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"
35 #include "qapi/qmp/qdict.h"
36 #include "qapi/qmp/qstring.h"
37 #include "qapi/qobject-input-visitor.h"
38 #include "qapi/qapi-visit-block-core.h"
39 #include "crypto/block.h"
40 #include "migration/blocker.h"
43 /**************************************************************/
44 /* QEMU COW block driver with compression and encryption support */
46 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
47 #define QCOW_VERSION 1
49 #define QCOW_CRYPT_NONE 0
50 #define QCOW_CRYPT_AES 1
52 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
54 typedef struct QCowHeader {
57 uint64_t backing_file_offset;
58 uint32_t backing_file_size;
60 uint64_t size; /* in bytes */
64 uint32_t crypt_method;
65 uint64_t l1_table_offset;
66 } QEMU_PACKED QCowHeader;
68 #define L2_CACHE_SIZE 16
70 typedef struct BDRVQcowState {
76 uint64_t cluster_offset_mask;
77 uint64_t l1_table_offset;
80 uint64_t l2_cache_offsets[L2_CACHE_SIZE];
81 uint32_t l2_cache_counts[L2_CACHE_SIZE];
82 uint8_t *cluster_cache;
83 uint8_t *cluster_data;
84 uint64_t cluster_cache_offset;
85 QCryptoBlock *crypto; /* Disk encryption format driver */
86 uint32_t crypt_method_header;
88 Error *migration_blocker;
91 static QemuOptsList qcow_create_opts;
93 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
95 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
97 const QCowHeader *cow_header = (const void *)buf;
99 if (buf_size >= sizeof(QCowHeader) &&
100 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
101 be32_to_cpu(cow_header->version) == QCOW_VERSION)
107 static QemuOptsList qcow_runtime_opts = {
109 .head = QTAILQ_HEAD_INITIALIZER(qcow_runtime_opts.head),
111 BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
112 { /* end of list */ }
116 static int qcow_open(BlockDriverState *bs, QDict *options, int flags,
119 BDRVQcowState *s = bs->opaque;
120 unsigned int len, i, shift;
123 Error *local_err = NULL;
124 QCryptoBlockOpenOptions *crypto_opts = NULL;
125 unsigned int cflags = 0;
126 QDict *encryptopts = NULL;
127 const char *encryptfmt;
129 qdict_extract_subqdict(options, &encryptopts, "encrypt.");
130 encryptfmt = qdict_get_try_str(encryptopts, "format");
132 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
139 ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
143 header.magic = be32_to_cpu(header.magic);
144 header.version = be32_to_cpu(header.version);
145 header.backing_file_offset = be64_to_cpu(header.backing_file_offset);
146 header.backing_file_size = be32_to_cpu(header.backing_file_size);
147 header.mtime = be32_to_cpu(header.mtime);
148 header.size = be64_to_cpu(header.size);
149 header.crypt_method = be32_to_cpu(header.crypt_method);
150 header.l1_table_offset = be64_to_cpu(header.l1_table_offset);
152 if (header.magic != QCOW_MAGIC) {
153 error_setg(errp, "Image not in qcow format");
157 if (header.version != QCOW_VERSION) {
158 error_setg(errp, "Unsupported qcow version %" PRIu32, header.version);
163 if (header.size <= 1) {
164 error_setg(errp, "Image size is too small (must be at least 2 bytes)");
168 if (header.cluster_bits < 9 || header.cluster_bits > 16) {
169 error_setg(errp, "Cluster size must be between 512 and 64k");
174 /* l2_bits specifies number of entries; storing a uint64_t in each entry,
175 * so bytes = num_entries << 3. */
176 if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) {
177 error_setg(errp, "L2 table size must be between 512 and 64k");
182 s->crypt_method_header = header.crypt_method;
183 if (s->crypt_method_header) {
184 if (bdrv_uses_whitelist() &&
185 s->crypt_method_header == QCOW_CRYPT_AES) {
187 "Use of AES-CBC encrypted qcow images is no longer "
188 "supported in system emulators");
189 error_append_hint(errp,
190 "You can use 'qemu-img convert' to convert your "
191 "image to an alternative supported format, such "
192 "as unencrypted qcow, or raw with the LUKS "
193 "format instead.\n");
197 if (s->crypt_method_header == QCOW_CRYPT_AES) {
198 if (encryptfmt && !g_str_equal(encryptfmt, "aes")) {
200 "Header reported 'aes' encryption format but "
201 "options specify '%s'", encryptfmt);
205 qdict_put_str(encryptopts, "format", "qcow");
206 crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
212 if (flags & BDRV_O_NO_IO) {
213 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
215 s->crypto = qcrypto_block_open(crypto_opts, "encrypt.",
216 NULL, NULL, cflags, 1, errp);
222 error_setg(errp, "invalid encryption method in qcow header");
226 bs->encrypted = true;
229 error_setg(errp, "No encryption in image header, but options "
230 "specified format '%s'", encryptfmt);
235 s->cluster_bits = header.cluster_bits;
236 s->cluster_size = 1 << s->cluster_bits;
237 s->l2_bits = header.l2_bits;
238 s->l2_size = 1 << s->l2_bits;
239 bs->total_sectors = header.size / 512;
240 s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
242 /* read the level 1 table */
243 shift = s->cluster_bits + s->l2_bits;
244 if (header.size > UINT64_MAX - (1LL << shift)) {
245 error_setg(errp, "Image too large");
249 uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift;
250 if (l1_size > INT_MAX / sizeof(uint64_t)) {
251 error_setg(errp, "Image too large");
255 s->l1_size = l1_size;
258 s->l1_table_offset = header.l1_table_offset;
259 s->l1_table = g_try_new(uint64_t, s->l1_size);
260 if (s->l1_table == NULL) {
261 error_setg(errp, "Could not allocate memory for L1 table");
266 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
267 s->l1_size * sizeof(uint64_t));
272 for(i = 0;i < s->l1_size; i++) {
273 s->l1_table[i] = be64_to_cpu(s->l1_table[i]);
276 /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
278 qemu_try_blockalign(bs->file->bs,
279 s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
280 if (s->l2_cache == NULL) {
281 error_setg(errp, "Could not allocate L2 table cache");
285 s->cluster_cache = g_malloc(s->cluster_size);
286 s->cluster_data = g_malloc(s->cluster_size);
287 s->cluster_cache_offset = -1;
289 /* read the backing file name */
290 if (header.backing_file_offset != 0) {
291 len = header.backing_file_size;
292 if (len > 1023 || len >= sizeof(bs->backing_file)) {
293 error_setg(errp, "Backing file name too long");
297 ret = bdrv_pread(bs->file, header.backing_file_offset,
298 bs->backing_file, len);
302 bs->backing_file[len] = '\0';
305 /* Disable migration when qcow images are used */
306 error_setg(&s->migration_blocker, "The qcow format used by node '%s' "
307 "does not support live migration",
308 bdrv_get_device_or_node_name(bs));
309 ret = migrate_add_blocker(s->migration_blocker, &local_err);
311 error_propagate(errp, local_err);
312 error_free(s->migration_blocker);
316 qobject_unref(encryptopts);
317 qapi_free_QCryptoBlockOpenOptions(crypto_opts);
318 qemu_co_mutex_init(&s->lock);
323 qemu_vfree(s->l2_cache);
324 g_free(s->cluster_cache);
325 g_free(s->cluster_data);
326 qcrypto_block_free(s->crypto);
327 qobject_unref(encryptopts);
328 qapi_free_QCryptoBlockOpenOptions(crypto_opts);
333 /* We have nothing to do for QCOW reopen, stubs just return
335 static int qcow_reopen_prepare(BDRVReopenState *state,
336 BlockReopenQueue *queue, Error **errp)
346 * 1 to allocate a normal cluster (for sector-aligned byte offsets 'n_start'
347 * to 'n_end' within the cluster)
349 * 2 to allocate a compressed cluster of size
350 * 'compressed_size'. 'compressed_size' must be > 0 and <
353 * return 0 if not allocated, 1 if *result is assigned, and negative
356 static int get_cluster_offset(BlockDriverState *bs,
357 uint64_t offset, int allocate,
359 int n_start, int n_end, uint64_t *result)
361 BDRVQcowState *s = bs->opaque;
362 int min_index, i, j, l1_index, l2_index, ret;
364 uint64_t *l2_table, cluster_offset, tmp;
369 l1_index = offset >> (s->l2_bits + s->cluster_bits);
370 l2_offset = s->l1_table[l1_index];
375 /* allocate a new l2 entry */
376 l2_offset = bdrv_getlength(bs->file->bs);
380 /* round to cluster size */
381 l2_offset = QEMU_ALIGN_UP(l2_offset, s->cluster_size);
382 /* update the L1 entry */
383 s->l1_table[l1_index] = l2_offset;
384 tmp = cpu_to_be64(l2_offset);
385 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
386 ret = bdrv_pwrite_sync(bs->file,
387 s->l1_table_offset + l1_index * sizeof(tmp),
394 for(i = 0; i < L2_CACHE_SIZE; i++) {
395 if (l2_offset == s->l2_cache_offsets[i]) {
396 /* increment the hit count */
397 if (++s->l2_cache_counts[i] == 0xffffffff) {
398 for(j = 0; j < L2_CACHE_SIZE; j++) {
399 s->l2_cache_counts[j] >>= 1;
402 l2_table = s->l2_cache + (i << s->l2_bits);
406 /* not found: load a new entry in the least used one */
408 min_count = 0xffffffff;
409 for(i = 0; i < L2_CACHE_SIZE; i++) {
410 if (s->l2_cache_counts[i] < min_count) {
411 min_count = s->l2_cache_counts[i];
415 l2_table = s->l2_cache + (min_index << s->l2_bits);
416 BLKDBG_EVENT(bs->file, BLKDBG_L2_LOAD);
418 memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
419 ret = bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
420 s->l2_size * sizeof(uint64_t));
425 ret = bdrv_pread(bs->file, l2_offset, l2_table,
426 s->l2_size * sizeof(uint64_t));
431 s->l2_cache_offsets[min_index] = l2_offset;
432 s->l2_cache_counts[min_index] = 1;
434 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
435 cluster_offset = be64_to_cpu(l2_table[l2_index]);
436 if (!cluster_offset ||
437 ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
440 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
441 assert(QEMU_IS_ALIGNED(n_start | n_end, BDRV_SECTOR_SIZE));
442 /* allocate a new cluster */
443 if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
444 (n_end - n_start) < s->cluster_size) {
445 /* if the cluster is already compressed, we must
446 decompress it in the case it is not completely
448 if (decompress_cluster(bs, cluster_offset) < 0) {
451 cluster_offset = bdrv_getlength(bs->file->bs);
452 if ((int64_t) cluster_offset < 0) {
453 return cluster_offset;
455 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
456 /* write the cluster content */
457 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
458 ret = bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache,
464 cluster_offset = bdrv_getlength(bs->file->bs);
465 if ((int64_t) cluster_offset < 0) {
466 return cluster_offset;
469 /* round to cluster size */
470 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
471 if (cluster_offset + s->cluster_size > INT64_MAX) {
474 ret = bdrv_truncate(bs->file, cluster_offset + s->cluster_size,
475 PREALLOC_MODE_OFF, NULL);
479 /* if encrypted, we must initialize the cluster
480 content which won't be written */
482 (n_end - n_start) < s->cluster_size) {
483 uint64_t start_offset;
485 start_offset = offset & ~(s->cluster_size - 1);
486 for (i = 0; i < s->cluster_size; i += BDRV_SECTOR_SIZE) {
487 if (i < n_start || i >= n_end) {
488 memset(s->cluster_data, 0x00, BDRV_SECTOR_SIZE);
489 if (qcrypto_block_encrypt(s->crypto,
496 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
497 ret = bdrv_pwrite(bs->file,
507 } else if (allocate == 2) {
508 cluster_offset |= QCOW_OFLAG_COMPRESSED |
509 (uint64_t)compressed_size << (63 - s->cluster_bits);
512 /* update L2 table */
513 tmp = cpu_to_be64(cluster_offset);
514 l2_table[l2_index] = tmp;
516 BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
518 BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE);
520 ret = bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
526 *result = cluster_offset;
530 static int coroutine_fn qcow_co_block_status(BlockDriverState *bs,
532 int64_t offset, int64_t bytes,
533 int64_t *pnum, int64_t *map,
534 BlockDriverState **file)
536 BDRVQcowState *s = bs->opaque;
537 int index_in_cluster, ret;
539 uint64_t cluster_offset;
541 qemu_co_mutex_lock(&s->lock);
542 ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset);
543 qemu_co_mutex_unlock(&s->lock);
547 index_in_cluster = offset & (s->cluster_size - 1);
548 n = s->cluster_size - index_in_cluster;
553 if (!cluster_offset) {
556 if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->crypto) {
557 return BDRV_BLOCK_DATA;
559 *map = cluster_offset | index_in_cluster;
560 *file = bs->file->bs;
561 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
564 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
565 const uint8_t *buf, int buf_size)
567 z_stream strm1, *strm = &strm1;
570 memset(strm, 0, sizeof(*strm));
572 strm->next_in = (uint8_t *)buf;
573 strm->avail_in = buf_size;
574 strm->next_out = out_buf;
575 strm->avail_out = out_buf_size;
577 ret = inflateInit2(strm, -12);
580 ret = inflate(strm, Z_FINISH);
581 out_len = strm->next_out - out_buf;
582 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
583 out_len != out_buf_size) {
591 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
593 BDRVQcowState *s = bs->opaque;
597 coffset = cluster_offset & s->cluster_offset_mask;
598 if (s->cluster_cache_offset != coffset) {
599 csize = cluster_offset >> (63 - s->cluster_bits);
600 csize &= (s->cluster_size - 1);
601 BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
602 ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
605 if (decompress_buffer(s->cluster_cache, s->cluster_size,
606 s->cluster_data, csize) < 0) {
609 s->cluster_cache_offset = coffset;
614 static void qcow_refresh_limits(BlockDriverState *bs, Error **errp)
616 /* At least encrypted images require 512-byte alignment. Apply the
617 * limit universally, rather than just on encrypted images, as
618 * it's easier to let the block layer handle rounding than to
619 * audit this code further. */
620 bs->bl.request_alignment = BDRV_SECTOR_SIZE;
623 static coroutine_fn int qcow_co_preadv(BlockDriverState *bs, uint64_t offset,
624 uint64_t bytes, QEMUIOVector *qiov,
627 BDRVQcowState *s = bs->opaque;
628 int offset_in_cluster;
630 uint64_t cluster_offset;
632 QEMUIOVector hd_qiov;
637 if (qiov->niov > 1) {
638 buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
644 buf = (uint8_t *)qiov->iov->iov_base;
647 qemu_co_mutex_lock(&s->lock);
650 /* prepare next request */
651 ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset);
655 offset_in_cluster = offset & (s->cluster_size - 1);
656 n = s->cluster_size - offset_in_cluster;
661 if (!cluster_offset) {
663 /* read from the base image */
664 hd_iov.iov_base = (void *)buf;
666 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
667 qemu_co_mutex_unlock(&s->lock);
668 /* qcow2 emits this on bs->file instead of bs->backing */
669 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
670 ret = bdrv_co_preadv(bs->backing, offset, n, &hd_qiov, 0);
671 qemu_co_mutex_lock(&s->lock);
676 /* Note: in this case, no need to wait */
679 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
680 /* add AIO support for compressed blocks ? */
681 if (decompress_cluster(bs, cluster_offset) < 0) {
685 memcpy(buf, s->cluster_cache + offset_in_cluster, n);
687 if ((cluster_offset & 511) != 0) {
691 hd_iov.iov_base = (void *)buf;
693 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
694 qemu_co_mutex_unlock(&s->lock);
695 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
696 ret = bdrv_co_preadv(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;
737 QEMUIOVector hd_qiov;
742 s->cluster_cache_offset = -1; /* disable compressed cache */
744 /* We must always copy the iov when encrypting, so we
745 * don't modify the original data buffer during encryption */
746 if (bs->encrypted || qiov->niov > 1) {
747 buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
751 qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
754 buf = (uint8_t *)qiov->iov->iov_base;
757 qemu_co_mutex_lock(&s->lock);
760 offset_in_cluster = offset & (s->cluster_size - 1);
761 n = s->cluster_size - offset_in_cluster;
765 ret = get_cluster_offset(bs, offset, 1, 0, offset_in_cluster,
766 offset_in_cluster + n, &cluster_offset);
770 if (!cluster_offset || (cluster_offset & 511) != 0) {
776 if (qcrypto_block_encrypt(s->crypto, offset, buf, n, NULL) < 0) {
782 hd_iov.iov_base = (void *)buf;
784 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
785 qemu_co_mutex_unlock(&s->lock);
786 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
787 ret = bdrv_co_pwritev(bs->file, cluster_offset + offset_in_cluster,
789 qemu_co_mutex_lock(&s->lock);
799 qemu_co_mutex_unlock(&s->lock);
801 qemu_vfree(orig_buf);
806 static void qcow_close(BlockDriverState *bs)
808 BDRVQcowState *s = bs->opaque;
810 qcrypto_block_free(s->crypto);
813 qemu_vfree(s->l2_cache);
814 g_free(s->cluster_cache);
815 g_free(s->cluster_data);
817 migrate_del_blocker(s->migration_blocker);
818 error_free(s->migration_blocker);
821 static int coroutine_fn qcow_co_create(BlockdevCreateOptions *opts,
824 BlockdevCreateOptionsQcow *qcow_opts;
825 int header_size, backing_filename_len, l1_size, shift, i;
828 int64_t total_size = 0;
830 BlockDriverState *bs;
831 BlockBackend *qcow_blk;
832 QCryptoBlock *crypto = NULL;
834 assert(opts->driver == BLOCKDEV_DRIVER_QCOW);
835 qcow_opts = &opts->u.qcow;
838 total_size = qcow_opts->size;
839 if (total_size == 0) {
840 error_setg(errp, "Image size is too small, cannot be zero length");
844 if (qcow_opts->has_encrypt &&
845 qcow_opts->encrypt->format != Q_CRYPTO_BLOCK_FORMAT_QCOW)
847 error_setg(errp, "Unsupported encryption format");
851 /* Create BlockBackend to write to the image */
852 bs = bdrv_open_blockdev_ref(qcow_opts->file, errp);
857 qcow_blk = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL);
858 ret = blk_insert_bs(qcow_blk, bs, errp);
862 blk_set_allow_write_beyond_eof(qcow_blk, true);
864 /* Create image format */
865 ret = blk_truncate(qcow_blk, 0, PREALLOC_MODE_OFF, errp);
870 memset(&header, 0, sizeof(header));
871 header.magic = cpu_to_be32(QCOW_MAGIC);
872 header.version = cpu_to_be32(QCOW_VERSION);
873 header.size = cpu_to_be64(total_size);
874 header_size = sizeof(header);
875 backing_filename_len = 0;
876 if (qcow_opts->has_backing_file) {
877 if (strcmp(qcow_opts->backing_file, "fat:")) {
878 header.backing_file_offset = cpu_to_be64(header_size);
879 backing_filename_len = strlen(qcow_opts->backing_file);
880 header.backing_file_size = cpu_to_be32(backing_filename_len);
881 header_size += backing_filename_len;
883 /* special backing file for vvfat */
884 qcow_opts->has_backing_file = false;
886 header.cluster_bits = 9; /* 512 byte cluster to avoid copying
887 unmodified sectors */
888 header.l2_bits = 12; /* 32 KB L2 tables */
890 header.cluster_bits = 12; /* 4 KB clusters */
891 header.l2_bits = 9; /* 4 KB L2 tables */
893 header_size = (header_size + 7) & ~7;
894 shift = header.cluster_bits + header.l2_bits;
895 l1_size = (total_size + (1LL << shift) - 1) >> shift;
897 header.l1_table_offset = cpu_to_be64(header_size);
899 if (qcow_opts->has_encrypt) {
900 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
902 crypto = qcrypto_block_create(qcow_opts->encrypt, "encrypt.",
903 NULL, NULL, NULL, errp);
909 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
912 /* write all the data */
913 ret = blk_pwrite(qcow_blk, 0, &header, sizeof(header), 0);
914 if (ret != sizeof(header)) {
918 if (qcow_opts->has_backing_file) {
919 ret = blk_pwrite(qcow_blk, sizeof(header),
920 qcow_opts->backing_file, backing_filename_len, 0);
921 if (ret != backing_filename_len) {
926 tmp = g_malloc0(BDRV_SECTOR_SIZE);
927 for (i = 0; i < DIV_ROUND_UP(sizeof(uint64_t) * l1_size, BDRV_SECTOR_SIZE);
929 ret = blk_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i,
930 tmp, BDRV_SECTOR_SIZE, 0);
931 if (ret != BDRV_SECTOR_SIZE) {
942 qcrypto_block_free(crypto);
946 static int coroutine_fn qcow_co_create_opts(const char *filename,
947 QemuOpts *opts, Error **errp)
949 BlockdevCreateOptions *create_options = NULL;
950 BlockDriverState *bs = NULL;
954 Error *local_err = NULL;
957 static const QDictRenames opt_renames[] = {
958 { BLOCK_OPT_BACKING_FILE, "backing-file" },
959 { BLOCK_OPT_ENCRYPT, BLOCK_OPT_ENCRYPT_FORMAT },
963 /* Parse options and convert legacy syntax */
964 qdict = qemu_opts_to_qdict_filtered(opts, NULL, &qcow_create_opts, true);
966 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT);
967 if (val && !strcmp(val, "on")) {
968 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow");
969 } else if (val && !strcmp(val, "off")) {
970 qdict_del(qdict, BLOCK_OPT_ENCRYPT);
973 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT);
974 if (val && !strcmp(val, "aes")) {
975 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow");
978 if (!qdict_rename_keys(qdict, opt_renames, errp)) {
983 /* Create and open the file (protocol layer) */
984 ret = bdrv_create_file(filename, opts, &local_err);
986 error_propagate(errp, local_err);
990 bs = bdrv_open(filename, NULL, NULL,
991 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
997 /* Now get the QAPI type BlockdevCreateOptions */
998 qdict_put_str(qdict, "driver", "qcow");
999 qdict_put_str(qdict, "file", bs->node_name);
1001 v = qobject_input_visitor_new_flat_confused(qdict, errp);
1007 visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err);
1011 error_propagate(errp, local_err);
1016 /* Silently round up size */
1017 assert(create_options->driver == BLOCKDEV_DRIVER_QCOW);
1018 create_options->u.qcow.size =
1019 ROUND_UP(create_options->u.qcow.size, BDRV_SECTOR_SIZE);
1021 /* Create the qcow image (format layer) */
1022 ret = qcow_co_create(create_options, errp);
1029 qobject_unref(qdict);
1031 qapi_free_BlockdevCreateOptions(create_options);
1035 static int qcow_make_empty(BlockDriverState *bs)
1037 BDRVQcowState *s = bs->opaque;
1038 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1041 memset(s->l1_table, 0, l1_length);
1042 if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
1045 ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length,
1046 PREALLOC_MODE_OFF, NULL);
1050 memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
1051 memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
1052 memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
1057 /* XXX: put compressed sectors first, then all the cluster aligned
1058 tables to avoid losing bytes in alignment */
1059 static coroutine_fn int
1060 qcow_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
1061 uint64_t bytes, QEMUIOVector *qiov)
1063 BDRVQcowState *s = bs->opaque;
1064 QEMUIOVector hd_qiov;
1068 uint8_t *buf, *out_buf;
1069 uint64_t cluster_offset;
1071 buf = qemu_blockalign(bs, s->cluster_size);
1072 if (bytes != s->cluster_size) {
1073 if (bytes > s->cluster_size ||
1074 offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
1079 /* Zero-pad last write if image size is not cluster aligned */
1080 memset(buf + bytes, 0, s->cluster_size - bytes);
1082 qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
1084 out_buf = g_malloc(s->cluster_size);
1086 /* best compression, small window, no zlib header */
1087 memset(&strm, 0, sizeof(strm));
1088 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1090 9, Z_DEFAULT_STRATEGY);
1096 strm.avail_in = s->cluster_size;
1097 strm.next_in = (uint8_t *)buf;
1098 strm.avail_out = s->cluster_size;
1099 strm.next_out = out_buf;
1101 ret = deflate(&strm, Z_FINISH);
1102 if (ret != Z_STREAM_END && ret != Z_OK) {
1107 out_len = strm.next_out - out_buf;
1111 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1112 /* could not compress: write normal cluster */
1113 ret = qcow_co_pwritev(bs, offset, bytes, qiov, 0);
1119 qemu_co_mutex_lock(&s->lock);
1120 ret = get_cluster_offset(bs, offset, 2, out_len, 0, 0, &cluster_offset);
1121 qemu_co_mutex_unlock(&s->lock);
1125 if (cluster_offset == 0) {
1129 cluster_offset &= s->cluster_offset_mask;
1131 iov = (struct iovec) {
1132 .iov_base = out_buf,
1135 qemu_iovec_init_external(&hd_qiov, &iov, 1);
1136 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
1137 ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0);
1149 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1151 BDRVQcowState *s = bs->opaque;
1152 bdi->cluster_size = s->cluster_size;
1156 static QemuOptsList qcow_create_opts = {
1157 .name = "qcow-create-opts",
1158 .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
1161 .name = BLOCK_OPT_SIZE,
1162 .type = QEMU_OPT_SIZE,
1163 .help = "Virtual disk size"
1166 .name = BLOCK_OPT_BACKING_FILE,
1167 .type = QEMU_OPT_STRING,
1168 .help = "File name of a base image"
1171 .name = BLOCK_OPT_ENCRYPT,
1172 .type = QEMU_OPT_BOOL,
1173 .help = "Encrypt the image with format 'aes'. (Deprecated "
1174 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",
1177 .name = BLOCK_OPT_ENCRYPT_FORMAT,
1178 .type = QEMU_OPT_STRING,
1179 .help = "Encrypt the image, format choices: 'aes'",
1181 BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
1182 { /* end of list */ }
1186 static BlockDriver bdrv_qcow = {
1187 .format_name = "qcow",
1188 .instance_size = sizeof(BDRVQcowState),
1189 .bdrv_probe = qcow_probe,
1190 .bdrv_open = qcow_open,
1191 .bdrv_close = qcow_close,
1192 .bdrv_child_perm = bdrv_format_default_perms,
1193 .bdrv_reopen_prepare = qcow_reopen_prepare,
1194 .bdrv_co_create = qcow_co_create,
1195 .bdrv_co_create_opts = qcow_co_create_opts,
1196 .bdrv_has_zero_init = bdrv_has_zero_init_1,
1197 .supports_backing = true,
1198 .bdrv_refresh_limits = qcow_refresh_limits,
1200 .bdrv_co_preadv = qcow_co_preadv,
1201 .bdrv_co_pwritev = qcow_co_pwritev,
1202 .bdrv_co_block_status = qcow_co_block_status,
1204 .bdrv_make_empty = qcow_make_empty,
1205 .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed,
1206 .bdrv_get_info = qcow_get_info,
1208 .create_opts = &qcow_create_opts,
1211 static void bdrv_qcow_init(void)
1213 bdrv_register(&bdrv_qcow);
1216 block_init(bdrv_qcow_init);