2 * Block driver for the QCOW version 2 format
4 * Copyright (c) 2004-2006 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24 #include "qemu-common.h"
25 #include "block_int.h"
29 #include "block/qcow2.h"
32 Differences with QCOW:
34 - Support for multiple incremental snapshots.
35 - Memory management by reference counts.
36 - Clusters which have a reference count of one have the bit
37 QCOW_OFLAG_COPIED to optimize write performance.
38 - Size of compressed clusters is stored in sectors to reduce bit usage
39 in the cluster offsets.
40 - Support for storing additional data (such as the VM state) in the
42 - If a backing store is used, the cluster size is not constrained
43 (could be backported to QCOW).
44 - L2 tables have always a size of one cluster.
52 #define QCOW_EXT_MAGIC_END 0
53 #define QCOW_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
55 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
57 const QCowHeader *cow_header = (const void *)buf;
59 if (buf_size >= sizeof(QCowHeader) &&
60 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
61 be32_to_cpu(cow_header->version) == QCOW_VERSION)
69 * read qcow2 extension and fill bs
70 * start reading from start_offset
71 * finish reading upon magic of value 0 or when end_offset reached
72 * unknown magic is skipped (future extension this version knows nothing about)
73 * return 0 upon success, non-0 otherwise
75 static int qcow_read_extensions(BlockDriverState *bs, uint64_t start_offset,
82 printf("qcow_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
84 offset = start_offset;
85 while (offset < end_offset) {
89 if (offset > s->cluster_size)
90 printf("qcow_handle_extension: suspicious offset %lu\n", offset);
92 printf("attemting to read extended header in offset %lu\n", offset);
95 if (bdrv_pread(bs->file, offset, &ext, sizeof(ext)) != sizeof(ext)) {
96 fprintf(stderr, "qcow_handle_extension: ERROR: "
97 "pread fail from offset %" PRIu64 "\n",
101 be32_to_cpus(&ext.magic);
102 be32_to_cpus(&ext.len);
103 offset += sizeof(ext);
105 printf("ext.magic = 0x%x\n", ext.magic);
108 case QCOW_EXT_MAGIC_END:
111 case QCOW_EXT_MAGIC_BACKING_FORMAT:
112 if (ext.len >= sizeof(bs->backing_format)) {
113 fprintf(stderr, "ERROR: ext_backing_format: len=%u too large"
115 ext.len, sizeof(bs->backing_format));
118 if (bdrv_pread(bs->file, offset , bs->backing_format,
121 bs->backing_format[ext.len] = '\0';
123 printf("Qcow2: Got format extension %s\n", bs->backing_format);
125 offset = ((offset + ext.len + 7) & ~7);
129 /* unknown magic -- just skip it */
130 offset = ((offset + ext.len + 7) & ~7);
139 static int qcow_open(BlockDriverState *bs, int flags)
141 BDRVQcowState *s = bs->opaque;
146 if (bdrv_pread(bs->file, 0, &header, sizeof(header)) != sizeof(header))
148 be32_to_cpus(&header.magic);
149 be32_to_cpus(&header.version);
150 be64_to_cpus(&header.backing_file_offset);
151 be32_to_cpus(&header.backing_file_size);
152 be64_to_cpus(&header.size);
153 be32_to_cpus(&header.cluster_bits);
154 be32_to_cpus(&header.crypt_method);
155 be64_to_cpus(&header.l1_table_offset);
156 be32_to_cpus(&header.l1_size);
157 be64_to_cpus(&header.refcount_table_offset);
158 be32_to_cpus(&header.refcount_table_clusters);
159 be64_to_cpus(&header.snapshots_offset);
160 be32_to_cpus(&header.nb_snapshots);
162 if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
164 if (header.cluster_bits < MIN_CLUSTER_BITS ||
165 header.cluster_bits > MAX_CLUSTER_BITS)
167 if (header.crypt_method > QCOW_CRYPT_AES)
169 s->crypt_method_header = header.crypt_method;
170 if (s->crypt_method_header)
172 s->cluster_bits = header.cluster_bits;
173 s->cluster_size = 1 << s->cluster_bits;
174 s->cluster_sectors = 1 << (s->cluster_bits - 9);
175 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
176 s->l2_size = 1 << s->l2_bits;
177 bs->total_sectors = header.size / 512;
178 s->csize_shift = (62 - (s->cluster_bits - 8));
179 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
180 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
181 s->refcount_table_offset = header.refcount_table_offset;
182 s->refcount_table_size =
183 header.refcount_table_clusters << (s->cluster_bits - 3);
185 s->snapshots_offset = header.snapshots_offset;
186 s->nb_snapshots = header.nb_snapshots;
188 /* read the level 1 table */
189 s->l1_size = header.l1_size;
190 s->l1_vm_state_index = size_to_l1(s, header.size);
191 /* the L1 table must contain at least enough entries to put
193 if (s->l1_size < s->l1_vm_state_index)
195 s->l1_table_offset = header.l1_table_offset;
196 if (s->l1_size > 0) {
197 s->l1_table = qemu_mallocz(
198 align_offset(s->l1_size * sizeof(uint64_t), 512));
199 if (bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
200 s->l1_size * sizeof(uint64_t))
202 for(i = 0;i < s->l1_size; i++) {
203 be64_to_cpus(&s->l1_table[i]);
207 s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
208 s->cluster_cache = qemu_malloc(s->cluster_size);
209 /* one more sector for decompressed data alignment */
210 s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
212 s->cluster_cache_offset = -1;
214 if (qcow2_refcount_init(bs) < 0)
217 QLIST_INIT(&s->cluster_allocs);
219 /* read qcow2 extensions */
220 if (header.backing_file_offset)
221 ext_end = header.backing_file_offset;
223 ext_end = s->cluster_size;
224 if (qcow_read_extensions(bs, sizeof(header), ext_end))
227 /* read the backing file name */
228 if (header.backing_file_offset != 0) {
229 len = header.backing_file_size;
232 if (bdrv_pread(bs->file, header.backing_file_offset, bs->backing_file, len) != len)
234 bs->backing_file[len] = '\0';
236 if (qcow2_read_snapshots(bs) < 0)
240 qcow2_check_refcounts(bs);
245 qcow2_free_snapshots(bs);
246 qcow2_refcount_close(bs);
247 qemu_free(s->l1_table);
248 qemu_free(s->l2_cache);
249 qemu_free(s->cluster_cache);
250 qemu_free(s->cluster_data);
254 static int qcow_set_key(BlockDriverState *bs, const char *key)
256 BDRVQcowState *s = bs->opaque;
260 memset(keybuf, 0, 16);
264 /* XXX: we could compress the chars to 7 bits to increase
266 for(i = 0;i < len;i++) {
269 s->crypt_method = s->crypt_method_header;
271 if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
273 if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
283 AES_encrypt(in, tmp, &s->aes_encrypt_key);
284 AES_decrypt(tmp, out, &s->aes_decrypt_key);
285 for(i = 0; i < 16; i++)
286 printf(" %02x", tmp[i]);
288 for(i = 0; i < 16; i++)
289 printf(" %02x", out[i]);
296 static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
297 int nb_sectors, int *pnum)
299 uint64_t cluster_offset;
303 /* FIXME We can get errors here, but the bdrv_is_allocated interface can't
304 * pass them on today */
305 ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset);
310 return (cluster_offset != 0);
313 /* handle reading after the end of the backing file */
314 int qcow2_backing_read1(BlockDriverState *bs,
315 int64_t sector_num, uint8_t *buf, int nb_sectors)
318 if ((sector_num + nb_sectors) <= bs->total_sectors)
320 if (sector_num >= bs->total_sectors)
323 n1 = bs->total_sectors - sector_num;
324 memset(buf + n1 * 512, 0, 512 * (nb_sectors - n1));
328 typedef struct QCowAIOCB {
329 BlockDriverAIOCB common;
334 int remaining_sectors;
335 int cur_nr_sectors; /* number of sectors in current iteration */
336 uint64_t cluster_offset;
337 uint8_t *cluster_data;
338 BlockDriverAIOCB *hd_aiocb;
340 QEMUIOVector hd_qiov;
343 QLIST_ENTRY(QCowAIOCB) next_depend;
346 static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
348 QCowAIOCB *acb = container_of(blockacb, QCowAIOCB, common);
350 bdrv_aio_cancel(acb->hd_aiocb);
351 qemu_aio_release(acb);
354 static AIOPool qcow_aio_pool = {
355 .aiocb_size = sizeof(QCowAIOCB),
356 .cancel = qcow_aio_cancel,
359 static void qcow_aio_read_cb(void *opaque, int ret);
360 static void qcow_aio_read_bh(void *opaque)
362 QCowAIOCB *acb = opaque;
363 qemu_bh_delete(acb->bh);
365 qcow_aio_read_cb(opaque, 0);
368 static int qcow_schedule_bh(QEMUBHFunc *cb, QCowAIOCB *acb)
373 acb->bh = qemu_bh_new(cb, acb);
377 qemu_bh_schedule(acb->bh);
382 static void qcow_aio_read_cb(void *opaque, int ret)
384 QCowAIOCB *acb = opaque;
385 BlockDriverState *bs = acb->common.bs;
386 BDRVQcowState *s = bs->opaque;
387 int index_in_cluster, n1;
389 acb->hd_aiocb = NULL;
393 /* post process the read buffer */
394 if (!acb->cluster_offset) {
396 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
399 if (s->crypt_method) {
400 qcow2_encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
401 acb->cur_nr_sectors, 0,
402 &s->aes_decrypt_key);
406 acb->remaining_sectors -= acb->cur_nr_sectors;
407 acb->sector_num += acb->cur_nr_sectors;
408 acb->buf += acb->cur_nr_sectors * 512;
410 if (acb->remaining_sectors == 0) {
411 /* request completed */
416 /* prepare next AIO request */
417 acb->cur_nr_sectors = acb->remaining_sectors;
418 ret = qcow2_get_cluster_offset(bs, acb->sector_num << 9,
419 &acb->cur_nr_sectors, &acb->cluster_offset);
424 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
426 if (!acb->cluster_offset) {
427 if (bs->backing_hd) {
428 /* read from the base image */
429 n1 = qcow2_backing_read1(bs->backing_hd, acb->sector_num,
430 acb->buf, acb->cur_nr_sectors);
432 acb->hd_iov.iov_base = (void *)acb->buf;
433 acb->hd_iov.iov_len = acb->cur_nr_sectors * 512;
434 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
435 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
436 acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num,
437 &acb->hd_qiov, acb->cur_nr_sectors,
438 qcow_aio_read_cb, acb);
439 if (acb->hd_aiocb == NULL)
442 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
447 /* Note: in this case, no need to wait */
448 memset(acb->buf, 0, 512 * acb->cur_nr_sectors);
449 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
453 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
454 /* add AIO support for compressed blocks ? */
455 if (qcow2_decompress_cluster(bs, acb->cluster_offset) < 0)
457 memcpy(acb->buf, s->cluster_cache + index_in_cluster * 512,
458 512 * acb->cur_nr_sectors);
459 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
463 if ((acb->cluster_offset & 511) != 0) {
468 acb->hd_iov.iov_base = (void *)acb->buf;
469 acb->hd_iov.iov_len = acb->cur_nr_sectors * 512;
470 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
471 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
472 acb->hd_aiocb = bdrv_aio_readv(bs->file,
473 (acb->cluster_offset >> 9) + index_in_cluster,
474 &acb->hd_qiov, acb->cur_nr_sectors,
475 qcow_aio_read_cb, acb);
476 if (acb->hd_aiocb == NULL) {
484 if (acb->qiov->niov > 1) {
485 qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size);
486 qemu_vfree(acb->orig_buf);
488 acb->common.cb(acb->common.opaque, ret);
489 qemu_aio_release(acb);
492 static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
493 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
494 BlockDriverCompletionFunc *cb, void *opaque, int is_write)
498 acb = qemu_aio_get(&qcow_aio_pool, bs, cb, opaque);
501 acb->hd_aiocb = NULL;
502 acb->sector_num = sector_num;
504 if (qiov->niov > 1) {
505 acb->buf = acb->orig_buf = qemu_blockalign(bs, qiov->size);
507 qemu_iovec_to_buffer(qiov, acb->buf);
509 acb->buf = (uint8_t *)qiov->iov->iov_base;
511 acb->remaining_sectors = nb_sectors;
512 acb->cur_nr_sectors = 0;
513 acb->cluster_offset = 0;
514 acb->l2meta.nb_clusters = 0;
515 QLIST_INIT(&acb->l2meta.dependent_requests);
519 static BlockDriverAIOCB *qcow_aio_readv(BlockDriverState *bs,
520 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
521 BlockDriverCompletionFunc *cb, void *opaque)
525 acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
529 qcow_aio_read_cb(acb, 0);
533 static void qcow_aio_write_cb(void *opaque, int ret);
535 static void run_dependent_requests(QCowL2Meta *m)
540 /* Take the request off the list of running requests */
541 if (m->nb_clusters != 0) {
542 QLIST_REMOVE(m, next_in_flight);
545 /* Restart all dependent requests */
546 QLIST_FOREACH_SAFE(req, &m->dependent_requests, next_depend, next) {
547 qcow_aio_write_cb(req, 0);
550 /* Empty the list for the next part of the request */
551 QLIST_INIT(&m->dependent_requests);
554 static void qcow_aio_write_cb(void *opaque, int ret)
556 QCowAIOCB *acb = opaque;
557 BlockDriverState *bs = acb->common.bs;
558 BDRVQcowState *s = bs->opaque;
559 int index_in_cluster;
560 const uint8_t *src_buf;
563 acb->hd_aiocb = NULL;
566 ret = qcow2_alloc_cluster_link_l2(bs, &acb->l2meta);
569 run_dependent_requests(&acb->l2meta);
574 acb->remaining_sectors -= acb->cur_nr_sectors;
575 acb->sector_num += acb->cur_nr_sectors;
576 acb->buf += acb->cur_nr_sectors * 512;
578 if (acb->remaining_sectors == 0) {
579 /* request completed */
584 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
585 n_end = index_in_cluster + acb->remaining_sectors;
586 if (s->crypt_method &&
587 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
588 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
590 ret = qcow2_alloc_cluster_offset(bs, acb->sector_num << 9,
591 index_in_cluster, n_end, &acb->cur_nr_sectors, &acb->l2meta);
596 acb->cluster_offset = acb->l2meta.cluster_offset;
598 /* Need to wait for another request? If so, we are done for now. */
599 if (acb->l2meta.nb_clusters == 0 && acb->l2meta.depends_on != NULL) {
600 QLIST_INSERT_HEAD(&acb->l2meta.depends_on->dependent_requests,
605 assert((acb->cluster_offset & 511) == 0);
607 if (s->crypt_method) {
608 if (!acb->cluster_data) {
609 acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS *
612 qcow2_encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
613 acb->cur_nr_sectors, 1, &s->aes_encrypt_key);
614 src_buf = acb->cluster_data;
618 acb->hd_iov.iov_base = (void *)src_buf;
619 acb->hd_iov.iov_len = acb->cur_nr_sectors * 512;
620 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
621 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
622 acb->hd_aiocb = bdrv_aio_writev(bs->file,
623 (acb->cluster_offset >> 9) + index_in_cluster,
624 &acb->hd_qiov, acb->cur_nr_sectors,
625 qcow_aio_write_cb, acb);
626 if (acb->hd_aiocb == NULL) {
634 if (acb->l2meta.nb_clusters != 0) {
635 QLIST_REMOVE(&acb->l2meta, next_in_flight);
638 if (acb->qiov->niov > 1)
639 qemu_vfree(acb->orig_buf);
640 acb->common.cb(acb->common.opaque, ret);
641 qemu_aio_release(acb);
644 static BlockDriverAIOCB *qcow_aio_writev(BlockDriverState *bs,
645 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
646 BlockDriverCompletionFunc *cb, void *opaque)
648 BDRVQcowState *s = bs->opaque;
651 s->cluster_cache_offset = -1; /* disable compressed cache */
653 acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
657 qcow_aio_write_cb(acb, 0);
661 static void qcow_close(BlockDriverState *bs)
663 BDRVQcowState *s = bs->opaque;
664 qemu_free(s->l1_table);
665 qemu_free(s->l2_cache);
666 qemu_free(s->cluster_cache);
667 qemu_free(s->cluster_data);
668 qcow2_refcount_close(bs);
672 * Updates the variable length parts of the qcow2 header, i.e. the backing file
673 * name and all extensions. qcow2 was not designed to allow such changes, so if
674 * we run out of space (we can only use the first cluster) this function may
677 * Returns 0 on success, -errno in error cases.
679 static int qcow2_update_ext_header(BlockDriverState *bs,
680 const char *backing_file, const char *backing_fmt)
682 size_t backing_file_len = 0;
683 size_t backing_fmt_len = 0;
684 BDRVQcowState *s = bs->opaque;
685 QCowExtension ext_backing_fmt = {0, 0};
688 /* Backing file format doesn't make sense without a backing file */
689 if (backing_fmt && !backing_file) {
693 /* Prepare the backing file format extension if needed */
695 ext_backing_fmt.len = cpu_to_be32(strlen(backing_fmt));
696 ext_backing_fmt.magic = cpu_to_be32(QCOW_EXT_MAGIC_BACKING_FORMAT);
697 backing_fmt_len = ((sizeof(ext_backing_fmt)
698 + strlen(backing_fmt) + 7) & ~7);
701 /* Check if we can fit the new header into the first cluster */
703 backing_file_len = strlen(backing_file);
706 size_t header_size = sizeof(QCowHeader) + backing_file_len
709 if (header_size > s->cluster_size) {
713 /* Rewrite backing file name and qcow2 extensions */
714 size_t ext_size = header_size - sizeof(QCowHeader);
715 uint8_t buf[ext_size];
717 size_t backing_file_offset = 0;
721 int padding = backing_fmt_len -
722 (sizeof(ext_backing_fmt) + strlen(backing_fmt));
724 memcpy(buf + offset, &ext_backing_fmt, sizeof(ext_backing_fmt));
725 offset += sizeof(ext_backing_fmt);
727 memcpy(buf + offset, backing_fmt, strlen(backing_fmt));
728 offset += strlen(backing_fmt);
730 memset(buf + offset, 0, padding);
734 memcpy(buf + offset, backing_file, backing_file_len);
735 backing_file_offset = sizeof(QCowHeader) + offset;
738 ret = bdrv_pwrite_sync(bs->file, sizeof(QCowHeader), buf, ext_size);
743 /* Update header fields */
744 uint64_t be_backing_file_offset = cpu_to_be64(backing_file_offset);
745 uint32_t be_backing_file_size = cpu_to_be32(backing_file_len);
747 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, backing_file_offset),
748 &be_backing_file_offset, sizeof(uint64_t));
753 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, backing_file_size),
754 &be_backing_file_size, sizeof(uint32_t));
764 static int qcow2_change_backing_file(BlockDriverState *bs,
765 const char *backing_file, const char *backing_fmt)
767 return qcow2_update_ext_header(bs, backing_file, backing_fmt);
770 static int get_bits_from_size(size_t size)
779 /* Not a power of two */
792 static int preallocate(BlockDriverState *bs)
800 nb_sectors = bdrv_getlength(bs) >> 9;
802 QLIST_INIT(&meta.dependent_requests);
803 meta.cluster_offset = 0;
806 num = MIN(nb_sectors, INT_MAX >> 9);
807 ret = qcow2_alloc_cluster_offset(bs, offset, 0, num, &num, &meta);
812 ret = qcow2_alloc_cluster_link_l2(bs, &meta);
814 qcow2_free_any_clusters(bs, meta.cluster_offset, meta.nb_clusters);
818 /* There are no dependent requests, but we need to remove our request
819 * from the list of in-flight requests */
820 run_dependent_requests(&meta);
822 /* TODO Preallocate data if requested */
829 * It is expected that the image file is large enough to actually contain
830 * all of the allocated clusters (otherwise we get failing reads after
831 * EOF). Extend the image to the last allocated sector.
833 if (meta.cluster_offset != 0) {
836 ret = bdrv_write(bs->file, (meta.cluster_offset >> 9) + num - 1, buf, 1);
845 static int qcow_create2(const char *filename, int64_t total_size,
846 const char *backing_file, const char *backing_format,
847 int flags, size_t cluster_size, int prealloc)
850 int fd, header_size, backing_filename_len, l1_size, i, shift, l2_bits;
851 int ref_clusters, reftable_clusters, backing_format_len = 0;
852 int rounded_ext_bf_len = 0;
854 uint64_t tmp, offset;
855 uint64_t old_ref_clusters;
856 QCowCreateState s1, *s = &s1;
857 QCowExtension ext_bf = {0, 0};
860 memset(s, 0, sizeof(*s));
862 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
865 memset(&header, 0, sizeof(header));
866 header.magic = cpu_to_be32(QCOW_MAGIC);
867 header.version = cpu_to_be32(QCOW_VERSION);
868 header.size = cpu_to_be64(total_size * 512);
869 header_size = sizeof(header);
870 backing_filename_len = 0;
872 if (backing_format) {
873 ext_bf.magic = QCOW_EXT_MAGIC_BACKING_FORMAT;
874 backing_format_len = strlen(backing_format);
875 ext_bf.len = backing_format_len;
876 rounded_ext_bf_len = (sizeof(ext_bf) + ext_bf.len + 7) & ~7;
877 header_size += rounded_ext_bf_len;
879 header.backing_file_offset = cpu_to_be64(header_size);
880 backing_filename_len = strlen(backing_file);
881 header.backing_file_size = cpu_to_be32(backing_filename_len);
882 header_size += backing_filename_len;
886 s->cluster_bits = get_bits_from_size(cluster_size);
887 if (s->cluster_bits < MIN_CLUSTER_BITS ||
888 s->cluster_bits > MAX_CLUSTER_BITS)
890 fprintf(stderr, "Cluster size must be a power of two between "
892 1 << MIN_CLUSTER_BITS,
893 1 << (MAX_CLUSTER_BITS - 10));
896 s->cluster_size = 1 << s->cluster_bits;
898 header.cluster_bits = cpu_to_be32(s->cluster_bits);
899 header_size = (header_size + 7) & ~7;
900 if (flags & BLOCK_FLAG_ENCRYPT) {
901 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
903 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
905 l2_bits = s->cluster_bits - 3;
906 shift = s->cluster_bits + l2_bits;
907 l1_size = (((total_size * 512) + (1LL << shift) - 1) >> shift);
908 offset = align_offset(header_size, s->cluster_size);
909 s->l1_table_offset = offset;
910 header.l1_table_offset = cpu_to_be64(s->l1_table_offset);
911 header.l1_size = cpu_to_be32(l1_size);
912 offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size);
914 /* count how many refcount blocks needed */
916 #define NUM_CLUSTERS(bytes) \
917 (((bytes) + (s->cluster_size) - 1) / (s->cluster_size))
919 ref_clusters = NUM_CLUSTERS(NUM_CLUSTERS(offset) * sizeof(uint16_t));
922 uint64_t image_clusters;
923 old_ref_clusters = ref_clusters;
925 /* Number of clusters used for the refcount table */
926 reftable_clusters = NUM_CLUSTERS(ref_clusters * sizeof(uint64_t));
928 /* Number of clusters that the whole image will have */
929 image_clusters = NUM_CLUSTERS(offset) + ref_clusters
932 /* Number of refcount blocks needed for the image */
933 ref_clusters = NUM_CLUSTERS(image_clusters * sizeof(uint16_t));
935 } while (ref_clusters != old_ref_clusters);
937 s->refcount_table = qemu_mallocz(reftable_clusters * s->cluster_size);
939 s->refcount_table_offset = offset;
940 header.refcount_table_offset = cpu_to_be64(offset);
941 header.refcount_table_clusters = cpu_to_be32(reftable_clusters);
942 offset += (reftable_clusters * s->cluster_size);
943 s->refcount_block_offset = offset;
945 for (i=0; i < ref_clusters; i++) {
946 s->refcount_table[i] = cpu_to_be64(offset);
947 offset += s->cluster_size;
950 s->refcount_block = qemu_mallocz(ref_clusters * s->cluster_size);
952 /* update refcounts */
953 qcow2_create_refcount_update(s, 0, header_size);
954 qcow2_create_refcount_update(s, s->l1_table_offset,
955 l1_size * sizeof(uint64_t));
956 qcow2_create_refcount_update(s, s->refcount_table_offset,
957 reftable_clusters * s->cluster_size);
958 qcow2_create_refcount_update(s, s->refcount_block_offset,
959 ref_clusters * s->cluster_size);
961 /* write all the data */
962 ret = qemu_write_full(fd, &header, sizeof(header));
963 if (ret != sizeof(header)) {
968 if (backing_format_len) {
970 int padding = rounded_ext_bf_len - (ext_bf.len + sizeof(ext_bf));
972 memset(zero, 0, sizeof(zero));
973 cpu_to_be32s(&ext_bf.magic);
974 cpu_to_be32s(&ext_bf.len);
975 ret = qemu_write_full(fd, &ext_bf, sizeof(ext_bf));
976 if (ret != sizeof(ext_bf)) {
980 ret = qemu_write_full(fd, backing_format, backing_format_len);
981 if (ret != backing_format_len) {
986 ret = qemu_write_full(fd, zero, padding);
987 if (ret != padding) {
993 ret = qemu_write_full(fd, backing_file, backing_filename_len);
994 if (ret != backing_filename_len) {
999 lseek(fd, s->l1_table_offset, SEEK_SET);
1001 for(i = 0;i < l1_size; i++) {
1002 ret = qemu_write_full(fd, &tmp, sizeof(tmp));
1003 if (ret != sizeof(tmp)) {
1008 lseek(fd, s->refcount_table_offset, SEEK_SET);
1009 ret = qemu_write_full(fd, s->refcount_table,
1010 reftable_clusters * s->cluster_size);
1011 if (ret != reftable_clusters * s->cluster_size) {
1016 lseek(fd, s->refcount_block_offset, SEEK_SET);
1017 ret = qemu_write_full(fd, s->refcount_block,
1018 ref_clusters * s->cluster_size);
1019 if (ret != ref_clusters * s->cluster_size) {
1026 qemu_free(s->refcount_table);
1027 qemu_free(s->refcount_block);
1030 /* Preallocate metadata */
1031 if (ret == 0 && prealloc) {
1032 BlockDriverState *bs;
1033 BlockDriver *drv = bdrv_find_format("qcow2");
1035 bdrv_open(bs, filename, BDRV_O_CACHE_WB | BDRV_O_RDWR, drv);
1036 ret = preallocate(bs);
1043 static int qcow_create(const char *filename, QEMUOptionParameter *options)
1045 const char *backing_file = NULL;
1046 const char *backing_fmt = NULL;
1047 uint64_t sectors = 0;
1049 size_t cluster_size = 65536;
1052 /* Read out options */
1053 while (options && options->name) {
1054 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1055 sectors = options->value.n / 512;
1056 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
1057 backing_file = options->value.s;
1058 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) {
1059 backing_fmt = options->value.s;
1060 } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
1061 flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
1062 } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
1063 if (options->value.n) {
1064 cluster_size = options->value.n;
1066 } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) {
1067 if (!options->value.s || !strcmp(options->value.s, "off")) {
1069 } else if (!strcmp(options->value.s, "metadata")) {
1072 fprintf(stderr, "Invalid preallocation mode: '%s'\n",
1080 if (backing_file && prealloc) {
1081 fprintf(stderr, "Backing file and preallocation cannot be used at "
1086 return qcow_create2(filename, sectors, backing_file, backing_fmt, flags,
1087 cluster_size, prealloc);
1090 static int qcow_make_empty(BlockDriverState *bs)
1093 /* XXX: not correct */
1094 BDRVQcowState *s = bs->opaque;
1095 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1098 memset(s->l1_table, 0, l1_length);
1099 if (bdrv_pwrite(bs->file, s->l1_table_offset, s->l1_table, l1_length) < 0)
1101 ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
1110 static int qcow2_truncate(BlockDriverState *bs, int64_t offset)
1112 BDRVQcowState *s = bs->opaque;
1113 int ret, new_l1_size;
1119 /* cannot proceed if image has snapshots */
1120 if (s->nb_snapshots) {
1124 /* shrinking is currently not supported */
1125 if (offset < bs->total_sectors * 512) {
1129 new_l1_size = size_to_l1(s, offset);
1130 ret = qcow2_grow_l1_table(bs, new_l1_size);
1135 /* write updated header.size */
1136 offset = cpu_to_be64(offset);
1137 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
1138 &offset, sizeof(uint64_t));
1143 s->l1_vm_state_index = new_l1_size;
1147 /* XXX: put compressed sectors first, then all the cluster aligned
1148 tables to avoid losing bytes in alignment */
1149 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
1150 const uint8_t *buf, int nb_sectors)
1152 BDRVQcowState *s = bs->opaque;
1156 uint64_t cluster_offset;
1158 if (nb_sectors == 0) {
1159 /* align end of file to a sector boundary to ease reading with
1160 sector based I/Os */
1161 cluster_offset = bdrv_getlength(bs->file);
1162 cluster_offset = (cluster_offset + 511) & ~511;
1163 bdrv_truncate(bs->file, cluster_offset);
1167 if (nb_sectors != s->cluster_sectors)
1170 out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1172 /* best compression, small window, no zlib header */
1173 memset(&strm, 0, sizeof(strm));
1174 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1176 9, Z_DEFAULT_STRATEGY);
1182 strm.avail_in = s->cluster_size;
1183 strm.next_in = (uint8_t *)buf;
1184 strm.avail_out = s->cluster_size;
1185 strm.next_out = out_buf;
1187 ret = deflate(&strm, Z_FINISH);
1188 if (ret != Z_STREAM_END && ret != Z_OK) {
1193 out_len = strm.next_out - out_buf;
1197 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1198 /* could not compress: write normal cluster */
1199 bdrv_write(bs, sector_num, buf, s->cluster_sectors);
1201 cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
1202 sector_num << 9, out_len);
1203 if (!cluster_offset)
1205 cluster_offset &= s->cluster_offset_mask;
1206 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
1207 if (bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len) != out_len) {
1217 static void qcow_flush(BlockDriverState *bs)
1219 bdrv_flush(bs->file);
1222 static BlockDriverAIOCB *qcow_aio_flush(BlockDriverState *bs,
1223 BlockDriverCompletionFunc *cb, void *opaque)
1225 return bdrv_aio_flush(bs->file, cb, opaque);
1228 static int64_t qcow_vm_state_offset(BDRVQcowState *s)
1230 return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits);
1233 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1235 BDRVQcowState *s = bs->opaque;
1236 bdi->cluster_size = s->cluster_size;
1237 bdi->vm_state_offset = qcow_vm_state_offset(s);
1242 static int qcow_check(BlockDriverState *bs, BdrvCheckResult *result)
1244 return qcow2_check_refcounts(bs, result);
1248 static void dump_refcounts(BlockDriverState *bs)
1250 BDRVQcowState *s = bs->opaque;
1251 int64_t nb_clusters, k, k1, size;
1254 size = bdrv_getlength(bs->file);
1255 nb_clusters = size_to_clusters(s, size);
1256 for(k = 0; k < nb_clusters;) {
1258 refcount = get_refcount(bs, k);
1260 while (k < nb_clusters && get_refcount(bs, k) == refcount)
1262 printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
1268 static int qcow_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1269 int64_t pos, int size)
1271 BDRVQcowState *s = bs->opaque;
1272 int growable = bs->growable;
1275 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
1277 ret = bdrv_pwrite(bs, qcow_vm_state_offset(s) + pos, buf, size);
1278 bs->growable = growable;
1283 static int qcow_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1284 int64_t pos, int size)
1286 BDRVQcowState *s = bs->opaque;
1287 int growable = bs->growable;
1290 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
1292 ret = bdrv_pread(bs, qcow_vm_state_offset(s) + pos, buf, size);
1293 bs->growable = growable;
1298 static QEMUOptionParameter qcow_create_options[] = {
1300 .name = BLOCK_OPT_SIZE,
1302 .help = "Virtual disk size"
1305 .name = BLOCK_OPT_BACKING_FILE,
1307 .help = "File name of a base image"
1310 .name = BLOCK_OPT_BACKING_FMT,
1312 .help = "Image format of the base image"
1315 .name = BLOCK_OPT_ENCRYPT,
1317 .help = "Encrypt the image"
1320 .name = BLOCK_OPT_CLUSTER_SIZE,
1322 .help = "qcow2 cluster size"
1325 .name = BLOCK_OPT_PREALLOC,
1327 .help = "Preallocation mode (allowed values: off, metadata)"
1332 static BlockDriver bdrv_qcow2 = {
1333 .format_name = "qcow2",
1334 .instance_size = sizeof(BDRVQcowState),
1335 .bdrv_probe = qcow_probe,
1336 .bdrv_open = qcow_open,
1337 .bdrv_close = qcow_close,
1338 .bdrv_create = qcow_create,
1339 .bdrv_flush = qcow_flush,
1340 .bdrv_is_allocated = qcow_is_allocated,
1341 .bdrv_set_key = qcow_set_key,
1342 .bdrv_make_empty = qcow_make_empty,
1344 .bdrv_aio_readv = qcow_aio_readv,
1345 .bdrv_aio_writev = qcow_aio_writev,
1346 .bdrv_aio_flush = qcow_aio_flush,
1348 .bdrv_truncate = qcow2_truncate,
1349 .bdrv_write_compressed = qcow_write_compressed,
1351 .bdrv_snapshot_create = qcow2_snapshot_create,
1352 .bdrv_snapshot_goto = qcow2_snapshot_goto,
1353 .bdrv_snapshot_delete = qcow2_snapshot_delete,
1354 .bdrv_snapshot_list = qcow2_snapshot_list,
1355 .bdrv_get_info = qcow_get_info,
1357 .bdrv_save_vmstate = qcow_save_vmstate,
1358 .bdrv_load_vmstate = qcow_load_vmstate,
1360 .bdrv_change_backing_file = qcow2_change_backing_file,
1362 .create_options = qcow_create_options,
1363 .bdrv_check = qcow_check,
1366 static void bdrv_qcow2_init(void)
1368 bdrv_register(&bdrv_qcow2);
1371 block_init(bdrv_qcow2_init);