? 0 : -GetLastError());
437 tmpdir
= getenv("TMPDIR");
440 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
443 fd
= mkstemp(filename
);
447 if (close(fd
) != 0) {
456 * Detect host devices. By convention, /dev/cdrom[N] is always
457 * recognized as a host CDROM.
459 static BlockDriver
*find_hdev_driver(const char *filename
)
461 int score_max
= 0, score
;
462 BlockDriver
*drv
= NULL
, *d
;
464 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
465 if (d
->bdrv_probe_device
) {
466 score
= d
->bdrv_probe_device(filename
);
467 if (score
> score_max
) {
477 BlockDriver
*bdrv_find_protocol(const char *filename
)
484 /* TODO Drivers without bdrv_file_open must be specified explicitly */
487 * XXX(hch): we really should not let host device detection
488 * override an explicit protocol specification, but moving this
489 * later breaks access to device names with colons in them.
490 * Thanks to the brain-dead persistent naming schemes on udev-
491 * based Linux systems those actually are quite common.
493 drv1
= find_hdev_driver(filename
);
498 if (!path_has_protocol(filename
)) {
499 return bdrv_find_format("file");
501 p
= strchr(filename
, ':');
504 if (len
> sizeof(protocol
) - 1)
505 len
= sizeof(protocol
) - 1;
506 memcpy(protocol
, filename
, len
);
507 protocol
[len
] = '\0';
508 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
509 if (drv1
->protocol_name
&&
510 !strcmp(drv1
->protocol_name
, protocol
)) {
517 static int find_image_format(BlockDriverState
*bs
, const char *filename
,
520 int score
, score_max
;
521 BlockDriver
*drv1
, *drv
;
525 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
526 if (bs
->sg
|| !bdrv_is_inserted(bs
) || bdrv_getlength(bs
) == 0) {
527 drv
= bdrv_find_format("raw");
535 ret
= bdrv_pread(bs
, 0, buf
, sizeof(buf
));
543 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
544 if (drv1
->bdrv_probe
) {
545 score
= drv1
->bdrv_probe(buf
, ret
, filename
);
546 if (score
> score_max
) {
560 * Set the current 'total_sectors' value
562 static int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
564 BlockDriver
*drv
= bs
->drv
;
566 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
570 /* query actual device if possible, otherwise just trust the hint */
571 if (drv
->bdrv_getlength
) {
572 int64_t length
= drv
->bdrv_getlength(bs
);
576 hint
= length
>> BDRV_SECTOR_BITS
;
579 bs
->total_sectors
= hint
;
584 * Set open flags for a given discard mode
586 * Return 0 on success, -1 if the discard mode was invalid.
588 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
590 *flags
&= ~BDRV_O_UNMAP
;
592 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
594 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
595 *flags
|= BDRV_O_UNMAP
;
604 * Set open flags for a given cache mode
606 * Return 0 on success, -1 if the cache mode was invalid.
608 int bdrv_parse_cache_flags(const char *mode
, int *flags
)
610 *flags
&= ~BDRV_O_CACHE_MASK
;
612 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
613 *flags
|= BDRV_O_NOCACHE
| BDRV_O_CACHE_WB
;
614 } else if (!strcmp(mode
, "directsync")) {
615 *flags
|= BDRV_O_NOCACHE
;
616 } else if (!strcmp(mode
, "writeback")) {
617 *flags
|= BDRV_O_CACHE_WB
;
618 } else if (!strcmp(mode
, "unsafe")) {
619 *flags
|= BDRV_O_CACHE_WB
;
620 *flags
|= BDRV_O_NO_FLUSH
;
621 } else if (!strcmp(mode
, "writethrough")) {
622 /* this is the default */
631 * The copy-on-read flag is actually a reference count so multiple users may
632 * use the feature without worrying about clobbering its previous state.
633 * Copy-on-read stays enabled until all users have called to disable it.
635 void bdrv_enable_copy_on_read(BlockDriverState
*bs
)
640 void bdrv_disable_copy_on_read(BlockDriverState
*bs
)
642 assert(bs
->copy_on_read
> 0);
646 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
648 int open_flags
= flags
| BDRV_O_CACHE_WB
;
651 * Clear flags that are internal to the block layer before opening the
654 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
657 * Snapshots should be writable.
659 if (bs
->is_temporary
) {
660 open_flags
|= BDRV_O_RDWR
;
667 * Common part for opening disk images and files
669 static int bdrv_open_common(BlockDriverState
*bs
, BlockDriverState
*file
,
670 const char *filename
,
671 int flags
, BlockDriver
*drv
)
676 assert(bs
->file
== NULL
);
678 trace_bdrv_open_common(bs
, filename
, flags
, drv
->format_name
);
680 bs
->open_flags
= flags
;
681 bs
->buffer_alignment
= 512;
683 assert(bs
->copy_on_read
== 0); /* bdrv_new() and bdrv_close() make it so */
684 if ((flags
& BDRV_O_RDWR
) && (flags
& BDRV_O_COPY_ON_READ
)) {
685 bdrv_enable_copy_on_read(bs
);
688 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
690 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
)) {
695 bs
->opaque
= g_malloc0(drv
->instance_size
);
697 bs
->enable_write_cache
= !!(flags
& BDRV_O_CACHE_WB
);
698 open_flags
= bdrv_open_flags(bs
, flags
);
700 bs
->read_only
= !(open_flags
& BDRV_O_RDWR
);
702 /* Open the image, either directly or using a protocol */
703 if (drv
->bdrv_file_open
) {
708 ret
= drv
->bdrv_file_open(bs
, filename
, open_flags
);
711 assert(file
!= NULL
);
713 ret
= drv
->bdrv_open(bs
, NULL
, open_flags
);
720 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
726 if (bs
->is_temporary
) {
741 * Opens a file using a protocol (file, host_device, nbd, ...)
743 int bdrv_file_open(BlockDriverState
**pbs
, const char *filename
, int flags
)
745 BlockDriverState
*bs
;
749 drv
= bdrv_find_protocol(filename
);
755 ret
= bdrv_open_common(bs
, NULL
, filename
, flags
, drv
);
765 int bdrv_open_backing_file(BlockDriverState
*bs
)
767 char backing_filename
[PATH_MAX
];
769 BlockDriver
*back_drv
= NULL
;
771 if (bs
->backing_hd
!= NULL
) {
775 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
776 if (bs
->backing_file
[0] == '\0') {
780 bs
->backing_hd
= bdrv_new("");
781 bdrv_get_full_backing_filename(bs
, backing_filename
,
782 sizeof(backing_filename
));
784 if (bs
->backing_format
[0] != '\0') {
785 back_drv
= bdrv_find_format(bs
->backing_format
);
788 /* backing files always opened read-only */
789 back_flags
= bs
->open_flags
& ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
);
791 ret
= bdrv_open(bs
->backing_hd
, backing_filename
, NULL
,
792 back_flags
, back_drv
);
794 bdrv_delete(bs
->backing_hd
);
795 bs
->backing_hd
= NULL
;
796 bs
->open_flags
|= BDRV_O_NO_BACKING
;
803 * Opens a disk image (raw, qcow2, vmdk, ...)
805 * options is a QDict of options to pass to the block drivers, or NULL for an
806 * empty set of options. The reference to the QDict belongs to the block layer
807 * after the call (even on failure), so if the caller intends to reuse the
808 * dictionary, it needs to use QINCREF() before calling bdrv_open.
810 int bdrv_open(BlockDriverState
*bs
, const char *filename
, QDict
*options
,
811 int flags
, BlockDriver
*drv
)
814 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
815 char tmp_filename
[PATH_MAX
+ 1];
816 BlockDriverState
*file
= NULL
;
818 /* NULL means an empty set of options */
819 if (options
== NULL
) {
820 options
= qdict_new();
823 bs
->options
= options
;
825 /* For snapshot=on, create a temporary qcow2 overlay */
826 if (flags
& BDRV_O_SNAPSHOT
) {
827 BlockDriverState
*bs1
;
830 BlockDriver
*bdrv_qcow2
;
831 QEMUOptionParameter
*options
;
832 char backing_filename
[PATH_MAX
];
834 /* if snapshot, we create a temporary backing file and open it
835 instead of opening 'filename' directly */
837 /* if there is a backing file, use it */
839 ret
= bdrv_open(bs1
, filename
, NULL
, 0, drv
);
844 total_size
= bdrv_getlength(bs1
) & BDRV_SECTOR_MASK
;
846 if (bs1
->drv
&& bs1
->drv
->protocol_name
)
851 ret
= get_tmp_filename(tmp_filename
, sizeof(tmp_filename
));
856 /* Real path is meaningless for protocols */
858 snprintf(backing_filename
, sizeof(backing_filename
),
860 } else if (!realpath(filename
, backing_filename
)) {
865 bdrv_qcow2
= bdrv_find_format("qcow2");
866 options
= parse_option_parameters("", bdrv_qcow2
->create_options
, NULL
);
868 set_option_parameter_int(options
, BLOCK_OPT_SIZE
, total_size
);
869 set_option_parameter(options
, BLOCK_OPT_BACKING_FILE
, backing_filename
);
871 set_option_parameter(options
, BLOCK_OPT_BACKING_FMT
,
875 ret
= bdrv_create(bdrv_qcow2
, tmp_filename
, options
);
876 free_option_parameters(options
);
881 filename
= tmp_filename
;
883 bs
->is_temporary
= 1;
886 /* Open image file without format layer */
887 if (flags
& BDRV_O_RDWR
) {
888 flags
|= BDRV_O_ALLOW_RDWR
;
891 ret
= bdrv_file_open(&file
, filename
, bdrv_open_flags(bs
, flags
));
896 /* Find the right image format driver */
898 ret
= find_image_format(file
, filename
, &drv
);
902 goto unlink_and_fail
;
906 ret
= bdrv_open_common(bs
, file
, filename
, flags
, drv
);
908 goto unlink_and_fail
;
911 if (bs
->file
!= file
) {
916 /* If there is a backing file, use it */
917 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
918 ret
= bdrv_open_backing_file(bs
);
925 if (!bdrv_key_required(bs
)) {
926 bdrv_dev_change_media_cb(bs
, true);
929 /* throttling disk I/O limits */
930 if (bs
->io_limits_enabled
) {
931 bdrv_io_limits_enable(bs
);
940 if (bs
->is_temporary
) {
944 QDECREF(bs
->options
);
950 typedef struct BlockReopenQueueEntry
{
952 BDRVReopenState state
;
953 QSIMPLEQ_ENTRY(BlockReopenQueueEntry
) entry
;
954 } BlockReopenQueueEntry
;
957 * Adds a BlockDriverState to a simple queue for an atomic, transactional
958 * reopen of multiple devices.
960 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
961 * already performed, or alternatively may be NULL a new BlockReopenQueue will
962 * be created and initialized. This newly created BlockReopenQueue should be
963 * passed back in for subsequent calls that are intended to be of the same
966 * bs is the BlockDriverState to add to the reopen queue.
968 * flags contains the open flags for the associated bs
970 * returns a pointer to bs_queue, which is either the newly allocated
971 * bs_queue, or the existing bs_queue being used.
974 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
975 BlockDriverState
*bs
, int flags
)
979 BlockReopenQueueEntry
*bs_entry
;
980 if (bs_queue
== NULL
) {
981 bs_queue
= g_new0(BlockReopenQueue
, 1);
982 QSIMPLEQ_INIT(bs_queue
);
986 bdrv_reopen_queue(bs_queue
, bs
->file
, flags
);
989 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
990 QSIMPLEQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
992 bs_entry
->state
.bs
= bs
;
993 bs_entry
->state
.flags
= flags
;
999 * Reopen multiple BlockDriverStates atomically & transactionally.
1001 * The queue passed in (bs_queue) must have been built up previous
1002 * via bdrv_reopen_queue().
1004 * Reopens all BDS specified in the queue, with the appropriate
1005 * flags. All devices are prepared for reopen, and failure of any
1006 * device will cause all device changes to be abandonded, and intermediate
1009 * If all devices prepare successfully, then the changes are committed
1013 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
1016 BlockReopenQueueEntry
*bs_entry
, *next
;
1017 Error
*local_err
= NULL
;
1019 assert(bs_queue
!= NULL
);
1023 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
1024 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, &local_err
)) {
1025 error_propagate(errp
, local_err
);
1028 bs_entry
->prepared
= true;
1031 /* If we reach this point, we have success and just need to apply the
1034 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
1035 bdrv_reopen_commit(&bs_entry
->state
);
1041 QSIMPLEQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
1042 if (ret
&& bs_entry
->prepared
) {
1043 bdrv_reopen_abort(&bs_entry
->state
);
1052 /* Reopen a single BlockDriverState with the specified flags. */
1053 int bdrv_reopen(BlockDriverState
*bs
, int bdrv_flags
, Error
**errp
)
1056 Error
*local_err
= NULL
;
1057 BlockReopenQueue
*queue
= bdrv_reopen_queue(NULL
, bs
, bdrv_flags
);
1059 ret
= bdrv_reopen_multiple(queue
, &local_err
);
1060 if (local_err
!= NULL
) {
1061 error_propagate(errp
, local_err
);
1068 * Prepares a BlockDriverState for reopen. All changes are staged in the
1069 * 'opaque' field of the BDRVReopenState, which is used and allocated by
1070 * the block driver layer .bdrv_reopen_prepare()
1072 * bs is the BlockDriverState to reopen
1073 * flags are the new open flags
1074 * queue is the reopen queue
1076 * Returns 0 on success, non-zero on error. On error errp will be set
1079 * On failure, bdrv_reopen_abort() will be called to clean up any data.
1080 * It is the responsibility of the caller to then call the abort() or
1081 * commit() for any other BDS that have been left in a prepare() state
1084 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
1088 Error
*local_err
= NULL
;
1091 assert(reopen_state
!= NULL
);
1092 assert(reopen_state
->bs
->drv
!= NULL
);
1093 drv
= reopen_state
->bs
->drv
;
1095 /* if we are to stay read-only, do not allow permission change
1097 if (!(reopen_state
->bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
1098 reopen_state
->flags
& BDRV_O_RDWR
) {
1099 error_set(errp
, QERR_DEVICE_IS_READ_ONLY
,
1100 reopen_state
->bs
->device_name
);
1105 ret
= bdrv_flush(reopen_state
->bs
);
1107 error_set(errp
, ERROR_CLASS_GENERIC_ERROR
, "Error (%s) flushing drive",
1112 if (drv
->bdrv_reopen_prepare
) {
1113 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
1115 if (local_err
!= NULL
) {
1116 error_propagate(errp
, local_err
);
1118 error_set(errp
, QERR_OPEN_FILE_FAILED
,
1119 reopen_state
->bs
->filename
);
1124 /* It is currently mandatory to have a bdrv_reopen_prepare()
1125 * handler for each supported drv. */
1126 error_set(errp
, QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED
,
1127 drv
->format_name
, reopen_state
->bs
->device_name
,
1128 "reopening of file");
1140 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
1141 * makes them final by swapping the staging BlockDriverState contents into
1142 * the active BlockDriverState contents.
1144 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
1148 assert(reopen_state
!= NULL
);
1149 drv
= reopen_state
->bs
->drv
;
1150 assert(drv
!= NULL
);
1152 /* If there are any driver level actions to take */
1153 if (drv
->bdrv_reopen_commit
) {
1154 drv
->bdrv_reopen_commit(reopen_state
);
1157 /* set BDS specific flags now */
1158 reopen_state
->bs
->open_flags
= reopen_state
->flags
;
1159 reopen_state
->bs
->enable_write_cache
= !!(reopen_state
->flags
&
1161 reopen_state
->bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
1165 * Abort the reopen, and delete and free the staged changes in
1168 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
1172 assert(reopen_state
!= NULL
);
1173 drv
= reopen_state
->bs
->drv
;
1174 assert(drv
!= NULL
);
1176 if (drv
->bdrv_reopen_abort
) {
1177 drv
->bdrv_reopen_abort(reopen_state
);
1182 void bdrv_close(BlockDriverState
*bs
)
1186 block_job_cancel_sync(bs
->job
);
1189 notifier_list_notify(&bs
->close_notifiers
, bs
);
1192 if (bs
== bs_snapshots
) {
1193 bs_snapshots
= NULL
;
1195 if (bs
->backing_hd
) {
1196 bdrv_delete(bs
->backing_hd
);
1197 bs
->backing_hd
= NULL
;
1199 bs
->drv
->bdrv_close(bs
);
1202 if (bs
->is_temporary
) {
1203 unlink(bs
->filename
);
1208 bs
->copy_on_read
= 0;
1209 bs
->backing_file
[0] = '\0';
1210 bs
->backing_format
[0] = '\0';
1211 bs
->total_sectors
= 0;
1216 QDECREF(bs
->options
);
1219 if (bs
->file
!= NULL
) {
1220 bdrv_delete(bs
->file
);
1225 bdrv_dev_change_media_cb(bs
, false);
1227 /*throttling disk I/O limits*/
1228 if (bs
->io_limits_enabled
) {
1229 bdrv_io_limits_disable(bs
);
1233 void bdrv_close_all(void)
1235 BlockDriverState
*bs
;
1237 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1243 * Wait for pending requests to complete across all BlockDriverStates
1245 * This function does not flush data to disk, use bdrv_flush_all() for that
1246 * after calling this function.
1248 * Note that completion of an asynchronous I/O operation can trigger any
1249 * number of other I/O operations on other devices---for example a coroutine
1250 * can be arbitrarily complex and a constant flow of I/O can come until the
1251 * coroutine is complete. Because of this, it is not possible to have a
1252 * function to drain a single device's I/O queue.
1254 void bdrv_drain_all(void)
1256 BlockDriverState
*bs
;
1260 busy
= qemu_aio_wait();
1262 /* FIXME: We do not have timer support here, so this is effectively
1265 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1266 if (!qemu_co_queue_empty(&bs
->throttled_reqs
)) {
1267 qemu_co_queue_restart_all(&bs
->throttled_reqs
);
1273 /* If requests are still pending there is a bug somewhere */
1274 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1275 assert(QLIST_EMPTY(&bs
->tracked_requests
));
1276 assert(qemu_co_queue_empty(&bs
->throttled_reqs
));
1280 /* make a BlockDriverState anonymous by removing from bdrv_state list.
1281 Also, NULL terminate the device_name to prevent double remove */
1282 void bdrv_make_anon(BlockDriverState
*bs
)
1284 if (bs
->device_name
[0] != '\0') {
1285 QTAILQ_REMOVE(&bdrv_states
, bs
, list
);
1287 bs
->device_name
[0] = '\0';
1290 static void bdrv_rebind(BlockDriverState
*bs
)
1292 if (bs
->drv
&& bs
->drv
->bdrv_rebind
) {
1293 bs
->drv
->bdrv_rebind(bs
);
1297 static void bdrv_move_feature_fields(BlockDriverState
*bs_dest
,
1298 BlockDriverState
*bs_src
)
1300 /* move some fields that need to stay attached to the device */
1301 bs_dest
->open_flags
= bs_src
->open_flags
;
1304 bs_dest
->dev_ops
= bs_src
->dev_ops
;
1305 bs_dest
->dev_opaque
= bs_src
->dev_opaque
;
1306 bs_dest
->dev
= bs_src
->dev
;
1307 bs_dest
->buffer_alignment
= bs_src
->buffer_alignment
;
1308 bs_dest
->copy_on_read
= bs_src
->copy_on_read
;
1310 bs_dest
->enable_write_cache
= bs_src
->enable_write_cache
;
1312 /* i/o timing parameters */
1313 bs_dest
->slice_time
= bs_src
->slice_time
;
1314 bs_dest
->slice_start
= bs_src
->slice_start
;
1315 bs_dest
->slice_end
= bs_src
->slice_end
;
1316 bs_dest
->io_limits
= bs_src
->io_limits
;
1317 bs_dest
->io_base
= bs_src
->io_base
;
1318 bs_dest
->throttled_reqs
= bs_src
->throttled_reqs
;
1319 bs_dest
->block_timer
= bs_src
->block_timer
;
1320 bs_dest
->io_limits_enabled
= bs_src
->io_limits_enabled
;
1323 bs_dest
->on_read_error
= bs_src
->on_read_error
;
1324 bs_dest
->on_write_error
= bs_src
->on_write_error
;
1327 bs_dest
->iostatus_enabled
= bs_src
->iostatus_enabled
;
1328 bs_dest
->iostatus
= bs_src
->iostatus
;
1331 bs_dest
->dirty_bitmap
= bs_src
->dirty_bitmap
;
1334 bs_dest
->in_use
= bs_src
->in_use
;
1335 bs_dest
->job
= bs_src
->job
;
1337 /* keep the same entry in bdrv_states */
1338 pstrcpy(bs_dest
->device_name
, sizeof(bs_dest
->device_name
),
1339 bs_src
->device_name
);
1340 bs_dest
->list
= bs_src
->list
;
1344 * Swap bs contents for two image chains while they are live,
1345 * while keeping required fields on the BlockDriverState that is
1346 * actually attached to a device.
1348 * This will modify the BlockDriverState fields, and swap contents
1349 * between bs_new and bs_old. Both bs_new and bs_old are modified.
1351 * bs_new is required to be anonymous.
1353 * This function does not create any image files.
1355 void bdrv_swap(BlockDriverState
*bs_new
, BlockDriverState
*bs_old
)
1357 BlockDriverState tmp
;
1359 /* bs_new must be anonymous and shouldn't have anything fancy enabled */
1360 assert(bs_new
->device_name
[0] == '\0');
1361 assert(bs_new
->dirty_bitmap
== NULL
);
1362 assert(bs_new
->job
== NULL
);
1363 assert(bs_new
->dev
== NULL
);
1364 assert(bs_new
->in_use
== 0);
1365 assert(bs_new
->io_limits_enabled
== false);
1366 assert(bs_new
->block_timer
== NULL
);
1372 /* there are some fields that should not be swapped, move them back */
1373 bdrv_move_feature_fields(&tmp
, bs_old
);
1374 bdrv_move_feature_fields(bs_old
, bs_new
);
1375 bdrv_move_feature_fields(bs_new
, &tmp
);
1377 /* bs_new shouldn't be in bdrv_states even after the swap! */
1378 assert(bs_new
->device_name
[0] == '\0');
1380 /* Check a few fields that should remain attached to the device */
1381 assert(bs_new
->dev
== NULL
);
1382 assert(bs_new
->job
== NULL
);
1383 assert(bs_new
->in_use
== 0);
1384 assert(bs_new
->io_limits_enabled
== false);
1385 assert(bs_new
->block_timer
== NULL
);
1387 bdrv_rebind(bs_new
);
1388 bdrv_rebind(bs_old
);
1392 * Add new bs contents at the top of an image chain while the chain is
1393 * live, while keeping required fields on the top layer.
1395 * This will modify the BlockDriverState fields, and swap contents
1396 * between bs_new and bs_top. Both bs_new and bs_top are modified.
1398 * bs_new is required to be anonymous.
1400 * This function does not create any image files.
1402 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
)
1404 bdrv_swap(bs_new
, bs_top
);
1406 /* The contents of 'tmp' will become bs_top, as we are
1407 * swapping bs_new and bs_top contents. */
1408 bs_top
->backing_hd
= bs_new
;
1409 bs_top
->open_flags
&= ~BDRV_O_NO_BACKING
;
1410 pstrcpy(bs_top
->backing_file
, sizeof(bs_top
->backing_file
),
1412 pstrcpy(bs_top
->backing_format
, sizeof(bs_top
->backing_format
),
1413 bs_new
->drv
? bs_new
->drv
->format_name
: "");
1416 void bdrv_delete(BlockDriverState
*bs
)
1420 assert(!bs
->in_use
);
1422 /* remove from list, if necessary */
1427 assert(bs
!= bs_snapshots
);
1431 int bdrv_attach_dev(BlockDriverState
*bs
, void *dev
)
1432 /* TODO change to DeviceState *dev when all users are qdevified */
1438 bdrv_iostatus_reset(bs
);
1442 /* TODO qdevified devices don't use this, remove when devices are qdevified */
1443 void bdrv_attach_dev_nofail(BlockDriverState
*bs
, void *dev
)
1445 if (bdrv_attach_dev(bs
, dev
) < 0) {
1450 void bdrv_detach_dev(BlockDriverState
*bs
, void *dev
)
1451 /* TODO change to DeviceState *dev when all users are qdevified */
1453 assert(bs
->dev
== dev
);
1456 bs
->dev_opaque
= NULL
;
1457 bs
->buffer_alignment
= 512;
1460 /* TODO change to return DeviceState * when all users are qdevified */
1461 void *bdrv_get_attached_dev(BlockDriverState
*bs
)
1466 void bdrv_set_dev_ops(BlockDriverState
*bs
, const BlockDevOps
*ops
,
1470 bs
->dev_opaque
= opaque
;
1471 if (bdrv_dev_has_removable_media(bs
) && bs
== bs_snapshots
) {
1472 bs_snapshots
= NULL
;
1476 void bdrv_emit_qmp_error_event(const BlockDriverState
*bdrv
,
1477 enum MonitorEvent ev
,
1478 BlockErrorAction action
, bool is_read
)
1481 const char *action_str
;
1484 case BDRV_ACTION_REPORT
:
1485 action_str
= "report";
1487 case BDRV_ACTION_IGNORE
:
1488 action_str
= "ignore";
1490 case BDRV_ACTION_STOP
:
1491 action_str
= "stop";
1497 data
= qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1500 is_read
? "read" : "write");
1501 monitor_protocol_event(ev
, data
);
1503 qobject_decref(data
);
1506 static void bdrv_emit_qmp_eject_event(BlockDriverState
*bs
, bool ejected
)
1510 data
= qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }",
1511 bdrv_get_device_name(bs
), ejected
);
1512 monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED
, data
);
1514 qobject_decref(data
);
1517 static void bdrv_dev_change_media_cb(BlockDriverState
*bs
, bool load
)
1519 if (bs
->dev_ops
&& bs
->dev_ops
->change_media_cb
) {
1520 bool tray_was_closed
= !bdrv_dev_is_tray_open(bs
);
1521 bs
->dev_ops
->change_media_cb(bs
->dev_opaque
, load
);
1522 if (tray_was_closed
) {
1524 bdrv_emit_qmp_eject_event(bs
, true);
1528 bdrv_emit_qmp_eject_event(bs
, false);
1533 bool bdrv_dev_has_removable_media(BlockDriverState
*bs
)
1535 return !bs
->dev
|| (bs
->dev_ops
&& bs
->dev_ops
->change_media_cb
);
1538 void bdrv_dev_eject_request(BlockDriverState
*bs
, bool force
)
1540 if (bs
->dev_ops
&& bs
->dev_ops
->eject_request_cb
) {
1541 bs
->dev_ops
->eject_request_cb(bs
->dev_opaque
, force
);
1545 bool bdrv_dev_is_tray_open(BlockDriverState
*bs
)
1547 if (bs
->dev_ops
&& bs
->dev_ops
->is_tray_open
) {
1548 return bs
->dev_ops
->is_tray_open(bs
->dev_opaque
);
1553 static void bdrv_dev_resize_cb(BlockDriverState
*bs
)
1555 if (bs
->dev_ops
&& bs
->dev_ops
->resize_cb
) {
1556 bs
->dev_ops
->resize_cb(bs
->dev_opaque
);
1560 bool bdrv_dev_is_medium_locked(BlockDriverState
*bs
)
1562 if (bs
->dev_ops
&& bs
->dev_ops
->is_medium_locked
) {
1563 return bs
->dev_ops
->is_medium_locked(bs
->dev_opaque
);
1569 * Run consistency checks on an image
1571 * Returns 0 if the check could be completed (it doesn't mean that the image is
1572 * free of errors) or -errno when an internal error occurred. The results of the
1573 * check are stored in res.
1575 int bdrv_check(BlockDriverState
*bs
, BdrvCheckResult
*res
, BdrvCheckMode fix
)
1577 if (bs
->drv
->bdrv_check
== NULL
) {
1581 memset(res
, 0, sizeof(*res
));
1582 return bs
->drv
->bdrv_check(bs
, res
, fix
);
1585 #define COMMIT_BUF_SECTORS 2048
1587 /* commit COW file into the raw image */
1588 int bdrv_commit(BlockDriverState
*bs
)
1590 BlockDriver
*drv
= bs
->drv
;
1591 int64_t sector
, total_sectors
;
1592 int n
, ro
, open_flags
;
1595 char filename
[PATH_MAX
];
1600 if (!bs
->backing_hd
) {
1604 if (bdrv_in_use(bs
) || bdrv_in_use(bs
->backing_hd
)) {
1608 ro
= bs
->backing_hd
->read_only
;
1609 /* Use pstrcpy (not strncpy): filename must be NUL-terminated. */
1610 pstrcpy(filename
, sizeof(filename
), bs
->backing_hd
->filename
);
1611 open_flags
= bs
->backing_hd
->open_flags
;
1614 if (bdrv_reopen(bs
->backing_hd
, open_flags
| BDRV_O_RDWR
, NULL
)) {
1619 total_sectors
= bdrv_getlength(bs
) >> BDRV_SECTOR_BITS
;
1620 buf
= g_malloc(COMMIT_BUF_SECTORS
* BDRV_SECTOR_SIZE
);
1622 for (sector
= 0; sector
< total_sectors
; sector
+= n
) {
1623 if (bdrv_is_allocated(bs
, sector
, COMMIT_BUF_SECTORS
, &n
)) {
1625 if (bdrv_read(bs
, sector
, buf
, n
) != 0) {
1630 if (bdrv_write(bs
->backing_hd
, sector
, buf
, n
) != 0) {
1637 if (drv
->bdrv_make_empty
) {
1638 ret
= drv
->bdrv_make_empty(bs
);
1643 * Make sure all data we wrote to the backing device is actually
1647 bdrv_flush(bs
->backing_hd
);
1653 /* ignoring error return here */
1654 bdrv_reopen(bs
->backing_hd
, open_flags
& ~BDRV_O_RDWR
, NULL
);
1660 int bdrv_commit_all(void)
1662 BlockDriverState
*bs
;
1664 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1665 if (bs
->drv
&& bs
->backing_hd
) {
1666 int ret
= bdrv_commit(bs
);
1675 struct BdrvTrackedRequest
{
1676 BlockDriverState
*bs
;
1680 QLIST_ENTRY(BdrvTrackedRequest
) list
;
1681 Coroutine
*co
; /* owner, used for deadlock detection */
1682 CoQueue wait_queue
; /* coroutines blocked on this request */
1686 * Remove an active request from the tracked requests list
1688 * This function should be called when a tracked request is completing.
1690 static void tracked_request_end(BdrvTrackedRequest
*req
)
1692 QLIST_REMOVE(req
, list
);
1693 qemu_co_queue_restart_all(&req
->wait_queue
);
1697 * Add an active request to the tracked requests list
1699 static void tracked_request_begin(BdrvTrackedRequest
*req
,
1700 BlockDriverState
*bs
,
1702 int nb_sectors
, bool is_write
)
1704 *req
= (BdrvTrackedRequest
){
1706 .sector_num
= sector_num
,
1707 .nb_sectors
= nb_sectors
,
1708 .is_write
= is_write
,
1709 .co
= qemu_coroutine_self(),
1712 qemu_co_queue_init(&req
->wait_queue
);
1714 QLIST_INSERT_HEAD(&bs
->tracked_requests
, req
, list
);
1718 * Round a region to cluster boundaries
1720 void bdrv_round_to_clusters(BlockDriverState
*bs
,
1721 int64_t sector_num
, int nb_sectors
,
1722 int64_t *cluster_sector_num
,
1723 int *cluster_nb_sectors
)
1725 BlockDriverInfo bdi
;
1727 if (bdrv_get_info(bs
, &bdi
) < 0 || bdi
.cluster_size
== 0) {
1728 *cluster_sector_num
= sector_num
;
1729 *cluster_nb_sectors
= nb_sectors
;
1731 int64_t c
= bdi
.cluster_size
/ BDRV_SECTOR_SIZE
;
1732 *cluster_sector_num
= QEMU_ALIGN_DOWN(sector_num
, c
);
1733 *cluster_nb_sectors
= QEMU_ALIGN_UP(sector_num
- *cluster_sector_num
+
1738 static bool tracked_request_overlaps(BdrvTrackedRequest
*req
,
1739 int64_t sector_num
, int nb_sectors
) {
1741 if (sector_num
>= req
->sector_num
+ req
->nb_sectors
) {
1745 if (req
->sector_num
>= sector_num
+ nb_sectors
) {
1751 static void coroutine_fn
wait_for_overlapping_requests(BlockDriverState
*bs
,
1752 int64_t sector_num
, int nb_sectors
)
1754 BdrvTrackedRequest
*req
;
1755 int64_t cluster_sector_num
;
1756 int cluster_nb_sectors
;
1759 /* If we touch the same cluster it counts as an overlap. This guarantees
1760 * that allocating writes will be serialized and not race with each other
1761 * for the same cluster. For example, in copy-on-read it ensures that the
1762 * CoR read and write operations are atomic and guest writes cannot
1763 * interleave between them.
1765 bdrv_round_to_clusters(bs
, sector_num
, nb_sectors
,
1766 &cluster_sector_num
, &cluster_nb_sectors
);
1770 QLIST_FOREACH(req
, &bs
->tracked_requests
, list
) {
1771 if (tracked_request_overlaps(req
, cluster_sector_num
,
1772 cluster_nb_sectors
)) {
1773 /* Hitting this means there was a reentrant request, for
1774 * example, a block driver issuing nested requests. This must
1775 * never happen since it means deadlock.
1777 assert(qemu_coroutine_self() != req
->co
);
1779 qemu_co_queue_wait(&req
->wait_queue
);
1790 * -EINVAL - backing format specified, but no file
1791 * -ENOSPC - can't update the backing file because no space is left in the
1793 * -ENOTSUP - format driver doesn't support changing the backing file
1795 int bdrv_change_backing_file(BlockDriverState
*bs
,
1796 const char *backing_file
, const char *backing_fmt
)
1798 BlockDriver
*drv
= bs
->drv
;
1801 /* Backing file format doesn't make sense without a backing file */
1802 if (backing_fmt
&& !backing_file
) {
1806 if (drv
->bdrv_change_backing_file
!= NULL
) {
1807 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
1813 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
1814 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
1820 * Finds the image layer in the chain that has 'bs' as its backing file.
1822 * active is the current topmost image.
1824 * Returns NULL if bs is not found in active's image chain,
1825 * or if active == bs.
1827 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
1828 BlockDriverState
*bs
)
1830 BlockDriverState
*overlay
= NULL
;
1831 BlockDriverState
*intermediate
;
1833 assert(active
!= NULL
);
1836 /* if bs is the same as active, then by definition it has no overlay
1842 intermediate
= active
;
1843 while (intermediate
->backing_hd
) {
1844 if (intermediate
->backing_hd
== bs
) {
1845 overlay
= intermediate
;
1848 intermediate
= intermediate
->backing_hd
;
1854 typedef struct BlkIntermediateStates
{
1855 BlockDriverState
*bs
;
1856 QSIMPLEQ_ENTRY(BlkIntermediateStates
) entry
;
1857 } BlkIntermediateStates
;
1861 * Drops images above 'base' up to and including 'top', and sets the image
1862 * above 'top' to have base as its backing file.
1864 * Requires that the overlay to 'top' is opened r/w, so that the backing file
1865 * information in 'bs' can be properly updated.
1867 * E.g., this will convert the following chain:
1868 * bottom <- base <- intermediate <- top <- active
1872 * bottom <- base <- active
1874 * It is allowed for bottom==base, in which case it converts:
1876 * base <- intermediate <- top <- active
1883 * if active == top, that is considered an error
1886 int bdrv_drop_intermediate(BlockDriverState
*active
, BlockDriverState
*top
,
1887 BlockDriverState
*base
)
1889 BlockDriverState
*intermediate
;
1890 BlockDriverState
*base_bs
= NULL
;
1891 BlockDriverState
*new_top_bs
= NULL
;
1892 BlkIntermediateStates
*intermediate_state
, *next
;
1895 QSIMPLEQ_HEAD(states_to_delete
, BlkIntermediateStates
) states_to_delete
;
1896 QSIMPLEQ_INIT(&states_to_delete
);
1898 if (!top
->drv
|| !base
->drv
) {
1902 new_top_bs
= bdrv_find_overlay(active
, top
);
1904 if (new_top_bs
== NULL
) {
1905 /* we could not find the image above 'top', this is an error */
1909 /* special case of new_top_bs->backing_hd already pointing to base - nothing
1910 * to do, no intermediate images */
1911 if (new_top_bs
->backing_hd
== base
) {
1918 /* now we will go down through the list, and add each BDS we find
1919 * into our deletion queue, until we hit the 'base'
1921 while (intermediate
) {
1922 intermediate_state
= g_malloc0(sizeof(BlkIntermediateStates
));
1923 intermediate_state
->bs
= intermediate
;
1924 QSIMPLEQ_INSERT_TAIL(&states_to_delete
, intermediate_state
, entry
);
1926 if (intermediate
->backing_hd
== base
) {
1927 base_bs
= intermediate
->backing_hd
;
1930 intermediate
= intermediate
->backing_hd
;
1932 if (base_bs
== NULL
) {
1933 /* something went wrong, we did not end at the base. safely
1934 * unravel everything, and exit with error */
1938 /* success - we can delete the intermediate states, and link top->base */
1939 ret
= bdrv_change_backing_file(new_top_bs
, base_bs
->filename
,
1940 base_bs
->drv
? base_bs
->drv
->format_name
: "");
1944 new_top_bs
->backing_hd
= base_bs
;
1947 QSIMPLEQ_FOREACH_SAFE(intermediate_state
, &states_to_delete
, entry
, next
) {
1948 /* so that bdrv_close() does not recursively close the chain */
1949 intermediate_state
->bs
->backing_hd
= NULL
;
1950 bdrv_delete(intermediate_state
->bs
);
1955 QSIMPLEQ_FOREACH_SAFE(intermediate_state
, &states_to_delete
, entry
, next
) {
1956 g_free(intermediate_state
);
1962 static int bdrv_check_byte_request(BlockDriverState
*bs
, int64_t offset
,
1967 if (!bdrv_is_inserted(bs
))
1973 len
= bdrv_getlength(bs
);
1978 if ((offset
> len
) || (len
- offset
< size
))
1984 static int bdrv_check_request(BlockDriverState
*bs
, int64_t sector_num
,
1987 return bdrv_check_byte_request(bs
, sector_num
* BDRV_SECTOR_SIZE
,
1988 nb_sectors
* BDRV_SECTOR_SIZE
);
1991 typedef struct RwCo
{
1992 BlockDriverState
*bs
;
2000 static void coroutine_fn
bdrv_rw_co_entry(void *opaque
)
2002 RwCo
*rwco
= opaque
;
2004 if (!rwco
->is_write
) {
2005 rwco
->ret
= bdrv_co_do_readv(rwco
->bs
, rwco
->sector_num
,
2006 rwco
->nb_sectors
, rwco
->qiov
, 0);
2008 rwco
->ret
= bdrv_co_do_writev(rwco
->bs
, rwco
->sector_num
,
2009 rwco
->nb_sectors
, rwco
->qiov
, 0);
2014 * Process a synchronous request using coroutines
2016 static int bdrv_rw_co(BlockDriverState
*bs
, int64_t sector_num
, uint8_t *buf
,
2017 int nb_sectors
, bool is_write
)
2020 struct iovec iov
= {
2021 .iov_base
= (void *)buf
,
2022 .iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
,
2027 .sector_num
= sector_num
,
2028 .nb_sectors
= nb_sectors
,
2030 .is_write
= is_write
,
2034 qemu_iovec_init_external(&qiov
, &iov
, 1);
2037 * In sync call context, when the vcpu is blocked, this throttling timer
2038 * will not fire; so the I/O throttling function has to be disabled here
2039 * if it has been enabled.
2041 if (bs
->io_limits_enabled
) {
2042 fprintf(stderr
, "Disabling I/O throttling on '%s' due "
2043 "to synchronous I/O.\n", bdrv_get_device_name(bs
));
2044 bdrv_io_limits_disable(bs
);
2047 if (qemu_in_coroutine()) {
2048 /* Fast-path if already in coroutine context */
2049 bdrv_rw_co_entry(&rwco
);
2051 co
= qemu_coroutine_create(bdrv_rw_co_entry
);
2052 qemu_coroutine_enter(co
, &rwco
);
2053 while (rwco
.ret
== NOT_DONE
) {
2060 /* return < 0 if error. See bdrv_write() for the return codes */
2061 int bdrv_read(BlockDriverState
*bs
, int64_t sector_num
,
2062 uint8_t *buf
, int nb_sectors
)
2064 return bdrv_rw_co(bs
, sector_num
, buf
, nb_sectors
, false);
2067 /* Just like bdrv_read(), but with I/O throttling temporarily disabled */
2068 int bdrv_read_unthrottled(BlockDriverState
*bs
, int64_t sector_num
,
2069 uint8_t *buf
, int nb_sectors
)
2074 enabled
= bs
->io_limits_enabled
;
2075 bs
->io_limits_enabled
= false;
2076 ret
= bdrv_read(bs
, 0, buf
, 1);
2077 bs
->io_limits_enabled
= enabled
;
2081 /* Return < 0 if error. Important errors are:
2082 -EIO generic I/O error (may happen for all errors)
2083 -ENOMEDIUM No media inserted.
2084 -EINVAL Invalid sector number or nb_sectors
2085 -EACCES Trying to write a read-only device
2087 int bdrv_write(BlockDriverState
*bs
, int64_t sector_num
,
2088 const uint8_t *buf
, int nb_sectors
)
2090 return bdrv_rw_co(bs
, sector_num
, (uint8_t *)buf
, nb_sectors
, true);
2093 int bdrv_pread(BlockDriverState
*bs
, int64_t offset
,
2094 void *buf
, int count1
)
2096 uint8_t tmp_buf
[BDRV_SECTOR_SIZE
];
2097 int len
, nb_sectors
, count
;
2102 /* first read to align to sector start */
2103 len
= (BDRV_SECTOR_SIZE
- offset
) & (BDRV_SECTOR_SIZE
- 1);
2106 sector_num
= offset
>> BDRV_SECTOR_BITS
;
2108 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
2110 memcpy(buf
, tmp_buf
+ (offset
& (BDRV_SECTOR_SIZE
- 1)), len
);
2118 /* read the sectors "in place" */
2119 nb_sectors
= count
>> BDRV_SECTOR_BITS
;
2120 if (nb_sectors
> 0) {
2121 if ((ret
= bdrv_read(bs
, sector_num
, buf
, nb_sectors
)) < 0)
2123 sector_num
+= nb_sectors
;
2124 len
= nb_sectors
<< BDRV_SECTOR_BITS
;
2129 /* add data from the last sector */
2131 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
2133 memcpy(buf
, tmp_buf
, count
);
2138 int bdrv_pwrite(BlockDriverState
*bs
, int64_t offset
,
2139 const void *buf
, int count1
)
2141 uint8_t tmp_buf
[BDRV_SECTOR_SIZE
];
2142 int len
, nb_sectors
, count
;
2147 /* first write to align to sector start */
2148 len
= (BDRV_SECTOR_SIZE
- offset
) & (BDRV_SECTOR_SIZE
- 1);
2151 sector_num
= offset
>> BDRV_SECTOR_BITS
;
2153 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
2155 memcpy(tmp_buf
+ (offset
& (BDRV_SECTOR_SIZE
- 1)), buf
, len
);
2156 if ((ret
= bdrv_write(bs
, sector_num
, tmp_buf
, 1)) < 0)
2165 /* write the sectors "in place" */
2166 nb_sectors
= count
>> BDRV_SECTOR_BITS
;
2167 if (nb_sectors
> 0) {
2168 if ((ret
= bdrv_write(bs
, sector_num
, buf
, nb_sectors
)) < 0)
2170 sector_num
+= nb_sectors
;
2171 len
= nb_sectors
<< BDRV_SECTOR_BITS
;
2176 /* add data from the last sector */
2178 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
2180 memcpy(tmp_buf
, buf
, count
);
2181 if ((ret
= bdrv_write(bs
, sector_num
, tmp_buf
, 1)) < 0)
2188 * Writes to the file and ensures that no writes are reordered across this
2189 * request (acts as a barrier)
2191 * Returns 0 on success, -errno in error cases.
2193 int bdrv_pwrite_sync(BlockDriverState
*bs
, int64_t offset
,
2194 const void *buf
, int count
)
2198 ret
= bdrv_pwrite(bs
, offset
, buf
, count
);
2203 /* No flush needed for cache modes that already do it */
2204 if (bs
->enable_write_cache
) {
2211 static int coroutine_fn
bdrv_co_do_copy_on_readv(BlockDriverState
*bs
,
2212 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
2214 /* Perform I/O through a temporary buffer so that users who scribble over
2215 * their read buffer while the operation is in progress do not end up
2216 * modifying the image file. This is critical for zero-copy guest I/O
2217 * where anything might happen inside guest memory.
2219 void *bounce_buffer
;
2221 BlockDriver
*drv
= bs
->drv
;
2223 QEMUIOVector bounce_qiov
;
2224 int64_t cluster_sector_num
;
2225 int cluster_nb_sectors
;
2229 /* Cover entire cluster so no additional backing file I/O is required when
2230 * allocating cluster in the image file.
2232 bdrv_round_to_clusters(bs
, sector_num
, nb_sectors
,
2233 &cluster_sector_num
, &cluster_nb_sectors
);
2235 trace_bdrv_co_do_copy_on_readv(bs
, sector_num
, nb_sectors
,
2236 cluster_sector_num
, cluster_nb_sectors
);
2238 iov
.iov_len
= cluster_nb_sectors
* BDRV_SECTOR_SIZE
;
2239 iov
.iov_base
= bounce_buffer
= qemu_blockalign(bs
, iov
.iov_len
);
2240 qemu_iovec_init_external(&bounce_qiov
, &iov
, 1);
2242 ret
= drv
->bdrv_co_readv(bs
, cluster_sector_num
, cluster_nb_sectors
,
2248 if (drv
->bdrv_co_write_zeroes
&&
2249 buffer_is_zero(bounce_buffer
, iov
.iov_len
)) {
2250 ret
= bdrv_co_do_write_zeroes(bs
, cluster_sector_num
,
2251 cluster_nb_sectors
);
2253 /* This does not change the data on the disk, it is not necessary
2254 * to flush even in cache=writethrough mode.
2256 ret
= drv
->bdrv_co_writev(bs
, cluster_sector_num
, cluster_nb_sectors
,
2261 /* It might be okay to ignore write errors for guest requests. If this
2262 * is a deliberate copy-on-read then we don't want to ignore the error.
2263 * Simply report it in all cases.
2268 skip_bytes
= (sector_num
- cluster_sector_num
) * BDRV_SECTOR_SIZE
;
2269 qemu_iovec_from_buf(qiov
, 0, bounce_buffer
+ skip_bytes
,
2270 nb_sectors
* BDRV_SECTOR_SIZE
);
2273 qemu_vfree(bounce_buffer
);
2278 * Handle a read request in coroutine context
2280 static int coroutine_fn
bdrv_co_do_readv(BlockDriverState
*bs
,
2281 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
2282 BdrvRequestFlags flags
)
2284 BlockDriver
*drv
= bs
->drv
;
2285 BdrvTrackedRequest req
;
2291 if (bdrv_check_request(bs
, sector_num
, nb_sectors
)) {
2295 /* throttling disk read I/O */
2296 if (bs
->io_limits_enabled
) {
2297 bdrv_io_limits_intercept(bs
, false, nb_sectors
);
2300 if (bs
->copy_on_read
) {
2301 flags
|= BDRV_REQ_COPY_ON_READ
;
2303 if (flags
& BDRV_REQ_COPY_ON_READ
) {
2304 bs
->copy_on_read_in_flight
++;
2307 if (bs
->copy_on_read_in_flight
) {
2308 wait_for_overlapping_requests(bs
, sector_num
, nb_sectors
);
2311 tracked_request_begin(&req
, bs
, sector_num
, nb_sectors
, false);
2313 if (flags
& BDRV_REQ_COPY_ON_READ
) {
2316 ret
= bdrv_co_is_allocated(bs
, sector_num
, nb_sectors
, &pnum
);
2321 if (!ret
|| pnum
!= nb_sectors
) {
2322 ret
= bdrv_co_do_copy_on_readv(bs
, sector_num
, nb_sectors
, qiov
);
2327 ret
= drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
2330 tracked_request_end(&req
);
2332 if (flags
& BDRV_REQ_COPY_ON_READ
) {
2333 bs
->copy_on_read_in_flight
--;
2339 int coroutine_fn
bdrv_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
2340 int nb_sectors
, QEMUIOVector
*qiov
)
2342 trace_bdrv_co_readv(bs
, sector_num
, nb_sectors
);
2344 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
, 0);
2347 int coroutine_fn
bdrv_co_copy_on_readv(BlockDriverState
*bs
,
2348 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
2350 trace_bdrv_co_copy_on_readv(bs
, sector_num
, nb_sectors
);
2352 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
,
2353 BDRV_REQ_COPY_ON_READ
);
2356 static int coroutine_fn
bdrv_co_do_write_zeroes(BlockDriverState
*bs
,
2357 int64_t sector_num
, int nb_sectors
)
2359 BlockDriver
*drv
= bs
->drv
;
2364 /* TODO Emulate only part of misaligned requests instead of letting block
2365 * drivers return -ENOTSUP and emulate everything */
2367 /* First try the efficient write zeroes operation */
2368 if (drv
->bdrv_co_write_zeroes
) {
2369 ret
= drv
->bdrv_co_write_zeroes(bs
, sector_num
, nb_sectors
);
2370 if (ret
!= -ENOTSUP
) {
2375 /* Fall back to bounce buffer if write zeroes is unsupported */
2376 iov
.iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
;
2377 iov
.iov_base
= qemu_blockalign(bs
, iov
.iov_len
);
2378 memset(iov
.iov_base
, 0, iov
.iov_len
);
2379 qemu_iovec_init_external(&qiov
, &iov
, 1);
2381 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, &qiov
);
2383 qemu_vfree(iov
.iov_base
);
2388 * Handle a write request in coroutine context
2390 static int coroutine_fn
bdrv_co_do_writev(BlockDriverState
*bs
,
2391 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
2392 BdrvRequestFlags flags
)
2394 BlockDriver
*drv
= bs
->drv
;
2395 BdrvTrackedRequest req
;
2401 if (bs
->read_only
) {
2404 if (bdrv_check_request(bs
, sector_num
, nb_sectors
)) {
2408 /* throttling disk write I/O */
2409 if (bs
->io_limits_enabled
) {
2410 bdrv_io_limits_intercept(bs
, true, nb_sectors
);
2413 if (bs
->copy_on_read_in_flight
) {
2414 wait_for_overlapping_requests(bs
, sector_num
, nb_sectors
);
2417 tracked_request_begin(&req
, bs
, sector_num
, nb_sectors
, true);
2419 if (flags
& BDRV_REQ_ZERO_WRITE
) {
2420 ret
= bdrv_co_do_write_zeroes(bs
, sector_num
, nb_sectors
);
2422 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, qiov
);
2425 if (ret
== 0 && !bs
->enable_write_cache
) {
2426 ret
= bdrv_co_flush(bs
);
2429 if (bs
->dirty_bitmap
) {
2430 bdrv_set_dirty(bs
, sector_num
, nb_sectors
);
2433 if (bs
->wr_highest_sector
< sector_num
+ nb_sectors
- 1) {
2434 bs
->wr_highest_sector
= sector_num
+ nb_sectors
- 1;
2437 tracked_request_end(&req
);
2442 int coroutine_fn
bdrv_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
2443 int nb_sectors
, QEMUIOVector
*qiov
)
2445 trace_bdrv_co_writev(bs
, sector_num
, nb_sectors
);
2447 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, qiov
, 0);
2450 int coroutine_fn
bdrv_co_write_zeroes(BlockDriverState
*bs
,
2451 int64_t sector_num
, int nb_sectors
)
2453 trace_bdrv_co_write_zeroes(bs
, sector_num
, nb_sectors
);
2455 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, NULL
,
2456 BDRV_REQ_ZERO_WRITE
);
2460 * Truncate file to 'offset' bytes (needed only for file protocols)
2462 int bdrv_truncate(BlockDriverState
*bs
, int64_t offset
)
2464 BlockDriver
*drv
= bs
->drv
;
2468 if (!drv
->bdrv_truncate
)
2472 if (bdrv_in_use(bs
))
2475 /* There better not be any in-flight IOs when we truncate the device. */
2478 ret
= drv
->bdrv_truncate(bs
, offset
);
2480 ret
= refresh_total_sectors(bs
, offset
>> BDRV_SECTOR_BITS
);
2481 bdrv_dev_resize_cb(bs
);
2487 * Length of a allocated file in bytes. Sparse files are counted by actual
2488 * allocated space. Return < 0 if error or unknown.
2490 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
2492 BlockDriver
*drv
= bs
->drv
;
2496 if (drv
->bdrv_get_allocated_file_size
) {
2497 return drv
->bdrv_get_allocated_file_size(bs
);
2500 return bdrv_get_allocated_file_size(bs
->file
);
2506 * Length of a file in bytes. Return < 0 if error or unknown.
2508 int64_t bdrv_getlength(BlockDriverState
*bs
)
2510 BlockDriver
*drv
= bs
->drv
;
2514 if (bs
->growable
|| bdrv_dev_has_removable_media(bs
)) {
2515 if (drv
->bdrv_getlength
) {
2516 return drv
->bdrv_getlength(bs
);
2519 return bs
->total_sectors
* BDRV_SECTOR_SIZE
;
2522 /* return 0 as number of sectors if no device present or error */
2523 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
2526 length
= bdrv_getlength(bs
);
2530 length
= length
>> BDRV_SECTOR_BITS
;
2531 *nb_sectors_ptr
= length
;
2534 /* throttling disk io limits */
2535 void bdrv_set_io_limits(BlockDriverState
*bs
,
2536 BlockIOLimit
*io_limits
)
2538 bs
->io_limits
= *io_limits
;
2539 bs
->io_limits_enabled
= bdrv_io_limits_enabled(bs
);
2542 void bdrv_set_on_error(BlockDriverState
*bs
, BlockdevOnError on_read_error
,
2543 BlockdevOnError on_write_error
)
2545 bs
->on_read_error
= on_read_error
;
2546 bs
->on_write_error
= on_write_error
;
2549 BlockdevOnError
bdrv_get_on_error(BlockDriverState
*bs
, bool is_read
)
2551 return is_read
? bs
->on_read_error
: bs
->on_write_error
;
2554 BlockErrorAction
bdrv_get_error_action(BlockDriverState
*bs
, bool is_read
, int error
)
2556 BlockdevOnError on_err
= is_read
? bs
->on_read_error
: bs
->on_write_error
;
2559 case BLOCKDEV_ON_ERROR_ENOSPC
:
2560 return (error
== ENOSPC
) ? BDRV_ACTION_STOP
: BDRV_ACTION_REPORT
;
2561 case BLOCKDEV_ON_ERROR_STOP
:
2562 return BDRV_ACTION_STOP
;
2563 case BLOCKDEV_ON_ERROR_REPORT
:
2564 return BDRV_ACTION_REPORT
;
2565 case BLOCKDEV_ON_ERROR_IGNORE
:
2566 return BDRV_ACTION_IGNORE
;
2572 /* This is done by device models because, while the block layer knows
2573 * about the error, it does not know whether an operation comes from
2574 * the device or the block layer (from a job, for example).
2576 void bdrv_error_action(BlockDriverState
*bs
, BlockErrorAction action
,
2577 bool is_read
, int error
)
2580 bdrv_emit_qmp_error_event(bs
, QEVENT_BLOCK_IO_ERROR
, action
, is_read
);
2581 if (action
== BDRV_ACTION_STOP
) {
2582 vm_stop(RUN_STATE_IO_ERROR
);
2583 bdrv_iostatus_set_err(bs
, error
);
2587 int bdrv_is_read_only(BlockDriverState
*bs
)
2589 return bs
->read_only
;
2592 int bdrv_is_sg(BlockDriverState
*bs
)
2597 int bdrv_enable_write_cache(BlockDriverState
*bs
)
2599 return bs
->enable_write_cache
;
2602 void bdrv_set_enable_write_cache(BlockDriverState
*bs
, bool wce
)
2604 bs
->enable_write_cache
= wce
;
2606 /* so a reopen() will preserve wce */
2608 bs
->open_flags
|= BDRV_O_CACHE_WB
;
2610 bs
->open_flags
&= ~BDRV_O_CACHE_WB
;
2614 int bdrv_is_encrypted(BlockDriverState
*bs
)
2616 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
)
2618 return bs
->encrypted
;
2621 int bdrv_key_required(BlockDriverState
*bs
)
2623 BlockDriverState
*backing_hd
= bs
->backing_hd
;
2625 if (backing_hd
&& backing_hd
->encrypted
&& !backing_hd
->valid_key
)
2627 return (bs
->encrypted
&& !bs
->valid_key
);
2630 int bdrv_set_key(BlockDriverState
*bs
, const char *key
)
2633 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
) {
2634 ret
= bdrv_set_key(bs
->backing_hd
, key
);
2640 if (!bs
->encrypted
) {
2642 } else if (!bs
->drv
|| !bs
->drv
->bdrv_set_key
) {
2645 ret
= bs
->drv
->bdrv_set_key(bs
, key
);
2648 } else if (!bs
->valid_key
) {
2650 /* call the change callback now, we skipped it on open */
2651 bdrv_dev_change_media_cb(bs
, true);
2656 const char *bdrv_get_format_name(BlockDriverState
*bs
)
2658 return bs
->drv
? bs
->drv
->format_name
: NULL
;
2661 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
2666 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
2667 it(opaque
, drv
->format_name
);
2671 BlockDriverState
*bdrv_find(const char *name
)
2673 BlockDriverState
*bs
;
2675 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2676 if (!strcmp(name
, bs
->device_name
)) {
2683 BlockDriverState
*bdrv_next(BlockDriverState
*bs
)
2686 return QTAILQ_FIRST(&bdrv_states
);
2688 return QTAILQ_NEXT(bs
, list
);
2691 void bdrv_iterate(void (*it
)(void *opaque
, BlockDriverState
*bs
), void *opaque
)
2693 BlockDriverState
*bs
;
2695 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2700 const char *bdrv_get_device_name(BlockDriverState
*bs
)
2702 return bs
->device_name
;
2705 int bdrv_get_flags(BlockDriverState
*bs
)
2707 return bs
->open_flags
;
2710 void bdrv_flush_all(void)
2712 BlockDriverState
*bs
;
2714 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2719 int bdrv_has_zero_init(BlockDriverState
*bs
)
2723 if (bs
->drv
->bdrv_has_zero_init
) {
2724 return bs
->drv
->bdrv_has_zero_init(bs
);
2730 typedef struct BdrvCoIsAllocatedData
{
2731 BlockDriverState
*bs
;
2732 BlockDriverState
*base
;
2738 } BdrvCoIsAllocatedData
;
2741 * Returns true iff the specified sector is present in the disk image. Drivers
2742 * not implementing the functionality are assumed to not support backing files,
2743 * hence all their sectors are reported as allocated.
2745 * If 'sector_num' is beyond the end of the disk image the return value is 0
2746 * and 'pnum' is set to 0.
2748 * 'pnum' is set to the number of sectors (including and immediately following
2749 * the specified sector) that are known to be in the same
2750 * allocated/unallocated state.
2752 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
2753 * beyond the end of the disk image it will be clamped.
2755 int coroutine_fn
bdrv_co_is_allocated(BlockDriverState
*bs
, int64_t sector_num
,
2756 int nb_sectors
, int *pnum
)
2760 if (sector_num
>= bs
->total_sectors
) {
2765 n
= bs
->total_sectors
- sector_num
;
2766 if (n
< nb_sectors
) {
2770 if (!bs
->drv
->bdrv_co_is_allocated
) {
2775 return bs
->drv
->bdrv_co_is_allocated(bs
, sector_num
, nb_sectors
, pnum
);
2778 /* Coroutine wrapper for bdrv_is_allocated() */
2779 static void coroutine_fn
bdrv_is_allocated_co_entry(void *opaque
)
2781 BdrvCoIsAllocatedData
*data
= opaque
;
2782 BlockDriverState
*bs
= data
->bs
;
2784 data
->ret
= bdrv_co_is_allocated(bs
, data
->sector_num
, data
->nb_sectors
,
2790 * Synchronous wrapper around bdrv_co_is_allocated().
2792 * See bdrv_co_is_allocated() for details.
2794 int bdrv_is_allocated(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
,
2798 BdrvCoIsAllocatedData data
= {
2800 .sector_num
= sector_num
,
2801 .nb_sectors
= nb_sectors
,
2806 co
= qemu_coroutine_create(bdrv_is_allocated_co_entry
);
2807 qemu_coroutine_enter(co
, &data
);
2808 while (!data
.done
) {
2815 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
2817 * Return true if the given sector is allocated in any image between
2818 * BASE and TOP (inclusive). BASE can be NULL to check if the given
2819 * sector is allocated in any image of the chain. Return false otherwise.
2821 * 'pnum' is set to the number of sectors (including and immediately following
2822 * the specified sector) that are known to be in the same
2823 * allocated/unallocated state.
2826 int coroutine_fn
bdrv_co_is_allocated_above(BlockDriverState
*top
,
2827 BlockDriverState
*base
,
2829 int nb_sectors
, int *pnum
)
2831 BlockDriverState
*intermediate
;
2832 int ret
, n
= nb_sectors
;
2835 while (intermediate
&& intermediate
!= base
) {
2837 ret
= bdrv_co_is_allocated(intermediate
, sector_num
, nb_sectors
,
2847 * [sector_num, nb_sectors] is unallocated on top but intermediate
2850 * [sector_num+x, nr_sectors] allocated.
2852 if (n
> pnum_inter
&&
2853 (intermediate
== top
||
2854 sector_num
+ pnum_inter
< intermediate
->total_sectors
)) {
2858 intermediate
= intermediate
->backing_hd
;
2865 /* Coroutine wrapper for bdrv_is_allocated_above() */
2866 static void coroutine_fn
bdrv_is_allocated_above_co_entry(void *opaque
)
2868 BdrvCoIsAllocatedData
*data
= opaque
;
2869 BlockDriverState
*top
= data
->bs
;
2870 BlockDriverState
*base
= data
->base
;
2872 data
->ret
= bdrv_co_is_allocated_above(top
, base
, data
->sector_num
,
2873 data
->nb_sectors
, data
->pnum
);
2878 * Synchronous wrapper around bdrv_co_is_allocated_above().
2880 * See bdrv_co_is_allocated_above() for details.
2882 int bdrv_is_allocated_above(BlockDriverState
*top
, BlockDriverState
*base
,
2883 int64_t sector_num
, int nb_sectors
, int *pnum
)
2886 BdrvCoIsAllocatedData data
= {
2889 .sector_num
= sector_num
,
2890 .nb_sectors
= nb_sectors
,
2895 co
= qemu_coroutine_create(bdrv_is_allocated_above_co_entry
);
2896 qemu_coroutine_enter(co
, &data
);
2897 while (!data
.done
) {
2903 BlockInfo
*bdrv_query_info(BlockDriverState
*bs
)
2905 BlockInfo
*info
= g_malloc0(sizeof(*info
));
2906 info
->device
= g_strdup(bs
->device_name
);
2907 info
->type
= g_strdup("unknown");
2908 info
->locked
= bdrv_dev_is_medium_locked(bs
);
2909 info
->removable
= bdrv_dev_has_removable_media(bs
);
2911 if (bdrv_dev_has_removable_media(bs
)) {
2912 info
->has_tray_open
= true;
2913 info
->tray_open
= bdrv_dev_is_tray_open(bs
);
2916 if (bdrv_iostatus_is_enabled(bs
)) {
2917 info
->has_io_status
= true;
2918 info
->io_status
= bs
->iostatus
;
2921 if (bs
->dirty_bitmap
) {
2922 info
->has_dirty
= true;
2923 info
->dirty
= g_malloc0(sizeof(*info
->dirty
));
2924 info
->dirty
->count
= bdrv_get_dirty_count(bs
) * BDRV_SECTOR_SIZE
;
2925 info
->dirty
->granularity
=
2926 ((int64_t) BDRV_SECTOR_SIZE
<< hbitmap_granularity(bs
->dirty_bitmap
));
2930 info
->has_inserted
= true;
2931 info
->inserted
= g_malloc0(sizeof(*info
->inserted
));
2932 info
->inserted
->file
= g_strdup(bs
->filename
);
2933 info
->inserted
->ro
= bs
->read_only
;
2934 info
->inserted
->drv
= g_strdup(bs
->drv
->format_name
);
2935 info
->inserted
->encrypted
= bs
->encrypted
;
2936 info
->inserted
->encryption_key_missing
= bdrv_key_required(bs
);
2938 if (bs
->backing_file
[0]) {
2939 info
->inserted
->has_backing_file
= true;
2940 info
->inserted
->backing_file
= g_strdup(bs
->backing_file
);
2943 info
->inserted
->backing_file_depth
= bdrv_get_backing_file_depth(bs
);
2945 if (bs
->io_limits_enabled
) {
2946 info
->inserted
->bps
=
2947 bs
->io_limits
.bps
[BLOCK_IO_LIMIT_TOTAL
];
2948 info
->inserted
->bps_rd
=
2949 bs
->io_limits
.bps
[BLOCK_IO_LIMIT_READ
];
2950 info
->inserted
->bps_wr
=
2951 bs
->io_limits
.bps
[BLOCK_IO_LIMIT_WRITE
];
2952 info
->inserted
->iops
=
2953 bs
->io_limits
.iops
[BLOCK_IO_LIMIT_TOTAL
];
2954 info
->inserted
->iops_rd
=
2955 bs
->io_limits
.iops
[BLOCK_IO_LIMIT_READ
];
2956 info
->inserted
->iops_wr
=
2957 bs
->io_limits
.iops
[BLOCK_IO_LIMIT_WRITE
];
2963 BlockInfoList
*qmp_query_block(Error
**errp
)
2965 BlockInfoList
*head
= NULL
, **p_next
= &head
;
2966 BlockDriverState
*bs
;
2968 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2969 BlockInfoList
*info
= g_malloc0(sizeof(*info
));
2970 info
->value
= bdrv_query_info(bs
);
2973 p_next
= &info
->next
;
2979 BlockStats
*bdrv_query_stats(const BlockDriverState
*bs
)
2983 s
= g_malloc0(sizeof(*s
));
2985 if (bs
->device_name
[0]) {
2986 s
->has_device
= true;
2987 s
->device
= g_strdup(bs
->device_name
);
2990 s
->stats
= g_malloc0(sizeof(*s
->stats
));
2991 s
->stats
->rd_bytes
= bs
->nr_bytes
[BDRV_ACCT_READ
];
2992 s
->stats
->wr_bytes
= bs
->nr_bytes
[BDRV_ACCT_WRITE
];
2993 s
->stats
->rd_operations
= bs
->nr_ops
[BDRV_ACCT_READ
];
2994 s
->stats
->wr_operations
= bs
->nr_ops
[BDRV_ACCT_WRITE
];
2995 s
->stats
->wr_highest_offset
= bs
->wr_highest_sector
* BDRV_SECTOR_SIZE
;
2996 s
->stats
->flush_operations
= bs
->nr_ops
[BDRV_ACCT_FLUSH
];
2997 s
->stats
->wr_total_time_ns
= bs
->total_time_ns
[BDRV_ACCT_WRITE
];
2998 s
->stats
->rd_total_time_ns
= bs
->total_time_ns
[BDRV_ACCT_READ
];
2999 s
->stats
->flush_total_time_ns
= bs
->total_time_ns
[BDRV_ACCT_FLUSH
];
3002 s
->has_parent
= true;
3003 s
->parent
= bdrv_query_stats(bs
->file
);
3009 BlockStatsList
*qmp_query_blockstats(Error
**errp
)
3011 BlockStatsList
*head
= NULL
, **p_next
= &head
;
3012 BlockDriverState
*bs
;
3014 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
3015 BlockStatsList
*info
= g_malloc0(sizeof(*info
));
3016 info
->value
= bdrv_query_stats(bs
);
3019 p_next
= &info
->next
;
3025 const char *bdrv_get_encrypted_filename(BlockDriverState
*bs
)
3027 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
)
3028 return bs
->backing_file
;
3029 else if (bs
->encrypted
)
3030 return bs
->filename
;
3035 void bdrv_get_backing_filename(BlockDriverState
*bs
,
3036 char *filename
, int filename_size
)
3038 pstrcpy(filename
, filename_size
, bs
->backing_file
);
3041 int bdrv_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
3042 const uint8_t *buf
, int nb_sectors
)
3044 BlockDriver
*drv
= bs
->drv
;
3047 if (!drv
->bdrv_write_compressed
)
3049 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
3052 assert(!bs
->dirty_bitmap
);
3054 return drv
->bdrv_write_compressed(bs
, sector_num
, buf
, nb_sectors
);
3057 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
3059 BlockDriver
*drv
= bs
->drv
;
3062 if (!drv
->bdrv_get_info
)
3064 memset(bdi
, 0, sizeof(*bdi
));
3065 return drv
->bdrv_get_info(bs
, bdi
);
3068 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
3069 int64_t pos
, int size
)
3071 BlockDriver
*drv
= bs
->drv
;
3074 if (drv
->bdrv_save_vmstate
)
3075 return drv
->bdrv_save_vmstate(bs
, buf
, pos
, size
);
3077 return bdrv_save_vmstate(bs
->file
, buf
, pos
, size
);
3081 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
3082 int64_t pos
, int size
)
3084 BlockDriver
*drv
= bs
->drv
;
3087 if (drv
->bdrv_load_vmstate
)
3088 return drv
->bdrv_load_vmstate(bs
, buf
, pos
, size
);
3090 return bdrv_load_vmstate(bs
->file
, buf
, pos
, size
);
3094 void bdrv_debug_event(BlockDriverState
*bs
, BlkDebugEvent event
)
3096 BlockDriver
*drv
= bs
->drv
;
3098 if (!drv
|| !drv
->bdrv_debug_event
) {
3102 drv
->bdrv_debug_event(bs
, event
);
3105 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
3108 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
3112 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
3113 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
3119 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
3121 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_resume
) {
3125 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
3126 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
3132 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
3134 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
3138 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
3139 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
3145 /**************************************************************/
3146 /* handling of snapshots */
3148 int bdrv_can_snapshot(BlockDriverState
*bs
)
3150 BlockDriver
*drv
= bs
->drv
;
3151 if (!drv
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
)) {
3155 if (!drv
->bdrv_snapshot_create
) {
3156 if (bs
->file
!= NULL
) {
3157 return bdrv_can_snapshot(bs
->file
);
3165 int bdrv_is_snapshot(BlockDriverState
*bs
)
3167 return !!(bs
->open_flags
& BDRV_O_SNAPSHOT
);
3170 BlockDriverState
*bdrv_snapshots(void)
3172 BlockDriverState
*bs
;
3175 return bs_snapshots
;
3179 while ((bs
= bdrv_next(bs
))) {
3180 if (bdrv_can_snapshot(bs
)) {
3188 int bdrv_snapshot_create(BlockDriverState
*bs
,
3189 QEMUSnapshotInfo
*sn_info
)
3191 BlockDriver
*drv
= bs
->drv
;
3194 if (drv
->bdrv_snapshot_create
)
3195 return drv
->bdrv_snapshot_create(bs
, sn_info
);
3197 return bdrv_snapshot_create(bs
->file
, sn_info
);
3201 int bdrv_snapshot_goto(BlockDriverState
*bs
,
3202 const char *snapshot_id
)
3204 BlockDriver
*drv
= bs
->drv
;
3209 if (drv
->bdrv_snapshot_goto
)
3210 return drv
->bdrv_snapshot_goto(bs
, snapshot_id
);
3213 drv
->bdrv_close(bs
);
3214 ret
= bdrv_snapshot_goto(bs
->file
, snapshot_id
);
3215 open_ret
= drv
->bdrv_open(bs
, NULL
, bs
->open_flags
);
3217 bdrv_delete(bs
->file
);
3227 int bdrv_snapshot_delete(BlockDriverState
*bs
, const char *snapshot_id
)
3229 BlockDriver
*drv
= bs
->drv
;
3232 if (drv
->bdrv_snapshot_delete
)
3233 return drv
->bdrv_snapshot_delete(bs
, snapshot_id
);
3235 return bdrv_snapshot_delete(bs
->file
, snapshot_id
);
3239 int bdrv_snapshot_list(BlockDriverState
*bs
,
3240 QEMUSnapshotInfo
**psn_info
)
3242 BlockDriver
*drv
= bs
->drv
;
3245 if (drv
->bdrv_snapshot_list
)
3246 return drv
->bdrv_snapshot_list(bs
, psn_info
);
3248 return bdrv_snapshot_list(bs
->file
, psn_info
);
3252 int bdrv_snapshot_load_tmp(BlockDriverState
*bs
,
3253 const char *snapshot_name
)
3255 BlockDriver
*drv
= bs
->drv
;
3259 if (!bs
->read_only
) {
3262 if (drv
->bdrv_snapshot_load_tmp
) {
3263 return drv
->bdrv_snapshot_load_tmp(bs
, snapshot_name
);
3268 /* backing_file can either be relative, or absolute, or a protocol. If it is
3269 * relative, it must be relative to the chain. So, passing in bs->filename
3270 * from a BDS as backing_file should not be done, as that may be relative to
3271 * the CWD rather than the chain. */
3272 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
3273 const char *backing_file
)
3275 char *filename_full
= NULL
;
3276 char *backing_file_full
= NULL
;
3277 char *filename_tmp
= NULL
;
3278 int is_protocol
= 0;
3279 BlockDriverState
*curr_bs
= NULL
;
3280 BlockDriverState
*retval
= NULL
;
3282 if (!bs
|| !bs
->drv
|| !backing_file
) {
3286 filename_full
= g_malloc(PATH_MAX
);
3287 backing_file_full
= g_malloc(PATH_MAX
);
3288 filename_tmp
= g_malloc(PATH_MAX
);
3290 is_protocol
= path_has_protocol(backing_file
);
3292 for (curr_bs
= bs
; curr_bs
->backing_hd
; curr_bs
= curr_bs
->backing_hd
) {
3294 /* If either of the filename paths is actually a protocol, then
3295 * compare unmodified paths; otherwise make paths relative */
3296 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
3297 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
3298 retval
= curr_bs
->backing_hd
;
3302 /* If not an absolute filename path, make it relative to the current
3303 * image's filename path */
3304 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
3307 /* We are going to compare absolute pathnames */
3308 if (!realpath(filename_tmp
, filename_full
)) {
3312 /* We need to make sure the backing filename we are comparing against
3313 * is relative to the current image filename (or absolute) */
3314 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
3315 curr_bs
->backing_file
);
3317 if (!realpath(filename_tmp
, backing_file_full
)) {
3321 if (strcmp(backing_file_full
, filename_full
) == 0) {
3322 retval
= curr_bs
->backing_hd
;
3328 g_free(filename_full
);
3329 g_free(backing_file_full
);
3330 g_free(filename_tmp
);
3334 int bdrv_get_backing_file_depth(BlockDriverState
*bs
)
3340 if (!bs
->backing_hd
) {
3344 return 1 + bdrv_get_backing_file_depth(bs
->backing_hd
);
3347 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
3349 BlockDriverState
*curr_bs
= NULL
;
3357 while (curr_bs
->backing_hd
) {
3358 curr_bs
= curr_bs
->backing_hd
;
3363 #define NB_SUFFIXES 4
3365 char *get_human_readable_size(char *buf
, int buf_size
, int64_t size
)
3367 static const char suffixes
[NB_SUFFIXES
] = "KMGT";
3372 snprintf(buf
, buf_size
, "%" PRId64
, size
);
3375 for(i
= 0; i
< NB_SUFFIXES
; i
++) {
3376 if (size
< (10 * base
)) {
3377 snprintf(buf
, buf_size
, "%0.1f%c",
3378 (double)size
/ base
,
3381 } else if (size
< (1000 * base
) || i
== (NB_SUFFIXES
- 1)) {
3382 snprintf(buf
, buf_size
, "%" PRId64
"%c",
3383 ((size
+ (base
>> 1)) / base
),
3393 char *bdrv_snapshot_dump(char *buf
, int buf_size
, QEMUSnapshotInfo
*sn
)
3395 char buf1
[128], date_buf
[128], clock_buf
[128];
3401 snprintf(buf
, buf_size
,
3402 "%-10s%-20s%7s%20s%15s",
3403 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
3406 localtime_r(&ti
, &tm
);
3407 strftime(date_buf
, sizeof(date_buf
),
3408 "%Y-%m-%d %H:%M:%S", &tm
);
3409 secs
= sn
->vm_clock_nsec
/ 1000000000;
3410 snprintf(clock_buf
, sizeof(clock_buf
),
3411 "%02d:%02d:%02d.%03d",
3413 (int)((secs
/ 60) % 60),
3415 (int)((sn
->vm_clock_nsec
/ 1000000) % 1000));
3416 snprintf(buf
, buf_size
,
3417 "%-10s%-20s%7s%20s%15s",
3418 sn
->id_str
, sn
->name
,
3419 get_human_readable_size(buf1
, sizeof(buf1
), sn
->vm_state_size
),
3426 /**************************************************************/
3429 BlockDriverAIOCB
*bdrv_aio_readv(BlockDriverState
*bs
, int64_t sector_num
,
3430 QEMUIOVector
*qiov
, int nb_sectors
,
3431 BlockDriverCompletionFunc
*cb
, void *opaque
)
3433 trace_bdrv_aio_readv(bs
, sector_num
, nb_sectors
, opaque
);
3435 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
,
3439 BlockDriverAIOCB
*bdrv_aio_writev(BlockDriverState
*bs
, int64_t sector_num
,
3440 QEMUIOVector
*qiov
, int nb_sectors
,
3441 BlockDriverCompletionFunc
*cb
, void *opaque
)
3443 trace_bdrv_aio_writev(bs
, sector_num
, nb_sectors
, opaque
);
3445 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
,
3450 typedef struct MultiwriteCB
{
3455 BlockDriverCompletionFunc
*cb
;
3457 QEMUIOVector
*free_qiov
;
3461 static void multiwrite_user_cb(MultiwriteCB
*mcb
)
3465 for (i
= 0; i
< mcb
->num_callbacks
; i
++) {
3466 mcb
->callbacks
[i
].cb(mcb
->callbacks
[i
].opaque
, mcb
->error
);
3467 if (mcb
->callbacks
[i
].free_qiov
) {
3468 qemu_iovec_destroy(mcb
->callbacks
[i
].free_qiov
);
3470 g_free(mcb
->callbacks
[i
].free_qiov
);
3474 static void multiwrite_cb(void *opaque
, int ret
)
3476 MultiwriteCB
*mcb
= opaque
;
3478 trace_multiwrite_cb(mcb
, ret
);
3480 if (ret
< 0 && !mcb
->error
) {
3484 mcb
->num_requests
--;
3485 if (mcb
->num_requests
== 0) {
3486 multiwrite_user_cb(mcb
);
3491 static int multiwrite_req_compare(const void *a
, const void *b
)
3493 const BlockRequest
*req1
= a
, *req2
= b
;
3496 * Note that we can't simply subtract req2->sector from req1->sector
3497 * here as that could overflow the return value.
3499 if (req1
->sector
> req2
->sector
) {
3501 } else if (req1
->sector
< req2
->sector
) {
3509 * Takes a bunch of requests and tries to merge them. Returns the number of
3510 * requests that remain after merging.
3512 static int multiwrite_merge(BlockDriverState
*bs
, BlockRequest
*reqs
,
3513 int num_reqs
, MultiwriteCB
*mcb
)
3517 // Sort requests by start sector
3518 qsort(reqs
, num_reqs
, sizeof(*reqs
), &multiwrite_req_compare
);
3520 // Check if adjacent requests touch the same clusters. If so, combine them,
3521 // filling up gaps with zero sectors.
3523 for (i
= 1; i
< num_reqs
; i
++) {
3525 int64_t oldreq_last
= reqs
[outidx
].sector
+ reqs
[outidx
].nb_sectors
;
3527 // Handle exactly sequential writes and overlapping writes.
3528 if (reqs
[i
].sector
<= oldreq_last
) {
3532 if (reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1 > IOV_MAX
) {
3538 QEMUIOVector
*qiov
= g_malloc0(sizeof(*qiov
));
3539 qemu_iovec_init(qiov
,
3540 reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1);
3542 // Add the first request to the merged one. If the requests are
3543 // overlapping, drop the last sectors of the first request.
3544 size
= (reqs
[i
].sector
- reqs
[outidx
].sector
) << 9;
3545 qemu_iovec_concat(qiov
, reqs
[outidx
].qiov
, 0, size
);
3547 // We should need to add any zeros between the two requests
3548 assert (reqs
[i
].sector
<= oldreq_last
);
3550 // Add the second request
3551 qemu_iovec_concat(qiov
, reqs
[i
].qiov
, 0, reqs
[i
].qiov
->size
);
3553 reqs
[outidx
].nb_sectors
= qiov
->size
>> 9;
3554 reqs
[outidx
].qiov
= qiov
;
3556 mcb
->callbacks
[i
].free_qiov
= reqs
[outidx
].qiov
;
3559 reqs
[outidx
].sector
= reqs
[i
].sector
;
3560 reqs
[outidx
].nb_sectors
= reqs
[i
].nb_sectors
;
3561 reqs
[outidx
].qiov
= reqs
[i
].qiov
;
3569 * Submit multiple AIO write requests at once.
3571 * On success, the function returns 0 and all requests in the reqs array have
3572 * been submitted. In error case this function returns -1, and any of the
3573 * requests may or may not be submitted yet. In particular, this means that the
3574 * callback will be called for some of the requests, for others it won't. The
3575 * caller must check the error field of the BlockRequest to wait for the right
3576 * callbacks (if error != 0, no callback will be called).
3578 * The implementation may modify the contents of the reqs array, e.g. to merge
3579 * requests. However, the fields opaque and error are left unmodified as they
3580 * are used to signal failure for a single request to the caller.
3582 int bdrv_aio_multiwrite(BlockDriverState
*bs
, BlockRequest
*reqs
, int num_reqs
)
3587 /* don't submit writes if we don't have a medium */
3588 if (bs
->drv
== NULL
) {
3589 for (i
= 0; i
< num_reqs
; i
++) {
3590 reqs
[i
].error
= -ENOMEDIUM
;
3595 if (num_reqs
== 0) {
3599 // Create MultiwriteCB structure
3600 mcb
= g_malloc0(sizeof(*mcb
) + num_reqs
* sizeof(*mcb
->callbacks
));
3601 mcb
->num_requests
= 0;
3602 mcb
->num_callbacks
= num_reqs
;
3604 for (i
= 0; i
< num_reqs
; i
++) {
3605 mcb
->callbacks
[i
].cb
= reqs
[i
].cb
;
3606 mcb
->callbacks
[i
].opaque
= reqs
[i
].opaque
;
3609 // Check for mergable requests
3610 num_reqs
= multiwrite_merge(bs
, reqs
, num_reqs
, mcb
);
3612 trace_bdrv_aio_multiwrite(mcb
, mcb
->num_callbacks
, num_reqs
);
3614 /* Run the aio requests. */
3615 mcb
->num_requests
= num_reqs
;
3616 for (i
= 0; i
< num_reqs
; i
++) {
3617 bdrv_aio_writev(bs
, reqs
[i
].sector
, reqs
[i
].qiov
,
3618 reqs
[i
].nb_sectors
, multiwrite_cb
, mcb
);
3624 void bdrv_aio_cancel(BlockDriverAIOCB
*acb
)
3626 acb
->aiocb_info
->cancel(acb
);
3629 /* block I/O throttling */
3630 static bool bdrv_exceed_bps_limits(BlockDriverState
*bs
, int nb_sectors
,
3631 bool is_write
, double elapsed_time
, uint64_t *wait
)
3633 uint64_t bps_limit
= 0;
3634 double bytes_limit
, bytes_base
, bytes_res
;
3635 double slice_time
, wait_time
;
3637 if (bs
->io_limits
.bps
[BLOCK_IO_LIMIT_TOTAL
]) {
3638 bps_limit
= bs
->io_limits
.bps
[BLOCK_IO_LIMIT_TOTAL
];
3639 } else if (bs
->io_limits
.bps
[is_write
]) {
3640 bps_limit
= bs
->io_limits
.bps
[is_write
];
3649 slice_time
= bs
->slice_end
- bs
->slice_start
;
3650 slice_time
/= (NANOSECONDS_PER_SECOND
);
3651 bytes_limit
= bps_limit
* slice_time
;
3652 bytes_base
= bs
->nr_bytes
[is_write
] - bs
->io_base
.bytes
[is_write
];
3653 if (bs
->io_limits
.bps
[BLOCK_IO_LIMIT_TOTAL
]) {
3654 bytes_base
+= bs
->nr_bytes
[!is_write
] - bs
->io_base
.bytes
[!is_write
];
3657 /* bytes_base: the bytes of data which have been read/written; and
3658 * it is obtained from the history statistic info.
3659 * bytes_res: the remaining bytes of data which need to be read/written.
3660 * (bytes_base + bytes_res) / bps_limit: used to calcuate
3661 * the total time for completing reading/writting all data.
3663 bytes_res
= (unsigned) nb_sectors
* BDRV_SECTOR_SIZE
;
3665 if (bytes_base
+ bytes_res
<= bytes_limit
) {
3673 /* Calc approx time to dispatch */
3674 wait_time
= (bytes_base
+ bytes_res
) / bps_limit
- elapsed_time
;
3676 /* When the I/O rate at runtime exceeds the limits,
3677 * bs->slice_end need to be extended in order that the current statistic
3678 * info can be kept until the timer fire, so it is increased and tuned
3679 * based on the result of experiment.
3681 bs
->slice_time
= wait_time
* BLOCK_IO_SLICE_TIME
* 10;
3682 bs
->slice_end
+= bs
->slice_time
- 3 * BLOCK_IO_SLICE_TIME
;
3684 *wait
= wait_time
* BLOCK_IO_SLICE_TIME
* 10;
3690 static bool bdrv_exceed_iops_limits(BlockDriverState
*bs
, bool is_write
,
3691 double elapsed_time
, uint64_t *wait
)
3693 uint64_t iops_limit
= 0;
3694 double ios_limit
, ios_base
;
3695 double slice_time
, wait_time
;
3697 if (bs
->io_limits
.iops
[BLOCK_IO_LIMIT_TOTAL
]) {
3698 iops_limit
= bs
->io_limits
.iops
[BLOCK_IO_LIMIT_TOTAL
];
3699 } else if (bs
->io_limits
.iops
[is_write
]) {
3700 iops_limit
= bs
->io_limits
.iops
[is_write
];
3709 slice_time
= bs
->slice_end
- bs
->slice_start
;
3710 slice_time
/= (NANOSECONDS_PER_SECOND
);
3711 ios_limit
= iops_limit
* slice_time
;
3712 ios_base
= bs
->nr_ops
[is_write
] - bs
->io_base
.ios
[is_write
];
3713 if (bs
->io_limits
.iops
[BLOCK_IO_LIMIT_TOTAL
]) {
3714 ios_base
+= bs
->nr_ops
[!is_write
] - bs
->io_base
.ios
[!is_write
];
3717 if (ios_base
+ 1 <= ios_limit
) {
3725 /* Calc approx time to dispatch */
3726 wait_time
= (ios_base
+ 1) / iops_limit
;
3727 if (wait_time
> elapsed_time
) {
3728 wait_time
= wait_time
- elapsed_time
;
3733 bs
->slice_time
= wait_time
* BLOCK_IO_SLICE_TIME
* 10;
3734 bs
->slice_end
+= bs
->slice_time
- 3 * BLOCK_IO_SLICE_TIME
;
3736 *wait
= wait_time
* BLOCK_IO_SLICE_TIME
* 10;
3742 static bool bdrv_exceed_io_limits(BlockDriverState
*bs
, int nb_sectors
,
3743 bool is_write
, int64_t *wait
)
3745 int64_t now
, max_wait
;
3746 uint64_t bps_wait
= 0, iops_wait
= 0;
3747 double elapsed_time
;
3748 int bps_ret
, iops_ret
;
3750 now
= qemu_get_clock_ns(vm_clock
);
3751 if ((bs
->slice_start
< now
)
3752 && (bs
->slice_end
> now
)) {
3753 bs
->slice_end
= now
+ bs
->slice_time
;
3755 bs
->slice_time
= 5 * BLOCK_IO_SLICE_TIME
;
3756 bs
->slice_start
= now
;
3757 bs
->slice_end
= now
+ bs
->slice_time
;
3759 bs
->io_base
.bytes
[is_write
] = bs
->nr_bytes
[is_write
];
3760 bs
->io_base
.bytes
[!is_write
] = bs
->nr_bytes
[!is_write
];
3762 bs
->io_base
.ios
[is_write
] = bs
->nr_ops
[is_write
];
3763 bs
->io_base
.ios
[!is_write
] = bs
->nr_ops
[!is_write
];
3766 elapsed_time
= now
- bs
->slice_start
;
3767 elapsed_time
/= (NANOSECONDS_PER_SECOND
);
3769 bps_ret
= bdrv_exceed_bps_limits(bs
, nb_sectors
,
3770 is_write
, elapsed_time
, &bps_wait
);
3771 iops_ret
= bdrv_exceed_iops_limits(bs
, is_write
,
3772 elapsed_time
, &iops_wait
);
3773 if (bps_ret
|| iops_ret
) {
3774 max_wait
= bps_wait
> iops_wait
? bps_wait
: iops_wait
;
3779 now
= qemu_get_clock_ns(vm_clock
);
3780 if (bs
->slice_end
< now
+ max_wait
) {
3781 bs
->slice_end
= now
+ max_wait
;
3794 /**************************************************************/
3795 /* async block device emulation */
3797 typedef struct BlockDriverAIOCBSync
{
3798 BlockDriverAIOCB common
;
3801 /* vector translation state */
3805 } BlockDriverAIOCBSync
;
3807 static void bdrv_aio_cancel_em(BlockDriverAIOCB
*blockacb
)
3809 BlockDriverAIOCBSync
*acb
=
3810 container_of(blockacb
, BlockDriverAIOCBSync
, common
);
3811 qemu_bh_delete(acb
->bh
);
3813 qemu_aio_release(acb
);
3816 static const AIOCBInfo bdrv_em_aiocb_info
= {
3817 .aiocb_size
= sizeof(BlockDriverAIOCBSync
),
3818 .cancel
= bdrv_aio_cancel_em
,
3821 static void bdrv_aio_bh_cb(void *opaque
)
3823 BlockDriverAIOCBSync
*acb
= opaque
;
3826 qemu_iovec_from_buf(acb
->qiov
, 0, acb
->bounce
, acb
->qiov
->size
);
3827 qemu_vfree(acb
->bounce
);
3828 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
3829 qemu_bh_delete(acb
->bh
);
3831 qemu_aio_release(acb
);
3834 static BlockDriverAIOCB
*bdrv_aio_rw_vector(BlockDriverState
*bs
,
3838 BlockDriverCompletionFunc
*cb
,
3843 BlockDriverAIOCBSync
*acb
;
3845 acb
= qemu_aio_get(&bdrv_em_aiocb_info
, bs
, cb
, opaque
);
3846 acb
->is_write
= is_write
;
3848 acb
->bounce
= qemu_blockalign(bs
, qiov
->size
);
3849 acb
->bh
= qemu_bh_new(bdrv_aio_bh_cb
, acb
);
3852 qemu_iovec_to_buf(acb
->qiov
, 0, acb
->bounce
, qiov
->size
);
3853 acb
->ret
= bs
->drv
->bdrv_write(bs
, sector_num
, acb
->bounce
, nb_sectors
);
3855 acb
->ret
= bs
->drv
->bdrv_read(bs
, sector_num
, acb
->bounce
, nb_sectors
);
3858 qemu_bh_schedule(acb
->bh
);
3860 return &acb
->common
;
3863 static BlockDriverAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
3864 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
3865 BlockDriverCompletionFunc
*cb
, void *opaque
)
3867 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 0);
3870 static BlockDriverAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
3871 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
3872 BlockDriverCompletionFunc
*cb
, void *opaque
)
3874 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 1);
3878 typedef struct BlockDriverAIOCBCoroutine
{
3879 BlockDriverAIOCB common
;
3884 } BlockDriverAIOCBCoroutine
;
3886 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB
*blockacb
)
3888 BlockDriverAIOCBCoroutine
*acb
=
3889 container_of(blockacb
, BlockDriverAIOCBCoroutine
, common
);
3898 static const AIOCBInfo bdrv_em_co_aiocb_info
= {
3899 .aiocb_size
= sizeof(BlockDriverAIOCBCoroutine
),
3900 .cancel
= bdrv_aio_co_cancel_em
,
3903 static void bdrv_co_em_bh(void *opaque
)
3905 BlockDriverAIOCBCoroutine
*acb
= opaque
;
3907 acb
->common
.cb(acb
->common
.opaque
, acb
->req
.error
);
3913 qemu_bh_delete(acb
->bh
);
3914 qemu_aio_release(acb
);
3917 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
3918 static void coroutine_fn
bdrv_co_do_rw(void *opaque
)
3920 BlockDriverAIOCBCoroutine
*acb
= opaque
;
3921 BlockDriverState
*bs
= acb
->common
.bs
;
3923 if (!acb
->is_write
) {
3924 acb
->req
.error
= bdrv_co_do_readv(bs
, acb
->req
.sector
,
3925 acb
->req
.nb_sectors
, acb
->req
.qiov
, 0);
3927 acb
->req
.error
= bdrv_co_do_writev(bs
, acb
->req
.sector
,
3928 acb
->req
.nb_sectors
, acb
->req
.qiov
, 0);
3931 acb
->bh
= qemu_bh_new(bdrv_co_em_bh
, acb
);
3932 qemu_bh_schedule(acb
->bh
);
3935 static BlockDriverAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
3939 BlockDriverCompletionFunc
*cb
,
3944 BlockDriverAIOCBCoroutine
*acb
;
3946 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
3947 acb
->req
.sector
= sector_num
;
3948 acb
->req
.nb_sectors
= nb_sectors
;
3949 acb
->req
.qiov
= qiov
;
3950 acb
->is_write
= is_write
;
3953 co
= qemu_coroutine_create(bdrv_co_do_rw
);
3954 qemu_coroutine_enter(co
, acb
);
3956 return &acb
->common
;
3959 static void coroutine_fn
bdrv_aio_flush_co_entry(void *opaque
)
3961 BlockDriverAIOCBCoroutine
*acb
= opaque
;
3962 BlockDriverState
*bs
= acb
->common
.bs
;
3964 acb
->req
.error
= bdrv_co_flush(bs
);
3965 acb
->bh
= qemu_bh_new(bdrv_co_em_bh
, acb
);
3966 qemu_bh_schedule(acb
->bh
);
3969 BlockDriverAIOCB
*bdrv_aio_flush(BlockDriverState
*bs
,
3970 BlockDriverCompletionFunc
*cb
, void *opaque
)
3972 trace_bdrv_aio_flush(bs
, opaque
);
3975 BlockDriverAIOCBCoroutine
*acb
;
3977 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
3980 co
= qemu_coroutine_create(bdrv_aio_flush_co_entry
);
3981 qemu_coroutine_enter(co
, acb
);
3983 return &acb
->common
;
3986 static void coroutine_fn
bdrv_aio_discard_co_entry(void *opaque
)
3988 BlockDriverAIOCBCoroutine
*acb
= opaque
;
3989 BlockDriverState
*bs
= acb
->common
.bs
;
3991 acb
->req
.error
= bdrv_co_discard(bs
, acb
->req
.sector
, acb
->req
.nb_sectors
);
3992 acb
->bh
= qemu_bh_new(bdrv_co_em_bh
, acb
);
3993 qemu_bh_schedule(acb
->bh
);
3996 BlockDriverAIOCB
*bdrv_aio_discard(BlockDriverState
*bs
,
3997 int64_t sector_num
, int nb_sectors
,
3998 BlockDriverCompletionFunc
*cb
, void *opaque
)
4001 BlockDriverAIOCBCoroutine
*acb
;
4003 trace_bdrv_aio_discard(bs
, sector_num
, nb_sectors
, opaque
);
4005 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
4006 acb
->req
.sector
= sector_num
;
4007 acb
->req
.nb_sectors
= nb_sectors
;
4009 co
= qemu_coroutine_create(bdrv_aio_discard_co_entry
);
4010 qemu_coroutine_enter(co
, acb
);
4012 return &acb
->common
;
4015 void bdrv_init(void)
4017 module_call_init(MODULE_INIT_BLOCK
);
4020 void bdrv_init_with_whitelist(void)
4022 use_bdrv_whitelist
= 1;
4026 void *qemu_aio_get(const AIOCBInfo
*aiocb_info
, BlockDriverState
*bs
,
4027 BlockDriverCompletionFunc
*cb
, void *opaque
)
4029 BlockDriverAIOCB
*acb
;
4031 acb
= g_slice_alloc(aiocb_info
->aiocb_size
);
4032 acb
->aiocb_info
= aiocb_info
;
4035 acb
->opaque
= opaque
;
4039 void qemu_aio_release(void *p
)
4041 BlockDriverAIOCB
*acb
= p
;
4042 g_slice_free1(acb
->aiocb_info
->aiocb_size
, acb
);
4045 /**************************************************************/
4046 /* Coroutine block device emulation */
4048 typedef struct CoroutineIOCompletion
{
4049 Coroutine
*coroutine
;
4051 } CoroutineIOCompletion
;
4053 static void bdrv_co_io_em_complete(void *opaque
, int ret
)
4055 CoroutineIOCompletion
*co
= opaque
;
4058 qemu_coroutine_enter(co
->coroutine
, NULL
);
4061 static int coroutine_fn
bdrv_co_io_em(BlockDriverState
*bs
, int64_t sector_num
,
4062 int nb_sectors
, QEMUIOVector
*iov
,
4065 CoroutineIOCompletion co
= {
4066 .coroutine
= qemu_coroutine_self(),
4068 BlockDriverAIOCB
*acb
;
4071 acb
= bs
->drv
->bdrv_aio_writev(bs
, sector_num
, iov
, nb_sectors
,
4072 bdrv_co_io_em_complete
, &co
);
4074 acb
= bs
->drv
->bdrv_aio_readv(bs
, sector_num
, iov
, nb_sectors
,
4075 bdrv_co_io_em_complete
, &co
);
4078 trace_bdrv_co_io_em(bs
, sector_num
, nb_sectors
, is_write
, acb
);
4082 qemu_coroutine_yield();
4087 static int coroutine_fn
bdrv_co_readv_em(BlockDriverState
*bs
,
4088 int64_t sector_num
, int nb_sectors
,
4091 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, false);
4094 static int coroutine_fn
bdrv_co_writev_em(BlockDriverState
*bs
,
4095 int64_t sector_num
, int nb_sectors
,
4098 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, true);
4101 static void coroutine_fn
bdrv_flush_co_entry(void *opaque
)
4103 RwCo
*rwco
= opaque
;
4105 rwco
->ret
= bdrv_co_flush(rwco
->bs
);
4108 int coroutine_fn
bdrv_co_flush(BlockDriverState
*bs
)
4112 if (!bs
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
)) {
4116 /* Write back cached data to the OS even with cache=unsafe */
4117 if (bs
->drv
->bdrv_co_flush_to_os
) {
4118 ret
= bs
->drv
->bdrv_co_flush_to_os(bs
);
4124 /* But don't actually force it to the disk with cache=unsafe */
4125 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
4129 if (bs
->drv
->bdrv_co_flush_to_disk
) {
4130 ret
= bs
->drv
->bdrv_co_flush_to_disk(bs
);
4131 } else if (bs
->drv
->bdrv_aio_flush
) {
4132 BlockDriverAIOCB
*acb
;
4133 CoroutineIOCompletion co
= {
4134 .coroutine
= qemu_coroutine_self(),
4137 acb
= bs
->drv
->bdrv_aio_flush(bs
, bdrv_co_io_em_complete
, &co
);
4141 qemu_coroutine_yield();
4146 * Some block drivers always operate in either writethrough or unsafe
4147 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
4148 * know how the server works (because the behaviour is hardcoded or
4149 * depends on server-side configuration), so we can't ensure that
4150 * everything is safe on disk. Returning an error doesn't work because
4151 * that would break guests even if the server operates in writethrough
4154 * Let's hope the user knows what he's doing.
4162 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
4163 * in the case of cache=unsafe, so there are no useless flushes.
4166 return bdrv_co_flush(bs
->file
);
4169 void bdrv_invalidate_cache(BlockDriverState
*bs
)
4171 if (bs
->drv
&& bs
->drv
->bdrv_invalidate_cache
) {
4172 bs
->drv
->bdrv_invalidate_cache(bs
);
4176 void bdrv_invalidate_cache_all(void)
4178 BlockDriverState
*bs
;
4180 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
4181 bdrv_invalidate_cache(bs
);
4185 void bdrv_clear_incoming_migration_all(void)
4187 BlockDriverState
*bs
;
4189 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
4190 bs
->open_flags
= bs
->open_flags
& ~(BDRV_O_INCOMING
);
4194 int bdrv_flush(BlockDriverState
*bs
)
4202 if (qemu_in_coroutine()) {
4203 /* Fast-path if already in coroutine context */
4204 bdrv_flush_co_entry(&rwco
);
4206 co
= qemu_coroutine_create(bdrv_flush_co_entry
);
4207 qemu_coroutine_enter(co
, &rwco
);
4208 while (rwco
.ret
== NOT_DONE
) {
4216 static void coroutine_fn
bdrv_discard_co_entry(void *opaque
)
4218 RwCo
*rwco
= opaque
;
4220 rwco
->ret
= bdrv_co_discard(rwco
->bs
, rwco
->sector_num
, rwco
->nb_sectors
);
4223 int coroutine_fn
bdrv_co_discard(BlockDriverState
*bs
, int64_t sector_num
,
4228 } else if (bdrv_check_request(bs
, sector_num
, nb_sectors
)) {
4230 } else if (bs
->read_only
) {
4234 if (bs
->dirty_bitmap
) {
4235 bdrv_reset_dirty(bs
, sector_num
, nb_sectors
);
4238 /* Do nothing if disabled. */
4239 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
4243 if (bs
->drv
->bdrv_co_discard
) {
4244 return bs
->drv
->bdrv_co_discard(bs
, sector_num
, nb_sectors
);
4245 } else if (bs
->drv
->bdrv_aio_discard
) {
4246 BlockDriverAIOCB
*acb
;
4247 CoroutineIOCompletion co
= {
4248 .coroutine
= qemu_coroutine_self(),
4251 acb
= bs
->drv
->bdrv_aio_discard(bs
, sector_num
, nb_sectors
,
4252 bdrv_co_io_em_complete
, &co
);
4256 qemu_coroutine_yield();
4264 int bdrv_discard(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
)
4269 .sector_num
= sector_num
,
4270 .nb_sectors
= nb_sectors
,
4274 if (qemu_in_coroutine()) {
4275 /* Fast-path if already in coroutine context */
4276 bdrv_discard_co_entry(&rwco
);
4278 co
= qemu_coroutine_create(bdrv_discard_co_entry
);
4279 qemu_coroutine_enter(co
, &rwco
);
4280 while (rwco
.ret
== NOT_DONE
) {
4288 /**************************************************************/
4289 /* removable device support */
4292 * Return TRUE if the media is present
4294 int bdrv_is_inserted(BlockDriverState
*bs
)
4296 BlockDriver
*drv
= bs
->drv
;
4300 if (!drv
->bdrv_is_inserted
)
4302 return drv
->bdrv_is_inserted(bs
);
4306 * Return whether the media changed since the last call to this
4307 * function, or -ENOTSUP if we don't know. Most drivers don't know.
4309 int bdrv_media_changed(BlockDriverState
*bs
)
4311 BlockDriver
*drv
= bs
->drv
;
4313 if (drv
&& drv
->bdrv_media_changed
) {
4314 return drv
->bdrv_media_changed(bs
);
4320 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4322 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
4324 BlockDriver
*drv
= bs
->drv
;
4326 if (drv
&& drv
->bdrv_eject
) {
4327 drv
->bdrv_eject(bs
, eject_flag
);
4330 if (bs
->device_name
[0] != '\0') {
4331 bdrv_emit_qmp_eject_event(bs
, eject_flag
);
4336 * Lock or unlock the media (if it is locked, the user won't be able
4337 * to eject it manually).
4339 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
4341 BlockDriver
*drv
= bs
->drv
;
4343 trace_bdrv_lock_medium(bs
, locked
);
4345 if (drv
&& drv
->bdrv_lock_medium
) {
4346 drv
->bdrv_lock_medium(bs
, locked
);
4350 /* needed for generic scsi interface */
4352 int bdrv_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
4354 BlockDriver
*drv
= bs
->drv
;
4356 if (drv
&& drv
->bdrv_ioctl
)
4357 return drv
->bdrv_ioctl(bs
, req
, buf
);
4361 BlockDriverAIOCB
*bdrv_aio_ioctl(BlockDriverState
*bs
,
4362 unsigned long int req
, void *buf
,
4363 BlockDriverCompletionFunc
*cb
, void *opaque
)
4365 BlockDriver
*drv
= bs
->drv
;
4367 if (drv
&& drv
->bdrv_aio_ioctl
)
4368 return drv
->bdrv_aio_ioctl(bs
, req
, buf
, cb
, opaque
);
4372 void bdrv_set_buffer_alignment(BlockDriverState
*bs
, int align
)
4374 bs
->buffer_alignment
= align
;
4377 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
4379 return qemu_memalign((bs
&& bs
->buffer_alignment
) ? bs
->buffer_alignment
: 512, size
);
4383 * Check if all memory in this vector is sector aligned.
4385 bool bdrv_qiov_is_aligned(BlockDriverState
*bs
, QEMUIOVector
*qiov
)
4389 for (i
= 0; i
< qiov
->niov
; i
++) {
4390 if ((uintptr_t) qiov
->iov
[i
].iov_base
% bs
->buffer_alignment
) {
4398 void bdrv_set_dirty_tracking(BlockDriverState
*bs
, int granularity
)
4400 int64_t bitmap_size
;
4402 assert((granularity
& (granularity
- 1)) == 0);
4405 granularity
>>= BDRV_SECTOR_BITS
;
4406 assert(!bs
->dirty_bitmap
);
4407 bitmap_size
= (bdrv_getlength(bs
) >> BDRV_SECTOR_BITS
);
4408 bs
->dirty_bitmap
= hbitmap_alloc(bitmap_size
, ffs(granularity
) - 1);
4410 if (bs
->dirty_bitmap
) {
4411 hbitmap_free(bs
->dirty_bitmap
);
4412 bs
->dirty_bitmap
= NULL
;
4417 int bdrv_get_dirty(BlockDriverState
*bs
, int64_t sector
)
4419 if (bs
->dirty_bitmap
) {
4420 return hbitmap_get(bs
->dirty_bitmap
, sector
);
4426 void bdrv_dirty_iter_init(BlockDriverState
*bs
, HBitmapIter
*hbi
)
4428 hbitmap_iter_init(hbi
, bs
->dirty_bitmap
, 0);
4431 void bdrv_set_dirty(BlockDriverState
*bs
, int64_t cur_sector
,
4434 hbitmap_set(bs
->dirty_bitmap
, cur_sector
, nr_sectors
);
4437 void bdrv_reset_dirty(BlockDriverState
*bs
, int64_t cur_sector
,
4440 hbitmap_reset(bs
->dirty_bitmap
, cur_sector
, nr_sectors
);
4443 int64_t bdrv_get_dirty_count(BlockDriverState
*bs
)
4445 if (bs
->dirty_bitmap
) {
4446 return hbitmap_count(bs
->dirty_bitmap
);
4452 void bdrv_set_in_use(BlockDriverState
*bs
, int in_use
)
4454 assert(bs
->in_use
!= in_use
);
4455 bs
->in_use
= in_use
;
4458 int bdrv_in_use(BlockDriverState
*bs
)
4463 void bdrv_iostatus_enable(BlockDriverState
*bs
)
4465 bs
->iostatus_enabled
= true;
4466 bs
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
4469 /* The I/O status is only enabled if the drive explicitly
4470 * enables it _and_ the VM is configured to stop on errors */
4471 bool bdrv_iostatus_is_enabled(const BlockDriverState
*bs
)
4473 return (bs
->iostatus_enabled
&&
4474 (bs
->on_write_error
== BLOCKDEV_ON_ERROR_ENOSPC
||
4475 bs
->on_write_error
== BLOCKDEV_ON_ERROR_STOP
||
4476 bs
->on_read_error
== BLOCKDEV_ON_ERROR_STOP
));
4479 void bdrv_iostatus_disable(BlockDriverState
*bs
)
4481 bs
->iostatus_enabled
= false;
4484 void bdrv_iostatus_reset(BlockDriverState
*bs
)
4486 if (bdrv_iostatus_is_enabled(bs
)) {
4487 bs
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
4489 block_job_iostatus_reset(bs
->job
);
4494 void bdrv_iostatus_set_err(BlockDriverState
*bs
, int error
)
4496 assert(bdrv_iostatus_is_enabled(bs
));
4497 if (bs
->iostatus
== BLOCK_DEVICE_IO_STATUS_OK
) {
4498 bs
->iostatus
= error
== ENOSPC
? BLOCK_DEVICE_IO_STATUS_NOSPACE
:
4499 BLOCK_DEVICE_IO_STATUS_FAILED
;
4504 bdrv_acct_start(BlockDriverState
*bs
, BlockAcctCookie
*cookie
, int64_t bytes
,
4505 enum BlockAcctType type
)
4507 assert(type
< BDRV_MAX_IOTYPE
);
4509 cookie
->bytes
= bytes
;
4510 cookie
->start_time_ns
= get_clock();
4511 cookie
->type
= type
;
4515 bdrv_acct_done(BlockDriverState
*bs
, BlockAcctCookie
*cookie
)
4517 assert(cookie
->type
< BDRV_MAX_IOTYPE
);
4519 bs
->nr_bytes
[cookie
->type
] += cookie
->bytes
;
4520 bs
->nr_ops
[cookie
->type
]++;
4521 bs
->total_time_ns
[cookie
->type
] += get_clock() - cookie
->start_time_ns
;
4524 void bdrv_img_create(const char *filename
, const char *fmt
,
4525 const char *base_filename
, const char *base_fmt
,
4526 char *options
, uint64_t img_size
, int flags
,
4527 Error
**errp
, bool quiet
)
4529 QEMUOptionParameter
*param
= NULL
, *create_options
= NULL
;
4530 QEMUOptionParameter
*backing_fmt
, *backing_file
, *size
;
4531 BlockDriverState
*bs
= NULL
;
4532 BlockDriver
*drv
, *proto_drv
;
4533 BlockDriver
*backing_drv
= NULL
;
4536 /* Find driver and parse its options */
4537 drv
= bdrv_find_format(fmt
);
4539 error_setg(errp
, "Unknown file format '%s'", fmt
);
4543 proto_drv
= bdrv_find_protocol(filename
);
4545 error_setg(errp
, "Unknown protocol '%s'", filename
);
4549 create_options
= append_option_parameters(create_options
,
4550 drv
->create_options
);
4551 create_options
= append_option_parameters(create_options
,
4552 proto_drv
->create_options
);
4554 /* Create parameter list with default values */
4555 param
= parse_option_parameters("", create_options
, param
);
4557 set_option_parameter_int(param
, BLOCK_OPT_SIZE
, img_size
);
4559 /* Parse -o options */
4561 param
= parse_option_parameters(options
, create_options
, param
);
4562 if (param
== NULL
) {
4563 error_setg(errp
, "Invalid options for file format '%s'.", fmt
);
4568 if (base_filename
) {
4569 if (set_option_parameter(param
, BLOCK_OPT_BACKING_FILE
,
4571 error_setg(errp
, "Backing file not supported for file format '%s'",
4578 if (set_option_parameter(param
, BLOCK_OPT_BACKING_FMT
, base_fmt
)) {
4579 error_setg(errp
, "Backing file format not supported for file "
4580 "format '%s'", fmt
);
4585 backing_file
= get_option_parameter(param
, BLOCK_OPT_BACKING_FILE
);
4586 if (backing_file
&& backing_file
->value
.s
) {
4587 if (!strcmp(filename
, backing_file
->value
.s
)) {
4588 error_setg(errp
, "Error: Trying to create an image with the "
4589 "same filename as the backing file");
4594 backing_fmt
= get_option_parameter(param
, BLOCK_OPT_BACKING_FMT
);
4595 if (backing_fmt
&& backing_fmt
->value
.s
) {
4596 backing_drv
= bdrv_find_format(backing_fmt
->value
.s
);
4598 error_setg(errp
, "Unknown backing file format '%s'",
4599 backing_fmt
->value
.s
);
4604 // The size for the image must always be specified, with one exception:
4605 // If we are using a backing file, we can obtain the size from there
4606 size
= get_option_parameter(param
, BLOCK_OPT_SIZE
);
4607 if (size
&& size
->value
.n
== -1) {
4608 if (backing_file
&& backing_file
->value
.s
) {
4613 /* backing files always opened read-only */
4615 flags
& ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
4619 ret
= bdrv_open(bs
, backing_file
->value
.s
, NULL
, back_flags
,
4622 error_setg_errno(errp
, -ret
, "Could not open '%s'",
4623 backing_file
->value
.s
);
4626 bdrv_get_geometry(bs
, &size
);
4629 snprintf(buf
, sizeof(buf
), "%" PRId64
, size
);
4630 set_option_parameter(param
, BLOCK_OPT_SIZE
, buf
);
4632 error_setg(errp
, "Image creation needs a size parameter");
4638 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
4639 print_option_parameters(param
);
4642 ret
= bdrv_create(drv
, filename
, param
);
4644 if (ret
== -ENOTSUP
) {
4645 error_setg(errp
,"Formatting or formatting option not supported for "
4646 "file format '%s'", fmt
);
4647 } else if (ret
== -EFBIG
) {
4648 error_setg(errp
, "The image size is too large for file format '%s'",
4651 error_setg(errp
, "%s: error while creating %s: %s", filename
, fmt
,
4657 free_option_parameters(create_options
);
4658 free_option_parameters(param
);