2 * Block driver for the QCOW version 2 format
4 * Copyright (c) 2004-2006 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
27 #include "qemu-common.h"
28 #include "block/block_int.h"
29 #include "block/qcow2.h"
32 int qcow2_grow_l1_table(BlockDriverState
*bs
, uint64_t min_size
,
35 BDRVQcowState
*s
= bs
->opaque
;
36 int new_l1_size2
, ret
, i
;
37 uint64_t *new_l1_table
;
38 int64_t old_l1_table_offset
, old_l1_size
;
39 int64_t new_l1_table_offset
, new_l1_size
;
42 if (min_size
<= s
->l1_size
)
45 /* Do a sanity check on min_size before trying to calculate new_l1_size
46 * (this prevents overflows during the while loop for the calculation of
48 if (min_size
> INT_MAX
/ sizeof(uint64_t)) {
53 new_l1_size
= min_size
;
55 /* Bump size up to reduce the number of times we have to grow */
56 new_l1_size
= s
->l1_size
;
57 if (new_l1_size
== 0) {
60 while (min_size
> new_l1_size
) {
61 new_l1_size
= (new_l1_size
* 3 + 1) / 2;
65 if (new_l1_size
> INT_MAX
/ sizeof(uint64_t)) {
70 fprintf(stderr
, "grow l1_table from %d to %" PRId64
"\n",
71 s
->l1_size
, new_l1_size
);
74 new_l1_size2
= sizeof(uint64_t) * new_l1_size
;
75 new_l1_table
= qemu_try_blockalign(bs
->file
,
76 align_offset(new_l1_size2
, 512));
77 if (new_l1_table
== NULL
) {
80 memset(new_l1_table
, 0, align_offset(new_l1_size2
, 512));
82 memcpy(new_l1_table
, s
->l1_table
, s
->l1_size
* sizeof(uint64_t));
84 /* write new table (align to cluster) */
85 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_GROW_ALLOC_TABLE
);
86 new_l1_table_offset
= qcow2_alloc_clusters(bs
, new_l1_size2
);
87 if (new_l1_table_offset
< 0) {
88 qemu_vfree(new_l1_table
);
89 return new_l1_table_offset
;
92 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
97 /* the L1 position has not yet been updated, so these clusters must
98 * indeed be completely free */
99 ret
= qcow2_pre_write_overlap_check(bs
, 0, new_l1_table_offset
,
105 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_GROW_WRITE_TABLE
);
106 for(i
= 0; i
< s
->l1_size
; i
++)
107 new_l1_table
[i
] = cpu_to_be64(new_l1_table
[i
]);
108 ret
= bdrv_pwrite_sync(bs
->file
, new_l1_table_offset
, new_l1_table
, new_l1_size2
);
111 for(i
= 0; i
< s
->l1_size
; i
++)
112 new_l1_table
[i
] = be64_to_cpu(new_l1_table
[i
]);
115 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_GROW_ACTIVATE_TABLE
);
116 cpu_to_be32w((uint32_t*)data
, new_l1_size
);
117 stq_be_p(data
+ 4, new_l1_table_offset
);
118 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, l1_size
), data
,sizeof(data
));
122 qemu_vfree(s
->l1_table
);
123 old_l1_table_offset
= s
->l1_table_offset
;
124 s
->l1_table_offset
= new_l1_table_offset
;
125 s
->l1_table
= new_l1_table
;
126 old_l1_size
= s
->l1_size
;
127 s
->l1_size
= new_l1_size
;
128 qcow2_free_clusters(bs
, old_l1_table_offset
, old_l1_size
* sizeof(uint64_t),
129 QCOW2_DISCARD_OTHER
);
132 qemu_vfree(new_l1_table
);
133 qcow2_free_clusters(bs
, new_l1_table_offset
, new_l1_size2
,
134 QCOW2_DISCARD_OTHER
);
141 * Loads a L2 table into memory. If the table is in the cache, the cache
142 * is used; otherwise the L2 table is loaded from the image file.
144 * Returns a pointer to the L2 table on success, or NULL if the read from
145 * the image file failed.
148 static int l2_load(BlockDriverState
*bs
, uint64_t l2_offset
,
151 BDRVQcowState
*s
= bs
->opaque
;
154 ret
= qcow2_cache_get(bs
, s
->l2_table_cache
, l2_offset
, (void**) l2_table
);
160 * Writes one sector of the L1 table to the disk (can't update single entries
161 * and we really don't want bdrv_pread to perform a read-modify-write)
163 #define L1_ENTRIES_PER_SECTOR (512 / 8)
164 int qcow2_write_l1_entry(BlockDriverState
*bs
, int l1_index
)
166 BDRVQcowState
*s
= bs
->opaque
;
167 uint64_t buf
[L1_ENTRIES_PER_SECTOR
] = { 0 };
171 l1_start_index
= l1_index
& ~(L1_ENTRIES_PER_SECTOR
- 1);
172 for (i
= 0; i
< L1_ENTRIES_PER_SECTOR
&& l1_start_index
+ i
< s
->l1_size
;
175 buf
[i
] = cpu_to_be64(s
->l1_table
[l1_start_index
+ i
]);
178 ret
= qcow2_pre_write_overlap_check(bs
, QCOW2_OL_ACTIVE_L1
,
179 s
->l1_table_offset
+ 8 * l1_start_index
, sizeof(buf
));
184 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
185 ret
= bdrv_pwrite_sync(bs
->file
, s
->l1_table_offset
+ 8 * l1_start_index
,
197 * Allocate a new l2 entry in the file. If l1_index points to an already
198 * used entry in the L2 table (i.e. we are doing a copy on write for the L2
199 * table) copy the contents of the old L2 table into the newly allocated one.
200 * Otherwise the new table is initialized with zeros.
204 static int l2_allocate(BlockDriverState
*bs
, int l1_index
, uint64_t **table
)
206 BDRVQcowState
*s
= bs
->opaque
;
207 uint64_t old_l2_offset
;
208 uint64_t *l2_table
= NULL
;
212 old_l2_offset
= s
->l1_table
[l1_index
];
214 trace_qcow2_l2_allocate(bs
, l1_index
);
216 /* allocate a new l2 entry */
218 l2_offset
= qcow2_alloc_clusters(bs
, s
->l2_size
* sizeof(uint64_t));
224 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
229 /* allocate a new entry in the l2 cache */
231 trace_qcow2_l2_allocate_get_empty(bs
, l1_index
);
232 ret
= qcow2_cache_get_empty(bs
, s
->l2_table_cache
, l2_offset
, (void**) table
);
239 if ((old_l2_offset
& L1E_OFFSET_MASK
) == 0) {
240 /* if there was no old l2 table, clear the new table */
241 memset(l2_table
, 0, s
->l2_size
* sizeof(uint64_t));
245 /* if there was an old l2 table, read it from the disk */
246 BLKDBG_EVENT(bs
->file
, BLKDBG_L2_ALLOC_COW_READ
);
247 ret
= qcow2_cache_get(bs
, s
->l2_table_cache
,
248 old_l2_offset
& L1E_OFFSET_MASK
,
249 (void**) &old_table
);
254 memcpy(l2_table
, old_table
, s
->cluster_size
);
256 qcow2_cache_put(bs
, s
->l2_table_cache
, (void **) &old_table
);
259 /* write the l2 table to the file */
260 BLKDBG_EVENT(bs
->file
, BLKDBG_L2_ALLOC_WRITE
);
262 trace_qcow2_l2_allocate_write_l2(bs
, l1_index
);
263 qcow2_cache_entry_mark_dirty(bs
, s
->l2_table_cache
, l2_table
);
264 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
269 /* update the L1 entry */
270 trace_qcow2_l2_allocate_write_l1(bs
, l1_index
);
271 s
->l1_table
[l1_index
] = l2_offset
| QCOW_OFLAG_COPIED
;
272 ret
= qcow2_write_l1_entry(bs
, l1_index
);
278 trace_qcow2_l2_allocate_done(bs
, l1_index
, 0);
282 trace_qcow2_l2_allocate_done(bs
, l1_index
, ret
);
283 if (l2_table
!= NULL
) {
284 qcow2_cache_put(bs
, s
->l2_table_cache
, (void**) table
);
286 s
->l1_table
[l1_index
] = old_l2_offset
;
288 qcow2_free_clusters(bs
, l2_offset
, s
->l2_size
* sizeof(uint64_t),
289 QCOW2_DISCARD_ALWAYS
);
295 * Checks how many clusters in a given L2 table are contiguous in the image
296 * file. As soon as one of the flags in the bitmask stop_flags changes compared
297 * to the first cluster, the search is stopped and the cluster is not counted
298 * as contiguous. (This allows it, for example, to stop at the first compressed
299 * cluster which may require a different handling)
301 static int count_contiguous_clusters(uint64_t nb_clusters
, int cluster_size
,
302 uint64_t *l2_table
, uint64_t stop_flags
)
305 uint64_t mask
= stop_flags
| L2E_OFFSET_MASK
| QCOW_OFLAG_COMPRESSED
;
306 uint64_t first_entry
= be64_to_cpu(l2_table
[0]);
307 uint64_t offset
= first_entry
& mask
;
312 assert(qcow2_get_cluster_type(first_entry
) != QCOW2_CLUSTER_COMPRESSED
);
314 for (i
= 0; i
< nb_clusters
; i
++) {
315 uint64_t l2_entry
= be64_to_cpu(l2_table
[i
]) & mask
;
316 if (offset
+ (uint64_t) i
* cluster_size
!= l2_entry
) {
324 static int count_contiguous_free_clusters(uint64_t nb_clusters
, uint64_t *l2_table
)
328 for (i
= 0; i
< nb_clusters
; i
++) {
329 int type
= qcow2_get_cluster_type(be64_to_cpu(l2_table
[i
]));
331 if (type
!= QCOW2_CLUSTER_UNALLOCATED
) {
339 /* The crypt function is compatible with the linux cryptoloop
340 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
342 void qcow2_encrypt_sectors(BDRVQcowState
*s
, int64_t sector_num
,
343 uint8_t *out_buf
, const uint8_t *in_buf
,
344 int nb_sectors
, int enc
,
353 for(i
= 0; i
< nb_sectors
; i
++) {
354 ivec
.ll
[0] = cpu_to_le64(sector_num
);
356 AES_cbc_encrypt(in_buf
, out_buf
, 512, key
,
364 static int coroutine_fn
copy_sectors(BlockDriverState
*bs
,
366 uint64_t cluster_offset
,
367 int n_start
, int n_end
)
369 BDRVQcowState
*s
= bs
->opaque
;
379 iov
.iov_len
= n
* BDRV_SECTOR_SIZE
;
380 iov
.iov_base
= qemu_try_blockalign(bs
, iov
.iov_len
);
381 if (iov
.iov_base
== NULL
) {
385 qemu_iovec_init_external(&qiov
, &iov
, 1);
387 BLKDBG_EVENT(bs
->file
, BLKDBG_COW_READ
);
394 /* Call .bdrv_co_readv() directly instead of using the public block-layer
395 * interface. This avoids double I/O throttling and request tracking,
396 * which can lead to deadlock when block layer copy-on-read is enabled.
398 ret
= bs
->drv
->bdrv_co_readv(bs
, start_sect
+ n_start
, n
, &qiov
);
403 if (s
->crypt_method
) {
404 qcow2_encrypt_sectors(s
, start_sect
+ n_start
,
405 iov
.iov_base
, iov
.iov_base
, n
, 1,
406 &s
->aes_encrypt_key
);
409 ret
= qcow2_pre_write_overlap_check(bs
, 0,
410 cluster_offset
+ n_start
* BDRV_SECTOR_SIZE
, n
* BDRV_SECTOR_SIZE
);
415 BLKDBG_EVENT(bs
->file
, BLKDBG_COW_WRITE
);
416 ret
= bdrv_co_writev(bs
->file
, (cluster_offset
>> 9) + n_start
, n
, &qiov
);
423 qemu_vfree(iov
.iov_base
);
431 * For a given offset of the disk image, find the cluster offset in
432 * qcow2 file. The offset is stored in *cluster_offset.
434 * on entry, *num is the number of contiguous sectors we'd like to
435 * access following offset.
437 * on exit, *num is the number of contiguous sectors we can read.
439 * Returns the cluster type (QCOW2_CLUSTER_*) on success, -errno in error
442 int qcow2_get_cluster_offset(BlockDriverState
*bs
, uint64_t offset
,
443 int *num
, uint64_t *cluster_offset
)
445 BDRVQcowState
*s
= bs
->opaque
;
446 unsigned int l2_index
;
447 uint64_t l1_index
, l2_offset
, *l2_table
;
449 unsigned int index_in_cluster
, nb_clusters
;
450 uint64_t nb_available
, nb_needed
;
453 index_in_cluster
= (offset
>> 9) & (s
->cluster_sectors
- 1);
454 nb_needed
= *num
+ index_in_cluster
;
456 l1_bits
= s
->l2_bits
+ s
->cluster_bits
;
458 /* compute how many bytes there are between the offset and
459 * the end of the l1 entry
462 nb_available
= (1ULL << l1_bits
) - (offset
& ((1ULL << l1_bits
) - 1));
464 /* compute the number of available sectors */
466 nb_available
= (nb_available
>> 9) + index_in_cluster
;
468 if (nb_needed
> nb_available
) {
469 nb_needed
= nb_available
;
474 /* seek the the l2 offset in the l1 table */
476 l1_index
= offset
>> l1_bits
;
477 if (l1_index
>= s
->l1_size
) {
478 ret
= QCOW2_CLUSTER_UNALLOCATED
;
482 l2_offset
= s
->l1_table
[l1_index
] & L1E_OFFSET_MASK
;
484 ret
= QCOW2_CLUSTER_UNALLOCATED
;
488 if (offset_into_cluster(s
, l2_offset
)) {
489 qcow2_signal_corruption(bs
, true, -1, -1, "L2 table offset %#" PRIx64
490 " unaligned (L1 index: %#" PRIx64
")",
491 l2_offset
, l1_index
);
495 /* load the l2 table in memory */
497 ret
= l2_load(bs
, l2_offset
, &l2_table
);
502 /* find the cluster offset for the given disk offset */
504 l2_index
= (offset
>> s
->cluster_bits
) & (s
->l2_size
- 1);
505 *cluster_offset
= be64_to_cpu(l2_table
[l2_index
]);
506 nb_clusters
= size_to_clusters(s
, nb_needed
<< 9);
508 ret
= qcow2_get_cluster_type(*cluster_offset
);
510 case QCOW2_CLUSTER_COMPRESSED
:
511 /* Compressed clusters can only be processed one by one */
513 *cluster_offset
&= L2E_COMPRESSED_OFFSET_SIZE_MASK
;
515 case QCOW2_CLUSTER_ZERO
:
516 if (s
->qcow_version
< 3) {
517 qcow2_signal_corruption(bs
, true, -1, -1, "Zero cluster entry found"
518 " in pre-v3 image (L2 offset: %#" PRIx64
519 ", L2 index: %#x)", l2_offset
, l2_index
);
523 c
= count_contiguous_clusters(nb_clusters
, s
->cluster_size
,
524 &l2_table
[l2_index
], QCOW_OFLAG_ZERO
);
527 case QCOW2_CLUSTER_UNALLOCATED
:
528 /* how many empty clusters ? */
529 c
= count_contiguous_free_clusters(nb_clusters
, &l2_table
[l2_index
]);
532 case QCOW2_CLUSTER_NORMAL
:
533 /* how many allocated clusters ? */
534 c
= count_contiguous_clusters(nb_clusters
, s
->cluster_size
,
535 &l2_table
[l2_index
], QCOW_OFLAG_ZERO
);
536 *cluster_offset
&= L2E_OFFSET_MASK
;
537 if (offset_into_cluster(s
, *cluster_offset
)) {
538 qcow2_signal_corruption(bs
, true, -1, -1, "Data cluster offset %#"
539 PRIx64
" unaligned (L2 offset: %#" PRIx64
540 ", L2 index: %#x)", *cluster_offset
,
541 l2_offset
, l2_index
);
550 qcow2_cache_put(bs
, s
->l2_table_cache
, (void**) &l2_table
);
552 nb_available
= (c
* s
->cluster_sectors
);
555 if (nb_available
> nb_needed
)
556 nb_available
= nb_needed
;
558 *num
= nb_available
- index_in_cluster
;
563 qcow2_cache_put(bs
, s
->l2_table_cache
, (void **)&l2_table
);
570 * for a given disk offset, load (and allocate if needed)
573 * the l2 table offset in the qcow2 file and the cluster index
574 * in the l2 table are given to the caller.
576 * Returns 0 on success, -errno in failure case
578 static int get_cluster_table(BlockDriverState
*bs
, uint64_t offset
,
579 uint64_t **new_l2_table
,
582 BDRVQcowState
*s
= bs
->opaque
;
583 unsigned int l2_index
;
584 uint64_t l1_index
, l2_offset
;
585 uint64_t *l2_table
= NULL
;
588 /* seek the the l2 offset in the l1 table */
590 l1_index
= offset
>> (s
->l2_bits
+ s
->cluster_bits
);
591 if (l1_index
>= s
->l1_size
) {
592 ret
= qcow2_grow_l1_table(bs
, l1_index
+ 1, false);
598 assert(l1_index
< s
->l1_size
);
599 l2_offset
= s
->l1_table
[l1_index
] & L1E_OFFSET_MASK
;
600 if (offset_into_cluster(s
, l2_offset
)) {
601 qcow2_signal_corruption(bs
, true, -1, -1, "L2 table offset %#" PRIx64
602 " unaligned (L1 index: %#" PRIx64
")",
603 l2_offset
, l1_index
);
607 /* seek the l2 table of the given l2 offset */
609 if (s
->l1_table
[l1_index
] & QCOW_OFLAG_COPIED
) {
610 /* load the l2 table in memory */
611 ret
= l2_load(bs
, l2_offset
, &l2_table
);
616 /* First allocate a new L2 table (and do COW if needed) */
617 ret
= l2_allocate(bs
, l1_index
, &l2_table
);
622 /* Then decrease the refcount of the old table */
624 qcow2_free_clusters(bs
, l2_offset
, s
->l2_size
* sizeof(uint64_t),
625 QCOW2_DISCARD_OTHER
);
629 /* find the cluster offset for the given disk offset */
631 l2_index
= (offset
>> s
->cluster_bits
) & (s
->l2_size
- 1);
633 *new_l2_table
= l2_table
;
634 *new_l2_index
= l2_index
;
640 * alloc_compressed_cluster_offset
642 * For a given offset of the disk image, return cluster offset in
645 * If the offset is not found, allocate a new compressed cluster.
647 * Return the cluster offset if successful,
648 * Return 0, otherwise.
652 uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState
*bs
,
656 BDRVQcowState
*s
= bs
->opaque
;
659 int64_t cluster_offset
;
662 ret
= get_cluster_table(bs
, offset
, &l2_table
, &l2_index
);
667 /* Compression can't overwrite anything. Fail if the cluster was already
669 cluster_offset
= be64_to_cpu(l2_table
[l2_index
]);
670 if (cluster_offset
& L2E_OFFSET_MASK
) {
671 qcow2_cache_put(bs
, s
->l2_table_cache
, (void**) &l2_table
);
675 cluster_offset
= qcow2_alloc_bytes(bs
, compressed_size
);
676 if (cluster_offset
< 0) {
677 qcow2_cache_put(bs
, s
->l2_table_cache
, (void**) &l2_table
);
681 nb_csectors
= ((cluster_offset
+ compressed_size
- 1) >> 9) -
682 (cluster_offset
>> 9);
684 cluster_offset
|= QCOW_OFLAG_COMPRESSED
|
685 ((uint64_t)nb_csectors
<< s
->csize_shift
);
687 /* update L2 table */
689 /* compressed clusters never have the copied flag */
691 BLKDBG_EVENT(bs
->file
, BLKDBG_L2_UPDATE_COMPRESSED
);
692 qcow2_cache_entry_mark_dirty(bs
, s
->l2_table_cache
, l2_table
);
693 l2_table
[l2_index
] = cpu_to_be64(cluster_offset
);
694 qcow2_cache_put(bs
, s
->l2_table_cache
, (void **) &l2_table
);
696 return cluster_offset
;
699 static int perform_cow(BlockDriverState
*bs
, QCowL2Meta
*m
, Qcow2COWRegion
*r
)
701 BDRVQcowState
*s
= bs
->opaque
;
704 if (r
->nb_sectors
== 0) {
708 qemu_co_mutex_unlock(&s
->lock
);
709 ret
= copy_sectors(bs
, m
->offset
/ BDRV_SECTOR_SIZE
, m
->alloc_offset
,
710 r
->offset
/ BDRV_SECTOR_SIZE
,
711 r
->offset
/ BDRV_SECTOR_SIZE
+ r
->nb_sectors
);
712 qemu_co_mutex_lock(&s
->lock
);
719 * Before we update the L2 table to actually point to the new cluster, we
720 * need to be sure that the refcounts have been increased and COW was
723 qcow2_cache_depends_on_flush(s
->l2_table_cache
);
728 int qcow2_alloc_cluster_link_l2(BlockDriverState
*bs
, QCowL2Meta
*m
)
730 BDRVQcowState
*s
= bs
->opaque
;
731 int i
, j
= 0, l2_index
, ret
;
732 uint64_t *old_cluster
, *l2_table
;
733 uint64_t cluster_offset
= m
->alloc_offset
;
735 trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m
->nb_clusters
);
736 assert(m
->nb_clusters
> 0);
738 old_cluster
= g_try_new(uint64_t, m
->nb_clusters
);
739 if (old_cluster
== NULL
) {
744 /* copy content of unmodified sectors */
745 ret
= perform_cow(bs
, m
, &m
->cow_start
);
750 ret
= perform_cow(bs
, m
, &m
->cow_end
);
755 /* Update L2 table. */
756 if (s
->use_lazy_refcounts
) {
757 qcow2_mark_dirty(bs
);
759 if (qcow2_need_accurate_refcounts(s
)) {
760 qcow2_cache_set_dependency(bs
, s
->l2_table_cache
,
761 s
->refcount_block_cache
);
764 ret
= get_cluster_table(bs
, m
->offset
, &l2_table
, &l2_index
);
768 qcow2_cache_entry_mark_dirty(bs
, s
->l2_table_cache
, l2_table
);
770 assert(l2_index
+ m
->nb_clusters
<= s
->l2_size
);
771 for (i
= 0; i
< m
->nb_clusters
; i
++) {
772 /* if two concurrent writes happen to the same unallocated cluster
773 * each write allocates separate cluster and writes data concurrently.
774 * The first one to complete updates l2 table with pointer to its
775 * cluster the second one has to do RMW (which is done above by
776 * copy_sectors()), update l2 table with its cluster pointer and free
777 * old cluster. This is what this loop does */
778 if(l2_table
[l2_index
+ i
] != 0)
779 old_cluster
[j
++] = l2_table
[l2_index
+ i
];
781 l2_table
[l2_index
+ i
] = cpu_to_be64((cluster_offset
+
782 (i
<< s
->cluster_bits
)) | QCOW_OFLAG_COPIED
);
786 qcow2_cache_put(bs
, s
->l2_table_cache
, (void **) &l2_table
);
789 * If this was a COW, we need to decrease the refcount of the old cluster.
790 * Also flush bs->file to get the right order for L2 and refcount update.
792 * Don't discard clusters that reach a refcount of 0 (e.g. compressed
793 * clusters), the next write will reuse them anyway.
796 for (i
= 0; i
< j
; i
++) {
797 qcow2_free_any_clusters(bs
, be64_to_cpu(old_cluster
[i
]), 1,
798 QCOW2_DISCARD_NEVER
);
809 * Returns the number of contiguous clusters that can be used for an allocating
810 * write, but require COW to be performed (this includes yet unallocated space,
811 * which must copy from the backing file)
813 static int count_cow_clusters(BDRVQcowState
*s
, int nb_clusters
,
814 uint64_t *l2_table
, int l2_index
)
818 for (i
= 0; i
< nb_clusters
; i
++) {
819 uint64_t l2_entry
= be64_to_cpu(l2_table
[l2_index
+ i
]);
820 int cluster_type
= qcow2_get_cluster_type(l2_entry
);
822 switch(cluster_type
) {
823 case QCOW2_CLUSTER_NORMAL
:
824 if (l2_entry
& QCOW_OFLAG_COPIED
) {
828 case QCOW2_CLUSTER_UNALLOCATED
:
829 case QCOW2_CLUSTER_COMPRESSED
:
830 case QCOW2_CLUSTER_ZERO
:
838 assert(i
<= nb_clusters
);
843 * Check if there already is an AIO write request in flight which allocates
844 * the same cluster. In this case we need to wait until the previous
845 * request has completed and updated the L2 table accordingly.
848 * 0 if there was no dependency. *cur_bytes indicates the number of
849 * bytes from guest_offset that can be read before the next
850 * dependency must be processed (or the request is complete)
852 * -EAGAIN if we had to wait for another request, previously gathered
853 * information on cluster allocation may be invalid now. The caller
854 * must start over anyway, so consider *cur_bytes undefined.
856 static int handle_dependencies(BlockDriverState
*bs
, uint64_t guest_offset
,
857 uint64_t *cur_bytes
, QCowL2Meta
**m
)
859 BDRVQcowState
*s
= bs
->opaque
;
860 QCowL2Meta
*old_alloc
;
861 uint64_t bytes
= *cur_bytes
;
863 QLIST_FOREACH(old_alloc
, &s
->cluster_allocs
, next_in_flight
) {
865 uint64_t start
= guest_offset
;
866 uint64_t end
= start
+ bytes
;
867 uint64_t old_start
= l2meta_cow_start(old_alloc
);
868 uint64_t old_end
= l2meta_cow_end(old_alloc
);
870 if (end
<= old_start
|| start
>= old_end
) {
871 /* No intersection */
873 if (start
< old_start
) {
874 /* Stop at the start of a running allocation */
875 bytes
= old_start
- start
;
880 /* Stop if already an l2meta exists. After yielding, it wouldn't
881 * be valid any more, so we'd have to clean up the old L2Metas
882 * and deal with requests depending on them before starting to
883 * gather new ones. Not worth the trouble. */
884 if (bytes
== 0 && *m
) {
890 /* Wait for the dependency to complete. We need to recheck
891 * the free/allocated clusters when we continue. */
892 qemu_co_mutex_unlock(&s
->lock
);
893 qemu_co_queue_wait(&old_alloc
->dependent_requests
);
894 qemu_co_mutex_lock(&s
->lock
);
900 /* Make sure that existing clusters and new allocations are only used up to
901 * the next dependency if we shortened the request above */
908 * Checks how many already allocated clusters that don't require a copy on
909 * write there are at the given guest_offset (up to *bytes). If
910 * *host_offset is not zero, only physically contiguous clusters beginning at
911 * this host offset are counted.
913 * Note that guest_offset may not be cluster aligned. In this case, the
914 * returned *host_offset points to exact byte referenced by guest_offset and
915 * therefore isn't cluster aligned as well.
918 * 0: if no allocated clusters are available at the given offset.
919 * *bytes is normally unchanged. It is set to 0 if the cluster
920 * is allocated and doesn't need COW, but doesn't have the right
923 * 1: if allocated clusters that don't require a COW are available at
924 * the requested offset. *bytes may have decreased and describes
925 * the length of the area that can be written to.
927 * -errno: in error cases
929 static int handle_copied(BlockDriverState
*bs
, uint64_t guest_offset
,
930 uint64_t *host_offset
, uint64_t *bytes
, QCowL2Meta
**m
)
932 BDRVQcowState
*s
= bs
->opaque
;
934 uint64_t cluster_offset
;
936 unsigned int nb_clusters
;
937 unsigned int keep_clusters
;
940 trace_qcow2_handle_copied(qemu_coroutine_self(), guest_offset
, *host_offset
,
943 assert(*host_offset
== 0 || offset_into_cluster(s
, guest_offset
)
944 == offset_into_cluster(s
, *host_offset
));
947 * Calculate the number of clusters to look for. We stop at L2 table
948 * boundaries to keep things simple.
951 size_to_clusters(s
, offset_into_cluster(s
, guest_offset
) + *bytes
);
953 l2_index
= offset_to_l2_index(s
, guest_offset
);
954 nb_clusters
= MIN(nb_clusters
, s
->l2_size
- l2_index
);
956 /* Find L2 entry for the first involved cluster */
957 ret
= get_cluster_table(bs
, guest_offset
, &l2_table
, &l2_index
);
962 cluster_offset
= be64_to_cpu(l2_table
[l2_index
]);
964 /* Check how many clusters are already allocated and don't need COW */
965 if (qcow2_get_cluster_type(cluster_offset
) == QCOW2_CLUSTER_NORMAL
966 && (cluster_offset
& QCOW_OFLAG_COPIED
))
968 /* If a specific host_offset is required, check it */
969 bool offset_matches
=
970 (cluster_offset
& L2E_OFFSET_MASK
) == *host_offset
;
972 if (offset_into_cluster(s
, cluster_offset
& L2E_OFFSET_MASK
)) {
973 qcow2_signal_corruption(bs
, true, -1, -1, "Data cluster offset "
974 "%#llx unaligned (guest offset: %#" PRIx64
975 ")", cluster_offset
& L2E_OFFSET_MASK
,
981 if (*host_offset
!= 0 && !offset_matches
) {
987 /* We keep all QCOW_OFLAG_COPIED clusters */
989 count_contiguous_clusters(nb_clusters
, s
->cluster_size
,
991 QCOW_OFLAG_COPIED
| QCOW_OFLAG_ZERO
);
992 assert(keep_clusters
<= nb_clusters
);
995 keep_clusters
* s
->cluster_size
996 - offset_into_cluster(s
, guest_offset
));
1005 qcow2_cache_put(bs
, s
->l2_table_cache
, (void **) &l2_table
);
1007 /* Only return a host offset if we actually made progress. Otherwise we
1008 * would make requirements for handle_alloc() that it can't fulfill */
1010 *host_offset
= (cluster_offset
& L2E_OFFSET_MASK
)
1011 + offset_into_cluster(s
, guest_offset
);
1018 * Allocates new clusters for the given guest_offset.
1020 * At most *nb_clusters are allocated, and on return *nb_clusters is updated to
1021 * contain the number of clusters that have been allocated and are contiguous
1022 * in the image file.
1024 * If *host_offset is non-zero, it specifies the offset in the image file at
1025 * which the new clusters must start. *nb_clusters can be 0 on return in this
1026 * case if the cluster at host_offset is already in use. If *host_offset is
1027 * zero, the clusters can be allocated anywhere in the image file.
1029 * *host_offset is updated to contain the offset into the image file at which
1030 * the first allocated cluster starts.
1032 * Return 0 on success and -errno in error cases. -EAGAIN means that the
1033 * function has been waiting for another request and the allocation must be
1034 * restarted, but the whole request should not be failed.
1036 static int do_alloc_cluster_offset(BlockDriverState
*bs
, uint64_t guest_offset
,
1037 uint64_t *host_offset
, unsigned int *nb_clusters
)
1039 BDRVQcowState
*s
= bs
->opaque
;
1041 trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset
,
1042 *host_offset
, *nb_clusters
);
1044 /* Allocate new clusters */
1045 trace_qcow2_cluster_alloc_phys(qemu_coroutine_self());
1046 if (*host_offset
== 0) {
1047 int64_t cluster_offset
=
1048 qcow2_alloc_clusters(bs
, *nb_clusters
* s
->cluster_size
);
1049 if (cluster_offset
< 0) {
1050 return cluster_offset
;
1052 *host_offset
= cluster_offset
;
1055 int ret
= qcow2_alloc_clusters_at(bs
, *host_offset
, *nb_clusters
);
1065 * Allocates new clusters for an area that either is yet unallocated or needs a
1066 * copy on write. If *host_offset is non-zero, clusters are only allocated if
1067 * the new allocation can match the specified host offset.
1069 * Note that guest_offset may not be cluster aligned. In this case, the
1070 * returned *host_offset points to exact byte referenced by guest_offset and
1071 * therefore isn't cluster aligned as well.
1074 * 0: if no clusters could be allocated. *bytes is set to 0,
1075 * *host_offset is left unchanged.
1077 * 1: if new clusters were allocated. *bytes may be decreased if the
1078 * new allocation doesn't cover all of the requested area.
1079 * *host_offset is updated to contain the host offset of the first
1080 * newly allocated cluster.
1082 * -errno: in error cases
1084 static int handle_alloc(BlockDriverState
*bs
, uint64_t guest_offset
,
1085 uint64_t *host_offset
, uint64_t *bytes
, QCowL2Meta
**m
)
1087 BDRVQcowState
*s
= bs
->opaque
;
1091 unsigned int nb_clusters
;
1094 uint64_t alloc_cluster_offset
;
1096 trace_qcow2_handle_alloc(qemu_coroutine_self(), guest_offset
, *host_offset
,
1101 * Calculate the number of clusters to look for. We stop at L2 table
1102 * boundaries to keep things simple.
1105 size_to_clusters(s
, offset_into_cluster(s
, guest_offset
) + *bytes
);
1107 l2_index
= offset_to_l2_index(s
, guest_offset
);
1108 nb_clusters
= MIN(nb_clusters
, s
->l2_size
- l2_index
);
1110 /* Find L2 entry for the first involved cluster */
1111 ret
= get_cluster_table(bs
, guest_offset
, &l2_table
, &l2_index
);
1116 entry
= be64_to_cpu(l2_table
[l2_index
]);
1118 /* For the moment, overwrite compressed clusters one by one */
1119 if (entry
& QCOW_OFLAG_COMPRESSED
) {
1122 nb_clusters
= count_cow_clusters(s
, nb_clusters
, l2_table
, l2_index
);
1125 /* This function is only called when there were no non-COW clusters, so if
1126 * we can't find any unallocated or COW clusters either, something is
1127 * wrong with our code. */
1128 assert(nb_clusters
> 0);
1130 qcow2_cache_put(bs
, s
->l2_table_cache
, (void **) &l2_table
);
1132 /* Allocate, if necessary at a given offset in the image file */
1133 alloc_cluster_offset
= start_of_cluster(s
, *host_offset
);
1134 ret
= do_alloc_cluster_offset(bs
, guest_offset
, &alloc_cluster_offset
,
1140 /* Can't extend contiguous allocation */
1141 if (nb_clusters
== 0) {
1146 /* !*host_offset would overwrite the image header and is reserved for "no
1147 * host offset preferred". If 0 was a valid host offset, it'd trigger the
1148 * following overlap check; do that now to avoid having an invalid value in
1150 if (!alloc_cluster_offset
) {
1151 ret
= qcow2_pre_write_overlap_check(bs
, 0, alloc_cluster_offset
,
1152 nb_clusters
* s
->cluster_size
);
1158 * Save info needed for meta data update.
1160 * requested_sectors: Number of sectors from the start of the first
1161 * newly allocated cluster to the end of the (possibly shortened
1162 * before) write request.
1164 * avail_sectors: Number of sectors from the start of the first
1165 * newly allocated to the end of the last newly allocated cluster.
1167 * nb_sectors: The number of sectors from the start of the first
1168 * newly allocated cluster to the end of the area that the write
1169 * request actually writes to (excluding COW at the end)
1171 int requested_sectors
=
1172 (*bytes
+ offset_into_cluster(s
, guest_offset
))
1173 >> BDRV_SECTOR_BITS
;
1174 int avail_sectors
= nb_clusters
1175 << (s
->cluster_bits
- BDRV_SECTOR_BITS
);
1176 int alloc_n_start
= offset_into_cluster(s
, guest_offset
)
1177 >> BDRV_SECTOR_BITS
;
1178 int nb_sectors
= MIN(requested_sectors
, avail_sectors
);
1179 QCowL2Meta
*old_m
= *m
;
1181 *m
= g_malloc0(sizeof(**m
));
1183 **m
= (QCowL2Meta
) {
1186 .alloc_offset
= alloc_cluster_offset
,
1187 .offset
= start_of_cluster(s
, guest_offset
),
1188 .nb_clusters
= nb_clusters
,
1189 .nb_available
= nb_sectors
,
1193 .nb_sectors
= alloc_n_start
,
1196 .offset
= nb_sectors
* BDRV_SECTOR_SIZE
,
1197 .nb_sectors
= avail_sectors
- nb_sectors
,
1200 qemu_co_queue_init(&(*m
)->dependent_requests
);
1201 QLIST_INSERT_HEAD(&s
->cluster_allocs
, *m
, next_in_flight
);
1203 *host_offset
= alloc_cluster_offset
+ offset_into_cluster(s
, guest_offset
);
1204 *bytes
= MIN(*bytes
, (nb_sectors
* BDRV_SECTOR_SIZE
)
1205 - offset_into_cluster(s
, guest_offset
));
1206 assert(*bytes
!= 0);
1211 if (*m
&& (*m
)->nb_clusters
> 0) {
1212 QLIST_REMOVE(*m
, next_in_flight
);
1218 * alloc_cluster_offset
1220 * For a given offset on the virtual disk, find the cluster offset in qcow2
1221 * file. If the offset is not found, allocate a new cluster.
1223 * If the cluster was already allocated, m->nb_clusters is set to 0 and
1224 * other fields in m are meaningless.
1226 * If the cluster is newly allocated, m->nb_clusters is set to the number of
1227 * contiguous clusters that have been allocated. In this case, the other
1228 * fields of m are valid and contain information about the first allocated
1231 * If the request conflicts with another write request in flight, the coroutine
1232 * is queued and will be reentered when the dependency has completed.
1234 * Return 0 on success and -errno in error cases
1236 int qcow2_alloc_cluster_offset(BlockDriverState
*bs
, uint64_t offset
,
1237 int *num
, uint64_t *host_offset
, QCowL2Meta
**m
)
1239 BDRVQcowState
*s
= bs
->opaque
;
1240 uint64_t start
, remaining
;
1241 uint64_t cluster_offset
;
1245 trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset
, *num
);
1247 assert((offset
& ~BDRV_SECTOR_MASK
) == 0);
1251 remaining
= (uint64_t)*num
<< BDRV_SECTOR_BITS
;
1259 if (!*host_offset
) {
1260 *host_offset
= start_of_cluster(s
, cluster_offset
);
1263 assert(remaining
>= cur_bytes
);
1266 remaining
-= cur_bytes
;
1267 cluster_offset
+= cur_bytes
;
1269 if (remaining
== 0) {
1273 cur_bytes
= remaining
;
1276 * Now start gathering as many contiguous clusters as possible:
1278 * 1. Check for overlaps with in-flight allocations
1280 * a) Overlap not in the first cluster -> shorten this request and
1281 * let the caller handle the rest in its next loop iteration.
1283 * b) Real overlaps of two requests. Yield and restart the search
1284 * for contiguous clusters (the situation could have changed
1285 * while we were sleeping)
1287 * c) TODO: Request starts in the same cluster as the in-flight
1288 * allocation ends. Shorten the COW of the in-fight allocation,
1289 * set cluster_offset to write to the same cluster and set up
1290 * the right synchronisation between the in-flight request and
1293 ret
= handle_dependencies(bs
, start
, &cur_bytes
, m
);
1294 if (ret
== -EAGAIN
) {
1295 /* Currently handle_dependencies() doesn't yield if we already had
1296 * an allocation. If it did, we would have to clean up the L2Meta
1297 * structs before starting over. */
1300 } else if (ret
< 0) {
1302 } else if (cur_bytes
== 0) {
1305 /* handle_dependencies() may have decreased cur_bytes (shortened
1306 * the allocations below) so that the next dependency is processed
1307 * correctly during the next loop iteration. */
1311 * 2. Count contiguous COPIED clusters.
1313 ret
= handle_copied(bs
, start
, &cluster_offset
, &cur_bytes
, m
);
1318 } else if (cur_bytes
== 0) {
1323 * 3. If the request still hasn't completed, allocate new clusters,
1324 * considering any cluster_offset of steps 1c or 2.
1326 ret
= handle_alloc(bs
, start
, &cluster_offset
, &cur_bytes
, m
);
1332 assert(cur_bytes
== 0);
1337 *num
-= remaining
>> BDRV_SECTOR_BITS
;
1339 assert(*host_offset
!= 0);
1344 static int decompress_buffer(uint8_t *out_buf
, int out_buf_size
,
1345 const uint8_t *buf
, int buf_size
)
1347 z_stream strm1
, *strm
= &strm1
;
1350 memset(strm
, 0, sizeof(*strm
));
1352 strm
->next_in
= (uint8_t *)buf
;
1353 strm
->avail_in
= buf_size
;
1354 strm
->next_out
= out_buf
;
1355 strm
->avail_out
= out_buf_size
;
1357 ret
= inflateInit2(strm
, -12);
1360 ret
= inflate(strm
, Z_FINISH
);
1361 out_len
= strm
->next_out
- out_buf
;
1362 if ((ret
!= Z_STREAM_END
&& ret
!= Z_BUF_ERROR
) ||
1363 out_len
!= out_buf_size
) {
1371 int qcow2_decompress_cluster(BlockDriverState
*bs
, uint64_t cluster_offset
)
1373 BDRVQcowState
*s
= bs
->opaque
;
1374 int ret
, csize
, nb_csectors
, sector_offset
;
1377 coffset
= cluster_offset
& s
->cluster_offset_mask
;
1378 if (s
->cluster_cache_offset
!= coffset
) {
1379 nb_csectors
= ((cluster_offset
>> s
->csize_shift
) & s
->csize_mask
) + 1;
1380 sector_offset
= coffset
& 511;
1381 csize
= nb_csectors
* 512 - sector_offset
;
1382 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_COMPRESSED
);
1383 ret
= bdrv_read(bs
->file
, coffset
>> 9, s
->cluster_data
, nb_csectors
);
1387 if (decompress_buffer(s
->cluster_cache
, s
->cluster_size
,
1388 s
->cluster_data
+ sector_offset
, csize
) < 0) {
1391 s
->cluster_cache_offset
= coffset
;
1397 * This discards as many clusters of nb_clusters as possible at once (i.e.
1398 * all clusters in the same L2 table) and returns the number of discarded
1401 static int discard_single_l2(BlockDriverState
*bs
, uint64_t offset
,
1402 unsigned int nb_clusters
, enum qcow2_discard_type type
, bool full_discard
)
1404 BDRVQcowState
*s
= bs
->opaque
;
1410 ret
= get_cluster_table(bs
, offset
, &l2_table
, &l2_index
);
1415 /* Limit nb_clusters to one L2 table */
1416 nb_clusters
= MIN(nb_clusters
, s
->l2_size
- l2_index
);
1418 for (i
= 0; i
< nb_clusters
; i
++) {
1419 uint64_t old_l2_entry
;
1421 old_l2_entry
= be64_to_cpu(l2_table
[l2_index
+ i
]);
1424 * If full_discard is false, make sure that a discarded area reads back
1425 * as zeroes for v3 images (we cannot do it for v2 without actually
1426 * writing a zero-filled buffer). We can skip the operation if the
1427 * cluster is already marked as zero, or if it's unallocated and we
1428 * don't have a backing file.
1430 * TODO We might want to use bdrv_get_block_status(bs) here, but we're
1431 * holding s->lock, so that doesn't work today.
1433 * If full_discard is true, the sector should not read back as zeroes,
1434 * but rather fall through to the backing file.
1436 switch (qcow2_get_cluster_type(old_l2_entry
)) {
1437 case QCOW2_CLUSTER_UNALLOCATED
:
1438 if (full_discard
|| !bs
->backing_hd
) {
1443 case QCOW2_CLUSTER_ZERO
:
1444 if (!full_discard
) {
1449 case QCOW2_CLUSTER_NORMAL
:
1450 case QCOW2_CLUSTER_COMPRESSED
:
1457 /* First remove L2 entries */
1458 qcow2_cache_entry_mark_dirty(bs
, s
->l2_table_cache
, l2_table
);
1459 if (!full_discard
&& s
->qcow_version
>= 3) {
1460 l2_table
[l2_index
+ i
] = cpu_to_be64(QCOW_OFLAG_ZERO
);
1462 l2_table
[l2_index
+ i
] = cpu_to_be64(0);
1465 /* Then decrease the refcount */
1466 qcow2_free_any_clusters(bs
, old_l2_entry
, 1, type
);
1469 qcow2_cache_put(bs
, s
->l2_table_cache
, (void **) &l2_table
);
1474 int qcow2_discard_clusters(BlockDriverState
*bs
, uint64_t offset
,
1475 int nb_sectors
, enum qcow2_discard_type type
, bool full_discard
)
1477 BDRVQcowState
*s
= bs
->opaque
;
1478 uint64_t end_offset
;
1479 unsigned int nb_clusters
;
1482 end_offset
= offset
+ (nb_sectors
<< BDRV_SECTOR_BITS
);
1484 /* Round start up and end down */
1485 offset
= align_offset(offset
, s
->cluster_size
);
1486 end_offset
= start_of_cluster(s
, end_offset
);
1488 if (offset
> end_offset
) {
1492 nb_clusters
= size_to_clusters(s
, end_offset
- offset
);
1494 s
->cache_discards
= true;
1496 /* Each L2 table is handled by its own loop iteration */
1497 while (nb_clusters
> 0) {
1498 ret
= discard_single_l2(bs
, offset
, nb_clusters
, type
, full_discard
);
1504 offset
+= (ret
* s
->cluster_size
);
1509 s
->cache_discards
= false;
1510 qcow2_process_discards(bs
, ret
);
1516 * This zeroes as many clusters of nb_clusters as possible at once (i.e.
1517 * all clusters in the same L2 table) and returns the number of zeroed
1520 static int zero_single_l2(BlockDriverState
*bs
, uint64_t offset
,
1521 unsigned int nb_clusters
)
1523 BDRVQcowState
*s
= bs
->opaque
;
1529 ret
= get_cluster_table(bs
, offset
, &l2_table
, &l2_index
);
1534 /* Limit nb_clusters to one L2 table */
1535 nb_clusters
= MIN(nb_clusters
, s
->l2_size
- l2_index
);
1537 for (i
= 0; i
< nb_clusters
; i
++) {
1538 uint64_t old_offset
;
1540 old_offset
= be64_to_cpu(l2_table
[l2_index
+ i
]);
1542 /* Update L2 entries */
1543 qcow2_cache_entry_mark_dirty(bs
, s
->l2_table_cache
, l2_table
);
1544 if (old_offset
& QCOW_OFLAG_COMPRESSED
) {
1545 l2_table
[l2_index
+ i
] = cpu_to_be64(QCOW_OFLAG_ZERO
);
1546 qcow2_free_any_clusters(bs
, old_offset
, 1, QCOW2_DISCARD_REQUEST
);
1548 l2_table
[l2_index
+ i
] |= cpu_to_be64(QCOW_OFLAG_ZERO
);
1552 qcow2_cache_put(bs
, s
->l2_table_cache
, (void **) &l2_table
);
1557 int qcow2_zero_clusters(BlockDriverState
*bs
, uint64_t offset
, int nb_sectors
)
1559 BDRVQcowState
*s
= bs
->opaque
;
1560 unsigned int nb_clusters
;
1563 /* The zero flag is only supported by version 3 and newer */
1564 if (s
->qcow_version
< 3) {
1568 /* Each L2 table is handled by its own loop iteration */
1569 nb_clusters
= size_to_clusters(s
, nb_sectors
<< BDRV_SECTOR_BITS
);
1571 s
->cache_discards
= true;
1573 while (nb_clusters
> 0) {
1574 ret
= zero_single_l2(bs
, offset
, nb_clusters
);
1580 offset
+= (ret
* s
->cluster_size
);
1585 s
->cache_discards
= false;
1586 qcow2_process_discards(bs
, ret
);
1592 * Expands all zero clusters in a specific L1 table (or deallocates them, for
1593 * non-backed non-pre-allocated zero clusters).
1595 * l1_entries and *visited_l1_entries are used to keep track of progress for
1596 * status_cb(). l1_entries contains the total number of L1 entries and
1597 * *visited_l1_entries counts all visited L1 entries.
1599 static int expand_zero_clusters_in_l1(BlockDriverState
*bs
, uint64_t *l1_table
,
1600 int l1_size
, int64_t *visited_l1_entries
,
1602 BlockDriverAmendStatusCB
*status_cb
)
1604 BDRVQcowState
*s
= bs
->opaque
;
1605 bool is_active_l1
= (l1_table
== s
->l1_table
);
1606 uint64_t *l2_table
= NULL
;
1610 if (!is_active_l1
) {
1611 /* inactive L2 tables require a buffer to be stored in when loading
1613 l2_table
= qemu_try_blockalign(bs
->file
, s
->cluster_size
);
1614 if (l2_table
== NULL
) {
1619 for (i
= 0; i
< l1_size
; i
++) {
1620 uint64_t l2_offset
= l1_table
[i
] & L1E_OFFSET_MASK
;
1621 bool l2_dirty
= false;
1622 uint64_t l2_refcount
;
1626 (*visited_l1_entries
)++;
1628 status_cb(bs
, *visited_l1_entries
, l1_entries
);
1633 if (offset_into_cluster(s
, l2_offset
)) {
1634 qcow2_signal_corruption(bs
, true, -1, -1, "L2 table offset %#"
1635 PRIx64
" unaligned (L1 index: %#x)",
1642 /* get active L2 tables from cache */
1643 ret
= qcow2_cache_get(bs
, s
->l2_table_cache
, l2_offset
,
1644 (void **)&l2_table
);
1646 /* load inactive L2 tables from disk */
1647 ret
= bdrv_read(bs
->file
, l2_offset
/ BDRV_SECTOR_SIZE
,
1648 (void *)l2_table
, s
->cluster_sectors
);
1654 ret
= qcow2_get_refcount(bs
, l2_offset
>> s
->cluster_bits
,
1660 for (j
= 0; j
< s
->l2_size
; j
++) {
1661 uint64_t l2_entry
= be64_to_cpu(l2_table
[j
]);
1662 int64_t offset
= l2_entry
& L2E_OFFSET_MASK
;
1663 int cluster_type
= qcow2_get_cluster_type(l2_entry
);
1664 bool preallocated
= offset
!= 0;
1666 if (cluster_type
!= QCOW2_CLUSTER_ZERO
) {
1670 if (!preallocated
) {
1671 if (!bs
->backing_hd
) {
1672 /* not backed; therefore we can simply deallocate the
1679 offset
= qcow2_alloc_clusters(bs
, s
->cluster_size
);
1685 if (l2_refcount
> 1) {
1686 /* For shared L2 tables, set the refcount accordingly (it is
1687 * already 1 and needs to be l2_refcount) */
1688 ret
= qcow2_update_cluster_refcount(bs
,
1689 offset
>> s
->cluster_bits
,
1690 refcount_diff(1, l2_refcount
), false,
1691 QCOW2_DISCARD_OTHER
);
1693 qcow2_free_clusters(bs
, offset
, s
->cluster_size
,
1694 QCOW2_DISCARD_OTHER
);
1700 if (offset_into_cluster(s
, offset
)) {
1701 qcow2_signal_corruption(bs
, true, -1, -1, "Data cluster offset "
1702 "%#" PRIx64
" unaligned (L2 offset: %#"
1703 PRIx64
", L2 index: %#x)", offset
,
1705 if (!preallocated
) {
1706 qcow2_free_clusters(bs
, offset
, s
->cluster_size
,
1707 QCOW2_DISCARD_ALWAYS
);
1713 ret
= qcow2_pre_write_overlap_check(bs
, 0, offset
, s
->cluster_size
);
1715 if (!preallocated
) {
1716 qcow2_free_clusters(bs
, offset
, s
->cluster_size
,
1717 QCOW2_DISCARD_ALWAYS
);
1722 ret
= bdrv_write_zeroes(bs
->file
, offset
/ BDRV_SECTOR_SIZE
,
1723 s
->cluster_sectors
, 0);
1725 if (!preallocated
) {
1726 qcow2_free_clusters(bs
, offset
, s
->cluster_size
,
1727 QCOW2_DISCARD_ALWAYS
);
1732 if (l2_refcount
== 1) {
1733 l2_table
[j
] = cpu_to_be64(offset
| QCOW_OFLAG_COPIED
);
1735 l2_table
[j
] = cpu_to_be64(offset
);
1742 qcow2_cache_entry_mark_dirty(bs
, s
->l2_table_cache
, l2_table
);
1743 qcow2_cache_depends_on_flush(s
->l2_table_cache
);
1745 qcow2_cache_put(bs
, s
->l2_table_cache
, (void **) &l2_table
);
1748 ret
= qcow2_pre_write_overlap_check(bs
,
1749 QCOW2_OL_INACTIVE_L2
| QCOW2_OL_ACTIVE_L2
, l2_offset
,
1755 ret
= bdrv_write(bs
->file
, l2_offset
/ BDRV_SECTOR_SIZE
,
1756 (void *)l2_table
, s
->cluster_sectors
);
1763 (*visited_l1_entries
)++;
1765 status_cb(bs
, *visited_l1_entries
, l1_entries
);
1773 if (!is_active_l1
) {
1774 qemu_vfree(l2_table
);
1776 qcow2_cache_put(bs
, s
->l2_table_cache
, (void **) &l2_table
);
1783 * For backed images, expands all zero clusters on the image. For non-backed
1784 * images, deallocates all non-pre-allocated zero clusters (and claims the
1785 * allocation for pre-allocated ones). This is important for downgrading to a
1786 * qcow2 version which doesn't yet support metadata zero clusters.
1788 int qcow2_expand_zero_clusters(BlockDriverState
*bs
,
1789 BlockDriverAmendStatusCB
*status_cb
)
1791 BDRVQcowState
*s
= bs
->opaque
;
1792 uint64_t *l1_table
= NULL
;
1793 int64_t l1_entries
= 0, visited_l1_entries
= 0;
1798 l1_entries
= s
->l1_size
;
1799 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
1800 l1_entries
+= s
->snapshots
[i
].l1_size
;
1804 ret
= expand_zero_clusters_in_l1(bs
, s
->l1_table
, s
->l1_size
,
1805 &visited_l1_entries
, l1_entries
,
1811 /* Inactive L1 tables may point to active L2 tables - therefore it is
1812 * necessary to flush the L2 table cache before trying to access the L2
1813 * tables pointed to by inactive L1 entries (else we might try to expand
1814 * zero clusters that have already been expanded); furthermore, it is also
1815 * necessary to empty the L2 table cache, since it may contain tables which
1816 * are now going to be modified directly on disk, bypassing the cache.
1817 * qcow2_cache_empty() does both for us. */
1818 ret
= qcow2_cache_empty(bs
, s
->l2_table_cache
);
1823 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
1824 int l1_sectors
= (s
->snapshots
[i
].l1_size
* sizeof(uint64_t) +
1825 BDRV_SECTOR_SIZE
- 1) / BDRV_SECTOR_SIZE
;
1827 l1_table
= g_realloc(l1_table
, l1_sectors
* BDRV_SECTOR_SIZE
);
1829 ret
= bdrv_read(bs
->file
, s
->snapshots
[i
].l1_table_offset
/
1830 BDRV_SECTOR_SIZE
, (void *)l1_table
, l1_sectors
);
1835 for (j
= 0; j
< s
->snapshots
[i
].l1_size
; j
++) {
1836 be64_to_cpus(&l1_table
[j
]);
1839 ret
= expand_zero_clusters_in_l1(bs
, l1_table
, s
->snapshots
[i
].l1_size
,
1840 &visited_l1_entries
, l1_entries
,