error: Eliminate error_propagate() with Coccinelle, part 2
[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 file_cluster_offset,
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 * sizeof(uint64_t),
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 / sizeof(uint64_t);
1046 r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size,
1047 l2_cache_entry_size);
1048 r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size,
1049 s->cluster_size);
1050 if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) {
1051 error_setg(errp, "Could not allocate metadata caches");
1052 ret = -ENOMEM;
1053 goto fail;
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 /* Check support for various header values */
1448 if (header.refcount_order > 6) {
1449 error_setg(errp, "Reference count entry width too large; may not "
1450 "exceed 64 bits");
1451 ret = -EINVAL;
1452 goto fail;
1453 }
1454 s->refcount_order = header.refcount_order;
1455 s->refcount_bits = 1 << s->refcount_order;
1456 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
1457 s->refcount_max += s->refcount_max - 1;
1458
1459 s->crypt_method_header = header.crypt_method;
1460 if (s->crypt_method_header) {
1461 if (bdrv_uses_whitelist() &&
1462 s->crypt_method_header == QCOW_CRYPT_AES) {
1463 error_setg(errp,
1464 "Use of AES-CBC encrypted qcow2 images is no longer "
1465 "supported in system emulators");
1466 error_append_hint(errp,
1467 "You can use 'qemu-img convert' to convert your "
1468 "image to an alternative supported format, such "
1469 "as unencrypted qcow2, or raw with the LUKS "
1470 "format instead.\n");
1471 ret = -ENOSYS;
1472 goto fail;
1473 }
1474
1475 if (s->crypt_method_header == QCOW_CRYPT_AES) {
1476 s->crypt_physical_offset = false;
1477 } else {
1478 /* Assuming LUKS and any future crypt methods we
1479 * add will all use physical offsets, due to the
1480 * fact that the alternative is insecure... */
1481 s->crypt_physical_offset = true;
1482 }
1483
1484 bs->encrypted = true;
1485 }
1486
1487 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
1488 s->l2_size = 1 << s->l2_bits;
1489 /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1490 s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
1491 s->refcount_block_size = 1 << s->refcount_block_bits;
1492 bs->total_sectors = header.size / BDRV_SECTOR_SIZE;
1493 s->csize_shift = (62 - (s->cluster_bits - 8));
1494 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
1495 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
1496
1497 s->refcount_table_offset = header.refcount_table_offset;
1498 s->refcount_table_size =
1499 header.refcount_table_clusters << (s->cluster_bits - 3);
1500
1501 if (header.refcount_table_clusters == 0 && !(flags & BDRV_O_CHECK)) {
1502 error_setg(errp, "Image does not contain a reference count table");
1503 ret = -EINVAL;
1504 goto fail;
1505 }
1506
1507 ret = qcow2_validate_table(bs, s->refcount_table_offset,
1508 header.refcount_table_clusters,
1509 s->cluster_size, QCOW_MAX_REFTABLE_SIZE,
1510 "Reference count table", errp);
1511 if (ret < 0) {
1512 goto fail;
1513 }
1514
1515 if (!(flags & BDRV_O_CHECK)) {
1516 /*
1517 * The total size in bytes of the snapshot table is checked in
1518 * qcow2_read_snapshots() because the size of each snapshot is
1519 * variable and we don't know it yet.
1520 * Here we only check the offset and number of snapshots.
1521 */
1522 ret = qcow2_validate_table(bs, header.snapshots_offset,
1523 header.nb_snapshots,
1524 sizeof(QCowSnapshotHeader),
1525 sizeof(QCowSnapshotHeader) *
1526 QCOW_MAX_SNAPSHOTS,
1527 "Snapshot table", errp);
1528 if (ret < 0) {
1529 goto fail;
1530 }
1531 }
1532
1533 /* read the level 1 table */
1534 ret = qcow2_validate_table(bs, header.l1_table_offset,
1535 header.l1_size, sizeof(uint64_t),
1536 QCOW_MAX_L1_SIZE, "Active L1 table", errp);
1537 if (ret < 0) {
1538 goto fail;
1539 }
1540 s->l1_size = header.l1_size;
1541 s->l1_table_offset = header.l1_table_offset;
1542
1543 l1_vm_state_index = size_to_l1(s, header.size);
1544 if (l1_vm_state_index > INT_MAX) {
1545 error_setg(errp, "Image is too big");
1546 ret = -EFBIG;
1547 goto fail;
1548 }
1549 s->l1_vm_state_index = l1_vm_state_index;
1550
1551 /* the L1 table must contain at least enough entries to put
1552 header.size bytes */
1553 if (s->l1_size < s->l1_vm_state_index) {
1554 error_setg(errp, "L1 table is too small");
1555 ret = -EINVAL;
1556 goto fail;
1557 }
1558
1559 if (s->l1_size > 0) {
1560 s->l1_table = qemu_try_blockalign(bs->file->bs,
1561 s->l1_size * sizeof(uint64_t));
1562 if (s->l1_table == NULL) {
1563 error_setg(errp, "Could not allocate L1 table");
1564 ret = -ENOMEM;
1565 goto fail;
1566 }
1567 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
1568 s->l1_size * sizeof(uint64_t));
1569 if (ret < 0) {
1570 error_setg_errno(errp, -ret, "Could not read L1 table");
1571 goto fail;
1572 }
1573 for(i = 0;i < s->l1_size; i++) {
1574 s->l1_table[i] = be64_to_cpu(s->l1_table[i]);
1575 }
1576 }
1577
1578 /* Parse driver-specific options */
1579 ret = qcow2_update_options(bs, options, flags, errp);
1580 if (ret < 0) {
1581 goto fail;
1582 }
1583
1584 s->flags = flags;
1585
1586 ret = qcow2_refcount_init(bs);
1587 if (ret != 0) {
1588 error_setg_errno(errp, -ret, "Could not initialize refcount handling");
1589 goto fail;
1590 }
1591
1592 QLIST_INIT(&s->cluster_allocs);
1593 QTAILQ_INIT(&s->discards);
1594
1595 /* read qcow2 extensions */
1596 if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
1597 flags, &update_header, errp)) {
1598 ret = -EINVAL;
1599 goto fail;
1600 }
1601
1602 /* Open external data file */
1603 s->data_file = bdrv_open_child(NULL, options, "data-file", bs,
1604 &child_of_bds, BDRV_CHILD_DATA,
1605 true, &local_err);
1606 if (local_err) {
1607 error_propagate(errp, local_err);
1608 ret = -EINVAL;
1609 goto fail;
1610 }
1611
1612 if (s->incompatible_features & QCOW2_INCOMPAT_DATA_FILE) {
1613 if (!s->data_file && s->image_data_file) {
1614 s->data_file = bdrv_open_child(s->image_data_file, options,
1615 "data-file", bs, &child_of_bds,
1616 BDRV_CHILD_DATA, false, errp);
1617 if (!s->data_file) {
1618 ret = -EINVAL;
1619 goto fail;
1620 }
1621 }
1622 if (!s->data_file) {
1623 error_setg(errp, "'data-file' is required for this image");
1624 ret = -EINVAL;
1625 goto fail;
1626 }
1627
1628 /* No data here */
1629 bs->file->role &= ~BDRV_CHILD_DATA;
1630
1631 /* Must succeed because we have given up permissions if anything */
1632 bdrv_child_refresh_perms(bs, bs->file, &error_abort);
1633 } else {
1634 if (s->data_file) {
1635 error_setg(errp, "'data-file' can only be set for images with an "
1636 "external data file");
1637 ret = -EINVAL;
1638 goto fail;
1639 }
1640
1641 s->data_file = bs->file;
1642
1643 if (data_file_is_raw(bs)) {
1644 error_setg(errp, "data-file-raw requires a data file");
1645 ret = -EINVAL;
1646 goto fail;
1647 }
1648 }
1649
1650 /* qcow2_read_extension may have set up the crypto context
1651 * if the crypt method needs a header region, some methods
1652 * don't need header extensions, so must check here
1653 */
1654 if (s->crypt_method_header && !s->crypto) {
1655 if (s->crypt_method_header == QCOW_CRYPT_AES) {
1656 unsigned int cflags = 0;
1657 if (flags & BDRV_O_NO_IO) {
1658 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
1659 }
1660 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
1661 NULL, NULL, cflags,
1662 QCOW2_MAX_THREADS, errp);
1663 if (!s->crypto) {
1664 ret = -EINVAL;
1665 goto fail;
1666 }
1667 } else if (!(flags & BDRV_O_NO_IO)) {
1668 error_setg(errp, "Missing CRYPTO header for crypt method %d",
1669 s->crypt_method_header);
1670 ret = -EINVAL;
1671 goto fail;
1672 }
1673 }
1674
1675 /* read the backing file name */
1676 if (header.backing_file_offset != 0) {
1677 len = header.backing_file_size;
1678 if (len > MIN(1023, s->cluster_size - header.backing_file_offset) ||
1679 len >= sizeof(bs->backing_file)) {
1680 error_setg(errp, "Backing file name too long");
1681 ret = -EINVAL;
1682 goto fail;
1683 }
1684 ret = bdrv_pread(bs->file, header.backing_file_offset,
1685 bs->auto_backing_file, len);
1686 if (ret < 0) {
1687 error_setg_errno(errp, -ret, "Could not read backing file name");
1688 goto fail;
1689 }
1690 bs->auto_backing_file[len] = '\0';
1691 pstrcpy(bs->backing_file, sizeof(bs->backing_file),
1692 bs->auto_backing_file);
1693 s->image_backing_file = g_strdup(bs->auto_backing_file);
1694 }
1695
1696 /*
1697 * Internal snapshots; skip reading them in check mode, because
1698 * we do not need them then, and we do not want to abort because
1699 * of a broken table.
1700 */
1701 if (!(flags & BDRV_O_CHECK)) {
1702 s->snapshots_offset = header.snapshots_offset;
1703 s->nb_snapshots = header.nb_snapshots;
1704
1705 ret = qcow2_read_snapshots(bs, errp);
1706 if (ret < 0) {
1707 goto fail;
1708 }
1709 }
1710
1711 /* Clear unknown autoclear feature bits */
1712 update_header |= s->autoclear_features & ~QCOW2_AUTOCLEAR_MASK;
1713 update_header =
1714 update_header && !bs->read_only && !(flags & BDRV_O_INACTIVE);
1715 if (update_header) {
1716 s->autoclear_features &= QCOW2_AUTOCLEAR_MASK;
1717 }
1718
1719 /* == Handle persistent dirty bitmaps ==
1720 *
1721 * We want load dirty bitmaps in three cases:
1722 *
1723 * 1. Normal open of the disk in active mode, not related to invalidation
1724 * after migration.
1725 *
1726 * 2. Invalidation of the target vm after pre-copy phase of migration, if
1727 * bitmaps are _not_ migrating through migration channel, i.e.
1728 * 'dirty-bitmaps' capability is disabled.
1729 *
1730 * 3. Invalidation of source vm after failed or canceled migration.
1731 * This is a very interesting case. There are two possible types of
1732 * bitmaps:
1733 *
1734 * A. Stored on inactivation and removed. They should be loaded from the
1735 * image.
1736 *
1737 * B. Not stored: not-persistent bitmaps and bitmaps, migrated through
1738 * the migration channel (with dirty-bitmaps capability).
1739 *
1740 * On the other hand, there are two possible sub-cases:
1741 *
1742 * 3.1 disk was changed by somebody else while were inactive. In this
1743 * case all in-RAM dirty bitmaps (both persistent and not) are
1744 * definitely invalid. And we don't have any method to determine
1745 * this.
1746 *
1747 * Simple and safe thing is to just drop all the bitmaps of type B on
1748 * inactivation. But in this case we lose bitmaps in valid 4.2 case.
1749 *
1750 * On the other hand, resuming source vm, if disk was already changed
1751 * is a bad thing anyway: not only bitmaps, the whole vm state is
1752 * out of sync with disk.
1753 *
1754 * This means, that user or management tool, who for some reason
1755 * decided to resume source vm, after disk was already changed by
1756 * target vm, should at least drop all dirty bitmaps by hand.
1757 *
1758 * So, we can ignore this case for now, but TODO: "generation"
1759 * extension for qcow2, to determine, that image was changed after
1760 * last inactivation. And if it is changed, we will drop (or at least
1761 * mark as 'invalid' all the bitmaps of type B, both persistent
1762 * and not).
1763 *
1764 * 3.2 disk was _not_ changed while were inactive. Bitmaps may be saved
1765 * to disk ('dirty-bitmaps' capability disabled), or not saved
1766 * ('dirty-bitmaps' capability enabled), but we don't need to care
1767 * of: let's load bitmaps as always: stored bitmaps will be loaded,
1768 * and not stored has flag IN_USE=1 in the image and will be skipped
1769 * on loading.
1770 *
1771 * One remaining possible case when we don't want load bitmaps:
1772 *
1773 * 4. Open disk in inactive mode in target vm (bitmaps are migrating or
1774 * will be loaded on invalidation, no needs try loading them before)
1775 */
1776
1777 if (!(bdrv_get_flags(bs) & BDRV_O_INACTIVE)) {
1778 /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */
1779 bool header_updated = qcow2_load_dirty_bitmaps(bs, &local_err);
1780 if (local_err != NULL) {
1781 error_propagate(errp, local_err);
1782 ret = -EINVAL;
1783 goto fail;
1784 }
1785
1786 update_header = update_header && !header_updated;
1787 }
1788
1789 if (update_header) {
1790 ret = qcow2_update_header(bs);
1791 if (ret < 0) {
1792 error_setg_errno(errp, -ret, "Could not update qcow2 header");
1793 goto fail;
1794 }
1795 }
1796
1797 bs->supported_zero_flags = header.version >= 3 ?
1798 BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK : 0;
1799 bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE;
1800
1801 /* Repair image if dirty */
1802 if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only &&
1803 (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
1804 BdrvCheckResult result = {0};
1805
1806 ret = qcow2_co_check_locked(bs, &result,
1807 BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
1808 if (ret < 0 || result.check_errors) {
1809 if (ret >= 0) {
1810 ret = -EIO;
1811 }
1812 error_setg_errno(errp, -ret, "Could not repair dirty image");
1813 goto fail;
1814 }
1815 }
1816
1817 #ifdef DEBUG_ALLOC
1818 {
1819 BdrvCheckResult result = {0};
1820 qcow2_check_refcounts(bs, &result, 0);
1821 }
1822 #endif
1823
1824 qemu_co_queue_init(&s->thread_task_queue);
1825
1826 return ret;
1827
1828 fail:
1829 g_free(s->image_data_file);
1830 if (has_data_file(bs)) {
1831 bdrv_unref_child(bs, s->data_file);
1832 s->data_file = NULL;
1833 }
1834 g_free(s->unknown_header_fields);
1835 cleanup_unknown_header_ext(bs);
1836 qcow2_free_snapshots(bs);
1837 qcow2_refcount_close(bs);
1838 qemu_vfree(s->l1_table);
1839 /* else pre-write overlap checks in cache_destroy may crash */
1840 s->l1_table = NULL;
1841 cache_clean_timer_del(bs);
1842 if (s->l2_table_cache) {
1843 qcow2_cache_destroy(s->l2_table_cache);
1844 }
1845 if (s->refcount_block_cache) {
1846 qcow2_cache_destroy(s->refcount_block_cache);
1847 }
1848 qcrypto_block_free(s->crypto);
1849 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1850 return ret;
1851 }
1852
1853 typedef struct QCow2OpenCo {
1854 BlockDriverState *bs;
1855 QDict *options;
1856 int flags;
1857 Error **errp;
1858 int ret;
1859 } QCow2OpenCo;
1860
1861 static void coroutine_fn qcow2_open_entry(void *opaque)
1862 {
1863 QCow2OpenCo *qoc = opaque;
1864 BDRVQcow2State *s = qoc->bs->opaque;
1865
1866 qemu_co_mutex_lock(&s->lock);
1867 qoc->ret = qcow2_do_open(qoc->bs, qoc->options, qoc->flags, qoc->errp);
1868 qemu_co_mutex_unlock(&s->lock);
1869 }
1870
1871 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
1872 Error **errp)
1873 {
1874 BDRVQcow2State *s = bs->opaque;
1875 QCow2OpenCo qoc = {
1876 .bs = bs,
1877 .options = options,
1878 .flags = flags,
1879 .errp = errp,
1880 .ret = -EINPROGRESS
1881 };
1882
1883 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds,
1884 BDRV_CHILD_IMAGE, false, errp);
1885 if (!bs->file) {
1886 return -EINVAL;
1887 }
1888
1889 /* Initialise locks */
1890 qemu_co_mutex_init(&s->lock);
1891
1892 if (qemu_in_coroutine()) {
1893 /* From bdrv_co_create. */
1894 qcow2_open_entry(&qoc);
1895 } else {
1896 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
1897 qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry, &qoc));
1898 BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS);
1899 }
1900 return qoc.ret;
1901 }
1902
1903 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1904 {
1905 BDRVQcow2State *s = bs->opaque;
1906
1907 if (bs->encrypted) {
1908 /* Encryption works on a sector granularity */
1909 bs->bl.request_alignment = qcrypto_block_get_sector_size(s->crypto);
1910 }
1911 bs->bl.pwrite_zeroes_alignment = s->cluster_size;
1912 bs->bl.pdiscard_alignment = s->cluster_size;
1913 }
1914
1915 static int qcow2_reopen_prepare(BDRVReopenState *state,
1916 BlockReopenQueue *queue, Error **errp)
1917 {
1918 Qcow2ReopenState *r;
1919 int ret;
1920
1921 r = g_new0(Qcow2ReopenState, 1);
1922 state->opaque = r;
1923
1924 ret = qcow2_update_options_prepare(state->bs, r, state->options,
1925 state->flags, errp);
1926 if (ret < 0) {
1927 goto fail;
1928 }
1929
1930 /* We need to write out any unwritten data if we reopen read-only. */
1931 if ((state->flags & BDRV_O_RDWR) == 0) {
1932 ret = qcow2_reopen_bitmaps_ro(state->bs, errp);
1933 if (ret < 0) {
1934 goto fail;
1935 }
1936
1937 ret = bdrv_flush(state->bs);
1938 if (ret < 0) {
1939 goto fail;
1940 }
1941
1942 ret = qcow2_mark_clean(state->bs);
1943 if (ret < 0) {
1944 goto fail;
1945 }
1946 }
1947
1948 return 0;
1949
1950 fail:
1951 qcow2_update_options_abort(state->bs, r);
1952 g_free(r);
1953 return ret;
1954 }
1955
1956 static void qcow2_reopen_commit(BDRVReopenState *state)
1957 {
1958 qcow2_update_options_commit(state->bs, state->opaque);
1959 g_free(state->opaque);
1960 }
1961
1962 static void qcow2_reopen_commit_post(BDRVReopenState *state)
1963 {
1964 if (state->flags & BDRV_O_RDWR) {
1965 Error *local_err = NULL;
1966
1967 if (qcow2_reopen_bitmaps_rw(state->bs, &local_err) < 0) {
1968 /*
1969 * This is not fatal, bitmaps just left read-only, so all following
1970 * writes will fail. User can remove read-only bitmaps to unblock
1971 * writes or retry reopen.
1972 */
1973 error_reportf_err(local_err,
1974 "%s: Failed to make dirty bitmaps writable: ",
1975 bdrv_get_node_name(state->bs));
1976 }
1977 }
1978 }
1979
1980 static void qcow2_reopen_abort(BDRVReopenState *state)
1981 {
1982 qcow2_update_options_abort(state->bs, state->opaque);
1983 g_free(state->opaque);
1984 }
1985
1986 static void qcow2_join_options(QDict *options, QDict *old_options)
1987 {
1988 bool has_new_overlap_template =
1989 qdict_haskey(options, QCOW2_OPT_OVERLAP) ||
1990 qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE);
1991 bool has_new_total_cache_size =
1992 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE);
1993 bool has_all_cache_options;
1994
1995 /* New overlap template overrides all old overlap options */
1996 if (has_new_overlap_template) {
1997 qdict_del(old_options, QCOW2_OPT_OVERLAP);
1998 qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE);
1999 qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER);
2000 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1);
2001 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2);
2002 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE);
2003 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK);
2004 qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE);
2005 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1);
2006 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2);
2007 }
2008
2009 /* New total cache size overrides all old options */
2010 if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) {
2011 qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE);
2012 qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
2013 }
2014
2015 qdict_join(options, old_options, false);
2016
2017 /*
2018 * If after merging all cache size options are set, an old total size is
2019 * overwritten. Do keep all options, however, if all three are new. The
2020 * resulting error message is what we want to happen.
2021 */
2022 has_all_cache_options =
2023 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) ||
2024 qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) ||
2025 qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
2026
2027 if (has_all_cache_options && !has_new_total_cache_size) {
2028 qdict_del(options, QCOW2_OPT_CACHE_SIZE);
2029 }
2030 }
2031
2032 static int coroutine_fn qcow2_co_block_status(BlockDriverState *bs,
2033 bool want_zero,
2034 int64_t offset, int64_t count,
2035 int64_t *pnum, int64_t *map,
2036 BlockDriverState **file)
2037 {
2038 BDRVQcow2State *s = bs->opaque;
2039 uint64_t cluster_offset;
2040 unsigned int bytes;
2041 int ret, status = 0;
2042
2043 qemu_co_mutex_lock(&s->lock);
2044
2045 if (!s->metadata_preallocation_checked) {
2046 ret = qcow2_detect_metadata_preallocation(bs);
2047 s->metadata_preallocation = (ret == 1);
2048 s->metadata_preallocation_checked = true;
2049 }
2050
2051 bytes = MIN(INT_MAX, count);
2052 ret = qcow2_get_cluster_offset(bs, offset, &bytes, &cluster_offset);
2053 qemu_co_mutex_unlock(&s->lock);
2054 if (ret < 0) {
2055 return ret;
2056 }
2057
2058 *pnum = bytes;
2059
2060 if ((ret == QCOW2_CLUSTER_NORMAL || ret == QCOW2_CLUSTER_ZERO_ALLOC) &&
2061 !s->crypto) {
2062 *map = cluster_offset | offset_into_cluster(s, offset);
2063 *file = s->data_file->bs;
2064 status |= BDRV_BLOCK_OFFSET_VALID;
2065 }
2066 if (ret == QCOW2_CLUSTER_ZERO_PLAIN || ret == QCOW2_CLUSTER_ZERO_ALLOC) {
2067 status |= BDRV_BLOCK_ZERO;
2068 } else if (ret != QCOW2_CLUSTER_UNALLOCATED) {
2069 status |= BDRV_BLOCK_DATA;
2070 }
2071 if (s->metadata_preallocation && (status & BDRV_BLOCK_DATA) &&
2072 (status & BDRV_BLOCK_OFFSET_VALID))
2073 {
2074 status |= BDRV_BLOCK_RECURSE;
2075 }
2076 return status;
2077 }
2078
2079 static coroutine_fn int qcow2_handle_l2meta(BlockDriverState *bs,
2080 QCowL2Meta **pl2meta,
2081 bool link_l2)
2082 {
2083 int ret = 0;
2084 QCowL2Meta *l2meta = *pl2meta;
2085
2086 while (l2meta != NULL) {
2087 QCowL2Meta *next;
2088
2089 if (link_l2) {
2090 ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
2091 if (ret) {
2092 goto out;
2093 }
2094 } else {
2095 qcow2_alloc_cluster_abort(bs, l2meta);
2096 }
2097
2098 /* Take the request off the list of running requests */
2099 if (l2meta->nb_clusters != 0) {
2100 QLIST_REMOVE(l2meta, next_in_flight);
2101 }
2102
2103 qemu_co_queue_restart_all(&l2meta->dependent_requests);
2104
2105 next = l2meta->next;
2106 g_free(l2meta);
2107 l2meta = next;
2108 }
2109 out:
2110 *pl2meta = l2meta;
2111 return ret;
2112 }
2113
2114 static coroutine_fn int
2115 qcow2_co_preadv_encrypted(BlockDriverState *bs,
2116 uint64_t file_cluster_offset,
2117 uint64_t offset,
2118 uint64_t bytes,
2119 QEMUIOVector *qiov,
2120 uint64_t qiov_offset)
2121 {
2122 int ret;
2123 BDRVQcow2State *s = bs->opaque;
2124 uint8_t *buf;
2125
2126 assert(bs->encrypted && s->crypto);
2127 assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2128
2129 /*
2130 * For encrypted images, read everything into a temporary
2131 * contiguous buffer on which the AES functions can work.
2132 * Also, decryption in a separate buffer is better as it
2133 * prevents the guest from learning information about the
2134 * encrypted nature of the virtual disk.
2135 */
2136
2137 buf = qemu_try_blockalign(s->data_file->bs, bytes);
2138 if (buf == NULL) {
2139 return -ENOMEM;
2140 }
2141
2142 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
2143 ret = bdrv_co_pread(s->data_file,
2144 file_cluster_offset + offset_into_cluster(s, offset),
2145 bytes, buf, 0);
2146 if (ret < 0) {
2147 goto fail;
2148 }
2149
2150 if (qcow2_co_decrypt(bs,
2151 file_cluster_offset + offset_into_cluster(s, offset),
2152 offset, buf, bytes) < 0)
2153 {
2154 ret = -EIO;
2155 goto fail;
2156 }
2157 qemu_iovec_from_buf(qiov, qiov_offset, buf, bytes);
2158
2159 fail:
2160 qemu_vfree(buf);
2161
2162 return ret;
2163 }
2164
2165 typedef struct Qcow2AioTask {
2166 AioTask task;
2167
2168 BlockDriverState *bs;
2169 QCow2ClusterType cluster_type; /* only for read */
2170 uint64_t file_cluster_offset;
2171 uint64_t offset;
2172 uint64_t bytes;
2173 QEMUIOVector *qiov;
2174 uint64_t qiov_offset;
2175 QCowL2Meta *l2meta; /* only for write */
2176 } Qcow2AioTask;
2177
2178 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task);
2179 static coroutine_fn int qcow2_add_task(BlockDriverState *bs,
2180 AioTaskPool *pool,
2181 AioTaskFunc func,
2182 QCow2ClusterType cluster_type,
2183 uint64_t file_cluster_offset,
2184 uint64_t offset,
2185 uint64_t bytes,
2186 QEMUIOVector *qiov,
2187 size_t qiov_offset,
2188 QCowL2Meta *l2meta)
2189 {
2190 Qcow2AioTask local_task;
2191 Qcow2AioTask *task = pool ? g_new(Qcow2AioTask, 1) : &local_task;
2192
2193 *task = (Qcow2AioTask) {
2194 .task.func = func,
2195 .bs = bs,
2196 .cluster_type = cluster_type,
2197 .qiov = qiov,
2198 .file_cluster_offset = file_cluster_offset,
2199 .offset = offset,
2200 .bytes = bytes,
2201 .qiov_offset = qiov_offset,
2202 .l2meta = l2meta,
2203 };
2204
2205 trace_qcow2_add_task(qemu_coroutine_self(), bs, pool,
2206 func == qcow2_co_preadv_task_entry ? "read" : "write",
2207 cluster_type, file_cluster_offset, offset, bytes,
2208 qiov, qiov_offset);
2209
2210 if (!pool) {
2211 return func(&task->task);
2212 }
2213
2214 aio_task_pool_start_task(pool, &task->task);
2215
2216 return 0;
2217 }
2218
2219 static coroutine_fn int qcow2_co_preadv_task(BlockDriverState *bs,
2220 QCow2ClusterType cluster_type,
2221 uint64_t file_cluster_offset,
2222 uint64_t offset, uint64_t bytes,
2223 QEMUIOVector *qiov,
2224 size_t qiov_offset)
2225 {
2226 BDRVQcow2State *s = bs->opaque;
2227 int offset_in_cluster = offset_into_cluster(s, offset);
2228
2229 switch (cluster_type) {
2230 case QCOW2_CLUSTER_ZERO_PLAIN:
2231 case QCOW2_CLUSTER_ZERO_ALLOC:
2232 /* Both zero types are handled in qcow2_co_preadv_part */
2233 g_assert_not_reached();
2234
2235 case QCOW2_CLUSTER_UNALLOCATED:
2236 assert(bs->backing); /* otherwise handled in qcow2_co_preadv_part */
2237
2238 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
2239 return bdrv_co_preadv_part(bs->backing, offset, bytes,
2240 qiov, qiov_offset, 0);
2241
2242 case QCOW2_CLUSTER_COMPRESSED:
2243 return qcow2_co_preadv_compressed(bs, file_cluster_offset,
2244 offset, bytes, qiov, qiov_offset);
2245
2246 case QCOW2_CLUSTER_NORMAL:
2247 assert(offset_into_cluster(s, file_cluster_offset) == 0);
2248 if (bs->encrypted) {
2249 return qcow2_co_preadv_encrypted(bs, file_cluster_offset,
2250 offset, bytes, qiov, qiov_offset);
2251 }
2252
2253 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
2254 return bdrv_co_preadv_part(s->data_file,
2255 file_cluster_offset + offset_in_cluster,
2256 bytes, qiov, qiov_offset, 0);
2257
2258 default:
2259 g_assert_not_reached();
2260 }
2261
2262 g_assert_not_reached();
2263 }
2264
2265 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task)
2266 {
2267 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
2268
2269 assert(!t->l2meta);
2270
2271 return qcow2_co_preadv_task(t->bs, t->cluster_type, t->file_cluster_offset,
2272 t->offset, t->bytes, t->qiov, t->qiov_offset);
2273 }
2274
2275 static coroutine_fn int qcow2_co_preadv_part(BlockDriverState *bs,
2276 uint64_t offset, uint64_t bytes,
2277 QEMUIOVector *qiov,
2278 size_t qiov_offset, int flags)
2279 {
2280 BDRVQcow2State *s = bs->opaque;
2281 int ret = 0;
2282 unsigned int cur_bytes; /* number of bytes in current iteration */
2283 uint64_t cluster_offset = 0;
2284 AioTaskPool *aio = NULL;
2285
2286 while (bytes != 0 && aio_task_pool_status(aio) == 0) {
2287 /* prepare next request */
2288 cur_bytes = MIN(bytes, INT_MAX);
2289 if (s->crypto) {
2290 cur_bytes = MIN(cur_bytes,
2291 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2292 }
2293
2294 qemu_co_mutex_lock(&s->lock);
2295 ret = qcow2_get_cluster_offset(bs, offset, &cur_bytes, &cluster_offset);
2296 qemu_co_mutex_unlock(&s->lock);
2297 if (ret < 0) {
2298 goto out;
2299 }
2300
2301 if (ret == QCOW2_CLUSTER_ZERO_PLAIN ||
2302 ret == QCOW2_CLUSTER_ZERO_ALLOC ||
2303 (ret == QCOW2_CLUSTER_UNALLOCATED && !bs->backing))
2304 {
2305 qemu_iovec_memset(qiov, qiov_offset, 0, cur_bytes);
2306 } else {
2307 if (!aio && cur_bytes != bytes) {
2308 aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
2309 }
2310 ret = qcow2_add_task(bs, aio, qcow2_co_preadv_task_entry, ret,
2311 cluster_offset, offset, cur_bytes,
2312 qiov, qiov_offset, NULL);
2313 if (ret < 0) {
2314 goto out;
2315 }
2316 }
2317
2318 bytes -= cur_bytes;
2319 offset += cur_bytes;
2320 qiov_offset += cur_bytes;
2321 }
2322
2323 out:
2324 if (aio) {
2325 aio_task_pool_wait_all(aio);
2326 if (ret == 0) {
2327 ret = aio_task_pool_status(aio);
2328 }
2329 g_free(aio);
2330 }
2331
2332 return ret;
2333 }
2334
2335 /* Check if it's possible to merge a write request with the writing of
2336 * the data from the COW regions */
2337 static bool merge_cow(uint64_t offset, unsigned bytes,
2338 QEMUIOVector *qiov, size_t qiov_offset,
2339 QCowL2Meta *l2meta)
2340 {
2341 QCowL2Meta *m;
2342
2343 for (m = l2meta; m != NULL; m = m->next) {
2344 /* If both COW regions are empty then there's nothing to merge */
2345 if (m->cow_start.nb_bytes == 0 && m->cow_end.nb_bytes == 0) {
2346 continue;
2347 }
2348
2349 /* If COW regions are handled already, skip this too */
2350 if (m->skip_cow) {
2351 continue;
2352 }
2353
2354 /* The data (middle) region must be immediately after the
2355 * start region */
2356 if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) {
2357 continue;
2358 }
2359
2360 /* The end region must be immediately after the data (middle)
2361 * region */
2362 if (m->offset + m->cow_end.offset != offset + bytes) {
2363 continue;
2364 }
2365
2366 /* Make sure that adding both COW regions to the QEMUIOVector
2367 * does not exceed IOV_MAX */
2368 if (qemu_iovec_subvec_niov(qiov, qiov_offset, bytes) > IOV_MAX - 2) {
2369 continue;
2370 }
2371
2372 m->data_qiov = qiov;
2373 m->data_qiov_offset = qiov_offset;
2374 return true;
2375 }
2376
2377 return false;
2378 }
2379
2380 static bool is_unallocated(BlockDriverState *bs, int64_t offset, int64_t bytes)
2381 {
2382 int64_t nr;
2383 return !bytes ||
2384 (!bdrv_is_allocated_above(bs, NULL, false, offset, bytes, &nr) &&
2385 nr == bytes);
2386 }
2387
2388 static bool is_zero_cow(BlockDriverState *bs, QCowL2Meta *m)
2389 {
2390 /*
2391 * This check is designed for optimization shortcut so it must be
2392 * efficient.
2393 * Instead of is_zero(), use is_unallocated() as it is faster (but not
2394 * as accurate and can result in false negatives).
2395 */
2396 return is_unallocated(bs, m->offset + m->cow_start.offset,
2397 m->cow_start.nb_bytes) &&
2398 is_unallocated(bs, m->offset + m->cow_end.offset,
2399 m->cow_end.nb_bytes);
2400 }
2401
2402 static int handle_alloc_space(BlockDriverState *bs, QCowL2Meta *l2meta)
2403 {
2404 BDRVQcow2State *s = bs->opaque;
2405 QCowL2Meta *m;
2406
2407 if (!(s->data_file->bs->supported_zero_flags & BDRV_REQ_NO_FALLBACK)) {
2408 return 0;
2409 }
2410
2411 if (bs->encrypted) {
2412 return 0;
2413 }
2414
2415 for (m = l2meta; m != NULL; m = m->next) {
2416 int ret;
2417
2418 if (!m->cow_start.nb_bytes && !m->cow_end.nb_bytes) {
2419 continue;
2420 }
2421
2422 if (!is_zero_cow(bs, m)) {
2423 continue;
2424 }
2425
2426 /*
2427 * instead of writing zero COW buffers,
2428 * efficiently zero out the whole clusters
2429 */
2430
2431 ret = qcow2_pre_write_overlap_check(bs, 0, m->alloc_offset,
2432 m->nb_clusters * s->cluster_size,
2433 true);
2434 if (ret < 0) {
2435 return ret;
2436 }
2437
2438 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_SPACE);
2439 ret = bdrv_co_pwrite_zeroes(s->data_file, m->alloc_offset,
2440 m->nb_clusters * s->cluster_size,
2441 BDRV_REQ_NO_FALLBACK);
2442 if (ret < 0) {
2443 if (ret != -ENOTSUP && ret != -EAGAIN) {
2444 return ret;
2445 }
2446 continue;
2447 }
2448
2449 trace_qcow2_skip_cow(qemu_coroutine_self(), m->offset, m->nb_clusters);
2450 m->skip_cow = true;
2451 }
2452 return 0;
2453 }
2454
2455 /*
2456 * qcow2_co_pwritev_task
2457 * Called with s->lock unlocked
2458 * l2meta - if not NULL, qcow2_co_pwritev_task() will consume it. Caller must
2459 * not use it somehow after qcow2_co_pwritev_task() call
2460 */
2461 static coroutine_fn int qcow2_co_pwritev_task(BlockDriverState *bs,
2462 uint64_t file_cluster_offset,
2463 uint64_t offset, uint64_t bytes,
2464 QEMUIOVector *qiov,
2465 uint64_t qiov_offset,
2466 QCowL2Meta *l2meta)
2467 {
2468 int ret;
2469 BDRVQcow2State *s = bs->opaque;
2470 void *crypt_buf = NULL;
2471 int offset_in_cluster = offset_into_cluster(s, offset);
2472 QEMUIOVector encrypted_qiov;
2473
2474 if (bs->encrypted) {
2475 assert(s->crypto);
2476 assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2477 crypt_buf = qemu_try_blockalign(bs->file->bs, bytes);
2478 if (crypt_buf == NULL) {
2479 ret = -ENOMEM;
2480 goto out_unlocked;
2481 }
2482 qemu_iovec_to_buf(qiov, qiov_offset, crypt_buf, bytes);
2483
2484 if (qcow2_co_encrypt(bs, file_cluster_offset + offset_in_cluster,
2485 offset, crypt_buf, bytes) < 0)
2486 {
2487 ret = -EIO;
2488 goto out_unlocked;
2489 }
2490
2491 qemu_iovec_init_buf(&encrypted_qiov, crypt_buf, bytes);
2492 qiov = &encrypted_qiov;
2493 qiov_offset = 0;
2494 }
2495
2496 /* Try to efficiently initialize the physical space with zeroes */
2497 ret = handle_alloc_space(bs, l2meta);
2498 if (ret < 0) {
2499 goto out_unlocked;
2500 }
2501
2502 /*
2503 * If we need to do COW, check if it's possible to merge the
2504 * writing of the guest data together with that of the COW regions.
2505 * If it's not possible (or not necessary) then write the
2506 * guest data now.
2507 */
2508 if (!merge_cow(offset, bytes, qiov, qiov_offset, l2meta)) {
2509 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
2510 trace_qcow2_writev_data(qemu_coroutine_self(),
2511 file_cluster_offset + offset_in_cluster);
2512 ret = bdrv_co_pwritev_part(s->data_file,
2513 file_cluster_offset + offset_in_cluster,
2514 bytes, qiov, qiov_offset, 0);
2515 if (ret < 0) {
2516 goto out_unlocked;
2517 }
2518 }
2519
2520 qemu_co_mutex_lock(&s->lock);
2521
2522 ret = qcow2_handle_l2meta(bs, &l2meta, true);
2523 goto out_locked;
2524
2525 out_unlocked:
2526 qemu_co_mutex_lock(&s->lock);
2527
2528 out_locked:
2529 qcow2_handle_l2meta(bs, &l2meta, false);
2530 qemu_co_mutex_unlock(&s->lock);
2531
2532 qemu_vfree(crypt_buf);
2533
2534 return ret;
2535 }
2536
2537 static coroutine_fn int qcow2_co_pwritev_task_entry(AioTask *task)
2538 {
2539 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
2540
2541 assert(!t->cluster_type);
2542
2543 return qcow2_co_pwritev_task(t->bs, t->file_cluster_offset,
2544 t->offset, t->bytes, t->qiov, t->qiov_offset,
2545 t->l2meta);
2546 }
2547
2548 static coroutine_fn int qcow2_co_pwritev_part(
2549 BlockDriverState *bs, uint64_t offset, uint64_t bytes,
2550 QEMUIOVector *qiov, size_t qiov_offset, int flags)
2551 {
2552 BDRVQcow2State *s = bs->opaque;
2553 int offset_in_cluster;
2554 int ret;
2555 unsigned int cur_bytes; /* number of sectors in current iteration */
2556 uint64_t cluster_offset;
2557 QCowL2Meta *l2meta = NULL;
2558 AioTaskPool *aio = NULL;
2559
2560 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes);
2561
2562 while (bytes != 0 && aio_task_pool_status(aio) == 0) {
2563
2564 l2meta = NULL;
2565
2566 trace_qcow2_writev_start_part(qemu_coroutine_self());
2567 offset_in_cluster = offset_into_cluster(s, offset);
2568 cur_bytes = MIN(bytes, INT_MAX);
2569 if (bs->encrypted) {
2570 cur_bytes = MIN(cur_bytes,
2571 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
2572 - offset_in_cluster);
2573 }
2574
2575 qemu_co_mutex_lock(&s->lock);
2576
2577 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
2578 &cluster_offset, &l2meta);
2579 if (ret < 0) {
2580 goto out_locked;
2581 }
2582
2583 assert(offset_into_cluster(s, cluster_offset) == 0);
2584
2585 ret = qcow2_pre_write_overlap_check(bs, 0,
2586 cluster_offset + offset_in_cluster,
2587 cur_bytes, true);
2588 if (ret < 0) {
2589 goto out_locked;
2590 }
2591
2592 qemu_co_mutex_unlock(&s->lock);
2593
2594 if (!aio && cur_bytes != bytes) {
2595 aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
2596 }
2597 ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_task_entry, 0,
2598 cluster_offset, offset, cur_bytes,
2599 qiov, qiov_offset, l2meta);
2600 l2meta = NULL; /* l2meta is consumed by qcow2_co_pwritev_task() */
2601 if (ret < 0) {
2602 goto fail_nometa;
2603 }
2604
2605 bytes -= cur_bytes;
2606 offset += cur_bytes;
2607 qiov_offset += cur_bytes;
2608 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes);
2609 }
2610 ret = 0;
2611
2612 qemu_co_mutex_lock(&s->lock);
2613
2614 out_locked:
2615 qcow2_handle_l2meta(bs, &l2meta, false);
2616
2617 qemu_co_mutex_unlock(&s->lock);
2618
2619 fail_nometa:
2620 if (aio) {
2621 aio_task_pool_wait_all(aio);
2622 if (ret == 0) {
2623 ret = aio_task_pool_status(aio);
2624 }
2625 g_free(aio);
2626 }
2627
2628 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
2629
2630 return ret;
2631 }
2632
2633 static int qcow2_inactivate(BlockDriverState *bs)
2634 {
2635 BDRVQcow2State *s = bs->opaque;
2636 int ret, result = 0;
2637 Error *local_err = NULL;
2638
2639 qcow2_store_persistent_dirty_bitmaps(bs, true, &local_err);
2640 if (local_err != NULL) {
2641 result = -EINVAL;
2642 error_reportf_err(local_err, "Lost persistent bitmaps during "
2643 "inactivation of node '%s': ",
2644 bdrv_get_device_or_node_name(bs));
2645 }
2646
2647 ret = qcow2_cache_flush(bs, s->l2_table_cache);
2648 if (ret) {
2649 result = ret;
2650 error_report("Failed to flush the L2 table cache: %s",
2651 strerror(-ret));
2652 }
2653
2654 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2655 if (ret) {
2656 result = ret;
2657 error_report("Failed to flush the refcount block cache: %s",
2658 strerror(-ret));
2659 }
2660
2661 if (result == 0) {
2662 qcow2_mark_clean(bs);
2663 }
2664
2665 return result;
2666 }
2667
2668 static void qcow2_close(BlockDriverState *bs)
2669 {
2670 BDRVQcow2State *s = bs->opaque;
2671 qemu_vfree(s->l1_table);
2672 /* else pre-write overlap checks in cache_destroy may crash */
2673 s->l1_table = NULL;
2674
2675 if (!(s->flags & BDRV_O_INACTIVE)) {
2676 qcow2_inactivate(bs);
2677 }
2678
2679 cache_clean_timer_del(bs);
2680 qcow2_cache_destroy(s->l2_table_cache);
2681 qcow2_cache_destroy(s->refcount_block_cache);
2682
2683 qcrypto_block_free(s->crypto);
2684 s->crypto = NULL;
2685 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
2686
2687 g_free(s->unknown_header_fields);
2688 cleanup_unknown_header_ext(bs);
2689
2690 g_free(s->image_data_file);
2691 g_free(s->image_backing_file);
2692 g_free(s->image_backing_format);
2693
2694 if (has_data_file(bs)) {
2695 bdrv_unref_child(bs, s->data_file);
2696 s->data_file = NULL;
2697 }
2698
2699 qcow2_refcount_close(bs);
2700 qcow2_free_snapshots(bs);
2701 }
2702
2703 static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs,
2704 Error **errp)
2705 {
2706 BDRVQcow2State *s = bs->opaque;
2707 int flags = s->flags;
2708 QCryptoBlock *crypto = NULL;
2709 QDict *options;
2710 Error *local_err = NULL;
2711 int ret;
2712
2713 /*
2714 * Backing files are read-only which makes all of their metadata immutable,
2715 * that means we don't have to worry about reopening them here.
2716 */
2717
2718 crypto = s->crypto;
2719 s->crypto = NULL;
2720
2721 qcow2_close(bs);
2722
2723 memset(s, 0, sizeof(BDRVQcow2State));
2724 options = qdict_clone_shallow(bs->options);
2725
2726 flags &= ~BDRV_O_INACTIVE;
2727 qemu_co_mutex_lock(&s->lock);
2728 ret = qcow2_do_open(bs, options, flags, &local_err);
2729 qemu_co_mutex_unlock(&s->lock);
2730 qobject_unref(options);
2731 if (local_err) {
2732 error_propagate_prepend(errp, local_err,
2733 "Could not reopen qcow2 layer: ");
2734 bs->drv = NULL;
2735 return;
2736 } else if (ret < 0) {
2737 error_setg_errno(errp, -ret, "Could not reopen qcow2 layer");
2738 bs->drv = NULL;
2739 return;
2740 }
2741
2742 s->crypto = crypto;
2743 }
2744
2745 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
2746 size_t len, size_t buflen)
2747 {
2748 QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
2749 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
2750
2751 if (buflen < ext_len) {
2752 return -ENOSPC;
2753 }
2754
2755 *ext_backing_fmt = (QCowExtension) {
2756 .magic = cpu_to_be32(magic),
2757 .len = cpu_to_be32(len),
2758 };
2759
2760 if (len) {
2761 memcpy(buf + sizeof(QCowExtension), s, len);
2762 }
2763
2764 return ext_len;
2765 }
2766
2767 /*
2768 * Updates the qcow2 header, including the variable length parts of it, i.e.
2769 * the backing file name and all extensions. qcow2 was not designed to allow
2770 * such changes, so if we run out of space (we can only use the first cluster)
2771 * this function may fail.
2772 *
2773 * Returns 0 on success, -errno in error cases.
2774 */
2775 int qcow2_update_header(BlockDriverState *bs)
2776 {
2777 BDRVQcow2State *s = bs->opaque;
2778 QCowHeader *header;
2779 char *buf;
2780 size_t buflen = s->cluster_size;
2781 int ret;
2782 uint64_t total_size;
2783 uint32_t refcount_table_clusters;
2784 size_t header_length;
2785 Qcow2UnknownHeaderExtension *uext;
2786
2787 buf = qemu_blockalign(bs, buflen);
2788
2789 /* Header structure */
2790 header = (QCowHeader*) buf;
2791
2792 if (buflen < sizeof(*header)) {
2793 ret = -ENOSPC;
2794 goto fail;
2795 }
2796
2797 header_length = sizeof(*header) + s->unknown_header_fields_size;
2798 total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
2799 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2800
2801 ret = validate_compression_type(s, NULL);
2802 if (ret) {
2803 goto fail;
2804 }
2805
2806 *header = (QCowHeader) {
2807 /* Version 2 fields */
2808 .magic = cpu_to_be32(QCOW_MAGIC),
2809 .version = cpu_to_be32(s->qcow_version),
2810 .backing_file_offset = 0,
2811 .backing_file_size = 0,
2812 .cluster_bits = cpu_to_be32(s->cluster_bits),
2813 .size = cpu_to_be64(total_size),
2814 .crypt_method = cpu_to_be32(s->crypt_method_header),
2815 .l1_size = cpu_to_be32(s->l1_size),
2816 .l1_table_offset = cpu_to_be64(s->l1_table_offset),
2817 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset),
2818 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
2819 .nb_snapshots = cpu_to_be32(s->nb_snapshots),
2820 .snapshots_offset = cpu_to_be64(s->snapshots_offset),
2821
2822 /* Version 3 fields */
2823 .incompatible_features = cpu_to_be64(s->incompatible_features),
2824 .compatible_features = cpu_to_be64(s->compatible_features),
2825 .autoclear_features = cpu_to_be64(s->autoclear_features),
2826 .refcount_order = cpu_to_be32(s->refcount_order),
2827 .header_length = cpu_to_be32(header_length),
2828 .compression_type = s->compression_type,
2829 };
2830
2831 /* For older versions, write a shorter header */
2832 switch (s->qcow_version) {
2833 case 2:
2834 ret = offsetof(QCowHeader, incompatible_features);
2835 break;
2836 case 3:
2837 ret = sizeof(*header);
2838 break;
2839 default:
2840 ret = -EINVAL;
2841 goto fail;
2842 }
2843
2844 buf += ret;
2845 buflen -= ret;
2846 memset(buf, 0, buflen);
2847
2848 /* Preserve any unknown field in the header */
2849 if (s->unknown_header_fields_size) {
2850 if (buflen < s->unknown_header_fields_size) {
2851 ret = -ENOSPC;
2852 goto fail;
2853 }
2854
2855 memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
2856 buf += s->unknown_header_fields_size;
2857 buflen -= s->unknown_header_fields_size;
2858 }
2859
2860 /* Backing file format header extension */
2861 if (s->image_backing_format) {
2862 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
2863 s->image_backing_format,
2864 strlen(s->image_backing_format),
2865 buflen);
2866 if (ret < 0) {
2867 goto fail;
2868 }
2869
2870 buf += ret;
2871 buflen -= ret;
2872 }
2873
2874 /* External data file header extension */
2875 if (has_data_file(bs) && s->image_data_file) {
2876 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_DATA_FILE,
2877 s->image_data_file, strlen(s->image_data_file),
2878 buflen);
2879 if (ret < 0) {
2880 goto fail;
2881 }
2882
2883 buf += ret;
2884 buflen -= ret;
2885 }
2886
2887 /* Full disk encryption header pointer extension */
2888 if (s->crypto_header.offset != 0) {
2889 s->crypto_header.offset = cpu_to_be64(s->crypto_header.offset);
2890 s->crypto_header.length = cpu_to_be64(s->crypto_header.length);
2891 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_CRYPTO_HEADER,
2892 &s->crypto_header, sizeof(s->crypto_header),
2893 buflen);
2894 s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset);
2895 s->crypto_header.length = be64_to_cpu(s->crypto_header.length);
2896 if (ret < 0) {
2897 goto fail;
2898 }
2899 buf += ret;
2900 buflen -= ret;
2901 }
2902
2903 /*
2904 * Feature table. A mere 8 feature names occupies 392 bytes, and
2905 * when coupled with the v3 minimum header of 104 bytes plus the
2906 * 8-byte end-of-extension marker, that would leave only 8 bytes
2907 * for a backing file name in an image with 512-byte clusters.
2908 * Thus, we choose to omit this header for cluster sizes 4k and
2909 * smaller.
2910 */
2911 if (s->qcow_version >= 3 && s->cluster_size > 4096) {
2912 static const Qcow2Feature features[] = {
2913 {
2914 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2915 .bit = QCOW2_INCOMPAT_DIRTY_BITNR,
2916 .name = "dirty bit",
2917 },
2918 {
2919 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2920 .bit = QCOW2_INCOMPAT_CORRUPT_BITNR,
2921 .name = "corrupt bit",
2922 },
2923 {
2924 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2925 .bit = QCOW2_INCOMPAT_DATA_FILE_BITNR,
2926 .name = "external data file",
2927 },
2928 {
2929 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2930 .bit = QCOW2_INCOMPAT_COMPRESSION_BITNR,
2931 .name = "compression type",
2932 },
2933 {
2934 .type = QCOW2_FEAT_TYPE_COMPATIBLE,
2935 .bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
2936 .name = "lazy refcounts",
2937 },
2938 {
2939 .type = QCOW2_FEAT_TYPE_AUTOCLEAR,
2940 .bit = QCOW2_AUTOCLEAR_BITMAPS_BITNR,
2941 .name = "bitmaps",
2942 },
2943 {
2944 .type = QCOW2_FEAT_TYPE_AUTOCLEAR,
2945 .bit = QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR,
2946 .name = "raw external data",
2947 },
2948 };
2949
2950 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
2951 features, sizeof(features), buflen);
2952 if (ret < 0) {
2953 goto fail;
2954 }
2955 buf += ret;
2956 buflen -= ret;
2957 }
2958
2959 /* Bitmap extension */
2960 if (s->nb_bitmaps > 0) {
2961 Qcow2BitmapHeaderExt bitmaps_header = {
2962 .nb_bitmaps = cpu_to_be32(s->nb_bitmaps),
2963 .bitmap_directory_size =
2964 cpu_to_be64(s->bitmap_directory_size),
2965 .bitmap_directory_offset =
2966 cpu_to_be64(s->bitmap_directory_offset)
2967 };
2968 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BITMAPS,
2969 &bitmaps_header, sizeof(bitmaps_header),
2970 buflen);
2971 if (ret < 0) {
2972 goto fail;
2973 }
2974 buf += ret;
2975 buflen -= ret;
2976 }
2977
2978 /* Keep unknown header extensions */
2979 QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
2980 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
2981 if (ret < 0) {
2982 goto fail;
2983 }
2984
2985 buf += ret;
2986 buflen -= ret;
2987 }
2988
2989 /* End of header extensions */
2990 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
2991 if (ret < 0) {
2992 goto fail;
2993 }
2994
2995 buf += ret;
2996 buflen -= ret;
2997
2998 /* Backing file name */
2999 if (s->image_backing_file) {
3000 size_t backing_file_len = strlen(s->image_backing_file);
3001
3002 if (buflen < backing_file_len) {
3003 ret = -ENOSPC;
3004 goto fail;
3005 }
3006
3007 /* Using strncpy is ok here, since buf is not NUL-terminated. */
3008 strncpy(buf, s->image_backing_file, buflen);
3009
3010 header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
3011 header->backing_file_size = cpu_to_be32(backing_file_len);
3012 }
3013
3014 /* Write the new header */
3015 ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
3016 if (ret < 0) {
3017 goto fail;
3018 }
3019
3020 ret = 0;
3021 fail:
3022 qemu_vfree(header);
3023 return ret;
3024 }
3025
3026 static int qcow2_change_backing_file(BlockDriverState *bs,
3027 const char *backing_file, const char *backing_fmt)
3028 {
3029 BDRVQcow2State *s = bs->opaque;
3030
3031 /* Adding a backing file means that the external data file alone won't be
3032 * enough to make sense of the content */
3033 if (backing_file && data_file_is_raw(bs)) {
3034 return -EINVAL;
3035 }
3036
3037 if (backing_file && strlen(backing_file) > 1023) {
3038 return -EINVAL;
3039 }
3040
3041 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3042 backing_file ?: "");
3043 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
3044 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
3045
3046 g_free(s->image_backing_file);
3047 g_free(s->image_backing_format);
3048
3049 s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
3050 s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
3051
3052 return qcow2_update_header(bs);
3053 }
3054
3055 static int qcow2_set_up_encryption(BlockDriverState *bs,
3056 QCryptoBlockCreateOptions *cryptoopts,
3057 Error **errp)
3058 {
3059 BDRVQcow2State *s = bs->opaque;
3060 QCryptoBlock *crypto = NULL;
3061 int fmt, ret;
3062
3063 switch (cryptoopts->format) {
3064 case Q_CRYPTO_BLOCK_FORMAT_LUKS:
3065 fmt = QCOW_CRYPT_LUKS;
3066 break;
3067 case Q_CRYPTO_BLOCK_FORMAT_QCOW:
3068 fmt = QCOW_CRYPT_AES;
3069 break;
3070 default:
3071 error_setg(errp, "Crypto format not supported in qcow2");
3072 return -EINVAL;
3073 }
3074
3075 s->crypt_method_header = fmt;
3076
3077 crypto = qcrypto_block_create(cryptoopts, "encrypt.",
3078 qcow2_crypto_hdr_init_func,
3079 qcow2_crypto_hdr_write_func,
3080 bs, errp);
3081 if (!crypto) {
3082 return -EINVAL;
3083 }
3084
3085 ret = qcow2_update_header(bs);
3086 if (ret < 0) {
3087 error_setg_errno(errp, -ret, "Could not write encryption header");
3088 goto out;
3089 }
3090
3091 ret = 0;
3092 out:
3093 qcrypto_block_free(crypto);
3094 return ret;
3095 }
3096
3097 /**
3098 * Preallocates metadata structures for data clusters between @offset (in the
3099 * guest disk) and @new_length (which is thus generally the new guest disk
3100 * size).
3101 *
3102 * Returns: 0 on success, -errno on failure.
3103 */
3104 static int coroutine_fn preallocate_co(BlockDriverState *bs, uint64_t offset,
3105 uint64_t new_length, PreallocMode mode,
3106 Error **errp)
3107 {
3108 BDRVQcow2State *s = bs->opaque;
3109 uint64_t bytes;
3110 uint64_t host_offset = 0;
3111 int64_t file_length;
3112 unsigned int cur_bytes;
3113 int ret;
3114 QCowL2Meta *meta;
3115
3116 assert(offset <= new_length);
3117 bytes = new_length - offset;
3118
3119 while (bytes) {
3120 cur_bytes = MIN(bytes, QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size));
3121 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
3122 &host_offset, &meta);
3123 if (ret < 0) {
3124 error_setg_errno(errp, -ret, "Allocating clusters failed");
3125 return ret;
3126 }
3127
3128 while (meta) {
3129 QCowL2Meta *next = meta->next;
3130
3131 ret = qcow2_alloc_cluster_link_l2(bs, meta);
3132 if (ret < 0) {
3133 error_setg_errno(errp, -ret, "Mapping clusters failed");
3134 qcow2_free_any_clusters(bs, meta->alloc_offset,
3135 meta->nb_clusters, QCOW2_DISCARD_NEVER);
3136 return ret;
3137 }
3138
3139 /* There are no dependent requests, but we need to remove our
3140 * request from the list of in-flight requests */
3141 QLIST_REMOVE(meta, next_in_flight);
3142
3143 g_free(meta);
3144 meta = next;
3145 }
3146
3147 /* TODO Preallocate data if requested */
3148
3149 bytes -= cur_bytes;
3150 offset += cur_bytes;
3151 }
3152
3153 /*
3154 * It is expected that the image file is large enough to actually contain
3155 * all of the allocated clusters (otherwise we get failing reads after
3156 * EOF). Extend the image to the last allocated sector.
3157 */
3158 file_length = bdrv_getlength(s->data_file->bs);
3159 if (file_length < 0) {
3160 error_setg_errno(errp, -file_length, "Could not get file size");
3161 return file_length;
3162 }
3163
3164 if (host_offset + cur_bytes > file_length) {
3165 if (mode == PREALLOC_MODE_METADATA) {
3166 mode = PREALLOC_MODE_OFF;
3167 }
3168 ret = bdrv_co_truncate(s->data_file, host_offset + cur_bytes, false,
3169 mode, 0, errp);
3170 if (ret < 0) {
3171 return ret;
3172 }
3173 }
3174
3175 return 0;
3176 }
3177
3178 /* qcow2_refcount_metadata_size:
3179 * @clusters: number of clusters to refcount (including data and L1/L2 tables)
3180 * @cluster_size: size of a cluster, in bytes
3181 * @refcount_order: refcount bits power-of-2 exponent
3182 * @generous_increase: allow for the refcount table to be 1.5x as large as it
3183 * needs to be
3184 *
3185 * Returns: Number of bytes required for refcount blocks and table metadata.
3186 */
3187 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size,
3188 int refcount_order, bool generous_increase,
3189 uint64_t *refblock_count)
3190 {
3191 /*
3192 * Every host cluster is reference-counted, including metadata (even
3193 * refcount metadata is recursively included).
3194 *
3195 * An accurate formula for the size of refcount metadata size is difficult
3196 * to derive. An easier method of calculation is finding the fixed point
3197 * where no further refcount blocks or table clusters are required to
3198 * reference count every cluster.
3199 */
3200 int64_t blocks_per_table_cluster = cluster_size / sizeof(uint64_t);
3201 int64_t refcounts_per_block = cluster_size * 8 / (1 << refcount_order);
3202 int64_t table = 0; /* number of refcount table clusters */
3203 int64_t blocks = 0; /* number of refcount block clusters */
3204 int64_t last;
3205 int64_t n = 0;
3206
3207 do {
3208 last = n;
3209 blocks = DIV_ROUND_UP(clusters + table + blocks, refcounts_per_block);
3210 table = DIV_ROUND_UP(blocks, blocks_per_table_cluster);
3211 n = clusters + blocks + table;
3212
3213 if (n == last && generous_increase) {
3214 clusters += DIV_ROUND_UP(table, 2);
3215 n = 0; /* force another loop */
3216 generous_increase = false;
3217 }
3218 } while (n != last);
3219
3220 if (refblock_count) {
3221 *refblock_count = blocks;
3222 }
3223
3224 return (blocks + table) * cluster_size;
3225 }
3226
3227 /**
3228 * qcow2_calc_prealloc_size:
3229 * @total_size: virtual disk size in bytes
3230 * @cluster_size: cluster size in bytes
3231 * @refcount_order: refcount bits power-of-2 exponent
3232 *
3233 * Returns: Total number of bytes required for the fully allocated image
3234 * (including metadata).
3235 */
3236 static int64_t qcow2_calc_prealloc_size(int64_t total_size,
3237 size_t cluster_size,
3238 int refcount_order)
3239 {
3240 int64_t meta_size = 0;
3241 uint64_t nl1e, nl2e;
3242 int64_t aligned_total_size = ROUND_UP(total_size, cluster_size);
3243
3244 /* header: 1 cluster */
3245 meta_size += cluster_size;
3246
3247 /* total size of L2 tables */
3248 nl2e = aligned_total_size / cluster_size;
3249 nl2e = ROUND_UP(nl2e, cluster_size / sizeof(uint64_t));
3250 meta_size += nl2e * sizeof(uint64_t);
3251
3252 /* total size of L1 tables */
3253 nl1e = nl2e * sizeof(uint64_t) / cluster_size;
3254 nl1e = ROUND_UP(nl1e, cluster_size / sizeof(uint64_t));
3255 meta_size += nl1e * sizeof(uint64_t);
3256
3257 /* total size of refcount table and blocks */
3258 meta_size += qcow2_refcount_metadata_size(
3259 (meta_size + aligned_total_size) / cluster_size,
3260 cluster_size, refcount_order, false, NULL);
3261
3262 return meta_size + aligned_total_size;
3263 }
3264
3265 static bool validate_cluster_size(size_t cluster_size, Error **errp)
3266 {
3267 int cluster_bits = ctz32(cluster_size);
3268 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
3269 (1 << cluster_bits) != cluster_size)
3270 {
3271 error_setg(errp, "Cluster size must be a power of two between %d and "
3272 "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
3273 return false;
3274 }
3275 return true;
3276 }
3277
3278 static size_t qcow2_opt_get_cluster_size_del(QemuOpts *opts, Error **errp)
3279 {
3280 size_t cluster_size;
3281
3282 cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
3283 DEFAULT_CLUSTER_SIZE);
3284 if (!validate_cluster_size(cluster_size, errp)) {
3285 return 0;
3286 }
3287 return cluster_size;
3288 }
3289
3290 static int qcow2_opt_get_version_del(QemuOpts *opts, Error **errp)
3291 {
3292 char *buf;
3293 int ret;
3294
3295 buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
3296 if (!buf) {
3297 ret = 3; /* default */
3298 } else if (!strcmp(buf, "0.10")) {
3299 ret = 2;
3300 } else if (!strcmp(buf, "1.1")) {
3301 ret = 3;
3302 } else {
3303 error_setg(errp, "Invalid compatibility level: '%s'", buf);
3304 ret = -EINVAL;
3305 }
3306 g_free(buf);
3307 return ret;
3308 }
3309
3310 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts *opts, int version,
3311 Error **errp)
3312 {
3313 uint64_t refcount_bits;
3314
3315 refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 16);
3316 if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
3317 error_setg(errp, "Refcount width must be a power of two and may not "
3318 "exceed 64 bits");
3319 return 0;
3320 }
3321
3322 if (version < 3 && refcount_bits != 16) {
3323 error_setg(errp, "Different refcount widths than 16 bits require "
3324 "compatibility level 1.1 or above (use compat=1.1 or "
3325 "greater)");
3326 return 0;
3327 }
3328
3329 return refcount_bits;
3330 }
3331
3332 static int coroutine_fn
3333 qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp)
3334 {
3335 BlockdevCreateOptionsQcow2 *qcow2_opts;
3336 QDict *options;
3337
3338 /*
3339 * Open the image file and write a minimal qcow2 header.
3340 *
3341 * We keep things simple and start with a zero-sized image. We also
3342 * do without refcount blocks or a L1 table for now. We'll fix the
3343 * inconsistency later.
3344 *
3345 * We do need a refcount table because growing the refcount table means
3346 * allocating two new refcount blocks - the second of which would be at
3347 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
3348 * size for any qcow2 image.
3349 */
3350 BlockBackend *blk = NULL;
3351 BlockDriverState *bs = NULL;
3352 BlockDriverState *data_bs = NULL;
3353 QCowHeader *header;
3354 size_t cluster_size;
3355 int version;
3356 int refcount_order;
3357 uint64_t* refcount_table;
3358 int ret;
3359 uint8_t compression_type = QCOW2_COMPRESSION_TYPE_ZLIB;
3360
3361 assert(create_options->driver == BLOCKDEV_DRIVER_QCOW2);
3362 qcow2_opts = &create_options->u.qcow2;
3363
3364 bs = bdrv_open_blockdev_ref(qcow2_opts->file, errp);
3365 if (bs == NULL) {
3366 return -EIO;
3367 }
3368
3369 /* Validate options and set default values */
3370 if (!QEMU_IS_ALIGNED(qcow2_opts->size, BDRV_SECTOR_SIZE)) {
3371 error_setg(errp, "Image size must be a multiple of %u bytes",
3372 (unsigned) BDRV_SECTOR_SIZE);
3373 ret = -EINVAL;
3374 goto out;
3375 }
3376
3377 if (qcow2_opts->has_version) {
3378 switch (qcow2_opts->version) {
3379 case BLOCKDEV_QCOW2_VERSION_V2:
3380 version = 2;
3381 break;
3382 case BLOCKDEV_QCOW2_VERSION_V3:
3383 version = 3;
3384 break;
3385 default:
3386 g_assert_not_reached();
3387 }
3388 } else {
3389 version = 3;
3390 }
3391
3392 if (qcow2_opts->has_cluster_size) {
3393 cluster_size = qcow2_opts->cluster_size;
3394 } else {
3395 cluster_size = DEFAULT_CLUSTER_SIZE;
3396 }
3397
3398 if (!validate_cluster_size(cluster_size, errp)) {
3399 ret = -EINVAL;
3400 goto out;
3401 }
3402
3403 if (!qcow2_opts->has_preallocation) {
3404 qcow2_opts->preallocation = PREALLOC_MODE_OFF;
3405 }
3406 if (qcow2_opts->has_backing_file &&
3407 qcow2_opts->preallocation != PREALLOC_MODE_OFF)
3408 {
3409 error_setg(errp, "Backing file and preallocation cannot be used at "
3410 "the same time");
3411 ret = -EINVAL;
3412 goto out;
3413 }
3414 if (qcow2_opts->has_backing_fmt && !qcow2_opts->has_backing_file) {
3415 error_setg(