2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24 #include "config-host.h"
25 #include "qemu-common.h"
28 #include "block_int.h"
33 #include "qemu-coroutine.h"
34 #include "qmp-commands.h"
35 #include "qemu-timer.h"
38 #include <sys/types.h>
40 #include <sys/ioctl.h>
41 #include <sys/queue.h>
51 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
54 BDRV_REQ_COPY_ON_READ
= 0x1,
55 BDRV_REQ_ZERO_WRITE
= 0x2,
58 static void bdrv_dev_change_media_cb(BlockDriverState
*bs
, bool load
);
59 static BlockDriverAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
60 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
61 BlockDriverCompletionFunc
*cb
, void *opaque
);
62 static BlockDriverAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
63 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
64 BlockDriverCompletionFunc
*cb
, void *opaque
);
65 static int coroutine_fn
bdrv_co_readv_em(BlockDriverState
*bs
,
66 int64_t sector_num
, int nb_sectors
,
68 static int coroutine_fn
bdrv_co_writev_em(BlockDriverState
*bs
,
69 int64_t sector_num
, int nb_sectors
,
71 static int coroutine_fn
bdrv_co_do_readv(BlockDriverState
*bs
,
72 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
73 BdrvRequestFlags flags
);
74 static int coroutine_fn
bdrv_co_do_writev(BlockDriverState
*bs
,
75 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
76 BdrvRequestFlags flags
);
77 static BlockDriverAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
81 BlockDriverCompletionFunc
*cb
,
84 static void coroutine_fn
bdrv_co_do_rw(void *opaque
);
85 static int coroutine_fn
bdrv_co_do_write_zeroes(BlockDriverState
*bs
,
86 int64_t sector_num
, int nb_sectors
);
88 static bool bdrv_exceed_bps_limits(BlockDriverState
*bs
, int nb_sectors
,
89 bool is_write
, double elapsed_time
, uint64_t *wait
);
90 static bool bdrv_exceed_iops_limits(BlockDriverState
*bs
, bool is_write
,
91 double elapsed_time
, uint64_t *wait
);
92 static bool bdrv_exceed_io_limits(BlockDriverState
*bs
, int nb_sectors
,
93 bool is_write
, int64_t *wait
);
95 static QTAILQ_HEAD(, BlockDriverState
) bdrv_states
=
96 QTAILQ_HEAD_INITIALIZER(bdrv_states
);
98 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
99 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
101 /* The device to use for VM snapshots */
102 static BlockDriverState
*bs_snapshots
;
104 /* If non-zero, use only whitelisted block drivers */
105 static int use_bdrv_whitelist
;
108 static int is_windows_drive_prefix(const char *filename
)
110 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
111 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
115 int is_windows_drive(const char *filename
)
117 if (is_windows_drive_prefix(filename
) &&
120 if (strstart(filename
, "\\\\.\\", NULL
) ||
121 strstart(filename
, "//./", NULL
))
127 /* throttling disk I/O limits */
128 void bdrv_io_limits_disable(BlockDriverState
*bs
)
130 bs
->io_limits_enabled
= false;
132 while (qemu_co_queue_next(&bs
->throttled_reqs
));
134 if (bs
->block_timer
) {
135 qemu_del_timer(bs
->block_timer
);
136 qemu_free_timer(bs
->block_timer
);
137 bs
->block_timer
= NULL
;
143 memset(&bs
->io_base
, 0, sizeof(bs
->io_base
));
146 static void bdrv_block_timer(void *opaque
)
148 BlockDriverState
*bs
= opaque
;
150 qemu_co_queue_next(&bs
->throttled_reqs
);
153 void bdrv_io_limits_enable(BlockDriverState
*bs
)
155 qemu_co_queue_init(&bs
->throttled_reqs
);
156 bs
->block_timer
= qemu_new_timer_ns(vm_clock
, bdrv_block_timer
, bs
);
157 bs
->slice_time
= 5 * BLOCK_IO_SLICE_TIME
;
158 bs
->slice_start
= qemu_get_clock_ns(vm_clock
);
159 bs
->slice_end
= bs
->slice_start
+ bs
->slice_time
;
160 memset(&bs
->io_base
, 0, sizeof(bs
->io_base
));
161 bs
->io_limits_enabled
= true;
164 bool bdrv_io_limits_enabled(BlockDriverState
*bs
)
166 BlockIOLimit
*io_limits
= &bs
->io_limits
;
167 return io_limits
->bps
[BLOCK_IO_LIMIT_READ
]
168 || io_limits
->bps
[BLOCK_IO_LIMIT_WRITE
]
169 || io_limits
->bps
[BLOCK_IO_LIMIT_TOTAL
]
170 || io_limits
->iops
[BLOCK_IO_LIMIT_READ
]
171 || io_limits
->iops
[BLOCK_IO_LIMIT_WRITE
]
172 || io_limits
->iops
[BLOCK_IO_LIMIT_TOTAL
];
175 static void bdrv_io_limits_intercept(BlockDriverState
*bs
,
176 bool is_write
, int nb_sectors
)
178 int64_t wait_time
= -1;
180 if (!qemu_co_queue_empty(&bs
->throttled_reqs
)) {
181 qemu_co_queue_wait(&bs
->throttled_reqs
);
184 /* In fact, we hope to keep each request's timing, in FIFO mode. The next
185 * throttled requests will not be dequeued until the current request is
186 * allowed to be serviced. So if the current request still exceeds the
187 * limits, it will be inserted to the head. All requests followed it will
188 * be still in throttled_reqs queue.
191 while (bdrv_exceed_io_limits(bs
, nb_sectors
, is_write
, &wait_time
)) {
192 qemu_mod_timer(bs
->block_timer
,
193 wait_time
+ qemu_get_clock_ns(vm_clock
));
194 qemu_co_queue_wait_insert_head(&bs
->throttled_reqs
);
197 qemu_co_queue_next(&bs
->throttled_reqs
);
200 /* check if the path starts with "<protocol>:" */
201 static int path_has_protocol(const char *path
)
206 if (is_windows_drive(path
) ||
207 is_windows_drive_prefix(path
)) {
210 p
= path
+ strcspn(path
, ":/\\");
212 p
= path
+ strcspn(path
, ":/");
218 int path_is_absolute(const char *path
)
221 /* specific case for names like: "\\.\d:" */
222 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
225 return (*path
== '/' || *path
== '\\');
227 return (*path
== '/');
231 /* if filename is absolute, just copy it to dest. Otherwise, build a
232 path to it by considering it is relative to base_path. URL are
234 void path_combine(char *dest
, int dest_size
,
235 const char *base_path
,
236 const char *filename
)
243 if (path_is_absolute(filename
)) {
244 pstrcpy(dest
, dest_size
, filename
);
246 p
= strchr(base_path
, ':');
251 p1
= strrchr(base_path
, '/');
255 p2
= strrchr(base_path
, '\\');
267 if (len
> dest_size
- 1)
269 memcpy(dest
, base_path
, len
);
271 pstrcat(dest
, dest_size
, filename
);
275 void bdrv_get_full_backing_filename(BlockDriverState
*bs
, char *dest
, size_t sz
)
277 if (bs
->backing_file
[0] == '\0' || path_has_protocol(bs
->backing_file
)) {
278 pstrcpy(dest
, sz
, bs
->backing_file
);
280 path_combine(dest
, sz
, bs
->filename
, bs
->backing_file
);
284 void bdrv_register(BlockDriver
*bdrv
)
286 /* Block drivers without coroutine functions need emulation */
287 if (!bdrv
->bdrv_co_readv
) {
288 bdrv
->bdrv_co_readv
= bdrv_co_readv_em
;
289 bdrv
->bdrv_co_writev
= bdrv_co_writev_em
;
291 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
292 * the block driver lacks aio we need to emulate that too.
294 if (!bdrv
->bdrv_aio_readv
) {
295 /* add AIO emulation layer */
296 bdrv
->bdrv_aio_readv
= bdrv_aio_readv_em
;
297 bdrv
->bdrv_aio_writev
= bdrv_aio_writev_em
;
301 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
304 /* create a new block device (by default it is empty) */
305 BlockDriverState
*bdrv_new(const char *device_name
)
307 BlockDriverState
*bs
;
309 bs
= g_malloc0(sizeof(BlockDriverState
));
310 pstrcpy(bs
->device_name
, sizeof(bs
->device_name
), device_name
);
311 if (device_name
[0] != '\0') {
312 QTAILQ_INSERT_TAIL(&bdrv_states
, bs
, list
);
314 bdrv_iostatus_disable(bs
);
318 BlockDriver
*bdrv_find_format(const char *format_name
)
321 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
322 if (!strcmp(drv1
->format_name
, format_name
)) {
329 static int bdrv_is_whitelisted(BlockDriver
*drv
)
331 static const char *whitelist
[] = {
332 CONFIG_BDRV_WHITELIST
337 return 1; /* no whitelist, anything goes */
339 for (p
= whitelist
; *p
; p
++) {
340 if (!strcmp(drv
->format_name
, *p
)) {
347 BlockDriver
*bdrv_find_whitelisted_format(const char *format_name
)
349 BlockDriver
*drv
= bdrv_find_format(format_name
);
350 return drv
&& bdrv_is_whitelisted(drv
) ? drv
: NULL
;
353 typedef struct CreateCo
{
356 QEMUOptionParameter
*options
;
360 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
362 CreateCo
*cco
= opaque
;
365 cco
->ret
= cco
->drv
->bdrv_create(cco
->filename
, cco
->options
);
368 int bdrv_create(BlockDriver
*drv
, const char* filename
,
369 QEMUOptionParameter
*options
)
376 .filename
= g_strdup(filename
),
381 if (!drv
->bdrv_create
) {
386 if (qemu_in_coroutine()) {
387 /* Fast-path if already in coroutine context */
388 bdrv_create_co_entry(&cco
);
390 co
= qemu_coroutine_create(bdrv_create_co_entry
);
391 qemu_coroutine_enter(co
, &cco
);
392 while (cco
.ret
== NOT_DONE
) {
400 g_free(cco
.filename
);
404 int bdrv_create_file(const char* filename
, QEMUOptionParameter
*options
)
408 drv
= bdrv_find_protocol(filename
);
413 return bdrv_create(drv
, filename
, options
);
417 * Create a uniquely-named empty temporary file.
418 * Return 0 upon success, otherwise a negative errno value.
420 int get_tmp_filename(char *filename
, int size
)
423 char temp_dir
[MAX_PATH
];
424 /* GetTempFileName requires that its output buffer (4th param)
425 have length MAX_PATH or greater. */
426 assert(size
>= MAX_PATH
);
427 return (GetTempPath(MAX_PATH
, temp_dir
)
428 && GetTempFileName(temp_dir
, "qem", 0, filename
)
429 ? 0 : -GetLastError());
433 tmpdir
= getenv("TMPDIR");
436 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
439 fd
= mkstemp(filename
);
443 if (close(fd
) != 0) {
452 * Detect host devices. By convention, /dev/cdrom[N] is always
453 * recognized as a host CDROM.
455 static BlockDriver
*find_hdev_driver(const char *filename
)
457 int score_max
= 0, score
;
458 BlockDriver
*drv
= NULL
, *d
;
460 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
461 if (d
->bdrv_probe_device
) {
462 score
= d
->bdrv_probe_device(filename
);
463 if (score
> score_max
) {
473 BlockDriver
*bdrv_find_protocol(const char *filename
)
480 /* TODO Drivers without bdrv_file_open must be specified explicitly */
483 * XXX(hch): we really should not let host device detection
484 * override an explicit protocol specification, but moving this
485 * later breaks access to device names with colons in them.
486 * Thanks to the brain-dead persistent naming schemes on udev-
487 * based Linux systems those actually are quite common.
489 drv1
= find_hdev_driver(filename
);
494 if (!path_has_protocol(filename
)) {
495 return bdrv_find_format("file");
497 p
= strchr(filename
, ':');
500 if (len
> sizeof(protocol
) - 1)
501 len
= sizeof(protocol
) - 1;
502 memcpy(protocol
, filename
, len
);
503 protocol
[len
] = '\0';
504 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
505 if (drv1
->protocol_name
&&
506 !strcmp(drv1
->protocol_name
, protocol
)) {
513 static int find_image_format(const char *filename
, BlockDriver
**pdrv
)
515 int ret
, score
, score_max
;
516 BlockDriver
*drv1
, *drv
;
518 BlockDriverState
*bs
;
520 ret
= bdrv_file_open(&bs
, filename
, 0);
526 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
527 if (bs
->sg
|| !bdrv_is_inserted(bs
)) {
529 drv
= bdrv_find_format("raw");
537 ret
= bdrv_pread(bs
, 0, buf
, sizeof(buf
));
546 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
547 if (drv1
->bdrv_probe
) {
548 score
= drv1
->bdrv_probe(buf
, ret
, filename
);
549 if (score
> score_max
) {
563 * Set the current 'total_sectors' value
565 static int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
567 BlockDriver
*drv
= bs
->drv
;
569 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
573 /* query actual device if possible, otherwise just trust the hint */
574 if (drv
->bdrv_getlength
) {
575 int64_t length
= drv
->bdrv_getlength(bs
);
579 hint
= length
>> BDRV_SECTOR_BITS
;
582 bs
->total_sectors
= hint
;
587 * Set open flags for a given cache mode
589 * Return 0 on success, -1 if the cache mode was invalid.
591 int bdrv_parse_cache_flags(const char *mode
, int *flags
)
593 *flags
&= ~BDRV_O_CACHE_MASK
;
595 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
596 *flags
|= BDRV_O_NOCACHE
| BDRV_O_CACHE_WB
;
597 } else if (!strcmp(mode
, "directsync")) {
598 *flags
|= BDRV_O_NOCACHE
;
599 } else if (!strcmp(mode
, "writeback")) {
600 *flags
|= BDRV_O_CACHE_WB
;
601 } else if (!strcmp(mode
, "unsafe")) {
602 *flags
|= BDRV_O_CACHE_WB
;
603 *flags
|= BDRV_O_NO_FLUSH
;
604 } else if (!strcmp(mode
, "writethrough")) {
605 /* this is the default */
614 * The copy-on-read flag is actually a reference count so multiple users may
615 * use the feature without worrying about clobbering its previous state.
616 * Copy-on-read stays enabled until all users have called to disable it.
618 void bdrv_enable_copy_on_read(BlockDriverState
*bs
)
623 void bdrv_disable_copy_on_read(BlockDriverState
*bs
)
625 assert(bs
->copy_on_read
> 0);
630 * Common part for opening disk images and files
632 static int bdrv_open_common(BlockDriverState
*bs
, const char *filename
,
633 int flags
, BlockDriver
*drv
)
638 assert(bs
->file
== NULL
);
640 trace_bdrv_open_common(bs
, filename
, flags
, drv
->format_name
);
642 bs
->open_flags
= flags
;
643 bs
->buffer_alignment
= 512;
645 assert(bs
->copy_on_read
== 0); /* bdrv_new() and bdrv_close() make it so */
646 if ((flags
& BDRV_O_RDWR
) && (flags
& BDRV_O_COPY_ON_READ
)) {
647 bdrv_enable_copy_on_read(bs
);
650 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
652 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
)) {
657 bs
->opaque
= g_malloc0(drv
->instance_size
);
659 bs
->enable_write_cache
= !!(flags
& BDRV_O_CACHE_WB
);
660 open_flags
= flags
| BDRV_O_CACHE_WB
;
663 * Clear flags that are internal to the block layer before opening the
666 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
669 * Snapshots should be writable.
671 if (bs
->is_temporary
) {
672 open_flags
|= BDRV_O_RDWR
;
675 bs
->read_only
= !(open_flags
& BDRV_O_RDWR
);
677 /* Open the image, either directly or using a protocol */
678 if (drv
->bdrv_file_open
) {
679 ret
= drv
->bdrv_file_open(bs
, filename
, open_flags
);
681 ret
= bdrv_file_open(&bs
->file
, filename
, open_flags
);
683 ret
= drv
->bdrv_open(bs
, open_flags
);
691 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
697 if (bs
->is_temporary
) {
705 bdrv_delete(bs
->file
);
715 * Opens a file using a protocol (file, host_device, nbd, ...)
717 int bdrv_file_open(BlockDriverState
**pbs
, const char *filename
, int flags
)
719 BlockDriverState
*bs
;
723 drv
= bdrv_find_protocol(filename
);
729 ret
= bdrv_open_common(bs
, filename
, flags
, drv
);
739 int bdrv_open_backing_file(BlockDriverState
*bs
)
741 char backing_filename
[PATH_MAX
];
743 BlockDriver
*back_drv
= NULL
;
745 if (bs
->backing_hd
!= NULL
) {
749 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
750 if (bs
->backing_file
[0] == '\0') {
754 bs
->backing_hd
= bdrv_new("");
755 bdrv_get_full_backing_filename(bs
, backing_filename
,
756 sizeof(backing_filename
));
758 if (bs
->backing_format
[0] != '\0') {
759 back_drv
= bdrv_find_format(bs
->backing_format
);
762 /* backing files always opened read-only */
763 back_flags
= bs
->open_flags
& ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
);
765 ret
= bdrv_open(bs
->backing_hd
, backing_filename
, back_flags
, back_drv
);
767 bdrv_delete(bs
->backing_hd
);
768 bs
->backing_hd
= NULL
;
769 bs
->open_flags
|= BDRV_O_NO_BACKING
;
776 * Opens a disk image (raw, qcow2, vmdk, ...)
778 int bdrv_open(BlockDriverState
*bs
, const char *filename
, int flags
,
782 char tmp_filename
[PATH_MAX
];
784 if (flags
& BDRV_O_SNAPSHOT
) {
785 BlockDriverState
*bs1
;
788 BlockDriver
*bdrv_qcow2
;
789 QEMUOptionParameter
*options
;
790 char backing_filename
[PATH_MAX
];
792 /* if snapshot, we create a temporary backing file and open it
793 instead of opening 'filename' directly */
795 /* if there is a backing file, use it */
797 ret
= bdrv_open(bs1
, filename
, 0, drv
);
802 total_size
= bdrv_getlength(bs1
) & BDRV_SECTOR_MASK
;
804 if (bs1
->drv
&& bs1
->drv
->protocol_name
)
809 ret
= get_tmp_filename(tmp_filename
, sizeof(tmp_filename
));
814 /* Real path is meaningless for protocols */
816 snprintf(backing_filename
, sizeof(backing_filename
),
818 else if (!realpath(filename
, backing_filename
))
821 bdrv_qcow2
= bdrv_find_format("qcow2");
822 options
= parse_option_parameters("", bdrv_qcow2
->create_options
, NULL
);
824 set_option_parameter_int(options
, BLOCK_OPT_SIZE
, total_size
);
825 set_option_parameter(options
, BLOCK_OPT_BACKING_FILE
, backing_filename
);
827 set_option_parameter(options
, BLOCK_OPT_BACKING_FMT
,
831 ret
= bdrv_create(bdrv_qcow2
, tmp_filename
, options
);
832 free_option_parameters(options
);
837 filename
= tmp_filename
;
839 bs
->is_temporary
= 1;
842 /* Find the right image format driver */
844 ret
= find_image_format(filename
, &drv
);
848 goto unlink_and_fail
;
851 if (flags
& BDRV_O_RDWR
) {
852 flags
|= BDRV_O_ALLOW_RDWR
;
856 ret
= bdrv_open_common(bs
, filename
, flags
, drv
);
858 goto unlink_and_fail
;
861 /* If there is a backing file, use it */
862 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
863 ret
= bdrv_open_backing_file(bs
);
870 if (!bdrv_key_required(bs
)) {
871 bdrv_dev_change_media_cb(bs
, true);
874 /* throttling disk I/O limits */
875 if (bs
->io_limits_enabled
) {
876 bdrv_io_limits_enable(bs
);
882 if (bs
->is_temporary
) {
888 typedef struct BlockReopenQueueEntry
{
890 BDRVReopenState state
;
891 QSIMPLEQ_ENTRY(BlockReopenQueueEntry
) entry
;
892 } BlockReopenQueueEntry
;
895 * Adds a BlockDriverState to a simple queue for an atomic, transactional
896 * reopen of multiple devices.
898 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
899 * already performed, or alternatively may be NULL a new BlockReopenQueue will
900 * be created and initialized. This newly created BlockReopenQueue should be
901 * passed back in for subsequent calls that are intended to be of the same
904 * bs is the BlockDriverState to add to the reopen queue.
906 * flags contains the open flags for the associated bs
908 * returns a pointer to bs_queue, which is either the newly allocated
909 * bs_queue, or the existing bs_queue being used.
912 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
913 BlockDriverState
*bs
, int flags
)
917 BlockReopenQueueEntry
*bs_entry
;
918 if (bs_queue
== NULL
) {
919 bs_queue
= g_new0(BlockReopenQueue
, 1);
920 QSIMPLEQ_INIT(bs_queue
);
924 bdrv_reopen_queue(bs_queue
, bs
->file
, flags
);
927 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
928 QSIMPLEQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
930 bs_entry
->state
.bs
= bs
;
931 bs_entry
->state
.flags
= flags
;
937 * Reopen multiple BlockDriverStates atomically & transactionally.
939 * The queue passed in (bs_queue) must have been built up previous
940 * via bdrv_reopen_queue().
942 * Reopens all BDS specified in the queue, with the appropriate
943 * flags. All devices are prepared for reopen, and failure of any
944 * device will cause all device changes to be abandonded, and intermediate
947 * If all devices prepare successfully, then the changes are committed
951 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
954 BlockReopenQueueEntry
*bs_entry
, *next
;
955 Error
*local_err
= NULL
;
957 assert(bs_queue
!= NULL
);
961 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
962 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, &local_err
)) {
963 error_propagate(errp
, local_err
);
966 bs_entry
->prepared
= true;
969 /* If we reach this point, we have success and just need to apply the
972 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
973 bdrv_reopen_commit(&bs_entry
->state
);
979 QSIMPLEQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
980 if (ret
&& bs_entry
->prepared
) {
981 bdrv_reopen_abort(&bs_entry
->state
);
990 /* Reopen a single BlockDriverState with the specified flags. */
991 int bdrv_reopen(BlockDriverState
*bs
, int bdrv_flags
, Error
**errp
)
994 Error
*local_err
= NULL
;
995 BlockReopenQueue
*queue
= bdrv_reopen_queue(NULL
, bs
, bdrv_flags
);
997 ret
= bdrv_reopen_multiple(queue
, &local_err
);
998 if (local_err
!= NULL
) {
999 error_propagate(errp
, local_err
);
1006 * Prepares a BlockDriverState for reopen. All changes are staged in the
1007 * 'opaque' field of the BDRVReopenState, which is used and allocated by
1008 * the block driver layer .bdrv_reopen_prepare()
1010 * bs is the BlockDriverState to reopen
1011 * flags are the new open flags
1012 * queue is the reopen queue
1014 * Returns 0 on success, non-zero on error. On error errp will be set
1017 * On failure, bdrv_reopen_abort() will be called to clean up any data.
1018 * It is the responsibility of the caller to then call the abort() or
1019 * commit() for any other BDS that have been left in a prepare() state
1022 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
1026 Error
*local_err
= NULL
;
1029 assert(reopen_state
!= NULL
);
1030 assert(reopen_state
->bs
->drv
!= NULL
);
1031 drv
= reopen_state
->bs
->drv
;
1033 /* if we are to stay read-only, do not allow permission change
1035 if (!(reopen_state
->bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
1036 reopen_state
->flags
& BDRV_O_RDWR
) {
1037 error_set(errp
, QERR_DEVICE_IS_READ_ONLY
,
1038 reopen_state
->bs
->device_name
);
1043 ret
= bdrv_flush(reopen_state
->bs
);
1045 error_set(errp
, ERROR_CLASS_GENERIC_ERROR
, "Error (%s) flushing drive",
1050 if (drv
->bdrv_reopen_prepare
) {
1051 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
1053 if (local_err
!= NULL
) {
1054 error_propagate(errp
, local_err
);
1056 error_set(errp
, QERR_OPEN_FILE_FAILED
,
1057 reopen_state
->bs
->filename
);
1062 /* It is currently mandatory to have a bdrv_reopen_prepare()
1063 * handler for each supported drv. */
1064 error_set(errp
, QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED
,
1065 drv
->format_name
, reopen_state
->bs
->device_name
,
1066 "reopening of file");
1078 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
1079 * makes them final by swapping the staging BlockDriverState contents into
1080 * the active BlockDriverState contents.
1082 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
1086 assert(reopen_state
!= NULL
);
1087 drv
= reopen_state
->bs
->drv
;
1088 assert(drv
!= NULL
);
1090 /* If there are any driver level actions to take */
1091 if (drv
->bdrv_reopen_commit
) {
1092 drv
->bdrv_reopen_commit(reopen_state
);
1095 /* set BDS specific flags now */
1096 reopen_state
->bs
->open_flags
= reopen_state
->flags
;
1097 reopen_state
->bs
->enable_write_cache
= !!(reopen_state
->flags
&
1099 reopen_state
->bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
1103 * Abort the reopen, and delete and free the staged changes in
1106 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
1110 assert(reopen_state
!= NULL
);
1111 drv
= reopen_state
->bs
->drv
;
1112 assert(drv
!= NULL
);
1114 if (drv
->bdrv_reopen_abort
) {
1115 drv
->bdrv_reopen_abort(reopen_state
);
1120 void bdrv_close(BlockDriverState
*bs
)
1125 block_job_cancel_sync(bs
->job
);
1129 if (bs
== bs_snapshots
) {
1130 bs_snapshots
= NULL
;
1132 if (bs
->backing_hd
) {
1133 bdrv_delete(bs
->backing_hd
);
1134 bs
->backing_hd
= NULL
;
1136 bs
->drv
->bdrv_close(bs
);
1139 if (bs
->is_temporary
) {
1140 unlink(bs
->filename
);
1145 bs
->copy_on_read
= 0;
1146 bs
->backing_file
[0] = '\0';
1147 bs
->backing_format
[0] = '\0';
1148 bs
->total_sectors
= 0;
1154 if (bs
->file
!= NULL
) {
1155 bdrv_delete(bs
->file
);
1160 bdrv_dev_change_media_cb(bs
, false);
1162 /*throttling disk I/O limits*/
1163 if (bs
->io_limits_enabled
) {
1164 bdrv_io_limits_disable(bs
);
1168 void bdrv_close_all(void)
1170 BlockDriverState
*bs
;
1172 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1178 * Wait for pending requests to complete across all BlockDriverStates
1180 * This function does not flush data to disk, use bdrv_flush_all() for that
1181 * after calling this function.
1183 * Note that completion of an asynchronous I/O operation can trigger any
1184 * number of other I/O operations on other devices---for example a coroutine
1185 * can be arbitrarily complex and a constant flow of I/O can come until the
1186 * coroutine is complete. Because of this, it is not possible to have a
1187 * function to drain a single device's I/O queue.
1189 void bdrv_drain_all(void)
1191 BlockDriverState
*bs
;
1195 busy
= qemu_aio_wait();
1197 /* FIXME: We do not have timer support here, so this is effectively
1200 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1201 if (!qemu_co_queue_empty(&bs
->throttled_reqs
)) {
1202 qemu_co_queue_restart_all(&bs
->throttled_reqs
);
1208 /* If requests are still pending there is a bug somewhere */
1209 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1210 assert(QLIST_EMPTY(&bs
->tracked_requests
));
1211 assert(qemu_co_queue_empty(&bs
->throttled_reqs
));
1215 /* make a BlockDriverState anonymous by removing from bdrv_state list.
1216 Also, NULL terminate the device_name to prevent double remove */
1217 void bdrv_make_anon(BlockDriverState
*bs
)
1219 if (bs
->device_name
[0] != '\0') {
1220 QTAILQ_REMOVE(&bdrv_states
, bs
, list
);
1222 bs
->device_name
[0] = '\0';
1225 static void bdrv_rebind(BlockDriverState
*bs
)
1227 if (bs
->drv
&& bs
->drv
->bdrv_rebind
) {
1228 bs
->drv
->bdrv_rebind(bs
);
1232 static void bdrv_move_feature_fields(BlockDriverState
*bs_dest
,
1233 BlockDriverState
*bs_src
)
1235 /* move some fields that need to stay attached to the device */
1236 bs_dest
->open_flags
= bs_src
->open_flags
;
1239 bs_dest
->dev_ops
= bs_src
->dev_ops
;
1240 bs_dest
->dev_opaque
= bs_src
->dev_opaque
;
1241 bs_dest
->dev
= bs_src
->dev
;
1242 bs_dest
->buffer_alignment
= bs_src
->buffer_alignment
;
1243 bs_dest
->copy_on_read
= bs_src
->copy_on_read
;
1245 bs_dest
->enable_write_cache
= bs_src
->enable_write_cache
;
1247 /* i/o timing parameters */
1248 bs_dest
->slice_time
= bs_src
->slice_time
;
1249 bs_dest
->slice_start
= bs_src
->slice_start
;
1250 bs_dest
->slice_end
= bs_src
->slice_end
;
1251 bs_dest
->io_limits
= bs_src
->io_limits
;
1252 bs_dest
->io_base
= bs_src
->io_base
;
1253 bs_dest
->throttled_reqs
= bs_src
->throttled_reqs
;
1254 bs_dest
->block_timer
= bs_src
->block_timer
;
1255 bs_dest
->io_limits_enabled
= bs_src
->io_limits_enabled
;
1258 bs_dest
->on_read_error
= bs_src
->on_read_error
;
1259 bs_dest
->on_write_error
= bs_src
->on_write_error
;
1262 bs_dest
->iostatus_enabled
= bs_src
->iostatus_enabled
;
1263 bs_dest
->iostatus
= bs_src
->iostatus
;
1266 bs_dest
->dirty_count
= bs_src
->dirty_count
;
1267 bs_dest
->dirty_bitmap
= bs_src
->dirty_bitmap
;
1270 bs_dest
->in_use
= bs_src
->in_use
;
1271 bs_dest
->job
= bs_src
->job
;
1273 /* keep the same entry in bdrv_states */
1274 pstrcpy(bs_dest
->device_name
, sizeof(bs_dest
->device_name
),
1275 bs_src
->device_name
);
1276 bs_dest
->list
= bs_src
->list
;
1280 * Swap bs contents for two image chains while they are live,
1281 * while keeping required fields on the BlockDriverState that is
1282 * actually attached to a device.
1284 * This will modify the BlockDriverState fields, and swap contents
1285 * between bs_new and bs_old. Both bs_new and bs_old are modified.
1287 * bs_new is required to be anonymous.
1289 * This function does not create any image files.
1291 void bdrv_swap(BlockDriverState
*bs_new
, BlockDriverState
*bs_old
)
1293 BlockDriverState tmp
;
1295 /* bs_new must be anonymous and shouldn't have anything fancy enabled */
1296 assert(bs_new
->device_name
[0] == '\0');
1297 assert(bs_new
->dirty_bitmap
== NULL
);
1298 assert(bs_new
->job
== NULL
);
1299 assert(bs_new
->dev
== NULL
);
1300 assert(bs_new
->in_use
== 0);
1301 assert(bs_new
->io_limits_enabled
== false);
1302 assert(bs_new
->block_timer
== NULL
);
1308 /* there are some fields that should not be swapped, move them back */
1309 bdrv_move_feature_fields(&tmp
, bs_old
);
1310 bdrv_move_feature_fields(bs_old
, bs_new
);
1311 bdrv_move_feature_fields(bs_new
, &tmp
);
1313 /* bs_new shouldn't be in bdrv_states even after the swap! */
1314 assert(bs_new
->device_name
[0] == '\0');
1316 /* Check a few fields that should remain attached to the device */
1317 assert(bs_new
->dev
== NULL
);
1318 assert(bs_new
->job
== NULL
);
1319 assert(bs_new
->in_use
== 0);
1320 assert(bs_new
->io_limits_enabled
== false);
1321 assert(bs_new
->block_timer
== NULL
);
1323 bdrv_rebind(bs_new
);
1324 bdrv_rebind(bs_old
);
1328 * Add new bs contents at the top of an image chain while the chain is
1329 * live, while keeping required fields on the top layer.
1331 * This will modify the BlockDriverState fields, and swap contents
1332 * between bs_new and bs_top. Both bs_new and bs_top are modified.
1334 * bs_new is required to be anonymous.
1336 * This function does not create any image files.
1338 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
)
1340 bdrv_swap(bs_new
, bs_top
);
1342 /* The contents of 'tmp' will become bs_top, as we are
1343 * swapping bs_new and bs_top contents. */
1344 bs_top
->backing_hd
= bs_new
;
1345 bs_top
->open_flags
&= ~BDRV_O_NO_BACKING
;
1346 pstrcpy(bs_top
->backing_file
, sizeof(bs_top
->backing_file
),
1348 pstrcpy(bs_top
->backing_format
, sizeof(bs_top
->backing_format
),
1349 bs_new
->drv
? bs_new
->drv
->format_name
: "");
1352 void bdrv_delete(BlockDriverState
*bs
)
1356 assert(!bs
->in_use
);
1358 /* remove from list, if necessary */
1363 assert(bs
!= bs_snapshots
);
1367 int bdrv_attach_dev(BlockDriverState
*bs
, void *dev
)
1368 /* TODO change to DeviceState *dev when all users are qdevified */
1374 bdrv_iostatus_reset(bs
);
1378 /* TODO qdevified devices don't use this, remove when devices are qdevified */
1379 void bdrv_attach_dev_nofail(BlockDriverState
*bs
, void *dev
)
1381 if (bdrv_attach_dev(bs
, dev
) < 0) {
1386 void bdrv_detach_dev(BlockDriverState
*bs
, void *dev
)
1387 /* TODO change to DeviceState *dev when all users are qdevified */
1389 assert(bs
->dev
== dev
);
1392 bs
->dev_opaque
= NULL
;
1393 bs
->buffer_alignment
= 512;
1396 /* TODO change to return DeviceState * when all users are qdevified */
1397 void *bdrv_get_attached_dev(BlockDriverState
*bs
)
1402 void bdrv_set_dev_ops(BlockDriverState
*bs
, const BlockDevOps
*ops
,
1406 bs
->dev_opaque
= opaque
;
1407 if (bdrv_dev_has_removable_media(bs
) && bs
== bs_snapshots
) {
1408 bs_snapshots
= NULL
;
1412 void bdrv_emit_qmp_error_event(const BlockDriverState
*bdrv
,
1413 enum MonitorEvent ev
,
1414 BlockErrorAction action
, bool is_read
)
1417 const char *action_str
;
1420 case BDRV_ACTION_REPORT
:
1421 action_str
= "report";
1423 case BDRV_ACTION_IGNORE
:
1424 action_str
= "ignore";
1426 case BDRV_ACTION_STOP
:
1427 action_str
= "stop";
1433 data
= qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1436 is_read
? "read" : "write");
1437 monitor_protocol_event(ev
, data
);
1439 qobject_decref(data
);
1442 static void bdrv_emit_qmp_eject_event(BlockDriverState
*bs
, bool ejected
)
1446 data
= qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }",
1447 bdrv_get_device_name(bs
), ejected
);
1448 monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED
, data
);
1450 qobject_decref(data
);
1453 static void bdrv_dev_change_media_cb(BlockDriverState
*bs
, bool load
)
1455 if (bs
->dev_ops
&& bs
->dev_ops
->change_media_cb
) {
1456 bool tray_was_closed
= !bdrv_dev_is_tray_open(bs
);
1457 bs
->dev_ops
->change_media_cb(bs
->dev_opaque
, load
);
1458 if (tray_was_closed
) {
1460 bdrv_emit_qmp_eject_event(bs
, true);
1464 bdrv_emit_qmp_eject_event(bs
, false);
1469 bool bdrv_dev_has_removable_media(BlockDriverState
*bs
)
1471 return !bs
->dev
|| (bs
->dev_ops
&& bs
->dev_ops
->change_media_cb
);
1474 void bdrv_dev_eject_request(BlockDriverState
*bs
, bool force
)
1476 if (bs
->dev_ops
&& bs
->dev_ops
->eject_request_cb
) {
1477 bs
->dev_ops
->eject_request_cb(bs
->dev_opaque
, force
);
1481 bool bdrv_dev_is_tray_open(BlockDriverState
*bs
)
1483 if (bs
->dev_ops
&& bs
->dev_ops
->is_tray_open
) {
1484 return bs
->dev_ops
->is_tray_open(bs
->dev_opaque
);
1489 static void bdrv_dev_resize_cb(BlockDriverState
*bs
)
1491 if (bs
->dev_ops
&& bs
->dev_ops
->resize_cb
) {
1492 bs
->dev_ops
->resize_cb(bs
->dev_opaque
);
1496 bool bdrv_dev_is_medium_locked(BlockDriverState
*bs
)
1498 if (bs
->dev_ops
&& bs
->dev_ops
->is_medium_locked
) {
1499 return bs
->dev_ops
->is_medium_locked(bs
->dev_opaque
);
1505 * Run consistency checks on an image
1507 * Returns 0 if the check could be completed (it doesn't mean that the image is
1508 * free of errors) or -errno when an internal error occurred. The results of the
1509 * check are stored in res.
1511 int bdrv_check(BlockDriverState
*bs
, BdrvCheckResult
*res
, BdrvCheckMode fix
)
1513 if (bs
->drv
->bdrv_check
== NULL
) {
1517 memset(res
, 0, sizeof(*res
));
1518 return bs
->drv
->bdrv_check(bs
, res
, fix
);
1521 #define COMMIT_BUF_SECTORS 2048
1523 /* commit COW file into the raw image */
1524 int bdrv_commit(BlockDriverState
*bs
)
1526 BlockDriver
*drv
= bs
->drv
;
1527 int64_t sector
, total_sectors
;
1528 int n
, ro
, open_flags
;
1531 char filename
[PATH_MAX
];
1536 if (!bs
->backing_hd
) {
1540 if (bdrv_in_use(bs
) || bdrv_in_use(bs
->backing_hd
)) {
1544 ro
= bs
->backing_hd
->read_only
;
1545 /* Use pstrcpy (not strncpy): filename must be NUL-terminated. */
1546 pstrcpy(filename
, sizeof(filename
), bs
->backing_hd
->filename
);
1547 open_flags
= bs
->backing_hd
->open_flags
;
1550 if (bdrv_reopen(bs
->backing_hd
, open_flags
| BDRV_O_RDWR
, NULL
)) {
1555 total_sectors
= bdrv_getlength(bs
) >> BDRV_SECTOR_BITS
;
1556 buf
= g_malloc(COMMIT_BUF_SECTORS
* BDRV_SECTOR_SIZE
);
1558 for (sector
= 0; sector
< total_sectors
; sector
+= n
) {
1559 if (bdrv_is_allocated(bs
, sector
, COMMIT_BUF_SECTORS
, &n
)) {
1561 if (bdrv_read(bs
, sector
, buf
, n
) != 0) {
1566 if (bdrv_write(bs
->backing_hd
, sector
, buf
, n
) != 0) {
1573 if (drv
->bdrv_make_empty
) {
1574 ret
= drv
->bdrv_make_empty(bs
);
1579 * Make sure all data we wrote to the backing device is actually
1583 bdrv_flush(bs
->backing_hd
);
1589 /* ignoring error return here */
1590 bdrv_reopen(bs
->backing_hd
, open_flags
& ~BDRV_O_RDWR
, NULL
);
1596 int bdrv_commit_all(void)
1598 BlockDriverState
*bs
;
1600 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
1601 int ret
= bdrv_commit(bs
);
1609 struct BdrvTrackedRequest
{
1610 BlockDriverState
*bs
;
1614 QLIST_ENTRY(BdrvTrackedRequest
) list
;
1615 Coroutine
*co
; /* owner, used for deadlock detection */
1616 CoQueue wait_queue
; /* coroutines blocked on this request */
1620 * Remove an active request from the tracked requests list
1622 * This function should be called when a tracked request is completing.
1624 static void tracked_request_end(BdrvTrackedRequest
*req
)
1626 QLIST_REMOVE(req
, list
);
1627 qemu_co_queue_restart_all(&req
->wait_queue
);
1631 * Add an active request to the tracked requests list
1633 static void tracked_request_begin(BdrvTrackedRequest
*req
,
1634 BlockDriverState
*bs
,
1636 int nb_sectors
, bool is_write
)
1638 *req
= (BdrvTrackedRequest
){
1640 .sector_num
= sector_num
,
1641 .nb_sectors
= nb_sectors
,
1642 .is_write
= is_write
,
1643 .co
= qemu_coroutine_self(),
1646 qemu_co_queue_init(&req
->wait_queue
);
1648 QLIST_INSERT_HEAD(&bs
->tracked_requests
, req
, list
);
1652 * Round a region to cluster boundaries
1654 static void round_to_clusters(BlockDriverState
*bs
,
1655 int64_t sector_num
, int nb_sectors
,
1656 int64_t *cluster_sector_num
,
1657 int *cluster_nb_sectors
)
1659 BlockDriverInfo bdi
;
1661 if (bdrv_get_info(bs
, &bdi
) < 0 || bdi
.cluster_size
== 0) {
1662 *cluster_sector_num
= sector_num
;
1663 *cluster_nb_sectors
= nb_sectors
;
1665 int64_t c
= bdi
.cluster_size
/ BDRV_SECTOR_SIZE
;
1666 *cluster_sector_num
= QEMU_ALIGN_DOWN(sector_num
, c
);
1667 *cluster_nb_sectors
= QEMU_ALIGN_UP(sector_num
- *cluster_sector_num
+
1672 static bool tracked_request_overlaps(BdrvTrackedRequest
*req
,
1673 int64_t sector_num
, int nb_sectors
) {
1675 if (sector_num
>= req
->sector_num
+ req
->nb_sectors
) {
1679 if (req
->sector_num
>= sector_num
+ nb_sectors
) {
1685 static void coroutine_fn
wait_for_overlapping_requests(BlockDriverState
*bs
,
1686 int64_t sector_num
, int nb_sectors
)
1688 BdrvTrackedRequest
*req
;
1689 int64_t cluster_sector_num
;
1690 int cluster_nb_sectors
;
1693 /* If we touch the same cluster it counts as an overlap. This guarantees
1694 * that allocating writes will be serialized and not race with each other
1695 * for the same cluster. For example, in copy-on-read it ensures that the
1696 * CoR read and write operations are atomic and guest writes cannot
1697 * interleave between them.
1699 round_to_clusters(bs
, sector_num
, nb_sectors
,
1700 &cluster_sector_num
, &cluster_nb_sectors
);
1704 QLIST_FOREACH(req
, &bs
->tracked_requests
, list
) {
1705 if (tracked_request_overlaps(req
, cluster_sector_num
,
1706 cluster_nb_sectors
)) {
1707 /* Hitting this means there was a reentrant request, for
1708 * example, a block driver issuing nested requests. This must
1709 * never happen since it means deadlock.
1711 assert(qemu_coroutine_self() != req
->co
);
1713 qemu_co_queue_wait(&req
->wait_queue
);
1724 * -EINVAL - backing format specified, but no file
1725 * -ENOSPC - can't update the backing file because no space is left in the
1727 * -ENOTSUP - format driver doesn't support changing the backing file
1729 int bdrv_change_backing_file(BlockDriverState
*bs
,
1730 const char *backing_file
, const char *backing_fmt
)
1732 BlockDriver
*drv
= bs
->drv
;
1735 /* Backing file format doesn't make sense without a backing file */
1736 if (backing_fmt
&& !backing_file
) {
1740 if (drv
->bdrv_change_backing_file
!= NULL
) {
1741 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
1747 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
1748 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
1754 * Finds the image layer in the chain that has 'bs' as its backing file.
1756 * active is the current topmost image.
1758 * Returns NULL if bs is not found in active's image chain,
1759 * or if active == bs.
1761 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
1762 BlockDriverState
*bs
)
1764 BlockDriverState
*overlay
= NULL
;
1765 BlockDriverState
*intermediate
;
1767 assert(active
!= NULL
);
1770 /* if bs is the same as active, then by definition it has no overlay
1776 intermediate
= active
;
1777 while (intermediate
->backing_hd
) {
1778 if (intermediate
->backing_hd
== bs
) {
1779 overlay
= intermediate
;
1782 intermediate
= intermediate
->backing_hd
;
1788 typedef struct BlkIntermediateStates
{
1789 BlockDriverState
*bs
;
1790 QSIMPLEQ_ENTRY(BlkIntermediateStates
) entry
;
1791 } BlkIntermediateStates
;
1795 * Drops images above 'base' up to and including 'top', and sets the image
1796 * above 'top' to have base as its backing file.
1798 * Requires that the overlay to 'top' is opened r/w, so that the backing file
1799 * information in 'bs' can be properly updated.
1801 * E.g., this will convert the following chain:
1802 * bottom <- base <- intermediate <- top <- active
1806 * bottom <- base <- active
1808 * It is allowed for bottom==base, in which case it converts:
1810 * base <- intermediate <- top <- active
1817 * if active == top, that is considered an error
1820 int bdrv_drop_intermediate(BlockDriverState
*active
, BlockDriverState
*top
,
1821 BlockDriverState
*base
)
1823 BlockDriverState
*intermediate
;
1824 BlockDriverState
*base_bs
= NULL
;
1825 BlockDriverState
*new_top_bs
= NULL
;
1826 BlkIntermediateStates
*intermediate_state
, *next
;
1829 QSIMPLEQ_HEAD(states_to_delete
, BlkIntermediateStates
) states_to_delete
;
1830 QSIMPLEQ_INIT(&states_to_delete
);
1832 if (!top
->drv
|| !base
->drv
) {
1836 new_top_bs
= bdrv_find_overlay(active
, top
);
1838 if (new_top_bs
== NULL
) {
1839 /* we could not find the image above 'top', this is an error */
1843 /* special case of new_top_bs->backing_hd already pointing to base - nothing
1844 * to do, no intermediate images */
1845 if (new_top_bs
->backing_hd
== base
) {
1852 /* now we will go down through the list, and add each BDS we find
1853 * into our deletion queue, until we hit the 'base'
1855 while (intermediate
) {
1856 intermediate_state
= g_malloc0(sizeof(BlkIntermediateStates
));
1857 intermediate_state
->bs
= intermediate
;
1858 QSIMPLEQ_INSERT_TAIL(&states_to_delete
, intermediate_state
, entry
);
1860 if (intermediate
->backing_hd
== base
) {
1861 base_bs
= intermediate
->backing_hd
;
1864 intermediate
= intermediate
->backing_hd
;
1866 if (base_bs
== NULL
) {
1867 /* something went wrong, we did not end at the base. safely
1868 * unravel everything, and exit with error */
1872 /* success - we can delete the intermediate states, and link top->base */
1873 ret
= bdrv_change_backing_file(new_top_bs
, base_bs
->filename
,
1874 base_bs
->drv
? base_bs
->drv
->format_name
: "");
1878 new_top_bs
->backing_hd
= base_bs
;
1881 QSIMPLEQ_FOREACH_SAFE(intermediate_state
, &states_to_delete
, entry
, next
) {
1882 /* so that bdrv_close() does not recursively close the chain */
1883 intermediate_state
->bs
->backing_hd
= NULL
;
1884 bdrv_delete(intermediate_state
->bs
);
1889 QSIMPLEQ_FOREACH_SAFE(intermediate_state
, &states_to_delete
, entry
, next
) {
1890 g_free(intermediate_state
);
1896 static int bdrv_check_byte_request(BlockDriverState
*bs
, int64_t offset
,
1901 if (!bdrv_is_inserted(bs
))
1907 len
= bdrv_getlength(bs
);
1912 if ((offset
> len
) || (len
- offset
< size
))
1918 static int bdrv_check_request(BlockDriverState
*bs
, int64_t sector_num
,
1921 return bdrv_check_byte_request(bs
, sector_num
* BDRV_SECTOR_SIZE
,
1922 nb_sectors
* BDRV_SECTOR_SIZE
);
1925 typedef struct RwCo
{
1926 BlockDriverState
*bs
;
1934 static void coroutine_fn
bdrv_rw_co_entry(void *opaque
)
1936 RwCo
*rwco
= opaque
;
1938 if (!rwco
->is_write
) {
1939 rwco
->ret
= bdrv_co_do_readv(rwco
->bs
, rwco
->sector_num
,
1940 rwco
->nb_sectors
, rwco
->qiov
, 0);
1942 rwco
->ret
= bdrv_co_do_writev(rwco
->bs
, rwco
->sector_num
,
1943 rwco
->nb_sectors
, rwco
->qiov
, 0);
1948 * Process a synchronous request using coroutines
1950 static int bdrv_rw_co(BlockDriverState
*bs
, int64_t sector_num
, uint8_t *buf
,
1951 int nb_sectors
, bool is_write
)
1954 struct iovec iov
= {
1955 .iov_base
= (void *)buf
,
1956 .iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
,
1961 .sector_num
= sector_num
,
1962 .nb_sectors
= nb_sectors
,
1964 .is_write
= is_write
,
1968 qemu_iovec_init_external(&qiov
, &iov
, 1);
1971 * In sync call context, when the vcpu is blocked, this throttling timer
1972 * will not fire; so the I/O throttling function has to be disabled here
1973 * if it has been enabled.
1975 if (bs
->io_limits_enabled
) {
1976 fprintf(stderr
, "Disabling I/O throttling on '%s' due "
1977 "to synchronous I/O.\n", bdrv_get_device_name(bs
));
1978 bdrv_io_limits_disable(bs
);
1981 if (qemu_in_coroutine()) {
1982 /* Fast-path if already in coroutine context */
1983 bdrv_rw_co_entry(&rwco
);
1985 co
= qemu_coroutine_create(bdrv_rw_co_entry
);
1986 qemu_coroutine_enter(co
, &rwco
);
1987 while (rwco
.ret
== NOT_DONE
) {
1994 /* return < 0 if error. See bdrv_write() for the return codes */
1995 int bdrv_read(BlockDriverState
*bs
, int64_t sector_num
,
1996 uint8_t *buf
, int nb_sectors
)
1998 return bdrv_rw_co(bs
, sector_num
, buf
, nb_sectors
, false);
2001 /* Just like bdrv_read(), but with I/O throttling temporarily disabled */
2002 int bdrv_read_unthrottled(BlockDriverState
*bs
, int64_t sector_num
,
2003 uint8_t *buf
, int nb_sectors
)
2008 enabled
= bs
->io_limits_enabled
;
2009 bs
->io_limits_enabled
= false;
2010 ret
= bdrv_read(bs
, 0, buf
, 1);
2011 bs
->io_limits_enabled
= enabled
;
2015 #define BITS_PER_LONG (sizeof(unsigned long) * 8)
2017 static void set_dirty_bitmap(BlockDriverState
*bs
, int64_t sector_num
,
2018 int nb_sectors
, int dirty
)
2021 unsigned long val
, idx
, bit
;
2023 start
= sector_num
/ BDRV_SECTORS_PER_DIRTY_CHUNK
;
2024 end
= (sector_num
+ nb_sectors
- 1) / BDRV_SECTORS_PER_DIRTY_CHUNK
;
2026 for (; start
<= end
; start
++) {
2027 idx
= start
/ BITS_PER_LONG
;
2028 bit
= start
% BITS_PER_LONG
;
2029 val
= bs
->dirty_bitmap
[idx
];
2031 if (!(val
& (1UL << bit
))) {
2036 if (val
& (1UL << bit
)) {
2038 val
&= ~(1UL << bit
);
2041 bs
->dirty_bitmap
[idx
] = val
;
2045 /* Return < 0 if error. Important errors are:
2046 -EIO generic I/O error (may happen for all errors)
2047 -ENOMEDIUM No media inserted.
2048 -EINVAL Invalid sector number or nb_sectors
2049 -EACCES Trying to write a read-only device
2051 int bdrv_write(BlockDriverState
*bs
, int64_t sector_num
,
2052 const uint8_t *buf
, int nb_sectors
)
2054 return bdrv_rw_co(bs
, sector_num
, (uint8_t *)buf
, nb_sectors
, true);
2057 int bdrv_pread(BlockDriverState
*bs
, int64_t offset
,
2058 void *buf
, int count1
)
2060 uint8_t tmp_buf
[BDRV_SECTOR_SIZE
];
2061 int len
, nb_sectors
, count
;
2066 /* first read to align to sector start */
2067 len
= (BDRV_SECTOR_SIZE
- offset
) & (BDRV_SECTOR_SIZE
- 1);
2070 sector_num
= offset
>> BDRV_SECTOR_BITS
;
2072 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
2074 memcpy(buf
, tmp_buf
+ (offset
& (BDRV_SECTOR_SIZE
- 1)), len
);
2082 /* read the sectors "in place" */
2083 nb_sectors
= count
>> BDRV_SECTOR_BITS
;
2084 if (nb_sectors
> 0) {
2085 if ((ret
= bdrv_read(bs
, sector_num
, buf
, nb_sectors
)) < 0)
2087 sector_num
+= nb_sectors
;
2088 len
= nb_sectors
<< BDRV_SECTOR_BITS
;
2093 /* add data from the last sector */
2095 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
2097 memcpy(buf
, tmp_buf
, count
);
2102 int bdrv_pwrite(BlockDriverState
*bs
, int64_t offset
,
2103 const void *buf
, int count1
)
2105 uint8_t tmp_buf
[BDRV_SECTOR_SIZE
];
2106 int len
, nb_sectors
, count
;
2111 /* first write to align to sector start */
2112 len
= (BDRV_SECTOR_SIZE
- offset
) & (BDRV_SECTOR_SIZE
- 1);
2115 sector_num
= offset
>> BDRV_SECTOR_BITS
;
2117 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
2119 memcpy(tmp_buf
+ (offset
& (BDRV_SECTOR_SIZE
- 1)), buf
, len
);
2120 if ((ret
= bdrv_write(bs
, sector_num
, tmp_buf
, 1)) < 0)
2129 /* write the sectors "in place" */
2130 nb_sectors
= count
>> BDRV_SECTOR_BITS
;
2131 if (nb_sectors
> 0) {
2132 if ((ret
= bdrv_write(bs
, sector_num
, buf
, nb_sectors
)) < 0)
2134 sector_num
+= nb_sectors
;
2135 len
= nb_sectors
<< BDRV_SECTOR_BITS
;
2140 /* add data from the last sector */
2142 if ((ret
= bdrv_read(bs
, sector_num
, tmp_buf
, 1)) < 0)
2144 memcpy(tmp_buf
, buf
, count
);
2145 if ((ret
= bdrv_write(bs
, sector_num
, tmp_buf
, 1)) < 0)
2152 * Writes to the file and ensures that no writes are reordered across this
2153 * request (acts as a barrier)
2155 * Returns 0 on success, -errno in error cases.
2157 int bdrv_pwrite_sync(BlockDriverState
*bs
, int64_t offset
,
2158 const void *buf
, int count
)
2162 ret
= bdrv_pwrite(bs
, offset
, buf
, count
);
2167 /* No flush needed for cache modes that already do it */
2168 if (bs
->enable_write_cache
) {
2175 static int coroutine_fn
bdrv_co_do_copy_on_readv(BlockDriverState
*bs
,
2176 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
2178 /* Perform I/O through a temporary buffer so that users who scribble over
2179 * their read buffer while the operation is in progress do not end up
2180 * modifying the image file. This is critical for zero-copy guest I/O
2181 * where anything might happen inside guest memory.
2183 void *bounce_buffer
;
2185 BlockDriver
*drv
= bs
->drv
;
2187 QEMUIOVector bounce_qiov
;
2188 int64_t cluster_sector_num
;
2189 int cluster_nb_sectors
;
2193 /* Cover entire cluster so no additional backing file I/O is required when
2194 * allocating cluster in the image file.
2196 round_to_clusters(bs
, sector_num
, nb_sectors
,
2197 &cluster_sector_num
, &cluster_nb_sectors
);
2199 trace_bdrv_co_do_copy_on_readv(bs
, sector_num
, nb_sectors
,
2200 cluster_sector_num
, cluster_nb_sectors
);
2202 iov
.iov_len
= cluster_nb_sectors
* BDRV_SECTOR_SIZE
;
2203 iov
.iov_base
= bounce_buffer
= qemu_blockalign(bs
, iov
.iov_len
);
2204 qemu_iovec_init_external(&bounce_qiov
, &iov
, 1);
2206 ret
= drv
->bdrv_co_readv(bs
, cluster_sector_num
, cluster_nb_sectors
,
2212 if (drv
->bdrv_co_write_zeroes
&&
2213 buffer_is_zero(bounce_buffer
, iov
.iov_len
)) {
2214 ret
= bdrv_co_do_write_zeroes(bs
, cluster_sector_num
,
2215 cluster_nb_sectors
);
2217 /* This does not change the data on the disk, it is not necessary
2218 * to flush even in cache=writethrough mode.
2220 ret
= drv
->bdrv_co_writev(bs
, cluster_sector_num
, cluster_nb_sectors
,
2225 /* It might be okay to ignore write errors for guest requests. If this
2226 * is a deliberate copy-on-read then we don't want to ignore the error.
2227 * Simply report it in all cases.
2232 skip_bytes
= (sector_num
- cluster_sector_num
) * BDRV_SECTOR_SIZE
;
2233 qemu_iovec_from_buf(qiov
, 0, bounce_buffer
+ skip_bytes
,
2234 nb_sectors
* BDRV_SECTOR_SIZE
);
2237 qemu_vfree(bounce_buffer
);
2242 * Handle a read request in coroutine context
2244 static int coroutine_fn
bdrv_co_do_readv(BlockDriverState
*bs
,
2245 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
2246 BdrvRequestFlags flags
)
2248 BlockDriver
*drv
= bs
->drv
;
2249 BdrvTrackedRequest req
;
2255 if (bdrv_check_request(bs
, sector_num
, nb_sectors
)) {
2259 /* throttling disk read I/O */
2260 if (bs
->io_limits_enabled
) {
2261 bdrv_io_limits_intercept(bs
, false, nb_sectors
);
2264 if (bs
->copy_on_read
) {
2265 flags
|= BDRV_REQ_COPY_ON_READ
;
2267 if (flags
& BDRV_REQ_COPY_ON_READ
) {
2268 bs
->copy_on_read_in_flight
++;
2271 if (bs
->copy_on_read_in_flight
) {
2272 wait_for_overlapping_requests(bs
, sector_num
, nb_sectors
);
2275 tracked_request_begin(&req
, bs
, sector_num
, nb_sectors
, false);
2277 if (flags
& BDRV_REQ_COPY_ON_READ
) {
2280 ret
= bdrv_co_is_allocated(bs
, sector_num
, nb_sectors
, &pnum
);
2285 if (!ret
|| pnum
!= nb_sectors
) {
2286 ret
= bdrv_co_do_copy_on_readv(bs
, sector_num
, nb_sectors
, qiov
);
2291 ret
= drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
2294 tracked_request_end(&req
);
2296 if (flags
& BDRV_REQ_COPY_ON_READ
) {
2297 bs
->copy_on_read_in_flight
--;
2303 int coroutine_fn
bdrv_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
2304 int nb_sectors
, QEMUIOVector
*qiov
)
2306 trace_bdrv_co_readv(bs
, sector_num
, nb_sectors
);
2308 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
, 0);
2311 int coroutine_fn
bdrv_co_copy_on_readv(BlockDriverState
*bs
,
2312 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
2314 trace_bdrv_co_copy_on_readv(bs
, sector_num
, nb_sectors
);
2316 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
,
2317 BDRV_REQ_COPY_ON_READ
);
2320 static int coroutine_fn
bdrv_co_do_write_zeroes(BlockDriverState
*bs
,
2321 int64_t sector_num
, int nb_sectors
)
2323 BlockDriver
*drv
= bs
->drv
;
2328 /* TODO Emulate only part of misaligned requests instead of letting block
2329 * drivers return -ENOTSUP and emulate everything */
2331 /* First try the efficient write zeroes operation */
2332 if (drv
->bdrv_co_write_zeroes
) {
2333 ret
= drv
->bdrv_co_write_zeroes(bs
, sector_num
, nb_sectors
);
2334 if (ret
!= -ENOTSUP
) {
2339 /* Fall back to bounce buffer if write zeroes is unsupported */
2340 iov
.iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
;
2341 iov
.iov_base
= qemu_blockalign(bs
, iov
.iov_len
);
2342 memset(iov
.iov_base
, 0, iov
.iov_len
);
2343 qemu_iovec_init_external(&qiov
, &iov
, 1);
2345 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, &qiov
);
2347 qemu_vfree(iov
.iov_base
);
2352 * Handle a write request in coroutine context
2354 static int coroutine_fn
bdrv_co_do_writev(BlockDriverState
*bs
,
2355 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
2356 BdrvRequestFlags flags
)
2358 BlockDriver
*drv
= bs
->drv
;
2359 BdrvTrackedRequest req
;
2365 if (bs
->read_only
) {
2368 if (bdrv_check_request(bs
, sector_num
, nb_sectors
)) {
2372 /* throttling disk write I/O */
2373 if (bs
->io_limits_enabled
) {
2374 bdrv_io_limits_intercept(bs
, true, nb_sectors
);
2377 if (bs
->copy_on_read_in_flight
) {
2378 wait_for_overlapping_requests(bs
, sector_num
, nb_sectors
);
2381 tracked_request_begin(&req
, bs
, sector_num
, nb_sectors
, true);
2383 if (flags
& BDRV_REQ_ZERO_WRITE
) {
2384 ret
= bdrv_co_do_write_zeroes(bs
, sector_num
, nb_sectors
);
2386 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, qiov
);
2389 if (ret
== 0 && !bs
->enable_write_cache
) {
2390 ret
= bdrv_co_flush(bs
);
2393 if (bs
->dirty_bitmap
) {
2394 set_dirty_bitmap(bs
, sector_num
, nb_sectors
, 1);
2397 if (bs
->wr_highest_sector
< sector_num
+ nb_sectors
- 1) {
2398 bs
->wr_highest_sector
= sector_num
+ nb_sectors
- 1;
2401 tracked_request_end(&req
);
2406 int coroutine_fn
bdrv_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
2407 int nb_sectors
, QEMUIOVector
*qiov
)
2409 trace_bdrv_co_writev(bs
, sector_num
, nb_sectors
);
2411 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, qiov
, 0);
2414 int coroutine_fn
bdrv_co_write_zeroes(BlockDriverState
*bs
,
2415 int64_t sector_num
, int nb_sectors
)
2417 trace_bdrv_co_write_zeroes(bs
, sector_num
, nb_sectors
);
2419 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, NULL
,
2420 BDRV_REQ_ZERO_WRITE
);
2424 * Truncate file to 'offset' bytes (needed only for file protocols)
2426 int bdrv_truncate(BlockDriverState
*bs
, int64_t offset
)
2428 BlockDriver
*drv
= bs
->drv
;
2432 if (!drv
->bdrv_truncate
)
2436 if (bdrv_in_use(bs
))
2438 ret
= drv
->bdrv_truncate(bs
, offset
);
2440 ret
= refresh_total_sectors(bs
, offset
>> BDRV_SECTOR_BITS
);
2441 bdrv_dev_resize_cb(bs
);
2447 * Length of a allocated file in bytes. Sparse files are counted by actual
2448 * allocated space. Return < 0 if error or unknown.
2450 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
2452 BlockDriver
*drv
= bs
->drv
;
2456 if (drv
->bdrv_get_allocated_file_size
) {
2457 return drv
->bdrv_get_allocated_file_size(bs
);
2460 return bdrv_get_allocated_file_size(bs
->file
);
2466 * Length of a file in bytes. Return < 0 if error or unknown.
2468 int64_t bdrv_getlength(BlockDriverState
*bs
)
2470 BlockDriver
*drv
= bs
->drv
;
2474 if (bs
->growable
|| bdrv_dev_has_removable_media(bs
)) {
2475 if (drv
->bdrv_getlength
) {
2476 return drv
->bdrv_getlength(bs
);
2479 return bs
->total_sectors
* BDRV_SECTOR_SIZE
;
2482 /* return 0 as number of sectors if no device present or error */
2483 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
2486 length
= bdrv_getlength(bs
);
2490 length
= length
>> BDRV_SECTOR_BITS
;
2491 *nb_sectors_ptr
= length
;
2494 /* throttling disk io limits */
2495 void bdrv_set_io_limits(BlockDriverState
*bs
,
2496 BlockIOLimit
*io_limits
)
2498 bs
->io_limits
= *io_limits
;
2499 bs
->io_limits_enabled
= bdrv_io_limits_enabled(bs
);
2502 void bdrv_set_on_error(BlockDriverState
*bs
, BlockdevOnError on_read_error
,
2503 BlockdevOnError on_write_error
)
2505 bs
->on_read_error
= on_read_error
;
2506 bs
->on_write_error
= on_write_error
;
2509 BlockdevOnError
bdrv_get_on_error(BlockDriverState
*bs
, bool is_read
)
2511 return is_read
? bs
->on_read_error
: bs
->on_write_error
;
2514 BlockErrorAction
bdrv_get_error_action(BlockDriverState
*bs
, bool is_read
, int error
)
2516 BlockdevOnError on_err
= is_read
? bs
->on_read_error
: bs
->on_write_error
;
2519 case BLOCKDEV_ON_ERROR_ENOSPC
:
2520 return (error
== ENOSPC
) ? BDRV_ACTION_STOP
: BDRV_ACTION_REPORT
;
2521 case BLOCKDEV_ON_ERROR_STOP
:
2522 return BDRV_ACTION_STOP
;
2523 case BLOCKDEV_ON_ERROR_REPORT
:
2524 return BDRV_ACTION_REPORT
;
2525 case BLOCKDEV_ON_ERROR_IGNORE
:
2526 return BDRV_ACTION_IGNORE
;
2532 /* This is done by device models because, while the block layer knows
2533 * about the error, it does not know whether an operation comes from
2534 * the device or the block layer (from a job, for example).
2536 void bdrv_error_action(BlockDriverState
*bs
, BlockErrorAction action
,
2537 bool is_read
, int error
)
2540 bdrv_emit_qmp_error_event(bs
, QEVENT_BLOCK_IO_ERROR
, action
, is_read
);
2541 if (action
== BDRV_ACTION_STOP
) {
2542 vm_stop(RUN_STATE_IO_ERROR
);
2543 bdrv_iostatus_set_err(bs
, error
);
2547 int bdrv_is_read_only(BlockDriverState
*bs
)
2549 return bs
->read_only
;
2552 int bdrv_is_sg(BlockDriverState
*bs
)
2557 int bdrv_enable_write_cache(BlockDriverState
*bs
)
2559 return bs
->enable_write_cache
;
2562 void bdrv_set_enable_write_cache(BlockDriverState
*bs
, bool wce
)
2564 bs
->enable_write_cache
= wce
;
2566 /* so a reopen() will preserve wce */
2568 bs
->open_flags
|= BDRV_O_CACHE_WB
;
2570 bs
->open_flags
&= ~BDRV_O_CACHE_WB
;
2574 int bdrv_is_encrypted(BlockDriverState
*bs
)
2576 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
)
2578 return bs
->encrypted
;
2581 int bdrv_key_required(BlockDriverState
*bs
)
2583 BlockDriverState
*backing_hd
= bs
->backing_hd
;
2585 if (backing_hd
&& backing_hd
->encrypted
&& !backing_hd
->valid_key
)
2587 return (bs
->encrypted
&& !bs
->valid_key
);
2590 int bdrv_set_key(BlockDriverState
*bs
, const char *key
)
2593 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
) {
2594 ret
= bdrv_set_key(bs
->backing_hd
, key
);
2600 if (!bs
->encrypted
) {
2602 } else if (!bs
->drv
|| !bs
->drv
->bdrv_set_key
) {
2605 ret
= bs
->drv
->bdrv_set_key(bs
, key
);
2608 } else if (!bs
->valid_key
) {
2610 /* call the change callback now, we skipped it on open */
2611 bdrv_dev_change_media_cb(bs
, true);
2616 const char *bdrv_get_format_name(BlockDriverState
*bs
)
2618 return bs
->drv
? bs
->drv
->format_name
: NULL
;
2621 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
2626 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
2627 it(opaque
, drv
->format_name
);
2631 BlockDriverState
*bdrv_find(const char *name
)
2633 BlockDriverState
*bs
;
2635 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2636 if (!strcmp(name
, bs
->device_name
)) {
2643 BlockDriverState
*bdrv_next(BlockDriverState
*bs
)
2646 return QTAILQ_FIRST(&bdrv_states
);
2648 return QTAILQ_NEXT(bs
, list
);
2651 void bdrv_iterate(void (*it
)(void *opaque
, BlockDriverState
*bs
), void *opaque
)
2653 BlockDriverState
*bs
;
2655 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2660 const char *bdrv_get_device_name(BlockDriverState
*bs
)
2662 return bs
->device_name
;
2665 int bdrv_get_flags(BlockDriverState
*bs
)
2667 return bs
->open_flags
;
2670 void bdrv_flush_all(void)
2672 BlockDriverState
*bs
;
2674 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2679 int bdrv_has_zero_init(BlockDriverState
*bs
)
2683 if (bs
->drv
->bdrv_has_zero_init
) {
2684 return bs
->drv
->bdrv_has_zero_init(bs
);
2690 typedef struct BdrvCoIsAllocatedData
{
2691 BlockDriverState
*bs
;
2697 } BdrvCoIsAllocatedData
;
2700 * Returns true iff the specified sector is present in the disk image. Drivers
2701 * not implementing the functionality are assumed to not support backing files,
2702 * hence all their sectors are reported as allocated.
2704 * If 'sector_num' is beyond the end of the disk image the return value is 0
2705 * and 'pnum' is set to 0.
2707 * 'pnum' is set to the number of sectors (including and immediately following
2708 * the specified sector) that are known to be in the same
2709 * allocated/unallocated state.
2711 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
2712 * beyond the end of the disk image it will be clamped.
2714 int coroutine_fn
bdrv_co_is_allocated(BlockDriverState
*bs
, int64_t sector_num
,
2715 int nb_sectors
, int *pnum
)
2719 if (sector_num
>= bs
->total_sectors
) {
2724 n
= bs
->total_sectors
- sector_num
;
2725 if (n
< nb_sectors
) {
2729 if (!bs
->drv
->bdrv_co_is_allocated
) {
2734 return bs
->drv
->bdrv_co_is_allocated(bs
, sector_num
, nb_sectors
, pnum
);
2737 /* Coroutine wrapper for bdrv_is_allocated() */
2738 static void coroutine_fn
bdrv_is_allocated_co_entry(void *opaque
)
2740 BdrvCoIsAllocatedData
*data
= opaque
;
2741 BlockDriverState
*bs
= data
->bs
;
2743 data
->ret
= bdrv_co_is_allocated(bs
, data
->sector_num
, data
->nb_sectors
,
2749 * Synchronous wrapper around bdrv_co_is_allocated().
2751 * See bdrv_co_is_allocated() for details.
2753 int bdrv_is_allocated(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
,
2757 BdrvCoIsAllocatedData data
= {
2759 .sector_num
= sector_num
,
2760 .nb_sectors
= nb_sectors
,
2765 co
= qemu_coroutine_create(bdrv_is_allocated_co_entry
);
2766 qemu_coroutine_enter(co
, &data
);
2767 while (!data
.done
) {
2774 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
2776 * Return true if the given sector is allocated in any image between
2777 * BASE and TOP (inclusive). BASE can be NULL to check if the given
2778 * sector is allocated in any image of the chain. Return false otherwise.
2780 * 'pnum' is set to the number of sectors (including and immediately following
2781 * the specified sector) that are known to be in the same
2782 * allocated/unallocated state.
2785 int coroutine_fn
bdrv_co_is_allocated_above(BlockDriverState
*top
,
2786 BlockDriverState
*base
,
2788 int nb_sectors
, int *pnum
)
2790 BlockDriverState
*intermediate
;
2791 int ret
, n
= nb_sectors
;
2794 while (intermediate
&& intermediate
!= base
) {
2796 ret
= bdrv_co_is_allocated(intermediate
, sector_num
, nb_sectors
,
2806 * [sector_num, nb_sectors] is unallocated on top but intermediate
2809 * [sector_num+x, nr_sectors] allocated.
2811 if (n
> pnum_inter
) {
2815 intermediate
= intermediate
->backing_hd
;
2822 BlockInfo
*bdrv_query_info(BlockDriverState
*bs
)
2824 BlockInfo
*info
= g_malloc0(sizeof(*info
));
2825 info
->device
= g_strdup(bs
->device_name
);
2826 info
->type
= g_strdup("unknown");
2827 info
->locked
= bdrv_dev_is_medium_locked(bs
);
2828 info
->removable
= bdrv_dev_has_removable_media(bs
);
2830 if (bdrv_dev_has_removable_media(bs
)) {
2831 info
->has_tray_open
= true;
2832 info
->tray_open
= bdrv_dev_is_tray_open(bs
);
2835 if (bdrv_iostatus_is_enabled(bs
)) {
2836 info
->has_io_status
= true;
2837 info
->io_status
= bs
->iostatus
;
2841 info
->has_inserted
= true;
2842 info
->inserted
= g_malloc0(sizeof(*info
->inserted
));
2843 info
->inserted
->file
= g_strdup(bs
->filename
);
2844 info
->inserted
->ro
= bs
->read_only
;
2845 info
->inserted
->drv
= g_strdup(bs
->drv
->format_name
);
2846 info
->inserted
->encrypted
= bs
->encrypted
;
2847 info
->inserted
->encryption_key_missing
= bdrv_key_required(bs
);
2849 if (bs
->backing_file
[0]) {
2850 info
->inserted
->has_backing_file
= true;
2851 info
->inserted
->backing_file
= g_strdup(bs
->backing_file
);
2854 info
->inserted
->backing_file_depth
= bdrv_get_backing_file_depth(bs
);
2856 if (bs
->io_limits_enabled
) {
2857 info
->inserted
->bps
=
2858 bs
->io_limits
.bps
[BLOCK_IO_LIMIT_TOTAL
];
2859 info
->inserted
->bps_rd
=
2860 bs
->io_limits
.bps
[BLOCK_IO_LIMIT_READ
];
2861 info
->inserted
->bps_wr
=
2862 bs
->io_limits
.bps
[BLOCK_IO_LIMIT_WRITE
];
2863 info
->inserted
->iops
=
2864 bs
->io_limits
.iops
[BLOCK_IO_LIMIT_TOTAL
];
2865 info
->inserted
->iops_rd
=
2866 bs
->io_limits
.iops
[BLOCK_IO_LIMIT_READ
];
2867 info
->inserted
->iops_wr
=
2868 bs
->io_limits
.iops
[BLOCK_IO_LIMIT_WRITE
];
2874 BlockInfoList
*qmp_query_block(Error
**errp
)
2876 BlockInfoList
*head
= NULL
, **p_next
= &head
;
2877 BlockDriverState
*bs
;
2879 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2880 BlockInfoList
*info
= g_malloc0(sizeof(*info
));
2881 info
->value
= bdrv_query_info(bs
);
2884 p_next
= &info
->next
;
2890 BlockStats
*bdrv_query_stats(const BlockDriverState
*bs
)
2894 s
= g_malloc0(sizeof(*s
));
2896 if (bs
->device_name
[0]) {
2897 s
->has_device
= true;
2898 s
->device
= g_strdup(bs
->device_name
);
2901 s
->stats
= g_malloc0(sizeof(*s
->stats
));
2902 s
->stats
->rd_bytes
= bs
->nr_bytes
[BDRV_ACCT_READ
];
2903 s
->stats
->wr_bytes
= bs
->nr_bytes
[BDRV_ACCT_WRITE
];
2904 s
->stats
->rd_operations
= bs
->nr_ops
[BDRV_ACCT_READ
];
2905 s
->stats
->wr_operations
= bs
->nr_ops
[BDRV_ACCT_WRITE
];
2906 s
->stats
->wr_highest_offset
= bs
->wr_highest_sector
* BDRV_SECTOR_SIZE
;
2907 s
->stats
->flush_operations
= bs
->nr_ops
[BDRV_ACCT_FLUSH
];
2908 s
->stats
->wr_total_time_ns
= bs
->total_time_ns
[BDRV_ACCT_WRITE
];
2909 s
->stats
->rd_total_time_ns
= bs
->total_time_ns
[BDRV_ACCT_READ
];
2910 s
->stats
->flush_total_time_ns
= bs
->total_time_ns
[BDRV_ACCT_FLUSH
];
2913 s
->has_parent
= true;
2914 s
->parent
= bdrv_query_stats(bs
->file
);
2920 BlockStatsList
*qmp_query_blockstats(Error
**errp
)
2922 BlockStatsList
*head
= NULL
, **p_next
= &head
;
2923 BlockDriverState
*bs
;
2925 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
2926 BlockStatsList
*info
= g_malloc0(sizeof(*info
));
2927 info
->value
= bdrv_query_stats(bs
);
2930 p_next
= &info
->next
;
2936 const char *bdrv_get_encrypted_filename(BlockDriverState
*bs
)
2938 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
)
2939 return bs
->backing_file
;
2940 else if (bs
->encrypted
)
2941 return bs
->filename
;
2946 void bdrv_get_backing_filename(BlockDriverState
*bs
,
2947 char *filename
, int filename_size
)
2949 pstrcpy(filename
, filename_size
, bs
->backing_file
);
2952 int bdrv_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
2953 const uint8_t *buf
, int nb_sectors
)
2955 BlockDriver
*drv
= bs
->drv
;
2958 if (!drv
->bdrv_write_compressed
)
2960 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
2963 if (bs
->dirty_bitmap
) {
2964 set_dirty_bitmap(bs
, sector_num
, nb_sectors
, 1);
2967 return drv
->bdrv_write_compressed(bs
, sector_num
, buf
, nb_sectors
);
2970 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2972 BlockDriver
*drv
= bs
->drv
;
2975 if (!drv
->bdrv_get_info
)
2977 memset(bdi
, 0, sizeof(*bdi
));
2978 return drv
->bdrv_get_info(bs
, bdi
);
2981 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
2982 int64_t pos
, int size
)
2984 BlockDriver
*drv
= bs
->drv
;
2987 if (drv
->bdrv_save_vmstate
)
2988 return drv
->bdrv_save_vmstate(bs
, buf
, pos
, size
);
2990 return bdrv_save_vmstate(bs
->file
, buf
, pos
, size
);
2994 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
2995 int64_t pos
, int size
)
2997 BlockDriver
*drv
= bs
->drv
;
3000 if (drv
->bdrv_load_vmstate
)
3001 return drv
->bdrv_load_vmstate(bs
, buf
, pos
, size
);
3003 return bdrv_load_vmstate(bs
->file
, buf
, pos
, size
);
3007 void bdrv_debug_event(BlockDriverState
*bs
, BlkDebugEvent event
)
3009 BlockDriver
*drv
= bs
->drv
;
3011 if (!drv
|| !drv
->bdrv_debug_event
) {
3015 drv
->bdrv_debug_event(bs
, event
);
3019 /**************************************************************/
3020 /* handling of snapshots */
3022 int bdrv_can_snapshot(BlockDriverState
*bs
)
3024 BlockDriver
*drv
= bs
->drv
;
3025 if (!drv
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
)) {
3029 if (!drv
->bdrv_snapshot_create
) {
3030 if (bs
->file
!= NULL
) {
3031 return bdrv_can_snapshot(bs
->file
);
3039 int bdrv_is_snapshot(BlockDriverState
*bs
)
3041 return !!(bs
->open_flags
& BDRV_O_SNAPSHOT
);
3044 BlockDriverState
*bdrv_snapshots(void)
3046 BlockDriverState
*bs
;
3049 return bs_snapshots
;
3053 while ((bs
= bdrv_next(bs
))) {
3054 if (bdrv_can_snapshot(bs
)) {
3062 int bdrv_snapshot_create(BlockDriverState
*bs
,
3063 QEMUSnapshotInfo
*sn_info
)
3065 BlockDriver
*drv
= bs
->drv
;
3068 if (drv
->bdrv_snapshot_create
)
3069 return drv
->bdrv_snapshot_create(bs
, sn_info
);
3071 return bdrv_snapshot_create(bs
->file
, sn_info
);
3075 int bdrv_snapshot_goto(BlockDriverState
*bs
,
3076 const char *snapshot_id
)
3078 BlockDriver
*drv
= bs
->drv
;
3083 if (drv
->bdrv_snapshot_goto
)
3084 return drv
->bdrv_snapshot_goto(bs
, snapshot_id
);
3087 drv
->bdrv_close(bs
);
3088 ret
= bdrv_snapshot_goto(bs
->file
, snapshot_id
);
3089 open_ret
= drv
->bdrv_open(bs
, bs
->open_flags
);
3091 bdrv_delete(bs
->file
);
3101 int bdrv_snapshot_delete(BlockDriverState
*bs
, const char *snapshot_id
)
3103 BlockDriver
*drv
= bs
->drv
;
3106 if (drv
->bdrv_snapshot_delete
)
3107 return drv
->bdrv_snapshot_delete(bs
, snapshot_id
);
3109 return bdrv_snapshot_delete(bs
->file
, snapshot_id
);
3113 int bdrv_snapshot_list(BlockDriverState
*bs
,
3114 QEMUSnapshotInfo
**psn_info
)
3116 BlockDriver
*drv
= bs
->drv
;
3119 if (drv
->bdrv_snapshot_list
)
3120 return drv
->bdrv_snapshot_list(bs
, psn_info
);
3122 return bdrv_snapshot_list(bs
->file
, psn_info
);
3126 int bdrv_snapshot_load_tmp(BlockDriverState
*bs
,
3127 const char *snapshot_name
)
3129 BlockDriver
*drv
= bs
->drv
;
3133 if (!bs
->read_only
) {
3136 if (drv
->bdrv_snapshot_load_tmp
) {
3137 return drv
->bdrv_snapshot_load_tmp(bs
, snapshot_name
);
3142 /* backing_file can either be relative, or absolute, or a protocol. If it is
3143 * relative, it must be relative to the chain. So, passing in bs->filename
3144 * from a BDS as backing_file should not be done, as that may be relative to
3145 * the CWD rather than the chain. */
3146 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
3147 const char *backing_file
)
3149 char *filename_full
= NULL
;
3150 char *backing_file_full
= NULL
;
3151 char *filename_tmp
= NULL
;
3152 int is_protocol
= 0;
3153 BlockDriverState
*curr_bs
= NULL
;
3154 BlockDriverState
*retval
= NULL
;
3156 if (!bs
|| !bs
->drv
|| !backing_file
) {
3160 filename_full
= g_malloc(PATH_MAX
);
3161 backing_file_full
= g_malloc(PATH_MAX
);
3162 filename_tmp
= g_malloc(PATH_MAX
);
3164 is_protocol
= path_has_protocol(backing_file
);
3166 for (curr_bs
= bs
; curr_bs
->backing_hd
; curr_bs
= curr_bs
->backing_hd
) {
3168 /* If either of the filename paths is actually a protocol, then
3169 * compare unmodified paths; otherwise make paths relative */
3170 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
3171 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
3172 retval
= curr_bs
->backing_hd
;
3176 /* If not an absolute filename path, make it relative to the current
3177 * image's filename path */
3178 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
3181 /* We are going to compare absolute pathnames */
3182 if (!realpath(filename_tmp
, filename_full
)) {
3186 /* We need to make sure the backing filename we are comparing against
3187 * is relative to the current image filename (or absolute) */
3188 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
3189 curr_bs
->backing_file
);
3191 if (!realpath(filename_tmp
, backing_file_full
)) {
3195 if (strcmp(backing_file_full
, filename_full
) == 0) {
3196 retval
= curr_bs
->backing_hd
;
3202 g_free(filename_full
);
3203 g_free(backing_file_full
);
3204 g_free(filename_tmp
);
3208 int bdrv_get_backing_file_depth(BlockDriverState
*bs
)
3214 if (!bs
->backing_hd
) {
3218 return 1 + bdrv_get_backing_file_depth(bs
->backing_hd
);
3221 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
3223 BlockDriverState
*curr_bs
= NULL
;
3231 while (curr_bs
->backing_hd
) {
3232 curr_bs
= curr_bs
->backing_hd
;
3237 #define NB_SUFFIXES 4
3239 char *get_human_readable_size(char *buf
, int buf_size
, int64_t size
)
3241 static const char suffixes
[NB_SUFFIXES
] = "KMGT";
3246 snprintf(buf
, buf_size
, "%" PRId64
, size
);
3249 for(i
= 0; i
< NB_SUFFIXES
; i
++) {
3250 if (size
< (10 * base
)) {
3251 snprintf(buf
, buf_size
, "%0.1f%c",
3252 (double)size
/ base
,
3255 } else if (size
< (1000 * base
) || i
== (NB_SUFFIXES
- 1)) {
3256 snprintf(buf
, buf_size
, "%" PRId64
"%c",
3257 ((size
+ (base
>> 1)) / base
),
3267 char *bdrv_snapshot_dump(char *buf
, int buf_size
, QEMUSnapshotInfo
*sn
)
3269 char buf1
[128], date_buf
[128], clock_buf
[128];
3279 snprintf(buf
, buf_size
,
3280 "%-10s%-20s%7s%20s%15s",
3281 "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
3285 ptm
= localtime(&ti
);
3286 strftime(date_buf
, sizeof(date_buf
),
3287 "%Y-%m-%d %H:%M:%S", ptm
);
3289 localtime_r(&ti
, &tm
);
3290 strftime(date_buf
, sizeof(date_buf
),
3291 "%Y-%m-%d %H:%M:%S", &tm
);
3293 secs
= sn
->vm_clock_nsec
/ 1000000000;
3294 snprintf(clock_buf
, sizeof(clock_buf
),
3295 "%02d:%02d:%02d.%03d",
3297 (int)((secs
/ 60) % 60),
3299 (int)((sn
->vm_clock_nsec
/ 1000000) % 1000));
3300 snprintf(buf
, buf_size
,
3301 "%-10s%-20s%7s%20s%15s",
3302 sn
->id_str
, sn
->name
,
3303 get_human_readable_size(buf1
, sizeof(buf1
), sn
->vm_state_size
),
3310 /**************************************************************/
3313 BlockDriverAIOCB
*bdrv_aio_readv(BlockDriverState
*bs
, int64_t sector_num
,
3314 QEMUIOVector
*qiov
, int nb_sectors
,
3315 BlockDriverCompletionFunc
*cb
, void *opaque
)
3317 trace_bdrv_aio_readv(bs
, sector_num
, nb_sectors
, opaque
);
3319 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
,
3323 BlockDriverAIOCB
*bdrv_aio_writev(BlockDriverState
*bs
, int64_t sector_num
,
3324 QEMUIOVector
*qiov
, int nb_sectors
,
3325 BlockDriverCompletionFunc
*cb
, void *opaque
)
3327 trace_bdrv_aio_writev(bs
, sector_num
, nb_sectors
, opaque
);
3329 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
,
3334 typedef struct MultiwriteCB
{
3339 BlockDriverCompletionFunc
*cb
;
3341 QEMUIOVector
*free_qiov
;
3345 static void multiwrite_user_cb(MultiwriteCB
*mcb
)
3349 for (i
= 0; i
< mcb
->num_callbacks
; i
++) {
3350 mcb
->callbacks
[i
].cb(mcb
->callbacks
[i
].opaque
, mcb
->error
);
3351 if (mcb
->callbacks
[i
].free_qiov
) {
3352 qemu_iovec_destroy(mcb
->callbacks
[i
].free_qiov
);
3354 g_free(mcb
->callbacks
[i
].free_qiov
);
3358 static void multiwrite_cb(void *opaque
, int ret
)
3360 MultiwriteCB
*mcb
= opaque
;
3362 trace_multiwrite_cb(mcb
, ret
);
3364 if (ret
< 0 && !mcb
->error
) {
3368 mcb
->num_requests
--;
3369 if (mcb
->num_requests
== 0) {
3370 multiwrite_user_cb(mcb
);
3375 static int multiwrite_req_compare(const void *a
, const void *b
)
3377 const BlockRequest
*req1
= a
, *req2
= b
;
3380 * Note that we can't simply subtract req2->sector from req1->sector
3381 * here as that could overflow the return value.
3383 if (req1
->sector
> req2
->sector
) {
3385 } else if (req1
->sector
< req2
->sector
) {
3393 * Takes a bunch of requests and tries to merge them. Returns the number of
3394 * requests that remain after merging.
3396 static int multiwrite_merge(BlockDriverState
*bs
, BlockRequest
*reqs
,
3397 int num_reqs
, MultiwriteCB
*mcb
)
3401 // Sort requests by start sector
3402 qsort(reqs
, num_reqs
, sizeof(*reqs
), &multiwrite_req_compare
);
3404 // Check if adjacent requests touch the same clusters. If so, combine them,
3405 // filling up gaps with zero sectors.
3407 for (i
= 1; i
< num_reqs
; i
++) {
3409 int64_t oldreq_last
= reqs
[outidx
].sector
+ reqs
[outidx
].nb_sectors
;
3411 // Handle exactly sequential writes and overlapping writes.
3412 if (reqs
[i
].sector
<= oldreq_last
) {
3416 if (reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1 > IOV_MAX
) {
3422 QEMUIOVector
*qiov
= g_malloc0(sizeof(*qiov
));
3423 qemu_iovec_init(qiov
,
3424 reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1);
3426 // Add the first request to the merged one. If the requests are
3427 // overlapping, drop the last sectors of the first request.
3428 size
= (reqs
[i
].sector
- reqs
[outidx
].sector
) << 9;
3429 qemu_iovec_concat(qiov
, reqs
[outidx
].qiov
, 0, size
);
3431 // We should need to add any zeros between the two requests
3432 assert (reqs
[i
].sector
<= oldreq_last
);
3434 // Add the second request
3435 qemu_iovec_concat(qiov
, reqs
[i
].qiov
, 0, reqs
[i
].qiov
->size
);
3437 reqs
[outidx
].nb_sectors
= qiov
->size
>> 9;
3438 reqs
[outidx
].qiov
= qiov
;
3440 mcb
->callbacks
[i
].free_qiov
= reqs
[outidx
].qiov
;
3443 reqs
[outidx
].sector
= reqs
[i
].sector
;
3444 reqs
[outidx
].nb_sectors
= reqs
[i
].nb_sectors
;
3445 reqs
[outidx
].qiov
= reqs
[i
].qiov
;
3453 * Submit multiple AIO write requests at once.
3455 * On success, the function returns 0 and all requests in the reqs array have
3456 * been submitted. In error case this function returns -1, and any of the
3457 * requests may or may not be submitted yet. In particular, this means that the
3458 * callback will be called for some of the requests, for others it won't. The
3459 * caller must check the error field of the BlockRequest to wait for the right
3460 * callbacks (if error != 0, no callback will be called).
3462 * The implementation may modify the contents of the reqs array, e.g. to merge
3463 * requests. However, the fields opaque and error are left unmodified as they
3464 * are used to signal failure for a single request to the caller.
3466 int bdrv_aio_multiwrite(BlockDriverState
*bs
, BlockRequest
*reqs
, int num_reqs
)
3471 /* don't submit writes if we don't have a medium */
3472 if (bs
->drv
== NULL
) {
3473 for (i
= 0; i
< num_reqs
; i
++) {
3474 reqs
[i
].error
= -ENOMEDIUM
;
3479 if (num_reqs
== 0) {
3483 // Create MultiwriteCB structure
3484 mcb
= g_malloc0(sizeof(*mcb
) + num_reqs
* sizeof(*mcb
->callbacks
));
3485 mcb
->num_requests
= 0;
3486 mcb
->num_callbacks
= num_reqs
;
3488 for (i
= 0; i
< num_reqs
; i
++) {
3489 mcb
->callbacks
[i
].cb
= reqs
[i
].cb
;
3490 mcb
->callbacks
[i
].opaque
= reqs
[i
].opaque
;
3493 // Check for mergable requests
3494 num_reqs
= multiwrite_merge(bs
, reqs
, num_reqs
, mcb
);
3496 trace_bdrv_aio_multiwrite(mcb
, mcb
->num_callbacks
, num_reqs
);
3498 /* Run the aio requests. */
3499 mcb
->num_requests
= num_reqs
;
3500 for (i
= 0; i
< num_reqs
; i
++) {
3501 bdrv_aio_writev(bs
, reqs
[i
].sector
, reqs
[i
].qiov
,
3502 reqs
[i
].nb_sectors
, multiwrite_cb
, mcb
);
3508 void bdrv_aio_cancel(BlockDriverAIOCB
*acb
)
3510 acb
->pool
->cancel(acb
);
3513 /* block I/O throttling */
3514 static bool bdrv_exceed_bps_limits(BlockDriverState
*bs
, int nb_sectors
,
3515 bool is_write
, double elapsed_time
, uint64_t *wait
)
3517 uint64_t bps_limit
= 0;
3518 double bytes_limit
, bytes_base
, bytes_res
;
3519 double slice_time
, wait_time
;
3521 if (bs
->io_limits
.bps
[BLOCK_IO_LIMIT_TOTAL
]) {
3522 bps_limit
= bs
->io_limits
.bps
[BLOCK_IO_LIMIT_TOTAL
];
3523 } else if (bs
->io_limits
.bps
[is_write
]) {
3524 bps_limit
= bs
->io_limits
.bps
[is_write
];
3533 slice_time
= bs
->slice_end
- bs
->slice_start
;
3534 slice_time
/= (NANOSECONDS_PER_SECOND
);
3535 bytes_limit
= bps_limit
* slice_time
;
3536 bytes_base
= bs
->nr_bytes
[is_write
] - bs
->io_base
.bytes
[is_write
];
3537 if (bs
->io_limits
.bps
[BLOCK_IO_LIMIT_TOTAL
]) {
3538 bytes_base
+= bs
->nr_bytes
[!is_write
] - bs
->io_base
.bytes
[!is_write
];
3541 /* bytes_base: the bytes of data which have been read/written; and
3542 * it is obtained from the history statistic info.
3543 * bytes_res: the remaining bytes of data which need to be read/written.
3544 * (bytes_base + bytes_res) / bps_limit: used to calcuate
3545 * the total time for completing reading/writting all data.
3547 bytes_res
= (unsigned) nb_sectors
* BDRV_SECTOR_SIZE
;
3549 if (bytes_base
+ bytes_res
<= bytes_limit
) {
3557 /* Calc approx time to dispatch */
3558 wait_time
= (bytes_base
+ bytes_res
) / bps_limit
- elapsed_time
;
3560 /* When the I/O rate at runtime exceeds the limits,
3561 * bs->slice_end need to be extended in order that the current statistic
3562 * info can be kept until the timer fire, so it is increased and tuned
3563 * based on the result of experiment.
3565 bs
->slice_time
= wait_time
* BLOCK_IO_SLICE_TIME
* 10;
3566 bs
->slice_end
+= bs
->slice_time
- 3 * BLOCK_IO_SLICE_TIME
;
3568 *wait
= wait_time
* BLOCK_IO_SLICE_TIME
* 10;
3574 static bool bdrv_exceed_iops_limits(BlockDriverState
*bs
, bool is_write
,
3575 double elapsed_time
, uint64_t *wait
)
3577 uint64_t iops_limit
= 0;
3578 double ios_limit
, ios_base
;
3579 double slice_time
, wait_time
;
3581 if (bs
->io_limits
.iops
[BLOCK_IO_LIMIT_TOTAL
]) {
3582 iops_limit
= bs
->io_limits
.iops
[BLOCK_IO_LIMIT_TOTAL
];
3583 } else if (bs
->io_limits
.iops
[is_write
]) {
3584 iops_limit
= bs
->io_limits
.iops
[is_write
];
3593 slice_time
= bs
->slice_end
- bs
->slice_start
;
3594 slice_time
/= (NANOSECONDS_PER_SECOND
);
3595 ios_limit
= iops_limit
* slice_time
;
3596 ios_base
= bs
->nr_ops
[is_write
] - bs
->io_base
.ios
[is_write
];
3597 if (bs
->io_limits
.iops
[BLOCK_IO_LIMIT_TOTAL
]) {
3598 ios_base
+= bs
->nr_ops
[!is_write
] - bs
->io_base
.ios
[!is_write
];
3601 if (ios_base
+ 1 <= ios_limit
) {
3609 /* Calc approx time to dispatch */
3610 wait_time
= (ios_base
+ 1) / iops_limit
;
3611 if (wait_time
> elapsed_time
) {
3612 wait_time
= wait_time
- elapsed_time
;
3617 bs
->slice_time
= wait_time
* BLOCK_IO_SLICE_TIME
* 10;
3618 bs
->slice_end
+= bs
->slice_time
- 3 * BLOCK_IO_SLICE_TIME
;
3620 *wait
= wait_time
* BLOCK_IO_SLICE_TIME
* 10;
3626 static bool bdrv_exceed_io_limits(BlockDriverState
*bs
, int nb_sectors
,
3627 bool is_write
, int64_t *wait
)
3629 int64_t now
, max_wait
;
3630 uint64_t bps_wait
= 0, iops_wait
= 0;
3631 double elapsed_time
;
3632 int bps_ret
, iops_ret
;
3634 now
= qemu_get_clock_ns(vm_clock
);
3635 if ((bs
->slice_start
< now
)
3636 && (bs
->slice_end
> now
)) {
3637 bs
->slice_end
= now
+ bs
->slice_time
;
3639 bs
->slice_time
= 5 * BLOCK_IO_SLICE_TIME
;
3640 bs
->slice_start
= now
;
3641 bs
->slice_end
= now
+ bs
->slice_time
;
3643 bs
->io_base
.bytes
[is_write
] = bs
->nr_bytes
[is_write
];
3644 bs
->io_base
.bytes
[!is_write
] = bs
->nr_bytes
[!is_write
];
3646 bs
->io_base
.ios
[is_write
] = bs
->nr_ops
[is_write
];
3647 bs
->io_base
.ios
[!is_write
] = bs
->nr_ops
[!is_write
];
3650 elapsed_time
= now
- bs
->slice_start
;
3651 elapsed_time
/= (NANOSECONDS_PER_SECOND
);
3653 bps_ret
= bdrv_exceed_bps_limits(bs
, nb_sectors
,
3654 is_write
, elapsed_time
, &bps_wait
);
3655 iops_ret
= bdrv_exceed_iops_limits(bs
, is_write
,
3656 elapsed_time
, &iops_wait
);
3657 if (bps_ret
|| iops_ret
) {
3658 max_wait
= bps_wait
> iops_wait
? bps_wait
: iops_wait
;
3663 now
= qemu_get_clock_ns(vm_clock
);
3664 if (bs
->slice_end
< now
+ max_wait
) {
3665 bs
->slice_end
= now
+ max_wait
;
3678 /**************************************************************/
3679 /* async block device emulation */
3681 typedef struct BlockDriverAIOCBSync
{
3682 BlockDriverAIOCB common
;
3685 /* vector translation state */
3689 } BlockDriverAIOCBSync
;
3691 static void bdrv_aio_cancel_em(BlockDriverAIOCB
*blockacb
)
3693 BlockDriverAIOCBSync
*acb
=
3694 container_of(blockacb
, BlockDriverAIOCBSync
, common
);
3695 qemu_bh_delete(acb
->bh
);
3697 qemu_aio_release(acb
);
3700 static AIOPool bdrv_em_aio_pool
= {
3701 .aiocb_size
= sizeof(BlockDriverAIOCBSync
),
3702 .cancel
= bdrv_aio_cancel_em
,
3705 static void bdrv_aio_bh_cb(void *opaque
)
3707 BlockDriverAIOCBSync
*acb
= opaque
;
3710 qemu_iovec_from_buf(acb
->qiov
, 0, acb
->bounce
, acb
->qiov
->size
);
3711 qemu_vfree(acb
->bounce
);
3712 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
3713 qemu_bh_delete(acb
->bh
);
3715 qemu_aio_release(acb
);
3718 static BlockDriverAIOCB
*bdrv_aio_rw_vector(BlockDriverState
*bs
,
3722 BlockDriverCompletionFunc
*cb
,
3727 BlockDriverAIOCBSync
*acb
;
3729 acb
= qemu_aio_get(&bdrv_em_aio_pool
, bs
, cb
, opaque
);
3730 acb
->is_write
= is_write
;
3732 acb
->bounce
= qemu_blockalign(bs
, qiov
->size
);
3733 acb
->bh
= qemu_bh_new(bdrv_aio_bh_cb
, acb
);
3736 qemu_iovec_to_buf(acb
->qiov
, 0, acb
->bounce
, qiov
->size
);
3737 acb
->ret
= bs
->drv
->bdrv_write(bs
, sector_num
, acb
->bounce
, nb_sectors
);
3739 acb
->ret
= bs
->drv
->bdrv_read(bs
, sector_num
, acb
->bounce
, nb_sectors
);
3742 qemu_bh_schedule(acb
->bh
);
3744 return &acb
->common
;
3747 static BlockDriverAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
3748 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
3749 BlockDriverCompletionFunc
*cb
, void *opaque
)
3751 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 0);
3754 static BlockDriverAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
3755 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
3756 BlockDriverCompletionFunc
*cb
, void *opaque
)
3758 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 1);
3762 typedef struct BlockDriverAIOCBCoroutine
{
3763 BlockDriverAIOCB common
;
3767 } BlockDriverAIOCBCoroutine
;
3769 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB
*blockacb
)
3774 static AIOPool bdrv_em_co_aio_pool
= {
3775 .aiocb_size
= sizeof(BlockDriverAIOCBCoroutine
),
3776 .cancel
= bdrv_aio_co_cancel_em
,
3779 static void bdrv_co_em_bh(void *opaque
)
3781 BlockDriverAIOCBCoroutine
*acb
= opaque
;
3783 acb
->common
.cb(acb
->common
.opaque
, acb
->req
.error
);
3784 qemu_bh_delete(acb
->bh
);
3785 qemu_aio_release(acb
);
3788 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
3789 static void coroutine_fn
bdrv_co_do_rw(void *opaque
)
3791 BlockDriverAIOCBCoroutine
*acb
= opaque
;
3792 BlockDriverState
*bs
= acb
->common
.bs
;
3794 if (!acb
->is_write
) {
3795 acb
->req
.error
= bdrv_co_do_readv(bs
, acb
->req
.sector
,
3796 acb
->req
.nb_sectors
, acb
->req
.qiov
, 0);
3798 acb
->req
.error
= bdrv_co_do_writev(bs
, acb
->req
.sector
,
3799 acb
->req
.nb_sectors
, acb
->req
.qiov
, 0);
3802 acb
->bh
= qemu_bh_new(bdrv_co_em_bh
, acb
);
3803 qemu_bh_schedule(acb
->bh
);
3806 static BlockDriverAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
3810 BlockDriverCompletionFunc
*cb
,
3815 BlockDriverAIOCBCoroutine
*acb
;
3817 acb
= qemu_aio_get(&bdrv_em_co_aio_pool
, bs
, cb
, opaque
);
3818 acb
->req
.sector
= sector_num
;
3819 acb
->req
.nb_sectors
= nb_sectors
;
3820 acb
->req
.qiov
= qiov
;
3821 acb
->is_write
= is_write
;
3823 co
= qemu_coroutine_create(bdrv_co_do_rw
);
3824 qemu_coroutine_enter(co
, acb
);
3826 return &acb
->common
;
3829 static void coroutine_fn
bdrv_aio_flush_co_entry(void *opaque
)
3831 BlockDriverAIOCBCoroutine
*acb
= opaque
;
3832 BlockDriverState
*bs
= acb
->common
.bs
;
3834 acb
->req
.error
= bdrv_co_flush(bs
);
3835 acb
->bh
= qemu_bh_new(bdrv_co_em_bh
, acb
);
3836 qemu_bh_schedule(acb
->bh
);
3839 BlockDriverAIOCB
*bdrv_aio_flush(BlockDriverState
*bs
,
3840 BlockDriverCompletionFunc
*cb
, void *opaque
)
3842 trace_bdrv_aio_flush(bs
, opaque
);
3845 BlockDriverAIOCBCoroutine
*acb
;
3847 acb
= qemu_aio_get(&bdrv_em_co_aio_pool
, bs
, cb
, opaque
);
3848 co
= qemu_coroutine_create(bdrv_aio_flush_co_entry
);
3849 qemu_coroutine_enter(co
, acb
);
3851 return &acb
->common
;
3854 static void coroutine_fn
bdrv_aio_discard_co_entry(void *opaque
)
3856 BlockDriverAIOCBCoroutine
*acb
= opaque
;
3857 BlockDriverState
*bs
= acb
->common
.bs
;
3859 acb
->req
.error
= bdrv_co_discard(bs
, acb
->req
.sector
, acb
->req
.nb_sectors
);
3860 acb
->bh
= qemu_bh_new(bdrv_co_em_bh
, acb
);
3861 qemu_bh_schedule(acb
->bh
);
3864 BlockDriverAIOCB
*bdrv_aio_discard(BlockDriverState
*bs
,
3865 int64_t sector_num
, int nb_sectors
,
3866 BlockDriverCompletionFunc
*cb
, void *opaque
)
3869 BlockDriverAIOCBCoroutine
*acb
;
3871 trace_bdrv_aio_discard(bs
, sector_num
, nb_sectors
, opaque
);
3873 acb
= qemu_aio_get(&bdrv_em_co_aio_pool
, bs
, cb
, opaque
);
3874 acb
->req
.sector
= sector_num
;
3875 acb
->req
.nb_sectors
= nb_sectors
;
3876 co
= qemu_coroutine_create(bdrv_aio_discard_co_entry
);
3877 qemu_coroutine_enter(co
, acb
);
3879 return &acb
->common
;
3882 void bdrv_init(void)
3884 module_call_init(MODULE_INIT_BLOCK
);
3887 void bdrv_init_with_whitelist(void)
3889 use_bdrv_whitelist
= 1;
3893 void *qemu_aio_get(AIOPool
*pool
, BlockDriverState
*bs
,
3894 BlockDriverCompletionFunc
*cb
, void *opaque
)
3896 BlockDriverAIOCB
*acb
;
3898 if (pool
->free_aiocb
) {
3899 acb
= pool
->free_aiocb
;
3900 pool
->free_aiocb
= acb
->next
;
3902 acb
= g_malloc0(pool
->aiocb_size
);
3907 acb
->opaque
= opaque
;
3911 void qemu_aio_release(void *p
)
3913 BlockDriverAIOCB
*acb
= (BlockDriverAIOCB
*)p
;
3914 AIOPool
*pool
= acb
->pool
;
3915 acb
->next
= pool
->free_aiocb
;
3916 pool
->free_aiocb
= acb
;
3919 /**************************************************************/
3920 /* Coroutine block device emulation */
3922 typedef struct CoroutineIOCompletion
{
3923 Coroutine
*coroutine
;
3925 } CoroutineIOCompletion
;
3927 static void bdrv_co_io_em_complete(void *opaque
, int ret
)
3929 CoroutineIOCompletion
*co
= opaque
;
3932 qemu_coroutine_enter(co
->coroutine
, NULL
);
3935 static int coroutine_fn
bdrv_co_io_em(BlockDriverState
*bs
, int64_t sector_num
,
3936 int nb_sectors
, QEMUIOVector
*iov
,
3939 CoroutineIOCompletion co
= {
3940 .coroutine
= qemu_coroutine_self(),
3942 BlockDriverAIOCB
*acb
;
3945 acb
= bs
->drv
->bdrv_aio_writev(bs
, sector_num
, iov
, nb_sectors
,
3946 bdrv_co_io_em_complete
, &co
);
3948 acb
= bs
->drv
->bdrv_aio_readv(bs
, sector_num
, iov
, nb_sectors
,
3949 bdrv_co_io_em_complete
, &co
);
3952 trace_bdrv_co_io_em(bs
, sector_num
, nb_sectors
, is_write
, acb
);
3956 qemu_coroutine_yield();
3961 static int coroutine_fn
bdrv_co_readv_em(BlockDriverState
*bs
,
3962 int64_t sector_num
, int nb_sectors
,
3965 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, false);
3968 static int coroutine_fn
bdrv_co_writev_em(BlockDriverState
*bs
,
3969 int64_t sector_num
, int nb_sectors
,
3972 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, true);
3975 static void coroutine_fn
bdrv_flush_co_entry(void *opaque
)
3977 RwCo
*rwco
= opaque
;
3979 rwco
->ret
= bdrv_co_flush(rwco
->bs
);
3982 int coroutine_fn
bdrv_co_flush(BlockDriverState
*bs
)
3986 if (!bs
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
)) {
3990 /* Write back cached data to the OS even with cache=unsafe */
3991 if (bs
->drv
->bdrv_co_flush_to_os
) {
3992 ret
= bs
->drv
->bdrv_co_flush_to_os(bs
);
3998 /* But don't actually force it to the disk with cache=unsafe */
3999 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
4003 if (bs
->drv
->bdrv_co_flush_to_disk
) {
4004 ret
= bs
->drv
->bdrv_co_flush_to_disk(bs
);
4005 } else if (bs
->drv
->bdrv_aio_flush
) {
4006 BlockDriverAIOCB
*acb
;
4007 CoroutineIOCompletion co
= {
4008 .coroutine
= qemu_coroutine_self(),
4011 acb
= bs
->drv
->bdrv_aio_flush(bs
, bdrv_co_io_em_complete
, &co
);
4015 qemu_coroutine_yield();
4020 * Some block drivers always operate in either writethrough or unsafe
4021 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
4022 * know how the server works (because the behaviour is hardcoded or
4023 * depends on server-side configuration), so we can't ensure that
4024 * everything is safe on disk. Returning an error doesn't work because
4025 * that would break guests even if the server operates in writethrough
4028 * Let's hope the user knows what he's doing.
4036 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
4037 * in the case of cache=unsafe, so there are no useless flushes.
4040 return bdrv_co_flush(bs
->file
);
4043 void bdrv_invalidate_cache(BlockDriverState
*bs
)
4045 if (bs
->drv
&& bs
->drv
->bdrv_invalidate_cache
) {
4046 bs
->drv
->bdrv_invalidate_cache(bs
);
4050 void bdrv_invalidate_cache_all(void)
4052 BlockDriverState
*bs
;
4054 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
4055 bdrv_invalidate_cache(bs
);
4059 void bdrv_clear_incoming_migration_all(void)
4061 BlockDriverState
*bs
;
4063 QTAILQ_FOREACH(bs
, &bdrv_states
, list
) {
4064 bs
->open_flags
= bs
->open_flags
& ~(BDRV_O_INCOMING
);
4068 int bdrv_flush(BlockDriverState
*bs
)
4076 if (qemu_in_coroutine()) {
4077 /* Fast-path if already in coroutine context */
4078 bdrv_flush_co_entry(&rwco
);
4080 co
= qemu_coroutine_create(bdrv_flush_co_entry
);
4081 qemu_coroutine_enter(co
, &rwco
);
4082 while (rwco
.ret
== NOT_DONE
) {
4090 static void coroutine_fn
bdrv_discard_co_entry(void *opaque
)
4092 RwCo
*rwco
= opaque
;
4094 rwco
->ret
= bdrv_co_discard(rwco
->bs
, rwco
->sector_num
, rwco
->nb_sectors
);
4097 int coroutine_fn
bdrv_co_discard(BlockDriverState
*bs
, int64_t sector_num
,
4102 } else if (bdrv_check_request(bs
, sector_num
, nb_sectors
)) {
4104 } else if (bs
->read_only
) {
4106 } else if (bs
->drv
->bdrv_co_discard
) {
4107 return bs
->drv
->bdrv_co_discard(bs
, sector_num
, nb_sectors
);
4108 } else if (bs
->drv
->bdrv_aio_discard
) {
4109 BlockDriverAIOCB
*acb
;
4110 CoroutineIOCompletion co
= {
4111 .coroutine
= qemu_coroutine_self(),
4114 acb
= bs
->drv
->bdrv_aio_discard(bs
, sector_num
, nb_sectors
,
4115 bdrv_co_io_em_complete
, &co
);
4119 qemu_coroutine_yield();
4127 int bdrv_discard(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
)
4132 .sector_num
= sector_num
,
4133 .nb_sectors
= nb_sectors
,
4137 if (qemu_in_coroutine()) {
4138 /* Fast-path if already in coroutine context */
4139 bdrv_discard_co_entry(&rwco
);
4141 co
= qemu_coroutine_create(bdrv_discard_co_entry
);
4142 qemu_coroutine_enter(co
, &rwco
);
4143 while (rwco
.ret
== NOT_DONE
) {
4151 /**************************************************************/
4152 /* removable device support */
4155 * Return TRUE if the media is present
4157 int bdrv_is_inserted(BlockDriverState
*bs
)
4159 BlockDriver
*drv
= bs
->drv
;
4163 if (!drv
->bdrv_is_inserted
)
4165 return drv
->bdrv_is_inserted(bs
);
4169 * Return whether the media changed since the last call to this
4170 * function, or -ENOTSUP if we don't know. Most drivers don't know.
4172 int bdrv_media_changed(BlockDriverState
*bs
)
4174 BlockDriver
*drv
= bs
->drv
;
4176 if (drv
&& drv
->bdrv_media_changed
) {
4177 return drv
->bdrv_media_changed(bs
);
4183 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4185 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
4187 BlockDriver
*drv
= bs
->drv
;
4189 if (drv
&& drv
->bdrv_eject
) {
4190 drv
->bdrv_eject(bs
, eject_flag
);
4193 if (bs
->device_name
[0] != '\0') {
4194 bdrv_emit_qmp_eject_event(bs
, eject_flag
);
4199 * Lock or unlock the media (if it is locked, the user won't be able
4200 * to eject it manually).
4202 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
4204 BlockDriver
*drv
= bs
->drv
;
4206 trace_bdrv_lock_medium(bs
, locked
);
4208 if (drv
&& drv
->bdrv_lock_medium
) {
4209 drv
->bdrv_lock_medium(bs
, locked
);
4213 /* needed for generic scsi interface */
4215 int bdrv_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
4217 BlockDriver
*drv
= bs
->drv
;
4219 if (drv
&& drv
->bdrv_ioctl
)
4220 return drv
->bdrv_ioctl(bs
, req
, buf
);
4224 BlockDriverAIOCB
*bdrv_aio_ioctl(BlockDriverState
*bs
,
4225 unsigned long int req
, void *buf
,
4226 BlockDriverCompletionFunc
*cb
, void *opaque
)
4228 BlockDriver
*drv
= bs
->drv
;
4230 if (drv
&& drv
->bdrv_aio_ioctl
)
4231 return drv
->bdrv_aio_ioctl(bs
, req
, buf
, cb
, opaque
);
4235 void bdrv_set_buffer_alignment(BlockDriverState
*bs
, int align
)
4237 bs
->buffer_alignment
= align
;
4240 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
4242 return qemu_memalign((bs
&& bs
->buffer_alignment
) ? bs
->buffer_alignment
: 512, size
);
4245 void bdrv_set_dirty_tracking(BlockDriverState
*bs
, int enable
)
4247 int64_t bitmap_size
;
4249 bs
->dirty_count
= 0;
4251 if (!bs
->dirty_bitmap
) {
4252 bitmap_size
= (bdrv_getlength(bs
) >> BDRV_SECTOR_BITS
) +
4253 BDRV_SECTORS_PER_DIRTY_CHUNK
* BITS_PER_LONG
- 1;
4254 bitmap_size
/= BDRV_SECTORS_PER_DIRTY_CHUNK
* BITS_PER_LONG
;
4256 bs
->dirty_bitmap
= g_new0(unsigned long, bitmap_size
);
4259 if (bs
->dirty_bitmap
) {
4260 g_free(bs
->dirty_bitmap
);
4261 bs
->dirty_bitmap
= NULL
;
4266 int bdrv_get_dirty(BlockDriverState
*bs
, int64_t sector
)
4268 int64_t chunk
= sector
/ (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK
;
4270 if (bs
->dirty_bitmap
&&
4271 (sector
<< BDRV_SECTOR_BITS
) < bdrv_getlength(bs
)) {
4272 return !!(bs
->dirty_bitmap
[chunk
/ (sizeof(unsigned long) * 8)] &
4273 (1UL << (chunk
% (sizeof(unsigned long) * 8))));
4279 void bdrv_reset_dirty(BlockDriverState
*bs
, int64_t cur_sector
,
4282 set_dirty_bitmap(bs
, cur_sector
, nr_sectors
, 0);
4285 int64_t bdrv_get_dirty_count(BlockDriverState
*bs
)
4287 return bs
->dirty_count
;
4290 void bdrv_set_in_use(BlockDriverState
*bs
, int in_use
)
4292 assert(bs
->in_use
!= in_use
);
4293 bs
->in_use
= in_use
;
4296 int bdrv_in_use(BlockDriverState
*bs
)
4301 void bdrv_iostatus_enable(BlockDriverState
*bs
)
4303 bs
->iostatus_enabled
= true;
4304 bs
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
4307 /* The I/O status is only enabled if the drive explicitly
4308 * enables it _and_ the VM is configured to stop on errors */
4309 bool bdrv_iostatus_is_enabled(const BlockDriverState
*bs
)
4311 return (bs
->iostatus_enabled
&&
4312 (bs
->on_write_error
== BLOCKDEV_ON_ERROR_ENOSPC
||
4313 bs
->on_write_error
== BLOCKDEV_ON_ERROR_STOP
||
4314 bs
->on_read_error
== BLOCKDEV_ON_ERROR_STOP
));
4317 void bdrv_iostatus_disable(BlockDriverState
*bs
)
4319 bs
->iostatus_enabled
= false;
4322 void bdrv_iostatus_reset(BlockDriverState
*bs
)
4324 if (bdrv_iostatus_is_enabled(bs
)) {
4325 bs
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
4329 void bdrv_iostatus_set_err(BlockDriverState
*bs
, int error
)
4331 assert(bdrv_iostatus_is_enabled(bs
));
4332 if (bs
->iostatus
== BLOCK_DEVICE_IO_STATUS_OK
) {
4333 bs
->iostatus
= error
== ENOSPC
? BLOCK_DEVICE_IO_STATUS_NOSPACE
:
4334 BLOCK_DEVICE_IO_STATUS_FAILED
;
4339 bdrv_acct_start(BlockDriverState
*bs
, BlockAcctCookie
*cookie
, int64_t bytes
,
4340 enum BlockAcctType type
)
4342 assert(type
< BDRV_MAX_IOTYPE
);
4344 cookie
->bytes
= bytes
;
4345 cookie
->start_time_ns
= get_clock();
4346 cookie
->type
= type
;
4350 bdrv_acct_done(BlockDriverState
*bs
, BlockAcctCookie
*cookie
)
4352 assert(cookie
->type
< BDRV_MAX_IOTYPE
);
4354 bs
->nr_bytes
[cookie
->type
] += cookie
->bytes
;
4355 bs
->nr_ops
[cookie
->type
]++;
4356 bs
->total_time_ns
[cookie
->type
] += get_clock() - cookie
->start_time_ns
;
4359 int bdrv_img_create(const char *filename
, const char *fmt
,
4360 const char *base_filename
, const char *base_fmt
,
4361 char *options
, uint64_t img_size
, int flags
)
4363 QEMUOptionParameter
*param
= NULL
, *create_options
= NULL
;
4364 QEMUOptionParameter
*backing_fmt
, *backing_file
, *size
;
4365 BlockDriverState
*bs
= NULL
;
4366 BlockDriver
*drv
, *proto_drv
;
4367 BlockDriver
*backing_drv
= NULL
;
4370 /* Find driver and parse its options */
4371 drv
= bdrv_find_format(fmt
);
4373 error_report("Unknown file format '%s'", fmt
);
4378 proto_drv
= bdrv_find_protocol(filename
);
4380 error_report("Unknown protocol '%s'", filename
);
4385 create_options
= append_option_parameters(create_options
,
4386 drv
->create_options
);
4387 create_options
= append_option_parameters(create_options
,
4388 proto_drv
->create_options
);
4390 /* Create parameter list with default values */
4391 param
= parse_option_parameters("", create_options
, param
);
4393 set_option_parameter_int(param
, BLOCK_OPT_SIZE
, img_size
);
4395 /* Parse -o options */
4397 param
= parse_option_parameters(options
, create_options
, param
);
4398 if (param
== NULL
) {
4399 error_report("Invalid options for file format '%s'.", fmt
);
4405 if (base_filename
) {
4406 if (set_option_parameter(param
, BLOCK_OPT_BACKING_FILE
,
4408 error_report("Backing file not supported for file format '%s'",
4416 if (set_option_parameter(param
, BLOCK_OPT_BACKING_FMT
, base_fmt
)) {
4417 error_report("Backing file format not supported for file "
4418 "format '%s'", fmt
);
4424 backing_file
= get_option_parameter(param
, BLOCK_OPT_BACKING_FILE
);
4425 if (backing_file
&& backing_file
->value
.s
) {
4426 if (!strcmp(filename
, backing_file
->value
.s
)) {
4427 error_report("Error: Trying to create an image with the "
4428 "same filename as the backing file");
4434 backing_fmt
= get_option_parameter(param
, BLOCK_OPT_BACKING_FMT
);
4435 if (backing_fmt
&& backing_fmt
->value
.s
) {
4436 backing_drv
= bdrv_find_format(backing_fmt
->value
.s
);
4438 error_report("Unknown backing file format '%s'",
4439 backing_fmt
->value
.s
);
4445 // The size for the image must always be specified, with one exception:
4446 // If we are using a backing file, we can obtain the size from there
4447 size
= get_option_parameter(param
, BLOCK_OPT_SIZE
);
4448 if (size
&& size
->value
.n
== -1) {
4449 if (backing_file
&& backing_file
->value
.s
) {
4454 /* backing files always opened read-only */
4456 flags
& ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
4460 ret
= bdrv_open(bs
, backing_file
->value
.s
, back_flags
, backing_drv
);
4462 error_report("Could not open '%s'", backing_file
->value
.s
);
4465 bdrv_get_geometry(bs
, &size
);
4468 snprintf(buf
, sizeof(buf
), "%" PRId64
, size
);
4469 set_option_parameter(param
, BLOCK_OPT_SIZE
, buf
);
4471 error_report("Image creation needs a size parameter");
4477 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
4478 print_option_parameters(param
);
4481 ret
= bdrv_create(drv
, filename
, param
);
4484 if (ret
== -ENOTSUP
) {
4485 error_report("Formatting or formatting option not supported for "
4486 "file format '%s'", fmt
);
4487 } else if (ret
== -EFBIG
) {
4488 error_report("The image size is too large for file format '%s'",
4491 error_report("%s: error while creating %s: %s", filename
, fmt
,
4497 free_option_parameters(create_options
);
4498 free_option_parameters(param
);