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"
35 Differences with QCOW:
37 - Support for multiple incremental snapshots.
38 - Memory management by reference counts.
39 - Clusters which have a reference count of one have the bit
40 QCOW_OFLAG_COPIED to optimize write performance.
41 - Size of compressed clusters is stored in sectors to reduce bit usage
42 in the cluster offsets.
43 - Support for storing additional data (such as the VM state) in the
45 - If a backing store is used, the cluster size is not constrained
46 (could be backported to QCOW).
47 - L2 tables have always a size of one cluster.
55 #define QCOW2_EXT_MAGIC_END 0
56 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
57 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
59 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
61 const QCowHeader *cow_header = (const void *)buf;
63 if (buf_size >= sizeof(QCowHeader) &&
64 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
65 be32_to_cpu(cow_header->version) >= 2)
73 * read qcow2 extension and fill bs
74 * start reading from start_offset
75 * finish reading upon magic of value 0 or when end_offset reached
76 * unknown magic is skipped (future extension this version knows nothing about)
77 * return 0 upon success, non-0 otherwise
79 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
80 uint64_t end_offset, void **p_feature_table)
82 BDRVQcowState *s = bs->opaque;
88 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
90 offset = start_offset;
91 while (offset < end_offset) {
95 if (offset > s->cluster_size)
96 printf("qcow2_read_extension: suspicious offset %lu\n", offset);
98 printf("attempting to read extended header in offset %lu\n", offset);
101 if (bdrv_pread(bs->file, offset, &ext, sizeof(ext)) != sizeof(ext)) {
102 fprintf(stderr, "qcow2_read_extension: ERROR: "
103 "pread fail from offset %" PRIu64 "\n",
107 be32_to_cpus(&ext.magic);
108 be32_to_cpus(&ext.len);
109 offset += sizeof(ext);
111 printf("ext.magic = 0x%x\n", ext.magic);
113 if (ext.len > end_offset - offset) {
114 error_report("Header extension too large");
119 case QCOW2_EXT_MAGIC_END:
122 case QCOW2_EXT_MAGIC_BACKING_FORMAT:
123 if (ext.len >= sizeof(bs->backing_format)) {
124 fprintf(stderr, "ERROR: ext_backing_format: len=%u too large"
126 ext.len, sizeof(bs->backing_format));
129 if (bdrv_pread(bs->file, offset , bs->backing_format,
132 bs->backing_format[ext.len] = '\0';
134 printf("Qcow2: Got format extension %s\n", bs->backing_format);
138 case QCOW2_EXT_MAGIC_FEATURE_TABLE:
139 if (p_feature_table != NULL) {
140 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
141 ret = bdrv_pread(bs->file, offset , feature_table, ext.len);
146 *p_feature_table = feature_table;
151 /* unknown magic - save it in case we need to rewrite the header */
153 Qcow2UnknownHeaderExtension *uext;
155 uext = g_malloc0(sizeof(*uext) + ext.len);
156 uext->magic = ext.magic;
158 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
160 ret = bdrv_pread(bs->file, offset , uext->data, uext->len);
168 offset += ((ext.len + 7) & ~7);
174 static void cleanup_unknown_header_ext(BlockDriverState *bs)
176 BDRVQcowState *s = bs->opaque;
177 Qcow2UnknownHeaderExtension *uext, *next;
179 QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
180 QLIST_REMOVE(uext, next);
185 static void GCC_FMT_ATTR(2, 3) report_unsupported(BlockDriverState *bs,
186 const char *fmt, ...)
192 vsnprintf(msg, sizeof(msg), fmt, ap);
195 qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
196 bs->device_name, "qcow2", msg);
199 static void report_unsupported_feature(BlockDriverState *bs,
200 Qcow2Feature *table, uint64_t mask)
202 while (table && table->name[0] != '\0') {
203 if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
204 if (mask & (1 << table->bit)) {
205 report_unsupported(bs, "%.46s",table->name);
206 mask &= ~(1 << table->bit);
213 report_unsupported(bs, "Unknown incompatible feature: %" PRIx64, mask);
217 static int qcow2_open(BlockDriverState *bs, int flags)
219 BDRVQcowState *s = bs->opaque;
225 ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
229 be32_to_cpus(&header.magic);
230 be32_to_cpus(&header.version);
231 be64_to_cpus(&header.backing_file_offset);
232 be32_to_cpus(&header.backing_file_size);
233 be64_to_cpus(&header.size);
234 be32_to_cpus(&header.cluster_bits);
235 be32_to_cpus(&header.crypt_method);
236 be64_to_cpus(&header.l1_table_offset);
237 be32_to_cpus(&header.l1_size);
238 be64_to_cpus(&header.refcount_table_offset);
239 be32_to_cpus(&header.refcount_table_clusters);
240 be64_to_cpus(&header.snapshots_offset);
241 be32_to_cpus(&header.nb_snapshots);
243 if (header.magic != QCOW_MAGIC) {
247 if (header.version < 2 || header.version > 3) {
248 report_unsupported(bs, "QCOW version %d", header.version);
253 s->qcow_version = header.version;
255 /* Initialise version 3 header fields */
256 if (header.version == 2) {
257 header.incompatible_features = 0;
258 header.compatible_features = 0;
259 header.autoclear_features = 0;
260 header.refcount_order = 4;
261 header.header_length = 72;
263 be64_to_cpus(&header.incompatible_features);
264 be64_to_cpus(&header.compatible_features);
265 be64_to_cpus(&header.autoclear_features);
266 be32_to_cpus(&header.refcount_order);
267 be32_to_cpus(&header.header_length);
270 if (header.header_length > sizeof(header)) {
271 s->unknown_header_fields_size = header.header_length - sizeof(header);
272 s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
273 ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields,
274 s->unknown_header_fields_size);
280 if (header.backing_file_offset) {
281 ext_end = header.backing_file_offset;
283 ext_end = 1 << header.cluster_bits;
286 /* Handle feature bits */
287 s->incompatible_features = header.incompatible_features;
288 s->compatible_features = header.compatible_features;
289 s->autoclear_features = header.autoclear_features;
291 if (s->incompatible_features != 0) {
292 void *feature_table = NULL;
293 qcow2_read_extensions(bs, header.header_length, ext_end,
295 report_unsupported_feature(bs, feature_table,
296 s->incompatible_features);
301 if (!bs->read_only && s->autoclear_features != 0) {
302 s->autoclear_features = 0;
303 ret = qcow2_update_header(bs);
309 /* Check support for various header values */
310 if (header.refcount_order != 4) {
311 report_unsupported(bs, "%d bit reference counts",
312 1 << header.refcount_order);
317 if (header.cluster_bits < MIN_CLUSTER_BITS ||
318 header.cluster_bits > MAX_CLUSTER_BITS) {
322 if (header.crypt_method > QCOW_CRYPT_AES) {
326 s->crypt_method_header = header.crypt_method;
327 if (s->crypt_method_header) {
330 s->cluster_bits = header.cluster_bits;
331 s->cluster_size = 1 << s->cluster_bits;
332 s->cluster_sectors = 1 << (s->cluster_bits - 9);
333 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
334 s->l2_size = 1 << s->l2_bits;
335 bs->total_sectors = header.size / 512;
336 s->csize_shift = (62 - (s->cluster_bits - 8));
337 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
338 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
339 s->refcount_table_offset = header.refcount_table_offset;
340 s->refcount_table_size =
341 header.refcount_table_clusters << (s->cluster_bits - 3);
343 s->snapshots_offset = header.snapshots_offset;
344 s->nb_snapshots = header.nb_snapshots;
346 /* read the level 1 table */
347 s->l1_size = header.l1_size;
348 s->l1_vm_state_index = size_to_l1(s, header.size);
349 /* the L1 table must contain at least enough entries to put
351 if (s->l1_size < s->l1_vm_state_index) {
355 s->l1_table_offset = header.l1_table_offset;
356 if (s->l1_size > 0) {
357 s->l1_table = g_malloc0(
358 align_offset(s->l1_size * sizeof(uint64_t), 512));
359 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
360 s->l1_size * sizeof(uint64_t));
364 for(i = 0;i < s->l1_size; i++) {
365 be64_to_cpus(&s->l1_table[i]);
369 /* alloc L2 table/refcount block cache */
370 writethrough = ((flags & BDRV_O_CACHE_WB) == 0);
371 s->l2_table_cache = qcow2_cache_create(bs, L2_CACHE_SIZE, writethrough);
372 s->refcount_block_cache = qcow2_cache_create(bs, REFCOUNT_CACHE_SIZE,
375 s->cluster_cache = g_malloc(s->cluster_size);
376 /* one more sector for decompressed data alignment */
377 s->cluster_data = qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
379 s->cluster_cache_offset = -1;
382 ret = qcow2_refcount_init(bs);
387 QLIST_INIT(&s->cluster_allocs);
389 /* read qcow2 extensions */
390 if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL)) {
395 /* read the backing file name */
396 if (header.backing_file_offset != 0) {
397 len = header.backing_file_size;
401 ret = bdrv_pread(bs->file, header.backing_file_offset,
402 bs->backing_file, len);
406 bs->backing_file[len] = '\0';
409 ret = qcow2_read_snapshots(bs);
414 /* Initialise locks */
415 qemu_co_mutex_init(&s->lock);
419 BdrvCheckResult result = {0};
420 qcow2_check_refcounts(bs, &result);
426 g_free(s->unknown_header_fields);
427 cleanup_unknown_header_ext(bs);
428 qcow2_free_snapshots(bs);
429 qcow2_refcount_close(bs);
431 if (s->l2_table_cache) {
432 qcow2_cache_destroy(bs, s->l2_table_cache);
434 g_free(s->cluster_cache);
435 qemu_vfree(s->cluster_data);
439 static int qcow2_set_key(BlockDriverState *bs, const char *key)
441 BDRVQcowState *s = bs->opaque;
445 memset(keybuf, 0, 16);
449 /* XXX: we could compress the chars to 7 bits to increase
451 for(i = 0;i < len;i++) {
454 s->crypt_method = s->crypt_method_header;
456 if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
458 if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
468 AES_encrypt(in, tmp, &s->aes_encrypt_key);
469 AES_decrypt(tmp, out, &s->aes_decrypt_key);
470 for(i = 0; i < 16; i++)
471 printf(" %02x", tmp[i]);
473 for(i = 0; i < 16; i++)
474 printf(" %02x", out[i]);
481 static int coroutine_fn qcow2_co_is_allocated(BlockDriverState *bs,
482 int64_t sector_num, int nb_sectors, int *pnum)
484 BDRVQcowState *s = bs->opaque;
485 uint64_t cluster_offset;
489 /* FIXME We can get errors here, but the bdrv_co_is_allocated interface
490 * can't pass them on today */
491 qemu_co_mutex_lock(&s->lock);
492 ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset);
493 qemu_co_mutex_unlock(&s->lock);
498 return (cluster_offset != 0);
501 /* handle reading after the end of the backing file */
502 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
503 int64_t sector_num, int nb_sectors)
506 if ((sector_num + nb_sectors) <= bs->total_sectors)
508 if (sector_num >= bs->total_sectors)
511 n1 = bs->total_sectors - sector_num;
513 qemu_iovec_memset_skip(qiov, 0, 512 * (nb_sectors - n1), 512 * n1);
518 static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num,
519 int remaining_sectors, QEMUIOVector *qiov)
521 BDRVQcowState *s = bs->opaque;
522 int index_in_cluster, n1;
524 int cur_nr_sectors; /* number of sectors in current iteration */
525 uint64_t cluster_offset = 0;
526 uint64_t bytes_done = 0;
527 QEMUIOVector hd_qiov;
528 uint8_t *cluster_data = NULL;
530 qemu_iovec_init(&hd_qiov, qiov->niov);
532 qemu_co_mutex_lock(&s->lock);
534 while (remaining_sectors != 0) {
536 /* prepare next request */
537 cur_nr_sectors = remaining_sectors;
538 if (s->crypt_method) {
539 cur_nr_sectors = MIN(cur_nr_sectors,
540 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
543 ret = qcow2_get_cluster_offset(bs, sector_num << 9,
544 &cur_nr_sectors, &cluster_offset);
549 index_in_cluster = sector_num & (s->cluster_sectors - 1);
551 qemu_iovec_reset(&hd_qiov);
552 qemu_iovec_copy(&hd_qiov, qiov, bytes_done,
553 cur_nr_sectors * 512);
556 case QCOW2_CLUSTER_UNALLOCATED:
558 if (bs->backing_hd) {
559 /* read from the base image */
560 n1 = qcow2_backing_read1(bs->backing_hd, &hd_qiov,
561 sector_num, cur_nr_sectors);
563 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
564 qemu_co_mutex_unlock(&s->lock);
565 ret = bdrv_co_readv(bs->backing_hd, sector_num,
567 qemu_co_mutex_lock(&s->lock);
573 /* Note: in this case, no need to wait */
574 qemu_iovec_memset(&hd_qiov, 0, 512 * cur_nr_sectors);
578 case QCOW2_CLUSTER_ZERO:
579 if (s->qcow_version < 3) {
583 qemu_iovec_memset(&hd_qiov, 0, 512 * cur_nr_sectors);
586 case QCOW2_CLUSTER_COMPRESSED:
587 /* add AIO support for compressed blocks ? */
588 ret = qcow2_decompress_cluster(bs, cluster_offset);
593 qemu_iovec_from_buffer(&hd_qiov,
594 s->cluster_cache + index_in_cluster * 512,
595 512 * cur_nr_sectors);
598 case QCOW2_CLUSTER_NORMAL:
599 if ((cluster_offset & 511) != 0) {
604 if (s->crypt_method) {
606 * For encrypted images, read everything into a temporary
607 * contiguous buffer on which the AES functions can work.
611 qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
614 assert(cur_nr_sectors <=
615 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
616 qemu_iovec_reset(&hd_qiov);
617 qemu_iovec_add(&hd_qiov, cluster_data,
618 512 * cur_nr_sectors);
621 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
622 qemu_co_mutex_unlock(&s->lock);
623 ret = bdrv_co_readv(bs->file,
624 (cluster_offset >> 9) + index_in_cluster,
625 cur_nr_sectors, &hd_qiov);
626 qemu_co_mutex_lock(&s->lock);
630 if (s->crypt_method) {
631 qcow2_encrypt_sectors(s, sector_num, cluster_data,
632 cluster_data, cur_nr_sectors, 0, &s->aes_decrypt_key);
633 qemu_iovec_reset(&hd_qiov);
634 qemu_iovec_copy(&hd_qiov, qiov, bytes_done,
635 cur_nr_sectors * 512);
636 qemu_iovec_from_buffer(&hd_qiov, cluster_data,
637 512 * cur_nr_sectors);
642 g_assert_not_reached();
647 remaining_sectors -= cur_nr_sectors;
648 sector_num += cur_nr_sectors;
649 bytes_done += cur_nr_sectors * 512;
654 qemu_co_mutex_unlock(&s->lock);
656 qemu_iovec_destroy(&hd_qiov);
657 qemu_vfree(cluster_data);
662 static void run_dependent_requests(BDRVQcowState *s, QCowL2Meta *m)
664 /* Take the request off the list of running requests */
665 if (m->nb_clusters != 0) {
666 QLIST_REMOVE(m, next_in_flight);
669 /* Restart all dependent requests */
670 if (!qemu_co_queue_empty(&m->dependent_requests)) {
671 qemu_co_mutex_unlock(&s->lock);
672 qemu_co_queue_restart_all(&m->dependent_requests);
673 qemu_co_mutex_lock(&s->lock);
677 static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
679 int remaining_sectors,
682 BDRVQcowState *s = bs->opaque;
683 int index_in_cluster;
686 int cur_nr_sectors; /* number of sectors in current iteration */
687 uint64_t cluster_offset;
688 QEMUIOVector hd_qiov;
689 uint64_t bytes_done = 0;
690 uint8_t *cluster_data = NULL;
691 QCowL2Meta l2meta = {
695 trace_qcow2_writev_start_req(qemu_coroutine_self(), sector_num,
698 qemu_co_queue_init(&l2meta.dependent_requests);
700 qemu_iovec_init(&hd_qiov, qiov->niov);
702 s->cluster_cache_offset = -1; /* disable compressed cache */
704 qemu_co_mutex_lock(&s->lock);
706 while (remaining_sectors != 0) {
708 trace_qcow2_writev_start_part(qemu_coroutine_self());
709 index_in_cluster = sector_num & (s->cluster_sectors - 1);
710 n_end = index_in_cluster + remaining_sectors;
711 if (s->crypt_method &&
712 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors) {
713 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
716 ret = qcow2_alloc_cluster_offset(bs, sector_num << 9,
717 index_in_cluster, n_end, &cur_nr_sectors, &l2meta);
722 cluster_offset = l2meta.cluster_offset;
723 assert((cluster_offset & 511) == 0);
725 qemu_iovec_reset(&hd_qiov);
726 qemu_iovec_copy(&hd_qiov, qiov, bytes_done,
727 cur_nr_sectors * 512);
729 if (s->crypt_method) {
731 cluster_data = qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS *
735 assert(hd_qiov.size <=
736 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
737 qemu_iovec_to_buffer(&hd_qiov, cluster_data);
739 qcow2_encrypt_sectors(s, sector_num, cluster_data,
740 cluster_data, cur_nr_sectors, 1, &s->aes_encrypt_key);
742 qemu_iovec_reset(&hd_qiov);
743 qemu_iovec_add(&hd_qiov, cluster_data,
744 cur_nr_sectors * 512);
747 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
748 qemu_co_mutex_unlock(&s->lock);
749 trace_qcow2_writev_data(qemu_coroutine_self(),
750 (cluster_offset >> 9) + index_in_cluster);
751 ret = bdrv_co_writev(bs->file,
752 (cluster_offset >> 9) + index_in_cluster,
753 cur_nr_sectors, &hd_qiov);
754 qemu_co_mutex_lock(&s->lock);
759 ret = qcow2_alloc_cluster_link_l2(bs, &l2meta);
764 run_dependent_requests(s, &l2meta);
766 remaining_sectors -= cur_nr_sectors;
767 sector_num += cur_nr_sectors;
768 bytes_done += cur_nr_sectors * 512;
769 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_nr_sectors);
774 run_dependent_requests(s, &l2meta);
776 qemu_co_mutex_unlock(&s->lock);
778 qemu_iovec_destroy(&hd_qiov);
779 qemu_vfree(cluster_data);
780 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
785 static void qcow2_close(BlockDriverState *bs)
787 BDRVQcowState *s = bs->opaque;
790 qcow2_cache_flush(bs, s->l2_table_cache);
791 qcow2_cache_flush(bs, s->refcount_block_cache);
793 qcow2_cache_destroy(bs, s->l2_table_cache);
794 qcow2_cache_destroy(bs, s->refcount_block_cache);
796 g_free(s->unknown_header_fields);
797 cleanup_unknown_header_ext(bs);
799 g_free(s->cluster_cache);
800 qemu_vfree(s->cluster_data);
801 qcow2_refcount_close(bs);
802 qcow2_free_snapshots(bs);
805 static void qcow2_invalidate_cache(BlockDriverState *bs)
807 BDRVQcowState *s = bs->opaque;
808 int flags = s->flags;
809 AES_KEY aes_encrypt_key;
810 AES_KEY aes_decrypt_key;
811 uint32_t crypt_method = 0;
814 * Backing files are read-only which makes all of their metadata immutable,
815 * that means we don't have to worry about reopening them here.
818 if (s->crypt_method) {
819 crypt_method = s->crypt_method;
820 memcpy(&aes_encrypt_key, &s->aes_encrypt_key, sizeof(aes_encrypt_key));
821 memcpy(&aes_decrypt_key, &s->aes_decrypt_key, sizeof(aes_decrypt_key));
826 memset(s, 0, sizeof(BDRVQcowState));
827 qcow2_open(bs, flags);
830 s->crypt_method = crypt_method;
831 memcpy(&s->aes_encrypt_key, &aes_encrypt_key, sizeof(aes_encrypt_key));
832 memcpy(&s->aes_decrypt_key, &aes_decrypt_key, sizeof(aes_decrypt_key));
836 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
837 size_t len, size_t buflen)
839 QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
840 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
842 if (buflen < ext_len) {
846 *ext_backing_fmt = (QCowExtension) {
847 .magic = cpu_to_be32(magic),
848 .len = cpu_to_be32(len),
850 memcpy(buf + sizeof(QCowExtension), s, len);
856 * Updates the qcow2 header, including the variable length parts of it, i.e.
857 * the backing file name and all extensions. qcow2 was not designed to allow
858 * such changes, so if we run out of space (we can only use the first cluster)
859 * this function may fail.
861 * Returns 0 on success, -errno in error cases.
863 int qcow2_update_header(BlockDriverState *bs)
865 BDRVQcowState *s = bs->opaque;
868 size_t buflen = s->cluster_size;
871 uint32_t refcount_table_clusters;
872 size_t header_length;
873 Qcow2UnknownHeaderExtension *uext;
875 buf = qemu_blockalign(bs, buflen);
877 /* Header structure */
878 header = (QCowHeader*) buf;
880 if (buflen < sizeof(*header)) {
885 header_length = sizeof(*header) + s->unknown_header_fields_size;
886 total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
887 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
889 *header = (QCowHeader) {
890 /* Version 2 fields */
891 .magic = cpu_to_be32(QCOW_MAGIC),
892 .version = cpu_to_be32(s->qcow_version),
893 .backing_file_offset = 0,
894 .backing_file_size = 0,
895 .cluster_bits = cpu_to_be32(s->cluster_bits),
896 .size = cpu_to_be64(total_size),
897 .crypt_method = cpu_to_be32(s->crypt_method_header),
898 .l1_size = cpu_to_be32(s->l1_size),
899 .l1_table_offset = cpu_to_be64(s->l1_table_offset),
900 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset),
901 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
902 .nb_snapshots = cpu_to_be32(s->nb_snapshots),
903 .snapshots_offset = cpu_to_be64(s->snapshots_offset),
905 /* Version 3 fields */
906 .incompatible_features = cpu_to_be64(s->incompatible_features),
907 .compatible_features = cpu_to_be64(s->compatible_features),
908 .autoclear_features = cpu_to_be64(s->autoclear_features),
909 .refcount_order = cpu_to_be32(3 + REFCOUNT_SHIFT),
910 .header_length = cpu_to_be32(header_length),
913 /* For older versions, write a shorter header */
914 switch (s->qcow_version) {
916 ret = offsetof(QCowHeader, incompatible_features);
919 ret = sizeof(*header);
928 memset(buf, 0, buflen);
930 /* Preserve any unknown field in the header */
931 if (s->unknown_header_fields_size) {
932 if (buflen < s->unknown_header_fields_size) {
937 memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
938 buf += s->unknown_header_fields_size;
939 buflen -= s->unknown_header_fields_size;
942 /* Backing file format header extension */
943 if (*bs->backing_format) {
944 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
945 bs->backing_format, strlen(bs->backing_format),
956 Qcow2Feature features[] = {
957 /* no feature defined yet */
960 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
961 features, sizeof(features), buflen);
968 /* Keep unknown header extensions */
969 QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
970 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
979 /* End of header extensions */
980 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
988 /* Backing file name */
989 if (*bs->backing_file) {
990 size_t backing_file_len = strlen(bs->backing_file);
992 if (buflen < backing_file_len) {
997 strncpy(buf, bs->backing_file, buflen);
999 header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
1000 header->backing_file_size = cpu_to_be32(backing_file_len);
1003 /* Write the new header */
1004 ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
1015 static int qcow2_change_backing_file(BlockDriverState *bs,
1016 const char *backing_file, const char *backing_fmt)
1018 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
1019 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
1021 return qcow2_update_header(bs);
1024 static int preallocate(BlockDriverState *bs)
1026 uint64_t nb_sectors;
1032 nb_sectors = bdrv_getlength(bs) >> 9;
1034 qemu_co_queue_init(&meta.dependent_requests);
1035 meta.cluster_offset = 0;
1037 while (nb_sectors) {
1038 num = MIN(nb_sectors, INT_MAX >> 9);
1039 ret = qcow2_alloc_cluster_offset(bs, offset, 0, num, &num, &meta);
1044 ret = qcow2_alloc_cluster_link_l2(bs, &meta);
1046 qcow2_free_any_clusters(bs, meta.cluster_offset, meta.nb_clusters);
1050 /* There are no dependent requests, but we need to remove our request
1051 * from the list of in-flight requests */
1052 run_dependent_requests(bs->opaque, &meta);
1054 /* TODO Preallocate data if requested */
1061 * It is expected that the image file is large enough to actually contain
1062 * all of the allocated clusters (otherwise we get failing reads after
1063 * EOF). Extend the image to the last allocated sector.
1065 if (meta.cluster_offset != 0) {
1067 memset(buf, 0, 512);
1068 ret = bdrv_write(bs->file, (meta.cluster_offset >> 9) + num - 1, buf, 1);
1077 static int qcow2_create2(const char *filename, int64_t total_size,
1078 const char *backing_file, const char *backing_format,
1079 int flags, size_t cluster_size, int prealloc,
1080 QEMUOptionParameter *options, int version)
1082 /* Calculate cluster_bits */
1084 cluster_bits = ffs(cluster_size) - 1;
1085 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
1086 (1 << cluster_bits) != cluster_size)
1089 "Cluster size must be a power of two between %d and %dk",
1090 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
1095 * Open the image file and write a minimal qcow2 header.
1097 * We keep things simple and start with a zero-sized image. We also
1098 * do without refcount blocks or a L1 table for now. We'll fix the
1099 * inconsistency later.
1101 * We do need a refcount table because growing the refcount table means
1102 * allocating two new refcount blocks - the seconds of which would be at
1103 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
1104 * size for any qcow2 image.
1106 BlockDriverState* bs;
1108 uint8_t* refcount_table;
1111 ret = bdrv_create_file(filename, options);
1116 ret = bdrv_file_open(&bs, filename, BDRV_O_RDWR);
1121 /* Write the header */
1122 memset(&header, 0, sizeof(header));
1123 header.magic = cpu_to_be32(QCOW_MAGIC);
1124 header.version = cpu_to_be32(version);
1125 header.cluster_bits = cpu_to_be32(cluster_bits);
1126 header.size = cpu_to_be64(0);
1127 header.l1_table_offset = cpu_to_be64(0);
1128 header.l1_size = cpu_to_be32(0);
1129 header.refcount_table_offset = cpu_to_be64(cluster_size);
1130 header.refcount_table_clusters = cpu_to_be32(1);
1131 header.refcount_order = cpu_to_be32(3 + REFCOUNT_SHIFT);
1132 header.header_length = cpu_to_be32(sizeof(header));
1134 if (flags & BLOCK_FLAG_ENCRYPT) {
1135 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
1137 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
1140 ret = bdrv_pwrite(bs, 0, &header, sizeof(header));
1145 /* Write an empty refcount table */
1146 refcount_table = g_malloc0(cluster_size);
1147 ret = bdrv_pwrite(bs, cluster_size, refcount_table, cluster_size);
1148 g_free(refcount_table);
1157 * And now open the image and make it consistent first (i.e. increase the
1158 * refcount of the cluster that is occupied by the header and the refcount
1161 BlockDriver* drv = bdrv_find_format("qcow2");
1162 assert(drv != NULL);
1163 ret = bdrv_open(bs, filename,
1164 BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_FLUSH, drv);
1169 ret = qcow2_alloc_clusters(bs, 2 * cluster_size);
1173 } else if (ret != 0) {
1174 error_report("Huh, first cluster in empty image is already in use?");
1178 /* Okay, now that we have a valid image, let's give it the right size */
1179 ret = bdrv_truncate(bs, total_size * BDRV_SECTOR_SIZE);
1184 /* Want a backing file? There you go.*/
1186 ret = bdrv_change_backing_file(bs, backing_file, backing_format);
1192 /* And if we're supposed to preallocate metadata, do that now */
1194 BDRVQcowState *s = bs->opaque;
1195 qemu_co_mutex_lock(&s->lock);
1196 ret = preallocate(bs);
1197 qemu_co_mutex_unlock(&s->lock);
1209 static int qcow2_create(const char *filename, QEMUOptionParameter *options)
1211 const char *backing_file = NULL;
1212 const char *backing_fmt = NULL;
1213 uint64_t sectors = 0;
1215 size_t cluster_size = DEFAULT_CLUSTER_SIZE;
1219 /* Read out options */
1220 while (options && options->name) {
1221 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1222 sectors = options->value.n / 512;
1223 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
1224 backing_file = options->value.s;
1225 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) {
1226 backing_fmt = options->value.s;
1227 } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
1228 flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
1229 } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
1230 if (options->value.n) {
1231 cluster_size = options->value.n;
1233 } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) {
1234 if (!options->value.s || !strcmp(options->value.s, "off")) {
1236 } else if (!strcmp(options->value.s, "metadata")) {
1239 fprintf(stderr, "Invalid preallocation mode: '%s'\n",
1243 } else if (!strcmp(options->name, BLOCK_OPT_COMPAT_LEVEL)) {
1244 if (!options->value.s || !strcmp(options->value.s, "0.10")) {
1246 } else if (!strcmp(options->value.s, "1.1")) {
1249 fprintf(stderr, "Invalid compatibility level: '%s'\n",
1257 if (backing_file && prealloc) {
1258 fprintf(stderr, "Backing file and preallocation cannot be used at "
1263 return qcow2_create2(filename, sectors, backing_file, backing_fmt, flags,
1264 cluster_size, prealloc, options, version);
1267 static int qcow2_make_empty(BlockDriverState *bs)
1270 /* XXX: not correct */
1271 BDRVQcowState *s = bs->opaque;
1272 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1275 memset(s->l1_table, 0, l1_length);
1276 if (bdrv_pwrite(bs->file, s->l1_table_offset, s->l1_table, l1_length) < 0)
1278 ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
1287 static coroutine_fn int qcow2_co_write_zeroes(BlockDriverState *bs,
1288 int64_t sector_num, int nb_sectors)
1291 BDRVQcowState *s = bs->opaque;
1293 /* Emulate misaligned zero writes */
1294 if (sector_num % s->cluster_sectors || nb_sectors % s->cluster_sectors) {
1298 /* Whatever is left can use real zero clusters */
1299 qemu_co_mutex_lock(&s->lock);
1300 ret = qcow2_zero_clusters(bs, sector_num << BDRV_SECTOR_BITS,
1302 qemu_co_mutex_unlock(&s->lock);
1307 static coroutine_fn int qcow2_co_discard(BlockDriverState *bs,
1308 int64_t sector_num, int nb_sectors)
1311 BDRVQcowState *s = bs->opaque;
1313 qemu_co_mutex_lock(&s->lock);
1314 ret = qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS,
1316 qemu_co_mutex_unlock(&s->lock);
1320 static int qcow2_truncate(BlockDriverState *bs, int64_t offset)
1322 BDRVQcowState *s = bs->opaque;
1323 int ret, new_l1_size;
1326 error_report("The new size must be a multiple of 512");
1330 /* cannot proceed if image has snapshots */
1331 if (s->nb_snapshots) {
1332 error_report("Can't resize an image which has snapshots");
1336 /* shrinking is currently not supported */
1337 if (offset < bs->total_sectors * 512) {
1338 error_report("qcow2 doesn't support shrinking images yet");
1342 new_l1_size = size_to_l1(s, offset);
1343 ret = qcow2_grow_l1_table(bs, new_l1_size, true);
1348 /* write updated header.size */
1349 offset = cpu_to_be64(offset);
1350 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
1351 &offset, sizeof(uint64_t));
1356 s->l1_vm_state_index = new_l1_size;
1360 /* XXX: put compressed sectors first, then all the cluster aligned
1361 tables to avoid losing bytes in alignment */
1362 static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num,
1363 const uint8_t *buf, int nb_sectors)
1365 BDRVQcowState *s = bs->opaque;
1369 uint64_t cluster_offset;
1371 if (nb_sectors == 0) {
1372 /* align end of file to a sector boundary to ease reading with
1373 sector based I/Os */
1374 cluster_offset = bdrv_getlength(bs->file);
1375 cluster_offset = (cluster_offset + 511) & ~511;
1376 bdrv_truncate(bs->file, cluster_offset);
1380 if (nb_sectors != s->cluster_sectors)
1383 out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1385 /* best compression, small window, no zlib header */
1386 memset(&strm, 0, sizeof(strm));
1387 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1389 9, Z_DEFAULT_STRATEGY);
1395 strm.avail_in = s->cluster_size;
1396 strm.next_in = (uint8_t *)buf;
1397 strm.avail_out = s->cluster_size;
1398 strm.next_out = out_buf;
1400 ret = deflate(&strm, Z_FINISH);
1401 if (ret != Z_STREAM_END && ret != Z_OK) {
1406 out_len = strm.next_out - out_buf;
1410 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1411 /* could not compress: write normal cluster */
1412 ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
1417 cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
1418 sector_num << 9, out_len);
1419 if (!cluster_offset) {
1423 cluster_offset &= s->cluster_offset_mask;
1424 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
1425 ret = bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len);
1437 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
1439 BDRVQcowState *s = bs->opaque;
1442 qemu_co_mutex_lock(&s->lock);
1443 ret = qcow2_cache_flush(bs, s->l2_table_cache);
1445 qemu_co_mutex_unlock(&s->lock);
1449 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
1451 qemu_co_mutex_unlock(&s->lock);
1454 qemu_co_mutex_unlock(&s->lock);
1459 static int64_t qcow2_vm_state_offset(BDRVQcowState *s)
1461 return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits);
1464 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1466 BDRVQcowState *s = bs->opaque;
1467 bdi->cluster_size = s->cluster_size;
1468 bdi->vm_state_offset = qcow2_vm_state_offset(s);
1473 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result)
1475 return qcow2_check_refcounts(bs, result);
1479 static void dump_refcounts(BlockDriverState *bs)
1481 BDRVQcowState *s = bs->opaque;
1482 int64_t nb_clusters, k, k1, size;
1485 size = bdrv_getlength(bs->file);
1486 nb_clusters = size_to_clusters(s, size);
1487 for(k = 0; k < nb_clusters;) {
1489 refcount = get_refcount(bs, k);
1491 while (k < nb_clusters && get_refcount(bs, k) == refcount)
1493 printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
1499 static int qcow2_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1500 int64_t pos, int size)
1502 BDRVQcowState *s = bs->opaque;
1503 int growable = bs->growable;
1506 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
1508 ret = bdrv_pwrite(bs, qcow2_vm_state_offset(s) + pos, buf, size);
1509 bs->growable = growable;
1514 static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1515 int64_t pos, int size)
1517 BDRVQcowState *s = bs->opaque;
1518 int growable = bs->growable;
1521 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
1523 ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size);
1524 bs->growable = growable;
1529 static QEMUOptionParameter qcow2_create_options[] = {
1531 .name = BLOCK_OPT_SIZE,
1533 .help = "Virtual disk size"
1536 .name = BLOCK_OPT_COMPAT_LEVEL,
1538 .help = "Compatibility level (0.10 or 1.1)"
1541 .name = BLOCK_OPT_BACKING_FILE,
1543 .help = "File name of a base image"
1546 .name = BLOCK_OPT_BACKING_FMT,
1548 .help = "Image format of the base image"
1551 .name = BLOCK_OPT_ENCRYPT,
1553 .help = "Encrypt the image"
1556 .name = BLOCK_OPT_CLUSTER_SIZE,
1558 .help = "qcow2 cluster size",
1559 .value = { .n = DEFAULT_CLUSTER_SIZE },
1562 .name = BLOCK_OPT_PREALLOC,
1564 .help = "Preallocation mode (allowed values: off, metadata)"
1569 static BlockDriver bdrv_qcow2 = {
1570 .format_name = "qcow2",
1571 .instance_size = sizeof(BDRVQcowState),
1572 .bdrv_probe = qcow2_probe,
1573 .bdrv_open = qcow2_open,
1574 .bdrv_close = qcow2_close,
1575 .bdrv_create = qcow2_create,
1576 .bdrv_co_is_allocated = qcow2_co_is_allocated,
1577 .bdrv_set_key = qcow2_set_key,
1578 .bdrv_make_empty = qcow2_make_empty,
1580 .bdrv_co_readv = qcow2_co_readv,
1581 .bdrv_co_writev = qcow2_co_writev,
1582 .bdrv_co_flush_to_os = qcow2_co_flush_to_os,
1584 .bdrv_co_write_zeroes = qcow2_co_write_zeroes,
1585 .bdrv_co_discard = qcow2_co_discard,
1586 .bdrv_truncate = qcow2_truncate,
1587 .bdrv_write_compressed = qcow2_write_compressed,
1589 .bdrv_snapshot_create = qcow2_snapshot_create,
1590 .bdrv_snapshot_goto = qcow2_snapshot_goto,
1591 .bdrv_snapshot_delete = qcow2_snapshot_delete,
1592 .bdrv_snapshot_list = qcow2_snapshot_list,
1593 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
1594 .bdrv_get_info = qcow2_get_info,
1596 .bdrv_save_vmstate = qcow2_save_vmstate,
1597 .bdrv_load_vmstate = qcow2_load_vmstate,
1599 .bdrv_change_backing_file = qcow2_change_backing_file,
1601 .bdrv_invalidate_cache = qcow2_invalidate_cache,
1603 .create_options = qcow2_create_options,
1604 .bdrv_check = qcow2_check,
1607 static void bdrv_qcow2_init(void)
1609 bdrv_register(&bdrv_qcow2);
1612 block_init(bdrv_qcow2_init);