2 * QEMU Enhanced Disk Format
4 * Copyright IBM, Corp. 2010
7 * Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
8 * Anthony Liguori <aliguori@us.ibm.com>
10 * This work is licensed under the terms of the GNU LGPL, version 2 or later.
11 * See the COPYING.LIB file in the top-level directory.
15 #include "qemu/osdep.h"
16 #include "qemu/timer.h"
19 #include "qapi/qmp/qerror.h"
20 #include "migration/migration.h"
22 static const AIOCBInfo qed_aiocb_info
= {
23 .aiocb_size
= sizeof(QEDAIOCB
),
26 static int bdrv_qed_probe(const uint8_t *buf
, int buf_size
,
29 const QEDHeader
*header
= (const QEDHeader
*)buf
;
31 if (buf_size
< sizeof(*header
)) {
34 if (le32_to_cpu(header
->magic
) != QED_MAGIC
) {
41 * Check whether an image format is raw
43 * @fmt: Backing file format, may be NULL
45 static bool qed_fmt_is_raw(const char *fmt
)
47 return fmt
&& strcmp(fmt
, "raw") == 0;
50 static void qed_header_le_to_cpu(const QEDHeader
*le
, QEDHeader
*cpu
)
52 cpu
->magic
= le32_to_cpu(le
->magic
);
53 cpu
->cluster_size
= le32_to_cpu(le
->cluster_size
);
54 cpu
->table_size
= le32_to_cpu(le
->table_size
);
55 cpu
->header_size
= le32_to_cpu(le
->header_size
);
56 cpu
->features
= le64_to_cpu(le
->features
);
57 cpu
->compat_features
= le64_to_cpu(le
->compat_features
);
58 cpu
->autoclear_features
= le64_to_cpu(le
->autoclear_features
);
59 cpu
->l1_table_offset
= le64_to_cpu(le
->l1_table_offset
);
60 cpu
->image_size
= le64_to_cpu(le
->image_size
);
61 cpu
->backing_filename_offset
= le32_to_cpu(le
->backing_filename_offset
);
62 cpu
->backing_filename_size
= le32_to_cpu(le
->backing_filename_size
);
65 static void qed_header_cpu_to_le(const QEDHeader
*cpu
, QEDHeader
*le
)
67 le
->magic
= cpu_to_le32(cpu
->magic
);
68 le
->cluster_size
= cpu_to_le32(cpu
->cluster_size
);
69 le
->table_size
= cpu_to_le32(cpu
->table_size
);
70 le
->header_size
= cpu_to_le32(cpu
->header_size
);
71 le
->features
= cpu_to_le64(cpu
->features
);
72 le
->compat_features
= cpu_to_le64(cpu
->compat_features
);
73 le
->autoclear_features
= cpu_to_le64(cpu
->autoclear_features
);
74 le
->l1_table_offset
= cpu_to_le64(cpu
->l1_table_offset
);
75 le
->image_size
= cpu_to_le64(cpu
->image_size
);
76 le
->backing_filename_offset
= cpu_to_le32(cpu
->backing_filename_offset
);
77 le
->backing_filename_size
= cpu_to_le32(cpu
->backing_filename_size
);
80 int qed_write_header_sync(BDRVQEDState
*s
)
85 qed_header_cpu_to_le(&s
->header
, &le
);
86 ret
= bdrv_pwrite(s
->bs
->file
->bs
, 0, &le
, sizeof(le
));
87 if (ret
!= sizeof(le
)) {
102 static void qed_write_header_cb(void *opaque
, int ret
)
104 QEDWriteHeaderCB
*write_header_cb
= opaque
;
106 qemu_vfree(write_header_cb
->buf
);
107 gencb_complete(write_header_cb
, ret
);
110 static void qed_write_header_read_cb(void *opaque
, int ret
)
112 QEDWriteHeaderCB
*write_header_cb
= opaque
;
113 BDRVQEDState
*s
= write_header_cb
->s
;
116 qed_write_header_cb(write_header_cb
, ret
);
121 qed_header_cpu_to_le(&s
->header
, (QEDHeader
*)write_header_cb
->buf
);
123 bdrv_aio_writev(s
->bs
->file
->bs
, 0, &write_header_cb
->qiov
,
124 write_header_cb
->nsectors
, qed_write_header_cb
,
129 * Update header in-place (does not rewrite backing filename or other strings)
131 * This function only updates known header fields in-place and does not affect
132 * extra data after the QED header.
134 static void qed_write_header(BDRVQEDState
*s
, BlockCompletionFunc cb
,
137 /* We must write full sectors for O_DIRECT but cannot necessarily generate
138 * the data following the header if an unrecognized compat feature is
139 * active. Therefore, first read the sectors containing the header, update
140 * them, and write back.
143 int nsectors
= (sizeof(QEDHeader
) + BDRV_SECTOR_SIZE
- 1) /
145 size_t len
= nsectors
* BDRV_SECTOR_SIZE
;
146 QEDWriteHeaderCB
*write_header_cb
= gencb_alloc(sizeof(*write_header_cb
),
149 write_header_cb
->s
= s
;
150 write_header_cb
->nsectors
= nsectors
;
151 write_header_cb
->buf
= qemu_blockalign(s
->bs
, len
);
152 write_header_cb
->iov
.iov_base
= write_header_cb
->buf
;
153 write_header_cb
->iov
.iov_len
= len
;
154 qemu_iovec_init_external(&write_header_cb
->qiov
, &write_header_cb
->iov
, 1);
156 bdrv_aio_readv(s
->bs
->file
->bs
, 0, &write_header_cb
->qiov
, nsectors
,
157 qed_write_header_read_cb
, write_header_cb
);
160 static uint64_t qed_max_image_size(uint32_t cluster_size
, uint32_t table_size
)
162 uint64_t table_entries
;
165 table_entries
= (table_size
* cluster_size
) / sizeof(uint64_t);
166 l2_size
= table_entries
* cluster_size
;
168 return l2_size
* table_entries
;
171 static bool qed_is_cluster_size_valid(uint32_t cluster_size
)
173 if (cluster_size
< QED_MIN_CLUSTER_SIZE
||
174 cluster_size
> QED_MAX_CLUSTER_SIZE
) {
177 if (cluster_size
& (cluster_size
- 1)) {
178 return false; /* not power of 2 */
183 static bool qed_is_table_size_valid(uint32_t table_size
)
185 if (table_size
< QED_MIN_TABLE_SIZE
||
186 table_size
> QED_MAX_TABLE_SIZE
) {
189 if (table_size
& (table_size
- 1)) {
190 return false; /* not power of 2 */
195 static bool qed_is_image_size_valid(uint64_t image_size
, uint32_t cluster_size
,
198 if (image_size
% BDRV_SECTOR_SIZE
!= 0) {
199 return false; /* not multiple of sector size */
201 if (image_size
> qed_max_image_size(cluster_size
, table_size
)) {
202 return false; /* image is too large */
208 * Read a string of known length from the image file
211 * @offset: File offset to start of string, in bytes
212 * @n: String length in bytes
213 * @buf: Destination buffer
214 * @buflen: Destination buffer length in bytes
215 * @ret: 0 on success, -errno on failure
217 * The string is NUL-terminated.
219 static int qed_read_string(BlockDriverState
*file
, uint64_t offset
, size_t n
,
220 char *buf
, size_t buflen
)
226 ret
= bdrv_pread(file
, offset
, buf
, n
);
235 * Allocate new clusters
238 * @n: Number of contiguous clusters to allocate
239 * @ret: Offset of first allocated cluster
241 * This function only produces the offset where the new clusters should be
242 * written. It updates BDRVQEDState but does not make any changes to the image
245 static uint64_t qed_alloc_clusters(BDRVQEDState
*s
, unsigned int n
)
247 uint64_t offset
= s
->file_size
;
248 s
->file_size
+= n
* s
->header
.cluster_size
;
252 QEDTable
*qed_alloc_table(BDRVQEDState
*s
)
254 /* Honor O_DIRECT memory alignment requirements */
255 return qemu_blockalign(s
->bs
,
256 s
->header
.cluster_size
* s
->header
.table_size
);
260 * Allocate a new zeroed L2 table
262 static CachedL2Table
*qed_new_l2_table(BDRVQEDState
*s
)
264 CachedL2Table
*l2_table
= qed_alloc_l2_cache_entry(&s
->l2_cache
);
266 l2_table
->table
= qed_alloc_table(s
);
267 l2_table
->offset
= qed_alloc_clusters(s
, s
->header
.table_size
);
269 memset(l2_table
->table
->offsets
, 0,
270 s
->header
.cluster_size
* s
->header
.table_size
);
274 static void qed_aio_next_io(void *opaque
, int ret
);
276 static void qed_plug_allocating_write_reqs(BDRVQEDState
*s
)
278 assert(!s
->allocating_write_reqs_plugged
);
280 s
->allocating_write_reqs_plugged
= true;
283 static void qed_unplug_allocating_write_reqs(BDRVQEDState
*s
)
287 assert(s
->allocating_write_reqs_plugged
);
289 s
->allocating_write_reqs_plugged
= false;
291 acb
= QSIMPLEQ_FIRST(&s
->allocating_write_reqs
);
293 qed_aio_next_io(acb
, 0);
297 static void qed_finish_clear_need_check(void *opaque
, int ret
)
302 static void qed_flush_after_clear_need_check(void *opaque
, int ret
)
304 BDRVQEDState
*s
= opaque
;
306 bdrv_aio_flush(s
->bs
, qed_finish_clear_need_check
, s
);
308 /* No need to wait until flush completes */
309 qed_unplug_allocating_write_reqs(s
);
312 static void qed_clear_need_check(void *opaque
, int ret
)
314 BDRVQEDState
*s
= opaque
;
317 qed_unplug_allocating_write_reqs(s
);
321 s
->header
.features
&= ~QED_F_NEED_CHECK
;
322 qed_write_header(s
, qed_flush_after_clear_need_check
, s
);
325 static void qed_need_check_timer_cb(void *opaque
)
327 BDRVQEDState
*s
= opaque
;
329 /* The timer should only fire when allocating writes have drained */
330 assert(!QSIMPLEQ_FIRST(&s
->allocating_write_reqs
));
332 trace_qed_need_check_timer_cb(s
);
334 qed_plug_allocating_write_reqs(s
);
336 /* Ensure writes are on disk before clearing flag */
337 bdrv_aio_flush(s
->bs
, qed_clear_need_check
, s
);
340 static void qed_start_need_check_timer(BDRVQEDState
*s
)
342 trace_qed_start_need_check_timer(s
);
344 /* Use QEMU_CLOCK_VIRTUAL so we don't alter the image file while suspended for
347 timer_mod(s
->need_check_timer
, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL
) +
348 get_ticks_per_sec() * QED_NEED_CHECK_TIMEOUT
);
351 /* It's okay to call this multiple times or when no timer is started */
352 static void qed_cancel_need_check_timer(BDRVQEDState
*s
)
354 trace_qed_cancel_need_check_timer(s
);
355 timer_del(s
->need_check_timer
);
358 static void bdrv_qed_detach_aio_context(BlockDriverState
*bs
)
360 BDRVQEDState
*s
= bs
->opaque
;
362 qed_cancel_need_check_timer(s
);
363 timer_free(s
->need_check_timer
);
366 static void bdrv_qed_attach_aio_context(BlockDriverState
*bs
,
367 AioContext
*new_context
)
369 BDRVQEDState
*s
= bs
->opaque
;
371 s
->need_check_timer
= aio_timer_new(new_context
,
372 QEMU_CLOCK_VIRTUAL
, SCALE_NS
,
373 qed_need_check_timer_cb
, s
);
374 if (s
->header
.features
& QED_F_NEED_CHECK
) {
375 qed_start_need_check_timer(s
);
379 static void bdrv_qed_drain(BlockDriverState
*bs
)
381 BDRVQEDState
*s
= bs
->opaque
;
383 /* Cancel timer and start doing I/O that were meant to happen as if it
384 * fired, that way we get bdrv_drain() taking care of the ongoing requests
386 qed_cancel_need_check_timer(s
);
387 qed_plug_allocating_write_reqs(s
);
388 bdrv_aio_flush(s
->bs
, qed_clear_need_check
, s
);
391 static int bdrv_qed_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
394 BDRVQEDState
*s
= bs
->opaque
;
400 QSIMPLEQ_INIT(&s
->allocating_write_reqs
);
402 ret
= bdrv_pread(bs
->file
->bs
, 0, &le_header
, sizeof(le_header
));
406 qed_header_le_to_cpu(&le_header
, &s
->header
);
408 if (s
->header
.magic
!= QED_MAGIC
) {
409 error_setg(errp
, "Image not in QED format");
412 if (s
->header
.features
& ~QED_FEATURE_MASK
) {
413 /* image uses unsupported feature bits */
415 snprintf(buf
, sizeof(buf
), "%" PRIx64
,
416 s
->header
.features
& ~QED_FEATURE_MASK
);
417 error_setg(errp
, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE
,
418 bdrv_get_device_or_node_name(bs
), "QED", buf
);
421 if (!qed_is_cluster_size_valid(s
->header
.cluster_size
)) {
425 /* Round down file size to the last cluster */
426 file_size
= bdrv_getlength(bs
->file
->bs
);
430 s
->file_size
= qed_start_of_cluster(s
, file_size
);
432 if (!qed_is_table_size_valid(s
->header
.table_size
)) {
435 if (!qed_is_image_size_valid(s
->header
.image_size
,
436 s
->header
.cluster_size
,
437 s
->header
.table_size
)) {
440 if (!qed_check_table_offset(s
, s
->header
.l1_table_offset
)) {
444 s
->table_nelems
= (s
->header
.cluster_size
* s
->header
.table_size
) /
446 s
->l2_shift
= ctz32(s
->header
.cluster_size
);
447 s
->l2_mask
= s
->table_nelems
- 1;
448 s
->l1_shift
= s
->l2_shift
+ ctz32(s
->table_nelems
);
450 /* Header size calculation must not overflow uint32_t */
451 if (s
->header
.header_size
> UINT32_MAX
/ s
->header
.cluster_size
) {
455 if ((s
->header
.features
& QED_F_BACKING_FILE
)) {
456 if ((uint64_t)s
->header
.backing_filename_offset
+
457 s
->header
.backing_filename_size
>
458 s
->header
.cluster_size
* s
->header
.header_size
) {
462 ret
= qed_read_string(bs
->file
->bs
, s
->header
.backing_filename_offset
,
463 s
->header
.backing_filename_size
, bs
->backing_file
,
464 sizeof(bs
->backing_file
));
469 if (s
->header
.features
& QED_F_BACKING_FORMAT_NO_PROBE
) {
470 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), "raw");
474 /* Reset unknown autoclear feature bits. This is a backwards
475 * compatibility mechanism that allows images to be opened by older
476 * programs, which "knock out" unknown feature bits. When an image is
477 * opened by a newer program again it can detect that the autoclear
478 * feature is no longer valid.
480 if ((s
->header
.autoclear_features
& ~QED_AUTOCLEAR_FEATURE_MASK
) != 0 &&
481 !bdrv_is_read_only(bs
->file
->bs
) && !(flags
& BDRV_O_INACTIVE
)) {
482 s
->header
.autoclear_features
&= QED_AUTOCLEAR_FEATURE_MASK
;
484 ret
= qed_write_header_sync(s
);
489 /* From here on only known autoclear feature bits are valid */
490 bdrv_flush(bs
->file
->bs
);
493 s
->l1_table
= qed_alloc_table(s
);
494 qed_init_l2_cache(&s
->l2_cache
);
496 ret
= qed_read_l1_table_sync(s
);
501 /* If image was not closed cleanly, check consistency */
502 if (!(flags
& BDRV_O_CHECK
) && (s
->header
.features
& QED_F_NEED_CHECK
)) {
503 /* Read-only images cannot be fixed. There is no risk of corruption
504 * since write operations are not possible. Therefore, allow
505 * potentially inconsistent images to be opened read-only. This can
506 * aid data recovery from an otherwise inconsistent image.
508 if (!bdrv_is_read_only(bs
->file
->bs
) &&
509 !(flags
& BDRV_O_INACTIVE
)) {
510 BdrvCheckResult result
= {0};
512 ret
= qed_check(s
, &result
, true);
519 bdrv_qed_attach_aio_context(bs
, bdrv_get_aio_context(bs
));
523 qed_free_l2_cache(&s
->l2_cache
);
524 qemu_vfree(s
->l1_table
);
529 static void bdrv_qed_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
531 BDRVQEDState
*s
= bs
->opaque
;
533 bs
->bl
.write_zeroes_alignment
= s
->header
.cluster_size
>> BDRV_SECTOR_BITS
;
536 /* We have nothing to do for QED reopen, stubs just return
538 static int bdrv_qed_reopen_prepare(BDRVReopenState
*state
,
539 BlockReopenQueue
*queue
, Error
**errp
)
544 static void bdrv_qed_close(BlockDriverState
*bs
)
546 BDRVQEDState
*s
= bs
->opaque
;
548 bdrv_qed_detach_aio_context(bs
);
550 /* Ensure writes reach stable storage */
551 bdrv_flush(bs
->file
->bs
);
553 /* Clean shutdown, no check required on next open */
554 if (s
->header
.features
& QED_F_NEED_CHECK
) {
555 s
->header
.features
&= ~QED_F_NEED_CHECK
;
556 qed_write_header_sync(s
);
559 qed_free_l2_cache(&s
->l2_cache
);
560 qemu_vfree(s
->l1_table
);
563 static int qed_create(const char *filename
, uint32_t cluster_size
,
564 uint64_t image_size
, uint32_t table_size
,
565 const char *backing_file
, const char *backing_fmt
,
566 QemuOpts
*opts
, Error
**errp
)
570 .cluster_size
= cluster_size
,
571 .table_size
= table_size
,
574 .compat_features
= 0,
575 .l1_table_offset
= cluster_size
,
576 .image_size
= image_size
,
579 uint8_t *l1_table
= NULL
;
580 size_t l1_size
= header
.cluster_size
* header
.table_size
;
581 Error
*local_err
= NULL
;
583 BlockDriverState
*bs
;
585 ret
= bdrv_create_file(filename
, opts
, &local_err
);
587 error_propagate(errp
, local_err
);
592 ret
= bdrv_open(&bs
, filename
, NULL
, NULL
,
593 BDRV_O_RDWR
| BDRV_O_CACHE_WB
| BDRV_O_PROTOCOL
,
596 error_propagate(errp
, local_err
);
600 /* File must start empty and grow, check truncate is supported */
601 ret
= bdrv_truncate(bs
, 0);
607 header
.features
|= QED_F_BACKING_FILE
;
608 header
.backing_filename_offset
= sizeof(le_header
);
609 header
.backing_filename_size
= strlen(backing_file
);
611 if (qed_fmt_is_raw(backing_fmt
)) {
612 header
.features
|= QED_F_BACKING_FORMAT_NO_PROBE
;
616 qed_header_cpu_to_le(&header
, &le_header
);
617 ret
= bdrv_pwrite(bs
, 0, &le_header
, sizeof(le_header
));
621 ret
= bdrv_pwrite(bs
, sizeof(le_header
), backing_file
,
622 header
.backing_filename_size
);
627 l1_table
= g_malloc0(l1_size
);
628 ret
= bdrv_pwrite(bs
, header
.l1_table_offset
, l1_table
, l1_size
);
633 ret
= 0; /* success */
640 static int bdrv_qed_create(const char *filename
, QemuOpts
*opts
, Error
**errp
)
642 uint64_t image_size
= 0;
643 uint32_t cluster_size
= QED_DEFAULT_CLUSTER_SIZE
;
644 uint32_t table_size
= QED_DEFAULT_TABLE_SIZE
;
645 char *backing_file
= NULL
;
646 char *backing_fmt
= NULL
;
649 image_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
651 backing_file
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
652 backing_fmt
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FMT
);
653 cluster_size
= qemu_opt_get_size_del(opts
,
654 BLOCK_OPT_CLUSTER_SIZE
,
655 QED_DEFAULT_CLUSTER_SIZE
);
656 table_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_TABLE_SIZE
,
657 QED_DEFAULT_TABLE_SIZE
);
659 if (!qed_is_cluster_size_valid(cluster_size
)) {
660 error_setg(errp
, "QED cluster size must be within range [%u, %u] "
662 QED_MIN_CLUSTER_SIZE
, QED_MAX_CLUSTER_SIZE
);
666 if (!qed_is_table_size_valid(table_size
)) {
667 error_setg(errp
, "QED table size must be within range [%u, %u] "
669 QED_MIN_TABLE_SIZE
, QED_MAX_TABLE_SIZE
);
673 if (!qed_is_image_size_valid(image_size
, cluster_size
, table_size
)) {
674 error_setg(errp
, "QED image size must be a non-zero multiple of "
675 "cluster size and less than %" PRIu64
" bytes",
676 qed_max_image_size(cluster_size
, table_size
));
681 ret
= qed_create(filename
, cluster_size
, image_size
, table_size
,
682 backing_file
, backing_fmt
, opts
, errp
);
685 g_free(backing_file
);
691 BlockDriverState
*bs
;
698 static void qed_is_allocated_cb(void *opaque
, int ret
, uint64_t offset
, size_t len
)
700 QEDIsAllocatedCB
*cb
= opaque
;
701 BDRVQEDState
*s
= cb
->bs
->opaque
;
702 *cb
->pnum
= len
/ BDRV_SECTOR_SIZE
;
704 case QED_CLUSTER_FOUND
:
705 offset
|= qed_offset_into_cluster(s
, cb
->pos
);
706 cb
->status
= BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
| offset
;
708 case QED_CLUSTER_ZERO
:
709 cb
->status
= BDRV_BLOCK_ZERO
;
722 qemu_coroutine_enter(cb
->co
, NULL
);
726 static int64_t coroutine_fn
bdrv_qed_co_get_block_status(BlockDriverState
*bs
,
728 int nb_sectors
, int *pnum
)
730 BDRVQEDState
*s
= bs
->opaque
;
731 size_t len
= (size_t)nb_sectors
* BDRV_SECTOR_SIZE
;
732 QEDIsAllocatedCB cb
= {
734 .pos
= (uint64_t)sector_num
* BDRV_SECTOR_SIZE
,
735 .status
= BDRV_BLOCK_OFFSET_MASK
,
738 QEDRequest request
= { .l2_table
= NULL
};
740 qed_find_cluster(s
, &request
, cb
.pos
, len
, qed_is_allocated_cb
, &cb
);
742 /* Now sleep if the callback wasn't invoked immediately */
743 while (cb
.status
== BDRV_BLOCK_OFFSET_MASK
) {
744 cb
.co
= qemu_coroutine_self();
745 qemu_coroutine_yield();
748 qed_unref_l2_cache_entry(request
.l2_table
);
753 static BDRVQEDState
*acb_to_s(QEDAIOCB
*acb
)
755 return acb
->common
.bs
->opaque
;
759 * Read from the backing file or zero-fill if no backing file
762 * @pos: Byte position in device
763 * @qiov: Destination I/O vector
764 * @backing_qiov: Possibly shortened copy of qiov, to be allocated here
765 * @cb: Completion function
766 * @opaque: User data for completion function
768 * This function reads qiov->size bytes starting at pos from the backing file.
769 * If there is no backing file then zeroes are read.
771 static void qed_read_backing_file(BDRVQEDState
*s
, uint64_t pos
,
773 QEMUIOVector
**backing_qiov
,
774 BlockCompletionFunc
*cb
, void *opaque
)
776 uint64_t backing_length
= 0;
779 /* If there is a backing file, get its length. Treat the absence of a
780 * backing file like a zero length backing file.
782 if (s
->bs
->backing
) {
783 int64_t l
= bdrv_getlength(s
->bs
->backing
->bs
);
791 /* Zero all sectors if reading beyond the end of the backing file */
792 if (pos
>= backing_length
||
793 pos
+ qiov
->size
> backing_length
) {
794 qemu_iovec_memset(qiov
, 0, 0, qiov
->size
);
797 /* Complete now if there are no backing file sectors to read */
798 if (pos
>= backing_length
) {
803 /* If the read straddles the end of the backing file, shorten it */
804 size
= MIN((uint64_t)backing_length
- pos
, qiov
->size
);
806 assert(*backing_qiov
== NULL
);
807 *backing_qiov
= g_new(QEMUIOVector
, 1);
808 qemu_iovec_init(*backing_qiov
, qiov
->niov
);
809 qemu_iovec_concat(*backing_qiov
, qiov
, 0, size
);
811 BLKDBG_EVENT(s
->bs
->file
, BLKDBG_READ_BACKING_AIO
);
812 bdrv_aio_readv(s
->bs
->backing
->bs
, pos
/ BDRV_SECTOR_SIZE
,
813 *backing_qiov
, size
/ BDRV_SECTOR_SIZE
, cb
, opaque
);
820 QEMUIOVector
*backing_qiov
;
823 } CopyFromBackingFileCB
;
825 static void qed_copy_from_backing_file_cb(void *opaque
, int ret
)
827 CopyFromBackingFileCB
*copy_cb
= opaque
;
828 qemu_vfree(copy_cb
->iov
.iov_base
);
829 gencb_complete(©_cb
->gencb
, ret
);
832 static void qed_copy_from_backing_file_write(void *opaque
, int ret
)
834 CopyFromBackingFileCB
*copy_cb
= opaque
;
835 BDRVQEDState
*s
= copy_cb
->s
;
837 if (copy_cb
->backing_qiov
) {
838 qemu_iovec_destroy(copy_cb
->backing_qiov
);
839 g_free(copy_cb
->backing_qiov
);
840 copy_cb
->backing_qiov
= NULL
;
844 qed_copy_from_backing_file_cb(copy_cb
, ret
);
848 BLKDBG_EVENT(s
->bs
->file
, BLKDBG_COW_WRITE
);
849 bdrv_aio_writev(s
->bs
->file
->bs
, copy_cb
->offset
/ BDRV_SECTOR_SIZE
,
850 ©_cb
->qiov
, copy_cb
->qiov
.size
/ BDRV_SECTOR_SIZE
,
851 qed_copy_from_backing_file_cb
, copy_cb
);
855 * Copy data from backing file into the image
858 * @pos: Byte position in device
859 * @len: Number of bytes
860 * @offset: Byte offset in image file
861 * @cb: Completion function
862 * @opaque: User data for completion function
864 static void qed_copy_from_backing_file(BDRVQEDState
*s
, uint64_t pos
,
865 uint64_t len
, uint64_t offset
,
866 BlockCompletionFunc
*cb
,
869 CopyFromBackingFileCB
*copy_cb
;
871 /* Skip copy entirely if there is no work to do */
877 copy_cb
= gencb_alloc(sizeof(*copy_cb
), cb
, opaque
);
879 copy_cb
->offset
= offset
;
880 copy_cb
->backing_qiov
= NULL
;
881 copy_cb
->iov
.iov_base
= qemu_blockalign(s
->bs
, len
);
882 copy_cb
->iov
.iov_len
= len
;
883 qemu_iovec_init_external(©_cb
->qiov
, ©_cb
->iov
, 1);
885 qed_read_backing_file(s
, pos
, ©_cb
->qiov
, ©_cb
->backing_qiov
,
886 qed_copy_from_backing_file_write
, copy_cb
);
890 * Link one or more contiguous clusters into a table
894 * @index: First cluster index
895 * @n: Number of contiguous clusters
896 * @cluster: First cluster offset
898 * The cluster offset may be an allocated byte offset in the image file, the
899 * zero cluster marker, or the unallocated cluster marker.
901 static void qed_update_l2_table(BDRVQEDState
*s
, QEDTable
*table
, int index
,
902 unsigned int n
, uint64_t cluster
)
905 for (i
= index
; i
< index
+ n
; i
++) {
906 table
->offsets
[i
] = cluster
;
907 if (!qed_offset_is_unalloc_cluster(cluster
) &&
908 !qed_offset_is_zero_cluster(cluster
)) {
909 cluster
+= s
->header
.cluster_size
;
914 static void qed_aio_complete_bh(void *opaque
)
916 QEDAIOCB
*acb
= opaque
;
917 BlockCompletionFunc
*cb
= acb
->common
.cb
;
918 void *user_opaque
= acb
->common
.opaque
;
919 int ret
= acb
->bh_ret
;
921 qemu_bh_delete(acb
->bh
);
924 /* Invoke callback */
925 cb(user_opaque
, ret
);
928 static void qed_aio_complete(QEDAIOCB
*acb
, int ret
)
930 BDRVQEDState
*s
= acb_to_s(acb
);
932 trace_qed_aio_complete(s
, acb
, ret
);
935 qemu_iovec_destroy(&acb
->cur_qiov
);
936 qed_unref_l2_cache_entry(acb
->request
.l2_table
);
938 /* Free the buffer we may have allocated for zero writes */
939 if (acb
->flags
& QED_AIOCB_ZERO
) {
940 qemu_vfree(acb
->qiov
->iov
[0].iov_base
);
941 acb
->qiov
->iov
[0].iov_base
= NULL
;
944 /* Arrange for a bh to invoke the completion function */
946 acb
->bh
= aio_bh_new(bdrv_get_aio_context(acb
->common
.bs
),
947 qed_aio_complete_bh
, acb
);
948 qemu_bh_schedule(acb
->bh
);
950 /* Start next allocating write request waiting behind this one. Note that
951 * requests enqueue themselves when they first hit an unallocated cluster
952 * but they wait until the entire request is finished before waking up the
953 * next request in the queue. This ensures that we don't cycle through
954 * requests multiple times but rather finish one at a time completely.
956 if (acb
== QSIMPLEQ_FIRST(&s
->allocating_write_reqs
)) {
957 QSIMPLEQ_REMOVE_HEAD(&s
->allocating_write_reqs
, next
);
958 acb
= QSIMPLEQ_FIRST(&s
->allocating_write_reqs
);
960 qed_aio_next_io(acb
, 0);
961 } else if (s
->header
.features
& QED_F_NEED_CHECK
) {
962 qed_start_need_check_timer(s
);
968 * Commit the current L2 table to the cache
970 static void qed_commit_l2_update(void *opaque
, int ret
)
972 QEDAIOCB
*acb
= opaque
;
973 BDRVQEDState
*s
= acb_to_s(acb
);
974 CachedL2Table
*l2_table
= acb
->request
.l2_table
;
975 uint64_t l2_offset
= l2_table
->offset
;
977 qed_commit_l2_cache_entry(&s
->l2_cache
, l2_table
);
979 /* This is guaranteed to succeed because we just committed the entry to the
982 acb
->request
.l2_table
= qed_find_l2_cache_entry(&s
->l2_cache
, l2_offset
);
983 assert(acb
->request
.l2_table
!= NULL
);
985 qed_aio_next_io(opaque
, ret
);
989 * Update L1 table with new L2 table offset and write it out
991 static void qed_aio_write_l1_update(void *opaque
, int ret
)
993 QEDAIOCB
*acb
= opaque
;
994 BDRVQEDState
*s
= acb_to_s(acb
);
998 qed_aio_complete(acb
, ret
);
1002 index
= qed_l1_index(s
, acb
->cur_pos
);
1003 s
->l1_table
->offsets
[index
] = acb
->request
.l2_table
->offset
;
1005 qed_write_l1_table(s
, index
, 1, qed_commit_l2_update
, acb
);
1009 * Update L2 table with new cluster offsets and write them out
1011 static void qed_aio_write_l2_update(QEDAIOCB
*acb
, int ret
, uint64_t offset
)
1013 BDRVQEDState
*s
= acb_to_s(acb
);
1014 bool need_alloc
= acb
->find_cluster_ret
== QED_CLUSTER_L1
;
1022 qed_unref_l2_cache_entry(acb
->request
.l2_table
);
1023 acb
->request
.l2_table
= qed_new_l2_table(s
);
1026 index
= qed_l2_index(s
, acb
->cur_pos
);
1027 qed_update_l2_table(s
, acb
->request
.l2_table
->table
, index
, acb
->cur_nclusters
,
1031 /* Write out the whole new L2 table */
1032 qed_write_l2_table(s
, &acb
->request
, 0, s
->table_nelems
, true,
1033 qed_aio_write_l1_update
, acb
);
1035 /* Write out only the updated part of the L2 table */
1036 qed_write_l2_table(s
, &acb
->request
, index
, acb
->cur_nclusters
, false,
1037 qed_aio_next_io
, acb
);
1042 qed_aio_complete(acb
, ret
);
1045 static void qed_aio_write_l2_update_cb(void *opaque
, int ret
)
1047 QEDAIOCB
*acb
= opaque
;
1048 qed_aio_write_l2_update(acb
, ret
, acb
->cur_cluster
);
1052 * Flush new data clusters before updating the L2 table
1054 * This flush is necessary when a backing file is in use. A crash during an
1055 * allocating write could result in empty clusters in the image. If the write
1056 * only touched a subregion of the cluster, then backing image sectors have
1057 * been lost in the untouched region. The solution is to flush after writing a
1058 * new data cluster and before updating the L2 table.
1060 static void qed_aio_write_flush_before_l2_update(void *opaque
, int ret
)
1062 QEDAIOCB
*acb
= opaque
;
1063 BDRVQEDState
*s
= acb_to_s(acb
);
1065 if (!bdrv_aio_flush(s
->bs
->file
->bs
, qed_aio_write_l2_update_cb
, opaque
)) {
1066 qed_aio_complete(acb
, -EIO
);
1071 * Write data to the image file
1073 static void qed_aio_write_main(void *opaque
, int ret
)
1075 QEDAIOCB
*acb
= opaque
;
1076 BDRVQEDState
*s
= acb_to_s(acb
);
1077 uint64_t offset
= acb
->cur_cluster
+
1078 qed_offset_into_cluster(s
, acb
->cur_pos
);
1079 BlockCompletionFunc
*next_fn
;
1081 trace_qed_aio_write_main(s
, acb
, ret
, offset
, acb
->cur_qiov
.size
);
1084 qed_aio_complete(acb
, ret
);
1088 if (acb
->find_cluster_ret
== QED_CLUSTER_FOUND
) {
1089 next_fn
= qed_aio_next_io
;
1091 if (s
->bs
->backing
) {
1092 next_fn
= qed_aio_write_flush_before_l2_update
;
1094 next_fn
= qed_aio_write_l2_update_cb
;
1098 BLKDBG_EVENT(s
->bs
->file
, BLKDBG_WRITE_AIO
);
1099 bdrv_aio_writev(s
->bs
->file
->bs
, offset
/ BDRV_SECTOR_SIZE
,
1100 &acb
->cur_qiov
, acb
->cur_qiov
.size
/ BDRV_SECTOR_SIZE
,
1105 * Populate back untouched region of new data cluster
1107 static void qed_aio_write_postfill(void *opaque
, int ret
)
1109 QEDAIOCB
*acb
= opaque
;
1110 BDRVQEDState
*s
= acb_to_s(acb
);
1111 uint64_t start
= acb
->cur_pos
+ acb
->cur_qiov
.size
;
1113 qed_start_of_cluster(s
, start
+ s
->header
.cluster_size
- 1) - start
;
1114 uint64_t offset
= acb
->cur_cluster
+
1115 qed_offset_into_cluster(s
, acb
->cur_pos
) +
1119 qed_aio_complete(acb
, ret
);
1123 trace_qed_aio_write_postfill(s
, acb
, start
, len
, offset
);
1124 qed_copy_from_backing_file(s
, start
, len
, offset
,
1125 qed_aio_write_main
, acb
);
1129 * Populate front untouched region of new data cluster
1131 static void qed_aio_write_prefill(void *opaque
, int ret
)
1133 QEDAIOCB
*acb
= opaque
;
1134 BDRVQEDState
*s
= acb_to_s(acb
);
1135 uint64_t start
= qed_start_of_cluster(s
, acb
->cur_pos
);
1136 uint64_t len
= qed_offset_into_cluster(s
, acb
->cur_pos
);
1138 trace_qed_aio_write_prefill(s
, acb
, start
, len
, acb
->cur_cluster
);
1139 qed_copy_from_backing_file(s
, start
, len
, acb
->cur_cluster
,
1140 qed_aio_write_postfill
, acb
);
1144 * Check if the QED_F_NEED_CHECK bit should be set during allocating write
1146 static bool qed_should_set_need_check(BDRVQEDState
*s
)
1148 /* The flush before L2 update path ensures consistency */
1149 if (s
->bs
->backing
) {
1153 return !(s
->header
.features
& QED_F_NEED_CHECK
);
1156 static void qed_aio_write_zero_cluster(void *opaque
, int ret
)
1158 QEDAIOCB
*acb
= opaque
;
1161 qed_aio_complete(acb
, ret
);
1165 qed_aio_write_l2_update(acb
, 0, 1);
1169 * Write new data cluster
1171 * @acb: Write request
1172 * @len: Length in bytes
1174 * This path is taken when writing to previously unallocated clusters.
1176 static void qed_aio_write_alloc(QEDAIOCB
*acb
, size_t len
)
1178 BDRVQEDState
*s
= acb_to_s(acb
);
1179 BlockCompletionFunc
*cb
;
1181 /* Cancel timer when the first allocating request comes in */
1182 if (QSIMPLEQ_EMPTY(&s
->allocating_write_reqs
)) {
1183 qed_cancel_need_check_timer(s
);
1186 /* Freeze this request if another allocating write is in progress */
1187 if (acb
!= QSIMPLEQ_FIRST(&s
->allocating_write_reqs
)) {
1188 QSIMPLEQ_INSERT_TAIL(&s
->allocating_write_reqs
, acb
, next
);
1190 if (acb
!= QSIMPLEQ_FIRST(&s
->allocating_write_reqs
) ||
1191 s
->allocating_write_reqs_plugged
) {
1192 return; /* wait for existing request to finish */
1195 acb
->cur_nclusters
= qed_bytes_to_clusters(s
,
1196 qed_offset_into_cluster(s
, acb
->cur_pos
) + len
);
1197 qemu_iovec_concat(&acb
->cur_qiov
, acb
->qiov
, acb
->qiov_offset
, len
);
1199 if (acb
->flags
& QED_AIOCB_ZERO
) {
1200 /* Skip ahead if the clusters are already zero */
1201 if (acb
->find_cluster_ret
== QED_CLUSTER_ZERO
) {
1202 qed_aio_next_io(acb
, 0);
1206 cb
= qed_aio_write_zero_cluster
;
1208 cb
= qed_aio_write_prefill
;
1209 acb
->cur_cluster
= qed_alloc_clusters(s
, acb
->cur_nclusters
);
1212 if (qed_should_set_need_check(s
)) {
1213 s
->header
.features
|= QED_F_NEED_CHECK
;
1214 qed_write_header(s
, cb
, acb
);
1221 * Write data cluster in place
1223 * @acb: Write request
1224 * @offset: Cluster offset in bytes
1225 * @len: Length in bytes
1227 * This path is taken when writing to already allocated clusters.
1229 static void qed_aio_write_inplace(QEDAIOCB
*acb
, uint64_t offset
, size_t len
)
1231 /* Allocate buffer for zero writes */
1232 if (acb
->flags
& QED_AIOCB_ZERO
) {
1233 struct iovec
*iov
= acb
->qiov
->iov
;
1235 if (!iov
->iov_base
) {
1236 iov
->iov_base
= qemu_try_blockalign(acb
->common
.bs
, iov
->iov_len
);
1237 if (iov
->iov_base
== NULL
) {
1238 qed_aio_complete(acb
, -ENOMEM
);
1241 memset(iov
->iov_base
, 0, iov
->iov_len
);
1245 /* Calculate the I/O vector */
1246 acb
->cur_cluster
= offset
;
1247 qemu_iovec_concat(&acb
->cur_qiov
, acb
->qiov
, acb
->qiov_offset
, len
);
1249 /* Do the actual write */
1250 qed_aio_write_main(acb
, 0);
1254 * Write data cluster
1256 * @opaque: Write request
1257 * @ret: QED_CLUSTER_FOUND, QED_CLUSTER_L2, QED_CLUSTER_L1,
1259 * @offset: Cluster offset in bytes
1260 * @len: Length in bytes
1262 * Callback from qed_find_cluster().
1264 static void qed_aio_write_data(void *opaque
, int ret
,
1265 uint64_t offset
, size_t len
)
1267 QEDAIOCB
*acb
= opaque
;
1269 trace_qed_aio_write_data(acb_to_s(acb
), acb
, ret
, offset
, len
);
1271 acb
->find_cluster_ret
= ret
;
1274 case QED_CLUSTER_FOUND
:
1275 qed_aio_write_inplace(acb
, offset
, len
);
1278 case QED_CLUSTER_L2
:
1279 case QED_CLUSTER_L1
:
1280 case QED_CLUSTER_ZERO
:
1281 qed_aio_write_alloc(acb
, len
);
1285 qed_aio_complete(acb
, ret
);
1293 * @opaque: Read request
1294 * @ret: QED_CLUSTER_FOUND, QED_CLUSTER_L2, QED_CLUSTER_L1,
1296 * @offset: Cluster offset in bytes
1297 * @len: Length in bytes
1299 * Callback from qed_find_cluster().
1301 static void qed_aio_read_data(void *opaque
, int ret
,
1302 uint64_t offset
, size_t len
)
1304 QEDAIOCB
*acb
= opaque
;
1305 BDRVQEDState
*s
= acb_to_s(acb
);
1306 BlockDriverState
*bs
= acb
->common
.bs
;
1308 /* Adjust offset into cluster */
1309 offset
+= qed_offset_into_cluster(s
, acb
->cur_pos
);
1311 trace_qed_aio_read_data(s
, acb
, ret
, offset
, len
);
1317 qemu_iovec_concat(&acb
->cur_qiov
, acb
->qiov
, acb
->qiov_offset
, len
);
1319 /* Handle zero cluster and backing file reads */
1320 if (ret
== QED_CLUSTER_ZERO
) {
1321 qemu_iovec_memset(&acb
->cur_qiov
, 0, 0, acb
->cur_qiov
.size
);
1322 qed_aio_next_io(acb
, 0);
1324 } else if (ret
!= QED_CLUSTER_FOUND
) {
1325 qed_read_backing_file(s
, acb
->cur_pos
, &acb
->cur_qiov
,
1326 &acb
->backing_qiov
, qed_aio_next_io
, acb
);
1330 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
1331 bdrv_aio_readv(bs
->file
->bs
, offset
/ BDRV_SECTOR_SIZE
,
1332 &acb
->cur_qiov
, acb
->cur_qiov
.size
/ BDRV_SECTOR_SIZE
,
1333 qed_aio_next_io
, acb
);
1337 qed_aio_complete(acb
, ret
);
1341 * Begin next I/O or complete the request
1343 static void qed_aio_next_io(void *opaque
, int ret
)
1345 QEDAIOCB
*acb
= opaque
;
1346 BDRVQEDState
*s
= acb_to_s(acb
);
1347 QEDFindClusterFunc
*io_fn
= (acb
->flags
& QED_AIOCB_WRITE
) ?
1348 qed_aio_write_data
: qed_aio_read_data
;
1350 trace_qed_aio_next_io(s
, acb
, ret
, acb
->cur_pos
+ acb
->cur_qiov
.size
);
1352 if (acb
->backing_qiov
) {
1353 qemu_iovec_destroy(acb
->backing_qiov
);
1354 g_free(acb
->backing_qiov
);
1355 acb
->backing_qiov
= NULL
;
1358 /* Handle I/O error */
1360 qed_aio_complete(acb
, ret
);
1364 acb
->qiov_offset
+= acb
->cur_qiov
.size
;
1365 acb
->cur_pos
+= acb
->cur_qiov
.size
;
1366 qemu_iovec_reset(&acb
->cur_qiov
);
1368 /* Complete request */
1369 if (acb
->cur_pos
>= acb
->end_pos
) {
1370 qed_aio_complete(acb
, 0);
1374 /* Find next cluster and start I/O */
1375 qed_find_cluster(s
, &acb
->request
,
1376 acb
->cur_pos
, acb
->end_pos
- acb
->cur_pos
,
1380 static BlockAIOCB
*qed_aio_setup(BlockDriverState
*bs
,
1382 QEMUIOVector
*qiov
, int nb_sectors
,
1383 BlockCompletionFunc
*cb
,
1384 void *opaque
, int flags
)
1386 QEDAIOCB
*acb
= qemu_aio_get(&qed_aiocb_info
, bs
, cb
, opaque
);
1388 trace_qed_aio_setup(bs
->opaque
, acb
, sector_num
, nb_sectors
,
1393 acb
->qiov_offset
= 0;
1394 acb
->cur_pos
= (uint64_t)sector_num
* BDRV_SECTOR_SIZE
;
1395 acb
->end_pos
= acb
->cur_pos
+ nb_sectors
* BDRV_SECTOR_SIZE
;
1396 acb
->backing_qiov
= NULL
;
1397 acb
->request
.l2_table
= NULL
;
1398 qemu_iovec_init(&acb
->cur_qiov
, qiov
->niov
);
1401 qed_aio_next_io(acb
, 0);
1402 return &acb
->common
;
1405 static BlockAIOCB
*bdrv_qed_aio_readv(BlockDriverState
*bs
,
1407 QEMUIOVector
*qiov
, int nb_sectors
,
1408 BlockCompletionFunc
*cb
,
1411 return qed_aio_setup(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 0);
1414 static BlockAIOCB
*bdrv_qed_aio_writev(BlockDriverState
*bs
,
1416 QEMUIOVector
*qiov
, int nb_sectors
,
1417 BlockCompletionFunc
*cb
,
1420 return qed_aio_setup(bs
, sector_num
, qiov
, nb_sectors
, cb
,
1421 opaque
, QED_AIOCB_WRITE
);
1430 static void coroutine_fn
qed_co_write_zeroes_cb(void *opaque
, int ret
)
1432 QEDWriteZeroesCB
*cb
= opaque
;
1437 qemu_coroutine_enter(cb
->co
, NULL
);
1441 static int coroutine_fn
bdrv_qed_co_write_zeroes(BlockDriverState
*bs
,
1444 BdrvRequestFlags flags
)
1446 BlockAIOCB
*blockacb
;
1447 BDRVQEDState
*s
= bs
->opaque
;
1448 QEDWriteZeroesCB cb
= { .done
= false };
1452 /* Refuse if there are untouched backing file sectors */
1454 if (qed_offset_into_cluster(s
, sector_num
* BDRV_SECTOR_SIZE
) != 0) {
1457 if (qed_offset_into_cluster(s
, nb_sectors
* BDRV_SECTOR_SIZE
) != 0) {
1462 /* Zero writes start without an I/O buffer. If a buffer becomes necessary
1463 * then it will be allocated during request processing.
1465 iov
.iov_base
= NULL
,
1466 iov
.iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
,
1468 qemu_iovec_init_external(&qiov
, &iov
, 1);
1469 blockacb
= qed_aio_setup(bs
, sector_num
, &qiov
, nb_sectors
,
1470 qed_co_write_zeroes_cb
, &cb
,
1471 QED_AIOCB_WRITE
| QED_AIOCB_ZERO
);
1476 cb
.co
= qemu_coroutine_self();
1477 qemu_coroutine_yield();
1483 static int bdrv_qed_truncate(BlockDriverState
*bs
, int64_t offset
)
1485 BDRVQEDState
*s
= bs
->opaque
;
1486 uint64_t old_image_size
;
1489 if (!qed_is_image_size_valid(offset
, s
->header
.cluster_size
,
1490 s
->header
.table_size
)) {
1494 /* Shrinking is currently not supported */
1495 if ((uint64_t)offset
< s
->header
.image_size
) {
1499 old_image_size
= s
->header
.image_size
;
1500 s
->header
.image_size
= offset
;
1501 ret
= qed_write_header_sync(s
);
1503 s
->header
.image_size
= old_image_size
;
1508 static int64_t bdrv_qed_getlength(BlockDriverState
*bs
)
1510 BDRVQEDState
*s
= bs
->opaque
;
1511 return s
->header
.image_size
;
1514 static int bdrv_qed_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
1516 BDRVQEDState
*s
= bs
->opaque
;
1518 memset(bdi
, 0, sizeof(*bdi
));
1519 bdi
->cluster_size
= s
->header
.cluster_size
;
1520 bdi
->is_dirty
= s
->header
.features
& QED_F_NEED_CHECK
;
1521 bdi
->unallocated_blocks_are_zero
= true;
1522 bdi
->can_write_zeroes_with_unmap
= true;
1526 static int bdrv_qed_change_backing_file(BlockDriverState
*bs
,
1527 const char *backing_file
,
1528 const char *backing_fmt
)
1530 BDRVQEDState
*s
= bs
->opaque
;
1531 QEDHeader new_header
, le_header
;
1533 size_t buffer_len
, backing_file_len
;
1536 /* Refuse to set backing filename if unknown compat feature bits are
1537 * active. If the image uses an unknown compat feature then we may not
1538 * know the layout of data following the header structure and cannot safely
1541 if (backing_file
&& (s
->header
.compat_features
&
1542 ~QED_COMPAT_FEATURE_MASK
)) {
1546 memcpy(&new_header
, &s
->header
, sizeof(new_header
));
1548 new_header
.features
&= ~(QED_F_BACKING_FILE
|
1549 QED_F_BACKING_FORMAT_NO_PROBE
);
1551 /* Adjust feature flags */
1553 new_header
.features
|= QED_F_BACKING_FILE
;
1555 if (qed_fmt_is_raw(backing_fmt
)) {
1556 new_header
.features
|= QED_F_BACKING_FORMAT_NO_PROBE
;
1560 /* Calculate new header size */
1561 backing_file_len
= 0;
1564 backing_file_len
= strlen(backing_file
);
1567 buffer_len
= sizeof(new_header
);
1568 new_header
.backing_filename_offset
= buffer_len
;
1569 new_header
.backing_filename_size
= backing_file_len
;
1570 buffer_len
+= backing_file_len
;
1572 /* Make sure we can rewrite header without failing */
1573 if (buffer_len
> new_header
.header_size
* new_header
.cluster_size
) {
1577 /* Prepare new header */
1578 buffer
= g_malloc(buffer_len
);
1580 qed_header_cpu_to_le(&new_header
, &le_header
);
1581 memcpy(buffer
, &le_header
, sizeof(le_header
));
1582 buffer_len
= sizeof(le_header
);
1585 memcpy(buffer
+ buffer_len
, backing_file
, backing_file_len
);
1586 buffer_len
+= backing_file_len
;
1589 /* Write new header */
1590 ret
= bdrv_pwrite_sync(bs
->file
->bs
, 0, buffer
, buffer_len
);
1593 memcpy(&s
->header
, &new_header
, sizeof(new_header
));
1598 static void bdrv_qed_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
1600 BDRVQEDState
*s
= bs
->opaque
;
1601 Error
*local_err
= NULL
;
1606 bdrv_invalidate_cache(bs
->file
->bs
, &local_err
);
1608 error_propagate(errp
, local_err
);
1612 memset(s
, 0, sizeof(BDRVQEDState
));
1613 ret
= bdrv_qed_open(bs
, NULL
, bs
->open_flags
, &local_err
);
1615 error_propagate(errp
, local_err
);
1616 error_prepend(errp
, "Could not reopen qed layer: ");
1618 } else if (ret
< 0) {
1619 error_setg_errno(errp
, -ret
, "Could not reopen qed layer");
1624 static int bdrv_qed_check(BlockDriverState
*bs
, BdrvCheckResult
*result
,
1627 BDRVQEDState
*s
= bs
->opaque
;
1629 return qed_check(s
, result
, !!fix
);
1632 static QemuOptsList qed_create_opts
= {
1633 .name
= "qed-create-opts",
1634 .head
= QTAILQ_HEAD_INITIALIZER(qed_create_opts
.head
),
1637 .name
= BLOCK_OPT_SIZE
,
1638 .type
= QEMU_OPT_SIZE
,
1639 .help
= "Virtual disk size"
1642 .name
= BLOCK_OPT_BACKING_FILE
,
1643 .type
= QEMU_OPT_STRING
,
1644 .help
= "File name of a base image"
1647 .name
= BLOCK_OPT_BACKING_FMT
,
1648 .type
= QEMU_OPT_STRING
,
1649 .help
= "Image format of the base image"
1652 .name
= BLOCK_OPT_CLUSTER_SIZE
,
1653 .type
= QEMU_OPT_SIZE
,
1654 .help
= "Cluster size (in bytes)",
1655 .def_value_str
= stringify(QED_DEFAULT_CLUSTER_SIZE
)
1658 .name
= BLOCK_OPT_TABLE_SIZE
,
1659 .type
= QEMU_OPT_SIZE
,
1660 .help
= "L1/L2 table size (in clusters)"
1662 { /* end of list */ }
1666 static BlockDriver bdrv_qed
= {
1667 .format_name
= "qed",
1668 .instance_size
= sizeof(BDRVQEDState
),
1669 .create_opts
= &qed_create_opts
,
1670 .supports_backing
= true,
1672 .bdrv_probe
= bdrv_qed_probe
,
1673 .bdrv_open
= bdrv_qed_open
,
1674 .bdrv_close
= bdrv_qed_close
,
1675 .bdrv_reopen_prepare
= bdrv_qed_reopen_prepare
,
1676 .bdrv_create
= bdrv_qed_create
,
1677 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
1678 .bdrv_co_get_block_status
= bdrv_qed_co_get_block_status
,
1679 .bdrv_aio_readv
= bdrv_qed_aio_readv
,
1680 .bdrv_aio_writev
= bdrv_qed_aio_writev
,
1681 .bdrv_co_write_zeroes
= bdrv_qed_co_write_zeroes
,
1682 .bdrv_truncate
= bdrv_qed_truncate
,
1683 .bdrv_getlength
= bdrv_qed_getlength
,
1684 .bdrv_get_info
= bdrv_qed_get_info
,
1685 .bdrv_refresh_limits
= bdrv_qed_refresh_limits
,
1686 .bdrv_change_backing_file
= bdrv_qed_change_backing_file
,
1687 .bdrv_invalidate_cache
= bdrv_qed_invalidate_cache
,
1688 .bdrv_check
= bdrv_qed_check
,
1689 .bdrv_detach_aio_context
= bdrv_qed_detach_aio_context
,
1690 .bdrv_attach_aio_context
= bdrv_qed_attach_aio_context
,
1691 .bdrv_drain
= bdrv_qed_drain
,
1694 static void bdrv_qed_init(void)
1696 bdrv_register(&bdrv_qed
);
1699 block_init(bdrv_qed_init
);