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,
84 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
86 offset = start_offset;
87 while (offset < end_offset) {
90 BDRVQcowState *s = bs->opaque;
92 if (offset > s->cluster_size)
93 printf("qcow2_read_extension: suspicious offset %lu\n", offset);
95 printf("attemting to read extended header in offset %lu\n", offset);
98 if (bdrv_pread(bs->file, offset, &ext, sizeof(ext)) != sizeof(ext)) {
99 fprintf(stderr, "qcow2_read_extension: ERROR: "
100 "pread fail from offset %" PRIu64 "\n",
104 be32_to_cpus(&ext.magic);
105 be32_to_cpus(&ext.len);
106 offset += sizeof(ext);
108 printf("ext.magic = 0x%x\n", ext.magic);
111 case QCOW2_EXT_MAGIC_END:
114 case QCOW2_EXT_MAGIC_BACKING_FORMAT:
115 if (ext.len >= sizeof(bs->backing_format)) {
116 fprintf(stderr, "ERROR: ext_backing_format: len=%u too large"
118 ext.len, sizeof(bs->backing_format));
121 if (bdrv_pread(bs->file, offset , bs->backing_format,
124 bs->backing_format[ext.len] = '\0';
126 printf("Qcow2: Got format extension %s\n", bs->backing_format);
128 offset = ((offset + ext.len + 7) & ~7);
132 /* unknown magic -- just skip it */
133 offset = ((offset + ext.len + 7) & ~7);
142 static int qcow2_open(BlockDriverState *bs, int flags)
144 BDRVQcowState *s = bs->opaque;
150 ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
154 be32_to_cpus(&header.magic);
155 be32_to_cpus(&header.version);
156 be64_to_cpus(&header.backing_file_offset);
157 be32_to_cpus(&header.backing_file_size);
158 be64_to_cpus(&header.size);
159 be32_to_cpus(&header.cluster_bits);
160 be32_to_cpus(&header.crypt_method);
161 be64_to_cpus(&header.l1_table_offset);
162 be32_to_cpus(&header.l1_size);
163 be64_to_cpus(&header.refcount_table_offset);
164 be32_to_cpus(&header.refcount_table_clusters);
165 be64_to_cpus(&header.snapshots_offset);
166 be32_to_cpus(&header.nb_snapshots);
168 if (header.magic != QCOW_MAGIC) {
172 if (header.version != QCOW_VERSION) {
174 snprintf(version, sizeof(version), "QCOW version %d", header.version);
175 qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
176 bs->device_name, "qcow2", version);
180 if (header.cluster_bits < MIN_CLUSTER_BITS ||
181 header.cluster_bits > MAX_CLUSTER_BITS) {
185 if (header.crypt_method > QCOW_CRYPT_AES) {
189 s->crypt_method_header = header.crypt_method;
190 if (s->crypt_method_header) {
193 s->cluster_bits = header.cluster_bits;
194 s->cluster_size = 1 << s->cluster_bits;
195 s->cluster_sectors = 1 << (s->cluster_bits - 9);
196 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
197 s->l2_size = 1 << s->l2_bits;
198 bs->total_sectors = header.size / 512;
199 s->csize_shift = (62 - (s->cluster_bits - 8));
200 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
201 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
202 s->refcount_table_offset = header.refcount_table_offset;
203 s->refcount_table_size =
204 header.refcount_table_clusters << (s->cluster_bits - 3);
206 s->snapshots_offset = header.snapshots_offset;
207 s->nb_snapshots = header.nb_snapshots;
209 /* read the level 1 table */
210 s->l1_size = header.l1_size;
211 s->l1_vm_state_index = size_to_l1(s, header.size);
212 /* the L1 table must contain at least enough entries to put
214 if (s->l1_size < s->l1_vm_state_index) {
218 s->l1_table_offset = header.l1_table_offset;
219 if (s->l1_size > 0) {
220 s->l1_table = g_malloc0(
221 align_offset(s->l1_size * sizeof(uint64_t), 512));
222 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
223 s->l1_size * sizeof(uint64_t));
227 for(i = 0;i < s->l1_size; i++) {
228 be64_to_cpus(&s->l1_table[i]);
232 /* alloc L2 table/refcount block cache */
233 writethrough = ((flags & BDRV_O_CACHE_WB) == 0);
234 s->l2_table_cache = qcow2_cache_create(bs, L2_CACHE_SIZE, writethrough);
235 s->refcount_block_cache = qcow2_cache_create(bs, REFCOUNT_CACHE_SIZE,
238 s->cluster_cache = g_malloc(s->cluster_size);
239 /* one more sector for decompressed data alignment */
240 s->cluster_data = g_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
242 s->cluster_cache_offset = -1;
244 ret = qcow2_refcount_init(bs);
249 QLIST_INIT(&s->cluster_allocs);
251 /* read qcow2 extensions */
252 if (header.backing_file_offset) {
253 ext_end = header.backing_file_offset;
255 ext_end = s->cluster_size;
257 if (qcow2_read_extensions(bs, sizeof(header), ext_end)) {
262 /* read the backing file name */
263 if (header.backing_file_offset != 0) {
264 len = header.backing_file_size;
268 ret = bdrv_pread(bs->file, header.backing_file_offset,
269 bs->backing_file, len);
273 bs->backing_file[len] = '\0';
275 if (qcow2_read_snapshots(bs) < 0) {
280 /* Initialise locks */
281 qemu_co_mutex_init(&s->lock);
285 BdrvCheckResult result = {0};
286 qcow2_check_refcounts(bs, &result);
292 qcow2_free_snapshots(bs);
293 qcow2_refcount_close(bs);
295 if (s->l2_table_cache) {
296 qcow2_cache_destroy(bs, s->l2_table_cache);
298 g_free(s->cluster_cache);
299 g_free(s->cluster_data);
303 static int qcow2_set_key(BlockDriverState *bs, const char *key)
305 BDRVQcowState *s = bs->opaque;
309 memset(keybuf, 0, 16);
313 /* XXX: we could compress the chars to 7 bits to increase
315 for(i = 0;i < len;i++) {
318 s->crypt_method = s->crypt_method_header;
320 if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
322 if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
332 AES_encrypt(in, tmp, &s->aes_encrypt_key);
333 AES_decrypt(tmp, out, &s->aes_decrypt_key);
334 for(i = 0; i < 16; i++)
335 printf(" %02x", tmp[i]);
337 for(i = 0; i < 16; i++)
338 printf(" %02x", out[i]);
345 static int qcow2_is_allocated(BlockDriverState *bs, int64_t sector_num,
346 int nb_sectors, int *pnum)
348 uint64_t cluster_offset;
352 /* FIXME We can get errors here, but the bdrv_is_allocated interface can't
353 * pass them on today */
354 ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset);
359 return (cluster_offset != 0);
362 /* handle reading after the end of the backing file */
363 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
364 int64_t sector_num, int nb_sectors)
367 if ((sector_num + nb_sectors) <= bs->total_sectors)
369 if (sector_num >= bs->total_sectors)
372 n1 = bs->total_sectors - sector_num;
374 qemu_iovec_memset_skip(qiov, 0, 512 * (nb_sectors - n1), 512 * n1);
379 typedef struct QCowAIOCB {
380 BlockDriverAIOCB common;
383 int remaining_sectors;
384 int cur_nr_sectors; /* number of sectors in current iteration */
386 uint64_t cluster_offset;
387 uint8_t *cluster_data;
389 QEMUIOVector hd_qiov;
392 QLIST_ENTRY(QCowAIOCB) next_depend;
396 * Returns 0 when the request is completed successfully, 1 when there is still
397 * a part left to do and -errno in error cases.
399 static int qcow2_aio_read_cb(QCowAIOCB *acb)
401 BlockDriverState *bs = acb->common.bs;
402 BDRVQcowState *s = bs->opaque;
403 int index_in_cluster, n1;
406 /* post process the read buffer */
407 if (!acb->cluster_offset) {
409 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
412 if (s->crypt_method) {
413 qcow2_encrypt_sectors(s, acb->sector_num, acb->cluster_data,
414 acb->cluster_data, acb->cur_nr_sectors, 0, &s->aes_decrypt_key);
415 qemu_iovec_reset(&acb->hd_qiov);
416 qemu_iovec_copy(&acb->hd_qiov, acb->qiov, acb->bytes_done,
417 acb->cur_nr_sectors * 512);
418 qemu_iovec_from_buffer(&acb->hd_qiov, acb->cluster_data,
419 512 * acb->cur_nr_sectors);
423 acb->remaining_sectors -= acb->cur_nr_sectors;
424 acb->sector_num += acb->cur_nr_sectors;
425 acb->bytes_done += acb->cur_nr_sectors * 512;
427 if (acb->remaining_sectors == 0) {
428 /* request completed */
432 /* prepare next AIO request */
433 acb->cur_nr_sectors = acb->remaining_sectors;
434 if (s->crypt_method) {
435 acb->cur_nr_sectors = MIN(acb->cur_nr_sectors,
436 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
439 ret = qcow2_get_cluster_offset(bs, acb->sector_num << 9,
440 &acb->cur_nr_sectors, &acb->cluster_offset);
445 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
447 qemu_iovec_reset(&acb->hd_qiov);
448 qemu_iovec_copy(&acb->hd_qiov, acb->qiov, acb->bytes_done,
449 acb->cur_nr_sectors * 512);
451 if (!acb->cluster_offset) {
453 if (bs->backing_hd) {
454 /* read from the base image */
455 n1 = qcow2_backing_read1(bs->backing_hd, &acb->hd_qiov,
456 acb->sector_num, acb->cur_nr_sectors);
458 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
459 qemu_co_mutex_unlock(&s->lock);
460 ret = bdrv_co_readv(bs->backing_hd, acb->sector_num,
462 qemu_co_mutex_lock(&s->lock);
469 /* Note: in this case, no need to wait */
470 qemu_iovec_memset(&acb->hd_qiov, 0, 512 * acb->cur_nr_sectors);
473 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
474 /* add AIO support for compressed blocks ? */
475 ret = qcow2_decompress_cluster(bs, acb->cluster_offset);
480 qemu_iovec_from_buffer(&acb->hd_qiov,
481 s->cluster_cache + index_in_cluster * 512,
482 512 * acb->cur_nr_sectors);
486 if ((acb->cluster_offset & 511) != 0) {
490 if (s->crypt_method) {
492 * For encrypted images, read everything into a temporary
493 * contiguous buffer on which the AES functions can work.
495 if (!acb->cluster_data) {
497 g_malloc0(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
500 assert(acb->cur_nr_sectors <=
501 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
502 qemu_iovec_reset(&acb->hd_qiov);
503 qemu_iovec_add(&acb->hd_qiov, acb->cluster_data,
504 512 * acb->cur_nr_sectors);
507 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
508 qemu_co_mutex_unlock(&s->lock);
509 ret = bdrv_co_readv(bs->file,
510 (acb->cluster_offset >> 9) + index_in_cluster,
511 acb->cur_nr_sectors, &acb->hd_qiov);
512 qemu_co_mutex_lock(&s->lock);
521 static QCowAIOCB *qcow2_aio_setup(BlockDriverState *bs, int64_t sector_num,
522 QEMUIOVector *qiov, int nb_sectors,
523 BlockDriverCompletionFunc *cb,
524 void *opaque, int is_write, QCowAIOCB *acb)
526 memset(acb, 0, sizeof(*acb));
528 acb->sector_num = sector_num;
530 acb->is_write = is_write;
532 qemu_iovec_init(&acb->hd_qiov, qiov->niov);
535 acb->remaining_sectors = nb_sectors;
536 acb->cur_nr_sectors = 0;
537 acb->cluster_offset = 0;
538 acb->l2meta.nb_clusters = 0;
539 qemu_co_queue_init(&acb->l2meta.dependent_requests);
543 static int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num,
544 int nb_sectors, QEMUIOVector *qiov)
546 BDRVQcowState *s = bs->opaque;
550 qcow2_aio_setup(bs, sector_num, qiov, nb_sectors, NULL, NULL, 0, &acb);
552 qemu_co_mutex_lock(&s->lock);
554 ret = qcow2_aio_read_cb(&acb);
556 qemu_co_mutex_unlock(&s->lock);
558 qemu_iovec_destroy(&acb.hd_qiov);
563 static void run_dependent_requests(BDRVQcowState *s, QCowL2Meta *m)
565 /* Take the request off the list of running requests */
566 if (m->nb_clusters != 0) {
567 QLIST_REMOVE(m, next_in_flight);
570 /* Restart all dependent requests */
571 if (!qemu_co_queue_empty(&m->dependent_requests)) {
572 qemu_co_mutex_unlock(&s->lock);
573 while(qemu_co_queue_next(&m->dependent_requests));
574 qemu_co_mutex_lock(&s->lock);
579 * Returns 0 when the request is completed successfully, 1 when there is still
580 * a part left to do and -errno in error cases.
582 static int qcow2_aio_write_cb(QCowAIOCB *acb)
584 BlockDriverState *bs = acb->common.bs;
585 BDRVQcowState *s = bs->opaque;
586 int index_in_cluster;
590 ret = qcow2_alloc_cluster_link_l2(bs, &acb->l2meta);
592 run_dependent_requests(s, &acb->l2meta);
598 acb->remaining_sectors -= acb->cur_nr_sectors;
599 acb->sector_num += acb->cur_nr_sectors;
600 acb->bytes_done += acb->cur_nr_sectors * 512;
602 if (acb->remaining_sectors == 0) {
603 /* request completed */
607 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
608 n_end = index_in_cluster + acb->remaining_sectors;
609 if (s->crypt_method &&
610 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
611 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
613 ret = qcow2_alloc_cluster_offset(bs, acb->sector_num << 9,
614 index_in_cluster, n_end, &acb->cur_nr_sectors, &acb->l2meta);
619 acb->cluster_offset = acb->l2meta.cluster_offset;
620 assert((acb->cluster_offset & 511) == 0);
622 qemu_iovec_reset(&acb->hd_qiov);
623 qemu_iovec_copy(&acb->hd_qiov, acb->qiov, acb->bytes_done,
624 acb->cur_nr_sectors * 512);
626 if (s->crypt_method) {
627 if (!acb->cluster_data) {
628 acb->cluster_data = g_malloc0(QCOW_MAX_CRYPT_CLUSTERS *
632 assert(acb->hd_qiov.size <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
633 qemu_iovec_to_buffer(&acb->hd_qiov, acb->cluster_data);
635 qcow2_encrypt_sectors(s, acb->sector_num, acb->cluster_data,
636 acb->cluster_data, acb->cur_nr_sectors, 1, &s->aes_encrypt_key);
638 qemu_iovec_reset(&acb->hd_qiov);
639 qemu_iovec_add(&acb->hd_qiov, acb->cluster_data,
640 acb->cur_nr_sectors * 512);
643 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
644 qemu_co_mutex_unlock(&s->lock);
645 ret = bdrv_co_writev(bs->file,
646 (acb->cluster_offset >> 9) + index_in_cluster,
647 acb->cur_nr_sectors, &acb->hd_qiov);
648 qemu_co_mutex_lock(&s->lock);
656 static int qcow2_co_writev(BlockDriverState *bs,
661 BDRVQcowState *s = bs->opaque;
665 qcow2_aio_setup(bs, sector_num, qiov, nb_sectors, NULL, NULL, 1, &acb);
666 s->cluster_cache_offset = -1; /* disable compressed cache */
668 qemu_co_mutex_lock(&s->lock);
670 ret = qcow2_aio_write_cb(&acb);
672 qemu_co_mutex_unlock(&s->lock);
674 qemu_iovec_destroy(&acb.hd_qiov);
679 static void qcow2_close(BlockDriverState *bs)
681 BDRVQcowState *s = bs->opaque;
684 qcow2_cache_flush(bs, s->l2_table_cache);
685 qcow2_cache_flush(bs, s->refcount_block_cache);
687 qcow2_cache_destroy(bs, s->l2_table_cache);
688 qcow2_cache_destroy(bs, s->refcount_block_cache);
690 g_free(s->cluster_cache);
691 g_free(s->cluster_data);
692 qcow2_refcount_close(bs);
696 * Updates the variable length parts of the qcow2 header, i.e. the backing file
697 * name and all extensions. qcow2 was not designed to allow such changes, so if
698 * we run out of space (we can only use the first cluster) this function may
701 * Returns 0 on success, -errno in error cases.
703 static int qcow2_update_ext_header(BlockDriverState *bs,
704 const char *backing_file, const char *backing_fmt)
706 size_t backing_file_len = 0;
707 size_t backing_fmt_len = 0;
708 BDRVQcowState *s = bs->opaque;
709 QCowExtension ext_backing_fmt = {0, 0};
712 /* Backing file format doesn't make sense without a backing file */
713 if (backing_fmt && !backing_file) {
717 /* Prepare the backing file format extension if needed */
719 ext_backing_fmt.len = cpu_to_be32(strlen(backing_fmt));
720 ext_backing_fmt.magic = cpu_to_be32(QCOW2_EXT_MAGIC_BACKING_FORMAT);
721 backing_fmt_len = ((sizeof(ext_backing_fmt)
722 + strlen(backing_fmt) + 7) & ~7);
725 /* Check if we can fit the new header into the first cluster */
727 backing_file_len = strlen(backing_file);
730 size_t header_size = sizeof(QCowHeader) + backing_file_len
733 if (header_size > s->cluster_size) {
737 /* Rewrite backing file name and qcow2 extensions */
738 size_t ext_size = header_size - sizeof(QCowHeader);
739 uint8_t buf[ext_size];
741 size_t backing_file_offset = 0;
745 int padding = backing_fmt_len -
746 (sizeof(ext_backing_fmt) + strlen(backing_fmt));
748 memcpy(buf + offset, &ext_backing_fmt, sizeof(ext_backing_fmt));
749 offset += sizeof(ext_backing_fmt);
751 memcpy(buf + offset, backing_fmt, strlen(backing_fmt));
752 offset += strlen(backing_fmt);
754 memset(buf + offset, 0, padding);
758 memcpy(buf + offset, backing_file, backing_file_len);
759 backing_file_offset = sizeof(QCowHeader) + offset;
762 ret = bdrv_pwrite_sync(bs->file, sizeof(QCowHeader), buf, ext_size);
767 /* Update header fields */
768 uint64_t be_backing_file_offset = cpu_to_be64(backing_file_offset);
769 uint32_t be_backing_file_size = cpu_to_be32(backing_file_len);
771 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, backing_file_offset),
772 &be_backing_file_offset, sizeof(uint64_t));
777 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, backing_file_size),
778 &be_backing_file_size, sizeof(uint32_t));
788 static int qcow2_change_backing_file(BlockDriverState *bs,
789 const char *backing_file, const char *backing_fmt)
791 return qcow2_update_ext_header(bs, backing_file, backing_fmt);
794 static int preallocate(BlockDriverState *bs)
802 nb_sectors = bdrv_getlength(bs) >> 9;
804 qemu_co_queue_init(&meta.dependent_requests);
805 meta.cluster_offset = 0;
808 num = MIN(nb_sectors, INT_MAX >> 9);
809 ret = qcow2_alloc_cluster_offset(bs, offset, 0, num, &num, &meta);
814 ret = qcow2_alloc_cluster_link_l2(bs, &meta);
816 qcow2_free_any_clusters(bs, meta.cluster_offset, meta.nb_clusters);
820 /* There are no dependent requests, but we need to remove our request
821 * from the list of in-flight requests */
822 run_dependent_requests(bs->opaque, &meta);
824 /* TODO Preallocate data if requested */
831 * It is expected that the image file is large enough to actually contain
832 * all of the allocated clusters (otherwise we get failing reads after
833 * EOF). Extend the image to the last allocated sector.
835 if (meta.cluster_offset != 0) {
838 ret = bdrv_write(bs->file, (meta.cluster_offset >> 9) + num - 1, buf, 1);
847 static int qcow2_create2(const char *filename, int64_t total_size,
848 const char *backing_file, const char *backing_format,
849 int flags, size_t cluster_size, int prealloc,
850 QEMUOptionParameter *options)
852 /* Calulate cluster_bits */
854 cluster_bits = ffs(cluster_size) - 1;
855 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
856 (1 << cluster_bits) != cluster_size)
859 "Cluster size must be a power of two between %d and %dk",
860 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
865 * Open the image file and write a minimal qcow2 header.
867 * We keep things simple and start with a zero-sized image. We also
868 * do without refcount blocks or a L1 table for now. We'll fix the
869 * inconsistency later.
871 * We do need a refcount table because growing the refcount table means
872 * allocating two new refcount blocks - the seconds of which would be at
873 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
874 * size for any qcow2 image.
876 BlockDriverState* bs;
878 uint8_t* refcount_table;
881 ret = bdrv_create_file(filename, options);
886 ret = bdrv_file_open(&bs, filename, BDRV_O_RDWR);
891 /* Write the header */
892 memset(&header, 0, sizeof(header));
893 header.magic = cpu_to_be32(QCOW_MAGIC);
894 header.version = cpu_to_be32(QCOW_VERSION);
895 header.cluster_bits = cpu_to_be32(cluster_bits);
896 header.size = cpu_to_be64(0);
897 header.l1_table_offset = cpu_to_be64(0);
898 header.l1_size = cpu_to_be32(0);
899 header.refcount_table_offset = cpu_to_be64(cluster_size);
900 header.refcount_table_clusters = cpu_to_be32(1);
902 if (flags & BLOCK_FLAG_ENCRYPT) {
903 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
905 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
908 ret = bdrv_pwrite(bs, 0, &header, sizeof(header));
913 /* Write an empty refcount table */
914 refcount_table = g_malloc0(cluster_size);
915 ret = bdrv_pwrite(bs, cluster_size, refcount_table, cluster_size);
916 g_free(refcount_table);
925 * And now open the image and make it consistent first (i.e. increase the
926 * refcount of the cluster that is occupied by the header and the refcount
929 BlockDriver* drv = bdrv_find_format("qcow2");
931 ret = bdrv_open(bs, filename,
932 BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_FLUSH, drv);
937 ret = qcow2_alloc_clusters(bs, 2 * cluster_size);
941 } else if (ret != 0) {
942 error_report("Huh, first cluster in empty image is already in use?");
946 /* Okay, now that we have a valid image, let's give it the right size */
947 ret = bdrv_truncate(bs, total_size * BDRV_SECTOR_SIZE);
952 /* Want a backing file? There you go.*/
954 ret = bdrv_change_backing_file(bs, backing_file, backing_format);
960 /* And if we're supposed to preallocate metadata, do that now */
962 ret = preallocate(bs);
974 static int qcow2_create(const char *filename, QEMUOptionParameter *options)
976 const char *backing_file = NULL;
977 const char *backing_fmt = NULL;
978 uint64_t sectors = 0;
980 size_t cluster_size = DEFAULT_CLUSTER_SIZE;
983 /* Read out options */
984 while (options && options->name) {
985 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
986 sectors = options->value.n / 512;
987 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
988 backing_file = options->value.s;
989 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) {
990 backing_fmt = options->value.s;
991 } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
992 flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
993 } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
994 if (options->value.n) {
995 cluster_size = options->value.n;
997 } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) {
998 if (!options->value.s || !strcmp(options->value.s, "off")) {
1000 } else if (!strcmp(options->value.s, "metadata")) {
1003 fprintf(stderr, "Invalid preallocation mode: '%s'\n",
1011 if (backing_file && prealloc) {
1012 fprintf(stderr, "Backing file and preallocation cannot be used at "
1017 return qcow2_create2(filename, sectors, backing_file, backing_fmt, flags,
1018 cluster_size, prealloc, options);
1021 static int qcow2_make_empty(BlockDriverState *bs)
1024 /* XXX: not correct */
1025 BDRVQcowState *s = bs->opaque;
1026 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1029 memset(s->l1_table, 0, l1_length);
1030 if (bdrv_pwrite(bs->file, s->l1_table_offset, s->l1_table, l1_length) < 0)
1032 ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
1041 static int qcow2_discard(BlockDriverState *bs, int64_t sector_num,
1044 return qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS,
1048 static int qcow2_truncate(BlockDriverState *bs, int64_t offset)
1050 BDRVQcowState *s = bs->opaque;
1051 int ret, new_l1_size;
1057 /* cannot proceed if image has snapshots */
1058 if (s->nb_snapshots) {
1062 /* shrinking is currently not supported */
1063 if (offset < bs->total_sectors * 512) {
1067 new_l1_size = size_to_l1(s, offset);
1068 ret = qcow2_grow_l1_table(bs, new_l1_size, true);
1073 /* write updated header.size */
1074 offset = cpu_to_be64(offset);
1075 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
1076 &offset, sizeof(uint64_t));
1081 s->l1_vm_state_index = new_l1_size;
1085 /* XXX: put compressed sectors first, then all the cluster aligned
1086 tables to avoid losing bytes in alignment */
1087 static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num,
1088 const uint8_t *buf, int nb_sectors)
1090 BDRVQcowState *s = bs->opaque;
1094 uint64_t cluster_offset;
1096 if (nb_sectors == 0) {
1097 /* align end of file to a sector boundary to ease reading with
1098 sector based I/Os */
1099 cluster_offset = bdrv_getlength(bs->file);
1100 cluster_offset = (cluster_offset + 511) & ~511;
1101 bdrv_truncate(bs->file, cluster_offset);
1105 if (nb_sectors != s->cluster_sectors)
1108 out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1110 /* best compression, small window, no zlib header */
1111 memset(&strm, 0, sizeof(strm));
1112 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1114 9, Z_DEFAULT_STRATEGY);
1120 strm.avail_in = s->cluster_size;
1121 strm.next_in = (uint8_t *)buf;
1122 strm.avail_out = s->cluster_size;
1123 strm.next_out = out_buf;
1125 ret = deflate(&strm, Z_FINISH);
1126 if (ret != Z_STREAM_END && ret != Z_OK) {
1131 out_len = strm.next_out - out_buf;
1135 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1136 /* could not compress: write normal cluster */
1137 bdrv_write(bs, sector_num, buf, s->cluster_sectors);
1139 cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
1140 sector_num << 9, out_len);
1141 if (!cluster_offset)
1143 cluster_offset &= s->cluster_offset_mask;
1144 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
1145 if (bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len) != out_len) {
1155 static int qcow2_flush(BlockDriverState *bs)
1157 BDRVQcowState *s = bs->opaque;
1160 ret = qcow2_cache_flush(bs, s->l2_table_cache);
1165 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
1170 return bdrv_flush(bs->file);
1173 static BlockDriverAIOCB *qcow2_aio_flush(BlockDriverState *bs,
1174 BlockDriverCompletionFunc *cb,
1177 BDRVQcowState *s = bs->opaque;
1180 ret = qcow2_cache_flush(bs, s->l2_table_cache);
1185 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
1190 return bdrv_aio_flush(bs->file, cb, opaque);
1193 static int64_t qcow2_vm_state_offset(BDRVQcowState *s)
1195 return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits);
1198 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1200 BDRVQcowState *s = bs->opaque;
1201 bdi->cluster_size = s->cluster_size;
1202 bdi->vm_state_offset = qcow2_vm_state_offset(s);
1207 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result)
1209 return qcow2_check_refcounts(bs, result);
1213 static void dump_refcounts(BlockDriverState *bs)
1215 BDRVQcowState *s = bs->opaque;
1216 int64_t nb_clusters, k, k1, size;
1219 size = bdrv_getlength(bs->file);
1220 nb_clusters = size_to_clusters(s, size);
1221 for(k = 0; k < nb_clusters;) {
1223 refcount = get_refcount(bs, k);
1225 while (k < nb_clusters && get_refcount(bs, k) == refcount)
1227 printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
1233 static int qcow2_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1234 int64_t pos, int size)
1236 BDRVQcowState *s = bs->opaque;
1237 int growable = bs->growable;
1240 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
1242 ret = bdrv_pwrite(bs, qcow2_vm_state_offset(s) + pos, buf, size);
1243 bs->growable = growable;
1248 static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1249 int64_t pos, int size)
1251 BDRVQcowState *s = bs->opaque;
1252 int growable = bs->growable;
1255 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
1257 ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size);
1258 bs->growable = growable;
1263 static QEMUOptionParameter qcow2_create_options[] = {
1265 .name = BLOCK_OPT_SIZE,
1267 .help = "Virtual disk size"
1270 .name = BLOCK_OPT_BACKING_FILE,
1272 .help = "File name of a base image"
1275 .name = BLOCK_OPT_BACKING_FMT,
1277 .help = "Image format of the base image"
1280 .name = BLOCK_OPT_ENCRYPT,
1282 .help = "Encrypt the image"
1285 .name = BLOCK_OPT_CLUSTER_SIZE,
1287 .help = "qcow2 cluster size",
1288 .value = { .n = DEFAULT_CLUSTER_SIZE },
1291 .name = BLOCK_OPT_PREALLOC,
1293 .help = "Preallocation mode (allowed values: off, metadata)"
1298 static BlockDriver bdrv_qcow2 = {
1299 .format_name = "qcow2",
1300 .instance_size = sizeof(BDRVQcowState),
1301 .bdrv_probe = qcow2_probe,
1302 .bdrv_open = qcow2_open,
1303 .bdrv_close = qcow2_close,
1304 .bdrv_create = qcow2_create,
1305 .bdrv_flush = qcow2_flush,
1306 .bdrv_is_allocated = qcow2_is_allocated,
1307 .bdrv_set_key = qcow2_set_key,
1308 .bdrv_make_empty = qcow2_make_empty,
1310 .bdrv_co_readv = qcow2_co_readv,
1311 .bdrv_co_writev = qcow2_co_writev,
1312 .bdrv_aio_flush = qcow2_aio_flush,
1314 .bdrv_discard = qcow2_discard,
1315 .bdrv_truncate = qcow2_truncate,
1316 .bdrv_write_compressed = qcow2_write_compressed,
1318 .bdrv_snapshot_create = qcow2_snapshot_create,
1319 .bdrv_snapshot_goto = qcow2_snapshot_goto,
1320 .bdrv_snapshot_delete = qcow2_snapshot_delete,
1321 .bdrv_snapshot_list = qcow2_snapshot_list,
1322 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
1323 .bdrv_get_info = qcow2_get_info,
1325 .bdrv_save_vmstate = qcow2_save_vmstate,
1326 .bdrv_load_vmstate = qcow2_load_vmstate,
1328 .bdrv_change_backing_file = qcow2_change_backing_file,
1330 .create_options = qcow2_create_options,
1331 .bdrv_check = qcow2_check,
1334 static void bdrv_qcow2_init(void)
1336 bdrv_register(&bdrv_qcow2);
1339 block_init(bdrv_qcow2_init);