xtensa: Avoid calling get_page_addr_code() from helper function
[qemu/ar7.git] / block / qcow2.c
blob2f9e58e0c4f6641d6bb1bad718f1980159b75820
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.
25 #include "qemu/osdep.h"
26 #include "block/block_int.h"
27 #include "block/qdict.h"
28 #include "sysemu/block-backend.h"
29 #include "qemu/module.h"
30 #include <zlib.h>
31 #include "qcow2.h"
32 #include "qemu/error-report.h"
33 #include "qapi/error.h"
34 #include "qapi/qapi-events-block-core.h"
35 #include "qapi/qmp/qdict.h"
36 #include "qapi/qmp/qstring.h"
37 #include "trace.h"
38 #include "qemu/option_int.h"
39 #include "qemu/cutils.h"
40 #include "qemu/bswap.h"
41 #include "qapi/qobject-input-visitor.h"
42 #include "qapi/qapi-visit-block-core.h"
43 #include "crypto.h"
46 Differences with QCOW:
48 - Support for multiple incremental snapshots.
49 - Memory management by reference counts.
50 - Clusters which have a reference count of one have the bit
51 QCOW_OFLAG_COPIED to optimize write performance.
52 - Size of compressed clusters is stored in sectors to reduce bit usage
53 in the cluster offsets.
54 - Support for storing additional data (such as the VM state) in the
55 snapshots.
56 - If a backing store is used, the cluster size is not constrained
57 (could be backported to QCOW).
58 - L2 tables have always a size of one cluster.
62 typedef struct {
63 uint32_t magic;
64 uint32_t len;
65 } QEMU_PACKED QCowExtension;
67 #define QCOW2_EXT_MAGIC_END 0
68 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
69 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
70 #define QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77
71 #define QCOW2_EXT_MAGIC_BITMAPS 0x23852875
73 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
75 const QCowHeader *cow_header = (const void *)buf;
77 if (buf_size >= sizeof(QCowHeader) &&
78 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
79 be32_to_cpu(cow_header->version) >= 2)
80 return 100;
81 else
82 return 0;
86 static ssize_t qcow2_crypto_hdr_read_func(QCryptoBlock *block, size_t offset,
87 uint8_t *buf, size_t buflen,
88 void *opaque, Error **errp)
90 BlockDriverState *bs = opaque;
91 BDRVQcow2State *s = bs->opaque;
92 ssize_t ret;
94 if ((offset + buflen) > s->crypto_header.length) {
95 error_setg(errp, "Request for data outside of extension header");
96 return -1;
99 ret = bdrv_pread(bs->file,
100 s->crypto_header.offset + offset, buf, buflen);
101 if (ret < 0) {
102 error_setg_errno(errp, -ret, "Could not read encryption header");
103 return -1;
105 return ret;
109 static ssize_t qcow2_crypto_hdr_init_func(QCryptoBlock *block, size_t headerlen,
110 void *opaque, Error **errp)
112 BlockDriverState *bs = opaque;
113 BDRVQcow2State *s = bs->opaque;
114 int64_t ret;
115 int64_t clusterlen;
117 ret = qcow2_alloc_clusters(bs, headerlen);
118 if (ret < 0) {
119 error_setg_errno(errp, -ret,
120 "Cannot allocate cluster for LUKS header size %zu",
121 headerlen);
122 return -1;
125 s->crypto_header.length = headerlen;
126 s->crypto_header.offset = ret;
128 /* Zero fill remaining space in cluster so it has predictable
129 * content in case of future spec changes */
130 clusterlen = size_to_clusters(s, headerlen) * s->cluster_size;
131 assert(qcow2_pre_write_overlap_check(bs, 0, ret, clusterlen) == 0);
132 ret = bdrv_pwrite_zeroes(bs->file,
133 ret + headerlen,
134 clusterlen - headerlen, 0);
135 if (ret < 0) {
136 error_setg_errno(errp, -ret, "Could not zero fill encryption header");
137 return -1;
140 return ret;
144 static ssize_t qcow2_crypto_hdr_write_func(QCryptoBlock *block, size_t offset,
145 const uint8_t *buf, size_t buflen,
146 void *opaque, Error **errp)
148 BlockDriverState *bs = opaque;
149 BDRVQcow2State *s = bs->opaque;
150 ssize_t ret;
152 if ((offset + buflen) > s->crypto_header.length) {
153 error_setg(errp, "Request for data outside of extension header");
154 return -1;
157 ret = bdrv_pwrite(bs->file,
158 s->crypto_header.offset + offset, buf, buflen);
159 if (ret < 0) {
160 error_setg_errno(errp, -ret, "Could not read encryption header");
161 return -1;
163 return ret;
168 * read qcow2 extension and fill bs
169 * start reading from start_offset
170 * finish reading upon magic of value 0 or when end_offset reached
171 * unknown magic is skipped (future extension this version knows nothing about)
172 * return 0 upon success, non-0 otherwise
174 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
175 uint64_t end_offset, void **p_feature_table,
176 int flags, bool *need_update_header,
177 Error **errp)
179 BDRVQcow2State *s = bs->opaque;
180 QCowExtension ext;
181 uint64_t offset;
182 int ret;
183 Qcow2BitmapHeaderExt bitmaps_ext;
185 if (need_update_header != NULL) {
186 *need_update_header = false;
189 #ifdef DEBUG_EXT
190 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
191 #endif
192 offset = start_offset;
193 while (offset < end_offset) {
195 #ifdef DEBUG_EXT
196 /* Sanity check */
197 if (offset > s->cluster_size)
198 printf("qcow2_read_extension: suspicious offset %lu\n", offset);
200 printf("attempting to read extended header in offset %lu\n", offset);
201 #endif
203 ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext));
204 if (ret < 0) {
205 error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: "
206 "pread fail from offset %" PRIu64, offset);
207 return 1;
209 be32_to_cpus(&ext.magic);
210 be32_to_cpus(&ext.len);
211 offset += sizeof(ext);
212 #ifdef DEBUG_EXT
213 printf("ext.magic = 0x%x\n", ext.magic);
214 #endif
215 if (offset > end_offset || ext.len > end_offset - offset) {
216 error_setg(errp, "Header extension too large");
217 return -EINVAL;
220 switch (ext.magic) {
221 case QCOW2_EXT_MAGIC_END:
222 return 0;
224 case QCOW2_EXT_MAGIC_BACKING_FORMAT:
225 if (ext.len >= sizeof(bs->backing_format)) {
226 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32
227 " too large (>=%zu)", ext.len,
228 sizeof(bs->backing_format));
229 return 2;
231 ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len);
232 if (ret < 0) {
233 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: "
234 "Could not read format name");
235 return 3;
237 bs->backing_format[ext.len] = '\0';
238 s->image_backing_format = g_strdup(bs->backing_format);
239 #ifdef DEBUG_EXT
240 printf("Qcow2: Got format extension %s\n", bs->backing_format);
241 #endif
242 break;
244 case QCOW2_EXT_MAGIC_FEATURE_TABLE:
245 if (p_feature_table != NULL) {
246 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
247 ret = bdrv_pread(bs->file, offset , feature_table, ext.len);
248 if (ret < 0) {
249 error_setg_errno(errp, -ret, "ERROR: ext_feature_table: "
250 "Could not read table");
251 return ret;
254 *p_feature_table = feature_table;
256 break;
258 case QCOW2_EXT_MAGIC_CRYPTO_HEADER: {
259 unsigned int cflags = 0;
260 if (s->crypt_method_header != QCOW_CRYPT_LUKS) {
261 error_setg(errp, "CRYPTO header extension only "
262 "expected with LUKS encryption method");
263 return -EINVAL;
265 if (ext.len != sizeof(Qcow2CryptoHeaderExtension)) {
266 error_setg(errp, "CRYPTO header extension size %u, "
267 "but expected size %zu", ext.len,
268 sizeof(Qcow2CryptoHeaderExtension));
269 return -EINVAL;
272 ret = bdrv_pread(bs->file, offset, &s->crypto_header, ext.len);
273 if (ret < 0) {
274 error_setg_errno(errp, -ret,
275 "Unable to read CRYPTO header extension");
276 return ret;
278 be64_to_cpus(&s->crypto_header.offset);
279 be64_to_cpus(&s->crypto_header.length);
281 if ((s->crypto_header.offset % s->cluster_size) != 0) {
282 error_setg(errp, "Encryption header offset '%" PRIu64 "' is "
283 "not a multiple of cluster size '%u'",
284 s->crypto_header.offset, s->cluster_size);
285 return -EINVAL;
288 if (flags & BDRV_O_NO_IO) {
289 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
291 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
292 qcow2_crypto_hdr_read_func,
293 bs, cflags, errp);
294 if (!s->crypto) {
295 return -EINVAL;
297 } break;
299 case QCOW2_EXT_MAGIC_BITMAPS:
300 if (ext.len != sizeof(bitmaps_ext)) {
301 error_setg_errno(errp, -ret, "bitmaps_ext: "
302 "Invalid extension length");
303 return -EINVAL;
306 if (!(s->autoclear_features & QCOW2_AUTOCLEAR_BITMAPS)) {
307 if (s->qcow_version < 3) {
308 /* Let's be a bit more specific */
309 warn_report("This qcow2 v2 image contains bitmaps, but "
310 "they may have been modified by a program "
311 "without persistent bitmap support; so now "
312 "they must all be considered inconsistent");
313 } else {
314 warn_report("a program lacking bitmap support "
315 "modified this file, so all bitmaps are now "
316 "considered inconsistent");
318 error_printf("Some clusters may be leaked, "
319 "run 'qemu-img check -r' on the image "
320 "file to fix.");
321 if (need_update_header != NULL) {
322 /* Updating is needed to drop invalid bitmap extension. */
323 *need_update_header = true;
325 break;
328 ret = bdrv_pread(bs->file, offset, &bitmaps_ext, ext.len);
329 if (ret < 0) {
330 error_setg_errno(errp, -ret, "bitmaps_ext: "
331 "Could not read ext header");
332 return ret;
335 if (bitmaps_ext.reserved32 != 0) {
336 error_setg_errno(errp, -ret, "bitmaps_ext: "
337 "Reserved field is not zero");
338 return -EINVAL;
341 be32_to_cpus(&bitmaps_ext.nb_bitmaps);
342 be64_to_cpus(&bitmaps_ext.bitmap_directory_size);
343 be64_to_cpus(&bitmaps_ext.bitmap_directory_offset);
345 if (bitmaps_ext.nb_bitmaps > QCOW2_MAX_BITMAPS) {
346 error_setg(errp,
347 "bitmaps_ext: Image has %" PRIu32 " bitmaps, "
348 "exceeding the QEMU supported maximum of %d",
349 bitmaps_ext.nb_bitmaps, QCOW2_MAX_BITMAPS);
350 return -EINVAL;
353 if (bitmaps_ext.nb_bitmaps == 0) {
354 error_setg(errp, "found bitmaps extension with zero bitmaps");
355 return -EINVAL;
358 if (bitmaps_ext.bitmap_directory_offset & (s->cluster_size - 1)) {
359 error_setg(errp, "bitmaps_ext: "
360 "invalid bitmap directory offset");
361 return -EINVAL;
364 if (bitmaps_ext.bitmap_directory_size >
365 QCOW2_MAX_BITMAP_DIRECTORY_SIZE) {
366 error_setg(errp, "bitmaps_ext: "
367 "bitmap directory size (%" PRIu64 ") exceeds "
368 "the maximum supported size (%d)",
369 bitmaps_ext.bitmap_directory_size,
370 QCOW2_MAX_BITMAP_DIRECTORY_SIZE);
371 return -EINVAL;
374 s->nb_bitmaps = bitmaps_ext.nb_bitmaps;
375 s->bitmap_directory_offset =
376 bitmaps_ext.bitmap_directory_offset;
377 s->bitmap_directory_size =
378 bitmaps_ext.bitmap_directory_size;
380 #ifdef DEBUG_EXT
381 printf("Qcow2: Got bitmaps extension: "
382 "offset=%" PRIu64 " nb_bitmaps=%" PRIu32 "\n",
383 s->bitmap_directory_offset, s->nb_bitmaps);
384 #endif
385 break;
387 default:
388 /* unknown magic - save it in case we need to rewrite the header */
389 /* If you add a new feature, make sure to also update the fast
390 * path of qcow2_make_empty() to deal with it. */
392 Qcow2UnknownHeaderExtension *uext;
394 uext = g_malloc0(sizeof(*uext) + ext.len);
395 uext->magic = ext.magic;
396 uext->len = ext.len;
397 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
399 ret = bdrv_pread(bs->file, offset , uext->data, uext->len);
400 if (ret < 0) {
401 error_setg_errno(errp, -ret, "ERROR: unknown extension: "
402 "Could not read data");
403 return ret;
406 break;
409 offset += ((ext.len + 7) & ~7);
412 return 0;
415 static void cleanup_unknown_header_ext(BlockDriverState *bs)
417 BDRVQcow2State *s = bs->opaque;
418 Qcow2UnknownHeaderExtension *uext, *next;
420 QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
421 QLIST_REMOVE(uext, next);
422 g_free(uext);
426 static void report_unsupported_feature(Error **errp, Qcow2Feature *table,
427 uint64_t mask)
429 char *features = g_strdup("");
430 char *old;
432 while (table && table->name[0] != '\0') {
433 if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
434 if (mask & (1ULL << table->bit)) {
435 old = features;
436 features = g_strdup_printf("%s%s%.46s", old, *old ? ", " : "",
437 table->name);
438 g_free(old);
439 mask &= ~(1ULL << table->bit);
442 table++;
445 if (mask) {
446 old = features;
447 features = g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64,
448 old, *old ? ", " : "", mask);
449 g_free(old);
452 error_setg(errp, "Unsupported qcow2 feature(s): %s", features);
453 g_free(features);
457 * Sets the dirty bit and flushes afterwards if necessary.
459 * The incompatible_features bit is only set if the image file header was
460 * updated successfully. Therefore it is not required to check the return
461 * value of this function.
463 int qcow2_mark_dirty(BlockDriverState *bs)
465 BDRVQcow2State *s = bs->opaque;
466 uint64_t val;
467 int ret;
469 assert(s->qcow_version >= 3);
471 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
472 return 0; /* already dirty */
475 val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY);
476 ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features),
477 &val, sizeof(val));
478 if (ret < 0) {
479 return ret;
481 ret = bdrv_flush(bs->file->bs);
482 if (ret < 0) {
483 return ret;
486 /* Only treat image as dirty if the header was updated successfully */
487 s->incompatible_features |= QCOW2_INCOMPAT_DIRTY;
488 return 0;
492 * Clears the dirty bit and flushes before if necessary. Only call this
493 * function when there are no pending requests, it does not guard against
494 * concurrent requests dirtying the image.
496 static int qcow2_mark_clean(BlockDriverState *bs)
498 BDRVQcow2State *s = bs->opaque;
500 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
501 int ret;
503 s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY;
505 ret = qcow2_flush_caches(bs);
506 if (ret < 0) {
507 return ret;
510 return qcow2_update_header(bs);
512 return 0;
516 * Marks the image as corrupt.
518 int qcow2_mark_corrupt(BlockDriverState *bs)
520 BDRVQcow2State *s = bs->opaque;
522 s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT;
523 return qcow2_update_header(bs);
527 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
528 * before if necessary.
530 int qcow2_mark_consistent(BlockDriverState *bs)
532 BDRVQcow2State *s = bs->opaque;
534 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
535 int ret = qcow2_flush_caches(bs);
536 if (ret < 0) {
537 return ret;
540 s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT;
541 return qcow2_update_header(bs);
543 return 0;
546 static int coroutine_fn qcow2_co_check_locked(BlockDriverState *bs,
547 BdrvCheckResult *result,
548 BdrvCheckMode fix)
550 int ret = qcow2_check_refcounts(bs, result, fix);
551 if (ret < 0) {
552 return ret;
555 if (fix && result->check_errors == 0 && result->corruptions == 0) {
556 ret = qcow2_mark_clean(bs);
557 if (ret < 0) {
558 return ret;
560 return qcow2_mark_consistent(bs);
562 return ret;
565 static int coroutine_fn qcow2_co_check(BlockDriverState *bs,
566 BdrvCheckResult *result,
567 BdrvCheckMode fix)
569 BDRVQcow2State *s = bs->opaque;
570 int ret;
572 qemu_co_mutex_lock(&s->lock);
573 ret = qcow2_co_check_locked(bs, result, fix);
574 qemu_co_mutex_unlock(&s->lock);
575 return ret;
578 int qcow2_validate_table(BlockDriverState *bs, uint64_t offset,
579 uint64_t entries, size_t entry_len,
580 int64_t max_size_bytes, const char *table_name,
581 Error **errp)
583 BDRVQcow2State *s = bs->opaque;
585 if (entries > max_size_bytes / entry_len) {
586 error_setg(errp, "%s too large", table_name);
587 return -EFBIG;
590 /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
591 * because values will be passed to qemu functions taking int64_t. */
592 if ((INT64_MAX - entries * entry_len < offset) ||
593 (offset_into_cluster(s, offset) != 0)) {
594 error_setg(errp, "%s offset invalid", table_name);
595 return -EINVAL;
598 return 0;
601 static QemuOptsList qcow2_runtime_opts = {
602 .name = "qcow2",
603 .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head),
604 .desc = {
606 .name = QCOW2_OPT_LAZY_REFCOUNTS,
607 .type = QEMU_OPT_BOOL,
608 .help = "Postpone refcount updates",
611 .name = QCOW2_OPT_DISCARD_REQUEST,
612 .type = QEMU_OPT_BOOL,
613 .help = "Pass guest discard requests to the layer below",
616 .name = QCOW2_OPT_DISCARD_SNAPSHOT,
617 .type = QEMU_OPT_BOOL,
618 .help = "Generate discard requests when snapshot related space "
619 "is freed",
622 .name = QCOW2_OPT_DISCARD_OTHER,
623 .type = QEMU_OPT_BOOL,
624 .help = "Generate discard requests when other clusters are freed",
627 .name = QCOW2_OPT_OVERLAP,
628 .type = QEMU_OPT_STRING,
629 .help = "Selects which overlap checks to perform from a range of "
630 "templates (none, constant, cached, all)",
633 .name = QCOW2_OPT_OVERLAP_TEMPLATE,
634 .type = QEMU_OPT_STRING,
635 .help = "Selects which overlap checks to perform from a range of "
636 "templates (none, constant, cached, all)",
639 .name = QCOW2_OPT_OVERLAP_MAIN_HEADER,
640 .type = QEMU_OPT_BOOL,
641 .help = "Check for unintended writes into the main qcow2 header",
644 .name = QCOW2_OPT_OVERLAP_ACTIVE_L1,
645 .type = QEMU_OPT_BOOL,
646 .help = "Check for unintended writes into the active L1 table",
649 .name = QCOW2_OPT_OVERLAP_ACTIVE_L2,
650 .type = QEMU_OPT_BOOL,
651 .help = "Check for unintended writes into an active L2 table",
654 .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
655 .type = QEMU_OPT_BOOL,
656 .help = "Check for unintended writes into the refcount table",
659 .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
660 .type = QEMU_OPT_BOOL,
661 .help = "Check for unintended writes into a refcount block",
664 .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
665 .type = QEMU_OPT_BOOL,
666 .help = "Check for unintended writes into the snapshot table",
669 .name = QCOW2_OPT_OVERLAP_INACTIVE_L1,
670 .type = QEMU_OPT_BOOL,
671 .help = "Check for unintended writes into an inactive L1 table",
674 .name = QCOW2_OPT_OVERLAP_INACTIVE_L2,
675 .type = QEMU_OPT_BOOL,
676 .help = "Check for unintended writes into an inactive L2 table",
679 .name = QCOW2_OPT_CACHE_SIZE,
680 .type = QEMU_OPT_SIZE,
681 .help = "Maximum combined metadata (L2 tables and refcount blocks) "
682 "cache size",
685 .name = QCOW2_OPT_L2_CACHE_SIZE,
686 .type = QEMU_OPT_SIZE,
687 .help = "Maximum L2 table cache size",
690 .name = QCOW2_OPT_L2_CACHE_ENTRY_SIZE,
691 .type = QEMU_OPT_SIZE,
692 .help = "Size of each entry in the L2 cache",
695 .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE,
696 .type = QEMU_OPT_SIZE,
697 .help = "Maximum refcount block cache size",
700 .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL,
701 .type = QEMU_OPT_NUMBER,
702 .help = "Clean unused cache entries after this time (in seconds)",
704 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
705 "ID of secret providing qcow2 AES key or LUKS passphrase"),
706 { /* end of list */ }
710 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = {
711 [QCOW2_OL_MAIN_HEADER_BITNR] = QCOW2_OPT_OVERLAP_MAIN_HEADER,
712 [QCOW2_OL_ACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L1,
713 [QCOW2_OL_ACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L2,
714 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
715 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
716 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
717 [QCOW2_OL_INACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L1,
718 [QCOW2_OL_INACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L2,
721 static void cache_clean_timer_cb(void *opaque)
723 BlockDriverState *bs = opaque;
724 BDRVQcow2State *s = bs->opaque;
725 qcow2_cache_clean_unused(s->l2_table_cache);
726 qcow2_cache_clean_unused(s->refcount_block_cache);
727 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
728 (int64_t) s->cache_clean_interval * 1000);
731 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context)
733 BDRVQcow2State *s = bs->opaque;
734 if (s->cache_clean_interval > 0) {
735 s->cache_clean_timer = aio_timer_new(context, QEMU_CLOCK_VIRTUAL,
736 SCALE_MS, cache_clean_timer_cb,
737 bs);
738 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
739 (int64_t) s->cache_clean_interval * 1000);
743 static void cache_clean_timer_del(BlockDriverState *bs)
745 BDRVQcow2State *s = bs->opaque;
746 if (s->cache_clean_timer) {
747 timer_del(s->cache_clean_timer);
748 timer_free(s->cache_clean_timer);
749 s->cache_clean_timer = NULL;
753 static void qcow2_detach_aio_context(BlockDriverState *bs)
755 cache_clean_timer_del(bs);
758 static void qcow2_attach_aio_context(BlockDriverState *bs,
759 AioContext *new_context)
761 cache_clean_timer_init(bs, new_context);
764 static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts,
765 uint64_t *l2_cache_size,
766 uint64_t *l2_cache_entry_size,
767 uint64_t *refcount_cache_size, Error **errp)
769 BDRVQcow2State *s = bs->opaque;
770 uint64_t combined_cache_size;
771 bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set;
772 int min_refcount_cache = MIN_REFCOUNT_CACHE_SIZE * s->cluster_size;
774 combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE);
775 l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE);
776 refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
778 combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0);
779 *l2_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 0);
780 *refcount_cache_size = qemu_opt_get_size(opts,
781 QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0);
783 *l2_cache_entry_size = qemu_opt_get_size(
784 opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE, s->cluster_size);
786 if (combined_cache_size_set) {
787 if (l2_cache_size_set && refcount_cache_size_set) {
788 error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE
789 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set "
790 "the same time");
791 return;
792 } else if (*l2_cache_size > combined_cache_size) {
793 error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed "
794 QCOW2_OPT_CACHE_SIZE);
795 return;
796 } else if (*refcount_cache_size > combined_cache_size) {
797 error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed "
798 QCOW2_OPT_CACHE_SIZE);
799 return;
802 if (l2_cache_size_set) {
803 *refcount_cache_size = combined_cache_size - *l2_cache_size;
804 } else if (refcount_cache_size_set) {
805 *l2_cache_size = combined_cache_size - *refcount_cache_size;
806 } else {
807 uint64_t virtual_disk_size = bs->total_sectors * BDRV_SECTOR_SIZE;
808 uint64_t max_l2_cache = virtual_disk_size / (s->cluster_size / 8);
810 /* Assign as much memory as possible to the L2 cache, and
811 * use the remainder for the refcount cache */
812 if (combined_cache_size >= max_l2_cache + min_refcount_cache) {
813 *l2_cache_size = max_l2_cache;
814 *refcount_cache_size = combined_cache_size - *l2_cache_size;
815 } else {
816 *refcount_cache_size =
817 MIN(combined_cache_size, min_refcount_cache);
818 *l2_cache_size = combined_cache_size - *refcount_cache_size;
821 } else {
822 if (!l2_cache_size_set) {
823 *l2_cache_size = MAX(DEFAULT_L2_CACHE_BYTE_SIZE,
824 (uint64_t)DEFAULT_L2_CACHE_CLUSTERS
825 * s->cluster_size);
827 if (!refcount_cache_size_set) {
828 *refcount_cache_size = min_refcount_cache;
832 if (*l2_cache_entry_size < (1 << MIN_CLUSTER_BITS) ||
833 *l2_cache_entry_size > s->cluster_size ||
834 !is_power_of_2(*l2_cache_entry_size)) {
835 error_setg(errp, "L2 cache entry size must be a power of two "
836 "between %d and the cluster size (%d)",
837 1 << MIN_CLUSTER_BITS, s->cluster_size);
838 return;
842 typedef struct Qcow2ReopenState {
843 Qcow2Cache *l2_table_cache;
844 Qcow2Cache *refcount_block_cache;
845 int l2_slice_size; /* Number of entries in a slice of the L2 table */
846 bool use_lazy_refcounts;
847 int overlap_check;
848 bool discard_passthrough[QCOW2_DISCARD_MAX];
849 uint64_t cache_clean_interval;
850 QCryptoBlockOpenOptions *crypto_opts; /* Disk encryption runtime options */
851 } Qcow2ReopenState;
853 static int qcow2_update_options_prepare(BlockDriverState *bs,
854 Qcow2ReopenState *r,
855 QDict *options, int flags,
856 Error **errp)
858 BDRVQcow2State *s = bs->opaque;
859 QemuOpts *opts = NULL;
860 const char *opt_overlap_check, *opt_overlap_check_template;
861 int overlap_check_template = 0;
862 uint64_t l2_cache_size, l2_cache_entry_size, refcount_cache_size;
863 int i;
864 const char *encryptfmt;
865 QDict *encryptopts = NULL;
866 Error *local_err = NULL;
867 int ret;
869 qdict_extract_subqdict(options, &encryptopts, "encrypt.");
870 encryptfmt = qdict_get_try_str(encryptopts, "format");
872 opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort);
873 qemu_opts_absorb_qdict(opts, options, &local_err);
874 if (local_err) {
875 error_propagate(errp, local_err);
876 ret = -EINVAL;
877 goto fail;
880 /* get L2 table/refcount block cache size from command line options */
881 read_cache_sizes(bs, opts, &l2_cache_size, &l2_cache_entry_size,
882 &refcount_cache_size, &local_err);
883 if (local_err) {
884 error_propagate(errp, local_err);
885 ret = -EINVAL;
886 goto fail;
889 l2_cache_size /= l2_cache_entry_size;
890 if (l2_cache_size < MIN_L2_CACHE_SIZE) {
891 l2_cache_size = MIN_L2_CACHE_SIZE;
893 if (l2_cache_size > INT_MAX) {
894 error_setg(errp, "L2 cache size too big");
895 ret = -EINVAL;
896 goto fail;
899 refcount_cache_size /= s->cluster_size;
900 if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) {
901 refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE;
903 if (refcount_cache_size > INT_MAX) {
904 error_setg(errp, "Refcount cache size too big");
905 ret = -EINVAL;
906 goto fail;
909 /* alloc new L2 table/refcount block cache, flush old one */
910 if (s->l2_table_cache) {
911 ret = qcow2_cache_flush(bs, s->l2_table_cache);
912 if (ret) {
913 error_setg_errno(errp, -ret, "Failed to flush the L2 table cache");
914 goto fail;
918 if (s->refcount_block_cache) {
919 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
920 if (ret) {
921 error_setg_errno(errp, -ret,
922 "Failed to flush the refcount block cache");
923 goto fail;
927 r->l2_slice_size = l2_cache_entry_size / sizeof(uint64_t);
928 r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size,
929 l2_cache_entry_size);
930 r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size,
931 s->cluster_size);
932 if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) {
933 error_setg(errp, "Could not allocate metadata caches");
934 ret = -ENOMEM;
935 goto fail;
938 /* New interval for cache cleanup timer */
939 r->cache_clean_interval =
940 qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL,
941 s->cache_clean_interval);
942 #ifndef CONFIG_LINUX
943 if (r->cache_clean_interval != 0) {
944 error_setg(errp, QCOW2_OPT_CACHE_CLEAN_INTERVAL
945 " not supported on this host");
946 ret = -EINVAL;
947 goto fail;
949 #endif
950 if (r->cache_clean_interval > UINT_MAX) {
951 error_setg(errp, "Cache clean interval too big");
952 ret = -EINVAL;
953 goto fail;
956 /* lazy-refcounts; flush if going from enabled to disabled */
957 r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS,
958 (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS));
959 if (r->use_lazy_refcounts && s->qcow_version < 3) {
960 error_setg(errp, "Lazy refcounts require a qcow2 image with at least "
961 "qemu 1.1 compatibility level");
962 ret = -EINVAL;
963 goto fail;
966 if (s->use_lazy_refcounts && !r->use_lazy_refcounts) {
967 ret = qcow2_mark_clean(bs);
968 if (ret < 0) {
969 error_setg_errno(errp, -ret, "Failed to disable lazy refcounts");
970 goto fail;
974 /* Overlap check options */
975 opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP);
976 opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE);
977 if (opt_overlap_check_template && opt_overlap_check &&
978 strcmp(opt_overlap_check_template, opt_overlap_check))
980 error_setg(errp, "Conflicting values for qcow2 options '"
981 QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
982 "' ('%s')", opt_overlap_check, opt_overlap_check_template);
983 ret = -EINVAL;
984 goto fail;
986 if (!opt_overlap_check) {
987 opt_overlap_check = opt_overlap_check_template ?: "cached";
990 if (!strcmp(opt_overlap_check, "none")) {
991 overlap_check_template = 0;
992 } else if (!strcmp(opt_overlap_check, "constant")) {
993 overlap_check_template = QCOW2_OL_CONSTANT;
994 } else if (!strcmp(opt_overlap_check, "cached")) {
995 overlap_check_template = QCOW2_OL_CACHED;
996 } else if (!strcmp(opt_overlap_check, "all")) {
997 overlap_check_template = QCOW2_OL_ALL;
998 } else {
999 error_setg(errp, "Unsupported value '%s' for qcow2 option "
1000 "'overlap-check'. Allowed are any of the following: "
1001 "none, constant, cached, all", opt_overlap_check);
1002 ret = -EINVAL;
1003 goto fail;
1006 r->overlap_check = 0;
1007 for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) {
1008 /* overlap-check defines a template bitmask, but every flag may be
1009 * overwritten through the associated boolean option */
1010 r->overlap_check |=
1011 qemu_opt_get_bool(opts, overlap_bool_option_names[i],
1012 overlap_check_template & (1 << i)) << i;
1015 r->discard_passthrough[QCOW2_DISCARD_NEVER] = false;
1016 r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true;
1017 r->discard_passthrough[QCOW2_DISCARD_REQUEST] =
1018 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST,
1019 flags & BDRV_O_UNMAP);
1020 r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] =
1021 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true);
1022 r->discard_passthrough[QCOW2_DISCARD_OTHER] =
1023 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false);
1025 switch (s->crypt_method_header) {
1026 case QCOW_CRYPT_NONE:
1027 if (encryptfmt) {
1028 error_setg(errp, "No encryption in image header, but options "
1029 "specified format '%s'", encryptfmt);
1030 ret = -EINVAL;
1031 goto fail;
1033 break;
1035 case QCOW_CRYPT_AES:
1036 if (encryptfmt && !g_str_equal(encryptfmt, "aes")) {
1037 error_setg(errp,
1038 "Header reported 'aes' encryption format but "
1039 "options specify '%s'", encryptfmt);
1040 ret = -EINVAL;
1041 goto fail;
1043 qdict_put_str(encryptopts, "format", "qcow");
1044 r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
1045 break;
1047 case QCOW_CRYPT_LUKS:
1048 if (encryptfmt && !g_str_equal(encryptfmt, "luks")) {
1049 error_setg(errp,
1050 "Header reported 'luks' encryption format but "
1051 "options specify '%s'", encryptfmt);
1052 ret = -EINVAL;
1053 goto fail;
1055 qdict_put_str(encryptopts, "format", "luks");
1056 r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
1057 break;
1059 default:
1060 error_setg(errp, "Unsupported encryption method %d",
1061 s->crypt_method_header);
1062 break;
1064 if (s->crypt_method_header != QCOW_CRYPT_NONE && !r->crypto_opts) {
1065 ret = -EINVAL;
1066 goto fail;
1069 ret = 0;
1070 fail:
1071 qobject_unref(encryptopts);
1072 qemu_opts_del(opts);
1073 opts = NULL;
1074 return ret;
1077 static void qcow2_update_options_commit(BlockDriverState *bs,
1078 Qcow2ReopenState *r)
1080 BDRVQcow2State *s = bs->opaque;
1081 int i;
1083 if (s->l2_table_cache) {
1084 qcow2_cache_destroy(s->l2_table_cache);
1086 if (s->refcount_block_cache) {
1087 qcow2_cache_destroy(s->refcount_block_cache);
1089 s->l2_table_cache = r->l2_table_cache;
1090 s->refcount_block_cache = r->refcount_block_cache;
1091 s->l2_slice_size = r->l2_slice_size;
1093 s->overlap_check = r->overlap_check;
1094 s->use_lazy_refcounts = r->use_lazy_refcounts;
1096 for (i = 0; i < QCOW2_DISCARD_MAX; i++) {
1097 s->discard_passthrough[i] = r->discard_passthrough[i];
1100 if (s->cache_clean_interval != r->cache_clean_interval) {
1101 cache_clean_timer_del(bs);
1102 s->cache_clean_interval = r->cache_clean_interval;
1103 cache_clean_timer_init(bs, bdrv_get_aio_context(bs));
1106 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1107 s->crypto_opts = r->crypto_opts;
1110 static void qcow2_update_options_abort(BlockDriverState *bs,
1111 Qcow2ReopenState *r)
1113 if (r->l2_table_cache) {
1114 qcow2_cache_destroy(r->l2_table_cache);
1116 if (r->refcount_block_cache) {
1117 qcow2_cache_destroy(r->refcount_block_cache);
1119 qapi_free_QCryptoBlockOpenOptions(r->crypto_opts);
1122 static int qcow2_update_options(BlockDriverState *bs, QDict *options,
1123 int flags, Error **errp)
1125 Qcow2ReopenState r = {};
1126 int ret;
1128 ret = qcow2_update_options_prepare(bs, &r, options, flags, errp);
1129 if (ret >= 0) {
1130 qcow2_update_options_commit(bs, &r);
1131 } else {
1132 qcow2_update_options_abort(bs, &r);
1135 return ret;
1138 /* Called with s->lock held. */
1139 static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options,
1140 int flags, Error **errp)
1142 BDRVQcow2State *s = bs->opaque;
1143 unsigned int len, i;
1144 int ret = 0;
1145 QCowHeader header;
1146 Error *local_err = NULL;
1147 uint64_t ext_end;
1148 uint64_t l1_vm_state_index;
1149 bool update_header = false;
1150 bool header_updated = false;
1152 ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
1153 if (ret < 0) {
1154 error_setg_errno(errp, -ret, "Could not read qcow2 header");
1155 goto fail;
1157 be32_to_cpus(&header.magic);
1158 be32_to_cpus(&header.version);
1159 be64_to_cpus(&header.backing_file_offset);
1160 be32_to_cpus(&header.backing_file_size);
1161 be64_to_cpus(&header.size);
1162 be32_to_cpus(&header.cluster_bits);
1163 be32_to_cpus(&header.crypt_method);
1164 be64_to_cpus(&header.l1_table_offset);
1165 be32_to_cpus(&header.l1_size);
1166 be64_to_cpus(&header.refcount_table_offset);
1167 be32_to_cpus(&header.refcount_table_clusters);
1168 be64_to_cpus(&header.snapshots_offset);
1169 be32_to_cpus(&header.nb_snapshots);
1171 if (header.magic != QCOW_MAGIC) {
1172 error_setg(errp, "Image is not in qcow2 format");
1173 ret = -EINVAL;
1174 goto fail;
1176 if (header.version < 2 || header.version > 3) {
1177 error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version);
1178 ret = -ENOTSUP;
1179 goto fail;
1182 s->qcow_version = header.version;
1184 /* Initialise cluster size */
1185 if (header.cluster_bits < MIN_CLUSTER_BITS ||
1186 header.cluster_bits > MAX_CLUSTER_BITS) {
1187 error_setg(errp, "Unsupported cluster size: 2^%" PRIu32,
1188 header.cluster_bits);
1189 ret = -EINVAL;
1190 goto fail;
1193 s->cluster_bits = header.cluster_bits;
1194 s->cluster_size = 1 << s->cluster_bits;
1195 s->cluster_sectors = 1 << (s->cluster_bits - BDRV_SECTOR_BITS);
1197 /* Initialise version 3 header fields */
1198 if (header.version == 2) {
1199 header.incompatible_features = 0;
1200 header.compatible_features = 0;
1201 header.autoclear_features = 0;
1202 header.refcount_order = 4;
1203 header.header_length = 72;
1204 } else {
1205 be64_to_cpus(&header.incompatible_features);
1206 be64_to_cpus(&header.compatible_features);
1207 be64_to_cpus(&header.autoclear_features);
1208 be32_to_cpus(&header.refcount_order);
1209 be32_to_cpus(&header.header_length);
1211 if (header.header_length < 104) {
1212 error_setg(errp, "qcow2 header too short");
1213 ret = -EINVAL;
1214 goto fail;
1218 if (header.header_length > s->cluster_size) {
1219 error_setg(errp, "qcow2 header exceeds cluster size");
1220 ret = -EINVAL;
1221 goto fail;
1224 if (header.header_length > sizeof(header)) {
1225 s->unknown_header_fields_size = header.header_length - sizeof(header);
1226 s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
1227 ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields,
1228 s->unknown_header_fields_size);
1229 if (ret < 0) {
1230 error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
1231 "fields");
1232 goto fail;
1236 if (header.backing_file_offset > s->cluster_size) {
1237 error_setg(errp, "Invalid backing file offset");
1238 ret = -EINVAL;
1239 goto fail;
1242 if (header.backing_file_offset) {
1243 ext_end = header.backing_file_offset;
1244 } else {
1245 ext_end = 1 << header.cluster_bits;
1248 /* Handle feature bits */
1249 s->incompatible_features = header.incompatible_features;
1250 s->compatible_features = header.compatible_features;
1251 s->autoclear_features = header.autoclear_features;
1253 if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
1254 void *feature_table = NULL;
1255 qcow2_read_extensions(bs, header.header_length, ext_end,
1256 &feature_table, flags, NULL, NULL);
1257 report_unsupported_feature(errp, feature_table,
1258 s->incompatible_features &
1259 ~QCOW2_INCOMPAT_MASK);
1260 ret = -ENOTSUP;
1261 g_free(feature_table);
1262 goto fail;
1265 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
1266 /* Corrupt images may not be written to unless they are being repaired
1268 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
1269 error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
1270 "read/write");
1271 ret = -EACCES;
1272 goto fail;
1276 /* Check support for various header values */
1277 if (header.refcount_order > 6) {
1278 error_setg(errp, "Reference count entry width too large; may not "
1279 "exceed 64 bits");
1280 ret = -EINVAL;
1281 goto fail;
1283 s->refcount_order = header.refcount_order;
1284 s->refcount_bits = 1 << s->refcount_order;
1285 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
1286 s->refcount_max += s->refcount_max - 1;
1288 s->crypt_method_header = header.crypt_method;
1289 if (s->crypt_method_header) {
1290 if (bdrv_uses_whitelist() &&
1291 s->crypt_method_header == QCOW_CRYPT_AES) {
1292 error_setg(errp,
1293 "Use of AES-CBC encrypted qcow2 images is no longer "
1294 "supported in system emulators");
1295 error_append_hint(errp,
1296 "You can use 'qemu-img convert' to convert your "
1297 "image to an alternative supported format, such "
1298 "as unencrypted qcow2, or raw with the LUKS "
1299 "format instead.\n");
1300 ret = -ENOSYS;
1301 goto fail;
1304 if (s->crypt_method_header == QCOW_CRYPT_AES) {
1305 s->crypt_physical_offset = false;
1306 } else {
1307 /* Assuming LUKS and any future crypt methods we
1308 * add will all use physical offsets, due to the
1309 * fact that the alternative is insecure... */
1310 s->crypt_physical_offset = true;
1313 bs->encrypted = true;
1316 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
1317 s->l2_size = 1 << s->l2_bits;
1318 /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1319 s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
1320 s->refcount_block_size = 1 << s->refcount_block_bits;
1321 bs->total_sectors = header.size / 512;
1322 s->csize_shift = (62 - (s->cluster_bits - 8));
1323 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
1324 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
1326 s->refcount_table_offset = header.refcount_table_offset;
1327 s->refcount_table_size =
1328 header.refcount_table_clusters << (s->cluster_bits - 3);
1330 if (header.refcount_table_clusters == 0 && !(flags & BDRV_O_CHECK)) {
1331 error_setg(errp, "Image does not contain a reference count table");
1332 ret = -EINVAL;
1333 goto fail;
1336 ret = qcow2_validate_table(bs, s->refcount_table_offset,
1337 header.refcount_table_clusters,
1338 s->cluster_size, QCOW_MAX_REFTABLE_SIZE,
1339 "Reference count table", errp);
1340 if (ret < 0) {
1341 goto fail;
1344 /* The total size in bytes of the snapshot table is checked in
1345 * qcow2_read_snapshots() because the size of each snapshot is
1346 * variable and we don't know it yet.
1347 * Here we only check the offset and number of snapshots. */
1348 ret = qcow2_validate_table(bs, header.snapshots_offset,
1349 header.nb_snapshots,
1350 sizeof(QCowSnapshotHeader),
1351 sizeof(QCowSnapshotHeader) * QCOW_MAX_SNAPSHOTS,
1352 "Snapshot table", errp);
1353 if (ret < 0) {
1354 goto fail;
1357 /* read the level 1 table */
1358 ret = qcow2_validate_table(bs, header.l1_table_offset,
1359 header.l1_size, sizeof(uint64_t),
1360 QCOW_MAX_L1_SIZE, "Active L1 table", errp);
1361 if (ret < 0) {
1362 goto fail;
1364 s->l1_size = header.l1_size;
1365 s->l1_table_offset = header.l1_table_offset;
1367 l1_vm_state_index = size_to_l1(s, header.size);
1368 if (l1_vm_state_index > INT_MAX) {
1369 error_setg(errp, "Image is too big");
1370 ret = -EFBIG;
1371 goto fail;
1373 s->l1_vm_state_index = l1_vm_state_index;
1375 /* the L1 table must contain at least enough entries to put
1376 header.size bytes */
1377 if (s->l1_size < s->l1_vm_state_index) {
1378 error_setg(errp, "L1 table is too small");
1379 ret = -EINVAL;
1380 goto fail;
1383 if (s->l1_size > 0) {
1384 s->l1_table = qemu_try_blockalign(bs->file->bs,
1385 ROUND_UP(s->l1_size * sizeof(uint64_t), 512));
1386 if (s->l1_table == NULL) {
1387 error_setg(errp, "Could not allocate L1 table");
1388 ret = -ENOMEM;
1389 goto fail;
1391 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
1392 s->l1_size * sizeof(uint64_t));
1393 if (ret < 0) {
1394 error_setg_errno(errp, -ret, "Could not read L1 table");
1395 goto fail;
1397 for(i = 0;i < s->l1_size; i++) {
1398 be64_to_cpus(&s->l1_table[i]);
1402 /* Parse driver-specific options */
1403 ret = qcow2_update_options(bs, options, flags, errp);
1404 if (ret < 0) {
1405 goto fail;
1408 s->cluster_cache_offset = -1;
1409 s->flags = flags;
1411 ret = qcow2_refcount_init(bs);
1412 if (ret != 0) {
1413 error_setg_errno(errp, -ret, "Could not initialize refcount handling");
1414 goto fail;
1417 QLIST_INIT(&s->cluster_allocs);
1418 QTAILQ_INIT(&s->discards);
1420 /* read qcow2 extensions */
1421 if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
1422 flags, &update_header, &local_err)) {
1423 error_propagate(errp, local_err);
1424 ret = -EINVAL;
1425 goto fail;
1428 /* qcow2_read_extension may have set up the crypto context
1429 * if the crypt method needs a header region, some methods
1430 * don't need header extensions, so must check here
1432 if (s->crypt_method_header && !s->crypto) {
1433 if (s->crypt_method_header == QCOW_CRYPT_AES) {
1434 unsigned int cflags = 0;
1435 if (flags & BDRV_O_NO_IO) {
1436 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
1438 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
1439 NULL, NULL, cflags, errp);
1440 if (!s->crypto) {
1441 ret = -EINVAL;
1442 goto fail;
1444 } else if (!(flags & BDRV_O_NO_IO)) {
1445 error_setg(errp, "Missing CRYPTO header for crypt method %d",
1446 s->crypt_method_header);
1447 ret = -EINVAL;
1448 goto fail;
1452 /* read the backing file name */
1453 if (header.backing_file_offset != 0) {
1454 len = header.backing_file_size;
1455 if (len > MIN(1023, s->cluster_size - header.backing_file_offset) ||
1456 len >= sizeof(bs->backing_file)) {
1457 error_setg(errp, "Backing file name too long");
1458 ret = -EINVAL;
1459 goto fail;
1461 ret = bdrv_pread(bs->file, header.backing_file_offset,
1462 bs->backing_file, len);
1463 if (ret < 0) {
1464 error_setg_errno(errp, -ret, "Could not read backing file name");
1465 goto fail;
1467 bs->backing_file[len] = '\0';
1468 s->image_backing_file = g_strdup(bs->backing_file);
1471 /* Internal snapshots */
1472 s->snapshots_offset = header.snapshots_offset;
1473 s->nb_snapshots = header.nb_snapshots;
1475 ret = qcow2_read_snapshots(bs);
1476 if (ret < 0) {
1477 error_setg_errno(errp, -ret, "Could not read snapshots");
1478 goto fail;
1481 /* Clear unknown autoclear feature bits */
1482 update_header |= s->autoclear_features & ~QCOW2_AUTOCLEAR_MASK;
1483 update_header =
1484 update_header && !bs->read_only && !(flags & BDRV_O_INACTIVE);
1485 if (update_header) {
1486 s->autoclear_features &= QCOW2_AUTOCLEAR_MASK;
1489 if (s->dirty_bitmaps_loaded) {
1490 /* It's some kind of reopen. There are no known cases where we need to
1491 * reload bitmaps in such a situation, so it's safer to skip them.
1493 * Moreover, if we have some readonly bitmaps and we are reopening for
1494 * rw we should reopen bitmaps correspondingly.
1496 if (bdrv_has_readonly_bitmaps(bs) &&
1497 !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE))
1499 qcow2_reopen_bitmaps_rw_hint(bs, &header_updated, &local_err);
1501 } else {
1502 header_updated = qcow2_load_dirty_bitmaps(bs, &local_err);
1503 s->dirty_bitmaps_loaded = true;
1505 update_header = update_header && !header_updated;
1506 if (local_err != NULL) {
1507 error_propagate(errp, local_err);
1508 ret = -EINVAL;
1509 goto fail;
1512 if (update_header) {
1513 ret = qcow2_update_header(bs);
1514 if (ret < 0) {
1515 error_setg_errno(errp, -ret, "Could not update qcow2 header");
1516 goto fail;
1520 bs->supported_zero_flags = header.version >= 3 ? BDRV_REQ_MAY_UNMAP : 0;
1522 /* Repair image if dirty */
1523 if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only &&
1524 (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
1525 BdrvCheckResult result = {0};
1527 ret = qcow2_co_check_locked(bs, &result,
1528 BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
1529 if (ret < 0 || result.check_errors) {
1530 if (ret >= 0) {
1531 ret = -EIO;
1533 error_setg_errno(errp, -ret, "Could not repair dirty image");
1534 goto fail;
1538 #ifdef DEBUG_ALLOC
1540 BdrvCheckResult result = {0};
1541 qcow2_check_refcounts(bs, &result, 0);
1543 #endif
1544 return ret;
1546 fail:
1547 g_free(s->unknown_header_fields);
1548 cleanup_unknown_header_ext(bs);
1549 qcow2_free_snapshots(bs);
1550 qcow2_refcount_close(bs);
1551 qemu_vfree(s->l1_table);
1552 /* else pre-write overlap checks in cache_destroy may crash */
1553 s->l1_table = NULL;
1554 cache_clean_timer_del(bs);
1555 if (s->l2_table_cache) {
1556 qcow2_cache_destroy(s->l2_table_cache);
1558 if (s->refcount_block_cache) {
1559 qcow2_cache_destroy(s->refcount_block_cache);
1561 qcrypto_block_free(s->crypto);
1562 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1563 return ret;
1566 typedef struct QCow2OpenCo {
1567 BlockDriverState *bs;
1568 QDict *options;
1569 int flags;
1570 Error **errp;
1571 int ret;
1572 } QCow2OpenCo;
1574 static void coroutine_fn qcow2_open_entry(void *opaque)
1576 QCow2OpenCo *qoc = opaque;
1577 BDRVQcow2State *s = qoc->bs->opaque;
1579 qemu_co_mutex_lock(&s->lock);
1580 qoc->ret = qcow2_do_open(qoc->bs, qoc->options, qoc->flags, qoc->errp);
1581 qemu_co_mutex_unlock(&s->lock);
1584 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
1585 Error **errp)
1587 BDRVQcow2State *s = bs->opaque;
1588 QCow2OpenCo qoc = {
1589 .bs = bs,
1590 .options = options,
1591 .flags = flags,
1592 .errp = errp,
1593 .ret = -EINPROGRESS
1596 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
1597 false, errp);
1598 if (!bs->file) {
1599 return -EINVAL;
1602 /* Initialise locks */
1603 qemu_co_mutex_init(&s->lock);
1605 if (qemu_in_coroutine()) {
1606 /* From bdrv_co_create. */
1607 qcow2_open_entry(&qoc);
1608 } else {
1609 qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry, &qoc));
1610 BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS);
1612 return qoc.ret;
1615 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1617 BDRVQcow2State *s = bs->opaque;
1619 if (bs->encrypted) {
1620 /* Encryption works on a sector granularity */
1621 bs->bl.request_alignment = BDRV_SECTOR_SIZE;
1623 bs->bl.pwrite_zeroes_alignment = s->cluster_size;
1624 bs->bl.pdiscard_alignment = s->cluster_size;
1627 static int qcow2_reopen_prepare(BDRVReopenState *state,
1628 BlockReopenQueue *queue, Error **errp)
1630 Qcow2ReopenState *r;
1631 int ret;
1633 r = g_new0(Qcow2ReopenState, 1);
1634 state->opaque = r;
1636 ret = qcow2_update_options_prepare(state->bs, r, state->options,
1637 state->flags, errp);
1638 if (ret < 0) {
1639 goto fail;
1642 /* We need to write out any unwritten data if we reopen read-only. */
1643 if ((state->flags & BDRV_O_RDWR) == 0) {
1644 ret = qcow2_reopen_bitmaps_ro(state->bs, errp);
1645 if (ret < 0) {
1646 goto fail;
1649 ret = bdrv_flush(state->bs);
1650 if (ret < 0) {
1651 goto fail;
1654 ret = qcow2_mark_clean(state->bs);
1655 if (ret < 0) {
1656 goto fail;
1660 return 0;
1662 fail:
1663 qcow2_update_options_abort(state->bs, r);
1664 g_free(r);
1665 return ret;
1668 static void qcow2_reopen_commit(BDRVReopenState *state)
1670 qcow2_update_options_commit(state->bs, state->opaque);
1671 g_free(state->opaque);
1674 static void qcow2_reopen_abort(BDRVReopenState *state)
1676 qcow2_update_options_abort(state->bs, state->opaque);
1677 g_free(state->opaque);
1680 static void qcow2_join_options(QDict *options, QDict *old_options)
1682 bool has_new_overlap_template =
1683 qdict_haskey(options, QCOW2_OPT_OVERLAP) ||
1684 qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE);
1685 bool has_new_total_cache_size =
1686 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE);
1687 bool has_all_cache_options;
1689 /* New overlap template overrides all old overlap options */
1690 if (has_new_overlap_template) {
1691 qdict_del(old_options, QCOW2_OPT_OVERLAP);
1692 qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE);
1693 qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER);
1694 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1);
1695 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2);
1696 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE);
1697 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK);
1698 qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE);
1699 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1);
1700 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2);
1703 /* New total cache size overrides all old options */
1704 if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) {
1705 qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE);
1706 qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1709 qdict_join(options, old_options, false);
1712 * If after merging all cache size options are set, an old total size is
1713 * overwritten. Do keep all options, however, if all three are new. The
1714 * resulting error message is what we want to happen.
1716 has_all_cache_options =
1717 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) ||
1718 qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) ||
1719 qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1721 if (has_all_cache_options && !has_new_total_cache_size) {
1722 qdict_del(options, QCOW2_OPT_CACHE_SIZE);
1726 static int coroutine_fn qcow2_co_block_status(BlockDriverState *bs,
1727 bool want_zero,
1728 int64_t offset, int64_t count,
1729 int64_t *pnum, int64_t *map,
1730 BlockDriverState **file)
1732 BDRVQcow2State *s = bs->opaque;
1733 uint64_t cluster_offset;
1734 int index_in_cluster, ret;
1735 unsigned int bytes;
1736 int status = 0;
1738 bytes = MIN(INT_MAX, count);
1739 qemu_co_mutex_lock(&s->lock);
1740 ret = qcow2_get_cluster_offset(bs, offset, &bytes, &cluster_offset);
1741 qemu_co_mutex_unlock(&s->lock);
1742 if (ret < 0) {
1743 return ret;
1746 *pnum = bytes;
1748 if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED &&
1749 !s->crypto) {
1750 index_in_cluster = offset & (s->cluster_size - 1);
1751 *map = cluster_offset | index_in_cluster;
1752 *file = bs->file->bs;
1753 status |= BDRV_BLOCK_OFFSET_VALID;
1755 if (ret == QCOW2_CLUSTER_ZERO_PLAIN || ret == QCOW2_CLUSTER_ZERO_ALLOC) {
1756 status |= BDRV_BLOCK_ZERO;
1757 } else if (ret != QCOW2_CLUSTER_UNALLOCATED) {
1758 status |= BDRV_BLOCK_DATA;
1760 return status;
1763 static coroutine_fn int qcow2_handle_l2meta(BlockDriverState *bs,
1764 QCowL2Meta **pl2meta,
1765 bool link_l2)
1767 int ret = 0;
1768 QCowL2Meta *l2meta = *pl2meta;
1770 while (l2meta != NULL) {
1771 QCowL2Meta *next;
1773 if (link_l2) {
1774 ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
1775 if (ret) {
1776 goto out;
1778 } else {
1779 qcow2_alloc_cluster_abort(bs, l2meta);
1782 /* Take the request off the list of running requests */
1783 if (l2meta->nb_clusters != 0) {
1784 QLIST_REMOVE(l2meta, next_in_flight);
1787 qemu_co_queue_restart_all(&l2meta->dependent_requests);
1789 next = l2meta->next;
1790 g_free(l2meta);
1791 l2meta = next;
1793 out:
1794 *pl2meta = l2meta;
1795 return ret;
1798 static coroutine_fn int qcow2_co_preadv(BlockDriverState *bs, uint64_t offset,
1799 uint64_t bytes, QEMUIOVector *qiov,
1800 int flags)
1802 BDRVQcow2State *s = bs->opaque;
1803 int offset_in_cluster;
1804 int ret;
1805 unsigned int cur_bytes; /* number of bytes in current iteration */
1806 uint64_t cluster_offset = 0;
1807 uint64_t bytes_done = 0;
1808 QEMUIOVector hd_qiov;
1809 uint8_t *cluster_data = NULL;
1811 qemu_iovec_init(&hd_qiov, qiov->niov);
1813 qemu_co_mutex_lock(&s->lock);
1815 while (bytes != 0) {
1817 /* prepare next request */
1818 cur_bytes = MIN(bytes, INT_MAX);
1819 if (s->crypto) {
1820 cur_bytes = MIN(cur_bytes,
1821 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1824 ret = qcow2_get_cluster_offset(bs, offset, &cur_bytes, &cluster_offset);
1825 if (ret < 0) {
1826 goto fail;
1829 offset_in_cluster = offset_into_cluster(s, offset);
1831 qemu_iovec_reset(&hd_qiov);
1832 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes);
1834 switch (ret) {
1835 case QCOW2_CLUSTER_UNALLOCATED:
1837 if (bs->backing) {
1838 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1839 qemu_co_mutex_unlock(&s->lock);
1840 ret = bdrv_co_preadv(bs->backing, offset, cur_bytes,
1841 &hd_qiov, 0);
1842 qemu_co_mutex_lock(&s->lock);
1843 if (ret < 0) {
1844 goto fail;
1846 } else {
1847 /* Note: in this case, no need to wait */
1848 qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes);
1850 break;
1852 case QCOW2_CLUSTER_ZERO_PLAIN:
1853 case QCOW2_CLUSTER_ZERO_ALLOC:
1854 qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes);
1855 break;
1857 case QCOW2_CLUSTER_COMPRESSED:
1858 /* add AIO support for compressed blocks ? */
1859 ret = qcow2_decompress_cluster(bs, cluster_offset);
1860 if (ret < 0) {
1861 goto fail;
1864 qemu_iovec_from_buf(&hd_qiov, 0,
1865 s->cluster_cache + offset_in_cluster,
1866 cur_bytes);
1867 break;
1869 case QCOW2_CLUSTER_NORMAL:
1870 if ((cluster_offset & 511) != 0) {
1871 ret = -EIO;
1872 goto fail;
1875 if (bs->encrypted) {
1876 assert(s->crypto);
1879 * For encrypted images, read everything into a temporary
1880 * contiguous buffer on which the AES functions can work.
1882 if (!cluster_data) {
1883 cluster_data =
1884 qemu_try_blockalign(bs->file->bs,
1885 QCOW_MAX_CRYPT_CLUSTERS
1886 * s->cluster_size);
1887 if (cluster_data == NULL) {
1888 ret = -ENOMEM;
1889 goto fail;
1893 assert(cur_bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1894 qemu_iovec_reset(&hd_qiov);
1895 qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes);
1898 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
1899 qemu_co_mutex_unlock(&s->lock);
1900 ret = bdrv_co_preadv(bs->file,
1901 cluster_offset + offset_in_cluster,
1902 cur_bytes, &hd_qiov, 0);
1903 qemu_co_mutex_lock(&s->lock);
1904 if (ret < 0) {
1905 goto fail;
1907 if (bs->encrypted) {
1908 assert(s->crypto);
1909 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
1910 assert((cur_bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
1911 if (qcrypto_block_decrypt(s->crypto,
1912 (s->crypt_physical_offset ?
1913 cluster_offset + offset_in_cluster :
1914 offset),
1915 cluster_data,
1916 cur_bytes,
1917 NULL) < 0) {
1918 ret = -EIO;
1919 goto fail;
1921 qemu_iovec_from_buf(qiov, bytes_done, cluster_data, cur_bytes);
1923 break;
1925 default:
1926 g_assert_not_reached();
1927 ret = -EIO;
1928 goto fail;
1931 bytes -= cur_bytes;
1932 offset += cur_bytes;
1933 bytes_done += cur_bytes;
1935 ret = 0;
1937 fail:
1938 qemu_co_mutex_unlock(&s->lock);
1940 qemu_iovec_destroy(&hd_qiov);
1941 qemu_vfree(cluster_data);
1943 return ret;
1946 /* Check if it's possible to merge a write request with the writing of
1947 * the data from the COW regions */
1948 static bool merge_cow(uint64_t offset, unsigned bytes,
1949 QEMUIOVector *hd_qiov, QCowL2Meta *l2meta)
1951 QCowL2Meta *m;
1953 for (m = l2meta; m != NULL; m = m->next) {
1954 /* If both COW regions are empty then there's nothing to merge */
1955 if (m->cow_start.nb_bytes == 0 && m->cow_end.nb_bytes == 0) {
1956 continue;
1959 /* The data (middle) region must be immediately after the
1960 * start region */
1961 if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) {
1962 continue;
1965 /* The end region must be immediately after the data (middle)
1966 * region */
1967 if (m->offset + m->cow_end.offset != offset + bytes) {
1968 continue;
1971 /* Make sure that adding both COW regions to the QEMUIOVector
1972 * does not exceed IOV_MAX */
1973 if (hd_qiov->niov > IOV_MAX - 2) {
1974 continue;
1977 m->data_qiov = hd_qiov;
1978 return true;
1981 return false;
1984 static coroutine_fn int qcow2_co_pwritev(BlockDriverState *bs, uint64_t offset,
1985 uint64_t bytes, QEMUIOVector *qiov,
1986 int flags)
1988 BDRVQcow2State *s = bs->opaque;
1989 int offset_in_cluster;
1990 int ret;
1991 unsigned int cur_bytes; /* number of sectors in current iteration */
1992 uint64_t cluster_offset;
1993 QEMUIOVector hd_qiov;
1994 uint64_t bytes_done = 0;
1995 uint8_t *cluster_data = NULL;
1996 QCowL2Meta *l2meta = NULL;
1998 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes);
2000 qemu_iovec_init(&hd_qiov, qiov->niov);
2002 s->cluster_cache_offset = -1; /* disable compressed cache */
2004 qemu_co_mutex_lock(&s->lock);
2006 while (bytes != 0) {
2008 l2meta = NULL;
2010 trace_qcow2_writev_start_part(qemu_coroutine_self());
2011 offset_in_cluster = offset_into_cluster(s, offset);
2012 cur_bytes = MIN(bytes, INT_MAX);
2013 if (bs->encrypted) {
2014 cur_bytes = MIN(cur_bytes,
2015 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
2016 - offset_in_cluster);
2019 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
2020 &cluster_offset, &l2meta);
2021 if (ret < 0) {
2022 goto fail;
2025 assert((cluster_offset & 511) == 0);
2027 qemu_iovec_reset(&hd_qiov);
2028 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes);
2030 if (bs->encrypted) {
2031 assert(s->crypto);
2032 if (!cluster_data) {
2033 cluster_data = qemu_try_blockalign(bs->file->bs,
2034 QCOW_MAX_CRYPT_CLUSTERS
2035 * s->cluster_size);
2036 if (cluster_data == NULL) {
2037 ret = -ENOMEM;
2038 goto fail;
2042 assert(hd_qiov.size <=
2043 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2044 qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size);
2046 if (qcrypto_block_encrypt(s->crypto,
2047 (s->crypt_physical_offset ?
2048 cluster_offset + offset_in_cluster :
2049 offset),
2050 cluster_data,
2051 cur_bytes, NULL) < 0) {
2052 ret = -EIO;
2053 goto fail;
2056 qemu_iovec_reset(&hd_qiov);
2057 qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes);
2060 ret = qcow2_pre_write_overlap_check(bs, 0,
2061 cluster_offset + offset_in_cluster, cur_bytes);
2062 if (ret < 0) {
2063 goto fail;
2066 /* If we need to do COW, check if it's possible to merge the
2067 * writing of the guest data together with that of the COW regions.
2068 * If it's not possible (or not necessary) then write the
2069 * guest data now. */
2070 if (!merge_cow(offset, cur_bytes, &hd_qiov, l2meta)) {
2071 qemu_co_mutex_unlock(&s->lock);
2072 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
2073 trace_qcow2_writev_data(qemu_coroutine_self(),
2074 cluster_offset + offset_in_cluster);
2075 ret = bdrv_co_pwritev(bs->file,
2076 cluster_offset + offset_in_cluster,
2077 cur_bytes, &hd_qiov, 0);
2078 qemu_co_mutex_lock(&s->lock);
2079 if (ret < 0) {
2080 goto fail;
2084 ret = qcow2_handle_l2meta(bs, &l2meta, true);
2085 if (ret) {
2086 goto fail;
2089 bytes -= cur_bytes;
2090 offset += cur_bytes;
2091 bytes_done += cur_bytes;
2092 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes);
2094 ret = 0;
2096 fail:
2097 qcow2_handle_l2meta(bs, &l2meta, false);
2099 qemu_co_mutex_unlock(&s->lock);
2101 qemu_iovec_destroy(&hd_qiov);
2102 qemu_vfree(cluster_data);
2103 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
2105 return ret;
2108 static int qcow2_inactivate(BlockDriverState *bs)
2110 BDRVQcow2State *s = bs->opaque;
2111 int ret, result = 0;
2112 Error *local_err = NULL;
2114 qcow2_store_persistent_dirty_bitmaps(bs, &local_err);
2115 if (local_err != NULL) {
2116 result = -EINVAL;
2117 error_report_err(local_err);
2118 error_report("Persistent bitmaps are lost for node '%s'",
2119 bdrv_get_device_or_node_name(bs));
2122 ret = qcow2_cache_flush(bs, s->l2_table_cache);
2123 if (ret) {
2124 result = ret;
2125 error_report("Failed to flush the L2 table cache: %s",
2126 strerror(-ret));
2129 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2130 if (ret) {
2131 result = ret;
2132 error_report("Failed to flush the refcount block cache: %s",
2133 strerror(-ret));
2136 if (result == 0) {
2137 qcow2_mark_clean(bs);
2140 return result;
2143 static void qcow2_close(BlockDriverState *bs)
2145 BDRVQcow2State *s = bs->opaque;
2146 qemu_vfree(s->l1_table);
2147 /* else pre-write overlap checks in cache_destroy may crash */
2148 s->l1_table = NULL;
2150 if (!(s->flags & BDRV_O_INACTIVE)) {
2151 qcow2_inactivate(bs);
2154 cache_clean_timer_del(bs);
2155 qcow2_cache_destroy(s->l2_table_cache);
2156 qcow2_cache_destroy(s->refcount_block_cache);
2158 qcrypto_block_free(s->crypto);
2159 s->crypto = NULL;
2161 g_free(s->unknown_header_fields);
2162 cleanup_unknown_header_ext(bs);
2164 g_free(s->image_backing_file);
2165 g_free(s->image_backing_format);
2167 g_free(s->cluster_cache);
2168 qemu_vfree(s->cluster_data);
2169 qcow2_refcount_close(bs);
2170 qcow2_free_snapshots(bs);
2173 static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs,
2174 Error **errp)
2176 BDRVQcow2State *s = bs->opaque;
2177 int flags = s->flags;
2178 QCryptoBlock *crypto = NULL;
2179 QDict *options;
2180 Error *local_err = NULL;
2181 int ret;
2184 * Backing files are read-only which makes all of their metadata immutable,
2185 * that means we don't have to worry about reopening them here.
2188 crypto = s->crypto;
2189 s->crypto = NULL;
2191 qcow2_close(bs);
2193 memset(s, 0, sizeof(BDRVQcow2State));
2194 options = qdict_clone_shallow(bs->options);
2196 flags &= ~BDRV_O_INACTIVE;
2197 qemu_co_mutex_lock(&s->lock);
2198 ret = qcow2_do_open(bs, options, flags, &local_err);
2199 qemu_co_mutex_unlock(&s->lock);
2200 qobject_unref(options);
2201 if (local_err) {
2202 error_propagate(errp, local_err);
2203 error_prepend(errp, "Could not reopen qcow2 layer: ");
2204 bs->drv = NULL;
2205 return;
2206 } else if (ret < 0) {
2207 error_setg_errno(errp, -ret, "Could not reopen qcow2 layer");
2208 bs->drv = NULL;
2209 return;
2212 s->crypto = crypto;
2215 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
2216 size_t len, size_t buflen)
2218 QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
2219 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
2221 if (buflen < ext_len) {
2222 return -ENOSPC;
2225 *ext_backing_fmt = (QCowExtension) {
2226 .magic = cpu_to_be32(magic),
2227 .len = cpu_to_be32(len),
2230 if (len) {
2231 memcpy(buf + sizeof(QCowExtension), s, len);
2234 return ext_len;
2238 * Updates the qcow2 header, including the variable length parts of it, i.e.
2239 * the backing file name and all extensions. qcow2 was not designed to allow
2240 * such changes, so if we run out of space (we can only use the first cluster)
2241 * this function may fail.
2243 * Returns 0 on success, -errno in error cases.
2245 int qcow2_update_header(BlockDriverState *bs)
2247 BDRVQcow2State *s = bs->opaque;
2248 QCowHeader *header;
2249 char *buf;
2250 size_t buflen = s->cluster_size;
2251 int ret;
2252 uint64_t total_size;
2253 uint32_t refcount_table_clusters;
2254 size_t header_length;
2255 Qcow2UnknownHeaderExtension *uext;
2257 buf = qemu_blockalign(bs, buflen);
2259 /* Header structure */
2260 header = (QCowHeader*) buf;
2262 if (buflen < sizeof(*header)) {
2263 ret = -ENOSPC;
2264 goto fail;
2267 header_length = sizeof(*header) + s->unknown_header_fields_size;
2268 total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
2269 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2271 *header = (QCowHeader) {
2272 /* Version 2 fields */
2273 .magic = cpu_to_be32(QCOW_MAGIC),
2274 .version = cpu_to_be32(s->qcow_version),
2275 .backing_file_offset = 0,
2276 .backing_file_size = 0,
2277 .cluster_bits = cpu_to_be32(s->cluster_bits),
2278 .size = cpu_to_be64(total_size),
2279 .crypt_method = cpu_to_be32(s->crypt_method_header),
2280 .l1_size = cpu_to_be32(s->l1_size),
2281 .l1_table_offset = cpu_to_be64(s->l1_table_offset),
2282 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset),
2283 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
2284 .nb_snapshots = cpu_to_be32(s->nb_snapshots),
2285 .snapshots_offset = cpu_to_be64(s->snapshots_offset),
2287 /* Version 3 fields */
2288 .incompatible_features = cpu_to_be64(s->incompatible_features),
2289 .compatible_features = cpu_to_be64(s->compatible_features),
2290 .autoclear_features = cpu_to_be64(s->autoclear_features),
2291 .refcount_order = cpu_to_be32(s->refcount_order),
2292 .header_length = cpu_to_be32(header_length),
2295 /* For older versions, write a shorter header */
2296 switch (s->qcow_version) {
2297 case 2:
2298 ret = offsetof(QCowHeader, incompatible_features);
2299 break;
2300 case 3:
2301 ret = sizeof(*header);
2302 break;
2303 default:
2304 ret = -EINVAL;
2305 goto fail;
2308 buf += ret;
2309 buflen -= ret;
2310 memset(buf, 0, buflen);
2312 /* Preserve any unknown field in the header */
2313 if (s->unknown_header_fields_size) {
2314 if (buflen < s->unknown_header_fields_size) {
2315 ret = -ENOSPC;
2316 goto fail;
2319 memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
2320 buf += s->unknown_header_fields_size;
2321 buflen -= s->unknown_header_fields_size;
2324 /* Backing file format header extension */
2325 if (s->image_backing_format) {
2326 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
2327 s->image_backing_format,
2328 strlen(s->image_backing_format),
2329 buflen);
2330 if (ret < 0) {
2331 goto fail;
2334 buf += ret;
2335 buflen -= ret;
2338 /* Full disk encryption header pointer extension */
2339 if (s->crypto_header.offset != 0) {
2340 cpu_to_be64s(&s->crypto_header.offset);
2341 cpu_to_be64s(&s->crypto_header.length);
2342 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_CRYPTO_HEADER,
2343 &s->crypto_header, sizeof(s->crypto_header),
2344 buflen);
2345 be64_to_cpus(&s->crypto_header.offset);
2346 be64_to_cpus(&s->crypto_header.length);
2347 if (ret < 0) {
2348 goto fail;
2350 buf += ret;
2351 buflen -= ret;
2354 /* Feature table */
2355 if (s->qcow_version >= 3) {
2356 Qcow2Feature features[] = {
2358 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2359 .bit = QCOW2_INCOMPAT_DIRTY_BITNR,
2360 .name = "dirty bit",
2363 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2364 .bit = QCOW2_INCOMPAT_CORRUPT_BITNR,
2365 .name = "corrupt bit",
2368 .type = QCOW2_FEAT_TYPE_COMPATIBLE,
2369 .bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
2370 .name = "lazy refcounts",
2374 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
2375 features, sizeof(features), buflen);
2376 if (ret < 0) {
2377 goto fail;
2379 buf += ret;
2380 buflen -= ret;
2383 /* Bitmap extension */
2384 if (s->nb_bitmaps > 0) {
2385 Qcow2BitmapHeaderExt bitmaps_header = {
2386 .nb_bitmaps = cpu_to_be32(s->nb_bitmaps),
2387 .bitmap_directory_size =
2388 cpu_to_be64(s->bitmap_directory_size),
2389 .bitmap_directory_offset =
2390 cpu_to_be64(s->bitmap_directory_offset)
2392 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BITMAPS,
2393 &bitmaps_header, sizeof(bitmaps_header),
2394 buflen);
2395 if (ret < 0) {
2396 goto fail;
2398 buf += ret;
2399 buflen -= ret;
2402 /* Keep unknown header extensions */
2403 QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
2404 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
2405 if (ret < 0) {
2406 goto fail;
2409 buf += ret;
2410 buflen -= ret;
2413 /* End of header extensions */
2414 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
2415 if (ret < 0) {
2416 goto fail;
2419 buf += ret;
2420 buflen -= ret;
2422 /* Backing file name */
2423 if (s->image_backing_file) {
2424 size_t backing_file_len = strlen(s->image_backing_file);
2426 if (buflen < backing_file_len) {
2427 ret = -ENOSPC;
2428 goto fail;
2431 /* Using strncpy is ok here, since buf is not NUL-terminated. */
2432 strncpy(buf, s->image_backing_file, buflen);
2434 header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
2435 header->backing_file_size = cpu_to_be32(backing_file_len);
2438 /* Write the new header */
2439 ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
2440 if (ret < 0) {
2441 goto fail;
2444 ret = 0;
2445 fail:
2446 qemu_vfree(header);
2447 return ret;
2450 static int qcow2_change_backing_file(BlockDriverState *bs,
2451 const char *backing_file, const char *backing_fmt)
2453 BDRVQcow2State *s = bs->opaque;
2455 if (backing_file && strlen(backing_file) > 1023) {
2456 return -EINVAL;
2459 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
2460 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
2462 g_free(s->image_backing_file);
2463 g_free(s->image_backing_format);
2465 s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
2466 s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
2468 return qcow2_update_header(bs);
2471 static int qcow2_crypt_method_from_format(const char *encryptfmt)
2473 if (g_str_equal(encryptfmt, "luks")) {
2474 return QCOW_CRYPT_LUKS;
2475 } else if (g_str_equal(encryptfmt, "aes")) {
2476 return QCOW_CRYPT_AES;
2477 } else {
2478 return -EINVAL;
2482 static int qcow2_set_up_encryption(BlockDriverState *bs,
2483 QCryptoBlockCreateOptions *cryptoopts,
2484 Error **errp)
2486 BDRVQcow2State *s = bs->opaque;
2487 QCryptoBlock *crypto = NULL;
2488 int fmt, ret;
2490 switch (cryptoopts->format) {
2491 case Q_CRYPTO_BLOCK_FORMAT_LUKS:
2492 fmt = QCOW_CRYPT_LUKS;
2493 break;
2494 case Q_CRYPTO_BLOCK_FORMAT_QCOW:
2495 fmt = QCOW_CRYPT_AES;
2496 break;
2497 default:
2498 error_setg(errp, "Crypto format not supported in qcow2");
2499 return -EINVAL;
2502 s->crypt_method_header = fmt;
2504 crypto = qcrypto_block_create(cryptoopts, "encrypt.",
2505 qcow2_crypto_hdr_init_func,
2506 qcow2_crypto_hdr_write_func,
2507 bs, errp);
2508 if (!crypto) {
2509 return -EINVAL;
2512 ret = qcow2_update_header(bs);
2513 if (ret < 0) {
2514 error_setg_errno(errp, -ret, "Could not write encryption header");
2515 goto out;
2518 ret = 0;
2519 out:
2520 qcrypto_block_free(crypto);
2521 return ret;
2525 * Preallocates metadata structures for data clusters between @offset (in the
2526 * guest disk) and @new_length (which is thus generally the new guest disk
2527 * size).
2529 * Returns: 0 on success, -errno on failure.
2531 static int coroutine_fn preallocate_co(BlockDriverState *bs, uint64_t offset,
2532 uint64_t new_length)
2534 uint64_t bytes;
2535 uint64_t host_offset = 0;
2536 unsigned int cur_bytes;
2537 int ret;
2538 QCowL2Meta *meta;
2540 assert(offset <= new_length);
2541 bytes = new_length - offset;
2543 while (bytes) {
2544 cur_bytes = MIN(bytes, INT_MAX);
2545 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
2546 &host_offset, &meta);
2547 if (ret < 0) {
2548 return ret;
2551 while (meta) {
2552 QCowL2Meta *next = meta->next;
2554 ret = qcow2_alloc_cluster_link_l2(bs, meta);
2555 if (ret < 0) {
2556 qcow2_free_any_clusters(bs, meta->alloc_offset,
2557 meta->nb_clusters, QCOW2_DISCARD_NEVER);
2558 return ret;
2561 /* There are no dependent requests, but we need to remove our
2562 * request from the list of in-flight requests */
2563 QLIST_REMOVE(meta, next_in_flight);
2565 g_free(meta);
2566 meta = next;
2569 /* TODO Preallocate data if requested */
2571 bytes -= cur_bytes;
2572 offset += cur_bytes;
2576 * It is expected that the image file is large enough to actually contain
2577 * all of the allocated clusters (otherwise we get failing reads after
2578 * EOF). Extend the image to the last allocated sector.
2580 if (host_offset != 0) {
2581 uint8_t data = 0;
2582 ret = bdrv_pwrite(bs->file, (host_offset + cur_bytes) - 1,
2583 &data, 1);
2584 if (ret < 0) {
2585 return ret;
2589 return 0;
2592 /* qcow2_refcount_metadata_size:
2593 * @clusters: number of clusters to refcount (including data and L1/L2 tables)
2594 * @cluster_size: size of a cluster, in bytes
2595 * @refcount_order: refcount bits power-of-2 exponent
2596 * @generous_increase: allow for the refcount table to be 1.5x as large as it
2597 * needs to be
2599 * Returns: Number of bytes required for refcount blocks and table metadata.
2601 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size,
2602 int refcount_order, bool generous_increase,
2603 uint64_t *refblock_count)
2606 * Every host cluster is reference-counted, including metadata (even
2607 * refcount metadata is recursively included).
2609 * An accurate formula for the size of refcount metadata size is difficult
2610 * to derive. An easier method of calculation is finding the fixed point
2611 * where no further refcount blocks or table clusters are required to
2612 * reference count every cluster.
2614 int64_t blocks_per_table_cluster = cluster_size / sizeof(uint64_t);
2615 int64_t refcounts_per_block = cluster_size * 8 / (1 << refcount_order);
2616 int64_t table = 0; /* number of refcount table clusters */
2617 int64_t blocks = 0; /* number of refcount block clusters */
2618 int64_t last;
2619 int64_t n = 0;
2621 do {
2622 last = n;
2623 blocks = DIV_ROUND_UP(clusters + table + blocks, refcounts_per_block);
2624 table = DIV_ROUND_UP(blocks, blocks_per_table_cluster);
2625 n = clusters + blocks + table;
2627 if (n == last && generous_increase) {
2628 clusters += DIV_ROUND_UP(table, 2);
2629 n = 0; /* force another loop */
2630 generous_increase = false;
2632 } while (n != last);
2634 if (refblock_count) {
2635 *refblock_count = blocks;
2638 return (blocks + table) * cluster_size;
2642 * qcow2_calc_prealloc_size:
2643 * @total_size: virtual disk size in bytes
2644 * @cluster_size: cluster size in bytes
2645 * @refcount_order: refcount bits power-of-2 exponent
2647 * Returns: Total number of bytes required for the fully allocated image
2648 * (including metadata).
2650 static int64_t qcow2_calc_prealloc_size(int64_t total_size,
2651 size_t cluster_size,
2652 int refcount_order)
2654 int64_t meta_size = 0;
2655 uint64_t nl1e, nl2e;
2656 int64_t aligned_total_size = ROUND_UP(total_size, cluster_size);
2658 /* header: 1 cluster */
2659 meta_size += cluster_size;
2661 /* total size of L2 tables */
2662 nl2e = aligned_total_size / cluster_size;
2663 nl2e = ROUND_UP(nl2e, cluster_size / sizeof(uint64_t));
2664 meta_size += nl2e * sizeof(uint64_t);
2666 /* total size of L1 tables */
2667 nl1e = nl2e * sizeof(uint64_t) / cluster_size;
2668 nl1e = ROUND_UP(nl1e, cluster_size / sizeof(uint64_t));
2669 meta_size += nl1e * sizeof(uint64_t);
2671 /* total size of refcount table and blocks */
2672 meta_size += qcow2_refcount_metadata_size(
2673 (meta_size + aligned_total_size) / cluster_size,
2674 cluster_size, refcount_order, false, NULL);
2676 return meta_size + aligned_total_size;
2679 static bool validate_cluster_size(size_t cluster_size, Error **errp)
2681 int cluster_bits = ctz32(cluster_size);
2682 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
2683 (1 << cluster_bits) != cluster_size)
2685 error_setg(errp, "Cluster size must be a power of two between %d and "
2686 "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
2687 return false;
2689 return true;
2692 static size_t qcow2_opt_get_cluster_size_del(QemuOpts *opts, Error **errp)
2694 size_t cluster_size;
2696 cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
2697 DEFAULT_CLUSTER_SIZE);
2698 if (!validate_cluster_size(cluster_size, errp)) {
2699 return 0;
2701 return cluster_size;
2704 static int qcow2_opt_get_version_del(QemuOpts *opts, Error **errp)
2706 char *buf;
2707 int ret;
2709 buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
2710 if (!buf) {
2711 ret = 3; /* default */
2712 } else if (!strcmp(buf, "0.10")) {
2713 ret = 2;
2714 } else if (!strcmp(buf, "1.1")) {
2715 ret = 3;
2716 } else {
2717 error_setg(errp, "Invalid compatibility level: '%s'", buf);
2718 ret = -EINVAL;
2720 g_free(buf);
2721 return ret;
2724 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts *opts, int version,
2725 Error **errp)
2727 uint64_t refcount_bits;
2729 refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 16);
2730 if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
2731 error_setg(errp, "Refcount width must be a power of two and may not "
2732 "exceed 64 bits");
2733 return 0;
2736 if (version < 3 && refcount_bits != 16) {
2737 error_setg(errp, "Different refcount widths than 16 bits require "
2738 "compatibility level 1.1 or above (use compat=1.1 or "
2739 "greater)");
2740 return 0;
2743 return refcount_bits;
2746 static int coroutine_fn
2747 qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp)
2749 BlockdevCreateOptionsQcow2 *qcow2_opts;
2750 QDict *options;
2753 * Open the image file and write a minimal qcow2 header.
2755 * We keep things simple and start with a zero-sized image. We also
2756 * do without refcount blocks or a L1 table for now. We'll fix the
2757 * inconsistency later.
2759 * We do need a refcount table because growing the refcount table means
2760 * allocating two new refcount blocks - the seconds of which would be at
2761 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
2762 * size for any qcow2 image.
2764 BlockBackend *blk = NULL;
2765 BlockDriverState *bs = NULL;
2766 QCowHeader *header;
2767 size_t cluster_size;
2768 int version;
2769 int refcount_order;
2770 uint64_t* refcount_table;
2771 Error *local_err = NULL;
2772 int ret;
2774 assert(create_options->driver == BLOCKDEV_DRIVER_QCOW2);
2775 qcow2_opts = &create_options->u.qcow2;
2777 bs = bdrv_open_blockdev_ref(qcow2_opts->file, errp);
2778 if (bs == NULL) {
2779 return -EIO;
2782 /* Validate options and set default values */
2783 if (!QEMU_IS_ALIGNED(qcow2_opts->size, BDRV_SECTOR_SIZE)) {
2784 error_setg(errp, "Image size must be a multiple of 512 bytes");
2785 ret = -EINVAL;
2786 goto out;
2789 if (qcow2_opts->has_version) {
2790 switch (qcow2_opts->version) {
2791 case BLOCKDEV_QCOW2_VERSION_V2:
2792 version = 2;
2793 break;
2794 case BLOCKDEV_QCOW2_VERSION_V3:
2795 version = 3;
2796 break;
2797 default:
2798 g_assert_not_reached();
2800 } else {
2801 version = 3;
2804 if (qcow2_opts->has_cluster_size) {
2805 cluster_size = qcow2_opts->cluster_size;
2806 } else {
2807 cluster_size = DEFAULT_CLUSTER_SIZE;
2810 if (!validate_cluster_size(cluster_size, errp)) {
2811 ret = -EINVAL;
2812 goto out;
2815 if (!qcow2_opts->has_preallocation) {
2816 qcow2_opts->preallocation = PREALLOC_MODE_OFF;
2818 if (qcow2_opts->has_backing_file &&
2819 qcow2_opts->preallocation != PREALLOC_MODE_OFF)
2821 error_setg(errp, "Backing file and preallocation cannot be used at "
2822 "the same time");
2823 ret = -EINVAL;
2824 goto out;
2826 if (qcow2_opts->has_backing_fmt && !qcow2_opts->has_backing_file) {
2827 error_setg(errp, "Backing format cannot be used without backing file");
2828 ret = -EINVAL;
2829 goto out;
2832 if (!qcow2_opts->has_lazy_refcounts) {
2833 qcow2_opts->lazy_refcounts = false;
2835 if (version < 3 && qcow2_opts->lazy_refcounts) {
2836 error_setg(errp, "Lazy refcounts only supported with compatibility "
2837 "level 1.1 and above (use version=v3 or greater)");
2838 ret = -EINVAL;
2839 goto out;
2842 if (!qcow2_opts->has_refcount_bits) {
2843 qcow2_opts->refcount_bits = 16;
2845 if (qcow2_opts->refcount_bits > 64 ||
2846 !is_power_of_2(qcow2_opts->refcount_bits))
2848 error_setg(errp, "Refcount width must be a power of two and may not "
2849 "exceed 64 bits");
2850 ret = -EINVAL;
2851 goto out;
2853 if (version < 3 && qcow2_opts->refcount_bits != 16) {
2854 error_setg(errp, "Different refcount widths than 16 bits require "
2855 "compatibility level 1.1 or above (use version=v3 or "
2856 "greater)");
2857 ret = -EINVAL;
2858 goto out;
2860 refcount_order = ctz32(qcow2_opts->refcount_bits);
2863 /* Create BlockBackend to write to the image */
2864 blk = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL);
2865 ret = blk_insert_bs(blk, bs, errp);
2866 if (ret < 0) {
2867 goto out;
2869 blk_set_allow_write_beyond_eof(blk, true);
2871 /* Clear the protocol layer and preallocate it if necessary */
2872 ret = blk_truncate(blk, 0, PREALLOC_MODE_OFF, errp);
2873 if (ret < 0) {
2874 goto out;
2877 if (qcow2_opts->preallocation == PREALLOC_MODE_FULL ||
2878 qcow2_opts->preallocation == PREALLOC_MODE_FALLOC)
2880 int64_t prealloc_size =
2881 qcow2_calc_prealloc_size(qcow2_opts->size, cluster_size,
2882 refcount_order);
2884 ret = blk_truncate(blk, prealloc_size, qcow2_opts->preallocation, errp);
2885 if (ret < 0) {
2886 goto out;
2890 /* Write the header */
2891 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
2892 header = g_malloc0(cluster_size);
2893 *header = (QCowHeader) {
2894 .magic = cpu_to_be32(QCOW_MAGIC),
2895 .version = cpu_to_be32(version),
2896 .cluster_bits = cpu_to_be32(ctz32(cluster_size)),
2897 .size = cpu_to_be64(0),
2898 .l1_table_offset = cpu_to_be64(0),
2899 .l1_size = cpu_to_be32(0),
2900 .refcount_table_offset = cpu_to_be64(cluster_size),
2901 .refcount_table_clusters = cpu_to_be32(1),
2902 .refcount_order = cpu_to_be32(refcount_order),
2903 .header_length = cpu_to_be32(sizeof(*header)),
2906 /* We'll update this to correct value later */
2907 header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
2909 if (qcow2_opts->lazy_refcounts) {
2910 header->compatible_features |=
2911 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
2914 ret = blk_pwrite(blk, 0, header, cluster_size, 0);
2915 g_free(header);
2916 if (ret < 0) {
2917 error_setg_errno(errp, -ret, "Could not write qcow2 header");
2918 goto out;
2921 /* Write a refcount table with one refcount block */
2922 refcount_table = g_malloc0(2 * cluster_size);
2923 refcount_table[0] = cpu_to_be64(2 * cluster_size);
2924 ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0);
2925 g_free(refcount_table);
2927 if (ret < 0) {
2928 error_setg_errno(errp, -ret, "Could not write refcount table");
2929 goto out;
2932 blk_unref(blk);
2933 blk = NULL;
2936 * And now open the image and make it consistent first (i.e. increase the
2937 * refcount of the cluster that is occupied by the header and the refcount
2938 * table)
2940 options = qdict_new();
2941 qdict_put_str(options, "driver", "qcow2");
2942 qdict_put_str(options, "file", bs->node_name);
2943 blk = blk_new_open(NULL, NULL, options,
2944 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH,
2945 &local_err);
2946 if (blk == NULL) {
2947 error_propagate(errp, local_err);
2948 ret = -EIO;
2949 goto out;
2952 ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size);
2953 if (ret < 0) {
2954 error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
2955 "header and refcount table");
2956 goto out;
2958 } else if (ret != 0) {
2959 error_report("Huh, first cluster in empty image is already in use?");
2960 abort();
2963 /* Create a full header (including things like feature table) */
2964 ret = qcow2_update_header(blk_bs(blk));
2965 if (ret < 0) {
2966 error_setg_errno(errp, -ret, "Could not update qcow2 header");
2967 goto out;
2970 /* Okay, now that we have a valid image, let's give it the right size */
2971 ret = blk_truncate(blk, qcow2_opts->size, PREALLOC_MODE_OFF, errp);
2972 if (ret < 0) {
2973 error_prepend(errp, "Could not resize image: ");
2974 goto out;
2977 /* Want a backing file? There you go.*/
2978 if (qcow2_opts->has_backing_file) {
2979 const char *backing_format = NULL;
2981 if (qcow2_opts->has_backing_fmt) {
2982 backing_format = BlockdevDriver_str(qcow2_opts->backing_fmt);
2985 ret = bdrv_change_backing_file(blk_bs(blk), qcow2_opts->backing_file,
2986 backing_format);
2987 if (ret < 0) {
2988 error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
2989 "with format '%s'", qcow2_opts->backing_file,
2990 backing_format);
2991 goto out;
2995 /* Want encryption? There you go. */
2996 if (qcow2_opts->has_encrypt) {
2997 ret = qcow2_set_up_encryption(blk_bs(blk), qcow2_opts->encrypt, errp);
2998 if (ret < 0) {
2999 goto out;
3003 /* And if we're supposed to preallocate metadata, do that now */
3004 if (qcow2_opts->preallocation != PREALLOC_MODE_OFF) {
3005 BDRVQcow2State *s = blk_bs(blk)->opaque;
3006 qemu_co_mutex_lock(&s->lock);
3007 ret = preallocate_co(blk_bs(blk), 0, qcow2_opts->size);
3008 qemu_co_mutex_unlock(&s->lock);
3010 if (ret < 0) {
3011 error_setg_errno(errp, -ret, "Could not preallocate metadata");
3012 goto out;
3016 blk_unref(blk);
3017 blk = NULL;
3019 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
3020 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
3021 * have to setup decryption context. We're not doing any I/O on the top
3022 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
3023 * not have effect.
3025 options = qdict_new();
3026 qdict_put_str(options, "driver", "qcow2");
3027 qdict_put_str(options, "file", bs->node_name);
3028 blk = blk_new_open(NULL, NULL, options,
3029 BDRV_O_RDWR | BDRV_O_NO_BACKING | BDRV_O_NO_IO,
3030 &local_err);
3031 if (blk == NULL) {
3032 error_propagate(errp, local_err);
3033 ret = -EIO;
3034 goto out;
3037 ret = 0;
3038 out:
3039 blk_unref(blk);
3040 bdrv_unref(bs);
3041 return ret;
3044 static int coroutine_fn qcow2_co_create_opts(const char *filename, QemuOpts *opts,
3045 Error **errp)
3047 BlockdevCreateOptions *create_options = NULL;
3048 QDict *qdict;
3049 Visitor *v;
3050 BlockDriverState *bs = NULL;
3051 Error *local_err = NULL;
3052 const char *val;
3053 int ret;
3055 /* Only the keyval visitor supports the dotted syntax needed for
3056 * encryption, so go through a QDict before getting a QAPI type. Ignore
3057 * options meant for the protocol layer so that the visitor doesn't
3058 * complain. */
3059 qdict = qemu_opts_to_qdict_filtered(opts, NULL, bdrv_qcow2.create_opts,
3060 true);
3062 /* Handle encryption options */
3063 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT);
3064 if (val && !strcmp(val, "on")) {
3065 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow");
3066 } else if (val && !strcmp(val, "off")) {
3067 qdict_del(qdict, BLOCK_OPT_ENCRYPT);
3070 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT);
3071 if (val && !strcmp(val, "aes")) {
3072 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow");
3075 /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into
3076 * version=v2/v3 below. */
3077 val = qdict_get_try_str(qdict, BLOCK_OPT_COMPAT_LEVEL);
3078 if (val && !strcmp(val, "0.10")) {
3079 qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v2");
3080 } else if (val && !strcmp(val, "1.1")) {
3081 qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v3");
3084 /* Change legacy command line options into QMP ones */
3085 static const QDictRenames opt_renames[] = {
3086 { BLOCK_OPT_BACKING_FILE, "backing-file" },
3087 { BLOCK_OPT_BACKING_FMT, "backing-fmt" },
3088 { BLOCK_OPT_CLUSTER_SIZE, "cluster-size" },
3089 { BLOCK_OPT_LAZY_REFCOUNTS, "lazy-refcounts" },
3090 { BLOCK_OPT_REFCOUNT_BITS, "refcount-bits" },
3091 { BLOCK_OPT_ENCRYPT, BLOCK_OPT_ENCRYPT_FORMAT },
3092 { BLOCK_OPT_COMPAT_LEVEL, "version" },
3093 { NULL, NULL },
3096 if (!qdict_rename_keys(qdict, opt_renames, errp)) {
3097 ret = -EINVAL;
3098 goto finish;
3101 /* Create and open the file (protocol layer) */
3102 ret = bdrv_create_file(filename, opts, errp);
3103 if (ret < 0) {
3104 goto finish;
3107 bs = bdrv_open(filename, NULL, NULL,
3108 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
3109 if (bs == NULL) {
3110 ret = -EIO;
3111 goto finish;
3114 /* Set 'driver' and 'node' options */
3115 qdict_put_str(qdict, "driver", "qcow2");
3116 qdict_put_str(qdict, "file", bs->node_name);
3118 /* Now get the QAPI type BlockdevCreateOptions */
3119 v = qobject_input_visitor_new_flat_confused(qdict, errp);
3120 if (!v) {
3121 ret = -EINVAL;
3122 goto finish;
3125 visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err);
3126 visit_free(v);
3128 if (local_err) {
3129 error_propagate(errp, local_err);
3130 ret = -EINVAL;
3131 goto finish;
3134 /* Silently round up size */
3135 create_options->u.qcow2.size = ROUND_UP(create_options->u.qcow2.size,
3136 BDRV_SECTOR_SIZE);
3138 /* Create the qcow2 image (format layer) */
3139 ret = qcow2_co_create(create_options, errp);
3140 if (ret < 0) {
3141 goto finish;
3144 ret = 0;
3145 finish:
3146 qobject_unref(qdict);
3147 bdrv_unref(bs);
3148 qapi_free_BlockdevCreateOptions(create_options);
3149 return ret;
3153 static bool is_zero(BlockDriverState *bs, int64_t offset, int64_t bytes)
3155 int64_t nr;
3156 int res;
3158 /* Clamp to image length, before checking status of underlying sectors */
3159 if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) {
3160 bytes = bs->total_sectors * BDRV_SECTOR_SIZE - offset;
3163 if (!bytes) {
3164 return true;
3166 res = bdrv_block_status_above(bs, NULL, offset, bytes, &nr, NULL, NULL);
3167 return res >= 0 && (res & BDRV_BLOCK_ZERO) && nr == bytes;
3170 static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs,
3171 int64_t offset, int bytes, BdrvRequestFlags flags)
3173 int ret;
3174 BDRVQcow2State *s = bs->opaque;
3176 uint32_t head = offset % s->cluster_size;
3177 uint32_t tail = (offset + bytes) % s->cluster_size;
3179 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, bytes);
3180 if (offset + bytes == bs->total_sectors * BDRV_SECTOR_SIZE) {
3181 tail = 0;
3184 if (head || tail) {
3185 uint64_t off;
3186 unsigned int nr;
3188 assert(head + bytes <= s->cluster_size);
3190 /* check whether remainder of cluster already reads as zero */
3191 if (!(is_zero(bs, offset - head, head) &&
3192 is_zero(bs, offset + bytes,
3193 tail ? s->cluster_size - tail : 0))) {
3194 return -ENOTSUP;
3197 qemu_co_mutex_lock(&s->lock);
3198 /* We can have new write after previous check */
3199 offset = QEMU_ALIGN_DOWN(offset, s->cluster_size);
3200 bytes = s->cluster_size;
3201 nr = s->cluster_size;
3202 ret = qcow2_get_cluster_offset(bs, offset, &nr, &off);
3203 if (ret != QCOW2_CLUSTER_UNALLOCATED &&
3204 ret != QCOW2_CLUSTER_ZERO_PLAIN &&
3205 ret != QCOW2_CLUSTER_ZERO_ALLOC) {
3206 qemu_co_mutex_unlock(&s->lock);
3207 return -ENOTSUP;
3209 } else {
3210 qemu_co_mutex_lock(&s->lock);
3213 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, bytes);
3215 /* Whatever is left can use real zero clusters */
3216 ret = qcow2_cluster_zeroize(bs, offset, bytes, flags);
3217 qemu_co_mutex_unlock(&s->lock);
3219 return ret;
3222 static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs,
3223 int64_t offset, int bytes)
3225 int ret;
3226 BDRVQcow2State *s = bs->opaque;
3228 if (!QEMU_IS_ALIGNED(offset | bytes, s->cluster_size)) {
3229 assert(bytes < s->cluster_size);
3230 /* Ignore partial clusters, except for the special case of the
3231 * complete partial cluster at the end of an unaligned file */
3232 if (!QEMU_IS_ALIGNED(offset, s->cluster_size) ||
3233 offset + bytes != bs->total_sectors * BDRV_SECTOR_SIZE) {
3234 return -ENOTSUP;
3238 qemu_co_mutex_lock(&s->lock);
3239 ret = qcow2_cluster_discard(bs, offset, bytes, QCOW2_DISCARD_REQUEST,
3240 false);
3241 qemu_co_mutex_unlock(&s->lock);
3242 return ret;
3245 static int coroutine_fn
3246 qcow2_co_copy_range_from(BlockDriverState *bs,
3247 BdrvChild *src, uint64_t src_offset,
3248 BdrvChild *dst, uint64_t dst_offset,
3249 uint64_t bytes, BdrvRequestFlags flags)
3251 BDRVQcow2State *s = bs->opaque;
3252 int ret;
3253 unsigned int cur_bytes; /* number of bytes in current iteration */
3254 BdrvChild *child = NULL;
3255 BdrvRequestFlags cur_flags;
3257 assert(!bs->encrypted);
3258 qemu_co_mutex_lock(&s->lock);
3260 while (bytes != 0) {
3261 uint64_t copy_offset = 0;
3262 /* prepare next request */
3263 cur_bytes = MIN(bytes, INT_MAX);
3264 cur_flags = flags;
3266 ret = qcow2_get_cluster_offset(bs, src_offset, &cur_bytes, &copy_offset);
3267 if (ret < 0) {
3268 goto out;
3271 switch (ret) {
3272 case QCOW2_CLUSTER_UNALLOCATED:
3273 if (bs->backing && bs->backing->bs) {
3274 int64_t backing_length = bdrv_getlength(bs->backing->bs);
3275 if (src_offset >= backing_length) {
3276 cur_flags |= BDRV_REQ_ZERO_WRITE;
3277 } else {
3278 child = bs->backing;
3279 cur_bytes = MIN(cur_bytes, backing_length - src_offset);
3280 copy_offset = src_offset;
3282 } else {
3283 cur_flags |= BDRV_REQ_ZERO_WRITE;
3285 break;
3287 case QCOW2_CLUSTER_ZERO_PLAIN:
3288 case QCOW2_CLUSTER_ZERO_ALLOC:
3289 cur_flags |= BDRV_REQ_ZERO_WRITE;
3290 break;
3292 case QCOW2_CLUSTER_COMPRESSED:
3293 ret = -ENOTSUP;
3294 goto out;
3295 break;
3297 case QCOW2_CLUSTER_NORMAL:
3298 child = bs->file;
3299 copy_offset += offset_into_cluster(s, src_offset);
3300 if ((copy_offset & 511) != 0) {
3301 ret = -EIO;
3302 goto out;
3304 break;
3306 default:
3307 abort();
3309 qemu_co_mutex_unlock(&s->lock);
3310 ret = bdrv_co_copy_range_from(child,
3311 copy_offset,
3312 dst, dst_offset,
3313 cur_bytes, cur_flags);
3314 qemu_co_mutex_lock(&s->lock);
3315 if (ret < 0) {
3316 goto out;
3319 bytes -= cur_bytes;
3320 src_offset += cur_bytes;
3321 dst_offset += cur_bytes;
3323 ret = 0;
3325 out:
3326 qemu_co_mutex_unlock(&s->lock);
3327 return ret;
3330 static int coroutine_fn
3331 qcow2_co_copy_range_to(BlockDriverState *bs,
3332 BdrvChild *src, uint64_t src_offset,
3333 BdrvChild *dst, uint64_t dst_offset,
3334 uint64_t bytes, BdrvRequestFlags flags)
3336 BDRVQcow2State *s = bs->opaque;
3337 int offset_in_cluster;
3338 int ret;
3339 unsigned int cur_bytes; /* number of sectors in current iteration */
3340 uint64_t cluster_offset;
3341 uint8_t *cluster_data = NULL;
3342 QCowL2Meta *l2meta = NULL;
3344 assert(!bs->encrypted);
3345 s->cluster_cache_offset = -1; /* disable compressed cache */
3347 qemu_co_mutex_lock(&s->lock);
3349 while (bytes != 0) {
3351 l2meta = NULL;
3353 offset_in_cluster = offset_into_cluster(s, dst_offset);
3354 cur_bytes = MIN(bytes, INT_MAX);
3356 /* TODO:
3357 * If src->bs == dst->bs, we could simply copy by incrementing
3358 * the refcnt, without copying user data.
3359 * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */
3360 ret = qcow2_alloc_cluster_offset(bs, dst_offset, &cur_bytes,
3361 &cluster_offset, &l2meta);
3362 if (ret < 0) {
3363 goto fail;
3366 assert((cluster_offset & 511) == 0);
3368 ret = qcow2_pre_write_overlap_check(bs, 0,
3369 cluster_offset + offset_in_cluster, cur_bytes);
3370 if (ret < 0) {
3371 goto fail;
3374 qemu_co_mutex_unlock(&s->lock);
3375 ret = bdrv_co_copy_range_to(src, src_offset,
3376 bs->file,
3377 cluster_offset + offset_in_cluster,
3378 cur_bytes, flags);
3379 qemu_co_mutex_lock(&s->lock);
3380 if (ret < 0) {
3381 goto fail;
3384 ret = qcow2_handle_l2meta(bs, &l2meta, true);
3385 if (ret) {
3386 goto fail;
3389 bytes -= cur_bytes;
3390 src_offset += cur_bytes;
3391 dst_offset += cur_bytes;
3393 ret = 0;
3395 fail:
3396 qcow2_handle_l2meta(bs, &l2meta, false);
3398 qemu_co_mutex_unlock(&s->lock);
3400 qemu_vfree(cluster_data);
3401 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
3403 return ret;
3406 static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset,
3407 PreallocMode prealloc, Error **errp)
3409 BDRVQcow2State *s = bs->opaque;
3410 uint64_t old_length;
3411 int64_t new_l1_size;
3412 int ret;
3414 if (prealloc != PREALLOC_MODE_OFF && prealloc != PREALLOC_MODE_METADATA &&
3415 prealloc != PREALLOC_MODE_FALLOC && prealloc != PREALLOC_MODE_FULL)
3417 error_setg(errp, "Unsupported preallocation mode '%s'",
3418 PreallocMode_str(prealloc));
3419 return -ENOTSUP;
3422 if (offset & 511) {
3423 error_setg(errp, "The new size must be a multiple of 512");
3424 return -EINVAL;
3427 qemu_co_mutex_lock(&s->lock);
3429 /* cannot proceed if image has snapshots */
3430 if (s->nb_snapshots) {
3431 error_setg(errp, "Can't resize an image which has snapshots");
3432 ret = -ENOTSUP;
3433 goto fail;
3436 /* cannot proceed if image has bitmaps */
3437 if (s->nb_bitmaps) {
3438 /* TODO: resize bitmaps in the image */
3439 error_setg(errp, "Can't resize an image which has bitmaps");
3440 ret = -ENOTSUP;
3441 goto fail;
3444 old_length = bs->total_sectors * 512;
3445 new_l1_size = size_to_l1(s, offset);
3447 if (offset < old_length) {
3448 int64_t last_cluster, old_file_size;
3449 if (prealloc != PREALLOC_MODE_OFF) {
3450 error_setg(errp,
3451 "Preallocation can't be used for shrinking an image");
3452 ret = -EINVAL;
3453 goto fail;
3456 ret = qcow2_cluster_discard(bs, ROUND_UP(offset, s->cluster_size),
3457 old_length - ROUND_UP(offset,
3458 s->cluster_size),
3459 QCOW2_DISCARD_ALWAYS, true);
3460 if (ret < 0) {
3461 error_setg_errno(errp, -ret, "Failed to discard cropped clusters");
3462 goto fail;
3465 ret = qcow2_shrink_l1_table(bs, new_l1_size);
3466 if (ret < 0) {
3467 error_setg_errno(errp, -ret,
3468 "Failed to reduce the number of L2 tables");
3469 goto fail;
3472 ret = qcow2_shrink_reftable(bs);
3473 if (ret < 0) {
3474 error_setg_errno(errp, -ret,
3475 "Failed to discard unused refblocks");
3476 goto fail;
3479 old_file_size = bdrv_getlength(bs->file->bs);
3480 if (old_file_size < 0) {
3481 error_setg_errno(errp, -old_file_size,
3482 "Failed to inquire current file length");
3483 ret = old_file_size;
3484 goto fail;
3486 last_cluster = qcow2_get_last_cluster(bs, old_file_size);
3487 if (last_cluster < 0) {
3488 error_setg_errno(errp, -last_cluster,
3489 "Failed to find the last cluster");
3490 ret = last_cluster;
3491 goto fail;
3493 if ((last_cluster + 1) * s->cluster_size < old_file_size) {
3494 Error *local_err = NULL;
3496 bdrv_co_truncate(bs->file, (last_cluster + 1) * s->cluster_size,
3497 PREALLOC_MODE_OFF, &local_err);
3498 if (local_err) {
3499 warn_reportf_err(local_err,
3500 "Failed to truncate the tail of the image: ");
3503 } else {
3504 ret = qcow2_grow_l1_table(bs, new_l1_size, true);
3505 if (ret < 0) {
3506 error_setg_errno(errp, -ret, "Failed to grow the L1 table");
3507 goto fail;
3511 switch (prealloc) {
3512 case PREALLOC_MODE_OFF:
3513 break;
3515 case PREALLOC_MODE_METADATA:
3516 ret = preallocate_co(bs, old_length, offset);
3517 if (ret < 0) {
3518 error_setg_errno(errp, -ret, "Preallocation failed");
3519 goto fail;
3521 break;
3523 case PREALLOC_MODE_FALLOC:
3524 case PREALLOC_MODE_FULL:
3526 int64_t allocation_start, host_offset, guest_offset;
3527 int64_t clusters_allocated;
3528 int64_t old_file_size, new_file_size;
3529 uint64_t nb_new_data_clusters, nb_new_l2_tables;
3531 old_file_size = bdrv_getlength(bs->file->bs);
3532 if (old_file_size < 0) {
3533 error_setg_errno(errp, -old_file_size,
3534 "Failed to inquire current file length");
3535 ret = old_file_size;
3536 goto fail;
3538 old_file_size = ROUND_UP(old_file_size, s->cluster_size);
3540 nb_new_data_clusters = DIV_ROUND_UP(offset - old_length,
3541 s->cluster_size);
3543 /* This is an overestimation; we will not actually allocate space for
3544 * these in the file but just make sure the new refcount structures are
3545 * able to cover them so we will not have to allocate new refblocks
3546 * while entering the data blocks in the potentially new L2 tables.
3547 * (We do not actually care where the L2 tables are placed. Maybe they
3548 * are already allocated or they can be placed somewhere before
3549 * @old_file_size. It does not matter because they will be fully
3550 * allocated automatically, so they do not need to be covered by the
3551 * preallocation. All that matters is that we will not have to allocate
3552 * new refcount structures for them.) */
3553 nb_new_l2_tables = DIV_ROUND_UP(nb_new_data_clusters,
3554 s->cluster_size / sizeof(uint64_t));
3555 /* The cluster range may not be aligned to L2 boundaries, so add one L2
3556 * table for a potential head/tail */
3557 nb_new_l2_tables++;
3559 allocation_start = qcow2_refcount_area(bs, old_file_size,
3560 nb_new_data_clusters +
3561 nb_new_l2_tables,
3562 true, 0, 0);
3563 if (allocation_start < 0) {
3564 error_setg_errno(errp, -allocation_start,
3565 "Failed to resize refcount structures");
3566 ret = allocation_start;
3567 goto fail;
3570 clusters_allocated = qcow2_alloc_clusters_at(bs, allocation_start,
3571 nb_new_data_clusters);
3572 if (clusters_allocated < 0) {
3573 error_setg_errno(errp, -clusters_allocated,
3574 "Failed to allocate data clusters");
3575 ret = clusters_allocated;
3576 goto fail;
3579 assert(clusters_allocated == nb_new_data_clusters);
3581 /* Allocate the data area */
3582 new_file_size = allocation_start +
3583 nb_new_data_clusters * s->cluster_size;
3584 ret = bdrv_co_truncate(bs->file, new_file_size, prealloc, errp);
3585 if (ret < 0) {
3586 error_prepend(errp, "Failed to resize underlying file: ");
3587 qcow2_free_clusters(bs, allocation_start,
3588 nb_new_data_clusters * s->cluster_size,
3589 QCOW2_DISCARD_OTHER);
3590 goto fail;
3593 /* Create the necessary L2 entries */
3594 host_offset = allocation_start;
3595 guest_offset = old_length;
3596 while (nb_new_data_clusters) {
3597 int64_t nb_clusters = MIN(
3598 nb_new_data_clusters,
3599 s->l2_slice_size - offset_to_l2_slice_index(s, guest_offset));
3600 QCowL2Meta allocation = {
3601 .offset = guest_offset,
3602 .alloc_offset = host_offset,
3603 .nb_clusters = nb_clusters,
3605 qemu_co_queue_init(&allocation.dependent_requests);
3607 ret = qcow2_alloc_cluster_link_l2(bs, &allocation);
3608 if (ret < 0) {
3609 error_setg_errno(errp, -ret, "Failed to update L2 tables");
3610 qcow2_free_clusters(bs, host_offset,
3611 nb_new_data_clusters * s->cluster_size,
3612 QCOW2_DISCARD_OTHER);
3613 goto fail;
3616 guest_offset += nb_clusters * s->cluster_size;
3617 host_offset += nb_clusters * s->cluster_size;
3618 nb_new_data_clusters -= nb_clusters;
3620 break;
3623 default:
3624 g_assert_not_reached();
3627 if (prealloc != PREALLOC_MODE_OFF) {
3628 /* Flush metadata before actually changing the image size */
3629 ret = qcow2_write_caches(bs);
3630 if (ret < 0) {
3631 error_setg_errno(errp, -ret,
3632 "Failed to flush the preallocated area to disk");
3633 goto fail;
3637 /* write updated header.size */
3638 offset = cpu_to_be64(offset);
3639 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
3640 &offset, sizeof(uint64_t));
3641 if (ret < 0) {
3642 error_setg_errno(errp, -ret, "Failed to update the image size");
3643 goto fail;
3646 s->l1_vm_state_index = new_l1_size;
3647 ret = 0;
3648 fail:
3649 qemu_co_mutex_unlock(&s->lock);
3650 return ret;
3653 /* XXX: put compressed sectors first, then all the cluster aligned
3654 tables to avoid losing bytes in alignment */
3655 static coroutine_fn int
3656 qcow2_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
3657 uint64_t bytes, QEMUIOVector *qiov)
3659 BDRVQcow2State *s = bs->opaque;
3660 QEMUIOVector hd_qiov;
3661 struct iovec iov;
3662 z_stream strm;
3663 int ret, out_len;
3664 uint8_t *buf, *out_buf;
3665 int64_t cluster_offset;
3667 if (bytes == 0) {
3668 /* align end of file to a sector boundary to ease reading with
3669 sector based I/Os */
3670 cluster_offset = bdrv_getlength(bs->file->bs);
3671 if (cluster_offset < 0) {
3672 return cluster_offset;
3674 return bdrv_co_truncate(bs->file, cluster_offset, PREALLOC_MODE_OFF,
3675 NULL);
3678 if (offset_into_cluster(s, offset)) {
3679 return -EINVAL;
3682 buf = qemu_blockalign(bs, s->cluster_size);
3683 if (bytes != s->cluster_size) {
3684 if (bytes > s->cluster_size ||
3685 offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
3687 qemu_vfree(buf);
3688 return -EINVAL;
3690 /* Zero-pad last write if image size is not cluster aligned */
3691 memset(buf + bytes, 0, s->cluster_size - bytes);
3693 qemu_iovec_to_buf(qiov, 0, buf, bytes);
3695 out_buf = g_malloc(s->cluster_size);
3697 /* best compression, small window, no zlib header */
3698 memset(&strm, 0, sizeof(strm));
3699 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
3700 Z_DEFLATED, -12,
3701 9, Z_DEFAULT_STRATEGY);
3702 if (ret != 0) {
3703 ret = -EINVAL;
3704 goto fail;
3707 strm.avail_in = s->cluster_size;
3708 strm.next_in = (uint8_t *)buf;
3709 strm.avail_out = s->cluster_size;
3710 strm.next_out = out_buf;
3712 ret = deflate(&strm, Z_FINISH);
3713 if (ret != Z_STREAM_END && ret != Z_OK) {
3714 deflateEnd(&strm);
3715 ret = -EINVAL;
3716 goto fail;
3718 out_len = strm.next_out - out_buf;
3720 deflateEnd(&strm);
3722 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
3723 /* could not compress: write normal cluster */
3724 ret = qcow2_co_pwritev(bs, offset, bytes, qiov, 0);
3725 if (ret < 0) {
3726 goto fail;
3728 goto success;
3731 qemu_co_mutex_lock(&s->lock);
3732 cluster_offset =
3733 qcow2_alloc_compressed_cluster_offset(bs, offset, out_len);
3734 if (!cluster_offset) {
3735 qemu_co_mutex_unlock(&s->lock);
3736 ret = -EIO;
3737 goto fail;
3739 cluster_offset &= s->cluster_offset_mask;
3741 ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len);
3742 qemu_co_mutex_unlock(&s->lock);
3743 if (ret < 0) {
3744 goto fail;
3747 iov = (struct iovec) {
3748 .iov_base = out_buf,
3749 .iov_len = out_len,
3751 qemu_iovec_init_external(&hd_qiov, &iov, 1);
3753 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
3754 ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0);
3755 if (ret < 0) {
3756 goto fail;
3758 success:
3759 ret = 0;
3760 fail:
3761 qemu_vfree(buf);
3762 g_free(out_buf);
3763 return ret;
3766 static int make_completely_empty(BlockDriverState *bs)
3768 BDRVQcow2State *s = bs->opaque;
3769 Error *local_err = NULL;
3770 int ret, l1_clusters;
3771 int64_t offset;
3772 uint64_t *new_reftable = NULL;
3773 uint64_t rt_entry, l1_size2;
3774 struct {
3775 uint64_t l1_offset;
3776 uint64_t reftable_offset;
3777 uint32_t reftable_clusters;
3778 } QEMU_PACKED l1_ofs_rt_ofs_cls;
3780 ret = qcow2_cache_empty(bs, s->l2_table_cache);
3781 if (ret < 0) {
3782 goto fail;
3785 ret = qcow2_cache_empty(bs, s->refcount_block_cache);
3786 if (ret < 0) {
3787 goto fail;
3790 /* Refcounts will be broken utterly */
3791 ret = qcow2_mark_dirty(bs);
3792 if (ret < 0) {
3793 goto fail;
3796 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
3798 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
3799 l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t);
3801 /* After this call, neither the in-memory nor the on-disk refcount
3802 * information accurately describe the actual references */
3804 ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset,
3805 l1_clusters * s->cluster_size, 0);
3806 if (ret < 0) {
3807 goto fail_broken_refcounts;
3809 memset(s->l1_table, 0, l1_size2);
3811 BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE);
3813 /* Overwrite enough clusters at the beginning of the sectors to place
3814 * the refcount table, a refcount block and the L1 table in; this may
3815 * overwrite parts of the existing refcount and L1 table, which is not
3816 * an issue because the dirty flag is set, complete data loss is in fact
3817 * desired and partial data loss is consequently fine as well */
3818 ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size,
3819 (2 + l1_clusters) * s->cluster_size, 0);
3820 /* This call (even if it failed overall) may have overwritten on-disk
3821 * refcount structures; in that case, the in-memory refcount information
3822 * will probably differ from the on-disk information which makes the BDS
3823 * unusable */
3824 if (ret < 0) {
3825 goto fail_broken_refcounts;
3828 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
3829 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
3831 /* "Create" an empty reftable (one cluster) directly after the image
3832 * header and an empty L1 table three clusters after the image header;
3833 * the cluster between those two will be used as the first refblock */
3834 l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size);
3835 l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size);
3836 l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1);
3837 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset),
3838 &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls));
3839 if (ret < 0) {
3840 goto fail_broken_refcounts;
3843 s->l1_table_offset = 3 * s->cluster_size;
3845 new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t));
3846 if (!new_reftable) {
3847 ret = -ENOMEM;
3848 goto fail_broken_refcounts;
3851 s->refcount_table_offset = s->cluster_size;
3852 s->refcount_table_size = s->cluster_size / sizeof(uint64_t);
3853 s->max_refcount_table_index = 0;
3855 g_free(s->refcount_table);
3856 s->refcount_table = new_reftable;
3857 new_reftable = NULL;
3859 /* Now the in-memory refcount information again corresponds to the on-disk
3860 * information (reftable is empty and no refblocks (the refblock cache is
3861 * empty)); however, this means some clusters (e.g. the image header) are
3862 * referenced, but not refcounted, but the normal qcow2 code assumes that
3863 * the in-memory information is always correct */
3865 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
3867 /* Enter the first refblock into the reftable */
3868 rt_entry = cpu_to_be64(2 * s->cluster_size);
3869 ret = bdrv_pwrite_sync(bs->file, s->cluster_size,
3870 &rt_entry, sizeof(rt_entry));
3871 if (ret < 0) {
3872 goto fail_broken_refcounts;
3874 s->refcount_table[0] = 2 * s->cluster_size;
3876 s->free_cluster_index = 0;
3877 assert(3 + l1_clusters <= s->refcount_block_size);
3878 offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2);
3879 if (offset < 0) {
3880 ret = offset;
3881 goto fail_broken_refcounts;
3882 } else if (offset > 0) {
3883 error_report("First cluster in emptied image is in use");
3884 abort();
3887 /* Now finally the in-memory information corresponds to the on-disk
3888 * structures and is correct */
3889 ret = qcow2_mark_clean(bs);
3890 if (ret < 0) {
3891 goto fail;
3894 ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size,
3895 PREALLOC_MODE_OFF, &local_err);
3896 if (ret < 0) {
3897 error_report_err(local_err);
3898 goto fail;
3901 return 0;
3903 fail_broken_refcounts:
3904 /* The BDS is unusable at this point. If we wanted to make it usable, we
3905 * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
3906 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
3907 * again. However, because the functions which could have caused this error
3908 * path to be taken are used by those functions as well, it's very likely
3909 * that that sequence will fail as well. Therefore, just eject the BDS. */
3910 bs->drv = NULL;
3912 fail:
3913 g_free(new_reftable);
3914 return ret;
3917 static int qcow2_make_empty(BlockDriverState *bs)
3919 BDRVQcow2State *s = bs->opaque;
3920 uint64_t offset, end_offset;
3921 int step = QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size);
3922 int l1_clusters, ret = 0;
3924 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
3926 if (s->qcow_version >= 3 && !s->snapshots && !s->nb_bitmaps &&
3927 3 + l1_clusters <= s->refcount_block_size &&
3928 s->crypt_method_header != QCOW_CRYPT_LUKS) {
3929 /* The following function only works for qcow2 v3 images (it
3930 * requires the dirty flag) and only as long as there are no
3931 * features that reserve extra clusters (such as snapshots,
3932 * LUKS header, or persistent bitmaps), because it completely
3933 * empties the image. Furthermore, the L1 table and three
3934 * additional clusters (image header, refcount table, one
3935 * refcount block) have to fit inside one refcount block. */
3936 return make_completely_empty(bs);
3939 /* This fallback code simply discards every active cluster; this is slow,
3940 * but works in all cases */
3941 end_offset = bs->total_sectors * BDRV_SECTOR_SIZE;
3942 for (offset = 0; offset < end_offset; offset += step) {
3943 /* As this function is generally used after committing an external
3944 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
3945 * default action for this kind of discard is to pass the discard,
3946 * which will ideally result in an actually smaller image file, as
3947 * is probably desired. */
3948 ret = qcow2_cluster_discard(bs, offset, MIN(step, end_offset - offset),
3949 QCOW2_DISCARD_SNAPSHOT, true);
3950 if (ret < 0) {
3951 break;
3955 return ret;
3958 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
3960 BDRVQcow2State *s = bs->opaque;
3961 int ret;
3963 qemu_co_mutex_lock(&s->lock);
3964 ret = qcow2_write_caches(bs);
3965 qemu_co_mutex_unlock(&s->lock);
3967 return ret;
3970 static BlockMeasureInfo *qcow2_measure(QemuOpts *opts, BlockDriverState *in_bs,
3971 Error **errp)
3973 Error *local_err = NULL;
3974 BlockMeasureInfo *info;
3975 uint64_t required = 0; /* bytes that contribute to required size */
3976 uint64_t virtual_size; /* disk size as seen by guest */
3977 uint64_t refcount_bits;
3978 uint64_t l2_tables;
3979 size_t cluster_size;
3980 int version;
3981 char *optstr;
3982 PreallocMode prealloc;
3983 bool has_backing_file;
3985 /* Parse image creation options */
3986 cluster_size = qcow2_opt_get_cluster_size_del(opts, &local_err);
3987 if (local_err) {
3988 goto err;
3991 version = qcow2_opt_get_version_del(opts, &local_err);
3992 if (local_err) {
3993 goto err;
3996 refcount_bits = qcow2_opt_get_refcount_bits_del(opts, version, &local_err);
3997 if (local_err) {
3998 goto err;
4001 optstr = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
4002 prealloc = qapi_enum_parse(&PreallocMode_lookup, optstr,
4003 PREALLOC_MODE_OFF, &local_err);
4004 g_free(optstr);
4005 if (local_err) {
4006 goto err;
4009 optstr = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
4010 has_backing_file = !!optstr;
4011 g_free(optstr);
4013 virtual_size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
4014 virtual_size = ROUND_UP(virtual_size, cluster_size);
4016 /* Check that virtual disk size is valid */
4017 l2_tables = DIV_ROUND_UP(virtual_size / cluster_size,
4018 cluster_size / sizeof(uint64_t));
4019 if (l2_tables * sizeof(uint64_t) > QCOW_MAX_L1_SIZE) {
4020 error_setg(&local_err, "The image size is too large "
4021 "(try using a larger cluster size)");
4022 goto err;
4025 /* Account for input image */
4026 if (in_bs) {
4027 int64_t ssize = bdrv_getlength(in_bs);
4028 if (ssize < 0) {
4029 error_setg_errno(&local_err, -ssize,
4030 "Unable to get image virtual_size");
4031 goto err;
4034 virtual_size = ROUND_UP(ssize, cluster_size);
4036 if (has_backing_file) {
4037 /* We don't how much of the backing chain is shared by the input
4038 * image and the new image file. In the worst case the new image's
4039 * backing file has nothing in common with the input image. Be
4040 * conservative and assume all clusters need to be written.
4042 required = virtual_size;
4043 } else {
4044 int64_t offset;
4045 int64_t pnum = 0;
4047 for (offset = 0; offset < ssize; offset += pnum) {
4048 int ret;
4050 ret = bdrv_block_status_above(in_bs, NULL, offset,
4051 ssize - offset, &pnum, NULL,
4052 NULL);
4053 if (ret < 0) {
4054 error_setg_errno(&local_err, -ret,
4055 "Unable to get block status");
4056 goto err;
4059 if (ret & BDRV_BLOCK_ZERO) {
4060 /* Skip zero regions (safe with no backing file) */
4061 } else if ((ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) ==
4062 (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) {
4063 /* Extend pnum to end of cluster for next iteration */
4064 pnum = ROUND_UP(offset + pnum, cluster_size) - offset;
4066 /* Count clusters we've seen */
4067 required += offset % cluster_size + pnum;
4073 /* Take into account preallocation. Nothing special is needed for
4074 * PREALLOC_MODE_METADATA since metadata is always counted.
4076 if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
4077 required = virtual_size;
4080 info = g_new(BlockMeasureInfo, 1);
4081 info->fully_allocated =
4082 qcow2_calc_prealloc_size(virtual_size, cluster_size,
4083 ctz32(refcount_bits));
4085 /* Remove data clusters that are not required. This overestimates the
4086 * required size because metadata needed for the fully allocated file is
4087 * still counted.
4089 info->required = info->fully_allocated - virtual_size + required;
4090 return info;
4092 err:
4093 error_propagate(errp, local_err);
4094 return NULL;
4097 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
4099 BDRVQcow2State *s = bs->opaque;
4100 bdi->unallocated_blocks_are_zero = true;
4101 bdi->cluster_size = s->cluster_size;
4102 bdi->vm_state_offset = qcow2_vm_state_offset(s);
4103 return 0;
4106 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs)
4108 BDRVQcow2State *s = bs->opaque;
4109 ImageInfoSpecific *spec_info;
4110 QCryptoBlockInfo *encrypt_info = NULL;
4112 if (s->crypto != NULL) {
4113 encrypt_info = qcrypto_block_get_info(s->crypto, &error_abort);
4116 spec_info = g_new(ImageInfoSpecific, 1);
4117 *spec_info = (ImageInfoSpecific){
4118 .type = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
4119 .u.qcow2.data = g_new(ImageInfoSpecificQCow2, 1),
4121 if (s->qcow_version == 2) {
4122 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
4123 .compat = g_strdup("0.10"),
4124 .refcount_bits = s->refcount_bits,
4126 } else if (s->qcow_version == 3) {
4127 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
4128 .compat = g_strdup("1.1"),
4129 .lazy_refcounts = s->compatible_features &
4130 QCOW2_COMPAT_LAZY_REFCOUNTS,
4131 .has_lazy_refcounts = true,
4132 .corrupt = s->incompatible_features &
4133 QCOW2_INCOMPAT_CORRUPT,
4134 .has_corrupt = true,
4135 .refcount_bits = s->refcount_bits,
4137 } else {
4138 /* if this assertion fails, this probably means a new version was
4139 * added without having it covered here */
4140 assert(false);
4143 if (encrypt_info) {
4144 ImageInfoSpecificQCow2Encryption *qencrypt =
4145 g_new(ImageInfoSpecificQCow2Encryption, 1);
4146 switch (encrypt_info->format) {
4147 case Q_CRYPTO_BLOCK_FORMAT_QCOW:
4148 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES;
4149 break;
4150 case Q_CRYPTO_BLOCK_FORMAT_LUKS:
4151 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS;
4152 qencrypt->u.luks = encrypt_info->u.luks;
4153 break;
4154 default:
4155 abort();
4157 /* Since we did shallow copy above, erase any pointers
4158 * in the original info */
4159 memset(&encrypt_info->u, 0, sizeof(encrypt_info->u));
4160 qapi_free_QCryptoBlockInfo(encrypt_info);
4162 spec_info->u.qcow2.data->has_encrypt = true;
4163 spec_info->u.qcow2.data->encrypt = qencrypt;
4166 return spec_info;
4169 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
4170 int64_t pos)
4172 BDRVQcow2State *s = bs->opaque;
4174 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
4175 return bs->drv->bdrv_co_pwritev(bs, qcow2_vm_state_offset(s) + pos,
4176 qiov->size, qiov, 0);
4179 static int qcow2_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
4180 int64_t pos)
4182 BDRVQcow2State *s = bs->opaque;
4184 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
4185 return bs->drv->bdrv_co_preadv(bs, qcow2_vm_state_offset(s) + pos,
4186 qiov->size, qiov, 0);
4190 * Downgrades an image's version. To achieve this, any incompatible features
4191 * have to be removed.
4193 static int qcow2_downgrade(BlockDriverState *bs, int target_version,
4194 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
4195 Error **errp)
4197 BDRVQcow2State *s = bs->opaque;
4198 int current_version = s->qcow_version;
4199 int ret;
4201 /* This is qcow2_downgrade(), not qcow2_upgrade() */
4202 assert(target_version < current_version);
4204 /* There are no other versions (now) that you can downgrade to */
4205 assert(target_version == 2);
4207 if (s->refcount_order != 4) {
4208 error_setg(errp, "compat=0.10 requires refcount_bits=16");
4209 return -ENOTSUP;
4212 /* clear incompatible features */
4213 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
4214 ret = qcow2_mark_clean(bs);
4215 if (ret < 0) {
4216 error_setg_errno(errp, -ret, "Failed to make the image clean");
4217 return ret;
4221 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
4222 * the first place; if that happens nonetheless, returning -ENOTSUP is the
4223 * best thing to do anyway */
4225 if (s->incompatible_features) {
4226 error_setg(errp, "Cannot downgrade an image with incompatible features "
4227 "%#" PRIx64 " set", s->incompatible_features);
4228 return -ENOTSUP;
4231 /* since we can ignore compatible features, we can set them to 0 as well */
4232 s->compatible_features = 0;
4233 /* if lazy refcounts have been used, they have already been fixed through
4234 * clearing the dirty flag */
4236 /* clearing autoclear features is trivial */
4237 s->autoclear_features = 0;
4239 ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque);
4240 if (ret < 0) {
4241 error_setg_errno(errp, -ret, "Failed to turn zero into data clusters");
4242 return ret;
4245 s->qcow_version = target_version;
4246 ret = qcow2_update_header(bs);
4247 if (ret < 0) {
4248 s->qcow_version = current_version;
4249 error_setg_errno(errp, -ret, "Failed to update the image header");
4250 return ret;
4252 return 0;
4255 typedef enum Qcow2AmendOperation {
4256 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
4257 * statically initialized to so that the helper CB can discern the first
4258 * invocation from an operation change */
4259 QCOW2_NO_OPERATION = 0,
4261 QCOW2_CHANGING_REFCOUNT_ORDER,
4262 QCOW2_DOWNGRADING,
4263 } Qcow2AmendOperation;
4265 typedef struct Qcow2AmendHelperCBInfo {
4266 /* The code coordinating the amend operations should only modify
4267 * these four fields; the rest will be managed by the CB */
4268 BlockDriverAmendStatusCB *original_status_cb;
4269 void *original_cb_opaque;
4271 Qcow2AmendOperation current_operation;
4273 /* Total number of operations to perform (only set once) */
4274 int total_operations;
4276 /* The following fields are managed by the CB */
4278 /* Number of operations completed */
4279 int operations_completed;
4281 /* Cumulative offset of all completed operations */
4282 int64_t offset_completed;
4284 Qcow2AmendOperation last_operation;
4285 int64_t last_work_size;
4286 } Qcow2AmendHelperCBInfo;
4288 static void qcow2_amend_helper_cb(BlockDriverState *bs,
4289 int64_t operation_offset,
4290 int64_t operation_work_size, void *opaque)
4292 Qcow2AmendHelperCBInfo *info = opaque;
4293 int64_t current_work_size;
4294 int64_t projected_work_size;
4296 if (info->current_operation != info->last_operation) {
4297 if (info->last_operation != QCOW2_NO_OPERATION) {
4298 info->offset_completed += info->last_work_size;
4299 info->operations_completed++;
4302 info->last_operation = info->current_operation;
4305 assert(info->total_operations > 0);
4306 assert(info->operations_completed < info->total_operations);
4308 info->last_work_size = operation_work_size;
4310 current_work_size = info->offset_completed + operation_work_size;
4312 /* current_work_size is the total work size for (operations_completed + 1)
4313 * operations (which includes this one), so multiply it by the number of
4314 * operations not covered and divide it by the number of operations
4315 * covered to get a projection for the operations not covered */
4316 projected_work_size = current_work_size * (info->total_operations -
4317 info->operations_completed - 1)
4318 / (info->operations_completed + 1);
4320 info->original_status_cb(bs, info->offset_completed + operation_offset,
4321 current_work_size + projected_work_size,
4322 info->original_cb_opaque);
4325 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,
4326 BlockDriverAmendStatusCB *status_cb,
4327 void *cb_opaque,
4328 Error **errp)
4330 BDRVQcow2State *s = bs->opaque;
4331 int old_version = s->qcow_version, new_version = old_version;
4332 uint64_t new_size = 0;
4333 const char *backing_file = NULL, *backing_format = NULL;
4334 bool lazy_refcounts = s->use_lazy_refcounts;
4335 const char *compat = NULL;
4336 uint64_t cluster_size = s->cluster_size;
4337 bool encrypt;
4338 int encformat;
4339 int refcount_bits = s->refcount_bits;
4340 int ret;
4341 QemuOptDesc *desc = opts->list->desc;
4342 Qcow2AmendHelperCBInfo helper_cb_info;
4344 while (desc && desc->name) {
4345 if (!qemu_opt_find(opts, desc->name)) {
4346 /* only change explicitly defined options */
4347 desc++;
4348 continue;
4351 if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) {
4352 compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL);
4353 if (!compat) {
4354 /* preserve default */
4355 } else if (!strcmp(compat, "0.10")) {
4356 new_version = 2;
4357 } else if (!strcmp(compat, "1.1")) {
4358 new_version = 3;
4359 } else {
4360 error_setg(errp, "Unknown compatibility level %s", compat);
4361 return -EINVAL;
4363 } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) {
4364 error_setg(errp, "Cannot change preallocation mode");
4365 return -ENOTSUP;
4366 } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) {
4367 new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
4368 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) {
4369 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
4370 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) {
4371 backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
4372 } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) {
4373 encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT,
4374 !!s->crypto);
4376 if (encrypt != !!s->crypto) {
4377 error_setg(errp,
4378 "Changing the encryption flag is not supported");
4379 return -ENOTSUP;
4381 } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT_FORMAT)) {
4382 encformat = qcow2_crypt_method_from_format(
4383 qemu_opt_get(opts, BLOCK_OPT_ENCRYPT_FORMAT));
4385 if (encformat != s->crypt_method_header) {
4386 error_setg(errp,
4387 "Changing the encryption format is not supported");
4388 return -ENOTSUP;
4390 } else if (g_str_has_prefix(desc->name, "encrypt.")) {
4391 error_setg(errp,
4392 "Changing the encryption parameters is not supported");
4393 return -ENOTSUP;
4394 } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) {
4395 cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE,
4396 cluster_size);
4397 if (cluster_size != s->cluster_size) {
4398 error_setg(errp, "Changing the cluster size is not supported");
4399 return -ENOTSUP;
4401 } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
4402 lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS,
4403 lazy_refcounts);
4404 } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) {
4405 refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS,
4406 refcount_bits);
4408 if (refcount_bits <= 0 || refcount_bits > 64 ||
4409 !is_power_of_2(refcount_bits))
4411 error_setg(errp, "Refcount width must be a power of two and "
4412 "may not exceed 64 bits");
4413 return -EINVAL;
4415 } else {
4416 /* if this point is reached, this probably means a new option was
4417 * added without having it covered here */
4418 abort();
4421 desc++;
4424 helper_cb_info = (Qcow2AmendHelperCBInfo){
4425 .original_status_cb = status_cb,
4426 .original_cb_opaque = cb_opaque,
4427 .total_operations = (new_version < old_version)
4428 + (s->refcount_bits != refcount_bits)
4431 /* Upgrade first (some features may require compat=1.1) */
4432 if (new_version > old_version) {
4433 s->qcow_version = new_version;
4434 ret = qcow2_update_header(bs);
4435 if (ret < 0) {
4436 s->qcow_version = old_version;
4437 error_setg_errno(errp, -ret, "Failed to update the image header");
4438 return ret;
4442 if (s->refcount_bits != refcount_bits) {
4443 int refcount_order = ctz32(refcount_bits);
4445 if (new_version < 3 && refcount_bits != 16) {
4446 error_setg(errp, "Refcount widths other than 16 bits require "
4447 "compatibility level 1.1 or above (use compat=1.1 or "
4448 "greater)");
4449 return -EINVAL;
4452 helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER;
4453 ret = qcow2_change_refcount_order(bs, refcount_order,
4454 &qcow2_amend_helper_cb,
4455 &helper_cb_info, errp);
4456 if (ret < 0) {
4457 return ret;
4461 if (backing_file || backing_format) {
4462 ret = qcow2_change_backing_file(bs,
4463 backing_file ?: s->image_backing_file,
4464 backing_format ?: s->image_backing_format);
4465 if (ret < 0) {
4466 error_setg_errno(errp, -ret, "Failed to change the backing file");
4467 return ret;
4471 if (s->use_lazy_refcounts != lazy_refcounts) {
4472 if (lazy_refcounts) {
4473 if (new_version < 3) {
4474 error_setg(errp, "Lazy refcounts only supported with "
4475 "compatibility level 1.1 and above (use compat=1.1 "
4476 "or greater)");
4477 return -EINVAL;
4479 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
4480 ret = qcow2_update_header(bs);
4481 if (ret < 0) {
4482 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
4483 error_setg_errno(errp, -ret, "Failed to update the image header");
4484 return ret;
4486 s->use_lazy_refcounts = true;
4487 } else {
4488 /* make image clean first */
4489 ret = qcow2_mark_clean(bs);
4490 if (ret < 0) {
4491 error_setg_errno(errp, -ret, "Failed to make the image clean");
4492 return ret;
4494 /* now disallow lazy refcounts */
4495 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
4496 ret = qcow2_update_header(bs);
4497 if (ret < 0) {
4498 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
4499 error_setg_errno(errp, -ret, "Failed to update the image header");
4500 return ret;
4502 s->use_lazy_refcounts = false;
4506 if (new_size) {
4507 BlockBackend *blk = blk_new(BLK_PERM_RESIZE, BLK_PERM_ALL);
4508 ret = blk_insert_bs(blk, bs, errp);
4509 if (ret < 0) {
4510 blk_unref(blk);
4511 return ret;
4514 ret = blk_truncate(blk, new_size, PREALLOC_MODE_OFF, errp);
4515 blk_unref(blk);
4516 if (ret < 0) {
4517 return ret;
4521 /* Downgrade last (so unsupported features can be removed before) */
4522 if (new_version < old_version) {
4523 helper_cb_info.current_operation = QCOW2_DOWNGRADING;
4524 ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb,
4525 &helper_cb_info, errp);
4526 if (ret < 0) {
4527 return ret;
4531 return 0;
4535 * If offset or size are negative, respectively, they will not be included in
4536 * the BLOCK_IMAGE_CORRUPTED event emitted.
4537 * fatal will be ignored for read-only BDS; corruptions found there will always
4538 * be considered non-fatal.
4540 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
4541 int64_t size, const char *message_format, ...)
4543 BDRVQcow2State *s = bs->opaque;
4544 const char *node_name;
4545 char *message;
4546 va_list ap;
4548 fatal = fatal && bdrv_is_writable(bs);
4550 if (s->signaled_corruption &&
4551 (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
4553 return;
4556 va_start(ap, message_format);
4557 message = g_strdup_vprintf(message_format, ap);
4558 va_end(ap);
4560 if (fatal) {
4561 fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
4562 "corruption events will be suppressed\n", message);
4563 } else {
4564 fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
4565 "corruption events will be suppressed\n", message);
4568 node_name = bdrv_get_node_name(bs);
4569 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs),
4570 *node_name != '\0', node_name,
4571 message, offset >= 0, offset,
4572 size >= 0, size,
4573 fatal, &error_abort);
4574 g_free(message);
4576 if (fatal) {
4577 qcow2_mark_corrupt(bs);
4578 bs->drv = NULL; /* make BDS unusable */
4581 s->signaled_corruption = true;
4584 static QemuOptsList qcow2_create_opts = {
4585 .name = "qcow2-create-opts",
4586 .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
4587 .desc = {
4589 .name = BLOCK_OPT_SIZE,
4590 .type = QEMU_OPT_SIZE,
4591 .help = "Virtual disk size"
4594 .name = BLOCK_OPT_COMPAT_LEVEL,
4595 .type = QEMU_OPT_STRING,
4596 .help = "Compatibility level (0.10 or 1.1)"
4599 .name = BLOCK_OPT_BACKING_FILE,
4600 .type = QEMU_OPT_STRING,
4601 .help = "File name of a base image"
4604 .name = BLOCK_OPT_BACKING_FMT,
4605 .type = QEMU_OPT_STRING,
4606 .help = "Image format of the base image"
4609 .name = BLOCK_OPT_ENCRYPT,
4610 .type = QEMU_OPT_BOOL,
4611 .help = "Encrypt the image with format 'aes'. (Deprecated "
4612 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",
4615 .name = BLOCK_OPT_ENCRYPT_FORMAT,
4616 .type = QEMU_OPT_STRING,
4617 .help = "Encrypt the image, format choices: 'aes', 'luks'",
4619 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
4620 "ID of secret providing qcow AES key or LUKS passphrase"),
4621 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."),
4622 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."),
4623 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."),
4624 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."),
4625 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."),
4626 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
4628 .name = BLOCK_OPT_CLUSTER_SIZE,
4629 .type = QEMU_OPT_SIZE,
4630 .help = "qcow2 cluster size",
4631 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
4634 .name = BLOCK_OPT_PREALLOC,
4635 .type = QEMU_OPT_STRING,
4636 .help = "Preallocation mode (allowed values: off, metadata, "
4637 "falloc, full)"
4640 .name = BLOCK_OPT_LAZY_REFCOUNTS,
4641 .type = QEMU_OPT_BOOL,
4642 .help = "Postpone refcount updates",
4643 .def_value_str = "off"
4646 .name = BLOCK_OPT_REFCOUNT_BITS,
4647 .type = QEMU_OPT_NUMBER,
4648 .help = "Width of a reference count entry in bits",
4649 .def_value_str = "16"
4651 { /* end of list */ }
4655 BlockDriver bdrv_qcow2 = {
4656 .format_name = "qcow2",
4657 .instance_size = sizeof(BDRVQcow2State),
4658 .bdrv_probe = qcow2_probe,
4659 .bdrv_open = qcow2_open,
4660 .bdrv_close = qcow2_close,
4661 .bdrv_reopen_prepare = qcow2_reopen_prepare,
4662 .bdrv_reopen_commit = qcow2_reopen_commit,
4663 .bdrv_reopen_abort = qcow2_reopen_abort,
4664 .bdrv_join_options = qcow2_join_options,
4665 .bdrv_child_perm = bdrv_format_default_perms,
4666 .bdrv_co_create_opts = qcow2_co_create_opts,
4667 .bdrv_co_create = qcow2_co_create,
4668 .bdrv_has_zero_init = bdrv_has_zero_init_1,
4669 .bdrv_co_block_status = qcow2_co_block_status,
4671 .bdrv_co_preadv = qcow2_co_preadv,
4672 .bdrv_co_pwritev = qcow2_co_pwritev,
4673 .bdrv_co_flush_to_os = qcow2_co_flush_to_os,
4675 .bdrv_co_pwrite_zeroes = qcow2_co_pwrite_zeroes,
4676 .bdrv_co_pdiscard = qcow2_co_pdiscard,
4677 .bdrv_co_copy_range_from = qcow2_co_copy_range_from,
4678 .bdrv_co_copy_range_to = qcow2_co_copy_range_to,
4679 .bdrv_co_truncate = qcow2_co_truncate,
4680 .bdrv_co_pwritev_compressed = qcow2_co_pwritev_compressed,
4681 .bdrv_make_empty = qcow2_make_empty,
4683 .bdrv_snapshot_create = qcow2_snapshot_create,
4684 .bdrv_snapshot_goto = qcow2_snapshot_goto,
4685 .bdrv_snapshot_delete = qcow2_snapshot_delete,
4686 .bdrv_snapshot_list = qcow2_snapshot_list,
4687 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
4688 .bdrv_measure = qcow2_measure,
4689 .bdrv_get_info = qcow2_get_info,
4690 .bdrv_get_specific_info = qcow2_get_specific_info,
4692 .bdrv_save_vmstate = qcow2_save_vmstate,
4693 .bdrv_load_vmstate = qcow2_load_vmstate,
4695 .supports_backing = true,
4696 .bdrv_change_backing_file = qcow2_change_backing_file,
4698 .bdrv_refresh_limits = qcow2_refresh_limits,
4699 .bdrv_co_invalidate_cache = qcow2_co_invalidate_cache,
4700 .bdrv_inactivate = qcow2_inactivate,
4702 .create_opts = &qcow2_create_opts,
4703 .bdrv_co_check = qcow2_co_check,
4704 .bdrv_amend_options = qcow2_amend_options,
4706 .bdrv_detach_aio_context = qcow2_detach_aio_context,
4707 .bdrv_attach_aio_context = qcow2_attach_aio_context,
4709 .bdrv_reopen_bitmaps_rw = qcow2_reopen_bitmaps_rw,
4710 .bdrv_can_store_new_dirty_bitmap = qcow2_can_store_new_dirty_bitmap,
4711 .bdrv_remove_persistent_dirty_bitmap = qcow2_remove_persistent_dirty_bitmap,
4714 static void bdrv_qcow2_init(void)
4716 bdrv_register(&bdrv_qcow2);
4719 block_init(bdrv_qcow2_init);