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/timer.h"
18 #include "qapi/qmp/qerror.h"
19 #include "migration/migration.h"
21 static void qed_aio_cancel(BlockDriverAIOCB
*blockacb
)
23 QEDAIOCB
*acb
= (QEDAIOCB
*)blockacb
;
24 bool finished
= false;
26 /* Wait for the request to finish */
27 acb
->finished
= &finished
;
33 static const AIOCBInfo qed_aiocb_info
= {
34 .aiocb_size
= sizeof(QEDAIOCB
),
35 .cancel
= qed_aio_cancel
,
38 static int bdrv_qed_probe(const uint8_t *buf
, int buf_size
,
41 const QEDHeader
*header
= (const QEDHeader
*)buf
;
43 if (buf_size
< sizeof(*header
)) {
46 if (le32_to_cpu(header
->magic
) != QED_MAGIC
) {
53 * Check whether an image format is raw
55 * @fmt: Backing file format, may be NULL
57 static bool qed_fmt_is_raw(const char *fmt
)
59 return fmt
&& strcmp(fmt
, "raw") == 0;
62 static void qed_header_le_to_cpu(const QEDHeader
*le
, QEDHeader
*cpu
)
64 cpu
->magic
= le32_to_cpu(le
->magic
);
65 cpu
->cluster_size
= le32_to_cpu(le
->cluster_size
);
66 cpu
->table_size
= le32_to_cpu(le
->table_size
);
67 cpu
->header_size
= le32_to_cpu(le
->header_size
);
68 cpu
->features
= le64_to_cpu(le
->features
);
69 cpu
->compat_features
= le64_to_cpu(le
->compat_features
);
70 cpu
->autoclear_features
= le64_to_cpu(le
->autoclear_features
);
71 cpu
->l1_table_offset
= le64_to_cpu(le
->l1_table_offset
);
72 cpu
->image_size
= le64_to_cpu(le
->image_size
);
73 cpu
->backing_filename_offset
= le32_to_cpu(le
->backing_filename_offset
);
74 cpu
->backing_filename_size
= le32_to_cpu(le
->backing_filename_size
);
77 static void qed_header_cpu_to_le(const QEDHeader
*cpu
, QEDHeader
*le
)
79 le
->magic
= cpu_to_le32(cpu
->magic
);
80 le
->cluster_size
= cpu_to_le32(cpu
->cluster_size
);
81 le
->table_size
= cpu_to_le32(cpu
->table_size
);
82 le
->header_size
= cpu_to_le32(cpu
->header_size
);
83 le
->features
= cpu_to_le64(cpu
->features
);
84 le
->compat_features
= cpu_to_le64(cpu
->compat_features
);
85 le
->autoclear_features
= cpu_to_le64(cpu
->autoclear_features
);
86 le
->l1_table_offset
= cpu_to_le64(cpu
->l1_table_offset
);
87 le
->image_size
= cpu_to_le64(cpu
->image_size
);
88 le
->backing_filename_offset
= cpu_to_le32(cpu
->backing_filename_offset
);
89 le
->backing_filename_size
= cpu_to_le32(cpu
->backing_filename_size
);
92 int qed_write_header_sync(BDRVQEDState
*s
)
97 qed_header_cpu_to_le(&s
->header
, &le
);
98 ret
= bdrv_pwrite(s
->bs
->file
, 0, &le
, sizeof(le
));
99 if (ret
!= sizeof(le
)) {
114 static void qed_write_header_cb(void *opaque
, int ret
)
116 QEDWriteHeaderCB
*write_header_cb
= opaque
;
118 qemu_vfree(write_header_cb
->buf
);
119 gencb_complete(write_header_cb
, ret
);
122 static void qed_write_header_read_cb(void *opaque
, int ret
)
124 QEDWriteHeaderCB
*write_header_cb
= opaque
;
125 BDRVQEDState
*s
= write_header_cb
->s
;
128 qed_write_header_cb(write_header_cb
, ret
);
133 qed_header_cpu_to_le(&s
->header
, (QEDHeader
*)write_header_cb
->buf
);
135 bdrv_aio_writev(s
->bs
->file
, 0, &write_header_cb
->qiov
,
136 write_header_cb
->nsectors
, qed_write_header_cb
,
141 * Update header in-place (does not rewrite backing filename or other strings)
143 * This function only updates known header fields in-place and does not affect
144 * extra data after the QED header.
146 static void qed_write_header(BDRVQEDState
*s
, BlockDriverCompletionFunc cb
,
149 /* We must write full sectors for O_DIRECT but cannot necessarily generate
150 * the data following the header if an unrecognized compat feature is
151 * active. Therefore, first read the sectors containing the header, update
152 * them, and write back.
155 int nsectors
= (sizeof(QEDHeader
) + BDRV_SECTOR_SIZE
- 1) /
157 size_t len
= nsectors
* BDRV_SECTOR_SIZE
;
158 QEDWriteHeaderCB
*write_header_cb
= gencb_alloc(sizeof(*write_header_cb
),
161 write_header_cb
->s
= s
;
162 write_header_cb
->nsectors
= nsectors
;
163 write_header_cb
->buf
= qemu_blockalign(s
->bs
, len
);
164 write_header_cb
->iov
.iov_base
= write_header_cb
->buf
;
165 write_header_cb
->iov
.iov_len
= len
;
166 qemu_iovec_init_external(&write_header_cb
->qiov
, &write_header_cb
->iov
, 1);
168 bdrv_aio_readv(s
->bs
->file
, 0, &write_header_cb
->qiov
, nsectors
,
169 qed_write_header_read_cb
, write_header_cb
);
172 static uint64_t qed_max_image_size(uint32_t cluster_size
, uint32_t table_size
)
174 uint64_t table_entries
;
177 table_entries
= (table_size
* cluster_size
) / sizeof(uint64_t);
178 l2_size
= table_entries
* cluster_size
;
180 return l2_size
* table_entries
;
183 static bool qed_is_cluster_size_valid(uint32_t cluster_size
)
185 if (cluster_size
< QED_MIN_CLUSTER_SIZE
||
186 cluster_size
> QED_MAX_CLUSTER_SIZE
) {
189 if (cluster_size
& (cluster_size
- 1)) {
190 return false; /* not power of 2 */
195 static bool qed_is_table_size_valid(uint32_t table_size
)
197 if (table_size
< QED_MIN_TABLE_SIZE
||
198 table_size
> QED_MAX_TABLE_SIZE
) {
201 if (table_size
& (table_size
- 1)) {
202 return false; /* not power of 2 */
207 static bool qed_is_image_size_valid(uint64_t image_size
, uint32_t cluster_size
,
210 if (image_size
% BDRV_SECTOR_SIZE
!= 0) {
211 return false; /* not multiple of sector size */
213 if (image_size
> qed_max_image_size(cluster_size
, table_size
)) {
214 return false; /* image is too large */
220 * Read a string of known length from the image file
223 * @offset: File offset to start of string, in bytes
224 * @n: String length in bytes
225 * @buf: Destination buffer
226 * @buflen: Destination buffer length in bytes
227 * @ret: 0 on success, -errno on failure
229 * The string is NUL-terminated.
231 static int qed_read_string(BlockDriverState
*file
, uint64_t offset
, size_t n
,
232 char *buf
, size_t buflen
)
238 ret
= bdrv_pread(file
, offset
, buf
, n
);
247 * Allocate new clusters
250 * @n: Number of contiguous clusters to allocate
251 * @ret: Offset of first allocated cluster
253 * This function only produces the offset where the new clusters should be
254 * written. It updates BDRVQEDState but does not make any changes to the image
257 static uint64_t qed_alloc_clusters(BDRVQEDState
*s
, unsigned int n
)
259 uint64_t offset
= s
->file_size
;
260 s
->file_size
+= n
* s
->header
.cluster_size
;
264 QEDTable
*qed_alloc_table(BDRVQEDState
*s
)
266 /* Honor O_DIRECT memory alignment requirements */
267 return qemu_blockalign(s
->bs
,
268 s
->header
.cluster_size
* s
->header
.table_size
);
272 * Allocate a new zeroed L2 table
274 static CachedL2Table
*qed_new_l2_table(BDRVQEDState
*s
)
276 CachedL2Table
*l2_table
= qed_alloc_l2_cache_entry(&s
->l2_cache
);
278 l2_table
->table
= qed_alloc_table(s
);
279 l2_table
->offset
= qed_alloc_clusters(s
, s
->header
.table_size
);
281 memset(l2_table
->table
->offsets
, 0,
282 s
->header
.cluster_size
* s
->header
.table_size
);
286 static void qed_aio_next_io(void *opaque
, int ret
);
288 static void qed_plug_allocating_write_reqs(BDRVQEDState
*s
)
290 assert(!s
->allocating_write_reqs_plugged
);
292 s
->allocating_write_reqs_plugged
= true;
295 static void qed_unplug_allocating_write_reqs(BDRVQEDState
*s
)
299 assert(s
->allocating_write_reqs_plugged
);
301 s
->allocating_write_reqs_plugged
= false;
303 acb
= QSIMPLEQ_FIRST(&s
->allocating_write_reqs
);
305 qed_aio_next_io(acb
, 0);
309 static void qed_finish_clear_need_check(void *opaque
, int ret
)
314 static void qed_flush_after_clear_need_check(void *opaque
, int ret
)
316 BDRVQEDState
*s
= opaque
;
318 bdrv_aio_flush(s
->bs
, qed_finish_clear_need_check
, s
);
320 /* No need to wait until flush completes */
321 qed_unplug_allocating_write_reqs(s
);
324 static void qed_clear_need_check(void *opaque
, int ret
)
326 BDRVQEDState
*s
= opaque
;
329 qed_unplug_allocating_write_reqs(s
);
333 s
->header
.features
&= ~QED_F_NEED_CHECK
;
334 qed_write_header(s
, qed_flush_after_clear_need_check
, s
);
337 static void qed_need_check_timer_cb(void *opaque
)
339 BDRVQEDState
*s
= opaque
;
341 /* The timer should only fire when allocating writes have drained */
342 assert(!QSIMPLEQ_FIRST(&s
->allocating_write_reqs
));
344 trace_qed_need_check_timer_cb(s
);
346 qed_plug_allocating_write_reqs(s
);
348 /* Ensure writes are on disk before clearing flag */
349 bdrv_aio_flush(s
->bs
, qed_clear_need_check
, s
);
352 static void qed_start_need_check_timer(BDRVQEDState
*s
)
354 trace_qed_start_need_check_timer(s
);
356 /* Use QEMU_CLOCK_VIRTUAL so we don't alter the image file while suspended for
359 timer_mod(s
->need_check_timer
, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL
) +
360 get_ticks_per_sec() * QED_NEED_CHECK_TIMEOUT
);
363 /* It's okay to call this multiple times or when no timer is started */
364 static void qed_cancel_need_check_timer(BDRVQEDState
*s
)
366 trace_qed_cancel_need_check_timer(s
);
367 timer_del(s
->need_check_timer
);
370 static void bdrv_qed_rebind(BlockDriverState
*bs
)
372 BDRVQEDState
*s
= bs
->opaque
;
376 static int bdrv_qed_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
379 BDRVQEDState
*s
= bs
->opaque
;
385 QSIMPLEQ_INIT(&s
->allocating_write_reqs
);
387 ret
= bdrv_pread(bs
->file
, 0, &le_header
, sizeof(le_header
));
391 qed_header_le_to_cpu(&le_header
, &s
->header
);
393 if (s
->header
.magic
!= QED_MAGIC
) {
394 error_setg(errp
, "Image not in QED format");
397 if (s
->header
.features
& ~QED_FEATURE_MASK
) {
398 /* image uses unsupported feature bits */
400 snprintf(buf
, sizeof(buf
), "%" PRIx64
,
401 s
->header
.features
& ~QED_FEATURE_MASK
);
402 error_set(errp
, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE
,
403 bs
->device_name
, "QED", buf
);
406 if (!qed_is_cluster_size_valid(s
->header
.cluster_size
)) {
410 /* Round down file size to the last cluster */
411 file_size
= bdrv_getlength(bs
->file
);
415 s
->file_size
= qed_start_of_cluster(s
, file_size
);
417 if (!qed_is_table_size_valid(s
->header
.table_size
)) {
420 if (!qed_is_image_size_valid(s
->header
.image_size
,
421 s
->header
.cluster_size
,
422 s
->header
.table_size
)) {
425 if (!qed_check_table_offset(s
, s
->header
.l1_table_offset
)) {
429 s
->table_nelems
= (s
->header
.cluster_size
* s
->header
.table_size
) /
431 s
->l2_shift
= ffs(s
->header
.cluster_size
) - 1;
432 s
->l2_mask
= s
->table_nelems
- 1;
433 s
->l1_shift
= s
->l2_shift
+ ffs(s
->table_nelems
) - 1;
435 if ((s
->header
.features
& QED_F_BACKING_FILE
)) {
436 if ((uint64_t)s
->header
.backing_filename_offset
+
437 s
->header
.backing_filename_size
>
438 s
->header
.cluster_size
* s
->header
.header_size
) {
442 ret
= qed_read_string(bs
->file
, s
->header
.backing_filename_offset
,
443 s
->header
.backing_filename_size
, bs
->backing_file
,
444 sizeof(bs
->backing_file
));
449 if (s
->header
.features
& QED_F_BACKING_FORMAT_NO_PROBE
) {
450 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), "raw");
454 /* Reset unknown autoclear feature bits. This is a backwards
455 * compatibility mechanism that allows images to be opened by older
456 * programs, which "knock out" unknown feature bits. When an image is
457 * opened by a newer program again it can detect that the autoclear
458 * feature is no longer valid.
460 if ((s
->header
.autoclear_features
& ~QED_AUTOCLEAR_FEATURE_MASK
) != 0 &&
461 !bdrv_is_read_only(bs
->file
) && !(flags
& BDRV_O_INCOMING
)) {
462 s
->header
.autoclear_features
&= QED_AUTOCLEAR_FEATURE_MASK
;
464 ret
= qed_write_header_sync(s
);
469 /* From here on only known autoclear feature bits are valid */
470 bdrv_flush(bs
->file
);
473 s
->l1_table
= qed_alloc_table(s
);
474 qed_init_l2_cache(&s
->l2_cache
);
476 ret
= qed_read_l1_table_sync(s
);
481 /* If image was not closed cleanly, check consistency */
482 if (!(flags
& BDRV_O_CHECK
) && (s
->header
.features
& QED_F_NEED_CHECK
)) {
483 /* Read-only images cannot be fixed. There is no risk of corruption
484 * since write operations are not possible. Therefore, allow
485 * potentially inconsistent images to be opened read-only. This can
486 * aid data recovery from an otherwise inconsistent image.
488 if (!bdrv_is_read_only(bs
->file
) &&
489 !(flags
& BDRV_O_INCOMING
)) {
490 BdrvCheckResult result
= {0};
492 ret
= qed_check(s
, &result
, true);
499 s
->need_check_timer
= timer_new_ns(QEMU_CLOCK_VIRTUAL
,
500 qed_need_check_timer_cb
, s
);
504 qed_free_l2_cache(&s
->l2_cache
);
505 qemu_vfree(s
->l1_table
);
510 static int bdrv_qed_refresh_limits(BlockDriverState
*bs
)
512 BDRVQEDState
*s
= bs
->opaque
;
514 bs
->bl
.write_zeroes_alignment
= s
->header
.cluster_size
>> BDRV_SECTOR_BITS
;
519 /* We have nothing to do for QED reopen, stubs just return
521 static int bdrv_qed_reopen_prepare(BDRVReopenState
*state
,
522 BlockReopenQueue
*queue
, Error
**errp
)
527 static void bdrv_qed_close(BlockDriverState
*bs
)
529 BDRVQEDState
*s
= bs
->opaque
;
531 qed_cancel_need_check_timer(s
);
532 timer_free(s
->need_check_timer
);
534 /* Ensure writes reach stable storage */
535 bdrv_flush(bs
->file
);
537 /* Clean shutdown, no check required on next open */
538 if (s
->header
.features
& QED_F_NEED_CHECK
) {
539 s
->header
.features
&= ~QED_F_NEED_CHECK
;
540 qed_write_header_sync(s
);
543 qed_free_l2_cache(&s
->l2_cache
);
544 qemu_vfree(s
->l1_table
);
547 static int qed_create(const char *filename
, uint32_t cluster_size
,
548 uint64_t image_size
, uint32_t table_size
,
549 const char *backing_file
, const char *backing_fmt
,
554 .cluster_size
= cluster_size
,
555 .table_size
= table_size
,
558 .compat_features
= 0,
559 .l1_table_offset
= cluster_size
,
560 .image_size
= image_size
,
563 uint8_t *l1_table
= NULL
;
564 size_t l1_size
= header
.cluster_size
* header
.table_size
;
565 Error
*local_err
= NULL
;
567 BlockDriverState
*bs
;
569 ret
= bdrv_create_file(filename
, NULL
, &local_err
);
571 error_propagate(errp
, local_err
);
576 ret
= bdrv_open(&bs
, filename
, NULL
, NULL
,
577 BDRV_O_RDWR
| BDRV_O_CACHE_WB
| BDRV_O_PROTOCOL
, NULL
,
580 error_propagate(errp
, local_err
);
584 /* File must start empty and grow, check truncate is supported */
585 ret
= bdrv_truncate(bs
, 0);
591 header
.features
|= QED_F_BACKING_FILE
;
592 header
.backing_filename_offset
= sizeof(le_header
);
593 header
.backing_filename_size
= strlen(backing_file
);
595 if (qed_fmt_is_raw(backing_fmt
)) {
596 header
.features
|= QED_F_BACKING_FORMAT_NO_PROBE
;
600 qed_header_cpu_to_le(&header
, &le_header
);
601 ret
= bdrv_pwrite(bs
, 0, &le_header
, sizeof(le_header
));
605 ret
= bdrv_pwrite(bs
, sizeof(le_header
), backing_file
,
606 header
.backing_filename_size
);
611 l1_table
= g_malloc0(l1_size
);
612 ret
= bdrv_pwrite(bs
, header
.l1_table_offset
, l1_table
, l1_size
);
617 ret
= 0; /* success */
624 static int bdrv_qed_create(const char *filename
, QEMUOptionParameter
*options
,
627 uint64_t image_size
= 0;
628 uint32_t cluster_size
= QED_DEFAULT_CLUSTER_SIZE
;
629 uint32_t table_size
= QED_DEFAULT_TABLE_SIZE
;
630 const char *backing_file
= NULL
;
631 const char *backing_fmt
= NULL
;
633 while (options
&& options
->name
) {
634 if (!strcmp(options
->name
, BLOCK_OPT_SIZE
)) {
635 image_size
= options
->value
.n
;
636 } else if (!strcmp(options
->name
, BLOCK_OPT_BACKING_FILE
)) {
637 backing_file
= options
->value
.s
;
638 } else if (!strcmp(options
->name
, BLOCK_OPT_BACKING_FMT
)) {
639 backing_fmt
= options
->value
.s
;
640 } else if (!strcmp(options
->name
, BLOCK_OPT_CLUSTER_SIZE
)) {
641 if (options
->value
.n
) {
642 cluster_size
= options
->value
.n
;
644 } else if (!strcmp(options
->name
, BLOCK_OPT_TABLE_SIZE
)) {
645 if (options
->value
.n
) {
646 table_size
= options
->value
.n
;
652 if (!qed_is_cluster_size_valid(cluster_size
)) {
653 error_setg(errp
, "QED cluster size must be within range [%u, %u] "
655 QED_MIN_CLUSTER_SIZE
, QED_MAX_CLUSTER_SIZE
);
658 if (!qed_is_table_size_valid(table_size
)) {
659 error_setg(errp
, "QED table size must be within range [%u, %u] "
661 QED_MIN_TABLE_SIZE
, QED_MAX_TABLE_SIZE
);
664 if (!qed_is_image_size_valid(image_size
, cluster_size
, table_size
)) {
665 error_setg(errp
, "QED image size must be a non-zero multiple of "
666 "cluster size and less than %" PRIu64
" bytes",
667 qed_max_image_size(cluster_size
, table_size
));
671 return qed_create(filename
, cluster_size
, image_size
, table_size
,
672 backing_file
, backing_fmt
, errp
);
676 BlockDriverState
*bs
;
683 static void qed_is_allocated_cb(void *opaque
, int ret
, uint64_t offset
, size_t len
)
685 QEDIsAllocatedCB
*cb
= opaque
;
686 BDRVQEDState
*s
= cb
->bs
->opaque
;
687 *cb
->pnum
= len
/ BDRV_SECTOR_SIZE
;
689 case QED_CLUSTER_FOUND
:
690 offset
|= qed_offset_into_cluster(s
, cb
->pos
);
691 cb
->status
= BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
| offset
;
693 case QED_CLUSTER_ZERO
:
694 cb
->status
= BDRV_BLOCK_ZERO
;
707 qemu_coroutine_enter(cb
->co
, NULL
);
711 static int64_t coroutine_fn
bdrv_qed_co_get_block_status(BlockDriverState
*bs
,
713 int nb_sectors
, int *pnum
)
715 BDRVQEDState
*s
= bs
->opaque
;
716 size_t len
= (size_t)nb_sectors
* BDRV_SECTOR_SIZE
;
717 QEDIsAllocatedCB cb
= {
719 .pos
= (uint64_t)sector_num
* BDRV_SECTOR_SIZE
,
720 .status
= BDRV_BLOCK_OFFSET_MASK
,
723 QEDRequest request
= { .l2_table
= NULL
};
725 qed_find_cluster(s
, &request
, cb
.pos
, len
, qed_is_allocated_cb
, &cb
);
727 /* Now sleep if the callback wasn't invoked immediately */
728 while (cb
.status
== BDRV_BLOCK_OFFSET_MASK
) {
729 cb
.co
= qemu_coroutine_self();
730 qemu_coroutine_yield();
733 qed_unref_l2_cache_entry(request
.l2_table
);
738 static BDRVQEDState
*acb_to_s(QEDAIOCB
*acb
)
740 return acb
->common
.bs
->opaque
;
744 * Read from the backing file or zero-fill if no backing file
747 * @pos: Byte position in device
748 * @qiov: Destination I/O vector
749 * @cb: Completion function
750 * @opaque: User data for completion function
752 * This function reads qiov->size bytes starting at pos from the backing file.
753 * If there is no backing file then zeroes are read.
755 static void qed_read_backing_file(BDRVQEDState
*s
, uint64_t pos
,
757 BlockDriverCompletionFunc
*cb
, void *opaque
)
759 uint64_t backing_length
= 0;
762 /* If there is a backing file, get its length. Treat the absence of a
763 * backing file like a zero length backing file.
765 if (s
->bs
->backing_hd
) {
766 int64_t l
= bdrv_getlength(s
->bs
->backing_hd
);
774 /* Zero all sectors if reading beyond the end of the backing file */
775 if (pos
>= backing_length
||
776 pos
+ qiov
->size
> backing_length
) {
777 qemu_iovec_memset(qiov
, 0, 0, qiov
->size
);
780 /* Complete now if there are no backing file sectors to read */
781 if (pos
>= backing_length
) {
786 /* If the read straddles the end of the backing file, shorten it */
787 size
= MIN((uint64_t)backing_length
- pos
, qiov
->size
);
789 BLKDBG_EVENT(s
->bs
->file
, BLKDBG_READ_BACKING_AIO
);
790 bdrv_aio_readv(s
->bs
->backing_hd
, pos
/ BDRV_SECTOR_SIZE
,
791 qiov
, size
/ BDRV_SECTOR_SIZE
, cb
, opaque
);
800 } CopyFromBackingFileCB
;
802 static void qed_copy_from_backing_file_cb(void *opaque
, int ret
)
804 CopyFromBackingFileCB
*copy_cb
= opaque
;
805 qemu_vfree(copy_cb
->iov
.iov_base
);
806 gencb_complete(©_cb
->gencb
, ret
);
809 static void qed_copy_from_backing_file_write(void *opaque
, int ret
)
811 CopyFromBackingFileCB
*copy_cb
= opaque
;
812 BDRVQEDState
*s
= copy_cb
->s
;
815 qed_copy_from_backing_file_cb(copy_cb
, ret
);
819 BLKDBG_EVENT(s
->bs
->file
, BLKDBG_COW_WRITE
);
820 bdrv_aio_writev(s
->bs
->file
, copy_cb
->offset
/ BDRV_SECTOR_SIZE
,
821 ©_cb
->qiov
, copy_cb
->qiov
.size
/ BDRV_SECTOR_SIZE
,
822 qed_copy_from_backing_file_cb
, copy_cb
);
826 * Copy data from backing file into the image
829 * @pos: Byte position in device
830 * @len: Number of bytes
831 * @offset: Byte offset in image file
832 * @cb: Completion function
833 * @opaque: User data for completion function
835 static void qed_copy_from_backing_file(BDRVQEDState
*s
, uint64_t pos
,
836 uint64_t len
, uint64_t offset
,
837 BlockDriverCompletionFunc
*cb
,
840 CopyFromBackingFileCB
*copy_cb
;
842 /* Skip copy entirely if there is no work to do */
848 copy_cb
= gencb_alloc(sizeof(*copy_cb
), cb
, opaque
);
850 copy_cb
->offset
= offset
;
851 copy_cb
->iov
.iov_base
= qemu_blockalign(s
->bs
, len
);
852 copy_cb
->iov
.iov_len
= len
;
853 qemu_iovec_init_external(©_cb
->qiov
, ©_cb
->iov
, 1);
855 qed_read_backing_file(s
, pos
, ©_cb
->qiov
,
856 qed_copy_from_backing_file_write
, copy_cb
);
860 * Link one or more contiguous clusters into a table
864 * @index: First cluster index
865 * @n: Number of contiguous clusters
866 * @cluster: First cluster offset
868 * The cluster offset may be an allocated byte offset in the image file, the
869 * zero cluster marker, or the unallocated cluster marker.
871 static void qed_update_l2_table(BDRVQEDState
*s
, QEDTable
*table
, int index
,
872 unsigned int n
, uint64_t cluster
)
875 for (i
= index
; i
< index
+ n
; i
++) {
876 table
->offsets
[i
] = cluster
;
877 if (!qed_offset_is_unalloc_cluster(cluster
) &&
878 !qed_offset_is_zero_cluster(cluster
)) {
879 cluster
+= s
->header
.cluster_size
;
884 static void qed_aio_complete_bh(void *opaque
)
886 QEDAIOCB
*acb
= opaque
;
887 BlockDriverCompletionFunc
*cb
= acb
->common
.cb
;
888 void *user_opaque
= acb
->common
.opaque
;
889 int ret
= acb
->bh_ret
;
890 bool *finished
= acb
->finished
;
892 qemu_bh_delete(acb
->bh
);
893 qemu_aio_release(acb
);
895 /* Invoke callback */
896 cb(user_opaque
, ret
);
898 /* Signal cancel completion */
904 static void qed_aio_complete(QEDAIOCB
*acb
, int ret
)
906 BDRVQEDState
*s
= acb_to_s(acb
);
908 trace_qed_aio_complete(s
, acb
, ret
);
911 qemu_iovec_destroy(&acb
->cur_qiov
);
912 qed_unref_l2_cache_entry(acb
->request
.l2_table
);
914 /* Free the buffer we may have allocated for zero writes */
915 if (acb
->flags
& QED_AIOCB_ZERO
) {
916 qemu_vfree(acb
->qiov
->iov
[0].iov_base
);
917 acb
->qiov
->iov
[0].iov_base
= NULL
;
920 /* Arrange for a bh to invoke the completion function */
922 acb
->bh
= qemu_bh_new(qed_aio_complete_bh
, acb
);
923 qemu_bh_schedule(acb
->bh
);
925 /* Start next allocating write request waiting behind this one. Note that
926 * requests enqueue themselves when they first hit an unallocated cluster
927 * but they wait until the entire request is finished before waking up the
928 * next request in the queue. This ensures that we don't cycle through
929 * requests multiple times but rather finish one at a time completely.
931 if (acb
== QSIMPLEQ_FIRST(&s
->allocating_write_reqs
)) {
932 QSIMPLEQ_REMOVE_HEAD(&s
->allocating_write_reqs
, next
);
933 acb
= QSIMPLEQ_FIRST(&s
->allocating_write_reqs
);
935 qed_aio_next_io(acb
, 0);
936 } else if (s
->header
.features
& QED_F_NEED_CHECK
) {
937 qed_start_need_check_timer(s
);
943 * Commit the current L2 table to the cache
945 static void qed_commit_l2_update(void *opaque
, int ret
)
947 QEDAIOCB
*acb
= opaque
;
948 BDRVQEDState
*s
= acb_to_s(acb
);
949 CachedL2Table
*l2_table
= acb
->request
.l2_table
;
950 uint64_t l2_offset
= l2_table
->offset
;
952 qed_commit_l2_cache_entry(&s
->l2_cache
, l2_table
);
954 /* This is guaranteed to succeed because we just committed the entry to the
957 acb
->request
.l2_table
= qed_find_l2_cache_entry(&s
->l2_cache
, l2_offset
);
958 assert(acb
->request
.l2_table
!= NULL
);
960 qed_aio_next_io(opaque
, ret
);
964 * Update L1 table with new L2 table offset and write it out
966 static void qed_aio_write_l1_update(void *opaque
, int ret
)
968 QEDAIOCB
*acb
= opaque
;
969 BDRVQEDState
*s
= acb_to_s(acb
);
973 qed_aio_complete(acb
, ret
);
977 index
= qed_l1_index(s
, acb
->cur_pos
);
978 s
->l1_table
->offsets
[index
] = acb
->request
.l2_table
->offset
;
980 qed_write_l1_table(s
, index
, 1, qed_commit_l2_update
, acb
);
984 * Update L2 table with new cluster offsets and write them out
986 static void qed_aio_write_l2_update(QEDAIOCB
*acb
, int ret
, uint64_t offset
)
988 BDRVQEDState
*s
= acb_to_s(acb
);
989 bool need_alloc
= acb
->find_cluster_ret
== QED_CLUSTER_L1
;
997 qed_unref_l2_cache_entry(acb
->request
.l2_table
);
998 acb
->request
.l2_table
= qed_new_l2_table(s
);
1001 index
= qed_l2_index(s
, acb
->cur_pos
);
1002 qed_update_l2_table(s
, acb
->request
.l2_table
->table
, index
, acb
->cur_nclusters
,
1006 /* Write out the whole new L2 table */
1007 qed_write_l2_table(s
, &acb
->request
, 0, s
->table_nelems
, true,
1008 qed_aio_write_l1_update
, acb
);
1010 /* Write out only the updated part of the L2 table */
1011 qed_write_l2_table(s
, &acb
->request
, index
, acb
->cur_nclusters
, false,
1012 qed_aio_next_io
, acb
);
1017 qed_aio_complete(acb
, ret
);
1020 static void qed_aio_write_l2_update_cb(void *opaque
, int ret
)
1022 QEDAIOCB
*acb
= opaque
;
1023 qed_aio_write_l2_update(acb
, ret
, acb
->cur_cluster
);
1027 * Flush new data clusters before updating the L2 table
1029 * This flush is necessary when a backing file is in use. A crash during an
1030 * allocating write could result in empty clusters in the image. If the write
1031 * only touched a subregion of the cluster, then backing image sectors have
1032 * been lost in the untouched region. The solution is to flush after writing a
1033 * new data cluster and before updating the L2 table.
1035 static void qed_aio_write_flush_before_l2_update(void *opaque
, int ret
)
1037 QEDAIOCB
*acb
= opaque
;
1038 BDRVQEDState
*s
= acb_to_s(acb
);
1040 if (!bdrv_aio_flush(s
->bs
->file
, qed_aio_write_l2_update_cb
, opaque
)) {
1041 qed_aio_complete(acb
, -EIO
);
1046 * Write data to the image file
1048 static void qed_aio_write_main(void *opaque
, int ret
)
1050 QEDAIOCB
*acb
= opaque
;
1051 BDRVQEDState
*s
= acb_to_s(acb
);
1052 uint64_t offset
= acb
->cur_cluster
+
1053 qed_offset_into_cluster(s
, acb
->cur_pos
);
1054 BlockDriverCompletionFunc
*next_fn
;
1056 trace_qed_aio_write_main(s
, acb
, ret
, offset
, acb
->cur_qiov
.size
);
1059 qed_aio_complete(acb
, ret
);
1063 if (acb
->find_cluster_ret
== QED_CLUSTER_FOUND
) {
1064 next_fn
= qed_aio_next_io
;
1066 if (s
->bs
->backing_hd
) {
1067 next_fn
= qed_aio_write_flush_before_l2_update
;
1069 next_fn
= qed_aio_write_l2_update_cb
;
1073 BLKDBG_EVENT(s
->bs
->file
, BLKDBG_WRITE_AIO
);
1074 bdrv_aio_writev(s
->bs
->file
, offset
/ BDRV_SECTOR_SIZE
,
1075 &acb
->cur_qiov
, acb
->cur_qiov
.size
/ BDRV_SECTOR_SIZE
,
1080 * Populate back untouched region of new data cluster
1082 static void qed_aio_write_postfill(void *opaque
, int ret
)
1084 QEDAIOCB
*acb
= opaque
;
1085 BDRVQEDState
*s
= acb_to_s(acb
);
1086 uint64_t start
= acb
->cur_pos
+ acb
->cur_qiov
.size
;
1088 qed_start_of_cluster(s
, start
+ s
->header
.cluster_size
- 1) - start
;
1089 uint64_t offset
= acb
->cur_cluster
+
1090 qed_offset_into_cluster(s
, acb
->cur_pos
) +
1094 qed_aio_complete(acb
, ret
);
1098 trace_qed_aio_write_postfill(s
, acb
, start
, len
, offset
);
1099 qed_copy_from_backing_file(s
, start
, len
, offset
,
1100 qed_aio_write_main
, acb
);
1104 * Populate front untouched region of new data cluster
1106 static void qed_aio_write_prefill(void *opaque
, int ret
)
1108 QEDAIOCB
*acb
= opaque
;
1109 BDRVQEDState
*s
= acb_to_s(acb
);
1110 uint64_t start
= qed_start_of_cluster(s
, acb
->cur_pos
);
1111 uint64_t len
= qed_offset_into_cluster(s
, acb
->cur_pos
);
1113 trace_qed_aio_write_prefill(s
, acb
, start
, len
, acb
->cur_cluster
);
1114 qed_copy_from_backing_file(s
, start
, len
, acb
->cur_cluster
,
1115 qed_aio_write_postfill
, acb
);
1119 * Check if the QED_F_NEED_CHECK bit should be set during allocating write
1121 static bool qed_should_set_need_check(BDRVQEDState
*s
)
1123 /* The flush before L2 update path ensures consistency */
1124 if (s
->bs
->backing_hd
) {
1128 return !(s
->header
.features
& QED_F_NEED_CHECK
);
1131 static void qed_aio_write_zero_cluster(void *opaque
, int ret
)
1133 QEDAIOCB
*acb
= opaque
;
1136 qed_aio_complete(acb
, ret
);
1140 qed_aio_write_l2_update(acb
, 0, 1);
1144 * Write new data cluster
1146 * @acb: Write request
1147 * @len: Length in bytes
1149 * This path is taken when writing to previously unallocated clusters.
1151 static void qed_aio_write_alloc(QEDAIOCB
*acb
, size_t len
)
1153 BDRVQEDState
*s
= acb_to_s(acb
);
1154 BlockDriverCompletionFunc
*cb
;
1156 /* Cancel timer when the first allocating request comes in */
1157 if (QSIMPLEQ_EMPTY(&s
->allocating_write_reqs
)) {
1158 qed_cancel_need_check_timer(s
);
1161 /* Freeze this request if another allocating write is in progress */
1162 if (acb
!= QSIMPLEQ_FIRST(&s
->allocating_write_reqs
)) {
1163 QSIMPLEQ_INSERT_TAIL(&s
->allocating_write_reqs
, acb
, next
);
1165 if (acb
!= QSIMPLEQ_FIRST(&s
->allocating_write_reqs
) ||
1166 s
->allocating_write_reqs_plugged
) {
1167 return; /* wait for existing request to finish */
1170 acb
->cur_nclusters
= qed_bytes_to_clusters(s
,
1171 qed_offset_into_cluster(s
, acb
->cur_pos
) + len
);
1172 qemu_iovec_concat(&acb
->cur_qiov
, acb
->qiov
, acb
->qiov_offset
, len
);
1174 if (acb
->flags
& QED_AIOCB_ZERO
) {
1175 /* Skip ahead if the clusters are already zero */
1176 if (acb
->find_cluster_ret
== QED_CLUSTER_ZERO
) {
1177 qed_aio_next_io(acb
, 0);
1181 cb
= qed_aio_write_zero_cluster
;
1183 cb
= qed_aio_write_prefill
;
1184 acb
->cur_cluster
= qed_alloc_clusters(s
, acb
->cur_nclusters
);
1187 if (qed_should_set_need_check(s
)) {
1188 s
->header
.features
|= QED_F_NEED_CHECK
;
1189 qed_write_header(s
, cb
, acb
);
1196 * Write data cluster in place
1198 * @acb: Write request
1199 * @offset: Cluster offset in bytes
1200 * @len: Length in bytes
1202 * This path is taken when writing to already allocated clusters.
1204 static void qed_aio_write_inplace(QEDAIOCB
*acb
, uint64_t offset
, size_t len
)
1206 /* Allocate buffer for zero writes */
1207 if (acb
->flags
& QED_AIOCB_ZERO
) {
1208 struct iovec
*iov
= acb
->qiov
->iov
;
1210 if (!iov
->iov_base
) {
1211 iov
->iov_base
= qemu_blockalign(acb
->common
.bs
, iov
->iov_len
);
1212 memset(iov
->iov_base
, 0, iov
->iov_len
);
1216 /* Calculate the I/O vector */
1217 acb
->cur_cluster
= offset
;
1218 qemu_iovec_concat(&acb
->cur_qiov
, acb
->qiov
, acb
->qiov_offset
, len
);
1220 /* Do the actual write */
1221 qed_aio_write_main(acb
, 0);
1225 * Write data cluster
1227 * @opaque: Write request
1228 * @ret: QED_CLUSTER_FOUND, QED_CLUSTER_L2, QED_CLUSTER_L1,
1230 * @offset: Cluster offset in bytes
1231 * @len: Length in bytes
1233 * Callback from qed_find_cluster().
1235 static void qed_aio_write_data(void *opaque
, int ret
,
1236 uint64_t offset
, size_t len
)
1238 QEDAIOCB
*acb
= opaque
;
1240 trace_qed_aio_write_data(acb_to_s(acb
), acb
, ret
, offset
, len
);
1242 acb
->find_cluster_ret
= ret
;
1245 case QED_CLUSTER_FOUND
:
1246 qed_aio_write_inplace(acb
, offset
, len
);
1249 case QED_CLUSTER_L2
:
1250 case QED_CLUSTER_L1
:
1251 case QED_CLUSTER_ZERO
:
1252 qed_aio_write_alloc(acb
, len
);
1256 qed_aio_complete(acb
, ret
);
1264 * @opaque: Read request
1265 * @ret: QED_CLUSTER_FOUND, QED_CLUSTER_L2, QED_CLUSTER_L1,
1267 * @offset: Cluster offset in bytes
1268 * @len: Length in bytes
1270 * Callback from qed_find_cluster().
1272 static void qed_aio_read_data(void *opaque
, int ret
,
1273 uint64_t offset
, size_t len
)
1275 QEDAIOCB
*acb
= opaque
;
1276 BDRVQEDState
*s
= acb_to_s(acb
);
1277 BlockDriverState
*bs
= acb
->common
.bs
;
1279 /* Adjust offset into cluster */
1280 offset
+= qed_offset_into_cluster(s
, acb
->cur_pos
);
1282 trace_qed_aio_read_data(s
, acb
, ret
, offset
, len
);
1288 qemu_iovec_concat(&acb
->cur_qiov
, acb
->qiov
, acb
->qiov_offset
, len
);
1290 /* Handle zero cluster and backing file reads */
1291 if (ret
== QED_CLUSTER_ZERO
) {
1292 qemu_iovec_memset(&acb
->cur_qiov
, 0, 0, acb
->cur_qiov
.size
);
1293 qed_aio_next_io(acb
, 0);
1295 } else if (ret
!= QED_CLUSTER_FOUND
) {
1296 qed_read_backing_file(s
, acb
->cur_pos
, &acb
->cur_qiov
,
1297 qed_aio_next_io
, acb
);
1301 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
1302 bdrv_aio_readv(bs
->file
, offset
/ BDRV_SECTOR_SIZE
,
1303 &acb
->cur_qiov
, acb
->cur_qiov
.size
/ BDRV_SECTOR_SIZE
,
1304 qed_aio_next_io
, acb
);
1308 qed_aio_complete(acb
, ret
);
1312 * Begin next I/O or complete the request
1314 static void qed_aio_next_io(void *opaque
, int ret
)
1316 QEDAIOCB
*acb
= opaque
;
1317 BDRVQEDState
*s
= acb_to_s(acb
);
1318 QEDFindClusterFunc
*io_fn
= (acb
->flags
& QED_AIOCB_WRITE
) ?
1319 qed_aio_write_data
: qed_aio_read_data
;
1321 trace_qed_aio_next_io(s
, acb
, ret
, acb
->cur_pos
+ acb
->cur_qiov
.size
);
1323 /* Handle I/O error */
1325 qed_aio_complete(acb
, ret
);
1329 acb
->qiov_offset
+= acb
->cur_qiov
.size
;
1330 acb
->cur_pos
+= acb
->cur_qiov
.size
;
1331 qemu_iovec_reset(&acb
->cur_qiov
);
1333 /* Complete request */
1334 if (acb
->cur_pos
>= acb
->end_pos
) {
1335 qed_aio_complete(acb
, 0);
1339 /* Find next cluster and start I/O */
1340 qed_find_cluster(s
, &acb
->request
,
1341 acb
->cur_pos
, acb
->end_pos
- acb
->cur_pos
,
1345 static BlockDriverAIOCB
*qed_aio_setup(BlockDriverState
*bs
,
1347 QEMUIOVector
*qiov
, int nb_sectors
,
1348 BlockDriverCompletionFunc
*cb
,
1349 void *opaque
, int flags
)
1351 QEDAIOCB
*acb
= qemu_aio_get(&qed_aiocb_info
, bs
, cb
, opaque
);
1353 trace_qed_aio_setup(bs
->opaque
, acb
, sector_num
, nb_sectors
,
1357 acb
->finished
= NULL
;
1359 acb
->qiov_offset
= 0;
1360 acb
->cur_pos
= (uint64_t)sector_num
* BDRV_SECTOR_SIZE
;
1361 acb
->end_pos
= acb
->cur_pos
+ nb_sectors
* BDRV_SECTOR_SIZE
;
1362 acb
->request
.l2_table
= NULL
;
1363 qemu_iovec_init(&acb
->cur_qiov
, qiov
->niov
);
1366 qed_aio_next_io(acb
, 0);
1367 return &acb
->common
;
1370 static BlockDriverAIOCB
*bdrv_qed_aio_readv(BlockDriverState
*bs
,
1372 QEMUIOVector
*qiov
, int nb_sectors
,
1373 BlockDriverCompletionFunc
*cb
,
1376 return qed_aio_setup(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 0);
1379 static BlockDriverAIOCB
*bdrv_qed_aio_writev(BlockDriverState
*bs
,
1381 QEMUIOVector
*qiov
, int nb_sectors
,
1382 BlockDriverCompletionFunc
*cb
,
1385 return qed_aio_setup(bs
, sector_num
, qiov
, nb_sectors
, cb
,
1386 opaque
, QED_AIOCB_WRITE
);
1395 static void coroutine_fn
qed_co_write_zeroes_cb(void *opaque
, int ret
)
1397 QEDWriteZeroesCB
*cb
= opaque
;
1402 qemu_coroutine_enter(cb
->co
, NULL
);
1406 static int coroutine_fn
bdrv_qed_co_write_zeroes(BlockDriverState
*bs
,
1409 BdrvRequestFlags flags
)
1411 BlockDriverAIOCB
*blockacb
;
1412 BDRVQEDState
*s
= bs
->opaque
;
1413 QEDWriteZeroesCB cb
= { .done
= false };
1417 /* Refuse if there are untouched backing file sectors */
1418 if (bs
->backing_hd
) {
1419 if (qed_offset_into_cluster(s
, sector_num
* BDRV_SECTOR_SIZE
) != 0) {
1422 if (qed_offset_into_cluster(s
, nb_sectors
* BDRV_SECTOR_SIZE
) != 0) {
1427 /* Zero writes start without an I/O buffer. If a buffer becomes necessary
1428 * then it will be allocated during request processing.
1430 iov
.iov_base
= NULL
,
1431 iov
.iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
,
1433 qemu_iovec_init_external(&qiov
, &iov
, 1);
1434 blockacb
= qed_aio_setup(bs
, sector_num
, &qiov
, nb_sectors
,
1435 qed_co_write_zeroes_cb
, &cb
,
1436 QED_AIOCB_WRITE
| QED_AIOCB_ZERO
);
1441 cb
.co
= qemu_coroutine_self();
1442 qemu_coroutine_yield();
1448 static int bdrv_qed_truncate(BlockDriverState
*bs
, int64_t offset
)
1450 BDRVQEDState
*s
= bs
->opaque
;
1451 uint64_t old_image_size
;
1454 if (!qed_is_image_size_valid(offset
, s
->header
.cluster_size
,
1455 s
->header
.table_size
)) {
1459 /* Shrinking is currently not supported */
1460 if ((uint64_t)offset
< s
->header
.image_size
) {
1464 old_image_size
= s
->header
.image_size
;
1465 s
->header
.image_size
= offset
;
1466 ret
= qed_write_header_sync(s
);
1468 s
->header
.image_size
= old_image_size
;
1473 static int64_t bdrv_qed_getlength(BlockDriverState
*bs
)
1475 BDRVQEDState
*s
= bs
->opaque
;
1476 return s
->header
.image_size
;
1479 static int bdrv_qed_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
1481 BDRVQEDState
*s
= bs
->opaque
;
1483 memset(bdi
, 0, sizeof(*bdi
));
1484 bdi
->cluster_size
= s
->header
.cluster_size
;
1485 bdi
->is_dirty
= s
->header
.features
& QED_F_NEED_CHECK
;
1486 bdi
->unallocated_blocks_are_zero
= true;
1487 bdi
->can_write_zeroes_with_unmap
= true;
1491 static int bdrv_qed_change_backing_file(BlockDriverState
*bs
,
1492 const char *backing_file
,
1493 const char *backing_fmt
)
1495 BDRVQEDState
*s
= bs
->opaque
;
1496 QEDHeader new_header
, le_header
;
1498 size_t buffer_len
, backing_file_len
;
1501 /* Refuse to set backing filename if unknown compat feature bits are
1502 * active. If the image uses an unknown compat feature then we may not
1503 * know the layout of data following the header structure and cannot safely
1506 if (backing_file
&& (s
->header
.compat_features
&
1507 ~QED_COMPAT_FEATURE_MASK
)) {
1511 memcpy(&new_header
, &s
->header
, sizeof(new_header
));
1513 new_header
.features
&= ~(QED_F_BACKING_FILE
|
1514 QED_F_BACKING_FORMAT_NO_PROBE
);
1516 /* Adjust feature flags */
1518 new_header
.features
|= QED_F_BACKING_FILE
;
1520 if (qed_fmt_is_raw(backing_fmt
)) {
1521 new_header
.features
|= QED_F_BACKING_FORMAT_NO_PROBE
;
1525 /* Calculate new header size */
1526 backing_file_len
= 0;
1529 backing_file_len
= strlen(backing_file
);
1532 buffer_len
= sizeof(new_header
);
1533 new_header
.backing_filename_offset
= buffer_len
;
1534 new_header
.backing_filename_size
= backing_file_len
;
1535 buffer_len
+= backing_file_len
;
1537 /* Make sure we can rewrite header without failing */
1538 if (buffer_len
> new_header
.header_size
* new_header
.cluster_size
) {
1542 /* Prepare new header */
1543 buffer
= g_malloc(buffer_len
);
1545 qed_header_cpu_to_le(&new_header
, &le_header
);
1546 memcpy(buffer
, &le_header
, sizeof(le_header
));
1547 buffer_len
= sizeof(le_header
);
1550 memcpy(buffer
+ buffer_len
, backing_file
, backing_file_len
);
1551 buffer_len
+= backing_file_len
;
1554 /* Write new header */
1555 ret
= bdrv_pwrite_sync(bs
->file
, 0, buffer
, buffer_len
);
1558 memcpy(&s
->header
, &new_header
, sizeof(new_header
));
1563 static void bdrv_qed_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
1565 BDRVQEDState
*s
= bs
->opaque
;
1566 Error
*local_err
= NULL
;
1571 bdrv_invalidate_cache(bs
->file
, &local_err
);
1573 error_propagate(errp
, local_err
);
1577 memset(s
, 0, sizeof(BDRVQEDState
));
1578 ret
= bdrv_qed_open(bs
, NULL
, bs
->open_flags
, &local_err
);
1580 error_setg(errp
, "Could not reopen qed layer: %s",
1581 error_get_pretty(local_err
));
1582 error_free(local_err
);
1584 } else if (ret
< 0) {
1585 error_setg_errno(errp
, -ret
, "Could not reopen qed layer");
1590 static int bdrv_qed_check(BlockDriverState
*bs
, BdrvCheckResult
*result
,
1593 BDRVQEDState
*s
= bs
->opaque
;
1595 return qed_check(s
, result
, !!fix
);
1598 static QEMUOptionParameter qed_create_options
[] = {
1600 .name
= BLOCK_OPT_SIZE
,
1602 .help
= "Virtual disk size (in bytes)"
1604 .name
= BLOCK_OPT_BACKING_FILE
,
1606 .help
= "File name of a base image"
1608 .name
= BLOCK_OPT_BACKING_FMT
,
1610 .help
= "Image format of the base image"
1612 .name
= BLOCK_OPT_CLUSTER_SIZE
,
1614 .help
= "Cluster size (in bytes)",
1615 .value
= { .n
= QED_DEFAULT_CLUSTER_SIZE
},
1617 .name
= BLOCK_OPT_TABLE_SIZE
,
1619 .help
= "L1/L2 table size (in clusters)"
1621 { /* end of list */ }
1624 static BlockDriver bdrv_qed
= {
1625 .format_name
= "qed",
1626 .instance_size
= sizeof(BDRVQEDState
),
1627 .create_options
= qed_create_options
,
1629 .bdrv_probe
= bdrv_qed_probe
,
1630 .bdrv_rebind
= bdrv_qed_rebind
,
1631 .bdrv_open
= bdrv_qed_open
,
1632 .bdrv_close
= bdrv_qed_close
,
1633 .bdrv_reopen_prepare
= bdrv_qed_reopen_prepare
,
1634 .bdrv_create
= bdrv_qed_create
,
1635 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
1636 .bdrv_co_get_block_status
= bdrv_qed_co_get_block_status
,
1637 .bdrv_aio_readv
= bdrv_qed_aio_readv
,
1638 .bdrv_aio_writev
= bdrv_qed_aio_writev
,
1639 .bdrv_co_write_zeroes
= bdrv_qed_co_write_zeroes
,
1640 .bdrv_truncate
= bdrv_qed_truncate
,
1641 .bdrv_getlength
= bdrv_qed_getlength
,
1642 .bdrv_get_info
= bdrv_qed_get_info
,
1643 .bdrv_refresh_limits
= bdrv_qed_refresh_limits
,
1644 .bdrv_change_backing_file
= bdrv_qed_change_backing_file
,
1645 .bdrv_invalidate_cache
= bdrv_qed_invalidate_cache
,
1646 .bdrv_check
= bdrv_qed_check
,
1649 static void bdrv_qed_init(void)
1651 bdrv_register(&bdrv_qed
);
1654 block_init(bdrv_qed_init
);