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, "Unsupported qcow version %" PRIu32, header.version);
164 if (header.size <= 1) {
165 error_setg(errp, "Image size is too small (must be at least 2 bytes)");
169 if (header.cluster_bits < 9 || header.cluster_bits > 16) {
170 error_setg(errp, "Cluster size must be between 512 and 64k");
175 /* l2_bits specifies number of entries; storing a uint64_t in each entry,
176 * so bytes = num_entries << 3. */
177 if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) {
178 error_setg(errp, "L2 table size must be between 512 and 64k");
183 s->crypt_method_header = header.crypt_method;
184 if (s->crypt_method_header) {
185 if (bdrv_uses_whitelist() &&
186 s->crypt_method_header == QCOW_CRYPT_AES) {
188 "Use of AES-CBC encrypted qcow images is no longer "
189 "supported in system emulators");
190 error_append_hint(errp,
191 "You can use 'qemu-img convert' to convert your "
192 "image to an alternative supported format, such "
193 "as unencrypted qcow, or raw with the LUKS "
194 "format instead.\n");
198 if (s->crypt_method_header == QCOW_CRYPT_AES) {
199 if (encryptfmt && !g_str_equal(encryptfmt, "aes")) {
201 "Header reported 'aes' encryption format but "
202 "options specify '%s'", encryptfmt);
206 qdict_put_str(encryptopts, "format", "qcow");
207 crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
213 if (flags & BDRV_O_NO_IO) {
214 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
216 s->crypto = qcrypto_block_open(crypto_opts, "encrypt.",
217 NULL, NULL, cflags, 1, errp);
223 error_setg(errp, "invalid encryption method in qcow header");
227 bs->encrypted = true;
230 error_setg(errp, "No encryption in image header, but options "
231 "specified format '%s'", encryptfmt);
236 s->cluster_bits = header.cluster_bits;
237 s->cluster_size = 1 << s->cluster_bits;
238 s->l2_bits = header.l2_bits;
239 s->l2_size = 1 << s->l2_bits;
240 bs->total_sectors = header.size / 512;
241 s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
243 /* read the level 1 table */
244 shift = s->cluster_bits + s->l2_bits;
245 if (header.size > UINT64_MAX - (1LL << shift)) {
246 error_setg(errp, "Image too large");
250 uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift;
251 if (l1_size > INT_MAX / sizeof(uint64_t)) {
252 error_setg(errp, "Image too large");
256 s->l1_size = l1_size;
259 s->l1_table_offset = header.l1_table_offset;
260 s->l1_table = g_try_new(uint64_t, s->l1_size);
261 if (s->l1_table == NULL) {
262 error_setg(errp, "Could not allocate memory for L1 table");
267 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
268 s->l1_size * sizeof(uint64_t));
273 for(i = 0;i < s->l1_size; i++) {
274 s->l1_table[i] = be64_to_cpu(s->l1_table[i]);
277 /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
279 qemu_try_blockalign(bs->file->bs,
280 s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
281 if (s->l2_cache == NULL) {
282 error_setg(errp, "Could not allocate L2 table cache");
286 s->cluster_cache = g_malloc(s->cluster_size);
287 s->cluster_data = g_malloc(s->cluster_size);
288 s->cluster_cache_offset = -1;
290 /* read the backing file name */
291 if (header.backing_file_offset != 0) {
292 len = header.backing_file_size;
293 if (len > 1023 || len >= sizeof(bs->backing_file)) {
294 error_setg(errp, "Backing file name too long");
298 ret = bdrv_pread(bs->file, header.backing_file_offset,
299 bs->auto_backing_file, len);
303 bs->auto_backing_file[len] = '\0';
304 pstrcpy(bs->backing_file, sizeof(bs->backing_file),
305 bs->auto_backing_file);
308 /* Disable migration when qcow images are used */
309 error_setg(&s->migration_blocker, "The qcow format used by node '%s' "
310 "does not support live migration",
311 bdrv_get_device_or_node_name(bs));
312 ret = migrate_add_blocker(s->migration_blocker, &local_err);
314 error_propagate(errp, local_err);
315 error_free(s->migration_blocker);
319 qobject_unref(encryptopts);
320 qapi_free_QCryptoBlockOpenOptions(crypto_opts);
321 qemu_co_mutex_init(&s->lock);
326 qemu_vfree(s->l2_cache);
327 g_free(s->cluster_cache);
328 g_free(s->cluster_data);
329 qcrypto_block_free(s->crypto);
330 qobject_unref(encryptopts);
331 qapi_free_QCryptoBlockOpenOptions(crypto_opts);
336 /* We have nothing to do for QCOW reopen, stubs just return
338 static int qcow_reopen_prepare(BDRVReopenState *state,
339 BlockReopenQueue *queue, Error **errp)
349 * 1 to allocate a normal cluster (for sector-aligned byte offsets 'n_start'
350 * to 'n_end' within the cluster)
352 * 2 to allocate a compressed cluster of size
353 * 'compressed_size'. 'compressed_size' must be > 0 and <
356 * return 0 if not allocated, 1 if *result is assigned, and negative
359 static int get_cluster_offset(BlockDriverState *bs,
360 uint64_t offset, int allocate,
362 int n_start, int n_end, uint64_t *result)
364 BDRVQcowState *s = bs->opaque;
365 int min_index, i, j, l1_index, l2_index, ret;
367 uint64_t *l2_table, cluster_offset, tmp;
372 l1_index = offset >> (s->l2_bits + s->cluster_bits);
373 l2_offset = s->l1_table[l1_index];
378 /* allocate a new l2 entry */
379 l2_offset = bdrv_getlength(bs->file->bs);
383 /* round to cluster size */
384 l2_offset = QEMU_ALIGN_UP(l2_offset, s->cluster_size);
385 /* update the L1 entry */
386 s->l1_table[l1_index] = l2_offset;
387 tmp = cpu_to_be64(l2_offset);
388 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
389 ret = bdrv_pwrite_sync(bs->file,
390 s->l1_table_offset + l1_index * sizeof(tmp),
397 for(i = 0; i < L2_CACHE_SIZE; i++) {
398 if (l2_offset == s->l2_cache_offsets[i]) {
399 /* increment the hit count */
400 if (++s->l2_cache_counts[i] == 0xffffffff) {
401 for(j = 0; j < L2_CACHE_SIZE; j++) {
402 s->l2_cache_counts[j] >>= 1;
405 l2_table = s->l2_cache + (i << s->l2_bits);
409 /* not found: load a new entry in the least used one */
411 min_count = 0xffffffff;
412 for(i = 0; i < L2_CACHE_SIZE; i++) {
413 if (s->l2_cache_counts[i] < min_count) {
414 min_count = s->l2_cache_counts[i];
418 l2_table = s->l2_cache + (min_index << s->l2_bits);
419 BLKDBG_EVENT(bs->file, BLKDBG_L2_LOAD);
421 memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
422 ret = bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
423 s->l2_size * sizeof(uint64_t));
428 ret = bdrv_pread(bs->file, l2_offset, l2_table,
429 s->l2_size * sizeof(uint64_t));
434 s->l2_cache_offsets[min_index] = l2_offset;
435 s->l2_cache_counts[min_index] = 1;
437 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
438 cluster_offset = be64_to_cpu(l2_table[l2_index]);
439 if (!cluster_offset ||
440 ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
443 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
444 assert(QEMU_IS_ALIGNED(n_start | n_end, BDRV_SECTOR_SIZE));
445 /* allocate a new cluster */
446 if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
447 (n_end - n_start) < s->cluster_size) {
448 /* if the cluster is already compressed, we must
449 decompress it in the case it is not completely
451 if (decompress_cluster(bs, cluster_offset) < 0) {
454 cluster_offset = bdrv_getlength(bs->file->bs);
455 if ((int64_t) cluster_offset < 0) {
456 return cluster_offset;
458 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
459 /* write the cluster content */
460 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
461 ret = bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache,
467 cluster_offset = bdrv_getlength(bs->file->bs);
468 if ((int64_t) cluster_offset < 0) {
469 return cluster_offset;
472 /* round to cluster size */
473 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
474 if (cluster_offset + s->cluster_size > INT64_MAX) {
477 ret = bdrv_truncate(bs->file, cluster_offset + s->cluster_size,
478 PREALLOC_MODE_OFF, NULL);
482 /* if encrypted, we must initialize the cluster
483 content which won't be written */
485 (n_end - n_start) < s->cluster_size) {
486 uint64_t start_offset;
488 start_offset = offset & ~(s->cluster_size - 1);
489 for (i = 0; i < s->cluster_size; i += BDRV_SECTOR_SIZE) {
490 if (i < n_start || i >= n_end) {
491 memset(s->cluster_data, 0x00, BDRV_SECTOR_SIZE);
492 if (qcrypto_block_encrypt(s->crypto,
499 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
500 ret = bdrv_pwrite(bs->file,
510 } else if (allocate == 2) {
511 cluster_offset |= QCOW_OFLAG_COMPRESSED |
512 (uint64_t)compressed_size << (63 - s->cluster_bits);
515 /* update L2 table */
516 tmp = cpu_to_be64(cluster_offset);
517 l2_table[l2_index] = tmp;
519 BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
521 BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE);
523 ret = bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
529 *result = cluster_offset;
533 static int coroutine_fn qcow_co_block_status(BlockDriverState *bs,
535 int64_t offset, int64_t bytes,
536 int64_t *pnum, int64_t *map,
537 BlockDriverState **file)
539 BDRVQcowState *s = bs->opaque;
540 int index_in_cluster, ret;
542 uint64_t cluster_offset;
544 qemu_co_mutex_lock(&s->lock);
545 ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset);
546 qemu_co_mutex_unlock(&s->lock);
550 index_in_cluster = offset & (s->cluster_size - 1);
551 n = s->cluster_size - index_in_cluster;
556 if (!cluster_offset) {
559 if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->crypto) {
560 return BDRV_BLOCK_DATA;
562 *map = cluster_offset | index_in_cluster;
563 *file = bs->file->bs;
564 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
567 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
568 const uint8_t *buf, int buf_size)
570 z_stream strm1, *strm = &strm1;
573 memset(strm, 0, sizeof(*strm));
575 strm->next_in = (uint8_t *)buf;
576 strm->avail_in = buf_size;
577 strm->next_out = out_buf;
578 strm->avail_out = out_buf_size;
580 ret = inflateInit2(strm, -12);
583 ret = inflate(strm, Z_FINISH);
584 out_len = strm->next_out - out_buf;
585 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
586 out_len != out_buf_size) {
594 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
596 BDRVQcowState *s = bs->opaque;
600 coffset = cluster_offset & s->cluster_offset_mask;
601 if (s->cluster_cache_offset != coffset) {
602 csize = cluster_offset >> (63 - s->cluster_bits);
603 csize &= (s->cluster_size - 1);
604 BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
605 ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
608 if (decompress_buffer(s->cluster_cache, s->cluster_size,
609 s->cluster_data, csize) < 0) {
612 s->cluster_cache_offset = coffset;
617 static void qcow_refresh_limits(BlockDriverState *bs, Error **errp)
619 /* At least encrypted images require 512-byte alignment. Apply the
620 * limit universally, rather than just on encrypted images, as
621 * it's easier to let the block layer handle rounding than to
622 * audit this code further. */
623 bs->bl.request_alignment = BDRV_SECTOR_SIZE;
626 static coroutine_fn int qcow_co_preadv(BlockDriverState *bs, uint64_t offset,
627 uint64_t bytes, QEMUIOVector *qiov,
630 BDRVQcowState *s = bs->opaque;
631 int offset_in_cluster;
633 uint64_t cluster_offset;
638 if (qiov->niov > 1) {
639 buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
645 buf = (uint8_t *)qiov->iov->iov_base;
648 qemu_co_mutex_lock(&s->lock);
651 /* prepare next request */
652 ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset);
656 offset_in_cluster = offset & (s->cluster_size - 1);
657 n = s->cluster_size - offset_in_cluster;
662 if (!cluster_offset) {
664 /* read from the base image */
665 qemu_co_mutex_unlock(&s->lock);
666 /* qcow2 emits this on bs->file instead of bs->backing */
667 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
668 ret = bdrv_co_pread(bs->backing, offset, n, buf, 0);
669 qemu_co_mutex_lock(&s->lock);
674 /* Note: in this case, no need to wait */
677 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
678 /* add AIO support for compressed blocks ? */
679 if (decompress_cluster(bs, cluster_offset) < 0) {
683 memcpy(buf, s->cluster_cache + offset_in_cluster, n);
685 if ((cluster_offset & 511) != 0) {
689 qemu_co_mutex_unlock(&s->lock);
690 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
691 ret = bdrv_co_pread(bs->file, cluster_offset + offset_in_cluster,
693 qemu_co_mutex_lock(&s->lock);
699 if (qcrypto_block_decrypt(s->crypto,
700 offset, buf, n, NULL) < 0) {
713 qemu_co_mutex_unlock(&s->lock);
715 if (qiov->niov > 1) {
716 qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
717 qemu_vfree(orig_buf);
723 static coroutine_fn int qcow_co_pwritev(BlockDriverState *bs, uint64_t offset,
724 uint64_t bytes, QEMUIOVector *qiov,
727 BDRVQcowState *s = bs->opaque;
728 int offset_in_cluster;
729 uint64_t cluster_offset;
735 s->cluster_cache_offset = -1; /* disable compressed cache */
737 /* We must always copy the iov when encrypting, so we
738 * don't modify the original data buffer during encryption */
739 if (bs->encrypted || qiov->niov > 1) {
740 buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
744 qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
747 buf = (uint8_t *)qiov->iov->iov_base;
750 qemu_co_mutex_lock(&s->lock);
753 offset_in_cluster = offset & (s->cluster_size - 1);
754 n = s->cluster_size - offset_in_cluster;
758 ret = get_cluster_offset(bs, offset, 1, 0, offset_in_cluster,
759 offset_in_cluster + n, &cluster_offset);
763 if (!cluster_offset || (cluster_offset & 511) != 0) {
769 if (qcrypto_block_encrypt(s->crypto, offset, buf, n, NULL) < 0) {
775 qemu_co_mutex_unlock(&s->lock);
776 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
777 ret = bdrv_co_pwrite(bs->file, cluster_offset + offset_in_cluster,
779 qemu_co_mutex_lock(&s->lock);
789 qemu_co_mutex_unlock(&s->lock);
791 qemu_vfree(orig_buf);
796 static void qcow_close(BlockDriverState *bs)
798 BDRVQcowState *s = bs->opaque;
800 qcrypto_block_free(s->crypto);
803 qemu_vfree(s->l2_cache);
804 g_free(s->cluster_cache);
805 g_free(s->cluster_data);
807 migrate_del_blocker(s->migration_blocker);
808 error_free(s->migration_blocker);
811 static int coroutine_fn qcow_co_create(BlockdevCreateOptions *opts,
814 BlockdevCreateOptionsQcow *qcow_opts;
815 int header_size, backing_filename_len, l1_size, shift, i;
818 int64_t total_size = 0;
820 BlockDriverState *bs;
821 BlockBackend *qcow_blk;
822 QCryptoBlock *crypto = NULL;
824 assert(opts->driver == BLOCKDEV_DRIVER_QCOW);
825 qcow_opts = &opts->u.qcow;
828 total_size = qcow_opts->size;
829 if (total_size == 0) {
830 error_setg(errp, "Image size is too small, cannot be zero length");
834 if (qcow_opts->has_encrypt &&
835 qcow_opts->encrypt->format != Q_CRYPTO_BLOCK_FORMAT_QCOW)
837 error_setg(errp, "Unsupported encryption format");
841 /* Create BlockBackend to write to the image */
842 bs = bdrv_open_blockdev_ref(qcow_opts->file, errp);
847 qcow_blk = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL);
848 ret = blk_insert_bs(qcow_blk, bs, errp);
852 blk_set_allow_write_beyond_eof(qcow_blk, true);
854 /* Create image format */
855 ret = blk_truncate(qcow_blk, 0, PREALLOC_MODE_OFF, errp);
860 memset(&header, 0, sizeof(header));
861 header.magic = cpu_to_be32(QCOW_MAGIC);
862 header.version = cpu_to_be32(QCOW_VERSION);
863 header.size = cpu_to_be64(total_size);
864 header_size = sizeof(header);
865 backing_filename_len = 0;
866 if (qcow_opts->has_backing_file) {
867 if (strcmp(qcow_opts->backing_file, "fat:")) {
868 header.backing_file_offset = cpu_to_be64(header_size);
869 backing_filename_len = strlen(qcow_opts->backing_file);
870 header.backing_file_size = cpu_to_be32(backing_filename_len);
871 header_size += backing_filename_len;
873 /* special backing file for vvfat */
874 qcow_opts->has_backing_file = false;
876 header.cluster_bits = 9; /* 512 byte cluster to avoid copying
877 unmodified sectors */
878 header.l2_bits = 12; /* 32 KB L2 tables */
880 header.cluster_bits = 12; /* 4 KB clusters */
881 header.l2_bits = 9; /* 4 KB L2 tables */
883 header_size = (header_size + 7) & ~7;
884 shift = header.cluster_bits + header.l2_bits;
885 l1_size = (total_size + (1LL << shift) - 1) >> shift;
887 header.l1_table_offset = cpu_to_be64(header_size);
889 if (qcow_opts->has_encrypt) {
890 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
892 crypto = qcrypto_block_create(qcow_opts->encrypt, "encrypt.",
893 NULL, NULL, NULL, errp);
899 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
902 /* write all the data */
903 ret = blk_pwrite(qcow_blk, 0, &header, sizeof(header), 0);
904 if (ret != sizeof(header)) {
908 if (qcow_opts->has_backing_file) {
909 ret = blk_pwrite(qcow_blk, sizeof(header),
910 qcow_opts->backing_file, backing_filename_len, 0);
911 if (ret != backing_filename_len) {
916 tmp = g_malloc0(BDRV_SECTOR_SIZE);
917 for (i = 0; i < DIV_ROUND_UP(sizeof(uint64_t) * l1_size, BDRV_SECTOR_SIZE);
919 ret = blk_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i,
920 tmp, BDRV_SECTOR_SIZE, 0);
921 if (ret != BDRV_SECTOR_SIZE) {
932 qcrypto_block_free(crypto);
936 static int coroutine_fn qcow_co_create_opts(const char *filename,
937 QemuOpts *opts, Error **errp)
939 BlockdevCreateOptions *create_options = NULL;
940 BlockDriverState *bs = NULL;
944 Error *local_err = NULL;
947 static const QDictRenames opt_renames[] = {
948 { BLOCK_OPT_BACKING_FILE, "backing-file" },
949 { BLOCK_OPT_ENCRYPT, BLOCK_OPT_ENCRYPT_FORMAT },
953 /* Parse options and convert legacy syntax */
954 qdict = qemu_opts_to_qdict_filtered(opts, NULL, &qcow_create_opts, true);
956 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT);
957 if (val && !strcmp(val, "on")) {
958 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow");
959 } else if (val && !strcmp(val, "off")) {
960 qdict_del(qdict, BLOCK_OPT_ENCRYPT);
963 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT);
964 if (val && !strcmp(val, "aes")) {
965 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow");
968 if (!qdict_rename_keys(qdict, opt_renames, errp)) {
973 /* Create and open the file (protocol layer) */
974 ret = bdrv_create_file(filename, opts, &local_err);
976 error_propagate(errp, local_err);
980 bs = bdrv_open(filename, NULL, NULL,
981 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
987 /* Now get the QAPI type BlockdevCreateOptions */
988 qdict_put_str(qdict, "driver", "qcow");
989 qdict_put_str(qdict, "file", bs->node_name);
991 v = qobject_input_visitor_new_flat_confused(qdict, errp);
997 visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err);
1001 error_propagate(errp, local_err);
1006 /* Silently round up size */
1007 assert(create_options->driver == BLOCKDEV_DRIVER_QCOW);
1008 create_options->u.qcow.size =
1009 ROUND_UP(create_options->u.qcow.size, BDRV_SECTOR_SIZE);
1011 /* Create the qcow image (format layer) */
1012 ret = qcow_co_create(create_options, errp);
1019 qobject_unref(qdict);
1021 qapi_free_BlockdevCreateOptions(create_options);
1025 static int qcow_make_empty(BlockDriverState *bs)
1027 BDRVQcowState *s = bs->opaque;
1028 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1031 memset(s->l1_table, 0, l1_length);
1032 if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
1035 ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length,
1036 PREALLOC_MODE_OFF, NULL);
1040 memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
1041 memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
1042 memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
1047 /* XXX: put compressed sectors first, then all the cluster aligned
1048 tables to avoid losing bytes in alignment */
1049 static coroutine_fn int
1050 qcow_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
1051 uint64_t bytes, QEMUIOVector *qiov)
1053 BDRVQcowState *s = bs->opaque;
1056 uint8_t *buf, *out_buf;
1057 uint64_t cluster_offset;
1059 buf = qemu_blockalign(bs, s->cluster_size);
1060 if (bytes != s->cluster_size) {
1061 if (bytes > s->cluster_size ||
1062 offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
1067 /* Zero-pad last write if image size is not cluster aligned */
1068 memset(buf + bytes, 0, s->cluster_size - bytes);
1070 qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
1072 out_buf = g_malloc(s->cluster_size);
1074 /* best compression, small window, no zlib header */
1075 memset(&strm, 0, sizeof(strm));
1076 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1078 9, Z_DEFAULT_STRATEGY);
1084 strm.avail_in = s->cluster_size;
1085 strm.next_in = (uint8_t *)buf;
1086 strm.avail_out = s->cluster_size;
1087 strm.next_out = out_buf;
1089 ret = deflate(&strm, Z_FINISH);
1090 if (ret != Z_STREAM_END && ret != Z_OK) {
1095 out_len = strm.next_out - out_buf;
1099 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1100 /* could not compress: write normal cluster */
1101 ret = qcow_co_pwritev(bs, offset, bytes, qiov, 0);
1107 qemu_co_mutex_lock(&s->lock);
1108 ret = get_cluster_offset(bs, offset, 2, out_len, 0, 0, &cluster_offset);
1109 qemu_co_mutex_unlock(&s->lock);
1113 if (cluster_offset == 0) {
1117 cluster_offset &= s->cluster_offset_mask;
1119 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
1120 ret = bdrv_co_pwrite(bs->file, cluster_offset, out_len, out_buf, 0);
1132 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1134 BDRVQcowState *s = bs->opaque;
1135 bdi->cluster_size = s->cluster_size;
1139 static QemuOptsList qcow_create_opts = {
1140 .name = "qcow-create-opts",
1141 .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
1144 .name = BLOCK_OPT_SIZE,
1145 .type = QEMU_OPT_SIZE,
1146 .help = "Virtual disk size"
1149 .name = BLOCK_OPT_BACKING_FILE,
1150 .type = QEMU_OPT_STRING,
1151 .help = "File name of a base image"
1154 .name = BLOCK_OPT_ENCRYPT,
1155 .type = QEMU_OPT_BOOL,
1156 .help = "Encrypt the image with format 'aes'. (Deprecated "
1157 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",
1160 .name = BLOCK_OPT_ENCRYPT_FORMAT,
1161 .type = QEMU_OPT_STRING,
1162 .help = "Encrypt the image, format choices: 'aes'",
1164 BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
1165 { /* end of list */ }
1169 static const char *const qcow_strong_runtime_opts[] = {
1170 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET,
1175 static BlockDriver bdrv_qcow = {
1176 .format_name = "qcow",
1177 .instance_size = sizeof(BDRVQcowState),
1178 .bdrv_probe = qcow_probe,
1179 .bdrv_open = qcow_open,
1180 .bdrv_close = qcow_close,
1181 .bdrv_child_perm = bdrv_format_default_perms,
1182 .bdrv_reopen_prepare = qcow_reopen_prepare,
1183 .bdrv_co_create = qcow_co_create,
1184 .bdrv_co_create_opts = qcow_co_create_opts,
1185 .bdrv_has_zero_init = bdrv_has_zero_init_1,
1186 .supports_backing = true,
1187 .bdrv_refresh_limits = qcow_refresh_limits,
1189 .bdrv_co_preadv = qcow_co_preadv,
1190 .bdrv_co_pwritev = qcow_co_pwritev,
1191 .bdrv_co_block_status = qcow_co_block_status,
1193 .bdrv_make_empty = qcow_make_empty,
1194 .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed,
1195 .bdrv_get_info = qcow_get_info,
1197 .create_opts = &qcow_create_opts,
1198 .strong_runtime_opts = qcow_strong_runtime_opts,
1201 static void bdrv_qcow_init(void)
1203 bdrv_register(&bdrv_qcow);
1206 block_init(bdrv_qcow_init);