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 "block/qdict.h"
17 #include "qapi/error.h"
18 #include "qemu/timer.h"
19 #include "qemu/bswap.h"
20 #include "qemu/option.h"
23 #include "sysemu/block-backend.h"
24 #include "qapi/qmp/qdict.h"
25 #include "qapi/qobject-input-visitor.h"
26 #include "qapi/qapi-visit-block-core.h"
28 static QemuOptsList qed_create_opts
;
30 static int bdrv_qed_probe(const uint8_t *buf
, int buf_size
,
33 const QEDHeader
*header
= (const QEDHeader
*)buf
;
35 if (buf_size
< sizeof(*header
)) {
38 if (le32_to_cpu(header
->magic
) != QED_MAGIC
) {
45 * Check whether an image format is raw
47 * @fmt: Backing file format, may be NULL
49 static bool qed_fmt_is_raw(const char *fmt
)
51 return fmt
&& strcmp(fmt
, "raw") == 0;
54 static void qed_header_le_to_cpu(const QEDHeader
*le
, QEDHeader
*cpu
)
56 cpu
->magic
= le32_to_cpu(le
->magic
);
57 cpu
->cluster_size
= le32_to_cpu(le
->cluster_size
);
58 cpu
->table_size
= le32_to_cpu(le
->table_size
);
59 cpu
->header_size
= le32_to_cpu(le
->header_size
);
60 cpu
->features
= le64_to_cpu(le
->features
);
61 cpu
->compat_features
= le64_to_cpu(le
->compat_features
);
62 cpu
->autoclear_features
= le64_to_cpu(le
->autoclear_features
);
63 cpu
->l1_table_offset
= le64_to_cpu(le
->l1_table_offset
);
64 cpu
->image_size
= le64_to_cpu(le
->image_size
);
65 cpu
->backing_filename_offset
= le32_to_cpu(le
->backing_filename_offset
);
66 cpu
->backing_filename_size
= le32_to_cpu(le
->backing_filename_size
);
69 static void qed_header_cpu_to_le(const QEDHeader
*cpu
, QEDHeader
*le
)
71 le
->magic
= cpu_to_le32(cpu
->magic
);
72 le
->cluster_size
= cpu_to_le32(cpu
->cluster_size
);
73 le
->table_size
= cpu_to_le32(cpu
->table_size
);
74 le
->header_size
= cpu_to_le32(cpu
->header_size
);
75 le
->features
= cpu_to_le64(cpu
->features
);
76 le
->compat_features
= cpu_to_le64(cpu
->compat_features
);
77 le
->autoclear_features
= cpu_to_le64(cpu
->autoclear_features
);
78 le
->l1_table_offset
= cpu_to_le64(cpu
->l1_table_offset
);
79 le
->image_size
= cpu_to_le64(cpu
->image_size
);
80 le
->backing_filename_offset
= cpu_to_le32(cpu
->backing_filename_offset
);
81 le
->backing_filename_size
= cpu_to_le32(cpu
->backing_filename_size
);
84 int qed_write_header_sync(BDRVQEDState
*s
)
89 qed_header_cpu_to_le(&s
->header
, &le
);
90 ret
= bdrv_pwrite(s
->bs
->file
, 0, &le
, sizeof(le
));
91 if (ret
!= sizeof(le
)) {
98 * Update header in-place (does not rewrite backing filename or other strings)
100 * This function only updates known header fields in-place and does not affect
101 * extra data after the QED header.
103 * No new allocating reqs can start while this function runs.
105 static int coroutine_fn
qed_write_header(BDRVQEDState
*s
)
107 /* We must write full sectors for O_DIRECT but cannot necessarily generate
108 * the data following the header if an unrecognized compat feature is
109 * active. Therefore, first read the sectors containing the header, update
110 * them, and write back.
113 int nsectors
= DIV_ROUND_UP(sizeof(QEDHeader
), BDRV_SECTOR_SIZE
);
114 size_t len
= nsectors
* BDRV_SECTOR_SIZE
;
119 assert(s
->allocating_acb
|| s
->allocating_write_reqs_plugged
);
121 buf
= qemu_blockalign(s
->bs
, len
);
122 qemu_iovec_init_buf(&qiov
, buf
, len
);
124 ret
= bdrv_co_preadv(s
->bs
->file
, 0, qiov
.size
, &qiov
, 0);
130 qed_header_cpu_to_le(&s
->header
, (QEDHeader
*) buf
);
132 ret
= bdrv_co_pwritev(s
->bs
->file
, 0, qiov
.size
, &qiov
, 0);
143 static uint64_t qed_max_image_size(uint32_t cluster_size
, uint32_t table_size
)
145 uint64_t table_entries
;
148 table_entries
= (table_size
* cluster_size
) / sizeof(uint64_t);
149 l2_size
= table_entries
* cluster_size
;
151 return l2_size
* table_entries
;
154 static bool qed_is_cluster_size_valid(uint32_t cluster_size
)
156 if (cluster_size
< QED_MIN_CLUSTER_SIZE
||
157 cluster_size
> QED_MAX_CLUSTER_SIZE
) {
160 if (cluster_size
& (cluster_size
- 1)) {
161 return false; /* not power of 2 */
166 static bool qed_is_table_size_valid(uint32_t table_size
)
168 if (table_size
< QED_MIN_TABLE_SIZE
||
169 table_size
> QED_MAX_TABLE_SIZE
) {
172 if (table_size
& (table_size
- 1)) {
173 return false; /* not power of 2 */
178 static bool qed_is_image_size_valid(uint64_t image_size
, uint32_t cluster_size
,
181 if (image_size
% BDRV_SECTOR_SIZE
!= 0) {
182 return false; /* not multiple of sector size */
184 if (image_size
> qed_max_image_size(cluster_size
, table_size
)) {
185 return false; /* image is too large */
191 * Read a string of known length from the image file
194 * @offset: File offset to start of string, in bytes
195 * @n: String length in bytes
196 * @buf: Destination buffer
197 * @buflen: Destination buffer length in bytes
198 * @ret: 0 on success, -errno on failure
200 * The string is NUL-terminated.
202 static int qed_read_string(BdrvChild
*file
, uint64_t offset
, size_t n
,
203 char *buf
, size_t buflen
)
209 ret
= bdrv_pread(file
, offset
, buf
, n
);
218 * Allocate new clusters
221 * @n: Number of contiguous clusters to allocate
222 * @ret: Offset of first allocated cluster
224 * This function only produces the offset where the new clusters should be
225 * written. It updates BDRVQEDState but does not make any changes to the image
228 * Called with table_lock held.
230 static uint64_t qed_alloc_clusters(BDRVQEDState
*s
, unsigned int n
)
232 uint64_t offset
= s
->file_size
;
233 s
->file_size
+= n
* s
->header
.cluster_size
;
237 QEDTable
*qed_alloc_table(BDRVQEDState
*s
)
239 /* Honor O_DIRECT memory alignment requirements */
240 return qemu_blockalign(s
->bs
,
241 s
->header
.cluster_size
* s
->header
.table_size
);
245 * Allocate a new zeroed L2 table
247 * Called with table_lock held.
249 static CachedL2Table
*qed_new_l2_table(BDRVQEDState
*s
)
251 CachedL2Table
*l2_table
= qed_alloc_l2_cache_entry(&s
->l2_cache
);
253 l2_table
->table
= qed_alloc_table(s
);
254 l2_table
->offset
= qed_alloc_clusters(s
, s
->header
.table_size
);
256 memset(l2_table
->table
->offsets
, 0,
257 s
->header
.cluster_size
* s
->header
.table_size
);
261 static bool qed_plug_allocating_write_reqs(BDRVQEDState
*s
)
263 qemu_co_mutex_lock(&s
->table_lock
);
265 /* No reentrancy is allowed. */
266 assert(!s
->allocating_write_reqs_plugged
);
267 if (s
->allocating_acb
!= NULL
) {
268 /* Another allocating write came concurrently. This cannot happen
269 * from bdrv_qed_co_drain_begin, but it can happen when the timer runs.
271 qemu_co_mutex_unlock(&s
->table_lock
);
275 s
->allocating_write_reqs_plugged
= true;
276 qemu_co_mutex_unlock(&s
->table_lock
);
280 static void qed_unplug_allocating_write_reqs(BDRVQEDState
*s
)
282 qemu_co_mutex_lock(&s
->table_lock
);
283 assert(s
->allocating_write_reqs_plugged
);
284 s
->allocating_write_reqs_plugged
= false;
285 qemu_co_queue_next(&s
->allocating_write_reqs
);
286 qemu_co_mutex_unlock(&s
->table_lock
);
289 static void coroutine_fn
qed_need_check_timer_entry(void *opaque
)
291 BDRVQEDState
*s
= opaque
;
294 trace_qed_need_check_timer_cb(s
);
296 if (!qed_plug_allocating_write_reqs(s
)) {
300 /* Ensure writes are on disk before clearing flag */
301 ret
= bdrv_co_flush(s
->bs
->file
->bs
);
303 qed_unplug_allocating_write_reqs(s
);
307 s
->header
.features
&= ~QED_F_NEED_CHECK
;
308 ret
= qed_write_header(s
);
311 qed_unplug_allocating_write_reqs(s
);
313 ret
= bdrv_co_flush(s
->bs
);
317 static void qed_need_check_timer_cb(void *opaque
)
319 Coroutine
*co
= qemu_coroutine_create(qed_need_check_timer_entry
, opaque
);
320 qemu_coroutine_enter(co
);
323 static void qed_start_need_check_timer(BDRVQEDState
*s
)
325 trace_qed_start_need_check_timer(s
);
327 /* Use QEMU_CLOCK_VIRTUAL so we don't alter the image file while suspended for
330 timer_mod(s
->need_check_timer
, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL
) +
331 NANOSECONDS_PER_SECOND
* QED_NEED_CHECK_TIMEOUT
);
334 /* It's okay to call this multiple times or when no timer is started */
335 static void qed_cancel_need_check_timer(BDRVQEDState
*s
)
337 trace_qed_cancel_need_check_timer(s
);
338 timer_del(s
->need_check_timer
);
341 static void bdrv_qed_detach_aio_context(BlockDriverState
*bs
)
343 BDRVQEDState
*s
= bs
->opaque
;
345 qed_cancel_need_check_timer(s
);
346 timer_free(s
->need_check_timer
);
349 static void bdrv_qed_attach_aio_context(BlockDriverState
*bs
,
350 AioContext
*new_context
)
352 BDRVQEDState
*s
= bs
->opaque
;
354 s
->need_check_timer
= aio_timer_new(new_context
,
355 QEMU_CLOCK_VIRTUAL
, SCALE_NS
,
356 qed_need_check_timer_cb
, s
);
357 if (s
->header
.features
& QED_F_NEED_CHECK
) {
358 qed_start_need_check_timer(s
);
362 static void coroutine_fn
bdrv_qed_co_drain_begin(BlockDriverState
*bs
)
364 BDRVQEDState
*s
= bs
->opaque
;
366 /* Fire the timer immediately in order to start doing I/O as soon as the
369 if (s
->need_check_timer
&& timer_pending(s
->need_check_timer
)) {
370 qed_cancel_need_check_timer(s
);
371 qed_need_check_timer_entry(s
);
375 static void bdrv_qed_init_state(BlockDriverState
*bs
)
377 BDRVQEDState
*s
= bs
->opaque
;
379 memset(s
, 0, sizeof(BDRVQEDState
));
381 qemu_co_mutex_init(&s
->table_lock
);
382 qemu_co_queue_init(&s
->allocating_write_reqs
);
385 /* Called with table_lock held. */
386 static int coroutine_fn
bdrv_qed_do_open(BlockDriverState
*bs
, QDict
*options
,
387 int flags
, Error
**errp
)
389 BDRVQEDState
*s
= bs
->opaque
;
394 ret
= bdrv_pread(bs
->file
, 0, &le_header
, sizeof(le_header
));
398 qed_header_le_to_cpu(&le_header
, &s
->header
);
400 if (s
->header
.magic
!= QED_MAGIC
) {
401 error_setg(errp
, "Image not in QED format");
404 if (s
->header
.features
& ~QED_FEATURE_MASK
) {
405 /* image uses unsupported feature bits */
406 error_setg(errp
, "Unsupported QED features: %" PRIx64
,
407 s
->header
.features
& ~QED_FEATURE_MASK
);
410 if (!qed_is_cluster_size_valid(s
->header
.cluster_size
)) {
414 /* Round down file size to the last cluster */
415 file_size
= bdrv_getlength(bs
->file
->bs
);
419 s
->file_size
= qed_start_of_cluster(s
, file_size
);
421 if (!qed_is_table_size_valid(s
->header
.table_size
)) {
424 if (!qed_is_image_size_valid(s
->header
.image_size
,
425 s
->header
.cluster_size
,
426 s
->header
.table_size
)) {
429 if (!qed_check_table_offset(s
, s
->header
.l1_table_offset
)) {
433 s
->table_nelems
= (s
->header
.cluster_size
* s
->header
.table_size
) /
435 s
->l2_shift
= ctz32(s
->header
.cluster_size
);
436 s
->l2_mask
= s
->table_nelems
- 1;
437 s
->l1_shift
= s
->l2_shift
+ ctz32(s
->table_nelems
);
439 /* Header size calculation must not overflow uint32_t */
440 if (s
->header
.header_size
> UINT32_MAX
/ s
->header
.cluster_size
) {
444 if ((s
->header
.features
& QED_F_BACKING_FILE
)) {
445 if ((uint64_t)s
->header
.backing_filename_offset
+
446 s
->header
.backing_filename_size
>
447 s
->header
.cluster_size
* s
->header
.header_size
) {
451 ret
= qed_read_string(bs
->file
, s
->header
.backing_filename_offset
,
452 s
->header
.backing_filename_size
, bs
->backing_file
,
453 sizeof(bs
->backing_file
));
458 if (s
->header
.features
& QED_F_BACKING_FORMAT_NO_PROBE
) {
459 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), "raw");
463 /* Reset unknown autoclear feature bits. This is a backwards
464 * compatibility mechanism that allows images to be opened by older
465 * programs, which "knock out" unknown feature bits. When an image is
466 * opened by a newer program again it can detect that the autoclear
467 * feature is no longer valid.
469 if ((s
->header
.autoclear_features
& ~QED_AUTOCLEAR_FEATURE_MASK
) != 0 &&
470 !bdrv_is_read_only(bs
->file
->bs
) && !(flags
& BDRV_O_INACTIVE
)) {
471 s
->header
.autoclear_features
&= QED_AUTOCLEAR_FEATURE_MASK
;
473 ret
= qed_write_header_sync(s
);
478 /* From here on only known autoclear feature bits are valid */
479 bdrv_flush(bs
->file
->bs
);
482 s
->l1_table
= qed_alloc_table(s
);
483 qed_init_l2_cache(&s
->l2_cache
);
485 ret
= qed_read_l1_table_sync(s
);
490 /* If image was not closed cleanly, check consistency */
491 if (!(flags
& BDRV_O_CHECK
) && (s
->header
.features
& QED_F_NEED_CHECK
)) {
492 /* Read-only images cannot be fixed. There is no risk of corruption
493 * since write operations are not possible. Therefore, allow
494 * potentially inconsistent images to be opened read-only. This can
495 * aid data recovery from an otherwise inconsistent image.
497 if (!bdrv_is_read_only(bs
->file
->bs
) &&
498 !(flags
& BDRV_O_INACTIVE
)) {
499 BdrvCheckResult result
= {0};
501 ret
= qed_check(s
, &result
, true);
508 bdrv_qed_attach_aio_context(bs
, bdrv_get_aio_context(bs
));
512 qed_free_l2_cache(&s
->l2_cache
);
513 qemu_vfree(s
->l1_table
);
518 typedef struct QEDOpenCo
{
519 BlockDriverState
*bs
;
526 static void coroutine_fn
bdrv_qed_open_entry(void *opaque
)
528 QEDOpenCo
*qoc
= opaque
;
529 BDRVQEDState
*s
= qoc
->bs
->opaque
;
531 qemu_co_mutex_lock(&s
->table_lock
);
532 qoc
->ret
= bdrv_qed_do_open(qoc
->bs
, qoc
->options
, qoc
->flags
, qoc
->errp
);
533 qemu_co_mutex_unlock(&s
->table_lock
);
536 static int bdrv_qed_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
547 bs
->file
= bdrv_open_child(NULL
, options
, "file", bs
, &child_file
,
553 bdrv_qed_init_state(bs
);
554 if (qemu_in_coroutine()) {
555 bdrv_qed_open_entry(&qoc
);
557 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
558 qemu_coroutine_enter(qemu_coroutine_create(bdrv_qed_open_entry
, &qoc
));
559 BDRV_POLL_WHILE(bs
, qoc
.ret
== -EINPROGRESS
);
561 BDRV_POLL_WHILE(bs
, qoc
.ret
== -EINPROGRESS
);
565 static void bdrv_qed_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
567 BDRVQEDState
*s
= bs
->opaque
;
569 bs
->bl
.pwrite_zeroes_alignment
= s
->header
.cluster_size
;
572 /* We have nothing to do for QED reopen, stubs just return
574 static int bdrv_qed_reopen_prepare(BDRVReopenState
*state
,
575 BlockReopenQueue
*queue
, Error
**errp
)
580 static void bdrv_qed_close(BlockDriverState
*bs
)
582 BDRVQEDState
*s
= bs
->opaque
;
584 bdrv_qed_detach_aio_context(bs
);
586 /* Ensure writes reach stable storage */
587 bdrv_flush(bs
->file
->bs
);
589 /* Clean shutdown, no check required on next open */
590 if (s
->header
.features
& QED_F_NEED_CHECK
) {
591 s
->header
.features
&= ~QED_F_NEED_CHECK
;
592 qed_write_header_sync(s
);
595 qed_free_l2_cache(&s
->l2_cache
);
596 qemu_vfree(s
->l1_table
);
599 static int coroutine_fn
bdrv_qed_co_create(BlockdevCreateOptions
*opts
,
602 BlockdevCreateOptionsQed
*qed_opts
;
603 BlockBackend
*blk
= NULL
;
604 BlockDriverState
*bs
= NULL
;
608 uint8_t *l1_table
= NULL
;
612 assert(opts
->driver
== BLOCKDEV_DRIVER_QED
);
613 qed_opts
= &opts
->u
.qed
;
615 /* Validate options and set default values */
616 if (!qed_opts
->has_cluster_size
) {
617 qed_opts
->cluster_size
= QED_DEFAULT_CLUSTER_SIZE
;
619 if (!qed_opts
->has_table_size
) {
620 qed_opts
->table_size
= QED_DEFAULT_TABLE_SIZE
;
623 if (!qed_is_cluster_size_valid(qed_opts
->cluster_size
)) {
624 error_setg(errp
, "QED cluster size must be within range [%u, %u] "
626 QED_MIN_CLUSTER_SIZE
, QED_MAX_CLUSTER_SIZE
);
629 if (!qed_is_table_size_valid(qed_opts
->table_size
)) {
630 error_setg(errp
, "QED table size must be within range [%u, %u] "
632 QED_MIN_TABLE_SIZE
, QED_MAX_TABLE_SIZE
);
635 if (!qed_is_image_size_valid(qed_opts
->size
, qed_opts
->cluster_size
,
636 qed_opts
->table_size
))
638 error_setg(errp
, "QED image size must be a non-zero multiple of "
639 "cluster size and less than %" PRIu64
" bytes",
640 qed_max_image_size(qed_opts
->cluster_size
,
641 qed_opts
->table_size
));
645 /* Create BlockBackend to write to the image */
646 bs
= bdrv_open_blockdev_ref(qed_opts
->file
, errp
);
651 blk
= blk_new(BLK_PERM_WRITE
| BLK_PERM_RESIZE
, BLK_PERM_ALL
);
652 ret
= blk_insert_bs(blk
, bs
, errp
);
656 blk_set_allow_write_beyond_eof(blk
, true);
658 /* Prepare image format */
659 header
= (QEDHeader
) {
661 .cluster_size
= qed_opts
->cluster_size
,
662 .table_size
= qed_opts
->table_size
,
665 .compat_features
= 0,
666 .l1_table_offset
= qed_opts
->cluster_size
,
667 .image_size
= qed_opts
->size
,
670 l1_size
= header
.cluster_size
* header
.table_size
;
672 /* File must start empty and grow, check truncate is supported */
673 ret
= blk_truncate(blk
, 0, PREALLOC_MODE_OFF
, errp
);
678 if (qed_opts
->has_backing_file
) {
679 header
.features
|= QED_F_BACKING_FILE
;
680 header
.backing_filename_offset
= sizeof(le_header
);
681 header
.backing_filename_size
= strlen(qed_opts
->backing_file
);
683 if (qed_opts
->has_backing_fmt
) {
684 const char *backing_fmt
= BlockdevDriver_str(qed_opts
->backing_fmt
);
685 if (qed_fmt_is_raw(backing_fmt
)) {
686 header
.features
|= QED_F_BACKING_FORMAT_NO_PROBE
;
691 qed_header_cpu_to_le(&header
, &le_header
);
692 ret
= blk_pwrite(blk
, 0, &le_header
, sizeof(le_header
), 0);
696 ret
= blk_pwrite(blk
, sizeof(le_header
), qed_opts
->backing_file
,
697 header
.backing_filename_size
, 0);
702 l1_table
= g_malloc0(l1_size
);
703 ret
= blk_pwrite(blk
, header
.l1_table_offset
, l1_table
, l1_size
, 0);
708 ret
= 0; /* success */
716 static int coroutine_fn
bdrv_qed_co_create_opts(const char *filename
,
720 BlockdevCreateOptions
*create_options
= NULL
;
723 BlockDriverState
*bs
= NULL
;
724 Error
*local_err
= NULL
;
727 static const QDictRenames opt_renames
[] = {
728 { BLOCK_OPT_BACKING_FILE
, "backing-file" },
729 { BLOCK_OPT_BACKING_FMT
, "backing-fmt" },
730 { BLOCK_OPT_CLUSTER_SIZE
, "cluster-size" },
731 { BLOCK_OPT_TABLE_SIZE
, "table-size" },
735 /* Parse options and convert legacy syntax */
736 qdict
= qemu_opts_to_qdict_filtered(opts
, NULL
, &qed_create_opts
, true);
738 if (!qdict_rename_keys(qdict
, opt_renames
, errp
)) {
743 /* Create and open the file (protocol layer) */
744 ret
= bdrv_create_file(filename
, opts
, &local_err
);
746 error_propagate(errp
, local_err
);
750 bs
= bdrv_open(filename
, NULL
, NULL
,
751 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
, errp
);
757 /* Now get the QAPI type BlockdevCreateOptions */
758 qdict_put_str(qdict
, "driver", "qed");
759 qdict_put_str(qdict
, "file", bs
->node_name
);
761 v
= qobject_input_visitor_new_flat_confused(qdict
, errp
);
767 visit_type_BlockdevCreateOptions(v
, NULL
, &create_options
, &local_err
);
771 error_propagate(errp
, local_err
);
776 /* Silently round up size */
777 assert(create_options
->driver
== BLOCKDEV_DRIVER_QED
);
778 create_options
->u
.qed
.size
=
779 ROUND_UP(create_options
->u
.qed
.size
, BDRV_SECTOR_SIZE
);
781 /* Create the qed image (format layer) */
782 ret
= bdrv_qed_co_create(create_options
, errp
);
785 qobject_unref(qdict
);
787 qapi_free_BlockdevCreateOptions(create_options
);
791 static int coroutine_fn
bdrv_qed_co_block_status(BlockDriverState
*bs
,
793 int64_t pos
, int64_t bytes
,
794 int64_t *pnum
, int64_t *map
,
795 BlockDriverState
**file
)
797 BDRVQEDState
*s
= bs
->opaque
;
798 size_t len
= MIN(bytes
, SIZE_MAX
);
800 QEDRequest request
= { .l2_table
= NULL
};
804 qemu_co_mutex_lock(&s
->table_lock
);
805 ret
= qed_find_cluster(s
, &request
, pos
, &len
, &offset
);
809 case QED_CLUSTER_FOUND
:
810 *map
= offset
| qed_offset_into_cluster(s
, pos
);
811 status
= BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
812 *file
= bs
->file
->bs
;
814 case QED_CLUSTER_ZERO
:
815 status
= BDRV_BLOCK_ZERO
;
827 qed_unref_l2_cache_entry(request
.l2_table
);
828 qemu_co_mutex_unlock(&s
->table_lock
);
833 static BDRVQEDState
*acb_to_s(QEDAIOCB
*acb
)
835 return acb
->bs
->opaque
;
839 * Read from the backing file or zero-fill if no backing file
842 * @pos: Byte position in device
843 * @qiov: Destination I/O vector
844 * @backing_qiov: Possibly shortened copy of qiov, to be allocated here
845 * @cb: Completion function
846 * @opaque: User data for completion function
848 * This function reads qiov->size bytes starting at pos from the backing file.
849 * If there is no backing file then zeroes are read.
851 static int coroutine_fn
qed_read_backing_file(BDRVQEDState
*s
, uint64_t pos
,
853 QEMUIOVector
**backing_qiov
)
855 uint64_t backing_length
= 0;
859 /* If there is a backing file, get its length. Treat the absence of a
860 * backing file like a zero length backing file.
862 if (s
->bs
->backing
) {
863 int64_t l
= bdrv_getlength(s
->bs
->backing
->bs
);
870 /* Zero all sectors if reading beyond the end of the backing file */
871 if (pos
>= backing_length
||
872 pos
+ qiov
->size
> backing_length
) {
873 qemu_iovec_memset(qiov
, 0, 0, qiov
->size
);
876 /* Complete now if there are no backing file sectors to read */
877 if (pos
>= backing_length
) {
881 /* If the read straddles the end of the backing file, shorten it */
882 size
= MIN((uint64_t)backing_length
- pos
, qiov
->size
);
884 assert(*backing_qiov
== NULL
);
885 *backing_qiov
= g_new(QEMUIOVector
, 1);
886 qemu_iovec_init(*backing_qiov
, qiov
->niov
);
887 qemu_iovec_concat(*backing_qiov
, qiov
, 0, size
);
889 BLKDBG_EVENT(s
->bs
->file
, BLKDBG_READ_BACKING_AIO
);
890 ret
= bdrv_co_preadv(s
->bs
->backing
, pos
, size
, *backing_qiov
, 0);
898 * Copy data from backing file into the image
901 * @pos: Byte position in device
902 * @len: Number of bytes
903 * @offset: Byte offset in image file
905 static int coroutine_fn
qed_copy_from_backing_file(BDRVQEDState
*s
,
906 uint64_t pos
, uint64_t len
,
910 QEMUIOVector
*backing_qiov
= NULL
;
913 /* Skip copy entirely if there is no work to do */
918 qemu_iovec_init_buf(&qiov
, qemu_blockalign(s
->bs
, len
), len
);
920 ret
= qed_read_backing_file(s
, pos
, &qiov
, &backing_qiov
);
923 qemu_iovec_destroy(backing_qiov
);
924 g_free(backing_qiov
);
932 BLKDBG_EVENT(s
->bs
->file
, BLKDBG_COW_WRITE
);
933 ret
= bdrv_co_pwritev(s
->bs
->file
, offset
, qiov
.size
, &qiov
, 0);
939 qemu_vfree(qemu_iovec_buf(&qiov
));
944 * Link one or more contiguous clusters into a table
948 * @index: First cluster index
949 * @n: Number of contiguous clusters
950 * @cluster: First cluster offset
952 * The cluster offset may be an allocated byte offset in the image file, the
953 * zero cluster marker, or the unallocated cluster marker.
955 * Called with table_lock held.
957 static void coroutine_fn
qed_update_l2_table(BDRVQEDState
*s
, QEDTable
*table
,
958 int index
, unsigned int n
,
962 for (i
= index
; i
< index
+ n
; i
++) {
963 table
->offsets
[i
] = cluster
;
964 if (!qed_offset_is_unalloc_cluster(cluster
) &&
965 !qed_offset_is_zero_cluster(cluster
)) {
966 cluster
+= s
->header
.cluster_size
;
971 /* Called with table_lock held. */
972 static void coroutine_fn
qed_aio_complete(QEDAIOCB
*acb
)
974 BDRVQEDState
*s
= acb_to_s(acb
);
977 qemu_iovec_destroy(&acb
->cur_qiov
);
978 qed_unref_l2_cache_entry(acb
->request
.l2_table
);
980 /* Free the buffer we may have allocated for zero writes */
981 if (acb
->flags
& QED_AIOCB_ZERO
) {
982 qemu_vfree(acb
->qiov
->iov
[0].iov_base
);
983 acb
->qiov
->iov
[0].iov_base
= NULL
;
986 /* Start next allocating write request waiting behind this one. Note that
987 * requests enqueue themselves when they first hit an unallocated cluster
988 * but they wait until the entire request is finished before waking up the
989 * next request in the queue. This ensures that we don't cycle through
990 * requests multiple times but rather finish one at a time completely.
992 if (acb
== s
->allocating_acb
) {
993 s
->allocating_acb
= NULL
;
994 if (!qemu_co_queue_empty(&s
->allocating_write_reqs
)) {
995 qemu_co_queue_next(&s
->allocating_write_reqs
);
996 } else if (s
->header
.features
& QED_F_NEED_CHECK
) {
997 qed_start_need_check_timer(s
);
1003 * Update L1 table with new L2 table offset and write it out
1005 * Called with table_lock held.
1007 static int coroutine_fn
qed_aio_write_l1_update(QEDAIOCB
*acb
)
1009 BDRVQEDState
*s
= acb_to_s(acb
);
1010 CachedL2Table
*l2_table
= acb
->request
.l2_table
;
1011 uint64_t l2_offset
= l2_table
->offset
;
1014 index
= qed_l1_index(s
, acb
->cur_pos
);
1015 s
->l1_table
->offsets
[index
] = l2_table
->offset
;
1017 ret
= qed_write_l1_table(s
, index
, 1);
1019 /* Commit the current L2 table to the cache */
1020 qed_commit_l2_cache_entry(&s
->l2_cache
, l2_table
);
1022 /* This is guaranteed to succeed because we just committed the entry to the
1025 acb
->request
.l2_table
= qed_find_l2_cache_entry(&s
->l2_cache
, l2_offset
);
1026 assert(acb
->request
.l2_table
!= NULL
);
1033 * Update L2 table with new cluster offsets and write them out
1035 * Called with table_lock held.
1037 static int coroutine_fn
qed_aio_write_l2_update(QEDAIOCB
*acb
, uint64_t offset
)
1039 BDRVQEDState
*s
= acb_to_s(acb
);
1040 bool need_alloc
= acb
->find_cluster_ret
== QED_CLUSTER_L1
;
1044 qed_unref_l2_cache_entry(acb
->request
.l2_table
);
1045 acb
->request
.l2_table
= qed_new_l2_table(s
);
1048 index
= qed_l2_index(s
, acb
->cur_pos
);
1049 qed_update_l2_table(s
, acb
->request
.l2_table
->table
, index
, acb
->cur_nclusters
,
1053 /* Write out the whole new L2 table */
1054 ret
= qed_write_l2_table(s
, &acb
->request
, 0, s
->table_nelems
, true);
1058 return qed_aio_write_l1_update(acb
);
1060 /* Write out only the updated part of the L2 table */
1061 ret
= qed_write_l2_table(s
, &acb
->request
, index
, acb
->cur_nclusters
,
1071 * Write data to the image file
1073 * Called with table_lock *not* held.
1075 static int coroutine_fn
qed_aio_write_main(QEDAIOCB
*acb
)
1077 BDRVQEDState
*s
= acb_to_s(acb
);
1078 uint64_t offset
= acb
->cur_cluster
+
1079 qed_offset_into_cluster(s
, acb
->cur_pos
);
1081 trace_qed_aio_write_main(s
, acb
, 0, offset
, acb
->cur_qiov
.size
);
1083 BLKDBG_EVENT(s
->bs
->file
, BLKDBG_WRITE_AIO
);
1084 return bdrv_co_pwritev(s
->bs
->file
, offset
, acb
->cur_qiov
.size
,
1089 * Populate untouched regions of new data cluster
1091 * Called with table_lock held.
1093 static int coroutine_fn
qed_aio_write_cow(QEDAIOCB
*acb
)
1095 BDRVQEDState
*s
= acb_to_s(acb
);
1096 uint64_t start
, len
, offset
;
1099 qemu_co_mutex_unlock(&s
->table_lock
);
1101 /* Populate front untouched region of new data cluster */
1102 start
= qed_start_of_cluster(s
, acb
->cur_pos
);
1103 len
= qed_offset_into_cluster(s
, acb
->cur_pos
);
1105 trace_qed_aio_write_prefill(s
, acb
, start
, len
, acb
->cur_cluster
);
1106 ret
= qed_copy_from_backing_file(s
, start
, len
, acb
->cur_cluster
);
1111 /* Populate back untouched region of new data cluster */
1112 start
= acb
->cur_pos
+ acb
->cur_qiov
.size
;
1113 len
= qed_start_of_cluster(s
, start
+ s
->header
.cluster_size
- 1) - start
;
1114 offset
= acb
->cur_cluster
+
1115 qed_offset_into_cluster(s
, acb
->cur_pos
) +
1118 trace_qed_aio_write_postfill(s
, acb
, start
, len
, offset
);
1119 ret
= qed_copy_from_backing_file(s
, start
, len
, offset
);
1124 ret
= qed_aio_write_main(acb
);
1129 if (s
->bs
->backing
) {
1131 * Flush new data clusters before updating the L2 table
1133 * This flush is necessary when a backing file is in use. A crash
1134 * during an allocating write could result in empty clusters in the
1135 * image. If the write only touched a subregion of the cluster,
1136 * then backing image sectors have been lost in the untouched
1137 * region. The solution is to flush after writing a new data
1138 * cluster and before updating the L2 table.
1140 ret
= bdrv_co_flush(s
->bs
->file
->bs
);
1144 qemu_co_mutex_lock(&s
->table_lock
);
1149 * Check if the QED_F_NEED_CHECK bit should be set during allocating write
1151 static bool qed_should_set_need_check(BDRVQEDState
*s
)
1153 /* The flush before L2 update path ensures consistency */
1154 if (s
->bs
->backing
) {
1158 return !(s
->header
.features
& QED_F_NEED_CHECK
);
1162 * Write new data cluster
1164 * @acb: Write request
1165 * @len: Length in bytes
1167 * This path is taken when writing to previously unallocated clusters.
1169 * Called with table_lock held.
1171 static int coroutine_fn
qed_aio_write_alloc(QEDAIOCB
*acb
, size_t len
)
1173 BDRVQEDState
*s
= acb_to_s(acb
);
1176 /* Cancel timer when the first allocating request comes in */
1177 if (s
->allocating_acb
== NULL
) {
1178 qed_cancel_need_check_timer(s
);
1181 /* Freeze this request if another allocating write is in progress */
1182 if (s
->allocating_acb
!= acb
|| s
->allocating_write_reqs_plugged
) {
1183 if (s
->allocating_acb
!= NULL
) {
1184 qemu_co_queue_wait(&s
->allocating_write_reqs
, &s
->table_lock
);
1185 assert(s
->allocating_acb
== NULL
);
1187 s
->allocating_acb
= acb
;
1188 return -EAGAIN
; /* start over with looking up table entries */
1191 acb
->cur_nclusters
= qed_bytes_to_clusters(s
,
1192 qed_offset_into_cluster(s
, acb
->cur_pos
) + len
);
1193 qemu_iovec_concat(&acb
->cur_qiov
, acb
->qiov
, acb
->qiov_offset
, len
);
1195 if (acb
->flags
& QED_AIOCB_ZERO
) {
1196 /* Skip ahead if the clusters are already zero */
1197 if (acb
->find_cluster_ret
== QED_CLUSTER_ZERO
) {
1200 acb
->cur_cluster
= 1;
1202 acb
->cur_cluster
= qed_alloc_clusters(s
, acb
->cur_nclusters
);
1205 if (qed_should_set_need_check(s
)) {
1206 s
->header
.features
|= QED_F_NEED_CHECK
;
1207 ret
= qed_write_header(s
);
1213 if (!(acb
->flags
& QED_AIOCB_ZERO
)) {
1214 ret
= qed_aio_write_cow(acb
);
1220 return qed_aio_write_l2_update(acb
, acb
->cur_cluster
);
1224 * Write data cluster in place
1226 * @acb: Write request
1227 * @offset: Cluster offset in bytes
1228 * @len: Length in bytes
1230 * This path is taken when writing to already allocated clusters.
1232 * Called with table_lock held.
1234 static int coroutine_fn
qed_aio_write_inplace(QEDAIOCB
*acb
, uint64_t offset
,
1237 BDRVQEDState
*s
= acb_to_s(acb
);
1240 qemu_co_mutex_unlock(&s
->table_lock
);
1242 /* Allocate buffer for zero writes */
1243 if (acb
->flags
& QED_AIOCB_ZERO
) {
1244 struct iovec
*iov
= acb
->qiov
->iov
;
1246 if (!iov
->iov_base
) {
1247 iov
->iov_base
= qemu_try_blockalign(acb
->bs
, iov
->iov_len
);
1248 if (iov
->iov_base
== NULL
) {
1252 memset(iov
->iov_base
, 0, iov
->iov_len
);
1256 /* Calculate the I/O vector */
1257 acb
->cur_cluster
= offset
;
1258 qemu_iovec_concat(&acb
->cur_qiov
, acb
->qiov
, acb
->qiov_offset
, len
);
1260 /* Do the actual write. */
1261 r
= qed_aio_write_main(acb
);
1263 qemu_co_mutex_lock(&s
->table_lock
);
1268 * Write data cluster
1270 * @opaque: Write request
1271 * @ret: QED_CLUSTER_FOUND, QED_CLUSTER_L2 or QED_CLUSTER_L1
1272 * @offset: Cluster offset in bytes
1273 * @len: Length in bytes
1275 * Called with table_lock held.
1277 static int coroutine_fn
qed_aio_write_data(void *opaque
, int ret
,
1278 uint64_t offset
, size_t len
)
1280 QEDAIOCB
*acb
= opaque
;
1282 trace_qed_aio_write_data(acb_to_s(acb
), acb
, ret
, offset
, len
);
1284 acb
->find_cluster_ret
= ret
;
1287 case QED_CLUSTER_FOUND
:
1288 return qed_aio_write_inplace(acb
, offset
, len
);
1290 case QED_CLUSTER_L2
:
1291 case QED_CLUSTER_L1
:
1292 case QED_CLUSTER_ZERO
:
1293 return qed_aio_write_alloc(acb
, len
);
1296 g_assert_not_reached();
1303 * @opaque: Read request
1304 * @ret: QED_CLUSTER_FOUND, QED_CLUSTER_L2 or QED_CLUSTER_L1
1305 * @offset: Cluster offset in bytes
1306 * @len: Length in bytes
1308 * Called with table_lock held.
1310 static int coroutine_fn
qed_aio_read_data(void *opaque
, int ret
,
1311 uint64_t offset
, size_t len
)
1313 QEDAIOCB
*acb
= opaque
;
1314 BDRVQEDState
*s
= acb_to_s(acb
);
1315 BlockDriverState
*bs
= acb
->bs
;
1318 qemu_co_mutex_unlock(&s
->table_lock
);
1320 /* Adjust offset into cluster */
1321 offset
+= qed_offset_into_cluster(s
, acb
->cur_pos
);
1323 trace_qed_aio_read_data(s
, acb
, ret
, offset
, len
);
1325 qemu_iovec_concat(&acb
->cur_qiov
, acb
->qiov
, acb
->qiov_offset
, len
);
1327 /* Handle zero cluster and backing file reads, otherwise read
1328 * data cluster directly.
1330 if (ret
== QED_CLUSTER_ZERO
) {
1331 qemu_iovec_memset(&acb
->cur_qiov
, 0, 0, acb
->cur_qiov
.size
);
1333 } else if (ret
!= QED_CLUSTER_FOUND
) {
1334 r
= qed_read_backing_file(s
, acb
->cur_pos
, &acb
->cur_qiov
,
1335 &acb
->backing_qiov
);
1337 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
1338 r
= bdrv_co_preadv(bs
->file
, offset
, acb
->cur_qiov
.size
,
1342 qemu_co_mutex_lock(&s
->table_lock
);
1347 * Begin next I/O or complete the request
1349 static int coroutine_fn
qed_aio_next_io(QEDAIOCB
*acb
)
1351 BDRVQEDState
*s
= acb_to_s(acb
);
1356 qemu_co_mutex_lock(&s
->table_lock
);
1358 trace_qed_aio_next_io(s
, acb
, 0, acb
->cur_pos
+ acb
->cur_qiov
.size
);
1360 if (acb
->backing_qiov
) {
1361 qemu_iovec_destroy(acb
->backing_qiov
);
1362 g_free(acb
->backing_qiov
);
1363 acb
->backing_qiov
= NULL
;
1366 acb
->qiov_offset
+= acb
->cur_qiov
.size
;
1367 acb
->cur_pos
+= acb
->cur_qiov
.size
;
1368 qemu_iovec_reset(&acb
->cur_qiov
);
1370 /* Complete request */
1371 if (acb
->cur_pos
>= acb
->end_pos
) {
1376 /* Find next cluster and start I/O */
1377 len
= acb
->end_pos
- acb
->cur_pos
;
1378 ret
= qed_find_cluster(s
, &acb
->request
, acb
->cur_pos
, &len
, &offset
);
1383 if (acb
->flags
& QED_AIOCB_WRITE
) {
1384 ret
= qed_aio_write_data(acb
, ret
, offset
, len
);
1386 ret
= qed_aio_read_data(acb
, ret
, offset
, len
);
1389 if (ret
< 0 && ret
!= -EAGAIN
) {
1394 trace_qed_aio_complete(s
, acb
, ret
);
1395 qed_aio_complete(acb
);
1396 qemu_co_mutex_unlock(&s
->table_lock
);
1400 static int coroutine_fn
qed_co_request(BlockDriverState
*bs
, int64_t sector_num
,
1401 QEMUIOVector
*qiov
, int nb_sectors
,
1406 .cur_pos
= (uint64_t) sector_num
* BDRV_SECTOR_SIZE
,
1407 .end_pos
= (sector_num
+ nb_sectors
) * BDRV_SECTOR_SIZE
,
1411 qemu_iovec_init(&acb
.cur_qiov
, qiov
->niov
);
1413 trace_qed_aio_setup(bs
->opaque
, &acb
, sector_num
, nb_sectors
, NULL
, flags
);
1416 return qed_aio_next_io(&acb
);
1419 static int coroutine_fn
bdrv_qed_co_readv(BlockDriverState
*bs
,
1420 int64_t sector_num
, int nb_sectors
,
1423 return qed_co_request(bs
, sector_num
, qiov
, nb_sectors
, 0);
1426 static int coroutine_fn
bdrv_qed_co_writev(BlockDriverState
*bs
,
1427 int64_t sector_num
, int nb_sectors
,
1428 QEMUIOVector
*qiov
, int flags
)
1431 return qed_co_request(bs
, sector_num
, qiov
, nb_sectors
, QED_AIOCB_WRITE
);
1434 static int coroutine_fn
bdrv_qed_co_pwrite_zeroes(BlockDriverState
*bs
,
1437 BdrvRequestFlags flags
)
1439 BDRVQEDState
*s
= bs
->opaque
;
1442 * Zero writes start without an I/O buffer. If a buffer becomes necessary
1443 * then it will be allocated during request processing.
1445 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, NULL
, bytes
);
1447 /* Fall back if the request is not aligned */
1448 if (qed_offset_into_cluster(s
, offset
) ||
1449 qed_offset_into_cluster(s
, bytes
)) {
1453 return qed_co_request(bs
, offset
>> BDRV_SECTOR_BITS
, &qiov
,
1454 bytes
>> BDRV_SECTOR_BITS
,
1455 QED_AIOCB_WRITE
| QED_AIOCB_ZERO
);
1458 static int coroutine_fn
bdrv_qed_co_truncate(BlockDriverState
*bs
,
1460 PreallocMode prealloc
,
1463 BDRVQEDState
*s
= bs
->opaque
;
1464 uint64_t old_image_size
;
1467 if (prealloc
!= PREALLOC_MODE_OFF
) {
1468 error_setg(errp
, "Unsupported preallocation mode '%s'",
1469 PreallocMode_str(prealloc
));
1473 if (!qed_is_image_size_valid(offset
, s
->header
.cluster_size
,
1474 s
->header
.table_size
)) {
1475 error_setg(errp
, "Invalid image size specified");
1479 if ((uint64_t)offset
< s
->header
.image_size
) {
1480 error_setg(errp
, "Shrinking images is currently not supported");
1484 old_image_size
= s
->header
.image_size
;
1485 s
->header
.image_size
= offset
;
1486 ret
= qed_write_header_sync(s
);
1488 s
->header
.image_size
= old_image_size
;
1489 error_setg_errno(errp
, -ret
, "Failed to update the image size");
1494 static int64_t bdrv_qed_getlength(BlockDriverState
*bs
)
1496 BDRVQEDState
*s
= bs
->opaque
;
1497 return s
->header
.image_size
;
1500 static int bdrv_qed_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
1502 BDRVQEDState
*s
= bs
->opaque
;
1504 memset(bdi
, 0, sizeof(*bdi
));
1505 bdi
->cluster_size
= s
->header
.cluster_size
;
1506 bdi
->is_dirty
= s
->header
.features
& QED_F_NEED_CHECK
;
1507 bdi
->unallocated_blocks_are_zero
= true;
1511 static int bdrv_qed_change_backing_file(BlockDriverState
*bs
,
1512 const char *backing_file
,
1513 const char *backing_fmt
)
1515 BDRVQEDState
*s
= bs
->opaque
;
1516 QEDHeader new_header
, le_header
;
1518 size_t buffer_len
, backing_file_len
;
1521 /* Refuse to set backing filename if unknown compat feature bits are
1522 * active. If the image uses an unknown compat feature then we may not
1523 * know the layout of data following the header structure and cannot safely
1526 if (backing_file
&& (s
->header
.compat_features
&
1527 ~QED_COMPAT_FEATURE_MASK
)) {
1531 memcpy(&new_header
, &s
->header
, sizeof(new_header
));
1533 new_header
.features
&= ~(QED_F_BACKING_FILE
|
1534 QED_F_BACKING_FORMAT_NO_PROBE
);
1536 /* Adjust feature flags */
1538 new_header
.features
|= QED_F_BACKING_FILE
;
1540 if (qed_fmt_is_raw(backing_fmt
)) {
1541 new_header
.features
|= QED_F_BACKING_FORMAT_NO_PROBE
;
1545 /* Calculate new header size */
1546 backing_file_len
= 0;
1549 backing_file_len
= strlen(backing_file
);
1552 buffer_len
= sizeof(new_header
);
1553 new_header
.backing_filename_offset
= buffer_len
;
1554 new_header
.backing_filename_size
= backing_file_len
;
1555 buffer_len
+= backing_file_len
;
1557 /* Make sure we can rewrite header without failing */
1558 if (buffer_len
> new_header
.header_size
* new_header
.cluster_size
) {
1562 /* Prepare new header */
1563 buffer
= g_malloc(buffer_len
);
1565 qed_header_cpu_to_le(&new_header
, &le_header
);
1566 memcpy(buffer
, &le_header
, sizeof(le_header
));
1567 buffer_len
= sizeof(le_header
);
1570 memcpy(buffer
+ buffer_len
, backing_file
, backing_file_len
);
1571 buffer_len
+= backing_file_len
;
1574 /* Write new header */
1575 ret
= bdrv_pwrite_sync(bs
->file
, 0, buffer
, buffer_len
);
1578 memcpy(&s
->header
, &new_header
, sizeof(new_header
));
1583 static void coroutine_fn
bdrv_qed_co_invalidate_cache(BlockDriverState
*bs
,
1586 BDRVQEDState
*s
= bs
->opaque
;
1587 Error
*local_err
= NULL
;
1592 bdrv_qed_init_state(bs
);
1593 qemu_co_mutex_lock(&s
->table_lock
);
1594 ret
= bdrv_qed_do_open(bs
, NULL
, bs
->open_flags
, &local_err
);
1595 qemu_co_mutex_unlock(&s
->table_lock
);
1597 error_propagate_prepend(errp
, local_err
,
1598 "Could not reopen qed layer: ");
1600 } else if (ret
< 0) {
1601 error_setg_errno(errp
, -ret
, "Could not reopen qed layer");
1606 static int bdrv_qed_co_check(BlockDriverState
*bs
, BdrvCheckResult
*result
,
1609 BDRVQEDState
*s
= bs
->opaque
;
1612 qemu_co_mutex_lock(&s
->table_lock
);
1613 ret
= qed_check(s
, result
, !!fix
);
1614 qemu_co_mutex_unlock(&s
->table_lock
);
1619 static QemuOptsList qed_create_opts
= {
1620 .name
= "qed-create-opts",
1621 .head
= QTAILQ_HEAD_INITIALIZER(qed_create_opts
.head
),
1624 .name
= BLOCK_OPT_SIZE
,
1625 .type
= QEMU_OPT_SIZE
,
1626 .help
= "Virtual disk size"
1629 .name
= BLOCK_OPT_BACKING_FILE
,
1630 .type
= QEMU_OPT_STRING
,
1631 .help
= "File name of a base image"
1634 .name
= BLOCK_OPT_BACKING_FMT
,
1635 .type
= QEMU_OPT_STRING
,
1636 .help
= "Image format of the base image"
1639 .name
= BLOCK_OPT_CLUSTER_SIZE
,
1640 .type
= QEMU_OPT_SIZE
,
1641 .help
= "Cluster size (in bytes)",
1642 .def_value_str
= stringify(QED_DEFAULT_CLUSTER_SIZE
)
1645 .name
= BLOCK_OPT_TABLE_SIZE
,
1646 .type
= QEMU_OPT_SIZE
,
1647 .help
= "L1/L2 table size (in clusters)"
1649 { /* end of list */ }
1653 static BlockDriver bdrv_qed
= {
1654 .format_name
= "qed",
1655 .instance_size
= sizeof(BDRVQEDState
),
1656 .create_opts
= &qed_create_opts
,
1657 .supports_backing
= true,
1659 .bdrv_probe
= bdrv_qed_probe
,
1660 .bdrv_open
= bdrv_qed_open
,
1661 .bdrv_close
= bdrv_qed_close
,
1662 .bdrv_reopen_prepare
= bdrv_qed_reopen_prepare
,
1663 .bdrv_child_perm
= bdrv_format_default_perms
,
1664 .bdrv_co_create
= bdrv_qed_co_create
,
1665 .bdrv_co_create_opts
= bdrv_qed_co_create_opts
,
1666 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
1667 .bdrv_co_block_status
= bdrv_qed_co_block_status
,
1668 .bdrv_co_readv
= bdrv_qed_co_readv
,
1669 .bdrv_co_writev
= bdrv_qed_co_writev
,
1670 .bdrv_co_pwrite_zeroes
= bdrv_qed_co_pwrite_zeroes
,
1671 .bdrv_co_truncate
= bdrv_qed_co_truncate
,
1672 .bdrv_getlength
= bdrv_qed_getlength
,
1673 .bdrv_get_info
= bdrv_qed_get_info
,
1674 .bdrv_refresh_limits
= bdrv_qed_refresh_limits
,
1675 .bdrv_change_backing_file
= bdrv_qed_change_backing_file
,
1676 .bdrv_co_invalidate_cache
= bdrv_qed_co_invalidate_cache
,
1677 .bdrv_co_check
= bdrv_qed_co_check
,
1678 .bdrv_detach_aio_context
= bdrv_qed_detach_aio_context
,
1679 .bdrv_attach_aio_context
= bdrv_qed_attach_aio_context
,
1680 .bdrv_co_drain_begin
= bdrv_qed_co_drain_begin
,
1683 static void bdrv_qed_init(void)
1685 bdrv_register(&bdrv_qed
);
1688 block_init(bdrv_qed_init
);