qcow2: Use macros for the L1, refcount and bitmap table entry sizes
[qemu.git] / block / qcow2.c
1 /*
2 * Block driver for the QCOW version 2 format
3 *
4 * Copyright (c) 2004-2006 Fabrice Bellard
5 *
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:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
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.
23 */
24
25 #include "qemu/osdep.h"
26
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"
45
46 /*
47 Differences with QCOW:
48
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.
60 */
61
62
63 typedef struct {
64 uint32_t magic;
65 uint32_t len;
66 } QEMU_PACKED QCowExtension;
67
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
74
75 static int coroutine_fn
76 qcow2_co_preadv_compressed(BlockDriverState *bs,
77 uint64_t cluster_descriptor,
78 uint64_t offset,
79 uint64_t bytes,
80 QEMUIOVector *qiov,
81 size_t qiov_offset);
82
83 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
84 {
85 const QCowHeader *cow_header = (const void *)buf;
86
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;
93 }
94
95
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)
99 {
100 BlockDriverState *bs = opaque;
101 BDRVQcow2State *s = bs->opaque;
102 ssize_t ret;
103
104 if ((offset + buflen) > s->crypto_header.length) {
105 error_setg(errp, "Request for data outside of extension header");
106 return -1;
107 }
108
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;
114 }
115 return ret;
116 }
117
118
119 static ssize_t qcow2_crypto_hdr_init_func(QCryptoBlock *block, size_t headerlen,
120 void *opaque, Error **errp)
121 {
122 BlockDriverState *bs = opaque;
123 BDRVQcow2State *s = bs->opaque;
124 int64_t ret;
125 int64_t clusterlen;
126
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;
133 }
134
135 s->crypto_header.length = headerlen;
136 s->crypto_header.offset = ret;
137
138 /*
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)
142 */
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;
151 }
152
153 return ret;
154 }
155
156
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)
160 {
161 BlockDriverState *bs = opaque;
162 BDRVQcow2State *s = bs->opaque;
163 ssize_t ret;
164
165 if ((offset + buflen) > s->crypto_header.length) {
166 error_setg(errp, "Request for data outside of extension header");
167 return -1;
168 }
169
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;
175 }
176 return ret;
177 }
178
179 static QDict*
180 qcow2_extract_crypto_opts(QemuOpts *opts, const char *fmt, Error **errp)
181 {
182 QDict *cryptoopts_qdict;
183 QDict *opts_qdict;
184
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;
191 }
192
193 /*
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
199 */
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)
204 {
205 BDRVQcow2State *s = bs->opaque;
206 QCowExtension ext;
207 uint64_t offset;
208 int ret;
209 Qcow2BitmapHeaderExt bitmaps_ext;
210
211 if (need_update_header != NULL) {
212 *need_update_header = false;
213 }
214
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) {
220
221 #ifdef DEBUG_EXT
222 /* Sanity check */
223 if (offset > s->cluster_size)
224 printf("qcow2_read_extension: suspicious offset %lu\n", offset);
225
226 printf("attempting to read extended header in offset %lu\n", offset);
227 #endif
228
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;
234 }
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;
244 }
245
246 switch (ext.magic) {
247 case QCOW2_EXT_MAGIC_END:
248 return 0;
249
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;
256 }
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;
262 }
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;
269
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;
278 }
279
280 *p_feature_table = feature_table;
281 }
282 break;
283
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;
290 }
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;
296 }
297
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;
303 }
304 s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset);
305 s->crypto_header.length = be64_to_cpu(s->crypto_header.length);
306
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;
312 }
313
314 if (flags & BDRV_O_NO_IO) {
315 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
316 }
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;
322 }
323 } break;
324
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;
330 }
331
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");
343 }
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;
350 }
351 break;
352 }
353
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;
359 }
360
361 if (bitmaps_ext.reserved32 != 0) {
362 error_setg_errno(errp, -ret, "bitmaps_ext: "
363 "Reserved field is not zero");
364 return -EINVAL;
365 }
366
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);
372
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;
379 }
380
381 if (bitmaps_ext.nb_bitmaps == 0) {
382 error_setg(errp, "found bitmaps extension with zero bitmaps");
383 return -EINVAL;
384 }
385
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;
390 }
391
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;
400 }
401
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;
407
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;
414
415 case QCOW2_EXT_MAGIC_DATA_FILE:
416 {
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;
423 }
424 #ifdef DEBUG_EXT
425 printf("Qcow2: Got external data file %s\n", s->image_data_file);
426 #endif
427 break;
428 }
429
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. */
434 {
435 Qcow2UnknownHeaderExtension *uext;
436
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);
441
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;
447 }
448 }
449 break;
450 }
451
452 offset += ((ext.len + 7) & ~7);
453 }
454
455 return 0;
456 }
457
458 static void cleanup_unknown_header_ext(BlockDriverState *bs)
459 {
460 BDRVQcow2State *s = bs->opaque;
461 Qcow2UnknownHeaderExtension *uext, *next;
462
463 QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
464 QLIST_REMOVE(uext, next);
465 g_free(uext);
466 }
467 }
468
469 static void report_unsupported_feature(Error **errp, Qcow2Feature *table,
470 uint64_t mask)
471 {
472 g_autoptr(GString) features = g_string_sized_new(60);
473
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, ", ");
479 }
480 g_string_append_printf(features, "%.46s", table->name);
481 mask &= ~(1ULL << table->bit);
482 }
483 }
484 table++;
485 }
486
487 if (mask) {
488 if (features->len > 0) {
489 g_string_append(features, ", ");
490 }
491 g_string_append_printf(features,
492 "Unknown incompatible feature: %" PRIx64, mask);
493 }
494
495 error_setg(errp, "Unsupported qcow2 feature(s): %s", features->str);
496 }
497
498 /*
499 * Sets the dirty bit and flushes afterwards if necessary.
500 *
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.
504 */
505 int qcow2_mark_dirty(BlockDriverState *bs)
506 {
507 BDRVQcow2State *s = bs->opaque;
508 uint64_t val;
509 int ret;
510
511 assert(s->qcow_version >= 3);
512
513 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
514 return 0; /* already dirty */
515 }
516
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;
522 }
523 ret = bdrv_flush(bs->file->bs);
524 if (ret < 0) {
525 return ret;
526 }
527
528 /* Only treat image as dirty if the header was updated successfully */
529 s->incompatible_features |= QCOW2_INCOMPAT_DIRTY;
530 return 0;
531 }
532
533 /*
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.
537 */
538 static int qcow2_mark_clean(BlockDriverState *bs)
539 {
540 BDRVQcow2State *s = bs->opaque;
541
542 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
543 int ret;
544
545 s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY;
546
547 ret = qcow2_flush_caches(bs);
548 if (ret < 0) {
549 return ret;
550 }
551
552 return qcow2_update_header(bs);
553 }
554 return 0;
555 }
556
557 /*
558 * Marks the image as corrupt.
559 */
560 int qcow2_mark_corrupt(BlockDriverState *bs)
561 {
562 BDRVQcow2State *s = bs->opaque;
563
564 s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT;
565 return qcow2_update_header(bs);
566 }
567
568 /*
569 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
570 * before if necessary.
571 */
572 int qcow2_mark_consistent(BlockDriverState *bs)
573 {
574 BDRVQcow2State *s = bs->opaque;
575
576 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
577 int ret = qcow2_flush_caches(bs);
578 if (ret < 0) {
579 return ret;
580 }
581
582 s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT;
583 return qcow2_update_header(bs);
584 }
585 return 0;
586 }
587
588 static void qcow2_add_check_result(BdrvCheckResult *out,
589 const BdrvCheckResult *src,
590 bool set_allocation_info)
591 {
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;
597
598 if (set_allocation_info) {
599 out->image_end_offset = src->image_end_offset;
600 out->bfi = src->bfi;
601 }
602 }
603
604 static int coroutine_fn qcow2_co_check_locked(BlockDriverState *bs,
605 BdrvCheckResult *result,
606 BdrvCheckMode fix)
607 {
608 BdrvCheckResult snapshot_res = {};
609 BdrvCheckResult refcount_res = {};
610 int ret;
611
612 memset(result, 0, sizeof(*result));
613
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;
618 }
619
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;
625 }
626
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;
631 }
632
633 if (fix && result->check_errors == 0 && result->corruptions == 0) {
634 ret = qcow2_mark_clean(bs);
635 if (ret < 0) {
636 return ret;
637 }
638 return qcow2_mark_consistent(bs);
639 }
640 return ret;
641 }
642
643 static int coroutine_fn qcow2_co_check(BlockDriverState *bs,
644 BdrvCheckResult *result,
645 BdrvCheckMode fix)
646 {
647 BDRVQcow2State *s = bs->opaque;
648 int ret;
649
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;
654 }
655
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)
660 {
661 BDRVQcow2State *s = bs->opaque;
662
663 if (entries > max_size_bytes / entry_len) {
664 error_setg(errp, "%s too large", table_name);
665 return -EFBIG;
666 }
667
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;
674 }
675
676 return 0;
677 }
678
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
701 };
702
703 static QemuOptsList qcow2_runtime_opts = {
704 .name = "qcow2",
705 .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head),
706 .desc = {
707 {
708 .name = QCOW2_OPT_LAZY_REFCOUNTS,
709 .type = QEMU_OPT_BOOL,
710 .help = "Postpone refcount updates",
711 },
712 {
713 .name = QCOW2_OPT_DISCARD_REQUEST,
714 .type = QEMU_OPT_BOOL,
715 .help = "Pass guest discard requests to the layer below",
716 },
717 {
718 .name = QCOW2_OPT_DISCARD_SNAPSHOT,
719 .type = QEMU_OPT_BOOL,
720 .help = "Generate discard requests when snapshot related space "
721 "is freed",
722 },
723 {
724 .name = QCOW2_OPT_DISCARD_OTHER,
725 .type = QEMU_OPT_BOOL,
726 .help = "Generate discard requests when other clusters are freed",
727 },
728 {
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)",
733 },
734 {
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)",
739 },
740 {
741 .name = QCOW2_OPT_OVERLAP_MAIN_HEADER,
742 .type = QEMU_OPT_BOOL,
743 .help = "Check for unintended writes into the main qcow2 header",
744 },
745 {
746 .name = QCOW2_OPT_OVERLAP_ACTIVE_L1,
747 .type = QEMU_OPT_BOOL,
748 .help = "Check for unintended writes into the active L1 table",
749 },
750 {
751 .name = QCOW2_OPT_OVERLAP_ACTIVE_L2,
752 .type = QEMU_OPT_BOOL,
753 .help = "Check for unintended writes into an active L2 table",
754 },
755 {
756 .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
757 .type = QEMU_OPT_BOOL,
758 .help = "Check for unintended writes into the refcount table",
759 },
760 {
761 .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
762 .type = QEMU_OPT_BOOL,
763 .help = "Check for unintended writes into a refcount block",
764 },
765 {
766 .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
767 .type = QEMU_OPT_BOOL,
768 .help = "Check for unintended writes into the snapshot table",
769 },
770 {
771 .name = QCOW2_OPT_OVERLAP_INACTIVE_L1,
772 .type = QEMU_OPT_BOOL,
773 .help = "Check for unintended writes into an inactive L1 table",
774 },
775 {
776 .name = QCOW2_OPT_OVERLAP_INACTIVE_L2,
777 .type = QEMU_OPT_BOOL,
778 .help = "Check for unintended writes into an inactive L2 table",
779 },
780 {
781 .name = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY,
782 .type = QEMU_OPT_BOOL,
783 .help = "Check for unintended writes into the bitmap directory",
784 },
785 {
786 .name = QCOW2_OPT_CACHE_SIZE,
787 .type = QEMU_OPT_SIZE,
788 .help = "Maximum combined metadata (L2 tables and refcount blocks) "
789 "cache size",
790 },
791 {
792 .name = QCOW2_OPT_L2_CACHE_SIZE,
793 .type = QEMU_OPT_SIZE,
794 .help = "Maximum L2 table cache size",
795 },
796 {
797 .name = QCOW2_OPT_L2_CACHE_ENTRY_SIZE,
798 .type = QEMU_OPT_SIZE,
799 .help = "Size of each entry in the L2 cache",
800 },
801 {
802 .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE,
803 .type = QEMU_OPT_SIZE,
804 .help = "Maximum refcount block cache size",
805 },
806 {
807 .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL,
808 .type = QEMU_OPT_NUMBER,
809 .help = "Clean unused cache entries after this time (in seconds)",
810 },
811 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
812 "ID of secret providing qcow2 AES key or LUKS passphrase"),
813 { /* end of list */ }
814 },
815 };
816
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,
827 };
828
829 static void cache_clean_timer_cb(void *opaque)
830 {
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);
837 }
838
839 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context)
840 {
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);
848 }
849 }
850
851 static void cache_clean_timer_del(BlockDriverState *bs)
852 {
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;
858 }
859 }
860
861 static void qcow2_detach_aio_context(BlockDriverState *bs)
862 {
863 cache_clean_timer_del(bs);
864 }
865
866 static void qcow2_attach_aio_context(BlockDriverState *bs,
867 AioContext *new_context)
868 {
869 cache_clean_timer_init(bs, new_context);
870 }
871
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)
876 {
877 BDRVQcow2State *s = bs->opaque;
878 uint64_t combined_cache_size, l2_cache_max_setting;
879 bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set;
880 bool l2_cache_entry_size_set;
881 int min_refcount_cache = MIN_REFCOUNT_CACHE_SIZE * s->cluster_size;
882 uint64_t virtual_disk_size = bs->total_sectors * BDRV_SECTOR_SIZE;
883 uint64_t max_l2_entries = DIV_ROUND_UP(virtual_disk_size, s->cluster_size);
884 /* An L2 table is always one cluster in size so the max cache size
885 * should be a multiple of the cluster size. */
886 uint64_t max_l2_cache = ROUND_UP(max_l2_entries * l2_entry_size(s),
887 s->cluster_size);
888
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);
893
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);
899
900 *l2_cache_entry_size = qemu_opt_get_size(
901 opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE, s->cluster_size);
902
903 *l2_cache_size = MIN(max_l2_cache, l2_cache_max_setting);
904
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;
920 }
921
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;
936 }
937 }
938 }
939
940 /*
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.
944 */
945 if (*l2_cache_size < max_l2_cache && !l2_cache_entry_size_set) {
946 *l2_cache_entry_size = MIN(s->cluster_size, 4096);
947 }
948
949 /* l2_cache_size and refcount_cache_size are ensured to have at least
950 * their minimum values in qcow2_update_options_prepare() */
951
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;
959 }
960 }
961
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;
972
973 static int qcow2_update_options_prepare(BlockDriverState *bs,
974 Qcow2ReopenState *r,
975 QDict *options, int flags,
976 Error **errp)
977 {
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;
988
989 qdict_extract_subqdict(options, &encryptopts, "encrypt.");
990 encryptfmt = qdict_get_try_str(encryptopts, "format");
991
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;
996 }
997
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;
1005 }
1006
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;
1010 }
1011 if (l2_cache_size > INT_MAX) {
1012 error_setg(errp, "L2 cache size too big");
1013 ret = -EINVAL;
1014 goto fail;
1015 }
1016
1017 refcount_cache_size /= s->cluster_size;
1018 if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) {
1019 refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE;
1020 }
1021 if (refcount_cache_size > INT_MAX) {
1022 error_setg(errp, "Refcount cache size too big");
1023 ret = -EINVAL;
1024 goto fail;
1025 }
1026
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;
1033 }
1034 }
1035
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;
1042 }
1043 }
1044
1045 r->l2_slice_size = l2_cache_entry_size / l2_entry_size(s);
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;
1054 }
1055
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;
1066 }
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;
1072 }
1073
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;
1082 }
1083
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;
1089 }
1090 }
1091
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))
1097 {
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;
1103 }
1104 if (!opt_overlap_check) {
1105 opt_overlap_check = opt_overlap_check_template ?: "cached";
1106 }
1107
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;
1122 }
1123
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;
1131 }
1132
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);
1142
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;
1150 }
1151 break;
1152
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;
1160 }
1161 qdict_put_str(encryptopts, "format", "qcow");
1162 r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
1163 break;
1164
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;
1172 }
1173 qdict_put_str(encryptopts, "format", "luks");
1174 r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
1175 break;
1176
1177 default:
1178 error_setg(errp, "Unsupported encryption method %d",
1179 s->crypt_method_header);
1180 break;
1181 }
1182 if (s->crypt_method_header != QCOW_CRYPT_NONE && !r->crypto_opts) {
1183 ret = -EINVAL;
1184 goto fail;
1185 }
1186
1187 ret = 0;
1188 fail:
1189 qobject_unref(encryptopts);
1190 qemu_opts_del(opts);
1191 opts = NULL;
1192 return ret;
1193 }
1194
1195 static void qcow2_update_options_commit(BlockDriverState *bs,
1196 Qcow2ReopenState *r)
1197 {
1198 BDRVQcow2State *s = bs->opaque;
1199 int i;
1200
1201 if (s->l2_table_cache) {
1202 qcow2_cache_destroy(s->l2_table_cache);
1203 }
1204 if (s->refcount_block_cache) {
1205 qcow2_cache_destroy(s->refcount_block_cache);
1206 }
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;
1210
1211 s->overlap_check = r->overlap_check;
1212 s->use_lazy_refcounts = r->use_lazy_refcounts;
1213
1214 for (i = 0; i < QCOW2_DISCARD_MAX; i++) {
1215 s->discard_passthrough[i] = r->discard_passthrough[i];
1216 }
1217
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));
1222 }
1223
1224 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1225 s->crypto_opts = r->crypto_opts;
1226 }
1227
1228 static void qcow2_update_options_abort(BlockDriverState *bs,
1229 Qcow2ReopenState *r)
1230 {
1231 if (r->l2_table_cache) {
1232 qcow2_cache_destroy(r->l2_table_cache);
1233 }
1234 if (r->refcount_block_cache) {
1235 qcow2_cache_destroy(r->refcount_block_cache);
1236 }
1237 qapi_free_QCryptoBlockOpenOptions(r->crypto_opts);
1238 }
1239
1240 static int qcow2_update_options(BlockDriverState *bs, QDict *options,
1241 int flags, Error **errp)
1242 {
1243 Qcow2ReopenState r = {};
1244 int ret;
1245
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);
1251 }
1252
1253 return ret;
1254 }
1255
1256 static int validate_compression_type(BDRVQcow2State *s, Error **errp)
1257 {
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;
1264
1265 default:
1266 error_setg(errp, "qcow2: unknown compression type: %u",
1267 s->compression_type);
1268 return -ENOTSUP;
1269 }
1270
1271 /*
1272 * if the compression type differs from QCOW2_COMPRESSION_TYPE_ZLIB
1273 * the incompatible feature flag must be set
1274 */
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;
1280 }
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;
1286 }
1287 }
1288
1289 return 0;
1290 }
1291
1292 /* Called with s->lock held. */
1293 static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options,
1294 int flags, Error **errp)
1295 {
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;
1304
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;
1309 }
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);
1324
1325 if (header.magic != QCOW_MAGIC) {
1326 error_setg(errp, "Image is not in qcow2 format");
1327 ret = -EINVAL;
1328 goto fail;
1329 }
1330 if (header.version < 2 || header.version > 3) {
1331 error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version);
1332 ret = -ENOTSUP;
1333 goto fail;
1334 }
1335
1336 s->qcow_version = header.version;
1337
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;
1345 }
1346
1347 s->cluster_bits = header.cluster_bits;
1348 s->cluster_size = 1 << s->cluster_bits;
1349
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);
1364
1365 if (header.header_length < 104) {
1366 error_setg(errp, "qcow2 header too short");
1367 ret = -EINVAL;
1368 goto fail;
1369 }
1370 }
1371
1372 if (header.header_length > s->cluster_size) {
1373 error_setg(errp, "qcow2 header exceeds cluster size");
1374 ret = -EINVAL;
1375 goto fail;
1376 }
1377
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;
1387 }
1388 }
1389
1390 if (header.backing_file_offset > s->cluster_size) {
1391 error_setg(errp, "Invalid backing file offset");
1392 ret = -EINVAL;
1393 goto fail;
1394 }
1395
1396 if (header.backing_file_offset) {
1397 ext_end = header.backing_file_offset;
1398 } else {
1399 ext_end = 1 << header.cluster_bits;
1400 }
1401
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;
1406
1407 /*
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
1412 */
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;
1417 }
1418
1419 ret = validate_compression_type(s, errp);
1420 if (ret) {
1421 goto fail;
1422 }
1423
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;
1434 }
1435
1436 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
1437 /* Corrupt images may not be written to unless they are being repaired
1438 */
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;
1444 }
1445 }
1446
1447 s->subclusters_per_cluster =
1448 has_subclusters(s) ? QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER : 1;
1449 s->subcluster_size = s->cluster_size / s->subclusters_per_cluster;
1450 s->subcluster_bits = ctz32(s->subcluster_size);
1451
1452 if (s->subcluster_size < (1 << MIN_CLUSTER_BITS)) {
1453 error_setg(errp, "Unsupported subcluster size: %d", s->subcluster_size);
1454 ret = -EINVAL;
1455 goto fail;
1456 }
1457
1458 /* Check support for various header values */
1459 if (header.refcount_order > 6) {
1460 error_setg(errp, "Reference count entry width too large; may not "
1461 "exceed 64 bits");
1462 ret = -EINVAL;
1463 goto fail;
1464 }
1465 s->refcount_order = header.refcount_order;
1466 s->refcount_bits = 1 << s->refcount_order;
1467 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
1468 s->refcount_max += s->refcount_max - 1;
1469
1470 s->crypt_method_header = header.crypt_method;
1471 if (s->crypt_method_header) {
1472 if (bdrv_uses_whitelist() &&
1473 s->crypt_method_header == QCOW_CRYPT_AES) {
1474 error_setg(errp,
1475 "Use of AES-CBC encrypted qcow2 images is no longer "
1476 "supported in system emulators");
1477 error_append_hint(errp,
1478 "You can use 'qemu-img convert' to convert your "
1479 "image to an alternative supported format, such "
1480 "as unencrypted qcow2, or raw with the LUKS "
1481 "format instead.\n");
1482 ret = -ENOSYS;
1483 goto fail;
1484 }
1485
1486 if (s->crypt_method_header == QCOW_CRYPT_AES) {
1487 s->crypt_physical_offset = false;
1488 } else {
1489 /* Assuming LUKS and any future crypt methods we
1490 * add will all use physical offsets, due to the
1491 * fact that the alternative is insecure... */
1492 s->crypt_physical_offset = true;
1493 }
1494
1495 bs->encrypted = true;
1496 }
1497
1498 s->l2_bits = s->cluster_bits - ctz32(l2_entry_size(s));
1499 s->l2_size = 1 << s->l2_bits;
1500 /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1501 s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
1502 s->refcount_block_size = 1 << s->refcount_block_bits;
1503 bs->total_sectors = header.size / BDRV_SECTOR_SIZE;
1504 s->csize_shift = (62 - (s->cluster_bits - 8));
1505 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
1506 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
1507
1508 s->refcount_table_offset = header.refcount_table_offset;
1509 s->refcount_table_size =
1510 header.refcount_table_clusters << (s->cluster_bits - 3);
1511
1512 if (header.refcount_table_clusters == 0 && !(flags & BDRV_O_CHECK)) {
1513 error_setg(errp, "Image does not contain a reference count table");
1514 ret = -EINVAL;
1515 goto fail;
1516 }
1517
1518 ret = qcow2_validate_table(bs, s->refcount_table_offset,
1519 header.refcount_table_clusters,
1520 s->cluster_size, QCOW_MAX_REFTABLE_SIZE,
1521 "Reference count table", errp);
1522 if (ret < 0) {
1523 goto fail;
1524 }
1525
1526 if (!(flags & BDRV_O_CHECK)) {
1527 /*
1528 * The total size in bytes of the snapshot table is checked in
1529 * qcow2_read_snapshots() because the size of each snapshot is
1530 * variable and we don't know it yet.
1531 * Here we only check the offset and number of snapshots.
1532 */
1533 ret = qcow2_validate_table(bs, header.snapshots_offset,
1534 header.nb_snapshots,
1535 sizeof(QCowSnapshotHeader),
1536 sizeof(QCowSnapshotHeader) *
1537 QCOW_MAX_SNAPSHOTS,
1538 "Snapshot table", errp);
1539 if (ret < 0) {
1540 goto fail;
1541 }
1542 }
1543
1544 /* read the level 1 table */
1545 ret = qcow2_validate_table(bs, header.l1_table_offset,
1546 header.l1_size, L1E_SIZE,
1547 QCOW_MAX_L1_SIZE, "Active L1 table", errp);
1548 if (ret < 0) {
1549 goto fail;
1550 }
1551 s->l1_size = header.l1_size;
1552 s->l1_table_offset = header.l1_table_offset;
1553
1554 l1_vm_state_index = size_to_l1(s, header.size);
1555 if (l1_vm_state_index > INT_MAX) {
1556 error_setg(errp, "Image is too big");
1557 ret = -EFBIG;
1558 goto fail;
1559 }
1560 s->l1_vm_state_index = l1_vm_state_index;
1561
1562 /* the L1 table must contain at least enough entries to put
1563 header.size bytes */
1564 if (s->l1_size < s->l1_vm_state_index) {
1565 error_setg(errp, "L1 table is too small");
1566 ret = -EINVAL;
1567 goto fail;
1568 }
1569
1570 if (s->l1_size > 0) {
1571 s->l1_table = qemu_try_blockalign(bs->file->bs, s->l1_size * L1E_SIZE);
1572 if (s->l1_table == NULL) {
1573 error_setg(errp, "Could not allocate L1 table");
1574 ret = -ENOMEM;
1575 goto fail;
1576 }
1577 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
1578 s->l1_size * L1E_SIZE);
1579 if (ret < 0) {
1580 error_setg_errno(errp, -ret, "Could not read L1 table");
1581 goto fail;
1582 }
1583 for(i = 0;i < s->l1_size; i++) {
1584 s->l1_table[i] = be64_to_cpu(s->l1_table[i]);
1585 }
1586 }
1587
1588 /* Parse driver-specific options */
1589 ret = qcow2_update_options(bs, options, flags, errp);
1590 if (ret < 0) {
1591 goto fail;
1592 }
1593
1594 s->flags = flags;
1595
1596 ret = qcow2_refcount_init(bs);
1597 if (ret != 0) {
1598 error_setg_errno(errp, -ret, "Could not initialize refcount handling");
1599 goto fail;
1600 }
1601
1602 QLIST_INIT(&s->cluster_allocs);
1603 QTAILQ_INIT(&s->discards);
1604
1605 /* read qcow2 extensions */
1606 if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
1607 flags, &update_header, errp)) {
1608 ret = -EINVAL;
1609 goto fail;
1610 }
1611
1612 /* Open external data file */
1613 s->data_file = bdrv_open_child(NULL, options, "data-file", bs,
1614 &child_of_bds, BDRV_CHILD_DATA,
1615 true, &local_err);
1616 if (local_err) {
1617 error_propagate(errp, local_err);
1618 ret = -EINVAL;
1619 goto fail;
1620 }
1621
1622 if (s->incompatible_features & QCOW2_INCOMPAT_DATA_FILE) {
1623 if (!s->data_file && s->image_data_file) {
1624 s->data_file = bdrv_open_child(s->image_data_file, options,
1625 "data-file", bs, &child_of_bds,
1626 BDRV_CHILD_DATA, false, errp);
1627 if (!s->data_file) {
1628 ret = -EINVAL;
1629 goto fail;
1630 }
1631 }
1632 if (!s->data_file) {
1633 error_setg(errp, "'data-file' is required for this image");
1634 ret = -EINVAL;
1635 goto fail;
1636 }
1637
1638 /* No data here */
1639 bs->file->role &= ~BDRV_CHILD_DATA;
1640
1641 /* Must succeed because we have given up permissions if anything */
1642 bdrv_child_refresh_perms(bs, bs->file, &error_abort);
1643 } else {
1644 if (s->data_file) {
1645 error_setg(errp, "'data-file' can only be set for images with an "
1646 "external data file");
1647 ret = -EINVAL;
1648 goto fail;
1649 }
1650
1651 s->data_file = bs->file;
1652
1653 if (data_file_is_raw(bs)) {
1654 error_setg(errp, "data-file-raw requires a data file");
1655 ret = -EINVAL;
1656 goto fail;
1657 }
1658 }
1659
1660 /* qcow2_read_extension may have set up the crypto context
1661 * if the crypt method needs a header region, some methods
1662 * don't need header extensions, so must check here
1663 */
1664 if (s->crypt_method_header && !s->crypto) {
1665 if (s->crypt_method_header == QCOW_CRYPT_AES) {
1666 unsigned int cflags = 0;
1667 if (flags & BDRV_O_NO_IO) {
1668 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
1669 }
1670 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
1671 NULL, NULL, cflags,
1672 QCOW2_MAX_THREADS, errp);
1673 if (!s->crypto) {
1674 ret = -EINVAL;
1675 goto fail;
1676 }
1677 } else if (!(flags & BDRV_O_NO_IO)) {
1678 error_setg(errp, "Missing CRYPTO header for crypt method %d",
1679 s->crypt_method_header);
1680 ret = -EINVAL;
1681 goto fail;
1682 }
1683 }
1684
1685 /* read the backing file name */
1686 if (header.backing_file_offset != 0) {
1687 len = header.backing_file_size;
1688 if (len > MIN(1023, s->cluster_size - header.backing_file_offset) ||
1689 len >= sizeof(bs->backing_file)) {
1690 error_setg(errp, "Backing file name too long");
1691 ret = -EINVAL;
1692 goto fail;
1693 }
1694 ret = bdrv_pread(bs->file, header.backing_file_offset,
1695 bs->auto_backing_file, len);
1696 if (ret < 0) {
1697 error_setg_errno(errp, -ret, "Could not read backing file name");
1698 goto fail;
1699 }
1700 bs->auto_backing_file[len] = '\0';
1701 pstrcpy(bs->backing_file, sizeof(bs->backing_file),
1702 bs->auto_backing_file);
1703 s->image_backing_file = g_strdup(bs->auto_backing_file);
1704 }
1705
1706 /*
1707 * Internal snapshots; skip reading them in check mode, because
1708 * we do not need them then, and we do not want to abort because
1709 * of a broken table.
1710 */
1711 if (!(flags & BDRV_O_CHECK)) {
1712 s->snapshots_offset = header.snapshots_offset;
1713 s->nb_snapshots = header.nb_snapshots;
1714
1715 ret = qcow2_read_snapshots(bs, errp);
1716 if (ret < 0) {
1717 goto fail;
1718 }
1719 }
1720
1721 /* Clear unknown autoclear feature bits */
1722 update_header |= s->autoclear_features & ~QCOW2_AUTOCLEAR_MASK;
1723 update_header =
1724 update_header && !bs->read_only && !(flags & BDRV_O_INACTIVE);
1725 if (update_header) {
1726 s->autoclear_features &= QCOW2_AUTOCLEAR_MASK;
1727 }
1728
1729 /* == Handle persistent dirty bitmaps ==
1730 *
1731 * We want load dirty bitmaps in three cases:
1732 *
1733 * 1. Normal open of the disk in active mode, not related to invalidation
1734 * after migration.
1735 *
1736 * 2. Invalidation of the target vm after pre-copy phase of migration, if
1737 * bitmaps are _not_ migrating through migration channel, i.e.
1738 * 'dirty-bitmaps' capability is disabled.
1739 *
1740 * 3. Invalidation of source vm after failed or canceled migration.
1741 * This is a very interesting case. There are two possible types of
1742 * bitmaps:
1743 *
1744 * A. Stored on inactivation and removed. They should be loaded from the
1745 * image.
1746 *
1747 * B. Not stored: not-persistent bitmaps and bitmaps, migrated through
1748 * the migration channel (with dirty-bitmaps capability).
1749 *
1750 * On the other hand, there are two possible sub-cases:
1751 *
1752 * 3.1 disk was changed by somebody else while were inactive. In this
1753 * case all in-RAM dirty bitmaps (both persistent and not) are
1754 * definitely invalid. And we don't have any method to determine
1755 * this.
1756 *
1757 * Simple and safe thing is to just drop all the bitmaps of type B on
1758 * inactivation. But in this case we lose bitmaps in valid 4.2 case.
1759 *
1760 * On the other hand, resuming source vm, if disk was already changed
1761 * is a bad thing anyway: not only bitmaps, the whole vm state is
1762 * out of sync with disk.
1763 *
1764 * This means, that user or management tool, who for some reason
1765 * decided to resume source vm, after disk was already changed by
1766 * target vm, should at least drop all dirty bitmaps by hand.
1767 *
1768 * So, we can ignore this case for now, but TODO: "generation"
1769 * extension for qcow2, to determine, that image was changed after
1770 * last inactivation. And if it is changed, we will drop (or at least
1771 * mark as 'invalid' all the bitmaps of type B, both persistent
1772 * and not).
1773 *
1774 * 3.2 disk was _not_ changed while were inactive. Bitmaps may be saved
1775 * to disk ('dirty-bitmaps' capability disabled), or not saved
1776 * ('dirty-bitmaps' capability enabled), but we don't need to care
1777 * of: let's load bitmaps as always: stored bitmaps will be loaded,
1778 * and not stored has flag IN_USE=1 in the image and will be skipped
1779 * on loading.
1780 *
1781 * One remaining possible case when we don't want load bitmaps:
1782 *
1783 * 4. Open disk in inactive mode in target vm (bitmaps are migrating or
1784 * will be loaded on invalidation, no needs try loading them before)
1785 */
1786
1787 if (!(bdrv_get_flags(bs) & BDRV_O_INACTIVE)) {
1788 /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */
1789 bool header_updated = qcow2_load_dirty_bitmaps(bs, &local_err);
1790 if (local_err != NULL) {
1791 error_propagate(errp, local_err);
1792 ret = -EINVAL;
1793 goto fail;
1794 }
1795
1796 update_header = update_header && !header_updated;
1797 }
1798
1799 if (update_header) {
1800 ret = qcow2_update_header(bs);
1801 if (ret < 0) {
1802 error_setg_errno(errp, -ret, "Could not update qcow2 header");
1803 goto fail;
1804 }
1805 }
1806
1807 bs->supported_zero_flags = header.version >= 3 ?
1808 BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK : 0;
1809 bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE;
1810
1811 /* Repair image if dirty */
1812 if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only &&
1813 (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
1814 BdrvCheckResult result = {0};
1815
1816 ret = qcow2_co_check_locked(bs, &result,
1817 BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
1818 if (ret < 0 || result.check_errors) {
1819 if (ret >= 0) {
1820 ret = -EIO;
1821 }
1822 error_setg_errno(errp, -ret, "Could not repair dirty image");
1823 goto fail;
1824 }
1825 }
1826
1827 #ifdef DEBUG_ALLOC
1828 {
1829 BdrvCheckResult result = {0};
1830 qcow2_check_refcounts(bs, &result, 0);
1831 }
1832 #endif
1833
1834 qemu_co_queue_init(&s->thread_task_queue);
1835
1836 return ret;
1837
1838 fail:
1839 g_free(s->image_data_file);
1840 if (has_data_file(bs)) {
1841 bdrv_unref_child(bs, s->data_file);
1842 s->data_file = NULL;
1843 }
1844 g_free(s->unknown_header_fields);
1845 cleanup_unknown_header_ext(bs);
1846 qcow2_free_snapshots(bs);
1847 qcow2_refcount_close(bs);
1848 qemu_vfree(s->l1_table);
1849 /* else pre-write overlap checks in cache_destroy may crash */
1850 s->l1_table = NULL;
1851 cache_clean_timer_del(bs);
1852 if (s->l2_table_cache) {
1853 qcow2_cache_destroy(s->l2_table_cache);
1854 }
1855 if (s->refcount_block_cache) {
1856 qcow2_cache_destroy(s->refcount_block_cache);
1857 }
1858 qcrypto_block_free(s->crypto);
1859 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1860 return ret;
1861 }
1862
1863 typedef struct QCow2OpenCo {
1864 BlockDriverState *bs;
1865 QDict *options;
1866 int flags;
1867 Error **errp;
1868 int ret;
1869 } QCow2OpenCo;
1870
1871 static void coroutine_fn qcow2_open_entry(void *opaque)
1872 {
1873 QCow2OpenCo *qoc = opaque;
1874 BDRVQcow2State *s = qoc->bs->opaque;
1875
1876 qemu_co_mutex_lock(&s->lock);
1877 qoc->ret = qcow2_do_open(qoc->bs, qoc->options, qoc->flags, qoc->errp);
1878 qemu_co_mutex_unlock(&s->lock);
1879 }
1880
1881 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
1882 Error **errp)
1883 {
1884 BDRVQcow2State *s = bs->opaque;
1885 QCow2OpenCo qoc = {
1886 .bs = bs,
1887 .options = options,
1888 .flags = flags,
1889 .errp = errp,
1890 .ret = -EINPROGRESS
1891 };
1892
1893 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds,
1894 BDRV_CHILD_IMAGE, false, errp);
1895 if (!bs->file) {
1896 return -EINVAL;
1897 }
1898
1899 /* Initialise locks */
1900 qemu_co_mutex_init(&s->lock);
1901
1902 if (qemu_in_coroutine()) {
1903 /* From bdrv_co_create. */
1904 qcow2_open_entry(&qoc);
1905 } else {
1906 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
1907 qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry, &qoc));
1908 BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS);
1909 }
1910 return qoc.ret;
1911 }
1912
1913 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1914 {
1915 BDRVQcow2State *s = bs->opaque;
1916
1917 if (bs->encrypted) {
1918 /* Encryption works on a sector granularity */
1919 bs->bl.request_alignment = qcrypto_block_get_sector_size(s->crypto);
1920 }
1921 bs->bl.pwrite_zeroes_alignment = s->subcluster_size;
1922 bs->bl.pdiscard_alignment = s->cluster_size;
1923 }
1924
1925 static int qcow2_reopen_prepare(BDRVReopenState *state,
1926 BlockReopenQueue *queue, Error **errp)
1927 {
1928 Qcow2ReopenState *r;
1929 int ret;
1930
1931 r = g_new0(Qcow2ReopenState, 1);
1932 state->opaque = r;
1933
1934 ret = qcow2_update_options_prepare(state->bs, r, state->options,
1935 state->flags, errp);
1936 if (ret < 0) {
1937 goto fail;
1938 }
1939
1940 /* We need to write out any unwritten data if we reopen read-only. */
1941 if ((state->flags & BDRV_O_RDWR) == 0) {
1942 ret = qcow2_reopen_bitmaps_ro(state->bs, errp);
1943 if (ret < 0) {
1944 goto fail;
1945 }
1946
1947 ret = bdrv_flush(state->bs);
1948 if (ret < 0) {
1949 goto fail;
1950 }
1951
1952 ret = qcow2_mark_clean(state->bs);
1953 if (ret < 0) {
1954 goto fail;
1955 }
1956 }
1957
1958 return 0;
1959
1960 fail:
1961 qcow2_update_options_abort(state->bs, r);
1962 g_free(r);
1963 return ret;
1964 }
1965
1966 static void qcow2_reopen_commit(BDRVReopenState *state)
1967 {
1968 qcow2_update_options_commit(state->bs, state->opaque);
1969 g_free(state->opaque);
1970 }
1971
1972 static void qcow2_reopen_commit_post(BDRVReopenState *state)
1973 {
1974 if (state->flags & BDRV_O_RDWR) {
1975 Error *local_err = NULL;
1976
1977 if (qcow2_reopen_bitmaps_rw(state->bs, &local_err) < 0) {
1978 /*
1979 * This is not fatal, bitmaps just left read-only, so all following
1980 * writes will fail. User can remove read-only bitmaps to unblock
1981 * writes or retry reopen.
1982 */
1983 error_reportf_err(local_err,
1984 "%s: Failed to make dirty bitmaps writable: ",
1985 bdrv_get_node_name(state->bs));
1986 }
1987 }
1988 }
1989
1990 static void qcow2_reopen_abort(BDRVReopenState *state)
1991 {
1992 qcow2_update_options_abort(state->bs, state->opaque);
1993 g_free(state->opaque);
1994 }
1995
1996 static void qcow2_join_options(QDict *options, QDict *old_options)
1997 {
1998 bool has_new_overlap_template =
1999 qdict_haskey(options, QCOW2_OPT_OVERLAP) ||
2000 qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE);
2001 bool has_new_total_cache_size =
2002 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE);
2003 bool has_all_cache_options;
2004
2005 /* New overlap template overrides all old overlap options */
2006 if (has_new_overlap_template) {
2007 qdict_del(old_options, QCOW2_OPT_OVERLAP);
2008 qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE);
2009 qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER);
2010 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1);
2011 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2);
2012 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE);
2013 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK);
2014 qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE);
2015 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1);
2016 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2);
2017 }
2018
2019 /* New total cache size overrides all old options */
2020 if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) {
2021 qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE);
2022 qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
2023 }
2024
2025 qdict_join(options, old_options, false);
2026
2027 /*
2028 * If after merging all cache size options are set, an old total size is
2029 * overwritten. Do keep all options, however, if all three are new. The
2030 * resulting error message is what we want to happen.
2031 */
2032 has_all_cache_options =
2033 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) ||
2034 qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) ||
2035 qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
2036
2037 if (has_all_cache_options && !has_new_total_cache_size) {
2038 qdict_del(options, QCOW2_OPT_CACHE_SIZE);
2039 }
2040 }
2041
2042 static int coroutine_fn qcow2_co_block_status(BlockDriverState *bs,
2043 bool want_zero,
2044 int64_t offset, int64_t count,
2045 int64_t *pnum, int64_t *map,
2046 BlockDriverState **file)
2047 {
2048 BDRVQcow2State *s = bs->opaque;
2049 uint64_t host_offset;
2050 unsigned int bytes;
2051 QCow2SubclusterType type;
2052 int ret, status = 0;
2053
2054 qemu_co_mutex_lock(&s->lock);
2055
2056 if (!s->metadata_preallocation_checked) {
2057 ret = qcow2_detect_metadata_preallocation(bs);
2058 s->metadata_preallocation = (ret == 1);
2059 s->metadata_preallocation_checked = true;
2060 }
2061
2062 bytes = MIN(INT_MAX, count);
2063 ret = qcow2_get_host_offset(bs, offset, &bytes, &host_offset, &type);
2064 qemu_co_mutex_unlock(&s->lock);
2065 if (ret < 0) {
2066 return ret;
2067 }
2068
2069 *pnum = bytes;
2070
2071 if ((type == QCOW2_SUBCLUSTER_NORMAL ||
2072 type == QCOW2_SUBCLUSTER_ZERO_ALLOC ||
2073 type == QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC) && !s->crypto) {
2074 *map = host_offset;
2075 *file = s->data_file->bs;
2076 status |= BDRV_BLOCK_OFFSET_VALID;
2077 }
2078 if (type == QCOW2_SUBCLUSTER_ZERO_PLAIN ||
2079 type == QCOW2_SUBCLUSTER_ZERO_ALLOC) {
2080 status |= BDRV_BLOCK_ZERO;
2081 } else if (type != QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN &&
2082 type != QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC) {
2083 status |= BDRV_BLOCK_DATA;
2084 }
2085 if (s->metadata_preallocation && (status & BDRV_BLOCK_DATA) &&
2086 (status & BDRV_BLOCK_OFFSET_VALID))
2087 {
2088 status |= BDRV_BLOCK_RECURSE;
2089 }
2090 return status;
2091 }
2092
2093 static coroutine_fn int qcow2_handle_l2meta(BlockDriverState *bs,
2094 QCowL2Meta **pl2meta,
2095 bool link_l2)
2096 {
2097 int ret = 0;
2098 QCowL2Meta *l2meta = *pl2meta;
2099
2100 while (l2meta != NULL) {
2101 QCowL2Meta *next;
2102
2103 if (link_l2) {
2104 assert(!l2meta->prealloc);
2105 ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
2106 if (ret) {
2107 goto out;
2108 }
2109 } else {
2110 qcow2_alloc_cluster_abort(bs, l2meta);
2111 }
2112
2113 /* Take the request off the list of running requests */
2114 if (l2meta->nb_clusters != 0) {
2115 QLIST_REMOVE(l2meta, next_in_flight);
2116 }
2117
2118 qemu_co_queue_restart_all(&l2meta->dependent_requests);
2119
2120 next = l2meta->next;
2121 g_free(l2meta);
2122 l2meta = next;
2123 }
2124 out:
2125 *pl2meta = l2meta;
2126 return ret;
2127 }
2128
2129 static coroutine_fn int
2130 qcow2_co_preadv_encrypted(BlockDriverState *bs,
2131 uint64_t host_offset,
2132 uint64_t offset,
2133 uint64_t bytes,
2134 QEMUIOVector *qiov,
2135 uint64_t qiov_offset)
2136 {
2137 int ret;
2138 BDRVQcow2State *s = bs->opaque;
2139 uint8_t *buf;
2140
2141 assert(bs->encrypted && s->crypto);
2142 assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2143
2144 /*
2145 * For encrypted images, read everything into a temporary
2146 * contiguous buffer on which the AES functions can work.
2147 * Also, decryption in a separate buffer is better as it
2148 * prevents the guest from learning information about the
2149 * encrypted nature of the virtual disk.
2150 */
2151
2152 buf = qemu_try_blockalign(s->data_file->bs, bytes);
2153 if (buf == NULL) {
2154 return -ENOMEM;
2155 }
2156
2157 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
2158 ret = bdrv_co_pread(s->data_file, host_offset, bytes, buf, 0);
2159 if (ret < 0) {
2160 goto fail;
2161 }
2162
2163 if (qcow2_co_decrypt(bs, host_offset, offset, buf, bytes) < 0)
2164 {
2165 ret = -EIO;
2166 goto fail;
2167 }
2168 qemu_iovec_from_buf(qiov, qiov_offset, buf, bytes);
2169
2170 fail:
2171 qemu_vfree(buf);
2172
2173 return ret;
2174 }
2175
2176 typedef struct Qcow2AioTask {
2177 AioTask task;
2178
2179 BlockDriverState *bs;
2180 QCow2SubclusterType subcluster_type; /* only for read */
2181 uint64_t host_offset; /* or full descriptor in compressed clusters */
2182 uint64_t offset;
2183 uint64_t bytes;
2184 QEMUIOVector *qiov;
2185 uint64_t qiov_offset;
2186 QCowL2Meta *l2meta; /* only for write */
2187 } Qcow2AioTask;
2188
2189 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task);
2190 static coroutine_fn int qcow2_add_task(BlockDriverState *bs,
2191 AioTaskPool *pool,
2192 AioTaskFunc func,
2193 QCow2SubclusterType subcluster_type,
2194 uint64_t host_offset,
2195 uint64_t offset,
2196 uint64_t bytes,
2197 QEMUIOVector *qiov,
2198 size_t qiov_offset,
2199 QCowL2Meta *l2meta)
2200 {
2201 Qcow2AioTask local_task;
2202 Qcow2AioTask *task = pool ? g_new(Qcow2AioTask, 1) : &local_task;
2203
2204 *task = (Qcow2AioTask) {
2205 .task.func = func,
2206 .bs = bs,
2207 .subcluster_type = subcluster_type,
2208 .qiov = qiov,
2209 .host_offset = host_offset,
2210 .offset = offset,
2211 .bytes = bytes,
2212 .qiov_offset = qiov_offset,
2213 .l2meta = l2meta,
2214 };
2215
2216 trace_qcow2_add_task(qemu_coroutine_self(), bs, pool,
2217 func == qcow2_co_preadv_task_entry ? "read" : "write",
2218 subcluster_type, host_offset, offset, bytes,
2219 qiov, qiov_offset);
2220
2221 if (!pool) {
2222 return func(&task->task);
2223 }
2224
2225 aio_task_pool_start_task(pool, &task->task);
2226
2227 return 0;
2228 }
2229
2230 static coroutine_fn int qcow2_co_preadv_task(BlockDriverState *bs,
2231 QCow2SubclusterType subc_type,
2232 uint64_t host_offset,
2233 uint64_t offset, uint64_t bytes,
2234 QEMUIOVector *qiov,
2235 size_t qiov_offset)
2236 {
2237 BDRVQcow2State *s = bs->opaque;
2238
2239 switch (subc_type) {
2240 case QCOW2_SUBCLUSTER_ZERO_PLAIN:
2241 case QCOW2_SUBCLUSTER_ZERO_ALLOC:
2242 /* Both zero types are handled in qcow2_co_preadv_part */
2243 g_assert_not_reached();
2244
2245 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN:
2246 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC:
2247 assert(bs->backing); /* otherwise handled in qcow2_co_preadv_part */
2248
2249 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
2250 return bdrv_co_preadv_part(bs->backing, offset, bytes,
2251 qiov, qiov_offset, 0);
2252
2253 case QCOW2_SUBCLUSTER_COMPRESSED:
2254 return qcow2_co_preadv_compressed(bs, host_offset,
2255 offset, bytes, qiov, qiov_offset);
2256
2257 case QCOW2_SUBCLUSTER_NORMAL:
2258 if (bs->encrypted) {
2259 return qcow2_co_preadv_encrypted(bs, host_offset,
2260 offset, bytes, qiov, qiov_offset);
2261 }
2262
2263 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
2264 return bdrv_co_preadv_part(s->data_file, host_offset,
2265 bytes, qiov, qiov_offset, 0);
2266
2267 default:
2268 g_assert_not_reached();
2269 }
2270
2271 g_assert_not_reached();
2272 }
2273
2274 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task)
2275 {
2276 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
2277
2278 assert(!t->l2meta);
2279
2280 return qcow2_co_preadv_task(t->bs, t->subcluster_type,
2281 t->host_offset, t->offset, t->bytes,
2282 t->qiov, t->qiov_offset);
2283 }
2284
2285 static coroutine_fn int qcow2_co_preadv_part(BlockDriverState *bs,
2286 uint64_t offset, uint64_t bytes,
2287 QEMUIOVector *qiov,
2288 size_t qiov_offset, int flags)
2289 {
2290 BDRVQcow2State *s = bs->opaque;
2291 int ret = 0;
2292 unsigned int cur_bytes; /* number of bytes in current iteration */
2293 uint64_t host_offset = 0;
2294 QCow2SubclusterType type;
2295 AioTaskPool *aio = NULL;
2296
2297 while (bytes != 0 && aio_task_pool_status(aio) == 0) {
2298 /* prepare next request */
2299 cur_bytes = MIN(bytes, INT_MAX);
2300 if (s->crypto) {
2301 cur_bytes = MIN(cur_bytes,
2302 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2303 }
2304
2305 qemu_co_mutex_lock(&s->lock);
2306 ret = qcow2_get_host_offset(bs, offset, &cur_bytes,
2307 &host_offset, &type);
2308 qemu_co_mutex_unlock(&s->lock);
2309 if (ret < 0) {
2310 goto out;
2311 }
2312
2313 if (type == QCOW2_SUBCLUSTER_ZERO_PLAIN ||
2314 type == QCOW2_SUBCLUSTER_ZERO_ALLOC ||
2315 (type == QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN && !bs->backing) ||
2316 (type == QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC && !bs->backing))
2317 {
2318 qemu_iovec_memset(qiov, qiov_offset, 0, cur_bytes);
2319 } else {
2320 if (!aio && cur_bytes != bytes) {
2321 aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
2322 }
2323 ret = qcow2_add_task(bs, aio, qcow2_co_preadv_task_entry, type,
2324 host_offset, offset, cur_bytes,
2325 qiov, qiov_offset, NULL);
2326 if (ret < 0) {
2327 goto out;
2328 }
2329 }
2330
2331 bytes -= cur_bytes;
2332 offset += cur_bytes;
2333 qiov_offset += cur_bytes;
2334 }
2335
2336 out:
2337 if (aio) {
2338 aio_task_pool_wait_all(aio);
2339 if (ret == 0) {
2340 ret = aio_task_pool_status(aio);
2341 }
2342 g_free(aio);
2343 }
2344
2345 return ret;
2346 }
2347
2348 /* Check if it's possible to merge a write request with the writing of
2349 * the data from the COW regions */
2350 static bool merge_cow(uint64_t offset, unsigned bytes,
2351 QEMUIOVector *qiov, size_t qiov_offset,
2352 QCowL2Meta *l2meta)
2353 {
2354 QCowL2Meta *m;
2355
2356 for (m = l2meta; m != NULL; m = m->next) {
2357 /* If both COW regions are empty then there's nothing to merge */
2358 if (m->cow_start.nb_bytes == 0 && m->cow_end.nb_bytes == 0) {
2359 continue;
2360 }
2361
2362 /* If COW regions are handled already, skip this too */
2363 if (m->skip_cow) {
2364 continue;
2365 }
2366
2367 /* The data (middle) region must be immediately after the
2368 * start region */
2369 if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) {
2370 continue;
2371 }
2372
2373 /* The end region must be immediately after the data (middle)
2374 * region */
2375 if (m->offset + m->cow_end.offset != offset + bytes) {
2376 continue;
2377 }
2378
2379 /* Make sure that adding both COW regions to the QEMUIOVector
2380 * does not exceed IOV_MAX */
2381 if (qemu_iovec_subvec_niov(qiov, qiov_offset, bytes) > IOV_MAX - 2) {
2382 continue;
2383 }
2384
2385 m->data_qiov = qiov;
2386 m->data_qiov_offset = qiov_offset;
2387 return true;
2388 }
2389
2390 return false;
2391 }
2392
2393 static bool is_unallocated(BlockDriverState *bs, int64_t offset, int64_t bytes)
2394 {
2395 int64_t nr;
2396 return !bytes ||
2397 (!bdrv_is_allocated_above(bs, NULL, false, offset, bytes, &nr) &&
2398 nr == bytes);
2399 }
2400
2401 static bool is_zero_cow(BlockDriverState *bs, QCowL2Meta *m)
2402 {
2403 /*
2404 * This check is designed for optimization shortcut so it must be
2405 * efficient.
2406 * Instead of is_zero(), use is_unallocated() as it is faster (but not
2407 * as accurate and can result in false negatives).
2408 */
2409 return is_unallocated(bs, m->offset + m->cow_start.offset,
2410 m->cow_start.nb_bytes) &&
2411 is_unallocated(bs, m->offset + m->cow_end.offset,
2412 m->cow_end.nb_bytes);
2413 }
2414
2415 static int handle_alloc_space(BlockDriverState *bs, QCowL2Meta *l2meta)
2416 {
2417 BDRVQcow2State *s = bs->opaque;
2418 QCowL2Meta *m;
2419
2420 if (!(s->data_file->bs->supported_zero_flags & BDRV_REQ_NO_FALLBACK)) {
2421 return 0;
2422 }
2423
2424 if (bs->encrypted) {
2425 return 0;
2426 }
2427
2428 for (m = l2meta; m != NULL; m = m->next) {
2429 int ret;
2430 uint64_t start_offset = m->alloc_offset + m->cow_start.offset;
2431 unsigned nb_bytes = m->cow_end.offset + m->cow_end.nb_bytes -
2432 m->cow_start.offset;
2433
2434 if (!m->cow_start.nb_bytes && !m->cow_end.nb_bytes) {
2435 continue;
2436 }
2437
2438 if (!is_zero_cow(bs, m)) {
2439 continue;
2440 }
2441
2442 /*
2443 * instead of writing zero COW buffers,
2444 * efficiently zero out the whole clusters
2445 */
2446
2447 ret = qcow2_pre_write_overlap_check(bs, 0, start_offset, nb_bytes,
2448 true);
2449 if (ret < 0) {
2450 return ret;
2451 }
2452
2453 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_SPACE);
2454 ret = bdrv_co_pwrite_zeroes(s->data_file, start_offset, nb_bytes,
2455 BDRV_REQ_NO_FALLBACK);
2456 if (ret < 0) {
2457 if (ret != -ENOTSUP && ret != -EAGAIN) {
2458 return ret;
2459 }
2460 continue;
2461 }
2462
2463 trace_qcow2_skip_cow(qemu_coroutine_self(), m->offset, m->nb_clusters);
2464 m->skip_cow = true;
2465 }
2466 return 0;
2467 }
2468
2469 /*
2470 * qcow2_co_pwritev_task
2471 * Called with s->lock unlocked
2472 * l2meta - if not NULL, qcow2_co_pwritev_task() will consume it. Caller must
2473 * not use it somehow after qcow2_co_pwritev_task() call
2474 */
2475 static coroutine_fn int qcow2_co_pwritev_task(BlockDriverState *bs,
2476 uint64_t host_offset,
2477 uint64_t offset, uint64_t bytes,
2478 QEMUIOVector *qiov,
2479 uint64_t qiov_offset,
2480 QCowL2Meta *l2meta)
2481 {
2482 int ret;
2483 BDRVQcow2State *s = bs->opaque;
2484 void *crypt_buf = NULL;
2485 QEMUIOVector encrypted_qiov;
2486
2487 if (bs->encrypted) {
2488 assert(s->crypto);
2489 assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2490 crypt_buf = qemu_try_blockalign(bs->file->bs, bytes);
2491 if (crypt_buf == NULL) {
2492 ret = -ENOMEM;
2493 goto out_unlocked;
2494 }
2495 qemu_iovec_to_buf(qiov, qiov_offset, crypt_buf, bytes);
2496
2497 if (qcow2_co_encrypt(bs, host_offset, offset, crypt_buf, bytes) < 0) {
2498 ret = -EIO;
2499 goto out_unlocked;
2500 }
2501
2502 qemu_iovec_init_buf(&encrypted_qiov, crypt_buf, bytes);
2503 qiov = &encrypted_qiov;
2504 qiov_offset = 0;
2505 }
2506
2507 /* Try to efficiently initialize the physical space with zeroes */
2508 ret = handle_alloc_space(bs, l2meta);
2509 if (ret < 0) {
2510 goto out_unlocked;
2511 }
2512
2513 /*
2514 * If we need to do COW, check if it's possible to merge the
2515 * writing of the guest data together with that of the COW regions.
2516 * If it's not possible (or not necessary) then write the
2517 * guest data now.
2518 */
2519 if (!merge_cow(offset, bytes, qiov, qiov_offset, l2meta)) {
2520 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
2521 trace_qcow2_writev_data(qemu_coroutine_self(), host_offset);
2522 ret = bdrv_co_pwritev_part(s->data_file, host_offset,
2523 bytes, qiov, qiov_offset, 0);
2524 if (ret < 0) {
2525 goto out_unlocked;
2526 }
2527 }
2528
2529 qemu_co_mutex_lock(&s->lock);
2530
2531 ret = qcow2_handle_l2meta(bs, &l2meta, true);
2532 goto out_locked;
2533
2534 out_unlocked:
2535 qemu_co_mutex_lock(&s->lock);
2536
2537 out_locked:
2538 qcow2_handle_l2meta(bs, &l2meta, false);
2539 qemu_co_mutex_unlock(&s->lock);
2540
2541 qemu_vfree(crypt_buf);
2542
2543 return ret;
2544 }
2545
2546 static coroutine_fn int qcow2_co_pwritev_task_entry(AioTask *task)
2547 {
2548 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
2549
2550 assert(!t->subcluster_type);
2551
2552 return qcow2_co_pwritev_task(t->bs, t->host_offset,
2553 t->offset, t->bytes, t->qiov, t->qiov_offset,
2554 t->l2meta);
2555 }
2556
2557 static coroutine_fn int qcow2_co_pwritev_part(
2558 BlockDriverState *bs, uint64_t offset, uint64_t bytes,
2559 QEMUIOVector *qiov, size_t qiov_offset, int flags)
2560 {
2561 BDRVQcow2State *s = bs->opaque;
2562 int offset_in_cluster;
2563 int ret;
2564 unsigned int cur_bytes; /* number of sectors in current iteration */
2565 uint64_t cluster_offset;
2566 QCowL2Meta *l2meta = NULL;
2567 AioTaskPool *aio = NULL;
2568
2569 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes);
2570
2571 while (bytes != 0 && aio_task_pool_status(aio) == 0) {
2572
2573 l2meta = NULL;
2574
2575 trace_qcow2_writev_start_part(qemu_coroutine_self());
2576 offset_in_cluster = offset_into_cluster(s, offset);
2577 cur_bytes = MIN(bytes, INT_MAX);
2578 if (bs->encrypted) {
2579 cur_bytes = MIN(cur_bytes,
2580 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
2581 - offset_in_cluster);
2582 }
2583
2584 qemu_co_mutex_lock(&s->lock);
2585
2586 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
2587 &cluster_offset, &l2meta);
2588 if (ret < 0) {
2589 goto out_locked;
2590 }
2591
2592 assert(offset_into_cluster(s, cluster_offset) == 0);
2593
2594 ret = qcow2_pre_write_overlap_check(bs, 0,
2595 cluster_offset + offset_in_cluster,
2596 cur_bytes, true);
2597 if (ret < 0) {
2598 goto out_locked;
2599 }
2600
2601 qemu_co_mutex_unlock(&s->lock);
2602
2603 if (!aio && cur_bytes != bytes) {
2604 aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
2605 }
2606 ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_task_entry, 0,
2607 cluster_offset + offset_in_cluster, offset,
2608 cur_bytes, qiov, qiov_offset, l2meta);
2609 l2meta = NULL; /* l2meta is consumed by qcow2_co_pwritev_task() */
2610 if (ret < 0) {
2611 goto fail_nometa;
2612 }
2613
2614 bytes -= cur_bytes;
2615 offset += cur_bytes;
2616 qiov_offset += cur_bytes;
2617 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes);
2618 }
2619 ret = 0;
2620
2621 qemu_co_mutex_lock(&s->lock);
2622
2623 out_locked:
2624 qcow2_handle_l2meta(bs, &l2meta, false);
2625
2626 qemu_co_mutex_unlock(&s->lock);
2627
2628 fail_nometa:
2629 if (aio) {
2630 aio_task_pool_wait_all(aio);
2631 if (ret == 0) {
2632 ret = aio_task_pool_status(aio);
2633 }
2634 g_free(aio);
2635 }
2636
2637 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
2638
2639 return ret;
2640 }
2641
2642 static int qcow2_inactivate(BlockDriverState *bs)
2643 {
2644 BDRVQcow2State *s = bs->opaque;
2645 int ret, result = 0;
2646 Error *local_err = NULL;
2647
2648 qcow2_store_persistent_dirty_bitmaps(bs, true, &local_err);
2649 if (local_err != NULL) {
2650 result = -EINVAL;
2651 error_reportf_err(local_err, "Lost persistent bitmaps during "
2652 "inactivation of node '%s': ",
2653 bdrv_get_device_or_node_name(bs));
2654 }
2655
2656 ret = qcow2_cache_flush(bs, s->l2_table_cache);
2657 if (ret) {
2658 result = ret;
2659 error_report("Failed to flush the L2 table cache: %s",
2660 strerror(-ret));
2661 }
2662
2663 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2664 if (ret) {
2665 result = ret;
2666 error_report("Failed to flush the refcount block cache: %s",
2667 strerror(-ret));
2668 }
2669
2670 if (result == 0) {
2671 qcow2_mark_clean(bs);
2672 }
2673
2674 return result;
2675 }
2676
2677 static void qcow2_close(BlockDriverState *bs)
2678 {
2679 BDRVQcow2State *s = bs->opaque;
2680 qemu_vfree(s->l1_table);
2681 /* else pre-write overlap checks in cache_destroy may crash */
2682 s->l1_table = NULL;
2683
2684 if (!(s->flags & BDRV_O_INACTIVE)) {
2685 qcow2_inactivate(bs);
2686 }
2687
2688 cache_clean_timer_del(bs);
2689 qcow2_cache_destroy(s->l2_table_cache);
2690 qcow2_cache_destroy(s->refcount_block_cache);
2691
2692 qcrypto_block_free(s->crypto);
2693 s->crypto = NULL;
2694 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
2695
2696 g_free(s->unknown_header_fields);
2697 cleanup_unknown_header_ext(bs);
2698
2699 g_free(s->image_data_file);
2700 g_free(s->image_backing_file);
2701 g_free(s->image_backing_format);
2702
2703 if (has_data_file(bs)) {
2704 bdrv_unref_child(bs, s->data_file);
2705 s->data_file = NULL;
2706 }
2707
2708 qcow2_refcount_close(bs);
2709 qcow2_free_snapshots(bs);
2710 }
2711
2712 static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs,
2713 Error **errp)
2714 {
2715 BDRVQcow2State *s = bs->opaque;
2716 int flags = s->flags;
2717 QCryptoBlock *crypto = NULL;
2718 QDict *options;
2719 Error *local_err = NULL;
2720 int ret;
2721
2722 /*
2723 * Backing files are read-only which makes all of their metadata immutable,
2724 * that means we don't have to worry about reopening them here.
2725 */
2726
2727 crypto = s->crypto;
2728 s->crypto = NULL;
2729
2730 qcow2_close(bs);
2731
2732 memset(s, 0, sizeof(BDRVQcow2State));
2733 options = qdict_clone_shallow(bs->options);
2734
2735 flags &= ~BDRV_O_INACTIVE;
2736 qemu_co_mutex_lock(&s->lock);
2737 ret = qcow2_do_open(bs, options, flags, &local_err);
2738 qemu_co_mutex_unlock(&s->lock);
2739 qobject_unref(options);
2740 if (local_err) {
2741 error_propagate_prepend(errp, local_err,
2742 "Could not reopen qcow2 layer: ");
2743 bs->drv = NULL;
2744 return;
2745 } else if (ret < 0) {
2746 error_setg_errno(errp, -ret, "Could not reopen qcow2 layer");
2747 bs->drv = NULL;
2748 return;
2749 }
2750
2751 s->crypto = crypto;
2752 }
2753
2754 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
2755 size_t len, size_t buflen)
2756 {
2757 QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
2758 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
2759
2760 if (buflen < ext_len) {
2761 return -ENOSPC;
2762 }
2763
2764 *ext_backing_fmt = (QCowExtension) {
2765 .magic = cpu_to_be32(magic),
2766 .len = cpu_to_be32(len),
2767 };
2768
2769 if (len) {
2770 memcpy(buf + sizeof(QCowExtension), s, len);
2771 }
2772
2773 return ext_len;
2774 }
2775
2776 /*
2777 * Updates the qcow2 header, including the variable length parts of it, i.e.
2778 * the backing file name and all extensions. qcow2 was not designed to allow
2779 * such changes, so if we run out of space (we can only use the first cluster)
2780 * this function may fail.
2781 *
2782 * Returns 0 on success, -errno in error cases.
2783 */
2784 int qcow2_update_header(BlockDriverState *bs)
2785 {
2786 BDRVQcow2State *s = bs->opaque;
2787 QCowHeader *header;
2788 char *buf;
2789 size_t buflen = s->cluster_size;
2790 int ret;
2791 uint64_t total_size;
2792 uint32_t refcount_table_clusters;
2793 size_t header_length;
2794 Qcow2UnknownHeaderExtension *uext;
2795
2796 buf = qemu_blockalign(bs, buflen);
2797
2798 /* Header structure */
2799 header = (QCowHeader*) buf;
2800
2801 if (buflen < sizeof(*header)) {
2802 ret = -ENOSPC;
2803 goto fail;
2804 }
2805
2806 header_length = sizeof(*header) + s->unknown_header_fields_size;
2807 total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
2808 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2809
2810 ret = validate_compression_type(s, NULL);
2811 if (ret) {
2812 goto fail;
2813 }
2814
2815 *header = (QCowHeader) {
2816 /* Version 2 fields */
2817 .magic = cpu_to_be32(QCOW_MAGIC),
2818 .version = cpu_to_be32(s->qcow_version),
2819 .backing_file_offset = 0,
2820 .backing_file_size = 0,
2821 .cluster_bits = cpu_to_be32(s->cluster_bits),
2822 .size = cpu_to_be64(total_size),
2823 .crypt_method = cpu_to_be32(s->crypt_method_header),
2824 .l1_size = cpu_to_be32(s->l1_size),
2825 .l1_table_offset = cpu_to_be64(s->l1_table_offset),
2826 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset),
2827 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
2828 .nb_snapshots = cpu_to_be32(s->nb_snapshots),
2829 .snapshots_offset = cpu_to_be64(s->snapshots_offset),
2830
2831 /* Version 3 fields */
2832 .incompatible_features = cpu_to_be64(s->incompatible_features),
2833 .compatible_features = cpu_to_be64(s->compatible_features),
2834 .autoclear_features = cpu_to_be64(s->autoclear_features),
2835 .refcount_order = cpu_to_be32(s->refcount_order),
2836 .header_length = cpu_to_be32(header_length),
2837 .compression_type = s->compression_type,
2838 };
2839
2840 /* For older versions, write a shorter header */
2841 switch (s->qcow_version) {
2842 case 2:
2843 ret = offsetof(QCowHeader, incompatible_features);
2844 break;
2845 case 3:
2846 ret = sizeof(*header);
2847 break;
2848 default:
2849 ret = -EINVAL;
2850 goto fail;
2851 }
2852
2853 buf += ret;
2854 buflen -= ret;
2855 memset(buf, 0, buflen);
2856
2857 /* Preserve any unknown field in the header */
2858 if (s->unknown_header_fields_size) {
2859 if (buflen < s->unknown_header_fields_size) {
2860 ret = -ENOSPC;
2861 goto fail;
2862 }
2863
2864 memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
2865 buf += s->unknown_header_fields_size;
2866 buflen -= s->unknown_header_fields_size;
2867 }
2868
2869 /* Backing file format header extension */
2870 if (s->image_backing_format) {
2871 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
2872 s->image_backing_format,
2873 strlen(s->image_backing_format),
2874 buflen);
2875 if (ret < 0) {
2876 goto fail;
2877 }
2878
2879 buf += ret;
2880 buflen -= ret;
2881 }
2882
2883 /* External data file header extension */
2884 if (has_data_file(bs) && s->image_data_file) {
2885 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_DATA_FILE,
2886 s->image_data_file, strlen(s->image_data_file),
2887 buflen);
2888 if (ret < 0) {
2889 goto fail;
2890 }
2891
2892 buf += ret;
2893 buflen -= ret;
2894 }
2895
2896 /* Full disk encryption header pointer extension */
2897 if (s->crypto_header.offset != 0) {
2898 s->crypto_header.offset = cpu_to_be64(s->crypto_header.offset);
2899 s->crypto_header.length = cpu_to_be64(s->crypto_header.length);
2900 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_CRYPTO_HEADER,
2901 &s->crypto_header, sizeof(s->crypto_header),
2902 buflen);
2903 s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset);
2904 s->crypto_header.length = be64_to_cpu(s->crypto_header.length);
2905 if (ret < 0) {
2906 goto fail;
2907 }
2908 buf += ret;
2909 buflen -= ret;
2910 }
2911
2912 /*
2913 * Feature table. A mere 8 feature names occupies 392 bytes, and
2914 * when coupled with the v3 minimum header of 104 bytes plus the
2915 * 8-byte end-of-extension marker, that would leave only 8 bytes
2916 * for a backing file name in an image with 512-byte clusters.
2917 * Thus, we choose to omit this header for cluster sizes 4k and
2918 * smaller.
2919 */
2920 if (s->qcow_version >= 3 && s->cluster_size > 4096) {
2921 static const Qcow2Feature features[] = {
2922 {
2923 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2924 .bit = QCOW2_INCOMPAT_DIRTY_BITNR,
2925 .name = "dirty bit",
2926 },
2927 {
2928 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2929 .bit = QCOW2_INCOMPAT_CORRUPT_BITNR,
2930 .name = "corrupt bit",
2931 },
2932 {
2933 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2934 .bit = QCOW2_INCOMPAT_DATA_FILE_BITNR,
2935 .name = "external data file",
2936 },
2937 {
2938 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2939 .bit = QCOW2_INCOMPAT_COMPRESSION_BITNR,
2940 .name = "compression type",
2941 },
2942 {
2943 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2944 .bit = QCOW2_INCOMPAT_EXTL2_BITNR,
2945 .name = "extended L2 entries",
2946 },
2947 {
2948 .type = QCOW2_FEAT_TYPE_COMPATIBLE,
2949 .bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
2950 .name = "lazy refcounts",
2951 },
2952 {
2953 .type = QCOW2_FEAT_TYPE_AUTOCLEAR,
2954 .bit = QCOW2_AUTOCLEAR_BITMAPS_BITNR,
2955 .name = "bitmaps",
2956 },
2957 {
2958 .type = QCOW2_FEAT_TYPE_AUTOCLEAR,
2959 .bit = QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR,
2960 .name = "raw external data",
2961 },
2962 };
2963
2964 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
2965 features, sizeof(features), buflen);
2966 if (ret < 0) {
2967 goto fail;
2968 }
2969 buf += ret;
2970 buflen -= ret;
2971 }
2972
2973 /* Bitmap extension */
2974 if (s->nb_bitmaps > 0) {
2975 Qcow2BitmapHeaderExt bitmaps_header = {
2976 .nb_bitmaps = cpu_to_be32(s->nb_bitmaps),
2977 .bitmap_directory_size =
2978 cpu_to_be64(s->bitmap_directory_size),
2979 .bitmap_directory_offset =
2980 cpu_to_be64(s->bitmap_directory_offset)
2981 };
2982 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BITMAPS,
2983 &bitmaps_header, sizeof(bitmaps_header),
2984 buflen);
2985 if (ret < 0) {
2986 goto fail;
2987 }
2988 buf += ret;
2989 buflen -= ret;
2990 }
2991
2992 /* Keep unknown header extensions */
2993 QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
2994 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
2995 if (ret < 0) {
2996 goto fail;
2997 }
2998
2999 buf += ret;
3000 buflen -= ret;
3001 }
3002
3003 /* End of header extensions */
3004 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
3005 if (ret < 0) {
3006 goto fail;
3007 }
3008
3009 buf += ret;
3010 buflen -= ret;
3011
3012 /* Backing file name */
3013 if (s->image_backing_file) {
3014 size_t backing_file_len = strlen(s->image_backing_file);
3015
3016 if (buflen < backing_file_len) {
3017 ret = -ENOSPC;
3018 goto fail;
3019 }
3020
3021 /* Using strncpy is ok here, since buf is not NUL-terminated. */
3022 strncpy(buf, s->image_backing_file, buflen);
3023
3024 header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
3025 header->backing_file_size = cpu_to_be32(backing_file_len);
3026 }
3027
3028 /* Write the new header */
3029 ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
3030 if (ret < 0) {
3031 goto fail;
3032 }
3033
3034 ret = 0;
3035 fail:
3036 qemu_vfree(header);
3037 return ret;
3038 }
3039
3040 static int qcow2_change_backing_file(BlockDriverState *bs,
3041 const char *backing_file, const char *backing_fmt)
3042 {
3043 BDRVQcow2State *s = bs->opaque;
3044
3045 /* Adding a backing file means that the external data file alone won't be
3046 * enough to make sense of the content */
3047 if (backing_file && data_file_is_raw(bs)) {
3048 return -EINVAL;
3049 }
3050
3051 if (backing_file && strlen(backing_file) > 1023) {
3052 return -EINVAL;
3053 }
3054
3055 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3056 backing_file ?: "");
3057 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
3058 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
3059
3060 g_free(s->image_backing_file);
3061 g_free(s->image_backing_format);
3062
3063 s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
3064 s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
3065
3066 return qcow2_update_header(bs);
3067 }
3068
3069 static int qcow2_set_up_encryption(BlockDriverState *bs,
3070 QCryptoBlockCreateOptions *cryptoopts,
3071 Error **errp)
3072 {
3073 BDRVQcow2State *s = bs->opaque;
3074 QCryptoBlock *crypto = NULL;
3075 int fmt, ret;
3076
3077 switch (cryptoopts->format) {
3078 case Q_CRYPTO_BLOCK_FORMAT_LUKS:
3079 fmt = QCOW_CRYPT_LUKS;
3080 break;
3081 case Q_CRYPTO_BLOCK_FORMAT_QCOW:
3082 fmt = QCOW_CRYPT_AES;
3083 break;
3084 default:
3085 error_setg(errp, "Crypto format not supported in qcow2");
3086 return -EINVAL;
3087 }
3088
3089 s->crypt_method_header = fmt;
3090
3091 crypto = qcrypto_block_create(cryptoopts, "encrypt.",
3092 qcow2_crypto_hdr_init_func,
3093 qcow2_crypto_hdr_write_func,
3094 bs, errp);
3095 if (!crypto) {
3096 return -EINVAL;
3097 }
3098
3099 ret = qcow2_update_header(bs);
3100 if (ret < 0) {
3101 error_setg_errno(errp, -ret, "Could not write encryption header");
3102 goto out;
3103 }
3104
3105 ret = 0;
3106 out:
3107 qcrypto_block_free(crypto);
3108 return ret;
3109 }
3110
3111 /**
3112 * Preallocates metadata structures for data clusters between @offset (in the
3113 * guest disk) and @new_length (which is thus generally the new guest disk
3114 * size).
3115 *
3116 * Returns: 0 on success, -errno on failure.
3117 */
3118 static int coroutine_fn preallocate_co(BlockDriverState *bs, uint64_t offset,
3119 uint64_t new_length, PreallocMode mode,
3120 Error **errp)
3121 {
3122 BDRVQcow2State *s = bs->opaque;
3123 uint64_t bytes;
3124 uint64_t host_offset = 0;
3125 int64_t file_length;
3126 unsigned int cur_bytes;
3127 int ret;
3128 QCowL2Meta *meta;
3129
3130 assert(offset <= new_length);
3131 bytes = new_length - offset;
3132
3133 while (bytes) {
3134 cur_bytes = MIN(bytes, QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size));
3135 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
3136 &host_offset, &meta);
3137 if (ret < 0) {
3138 error_setg_errno(errp, -ret, "Allocating clusters failed");
3139 return ret;
3140 }
3141
3142 while (meta) {
3143 QCowL2Meta *next = meta->next;
3144 meta->prealloc = true;
3145
3146 ret = qcow2_alloc_cluster_link_l2(bs, meta);
3147 if (ret < 0) {
3148 error_setg_errno(errp, -ret, "Mapping clusters failed");
3149 qcow2_free_any_clusters(bs, meta->alloc_offset,
3150 meta->nb_clusters, QCOW2_DISCARD_NEVER);
3151 return ret;
3152 }
3153
3154 /* There are no dependent requests, but we need to remove our
3155 * request from the list of in-flight requests */
3156 QLIST_REMOVE(meta, next_in_flight);
3157
3158 g_free(meta);
3159 meta = next;
3160 }
3161
3162 /* TODO Preallocate data if requested */
3163
3164 bytes -= cur_bytes;
3165 offset += cur_bytes;
3166 }
3167
3168 /*
3169 * It is expected that the image file is large enough to actually contain
3170 * all of the allocated clusters (otherwise we get failing reads after
3171 * EOF). Extend the image to the last allocated sector.
3172 */
3173 file_length = bdrv_getlength(s->data_file->bs);
3174 if (file_length < 0) {
3175 error_setg_errno(errp, -file_length, "Could not get file size");
3176 return file_length;
3177 }
3178
3179 if (host_offset + cur_bytes > file_length) {
3180 if (mode == PREALLOC_MODE_METADATA) {
3181 mode = PREALLOC_MODE_OFF;
3182 }
3183 ret = bdrv_co_truncate(s->data_file, host_offset + cur_bytes, false,
3184 mode, 0, errp);
3185 if (ret < 0) {
3186 return ret;
3187 }
3188 }
3189
3190 return 0;
3191 }
3192
3193 /* qcow2_refcount_metadata_size:
3194 * @clusters: number of clusters to refcount (including data and L1/L2 tables)
3195 * @cluster_size: size of a cluster, in bytes
3196 * @refcount_order: refcount bits power-of-2 exponent
3197 * @generous_increase: allow for the refcount table to be 1.5x as large as it
3198 * needs to be
3199 *
3200 * Returns: Number of bytes required for refcount blocks and table metadata.
3201 */
3202 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size,
3203 int refcount_order, bool generous_increase,
3204 uint64_t *refblock_count)
3205 {
3206 /*
3207 * Every host cluster is reference-counted, including metadata (even
3208 * refcount metadata is recursively included).
3209 *
3210 * An accurate formula for the size of refcount metadata size is difficult
3211 * to derive. An easier method of calculation is finding the fixed point
3212 * where no further refcount blocks or table clusters are required to
3213 * reference count every cluster.
3214 */
3215 int64_t blocks_per_table_cluster = cluster_size / REFTABLE_ENTRY_SIZE;
3216 int64_t refcounts_per_block = cluster_size * 8 / (1 << refcount_order);
3217 int64_t table = 0; /* number of refcount table clusters */
3218 int64_t blocks = 0; /* number of refcount block clusters */
3219 int64_t last;
3220 int64_t n = 0;
3221
3222 do {
3223 last = n;
3224 blocks = DIV_ROUND_UP(clusters + table + blocks, refcounts_per_block);
3225 table = DIV_ROUND_UP(blocks, blocks_per_table_cluster);
3226 n = clusters + blocks + table;
3227
3228 if (n == last && generous_increase) {
3229 clusters += DIV_ROUND_UP(table, 2);
3230 n = 0; /* force another loop */
3231 generous_increase = false;
3232 }
3233 } while (n != last);
3234
3235 if (refblock_count) {
3236 *refblock_count = blocks;
3237 }
3238
3239 return (blocks + table) * cluster_size;
3240 }
3241
3242 /**
3243 * qcow2_calc_prealloc_size:
3244 * @total_size: virtual disk size in bytes
3245 * @cluster_size: cluster size in bytes
3246 * @refcount_order: refcount bits power-of-2 exponent
3247 * @extended_l2: true if the image has extended L2 entries
3248 *
3249 * Returns: Total number of bytes required for the fully allocated image
3250 * (including metadata).
3251 */
3252 static int64_t qcow2_calc_prealloc_size(int64_t total_size,
3253 size_t cluster_size,
3254 int refcount_order,
3255 bool extended_l2)
3256 {
3257 int64_t meta_size = 0;
3258 uint64_t nl1e, nl2e;
3259 int64_t aligned_total_size = ROUND_UP(total_size, cluster_size);
3260 size_t l2e_size = extended_l2 ? L2E_SIZE_EXTENDED : L2E_SIZE_NORMAL;
3261
3262 /* header: 1 cluster */
3263 meta_size += cluster_size;
3264
3265 /* total size of L2 tables */
3266 nl2e = aligned_total_size / cluster_size;
3267 nl2e = ROUND_UP(nl2e, cluster_size / l2e_size);
3268 meta_size += nl2e * l2e_size;
3269
3270 /* total size of L1 tables */
3271 nl1e = nl2e * l2e_size / cluster_size;
3272 nl1e = ROUND_UP(nl1e, cluster_size / L1E_SIZE);
3273 meta_size += nl1e * L1E_SIZE;
3274
3275 /* total size of refcount table and blocks */
3276 meta_size += qcow2_refcount_metadata_size(
3277 (meta_size + aligned_total_size) / cluster_size,
3278 cluster_size, refcount_order, false, NULL);
3279
3280 return meta_size + aligned_total_size;
3281 }
3282
3283 static bool validate_cluster_size(size_t cluster_size, bool extended_l2,
3284 Error **errp)
3285 {
3286 int cluster_bits = ctz32(cluster_size);
3287 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
3288 (1 << cluster_bits) != cluster_size)
3289 {
3290 error_setg(errp, "Cluster size must be a power of two between %d and "
3291 "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
3292 return false;
3293 }
3294
3295 if (extended_l2) {
3296 unsigned min_cluster_size =
3297 (1 << MIN_CLUSTER_BITS) * QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER;
3298 if (cluster_size < min_cluster_size) {
3299 error_setg(errp, "Extended L2 entries are only supported with "
3300 "cluster sizes of at least %u bytes", min_cluster_size);
3301 return false;
3302 }
3303 }
3304
3305 return true;
3306 }
3307
3308 static size_t qcow2_opt_get_cluster_size_del(QemuOpts *opts, bool extended_l2,
3309 Error **errp)
3310 {
3311 size_t cluster_size;
3312
3313 cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
3314 DEFAULT_CLUSTER_SIZE);
3315 if (!validate_cluster_size(cluster_size, extended_l2, errp)) {
3316 return 0;
3317 }
3318 return cluster_size;
3319 }
3320
3321 static int qcow2_opt_get_version_del(QemuOpts *opts, Error **errp)
3322 {
3323 char *buf;
3324 int ret;
3325
3326 buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
3327 if (!buf) {
3328 ret = 3; /* default */
3329 } else if (!strcmp(buf, "0.10")) {
3330 ret = 2;
3331 } else if (!strcmp(buf, "1.1")) {
3332 ret = 3;
3333 } else {
3334 error_setg(errp, "Invalid compatibility level: '%s'", buf);
3335 ret = -EINVAL;
3336 }
3337 g_free(buf);
3338 return ret;
3339 }
3340
3341 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts *opts, int version,
3342 Error **errp)
3343 {
3344 uint64_t refcount_bits;
3345
3346 refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 16);
3347 if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
3348 error_setg(errp, "Refcount width must be a power of two and may not "
3349 "exceed 64 bits");
3350 return 0;
3351 }
3352
3353 if (version < 3 && refcount_bits != 16) {
3354 error_setg(errp, "Different refcount widths than 16 bits require "
3355 "compatibility level 1.1 or above (use compat=1.1 or "
3356 "greater)");
3357 return 0;
3358 }
3359
3360 return refcount_bits;
3361 }
3362
3363 static int coroutine_fn
3364 qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp)
3365 {
3366 BlockdevCreateOptionsQcow2 *qcow2_opts;
3367 QDict *options;
3368
3369 /*
3370 * Open the image file and write a minimal qcow2 header.
3371 *
3372 * We keep things simple and start with a zero-sized image. We also
3373 * do without refcount blocks or a L1 table for now. We'll fix the
3374 * inconsistency later.
3375 *
3376 * We do need a refcount table because growing the refcount table means
3377 * allocating two new refcount blocks - the second of which would be at
3378 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
3379 * size for any qcow2 image.
3380 */
3381 BlockBackend *blk = NULL;
3382 BlockDriverState *bs = NULL;
3383 BlockDriverState *data_bs = NULL;
3384 QCowHeader *header;
3385 size_t cluster_size;
3386 int version;
3387 int refcount_order;
3388 uint64_t* refcount_table;
3389 int ret;
3390 uint8_t compression_type = QCOW2_COMPRESSION_TYPE_ZLIB;
3391
3392 assert(create_options->driver == BLOCKDEV_DRIVER_QCOW2);
3393 qcow2_opts = &create_options->u.qcow2;
3394
3395 bs = bdrv_open_blockdev_ref(qcow2_opts->file, errp);
3396 if (bs == NULL) {
3397 return -EIO;
3398 }
3399
3400 /* Validate options and set default values */
3401 if (!QEMU_IS_ALIGNED(qcow2_opts->size, BDRV_SECTOR_SIZE)) {
3402 error_setg(errp, "Image size must be a multiple of %u bytes",
3403 (unsigned) BDRV_SECTOR_SIZE);
3404 ret = -EINVAL;
3405 goto out;
3406 }
3407
3408 if (qcow2_opts->has_version) {
3409 switch (qcow2_opts->version) {
3410 case BLOCKDEV_QCOW2_VERSION_V2:
3411 version = 2;
3412 break;
3413 case BLOCKDEV_QCOW2_VERSION_V3:
3414 version = 3;
3415 break;