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"
31 int qcow2_grow_l1_table(BlockDriverState *bs, int min_size)
33 BDRVQcowState *s = bs->opaque;
34 int new_l1_size, new_l1_size2, ret, i;
35 uint64_t *new_l1_table;
36 uint64_t new_l1_table_offset;
39 new_l1_size = s->l1_size;
40 if (min_size <= new_l1_size)
42 while (min_size > new_l1_size) {
43 new_l1_size = (new_l1_size * 3 + 1) / 2;
46 printf("grow l1_table from %d to %d\n", s->l1_size, new_l1_size);
49 new_l1_size2 = sizeof(uint64_t) * new_l1_size;
50 new_l1_table = qemu_mallocz(align_offset(new_l1_size2, 512));
51 memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
53 /* write new table (align to cluster) */
54 new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2);
56 for(i = 0; i < s->l1_size; i++)
57 new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
58 ret = bdrv_pwrite(s->hd, new_l1_table_offset, new_l1_table, new_l1_size2);
59 if (ret != new_l1_size2)
61 for(i = 0; i < s->l1_size; i++)
62 new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
65 cpu_to_be32w((uint32_t*)data, new_l1_size);
66 cpu_to_be64w((uint64_t*)(data + 4), new_l1_table_offset);
67 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, l1_size), data,
68 sizeof(data)) != sizeof(data))
70 qemu_free(s->l1_table);
71 qcow2_free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t));
72 s->l1_table_offset = new_l1_table_offset;
73 s->l1_table = new_l1_table;
74 s->l1_size = new_l1_size;
77 qemu_free(s->l1_table);
81 void qcow2_l2_cache_reset(BlockDriverState *bs)
83 BDRVQcowState *s = bs->opaque;
85 memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
86 memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
87 memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
90 static inline int l2_cache_new_entry(BlockDriverState *bs)
92 BDRVQcowState *s = bs->opaque;
96 /* find a new entry in the least used one */
98 min_count = 0xffffffff;
99 for(i = 0; i < L2_CACHE_SIZE; i++) {
100 if (s->l2_cache_counts[i] < min_count) {
101 min_count = s->l2_cache_counts[i];
111 * seek l2_offset in the l2_cache table
112 * if not found, return NULL,
114 * increments the l2 cache hit count of the entry,
115 * if counter overflow, divide by two all counters
116 * return the pointer to the l2 cache entry
120 static uint64_t *seek_l2_table(BDRVQcowState *s, uint64_t l2_offset)
124 for(i = 0; i < L2_CACHE_SIZE; i++) {
125 if (l2_offset == s->l2_cache_offsets[i]) {
126 /* increment the hit count */
127 if (++s->l2_cache_counts[i] == 0xffffffff) {
128 for(j = 0; j < L2_CACHE_SIZE; j++) {
129 s->l2_cache_counts[j] >>= 1;
132 return s->l2_cache + (i << s->l2_bits);
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 uint64_t *l2_load(BlockDriverState *bs, uint64_t l2_offset)
150 BDRVQcowState *s = bs->opaque;
154 /* seek if the table for the given offset is in the cache */
156 l2_table = seek_l2_table(s, l2_offset);
157 if (l2_table != NULL)
160 /* not found: load a new entry in the least used one */
162 min_index = l2_cache_new_entry(bs);
163 l2_table = s->l2_cache + (min_index << s->l2_bits);
164 if (bdrv_pread(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
165 s->l2_size * sizeof(uint64_t))
167 s->l2_cache_offsets[min_index] = l2_offset;
168 s->l2_cache_counts[min_index] = 1;
174 * Writes one sector of the L1 table to the disk (can't update single entries
175 * and we really don't want bdrv_pread to perform a read-modify-write)
177 #define L1_ENTRIES_PER_SECTOR (512 / 8)
178 static int write_l1_entry(BDRVQcowState *s, int l1_index)
180 uint64_t buf[L1_ENTRIES_PER_SECTOR];
184 l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1);
185 for (i = 0; i < L1_ENTRIES_PER_SECTOR; i++) {
186 buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]);
189 if (bdrv_pwrite(s->hd, s->l1_table_offset + 8 * l1_start_index,
190 buf, sizeof(buf)) != sizeof(buf))
201 * Allocate a new l2 entry in the file. If l1_index points to an already
202 * used entry in the L2 table (i.e. we are doing a copy on write for the L2
203 * table) copy the contents of the old L2 table into the newly allocated one.
204 * Otherwise the new table is initialized with zeros.
208 static uint64_t *l2_allocate(BlockDriverState *bs, int l1_index)
210 BDRVQcowState *s = bs->opaque;
212 uint64_t old_l2_offset;
213 uint64_t *l2_table, l2_offset;
215 old_l2_offset = s->l1_table[l1_index];
217 /* allocate a new l2 entry */
219 l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
221 /* update the L1 entry */
223 s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
224 if (write_l1_entry(s, l1_index) < 0) {
228 /* allocate a new entry in the l2 cache */
230 min_index = l2_cache_new_entry(bs);
231 l2_table = s->l2_cache + (min_index << s->l2_bits);
233 if (old_l2_offset == 0) {
234 /* if there was no old l2 table, clear the new table */
235 memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
237 /* if there was an old l2 table, read it from the disk */
238 if (bdrv_pread(s->hd, old_l2_offset,
239 l2_table, s->l2_size * sizeof(uint64_t)) !=
240 s->l2_size * sizeof(uint64_t))
243 /* write the l2 table to the file */
244 if (bdrv_pwrite(s->hd, l2_offset,
245 l2_table, s->l2_size * sizeof(uint64_t)) !=
246 s->l2_size * sizeof(uint64_t))
249 /* update the l2 cache entry */
251 s->l2_cache_offsets[min_index] = l2_offset;
252 s->l2_cache_counts[min_index] = 1;
257 static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size,
258 uint64_t *l2_table, uint64_t start, uint64_t mask)
261 uint64_t offset = be64_to_cpu(l2_table[0]) & ~mask;
266 for (i = start; i < start + nb_clusters; i++)
267 if (offset + i * cluster_size != (be64_to_cpu(l2_table[i]) & ~mask))
273 static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table)
277 while(nb_clusters-- && l2_table[i] == 0)
283 /* The crypt function is compatible with the linux cryptoloop
284 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
286 void qcow2_encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
287 uint8_t *out_buf, const uint8_t *in_buf,
288 int nb_sectors, int enc,
297 for(i = 0; i < nb_sectors; i++) {
298 ivec.ll[0] = cpu_to_le64(sector_num);
300 AES_cbc_encrypt(in_buf, out_buf, 512, key,
309 static int qcow_read(BlockDriverState *bs, int64_t sector_num,
310 uint8_t *buf, int nb_sectors)
312 BDRVQcowState *s = bs->opaque;
313 int ret, index_in_cluster, n, n1;
314 uint64_t cluster_offset;
316 while (nb_sectors > 0) {
318 cluster_offset = qcow2_get_cluster_offset(bs, sector_num << 9, &n);
319 index_in_cluster = sector_num & (s->cluster_sectors - 1);
320 if (!cluster_offset) {
321 if (bs->backing_hd) {
322 /* read from the base image */
323 n1 = qcow2_backing_read1(bs->backing_hd, sector_num, buf, n);
325 ret = bdrv_read(bs->backing_hd, sector_num, buf, n1);
330 memset(buf, 0, 512 * n);
332 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
333 if (qcow2_decompress_cluster(s, cluster_offset) < 0)
335 memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
337 ret = bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
340 if (s->crypt_method) {
341 qcow2_encrypt_sectors(s, sector_num, buf, buf, n, 0,
342 &s->aes_decrypt_key);
352 static int copy_sectors(BlockDriverState *bs, uint64_t start_sect,
353 uint64_t cluster_offset, int n_start, int n_end)
355 BDRVQcowState *s = bs->opaque;
361 ret = qcow_read(bs, start_sect + n_start, s->cluster_data, n);
364 if (s->crypt_method) {
365 qcow2_encrypt_sectors(s, start_sect + n_start,
367 s->cluster_data, n, 1,
368 &s->aes_encrypt_key);
370 ret = bdrv_write(s->hd, (cluster_offset >> 9) + n_start,
381 * For a given offset of the disk image, return cluster offset in
384 * on entry, *num is the number of contiguous clusters we'd like to
385 * access following offset.
387 * on exit, *num is the number of contiguous clusters we can read.
389 * Return 1, if the offset is found
390 * Return 0, otherwise.
394 uint64_t qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
397 BDRVQcowState *s = bs->opaque;
398 int l1_index, l2_index;
399 uint64_t l2_offset, *l2_table, cluster_offset;
401 int index_in_cluster, nb_available, nb_needed, nb_clusters;
403 index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
404 nb_needed = *num + index_in_cluster;
406 l1_bits = s->l2_bits + s->cluster_bits;
408 /* compute how many bytes there are between the offset and
409 * the end of the l1 entry
412 nb_available = (1 << l1_bits) - (offset & ((1 << l1_bits) - 1));
414 /* compute the number of available sectors */
416 nb_available = (nb_available >> 9) + index_in_cluster;
418 if (nb_needed > nb_available) {
419 nb_needed = nb_available;
424 /* seek the the l2 offset in the l1 table */
426 l1_index = offset >> l1_bits;
427 if (l1_index >= s->l1_size)
430 l2_offset = s->l1_table[l1_index];
432 /* seek the l2 table of the given l2 offset */
437 /* load the l2 table in memory */
439 l2_offset &= ~QCOW_OFLAG_COPIED;
440 l2_table = l2_load(bs, l2_offset);
441 if (l2_table == NULL)
444 /* find the cluster offset for the given disk offset */
446 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
447 cluster_offset = be64_to_cpu(l2_table[l2_index]);
448 nb_clusters = size_to_clusters(s, nb_needed << 9);
450 if (!cluster_offset) {
451 /* how many empty clusters ? */
452 c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]);
454 /* how many allocated clusters ? */
455 c = count_contiguous_clusters(nb_clusters, s->cluster_size,
456 &l2_table[l2_index], 0, QCOW_OFLAG_COPIED);
459 nb_available = (c * s->cluster_sectors);
461 if (nb_available > nb_needed)
462 nb_available = nb_needed;
464 *num = nb_available - index_in_cluster;
466 return cluster_offset & ~QCOW_OFLAG_COPIED;
472 * for a given disk offset, load (and allocate if needed)
475 * the l2 table offset in the qcow2 file and the cluster index
476 * in the l2 table are given to the caller.
480 static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
481 uint64_t **new_l2_table,
482 uint64_t *new_l2_offset,
485 BDRVQcowState *s = bs->opaque;
486 int l1_index, l2_index, ret;
487 uint64_t l2_offset, *l2_table;
489 /* seek the the l2 offset in the l1 table */
491 l1_index = offset >> (s->l2_bits + s->cluster_bits);
492 if (l1_index >= s->l1_size) {
493 ret = qcow2_grow_l1_table(bs, l1_index + 1);
497 l2_offset = s->l1_table[l1_index];
499 /* seek the l2 table of the given l2 offset */
501 if (l2_offset & QCOW_OFLAG_COPIED) {
502 /* load the l2 table in memory */
503 l2_offset &= ~QCOW_OFLAG_COPIED;
504 l2_table = l2_load(bs, l2_offset);
505 if (l2_table == NULL)
509 qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
510 l2_table = l2_allocate(bs, l1_index);
511 if (l2_table == NULL)
513 l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED;
516 /* find the cluster offset for the given disk offset */
518 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
520 *new_l2_table = l2_table;
521 *new_l2_offset = l2_offset;
522 *new_l2_index = l2_index;
528 * alloc_compressed_cluster_offset
530 * For a given offset of the disk image, return cluster offset in
533 * If the offset is not found, allocate a new compressed cluster.
535 * Return the cluster offset if successful,
536 * Return 0, otherwise.
540 uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs,
544 BDRVQcowState *s = bs->opaque;
546 uint64_t l2_offset, *l2_table, cluster_offset;
549 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
553 cluster_offset = be64_to_cpu(l2_table[l2_index]);
554 if (cluster_offset & QCOW_OFLAG_COPIED)
555 return cluster_offset & ~QCOW_OFLAG_COPIED;
558 qcow2_free_any_clusters(bs, cluster_offset, 1);
560 cluster_offset = qcow2_alloc_bytes(bs, compressed_size);
561 nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
562 (cluster_offset >> 9);
564 cluster_offset |= QCOW_OFLAG_COMPRESSED |
565 ((uint64_t)nb_csectors << s->csize_shift);
567 /* update L2 table */
569 /* compressed clusters never have the copied flag */
571 l2_table[l2_index] = cpu_to_be64(cluster_offset);
572 if (bdrv_pwrite(s->hd,
573 l2_offset + l2_index * sizeof(uint64_t),
575 sizeof(uint64_t)) != sizeof(uint64_t))
578 return cluster_offset;
582 * Write L2 table updates to disk, writing whole sectors to avoid a
583 * read-modify-write in bdrv_pwrite
585 #define L2_ENTRIES_PER_SECTOR (512 / 8)
586 static int write_l2_entries(BDRVQcowState *s, uint64_t *l2_table,
587 uint64_t l2_offset, int l2_index, int num)
589 int l2_start_index = l2_index & ~(L1_ENTRIES_PER_SECTOR - 1);
590 int start_offset = (8 * l2_index) & ~511;
591 int end_offset = (8 * (l2_index + num) + 511) & ~511;
592 size_t len = end_offset - start_offset;
594 if (bdrv_pwrite(s->hd, l2_offset + start_offset, &l2_table[l2_start_index],
603 int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, uint64_t cluster_offset,
606 BDRVQcowState *s = bs->opaque;
607 int i, j = 0, l2_index, ret;
608 uint64_t *old_cluster, start_sect, l2_offset, *l2_table;
610 if (m->nb_clusters == 0)
613 old_cluster = qemu_malloc(m->nb_clusters * sizeof(uint64_t));
615 /* copy content of unmodified sectors */
616 start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9;
618 ret = copy_sectors(bs, start_sect, cluster_offset, 0, m->n_start);
623 if (m->nb_available & (s->cluster_sectors - 1)) {
624 uint64_t end = m->nb_available & ~(uint64_t)(s->cluster_sectors - 1);
625 ret = copy_sectors(bs, start_sect + end, cluster_offset + (end << 9),
626 m->nb_available - end, s->cluster_sectors);
632 /* update L2 table */
633 if (!get_cluster_table(bs, m->offset, &l2_table, &l2_offset, &l2_index))
636 for (i = 0; i < m->nb_clusters; i++) {
637 /* if two concurrent writes happen to the same unallocated cluster
638 * each write allocates separate cluster and writes data concurrently.
639 * The first one to complete updates l2 table with pointer to its
640 * cluster the second one has to do RMW (which is done above by
641 * copy_sectors()), update l2 table with its cluster pointer and free
642 * old cluster. This is what this loop does */
643 if(l2_table[l2_index + i] != 0)
644 old_cluster[j++] = l2_table[l2_index + i];
646 l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
647 (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
650 if (write_l2_entries(s, l2_table, l2_offset, l2_index, m->nb_clusters) < 0) {
655 for (i = 0; i < j; i++)
656 qcow2_free_any_clusters(bs,
657 be64_to_cpu(old_cluster[i]) & ~QCOW_OFLAG_COPIED, 1);
661 qemu_free(old_cluster);
666 * alloc_cluster_offset
668 * For a given offset of the disk image, return cluster offset in
671 * If the offset is not found, allocate a new cluster.
673 * Return the cluster offset if successful,
674 * Return 0, otherwise.
678 uint64_t qcow2_alloc_cluster_offset(BlockDriverState *bs,
680 int n_start, int n_end,
681 int *num, QCowL2Meta *m)
683 BDRVQcowState *s = bs->opaque;
685 uint64_t l2_offset, *l2_table, cluster_offset;
686 int nb_clusters, i = 0;
687 QCowL2Meta *old_alloc;
689 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
693 nb_clusters = size_to_clusters(s, n_end << 9);
695 nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
697 cluster_offset = be64_to_cpu(l2_table[l2_index]);
699 /* We keep all QCOW_OFLAG_COPIED clusters */
701 if (cluster_offset & QCOW_OFLAG_COPIED) {
702 nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size,
703 &l2_table[l2_index], 0, 0);
705 cluster_offset &= ~QCOW_OFLAG_COPIED;
711 /* for the moment, multiple compressed clusters are not managed */
713 if (cluster_offset & QCOW_OFLAG_COMPRESSED)
716 /* how many available clusters ? */
718 while (i < nb_clusters) {
719 i += count_contiguous_clusters(nb_clusters - i, s->cluster_size,
720 &l2_table[l2_index], i, 0);
722 if(be64_to_cpu(l2_table[l2_index + i]))
725 i += count_contiguous_free_clusters(nb_clusters - i,
726 &l2_table[l2_index + i]);
728 cluster_offset = be64_to_cpu(l2_table[l2_index + i]);
730 if ((cluster_offset & QCOW_OFLAG_COPIED) ||
731 (cluster_offset & QCOW_OFLAG_COMPRESSED))
737 * Check if there already is an AIO write request in flight which allocates
738 * the same cluster. In this case we need to wait until the previous
739 * request has completed and updated the L2 table accordingly.
741 LIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) {
743 uint64_t end_offset = offset + nb_clusters * s->cluster_size;
744 uint64_t old_offset = old_alloc->offset;
745 uint64_t old_end_offset = old_alloc->offset +
746 old_alloc->nb_clusters * s->cluster_size;
748 if (end_offset < old_offset || offset > old_end_offset) {
749 /* No intersection */
751 if (offset < old_offset) {
752 /* Stop at the start of a running allocation */
753 nb_clusters = (old_offset - offset) >> s->cluster_bits;
758 if (nb_clusters == 0) {
759 /* Set dependency and wait for a callback */
760 m->depends_on = old_alloc;
772 LIST_INSERT_HEAD(&s->cluster_allocs, m, next_in_flight);
774 /* allocate a new cluster */
776 cluster_offset = qcow2_alloc_clusters(bs, nb_clusters * s->cluster_size);
778 /* save info needed for meta data update */
780 m->n_start = n_start;
781 m->nb_clusters = nb_clusters;
784 m->nb_available = MIN(nb_clusters << (s->cluster_bits - 9), n_end);
786 *num = m->nb_available - n_start;
788 return cluster_offset;
791 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
792 const uint8_t *buf, int buf_size)
794 z_stream strm1, *strm = &strm1;
797 memset(strm, 0, sizeof(*strm));
799 strm->next_in = (uint8_t *)buf;
800 strm->avail_in = buf_size;
801 strm->next_out = out_buf;
802 strm->avail_out = out_buf_size;
804 ret = inflateInit2(strm, -12);
807 ret = inflate(strm, Z_FINISH);
808 out_len = strm->next_out - out_buf;
809 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
810 out_len != out_buf_size) {
818 int qcow2_decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset)
820 int ret, csize, nb_csectors, sector_offset;
823 coffset = cluster_offset & s->cluster_offset_mask;
824 if (s->cluster_cache_offset != coffset) {
825 nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
826 sector_offset = coffset & 511;
827 csize = nb_csectors * 512 - sector_offset;
828 ret = bdrv_read(s->hd, coffset >> 9, s->cluster_data, nb_csectors);
832 if (decompress_buffer(s->cluster_cache, s->cluster_size,
833 s->cluster_data + sector_offset, csize) < 0) {
836 s->cluster_cache_offset = coffset;