2 * Block driver for the QCOW version 2 format
4 * Copyright (c) 2004-2006 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "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 /* indicate that the refcount of the referenced cluster is exactly one. */
56 #define QCOW_OFLAG_COPIED (1LL << 63)
57 /* indicate that the cluster is compressed (they never have the copied flag) */
58 #define QCOW_OFLAG_COMPRESSED (1LL << 62)
60 #define REFCOUNT_SHIFT 1 /* refcount size is 2 bytes */
63 #define offsetof(type, field) ((size_t) &((type *)0)->field)
66 typedef struct QCowHeader {
69 uint64_t backing_file_offset;
70 uint32_t backing_file_size;
71 uint32_t cluster_bits;
72 uint64_t size; /* in bytes */
73 uint32_t crypt_method;
74 uint32_t l1_size; /* XXX: save number of clusters instead ? */
75 uint64_t l1_table_offset;
76 uint64_t refcount_table_offset;
77 uint32_t refcount_table_clusters;
78 uint32_t nb_snapshots;
79 uint64_t snapshots_offset;
82 typedef struct __attribute__((packed)) QCowSnapshotHeader {
83 /* header is 8 byte aligned */
84 uint64_t l1_table_offset;
93 uint64_t vm_clock_nsec;
95 uint32_t vm_state_size;
96 uint32_t extra_data_size; /* for extension */
97 /* extra data follows */
100 } QCowSnapshotHeader;
102 #define L2_CACHE_SIZE 16
104 typedef struct QCowSnapshot {
105 uint64_t l1_table_offset;
109 uint32_t vm_state_size;
112 uint64_t vm_clock_nsec;
115 typedef struct BDRVQcowState {
116 BlockDriverState *hd;
123 int l1_vm_state_index;
126 uint64_t cluster_offset_mask;
127 uint64_t l1_table_offset;
130 uint64_t l2_cache_offsets[L2_CACHE_SIZE];
131 uint32_t l2_cache_counts[L2_CACHE_SIZE];
132 uint8_t *cluster_cache;
133 uint8_t *cluster_data;
134 uint64_t cluster_cache_offset;
136 uint64_t *refcount_table;
137 uint64_t refcount_table_offset;
138 uint32_t refcount_table_size;
139 uint64_t refcount_block_cache_offset;
140 uint16_t *refcount_block_cache;
141 int64_t free_cluster_index;
142 int64_t free_byte_offset;
144 uint32_t crypt_method; /* current crypt method, 0 if no key yet */
145 uint32_t crypt_method_header;
146 AES_KEY aes_encrypt_key;
147 AES_KEY aes_decrypt_key;
148 uint64_t snapshots_offset;
151 QCowSnapshot *snapshots;
154 static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset);
155 static int qcow_read(BlockDriverState *bs, int64_t sector_num,
156 uint8_t *buf, int nb_sectors);
157 static int qcow_read_snapshots(BlockDriverState *bs);
158 static void qcow_free_snapshots(BlockDriverState *bs);
159 static int refcount_init(BlockDriverState *bs);
160 static void refcount_close(BlockDriverState *bs);
161 static int get_refcount(BlockDriverState *bs, int64_t cluster_index);
162 static int update_cluster_refcount(BlockDriverState *bs,
163 int64_t cluster_index,
165 static void update_refcount(BlockDriverState *bs,
166 int64_t offset, int64_t length,
168 static int64_t alloc_clusters(BlockDriverState *bs, int64_t size);
169 static int64_t alloc_bytes(BlockDriverState *bs, int size);
170 static void free_clusters(BlockDriverState *bs,
171 int64_t offset, int64_t size);
173 static void check_refcounts(BlockDriverState *bs);
176 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
178 const QCowHeader *cow_header = (const void *)buf;
180 if (buf_size >= sizeof(QCowHeader) &&
181 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
182 be32_to_cpu(cow_header->version) == QCOW_VERSION)
188 static int qcow_open(BlockDriverState *bs, const char *filename, int flags)
190 BDRVQcowState *s = bs->opaque;
191 int len, i, shift, ret;
194 ret = bdrv_file_open(&s->hd, filename, flags);
197 if (bdrv_pread(s->hd, 0, &header, sizeof(header)) != sizeof(header))
199 be32_to_cpus(&header.magic);
200 be32_to_cpus(&header.version);
201 be64_to_cpus(&header.backing_file_offset);
202 be32_to_cpus(&header.backing_file_size);
203 be64_to_cpus(&header.size);
204 be32_to_cpus(&header.cluster_bits);
205 be32_to_cpus(&header.crypt_method);
206 be64_to_cpus(&header.l1_table_offset);
207 be32_to_cpus(&header.l1_size);
208 be64_to_cpus(&header.refcount_table_offset);
209 be32_to_cpus(&header.refcount_table_clusters);
210 be64_to_cpus(&header.snapshots_offset);
211 be32_to_cpus(&header.nb_snapshots);
213 if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
215 if (header.size <= 1 ||
216 header.cluster_bits < 9 ||
217 header.cluster_bits > 16)
219 if (header.crypt_method > QCOW_CRYPT_AES)
221 s->crypt_method_header = header.crypt_method;
222 if (s->crypt_method_header)
224 s->cluster_bits = header.cluster_bits;
225 s->cluster_size = 1 << s->cluster_bits;
226 s->cluster_sectors = 1 << (s->cluster_bits - 9);
227 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
228 s->l2_size = 1 << s->l2_bits;
229 bs->total_sectors = header.size / 512;
230 s->csize_shift = (62 - (s->cluster_bits - 8));
231 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
232 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
233 s->refcount_table_offset = header.refcount_table_offset;
234 s->refcount_table_size =
235 header.refcount_table_clusters << (s->cluster_bits - 3);
237 s->snapshots_offset = header.snapshots_offset;
238 s->nb_snapshots = header.nb_snapshots;
240 /* read the level 1 table */
241 s->l1_size = header.l1_size;
242 shift = s->cluster_bits + s->l2_bits;
243 s->l1_vm_state_index = (header.size + (1LL << shift) - 1) >> shift;
244 /* the L1 table must contain at least enough entries to put
246 if (s->l1_size < s->l1_vm_state_index)
248 s->l1_table_offset = header.l1_table_offset;
249 s->l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
252 if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
253 s->l1_size * sizeof(uint64_t))
255 for(i = 0;i < s->l1_size; i++) {
256 be64_to_cpus(&s->l1_table[i]);
259 s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
262 s->cluster_cache = qemu_malloc(s->cluster_size);
263 if (!s->cluster_cache)
265 /* one more sector for decompressed data alignment */
266 s->cluster_data = qemu_malloc(s->cluster_size + 512);
267 if (!s->cluster_data)
269 s->cluster_cache_offset = -1;
271 if (refcount_init(bs) < 0)
274 /* read the backing file name */
275 if (header.backing_file_offset != 0) {
276 len = header.backing_file_size;
279 if (bdrv_pread(s->hd, header.backing_file_offset, bs->backing_file, len) != len)
281 bs->backing_file[len] = '\0';
283 if (qcow_read_snapshots(bs) < 0)
292 qcow_free_snapshots(bs);
294 qemu_free(s->l1_table);
295 qemu_free(s->l2_cache);
296 qemu_free(s->cluster_cache);
297 qemu_free(s->cluster_data);
302 static int qcow_set_key(BlockDriverState *bs, const char *key)
304 BDRVQcowState *s = bs->opaque;
308 memset(keybuf, 0, 16);
312 /* XXX: we could compress the chars to 7 bits to increase
314 for(i = 0;i < len;i++) {
317 s->crypt_method = s->crypt_method_header;
319 if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
321 if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
331 AES_encrypt(in, tmp, &s->aes_encrypt_key);
332 AES_decrypt(tmp, out, &s->aes_decrypt_key);
333 for(i = 0; i < 16; i++)
334 printf(" %02x", tmp[i]);
336 for(i = 0; i < 16; i++)
337 printf(" %02x", out[i]);
344 /* The crypt function is compatible with the linux cryptoloop
345 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
347 static void encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
348 uint8_t *out_buf, const uint8_t *in_buf,
349 int nb_sectors, int enc,
358 for(i = 0; i < nb_sectors; i++) {
359 ivec.ll[0] = cpu_to_le64(sector_num);
361 AES_cbc_encrypt(in_buf, out_buf, 512, key,
369 static int copy_sectors(BlockDriverState *bs, uint64_t start_sect,
370 uint64_t cluster_offset, int n_start, int n_end)
372 BDRVQcowState *s = bs->opaque;
378 ret = qcow_read(bs, start_sect + n_start, s->cluster_data, n);
381 if (s->crypt_method) {
382 encrypt_sectors(s, start_sect + n_start,
384 s->cluster_data, n, 1,
385 &s->aes_encrypt_key);
387 ret = bdrv_write(s->hd, (cluster_offset >> 9) + n_start,
394 static void l2_cache_reset(BlockDriverState *bs)
396 BDRVQcowState *s = bs->opaque;
398 memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
399 memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
400 memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
403 static inline int l2_cache_new_entry(BlockDriverState *bs)
405 BDRVQcowState *s = bs->opaque;
409 /* find a new entry in the least used one */
411 min_count = 0xffffffff;
412 for(i = 0; i < L2_CACHE_SIZE; i++) {
413 if (s->l2_cache_counts[i] < min_count) {
414 min_count = s->l2_cache_counts[i];
421 static int64_t align_offset(int64_t offset, int n)
423 offset = (offset + n - 1) & ~(n - 1);
427 static int grow_l1_table(BlockDriverState *bs, int min_size)
429 BDRVQcowState *s = bs->opaque;
430 int new_l1_size, new_l1_size2, ret, i;
431 uint64_t *new_l1_table;
432 uint64_t new_l1_table_offset;
436 new_l1_size = s->l1_size;
437 if (min_size <= new_l1_size)
439 while (min_size > new_l1_size) {
440 new_l1_size = (new_l1_size * 3 + 1) / 2;
443 printf("grow l1_table from %d to %d\n", s->l1_size, new_l1_size);
446 new_l1_size2 = sizeof(uint64_t) * new_l1_size;
447 new_l1_table = qemu_mallocz(new_l1_size2);
450 memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
452 /* write new table (align to cluster) */
453 new_l1_table_offset = alloc_clusters(bs, new_l1_size2);
455 for(i = 0; i < s->l1_size; i++)
456 new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
457 ret = bdrv_pwrite(s->hd, new_l1_table_offset, new_l1_table, new_l1_size2);
458 if (ret != new_l1_size2)
460 for(i = 0; i < s->l1_size; i++)
461 new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
464 data64 = cpu_to_be64(new_l1_table_offset);
465 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, l1_table_offset),
466 &data64, sizeof(data64)) != sizeof(data64))
468 data32 = cpu_to_be32(new_l1_size);
469 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, l1_size),
470 &data32, sizeof(data32)) != sizeof(data32))
472 qemu_free(s->l1_table);
473 free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t));
474 s->l1_table_offset = new_l1_table_offset;
475 s->l1_table = new_l1_table;
476 s->l1_size = new_l1_size;
479 qemu_free(s->l1_table);
487 * 1 to allocate a normal cluster (for sector indexes 'n_start' to
490 * 2 to allocate a compressed cluster of size
491 * 'compressed_size'. 'compressed_size' must be > 0 and <
494 * return 0 if not allocated.
496 static uint64_t get_cluster_offset(BlockDriverState *bs,
497 uint64_t offset, int allocate,
499 int n_start, int n_end)
501 BDRVQcowState *s = bs->opaque;
502 int min_index, i, j, l1_index, l2_index, ret;
503 uint64_t l2_offset, *l2_table, cluster_offset, tmp, old_l2_offset;
505 l1_index = offset >> (s->l2_bits + s->cluster_bits);
506 if (l1_index >= s->l1_size) {
507 /* outside l1 table is allowed: we grow the table if needed */
510 if (grow_l1_table(bs, l1_index + 1) < 0)
513 l2_offset = s->l1_table[l1_index];
518 old_l2_offset = l2_offset;
519 /* allocate a new l2 entry */
520 l2_offset = alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
521 /* update the L1 entry */
522 s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
523 tmp = cpu_to_be64(l2_offset | QCOW_OFLAG_COPIED);
524 if (bdrv_pwrite(s->hd, s->l1_table_offset + l1_index * sizeof(tmp),
525 &tmp, sizeof(tmp)) != sizeof(tmp))
527 min_index = l2_cache_new_entry(bs);
528 l2_table = s->l2_cache + (min_index << s->l2_bits);
530 if (old_l2_offset == 0) {
531 memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
533 if (bdrv_pread(s->hd, old_l2_offset,
534 l2_table, s->l2_size * sizeof(uint64_t)) !=
535 s->l2_size * sizeof(uint64_t))
538 if (bdrv_pwrite(s->hd, l2_offset,
539 l2_table, s->l2_size * sizeof(uint64_t)) !=
540 s->l2_size * sizeof(uint64_t))
543 if (!(l2_offset & QCOW_OFLAG_COPIED)) {
545 free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
549 l2_offset &= ~QCOW_OFLAG_COPIED;
551 for(i = 0; i < L2_CACHE_SIZE; i++) {
552 if (l2_offset == s->l2_cache_offsets[i]) {
553 /* increment the hit count */
554 if (++s->l2_cache_counts[i] == 0xffffffff) {
555 for(j = 0; j < L2_CACHE_SIZE; j++) {
556 s->l2_cache_counts[j] >>= 1;
559 l2_table = s->l2_cache + (i << s->l2_bits);
563 /* not found: load a new entry in the least used one */
564 min_index = l2_cache_new_entry(bs);
565 l2_table = s->l2_cache + (min_index << s->l2_bits);
566 if (bdrv_pread(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
567 s->l2_size * sizeof(uint64_t))
570 s->l2_cache_offsets[min_index] = l2_offset;
571 s->l2_cache_counts[min_index] = 1;
573 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
574 cluster_offset = be64_to_cpu(l2_table[l2_index]);
575 if (!cluster_offset) {
577 return cluster_offset;
578 } else if (!(cluster_offset & QCOW_OFLAG_COPIED)) {
580 return cluster_offset;
581 /* free the cluster */
582 if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
584 nb_csectors = ((cluster_offset >> s->csize_shift) &
586 free_clusters(bs, (cluster_offset & s->cluster_offset_mask) & ~511,
589 free_clusters(bs, cluster_offset, s->cluster_size);
592 cluster_offset &= ~QCOW_OFLAG_COPIED;
593 return cluster_offset;
596 /* allocate a new cluster */
597 cluster_offset = alloc_clusters(bs, s->cluster_size);
599 /* we must initialize the cluster content which won't be
601 if ((n_end - n_start) < s->cluster_sectors) {
604 start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
605 ret = copy_sectors(bs, start_sect,
606 cluster_offset, 0, n_start);
609 ret = copy_sectors(bs, start_sect,
610 cluster_offset, n_end, s->cluster_sectors);
614 tmp = cpu_to_be64(cluster_offset | QCOW_OFLAG_COPIED);
617 cluster_offset = alloc_bytes(bs, compressed_size);
618 nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
619 (cluster_offset >> 9);
620 cluster_offset |= QCOW_OFLAG_COMPRESSED |
621 ((uint64_t)nb_csectors << s->csize_shift);
622 /* compressed clusters never have the copied flag */
623 tmp = cpu_to_be64(cluster_offset);
625 /* update L2 table */
626 l2_table[l2_index] = tmp;
627 if (bdrv_pwrite(s->hd,
628 l2_offset + l2_index * sizeof(tmp), &tmp, sizeof(tmp)) != sizeof(tmp))
630 return cluster_offset;
633 static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
634 int nb_sectors, int *pnum)
636 BDRVQcowState *s = bs->opaque;
637 int index_in_cluster, n;
638 uint64_t cluster_offset;
640 cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
641 index_in_cluster = sector_num & (s->cluster_sectors - 1);
642 n = s->cluster_sectors - index_in_cluster;
646 return (cluster_offset != 0);
649 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
650 const uint8_t *buf, int buf_size)
652 z_stream strm1, *strm = &strm1;
655 memset(strm, 0, sizeof(*strm));
657 strm->next_in = (uint8_t *)buf;
658 strm->avail_in = buf_size;
659 strm->next_out = out_buf;
660 strm->avail_out = out_buf_size;
662 ret = inflateInit2(strm, -12);
665 ret = inflate(strm, Z_FINISH);
666 out_len = strm->next_out - out_buf;
667 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
668 out_len != out_buf_size) {
676 static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset)
678 int ret, csize, nb_csectors, sector_offset;
681 coffset = cluster_offset & s->cluster_offset_mask;
682 if (s->cluster_cache_offset != coffset) {
683 nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
684 sector_offset = coffset & 511;
685 csize = nb_csectors * 512 - sector_offset;
686 ret = bdrv_read(s->hd, coffset >> 9, s->cluster_data, nb_csectors);
690 if (decompress_buffer(s->cluster_cache, s->cluster_size,
691 s->cluster_data + sector_offset, csize) < 0) {
694 s->cluster_cache_offset = coffset;
699 /* handle reading after the end of the backing file */
700 static int backing_read1(BlockDriverState *bs,
701 int64_t sector_num, uint8_t *buf, int nb_sectors)
704 if ((sector_num + nb_sectors) <= bs->total_sectors)
706 if (sector_num >= bs->total_sectors)
709 n1 = bs->total_sectors - sector_num;
710 memset(buf + n1 * 512, 0, 512 * (nb_sectors - n1));
714 static int qcow_read(BlockDriverState *bs, int64_t sector_num,
715 uint8_t *buf, int nb_sectors)
717 BDRVQcowState *s = bs->opaque;
718 int ret, index_in_cluster, n, n1;
719 uint64_t cluster_offset;
721 while (nb_sectors > 0) {
722 cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
723 index_in_cluster = sector_num & (s->cluster_sectors - 1);
724 n = s->cluster_sectors - index_in_cluster;
727 if (!cluster_offset) {
728 if (bs->backing_hd) {
729 /* read from the base image */
730 n1 = backing_read1(bs->backing_hd, sector_num, buf, n);
732 ret = bdrv_read(bs->backing_hd, sector_num, buf, n1);
737 memset(buf, 0, 512 * n);
739 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
740 if (decompress_cluster(s, cluster_offset) < 0)
742 memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
744 ret = bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
747 if (s->crypt_method) {
748 encrypt_sectors(s, sector_num, buf, buf, n, 0,
749 &s->aes_decrypt_key);
759 static int qcow_write(BlockDriverState *bs, int64_t sector_num,
760 const uint8_t *buf, int nb_sectors)
762 BDRVQcowState *s = bs->opaque;
763 int ret, index_in_cluster, n;
764 uint64_t cluster_offset;
766 while (nb_sectors > 0) {
767 index_in_cluster = sector_num & (s->cluster_sectors - 1);
768 n = s->cluster_sectors - index_in_cluster;
771 cluster_offset = get_cluster_offset(bs, sector_num << 9, 1, 0,
773 index_in_cluster + n);
776 if (s->crypt_method) {
777 encrypt_sectors(s, sector_num, s->cluster_data, buf, n, 1,
778 &s->aes_encrypt_key);
779 ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512,
780 s->cluster_data, n * 512);
782 ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
790 s->cluster_cache_offset = -1; /* disable compressed cache */
799 uint64_t cluster_offset;
800 uint8_t *cluster_data;
801 BlockDriverAIOCB *hd_aiocb;
802 BlockDriverAIOCB *backing_hd_aiocb;
805 static void qcow_aio_delete(BlockDriverAIOCB *acb);
807 static int qcow_aio_new(BlockDriverAIOCB *acb)
809 BlockDriverState *bs = acb->bs;
810 BDRVQcowState *s = bs->opaque;
812 acb1 = qemu_mallocz(sizeof(QCowAIOCB));
816 acb1->hd_aiocb = bdrv_aio_new(s->hd);
819 if (bs->backing_hd) {
820 acb1->backing_hd_aiocb = bdrv_aio_new(bs->backing_hd);
821 if (!acb1->backing_hd_aiocb)
826 qcow_aio_delete(acb);
830 static void qcow_aio_read_cb(void *opaque, int ret)
832 BlockDriverAIOCB *acb = opaque;
833 BlockDriverState *bs = acb->bs;
834 BDRVQcowState *s = bs->opaque;
835 QCowAIOCB *acb1 = acb->opaque;
836 int index_in_cluster, n1;
840 acb->cb(acb->cb_opaque, ret);
845 /* post process the read buffer */
846 if (!acb1->cluster_offset) {
848 } else if (acb1->cluster_offset & QCOW_OFLAG_COMPRESSED) {
851 if (s->crypt_method) {
852 encrypt_sectors(s, acb1->sector_num, acb1->buf, acb1->buf,
854 &s->aes_decrypt_key);
858 acb1->nb_sectors -= acb1->n;
859 acb1->sector_num += acb1->n;
860 acb1->buf += acb1->n * 512;
862 if (acb1->nb_sectors == 0) {
863 /* request completed */
864 acb->cb(acb->cb_opaque, 0);
868 /* prepare next AIO request */
869 acb1->cluster_offset = get_cluster_offset(bs,
870 acb1->sector_num << 9,
872 index_in_cluster = acb1->sector_num & (s->cluster_sectors - 1);
873 acb1->n = s->cluster_sectors - index_in_cluster;
874 if (acb1->n > acb1->nb_sectors)
875 acb1->n = acb1->nb_sectors;
877 if (!acb1->cluster_offset) {
878 if (bs->backing_hd) {
879 /* read from the base image */
880 n1 = backing_read1(bs->backing_hd, acb1->sector_num,
883 ret = bdrv_aio_read(acb1->backing_hd_aiocb, acb1->sector_num,
884 acb1->buf, n1, qcow_aio_read_cb, acb);
891 /* Note: in this case, no need to wait */
892 memset(acb1->buf, 0, 512 * acb1->n);
895 } else if (acb1->cluster_offset & QCOW_OFLAG_COMPRESSED) {
896 /* add AIO support for compressed blocks ? */
897 if (decompress_cluster(s, acb1->cluster_offset) < 0)
900 s->cluster_cache + index_in_cluster * 512, 512 * acb1->n);
903 if ((acb1->cluster_offset & 511) != 0) {
907 ret = bdrv_aio_read(acb1->hd_aiocb,
908 (acb1->cluster_offset >> 9) + index_in_cluster,
909 acb1->buf, acb1->n, qcow_aio_read_cb, acb);
915 static int qcow_aio_read(BlockDriverAIOCB *acb, int64_t sector_num,
916 uint8_t *buf, int nb_sectors)
918 QCowAIOCB *acb1 = acb->opaque;
920 acb1->sector_num = sector_num;
922 acb1->nb_sectors = nb_sectors;
924 acb1->cluster_offset = 0;
926 qcow_aio_read_cb(acb, 0);
930 static void qcow_aio_write_cb(void *opaque, int ret)
932 BlockDriverAIOCB *acb = opaque;
933 BlockDriverState *bs = acb->bs;
934 BDRVQcowState *s = bs->opaque;
935 QCowAIOCB *acb1 = acb->opaque;
936 int index_in_cluster;
937 uint64_t cluster_offset;
938 const uint8_t *src_buf;
942 acb->cb(acb->cb_opaque, ret);
946 acb1->nb_sectors -= acb1->n;
947 acb1->sector_num += acb1->n;
948 acb1->buf += acb1->n * 512;
950 if (acb1->nb_sectors == 0) {
951 /* request completed */
952 acb->cb(acb->cb_opaque, 0);
956 index_in_cluster = acb1->sector_num & (s->cluster_sectors - 1);
957 acb1->n = s->cluster_sectors - index_in_cluster;
958 if (acb1->n > acb1->nb_sectors)
959 acb1->n = acb1->nb_sectors;
960 cluster_offset = get_cluster_offset(bs, acb1->sector_num << 9, 1, 0,
962 index_in_cluster + acb1->n);
963 if (!cluster_offset || (cluster_offset & 511) != 0) {
967 if (s->crypt_method) {
968 if (!acb1->cluster_data) {
969 acb1->cluster_data = qemu_mallocz(s->cluster_size);
970 if (!acb1->cluster_data) {
975 encrypt_sectors(s, acb1->sector_num, acb1->cluster_data, acb1->buf,
976 acb1->n, 1, &s->aes_encrypt_key);
977 src_buf = acb1->cluster_data;
981 ret = bdrv_aio_write(acb1->hd_aiocb,
982 (cluster_offset >> 9) + index_in_cluster,
984 qcow_aio_write_cb, acb);
989 static int qcow_aio_write(BlockDriverAIOCB *acb, int64_t sector_num,
990 const uint8_t *buf, int nb_sectors)
992 QCowAIOCB *acb1 = acb->opaque;
993 BlockDriverState *bs = acb->bs;
994 BDRVQcowState *s = bs->opaque;
996 s->cluster_cache_offset = -1; /* disable compressed cache */
998 acb1->sector_num = sector_num;
999 acb1->buf = (uint8_t *)buf;
1000 acb1->nb_sectors = nb_sectors;
1003 qcow_aio_write_cb(acb, 0);
1007 static void qcow_aio_cancel(BlockDriverAIOCB *acb)
1009 QCowAIOCB *acb1 = acb->opaque;
1011 bdrv_aio_cancel(acb1->hd_aiocb);
1012 if (acb1->backing_hd_aiocb)
1013 bdrv_aio_cancel(acb1->backing_hd_aiocb);
1016 static void qcow_aio_delete(BlockDriverAIOCB *acb)
1018 QCowAIOCB *acb1 = acb->opaque;
1020 bdrv_aio_delete(acb1->hd_aiocb);
1021 if (acb1->backing_hd_aiocb)
1022 bdrv_aio_delete(acb1->backing_hd_aiocb);
1023 qemu_free(acb1->cluster_data);
1027 static void qcow_close(BlockDriverState *bs)
1029 BDRVQcowState *s = bs->opaque;
1030 qemu_free(s->l1_table);
1031 qemu_free(s->l2_cache);
1032 qemu_free(s->cluster_cache);
1033 qemu_free(s->cluster_data);
1038 /* XXX: use std qcow open function ? */
1039 typedef struct QCowCreateState {
1042 uint16_t *refcount_block;
1043 uint64_t *refcount_table;
1044 int64_t l1_table_offset;
1045 int64_t refcount_table_offset;
1046 int64_t refcount_block_offset;
1049 static void create_refcount_update(QCowCreateState *s,
1050 int64_t offset, int64_t size)
1053 int64_t start, last, cluster_offset;
1056 start = offset & ~(s->cluster_size - 1);
1057 last = (offset + size - 1) & ~(s->cluster_size - 1);
1058 for(cluster_offset = start; cluster_offset <= last;
1059 cluster_offset += s->cluster_size) {
1060 p = &s->refcount_block[cluster_offset >> s->cluster_bits];
1061 refcount = be16_to_cpu(*p);
1063 *p = cpu_to_be16(refcount);
1067 static int qcow_create(const char *filename, int64_t total_size,
1068 const char *backing_file, int flags)
1070 int fd, header_size, backing_filename_len, l1_size, i, shift, l2_bits;
1072 uint64_t tmp, offset;
1073 QCowCreateState s1, *s = &s1;
1075 memset(s, 0, sizeof(*s));
1077 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
1080 memset(&header, 0, sizeof(header));
1081 header.magic = cpu_to_be32(QCOW_MAGIC);
1082 header.version = cpu_to_be32(QCOW_VERSION);
1083 header.size = cpu_to_be64(total_size * 512);
1084 header_size = sizeof(header);
1085 backing_filename_len = 0;
1087 header.backing_file_offset = cpu_to_be64(header_size);
1088 backing_filename_len = strlen(backing_file);
1089 header.backing_file_size = cpu_to_be32(backing_filename_len);
1090 header_size += backing_filename_len;
1092 s->cluster_bits = 12; /* 4 KB clusters */
1093 s->cluster_size = 1 << s->cluster_bits;
1094 header.cluster_bits = cpu_to_be32(s->cluster_bits);
1095 header_size = (header_size + 7) & ~7;
1097 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
1099 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
1101 l2_bits = s->cluster_bits - 3;
1102 shift = s->cluster_bits + l2_bits;
1103 l1_size = (((total_size * 512) + (1LL << shift) - 1) >> shift);
1104 offset = align_offset(header_size, s->cluster_size);
1105 s->l1_table_offset = offset;
1106 header.l1_table_offset = cpu_to_be64(s->l1_table_offset);
1107 header.l1_size = cpu_to_be32(l1_size);
1108 offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size);
1110 s->refcount_table = qemu_mallocz(s->cluster_size);
1111 if (!s->refcount_table)
1113 s->refcount_block = qemu_mallocz(s->cluster_size);
1114 if (!s->refcount_block)
1117 s->refcount_table_offset = offset;
1118 header.refcount_table_offset = cpu_to_be64(offset);
1119 header.refcount_table_clusters = cpu_to_be32(1);
1120 offset += s->cluster_size;
1122 s->refcount_table[0] = cpu_to_be64(offset);
1123 s->refcount_block_offset = offset;
1124 offset += s->cluster_size;
1126 /* update refcounts */
1127 create_refcount_update(s, 0, header_size);
1128 create_refcount_update(s, s->l1_table_offset, l1_size * sizeof(uint64_t));
1129 create_refcount_update(s, s->refcount_table_offset, s->cluster_size);
1130 create_refcount_update(s, s->refcount_block_offset, s->cluster_size);
1132 /* write all the data */
1133 write(fd, &header, sizeof(header));
1135 write(fd, backing_file, backing_filename_len);
1137 lseek(fd, s->l1_table_offset, SEEK_SET);
1139 for(i = 0;i < l1_size; i++) {
1140 write(fd, &tmp, sizeof(tmp));
1142 lseek(fd, s->refcount_table_offset, SEEK_SET);
1143 write(fd, s->refcount_table, s->cluster_size);
1145 lseek(fd, s->refcount_block_offset, SEEK_SET);
1146 write(fd, s->refcount_block, s->cluster_size);
1148 qemu_free(s->refcount_table);
1149 qemu_free(s->refcount_block);
1153 qemu_free(s->refcount_table);
1154 qemu_free(s->refcount_block);
1159 static int qcow_make_empty(BlockDriverState *bs)
1162 /* XXX: not correct */
1163 BDRVQcowState *s = bs->opaque;
1164 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1167 memset(s->l1_table, 0, l1_length);
1168 if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
1170 ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
1179 /* XXX: put compressed sectors first, then all the cluster aligned
1180 tables to avoid losing bytes in alignment */
1181 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
1182 const uint8_t *buf, int nb_sectors)
1184 BDRVQcowState *s = bs->opaque;
1188 uint64_t cluster_offset;
1190 if (nb_sectors == 0) {
1191 /* align end of file to a sector boundary to ease reading with
1192 sector based I/Os */
1193 cluster_offset = bdrv_getlength(s->hd);
1194 cluster_offset = (cluster_offset + 511) & ~511;
1195 bdrv_truncate(s->hd, cluster_offset);
1199 if (nb_sectors != s->cluster_sectors)
1202 out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1206 /* best compression, small window, no zlib header */
1207 memset(&strm, 0, sizeof(strm));
1208 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1210 9, Z_DEFAULT_STRATEGY);
1216 strm.avail_in = s->cluster_size;
1217 strm.next_in = (uint8_t *)buf;
1218 strm.avail_out = s->cluster_size;
1219 strm.next_out = out_buf;
1221 ret = deflate(&strm, Z_FINISH);
1222 if (ret != Z_STREAM_END && ret != Z_OK) {
1227 out_len = strm.next_out - out_buf;
1231 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1232 /* could not compress: write normal cluster */
1233 qcow_write(bs, sector_num, buf, s->cluster_sectors);
1235 cluster_offset = get_cluster_offset(bs, sector_num << 9, 2,
1237 cluster_offset &= s->cluster_offset_mask;
1238 if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
1248 static void qcow_flush(BlockDriverState *bs)
1250 BDRVQcowState *s = bs->opaque;
1254 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1256 BDRVQcowState *s = bs->opaque;
1257 bdi->cluster_size = s->cluster_size;
1258 bdi->vm_state_offset = (int64_t)s->l1_vm_state_index <<
1259 (s->cluster_bits + s->l2_bits);
1263 /*********************************************************/
1264 /* snapshot support */
1266 /* update the refcounts of snapshots and the copied flag */
1267 static int update_snapshot_refcount(BlockDriverState *bs,
1268 int64_t l1_table_offset,
1272 BDRVQcowState *s = bs->opaque;
1273 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
1274 int64_t old_offset, old_l2_offset;
1275 int l2_size, i, j, l1_modified, l2_modified, nb_csectors, refcount;
1281 l1_size2 = l1_size * sizeof(uint64_t);
1283 if (l1_table_offset != s->l1_table_offset) {
1284 l1_table = qemu_malloc(l1_size2);
1288 if (bdrv_pread(s->hd, l1_table_offset,
1289 l1_table, l1_size2) != l1_size2)
1291 for(i = 0;i < l1_size; i++)
1292 be64_to_cpus(&l1_table[i]);
1294 assert(l1_size == s->l1_size);
1295 l1_table = s->l1_table;
1299 l2_size = s->l2_size * sizeof(uint64_t);
1300 l2_table = qemu_malloc(l2_size);
1304 for(i = 0; i < l1_size; i++) {
1305 l2_offset = l1_table[i];
1307 old_l2_offset = l2_offset;
1308 l2_offset &= ~QCOW_OFLAG_COPIED;
1310 if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
1312 for(j = 0; j < s->l2_size; j++) {
1313 offset = be64_to_cpu(l2_table[j]);
1315 old_offset = offset;
1316 offset &= ~QCOW_OFLAG_COPIED;
1317 if (offset & QCOW_OFLAG_COMPRESSED) {
1318 nb_csectors = ((offset >> s->csize_shift) &
1321 update_refcount(bs, (offset & s->cluster_offset_mask) & ~511,
1322 nb_csectors * 512, addend);
1323 /* compressed clusters are never modified */
1327 refcount = update_cluster_refcount(bs, offset >> s->cluster_bits, addend);
1329 refcount = get_refcount(bs, offset >> s->cluster_bits);
1333 if (refcount == 1) {
1334 offset |= QCOW_OFLAG_COPIED;
1336 if (offset != old_offset) {
1337 l2_table[j] = cpu_to_be64(offset);
1343 if (bdrv_pwrite(s->hd,
1344 l2_offset, l2_table, l2_size) != l2_size)
1349 refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend);
1351 refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
1353 if (refcount == 1) {
1354 l2_offset |= QCOW_OFLAG_COPIED;
1356 if (l2_offset != old_l2_offset) {
1357 l1_table[i] = l2_offset;
1363 for(i = 0; i < l1_size; i++)
1364 cpu_to_be64s(&l1_table[i]);
1365 if (bdrv_pwrite(s->hd, l1_table_offset, l1_table,
1366 l1_size2) != l1_size2)
1368 for(i = 0; i < l1_size; i++)
1369 be64_to_cpus(&l1_table[i]);
1372 qemu_free(l1_table);
1373 qemu_free(l2_table);
1377 qemu_free(l1_table);
1378 qemu_free(l2_table);
1382 static void qcow_free_snapshots(BlockDriverState *bs)
1384 BDRVQcowState *s = bs->opaque;
1387 for(i = 0; i < s->nb_snapshots; i++) {
1388 qemu_free(s->snapshots[i].name);
1389 qemu_free(s->snapshots[i].id_str);
1391 qemu_free(s->snapshots);
1392 s->snapshots = NULL;
1393 s->nb_snapshots = 0;
1396 static int qcow_read_snapshots(BlockDriverState *bs)
1398 BDRVQcowState *s = bs->opaque;
1399 QCowSnapshotHeader h;
1401 int i, id_str_size, name_size;
1403 uint32_t extra_data_size;
1405 offset = s->snapshots_offset;
1406 s->snapshots = qemu_mallocz(s->nb_snapshots * sizeof(QCowSnapshot));
1409 for(i = 0; i < s->nb_snapshots; i++) {
1410 offset = align_offset(offset, 8);
1411 if (bdrv_pread(s->hd, offset, &h, sizeof(h)) != sizeof(h))
1413 offset += sizeof(h);
1414 sn = s->snapshots + i;
1415 sn->l1_table_offset = be64_to_cpu(h.l1_table_offset);
1416 sn->l1_size = be32_to_cpu(h.l1_size);
1417 sn->vm_state_size = be32_to_cpu(h.vm_state_size);
1418 sn->date_sec = be32_to_cpu(h.date_sec);
1419 sn->date_nsec = be32_to_cpu(h.date_nsec);
1420 sn->vm_clock_nsec = be64_to_cpu(h.vm_clock_nsec);
1421 extra_data_size = be32_to_cpu(h.extra_data_size);
1423 id_str_size = be16_to_cpu(h.id_str_size);
1424 name_size = be16_to_cpu(h.name_size);
1426 offset += extra_data_size;
1428 sn->id_str = qemu_malloc(id_str_size + 1);
1431 if (bdrv_pread(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
1433 offset += id_str_size;
1434 sn->id_str[id_str_size] = '\0';
1436 sn->name = qemu_malloc(name_size + 1);
1439 if (bdrv_pread(s->hd, offset, sn->name, name_size) != name_size)
1441 offset += name_size;
1442 sn->name[name_size] = '\0';
1444 s->snapshots_size = offset - s->snapshots_offset;
1447 qcow_free_snapshots(bs);
1451 /* add at the end of the file a new list of snapshots */
1452 static int qcow_write_snapshots(BlockDriverState *bs)
1454 BDRVQcowState *s = bs->opaque;
1456 QCowSnapshotHeader h;
1457 int i, name_size, id_str_size, snapshots_size;
1460 int64_t offset, snapshots_offset;
1462 /* compute the size of the snapshots */
1464 for(i = 0; i < s->nb_snapshots; i++) {
1465 sn = s->snapshots + i;
1466 offset = align_offset(offset, 8);
1467 offset += sizeof(h);
1468 offset += strlen(sn->id_str);
1469 offset += strlen(sn->name);
1471 snapshots_size = offset;
1473 snapshots_offset = alloc_clusters(bs, snapshots_size);
1474 offset = snapshots_offset;
1476 for(i = 0; i < s->nb_snapshots; i++) {
1477 sn = s->snapshots + i;
1478 memset(&h, 0, sizeof(h));
1479 h.l1_table_offset = cpu_to_be64(sn->l1_table_offset);
1480 h.l1_size = cpu_to_be32(sn->l1_size);
1481 h.vm_state_size = cpu_to_be32(sn->vm_state_size);
1482 h.date_sec = cpu_to_be32(sn->date_sec);
1483 h.date_nsec = cpu_to_be32(sn->date_nsec);
1484 h.vm_clock_nsec = cpu_to_be64(sn->vm_clock_nsec);
1486 id_str_size = strlen(sn->id_str);
1487 name_size = strlen(sn->name);
1488 h.id_str_size = cpu_to_be16(id_str_size);
1489 h.name_size = cpu_to_be16(name_size);
1490 offset = align_offset(offset, 8);
1491 if (bdrv_pwrite(s->hd, offset, &h, sizeof(h)) != sizeof(h))
1493 offset += sizeof(h);
1494 if (bdrv_pwrite(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
1496 offset += id_str_size;
1497 if (bdrv_pwrite(s->hd, offset, sn->name, name_size) != name_size)
1499 offset += name_size;
1502 /* update the various header fields */
1503 data64 = cpu_to_be64(snapshots_offset);
1504 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, snapshots_offset),
1505 &data64, sizeof(data64)) != sizeof(data64))
1507 data32 = cpu_to_be32(s->nb_snapshots);
1508 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, nb_snapshots),
1509 &data32, sizeof(data32)) != sizeof(data32))
1512 /* free the old snapshot table */
1513 free_clusters(bs, s->snapshots_offset, s->snapshots_size);
1514 s->snapshots_offset = snapshots_offset;
1515 s->snapshots_size = snapshots_size;
1521 static void find_new_snapshot_id(BlockDriverState *bs,
1522 char *id_str, int id_str_size)
1524 BDRVQcowState *s = bs->opaque;
1526 int i, id, id_max = 0;
1528 for(i = 0; i < s->nb_snapshots; i++) {
1529 sn = s->snapshots + i;
1530 id = strtoul(sn->id_str, NULL, 10);
1534 snprintf(id_str, id_str_size, "%d", id_max + 1);
1537 static int find_snapshot_by_id(BlockDriverState *bs, const char *id_str)
1539 BDRVQcowState *s = bs->opaque;
1542 for(i = 0; i < s->nb_snapshots; i++) {
1543 if (!strcmp(s->snapshots[i].id_str, id_str))
1549 static int find_snapshot_by_id_or_name(BlockDriverState *bs, const char *name)
1551 BDRVQcowState *s = bs->opaque;
1554 ret = find_snapshot_by_id(bs, name);
1557 for(i = 0; i < s->nb_snapshots; i++) {
1558 if (!strcmp(s->snapshots[i].name, name))
1564 /* if no id is provided, a new one is constructed */
1565 static int qcow_snapshot_create(BlockDriverState *bs,
1566 QEMUSnapshotInfo *sn_info)
1568 BDRVQcowState *s = bs->opaque;
1569 QCowSnapshot *snapshots1, sn1, *sn = &sn1;
1571 uint64_t *l1_table = NULL;
1573 memset(sn, 0, sizeof(*sn));
1575 if (sn_info->id_str[0] == '\0') {
1576 /* compute a new id */
1577 find_new_snapshot_id(bs, sn_info->id_str, sizeof(sn_info->id_str));
1580 /* check that the ID is unique */
1581 if (find_snapshot_by_id(bs, sn_info->id_str) >= 0)
1584 sn->id_str = qemu_strdup(sn_info->id_str);
1587 sn->name = qemu_strdup(sn_info->name);
1590 sn->vm_state_size = sn_info->vm_state_size;
1591 sn->date_sec = sn_info->date_sec;
1592 sn->date_nsec = sn_info->date_nsec;
1593 sn->vm_clock_nsec = sn_info->vm_clock_nsec;
1595 ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1);
1599 /* create the L1 table of the snapshot */
1600 sn->l1_table_offset = alloc_clusters(bs, s->l1_size * sizeof(uint64_t));
1601 sn->l1_size = s->l1_size;
1603 l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
1606 for(i = 0; i < s->l1_size; i++) {
1607 l1_table[i] = cpu_to_be64(s->l1_table[i]);
1609 if (bdrv_pwrite(s->hd, sn->l1_table_offset,
1610 l1_table, s->l1_size * sizeof(uint64_t)) !=
1611 (s->l1_size * sizeof(uint64_t)))
1613 qemu_free(l1_table);
1616 snapshots1 = qemu_malloc((s->nb_snapshots + 1) * sizeof(QCowSnapshot));
1619 memcpy(snapshots1, s->snapshots, s->nb_snapshots * sizeof(QCowSnapshot));
1620 s->snapshots = snapshots1;
1621 s->snapshots[s->nb_snapshots++] = *sn;
1623 if (qcow_write_snapshots(bs) < 0)
1626 check_refcounts(bs);
1630 qemu_free(sn->name);
1631 qemu_free(l1_table);
1635 /* copy the snapshot 'snapshot_name' into the current disk image */
1636 static int qcow_snapshot_goto(BlockDriverState *bs,
1637 const char *snapshot_id)
1639 BDRVQcowState *s = bs->opaque;
1641 int i, snapshot_index, l1_size2;
1643 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
1644 if (snapshot_index < 0)
1646 sn = &s->snapshots[snapshot_index];
1648 if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, -1) < 0)
1651 if (grow_l1_table(bs, sn->l1_size) < 0)
1654 s->l1_size = sn->l1_size;
1655 l1_size2 = s->l1_size * sizeof(uint64_t);
1656 /* copy the snapshot l1 table to the current l1 table */
1657 if (bdrv_pread(s->hd, sn->l1_table_offset,
1658 s->l1_table, l1_size2) != l1_size2)
1660 if (bdrv_pwrite(s->hd, s->l1_table_offset,
1661 s->l1_table, l1_size2) != l1_size2)
1663 for(i = 0;i < s->l1_size; i++) {
1664 be64_to_cpus(&s->l1_table[i]);
1667 if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1) < 0)
1671 check_refcounts(bs);
1678 static int qcow_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1680 BDRVQcowState *s = bs->opaque;
1682 int snapshot_index, ret;
1684 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
1685 if (snapshot_index < 0)
1687 sn = &s->snapshots[snapshot_index];
1689 ret = update_snapshot_refcount(bs, sn->l1_table_offset, sn->l1_size, -1);
1692 /* must update the copied flag on the current cluster offsets */
1693 ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
1696 free_clusters(bs, sn->l1_table_offset, sn->l1_size * sizeof(uint64_t));
1698 qemu_free(sn->id_str);
1699 qemu_free(sn->name);
1700 memmove(sn, sn + 1, (s->nb_snapshots - snapshot_index - 1) * sizeof(*sn));
1702 ret = qcow_write_snapshots(bs);
1704 /* XXX: restore snapshot if error ? */
1708 check_refcounts(bs);
1713 static int qcow_snapshot_list(BlockDriverState *bs,
1714 QEMUSnapshotInfo **psn_tab)
1716 BDRVQcowState *s = bs->opaque;
1717 QEMUSnapshotInfo *sn_tab, *sn_info;
1721 sn_tab = qemu_mallocz(s->nb_snapshots * sizeof(QEMUSnapshotInfo));
1724 for(i = 0; i < s->nb_snapshots; i++) {
1725 sn_info = sn_tab + i;
1726 sn = s->snapshots + i;
1727 pstrcpy(sn_info->id_str, sizeof(sn_info->id_str),
1729 pstrcpy(sn_info->name, sizeof(sn_info->name),
1731 sn_info->vm_state_size = sn->vm_state_size;
1732 sn_info->date_sec = sn->date_sec;
1733 sn_info->date_nsec = sn->date_nsec;
1734 sn_info->vm_clock_nsec = sn->vm_clock_nsec;
1737 return s->nb_snapshots;
1744 /*********************************************************/
1745 /* refcount handling */
1747 static int refcount_init(BlockDriverState *bs)
1749 BDRVQcowState *s = bs->opaque;
1750 int ret, refcount_table_size2, i;
1752 s->refcount_block_cache = qemu_malloc(s->cluster_size);
1753 if (!s->refcount_block_cache)
1755 refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
1756 s->refcount_table = qemu_malloc(refcount_table_size2);
1757 if (!s->refcount_table)
1759 if (s->refcount_table_size > 0) {
1760 ret = bdrv_pread(s->hd, s->refcount_table_offset,
1761 s->refcount_table, refcount_table_size2);
1762 if (ret != refcount_table_size2)
1764 for(i = 0; i < s->refcount_table_size; i++)
1765 be64_to_cpus(&s->refcount_table[i]);
1772 static void refcount_close(BlockDriverState *bs)
1774 BDRVQcowState *s = bs->opaque;
1775 qemu_free(s->refcount_block_cache);
1776 qemu_free(s->refcount_table);
1780 static int load_refcount_block(BlockDriverState *bs,
1781 int64_t refcount_block_offset)
1783 BDRVQcowState *s = bs->opaque;
1785 ret = bdrv_pread(s->hd, refcount_block_offset, s->refcount_block_cache,
1787 if (ret != s->cluster_size)
1789 s->refcount_block_cache_offset = refcount_block_offset;
1793 static int get_refcount(BlockDriverState *bs, int64_t cluster_index)
1795 BDRVQcowState *s = bs->opaque;
1796 int refcount_table_index, block_index;
1797 int64_t refcount_block_offset;
1799 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
1800 if (refcount_table_index >= s->refcount_table_size)
1802 refcount_block_offset = s->refcount_table[refcount_table_index];
1803 if (!refcount_block_offset)
1805 if (refcount_block_offset != s->refcount_block_cache_offset) {
1806 /* better than nothing: return allocated if read error */
1807 if (load_refcount_block(bs, refcount_block_offset) < 0)
1810 block_index = cluster_index &
1811 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
1812 return be16_to_cpu(s->refcount_block_cache[block_index]);
1815 /* return < 0 if error */
1816 static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
1818 BDRVQcowState *s = bs->opaque;
1821 nb_clusters = (size + s->cluster_size - 1) >> s->cluster_bits;
1823 if (get_refcount(bs, s->free_cluster_index) == 0) {
1824 s->free_cluster_index++;
1825 for(i = 1; i < nb_clusters; i++) {
1826 if (get_refcount(bs, s->free_cluster_index) != 0)
1828 s->free_cluster_index++;
1831 printf("alloc_clusters: size=%lld -> %lld\n",
1833 (s->free_cluster_index - nb_clusters) << s->cluster_bits);
1835 return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
1838 s->free_cluster_index++;
1843 static int64_t alloc_clusters(BlockDriverState *bs, int64_t size)
1847 offset = alloc_clusters_noref(bs, size);
1848 update_refcount(bs, offset, size, 1);
1852 /* only used to allocate compressed sectors. We try to allocate
1853 contiguous sectors. size must be <= cluster_size */
1854 static int64_t alloc_bytes(BlockDriverState *bs, int size)
1856 BDRVQcowState *s = bs->opaque;
1857 int64_t offset, cluster_offset;
1858 int free_in_cluster;
1860 assert(size > 0 && size <= s->cluster_size);
1861 if (s->free_byte_offset == 0) {
1862 s->free_byte_offset = alloc_clusters(bs, s->cluster_size);
1865 free_in_cluster = s->cluster_size -
1866 (s->free_byte_offset & (s->cluster_size - 1));
1867 if (size <= free_in_cluster) {
1868 /* enough space in current cluster */
1869 offset = s->free_byte_offset;
1870 s->free_byte_offset += size;
1871 free_in_cluster -= size;
1872 if (free_in_cluster == 0)
1873 s->free_byte_offset = 0;
1874 if ((offset & (s->cluster_size - 1)) != 0)
1875 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
1877 offset = alloc_clusters(bs, s->cluster_size);
1878 cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
1879 if ((cluster_offset + s->cluster_size) == offset) {
1880 /* we are lucky: contiguous data */
1881 offset = s->free_byte_offset;
1882 update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
1883 s->free_byte_offset += size;
1885 s->free_byte_offset = offset;
1892 static void free_clusters(BlockDriverState *bs,
1893 int64_t offset, int64_t size)
1895 update_refcount(bs, offset, size, -1);
1898 static int grow_refcount_table(BlockDriverState *bs, int min_size)
1900 BDRVQcowState *s = bs->opaque;
1901 int new_table_size, new_table_size2, refcount_table_clusters, i, ret;
1902 uint64_t *new_table;
1903 int64_t table_offset;
1907 if (min_size <= s->refcount_table_size)
1909 /* compute new table size */
1910 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
1912 if (refcount_table_clusters == 0) {
1913 refcount_table_clusters = 1;
1915 refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
1917 new_table_size = refcount_table_clusters << (s->cluster_bits - 3);
1918 if (min_size <= new_table_size)
1922 printf("grow_refcount_table from %d to %d\n",
1923 s->refcount_table_size,
1926 new_table_size2 = new_table_size * sizeof(uint64_t);
1927 new_table = qemu_mallocz(new_table_size2);
1930 memcpy(new_table, s->refcount_table,
1931 s->refcount_table_size * sizeof(uint64_t));
1932 for(i = 0; i < s->refcount_table_size; i++)
1933 cpu_to_be64s(&new_table[i]);
1934 /* Note: we cannot update the refcount now to avoid recursion */
1935 table_offset = alloc_clusters_noref(bs, new_table_size2);
1936 ret = bdrv_pwrite(s->hd, table_offset, new_table, new_table_size2);
1937 if (ret != new_table_size2)
1939 for(i = 0; i < s->refcount_table_size; i++)
1940 be64_to_cpus(&new_table[i]);
1942 data64 = cpu_to_be64(table_offset);
1943 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, refcount_table_offset),
1944 &data64, sizeof(data64)) != sizeof(data64))
1946 data32 = cpu_to_be32(refcount_table_clusters);
1947 if (bdrv_pwrite(s->hd, offsetof(QCowHeader, refcount_table_clusters),
1948 &data32, sizeof(data32)) != sizeof(data32))
1950 qemu_free(s->refcount_table);
1951 s->refcount_table = new_table;
1952 s->refcount_table_size = new_table_size;
1954 update_refcount(bs, table_offset, new_table_size2, 1);
1957 free_clusters(bs, table_offset, new_table_size2);
1958 qemu_free(new_table);
1962 /* addend must be 1 or -1 */
1963 /* XXX: cache several refcount block clusters ? */
1964 static int update_cluster_refcount(BlockDriverState *bs,
1965 int64_t cluster_index,
1968 BDRVQcowState *s = bs->opaque;
1969 int64_t offset, refcount_block_offset;
1970 int ret, refcount_table_index, block_index, refcount;
1973 refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
1974 if (refcount_table_index >= s->refcount_table_size) {
1977 ret = grow_refcount_table(bs, refcount_table_index + 1);
1981 refcount_block_offset = s->refcount_table[refcount_table_index];
1982 if (!refcount_block_offset) {
1985 /* create a new refcount block */
1986 /* Note: we cannot update the refcount now to avoid recursion */
1987 offset = alloc_clusters_noref(bs, s->cluster_size);
1988 memset(s->refcount_block_cache, 0, s->cluster_size);
1989 ret = bdrv_pwrite(s->hd, offset, s->refcount_block_cache, s->cluster_size);
1990 if (ret != s->cluster_size)
1992 s->refcount_table[refcount_table_index] = offset;
1993 data64 = cpu_to_be64(offset);
1994 ret = bdrv_pwrite(s->hd, s->refcount_table_offset +
1995 refcount_table_index * sizeof(uint64_t),
1996 &data64, sizeof(data64));
1997 if (ret != sizeof(data64))
2000 refcount_block_offset = offset;
2001 s->refcount_block_cache_offset = offset;
2002 update_refcount(bs, offset, s->cluster_size, 1);
2004 if (refcount_block_offset != s->refcount_block_cache_offset) {
2005 if (load_refcount_block(bs, refcount_block_offset) < 0)
2009 /* we can update the count and save it */
2010 block_index = cluster_index &
2011 ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2012 refcount = be16_to_cpu(s->refcount_block_cache[block_index]);
2014 if (refcount < 0 || refcount > 0xffff)
2016 if (refcount == 0 && cluster_index < s->free_cluster_index) {
2017 s->free_cluster_index = cluster_index;
2019 s->refcount_block_cache[block_index] = cpu_to_be16(refcount);
2020 if (bdrv_pwrite(s->hd,
2021 refcount_block_offset + (block_index << REFCOUNT_SHIFT),
2022 &s->refcount_block_cache[block_index], 2) != 2)
2027 static void update_refcount(BlockDriverState *bs,
2028 int64_t offset, int64_t length,
2031 BDRVQcowState *s = bs->opaque;
2032 int64_t start, last, cluster_offset;
2035 printf("update_refcount: offset=%lld size=%lld addend=%d\n",
2036 offset, length, addend);
2040 start = offset & ~(s->cluster_size - 1);
2041 last = (offset + length - 1) & ~(s->cluster_size - 1);
2042 for(cluster_offset = start; cluster_offset <= last;
2043 cluster_offset += s->cluster_size) {
2044 update_cluster_refcount(bs, cluster_offset >> s->cluster_bits, addend);
2049 static void inc_refcounts(BlockDriverState *bs,
2050 uint16_t *refcount_table,
2051 int refcount_table_size,
2052 int64_t offset, int64_t size)
2054 BDRVQcowState *s = bs->opaque;
2055 int64_t start, last, cluster_offset;
2061 start = offset & ~(s->cluster_size - 1);
2062 last = (offset + size - 1) & ~(s->cluster_size - 1);
2063 for(cluster_offset = start; cluster_offset <= last;
2064 cluster_offset += s->cluster_size) {
2065 k = cluster_offset >> s->cluster_bits;
2066 if (k < 0 || k >= refcount_table_size) {
2067 printf("ERROR: invalid cluster offset=0x%llx\n", cluster_offset);
2069 if (++refcount_table[k] == 0) {
2070 printf("ERROR: overflow cluster offset=0x%llx\n", cluster_offset);
2076 static int check_refcounts_l1(BlockDriverState *bs,
2077 uint16_t *refcount_table,
2078 int refcount_table_size,
2079 int64_t l1_table_offset, int l1_size,
2082 BDRVQcowState *s = bs->opaque;
2083 uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2;
2084 int l2_size, i, j, nb_csectors, refcount;
2087 l1_size2 = l1_size * sizeof(uint64_t);
2089 inc_refcounts(bs, refcount_table, refcount_table_size,
2090 l1_table_offset, l1_size2);
2092 l1_table = qemu_malloc(l1_size2);
2095 if (bdrv_pread(s->hd, l1_table_offset,
2096 l1_table, l1_size2) != l1_size2)
2098 for(i = 0;i < l1_size; i++)
2099 be64_to_cpus(&l1_table[i]);
2101 l2_size = s->l2_size * sizeof(uint64_t);
2102 l2_table = qemu_malloc(l2_size);
2105 for(i = 0; i < l1_size; i++) {
2106 l2_offset = l1_table[i];
2109 refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED) >> s->cluster_bits);
2110 if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
2111 printf("ERROR OFLAG_COPIED: l2_offset=%llx refcount=%d\n",
2112 l2_offset, refcount);
2115 l2_offset &= ~QCOW_OFLAG_COPIED;
2116 if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
2118 for(j = 0; j < s->l2_size; j++) {
2119 offset = be64_to_cpu(l2_table[j]);
2121 if (offset & QCOW_OFLAG_COMPRESSED) {
2122 if (offset & QCOW_OFLAG_COPIED) {
2123 printf("ERROR: cluster %lld: copied flag must never be set for compressed clusters\n",
2124 offset >> s->cluster_bits);
2125 offset &= ~QCOW_OFLAG_COPIED;
2127 nb_csectors = ((offset >> s->csize_shift) &
2129 offset &= s->cluster_offset_mask;
2130 inc_refcounts(bs, refcount_table,
2131 refcount_table_size,
2132 offset & ~511, nb_csectors * 512);
2135 refcount = get_refcount(bs, (offset & ~QCOW_OFLAG_COPIED) >> s->cluster_bits);
2136 if ((refcount == 1) != ((offset & QCOW_OFLAG_COPIED) != 0)) {
2137 printf("ERROR OFLAG_COPIED: offset=%llx refcount=%d\n",
2141 offset &= ~QCOW_OFLAG_COPIED;
2142 inc_refcounts(bs, refcount_table,
2143 refcount_table_size,
2144 offset, s->cluster_size);
2148 inc_refcounts(bs, refcount_table,
2149 refcount_table_size,
2154 qemu_free(l1_table);
2155 qemu_free(l2_table);
2158 printf("ERROR: I/O error in check_refcounts_l1\n");
2159 qemu_free(l1_table);
2160 qemu_free(l2_table);
2164 static void check_refcounts(BlockDriverState *bs)
2166 BDRVQcowState *s = bs->opaque;
2168 int nb_clusters, refcount1, refcount2, i;
2170 uint16_t *refcount_table;
2172 size = bdrv_getlength(s->hd);
2173 nb_clusters = (size + s->cluster_size - 1) >> s->cluster_bits;
2174 refcount_table = qemu_mallocz(nb_clusters * sizeof(uint16_t));
2177 inc_refcounts(bs, refcount_table, nb_clusters,
2178 0, s->cluster_size);
2180 check_refcounts_l1(bs, refcount_table, nb_clusters,
2181 s->l1_table_offset, s->l1_size, 1);
2184 for(i = 0; i < s->nb_snapshots; i++) {
2185 sn = s->snapshots + i;
2186 check_refcounts_l1(bs, refcount_table, nb_clusters,
2187 sn->l1_table_offset, sn->l1_size, 0);
2189 inc_refcounts(bs, refcount_table, nb_clusters,
2190 s->snapshots_offset, s->snapshots_size);
2193 inc_refcounts(bs, refcount_table, nb_clusters,
2194 s->refcount_table_offset,
2195 s->refcount_table_size * sizeof(uint64_t));
2196 for(i = 0; i < s->refcount_table_size; i++) {
2198 offset = s->refcount_table[i];
2200 inc_refcounts(bs, refcount_table, nb_clusters,
2201 offset, s->cluster_size);
2205 /* compare ref counts */
2206 for(i = 0; i < nb_clusters; i++) {
2207 refcount1 = get_refcount(bs, i);
2208 refcount2 = refcount_table[i];
2209 if (refcount1 != refcount2)
2210 printf("ERROR cluster %d refcount=%d reference=%d\n",
2211 i, refcount1, refcount2);
2214 qemu_free(refcount_table);
2218 static void dump_refcounts(BlockDriverState *bs)
2220 BDRVQcowState *s = bs->opaque;
2221 int64_t nb_clusters, k, k1, size;
2224 size = bdrv_getlength(s->hd);
2225 nb_clusters = (size + s->cluster_size - 1) >> s->cluster_bits;
2226 for(k = 0; k < nb_clusters;) {
2228 refcount = get_refcount(bs, k);
2230 while (k < nb_clusters && get_refcount(bs, k) == refcount)
2232 printf("%lld: refcount=%d nb=%lld\n", k, refcount, k - k1);
2238 BlockDriver bdrv_qcow2 = {
2240 sizeof(BDRVQcowState),
2252 .bdrv_aio_new = qcow_aio_new,
2253 .bdrv_aio_read = qcow_aio_read,
2254 .bdrv_aio_write = qcow_aio_write,
2255 .bdrv_aio_cancel = qcow_aio_cancel,
2256 .bdrv_aio_delete = qcow_aio_delete,
2257 .bdrv_write_compressed = qcow_write_compressed,
2259 .bdrv_snapshot_create = qcow_snapshot_create,
2260 .bdrv_snapshot_goto = qcow_snapshot_goto,
2261 .bdrv_snapshot_delete = qcow_snapshot_delete,
2262 .bdrv_snapshot_list = qcow_snapshot_list,
2263 .bdrv_get_info = qcow_get_info,