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
24 #include "qemu-common.h"
25 #include "block_int.h"
31 Differences with QCOW:
33 - Support for multiple incremental snapshots.
34 - Memory management by reference counts.
35 - Clusters which have a reference count of one have the bit
36 QCOW_OFLAG_COPIED to optimize write performance.
37 - Size of compressed clusters is stored in sectors to reduce bit usage
38 in the cluster offsets.
39 - Support for storing additional data (such as the VM state) in the
41 - If a backing store is used, the cluster size is not constrained
42 (could be backported to QCOW).
43 - L2 tables have always a size of one cluster.
47 //#define DEBUG_ALLOC2
49 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
50 #define QCOW_VERSION 2
52 #define QCOW_CRYPT_NONE 0
53 #define QCOW_CRYPT_AES 1
55 #define QCOW_MAX_CRYPT_CLUSTERS 32
57 /* indicate that the refcount of the referenced cluster is exactly one. */
58 #define QCOW_OFLAG_COPIED (1LL << 63)
59 /* indicate that the cluster is compressed (they never have the copied flag) */
60 #define QCOW_OFLAG_COMPRESSED (1LL << 62)
62 #define REFCOUNT_SHIFT 1 /* refcount size is 2 bytes */
64 typedef struct QCowHeader {
67 uint64_t backing_file_offset;
68 uint32_t backing_file_size;
69 uint32_t cluster_bits;
70 uint64_t size; /* in bytes */
71 uint32_t crypt_method;
72 uint32_t l1_size; /* XXX: save number of clusters instead ? */
73 uint64_t l1_table_offset;
74 uint64_t refcount_table_offset;
75 uint32_t refcount_table_clusters;
76 uint32_t nb_snapshots;
77 uint64_t snapshots_offset;
80 typedef struct __attribute__((packed)) QCowSnapshotHeader {
81 /* header is 8 byte aligned */
82 uint64_t l1_table_offset;
91 uint64_t vm_clock_nsec;
93 uint32_t vm_state_size;
94 uint32_t extra_data_size; /* for extension */
95 /* extra data follows */
100 #define L2_CACHE_SIZE 16
102 typedef struct QCowSnapshot {
103 uint64_t l1_table_offset;
107 uint32_t vm_state_size;
110 uint64_t vm_clock_nsec;
113 typedef struct BDRVQcowState {
114 BlockDriverState *hd;
121 int l1_vm_state_index;
124 uint64_t cluster_offset_mask;
125 uint64_t l1_table_offset;
128 uint64_t l2_cache_offsets[L2_CACHE_SIZE];
129 uint32_t l2_cache_counts[L2_CACHE_SIZE];
130 uint8_t *cluster_cache;
131 uint8_t *cluster_data;
132 uint64_t cluster_cache_offset;
134 uint64_t *refcount_table;
135 uint64_t refcount_table_offset;
136 uint32_t refcount_table_size;
137 uint64_t refcount_block_cache_offset;
138 uint16_t *refcount_block_cache;
139 int64_t free_cluster_index;
140 int64_t free_byte_offset;
142 uint32_t crypt_method; /* current crypt method, 0 if no key yet */
143 uint32_t crypt_method_header;
144 AES_KEY aes_encrypt_key;
145 AES_KEY aes_decrypt_key;
146 uint64_t snapshots_offset;
149 QCowSnapshot *snapshots;
152 static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset);
153 static int qcow_read(BlockDriverState *bs, int64_t sector_num,
154 uint8_t *buf, int nb_sectors);
155 static int qcow_read_snapshots(BlockDriverState *bs);
156 static void qcow_free_snapshots(BlockDriverState *bs);
157 static int refcount_init(BlockDriverState *bs);
158 static void refcount_close(BlockDriverState *bs);
159 static int get_refcount(BlockDriverState *bs, int64_t cluster_index);
160 static int update_cluster_refcount(BlockDriverState *bs,
161 int64_t cluster_index,
163 static void update_refcount(BlockDriverState *bs,
164 int64_t offset, int64_t length,
166 static int64_t alloc_clusters(BlockDriverState *bs, int64_t size);
167 static int64_t alloc_bytes(BlockDriverState *bs, int size);
168 static void free_clusters(BlockDriverState *bs,
169 int64_t offset, int64_t size);
171 static void check_refcounts(BlockDriverState *bs);
174 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
176 const QCowHeader *cow_header = (const void *)buf;
178 if (buf_size >= sizeof(QCowHeader) &&
179 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
180 be32_to_cpu(cow_header->version) == QCOW_VERSION)
186 static int qcow_open(BlockDriverState *bs, const char *filename, int flags)
188 BDRVQcowState *s = bs->opaque;
189 int len, i, shift, ret;
192 ret = bdrv_file_open(&s->hd, filename, flags);
195 if (bdrv_pread(s->hd, 0, &header, sizeof(header)) != sizeof(header))
197 be32_to_cpus(&header.magic);
198 be32_to_cpus(&header.version);
199 be64_to_cpus(&header.backing_file_offset);
200 be32_to_cpus(&header.backing_file_size);
201 be64_to_cpus(&header.size);
202 be32_to_cpus(&header.cluster_bits);
203 be32_to_cpus(&header.crypt_method);
204 be64_to_cpus(&header.l1_table_offset);
205 be32_to_cpus(&header.l1_size);
206 be64_to_cpus(&header.refcount_table_offset);
207 be32_to_cpus(&header.refcount_table_clusters);
208 be64_to_cpus(&header.snapshots_offset);
209 be32_to_cpus(&header.nb_snapshots);
211 if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
213 if (header.size <= 1 ||
214 header.cluster_bits < 9 ||
215 header.cluster_bits > 16)
217 if (header.crypt_method > QCOW_CRYPT_AES)
219 s->crypt_method_header = header.crypt_method;
220 if (s->crypt_method_header)
222 s->cluster_bits = header.cluster_bits;
223 s->cluster_size = 1 << s->cluster_bits;
224 s->cluster_sectors = 1 << (s->cluster_bits - 9);
225 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
226 s->l2_size = 1 << s->l2_bits;
227 bs->total_sectors = header.size / 512;
228 s->csize_shift = (62 - (s->cluster_bits - 8));
229 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
230 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
231 s->refcount_table_offset = header.refcount_table_offset;
232 s->refcount_table_size =
233 header.refcount_table_clusters << (s->cluster_bits - 3);
235 s->snapshots_offset = header.snapshots_offset;
236 s->nb_snapshots = header.nb_snapshots;
238 /* read the level 1 table */
239 s->l1_size = header.l1_size;
240 shift = s->cluster_bits + s->l2_bits;
241 s->l1_vm_state_index = (header.size + (1LL << shift) - 1) >> shift;
242 /* the L1 table must contain at least enough entries to put
244 if (s->l1_size < s->l1_vm_state_index)
246 s->l1_table_offset = header.l1_table_offset;
247 s->l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
250 if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
251 s->l1_size * sizeof(uint64_t))
253 for(i = 0;i < s->l1_size; i++) {
254 be64_to_cpus(&s->l1_table[i]);
257 s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
260 s->cluster_cache = qemu_malloc(s->cluster_size);
261 if (!s->cluster_cache)
263 /* one more sector for decompressed data alignment */
264 s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
266 if (!s->cluster_data)
268 s->cluster_cache_offset = -1;
270 if (refcount_init(bs) < 0)
273 /* read the backing file name */
274 if (header.backing_file_offset != 0) {
275 len = header.backing_file_size;
278 if (bdrv_pread(s->hd, header.backing_file_offset, bs->backing_file, len) != len)
280 bs->backing_file[len] = '\0';
282 if (qcow_read_snapshots(bs) < 0)
291 qcow_free_snapshots(bs);
293 qemu_free(s->l1_table);
294 qemu_free(s->l2_cache);
295 qemu_free(s->cluster_cache);
296 qemu_free(s->cluster_data);
301 static int qcow_set_key(BlockDriverState *bs, const char *key)
303 BDRVQcowState *s = bs->opaque;
307 memset(keybuf, 0, 16);
311 /* XXX: we could compress the chars to 7 bits to increase
313 for(i = 0;i < len;i++) {
316 s->crypt_method = s->crypt_method_header;
318 if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
320 if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
330 AES_encrypt(in, tmp, &s->aes_encrypt_key);
331 AES_decrypt(tmp, out, &s->aes_decrypt_key);
332 for(i = 0; i < 16; i++)
333 printf(" %02x", tmp[i]);
335 for(i = 0; i < 16; i++)
336 printf(" %02x", out[i]);
343 /* The crypt function is compatible with the linux cryptoloop
344 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
346 static void encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
347 uint8_t *out_buf, const uint8_t *in_buf,
348 int nb_sectors, int enc,
357 for(i = 0; i < nb_sectors; i++) {
358 ivec.ll[0] = cpu_to_le64(sector_num);
360 AES_cbc_encrypt(in_buf, out_buf, 512, key,
368 static int copy_sectors(BlockDriverState *bs, uint64_t start_sect,
369 uint64_t cluster_offset, int n_start, int n_end)
371 BDRVQcowState *s = bs->opaque;
377 ret = qcow_read(bs, start_sect + n_start, s->cluster_data, n);
380 if (s->crypt_method) {
381 encrypt_sectors(s, start_sect + n_start,
383 s->cluster_data, n, 1,
384 &s->aes_encrypt_key);
386 ret = bdrv_write(s->hd, (cluster_offset >> 9) + n_start,
393 static void l2_cache_reset(BlockDriverState *bs)
395 BDRVQcowState *s = bs->opaque;
397 memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
398 memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
399 memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
402 static inline int l2_cache_new_entry(BlockDriverState *bs)
404 BDRVQcowState *s = bs->opaque;
408 /* find a new entry in the least used one */
410 min_count = 0xffffffff;
411 for(i = 0; i < L2_CACHE_SIZE; i++) {
412 if (s->l2_cache_counts[i] < min_count) {
413 min_count = s->l2_cache_counts[i];
420 static int64_t align_offset(int64_t offset, int n)
422 offset = (offset + n - 1) & ~(n - 1);
426 static int grow_l1_table(BlockDriverState *bs, int min_size)
428 BDRVQcowState *s = bs->opaque;
429 int new_l1_size, new_l1_size2, ret, i;
430 uint64_t *new_l1_table;
431 uint64_t new_l1_table_offset;
434 new_l1_size = s->l1_size;
435 if (min_size <= new_l1_size)
437 while (min_size > new_l1_size) {
438 new_l1_size = (new_l1_size * 3 + 1) / 2;
441 printf("grow l1_table from %d to %d\n", s->l1_size, new_l1_size);
444 new_l1_size2 = sizeof(uint64_t) * new_l1_size;
445 new_l1_table = qemu_mallocz(new_l1_size2);
448 memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
450 /* write new table (align to cluster) */
451 new_l1_table_offset = alloc_clusters(bs, new_l1_size2);
453 for(i = 0; i < s->l1_size; i++)
454 new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
455 ret = bdrv_pwrite(s->hd, new_l1_table_offset, new_l1_table, new_l1_size2);
456 if (ret != new_l1_size2)
458 for(i = 0; i < s->l1_size; i++)
459 new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
462 cpu_to_be32w((uint32_t*)data, new_l1_size);
463 cpu_to_be64w((uint64_t*)(data + 4), new_l1_table_offset);
464 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, l1_size), data,
465 sizeof(data)) != sizeof(data))
467 qemu_free(s->l1_table);
468 free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t));
469 s->l1_table_offset = new_l1_table_offset;
470 s->l1_table = new_l1_table;
471 s->l1_size = new_l1_size;
474 qemu_free(s->l1_table);
481 * seek l2_offset in the l2_cache table
482 * if not found, return NULL,
484 * increments the l2 cache hit count of the entry,
485 * if counter overflow, divide by two all counters
486 * return the pointer to the l2 cache entry
490 static uint64_t *seek_l2_table(BDRVQcowState *s, uint64_t l2_offset)
494 for(i = 0; i < L2_CACHE_SIZE; i++) {
495 if (l2_offset == s->l2_cache_offsets[i]) {
496 /* increment the hit count */
497 if (++s->l2_cache_counts[i] == 0xffffffff) {
498 for(j = 0; j < L2_CACHE_SIZE; j++) {
499 s->l2_cache_counts[j] >>= 1;
502 return s->l2_cache + (i << s->l2_bits);
511 * Loads a L2 table into memory. If the table is in the cache, the cache
512 * is used; otherwise the L2 table is loaded from the image file.
514 * Returns a pointer to the L2 table on success, or NULL if the read from
515 * the image file failed.
518 static uint64_t *l2_load(BlockDriverState *bs, uint64_t l2_offset)
520 BDRVQcowState *s = bs->opaque;
524 /* seek if the table for the given offset is in the cache */
526 l2_table = seek_l2_table(s, l2_offset);
527 if (l2_table != NULL)
530 /* not found: load a new entry in the least used one */
532 min_index = l2_cache_new_entry(bs);
533 l2_table = s->l2_cache + (min_index << s->l2_bits);
534 if (bdrv_pread(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
535 s->l2_size * sizeof(uint64_t))
537 s->l2_cache_offsets[min_index] = l2_offset;
538 s->l2_cache_counts[min_index] = 1;
546 * Allocate a new l2 entry in the file. If l1_index points to an already
547 * used entry in the L2 table (i.e. we are doing a copy on write for the L2
548 * table) copy the contents of the old L2 table into the newly allocated one.
549 * Otherwise the new table is initialized with zeros.
553 static uint64_t *l2_allocate(BlockDriverState *bs, int l1_index)
555 BDRVQcowState *s = bs->opaque;
557 uint64_t old_l2_offset, tmp;
558 uint64_t *l2_table, l2_offset;
560 old_l2_offset = s->l1_table[l1_index];
562 /* allocate a new l2 entry */
564 l2_offset = alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
566 /* update the L1 entry */
568 s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
570 tmp = cpu_to_be64(l2_offset | QCOW_OFLAG_COPIED);
571 if (bdrv_pwrite(s->hd, s->l1_table_offset + l1_index * sizeof(tmp),
572 &tmp, sizeof(tmp)) != sizeof(tmp))
575 /* allocate a new entry in the l2 cache */
577 min_index = l2_cache_new_entry(bs);
578 l2_table = s->l2_cache + (min_index << s->l2_bits);
580 if (old_l2_offset == 0) {
581 /* if there was no old l2 table, clear the new table */
582 memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
584 /* if there was an old l2 table, read it from the disk */
585 if (bdrv_pread(s->hd, old_l2_offset,
586 l2_table, s->l2_size * sizeof(uint64_t)) !=
587 s->l2_size * sizeof(uint64_t))
590 /* write the l2 table to the file */
591 if (bdrv_pwrite(s->hd, l2_offset,
592 l2_table, s->l2_size * sizeof(uint64_t)) !=
593 s->l2_size * sizeof(uint64_t))
596 /* update the l2 cache entry */
598 s->l2_cache_offsets[min_index] = l2_offset;
599 s->l2_cache_counts[min_index] = 1;
604 static int size_to_clusters(BDRVQcowState *s, int64_t size)
606 return (size + (s->cluster_size - 1)) >> s->cluster_bits;
609 static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size,
610 uint64_t *l2_table, uint64_t mask)
613 uint64_t offset = be64_to_cpu(l2_table[0]) & ~mask;
615 for (i = 0; i < nb_clusters; i++)
616 if (offset + i * cluster_size != (be64_to_cpu(l2_table[i]) & ~mask))
622 static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table)
626 while(nb_clusters-- && l2_table[i] == 0)
635 * For a given offset of the disk image, return cluster offset in
638 * on entry, *num is the number of contiguous clusters we'd like to
639 * access following offset.
641 * on exit, *num is the number of contiguous clusters we can read.
643 * Return 1, if the offset is found
644 * Return 0, otherwise.
648 static uint64_t get_cluster_offset(BlockDriverState *bs,
649 uint64_t offset, int *num)
651 BDRVQcowState *s = bs->opaque;
652 int l1_index, l2_index;
653 uint64_t l2_offset, *l2_table, cluster_offset;
655 int index_in_cluster, nb_available, nb_needed, nb_clusters;
657 index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
658 nb_needed = *num + index_in_cluster;
660 l1_bits = s->l2_bits + s->cluster_bits;
662 /* compute how many bytes there are between the offset and
663 * the end of the l1 entry
666 nb_available = (1 << l1_bits) - (offset & ((1 << l1_bits) - 1));
668 /* compute the number of available sectors */
670 nb_available = (nb_available >> 9) + index_in_cluster;
674 /* seek the the l2 offset in the l1 table */
676 l1_index = offset >> l1_bits;
677 if (l1_index >= s->l1_size)
680 l2_offset = s->l1_table[l1_index];
682 /* seek the l2 table of the given l2 offset */
687 /* load the l2 table in memory */
689 l2_offset &= ~QCOW_OFLAG_COPIED;
690 l2_table = l2_load(bs, l2_offset);
691 if (l2_table == NULL)
694 /* find the cluster offset for the given disk offset */
696 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
697 cluster_offset = be64_to_cpu(l2_table[l2_index]);
698 nb_clusters = size_to_clusters(s, nb_needed << 9);
700 if (!cluster_offset) {
701 /* how many empty clusters ? */
702 c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]);
704 /* how many allocated clusters ? */
705 c = count_contiguous_clusters(nb_clusters, s->cluster_size,
706 &l2_table[l2_index], QCOW_OFLAG_COPIED);
709 nb_available = (c * s->cluster_sectors);
711 if (nb_available > nb_needed)
712 nb_available = nb_needed;
714 *num = nb_available - index_in_cluster;
716 return cluster_offset & ~QCOW_OFLAG_COPIED;
722 * free clusters according to its type: compressed or not
726 static void free_any_clusters(BlockDriverState *bs,
727 uint64_t cluster_offset, int nb_clusters)
729 BDRVQcowState *s = bs->opaque;
731 /* free the cluster */
733 if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
735 nb_csectors = ((cluster_offset >> s->csize_shift) &
737 free_clusters(bs, (cluster_offset & s->cluster_offset_mask) & ~511,
742 free_clusters(bs, cluster_offset, nb_clusters << s->cluster_bits);
750 * for a given disk offset, load (and allocate if needed)
753 * the l2 table offset in the qcow2 file and the cluster index
754 * in the l2 table are given to the caller.
758 static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
759 uint64_t **new_l2_table,
760 uint64_t *new_l2_offset,
763 BDRVQcowState *s = bs->opaque;
764 int l1_index, l2_index, ret;
765 uint64_t l2_offset, *l2_table;
767 /* seek the the l2 offset in the l1 table */
769 l1_index = offset >> (s->l2_bits + s->cluster_bits);
770 if (l1_index >= s->l1_size) {
771 ret = grow_l1_table(bs, l1_index + 1);
775 l2_offset = s->l1_table[l1_index];
777 /* seek the l2 table of the given l2 offset */
779 if (l2_offset & QCOW_OFLAG_COPIED) {
780 /* load the l2 table in memory */
781 l2_offset &= ~QCOW_OFLAG_COPIED;
782 l2_table = l2_load(bs, l2_offset);
783 if (l2_table == NULL)
787 free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
788 l2_table = l2_allocate(bs, l1_index);
789 if (l2_table == NULL)
791 l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED;
794 /* find the cluster offset for the given disk offset */
796 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
798 *new_l2_table = l2_table;
799 *new_l2_offset = l2_offset;
800 *new_l2_index = l2_index;
806 * alloc_compressed_cluster_offset
808 * For a given offset of the disk image, return cluster offset in
811 * If the offset is not found, allocate a new compressed cluster.
813 * Return the cluster offset if successful,
814 * Return 0, otherwise.
818 static uint64_t alloc_compressed_cluster_offset(BlockDriverState *bs,
822 BDRVQcowState *s = bs->opaque;
824 uint64_t l2_offset, *l2_table, cluster_offset;
827 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
831 cluster_offset = be64_to_cpu(l2_table[l2_index]);
832 if (cluster_offset & QCOW_OFLAG_COPIED)
833 return cluster_offset & ~QCOW_OFLAG_COPIED;
836 free_any_clusters(bs, cluster_offset, 1);
838 cluster_offset = alloc_bytes(bs, compressed_size);
839 nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
840 (cluster_offset >> 9);
842 cluster_offset |= QCOW_OFLAG_COMPRESSED |
843 ((uint64_t)nb_csectors << s->csize_shift);
845 /* update L2 table */
847 /* compressed clusters never have the copied flag */
849 l2_table[l2_index] = cpu_to_be64(cluster_offset);
850 if (bdrv_pwrite(s->hd,
851 l2_offset + l2_index * sizeof(uint64_t),
853 sizeof(uint64_t)) != sizeof(uint64_t))
856 return cluster_offset;
860 * alloc_cluster_offset
862 * For a given offset of the disk image, return cluster offset in
865 * If the offset is not found, allocate a new cluster.
867 * Return the cluster offset if successful,
868 * Return 0, otherwise.
872 static uint64_t alloc_cluster_offset(BlockDriverState *bs,
874 int n_start, int n_end,
877 BDRVQcowState *s = bs->opaque;
879 uint64_t l2_offset, *l2_table, cluster_offset;
880 int nb_available, nb_clusters, i = 0;
883 ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
887 nb_clusters = size_to_clusters(s, n_end << 9);
889 if (nb_clusters > s->l2_size - l2_index)
890 nb_clusters = s->l2_size - l2_index;
892 cluster_offset = be64_to_cpu(l2_table[l2_index]);
894 /* We keep all QCOW_OFLAG_COPIED clusters */
896 if (cluster_offset & QCOW_OFLAG_COPIED) {
897 nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size,
898 &l2_table[l2_index], 0);
900 nb_available = nb_clusters << (s->cluster_bits - 9);
901 if (nb_available > n_end)
902 nb_available = n_end;
904 cluster_offset &= ~QCOW_OFLAG_COPIED;
909 /* for the moment, multiple compressed clusters are not managed */
911 if (cluster_offset & QCOW_OFLAG_COMPRESSED)
914 /* how many available clusters ? */
916 while (i < nb_clusters) {
918 i += count_contiguous_free_clusters(nb_clusters - i,
919 &l2_table[l2_index + i]);
921 cluster_offset = be64_to_cpu(l2_table[l2_index + i]);
923 if ((cluster_offset & QCOW_OFLAG_COPIED) ||
924 (cluster_offset & QCOW_OFLAG_COMPRESSED))
927 j = count_contiguous_clusters(nb_clusters - i, s->cluster_size,
928 &l2_table[l2_index + i], 0);
931 free_any_clusters(bs, cluster_offset, j);
935 if(be64_to_cpu(l2_table[l2_index + i]))
940 /* allocate a new cluster */
942 cluster_offset = alloc_clusters(bs, nb_clusters * s->cluster_size);
944 /* we must initialize the cluster content which won't be
947 nb_available = nb_clusters << (s->cluster_bits - 9);
948 if (nb_available > n_end)
949 nb_available = n_end;
951 /* copy content of unmodified sectors */
953 start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
955 ret = copy_sectors(bs, start_sect, cluster_offset, 0, n_start);
960 if (nb_available & (s->cluster_sectors - 1)) {
961 uint64_t end = nb_available & ~(uint64_t)(s->cluster_sectors - 1);
962 ret = copy_sectors(bs, start_sect + end,
963 cluster_offset + (end << 9),
970 /* update L2 table */
972 for (i = 0; i < nb_clusters; i++)
973 l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
974 (i << s->cluster_bits)) |
977 if (bdrv_pwrite(s->hd,
978 l2_offset + l2_index * sizeof(uint64_t),
980 nb_clusters * sizeof(uint64_t)) !=
981 nb_clusters * sizeof(uint64_t))
985 *num = nb_available - n_start;
987 return cluster_offset;
990 static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
991 int nb_sectors, int *pnum)
993 uint64_t cluster_offset;
996 cluster_offset = get_cluster_offset(bs, sector_num << 9, pnum);
998 return (cluster_offset != 0);
1001 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
1002 const uint8_t *buf, int buf_size)
1004 z_stream strm1, *strm = &strm1;
1007 memset(strm, 0, sizeof(*strm));
1009 strm->next_in = (uint8_t *)buf;
1010 strm->avail_in = buf_size;
1011 strm->next_out = out_buf;
1012 strm->avail_out = out_buf_size;
1014 ret = inflateInit2(strm, -12);
1017 ret = inflate(strm, Z_FINISH);
1018 out_len = strm->next_out - out_buf;
1019 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
1020 out_len != out_buf_size) {
1028 static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset)
1030 int ret, csize, nb_csectors, sector_offset;
1033 coffset = cluster_offset & s->cluster_offset_mask;
1034 if (s->cluster_cache_offset != coffset) {
1035 nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
1036 sector_offset = coffset & 511;
1037 csize = nb_csectors * 512 - sector_offset;
1038 ret = bdrv_read(s->hd, coffset >> 9, s->cluster_data, nb_csectors);
1042 if (decompress_buffer(s->cluster_cache, s->cluster_size,
1043 s->cluster_data + sector_offset, csize) < 0) {
1046 s->cluster_cache_offset = coffset;
1051 /* handle reading after the end of the backing file */
1052 static int backing_read1(BlockDriverState *bs,
1053 int64_t sector_num, uint8_t *buf, int nb_sectors)
1056 if ((sector_num + nb_sectors) <= bs->total_sectors)
1058 if (sector_num >= bs->total_sectors)
1061 n1 = bs->total_sectors - sector_num;
1062 memset(buf + n1 * 512, 0, 512 * (nb_sectors - n1));
1066 static int qcow_read(BlockDriverState *bs, int64_t sector_num,
1067 uint8_t *buf, int nb_sectors)
1069 BDRVQcowState *s = bs->opaque;
1070 int ret, index_in_cluster, n, n1;
1071 uint64_t cluster_offset;
1073 while (nb_sectors > 0) {
1075 cluster_offset = get_cluster_offset(bs, sector_num << 9, &n);
1076 index_in_cluster = sector_num & (s->cluster_sectors - 1);
1077 if (!cluster_offset) {
1078 if (bs->backing_hd) {
1079 /* read from the base image */
1080 n1 = backing_read1(bs->backing_hd, sector_num, buf, n);
1082 ret = bdrv_read(bs->backing_hd, sector_num, buf, n1);
1087 memset(buf, 0, 512 * n);
1089 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
1090 if (decompress_cluster(s, cluster_offset) < 0)
1092 memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
1094 ret = bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
1097 if (s->crypt_method) {
1098 encrypt_sectors(s, sector_num, buf, buf, n, 0,
1099 &s->aes_decrypt_key);
1109 static int qcow_write(BlockDriverState *bs, int64_t sector_num,
1110 const uint8_t *buf, int nb_sectors)
1112 BDRVQcowState *s = bs->opaque;
1113 int ret, index_in_cluster, n;
1114 uint64_t cluster_offset;
1117 while (nb_sectors > 0) {
1118 index_in_cluster = sector_num & (s->cluster_sectors - 1);
1119 n_end = index_in_cluster + nb_sectors;
1120 if (s->crypt_method &&
1121 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
1122 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
1123 cluster_offset = alloc_cluster_offset(bs, sector_num << 9,
1126 if (!cluster_offset)
1128 if (s->crypt_method) {
1129 encrypt_sectors(s, sector_num, s->cluster_data, buf, n, 1,
1130 &s->aes_encrypt_key);
1131 ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512,
1132 s->cluster_data, n * 512);
1134 ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
1142 s->cluster_cache_offset = -1; /* disable compressed cache */
1146 typedef struct QCowAIOCB {
1147 BlockDriverAIOCB common;
1152 uint64_t cluster_offset;
1153 uint8_t *cluster_data;
1154 BlockDriverAIOCB *hd_aiocb;
1158 static void qcow_aio_read_cb(void *opaque, int ret);
1159 static void qcow_aio_read_bh(void *opaque)
1161 QCowAIOCB *acb = opaque;
1162 qemu_bh_delete(acb->bh);
1164 qcow_aio_read_cb(opaque, 0);
1167 static int qcow_schedule_bh(QEMUBHFunc *cb, QCowAIOCB *acb)
1172 acb->bh = qemu_bh_new(cb, acb);
1176 qemu_bh_schedule(acb->bh);
1181 static void qcow_aio_read_cb(void *opaque, int ret)
1183 QCowAIOCB *acb = opaque;
1184 BlockDriverState *bs = acb->common.bs;
1185 BDRVQcowState *s = bs->opaque;
1186 int index_in_cluster, n1;
1188 acb->hd_aiocb = NULL;
1191 acb->common.cb(acb->common.opaque, ret);
1192 qemu_aio_release(acb);
1196 /* post process the read buffer */
1197 if (!acb->cluster_offset) {
1199 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
1202 if (s->crypt_method) {
1203 encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
1205 &s->aes_decrypt_key);
1209 acb->nb_sectors -= acb->n;
1210 acb->sector_num += acb->n;
1211 acb->buf += acb->n * 512;
1213 if (acb->nb_sectors == 0) {
1214 /* request completed */
1215 acb->common.cb(acb->common.opaque, 0);
1216 qemu_aio_release(acb);
1220 /* prepare next AIO request */
1221 acb->n = acb->nb_sectors;
1222 acb->cluster_offset = get_cluster_offset(bs, acb->sector_num << 9, &acb->n);
1223 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
1225 if (!acb->cluster_offset) {
1226 if (bs->backing_hd) {
1227 /* read from the base image */
1228 n1 = backing_read1(bs->backing_hd, acb->sector_num,
1231 acb->hd_aiocb = bdrv_aio_read(bs->backing_hd, acb->sector_num,
1232 acb->buf, acb->n, qcow_aio_read_cb, acb);
1233 if (acb->hd_aiocb == NULL)
1236 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1241 /* Note: in this case, no need to wait */
1242 memset(acb->buf, 0, 512 * acb->n);
1243 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1247 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
1248 /* add AIO support for compressed blocks ? */
1249 if (decompress_cluster(s, acb->cluster_offset) < 0)
1252 s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
1253 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
1257 if ((acb->cluster_offset & 511) != 0) {
1261 acb->hd_aiocb = bdrv_aio_read(s->hd,
1262 (acb->cluster_offset >> 9) + index_in_cluster,
1263 acb->buf, acb->n, qcow_aio_read_cb, acb);
1264 if (acb->hd_aiocb == NULL)
1269 static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
1270 int64_t sector_num, uint8_t *buf, int nb_sectors,
1271 BlockDriverCompletionFunc *cb, void *opaque)
1275 acb = qemu_aio_get(bs, cb, opaque);
1278 acb->hd_aiocb = NULL;
1279 acb->sector_num = sector_num;
1281 acb->nb_sectors = nb_sectors;
1283 acb->cluster_offset = 0;
1287 static BlockDriverAIOCB *qcow_aio_read(BlockDriverState *bs,
1288 int64_t sector_num, uint8_t *buf, int nb_sectors,
1289 BlockDriverCompletionFunc *cb, void *opaque)
1293 acb = qcow_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
1297 qcow_aio_read_cb(acb, 0);
1298 return &acb->common;
1301 static void qcow_aio_write_cb(void *opaque, int ret)
1303 QCowAIOCB *acb = opaque;
1304 BlockDriverState *bs = acb->common.bs;
1305 BDRVQcowState *s = bs->opaque;
1306 int index_in_cluster;
1307 uint64_t cluster_offset;
1308 const uint8_t *src_buf;
1311 acb->hd_aiocb = NULL;
1315 acb->common.cb(acb->common.opaque, ret);
1316 qemu_aio_release(acb);
1320 acb->nb_sectors -= acb->n;
1321 acb->sector_num += acb->n;
1322 acb->buf += acb->n * 512;
1324 if (acb->nb_sectors == 0) {
1325 /* request completed */
1326 acb->common.cb(acb->common.opaque, 0);
1327 qemu_aio_release(acb);
1331 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
1332 n_end = index_in_cluster + acb->nb_sectors;
1333 if (s->crypt_method &&
1334 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
1335 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
1337 cluster_offset = alloc_cluster_offset(bs, acb->sector_num << 9,
1340 if (!cluster_offset || (cluster_offset & 511) != 0) {
1344 if (s->crypt_method) {
1345 if (!acb->cluster_data) {
1346 acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS *
1348 if (!acb->cluster_data) {
1353 encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
1354 acb->n, 1, &s->aes_encrypt_key);
1355 src_buf = acb->cluster_data;
1359 acb->hd_aiocb = bdrv_aio_write(s->hd,
1360 (cluster_offset >> 9) + index_in_cluster,
1362 qcow_aio_write_cb, acb);
1363 if (acb->hd_aiocb == NULL)
1367 static BlockDriverAIOCB *qcow_aio_write(BlockDriverState *bs,
1368 int64_t sector_num, const uint8_t *buf, int nb_sectors,
1369 BlockDriverCompletionFunc *cb, void *opaque)
1371 BDRVQcowState *s = bs->opaque;
1374 s->cluster_cache_offset = -1; /* disable compressed cache */
1376 acb = qcow_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
1380 qcow_aio_write_cb(acb, 0);
1381 return &acb->common;
1384 static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
1386 QCowAIOCB *acb = (QCowAIOCB *)blockacb;
1388 bdrv_aio_cancel(acb->hd_aiocb);
1389 qemu_aio_release(acb);
1392 static void qcow_close(BlockDriverState *bs)
1394 BDRVQcowState *s = bs->opaque;
1395 qemu_free(s->l1_table);
1396 qemu_free(s->l2_cache);
1397 qemu_free(s->cluster_cache);
1398 qemu_free(s->cluster_data);
1403 /* XXX: use std qcow open function ? */
1404 typedef struct QCowCreateState {
1407 uint16_t *refcount_block;
1408 uint64_t *refcount_table;
1409 int64_t l1_table_offset;
1410 int64_t refcount_table_offset;
1411 int64_t refcount_block_offset;
1414 static void create_refcount_update(QCowCreateState *s,
1415 int64_t offset, int64_t size)
1418 int64_t start, last, cluster_offset;
1421 start = offset & ~(s->cluster_size - 1);
1422 last = (offset + size - 1) & ~(s->cluster_size - 1);
1423 for(cluster_offset = start; cluster_offset <= last;
1424 cluster_offset += s->cluster_size) {
1425 p = &s->refcount_block[cluster_offset >> s->cluster_bits];
1426 refcount = be16_to_cpu(*p);
1428 *p = cpu_to_be16(refcount);
1432 static int qcow_create(const char *filename, int64_t total_size,
1433 const char *backing_file, int flags)
1435 int fd, header_size, backing_filename_len, l1_size, i, shift, l2_bits;
1437 uint64_t tmp, offset;
1438 QCowCreateState s1, *s = &s1;
1440 memset(s, 0, sizeof(*s));
1442 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
1445 memset(&header, 0, sizeof(header));
1446 header.magic = cpu_to_be32(QCOW_MAGIC);
1447 header.version = cpu_to_be32(QCOW_VERSION);
1448 header.size = cpu_to_be64(total_size * 512);
1449 header_size = sizeof(header);
1450 backing_filename_len = 0;
1452 header.backing_file_offset = cpu_to_be64(header_size);
1453 backing_filename_len = strlen(backing_file);
1454 header.backing_file_size = cpu_to_be32(backing_filename_len);
1455 header_size += backing_filename_len;
1457 s->cluster_bits = 12; /* 4 KB clusters */
1458 s->cluster_size = 1 << s->cluster_bits;
1459 header.cluster_bits = cpu_to_be32(s->cluster_bits);
1460 header_size = (header_size + 7) & ~7;
1461 if (flags & BLOCK_FLAG_ENCRYPT) {
1462 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
1464 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
1466 l2_bits = s->cluster_bits - 3;
1467 shift = s->cluster_bits + l2_bits;
1468 l1_size = (((total_size * 512) + (1LL << shift) - 1) >> shift);
1469 offset = align_offset(header_size, s->cluster_size);
1470 s->l1_table_offset = offset;
1471 header.l1_table_offset = cpu_to_be64(s->l1_table_offset);
1472 header.l1_size = cpu_to_be32(l1_size);
1473 offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size);
1475 s->refcount_table = qemu_mallocz(s->cluster_size);
1476 if (!s->refcount_table)
1478 s->refcount_block = qemu_mallocz(s->cluster_size);
1479 if (!s->refcount_block)
1482 s->refcount_table_offset = offset;
1483 header.refcount_table_offset = cpu_to_be64(offset);
1484 header.refcount_table_clusters = cpu_to_be32(1);
1485 offset += s->cluster_size;
1487 s->refcount_table[0] = cpu_to_be64(offset);
1488 s->refcount_block_offset = offset;
1489 offset += s->cluster_size;
1491 /* update refcounts */
1492 create_refcount_update(s, 0, header_size);
1493 create_refcount_update(s, s->l1_table_offset, l1_size * sizeof(uint64_t));
1494 create_refcount_update(s, s->refcount_table_offset, s->cluster_size);
1495 create_refcount_update(s, s->refcount_block_offset, s->cluster_size);
1497 /* write all the data */
1498 write(fd, &header, sizeof(header));
1500 write(fd, backing_file, backing_filename_len);
1502 lseek(fd, s->l1_table_offset, SEEK_SET);
1504 for(i = 0;i < l1_size; i++) {
1505 write(fd, &tmp, sizeof(tmp));
1507 lseek(fd, s->refcount_table_offset, SEEK_SET);
1508 write(fd, s->refcount_table, s->cluster_size);
1510 lseek(fd, s->refcount_block_offset, SEEK_SET);
1511 write(fd, s->refcount_block, s->cluster_size);
1513 qemu_free(s->refcount_table);
1514 qemu_free(s->refcount_block);
1518 qemu_free(s->refcount_table);
1519 qemu_free(s->refcount_block);
1524 static int qcow_make_empty(BlockDriverState *bs)
1527 /* XXX: not correct */
1528 BDRVQcowState *s = bs->opaque;
1529 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1532 memset(s->l1_table, 0, l1_length);
1533 if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
1535 ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
1544 /* XXX: put compressed sectors first, then all the cluster aligned
1545 tables to avoid losing bytes in alignment */
1546 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
1547 const uint8_t *buf, int nb_sectors)
1549 BDRVQcowState *s = bs->opaque;
1553 uint64_t cluster_offset;
1555 if (nb_sectors == 0) {
1556 /* align end of file to a sector boundary to ease reading with
1557 sector based I/Os */
1558 cluster_offset = bdrv_getlength(s->hd);
1559 cluster_offset = (cluster_offset + 511) & ~511;
1560 bdrv_truncate(s->hd, cluster_offset);
1564 if (nb_sectors != s->cluster_sectors)
1567 out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1571 /* best compression, small window, no zlib header */
1572 memset(&strm, 0, sizeof(strm));
1573 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1575 9, Z_DEFAULT_STRATEGY);
1581 strm.avail_in = s->cluster_size;
1582 strm.next_in = (uint8_t *)buf;
1583 strm.avail_out = s->cluster_size;
1584 strm.next_out = out_buf;
1586 ret = deflate(&strm, Z_FINISH);
1587 if (ret != Z_STREAM_END && ret != Z_OK) {
1592 out_len = strm.next_out - out_buf;
1596 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1597 /* could not compress: write normal cluster */
1598 qcow_write(bs, sector_num, buf, s->cluster_sectors);
1600 cluster_offset = alloc_compressed_cluster_offset(bs, sector_num << 9,
1602 if (!cluster_offset)
1604 cluster_offset &= s->cluster_offset_mask;
1605 if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
1615 static void qcow_flush(BlockDriverState *bs)
1617 BDRVQcowState *s = bs->opaque;
1621 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1623 BDRVQcowState *s = bs->opaque;
1624 bdi->cluster_size = s->cluster_size;
1625 bdi->vm_state_offset = (int64_t)s->l1_vm_state_index <<
1626 (s->cluster_bits + s->l2_bits);
1630 /*********************************************************/
1631 /* snapshot support */
1633 /* update the refcounts of snapshots and the copied flag */
1634 static int update_snapshot_refcount(BlockDriverState *bs,
1635 int64_t l1_table_offset,
1639 BDRVQcowState *s = bs->opaque;
1640 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
1641 int64_t old_offset, old_l2_offset;
1642 int l2_size, i, j, l1_modified, l2_modified, nb_csectors, refcount;
1648 l1_size2 = l1_size * sizeof(uint64_t);
1650 if (l1_table_offset != s->l1_table_offset) {
1651 l1_table = qemu_malloc(l1_size2);
1655 if (bdrv_pread(s->hd, l1_table_offset,
1656 l1_table, l1_size2) != l1_size2)
1658 for(i = 0;i < l1_size; i++)
1659 be64_to_cpus(&l1_table[i]);
1661 assert(l1_size == s->l1_size);
1662 l1_table = s->l1_table;
1666 l2_size = s->l2_size * sizeof(uint64_t);
1667 l2_table = qemu_malloc(l2_size);
1671 for(i = 0; i < l1_size; i++) {
1672 l2_offset = l1_table[i];
1674 old_l2_offset = l2_offset;
1675 l2_offset &= ~QCOW_OFLAG_COPIED;
1677 if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
1679 for(j = 0; j < s->l2_size; j++) {
1680 offset = be64_to_cpu(l2_table[j]);
1682 old_offset = offset;
1683 offset &= ~QCOW_OFLAG_COPIED;
1684 if (offset & QCOW_OFLAG_COMPRESSED) {
1685 nb_csectors = ((offset >> s->csize_shift) &
1688 update_refcount(bs, (offset & s->cluster_offset_mask) & ~511,
1689 nb_csectors * 512, addend);
1690 /* compressed clusters are never modified */
1694 refcount = update_cluster_refcount(bs, offset >> s->cluster_bits, addend);
1696 refcount = get_refcount(bs, offset >> s->cluster_bits);
1700 if (refcount == 1) {
1701 offset |= QCOW_OFLAG_COPIED;
1703 if (offset != old_offset) {
1704 l2_table[j] = cpu_to_be64(offset);
1710 if (bdrv_pwrite(s->hd,
1711 l2_offset, l2_table, l2_size) != l2_size)
1716 refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend);
1718 refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
1720 if (refcount == 1) {
1721 l2_offset |= QCOW_OFLAG_COPIED;
1723 if (l2_offset != old_l2_offset) {
1724 l1_table[i] = l2_offset;
1730 for(i = 0; i < l1_size; i++)
1731 cpu_to_be64s(&l1_table[i]);
1732 if (bdrv_pwrite(s->hd, l1_table_offset, l1_table,
1733 l1_size2) != l1_size2)
1735 for(i = 0; i < l1_size; i++)
1736 be64_to_cpus(&l1_table[i]);
1739 qemu_free(l1_table);
1740 qemu_free(l2_table);
1744 qemu_free(l1_table);
1745 qemu_free(l2_table);
1749 static void qcow_free_snapshots(BlockDriverState *bs)
1751 BDRVQcowState *s = bs->opaque;
1754 for(i = 0; i < s->nb_snapshots; i++) {
1755 qemu_free(s->snapshots[i].name);
1756 qemu_free(s->snapshots[i].id_str);
1758 qemu_free(s->snapshots);
1759 s->snapshots = NULL;
1760 s->nb_snapshots = 0;
1763 static int qcow_read_snapshots(BlockDriverState *bs)
1765 BDRVQcowState *s = bs->opaque;
1766 QCowSnapshotHeader h;
1768 int i, id_str_size, name_size;
1770 uint32_t extra_data_size;
1772 offset = s->snapshots_offset;
1773 s->snapshots = qemu_mallocz(s->nb_snapshots * sizeof(QCowSnapshot));
1776 for(i = 0; i < s->nb_snapshots; i++) {
1777 offset = align_offset(offset, 8);
1778 if (bdrv_pread(s->hd, offset, &h, sizeof(h)) != sizeof(h))
1780 offset += sizeof(h);
1781 sn = s->snapshots + i;
1782 sn->l1_table_offset = be64_to_cpu(h.l1_table_offset);
1783 sn->l1_size = be32_to_cpu(h.l1_size);
1784 sn->vm_state_size = be32_to_cpu(h.vm_state_size);
1785 sn->date_sec = be32_to_cpu(h.date_sec);
1786 sn->date_nsec = be32_to_cpu(h.date_nsec);
1787 sn->vm_clock_nsec = be64_to_cpu(h.vm_clock_nsec);
1788 extra_data_size = be32_to_cpu(h.extra_data_size);
1790 id_str_size = be16_to_cpu(h.id_str_size);
1791 name_size = be16_to_cpu(h.name_size);
1793 offset += extra_data_size;
1795 sn->id_str = qemu_malloc(id_str_size + 1);
1798 if (bdrv_pread(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
1800 offset += id_str_size;
1801 sn->id_str[id_str_size] = '\0';
1803 sn->name = qemu_malloc(name_size + 1);
1806 if (bdrv_pread(s->hd, offset, sn->name, name_size) != name_size)
1808 offset += name_size;
1809 sn->name[name_size] = '\0';
1811 s->snapshots_size = offset - s->snapshots_offset;
1814 qcow_free_snapshots(bs);
1818 /* add at the end of the file a new list of snapshots */
1819 static int qcow_write_snapshots(BlockDriverState *bs)
1821 BDRVQcowState *s = bs->opaque;
1823 QCowSnapshotHeader h;
1824 int i, name_size, id_str_size, snapshots_size;
1827 int64_t offset, snapshots_offset;
1829 /* compute the size of the snapshots */
1831 for(i = 0; i < s->nb_snapshots; i++) {
1832 sn = s->snapshots + i;
1833 offset = align_offset(offset, 8);
1834 offset += sizeof(h);
1835 offset += strlen(sn->id_str);
1836 offset += strlen(sn->name);
1838 snapshots_size = offset;
1840 snapshots_offset = alloc_clusters(bs, snapshots_size);
1841 offset = snapshots_offset;
1843 for(i = 0; i < s->nb_snapshots; i++) {
1844 sn = s->snapshots + i;
1845 memset(&h, 0, sizeof(h));
1846 h.l1_table_offset = cpu_to_be64(sn->l1_table_offset);
1847 h.l1_size = cpu_to_be32(sn->l1_size);
1848 h.vm_state_size = cpu_to_be32(sn->vm_state_size);
1849 h.date_sec = cpu_to_be32(sn->date_sec);
1850 h.date_nsec = cpu_to_be32(sn->date_nsec);
1851 h.vm_clock_nsec = cpu_to_be64(sn->vm_clock_nsec);
1853 id_str_size = strlen(sn->id_str);
1854 name_size = strlen(sn->name);
1855 h.id_str_size = cpu_to_be16(id_str_size);
1856 h.name_size = cpu_to_be16(name_size);
1857 offset = align_offset(offset, 8);
1858 if (bdrv_pwrite(s->hd, offset, &h, sizeof(h)) != sizeof(h))
1860 offset += sizeof(h);
1861 if (bdrv_pwrite(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
1863 offset += id_str_size;
1864 if (bdrv_pwrite(s->hd, offset, sn->name, name_size) != name_size)
1866 offset += name_size;
1869 /* update the various header fields */
1870 data64 = cpu_to_be64(snapshots_offset);
1871 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, snapshots_offset),
1872 &data64, sizeof(data64)) != sizeof(data64))
1874 data32 = cpu_to_be32(s->nb_snapshots);
1875 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, nb_snapshots),
1876 &data32, sizeof(data32)) != sizeof(data32))
1879 /* free the old snapshot table */
1880 free_clusters(bs, s->snapshots_offset, s->snapshots_size);
1881 s->snapshots_offset = snapshots_offset;
1882 s->snapshots_size = snapshots_size;
1888 static void find_new_snapshot_id(BlockDriverState *bs,
1889 char *id_str, int id_str_size)
1891 BDRVQcowState *s = bs->opaque;
1893 int i, id, id_max = 0;
1895 for(i = 0; i < s->nb_snapshots; i++) {
1896 sn = s->snapshots + i;
1897 id = strtoul(sn->id_str, NULL, 10);
1901 snprintf(id_str, id_str_size, "%d", id_max + 1);
1904 static int find_snapshot_by_id(BlockDriverState *bs, const char *id_str)
1906 BDRVQcowState *s = bs->opaque;
1909 for(i = 0; i < s->nb_snapshots; i++) {
1910 if (!strcmp(s->snapshots[i].id_str, id_str))
1916 static int find_snapshot_by_id_or_name(BlockDriverState *bs, const char *name)
1918 BDRVQcowState *s = bs->opaque;
1921 ret = find_snapshot_by_id(bs, name);
1924 for(i = 0; i < s->nb_snapshots; i++) {
1925 if (!strcmp(s->snapshots[i].name, name))
1931 /* if no id is provided, a new one is constructed */
1932 static int qcow_snapshot_create(BlockDriverState *bs,
1933 QEMUSnapshotInfo *sn_info)
1935 BDRVQcowState *s = bs->opaque;
1936 QCowSnapshot *snapshots1, sn1, *sn = &sn1;
1938 uint64_t *l1_table = NULL;
1940 memset(sn, 0, sizeof(*sn));
1942 if (sn_info->id_str[0] == '\0') {
1943 /* compute a new id */
1944 find_new_snapshot_id(bs, sn_info->id_str, sizeof(sn_info->id_str));
1947 /* check that the ID is unique */
1948 if (find_snapshot_by_id(bs, sn_info->id_str) >= 0)
1951 sn->id_str = qemu_strdup(sn_info->id_str);
1954 sn->name = qemu_strdup(sn_info->name);
1957 sn->vm_state_size = sn_info->vm_state_size;
1958 sn->date_sec = sn_info->date_sec;
1959 sn->date_nsec = sn_info->date_nsec;
1960 sn->vm_clock_nsec = sn_info->vm_clock_nsec;
1962 ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1);
1966 /* create the L1 table of the snapshot */
1967 sn->l1_table_offset = alloc_clusters(bs, s->l1_size * sizeof(uint64_t));
1968 sn->l1_size = s->l1_size;
1970 l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
1973 for(i = 0; i < s->l1_size; i++) {
1974 l1_table[i] = cpu_to_be64(s->l1_table[i]);
1976 if (bdrv_pwrite(s->hd, sn->l1_table_offset,
1977 l1_table, s->l1_size * sizeof(uint64_t)) !=
1978 (s->l1_size * sizeof(uint64_t)))
1980 qemu_free(l1_table);
1983 snapshots1 = qemu_malloc((s->nb_snapshots + 1) * sizeof(QCowSnapshot));
1986 memcpy(snapshots1, s->snapshots, s->nb_snapshots * sizeof(QCowSnapshot));
1987 s->snapshots = snapshots1;
1988 s->snapshots[s->nb_snapshots++] = *sn;
1990 if (qcow_write_snapshots(bs) < 0)
1993 check_refcounts(bs);
1997 qemu_free(sn->name);
1998 qemu_free(l1_table);
2002 /* copy the snapshot 'snapshot_name' into the current disk image */
2003 static int qcow_snapshot_goto(BlockDriverState *bs,
2004 const char *snapshot_id)
2006 BDRVQcowState *s = bs->opaque;
2008 int i, snapshot_index, l1_size2;
2010 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
2011 if (snapshot_index < 0)
2013 sn = &s->snapshots[snapshot_index];
2015 if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, -1) < 0)
2018 if (grow_l1_table(bs, sn->l1_size) < 0)
2021 s->l1_size = sn->l1_size;
2022 l1_size2 = s->l1_size * sizeof(uint64_t);
2023 /* copy the snapshot l1 table to the current l1 table */
2024 if (bdrv_pread(s->hd, sn->l1_table_offset,
2025 s->l1_table, l1_size2) != l1_size2)
2027 if (bdrv_pwrite(s->hd, s->l1_table_offset,
2028 s->l1_table, l1_size2) != l1_size2)
2030 for(i = 0;i < s->l1_size; i++) {
2031 be64_to_cpus(&s->l1_table[i]);
2034 if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1) < 0)
2038 check_refcounts(bs);
2045 static int qcow_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2047 BDRVQcowState *s = bs->opaque;
2049 int snapshot_index, ret;
2051 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
2052 if (snapshot_index < 0)
2054 sn = &s->snapshots[snapshot_index];
2056 ret = update_snapshot_refcount(bs, sn->l1_table_offset, sn->l1_size, -1);
2059 /* must update the copied flag on the current cluster offsets */
2060 ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
2063 free_clusters(bs, sn->l1_table_offset, sn->l1_size * sizeof(uint64_t));
2065 qemu_free(sn->id_str);
2066 qemu_free(sn->name);
2067 memmove(sn, sn + 1, (s->nb_snapshots - snapshot_index - 1) * sizeof(*sn));
2069 ret = qcow_write_snapshots(bs);
2071 /* XXX: restore snapshot if error ? */
2075 check_refcounts(bs);
2080 static int qcow_snapshot_list(BlockDriverState *bs,
2081 QEMUSnapshotInfo **psn_tab)
2083 BDRVQcowState *s = bs->opaque;
2084 QEMUSnapshotInfo *sn_tab, *sn_info;
2088 sn_tab = qemu_mallocz(s->nb_snapshots * sizeof(QEMUSnapshotInfo));
2091 for(i = 0; i < s->nb_snapshots; i++) {
2092 sn_info = sn_tab + i;
2093 sn = s->snapshots + i;
2094 pstrcpy(sn_info->id_str, sizeof(sn_info->id_str),
2096 pstrcpy(sn_info->name, sizeof(sn_info->name),
2098 sn_info->vm_state_size = sn->vm_state_size;
2099 sn_info->date_sec = sn->date_sec;
2100 sn_info->date_nsec = sn->date_nsec;
2101 sn_info->vm_clock_nsec = sn->vm_clock_nsec;
2104 return s->nb_snapshots;
2111 /*********************************************************/
2112 /* refcount handling */
2114 static int refcount_init(BlockDriverState *bs)
2116 BDRVQcowState *s = bs->opaque;
2117 int ret, refcount_table_size2, i;
2119 s->refcount_block_cache = qemu_malloc(s->cluster_size);
2120 if (!s->refcount_block_cache)
2122 refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
2123 s->refcount_table = qemu_malloc(refcount_table_size2);
2124 if (!s->refcount_table)
2126 if (s->refcount_table_size > 0) {
2127 ret = bdrv_pread(s->hd, s->refcount_table_offset,
2128 s->refcount_table, refcount_table_size2);
2129 if (ret != refcount_table_size2)
2131 for(i = 0; i < s->refcount_table_size; i++)
2132 be64_to_cpus(&s->refcount_table[i]);
2139 static void refcount_close(BlockDriverState *bs)
2141 BDRVQcowState *s = bs->opaque;
2142 qemu_free(s->refcount_block_cache);
2143 qemu_free(s->refcount_table);
2147 static int load_refcount_block(BlockDriverState *bs,
2148 int64_t refcount_block_offset)
2150 BDRVQcowState *s = bs->opaque;
2152 ret = bdrv_pread(s->hd, refcount_block_offset, s->refcount_block_cache,
2154 if (ret != s->cluster_size)
2156 s->refcount_block_cache_offset = refcount_block_offset;
2160 static int get_refcount(BlockDriverState *bs, int64_t cluster_index)
2162 BDRVQcowState *s = bs->opaque;
2163 int refcount_table_index, block_index;
2164 int64_t refcount_block_offset;
2166 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
2167 if (refcount_table_index >= s->refcount_table_size)
2169 refcount_block_offset = s->refcount_table[refcount_table_index];
2170 if (!refcount_block_offset)
2172 if (refcount_block_offset != s->refcount_block_cache_offset) {
2173 /* better than nothing: return allocated if read error */
2174 if (load_refcount_block(bs, refcount_block_offset) < 0)
2177 block_index = cluster_index &
2178 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2179 return be16_to_cpu(s->refcount_block_cache[block_index]);
2182 /* return < 0 if error */
2183 static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
2185 BDRVQcowState *s = bs->opaque;
2188 nb_clusters = size_to_clusters(s, size);
2190 for(i = 0; i < nb_clusters; i++) {
2191 int64_t i = s->free_cluster_index++;
2192 if (get_refcount(bs, i) != 0)
2196 printf("alloc_clusters: size=%lld -> %lld\n",
2198 (s->free_cluster_index - nb_clusters) << s->cluster_bits);
2200 return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
2203 static int64_t alloc_clusters(BlockDriverState *bs, int64_t size)
2207 offset = alloc_clusters_noref(bs, size);
2208 update_refcount(bs, offset, size, 1);
2212 /* only used to allocate compressed sectors. We try to allocate
2213 contiguous sectors. size must be <= cluster_size */
2214 static int64_t alloc_bytes(BlockDriverState *bs, int size)
2216 BDRVQcowState *s = bs->opaque;
2217 int64_t offset, cluster_offset;
2218 int free_in_cluster;
2220 assert(size > 0 && size <= s->cluster_size);
2221 if (s->free_byte_offset == 0) {
2222 s->free_byte_offset = alloc_clusters(bs, s->cluster_size);
2225 free_in_cluster = s->cluster_size -
2226 (s->free_byte_offset & (s->cluster_size - 1));
2227 if (size <= free_in_cluster) {
2228 /* enough space in current cluster */
2229 offset = s->free_byte_offset;
2230 s->free_byte_offset += size;
2231 free_in_cluster -= size;
2232 if (free_in_cluster == 0)
2233 s->free_byte_offset = 0;
2234 if ((offset & (s->cluster_size - 1)) != 0)
2235 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
2237 offset = alloc_clusters(bs, s->cluster_size);
2238 cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
2239 if ((cluster_offset + s->cluster_size) == offset) {
2240 /* we are lucky: contiguous data */
2241 offset = s->free_byte_offset;
2242 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
2243 s->free_byte_offset += size;
2245 s->free_byte_offset = offset;
2252 static void free_clusters(BlockDriverState *bs,
2253 int64_t offset, int64_t size)
2255 update_refcount(bs, offset, size, -1);
2258 static int grow_refcount_table(BlockDriverState *bs, int min_size)
2260 BDRVQcowState *s = bs->opaque;
2261 int new_table_size, new_table_size2, refcount_table_clusters, i, ret;
2262 uint64_t *new_table;
2263 int64_t table_offset;
2266 int64_t old_table_offset;
2268 if (min_size <= s->refcount_table_size)
2270 /* compute new table size */
2271 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2273 if (refcount_table_clusters == 0) {
2274 refcount_table_clusters = 1;
2276 refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
2278 new_table_size = refcount_table_clusters << (s->cluster_bits - 3);
2279 if (min_size <= new_table_size)
2283 printf("grow_refcount_table from %d to %d\n",
2284 s->refcount_table_size,
2287 new_table_size2 = new_table_size * sizeof(uint64_t);
2288 new_table = qemu_mallocz(new_table_size2);
2291 memcpy(new_table, s->refcount_table,
2292 s->refcount_table_size * sizeof(uint64_t));
2293 for(i = 0; i < s->refcount_table_size; i++)
2294 cpu_to_be64s(&new_table[i]);
2295 /* Note: we cannot update the refcount now to avoid recursion */
2296 table_offset = alloc_clusters_noref(bs, new_table_size2);
2297 ret = bdrv_pwrite(s->hd, table_offset, new_table, new_table_size2);
2298 if (ret != new_table_size2)
2300 for(i = 0; i < s->refcount_table_size; i++)
2301 be64_to_cpus(&new_table[i]);
2303 cpu_to_be64w((uint64_t*)data, table_offset);
2304 cpu_to_be32w((uint32_t*)(data + 8), refcount_table_clusters);
2305 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, refcount_table_offset),
2306 data, sizeof(data)) != sizeof(data))
2308 qemu_free(s->refcount_table);
2309 old_table_offset = s->refcount_table_offset;
2310 old_table_size = s->refcount_table_size;
2311 s->refcount_table = new_table;
2312 s->refcount_table_size = new_table_size;
2313 s->refcount_table_offset = table_offset;
2315 update_refcount(bs, table_offset, new_table_size2, 1);
2316 free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
2319 free_clusters(bs, table_offset, new_table_size2);
2320 qemu_free(new_table);
2324 /* addend must be 1 or -1 */
2325 /* XXX: cache several refcount block clusters ? */
2326 static int update_cluster_refcount(BlockDriverState *bs,
2327 int64_t cluster_index,
2330 BDRVQcowState *s = bs->opaque;
2331 int64_t offset, refcount_block_offset;
2332 int ret, refcount_table_index, block_index, refcount;
2335 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
2336 if (refcount_table_index >= s->refcount_table_size) {
2339 ret = grow_refcount_table(bs, refcount_table_index + 1);
2343 refcount_block_offset = s->refcount_table[refcount_table_index];
2344 if (!refcount_block_offset) {
2347 /* create a new refcount block */
2348 /* Note: we cannot update the refcount now to avoid recursion */
2349 offset = alloc_clusters_noref(bs, s->cluster_size);
2350 memset(s->refcount_block_cache, 0, s->cluster_size);
2351 ret = bdrv_pwrite(s->hd, offset, s->refcount_block_cache, s->cluster_size);
2352 if (ret != s->cluster_size)
2354 s->refcount_table[refcount_table_index] = offset;
2355 data64 = cpu_to_be64(offset);
2356 ret = bdrv_pwrite(s->hd, s->refcount_table_offset +
2357 refcount_table_index * sizeof(uint64_t),
2358 &data64, sizeof(data64));
2359 if (ret != sizeof(data64))
2362 refcount_block_offset = offset;
2363 s->refcount_block_cache_offset = offset;
2364 update_refcount(bs, offset, s->cluster_size, 1);
2366 if (refcount_block_offset != s->refcount_block_cache_offset) {
2367 if (load_refcount_block(bs, refcount_block_offset) < 0)
2371 /* we can update the count and save it */
2372 block_index = cluster_index &
2373 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2374 refcount = be16_to_cpu(s->refcount_block_cache[block_index]);
2376 if (refcount < 0 || refcount > 0xffff)
2378 if (refcount == 0 && cluster_index < s->free_cluster_index) {
2379 s->free_cluster_index = cluster_index;
2381 s->refcount_block_cache[block_index] = cpu_to_be16(refcount);
2382 if (bdrv_pwrite(s->hd,
2383 refcount_block_offset + (block_index << REFCOUNT_SHIFT),
2384 &s->refcount_block_cache[block_index], 2) != 2)
2389 static void update_refcount(BlockDriverState *bs,
2390 int64_t offset, int64_t length,
2393 BDRVQcowState *s = bs->opaque;
2394 int64_t start, last, cluster_offset;
2397 printf("update_refcount: offset=%lld size=%lld addend=%d\n",
2398 offset, length, addend);
2402 start = offset & ~(s->cluster_size - 1);
2403 last = (offset + length - 1) & ~(s->cluster_size - 1);
2404 for(cluster_offset = start; cluster_offset <= last;
2405 cluster_offset += s->cluster_size) {
2406 update_cluster_refcount(bs, cluster_offset >> s->cluster_bits, addend);
2411 static void inc_refcounts(BlockDriverState *bs,
2412 uint16_t *refcount_table,
2413 int refcount_table_size,
2414 int64_t offset, int64_t size)
2416 BDRVQcowState *s = bs->opaque;
2417 int64_t start, last, cluster_offset;
2423 start = offset & ~(s->cluster_size - 1);
2424 last = (offset + size - 1) & ~(s->cluster_size - 1);
2425 for(cluster_offset = start; cluster_offset <= last;
2426 cluster_offset += s->cluster_size) {
2427 k = cluster_offset >> s->cluster_bits;
2428 if (k < 0 || k >= refcount_table_size) {
2429 printf("ERROR: invalid cluster offset=0x%llx\n", cluster_offset);
2431 if (++refcount_table[k] == 0) {
2432 printf("ERROR: overflow cluster offset=0x%llx\n", cluster_offset);
2438 static int check_refcounts_l1(BlockDriverState *bs,
2439 uint16_t *refcount_table,
2440 int refcount_table_size,
2441 int64_t l1_table_offset, int l1_size,
2444 BDRVQcowState *s = bs->opaque;
2445 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2;
2446 int l2_size, i, j, nb_csectors, refcount;
2449 l1_size2 = l1_size * sizeof(uint64_t);
2451 inc_refcounts(bs, refcount_table, refcount_table_size,
2452 l1_table_offset, l1_size2);
2454 l1_table = qemu_malloc(l1_size2);
2457 if (bdrv_pread(s->hd, l1_table_offset,
2458 l1_table, l1_size2) != l1_size2)
2460 for(i = 0;i < l1_size; i++)
2461 be64_to_cpus(&l1_table[i]);
2463 l2_size = s->l2_size * sizeof(uint64_t);
2464 l2_table = qemu_malloc(l2_size);
2467 for(i = 0; i < l1_size; i++) {
2468 l2_offset = l1_table[i];
2471 refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED) >> s->cluster_bits);
2472 if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
2473 printf("ERROR OFLAG_COPIED: l2_offset=%llx refcount=%d\n",
2474 l2_offset, refcount);
2477 l2_offset &= ~QCOW_OFLAG_COPIED;
2478 if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
2480 for(j = 0; j < s->l2_size; j++) {
2481 offset = be64_to_cpu(l2_table[j]);
2483 if (offset & QCOW_OFLAG_COMPRESSED) {
2484 if (offset & QCOW_OFLAG_COPIED) {
2485 printf("ERROR: cluster %lld: copied flag must never be set for compressed clusters\n",
2486 offset >> s->cluster_bits);
2487 offset &= ~QCOW_OFLAG_COPIED;
2489 nb_csectors = ((offset >> s->csize_shift) &
2491 offset &= s->cluster_offset_mask;
2492 inc_refcounts(bs, refcount_table,
2493 refcount_table_size,
2494 offset & ~511, nb_csectors * 512);
2497 refcount = get_refcount(bs, (offset & ~QCOW_OFLAG_COPIED) >> s->cluster_bits);
2498 if ((refcount == 1) != ((offset & QCOW_OFLAG_COPIED) != 0)) {
2499 printf("ERROR OFLAG_COPIED: offset=%llx refcount=%d\n",
2503 offset &= ~QCOW_OFLAG_COPIED;
2504 inc_refcounts(bs, refcount_table,
2505 refcount_table_size,
2506 offset, s->cluster_size);
2510 inc_refcounts(bs, refcount_table,
2511 refcount_table_size,
2516 qemu_free(l1_table);
2517 qemu_free(l2_table);
2520 printf("ERROR: I/O error in check_refcounts_l1\n");
2521 qemu_free(l1_table);
2522 qemu_free(l2_table);
2526 static void check_refcounts(BlockDriverState *bs)
2528 BDRVQcowState *s = bs->opaque;
2530 int nb_clusters, refcount1, refcount2, i;
2532 uint16_t *refcount_table;
2534 size = bdrv_getlength(s->hd);
2535 nb_clusters = size_to_clusters(s, size);
2536 refcount_table = qemu_mallocz(nb_clusters * sizeof(uint16_t));
2539 inc_refcounts(bs, refcount_table, nb_clusters,
2540 0, s->cluster_size);
2542 check_refcounts_l1(bs, refcount_table, nb_clusters,
2543 s->l1_table_offset, s->l1_size, 1);
2546 for(i = 0; i < s->nb_snapshots; i++) {
2547 sn = s->snapshots + i;
2548 check_refcounts_l1(bs, refcount_table, nb_clusters,
2549 sn->l1_table_offset, sn->l1_size, 0);
2551 inc_refcounts(bs, refcount_table, nb_clusters,
2552 s->snapshots_offset, s->snapshots_size);
2555 inc_refcounts(bs, refcount_table, nb_clusters,
2556 s->refcount_table_offset,
2557 s->refcount_table_size * sizeof(uint64_t));
2558 for(i = 0; i < s->refcount_table_size; i++) {
2560 offset = s->refcount_table[i];
2562 inc_refcounts(bs, refcount_table, nb_clusters,
2563 offset, s->cluster_size);
2567 /* compare ref counts */
2568 for(i = 0; i < nb_clusters; i++) {
2569 refcount1 = get_refcount(bs, i);
2570 refcount2 = refcount_table[i];
2571 if (refcount1 != refcount2)
2572 printf("ERROR cluster %d refcount=%d reference=%d\n",
2573 i, refcount1, refcount2);
2576 qemu_free(refcount_table);
2580 static void dump_refcounts(BlockDriverState *bs)
2582 BDRVQcowState *s = bs->opaque;
2583 int64_t nb_clusters, k, k1, size;
2586 size = bdrv_getlength(s->hd);
2587 nb_clusters = size_to_clusters(s, size);
2588 for(k = 0; k < nb_clusters;) {
2590 refcount = get_refcount(bs, k);
2592 while (k < nb_clusters && get_refcount(bs, k) == refcount)
2594 printf("%lld: refcount=%d nb=%lld\n", k, refcount, k - k1);
2600 BlockDriver bdrv_qcow2 = {
2602 sizeof(BDRVQcowState),
2614 .bdrv_aio_read = qcow_aio_read,
2615 .bdrv_aio_write = qcow_aio_write,
2616 .bdrv_aio_cancel = qcow_aio_cancel,
2617 .aiocb_size = sizeof(QCowAIOCB),
2618 .bdrv_write_compressed = qcow_write_compressed,
2620 .bdrv_snapshot_create = qcow_snapshot_create,
2621 .bdrv_snapshot_goto = qcow_snapshot_goto,
2622 .bdrv_snapshot_delete = qcow_snapshot_delete,
2623 .bdrv_snapshot_list = qcow_snapshot_list,
2624 .bdrv_get_info = qcow_get_info,