2 * Block driver for the QCOW format
4 * Copyright (c) 2004-2006 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "qemu/osdep.h"
26 #include "qapi/error.h"
27 #include "qemu/error-report.h"
28 #include "block/block_int.h"
29 #include "block/qdict.h"
30 #include "sysemu/block-backend.h"
31 #include "qemu/module.h"
32 #include "qemu/option.h"
33 #include "qemu/bswap.h"
34 #include "qemu/cutils.h"
36 #include "qapi/qmp/qdict.h"
37 #include "qapi/qmp/qstring.h"
38 #include "qapi/qobject-input-visitor.h"
39 #include "qapi/qapi-visit-block-core.h"
40 #include "crypto/block.h"
41 #include "migration/blocker.h"
44 /**************************************************************/
45 /* QEMU COW block driver with compression and encryption support */
47 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
48 #define QCOW_VERSION 1
50 #define QCOW_CRYPT_NONE 0
51 #define QCOW_CRYPT_AES 1
53 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
55 typedef struct QCowHeader {
58 uint64_t backing_file_offset;
59 uint32_t backing_file_size;
61 uint64_t size; /* in bytes */
65 uint32_t crypt_method;
66 uint64_t l1_table_offset;
67 } QEMU_PACKED QCowHeader;
69 #define L2_CACHE_SIZE 16
71 typedef struct BDRVQcowState {
77 uint64_t cluster_offset_mask;
78 uint64_t l1_table_offset;
81 uint64_t l2_cache_offsets[L2_CACHE_SIZE];
82 uint32_t l2_cache_counts[L2_CACHE_SIZE];
83 uint8_t *cluster_cache;
84 uint8_t *cluster_data;
85 uint64_t cluster_cache_offset;
86 QCryptoBlock *crypto; /* Disk encryption format driver */
87 uint32_t crypt_method_header;
89 Error *migration_blocker;
92 static QemuOptsList qcow_create_opts;
94 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
96 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
98 const QCowHeader *cow_header = (const void *)buf;
100 if (buf_size >= sizeof(QCowHeader) &&
101 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
102 be32_to_cpu(cow_header->version) == QCOW_VERSION)
108 static QemuOptsList qcow_runtime_opts = {
110 .head = QTAILQ_HEAD_INITIALIZER(qcow_runtime_opts.head),
112 BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
113 { /* end of list */ }
117 static int qcow_open(BlockDriverState *bs, QDict *options, int flags,
120 BDRVQcowState *s = bs->opaque;
121 unsigned int len, i, shift;
124 Error *local_err = NULL;
125 QCryptoBlockOpenOptions *crypto_opts = NULL;
126 unsigned int cflags = 0;
127 QDict *encryptopts = NULL;
128 const char *encryptfmt;
130 qdict_extract_subqdict(options, &encryptopts, "encrypt.");
131 encryptfmt = qdict_get_try_str(encryptopts, "format");
133 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
140 ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
144 header.magic = be32_to_cpu(header.magic);
145 header.version = be32_to_cpu(header.version);
146 header.backing_file_offset = be64_to_cpu(header.backing_file_offset);
147 header.backing_file_size = be32_to_cpu(header.backing_file_size);
148 header.mtime = be32_to_cpu(header.mtime);
149 header.size = be64_to_cpu(header.size);
150 header.crypt_method = be32_to_cpu(header.crypt_method);
151 header.l1_table_offset = be64_to_cpu(header.l1_table_offset);
153 if (header.magic != QCOW_MAGIC) {
154 error_setg(errp, "Image not in qcow format");
158 if (header.version != QCOW_VERSION) {
159 error_setg(errp, "qcow (v%d) does not support qcow version %" PRIu32,
160 QCOW_VERSION, header.version);
161 if (header.version == 2 || header.version == 3) {
162 error_append_hint(errp, "Try the 'qcow2' driver instead.\n");
169 if (header.size <= 1) {
170 error_setg(errp, "Image size is too small (must be at least 2 bytes)");
174 if (header.cluster_bits < 9 || header.cluster_bits > 16) {
175 error_setg(errp, "Cluster size must be between 512 and 64k");
180 /* l2_bits specifies number of entries; storing a uint64_t in each entry,
181 * so bytes = num_entries << 3. */
182 if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) {
183 error_setg(errp, "L2 table size must be between 512 and 64k");
188 s->crypt_method_header = header.crypt_method;
189 if (s->crypt_method_header) {
190 if (bdrv_uses_whitelist() &&
191 s->crypt_method_header == QCOW_CRYPT_AES) {
193 "Use of AES-CBC encrypted qcow images is no longer "
194 "supported in system emulators");
195 error_append_hint(errp,
196 "You can use 'qemu-img convert' to convert your "
197 "image to an alternative supported format, such "
198 "as unencrypted qcow, or raw with the LUKS "
199 "format instead.\n");
203 if (s->crypt_method_header == QCOW_CRYPT_AES) {
204 if (encryptfmt && !g_str_equal(encryptfmt, "aes")) {
206 "Header reported 'aes' encryption format but "
207 "options specify '%s'", encryptfmt);
211 qdict_put_str(encryptopts, "format", "qcow");
212 crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
218 if (flags & BDRV_O_NO_IO) {
219 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
221 s->crypto = qcrypto_block_open(crypto_opts, "encrypt.",
222 NULL, NULL, cflags, 1, errp);
228 error_setg(errp, "invalid encryption method in qcow header");
232 bs->encrypted = true;
235 error_setg(errp, "No encryption in image header, but options "
236 "specified format '%s'", encryptfmt);
241 s->cluster_bits = header.cluster_bits;
242 s->cluster_size = 1 << s->cluster_bits;
243 s->l2_bits = header.l2_bits;
244 s->l2_size = 1 << s->l2_bits;
245 bs->total_sectors = header.size / 512;
246 s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
248 /* read the level 1 table */
249 shift = s->cluster_bits + s->l2_bits;
250 if (header.size > UINT64_MAX - (1LL << shift)) {
251 error_setg(errp, "Image too large");
255 uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift;
256 if (l1_size > INT_MAX / sizeof(uint64_t)) {
257 error_setg(errp, "Image too large");
261 s->l1_size = l1_size;
264 s->l1_table_offset = header.l1_table_offset;
265 s->l1_table = g_try_new(uint64_t, s->l1_size);
266 if (s->l1_table == NULL) {
267 error_setg(errp, "Could not allocate memory for L1 table");
272 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
273 s->l1_size * sizeof(uint64_t));
278 for(i = 0;i < s->l1_size; i++) {
279 s->l1_table[i] = be64_to_cpu(s->l1_table[i]);
282 /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
284 qemu_try_blockalign(bs->file->bs,
285 s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
286 if (s->l2_cache == NULL) {
287 error_setg(errp, "Could not allocate L2 table cache");
291 s->cluster_cache = g_malloc(s->cluster_size);
292 s->cluster_data = g_malloc(s->cluster_size);
293 s->cluster_cache_offset = -1;
295 /* read the backing file name */
296 if (header.backing_file_offset != 0) {
297 len = header.backing_file_size;
298 if (len > 1023 || len >= sizeof(bs->backing_file)) {
299 error_setg(errp, "Backing file name too long");
303 ret = bdrv_pread(bs->file, header.backing_file_offset,
304 bs->auto_backing_file, len);
308 bs->auto_backing_file[len] = '\0';
309 pstrcpy(bs->backing_file, sizeof(bs->backing_file),
310 bs->auto_backing_file);
313 /* Disable migration when qcow images are used */
314 error_setg(&s->migration_blocker, "The qcow format used by node '%s' "
315 "does not support live migration",
316 bdrv_get_device_or_node_name(bs));
317 ret = migrate_add_blocker(s->migration_blocker, &local_err);
319 error_propagate(errp, local_err);
320 error_free(s->migration_blocker);
324 qobject_unref(encryptopts);
325 qapi_free_QCryptoBlockOpenOptions(crypto_opts);
326 qemu_co_mutex_init(&s->lock);
331 qemu_vfree(s->l2_cache);
332 g_free(s->cluster_cache);
333 g_free(s->cluster_data);
334 qcrypto_block_free(s->crypto);
335 qobject_unref(encryptopts);
336 qapi_free_QCryptoBlockOpenOptions(crypto_opts);
341 /* We have nothing to do for QCOW reopen, stubs just return
343 static int qcow_reopen_prepare(BDRVReopenState *state,
344 BlockReopenQueue *queue, Error **errp)
354 * 1 to allocate a normal cluster (for sector-aligned byte offsets 'n_start'
355 * to 'n_end' within the cluster)
357 * 2 to allocate a compressed cluster of size
358 * 'compressed_size'. 'compressed_size' must be > 0 and <
361 * return 0 if not allocated, 1 if *result is assigned, and negative
364 static int get_cluster_offset(BlockDriverState *bs,
365 uint64_t offset, int allocate,
367 int n_start, int n_end, uint64_t *result)
369 BDRVQcowState *s = bs->opaque;
370 int min_index, i, j, l1_index, l2_index, ret;
372 uint64_t *l2_table, cluster_offset, tmp;
377 l1_index = offset >> (s->l2_bits + s->cluster_bits);
378 l2_offset = s->l1_table[l1_index];
383 /* allocate a new l2 entry */
384 l2_offset = bdrv_getlength(bs->file->bs);
388 /* round to cluster size */
389 l2_offset = QEMU_ALIGN_UP(l2_offset, s->cluster_size);
390 /* update the L1 entry */
391 s->l1_table[l1_index] = l2_offset;
392 tmp = cpu_to_be64(l2_offset);
393 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
394 ret = bdrv_pwrite_sync(bs->file,
395 s->l1_table_offset + l1_index * sizeof(tmp),
402 for(i = 0; i < L2_CACHE_SIZE; i++) {
403 if (l2_offset == s->l2_cache_offsets[i]) {
404 /* increment the hit count */
405 if (++s->l2_cache_counts[i] == 0xffffffff) {
406 for(j = 0; j < L2_CACHE_SIZE; j++) {
407 s->l2_cache_counts[j] >>= 1;
410 l2_table = s->l2_cache + (i << s->l2_bits);
414 /* not found: load a new entry in the least used one */
416 min_count = 0xffffffff;
417 for(i = 0; i < L2_CACHE_SIZE; i++) {
418 if (s->l2_cache_counts[i] < min_count) {
419 min_count = s->l2_cache_counts[i];
423 l2_table = s->l2_cache + (min_index << s->l2_bits);
424 BLKDBG_EVENT(bs->file, BLKDBG_L2_LOAD);
426 memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
427 ret = bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
428 s->l2_size * sizeof(uint64_t));
433 ret = bdrv_pread(bs->file, l2_offset, l2_table,
434 s->l2_size * sizeof(uint64_t));
439 s->l2_cache_offsets[min_index] = l2_offset;
440 s->l2_cache_counts[min_index] = 1;
442 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
443 cluster_offset = be64_to_cpu(l2_table[l2_index]);
444 if (!cluster_offset ||
445 ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
448 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
449 assert(QEMU_IS_ALIGNED(n_start | n_end, BDRV_SECTOR_SIZE));
450 /* allocate a new cluster */
451 if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
452 (n_end - n_start) < s->cluster_size) {
453 /* if the cluster is already compressed, we must
454 decompress it in the case it is not completely
456 if (decompress_cluster(bs, cluster_offset) < 0) {
459 cluster_offset = bdrv_getlength(bs->file->bs);
460 if ((int64_t) cluster_offset < 0) {
461 return cluster_offset;
463 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
464 /* write the cluster content */
465 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
466 ret = bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache,
472 cluster_offset = bdrv_getlength(bs->file->bs);
473 if ((int64_t) cluster_offset < 0) {
474 return cluster_offset;
477 /* round to cluster size */
478 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
479 if (cluster_offset + s->cluster_size > INT64_MAX) {
482 ret = bdrv_truncate(bs->file, cluster_offset + s->cluster_size,
483 PREALLOC_MODE_OFF, NULL);
487 /* if encrypted, we must initialize the cluster
488 content which won't be written */
490 (n_end - n_start) < s->cluster_size) {
491 uint64_t start_offset;
493 start_offset = offset & ~(s->cluster_size - 1);
494 for (i = 0; i < s->cluster_size; i += BDRV_SECTOR_SIZE) {
495 if (i < n_start || i >= n_end) {
496 memset(s->cluster_data, 0x00, BDRV_SECTOR_SIZE);
497 if (qcrypto_block_encrypt(s->crypto,
504 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
505 ret = bdrv_pwrite(bs->file,
515 } else if (allocate == 2) {
516 cluster_offset |= QCOW_OFLAG_COMPRESSED |
517 (uint64_t)compressed_size << (63 - s->cluster_bits);
520 /* update L2 table */
521 tmp = cpu_to_be64(cluster_offset);
522 l2_table[l2_index] = tmp;
524 BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
526 BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE);
528 ret = bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
534 *result = cluster_offset;
538 static int coroutine_fn qcow_co_block_status(BlockDriverState *bs,
540 int64_t offset, int64_t bytes,
541 int64_t *pnum, int64_t *map,
542 BlockDriverState **file)
544 BDRVQcowState *s = bs->opaque;
545 int index_in_cluster, ret;
547 uint64_t cluster_offset;
549 qemu_co_mutex_lock(&s->lock);
550 ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset);
551 qemu_co_mutex_unlock(&s->lock);
555 index_in_cluster = offset & (s->cluster_size - 1);
556 n = s->cluster_size - index_in_cluster;
561 if (!cluster_offset) {
564 if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->crypto) {
565 return BDRV_BLOCK_DATA;
567 *map = cluster_offset | index_in_cluster;
568 *file = bs->file->bs;
569 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
572 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
573 const uint8_t *buf, int buf_size)
575 z_stream strm1, *strm = &strm1;
578 memset(strm, 0, sizeof(*strm));
580 strm->next_in = (uint8_t *)buf;
581 strm->avail_in = buf_size;
582 strm->next_out = out_buf;
583 strm->avail_out = out_buf_size;
585 ret = inflateInit2(strm, -12);
588 ret = inflate(strm, Z_FINISH);
589 out_len = strm->next_out - out_buf;
590 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
591 out_len != out_buf_size) {
599 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
601 BDRVQcowState *s = bs->opaque;
605 coffset = cluster_offset & s->cluster_offset_mask;
606 if (s->cluster_cache_offset != coffset) {
607 csize = cluster_offset >> (63 - s->cluster_bits);
608 csize &= (s->cluster_size - 1);
609 BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
610 ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
613 if (decompress_buffer(s->cluster_cache, s->cluster_size,
614 s->cluster_data, csize) < 0) {
617 s->cluster_cache_offset = coffset;
622 static void qcow_refresh_limits(BlockDriverState *bs, Error **errp)
624 /* At least encrypted images require 512-byte alignment. Apply the
625 * limit universally, rather than just on encrypted images, as
626 * it's easier to let the block layer handle rounding than to
627 * audit this code further. */
628 bs->bl.request_alignment = BDRV_SECTOR_SIZE;
631 static coroutine_fn int qcow_co_preadv(BlockDriverState *bs, uint64_t offset,
632 uint64_t bytes, QEMUIOVector *qiov,
635 BDRVQcowState *s = bs->opaque;
636 int offset_in_cluster;
638 uint64_t cluster_offset;
643 if (qiov->niov > 1) {
644 buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
650 buf = (uint8_t *)qiov->iov->iov_base;
653 qemu_co_mutex_lock(&s->lock);
656 /* prepare next request */
657 ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset);
661 offset_in_cluster = offset & (s->cluster_size - 1);
662 n = s->cluster_size - offset_in_cluster;
667 if (!cluster_offset) {
669 /* read from the base image */
670 qemu_co_mutex_unlock(&s->lock);
671 /* qcow2 emits this on bs->file instead of bs->backing */
672 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
673 ret = bdrv_co_pread(bs->backing, offset, n, buf, 0);
674 qemu_co_mutex_lock(&s->lock);
679 /* Note: in this case, no need to wait */
682 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
683 /* add AIO support for compressed blocks ? */
684 if (decompress_cluster(bs, cluster_offset) < 0) {
688 memcpy(buf, s->cluster_cache + offset_in_cluster, n);
690 if ((cluster_offset & 511) != 0) {
694 qemu_co_mutex_unlock(&s->lock);
695 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
696 ret = bdrv_co_pread(bs->file, cluster_offset + offset_in_cluster,
698 qemu_co_mutex_lock(&s->lock);
704 if (qcrypto_block_decrypt(s->crypto,
705 offset, buf, n, NULL) < 0) {
718 qemu_co_mutex_unlock(&s->lock);
720 if (qiov->niov > 1) {
721 qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
722 qemu_vfree(orig_buf);
728 static coroutine_fn int qcow_co_pwritev(BlockDriverState *bs, uint64_t offset,
729 uint64_t bytes, QEMUIOVector *qiov,
732 BDRVQcowState *s = bs->opaque;
733 int offset_in_cluster;
734 uint64_t cluster_offset;
740 s->cluster_cache_offset = -1; /* disable compressed cache */
742 /* We must always copy the iov when encrypting, so we
743 * don't modify the original data buffer during encryption */
744 if (bs->encrypted || qiov->niov > 1) {
745 buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
749 qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
752 buf = (uint8_t *)qiov->iov->iov_base;
755 qemu_co_mutex_lock(&s->lock);
758 offset_in_cluster = offset & (s->cluster_size - 1);
759 n = s->cluster_size - offset_in_cluster;
763 ret = get_cluster_offset(bs, offset, 1, 0, offset_in_cluster,
764 offset_in_cluster + n, &cluster_offset);
768 if (!cluster_offset || (cluster_offset & 511) != 0) {
774 if (qcrypto_block_encrypt(s->crypto, offset, buf, n, NULL) < 0) {
780 qemu_co_mutex_unlock(&s->lock);
781 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
782 ret = bdrv_co_pwrite(bs->file, cluster_offset + offset_in_cluster,
784 qemu_co_mutex_lock(&s->lock);
794 qemu_co_mutex_unlock(&s->lock);
796 qemu_vfree(orig_buf);
801 static void qcow_close(BlockDriverState *bs)
803 BDRVQcowState *s = bs->opaque;
805 qcrypto_block_free(s->crypto);
808 qemu_vfree(s->l2_cache);
809 g_free(s->cluster_cache);
810 g_free(s->cluster_data);
812 migrate_del_blocker(s->migration_blocker);
813 error_free(s->migration_blocker);
816 static int coroutine_fn qcow_co_create(BlockdevCreateOptions *opts,
819 BlockdevCreateOptionsQcow *qcow_opts;
820 int header_size, backing_filename_len, l1_size, shift, i;
823 int64_t total_size = 0;
825 BlockDriverState *bs;
826 BlockBackend *qcow_blk;
827 QCryptoBlock *crypto = NULL;
829 assert(opts->driver == BLOCKDEV_DRIVER_QCOW);
830 qcow_opts = &opts->u.qcow;
833 total_size = qcow_opts->size;
834 if (total_size == 0) {
835 error_setg(errp, "Image size is too small, cannot be zero length");
839 if (qcow_opts->has_encrypt &&
840 qcow_opts->encrypt->format != Q_CRYPTO_BLOCK_FORMAT_QCOW)
842 error_setg(errp, "Unsupported encryption format");
846 /* Create BlockBackend to write to the image */
847 bs = bdrv_open_blockdev_ref(qcow_opts->file, errp);
852 qcow_blk = blk_new(bdrv_get_aio_context(bs),
853 BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL);
854 ret = blk_insert_bs(qcow_blk, bs, errp);
858 blk_set_allow_write_beyond_eof(qcow_blk, true);
860 /* Create image format */
861 ret = blk_truncate(qcow_blk, 0, PREALLOC_MODE_OFF, errp);
866 memset(&header, 0, sizeof(header));
867 header.magic = cpu_to_be32(QCOW_MAGIC);
868 header.version = cpu_to_be32(QCOW_VERSION);
869 header.size = cpu_to_be64(total_size);
870 header_size = sizeof(header);
871 backing_filename_len = 0;
872 if (qcow_opts->has_backing_file) {
873 if (strcmp(qcow_opts->backing_file, "fat:")) {
874 header.backing_file_offset = cpu_to_be64(header_size);
875 backing_filename_len = strlen(qcow_opts->backing_file);
876 header.backing_file_size = cpu_to_be32(backing_filename_len);
877 header_size += backing_filename_len;
879 /* special backing file for vvfat */
880 qcow_opts->has_backing_file = false;
882 header.cluster_bits = 9; /* 512 byte cluster to avoid copying
883 unmodified sectors */
884 header.l2_bits = 12; /* 32 KB L2 tables */
886 header.cluster_bits = 12; /* 4 KB clusters */
887 header.l2_bits = 9; /* 4 KB L2 tables */
889 header_size = (header_size + 7) & ~7;
890 shift = header.cluster_bits + header.l2_bits;
891 l1_size = (total_size + (1LL << shift) - 1) >> shift;
893 header.l1_table_offset = cpu_to_be64(header_size);
895 if (qcow_opts->has_encrypt) {
896 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
898 crypto = qcrypto_block_create(qcow_opts->encrypt, "encrypt.",
899 NULL, NULL, NULL, errp);
905 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
908 /* write all the data */
909 ret = blk_pwrite(qcow_blk, 0, &header, sizeof(header), 0);
910 if (ret != sizeof(header)) {
914 if (qcow_opts->has_backing_file) {
915 ret = blk_pwrite(qcow_blk, sizeof(header),
916 qcow_opts->backing_file, backing_filename_len, 0);
917 if (ret != backing_filename_len) {
922 tmp = g_malloc0(BDRV_SECTOR_SIZE);
923 for (i = 0; i < DIV_ROUND_UP(sizeof(uint64_t) * l1_size, BDRV_SECTOR_SIZE);
925 ret = blk_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i,
926 tmp, BDRV_SECTOR_SIZE, 0);
927 if (ret != BDRV_SECTOR_SIZE) {
938 qcrypto_block_free(crypto);
942 static int coroutine_fn qcow_co_create_opts(const char *filename,
943 QemuOpts *opts, Error **errp)
945 BlockdevCreateOptions *create_options = NULL;
946 BlockDriverState *bs = NULL;
950 Error *local_err = NULL;
953 static const QDictRenames opt_renames[] = {
954 { BLOCK_OPT_BACKING_FILE, "backing-file" },
955 { BLOCK_OPT_ENCRYPT, BLOCK_OPT_ENCRYPT_FORMAT },
959 /* Parse options and convert legacy syntax */
960 qdict = qemu_opts_to_qdict_filtered(opts, NULL, &qcow_create_opts, true);
962 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT);
963 if (val && !strcmp(val, "on")) {
964 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow");
965 } else if (val && !strcmp(val, "off")) {
966 qdict_del(qdict, BLOCK_OPT_ENCRYPT);
969 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT);
970 if (val && !strcmp(val, "aes")) {
971 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow");
974 if (!qdict_rename_keys(qdict, opt_renames, errp)) {
979 /* Create and open the file (protocol layer) */
980 ret = bdrv_create_file(filename, opts, &local_err);
982 error_propagate(errp, local_err);
986 bs = bdrv_open(filename, NULL, NULL,
987 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
993 /* Now get the QAPI type BlockdevCreateOptions */
994 qdict_put_str(qdict, "driver", "qcow");
995 qdict_put_str(qdict, "file", bs->node_name);
997 v = qobject_input_visitor_new_flat_confused(qdict, errp);
1003 visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err);
1007 error_propagate(errp, local_err);
1012 /* Silently round up size */
1013 assert(create_options->driver == BLOCKDEV_DRIVER_QCOW);
1014 create_options->u.qcow.size =
1015 ROUND_UP(create_options->u.qcow.size, BDRV_SECTOR_SIZE);
1017 /* Create the qcow image (format layer) */
1018 ret = qcow_co_create(create_options, errp);
1025 qobject_unref(qdict);
1027 qapi_free_BlockdevCreateOptions(create_options);
1031 static int qcow_make_empty(BlockDriverState *bs)
1033 BDRVQcowState *s = bs->opaque;
1034 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1037 memset(s->l1_table, 0, l1_length);
1038 if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
1041 ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length,
1042 PREALLOC_MODE_OFF, NULL);
1046 memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
1047 memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
1048 memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
1053 /* XXX: put compressed sectors first, then all the cluster aligned
1054 tables to avoid losing bytes in alignment */
1055 static coroutine_fn int
1056 qcow_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
1057 uint64_t bytes, QEMUIOVector *qiov)
1059 BDRVQcowState *s = bs->opaque;
1062 uint8_t *buf, *out_buf;
1063 uint64_t cluster_offset;
1065 buf = qemu_blockalign(bs, s->cluster_size);
1066 if (bytes != s->cluster_size) {
1067 if (bytes > s->cluster_size ||
1068 offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
1073 /* Zero-pad last write if image size is not cluster aligned */
1074 memset(buf + bytes, 0, s->cluster_size - bytes);
1076 qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
1078 out_buf = g_malloc(s->cluster_size);
1080 /* best compression, small window, no zlib header */
1081 memset(&strm, 0, sizeof(strm));
1082 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1084 9, Z_DEFAULT_STRATEGY);
1090 strm.avail_in = s->cluster_size;
1091 strm.next_in = (uint8_t *)buf;
1092 strm.avail_out = s->cluster_size;
1093 strm.next_out = out_buf;
1095 ret = deflate(&strm, Z_FINISH);
1096 if (ret != Z_STREAM_END && ret != Z_OK) {
1101 out_len = strm.next_out - out_buf;
1105 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1106 /* could not compress: write normal cluster */
1107 ret = qcow_co_pwritev(bs, offset, bytes, qiov, 0);
1113 qemu_co_mutex_lock(&s->lock);
1114 ret = get_cluster_offset(bs, offset, 2, out_len, 0, 0, &cluster_offset);
1115 qemu_co_mutex_unlock(&s->lock);
1119 if (cluster_offset == 0) {
1123 cluster_offset &= s->cluster_offset_mask;
1125 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
1126 ret = bdrv_co_pwrite(bs->file, cluster_offset, out_len, out_buf, 0);
1138 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1140 BDRVQcowState *s = bs->opaque;
1141 bdi->cluster_size = s->cluster_size;
1145 static QemuOptsList qcow_create_opts = {
1146 .name = "qcow-create-opts",
1147 .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
1150 .name = BLOCK_OPT_SIZE,
1151 .type = QEMU_OPT_SIZE,
1152 .help = "Virtual disk size"
1155 .name = BLOCK_OPT_BACKING_FILE,
1156 .type = QEMU_OPT_STRING,
1157 .help = "File name of a base image"
1160 .name = BLOCK_OPT_ENCRYPT,
1161 .type = QEMU_OPT_BOOL,
1162 .help = "Encrypt the image with format 'aes'. (Deprecated "
1163 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",
1166 .name = BLOCK_OPT_ENCRYPT_FORMAT,
1167 .type = QEMU_OPT_STRING,
1168 .help = "Encrypt the image, format choices: 'aes'",
1170 BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
1171 { /* end of list */ }
1175 static const char *const qcow_strong_runtime_opts[] = {
1176 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET,
1181 static BlockDriver bdrv_qcow = {
1182 .format_name = "qcow",
1183 .instance_size = sizeof(BDRVQcowState),
1184 .bdrv_probe = qcow_probe,
1185 .bdrv_open = qcow_open,
1186 .bdrv_close = qcow_close,
1187 .bdrv_child_perm = bdrv_format_default_perms,
1188 .bdrv_reopen_prepare = qcow_reopen_prepare,
1189 .bdrv_co_create = qcow_co_create,
1190 .bdrv_co_create_opts = qcow_co_create_opts,
1191 .bdrv_has_zero_init = bdrv_has_zero_init_1,
1192 .supports_backing = true,
1193 .bdrv_refresh_limits = qcow_refresh_limits,
1195 .bdrv_co_preadv = qcow_co_preadv,
1196 .bdrv_co_pwritev = qcow_co_pwritev,
1197 .bdrv_co_block_status = qcow_co_block_status,
1199 .bdrv_make_empty = qcow_make_empty,
1200 .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed,
1201 .bdrv_get_info = qcow_get_info,
1203 .create_opts = &qcow_create_opts,
1204 .strong_runtime_opts = qcow_strong_runtime_opts,
1207 static void bdrv_qcow_init(void)
1209 bdrv_register(&bdrv_qcow);
1212 block_init(bdrv_qcow_init);