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_int.h"
29 #include "block/qcow2.h"
32 int qcow2_grow_l1_table(BlockDriverState *bs, int min_size, bool exact_size)
34 BDRVQcowState *s = bs->opaque;
35 int new_l1_size, new_l1_size2, ret, i;
36 uint64_t *new_l1_table;
37 int64_t new_l1_table_offset;
40 if (min_size <= s->l1_size)
44 new_l1_size = min_size;
46 /* Bump size up to reduce the number of times we have to grow */
47 new_l1_size = s->l1_size;
48 if (new_l1_size == 0) {
51 while (min_size > new_l1_size) {
52 new_l1_size = (new_l1_size * 3 + 1) / 2;
57 fprintf(stderr, "grow l1_table from %d to %d\n", s->l1_size, new_l1_size);
60 new_l1_size2 = sizeof(uint64_t) * new_l1_size;
61 new_l1_table = g_malloc0(align_offset(new_l1_size2, 512));
62 memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
64 /* write new table (align to cluster) */
65 BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ALLOC_TABLE);
66 new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2);
67 if (new_l1_table_offset < 0) {
69 return new_l1_table_offset;
72 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
77 BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_WRITE_TABLE);
78 for(i = 0; i < s->l1_size; i++)
79 new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
80 ret = bdrv_pwrite_sync(bs->file, new_l1_table_offset, new_l1_table, new_l1_size2);
83 for(i = 0; i < s->l1_size; i++)
84 new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
87 BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ACTIVATE_TABLE);
88 cpu_to_be32w((uint32_t*)data, new_l1_size);
89 cpu_to_be64wu((uint64_t*)(data + 4), new_l1_table_offset);
90 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_size), data,sizeof(data));
95 qcow2_free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t));
96 s->l1_table_offset = new_l1_table_offset;
97 s->l1_table = new_l1_table;
98 s->l1_size = new_l1_size;
101 g_free(new_l1_table);
102 qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2);
109 * Loads a L2 table into memory. If the table is in the cache, the cache
110 * is used; otherwise the L2 table is loaded from the image file.
112 * Returns a pointer to the L2 table on success, or NULL if the read from
113 * the image file failed.
116 static int l2_load(BlockDriverState *bs, uint64_t l2_offset,
119 BDRVQcowState *s = bs->opaque;
122 ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset, (void**) l2_table);
128 * Writes one sector of the L1 table to the disk (can't update single entries
129 * and we really don't want bdrv_pread to perform a read-modify-write)
131 #define L1_ENTRIES_PER_SECTOR (512 / 8)
132 static int write_l1_entry(BlockDriverState *bs, int l1_index)
134 BDRVQcowState *s = bs->opaque;
135 uint64_t buf[L1_ENTRIES_PER_SECTOR];
139 l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1);
140 for (i = 0; i < L1_ENTRIES_PER_SECTOR; i++) {
141 buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]);
144 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
145 ret = bdrv_pwrite_sync(bs->file, s->l1_table_offset + 8 * l1_start_index,
157 * Allocate a new l2 entry in the file. If l1_index points to an already
158 * used entry in the L2 table (i.e. we are doing a copy on write for the L2
159 * table) copy the contents of the old L2 table into the newly allocated one.
160 * Otherwise the new table is initialized with zeros.
164 static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table)
166 BDRVQcowState *s = bs->opaque;
167 uint64_t old_l2_offset;
172 old_l2_offset = s->l1_table[l1_index];
174 trace_qcow2_l2_allocate(bs, l1_index);
176 /* allocate a new l2 entry */
178 l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
183 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
188 /* allocate a new entry in the l2 cache */
190 trace_qcow2_l2_allocate_get_empty(bs, l1_index);
191 ret = qcow2_cache_get_empty(bs, s->l2_table_cache, l2_offset, (void**) table);
198 if (old_l2_offset == 0) {
199 /* if there was no old l2 table, clear the new table */
200 memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
204 /* if there was an old l2 table, read it from the disk */
205 BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_COW_READ);
206 ret = qcow2_cache_get(bs, s->l2_table_cache, old_l2_offset,
207 (void**) &old_table);
212 memcpy(l2_table, old_table, s->cluster_size);
214 ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &old_table);
220 /* write the l2 table to the file */
221 BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE);
223 trace_qcow2_l2_allocate_write_l2(bs, l1_index);
224 qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
225 ret = qcow2_cache_flush(bs, s->l2_table_cache);
230 /* update the L1 entry */
231 trace_qcow2_l2_allocate_write_l1(bs, l1_index);
232 s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
233 ret = write_l1_entry(bs, l1_index);
239 trace_qcow2_l2_allocate_done(bs, l1_index, 0);
243 trace_qcow2_l2_allocate_done(bs, l1_index, ret);
244 qcow2_cache_put(bs, s->l2_table_cache, (void**) table);
245 s->l1_table[l1_index] = old_l2_offset;
249 static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size,
250 uint64_t *l2_table, uint64_t start, uint64_t mask)
253 uint64_t offset = be64_to_cpu(l2_table[0]) & ~mask;
258 for (i = start; i < start + nb_clusters; i++)
259 if (offset + (uint64_t) i * cluster_size != (be64_to_cpu(l2_table[i]) & ~mask))
265 static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table)
269 while(nb_clusters-- && l2_table[i] == 0)
275 /* The crypt function is compatible with the linux cryptoloop
276 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
278 void qcow2_encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
279 uint8_t *out_buf, const uint8_t *in_buf,
280 int nb_sectors, int enc,
289 for(i = 0; i < nb_sectors; i++) {
290 ivec.ll[0] = cpu_to_le64(sector_num);
292 AES_cbc_encrypt(in_buf, out_buf, 512, key,
300 static int coroutine_fn copy_sectors(BlockDriverState *bs,
302 uint64_t cluster_offset,
303 int n_start, int n_end)
305 BDRVQcowState *s = bs->opaque;
311 * If this is the last cluster and it is only partially used, we must only
312 * copy until the end of the image, or bdrv_check_request will fail for the
313 * bdrv_read/write calls below.
315 if (start_sect + n_end > bs->total_sectors) {
316 n_end = bs->total_sectors - start_sect;
324 iov.iov_len = n * BDRV_SECTOR_SIZE;
325 iov.iov_base = qemu_blockalign(bs, iov.iov_len);
327 qemu_iovec_init_external(&qiov, &iov, 1);
329 BLKDBG_EVENT(bs->file, BLKDBG_COW_READ);
331 /* Call .bdrv_co_readv() directly instead of using the public block-layer
332 * interface. This avoids double I/O throttling and request tracking,
333 * which can lead to deadlock when block layer copy-on-read is enabled.
335 ret = bs->drv->bdrv_co_readv(bs, start_sect + n_start, n, &qiov);
340 if (s->crypt_method) {
341 qcow2_encrypt_sectors(s, start_sect + n_start,
342 iov.iov_base, iov.iov_base, n, 1,
343 &s->aes_encrypt_key);
346 BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE);
347 ret = bdrv_co_writev(bs->file, (cluster_offset >> 9) + n_start, n, &qiov);
354 qemu_vfree(iov.iov_base);
362 * For a given offset of the disk image, find the cluster offset in
363 * qcow2 file. The offset is stored in *cluster_offset.
365 * on entry, *num is the number of contiguous sectors we'd like to
366 * access following offset.
368 * on exit, *num is the number of contiguous sectors we can read.
370 * Return 0, if the offset is found
371 * Return -errno, otherwise.
375 int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
376 int *num, uint64_t *cluster_offset)
378 BDRVQcowState *s = bs->opaque;
379 unsigned int l1_index, l2_index;
380 uint64_t l2_offset, *l2_table;
382 unsigned int index_in_cluster, nb_clusters;
383 uint64_t nb_available, nb_needed;
386 index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
387 nb_needed = *num + index_in_cluster;
389 l1_bits = s->l2_bits + s->cluster_bits;
391 /* compute how many bytes there are between the offset and
392 * the end of the l1 entry
395 nb_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1));
397 /* compute the number of available sectors */
399 nb_available = (nb_available >> 9) + index_in_cluster;
401 if (nb_needed > nb_available) {
402 nb_needed = nb_available;
407 /* seek the the l2 offset in the l1 table */
409 l1_index = offset >> l1_bits;
410 if (l1_index >= s->l1_size)
413 l2_offset = s->l1_table[l1_index];
415 /* seek the l2 table of the given l2 offset */
420 /* load the l2 table in memory */
422 l2_offset &= ~QCOW_OFLAG_COPIED;
423 ret = l2_load(bs, l2_offset, &l2_table);
428 /* find the cluster offset for the given disk offset */
430 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
431 *cluster_offset = be64_to_cpu(l2_table[l2_index]);
432 nb_clusters = size_to_clusters(s, nb_needed << 9);
434 if (!*cluster_offset) {
435 /* how many empty clusters ? */
436 c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]);
438 /* how many allocated clusters ? */
439 c = count_contiguous_clusters(nb_clusters, s->cluster_size,
440 &l2_table[l2_index], 0, QCOW_OFLAG_COPIED);
443 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
445 nb_available = (c * s->cluster_sectors);
447 if (nb_available > nb_needed)
448 nb_available = nb_needed;
450 *num = nb_available - index_in_cluster;
452 *cluster_offset &=~QCOW_OFLAG_COPIED;
459 * for a given disk offset, load (and allocate if needed)
462 * the l2 table offset in the qcow2 file and the cluster index
463 * in the l2 table are given to the caller.
465 * Returns 0 on success, -errno in failure case
467 static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
468 uint64_t **new_l2_table,
469 uint64_t *new_l2_offset,
472 BDRVQcowState *s = bs->opaque;
473 unsigned int l1_index, l2_index;
475 uint64_t *l2_table = NULL;
478 /* seek the the l2 offset in the l1 table */
480 l1_index = offset >> (s->l2_bits + s->cluster_bits);
481 if (l1_index >= s->l1_size) {
482 ret = qcow2_grow_l1_table(bs, l1_index + 1, false);
487 l2_offset = s->l1_table[l1_index];
489 /* seek the l2 table of the given l2 offset */
491 if (l2_offset & QCOW_OFLAG_COPIED) {
492 /* load the l2 table in memory */
493 l2_offset &= ~QCOW_OFLAG_COPIED;
494 ret = l2_load(bs, l2_offset, &l2_table);
499 /* First allocate a new L2 table (and do COW if needed) */
500 ret = l2_allocate(bs, l1_index, &l2_table);
505 /* Then decrease the refcount of the old table */
507 qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
509 l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED;
512 /* find the cluster offset for the given disk offset */
514 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
516 *new_l2_table = l2_table;
517 *new_l2_offset = l2_offset;
518 *new_l2_index = l2_index;
524 * alloc_compressed_cluster_offset
526 * For a given offset of the disk image, return cluster offset in
529 * If the offset is not found, allocate a new compressed cluster.
531 * Return the cluster offset if successful,
532 * Return 0, otherwise.
536 uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs,
540 BDRVQcowState *s = bs->opaque;
542 uint64_t l2_offset, *l2_table;
543 int64_t cluster_offset;
546 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
551 cluster_offset = be64_to_cpu(l2_table[l2_index]);
552 if (cluster_offset & QCOW_OFLAG_COPIED) {
553 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
558 qcow2_free_any_clusters(bs, cluster_offset, 1);
560 cluster_offset = qcow2_alloc_bytes(bs, compressed_size);
561 if (cluster_offset < 0) {
562 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
566 nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
567 (cluster_offset >> 9);
569 cluster_offset |= QCOW_OFLAG_COMPRESSED |
570 ((uint64_t)nb_csectors << s->csize_shift);
572 /* update L2 table */
574 /* compressed clusters never have the copied flag */
576 BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
577 qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
578 l2_table[l2_index] = cpu_to_be64(cluster_offset);
579 ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
584 return cluster_offset;
587 int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
589 BDRVQcowState *s = bs->opaque;
590 int i, j = 0, l2_index, ret;
591 uint64_t *old_cluster, start_sect, l2_offset, *l2_table;
592 uint64_t cluster_offset = m->cluster_offset;
595 trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters);
597 if (m->nb_clusters == 0)
600 old_cluster = g_malloc(m->nb_clusters * sizeof(uint64_t));
602 /* copy content of unmodified sectors */
603 start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9;
606 qemu_co_mutex_unlock(&s->lock);
607 ret = copy_sectors(bs, start_sect, cluster_offset, 0, m->n_start);
608 qemu_co_mutex_lock(&s->lock);
613 if (m->nb_available & (s->cluster_sectors - 1)) {
614 uint64_t end = m->nb_available & ~(uint64_t)(s->cluster_sectors - 1);
616 qemu_co_mutex_unlock(&s->lock);
617 ret = copy_sectors(bs, start_sect + end, cluster_offset + (end << 9),
618 m->nb_available - end, s->cluster_sectors);
619 qemu_co_mutex_lock(&s->lock);
627 * Before we update the L2 table to actually point to the new cluster, we
628 * need to be sure that the refcounts have been increased and COW was
632 qcow2_cache_depends_on_flush(s->l2_table_cache);
635 qcow2_cache_set_dependency(bs, s->l2_table_cache, s->refcount_block_cache);
636 ret = get_cluster_table(bs, m->offset, &l2_table, &l2_offset, &l2_index);
640 qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
642 for (i = 0; i < m->nb_clusters; i++) {
643 /* if two concurrent writes happen to the same unallocated cluster
644 * each write allocates separate cluster and writes data concurrently.
645 * The first one to complete updates l2 table with pointer to its
646 * cluster the second one has to do RMW (which is done above by
647 * copy_sectors()), update l2 table with its cluster pointer and free
648 * old cluster. This is what this loop does */
649 if(l2_table[l2_index + i] != 0)
650 old_cluster[j++] = l2_table[l2_index + i];
652 l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
653 (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
657 ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
663 * If this was a COW, we need to decrease the refcount of the old cluster.
664 * Also flush bs->file to get the right order for L2 and refcount update.
667 for (i = 0; i < j; i++) {
668 qcow2_free_any_clusters(bs,
669 be64_to_cpu(old_cluster[i]) & ~QCOW_OFLAG_COPIED, 1);
680 * Returns the number of contiguous clusters that can be used for an allocating
681 * write, but require COW to be performed (this includes yet unallocated space,
682 * which must copy from the backing file)
684 static int count_cow_clusters(BDRVQcowState *s, int nb_clusters,
685 uint64_t *l2_table, int l2_index)
688 uint64_t cluster_offset;
690 while (i < nb_clusters) {
691 i += count_contiguous_clusters(nb_clusters - i, s->cluster_size,
692 &l2_table[l2_index], i, 0);
693 if ((i >= nb_clusters) || be64_to_cpu(l2_table[l2_index + i])) {
697 i += count_contiguous_free_clusters(nb_clusters - i,
698 &l2_table[l2_index + i]);
699 if (i >= nb_clusters) {
703 cluster_offset = be64_to_cpu(l2_table[l2_index + i]);
705 if ((cluster_offset & QCOW_OFLAG_COPIED) ||
706 (cluster_offset & QCOW_OFLAG_COMPRESSED))
710 assert(i <= nb_clusters);
715 * alloc_cluster_offset
717 * For a given offset of the disk image, return cluster offset in qcow2 file.
718 * If the offset is not found, allocate a new cluster.
720 * If the cluster was already allocated, m->nb_clusters is set to 0,
721 * other fields in m are meaningless.
723 * If the cluster is newly allocated, m->nb_clusters is set to the number of
724 * contiguous clusters that have been allocated. In this case, the other
725 * fields of m are valid and contain information about the first allocated
728 * If the request conflicts with another write request in flight, the coroutine
729 * is queued and will be reentered when the dependency has completed.
731 * Return 0 on success and -errno in error cases
733 int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset,
734 int n_start, int n_end, int *num, QCowL2Meta *m)
736 BDRVQcowState *s = bs->opaque;
738 uint64_t l2_offset, *l2_table;
739 int64_t cluster_offset;
740 unsigned int nb_clusters, i = 0;
741 QCowL2Meta *old_alloc;
743 trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset,
746 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
752 nb_clusters = size_to_clusters(s, n_end << 9);
754 nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
756 cluster_offset = be64_to_cpu(l2_table[l2_index]);
758 /* We keep all QCOW_OFLAG_COPIED clusters */
760 if (cluster_offset & QCOW_OFLAG_COPIED) {
761 nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size,
762 &l2_table[l2_index], 0, 0);
764 cluster_offset &= ~QCOW_OFLAG_COPIED;
770 /* for the moment, multiple compressed clusters are not managed */
772 if (cluster_offset & QCOW_OFLAG_COMPRESSED)
775 /* how many available clusters ? */
777 i = count_cow_clusters(s, nb_clusters, l2_table, l2_index);
778 assert(i <= nb_clusters);
782 * Check if there already is an AIO write request in flight which allocates
783 * the same cluster. In this case we need to wait until the previous
784 * request has completed and updated the L2 table accordingly.
786 QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) {
788 uint64_t start = offset >> s->cluster_bits;
789 uint64_t end = start + nb_clusters;
790 uint64_t old_start = old_alloc->offset >> s->cluster_bits;
791 uint64_t old_end = old_start + old_alloc->nb_clusters;
793 if (end < old_start || start > old_end) {
794 /* No intersection */
796 if (start < old_start) {
797 /* Stop at the start of a running allocation */
798 nb_clusters = old_start - start;
803 if (nb_clusters == 0) {
804 /* Wait for the dependency to complete. We need to recheck
805 * the free/allocated clusters when we continue. */
806 qemu_co_mutex_unlock(&s->lock);
807 qemu_co_queue_wait(&old_alloc->dependent_requests);
808 qemu_co_mutex_lock(&s->lock);
818 /* save info needed for meta data update */
820 m->n_start = n_start;
821 m->nb_clusters = nb_clusters;
823 QLIST_INSERT_HEAD(&s->cluster_allocs, m, next_in_flight);
825 /* allocate a new cluster */
826 trace_qcow2_cluster_alloc_phys(qemu_coroutine_self());
827 cluster_offset = qcow2_alloc_clusters(bs, nb_clusters * s->cluster_size);
828 if (cluster_offset < 0) {
829 ret = cluster_offset;
834 ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
839 m->nb_available = MIN(nb_clusters << (s->cluster_bits - 9), n_end);
840 m->cluster_offset = cluster_offset;
842 *num = m->nb_available - n_start;
847 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
849 QLIST_REMOVE(m, next_in_flight);
853 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
854 const uint8_t *buf, int buf_size)
856 z_stream strm1, *strm = &strm1;
859 memset(strm, 0, sizeof(*strm));
861 strm->next_in = (uint8_t *)buf;
862 strm->avail_in = buf_size;
863 strm->next_out = out_buf;
864 strm->avail_out = out_buf_size;
866 ret = inflateInit2(strm, -12);
869 ret = inflate(strm, Z_FINISH);
870 out_len = strm->next_out - out_buf;
871 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
872 out_len != out_buf_size) {
880 int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
882 BDRVQcowState *s = bs->opaque;
883 int ret, csize, nb_csectors, sector_offset;
886 coffset = cluster_offset & s->cluster_offset_mask;
887 if (s->cluster_cache_offset != coffset) {
888 nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
889 sector_offset = coffset & 511;
890 csize = nb_csectors * 512 - sector_offset;
891 BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
892 ret = bdrv_read(bs->file, coffset >> 9, s->cluster_data, nb_csectors);
896 if (decompress_buffer(s->cluster_cache, s->cluster_size,
897 s->cluster_data + sector_offset, csize) < 0) {
900 s->cluster_cache_offset = coffset;
906 * This discards as many clusters of nb_clusters as possible at once (i.e.
907 * all clusters in the same L2 table) and returns the number of discarded
910 static int discard_single_l2(BlockDriverState *bs, uint64_t offset,
911 unsigned int nb_clusters)
913 BDRVQcowState *s = bs->opaque;
914 uint64_t l2_offset, *l2_table;
919 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
924 /* Limit nb_clusters to one L2 table */
925 nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
927 for (i = 0; i < nb_clusters; i++) {
930 old_offset = be64_to_cpu(l2_table[l2_index + i]);
931 old_offset &= ~QCOW_OFLAG_COPIED;
933 if (old_offset == 0) {
937 /* First remove L2 entries */
938 qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
939 l2_table[l2_index + i] = cpu_to_be64(0);
941 /* Then decrease the refcount */
942 qcow2_free_any_clusters(bs, old_offset, 1);
945 ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
953 int qcow2_discard_clusters(BlockDriverState *bs, uint64_t offset,
956 BDRVQcowState *s = bs->opaque;
958 unsigned int nb_clusters;
961 end_offset = offset + (nb_sectors << BDRV_SECTOR_BITS);
963 /* Round start up and end down */
964 offset = align_offset(offset, s->cluster_size);
965 end_offset &= ~(s->cluster_size - 1);
967 if (offset > end_offset) {
971 nb_clusters = size_to_clusters(s, end_offset - offset);
973 /* Each L2 table is handled by its own loop iteration */
974 while (nb_clusters > 0) {
975 ret = discard_single_l2(bs, offset, nb_clusters);
981 offset += (ret * s->cluster_size);