vl: Tighten parsing of -m argument
[qemu.git] / block / qcow2.c
blobd7805ce9433116f54438da40020c857787894f04
1 /*
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
22 * THE SOFTWARE.
24 #include "qemu-common.h"
25 #include "block_int.h"
26 #include "module.h"
27 #include <zlib.h>
28 #include "aes.h"
29 #include "block/qcow2.h"
30 #include "qemu-error.h"
31 #include "qerror.h"
34 Differences with QCOW:
36 - Support for multiple incremental snapshots.
37 - Memory management by reference counts.
38 - Clusters which have a reference count of one have the bit
39 QCOW_OFLAG_COPIED to optimize write performance.
40 - Size of compressed clusters is stored in sectors to reduce bit usage
41 in the cluster offsets.
42 - Support for storing additional data (such as the VM state) in the
43 snapshots.
44 - If a backing store is used, the cluster size is not constrained
45 (could be backported to QCOW).
46 - L2 tables have always a size of one cluster.
50 typedef struct {
51 uint32_t magic;
52 uint32_t len;
53 } QCowExtension;
54 #define QCOW2_EXT_MAGIC_END 0
55 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
57 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
59 const QCowHeader *cow_header = (const void *)buf;
61 if (buf_size >= sizeof(QCowHeader) &&
62 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
63 be32_to_cpu(cow_header->version) >= QCOW_VERSION)
64 return 100;
65 else
66 return 0;
70 /*
71 * read qcow2 extension and fill bs
72 * start reading from start_offset
73 * finish reading upon magic of value 0 or when end_offset reached
74 * unknown magic is skipped (future extension this version knows nothing about)
75 * return 0 upon success, non-0 otherwise
77 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
78 uint64_t end_offset)
80 QCowExtension ext;
81 uint64_t offset;
83 #ifdef DEBUG_EXT
84 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
85 #endif
86 offset = start_offset;
87 while (offset < end_offset) {
89 #ifdef DEBUG_EXT
90 BDRVQcowState *s = bs->opaque;
91 /* Sanity check */
92 if (offset > s->cluster_size)
93 printf("qcow2_read_extension: suspicious offset %lu\n", offset);
95 printf("attemting to read extended header in offset %lu\n", offset);
96 #endif
98 if (bdrv_pread(bs->file, offset, &ext, sizeof(ext)) != sizeof(ext)) {
99 fprintf(stderr, "qcow2_read_extension: ERROR: "
100 "pread fail from offset %" PRIu64 "\n",
101 offset);
102 return 1;
104 be32_to_cpus(&ext.magic);
105 be32_to_cpus(&ext.len);
106 offset += sizeof(ext);
107 #ifdef DEBUG_EXT
108 printf("ext.magic = 0x%x\n", ext.magic);
109 #endif
110 switch (ext.magic) {
111 case QCOW2_EXT_MAGIC_END:
112 return 0;
114 case QCOW2_EXT_MAGIC_BACKING_FORMAT:
115 if (ext.len >= sizeof(bs->backing_format)) {
116 fprintf(stderr, "ERROR: ext_backing_format: len=%u too large"
117 " (>=%zu)\n",
118 ext.len, sizeof(bs->backing_format));
119 return 2;
121 if (bdrv_pread(bs->file, offset , bs->backing_format,
122 ext.len) != ext.len)
123 return 3;
124 bs->backing_format[ext.len] = '\0';
125 #ifdef DEBUG_EXT
126 printf("Qcow2: Got format extension %s\n", bs->backing_format);
127 #endif
128 offset = ((offset + ext.len + 7) & ~7);
129 break;
131 default:
132 /* unknown magic -- just skip it */
133 offset = ((offset + ext.len + 7) & ~7);
134 break;
138 return 0;
142 static int qcow2_open(BlockDriverState *bs, int flags)
144 BDRVQcowState *s = bs->opaque;
145 int len, i, ret = 0;
146 QCowHeader header;
147 uint64_t ext_end;
148 bool writethrough;
150 ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
151 if (ret < 0) {
152 goto fail;
154 be32_to_cpus(&header.magic);
155 be32_to_cpus(&header.version);
156 be64_to_cpus(&header.backing_file_offset);
157 be32_to_cpus(&header.backing_file_size);
158 be64_to_cpus(&header.size);
159 be32_to_cpus(&header.cluster_bits);
160 be32_to_cpus(&header.crypt_method);
161 be64_to_cpus(&header.l1_table_offset);
162 be32_to_cpus(&header.l1_size);
163 be64_to_cpus(&header.refcount_table_offset);
164 be32_to_cpus(&header.refcount_table_clusters);
165 be64_to_cpus(&header.snapshots_offset);
166 be32_to_cpus(&header.nb_snapshots);
168 if (header.magic != QCOW_MAGIC) {
169 ret = -EINVAL;
170 goto fail;
172 if (header.version != QCOW_VERSION) {
173 char version[64];
174 snprintf(version, sizeof(version), "QCOW version %d", header.version);
175 qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
176 bs->device_name, "qcow2", version);
177 ret = -ENOTSUP;
178 goto fail;
180 if (header.cluster_bits < MIN_CLUSTER_BITS ||
181 header.cluster_bits > MAX_CLUSTER_BITS) {
182 ret = -EINVAL;
183 goto fail;
185 if (header.crypt_method > QCOW_CRYPT_AES) {
186 ret = -EINVAL;
187 goto fail;
189 s->crypt_method_header = header.crypt_method;
190 if (s->crypt_method_header) {
191 bs->encrypted = 1;
193 s->cluster_bits = header.cluster_bits;
194 s->cluster_size = 1 << s->cluster_bits;
195 s->cluster_sectors = 1 << (s->cluster_bits - 9);
196 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
197 s->l2_size = 1 << s->l2_bits;
198 bs->total_sectors = header.size / 512;
199 s->csize_shift = (62 - (s->cluster_bits - 8));
200 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
201 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
202 s->refcount_table_offset = header.refcount_table_offset;
203 s->refcount_table_size =
204 header.refcount_table_clusters << (s->cluster_bits - 3);
206 s->snapshots_offset = header.snapshots_offset;
207 s->nb_snapshots = header.nb_snapshots;
209 /* read the level 1 table */
210 s->l1_size = header.l1_size;
211 s->l1_vm_state_index = size_to_l1(s, header.size);
212 /* the L1 table must contain at least enough entries to put
213 header.size bytes */
214 if (s->l1_size < s->l1_vm_state_index) {
215 ret = -EINVAL;
216 goto fail;
218 s->l1_table_offset = header.l1_table_offset;
219 if (s->l1_size > 0) {
220 s->l1_table = g_malloc0(
221 align_offset(s->l1_size * sizeof(uint64_t), 512));
222 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
223 s->l1_size * sizeof(uint64_t));
224 if (ret < 0) {
225 goto fail;
227 for(i = 0;i < s->l1_size; i++) {
228 be64_to_cpus(&s->l1_table[i]);
232 /* alloc L2 table/refcount block cache */
233 writethrough = ((flags & BDRV_O_CACHE_WB) == 0);
234 s->l2_table_cache = qcow2_cache_create(bs, L2_CACHE_SIZE, writethrough);
235 s->refcount_block_cache = qcow2_cache_create(bs, REFCOUNT_CACHE_SIZE,
236 writethrough);
238 s->cluster_cache = g_malloc(s->cluster_size);
239 /* one more sector for decompressed data alignment */
240 s->cluster_data = qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
241 + 512);
242 s->cluster_cache_offset = -1;
243 s->flags = flags;
245 ret = qcow2_refcount_init(bs);
246 if (ret != 0) {
247 goto fail;
250 QLIST_INIT(&s->cluster_allocs);
252 /* read qcow2 extensions */
253 if (header.backing_file_offset) {
254 ext_end = header.backing_file_offset;
255 } else {
256 ext_end = s->cluster_size;
258 if (qcow2_read_extensions(bs, sizeof(header), ext_end)) {
259 ret = -EINVAL;
260 goto fail;
263 /* read the backing file name */
264 if (header.backing_file_offset != 0) {
265 len = header.backing_file_size;
266 if (len > 1023) {
267 len = 1023;
269 ret = bdrv_pread(bs->file, header.backing_file_offset,
270 bs->backing_file, len);
271 if (ret < 0) {
272 goto fail;
274 bs->backing_file[len] = '\0';
276 if (qcow2_read_snapshots(bs) < 0) {
277 ret = -EINVAL;
278 goto fail;
281 /* Initialise locks */
282 qemu_co_mutex_init(&s->lock);
284 #ifdef DEBUG_ALLOC
286 BdrvCheckResult result = {0};
287 qcow2_check_refcounts(bs, &result);
289 #endif
290 return ret;
292 fail:
293 qcow2_free_snapshots(bs);
294 qcow2_refcount_close(bs);
295 g_free(s->l1_table);
296 if (s->l2_table_cache) {
297 qcow2_cache_destroy(bs, s->l2_table_cache);
299 g_free(s->cluster_cache);
300 qemu_vfree(s->cluster_data);
301 return ret;
304 static int qcow2_set_key(BlockDriverState *bs, const char *key)
306 BDRVQcowState *s = bs->opaque;
307 uint8_t keybuf[16];
308 int len, i;
310 memset(keybuf, 0, 16);
311 len = strlen(key);
312 if (len > 16)
313 len = 16;
314 /* XXX: we could compress the chars to 7 bits to increase
315 entropy */
316 for(i = 0;i < len;i++) {
317 keybuf[i] = key[i];
319 s->crypt_method = s->crypt_method_header;
321 if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
322 return -1;
323 if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
324 return -1;
325 #if 0
326 /* test */
328 uint8_t in[16];
329 uint8_t out[16];
330 uint8_t tmp[16];
331 for(i=0;i<16;i++)
332 in[i] = i;
333 AES_encrypt(in, tmp, &s->aes_encrypt_key);
334 AES_decrypt(tmp, out, &s->aes_decrypt_key);
335 for(i = 0; i < 16; i++)
336 printf(" %02x", tmp[i]);
337 printf("\n");
338 for(i = 0; i < 16; i++)
339 printf(" %02x", out[i]);
340 printf("\n");
342 #endif
343 return 0;
346 static int qcow2_is_allocated(BlockDriverState *bs, int64_t sector_num,
347 int nb_sectors, int *pnum)
349 uint64_t cluster_offset;
350 int ret;
352 *pnum = nb_sectors;
353 /* FIXME We can get errors here, but the bdrv_is_allocated interface can't
354 * pass them on today */
355 ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset);
356 if (ret < 0) {
357 *pnum = 0;
360 return (cluster_offset != 0);
363 /* handle reading after the end of the backing file */
364 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
365 int64_t sector_num, int nb_sectors)
367 int n1;
368 if ((sector_num + nb_sectors) <= bs->total_sectors)
369 return nb_sectors;
370 if (sector_num >= bs->total_sectors)
371 n1 = 0;
372 else
373 n1 = bs->total_sectors - sector_num;
375 qemu_iovec_memset_skip(qiov, 0, 512 * (nb_sectors - n1), 512 * n1);
377 return n1;
380 static int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num,
381 int remaining_sectors, QEMUIOVector *qiov)
383 BDRVQcowState *s = bs->opaque;
384 int index_in_cluster, n1;
385 int ret;
386 int cur_nr_sectors; /* number of sectors in current iteration */
387 uint64_t cluster_offset = 0;
388 uint64_t bytes_done = 0;
389 QEMUIOVector hd_qiov;
390 uint8_t *cluster_data = NULL;
392 qemu_iovec_init(&hd_qiov, qiov->niov);
394 qemu_co_mutex_lock(&s->lock);
396 while (remaining_sectors != 0) {
398 /* prepare next request */
399 cur_nr_sectors = remaining_sectors;
400 if (s->crypt_method) {
401 cur_nr_sectors = MIN(cur_nr_sectors,
402 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
405 ret = qcow2_get_cluster_offset(bs, sector_num << 9,
406 &cur_nr_sectors, &cluster_offset);
407 if (ret < 0) {
408 goto fail;
411 index_in_cluster = sector_num & (s->cluster_sectors - 1);
413 qemu_iovec_reset(&hd_qiov);
414 qemu_iovec_copy(&hd_qiov, qiov, bytes_done,
415 cur_nr_sectors * 512);
417 if (!cluster_offset) {
419 if (bs->backing_hd) {
420 /* read from the base image */
421 n1 = qcow2_backing_read1(bs->backing_hd, &hd_qiov,
422 sector_num, cur_nr_sectors);
423 if (n1 > 0) {
424 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
425 qemu_co_mutex_unlock(&s->lock);
426 ret = bdrv_co_readv(bs->backing_hd, sector_num,
427 n1, &hd_qiov);
428 qemu_co_mutex_lock(&s->lock);
429 if (ret < 0) {
430 goto fail;
433 } else {
434 /* Note: in this case, no need to wait */
435 qemu_iovec_memset(&hd_qiov, 0, 512 * cur_nr_sectors);
437 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
438 /* add AIO support for compressed blocks ? */
439 ret = qcow2_decompress_cluster(bs, cluster_offset);
440 if (ret < 0) {
441 goto fail;
444 qemu_iovec_from_buffer(&hd_qiov,
445 s->cluster_cache + index_in_cluster * 512,
446 512 * cur_nr_sectors);
447 } else {
448 if ((cluster_offset & 511) != 0) {
449 ret = -EIO;
450 goto fail;
453 if (s->crypt_method) {
455 * For encrypted images, read everything into a temporary
456 * contiguous buffer on which the AES functions can work.
458 if (!cluster_data) {
459 cluster_data =
460 qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
463 assert(cur_nr_sectors <=
464 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
465 qemu_iovec_reset(&hd_qiov);
466 qemu_iovec_add(&hd_qiov, cluster_data,
467 512 * cur_nr_sectors);
470 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
471 qemu_co_mutex_unlock(&s->lock);
472 ret = bdrv_co_readv(bs->file,
473 (cluster_offset >> 9) + index_in_cluster,
474 cur_nr_sectors, &hd_qiov);
475 qemu_co_mutex_lock(&s->lock);
476 if (ret < 0) {
477 goto fail;
479 if (s->crypt_method) {
480 qcow2_encrypt_sectors(s, sector_num, cluster_data,
481 cluster_data, cur_nr_sectors, 0, &s->aes_decrypt_key);
482 qemu_iovec_reset(&hd_qiov);
483 qemu_iovec_copy(&hd_qiov, qiov, bytes_done,
484 cur_nr_sectors * 512);
485 qemu_iovec_from_buffer(&hd_qiov, cluster_data,
486 512 * cur_nr_sectors);
490 remaining_sectors -= cur_nr_sectors;
491 sector_num += cur_nr_sectors;
492 bytes_done += cur_nr_sectors * 512;
494 ret = 0;
496 fail:
497 qemu_co_mutex_unlock(&s->lock);
499 qemu_iovec_destroy(&hd_qiov);
500 qemu_vfree(cluster_data);
502 return ret;
505 static void run_dependent_requests(BDRVQcowState *s, QCowL2Meta *m)
507 /* Take the request off the list of running requests */
508 if (m->nb_clusters != 0) {
509 QLIST_REMOVE(m, next_in_flight);
512 /* Restart all dependent requests */
513 if (!qemu_co_queue_empty(&m->dependent_requests)) {
514 qemu_co_mutex_unlock(&s->lock);
515 while(qemu_co_queue_next(&m->dependent_requests));
516 qemu_co_mutex_lock(&s->lock);
520 static int qcow2_co_writev(BlockDriverState *bs,
521 int64_t sector_num,
522 int remaining_sectors,
523 QEMUIOVector *qiov)
525 BDRVQcowState *s = bs->opaque;
526 int index_in_cluster;
527 int n_end;
528 int ret;
529 int cur_nr_sectors; /* number of sectors in current iteration */
530 uint64_t cluster_offset;
531 QEMUIOVector hd_qiov;
532 uint64_t bytes_done = 0;
533 uint8_t *cluster_data = NULL;
534 QCowL2Meta l2meta = {
535 .nb_clusters = 0,
538 qemu_co_queue_init(&l2meta.dependent_requests);
540 qemu_iovec_init(&hd_qiov, qiov->niov);
542 s->cluster_cache_offset = -1; /* disable compressed cache */
544 qemu_co_mutex_lock(&s->lock);
546 while (remaining_sectors != 0) {
548 index_in_cluster = sector_num & (s->cluster_sectors - 1);
549 n_end = index_in_cluster + remaining_sectors;
550 if (s->crypt_method &&
551 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors) {
552 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
555 ret = qcow2_alloc_cluster_offset(bs, sector_num << 9,
556 index_in_cluster, n_end, &cur_nr_sectors, &l2meta);
557 if (ret < 0) {
558 goto fail;
561 cluster_offset = l2meta.cluster_offset;
562 assert((cluster_offset & 511) == 0);
564 qemu_iovec_reset(&hd_qiov);
565 qemu_iovec_copy(&hd_qiov, qiov, bytes_done,
566 cur_nr_sectors * 512);
568 if (s->crypt_method) {
569 if (!cluster_data) {
570 cluster_data = qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS *
571 s->cluster_size);
574 assert(hd_qiov.size <=
575 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
576 qemu_iovec_to_buffer(&hd_qiov, cluster_data);
578 qcow2_encrypt_sectors(s, sector_num, cluster_data,
579 cluster_data, cur_nr_sectors, 1, &s->aes_encrypt_key);
581 qemu_iovec_reset(&hd_qiov);
582 qemu_iovec_add(&hd_qiov, cluster_data,
583 cur_nr_sectors * 512);
586 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
587 qemu_co_mutex_unlock(&s->lock);
588 ret = bdrv_co_writev(bs->file,
589 (cluster_offset >> 9) + index_in_cluster,
590 cur_nr_sectors, &hd_qiov);
591 qemu_co_mutex_lock(&s->lock);
592 if (ret < 0) {
593 goto fail;
596 ret = qcow2_alloc_cluster_link_l2(bs, &l2meta);
597 if (ret < 0) {
598 goto fail;
601 run_dependent_requests(s, &l2meta);
603 remaining_sectors -= cur_nr_sectors;
604 sector_num += cur_nr_sectors;
605 bytes_done += cur_nr_sectors * 512;
607 ret = 0;
609 fail:
610 run_dependent_requests(s, &l2meta);
612 qemu_co_mutex_unlock(&s->lock);
614 qemu_iovec_destroy(&hd_qiov);
615 qemu_vfree(cluster_data);
617 return ret;
620 static void qcow2_close(BlockDriverState *bs)
622 BDRVQcowState *s = bs->opaque;
623 g_free(s->l1_table);
625 qcow2_cache_flush(bs, s->l2_table_cache);
626 qcow2_cache_flush(bs, s->refcount_block_cache);
628 qcow2_cache_destroy(bs, s->l2_table_cache);
629 qcow2_cache_destroy(bs, s->refcount_block_cache);
631 g_free(s->cluster_cache);
632 qemu_vfree(s->cluster_data);
633 qcow2_refcount_close(bs);
636 static void qcow2_invalidate_cache(BlockDriverState *bs)
638 BDRVQcowState *s = bs->opaque;
639 int flags = s->flags;
640 AES_KEY aes_encrypt_key;
641 AES_KEY aes_decrypt_key;
642 uint32_t crypt_method = 0;
645 * Backing files are read-only which makes all of their metadata immutable,
646 * that means we don't have to worry about reopening them here.
649 if (s->crypt_method) {
650 crypt_method = s->crypt_method;
651 memcpy(&aes_encrypt_key, &s->aes_encrypt_key, sizeof(aes_encrypt_key));
652 memcpy(&aes_decrypt_key, &s->aes_decrypt_key, sizeof(aes_decrypt_key));
655 qcow2_close(bs);
657 memset(s, 0, sizeof(BDRVQcowState));
658 qcow2_open(bs, flags);
660 if (crypt_method) {
661 s->crypt_method = crypt_method;
662 memcpy(&s->aes_encrypt_key, &aes_encrypt_key, sizeof(aes_encrypt_key));
663 memcpy(&s->aes_decrypt_key, &aes_decrypt_key, sizeof(aes_decrypt_key));
668 * Updates the variable length parts of the qcow2 header, i.e. the backing file
669 * name and all extensions. qcow2 was not designed to allow such changes, so if
670 * we run out of space (we can only use the first cluster) this function may
671 * fail.
673 * Returns 0 on success, -errno in error cases.
675 static int qcow2_update_ext_header(BlockDriverState *bs,
676 const char *backing_file, const char *backing_fmt)
678 size_t backing_file_len = 0;
679 size_t backing_fmt_len = 0;
680 BDRVQcowState *s = bs->opaque;
681 QCowExtension ext_backing_fmt = {0, 0};
682 int ret;
684 /* Backing file format doesn't make sense without a backing file */
685 if (backing_fmt && !backing_file) {
686 return -EINVAL;
689 /* Prepare the backing file format extension if needed */
690 if (backing_fmt) {
691 ext_backing_fmt.len = cpu_to_be32(strlen(backing_fmt));
692 ext_backing_fmt.magic = cpu_to_be32(QCOW2_EXT_MAGIC_BACKING_FORMAT);
693 backing_fmt_len = ((sizeof(ext_backing_fmt)
694 + strlen(backing_fmt) + 7) & ~7);
697 /* Check if we can fit the new header into the first cluster */
698 if (backing_file) {
699 backing_file_len = strlen(backing_file);
702 size_t header_size = sizeof(QCowHeader) + backing_file_len
703 + backing_fmt_len;
705 if (header_size > s->cluster_size) {
706 return -ENOSPC;
709 /* Rewrite backing file name and qcow2 extensions */
710 size_t ext_size = header_size - sizeof(QCowHeader);
711 uint8_t buf[ext_size];
712 size_t offset = 0;
713 size_t backing_file_offset = 0;
715 if (backing_file) {
716 if (backing_fmt) {
717 int padding = backing_fmt_len -
718 (sizeof(ext_backing_fmt) + strlen(backing_fmt));
720 memcpy(buf + offset, &ext_backing_fmt, sizeof(ext_backing_fmt));
721 offset += sizeof(ext_backing_fmt);
723 memcpy(buf + offset, backing_fmt, strlen(backing_fmt));
724 offset += strlen(backing_fmt);
726 memset(buf + offset, 0, padding);
727 offset += padding;
730 memcpy(buf + offset, backing_file, backing_file_len);
731 backing_file_offset = sizeof(QCowHeader) + offset;
734 ret = bdrv_pwrite_sync(bs->file, sizeof(QCowHeader), buf, ext_size);
735 if (ret < 0) {
736 goto fail;
739 /* Update header fields */
740 uint64_t be_backing_file_offset = cpu_to_be64(backing_file_offset);
741 uint32_t be_backing_file_size = cpu_to_be32(backing_file_len);
743 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, backing_file_offset),
744 &be_backing_file_offset, sizeof(uint64_t));
745 if (ret < 0) {
746 goto fail;
749 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, backing_file_size),
750 &be_backing_file_size, sizeof(uint32_t));
751 if (ret < 0) {
752 goto fail;
755 ret = 0;
756 fail:
757 return ret;
760 static int qcow2_change_backing_file(BlockDriverState *bs,
761 const char *backing_file, const char *backing_fmt)
763 return qcow2_update_ext_header(bs, backing_file, backing_fmt);
766 static int preallocate(BlockDriverState *bs)
768 uint64_t nb_sectors;
769 uint64_t offset;
770 int num;
771 int ret;
772 QCowL2Meta meta;
774 nb_sectors = bdrv_getlength(bs) >> 9;
775 offset = 0;
776 qemu_co_queue_init(&meta.dependent_requests);
777 meta.cluster_offset = 0;
779 while (nb_sectors) {
780 num = MIN(nb_sectors, INT_MAX >> 9);
781 ret = qcow2_alloc_cluster_offset(bs, offset, 0, num, &num, &meta);
782 if (ret < 0) {
783 return ret;
786 ret = qcow2_alloc_cluster_link_l2(bs, &meta);
787 if (ret < 0) {
788 qcow2_free_any_clusters(bs, meta.cluster_offset, meta.nb_clusters);
789 return ret;
792 /* There are no dependent requests, but we need to remove our request
793 * from the list of in-flight requests */
794 run_dependent_requests(bs->opaque, &meta);
796 /* TODO Preallocate data if requested */
798 nb_sectors -= num;
799 offset += num << 9;
803 * It is expected that the image file is large enough to actually contain
804 * all of the allocated clusters (otherwise we get failing reads after
805 * EOF). Extend the image to the last allocated sector.
807 if (meta.cluster_offset != 0) {
808 uint8_t buf[512];
809 memset(buf, 0, 512);
810 ret = bdrv_write(bs->file, (meta.cluster_offset >> 9) + num - 1, buf, 1);
811 if (ret < 0) {
812 return ret;
816 return 0;
819 static int qcow2_create2(const char *filename, int64_t total_size,
820 const char *backing_file, const char *backing_format,
821 int flags, size_t cluster_size, int prealloc,
822 QEMUOptionParameter *options)
824 /* Calulate cluster_bits */
825 int cluster_bits;
826 cluster_bits = ffs(cluster_size) - 1;
827 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
828 (1 << cluster_bits) != cluster_size)
830 error_report(
831 "Cluster size must be a power of two between %d and %dk",
832 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
833 return -EINVAL;
837 * Open the image file and write a minimal qcow2 header.
839 * We keep things simple and start with a zero-sized image. We also
840 * do without refcount blocks or a L1 table for now. We'll fix the
841 * inconsistency later.
843 * We do need a refcount table because growing the refcount table means
844 * allocating two new refcount blocks - the seconds of which would be at
845 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
846 * size for any qcow2 image.
848 BlockDriverState* bs;
849 QCowHeader header;
850 uint8_t* refcount_table;
851 int ret;
853 ret = bdrv_create_file(filename, options);
854 if (ret < 0) {
855 return ret;
858 ret = bdrv_file_open(&bs, filename, BDRV_O_RDWR);
859 if (ret < 0) {
860 return ret;
863 /* Write the header */
864 memset(&header, 0, sizeof(header));
865 header.magic = cpu_to_be32(QCOW_MAGIC);
866 header.version = cpu_to_be32(QCOW_VERSION);
867 header.cluster_bits = cpu_to_be32(cluster_bits);
868 header.size = cpu_to_be64(0);
869 header.l1_table_offset = cpu_to_be64(0);
870 header.l1_size = cpu_to_be32(0);
871 header.refcount_table_offset = cpu_to_be64(cluster_size);
872 header.refcount_table_clusters = cpu_to_be32(1);
874 if (flags & BLOCK_FLAG_ENCRYPT) {
875 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
876 } else {
877 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
880 ret = bdrv_pwrite(bs, 0, &header, sizeof(header));
881 if (ret < 0) {
882 goto out;
885 /* Write an empty refcount table */
886 refcount_table = g_malloc0(cluster_size);
887 ret = bdrv_pwrite(bs, cluster_size, refcount_table, cluster_size);
888 g_free(refcount_table);
890 if (ret < 0) {
891 goto out;
894 bdrv_close(bs);
897 * And now open the image and make it consistent first (i.e. increase the
898 * refcount of the cluster that is occupied by the header and the refcount
899 * table)
901 BlockDriver* drv = bdrv_find_format("qcow2");
902 assert(drv != NULL);
903 ret = bdrv_open(bs, filename,
904 BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_FLUSH, drv);
905 if (ret < 0) {
906 goto out;
909 ret = qcow2_alloc_clusters(bs, 2 * cluster_size);
910 if (ret < 0) {
911 goto out;
913 } else if (ret != 0) {
914 error_report("Huh, first cluster in empty image is already in use?");
915 abort();
918 /* Okay, now that we have a valid image, let's give it the right size */
919 ret = bdrv_truncate(bs, total_size * BDRV_SECTOR_SIZE);
920 if (ret < 0) {
921 goto out;
924 /* Want a backing file? There you go.*/
925 if (backing_file) {
926 ret = bdrv_change_backing_file(bs, backing_file, backing_format);
927 if (ret < 0) {
928 goto out;
932 /* And if we're supposed to preallocate metadata, do that now */
933 if (prealloc) {
934 ret = preallocate(bs);
935 if (ret < 0) {
936 goto out;
940 ret = 0;
941 out:
942 bdrv_delete(bs);
943 return ret;
946 static int qcow2_create(const char *filename, QEMUOptionParameter *options)
948 const char *backing_file = NULL;
949 const char *backing_fmt = NULL;
950 uint64_t sectors = 0;
951 int flags = 0;
952 size_t cluster_size = DEFAULT_CLUSTER_SIZE;
953 int prealloc = 0;
955 /* Read out options */
956 while (options && options->name) {
957 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
958 sectors = options->value.n / 512;
959 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
960 backing_file = options->value.s;
961 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) {
962 backing_fmt = options->value.s;
963 } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
964 flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
965 } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
966 if (options->value.n) {
967 cluster_size = options->value.n;
969 } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) {
970 if (!options->value.s || !strcmp(options->value.s, "off")) {
971 prealloc = 0;
972 } else if (!strcmp(options->value.s, "metadata")) {
973 prealloc = 1;
974 } else {
975 fprintf(stderr, "Invalid preallocation mode: '%s'\n",
976 options->value.s);
977 return -EINVAL;
980 options++;
983 if (backing_file && prealloc) {
984 fprintf(stderr, "Backing file and preallocation cannot be used at "
985 "the same time\n");
986 return -EINVAL;
989 return qcow2_create2(filename, sectors, backing_file, backing_fmt, flags,
990 cluster_size, prealloc, options);
993 static int qcow2_make_empty(BlockDriverState *bs)
995 #if 0
996 /* XXX: not correct */
997 BDRVQcowState *s = bs->opaque;
998 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
999 int ret;
1001 memset(s->l1_table, 0, l1_length);
1002 if (bdrv_pwrite(bs->file, s->l1_table_offset, s->l1_table, l1_length) < 0)
1003 return -1;
1004 ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
1005 if (ret < 0)
1006 return ret;
1008 l2_cache_reset(bs);
1009 #endif
1010 return 0;
1013 static coroutine_fn int qcow2_co_discard(BlockDriverState *bs,
1014 int64_t sector_num, int nb_sectors)
1016 int ret;
1017 BDRVQcowState *s = bs->opaque;
1019 qemu_co_mutex_lock(&s->lock);
1020 ret = qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS,
1021 nb_sectors);
1022 qemu_co_mutex_unlock(&s->lock);
1023 return ret;
1026 static int qcow2_truncate(BlockDriverState *bs, int64_t offset)
1028 BDRVQcowState *s = bs->opaque;
1029 int ret, new_l1_size;
1031 if (offset & 511) {
1032 return -EINVAL;
1035 /* cannot proceed if image has snapshots */
1036 if (s->nb_snapshots) {
1037 return -ENOTSUP;
1040 /* shrinking is currently not supported */
1041 if (offset < bs->total_sectors * 512) {
1042 return -ENOTSUP;
1045 new_l1_size = size_to_l1(s, offset);
1046 ret = qcow2_grow_l1_table(bs, new_l1_size, true);
1047 if (ret < 0) {
1048 return ret;
1051 /* write updated header.size */
1052 offset = cpu_to_be64(offset);
1053 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
1054 &offset, sizeof(uint64_t));
1055 if (ret < 0) {
1056 return ret;
1059 s->l1_vm_state_index = new_l1_size;
1060 return 0;
1063 /* XXX: put compressed sectors first, then all the cluster aligned
1064 tables to avoid losing bytes in alignment */
1065 static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num,
1066 const uint8_t *buf, int nb_sectors)
1068 BDRVQcowState *s = bs->opaque;
1069 z_stream strm;
1070 int ret, out_len;
1071 uint8_t *out_buf;
1072 uint64_t cluster_offset;
1074 if (nb_sectors == 0) {
1075 /* align end of file to a sector boundary to ease reading with
1076 sector based I/Os */
1077 cluster_offset = bdrv_getlength(bs->file);
1078 cluster_offset = (cluster_offset + 511) & ~511;
1079 bdrv_truncate(bs->file, cluster_offset);
1080 return 0;
1083 if (nb_sectors != s->cluster_sectors)
1084 return -EINVAL;
1086 out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1088 /* best compression, small window, no zlib header */
1089 memset(&strm, 0, sizeof(strm));
1090 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1091 Z_DEFLATED, -12,
1092 9, Z_DEFAULT_STRATEGY);
1093 if (ret != 0) {
1094 ret = -EINVAL;
1095 goto fail;
1098 strm.avail_in = s->cluster_size;
1099 strm.next_in = (uint8_t *)buf;
1100 strm.avail_out = s->cluster_size;
1101 strm.next_out = out_buf;
1103 ret = deflate(&strm, Z_FINISH);
1104 if (ret != Z_STREAM_END && ret != Z_OK) {
1105 deflateEnd(&strm);
1106 ret = -EINVAL;
1107 goto fail;
1109 out_len = strm.next_out - out_buf;
1111 deflateEnd(&strm);
1113 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1114 /* could not compress: write normal cluster */
1115 ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
1116 if (ret < 0) {
1117 goto fail;
1119 } else {
1120 cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
1121 sector_num << 9, out_len);
1122 if (!cluster_offset) {
1123 ret = -EIO;
1124 goto fail;
1126 cluster_offset &= s->cluster_offset_mask;
1127 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
1128 ret = bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len);
1129 if (ret < 0) {
1130 goto fail;
1134 ret = 0;
1135 fail:
1136 g_free(out_buf);
1137 return ret;
1140 static int qcow2_co_flush_to_os(BlockDriverState *bs)
1142 BDRVQcowState *s = bs->opaque;
1143 int ret;
1145 qemu_co_mutex_lock(&s->lock);
1146 ret = qcow2_cache_flush(bs, s->l2_table_cache);
1147 if (ret < 0) {
1148 qemu_co_mutex_unlock(&s->lock);
1149 return ret;
1152 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
1153 if (ret < 0) {
1154 qemu_co_mutex_unlock(&s->lock);
1155 return ret;
1157 qemu_co_mutex_unlock(&s->lock);
1159 return 0;
1162 static int qcow2_co_flush_to_disk(BlockDriverState *bs)
1164 return bdrv_co_flush(bs->file);
1167 static int64_t qcow2_vm_state_offset(BDRVQcowState *s)
1169 return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits);
1172 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1174 BDRVQcowState *s = bs->opaque;
1175 bdi->cluster_size = s->cluster_size;
1176 bdi->vm_state_offset = qcow2_vm_state_offset(s);
1177 return 0;
1181 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result)
1183 return qcow2_check_refcounts(bs, result);
1186 #if 0
1187 static void dump_refcounts(BlockDriverState *bs)
1189 BDRVQcowState *s = bs->opaque;
1190 int64_t nb_clusters, k, k1, size;
1191 int refcount;
1193 size = bdrv_getlength(bs->file);
1194 nb_clusters = size_to_clusters(s, size);
1195 for(k = 0; k < nb_clusters;) {
1196 k1 = k;
1197 refcount = get_refcount(bs, k);
1198 k++;
1199 while (k < nb_clusters && get_refcount(bs, k) == refcount)
1200 k++;
1201 printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
1202 k - k1);
1205 #endif
1207 static int qcow2_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1208 int64_t pos, int size)
1210 BDRVQcowState *s = bs->opaque;
1211 int growable = bs->growable;
1212 int ret;
1214 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
1215 bs->growable = 1;
1216 ret = bdrv_pwrite(bs, qcow2_vm_state_offset(s) + pos, buf, size);
1217 bs->growable = growable;
1219 return ret;
1222 static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1223 int64_t pos, int size)
1225 BDRVQcowState *s = bs->opaque;
1226 int growable = bs->growable;
1227 int ret;
1229 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
1230 bs->growable = 1;
1231 ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size);
1232 bs->growable = growable;
1234 return ret;
1237 static QEMUOptionParameter qcow2_create_options[] = {
1239 .name = BLOCK_OPT_SIZE,
1240 .type = OPT_SIZE,
1241 .help = "Virtual disk size"
1244 .name = BLOCK_OPT_BACKING_FILE,
1245 .type = OPT_STRING,
1246 .help = "File name of a base image"
1249 .name = BLOCK_OPT_BACKING_FMT,
1250 .type = OPT_STRING,
1251 .help = "Image format of the base image"
1254 .name = BLOCK_OPT_ENCRYPT,
1255 .type = OPT_FLAG,
1256 .help = "Encrypt the image"
1259 .name = BLOCK_OPT_CLUSTER_SIZE,
1260 .type = OPT_SIZE,
1261 .help = "qcow2 cluster size",
1262 .value = { .n = DEFAULT_CLUSTER_SIZE },
1265 .name = BLOCK_OPT_PREALLOC,
1266 .type = OPT_STRING,
1267 .help = "Preallocation mode (allowed values: off, metadata)"
1269 { NULL }
1272 static BlockDriver bdrv_qcow2 = {
1273 .format_name = "qcow2",
1274 .instance_size = sizeof(BDRVQcowState),
1275 .bdrv_probe = qcow2_probe,
1276 .bdrv_open = qcow2_open,
1277 .bdrv_close = qcow2_close,
1278 .bdrv_create = qcow2_create,
1279 .bdrv_is_allocated = qcow2_is_allocated,
1280 .bdrv_set_key = qcow2_set_key,
1281 .bdrv_make_empty = qcow2_make_empty,
1283 .bdrv_co_readv = qcow2_co_readv,
1284 .bdrv_co_writev = qcow2_co_writev,
1285 .bdrv_co_flush_to_os = qcow2_co_flush_to_os,
1286 .bdrv_co_flush_to_disk = qcow2_co_flush_to_disk,
1288 .bdrv_co_discard = qcow2_co_discard,
1289 .bdrv_truncate = qcow2_truncate,
1290 .bdrv_write_compressed = qcow2_write_compressed,
1292 .bdrv_snapshot_create = qcow2_snapshot_create,
1293 .bdrv_snapshot_goto = qcow2_snapshot_goto,
1294 .bdrv_snapshot_delete = qcow2_snapshot_delete,
1295 .bdrv_snapshot_list = qcow2_snapshot_list,
1296 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
1297 .bdrv_get_info = qcow2_get_info,
1299 .bdrv_save_vmstate = qcow2_save_vmstate,
1300 .bdrv_load_vmstate = qcow2_load_vmstate,
1302 .bdrv_change_backing_file = qcow2_change_backing_file,
1304 .bdrv_invalidate_cache = qcow2_invalidate_cache,
1306 .create_options = qcow2_create_options,
1307 .bdrv_check = qcow2_check,
1310 static void bdrv_qcow2_init(void)
1312 bdrv_register(&bdrv_qcow2);
1315 block_init(bdrv_qcow2_init);