tests/acceptance: Test the Arduino MEGA2560 board
[qemu/ar7.git] / block / qcow2.c
blobea33673c552c57091e62e950e38701d71300a06c
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 file_cluster_offset,
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_del(s->cache_clean_timer);
856 timer_free(s->cache_clean_timer);
857 s->cache_clean_timer = NULL;
861 static void qcow2_detach_aio_context(BlockDriverState *bs)
863 cache_clean_timer_del(bs);
866 static void qcow2_attach_aio_context(BlockDriverState *bs,
867 AioContext *new_context)
869 cache_clean_timer_init(bs, new_context);
872 static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts,
873 uint64_t *l2_cache_size,
874 uint64_t *l2_cache_entry_size,
875 uint64_t *refcount_cache_size, Error **errp)
877 BDRVQcow2State *s = bs->opaque;
878 uint64_t combined_cache_size, l2_cache_max_setting;
879 bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set;
880 bool l2_cache_entry_size_set;
881 int min_refcount_cache = MIN_REFCOUNT_CACHE_SIZE * s->cluster_size;
882 uint64_t virtual_disk_size = bs->total_sectors * BDRV_SECTOR_SIZE;
883 uint64_t max_l2_entries = DIV_ROUND_UP(virtual_disk_size, s->cluster_size);
884 /* An L2 table is always one cluster in size so the max cache size
885 * should be a multiple of the cluster size. */
886 uint64_t max_l2_cache = ROUND_UP(max_l2_entries * sizeof(uint64_t),
887 s->cluster_size);
889 combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE);
890 l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE);
891 refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
892 l2_cache_entry_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE);
894 combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0);
895 l2_cache_max_setting = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE,
896 DEFAULT_L2_CACHE_MAX_SIZE);
897 *refcount_cache_size = qemu_opt_get_size(opts,
898 QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0);
900 *l2_cache_entry_size = qemu_opt_get_size(
901 opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE, s->cluster_size);
903 *l2_cache_size = MIN(max_l2_cache, l2_cache_max_setting);
905 if (combined_cache_size_set) {
906 if (l2_cache_size_set && refcount_cache_size_set) {
907 error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE
908 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set "
909 "at the same time");
910 return;
911 } else if (l2_cache_size_set &&
912 (l2_cache_max_setting > combined_cache_size)) {
913 error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed "
914 QCOW2_OPT_CACHE_SIZE);
915 return;
916 } else if (*refcount_cache_size > combined_cache_size) {
917 error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed "
918 QCOW2_OPT_CACHE_SIZE);
919 return;
922 if (l2_cache_size_set) {
923 *refcount_cache_size = combined_cache_size - *l2_cache_size;
924 } else if (refcount_cache_size_set) {
925 *l2_cache_size = combined_cache_size - *refcount_cache_size;
926 } else {
927 /* Assign as much memory as possible to the L2 cache, and
928 * use the remainder for the refcount cache */
929 if (combined_cache_size >= max_l2_cache + min_refcount_cache) {
930 *l2_cache_size = max_l2_cache;
931 *refcount_cache_size = combined_cache_size - *l2_cache_size;
932 } else {
933 *refcount_cache_size =
934 MIN(combined_cache_size, min_refcount_cache);
935 *l2_cache_size = combined_cache_size - *refcount_cache_size;
941 * If the L2 cache is not enough to cover the whole disk then
942 * default to 4KB entries. Smaller entries reduce the cost of
943 * loads and evictions and increase I/O performance.
945 if (*l2_cache_size < max_l2_cache && !l2_cache_entry_size_set) {
946 *l2_cache_entry_size = MIN(s->cluster_size, 4096);
949 /* l2_cache_size and refcount_cache_size are ensured to have at least
950 * their minimum values in qcow2_update_options_prepare() */
952 if (*l2_cache_entry_size < (1 << MIN_CLUSTER_BITS) ||
953 *l2_cache_entry_size > s->cluster_size ||
954 !is_power_of_2(*l2_cache_entry_size)) {
955 error_setg(errp, "L2 cache entry size must be a power of two "
956 "between %d and the cluster size (%d)",
957 1 << MIN_CLUSTER_BITS, s->cluster_size);
958 return;
962 typedef struct Qcow2ReopenState {
963 Qcow2Cache *l2_table_cache;
964 Qcow2Cache *refcount_block_cache;
965 int l2_slice_size; /* Number of entries in a slice of the L2 table */
966 bool use_lazy_refcounts;
967 int overlap_check;
968 bool discard_passthrough[QCOW2_DISCARD_MAX];
969 uint64_t cache_clean_interval;
970 QCryptoBlockOpenOptions *crypto_opts; /* Disk encryption runtime options */
971 } Qcow2ReopenState;
973 static int qcow2_update_options_prepare(BlockDriverState *bs,
974 Qcow2ReopenState *r,
975 QDict *options, int flags,
976 Error **errp)
978 BDRVQcow2State *s = bs->opaque;
979 QemuOpts *opts = NULL;
980 const char *opt_overlap_check, *opt_overlap_check_template;
981 int overlap_check_template = 0;
982 uint64_t l2_cache_size, l2_cache_entry_size, refcount_cache_size;
983 int i;
984 const char *encryptfmt;
985 QDict *encryptopts = NULL;
986 Error *local_err = NULL;
987 int ret;
989 qdict_extract_subqdict(options, &encryptopts, "encrypt.");
990 encryptfmt = qdict_get_try_str(encryptopts, "format");
992 opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort);
993 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
994 ret = -EINVAL;
995 goto fail;
998 /* get L2 table/refcount block cache size from command line options */
999 read_cache_sizes(bs, opts, &l2_cache_size, &l2_cache_entry_size,
1000 &refcount_cache_size, &local_err);
1001 if (local_err) {
1002 error_propagate(errp, local_err);
1003 ret = -EINVAL;
1004 goto fail;
1007 l2_cache_size /= l2_cache_entry_size;
1008 if (l2_cache_size < MIN_L2_CACHE_SIZE) {
1009 l2_cache_size = MIN_L2_CACHE_SIZE;
1011 if (l2_cache_size > INT_MAX) {
1012 error_setg(errp, "L2 cache size too big");
1013 ret = -EINVAL;
1014 goto fail;
1017 refcount_cache_size /= s->cluster_size;
1018 if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) {
1019 refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE;
1021 if (refcount_cache_size > INT_MAX) {
1022 error_setg(errp, "Refcount cache size too big");
1023 ret = -EINVAL;
1024 goto fail;
1027 /* alloc new L2 table/refcount block cache, flush old one */
1028 if (s->l2_table_cache) {
1029 ret = qcow2_cache_flush(bs, s->l2_table_cache);
1030 if (ret) {
1031 error_setg_errno(errp, -ret, "Failed to flush the L2 table cache");
1032 goto fail;
1036 if (s->refcount_block_cache) {
1037 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
1038 if (ret) {
1039 error_setg_errno(errp, -ret,
1040 "Failed to flush the refcount block cache");
1041 goto fail;
1045 r->l2_slice_size = l2_cache_entry_size / sizeof(uint64_t);
1046 r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size,
1047 l2_cache_entry_size);
1048 r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size,
1049 s->cluster_size);
1050 if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) {
1051 error_setg(errp, "Could not allocate metadata caches");
1052 ret = -ENOMEM;
1053 goto fail;
1056 /* New interval for cache cleanup timer */
1057 r->cache_clean_interval =
1058 qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL,
1059 DEFAULT_CACHE_CLEAN_INTERVAL);
1060 #ifndef CONFIG_LINUX
1061 if (r->cache_clean_interval != 0) {
1062 error_setg(errp, QCOW2_OPT_CACHE_CLEAN_INTERVAL
1063 " not supported on this host");
1064 ret = -EINVAL;
1065 goto fail;
1067 #endif
1068 if (r->cache_clean_interval > UINT_MAX) {
1069 error_setg(errp, "Cache clean interval too big");
1070 ret = -EINVAL;
1071 goto fail;
1074 /* lazy-refcounts; flush if going from enabled to disabled */
1075 r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS,
1076 (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS));
1077 if (r->use_lazy_refcounts && s->qcow_version < 3) {
1078 error_setg(errp, "Lazy refcounts require a qcow2 image with at least "
1079 "qemu 1.1 compatibility level");
1080 ret = -EINVAL;
1081 goto fail;
1084 if (s->use_lazy_refcounts && !r->use_lazy_refcounts) {
1085 ret = qcow2_mark_clean(bs);
1086 if (ret < 0) {
1087 error_setg_errno(errp, -ret, "Failed to disable lazy refcounts");
1088 goto fail;
1092 /* Overlap check options */
1093 opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP);
1094 opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE);
1095 if (opt_overlap_check_template && opt_overlap_check &&
1096 strcmp(opt_overlap_check_template, opt_overlap_check))
1098 error_setg(errp, "Conflicting values for qcow2 options '"
1099 QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
1100 "' ('%s')", opt_overlap_check, opt_overlap_check_template);
1101 ret = -EINVAL;
1102 goto fail;
1104 if (!opt_overlap_check) {
1105 opt_overlap_check = opt_overlap_check_template ?: "cached";
1108 if (!strcmp(opt_overlap_check, "none")) {
1109 overlap_check_template = 0;
1110 } else if (!strcmp(opt_overlap_check, "constant")) {
1111 overlap_check_template = QCOW2_OL_CONSTANT;
1112 } else if (!strcmp(opt_overlap_check, "cached")) {
1113 overlap_check_template = QCOW2_OL_CACHED;
1114 } else if (!strcmp(opt_overlap_check, "all")) {
1115 overlap_check_template = QCOW2_OL_ALL;
1116 } else {
1117 error_setg(errp, "Unsupported value '%s' for qcow2 option "
1118 "'overlap-check'. Allowed are any of the following: "
1119 "none, constant, cached, all", opt_overlap_check);
1120 ret = -EINVAL;
1121 goto fail;
1124 r->overlap_check = 0;
1125 for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) {
1126 /* overlap-check defines a template bitmask, but every flag may be
1127 * overwritten through the associated boolean option */
1128 r->overlap_check |=
1129 qemu_opt_get_bool(opts, overlap_bool_option_names[i],
1130 overlap_check_template & (1 << i)) << i;
1133 r->discard_passthrough[QCOW2_DISCARD_NEVER] = false;
1134 r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true;
1135 r->discard_passthrough[QCOW2_DISCARD_REQUEST] =
1136 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST,
1137 flags & BDRV_O_UNMAP);
1138 r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] =
1139 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true);
1140 r->discard_passthrough[QCOW2_DISCARD_OTHER] =
1141 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false);
1143 switch (s->crypt_method_header) {
1144 case QCOW_CRYPT_NONE:
1145 if (encryptfmt) {
1146 error_setg(errp, "No encryption in image header, but options "
1147 "specified format '%s'", encryptfmt);
1148 ret = -EINVAL;
1149 goto fail;
1151 break;
1153 case QCOW_CRYPT_AES:
1154 if (encryptfmt && !g_str_equal(encryptfmt, "aes")) {
1155 error_setg(errp,
1156 "Header reported 'aes' encryption format but "
1157 "options specify '%s'", encryptfmt);
1158 ret = -EINVAL;
1159 goto fail;
1161 qdict_put_str(encryptopts, "format", "qcow");
1162 r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
1163 break;
1165 case QCOW_CRYPT_LUKS:
1166 if (encryptfmt && !g_str_equal(encryptfmt, "luks")) {
1167 error_setg(errp,
1168 "Header reported 'luks' encryption format but "
1169 "options specify '%s'", encryptfmt);
1170 ret = -EINVAL;
1171 goto fail;
1173 qdict_put_str(encryptopts, "format", "luks");
1174 r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
1175 break;
1177 default:
1178 error_setg(errp, "Unsupported encryption method %d",
1179 s->crypt_method_header);
1180 break;
1182 if (s->crypt_method_header != QCOW_CRYPT_NONE && !r->crypto_opts) {
1183 ret = -EINVAL;
1184 goto fail;
1187 ret = 0;
1188 fail:
1189 qobject_unref(encryptopts);
1190 qemu_opts_del(opts);
1191 opts = NULL;
1192 return ret;
1195 static void qcow2_update_options_commit(BlockDriverState *bs,
1196 Qcow2ReopenState *r)
1198 BDRVQcow2State *s = bs->opaque;
1199 int i;
1201 if (s->l2_table_cache) {
1202 qcow2_cache_destroy(s->l2_table_cache);
1204 if (s->refcount_block_cache) {
1205 qcow2_cache_destroy(s->refcount_block_cache);
1207 s->l2_table_cache = r->l2_table_cache;
1208 s->refcount_block_cache = r->refcount_block_cache;
1209 s->l2_slice_size = r->l2_slice_size;
1211 s->overlap_check = r->overlap_check;
1212 s->use_lazy_refcounts = r->use_lazy_refcounts;
1214 for (i = 0; i < QCOW2_DISCARD_MAX; i++) {
1215 s->discard_passthrough[i] = r->discard_passthrough[i];
1218 if (s->cache_clean_interval != r->cache_clean_interval) {
1219 cache_clean_timer_del(bs);
1220 s->cache_clean_interval = r->cache_clean_interval;
1221 cache_clean_timer_init(bs, bdrv_get_aio_context(bs));
1224 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1225 s->crypto_opts = r->crypto_opts;
1228 static void qcow2_update_options_abort(BlockDriverState *bs,
1229 Qcow2ReopenState *r)
1231 if (r->l2_table_cache) {
1232 qcow2_cache_destroy(r->l2_table_cache);
1234 if (r->refcount_block_cache) {
1235 qcow2_cache_destroy(r->refcount_block_cache);
1237 qapi_free_QCryptoBlockOpenOptions(r->crypto_opts);
1240 static int qcow2_update_options(BlockDriverState *bs, QDict *options,
1241 int flags, Error **errp)
1243 Qcow2ReopenState r = {};
1244 int ret;
1246 ret = qcow2_update_options_prepare(bs, &r, options, flags, errp);
1247 if (ret >= 0) {
1248 qcow2_update_options_commit(bs, &r);
1249 } else {
1250 qcow2_update_options_abort(bs, &r);
1253 return ret;
1256 static int validate_compression_type(BDRVQcow2State *s, Error **errp)
1258 switch (s->compression_type) {
1259 case QCOW2_COMPRESSION_TYPE_ZLIB:
1260 #ifdef CONFIG_ZSTD
1261 case QCOW2_COMPRESSION_TYPE_ZSTD:
1262 #endif
1263 break;
1265 default:
1266 error_setg(errp, "qcow2: unknown compression type: %u",
1267 s->compression_type);
1268 return -ENOTSUP;
1272 * if the compression type differs from QCOW2_COMPRESSION_TYPE_ZLIB
1273 * the incompatible feature flag must be set
1275 if (s->compression_type == QCOW2_COMPRESSION_TYPE_ZLIB) {
1276 if (s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION) {
1277 error_setg(errp, "qcow2: Compression type incompatible feature "
1278 "bit must not be set");
1279 return -EINVAL;
1281 } else {
1282 if (!(s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION)) {
1283 error_setg(errp, "qcow2: Compression type incompatible feature "
1284 "bit must be set");
1285 return -EINVAL;
1289 return 0;
1292 /* Called with s->lock held. */
1293 static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options,
1294 int flags, Error **errp)
1296 BDRVQcow2State *s = bs->opaque;
1297 unsigned int len, i;
1298 int ret = 0;
1299 QCowHeader header;
1300 Error *local_err = NULL;
1301 uint64_t ext_end;
1302 uint64_t l1_vm_state_index;
1303 bool update_header = false;
1305 ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
1306 if (ret < 0) {
1307 error_setg_errno(errp, -ret, "Could not read qcow2 header");
1308 goto fail;
1310 header.magic = be32_to_cpu(header.magic);
1311 header.version = be32_to_cpu(header.version);
1312 header.backing_file_offset = be64_to_cpu(header.backing_file_offset);
1313 header.backing_file_size = be32_to_cpu(header.backing_file_size);
1314 header.size = be64_to_cpu(header.size);
1315 header.cluster_bits = be32_to_cpu(header.cluster_bits);
1316 header.crypt_method = be32_to_cpu(header.crypt_method);
1317 header.l1_table_offset = be64_to_cpu(header.l1_table_offset);
1318 header.l1_size = be32_to_cpu(header.l1_size);
1319 header.refcount_table_offset = be64_to_cpu(header.refcount_table_offset);
1320 header.refcount_table_clusters =
1321 be32_to_cpu(header.refcount_table_clusters);
1322 header.snapshots_offset = be64_to_cpu(header.snapshots_offset);
1323 header.nb_snapshots = be32_to_cpu(header.nb_snapshots);
1325 if (header.magic != QCOW_MAGIC) {
1326 error_setg(errp, "Image is not in qcow2 format");
1327 ret = -EINVAL;
1328 goto fail;
1330 if (header.version < 2 || header.version > 3) {
1331 error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version);
1332 ret = -ENOTSUP;
1333 goto fail;
1336 s->qcow_version = header.version;
1338 /* Initialise cluster size */
1339 if (header.cluster_bits < MIN_CLUSTER_BITS ||
1340 header.cluster_bits > MAX_CLUSTER_BITS) {
1341 error_setg(errp, "Unsupported cluster size: 2^%" PRIu32,
1342 header.cluster_bits);
1343 ret = -EINVAL;
1344 goto fail;
1347 s->cluster_bits = header.cluster_bits;
1348 s->cluster_size = 1 << s->cluster_bits;
1350 /* Initialise version 3 header fields */
1351 if (header.version == 2) {
1352 header.incompatible_features = 0;
1353 header.compatible_features = 0;
1354 header.autoclear_features = 0;
1355 header.refcount_order = 4;
1356 header.header_length = 72;
1357 } else {
1358 header.incompatible_features =
1359 be64_to_cpu(header.incompatible_features);
1360 header.compatible_features = be64_to_cpu(header.compatible_features);
1361 header.autoclear_features = be64_to_cpu(header.autoclear_features);
1362 header.refcount_order = be32_to_cpu(header.refcount_order);
1363 header.header_length = be32_to_cpu(header.header_length);
1365 if (header.header_length < 104) {
1366 error_setg(errp, "qcow2 header too short");
1367 ret = -EINVAL;
1368 goto fail;
1372 if (header.header_length > s->cluster_size) {
1373 error_setg(errp, "qcow2 header exceeds cluster size");
1374 ret = -EINVAL;
1375 goto fail;
1378 if (header.header_length > sizeof(header)) {
1379 s->unknown_header_fields_size = header.header_length - sizeof(header);
1380 s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
1381 ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields,
1382 s->unknown_header_fields_size);
1383 if (ret < 0) {
1384 error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
1385 "fields");
1386 goto fail;
1390 if (header.backing_file_offset > s->cluster_size) {
1391 error_setg(errp, "Invalid backing file offset");
1392 ret = -EINVAL;
1393 goto fail;
1396 if (header.backing_file_offset) {
1397 ext_end = header.backing_file_offset;
1398 } else {
1399 ext_end = 1 << header.cluster_bits;
1402 /* Handle feature bits */
1403 s->incompatible_features = header.incompatible_features;
1404 s->compatible_features = header.compatible_features;
1405 s->autoclear_features = header.autoclear_features;
1408 * Handle compression type
1409 * Older qcow2 images don't contain the compression type header.
1410 * Distinguish them by the header length and use
1411 * the only valid (default) compression type in that case
1413 if (header.header_length > offsetof(QCowHeader, compression_type)) {
1414 s->compression_type = header.compression_type;
1415 } else {
1416 s->compression_type = QCOW2_COMPRESSION_TYPE_ZLIB;
1419 ret = validate_compression_type(s, errp);
1420 if (ret) {
1421 goto fail;
1424 if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
1425 void *feature_table = NULL;
1426 qcow2_read_extensions(bs, header.header_length, ext_end,
1427 &feature_table, flags, NULL, NULL);
1428 report_unsupported_feature(errp, feature_table,
1429 s->incompatible_features &
1430 ~QCOW2_INCOMPAT_MASK);
1431 ret = -ENOTSUP;
1432 g_free(feature_table);
1433 goto fail;
1436 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
1437 /* Corrupt images may not be written to unless they are being repaired
1439 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
1440 error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
1441 "read/write");
1442 ret = -EACCES;
1443 goto fail;
1447 /* Check support for various header values */
1448 if (header.refcount_order > 6) {
1449 error_setg(errp, "Reference count entry width too large; may not "
1450 "exceed 64 bits");
1451 ret = -EINVAL;
1452 goto fail;
1454 s->refcount_order = header.refcount_order;
1455 s->refcount_bits = 1 << s->refcount_order;
1456 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
1457 s->refcount_max += s->refcount_max - 1;
1459 s->crypt_method_header = header.crypt_method;
1460 if (s->crypt_method_header) {
1461 if (bdrv_uses_whitelist() &&
1462 s->crypt_method_header == QCOW_CRYPT_AES) {
1463 error_setg(errp,
1464 "Use of AES-CBC encrypted qcow2 images is no longer "
1465 "supported in system emulators");
1466 error_append_hint(errp,
1467 "You can use 'qemu-img convert' to convert your "
1468 "image to an alternative supported format, such "
1469 "as unencrypted qcow2, or raw with the LUKS "
1470 "format instead.\n");
1471 ret = -ENOSYS;
1472 goto fail;
1475 if (s->crypt_method_header == QCOW_CRYPT_AES) {
1476 s->crypt_physical_offset = false;
1477 } else {
1478 /* Assuming LUKS and any future crypt methods we
1479 * add will all use physical offsets, due to the
1480 * fact that the alternative is insecure... */
1481 s->crypt_physical_offset = true;
1484 bs->encrypted = true;
1487 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
1488 s->l2_size = 1 << s->l2_bits;
1489 /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1490 s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
1491 s->refcount_block_size = 1 << s->refcount_block_bits;
1492 bs->total_sectors = header.size / BDRV_SECTOR_SIZE;
1493 s->csize_shift = (62 - (s->cluster_bits - 8));
1494 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
1495 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
1497 s->refcount_table_offset = header.refcount_table_offset;
1498 s->refcount_table_size =
1499 header.refcount_table_clusters << (s->cluster_bits - 3);
1501 if (header.refcount_table_clusters == 0 && !(flags & BDRV_O_CHECK)) {
1502 error_setg(errp, "Image does not contain a reference count table");
1503 ret = -EINVAL;
1504 goto fail;
1507 ret = qcow2_validate_table(bs, s->refcount_table_offset,
1508 header.refcount_table_clusters,
1509 s->cluster_size, QCOW_MAX_REFTABLE_SIZE,
1510 "Reference count table", errp);
1511 if (ret < 0) {
1512 goto fail;
1515 if (!(flags & BDRV_O_CHECK)) {
1517 * The total size in bytes of the snapshot table is checked in
1518 * qcow2_read_snapshots() because the size of each snapshot is
1519 * variable and we don't know it yet.
1520 * Here we only check the offset and number of snapshots.
1522 ret = qcow2_validate_table(bs, header.snapshots_offset,
1523 header.nb_snapshots,
1524 sizeof(QCowSnapshotHeader),
1525 sizeof(QCowSnapshotHeader) *
1526 QCOW_MAX_SNAPSHOTS,
1527 "Snapshot table", errp);
1528 if (ret < 0) {
1529 goto fail;
1533 /* read the level 1 table */
1534 ret = qcow2_validate_table(bs, header.l1_table_offset,
1535 header.l1_size, sizeof(uint64_t),
1536 QCOW_MAX_L1_SIZE, "Active L1 table", errp);
1537 if (ret < 0) {
1538 goto fail;
1540 s->l1_size = header.l1_size;
1541 s->l1_table_offset = header.l1_table_offset;
1543 l1_vm_state_index = size_to_l1(s, header.size);
1544 if (l1_vm_state_index > INT_MAX) {
1545 error_setg(errp, "Image is too big");
1546 ret = -EFBIG;
1547 goto fail;
1549 s->l1_vm_state_index = l1_vm_state_index;
1551 /* the L1 table must contain at least enough entries to put
1552 header.size bytes */
1553 if (s->l1_size < s->l1_vm_state_index) {
1554 error_setg(errp, "L1 table is too small");
1555 ret = -EINVAL;
1556 goto fail;
1559 if (s->l1_size > 0) {
1560 s->l1_table = qemu_try_blockalign(bs->file->bs,
1561 s->l1_size * sizeof(uint64_t));
1562 if (s->l1_table == NULL) {
1563 error_setg(errp, "Could not allocate L1 table");
1564 ret = -ENOMEM;
1565 goto fail;
1567 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
1568 s->l1_size * sizeof(uint64_t));
1569 if (ret < 0) {
1570 error_setg_errno(errp, -ret, "Could not read L1 table");
1571 goto fail;
1573 for(i = 0;i < s->l1_size; i++) {
1574 s->l1_table[i] = be64_to_cpu(s->l1_table[i]);
1578 /* Parse driver-specific options */
1579 ret = qcow2_update_options(bs, options, flags, errp);
1580 if (ret < 0) {
1581 goto fail;
1584 s->flags = flags;
1586 ret = qcow2_refcount_init(bs);
1587 if (ret != 0) {
1588 error_setg_errno(errp, -ret, "Could not initialize refcount handling");
1589 goto fail;
1592 QLIST_INIT(&s->cluster_allocs);
1593 QTAILQ_INIT(&s->discards);
1595 /* read qcow2 extensions */
1596 if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
1597 flags, &update_header, errp)) {
1598 ret = -EINVAL;
1599 goto fail;
1602 /* Open external data file */
1603 s->data_file = bdrv_open_child(NULL, options, "data-file", bs,
1604 &child_of_bds, BDRV_CHILD_DATA,
1605 true, &local_err);
1606 if (local_err) {
1607 error_propagate(errp, local_err);
1608 ret = -EINVAL;
1609 goto fail;
1612 if (s->incompatible_features & QCOW2_INCOMPAT_DATA_FILE) {
1613 if (!s->data_file && s->image_data_file) {
1614 s->data_file = bdrv_open_child(s->image_data_file, options,
1615 "data-file", bs, &child_of_bds,
1616 BDRV_CHILD_DATA, false, errp);
1617 if (!s->data_file) {
1618 ret = -EINVAL;
1619 goto fail;
1622 if (!s->data_file) {
1623 error_setg(errp, "'data-file' is required for this image");
1624 ret = -EINVAL;
1625 goto fail;
1628 /* No data here */
1629 bs->file->role &= ~BDRV_CHILD_DATA;
1631 /* Must succeed because we have given up permissions if anything */
1632 bdrv_child_refresh_perms(bs, bs->file, &error_abort);
1633 } else {
1634 if (s->data_file) {
1635 error_setg(errp, "'data-file' can only be set for images with an "
1636 "external data file");
1637 ret = -EINVAL;
1638 goto fail;
1641 s->data_file = bs->file;
1643 if (data_file_is_raw(bs)) {
1644 error_setg(errp, "data-file-raw requires a data file");
1645 ret = -EINVAL;
1646 goto fail;
1650 /* qcow2_read_extension may have set up the crypto context
1651 * if the crypt method needs a header region, some methods
1652 * don't need header extensions, so must check here
1654 if (s->crypt_method_header && !s->crypto) {
1655 if (s->crypt_method_header == QCOW_CRYPT_AES) {
1656 unsigned int cflags = 0;
1657 if (flags & BDRV_O_NO_IO) {
1658 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
1660 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
1661 NULL, NULL, cflags,
1662 QCOW2_MAX_THREADS, errp);
1663 if (!s->crypto) {
1664 ret = -EINVAL;
1665 goto fail;
1667 } else if (!(flags & BDRV_O_NO_IO)) {
1668 error_setg(errp, "Missing CRYPTO header for crypt method %d",
1669 s->crypt_method_header);
1670 ret = -EINVAL;
1671 goto fail;
1675 /* read the backing file name */
1676 if (header.backing_file_offset != 0) {
1677 len = header.backing_file_size;
1678 if (len > MIN(1023, s->cluster_size - header.backing_file_offset) ||
1679 len >= sizeof(bs->backing_file)) {
1680 error_setg(errp, "Backing file name too long");
1681 ret = -EINVAL;
1682 goto fail;
1684 ret = bdrv_pread(bs->file, header.backing_file_offset,
1685 bs->auto_backing_file, len);
1686 if (ret < 0) {
1687 error_setg_errno(errp, -ret, "Could not read backing file name");
1688 goto fail;
1690 bs->auto_backing_file[len] = '\0';
1691 pstrcpy(bs->backing_file, sizeof(bs->backing_file),
1692 bs->auto_backing_file);
1693 s->image_backing_file = g_strdup(bs->auto_backing_file);
1697 * Internal snapshots; skip reading them in check mode, because
1698 * we do not need them then, and we do not want to abort because
1699 * of a broken table.
1701 if (!(flags & BDRV_O_CHECK)) {
1702 s->snapshots_offset = header.snapshots_offset;
1703 s->nb_snapshots = header.nb_snapshots;
1705 ret = qcow2_read_snapshots(bs, errp);
1706 if (ret < 0) {
1707 goto fail;
1711 /* Clear unknown autoclear feature bits */
1712 update_header |= s->autoclear_features & ~QCOW2_AUTOCLEAR_MASK;
1713 update_header =
1714 update_header && !bs->read_only && !(flags & BDRV_O_INACTIVE);
1715 if (update_header) {
1716 s->autoclear_features &= QCOW2_AUTOCLEAR_MASK;
1719 /* == Handle persistent dirty bitmaps ==
1721 * We want load dirty bitmaps in three cases:
1723 * 1. Normal open of the disk in active mode, not related to invalidation
1724 * after migration.
1726 * 2. Invalidation of the target vm after pre-copy phase of migration, if
1727 * bitmaps are _not_ migrating through migration channel, i.e.
1728 * 'dirty-bitmaps' capability is disabled.
1730 * 3. Invalidation of source vm after failed or canceled migration.
1731 * This is a very interesting case. There are two possible types of
1732 * bitmaps:
1734 * A. Stored on inactivation and removed. They should be loaded from the
1735 * image.
1737 * B. Not stored: not-persistent bitmaps and bitmaps, migrated through
1738 * the migration channel (with dirty-bitmaps capability).
1740 * On the other hand, there are two possible sub-cases:
1742 * 3.1 disk was changed by somebody else while were inactive. In this
1743 * case all in-RAM dirty bitmaps (both persistent and not) are
1744 * definitely invalid. And we don't have any method to determine
1745 * this.
1747 * Simple and safe thing is to just drop all the bitmaps of type B on
1748 * inactivation. But in this case we lose bitmaps in valid 4.2 case.
1750 * On the other hand, resuming source vm, if disk was already changed
1751 * is a bad thing anyway: not only bitmaps, the whole vm state is
1752 * out of sync with disk.
1754 * This means, that user or management tool, who for some reason
1755 * decided to resume source vm, after disk was already changed by
1756 * target vm, should at least drop all dirty bitmaps by hand.
1758 * So, we can ignore this case for now, but TODO: "generation"
1759 * extension for qcow2, to determine, that image was changed after
1760 * last inactivation. And if it is changed, we will drop (or at least
1761 * mark as 'invalid' all the bitmaps of type B, both persistent
1762 * and not).
1764 * 3.2 disk was _not_ changed while were inactive. Bitmaps may be saved
1765 * to disk ('dirty-bitmaps' capability disabled), or not saved
1766 * ('dirty-bitmaps' capability enabled), but we don't need to care
1767 * of: let's load bitmaps as always: stored bitmaps will be loaded,
1768 * and not stored has flag IN_USE=1 in the image and will be skipped
1769 * on loading.
1771 * One remaining possible case when we don't want load bitmaps:
1773 * 4. Open disk in inactive mode in target vm (bitmaps are migrating or
1774 * will be loaded on invalidation, no needs try loading them before)
1777 if (!(bdrv_get_flags(bs) & BDRV_O_INACTIVE)) {
1778 /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */
1779 bool header_updated = qcow2_load_dirty_bitmaps(bs, &local_err);
1780 if (local_err != NULL) {
1781 error_propagate(errp, local_err);
1782 ret = -EINVAL;
1783 goto fail;
1786 update_header = update_header && !header_updated;
1789 if (update_header) {
1790 ret = qcow2_update_header(bs);
1791 if (ret < 0) {
1792 error_setg_errno(errp, -ret, "Could not update qcow2 header");
1793 goto fail;
1797 bs->supported_zero_flags = header.version >= 3 ?
1798 BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK : 0;
1799 bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE;
1801 /* Repair image if dirty */
1802 if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only &&
1803 (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
1804 BdrvCheckResult result = {0};
1806 ret = qcow2_co_check_locked(bs, &result,
1807 BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
1808 if (ret < 0 || result.check_errors) {
1809 if (ret >= 0) {
1810 ret = -EIO;
1812 error_setg_errno(errp, -ret, "Could not repair dirty image");
1813 goto fail;
1817 #ifdef DEBUG_ALLOC
1819 BdrvCheckResult result = {0};
1820 qcow2_check_refcounts(bs, &result, 0);
1822 #endif
1824 qemu_co_queue_init(&s->thread_task_queue);
1826 return ret;
1828 fail:
1829 g_free(s->image_data_file);
1830 if (has_data_file(bs)) {
1831 bdrv_unref_child(bs, s->data_file);
1832 s->data_file = NULL;
1834 g_free(s->unknown_header_fields);
1835 cleanup_unknown_header_ext(bs);
1836 qcow2_free_snapshots(bs);
1837 qcow2_refcount_close(bs);
1838 qemu_vfree(s->l1_table);
1839 /* else pre-write overlap checks in cache_destroy may crash */
1840 s->l1_table = NULL;
1841 cache_clean_timer_del(bs);
1842 if (s->l2_table_cache) {
1843 qcow2_cache_destroy(s->l2_table_cache);
1845 if (s->refcount_block_cache) {
1846 qcow2_cache_destroy(s->refcount_block_cache);
1848 qcrypto_block_free(s->crypto);
1849 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1850 return ret;
1853 typedef struct QCow2OpenCo {
1854 BlockDriverState *bs;
1855 QDict *options;
1856 int flags;
1857 Error **errp;
1858 int ret;
1859 } QCow2OpenCo;
1861 static void coroutine_fn qcow2_open_entry(void *opaque)
1863 QCow2OpenCo *qoc = opaque;
1864 BDRVQcow2State *s = qoc->bs->opaque;
1866 qemu_co_mutex_lock(&s->lock);
1867 qoc->ret = qcow2_do_open(qoc->bs, qoc->options, qoc->flags, qoc->errp);
1868 qemu_co_mutex_unlock(&s->lock);
1871 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
1872 Error **errp)
1874 BDRVQcow2State *s = bs->opaque;
1875 QCow2OpenCo qoc = {
1876 .bs = bs,
1877 .options = options,
1878 .flags = flags,
1879 .errp = errp,
1880 .ret = -EINPROGRESS
1883 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds,
1884 BDRV_CHILD_IMAGE, false, errp);
1885 if (!bs->file) {
1886 return -EINVAL;
1889 /* Initialise locks */
1890 qemu_co_mutex_init(&s->lock);
1892 if (qemu_in_coroutine()) {
1893 /* From bdrv_co_create. */
1894 qcow2_open_entry(&qoc);
1895 } else {
1896 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
1897 qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry, &qoc));
1898 BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS);
1900 return qoc.ret;
1903 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1905 BDRVQcow2State *s = bs->opaque;
1907 if (bs->encrypted) {
1908 /* Encryption works on a sector granularity */
1909 bs->bl.request_alignment = qcrypto_block_get_sector_size(s->crypto);
1911 bs->bl.pwrite_zeroes_alignment = s->cluster_size;
1912 bs->bl.pdiscard_alignment = s->cluster_size;
1915 static int qcow2_reopen_prepare(BDRVReopenState *state,
1916 BlockReopenQueue *queue, Error **errp)
1918 Qcow2ReopenState *r;
1919 int ret;
1921 r = g_new0(Qcow2ReopenState, 1);
1922 state->opaque = r;
1924 ret = qcow2_update_options_prepare(state->bs, r, state->options,
1925 state->flags, errp);
1926 if (ret < 0) {
1927 goto fail;
1930 /* We need to write out any unwritten data if we reopen read-only. */
1931 if ((state->flags & BDRV_O_RDWR) == 0) {
1932 ret = qcow2_reopen_bitmaps_ro(state->bs, errp);
1933 if (ret < 0) {
1934 goto fail;
1937 ret = bdrv_flush(state->bs);
1938 if (ret < 0) {
1939 goto fail;
1942 ret = qcow2_mark_clean(state->bs);
1943 if (ret < 0) {
1944 goto fail;
1948 return 0;
1950 fail:
1951 qcow2_update_options_abort(state->bs, r);
1952 g_free(r);
1953 return ret;
1956 static void qcow2_reopen_commit(BDRVReopenState *state)
1958 qcow2_update_options_commit(state->bs, state->opaque);
1959 g_free(state->opaque);
1962 static void qcow2_reopen_commit_post(BDRVReopenState *state)
1964 if (state->flags & BDRV_O_RDWR) {
1965 Error *local_err = NULL;
1967 if (qcow2_reopen_bitmaps_rw(state->bs, &local_err) < 0) {
1969 * This is not fatal, bitmaps just left read-only, so all following
1970 * writes will fail. User can remove read-only bitmaps to unblock
1971 * writes or retry reopen.
1973 error_reportf_err(local_err,
1974 "%s: Failed to make dirty bitmaps writable: ",
1975 bdrv_get_node_name(state->bs));
1980 static void qcow2_reopen_abort(BDRVReopenState *state)
1982 qcow2_update_options_abort(state->bs, state->opaque);
1983 g_free(state->opaque);
1986 static void qcow2_join_options(QDict *options, QDict *old_options)
1988 bool has_new_overlap_template =
1989 qdict_haskey(options, QCOW2_OPT_OVERLAP) ||
1990 qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE);
1991 bool has_new_total_cache_size =
1992 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE);
1993 bool has_all_cache_options;
1995 /* New overlap template overrides all old overlap options */
1996 if (has_new_overlap_template) {
1997 qdict_del(old_options, QCOW2_OPT_OVERLAP);
1998 qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE);
1999 qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER);
2000 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1);
2001 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2);
2002 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE);
2003 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK);
2004 qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE);
2005 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1);
2006 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2);
2009 /* New total cache size overrides all old options */
2010 if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) {
2011 qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE);
2012 qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
2015 qdict_join(options, old_options, false);
2018 * If after merging all cache size options are set, an old total size is
2019 * overwritten. Do keep all options, however, if all three are new. The
2020 * resulting error message is what we want to happen.
2022 has_all_cache_options =
2023 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) ||
2024 qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) ||
2025 qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
2027 if (has_all_cache_options && !has_new_total_cache_size) {
2028 qdict_del(options, QCOW2_OPT_CACHE_SIZE);
2032 static int coroutine_fn qcow2_co_block_status(BlockDriverState *bs,
2033 bool want_zero,
2034 int64_t offset, int64_t count,
2035 int64_t *pnum, int64_t *map,
2036 BlockDriverState **file)
2038 BDRVQcow2State *s = bs->opaque;
2039 uint64_t cluster_offset;
2040 unsigned int bytes;
2041 int ret, status = 0;
2043 qemu_co_mutex_lock(&s->lock);
2045 if (!s->metadata_preallocation_checked) {
2046 ret = qcow2_detect_metadata_preallocation(bs);
2047 s->metadata_preallocation = (ret == 1);
2048 s->metadata_preallocation_checked = true;
2051 bytes = MIN(INT_MAX, count);
2052 ret = qcow2_get_cluster_offset(bs, offset, &bytes, &cluster_offset);
2053 qemu_co_mutex_unlock(&s->lock);
2054 if (ret < 0) {
2055 return ret;
2058 *pnum = bytes;
2060 if ((ret == QCOW2_CLUSTER_NORMAL || ret == QCOW2_CLUSTER_ZERO_ALLOC) &&
2061 !s->crypto) {
2062 *map = cluster_offset | offset_into_cluster(s, offset);
2063 *file = s->data_file->bs;
2064 status |= BDRV_BLOCK_OFFSET_VALID;
2066 if (ret == QCOW2_CLUSTER_ZERO_PLAIN || ret == QCOW2_CLUSTER_ZERO_ALLOC) {
2067 status |= BDRV_BLOCK_ZERO;
2068 } else if (ret != QCOW2_CLUSTER_UNALLOCATED) {
2069 status |= BDRV_BLOCK_DATA;
2071 if (s->metadata_preallocation && (status & BDRV_BLOCK_DATA) &&
2072 (status & BDRV_BLOCK_OFFSET_VALID))
2074 status |= BDRV_BLOCK_RECURSE;
2076 return status;
2079 static coroutine_fn int qcow2_handle_l2meta(BlockDriverState *bs,
2080 QCowL2Meta **pl2meta,
2081 bool link_l2)
2083 int ret = 0;
2084 QCowL2Meta *l2meta = *pl2meta;
2086 while (l2meta != NULL) {
2087 QCowL2Meta *next;
2089 if (link_l2) {
2090 ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
2091 if (ret) {
2092 goto out;
2094 } else {
2095 qcow2_alloc_cluster_abort(bs, l2meta);
2098 /* Take the request off the list of running requests */
2099 if (l2meta->nb_clusters != 0) {
2100 QLIST_REMOVE(l2meta, next_in_flight);
2103 qemu_co_queue_restart_all(&l2meta->dependent_requests);
2105 next = l2meta->next;
2106 g_free(l2meta);
2107 l2meta = next;
2109 out:
2110 *pl2meta = l2meta;
2111 return ret;
2114 static coroutine_fn int
2115 qcow2_co_preadv_encrypted(BlockDriverState *bs,
2116 uint64_t file_cluster_offset,
2117 uint64_t offset,
2118 uint64_t bytes,
2119 QEMUIOVector *qiov,
2120 uint64_t qiov_offset)
2122 int ret;
2123 BDRVQcow2State *s = bs->opaque;
2124 uint8_t *buf;
2126 assert(bs->encrypted && s->crypto);
2127 assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2130 * For encrypted images, read everything into a temporary
2131 * contiguous buffer on which the AES functions can work.
2132 * Also, decryption in a separate buffer is better as it
2133 * prevents the guest from learning information about the
2134 * encrypted nature of the virtual disk.
2137 buf = qemu_try_blockalign(s->data_file->bs, bytes);
2138 if (buf == NULL) {
2139 return -ENOMEM;
2142 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
2143 ret = bdrv_co_pread(s->data_file,
2144 file_cluster_offset + offset_into_cluster(s, offset),
2145 bytes, buf, 0);
2146 if (ret < 0) {
2147 goto fail;
2150 if (qcow2_co_decrypt(bs,
2151 file_cluster_offset + offset_into_cluster(s, offset),
2152 offset, buf, bytes) < 0)
2154 ret = -EIO;
2155 goto fail;
2157 qemu_iovec_from_buf(qiov, qiov_offset, buf, bytes);
2159 fail:
2160 qemu_vfree(buf);
2162 return ret;
2165 typedef struct Qcow2AioTask {
2166 AioTask task;
2168 BlockDriverState *bs;
2169 QCow2ClusterType cluster_type; /* only for read */
2170 uint64_t file_cluster_offset;
2171 uint64_t offset;
2172 uint64_t bytes;
2173 QEMUIOVector *qiov;
2174 uint64_t qiov_offset;
2175 QCowL2Meta *l2meta; /* only for write */
2176 } Qcow2AioTask;
2178 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task);
2179 static coroutine_fn int qcow2_add_task(BlockDriverState *bs,
2180 AioTaskPool *pool,
2181 AioTaskFunc func,
2182 QCow2ClusterType cluster_type,
2183 uint64_t file_cluster_offset,
2184 uint64_t offset,
2185 uint64_t bytes,
2186 QEMUIOVector *qiov,
2187 size_t qiov_offset,
2188 QCowL2Meta *l2meta)
2190 Qcow2AioTask local_task;
2191 Qcow2AioTask *task = pool ? g_new(Qcow2AioTask, 1) : &local_task;
2193 *task = (Qcow2AioTask) {
2194 .task.func = func,
2195 .bs = bs,
2196 .cluster_type = cluster_type,
2197 .qiov = qiov,
2198 .file_cluster_offset = file_cluster_offset,
2199 .offset = offset,
2200 .bytes = bytes,
2201 .qiov_offset = qiov_offset,
2202 .l2meta = l2meta,
2205 trace_qcow2_add_task(qemu_coroutine_self(), bs, pool,
2206 func == qcow2_co_preadv_task_entry ? "read" : "write",
2207 cluster_type, file_cluster_offset, offset, bytes,
2208 qiov, qiov_offset);
2210 if (!pool) {
2211 return func(&task->task);
2214 aio_task_pool_start_task(pool, &task->task);
2216 return 0;
2219 static coroutine_fn int qcow2_co_preadv_task(BlockDriverState *bs,
2220 QCow2ClusterType cluster_type,
2221 uint64_t file_cluster_offset,
2222 uint64_t offset, uint64_t bytes,
2223 QEMUIOVector *qiov,
2224 size_t qiov_offset)
2226 BDRVQcow2State *s = bs->opaque;
2227 int offset_in_cluster = offset_into_cluster(s, offset);
2229 switch (cluster_type) {
2230 case QCOW2_CLUSTER_ZERO_PLAIN:
2231 case QCOW2_CLUSTER_ZERO_ALLOC:
2232 /* Both zero types are handled in qcow2_co_preadv_part */
2233 g_assert_not_reached();
2235 case QCOW2_CLUSTER_UNALLOCATED:
2236 assert(bs->backing); /* otherwise handled in qcow2_co_preadv_part */
2238 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
2239 return bdrv_co_preadv_part(bs->backing, offset, bytes,
2240 qiov, qiov_offset, 0);
2242 case QCOW2_CLUSTER_COMPRESSED:
2243 return qcow2_co_preadv_compressed(bs, file_cluster_offset,
2244 offset, bytes, qiov, qiov_offset);
2246 case QCOW2_CLUSTER_NORMAL:
2247 assert(offset_into_cluster(s, file_cluster_offset) == 0);
2248 if (bs->encrypted) {
2249 return qcow2_co_preadv_encrypted(bs, file_cluster_offset,
2250 offset, bytes, qiov, qiov_offset);
2253 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
2254 return bdrv_co_preadv_part(s->data_file,
2255 file_cluster_offset + offset_in_cluster,
2256 bytes, qiov, qiov_offset, 0);
2258 default:
2259 g_assert_not_reached();
2262 g_assert_not_reached();
2265 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task)
2267 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
2269 assert(!t->l2meta);
2271 return qcow2_co_preadv_task(t->bs, t->cluster_type, t->file_cluster_offset,
2272 t->offset, t->bytes, t->qiov, t->qiov_offset);
2275 static coroutine_fn int qcow2_co_preadv_part(BlockDriverState *bs,
2276 uint64_t offset, uint64_t bytes,
2277 QEMUIOVector *qiov,
2278 size_t qiov_offset, int flags)
2280 BDRVQcow2State *s = bs->opaque;
2281 int ret = 0;
2282 unsigned int cur_bytes; /* number of bytes in current iteration */
2283 uint64_t cluster_offset = 0;
2284 AioTaskPool *aio = NULL;
2286 while (bytes != 0 && aio_task_pool_status(aio) == 0) {
2287 /* prepare next request */
2288 cur_bytes = MIN(bytes, INT_MAX);
2289 if (s->crypto) {
2290 cur_bytes = MIN(cur_bytes,
2291 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2294 qemu_co_mutex_lock(&s->lock);
2295 ret = qcow2_get_cluster_offset(bs, offset, &cur_bytes, &cluster_offset);
2296 qemu_co_mutex_unlock(&s->lock);
2297 if (ret < 0) {
2298 goto out;
2301 if (ret == QCOW2_CLUSTER_ZERO_PLAIN ||
2302 ret == QCOW2_CLUSTER_ZERO_ALLOC ||
2303 (ret == QCOW2_CLUSTER_UNALLOCATED && !bs->backing))
2305 qemu_iovec_memset(qiov, qiov_offset, 0, cur_bytes);
2306 } else {
2307 if (!aio && cur_bytes != bytes) {
2308 aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
2310 ret = qcow2_add_task(bs, aio, qcow2_co_preadv_task_entry, ret,
2311 cluster_offset, offset, cur_bytes,
2312 qiov, qiov_offset, NULL);
2313 if (ret < 0) {
2314 goto out;
2318 bytes -= cur_bytes;
2319 offset += cur_bytes;
2320 qiov_offset += cur_bytes;
2323 out:
2324 if (aio) {
2325 aio_task_pool_wait_all(aio);
2326 if (ret == 0) {
2327 ret = aio_task_pool_status(aio);
2329 g_free(aio);
2332 return ret;
2335 /* Check if it's possible to merge a write request with the writing of
2336 * the data from the COW regions */
2337 static bool merge_cow(uint64_t offset, unsigned bytes,
2338 QEMUIOVector *qiov, size_t qiov_offset,
2339 QCowL2Meta *l2meta)
2341 QCowL2Meta *m;
2343 for (m = l2meta; m != NULL; m = m->next) {
2344 /* If both COW regions are empty then there's nothing to merge */
2345 if (m->cow_start.nb_bytes == 0 && m->cow_end.nb_bytes == 0) {
2346 continue;
2349 /* If COW regions are handled already, skip this too */
2350 if (m->skip_cow) {
2351 continue;
2354 /* The data (middle) region must be immediately after the
2355 * start region */
2356 if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) {
2357 continue;
2360 /* The end region must be immediately after the data (middle)
2361 * region */
2362 if (m->offset + m->cow_end.offset != offset + bytes) {
2363 continue;
2366 /* Make sure that adding both COW regions to the QEMUIOVector
2367 * does not exceed IOV_MAX */
2368 if (qemu_iovec_subvec_niov(qiov, qiov_offset, bytes) > IOV_MAX - 2) {
2369 continue;
2372 m->data_qiov = qiov;
2373 m->data_qiov_offset = qiov_offset;
2374 return true;
2377 return false;
2380 static bool is_unallocated(BlockDriverState *bs, int64_t offset, int64_t bytes)
2382 int64_t nr;
2383 return !bytes ||
2384 (!bdrv_is_allocated_above(bs, NULL, false, offset, bytes, &nr) &&
2385 nr == bytes);
2388 static bool is_zero_cow(BlockDriverState *bs, QCowL2Meta *m)
2391 * This check is designed for optimization shortcut so it must be
2392 * efficient.
2393 * Instead of is_zero(), use is_unallocated() as it is faster (but not
2394 * as accurate and can result in false negatives).
2396 return is_unallocated(bs, m->offset + m->cow_start.offset,
2397 m->cow_start.nb_bytes) &&
2398 is_unallocated(bs, m->offset + m->cow_end.offset,
2399 m->cow_end.nb_bytes);
2402 static int handle_alloc_space(BlockDriverState *bs, QCowL2Meta *l2meta)
2404 BDRVQcow2State *s = bs->opaque;
2405 QCowL2Meta *m;
2407 if (!(s->data_file->bs->supported_zero_flags & BDRV_REQ_NO_FALLBACK)) {
2408 return 0;
2411 if (bs->encrypted) {
2412 return 0;
2415 for (m = l2meta; m != NULL; m = m->next) {
2416 int ret;
2418 if (!m->cow_start.nb_bytes && !m->cow_end.nb_bytes) {
2419 continue;
2422 if (!is_zero_cow(bs, m)) {
2423 continue;
2427 * instead of writing zero COW buffers,
2428 * efficiently zero out the whole clusters
2431 ret = qcow2_pre_write_overlap_check(bs, 0, m->alloc_offset,
2432 m->nb_clusters * s->cluster_size,
2433 true);
2434 if (ret < 0) {
2435 return ret;
2438 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_SPACE);
2439 ret = bdrv_co_pwrite_zeroes(s->data_file, m->alloc_offset,
2440 m->nb_clusters * s->cluster_size,
2441 BDRV_REQ_NO_FALLBACK);
2442 if (ret < 0) {
2443 if (ret != -ENOTSUP && ret != -EAGAIN) {
2444 return ret;
2446 continue;
2449 trace_qcow2_skip_cow(qemu_coroutine_self(), m->offset, m->nb_clusters);
2450 m->skip_cow = true;
2452 return 0;
2456 * qcow2_co_pwritev_task
2457 * Called with s->lock unlocked
2458 * l2meta - if not NULL, qcow2_co_pwritev_task() will consume it. Caller must
2459 * not use it somehow after qcow2_co_pwritev_task() call
2461 static coroutine_fn int qcow2_co_pwritev_task(BlockDriverState *bs,
2462 uint64_t file_cluster_offset,
2463 uint64_t offset, uint64_t bytes,
2464 QEMUIOVector *qiov,
2465 uint64_t qiov_offset,
2466 QCowL2Meta *l2meta)
2468 int ret;
2469 BDRVQcow2State *s = bs->opaque;
2470 void *crypt_buf = NULL;
2471 int offset_in_cluster = offset_into_cluster(s, offset);
2472 QEMUIOVector encrypted_qiov;
2474 if (bs->encrypted) {
2475 assert(s->crypto);
2476 assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2477 crypt_buf = qemu_try_blockalign(bs->file->bs, bytes);
2478 if (crypt_buf == NULL) {
2479 ret = -ENOMEM;
2480 goto out_unlocked;
2482 qemu_iovec_to_buf(qiov, qiov_offset, crypt_buf, bytes);
2484 if (qcow2_co_encrypt(bs, file_cluster_offset + offset_in_cluster,
2485 offset, crypt_buf, bytes) < 0)
2487 ret = -EIO;
2488 goto out_unlocked;
2491 qemu_iovec_init_buf(&encrypted_qiov, crypt_buf, bytes);
2492 qiov = &encrypted_qiov;
2493 qiov_offset = 0;
2496 /* Try to efficiently initialize the physical space with zeroes */
2497 ret = handle_alloc_space(bs, l2meta);
2498 if (ret < 0) {
2499 goto out_unlocked;
2503 * If we need to do COW, check if it's possible to merge the
2504 * writing of the guest data together with that of the COW regions.
2505 * If it's not possible (or not necessary) then write the
2506 * guest data now.
2508 if (!merge_cow(offset, bytes, qiov, qiov_offset, l2meta)) {
2509 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
2510 trace_qcow2_writev_data(qemu_coroutine_self(),
2511 file_cluster_offset + offset_in_cluster);
2512 ret = bdrv_co_pwritev_part(s->data_file,
2513 file_cluster_offset + offset_in_cluster,
2514 bytes, qiov, qiov_offset, 0);
2515 if (ret < 0) {
2516 goto out_unlocked;
2520 qemu_co_mutex_lock(&s->lock);
2522 ret = qcow2_handle_l2meta(bs, &l2meta, true);
2523 goto out_locked;
2525 out_unlocked:
2526 qemu_co_mutex_lock(&s->lock);
2528 out_locked:
2529 qcow2_handle_l2meta(bs, &l2meta, false);
2530 qemu_co_mutex_unlock(&s->lock);
2532 qemu_vfree(crypt_buf);
2534 return ret;
2537 static coroutine_fn int qcow2_co_pwritev_task_entry(AioTask *task)
2539 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
2541 assert(!t->cluster_type);
2543 return qcow2_co_pwritev_task(t->bs, t->file_cluster_offset,
2544 t->offset, t->bytes, t->qiov, t->qiov_offset,
2545 t->l2meta);
2548 static coroutine_fn int qcow2_co_pwritev_part(
2549 BlockDriverState *bs, uint64_t offset, uint64_t bytes,
2550 QEMUIOVector *qiov, size_t qiov_offset, int flags)
2552 BDRVQcow2State *s = bs->opaque;
2553 int offset_in_cluster;
2554 int ret;
2555 unsigned int cur_bytes; /* number of sectors in current iteration */
2556 uint64_t cluster_offset;
2557 QCowL2Meta *l2meta = NULL;
2558 AioTaskPool *aio = NULL;
2560 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes);
2562 while (bytes != 0 && aio_task_pool_status(aio) == 0) {
2564 l2meta = NULL;
2566 trace_qcow2_writev_start_part(qemu_coroutine_self());
2567 offset_in_cluster = offset_into_cluster(s, offset);
2568 cur_bytes = MIN(bytes, INT_MAX);
2569 if (bs->encrypted) {
2570 cur_bytes = MIN(cur_bytes,
2571 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
2572 - offset_in_cluster);
2575 qemu_co_mutex_lock(&s->lock);
2577 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
2578 &cluster_offset, &l2meta);
2579 if (ret < 0) {
2580 goto out_locked;
2583 assert(offset_into_cluster(s, cluster_offset) == 0);
2585 ret = qcow2_pre_write_overlap_check(bs, 0,
2586 cluster_offset + offset_in_cluster,
2587 cur_bytes, true);
2588 if (ret < 0) {
2589 goto out_locked;
2592 qemu_co_mutex_unlock(&s->lock);
2594 if (!aio && cur_bytes != bytes) {
2595 aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
2597 ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_task_entry, 0,
2598 cluster_offset, offset, cur_bytes,
2599 qiov, qiov_offset, l2meta);
2600 l2meta = NULL; /* l2meta is consumed by qcow2_co_pwritev_task() */
2601 if (ret < 0) {
2602 goto fail_nometa;
2605 bytes -= cur_bytes;
2606 offset += cur_bytes;
2607 qiov_offset += cur_bytes;
2608 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes);
2610 ret = 0;
2612 qemu_co_mutex_lock(&s->lock);
2614 out_locked:
2615 qcow2_handle_l2meta(bs, &l2meta, false);
2617 qemu_co_mutex_unlock(&s->lock);
2619 fail_nometa:
2620 if (aio) {
2621 aio_task_pool_wait_all(aio);
2622 if (ret == 0) {
2623 ret = aio_task_pool_status(aio);
2625 g_free(aio);
2628 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
2630 return ret;
2633 static int qcow2_inactivate(BlockDriverState *bs)
2635 BDRVQcow2State *s = bs->opaque;
2636 int ret, result = 0;
2637 Error *local_err = NULL;
2639 qcow2_store_persistent_dirty_bitmaps(bs, true, &local_err);
2640 if (local_err != NULL) {
2641 result = -EINVAL;
2642 error_reportf_err(local_err, "Lost persistent bitmaps during "
2643 "inactivation of node '%s': ",
2644 bdrv_get_device_or_node_name(bs));
2647 ret = qcow2_cache_flush(bs, s->l2_table_cache);
2648 if (ret) {
2649 result = ret;
2650 error_report("Failed to flush the L2 table cache: %s",
2651 strerror(-ret));
2654 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2655 if (ret) {
2656 result = ret;
2657 error_report("Failed to flush the refcount block cache: %s",
2658 strerror(-ret));
2661 if (result == 0) {
2662 qcow2_mark_clean(bs);
2665 return result;
2668 static void qcow2_close(BlockDriverState *bs)
2670 BDRVQcow2State *s = bs->opaque;
2671 qemu_vfree(s->l1_table);
2672 /* else pre-write overlap checks in cache_destroy may crash */
2673 s->l1_table = NULL;
2675 if (!(s->flags & BDRV_O_INACTIVE)) {
2676 qcow2_inactivate(bs);
2679 cache_clean_timer_del(bs);
2680 qcow2_cache_destroy(s->l2_table_cache);
2681 qcow2_cache_destroy(s->refcount_block_cache);
2683 qcrypto_block_free(s->crypto);
2684 s->crypto = NULL;
2685 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
2687 g_free(s->unknown_header_fields);
2688 cleanup_unknown_header_ext(bs);
2690 g_free(s->image_data_file);
2691 g_free(s->image_backing_file);
2692 g_free(s->image_backing_format);
2694 if (has_data_file(bs)) {
2695 bdrv_unref_child(bs, s->data_file);
2696 s->data_file = NULL;
2699 qcow2_refcount_close(bs);
2700 qcow2_free_snapshots(bs);
2703 static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs,
2704 Error **errp)
2706 BDRVQcow2State *s = bs->opaque;
2707 int flags = s->flags;
2708 QCryptoBlock *crypto = NULL;
2709 QDict *options;
2710 Error *local_err = NULL;
2711 int ret;
2714 * Backing files are read-only which makes all of their metadata immutable,
2715 * that means we don't have to worry about reopening them here.
2718 crypto = s->crypto;
2719 s->crypto = NULL;
2721 qcow2_close(bs);
2723 memset(s, 0, sizeof(BDRVQcow2State));
2724 options = qdict_clone_shallow(bs->options);
2726 flags &= ~BDRV_O_INACTIVE;
2727 qemu_co_mutex_lock(&s->lock);
2728 ret = qcow2_do_open(bs, options, flags, &local_err);
2729 qemu_co_mutex_unlock(&s->lock);
2730 qobject_unref(options);
2731 if (local_err) {
2732 error_propagate_prepend(errp, local_err,
2733 "Could not reopen qcow2 layer: ");
2734 bs->drv = NULL;
2735 return;
2736 } else if (ret < 0) {
2737 error_setg_errno(errp, -ret, "Could not reopen qcow2 layer");
2738 bs->drv = NULL;
2739 return;
2742 s->crypto = crypto;
2745 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
2746 size_t len, size_t buflen)
2748 QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
2749 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
2751 if (buflen < ext_len) {
2752 return -ENOSPC;
2755 *ext_backing_fmt = (QCowExtension) {
2756 .magic = cpu_to_be32(magic),
2757 .len = cpu_to_be32(len),
2760 if (len) {
2761 memcpy(buf + sizeof(QCowExtension), s, len);
2764 return ext_len;
2768 * Updates the qcow2 header, including the variable length parts of it, i.e.
2769 * the backing file name and all extensions. qcow2 was not designed to allow
2770 * such changes, so if we run out of space (we can only use the first cluster)
2771 * this function may fail.
2773 * Returns 0 on success, -errno in error cases.
2775 int qcow2_update_header(BlockDriverState *bs)
2777 BDRVQcow2State *s = bs->opaque;
2778 QCowHeader *header;
2779 char *buf;
2780 size_t buflen = s->cluster_size;
2781 int ret;
2782 uint64_t total_size;
2783 uint32_t refcount_table_clusters;
2784 size_t header_length;
2785 Qcow2UnknownHeaderExtension *uext;
2787 buf = qemu_blockalign(bs, buflen);
2789 /* Header structure */
2790 header = (QCowHeader*) buf;
2792 if (buflen < sizeof(*header)) {
2793 ret = -ENOSPC;
2794 goto fail;
2797 header_length = sizeof(*header) + s->unknown_header_fields_size;
2798 total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
2799 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2801 ret = validate_compression_type(s, NULL);
2802 if (ret) {
2803 goto fail;
2806 *header = (QCowHeader) {
2807 /* Version 2 fields */
2808 .magic = cpu_to_be32(QCOW_MAGIC),
2809 .version = cpu_to_be32(s->qcow_version),
2810 .backing_file_offset = 0,
2811 .backing_file_size = 0,
2812 .cluster_bits = cpu_to_be32(s->cluster_bits),
2813 .size = cpu_to_be64(total_size),
2814 .crypt_method = cpu_to_be32(s->crypt_method_header),
2815 .l1_size = cpu_to_be32(s->l1_size),
2816 .l1_table_offset = cpu_to_be64(s->l1_table_offset),
2817 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset),
2818 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
2819 .nb_snapshots = cpu_to_be32(s->nb_snapshots),
2820 .snapshots_offset = cpu_to_be64(s->snapshots_offset),
2822 /* Version 3 fields */
2823 .incompatible_features = cpu_to_be64(s->incompatible_features),
2824 .compatible_features = cpu_to_be64(s->compatible_features),
2825 .autoclear_features = cpu_to_be64(s->autoclear_features),
2826 .refcount_order = cpu_to_be32(s->refcount_order),
2827 .header_length = cpu_to_be32(header_length),
2828 .compression_type = s->compression_type,
2831 /* For older versions, write a shorter header */
2832 switch (s->qcow_version) {
2833 case 2:
2834 ret = offsetof(QCowHeader, incompatible_features);
2835 break;
2836 case 3:
2837 ret = sizeof(*header);
2838 break;
2839 default:
2840 ret = -EINVAL;
2841 goto fail;
2844 buf += ret;
2845 buflen -= ret;
2846 memset(buf, 0, buflen);
2848 /* Preserve any unknown field in the header */
2849 if (s->unknown_header_fields_size) {
2850 if (buflen < s->unknown_header_fields_size) {
2851 ret = -ENOSPC;
2852 goto fail;
2855 memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
2856 buf += s->unknown_header_fields_size;
2857 buflen -= s->unknown_header_fields_size;
2860 /* Backing file format header extension */
2861 if (s->image_backing_format) {
2862 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
2863 s->image_backing_format,
2864 strlen(s->image_backing_format),
2865 buflen);
2866 if (ret < 0) {
2867 goto fail;
2870 buf += ret;
2871 buflen -= ret;
2874 /* External data file header extension */
2875 if (has_data_file(bs) && s->image_data_file) {
2876 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_DATA_FILE,
2877 s->image_data_file, strlen(s->image_data_file),
2878 buflen);
2879 if (ret < 0) {
2880 goto fail;
2883 buf += ret;
2884 buflen -= ret;
2887 /* Full disk encryption header pointer extension */
2888 if (s->crypto_header.offset != 0) {
2889 s->crypto_header.offset = cpu_to_be64(s->crypto_header.offset);
2890 s->crypto_header.length = cpu_to_be64(s->crypto_header.length);
2891 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_CRYPTO_HEADER,
2892 &s->crypto_header, sizeof(s->crypto_header),
2893 buflen);
2894 s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset);
2895 s->crypto_header.length = be64_to_cpu(s->crypto_header.length);
2896 if (ret < 0) {
2897 goto fail;
2899 buf += ret;
2900 buflen -= ret;
2904 * Feature table. A mere 8 feature names occupies 392 bytes, and
2905 * when coupled with the v3 minimum header of 104 bytes plus the
2906 * 8-byte end-of-extension marker, that would leave only 8 bytes
2907 * for a backing file name in an image with 512-byte clusters.
2908 * Thus, we choose to omit this header for cluster sizes 4k and
2909 * smaller.
2911 if (s->qcow_version >= 3 && s->cluster_size > 4096) {
2912 static const Qcow2Feature features[] = {
2914 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2915 .bit = QCOW2_INCOMPAT_DIRTY_BITNR,
2916 .name = "dirty bit",
2919 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2920 .bit = QCOW2_INCOMPAT_CORRUPT_BITNR,
2921 .name = "corrupt bit",
2924 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2925 .bit = QCOW2_INCOMPAT_DATA_FILE_BITNR,
2926 .name = "external data file",
2929 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2930 .bit = QCOW2_INCOMPAT_COMPRESSION_BITNR,
2931 .name = "compression type",
2934 .type = QCOW2_FEAT_TYPE_COMPATIBLE,
2935 .bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
2936 .name = "lazy refcounts",
2939 .type = QCOW2_FEAT_TYPE_AUTOCLEAR,
2940 .bit = QCOW2_AUTOCLEAR_BITMAPS_BITNR,
2941 .name = "bitmaps",
2944 .type = QCOW2_FEAT_TYPE_AUTOCLEAR,
2945 .bit = QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR,
2946 .name = "raw external data",
2950 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
2951 features, sizeof(features), buflen);
2952 if (ret < 0) {
2953 goto fail;
2955 buf += ret;
2956 buflen -= ret;
2959 /* Bitmap extension */
2960 if (s->nb_bitmaps > 0) {
2961 Qcow2BitmapHeaderExt bitmaps_header = {
2962 .nb_bitmaps = cpu_to_be32(s->nb_bitmaps),
2963 .bitmap_directory_size =
2964 cpu_to_be64(s->bitmap_directory_size),
2965 .bitmap_directory_offset =
2966 cpu_to_be64(s->bitmap_directory_offset)
2968 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BITMAPS,
2969 &bitmaps_header, sizeof(bitmaps_header),
2970 buflen);
2971 if (ret < 0) {
2972 goto fail;
2974 buf += ret;
2975 buflen -= ret;
2978 /* Keep unknown header extensions */
2979 QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
2980 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
2981 if (ret < 0) {
2982 goto fail;
2985 buf += ret;
2986 buflen -= ret;
2989 /* End of header extensions */
2990 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
2991 if (ret < 0) {
2992 goto fail;
2995 buf += ret;
2996 buflen -= ret;
2998 /* Backing file name */
2999 if (s->image_backing_file) {
3000 size_t backing_file_len = strlen(s->image_backing_file);
3002 if (buflen < backing_file_len) {
3003 ret = -ENOSPC;
3004 goto fail;
3007 /* Using strncpy is ok here, since buf is not NUL-terminated. */
3008 strncpy(buf, s->image_backing_file, buflen);
3010 header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
3011 header->backing_file_size = cpu_to_be32(backing_file_len);
3014 /* Write the new header */
3015 ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
3016 if (ret < 0) {
3017 goto fail;
3020 ret = 0;
3021 fail:
3022 qemu_vfree(header);
3023 return ret;
3026 static int qcow2_change_backing_file(BlockDriverState *bs,
3027 const char *backing_file, const char *backing_fmt)
3029 BDRVQcow2State *s = bs->opaque;
3031 /* Adding a backing file means that the external data file alone won't be
3032 * enough to make sense of the content */
3033 if (backing_file && data_file_is_raw(bs)) {
3034 return -EINVAL;
3037 if (backing_file && strlen(backing_file) > 1023) {
3038 return -EINVAL;
3041 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3042 backing_file ?: "");
3043 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
3044 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
3046 g_free(s->image_backing_file);
3047 g_free(s->image_backing_format);
3049 s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
3050 s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
3052 return qcow2_update_header(bs);
3055 static int qcow2_set_up_encryption(BlockDriverState *bs,
3056 QCryptoBlockCreateOptions *cryptoopts,
3057 Error **errp)
3059 BDRVQcow2State *s = bs->opaque;
3060 QCryptoBlock *crypto = NULL;
3061 int fmt, ret;
3063 switch (cryptoopts->format) {
3064 case Q_CRYPTO_BLOCK_FORMAT_LUKS:
3065 fmt = QCOW_CRYPT_LUKS;
3066 break;
3067 case Q_CRYPTO_BLOCK_FORMAT_QCOW:
3068 fmt = QCOW_CRYPT_AES;
3069 break;
3070 default:
3071 error_setg(errp, "Crypto format not supported in qcow2");
3072 return -EINVAL;
3075 s->crypt_method_header = fmt;
3077 crypto = qcrypto_block_create(cryptoopts, "encrypt.",
3078 qcow2_crypto_hdr_init_func,
3079 qcow2_crypto_hdr_write_func,
3080 bs, errp);
3081 if (!crypto) {
3082 return -EINVAL;
3085 ret = qcow2_update_header(bs);
3086 if (ret < 0) {
3087 error_setg_errno(errp, -ret, "Could not write encryption header");
3088 goto out;
3091 ret = 0;
3092 out:
3093 qcrypto_block_free(crypto);
3094 return ret;
3098 * Preallocates metadata structures for data clusters between @offset (in the
3099 * guest disk) and @new_length (which is thus generally the new guest disk
3100 * size).
3102 * Returns: 0 on success, -errno on failure.
3104 static int coroutine_fn preallocate_co(BlockDriverState *bs, uint64_t offset,
3105 uint64_t new_length, PreallocMode mode,
3106 Error **errp)
3108 BDRVQcow2State *s = bs->opaque;
3109 uint64_t bytes;
3110 uint64_t host_offset = 0;
3111 int64_t file_length;
3112 unsigned int cur_bytes;
3113 int ret;
3114 QCowL2Meta *meta;
3116 assert(offset <= new_length);
3117 bytes = new_length - offset;
3119 while (bytes) {
3120 cur_bytes = MIN(bytes, QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size));
3121 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
3122 &host_offset, &meta);
3123 if (ret < 0) {
3124 error_setg_errno(errp, -ret, "Allocating clusters failed");
3125 return ret;
3128 while (meta) {
3129 QCowL2Meta *next = meta->next;
3131 ret = qcow2_alloc_cluster_link_l2(bs, meta);
3132 if (ret < 0) {
3133 error_setg_errno(errp, -ret, "Mapping clusters failed");
3134 qcow2_free_any_clusters(bs, meta->alloc_offset,
3135 meta->nb_clusters, QCOW2_DISCARD_NEVER);
3136 return ret;
3139 /* There are no dependent requests, but we need to remove our
3140 * request from the list of in-flight requests */
3141 QLIST_REMOVE(meta, next_in_flight);
3143 g_free(meta);
3144 meta = next;
3147 /* TODO Preallocate data if requested */
3149 bytes -= cur_bytes;
3150 offset += cur_bytes;
3154 * It is expected that the image file is large enough to actually contain
3155 * all of the allocated clusters (otherwise we get failing reads after
3156 * EOF). Extend the image to the last allocated sector.
3158 file_length = bdrv_getlength(s->data_file->bs);
3159 if (file_length < 0) {
3160 error_setg_errno(errp, -file_length, "Could not get file size");
3161 return file_length;
3164 if (host_offset + cur_bytes > file_length) {
3165 if (mode == PREALLOC_MODE_METADATA) {
3166 mode = PREALLOC_MODE_OFF;
3168 ret = bdrv_co_truncate(s->data_file, host_offset + cur_bytes, false,
3169 mode, 0, errp);
3170 if (ret < 0) {
3171 return ret;
3175 return 0;
3178 /* qcow2_refcount_metadata_size:
3179 * @clusters: number of clusters to refcount (including data and L1/L2 tables)
3180 * @cluster_size: size of a cluster, in bytes
3181 * @refcount_order: refcount bits power-of-2 exponent
3182 * @generous_increase: allow for the refcount table to be 1.5x as large as it
3183 * needs to be
3185 * Returns: Number of bytes required for refcount blocks and table metadata.
3187 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size,
3188 int refcount_order, bool generous_increase,
3189 uint64_t *refblock_count)
3192 * Every host cluster is reference-counted, including metadata (even
3193 * refcount metadata is recursively included).
3195 * An accurate formula for the size of refcount metadata size is difficult
3196 * to derive. An easier method of calculation is finding the fixed point
3197 * where no further refcount blocks or table clusters are required to
3198 * reference count every cluster.
3200 int64_t blocks_per_table_cluster = cluster_size / sizeof(uint64_t);
3201 int64_t refcounts_per_block = cluster_size * 8 / (1 << refcount_order);
3202 int64_t table = 0; /* number of refcount table clusters */
3203 int64_t blocks = 0; /* number of refcount block clusters */
3204 int64_t last;
3205 int64_t n = 0;
3207 do {
3208 last = n;
3209 blocks = DIV_ROUND_UP(clusters + table + blocks, refcounts_per_block);
3210 table = DIV_ROUND_UP(blocks, blocks_per_table_cluster);
3211 n = clusters + blocks + table;
3213 if (n == last && generous_increase) {
3214 clusters += DIV_ROUND_UP(table, 2);
3215 n = 0; /* force another loop */
3216 generous_increase = false;
3218 } while (n != last);
3220 if (refblock_count) {
3221 *refblock_count = blocks;
3224 return (blocks + table) * cluster_size;
3228 * qcow2_calc_prealloc_size:
3229 * @total_size: virtual disk size in bytes
3230 * @cluster_size: cluster size in bytes
3231 * @refcount_order: refcount bits power-of-2 exponent
3233 * Returns: Total number of bytes required for the fully allocated image
3234 * (including metadata).
3236 static int64_t qcow2_calc_prealloc_size(int64_t total_size,
3237 size_t cluster_size,
3238 int refcount_order)
3240 int64_t meta_size = 0;
3241 uint64_t nl1e, nl2e;
3242 int64_t aligned_total_size = ROUND_UP(total_size, cluster_size);
3244 /* header: 1 cluster */
3245 meta_size += cluster_size;
3247 /* total size of L2 tables */
3248 nl2e = aligned_total_size / cluster_size;
3249 nl2e = ROUND_UP(nl2e, cluster_size / sizeof(uint64_t));
3250 meta_size += nl2e * sizeof(uint64_t);
3252 /* total size of L1 tables */
3253 nl1e = nl2e * sizeof(uint64_t) / cluster_size;
3254 nl1e = ROUND_UP(nl1e, cluster_size / sizeof(uint64_t));
3255 meta_size += nl1e * sizeof(uint64_t);
3257 /* total size of refcount table and blocks */
3258 meta_size += qcow2_refcount_metadata_size(
3259 (meta_size + aligned_total_size) / cluster_size,
3260 cluster_size, refcount_order, false, NULL);
3262 return meta_size + aligned_total_size;
3265 static bool validate_cluster_size(size_t cluster_size, Error **errp)
3267 int cluster_bits = ctz32(cluster_size);
3268 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
3269 (1 << cluster_bits) != cluster_size)
3271 error_setg(errp, "Cluster size must be a power of two between %d and "
3272 "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
3273 return false;
3275 return true;
3278 static size_t qcow2_opt_get_cluster_size_del(QemuOpts *opts, Error **errp)
3280 size_t cluster_size;
3282 cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
3283 DEFAULT_CLUSTER_SIZE);
3284 if (!validate_cluster_size(cluster_size, errp)) {
3285 return 0;
3287 return cluster_size;
3290 static int qcow2_opt_get_version_del(QemuOpts *opts, Error **errp)
3292 char *buf;
3293 int ret;
3295 buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
3296 if (!buf) {
3297 ret = 3; /* default */
3298 } else if (!strcmp(buf, "0.10")) {
3299 ret = 2;
3300 } else if (!strcmp(buf, "1.1")) {
3301 ret = 3;
3302 } else {
3303 error_setg(errp, "Invalid compatibility level: '%s'", buf);
3304 ret = -EINVAL;
3306 g_free(buf);
3307 return ret;
3310 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts *opts, int version,
3311 Error **errp)
3313 uint64_t refcount_bits;
3315 refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 16);
3316 if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
3317 error_setg(errp, "Refcount width must be a power of two and may not "
3318 "exceed 64 bits");
3319 return 0;
3322 if (version < 3 && refcount_bits != 16) {
3323 error_setg(errp, "Different refcount widths than 16 bits require "
3324 "compatibility level 1.1 or above (use compat=1.1 or "
3325 "greater)");
3326 return 0;
3329 return refcount_bits;
3332 static int coroutine_fn
3333 qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp)
3335 BlockdevCreateOptionsQcow2 *qcow2_opts;
3336 QDict *options;
3339 * Open the image file and write a minimal qcow2 header.
3341 * We keep things simple and start with a zero-sized image. We also
3342 * do without refcount blocks or a L1 table for now. We'll fix the
3343 * inconsistency later.
3345 * We do need a refcount table because growing the refcount table means
3346 * allocating two new refcount blocks - the second of which would be at
3347 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
3348 * size for any qcow2 image.
3350 BlockBackend *blk = NULL;
3351 BlockDriverState *bs = NULL;
3352 BlockDriverState *data_bs = NULL;
3353 QCowHeader *header;
3354 size_t cluster_size;
3355 int version;
3356 int refcount_order;
3357 uint64_t* refcount_table;
3358 int ret;
3359 uint8_t compression_type = QCOW2_COMPRESSION_TYPE_ZLIB;
3361 assert(create_options->driver == BLOCKDEV_DRIVER_QCOW2);
3362 qcow2_opts = &create_options->u.qcow2;
3364 bs = bdrv_open_blockdev_ref(qcow2_opts->file, errp);
3365 if (bs == NULL) {
3366 return -EIO;
3369 /* Validate options and set default values */
3370 if (!QEMU_IS_ALIGNED(qcow2_opts->size, BDRV_SECTOR_SIZE)) {
3371 error_setg(errp, "Image size must be a multiple of %u bytes",
3372 (unsigned) BDRV_SECTOR_SIZE);
3373 ret = -EINVAL;
3374 goto out;
3377 if (qcow2_opts->has_version) {
3378 switch (qcow2_opts->version) {
3379 case BLOCKDEV_QCOW2_VERSION_V2:
3380 version = 2;
3381 break;
3382 case BLOCKDEV_QCOW2_VERSION_V3:
3383 version = 3;
3384 break;
3385 default:
3386 g_assert_not_reached();
3388 } else {
3389 version = 3;
3392 if (qcow2_opts->has_cluster_size) {
3393 cluster_size = qcow2_opts->cluster_size;
3394 } else {
3395 cluster_size = DEFAULT_CLUSTER_SIZE;
3398 if (!validate_cluster_size(cluster_size, errp)) {
3399 ret = -EINVAL;
3400 goto out;
3403 if (!qcow2_opts->has_preallocation) {
3404 qcow2_opts->preallocation = PREALLOC_MODE_OFF;
3406 if (qcow2_opts->has_backing_file &&
3407 qcow2_opts->preallocation != PREALLOC_MODE_OFF)
3409 error_setg(errp, "Backing file and preallocation cannot be used at "
3410 "the same time");
3411 ret = -EINVAL;
3412 goto out;
3414 if (qcow2_opts->has_backing_fmt && !qcow2_opts->has_backing_file) {
3415 error_setg(errp, "Backing format cannot be used without backing file");
3416 ret = -EINVAL;
3417 goto out;
3420 if (!qcow2_opts->has_lazy_refcounts) {
3421 qcow2_opts->lazy_refcounts = false;
3423 if (version < 3 && qcow2_opts->lazy_refcounts) {
3424 error_setg(errp, "Lazy refcounts only supported with compatibility "
3425 "level 1.1 and above (use version=v3 or greater)");
3426 ret = -EINVAL;
3427 goto out;
3430 if (!qcow2_opts->has_refcount_bits) {
3431 qcow2_opts->refcount_bits = 16;
3433 if (qcow2_opts->refcount_bits > 64 ||
3434 !is_power_of_2(qcow2_opts->refcount_bits))
3436 error_setg(errp, "Refcount width must be a power of two and may not "
3437 "exceed 64 bits");
3438 ret = -EINVAL;
3439 goto out;
3441 if (version < 3 && qcow2_opts->refcount_bits != 16) {
3442 error_setg(errp, "Different refcount widths than 16 bits require "
3443 "compatibility level 1.1 or above (use version=v3 or "
3444 "greater)");
3445 ret = -EINVAL;
3446 goto out;
3448 refcount_order = ctz32(qcow2_opts->refcount_bits);
3450 if (qcow2_opts->data_file_raw && !qcow2_opts->data_file) {
3451 error_setg(errp, "data-file-raw requires data-file");
3452 ret = -EINVAL;
3453 goto out;
3455 if (qcow2_opts->data_file_raw && qcow2_opts->has_backing_file) {
3456 error_setg(errp, "Backing file and data-file-raw cannot be used at "
3457 "the same time");
3458 ret = -EINVAL;
3459 goto out;
3462 if (qcow2_opts->data_file) {
3463 if (version < 3) {
3464 error_setg(errp, "External data files are only supported with "
3465 "compatibility level 1.1 and above (use version=v3 or "
3466 "greater)");
3467 ret = -EINVAL;
3468 goto out;
3470 data_bs = bdrv_open_blockdev_ref(qcow2_opts->data_file, errp);
3471 if (data_bs == NULL) {
3472 ret = -EIO;
3473 goto out;
3477 if (qcow2_opts->has_compression_type &&
3478 qcow2_opts->compression_type != QCOW2_COMPRESSION_TYPE_ZLIB) {
3480 ret = -EINVAL;
3482 if (version < 3) {
3483 error_setg(errp, "Non-zlib compression type is only supported with "
3484 "compatibility level 1.1 and above (use version=v3 or "
3485 "greater)");
3486 goto out;
3489 switch (qcow2_opts->compression_type) {
3490 #ifdef CONFIG_ZSTD
3491 case QCOW2_COMPRESSION_TYPE_ZSTD:
3492 break;
3493 #endif
3494 default:
3495 error_setg(errp, "Unknown compression type");
3496 goto out;
3499 compression_type = qcow2_opts->compression_type;
3502 /* Create BlockBackend to write to the image */
3503 blk = blk_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL,
3504 errp);
3505 if (!blk) {
3506 ret = -EPERM;
3507 goto out;
3509 blk_set_allow_write_beyond_eof(blk, true);
3511 /* Write the header */
3512 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
3513 header = g_malloc0(cluster_size);
3514 *header = (QCowHeader) {
3515 .magic = cpu_to_be32(QCOW_MAGIC),
3516 .version = cpu_to_be32(version),
3517 .cluster_bits = cpu_to_be32(ctz32(cluster_size)),
3518 .size = cpu_to_be64(0),
3519 .l1_table_offset = cpu_to_be64(0),
3520 .l1_size = cpu_to_be32(0),
3521 .refcount_table_offset = cpu_to_be64(cluster_size),
3522 .refcount_table_clusters = cpu_to_be32(1),
3523 .refcount_order = cpu_to_be32(refcount_order),
3524 /* don't deal with endianness since compression_type is 1 byte long */
3525 .compression_type = compression_type,
3526 .header_length = cpu_to_be32(sizeof(*header)),
3529 /* We'll update this to correct value later */
3530 header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
3532 if (qcow2_opts->lazy_refcounts) {
3533 header->compatible_features |=
3534 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
3536 if (data_bs) {
3537 header->incompatible_features |=
3538 cpu_to_be64(QCOW2_INCOMPAT_DATA_FILE);
3540 if (qcow2_opts->data_file_raw) {
3541 header->autoclear_features |=
3542 cpu_to_be64(QCOW2_AUTOCLEAR_DATA_FILE_RAW);
3544 if (compression_type != QCOW2_COMPRESSION_TYPE_ZLIB) {
3545 header->incompatible_features |=
3546 cpu_to_be64(QCOW2_INCOMPAT_COMPRESSION);
3549 ret = blk_pwrite(blk, 0, header, cluster_size, 0);
3550 g_free(header);
3551 if (ret < 0) {
3552 error_setg_errno(errp, -ret, "Could not write qcow2 header");
3553 goto out;
3556 /* Write a refcount table with one refcount block */
3557 refcount_table = g_malloc0(2 * cluster_size);
3558 refcount_table[0] = cpu_to_be64(2 * cluster_size);
3559 ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0);
3560 g_free(refcount_table);
3562 if (ret < 0) {
3563 error_setg_errno(errp, -ret, "Could not write refcount table");
3564 goto out;
3567 blk_unref(blk);
3568 blk = NULL;
3571 * And now open the image and make it consistent first (i.e. increase the
3572 * refcount of the cluster that is occupied by the header and the refcount
3573 * table)
3575 options = qdict_new();
3576 qdict_put_str(options, "driver", "qcow2");
3577 qdict_put_str(options, "file", bs->node_name);
3578 if (data_bs) {
3579 qdict_put_str(options, "data-file", data_bs->node_name);
3581 blk = blk_new_open(NULL, NULL, options,
3582 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH,
3583 errp);
3584 if (blk == NULL) {
3585 ret = -EIO;
3586 goto out;
3589 ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size);
3590 if (ret < 0) {
3591 error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
3592 "header and refcount table");
3593 goto out;
3595 } else if (ret != 0) {
3596 error_report("Huh, first cluster in empty image is already in use?");
3597 abort();
3600 /* Set the external data file if necessary */
3601 if (data_bs) {
3602 BDRVQcow2State *s = blk_bs(blk)->opaque;
3603 s->image_data_file = g_strdup(data_bs->filename);
3606 /* Create a full header (including things like feature table) */
3607 ret = qcow2_update_header(blk_bs(blk));
3608 if (ret < 0) {
3609 error_setg_errno(errp, -ret, "Could not update qcow2 header");
3610 goto out;
3613 /* Okay, now that we have a valid image, let's give it the right size */
3614 ret = blk_truncate(blk, qcow2_opts->size, false, qcow2_opts->preallocation,
3615 0, errp);
3616 if (ret < 0) {
3617 error_prepend(errp, "Could not resize image: ");
3618 goto out;
3621 /* Want a backing file? There you go. */
3622 if (qcow2_opts->has_backing_file) {
3623 const char *backing_format = NULL;
3625 if (qcow2_opts->has_backing_fmt) {
3626 backing_format = BlockdevDriver_str(qcow2_opts->backing_fmt);
3629 ret = bdrv_change_backing_file(blk_bs(blk), qcow2_opts->backing_file,
3630 backing_format);
3631 if (ret < 0) {
3632 error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
3633 "with format '%s'", qcow2_opts->backing_file,
3634 backing_format);
3635 goto out;
3639 /* Want encryption? There you go. */
3640 if (qcow2_opts->has_encrypt) {
3641 ret = qcow2_set_up_encryption(blk_bs(blk), qcow2_opts->encrypt, errp);
3642 if (ret < 0) {
3643 goto out;
3647 blk_unref(blk);
3648 blk = NULL;
3650 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
3651 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
3652 * have to setup decryption context. We're not doing any I/O on the top
3653 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
3654 * not have effect.
3656 options = qdict_new();
3657 qdict_put_str(options, "driver", "qcow2");
3658 qdict_put_str(options, "file", bs->node_name);
3659 if (data_bs) {
3660 qdict_put_str(options, "data-file", data_bs->node_name);
3662 blk = blk_new_open(NULL, NULL, options,
3663 BDRV_O_RDWR | BDRV_O_NO_BACKING | BDRV_O_NO_IO,
3664 errp);
3665 if (blk == NULL) {
3666 ret = -EIO;
3667 goto out;
3670 ret = 0;
3671 out:
3672 blk_unref(blk);
3673 bdrv_unref(bs);
3674 bdrv_unref(data_bs);
3675 return ret;
3678 static int coroutine_fn qcow2_co_create_opts(BlockDriver *drv,
3679 const char *filename,
3680 QemuOpts *opts,
3681 Error **errp)
3683 BlockdevCreateOptions *create_options = NULL;
3684 QDict *qdict;
3685 Visitor *v;
3686 BlockDriverState *bs = NULL;
3687 BlockDriverState *data_bs = NULL;
3688 const char *val;
3689 int ret;
3691 /* Only the keyval visitor supports the dotted syntax needed for
3692 * encryption, so go through a QDict before getting a QAPI type. Ignore
3693 * options meant for the protocol layer so that the visitor doesn't
3694 * complain. */
3695 qdict = qemu_opts_to_qdict_filtered(opts, NULL, bdrv_qcow2.create_opts,
3696 true);
3698 /* Handle encryption options */
3699 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT);
3700 if (val && !strcmp(val, "on")) {
3701 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow");
3702 } else if (val && !strcmp(val, "off")) {
3703 qdict_del(qdict, BLOCK_OPT_ENCRYPT);
3706 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT);
3707 if (val && !strcmp(val, "aes")) {
3708 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow");
3711 /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into
3712 * version=v2/v3 below. */
3713 val = qdict_get_try_str(qdict, BLOCK_OPT_COMPAT_LEVEL);
3714 if (val && !strcmp(val, "0.10")) {
3715 qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v2");
3716 } else if (val && !strcmp(val, "1.1")) {
3717 qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v3");
3720 /* Change legacy command line options into QMP ones */
3721 static const QDictRenames opt_renames[] = {
3722 { BLOCK_OPT_BACKING_FILE, "backing-file" },
3723 { BLOCK_OPT_BACKING_FMT, "backing-fmt" },
3724 { BLOCK_OPT_CLUSTER_SIZE, "cluster-size" },
3725 { BLOCK_OPT_LAZY_REFCOUNTS, "lazy-refcounts" },
3726 { BLOCK_OPT_REFCOUNT_BITS, "refcount-bits" },
3727 { BLOCK_OPT_ENCRYPT, BLOCK_OPT_ENCRYPT_FORMAT },
3728 { BLOCK_OPT_COMPAT_LEVEL, "version" },
3729 { BLOCK_OPT_DATA_FILE_RAW, "data-file-raw" },
3730 { BLOCK_OPT_COMPRESSION_TYPE, "compression-type" },
3731 { NULL, NULL },
3734 if (!qdict_rename_keys(qdict, opt_renames, errp)) {
3735 ret = -EINVAL;
3736 goto finish;
3739 /* Create and open the file (protocol layer) */
3740 ret = bdrv_create_file(filename, opts, errp);
3741 if (ret < 0) {
3742 goto finish;
3745 bs = bdrv_open(filename, NULL, NULL,
3746 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
3747 if (bs == NULL) {
3748 ret = -EIO;
3749 goto finish;
3752 /* Create and open an external data file (protocol layer) */
3753 val = qdict_get_try_str(qdict, BLOCK_OPT_DATA_FILE);
3754 if (val) {
3755 ret = bdrv_create_file(val, opts, errp);
3756 if (ret < 0) {
3757 goto finish;
3760 data_bs = bdrv_open(val, NULL, NULL,
3761 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
3762 errp);
3763 if (data_bs == NULL) {
3764 ret = -EIO;
3765 goto finish;
3768 qdict_del(qdict, BLOCK_OPT_DATA_FILE);
3769 qdict_put_str(qdict, "data-file", data_bs->node_name);
3772 /* Set 'driver' and 'node' options */
3773 qdict_put_str(qdict, "driver", "qcow2");
3774 qdict_put_str(qdict, "file", bs->node_name);
3776 /* Now get the QAPI type BlockdevCreateOptions */
3777 v = qobject_input_visitor_new_flat_confused(qdict, errp);
3778 if (!v) {
3779 ret = -EINVAL;
3780 goto finish;
3783 visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp);
3784 visit_free(v);
3785 if (!create_options) {
3786 ret = -EINVAL;
3787 goto finish;
3790 /* Silently round up size */
3791 create_options->u.qcow2.size = ROUND_UP(create_options->u.qcow2.size,
3792 BDRV_SECTOR_SIZE);
3794 /* Create the qcow2 image (format layer) */
3795 ret = qcow2_co_create(create_options, errp);
3796 if (ret < 0) {
3797 goto finish;
3800 ret = 0;
3801 finish:
3802 qobject_unref(qdict);
3803 bdrv_unref(bs);
3804 bdrv_unref(data_bs);
3805 qapi_free_BlockdevCreateOptions(create_options);
3806 return ret;
3810 static bool is_zero(BlockDriverState *bs, int64_t offset, int64_t bytes)
3812 int64_t nr;
3813 int res;
3815 /* Clamp to image length, before checking status of underlying sectors */
3816 if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) {
3817 bytes = bs->total_sectors * BDRV_SECTOR_SIZE - offset;
3820 if (!bytes) {
3821 return true;
3823 res = bdrv_block_status_above(bs, NULL, offset, bytes, &nr, NULL, NULL);
3824 return res >= 0 && (res & BDRV_BLOCK_ZERO) && nr == bytes;
3827 static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs,
3828 int64_t offset, int bytes, BdrvRequestFlags flags)
3830 int ret;
3831 BDRVQcow2State *s = bs->opaque;
3833 uint32_t head = offset % s->cluster_size;
3834 uint32_t tail = (offset + bytes) % s->cluster_size;
3836 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, bytes);
3837 if (offset + bytes == bs->total_sectors * BDRV_SECTOR_SIZE) {
3838 tail = 0;
3841 if (head || tail) {
3842 uint64_t off;
3843 unsigned int nr;
3845 assert(head + bytes <= s->cluster_size);
3847 /* check whether remainder of cluster already reads as zero */
3848 if (!(is_zero(bs, offset - head, head) &&
3849 is_zero(bs, offset + bytes,
3850 tail ? s->cluster_size - tail : 0))) {
3851 return -ENOTSUP;
3854 qemu_co_mutex_lock(&s->lock);
3855 /* We can have new write after previous check */
3856 offset = QEMU_ALIGN_DOWN(offset, s->cluster_size);
3857 bytes = s->cluster_size;
3858 nr = s->cluster_size;
3859 ret = qcow2_get_cluster_offset(bs, offset, &nr, &off);
3860 if (ret != QCOW2_CLUSTER_UNALLOCATED &&
3861 ret != QCOW2_CLUSTER_ZERO_PLAIN &&
3862 ret != QCOW2_CLUSTER_ZERO_ALLOC) {
3863 qemu_co_mutex_unlock(&s->lock);
3864 return -ENOTSUP;
3866 } else {
3867 qemu_co_mutex_lock(&s->lock);
3870 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, bytes);
3872 /* Whatever is left can use real zero clusters */
3873 ret = qcow2_cluster_zeroize(bs, offset, bytes, flags);
3874 qemu_co_mutex_unlock(&s->lock);
3876 return ret;
3879 static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs,
3880 int64_t offset, int bytes)
3882 int ret;
3883 BDRVQcow2State *s = bs->opaque;
3885 /* If the image does not support QCOW_OFLAG_ZERO then discarding
3886 * clusters could expose stale data from the backing file. */
3887 if (s->qcow_version < 3 && bs->backing) {
3888 return -ENOTSUP;
3891 if (!QEMU_IS_ALIGNED(offset | bytes, s->cluster_size)) {
3892 assert(bytes < s->cluster_size);
3893 /* Ignore partial clusters, except for the special case of the
3894 * complete partial cluster at the end of an unaligned file */
3895 if (!QEMU_IS_ALIGNED(offset, s->cluster_size) ||
3896 offset + bytes != bs->total_sectors * BDRV_SECTOR_SIZE) {
3897 return -ENOTSUP;
3901 qemu_co_mutex_lock(&s->lock);
3902 ret = qcow2_cluster_discard(bs, offset, bytes, QCOW2_DISCARD_REQUEST,
3903 false);
3904 qemu_co_mutex_unlock(&s->lock);
3905 return ret;
3908 static int coroutine_fn
3909 qcow2_co_copy_range_from(BlockDriverState *bs,
3910 BdrvChild *src, uint64_t src_offset,
3911 BdrvChild *dst, uint64_t dst_offset,
3912 uint64_t bytes, BdrvRequestFlags read_flags,
3913 BdrvRequestFlags write_flags)
3915 BDRVQcow2State *s = bs->opaque;
3916 int ret;
3917 unsigned int cur_bytes; /* number of bytes in current iteration */
3918 BdrvChild *child = NULL;
3919 BdrvRequestFlags cur_write_flags;
3921 assert(!bs->encrypted);
3922 qemu_co_mutex_lock(&s->lock);
3924 while (bytes != 0) {
3925 uint64_t copy_offset = 0;
3926 /* prepare next request */
3927 cur_bytes = MIN(bytes, INT_MAX);
3928 cur_write_flags = write_flags;
3930 ret = qcow2_get_cluster_offset(bs, src_offset, &cur_bytes, &copy_offset);
3931 if (ret < 0) {
3932 goto out;
3935 switch (ret) {
3936 case QCOW2_CLUSTER_UNALLOCATED:
3937 if (bs->backing && bs->backing->bs) {
3938 int64_t backing_length = bdrv_getlength(bs->backing->bs);
3939 if (src_offset >= backing_length) {
3940 cur_write_flags |= BDRV_REQ_ZERO_WRITE;
3941 } else {
3942 child = bs->backing;
3943 cur_bytes = MIN(cur_bytes, backing_length - src_offset);
3944 copy_offset = src_offset;
3946 } else {
3947 cur_write_flags |= BDRV_REQ_ZERO_WRITE;
3949 break;
3951 case QCOW2_CLUSTER_ZERO_PLAIN:
3952 case QCOW2_CLUSTER_ZERO_ALLOC:
3953 cur_write_flags |= BDRV_REQ_ZERO_WRITE;
3954 break;
3956 case QCOW2_CLUSTER_COMPRESSED:
3957 ret = -ENOTSUP;
3958 goto out;
3960 case QCOW2_CLUSTER_NORMAL:
3961 child = s->data_file;
3962 copy_offset += offset_into_cluster(s, src_offset);
3963 break;
3965 default:
3966 abort();
3968 qemu_co_mutex_unlock(&s->lock);
3969 ret = bdrv_co_copy_range_from(child,
3970 copy_offset,
3971 dst, dst_offset,
3972 cur_bytes, read_flags, cur_write_flags);
3973 qemu_co_mutex_lock(&s->lock);
3974 if (ret < 0) {
3975 goto out;
3978 bytes -= cur_bytes;
3979 src_offset += cur_bytes;
3980 dst_offset += cur_bytes;
3982 ret = 0;
3984 out:
3985 qemu_co_mutex_unlock(&s->lock);
3986 return ret;
3989 static int coroutine_fn
3990 qcow2_co_copy_range_to(BlockDriverState *bs,
3991 BdrvChild *src, uint64_t src_offset,
3992 BdrvChild *dst, uint64_t dst_offset,
3993 uint64_t bytes, BdrvRequestFlags read_flags,
3994 BdrvRequestFlags write_flags)
3996 BDRVQcow2State *s = bs->opaque;
3997 int offset_in_cluster;
3998 int ret;
3999 unsigned int cur_bytes; /* number of sectors in current iteration */
4000 uint64_t cluster_offset;
4001 QCowL2Meta *l2meta = NULL;
4003 assert(!bs->encrypted);
4005 qemu_co_mutex_lock(&s->lock);
4007 while (bytes != 0) {
4009 l2meta = NULL;
4011 offset_in_cluster = offset_into_cluster(s, dst_offset);
4012 cur_bytes = MIN(bytes, INT_MAX);
4014 /* TODO:
4015 * If src->bs == dst->bs, we could simply copy by incrementing
4016 * the refcnt, without copying user data.
4017 * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */
4018 ret = qcow2_alloc_cluster_offset(bs, dst_offset, &cur_bytes,
4019 &cluster_offset, &l2meta);
4020 if (ret < 0) {
4021 goto fail;
4024 assert(offset_into_cluster(s, cluster_offset) == 0);
4026 ret = qcow2_pre_write_overlap_check(bs, 0,
4027 cluster_offset + offset_in_cluster, cur_bytes, true);
4028 if (ret < 0) {
4029 goto fail;
4032 qemu_co_mutex_unlock(&s->lock);
4033 ret = bdrv_co_copy_range_to(src, src_offset,
4034 s->data_file,
4035 cluster_offset + offset_in_cluster,
4036 cur_bytes, read_flags, write_flags);
4037 qemu_co_mutex_lock(&s->lock);
4038 if (ret < 0) {
4039 goto fail;
4042 ret = qcow2_handle_l2meta(bs, &l2meta, true);
4043 if (ret) {
4044 goto fail;
4047 bytes -= cur_bytes;
4048 src_offset += cur_bytes;
4049 dst_offset += cur_bytes;
4051 ret = 0;
4053 fail:
4054 qcow2_handle_l2meta(bs, &l2meta, false);
4056 qemu_co_mutex_unlock(&s->lock);
4058 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
4060 return ret;
4063 static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset,
4064 bool exact, PreallocMode prealloc,
4065 BdrvRequestFlags flags, Error **errp)
4067 BDRVQcow2State *s = bs->opaque;
4068 uint64_t old_length;
4069 int64_t new_l1_size;
4070 int ret;
4071 QDict *options;
4073 if (prealloc != PREALLOC_MODE_OFF && prealloc != PREALLOC_MODE_METADATA &&
4074 prealloc != PREALLOC_MODE_FALLOC && prealloc != PREALLOC_MODE_FULL)
4076 error_setg(errp, "Unsupported preallocation mode '%s'",
4077 PreallocMode_str(prealloc));
4078 return -ENOTSUP;
4081 if (!QEMU_IS_ALIGNED(offset, BDRV_SECTOR_SIZE)) {
4082 error_setg(errp, "The new size must be a multiple of %u",
4083 (unsigned) BDRV_SECTOR_SIZE);
4084 return -EINVAL;
4087 qemu_co_mutex_lock(&s->lock);
4090 * Even though we store snapshot size for all images, it was not
4091 * required until v3, so it is not safe to proceed for v2.
4093 if (s->nb_snapshots && s->qcow_version < 3) {
4094 error_setg(errp, "Can't resize a v2 image which has snapshots");
4095 ret = -ENOTSUP;
4096 goto fail;
4099 /* See qcow2-bitmap.c for which bitmap scenarios prevent a resize. */
4100 if (qcow2_truncate_bitmaps_check(bs, errp)) {
4101 ret = -ENOTSUP;
4102 goto fail;
4105 old_length = bs->total_sectors * BDRV_SECTOR_SIZE;
4106 new_l1_size = size_to_l1(s, offset);
4108 if (offset < old_length) {
4109 int64_t last_cluster, old_file_size;
4110 if (prealloc != PREALLOC_MODE_OFF) {
4111 error_setg(errp,
4112 "Preallocation can't be used for shrinking an image");
4113 ret = -EINVAL;
4114 goto fail;
4117 ret = qcow2_cluster_discard(bs, ROUND_UP(offset, s->cluster_size),
4118 old_length - ROUND_UP(offset,
4119 s->cluster_size),
4120 QCOW2_DISCARD_ALWAYS, true);
4121 if (ret < 0) {
4122 error_setg_errno(errp, -ret, "Failed to discard cropped clusters");
4123 goto fail;
4126 ret = qcow2_shrink_l1_table(bs, new_l1_size);
4127 if (ret < 0) {
4128 error_setg_errno(errp, -ret,
4129 "Failed to reduce the number of L2 tables");
4130 goto fail;
4133 ret = qcow2_shrink_reftable(bs);
4134 if (ret < 0) {
4135 error_setg_errno(errp, -ret,
4136 "Failed to discard unused refblocks");
4137 goto fail;
4140 old_file_size = bdrv_getlength(bs->file->bs);
4141 if (old_file_size < 0) {
4142 error_setg_errno(errp, -old_file_size,
4143 "Failed to inquire current file length");
4144 ret = old_file_size;
4145 goto fail;
4147 last_cluster = qcow2_get_last_cluster(bs, old_file_size);
4148 if (last_cluster < 0) {
4149 error_setg_errno(errp, -last_cluster,
4150 "Failed to find the last cluster");
4151 ret = last_cluster;
4152 goto fail;
4154 if ((last_cluster + 1) * s->cluster_size < old_file_size) {
4155 Error *local_err = NULL;
4158 * Do not pass @exact here: It will not help the user if
4159 * we get an error here just because they wanted to shrink
4160 * their qcow2 image (on a block device) with qemu-img.
4161 * (And on the qcow2 layer, the @exact requirement is
4162 * always fulfilled, so there is no need to pass it on.)
4164 bdrv_co_truncate(bs->file, (last_cluster + 1) * s->cluster_size,
4165 false, PREALLOC_MODE_OFF, 0, &local_err);
4166 if (local_err) {
4167 warn_reportf_err(local_err,
4168 "Failed to truncate the tail of the image: ");
4171 } else {
4172 ret = qcow2_grow_l1_table(bs, new_l1_size, true);
4173 if (ret < 0) {
4174 error_setg_errno(errp, -ret, "Failed to grow the L1 table");
4175 goto fail;
4179 switch (prealloc) {
4180 case PREALLOC_MODE_OFF:
4181 if (has_data_file(bs)) {
4183 * If the caller wants an exact resize, the external data
4184 * file should be resized to the exact target size, too,
4185 * so we pass @exact here.
4187 ret = bdrv_co_truncate(s->data_file, offset, exact, prealloc, 0,
4188 errp);
4189 if (ret < 0) {
4190 goto fail;
4193 break;
4195 case PREALLOC_MODE_METADATA:
4196 ret = preallocate_co(bs, old_length, offset, prealloc, errp);
4197 if (ret < 0) {
4198 goto fail;
4200 break;
4202 case PREALLOC_MODE_FALLOC:
4203 case PREALLOC_MODE_FULL:
4205 int64_t allocation_start, host_offset, guest_offset;
4206 int64_t clusters_allocated;
4207 int64_t old_file_size, last_cluster, new_file_size;
4208 uint64_t nb_new_data_clusters, nb_new_l2_tables;
4210 /* With a data file, preallocation means just allocating the metadata
4211 * and forwarding the truncate request to the data file */
4212 if (has_data_file(bs)) {
4213 ret = preallocate_co(bs, old_length, offset, prealloc, errp);
4214 if (ret < 0) {
4215 goto fail;
4217 break;
4220 old_file_size = bdrv_getlength(bs->file->bs);
4221 if (old_file_size < 0) {
4222 error_setg_errno(errp, -old_file_size,
4223 "Failed to inquire current file length");
4224 ret = old_file_size;
4225 goto fail;
4228 last_cluster = qcow2_get_last_cluster(bs, old_file_size);
4229 if (last_cluster >= 0) {
4230 old_file_size = (last_cluster + 1) * s->cluster_size;
4231 } else {
4232 old_file_size = ROUND_UP(old_file_size, s->cluster_size);
4235 nb_new_data_clusters = (ROUND_UP(offset, s->cluster_size) -
4236 start_of_cluster(s, old_length)) >> s->cluster_bits;
4238 /* This is an overestimation; we will not actually allocate space for
4239 * these in the file but just make sure the new refcount structures are
4240 * able to cover them so we will not have to allocate new refblocks
4241 * while entering the data blocks in the potentially new L2 tables.
4242 * (We do not actually care where the L2 tables are placed. Maybe they
4243 * are already allocated or they can be placed somewhere before
4244 * @old_file_size. It does not matter because they will be fully
4245 * allocated automatically, so they do not need to be covered by the
4246 * preallocation. All that matters is that we will not have to allocate
4247 * new refcount structures for them.) */
4248 nb_new_l2_tables = DIV_ROUND_UP(nb_new_data_clusters,
4249 s->cluster_size / sizeof(uint64_t));
4250 /* The cluster range may not be aligned to L2 boundaries, so add one L2
4251 * table for a potential head/tail */
4252 nb_new_l2_tables++;
4254 allocation_start = qcow2_refcount_area(bs, old_file_size,
4255 nb_new_data_clusters +
4256 nb_new_l2_tables,
4257 true, 0, 0);
4258 if (allocation_start < 0) {
4259 error_setg_errno(errp, -allocation_start,
4260 "Failed to resize refcount structures");
4261 ret = allocation_start;
4262 goto fail;
4265 clusters_allocated = qcow2_alloc_clusters_at(bs, allocation_start,
4266 nb_new_data_clusters);
4267 if (clusters_allocated < 0) {
4268 error_setg_errno(errp, -clusters_allocated,
4269 "Failed to allocate data clusters");
4270 ret = clusters_allocated;
4271 goto fail;
4274 assert(clusters_allocated == nb_new_data_clusters);
4276 /* Allocate the data area */
4277 new_file_size = allocation_start +
4278 nb_new_data_clusters * s->cluster_size;
4280 * Image file grows, so @exact does not matter.
4282 * If we need to zero out the new area, try first whether the protocol
4283 * driver can already take care of this.
4285 if (flags & BDRV_REQ_ZERO_WRITE) {
4286 ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc,
4287 BDRV_REQ_ZERO_WRITE, NULL);
4288 if (ret >= 0) {
4289 flags &= ~BDRV_REQ_ZERO_WRITE;
4291 } else {
4292 ret = -1;
4294 if (ret < 0) {
4295 ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc, 0,
4296 errp);
4298 if (ret < 0) {
4299 error_prepend(errp, "Failed to resize underlying file: ");
4300 qcow2_free_clusters(bs, allocation_start,
4301 nb_new_data_clusters * s->cluster_size,
4302 QCOW2_DISCARD_OTHER);
4303 goto fail;
4306 /* Create the necessary L2 entries */
4307 host_offset = allocation_start;
4308 guest_offset = old_length;
4309 while (nb_new_data_clusters) {
4310 int64_t nb_clusters = MIN(
4311 nb_new_data_clusters,
4312 s->l2_slice_size - offset_to_l2_slice_index(s, guest_offset));
4313 unsigned cow_start_length = offset_into_cluster(s, guest_offset);
4314 QCowL2Meta allocation;
4315 guest_offset = start_of_cluster(s, guest_offset);
4316 allocation = (QCowL2Meta) {
4317 .offset = guest_offset,
4318 .alloc_offset = host_offset,
4319 .nb_clusters = nb_clusters,
4320 .cow_start = {
4321 .offset = 0,
4322 .nb_bytes = cow_start_length,
4324 .cow_end = {
4325 .offset = nb_clusters << s->cluster_bits,
4326 .nb_bytes = 0,
4329 qemu_co_queue_init(&allocation.dependent_requests);
4331 ret = qcow2_alloc_cluster_link_l2(bs, &allocation);
4332 if (ret < 0) {
4333 error_setg_errno(errp, -ret, "Failed to update L2 tables");
4334 qcow2_free_clusters(bs, host_offset,
4335 nb_new_data_clusters * s->cluster_size,
4336 QCOW2_DISCARD_OTHER);
4337 goto fail;
4340 guest_offset += nb_clusters * s->cluster_size;
4341 host_offset += nb_clusters * s->cluster_size;
4342 nb_new_data_clusters -= nb_clusters;
4344 break;
4347 default:
4348 g_assert_not_reached();
4351 if ((flags & BDRV_REQ_ZERO_WRITE) && offset > old_length) {
4352 uint64_t zero_start = QEMU_ALIGN_UP(old_length, s->cluster_size);
4355 * Use zero clusters as much as we can. qcow2_cluster_zeroize()
4356 * requires a cluster-aligned start. The end may be unaligned if it is
4357 * at the end of the image (which it is here).
4359 if (offset > zero_start) {
4360 ret = qcow2_cluster_zeroize(bs, zero_start, offset - zero_start, 0);
4361 if (ret < 0) {
4362 error_setg_errno(errp, -ret, "Failed to zero out new clusters");
4363 goto fail;
4367 /* Write explicit zeros for the unaligned head */
4368 if (zero_start > old_length) {
4369 uint64_t len = MIN(zero_start, offset) - old_length;
4370 uint8_t *buf = qemu_blockalign0(bs, len);
4371 QEMUIOVector qiov;
4372 qemu_iovec_init_buf(&qiov, buf, len);
4374 qemu_co_mutex_unlock(&s->lock);
4375 ret = qcow2_co_pwritev_part(bs, old_length, len, &qiov, 0, 0);
4376 qemu_co_mutex_lock(&s->lock);
4378 qemu_vfree(buf);
4379 if (ret < 0) {
4380 error_setg_errno(errp, -ret, "Failed to zero out the new area");
4381 goto fail;
4386 if (prealloc != PREALLOC_MODE_OFF) {
4387 /* Flush metadata before actually changing the image size */
4388 ret = qcow2_write_caches(bs);
4389 if (ret < 0) {
4390 error_setg_errno(errp, -ret,
4391 "Failed to flush the preallocated area to disk");
4392 goto fail;
4396 bs->total_sectors = offset / BDRV_SECTOR_SIZE;
4398 /* write updated header.size */
4399 offset = cpu_to_be64(offset);
4400 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
4401 &offset, sizeof(uint64_t));
4402 if (ret < 0) {
4403 error_setg_errno(errp, -ret, "Failed to update the image size");
4404 goto fail;
4407 s->l1_vm_state_index = new_l1_size;
4409 /* Update cache sizes */
4410 options = qdict_clone_shallow(bs->options);
4411 ret = qcow2_update_options(bs, options, s->flags, errp);
4412 qobject_unref(options);
4413 if (ret < 0) {
4414 goto fail;
4416 ret = 0;
4417 fail:
4418 qemu_co_mutex_unlock(&s->lock);
4419 return ret;
4422 static coroutine_fn int
4423 qcow2_co_pwritev_compressed_task(BlockDriverState *bs,
4424 uint64_t offset, uint64_t bytes,
4425 QEMUIOVector *qiov, size_t qiov_offset)
4427 BDRVQcow2State *s = bs->opaque;
4428 int ret;
4429 ssize_t out_len;
4430 uint8_t *buf, *out_buf;
4431 uint64_t cluster_offset;
4433 assert(bytes == s->cluster_size || (bytes < s->cluster_size &&
4434 (offset + bytes == bs->total_sectors << BDRV_SECTOR_BITS)));
4436 buf = qemu_blockalign(bs, s->cluster_size);
4437 if (bytes < s->cluster_size) {
4438 /* Zero-pad last write if image size is not cluster aligned */
4439 memset(buf + bytes, 0, s->cluster_size - bytes);
4441 qemu_iovec_to_buf(qiov, qiov_offset, buf, bytes);
4443 out_buf = g_malloc(s->cluster_size);
4445 out_len = qcow2_co_compress(bs, out_buf, s->cluster_size - 1,
4446 buf, s->cluster_size);
4447 if (out_len == -ENOMEM) {
4448 /* could not compress: write normal cluster */
4449 ret = qcow2_co_pwritev_part(bs, offset, bytes, qiov, qiov_offset, 0);
4450 if (ret < 0) {
4451 goto fail;
4453 goto success;
4454 } else if (out_len < 0) {
4455 ret = -EINVAL;
4456 goto fail;
4459 qemu_co_mutex_lock(&s->lock);
4460 ret = qcow2_alloc_compressed_cluster_offset(bs, offset, out_len,
4461 &cluster_offset);
4462 if (ret < 0) {
4463 qemu_co_mutex_unlock(&s->lock);
4464 goto fail;
4467 ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len, true);
4468 qemu_co_mutex_unlock(&s->lock);
4469 if (ret < 0) {
4470 goto fail;
4473 BLKDBG_EVENT(s->data_file, BLKDBG_WRITE_COMPRESSED);
4474 ret = bdrv_co_pwrite(s->data_file, cluster_offset, out_len, out_buf, 0);
4475 if (ret < 0) {
4476 goto fail;
4478 success:
4479 ret = 0;
4480 fail:
4481 qemu_vfree(buf);
4482 g_free(out_buf);
4483 return ret;
4486 static coroutine_fn int qcow2_co_pwritev_compressed_task_entry(AioTask *task)
4488 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
4490 assert(!t->cluster_type && !t->l2meta);
4492 return qcow2_co_pwritev_compressed_task(t->bs, t->offset, t->bytes, t->qiov,
4493 t->qiov_offset);
4497 * XXX: put compressed sectors first, then all the cluster aligned
4498 * tables to avoid losing bytes in alignment
4500 static coroutine_fn int
4501 qcow2_co_pwritev_compressed_part(BlockDriverState *bs,
4502 uint64_t offset, uint64_t bytes,
4503 QEMUIOVector *qiov, size_t qiov_offset)
4505 BDRVQcow2State *s = bs->opaque;
4506 AioTaskPool *aio = NULL;
4507 int ret = 0;
4509 if (has_data_file(bs)) {
4510 return -ENOTSUP;
4513 if (bytes == 0) {
4515 * align end of file to a sector boundary to ease reading with
4516 * sector based I/Os
4518 int64_t len = bdrv_getlength(bs->file->bs);
4519 if (len < 0) {
4520 return len;
4522 return bdrv_co_truncate(bs->file, len, false, PREALLOC_MODE_OFF, 0,
4523 NULL);
4526 if (offset_into_cluster(s, offset)) {
4527 return -EINVAL;
4530 if (offset_into_cluster(s, bytes) &&
4531 (offset + bytes) != (bs->total_sectors << BDRV_SECTOR_BITS)) {
4532 return -EINVAL;
4535 while (bytes && aio_task_pool_status(aio) == 0) {
4536 uint64_t chunk_size = MIN(bytes, s->cluster_size);
4538 if (!aio && chunk_size != bytes) {
4539 aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
4542 ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_compressed_task_entry,
4543 0, 0, offset, chunk_size, qiov, qiov_offset, NULL);
4544 if (ret < 0) {
4545 break;
4547 qiov_offset += chunk_size;
4548 offset += chunk_size;
4549 bytes -= chunk_size;
4552 if (aio) {
4553 aio_task_pool_wait_all(aio);
4554 if (ret == 0) {
4555 ret = aio_task_pool_status(aio);
4557 g_free(aio);
4560 return ret;
4563 static int coroutine_fn
4564 qcow2_co_preadv_compressed(BlockDriverState *bs,
4565 uint64_t file_cluster_offset,
4566 uint64_t offset,
4567 uint64_t bytes,
4568 QEMUIOVector *qiov,
4569 size_t qiov_offset)
4571 BDRVQcow2State *s = bs->opaque;
4572 int ret = 0, csize, nb_csectors;
4573 uint64_t coffset;
4574 uint8_t *buf, *out_buf;
4575 int offset_in_cluster = offset_into_cluster(s, offset);
4577 coffset = file_cluster_offset & s->cluster_offset_mask;
4578 nb_csectors = ((file_cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
4579 csize = nb_csectors * QCOW2_COMPRESSED_SECTOR_SIZE -
4580 (coffset & ~QCOW2_COMPRESSED_SECTOR_MASK);
4582 buf = g_try_malloc(csize);
4583 if (!buf) {
4584 return -ENOMEM;
4587 out_buf = qemu_blockalign(bs, s->cluster_size);
4589 BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
4590 ret = bdrv_co_pread(bs->file, coffset, csize, buf, 0);
4591 if (ret < 0) {
4592 goto fail;
4595 if (qcow2_co_decompress(bs, out_buf, s->cluster_size, buf, csize) < 0) {
4596 ret = -EIO;
4597 goto fail;
4600 qemu_iovec_from_buf(qiov, qiov_offset, out_buf + offset_in_cluster, bytes);
4602 fail:
4603 qemu_vfree(out_buf);
4604 g_free(buf);
4606 return ret;
4609 static int make_completely_empty(BlockDriverState *bs)
4611 BDRVQcow2State *s = bs->opaque;
4612 Error *local_err = NULL;
4613 int ret, l1_clusters;
4614 int64_t offset;
4615 uint64_t *new_reftable = NULL;
4616 uint64_t rt_entry, l1_size2;
4617 struct {
4618 uint64_t l1_offset;
4619 uint64_t reftable_offset;
4620 uint32_t reftable_clusters;
4621 } QEMU_PACKED l1_ofs_rt_ofs_cls;
4623 ret = qcow2_cache_empty(bs, s->l2_table_cache);
4624 if (ret < 0) {
4625 goto fail;
4628 ret = qcow2_cache_empty(bs, s->refcount_block_cache);
4629 if (ret < 0) {
4630 goto fail;
4633 /* Refcounts will be broken utterly */
4634 ret = qcow2_mark_dirty(bs);
4635 if (ret < 0) {
4636 goto fail;
4639 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
4641 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
4642 l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t);
4644 /* After this call, neither the in-memory nor the on-disk refcount
4645 * information accurately describe the actual references */
4647 ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset,
4648 l1_clusters * s->cluster_size, 0);
4649 if (ret < 0) {
4650 goto fail_broken_refcounts;
4652 memset(s->l1_table, 0, l1_size2);
4654 BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE);
4656 /* Overwrite enough clusters at the beginning of the sectors to place
4657 * the refcount table, a refcount block and the L1 table in; this may
4658 * overwrite parts of the existing refcount and L1 table, which is not
4659 * an issue because the dirty flag is set, complete data loss is in fact
4660 * desired and partial data loss is consequently fine as well */
4661 ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size,
4662 (2 + l1_clusters) * s->cluster_size, 0);
4663 /* This call (even if it failed overall) may have overwritten on-disk
4664 * refcount structures; in that case, the in-memory refcount information
4665 * will probably differ from the on-disk information which makes the BDS
4666 * unusable */
4667 if (ret < 0) {
4668 goto fail_broken_refcounts;
4671 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
4672 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
4674 /* "Create" an empty reftable (one cluster) directly after the image
4675 * header and an empty L1 table three clusters after the image header;
4676 * the cluster between those two will be used as the first refblock */
4677 l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size);
4678 l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size);
4679 l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1);
4680 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset),
4681 &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls));
4682 if (ret < 0) {
4683 goto fail_broken_refcounts;
4686 s->l1_table_offset = 3 * s->cluster_size;
4688 new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t));
4689 if (!new_reftable) {
4690 ret = -ENOMEM;
4691 goto fail_broken_refcounts;
4694 s->refcount_table_offset = s->cluster_size;
4695 s->refcount_table_size = s->cluster_size / sizeof(uint64_t);
4696 s->max_refcount_table_index = 0;
4698 g_free(s->refcount_table);
4699 s->refcount_table = new_reftable;
4700 new_reftable = NULL;
4702 /* Now the in-memory refcount information again corresponds to the on-disk
4703 * information (reftable is empty and no refblocks (the refblock cache is
4704 * empty)); however, this means some clusters (e.g. the image header) are
4705 * referenced, but not refcounted, but the normal qcow2 code assumes that
4706 * the in-memory information is always correct */
4708 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
4710 /* Enter the first refblock into the reftable */
4711 rt_entry = cpu_to_be64(2 * s->cluster_size);
4712 ret = bdrv_pwrite_sync(bs->file, s->cluster_size,
4713 &rt_entry, sizeof(rt_entry));
4714 if (ret < 0) {
4715 goto fail_broken_refcounts;
4717 s->refcount_table[0] = 2 * s->cluster_size;
4719 s->free_cluster_index = 0;
4720 assert(3 + l1_clusters <= s->refcount_block_size);
4721 offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2);
4722 if (offset < 0) {
4723 ret = offset;
4724 goto fail_broken_refcounts;
4725 } else if (offset > 0) {
4726 error_report("First cluster in emptied image is in use");
4727 abort();
4730 /* Now finally the in-memory information corresponds to the on-disk
4731 * structures and is correct */
4732 ret = qcow2_mark_clean(bs);
4733 if (ret < 0) {
4734 goto fail;
4737 ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size, false,
4738 PREALLOC_MODE_OFF, 0, &local_err);
4739 if (ret < 0) {
4740 error_report_err(local_err);
4741 goto fail;
4744 return 0;
4746 fail_broken_refcounts:
4747 /* The BDS is unusable at this point. If we wanted to make it usable, we
4748 * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
4749 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
4750 * again. However, because the functions which could have caused this error
4751 * path to be taken are used by those functions as well, it's very likely
4752 * that that sequence will fail as well. Therefore, just eject the BDS. */
4753 bs->drv = NULL;
4755 fail:
4756 g_free(new_reftable);
4757 return ret;
4760 static int qcow2_make_empty(BlockDriverState *bs)
4762 BDRVQcow2State *s = bs->opaque;
4763 uint64_t offset, end_offset;
4764 int step = QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size);
4765 int l1_clusters, ret = 0;
4767 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
4769 if (s->qcow_version >= 3 && !s->snapshots && !s->nb_bitmaps &&
4770 3 + l1_clusters <= s->refcount_block_size &&
4771 s->crypt_method_header != QCOW_CRYPT_LUKS &&
4772 !has_data_file(bs)) {
4773 /* The following function only works for qcow2 v3 images (it
4774 * requires the dirty flag) and only as long as there are no
4775 * features that reserve extra clusters (such as snapshots,
4776 * LUKS header, or persistent bitmaps), because it completely
4777 * empties the image. Furthermore, the L1 table and three
4778 * additional clusters (image header, refcount table, one
4779 * refcount block) have to fit inside one refcount block. It
4780 * only resets the image file, i.e. does not work with an
4781 * external data file. */
4782 return make_completely_empty(bs);
4785 /* This fallback code simply discards every active cluster; this is slow,
4786 * but works in all cases */
4787 end_offset = bs->total_sectors * BDRV_SECTOR_SIZE;
4788 for (offset = 0; offset < end_offset; offset += step) {
4789 /* As this function is generally used after committing an external
4790 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
4791 * default action for this kind of discard is to pass the discard,
4792 * which will ideally result in an actually smaller image file, as
4793 * is probably desired. */
4794 ret = qcow2_cluster_discard(bs, offset, MIN(step, end_offset - offset),
4795 QCOW2_DISCARD_SNAPSHOT, true);
4796 if (ret < 0) {
4797 break;
4801 return ret;
4804 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
4806 BDRVQcow2State *s = bs->opaque;
4807 int ret;
4809 qemu_co_mutex_lock(&s->lock);
4810 ret = qcow2_write_caches(bs);
4811 qemu_co_mutex_unlock(&s->lock);
4813 return ret;
4816 static BlockMeasureInfo *qcow2_measure(QemuOpts *opts, BlockDriverState *in_bs,
4817 Error **errp)
4819 Error *local_err = NULL;
4820 BlockMeasureInfo *info;
4821 uint64_t required = 0; /* bytes that contribute to required size */
4822 uint64_t virtual_size; /* disk size as seen by guest */
4823 uint64_t refcount_bits;
4824 uint64_t l2_tables;
4825 uint64_t luks_payload_size = 0;
4826 size_t cluster_size;
4827 int version;
4828 char *optstr;
4829 PreallocMode prealloc;
4830 bool has_backing_file;
4831 bool has_luks;
4833 /* Parse image creation options */
4834 cluster_size = qcow2_opt_get_cluster_size_del(opts, &local_err);
4835 if (local_err) {
4836 goto err;
4839 version = qcow2_opt_get_version_del(opts, &local_err);
4840 if (local_err) {
4841 goto err;
4844 refcount_bits = qcow2_opt_get_refcount_bits_del(opts, version, &local_err);
4845 if (local_err) {
4846 goto err;
4849 optstr = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
4850 prealloc = qapi_enum_parse(&PreallocMode_lookup, optstr,
4851 PREALLOC_MODE_OFF, &local_err);
4852 g_free(optstr);
4853 if (local_err) {
4854 goto err;
4857 optstr = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
4858 has_backing_file = !!optstr;
4859 g_free(optstr);
4861 optstr = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT);
4862 has_luks = optstr && strcmp(optstr, "luks") == 0;
4863 g_free(optstr);
4865 if (has_luks) {
4866 g_autoptr(QCryptoBlockCreateOptions) create_opts = NULL;
4867 QDict *cryptoopts = qcow2_extract_crypto_opts(opts, "luks", errp);
4868 size_t headerlen;
4870 create_opts = block_crypto_create_opts_init(cryptoopts, errp);
4871 qobject_unref(cryptoopts);
4872 if (!create_opts) {
4873 goto err;
4876 if (!qcrypto_block_calculate_payload_offset(create_opts,
4877 "encrypt.",
4878 &headerlen,
4879 &local_err)) {
4880 goto err;
4883 luks_payload_size = ROUND_UP(headerlen, cluster_size);
4886 virtual_size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
4887 virtual_size = ROUND_UP(virtual_size, cluster_size);
4889 /* Check that virtual disk size is valid */
4890 l2_tables = DIV_ROUND_UP(virtual_size / cluster_size,
4891 cluster_size / sizeof(uint64_t));
4892 if (l2_tables * sizeof(uint64_t) > QCOW_MAX_L1_SIZE) {
4893 error_setg(&local_err, "The image size is too large "
4894 "(try using a larger cluster size)");
4895 goto err;
4898 /* Account for input image */
4899 if (in_bs) {
4900 int64_t ssize = bdrv_getlength(in_bs);
4901 if (ssize < 0) {
4902 error_setg_errno(&local_err, -ssize,
4903 "Unable to get image virtual_size");
4904 goto err;
4907 virtual_size = ROUND_UP(ssize, cluster_size);
4909 if (has_backing_file) {
4910 /* We don't how much of the backing chain is shared by the input
4911 * image and the new image file. In the worst case the new image's
4912 * backing file has nothing in common with the input image. Be
4913 * conservative and assume all clusters need to be written.
4915 required = virtual_size;
4916 } else {
4917 int64_t offset;
4918 int64_t pnum = 0;
4920 for (offset = 0; offset < ssize; offset += pnum) {
4921 int ret;
4923 ret = bdrv_block_status_above(in_bs, NULL, offset,
4924 ssize - offset, &pnum, NULL,
4925 NULL);
4926 if (ret < 0) {
4927 error_setg_errno(&local_err, -ret,
4928 "Unable to get block status");
4929 goto err;
4932 if (ret & BDRV_BLOCK_ZERO) {
4933 /* Skip zero regions (safe with no backing file) */
4934 } else if ((ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) ==
4935 (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) {
4936 /* Extend pnum to end of cluster for next iteration */
4937 pnum = ROUND_UP(offset + pnum, cluster_size) - offset;
4939 /* Count clusters we've seen */
4940 required += offset % cluster_size + pnum;
4946 /* Take into account preallocation. Nothing special is needed for
4947 * PREALLOC_MODE_METADATA since metadata is always counted.
4949 if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
4950 required = virtual_size;
4953 info = g_new0(BlockMeasureInfo, 1);
4954 info->fully_allocated =
4955 qcow2_calc_prealloc_size(virtual_size, cluster_size,
4956 ctz32(refcount_bits)) + luks_payload_size;
4959 * Remove data clusters that are not required. This overestimates the
4960 * required size because metadata needed for the fully allocated file is
4961 * still counted. Show bitmaps only if both source and destination
4962 * would support them.
4964 info->required = info->fully_allocated - virtual_size + required;
4965 info->has_bitmaps = version >= 3 && in_bs &&
4966 bdrv_supports_persistent_dirty_bitmap(in_bs);
4967 if (info->has_bitmaps) {
4968 info->bitmaps = qcow2_get_persistent_dirty_bitmap_size(in_bs,
4969 cluster_size);
4971 return info;
4973 err:
4974 error_propagate(errp, local_err);
4975 return NULL;
4978 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
4980 BDRVQcow2State *s = bs->opaque;
4981 bdi->cluster_size = s->cluster_size;
4982 bdi->vm_state_offset = qcow2_vm_state_offset(s);
4983 return 0;
4986 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs,
4987 Error **errp)
4989 BDRVQcow2State *s = bs->opaque;
4990 ImageInfoSpecific *spec_info;
4991 QCryptoBlockInfo *encrypt_info = NULL;
4992 Error *local_err = NULL;
4994 if (s->crypto != NULL) {
4995 encrypt_info = qcrypto_block_get_info(s->crypto, &local_err);
4996 if (local_err) {
4997 error_propagate(errp, local_err);
4998 return NULL;
5002 spec_info = g_new(ImageInfoSpecific, 1);
5003 *spec_info = (ImageInfoSpecific){
5004 .type = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
5005 .u.qcow2.data = g_new0(ImageInfoSpecificQCow2, 1),
5007 if (s->qcow_version == 2) {
5008 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
5009 .compat = g_strdup("0.10"),
5010 .refcount_bits = s->refcount_bits,
5012 } else if (s->qcow_version == 3) {
5013 Qcow2BitmapInfoList *bitmaps;
5014 bitmaps = qcow2_get_bitmap_info_list(bs, &local_err);
5015 if (local_err) {
5016 error_propagate(errp, local_err);
5017 qapi_free_ImageInfoSpecific(spec_info);
5018 qapi_free_QCryptoBlockInfo(encrypt_info);
5019 return NULL;
5021 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
5022 .compat = g_strdup("1.1"),
5023 .lazy_refcounts = s->compatible_features &
5024 QCOW2_COMPAT_LAZY_REFCOUNTS,
5025 .has_lazy_refcounts = true,
5026 .corrupt = s->incompatible_features &
5027 QCOW2_INCOMPAT_CORRUPT,
5028 .has_corrupt = true,
5029 .refcount_bits = s->refcount_bits,
5030 .has_bitmaps = !!bitmaps,
5031 .bitmaps = bitmaps,
5032 .has_data_file = !!s->image_data_file,
5033 .data_file = g_strdup(s->image_data_file),
5034 .has_data_file_raw = has_data_file(bs),
5035 .data_file_raw = data_file_is_raw(bs),
5036 .compression_type = s->compression_type,
5038 } else {
5039 /* if this assertion fails, this probably means a new version was
5040 * added without having it covered here */
5041 assert(false);
5044 if (encrypt_info) {
5045 ImageInfoSpecificQCow2Encryption *qencrypt =
5046 g_new(ImageInfoSpecificQCow2Encryption, 1);
5047 switch (encrypt_info->format) {
5048 case Q_CRYPTO_BLOCK_FORMAT_QCOW:
5049 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES;
5050 break;
5051 case Q_CRYPTO_BLOCK_FORMAT_LUKS:
5052 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS;
5053 qencrypt->u.luks = encrypt_info->u.luks;
5054 break;
5055 default:
5056 abort();
5058 /* Since we did shallow copy above, erase any pointers
5059 * in the original info */
5060 memset(&encrypt_info->u, 0, sizeof(encrypt_info->u));
5061 qapi_free_QCryptoBlockInfo(encrypt_info);
5063 spec_info->u.qcow2.data->has_encrypt = true;
5064 spec_info->u.qcow2.data->encrypt = qencrypt;
5067 return spec_info;
5070 static int qcow2_has_zero_init(BlockDriverState *bs)
5072 BDRVQcow2State *s = bs->opaque;
5073 bool preallocated;
5075 if (qemu_in_coroutine()) {
5076 qemu_co_mutex_lock(&s->lock);
5079 * Check preallocation status: Preallocated images have all L2
5080 * tables allocated, nonpreallocated images have none. It is
5081 * therefore enough to check the first one.
5083 preallocated = s->l1_size > 0 && s->l1_table[0] != 0;
5084 if (qemu_in_coroutine()) {
5085 qemu_co_mutex_unlock(&s->lock);
5088 if (!preallocated) {
5089 return 1;
5090 } else if (bs->encrypted) {
5091 return 0;
5092 } else {
5093 return bdrv_has_zero_init(s->data_file->bs);
5097 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
5098 int64_t pos)
5100 BDRVQcow2State *s = bs->opaque;
5102 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
5103 return bs->drv->bdrv_co_pwritev_part(bs, qcow2_vm_state_offset(s) + pos,
5104 qiov->size, qiov, 0, 0);
5107 static int qcow2_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
5108 int64_t pos)
5110 BDRVQcow2State *s = bs->opaque;
5112 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
5113 return bs->drv->bdrv_co_preadv_part(bs, qcow2_vm_state_offset(s) + pos,
5114 qiov->size, qiov, 0, 0);
5118 * Downgrades an image's version. To achieve this, any incompatible features
5119 * have to be removed.
5121 static int qcow2_downgrade(BlockDriverState *bs, int target_version,
5122 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
5123 Error **errp)
5125 BDRVQcow2State *s = bs->opaque;
5126 int current_version = s->qcow_version;
5127 int ret;
5128 int i;
5130 /* This is qcow2_downgrade(), not qcow2_upgrade() */
5131 assert(target_version < current_version);
5133 /* There are no other versions (now) that you can downgrade to */
5134 assert(target_version == 2);
5136 if (s->refcount_order != 4) {
5137 error_setg(errp, "compat=0.10 requires refcount_bits=16");
5138 return -ENOTSUP;
5141 if (has_data_file(bs)) {
5142 error_setg(errp, "Cannot downgrade an image with a data file");
5143 return -ENOTSUP;
5147 * If any internal snapshot has a different size than the current
5148 * image size, or VM state size that exceeds 32 bits, downgrading
5149 * is unsafe. Even though we would still use v3-compliant output
5150 * to preserve that data, other v2 programs might not realize
5151 * those optional fields are important.
5153 for (i = 0; i < s->nb_snapshots; i++) {
5154 if (s->snapshots[i].vm_state_size > UINT32_MAX ||
5155 s->snapshots[i].disk_size != bs->total_sectors * BDRV_SECTOR_SIZE) {
5156 error_setg(errp, "Internal snapshots prevent downgrade of image");
5157 return -ENOTSUP;
5161 /* clear incompatible features */
5162 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
5163 ret = qcow2_mark_clean(bs);
5164 if (ret < 0) {
5165 error_setg_errno(errp, -ret, "Failed to make the image clean");
5166 return ret;
5170 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
5171 * the first place; if that happens nonetheless, returning -ENOTSUP is the
5172 * best thing to do anyway */
5174 if (s->incompatible_features) {
5175 error_setg(errp, "Cannot downgrade an image with incompatible features "
5176 "%#" PRIx64 " set", s->incompatible_features);
5177 return -ENOTSUP;
5180 /* since we can ignore compatible features, we can set them to 0 as well */
5181 s->compatible_features = 0;
5182 /* if lazy refcounts have been used, they have already been fixed through
5183 * clearing the dirty flag */
5185 /* clearing autoclear features is trivial */
5186 s->autoclear_features = 0;
5188 ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque);
5189 if (ret < 0) {
5190 error_setg_errno(errp, -ret, "Failed to turn zero into data clusters");
5191 return ret;
5194 s->qcow_version = target_version;
5195 ret = qcow2_update_header(bs);
5196 if (ret < 0) {
5197 s->qcow_version = current_version;
5198 error_setg_errno(errp, -ret, "Failed to update the image header");
5199 return ret;
5201 return 0;
5205 * Upgrades an image's version. While newer versions encompass all
5206 * features of older versions, some things may have to be presented
5207 * differently.
5209 static int qcow2_upgrade(BlockDriverState *bs, int target_version,
5210 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
5211 Error **errp)
5213 BDRVQcow2State *s = bs->opaque;
5214 bool need_snapshot_update;
5215 int current_version = s->qcow_version;
5216 int i;
5217 int ret;
5219 /* This is qcow2_upgrade(), not qcow2_downgrade() */
5220 assert(target_version > current_version);
5222 /* There are no other versions (yet) that you can upgrade to */
5223 assert(target_version == 3);
5225 status_cb(bs, 0, 2, cb_opaque);
5228 * In v2, snapshots do not need to have extra data. v3 requires
5229 * the 64-bit VM state size and the virtual disk size to be
5230 * present.
5231 * qcow2_write_snapshots() will always write the list in the
5232 * v3-compliant format.
5234 need_snapshot_update = false;
5235 for (i = 0; i < s->nb_snapshots; i++) {
5236 if (s->snapshots[i].extra_data_size <
5237 sizeof_field(QCowSnapshotExtraData, vm_state_size_large) +
5238 sizeof_field(QCowSnapshotExtraData, disk_size))
5240 need_snapshot_update = true;
5241 break;
5244 if (need_snapshot_update) {
5245 ret = qcow2_write_snapshots(bs);
5246 if (ret < 0) {
5247 error_setg_errno(errp, -ret, "Failed to update the snapshot table");
5248 return ret;
5251 status_cb(bs, 1, 2, cb_opaque);
5253 s->qcow_version = target_version;
5254 ret = qcow2_update_header(bs);
5255 if (ret < 0) {
5256 s->qcow_version = current_version;
5257 error_setg_errno(errp, -ret, "Failed to update the image header");
5258 return ret;
5260 status_cb(bs, 2, 2, cb_opaque);
5262 return 0;
5265 typedef enum Qcow2AmendOperation {
5266 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
5267 * statically initialized to so that the helper CB can discern the first
5268 * invocation from an operation change */
5269 QCOW2_NO_OPERATION = 0,
5271 QCOW2_UPGRADING,
5272 QCOW2_UPDATING_ENCRYPTION,
5273 QCOW2_CHANGING_REFCOUNT_ORDER,
5274 QCOW2_DOWNGRADING,
5275 } Qcow2AmendOperation;
5277 typedef struct Qcow2AmendHelperCBInfo {
5278 /* The code coordinating the amend operations should only modify
5279 * these four fields; the rest will be managed by the CB */
5280 BlockDriverAmendStatusCB *original_status_cb;
5281 void *original_cb_opaque;
5283 Qcow2AmendOperation current_operation;
5285 /* Total number of operations to perform (only set once) */
5286 int total_operations;
5288 /* The following fields are managed by the CB */
5290 /* Number of operations completed */
5291 int operations_completed;
5293 /* Cumulative offset of all completed operations */
5294 int64_t offset_completed;
5296 Qcow2AmendOperation last_operation;
5297 int64_t last_work_size;
5298 } Qcow2AmendHelperCBInfo;
5300 static void qcow2_amend_helper_cb(BlockDriverState *bs,
5301 int64_t operation_offset,
5302 int64_t operation_work_size, void *opaque)
5304 Qcow2AmendHelperCBInfo *info = opaque;
5305 int64_t current_work_size;
5306 int64_t projected_work_size;
5308 if (info->current_operation != info->last_operation) {
5309 if (info->last_operation != QCOW2_NO_OPERATION) {
5310 info->offset_completed += info->last_work_size;
5311 info->operations_completed++;
5314 info->last_operation = info->current_operation;
5317 assert(info->total_operations > 0);
5318 assert(info->operations_completed < info->total_operations);
5320 info->last_work_size = operation_work_size;
5322 current_work_size = info->offset_completed + operation_work_size;
5324 /* current_work_size is the total work size for (operations_completed + 1)
5325 * operations (which includes this one), so multiply it by the number of
5326 * operations not covered and divide it by the number of operations
5327 * covered to get a projection for the operations not covered */
5328 projected_work_size = current_work_size * (info->total_operations -
5329 info->operations_completed - 1)
5330 / (info->operations_completed + 1);
5332 info->original_status_cb(bs, info->offset_completed + operation_offset,
5333 current_work_size + projected_work_size,
5334 info->original_cb_opaque);
5337 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,
5338 BlockDriverAmendStatusCB *status_cb,
5339 void *cb_opaque,
5340 bool force,
5341 Error **errp)
5343 BDRVQcow2State *s = bs->opaque;
5344 int old_version = s->qcow_version, new_version = old_version;
5345 uint64_t new_size = 0;
5346 const char *backing_file = NULL, *backing_format = NULL, *data_file = NULL;
5347 bool lazy_refcounts = s->use_lazy_refcounts;
5348 bool data_file_raw = data_file_is_raw(bs);
5349 const char *compat = NULL;
5350 int refcount_bits = s->refcount_bits;
5351 int ret;
5352 QemuOptDesc *desc = opts->list->desc;
5353 Qcow2AmendHelperCBInfo helper_cb_info;
5354 bool encryption_update = false;
5356 while (desc && desc->name) {
5357 if (!qemu_opt_find(opts, desc->name)) {
5358 /* only change explicitly defined options */
5359 desc++;
5360 continue;
5363 if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) {
5364 compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL);
5365 if (!compat) {
5366 /* preserve default */
5367 } else if (!strcmp(compat, "0.10") || !strcmp(compat, "v2")) {
5368 new_version = 2;
5369 } else if (!strcmp(compat, "1.1") || !strcmp(compat, "v3")) {
5370 new_version = 3;
5371 } else {
5372 error_setg(errp, "Unknown compatibility level %s", compat);
5373 return -EINVAL;
5375 } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) {
5376 new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
5377 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) {
5378 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
5379 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) {
5380 backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
5381 } else if (g_str_has_prefix(desc->name, "encrypt.")) {
5382 if (!s->crypto) {
5383 error_setg(errp,
5384 "Can't amend encryption options - encryption not present");
5385 return -EINVAL;
5387 if (s->crypt_method_header != QCOW_CRYPT_LUKS) {
5388 error_setg(errp,
5389 "Only LUKS encryption options can be amended");
5390 return -ENOTSUP;
5392 encryption_update = true;
5393 } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
5394 lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS,
5395 lazy_refcounts);
5396 } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) {
5397 refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS,
5398 refcount_bits);
5400 if (refcount_bits <= 0 || refcount_bits > 64 ||
5401 !is_power_of_2(refcount_bits))
5403 error_setg(errp, "Refcount width must be a power of two and "
5404 "may not exceed 64 bits");
5405 return -EINVAL;
5407 } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE)) {
5408 data_file = qemu_opt_get(opts, BLOCK_OPT_DATA_FILE);
5409 if (data_file && !has_data_file(bs)) {
5410 error_setg(errp, "data-file can only be set for images that "
5411 "use an external data file");
5412 return -EINVAL;
5414 } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE_RAW)) {
5415 data_file_raw = qemu_opt_get_bool(opts, BLOCK_OPT_DATA_FILE_RAW,
5416 data_file_raw);
5417 if (data_file_raw && !data_file_is_raw(bs)) {
5418 error_setg(errp, "data-file-raw cannot be set on existing "
5419 "images");
5420 return -EINVAL;
5422 } else {
5423 /* if this point is reached, this probably means a new option was
5424 * added without having it covered here */
5425 abort();
5428 desc++;
5431 helper_cb_info = (Qcow2AmendHelperCBInfo){
5432 .original_status_cb = status_cb,
5433 .original_cb_opaque = cb_opaque,
5434 .total_operations = (new_version != old_version)
5435 + (s->refcount_bits != refcount_bits) +
5436 (encryption_update == true)
5439 /* Upgrade first (some features may require compat=1.1) */
5440 if (new_version > old_version) {
5441 helper_cb_info.current_operation = QCOW2_UPGRADING;
5442 ret = qcow2_upgrade(bs, new_version, &qcow2_amend_helper_cb,
5443 &helper_cb_info, errp);
5444 if (ret < 0) {
5445 return ret;
5449 if (encryption_update) {
5450 QDict *amend_opts_dict;
5451 QCryptoBlockAmendOptions *amend_opts;
5453 helper_cb_info.current_operation = QCOW2_UPDATING_ENCRYPTION;
5454 amend_opts_dict = qcow2_extract_crypto_opts(opts, "luks", errp);
5455 if (!amend_opts_dict) {
5456 return -EINVAL;
5458 amend_opts = block_crypto_amend_opts_init(amend_opts_dict, errp);
5459 qobject_unref(amend_opts_dict);
5460 if (!amend_opts) {
5461 return -EINVAL;
5463 ret = qcrypto_block_amend_options(s->crypto,
5464 qcow2_crypto_hdr_read_func,
5465 qcow2_crypto_hdr_write_func,
5467 amend_opts,
5468 force,
5469 errp);
5470 qapi_free_QCryptoBlockAmendOptions(amend_opts);
5471 if (ret < 0) {
5472 return ret;
5476 if (s->refcount_bits != refcount_bits) {
5477 int refcount_order = ctz32(refcount_bits);
5479 if (new_version < 3 && refcount_bits != 16) {
5480 error_setg(errp, "Refcount widths other than 16 bits require "
5481 "compatibility level 1.1 or above (use compat=1.1 or "
5482 "greater)");
5483 return -EINVAL;
5486 helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER;
5487 ret = qcow2_change_refcount_order(bs, refcount_order,
5488 &qcow2_amend_helper_cb,
5489 &helper_cb_info, errp);
5490 if (ret < 0) {
5491 return ret;
5495 /* data-file-raw blocks backing files, so clear it first if requested */
5496 if (data_file_raw) {
5497 s->autoclear_features |= QCOW2_AUTOCLEAR_DATA_FILE_RAW;
5498 } else {
5499 s->autoclear_features &= ~QCOW2_AUTOCLEAR_DATA_FILE_RAW;
5502 if (data_file) {
5503 g_free(s->image_data_file);
5504 s->image_data_file = *data_file ? g_strdup(data_file) : NULL;
5507 ret = qcow2_update_header(bs);
5508 if (ret < 0) {
5509 error_setg_errno(errp, -ret, "Failed to update the image header");
5510 return ret;
5513 if (backing_file || backing_format) {
5514 ret = qcow2_change_backing_file(bs,
5515 backing_file ?: s->image_backing_file,
5516 backing_format ?: s->image_backing_format);
5517 if (ret < 0) {
5518 error_setg_errno(errp, -ret, "Failed to change the backing file");
5519 return ret;
5523 if (s->use_lazy_refcounts != lazy_refcounts) {
5524 if (lazy_refcounts) {
5525 if (new_version < 3) {
5526 error_setg(errp, "Lazy refcounts only supported with "
5527 "compatibility level 1.1 and above (use compat=1.1 "
5528 "or greater)");
5529 return -EINVAL;
5531 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
5532 ret = qcow2_update_header(bs);
5533 if (ret < 0) {
5534 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
5535 error_setg_errno(errp, -ret, "Failed to update the image header");
5536 return ret;
5538 s->use_lazy_refcounts = true;
5539 } else {
5540 /* make image clean first */
5541 ret = qcow2_mark_clean(bs);
5542 if (ret < 0) {
5543 error_setg_errno(errp, -ret, "Failed to make the image clean");
5544 return ret;
5546 /* now disallow lazy refcounts */
5547 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
5548 ret = qcow2_update_header(bs);
5549 if (ret < 0) {
5550 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
5551 error_setg_errno(errp, -ret, "Failed to update the image header");
5552 return ret;
5554 s->use_lazy_refcounts = false;
5558 if (new_size) {
5559 BlockBackend *blk = blk_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL,
5560 errp);
5561 if (!blk) {
5562 return -EPERM;
5566 * Amending image options should ensure that the image has
5567 * exactly the given new values, so pass exact=true here.
5569 ret = blk_truncate(blk, new_size, true, PREALLOC_MODE_OFF, 0, errp);
5570 blk_unref(blk);
5571 if (ret < 0) {
5572 return ret;
5576 /* Downgrade last (so unsupported features can be removed before) */
5577 if (new_version < old_version) {
5578 helper_cb_info.current_operation = QCOW2_DOWNGRADING;
5579 ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb,
5580 &helper_cb_info, errp);
5581 if (ret < 0) {
5582 return ret;
5586 return 0;
5589 static int coroutine_fn qcow2_co_amend(BlockDriverState *bs,
5590 BlockdevAmendOptions *opts,
5591 bool force,
5592 Error **errp)
5594 BlockdevAmendOptionsQcow2 *qopts = &opts->u.qcow2;
5595 BDRVQcow2State *s = bs->opaque;
5596 int ret = 0;
5598 if (qopts->has_encrypt) {
5599 if (!s->crypto) {
5600 error_setg(errp, "image is not encrypted, can't amend");
5601 return -EOPNOTSUPP;
5604 if (qopts->encrypt->format != Q_CRYPTO_BLOCK_FORMAT_LUKS) {
5605 error_setg(errp,
5606 "Amend can't be used to change the qcow2 encryption format");
5607 return -EOPNOTSUPP;
5610 if (s->crypt_method_header != QCOW_CRYPT_LUKS) {
5611 error_setg(errp,
5612 "Only LUKS encryption options can be amended for qcow2 with blockdev-amend");
5613 return -EOPNOTSUPP;
5616 ret = qcrypto_block_amend_options(s->crypto,
5617 qcow2_crypto_hdr_read_func,
5618 qcow2_crypto_hdr_write_func,
5620 qopts->encrypt,
5621 force,
5622 errp);
5624 return ret;
5628 * If offset or size are negative, respectively, they will not be included in
5629 * the BLOCK_IMAGE_CORRUPTED event emitted.
5630 * fatal will be ignored for read-only BDS; corruptions found there will always
5631 * be considered non-fatal.
5633 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
5634 int64_t size, const char *message_format, ...)
5636 BDRVQcow2State *s = bs->opaque;
5637 const char *node_name;
5638 char *message;
5639 va_list ap;
5641 fatal = fatal && bdrv_is_writable(bs);
5643 if (s->signaled_corruption &&
5644 (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
5646 return;
5649 va_start(ap, message_format);
5650 message = g_strdup_vprintf(message_format, ap);
5651 va_end(ap);
5653 if (fatal) {
5654 fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
5655 "corruption events will be suppressed\n", message);
5656 } else {
5657 fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
5658 "corruption events will be suppressed\n", message);
5661 node_name = bdrv_get_node_name(bs);
5662 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs),
5663 *node_name != '\0', node_name,
5664 message, offset >= 0, offset,
5665 size >= 0, size,
5666 fatal);
5667 g_free(message);
5669 if (fatal) {
5670 qcow2_mark_corrupt(bs);
5671 bs->drv = NULL; /* make BDS unusable */
5674 s->signaled_corruption = true;
5677 #define QCOW_COMMON_OPTIONS \
5679 .name = BLOCK_OPT_SIZE, \
5680 .type = QEMU_OPT_SIZE, \
5681 .help = "Virtual disk size" \
5682 }, \
5684 .name = BLOCK_OPT_COMPAT_LEVEL, \
5685 .type = QEMU_OPT_STRING, \
5686 .help = "Compatibility level (v2 [0.10] or v3 [1.1])" \
5687 }, \
5689 .name = BLOCK_OPT_BACKING_FILE, \
5690 .type = QEMU_OPT_STRING, \
5691 .help = "File name of a base image" \
5692 }, \
5694 .name = BLOCK_OPT_BACKING_FMT, \
5695 .type = QEMU_OPT_STRING, \
5696 .help = "Image format of the base image" \
5697 }, \
5699 .name = BLOCK_OPT_DATA_FILE, \
5700 .type = QEMU_OPT_STRING, \
5701 .help = "File name of an external data file" \
5702 }, \
5704 .name = BLOCK_OPT_DATA_FILE_RAW, \
5705 .type = QEMU_OPT_BOOL, \
5706 .help = "The external data file must stay valid " \
5707 "as a raw image" \
5708 }, \
5710 .name = BLOCK_OPT_LAZY_REFCOUNTS, \
5711 .type = QEMU_OPT_BOOL, \
5712 .help = "Postpone refcount updates", \
5713 .def_value_str = "off" \
5714 }, \
5716 .name = BLOCK_OPT_REFCOUNT_BITS, \
5717 .type = QEMU_OPT_NUMBER, \
5718 .help = "Width of a reference count entry in bits", \
5719 .def_value_str = "16" \
5722 static QemuOptsList qcow2_create_opts = {
5723 .name = "qcow2-create-opts",
5724 .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
5725 .desc = {
5727 .name = BLOCK_OPT_ENCRYPT, \
5728 .type = QEMU_OPT_BOOL, \
5729 .help = "Encrypt the image with format 'aes'. (Deprecated " \
5730 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)", \
5731 }, \
5733 .name = BLOCK_OPT_ENCRYPT_FORMAT, \
5734 .type = QEMU_OPT_STRING, \
5735 .help = "Encrypt the image, format choices: 'aes', 'luks'", \
5736 }, \
5737 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", \
5738 "ID of secret providing qcow AES key or LUKS passphrase"), \
5739 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."), \
5740 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."), \
5741 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."), \
5742 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."), \
5743 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."), \
5744 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."), \
5746 .name = BLOCK_OPT_CLUSTER_SIZE, \
5747 .type = QEMU_OPT_SIZE, \
5748 .help = "qcow2 cluster size", \
5749 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE) \
5750 }, \
5752 .name = BLOCK_OPT_PREALLOC, \
5753 .type = QEMU_OPT_STRING, \
5754 .help = "Preallocation mode (allowed values: off, " \
5755 "metadata, falloc, full)" \
5756 }, \
5758 .name = BLOCK_OPT_COMPRESSION_TYPE, \
5759 .type = QEMU_OPT_STRING, \
5760 .help = "Compression method used for image cluster " \
5761 "compression", \
5762 .def_value_str = "zlib" \
5764 QCOW_COMMON_OPTIONS,
5765 { /* end of list */ }
5769 static QemuOptsList qcow2_amend_opts = {
5770 .name = "qcow2-amend-opts",
5771 .head = QTAILQ_HEAD_INITIALIZER(qcow2_amend_opts.head),
5772 .desc = {
5773 BLOCK_CRYPTO_OPT_DEF_LUKS_STATE("encrypt."),
5774 BLOCK_CRYPTO_OPT_DEF_LUKS_KEYSLOT("encrypt."),
5775 BLOCK_CRYPTO_OPT_DEF_LUKS_OLD_SECRET("encrypt."),
5776 BLOCK_CRYPTO_OPT_DEF_LUKS_NEW_SECRET("encrypt."),
5777 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
5778 QCOW_COMMON_OPTIONS,
5779 { /* end of list */ }
5783 static const char *const qcow2_strong_runtime_opts[] = {
5784 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET,
5786 NULL
5789 BlockDriver bdrv_qcow2 = {
5790 .format_name = "qcow2",
5791 .instance_size = sizeof(BDRVQcow2State),
5792 .bdrv_probe = qcow2_probe,
5793 .bdrv_open = qcow2_open,
5794 .bdrv_close = qcow2_close,
5795 .bdrv_reopen_prepare = qcow2_reopen_prepare,
5796 .bdrv_reopen_commit = qcow2_reopen_commit,
5797 .bdrv_reopen_commit_post = qcow2_reopen_commit_post,
5798 .bdrv_reopen_abort = qcow2_reopen_abort,
5799 .bdrv_join_options = qcow2_join_options,
5800 .bdrv_child_perm = bdrv_default_perms,
5801 .bdrv_co_create_opts = qcow2_co_create_opts,
5802 .bdrv_co_create = qcow2_co_create,
5803 .bdrv_has_zero_init = qcow2_has_zero_init,
5804 .bdrv_co_block_status = qcow2_co_block_status,
5806 .bdrv_co_preadv_part = qcow2_co_preadv_part,
5807 .bdrv_co_pwritev_part = qcow2_co_pwritev_part,
5808 .bdrv_co_flush_to_os = qcow2_co_flush_to_os,
5810 .bdrv_co_pwrite_zeroes = qcow2_co_pwrite_zeroes,
5811 .bdrv_co_pdiscard = qcow2_co_pdiscard,
5812 .bdrv_co_copy_range_from = qcow2_co_copy_range_from,
5813 .bdrv_co_copy_range_to = qcow2_co_copy_range_to,
5814 .bdrv_co_truncate = qcow2_co_truncate,
5815 .bdrv_co_pwritev_compressed_part = qcow2_co_pwritev_compressed_part,
5816 .bdrv_make_empty = qcow2_make_empty,
5818 .bdrv_snapshot_create = qcow2_snapshot_create,
5819 .bdrv_snapshot_goto = qcow2_snapshot_goto,
5820 .bdrv_snapshot_delete = qcow2_snapshot_delete,
5821 .bdrv_snapshot_list = qcow2_snapshot_list,
5822 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
5823 .bdrv_measure = qcow2_measure,
5824 .bdrv_get_info = qcow2_get_info,
5825 .bdrv_get_specific_info = qcow2_get_specific_info,
5827 .bdrv_save_vmstate = qcow2_save_vmstate,
5828 .bdrv_load_vmstate = qcow2_load_vmstate,
5830 .is_format = true,
5831 .supports_backing = true,
5832 .bdrv_change_backing_file = qcow2_change_backing_file,
5834 .bdrv_refresh_limits = qcow2_refresh_limits,
5835 .bdrv_co_invalidate_cache = qcow2_co_invalidate_cache,
5836 .bdrv_inactivate = qcow2_inactivate,
5838 .create_opts = &qcow2_create_opts,
5839 .amend_opts = &qcow2_amend_opts,
5840 .strong_runtime_opts = qcow2_strong_runtime_opts,
5841 .mutable_opts = mutable_opts,
5842 .bdrv_co_check = qcow2_co_check,
5843 .bdrv_amend_options = qcow2_amend_options,
5844 .bdrv_co_amend = qcow2_co_amend,
5846 .bdrv_detach_aio_context = qcow2_detach_aio_context,
5847 .bdrv_attach_aio_context = qcow2_attach_aio_context,
5849 .bdrv_supports_persistent_dirty_bitmap =
5850 qcow2_supports_persistent_dirty_bitmap,
5851 .bdrv_co_can_store_new_dirty_bitmap = qcow2_co_can_store_new_dirty_bitmap,
5852 .bdrv_co_remove_persistent_dirty_bitmap =
5853 qcow2_co_remove_persistent_dirty_bitmap,
5856 static void bdrv_qcow2_init(void)
5858 bdrv_register(&bdrv_qcow2);
5861 block_init(bdrv_qcow2_init);