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"
29 #include "block/qcow2.h"
30 #include "qemu-error.h"
34 Differences with QCOW:
36 - Support for multiple incremental snapshots.
37 - Memory management by reference counts.
38 - Clusters which have a reference count of one have the bit
39 QCOW_OFLAG_COPIED to optimize write performance.
40 - Size of compressed clusters is stored in sectors to reduce bit usage
41 in the cluster offsets.
42 - Support for storing additional data (such as the VM state) in the
44 - If a backing store is used, the cluster size is not constrained
45 (could be backported to QCOW).
46 - L2 tables have always a size of one cluster.
54 #define QCOW2_EXT_MAGIC_END 0
55 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
57 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
59 const QCowHeader *cow_header = (const void *)buf;
61 if (buf_size >= sizeof(QCowHeader) &&
62 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
63 be32_to_cpu(cow_header->version) >= QCOW_VERSION)
71 * read qcow2 extension and fill bs
72 * start reading from start_offset
73 * finish reading upon magic of value 0 or when end_offset reached
74 * unknown magic is skipped (future extension this version knows nothing about)
75 * return 0 upon success, non-0 otherwise
77 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
80 BDRVQcowState *s = bs->opaque;
86 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
88 offset = start_offset;
89 while (offset < end_offset) {
93 if (offset > s->cluster_size)
94 printf("qcow2_read_extension: suspicious offset %lu\n", offset);
96 printf("attempting to read extended header in offset %lu\n", offset);
99 if (bdrv_pread(bs->file, offset, &ext, sizeof(ext)) != sizeof(ext)) {
100 fprintf(stderr, "qcow2_read_extension: ERROR: "
101 "pread fail from offset %" PRIu64 "\n",
105 be32_to_cpus(&ext.magic);
106 be32_to_cpus(&ext.len);
107 offset += sizeof(ext);
109 printf("ext.magic = 0x%x\n", ext.magic);
112 case QCOW2_EXT_MAGIC_END:
115 case QCOW2_EXT_MAGIC_BACKING_FORMAT:
116 if (ext.len >= sizeof(bs->backing_format)) {
117 fprintf(stderr, "ERROR: ext_backing_format: len=%u too large"
119 ext.len, sizeof(bs->backing_format));
122 if (bdrv_pread(bs->file, offset , bs->backing_format,
125 bs->backing_format[ext.len] = '\0';
127 printf("Qcow2: Got format extension %s\n", bs->backing_format);
132 /* unknown magic - save it in case we need to rewrite the header */
134 Qcow2UnknownHeaderExtension *uext;
136 uext = g_malloc0(sizeof(*uext) + ext.len);
137 uext->magic = ext.magic;
139 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
141 ret = bdrv_pread(bs->file, offset , uext->data, uext->len);
149 offset += ((ext.len + 7) & ~7);
155 static void cleanup_unknown_header_ext(BlockDriverState *bs)
157 BDRVQcowState *s = bs->opaque;
158 Qcow2UnknownHeaderExtension *uext, *next;
160 QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
161 QLIST_REMOVE(uext, next);
166 static int qcow2_open(BlockDriverState *bs, int flags)
168 BDRVQcowState *s = bs->opaque;
174 ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
178 be32_to_cpus(&header.magic);
179 be32_to_cpus(&header.version);
180 be64_to_cpus(&header.backing_file_offset);
181 be32_to_cpus(&header.backing_file_size);
182 be64_to_cpus(&header.size);
183 be32_to_cpus(&header.cluster_bits);
184 be32_to_cpus(&header.crypt_method);
185 be64_to_cpus(&header.l1_table_offset);
186 be32_to_cpus(&header.l1_size);
187 be64_to_cpus(&header.refcount_table_offset);
188 be32_to_cpus(&header.refcount_table_clusters);
189 be64_to_cpus(&header.snapshots_offset);
190 be32_to_cpus(&header.nb_snapshots);
192 if (header.magic != QCOW_MAGIC) {
196 if (header.version != QCOW_VERSION) {
198 snprintf(version, sizeof(version), "QCOW version %d", header.version);
199 qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
200 bs->device_name, "qcow2", version);
204 if (header.cluster_bits < MIN_CLUSTER_BITS ||
205 header.cluster_bits > MAX_CLUSTER_BITS) {
209 if (header.crypt_method > QCOW_CRYPT_AES) {
213 s->crypt_method_header = header.crypt_method;
214 if (s->crypt_method_header) {
217 s->cluster_bits = header.cluster_bits;
218 s->cluster_size = 1 << s->cluster_bits;
219 s->cluster_sectors = 1 << (s->cluster_bits - 9);
220 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
221 s->l2_size = 1 << s->l2_bits;
222 bs->total_sectors = header.size / 512;
223 s->csize_shift = (62 - (s->cluster_bits - 8));
224 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
225 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
226 s->refcount_table_offset = header.refcount_table_offset;
227 s->refcount_table_size =
228 header.refcount_table_clusters << (s->cluster_bits - 3);
230 s->snapshots_offset = header.snapshots_offset;
231 s->nb_snapshots = header.nb_snapshots;
233 /* read the level 1 table */
234 s->l1_size = header.l1_size;
235 s->l1_vm_state_index = size_to_l1(s, header.size);
236 /* the L1 table must contain at least enough entries to put
238 if (s->l1_size < s->l1_vm_state_index) {
242 s->l1_table_offset = header.l1_table_offset;
243 if (s->l1_size > 0) {
244 s->l1_table = g_malloc0(
245 align_offset(s->l1_size * sizeof(uint64_t), 512));
246 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
247 s->l1_size * sizeof(uint64_t));
251 for(i = 0;i < s->l1_size; i++) {
252 be64_to_cpus(&s->l1_table[i]);
256 /* alloc L2 table/refcount block cache */
257 writethrough = ((flags & BDRV_O_CACHE_WB) == 0);
258 s->l2_table_cache = qcow2_cache_create(bs, L2_CACHE_SIZE, writethrough);
259 s->refcount_block_cache = qcow2_cache_create(bs, REFCOUNT_CACHE_SIZE,
262 s->cluster_cache = g_malloc(s->cluster_size);
263 /* one more sector for decompressed data alignment */
264 s->cluster_data = qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
266 s->cluster_cache_offset = -1;
269 ret = qcow2_refcount_init(bs);
274 QLIST_INIT(&s->cluster_allocs);
276 /* read qcow2 extensions */
277 if (header.backing_file_offset) {
278 ext_end = header.backing_file_offset;
280 ext_end = s->cluster_size;
282 if (qcow2_read_extensions(bs, sizeof(header), ext_end)) {
287 /* read the backing file name */
288 if (header.backing_file_offset != 0) {
289 len = header.backing_file_size;
293 ret = bdrv_pread(bs->file, header.backing_file_offset,
294 bs->backing_file, len);
298 bs->backing_file[len] = '\0';
301 ret = qcow2_read_snapshots(bs);
306 /* Initialise locks */
307 qemu_co_mutex_init(&s->lock);
311 BdrvCheckResult result = {0};
312 qcow2_check_refcounts(bs, &result);
318 cleanup_unknown_header_ext(bs);
319 qcow2_free_snapshots(bs);
320 qcow2_refcount_close(bs);
322 if (s->l2_table_cache) {
323 qcow2_cache_destroy(bs, s->l2_table_cache);
325 g_free(s->cluster_cache);
326 qemu_vfree(s->cluster_data);
330 static int qcow2_set_key(BlockDriverState *bs, const char *key)
332 BDRVQcowState *s = bs->opaque;
336 memset(keybuf, 0, 16);
340 /* XXX: we could compress the chars to 7 bits to increase
342 for(i = 0;i < len;i++) {
345 s->crypt_method = s->crypt_method_header;
347 if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
349 if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
359 AES_encrypt(in, tmp, &s->aes_encrypt_key);
360 AES_decrypt(tmp, out, &s->aes_decrypt_key);
361 for(i = 0; i < 16; i++)
362 printf(" %02x", tmp[i]);
364 for(i = 0; i < 16; i++)
365 printf(" %02x", out[i]);
372 static int coroutine_fn qcow2_co_is_allocated(BlockDriverState *bs,
373 int64_t sector_num, int nb_sectors, int *pnum)
375 BDRVQcowState *s = bs->opaque;
376 uint64_t cluster_offset;
380 /* FIXME We can get errors here, but the bdrv_co_is_allocated interface
381 * can't pass them on today */
382 qemu_co_mutex_lock(&s->lock);
383 ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset);
384 qemu_co_mutex_unlock(&s->lock);
389 return (cluster_offset != 0);
392 /* handle reading after the end of the backing file */
393 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
394 int64_t sector_num, int nb_sectors)
397 if ((sector_num + nb_sectors) <= bs->total_sectors)
399 if (sector_num >= bs->total_sectors)
402 n1 = bs->total_sectors - sector_num;
404 qemu_iovec_memset_skip(qiov, 0, 512 * (nb_sectors - n1), 512 * n1);
409 static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num,
410 int remaining_sectors, QEMUIOVector *qiov)
412 BDRVQcowState *s = bs->opaque;
413 int index_in_cluster, n1;
415 int cur_nr_sectors; /* number of sectors in current iteration */
416 uint64_t cluster_offset = 0;
417 uint64_t bytes_done = 0;
418 QEMUIOVector hd_qiov;
419 uint8_t *cluster_data = NULL;
421 qemu_iovec_init(&hd_qiov, qiov->niov);
423 qemu_co_mutex_lock(&s->lock);
425 while (remaining_sectors != 0) {
427 /* prepare next request */
428 cur_nr_sectors = remaining_sectors;
429 if (s->crypt_method) {
430 cur_nr_sectors = MIN(cur_nr_sectors,
431 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
434 ret = qcow2_get_cluster_offset(bs, sector_num << 9,
435 &cur_nr_sectors, &cluster_offset);
440 index_in_cluster = sector_num & (s->cluster_sectors - 1);
442 qemu_iovec_reset(&hd_qiov);
443 qemu_iovec_copy(&hd_qiov, qiov, bytes_done,
444 cur_nr_sectors * 512);
446 if (!cluster_offset) {
448 if (bs->backing_hd) {
449 /* read from the base image */
450 n1 = qcow2_backing_read1(bs->backing_hd, &hd_qiov,
451 sector_num, cur_nr_sectors);
453 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
454 qemu_co_mutex_unlock(&s->lock);
455 ret = bdrv_co_readv(bs->backing_hd, sector_num,
457 qemu_co_mutex_lock(&s->lock);
463 /* Note: in this case, no need to wait */
464 qemu_iovec_memset(&hd_qiov, 0, 512 * cur_nr_sectors);
466 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
467 /* add AIO support for compressed blocks ? */
468 ret = qcow2_decompress_cluster(bs, cluster_offset);
473 qemu_iovec_from_buffer(&hd_qiov,
474 s->cluster_cache + index_in_cluster * 512,
475 512 * cur_nr_sectors);
477 if ((cluster_offset & 511) != 0) {
482 if (s->crypt_method) {
484 * For encrypted images, read everything into a temporary
485 * contiguous buffer on which the AES functions can work.
489 qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
492 assert(cur_nr_sectors <=
493 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
494 qemu_iovec_reset(&hd_qiov);
495 qemu_iovec_add(&hd_qiov, cluster_data,
496 512 * cur_nr_sectors);
499 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
500 qemu_co_mutex_unlock(&s->lock);
501 ret = bdrv_co_readv(bs->file,
502 (cluster_offset >> 9) + index_in_cluster,
503 cur_nr_sectors, &hd_qiov);
504 qemu_co_mutex_lock(&s->lock);
508 if (s->crypt_method) {
509 qcow2_encrypt_sectors(s, sector_num, cluster_data,
510 cluster_data, cur_nr_sectors, 0, &s->aes_decrypt_key);
511 qemu_iovec_reset(&hd_qiov);
512 qemu_iovec_copy(&hd_qiov, qiov, bytes_done,
513 cur_nr_sectors * 512);
514 qemu_iovec_from_buffer(&hd_qiov, cluster_data,
515 512 * cur_nr_sectors);
519 remaining_sectors -= cur_nr_sectors;
520 sector_num += cur_nr_sectors;
521 bytes_done += cur_nr_sectors * 512;
526 qemu_co_mutex_unlock(&s->lock);
528 qemu_iovec_destroy(&hd_qiov);
529 qemu_vfree(cluster_data);
534 static void run_dependent_requests(BDRVQcowState *s, QCowL2Meta *m)
536 /* Take the request off the list of running requests */
537 if (m->nb_clusters != 0) {
538 QLIST_REMOVE(m, next_in_flight);
541 /* Restart all dependent requests */
542 if (!qemu_co_queue_empty(&m->dependent_requests)) {
543 qemu_co_mutex_unlock(&s->lock);
544 qemu_co_queue_restart_all(&m->dependent_requests);
545 qemu_co_mutex_lock(&s->lock);
549 static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
551 int remaining_sectors,
554 BDRVQcowState *s = bs->opaque;
555 int index_in_cluster;
558 int cur_nr_sectors; /* number of sectors in current iteration */
559 uint64_t cluster_offset;
560 QEMUIOVector hd_qiov;
561 uint64_t bytes_done = 0;
562 uint8_t *cluster_data = NULL;
563 QCowL2Meta l2meta = {
567 qemu_co_queue_init(&l2meta.dependent_requests);
569 qemu_iovec_init(&hd_qiov, qiov->niov);
571 s->cluster_cache_offset = -1; /* disable compressed cache */
573 qemu_co_mutex_lock(&s->lock);
575 while (remaining_sectors != 0) {
577 index_in_cluster = sector_num & (s->cluster_sectors - 1);
578 n_end = index_in_cluster + remaining_sectors;
579 if (s->crypt_method &&
580 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors) {
581 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
584 ret = qcow2_alloc_cluster_offset(bs, sector_num << 9,
585 index_in_cluster, n_end, &cur_nr_sectors, &l2meta);
590 cluster_offset = l2meta.cluster_offset;
591 assert((cluster_offset & 511) == 0);
593 qemu_iovec_reset(&hd_qiov);
594 qemu_iovec_copy(&hd_qiov, qiov, bytes_done,
595 cur_nr_sectors * 512);
597 if (s->crypt_method) {
599 cluster_data = qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS *
603 assert(hd_qiov.size <=
604 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
605 qemu_iovec_to_buffer(&hd_qiov, cluster_data);
607 qcow2_encrypt_sectors(s, sector_num, cluster_data,
608 cluster_data, cur_nr_sectors, 1, &s->aes_encrypt_key);
610 qemu_iovec_reset(&hd_qiov);
611 qemu_iovec_add(&hd_qiov, cluster_data,
612 cur_nr_sectors * 512);
615 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
616 qemu_co_mutex_unlock(&s->lock);
617 ret = bdrv_co_writev(bs->file,
618 (cluster_offset >> 9) + index_in_cluster,
619 cur_nr_sectors, &hd_qiov);
620 qemu_co_mutex_lock(&s->lock);
625 ret = qcow2_alloc_cluster_link_l2(bs, &l2meta);
630 run_dependent_requests(s, &l2meta);
632 remaining_sectors -= cur_nr_sectors;
633 sector_num += cur_nr_sectors;
634 bytes_done += cur_nr_sectors * 512;
639 run_dependent_requests(s, &l2meta);
641 qemu_co_mutex_unlock(&s->lock);
643 qemu_iovec_destroy(&hd_qiov);
644 qemu_vfree(cluster_data);
649 static void qcow2_close(BlockDriverState *bs)
651 BDRVQcowState *s = bs->opaque;
654 qcow2_cache_flush(bs, s->l2_table_cache);
655 qcow2_cache_flush(bs, s->refcount_block_cache);
657 qcow2_cache_destroy(bs, s->l2_table_cache);
658 qcow2_cache_destroy(bs, s->refcount_block_cache);
660 cleanup_unknown_header_ext(bs);
661 g_free(s->cluster_cache);
662 qemu_vfree(s->cluster_data);
663 qcow2_refcount_close(bs);
664 qcow2_free_snapshots(bs);
667 static void qcow2_invalidate_cache(BlockDriverState *bs)
669 BDRVQcowState *s = bs->opaque;
670 int flags = s->flags;
671 AES_KEY aes_encrypt_key;
672 AES_KEY aes_decrypt_key;
673 uint32_t crypt_method = 0;
676 * Backing files are read-only which makes all of their metadata immutable,
677 * that means we don't have to worry about reopening them here.
680 if (s->crypt_method) {
681 crypt_method = s->crypt_method;
682 memcpy(&aes_encrypt_key, &s->aes_encrypt_key, sizeof(aes_encrypt_key));
683 memcpy(&aes_decrypt_key, &s->aes_decrypt_key, sizeof(aes_decrypt_key));
688 memset(s, 0, sizeof(BDRVQcowState));
689 qcow2_open(bs, flags);
692 s->crypt_method = crypt_method;
693 memcpy(&s->aes_encrypt_key, &aes_encrypt_key, sizeof(aes_encrypt_key));
694 memcpy(&s->aes_decrypt_key, &aes_decrypt_key, sizeof(aes_decrypt_key));
698 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
699 size_t len, size_t buflen)
701 QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
702 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
704 if (buflen < ext_len) {
708 *ext_backing_fmt = (QCowExtension) {
709 .magic = cpu_to_be32(magic),
710 .len = cpu_to_be32(len),
712 memcpy(buf + sizeof(QCowExtension), s, len);
718 * Updates the qcow2 header, including the variable length parts of it, i.e.
719 * the backing file name and all extensions. qcow2 was not designed to allow
720 * such changes, so if we run out of space (we can only use the first cluster)
721 * this function may fail.
723 * Returns 0 on success, -errno in error cases.
725 int qcow2_update_header(BlockDriverState *bs)
727 BDRVQcowState *s = bs->opaque;
730 size_t buflen = s->cluster_size;
733 uint32_t refcount_table_clusters;
734 Qcow2UnknownHeaderExtension *uext;
736 buf = qemu_blockalign(bs, buflen);
737 memset(buf, 0, s->cluster_size);
739 /* Header structure */
740 header = (QCowHeader*) buf;
742 if (buflen < sizeof(*header)) {
747 total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
748 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
750 *header = (QCowHeader) {
751 .magic = cpu_to_be32(QCOW_MAGIC),
752 .version = cpu_to_be32(QCOW_VERSION),
753 .backing_file_offset = 0,
754 .backing_file_size = 0,
755 .cluster_bits = cpu_to_be32(s->cluster_bits),
756 .size = cpu_to_be64(total_size),
757 .crypt_method = cpu_to_be32(s->crypt_method_header),
758 .l1_size = cpu_to_be32(s->l1_size),
759 .l1_table_offset = cpu_to_be64(s->l1_table_offset),
760 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset),
761 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
762 .nb_snapshots = cpu_to_be32(s->nb_snapshots),
763 .snapshots_offset = cpu_to_be64(s->snapshots_offset),
766 buf += sizeof(*header);
767 buflen -= sizeof(*header);
769 /* Backing file format header extension */
770 if (*bs->backing_format) {
771 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
772 bs->backing_format, strlen(bs->backing_format),
782 /* Keep unknown header extensions */
783 QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
784 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
793 /* End of header extensions */
794 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
802 /* Backing file name */
803 if (*bs->backing_file) {
804 size_t backing_file_len = strlen(bs->backing_file);
806 if (buflen < backing_file_len) {
811 strncpy(buf, bs->backing_file, buflen);
813 header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
814 header->backing_file_size = cpu_to_be32(backing_file_len);
817 /* Write the new header */
818 ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
829 static int qcow2_change_backing_file(BlockDriverState *bs,
830 const char *backing_file, const char *backing_fmt)
832 /* Backing file format doesn't make sense without a backing file */
833 if (backing_fmt && !backing_file) {
837 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
838 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
840 return qcow2_update_header(bs);
843 static int preallocate(BlockDriverState *bs)
851 nb_sectors = bdrv_getlength(bs) >> 9;
853 qemu_co_queue_init(&meta.dependent_requests);
854 meta.cluster_offset = 0;
857 num = MIN(nb_sectors, INT_MAX >> 9);
858 ret = qcow2_alloc_cluster_offset(bs, offset, 0, num, &num, &meta);
863 ret = qcow2_alloc_cluster_link_l2(bs, &meta);
865 qcow2_free_any_clusters(bs, meta.cluster_offset, meta.nb_clusters);
869 /* There are no dependent requests, but we need to remove our request
870 * from the list of in-flight requests */
871 run_dependent_requests(bs->opaque, &meta);
873 /* TODO Preallocate data if requested */
880 * It is expected that the image file is large enough to actually contain
881 * all of the allocated clusters (otherwise we get failing reads after
882 * EOF). Extend the image to the last allocated sector.
884 if (meta.cluster_offset != 0) {
887 ret = bdrv_write(bs->file, (meta.cluster_offset >> 9) + num - 1, buf, 1);
896 static int qcow2_create2(const char *filename, int64_t total_size,
897 const char *backing_file, const char *backing_format,
898 int flags, size_t cluster_size, int prealloc,
899 QEMUOptionParameter *options)
901 /* Calculate cluster_bits */
903 cluster_bits = ffs(cluster_size) - 1;
904 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
905 (1 << cluster_bits) != cluster_size)
908 "Cluster size must be a power of two between %d and %dk",
909 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
914 * Open the image file and write a minimal qcow2 header.
916 * We keep things simple and start with a zero-sized image. We also
917 * do without refcount blocks or a L1 table for now. We'll fix the
918 * inconsistency later.
920 * We do need a refcount table because growing the refcount table means
921 * allocating two new refcount blocks - the seconds of which would be at
922 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
923 * size for any qcow2 image.
925 BlockDriverState* bs;
927 uint8_t* refcount_table;
930 ret = bdrv_create_file(filename, options);
935 ret = bdrv_file_open(&bs, filename, BDRV_O_RDWR);
940 /* Write the header */
941 memset(&header, 0, sizeof(header));
942 header.magic = cpu_to_be32(QCOW_MAGIC);
943 header.version = cpu_to_be32(QCOW_VERSION);
944 header.cluster_bits = cpu_to_be32(cluster_bits);
945 header.size = cpu_to_be64(0);
946 header.l1_table_offset = cpu_to_be64(0);
947 header.l1_size = cpu_to_be32(0);
948 header.refcount_table_offset = cpu_to_be64(cluster_size);
949 header.refcount_table_clusters = cpu_to_be32(1);
951 if (flags & BLOCK_FLAG_ENCRYPT) {
952 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
954 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
957 ret = bdrv_pwrite(bs, 0, &header, sizeof(header));
962 /* Write an empty refcount table */
963 refcount_table = g_malloc0(cluster_size);
964 ret = bdrv_pwrite(bs, cluster_size, refcount_table, cluster_size);
965 g_free(refcount_table);
974 * And now open the image and make it consistent first (i.e. increase the
975 * refcount of the cluster that is occupied by the header and the refcount
978 BlockDriver* drv = bdrv_find_format("qcow2");
980 ret = bdrv_open(bs, filename,
981 BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_FLUSH, drv);
986 ret = qcow2_alloc_clusters(bs, 2 * cluster_size);
990 } else if (ret != 0) {
991 error_report("Huh, first cluster in empty image is already in use?");
995 /* Okay, now that we have a valid image, let's give it the right size */
996 ret = bdrv_truncate(bs, total_size * BDRV_SECTOR_SIZE);
1001 /* Want a backing file? There you go.*/
1003 ret = bdrv_change_backing_file(bs, backing_file, backing_format);
1009 /* And if we're supposed to preallocate metadata, do that now */
1011 ret = preallocate(bs);
1023 static int qcow2_create(const char *filename, QEMUOptionParameter *options)
1025 const char *backing_file = NULL;
1026 const char *backing_fmt = NULL;
1027 uint64_t sectors = 0;
1029 size_t cluster_size = DEFAULT_CLUSTER_SIZE;
1032 /* Read out options */
1033 while (options && options->name) {
1034 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1035 sectors = options->value.n / 512;
1036 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
1037 backing_file = options->value.s;
1038 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) {
1039 backing_fmt = options->value.s;
1040 } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
1041 flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
1042 } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
1043 if (options->value.n) {
1044 cluster_size = options->value.n;
1046 } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) {
1047 if (!options->value.s || !strcmp(options->value.s, "off")) {
1049 } else if (!strcmp(options->value.s, "metadata")) {
1052 fprintf(stderr, "Invalid preallocation mode: '%s'\n",
1060 if (backing_file && prealloc) {
1061 fprintf(stderr, "Backing file and preallocation cannot be used at "
1066 return qcow2_create2(filename, sectors, backing_file, backing_fmt, flags,
1067 cluster_size, prealloc, options);
1070 static int qcow2_make_empty(BlockDriverState *bs)
1073 /* XXX: not correct */
1074 BDRVQcowState *s = bs->opaque;
1075 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1078 memset(s->l1_table, 0, l1_length);
1079 if (bdrv_pwrite(bs->file, s->l1_table_offset, s->l1_table, l1_length) < 0)
1081 ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
1090 static coroutine_fn int qcow2_co_discard(BlockDriverState *bs,
1091 int64_t sector_num, int nb_sectors)
1094 BDRVQcowState *s = bs->opaque;
1096 qemu_co_mutex_lock(&s->lock);
1097 ret = qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS,
1099 qemu_co_mutex_unlock(&s->lock);
1103 static int qcow2_truncate(BlockDriverState *bs, int64_t offset)
1105 BDRVQcowState *s = bs->opaque;
1106 int ret, new_l1_size;
1112 /* cannot proceed if image has snapshots */
1113 if (s->nb_snapshots) {
1117 /* shrinking is currently not supported */
1118 if (offset < bs->total_sectors * 512) {
1122 new_l1_size = size_to_l1(s, offset);
1123 ret = qcow2_grow_l1_table(bs, new_l1_size, true);
1128 /* write updated header.size */
1129 offset = cpu_to_be64(offset);
1130 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
1131 &offset, sizeof(uint64_t));
1136 s->l1_vm_state_index = new_l1_size;
1140 /* XXX: put compressed sectors first, then all the cluster aligned
1141 tables to avoid losing bytes in alignment */
1142 static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num,
1143 const uint8_t *buf, int nb_sectors)
1145 BDRVQcowState *s = bs->opaque;
1149 uint64_t cluster_offset;
1151 if (nb_sectors == 0) {
1152 /* align end of file to a sector boundary to ease reading with
1153 sector based I/Os */
1154 cluster_offset = bdrv_getlength(bs->file);
1155 cluster_offset = (cluster_offset + 511) & ~511;
1156 bdrv_truncate(bs->file, cluster_offset);
1160 if (nb_sectors != s->cluster_sectors)
1163 out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1165 /* best compression, small window, no zlib header */
1166 memset(&strm, 0, sizeof(strm));
1167 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1169 9, Z_DEFAULT_STRATEGY);
1175 strm.avail_in = s->cluster_size;
1176 strm.next_in = (uint8_t *)buf;
1177 strm.avail_out = s->cluster_size;
1178 strm.next_out = out_buf;
1180 ret = deflate(&strm, Z_FINISH);
1181 if (ret != Z_STREAM_END && ret != Z_OK) {
1186 out_len = strm.next_out - out_buf;
1190 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1191 /* could not compress: write normal cluster */
1192 ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
1197 cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
1198 sector_num << 9, out_len);
1199 if (!cluster_offset) {
1203 cluster_offset &= s->cluster_offset_mask;
1204 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
1205 ret = bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len);
1217 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
1219 BDRVQcowState *s = bs->opaque;
1222 qemu_co_mutex_lock(&s->lock);
1223 ret = qcow2_cache_flush(bs, s->l2_table_cache);
1225 qemu_co_mutex_unlock(&s->lock);
1229 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
1231 qemu_co_mutex_unlock(&s->lock);
1234 qemu_co_mutex_unlock(&s->lock);
1239 static coroutine_fn int qcow2_co_flush_to_disk(BlockDriverState *bs)
1241 return bdrv_co_flush(bs->file);
1244 static int64_t qcow2_vm_state_offset(BDRVQcowState *s)
1246 return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits);
1249 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1251 BDRVQcowState *s = bs->opaque;
1252 bdi->cluster_size = s->cluster_size;
1253 bdi->vm_state_offset = qcow2_vm_state_offset(s);
1258 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result)
1260 return qcow2_check_refcounts(bs, result);
1264 static void dump_refcounts(BlockDriverState *bs)
1266 BDRVQcowState *s = bs->opaque;
1267 int64_t nb_clusters, k, k1, size;
1270 size = bdrv_getlength(bs->file);
1271 nb_clusters = size_to_clusters(s, size);
1272 for(k = 0; k < nb_clusters;) {
1274 refcount = get_refcount(bs, k);
1276 while (k < nb_clusters && get_refcount(bs, k) == refcount)
1278 printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
1284 static int qcow2_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1285 int64_t pos, int size)
1287 BDRVQcowState *s = bs->opaque;
1288 int growable = bs->growable;
1291 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
1293 ret = bdrv_pwrite(bs, qcow2_vm_state_offset(s) + pos, buf, size);
1294 bs->growable = growable;
1299 static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1300 int64_t pos, int size)
1302 BDRVQcowState *s = bs->opaque;
1303 int growable = bs->growable;
1306 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
1308 ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size);
1309 bs->growable = growable;
1314 static QEMUOptionParameter qcow2_create_options[] = {
1316 .name = BLOCK_OPT_SIZE,
1318 .help = "Virtual disk size"
1321 .name = BLOCK_OPT_BACKING_FILE,
1323 .help = "File name of a base image"
1326 .name = BLOCK_OPT_BACKING_FMT,
1328 .help = "Image format of the base image"
1331 .name = BLOCK_OPT_ENCRYPT,
1333 .help = "Encrypt the image"
1336 .name = BLOCK_OPT_CLUSTER_SIZE,
1338 .help = "qcow2 cluster size",
1339 .value = { .n = DEFAULT_CLUSTER_SIZE },
1342 .name = BLOCK_OPT_PREALLOC,
1344 .help = "Preallocation mode (allowed values: off, metadata)"
1349 static BlockDriver bdrv_qcow2 = {
1350 .format_name = "qcow2",
1351 .instance_size = sizeof(BDRVQcowState),
1352 .bdrv_probe = qcow2_probe,
1353 .bdrv_open = qcow2_open,
1354 .bdrv_close = qcow2_close,
1355 .bdrv_create = qcow2_create,
1356 .bdrv_co_is_allocated = qcow2_co_is_allocated,
1357 .bdrv_set_key = qcow2_set_key,
1358 .bdrv_make_empty = qcow2_make_empty,
1360 .bdrv_co_readv = qcow2_co_readv,
1361 .bdrv_co_writev = qcow2_co_writev,
1362 .bdrv_co_flush_to_os = qcow2_co_flush_to_os,
1363 .bdrv_co_flush_to_disk = qcow2_co_flush_to_disk,
1365 .bdrv_co_discard = qcow2_co_discard,
1366 .bdrv_truncate = qcow2_truncate,
1367 .bdrv_write_compressed = qcow2_write_compressed,
1369 .bdrv_snapshot_create = qcow2_snapshot_create,
1370 .bdrv_snapshot_goto = qcow2_snapshot_goto,
1371 .bdrv_snapshot_delete = qcow2_snapshot_delete,
1372 .bdrv_snapshot_list = qcow2_snapshot_list,
1373 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
1374 .bdrv_get_info = qcow2_get_info,
1376 .bdrv_save_vmstate = qcow2_save_vmstate,
1377 .bdrv_load_vmstate = qcow2_load_vmstate,
1379 .bdrv_change_backing_file = qcow2_change_backing_file,
1381 .bdrv_invalidate_cache = qcow2_invalidate_cache,
1383 .create_options = qcow2_create_options,
1384 .bdrv_check = qcow2_check,
1387 static void bdrv_qcow2_init(void)
1389 bdrv_register(&bdrv_qcow2);
1392 block_init(bdrv_qcow2_init);