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
25 #include "qemu/osdep.h"
28 #include "qapi/error.h"
29 #include "qemu-common.h"
30 #include "block/block_int.h"
31 #include "block/qcow2.h"
34 int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size,
37 BDRVQcow2State *s = bs->opaque;
38 int new_l1_size2, ret, i;
39 uint64_t *new_l1_table;
40 int64_t old_l1_table_offset, old_l1_size;
41 int64_t new_l1_table_offset, new_l1_size;
44 if (min_size <= s->l1_size)
47 /* Do a sanity check on min_size before trying to calculate new_l1_size
48 * (this prevents overflows during the while loop for the calculation of
50 if (min_size > INT_MAX / sizeof(uint64_t)) {
55 new_l1_size = min_size;
57 /* Bump size up to reduce the number of times we have to grow */
58 new_l1_size = s->l1_size;
59 if (new_l1_size == 0) {
62 while (min_size > new_l1_size) {
63 new_l1_size = (new_l1_size * 3 + 1) / 2;
67 if (new_l1_size > INT_MAX / sizeof(uint64_t)) {
72 fprintf(stderr, "grow l1_table from %d to %" PRId64 "\n",
73 s->l1_size, new_l1_size);
76 new_l1_size2 = sizeof(uint64_t) * new_l1_size;
77 new_l1_table = qemu_try_blockalign(bs->file->bs,
78 align_offset(new_l1_size2, 512));
79 if (new_l1_table == NULL) {
82 memset(new_l1_table, 0, align_offset(new_l1_size2, 512));
84 memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
86 /* write new table (align to cluster) */
87 BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ALLOC_TABLE);
88 new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2);
89 if (new_l1_table_offset < 0) {
90 qemu_vfree(new_l1_table);
91 return new_l1_table_offset;
94 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
99 /* the L1 position has not yet been updated, so these clusters must
100 * indeed be completely free */
101 ret = qcow2_pre_write_overlap_check(bs, 0, new_l1_table_offset,
107 BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_WRITE_TABLE);
108 for(i = 0; i < s->l1_size; i++)
109 new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
110 ret = bdrv_pwrite_sync(bs->file->bs, new_l1_table_offset,
111 new_l1_table, new_l1_size2);
114 for(i = 0; i < s->l1_size; i++)
115 new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
118 BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ACTIVATE_TABLE);
119 cpu_to_be32w((uint32_t*)data, new_l1_size);
120 stq_be_p(data + 4, new_l1_table_offset);
121 ret = bdrv_pwrite_sync(bs->file->bs, offsetof(QCowHeader, l1_size),
126 qemu_vfree(s->l1_table);
127 old_l1_table_offset = s->l1_table_offset;
128 s->l1_table_offset = new_l1_table_offset;
129 s->l1_table = new_l1_table;
130 old_l1_size = s->l1_size;
131 s->l1_size = new_l1_size;
132 qcow2_free_clusters(bs, old_l1_table_offset, old_l1_size * sizeof(uint64_t),
133 QCOW2_DISCARD_OTHER);
136 qemu_vfree(new_l1_table);
137 qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2,
138 QCOW2_DISCARD_OTHER);
145 * Loads a L2 table into memory. If the table is in the cache, the cache
146 * is used; otherwise the L2 table is loaded from the image file.
148 * Returns a pointer to the L2 table on success, or NULL if the read from
149 * the image file failed.
152 static int l2_load(BlockDriverState *bs, uint64_t l2_offset,
155 BDRVQcow2State *s = bs->opaque;
158 ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset, (void**) l2_table);
164 * Writes one sector of the L1 table to the disk (can't update single entries
165 * and we really don't want bdrv_pread to perform a read-modify-write)
167 #define L1_ENTRIES_PER_SECTOR (512 / 8)
168 int qcow2_write_l1_entry(BlockDriverState *bs, int l1_index)
170 BDRVQcow2State *s = bs->opaque;
171 uint64_t buf[L1_ENTRIES_PER_SECTOR] = { 0 };
175 l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1);
176 for (i = 0; i < L1_ENTRIES_PER_SECTOR && l1_start_index + i < s->l1_size;
179 buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]);
182 ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L1,
183 s->l1_table_offset + 8 * l1_start_index, sizeof(buf));
188 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
189 ret = bdrv_pwrite_sync(bs->file->bs,
190 s->l1_table_offset + 8 * l1_start_index,
202 * Allocate a new l2 entry in the file. If l1_index points to an already
203 * used entry in the L2 table (i.e. we are doing a copy on write for the L2
204 * table) copy the contents of the old L2 table into the newly allocated one.
205 * Otherwise the new table is initialized with zeros.
209 static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table)
211 BDRVQcow2State *s = bs->opaque;
212 uint64_t old_l2_offset;
213 uint64_t *l2_table = NULL;
217 old_l2_offset = s->l1_table[l1_index];
219 trace_qcow2_l2_allocate(bs, l1_index);
221 /* allocate a new l2 entry */
223 l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
229 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
234 /* allocate a new entry in the l2 cache */
236 trace_qcow2_l2_allocate_get_empty(bs, l1_index);
237 ret = qcow2_cache_get_empty(bs, s->l2_table_cache, l2_offset, (void**) table);
244 if ((old_l2_offset & L1E_OFFSET_MASK) == 0) {
245 /* if there was no old l2 table, clear the new table */
246 memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
250 /* if there was an old l2 table, read it from the disk */
251 BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_COW_READ);
252 ret = qcow2_cache_get(bs, s->l2_table_cache,
253 old_l2_offset & L1E_OFFSET_MASK,
254 (void**) &old_table);
259 memcpy(l2_table, old_table, s->cluster_size);
261 qcow2_cache_put(bs, s->l2_table_cache, (void **) &old_table);
264 /* write the l2 table to the file */
265 BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE);
267 trace_qcow2_l2_allocate_write_l2(bs, l1_index);
268 qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
269 ret = qcow2_cache_flush(bs, s->l2_table_cache);
274 /* update the L1 entry */
275 trace_qcow2_l2_allocate_write_l1(bs, l1_index);
276 s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
277 ret = qcow2_write_l1_entry(bs, l1_index);
283 trace_qcow2_l2_allocate_done(bs, l1_index, 0);
287 trace_qcow2_l2_allocate_done(bs, l1_index, ret);
288 if (l2_table != NULL) {
289 qcow2_cache_put(bs, s->l2_table_cache, (void**) table);
291 s->l1_table[l1_index] = old_l2_offset;
293 qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t),
294 QCOW2_DISCARD_ALWAYS);
300 * Checks how many clusters in a given L2 table are contiguous in the image
301 * file. As soon as one of the flags in the bitmask stop_flags changes compared
302 * to the first cluster, the search is stopped and the cluster is not counted
303 * as contiguous. (This allows it, for example, to stop at the first compressed
304 * cluster which may require a different handling)
306 static int count_contiguous_clusters(int nb_clusters, int cluster_size,
307 uint64_t *l2_table, uint64_t stop_flags)
310 uint64_t mask = stop_flags | L2E_OFFSET_MASK | QCOW_OFLAG_COMPRESSED;
311 uint64_t first_entry = be64_to_cpu(l2_table[0]);
312 uint64_t offset = first_entry & mask;
317 assert(qcow2_get_cluster_type(first_entry) == QCOW2_CLUSTER_NORMAL);
319 for (i = 0; i < nb_clusters; i++) {
320 uint64_t l2_entry = be64_to_cpu(l2_table[i]) & mask;
321 if (offset + (uint64_t) i * cluster_size != l2_entry) {
329 static int count_contiguous_clusters_by_type(int nb_clusters,
335 for (i = 0; i < nb_clusters; i++) {
336 int type = qcow2_get_cluster_type(be64_to_cpu(l2_table[i]));
338 if (type != wanted_type) {
346 /* The crypt function is compatible with the linux cryptoloop
347 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
349 int qcow2_encrypt_sectors(BDRVQcow2State *s, int64_t sector_num,
350 uint8_t *out_buf, const uint8_t *in_buf,
351 int nb_sectors, bool enc,
361 for(i = 0; i < nb_sectors; i++) {
362 ivec.ll[0] = cpu_to_le64(sector_num);
364 if (qcrypto_cipher_setiv(s->cipher,
365 ivec.b, G_N_ELEMENTS(ivec.b),
370 ret = qcrypto_cipher_encrypt(s->cipher,
376 ret = qcrypto_cipher_decrypt(s->cipher,
392 static int coroutine_fn copy_sectors(BlockDriverState *bs,
394 uint64_t cluster_offset,
395 int n_start, int n_end)
397 BDRVQcow2State *s = bs->opaque;
407 iov.iov_len = n * BDRV_SECTOR_SIZE;
408 iov.iov_base = qemu_try_blockalign(bs, iov.iov_len);
409 if (iov.iov_base == NULL) {
413 qemu_iovec_init_external(&qiov, &iov, 1);
415 BLKDBG_EVENT(bs->file, BLKDBG_COW_READ);
422 /* Call .bdrv_co_readv() directly instead of using the public block-layer
423 * interface. This avoids double I/O throttling and request tracking,
424 * which can lead to deadlock when block layer copy-on-read is enabled.
426 ret = bs->drv->bdrv_co_readv(bs, start_sect + n_start, n, &qiov);
434 if (qcow2_encrypt_sectors(s, start_sect + n_start,
435 iov.iov_base, iov.iov_base, n,
443 ret = qcow2_pre_write_overlap_check(bs, 0,
444 cluster_offset + n_start * BDRV_SECTOR_SIZE, n * BDRV_SECTOR_SIZE);
449 BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE);
450 ret = bdrv_co_writev(bs->file->bs, (cluster_offset >> 9) + n_start, n,
458 qemu_vfree(iov.iov_base);
466 * For a given offset of the disk image, find the cluster offset in
467 * qcow2 file. The offset is stored in *cluster_offset.
469 * on entry, *num is the number of contiguous sectors we'd like to
470 * access following offset.
472 * on exit, *num is the number of contiguous sectors we can read.
474 * Returns the cluster type (QCOW2_CLUSTER_*) on success, -errno in error
477 int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
478 int *num, uint64_t *cluster_offset)
480 BDRVQcow2State *s = bs->opaque;
481 unsigned int l2_index;
482 uint64_t l1_index, l2_offset, *l2_table;
484 unsigned int index_in_cluster, nb_clusters;
485 uint64_t nb_available, nb_needed;
488 index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
489 nb_needed = *num + index_in_cluster;
491 l1_bits = s->l2_bits + s->cluster_bits;
493 /* compute how many bytes there are between the offset and
494 * the end of the l1 entry
497 nb_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1));
499 /* compute the number of available sectors */
501 nb_available = (nb_available >> 9) + index_in_cluster;
503 if (nb_needed > nb_available) {
504 nb_needed = nb_available;
506 assert(nb_needed <= INT_MAX);
510 /* seek to the l2 offset in the l1 table */
512 l1_index = offset >> l1_bits;
513 if (l1_index >= s->l1_size) {
514 ret = QCOW2_CLUSTER_UNALLOCATED;
518 l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK;
520 ret = QCOW2_CLUSTER_UNALLOCATED;
524 if (offset_into_cluster(s, l2_offset)) {
525 qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64
526 " unaligned (L1 index: %#" PRIx64 ")",
527 l2_offset, l1_index);
531 /* load the l2 table in memory */
533 ret = l2_load(bs, l2_offset, &l2_table);
538 /* find the cluster offset for the given disk offset */
540 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
541 *cluster_offset = be64_to_cpu(l2_table[l2_index]);
543 /* nb_needed <= INT_MAX, thus nb_clusters <= INT_MAX, too */
544 nb_clusters = size_to_clusters(s, nb_needed << 9);
546 ret = qcow2_get_cluster_type(*cluster_offset);
548 case QCOW2_CLUSTER_COMPRESSED:
549 /* Compressed clusters can only be processed one by one */
551 *cluster_offset &= L2E_COMPRESSED_OFFSET_SIZE_MASK;
553 case QCOW2_CLUSTER_ZERO:
554 if (s->qcow_version < 3) {
555 qcow2_signal_corruption(bs, true, -1, -1, "Zero cluster entry found"
556 " in pre-v3 image (L2 offset: %#" PRIx64
557 ", L2 index: %#x)", l2_offset, l2_index);
561 c = count_contiguous_clusters_by_type(nb_clusters, &l2_table[l2_index],
565 case QCOW2_CLUSTER_UNALLOCATED:
566 /* how many empty clusters ? */
567 c = count_contiguous_clusters_by_type(nb_clusters, &l2_table[l2_index],
568 QCOW2_CLUSTER_UNALLOCATED);
571 case QCOW2_CLUSTER_NORMAL:
572 /* how many allocated clusters ? */
573 c = count_contiguous_clusters(nb_clusters, s->cluster_size,
574 &l2_table[l2_index], QCOW_OFLAG_ZERO);
575 *cluster_offset &= L2E_OFFSET_MASK;
576 if (offset_into_cluster(s, *cluster_offset)) {
577 qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset %#"
578 PRIx64 " unaligned (L2 offset: %#" PRIx64
579 ", L2 index: %#x)", *cluster_offset,
580 l2_offset, l2_index);
589 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
591 nb_available = (c * s->cluster_sectors);
594 if (nb_available > nb_needed)
595 nb_available = nb_needed;
597 *num = nb_available - index_in_cluster;
602 qcow2_cache_put(bs, s->l2_table_cache, (void **)&l2_table);
609 * for a given disk offset, load (and allocate if needed)
612 * the l2 table offset in the qcow2 file and the cluster index
613 * in the l2 table are given to the caller.
615 * Returns 0 on success, -errno in failure case
617 static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
618 uint64_t **new_l2_table,
621 BDRVQcow2State *s = bs->opaque;
622 unsigned int l2_index;
623 uint64_t l1_index, l2_offset;
624 uint64_t *l2_table = NULL;
627 /* seek to the l2 offset in the l1 table */
629 l1_index = offset >> (s->l2_bits + s->cluster_bits);
630 if (l1_index >= s->l1_size) {
631 ret = qcow2_grow_l1_table(bs, l1_index + 1, false);
637 assert(l1_index < s->l1_size);
638 l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK;
639 if (offset_into_cluster(s, l2_offset)) {
640 qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64
641 " unaligned (L1 index: %#" PRIx64 ")",
642 l2_offset, l1_index);
646 /* seek the l2 table of the given l2 offset */
648 if (s->l1_table[l1_index] & QCOW_OFLAG_COPIED) {
649 /* load the l2 table in memory */
650 ret = l2_load(bs, l2_offset, &l2_table);
655 /* First allocate a new L2 table (and do COW if needed) */
656 ret = l2_allocate(bs, l1_index, &l2_table);
661 /* Then decrease the refcount of the old table */
663 qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t),
664 QCOW2_DISCARD_OTHER);
668 /* find the cluster offset for the given disk offset */
670 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
672 *new_l2_table = l2_table;
673 *new_l2_index = l2_index;
679 * alloc_compressed_cluster_offset
681 * For a given offset of the disk image, return cluster offset in
684 * If the offset is not found, allocate a new compressed cluster.
686 * Return the cluster offset if successful,
687 * Return 0, otherwise.
691 uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs,
695 BDRVQcow2State *s = bs->opaque;
698 int64_t cluster_offset;
701 ret = get_cluster_table(bs, offset, &l2_table, &l2_index);
706 /* Compression can't overwrite anything. Fail if the cluster was already
708 cluster_offset = be64_to_cpu(l2_table[l2_index]);
709 if (cluster_offset & L2E_OFFSET_MASK) {
710 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
714 cluster_offset = qcow2_alloc_bytes(bs, compressed_size);
715 if (cluster_offset < 0) {
716 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
720 nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
721 (cluster_offset >> 9);
723 cluster_offset |= QCOW_OFLAG_COMPRESSED |
724 ((uint64_t)nb_csectors << s->csize_shift);
726 /* update L2 table */
728 /* compressed clusters never have the copied flag */
730 BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
731 qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
732 l2_table[l2_index] = cpu_to_be64(cluster_offset);
733 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
735 return cluster_offset;
738 static int perform_cow(BlockDriverState *bs, QCowL2Meta *m, Qcow2COWRegion *r)
740 BDRVQcow2State *s = bs->opaque;
743 if (r->nb_sectors == 0) {
747 qemu_co_mutex_unlock(&s->lock);
748 ret = copy_sectors(bs, m->offset / BDRV_SECTOR_SIZE, m->alloc_offset,
749 r->offset / BDRV_SECTOR_SIZE,
750 r->offset / BDRV_SECTOR_SIZE + r->nb_sectors);
751 qemu_co_mutex_lock(&s->lock);
758 * Before we update the L2 table to actually point to the new cluster, we
759 * need to be sure that the refcounts have been increased and COW was
762 qcow2_cache_depends_on_flush(s->l2_table_cache);
767 int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
769 BDRVQcow2State *s = bs->opaque;
770 int i, j = 0, l2_index, ret;
771 uint64_t *old_cluster, *l2_table;
772 uint64_t cluster_offset = m->alloc_offset;
774 trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters);
775 assert(m->nb_clusters > 0);
777 old_cluster = g_try_new(uint64_t, m->nb_clusters);
778 if (old_cluster == NULL) {
783 /* copy content of unmodified sectors */
784 ret = perform_cow(bs, m, &m->cow_start);
789 ret = perform_cow(bs, m, &m->cow_end);
794 /* Update L2 table. */
795 if (s->use_lazy_refcounts) {
796 qcow2_mark_dirty(bs);
798 if (qcow2_need_accurate_refcounts(s)) {
799 qcow2_cache_set_dependency(bs, s->l2_table_cache,
800 s->refcount_block_cache);
803 ret = get_cluster_table(bs, m->offset, &l2_table, &l2_index);
807 qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
809 assert(l2_index + m->nb_clusters <= s->l2_size);
810 for (i = 0; i < m->nb_clusters; i++) {
811 /* if two concurrent writes happen to the same unallocated cluster
812 * each write allocates separate cluster and writes data concurrently.
813 * The first one to complete updates l2 table with pointer to its
814 * cluster the second one has to do RMW (which is done above by
815 * copy_sectors()), update l2 table with its cluster pointer and free
816 * old cluster. This is what this loop does */
817 if(l2_table[l2_index + i] != 0)
818 old_cluster[j++] = l2_table[l2_index + i];
820 l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
821 (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
825 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
828 * If this was a COW, we need to decrease the refcount of the old cluster.
830 * Don't discard clusters that reach a refcount of 0 (e.g. compressed
831 * clusters), the next write will reuse them anyway.
834 for (i = 0; i < j; i++) {
835 qcow2_free_any_clusters(bs, be64_to_cpu(old_cluster[i]), 1,
836 QCOW2_DISCARD_NEVER);
847 * Returns the number of contiguous clusters that can be used for an allocating
848 * write, but require COW to be performed (this includes yet unallocated space,
849 * which must copy from the backing file)
851 static int count_cow_clusters(BDRVQcow2State *s, int nb_clusters,
852 uint64_t *l2_table, int l2_index)
856 for (i = 0; i < nb_clusters; i++) {
857 uint64_t l2_entry = be64_to_cpu(l2_table[l2_index + i]);
858 int cluster_type = qcow2_get_cluster_type(l2_entry);
860 switch(cluster_type) {
861 case QCOW2_CLUSTER_NORMAL:
862 if (l2_entry & QCOW_OFLAG_COPIED) {
866 case QCOW2_CLUSTER_UNALLOCATED:
867 case QCOW2_CLUSTER_COMPRESSED:
868 case QCOW2_CLUSTER_ZERO:
876 assert(i <= nb_clusters);
881 * Check if there already is an AIO write request in flight which allocates
882 * the same cluster. In this case we need to wait until the previous
883 * request has completed and updated the L2 table accordingly.
886 * 0 if there was no dependency. *cur_bytes indicates the number of
887 * bytes from guest_offset that can be read before the next
888 * dependency must be processed (or the request is complete)
890 * -EAGAIN if we had to wait for another request, previously gathered
891 * information on cluster allocation may be invalid now. The caller
892 * must start over anyway, so consider *cur_bytes undefined.
894 static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset,
895 uint64_t *cur_bytes, QCowL2Meta **m)
897 BDRVQcow2State *s = bs->opaque;
898 QCowL2Meta *old_alloc;
899 uint64_t bytes = *cur_bytes;
901 QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) {
903 uint64_t start = guest_offset;
904 uint64_t end = start + bytes;
905 uint64_t old_start = l2meta_cow_start(old_alloc);
906 uint64_t old_end = l2meta_cow_end(old_alloc);
908 if (end <= old_start || start >= old_end) {
909 /* No intersection */
911 if (start < old_start) {
912 /* Stop at the start of a running allocation */
913 bytes = old_start - start;
918 /* Stop if already an l2meta exists. After yielding, it wouldn't
919 * be valid any more, so we'd have to clean up the old L2Metas
920 * and deal with requests depending on them before starting to
921 * gather new ones. Not worth the trouble. */
922 if (bytes == 0 && *m) {
928 /* Wait for the dependency to complete. We need to recheck
929 * the free/allocated clusters when we continue. */
930 qemu_co_mutex_unlock(&s->lock);
931 qemu_co_queue_wait(&old_alloc->dependent_requests);
932 qemu_co_mutex_lock(&s->lock);
938 /* Make sure that existing clusters and new allocations are only used up to
939 * the next dependency if we shortened the request above */
946 * Checks how many already allocated clusters that don't require a copy on
947 * write there are at the given guest_offset (up to *bytes). If
948 * *host_offset is not zero, only physically contiguous clusters beginning at
949 * this host offset are counted.
951 * Note that guest_offset may not be cluster aligned. In this case, the
952 * returned *host_offset points to exact byte referenced by guest_offset and
953 * therefore isn't cluster aligned as well.
956 * 0: if no allocated clusters are available at the given offset.
957 * *bytes is normally unchanged. It is set to 0 if the cluster
958 * is allocated and doesn't need COW, but doesn't have the right
961 * 1: if allocated clusters that don't require a COW are available at
962 * the requested offset. *bytes may have decreased and describes
963 * the length of the area that can be written to.
965 * -errno: in error cases
967 static int handle_copied(BlockDriverState *bs, uint64_t guest_offset,
968 uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m)
970 BDRVQcow2State *s = bs->opaque;
972 uint64_t cluster_offset;
974 uint64_t nb_clusters;
975 unsigned int keep_clusters;
978 trace_qcow2_handle_copied(qemu_coroutine_self(), guest_offset, *host_offset,
981 assert(*host_offset == 0 || offset_into_cluster(s, guest_offset)
982 == offset_into_cluster(s, *host_offset));
985 * Calculate the number of clusters to look for. We stop at L2 table
986 * boundaries to keep things simple.
989 size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes);
991 l2_index = offset_to_l2_index(s, guest_offset);
992 nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
993 assert(nb_clusters <= INT_MAX);
995 /* Find L2 entry for the first involved cluster */
996 ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index);
1001 cluster_offset = be64_to_cpu(l2_table[l2_index]);
1003 /* Check how many clusters are already allocated and don't need COW */
1004 if (qcow2_get_cluster_type(cluster_offset) == QCOW2_CLUSTER_NORMAL
1005 && (cluster_offset & QCOW_OFLAG_COPIED))
1007 /* If a specific host_offset is required, check it */
1008 bool offset_matches =
1009 (cluster_offset & L2E_OFFSET_MASK) == *host_offset;
1011 if (offset_into_cluster(s, cluster_offset & L2E_OFFSET_MASK)) {
1012 qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset "
1013 "%#llx unaligned (guest offset: %#" PRIx64
1014 ")", cluster_offset & L2E_OFFSET_MASK,
1020 if (*host_offset != 0 && !offset_matches) {
1026 /* We keep all QCOW_OFLAG_COPIED clusters */
1028 count_contiguous_clusters(nb_clusters, s->cluster_size,
1029 &l2_table[l2_index],
1030 QCOW_OFLAG_COPIED | QCOW_OFLAG_ZERO);
1031 assert(keep_clusters <= nb_clusters);
1033 *bytes = MIN(*bytes,
1034 keep_clusters * s->cluster_size
1035 - offset_into_cluster(s, guest_offset));
1044 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1046 /* Only return a host offset if we actually made progress. Otherwise we
1047 * would make requirements for handle_alloc() that it can't fulfill */
1049 *host_offset = (cluster_offset & L2E_OFFSET_MASK)
1050 + offset_into_cluster(s, guest_offset);
1057 * Allocates new clusters for the given guest_offset.
1059 * At most *nb_clusters are allocated, and on return *nb_clusters is updated to
1060 * contain the number of clusters that have been allocated and are contiguous
1061 * in the image file.
1063 * If *host_offset is non-zero, it specifies the offset in the image file at
1064 * which the new clusters must start. *nb_clusters can be 0 on return in this
1065 * case if the cluster at host_offset is already in use. If *host_offset is
1066 * zero, the clusters can be allocated anywhere in the image file.
1068 * *host_offset is updated to contain the offset into the image file at which
1069 * the first allocated cluster starts.
1071 * Return 0 on success and -errno in error cases. -EAGAIN means that the
1072 * function has been waiting for another request and the allocation must be
1073 * restarted, but the whole request should not be failed.
1075 static int do_alloc_cluster_offset(BlockDriverState *bs, uint64_t guest_offset,
1076 uint64_t *host_offset, uint64_t *nb_clusters)
1078 BDRVQcow2State *s = bs->opaque;
1080 trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset,
1081 *host_offset, *nb_clusters);
1083 /* Allocate new clusters */
1084 trace_qcow2_cluster_alloc_phys(qemu_coroutine_self());
1085 if (*host_offset == 0) {
1086 int64_t cluster_offset =
1087 qcow2_alloc_clusters(bs, *nb_clusters * s->cluster_size);
1088 if (cluster_offset < 0) {
1089 return cluster_offset;
1091 *host_offset = cluster_offset;
1094 int64_t ret = qcow2_alloc_clusters_at(bs, *host_offset, *nb_clusters);
1104 * Allocates new clusters for an area that either is yet unallocated or needs a
1105 * copy on write. If *host_offset is non-zero, clusters are only allocated if
1106 * the new allocation can match the specified host offset.
1108 * Note that guest_offset may not be cluster aligned. In this case, the
1109 * returned *host_offset points to exact byte referenced by guest_offset and
1110 * therefore isn't cluster aligned as well.
1113 * 0: if no clusters could be allocated. *bytes is set to 0,
1114 * *host_offset is left unchanged.
1116 * 1: if new clusters were allocated. *bytes may be decreased if the
1117 * new allocation doesn't cover all of the requested area.
1118 * *host_offset is updated to contain the host offset of the first
1119 * newly allocated cluster.
1121 * -errno: in error cases
1123 static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset,
1124 uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m)
1126 BDRVQcow2State *s = bs->opaque;
1130 uint64_t nb_clusters;
1133 uint64_t alloc_cluster_offset;
1135 trace_qcow2_handle_alloc(qemu_coroutine_self(), guest_offset, *host_offset,
1140 * Calculate the number of clusters to look for. We stop at L2 table
1141 * boundaries to keep things simple.
1144 size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes);
1146 l2_index = offset_to_l2_index(s, guest_offset);
1147 nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
1148 assert(nb_clusters <= INT_MAX);
1150 /* Find L2 entry for the first involved cluster */
1151 ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index);
1156 entry = be64_to_cpu(l2_table[l2_index]);
1158 /* For the moment, overwrite compressed clusters one by one */
1159 if (entry & QCOW_OFLAG_COMPRESSED) {
1162 nb_clusters = count_cow_clusters(s, nb_clusters, l2_table, l2_index);
1165 /* This function is only called when there were no non-COW clusters, so if
1166 * we can't find any unallocated or COW clusters either, something is
1167 * wrong with our code. */
1168 assert(nb_clusters > 0);
1170 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1172 /* Allocate, if necessary at a given offset in the image file */
1173 alloc_cluster_offset = start_of_cluster(s, *host_offset);
1174 ret = do_alloc_cluster_offset(bs, guest_offset, &alloc_cluster_offset,
1180 /* Can't extend contiguous allocation */
1181 if (nb_clusters == 0) {
1186 /* !*host_offset would overwrite the image header and is reserved for "no
1187 * host offset preferred". If 0 was a valid host offset, it'd trigger the
1188 * following overlap check; do that now to avoid having an invalid value in
1190 if (!alloc_cluster_offset) {
1191 ret = qcow2_pre_write_overlap_check(bs, 0, alloc_cluster_offset,
1192 nb_clusters * s->cluster_size);
1198 * Save info needed for meta data update.
1200 * requested_sectors: Number of sectors from the start of the first
1201 * newly allocated cluster to the end of the (possibly shortened
1202 * before) write request.
1204 * avail_sectors: Number of sectors from the start of the first
1205 * newly allocated to the end of the last newly allocated cluster.
1207 * nb_sectors: The number of sectors from the start of the first
1208 * newly allocated cluster to the end of the area that the write
1209 * request actually writes to (excluding COW at the end)
1211 int requested_sectors =
1212 (*bytes + offset_into_cluster(s, guest_offset))
1213 >> BDRV_SECTOR_BITS;
1214 int avail_sectors = nb_clusters
1215 << (s->cluster_bits - BDRV_SECTOR_BITS);
1216 int alloc_n_start = offset_into_cluster(s, guest_offset)
1217 >> BDRV_SECTOR_BITS;
1218 int nb_sectors = MIN(requested_sectors, avail_sectors);
1219 QCowL2Meta *old_m = *m;
1221 *m = g_malloc0(sizeof(**m));
1223 **m = (QCowL2Meta) {
1226 .alloc_offset = alloc_cluster_offset,
1227 .offset = start_of_cluster(s, guest_offset),
1228 .nb_clusters = nb_clusters,
1229 .nb_available = nb_sectors,
1233 .nb_sectors = alloc_n_start,
1236 .offset = nb_sectors * BDRV_SECTOR_SIZE,
1237 .nb_sectors = avail_sectors - nb_sectors,
1240 qemu_co_queue_init(&(*m)->dependent_requests);
1241 QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight);
1243 *host_offset = alloc_cluster_offset + offset_into_cluster(s, guest_offset);
1244 *bytes = MIN(*bytes, (nb_sectors * BDRV_SECTOR_SIZE)
1245 - offset_into_cluster(s, guest_offset));
1246 assert(*bytes != 0);
1251 if (*m && (*m)->nb_clusters > 0) {
1252 QLIST_REMOVE(*m, next_in_flight);
1258 * alloc_cluster_offset
1260 * For a given offset on the virtual disk, find the cluster offset in qcow2
1261 * file. If the offset is not found, allocate a new cluster.
1263 * If the cluster was already allocated, m->nb_clusters is set to 0 and
1264 * other fields in m are meaningless.
1266 * If the cluster is newly allocated, m->nb_clusters is set to the number of
1267 * contiguous clusters that have been allocated. In this case, the other
1268 * fields of m are valid and contain information about the first allocated
1271 * If the request conflicts with another write request in flight, the coroutine
1272 * is queued and will be reentered when the dependency has completed.
1274 * Return 0 on success and -errno in error cases
1276 int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset,
1277 int *num, uint64_t *host_offset, QCowL2Meta **m)
1279 BDRVQcow2State *s = bs->opaque;
1280 uint64_t start, remaining;
1281 uint64_t cluster_offset;
1285 trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset, *num);
1287 assert((offset & ~BDRV_SECTOR_MASK) == 0);
1291 remaining = (uint64_t)*num << BDRV_SECTOR_BITS;
1299 if (!*host_offset) {
1300 *host_offset = start_of_cluster(s, cluster_offset);
1303 assert(remaining >= cur_bytes);
1306 remaining -= cur_bytes;
1307 cluster_offset += cur_bytes;
1309 if (remaining == 0) {
1313 cur_bytes = remaining;
1316 * Now start gathering as many contiguous clusters as possible:
1318 * 1. Check for overlaps with in-flight allocations
1320 * a) Overlap not in the first cluster -> shorten this request and
1321 * let the caller handle the rest in its next loop iteration.
1323 * b) Real overlaps of two requests. Yield and restart the search
1324 * for contiguous clusters (the situation could have changed
1325 * while we were sleeping)
1327 * c) TODO: Request starts in the same cluster as the in-flight
1328 * allocation ends. Shorten the COW of the in-fight allocation,
1329 * set cluster_offset to write to the same cluster and set up
1330 * the right synchronisation between the in-flight request and
1333 ret = handle_dependencies(bs, start, &cur_bytes, m);
1334 if (ret == -EAGAIN) {
1335 /* Currently handle_dependencies() doesn't yield if we already had
1336 * an allocation. If it did, we would have to clean up the L2Meta
1337 * structs before starting over. */
1340 } else if (ret < 0) {
1342 } else if (cur_bytes == 0) {
1345 /* handle_dependencies() may have decreased cur_bytes (shortened
1346 * the allocations below) so that the next dependency is processed
1347 * correctly during the next loop iteration. */
1351 * 2. Count contiguous COPIED clusters.
1353 ret = handle_copied(bs, start, &cluster_offset, &cur_bytes, m);
1358 } else if (cur_bytes == 0) {
1363 * 3. If the request still hasn't completed, allocate new clusters,
1364 * considering any cluster_offset of steps 1c or 2.
1366 ret = handle_alloc(bs, start, &cluster_offset, &cur_bytes, m);
1372 assert(cur_bytes == 0);
1377 *num -= remaining >> BDRV_SECTOR_BITS;
1379 assert(*host_offset != 0);
1384 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
1385 const uint8_t *buf, int buf_size)
1387 z_stream strm1, *strm = &strm1;
1390 memset(strm, 0, sizeof(*strm));
1392 strm->next_in = (uint8_t *)buf;
1393 strm->avail_in = buf_size;
1394 strm->next_out = out_buf;
1395 strm->avail_out = out_buf_size;
1397 ret = inflateInit2(strm, -12);
1400 ret = inflate(strm, Z_FINISH);
1401 out_len = strm->next_out - out_buf;
1402 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
1403 out_len != out_buf_size) {
1411 int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
1413 BDRVQcow2State *s = bs->opaque;
1414 int ret, csize, nb_csectors, sector_offset;
1417 coffset = cluster_offset & s->cluster_offset_mask;
1418 if (s->cluster_cache_offset != coffset) {
1419 nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
1420 sector_offset = coffset & 511;
1421 csize = nb_csectors * 512 - sector_offset;
1422 BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
1423 ret = bdrv_read(bs->file->bs, coffset >> 9, s->cluster_data,
1428 if (decompress_buffer(s->cluster_cache, s->cluster_size,
1429 s->cluster_data + sector_offset, csize) < 0) {
1432 s->cluster_cache_offset = coffset;
1438 * This discards as many clusters of nb_clusters as possible at once (i.e.
1439 * all clusters in the same L2 table) and returns the number of discarded
1442 static int discard_single_l2(BlockDriverState *bs, uint64_t offset,
1443 uint64_t nb_clusters, enum qcow2_discard_type type,
1446 BDRVQcow2State *s = bs->opaque;
1452 ret = get_cluster_table(bs, offset, &l2_table, &l2_index);
1457 /* Limit nb_clusters to one L2 table */
1458 nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
1459 assert(nb_clusters <= INT_MAX);
1461 for (i = 0; i < nb_clusters; i++) {
1462 uint64_t old_l2_entry;
1464 old_l2_entry = be64_to_cpu(l2_table[l2_index + i]);
1467 * If full_discard is false, make sure that a discarded area reads back
1468 * as zeroes for v3 images (we cannot do it for v2 without actually
1469 * writing a zero-filled buffer). We can skip the operation if the
1470 * cluster is already marked as zero, or if it's unallocated and we
1471 * don't have a backing file.
1473 * TODO We might want to use bdrv_get_block_status(bs) here, but we're
1474 * holding s->lock, so that doesn't work today.
1476 * If full_discard is true, the sector should not read back as zeroes,
1477 * but rather fall through to the backing file.
1479 switch (qcow2_get_cluster_type(old_l2_entry)) {
1480 case QCOW2_CLUSTER_UNALLOCATED:
1481 if (full_discard || !bs->backing) {
1486 case QCOW2_CLUSTER_ZERO:
1487 if (!full_discard) {
1492 case QCOW2_CLUSTER_NORMAL:
1493 case QCOW2_CLUSTER_COMPRESSED:
1500 /* First remove L2 entries */
1501 qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
1502 if (!full_discard && s->qcow_version >= 3) {
1503 l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO);
1505 l2_table[l2_index + i] = cpu_to_be64(0);
1508 /* Then decrease the refcount */
1509 qcow2_free_any_clusters(bs, old_l2_entry, 1, type);
1512 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1517 int qcow2_discard_clusters(BlockDriverState *bs, uint64_t offset,
1518 int nb_sectors, enum qcow2_discard_type type, bool full_discard)
1520 BDRVQcow2State *s = bs->opaque;
1521 uint64_t end_offset;
1522 uint64_t nb_clusters;
1525 end_offset = offset + (nb_sectors << BDRV_SECTOR_BITS);
1527 /* Round start up and end down */
1528 offset = align_offset(offset, s->cluster_size);
1529 end_offset = start_of_cluster(s, end_offset);
1531 if (offset > end_offset) {
1535 nb_clusters = size_to_clusters(s, end_offset - offset);
1537 s->cache_discards = true;
1539 /* Each L2 table is handled by its own loop iteration */
1540 while (nb_clusters > 0) {
1541 ret = discard_single_l2(bs, offset, nb_clusters, type, full_discard);
1547 offset += (ret * s->cluster_size);
1552 s->cache_discards = false;
1553 qcow2_process_discards(bs, ret);
1559 * This zeroes as many clusters of nb_clusters as possible at once (i.e.
1560 * all clusters in the same L2 table) and returns the number of zeroed
1563 static int zero_single_l2(BlockDriverState *bs, uint64_t offset,
1564 uint64_t nb_clusters)
1566 BDRVQcow2State *s = bs->opaque;
1572 ret = get_cluster_table(bs, offset, &l2_table, &l2_index);
1577 /* Limit nb_clusters to one L2 table */
1578 nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
1579 assert(nb_clusters <= INT_MAX);
1581 for (i = 0; i < nb_clusters; i++) {
1582 uint64_t old_offset;
1584 old_offset = be64_to_cpu(l2_table[l2_index + i]);
1586 /* Update L2 entries */
1587 qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
1588 if (old_offset & QCOW_OFLAG_COMPRESSED) {
1589 l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO);
1590 qcow2_free_any_clusters(bs, old_offset, 1, QCOW2_DISCARD_REQUEST);
1592 l2_table[l2_index + i] |= cpu_to_be64(QCOW_OFLAG_ZERO);
1596 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1601 int qcow2_zero_clusters(BlockDriverState *bs, uint64_t offset, int nb_sectors)
1603 BDRVQcow2State *s = bs->opaque;
1604 uint64_t nb_clusters;
1607 /* The zero flag is only supported by version 3 and newer */
1608 if (s->qcow_version < 3) {
1612 /* Each L2 table is handled by its own loop iteration */
1613 nb_clusters = size_to_clusters(s, nb_sectors << BDRV_SECTOR_BITS);
1615 s->cache_discards = true;
1617 while (nb_clusters > 0) {
1618 ret = zero_single_l2(bs, offset, nb_clusters);
1624 offset += (ret * s->cluster_size);
1629 s->cache_discards = false;
1630 qcow2_process_discards(bs, ret);
1636 * Expands all zero clusters in a specific L1 table (or deallocates them, for
1637 * non-backed non-pre-allocated zero clusters).
1639 * l1_entries and *visited_l1_entries are used to keep track of progress for
1640 * status_cb(). l1_entries contains the total number of L1 entries and
1641 * *visited_l1_entries counts all visited L1 entries.
1643 static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table,
1644 int l1_size, int64_t *visited_l1_entries,
1646 BlockDriverAmendStatusCB *status_cb,
1649 BDRVQcow2State *s = bs->opaque;
1650 bool is_active_l1 = (l1_table == s->l1_table);
1651 uint64_t *l2_table = NULL;
1655 if (!is_active_l1) {
1656 /* inactive L2 tables require a buffer to be stored in when loading
1658 l2_table = qemu_try_blockalign(bs->file->bs, s->cluster_size);
1659 if (l2_table == NULL) {
1664 for (i = 0; i < l1_size; i++) {
1665 uint64_t l2_offset = l1_table[i] & L1E_OFFSET_MASK;
1666 bool l2_dirty = false;
1667 uint64_t l2_refcount;
1671 (*visited_l1_entries)++;
1673 status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque);
1678 if (offset_into_cluster(s, l2_offset)) {
1679 qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#"
1680 PRIx64 " unaligned (L1 index: %#x)",
1687 /* get active L2 tables from cache */
1688 ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset,
1689 (void **)&l2_table);
1691 /* load inactive L2 tables from disk */
1692 ret = bdrv_read(bs->file->bs, l2_offset / BDRV_SECTOR_SIZE,
1693 (void *)l2_table, s->cluster_sectors);
1699 ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits,
1705 for (j = 0; j < s->l2_size; j++) {
1706 uint64_t l2_entry = be64_to_cpu(l2_table[j]);
1707 int64_t offset = l2_entry & L2E_OFFSET_MASK;
1708 int cluster_type = qcow2_get_cluster_type(l2_entry);
1709 bool preallocated = offset != 0;
1711 if (cluster_type != QCOW2_CLUSTER_ZERO) {
1715 if (!preallocated) {
1717 /* not backed; therefore we can simply deallocate the
1724 offset = qcow2_alloc_clusters(bs, s->cluster_size);
1730 if (l2_refcount > 1) {
1731 /* For shared L2 tables, set the refcount accordingly (it is
1732 * already 1 and needs to be l2_refcount) */
1733 ret = qcow2_update_cluster_refcount(bs,
1734 offset >> s->cluster_bits,
1735 refcount_diff(1, l2_refcount), false,
1736 QCOW2_DISCARD_OTHER);
1738 qcow2_free_clusters(bs, offset, s->cluster_size,
1739 QCOW2_DISCARD_OTHER);
1745 if (offset_into_cluster(s, offset)) {
1746 qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset "
1747 "%#" PRIx64 " unaligned (L2 offset: %#"
1748 PRIx64 ", L2 index: %#x)", offset,
1750 if (!preallocated) {
1751 qcow2_free_clusters(bs, offset, s->cluster_size,
1752 QCOW2_DISCARD_ALWAYS);
1758 ret = qcow2_pre_write_overlap_check(bs, 0, offset, s->cluster_size);
1760 if (!preallocated) {
1761 qcow2_free_clusters(bs, offset, s->cluster_size,
1762 QCOW2_DISCARD_ALWAYS);
1767 ret = bdrv_write_zeroes(bs->file->bs, offset / BDRV_SECTOR_SIZE,
1768 s->cluster_sectors, 0);
1770 if (!preallocated) {
1771 qcow2_free_clusters(bs, offset, s->cluster_size,
1772 QCOW2_DISCARD_ALWAYS);
1777 if (l2_refcount == 1) {
1778 l2_table[j] = cpu_to_be64(offset | QCOW_OFLAG_COPIED);
1780 l2_table[j] = cpu_to_be64(offset);
1787 qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
1788 qcow2_cache_depends_on_flush(s->l2_table_cache);
1790 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1793 ret = qcow2_pre_write_overlap_check(bs,
1794 QCOW2_OL_INACTIVE_L2 | QCOW2_OL_ACTIVE_L2, l2_offset,
1800 ret = bdrv_write(bs->file->bs, l2_offset / BDRV_SECTOR_SIZE,
1801 (void *)l2_table, s->cluster_sectors);
1808 (*visited_l1_entries)++;
1810 status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque);
1818 if (!is_active_l1) {
1819 qemu_vfree(l2_table);
1821 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1828 * For backed images, expands all zero clusters on the image. For non-backed
1829 * images, deallocates all non-pre-allocated zero clusters (and claims the
1830 * allocation for pre-allocated ones). This is important for downgrading to a
1831 * qcow2 version which doesn't yet support metadata zero clusters.
1833 int qcow2_expand_zero_clusters(BlockDriverState *bs,
1834 BlockDriverAmendStatusCB *status_cb,
1837 BDRVQcow2State *s = bs->opaque;
1838 uint64_t *l1_table = NULL;
1839 int64_t l1_entries = 0, visited_l1_entries = 0;
1844 l1_entries = s->l1_size;
1845 for (i = 0; i < s->nb_snapshots; i++) {
1846 l1_entries += s->snapshots[i].l1_size;
1850 ret = expand_zero_clusters_in_l1(bs, s->l1_table, s->l1_size,
1851 &visited_l1_entries, l1_entries,
1852 status_cb, cb_opaque);
1857 /* Inactive L1 tables may point to active L2 tables - therefore it is
1858 * necessary to flush the L2 table cache before trying to access the L2
1859 * tables pointed to by inactive L1 entries (else we might try to expand
1860 * zero clusters that have already been expanded); furthermore, it is also
1861 * necessary to empty the L2 table cache, since it may contain tables which
1862 * are now going to be modified directly on disk, bypassing the cache.
1863 * qcow2_cache_empty() does both for us. */
1864 ret = qcow2_cache_empty(bs, s->l2_table_cache);
1869 for (i = 0; i < s->nb_snapshots; i++) {
1870 int l1_sectors = (s->snapshots[i].l1_size * sizeof(uint64_t) +
1871 BDRV_SECTOR_SIZE - 1) / BDRV_SECTOR_SIZE;
1873 l1_table = g_realloc(l1_table, l1_sectors * BDRV_SECTOR_SIZE);
1875 ret = bdrv_read(bs->file->bs,
1876 s->snapshots[i].l1_table_offset / BDRV_SECTOR_SIZE,
1877 (void *)l1_table, l1_sectors);
1882 for (j = 0; j < s->snapshots[i].l1_size; j++) {
1883 be64_to_cpus(&l1_table[j]);
1886 ret = expand_zero_clusters_in_l1(bs, l1_table, s->snapshots[i].l1_size,
1887 &visited_l1_entries, l1_entries,
1888 status_cb, cb_opaque);