ui/sdl2: disable SDL_HINT_GRAB_KEYBOARD on Windows
[qemu/ar7.git] / block / qcow2.c
blob30fd53fa64bcd11eaddab8a53b558548b0c532fb
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"
27 #include "block/qdict.h"
28 #include "sysemu/block-backend.h"
29 #include "qemu/main-loop.h"
30 #include "qemu/module.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 "qemu/memalign.h"
42 #include "qapi/qobject-input-visitor.h"
43 #include "qapi/qapi-visit-block-core.h"
44 #include "crypto.h"
45 #include "block/aio_task.h"
46 #include "block/dirty-bitmap.h"
49 Differences with QCOW:
51 - Support for multiple incremental snapshots.
52 - Memory management by reference counts.
53 - Clusters which have a reference count of one have the bit
54 QCOW_OFLAG_COPIED to optimize write performance.
55 - Size of compressed clusters is stored in sectors to reduce bit usage
56 in the cluster offsets.
57 - Support for storing additional data (such as the VM state) in the
58 snapshots.
59 - If a backing store is used, the cluster size is not constrained
60 (could be backported to QCOW).
61 - L2 tables have always a size of one cluster.
65 typedef struct {
66 uint32_t magic;
67 uint32_t len;
68 } QEMU_PACKED QCowExtension;
70 #define QCOW2_EXT_MAGIC_END 0
71 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xe2792aca
72 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
73 #define QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77
74 #define QCOW2_EXT_MAGIC_BITMAPS 0x23852875
75 #define QCOW2_EXT_MAGIC_DATA_FILE 0x44415441
77 static int coroutine_fn
78 qcow2_co_preadv_compressed(BlockDriverState *bs,
79 uint64_t l2_entry,
80 uint64_t offset,
81 uint64_t bytes,
82 QEMUIOVector *qiov,
83 size_t qiov_offset);
85 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
87 const QCowHeader *cow_header = (const void *)buf;
89 if (buf_size >= sizeof(QCowHeader) &&
90 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
91 be32_to_cpu(cow_header->version) >= 2)
92 return 100;
93 else
94 return 0;
98 static int qcow2_crypto_hdr_read_func(QCryptoBlock *block, size_t offset,
99 uint8_t *buf, size_t buflen,
100 void *opaque, Error **errp)
102 BlockDriverState *bs = opaque;
103 BDRVQcow2State *s = bs->opaque;
104 ssize_t ret;
106 if ((offset + buflen) > s->crypto_header.length) {
107 error_setg(errp, "Request for data outside of extension header");
108 return -1;
111 ret = bdrv_pread(bs->file, s->crypto_header.offset + offset, buflen, buf,
113 if (ret < 0) {
114 error_setg_errno(errp, -ret, "Could not read encryption header");
115 return -1;
117 return 0;
121 static int qcow2_crypto_hdr_init_func(QCryptoBlock *block, size_t headerlen,
122 void *opaque, Error **errp)
124 BlockDriverState *bs = opaque;
125 BDRVQcow2State *s = bs->opaque;
126 int64_t ret;
127 int64_t clusterlen;
129 ret = qcow2_alloc_clusters(bs, headerlen);
130 if (ret < 0) {
131 error_setg_errno(errp, -ret,
132 "Cannot allocate cluster for LUKS header size %zu",
133 headerlen);
134 return -1;
137 s->crypto_header.length = headerlen;
138 s->crypto_header.offset = ret;
141 * Zero fill all space in cluster so it has predictable
142 * content, as we may not initialize some regions of the
143 * header (eg only 1 out of 8 key slots will be initialized)
145 clusterlen = size_to_clusters(s, headerlen) * s->cluster_size;
146 assert(qcow2_pre_write_overlap_check(bs, 0, ret, clusterlen, false) == 0);
147 ret = bdrv_pwrite_zeroes(bs->file,
148 ret,
149 clusterlen, 0);
150 if (ret < 0) {
151 error_setg_errno(errp, -ret, "Could not zero fill encryption header");
152 return -1;
155 return 0;
159 static int qcow2_crypto_hdr_write_func(QCryptoBlock *block, size_t offset,
160 const uint8_t *buf, size_t buflen,
161 void *opaque, Error **errp)
163 BlockDriverState *bs = opaque;
164 BDRVQcow2State *s = bs->opaque;
165 ssize_t ret;
167 if ((offset + buflen) > s->crypto_header.length) {
168 error_setg(errp, "Request for data outside of extension header");
169 return -1;
172 ret = bdrv_pwrite(bs->file, s->crypto_header.offset + offset, buflen, buf,
174 if (ret < 0) {
175 error_setg_errno(errp, -ret, "Could not read encryption header");
176 return -1;
178 return 0;
181 static QDict*
182 qcow2_extract_crypto_opts(QemuOpts *opts, const char *fmt, Error **errp)
184 QDict *cryptoopts_qdict;
185 QDict *opts_qdict;
187 /* Extract "encrypt." options into a qdict */
188 opts_qdict = qemu_opts_to_qdict(opts, NULL);
189 qdict_extract_subqdict(opts_qdict, &cryptoopts_qdict, "encrypt.");
190 qobject_unref(opts_qdict);
191 qdict_put_str(cryptoopts_qdict, "format", fmt);
192 return cryptoopts_qdict;
196 * read qcow2 extension and fill bs
197 * start reading from start_offset
198 * finish reading upon magic of value 0 or when end_offset reached
199 * unknown magic is skipped (future extension this version knows nothing about)
200 * return 0 upon success, non-0 otherwise
202 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
203 uint64_t end_offset, void **p_feature_table,
204 int flags, bool *need_update_header,
205 Error **errp)
207 BDRVQcow2State *s = bs->opaque;
208 QCowExtension ext;
209 uint64_t offset;
210 int ret;
211 Qcow2BitmapHeaderExt bitmaps_ext;
213 if (need_update_header != NULL) {
214 *need_update_header = false;
217 #ifdef DEBUG_EXT
218 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
219 #endif
220 offset = start_offset;
221 while (offset < end_offset) {
223 #ifdef DEBUG_EXT
224 /* Sanity check */
225 if (offset > s->cluster_size)
226 printf("qcow2_read_extension: suspicious offset %lu\n", offset);
228 printf("attempting to read extended header in offset %lu\n", offset);
229 #endif
231 ret = bdrv_pread(bs->file, offset, sizeof(ext), &ext, 0);
232 if (ret < 0) {
233 error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: "
234 "pread fail from offset %" PRIu64, offset);
235 return 1;
237 ext.magic = be32_to_cpu(ext.magic);
238 ext.len = be32_to_cpu(ext.len);
239 offset += sizeof(ext);
240 #ifdef DEBUG_EXT
241 printf("ext.magic = 0x%x\n", ext.magic);
242 #endif
243 if (offset > end_offset || ext.len > end_offset - offset) {
244 error_setg(errp, "Header extension too large");
245 return -EINVAL;
248 switch (ext.magic) {
249 case QCOW2_EXT_MAGIC_END:
250 return 0;
252 case QCOW2_EXT_MAGIC_BACKING_FORMAT:
253 if (ext.len >= sizeof(bs->backing_format)) {
254 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32
255 " too large (>=%zu)", ext.len,
256 sizeof(bs->backing_format));
257 return 2;
259 ret = bdrv_pread(bs->file, offset, ext.len, bs->backing_format, 0);
260 if (ret < 0) {
261 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: "
262 "Could not read format name");
263 return 3;
265 bs->backing_format[ext.len] = '\0';
266 s->image_backing_format = g_strdup(bs->backing_format);
267 #ifdef DEBUG_EXT
268 printf("Qcow2: Got format extension %s\n", bs->backing_format);
269 #endif
270 break;
272 case QCOW2_EXT_MAGIC_FEATURE_TABLE:
273 if (p_feature_table != NULL) {
274 void *feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
275 ret = bdrv_pread(bs->file, offset, ext.len, feature_table, 0);
276 if (ret < 0) {
277 error_setg_errno(errp, -ret, "ERROR: ext_feature_table: "
278 "Could not read table");
279 g_free(feature_table);
280 return ret;
283 *p_feature_table = feature_table;
285 break;
287 case QCOW2_EXT_MAGIC_CRYPTO_HEADER: {
288 unsigned int cflags = 0;
289 if (s->crypt_method_header != QCOW_CRYPT_LUKS) {
290 error_setg(errp, "CRYPTO header extension only "
291 "expected with LUKS encryption method");
292 return -EINVAL;
294 if (ext.len != sizeof(Qcow2CryptoHeaderExtension)) {
295 error_setg(errp, "CRYPTO header extension size %u, "
296 "but expected size %zu", ext.len,
297 sizeof(Qcow2CryptoHeaderExtension));
298 return -EINVAL;
301 ret = bdrv_pread(bs->file, offset, ext.len, &s->crypto_header, 0);
302 if (ret < 0) {
303 error_setg_errno(errp, -ret,
304 "Unable to read CRYPTO header extension");
305 return ret;
307 s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset);
308 s->crypto_header.length = be64_to_cpu(s->crypto_header.length);
310 if ((s->crypto_header.offset % s->cluster_size) != 0) {
311 error_setg(errp, "Encryption header offset '%" PRIu64 "' is "
312 "not a multiple of cluster size '%u'",
313 s->crypto_header.offset, s->cluster_size);
314 return -EINVAL;
317 if (flags & BDRV_O_NO_IO) {
318 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
320 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
321 qcow2_crypto_hdr_read_func,
322 bs, cflags, QCOW2_MAX_THREADS, errp);
323 if (!s->crypto) {
324 return -EINVAL;
326 } break;
328 case QCOW2_EXT_MAGIC_BITMAPS:
329 if (ext.len != sizeof(bitmaps_ext)) {
330 error_setg_errno(errp, -ret, "bitmaps_ext: "
331 "Invalid extension length");
332 return -EINVAL;
335 if (!(s->autoclear_features & QCOW2_AUTOCLEAR_BITMAPS)) {
336 if (s->qcow_version < 3) {
337 /* Let's be a bit more specific */
338 warn_report("This qcow2 v2 image contains bitmaps, but "
339 "they may have been modified by a program "
340 "without persistent bitmap support; so now "
341 "they must all be considered inconsistent");
342 } else {
343 warn_report("a program lacking bitmap support "
344 "modified this file, so all bitmaps are now "
345 "considered inconsistent");
347 error_printf("Some clusters may be leaked, "
348 "run 'qemu-img check -r' on the image "
349 "file to fix.");
350 if (need_update_header != NULL) {
351 /* Updating is needed to drop invalid bitmap extension. */
352 *need_update_header = true;
354 break;
357 ret = bdrv_pread(bs->file, offset, ext.len, &bitmaps_ext, 0);
358 if (ret < 0) {
359 error_setg_errno(errp, -ret, "bitmaps_ext: "
360 "Could not read ext header");
361 return ret;
364 if (bitmaps_ext.reserved32 != 0) {
365 error_setg_errno(errp, -ret, "bitmaps_ext: "
366 "Reserved field is not zero");
367 return -EINVAL;
370 bitmaps_ext.nb_bitmaps = be32_to_cpu(bitmaps_ext.nb_bitmaps);
371 bitmaps_ext.bitmap_directory_size =
372 be64_to_cpu(bitmaps_ext.bitmap_directory_size);
373 bitmaps_ext.bitmap_directory_offset =
374 be64_to_cpu(bitmaps_ext.bitmap_directory_offset);
376 if (bitmaps_ext.nb_bitmaps > QCOW2_MAX_BITMAPS) {
377 error_setg(errp,
378 "bitmaps_ext: Image has %" PRIu32 " bitmaps, "
379 "exceeding the QEMU supported maximum of %d",
380 bitmaps_ext.nb_bitmaps, QCOW2_MAX_BITMAPS);
381 return -EINVAL;
384 if (bitmaps_ext.nb_bitmaps == 0) {
385 error_setg(errp, "found bitmaps extension with zero bitmaps");
386 return -EINVAL;
389 if (offset_into_cluster(s, bitmaps_ext.bitmap_directory_offset)) {
390 error_setg(errp, "bitmaps_ext: "
391 "invalid bitmap directory offset");
392 return -EINVAL;
395 if (bitmaps_ext.bitmap_directory_size >
396 QCOW2_MAX_BITMAP_DIRECTORY_SIZE) {
397 error_setg(errp, "bitmaps_ext: "
398 "bitmap directory size (%" PRIu64 ") exceeds "
399 "the maximum supported size (%d)",
400 bitmaps_ext.bitmap_directory_size,
401 QCOW2_MAX_BITMAP_DIRECTORY_SIZE);
402 return -EINVAL;
405 s->nb_bitmaps = bitmaps_ext.nb_bitmaps;
406 s->bitmap_directory_offset =
407 bitmaps_ext.bitmap_directory_offset;
408 s->bitmap_directory_size =
409 bitmaps_ext.bitmap_directory_size;
411 #ifdef DEBUG_EXT
412 printf("Qcow2: Got bitmaps extension: "
413 "offset=%" PRIu64 " nb_bitmaps=%" PRIu32 "\n",
414 s->bitmap_directory_offset, s->nb_bitmaps);
415 #endif
416 break;
418 case QCOW2_EXT_MAGIC_DATA_FILE:
420 s->image_data_file = g_malloc0(ext.len + 1);
421 ret = bdrv_pread(bs->file, offset, ext.len, s->image_data_file, 0);
422 if (ret < 0) {
423 error_setg_errno(errp, -ret,
424 "ERROR: Could not read data file name");
425 return ret;
427 #ifdef DEBUG_EXT
428 printf("Qcow2: Got external data file %s\n", s->image_data_file);
429 #endif
430 break;
433 default:
434 /* unknown magic - save it in case we need to rewrite the header */
435 /* If you add a new feature, make sure to also update the fast
436 * path of qcow2_make_empty() to deal with it. */
438 Qcow2UnknownHeaderExtension *uext;
440 uext = g_malloc0(sizeof(*uext) + ext.len);
441 uext->magic = ext.magic;
442 uext->len = ext.len;
443 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
445 ret = bdrv_pread(bs->file, offset, uext->len, uext->data, 0);
446 if (ret < 0) {
447 error_setg_errno(errp, -ret, "ERROR: unknown extension: "
448 "Could not read data");
449 return ret;
452 break;
455 offset += ((ext.len + 7) & ~7);
458 return 0;
461 static void cleanup_unknown_header_ext(BlockDriverState *bs)
463 BDRVQcow2State *s = bs->opaque;
464 Qcow2UnknownHeaderExtension *uext, *next;
466 QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
467 QLIST_REMOVE(uext, next);
468 g_free(uext);
472 static void report_unsupported_feature(Error **errp, Qcow2Feature *table,
473 uint64_t mask)
475 g_autoptr(GString) features = g_string_sized_new(60);
477 while (table && table->name[0] != '\0') {
478 if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
479 if (mask & (1ULL << table->bit)) {
480 if (features->len > 0) {
481 g_string_append(features, ", ");
483 g_string_append_printf(features, "%.46s", table->name);
484 mask &= ~(1ULL << table->bit);
487 table++;
490 if (mask) {
491 if (features->len > 0) {
492 g_string_append(features, ", ");
494 g_string_append_printf(features,
495 "Unknown incompatible feature: %" PRIx64, mask);
498 error_setg(errp, "Unsupported qcow2 feature(s): %s", features->str);
502 * Sets the dirty bit and flushes afterwards if necessary.
504 * The incompatible_features bit is only set if the image file header was
505 * updated successfully. Therefore it is not required to check the return
506 * value of this function.
508 int qcow2_mark_dirty(BlockDriverState *bs)
510 BDRVQcow2State *s = bs->opaque;
511 uint64_t val;
512 int ret;
514 assert(s->qcow_version >= 3);
516 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
517 return 0; /* already dirty */
520 val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY);
521 ret = bdrv_pwrite_sync(bs->file,
522 offsetof(QCowHeader, incompatible_features),
523 sizeof(val), &val, 0);
524 if (ret < 0) {
525 return ret;
528 /* Only treat image as dirty if the header was updated successfully */
529 s->incompatible_features |= QCOW2_INCOMPAT_DIRTY;
530 return 0;
534 * Clears the dirty bit and flushes before if necessary. Only call this
535 * function when there are no pending requests, it does not guard against
536 * concurrent requests dirtying the image.
538 static int qcow2_mark_clean(BlockDriverState *bs)
540 BDRVQcow2State *s = bs->opaque;
542 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
543 int ret;
545 s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY;
547 ret = qcow2_flush_caches(bs);
548 if (ret < 0) {
549 return ret;
552 return qcow2_update_header(bs);
554 return 0;
558 * Marks the image as corrupt.
560 int qcow2_mark_corrupt(BlockDriverState *bs)
562 BDRVQcow2State *s = bs->opaque;
564 s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT;
565 return qcow2_update_header(bs);
569 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
570 * before if necessary.
572 int qcow2_mark_consistent(BlockDriverState *bs)
574 BDRVQcow2State *s = bs->opaque;
576 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
577 int ret = qcow2_flush_caches(bs);
578 if (ret < 0) {
579 return ret;
582 s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT;
583 return qcow2_update_header(bs);
585 return 0;
588 static void qcow2_add_check_result(BdrvCheckResult *out,
589 const BdrvCheckResult *src,
590 bool set_allocation_info)
592 out->corruptions += src->corruptions;
593 out->leaks += src->leaks;
594 out->check_errors += src->check_errors;
595 out->corruptions_fixed += src->corruptions_fixed;
596 out->leaks_fixed += src->leaks_fixed;
598 if (set_allocation_info) {
599 out->image_end_offset = src->image_end_offset;
600 out->bfi = src->bfi;
604 static int coroutine_fn GRAPH_RDLOCK
605 qcow2_co_check_locked(BlockDriverState *bs, BdrvCheckResult *result,
606 BdrvCheckMode fix)
608 BdrvCheckResult snapshot_res = {};
609 BdrvCheckResult refcount_res = {};
610 int ret;
612 memset(result, 0, sizeof(*result));
614 ret = qcow2_check_read_snapshot_table(bs, &snapshot_res, fix);
615 if (ret < 0) {
616 qcow2_add_check_result(result, &snapshot_res, false);
617 return ret;
620 ret = qcow2_check_refcounts(bs, &refcount_res, fix);
621 qcow2_add_check_result(result, &refcount_res, true);
622 if (ret < 0) {
623 qcow2_add_check_result(result, &snapshot_res, false);
624 return ret;
627 ret = qcow2_check_fix_snapshot_table(bs, &snapshot_res, fix);
628 qcow2_add_check_result(result, &snapshot_res, false);
629 if (ret < 0) {
630 return ret;
633 if (fix && result->check_errors == 0 && result->corruptions == 0) {
634 ret = qcow2_mark_clean(bs);
635 if (ret < 0) {
636 return ret;
638 return qcow2_mark_consistent(bs);
640 return ret;
643 static int coroutine_fn GRAPH_RDLOCK
644 qcow2_co_check(BlockDriverState *bs, BdrvCheckResult *result,
645 BdrvCheckMode fix)
647 BDRVQcow2State *s = bs->opaque;
648 int ret;
650 qemu_co_mutex_lock(&s->lock);
651 ret = qcow2_co_check_locked(bs, result, fix);
652 qemu_co_mutex_unlock(&s->lock);
653 return ret;
656 int qcow2_validate_table(BlockDriverState *bs, uint64_t offset,
657 uint64_t entries, size_t entry_len,
658 int64_t max_size_bytes, const char *table_name,
659 Error **errp)
661 BDRVQcow2State *s = bs->opaque;
663 if (entries > max_size_bytes / entry_len) {
664 error_setg(errp, "%s too large", table_name);
665 return -EFBIG;
668 /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
669 * because values will be passed to qemu functions taking int64_t. */
670 if ((INT64_MAX - entries * entry_len < offset) ||
671 (offset_into_cluster(s, offset) != 0)) {
672 error_setg(errp, "%s offset invalid", table_name);
673 return -EINVAL;
676 return 0;
679 static const char *const mutable_opts[] = {
680 QCOW2_OPT_LAZY_REFCOUNTS,
681 QCOW2_OPT_DISCARD_REQUEST,
682 QCOW2_OPT_DISCARD_SNAPSHOT,
683 QCOW2_OPT_DISCARD_OTHER,
684 QCOW2_OPT_OVERLAP,
685 QCOW2_OPT_OVERLAP_TEMPLATE,
686 QCOW2_OPT_OVERLAP_MAIN_HEADER,
687 QCOW2_OPT_OVERLAP_ACTIVE_L1,
688 QCOW2_OPT_OVERLAP_ACTIVE_L2,
689 QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
690 QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
691 QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
692 QCOW2_OPT_OVERLAP_INACTIVE_L1,
693 QCOW2_OPT_OVERLAP_INACTIVE_L2,
694 QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY,
695 QCOW2_OPT_CACHE_SIZE,
696 QCOW2_OPT_L2_CACHE_SIZE,
697 QCOW2_OPT_L2_CACHE_ENTRY_SIZE,
698 QCOW2_OPT_REFCOUNT_CACHE_SIZE,
699 QCOW2_OPT_CACHE_CLEAN_INTERVAL,
700 NULL
703 static QemuOptsList qcow2_runtime_opts = {
704 .name = "qcow2",
705 .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head),
706 .desc = {
708 .name = QCOW2_OPT_LAZY_REFCOUNTS,
709 .type = QEMU_OPT_BOOL,
710 .help = "Postpone refcount updates",
713 .name = QCOW2_OPT_DISCARD_REQUEST,
714 .type = QEMU_OPT_BOOL,
715 .help = "Pass guest discard requests to the layer below",
718 .name = QCOW2_OPT_DISCARD_SNAPSHOT,
719 .type = QEMU_OPT_BOOL,
720 .help = "Generate discard requests when snapshot related space "
721 "is freed",
724 .name = QCOW2_OPT_DISCARD_OTHER,
725 .type = QEMU_OPT_BOOL,
726 .help = "Generate discard requests when other clusters are freed",
729 .name = QCOW2_OPT_OVERLAP,
730 .type = QEMU_OPT_STRING,
731 .help = "Selects which overlap checks to perform from a range of "
732 "templates (none, constant, cached, all)",
735 .name = QCOW2_OPT_OVERLAP_TEMPLATE,
736 .type = QEMU_OPT_STRING,
737 .help = "Selects which overlap checks to perform from a range of "
738 "templates (none, constant, cached, all)",
741 .name = QCOW2_OPT_OVERLAP_MAIN_HEADER,
742 .type = QEMU_OPT_BOOL,
743 .help = "Check for unintended writes into the main qcow2 header",
746 .name = QCOW2_OPT_OVERLAP_ACTIVE_L1,
747 .type = QEMU_OPT_BOOL,
748 .help = "Check for unintended writes into the active L1 table",
751 .name = QCOW2_OPT_OVERLAP_ACTIVE_L2,
752 .type = QEMU_OPT_BOOL,
753 .help = "Check for unintended writes into an active L2 table",
756 .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
757 .type = QEMU_OPT_BOOL,
758 .help = "Check for unintended writes into the refcount table",
761 .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
762 .type = QEMU_OPT_BOOL,
763 .help = "Check for unintended writes into a refcount block",
766 .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
767 .type = QEMU_OPT_BOOL,
768 .help = "Check for unintended writes into the snapshot table",
771 .name = QCOW2_OPT_OVERLAP_INACTIVE_L1,
772 .type = QEMU_OPT_BOOL,
773 .help = "Check for unintended writes into an inactive L1 table",
776 .name = QCOW2_OPT_OVERLAP_INACTIVE_L2,
777 .type = QEMU_OPT_BOOL,
778 .help = "Check for unintended writes into an inactive L2 table",
781 .name = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY,
782 .type = QEMU_OPT_BOOL,
783 .help = "Check for unintended writes into the bitmap directory",
786 .name = QCOW2_OPT_CACHE_SIZE,
787 .type = QEMU_OPT_SIZE,
788 .help = "Maximum combined metadata (L2 tables and refcount blocks) "
789 "cache size",
792 .name = QCOW2_OPT_L2_CACHE_SIZE,
793 .type = QEMU_OPT_SIZE,
794 .help = "Maximum L2 table cache size",
797 .name = QCOW2_OPT_L2_CACHE_ENTRY_SIZE,
798 .type = QEMU_OPT_SIZE,
799 .help = "Size of each entry in the L2 cache",
802 .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE,
803 .type = QEMU_OPT_SIZE,
804 .help = "Maximum refcount block cache size",
807 .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL,
808 .type = QEMU_OPT_NUMBER,
809 .help = "Clean unused cache entries after this time (in seconds)",
811 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
812 "ID of secret providing qcow2 AES key or LUKS passphrase"),
813 { /* end of list */ }
817 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = {
818 [QCOW2_OL_MAIN_HEADER_BITNR] = QCOW2_OPT_OVERLAP_MAIN_HEADER,
819 [QCOW2_OL_ACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L1,
820 [QCOW2_OL_ACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L2,
821 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
822 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
823 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
824 [QCOW2_OL_INACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L1,
825 [QCOW2_OL_INACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L2,
826 [QCOW2_OL_BITMAP_DIRECTORY_BITNR] = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY,
829 static void cache_clean_timer_cb(void *opaque)
831 BlockDriverState *bs = opaque;
832 BDRVQcow2State *s = bs->opaque;
833 qcow2_cache_clean_unused(s->l2_table_cache);
834 qcow2_cache_clean_unused(s->refcount_block_cache);
835 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
836 (int64_t) s->cache_clean_interval * 1000);
839 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context)
841 BDRVQcow2State *s = bs->opaque;
842 if (s->cache_clean_interval > 0) {
843 s->cache_clean_timer =
844 aio_timer_new_with_attrs(context, QEMU_CLOCK_VIRTUAL,
845 SCALE_MS, QEMU_TIMER_ATTR_EXTERNAL,
846 cache_clean_timer_cb, bs);
847 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
848 (int64_t) s->cache_clean_interval * 1000);
852 static void cache_clean_timer_del(BlockDriverState *bs)
854 BDRVQcow2State *s = bs->opaque;
855 if (s->cache_clean_timer) {
856 timer_free(s->cache_clean_timer);
857 s->cache_clean_timer = NULL;
861 static void qcow2_detach_aio_context(BlockDriverState *bs)
863 cache_clean_timer_del(bs);
866 static void qcow2_attach_aio_context(BlockDriverState *bs,
867 AioContext *new_context)
869 cache_clean_timer_init(bs, new_context);
872 static bool read_cache_sizes(BlockDriverState *bs, QemuOpts *opts,
873 uint64_t *l2_cache_size,
874 uint64_t *l2_cache_entry_size,
875 uint64_t *refcount_cache_size, Error **errp)
877 BDRVQcow2State *s = bs->opaque;
878 uint64_t combined_cache_size, l2_cache_max_setting;
879 bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set;
880 bool l2_cache_entry_size_set;
881 int min_refcount_cache = MIN_REFCOUNT_CACHE_SIZE * s->cluster_size;
882 uint64_t virtual_disk_size = bs->total_sectors * BDRV_SECTOR_SIZE;
883 uint64_t max_l2_entries = DIV_ROUND_UP(virtual_disk_size, s->cluster_size);
884 /* An L2 table is always one cluster in size so the max cache size
885 * should be a multiple of the cluster size. */
886 uint64_t max_l2_cache = ROUND_UP(max_l2_entries * l2_entry_size(s),
887 s->cluster_size);
889 combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE);
890 l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE);
891 refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
892 l2_cache_entry_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE);
894 combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0);
895 l2_cache_max_setting = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE,
896 DEFAULT_L2_CACHE_MAX_SIZE);
897 *refcount_cache_size = qemu_opt_get_size(opts,
898 QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0);
900 *l2_cache_entry_size = qemu_opt_get_size(
901 opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE, s->cluster_size);
903 *l2_cache_size = MIN(max_l2_cache, l2_cache_max_setting);
905 if (combined_cache_size_set) {
906 if (l2_cache_size_set && refcount_cache_size_set) {
907 error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE
908 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set "
909 "at the same time");
910 return false;
911 } else if (l2_cache_size_set &&
912 (l2_cache_max_setting > combined_cache_size)) {
913 error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed "
914 QCOW2_OPT_CACHE_SIZE);
915 return false;
916 } else if (*refcount_cache_size > combined_cache_size) {
917 error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed "
918 QCOW2_OPT_CACHE_SIZE);
919 return false;
922 if (l2_cache_size_set) {
923 *refcount_cache_size = combined_cache_size - *l2_cache_size;
924 } else if (refcount_cache_size_set) {
925 *l2_cache_size = combined_cache_size - *refcount_cache_size;
926 } else {
927 /* Assign as much memory as possible to the L2 cache, and
928 * use the remainder for the refcount cache */
929 if (combined_cache_size >= max_l2_cache + min_refcount_cache) {
930 *l2_cache_size = max_l2_cache;
931 *refcount_cache_size = combined_cache_size - *l2_cache_size;
932 } else {
933 *refcount_cache_size =
934 MIN(combined_cache_size, min_refcount_cache);
935 *l2_cache_size = combined_cache_size - *refcount_cache_size;
941 * If the L2 cache is not enough to cover the whole disk then
942 * default to 4KB entries. Smaller entries reduce the cost of
943 * loads and evictions and increase I/O performance.
945 if (*l2_cache_size < max_l2_cache && !l2_cache_entry_size_set) {
946 *l2_cache_entry_size = MIN(s->cluster_size, 4096);
949 /* l2_cache_size and refcount_cache_size are ensured to have at least
950 * their minimum values in qcow2_update_options_prepare() */
952 if (*l2_cache_entry_size < (1 << MIN_CLUSTER_BITS) ||
953 *l2_cache_entry_size > s->cluster_size ||
954 !is_power_of_2(*l2_cache_entry_size)) {
955 error_setg(errp, "L2 cache entry size must be a power of two "
956 "between %d and the cluster size (%d)",
957 1 << MIN_CLUSTER_BITS, s->cluster_size);
958 return false;
961 return true;
964 typedef struct Qcow2ReopenState {
965 Qcow2Cache *l2_table_cache;
966 Qcow2Cache *refcount_block_cache;
967 int l2_slice_size; /* Number of entries in a slice of the L2 table */
968 bool use_lazy_refcounts;
969 int overlap_check;
970 bool discard_passthrough[QCOW2_DISCARD_MAX];
971 uint64_t cache_clean_interval;
972 QCryptoBlockOpenOptions *crypto_opts; /* Disk encryption runtime options */
973 } Qcow2ReopenState;
975 static int qcow2_update_options_prepare(BlockDriverState *bs,
976 Qcow2ReopenState *r,
977 QDict *options, int flags,
978 Error **errp)
980 BDRVQcow2State *s = bs->opaque;
981 QemuOpts *opts = NULL;
982 const char *opt_overlap_check, *opt_overlap_check_template;
983 int overlap_check_template = 0;
984 uint64_t l2_cache_size, l2_cache_entry_size, refcount_cache_size;
985 int i;
986 const char *encryptfmt;
987 QDict *encryptopts = NULL;
988 int ret;
990 qdict_extract_subqdict(options, &encryptopts, "encrypt.");
991 encryptfmt = qdict_get_try_str(encryptopts, "format");
993 opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort);
994 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
995 ret = -EINVAL;
996 goto fail;
999 /* get L2 table/refcount block cache size from command line options */
1000 if (!read_cache_sizes(bs, opts, &l2_cache_size, &l2_cache_entry_size,
1001 &refcount_cache_size, errp)) {
1002 ret = -EINVAL;
1003 goto fail;
1006 l2_cache_size /= l2_cache_entry_size;
1007 if (l2_cache_size < MIN_L2_CACHE_SIZE) {
1008 l2_cache_size = MIN_L2_CACHE_SIZE;
1010 if (l2_cache_size > INT_MAX) {
1011 error_setg(errp, "L2 cache size too big");
1012 ret = -EINVAL;
1013 goto fail;
1016 refcount_cache_size /= s->cluster_size;
1017 if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) {
1018 refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE;
1020 if (refcount_cache_size > INT_MAX) {
1021 error_setg(errp, "Refcount cache size too big");
1022 ret = -EINVAL;
1023 goto fail;
1026 /* alloc new L2 table/refcount block cache, flush old one */
1027 if (s->l2_table_cache) {
1028 ret = qcow2_cache_flush(bs, s->l2_table_cache);
1029 if (ret) {
1030 error_setg_errno(errp, -ret, "Failed to flush the L2 table cache");
1031 goto fail;
1035 if (s->refcount_block_cache) {
1036 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
1037 if (ret) {
1038 error_setg_errno(errp, -ret,
1039 "Failed to flush the refcount block cache");
1040 goto fail;
1044 r->l2_slice_size = l2_cache_entry_size / l2_entry_size(s);
1045 r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size,
1046 l2_cache_entry_size);
1047 r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size,
1048 s->cluster_size);
1049 if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) {
1050 error_setg(errp, "Could not allocate metadata caches");
1051 ret = -ENOMEM;
1052 goto fail;
1055 /* New interval for cache cleanup timer */
1056 r->cache_clean_interval =
1057 qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL,
1058 DEFAULT_CACHE_CLEAN_INTERVAL);
1059 #ifndef CONFIG_LINUX
1060 if (r->cache_clean_interval != 0) {
1061 error_setg(errp, QCOW2_OPT_CACHE_CLEAN_INTERVAL
1062 " not supported on this host");
1063 ret = -EINVAL;
1064 goto fail;
1066 #endif
1067 if (r->cache_clean_interval > UINT_MAX) {
1068 error_setg(errp, "Cache clean interval too big");
1069 ret = -EINVAL;
1070 goto fail;
1073 /* lazy-refcounts; flush if going from enabled to disabled */
1074 r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS,
1075 (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS));
1076 if (r->use_lazy_refcounts && s->qcow_version < 3) {
1077 error_setg(errp, "Lazy refcounts require a qcow2 image with at least "
1078 "qemu 1.1 compatibility level");
1079 ret = -EINVAL;
1080 goto fail;
1083 if (s->use_lazy_refcounts && !r->use_lazy_refcounts) {
1084 ret = qcow2_mark_clean(bs);
1085 if (ret < 0) {
1086 error_setg_errno(errp, -ret, "Failed to disable lazy refcounts");
1087 goto fail;
1091 /* Overlap check options */
1092 opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP);
1093 opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE);
1094 if (opt_overlap_check_template && opt_overlap_check &&
1095 strcmp(opt_overlap_check_template, opt_overlap_check))
1097 error_setg(errp, "Conflicting values for qcow2 options '"
1098 QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
1099 "' ('%s')", opt_overlap_check, opt_overlap_check_template);
1100 ret = -EINVAL;
1101 goto fail;
1103 if (!opt_overlap_check) {
1104 opt_overlap_check = opt_overlap_check_template ?: "cached";
1107 if (!strcmp(opt_overlap_check, "none")) {
1108 overlap_check_template = 0;
1109 } else if (!strcmp(opt_overlap_check, "constant")) {
1110 overlap_check_template = QCOW2_OL_CONSTANT;
1111 } else if (!strcmp(opt_overlap_check, "cached")) {
1112 overlap_check_template = QCOW2_OL_CACHED;
1113 } else if (!strcmp(opt_overlap_check, "all")) {
1114 overlap_check_template = QCOW2_OL_ALL;
1115 } else {
1116 error_setg(errp, "Unsupported value '%s' for qcow2 option "
1117 "'overlap-check'. Allowed are any of the following: "
1118 "none, constant, cached, all", opt_overlap_check);
1119 ret = -EINVAL;
1120 goto fail;
1123 r->overlap_check = 0;
1124 for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) {
1125 /* overlap-check defines a template bitmask, but every flag may be
1126 * overwritten through the associated boolean option */
1127 r->overlap_check |=
1128 qemu_opt_get_bool(opts, overlap_bool_option_names[i],
1129 overlap_check_template & (1 << i)) << i;
1132 r->discard_passthrough[QCOW2_DISCARD_NEVER] = false;
1133 r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true;
1134 r->discard_passthrough[QCOW2_DISCARD_REQUEST] =
1135 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST,
1136 flags & BDRV_O_UNMAP);
1137 r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] =
1138 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true);
1139 r->discard_passthrough[QCOW2_DISCARD_OTHER] =
1140 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false);
1142 switch (s->crypt_method_header) {
1143 case QCOW_CRYPT_NONE:
1144 if (encryptfmt) {
1145 error_setg(errp, "No encryption in image header, but options "
1146 "specified format '%s'", encryptfmt);
1147 ret = -EINVAL;
1148 goto fail;
1150 break;
1152 case QCOW_CRYPT_AES:
1153 if (encryptfmt && !g_str_equal(encryptfmt, "aes")) {
1154 error_setg(errp,
1155 "Header reported 'aes' encryption format but "
1156 "options specify '%s'", encryptfmt);
1157 ret = -EINVAL;
1158 goto fail;
1160 qdict_put_str(encryptopts, "format", "qcow");
1161 r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
1162 if (!r->crypto_opts) {
1163 ret = -EINVAL;
1164 goto fail;
1166 break;
1168 case QCOW_CRYPT_LUKS:
1169 if (encryptfmt && !g_str_equal(encryptfmt, "luks")) {
1170 error_setg(errp,
1171 "Header reported 'luks' encryption format but "
1172 "options specify '%s'", encryptfmt);
1173 ret = -EINVAL;
1174 goto fail;
1176 qdict_put_str(encryptopts, "format", "luks");
1177 r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
1178 if (!r->crypto_opts) {
1179 ret = -EINVAL;
1180 goto fail;
1182 break;
1184 default:
1185 error_setg(errp, "Unsupported encryption method %d",
1186 s->crypt_method_header);
1187 ret = -EINVAL;
1188 goto fail;
1191 ret = 0;
1192 fail:
1193 qobject_unref(encryptopts);
1194 qemu_opts_del(opts);
1195 opts = NULL;
1196 return ret;
1199 static void qcow2_update_options_commit(BlockDriverState *bs,
1200 Qcow2ReopenState *r)
1202 BDRVQcow2State *s = bs->opaque;
1203 int i;
1205 if (s->l2_table_cache) {
1206 qcow2_cache_destroy(s->l2_table_cache);
1208 if (s->refcount_block_cache) {
1209 qcow2_cache_destroy(s->refcount_block_cache);
1211 s->l2_table_cache = r->l2_table_cache;
1212 s->refcount_block_cache = r->refcount_block_cache;
1213 s->l2_slice_size = r->l2_slice_size;
1215 s->overlap_check = r->overlap_check;
1216 s->use_lazy_refcounts = r->use_lazy_refcounts;
1218 for (i = 0; i < QCOW2_DISCARD_MAX; i++) {
1219 s->discard_passthrough[i] = r->discard_passthrough[i];
1222 if (s->cache_clean_interval != r->cache_clean_interval) {
1223 cache_clean_timer_del(bs);
1224 s->cache_clean_interval = r->cache_clean_interval;
1225 cache_clean_timer_init(bs, bdrv_get_aio_context(bs));
1228 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1229 s->crypto_opts = r->crypto_opts;
1232 static void qcow2_update_options_abort(BlockDriverState *bs,
1233 Qcow2ReopenState *r)
1235 if (r->l2_table_cache) {
1236 qcow2_cache_destroy(r->l2_table_cache);
1238 if (r->refcount_block_cache) {
1239 qcow2_cache_destroy(r->refcount_block_cache);
1241 qapi_free_QCryptoBlockOpenOptions(r->crypto_opts);
1244 static int qcow2_update_options(BlockDriverState *bs, QDict *options,
1245 int flags, Error **errp)
1247 Qcow2ReopenState r = {};
1248 int ret;
1250 ret = qcow2_update_options_prepare(bs, &r, options, flags, errp);
1251 if (ret >= 0) {
1252 qcow2_update_options_commit(bs, &r);
1253 } else {
1254 qcow2_update_options_abort(bs, &r);
1257 return ret;
1260 static int validate_compression_type(BDRVQcow2State *s, Error **errp)
1262 switch (s->compression_type) {
1263 case QCOW2_COMPRESSION_TYPE_ZLIB:
1264 #ifdef CONFIG_ZSTD
1265 case QCOW2_COMPRESSION_TYPE_ZSTD:
1266 #endif
1267 break;
1269 default:
1270 error_setg(errp, "qcow2: unknown compression type: %u",
1271 s->compression_type);
1272 return -ENOTSUP;
1276 * if the compression type differs from QCOW2_COMPRESSION_TYPE_ZLIB
1277 * the incompatible feature flag must be set
1279 if (s->compression_type == QCOW2_COMPRESSION_TYPE_ZLIB) {
1280 if (s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION) {
1281 error_setg(errp, "qcow2: Compression type incompatible feature "
1282 "bit must not be set");
1283 return -EINVAL;
1285 } else {
1286 if (!(s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION)) {
1287 error_setg(errp, "qcow2: Compression type incompatible feature "
1288 "bit must be set");
1289 return -EINVAL;
1293 return 0;
1296 /* Called with s->lock held. */
1297 static int coroutine_fn GRAPH_RDLOCK
1298 qcow2_do_open(BlockDriverState *bs, QDict *options, int flags,
1299 bool open_data_file, Error **errp)
1301 ERRP_GUARD();
1302 BDRVQcow2State *s = bs->opaque;
1303 unsigned int len, i;
1304 int ret = 0;
1305 QCowHeader header;
1306 uint64_t ext_end;
1307 uint64_t l1_vm_state_index;
1308 bool update_header = false;
1310 ret = bdrv_co_pread(bs->file, 0, sizeof(header), &header, 0);
1311 if (ret < 0) {
1312 error_setg_errno(errp, -ret, "Could not read qcow2 header");
1313 goto fail;
1315 header.magic = be32_to_cpu(header.magic);
1316 header.version = be32_to_cpu(header.version);
1317 header.backing_file_offset = be64_to_cpu(header.backing_file_offset);
1318 header.backing_file_size = be32_to_cpu(header.backing_file_size);
1319 header.size = be64_to_cpu(header.size);
1320 header.cluster_bits = be32_to_cpu(header.cluster_bits);
1321 header.crypt_method = be32_to_cpu(header.crypt_method);
1322 header.l1_table_offset = be64_to_cpu(header.l1_table_offset);
1323 header.l1_size = be32_to_cpu(header.l1_size);
1324 header.refcount_table_offset = be64_to_cpu(header.refcount_table_offset);
1325 header.refcount_table_clusters =
1326 be32_to_cpu(header.refcount_table_clusters);
1327 header.snapshots_offset = be64_to_cpu(header.snapshots_offset);
1328 header.nb_snapshots = be32_to_cpu(header.nb_snapshots);
1330 if (header.magic != QCOW_MAGIC) {
1331 error_setg(errp, "Image is not in qcow2 format");
1332 ret = -EINVAL;
1333 goto fail;
1335 if (header.version < 2 || header.version > 3) {
1336 error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version);
1337 ret = -ENOTSUP;
1338 goto fail;
1341 s->qcow_version = header.version;
1343 /* Initialise cluster size */
1344 if (header.cluster_bits < MIN_CLUSTER_BITS ||
1345 header.cluster_bits > MAX_CLUSTER_BITS) {
1346 error_setg(errp, "Unsupported cluster size: 2^%" PRIu32,
1347 header.cluster_bits);
1348 ret = -EINVAL;
1349 goto fail;
1352 s->cluster_bits = header.cluster_bits;
1353 s->cluster_size = 1 << s->cluster_bits;
1355 /* Initialise version 3 header fields */
1356 if (header.version == 2) {
1357 header.incompatible_features = 0;
1358 header.compatible_features = 0;
1359 header.autoclear_features = 0;
1360 header.refcount_order = 4;
1361 header.header_length = 72;
1362 } else {
1363 header.incompatible_features =
1364 be64_to_cpu(header.incompatible_features);
1365 header.compatible_features = be64_to_cpu(header.compatible_features);
1366 header.autoclear_features = be64_to_cpu(header.autoclear_features);
1367 header.refcount_order = be32_to_cpu(header.refcount_order);
1368 header.header_length = be32_to_cpu(header.header_length);
1370 if (header.header_length < 104) {
1371 error_setg(errp, "qcow2 header too short");
1372 ret = -EINVAL;
1373 goto fail;
1377 if (header.header_length > s->cluster_size) {
1378 error_setg(errp, "qcow2 header exceeds cluster size");
1379 ret = -EINVAL;
1380 goto fail;
1383 if (header.header_length > sizeof(header)) {
1384 s->unknown_header_fields_size = header.header_length - sizeof(header);
1385 s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
1386 ret = bdrv_co_pread(bs->file, sizeof(header),
1387 s->unknown_header_fields_size,
1388 s->unknown_header_fields, 0);
1389 if (ret < 0) {
1390 error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
1391 "fields");
1392 goto fail;
1396 if (header.backing_file_offset > s->cluster_size) {
1397 error_setg(errp, "Invalid backing file offset");
1398 ret = -EINVAL;
1399 goto fail;
1402 if (header.backing_file_offset) {
1403 ext_end = header.backing_file_offset;
1404 } else {
1405 ext_end = 1 << header.cluster_bits;
1408 /* Handle feature bits */
1409 s->incompatible_features = header.incompatible_features;
1410 s->compatible_features = header.compatible_features;
1411 s->autoclear_features = header.autoclear_features;
1414 * Handle compression type
1415 * Older qcow2 images don't contain the compression type header.
1416 * Distinguish them by the header length and use
1417 * the only valid (default) compression type in that case
1419 if (header.header_length > offsetof(QCowHeader, compression_type)) {
1420 s->compression_type = header.compression_type;
1421 } else {
1422 s->compression_type = QCOW2_COMPRESSION_TYPE_ZLIB;
1425 ret = validate_compression_type(s, errp);
1426 if (ret) {
1427 goto fail;
1430 if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
1431 void *feature_table = NULL;
1432 qcow2_read_extensions(bs, header.header_length, ext_end,
1433 &feature_table, flags, NULL, NULL);
1434 report_unsupported_feature(errp, feature_table,
1435 s->incompatible_features &
1436 ~QCOW2_INCOMPAT_MASK);
1437 ret = -ENOTSUP;
1438 g_free(feature_table);
1439 goto fail;
1442 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
1443 /* Corrupt images may not be written to unless they are being repaired
1445 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
1446 error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
1447 "read/write");
1448 ret = -EACCES;
1449 goto fail;
1453 s->subclusters_per_cluster =
1454 has_subclusters(s) ? QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER : 1;
1455 s->subcluster_size = s->cluster_size / s->subclusters_per_cluster;
1456 s->subcluster_bits = ctz32(s->subcluster_size);
1458 if (s->subcluster_size < (1 << MIN_CLUSTER_BITS)) {
1459 error_setg(errp, "Unsupported subcluster size: %d", s->subcluster_size);
1460 ret = -EINVAL;
1461 goto fail;
1464 /* Check support for various header values */
1465 if (header.refcount_order > 6) {
1466 error_setg(errp, "Reference count entry width too large; may not "
1467 "exceed 64 bits");
1468 ret = -EINVAL;
1469 goto fail;
1471 s->refcount_order = header.refcount_order;
1472 s->refcount_bits = 1 << s->refcount_order;
1473 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
1474 s->refcount_max += s->refcount_max - 1;
1476 s->crypt_method_header = header.crypt_method;
1477 if (s->crypt_method_header) {
1478 if (bdrv_uses_whitelist() &&
1479 s->crypt_method_header == QCOW_CRYPT_AES) {
1480 error_setg(errp,
1481 "Use of AES-CBC encrypted qcow2 images is no longer "
1482 "supported in system emulators");
1483 error_append_hint(errp,
1484 "You can use 'qemu-img convert' to convert your "
1485 "image to an alternative supported format, such "
1486 "as unencrypted qcow2, or raw with the LUKS "
1487 "format instead.\n");
1488 ret = -ENOSYS;
1489 goto fail;
1492 if (s->crypt_method_header == QCOW_CRYPT_AES) {
1493 s->crypt_physical_offset = false;
1494 } else {
1495 /* Assuming LUKS and any future crypt methods we
1496 * add will all use physical offsets, due to the
1497 * fact that the alternative is insecure... */
1498 s->crypt_physical_offset = true;
1501 bs->encrypted = true;
1504 s->l2_bits = s->cluster_bits - ctz32(l2_entry_size(s));
1505 s->l2_size = 1 << s->l2_bits;
1506 /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1507 s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
1508 s->refcount_block_size = 1 << s->refcount_block_bits;
1509 bs->total_sectors = header.size / BDRV_SECTOR_SIZE;
1510 s->csize_shift = (62 - (s->cluster_bits - 8));
1511 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
1512 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
1514 s->refcount_table_offset = header.refcount_table_offset;
1515 s->refcount_table_size =
1516 header.refcount_table_clusters << (s->cluster_bits - 3);
1518 if (header.refcount_table_clusters == 0 && !(flags & BDRV_O_CHECK)) {
1519 error_setg(errp, "Image does not contain a reference count table");
1520 ret = -EINVAL;
1521 goto fail;
1524 ret = qcow2_validate_table(bs, s->refcount_table_offset,
1525 header.refcount_table_clusters,
1526 s->cluster_size, QCOW_MAX_REFTABLE_SIZE,
1527 "Reference count table", errp);
1528 if (ret < 0) {
1529 goto fail;
1532 if (!(flags & BDRV_O_CHECK)) {
1534 * The total size in bytes of the snapshot table is checked in
1535 * qcow2_read_snapshots() because the size of each snapshot is
1536 * variable and we don't know it yet.
1537 * Here we only check the offset and number of snapshots.
1539 ret = qcow2_validate_table(bs, header.snapshots_offset,
1540 header.nb_snapshots,
1541 sizeof(QCowSnapshotHeader),
1542 sizeof(QCowSnapshotHeader) *
1543 QCOW_MAX_SNAPSHOTS,
1544 "Snapshot table", errp);
1545 if (ret < 0) {
1546 goto fail;
1550 /* read the level 1 table */
1551 ret = qcow2_validate_table(bs, header.l1_table_offset,
1552 header.l1_size, L1E_SIZE,
1553 QCOW_MAX_L1_SIZE, "Active L1 table", errp);
1554 if (ret < 0) {
1555 goto fail;
1557 s->l1_size = header.l1_size;
1558 s->l1_table_offset = header.l1_table_offset;
1560 l1_vm_state_index = size_to_l1(s, header.size);
1561 if (l1_vm_state_index > INT_MAX) {
1562 error_setg(errp, "Image is too big");
1563 ret = -EFBIG;
1564 goto fail;
1566 s->l1_vm_state_index = l1_vm_state_index;
1568 /* the L1 table must contain at least enough entries to put
1569 header.size bytes */
1570 if (s->l1_size < s->l1_vm_state_index) {
1571 error_setg(errp, "L1 table is too small");
1572 ret = -EINVAL;
1573 goto fail;
1576 if (s->l1_size > 0) {
1577 s->l1_table = qemu_try_blockalign(bs->file->bs, s->l1_size * L1E_SIZE);
1578 if (s->l1_table == NULL) {
1579 error_setg(errp, "Could not allocate L1 table");
1580 ret = -ENOMEM;
1581 goto fail;
1583 ret = bdrv_co_pread(bs->file, s->l1_table_offset, s->l1_size * L1E_SIZE,
1584 s->l1_table, 0);
1585 if (ret < 0) {
1586 error_setg_errno(errp, -ret, "Could not read L1 table");
1587 goto fail;
1589 for(i = 0;i < s->l1_size; i++) {
1590 s->l1_table[i] = be64_to_cpu(s->l1_table[i]);
1594 /* Parse driver-specific options */
1595 ret = qcow2_update_options(bs, options, flags, errp);
1596 if (ret < 0) {
1597 goto fail;
1600 s->flags = flags;
1602 ret = qcow2_refcount_init(bs);
1603 if (ret != 0) {
1604 error_setg_errno(errp, -ret, "Could not initialize refcount handling");
1605 goto fail;
1608 QLIST_INIT(&s->cluster_allocs);
1609 QTAILQ_INIT(&s->discards);
1611 /* read qcow2 extensions */
1612 if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
1613 flags, &update_header, errp)) {
1614 ret = -EINVAL;
1615 goto fail;
1618 if (open_data_file) {
1619 /* Open external data file */
1620 s->data_file = bdrv_co_open_child(NULL, options, "data-file", bs,
1621 &child_of_bds, BDRV_CHILD_DATA,
1622 true, errp);
1623 if (*errp) {
1624 ret = -EINVAL;
1625 goto fail;
1628 if (s->incompatible_features & QCOW2_INCOMPAT_DATA_FILE) {
1629 if (!s->data_file && s->image_data_file) {
1630 s->data_file = bdrv_co_open_child(s->image_data_file, options,
1631 "data-file", bs,
1632 &child_of_bds,
1633 BDRV_CHILD_DATA, false, errp);
1634 if (!s->data_file) {
1635 ret = -EINVAL;
1636 goto fail;
1639 if (!s->data_file) {
1640 error_setg(errp, "'data-file' is required for this image");
1641 ret = -EINVAL;
1642 goto fail;
1645 /* No data here */
1646 bs->file->role &= ~BDRV_CHILD_DATA;
1648 /* Must succeed because we have given up permissions if anything */
1649 bdrv_child_refresh_perms(bs, bs->file, &error_abort);
1650 } else {
1651 if (s->data_file) {
1652 error_setg(errp, "'data-file' can only be set for images with "
1653 "an external data file");
1654 ret = -EINVAL;
1655 goto fail;
1658 s->data_file = bs->file;
1660 if (data_file_is_raw(bs)) {
1661 error_setg(errp, "data-file-raw requires a data file");
1662 ret = -EINVAL;
1663 goto fail;
1668 /* qcow2_read_extension may have set up the crypto context
1669 * if the crypt method needs a header region, some methods
1670 * don't need header extensions, so must check here
1672 if (s->crypt_method_header && !s->crypto) {
1673 if (s->crypt_method_header == QCOW_CRYPT_AES) {
1674 unsigned int cflags = 0;
1675 if (flags & BDRV_O_NO_IO) {
1676 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
1678 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
1679 NULL, NULL, cflags,
1680 QCOW2_MAX_THREADS, errp);
1681 if (!s->crypto) {
1682 ret = -EINVAL;
1683 goto fail;
1685 } else if (!(flags & BDRV_O_NO_IO)) {
1686 error_setg(errp, "Missing CRYPTO header for crypt method %d",
1687 s->crypt_method_header);
1688 ret = -EINVAL;
1689 goto fail;
1693 /* read the backing file name */
1694 if (header.backing_file_offset != 0) {
1695 len = header.backing_file_size;
1696 if (len > MIN(1023, s->cluster_size - header.backing_file_offset) ||
1697 len >= sizeof(bs->backing_file)) {
1698 error_setg(errp, "Backing file name too long");
1699 ret = -EINVAL;
1700 goto fail;
1703 s->image_backing_file = g_malloc(len + 1);
1704 ret = bdrv_co_pread(bs->file, header.backing_file_offset, len,
1705 s->image_backing_file, 0);
1706 if (ret < 0) {
1707 error_setg_errno(errp, -ret, "Could not read backing file name");
1708 goto fail;
1710 s->image_backing_file[len] = '\0';
1713 * Update only when something has changed. This function is called by
1714 * qcow2_co_invalidate_cache(), and we do not want to reset
1715 * auto_backing_file unless necessary.
1717 if (!g_str_equal(s->image_backing_file, bs->backing_file)) {
1718 pstrcpy(bs->backing_file, sizeof(bs->backing_file),
1719 s->image_backing_file);
1720 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
1721 s->image_backing_file);
1726 * Internal snapshots; skip reading them in check mode, because
1727 * we do not need them then, and we do not want to abort because
1728 * of a broken table.
1730 if (!(flags & BDRV_O_CHECK)) {
1731 s->snapshots_offset = header.snapshots_offset;
1732 s->nb_snapshots = header.nb_snapshots;
1734 ret = qcow2_read_snapshots(bs, errp);
1735 if (ret < 0) {
1736 goto fail;
1740 /* Clear unknown autoclear feature bits */
1741 update_header |= s->autoclear_features & ~QCOW2_AUTOCLEAR_MASK;
1742 update_header = update_header && bdrv_is_writable(bs);
1743 if (update_header) {
1744 s->autoclear_features &= QCOW2_AUTOCLEAR_MASK;
1747 /* == Handle persistent dirty bitmaps ==
1749 * We want load dirty bitmaps in three cases:
1751 * 1. Normal open of the disk in active mode, not related to invalidation
1752 * after migration.
1754 * 2. Invalidation of the target vm after pre-copy phase of migration, if
1755 * bitmaps are _not_ migrating through migration channel, i.e.
1756 * 'dirty-bitmaps' capability is disabled.
1758 * 3. Invalidation of source vm after failed or canceled migration.
1759 * This is a very interesting case. There are two possible types of
1760 * bitmaps:
1762 * A. Stored on inactivation and removed. They should be loaded from the
1763 * image.
1765 * B. Not stored: not-persistent bitmaps and bitmaps, migrated through
1766 * the migration channel (with dirty-bitmaps capability).
1768 * On the other hand, there are two possible sub-cases:
1770 * 3.1 disk was changed by somebody else while were inactive. In this
1771 * case all in-RAM dirty bitmaps (both persistent and not) are
1772 * definitely invalid. And we don't have any method to determine
1773 * this.
1775 * Simple and safe thing is to just drop all the bitmaps of type B on
1776 * inactivation. But in this case we lose bitmaps in valid 4.2 case.
1778 * On the other hand, resuming source vm, if disk was already changed
1779 * is a bad thing anyway: not only bitmaps, the whole vm state is
1780 * out of sync with disk.
1782 * This means, that user or management tool, who for some reason
1783 * decided to resume source vm, after disk was already changed by
1784 * target vm, should at least drop all dirty bitmaps by hand.
1786 * So, we can ignore this case for now, but TODO: "generation"
1787 * extension for qcow2, to determine, that image was changed after
1788 * last inactivation. And if it is changed, we will drop (or at least
1789 * mark as 'invalid' all the bitmaps of type B, both persistent
1790 * and not).
1792 * 3.2 disk was _not_ changed while were inactive. Bitmaps may be saved
1793 * to disk ('dirty-bitmaps' capability disabled), or not saved
1794 * ('dirty-bitmaps' capability enabled), but we don't need to care
1795 * of: let's load bitmaps as always: stored bitmaps will be loaded,
1796 * and not stored has flag IN_USE=1 in the image and will be skipped
1797 * on loading.
1799 * One remaining possible case when we don't want load bitmaps:
1801 * 4. Open disk in inactive mode in target vm (bitmaps are migrating or
1802 * will be loaded on invalidation, no needs try loading them before)
1805 if (!(bdrv_get_flags(bs) & BDRV_O_INACTIVE)) {
1806 /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */
1807 bool header_updated;
1808 if (!qcow2_load_dirty_bitmaps(bs, &header_updated, errp)) {
1809 ret = -EINVAL;
1810 goto fail;
1813 update_header = update_header && !header_updated;
1816 if (update_header) {
1817 ret = qcow2_update_header(bs);
1818 if (ret < 0) {
1819 error_setg_errno(errp, -ret, "Could not update qcow2 header");
1820 goto fail;
1824 bs->supported_zero_flags = header.version >= 3 ?
1825 BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK : 0;
1826 bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE;
1828 /* Repair image if dirty */
1829 if (!(flags & BDRV_O_CHECK) && bdrv_is_writable(bs) &&
1830 (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
1831 BdrvCheckResult result = {0};
1833 ret = qcow2_co_check_locked(bs, &result,
1834 BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
1835 if (ret < 0 || result.check_errors) {
1836 if (ret >= 0) {
1837 ret = -EIO;
1839 error_setg_errno(errp, -ret, "Could not repair dirty image");
1840 goto fail;
1844 #ifdef DEBUG_ALLOC
1846 BdrvCheckResult result = {0};
1847 qcow2_check_refcounts(bs, &result, 0);
1849 #endif
1851 qemu_co_queue_init(&s->thread_task_queue);
1853 return ret;
1855 fail:
1856 g_free(s->image_data_file);
1857 if (open_data_file && has_data_file(bs)) {
1858 bdrv_unref_child(bs, s->data_file);
1859 s->data_file = NULL;
1861 g_free(s->unknown_header_fields);
1862 cleanup_unknown_header_ext(bs);
1863 qcow2_free_snapshots(bs);
1864 qcow2_refcount_close(bs);
1865 qemu_vfree(s->l1_table);
1866 /* else pre-write overlap checks in cache_destroy may crash */
1867 s->l1_table = NULL;
1868 cache_clean_timer_del(bs);
1869 if (s->l2_table_cache) {
1870 qcow2_cache_destroy(s->l2_table_cache);
1872 if (s->refcount_block_cache) {
1873 qcow2_cache_destroy(s->refcount_block_cache);
1875 qcrypto_block_free(s->crypto);
1876 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1877 return ret;
1880 typedef struct QCow2OpenCo {
1881 BlockDriverState *bs;
1882 QDict *options;
1883 int flags;
1884 Error **errp;
1885 int ret;
1886 } QCow2OpenCo;
1888 static void coroutine_fn qcow2_open_entry(void *opaque)
1890 QCow2OpenCo *qoc = opaque;
1891 BDRVQcow2State *s = qoc->bs->opaque;
1893 assume_graph_lock(); /* FIXME */
1895 qemu_co_mutex_lock(&s->lock);
1896 qoc->ret = qcow2_do_open(qoc->bs, qoc->options, qoc->flags, true,
1897 qoc->errp);
1898 qemu_co_mutex_unlock(&s->lock);
1901 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
1902 Error **errp)
1904 BDRVQcow2State *s = bs->opaque;
1905 QCow2OpenCo qoc = {
1906 .bs = bs,
1907 .options = options,
1908 .flags = flags,
1909 .errp = errp,
1910 .ret = -EINPROGRESS
1912 int ret;
1914 ret = bdrv_open_file_child(NULL, options, "file", bs, errp);
1915 if (ret < 0) {
1916 return ret;
1919 /* Initialise locks */
1920 qemu_co_mutex_init(&s->lock);
1922 if (qemu_in_coroutine()) {
1923 /* From bdrv_co_create. */
1924 qcow2_open_entry(&qoc);
1925 } else {
1926 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
1927 qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry, &qoc));
1928 BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS);
1930 return qoc.ret;
1933 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1935 BDRVQcow2State *s = bs->opaque;
1937 if (bs->encrypted) {
1938 /* Encryption works on a sector granularity */
1939 bs->bl.request_alignment = qcrypto_block_get_sector_size(s->crypto);
1941 bs->bl.pwrite_zeroes_alignment = s->subcluster_size;
1942 bs->bl.pdiscard_alignment = s->cluster_size;
1945 static int qcow2_reopen_prepare(BDRVReopenState *state,
1946 BlockReopenQueue *queue, Error **errp)
1948 BDRVQcow2State *s = state->bs->opaque;
1949 Qcow2ReopenState *r;
1950 int ret;
1952 r = g_new0(Qcow2ReopenState, 1);
1953 state->opaque = r;
1955 ret = qcow2_update_options_prepare(state->bs, r, state->options,
1956 state->flags, errp);
1957 if (ret < 0) {
1958 goto fail;
1961 /* We need to write out any unwritten data if we reopen read-only. */
1962 if ((state->flags & BDRV_O_RDWR) == 0) {
1963 ret = qcow2_reopen_bitmaps_ro(state->bs, errp);
1964 if (ret < 0) {
1965 goto fail;
1968 ret = bdrv_flush(state->bs);
1969 if (ret < 0) {
1970 goto fail;
1973 ret = qcow2_mark_clean(state->bs);
1974 if (ret < 0) {
1975 goto fail;
1980 * Without an external data file, s->data_file points to the same BdrvChild
1981 * as bs->file. It needs to be resynced after reopen because bs->file may
1982 * be changed. We can't use it in the meantime.
1984 if (!has_data_file(state->bs)) {
1985 assert(s->data_file == state->bs->file);
1986 s->data_file = NULL;
1989 return 0;
1991 fail:
1992 qcow2_update_options_abort(state->bs, r);
1993 g_free(r);
1994 return ret;
1997 static void qcow2_reopen_commit(BDRVReopenState *state)
1999 BDRVQcow2State *s = state->bs->opaque;
2001 qcow2_update_options_commit(state->bs, state->opaque);
2002 if (!s->data_file) {
2004 * If we don't have an external data file, s->data_file was cleared by
2005 * qcow2_reopen_prepare() and needs to be updated.
2007 s->data_file = state->bs->file;
2009 g_free(state->opaque);
2012 static void qcow2_reopen_commit_post(BDRVReopenState *state)
2014 if (state->flags & BDRV_O_RDWR) {
2015 Error *local_err = NULL;
2017 if (qcow2_reopen_bitmaps_rw(state->bs, &local_err) < 0) {
2019 * This is not fatal, bitmaps just left read-only, so all following
2020 * writes will fail. User can remove read-only bitmaps to unblock
2021 * writes or retry reopen.
2023 error_reportf_err(local_err,
2024 "%s: Failed to make dirty bitmaps writable: ",
2025 bdrv_get_node_name(state->bs));
2030 static void qcow2_reopen_abort(BDRVReopenState *state)
2032 BDRVQcow2State *s = state->bs->opaque;
2034 if (!s->data_file) {
2036 * If we don't have an external data file, s->data_file was cleared by
2037 * qcow2_reopen_prepare() and needs to be restored.
2039 s->data_file = state->bs->file;
2041 qcow2_update_options_abort(state->bs, state->opaque);
2042 g_free(state->opaque);
2045 static void qcow2_join_options(QDict *options, QDict *old_options)
2047 bool has_new_overlap_template =
2048 qdict_haskey(options, QCOW2_OPT_OVERLAP) ||
2049 qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE);
2050 bool has_new_total_cache_size =
2051 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE);
2052 bool has_all_cache_options;
2054 /* New overlap template overrides all old overlap options */
2055 if (has_new_overlap_template) {
2056 qdict_del(old_options, QCOW2_OPT_OVERLAP);
2057 qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE);
2058 qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER);
2059 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1);
2060 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2);
2061 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE);
2062 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK);
2063 qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE);
2064 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1);
2065 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2);
2068 /* New total cache size overrides all old options */
2069 if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) {
2070 qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE);
2071 qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
2074 qdict_join(options, old_options, false);
2077 * If after merging all cache size options are set, an old total size is
2078 * overwritten. Do keep all options, however, if all three are new. The
2079 * resulting error message is what we want to happen.
2081 has_all_cache_options =
2082 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) ||
2083 qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) ||
2084 qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
2086 if (has_all_cache_options && !has_new_total_cache_size) {
2087 qdict_del(options, QCOW2_OPT_CACHE_SIZE);
2091 static int coroutine_fn qcow2_co_block_status(BlockDriverState *bs,
2092 bool want_zero,
2093 int64_t offset, int64_t count,
2094 int64_t *pnum, int64_t *map,
2095 BlockDriverState **file)
2097 BDRVQcow2State *s = bs->opaque;
2098 uint64_t host_offset;
2099 unsigned int bytes;
2100 QCow2SubclusterType type;
2101 int ret, status = 0;
2103 qemu_co_mutex_lock(&s->lock);
2105 if (!s->metadata_preallocation_checked) {
2106 ret = qcow2_detect_metadata_preallocation(bs);
2107 s->metadata_preallocation = (ret == 1);
2108 s->metadata_preallocation_checked = true;
2111 bytes = MIN(INT_MAX, count);
2112 ret = qcow2_get_host_offset(bs, offset, &bytes, &host_offset, &type);
2113 qemu_co_mutex_unlock(&s->lock);
2114 if (ret < 0) {
2115 return ret;
2118 *pnum = bytes;
2120 if ((type == QCOW2_SUBCLUSTER_NORMAL ||
2121 type == QCOW2_SUBCLUSTER_ZERO_ALLOC ||
2122 type == QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC) && !s->crypto) {
2123 *map = host_offset;
2124 *file = s->data_file->bs;
2125 status |= BDRV_BLOCK_OFFSET_VALID;
2127 if (type == QCOW2_SUBCLUSTER_ZERO_PLAIN ||
2128 type == QCOW2_SUBCLUSTER_ZERO_ALLOC) {
2129 status |= BDRV_BLOCK_ZERO;
2130 } else if (type != QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN &&
2131 type != QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC) {
2132 status |= BDRV_BLOCK_DATA;
2134 if (s->metadata_preallocation && (status & BDRV_BLOCK_DATA) &&
2135 (status & BDRV_BLOCK_OFFSET_VALID))
2137 status |= BDRV_BLOCK_RECURSE;
2139 return status;
2142 static int coroutine_fn GRAPH_RDLOCK
2143 qcow2_handle_l2meta(BlockDriverState *bs, QCowL2Meta **pl2meta, bool link_l2)
2145 int ret = 0;
2146 QCowL2Meta *l2meta = *pl2meta;
2148 while (l2meta != NULL) {
2149 QCowL2Meta *next;
2151 if (link_l2) {
2152 ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
2153 if (ret) {
2154 goto out;
2156 } else {
2157 qcow2_alloc_cluster_abort(bs, l2meta);
2160 /* Take the request off the list of running requests */
2161 QLIST_REMOVE(l2meta, next_in_flight);
2163 qemu_co_queue_restart_all(&l2meta->dependent_requests);
2165 next = l2meta->next;
2166 g_free(l2meta);
2167 l2meta = next;
2169 out:
2170 *pl2meta = l2meta;
2171 return ret;
2174 static int coroutine_fn GRAPH_RDLOCK
2175 qcow2_co_preadv_encrypted(BlockDriverState *bs,
2176 uint64_t host_offset,
2177 uint64_t offset,
2178 uint64_t bytes,
2179 QEMUIOVector *qiov,
2180 uint64_t qiov_offset)
2182 int ret;
2183 BDRVQcow2State *s = bs->opaque;
2184 uint8_t *buf;
2186 assert(bs->encrypted && s->crypto);
2187 assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2190 * For encrypted images, read everything into a temporary
2191 * contiguous buffer on which the AES functions can work.
2192 * Also, decryption in a separate buffer is better as it
2193 * prevents the guest from learning information about the
2194 * encrypted nature of the virtual disk.
2197 buf = qemu_try_blockalign(s->data_file->bs, bytes);
2198 if (buf == NULL) {
2199 return -ENOMEM;
2202 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
2203 ret = bdrv_co_pread(s->data_file, host_offset, bytes, buf, 0);
2204 if (ret < 0) {
2205 goto fail;
2208 if (qcow2_co_decrypt(bs, host_offset, offset, buf, bytes) < 0)
2210 ret = -EIO;
2211 goto fail;
2213 qemu_iovec_from_buf(qiov, qiov_offset, buf, bytes);
2215 fail:
2216 qemu_vfree(buf);
2218 return ret;
2221 typedef struct Qcow2AioTask {
2222 AioTask task;
2224 BlockDriverState *bs;
2225 QCow2SubclusterType subcluster_type; /* only for read */
2226 uint64_t host_offset; /* or l2_entry for compressed read */
2227 uint64_t offset;
2228 uint64_t bytes;
2229 QEMUIOVector *qiov;
2230 uint64_t qiov_offset;
2231 QCowL2Meta *l2meta; /* only for write */
2232 } Qcow2AioTask;
2234 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task);
2235 static coroutine_fn int qcow2_add_task(BlockDriverState *bs,
2236 AioTaskPool *pool,
2237 AioTaskFunc func,
2238 QCow2SubclusterType subcluster_type,
2239 uint64_t host_offset,
2240 uint64_t offset,
2241 uint64_t bytes,
2242 QEMUIOVector *qiov,
2243 size_t qiov_offset,
2244 QCowL2Meta *l2meta)
2246 Qcow2AioTask local_task;
2247 Qcow2AioTask *task = pool ? g_new(Qcow2AioTask, 1) : &local_task;
2249 *task = (Qcow2AioTask) {
2250 .task.func = func,
2251 .bs = bs,
2252 .subcluster_type = subcluster_type,
2253 .qiov = qiov,
2254 .host_offset = host_offset,
2255 .offset = offset,
2256 .bytes = bytes,
2257 .qiov_offset = qiov_offset,
2258 .l2meta = l2meta,
2261 trace_qcow2_add_task(qemu_coroutine_self(), bs, pool,
2262 func == qcow2_co_preadv_task_entry ? "read" : "write",
2263 subcluster_type, host_offset, offset, bytes,
2264 qiov, qiov_offset);
2266 if (!pool) {
2267 return func(&task->task);
2270 aio_task_pool_start_task(pool, &task->task);
2272 return 0;
2275 static int coroutine_fn GRAPH_RDLOCK
2276 qcow2_co_preadv_task(BlockDriverState *bs, QCow2SubclusterType subc_type,
2277 uint64_t host_offset, uint64_t offset, uint64_t bytes,
2278 QEMUIOVector *qiov, size_t qiov_offset)
2280 BDRVQcow2State *s = bs->opaque;
2282 switch (subc_type) {
2283 case QCOW2_SUBCLUSTER_ZERO_PLAIN:
2284 case QCOW2_SUBCLUSTER_ZERO_ALLOC:
2285 /* Both zero types are handled in qcow2_co_preadv_part */
2286 g_assert_not_reached();
2288 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN:
2289 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC:
2290 assert(bs->backing); /* otherwise handled in qcow2_co_preadv_part */
2292 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
2293 return bdrv_co_preadv_part(bs->backing, offset, bytes,
2294 qiov, qiov_offset, 0);
2296 case QCOW2_SUBCLUSTER_COMPRESSED:
2297 return qcow2_co_preadv_compressed(bs, host_offset,
2298 offset, bytes, qiov, qiov_offset);
2300 case QCOW2_SUBCLUSTER_NORMAL:
2301 if (bs->encrypted) {
2302 return qcow2_co_preadv_encrypted(bs, host_offset,
2303 offset, bytes, qiov, qiov_offset);
2306 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
2307 return bdrv_co_preadv_part(s->data_file, host_offset,
2308 bytes, qiov, qiov_offset, 0);
2310 default:
2311 g_assert_not_reached();
2314 g_assert_not_reached();
2318 * This function can count as GRAPH_RDLOCK because qcow2_co_preadv_part() holds
2319 * the graph lock and keeps it until this coroutine has terminated.
2321 static int coroutine_fn GRAPH_RDLOCK qcow2_co_preadv_task_entry(AioTask *task)
2323 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
2325 assert(!t->l2meta);
2327 return qcow2_co_preadv_task(t->bs, t->subcluster_type,
2328 t->host_offset, t->offset, t->bytes,
2329 t->qiov, t->qiov_offset);
2332 static int coroutine_fn GRAPH_RDLOCK
2333 qcow2_co_preadv_part(BlockDriverState *bs, int64_t offset, int64_t bytes,
2334 QEMUIOVector *qiov, size_t qiov_offset,
2335 BdrvRequestFlags flags)
2337 BDRVQcow2State *s = bs->opaque;
2338 int ret = 0;
2339 unsigned int cur_bytes; /* number of bytes in current iteration */
2340 uint64_t host_offset = 0;
2341 QCow2SubclusterType type;
2342 AioTaskPool *aio = NULL;
2344 while (bytes != 0 && aio_task_pool_status(aio) == 0) {
2345 /* prepare next request */
2346 cur_bytes = MIN(bytes, INT_MAX);
2347 if (s->crypto) {
2348 cur_bytes = MIN(cur_bytes,
2349 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2352 qemu_co_mutex_lock(&s->lock);
2353 ret = qcow2_get_host_offset(bs, offset, &cur_bytes,
2354 &host_offset, &type);
2355 qemu_co_mutex_unlock(&s->lock);
2356 if (ret < 0) {
2357 goto out;
2360 if (type == QCOW2_SUBCLUSTER_ZERO_PLAIN ||
2361 type == QCOW2_SUBCLUSTER_ZERO_ALLOC ||
2362 (type == QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN && !bs->backing) ||
2363 (type == QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC && !bs->backing))
2365 qemu_iovec_memset(qiov, qiov_offset, 0, cur_bytes);
2366 } else {
2367 if (!aio && cur_bytes != bytes) {
2368 aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
2370 ret = qcow2_add_task(bs, aio, qcow2_co_preadv_task_entry, type,
2371 host_offset, offset, cur_bytes,
2372 qiov, qiov_offset, NULL);
2373 if (ret < 0) {
2374 goto out;
2378 bytes -= cur_bytes;
2379 offset += cur_bytes;
2380 qiov_offset += cur_bytes;
2383 out:
2384 if (aio) {
2385 aio_task_pool_wait_all(aio);
2386 if (ret == 0) {
2387 ret = aio_task_pool_status(aio);
2389 g_free(aio);
2392 return ret;
2395 /* Check if it's possible to merge a write request with the writing of
2396 * the data from the COW regions */
2397 static bool merge_cow(uint64_t offset, unsigned bytes,
2398 QEMUIOVector *qiov, size_t qiov_offset,
2399 QCowL2Meta *l2meta)
2401 QCowL2Meta *m;
2403 for (m = l2meta; m != NULL; m = m->next) {
2404 /* If both COW regions are empty then there's nothing to merge */
2405 if (m->cow_start.nb_bytes == 0 && m->cow_end.nb_bytes == 0) {
2406 continue;
2409 /* If COW regions are handled already, skip this too */
2410 if (m->skip_cow) {
2411 continue;
2415 * The write request should start immediately after the first
2416 * COW region. This does not always happen because the area
2417 * touched by the request can be larger than the one defined
2418 * by @m (a single request can span an area consisting of a
2419 * mix of previously unallocated and allocated clusters, that
2420 * is why @l2meta is a list).
2422 if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) {
2423 /* In this case the request starts before this region */
2424 assert(offset < l2meta_cow_start(m));
2425 assert(m->cow_start.nb_bytes == 0);
2426 continue;
2429 /* The write request should end immediately before the second
2430 * COW region (see above for why it does not always happen) */
2431 if (m->offset + m->cow_end.offset != offset + bytes) {
2432 assert(offset + bytes > m->offset + m->cow_end.offset);
2433 assert(m->cow_end.nb_bytes == 0);
2434 continue;
2437 /* Make sure that adding both COW regions to the QEMUIOVector
2438 * does not exceed IOV_MAX */
2439 if (qemu_iovec_subvec_niov(qiov, qiov_offset, bytes) > IOV_MAX - 2) {
2440 continue;
2443 m->data_qiov = qiov;
2444 m->data_qiov_offset = qiov_offset;
2445 return true;
2448 return false;
2452 * Return 1 if the COW regions read as zeroes, 0 if not, < 0 on error.
2453 * Note that returning 0 does not guarantee non-zero data.
2455 static int coroutine_fn GRAPH_RDLOCK
2456 is_zero_cow(BlockDriverState *bs, QCowL2Meta *m)
2459 * This check is designed for optimization shortcut so it must be
2460 * efficient.
2461 * Instead of is_zero(), use bdrv_co_is_zero_fast() as it is
2462 * faster (but not as accurate and can result in false negatives).
2464 int ret = bdrv_co_is_zero_fast(bs, m->offset + m->cow_start.offset,
2465 m->cow_start.nb_bytes);
2466 if (ret <= 0) {
2467 return ret;
2470 return bdrv_co_is_zero_fast(bs, m->offset + m->cow_end.offset,
2471 m->cow_end.nb_bytes);
2474 static int coroutine_fn GRAPH_RDLOCK
2475 handle_alloc_space(BlockDriverState *bs, QCowL2Meta *l2meta)
2477 BDRVQcow2State *s = bs->opaque;
2478 QCowL2Meta *m;
2480 if (!(s->data_file->bs->supported_zero_flags & BDRV_REQ_NO_FALLBACK)) {
2481 return 0;
2484 if (bs->encrypted) {
2485 return 0;
2488 for (m = l2meta; m != NULL; m = m->next) {
2489 int ret;
2490 uint64_t start_offset = m->alloc_offset + m->cow_start.offset;
2491 unsigned nb_bytes = m->cow_end.offset + m->cow_end.nb_bytes -
2492 m->cow_start.offset;
2494 if (!m->cow_start.nb_bytes && !m->cow_end.nb_bytes) {
2495 continue;
2498 ret = is_zero_cow(bs, m);
2499 if (ret < 0) {
2500 return ret;
2501 } else if (ret == 0) {
2502 continue;
2506 * instead of writing zero COW buffers,
2507 * efficiently zero out the whole clusters
2510 ret = qcow2_pre_write_overlap_check(bs, 0, start_offset, nb_bytes,
2511 true);
2512 if (ret < 0) {
2513 return ret;
2516 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_SPACE);
2517 ret = bdrv_co_pwrite_zeroes(s->data_file, start_offset, nb_bytes,
2518 BDRV_REQ_NO_FALLBACK);
2519 if (ret < 0) {
2520 if (ret != -ENOTSUP && ret != -EAGAIN) {
2521 return ret;
2523 continue;
2526 trace_qcow2_skip_cow(qemu_coroutine_self(), m->offset, m->nb_clusters);
2527 m->skip_cow = true;
2529 return 0;
2533 * qcow2_co_pwritev_task
2534 * Called with s->lock unlocked
2535 * l2meta - if not NULL, qcow2_co_pwritev_task() will consume it. Caller must
2536 * not use it somehow after qcow2_co_pwritev_task() call
2538 static coroutine_fn GRAPH_RDLOCK
2539 int qcow2_co_pwritev_task(BlockDriverState *bs, uint64_t host_offset,
2540 uint64_t offset, uint64_t bytes, QEMUIOVector *qiov,
2541 uint64_t qiov_offset, QCowL2Meta *l2meta)
2543 int ret;
2544 BDRVQcow2State *s = bs->opaque;
2545 void *crypt_buf = NULL;
2546 QEMUIOVector encrypted_qiov;
2548 if (bs->encrypted) {
2549 assert(s->crypto);
2550 assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2551 crypt_buf = qemu_try_blockalign(bs->file->bs, bytes);
2552 if (crypt_buf == NULL) {
2553 ret = -ENOMEM;
2554 goto out_unlocked;
2556 qemu_iovec_to_buf(qiov, qiov_offset, crypt_buf, bytes);
2558 if (qcow2_co_encrypt(bs, host_offset, offset, crypt_buf, bytes) < 0) {
2559 ret = -EIO;
2560 goto out_unlocked;
2563 qemu_iovec_init_buf(&encrypted_qiov, crypt_buf, bytes);
2564 qiov = &encrypted_qiov;
2565 qiov_offset = 0;
2568 /* Try to efficiently initialize the physical space with zeroes */
2569 ret = handle_alloc_space(bs, l2meta);
2570 if (ret < 0) {
2571 goto out_unlocked;
2575 * If we need to do COW, check if it's possible to merge the
2576 * writing of the guest data together with that of the COW regions.
2577 * If it's not possible (or not necessary) then write the
2578 * guest data now.
2580 if (!merge_cow(offset, bytes, qiov, qiov_offset, l2meta)) {
2581 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
2582 trace_qcow2_writev_data(qemu_coroutine_self(), host_offset);
2583 ret = bdrv_co_pwritev_part(s->data_file, host_offset,
2584 bytes, qiov, qiov_offset, 0);
2585 if (ret < 0) {
2586 goto out_unlocked;
2590 qemu_co_mutex_lock(&s->lock);
2592 ret = qcow2_handle_l2meta(bs, &l2meta, true);
2593 goto out_locked;
2595 out_unlocked:
2596 qemu_co_mutex_lock(&s->lock);
2598 out_locked:
2599 qcow2_handle_l2meta(bs, &l2meta, false);
2600 qemu_co_mutex_unlock(&s->lock);
2602 qemu_vfree(crypt_buf);
2604 return ret;
2608 * This function can count as GRAPH_RDLOCK because qcow2_co_pwritev_part() holds
2609 * the graph lock and keeps it until this coroutine has terminated.
2611 static coroutine_fn GRAPH_RDLOCK int qcow2_co_pwritev_task_entry(AioTask *task)
2613 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
2615 assert(!t->subcluster_type);
2617 return qcow2_co_pwritev_task(t->bs, t->host_offset,
2618 t->offset, t->bytes, t->qiov, t->qiov_offset,
2619 t->l2meta);
2622 static int coroutine_fn GRAPH_RDLOCK
2623 qcow2_co_pwritev_part(BlockDriverState *bs, int64_t offset, int64_t bytes,
2624 QEMUIOVector *qiov, size_t qiov_offset,
2625 BdrvRequestFlags flags)
2627 BDRVQcow2State *s = bs->opaque;
2628 int offset_in_cluster;
2629 int ret;
2630 unsigned int cur_bytes; /* number of sectors in current iteration */
2631 uint64_t host_offset;
2632 QCowL2Meta *l2meta = NULL;
2633 AioTaskPool *aio = NULL;
2635 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes);
2637 while (bytes != 0 && aio_task_pool_status(aio) == 0) {
2639 l2meta = NULL;
2641 trace_qcow2_writev_start_part(qemu_coroutine_self());
2642 offset_in_cluster = offset_into_cluster(s, offset);
2643 cur_bytes = MIN(bytes, INT_MAX);
2644 if (bs->encrypted) {
2645 cur_bytes = MIN(cur_bytes,
2646 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
2647 - offset_in_cluster);
2650 qemu_co_mutex_lock(&s->lock);
2652 ret = qcow2_alloc_host_offset(bs, offset, &cur_bytes,
2653 &host_offset, &l2meta);
2654 if (ret < 0) {
2655 goto out_locked;
2658 ret = qcow2_pre_write_overlap_check(bs, 0, host_offset,
2659 cur_bytes, true);
2660 if (ret < 0) {
2661 goto out_locked;
2664 qemu_co_mutex_unlock(&s->lock);
2666 if (!aio && cur_bytes != bytes) {
2667 aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
2669 ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_task_entry, 0,
2670 host_offset, offset,
2671 cur_bytes, qiov, qiov_offset, l2meta);
2672 l2meta = NULL; /* l2meta is consumed by qcow2_co_pwritev_task() */
2673 if (ret < 0) {
2674 goto fail_nometa;
2677 bytes -= cur_bytes;
2678 offset += cur_bytes;
2679 qiov_offset += cur_bytes;
2680 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes);
2682 ret = 0;
2684 qemu_co_mutex_lock(&s->lock);
2686 out_locked:
2687 qcow2_handle_l2meta(bs, &l2meta, false);
2689 qemu_co_mutex_unlock(&s->lock);
2691 fail_nometa:
2692 if (aio) {
2693 aio_task_pool_wait_all(aio);
2694 if (ret == 0) {
2695 ret = aio_task_pool_status(aio);
2697 g_free(aio);
2700 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
2702 return ret;
2705 static int qcow2_inactivate(BlockDriverState *bs)
2707 BDRVQcow2State *s = bs->opaque;
2708 int ret, result = 0;
2709 Error *local_err = NULL;
2711 qcow2_store_persistent_dirty_bitmaps(bs, true, &local_err);
2712 if (local_err != NULL) {
2713 result = -EINVAL;
2714 error_reportf_err(local_err, "Lost persistent bitmaps during "
2715 "inactivation of node '%s': ",
2716 bdrv_get_device_or_node_name(bs));
2719 ret = qcow2_cache_flush(bs, s->l2_table_cache);
2720 if (ret) {
2721 result = ret;
2722 error_report("Failed to flush the L2 table cache: %s",
2723 strerror(-ret));
2726 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2727 if (ret) {
2728 result = ret;
2729 error_report("Failed to flush the refcount block cache: %s",
2730 strerror(-ret));
2733 if (result == 0) {
2734 qcow2_mark_clean(bs);
2737 return result;
2740 static void qcow2_do_close(BlockDriverState *bs, bool close_data_file)
2742 BDRVQcow2State *s = bs->opaque;
2743 qemu_vfree(s->l1_table);
2744 /* else pre-write overlap checks in cache_destroy may crash */
2745 s->l1_table = NULL;
2747 if (!(s->flags & BDRV_O_INACTIVE)) {
2748 qcow2_inactivate(bs);
2751 cache_clean_timer_del(bs);
2752 qcow2_cache_destroy(s->l2_table_cache);
2753 qcow2_cache_destroy(s->refcount_block_cache);
2755 qcrypto_block_free(s->crypto);
2756 s->crypto = NULL;
2757 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
2759 g_free(s->unknown_header_fields);
2760 cleanup_unknown_header_ext(bs);
2762 g_free(s->image_data_file);
2763 g_free(s->image_backing_file);
2764 g_free(s->image_backing_format);
2766 if (close_data_file && has_data_file(bs)) {
2767 bdrv_unref_child(bs, s->data_file);
2768 s->data_file = NULL;
2771 qcow2_refcount_close(bs);
2772 qcow2_free_snapshots(bs);
2775 static void qcow2_close(BlockDriverState *bs)
2777 qcow2_do_close(bs, true);
2780 static void coroutine_fn GRAPH_RDLOCK
2781 qcow2_co_invalidate_cache(BlockDriverState *bs, Error **errp)
2783 ERRP_GUARD();
2784 BDRVQcow2State *s = bs->opaque;
2785 BdrvChild *data_file;
2786 int flags = s->flags;
2787 QCryptoBlock *crypto = NULL;
2788 QDict *options;
2789 int ret;
2792 * Backing files are read-only which makes all of their metadata immutable,
2793 * that means we don't have to worry about reopening them here.
2796 crypto = s->crypto;
2797 s->crypto = NULL;
2800 * Do not reopen s->data_file (i.e., have qcow2_do_close() not close it,
2801 * and then prevent qcow2_do_open() from opening it), because this function
2802 * runs in the I/O path and as such we must not invoke global-state
2803 * functions like bdrv_unref_child() and bdrv_open_child().
2806 qcow2_do_close(bs, false);
2808 data_file = s->data_file;
2809 memset(s, 0, sizeof(BDRVQcow2State));
2810 s->data_file = data_file;
2812 options = qdict_clone_shallow(bs->options);
2814 flags &= ~BDRV_O_INACTIVE;
2815 qemu_co_mutex_lock(&s->lock);
2816 ret = qcow2_do_open(bs, options, flags, false, errp);
2817 qemu_co_mutex_unlock(&s->lock);
2818 qobject_unref(options);
2819 if (ret < 0) {
2820 error_prepend(errp, "Could not reopen qcow2 layer: ");
2821 bs->drv = NULL;
2822 return;
2825 s->crypto = crypto;
2828 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
2829 size_t len, size_t buflen)
2831 QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
2832 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
2834 if (buflen < ext_len) {
2835 return -ENOSPC;
2838 *ext_backing_fmt = (QCowExtension) {
2839 .magic = cpu_to_be32(magic),
2840 .len = cpu_to_be32(len),
2843 if (len) {
2844 memcpy(buf + sizeof(QCowExtension), s, len);
2847 return ext_len;
2851 * Updates the qcow2 header, including the variable length parts of it, i.e.
2852 * the backing file name and all extensions. qcow2 was not designed to allow
2853 * such changes, so if we run out of space (we can only use the first cluster)
2854 * this function may fail.
2856 * Returns 0 on success, -errno in error cases.
2858 int qcow2_update_header(BlockDriverState *bs)
2860 BDRVQcow2State *s = bs->opaque;
2861 QCowHeader *header;
2862 char *buf;
2863 size_t buflen = s->cluster_size;
2864 int ret;
2865 uint64_t total_size;
2866 uint32_t refcount_table_clusters;
2867 size_t header_length;
2868 Qcow2UnknownHeaderExtension *uext;
2870 buf = qemu_blockalign(bs, buflen);
2872 /* Header structure */
2873 header = (QCowHeader*) buf;
2875 if (buflen < sizeof(*header)) {
2876 ret = -ENOSPC;
2877 goto fail;
2880 header_length = sizeof(*header) + s->unknown_header_fields_size;
2881 total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
2882 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2884 ret = validate_compression_type(s, NULL);
2885 if (ret) {
2886 goto fail;
2889 *header = (QCowHeader) {
2890 /* Version 2 fields */
2891 .magic = cpu_to_be32(QCOW_MAGIC),
2892 .version = cpu_to_be32(s->qcow_version),
2893 .backing_file_offset = 0,
2894 .backing_file_size = 0,
2895 .cluster_bits = cpu_to_be32(s->cluster_bits),
2896 .size = cpu_to_be64(total_size),
2897 .crypt_method = cpu_to_be32(s->crypt_method_header),
2898 .l1_size = cpu_to_be32(s->l1_size),
2899 .l1_table_offset = cpu_to_be64(s->l1_table_offset),
2900 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset),
2901 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
2902 .nb_snapshots = cpu_to_be32(s->nb_snapshots),
2903 .snapshots_offset = cpu_to_be64(s->snapshots_offset),
2905 /* Version 3 fields */
2906 .incompatible_features = cpu_to_be64(s->incompatible_features),
2907 .compatible_features = cpu_to_be64(s->compatible_features),
2908 .autoclear_features = cpu_to_be64(s->autoclear_features),
2909 .refcount_order = cpu_to_be32(s->refcount_order),
2910 .header_length = cpu_to_be32(header_length),
2911 .compression_type = s->compression_type,
2914 /* For older versions, write a shorter header */
2915 switch (s->qcow_version) {
2916 case 2:
2917 ret = offsetof(QCowHeader, incompatible_features);
2918 break;
2919 case 3:
2920 ret = sizeof(*header);
2921 break;
2922 default:
2923 ret = -EINVAL;
2924 goto fail;
2927 buf += ret;
2928 buflen -= ret;
2929 memset(buf, 0, buflen);
2931 /* Preserve any unknown field in the header */
2932 if (s->unknown_header_fields_size) {
2933 if (buflen < s->unknown_header_fields_size) {
2934 ret = -ENOSPC;
2935 goto fail;
2938 memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
2939 buf += s->unknown_header_fields_size;
2940 buflen -= s->unknown_header_fields_size;
2943 /* Backing file format header extension */
2944 if (s->image_backing_format) {
2945 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
2946 s->image_backing_format,
2947 strlen(s->image_backing_format),
2948 buflen);
2949 if (ret < 0) {
2950 goto fail;
2953 buf += ret;
2954 buflen -= ret;
2957 /* External data file header extension */
2958 if (has_data_file(bs) && s->image_data_file) {
2959 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_DATA_FILE,
2960 s->image_data_file, strlen(s->image_data_file),
2961 buflen);
2962 if (ret < 0) {
2963 goto fail;
2966 buf += ret;
2967 buflen -= ret;
2970 /* Full disk encryption header pointer extension */
2971 if (s->crypto_header.offset != 0) {
2972 s->crypto_header.offset = cpu_to_be64(s->crypto_header.offset);
2973 s->crypto_header.length = cpu_to_be64(s->crypto_header.length);
2974 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_CRYPTO_HEADER,
2975 &s->crypto_header, sizeof(s->crypto_header),
2976 buflen);
2977 s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset);
2978 s->crypto_header.length = be64_to_cpu(s->crypto_header.length);
2979 if (ret < 0) {
2980 goto fail;
2982 buf += ret;
2983 buflen -= ret;
2987 * Feature table. A mere 8 feature names occupies 392 bytes, and
2988 * when coupled with the v3 minimum header of 104 bytes plus the
2989 * 8-byte end-of-extension marker, that would leave only 8 bytes
2990 * for a backing file name in an image with 512-byte clusters.
2991 * Thus, we choose to omit this header for cluster sizes 4k and
2992 * smaller.
2994 if (s->qcow_version >= 3 && s->cluster_size > 4096) {
2995 static const Qcow2Feature features[] = {
2997 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2998 .bit = QCOW2_INCOMPAT_DIRTY_BITNR,
2999 .name = "dirty bit",
3002 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
3003 .bit = QCOW2_INCOMPAT_CORRUPT_BITNR,
3004 .name = "corrupt bit",
3007 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
3008 .bit = QCOW2_INCOMPAT_DATA_FILE_BITNR,
3009 .name = "external data file",
3012 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
3013 .bit = QCOW2_INCOMPAT_COMPRESSION_BITNR,
3014 .name = "compression type",
3017 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
3018 .bit = QCOW2_INCOMPAT_EXTL2_BITNR,
3019 .name = "extended L2 entries",
3022 .type = QCOW2_FEAT_TYPE_COMPATIBLE,
3023 .bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
3024 .name = "lazy refcounts",
3027 .type = QCOW2_FEAT_TYPE_AUTOCLEAR,
3028 .bit = QCOW2_AUTOCLEAR_BITMAPS_BITNR,
3029 .name = "bitmaps",
3032 .type = QCOW2_FEAT_TYPE_AUTOCLEAR,
3033 .bit = QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR,
3034 .name = "raw external data",
3038 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
3039 features, sizeof(features), buflen);
3040 if (ret < 0) {
3041 goto fail;
3043 buf += ret;
3044 buflen -= ret;
3047 /* Bitmap extension */
3048 if (s->nb_bitmaps > 0) {
3049 Qcow2BitmapHeaderExt bitmaps_header = {
3050 .nb_bitmaps = cpu_to_be32(s->nb_bitmaps),
3051 .bitmap_directory_size =
3052 cpu_to_be64(s->bitmap_directory_size),
3053 .bitmap_directory_offset =
3054 cpu_to_be64(s->bitmap_directory_offset)
3056 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BITMAPS,
3057 &bitmaps_header, sizeof(bitmaps_header),
3058 buflen);
3059 if (ret < 0) {
3060 goto fail;
3062 buf += ret;
3063 buflen -= ret;
3066 /* Keep unknown header extensions */
3067 QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
3068 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
3069 if (ret < 0) {
3070 goto fail;
3073 buf += ret;
3074 buflen -= ret;
3077 /* End of header extensions */
3078 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
3079 if (ret < 0) {
3080 goto fail;
3083 buf += ret;
3084 buflen -= ret;
3086 /* Backing file name */
3087 if (s->image_backing_file) {
3088 size_t backing_file_len = strlen(s->image_backing_file);
3090 if (buflen < backing_file_len) {
3091 ret = -ENOSPC;
3092 goto fail;
3095 /* Using strncpy is ok here, since buf is not NUL-terminated. */
3096 strncpy(buf, s->image_backing_file, buflen);
3098 header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
3099 header->backing_file_size = cpu_to_be32(backing_file_len);
3102 /* Write the new header */
3103 ret = bdrv_pwrite(bs->file, 0, s->cluster_size, header, 0);
3104 if (ret < 0) {
3105 goto fail;
3108 ret = 0;
3109 fail:
3110 qemu_vfree(header);
3111 return ret;
3114 static int qcow2_change_backing_file(BlockDriverState *bs,
3115 const char *backing_file, const char *backing_fmt)
3117 BDRVQcow2State *s = bs->opaque;
3119 /* Adding a backing file means that the external data file alone won't be
3120 * enough to make sense of the content */
3121 if (backing_file && data_file_is_raw(bs)) {
3122 return -EINVAL;
3125 if (backing_file && strlen(backing_file) > 1023) {
3126 return -EINVAL;
3129 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3130 backing_file ?: "");
3131 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
3132 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
3134 g_free(s->image_backing_file);
3135 g_free(s->image_backing_format);
3137 s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
3138 s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
3140 return qcow2_update_header(bs);
3143 static int qcow2_set_up_encryption(BlockDriverState *bs,
3144 QCryptoBlockCreateOptions *cryptoopts,
3145 Error **errp)
3147 BDRVQcow2State *s = bs->opaque;
3148 QCryptoBlock *crypto = NULL;
3149 int fmt, ret;
3151 switch (cryptoopts->format) {
3152 case Q_CRYPTO_BLOCK_FORMAT_LUKS:
3153 fmt = QCOW_CRYPT_LUKS;
3154 break;
3155 case Q_CRYPTO_BLOCK_FORMAT_QCOW:
3156 fmt = QCOW_CRYPT_AES;
3157 break;
3158 default:
3159 error_setg(errp, "Crypto format not supported in qcow2");
3160 return -EINVAL;
3163 s->crypt_method_header = fmt;
3165 crypto = qcrypto_block_create(cryptoopts, "encrypt.",
3166 qcow2_crypto_hdr_init_func,
3167 qcow2_crypto_hdr_write_func,
3168 bs, errp);
3169 if (!crypto) {
3170 return -EINVAL;
3173 ret = qcow2_update_header(bs);
3174 if (ret < 0) {
3175 error_setg_errno(errp, -ret, "Could not write encryption header");
3176 goto out;
3179 ret = 0;
3180 out:
3181 qcrypto_block_free(crypto);
3182 return ret;
3186 * Preallocates metadata structures for data clusters between @offset (in the
3187 * guest disk) and @new_length (which is thus generally the new guest disk
3188 * size).
3190 * Returns: 0 on success, -errno on failure.
3192 static int coroutine_fn GRAPH_RDLOCK
3193 preallocate_co(BlockDriverState *bs, uint64_t offset, uint64_t new_length,
3194 PreallocMode mode, Error **errp)
3196 BDRVQcow2State *s = bs->opaque;
3197 uint64_t bytes;
3198 uint64_t host_offset = 0;
3199 int64_t file_length;
3200 unsigned int cur_bytes;
3201 int ret;
3202 QCowL2Meta *meta = NULL, *m;
3204 assert(offset <= new_length);
3205 bytes = new_length - offset;
3207 while (bytes) {
3208 cur_bytes = MIN(bytes, QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size));
3209 ret = qcow2_alloc_host_offset(bs, offset, &cur_bytes,
3210 &host_offset, &meta);
3211 if (ret < 0) {
3212 error_setg_errno(errp, -ret, "Allocating clusters failed");
3213 goto out;
3216 for (m = meta; m != NULL; m = m->next) {
3217 m->prealloc = true;
3220 ret = qcow2_handle_l2meta(bs, &meta, true);
3221 if (ret < 0) {
3222 error_setg_errno(errp, -ret, "Mapping clusters failed");
3223 goto out;
3226 /* TODO Preallocate data if requested */
3228 bytes -= cur_bytes;
3229 offset += cur_bytes;
3233 * It is expected that the image file is large enough to actually contain
3234 * all of the allocated clusters (otherwise we get failing reads after
3235 * EOF). Extend the image to the last allocated sector.
3237 file_length = bdrv_getlength(s->data_file->bs);
3238 if (file_length < 0) {
3239 error_setg_errno(errp, -file_length, "Could not get file size");
3240 ret = file_length;
3241 goto out;
3244 if (host_offset + cur_bytes > file_length) {
3245 if (mode == PREALLOC_MODE_METADATA) {
3246 mode = PREALLOC_MODE_OFF;
3248 ret = bdrv_co_truncate(s->data_file, host_offset + cur_bytes, false,
3249 mode, 0, errp);
3250 if (ret < 0) {
3251 goto out;
3255 ret = 0;
3257 out:
3258 qcow2_handle_l2meta(bs, &meta, false);
3259 return ret;
3262 /* qcow2_refcount_metadata_size:
3263 * @clusters: number of clusters to refcount (including data and L1/L2 tables)
3264 * @cluster_size: size of a cluster, in bytes
3265 * @refcount_order: refcount bits power-of-2 exponent
3266 * @generous_increase: allow for the refcount table to be 1.5x as large as it
3267 * needs to be
3269 * Returns: Number of bytes required for refcount blocks and table metadata.
3271 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size,
3272 int refcount_order, bool generous_increase,
3273 uint64_t *refblock_count)
3276 * Every host cluster is reference-counted, including metadata (even
3277 * refcount metadata is recursively included).
3279 * An accurate formula for the size of refcount metadata size is difficult
3280 * to derive. An easier method of calculation is finding the fixed point
3281 * where no further refcount blocks or table clusters are required to
3282 * reference count every cluster.
3284 int64_t blocks_per_table_cluster = cluster_size / REFTABLE_ENTRY_SIZE;
3285 int64_t refcounts_per_block = cluster_size * 8 / (1 << refcount_order);
3286 int64_t table = 0; /* number of refcount table clusters */
3287 int64_t blocks = 0; /* number of refcount block clusters */
3288 int64_t last;
3289 int64_t n = 0;
3291 do {
3292 last = n;
3293 blocks = DIV_ROUND_UP(clusters + table + blocks, refcounts_per_block);
3294 table = DIV_ROUND_UP(blocks, blocks_per_table_cluster);
3295 n = clusters + blocks + table;
3297 if (n == last && generous_increase) {
3298 clusters += DIV_ROUND_UP(table, 2);
3299 n = 0; /* force another loop */
3300 generous_increase = false;
3302 } while (n != last);
3304 if (refblock_count) {
3305 *refblock_count = blocks;
3308 return (blocks + table) * cluster_size;
3312 * qcow2_calc_prealloc_size:
3313 * @total_size: virtual disk size in bytes
3314 * @cluster_size: cluster size in bytes
3315 * @refcount_order: refcount bits power-of-2 exponent
3316 * @extended_l2: true if the image has extended L2 entries
3318 * Returns: Total number of bytes required for the fully allocated image
3319 * (including metadata).
3321 static int64_t qcow2_calc_prealloc_size(int64_t total_size,
3322 size_t cluster_size,
3323 int refcount_order,
3324 bool extended_l2)
3326 int64_t meta_size = 0;
3327 uint64_t nl1e, nl2e;
3328 int64_t aligned_total_size = ROUND_UP(total_size, cluster_size);
3329 size_t l2e_size = extended_l2 ? L2E_SIZE_EXTENDED : L2E_SIZE_NORMAL;
3331 /* header: 1 cluster */
3332 meta_size += cluster_size;
3334 /* total size of L2 tables */
3335 nl2e = aligned_total_size / cluster_size;
3336 nl2e = ROUND_UP(nl2e, cluster_size / l2e_size);
3337 meta_size += nl2e * l2e_size;
3339 /* total size of L1 tables */
3340 nl1e = nl2e * l2e_size / cluster_size;
3341 nl1e = ROUND_UP(nl1e, cluster_size / L1E_SIZE);
3342 meta_size += nl1e * L1E_SIZE;
3344 /* total size of refcount table and blocks */
3345 meta_size += qcow2_refcount_metadata_size(
3346 (meta_size + aligned_total_size) / cluster_size,
3347 cluster_size, refcount_order, false, NULL);
3349 return meta_size + aligned_total_size;
3352 static bool validate_cluster_size(size_t cluster_size, bool extended_l2,
3353 Error **errp)
3355 int cluster_bits = ctz32(cluster_size);
3356 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
3357 (1 << cluster_bits) != cluster_size)
3359 error_setg(errp, "Cluster size must be a power of two between %d and "
3360 "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
3361 return false;
3364 if (extended_l2) {
3365 unsigned min_cluster_size =
3366 (1 << MIN_CLUSTER_BITS) * QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER;
3367 if (cluster_size < min_cluster_size) {
3368 error_setg(errp, "Extended L2 entries are only supported with "
3369 "cluster sizes of at least %u bytes", min_cluster_size);
3370 return false;
3374 return true;
3377 static size_t qcow2_opt_get_cluster_size_del(QemuOpts *opts, bool extended_l2,
3378 Error **errp)
3380 size_t cluster_size;
3382 cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
3383 DEFAULT_CLUSTER_SIZE);
3384 if (!validate_cluster_size(cluster_size, extended_l2, errp)) {
3385 return 0;
3387 return cluster_size;
3390 static int qcow2_opt_get_version_del(QemuOpts *opts, Error **errp)
3392 char *buf;
3393 int ret;
3395 buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
3396 if (!buf) {
3397 ret = 3; /* default */
3398 } else if (!strcmp(buf, "0.10")) {
3399 ret = 2;
3400 } else if (!strcmp(buf, "1.1")) {
3401 ret = 3;
3402 } else {
3403 error_setg(errp, "Invalid compatibility level: '%s'", buf);
3404 ret = -EINVAL;
3406 g_free(buf);
3407 return ret;
3410 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts *opts, int version,
3411 Error **errp)
3413 uint64_t refcount_bits;
3415 refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 16);
3416 if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
3417 error_setg(errp, "Refcount width must be a power of two and may not "
3418 "exceed 64 bits");
3419 return 0;
3422 if (version < 3 && refcount_bits != 16) {
3423 error_setg(errp, "Different refcount widths than 16 bits require "
3424 "compatibility level 1.1 or above (use compat=1.1 or "
3425 "greater)");
3426 return 0;
3429 return refcount_bits;
3432 static int coroutine_fn
3433 qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp)
3435 BlockdevCreateOptionsQcow2 *qcow2_opts;
3436 QDict *options;
3439 * Open the image file and write a minimal qcow2 header.
3441 * We keep things simple and start with a zero-sized image. We also
3442 * do without refcount blocks or a L1 table for now. We'll fix the
3443 * inconsistency later.
3445 * We do need a refcount table because growing the refcount table means
3446 * allocating two new refcount blocks - the second of which would be at
3447 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
3448 * size for any qcow2 image.
3450 BlockBackend *blk = NULL;
3451 BlockDriverState *bs = NULL;
3452 BlockDriverState *data_bs = NULL;
3453 QCowHeader *header;
3454 size_t cluster_size;
3455 int version;
3456 int refcount_order;
3457 uint64_t *refcount_table;
3458 int ret;
3459 uint8_t compression_type = QCOW2_COMPRESSION_TYPE_ZLIB;
3461 assert(create_options->driver == BLOCKDEV_DRIVER_QCOW2);
3462 qcow2_opts = &create_options->u.qcow2;
3464 bs = bdrv_co_open_blockdev_ref(qcow2_opts->file, errp);
3465 if (bs == NULL) {
3466 return -EIO;
3469 /* Validate options and set default values */
3470 if (!QEMU_IS_ALIGNED(qcow2_opts->size, BDRV_SECTOR_SIZE)) {
3471 error_setg(errp, "Image size must be a multiple of %u bytes",
3472 (unsigned) BDRV_SECTOR_SIZE);
3473 ret = -EINVAL;
3474 goto out;
3477 if (qcow2_opts->has_version) {
3478 switch (qcow2_opts->version) {
3479 case BLOCKDEV_QCOW2_VERSION_V2:
3480 version = 2;
3481 break;
3482 case BLOCKDEV_QCOW2_VERSION_V3:
3483 version = 3;
3484 break;
3485 default:
3486 g_assert_not_reached();
3488 } else {
3489 version = 3;
3492 if (qcow2_opts->has_cluster_size) {
3493 cluster_size = qcow2_opts->cluster_size;
3494 } else {
3495 cluster_size = DEFAULT_CLUSTER_SIZE;
3498 if (!qcow2_opts->has_extended_l2) {
3499 qcow2_opts->extended_l2 = false;
3501 if (qcow2_opts->extended_l2) {
3502 if (version < 3) {
3503 error_setg(errp, "Extended L2 entries are only supported with "
3504 "compatibility level 1.1 and above (use version=v3 or "
3505 "greater)");
3506 ret = -EINVAL;
3507 goto out;
3511 if (!validate_cluster_size(cluster_size, qcow2_opts->extended_l2, errp)) {
3512 ret = -EINVAL;
3513 goto out;
3516 if (!qcow2_opts->has_preallocation) {
3517 qcow2_opts->preallocation = PREALLOC_MODE_OFF;
3519 if (qcow2_opts->backing_file &&
3520 qcow2_opts->preallocation != PREALLOC_MODE_OFF &&
3521 !qcow2_opts->extended_l2)
3523 error_setg(errp, "Backing file and preallocation can only be used at "
3524 "the same time if extended_l2 is on");
3525 ret = -EINVAL;
3526 goto out;
3528 if (qcow2_opts->has_backing_fmt && !qcow2_opts->backing_file) {
3529 error_setg(errp, "Backing format cannot be used without backing file");
3530 ret = -EINVAL;
3531 goto out;
3534 if (!qcow2_opts->has_lazy_refcounts) {
3535 qcow2_opts->lazy_refcounts = false;
3537 if (version < 3 && qcow2_opts->lazy_refcounts) {
3538 error_setg(errp, "Lazy refcounts only supported with compatibility "
3539 "level 1.1 and above (use version=v3 or greater)");
3540 ret = -EINVAL;
3541 goto out;
3544 if (!qcow2_opts->has_refcount_bits) {
3545 qcow2_opts->refcount_bits = 16;
3547 if (qcow2_opts->refcount_bits > 64 ||
3548 !is_power_of_2(qcow2_opts->refcount_bits))
3550 error_setg(errp, "Refcount width must be a power of two and may not "
3551 "exceed 64 bits");
3552 ret = -EINVAL;
3553 goto out;
3555 if (version < 3 && qcow2_opts->refcount_bits != 16) {
3556 error_setg(errp, "Different refcount widths than 16 bits require "
3557 "compatibility level 1.1 or above (use version=v3 or "
3558 "greater)");
3559 ret = -EINVAL;
3560 goto out;
3562 refcount_order = ctz32(qcow2_opts->refcount_bits);
3564 if (qcow2_opts->data_file_raw && !qcow2_opts->data_file) {
3565 error_setg(errp, "data-file-raw requires data-file");
3566 ret = -EINVAL;
3567 goto out;
3569 if (qcow2_opts->data_file_raw && qcow2_opts->backing_file) {
3570 error_setg(errp, "Backing file and data-file-raw cannot be used at "
3571 "the same time");
3572 ret = -EINVAL;
3573 goto out;
3575 if (qcow2_opts->data_file_raw &&
3576 qcow2_opts->preallocation == PREALLOC_MODE_OFF)
3579 * data-file-raw means that "the external data file can be
3580 * read as a consistent standalone raw image without looking
3581 * at the qcow2 metadata." It does not say that the metadata
3582 * must be ignored, though (and the qcow2 driver in fact does
3583 * not ignore it), so the L1/L2 tables must be present and
3584 * give a 1:1 mapping, so you get the same result regardless
3585 * of whether you look at the metadata or whether you ignore
3586 * it.
3588 qcow2_opts->preallocation = PREALLOC_MODE_METADATA;
3591 * Cannot use preallocation with backing files, but giving a
3592 * backing file when specifying data_file_raw is an error
3593 * anyway.
3595 assert(!qcow2_opts->backing_file);
3598 if (qcow2_opts->data_file) {
3599 if (version < 3) {
3600 error_setg(errp, "External data files are only supported with "
3601 "compatibility level 1.1 and above (use version=v3 or "
3602 "greater)");
3603 ret = -EINVAL;
3604 goto out;
3606 data_bs = bdrv_co_open_blockdev_ref(qcow2_opts->data_file, errp);
3607 if (data_bs == NULL) {
3608 ret = -EIO;
3609 goto out;
3613 if (qcow2_opts->has_compression_type &&
3614 qcow2_opts->compression_type != QCOW2_COMPRESSION_TYPE_ZLIB) {
3616 ret = -EINVAL;
3618 if (version < 3) {
3619 error_setg(errp, "Non-zlib compression type is only supported with "
3620 "compatibility level 1.1 and above (use version=v3 or "
3621 "greater)");
3622 goto out;
3625 switch (qcow2_opts->compression_type) {
3626 #ifdef CONFIG_ZSTD
3627 case QCOW2_COMPRESSION_TYPE_ZSTD:
3628 break;
3629 #endif
3630 default:
3631 error_setg(errp, "Unknown compression type");
3632 goto out;
3635 compression_type = qcow2_opts->compression_type;
3638 /* Create BlockBackend to write to the image */
3639 blk = blk_co_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL,
3640 errp);
3641 if (!blk) {
3642 ret = -EPERM;
3643 goto out;
3645 blk_set_allow_write_beyond_eof(blk, true);
3647 /* Write the header */
3648 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
3649 header = g_malloc0(cluster_size);
3650 *header = (QCowHeader) {
3651 .magic = cpu_to_be32(QCOW_MAGIC),
3652 .version = cpu_to_be32(version),
3653 .cluster_bits = cpu_to_be32(ctz32(cluster_size)),
3654 .size = cpu_to_be64(0),
3655 .l1_table_offset = cpu_to_be64(0),
3656 .l1_size = cpu_to_be32(0),
3657 .refcount_table_offset = cpu_to_be64(cluster_size),
3658 .refcount_table_clusters = cpu_to_be32(1),
3659 .refcount_order = cpu_to_be32(refcount_order),
3660 /* don't deal with endianness since compression_type is 1 byte long */
3661 .compression_type = compression_type,
3662 .header_length = cpu_to_be32(sizeof(*header)),
3665 /* We'll update this to correct value later */
3666 header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
3668 if (qcow2_opts->lazy_refcounts) {
3669 header->compatible_features |=
3670 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
3672 if (data_bs) {
3673 header->incompatible_features |=
3674 cpu_to_be64(QCOW2_INCOMPAT_DATA_FILE);
3676 if (qcow2_opts->data_file_raw) {
3677 header->autoclear_features |=
3678 cpu_to_be64(QCOW2_AUTOCLEAR_DATA_FILE_RAW);
3680 if (compression_type != QCOW2_COMPRESSION_TYPE_ZLIB) {
3681 header->incompatible_features |=
3682 cpu_to_be64(QCOW2_INCOMPAT_COMPRESSION);
3685 if (qcow2_opts->extended_l2) {
3686 header->incompatible_features |=
3687 cpu_to_be64(QCOW2_INCOMPAT_EXTL2);
3690 ret = blk_co_pwrite(blk, 0, cluster_size, header, 0);
3691 g_free(header);
3692 if (ret < 0) {
3693 error_setg_errno(errp, -ret, "Could not write qcow2 header");
3694 goto out;
3697 /* Write a refcount table with one refcount block */
3698 refcount_table = g_malloc0(2 * cluster_size);
3699 refcount_table[0] = cpu_to_be64(2 * cluster_size);
3700 ret = blk_co_pwrite(blk, cluster_size, 2 * cluster_size, refcount_table, 0);
3701 g_free(refcount_table);
3703 if (ret < 0) {
3704 error_setg_errno(errp, -ret, "Could not write refcount table");
3705 goto out;
3708 blk_unref(blk);
3709 blk = NULL;
3712 * And now open the image and make it consistent first (i.e. increase the
3713 * refcount of the cluster that is occupied by the header and the refcount
3714 * table)
3716 options = qdict_new();
3717 qdict_put_str(options, "driver", "qcow2");
3718 qdict_put_str(options, "file", bs->node_name);
3719 if (data_bs) {
3720 qdict_put_str(options, "data-file", data_bs->node_name);
3722 blk = blk_co_new_open(NULL, NULL, options,
3723 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH,
3724 errp);
3725 if (blk == NULL) {
3726 ret = -EIO;
3727 goto out;
3730 ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size);
3731 if (ret < 0) {
3732 error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
3733 "header and refcount table");
3734 goto out;
3736 } else if (ret != 0) {
3737 error_report("Huh, first cluster in empty image is already in use?");
3738 abort();
3741 /* Set the external data file if necessary */
3742 if (data_bs) {
3743 BDRVQcow2State *s = blk_bs(blk)->opaque;
3744 s->image_data_file = g_strdup(data_bs->filename);
3747 /* Create a full header (including things like feature table) */
3748 ret = qcow2_update_header(blk_bs(blk));
3749 if (ret < 0) {
3750 error_setg_errno(errp, -ret, "Could not update qcow2 header");
3751 goto out;
3754 /* Okay, now that we have a valid image, let's give it the right size */
3755 ret = blk_co_truncate(blk, qcow2_opts->size, false,
3756 qcow2_opts->preallocation, 0, errp);
3757 if (ret < 0) {
3758 error_prepend(errp, "Could not resize image: ");
3759 goto out;
3762 /* Want a backing file? There you go. */
3763 if (qcow2_opts->backing_file) {
3764 const char *backing_format = NULL;
3766 if (qcow2_opts->has_backing_fmt) {
3767 backing_format = BlockdevDriver_str(qcow2_opts->backing_fmt);
3770 ret = bdrv_change_backing_file(blk_bs(blk), qcow2_opts->backing_file,
3771 backing_format, false);
3772 if (ret < 0) {
3773 error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
3774 "with format '%s'", qcow2_opts->backing_file,
3775 backing_format);
3776 goto out;
3780 /* Want encryption? There you go. */
3781 if (qcow2_opts->encrypt) {
3782 ret = qcow2_set_up_encryption(blk_bs(blk), qcow2_opts->encrypt, errp);
3783 if (ret < 0) {
3784 goto out;
3788 blk_unref(blk);
3789 blk = NULL;
3791 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
3792 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
3793 * have to setup decryption context. We're not doing any I/O on the top
3794 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
3795 * not have effect.
3797 options = qdict_new();
3798 qdict_put_str(options, "driver", "qcow2");
3799 qdict_put_str(options, "file", bs->node_name);
3800 if (data_bs) {
3801 qdict_put_str(options, "data-file", data_bs->node_name);
3803 blk = blk_co_new_open(NULL, NULL, options,
3804 BDRV_O_RDWR | BDRV_O_NO_BACKING | BDRV_O_NO_IO,
3805 errp);
3806 if (blk == NULL) {
3807 ret = -EIO;
3808 goto out;
3811 ret = 0;
3812 out:
3813 blk_unref(blk);
3814 bdrv_unref(bs);
3815 bdrv_unref(data_bs);
3816 return ret;
3819 static int coroutine_fn GRAPH_RDLOCK
3820 qcow2_co_create_opts(BlockDriver *drv, const char *filename, QemuOpts *opts,
3821 Error **errp)
3823 BlockdevCreateOptions *create_options = NULL;
3824 QDict *qdict;
3825 Visitor *v;
3826 BlockDriverState *bs = NULL;
3827 BlockDriverState *data_bs = NULL;
3828 const char *val;
3829 int ret;
3831 /* Only the keyval visitor supports the dotted syntax needed for
3832 * encryption, so go through a QDict before getting a QAPI type. Ignore
3833 * options meant for the protocol layer so that the visitor doesn't
3834 * complain. */
3835 qdict = qemu_opts_to_qdict_filtered(opts, NULL, bdrv_qcow2.create_opts,
3836 true);
3838 /* Handle encryption options */
3839 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT);
3840 if (val && !strcmp(val, "on")) {
3841 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow");
3842 } else if (val && !strcmp(val, "off")) {
3843 qdict_del(qdict, BLOCK_OPT_ENCRYPT);
3846 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT);
3847 if (val && !strcmp(val, "aes")) {
3848 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow");
3851 /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into
3852 * version=v2/v3 below. */
3853 val = qdict_get_try_str(qdict, BLOCK_OPT_COMPAT_LEVEL);
3854 if (val && !strcmp(val, "0.10")) {
3855 qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v2");
3856 } else if (val && !strcmp(val, "1.1")) {
3857 qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v3");
3860 /* Change legacy command line options into QMP ones */
3861 static const QDictRenames opt_renames[] = {
3862 { BLOCK_OPT_BACKING_FILE, "backing-file" },
3863 { BLOCK_OPT_BACKING_FMT, "backing-fmt" },
3864 { BLOCK_OPT_CLUSTER_SIZE, "cluster-size" },
3865 { BLOCK_OPT_LAZY_REFCOUNTS, "lazy-refcounts" },
3866 { BLOCK_OPT_EXTL2, "extended-l2" },
3867 { BLOCK_OPT_REFCOUNT_BITS, "refcount-bits" },
3868 { BLOCK_OPT_ENCRYPT, BLOCK_OPT_ENCRYPT_FORMAT },
3869 { BLOCK_OPT_COMPAT_LEVEL, "version" },
3870 { BLOCK_OPT_DATA_FILE_RAW, "data-file-raw" },
3871 { BLOCK_OPT_COMPRESSION_TYPE, "compression-type" },
3872 { NULL, NULL },
3875 if (!qdict_rename_keys(qdict, opt_renames, errp)) {
3876 ret = -EINVAL;
3877 goto finish;
3880 /* Create and open the file (protocol layer) */
3881 ret = bdrv_co_create_file(filename, opts, errp);
3882 if (ret < 0) {
3883 goto finish;
3886 bs = bdrv_co_open(filename, NULL, NULL,
3887 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
3888 if (bs == NULL) {
3889 ret = -EIO;
3890 goto finish;
3893 /* Create and open an external data file (protocol layer) */
3894 val = qdict_get_try_str(qdict, BLOCK_OPT_DATA_FILE);
3895 if (val) {
3896 ret = bdrv_co_create_file(val, opts, errp);
3897 if (ret < 0) {
3898 goto finish;
3901 data_bs = bdrv_co_open(val, NULL, NULL,
3902 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
3903 errp);
3904 if (data_bs == NULL) {
3905 ret = -EIO;
3906 goto finish;
3909 qdict_del(qdict, BLOCK_OPT_DATA_FILE);
3910 qdict_put_str(qdict, "data-file", data_bs->node_name);
3913 /* Set 'driver' and 'node' options */
3914 qdict_put_str(qdict, "driver", "qcow2");
3915 qdict_put_str(qdict, "file", bs->node_name);
3917 /* Now get the QAPI type BlockdevCreateOptions */
3918 v = qobject_input_visitor_new_flat_confused(qdict, errp);
3919 if (!v) {
3920 ret = -EINVAL;
3921 goto finish;
3924 visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp);
3925 visit_free(v);
3926 if (!create_options) {
3927 ret = -EINVAL;
3928 goto finish;
3931 /* Silently round up size */
3932 create_options->u.qcow2.size = ROUND_UP(create_options->u.qcow2.size,
3933 BDRV_SECTOR_SIZE);
3935 /* Create the qcow2 image (format layer) */
3936 ret = qcow2_co_create(create_options, errp);
3937 finish:
3938 if (ret < 0) {
3939 bdrv_co_delete_file_noerr(bs);
3940 bdrv_co_delete_file_noerr(data_bs);
3941 } else {
3942 ret = 0;
3945 qobject_unref(qdict);
3946 bdrv_unref(bs);
3947 bdrv_unref(data_bs);
3948 qapi_free_BlockdevCreateOptions(create_options);
3949 return ret;
3953 static bool is_zero(BlockDriverState *bs, int64_t offset, int64_t bytes)
3955 int64_t nr;
3956 int res;
3958 /* Clamp to image length, before checking status of underlying sectors */
3959 if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) {
3960 bytes = bs->total_sectors * BDRV_SECTOR_SIZE - offset;
3963 if (!bytes) {
3964 return true;
3968 * bdrv_block_status_above doesn't merge different types of zeros, for
3969 * example, zeros which come from the region which is unallocated in
3970 * the whole backing chain, and zeros which come because of a short
3971 * backing file. So, we need a loop.
3973 do {
3974 res = bdrv_block_status_above(bs, NULL, offset, bytes, &nr, NULL, NULL);
3975 offset += nr;
3976 bytes -= nr;
3977 } while (res >= 0 && (res & BDRV_BLOCK_ZERO) && nr && bytes);
3979 return res >= 0 && (res & BDRV_BLOCK_ZERO) && bytes == 0;
3982 static int coroutine_fn GRAPH_RDLOCK
3983 qcow2_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes,
3984 BdrvRequestFlags flags)
3986 int ret;
3987 BDRVQcow2State *s = bs->opaque;
3989 uint32_t head = offset_into_subcluster(s, offset);
3990 uint32_t tail = ROUND_UP(offset + bytes, s->subcluster_size) -
3991 (offset + bytes);
3993 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, bytes);
3994 if (offset + bytes == bs->total_sectors * BDRV_SECTOR_SIZE) {
3995 tail = 0;
3998 if (head || tail) {
3999 uint64_t off;
4000 unsigned int nr;
4001 QCow2SubclusterType type;
4003 assert(head + bytes + tail <= s->subcluster_size);
4005 /* check whether remainder of cluster already reads as zero */
4006 if (!(is_zero(bs, offset - head, head) &&
4007 is_zero(bs, offset + bytes, tail))) {
4008 return -ENOTSUP;
4011 qemu_co_mutex_lock(&s->lock);
4012 /* We can have new write after previous check */
4013 offset -= head;
4014 bytes = s->subcluster_size;
4015 nr = s->subcluster_size;
4016 ret = qcow2_get_host_offset(bs, offset, &nr, &off, &type);
4017 if (ret < 0 ||
4018 (type != QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN &&
4019 type != QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC &&
4020 type != QCOW2_SUBCLUSTER_ZERO_PLAIN &&
4021 type != QCOW2_SUBCLUSTER_ZERO_ALLOC)) {
4022 qemu_co_mutex_unlock(&s->lock);
4023 return ret < 0 ? ret : -ENOTSUP;
4025 } else {
4026 qemu_co_mutex_lock(&s->lock);
4029 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, bytes);
4031 /* Whatever is left can use real zero subclusters */
4032 ret = qcow2_subcluster_zeroize(bs, offset, bytes, flags);
4033 qemu_co_mutex_unlock(&s->lock);
4035 return ret;
4038 static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs,
4039 int64_t offset, int64_t bytes)
4041 int ret;
4042 BDRVQcow2State *s = bs->opaque;
4044 /* If the image does not support QCOW_OFLAG_ZERO then discarding
4045 * clusters could expose stale data from the backing file. */
4046 if (s->qcow_version < 3 && bs->backing) {
4047 return -ENOTSUP;
4050 if (!QEMU_IS_ALIGNED(offset | bytes, s->cluster_size)) {
4051 assert(bytes < s->cluster_size);
4052 /* Ignore partial clusters, except for the special case of the
4053 * complete partial cluster at the end of an unaligned file */
4054 if (!QEMU_IS_ALIGNED(offset, s->cluster_size) ||
4055 offset + bytes != bs->total_sectors * BDRV_SECTOR_SIZE) {
4056 return -ENOTSUP;
4060 qemu_co_mutex_lock(&s->lock);
4061 ret = qcow2_cluster_discard(bs, offset, bytes, QCOW2_DISCARD_REQUEST,
4062 false);
4063 qemu_co_mutex_unlock(&s->lock);
4064 return ret;
4067 static int coroutine_fn GRAPH_RDLOCK
4068 qcow2_co_copy_range_from(BlockDriverState *bs,
4069 BdrvChild *src, int64_t src_offset,
4070 BdrvChild *dst, int64_t dst_offset,
4071 int64_t bytes, BdrvRequestFlags read_flags,
4072 BdrvRequestFlags write_flags)
4074 BDRVQcow2State *s = bs->opaque;
4075 int ret;
4076 unsigned int cur_bytes; /* number of bytes in current iteration */
4077 BdrvChild *child = NULL;
4078 BdrvRequestFlags cur_write_flags;
4080 assert(!bs->encrypted);
4081 qemu_co_mutex_lock(&s->lock);
4083 while (bytes != 0) {
4084 uint64_t copy_offset = 0;
4085 QCow2SubclusterType type;
4086 /* prepare next request */
4087 cur_bytes = MIN(bytes, INT_MAX);
4088 cur_write_flags = write_flags;
4090 ret = qcow2_get_host_offset(bs, src_offset, &cur_bytes,
4091 &copy_offset, &type);
4092 if (ret < 0) {
4093 goto out;
4096 switch (type) {
4097 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN:
4098 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC:
4099 if (bs->backing && bs->backing->bs) {
4100 int64_t backing_length = bdrv_getlength(bs->backing->bs);
4101 if (src_offset >= backing_length) {
4102 cur_write_flags |= BDRV_REQ_ZERO_WRITE;
4103 } else {
4104 child = bs->backing;
4105 cur_bytes = MIN(cur_bytes, backing_length - src_offset);
4106 copy_offset = src_offset;
4108 } else {
4109 cur_write_flags |= BDRV_REQ_ZERO_WRITE;
4111 break;
4113 case QCOW2_SUBCLUSTER_ZERO_PLAIN:
4114 case QCOW2_SUBCLUSTER_ZERO_ALLOC:
4115 cur_write_flags |= BDRV_REQ_ZERO_WRITE;
4116 break;
4118 case QCOW2_SUBCLUSTER_COMPRESSED:
4119 ret = -ENOTSUP;
4120 goto out;
4122 case QCOW2_SUBCLUSTER_NORMAL:
4123 child = s->data_file;
4124 break;
4126 default:
4127 abort();
4129 qemu_co_mutex_unlock(&s->lock);
4130 ret = bdrv_co_copy_range_from(child,
4131 copy_offset,
4132 dst, dst_offset,
4133 cur_bytes, read_flags, cur_write_flags);
4134 qemu_co_mutex_lock(&s->lock);
4135 if (ret < 0) {
4136 goto out;
4139 bytes -= cur_bytes;
4140 src_offset += cur_bytes;
4141 dst_offset += cur_bytes;
4143 ret = 0;
4145 out:
4146 qemu_co_mutex_unlock(&s->lock);
4147 return ret;
4150 static int coroutine_fn GRAPH_RDLOCK
4151 qcow2_co_copy_range_to(BlockDriverState *bs,
4152 BdrvChild *src, int64_t src_offset,
4153 BdrvChild *dst, int64_t dst_offset,
4154 int64_t bytes, BdrvRequestFlags read_flags,
4155 BdrvRequestFlags write_flags)
4157 BDRVQcow2State *s = bs->opaque;
4158 int ret;
4159 unsigned int cur_bytes; /* number of sectors in current iteration */
4160 uint64_t host_offset;
4161 QCowL2Meta *l2meta = NULL;
4163 assert(!bs->encrypted);
4165 qemu_co_mutex_lock(&s->lock);
4167 while (bytes != 0) {
4169 l2meta = NULL;
4171 cur_bytes = MIN(bytes, INT_MAX);
4173 /* TODO:
4174 * If src->bs == dst->bs, we could simply copy by incrementing
4175 * the refcnt, without copying user data.
4176 * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */
4177 ret = qcow2_alloc_host_offset(bs, dst_offset, &cur_bytes,
4178 &host_offset, &l2meta);
4179 if (ret < 0) {
4180 goto fail;
4183 ret = qcow2_pre_write_overlap_check(bs, 0, host_offset, cur_bytes,
4184 true);
4185 if (ret < 0) {
4186 goto fail;
4189 qemu_co_mutex_unlock(&s->lock);
4190 ret = bdrv_co_copy_range_to(src, src_offset, s->data_file, host_offset,
4191 cur_bytes, read_flags, write_flags);
4192 qemu_co_mutex_lock(&s->lock);
4193 if (ret < 0) {
4194 goto fail;
4197 ret = qcow2_handle_l2meta(bs, &l2meta, true);
4198 if (ret) {
4199 goto fail;
4202 bytes -= cur_bytes;
4203 src_offset += cur_bytes;
4204 dst_offset += cur_bytes;
4206 ret = 0;
4208 fail:
4209 qcow2_handle_l2meta(bs, &l2meta, false);
4211 qemu_co_mutex_unlock(&s->lock);
4213 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
4215 return ret;
4218 static int coroutine_fn GRAPH_RDLOCK
4219 qcow2_co_truncate(BlockDriverState *bs, int64_t offset, bool exact,
4220 PreallocMode prealloc, BdrvRequestFlags flags, Error **errp)
4222 BDRVQcow2State *s = bs->opaque;
4223 uint64_t old_length;
4224 int64_t new_l1_size;
4225 int ret;
4226 QDict *options;
4228 if (prealloc != PREALLOC_MODE_OFF && prealloc != PREALLOC_MODE_METADATA &&
4229 prealloc != PREALLOC_MODE_FALLOC && prealloc != PREALLOC_MODE_FULL)
4231 error_setg(errp, "Unsupported preallocation mode '%s'",
4232 PreallocMode_str(prealloc));
4233 return -ENOTSUP;
4236 if (!QEMU_IS_ALIGNED(offset, BDRV_SECTOR_SIZE)) {
4237 error_setg(errp, "The new size must be a multiple of %u",
4238 (unsigned) BDRV_SECTOR_SIZE);
4239 return -EINVAL;
4242 qemu_co_mutex_lock(&s->lock);
4245 * Even though we store snapshot size for all images, it was not
4246 * required until v3, so it is not safe to proceed for v2.
4248 if (s->nb_snapshots && s->qcow_version < 3) {
4249 error_setg(errp, "Can't resize a v2 image which has snapshots");
4250 ret = -ENOTSUP;
4251 goto fail;
4254 /* See qcow2-bitmap.c for which bitmap scenarios prevent a resize. */
4255 if (qcow2_truncate_bitmaps_check(bs, errp)) {
4256 ret = -ENOTSUP;
4257 goto fail;
4260 old_length = bs->total_sectors * BDRV_SECTOR_SIZE;
4261 new_l1_size = size_to_l1(s, offset);
4263 if (offset < old_length) {
4264 int64_t last_cluster, old_file_size;
4265 if (prealloc != PREALLOC_MODE_OFF) {
4266 error_setg(errp,
4267 "Preallocation can't be used for shrinking an image");
4268 ret = -EINVAL;
4269 goto fail;
4272 ret = qcow2_cluster_discard(bs, ROUND_UP(offset, s->cluster_size),
4273 old_length - ROUND_UP(offset,
4274 s->cluster_size),
4275 QCOW2_DISCARD_ALWAYS, true);
4276 if (ret < 0) {
4277 error_setg_errno(errp, -ret, "Failed to discard cropped clusters");
4278 goto fail;
4281 ret = qcow2_shrink_l1_table(bs, new_l1_size);
4282 if (ret < 0) {
4283 error_setg_errno(errp, -ret,
4284 "Failed to reduce the number of L2 tables");
4285 goto fail;
4288 ret = qcow2_shrink_reftable(bs);
4289 if (ret < 0) {
4290 error_setg_errno(errp, -ret,
4291 "Failed to discard unused refblocks");
4292 goto fail;
4295 old_file_size = bdrv_getlength(bs->file->bs);
4296 if (old_file_size < 0) {
4297 error_setg_errno(errp, -old_file_size,
4298 "Failed to inquire current file length");
4299 ret = old_file_size;
4300 goto fail;
4302 last_cluster = qcow2_get_last_cluster(bs, old_file_size);
4303 if (last_cluster < 0) {
4304 error_setg_errno(errp, -last_cluster,
4305 "Failed to find the last cluster");
4306 ret = last_cluster;
4307 goto fail;
4309 if ((last_cluster + 1) * s->cluster_size < old_file_size) {
4310 Error *local_err = NULL;
4313 * Do not pass @exact here: It will not help the user if
4314 * we get an error here just because they wanted to shrink
4315 * their qcow2 image (on a block device) with qemu-img.
4316 * (And on the qcow2 layer, the @exact requirement is
4317 * always fulfilled, so there is no need to pass it on.)
4319 bdrv_co_truncate(bs->file, (last_cluster + 1) * s->cluster_size,
4320 false, PREALLOC_MODE_OFF, 0, &local_err);
4321 if (local_err) {
4322 warn_reportf_err(local_err,
4323 "Failed to truncate the tail of the image: ");
4326 } else {
4327 ret = qcow2_grow_l1_table(bs, new_l1_size, true);
4328 if (ret < 0) {
4329 error_setg_errno(errp, -ret, "Failed to grow the L1 table");
4330 goto fail;
4333 if (data_file_is_raw(bs) && prealloc == PREALLOC_MODE_OFF) {
4335 * When creating a qcow2 image with data-file-raw, we enforce
4336 * at least prealloc=metadata, so that the L1/L2 tables are
4337 * fully allocated and reading from the data file will return
4338 * the same data as reading from the qcow2 image. When the
4339 * image is grown, we must consequently preallocate the
4340 * metadata structures to cover the added area.
4342 prealloc = PREALLOC_MODE_METADATA;
4346 switch (prealloc) {
4347 case PREALLOC_MODE_OFF:
4348 if (has_data_file(bs)) {
4350 * If the caller wants an exact resize, the external data
4351 * file should be resized to the exact target size, too,
4352 * so we pass @exact here.
4354 ret = bdrv_co_truncate(s->data_file, offset, exact, prealloc, 0,
4355 errp);
4356 if (ret < 0) {
4357 goto fail;
4360 break;
4362 case PREALLOC_MODE_METADATA:
4363 ret = preallocate_co(bs, old_length, offset, prealloc, errp);
4364 if (ret < 0) {
4365 goto fail;
4367 break;
4369 case PREALLOC_MODE_FALLOC:
4370 case PREALLOC_MODE_FULL:
4372 int64_t allocation_start, host_offset, guest_offset;
4373 int64_t clusters_allocated;
4374 int64_t old_file_size, last_cluster, new_file_size;
4375 uint64_t nb_new_data_clusters, nb_new_l2_tables;
4376 bool subclusters_need_allocation = false;
4378 /* With a data file, preallocation means just allocating the metadata
4379 * and forwarding the truncate request to the data file */
4380 if (has_data_file(bs)) {
4381 ret = preallocate_co(bs, old_length, offset, prealloc, errp);
4382 if (ret < 0) {
4383 goto fail;
4385 break;
4388 old_file_size = bdrv_getlength(bs->file->bs);
4389 if (old_file_size < 0) {
4390 error_setg_errno(errp, -old_file_size,
4391 "Failed to inquire current file length");
4392 ret = old_file_size;
4393 goto fail;
4396 last_cluster = qcow2_get_last_cluster(bs, old_file_size);
4397 if (last_cluster >= 0) {
4398 old_file_size = (last_cluster + 1) * s->cluster_size;
4399 } else {
4400 old_file_size = ROUND_UP(old_file_size, s->cluster_size);
4403 nb_new_data_clusters = (ROUND_UP(offset, s->cluster_size) -
4404 start_of_cluster(s, old_length)) >> s->cluster_bits;
4406 /* This is an overestimation; we will not actually allocate space for
4407 * these in the file but just make sure the new refcount structures are
4408 * able to cover them so we will not have to allocate new refblocks
4409 * while entering the data blocks in the potentially new L2 tables.
4410 * (We do not actually care where the L2 tables are placed. Maybe they
4411 * are already allocated or they can be placed somewhere before
4412 * @old_file_size. It does not matter because they will be fully
4413 * allocated automatically, so they do not need to be covered by the
4414 * preallocation. All that matters is that we will not have to allocate
4415 * new refcount structures for them.) */
4416 nb_new_l2_tables = DIV_ROUND_UP(nb_new_data_clusters,
4417 s->cluster_size / l2_entry_size(s));
4418 /* The cluster range may not be aligned to L2 boundaries, so add one L2
4419 * table for a potential head/tail */
4420 nb_new_l2_tables++;
4422 allocation_start = qcow2_refcount_area(bs, old_file_size,
4423 nb_new_data_clusters +
4424 nb_new_l2_tables,
4425 true, 0, 0);
4426 if (allocation_start < 0) {
4427 error_setg_errno(errp, -allocation_start,
4428 "Failed to resize refcount structures");
4429 ret = allocation_start;
4430 goto fail;
4433 clusters_allocated = qcow2_alloc_clusters_at(bs, allocation_start,
4434 nb_new_data_clusters);
4435 if (clusters_allocated < 0) {
4436 error_setg_errno(errp, -clusters_allocated,
4437 "Failed to allocate data clusters");
4438 ret = clusters_allocated;
4439 goto fail;
4442 assert(clusters_allocated == nb_new_data_clusters);
4444 /* Allocate the data area */
4445 new_file_size = allocation_start +
4446 nb_new_data_clusters * s->cluster_size;
4448 * Image file grows, so @exact does not matter.
4450 * If we need to zero out the new area, try first whether the protocol
4451 * driver can already take care of this.
4453 if (flags & BDRV_REQ_ZERO_WRITE) {
4454 ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc,
4455 BDRV_REQ_ZERO_WRITE, NULL);
4456 if (ret >= 0) {
4457 flags &= ~BDRV_REQ_ZERO_WRITE;
4458 /* Ensure that we read zeroes and not backing file data */
4459 subclusters_need_allocation = true;
4461 } else {
4462 ret = -1;
4464 if (ret < 0) {
4465 ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc, 0,
4466 errp);
4468 if (ret < 0) {
4469 error_prepend(errp, "Failed to resize underlying file: ");
4470 qcow2_free_clusters(bs, allocation_start,
4471 nb_new_data_clusters * s->cluster_size,
4472 QCOW2_DISCARD_OTHER);
4473 goto fail;
4476 /* Create the necessary L2 entries */
4477 host_offset = allocation_start;
4478 guest_offset = old_length;
4479 while (nb_new_data_clusters) {
4480 int64_t nb_clusters = MIN(
4481 nb_new_data_clusters,
4482 s->l2_slice_size - offset_to_l2_slice_index(s, guest_offset));
4483 unsigned cow_start_length = offset_into_cluster(s, guest_offset);
4484 QCowL2Meta allocation;
4485 guest_offset = start_of_cluster(s, guest_offset);
4486 allocation = (QCowL2Meta) {
4487 .offset = guest_offset,
4488 .alloc_offset = host_offset,
4489 .nb_clusters = nb_clusters,
4490 .cow_start = {
4491 .offset = 0,
4492 .nb_bytes = cow_start_length,
4494 .cow_end = {
4495 .offset = nb_clusters << s->cluster_bits,
4496 .nb_bytes = 0,
4498 .prealloc = !subclusters_need_allocation,
4500 qemu_co_queue_init(&allocation.dependent_requests);
4502 ret = qcow2_alloc_cluster_link_l2(bs, &allocation);
4503 if (ret < 0) {
4504 error_setg_errno(errp, -ret, "Failed to update L2 tables");
4505 qcow2_free_clusters(bs, host_offset,
4506 nb_new_data_clusters * s->cluster_size,
4507 QCOW2_DISCARD_OTHER);
4508 goto fail;
4511 guest_offset += nb_clusters * s->cluster_size;
4512 host_offset += nb_clusters * s->cluster_size;
4513 nb_new_data_clusters -= nb_clusters;
4515 break;
4518 default:
4519 g_assert_not_reached();
4522 if ((flags & BDRV_REQ_ZERO_WRITE) && offset > old_length) {
4523 uint64_t zero_start = QEMU_ALIGN_UP(old_length, s->subcluster_size);
4526 * Use zero clusters as much as we can. qcow2_subcluster_zeroize()
4527 * requires a subcluster-aligned start. The end may be unaligned if
4528 * it is at the end of the image (which it is here).
4530 if (offset > zero_start) {
4531 ret = qcow2_subcluster_zeroize(bs, zero_start, offset - zero_start,
4533 if (ret < 0) {
4534 error_setg_errno(errp, -ret, "Failed to zero out new clusters");
4535 goto fail;
4539 /* Write explicit zeros for the unaligned head */
4540 if (zero_start > old_length) {
4541 uint64_t len = MIN(zero_start, offset) - old_length;
4542 uint8_t *buf = qemu_blockalign0(bs, len);
4543 QEMUIOVector qiov;
4544 qemu_iovec_init_buf(&qiov, buf, len);
4546 qemu_co_mutex_unlock(&s->lock);
4547 ret = qcow2_co_pwritev_part(bs, old_length, len, &qiov, 0, 0);
4548 qemu_co_mutex_lock(&s->lock);
4550 qemu_vfree(buf);
4551 if (ret < 0) {
4552 error_setg_errno(errp, -ret, "Failed to zero out the new area");
4553 goto fail;
4558 if (prealloc != PREALLOC_MODE_OFF) {
4559 /* Flush metadata before actually changing the image size */
4560 ret = qcow2_write_caches(bs);
4561 if (ret < 0) {
4562 error_setg_errno(errp, -ret,
4563 "Failed to flush the preallocated area to disk");
4564 goto fail;
4568 bs->total_sectors = offset / BDRV_SECTOR_SIZE;
4570 /* write updated header.size */
4571 offset = cpu_to_be64(offset);
4572 ret = bdrv_co_pwrite_sync(bs->file, offsetof(QCowHeader, size),
4573 sizeof(offset), &offset, 0);
4574 if (ret < 0) {
4575 error_setg_errno(errp, -ret, "Failed to update the image size");
4576 goto fail;
4579 s->l1_vm_state_index = new_l1_size;
4581 /* Update cache sizes */
4582 options = qdict_clone_shallow(bs->options);
4583 ret = qcow2_update_options(bs, options, s->flags, errp);
4584 qobject_unref(options);
4585 if (ret < 0) {
4586 goto fail;
4588 ret = 0;
4589 fail:
4590 qemu_co_mutex_unlock(&s->lock);
4591 return ret;
4594 static int coroutine_fn GRAPH_RDLOCK
4595 qcow2_co_pwritev_compressed_task(BlockDriverState *bs,
4596 uint64_t offset, uint64_t bytes,
4597 QEMUIOVector *qiov, size_t qiov_offset)
4599 BDRVQcow2State *s = bs->opaque;
4600 int ret;
4601 ssize_t out_len;
4602 uint8_t *buf, *out_buf;
4603 uint64_t cluster_offset;
4605 assert(bytes == s->cluster_size || (bytes < s->cluster_size &&
4606 (offset + bytes == bs->total_sectors << BDRV_SECTOR_BITS)));
4608 buf = qemu_blockalign(bs, s->cluster_size);
4609 if (bytes < s->cluster_size) {
4610 /* Zero-pad last write if image size is not cluster aligned */
4611 memset(buf + bytes, 0, s->cluster_size - bytes);
4613 qemu_iovec_to_buf(qiov, qiov_offset, buf, bytes);
4615 out_buf = g_malloc(s->cluster_size);
4617 out_len = qcow2_co_compress(bs, out_buf, s->cluster_size - 1,
4618 buf, s->cluster_size);
4619 if (out_len == -ENOMEM) {
4620 /* could not compress: write normal cluster */
4621 ret = qcow2_co_pwritev_part(bs, offset, bytes, qiov, qiov_offset, 0);
4622 if (ret < 0) {
4623 goto fail;
4625 goto success;
4626 } else if (out_len < 0) {
4627 ret = -EINVAL;
4628 goto fail;
4631 qemu_co_mutex_lock(&s->lock);
4632 ret = qcow2_alloc_compressed_cluster_offset(bs, offset, out_len,
4633 &cluster_offset);
4634 if (ret < 0) {
4635 qemu_co_mutex_unlock(&s->lock);
4636 goto fail;
4639 ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len, true);
4640 qemu_co_mutex_unlock(&s->lock);
4641 if (ret < 0) {
4642 goto fail;
4645 BLKDBG_EVENT(s->data_file, BLKDBG_WRITE_COMPRESSED);
4646 ret = bdrv_co_pwrite(s->data_file, cluster_offset, out_len, out_buf, 0);
4647 if (ret < 0) {
4648 goto fail;
4650 success:
4651 ret = 0;
4652 fail:
4653 qemu_vfree(buf);
4654 g_free(out_buf);
4655 return ret;
4659 * This function can count as GRAPH_RDLOCK because
4660 * qcow2_co_pwritev_compressed_part() holds the graph lock and keeps it until
4661 * this coroutine has terminated.
4663 static int coroutine_fn GRAPH_RDLOCK
4664 qcow2_co_pwritev_compressed_task_entry(AioTask *task)
4666 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
4668 assert(!t->subcluster_type && !t->l2meta);
4670 return qcow2_co_pwritev_compressed_task(t->bs, t->offset, t->bytes, t->qiov,
4671 t->qiov_offset);
4675 * XXX: put compressed sectors first, then all the cluster aligned
4676 * tables to avoid losing bytes in alignment
4678 static int coroutine_fn GRAPH_RDLOCK
4679 qcow2_co_pwritev_compressed_part(BlockDriverState *bs,
4680 int64_t offset, int64_t bytes,
4681 QEMUIOVector *qiov, size_t qiov_offset)
4683 BDRVQcow2State *s = bs->opaque;
4684 AioTaskPool *aio = NULL;
4685 int ret = 0;
4687 if (has_data_file(bs)) {
4688 return -ENOTSUP;
4691 if (bytes == 0) {
4693 * align end of file to a sector boundary to ease reading with
4694 * sector based I/Os
4696 int64_t len = bdrv_getlength(bs->file->bs);
4697 if (len < 0) {
4698 return len;
4700 return bdrv_co_truncate(bs->file, len, false, PREALLOC_MODE_OFF, 0,
4701 NULL);
4704 if (offset_into_cluster(s, offset)) {
4705 return -EINVAL;
4708 if (offset_into_cluster(s, bytes) &&
4709 (offset + bytes) != (bs->total_sectors << BDRV_SECTOR_BITS)) {
4710 return -EINVAL;
4713 while (bytes && aio_task_pool_status(aio) == 0) {
4714 uint64_t chunk_size = MIN(bytes, s->cluster_size);
4716 if (!aio && chunk_size != bytes) {
4717 aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
4720 ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_compressed_task_entry,
4721 0, 0, offset, chunk_size, qiov, qiov_offset, NULL);
4722 if (ret < 0) {
4723 break;
4725 qiov_offset += chunk_size;
4726 offset += chunk_size;
4727 bytes -= chunk_size;
4730 if (aio) {
4731 aio_task_pool_wait_all(aio);
4732 if (ret == 0) {
4733 ret = aio_task_pool_status(aio);
4735 g_free(aio);
4738 return ret;
4741 static int coroutine_fn GRAPH_RDLOCK
4742 qcow2_co_preadv_compressed(BlockDriverState *bs,
4743 uint64_t l2_entry,
4744 uint64_t offset,
4745 uint64_t bytes,
4746 QEMUIOVector *qiov,
4747 size_t qiov_offset)
4749 BDRVQcow2State *s = bs->opaque;
4750 int ret = 0, csize;
4751 uint64_t coffset;
4752 uint8_t *buf, *out_buf;
4753 int offset_in_cluster = offset_into_cluster(s, offset);
4755 qcow2_parse_compressed_l2_entry(bs, l2_entry, &coffset, &csize);
4757 buf = g_try_malloc(csize);
4758 if (!buf) {
4759 return -ENOMEM;
4762 out_buf = qemu_blockalign(bs, s->cluster_size);
4764 BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
4765 ret = bdrv_co_pread(bs->file, coffset, csize, buf, 0);
4766 if (ret < 0) {
4767 goto fail;
4770 if (qcow2_co_decompress(bs, out_buf, s->cluster_size, buf, csize) < 0) {
4771 ret = -EIO;
4772 goto fail;
4775 qemu_iovec_from_buf(qiov, qiov_offset, out_buf + offset_in_cluster, bytes);
4777 fail:
4778 qemu_vfree(out_buf);
4779 g_free(buf);
4781 return ret;
4784 static int make_completely_empty(BlockDriverState *bs)
4786 BDRVQcow2State *s = bs->opaque;
4787 Error *local_err = NULL;
4788 int ret, l1_clusters;
4789 int64_t offset;
4790 uint64_t *new_reftable = NULL;
4791 uint64_t rt_entry, l1_size2;
4792 struct {
4793 uint64_t l1_offset;
4794 uint64_t reftable_offset;
4795 uint32_t reftable_clusters;
4796 } QEMU_PACKED l1_ofs_rt_ofs_cls;
4798 ret = qcow2_cache_empty(bs, s->l2_table_cache);
4799 if (ret < 0) {
4800 goto fail;
4803 ret = qcow2_cache_empty(bs, s->refcount_block_cache);
4804 if (ret < 0) {
4805 goto fail;
4808 /* Refcounts will be broken utterly */
4809 ret = qcow2_mark_dirty(bs);
4810 if (ret < 0) {
4811 goto fail;
4814 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
4816 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / L1E_SIZE);
4817 l1_size2 = (uint64_t)s->l1_size * L1E_SIZE;
4819 /* After this call, neither the in-memory nor the on-disk refcount
4820 * information accurately describe the actual references */
4822 ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset,
4823 l1_clusters * s->cluster_size, 0);
4824 if (ret < 0) {
4825 goto fail_broken_refcounts;
4827 memset(s->l1_table, 0, l1_size2);
4829 BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE);
4831 /* Overwrite enough clusters at the beginning of the sectors to place
4832 * the refcount table, a refcount block and the L1 table in; this may
4833 * overwrite parts of the existing refcount and L1 table, which is not
4834 * an issue because the dirty flag is set, complete data loss is in fact
4835 * desired and partial data loss is consequently fine as well */
4836 ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size,
4837 (2 + l1_clusters) * s->cluster_size, 0);
4838 /* This call (even if it failed overall) may have overwritten on-disk
4839 * refcount structures; in that case, the in-memory refcount information
4840 * will probably differ from the on-disk information which makes the BDS
4841 * unusable */
4842 if (ret < 0) {
4843 goto fail_broken_refcounts;
4846 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
4847 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
4849 /* "Create" an empty reftable (one cluster) directly after the image
4850 * header and an empty L1 table three clusters after the image header;
4851 * the cluster between those two will be used as the first refblock */
4852 l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size);
4853 l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size);
4854 l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1);
4855 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset),
4856 sizeof(l1_ofs_rt_ofs_cls), &l1_ofs_rt_ofs_cls, 0);
4857 if (ret < 0) {
4858 goto fail_broken_refcounts;
4861 s->l1_table_offset = 3 * s->cluster_size;
4863 new_reftable = g_try_new0(uint64_t, s->cluster_size / REFTABLE_ENTRY_SIZE);
4864 if (!new_reftable) {
4865 ret = -ENOMEM;
4866 goto fail_broken_refcounts;
4869 s->refcount_table_offset = s->cluster_size;
4870 s->refcount_table_size = s->cluster_size / REFTABLE_ENTRY_SIZE;
4871 s->max_refcount_table_index = 0;
4873 g_free(s->refcount_table);
4874 s->refcount_table = new_reftable;
4875 new_reftable = NULL;
4877 /* Now the in-memory refcount information again corresponds to the on-disk
4878 * information (reftable is empty and no refblocks (the refblock cache is
4879 * empty)); however, this means some clusters (e.g. the image header) are
4880 * referenced, but not refcounted, but the normal qcow2 code assumes that
4881 * the in-memory information is always correct */
4883 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
4885 /* Enter the first refblock into the reftable */
4886 rt_entry = cpu_to_be64(2 * s->cluster_size);
4887 ret = bdrv_pwrite_sync(bs->file, s->cluster_size, sizeof(rt_entry),
4888 &rt_entry, 0);
4889 if (ret < 0) {
4890 goto fail_broken_refcounts;
4892 s->refcount_table[0] = 2 * s->cluster_size;
4894 s->free_cluster_index = 0;
4895 assert(3 + l1_clusters <= s->refcount_block_size);
4896 offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2);
4897 if (offset < 0) {
4898 ret = offset;
4899 goto fail_broken_refcounts;
4900 } else if (offset > 0) {
4901 error_report("First cluster in emptied image is in use");
4902 abort();
4905 /* Now finally the in-memory information corresponds to the on-disk
4906 * structures and is correct */
4907 ret = qcow2_mark_clean(bs);
4908 if (ret < 0) {
4909 goto fail;
4912 ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size, false,
4913 PREALLOC_MODE_OFF, 0, &local_err);
4914 if (ret < 0) {
4915 error_report_err(local_err);
4916 goto fail;
4919 return 0;
4921 fail_broken_refcounts:
4922 /* The BDS is unusable at this point. If we wanted to make it usable, we
4923 * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
4924 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
4925 * again. However, because the functions which could have caused this error
4926 * path to be taken are used by those functions as well, it's very likely
4927 * that that sequence will fail as well. Therefore, just eject the BDS. */
4928 bs->drv = NULL;
4930 fail:
4931 g_free(new_reftable);
4932 return ret;
4935 static int qcow2_make_empty(BlockDriverState *bs)
4937 BDRVQcow2State *s = bs->opaque;
4938 uint64_t offset, end_offset;
4939 int step = QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size);
4940 int l1_clusters, ret = 0;
4942 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / L1E_SIZE);
4944 if (s->qcow_version >= 3 && !s->snapshots && !s->nb_bitmaps &&
4945 3 + l1_clusters <= s->refcount_block_size &&
4946 s->crypt_method_header != QCOW_CRYPT_LUKS &&
4947 !has_data_file(bs)) {
4948 /* The following function only works for qcow2 v3 images (it
4949 * requires the dirty flag) and only as long as there are no
4950 * features that reserve extra clusters (such as snapshots,
4951 * LUKS header, or persistent bitmaps), because it completely
4952 * empties the image. Furthermore, the L1 table and three
4953 * additional clusters (image header, refcount table, one
4954 * refcount block) have to fit inside one refcount block. It
4955 * only resets the image file, i.e. does not work with an
4956 * external data file. */
4957 return make_completely_empty(bs);
4960 /* This fallback code simply discards every active cluster; this is slow,
4961 * but works in all cases */
4962 end_offset = bs->total_sectors * BDRV_SECTOR_SIZE;
4963 for (offset = 0; offset < end_offset; offset += step) {
4964 /* As this function is generally used after committing an external
4965 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
4966 * default action for this kind of discard is to pass the discard,
4967 * which will ideally result in an actually smaller image file, as
4968 * is probably desired. */
4969 ret = qcow2_cluster_discard(bs, offset, MIN(step, end_offset - offset),
4970 QCOW2_DISCARD_SNAPSHOT, true);
4971 if (ret < 0) {
4972 break;
4976 return ret;
4979 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
4981 BDRVQcow2State *s = bs->opaque;
4982 int ret;
4984 qemu_co_mutex_lock(&s->lock);
4985 ret = qcow2_write_caches(bs);
4986 qemu_co_mutex_unlock(&s->lock);
4988 return ret;
4991 static BlockMeasureInfo *qcow2_measure(QemuOpts *opts, BlockDriverState *in_bs,
4992 Error **errp)
4994 Error *local_err = NULL;
4995 BlockMeasureInfo *info;
4996 uint64_t required = 0; /* bytes that contribute to required size */
4997 uint64_t virtual_size; /* disk size as seen by guest */
4998 uint64_t refcount_bits;
4999 uint64_t l2_tables;
5000 uint64_t luks_payload_size = 0;
5001 size_t cluster_size;
5002 int version;
5003 char *optstr;
5004 PreallocMode prealloc;
5005 bool has_backing_file;
5006 bool has_luks;
5007 bool extended_l2;
5008 size_t l2e_size;
5010 /* Parse image creation options */
5011 extended_l2 = qemu_opt_get_bool_del(opts, BLOCK_OPT_EXTL2, false);
5013 cluster_size = qcow2_opt_get_cluster_size_del(opts, extended_l2,
5014 &local_err);
5015 if (local_err) {
5016 goto err;
5019 version = qcow2_opt_get_version_del(opts, &local_err);
5020 if (local_err) {
5021 goto err;
5024 refcount_bits = qcow2_opt_get_refcount_bits_del(opts, version, &local_err);
5025 if (local_err) {
5026 goto err;
5029 optstr = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
5030 prealloc = qapi_enum_parse(&PreallocMode_lookup, optstr,
5031 PREALLOC_MODE_OFF, &local_err);
5032 g_free(optstr);
5033 if (local_err) {
5034 goto err;
5037 optstr = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
5038 has_backing_file = !!optstr;
5039 g_free(optstr);
5041 optstr = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT);
5042 has_luks = optstr && strcmp(optstr, "luks") == 0;
5043 g_free(optstr);
5045 if (has_luks) {
5046 g_autoptr(QCryptoBlockCreateOptions) create_opts = NULL;
5047 QDict *cryptoopts = qcow2_extract_crypto_opts(opts, "luks", errp);
5048 size_t headerlen;
5050 create_opts = block_crypto_create_opts_init(cryptoopts, errp);
5051 qobject_unref(cryptoopts);
5052 if (!create_opts) {
5053 goto err;
5056 if (!qcrypto_block_calculate_payload_offset(create_opts,
5057 "encrypt.",
5058 &headerlen,
5059 &local_err)) {
5060 goto err;
5063 luks_payload_size = ROUND_UP(headerlen, cluster_size);
5066 virtual_size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
5067 virtual_size = ROUND_UP(virtual_size, cluster_size);
5069 /* Check that virtual disk size is valid */
5070 l2e_size = extended_l2 ? L2E_SIZE_EXTENDED : L2E_SIZE_NORMAL;
5071 l2_tables = DIV_ROUND_UP(virtual_size / cluster_size,
5072 cluster_size / l2e_size);
5073 if (l2_tables * L1E_SIZE > QCOW_MAX_L1_SIZE) {
5074 error_setg(&local_err, "The image size is too large "
5075 "(try using a larger cluster size)");
5076 goto err;
5079 /* Account for input image */
5080 if (in_bs) {
5081 int64_t ssize = bdrv_getlength(in_bs);
5082 if (ssize < 0) {
5083 error_setg_errno(&local_err, -ssize,
5084 "Unable to get image virtual_size");
5085 goto err;
5088 virtual_size = ROUND_UP(ssize, cluster_size);
5090 if (has_backing_file) {
5091 /* We don't how much of the backing chain is shared by the input
5092 * image and the new image file. In the worst case the new image's
5093 * backing file has nothing in common with the input image. Be
5094 * conservative and assume all clusters need to be written.
5096 required = virtual_size;
5097 } else {
5098 int64_t offset;
5099 int64_t pnum = 0;
5101 for (offset = 0; offset < ssize; offset += pnum) {
5102 int ret;
5104 ret = bdrv_block_status_above(in_bs, NULL, offset,
5105 ssize - offset, &pnum, NULL,
5106 NULL);
5107 if (ret < 0) {
5108 error_setg_errno(&local_err, -ret,
5109 "Unable to get block status");
5110 goto err;
5113 if (ret & BDRV_BLOCK_ZERO) {
5114 /* Skip zero regions (safe with no backing file) */
5115 } else if ((ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) ==
5116 (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) {
5117 /* Extend pnum to end of cluster for next iteration */
5118 pnum = ROUND_UP(offset + pnum, cluster_size) - offset;
5120 /* Count clusters we've seen */
5121 required += offset % cluster_size + pnum;
5127 /* Take into account preallocation. Nothing special is needed for
5128 * PREALLOC_MODE_METADATA since metadata is always counted.
5130 if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
5131 required = virtual_size;
5134 info = g_new0(BlockMeasureInfo, 1);
5135 info->fully_allocated = luks_payload_size +
5136 qcow2_calc_prealloc_size(virtual_size, cluster_size,
5137 ctz32(refcount_bits), extended_l2);
5140 * Remove data clusters that are not required. This overestimates the
5141 * required size because metadata needed for the fully allocated file is
5142 * still counted. Show bitmaps only if both source and destination
5143 * would support them.
5145 info->required = info->fully_allocated - virtual_size + required;
5146 info->has_bitmaps = version >= 3 && in_bs &&
5147 bdrv_supports_persistent_dirty_bitmap(in_bs);
5148 if (info->has_bitmaps) {
5149 info->bitmaps = qcow2_get_persistent_dirty_bitmap_size(in_bs,
5150 cluster_size);
5152 return info;
5154 err:
5155 error_propagate(errp, local_err);
5156 return NULL;
5159 static int coroutine_fn
5160 qcow2_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
5162 BDRVQcow2State *s = bs->opaque;
5163 bdi->cluster_size = s->cluster_size;
5164 bdi->vm_state_offset = qcow2_vm_state_offset(s);
5165 bdi->is_dirty = s->incompatible_features & QCOW2_INCOMPAT_DIRTY;
5166 return 0;
5169 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs,
5170 Error **errp)
5172 BDRVQcow2State *s = bs->opaque;
5173 ImageInfoSpecific *spec_info;
5174 QCryptoBlockInfo *encrypt_info = NULL;
5176 if (s->crypto != NULL) {
5177 encrypt_info = qcrypto_block_get_info(s->crypto, errp);
5178 if (!encrypt_info) {
5179 return NULL;
5183 spec_info = g_new(ImageInfoSpecific, 1);
5184 *spec_info = (ImageInfoSpecific){
5185 .type = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
5186 .u.qcow2.data = g_new0(ImageInfoSpecificQCow2, 1),
5188 if (s->qcow_version == 2) {
5189 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
5190 .compat = g_strdup("0.10"),
5191 .refcount_bits = s->refcount_bits,
5193 } else if (s->qcow_version == 3) {
5194 Qcow2BitmapInfoList *bitmaps;
5195 if (!qcow2_get_bitmap_info_list(bs, &bitmaps, errp)) {
5196 qapi_free_ImageInfoSpecific(spec_info);
5197 qapi_free_QCryptoBlockInfo(encrypt_info);
5198 return NULL;
5200 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
5201 .compat = g_strdup("1.1"),
5202 .lazy_refcounts = s->compatible_features &
5203 QCOW2_COMPAT_LAZY_REFCOUNTS,
5204 .has_lazy_refcounts = true,
5205 .corrupt = s->incompatible_features &
5206 QCOW2_INCOMPAT_CORRUPT,
5207 .has_corrupt = true,
5208 .has_extended_l2 = true,
5209 .extended_l2 = has_subclusters(s),
5210 .refcount_bits = s->refcount_bits,
5211 .has_bitmaps = !!bitmaps,
5212 .bitmaps = bitmaps,
5213 .data_file = g_strdup(s->image_data_file),
5214 .has_data_file_raw = has_data_file(bs),
5215 .data_file_raw = data_file_is_raw(bs),
5216 .compression_type = s->compression_type,
5218 } else {
5219 /* if this assertion fails, this probably means a new version was
5220 * added without having it covered here */
5221 assert(false);
5224 if (encrypt_info) {
5225 ImageInfoSpecificQCow2Encryption *qencrypt =
5226 g_new(ImageInfoSpecificQCow2Encryption, 1);
5227 switch (encrypt_info->format) {
5228 case Q_CRYPTO_BLOCK_FORMAT_QCOW:
5229 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES;
5230 break;
5231 case Q_CRYPTO_BLOCK_FORMAT_LUKS:
5232 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS;
5233 qencrypt->u.luks = encrypt_info->u.luks;
5234 break;
5235 default:
5236 abort();
5238 /* Since we did shallow copy above, erase any pointers
5239 * in the original info */
5240 memset(&encrypt_info->u, 0, sizeof(encrypt_info->u));
5241 qapi_free_QCryptoBlockInfo(encrypt_info);
5243 spec_info->u.qcow2.data->encrypt = qencrypt;
5246 return spec_info;
5249 static int qcow2_has_zero_init(BlockDriverState *bs)
5251 BDRVQcow2State *s = bs->opaque;
5252 bool preallocated;
5254 if (qemu_in_coroutine()) {
5255 qemu_co_mutex_lock(&s->lock);
5258 * Check preallocation status: Preallocated images have all L2
5259 * tables allocated, nonpreallocated images have none. It is
5260 * therefore enough to check the first one.
5262 preallocated = s->l1_size > 0 && s->l1_table[0] != 0;
5263 if (qemu_in_coroutine()) {
5264 qemu_co_mutex_unlock(&s->lock);
5267 if (!preallocated) {
5268 return 1;
5269 } else if (bs->encrypted) {
5270 return 0;
5271 } else {
5272 return bdrv_has_zero_init(s->data_file->bs);
5277 * Check the request to vmstate. On success return
5278 * qcow2_vm_state_offset(bs) + @pos
5280 static int64_t qcow2_check_vmstate_request(BlockDriverState *bs,
5281 QEMUIOVector *qiov, int64_t pos)
5283 BDRVQcow2State *s = bs->opaque;
5284 int64_t vmstate_offset = qcow2_vm_state_offset(s);
5285 int ret;
5287 /* Incoming requests must be OK */
5288 bdrv_check_qiov_request(pos, qiov->size, qiov, 0, &error_abort);
5290 if (INT64_MAX - pos < vmstate_offset) {
5291 return -EIO;
5294 pos += vmstate_offset;
5295 ret = bdrv_check_qiov_request(pos, qiov->size, qiov, 0, NULL);
5296 if (ret < 0) {
5297 return ret;
5300 return pos;
5303 static int coroutine_fn GRAPH_RDLOCK
5304 qcow2_co_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
5306 int64_t offset = qcow2_check_vmstate_request(bs, qiov, pos);
5307 if (offset < 0) {
5308 return offset;
5311 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
5312 return bs->drv->bdrv_co_pwritev_part(bs, offset, qiov->size, qiov, 0, 0);
5315 static int coroutine_fn GRAPH_RDLOCK
5316 qcow2_co_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
5318 int64_t offset = qcow2_check_vmstate_request(bs, qiov, pos);
5319 if (offset < 0) {
5320 return offset;
5323 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
5324 return bs->drv->bdrv_co_preadv_part(bs, offset, qiov->size, qiov, 0, 0);
5327 static int qcow2_has_compressed_clusters(BlockDriverState *bs)
5329 int64_t offset = 0;
5330 int64_t bytes = bdrv_getlength(bs);
5332 if (bytes < 0) {
5333 return bytes;
5336 while (bytes != 0) {
5337 int ret;
5338 QCow2SubclusterType type;
5339 unsigned int cur_bytes = MIN(INT_MAX, bytes);
5340 uint64_t host_offset;
5342 ret = qcow2_get_host_offset(bs, offset, &cur_bytes, &host_offset,
5343 &type);
5344 if (ret < 0) {
5345 return ret;
5348 if (type == QCOW2_SUBCLUSTER_COMPRESSED) {
5349 return 1;
5352 offset += cur_bytes;
5353 bytes -= cur_bytes;
5356 return 0;
5360 * Downgrades an image's version. To achieve this, any incompatible features
5361 * have to be removed.
5363 static int qcow2_downgrade(BlockDriverState *bs, int target_version,
5364 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
5365 Error **errp)
5367 BDRVQcow2State *s = bs->opaque;
5368 int current_version = s->qcow_version;
5369 int ret;
5370 int i;
5372 /* This is qcow2_downgrade(), not qcow2_upgrade() */
5373 assert(target_version < current_version);
5375 /* There are no other versions (now) that you can downgrade to */
5376 assert(target_version == 2);
5378 if (s->refcount_order != 4) {
5379 error_setg(errp, "compat=0.10 requires refcount_bits=16");
5380 return -ENOTSUP;
5383 if (has_data_file(bs)) {
5384 error_setg(errp, "Cannot downgrade an image with a data file");
5385 return -ENOTSUP;
5389 * If any internal snapshot has a different size than the current
5390 * image size, or VM state size that exceeds 32 bits, downgrading
5391 * is unsafe. Even though we would still use v3-compliant output
5392 * to preserve that data, other v2 programs might not realize
5393 * those optional fields are important.
5395 for (i = 0; i < s->nb_snapshots; i++) {
5396 if (s->snapshots[i].vm_state_size > UINT32_MAX ||
5397 s->snapshots[i].disk_size != bs->total_sectors * BDRV_SECTOR_SIZE) {
5398 error_setg(errp, "Internal snapshots prevent downgrade of image");
5399 return -ENOTSUP;
5403 /* clear incompatible features */
5404 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
5405 ret = qcow2_mark_clean(bs);
5406 if (ret < 0) {
5407 error_setg_errno(errp, -ret, "Failed to make the image clean");
5408 return ret;
5412 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
5413 * the first place; if that happens nonetheless, returning -ENOTSUP is the
5414 * best thing to do anyway */
5416 if (s->incompatible_features & ~QCOW2_INCOMPAT_COMPRESSION) {
5417 error_setg(errp, "Cannot downgrade an image with incompatible features "
5418 "0x%" PRIx64 " set",
5419 s->incompatible_features & ~QCOW2_INCOMPAT_COMPRESSION);
5420 return -ENOTSUP;
5423 /* since we can ignore compatible features, we can set them to 0 as well */
5424 s->compatible_features = 0;
5425 /* if lazy refcounts have been used, they have already been fixed through
5426 * clearing the dirty flag */
5428 /* clearing autoclear features is trivial */
5429 s->autoclear_features = 0;
5431 ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque);
5432 if (ret < 0) {
5433 error_setg_errno(errp, -ret, "Failed to turn zero into data clusters");
5434 return ret;
5437 if (s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION) {
5438 ret = qcow2_has_compressed_clusters(bs);
5439 if (ret < 0) {
5440 error_setg(errp, "Failed to check block status");
5441 return -EINVAL;
5443 if (ret) {
5444 error_setg(errp, "Cannot downgrade an image with zstd compression "
5445 "type and existing compressed clusters");
5446 return -ENOTSUP;
5449 * No compressed clusters for now, so just chose default zlib
5450 * compression.
5452 s->incompatible_features &= ~QCOW2_INCOMPAT_COMPRESSION;
5453 s->compression_type = QCOW2_COMPRESSION_TYPE_ZLIB;
5456 assert(s->incompatible_features == 0);
5458 s->qcow_version = target_version;
5459 ret = qcow2_update_header(bs);
5460 if (ret < 0) {
5461 s->qcow_version = current_version;
5462 error_setg_errno(errp, -ret, "Failed to update the image header");
5463 return ret;
5465 return 0;
5469 * Upgrades an image's version. While newer versions encompass all
5470 * features of older versions, some things may have to be presented
5471 * differently.
5473 static int qcow2_upgrade(BlockDriverState *bs, int target_version,
5474 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
5475 Error **errp)
5477 BDRVQcow2State *s = bs->opaque;
5478 bool need_snapshot_update;
5479 int current_version = s->qcow_version;
5480 int i;
5481 int ret;
5483 /* This is qcow2_upgrade(), not qcow2_downgrade() */
5484 assert(target_version > current_version);
5486 /* There are no other versions (yet) that you can upgrade to */
5487 assert(target_version == 3);
5489 status_cb(bs, 0, 2, cb_opaque);
5492 * In v2, snapshots do not need to have extra data. v3 requires
5493 * the 64-bit VM state size and the virtual disk size to be
5494 * present.
5495 * qcow2_write_snapshots() will always write the list in the
5496 * v3-compliant format.
5498 need_snapshot_update = false;
5499 for (i = 0; i < s->nb_snapshots; i++) {
5500 if (s->snapshots[i].extra_data_size <
5501 sizeof_field(QCowSnapshotExtraData, vm_state_size_large) +
5502 sizeof_field(QCowSnapshotExtraData, disk_size))
5504 need_snapshot_update = true;
5505 break;
5508 if (need_snapshot_update) {
5509 ret = qcow2_write_snapshots(bs);
5510 if (ret < 0) {
5511 error_setg_errno(errp, -ret, "Failed to update the snapshot table");
5512 return ret;
5515 status_cb(bs, 1, 2, cb_opaque);
5517 s->qcow_version = target_version;
5518 ret = qcow2_update_header(bs);
5519 if (ret < 0) {
5520 s->qcow_version = current_version;
5521 error_setg_errno(errp, -ret, "Failed to update the image header");
5522 return ret;
5524 status_cb(bs, 2, 2, cb_opaque);
5526 return 0;
5529 typedef enum Qcow2AmendOperation {
5530 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
5531 * statically initialized to so that the helper CB can discern the first
5532 * invocation from an operation change */
5533 QCOW2_NO_OPERATION = 0,
5535 QCOW2_UPGRADING,
5536 QCOW2_UPDATING_ENCRYPTION,
5537 QCOW2_CHANGING_REFCOUNT_ORDER,
5538 QCOW2_DOWNGRADING,
5539 } Qcow2AmendOperation;
5541 typedef struct Qcow2AmendHelperCBInfo {
5542 /* The code coordinating the amend operations should only modify
5543 * these four fields; the rest will be managed by the CB */
5544 BlockDriverAmendStatusCB *original_status_cb;
5545 void *original_cb_opaque;
5547 Qcow2AmendOperation current_operation;
5549 /* Total number of operations to perform (only set once) */
5550 int total_operations;
5552 /* The following fields are managed by the CB */
5554 /* Number of operations completed */
5555 int operations_completed;
5557 /* Cumulative offset of all completed operations */
5558 int64_t offset_completed;
5560 Qcow2AmendOperation last_operation;
5561 int64_t last_work_size;
5562 } Qcow2AmendHelperCBInfo;
5564 static void qcow2_amend_helper_cb(BlockDriverState *bs,
5565 int64_t operation_offset,
5566 int64_t operation_work_size, void *opaque)
5568 Qcow2AmendHelperCBInfo *info = opaque;
5569 int64_t current_work_size;
5570 int64_t projected_work_size;
5572 if (info->current_operation != info->last_operation) {
5573 if (info->last_operation != QCOW2_NO_OPERATION) {
5574 info->offset_completed += info->last_work_size;
5575 info->operations_completed++;
5578 info->last_operation = info->current_operation;
5581 assert(info->total_operations > 0);
5582 assert(info->operations_completed < info->total_operations);
5584 info->last_work_size = operation_work_size;
5586 current_work_size = info->offset_completed + operation_work_size;
5588 /* current_work_size is the total work size for (operations_completed + 1)
5589 * operations (which includes this one), so multiply it by the number of
5590 * operations not covered and divide it by the number of operations
5591 * covered to get a projection for the operations not covered */
5592 projected_work_size = current_work_size * (info->total_operations -
5593 info->operations_completed - 1)
5594 / (info->operations_completed + 1);
5596 info->original_status_cb(bs, info->offset_completed + operation_offset,
5597 current_work_size + projected_work_size,
5598 info->original_cb_opaque);
5601 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,
5602 BlockDriverAmendStatusCB *status_cb,
5603 void *cb_opaque,
5604 bool force,
5605 Error **errp)
5607 BDRVQcow2State *s = bs->opaque;
5608 int old_version = s->qcow_version, new_version = old_version;
5609 uint64_t new_size = 0;
5610 const char *backing_file = NULL, *backing_format = NULL, *data_file = NULL;
5611 bool lazy_refcounts = s->use_lazy_refcounts;
5612 bool data_file_raw = data_file_is_raw(bs);
5613 const char *compat = NULL;
5614 int refcount_bits = s->refcount_bits;
5615 int ret;
5616 QemuOptDesc *desc = opts->list->desc;
5617 Qcow2AmendHelperCBInfo helper_cb_info;
5618 bool encryption_update = false;
5620 while (desc && desc->name) {
5621 if (!qemu_opt_find(opts, desc->name)) {
5622 /* only change explicitly defined options */
5623 desc++;
5624 continue;
5627 if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) {
5628 compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL);
5629 if (!compat) {
5630 /* preserve default */
5631 } else if (!strcmp(compat, "0.10") || !strcmp(compat, "v2")) {
5632 new_version = 2;
5633 } else if (!strcmp(compat, "1.1") || !strcmp(compat, "v3")) {
5634 new_version = 3;
5635 } else {
5636 error_setg(errp, "Unknown compatibility level %s", compat);
5637 return -EINVAL;
5639 } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) {
5640 new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
5641 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) {
5642 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
5643 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) {
5644 backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
5645 } else if (g_str_has_prefix(desc->name, "encrypt.")) {
5646 if (!s->crypto) {
5647 error_setg(errp,
5648 "Can't amend encryption options - encryption not present");
5649 return -EINVAL;
5651 if (s->crypt_method_header != QCOW_CRYPT_LUKS) {
5652 error_setg(errp,
5653 "Only LUKS encryption options can be amended");
5654 return -ENOTSUP;
5656 encryption_update = true;
5657 } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
5658 lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS,
5659 lazy_refcounts);
5660 } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) {
5661 refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS,
5662 refcount_bits);
5664 if (refcount_bits <= 0 || refcount_bits > 64 ||
5665 !is_power_of_2(refcount_bits))
5667 error_setg(errp, "Refcount width must be a power of two and "
5668 "may not exceed 64 bits");
5669 return -EINVAL;
5671 } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE)) {
5672 data_file = qemu_opt_get(opts, BLOCK_OPT_DATA_FILE);
5673 if (data_file && !has_data_file(bs)) {
5674 error_setg(errp, "data-file can only be set for images that "
5675 "use an external data file");
5676 return -EINVAL;
5678 } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE_RAW)) {
5679 data_file_raw = qemu_opt_get_bool(opts, BLOCK_OPT_DATA_FILE_RAW,
5680 data_file_raw);
5681 if (data_file_raw && !data_file_is_raw(bs)) {
5682 error_setg(errp, "data-file-raw cannot be set on existing "
5683 "images");
5684 return -EINVAL;
5686 } else {
5687 /* if this point is reached, this probably means a new option was
5688 * added without having it covered here */
5689 abort();
5692 desc++;
5695 helper_cb_info = (Qcow2AmendHelperCBInfo){
5696 .original_status_cb = status_cb,
5697 .original_cb_opaque = cb_opaque,
5698 .total_operations = (new_version != old_version)
5699 + (s->refcount_bits != refcount_bits) +
5700 (encryption_update == true)
5703 /* Upgrade first (some features may require compat=1.1) */
5704 if (new_version > old_version) {
5705 helper_cb_info.current_operation = QCOW2_UPGRADING;
5706 ret = qcow2_upgrade(bs, new_version, &qcow2_amend_helper_cb,
5707 &helper_cb_info, errp);
5708 if (ret < 0) {
5709 return ret;
5713 if (encryption_update) {
5714 QDict *amend_opts_dict;
5715 QCryptoBlockAmendOptions *amend_opts;
5717 helper_cb_info.current_operation = QCOW2_UPDATING_ENCRYPTION;
5718 amend_opts_dict = qcow2_extract_crypto_opts(opts, "luks", errp);
5719 if (!amend_opts_dict) {
5720 return -EINVAL;
5722 amend_opts = block_crypto_amend_opts_init(amend_opts_dict, errp);
5723 qobject_unref(amend_opts_dict);
5724 if (!amend_opts) {
5725 return -EINVAL;
5727 ret = qcrypto_block_amend_options(s->crypto,
5728 qcow2_crypto_hdr_read_func,
5729 qcow2_crypto_hdr_write_func,
5731 amend_opts,
5732 force,
5733 errp);
5734 qapi_free_QCryptoBlockAmendOptions(amend_opts);
5735 if (ret < 0) {
5736 return ret;
5740 if (s->refcount_bits != refcount_bits) {
5741 int refcount_order = ctz32(refcount_bits);
5743 if (new_version < 3 && refcount_bits != 16) {
5744 error_setg(errp, "Refcount widths other than 16 bits require "
5745 "compatibility level 1.1 or above (use compat=1.1 or "
5746 "greater)");
5747 return -EINVAL;
5750 helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER;
5751 ret = qcow2_change_refcount_order(bs, refcount_order,
5752 &qcow2_amend_helper_cb,
5753 &helper_cb_info, errp);
5754 if (ret < 0) {
5755 return ret;
5759 /* data-file-raw blocks backing files, so clear it first if requested */
5760 if (data_file_raw) {
5761 s->autoclear_features |= QCOW2_AUTOCLEAR_DATA_FILE_RAW;
5762 } else {
5763 s->autoclear_features &= ~QCOW2_AUTOCLEAR_DATA_FILE_RAW;
5766 if (data_file) {
5767 g_free(s->image_data_file);
5768 s->image_data_file = *data_file ? g_strdup(data_file) : NULL;
5771 ret = qcow2_update_header(bs);
5772 if (ret < 0) {
5773 error_setg_errno(errp, -ret, "Failed to update the image header");
5774 return ret;
5777 if (backing_file || backing_format) {
5778 if (g_strcmp0(backing_file, s->image_backing_file) ||
5779 g_strcmp0(backing_format, s->image_backing_format)) {
5780 error_setg(errp, "Cannot amend the backing file");
5781 error_append_hint(errp,
5782 "You can use 'qemu-img rebase' instead.\n");
5783 return -EINVAL;
5787 if (s->use_lazy_refcounts != lazy_refcounts) {
5788 if (lazy_refcounts) {
5789 if (new_version < 3) {
5790 error_setg(errp, "Lazy refcounts only supported with "
5791 "compatibility level 1.1 and above (use compat=1.1 "
5792 "or greater)");
5793 return -EINVAL;
5795 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
5796 ret = qcow2_update_header(bs);
5797 if (ret < 0) {
5798 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
5799 error_setg_errno(errp, -ret, "Failed to update the image header");
5800 return ret;
5802 s->use_lazy_refcounts = true;
5803 } else {
5804 /* make image clean first */
5805 ret = qcow2_mark_clean(bs);
5806 if (ret < 0) {
5807 error_setg_errno(errp, -ret, "Failed to make the image clean");
5808 return ret;
5810 /* now disallow lazy refcounts */
5811 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
5812 ret = qcow2_update_header(bs);
5813 if (ret < 0) {
5814 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
5815 error_setg_errno(errp, -ret, "Failed to update the image header");
5816 return ret;
5818 s->use_lazy_refcounts = false;
5822 if (new_size) {
5823 BlockBackend *blk = blk_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL,
5824 errp);
5825 if (!blk) {
5826 return -EPERM;
5830 * Amending image options should ensure that the image has
5831 * exactly the given new values, so pass exact=true here.
5833 ret = blk_truncate(blk, new_size, true, PREALLOC_MODE_OFF, 0, errp);
5834 blk_unref(blk);
5835 if (ret < 0) {
5836 return ret;
5840 /* Downgrade last (so unsupported features can be removed before) */
5841 if (new_version < old_version) {
5842 helper_cb_info.current_operation = QCOW2_DOWNGRADING;
5843 ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb,
5844 &helper_cb_info, errp);
5845 if (ret < 0) {
5846 return ret;
5850 return 0;
5853 static int coroutine_fn qcow2_co_amend(BlockDriverState *bs,
5854 BlockdevAmendOptions *opts,
5855 bool force,
5856 Error **errp)
5858 BlockdevAmendOptionsQcow2 *qopts = &opts->u.qcow2;
5859 BDRVQcow2State *s = bs->opaque;
5860 int ret = 0;
5862 if (qopts->encrypt) {
5863 if (!s->crypto) {
5864 error_setg(errp, "image is not encrypted, can't amend");
5865 return -EOPNOTSUPP;
5868 if (qopts->encrypt->format != Q_CRYPTO_BLOCK_FORMAT_LUKS) {
5869 error_setg(errp,
5870 "Amend can't be used to change the qcow2 encryption format");
5871 return -EOPNOTSUPP;
5874 if (s->crypt_method_header != QCOW_CRYPT_LUKS) {
5875 error_setg(errp,
5876 "Only LUKS encryption options can be amended for qcow2 with blockdev-amend");
5877 return -EOPNOTSUPP;
5880 ret = qcrypto_block_amend_options(s->crypto,
5881 qcow2_crypto_hdr_read_func,
5882 qcow2_crypto_hdr_write_func,
5884 qopts->encrypt,
5885 force,
5886 errp);
5888 return ret;
5892 * If offset or size are negative, respectively, they will not be included in
5893 * the BLOCK_IMAGE_CORRUPTED event emitted.
5894 * fatal will be ignored for read-only BDS; corruptions found there will always
5895 * be considered non-fatal.
5897 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
5898 int64_t size, const char *message_format, ...)
5900 BDRVQcow2State *s = bs->opaque;
5901 const char *node_name;
5902 char *message;
5903 va_list ap;
5905 fatal = fatal && bdrv_is_writable(bs);
5907 if (s->signaled_corruption &&
5908 (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
5910 return;
5913 va_start(ap, message_format);
5914 message = g_strdup_vprintf(message_format, ap);
5915 va_end(ap);
5917 if (fatal) {
5918 fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
5919 "corruption events will be suppressed\n", message);
5920 } else {
5921 fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
5922 "corruption events will be suppressed\n", message);
5925 node_name = bdrv_get_node_name(bs);
5926 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs),
5927 *node_name ? node_name : NULL,
5928 message, offset >= 0, offset,
5929 size >= 0, size,
5930 fatal);
5931 g_free(message);
5933 if (fatal) {
5934 qcow2_mark_corrupt(bs);
5935 bs->drv = NULL; /* make BDS unusable */
5938 s->signaled_corruption = true;
5941 #define QCOW_COMMON_OPTIONS \
5943 .name = BLOCK_OPT_SIZE, \
5944 .type = QEMU_OPT_SIZE, \
5945 .help = "Virtual disk size" \
5946 }, \
5948 .name = BLOCK_OPT_COMPAT_LEVEL, \
5949 .type = QEMU_OPT_STRING, \
5950 .help = "Compatibility level (v2 [0.10] or v3 [1.1])" \
5951 }, \
5953 .name = BLOCK_OPT_BACKING_FILE, \
5954 .type = QEMU_OPT_STRING, \
5955 .help = "File name of a base image" \
5956 }, \
5958 .name = BLOCK_OPT_BACKING_FMT, \
5959 .type = QEMU_OPT_STRING, \
5960 .help = "Image format of the base image" \
5961 }, \
5963 .name = BLOCK_OPT_DATA_FILE, \
5964 .type = QEMU_OPT_STRING, \
5965 .help = "File name of an external data file" \
5966 }, \
5968 .name = BLOCK_OPT_DATA_FILE_RAW, \
5969 .type = QEMU_OPT_BOOL, \
5970 .help = "The external data file must stay valid " \
5971 "as a raw image" \
5972 }, \
5974 .name = BLOCK_OPT_LAZY_REFCOUNTS, \
5975 .type = QEMU_OPT_BOOL, \
5976 .help = "Postpone refcount updates", \
5977 .def_value_str = "off" \
5978 }, \
5980 .name = BLOCK_OPT_REFCOUNT_BITS, \
5981 .type = QEMU_OPT_NUMBER, \
5982 .help = "Width of a reference count entry in bits", \
5983 .def_value_str = "16" \
5986 static QemuOptsList qcow2_create_opts = {
5987 .name = "qcow2-create-opts",
5988 .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
5989 .desc = {
5991 .name = BLOCK_OPT_ENCRYPT, \
5992 .type = QEMU_OPT_BOOL, \
5993 .help = "Encrypt the image with format 'aes'. (Deprecated " \
5994 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)", \
5995 }, \
5997 .name = BLOCK_OPT_ENCRYPT_FORMAT, \
5998 .type = QEMU_OPT_STRING, \
5999 .help = "Encrypt the image, format choices: 'aes', 'luks'", \
6000 }, \
6001 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", \
6002 "ID of secret providing qcow AES key or LUKS passphrase"), \
6003 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."), \
6004 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."), \
6005 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."), \
6006 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."), \
6007 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."), \
6008 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."), \
6010 .name = BLOCK_OPT_CLUSTER_SIZE, \
6011 .type = QEMU_OPT_SIZE, \
6012 .help = "qcow2 cluster size", \
6013 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE) \
6014 }, \
6016 .name = BLOCK_OPT_EXTL2, \
6017 .type = QEMU_OPT_BOOL, \
6018 .help = "Extended L2 tables", \
6019 .def_value_str = "off" \
6020 }, \
6022 .name = BLOCK_OPT_PREALLOC, \
6023 .type = QEMU_OPT_STRING, \
6024 .help = "Preallocation mode (allowed values: off, " \
6025 "metadata, falloc, full)" \
6026 }, \
6028 .name = BLOCK_OPT_COMPRESSION_TYPE, \
6029 .type = QEMU_OPT_STRING, \
6030 .help = "Compression method used for image cluster " \
6031 "compression", \
6032 .def_value_str = "zlib" \
6034 QCOW_COMMON_OPTIONS,
6035 { /* end of list */ }
6039 static QemuOptsList qcow2_amend_opts = {
6040 .name = "qcow2-amend-opts",
6041 .head = QTAILQ_HEAD_INITIALIZER(qcow2_amend_opts.head),
6042 .desc = {
6043 BLOCK_CRYPTO_OPT_DEF_LUKS_STATE("encrypt."),
6044 BLOCK_CRYPTO_OPT_DEF_LUKS_KEYSLOT("encrypt."),
6045 BLOCK_CRYPTO_OPT_DEF_LUKS_OLD_SECRET("encrypt."),
6046 BLOCK_CRYPTO_OPT_DEF_LUKS_NEW_SECRET("encrypt."),
6047 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
6048 QCOW_COMMON_OPTIONS,
6049 { /* end of list */ }
6053 static const char *const qcow2_strong_runtime_opts[] = {
6054 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET,
6056 NULL
6059 BlockDriver bdrv_qcow2 = {
6060 .format_name = "qcow2",
6061 .instance_size = sizeof(BDRVQcow2State),
6062 .bdrv_probe = qcow2_probe,
6063 .bdrv_open = qcow2_open,
6064 .bdrv_close = qcow2_close,
6065 .bdrv_reopen_prepare = qcow2_reopen_prepare,
6066 .bdrv_reopen_commit = qcow2_reopen_commit,
6067 .bdrv_reopen_commit_post = qcow2_reopen_commit_post,
6068 .bdrv_reopen_abort = qcow2_reopen_abort,
6069 .bdrv_join_options = qcow2_join_options,
6070 .bdrv_child_perm = bdrv_default_perms,
6071 .bdrv_co_create_opts = qcow2_co_create_opts,
6072 .bdrv_co_create = qcow2_co_create,
6073 .bdrv_has_zero_init = qcow2_has_zero_init,
6074 .bdrv_co_block_status = qcow2_co_block_status,
6076 .bdrv_co_preadv_part = qcow2_co_preadv_part,
6077 .bdrv_co_pwritev_part = qcow2_co_pwritev_part,
6078 .bdrv_co_flush_to_os = qcow2_co_flush_to_os,
6080 .bdrv_co_pwrite_zeroes = qcow2_co_pwrite_zeroes,
6081 .bdrv_co_pdiscard = qcow2_co_pdiscard,
6082 .bdrv_co_copy_range_from = qcow2_co_copy_range_from,
6083 .bdrv_co_copy_range_to = qcow2_co_copy_range_to,
6084 .bdrv_co_truncate = qcow2_co_truncate,
6085 .bdrv_co_pwritev_compressed_part = qcow2_co_pwritev_compressed_part,
6086 .bdrv_make_empty = qcow2_make_empty,
6088 .bdrv_snapshot_create = qcow2_snapshot_create,
6089 .bdrv_snapshot_goto = qcow2_snapshot_goto,
6090 .bdrv_snapshot_delete = qcow2_snapshot_delete,
6091 .bdrv_snapshot_list = qcow2_snapshot_list,
6092 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
6093 .bdrv_measure = qcow2_measure,
6094 .bdrv_co_get_info = qcow2_co_get_info,
6095 .bdrv_get_specific_info = qcow2_get_specific_info,
6097 .bdrv_co_save_vmstate = qcow2_co_save_vmstate,
6098 .bdrv_co_load_vmstate = qcow2_co_load_vmstate,
6100 .is_format = true,
6101 .supports_backing = true,
6102 .bdrv_change_backing_file = qcow2_change_backing_file,
6104 .bdrv_refresh_limits = qcow2_refresh_limits,
6105 .bdrv_co_invalidate_cache = qcow2_co_invalidate_cache,
6106 .bdrv_inactivate = qcow2_inactivate,
6108 .create_opts = &qcow2_create_opts,
6109 .amend_opts = &qcow2_amend_opts,
6110 .strong_runtime_opts = qcow2_strong_runtime_opts,
6111 .mutable_opts = mutable_opts,
6112 .bdrv_co_check = qcow2_co_check,
6113 .bdrv_amend_options = qcow2_amend_options,
6114 .bdrv_co_amend = qcow2_co_amend,
6116 .bdrv_detach_aio_context = qcow2_detach_aio_context,
6117 .bdrv_attach_aio_context = qcow2_attach_aio_context,
6119 .bdrv_supports_persistent_dirty_bitmap =
6120 qcow2_supports_persistent_dirty_bitmap,
6121 .bdrv_co_can_store_new_dirty_bitmap = qcow2_co_can_store_new_dirty_bitmap,
6122 .bdrv_co_remove_persistent_dirty_bitmap =
6123 qcow2_co_remove_persistent_dirty_bitmap,
6126 static void bdrv_qcow2_init(void)
6128 bdrv_register(&bdrv_qcow2);
6131 block_init(bdrv_qcow2_init);