iotests: fix loading of common.config from tests/ subdir
[qemu/ar7.git] / block / qcow2.c
blob5d94f45be9507bd5f908591d3c59b3500b682257
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 "qapi/qobject-input-visitor.h"
42 #include "qapi/qapi-visit-block-core.h"
43 #include "crypto.h"
44 #include "block/aio_task.h"
47 Differences with QCOW:
49 - Support for multiple incremental snapshots.
50 - Memory management by reference counts.
51 - Clusters which have a reference count of one have the bit
52 QCOW_OFLAG_COPIED to optimize write performance.
53 - Size of compressed clusters is stored in sectors to reduce bit usage
54 in the cluster offsets.
55 - Support for storing additional data (such as the VM state) in the
56 snapshots.
57 - If a backing store is used, the cluster size is not constrained
58 (could be backported to QCOW).
59 - L2 tables have always a size of one cluster.
63 typedef struct {
64 uint32_t magic;
65 uint32_t len;
66 } QEMU_PACKED QCowExtension;
68 #define QCOW2_EXT_MAGIC_END 0
69 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xe2792aca
70 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
71 #define QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77
72 #define QCOW2_EXT_MAGIC_BITMAPS 0x23852875
73 #define QCOW2_EXT_MAGIC_DATA_FILE 0x44415441
75 static int coroutine_fn
76 qcow2_co_preadv_compressed(BlockDriverState *bs,
77 uint64_t cluster_descriptor,
78 uint64_t offset,
79 uint64_t bytes,
80 QEMUIOVector *qiov,
81 size_t qiov_offset);
83 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
85 const QCowHeader *cow_header = (const void *)buf;
87 if (buf_size >= sizeof(QCowHeader) &&
88 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
89 be32_to_cpu(cow_header->version) >= 2)
90 return 100;
91 else
92 return 0;
96 static ssize_t qcow2_crypto_hdr_read_func(QCryptoBlock *block, size_t offset,
97 uint8_t *buf, size_t buflen,
98 void *opaque, Error **errp)
100 BlockDriverState *bs = opaque;
101 BDRVQcow2State *s = bs->opaque;
102 ssize_t ret;
104 if ((offset + buflen) > s->crypto_header.length) {
105 error_setg(errp, "Request for data outside of extension header");
106 return -1;
109 ret = bdrv_pread(bs->file,
110 s->crypto_header.offset + offset, buf, buflen);
111 if (ret < 0) {
112 error_setg_errno(errp, -ret, "Could not read encryption header");
113 return -1;
115 return ret;
119 static ssize_t qcow2_crypto_hdr_init_func(QCryptoBlock *block, size_t headerlen,
120 void *opaque, Error **errp)
122 BlockDriverState *bs = opaque;
123 BDRVQcow2State *s = bs->opaque;
124 int64_t ret;
125 int64_t clusterlen;
127 ret = qcow2_alloc_clusters(bs, headerlen);
128 if (ret < 0) {
129 error_setg_errno(errp, -ret,
130 "Cannot allocate cluster for LUKS header size %zu",
131 headerlen);
132 return -1;
135 s->crypto_header.length = headerlen;
136 s->crypto_header.offset = ret;
139 * Zero fill all space in cluster so it has predictable
140 * content, as we may not initialize some regions of the
141 * header (eg only 1 out of 8 key slots will be initialized)
143 clusterlen = size_to_clusters(s, headerlen) * s->cluster_size;
144 assert(qcow2_pre_write_overlap_check(bs, 0, ret, clusterlen, false) == 0);
145 ret = bdrv_pwrite_zeroes(bs->file,
146 ret,
147 clusterlen, 0);
148 if (ret < 0) {
149 error_setg_errno(errp, -ret, "Could not zero fill encryption header");
150 return -1;
153 return ret;
157 static ssize_t qcow2_crypto_hdr_write_func(QCryptoBlock *block, size_t offset,
158 const uint8_t *buf, size_t buflen,
159 void *opaque, Error **errp)
161 BlockDriverState *bs = opaque;
162 BDRVQcow2State *s = bs->opaque;
163 ssize_t ret;
165 if ((offset + buflen) > s->crypto_header.length) {
166 error_setg(errp, "Request for data outside of extension header");
167 return -1;
170 ret = bdrv_pwrite(bs->file,
171 s->crypto_header.offset + offset, buf, buflen);
172 if (ret < 0) {
173 error_setg_errno(errp, -ret, "Could not read encryption header");
174 return -1;
176 return ret;
179 static QDict*
180 qcow2_extract_crypto_opts(QemuOpts *opts, const char *fmt, Error **errp)
182 QDict *cryptoopts_qdict;
183 QDict *opts_qdict;
185 /* Extract "encrypt." options into a qdict */
186 opts_qdict = qemu_opts_to_qdict(opts, NULL);
187 qdict_extract_subqdict(opts_qdict, &cryptoopts_qdict, "encrypt.");
188 qobject_unref(opts_qdict);
189 qdict_put_str(cryptoopts_qdict, "format", fmt);
190 return cryptoopts_qdict;
194 * read qcow2 extension and fill bs
195 * start reading from start_offset
196 * finish reading upon magic of value 0 or when end_offset reached
197 * unknown magic is skipped (future extension this version knows nothing about)
198 * return 0 upon success, non-0 otherwise
200 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
201 uint64_t end_offset, void **p_feature_table,
202 int flags, bool *need_update_header,
203 Error **errp)
205 BDRVQcow2State *s = bs->opaque;
206 QCowExtension ext;
207 uint64_t offset;
208 int ret;
209 Qcow2BitmapHeaderExt bitmaps_ext;
211 if (need_update_header != NULL) {
212 *need_update_header = false;
215 #ifdef DEBUG_EXT
216 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
217 #endif
218 offset = start_offset;
219 while (offset < end_offset) {
221 #ifdef DEBUG_EXT
222 /* Sanity check */
223 if (offset > s->cluster_size)
224 printf("qcow2_read_extension: suspicious offset %lu\n", offset);
226 printf("attempting to read extended header in offset %lu\n", offset);
227 #endif
229 ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext));
230 if (ret < 0) {
231 error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: "
232 "pread fail from offset %" PRIu64, offset);
233 return 1;
235 ext.magic = be32_to_cpu(ext.magic);
236 ext.len = be32_to_cpu(ext.len);
237 offset += sizeof(ext);
238 #ifdef DEBUG_EXT
239 printf("ext.magic = 0x%x\n", ext.magic);
240 #endif
241 if (offset > end_offset || ext.len > end_offset - offset) {
242 error_setg(errp, "Header extension too large");
243 return -EINVAL;
246 switch (ext.magic) {
247 case QCOW2_EXT_MAGIC_END:
248 return 0;
250 case QCOW2_EXT_MAGIC_BACKING_FORMAT:
251 if (ext.len >= sizeof(bs->backing_format)) {
252 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32
253 " too large (>=%zu)", ext.len,
254 sizeof(bs->backing_format));
255 return 2;
257 ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len);
258 if (ret < 0) {
259 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: "
260 "Could not read format name");
261 return 3;
263 bs->backing_format[ext.len] = '\0';
264 s->image_backing_format = g_strdup(bs->backing_format);
265 #ifdef DEBUG_EXT
266 printf("Qcow2: Got format extension %s\n", bs->backing_format);
267 #endif
268 break;
270 case QCOW2_EXT_MAGIC_FEATURE_TABLE:
271 if (p_feature_table != NULL) {
272 void *feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
273 ret = bdrv_pread(bs->file, offset , feature_table, ext.len);
274 if (ret < 0) {
275 error_setg_errno(errp, -ret, "ERROR: ext_feature_table: "
276 "Could not read table");
277 return ret;
280 *p_feature_table = feature_table;
282 break;
284 case QCOW2_EXT_MAGIC_CRYPTO_HEADER: {
285 unsigned int cflags = 0;
286 if (s->crypt_method_header != QCOW_CRYPT_LUKS) {
287 error_setg(errp, "CRYPTO header extension only "
288 "expected with LUKS encryption method");
289 return -EINVAL;
291 if (ext.len != sizeof(Qcow2CryptoHeaderExtension)) {
292 error_setg(errp, "CRYPTO header extension size %u, "
293 "but expected size %zu", ext.len,
294 sizeof(Qcow2CryptoHeaderExtension));
295 return -EINVAL;
298 ret = bdrv_pread(bs->file, offset, &s->crypto_header, ext.len);
299 if (ret < 0) {
300 error_setg_errno(errp, -ret,
301 "Unable to read CRYPTO header extension");
302 return ret;
304 s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset);
305 s->crypto_header.length = be64_to_cpu(s->crypto_header.length);
307 if ((s->crypto_header.offset % s->cluster_size) != 0) {
308 error_setg(errp, "Encryption header offset '%" PRIu64 "' is "
309 "not a multiple of cluster size '%u'",
310 s->crypto_header.offset, s->cluster_size);
311 return -EINVAL;
314 if (flags & BDRV_O_NO_IO) {
315 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
317 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
318 qcow2_crypto_hdr_read_func,
319 bs, cflags, QCOW2_MAX_THREADS, errp);
320 if (!s->crypto) {
321 return -EINVAL;
323 } break;
325 case QCOW2_EXT_MAGIC_BITMAPS:
326 if (ext.len != sizeof(bitmaps_ext)) {
327 error_setg_errno(errp, -ret, "bitmaps_ext: "
328 "Invalid extension length");
329 return -EINVAL;
332 if (!(s->autoclear_features & QCOW2_AUTOCLEAR_BITMAPS)) {
333 if (s->qcow_version < 3) {
334 /* Let's be a bit more specific */
335 warn_report("This qcow2 v2 image contains bitmaps, but "
336 "they may have been modified by a program "
337 "without persistent bitmap support; so now "
338 "they must all be considered inconsistent");
339 } else {
340 warn_report("a program lacking bitmap support "
341 "modified this file, so all bitmaps are now "
342 "considered inconsistent");
344 error_printf("Some clusters may be leaked, "
345 "run 'qemu-img check -r' on the image "
346 "file to fix.");
347 if (need_update_header != NULL) {
348 /* Updating is needed to drop invalid bitmap extension. */
349 *need_update_header = true;
351 break;
354 ret = bdrv_pread(bs->file, offset, &bitmaps_ext, ext.len);
355 if (ret < 0) {
356 error_setg_errno(errp, -ret, "bitmaps_ext: "
357 "Could not read ext header");
358 return ret;
361 if (bitmaps_ext.reserved32 != 0) {
362 error_setg_errno(errp, -ret, "bitmaps_ext: "
363 "Reserved field is not zero");
364 return -EINVAL;
367 bitmaps_ext.nb_bitmaps = be32_to_cpu(bitmaps_ext.nb_bitmaps);
368 bitmaps_ext.bitmap_directory_size =
369 be64_to_cpu(bitmaps_ext.bitmap_directory_size);
370 bitmaps_ext.bitmap_directory_offset =
371 be64_to_cpu(bitmaps_ext.bitmap_directory_offset);
373 if (bitmaps_ext.nb_bitmaps > QCOW2_MAX_BITMAPS) {
374 error_setg(errp,
375 "bitmaps_ext: Image has %" PRIu32 " bitmaps, "
376 "exceeding the QEMU supported maximum of %d",
377 bitmaps_ext.nb_bitmaps, QCOW2_MAX_BITMAPS);
378 return -EINVAL;
381 if (bitmaps_ext.nb_bitmaps == 0) {
382 error_setg(errp, "found bitmaps extension with zero bitmaps");
383 return -EINVAL;
386 if (offset_into_cluster(s, bitmaps_ext.bitmap_directory_offset)) {
387 error_setg(errp, "bitmaps_ext: "
388 "invalid bitmap directory offset");
389 return -EINVAL;
392 if (bitmaps_ext.bitmap_directory_size >
393 QCOW2_MAX_BITMAP_DIRECTORY_SIZE) {
394 error_setg(errp, "bitmaps_ext: "
395 "bitmap directory size (%" PRIu64 ") exceeds "
396 "the maximum supported size (%d)",
397 bitmaps_ext.bitmap_directory_size,
398 QCOW2_MAX_BITMAP_DIRECTORY_SIZE);
399 return -EINVAL;
402 s->nb_bitmaps = bitmaps_ext.nb_bitmaps;
403 s->bitmap_directory_offset =
404 bitmaps_ext.bitmap_directory_offset;
405 s->bitmap_directory_size =
406 bitmaps_ext.bitmap_directory_size;
408 #ifdef DEBUG_EXT
409 printf("Qcow2: Got bitmaps extension: "
410 "offset=%" PRIu64 " nb_bitmaps=%" PRIu32 "\n",
411 s->bitmap_directory_offset, s->nb_bitmaps);
412 #endif
413 break;
415 case QCOW2_EXT_MAGIC_DATA_FILE:
417 s->image_data_file = g_malloc0(ext.len + 1);
418 ret = bdrv_pread(bs->file, offset, s->image_data_file, ext.len);
419 if (ret < 0) {
420 error_setg_errno(errp, -ret,
421 "ERROR: Could not read data file name");
422 return ret;
424 #ifdef DEBUG_EXT
425 printf("Qcow2: Got external data file %s\n", s->image_data_file);
426 #endif
427 break;
430 default:
431 /* unknown magic - save it in case we need to rewrite the header */
432 /* If you add a new feature, make sure to also update the fast
433 * path of qcow2_make_empty() to deal with it. */
435 Qcow2UnknownHeaderExtension *uext;
437 uext = g_malloc0(sizeof(*uext) + ext.len);
438 uext->magic = ext.magic;
439 uext->len = ext.len;
440 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
442 ret = bdrv_pread(bs->file, offset , uext->data, uext->len);
443 if (ret < 0) {
444 error_setg_errno(errp, -ret, "ERROR: unknown extension: "
445 "Could not read data");
446 return ret;
449 break;
452 offset += ((ext.len + 7) & ~7);
455 return 0;
458 static void cleanup_unknown_header_ext(BlockDriverState *bs)
460 BDRVQcow2State *s = bs->opaque;
461 Qcow2UnknownHeaderExtension *uext, *next;
463 QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
464 QLIST_REMOVE(uext, next);
465 g_free(uext);
469 static void report_unsupported_feature(Error **errp, Qcow2Feature *table,
470 uint64_t mask)
472 g_autoptr(GString) features = g_string_sized_new(60);
474 while (table && table->name[0] != '\0') {
475 if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
476 if (mask & (1ULL << table->bit)) {
477 if (features->len > 0) {
478 g_string_append(features, ", ");
480 g_string_append_printf(features, "%.46s", table->name);
481 mask &= ~(1ULL << table->bit);
484 table++;
487 if (mask) {
488 if (features->len > 0) {
489 g_string_append(features, ", ");
491 g_string_append_printf(features,
492 "Unknown incompatible feature: %" PRIx64, mask);
495 error_setg(errp, "Unsupported qcow2 feature(s): %s", features->str);
499 * Sets the dirty bit and flushes afterwards if necessary.
501 * The incompatible_features bit is only set if the image file header was
502 * updated successfully. Therefore it is not required to check the return
503 * value of this function.
505 int qcow2_mark_dirty(BlockDriverState *bs)
507 BDRVQcow2State *s = bs->opaque;
508 uint64_t val;
509 int ret;
511 assert(s->qcow_version >= 3);
513 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
514 return 0; /* already dirty */
517 val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY);
518 ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features),
519 &val, sizeof(val));
520 if (ret < 0) {
521 return ret;
523 ret = bdrv_flush(bs->file->bs);
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 qcow2_co_check_locked(BlockDriverState *bs,
605 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 qcow2_co_check(BlockDriverState *bs,
644 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 = aio_timer_new(context, QEMU_CLOCK_VIRTUAL,
844 SCALE_MS, cache_clean_timer_cb,
845 bs);
846 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
847 (int64_t) s->cache_clean_interval * 1000);
851 static void cache_clean_timer_del(BlockDriverState *bs)
853 BDRVQcow2State *s = bs->opaque;
854 if (s->cache_clean_timer) {
855 timer_free(s->cache_clean_timer);
856 s->cache_clean_timer = NULL;
860 static void qcow2_detach_aio_context(BlockDriverState *bs)
862 cache_clean_timer_del(bs);
865 static void qcow2_attach_aio_context(BlockDriverState *bs,
866 AioContext *new_context)
868 cache_clean_timer_init(bs, new_context);
871 static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts,
872 uint64_t *l2_cache_size,
873 uint64_t *l2_cache_entry_size,
874 uint64_t *refcount_cache_size, Error **errp)
876 BDRVQcow2State *s = bs->opaque;
877 uint64_t combined_cache_size, l2_cache_max_setting;
878 bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set;
879 bool l2_cache_entry_size_set;
880 int min_refcount_cache = MIN_REFCOUNT_CACHE_SIZE * s->cluster_size;
881 uint64_t virtual_disk_size = bs->total_sectors * BDRV_SECTOR_SIZE;
882 uint64_t max_l2_entries = DIV_ROUND_UP(virtual_disk_size, s->cluster_size);
883 /* An L2 table is always one cluster in size so the max cache size
884 * should be a multiple of the cluster size. */
885 uint64_t max_l2_cache = ROUND_UP(max_l2_entries * l2_entry_size(s),
886 s->cluster_size);
888 combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE);
889 l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE);
890 refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
891 l2_cache_entry_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE);
893 combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0);
894 l2_cache_max_setting = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE,
895 DEFAULT_L2_CACHE_MAX_SIZE);
896 *refcount_cache_size = qemu_opt_get_size(opts,
897 QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0);
899 *l2_cache_entry_size = qemu_opt_get_size(
900 opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE, s->cluster_size);
902 *l2_cache_size = MIN(max_l2_cache, l2_cache_max_setting);
904 if (combined_cache_size_set) {
905 if (l2_cache_size_set && refcount_cache_size_set) {
906 error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE
907 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set "
908 "at the same time");
909 return;
910 } else if (l2_cache_size_set &&
911 (l2_cache_max_setting > combined_cache_size)) {
912 error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed "
913 QCOW2_OPT_CACHE_SIZE);
914 return;
915 } else if (*refcount_cache_size > combined_cache_size) {
916 error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed "
917 QCOW2_OPT_CACHE_SIZE);
918 return;
921 if (l2_cache_size_set) {
922 *refcount_cache_size = combined_cache_size - *l2_cache_size;
923 } else if (refcount_cache_size_set) {
924 *l2_cache_size = combined_cache_size - *refcount_cache_size;
925 } else {
926 /* Assign as much memory as possible to the L2 cache, and
927 * use the remainder for the refcount cache */
928 if (combined_cache_size >= max_l2_cache + min_refcount_cache) {
929 *l2_cache_size = max_l2_cache;
930 *refcount_cache_size = combined_cache_size - *l2_cache_size;
931 } else {
932 *refcount_cache_size =
933 MIN(combined_cache_size, min_refcount_cache);
934 *l2_cache_size = combined_cache_size - *refcount_cache_size;
940 * If the L2 cache is not enough to cover the whole disk then
941 * default to 4KB entries. Smaller entries reduce the cost of
942 * loads and evictions and increase I/O performance.
944 if (*l2_cache_size < max_l2_cache && !l2_cache_entry_size_set) {
945 *l2_cache_entry_size = MIN(s->cluster_size, 4096);
948 /* l2_cache_size and refcount_cache_size are ensured to have at least
949 * their minimum values in qcow2_update_options_prepare() */
951 if (*l2_cache_entry_size < (1 << MIN_CLUSTER_BITS) ||
952 *l2_cache_entry_size > s->cluster_size ||
953 !is_power_of_2(*l2_cache_entry_size)) {
954 error_setg(errp, "L2 cache entry size must be a power of two "
955 "between %d and the cluster size (%d)",
956 1 << MIN_CLUSTER_BITS, s->cluster_size);
957 return;
961 typedef struct Qcow2ReopenState {
962 Qcow2Cache *l2_table_cache;
963 Qcow2Cache *refcount_block_cache;
964 int l2_slice_size; /* Number of entries in a slice of the L2 table */
965 bool use_lazy_refcounts;
966 int overlap_check;
967 bool discard_passthrough[QCOW2_DISCARD_MAX];
968 uint64_t cache_clean_interval;
969 QCryptoBlockOpenOptions *crypto_opts; /* Disk encryption runtime options */
970 } Qcow2ReopenState;
972 static int qcow2_update_options_prepare(BlockDriverState *bs,
973 Qcow2ReopenState *r,
974 QDict *options, int flags,
975 Error **errp)
977 BDRVQcow2State *s = bs->opaque;
978 QemuOpts *opts = NULL;
979 const char *opt_overlap_check, *opt_overlap_check_template;
980 int overlap_check_template = 0;
981 uint64_t l2_cache_size, l2_cache_entry_size, refcount_cache_size;
982 int i;
983 const char *encryptfmt;
984 QDict *encryptopts = NULL;
985 Error *local_err = NULL;
986 int ret;
988 qdict_extract_subqdict(options, &encryptopts, "encrypt.");
989 encryptfmt = qdict_get_try_str(encryptopts, "format");
991 opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort);
992 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
993 ret = -EINVAL;
994 goto fail;
997 /* get L2 table/refcount block cache size from command line options */
998 read_cache_sizes(bs, opts, &l2_cache_size, &l2_cache_entry_size,
999 &refcount_cache_size, &local_err);
1000 if (local_err) {
1001 error_propagate(errp, local_err);
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 break;
1164 case QCOW_CRYPT_LUKS:
1165 if (encryptfmt && !g_str_equal(encryptfmt, "luks")) {
1166 error_setg(errp,
1167 "Header reported 'luks' encryption format but "
1168 "options specify '%s'", encryptfmt);
1169 ret = -EINVAL;
1170 goto fail;
1172 qdict_put_str(encryptopts, "format", "luks");
1173 r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
1174 break;
1176 default:
1177 error_setg(errp, "Unsupported encryption method %d",
1178 s->crypt_method_header);
1179 break;
1181 if (s->crypt_method_header != QCOW_CRYPT_NONE && !r->crypto_opts) {
1182 ret = -EINVAL;
1183 goto fail;
1186 ret = 0;
1187 fail:
1188 qobject_unref(encryptopts);
1189 qemu_opts_del(opts);
1190 opts = NULL;
1191 return ret;
1194 static void qcow2_update_options_commit(BlockDriverState *bs,
1195 Qcow2ReopenState *r)
1197 BDRVQcow2State *s = bs->opaque;
1198 int i;
1200 if (s->l2_table_cache) {
1201 qcow2_cache_destroy(s->l2_table_cache);
1203 if (s->refcount_block_cache) {
1204 qcow2_cache_destroy(s->refcount_block_cache);
1206 s->l2_table_cache = r->l2_table_cache;
1207 s->refcount_block_cache = r->refcount_block_cache;
1208 s->l2_slice_size = r->l2_slice_size;
1210 s->overlap_check = r->overlap_check;
1211 s->use_lazy_refcounts = r->use_lazy_refcounts;
1213 for (i = 0; i < QCOW2_DISCARD_MAX; i++) {
1214 s->discard_passthrough[i] = r->discard_passthrough[i];
1217 if (s->cache_clean_interval != r->cache_clean_interval) {
1218 cache_clean_timer_del(bs);
1219 s->cache_clean_interval = r->cache_clean_interval;
1220 cache_clean_timer_init(bs, bdrv_get_aio_context(bs));
1223 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1224 s->crypto_opts = r->crypto_opts;
1227 static void qcow2_update_options_abort(BlockDriverState *bs,
1228 Qcow2ReopenState *r)
1230 if (r->l2_table_cache) {
1231 qcow2_cache_destroy(r->l2_table_cache);
1233 if (r->refcount_block_cache) {
1234 qcow2_cache_destroy(r->refcount_block_cache);
1236 qapi_free_QCryptoBlockOpenOptions(r->crypto_opts);
1239 static int qcow2_update_options(BlockDriverState *bs, QDict *options,
1240 int flags, Error **errp)
1242 Qcow2ReopenState r = {};
1243 int ret;
1245 ret = qcow2_update_options_prepare(bs, &r, options, flags, errp);
1246 if (ret >= 0) {
1247 qcow2_update_options_commit(bs, &r);
1248 } else {
1249 qcow2_update_options_abort(bs, &r);
1252 return ret;
1255 static int validate_compression_type(BDRVQcow2State *s, Error **errp)
1257 switch (s->compression_type) {
1258 case QCOW2_COMPRESSION_TYPE_ZLIB:
1259 #ifdef CONFIG_ZSTD
1260 case QCOW2_COMPRESSION_TYPE_ZSTD:
1261 #endif
1262 break;
1264 default:
1265 error_setg(errp, "qcow2: unknown compression type: %u",
1266 s->compression_type);
1267 return -ENOTSUP;
1271 * if the compression type differs from QCOW2_COMPRESSION_TYPE_ZLIB
1272 * the incompatible feature flag must be set
1274 if (s->compression_type == QCOW2_COMPRESSION_TYPE_ZLIB) {
1275 if (s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION) {
1276 error_setg(errp, "qcow2: Compression type incompatible feature "
1277 "bit must not be set");
1278 return -EINVAL;
1280 } else {
1281 if (!(s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION)) {
1282 error_setg(errp, "qcow2: Compression type incompatible feature "
1283 "bit must be set");
1284 return -EINVAL;
1288 return 0;
1291 /* Called with s->lock held. */
1292 static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options,
1293 int flags, Error **errp)
1295 BDRVQcow2State *s = bs->opaque;
1296 unsigned int len, i;
1297 int ret = 0;
1298 QCowHeader header;
1299 Error *local_err = NULL;
1300 uint64_t ext_end;
1301 uint64_t l1_vm_state_index;
1302 bool update_header = false;
1304 ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
1305 if (ret < 0) {
1306 error_setg_errno(errp, -ret, "Could not read qcow2 header");
1307 goto fail;
1309 header.magic = be32_to_cpu(header.magic);
1310 header.version = be32_to_cpu(header.version);
1311 header.backing_file_offset = be64_to_cpu(header.backing_file_offset);
1312 header.backing_file_size = be32_to_cpu(header.backing_file_size);
1313 header.size = be64_to_cpu(header.size);
1314 header.cluster_bits = be32_to_cpu(header.cluster_bits);
1315 header.crypt_method = be32_to_cpu(header.crypt_method);
1316 header.l1_table_offset = be64_to_cpu(header.l1_table_offset);
1317 header.l1_size = be32_to_cpu(header.l1_size);
1318 header.refcount_table_offset = be64_to_cpu(header.refcount_table_offset);
1319 header.refcount_table_clusters =
1320 be32_to_cpu(header.refcount_table_clusters);
1321 header.snapshots_offset = be64_to_cpu(header.snapshots_offset);
1322 header.nb_snapshots = be32_to_cpu(header.nb_snapshots);
1324 if (header.magic != QCOW_MAGIC) {
1325 error_setg(errp, "Image is not in qcow2 format");
1326 ret = -EINVAL;
1327 goto fail;
1329 if (header.version < 2 || header.version > 3) {
1330 error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version);
1331 ret = -ENOTSUP;
1332 goto fail;
1335 s->qcow_version = header.version;
1337 /* Initialise cluster size */
1338 if (header.cluster_bits < MIN_CLUSTER_BITS ||
1339 header.cluster_bits > MAX_CLUSTER_BITS) {
1340 error_setg(errp, "Unsupported cluster size: 2^%" PRIu32,
1341 header.cluster_bits);
1342 ret = -EINVAL;
1343 goto fail;
1346 s->cluster_bits = header.cluster_bits;
1347 s->cluster_size = 1 << s->cluster_bits;
1349 /* Initialise version 3 header fields */
1350 if (header.version == 2) {
1351 header.incompatible_features = 0;
1352 header.compatible_features = 0;
1353 header.autoclear_features = 0;
1354 header.refcount_order = 4;
1355 header.header_length = 72;
1356 } else {
1357 header.incompatible_features =
1358 be64_to_cpu(header.incompatible_features);
1359 header.compatible_features = be64_to_cpu(header.compatible_features);
1360 header.autoclear_features = be64_to_cpu(header.autoclear_features);
1361 header.refcount_order = be32_to_cpu(header.refcount_order);
1362 header.header_length = be32_to_cpu(header.header_length);
1364 if (header.header_length < 104) {
1365 error_setg(errp, "qcow2 header too short");
1366 ret = -EINVAL;
1367 goto fail;
1371 if (header.header_length > s->cluster_size) {
1372 error_setg(errp, "qcow2 header exceeds cluster size");
1373 ret = -EINVAL;
1374 goto fail;
1377 if (header.header_length > sizeof(header)) {
1378 s->unknown_header_fields_size = header.header_length - sizeof(header);
1379 s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
1380 ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields,
1381 s->unknown_header_fields_size);
1382 if (ret < 0) {
1383 error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
1384 "fields");
1385 goto fail;
1389 if (header.backing_file_offset > s->cluster_size) {
1390 error_setg(errp, "Invalid backing file offset");
1391 ret = -EINVAL;
1392 goto fail;
1395 if (header.backing_file_offset) {
1396 ext_end = header.backing_file_offset;
1397 } else {
1398 ext_end = 1 << header.cluster_bits;
1401 /* Handle feature bits */
1402 s->incompatible_features = header.incompatible_features;
1403 s->compatible_features = header.compatible_features;
1404 s->autoclear_features = header.autoclear_features;
1407 * Handle compression type
1408 * Older qcow2 images don't contain the compression type header.
1409 * Distinguish them by the header length and use
1410 * the only valid (default) compression type in that case
1412 if (header.header_length > offsetof(QCowHeader, compression_type)) {
1413 s->compression_type = header.compression_type;
1414 } else {
1415 s->compression_type = QCOW2_COMPRESSION_TYPE_ZLIB;
1418 ret = validate_compression_type(s, errp);
1419 if (ret) {
1420 goto fail;
1423 if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
1424 void *feature_table = NULL;
1425 qcow2_read_extensions(bs, header.header_length, ext_end,
1426 &feature_table, flags, NULL, NULL);
1427 report_unsupported_feature(errp, feature_table,
1428 s->incompatible_features &
1429 ~QCOW2_INCOMPAT_MASK);
1430 ret = -ENOTSUP;
1431 g_free(feature_table);
1432 goto fail;
1435 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
1436 /* Corrupt images may not be written to unless they are being repaired
1438 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
1439 error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
1440 "read/write");
1441 ret = -EACCES;
1442 goto fail;
1446 s->subclusters_per_cluster =
1447 has_subclusters(s) ? QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER : 1;
1448 s->subcluster_size = s->cluster_size / s->subclusters_per_cluster;
1449 s->subcluster_bits = ctz32(s->subcluster_size);
1451 if (s->subcluster_size < (1 << MIN_CLUSTER_BITS)) {
1452 error_setg(errp, "Unsupported subcluster size: %d", s->subcluster_size);
1453 ret = -EINVAL;
1454 goto fail;
1457 /* Check support for various header values */
1458 if (header.refcount_order > 6) {
1459 error_setg(errp, "Reference count entry width too large; may not "
1460 "exceed 64 bits");
1461 ret = -EINVAL;
1462 goto fail;
1464 s->refcount_order = header.refcount_order;
1465 s->refcount_bits = 1 << s->refcount_order;
1466 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
1467 s->refcount_max += s->refcount_max - 1;
1469 s->crypt_method_header = header.crypt_method;
1470 if (s->crypt_method_header) {
1471 if (bdrv_uses_whitelist() &&
1472 s->crypt_method_header == QCOW_CRYPT_AES) {
1473 error_setg(errp,
1474 "Use of AES-CBC encrypted qcow2 images is no longer "
1475 "supported in system emulators");
1476 error_append_hint(errp,
1477 "You can use 'qemu-img convert' to convert your "
1478 "image to an alternative supported format, such "
1479 "as unencrypted qcow2, or raw with the LUKS "
1480 "format instead.\n");
1481 ret = -ENOSYS;
1482 goto fail;
1485 if (s->crypt_method_header == QCOW_CRYPT_AES) {
1486 s->crypt_physical_offset = false;
1487 } else {
1488 /* Assuming LUKS and any future crypt methods we
1489 * add will all use physical offsets, due to the
1490 * fact that the alternative is insecure... */
1491 s->crypt_physical_offset = true;
1494 bs->encrypted = true;
1497 s->l2_bits = s->cluster_bits - ctz32(l2_entry_size(s));
1498 s->l2_size = 1 << s->l2_bits;
1499 /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1500 s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
1501 s->refcount_block_size = 1 << s->refcount_block_bits;
1502 bs->total_sectors = header.size / BDRV_SECTOR_SIZE;
1503 s->csize_shift = (62 - (s->cluster_bits - 8));
1504 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
1505 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
1507 s->refcount_table_offset = header.refcount_table_offset;
1508 s->refcount_table_size =
1509 header.refcount_table_clusters << (s->cluster_bits - 3);
1511 if (header.refcount_table_clusters == 0 && !(flags & BDRV_O_CHECK)) {
1512 error_setg(errp, "Image does not contain a reference count table");
1513 ret = -EINVAL;
1514 goto fail;
1517 ret = qcow2_validate_table(bs, s->refcount_table_offset,
1518 header.refcount_table_clusters,
1519 s->cluster_size, QCOW_MAX_REFTABLE_SIZE,
1520 "Reference count table", errp);
1521 if (ret < 0) {
1522 goto fail;
1525 if (!(flags & BDRV_O_CHECK)) {
1527 * The total size in bytes of the snapshot table is checked in
1528 * qcow2_read_snapshots() because the size of each snapshot is
1529 * variable and we don't know it yet.
1530 * Here we only check the offset and number of snapshots.
1532 ret = qcow2_validate_table(bs, header.snapshots_offset,
1533 header.nb_snapshots,
1534 sizeof(QCowSnapshotHeader),
1535 sizeof(QCowSnapshotHeader) *
1536 QCOW_MAX_SNAPSHOTS,
1537 "Snapshot table", errp);
1538 if (ret < 0) {
1539 goto fail;
1543 /* read the level 1 table */
1544 ret = qcow2_validate_table(bs, header.l1_table_offset,
1545 header.l1_size, L1E_SIZE,
1546 QCOW_MAX_L1_SIZE, "Active L1 table", errp);
1547 if (ret < 0) {
1548 goto fail;
1550 s->l1_size = header.l1_size;
1551 s->l1_table_offset = header.l1_table_offset;
1553 l1_vm_state_index = size_to_l1(s, header.size);
1554 if (l1_vm_state_index > INT_MAX) {
1555 error_setg(errp, "Image is too big");
1556 ret = -EFBIG;
1557 goto fail;
1559 s->l1_vm_state_index = l1_vm_state_index;
1561 /* the L1 table must contain at least enough entries to put
1562 header.size bytes */
1563 if (s->l1_size < s->l1_vm_state_index) {
1564 error_setg(errp, "L1 table is too small");
1565 ret = -EINVAL;
1566 goto fail;
1569 if (s->l1_size > 0) {
1570 s->l1_table = qemu_try_blockalign(bs->file->bs, s->l1_size * L1E_SIZE);
1571 if (s->l1_table == NULL) {
1572 error_setg(errp, "Could not allocate L1 table");
1573 ret = -ENOMEM;
1574 goto fail;
1576 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
1577 s->l1_size * L1E_SIZE);
1578 if (ret < 0) {
1579 error_setg_errno(errp, -ret, "Could not read L1 table");
1580 goto fail;
1582 for(i = 0;i < s->l1_size; i++) {
1583 s->l1_table[i] = be64_to_cpu(s->l1_table[i]);
1587 /* Parse driver-specific options */
1588 ret = qcow2_update_options(bs, options, flags, errp);
1589 if (ret < 0) {
1590 goto fail;
1593 s->flags = flags;
1595 ret = qcow2_refcount_init(bs);
1596 if (ret != 0) {
1597 error_setg_errno(errp, -ret, "Could not initialize refcount handling");
1598 goto fail;
1601 QLIST_INIT(&s->cluster_allocs);
1602 QTAILQ_INIT(&s->discards);
1604 /* read qcow2 extensions */
1605 if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
1606 flags, &update_header, errp)) {
1607 ret = -EINVAL;
1608 goto fail;
1611 /* Open external data file */
1612 s->data_file = bdrv_open_child(NULL, options, "data-file", bs,
1613 &child_of_bds, BDRV_CHILD_DATA,
1614 true, &local_err);
1615 if (local_err) {
1616 error_propagate(errp, local_err);
1617 ret = -EINVAL;
1618 goto fail;
1621 if (s->incompatible_features & QCOW2_INCOMPAT_DATA_FILE) {
1622 if (!s->data_file && s->image_data_file) {
1623 s->data_file = bdrv_open_child(s->image_data_file, options,
1624 "data-file", bs, &child_of_bds,
1625 BDRV_CHILD_DATA, false, errp);
1626 if (!s->data_file) {
1627 ret = -EINVAL;
1628 goto fail;
1631 if (!s->data_file) {
1632 error_setg(errp, "'data-file' is required for this image");
1633 ret = -EINVAL;
1634 goto fail;
1637 /* No data here */
1638 bs->file->role &= ~BDRV_CHILD_DATA;
1640 /* Must succeed because we have given up permissions if anything */
1641 bdrv_child_refresh_perms(bs, bs->file, &error_abort);
1642 } else {
1643 if (s->data_file) {
1644 error_setg(errp, "'data-file' can only be set for images with an "
1645 "external data file");
1646 ret = -EINVAL;
1647 goto fail;
1650 s->data_file = bs->file;
1652 if (data_file_is_raw(bs)) {
1653 error_setg(errp, "data-file-raw requires a data file");
1654 ret = -EINVAL;
1655 goto fail;
1659 /* qcow2_read_extension may have set up the crypto context
1660 * if the crypt method needs a header region, some methods
1661 * don't need header extensions, so must check here
1663 if (s->crypt_method_header && !s->crypto) {
1664 if (s->crypt_method_header == QCOW_CRYPT_AES) {
1665 unsigned int cflags = 0;
1666 if (flags & BDRV_O_NO_IO) {
1667 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
1669 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
1670 NULL, NULL, cflags,
1671 QCOW2_MAX_THREADS, errp);
1672 if (!s->crypto) {
1673 ret = -EINVAL;
1674 goto fail;
1676 } else if (!(flags & BDRV_O_NO_IO)) {
1677 error_setg(errp, "Missing CRYPTO header for crypt method %d",
1678 s->crypt_method_header);
1679 ret = -EINVAL;
1680 goto fail;
1684 /* read the backing file name */
1685 if (header.backing_file_offset != 0) {
1686 len = header.backing_file_size;
1687 if (len > MIN(1023, s->cluster_size - header.backing_file_offset) ||
1688 len >= sizeof(bs->backing_file)) {
1689 error_setg(errp, "Backing file name too long");
1690 ret = -EINVAL;
1691 goto fail;
1693 ret = bdrv_pread(bs->file, header.backing_file_offset,
1694 bs->auto_backing_file, len);
1695 if (ret < 0) {
1696 error_setg_errno(errp, -ret, "Could not read backing file name");
1697 goto fail;
1699 bs->auto_backing_file[len] = '\0';
1700 pstrcpy(bs->backing_file, sizeof(bs->backing_file),
1701 bs->auto_backing_file);
1702 s->image_backing_file = g_strdup(bs->auto_backing_file);
1706 * Internal snapshots; skip reading them in check mode, because
1707 * we do not need them then, and we do not want to abort because
1708 * of a broken table.
1710 if (!(flags & BDRV_O_CHECK)) {
1711 s->snapshots_offset = header.snapshots_offset;
1712 s->nb_snapshots = header.nb_snapshots;
1714 ret = qcow2_read_snapshots(bs, errp);
1715 if (ret < 0) {
1716 goto fail;
1720 /* Clear unknown autoclear feature bits */
1721 update_header |= s->autoclear_features & ~QCOW2_AUTOCLEAR_MASK;
1722 update_header =
1723 update_header && !bs->read_only && !(flags & BDRV_O_INACTIVE);
1724 if (update_header) {
1725 s->autoclear_features &= QCOW2_AUTOCLEAR_MASK;
1728 /* == Handle persistent dirty bitmaps ==
1730 * We want load dirty bitmaps in three cases:
1732 * 1. Normal open of the disk in active mode, not related to invalidation
1733 * after migration.
1735 * 2. Invalidation of the target vm after pre-copy phase of migration, if
1736 * bitmaps are _not_ migrating through migration channel, i.e.
1737 * 'dirty-bitmaps' capability is disabled.
1739 * 3. Invalidation of source vm after failed or canceled migration.
1740 * This is a very interesting case. There are two possible types of
1741 * bitmaps:
1743 * A. Stored on inactivation and removed. They should be loaded from the
1744 * image.
1746 * B. Not stored: not-persistent bitmaps and bitmaps, migrated through
1747 * the migration channel (with dirty-bitmaps capability).
1749 * On the other hand, there are two possible sub-cases:
1751 * 3.1 disk was changed by somebody else while were inactive. In this
1752 * case all in-RAM dirty bitmaps (both persistent and not) are
1753 * definitely invalid. And we don't have any method to determine
1754 * this.
1756 * Simple and safe thing is to just drop all the bitmaps of type B on
1757 * inactivation. But in this case we lose bitmaps in valid 4.2 case.
1759 * On the other hand, resuming source vm, if disk was already changed
1760 * is a bad thing anyway: not only bitmaps, the whole vm state is
1761 * out of sync with disk.
1763 * This means, that user or management tool, who for some reason
1764 * decided to resume source vm, after disk was already changed by
1765 * target vm, should at least drop all dirty bitmaps by hand.
1767 * So, we can ignore this case for now, but TODO: "generation"
1768 * extension for qcow2, to determine, that image was changed after
1769 * last inactivation. And if it is changed, we will drop (or at least
1770 * mark as 'invalid' all the bitmaps of type B, both persistent
1771 * and not).
1773 * 3.2 disk was _not_ changed while were inactive. Bitmaps may be saved
1774 * to disk ('dirty-bitmaps' capability disabled), or not saved
1775 * ('dirty-bitmaps' capability enabled), but we don't need to care
1776 * of: let's load bitmaps as always: stored bitmaps will be loaded,
1777 * and not stored has flag IN_USE=1 in the image and will be skipped
1778 * on loading.
1780 * One remaining possible case when we don't want load bitmaps:
1782 * 4. Open disk in inactive mode in target vm (bitmaps are migrating or
1783 * will be loaded on invalidation, no needs try loading them before)
1786 if (!(bdrv_get_flags(bs) & BDRV_O_INACTIVE)) {
1787 /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */
1788 bool header_updated = qcow2_load_dirty_bitmaps(bs, &local_err);
1789 if (local_err != NULL) {
1790 error_propagate(errp, local_err);
1791 ret = -EINVAL;
1792 goto fail;
1795 update_header = update_header && !header_updated;
1798 if (update_header) {
1799 ret = qcow2_update_header(bs);
1800 if (ret < 0) {
1801 error_setg_errno(errp, -ret, "Could not update qcow2 header");
1802 goto fail;
1806 bs->supported_zero_flags = header.version >= 3 ?
1807 BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK : 0;
1808 bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE;
1810 /* Repair image if dirty */
1811 if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only &&
1812 (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
1813 BdrvCheckResult result = {0};
1815 ret = qcow2_co_check_locked(bs, &result,
1816 BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
1817 if (ret < 0 || result.check_errors) {
1818 if (ret >= 0) {
1819 ret = -EIO;
1821 error_setg_errno(errp, -ret, "Could not repair dirty image");
1822 goto fail;
1826 #ifdef DEBUG_ALLOC
1828 BdrvCheckResult result = {0};
1829 qcow2_check_refcounts(bs, &result, 0);
1831 #endif
1833 qemu_co_queue_init(&s->thread_task_queue);
1835 return ret;
1837 fail:
1838 g_free(s->image_data_file);
1839 if (has_data_file(bs)) {
1840 bdrv_unref_child(bs, s->data_file);
1841 s->data_file = NULL;
1843 g_free(s->unknown_header_fields);
1844 cleanup_unknown_header_ext(bs);
1845 qcow2_free_snapshots(bs);
1846 qcow2_refcount_close(bs);
1847 qemu_vfree(s->l1_table);
1848 /* else pre-write overlap checks in cache_destroy may crash */
1849 s->l1_table = NULL;
1850 cache_clean_timer_del(bs);
1851 if (s->l2_table_cache) {
1852 qcow2_cache_destroy(s->l2_table_cache);
1854 if (s->refcount_block_cache) {
1855 qcow2_cache_destroy(s->refcount_block_cache);
1857 qcrypto_block_free(s->crypto);
1858 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1859 return ret;
1862 typedef struct QCow2OpenCo {
1863 BlockDriverState *bs;
1864 QDict *options;
1865 int flags;
1866 Error **errp;
1867 int ret;
1868 } QCow2OpenCo;
1870 static void coroutine_fn qcow2_open_entry(void *opaque)
1872 QCow2OpenCo *qoc = opaque;
1873 BDRVQcow2State *s = qoc->bs->opaque;
1875 qemu_co_mutex_lock(&s->lock);
1876 qoc->ret = qcow2_do_open(qoc->bs, qoc->options, qoc->flags, qoc->errp);
1877 qemu_co_mutex_unlock(&s->lock);
1880 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
1881 Error **errp)
1883 BDRVQcow2State *s = bs->opaque;
1884 QCow2OpenCo qoc = {
1885 .bs = bs,
1886 .options = options,
1887 .flags = flags,
1888 .errp = errp,
1889 .ret = -EINPROGRESS
1892 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds,
1893 BDRV_CHILD_IMAGE, false, errp);
1894 if (!bs->file) {
1895 return -EINVAL;
1898 /* Initialise locks */
1899 qemu_co_mutex_init(&s->lock);
1901 if (qemu_in_coroutine()) {
1902 /* From bdrv_co_create. */
1903 qcow2_open_entry(&qoc);
1904 } else {
1905 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
1906 qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry, &qoc));
1907 BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS);
1909 return qoc.ret;
1912 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1914 BDRVQcow2State *s = bs->opaque;
1916 if (bs->encrypted) {
1917 /* Encryption works on a sector granularity */
1918 bs->bl.request_alignment = qcrypto_block_get_sector_size(s->crypto);
1920 bs->bl.pwrite_zeroes_alignment = s->subcluster_size;
1921 bs->bl.pdiscard_alignment = s->cluster_size;
1924 static int qcow2_reopen_prepare(BDRVReopenState *state,
1925 BlockReopenQueue *queue, Error **errp)
1927 Qcow2ReopenState *r;
1928 int ret;
1930 r = g_new0(Qcow2ReopenState, 1);
1931 state->opaque = r;
1933 ret = qcow2_update_options_prepare(state->bs, r, state->options,
1934 state->flags, errp);
1935 if (ret < 0) {
1936 goto fail;
1939 /* We need to write out any unwritten data if we reopen read-only. */
1940 if ((state->flags & BDRV_O_RDWR) == 0) {
1941 ret = qcow2_reopen_bitmaps_ro(state->bs, errp);
1942 if (ret < 0) {
1943 goto fail;
1946 ret = bdrv_flush(state->bs);
1947 if (ret < 0) {
1948 goto fail;
1951 ret = qcow2_mark_clean(state->bs);
1952 if (ret < 0) {
1953 goto fail;
1957 return 0;
1959 fail:
1960 qcow2_update_options_abort(state->bs, r);
1961 g_free(r);
1962 return ret;
1965 static void qcow2_reopen_commit(BDRVReopenState *state)
1967 qcow2_update_options_commit(state->bs, state->opaque);
1968 g_free(state->opaque);
1971 static void qcow2_reopen_commit_post(BDRVReopenState *state)
1973 if (state->flags & BDRV_O_RDWR) {
1974 Error *local_err = NULL;
1976 if (qcow2_reopen_bitmaps_rw(state->bs, &local_err) < 0) {
1978 * This is not fatal, bitmaps just left read-only, so all following
1979 * writes will fail. User can remove read-only bitmaps to unblock
1980 * writes or retry reopen.
1982 error_reportf_err(local_err,
1983 "%s: Failed to make dirty bitmaps writable: ",
1984 bdrv_get_node_name(state->bs));
1989 static void qcow2_reopen_abort(BDRVReopenState *state)
1991 qcow2_update_options_abort(state->bs, state->opaque);
1992 g_free(state->opaque);
1995 static void qcow2_join_options(QDict *options, QDict *old_options)
1997 bool has_new_overlap_template =
1998 qdict_haskey(options, QCOW2_OPT_OVERLAP) ||
1999 qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE);
2000 bool has_new_total_cache_size =
2001 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE);
2002 bool has_all_cache_options;
2004 /* New overlap template overrides all old overlap options */
2005 if (has_new_overlap_template) {
2006 qdict_del(old_options, QCOW2_OPT_OVERLAP);
2007 qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE);
2008 qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER);
2009 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1);
2010 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2);
2011 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE);
2012 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK);
2013 qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE);
2014 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1);
2015 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2);
2018 /* New total cache size overrides all old options */
2019 if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) {
2020 qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE);
2021 qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
2024 qdict_join(options, old_options, false);
2027 * If after merging all cache size options are set, an old total size is
2028 * overwritten. Do keep all options, however, if all three are new. The
2029 * resulting error message is what we want to happen.
2031 has_all_cache_options =
2032 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) ||
2033 qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) ||
2034 qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
2036 if (has_all_cache_options && !has_new_total_cache_size) {
2037 qdict_del(options, QCOW2_OPT_CACHE_SIZE);
2041 static int coroutine_fn qcow2_co_block_status(BlockDriverState *bs,
2042 bool want_zero,
2043 int64_t offset, int64_t count,
2044 int64_t *pnum, int64_t *map,
2045 BlockDriverState **file)
2047 BDRVQcow2State *s = bs->opaque;
2048 uint64_t host_offset;
2049 unsigned int bytes;
2050 QCow2SubclusterType type;
2051 int ret, status = 0;
2053 qemu_co_mutex_lock(&s->lock);
2055 if (!s->metadata_preallocation_checked) {
2056 ret = qcow2_detect_metadata_preallocation(bs);
2057 s->metadata_preallocation = (ret == 1);
2058 s->metadata_preallocation_checked = true;
2061 bytes = MIN(INT_MAX, count);
2062 ret = qcow2_get_host_offset(bs, offset, &bytes, &host_offset, &type);
2063 qemu_co_mutex_unlock(&s->lock);
2064 if (ret < 0) {
2065 return ret;
2068 *pnum = bytes;
2070 if ((type == QCOW2_SUBCLUSTER_NORMAL ||
2071 type == QCOW2_SUBCLUSTER_ZERO_ALLOC ||
2072 type == QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC) && !s->crypto) {
2073 *map = host_offset;
2074 *file = s->data_file->bs;
2075 status |= BDRV_BLOCK_OFFSET_VALID;
2077 if (type == QCOW2_SUBCLUSTER_ZERO_PLAIN ||
2078 type == QCOW2_SUBCLUSTER_ZERO_ALLOC) {
2079 status |= BDRV_BLOCK_ZERO;
2080 } else if (type != QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN &&
2081 type != QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC) {
2082 status |= BDRV_BLOCK_DATA;
2084 if (s->metadata_preallocation && (status & BDRV_BLOCK_DATA) &&
2085 (status & BDRV_BLOCK_OFFSET_VALID))
2087 status |= BDRV_BLOCK_RECURSE;
2089 return status;
2092 static coroutine_fn int qcow2_handle_l2meta(BlockDriverState *bs,
2093 QCowL2Meta **pl2meta,
2094 bool link_l2)
2096 int ret = 0;
2097 QCowL2Meta *l2meta = *pl2meta;
2099 while (l2meta != NULL) {
2100 QCowL2Meta *next;
2102 if (link_l2) {
2103 ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
2104 if (ret) {
2105 goto out;
2107 } else {
2108 qcow2_alloc_cluster_abort(bs, l2meta);
2111 /* Take the request off the list of running requests */
2112 QLIST_REMOVE(l2meta, next_in_flight);
2114 qemu_co_queue_restart_all(&l2meta->dependent_requests);
2116 next = l2meta->next;
2117 g_free(l2meta);
2118 l2meta = next;
2120 out:
2121 *pl2meta = l2meta;
2122 return ret;
2125 static coroutine_fn int
2126 qcow2_co_preadv_encrypted(BlockDriverState *bs,
2127 uint64_t host_offset,
2128 uint64_t offset,
2129 uint64_t bytes,
2130 QEMUIOVector *qiov,
2131 uint64_t qiov_offset)
2133 int ret;
2134 BDRVQcow2State *s = bs->opaque;
2135 uint8_t *buf;
2137 assert(bs->encrypted && s->crypto);
2138 assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2141 * For encrypted images, read everything into a temporary
2142 * contiguous buffer on which the AES functions can work.
2143 * Also, decryption in a separate buffer is better as it
2144 * prevents the guest from learning information about the
2145 * encrypted nature of the virtual disk.
2148 buf = qemu_try_blockalign(s->data_file->bs, bytes);
2149 if (buf == NULL) {
2150 return -ENOMEM;
2153 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
2154 ret = bdrv_co_pread(s->data_file, host_offset, bytes, buf, 0);
2155 if (ret < 0) {
2156 goto fail;
2159 if (qcow2_co_decrypt(bs, host_offset, offset, buf, bytes) < 0)
2161 ret = -EIO;
2162 goto fail;
2164 qemu_iovec_from_buf(qiov, qiov_offset, buf, bytes);
2166 fail:
2167 qemu_vfree(buf);
2169 return ret;
2172 typedef struct Qcow2AioTask {
2173 AioTask task;
2175 BlockDriverState *bs;
2176 QCow2SubclusterType subcluster_type; /* only for read */
2177 uint64_t host_offset; /* or full descriptor in compressed clusters */
2178 uint64_t offset;
2179 uint64_t bytes;
2180 QEMUIOVector *qiov;
2181 uint64_t qiov_offset;
2182 QCowL2Meta *l2meta; /* only for write */
2183 } Qcow2AioTask;
2185 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task);
2186 static coroutine_fn int qcow2_add_task(BlockDriverState *bs,
2187 AioTaskPool *pool,
2188 AioTaskFunc func,
2189 QCow2SubclusterType subcluster_type,
2190 uint64_t host_offset,
2191 uint64_t offset,
2192 uint64_t bytes,
2193 QEMUIOVector *qiov,
2194 size_t qiov_offset,
2195 QCowL2Meta *l2meta)
2197 Qcow2AioTask local_task;
2198 Qcow2AioTask *task = pool ? g_new(Qcow2AioTask, 1) : &local_task;
2200 *task = (Qcow2AioTask) {
2201 .task.func = func,
2202 .bs = bs,
2203 .subcluster_type = subcluster_type,
2204 .qiov = qiov,
2205 .host_offset = host_offset,
2206 .offset = offset,
2207 .bytes = bytes,
2208 .qiov_offset = qiov_offset,
2209 .l2meta = l2meta,
2212 trace_qcow2_add_task(qemu_coroutine_self(), bs, pool,
2213 func == qcow2_co_preadv_task_entry ? "read" : "write",
2214 subcluster_type, host_offset, offset, bytes,
2215 qiov, qiov_offset);
2217 if (!pool) {
2218 return func(&task->task);
2221 aio_task_pool_start_task(pool, &task->task);
2223 return 0;
2226 static coroutine_fn int qcow2_co_preadv_task(BlockDriverState *bs,
2227 QCow2SubclusterType subc_type,
2228 uint64_t host_offset,
2229 uint64_t offset, uint64_t bytes,
2230 QEMUIOVector *qiov,
2231 size_t qiov_offset)
2233 BDRVQcow2State *s = bs->opaque;
2235 switch (subc_type) {
2236 case QCOW2_SUBCLUSTER_ZERO_PLAIN:
2237 case QCOW2_SUBCLUSTER_ZERO_ALLOC:
2238 /* Both zero types are handled in qcow2_co_preadv_part */
2239 g_assert_not_reached();
2241 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN:
2242 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC:
2243 assert(bs->backing); /* otherwise handled in qcow2_co_preadv_part */
2245 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
2246 return bdrv_co_preadv_part(bs->backing, offset, bytes,
2247 qiov, qiov_offset, 0);
2249 case QCOW2_SUBCLUSTER_COMPRESSED:
2250 return qcow2_co_preadv_compressed(bs, host_offset,
2251 offset, bytes, qiov, qiov_offset);
2253 case QCOW2_SUBCLUSTER_NORMAL:
2254 if (bs->encrypted) {
2255 return qcow2_co_preadv_encrypted(bs, host_offset,
2256 offset, bytes, qiov, qiov_offset);
2259 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
2260 return bdrv_co_preadv_part(s->data_file, host_offset,
2261 bytes, qiov, qiov_offset, 0);
2263 default:
2264 g_assert_not_reached();
2267 g_assert_not_reached();
2270 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task)
2272 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
2274 assert(!t->l2meta);
2276 return qcow2_co_preadv_task(t->bs, t->subcluster_type,
2277 t->host_offset, t->offset, t->bytes,
2278 t->qiov, t->qiov_offset);
2281 static coroutine_fn int qcow2_co_preadv_part(BlockDriverState *bs,
2282 uint64_t offset, uint64_t bytes,
2283 QEMUIOVector *qiov,
2284 size_t qiov_offset, int flags)
2286 BDRVQcow2State *s = bs->opaque;
2287 int ret = 0;
2288 unsigned int cur_bytes; /* number of bytes in current iteration */
2289 uint64_t host_offset = 0;
2290 QCow2SubclusterType type;
2291 AioTaskPool *aio = NULL;
2293 while (bytes != 0 && aio_task_pool_status(aio) == 0) {
2294 /* prepare next request */
2295 cur_bytes = MIN(bytes, INT_MAX);
2296 if (s->crypto) {
2297 cur_bytes = MIN(cur_bytes,
2298 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2301 qemu_co_mutex_lock(&s->lock);
2302 ret = qcow2_get_host_offset(bs, offset, &cur_bytes,
2303 &host_offset, &type);
2304 qemu_co_mutex_unlock(&s->lock);
2305 if (ret < 0) {
2306 goto out;
2309 if (type == QCOW2_SUBCLUSTER_ZERO_PLAIN ||
2310 type == QCOW2_SUBCLUSTER_ZERO_ALLOC ||
2311 (type == QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN && !bs->backing) ||
2312 (type == QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC && !bs->backing))
2314 qemu_iovec_memset(qiov, qiov_offset, 0, cur_bytes);
2315 } else {
2316 if (!aio && cur_bytes != bytes) {
2317 aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
2319 ret = qcow2_add_task(bs, aio, qcow2_co_preadv_task_entry, type,
2320 host_offset, offset, cur_bytes,
2321 qiov, qiov_offset, NULL);
2322 if (ret < 0) {
2323 goto out;
2327 bytes -= cur_bytes;
2328 offset += cur_bytes;
2329 qiov_offset += cur_bytes;
2332 out:
2333 if (aio) {
2334 aio_task_pool_wait_all(aio);
2335 if (ret == 0) {
2336 ret = aio_task_pool_status(aio);
2338 g_free(aio);
2341 return ret;
2344 /* Check if it's possible to merge a write request with the writing of
2345 * the data from the COW regions */
2346 static bool merge_cow(uint64_t offset, unsigned bytes,
2347 QEMUIOVector *qiov, size_t qiov_offset,
2348 QCowL2Meta *l2meta)
2350 QCowL2Meta *m;
2352 for (m = l2meta; m != NULL; m = m->next) {
2353 /* If both COW regions are empty then there's nothing to merge */
2354 if (m->cow_start.nb_bytes == 0 && m->cow_end.nb_bytes == 0) {
2355 continue;
2358 /* If COW regions are handled already, skip this too */
2359 if (m->skip_cow) {
2360 continue;
2364 * The write request should start immediately after the first
2365 * COW region. This does not always happen because the area
2366 * touched by the request can be larger than the one defined
2367 * by @m (a single request can span an area consisting of a
2368 * mix of previously unallocated and allocated clusters, that
2369 * is why @l2meta is a list).
2371 if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) {
2372 /* In this case the request starts before this region */
2373 assert(offset < l2meta_cow_start(m));
2374 assert(m->cow_start.nb_bytes == 0);
2375 continue;
2378 /* The write request should end immediately before the second
2379 * COW region (see above for why it does not always happen) */
2380 if (m->offset + m->cow_end.offset != offset + bytes) {
2381 assert(offset + bytes > m->offset + m->cow_end.offset);
2382 assert(m->cow_end.nb_bytes == 0);
2383 continue;
2386 /* Make sure that adding both COW regions to the QEMUIOVector
2387 * does not exceed IOV_MAX */
2388 if (qemu_iovec_subvec_niov(qiov, qiov_offset, bytes) > IOV_MAX - 2) {
2389 continue;
2392 m->data_qiov = qiov;
2393 m->data_qiov_offset = qiov_offset;
2394 return true;
2397 return false;
2401 * Return 1 if the COW regions read as zeroes, 0 if not, < 0 on error.
2402 * Note that returning 0 does not guarantee non-zero data.
2404 static int is_zero_cow(BlockDriverState *bs, QCowL2Meta *m)
2407 * This check is designed for optimization shortcut so it must be
2408 * efficient.
2409 * Instead of is_zero(), use bdrv_co_is_zero_fast() as it is
2410 * faster (but not as accurate and can result in false negatives).
2412 int ret = bdrv_co_is_zero_fast(bs, m->offset + m->cow_start.offset,
2413 m->cow_start.nb_bytes);
2414 if (ret <= 0) {
2415 return ret;
2418 return bdrv_co_is_zero_fast(bs, m->offset + m->cow_end.offset,
2419 m->cow_end.nb_bytes);
2422 static int handle_alloc_space(BlockDriverState *bs, QCowL2Meta *l2meta)
2424 BDRVQcow2State *s = bs->opaque;
2425 QCowL2Meta *m;
2427 if (!(s->data_file->bs->supported_zero_flags & BDRV_REQ_NO_FALLBACK)) {
2428 return 0;
2431 if (bs->encrypted) {
2432 return 0;
2435 for (m = l2meta; m != NULL; m = m->next) {
2436 int ret;
2437 uint64_t start_offset = m->alloc_offset + m->cow_start.offset;
2438 unsigned nb_bytes = m->cow_end.offset + m->cow_end.nb_bytes -
2439 m->cow_start.offset;
2441 if (!m->cow_start.nb_bytes && !m->cow_end.nb_bytes) {
2442 continue;
2445 ret = is_zero_cow(bs, m);
2446 if (ret < 0) {
2447 return ret;
2448 } else if (ret == 0) {
2449 continue;
2453 * instead of writing zero COW buffers,
2454 * efficiently zero out the whole clusters
2457 ret = qcow2_pre_write_overlap_check(bs, 0, start_offset, nb_bytes,
2458 true);
2459 if (ret < 0) {
2460 return ret;
2463 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_SPACE);
2464 ret = bdrv_co_pwrite_zeroes(s->data_file, start_offset, nb_bytes,
2465 BDRV_REQ_NO_FALLBACK);
2466 if (ret < 0) {
2467 if (ret != -ENOTSUP && ret != -EAGAIN) {
2468 return ret;
2470 continue;
2473 trace_qcow2_skip_cow(qemu_coroutine_self(), m->offset, m->nb_clusters);
2474 m->skip_cow = true;
2476 return 0;
2480 * qcow2_co_pwritev_task
2481 * Called with s->lock unlocked
2482 * l2meta - if not NULL, qcow2_co_pwritev_task() will consume it. Caller must
2483 * not use it somehow after qcow2_co_pwritev_task() call
2485 static coroutine_fn int qcow2_co_pwritev_task(BlockDriverState *bs,
2486 uint64_t host_offset,
2487 uint64_t offset, uint64_t bytes,
2488 QEMUIOVector *qiov,
2489 uint64_t qiov_offset,
2490 QCowL2Meta *l2meta)
2492 int ret;
2493 BDRVQcow2State *s = bs->opaque;
2494 void *crypt_buf = NULL;
2495 QEMUIOVector encrypted_qiov;
2497 if (bs->encrypted) {
2498 assert(s->crypto);
2499 assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2500 crypt_buf = qemu_try_blockalign(bs->file->bs, bytes);
2501 if (crypt_buf == NULL) {
2502 ret = -ENOMEM;
2503 goto out_unlocked;
2505 qemu_iovec_to_buf(qiov, qiov_offset, crypt_buf, bytes);
2507 if (qcow2_co_encrypt(bs, host_offset, offset, crypt_buf, bytes) < 0) {
2508 ret = -EIO;
2509 goto out_unlocked;
2512 qemu_iovec_init_buf(&encrypted_qiov, crypt_buf, bytes);
2513 qiov = &encrypted_qiov;
2514 qiov_offset = 0;
2517 /* Try to efficiently initialize the physical space with zeroes */
2518 ret = handle_alloc_space(bs, l2meta);
2519 if (ret < 0) {
2520 goto out_unlocked;
2524 * If we need to do COW, check if it's possible to merge the
2525 * writing of the guest data together with that of the COW regions.
2526 * If it's not possible (or not necessary) then write the
2527 * guest data now.
2529 if (!merge_cow(offset, bytes, qiov, qiov_offset, l2meta)) {
2530 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
2531 trace_qcow2_writev_data(qemu_coroutine_self(), host_offset);
2532 ret = bdrv_co_pwritev_part(s->data_file, host_offset,
2533 bytes, qiov, qiov_offset, 0);
2534 if (ret < 0) {
2535 goto out_unlocked;
2539 qemu_co_mutex_lock(&s->lock);
2541 ret = qcow2_handle_l2meta(bs, &l2meta, true);
2542 goto out_locked;
2544 out_unlocked:
2545 qemu_co_mutex_lock(&s->lock);
2547 out_locked:
2548 qcow2_handle_l2meta(bs, &l2meta, false);
2549 qemu_co_mutex_unlock(&s->lock);
2551 qemu_vfree(crypt_buf);
2553 return ret;
2556 static coroutine_fn int qcow2_co_pwritev_task_entry(AioTask *task)
2558 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
2560 assert(!t->subcluster_type);
2562 return qcow2_co_pwritev_task(t->bs, t->host_offset,
2563 t->offset, t->bytes, t->qiov, t->qiov_offset,
2564 t->l2meta);
2567 static coroutine_fn int qcow2_co_pwritev_part(
2568 BlockDriverState *bs, uint64_t offset, uint64_t bytes,
2569 QEMUIOVector *qiov, size_t qiov_offset, int flags)
2571 BDRVQcow2State *s = bs->opaque;
2572 int offset_in_cluster;
2573 int ret;
2574 unsigned int cur_bytes; /* number of sectors in current iteration */
2575 uint64_t host_offset;
2576 QCowL2Meta *l2meta = NULL;
2577 AioTaskPool *aio = NULL;
2579 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes);
2581 while (bytes != 0 && aio_task_pool_status(aio) == 0) {
2583 l2meta = NULL;
2585 trace_qcow2_writev_start_part(qemu_coroutine_self());
2586 offset_in_cluster = offset_into_cluster(s, offset);
2587 cur_bytes = MIN(bytes, INT_MAX);
2588 if (bs->encrypted) {
2589 cur_bytes = MIN(cur_bytes,
2590 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
2591 - offset_in_cluster);
2594 qemu_co_mutex_lock(&s->lock);
2596 ret = qcow2_alloc_host_offset(bs, offset, &cur_bytes,
2597 &host_offset, &l2meta);
2598 if (ret < 0) {
2599 goto out_locked;
2602 ret = qcow2_pre_write_overlap_check(bs, 0, host_offset,
2603 cur_bytes, true);
2604 if (ret < 0) {
2605 goto out_locked;
2608 qemu_co_mutex_unlock(&s->lock);
2610 if (!aio && cur_bytes != bytes) {
2611 aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
2613 ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_task_entry, 0,
2614 host_offset, offset,
2615 cur_bytes, qiov, qiov_offset, l2meta);
2616 l2meta = NULL; /* l2meta is consumed by qcow2_co_pwritev_task() */
2617 if (ret < 0) {
2618 goto fail_nometa;
2621 bytes -= cur_bytes;
2622 offset += cur_bytes;
2623 qiov_offset += cur_bytes;
2624 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes);
2626 ret = 0;
2628 qemu_co_mutex_lock(&s->lock);
2630 out_locked:
2631 qcow2_handle_l2meta(bs, &l2meta, false);
2633 qemu_co_mutex_unlock(&s->lock);
2635 fail_nometa:
2636 if (aio) {
2637 aio_task_pool_wait_all(aio);
2638 if (ret == 0) {
2639 ret = aio_task_pool_status(aio);
2641 g_free(aio);
2644 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
2646 return ret;
2649 static int qcow2_inactivate(BlockDriverState *bs)
2651 BDRVQcow2State *s = bs->opaque;
2652 int ret, result = 0;
2653 Error *local_err = NULL;
2655 qcow2_store_persistent_dirty_bitmaps(bs, true, &local_err);
2656 if (local_err != NULL) {
2657 result = -EINVAL;
2658 error_reportf_err(local_err, "Lost persistent bitmaps during "
2659 "inactivation of node '%s': ",
2660 bdrv_get_device_or_node_name(bs));
2663 ret = qcow2_cache_flush(bs, s->l2_table_cache);
2664 if (ret) {
2665 result = ret;
2666 error_report("Failed to flush the L2 table cache: %s",
2667 strerror(-ret));
2670 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2671 if (ret) {
2672 result = ret;
2673 error_report("Failed to flush the refcount block cache: %s",
2674 strerror(-ret));
2677 if (result == 0) {
2678 qcow2_mark_clean(bs);
2681 return result;
2684 static void qcow2_close(BlockDriverState *bs)
2686 BDRVQcow2State *s = bs->opaque;
2687 qemu_vfree(s->l1_table);
2688 /* else pre-write overlap checks in cache_destroy may crash */
2689 s->l1_table = NULL;
2691 if (!(s->flags & BDRV_O_INACTIVE)) {
2692 qcow2_inactivate(bs);
2695 cache_clean_timer_del(bs);
2696 qcow2_cache_destroy(s->l2_table_cache);
2697 qcow2_cache_destroy(s->refcount_block_cache);
2699 qcrypto_block_free(s->crypto);
2700 s->crypto = NULL;
2701 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
2703 g_free(s->unknown_header_fields);
2704 cleanup_unknown_header_ext(bs);
2706 g_free(s->image_data_file);
2707 g_free(s->image_backing_file);
2708 g_free(s->image_backing_format);
2710 if (has_data_file(bs)) {
2711 bdrv_unref_child(bs, s->data_file);
2712 s->data_file = NULL;
2715 qcow2_refcount_close(bs);
2716 qcow2_free_snapshots(bs);
2719 static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs,
2720 Error **errp)
2722 BDRVQcow2State *s = bs->opaque;
2723 int flags = s->flags;
2724 QCryptoBlock *crypto = NULL;
2725 QDict *options;
2726 Error *local_err = NULL;
2727 int ret;
2730 * Backing files are read-only which makes all of their metadata immutable,
2731 * that means we don't have to worry about reopening them here.
2734 crypto = s->crypto;
2735 s->crypto = NULL;
2737 qcow2_close(bs);
2739 memset(s, 0, sizeof(BDRVQcow2State));
2740 options = qdict_clone_shallow(bs->options);
2742 flags &= ~BDRV_O_INACTIVE;
2743 qemu_co_mutex_lock(&s->lock);
2744 ret = qcow2_do_open(bs, options, flags, &local_err);
2745 qemu_co_mutex_unlock(&s->lock);
2746 qobject_unref(options);
2747 if (local_err) {
2748 error_propagate_prepend(errp, local_err,
2749 "Could not reopen qcow2 layer: ");
2750 bs->drv = NULL;
2751 return;
2752 } else if (ret < 0) {
2753 error_setg_errno(errp, -ret, "Could not reopen qcow2 layer");
2754 bs->drv = NULL;
2755 return;
2758 s->crypto = crypto;
2761 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
2762 size_t len, size_t buflen)
2764 QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
2765 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
2767 if (buflen < ext_len) {
2768 return -ENOSPC;
2771 *ext_backing_fmt = (QCowExtension) {
2772 .magic = cpu_to_be32(magic),
2773 .len = cpu_to_be32(len),
2776 if (len) {
2777 memcpy(buf + sizeof(QCowExtension), s, len);
2780 return ext_len;
2784 * Updates the qcow2 header, including the variable length parts of it, i.e.
2785 * the backing file name and all extensions. qcow2 was not designed to allow
2786 * such changes, so if we run out of space (we can only use the first cluster)
2787 * this function may fail.
2789 * Returns 0 on success, -errno in error cases.
2791 int qcow2_update_header(BlockDriverState *bs)
2793 BDRVQcow2State *s = bs->opaque;
2794 QCowHeader *header;
2795 char *buf;
2796 size_t buflen = s->cluster_size;
2797 int ret;
2798 uint64_t total_size;
2799 uint32_t refcount_table_clusters;
2800 size_t header_length;
2801 Qcow2UnknownHeaderExtension *uext;
2803 buf = qemu_blockalign(bs, buflen);
2805 /* Header structure */
2806 header = (QCowHeader*) buf;
2808 if (buflen < sizeof(*header)) {
2809 ret = -ENOSPC;
2810 goto fail;
2813 header_length = sizeof(*header) + s->unknown_header_fields_size;
2814 total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
2815 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2817 ret = validate_compression_type(s, NULL);
2818 if (ret) {
2819 goto fail;
2822 *header = (QCowHeader) {
2823 /* Version 2 fields */
2824 .magic = cpu_to_be32(QCOW_MAGIC),
2825 .version = cpu_to_be32(s->qcow_version),
2826 .backing_file_offset = 0,
2827 .backing_file_size = 0,
2828 .cluster_bits = cpu_to_be32(s->cluster_bits),
2829 .size = cpu_to_be64(total_size),
2830 .crypt_method = cpu_to_be32(s->crypt_method_header),
2831 .l1_size = cpu_to_be32(s->l1_size),
2832 .l1_table_offset = cpu_to_be64(s->l1_table_offset),
2833 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset),
2834 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
2835 .nb_snapshots = cpu_to_be32(s->nb_snapshots),
2836 .snapshots_offset = cpu_to_be64(s->snapshots_offset),
2838 /* Version 3 fields */
2839 .incompatible_features = cpu_to_be64(s->incompatible_features),
2840 .compatible_features = cpu_to_be64(s->compatible_features),
2841 .autoclear_features = cpu_to_be64(s->autoclear_features),
2842 .refcount_order = cpu_to_be32(s->refcount_order),
2843 .header_length = cpu_to_be32(header_length),
2844 .compression_type = s->compression_type,
2847 /* For older versions, write a shorter header */
2848 switch (s->qcow_version) {
2849 case 2:
2850 ret = offsetof(QCowHeader, incompatible_features);
2851 break;
2852 case 3:
2853 ret = sizeof(*header);
2854 break;
2855 default:
2856 ret = -EINVAL;
2857 goto fail;
2860 buf += ret;
2861 buflen -= ret;
2862 memset(buf, 0, buflen);
2864 /* Preserve any unknown field in the header */
2865 if (s->unknown_header_fields_size) {
2866 if (buflen < s->unknown_header_fields_size) {
2867 ret = -ENOSPC;
2868 goto fail;
2871 memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
2872 buf += s->unknown_header_fields_size;
2873 buflen -= s->unknown_header_fields_size;
2876 /* Backing file format header extension */
2877 if (s->image_backing_format) {
2878 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
2879 s->image_backing_format,
2880 strlen(s->image_backing_format),
2881 buflen);
2882 if (ret < 0) {
2883 goto fail;
2886 buf += ret;
2887 buflen -= ret;
2890 /* External data file header extension */
2891 if (has_data_file(bs) && s->image_data_file) {
2892 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_DATA_FILE,
2893 s->image_data_file, strlen(s->image_data_file),
2894 buflen);
2895 if (ret < 0) {
2896 goto fail;
2899 buf += ret;
2900 buflen -= ret;
2903 /* Full disk encryption header pointer extension */
2904 if (s->crypto_header.offset != 0) {
2905 s->crypto_header.offset = cpu_to_be64(s->crypto_header.offset);
2906 s->crypto_header.length = cpu_to_be64(s->crypto_header.length);
2907 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_CRYPTO_HEADER,
2908 &s->crypto_header, sizeof(s->crypto_header),
2909 buflen);
2910 s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset);
2911 s->crypto_header.length = be64_to_cpu(s->crypto_header.length);
2912 if (ret < 0) {
2913 goto fail;
2915 buf += ret;
2916 buflen -= ret;
2920 * Feature table. A mere 8 feature names occupies 392 bytes, and
2921 * when coupled with the v3 minimum header of 104 bytes plus the
2922 * 8-byte end-of-extension marker, that would leave only 8 bytes
2923 * for a backing file name in an image with 512-byte clusters.
2924 * Thus, we choose to omit this header for cluster sizes 4k and
2925 * smaller.
2927 if (s->qcow_version >= 3 && s->cluster_size > 4096) {
2928 static const Qcow2Feature features[] = {
2930 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2931 .bit = QCOW2_INCOMPAT_DIRTY_BITNR,
2932 .name = "dirty bit",
2935 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2936 .bit = QCOW2_INCOMPAT_CORRUPT_BITNR,
2937 .name = "corrupt bit",
2940 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2941 .bit = QCOW2_INCOMPAT_DATA_FILE_BITNR,
2942 .name = "external data file",
2945 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2946 .bit = QCOW2_INCOMPAT_COMPRESSION_BITNR,
2947 .name = "compression type",
2950 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2951 .bit = QCOW2_INCOMPAT_EXTL2_BITNR,
2952 .name = "extended L2 entries",
2955 .type = QCOW2_FEAT_TYPE_COMPATIBLE,
2956 .bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
2957 .name = "lazy refcounts",
2960 .type = QCOW2_FEAT_TYPE_AUTOCLEAR,
2961 .bit = QCOW2_AUTOCLEAR_BITMAPS_BITNR,
2962 .name = "bitmaps",
2965 .type = QCOW2_FEAT_TYPE_AUTOCLEAR,
2966 .bit = QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR,
2967 .name = "raw external data",
2971 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
2972 features, sizeof(features), buflen);
2973 if (ret < 0) {
2974 goto fail;
2976 buf += ret;
2977 buflen -= ret;
2980 /* Bitmap extension */
2981 if (s->nb_bitmaps > 0) {
2982 Qcow2BitmapHeaderExt bitmaps_header = {
2983 .nb_bitmaps = cpu_to_be32(s->nb_bitmaps),
2984 .bitmap_directory_size =
2985 cpu_to_be64(s->bitmap_directory_size),
2986 .bitmap_directory_offset =
2987 cpu_to_be64(s->bitmap_directory_offset)
2989 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BITMAPS,
2990 &bitmaps_header, sizeof(bitmaps_header),
2991 buflen);
2992 if (ret < 0) {
2993 goto fail;
2995 buf += ret;
2996 buflen -= ret;
2999 /* Keep unknown header extensions */
3000 QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
3001 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
3002 if (ret < 0) {
3003 goto fail;
3006 buf += ret;
3007 buflen -= ret;
3010 /* End of header extensions */
3011 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
3012 if (ret < 0) {
3013 goto fail;
3016 buf += ret;
3017 buflen -= ret;
3019 /* Backing file name */
3020 if (s->image_backing_file) {
3021 size_t backing_file_len = strlen(s->image_backing_file);
3023 if (buflen < backing_file_len) {
3024 ret = -ENOSPC;
3025 goto fail;
3028 /* Using strncpy is ok here, since buf is not NUL-terminated. */
3029 strncpy(buf, s->image_backing_file, buflen);
3031 header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
3032 header->backing_file_size = cpu_to_be32(backing_file_len);
3035 /* Write the new header */
3036 ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
3037 if (ret < 0) {
3038 goto fail;
3041 ret = 0;
3042 fail:
3043 qemu_vfree(header);
3044 return ret;
3047 static int qcow2_change_backing_file(BlockDriverState *bs,
3048 const char *backing_file, const char *backing_fmt)
3050 BDRVQcow2State *s = bs->opaque;
3052 /* Adding a backing file means that the external data file alone won't be
3053 * enough to make sense of the content */
3054 if (backing_file && data_file_is_raw(bs)) {
3055 return -EINVAL;
3058 if (backing_file && strlen(backing_file) > 1023) {
3059 return -EINVAL;
3062 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3063 backing_file ?: "");
3064 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
3065 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
3067 g_free(s->image_backing_file);
3068 g_free(s->image_backing_format);
3070 s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
3071 s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
3073 return qcow2_update_header(bs);
3076 static int qcow2_set_up_encryption(BlockDriverState *bs,
3077 QCryptoBlockCreateOptions *cryptoopts,
3078 Error **errp)
3080 BDRVQcow2State *s = bs->opaque;
3081 QCryptoBlock *crypto = NULL;
3082 int fmt, ret;
3084 switch (cryptoopts->format) {
3085 case Q_CRYPTO_BLOCK_FORMAT_LUKS:
3086 fmt = QCOW_CRYPT_LUKS;
3087 break;
3088 case Q_CRYPTO_BLOCK_FORMAT_QCOW:
3089 fmt = QCOW_CRYPT_AES;
3090 break;
3091 default:
3092 error_setg(errp, "Crypto format not supported in qcow2");
3093 return -EINVAL;
3096 s->crypt_method_header = fmt;
3098 crypto = qcrypto_block_create(cryptoopts, "encrypt.",
3099 qcow2_crypto_hdr_init_func,
3100 qcow2_crypto_hdr_write_func,
3101 bs, errp);
3102 if (!crypto) {
3103 return -EINVAL;
3106 ret = qcow2_update_header(bs);
3107 if (ret < 0) {
3108 error_setg_errno(errp, -ret, "Could not write encryption header");
3109 goto out;
3112 ret = 0;
3113 out:
3114 qcrypto_block_free(crypto);
3115 return ret;
3119 * Preallocates metadata structures for data clusters between @offset (in the
3120 * guest disk) and @new_length (which is thus generally the new guest disk
3121 * size).
3123 * Returns: 0 on success, -errno on failure.
3125 static int coroutine_fn preallocate_co(BlockDriverState *bs, uint64_t offset,
3126 uint64_t new_length, PreallocMode mode,
3127 Error **errp)
3129 BDRVQcow2State *s = bs->opaque;
3130 uint64_t bytes;
3131 uint64_t host_offset = 0;
3132 int64_t file_length;
3133 unsigned int cur_bytes;
3134 int ret;
3135 QCowL2Meta *meta = NULL, *m;
3137 assert(offset <= new_length);
3138 bytes = new_length - offset;
3140 while (bytes) {
3141 cur_bytes = MIN(bytes, QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size));
3142 ret = qcow2_alloc_host_offset(bs, offset, &cur_bytes,
3143 &host_offset, &meta);
3144 if (ret < 0) {
3145 error_setg_errno(errp, -ret, "Allocating clusters failed");
3146 goto out;
3149 for (m = meta; m != NULL; m = m->next) {
3150 m->prealloc = true;
3153 ret = qcow2_handle_l2meta(bs, &meta, true);
3154 if (ret < 0) {
3155 error_setg_errno(errp, -ret, "Mapping clusters failed");
3156 goto out;
3159 /* TODO Preallocate data if requested */
3161 bytes -= cur_bytes;
3162 offset += cur_bytes;
3166 * It is expected that the image file is large enough to actually contain
3167 * all of the allocated clusters (otherwise we get failing reads after
3168 * EOF). Extend the image to the last allocated sector.
3170 file_length = bdrv_getlength(s->data_file->bs);
3171 if (file_length < 0) {
3172 error_setg_errno(errp, -file_length, "Could not get file size");
3173 ret = file_length;
3174 goto out;
3177 if (host_offset + cur_bytes > file_length) {
3178 if (mode == PREALLOC_MODE_METADATA) {
3179 mode = PREALLOC_MODE_OFF;
3181 ret = bdrv_co_truncate(s->data_file, host_offset + cur_bytes, false,
3182 mode, 0, errp);
3183 if (ret < 0) {
3184 goto out;
3188 ret = 0;
3190 out:
3191 qcow2_handle_l2meta(bs, &meta, false);
3192 return ret;
3195 /* qcow2_refcount_metadata_size:
3196 * @clusters: number of clusters to refcount (including data and L1/L2 tables)
3197 * @cluster_size: size of a cluster, in bytes
3198 * @refcount_order: refcount bits power-of-2 exponent
3199 * @generous_increase: allow for the refcount table to be 1.5x as large as it
3200 * needs to be
3202 * Returns: Number of bytes required for refcount blocks and table metadata.
3204 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size,
3205 int refcount_order, bool generous_increase,
3206 uint64_t *refblock_count)
3209 * Every host cluster is reference-counted, including metadata (even
3210 * refcount metadata is recursively included).
3212 * An accurate formula for the size of refcount metadata size is difficult
3213 * to derive. An easier method of calculation is finding the fixed point
3214 * where no further refcount blocks or table clusters are required to
3215 * reference count every cluster.
3217 int64_t blocks_per_table_cluster = cluster_size / REFTABLE_ENTRY_SIZE;
3218 int64_t refcounts_per_block = cluster_size * 8 / (1 << refcount_order);
3219 int64_t table = 0; /* number of refcount table clusters */
3220 int64_t blocks = 0; /* number of refcount block clusters */
3221 int64_t last;
3222 int64_t n = 0;
3224 do {
3225 last = n;
3226 blocks = DIV_ROUND_UP(clusters + table + blocks, refcounts_per_block);
3227 table = DIV_ROUND_UP(blocks, blocks_per_table_cluster);
3228 n = clusters + blocks + table;
3230 if (n == last && generous_increase) {
3231 clusters += DIV_ROUND_UP(table, 2);
3232 n = 0; /* force another loop */
3233 generous_increase = false;
3235 } while (n != last);
3237 if (refblock_count) {
3238 *refblock_count = blocks;
3241 return (blocks + table) * cluster_size;
3245 * qcow2_calc_prealloc_size:
3246 * @total_size: virtual disk size in bytes
3247 * @cluster_size: cluster size in bytes
3248 * @refcount_order: refcount bits power-of-2 exponent
3249 * @extended_l2: true if the image has extended L2 entries
3251 * Returns: Total number of bytes required for the fully allocated image
3252 * (including metadata).
3254 static int64_t qcow2_calc_prealloc_size(int64_t total_size,
3255 size_t cluster_size,
3256 int refcount_order,
3257 bool extended_l2)
3259 int64_t meta_size = 0;
3260 uint64_t nl1e, nl2e;
3261 int64_t aligned_total_size = ROUND_UP(total_size, cluster_size);
3262 size_t l2e_size = extended_l2 ? L2E_SIZE_EXTENDED : L2E_SIZE_NORMAL;
3264 /* header: 1 cluster */
3265 meta_size += cluster_size;
3267 /* total size of L2 tables */
3268 nl2e = aligned_total_size / cluster_size;
3269 nl2e = ROUND_UP(nl2e, cluster_size / l2e_size);
3270 meta_size += nl2e * l2e_size;
3272 /* total size of L1 tables */
3273 nl1e = nl2e * l2e_size / cluster_size;
3274 nl1e = ROUND_UP(nl1e, cluster_size / L1E_SIZE);
3275 meta_size += nl1e * L1E_SIZE;
3277 /* total size of refcount table and blocks */
3278 meta_size += qcow2_refcount_metadata_size(
3279 (meta_size + aligned_total_size) / cluster_size,
3280 cluster_size, refcount_order, false, NULL);
3282 return meta_size + aligned_total_size;
3285 static bool validate_cluster_size(size_t cluster_size, bool extended_l2,
3286 Error **errp)
3288 int cluster_bits = ctz32(cluster_size);
3289 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
3290 (1 << cluster_bits) != cluster_size)
3292 error_setg(errp, "Cluster size must be a power of two between %d and "
3293 "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
3294 return false;
3297 if (extended_l2) {
3298 unsigned min_cluster_size =
3299 (1 << MIN_CLUSTER_BITS) * QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER;
3300 if (cluster_size < min_cluster_size) {
3301 error_setg(errp, "Extended L2 entries are only supported with "
3302 "cluster sizes of at least %u bytes", min_cluster_size);
3303 return false;
3307 return true;
3310 static size_t qcow2_opt_get_cluster_size_del(QemuOpts *opts, bool extended_l2,
3311 Error **errp)
3313 size_t cluster_size;
3315 cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
3316 DEFAULT_CLUSTER_SIZE);
3317 if (!validate_cluster_size(cluster_size, extended_l2, errp)) {
3318 return 0;
3320 return cluster_size;
3323 static int qcow2_opt_get_version_del(QemuOpts *opts, Error **errp)
3325 char *buf;
3326 int ret;
3328 buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
3329 if (!buf) {
3330 ret = 3; /* default */
3331 } else if (!strcmp(buf, "0.10")) {
3332 ret = 2;
3333 } else if (!strcmp(buf, "1.1")) {
3334 ret = 3;
3335 } else {
3336 error_setg(errp, "Invalid compatibility level: '%s'", buf);
3337 ret = -EINVAL;
3339 g_free(buf);
3340 return ret;
3343 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts *opts, int version,
3344 Error **errp)
3346 uint64_t refcount_bits;
3348 refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 16);
3349 if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
3350 error_setg(errp, "Refcount width must be a power of two and may not "
3351 "exceed 64 bits");
3352 return 0;
3355 if (version < 3 && refcount_bits != 16) {
3356 error_setg(errp, "Different refcount widths than 16 bits require "
3357 "compatibility level 1.1 or above (use compat=1.1 or "
3358 "greater)");
3359 return 0;
3362 return refcount_bits;
3365 static int coroutine_fn
3366 qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp)
3368 BlockdevCreateOptionsQcow2 *qcow2_opts;
3369 QDict *options;
3372 * Open the image file and write a minimal qcow2 header.
3374 * We keep things simple and start with a zero-sized image. We also
3375 * do without refcount blocks or a L1 table for now. We'll fix the
3376 * inconsistency later.
3378 * We do need a refcount table because growing the refcount table means
3379 * allocating two new refcount blocks - the second of which would be at
3380 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
3381 * size for any qcow2 image.
3383 BlockBackend *blk = NULL;
3384 BlockDriverState *bs = NULL;
3385 BlockDriverState *data_bs = NULL;
3386 QCowHeader *header;
3387 size_t cluster_size;
3388 int version;
3389 int refcount_order;
3390 uint64_t *refcount_table;
3391 int ret;
3392 uint8_t compression_type = QCOW2_COMPRESSION_TYPE_ZLIB;
3394 assert(create_options->driver == BLOCKDEV_DRIVER_QCOW2);
3395 qcow2_opts = &create_options->u.qcow2;
3397 bs = bdrv_open_blockdev_ref(qcow2_opts->file, errp);
3398 if (bs == NULL) {
3399 return -EIO;
3402 /* Validate options and set default values */
3403 if (!QEMU_IS_ALIGNED(qcow2_opts->size, BDRV_SECTOR_SIZE)) {
3404 error_setg(errp, "Image size must be a multiple of %u bytes",
3405 (unsigned) BDRV_SECTOR_SIZE);
3406 ret = -EINVAL;
3407 goto out;
3410 if (qcow2_opts->has_version) {
3411 switch (qcow2_opts->version) {
3412 case BLOCKDEV_QCOW2_VERSION_V2:
3413 version = 2;
3414 break;
3415 case BLOCKDEV_QCOW2_VERSION_V3:
3416 version = 3;
3417 break;
3418 default:
3419 g_assert_not_reached();
3421 } else {
3422 version = 3;
3425 if (qcow2_opts->has_cluster_size) {
3426 cluster_size = qcow2_opts->cluster_size;
3427 } else {
3428 cluster_size = DEFAULT_CLUSTER_SIZE;
3431 if (!qcow2_opts->has_extended_l2) {
3432 qcow2_opts->extended_l2 = false;
3434 if (qcow2_opts->extended_l2) {
3435 if (version < 3) {
3436 error_setg(errp, "Extended L2 entries are only supported with "
3437 "compatibility level 1.1 and above (use version=v3 or "
3438 "greater)");
3439 ret = -EINVAL;
3440 goto out;
3444 if (!validate_cluster_size(cluster_size, qcow2_opts->extended_l2, errp)) {
3445 ret = -EINVAL;
3446 goto out;
3449 if (!qcow2_opts->has_preallocation) {
3450 qcow2_opts->preallocation = PREALLOC_MODE_OFF;
3452 if (qcow2_opts->has_backing_file &&
3453 qcow2_opts->preallocation != PREALLOC_MODE_OFF &&
3454 !qcow2_opts->extended_l2)
3456 error_setg(errp, "Backing file and preallocation can only be used at "
3457 "the same time if extended_l2 is on");
3458 ret = -EINVAL;
3459 goto out;
3461 if (qcow2_opts->has_backing_fmt && !qcow2_opts->has_backing_file) {
3462 error_setg(errp, "Backing format cannot be used without backing file");
3463 ret = -EINVAL;
3464 goto out;
3467 if (!qcow2_opts->has_lazy_refcounts) {
3468 qcow2_opts->lazy_refcounts = false;
3470 if (version < 3 && qcow2_opts->lazy_refcounts) {
3471 error_setg(errp, "Lazy refcounts only supported with compatibility "
3472 "level 1.1 and above (use version=v3 or greater)");
3473 ret = -EINVAL;
3474 goto out;
3477 if (!qcow2_opts->has_refcount_bits) {
3478 qcow2_opts->refcount_bits = 16;
3480 if (qcow2_opts->refcount_bits > 64 ||
3481 !is_power_of_2(qcow2_opts->refcount_bits))
3483 error_setg(errp, "Refcount width must be a power of two and may not "
3484 "exceed 64 bits");
3485 ret = -EINVAL;
3486 goto out;
3488 if (version < 3 && qcow2_opts->refcount_bits != 16) {
3489 error_setg(errp, "Different refcount widths than 16 bits require "
3490 "compatibility level 1.1 or above (use version=v3 or "
3491 "greater)");
3492 ret = -EINVAL;
3493 goto out;
3495 refcount_order = ctz32(qcow2_opts->refcount_bits);
3497 if (qcow2_opts->data_file_raw && !qcow2_opts->data_file) {
3498 error_setg(errp, "data-file-raw requires data-file");
3499 ret = -EINVAL;
3500 goto out;
3502 if (qcow2_opts->data_file_raw && qcow2_opts->has_backing_file) {
3503 error_setg(errp, "Backing file and data-file-raw cannot be used at "
3504 "the same time");
3505 ret = -EINVAL;
3506 goto out;
3509 if (qcow2_opts->data_file) {
3510 if (version < 3) {
3511 error_setg(errp, "External data files are only supported with "
3512 "compatibility level 1.1 and above (use version=v3 or "
3513 "greater)");
3514 ret = -EINVAL;
3515 goto out;
3517 data_bs = bdrv_open_blockdev_ref(qcow2_opts->data_file, errp);
3518 if (data_bs == NULL) {
3519 ret = -EIO;
3520 goto out;
3524 if (qcow2_opts->has_compression_type &&
3525 qcow2_opts->compression_type != QCOW2_COMPRESSION_TYPE_ZLIB) {
3527 ret = -EINVAL;
3529 if (version < 3) {
3530 error_setg(errp, "Non-zlib compression type is only supported with "
3531 "compatibility level 1.1 and above (use version=v3 or "
3532 "greater)");
3533 goto out;
3536 switch (qcow2_opts->compression_type) {
3537 #ifdef CONFIG_ZSTD
3538 case QCOW2_COMPRESSION_TYPE_ZSTD:
3539 break;
3540 #endif
3541 default:
3542 error_setg(errp, "Unknown compression type");
3543 goto out;
3546 compression_type = qcow2_opts->compression_type;
3549 /* Create BlockBackend to write to the image */
3550 blk = blk_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL,
3551 errp);
3552 if (!blk) {
3553 ret = -EPERM;
3554 goto out;
3556 blk_set_allow_write_beyond_eof(blk, true);
3558 /* Write the header */
3559 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
3560 header = g_malloc0(cluster_size);
3561 *header = (QCowHeader) {
3562 .magic = cpu_to_be32(QCOW_MAGIC),
3563 .version = cpu_to_be32(version),
3564 .cluster_bits = cpu_to_be32(ctz32(cluster_size)),
3565 .size = cpu_to_be64(0),
3566 .l1_table_offset = cpu_to_be64(0),
3567 .l1_size = cpu_to_be32(0),
3568 .refcount_table_offset = cpu_to_be64(cluster_size),
3569 .refcount_table_clusters = cpu_to_be32(1),
3570 .refcount_order = cpu_to_be32(refcount_order),
3571 /* don't deal with endianness since compression_type is 1 byte long */
3572 .compression_type = compression_type,
3573 .header_length = cpu_to_be32(sizeof(*header)),
3576 /* We'll update this to correct value later */
3577 header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
3579 if (qcow2_opts->lazy_refcounts) {
3580 header->compatible_features |=
3581 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
3583 if (data_bs) {
3584 header->incompatible_features |=
3585 cpu_to_be64(QCOW2_INCOMPAT_DATA_FILE);
3587 if (qcow2_opts->data_file_raw) {
3588 header->autoclear_features |=
3589 cpu_to_be64(QCOW2_AUTOCLEAR_DATA_FILE_RAW);
3591 if (compression_type != QCOW2_COMPRESSION_TYPE_ZLIB) {
3592 header->incompatible_features |=
3593 cpu_to_be64(QCOW2_INCOMPAT_COMPRESSION);
3596 if (qcow2_opts->extended_l2) {
3597 header->incompatible_features |=
3598 cpu_to_be64(QCOW2_INCOMPAT_EXTL2);
3601 ret = blk_pwrite(blk, 0, header, cluster_size, 0);
3602 g_free(header);
3603 if (ret < 0) {
3604 error_setg_errno(errp, -ret, "Could not write qcow2 header");
3605 goto out;
3608 /* Write a refcount table with one refcount block */
3609 refcount_table = g_malloc0(2 * cluster_size);
3610 refcount_table[0] = cpu_to_be64(2 * cluster_size);
3611 ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0);
3612 g_free(refcount_table);
3614 if (ret < 0) {
3615 error_setg_errno(errp, -ret, "Could not write refcount table");
3616 goto out;
3619 blk_unref(blk);
3620 blk = NULL;
3623 * And now open the image and make it consistent first (i.e. increase the
3624 * refcount of the cluster that is occupied by the header and the refcount
3625 * table)
3627 options = qdict_new();
3628 qdict_put_str(options, "driver", "qcow2");
3629 qdict_put_str(options, "file", bs->node_name);
3630 if (data_bs) {
3631 qdict_put_str(options, "data-file", data_bs->node_name);
3633 blk = blk_new_open(NULL, NULL, options,
3634 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH,
3635 errp);
3636 if (blk == NULL) {
3637 ret = -EIO;
3638 goto out;
3641 ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size);
3642 if (ret < 0) {
3643 error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
3644 "header and refcount table");
3645 goto out;
3647 } else if (ret != 0) {
3648 error_report("Huh, first cluster in empty image is already in use?");
3649 abort();
3652 /* Set the external data file if necessary */
3653 if (data_bs) {
3654 BDRVQcow2State *s = blk_bs(blk)->opaque;
3655 s->image_data_file = g_strdup(data_bs->filename);
3658 /* Create a full header (including things like feature table) */
3659 ret = qcow2_update_header(blk_bs(blk));
3660 if (ret < 0) {
3661 error_setg_errno(errp, -ret, "Could not update qcow2 header");
3662 goto out;
3665 /* Okay, now that we have a valid image, let's give it the right size */
3666 ret = blk_truncate(blk, qcow2_opts->size, false, qcow2_opts->preallocation,
3667 0, errp);
3668 if (ret < 0) {
3669 error_prepend(errp, "Could not resize image: ");
3670 goto out;
3673 /* Want a backing file? There you go. */
3674 if (qcow2_opts->has_backing_file) {
3675 const char *backing_format = NULL;
3677 if (qcow2_opts->has_backing_fmt) {
3678 backing_format = BlockdevDriver_str(qcow2_opts->backing_fmt);
3681 ret = bdrv_change_backing_file(blk_bs(blk), qcow2_opts->backing_file,
3682 backing_format, false);
3683 if (ret < 0) {
3684 error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
3685 "with format '%s'", qcow2_opts->backing_file,
3686 backing_format);
3687 goto out;
3691 /* Want encryption? There you go. */
3692 if (qcow2_opts->has_encrypt) {
3693 ret = qcow2_set_up_encryption(blk_bs(blk), qcow2_opts->encrypt, errp);
3694 if (ret < 0) {
3695 goto out;
3699 blk_unref(blk);
3700 blk = NULL;
3702 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
3703 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
3704 * have to setup decryption context. We're not doing any I/O on the top
3705 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
3706 * not have effect.
3708 options = qdict_new();
3709 qdict_put_str(options, "driver", "qcow2");
3710 qdict_put_str(options, "file", bs->node_name);
3711 if (data_bs) {
3712 qdict_put_str(options, "data-file", data_bs->node_name);
3714 blk = blk_new_open(NULL, NULL, options,
3715 BDRV_O_RDWR | BDRV_O_NO_BACKING | BDRV_O_NO_IO,
3716 errp);
3717 if (blk == NULL) {
3718 ret = -EIO;
3719 goto out;
3722 ret = 0;
3723 out:
3724 blk_unref(blk);
3725 bdrv_unref(bs);
3726 bdrv_unref(data_bs);
3727 return ret;
3730 static int coroutine_fn qcow2_co_create_opts(BlockDriver *drv,
3731 const char *filename,
3732 QemuOpts *opts,
3733 Error **errp)
3735 BlockdevCreateOptions *create_options = NULL;
3736 QDict *qdict;
3737 Visitor *v;
3738 BlockDriverState *bs = NULL;
3739 BlockDriverState *data_bs = NULL;
3740 const char *val;
3741 int ret;
3743 /* Only the keyval visitor supports the dotted syntax needed for
3744 * encryption, so go through a QDict before getting a QAPI type. Ignore
3745 * options meant for the protocol layer so that the visitor doesn't
3746 * complain. */
3747 qdict = qemu_opts_to_qdict_filtered(opts, NULL, bdrv_qcow2.create_opts,
3748 true);
3750 /* Handle encryption options */
3751 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT);
3752 if (val && !strcmp(val, "on")) {
3753 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow");
3754 } else if (val && !strcmp(val, "off")) {
3755 qdict_del(qdict, BLOCK_OPT_ENCRYPT);
3758 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT);
3759 if (val && !strcmp(val, "aes")) {
3760 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow");
3763 /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into
3764 * version=v2/v3 below. */
3765 val = qdict_get_try_str(qdict, BLOCK_OPT_COMPAT_LEVEL);
3766 if (val && !strcmp(val, "0.10")) {
3767 qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v2");
3768 } else if (val && !strcmp(val, "1.1")) {
3769 qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v3");
3772 /* Change legacy command line options into QMP ones */
3773 static const QDictRenames opt_renames[] = {
3774 { BLOCK_OPT_BACKING_FILE, "backing-file" },
3775 { BLOCK_OPT_BACKING_FMT, "backing-fmt" },
3776 { BLOCK_OPT_CLUSTER_SIZE, "cluster-size" },
3777 { BLOCK_OPT_LAZY_REFCOUNTS, "lazy-refcounts" },
3778 { BLOCK_OPT_EXTL2, "extended-l2" },
3779 { BLOCK_OPT_REFCOUNT_BITS, "refcount-bits" },
3780 { BLOCK_OPT_ENCRYPT, BLOCK_OPT_ENCRYPT_FORMAT },
3781 { BLOCK_OPT_COMPAT_LEVEL, "version" },
3782 { BLOCK_OPT_DATA_FILE_RAW, "data-file-raw" },
3783 { BLOCK_OPT_COMPRESSION_TYPE, "compression-type" },
3784 { NULL, NULL },
3787 if (!qdict_rename_keys(qdict, opt_renames, errp)) {
3788 ret = -EINVAL;
3789 goto finish;
3792 /* Create and open the file (protocol layer) */
3793 ret = bdrv_create_file(filename, opts, errp);
3794 if (ret < 0) {
3795 goto finish;
3798 bs = bdrv_open(filename, NULL, NULL,
3799 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
3800 if (bs == NULL) {
3801 ret = -EIO;
3802 goto finish;
3805 /* Create and open an external data file (protocol layer) */
3806 val = qdict_get_try_str(qdict, BLOCK_OPT_DATA_FILE);
3807 if (val) {
3808 ret = bdrv_create_file(val, opts, errp);
3809 if (ret < 0) {
3810 goto finish;
3813 data_bs = bdrv_open(val, NULL, NULL,
3814 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
3815 errp);
3816 if (data_bs == NULL) {
3817 ret = -EIO;
3818 goto finish;
3821 qdict_del(qdict, BLOCK_OPT_DATA_FILE);
3822 qdict_put_str(qdict, "data-file", data_bs->node_name);
3825 /* Set 'driver' and 'node' options */
3826 qdict_put_str(qdict, "driver", "qcow2");
3827 qdict_put_str(qdict, "file", bs->node_name);
3829 /* Now get the QAPI type BlockdevCreateOptions */
3830 v = qobject_input_visitor_new_flat_confused(qdict, errp);
3831 if (!v) {
3832 ret = -EINVAL;
3833 goto finish;
3836 visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp);
3837 visit_free(v);
3838 if (!create_options) {
3839 ret = -EINVAL;
3840 goto finish;
3843 /* Silently round up size */
3844 create_options->u.qcow2.size = ROUND_UP(create_options->u.qcow2.size,
3845 BDRV_SECTOR_SIZE);
3847 /* Create the qcow2 image (format layer) */
3848 ret = qcow2_co_create(create_options, errp);
3849 if (ret < 0) {
3850 goto finish;
3853 ret = 0;
3854 finish:
3855 qobject_unref(qdict);
3856 bdrv_unref(bs);
3857 bdrv_unref(data_bs);
3858 qapi_free_BlockdevCreateOptions(create_options);
3859 return ret;
3863 static bool is_zero(BlockDriverState *bs, int64_t offset, int64_t bytes)
3865 int64_t nr;
3866 int res;
3868 /* Clamp to image length, before checking status of underlying sectors */
3869 if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) {
3870 bytes = bs->total_sectors * BDRV_SECTOR_SIZE - offset;
3873 if (!bytes) {
3874 return true;
3878 * bdrv_block_status_above doesn't merge different types of zeros, for
3879 * example, zeros which come from the region which is unallocated in
3880 * the whole backing chain, and zeros which come because of a short
3881 * backing file. So, we need a loop.
3883 do {
3884 res = bdrv_block_status_above(bs, NULL, offset, bytes, &nr, NULL, NULL);
3885 offset += nr;
3886 bytes -= nr;
3887 } while (res >= 0 && (res & BDRV_BLOCK_ZERO) && nr && bytes);
3889 return res >= 0 && (res & BDRV_BLOCK_ZERO) && bytes == 0;
3892 static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs,
3893 int64_t offset, int bytes, BdrvRequestFlags flags)
3895 int ret;
3896 BDRVQcow2State *s = bs->opaque;
3898 uint32_t head = offset_into_subcluster(s, offset);
3899 uint32_t tail = ROUND_UP(offset + bytes, s->subcluster_size) -
3900 (offset + bytes);
3902 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, bytes);
3903 if (offset + bytes == bs->total_sectors * BDRV_SECTOR_SIZE) {
3904 tail = 0;
3907 if (head || tail) {
3908 uint64_t off;
3909 unsigned int nr;
3910 QCow2SubclusterType type;
3912 assert(head + bytes + tail <= s->subcluster_size);
3914 /* check whether remainder of cluster already reads as zero */
3915 if (!(is_zero(bs, offset - head, head) &&
3916 is_zero(bs, offset + bytes, tail))) {
3917 return -ENOTSUP;
3920 qemu_co_mutex_lock(&s->lock);
3921 /* We can have new write after previous check */
3922 offset -= head;
3923 bytes = s->subcluster_size;
3924 nr = s->subcluster_size;
3925 ret = qcow2_get_host_offset(bs, offset, &nr, &off, &type);
3926 if (ret < 0 ||
3927 (type != QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN &&
3928 type != QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC &&
3929 type != QCOW2_SUBCLUSTER_ZERO_PLAIN &&
3930 type != QCOW2_SUBCLUSTER_ZERO_ALLOC)) {
3931 qemu_co_mutex_unlock(&s->lock);
3932 return ret < 0 ? ret : -ENOTSUP;
3934 } else {
3935 qemu_co_mutex_lock(&s->lock);
3938 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, bytes);
3940 /* Whatever is left can use real zero subclusters */
3941 ret = qcow2_subcluster_zeroize(bs, offset, bytes, flags);
3942 qemu_co_mutex_unlock(&s->lock);
3944 return ret;
3947 static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs,
3948 int64_t offset, int bytes)
3950 int ret;
3951 BDRVQcow2State *s = bs->opaque;
3953 /* If the image does not support QCOW_OFLAG_ZERO then discarding
3954 * clusters could expose stale data from the backing file. */
3955 if (s->qcow_version < 3 && bs->backing) {
3956 return -ENOTSUP;
3959 if (!QEMU_IS_ALIGNED(offset | bytes, s->cluster_size)) {
3960 assert(bytes < s->cluster_size);
3961 /* Ignore partial clusters, except for the special case of the
3962 * complete partial cluster at the end of an unaligned file */
3963 if (!QEMU_IS_ALIGNED(offset, s->cluster_size) ||
3964 offset + bytes != bs->total_sectors * BDRV_SECTOR_SIZE) {
3965 return -ENOTSUP;
3969 qemu_co_mutex_lock(&s->lock);
3970 ret = qcow2_cluster_discard(bs, offset, bytes, QCOW2_DISCARD_REQUEST,
3971 false);
3972 qemu_co_mutex_unlock(&s->lock);
3973 return ret;
3976 static int coroutine_fn
3977 qcow2_co_copy_range_from(BlockDriverState *bs,
3978 BdrvChild *src, uint64_t src_offset,
3979 BdrvChild *dst, uint64_t dst_offset,
3980 uint64_t bytes, BdrvRequestFlags read_flags,
3981 BdrvRequestFlags write_flags)
3983 BDRVQcow2State *s = bs->opaque;
3984 int ret;
3985 unsigned int cur_bytes; /* number of bytes in current iteration */
3986 BdrvChild *child = NULL;
3987 BdrvRequestFlags cur_write_flags;
3989 assert(!bs->encrypted);
3990 qemu_co_mutex_lock(&s->lock);
3992 while (bytes != 0) {
3993 uint64_t copy_offset = 0;
3994 QCow2SubclusterType type;
3995 /* prepare next request */
3996 cur_bytes = MIN(bytes, INT_MAX);
3997 cur_write_flags = write_flags;
3999 ret = qcow2_get_host_offset(bs, src_offset, &cur_bytes,
4000 &copy_offset, &type);
4001 if (ret < 0) {
4002 goto out;
4005 switch (type) {
4006 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN:
4007 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC:
4008 if (bs->backing && bs->backing->bs) {
4009 int64_t backing_length = bdrv_getlength(bs->backing->bs);
4010 if (src_offset >= backing_length) {
4011 cur_write_flags |= BDRV_REQ_ZERO_WRITE;
4012 } else {
4013 child = bs->backing;
4014 cur_bytes = MIN(cur_bytes, backing_length - src_offset);
4015 copy_offset = src_offset;
4017 } else {
4018 cur_write_flags |= BDRV_REQ_ZERO_WRITE;
4020 break;
4022 case QCOW2_SUBCLUSTER_ZERO_PLAIN:
4023 case QCOW2_SUBCLUSTER_ZERO_ALLOC:
4024 cur_write_flags |= BDRV_REQ_ZERO_WRITE;
4025 break;
4027 case QCOW2_SUBCLUSTER_COMPRESSED:
4028 ret = -ENOTSUP;
4029 goto out;
4031 case QCOW2_SUBCLUSTER_NORMAL:
4032 child = s->data_file;
4033 break;
4035 default:
4036 abort();
4038 qemu_co_mutex_unlock(&s->lock);
4039 ret = bdrv_co_copy_range_from(child,
4040 copy_offset,
4041 dst, dst_offset,
4042 cur_bytes, read_flags, cur_write_flags);
4043 qemu_co_mutex_lock(&s->lock);
4044 if (ret < 0) {
4045 goto out;
4048 bytes -= cur_bytes;
4049 src_offset += cur_bytes;
4050 dst_offset += cur_bytes;
4052 ret = 0;
4054 out:
4055 qemu_co_mutex_unlock(&s->lock);
4056 return ret;
4059 static int coroutine_fn
4060 qcow2_co_copy_range_to(BlockDriverState *bs,
4061 BdrvChild *src, uint64_t src_offset,
4062 BdrvChild *dst, uint64_t dst_offset,
4063 uint64_t bytes, BdrvRequestFlags read_flags,
4064 BdrvRequestFlags write_flags)
4066 BDRVQcow2State *s = bs->opaque;
4067 int ret;
4068 unsigned int cur_bytes; /* number of sectors in current iteration */
4069 uint64_t host_offset;
4070 QCowL2Meta *l2meta = NULL;
4072 assert(!bs->encrypted);
4074 qemu_co_mutex_lock(&s->lock);
4076 while (bytes != 0) {
4078 l2meta = NULL;
4080 cur_bytes = MIN(bytes, INT_MAX);
4082 /* TODO:
4083 * If src->bs == dst->bs, we could simply copy by incrementing
4084 * the refcnt, without copying user data.
4085 * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */
4086 ret = qcow2_alloc_host_offset(bs, dst_offset, &cur_bytes,
4087 &host_offset, &l2meta);
4088 if (ret < 0) {
4089 goto fail;
4092 ret = qcow2_pre_write_overlap_check(bs, 0, host_offset, cur_bytes,
4093 true);
4094 if (ret < 0) {
4095 goto fail;
4098 qemu_co_mutex_unlock(&s->lock);
4099 ret = bdrv_co_copy_range_to(src, src_offset, s->data_file, host_offset,
4100 cur_bytes, read_flags, write_flags);
4101 qemu_co_mutex_lock(&s->lock);
4102 if (ret < 0) {
4103 goto fail;
4106 ret = qcow2_handle_l2meta(bs, &l2meta, true);
4107 if (ret) {
4108 goto fail;
4111 bytes -= cur_bytes;
4112 src_offset += cur_bytes;
4113 dst_offset += cur_bytes;
4115 ret = 0;
4117 fail:
4118 qcow2_handle_l2meta(bs, &l2meta, false);
4120 qemu_co_mutex_unlock(&s->lock);
4122 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
4124 return ret;
4127 static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset,
4128 bool exact, PreallocMode prealloc,
4129 BdrvRequestFlags flags, Error **errp)
4131 BDRVQcow2State *s = bs->opaque;
4132 uint64_t old_length;
4133 int64_t new_l1_size;
4134 int ret;
4135 QDict *options;
4137 if (prealloc != PREALLOC_MODE_OFF && prealloc != PREALLOC_MODE_METADATA &&
4138 prealloc != PREALLOC_MODE_FALLOC && prealloc != PREALLOC_MODE_FULL)
4140 error_setg(errp, "Unsupported preallocation mode '%s'",
4141 PreallocMode_str(prealloc));
4142 return -ENOTSUP;
4145 if (!QEMU_IS_ALIGNED(offset, BDRV_SECTOR_SIZE)) {
4146 error_setg(errp, "The new size must be a multiple of %u",
4147 (unsigned) BDRV_SECTOR_SIZE);
4148 return -EINVAL;
4151 qemu_co_mutex_lock(&s->lock);
4154 * Even though we store snapshot size for all images, it was not
4155 * required until v3, so it is not safe to proceed for v2.
4157 if (s->nb_snapshots && s->qcow_version < 3) {
4158 error_setg(errp, "Can't resize a v2 image which has snapshots");
4159 ret = -ENOTSUP;
4160 goto fail;
4163 /* See qcow2-bitmap.c for which bitmap scenarios prevent a resize. */
4164 if (qcow2_truncate_bitmaps_check(bs, errp)) {
4165 ret = -ENOTSUP;
4166 goto fail;
4169 old_length = bs->total_sectors * BDRV_SECTOR_SIZE;
4170 new_l1_size = size_to_l1(s, offset);
4172 if (offset < old_length) {
4173 int64_t last_cluster, old_file_size;
4174 if (prealloc != PREALLOC_MODE_OFF) {
4175 error_setg(errp,
4176 "Preallocation can't be used for shrinking an image");
4177 ret = -EINVAL;
4178 goto fail;
4181 ret = qcow2_cluster_discard(bs, ROUND_UP(offset, s->cluster_size),
4182 old_length - ROUND_UP(offset,
4183 s->cluster_size),
4184 QCOW2_DISCARD_ALWAYS, true);
4185 if (ret < 0) {
4186 error_setg_errno(errp, -ret, "Failed to discard cropped clusters");
4187 goto fail;
4190 ret = qcow2_shrink_l1_table(bs, new_l1_size);
4191 if (ret < 0) {
4192 error_setg_errno(errp, -ret,
4193 "Failed to reduce the number of L2 tables");
4194 goto fail;
4197 ret = qcow2_shrink_reftable(bs);
4198 if (ret < 0) {
4199 error_setg_errno(errp, -ret,
4200 "Failed to discard unused refblocks");
4201 goto fail;
4204 old_file_size = bdrv_getlength(bs->file->bs);
4205 if (old_file_size < 0) {
4206 error_setg_errno(errp, -old_file_size,
4207 "Failed to inquire current file length");
4208 ret = old_file_size;
4209 goto fail;
4211 last_cluster = qcow2_get_last_cluster(bs, old_file_size);
4212 if (last_cluster < 0) {
4213 error_setg_errno(errp, -last_cluster,
4214 "Failed to find the last cluster");
4215 ret = last_cluster;
4216 goto fail;
4218 if ((last_cluster + 1) * s->cluster_size < old_file_size) {
4219 Error *local_err = NULL;
4222 * Do not pass @exact here: It will not help the user if
4223 * we get an error here just because they wanted to shrink
4224 * their qcow2 image (on a block device) with qemu-img.
4225 * (And on the qcow2 layer, the @exact requirement is
4226 * always fulfilled, so there is no need to pass it on.)
4228 bdrv_co_truncate(bs->file, (last_cluster + 1) * s->cluster_size,
4229 false, PREALLOC_MODE_OFF, 0, &local_err);
4230 if (local_err) {
4231 warn_reportf_err(local_err,
4232 "Failed to truncate the tail of the image: ");
4235 } else {
4236 ret = qcow2_grow_l1_table(bs, new_l1_size, true);
4237 if (ret < 0) {
4238 error_setg_errno(errp, -ret, "Failed to grow the L1 table");
4239 goto fail;
4243 switch (prealloc) {
4244 case PREALLOC_MODE_OFF:
4245 if (has_data_file(bs)) {
4247 * If the caller wants an exact resize, the external data
4248 * file should be resized to the exact target size, too,
4249 * so we pass @exact here.
4251 ret = bdrv_co_truncate(s->data_file, offset, exact, prealloc, 0,
4252 errp);
4253 if (ret < 0) {
4254 goto fail;
4257 break;
4259 case PREALLOC_MODE_METADATA:
4260 ret = preallocate_co(bs, old_length, offset, prealloc, errp);
4261 if (ret < 0) {
4262 goto fail;
4264 break;
4266 case PREALLOC_MODE_FALLOC:
4267 case PREALLOC_MODE_FULL:
4269 int64_t allocation_start, host_offset, guest_offset;
4270 int64_t clusters_allocated;
4271 int64_t old_file_size, last_cluster, new_file_size;
4272 uint64_t nb_new_data_clusters, nb_new_l2_tables;
4273 bool subclusters_need_allocation = false;
4275 /* With a data file, preallocation means just allocating the metadata
4276 * and forwarding the truncate request to the data file */
4277 if (has_data_file(bs)) {
4278 ret = preallocate_co(bs, old_length, offset, prealloc, errp);
4279 if (ret < 0) {
4280 goto fail;
4282 break;
4285 old_file_size = bdrv_getlength(bs->file->bs);
4286 if (old_file_size < 0) {
4287 error_setg_errno(errp, -old_file_size,
4288 "Failed to inquire current file length");
4289 ret = old_file_size;
4290 goto fail;
4293 last_cluster = qcow2_get_last_cluster(bs, old_file_size);
4294 if (last_cluster >= 0) {
4295 old_file_size = (last_cluster + 1) * s->cluster_size;
4296 } else {
4297 old_file_size = ROUND_UP(old_file_size, s->cluster_size);
4300 nb_new_data_clusters = (ROUND_UP(offset, s->cluster_size) -
4301 start_of_cluster(s, old_length)) >> s->cluster_bits;
4303 /* This is an overestimation; we will not actually allocate space for
4304 * these in the file but just make sure the new refcount structures are
4305 * able to cover them so we will not have to allocate new refblocks
4306 * while entering the data blocks in the potentially new L2 tables.
4307 * (We do not actually care where the L2 tables are placed. Maybe they
4308 * are already allocated or they can be placed somewhere before
4309 * @old_file_size. It does not matter because they will be fully
4310 * allocated automatically, so they do not need to be covered by the
4311 * preallocation. All that matters is that we will not have to allocate
4312 * new refcount structures for them.) */
4313 nb_new_l2_tables = DIV_ROUND_UP(nb_new_data_clusters,
4314 s->cluster_size / l2_entry_size(s));
4315 /* The cluster range may not be aligned to L2 boundaries, so add one L2
4316 * table for a potential head/tail */
4317 nb_new_l2_tables++;
4319 allocation_start = qcow2_refcount_area(bs, old_file_size,
4320 nb_new_data_clusters +
4321 nb_new_l2_tables,
4322 true, 0, 0);
4323 if (allocation_start < 0) {
4324 error_setg_errno(errp, -allocation_start,
4325 "Failed to resize refcount structures");
4326 ret = allocation_start;
4327 goto fail;
4330 clusters_allocated = qcow2_alloc_clusters_at(bs, allocation_start,
4331 nb_new_data_clusters);
4332 if (clusters_allocated < 0) {
4333 error_setg_errno(errp, -clusters_allocated,
4334 "Failed to allocate data clusters");
4335 ret = clusters_allocated;
4336 goto fail;
4339 assert(clusters_allocated == nb_new_data_clusters);
4341 /* Allocate the data area */
4342 new_file_size = allocation_start +
4343 nb_new_data_clusters * s->cluster_size;
4345 * Image file grows, so @exact does not matter.
4347 * If we need to zero out the new area, try first whether the protocol
4348 * driver can already take care of this.
4350 if (flags & BDRV_REQ_ZERO_WRITE) {
4351 ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc,
4352 BDRV_REQ_ZERO_WRITE, NULL);
4353 if (ret >= 0) {
4354 flags &= ~BDRV_REQ_ZERO_WRITE;
4355 /* Ensure that we read zeroes and not backing file data */
4356 subclusters_need_allocation = true;
4358 } else {
4359 ret = -1;
4361 if (ret < 0) {
4362 ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc, 0,
4363 errp);
4365 if (ret < 0) {
4366 error_prepend(errp, "Failed to resize underlying file: ");
4367 qcow2_free_clusters(bs, allocation_start,
4368 nb_new_data_clusters * s->cluster_size,
4369 QCOW2_DISCARD_OTHER);
4370 goto fail;
4373 /* Create the necessary L2 entries */
4374 host_offset = allocation_start;
4375 guest_offset = old_length;
4376 while (nb_new_data_clusters) {
4377 int64_t nb_clusters = MIN(
4378 nb_new_data_clusters,
4379 s->l2_slice_size - offset_to_l2_slice_index(s, guest_offset));
4380 unsigned cow_start_length = offset_into_cluster(s, guest_offset);
4381 QCowL2Meta allocation;
4382 guest_offset = start_of_cluster(s, guest_offset);
4383 allocation = (QCowL2Meta) {
4384 .offset = guest_offset,
4385 .alloc_offset = host_offset,
4386 .nb_clusters = nb_clusters,
4387 .cow_start = {
4388 .offset = 0,
4389 .nb_bytes = cow_start_length,
4391 .cow_end = {
4392 .offset = nb_clusters << s->cluster_bits,
4393 .nb_bytes = 0,
4395 .prealloc = !subclusters_need_allocation,
4397 qemu_co_queue_init(&allocation.dependent_requests);
4399 ret = qcow2_alloc_cluster_link_l2(bs, &allocation);
4400 if (ret < 0) {
4401 error_setg_errno(errp, -ret, "Failed to update L2 tables");
4402 qcow2_free_clusters(bs, host_offset,
4403 nb_new_data_clusters * s->cluster_size,
4404 QCOW2_DISCARD_OTHER);
4405 goto fail;
4408 guest_offset += nb_clusters * s->cluster_size;
4409 host_offset += nb_clusters * s->cluster_size;
4410 nb_new_data_clusters -= nb_clusters;
4412 break;
4415 default:
4416 g_assert_not_reached();
4419 if ((flags & BDRV_REQ_ZERO_WRITE) && offset > old_length) {
4420 uint64_t zero_start = QEMU_ALIGN_UP(old_length, s->subcluster_size);
4423 * Use zero clusters as much as we can. qcow2_subcluster_zeroize()
4424 * requires a subcluster-aligned start. The end may be unaligned if
4425 * it is at the end of the image (which it is here).
4427 if (offset > zero_start) {
4428 ret = qcow2_subcluster_zeroize(bs, zero_start, offset - zero_start,
4430 if (ret < 0) {
4431 error_setg_errno(errp, -ret, "Failed to zero out new clusters");
4432 goto fail;
4436 /* Write explicit zeros for the unaligned head */
4437 if (zero_start > old_length) {
4438 uint64_t len = MIN(zero_start, offset) - old_length;
4439 uint8_t *buf = qemu_blockalign0(bs, len);
4440 QEMUIOVector qiov;
4441 qemu_iovec_init_buf(&qiov, buf, len);
4443 qemu_co_mutex_unlock(&s->lock);
4444 ret = qcow2_co_pwritev_part(bs, old_length, len, &qiov, 0, 0);
4445 qemu_co_mutex_lock(&s->lock);
4447 qemu_vfree(buf);
4448 if (ret < 0) {
4449 error_setg_errno(errp, -ret, "Failed to zero out the new area");
4450 goto fail;
4455 if (prealloc != PREALLOC_MODE_OFF) {
4456 /* Flush metadata before actually changing the image size */
4457 ret = qcow2_write_caches(bs);
4458 if (ret < 0) {
4459 error_setg_errno(errp, -ret,
4460 "Failed to flush the preallocated area to disk");
4461 goto fail;
4465 bs->total_sectors = offset / BDRV_SECTOR_SIZE;
4467 /* write updated header.size */
4468 offset = cpu_to_be64(offset);
4469 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
4470 &offset, sizeof(offset));
4471 if (ret < 0) {
4472 error_setg_errno(errp, -ret, "Failed to update the image size");
4473 goto fail;
4476 s->l1_vm_state_index = new_l1_size;
4478 /* Update cache sizes */
4479 options = qdict_clone_shallow(bs->options);
4480 ret = qcow2_update_options(bs, options, s->flags, errp);
4481 qobject_unref(options);
4482 if (ret < 0) {
4483 goto fail;
4485 ret = 0;
4486 fail:
4487 qemu_co_mutex_unlock(&s->lock);
4488 return ret;
4491 static coroutine_fn int
4492 qcow2_co_pwritev_compressed_task(BlockDriverState *bs,
4493 uint64_t offset, uint64_t bytes,
4494 QEMUIOVector *qiov, size_t qiov_offset)
4496 BDRVQcow2State *s = bs->opaque;
4497 int ret;
4498 ssize_t out_len;
4499 uint8_t *buf, *out_buf;
4500 uint64_t cluster_offset;
4502 assert(bytes == s->cluster_size || (bytes < s->cluster_size &&
4503 (offset + bytes == bs->total_sectors << BDRV_SECTOR_BITS)));
4505 buf = qemu_blockalign(bs, s->cluster_size);
4506 if (bytes < s->cluster_size) {
4507 /* Zero-pad last write if image size is not cluster aligned */
4508 memset(buf + bytes, 0, s->cluster_size - bytes);
4510 qemu_iovec_to_buf(qiov, qiov_offset, buf, bytes);
4512 out_buf = g_malloc(s->cluster_size);
4514 out_len = qcow2_co_compress(bs, out_buf, s->cluster_size - 1,
4515 buf, s->cluster_size);
4516 if (out_len == -ENOMEM) {
4517 /* could not compress: write normal cluster */
4518 ret = qcow2_co_pwritev_part(bs, offset, bytes, qiov, qiov_offset, 0);
4519 if (ret < 0) {
4520 goto fail;
4522 goto success;
4523 } else if (out_len < 0) {
4524 ret = -EINVAL;
4525 goto fail;
4528 qemu_co_mutex_lock(&s->lock);
4529 ret = qcow2_alloc_compressed_cluster_offset(bs, offset, out_len,
4530 &cluster_offset);
4531 if (ret < 0) {
4532 qemu_co_mutex_unlock(&s->lock);
4533 goto fail;
4536 ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len, true);
4537 qemu_co_mutex_unlock(&s->lock);
4538 if (ret < 0) {
4539 goto fail;
4542 BLKDBG_EVENT(s->data_file, BLKDBG_WRITE_COMPRESSED);
4543 ret = bdrv_co_pwrite(s->data_file, cluster_offset, out_len, out_buf, 0);
4544 if (ret < 0) {
4545 goto fail;
4547 success:
4548 ret = 0;
4549 fail:
4550 qemu_vfree(buf);
4551 g_free(out_buf);
4552 return ret;
4555 static coroutine_fn int qcow2_co_pwritev_compressed_task_entry(AioTask *task)
4557 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
4559 assert(!t->subcluster_type && !t->l2meta);
4561 return qcow2_co_pwritev_compressed_task(t->bs, t->offset, t->bytes, t->qiov,
4562 t->qiov_offset);
4566 * XXX: put compressed sectors first, then all the cluster aligned
4567 * tables to avoid losing bytes in alignment
4569 static coroutine_fn int
4570 qcow2_co_pwritev_compressed_part(BlockDriverState *bs,
4571 uint64_t offset, uint64_t bytes,
4572 QEMUIOVector *qiov, size_t qiov_offset)
4574 BDRVQcow2State *s = bs->opaque;
4575 AioTaskPool *aio = NULL;
4576 int ret = 0;
4578 if (has_data_file(bs)) {
4579 return -ENOTSUP;
4582 if (bytes == 0) {
4584 * align end of file to a sector boundary to ease reading with
4585 * sector based I/Os
4587 int64_t len = bdrv_getlength(bs->file->bs);
4588 if (len < 0) {
4589 return len;
4591 return bdrv_co_truncate(bs->file, len, false, PREALLOC_MODE_OFF, 0,
4592 NULL);
4595 if (offset_into_cluster(s, offset)) {
4596 return -EINVAL;
4599 if (offset_into_cluster(s, bytes) &&
4600 (offset + bytes) != (bs->total_sectors << BDRV_SECTOR_BITS)) {
4601 return -EINVAL;
4604 while (bytes && aio_task_pool_status(aio) == 0) {
4605 uint64_t chunk_size = MIN(bytes, s->cluster_size);
4607 if (!aio && chunk_size != bytes) {
4608 aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
4611 ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_compressed_task_entry,
4612 0, 0, offset, chunk_size, qiov, qiov_offset, NULL);
4613 if (ret < 0) {
4614 break;
4616 qiov_offset += chunk_size;
4617 offset += chunk_size;
4618 bytes -= chunk_size;
4621 if (aio) {
4622 aio_task_pool_wait_all(aio);
4623 if (ret == 0) {
4624 ret = aio_task_pool_status(aio);
4626 g_free(aio);
4629 return ret;
4632 static int coroutine_fn
4633 qcow2_co_preadv_compressed(BlockDriverState *bs,
4634 uint64_t cluster_descriptor,
4635 uint64_t offset,
4636 uint64_t bytes,
4637 QEMUIOVector *qiov,
4638 size_t qiov_offset)
4640 BDRVQcow2State *s = bs->opaque;
4641 int ret = 0, csize, nb_csectors;
4642 uint64_t coffset;
4643 uint8_t *buf, *out_buf;
4644 int offset_in_cluster = offset_into_cluster(s, offset);
4646 coffset = cluster_descriptor & s->cluster_offset_mask;
4647 nb_csectors = ((cluster_descriptor >> s->csize_shift) & s->csize_mask) + 1;
4648 csize = nb_csectors * QCOW2_COMPRESSED_SECTOR_SIZE -
4649 (coffset & ~QCOW2_COMPRESSED_SECTOR_MASK);
4651 buf = g_try_malloc(csize);
4652 if (!buf) {
4653 return -ENOMEM;
4656 out_buf = qemu_blockalign(bs, s->cluster_size);
4658 BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
4659 ret = bdrv_co_pread(bs->file, coffset, csize, buf, 0);
4660 if (ret < 0) {
4661 goto fail;
4664 if (qcow2_co_decompress(bs, out_buf, s->cluster_size, buf, csize) < 0) {
4665 ret = -EIO;
4666 goto fail;
4669 qemu_iovec_from_buf(qiov, qiov_offset, out_buf + offset_in_cluster, bytes);
4671 fail:
4672 qemu_vfree(out_buf);
4673 g_free(buf);
4675 return ret;
4678 static int make_completely_empty(BlockDriverState *bs)
4680 BDRVQcow2State *s = bs->opaque;
4681 Error *local_err = NULL;
4682 int ret, l1_clusters;
4683 int64_t offset;
4684 uint64_t *new_reftable = NULL;
4685 uint64_t rt_entry, l1_size2;
4686 struct {
4687 uint64_t l1_offset;
4688 uint64_t reftable_offset;
4689 uint32_t reftable_clusters;
4690 } QEMU_PACKED l1_ofs_rt_ofs_cls;
4692 ret = qcow2_cache_empty(bs, s->l2_table_cache);
4693 if (ret < 0) {
4694 goto fail;
4697 ret = qcow2_cache_empty(bs, s->refcount_block_cache);
4698 if (ret < 0) {
4699 goto fail;
4702 /* Refcounts will be broken utterly */
4703 ret = qcow2_mark_dirty(bs);
4704 if (ret < 0) {
4705 goto fail;
4708 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
4710 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / L1E_SIZE);
4711 l1_size2 = (uint64_t)s->l1_size * L1E_SIZE;
4713 /* After this call, neither the in-memory nor the on-disk refcount
4714 * information accurately describe the actual references */
4716 ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset,
4717 l1_clusters * s->cluster_size, 0);
4718 if (ret < 0) {
4719 goto fail_broken_refcounts;
4721 memset(s->l1_table, 0, l1_size2);
4723 BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE);
4725 /* Overwrite enough clusters at the beginning of the sectors to place
4726 * the refcount table, a refcount block and the L1 table in; this may
4727 * overwrite parts of the existing refcount and L1 table, which is not
4728 * an issue because the dirty flag is set, complete data loss is in fact
4729 * desired and partial data loss is consequently fine as well */
4730 ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size,
4731 (2 + l1_clusters) * s->cluster_size, 0);
4732 /* This call (even if it failed overall) may have overwritten on-disk
4733 * refcount structures; in that case, the in-memory refcount information
4734 * will probably differ from the on-disk information which makes the BDS
4735 * unusable */
4736 if (ret < 0) {
4737 goto fail_broken_refcounts;
4740 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
4741 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
4743 /* "Create" an empty reftable (one cluster) directly after the image
4744 * header and an empty L1 table three clusters after the image header;
4745 * the cluster between those two will be used as the first refblock */
4746 l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size);
4747 l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size);
4748 l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1);
4749 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset),
4750 &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls));
4751 if (ret < 0) {
4752 goto fail_broken_refcounts;
4755 s->l1_table_offset = 3 * s->cluster_size;
4757 new_reftable = g_try_new0(uint64_t, s->cluster_size / REFTABLE_ENTRY_SIZE);
4758 if (!new_reftable) {
4759 ret = -ENOMEM;
4760 goto fail_broken_refcounts;
4763 s->refcount_table_offset = s->cluster_size;
4764 s->refcount_table_size = s->cluster_size / REFTABLE_ENTRY_SIZE;
4765 s->max_refcount_table_index = 0;
4767 g_free(s->refcount_table);
4768 s->refcount_table = new_reftable;
4769 new_reftable = NULL;
4771 /* Now the in-memory refcount information again corresponds to the on-disk
4772 * information (reftable is empty and no refblocks (the refblock cache is
4773 * empty)); however, this means some clusters (e.g. the image header) are
4774 * referenced, but not refcounted, but the normal qcow2 code assumes that
4775 * the in-memory information is always correct */
4777 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
4779 /* Enter the first refblock into the reftable */
4780 rt_entry = cpu_to_be64(2 * s->cluster_size);
4781 ret = bdrv_pwrite_sync(bs->file, s->cluster_size,
4782 &rt_entry, sizeof(rt_entry));
4783 if (ret < 0) {
4784 goto fail_broken_refcounts;
4786 s->refcount_table[0] = 2 * s->cluster_size;
4788 s->free_cluster_index = 0;
4789 assert(3 + l1_clusters <= s->refcount_block_size);
4790 offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2);
4791 if (offset < 0) {
4792 ret = offset;
4793 goto fail_broken_refcounts;
4794 } else if (offset > 0) {
4795 error_report("First cluster in emptied image is in use");
4796 abort();
4799 /* Now finally the in-memory information corresponds to the on-disk
4800 * structures and is correct */
4801 ret = qcow2_mark_clean(bs);
4802 if (ret < 0) {
4803 goto fail;
4806 ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size, false,
4807 PREALLOC_MODE_OFF, 0, &local_err);
4808 if (ret < 0) {
4809 error_report_err(local_err);
4810 goto fail;
4813 return 0;
4815 fail_broken_refcounts:
4816 /* The BDS is unusable at this point. If we wanted to make it usable, we
4817 * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
4818 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
4819 * again. However, because the functions which could have caused this error
4820 * path to be taken are used by those functions as well, it's very likely
4821 * that that sequence will fail as well. Therefore, just eject the BDS. */
4822 bs->drv = NULL;
4824 fail:
4825 g_free(new_reftable);
4826 return ret;
4829 static int qcow2_make_empty(BlockDriverState *bs)
4831 BDRVQcow2State *s = bs->opaque;
4832 uint64_t offset, end_offset;
4833 int step = QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size);
4834 int l1_clusters, ret = 0;
4836 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / L1E_SIZE);
4838 if (s->qcow_version >= 3 && !s->snapshots && !s->nb_bitmaps &&
4839 3 + l1_clusters <= s->refcount_block_size &&
4840 s->crypt_method_header != QCOW_CRYPT_LUKS &&
4841 !has_data_file(bs)) {
4842 /* The following function only works for qcow2 v3 images (it
4843 * requires the dirty flag) and only as long as there are no
4844 * features that reserve extra clusters (such as snapshots,
4845 * LUKS header, or persistent bitmaps), because it completely
4846 * empties the image. Furthermore, the L1 table and three
4847 * additional clusters (image header, refcount table, one
4848 * refcount block) have to fit inside one refcount block. It
4849 * only resets the image file, i.e. does not work with an
4850 * external data file. */
4851 return make_completely_empty(bs);
4854 /* This fallback code simply discards every active cluster; this is slow,
4855 * but works in all cases */
4856 end_offset = bs->total_sectors * BDRV_SECTOR_SIZE;
4857 for (offset = 0; offset < end_offset; offset += step) {
4858 /* As this function is generally used after committing an external
4859 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
4860 * default action for this kind of discard is to pass the discard,
4861 * which will ideally result in an actually smaller image file, as
4862 * is probably desired. */
4863 ret = qcow2_cluster_discard(bs, offset, MIN(step, end_offset - offset),
4864 QCOW2_DISCARD_SNAPSHOT, true);
4865 if (ret < 0) {
4866 break;
4870 return ret;
4873 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
4875 BDRVQcow2State *s = bs->opaque;
4876 int ret;
4878 qemu_co_mutex_lock(&s->lock);
4879 ret = qcow2_write_caches(bs);
4880 qemu_co_mutex_unlock(&s->lock);
4882 return ret;
4885 static BlockMeasureInfo *qcow2_measure(QemuOpts *opts, BlockDriverState *in_bs,
4886 Error **errp)
4888 Error *local_err = NULL;
4889 BlockMeasureInfo *info;
4890 uint64_t required = 0; /* bytes that contribute to required size */
4891 uint64_t virtual_size; /* disk size as seen by guest */
4892 uint64_t refcount_bits;
4893 uint64_t l2_tables;
4894 uint64_t luks_payload_size = 0;
4895 size_t cluster_size;
4896 int version;
4897 char *optstr;
4898 PreallocMode prealloc;
4899 bool has_backing_file;
4900 bool has_luks;
4901 bool extended_l2;
4902 size_t l2e_size;
4904 /* Parse image creation options */
4905 extended_l2 = qemu_opt_get_bool_del(opts, BLOCK_OPT_EXTL2, false);
4907 cluster_size = qcow2_opt_get_cluster_size_del(opts, extended_l2,
4908 &local_err);
4909 if (local_err) {
4910 goto err;
4913 version = qcow2_opt_get_version_del(opts, &local_err);
4914 if (local_err) {
4915 goto err;
4918 refcount_bits = qcow2_opt_get_refcount_bits_del(opts, version, &local_err);
4919 if (local_err) {
4920 goto err;
4923 optstr = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
4924 prealloc = qapi_enum_parse(&PreallocMode_lookup, optstr,
4925 PREALLOC_MODE_OFF, &local_err);
4926 g_free(optstr);
4927 if (local_err) {
4928 goto err;
4931 optstr = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
4932 has_backing_file = !!optstr;
4933 g_free(optstr);
4935 optstr = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT);
4936 has_luks = optstr && strcmp(optstr, "luks") == 0;
4937 g_free(optstr);
4939 if (has_luks) {
4940 g_autoptr(QCryptoBlockCreateOptions) create_opts = NULL;
4941 QDict *cryptoopts = qcow2_extract_crypto_opts(opts, "luks", errp);
4942 size_t headerlen;
4944 create_opts = block_crypto_create_opts_init(cryptoopts, errp);
4945 qobject_unref(cryptoopts);
4946 if (!create_opts) {
4947 goto err;
4950 if (!qcrypto_block_calculate_payload_offset(create_opts,
4951 "encrypt.",
4952 &headerlen,
4953 &local_err)) {
4954 goto err;
4957 luks_payload_size = ROUND_UP(headerlen, cluster_size);
4960 virtual_size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
4961 virtual_size = ROUND_UP(virtual_size, cluster_size);
4963 /* Check that virtual disk size is valid */
4964 l2e_size = extended_l2 ? L2E_SIZE_EXTENDED : L2E_SIZE_NORMAL;
4965 l2_tables = DIV_ROUND_UP(virtual_size / cluster_size,
4966 cluster_size / l2e_size);
4967 if (l2_tables * L1E_SIZE > QCOW_MAX_L1_SIZE) {
4968 error_setg(&local_err, "The image size is too large "
4969 "(try using a larger cluster size)");
4970 goto err;
4973 /* Account for input image */
4974 if (in_bs) {
4975 int64_t ssize = bdrv_getlength(in_bs);
4976 if (ssize < 0) {
4977 error_setg_errno(&local_err, -ssize,
4978 "Unable to get image virtual_size");
4979 goto err;
4982 virtual_size = ROUND_UP(ssize, cluster_size);
4984 if (has_backing_file) {
4985 /* We don't how much of the backing chain is shared by the input
4986 * image and the new image file. In the worst case the new image's
4987 * backing file has nothing in common with the input image. Be
4988 * conservative and assume all clusters need to be written.
4990 required = virtual_size;
4991 } else {
4992 int64_t offset;
4993 int64_t pnum = 0;
4995 for (offset = 0; offset < ssize; offset += pnum) {
4996 int ret;
4998 ret = bdrv_block_status_above(in_bs, NULL, offset,
4999 ssize - offset, &pnum, NULL,
5000 NULL);
5001 if (ret < 0) {
5002 error_setg_errno(&local_err, -ret,
5003 "Unable to get block status");
5004 goto err;
5007 if (ret & BDRV_BLOCK_ZERO) {
5008 /* Skip zero regions (safe with no backing file) */
5009 } else if ((ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) ==
5010 (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) {
5011 /* Extend pnum to end of cluster for next iteration */
5012 pnum = ROUND_UP(offset + pnum, cluster_size) - offset;
5014 /* Count clusters we've seen */
5015 required += offset % cluster_size + pnum;
5021 /* Take into account preallocation. Nothing special is needed for
5022 * PREALLOC_MODE_METADATA since metadata is always counted.
5024 if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
5025 required = virtual_size;
5028 info = g_new0(BlockMeasureInfo, 1);
5029 info->fully_allocated = luks_payload_size +
5030 qcow2_calc_prealloc_size(virtual_size, cluster_size,
5031 ctz32(refcount_bits), extended_l2);
5034 * Remove data clusters that are not required. This overestimates the
5035 * required size because metadata needed for the fully allocated file is
5036 * still counted. Show bitmaps only if both source and destination
5037 * would support them.
5039 info->required = info->fully_allocated - virtual_size + required;
5040 info->has_bitmaps = version >= 3 && in_bs &&
5041 bdrv_supports_persistent_dirty_bitmap(in_bs);
5042 if (info->has_bitmaps) {
5043 info->bitmaps = qcow2_get_persistent_dirty_bitmap_size(in_bs,
5044 cluster_size);
5046 return info;
5048 err:
5049 error_propagate(errp, local_err);
5050 return NULL;
5053 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
5055 BDRVQcow2State *s = bs->opaque;
5056 bdi->cluster_size = s->cluster_size;
5057 bdi->vm_state_offset = qcow2_vm_state_offset(s);
5058 return 0;
5061 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs,
5062 Error **errp)
5064 BDRVQcow2State *s = bs->opaque;
5065 ImageInfoSpecific *spec_info;
5066 QCryptoBlockInfo *encrypt_info = NULL;
5067 Error *local_err = NULL;
5069 if (s->crypto != NULL) {
5070 encrypt_info = qcrypto_block_get_info(s->crypto, &local_err);
5071 if (local_err) {
5072 error_propagate(errp, local_err);
5073 return NULL;
5077 spec_info = g_new(ImageInfoSpecific, 1);
5078 *spec_info = (ImageInfoSpecific){
5079 .type = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
5080 .u.qcow2.data = g_new0(ImageInfoSpecificQCow2, 1),
5082 if (s->qcow_version == 2) {
5083 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
5084 .compat = g_strdup("0.10"),
5085 .refcount_bits = s->refcount_bits,
5087 } else if (s->qcow_version == 3) {
5088 Qcow2BitmapInfoList *bitmaps;
5089 bitmaps = qcow2_get_bitmap_info_list(bs, &local_err);
5090 if (local_err) {
5091 error_propagate(errp, local_err);
5092 qapi_free_ImageInfoSpecific(spec_info);
5093 qapi_free_QCryptoBlockInfo(encrypt_info);
5094 return NULL;
5096 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
5097 .compat = g_strdup("1.1"),
5098 .lazy_refcounts = s->compatible_features &
5099 QCOW2_COMPAT_LAZY_REFCOUNTS,
5100 .has_lazy_refcounts = true,
5101 .corrupt = s->incompatible_features &
5102 QCOW2_INCOMPAT_CORRUPT,
5103 .has_corrupt = true,
5104 .has_extended_l2 = true,
5105 .extended_l2 = has_subclusters(s),
5106 .refcount_bits = s->refcount_bits,
5107 .has_bitmaps = !!bitmaps,
5108 .bitmaps = bitmaps,
5109 .has_data_file = !!s->image_data_file,
5110 .data_file = g_strdup(s->image_data_file),
5111 .has_data_file_raw = has_data_file(bs),
5112 .data_file_raw = data_file_is_raw(bs),
5113 .compression_type = s->compression_type,
5115 } else {
5116 /* if this assertion fails, this probably means a new version was
5117 * added without having it covered here */
5118 assert(false);
5121 if (encrypt_info) {
5122 ImageInfoSpecificQCow2Encryption *qencrypt =
5123 g_new(ImageInfoSpecificQCow2Encryption, 1);
5124 switch (encrypt_info->format) {
5125 case Q_CRYPTO_BLOCK_FORMAT_QCOW:
5126 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES;
5127 break;
5128 case Q_CRYPTO_BLOCK_FORMAT_LUKS:
5129 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS;
5130 qencrypt->u.luks = encrypt_info->u.luks;
5131 break;
5132 default:
5133 abort();
5135 /* Since we did shallow copy above, erase any pointers
5136 * in the original info */
5137 memset(&encrypt_info->u, 0, sizeof(encrypt_info->u));
5138 qapi_free_QCryptoBlockInfo(encrypt_info);
5140 spec_info->u.qcow2.data->has_encrypt = true;
5141 spec_info->u.qcow2.data->encrypt = qencrypt;
5144 return spec_info;
5147 static int qcow2_has_zero_init(BlockDriverState *bs)
5149 BDRVQcow2State *s = bs->opaque;
5150 bool preallocated;
5152 if (qemu_in_coroutine()) {
5153 qemu_co_mutex_lock(&s->lock);
5156 * Check preallocation status: Preallocated images have all L2
5157 * tables allocated, nonpreallocated images have none. It is
5158 * therefore enough to check the first one.
5160 preallocated = s->l1_size > 0 && s->l1_table[0] != 0;
5161 if (qemu_in_coroutine()) {
5162 qemu_co_mutex_unlock(&s->lock);
5165 if (!preallocated) {
5166 return 1;
5167 } else if (bs->encrypted) {
5168 return 0;
5169 } else {
5170 return bdrv_has_zero_init(s->data_file->bs);
5174 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
5175 int64_t pos)
5177 BDRVQcow2State *s = bs->opaque;
5179 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
5180 return bs->drv->bdrv_co_pwritev_part(bs, qcow2_vm_state_offset(s) + pos,
5181 qiov->size, qiov, 0, 0);
5184 static int qcow2_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
5185 int64_t pos)
5187 BDRVQcow2State *s = bs->opaque;
5189 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
5190 return bs->drv->bdrv_co_preadv_part(bs, qcow2_vm_state_offset(s) + pos,
5191 qiov->size, qiov, 0, 0);
5195 * Downgrades an image's version. To achieve this, any incompatible features
5196 * have to be removed.
5198 static int qcow2_downgrade(BlockDriverState *bs, int target_version,
5199 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
5200 Error **errp)
5202 BDRVQcow2State *s = bs->opaque;
5203 int current_version = s->qcow_version;
5204 int ret;
5205 int i;
5207 /* This is qcow2_downgrade(), not qcow2_upgrade() */
5208 assert(target_version < current_version);
5210 /* There are no other versions (now) that you can downgrade to */
5211 assert(target_version == 2);
5213 if (s->refcount_order != 4) {
5214 error_setg(errp, "compat=0.10 requires refcount_bits=16");
5215 return -ENOTSUP;
5218 if (has_data_file(bs)) {
5219 error_setg(errp, "Cannot downgrade an image with a data file");
5220 return -ENOTSUP;
5224 * If any internal snapshot has a different size than the current
5225 * image size, or VM state size that exceeds 32 bits, downgrading
5226 * is unsafe. Even though we would still use v3-compliant output
5227 * to preserve that data, other v2 programs might not realize
5228 * those optional fields are important.
5230 for (i = 0; i < s->nb_snapshots; i++) {
5231 if (s->snapshots[i].vm_state_size > UINT32_MAX ||
5232 s->snapshots[i].disk_size != bs->total_sectors * BDRV_SECTOR_SIZE) {
5233 error_setg(errp, "Internal snapshots prevent downgrade of image");
5234 return -ENOTSUP;
5238 /* clear incompatible features */
5239 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
5240 ret = qcow2_mark_clean(bs);
5241 if (ret < 0) {
5242 error_setg_errno(errp, -ret, "Failed to make the image clean");
5243 return ret;
5247 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
5248 * the first place; if that happens nonetheless, returning -ENOTSUP is the
5249 * best thing to do anyway */
5251 if (s->incompatible_features) {
5252 error_setg(errp, "Cannot downgrade an image with incompatible features "
5253 "%#" PRIx64 " set", s->incompatible_features);
5254 return -ENOTSUP;
5257 /* since we can ignore compatible features, we can set them to 0 as well */
5258 s->compatible_features = 0;
5259 /* if lazy refcounts have been used, they have already been fixed through
5260 * clearing the dirty flag */
5262 /* clearing autoclear features is trivial */
5263 s->autoclear_features = 0;
5265 ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque);
5266 if (ret < 0) {
5267 error_setg_errno(errp, -ret, "Failed to turn zero into data clusters");
5268 return ret;
5271 s->qcow_version = target_version;
5272 ret = qcow2_update_header(bs);
5273 if (ret < 0) {
5274 s->qcow_version = current_version;
5275 error_setg_errno(errp, -ret, "Failed to update the image header");
5276 return ret;
5278 return 0;
5282 * Upgrades an image's version. While newer versions encompass all
5283 * features of older versions, some things may have to be presented
5284 * differently.
5286 static int qcow2_upgrade(BlockDriverState *bs, int target_version,
5287 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
5288 Error **errp)
5290 BDRVQcow2State *s = bs->opaque;
5291 bool need_snapshot_update;
5292 int current_version = s->qcow_version;
5293 int i;
5294 int ret;
5296 /* This is qcow2_upgrade(), not qcow2_downgrade() */
5297 assert(target_version > current_version);
5299 /* There are no other versions (yet) that you can upgrade to */
5300 assert(target_version == 3);
5302 status_cb(bs, 0, 2, cb_opaque);
5305 * In v2, snapshots do not need to have extra data. v3 requires
5306 * the 64-bit VM state size and the virtual disk size to be
5307 * present.
5308 * qcow2_write_snapshots() will always write the list in the
5309 * v3-compliant format.
5311 need_snapshot_update = false;
5312 for (i = 0; i < s->nb_snapshots; i++) {
5313 if (s->snapshots[i].extra_data_size <
5314 sizeof_field(QCowSnapshotExtraData, vm_state_size_large) +
5315 sizeof_field(QCowSnapshotExtraData, disk_size))
5317 need_snapshot_update = true;
5318 break;
5321 if (need_snapshot_update) {
5322 ret = qcow2_write_snapshots(bs);
5323 if (ret < 0) {
5324 error_setg_errno(errp, -ret, "Failed to update the snapshot table");
5325 return ret;
5328 status_cb(bs, 1, 2, cb_opaque);
5330 s->qcow_version = target_version;
5331 ret = qcow2_update_header(bs);
5332 if (ret < 0) {
5333 s->qcow_version = current_version;
5334 error_setg_errno(errp, -ret, "Failed to update the image header");
5335 return ret;
5337 status_cb(bs, 2, 2, cb_opaque);
5339 return 0;
5342 typedef enum Qcow2AmendOperation {
5343 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
5344 * statically initialized to so that the helper CB can discern the first
5345 * invocation from an operation change */
5346 QCOW2_NO_OPERATION = 0,
5348 QCOW2_UPGRADING,
5349 QCOW2_UPDATING_ENCRYPTION,
5350 QCOW2_CHANGING_REFCOUNT_ORDER,
5351 QCOW2_DOWNGRADING,
5352 } Qcow2AmendOperation;
5354 typedef struct Qcow2AmendHelperCBInfo {
5355 /* The code coordinating the amend operations should only modify
5356 * these four fields; the rest will be managed by the CB */
5357 BlockDriverAmendStatusCB *original_status_cb;
5358 void *original_cb_opaque;
5360 Qcow2AmendOperation current_operation;
5362 /* Total number of operations to perform (only set once) */
5363 int total_operations;
5365 /* The following fields are managed by the CB */
5367 /* Number of operations completed */
5368 int operations_completed;
5370 /* Cumulative offset of all completed operations */
5371 int64_t offset_completed;
5373 Qcow2AmendOperation last_operation;
5374 int64_t last_work_size;
5375 } Qcow2AmendHelperCBInfo;
5377 static void qcow2_amend_helper_cb(BlockDriverState *bs,
5378 int64_t operation_offset,
5379 int64_t operation_work_size, void *opaque)
5381 Qcow2AmendHelperCBInfo *info = opaque;
5382 int64_t current_work_size;
5383 int64_t projected_work_size;
5385 if (info->current_operation != info->last_operation) {
5386 if (info->last_operation != QCOW2_NO_OPERATION) {
5387 info->offset_completed += info->last_work_size;
5388 info->operations_completed++;
5391 info->last_operation = info->current_operation;
5394 assert(info->total_operations > 0);
5395 assert(info->operations_completed < info->total_operations);
5397 info->last_work_size = operation_work_size;
5399 current_work_size = info->offset_completed + operation_work_size;
5401 /* current_work_size is the total work size for (operations_completed + 1)
5402 * operations (which includes this one), so multiply it by the number of
5403 * operations not covered and divide it by the number of operations
5404 * covered to get a projection for the operations not covered */
5405 projected_work_size = current_work_size * (info->total_operations -
5406 info->operations_completed - 1)
5407 / (info->operations_completed + 1);
5409 info->original_status_cb(bs, info->offset_completed + operation_offset,
5410 current_work_size + projected_work_size,
5411 info->original_cb_opaque);
5414 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,
5415 BlockDriverAmendStatusCB *status_cb,
5416 void *cb_opaque,
5417 bool force,
5418 Error **errp)
5420 BDRVQcow2State *s = bs->opaque;
5421 int old_version = s->qcow_version, new_version = old_version;
5422 uint64_t new_size = 0;
5423 const char *backing_file = NULL, *backing_format = NULL, *data_file = NULL;
5424 bool lazy_refcounts = s->use_lazy_refcounts;
5425 bool data_file_raw = data_file_is_raw(bs);
5426 const char *compat = NULL;
5427 int refcount_bits = s->refcount_bits;
5428 int ret;
5429 QemuOptDesc *desc = opts->list->desc;
5430 Qcow2AmendHelperCBInfo helper_cb_info;
5431 bool encryption_update = false;
5433 while (desc && desc->name) {
5434 if (!qemu_opt_find(opts, desc->name)) {
5435 /* only change explicitly defined options */
5436 desc++;
5437 continue;
5440 if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) {
5441 compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL);
5442 if (!compat) {
5443 /* preserve default */
5444 } else if (!strcmp(compat, "0.10") || !strcmp(compat, "v2")) {
5445 new_version = 2;
5446 } else if (!strcmp(compat, "1.1") || !strcmp(compat, "v3")) {
5447 new_version = 3;
5448 } else {
5449 error_setg(errp, "Unknown compatibility level %s", compat);
5450 return -EINVAL;
5452 } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) {
5453 new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
5454 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) {
5455 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
5456 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) {
5457 backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
5458 } else if (g_str_has_prefix(desc->name, "encrypt.")) {
5459 if (!s->crypto) {
5460 error_setg(errp,
5461 "Can't amend encryption options - encryption not present");
5462 return -EINVAL;
5464 if (s->crypt_method_header != QCOW_CRYPT_LUKS) {
5465 error_setg(errp,
5466 "Only LUKS encryption options can be amended");
5467 return -ENOTSUP;
5469 encryption_update = true;
5470 } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
5471 lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS,
5472 lazy_refcounts);
5473 } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) {
5474 refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS,
5475 refcount_bits);
5477 if (refcount_bits <= 0 || refcount_bits > 64 ||
5478 !is_power_of_2(refcount_bits))
5480 error_setg(errp, "Refcount width must be a power of two and "
5481 "may not exceed 64 bits");
5482 return -EINVAL;
5484 } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE)) {
5485 data_file = qemu_opt_get(opts, BLOCK_OPT_DATA_FILE);
5486 if (data_file && !has_data_file(bs)) {
5487 error_setg(errp, "data-file can only be set for images that "
5488 "use an external data file");
5489 return -EINVAL;
5491 } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE_RAW)) {
5492 data_file_raw = qemu_opt_get_bool(opts, BLOCK_OPT_DATA_FILE_RAW,
5493 data_file_raw);
5494 if (data_file_raw && !data_file_is_raw(bs)) {
5495 error_setg(errp, "data-file-raw cannot be set on existing "
5496 "images");
5497 return -EINVAL;
5499 } else {
5500 /* if this point is reached, this probably means a new option was
5501 * added without having it covered here */
5502 abort();
5505 desc++;
5508 helper_cb_info = (Qcow2AmendHelperCBInfo){
5509 .original_status_cb = status_cb,
5510 .original_cb_opaque = cb_opaque,
5511 .total_operations = (new_version != old_version)
5512 + (s->refcount_bits != refcount_bits) +
5513 (encryption_update == true)
5516 /* Upgrade first (some features may require compat=1.1) */
5517 if (new_version > old_version) {
5518 helper_cb_info.current_operation = QCOW2_UPGRADING;
5519 ret = qcow2_upgrade(bs, new_version, &qcow2_amend_helper_cb,
5520 &helper_cb_info, errp);
5521 if (ret < 0) {
5522 return ret;
5526 if (encryption_update) {
5527 QDict *amend_opts_dict;
5528 QCryptoBlockAmendOptions *amend_opts;
5530 helper_cb_info.current_operation = QCOW2_UPDATING_ENCRYPTION;
5531 amend_opts_dict = qcow2_extract_crypto_opts(opts, "luks", errp);
5532 if (!amend_opts_dict) {
5533 return -EINVAL;
5535 amend_opts = block_crypto_amend_opts_init(amend_opts_dict, errp);
5536 qobject_unref(amend_opts_dict);
5537 if (!amend_opts) {
5538 return -EINVAL;
5540 ret = qcrypto_block_amend_options(s->crypto,
5541 qcow2_crypto_hdr_read_func,
5542 qcow2_crypto_hdr_write_func,
5544 amend_opts,
5545 force,
5546 errp);
5547 qapi_free_QCryptoBlockAmendOptions(amend_opts);
5548 if (ret < 0) {
5549 return ret;
5553 if (s->refcount_bits != refcount_bits) {
5554 int refcount_order = ctz32(refcount_bits);
5556 if (new_version < 3 && refcount_bits != 16) {
5557 error_setg(errp, "Refcount widths other than 16 bits require "
5558 "compatibility level 1.1 or above (use compat=1.1 or "
5559 "greater)");
5560 return -EINVAL;
5563 helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER;
5564 ret = qcow2_change_refcount_order(bs, refcount_order,
5565 &qcow2_amend_helper_cb,
5566 &helper_cb_info, errp);
5567 if (ret < 0) {
5568 return ret;
5572 /* data-file-raw blocks backing files, so clear it first if requested */
5573 if (data_file_raw) {
5574 s->autoclear_features |= QCOW2_AUTOCLEAR_DATA_FILE_RAW;
5575 } else {
5576 s->autoclear_features &= ~QCOW2_AUTOCLEAR_DATA_FILE_RAW;
5579 if (data_file) {
5580 g_free(s->image_data_file);
5581 s->image_data_file = *data_file ? g_strdup(data_file) : NULL;
5584 ret = qcow2_update_header(bs);
5585 if (ret < 0) {
5586 error_setg_errno(errp, -ret, "Failed to update the image header");
5587 return ret;
5590 if (backing_file || backing_format) {
5591 if (g_strcmp0(backing_file, s->image_backing_file) ||
5592 g_strcmp0(backing_format, s->image_backing_format)) {
5593 warn_report("Deprecated use of amend to alter the backing file; "
5594 "use qemu-img rebase instead");
5596 ret = qcow2_change_backing_file(bs,
5597 backing_file ?: s->image_backing_file,
5598 backing_format ?: s->image_backing_format);
5599 if (ret < 0) {
5600 error_setg_errno(errp, -ret, "Failed to change the backing file");
5601 return ret;
5605 if (s->use_lazy_refcounts != lazy_refcounts) {
5606 if (lazy_refcounts) {
5607 if (new_version < 3) {
5608 error_setg(errp, "Lazy refcounts only supported with "
5609 "compatibility level 1.1 and above (use compat=1.1 "
5610 "or greater)");
5611 return -EINVAL;
5613 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
5614 ret = qcow2_update_header(bs);
5615 if (ret < 0) {
5616 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
5617 error_setg_errno(errp, -ret, "Failed to update the image header");
5618 return ret;
5620 s->use_lazy_refcounts = true;
5621 } else {
5622 /* make image clean first */
5623 ret = qcow2_mark_clean(bs);
5624 if (ret < 0) {
5625 error_setg_errno(errp, -ret, "Failed to make the image clean");
5626 return ret;
5628 /* now disallow lazy refcounts */
5629 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
5630 ret = qcow2_update_header(bs);
5631 if (ret < 0) {
5632 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
5633 error_setg_errno(errp, -ret, "Failed to update the image header");
5634 return ret;
5636 s->use_lazy_refcounts = false;
5640 if (new_size) {
5641 BlockBackend *blk = blk_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL,
5642 errp);
5643 if (!blk) {
5644 return -EPERM;
5648 * Amending image options should ensure that the image has
5649 * exactly the given new values, so pass exact=true here.
5651 ret = blk_truncate(blk, new_size, true, PREALLOC_MODE_OFF, 0, errp);
5652 blk_unref(blk);
5653 if (ret < 0) {
5654 return ret;
5658 /* Downgrade last (so unsupported features can be removed before) */
5659 if (new_version < old_version) {
5660 helper_cb_info.current_operation = QCOW2_DOWNGRADING;
5661 ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb,
5662 &helper_cb_info, errp);
5663 if (ret < 0) {
5664 return ret;
5668 return 0;
5671 static int coroutine_fn qcow2_co_amend(BlockDriverState *bs,
5672 BlockdevAmendOptions *opts,
5673 bool force,
5674 Error **errp)
5676 BlockdevAmendOptionsQcow2 *qopts = &opts->u.qcow2;
5677 BDRVQcow2State *s = bs->opaque;
5678 int ret = 0;
5680 if (qopts->has_encrypt) {
5681 if (!s->crypto) {
5682 error_setg(errp, "image is not encrypted, can't amend");
5683 return -EOPNOTSUPP;
5686 if (qopts->encrypt->format != Q_CRYPTO_BLOCK_FORMAT_LUKS) {
5687 error_setg(errp,
5688 "Amend can't be used to change the qcow2 encryption format");
5689 return -EOPNOTSUPP;
5692 if (s->crypt_method_header != QCOW_CRYPT_LUKS) {
5693 error_setg(errp,
5694 "Only LUKS encryption options can be amended for qcow2 with blockdev-amend");
5695 return -EOPNOTSUPP;
5698 ret = qcrypto_block_amend_options(s->crypto,
5699 qcow2_crypto_hdr_read_func,
5700 qcow2_crypto_hdr_write_func,
5702 qopts->encrypt,
5703 force,
5704 errp);
5706 return ret;
5710 * If offset or size are negative, respectively, they will not be included in
5711 * the BLOCK_IMAGE_CORRUPTED event emitted.
5712 * fatal will be ignored for read-only BDS; corruptions found there will always
5713 * be considered non-fatal.
5715 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
5716 int64_t size, const char *message_format, ...)
5718 BDRVQcow2State *s = bs->opaque;
5719 const char *node_name;
5720 char *message;
5721 va_list ap;
5723 fatal = fatal && bdrv_is_writable(bs);
5725 if (s->signaled_corruption &&
5726 (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
5728 return;
5731 va_start(ap, message_format);
5732 message = g_strdup_vprintf(message_format, ap);
5733 va_end(ap);
5735 if (fatal) {
5736 fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
5737 "corruption events will be suppressed\n", message);
5738 } else {
5739 fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
5740 "corruption events will be suppressed\n", message);
5743 node_name = bdrv_get_node_name(bs);
5744 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs),
5745 *node_name != '\0', node_name,
5746 message, offset >= 0, offset,
5747 size >= 0, size,
5748 fatal);
5749 g_free(message);
5751 if (fatal) {
5752 qcow2_mark_corrupt(bs);
5753 bs->drv = NULL; /* make BDS unusable */
5756 s->signaled_corruption = true;
5759 #define QCOW_COMMON_OPTIONS \
5761 .name = BLOCK_OPT_SIZE, \
5762 .type = QEMU_OPT_SIZE, \
5763 .help = "Virtual disk size" \
5764 }, \
5766 .name = BLOCK_OPT_COMPAT_LEVEL, \
5767 .type = QEMU_OPT_STRING, \
5768 .help = "Compatibility level (v2 [0.10] or v3 [1.1])" \
5769 }, \
5771 .name = BLOCK_OPT_BACKING_FILE, \
5772 .type = QEMU_OPT_STRING, \
5773 .help = "File name of a base image" \
5774 }, \
5776 .name = BLOCK_OPT_BACKING_FMT, \
5777 .type = QEMU_OPT_STRING, \
5778 .help = "Image format of the base image" \
5779 }, \
5781 .name = BLOCK_OPT_DATA_FILE, \
5782 .type = QEMU_OPT_STRING, \
5783 .help = "File name of an external data file" \
5784 }, \
5786 .name = BLOCK_OPT_DATA_FILE_RAW, \
5787 .type = QEMU_OPT_BOOL, \
5788 .help = "The external data file must stay valid " \
5789 "as a raw image" \
5790 }, \
5792 .name = BLOCK_OPT_LAZY_REFCOUNTS, \
5793 .type = QEMU_OPT_BOOL, \
5794 .help = "Postpone refcount updates", \
5795 .def_value_str = "off" \
5796 }, \
5798 .name = BLOCK_OPT_REFCOUNT_BITS, \
5799 .type = QEMU_OPT_NUMBER, \
5800 .help = "Width of a reference count entry in bits", \
5801 .def_value_str = "16" \
5804 static QemuOptsList qcow2_create_opts = {
5805 .name = "qcow2-create-opts",
5806 .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
5807 .desc = {
5809 .name = BLOCK_OPT_ENCRYPT, \
5810 .type = QEMU_OPT_BOOL, \
5811 .help = "Encrypt the image with format 'aes'. (Deprecated " \
5812 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)", \
5813 }, \
5815 .name = BLOCK_OPT_ENCRYPT_FORMAT, \
5816 .type = QEMU_OPT_STRING, \
5817 .help = "Encrypt the image, format choices: 'aes', 'luks'", \
5818 }, \
5819 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", \
5820 "ID of secret providing qcow AES key or LUKS passphrase"), \
5821 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."), \
5822 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."), \
5823 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."), \
5824 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."), \
5825 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."), \
5826 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."), \
5828 .name = BLOCK_OPT_CLUSTER_SIZE, \
5829 .type = QEMU_OPT_SIZE, \
5830 .help = "qcow2 cluster size", \
5831 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE) \
5832 }, \
5834 .name = BLOCK_OPT_EXTL2, \
5835 .type = QEMU_OPT_BOOL, \
5836 .help = "Extended L2 tables", \
5837 .def_value_str = "off" \
5838 }, \
5840 .name = BLOCK_OPT_PREALLOC, \
5841 .type = QEMU_OPT_STRING, \
5842 .help = "Preallocation mode (allowed values: off, " \
5843 "metadata, falloc, full)" \
5844 }, \
5846 .name = BLOCK_OPT_COMPRESSION_TYPE, \
5847 .type = QEMU_OPT_STRING, \
5848 .help = "Compression method used for image cluster " \
5849 "compression", \
5850 .def_value_str = "zlib" \
5852 QCOW_COMMON_OPTIONS,
5853 { /* end of list */ }
5857 static QemuOptsList qcow2_amend_opts = {
5858 .name = "qcow2-amend-opts",
5859 .head = QTAILQ_HEAD_INITIALIZER(qcow2_amend_opts.head),
5860 .desc = {
5861 BLOCK_CRYPTO_OPT_DEF_LUKS_STATE("encrypt."),
5862 BLOCK_CRYPTO_OPT_DEF_LUKS_KEYSLOT("encrypt."),
5863 BLOCK_CRYPTO_OPT_DEF_LUKS_OLD_SECRET("encrypt."),
5864 BLOCK_CRYPTO_OPT_DEF_LUKS_NEW_SECRET("encrypt."),
5865 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
5866 QCOW_COMMON_OPTIONS,
5867 { /* end of list */ }
5871 static const char *const qcow2_strong_runtime_opts[] = {
5872 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET,
5874 NULL
5877 BlockDriver bdrv_qcow2 = {
5878 .format_name = "qcow2",
5879 .instance_size = sizeof(BDRVQcow2State),
5880 .bdrv_probe = qcow2_probe,
5881 .bdrv_open = qcow2_open,
5882 .bdrv_close = qcow2_close,
5883 .bdrv_reopen_prepare = qcow2_reopen_prepare,
5884 .bdrv_reopen_commit = qcow2_reopen_commit,
5885 .bdrv_reopen_commit_post = qcow2_reopen_commit_post,
5886 .bdrv_reopen_abort = qcow2_reopen_abort,
5887 .bdrv_join_options = qcow2_join_options,
5888 .bdrv_child_perm = bdrv_default_perms,
5889 .bdrv_co_create_opts = qcow2_co_create_opts,
5890 .bdrv_co_create = qcow2_co_create,
5891 .bdrv_has_zero_init = qcow2_has_zero_init,
5892 .bdrv_co_block_status = qcow2_co_block_status,
5894 .bdrv_co_preadv_part = qcow2_co_preadv_part,
5895 .bdrv_co_pwritev_part = qcow2_co_pwritev_part,
5896 .bdrv_co_flush_to_os = qcow2_co_flush_to_os,
5898 .bdrv_co_pwrite_zeroes = qcow2_co_pwrite_zeroes,
5899 .bdrv_co_pdiscard = qcow2_co_pdiscard,
5900 .bdrv_co_copy_range_from = qcow2_co_copy_range_from,
5901 .bdrv_co_copy_range_to = qcow2_co_copy_range_to,
5902 .bdrv_co_truncate = qcow2_co_truncate,
5903 .bdrv_co_pwritev_compressed_part = qcow2_co_pwritev_compressed_part,
5904 .bdrv_make_empty = qcow2_make_empty,
5906 .bdrv_snapshot_create = qcow2_snapshot_create,
5907 .bdrv_snapshot_goto = qcow2_snapshot_goto,
5908 .bdrv_snapshot_delete = qcow2_snapshot_delete,
5909 .bdrv_snapshot_list = qcow2_snapshot_list,
5910 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
5911 .bdrv_measure = qcow2_measure,
5912 .bdrv_get_info = qcow2_get_info,
5913 .bdrv_get_specific_info = qcow2_get_specific_info,
5915 .bdrv_save_vmstate = qcow2_save_vmstate,
5916 .bdrv_load_vmstate = qcow2_load_vmstate,
5918 .is_format = true,
5919 .supports_backing = true,
5920 .bdrv_change_backing_file = qcow2_change_backing_file,
5922 .bdrv_refresh_limits = qcow2_refresh_limits,
5923 .bdrv_co_invalidate_cache = qcow2_co_invalidate_cache,
5924 .bdrv_inactivate = qcow2_inactivate,
5926 .create_opts = &qcow2_create_opts,
5927 .amend_opts = &qcow2_amend_opts,
5928 .strong_runtime_opts = qcow2_strong_runtime_opts,
5929 .mutable_opts = mutable_opts,
5930 .bdrv_co_check = qcow2_co_check,
5931 .bdrv_amend_options = qcow2_amend_options,
5932 .bdrv_co_amend = qcow2_co_amend,
5934 .bdrv_detach_aio_context = qcow2_detach_aio_context,
5935 .bdrv_attach_aio_context = qcow2_attach_aio_context,
5937 .bdrv_supports_persistent_dirty_bitmap =
5938 qcow2_supports_persistent_dirty_bitmap,
5939 .bdrv_co_can_store_new_dirty_bitmap = qcow2_co_can_store_new_dirty_bitmap,
5940 .bdrv_co_remove_persistent_dirty_bitmap =
5941 qcow2_co_remove_persistent_dirty_bitmap,
5944 static void bdrv_qcow2_init(void)
5946 bdrv_register(&bdrv_qcow2);
5949 block_init(bdrv_qcow2_init);