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
24 #include "qemu-common.h"
25 #include "block_int.h"
30 /**************************************************************/
31 /* QEMU COW block driver with compression and encryption support */
33 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
34 #define QCOW_VERSION 1
36 #define QCOW_CRYPT_NONE 0
37 #define QCOW_CRYPT_AES 1
39 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
41 typedef struct QCowHeader {
44 uint64_t backing_file_offset;
45 uint32_t backing_file_size;
47 uint64_t size; /* in bytes */
50 uint32_t crypt_method;
51 uint64_t l1_table_offset;
54 #define L2_CACHE_SIZE 16
56 typedef struct BDRVQcowState {
64 uint64_t cluster_offset_mask;
65 uint64_t l1_table_offset;
68 uint64_t l2_cache_offsets[L2_CACHE_SIZE];
69 uint32_t l2_cache_counts[L2_CACHE_SIZE];
70 uint8_t *cluster_cache;
71 uint8_t *cluster_data;
72 uint64_t cluster_cache_offset;
73 uint32_t crypt_method; /* current crypt method, 0 if no key yet */
74 uint32_t crypt_method_header;
75 AES_KEY aes_encrypt_key;
76 AES_KEY aes_decrypt_key;
79 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
81 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
83 const QCowHeader *cow_header = (const void *)buf;
85 if (buf_size >= sizeof(QCowHeader) &&
86 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
87 be32_to_cpu(cow_header->version) == QCOW_VERSION)
93 static int qcow_open(BlockDriverState *bs, int flags)
95 BDRVQcowState *s = bs->opaque;
99 if (bdrv_pread(bs->file, 0, &header, sizeof(header)) != sizeof(header))
101 be32_to_cpus(&header.magic);
102 be32_to_cpus(&header.version);
103 be64_to_cpus(&header.backing_file_offset);
104 be32_to_cpus(&header.backing_file_size);
105 be32_to_cpus(&header.mtime);
106 be64_to_cpus(&header.size);
107 be32_to_cpus(&header.crypt_method);
108 be64_to_cpus(&header.l1_table_offset);
110 if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
112 if (header.size <= 1 || header.cluster_bits < 9)
114 if (header.crypt_method > QCOW_CRYPT_AES)
116 s->crypt_method_header = header.crypt_method;
117 if (s->crypt_method_header)
119 s->cluster_bits = header.cluster_bits;
120 s->cluster_size = 1 << s->cluster_bits;
121 s->cluster_sectors = 1 << (s->cluster_bits - 9);
122 s->l2_bits = header.l2_bits;
123 s->l2_size = 1 << s->l2_bits;
124 bs->total_sectors = header.size / 512;
125 s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
127 /* read the level 1 table */
128 shift = s->cluster_bits + s->l2_bits;
129 s->l1_size = (header.size + (1LL << shift) - 1) >> shift;
131 s->l1_table_offset = header.l1_table_offset;
132 s->l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
135 if (bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
136 s->l1_size * sizeof(uint64_t))
138 for(i = 0;i < s->l1_size; i++) {
139 be64_to_cpus(&s->l1_table[i]);
142 s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
145 s->cluster_cache = qemu_malloc(s->cluster_size);
146 if (!s->cluster_cache)
148 s->cluster_data = qemu_malloc(s->cluster_size);
149 if (!s->cluster_data)
151 s->cluster_cache_offset = -1;
153 /* read the backing file name */
154 if (header.backing_file_offset != 0) {
155 len = header.backing_file_size;
158 if (bdrv_pread(bs->file, header.backing_file_offset, bs->backing_file, len) != len)
160 bs->backing_file[len] = '\0';
165 qemu_free(s->l1_table);
166 qemu_free(s->l2_cache);
167 qemu_free(s->cluster_cache);
168 qemu_free(s->cluster_data);
172 static int qcow_set_key(BlockDriverState *bs, const char *key)
174 BDRVQcowState *s = bs->opaque;
178 memset(keybuf, 0, 16);
182 /* XXX: we could compress the chars to 7 bits to increase
184 for(i = 0;i < len;i++) {
187 s->crypt_method = s->crypt_method_header;
189 if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
191 if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
201 AES_encrypt(in, tmp, &s->aes_encrypt_key);
202 AES_decrypt(tmp, out, &s->aes_decrypt_key);
203 for(i = 0; i < 16; i++)
204 printf(" %02x", tmp[i]);
206 for(i = 0; i < 16; i++)
207 printf(" %02x", out[i]);
214 /* The crypt function is compatible with the linux cryptoloop
215 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
217 static void encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
218 uint8_t *out_buf, const uint8_t *in_buf,
219 int nb_sectors, int enc,
228 for(i = 0; i < nb_sectors; i++) {
229 ivec.ll[0] = cpu_to_le64(sector_num);
231 AES_cbc_encrypt(in_buf, out_buf, 512, key,
243 * 1 to allocate a normal cluster (for sector indexes 'n_start' to
246 * 2 to allocate a compressed cluster of size
247 * 'compressed_size'. 'compressed_size' must be > 0 and <
250 * return 0 if not allocated.
252 static uint64_t get_cluster_offset(BlockDriverState *bs,
253 uint64_t offset, int allocate,
255 int n_start, int n_end)
257 BDRVQcowState *s = bs->opaque;
258 int min_index, i, j, l1_index, l2_index;
259 uint64_t l2_offset, *l2_table, cluster_offset, tmp;
263 l1_index = offset >> (s->l2_bits + s->cluster_bits);
264 l2_offset = s->l1_table[l1_index];
269 /* allocate a new l2 entry */
270 l2_offset = bdrv_getlength(bs->file);
271 /* round to cluster size */
272 l2_offset = (l2_offset + s->cluster_size - 1) & ~(s->cluster_size - 1);
273 /* update the L1 entry */
274 s->l1_table[l1_index] = l2_offset;
275 tmp = cpu_to_be64(l2_offset);
276 if (bdrv_pwrite_sync(bs->file,
277 s->l1_table_offset + l1_index * sizeof(tmp),
278 &tmp, sizeof(tmp)) < 0)
282 for(i = 0; i < L2_CACHE_SIZE; i++) {
283 if (l2_offset == s->l2_cache_offsets[i]) {
284 /* increment the hit count */
285 if (++s->l2_cache_counts[i] == 0xffffffff) {
286 for(j = 0; j < L2_CACHE_SIZE; j++) {
287 s->l2_cache_counts[j] >>= 1;
290 l2_table = s->l2_cache + (i << s->l2_bits);
294 /* not found: load a new entry in the least used one */
296 min_count = 0xffffffff;
297 for(i = 0; i < L2_CACHE_SIZE; i++) {
298 if (s->l2_cache_counts[i] < min_count) {
299 min_count = s->l2_cache_counts[i];
303 l2_table = s->l2_cache + (min_index << s->l2_bits);
305 memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
306 if (bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
307 s->l2_size * sizeof(uint64_t)) < 0)
310 if (bdrv_pread(bs->file, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
311 s->l2_size * sizeof(uint64_t))
314 s->l2_cache_offsets[min_index] = l2_offset;
315 s->l2_cache_counts[min_index] = 1;
317 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
318 cluster_offset = be64_to_cpu(l2_table[l2_index]);
319 if (!cluster_offset ||
320 ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
323 /* allocate a new cluster */
324 if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
325 (n_end - n_start) < s->cluster_sectors) {
326 /* if the cluster is already compressed, we must
327 decompress it in the case it is not completely
329 if (decompress_cluster(bs, cluster_offset) < 0)
331 cluster_offset = bdrv_getlength(bs->file);
332 cluster_offset = (cluster_offset + s->cluster_size - 1) &
333 ~(s->cluster_size - 1);
334 /* write the cluster content */
335 if (bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache, s->cluster_size) !=
339 cluster_offset = bdrv_getlength(bs->file);
341 /* round to cluster size */
342 cluster_offset = (cluster_offset + s->cluster_size - 1) &
343 ~(s->cluster_size - 1);
344 bdrv_truncate(bs->file, cluster_offset + s->cluster_size);
345 /* if encrypted, we must initialize the cluster
346 content which won't be written */
347 if (s->crypt_method &&
348 (n_end - n_start) < s->cluster_sectors) {
350 start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
351 memset(s->cluster_data + 512, 0x00, 512);
352 for(i = 0; i < s->cluster_sectors; i++) {
353 if (i < n_start || i >= n_end) {
354 encrypt_sectors(s, start_sect + i,
356 s->cluster_data + 512, 1, 1,
357 &s->aes_encrypt_key);
358 if (bdrv_pwrite(bs->file, cluster_offset + i * 512,
359 s->cluster_data, 512) != 512)
364 } else if (allocate == 2) {
365 cluster_offset |= QCOW_OFLAG_COMPRESSED |
366 (uint64_t)compressed_size << (63 - s->cluster_bits);
369 /* update L2 table */
370 tmp = cpu_to_be64(cluster_offset);
371 l2_table[l2_index] = tmp;
372 if (bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
373 &tmp, sizeof(tmp)) < 0)
376 return cluster_offset;
379 static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
380 int nb_sectors, int *pnum)
382 BDRVQcowState *s = bs->opaque;
383 int index_in_cluster, n;
384 uint64_t cluster_offset;
386 cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
387 index_in_cluster = sector_num & (s->cluster_sectors - 1);
388 n = s->cluster_sectors - index_in_cluster;
392 return (cluster_offset != 0);
395 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
396 const uint8_t *buf, int buf_size)
398 z_stream strm1, *strm = &strm1;
401 memset(strm, 0, sizeof(*strm));
403 strm->next_in = (uint8_t *)buf;
404 strm->avail_in = buf_size;
405 strm->next_out = out_buf;
406 strm->avail_out = out_buf_size;
408 ret = inflateInit2(strm, -12);
411 ret = inflate(strm, Z_FINISH);
412 out_len = strm->next_out - out_buf;
413 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
414 out_len != out_buf_size) {
422 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
424 BDRVQcowState *s = bs->opaque;
428 coffset = cluster_offset & s->cluster_offset_mask;
429 if (s->cluster_cache_offset != coffset) {
430 csize = cluster_offset >> (63 - s->cluster_bits);
431 csize &= (s->cluster_size - 1);
432 ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
435 if (decompress_buffer(s->cluster_cache, s->cluster_size,
436 s->cluster_data, csize) < 0) {
439 s->cluster_cache_offset = coffset;
446 static int qcow_read(BlockDriverState *bs, int64_t sector_num,
447 uint8_t *buf, int nb_sectors)
449 BDRVQcowState *s = bs->opaque;
450 int ret, index_in_cluster, n;
451 uint64_t cluster_offset;
453 while (nb_sectors > 0) {
454 cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
455 index_in_cluster = sector_num & (s->cluster_sectors - 1);
456 n = s->cluster_sectors - index_in_cluster;
459 if (!cluster_offset) {
460 if (bs->backing_hd) {
461 /* read from the base image */
462 ret = bdrv_read(bs->backing_hd, sector_num, buf, n);
466 memset(buf, 0, 512 * n);
468 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
469 if (decompress_cluster(bs, cluster_offset) < 0)
471 memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
473 ret = bdrv_pread(bs->file, cluster_offset + index_in_cluster * 512, buf, n * 512);
476 if (s->crypt_method) {
477 encrypt_sectors(s, sector_num, buf, buf, n, 0,
478 &s->aes_decrypt_key);
489 typedef struct QCowAIOCB {
490 BlockDriverAIOCB common;
497 uint64_t cluster_offset;
498 uint8_t *cluster_data;
500 QEMUIOVector hd_qiov;
501 BlockDriverAIOCB *hd_aiocb;
504 static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
506 QCowAIOCB *acb = container_of(blockacb, QCowAIOCB, common);
508 bdrv_aio_cancel(acb->hd_aiocb);
509 qemu_aio_release(acb);
512 static AIOPool qcow_aio_pool = {
513 .aiocb_size = sizeof(QCowAIOCB),
514 .cancel = qcow_aio_cancel,
517 static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
518 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
519 BlockDriverCompletionFunc *cb, void *opaque, int is_write)
523 acb = qemu_aio_get(&qcow_aio_pool, bs, cb, opaque);
526 acb->hd_aiocb = NULL;
527 acb->sector_num = sector_num;
529 if (qiov->niov > 1) {
530 acb->buf = acb->orig_buf = qemu_blockalign(bs, qiov->size);
532 qemu_iovec_to_buffer(qiov, acb->buf);
534 acb->buf = (uint8_t *)qiov->iov->iov_base;
536 acb->nb_sectors = nb_sectors;
538 acb->cluster_offset = 0;
542 static void qcow_aio_read_cb(void *opaque, int ret)
544 QCowAIOCB *acb = opaque;
545 BlockDriverState *bs = acb->common.bs;
546 BDRVQcowState *s = bs->opaque;
547 int index_in_cluster;
549 acb->hd_aiocb = NULL;
554 /* post process the read buffer */
555 if (!acb->cluster_offset) {
557 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
560 if (s->crypt_method) {
561 encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
563 &s->aes_decrypt_key);
567 acb->nb_sectors -= acb->n;
568 acb->sector_num += acb->n;
569 acb->buf += acb->n * 512;
571 if (acb->nb_sectors == 0) {
572 /* request completed */
577 /* prepare next AIO request */
578 acb->cluster_offset = get_cluster_offset(bs, acb->sector_num << 9,
580 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
581 acb->n = s->cluster_sectors - index_in_cluster;
582 if (acb->n > acb->nb_sectors)
583 acb->n = acb->nb_sectors;
585 if (!acb->cluster_offset) {
586 if (bs->backing_hd) {
587 /* read from the base image */
588 acb->hd_iov.iov_base = (void *)acb->buf;
589 acb->hd_iov.iov_len = acb->n * 512;
590 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
591 acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num,
592 &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
593 if (acb->hd_aiocb == NULL)
596 /* Note: in this case, no need to wait */
597 memset(acb->buf, 0, 512 * acb->n);
600 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
601 /* add AIO support for compressed blocks ? */
602 if (decompress_cluster(bs, acb->cluster_offset) < 0)
605 s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
608 if ((acb->cluster_offset & 511) != 0) {
612 acb->hd_iov.iov_base = (void *)acb->buf;
613 acb->hd_iov.iov_len = acb->n * 512;
614 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
615 acb->hd_aiocb = bdrv_aio_readv(bs->file,
616 (acb->cluster_offset >> 9) + index_in_cluster,
617 &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
618 if (acb->hd_aiocb == NULL)
625 if (acb->qiov->niov > 1) {
626 qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size);
627 qemu_vfree(acb->orig_buf);
629 acb->common.cb(acb->common.opaque, ret);
630 qemu_aio_release(acb);
633 static BlockDriverAIOCB *qcow_aio_readv(BlockDriverState *bs,
634 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
635 BlockDriverCompletionFunc *cb, void *opaque)
639 acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
643 qcow_aio_read_cb(acb, 0);
647 static void qcow_aio_write_cb(void *opaque, int ret)
649 QCowAIOCB *acb = opaque;
650 BlockDriverState *bs = acb->common.bs;
651 BDRVQcowState *s = bs->opaque;
652 int index_in_cluster;
653 uint64_t cluster_offset;
654 const uint8_t *src_buf;
656 acb->hd_aiocb = NULL;
661 acb->nb_sectors -= acb->n;
662 acb->sector_num += acb->n;
663 acb->buf += acb->n * 512;
665 if (acb->nb_sectors == 0) {
666 /* request completed */
671 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
672 acb->n = s->cluster_sectors - index_in_cluster;
673 if (acb->n > acb->nb_sectors)
674 acb->n = acb->nb_sectors;
675 cluster_offset = get_cluster_offset(bs, acb->sector_num << 9, 1, 0,
677 index_in_cluster + acb->n);
678 if (!cluster_offset || (cluster_offset & 511) != 0) {
682 if (s->crypt_method) {
683 if (!acb->cluster_data) {
684 acb->cluster_data = qemu_mallocz(s->cluster_size);
685 if (!acb->cluster_data) {
690 encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
691 acb->n, 1, &s->aes_encrypt_key);
692 src_buf = acb->cluster_data;
697 acb->hd_iov.iov_base = (void *)src_buf;
698 acb->hd_iov.iov_len = acb->n * 512;
699 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
700 acb->hd_aiocb = bdrv_aio_writev(bs->file,
701 (cluster_offset >> 9) + index_in_cluster,
702 &acb->hd_qiov, acb->n,
703 qcow_aio_write_cb, acb);
704 if (acb->hd_aiocb == NULL)
709 if (acb->qiov->niov > 1)
710 qemu_vfree(acb->orig_buf);
711 acb->common.cb(acb->common.opaque, ret);
712 qemu_aio_release(acb);
715 static BlockDriverAIOCB *qcow_aio_writev(BlockDriverState *bs,
716 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
717 BlockDriverCompletionFunc *cb, void *opaque)
719 BDRVQcowState *s = bs->opaque;
722 s->cluster_cache_offset = -1; /* disable compressed cache */
724 acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
729 qcow_aio_write_cb(acb, 0);
733 static void qcow_close(BlockDriverState *bs)
735 BDRVQcowState *s = bs->opaque;
736 qemu_free(s->l1_table);
737 qemu_free(s->l2_cache);
738 qemu_free(s->cluster_cache);
739 qemu_free(s->cluster_data);
742 static int qcow_create(const char *filename, QEMUOptionParameter *options)
744 int fd, header_size, backing_filename_len, l1_size, i, shift;
747 int64_t total_size = 0;
748 const char *backing_file = NULL;
752 /* Read out options */
753 while (options && options->name) {
754 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
755 total_size = options->value.n / 512;
756 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
757 backing_file = options->value.s;
758 } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
759 flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
764 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
767 memset(&header, 0, sizeof(header));
768 header.magic = cpu_to_be32(QCOW_MAGIC);
769 header.version = cpu_to_be32(QCOW_VERSION);
770 header.size = cpu_to_be64(total_size * 512);
771 header_size = sizeof(header);
772 backing_filename_len = 0;
774 if (strcmp(backing_file, "fat:")) {
775 header.backing_file_offset = cpu_to_be64(header_size);
776 backing_filename_len = strlen(backing_file);
777 header.backing_file_size = cpu_to_be32(backing_filename_len);
778 header_size += backing_filename_len;
780 /* special backing file for vvfat */
783 header.cluster_bits = 9; /* 512 byte cluster to avoid copying
784 unmodifyed sectors */
785 header.l2_bits = 12; /* 32 KB L2 tables */
787 header.cluster_bits = 12; /* 4 KB clusters */
788 header.l2_bits = 9; /* 4 KB L2 tables */
790 header_size = (header_size + 7) & ~7;
791 shift = header.cluster_bits + header.l2_bits;
792 l1_size = ((total_size * 512) + (1LL << shift) - 1) >> shift;
794 header.l1_table_offset = cpu_to_be64(header_size);
795 if (flags & BLOCK_FLAG_ENCRYPT) {
796 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
798 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
801 /* write all the data */
802 ret = qemu_write_full(fd, &header, sizeof(header));
803 if (ret != sizeof(header)) {
809 ret = qemu_write_full(fd, backing_file, backing_filename_len);
810 if (ret != backing_filename_len) {
816 lseek(fd, header_size, SEEK_SET);
818 for(i = 0;i < l1_size; i++) {
819 ret = qemu_write_full(fd, &tmp, sizeof(tmp));
820 if (ret != sizeof(tmp)) {
832 static int qcow_make_empty(BlockDriverState *bs)
834 BDRVQcowState *s = bs->opaque;
835 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
838 memset(s->l1_table, 0, l1_length);
839 if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
842 ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
846 memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
847 memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
848 memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
853 /* XXX: put compressed sectors first, then all the cluster aligned
854 tables to avoid losing bytes in alignment */
855 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
856 const uint8_t *buf, int nb_sectors)
858 BDRVQcowState *s = bs->opaque;
862 uint64_t cluster_offset;
864 if (nb_sectors != s->cluster_sectors)
867 out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
871 /* best compression, small window, no zlib header */
872 memset(&strm, 0, sizeof(strm));
873 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
875 9, Z_DEFAULT_STRATEGY);
881 strm.avail_in = s->cluster_size;
882 strm.next_in = (uint8_t *)buf;
883 strm.avail_out = s->cluster_size;
884 strm.next_out = out_buf;
886 ret = deflate(&strm, Z_FINISH);
887 if (ret != Z_STREAM_END && ret != Z_OK) {
892 out_len = strm.next_out - out_buf;
896 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
897 /* could not compress: write normal cluster */
898 bdrv_write(bs, sector_num, buf, s->cluster_sectors);
900 cluster_offset = get_cluster_offset(bs, sector_num << 9, 2,
902 cluster_offset &= s->cluster_offset_mask;
903 if (bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len) != out_len) {
913 static void qcow_flush(BlockDriverState *bs)
915 bdrv_flush(bs->file);
918 static BlockDriverAIOCB *qcow_aio_flush(BlockDriverState *bs,
919 BlockDriverCompletionFunc *cb, void *opaque)
921 return bdrv_aio_flush(bs->file, cb, opaque);
924 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
926 BDRVQcowState *s = bs->opaque;
927 bdi->cluster_size = s->cluster_size;
932 static QEMUOptionParameter qcow_create_options[] = {
934 .name = BLOCK_OPT_SIZE,
936 .help = "Virtual disk size"
939 .name = BLOCK_OPT_BACKING_FILE,
941 .help = "File name of a base image"
944 .name = BLOCK_OPT_ENCRYPT,
946 .help = "Encrypt the image"
951 static BlockDriver bdrv_qcow = {
952 .format_name = "qcow",
953 .instance_size = sizeof(BDRVQcowState),
954 .bdrv_probe = qcow_probe,
955 .bdrv_open = qcow_open,
956 .bdrv_close = qcow_close,
957 .bdrv_create = qcow_create,
958 .bdrv_flush = qcow_flush,
959 .bdrv_is_allocated = qcow_is_allocated,
960 .bdrv_set_key = qcow_set_key,
961 .bdrv_make_empty = qcow_make_empty,
962 .bdrv_aio_readv = qcow_aio_readv,
963 .bdrv_aio_writev = qcow_aio_writev,
964 .bdrv_aio_flush = qcow_aio_flush,
965 .bdrv_write_compressed = qcow_write_compressed,
966 .bdrv_get_info = qcow_get_info,
968 .create_options = qcow_create_options,
971 static void bdrv_qcow_init(void)
973 bdrv_register(&bdrv_qcow);
976 block_init(bdrv_qcow_init);