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 "system/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 "qobject/qdict.h"
36 #include "qobject/qstring.h"
37 #include "trace.h"
38 #include "qemu/option_int.h"
39 #include "qemu/cutils.h"
40 #include "qemu/bswap.h"
41 #include "qemu/memalign.h"
42 #include "qapi/qobject-input-visitor.h"
43 #include "qapi/qapi-visit-block-core.h"
44 #include "crypto.h"
45 #include "block/aio_task.h"
46 #include "block/dirty-bitmap.h"
47
48 /*
49 Differences with QCOW:
50
51 - Support for multiple incremental snapshots.
52 - Memory management by reference counts.
53 - Clusters which have a reference count of one have the bit
54 QCOW_OFLAG_COPIED to optimize write performance.
55 - Size of compressed clusters is stored in sectors to reduce bit usage
56 in the cluster offsets.
57 - Support for storing additional data (such as the VM state) in the
58 snapshots.
59 - If a backing store is used, the cluster size is not constrained
60 (could be backported to QCOW).
61 - L2 tables have always a size of one cluster.
62 */
63
64
65 typedef struct {
66 uint32_t magic;
67 uint32_t len;
68 } QEMU_PACKED QCowExtension;
69
70 #define QCOW2_EXT_MAGIC_END 0
71 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xe2792aca
72 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
73 #define QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77
74 #define QCOW2_EXT_MAGIC_BITMAPS 0x23852875
75 #define QCOW2_EXT_MAGIC_DATA_FILE 0x44415441
76
77 static int coroutine_fn
78 qcow2_co_preadv_compressed(BlockDriverState *bs,
79 uint64_t l2_entry,
80 uint64_t offset,
81 uint64_t bytes,
82 QEMUIOVector *qiov,
83 size_t qiov_offset);
84
qcow2_probe(const uint8_t * buf,int buf_size,const char * filename)85 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
86 {
87 const QCowHeader *cow_header = (const void *)buf;
88
89 if (buf_size >= sizeof(QCowHeader) &&
90 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
91 be32_to_cpu(cow_header->version) >= 2)
92 return 100;
93 else
94 return 0;
95 }
96
97
98 static int GRAPH_RDLOCK
qcow2_crypto_hdr_read_func(QCryptoBlock * block,size_t offset,uint8_t * buf,size_t buflen,void * opaque,Error ** errp)99 qcow2_crypto_hdr_read_func(QCryptoBlock *block, size_t offset,
100 uint8_t *buf, size_t buflen,
101 void *opaque, Error **errp)
102 {
103 BlockDriverState *bs = opaque;
104 BDRVQcow2State *s = bs->opaque;
105 ssize_t ret;
106
107 if ((offset + buflen) > s->crypto_header.length) {
108 error_setg(errp, "Request for data outside of extension header");
109 return -1;
110 }
111
112 ret = bdrv_pread(bs->file, s->crypto_header.offset + offset, buflen, buf,
113 0);
114 if (ret < 0) {
115 error_setg_errno(errp, -ret, "Could not read encryption header");
116 return -1;
117 }
118 return 0;
119 }
120
121
122 static int coroutine_fn GRAPH_RDLOCK
qcow2_crypto_hdr_init_func(QCryptoBlock * block,size_t headerlen,void * opaque,Error ** errp)123 qcow2_crypto_hdr_init_func(QCryptoBlock *block, size_t headerlen, void *opaque,
124 Error **errp)
125 {
126 BlockDriverState *bs = opaque;
127 BDRVQcow2State *s = bs->opaque;
128 int64_t ret;
129 int64_t clusterlen;
130
131 ret = qcow2_alloc_clusters(bs, headerlen);
132 if (ret < 0) {
133 error_setg_errno(errp, -ret,
134 "Cannot allocate cluster for LUKS header size %zu",
135 headerlen);
136 return -1;
137 }
138
139 s->crypto_header.length = headerlen;
140 s->crypto_header.offset = ret;
141
142 /*
143 * Zero fill all space in cluster so it has predictable
144 * content, as we may not initialize some regions of the
145 * header (eg only 1 out of 8 key slots will be initialized)
146 */
147 clusterlen = size_to_clusters(s, headerlen) * s->cluster_size;
148 assert(qcow2_pre_write_overlap_check(bs, 0, ret, clusterlen, false) == 0);
149 ret = bdrv_co_pwrite_zeroes(bs->file, ret, clusterlen, 0);
150 if (ret < 0) {
151 error_setg_errno(errp, -ret, "Could not zero fill encryption header");
152 return -1;
153 }
154
155 return 0;
156 }
157
158
159 /* The graph lock must be held when called in coroutine context */
160 static int coroutine_mixed_fn GRAPH_RDLOCK
qcow2_crypto_hdr_write_func(QCryptoBlock * block,size_t offset,const uint8_t * buf,size_t buflen,void * opaque,Error ** errp)161 qcow2_crypto_hdr_write_func(QCryptoBlock *block, size_t offset,
162 const uint8_t *buf, size_t buflen,
163 void *opaque, Error **errp)
164 {
165 BlockDriverState *bs = opaque;
166 BDRVQcow2State *s = bs->opaque;
167 ssize_t ret;
168
169 if ((offset + buflen) > s->crypto_header.length) {
170 error_setg(errp, "Request for data outside of extension header");
171 return -1;
172 }
173
174 ret = bdrv_pwrite(bs->file, s->crypto_header.offset + offset, buflen, buf,
175 0);
176 if (ret < 0) {
177 error_setg_errno(errp, -ret, "Could not read encryption header");
178 return -1;
179 }
180 return 0;
181 }
182
183 static QDict*
qcow2_extract_crypto_opts(QemuOpts * opts,const char * fmt,Error ** errp)184 qcow2_extract_crypto_opts(QemuOpts *opts, const char *fmt, Error **errp)
185 {
186 QDict *cryptoopts_qdict;
187 QDict *opts_qdict;
188
189 /* Extract "encrypt." options into a qdict */
190 opts_qdict = qemu_opts_to_qdict(opts, NULL);
191 qdict_extract_subqdict(opts_qdict, &cryptoopts_qdict, "encrypt.");
192 qobject_unref(opts_qdict);
193 qdict_put_str(cryptoopts_qdict, "format", fmt);
194 return cryptoopts_qdict;
195 }
196
197 /*
198 * read qcow2 extension and fill bs
199 * start reading from start_offset
200 * finish reading upon magic of value 0 or when end_offset reached
201 * unknown magic is skipped (future extension this version knows nothing about)
202 * return 0 upon success, non-0 otherwise
203 */
204 static int coroutine_fn GRAPH_RDLOCK
qcow2_read_extensions(BlockDriverState * bs,uint64_t start_offset,uint64_t end_offset,void ** p_feature_table,int flags,bool * need_update_header,Error ** errp)205 qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
206 uint64_t end_offset, void **p_feature_table,
207 int flags, bool *need_update_header, Error **errp)
208 {
209 BDRVQcow2State *s = bs->opaque;
210 QCowExtension ext;
211 uint64_t offset;
212 int ret;
213 Qcow2BitmapHeaderExt bitmaps_ext;
214
215 if (need_update_header != NULL) {
216 *need_update_header = false;
217 }
218
219 #ifdef DEBUG_EXT
220 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
221 #endif
222 offset = start_offset;
223 while (offset < end_offset) {
224
225 #ifdef DEBUG_EXT
226 /* Sanity check */
227 if (offset > s->cluster_size)
228 printf("qcow2_read_extension: suspicious offset %lu\n", offset);
229
230 printf("attempting to read extended header in offset %lu\n", offset);
231 #endif
232
233 ret = bdrv_co_pread(bs->file, offset, sizeof(ext), &ext, 0);
234 if (ret < 0) {
235 error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: "
236 "pread fail from offset %" PRIu64, offset);
237 return 1;
238 }
239 ext.magic = be32_to_cpu(ext.magic);
240 ext.len = be32_to_cpu(ext.len);
241 offset += sizeof(ext);
242 #ifdef DEBUG_EXT
243 printf("ext.magic = 0x%x\n", ext.magic);
244 #endif
245 if (offset > end_offset || ext.len > end_offset - offset) {
246 error_setg(errp, "Header extension too large");
247 return -EINVAL;
248 }
249
250 switch (ext.magic) {
251 case QCOW2_EXT_MAGIC_END:
252 return 0;
253
254 case QCOW2_EXT_MAGIC_BACKING_FORMAT:
255 if (ext.len >= sizeof(bs->backing_format)) {
256 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32
257 " too large (>=%zu)", ext.len,
258 sizeof(bs->backing_format));
259 return 2;
260 }
261 ret = bdrv_co_pread(bs->file, offset, ext.len, bs->backing_format, 0);
262 if (ret < 0) {
263 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: "
264 "Could not read format name");
265 return 3;
266 }
267 bs->backing_format[ext.len] = '\0';
268 s->image_backing_format = g_strdup(bs->backing_format);
269 #ifdef DEBUG_EXT
270 printf("Qcow2: Got format extension %s\n", bs->backing_format);
271 #endif
272 break;
273
274 case QCOW2_EXT_MAGIC_FEATURE_TABLE:
275 if (p_feature_table != NULL) {
276 void *feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
277 ret = bdrv_co_pread(bs->file, offset, ext.len, feature_table, 0);
278 if (ret < 0) {
279 error_setg_errno(errp, -ret, "ERROR: ext_feature_table: "
280 "Could not read table");
281 g_free(feature_table);
282 return ret;
283 }
284
285 *p_feature_table = feature_table;
286 }
287 break;
288
289 case QCOW2_EXT_MAGIC_CRYPTO_HEADER: {
290 unsigned int cflags = 0;
291 if (s->crypt_method_header != QCOW_CRYPT_LUKS) {
292 error_setg(errp, "CRYPTO header extension only "
293 "expected with LUKS encryption method");
294 return -EINVAL;
295 }
296 if (ext.len != sizeof(Qcow2CryptoHeaderExtension)) {
297 error_setg(errp, "CRYPTO header extension size %u, "
298 "but expected size %zu", ext.len,
299 sizeof(Qcow2CryptoHeaderExtension));
300 return -EINVAL;
301 }
302
303 ret = bdrv_co_pread(bs->file, offset, ext.len, &s->crypto_header, 0);
304 if (ret < 0) {
305 error_setg_errno(errp, -ret,
306 "Unable to read CRYPTO header extension");
307 return ret;
308 }
309 s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset);
310 s->crypto_header.length = be64_to_cpu(s->crypto_header.length);
311
312 if ((s->crypto_header.offset % s->cluster_size) != 0) {
313 error_setg(errp, "Encryption header offset '%" PRIu64 "' is "
314 "not a multiple of cluster size '%u'",
315 s->crypto_header.offset, s->cluster_size);
316 return -EINVAL;
317 }
318
319 if (flags & BDRV_O_NO_IO) {
320 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
321 }
322 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
323 qcow2_crypto_hdr_read_func,
324 bs, cflags, errp);
325 if (!s->crypto) {
326 return -EINVAL;
327 }
328 } break;
329
330 case QCOW2_EXT_MAGIC_BITMAPS:
331 if (ext.len != sizeof(bitmaps_ext)) {
332 error_setg_errno(errp, -ret, "bitmaps_ext: "
333 "Invalid extension length");
334 return -EINVAL;
335 }
336
337 if (!(s->autoclear_features & QCOW2_AUTOCLEAR_BITMAPS)) {
338 if (s->qcow_version < 3) {
339 /* Let's be a bit more specific */
340 warn_report("This qcow2 v2 image contains bitmaps, but "
341 "they may have been modified by a program "
342 "without persistent bitmap support; so now "
343 "they must all be considered inconsistent");
344 } else {
345 warn_report("a program lacking bitmap support "
346 "modified this file, so all bitmaps are now "
347 "considered inconsistent");
348 }
349 error_printf("Some clusters may be leaked, "
350 "run 'qemu-img check -r' on the image "
351 "file to fix.");
352 if (need_update_header != NULL) {
353 /* Updating is needed to drop invalid bitmap extension. */
354 *need_update_header = true;
355 }
356 break;
357 }
358
359 ret = bdrv_co_pread(bs->file, offset, ext.len, &bitmaps_ext, 0);
360 if (ret < 0) {
361 error_setg_errno(errp, -ret, "bitmaps_ext: "
362 "Could not read ext header");
363 return ret;
364 }
365
366 if (bitmaps_ext.reserved32 != 0) {
367 error_setg_errno(errp, -ret, "bitmaps_ext: "
368 "Reserved field is not zero");
369 return -EINVAL;
370 }
371
372 bitmaps_ext.nb_bitmaps = be32_to_cpu(bitmaps_ext.nb_bitmaps);
373 bitmaps_ext.bitmap_directory_size =
374 be64_to_cpu(bitmaps_ext.bitmap_directory_size);
375 bitmaps_ext.bitmap_directory_offset =
376 be64_to_cpu(bitmaps_ext.bitmap_directory_offset);
377
378 if (bitmaps_ext.nb_bitmaps > QCOW2_MAX_BITMAPS) {
379 error_setg(errp,
380 "bitmaps_ext: Image has %" PRIu32 " bitmaps, "
381 "exceeding the QEMU supported maximum of %d",
382 bitmaps_ext.nb_bitmaps, QCOW2_MAX_BITMAPS);
383 return -EINVAL;
384 }
385
386 if (bitmaps_ext.nb_bitmaps == 0) {
387 error_setg(errp, "found bitmaps extension with zero bitmaps");
388 return -EINVAL;
389 }
390
391 if (offset_into_cluster(s, bitmaps_ext.bitmap_directory_offset)) {
392 error_setg(errp, "bitmaps_ext: "
393 "invalid bitmap directory offset");
394 return -EINVAL;
395 }
396
397 if (bitmaps_ext.bitmap_directory_size >
398 QCOW2_MAX_BITMAP_DIRECTORY_SIZE) {
399 error_setg(errp, "bitmaps_ext: "
400 "bitmap directory size (%" PRIu64 ") exceeds "
401 "the maximum supported size (%d)",
402 bitmaps_ext.bitmap_directory_size,
403 QCOW2_MAX_BITMAP_DIRECTORY_SIZE);
404 return -EINVAL;
405 }
406
407 s->nb_bitmaps = bitmaps_ext.nb_bitmaps;
408 s->bitmap_directory_offset =
409 bitmaps_ext.bitmap_directory_offset;
410 s->bitmap_directory_size =
411 bitmaps_ext.bitmap_directory_size;
412
413 #ifdef DEBUG_EXT
414 printf("Qcow2: Got bitmaps extension: "
415 "offset=%" PRIu64 " nb_bitmaps=%" PRIu32 "\n",
416 s->bitmap_directory_offset, s->nb_bitmaps);
417 #endif
418 break;
419
420 case QCOW2_EXT_MAGIC_DATA_FILE:
421 {
422 s->image_data_file = g_malloc0(ext.len + 1);
423 ret = bdrv_co_pread(bs->file, offset, ext.len, s->image_data_file, 0);
424 if (ret < 0) {
425 error_setg_errno(errp, -ret,
426 "ERROR: Could not read data file name");
427 return ret;
428 }
429 #ifdef DEBUG_EXT
430 printf("Qcow2: Got external data file %s\n", s->image_data_file);
431 #endif
432 break;
433 }
434
435 default:
436 /* unknown magic - save it in case we need to rewrite the header */
437 /* If you add a new feature, make sure to also update the fast
438 * path of qcow2_make_empty() to deal with it. */
439 {
440 Qcow2UnknownHeaderExtension *uext;
441
442 uext = g_malloc0(sizeof(*uext) + ext.len);
443 uext->magic = ext.magic;
444 uext->len = ext.len;
445 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
446
447 ret = bdrv_co_pread(bs->file, offset, uext->len, uext->data, 0);
448 if (ret < 0) {
449 error_setg_errno(errp, -ret, "ERROR: unknown extension: "
450 "Could not read data");
451 return ret;
452 }
453 }
454 break;
455 }
456
457 offset += ((ext.len + 7) & ~7);
458 }
459
460 return 0;
461 }
462
cleanup_unknown_header_ext(BlockDriverState * bs)463 static void cleanup_unknown_header_ext(BlockDriverState *bs)
464 {
465 BDRVQcow2State *s = bs->opaque;
466 Qcow2UnknownHeaderExtension *uext, *next;
467
468 QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
469 QLIST_REMOVE(uext, next);
470 g_free(uext);
471 }
472 }
473
report_unsupported_feature(Error ** errp,Qcow2Feature * table,uint64_t mask)474 static void report_unsupported_feature(Error **errp, Qcow2Feature *table,
475 uint64_t mask)
476 {
477 g_autoptr(GString) features = g_string_sized_new(60);
478
479 while (table && table->name[0] != '\0') {
480 if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
481 if (mask & (1ULL << table->bit)) {
482 if (features->len > 0) {
483 g_string_append(features, ", ");
484 }
485 g_string_append_printf(features, "%.46s", table->name);
486 mask &= ~(1ULL << table->bit);
487 }
488 }
489 table++;
490 }
491
492 if (mask) {
493 if (features->len > 0) {
494 g_string_append(features, ", ");
495 }
496 g_string_append_printf(features,
497 "Unknown incompatible feature: %" PRIx64, mask);
498 }
499
500 error_setg(errp, "Unsupported qcow2 feature(s): %s", features->str);
501 }
502
503 /*
504 * Sets the dirty bit and flushes afterwards if necessary.
505 *
506 * The incompatible_features bit is only set if the image file header was
507 * updated successfully. Therefore it is not required to check the return
508 * value of this function.
509 */
qcow2_mark_dirty(BlockDriverState * bs)510 int qcow2_mark_dirty(BlockDriverState *bs)
511 {
512 BDRVQcow2State *s = bs->opaque;
513 uint64_t val;
514 int ret;
515
516 assert(s->qcow_version >= 3);
517
518 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
519 return 0; /* already dirty */
520 }
521
522 val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY);
523 ret = bdrv_pwrite_sync(bs->file,
524 offsetof(QCowHeader, incompatible_features),
525 sizeof(val), &val, 0);
526 if (ret < 0) {
527 return ret;
528 }
529
530 /* Only treat image as dirty if the header was updated successfully */
531 s->incompatible_features |= QCOW2_INCOMPAT_DIRTY;
532 return 0;
533 }
534
535 /*
536 * Clears the dirty bit and flushes before if necessary. Only call this
537 * function when there are no pending requests, it does not guard against
538 * concurrent requests dirtying the image.
539 */
qcow2_mark_clean(BlockDriverState * bs)540 static int GRAPH_RDLOCK qcow2_mark_clean(BlockDriverState *bs)
541 {
542 BDRVQcow2State *s = bs->opaque;
543
544 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
545 int ret;
546
547 s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY;
548
549 ret = qcow2_flush_caches(bs);
550 if (ret < 0) {
551 return ret;
552 }
553
554 return qcow2_update_header(bs);
555 }
556 return 0;
557 }
558
559 /*
560 * Marks the image as corrupt.
561 */
qcow2_mark_corrupt(BlockDriverState * bs)562 int qcow2_mark_corrupt(BlockDriverState *bs)
563 {
564 BDRVQcow2State *s = bs->opaque;
565
566 s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT;
567 return qcow2_update_header(bs);
568 }
569
570 /*
571 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
572 * before if necessary.
573 */
574 static int coroutine_fn GRAPH_RDLOCK
qcow2_mark_consistent(BlockDriverState * bs)575 qcow2_mark_consistent(BlockDriverState *bs)
576 {
577 BDRVQcow2State *s = bs->opaque;
578
579 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
580 int ret = qcow2_flush_caches(bs);
581 if (ret < 0) {
582 return ret;
583 }
584
585 s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT;
586 return qcow2_update_header(bs);
587 }
588 return 0;
589 }
590
qcow2_add_check_result(BdrvCheckResult * out,const BdrvCheckResult * src,bool set_allocation_info)591 static void qcow2_add_check_result(BdrvCheckResult *out,
592 const BdrvCheckResult *src,
593 bool set_allocation_info)
594 {
595 out->corruptions += src->corruptions;
596 out->leaks += src->leaks;
597 out->check_errors += src->check_errors;
598 out->corruptions_fixed += src->corruptions_fixed;
599 out->leaks_fixed += src->leaks_fixed;
600
601 if (set_allocation_info) {
602 out->image_end_offset = src->image_end_offset;
603 out->bfi = src->bfi;
604 }
605 }
606
607 static int coroutine_fn GRAPH_RDLOCK
qcow2_co_check_locked(BlockDriverState * bs,BdrvCheckResult * result,BdrvCheckMode fix)608 qcow2_co_check_locked(BlockDriverState *bs, BdrvCheckResult *result,
609 BdrvCheckMode fix)
610 {
611 BdrvCheckResult snapshot_res = {};
612 BdrvCheckResult refcount_res = {};
613 int ret;
614
615 memset(result, 0, sizeof(*result));
616
617 ret = qcow2_check_read_snapshot_table(bs, &snapshot_res, fix);
618 if (ret < 0) {
619 qcow2_add_check_result(result, &snapshot_res, false);
620 return ret;
621 }
622
623 ret = qcow2_check_refcounts(bs, &refcount_res, fix);
624 qcow2_add_check_result(result, &refcount_res, true);
625 if (ret < 0) {
626 qcow2_add_check_result(result, &snapshot_res, false);
627 return ret;
628 }
629
630 ret = qcow2_check_fix_snapshot_table(bs, &snapshot_res, fix);
631 qcow2_add_check_result(result, &snapshot_res, false);
632 if (ret < 0) {
633 return ret;
634 }
635
636 if (fix && result->check_errors == 0 && result->corruptions == 0) {
637 ret = qcow2_mark_clean(bs);
638 if (ret < 0) {
639 return ret;
640 }
641 return qcow2_mark_consistent(bs);
642 }
643 return ret;
644 }
645
646 static int coroutine_fn GRAPH_RDLOCK
qcow2_co_check(BlockDriverState * bs,BdrvCheckResult * result,BdrvCheckMode fix)647 qcow2_co_check(BlockDriverState *bs, BdrvCheckResult *result,
648 BdrvCheckMode fix)
649 {
650 BDRVQcow2State *s = bs->opaque;
651 int ret;
652
653 qemu_co_mutex_lock(&s->lock);
654 ret = qcow2_co_check_locked(bs, result, fix);
655 qemu_co_mutex_unlock(&s->lock);
656 return ret;
657 }
658
qcow2_validate_table(BlockDriverState * bs,uint64_t offset,uint64_t entries,size_t entry_len,int64_t max_size_bytes,const char * table_name,Error ** errp)659 int qcow2_validate_table(BlockDriverState *bs, uint64_t offset,
660 uint64_t entries, size_t entry_len,
661 int64_t max_size_bytes, const char *table_name,
662 Error **errp)
663 {
664 BDRVQcow2State *s = bs->opaque;
665
666 if (entries > max_size_bytes / entry_len) {
667 error_setg(errp, "%s too large", table_name);
668 return -EFBIG;
669 }
670
671 /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
672 * because values will be passed to qemu functions taking int64_t. */
673 if ((INT64_MAX - entries * entry_len < offset) ||
674 (offset_into_cluster(s, offset) != 0)) {
675 error_setg(errp, "%s offset invalid", table_name);
676 return -EINVAL;
677 }
678
679 return 0;
680 }
681
682 static const char *const mutable_opts[] = {
683 QCOW2_OPT_LAZY_REFCOUNTS,
684 QCOW2_OPT_DISCARD_REQUEST,
685 QCOW2_OPT_DISCARD_SNAPSHOT,
686 QCOW2_OPT_DISCARD_OTHER,
687 QCOW2_OPT_DISCARD_NO_UNREF,
688 QCOW2_OPT_OVERLAP,
689 QCOW2_OPT_OVERLAP_TEMPLATE,
690 QCOW2_OPT_OVERLAP_MAIN_HEADER,
691 QCOW2_OPT_OVERLAP_ACTIVE_L1,
692 QCOW2_OPT_OVERLAP_ACTIVE_L2,
693 QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
694 QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
695 QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
696 QCOW2_OPT_OVERLAP_INACTIVE_L1,
697 QCOW2_OPT_OVERLAP_INACTIVE_L2,
698 QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY,
699 QCOW2_OPT_CACHE_SIZE,
700 QCOW2_OPT_L2_CACHE_SIZE,
701 QCOW2_OPT_L2_CACHE_ENTRY_SIZE,
702 QCOW2_OPT_REFCOUNT_CACHE_SIZE,
703 QCOW2_OPT_CACHE_CLEAN_INTERVAL,
704 NULL
705 };
706
707 static QemuOptsList qcow2_runtime_opts = {
708 .name = "qcow2",
709 .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head),
710 .desc = {
711 {
712 .name = QCOW2_OPT_LAZY_REFCOUNTS,
713 .type = QEMU_OPT_BOOL,
714 .help = "Postpone refcount updates",
715 },
716 {
717 .name = QCOW2_OPT_DISCARD_REQUEST,
718 .type = QEMU_OPT_BOOL,
719 .help = "Pass guest discard requests to the layer below",
720 },
721 {
722 .name = QCOW2_OPT_DISCARD_SNAPSHOT,
723 .type = QEMU_OPT_BOOL,
724 .help = "Generate discard requests when snapshot related space "
725 "is freed",
726 },
727 {
728 .name = QCOW2_OPT_DISCARD_OTHER,
729 .type = QEMU_OPT_BOOL,
730 .help = "Generate discard requests when other clusters are freed",
731 },
732 {
733 .name = QCOW2_OPT_DISCARD_NO_UNREF,
734 .type = QEMU_OPT_BOOL,
735 .help = "Do not unreference discarded clusters",
736 },
737 {
738 .name = QCOW2_OPT_OVERLAP,
739 .type = QEMU_OPT_STRING,
740 .help = "Selects which overlap checks to perform from a range of "
741 "templates (none, constant, cached, all)",
742 },
743 {
744 .name = QCOW2_OPT_OVERLAP_TEMPLATE,
745 .type = QEMU_OPT_STRING,
746 .help = "Selects which overlap checks to perform from a range of "
747 "templates (none, constant, cached, all)",
748 },
749 {
750 .name = QCOW2_OPT_OVERLAP_MAIN_HEADER,
751 .type = QEMU_OPT_BOOL,
752 .help = "Check for unintended writes into the main qcow2 header",
753 },
754 {
755 .name = QCOW2_OPT_OVERLAP_ACTIVE_L1,
756 .type = QEMU_OPT_BOOL,
757 .help = "Check for unintended writes into the active L1 table",
758 },
759 {
760 .name = QCOW2_OPT_OVERLAP_ACTIVE_L2,
761 .type = QEMU_OPT_BOOL,
762 .help = "Check for unintended writes into an active L2 table",
763 },
764 {
765 .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
766 .type = QEMU_OPT_BOOL,
767 .help = "Check for unintended writes into the refcount table",
768 },
769 {
770 .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
771 .type = QEMU_OPT_BOOL,
772 .help = "Check for unintended writes into a refcount block",
773 },
774 {
775 .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
776 .type = QEMU_OPT_BOOL,
777 .help = "Check for unintended writes into the snapshot table",
778 },
779 {
780 .name = QCOW2_OPT_OVERLAP_INACTIVE_L1,
781 .type = QEMU_OPT_BOOL,
782 .help = "Check for unintended writes into an inactive L1 table",
783 },
784 {
785 .name = QCOW2_OPT_OVERLAP_INACTIVE_L2,
786 .type = QEMU_OPT_BOOL,
787 .help = "Check for unintended writes into an inactive L2 table",
788 },
789 {
790 .name = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY,
791 .type = QEMU_OPT_BOOL,
792 .help = "Check for unintended writes into the bitmap directory",
793 },
794 {
795 .name = QCOW2_OPT_CACHE_SIZE,
796 .type = QEMU_OPT_SIZE,
797 .help = "Maximum combined metadata (L2 tables and refcount blocks) "
798 "cache size",
799 },
800 {
801 .name = QCOW2_OPT_L2_CACHE_SIZE,
802 .type = QEMU_OPT_SIZE,
803 .help = "Maximum L2 table cache size",
804 },
805 {
806 .name = QCOW2_OPT_L2_CACHE_ENTRY_SIZE,
807 .type = QEMU_OPT_SIZE,
808 .help = "Size of each entry in the L2 cache",
809 },
810 {
811 .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE,
812 .type = QEMU_OPT_SIZE,
813 .help = "Maximum refcount block cache size",
814 },
815 {
816 .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL,
817 .type = QEMU_OPT_NUMBER,
818 .help = "Clean unused cache entries after this time (in seconds)",
819 },
820 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
821 "ID of secret providing qcow2 AES key or LUKS passphrase"),
822 { /* end of list */ }
823 },
824 };
825
826 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = {
827 [QCOW2_OL_MAIN_HEADER_BITNR] = QCOW2_OPT_OVERLAP_MAIN_HEADER,
828 [QCOW2_OL_ACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L1,
829 [QCOW2_OL_ACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L2,
830 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
831 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
832 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
833 [QCOW2_OL_INACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L1,
834 [QCOW2_OL_INACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L2,
835 [QCOW2_OL_BITMAP_DIRECTORY_BITNR] = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY,
836 };
837
cache_clean_timer_cb(void * opaque)838 static void cache_clean_timer_cb(void *opaque)
839 {
840 BlockDriverState *bs = opaque;
841 BDRVQcow2State *s = bs->opaque;
842 qcow2_cache_clean_unused(s->l2_table_cache);
843 qcow2_cache_clean_unused(s->refcount_block_cache);
844 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
845 (int64_t) s->cache_clean_interval * 1000);
846 }
847
cache_clean_timer_init(BlockDriverState * bs,AioContext * context)848 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context)
849 {
850 BDRVQcow2State *s = bs->opaque;
851 if (s->cache_clean_interval > 0) {
852 s->cache_clean_timer =
853 aio_timer_new_with_attrs(context, QEMU_CLOCK_VIRTUAL,
854 SCALE_MS, QEMU_TIMER_ATTR_EXTERNAL,
855 cache_clean_timer_cb, bs);
856 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
857 (int64_t) s->cache_clean_interval * 1000);
858 }
859 }
860
cache_clean_timer_del(BlockDriverState * bs)861 static void cache_clean_timer_del(BlockDriverState *bs)
862 {
863 BDRVQcow2State *s = bs->opaque;
864 if (s->cache_clean_timer) {
865 timer_free(s->cache_clean_timer);
866 s->cache_clean_timer = NULL;
867 }
868 }
869
qcow2_detach_aio_context(BlockDriverState * bs)870 static void qcow2_detach_aio_context(BlockDriverState *bs)
871 {
872 cache_clean_timer_del(bs);
873 }
874
qcow2_attach_aio_context(BlockDriverState * bs,AioContext * new_context)875 static void qcow2_attach_aio_context(BlockDriverState *bs,
876 AioContext *new_context)
877 {
878 cache_clean_timer_init(bs, new_context);
879 }
880
read_cache_sizes(BlockDriverState * bs,QemuOpts * opts,uint64_t * l2_cache_size,uint64_t * l2_cache_entry_size,uint64_t * refcount_cache_size,Error ** errp)881 static bool read_cache_sizes(BlockDriverState *bs, QemuOpts *opts,
882 uint64_t *l2_cache_size,
883 uint64_t *l2_cache_entry_size,
884 uint64_t *refcount_cache_size, Error **errp)
885 {
886 BDRVQcow2State *s = bs->opaque;
887 uint64_t combined_cache_size, l2_cache_max_setting;
888 bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set;
889 bool l2_cache_entry_size_set;
890 int min_refcount_cache = MIN_REFCOUNT_CACHE_SIZE * s->cluster_size;
891 uint64_t virtual_disk_size = bs->total_sectors * BDRV_SECTOR_SIZE;
892 uint64_t max_l2_entries = DIV_ROUND_UP(virtual_disk_size, s->cluster_size);
893 /* An L2 table is always one cluster in size so the max cache size
894 * should be a multiple of the cluster size. */
895 uint64_t max_l2_cache = ROUND_UP(max_l2_entries * l2_entry_size(s),
896 s->cluster_size);
897
898 combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE);
899 l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE);
900 refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
901 l2_cache_entry_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE);
902
903 combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0);
904 l2_cache_max_setting = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE,
905 DEFAULT_L2_CACHE_MAX_SIZE);
906 *refcount_cache_size = qemu_opt_get_size(opts,
907 QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0);
908
909 *l2_cache_entry_size = qemu_opt_get_size(
910 opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE, s->cluster_size);
911
912 *l2_cache_size = MIN(max_l2_cache, l2_cache_max_setting);
913
914 if (combined_cache_size_set) {
915 if (l2_cache_size_set && refcount_cache_size_set) {
916 error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE
917 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set "
918 "at the same time");
919 return false;
920 } else if (l2_cache_size_set &&
921 (l2_cache_max_setting > combined_cache_size)) {
922 error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed "
923 QCOW2_OPT_CACHE_SIZE);
924 return false;
925 } else if (*refcount_cache_size > combined_cache_size) {
926 error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed "
927 QCOW2_OPT_CACHE_SIZE);
928 return false;
929 }
930
931 if (l2_cache_size_set) {
932 *refcount_cache_size = combined_cache_size - *l2_cache_size;
933 } else if (refcount_cache_size_set) {
934 *l2_cache_size = combined_cache_size - *refcount_cache_size;
935 } else {
936 /* Assign as much memory as possible to the L2 cache, and
937 * use the remainder for the refcount cache */
938 if (combined_cache_size >= max_l2_cache + min_refcount_cache) {
939 *l2_cache_size = max_l2_cache;
940 *refcount_cache_size = combined_cache_size - *l2_cache_size;
941 } else {
942 *refcount_cache_size =
943 MIN(combined_cache_size, min_refcount_cache);
944 *l2_cache_size = combined_cache_size - *refcount_cache_size;
945 }
946 }
947 }
948
949 /*
950 * If the L2 cache is not enough to cover the whole disk then
951 * default to 4KB entries. Smaller entries reduce the cost of
952 * loads and evictions and increase I/O performance.
953 */
954 if (*l2_cache_size < max_l2_cache && !l2_cache_entry_size_set) {
955 *l2_cache_entry_size = MIN(s->cluster_size, 4096);
956 }
957
958 /* l2_cache_size and refcount_cache_size are ensured to have at least
959 * their minimum values in qcow2_update_options_prepare() */
960
961 if (*l2_cache_entry_size < (1 << MIN_CLUSTER_BITS) ||
962 *l2_cache_entry_size > s->cluster_size ||
963 !is_power_of_2(*l2_cache_entry_size)) {
964 error_setg(errp, "L2 cache entry size must be a power of two "
965 "between %d and the cluster size (%d)",
966 1 << MIN_CLUSTER_BITS, s->cluster_size);
967 return false;
968 }
969
970 return true;
971 }
972
973 typedef struct Qcow2ReopenState {
974 Qcow2Cache *l2_table_cache;
975 Qcow2Cache *refcount_block_cache;
976 int l2_slice_size; /* Number of entries in a slice of the L2 table */
977 bool use_lazy_refcounts;
978 int overlap_check;
979 bool discard_passthrough[QCOW2_DISCARD_MAX];
980 bool discard_no_unref;
981 uint64_t cache_clean_interval;
982 QCryptoBlockOpenOptions *crypto_opts; /* Disk encryption runtime options */
983 } Qcow2ReopenState;
984
985 static int GRAPH_RDLOCK
qcow2_update_options_prepare(BlockDriverState * bs,Qcow2ReopenState * r,QDict * options,int flags,Error ** errp)986 qcow2_update_options_prepare(BlockDriverState *bs, Qcow2ReopenState *r,
987 QDict *options, int flags, Error **errp)
988 {
989 BDRVQcow2State *s = bs->opaque;
990 QemuOpts *opts = NULL;
991 const char *opt_overlap_check, *opt_overlap_check_template;
992 int overlap_check_template = 0;
993 uint64_t l2_cache_size, l2_cache_entry_size, refcount_cache_size;
994 int i;
995 const char *encryptfmt;
996 QDict *encryptopts = NULL;
997 int ret;
998
999 qdict_extract_subqdict(options, &encryptopts, "encrypt.");
1000 encryptfmt = qdict_get_try_str(encryptopts, "format");
1001
1002 opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort);
1003 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1004 ret = -EINVAL;
1005 goto fail;
1006 }
1007
1008 /* get L2 table/refcount block cache size from command line options */
1009 if (!read_cache_sizes(bs, opts, &l2_cache_size, &l2_cache_entry_size,
1010 &refcount_cache_size, errp)) {
1011 ret = -EINVAL;
1012 goto fail;
1013 }
1014
1015 l2_cache_size /= l2_cache_entry_size;
1016 if (l2_cache_size < MIN_L2_CACHE_SIZE) {
1017 l2_cache_size = MIN_L2_CACHE_SIZE;
1018 }
1019 if (l2_cache_size > INT_MAX) {
1020 error_setg(errp, "L2 cache size too big");
1021 ret = -EINVAL;
1022 goto fail;
1023 }
1024
1025 refcount_cache_size /= s->cluster_size;
1026 if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) {
1027 refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE;
1028 }
1029 if (refcount_cache_size > INT_MAX) {
1030 error_setg(errp, "Refcount cache size too big");
1031 ret = -EINVAL;
1032 goto fail;
1033 }
1034
1035 /* alloc new L2 table/refcount block cache, flush old one */
1036 if (s->l2_table_cache) {
1037 ret = qcow2_cache_flush(bs, s->l2_table_cache);
1038 if (ret) {
1039 error_setg_errno(errp, -ret, "Failed to flush the L2 table cache");
1040 goto fail;
1041 }
1042 }
1043
1044 if (s->refcount_block_cache) {
1045 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
1046 if (ret) {
1047 error_setg_errno(errp, -ret,
1048 "Failed to flush the refcount block cache");
1049 goto fail;
1050 }
1051 }
1052
1053 r->l2_slice_size = l2_cache_entry_size / l2_entry_size(s);
1054 r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size,
1055 l2_cache_entry_size);
1056 r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size,
1057 s->cluster_size);
1058 if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) {
1059 error_setg(errp, "Could not allocate metadata caches");
1060 ret = -ENOMEM;
1061 goto fail;
1062 }
1063
1064 /* New interval for cache cleanup timer */
1065 r->cache_clean_interval =
1066 qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL,
1067 DEFAULT_CACHE_CLEAN_INTERVAL);
1068 #ifndef CONFIG_LINUX
1069 if (r->cache_clean_interval != 0) {
1070 error_setg(errp, QCOW2_OPT_CACHE_CLEAN_INTERVAL
1071 " not supported on this host");
1072 ret = -EINVAL;
1073 goto fail;
1074 }
1075 #endif
1076 if (r->cache_clean_interval > UINT_MAX) {
1077 error_setg(errp, "Cache clean interval too big");
1078 ret = -EINVAL;
1079 goto fail;
1080 }
1081
1082 /* lazy-refcounts; flush if going from enabled to disabled */
1083 r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS,
1084 (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS));
1085 if (r->use_lazy_refcounts && s->qcow_version < 3) {
1086 error_setg(errp, "Lazy refcounts require a qcow2 image with at least "
1087 "qemu 1.1 compatibility level");
1088 ret = -EINVAL;
1089 goto fail;
1090 }
1091
1092 if (s->use_lazy_refcounts && !r->use_lazy_refcounts) {
1093 ret = qcow2_mark_clean(bs);
1094 if (ret < 0) {
1095 error_setg_errno(errp, -ret, "Failed to disable lazy refcounts");
1096 goto fail;
1097 }
1098 }
1099
1100 /* Overlap check options */
1101 opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP);
1102 opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE);
1103 if (opt_overlap_check_template && opt_overlap_check &&
1104 strcmp(opt_overlap_check_template, opt_overlap_check))
1105 {
1106 error_setg(errp, "Conflicting values for qcow2 options '"
1107 QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
1108 "' ('%s')", opt_overlap_check, opt_overlap_check_template);
1109 ret = -EINVAL;
1110 goto fail;
1111 }
1112 if (!opt_overlap_check) {
1113 opt_overlap_check = opt_overlap_check_template ?: "cached";
1114 }
1115
1116 if (!strcmp(opt_overlap_check, "none")) {
1117 overlap_check_template = 0;
1118 } else if (!strcmp(opt_overlap_check, "constant")) {
1119 overlap_check_template = QCOW2_OL_CONSTANT;
1120 } else if (!strcmp(opt_overlap_check, "cached")) {
1121 overlap_check_template = QCOW2_OL_CACHED;
1122 } else if (!strcmp(opt_overlap_check, "all")) {
1123 overlap_check_template = QCOW2_OL_ALL;
1124 } else {
1125 error_setg(errp, "Unsupported value '%s' for qcow2 option "
1126 "'overlap-check'. Allowed are any of the following: "
1127 "none, constant, cached, all", opt_overlap_check);
1128 ret = -EINVAL;
1129 goto fail;
1130 }
1131
1132 r->overlap_check = 0;
1133 for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) {
1134 /* overlap-check defines a template bitmask, but every flag may be
1135 * overwritten through the associated boolean option */
1136 r->overlap_check |=
1137 qemu_opt_get_bool(opts, overlap_bool_option_names[i],
1138 overlap_check_template & (1 << i)) << i;
1139 }
1140
1141 r->discard_passthrough[QCOW2_DISCARD_NEVER] = false;
1142 r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true;
1143 r->discard_passthrough[QCOW2_DISCARD_REQUEST] =
1144 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST,
1145 flags & BDRV_O_UNMAP);
1146 r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] =
1147 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true);
1148 r->discard_passthrough[QCOW2_DISCARD_OTHER] =
1149 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false);
1150
1151 r->discard_no_unref = qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_NO_UNREF,
1152 false);
1153 if (r->discard_no_unref && s->qcow_version < 3) {
1154 error_setg(errp,
1155 "discard-no-unref is only supported since qcow2 version 3");
1156 ret = -EINVAL;
1157 goto fail;
1158 }
1159
1160 switch (s->crypt_method_header) {
1161 case QCOW_CRYPT_NONE:
1162 if (encryptfmt) {
1163 error_setg(errp, "No encryption in image header, but options "
1164 "specified format '%s'", encryptfmt);
1165 ret = -EINVAL;
1166 goto fail;
1167 }
1168 break;
1169
1170 case QCOW_CRYPT_AES:
1171 if (encryptfmt && !g_str_equal(encryptfmt, "aes")) {
1172 error_setg(errp,
1173 "Header reported 'aes' encryption format but "
1174 "options specify '%s'", encryptfmt);
1175 ret = -EINVAL;
1176 goto fail;
1177 }
1178 qdict_put_str(encryptopts, "format", "qcow");
1179 r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
1180 if (!r->crypto_opts) {
1181 ret = -EINVAL;
1182 goto fail;
1183 }
1184 break;
1185
1186 case QCOW_CRYPT_LUKS:
1187 if (encryptfmt && !g_str_equal(encryptfmt, "luks")) {
1188 error_setg(errp,
1189 "Header reported 'luks' encryption format but "
1190 "options specify '%s'", encryptfmt);
1191 ret = -EINVAL;
1192 goto fail;
1193 }
1194 qdict_put_str(encryptopts, "format", "luks");
1195 r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
1196 if (!r->crypto_opts) {
1197 ret = -EINVAL;
1198 goto fail;
1199 }
1200 break;
1201
1202 default:
1203 error_setg(errp, "Unsupported encryption method %d",
1204 s->crypt_method_header);
1205 ret = -EINVAL;
1206 goto fail;
1207 }
1208
1209 ret = 0;
1210 fail:
1211 qobject_unref(encryptopts);
1212 qemu_opts_del(opts);
1213 opts = NULL;
1214 return ret;
1215 }
1216
qcow2_update_options_commit(BlockDriverState * bs,Qcow2ReopenState * r)1217 static void qcow2_update_options_commit(BlockDriverState *bs,
1218 Qcow2ReopenState *r)
1219 {
1220 BDRVQcow2State *s = bs->opaque;
1221 int i;
1222
1223 if (s->l2_table_cache) {
1224 qcow2_cache_destroy(s->l2_table_cache);
1225 }
1226 if (s->refcount_block_cache) {
1227 qcow2_cache_destroy(s->refcount_block_cache);
1228 }
1229 s->l2_table_cache = r->l2_table_cache;
1230 s->refcount_block_cache = r->refcount_block_cache;
1231 s->l2_slice_size = r->l2_slice_size;
1232
1233 s->overlap_check = r->overlap_check;
1234 s->use_lazy_refcounts = r->use_lazy_refcounts;
1235
1236 for (i = 0; i < QCOW2_DISCARD_MAX; i++) {
1237 s->discard_passthrough[i] = r->discard_passthrough[i];
1238 }
1239
1240 s->discard_no_unref = r->discard_no_unref;
1241
1242 if (s->cache_clean_interval != r->cache_clean_interval) {
1243 cache_clean_timer_del(bs);
1244 s->cache_clean_interval = r->cache_clean_interval;
1245 cache_clean_timer_init(bs, bdrv_get_aio_context(bs));
1246 }
1247
1248 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1249 s->crypto_opts = r->crypto_opts;
1250 }
1251
qcow2_update_options_abort(BlockDriverState * bs,Qcow2ReopenState * r)1252 static void qcow2_update_options_abort(BlockDriverState *bs,
1253 Qcow2ReopenState *r)
1254 {
1255 if (r->l2_table_cache) {
1256 qcow2_cache_destroy(r->l2_table_cache);
1257 }
1258 if (r->refcount_block_cache) {
1259 qcow2_cache_destroy(r->refcount_block_cache);
1260 }
1261 qapi_free_QCryptoBlockOpenOptions(r->crypto_opts);
1262 }
1263
1264 static int coroutine_fn GRAPH_RDLOCK
qcow2_update_options(BlockDriverState * bs,QDict * options,int flags,Error ** errp)1265 qcow2_update_options(BlockDriverState *bs, QDict *options, int flags,
1266 Error **errp)
1267 {
1268 Qcow2ReopenState r = {};
1269 int ret;
1270
1271 ret = qcow2_update_options_prepare(bs, &r, options, flags, errp);
1272 if (ret >= 0) {
1273 qcow2_update_options_commit(bs, &r);
1274 } else {
1275 qcow2_update_options_abort(bs, &r);
1276 }
1277
1278 return ret;
1279 }
1280
validate_compression_type(BDRVQcow2State * s,Error ** errp)1281 static int validate_compression_type(BDRVQcow2State *s, Error **errp)
1282 {
1283 switch (s->compression_type) {
1284 case QCOW2_COMPRESSION_TYPE_ZLIB:
1285 #ifdef CONFIG_ZSTD
1286 case QCOW2_COMPRESSION_TYPE_ZSTD:
1287 #endif
1288 break;
1289
1290 default:
1291 error_setg(errp, "qcow2: unknown compression type: %u",
1292 s->compression_type);
1293 return -ENOTSUP;
1294 }
1295
1296 /*
1297 * if the compression type differs from QCOW2_COMPRESSION_TYPE_ZLIB
1298 * the incompatible feature flag must be set
1299 */
1300 if (s->compression_type == QCOW2_COMPRESSION_TYPE_ZLIB) {
1301 if (s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION) {
1302 error_setg(errp, "qcow2: Compression type incompatible feature "
1303 "bit must not be set");
1304 return -EINVAL;
1305 }
1306 } else {
1307 if (!(s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION)) {
1308 error_setg(errp, "qcow2: Compression type incompatible feature "
1309 "bit must be set");
1310 return -EINVAL;
1311 }
1312 }
1313
1314 return 0;
1315 }
1316
1317 /* Called with s->lock held. */
1318 static int coroutine_fn GRAPH_RDLOCK
qcow2_do_open(BlockDriverState * bs,QDict * options,int flags,bool open_data_file,Error ** errp)1319 qcow2_do_open(BlockDriverState *bs, QDict *options, int flags,
1320 bool open_data_file, Error **errp)
1321 {
1322 ERRP_GUARD();
1323 BDRVQcow2State *s = bs->opaque;
1324 unsigned int len, i;
1325 int ret = 0;
1326 QCowHeader header;
1327 uint64_t ext_end;
1328 uint64_t l1_vm_state_index;
1329 bool update_header = false;
1330
1331 ret = bdrv_co_pread(bs->file, 0, sizeof(header), &header, 0);
1332 if (ret < 0) {
1333 error_setg_errno(errp, -ret, "Could not read qcow2 header");
1334 goto fail;
1335 }
1336 header.magic = be32_to_cpu(header.magic);
1337 header.version = be32_to_cpu(header.version);
1338 header.backing_file_offset = be64_to_cpu(header.backing_file_offset);
1339 header.backing_file_size = be32_to_cpu(header.backing_file_size);
1340 header.size = be64_to_cpu(header.size);
1341 header.cluster_bits = be32_to_cpu(header.cluster_bits);
1342 header.crypt_method = be32_to_cpu(header.crypt_method);
1343 header.l1_table_offset = be64_to_cpu(header.l1_table_offset);
1344 header.l1_size = be32_to_cpu(header.l1_size);
1345 header.refcount_table_offset = be64_to_cpu(header.refcount_table_offset);
1346 header.refcount_table_clusters =
1347 be32_to_cpu(header.refcount_table_clusters);
1348 header.snapshots_offset = be64_to_cpu(header.snapshots_offset);
1349 header.nb_snapshots = be32_to_cpu(header.nb_snapshots);
1350
1351 if (header.magic != QCOW_MAGIC) {
1352 error_setg(errp, "Image is not in qcow2 format");
1353 ret = -EINVAL;
1354 goto fail;
1355 }
1356 if (header.version < 2 || header.version > 3) {
1357 error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version);
1358 ret = -ENOTSUP;
1359 goto fail;
1360 }
1361
1362 s->qcow_version = header.version;
1363
1364 /* Initialise cluster size */
1365 if (header.cluster_bits < MIN_CLUSTER_BITS ||
1366 header.cluster_bits > MAX_CLUSTER_BITS) {
1367 error_setg(errp, "Unsupported cluster size: 2^%" PRIu32,
1368 header.cluster_bits);
1369 ret = -EINVAL;
1370 goto fail;
1371 }
1372
1373 s->cluster_bits = header.cluster_bits;
1374 s->cluster_size = 1 << s->cluster_bits;
1375
1376 /* Initialise version 3 header fields */
1377 if (header.version == 2) {
1378 header.incompatible_features = 0;
1379 header.compatible_features = 0;
1380 header.autoclear_features = 0;
1381 header.refcount_order = 4;
1382 header.header_length = 72;
1383 } else {
1384 header.incompatible_features =
1385 be64_to_cpu(header.incompatible_features);
1386 header.compatible_features = be64_to_cpu(header.compatible_features);
1387 header.autoclear_features = be64_to_cpu(header.autoclear_features);
1388 header.refcount_order = be32_to_cpu(header.refcount_order);
1389 header.header_length = be32_to_cpu(header.header_length);
1390
1391 if (header.header_length < 104) {
1392 error_setg(errp, "qcow2 header too short");
1393 ret = -EINVAL;
1394 goto fail;
1395 }
1396 }
1397
1398 if (header.header_length > s->cluster_size) {
1399 error_setg(errp, "qcow2 header exceeds cluster size");
1400 ret = -EINVAL;
1401 goto fail;
1402 }
1403
1404 if (header.header_length > sizeof(header)) {
1405 s->unknown_header_fields_size = header.header_length - sizeof(header);
1406 s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
1407 ret = bdrv_co_pread(bs->file, sizeof(header),
1408 s->unknown_header_fields_size,
1409 s->unknown_header_fields, 0);
1410 if (ret < 0) {
1411 error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
1412 "fields");
1413 goto fail;
1414 }
1415 }
1416
1417 if (header.backing_file_offset > s->cluster_size) {
1418 error_setg(errp, "Invalid backing file offset");
1419 ret = -EINVAL;
1420 goto fail;
1421 }
1422
1423 if (header.backing_file_offset) {
1424 ext_end = header.backing_file_offset;
1425 } else {
1426 ext_end = 1 << header.cluster_bits;
1427 }
1428
1429 /* Handle feature bits */
1430 s->incompatible_features = header.incompatible_features;
1431 s->compatible_features = header.compatible_features;
1432 s->autoclear_features = header.autoclear_features;
1433
1434 /*
1435 * Handle compression type
1436 * Older qcow2 images don't contain the compression type header.
1437 * Distinguish them by the header length and use
1438 * the only valid (default) compression type in that case
1439 */
1440 if (header.header_length > offsetof(QCowHeader, compression_type)) {
1441 s->compression_type = header.compression_type;
1442 } else {
1443 s->compression_type = QCOW2_COMPRESSION_TYPE_ZLIB;
1444 }
1445
1446 ret = validate_compression_type(s, errp);
1447 if (ret) {
1448 goto fail;
1449 }
1450
1451 if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
1452 void *feature_table = NULL;
1453 qcow2_read_extensions(bs, header.header_length, ext_end,
1454 &feature_table, flags, NULL, NULL);
1455 report_unsupported_feature(errp, feature_table,
1456 s->incompatible_features &
1457 ~QCOW2_INCOMPAT_MASK);
1458 ret = -ENOTSUP;
1459 g_free(feature_table);
1460 goto fail;
1461 }
1462
1463 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
1464 /* Corrupt images may not be written to unless they are being repaired
1465 */
1466 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
1467 error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
1468 "read/write");
1469 ret = -EACCES;
1470 goto fail;
1471 }
1472 }
1473
1474 s->subclusters_per_cluster =
1475 has_subclusters(s) ? QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER : 1;
1476 s->subcluster_size = s->cluster_size / s->subclusters_per_cluster;
1477 s->subcluster_bits = ctz32(s->subcluster_size);
1478
1479 if (s->subcluster_size < (1 << MIN_CLUSTER_BITS)) {
1480 error_setg(errp, "Unsupported subcluster size: %d", s->subcluster_size);
1481 ret = -EINVAL;
1482 goto fail;
1483 }
1484
1485 /* Check support for various header values */
1486 if (header.refcount_order > 6) {
1487 error_setg(errp, "Reference count entry width too large; may not "
1488 "exceed 64 bits");
1489 ret = -EINVAL;
1490 goto fail;
1491 }
1492 s->refcount_order = header.refcount_order;
1493 s->refcount_bits = 1 << s->refcount_order;
1494 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
1495 s->refcount_max += s->refcount_max - 1;
1496
1497 s->crypt_method_header = header.crypt_method;
1498 if (s->crypt_method_header) {
1499 if (bdrv_uses_whitelist() &&
1500 s->crypt_method_header == QCOW_CRYPT_AES) {
1501 error_setg(errp,
1502 "Use of AES-CBC encrypted qcow2 images is no longer "
1503 "supported in system emulators");
1504 error_append_hint(errp,
1505 "You can use 'qemu-img convert' to convert your "
1506 "image to an alternative supported format, such "
1507 "as unencrypted qcow2, or raw with the LUKS "
1508 "format instead.\n");
1509 ret = -ENOSYS;
1510 goto fail;
1511 }
1512
1513 if (s->crypt_method_header == QCOW_CRYPT_AES) {
1514 s->crypt_physical_offset = false;
1515 } else {
1516 /* Assuming LUKS and any future crypt methods we
1517 * add will all use physical offsets, due to the
1518 * fact that the alternative is insecure... */
1519 s->crypt_physical_offset = true;
1520 }
1521
1522 bs->encrypted = true;
1523 }
1524
1525 s->l2_bits = s->cluster_bits - ctz32(l2_entry_size(s));
1526 s->l2_size = 1 << s->l2_bits;
1527 /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1528 s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
1529 s->refcount_block_size = 1 << s->refcount_block_bits;
1530 bs->total_sectors = header.size / BDRV_SECTOR_SIZE;
1531 s->csize_shift = (62 - (s->cluster_bits - 8));
1532 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
1533 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
1534
1535 s->refcount_table_offset = header.refcount_table_offset;
1536 s->refcount_table_size =
1537 header.refcount_table_clusters << (s->cluster_bits - 3);
1538
1539 if (header.refcount_table_clusters == 0 && !(flags & BDRV_O_CHECK)) {
1540 error_setg(errp, "Image does not contain a reference count table");
1541 ret = -EINVAL;
1542 goto fail;
1543 }
1544
1545 ret = qcow2_validate_table(bs, s->refcount_table_offset,
1546 header.refcount_table_clusters,
1547 s->cluster_size, QCOW_MAX_REFTABLE_SIZE,
1548 "Reference count table", errp);
1549 if (ret < 0) {
1550 goto fail;
1551 }
1552
1553 if (!(flags & BDRV_O_CHECK)) {
1554 /*
1555 * The total size in bytes of the snapshot table is checked in
1556 * qcow2_read_snapshots() because the size of each snapshot is
1557 * variable and we don't know it yet.
1558 * Here we only check the offset and number of snapshots.
1559 */
1560 ret = qcow2_validate_table(bs, header.snapshots_offset,
1561 header.nb_snapshots,
1562 sizeof(QCowSnapshotHeader),
1563 sizeof(QCowSnapshotHeader) *
1564 QCOW_MAX_SNAPSHOTS,
1565 "Snapshot table", errp);
1566 if (ret < 0) {
1567 goto fail;
1568 }
1569 }
1570
1571 /* read the level 1 table */
1572 ret = qcow2_validate_table(bs, header.l1_table_offset,
1573 header.l1_size, L1E_SIZE,
1574 QCOW_MAX_L1_SIZE, "Active L1 table", errp);
1575 if (ret < 0) {
1576 goto fail;
1577 }
1578 s->l1_size = header.l1_size;
1579 s->l1_table_offset = header.l1_table_offset;
1580
1581 l1_vm_state_index = size_to_l1(s, header.size);
1582 if (l1_vm_state_index > INT_MAX) {
1583 error_setg(errp, "Image is too big");
1584 ret = -EFBIG;
1585 goto fail;
1586 }
1587 s->l1_vm_state_index = l1_vm_state_index;
1588
1589 /* the L1 table must contain at least enough entries to put
1590 header.size bytes */
1591 if (s->l1_size < s->l1_vm_state_index) {
1592 error_setg(errp, "L1 table is too small");
1593 ret = -EINVAL;
1594 goto fail;
1595 }
1596
1597 if (s->l1_size > 0) {
1598 s->l1_table = qemu_try_blockalign(bs->file->bs, s->l1_size * L1E_SIZE);
1599 if (s->l1_table == NULL) {
1600 error_setg(errp, "Could not allocate L1 table");
1601 ret = -ENOMEM;
1602 goto fail;
1603 }
1604 ret = bdrv_co_pread(bs->file, s->l1_table_offset, s->l1_size * L1E_SIZE,
1605 s->l1_table, 0);
1606 if (ret < 0) {
1607 error_setg_errno(errp, -ret, "Could not read L1 table");
1608 goto fail;
1609 }
1610 for(i = 0;i < s->l1_size; i++) {
1611 s->l1_table[i] = be64_to_cpu(s->l1_table[i]);
1612 }
1613 }
1614
1615 /* Parse driver-specific options */
1616 ret = qcow2_update_options(bs, options, flags, errp);
1617 if (ret < 0) {
1618 goto fail;
1619 }
1620
1621 s->flags = flags;
1622
1623 ret = qcow2_refcount_init(bs);
1624 if (ret != 0) {
1625 error_setg_errno(errp, -ret, "Could not initialize refcount handling");
1626 goto fail;
1627 }
1628
1629 QLIST_INIT(&s->cluster_allocs);
1630 QTAILQ_INIT(&s->discards);
1631
1632 /* read qcow2 extensions */
1633 if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
1634 flags, &update_header, errp)) {
1635 ret = -EINVAL;
1636 goto fail;
1637 }
1638
1639 if (open_data_file && (flags & BDRV_O_NO_IO)) {
1640 /*
1641 * Don't open the data file for 'qemu-img info' so that it can be used
1642 * to verify that an untrusted qcow2 image doesn't refer to external
1643 * files.
1644 *
1645 * Note: This still makes has_data_file() return true.
1646 */
1647 if (s->incompatible_features & QCOW2_INCOMPAT_DATA_FILE) {
1648 s->data_file = NULL;
1649 } else {
1650 s->data_file = bs->file;
1651 }
1652 qdict_extract_subqdict(options, NULL, "data-file.");
1653 qdict_del(options, "data-file");
1654 } else if (open_data_file) {
1655 /* Open external data file */
1656 bdrv_graph_co_rdunlock();
1657 s->data_file = bdrv_co_open_child(NULL, options, "data-file", bs,
1658 &child_of_bds, BDRV_CHILD_DATA,
1659 true, errp);
1660 bdrv_graph_co_rdlock();
1661 if (*errp) {
1662 ret = -EINVAL;
1663 goto fail;
1664 }
1665
1666 if (s->incompatible_features & QCOW2_INCOMPAT_DATA_FILE) {
1667 if (!s->data_file && s->image_data_file) {
1668 bdrv_graph_co_rdunlock();
1669 s->data_file = bdrv_co_open_child(s->image_data_file, options,
1670 "data-file", bs,
1671 &child_of_bds,
1672 BDRV_CHILD_DATA, false, errp);
1673 bdrv_graph_co_rdlock();
1674 if (!s->data_file) {
1675 ret = -EINVAL;
1676 goto fail;
1677 }
1678 }
1679 if (!s->data_file) {
1680 error_setg(errp, "'data-file' is required for this image");
1681 ret = -EINVAL;
1682 goto fail;
1683 }
1684
1685 /* No data here */
1686 bs->file->role &= ~BDRV_CHILD_DATA;
1687
1688 /* Must succeed because we have given up permissions if anything */
1689 bdrv_child_refresh_perms(bs, bs->file, &error_abort);
1690 } else {
1691 if (s->data_file) {
1692 error_setg(errp, "'data-file' can only be set for images with "
1693 "an external data file");
1694 ret = -EINVAL;
1695 goto fail;
1696 }
1697
1698 s->data_file = bs->file;
1699
1700 if (data_file_is_raw(bs)) {
1701 error_setg(errp, "data-file-raw requires a data file");
1702 ret = -EINVAL;
1703 goto fail;
1704 }
1705 }
1706 }
1707
1708 /* qcow2_read_extension may have set up the crypto context
1709 * if the crypt method needs a header region, some methods
1710 * don't need header extensions, so must check here
1711 */
1712 if (s->crypt_method_header && !s->crypto) {
1713 if (s->crypt_method_header == QCOW_CRYPT_AES) {
1714 unsigned int cflags = 0;
1715 if (flags & BDRV_O_NO_IO) {
1716 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
1717 }
1718 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
1719 NULL, NULL, cflags, errp);
1720 if (!s->crypto) {
1721 ret = -EINVAL;
1722 goto fail;
1723 }
1724 } else {
1725 error_setg(errp, "Missing CRYPTO header for crypt method %d",
1726 s->crypt_method_header);
1727 ret = -EINVAL;
1728 goto fail;
1729 }
1730 }
1731
1732 /* read the backing file name */
1733 if (header.backing_file_offset != 0) {
1734 len = header.backing_file_size;
1735 if (len > MIN(1023, s->cluster_size - header.backing_file_offset) ||
1736 len >= sizeof(bs->backing_file)) {
1737 error_setg(errp, "Backing file name too long");
1738 ret = -EINVAL;
1739 goto fail;
1740 }
1741
1742 s->image_backing_file = g_malloc(len + 1);
1743 ret = bdrv_co_pread(bs->file, header.backing_file_offset, len,
1744 s->image_backing_file, 0);
1745 if (ret < 0) {
1746 error_setg_errno(errp, -ret, "Could not read backing file name");
1747 goto fail;
1748 }
1749 s->image_backing_file[len] = '\0';
1750
1751 /*
1752 * Update only when something has changed. This function is called by
1753 * qcow2_co_invalidate_cache(), and we do not want to reset
1754 * auto_backing_file unless necessary.
1755 */
1756 if (!g_str_equal(s->image_backing_file, bs->backing_file)) {
1757 pstrcpy(bs->backing_file, sizeof(bs->backing_file),
1758 s->image_backing_file);
1759 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
1760 s->image_backing_file);
1761 }
1762 }
1763
1764 /*
1765 * Internal snapshots; skip reading them in check mode, because
1766 * we do not need them then, and we do not want to abort because
1767 * of a broken table.
1768 */
1769 if (!(flags & BDRV_O_CHECK)) {
1770 s->snapshots_offset = header.snapshots_offset;
1771 s->nb_snapshots = header.nb_snapshots;
1772
1773 ret = qcow2_read_snapshots(bs, errp);
1774 if (ret < 0) {
1775 goto fail;
1776 }
1777 }
1778
1779 /* Clear unknown autoclear feature bits */
1780 update_header |= s->autoclear_features & ~QCOW2_AUTOCLEAR_MASK;
1781 update_header = update_header && bdrv_is_writable(bs);
1782 if (update_header) {
1783 s->autoclear_features &= QCOW2_AUTOCLEAR_MASK;
1784 }
1785
1786 /* == Handle persistent dirty bitmaps ==
1787 *
1788 * We want load dirty bitmaps in three cases:
1789 *
1790 * 1. Normal open of the disk in active mode, not related to invalidation
1791 * after migration.
1792 *
1793 * 2. Invalidation of the target vm after pre-copy phase of migration, if
1794 * bitmaps are _not_ migrating through migration channel, i.e.
1795 * 'dirty-bitmaps' capability is disabled.
1796 *
1797 * 3. Invalidation of source vm after failed or canceled migration.
1798 * This is a very interesting case. There are two possible types of
1799 * bitmaps:
1800 *
1801 * A. Stored on inactivation and removed. They should be loaded from the
1802 * image.
1803 *
1804 * B. Not stored: not-persistent bitmaps and bitmaps, migrated through
1805 * the migration channel (with dirty-bitmaps capability).
1806 *
1807 * On the other hand, there are two possible sub-cases:
1808 *
1809 * 3.1 disk was changed by somebody else while were inactive. In this
1810 * case all in-RAM dirty bitmaps (both persistent and not) are
1811 * definitely invalid. And we don't have any method to determine
1812 * this.
1813 *
1814 * Simple and safe thing is to just drop all the bitmaps of type B on
1815 * inactivation. But in this case we lose bitmaps in valid 4.2 case.
1816 *
1817 * On the other hand, resuming source vm, if disk was already changed
1818 * is a bad thing anyway: not only bitmaps, the whole vm state is
1819 * out of sync with disk.
1820 *
1821 * This means, that user or management tool, who for some reason
1822 * decided to resume source vm, after disk was already changed by
1823 * target vm, should at least drop all dirty bitmaps by hand.
1824 *
1825 * So, we can ignore this case for now, but TODO: "generation"
1826 * extension for qcow2, to determine, that image was changed after
1827 * last inactivation. And if it is changed, we will drop (or at least
1828 * mark as 'invalid' all the bitmaps of type B, both persistent
1829 * and not).
1830 *
1831 * 3.2 disk was _not_ changed while were inactive. Bitmaps may be saved
1832 * to disk ('dirty-bitmaps' capability disabled), or not saved
1833 * ('dirty-bitmaps' capability enabled), but we don't need to care
1834 * of: let's load bitmaps as always: stored bitmaps will be loaded,
1835 * and not stored has flag IN_USE=1 in the image and will be skipped
1836 * on loading.
1837 *
1838 * One remaining possible case when we don't want load bitmaps:
1839 *
1840 * 4. Open disk in inactive mode in target vm (bitmaps are migrating or
1841 * will be loaded on invalidation, no needs try loading them before)
1842 */
1843
1844 if (!(bdrv_get_flags(bs) & BDRV_O_INACTIVE)) {
1845 /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */
1846 bool header_updated;
1847 if (!qcow2_load_dirty_bitmaps(bs, &header_updated, errp)) {
1848 ret = -EINVAL;
1849 goto fail;
1850 }
1851
1852 update_header = update_header && !header_updated;
1853 }
1854
1855 if (update_header) {
1856 ret = qcow2_update_header(bs);
1857 if (ret < 0) {
1858 error_setg_errno(errp, -ret, "Could not update qcow2 header");
1859 goto fail;
1860 }
1861 }
1862
1863 bs->supported_zero_flags = header.version >= 3 ?
1864 BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK : 0;
1865 bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE;
1866
1867 /* Repair image if dirty */
1868 if (!(flags & BDRV_O_CHECK) && bdrv_is_writable(bs) &&
1869 (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
1870 BdrvCheckResult result = {0};
1871
1872 ret = qcow2_co_check_locked(bs, &result,
1873 BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
1874 if (ret < 0 || result.check_errors) {
1875 if (ret >= 0) {
1876 ret = -EIO;
1877 }
1878 error_setg_errno(errp, -ret, "Could not repair dirty image");
1879 goto fail;
1880 }
1881 }
1882
1883 #ifdef DEBUG_ALLOC
1884 {
1885 BdrvCheckResult result = {0};
1886 qcow2_check_refcounts(bs, &result, 0);
1887 }
1888 #endif
1889
1890 qemu_co_queue_init(&s->thread_task_queue);
1891
1892 return ret;
1893
1894 fail:
1895 g_free(s->image_data_file);
1896 if (open_data_file && has_data_file(bs)) {
1897 bdrv_graph_co_rdunlock();
1898 bdrv_drain_all_begin();
1899 bdrv_co_unref_child(bs, s->data_file);
1900 bdrv_drain_all_end();
1901 bdrv_graph_co_rdlock();
1902 s->data_file = NULL;
1903 }
1904 g_free(s->unknown_header_fields);
1905 cleanup_unknown_header_ext(bs);
1906 qcow2_free_snapshots(bs);
1907 qcow2_refcount_close(bs);
1908 qemu_vfree(s->l1_table);
1909 /* else pre-write overlap checks in cache_destroy may crash */
1910 s->l1_table = NULL;
1911 cache_clean_timer_del(bs);
1912 if (s->l2_table_cache) {
1913 qcow2_cache_destroy(s->l2_table_cache);
1914 }
1915 if (s->refcount_block_cache) {
1916 qcow2_cache_destroy(s->refcount_block_cache);
1917 }
1918 qcrypto_block_free(s->crypto);
1919 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1920 return ret;
1921 }
1922
1923 typedef struct QCow2OpenCo {
1924 BlockDriverState *bs;
1925 QDict *options;
1926 int flags;
1927 Error **errp;
1928 int ret;
1929 } QCow2OpenCo;
1930
qcow2_open_entry(void * opaque)1931 static void coroutine_fn qcow2_open_entry(void *opaque)
1932 {
1933 QCow2OpenCo *qoc = opaque;
1934 BDRVQcow2State *s = qoc->bs->opaque;
1935
1936 GRAPH_RDLOCK_GUARD();
1937
1938 qemu_co_mutex_lock(&s->lock);
1939 qoc->ret = qcow2_do_open(qoc->bs, qoc->options, qoc->flags, true,
1940 qoc->errp);
1941 qemu_co_mutex_unlock(&s->lock);
1942
1943 aio_wait_kick();
1944 }
1945
qcow2_open(BlockDriverState * bs,QDict * options,int flags,Error ** errp)1946 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
1947 Error **errp)
1948 {
1949 BDRVQcow2State *s = bs->opaque;
1950 QCow2OpenCo qoc = {
1951 .bs = bs,
1952 .options = options,
1953 .flags = flags,
1954 .errp = errp,
1955 .ret = -EINPROGRESS
1956 };
1957 int ret;
1958
1959 ret = bdrv_open_file_child(NULL, options, "file", bs, errp);
1960 if (ret < 0) {
1961 return ret;
1962 }
1963
1964 /* Initialise locks */
1965 qemu_co_mutex_init(&s->lock);
1966
1967 assert(!qemu_in_coroutine());
1968 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
1969
1970 aio_co_enter(bdrv_get_aio_context(bs),
1971 qemu_coroutine_create(qcow2_open_entry, &qoc));
1972 AIO_WAIT_WHILE_UNLOCKED(NULL, qoc.ret == -EINPROGRESS);
1973
1974 return qoc.ret;
1975 }
1976
qcow2_refresh_limits(BlockDriverState * bs,Error ** errp)1977 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1978 {
1979 BDRVQcow2State *s = bs->opaque;
1980
1981 if (s->crypto) {
1982 /* Encryption works on a sector granularity */
1983 bs->bl.request_alignment = qcrypto_block_get_sector_size(s->crypto);
1984 }
1985 bs->bl.pwrite_zeroes_alignment = s->subcluster_size;
1986 bs->bl.pdiscard_alignment = s->cluster_size;
1987 }
1988
1989 static int GRAPH_UNLOCKED
qcow2_reopen_prepare(BDRVReopenState * state,BlockReopenQueue * queue,Error ** errp)1990 qcow2_reopen_prepare(BDRVReopenState *state,BlockReopenQueue *queue,
1991 Error **errp)
1992 {
1993 BDRVQcow2State *s = state->bs->opaque;
1994 Qcow2ReopenState *r;
1995 int ret;
1996
1997 GLOBAL_STATE_CODE();
1998 GRAPH_RDLOCK_GUARD_MAINLOOP();
1999
2000 r = g_new0(Qcow2ReopenState, 1);
2001 state->opaque = r;
2002
2003 ret = qcow2_update_options_prepare(state->bs, r, state->options,
2004 state->flags, errp);
2005 if (ret < 0) {
2006 goto fail;
2007 }
2008
2009 /* We need to write out any unwritten data if we reopen read-only. */
2010 if ((state->flags & BDRV_O_RDWR) == 0) {
2011 ret = qcow2_reopen_bitmaps_ro(state->bs, errp);
2012 if (ret < 0) {
2013 goto fail;
2014 }
2015
2016 ret = bdrv_flush(state->bs);
2017 if (ret < 0) {
2018 goto fail;
2019 }
2020
2021 ret = qcow2_mark_clean(state->bs);
2022 if (ret < 0) {
2023 goto fail;
2024 }
2025 }
2026
2027 /*
2028 * Without an external data file, s->data_file points to the same BdrvChild
2029 * as bs->file. It needs to be resynced after reopen because bs->file may
2030 * be changed. We can't use it in the meantime.
2031 */
2032 if (!has_data_file(state->bs)) {
2033 assert(s->data_file == state->bs->file);
2034 s->data_file = NULL;
2035 }
2036
2037 return 0;
2038
2039 fail:
2040 qcow2_update_options_abort(state->bs, r);
2041 g_free(r);
2042 return ret;
2043 }
2044
qcow2_reopen_commit(BDRVReopenState * state)2045 static void qcow2_reopen_commit(BDRVReopenState *state)
2046 {
2047 BDRVQcow2State *s = state->bs->opaque;
2048
2049 GRAPH_RDLOCK_GUARD_MAINLOOP();
2050
2051 qcow2_update_options_commit(state->bs, state->opaque);
2052 if (!s->data_file) {
2053 /*
2054 * If we don't have an external data file, s->data_file was cleared by
2055 * qcow2_reopen_prepare() and needs to be updated.
2056 */
2057 s->data_file = state->bs->file;
2058 }
2059 g_free(state->opaque);
2060 }
2061
qcow2_reopen_commit_post(BDRVReopenState * state)2062 static void qcow2_reopen_commit_post(BDRVReopenState *state)
2063 {
2064 GRAPH_RDLOCK_GUARD_MAINLOOP();
2065
2066 if (state->flags & BDRV_O_RDWR) {
2067 Error *local_err = NULL;
2068
2069 if (qcow2_reopen_bitmaps_rw(state->bs, &local_err) < 0) {
2070 /*
2071 * This is not fatal, bitmaps just left read-only, so all following
2072 * writes will fail. User can remove read-only bitmaps to unblock
2073 * writes or retry reopen.
2074 */
2075 error_reportf_err(local_err,
2076 "%s: Failed to make dirty bitmaps writable: ",
2077 bdrv_get_node_name(state->bs));
2078 }
2079 }
2080 }
2081
qcow2_reopen_abort(BDRVReopenState * state)2082 static void qcow2_reopen_abort(BDRVReopenState *state)
2083 {
2084 BDRVQcow2State *s = state->bs->opaque;
2085
2086 GRAPH_RDLOCK_GUARD_MAINLOOP();
2087
2088 if (!s->data_file) {
2089 /*
2090 * If we don't have an external data file, s->data_file was cleared by
2091 * qcow2_reopen_prepare() and needs to be restored.
2092 */
2093 s->data_file = state->bs->file;
2094 }
2095 qcow2_update_options_abort(state->bs, state->opaque);
2096 g_free(state->opaque);
2097 }
2098
qcow2_join_options(QDict * options,QDict * old_options)2099 static void qcow2_join_options(QDict *options, QDict *old_options)
2100 {
2101 bool has_new_overlap_template =
2102 qdict_haskey(options, QCOW2_OPT_OVERLAP) ||
2103 qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE);
2104 bool has_new_total_cache_size =
2105 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE);
2106 bool has_all_cache_options;
2107
2108 /* New overlap template overrides all old overlap options */
2109 if (has_new_overlap_template) {
2110 qdict_del(old_options, QCOW2_OPT_OVERLAP);
2111 qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE);
2112 qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER);
2113 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1);
2114 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2);
2115 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE);
2116 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK);
2117 qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE);
2118 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1);
2119 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2);
2120 }
2121
2122 /* New total cache size overrides all old options */
2123 if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) {
2124 qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE);
2125 qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
2126 }
2127
2128 qdict_join(options, old_options, false);
2129
2130 /*
2131 * If after merging all cache size options are set, an old total size is
2132 * overwritten. Do keep all options, however, if all three are new. The
2133 * resulting error message is what we want to happen.
2134 */
2135 has_all_cache_options =
2136 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) ||
2137 qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) ||
2138 qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
2139
2140 if (has_all_cache_options && !has_new_total_cache_size) {
2141 qdict_del(options, QCOW2_OPT_CACHE_SIZE);
2142 }
2143 }
2144
2145 static int coroutine_fn GRAPH_RDLOCK
qcow2_co_block_status(BlockDriverState * bs,unsigned int mode,int64_t offset,int64_t count,int64_t * pnum,int64_t * map,BlockDriverState ** file)2146 qcow2_co_block_status(BlockDriverState *bs, unsigned int mode,
2147 int64_t offset, int64_t count, int64_t *pnum,
2148 int64_t *map, BlockDriverState **file)
2149 {
2150 BDRVQcow2State *s = bs->opaque;
2151 uint64_t host_offset;
2152 unsigned int bytes;
2153 QCow2SubclusterType type;
2154 int ret, status = 0;
2155
2156 qemu_co_mutex_lock(&s->lock);
2157
2158 if (!s->metadata_preallocation_checked) {
2159 ret = qcow2_detect_metadata_preallocation(bs);
2160 s->metadata_preallocation = (ret == 1);
2161 s->metadata_preallocation_checked = true;
2162 }
2163
2164 bytes = MIN(INT_MAX, count);
2165 ret = qcow2_get_host_offset(bs, offset, &bytes, &host_offset, &type);
2166 qemu_co_mutex_unlock(&s->lock);
2167 if (ret < 0) {
2168 return ret;
2169 }
2170
2171 *pnum = bytes;
2172
2173 if ((type == QCOW2_SUBCLUSTER_NORMAL ||
2174 type == QCOW2_SUBCLUSTER_ZERO_ALLOC ||
2175 type == QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC) && !s->crypto) {
2176 *map = host_offset;
2177 *file = s->data_file->bs;
2178 status |= BDRV_BLOCK_OFFSET_VALID;
2179 }
2180 if (type == QCOW2_SUBCLUSTER_ZERO_PLAIN ||
2181 type == QCOW2_SUBCLUSTER_ZERO_ALLOC) {
2182 status |= BDRV_BLOCK_ZERO;
2183 } else if (type != QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN &&
2184 type != QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC) {
2185 status |= BDRV_BLOCK_DATA;
2186 }
2187 if (s->metadata_preallocation && (status & BDRV_BLOCK_DATA) &&
2188 (status & BDRV_BLOCK_OFFSET_VALID))
2189 {
2190 status |= BDRV_BLOCK_RECURSE;
2191 }
2192 if (type == QCOW2_SUBCLUSTER_COMPRESSED) {
2193 status |= BDRV_BLOCK_COMPRESSED;
2194 }
2195 return status;
2196 }
2197
2198 static int coroutine_fn GRAPH_RDLOCK
qcow2_handle_l2meta(BlockDriverState * bs,QCowL2Meta ** pl2meta,bool link_l2)2199 qcow2_handle_l2meta(BlockDriverState *bs, QCowL2Meta **pl2meta, bool link_l2)
2200 {
2201 int ret = 0;
2202 QCowL2Meta *l2meta = *pl2meta;
2203
2204 while (l2meta != NULL) {
2205 QCowL2Meta *next;
2206
2207 if (link_l2) {
2208 ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
2209 if (ret) {
2210 goto out;
2211 }
2212 } else {
2213 qcow2_alloc_cluster_abort(bs, l2meta);
2214 }
2215
2216 /* Take the request off the list of running requests */
2217 QLIST_REMOVE(l2meta, next_in_flight);
2218
2219 qemu_co_queue_restart_all(&l2meta->dependent_requests);
2220
2221 next = l2meta->next;
2222 g_free(l2meta);
2223 l2meta = next;
2224 }
2225 out:
2226 *pl2meta = l2meta;
2227 return ret;
2228 }
2229
2230 static int coroutine_fn GRAPH_RDLOCK
qcow2_co_preadv_encrypted(BlockDriverState * bs,uint64_t host_offset,uint64_t offset,uint64_t bytes,QEMUIOVector * qiov,uint64_t qiov_offset)2231 qcow2_co_preadv_encrypted(BlockDriverState *bs,
2232 uint64_t host_offset,
2233 uint64_t offset,
2234 uint64_t bytes,
2235 QEMUIOVector *qiov,
2236 uint64_t qiov_offset)
2237 {
2238 int ret;
2239 BDRVQcow2State *s = bs->opaque;
2240 uint8_t *buf;
2241
2242 assert(bs->encrypted && s->crypto);
2243 assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2244
2245 /*
2246 * For encrypted images, read everything into a temporary
2247 * contiguous buffer on which the AES functions can work.
2248 * Also, decryption in a separate buffer is better as it
2249 * prevents the guest from learning information about the
2250 * encrypted nature of the virtual disk.
2251 */
2252
2253 buf = qemu_try_blockalign(s->data_file->bs, bytes);
2254 if (buf == NULL) {
2255 return -ENOMEM;
2256 }
2257
2258 BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_AIO);
2259 ret = bdrv_co_pread(s->data_file, host_offset, bytes, buf, 0);
2260 if (ret < 0) {
2261 goto fail;
2262 }
2263
2264 if (qcow2_co_decrypt(bs, host_offset, offset, buf, bytes) < 0)
2265 {
2266 ret = -EIO;
2267 goto fail;
2268 }
2269 qemu_iovec_from_buf(qiov, qiov_offset, buf, bytes);
2270
2271 fail:
2272 qemu_vfree(buf);
2273
2274 return ret;
2275 }
2276
2277 typedef struct Qcow2AioTask {
2278 AioTask task;
2279
2280 BlockDriverState *bs;
2281 QCow2SubclusterType subcluster_type; /* only for read */
2282 uint64_t host_offset; /* or l2_entry for compressed read */
2283 uint64_t offset;
2284 uint64_t bytes;
2285 QEMUIOVector *qiov;
2286 uint64_t qiov_offset;
2287 QCowL2Meta *l2meta; /* only for write */
2288 } Qcow2AioTask;
2289
2290 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task);
qcow2_add_task(BlockDriverState * bs,AioTaskPool * pool,AioTaskFunc func,QCow2SubclusterType subcluster_type,uint64_t host_offset,uint64_t offset,uint64_t bytes,QEMUIOVector * qiov,size_t qiov_offset,QCowL2Meta * l2meta)2291 static coroutine_fn int qcow2_add_task(BlockDriverState *bs,
2292 AioTaskPool *pool,
2293 AioTaskFunc func,
2294 QCow2SubclusterType subcluster_type,
2295 uint64_t host_offset,
2296 uint64_t offset,
2297 uint64_t bytes,
2298 QEMUIOVector *qiov,
2299 size_t qiov_offset,
2300 QCowL2Meta *l2meta)
2301 {
2302 Qcow2AioTask local_task;
2303 Qcow2AioTask *task = pool ? g_new(Qcow2AioTask, 1) : &local_task;
2304
2305 *task = (Qcow2AioTask) {
2306 .task.func = func,
2307 .bs = bs,
2308 .subcluster_type = subcluster_type,
2309 .qiov = qiov,
2310 .host_offset = host_offset,
2311 .offset = offset,
2312 .bytes = bytes,
2313 .qiov_offset = qiov_offset,
2314 .l2meta = l2meta,
2315 };
2316
2317 trace_qcow2_add_task(qemu_coroutine_self(), bs, pool,
2318 func == qcow2_co_preadv_task_entry ? "read" : "write",
2319 subcluster_type, host_offset, offset, bytes,
2320 qiov, qiov_offset);
2321
2322 if (!pool) {
2323 return func(&task->task);
2324 }
2325
2326 aio_task_pool_start_task(pool, &task->task);
2327
2328 return 0;
2329 }
2330
2331 static int coroutine_fn GRAPH_RDLOCK
qcow2_co_preadv_task(BlockDriverState * bs,QCow2SubclusterType subc_type,uint64_t host_offset,uint64_t offset,uint64_t bytes,QEMUIOVector * qiov,size_t qiov_offset)2332 qcow2_co_preadv_task(BlockDriverState *bs, QCow2SubclusterType subc_type,
2333 uint64_t host_offset, uint64_t offset, uint64_t bytes,
2334 QEMUIOVector *qiov, size_t qiov_offset)
2335 {
2336 BDRVQcow2State *s = bs->opaque;
2337
2338 switch (subc_type) {
2339 case QCOW2_SUBCLUSTER_ZERO_PLAIN:
2340 case QCOW2_SUBCLUSTER_ZERO_ALLOC:
2341 /* Both zero types are handled in qcow2_co_preadv_part */
2342 g_assert_not_reached();
2343
2344 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN:
2345 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC:
2346 assert(bs->backing); /* otherwise handled in qcow2_co_preadv_part */
2347
2348 BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
2349 return bdrv_co_preadv_part(bs->backing, offset, bytes,
2350 qiov, qiov_offset, 0);
2351
2352 case QCOW2_SUBCLUSTER_COMPRESSED:
2353 return qcow2_co_preadv_compressed(bs, host_offset,
2354 offset, bytes, qiov, qiov_offset);
2355
2356 case QCOW2_SUBCLUSTER_NORMAL:
2357 if (bs->encrypted) {
2358 return qcow2_co_preadv_encrypted(bs, host_offset,
2359 offset, bytes, qiov, qiov_offset);
2360 }
2361
2362 BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_AIO);
2363 return bdrv_co_preadv_part(s->data_file, host_offset,
2364 bytes, qiov, qiov_offset, 0);
2365
2366 default:
2367 g_assert_not_reached();
2368 }
2369
2370 g_assert_not_reached();
2371 }
2372
2373 /*
2374 * This function can count as GRAPH_RDLOCK because qcow2_co_preadv_part() holds
2375 * the graph lock and keeps it until this coroutine has terminated.
2376 */
qcow2_co_preadv_task_entry(AioTask * task)2377 static int coroutine_fn GRAPH_RDLOCK qcow2_co_preadv_task_entry(AioTask *task)
2378 {
2379 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
2380
2381 assert(!t->l2meta);
2382
2383 return qcow2_co_preadv_task(t->bs, t->subcluster_type,
2384 t->host_offset, t->offset, t->bytes,
2385 t->qiov, t->qiov_offset);
2386 }
2387
2388 static int coroutine_fn GRAPH_RDLOCK
qcow2_co_preadv_part(BlockDriverState * bs,int64_t offset,int64_t bytes,QEMUIOVector * qiov,size_t qiov_offset,BdrvRequestFlags flags)2389 qcow2_co_preadv_part(BlockDriverState *bs, int64_t offset, int64_t bytes,
2390 QEMUIOVector *qiov, size_t qiov_offset,
2391 BdrvRequestFlags flags)
2392 {
2393 BDRVQcow2State *s = bs->opaque;
2394 int ret = 0;
2395 unsigned int cur_bytes; /* number of bytes in current iteration */
2396 uint64_t host_offset = 0;
2397 QCow2SubclusterType type;
2398 AioTaskPool *aio = NULL;
2399
2400 while (bytes != 0 && aio_task_pool_status(aio) == 0) {
2401 /* prepare next request */
2402 cur_bytes = MIN(bytes, INT_MAX);
2403 if (s->crypto) {
2404 cur_bytes = MIN(cur_bytes,
2405 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2406 }
2407
2408 qemu_co_mutex_lock(&s->lock);
2409 ret = qcow2_get_host_offset(bs, offset, &cur_bytes,
2410 &host_offset, &type);
2411 qemu_co_mutex_unlock(&s->lock);
2412 if (ret < 0) {
2413 goto out;
2414 }
2415
2416 if (type == QCOW2_SUBCLUSTER_ZERO_PLAIN ||
2417 type == QCOW2_SUBCLUSTER_ZERO_ALLOC ||
2418 (type == QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN && !bs->backing) ||
2419 (type == QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC && !bs->backing))
2420 {
2421 qemu_iovec_memset(qiov, qiov_offset, 0, cur_bytes);
2422 } else {
2423 if (!aio && cur_bytes != bytes) {
2424 aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
2425 }
2426 ret = qcow2_add_task(bs, aio, qcow2_co_preadv_task_entry, type,
2427 host_offset, offset, cur_bytes,
2428 qiov, qiov_offset, NULL);
2429 if (ret < 0) {
2430 goto out;
2431 }
2432 }
2433
2434 bytes -= cur_bytes;
2435 offset += cur_bytes;
2436 qiov_offset += cur_bytes;
2437 }
2438
2439 out:
2440 if (aio) {
2441 aio_task_pool_wait_all(aio);
2442 if (ret == 0) {
2443 ret = aio_task_pool_status(aio);
2444 }
2445 g_free(aio);
2446 }
2447
2448 return ret;
2449 }
2450
2451 /* Check if it's possible to merge a write request with the writing of
2452 * the data from the COW regions */
merge_cow(uint64_t offset,unsigned bytes,QEMUIOVector * qiov,size_t qiov_offset,QCowL2Meta * l2meta)2453 static bool merge_cow(uint64_t offset, unsigned bytes,
2454 QEMUIOVector *qiov, size_t qiov_offset,
2455 QCowL2Meta *l2meta)
2456 {
2457 QCowL2Meta *m;
2458
2459 for (m = l2meta; m != NULL; m = m->next) {
2460 /* If both COW regions are empty then there's nothing to merge */
2461 if (m->cow_start.nb_bytes == 0 && m->cow_end.nb_bytes == 0) {
2462 continue;
2463 }
2464
2465 /* If COW regions are handled already, skip this too */
2466 if (m->skip_cow) {
2467 continue;
2468 }
2469
2470 /*
2471 * The write request should start immediately after the first
2472 * COW region. This does not always happen because the area
2473 * touched by the request can be larger than the one defined
2474 * by @m (a single request can span an area consisting of a
2475 * mix of previously unallocated and allocated clusters, that
2476 * is why @l2meta is a list).
2477 */
2478 if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) {
2479 /* In this case the request starts before this region */
2480 assert(offset < l2meta_cow_start(m));
2481 assert(m->cow_start.nb_bytes == 0);
2482 continue;
2483 }
2484
2485 /* The write request should end immediately before the second
2486 * COW region (see above for why it does not always happen) */
2487 if (m->offset + m->cow_end.offset != offset + bytes) {
2488 assert(offset + bytes > m->offset + m->cow_end.offset);
2489 assert(m->cow_end.nb_bytes == 0);
2490 continue;
2491 }
2492
2493 /* Make sure that adding both COW regions to the QEMUIOVector
2494 * does not exceed IOV_MAX */
2495 if (qemu_iovec_subvec_niov(qiov, qiov_offset, bytes) > IOV_MAX - 2) {
2496 continue;
2497 }
2498
2499 m->data_qiov = qiov;
2500 m->data_qiov_offset = qiov_offset;
2501 return true;
2502 }
2503
2504 return false;
2505 }
2506
2507 /*
2508 * Return 1 if the COW regions read as zeroes, 0 if not, < 0 on error.
2509 * Note that returning 0 does not guarantee non-zero data.
2510 */
2511 static int coroutine_fn GRAPH_RDLOCK
is_zero_cow(BlockDriverState * bs,QCowL2Meta * m)2512 is_zero_cow(BlockDriverState *bs, QCowL2Meta *m)
2513 {
2514 /*
2515 * This check is designed for optimization shortcut so it must be
2516 * efficient.
2517 * Instead of is_zero(), use bdrv_co_is_zero_fast() as it is
2518 * faster (but not as accurate and can result in false negatives).
2519 */
2520 int ret = bdrv_co_is_zero_fast(bs, m->offset + m->cow_start.offset,
2521 m->cow_start.nb_bytes);
2522 if (ret <= 0) {
2523 return ret;
2524 }
2525
2526 return bdrv_co_is_zero_fast(bs, m->offset + m->cow_end.offset,
2527 m->cow_end.nb_bytes);
2528 }
2529
2530 static int coroutine_fn GRAPH_RDLOCK
handle_alloc_space(BlockDriverState * bs,QCowL2Meta * l2meta)2531 handle_alloc_space(BlockDriverState *bs, QCowL2Meta *l2meta)
2532 {
2533 BDRVQcow2State *s = bs->opaque;
2534 QCowL2Meta *m;
2535
2536 if (!(s->data_file->bs->supported_zero_flags & BDRV_REQ_NO_FALLBACK)) {
2537 return 0;
2538 }
2539
2540 if (bs->encrypted) {
2541 return 0;
2542 }
2543
2544 for (m = l2meta; m != NULL; m = m->next) {
2545 int ret;
2546 uint64_t start_offset = m->alloc_offset + m->cow_start.offset;
2547 unsigned nb_bytes = m->cow_end.offset + m->cow_end.nb_bytes -
2548 m->cow_start.offset;
2549
2550 if (!m->cow_start.nb_bytes && !m->cow_end.nb_bytes) {
2551 continue;
2552 }
2553
2554 ret = is_zero_cow(bs, m);
2555 if (ret < 0) {
2556 return ret;
2557 } else if (ret == 0) {
2558 continue;
2559 }
2560
2561 /*
2562 * instead of writing zero COW buffers,
2563 * efficiently zero out the whole clusters
2564 */
2565
2566 ret = qcow2_pre_write_overlap_check(bs, 0, start_offset, nb_bytes,
2567 true);
2568 if (ret < 0) {
2569 return ret;
2570 }
2571
2572 BLKDBG_CO_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_SPACE);
2573 ret = bdrv_co_pwrite_zeroes(s->data_file, start_offset, nb_bytes,
2574 BDRV_REQ_NO_FALLBACK);
2575 if (ret < 0) {
2576 if (ret != -ENOTSUP && ret != -EAGAIN) {
2577 return ret;
2578 }
2579 continue;
2580 }
2581
2582 trace_qcow2_skip_cow(qemu_coroutine_self(), m->offset, m->nb_clusters);
2583 m->skip_cow = true;
2584 }
2585 return 0;
2586 }
2587
2588 /*
2589 * qcow2_co_pwritev_task
2590 * Called with s->lock unlocked
2591 * l2meta - if not NULL, qcow2_co_pwritev_task() will consume it. Caller must
2592 * not use it somehow after qcow2_co_pwritev_task() call
2593 */
2594 static coroutine_fn GRAPH_RDLOCK
qcow2_co_pwritev_task(BlockDriverState * bs,uint64_t host_offset,uint64_t offset,uint64_t bytes,QEMUIOVector * qiov,uint64_t qiov_offset,QCowL2Meta * l2meta)2595 int qcow2_co_pwritev_task(BlockDriverState *bs, uint64_t host_offset,
2596 uint64_t offset, uint64_t bytes, QEMUIOVector *qiov,
2597 uint64_t qiov_offset, QCowL2Meta *l2meta)
2598 {
2599 int ret;
2600 BDRVQcow2State *s = bs->opaque;
2601 void *crypt_buf = NULL;
2602 QEMUIOVector encrypted_qiov;
2603
2604 if (bs->encrypted) {
2605 assert(s->crypto);
2606 assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2607 crypt_buf = qemu_try_blockalign(bs->file->bs, bytes);
2608 if (crypt_buf == NULL) {
2609 ret = -ENOMEM;
2610 goto out_unlocked;
2611 }
2612 qemu_iovec_to_buf(qiov, qiov_offset, crypt_buf, bytes);
2613
2614 if (qcow2_co_encrypt(bs, host_offset, offset, crypt_buf, bytes) < 0) {
2615 ret = -EIO;
2616 goto out_unlocked;
2617 }
2618
2619 qemu_iovec_init_buf(&encrypted_qiov, crypt_buf, bytes);
2620 qiov = &encrypted_qiov;
2621 qiov_offset = 0;
2622 }
2623
2624 /* Try to efficiently initialize the physical space with zeroes */
2625 ret = handle_alloc_space(bs, l2meta);
2626 if (ret < 0) {
2627 goto out_unlocked;
2628 }
2629
2630 /*
2631 * If we need to do COW, check if it's possible to merge the
2632 * writing of the guest data together with that of the COW regions.
2633 * If it's not possible (or not necessary) then write the
2634 * guest data now.
2635 */
2636 if (!merge_cow(offset, bytes, qiov, qiov_offset, l2meta)) {
2637 BLKDBG_CO_EVENT(bs->file, BLKDBG_WRITE_AIO);
2638 trace_qcow2_writev_data(qemu_coroutine_self(), host_offset);
2639 ret = bdrv_co_pwritev_part(s->data_file, host_offset,
2640 bytes, qiov, qiov_offset, 0);
2641 if (ret < 0) {
2642 goto out_unlocked;
2643 }
2644 }
2645
2646 qemu_co_mutex_lock(&s->lock);
2647
2648 ret = qcow2_handle_l2meta(bs, &l2meta, true);
2649 goto out_locked;
2650
2651 out_unlocked:
2652 qemu_co_mutex_lock(&s->lock);
2653
2654 out_locked:
2655 qcow2_handle_l2meta(bs, &l2meta, false);
2656 qemu_co_mutex_unlock(&s->lock);
2657
2658 qemu_vfree(crypt_buf);
2659
2660 return ret;
2661 }
2662
2663 /*
2664 * This function can count as GRAPH_RDLOCK because qcow2_co_pwritev_part() holds
2665 * the graph lock and keeps it until this coroutine has terminated.
2666 */
qcow2_co_pwritev_task_entry(AioTask * task)2667 static coroutine_fn GRAPH_RDLOCK int qcow2_co_pwritev_task_entry(AioTask *task)
2668 {
2669 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
2670
2671 assert(!t->subcluster_type);
2672
2673 return qcow2_co_pwritev_task(t->bs, t->host_offset,
2674 t->offset, t->bytes, t->qiov, t->qiov_offset,
2675 t->l2meta);
2676 }
2677
2678 static int coroutine_fn GRAPH_RDLOCK
qcow2_co_pwritev_part(BlockDriverState * bs,int64_t offset,int64_t bytes,QEMUIOVector * qiov,size_t qiov_offset,BdrvRequestFlags flags)2679 qcow2_co_pwritev_part(BlockDriverState *bs, int64_t offset, int64_t bytes,
2680 QEMUIOVector *qiov, size_t qiov_offset,
2681 BdrvRequestFlags flags)
2682 {
2683 BDRVQcow2State *s = bs->opaque;
2684 int offset_in_cluster;
2685 int ret;
2686 unsigned int cur_bytes; /* number of sectors in current iteration */
2687 uint64_t host_offset;
2688 QCowL2Meta *l2meta = NULL;
2689 AioTaskPool *aio = NULL;
2690
2691 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes);
2692
2693 while (bytes != 0 && aio_task_pool_status(aio) == 0) {
2694
2695 l2meta = NULL;
2696
2697 trace_qcow2_writev_start_part(qemu_coroutine_self());
2698 offset_in_cluster = offset_into_cluster(s, offset);
2699 cur_bytes = MIN(bytes, INT_MAX);
2700 if (bs->encrypted) {
2701 cur_bytes = MIN(cur_bytes,
2702 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
2703 - offset_in_cluster);
2704 }
2705
2706 qemu_co_mutex_lock(&s->lock);
2707
2708 ret = qcow2_alloc_host_offset(bs, offset, &cur_bytes,
2709 &host_offset, &l2meta);
2710 if (ret < 0) {
2711 goto out_locked;
2712 }
2713
2714 ret = qcow2_pre_write_overlap_check(bs, 0, host_offset,
2715 cur_bytes, true);
2716 if (ret < 0) {
2717 goto out_locked;
2718 }
2719
2720 qemu_co_mutex_unlock(&s->lock);
2721
2722 if (!aio && cur_bytes != bytes) {
2723 aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
2724 }
2725 ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_task_entry, 0,
2726 host_offset, offset,
2727 cur_bytes, qiov, qiov_offset, l2meta);
2728 l2meta = NULL; /* l2meta is consumed by qcow2_co_pwritev_task() */
2729 if (ret < 0) {
2730 goto fail_nometa;
2731 }
2732
2733 bytes -= cur_bytes;
2734 offset += cur_bytes;
2735 qiov_offset += cur_bytes;
2736 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes);
2737 }
2738 ret = 0;
2739
2740 qemu_co_mutex_lock(&s->lock);
2741
2742 out_locked:
2743 qcow2_handle_l2meta(bs, &l2meta, false);
2744
2745 qemu_co_mutex_unlock(&s->lock);
2746
2747 fail_nometa:
2748 if (aio) {
2749 aio_task_pool_wait_all(aio);
2750 if (ret == 0) {
2751 ret = aio_task_pool_status(aio);
2752 }
2753 g_free(aio);
2754 }
2755
2756 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
2757
2758 return ret;
2759 }
2760
qcow2_inactivate(BlockDriverState * bs)2761 static int GRAPH_RDLOCK qcow2_inactivate(BlockDriverState *bs)
2762 {
2763 BDRVQcow2State *s = bs->opaque;
2764 int ret, result = 0;
2765 Error *local_err = NULL;
2766
2767 qcow2_store_persistent_dirty_bitmaps(bs, true, &local_err);
2768 if (local_err != NULL) {
2769 result = -EINVAL;
2770 error_reportf_err(local_err, "Lost persistent bitmaps during "
2771 "inactivation of node '%s': ",
2772 bdrv_get_device_or_node_name(bs));
2773 }
2774
2775 ret = qcow2_cache_flush(bs, s->l2_table_cache);
2776 if (ret) {
2777 result = ret;
2778 error_report("Failed to flush the L2 table cache: %s",
2779 strerror(-ret));
2780 }
2781
2782 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2783 if (ret) {
2784 result = ret;
2785 error_report("Failed to flush the refcount block cache: %s",
2786 strerror(-ret));
2787 }
2788
2789 if (result == 0) {
2790 qcow2_mark_clean(bs);
2791 }
2792
2793 return result;
2794 }
2795
2796 static void coroutine_mixed_fn GRAPH_RDLOCK
qcow2_do_close(BlockDriverState * bs,bool close_data_file)2797 qcow2_do_close(BlockDriverState *bs, bool close_data_file)
2798 {
2799 BDRVQcow2State *s = bs->opaque;
2800 qemu_vfree(s->l1_table);
2801 /* else pre-write overlap checks in cache_destroy may crash */
2802 s->l1_table = NULL;
2803
2804 if (!(s->flags & BDRV_O_INACTIVE)) {
2805 qcow2_inactivate(bs);
2806 }
2807
2808 cache_clean_timer_del(bs);
2809 qcow2_cache_destroy(s->l2_table_cache);
2810 qcow2_cache_destroy(s->refcount_block_cache);
2811
2812 qcrypto_block_free(s->crypto);
2813 s->crypto = NULL;
2814 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
2815
2816 g_free(s->unknown_header_fields);
2817 cleanup_unknown_header_ext(bs);
2818
2819 g_free(s->image_data_file);
2820 g_free(s->image_backing_file);
2821 g_free(s->image_backing_format);
2822
2823 if (close_data_file && has_data_file(bs)) {
2824 GLOBAL_STATE_CODE();
2825 bdrv_graph_rdunlock_main_loop();
2826 bdrv_graph_wrlock_drained();
2827 bdrv_unref_child(bs, s->data_file);
2828 bdrv_graph_wrunlock();
2829 s->data_file = NULL;
2830 bdrv_graph_rdlock_main_loop();
2831 }
2832
2833 qcow2_refcount_close(bs);
2834 qcow2_free_snapshots(bs);
2835 }
2836
qcow2_close(BlockDriverState * bs)2837 static void GRAPH_UNLOCKED qcow2_close(BlockDriverState *bs)
2838 {
2839 GLOBAL_STATE_CODE();
2840 GRAPH_RDLOCK_GUARD_MAINLOOP();
2841
2842 qcow2_do_close(bs, true);
2843 }
2844
2845 static void coroutine_fn GRAPH_RDLOCK
qcow2_co_invalidate_cache(BlockDriverState * bs,Error ** errp)2846 qcow2_co_invalidate_cache(BlockDriverState *bs, Error **errp)
2847 {
2848 ERRP_GUARD();
2849 BDRVQcow2State *s = bs->opaque;
2850 BdrvChild *data_file;
2851 int flags = s->flags;
2852 QCryptoBlock *crypto = NULL;
2853 QDict *options;
2854 int ret;
2855
2856 /*
2857 * Backing files are read-only which makes all of their metadata immutable,
2858 * that means we don't have to worry about reopening them here.
2859 */
2860
2861 crypto = s->crypto;
2862 s->crypto = NULL;
2863
2864 /*
2865 * Do not reopen s->data_file (i.e., have qcow2_do_close() not close it,
2866 * and then prevent qcow2_do_open() from opening it), because this function
2867 * runs in the I/O path and as such we must not invoke global-state
2868 * functions like bdrv_unref_child() and bdrv_open_child().
2869 */
2870
2871 qcow2_do_close(bs, false);
2872
2873 data_file = s->data_file;
2874 memset(s, 0, sizeof(BDRVQcow2State));
2875 s->data_file = data_file;
2876
2877 options = qdict_clone_shallow(bs->options);
2878
2879 flags &= ~BDRV_O_INACTIVE;
2880 qemu_co_mutex_lock(&s->lock);
2881 ret = qcow2_do_open(bs, options, flags, false, errp);
2882 qemu_co_mutex_unlock(&s->lock);
2883 qobject_unref(options);
2884 if (ret < 0) {
2885 error_prepend(errp, "Could not reopen qcow2 layer: ");
2886 bs->drv = NULL;
2887 return;
2888 }
2889
2890 s->crypto = crypto;
2891 }
2892
header_ext_add(char * buf,uint32_t magic,const void * s,size_t len,size_t buflen)2893 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
2894 size_t len, size_t buflen)
2895 {
2896 QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
2897 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
2898
2899 if (buflen < ext_len) {
2900 return -ENOSPC;
2901 }
2902
2903 *ext_backing_fmt = (QCowExtension) {
2904 .magic = cpu_to_be32(magic),
2905 .len = cpu_to_be32(len),
2906 };
2907
2908 if (len) {
2909 memcpy(buf + sizeof(QCowExtension), s, len);
2910 }
2911
2912 return ext_len;
2913 }
2914
2915 /*
2916 * Updates the qcow2 header, including the variable length parts of it, i.e.
2917 * the backing file name and all extensions. qcow2 was not designed to allow
2918 * such changes, so if we run out of space (we can only use the first cluster)
2919 * this function may fail.
2920 *
2921 * Returns 0 on success, -errno in error cases.
2922 */
qcow2_update_header(BlockDriverState * bs)2923 int qcow2_update_header(BlockDriverState *bs)
2924 {
2925 BDRVQcow2State *s = bs->opaque;
2926 QCowHeader *header;
2927 char *buf;
2928 size_t buflen = s->cluster_size;
2929 int ret;
2930 uint64_t total_size;
2931 uint32_t refcount_table_clusters;
2932 size_t header_length;
2933 Qcow2UnknownHeaderExtension *uext;
2934
2935 buf = qemu_blockalign(bs, buflen);
2936
2937 /* Header structure */
2938 header = (QCowHeader*) buf;
2939
2940 if (buflen < sizeof(*header)) {
2941 ret = -ENOSPC;
2942 goto fail;
2943 }
2944
2945 header_length = sizeof(*header) + s->unknown_header_fields_size;
2946 total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
2947 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2948
2949 ret = validate_compression_type(s, NULL);
2950 if (ret) {
2951 goto fail;
2952 }
2953
2954 *header = (QCowHeader) {
2955 /* Version 2 fields */
2956 .magic = cpu_to_be32(QCOW_MAGIC),
2957 .version = cpu_to_be32(s->qcow_version),
2958 .backing_file_offset = 0,
2959 .backing_file_size = 0,
2960 .cluster_bits = cpu_to_be32(s->cluster_bits),
2961 .size = cpu_to_be64(total_size),
2962 .crypt_method = cpu_to_be32(s->crypt_method_header),
2963 .l1_size = cpu_to_be32(s->l1_size),
2964 .l1_table_offset = cpu_to_be64(s->l1_table_offset),
2965 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset),
2966 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
2967 .nb_snapshots = cpu_to_be32(s->nb_snapshots),
2968 .snapshots_offset = cpu_to_be64(s->snapshots_offset),
2969
2970 /* Version 3 fields */
2971 .incompatible_features = cpu_to_be64(s->incompatible_features),
2972 .compatible_features = cpu_to_be64(s->compatible_features),
2973 .autoclear_features = cpu_to_be64(s->autoclear_features),
2974 .refcount_order = cpu_to_be32(s->refcount_order),
2975 .header_length = cpu_to_be32(header_length),
2976 .compression_type = s->compression_type,
2977 };
2978
2979 /* For older versions, write a shorter header */
2980 switch (s->qcow_version) {
2981 case 2:
2982 ret = offsetof(QCowHeader, incompatible_features);
2983 break;
2984 case 3:
2985 ret = sizeof(*header);
2986 break;
2987 default:
2988 ret = -EINVAL;
2989 goto fail;
2990 }
2991
2992 buf += ret;
2993 buflen -= ret;
2994 memset(buf, 0, buflen);
2995
2996 /* Preserve any unknown field in the header */
2997 if (s->unknown_header_fields_size) {
2998 if (buflen < s->unknown_header_fields_size) {
2999 ret = -ENOSPC;
3000 goto fail;
3001 }
3002
3003 memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
3004 buf += s->unknown_header_fields_size;
3005 buflen -= s->unknown_header_fields_size;
3006 }
3007
3008 /* Backing file format header extension */
3009 if (s->image_backing_format) {
3010 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
3011 s->image_backing_format,
3012 strlen(s->image_backing_format),
3013 buflen);
3014 if (ret < 0) {
3015 goto fail;
3016 }
3017
3018 buf += ret;
3019 buflen -= ret;
3020 }
3021
3022 /* External data file header extension */
3023 if (has_data_file(bs) && s->image_data_file) {
3024 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_DATA_FILE,
3025 s->image_data_file, strlen(s->image_data_file),
3026 buflen);
3027 if (ret < 0) {
3028 goto fail;
3029 }
3030
3031 buf += ret;
3032 buflen -= ret;
3033 }
3034
3035 /* Full disk encryption header pointer extension */
3036 if (s->crypto_header.offset != 0) {
3037 s->crypto_header.offset = cpu_to_be64(s->crypto_header.offset);
3038 s->crypto_header.length = cpu_to_be64(s->crypto_header.length);
3039 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_CRYPTO_HEADER,
3040 &s->crypto_header, sizeof(s->crypto_header),
3041 buflen);
3042 s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset);
3043 s->crypto_header.length = be64_to_cpu(s->crypto_header.length);
3044 if (ret < 0) {
3045 goto fail;
3046 }
3047 buf += ret;
3048 buflen -= ret;
3049 }
3050
3051 /*
3052 * Feature table. A mere 8 feature names occupies 392 bytes, and
3053 * when coupled with the v3 minimum header of 104 bytes plus the
3054 * 8-byte end-of-extension marker, that would leave only 8 bytes
3055 * for a backing file name in an image with 512-byte clusters.
3056 * Thus, we choose to omit this header for cluster sizes 4k and
3057 * smaller.
3058 */
3059 if (s->qcow_version >= 3 && s->cluster_size > 4096) {
3060 static const Qcow2Feature features[] = {
3061 {
3062 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
3063 .bit = QCOW2_INCOMPAT_DIRTY_BITNR,
3064 .name = "dirty bit",
3065 },
3066 {
3067 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
3068 .bit = QCOW2_INCOMPAT_CORRUPT_BITNR,
3069 .name = "corrupt bit",
3070 },
3071 {
3072 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
3073 .bit = QCOW2_INCOMPAT_DATA_FILE_BITNR,
3074 .name = "external data file",
3075 },
3076 {
3077 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
3078 .bit = QCOW2_INCOMPAT_COMPRESSION_BITNR,
3079 .name = "compression type",
3080 },
3081 {
3082 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
3083 .bit = QCOW2_INCOMPAT_EXTL2_BITNR,
3084 .name = "extended L2 entries",
3085 },
3086 {
3087 .type = QCOW2_FEAT_TYPE_COMPATIBLE,
3088 .bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
3089 .name = "lazy refcounts",
3090 },
3091 {
3092 .type = QCOW2_FEAT_TYPE_AUTOCLEAR,
3093 .bit = QCOW2_AUTOCLEAR_BITMAPS_BITNR,
3094 .name = "bitmaps",
3095 },
3096 {
3097 .type = QCOW2_FEAT_TYPE_AUTOCLEAR,
3098 .bit = QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR,
3099 .name = "raw external data",
3100 },
3101 };
3102
3103 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
3104 features, sizeof(features), buflen);
3105 if (ret < 0) {
3106 goto fail;
3107 }
3108 buf += ret;
3109 buflen -= ret;
3110 }
3111
3112 /* Bitmap extension */
3113 if (s->nb_bitmaps > 0) {
3114 Qcow2BitmapHeaderExt bitmaps_header = {
3115 .nb_bitmaps = cpu_to_be32(s->nb_bitmaps),
3116 .bitmap_directory_size =
3117 cpu_to_be64(s->bitmap_directory_size),
3118 .bitmap_directory_offset =
3119 cpu_to_be64(s->bitmap_directory_offset)
3120 };
3121 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BITMAPS,
3122 &bitmaps_header, sizeof(bitmaps_header),
3123 buflen);
3124 if (ret < 0) {
3125 goto fail;
3126 }
3127 buf += ret;
3128 buflen -= ret;
3129 }
3130
3131 /* Keep unknown header extensions */
3132 QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
3133 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
3134 if (ret < 0) {
3135 goto fail;
3136 }
3137
3138 buf += ret;
3139 buflen -= ret;
3140 }
3141
3142 /* End of header extensions */
3143 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
3144 if (ret < 0) {
3145 goto fail;
3146 }
3147
3148 buf += ret;
3149 buflen -= ret;
3150
3151 /* Backing file name */
3152 if (s->image_backing_file) {
3153 size_t backing_file_len = strlen(s->image_backing_file);
3154
3155 if (buflen < backing_file_len) {
3156 ret = -ENOSPC;
3157 goto fail;
3158 }
3159
3160 /* Using strncpy is ok here, since buf is not NUL-terminated. */
3161 strncpy(buf, s->image_backing_file, buflen);
3162
3163 header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
3164 header->backing_file_size = cpu_to_be32(backing_file_len);
3165 }
3166
3167 /* Write the new header */
3168 ret = bdrv_pwrite(bs->file, 0, s->cluster_size, header, 0);
3169 if (ret < 0) {
3170 goto fail;
3171 }
3172
3173 ret = 0;
3174 fail:
3175 qemu_vfree(header);
3176 return ret;
3177 }
3178
3179 static int coroutine_fn GRAPH_RDLOCK
qcow2_co_change_backing_file(BlockDriverState * bs,const char * backing_file,const char * backing_fmt)3180 qcow2_co_change_backing_file(BlockDriverState *bs, const char *backing_file,
3181 const char *backing_fmt)
3182 {
3183 BDRVQcow2State *s = bs->opaque;
3184
3185 /* Adding a backing file means that the external data file alone won't be
3186 * enough to make sense of the content */
3187 if (backing_file && data_file_is_raw(bs)) {
3188 return -EINVAL;
3189 }
3190
3191 if (backing_file && strlen(backing_file) > 1023) {
3192 return -EINVAL;
3193 }
3194
3195 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3196 backing_file ?: "");
3197 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
3198 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
3199
3200 g_free(s->image_backing_file);
3201 g_free(s->image_backing_format);
3202
3203 s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
3204 s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
3205
3206 return qcow2_update_header(bs);
3207 }
3208
3209 static int coroutine_fn GRAPH_RDLOCK
qcow2_set_up_encryption(BlockDriverState * bs,QCryptoBlockCreateOptions * cryptoopts,Error ** errp)3210 qcow2_set_up_encryption(BlockDriverState *bs,
3211 QCryptoBlockCreateOptions *cryptoopts,
3212 Error **errp)
3213 {
3214 BDRVQcow2State *s = bs->opaque;
3215 QCryptoBlock *crypto = NULL;
3216 int fmt, ret;
3217
3218 switch (cryptoopts->format) {
3219 case QCRYPTO_BLOCK_FORMAT_LUKS:
3220 fmt = QCOW_CRYPT_LUKS;
3221 break;
3222 case QCRYPTO_BLOCK_FORMAT_QCOW:
3223 fmt = QCOW_CRYPT_AES;
3224 break;
3225 default:
3226 error_setg(errp, "Crypto format not supported in qcow2");
3227 return -EINVAL;
3228 }
3229
3230 s->crypt_method_header = fmt;
3231
3232 crypto = qcrypto_block_create(cryptoopts, "encrypt.",
3233 qcow2_crypto_hdr_init_func,
3234 qcow2_crypto_hdr_write_func,
3235 bs, 0, errp);
3236 if (!crypto) {
3237 return -EINVAL;
3238 }
3239
3240 ret = qcow2_update_header(bs);
3241 if (ret < 0) {
3242 error_setg_errno(errp, -ret, "Could not write encryption header");
3243 goto out;
3244 }
3245
3246 ret = 0;
3247 out:
3248 qcrypto_block_free(crypto);
3249 return ret;
3250 }
3251
3252 /**
3253 * Preallocates metadata structures for data clusters between @offset (in the
3254 * guest disk) and @new_length (which is thus generally the new guest disk
3255 * size).
3256 *
3257 * Returns: 0 on success, -errno on failure.
3258 */
3259 static int coroutine_fn GRAPH_RDLOCK
preallocate_co(BlockDriverState * bs,uint64_t offset,uint64_t new_length,PreallocMode mode,Error ** errp)3260 preallocate_co(BlockDriverState *bs, uint64_t offset, uint64_t new_length,
3261 PreallocMode mode, Error **errp)
3262 {
3263 BDRVQcow2State *s = bs->opaque;
3264 uint64_t bytes;
3265 uint64_t host_offset = 0;
3266 int64_t file_length;
3267 unsigned int cur_bytes;
3268 int ret;
3269 QCowL2Meta *meta = NULL, *m;
3270
3271 assert(offset <= new_length);
3272 bytes = new_length - offset;
3273
3274 while (bytes) {
3275 cur_bytes = MIN(bytes, QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size));
3276 ret = qcow2_alloc_host_offset(bs, offset, &cur_bytes,
3277 &host_offset, &meta);
3278 if (ret < 0) {
3279 error_setg_errno(errp, -ret, "Allocating clusters failed");
3280 goto out;
3281 }
3282
3283 for (m = meta; m != NULL; m = m->next) {
3284 m->prealloc = true;
3285 }
3286
3287 ret = qcow2_handle_l2meta(bs, &meta, true);
3288 if (ret < 0) {
3289 error_setg_errno(errp, -ret, "Mapping clusters failed");
3290 goto out;
3291 }
3292
3293 /* TODO Preallocate data if requested */
3294
3295 bytes -= cur_bytes;
3296 offset += cur_bytes;
3297 }
3298
3299 /*
3300 * It is expected that the image file is large enough to actually contain
3301 * all of the allocated clusters (otherwise we get failing reads after
3302 * EOF). Extend the image to the last allocated sector.
3303 */
3304 file_length = bdrv_co_getlength(s->data_file->bs);
3305 if (file_length < 0) {
3306 error_setg_errno(errp, -file_length, "Could not get file size");
3307 ret = file_length;
3308 goto out;
3309 }
3310
3311 if (host_offset + cur_bytes > file_length) {
3312 if (mode == PREALLOC_MODE_METADATA) {
3313 mode = PREALLOC_MODE_OFF;
3314 }
3315 ret = bdrv_co_truncate(s->data_file, host_offset + cur_bytes, false,
3316 mode, 0, errp);
3317 if (ret < 0) {
3318 goto out;
3319 }
3320 }
3321
3322 ret = 0;
3323
3324 out:
3325 qcow2_handle_l2meta(bs, &meta, false);
3326 return ret;
3327 }
3328
3329 /* qcow2_refcount_metadata_size:
3330 * @clusters: number of clusters to refcount (including data and L1/L2 tables)
3331 * @cluster_size: size of a cluster, in bytes
3332 * @refcount_order: refcount bits power-of-2 exponent
3333 * @generous_increase: allow for the refcount table to be 1.5x as large as it
3334 * needs to be
3335 *
3336 * Returns: Number of bytes required for refcount blocks and table metadata.
3337 */
qcow2_refcount_metadata_size(int64_t clusters,size_t cluster_size,int refcount_order,bool generous_increase,uint64_t * refblock_count)3338 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size,
3339 int refcount_order, bool generous_increase,
3340 uint64_t *refblock_count)
3341 {
3342 /*
3343 * Every host cluster is reference-counted, including metadata (even
3344 * refcount metadata is recursively included).
3345 *
3346 * An accurate formula for the size of refcount metadata size is difficult
3347 * to derive. An easier method of calculation is finding the fixed point
3348 * where no further refcount blocks or table clusters are required to
3349 * reference count every cluster.
3350 */
3351 int64_t blocks_per_table_cluster = cluster_size / REFTABLE_ENTRY_SIZE;
3352 int64_t refcounts_per_block = cluster_size * 8 / (1 << refcount_order);
3353 int64_t table = 0; /* number of refcount table clusters */
3354 int64_t blocks = 0; /* number of refcount block clusters */
3355 int64_t last;
3356 int64_t n = 0;
3357
3358 do {
3359 last = n;
3360 blocks = DIV_ROUND_UP(clusters + table + blocks, refcounts_per_block);
3361 table = DIV_ROUND_UP(blocks, blocks_per_table_cluster);
3362 n = clusters + blocks + table;
3363
3364 if (n == last && generous_increase) {
3365 clusters += DIV_ROUND_UP(table, 2);
3366 n = 0; /* force another loop */
3367 generous_increase = false;
3368 }
3369 } while (n != last);
3370
3371 if (refblock_count) {
3372 *refblock_count = blocks;
3373 }
3374
3375 return (blocks + table) * cluster_size;
3376 }
3377
3378 /**
3379 * qcow2_calc_prealloc_size:
3380 * @total_size: virtual disk size in bytes
3381 * @cluster_size: cluster size in bytes
3382 * @refcount_order: refcount bits power-of-2 exponent
3383 * @extended_l2: true if the image has extended L2 entries
3384 *
3385 * Returns: Total number of bytes required for the fully allocated image
3386 * (including metadata).
3387 */
qcow2_calc_prealloc_size(int64_t total_size,size_t cluster_size,int refcount_order,bool extended_l2)3388 static int64_t qcow2_calc_prealloc_size(int64_t total_size,
3389 size_t cluster_size,
3390 int refcount_order,
3391 bool extended_l2)
3392 {
3393 int64_t meta_size = 0;
3394 uint64_t nl1e, nl2e;
3395 int64_t aligned_total_size = ROUND_UP(total_size, cluster_size);
3396 size_t l2e_size = extended_l2 ? L2E_SIZE_EXTENDED : L2E_SIZE_NORMAL;
3397
3398 /* header: 1 cluster */
3399 meta_size += cluster_size;
3400
3401 /* total size of L2 tables */
3402 nl2e = aligned_total_size / cluster_size;
3403 nl2e = ROUND_UP(nl2e, cluster_size / l2e_size);
3404 meta_size += nl2e * l2e_size;
3405
3406 /* total size of L1 tables */
3407 nl1e = nl2e * l2e_size / cluster_size;
3408 nl1e = ROUND_UP(nl1e, cluster_size / L1E_SIZE);
3409 meta_size += nl1e * L1E_SIZE;
3410
3411 /* total size of refcount table and blocks */
3412 meta_size += qcow2_refcount_metadata_size(
3413 (meta_size + aligned_total_size) / cluster_size,
3414 cluster_size, refcount_order, false, NULL);
3415
3416 return meta_size + aligned_total_size;
3417 }
3418
validate_cluster_size(size_t cluster_size,bool extended_l2,Error ** errp)3419 static bool validate_cluster_size(size_t cluster_size, bool extended_l2,
3420 Error **errp)
3421 {
3422 int cluster_bits = ctz32(cluster_size);
3423 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
3424 (1 << cluster_bits) != cluster_size)
3425 {
3426 error_setg(errp, "Cluster size must be a power of two between %d and "
3427 "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
3428 return false;
3429 }
3430
3431 if (extended_l2) {
3432 unsigned min_cluster_size =
3433 (1 << MIN_CLUSTER_BITS) * QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER;
3434 if (cluster_size < min_cluster_size) {
3435 error_setg(errp, "Extended L2 entries are only supported with "
3436 "cluster sizes of at least %u bytes", min_cluster_size);
3437 return false;
3438 }
3439 }
3440
3441 return true;
3442 }
3443
qcow2_opt_get_cluster_size_del(QemuOpts * opts,bool extended_l2,Error ** errp)3444 static size_t qcow2_opt_get_cluster_size_del(QemuOpts *opts, bool extended_l2,
3445 Error **errp)
3446 {
3447 size_t cluster_size;
3448
3449 cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
3450 DEFAULT_CLUSTER_SIZE);
3451 if (!validate_cluster_size(cluster_size, extended_l2, errp)) {
3452 return 0;
3453 }
3454 return cluster_size;
3455 }
3456
qcow2_opt_get_version_del(QemuOpts * opts,Error ** errp)3457 static int qcow2_opt_get_version_del(QemuOpts *opts, Error **errp)
3458 {
3459 char *buf;
3460 int ret;
3461
3462 buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
3463 if (!buf) {
3464 ret = 3; /* default */
3465 } else if (!strcmp(buf, "0.10")) {
3466 ret = 2;
3467 } else if (!strcmp(buf, "1.1")) {
3468 ret = 3;
3469 } else {
3470 error_setg(errp, "Invalid compatibility level: '%s'", buf);
3471 ret = -EINVAL;
3472 }
3473 g_free(buf);
3474 return ret;
3475 }
3476
qcow2_opt_get_refcount_bits_del(QemuOpts * opts,int version,Error ** errp)3477 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts *opts, int version,
3478 Error **errp)
3479 {
3480 uint64_t refcount_bits;
3481
3482 refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 16);
3483 if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
3484 error_setg(errp, "Refcount width must be a power of two and may not "
3485 "exceed 64 bits");
3486 return 0;
3487 }
3488
3489 if (version < 3 && refcount_bits != 16) {
3490 error_setg(errp, "Different refcount widths than 16 bits require "
3491 "compatibility level 1.1 or above (use compat=1.1 or "
3492 "greater)");
3493 return 0;
3494 }
3495
3496 return refcount_bits;
3497 }
3498
3499 static int coroutine_fn GRAPH_UNLOCKED
qcow2_co_create(BlockdevCreateOptions * create_options,Error ** errp)3500 qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp)
3501 {
3502 ERRP_GUARD();
3503 BlockdevCreateOptionsQcow2 *qcow2_opts;
3504 QDict *options;
3505
3506 /*
3507 * Open the image file and write a minimal qcow2 header.
3508 *
3509 * We keep things simple and start with a zero-sized image. We also
3510 * do without refcount blocks or a L1 table for now. We'll fix the
3511 * inconsistency later.
3512 *
3513 * We do need a refcount table because growing the refcount table means
3514 * allocating two new refcount blocks - the second of which would be at
3515 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
3516 * size for any qcow2 image.
3517 */
3518 BlockBackend *blk = NULL;
3519 BlockDriverState *bs = NULL;
3520 BlockDriverState *data_bs = NULL;
3521 QCowHeader *header;
3522 size_t cluster_size;
3523 int version;
3524 int refcount_order;
3525 uint64_t *refcount_table;
3526 int ret;
3527 uint8_t compression_type = QCOW2_COMPRESSION_TYPE_ZLIB;
3528
3529 assert(create_options->driver == BLOCKDEV_DRIVER_QCOW2);
3530 qcow2_opts = &create_options->u.qcow2;
3531
3532 bs = bdrv_co_open_blockdev_ref(qcow2_opts->file, errp);
3533 if (bs == NULL) {
3534 return -EIO;
3535 }
3536
3537 /* Validate options and set default values */
3538 if (!QEMU_IS_ALIGNED(qcow2_opts->size, BDRV_SECTOR_SIZE)) {
3539 error_setg(errp, "Image size must be a multiple of %u bytes",
3540 (unsigned) BDRV_SECTOR_SIZE);
3541 ret = -EINVAL;
3542 goto out;
3543 }
3544
3545 if (qcow2_opts->has_version) {
3546 switch (qcow2_opts->version) {
3547 case BLOCKDEV_QCOW2_VERSION_V2:
3548 version = 2;
3549 break;
3550 case BLOCKDEV_QCOW2_VERSION_V3:
3551 version = 3;
3552 break;
3553 default:
3554 g_assert_not_reached();
3555 }
3556 } else {
3557 version = 3;
3558 }
3559
3560 if (qcow2_opts->has_cluster_size) {
3561 cluster_size = qcow2_opts->cluster_size;
3562 } else {
3563 cluster_size = DEFAULT_CLUSTER_SIZE;
3564 }
3565
3566 if (!qcow2_opts->has_extended_l2) {
3567 qcow2_opts->extended_l2 = false;
3568 }
3569 if (qcow2_opts->extended_l2) {
3570 if (version < 3) {
3571 error_setg(errp, "Extended L2 entries are only supported with "
3572 "compatibility level 1.1 and above (use version=v3 or "
3573 "greater)");
3574 ret = -EINVAL;
3575 goto out;
3576 }
3577 }
3578
3579 if (!validate_cluster_size(cluster_size, qcow2_opts->extended_l2, errp)) {
3580 ret = -EINVAL;
3581 goto out;
3582 }
3583
3584 if (!qcow2_opts->has_preallocation) {
3585 qcow2_opts->preallocation = PREALLOC_MODE_OFF;
3586 }
3587 if (qcow2_opts->backing_file &&
3588 qcow2_opts->preallocation != PREALLOC_MODE_OFF &&
3589 !qcow2_opts->extended_l2)
3590 {
3591 error_setg(errp, "Backing file and preallocation can only be used at "
3592 "the same time if extended_l2 is on");
3593 ret = -EINVAL;
3594 goto out;
3595 }
3596 if (qcow2_opts->has_backing_fmt && !qcow2_opts->backing_file) {
3597 error_setg(errp, "Backing format cannot be used without backing file");
3598 ret = -EINVAL;
3599 goto out;
3600 }
3601
3602 if (!qcow2_opts->has_lazy_refcounts) {
3603 qcow2_opts->lazy_refcounts = false;
3604 }
3605 if (version < 3 && qcow2_opts->lazy_refcounts) {
3606 error_setg(errp, "Lazy refcounts only supported with compatibility "
3607 "level 1.1 and above (use version=v3 or greater)");
3608 ret = -EINVAL;
3609 goto out;
3610 }
3611
3612 if (!qcow2_opts->has_refcount_bits) {
3613 qcow2_opts->refcount_bits = 16;
3614 }
3615 if (qcow2_opts->refcount_bits > 64 ||
3616 !is_power_of_2(qcow2_opts->refcount_bits))
3617 {
3618 error_setg(errp, "Refcount width must be a power of two and may not "
3619 "exceed 64 bits");
3620 ret = -EINVAL;
3621 goto out;
3622 }
3623 if (version < 3 && qcow2_opts->refcount_bits != 16) {
3624 error_setg(errp, "Different refcount widths than 16 bits require "
3625 "compatibility level 1.1 or above (use version=v3 or "
3626 "greater)");
3627 ret = -EINVAL;
3628 goto out;
3629 }
3630 refcount_order = ctz32(qcow2_opts->refcount_bits);
3631
3632 if (qcow2_opts->data_file_raw && !qcow2_opts->data_file) {
3633 error_setg(errp, "data-file-raw requires data-file");
3634 ret = -EINVAL;
3635 goto out;
3636 }
3637 if (qcow2_opts->data_file_raw && qcow2_opts->backing_file) {
3638 error_setg(errp, "Backing file and data-file-raw cannot be used at "
3639 "the same time");
3640 ret = -EINVAL;
3641 goto out;
3642 }
3643 if (qcow2_opts->data_file_raw &&
3644 qcow2_opts->preallocation == PREALLOC_MODE_OFF)
3645 {
3646 /*
3647 * data-file-raw means that "the external data file can be
3648 * read as a consistent standalone raw image without looking
3649 * at the qcow2 metadata." It does not say that the metadata
3650 * must be ignored, though (and the qcow2 driver in fact does
3651 * not ignore it), so the L1/L2 tables must be present and
3652 * give a 1:1 mapping, so you get the same result regardless
3653 * of whether you look at the metadata or whether you ignore
3654 * it.
3655 */
3656 qcow2_opts->preallocation = PREALLOC_MODE_METADATA;
3657
3658 /*
3659 * Cannot use preallocation with backing files, but giving a
3660 * backing file when specifying data_file_raw is an error
3661 * anyway.
3662 */
3663 assert(!qcow2_opts->backing_file);
3664 }
3665
3666 if (qcow2_opts->data_file) {
3667 if (version < 3) {
3668 error_setg(errp, "External data files are only supported with "
3669 "compatibility level 1.1 and above (use version=v3 or "
3670 "greater)");
3671 ret = -EINVAL;
3672 goto out;
3673 }
3674 data_bs = bdrv_co_open_blockdev_ref(qcow2_opts->data_file, errp);
3675 if (data_bs == NULL) {
3676 ret = -EIO;
3677 goto out;
3678 }
3679 }
3680
3681 if (qcow2_opts->has_compression_type &&
3682 qcow2_opts->compression_type != QCOW2_COMPRESSION_TYPE_ZLIB) {
3683
3684 ret = -EINVAL;
3685
3686 if (version < 3) {
3687 error_setg(errp, "Non-zlib compression type is only supported with "
3688 "compatibility level 1.1 and above (use version=v3 or "
3689 "greater)");
3690 goto out;
3691 }
3692
3693 switch (qcow2_opts->compression_type) {
3694 #ifdef CONFIG_ZSTD
3695 case QCOW2_COMPRESSION_TYPE_ZSTD:
3696 break;
3697 #endif
3698 default:
3699 error_setg(errp, "Unknown compression type");
3700 goto out;
3701 }
3702
3703 compression_type = qcow2_opts->compression_type;
3704 }
3705
3706 /* Create BlockBackend to write to the image */
3707 blk = blk_co_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL,
3708 errp);
3709 if (!blk) {
3710 ret = -EPERM;
3711 goto out;
3712 }
3713 blk_set_allow_write_beyond_eof(blk, true);
3714
3715 /* Write the header */
3716 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
3717 header = g_malloc0(cluster_size);
3718 *header = (QCowHeader) {
3719 .magic = cpu_to_be32(QCOW_MAGIC),
3720 .version = cpu_to_be32(version),
3721 .cluster_bits = cpu_to_be32(ctz32(cluster_size)),
3722 .size = cpu_to_be64(0),
3723 .l1_table_offset = cpu_to_be64(0),
3724 .l1_size = cpu_to_be32(0),
3725 .refcount_table_offset = cpu_to_be64(cluster_size),
3726 .refcount_table_clusters = cpu_to_be32(1),
3727 .refcount_order = cpu_to_be32(refcount_order),
3728 /* don't deal with endianness since compression_type is 1 byte long */
3729 .compression_type = compression_type,
3730 .header_length = cpu_to_be32(sizeof(*header)),
3731 };
3732
3733 /* We'll update this to correct value later */
3734 header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
3735
3736 if (qcow2_opts->lazy_refcounts) {
3737 header->compatible_features |=
3738 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
3739 }
3740 if (data_bs) {
3741 header->incompatible_features |=
3742 cpu_to_be64(QCOW2_INCOMPAT_DATA_FILE);
3743 }
3744 if (qcow2_opts->data_file_raw) {
3745 header->autoclear_features |=
3746 cpu_to_be64(QCOW2_AUTOCLEAR_DATA_FILE_RAW);
3747 }
3748 if (compression_type != QCOW2_COMPRESSION_TYPE_ZLIB) {
3749 header->incompatible_features |=
3750 cpu_to_be64(QCOW2_INCOMPAT_COMPRESSION);
3751 }
3752
3753 if (qcow2_opts->extended_l2) {
3754 header->incompatible_features |=
3755 cpu_to_be64(QCOW2_INCOMPAT_EXTL2);
3756 }
3757
3758 ret = blk_co_pwrite(blk, 0, cluster_size, header, 0);
3759 g_free(header);
3760 if (ret < 0) {
3761 error_setg_errno(errp, -ret, "Could not write qcow2 header");
3762 goto out;
3763 }
3764
3765 /* Write a refcount table with one refcount block */
3766 refcount_table = g_malloc0(2 * cluster_size);
3767 refcount_table[0] = cpu_to_be64(2 * cluster_size);
3768 ret = blk_co_pwrite(blk, cluster_size, 2 * cluster_size, refcount_table, 0);
3769 g_free(refcount_table);
3770
3771 if (ret < 0) {
3772 error_setg_errno(errp, -ret, "Could not write refcount table");
3773 goto out;
3774 }
3775
3776 blk_co_unref(blk);
3777 blk = NULL;
3778
3779 /*
3780 * And now open the image and make it consistent first (i.e. increase the
3781 * refcount of the cluster that is occupied by the header and the refcount
3782 * table)
3783 */
3784 options = qdict_new();
3785 qdict_put_str(options, "driver", "qcow2");
3786 qdict_put_str(options, "file", bs->node_name);
3787 if (data_bs) {
3788 qdict_put_str(options, "data-file", data_bs->node_name);
3789 }
3790 blk = blk_co_new_open(NULL, NULL, options,
3791 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH,
3792 errp);
3793 if (blk == NULL) {
3794 ret = -EIO;
3795 goto out;
3796 }
3797
3798 bdrv_graph_co_rdlock();
3799 ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size);
3800 if (ret < 0) {
3801 bdrv_graph_co_rdunlock();
3802 error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
3803 "header and refcount table");
3804 goto out;
3805
3806 } else if (ret != 0) {
3807 error_report("Huh, first cluster in empty image is already in use?");
3808 abort();
3809 }
3810
3811 /* Set the external data file if necessary */
3812 if (data_bs) {
3813 BDRVQcow2State *s = blk_bs(blk)->opaque;
3814 s->image_data_file = g_strdup(data_bs->filename);
3815 }
3816
3817 /* Create a full header (including things like feature table) */
3818 ret = qcow2_update_header(blk_bs(blk));
3819 bdrv_graph_co_rdunlock();
3820
3821 if (ret < 0) {
3822 error_setg_errno(errp, -ret, "Could not update qcow2 header");
3823 goto out;
3824 }
3825
3826 /* Okay, now that we have a valid image, let's give it the right size */
3827 ret = blk_co_truncate(blk, qcow2_opts->size, false,
3828 qcow2_opts->preallocation, 0, errp);
3829 if (ret < 0) {
3830 error_prepend(errp, "Could not resize image: ");
3831 goto out;
3832 }
3833
3834 /* Want a backing file? There you go. */
3835 if (qcow2_opts->backing_file) {
3836 const char *backing_format = NULL;
3837
3838 if (qcow2_opts->has_backing_fmt) {
3839 backing_format = BlockdevDriver_str(qcow2_opts->backing_fmt);
3840 }
3841
3842 bdrv_graph_co_rdlock();
3843 ret = bdrv_co_change_backing_file(blk_bs(blk), qcow2_opts->backing_file,
3844 backing_format, false);
3845 bdrv_graph_co_rdunlock();
3846
3847 if (ret < 0) {
3848 error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
3849 "with format '%s'", qcow2_opts->backing_file,
3850 backing_format);
3851 goto out;
3852 }
3853 }
3854
3855 /* Want encryption? There you go. */
3856 if (qcow2_opts->encrypt) {
3857 bdrv_graph_co_rdlock();
3858 ret = qcow2_set_up_encryption(blk_bs(blk), qcow2_opts->encrypt, errp);
3859 bdrv_graph_co_rdunlock();
3860
3861 if (ret < 0) {
3862 goto out;
3863 }
3864 }
3865
3866 blk_co_unref(blk);
3867 blk = NULL;
3868
3869 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
3870 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
3871 * have to setup decryption context. We're not doing any I/O on the top
3872 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
3873 * not have effect.
3874 */
3875 options = qdict_new();
3876 qdict_put_str(options, "driver", "qcow2");
3877 qdict_put_str(options, "file", bs->node_name);
3878 if (data_bs) {
3879 qdict_put_str(options, "data-file", data_bs->node_name);
3880 }
3881 blk = blk_co_new_open(NULL, NULL, options,
3882 BDRV_O_RDWR | BDRV_O_NO_BACKING | BDRV_O_NO_IO,
3883 errp);
3884 if (blk == NULL) {
3885 ret = -EIO;
3886 goto out;
3887 }
3888
3889 ret = 0;
3890 out:
3891 blk_co_unref(blk);
3892 bdrv_co_unref(bs);
3893 bdrv_co_unref(data_bs);
3894 return ret;
3895 }
3896
3897 static int coroutine_fn GRAPH_UNLOCKED
qcow2_co_create_opts(BlockDriver * drv,const char * filename,QemuOpts * opts,Error ** errp)3898 qcow2_co_create_opts(BlockDriver *drv, const char *filename, QemuOpts *opts,
3899 Error **errp)
3900 {
3901 BlockdevCreateOptions *create_options = NULL;
3902 QDict *qdict;
3903 Visitor *v;
3904 BlockDriverState *bs = NULL;
3905 BlockDriverState *data_bs = NULL;
3906 const char *val;
3907 int ret;
3908
3909 /* Only the keyval visitor supports the dotted syntax needed for
3910 * encryption, so go through a QDict before getting a QAPI type. Ignore
3911 * options meant for the protocol layer so that the visitor doesn't
3912 * complain. */
3913 qdict = qemu_opts_to_qdict_filtered(opts, NULL, bdrv_qcow2.create_opts,
3914 true);
3915
3916 /* Handle encryption options */
3917 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT);
3918 if (val && !strcmp(val, "on")) {
3919 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow");
3920 } else if (val && !strcmp(val, "off")) {
3921 qdict_del(qdict, BLOCK_OPT_ENCRYPT);
3922 }
3923
3924 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT);
3925 if (val && !strcmp(val, "aes")) {
3926 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow");
3927 }
3928
3929 /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into
3930 * version=v2/v3 below. */
3931 val = qdict_get_try_str(qdict, BLOCK_OPT_COMPAT_LEVEL);
3932 if (val && !strcmp(val, "0.10")) {
3933 qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v2");
3934 } else if (val && !strcmp(val, "1.1")) {
3935 qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v3");
3936 }
3937
3938 /* Change legacy command line options into QMP ones */
3939 static const QDictRenames opt_renames[] = {
3940 { BLOCK_OPT_BACKING_FILE, "backing-file" },
3941 { BLOCK_OPT_BACKING_FMT, "backing-fmt" },
3942 { BLOCK_OPT_CLUSTER_SIZE, "cluster-size" },
3943 { BLOCK_OPT_LAZY_REFCOUNTS, "lazy-refcounts" },
3944 { BLOCK_OPT_EXTL2, "extended-l2" },
3945 { BLOCK_OPT_REFCOUNT_BITS, "refcount-bits" },
3946 { BLOCK_OPT_ENCRYPT, BLOCK_OPT_ENCRYPT_FORMAT },
3947 { BLOCK_OPT_COMPAT_LEVEL, "version" },
3948 { BLOCK_OPT_DATA_FILE_RAW, "data-file-raw" },
3949 { BLOCK_OPT_COMPRESSION_TYPE, "compression-type" },
3950 { NULL, NULL },
3951 };
3952
3953 if (!qdict_rename_keys(qdict, opt_renames, errp)) {
3954 ret = -EINVAL;
3955 goto finish;
3956 }
3957
3958 /* Create and open the file (protocol layer) */
3959 ret = bdrv_co_create_file(filename, opts, errp);
3960 if (ret < 0) {
3961 goto finish;
3962 }
3963
3964 bs = bdrv_co_open(filename, NULL, NULL,
3965 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
3966 if (bs == NULL) {
3967 ret = -EIO;
3968 goto finish;
3969 }
3970
3971 /* Create and open an external data file (protocol layer) */
3972 val = qdict_get_try_str(qdict, BLOCK_OPT_DATA_FILE);
3973 if (val) {
3974 ret = bdrv_co_create_file(val, opts, errp);
3975 if (ret < 0) {
3976 goto finish;
3977 }
3978
3979 data_bs = bdrv_co_open(val, NULL, NULL,
3980 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
3981 errp);
3982 if (data_bs == NULL) {
3983 ret = -EIO;
3984 goto finish;
3985 }
3986
3987 qdict_del(qdict, BLOCK_OPT_DATA_FILE);
3988 qdict_put_str(qdict, "data-file", data_bs->node_name);
3989 }
3990
3991 /* Set 'driver' and 'node' options */
3992 qdict_put_str(qdict, "driver", "qcow2");
3993 qdict_put_str(qdict, "file", bs->node_name);
3994
3995 /* Now get the QAPI type BlockdevCreateOptions */
3996 v = qobject_input_visitor_new_flat_confused(qdict, errp);
3997 if (!v) {
3998 ret = -EINVAL;
3999 goto finish;
4000 }
4001
4002 visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp);
4003 visit_free(v);
4004 if (!create_options) {
4005 ret = -EINVAL;
4006 goto finish;
4007 }
4008
4009 /* Silently round up size */
4010 create_options->u.qcow2.size = ROUND_UP(create_options->u.qcow2.size,
4011 BDRV_SECTOR_SIZE);
4012
4013 /* Create the qcow2 image (format layer) */
4014 ret = qcow2_co_create(create_options, errp);
4015 finish:
4016 if (ret < 0) {
4017 bdrv_graph_co_rdlock();
4018 bdrv_co_delete_file_noerr(bs);
4019 bdrv_co_delete_file_noerr(data_bs);
4020 bdrv_graph_co_rdunlock();
4021 } else {
4022 ret = 0;
4023 }
4024
4025 qobject_unref(qdict);
4026 bdrv_co_unref(bs);
4027 bdrv_co_unref(data_bs);
4028 qapi_free_BlockdevCreateOptions(create_options);
4029 return ret;
4030 }
4031
4032
4033 static bool coroutine_fn GRAPH_RDLOCK
is_zero(BlockDriverState * bs,int64_t offset,int64_t bytes)4034 is_zero(BlockDriverState *bs, int64_t offset, int64_t bytes)
4035 {
4036 int64_t nr;
4037 int res;
4038
4039 /* Clamp to image length, before checking status of underlying sectors */
4040 if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) {
4041 bytes = bs->total_sectors * BDRV_SECTOR_SIZE - offset;
4042 }
4043
4044 if (!bytes) {
4045 return true;
4046 }
4047
4048 /*
4049 * bdrv_block_status_above doesn't merge different types of zeros, for
4050 * example, zeros which come from the region which is unallocated in
4051 * the whole backing chain, and zeros which come because of a short
4052 * backing file. So, we need a loop.
4053 */
4054 do {
4055 res = bdrv_co_block_status_above(bs, NULL, offset, bytes, &nr, NULL, NULL);
4056 offset += nr;
4057 bytes -= nr;
4058 } while (res >= 0 && (res & BDRV_BLOCK_ZERO) && nr && bytes);
4059
4060 return res >= 0 && (res & BDRV_BLOCK_ZERO) && bytes == 0;
4061 }
4062
4063 static int coroutine_fn GRAPH_RDLOCK
qcow2_co_pwrite_zeroes(BlockDriverState * bs,int64_t offset,int64_t bytes,BdrvRequestFlags flags)4064 qcow2_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes,
4065 BdrvRequestFlags flags)
4066 {
4067 int ret;
4068 BDRVQcow2State *s = bs->opaque;
4069
4070 uint32_t head = offset_into_subcluster(s, offset);
4071 uint32_t tail = ROUND_UP(offset + bytes, s->subcluster_size) -
4072 (offset + bytes);
4073
4074 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, bytes);
4075 if (offset + bytes == bs->total_sectors * BDRV_SECTOR_SIZE) {
4076 tail = 0;
4077 }
4078
4079 if (head || tail) {
4080 uint64_t off;
4081 unsigned int nr;
4082 QCow2SubclusterType type;
4083
4084 assert(head + bytes + tail <= s->subcluster_size);
4085
4086 /* check whether remainder of cluster already reads as zero */
4087 if (!(is_zero(bs, offset - head, head) &&
4088 is_zero(bs, offset + bytes, tail))) {
4089 return -ENOTSUP;
4090 }
4091
4092 qemu_co_mutex_lock(&s->lock);
4093 /* We can have new write after previous check */
4094 offset -= head;
4095 bytes = s->subcluster_size;
4096 nr = s->subcluster_size;
4097 ret = qcow2_get_host_offset(bs, offset, &nr, &off, &type);
4098 if (ret < 0 ||
4099 (type != QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN &&
4100 type != QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC &&
4101 type != QCOW2_SUBCLUSTER_ZERO_PLAIN &&
4102 type != QCOW2_SUBCLUSTER_ZERO_ALLOC)) {
4103 qemu_co_mutex_unlock(&s->lock);
4104 return ret < 0 ? ret : -ENOTSUP;
4105 }
4106 } else {
4107 qemu_co_mutex_lock(&s->lock);
4108 }
4109
4110 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, bytes);
4111
4112 /* Whatever is left can use real zero subclusters */
4113 ret = qcow2_subcluster_zeroize(bs, offset, bytes, flags);
4114 qemu_co_mutex_unlock(&s->lock);
4115
4116 return ret;
4117 }
4118
4119 static int coroutine_fn GRAPH_RDLOCK
qcow2_co_pdiscard(BlockDriverState * bs,int64_t offset,int64_t bytes)4120 qcow2_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes)
4121 {
4122 int ret;
4123 BDRVQcow2State *s = bs->opaque;
4124
4125 /* If the image does not support QCOW_OFLAG_ZERO then discarding
4126 * clusters could expose stale data from the backing file. */
4127 if (s->qcow_version < 3 && bs->backing) {
4128 return -ENOTSUP;
4129 }
4130
4131 if (!QEMU_IS_ALIGNED(offset | bytes, s->cluster_size)) {
4132 assert(bytes < s->cluster_size);
4133 /* Ignore partial clusters, except for the special case of the
4134 * complete partial cluster at the end of an unaligned file */
4135 if (!QEMU_IS_ALIGNED(offset, s->cluster_size) ||
4136 offset + bytes != bs->total_sectors * BDRV_SECTOR_SIZE) {
4137 return -ENOTSUP;
4138 }
4139 }
4140
4141 qemu_co_mutex_lock(&s->lock);
4142 ret = qcow2_cluster_discard(bs, offset, bytes, QCOW2_DISCARD_REQUEST,
4143 false);
4144 qemu_co_mutex_unlock(&s->lock);
4145 return ret;
4146 }
4147
4148 static int coroutine_fn GRAPH_RDLOCK
qcow2_co_copy_range_from(BlockDriverState * bs,BdrvChild * src,int64_t src_offset,BdrvChild * dst,int64_t dst_offset,int64_t bytes,BdrvRequestFlags read_flags,BdrvRequestFlags write_flags)4149 qcow2_co_copy_range_from(BlockDriverState *bs,
4150 BdrvChild *src, int64_t src_offset,
4151 BdrvChild *dst, int64_t dst_offset,
4152 int64_t bytes, BdrvRequestFlags read_flags,
4153 BdrvRequestFlags write_flags)
4154 {
4155 BDRVQcow2State *s = bs->opaque;
4156 int ret;
4157 unsigned int cur_bytes; /* number of bytes in current iteration */
4158 BdrvChild *child = NULL;
4159 BdrvRequestFlags cur_write_flags;
4160
4161 assert(!bs->encrypted);
4162 qemu_co_mutex_lock(&s->lock);
4163
4164 while (bytes != 0) {
4165 uint64_t copy_offset = 0;
4166 QCow2SubclusterType type;
4167 /* prepare next request */
4168 cur_bytes = MIN(bytes, INT_MAX);
4169 cur_write_flags = write_flags;
4170
4171 ret = qcow2_get_host_offset(bs, src_offset, &cur_bytes,
4172 ©_offset, &type);
4173 if (ret < 0) {
4174 goto out;
4175 }
4176
4177 switch (type) {
4178 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN:
4179 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC:
4180 if (bs->backing && bs->backing->bs) {
4181 int64_t backing_length = bdrv_co_getlength(bs->backing->bs);
4182 if (src_offset >= backing_length) {
4183 cur_write_flags |= BDRV_REQ_ZERO_WRITE;
4184 } else {
4185 child = bs->backing;
4186 cur_bytes = MIN(cur_bytes, backing_length - src_offset);
4187 copy_offset = src_offset;
4188 }
4189 } else {
4190 cur_write_flags |= BDRV_REQ_ZERO_WRITE;
4191 }
4192 break;
4193
4194 case QCOW2_SUBCLUSTER_ZERO_PLAIN:
4195 case QCOW2_SUBCLUSTER_ZERO_ALLOC:
4196 cur_write_flags |= BDRV_REQ_ZERO_WRITE;
4197 break;
4198
4199 case QCOW2_SUBCLUSTER_COMPRESSED:
4200 ret = -ENOTSUP;
4201 goto out;
4202
4203 case QCOW2_SUBCLUSTER_NORMAL:
4204 child = s->data_file;
4205 break;
4206
4207 default:
4208 abort();
4209 }
4210 qemu_co_mutex_unlock(&s->lock);
4211 ret = bdrv_co_copy_range_from(child,
4212 copy_offset,
4213 dst, dst_offset,
4214 cur_bytes, read_flags, cur_write_flags);
4215 qemu_co_mutex_lock(&s->lock);
4216 if (ret < 0) {
4217 goto out;
4218 }
4219
4220 bytes -= cur_bytes;
4221 src_offset += cur_bytes;
4222 dst_offset += cur_bytes;
4223 }
4224 ret = 0;
4225
4226 out:
4227 qemu_co_mutex_unlock(&s->lock);
4228 return ret;
4229 }
4230
4231 static int coroutine_fn GRAPH_RDLOCK
qcow2_co_copy_range_to(BlockDriverState * bs,BdrvChild * src,int64_t src_offset,BdrvChild * dst,int64_t dst_offset,int64_t bytes,BdrvRequestFlags read_flags,BdrvRequestFlags write_flags)4232 qcow2_co_copy_range_to(BlockDriverState *bs,
4233 BdrvChild *src, int64_t src_offset,
4234 BdrvChild *dst, int64_t dst_offset,
4235 int64_t bytes, BdrvRequestFlags read_flags,
4236 BdrvRequestFlags write_flags)
4237 {
4238 BDRVQcow2State *s = bs->opaque;
4239 int ret;
4240 unsigned int cur_bytes; /* number of sectors in current iteration */
4241 uint64_t host_offset;
4242 QCowL2Meta *l2meta = NULL;
4243
4244 assert(!bs->encrypted);
4245
4246 qemu_co_mutex_lock(&s->lock);
4247
4248 while (bytes != 0) {
4249
4250 l2meta = NULL;
4251
4252 cur_bytes = MIN(bytes, INT_MAX);
4253
4254 /* TODO:
4255 * If src->bs == dst->bs, we could simply copy by incrementing
4256 * the refcnt, without copying user data.
4257 * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */
4258 ret = qcow2_alloc_host_offset(bs, dst_offset, &cur_bytes,
4259 &host_offset, &l2meta);
4260 if (ret < 0) {
4261 goto fail;
4262 }
4263
4264 ret = qcow2_pre_write_overlap_check(bs, 0, host_offset, cur_bytes,
4265 true);
4266 if (ret < 0) {
4267 goto fail;
4268 }
4269
4270 qemu_co_mutex_unlock(&s->lock);
4271 ret = bdrv_co_copy_range_to(src, src_offset, s->data_file, host_offset,
4272 cur_bytes, read_flags, write_flags);
4273 qemu_co_mutex_lock(&s->lock);
4274 if (ret < 0) {
4275 goto fail;
4276 }
4277
4278 ret = qcow2_handle_l2meta(bs, &l2meta, true);
4279 if (ret) {
4280 goto fail;
4281 }
4282
4283 bytes -= cur_bytes;
4284 src_offset += cur_bytes;
4285 dst_offset += cur_bytes;
4286 }
4287 ret = 0;
4288
4289 fail:
4290 qcow2_handle_l2meta(bs, &l2meta, false);
4291
4292 qemu_co_mutex_unlock(&s->lock);
4293
4294 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
4295
4296 return ret;
4297 }
4298
4299 static int coroutine_fn GRAPH_RDLOCK
qcow2_co_truncate(BlockDriverState * bs,int64_t offset,bool exact,PreallocMode prealloc,BdrvRequestFlags flags,Error ** errp)4300 qcow2_co_truncate(BlockDriverState *bs, int64_t offset, bool exact,
4301 PreallocMode prealloc, BdrvRequestFlags flags, Error **errp)
4302 {
4303 ERRP_GUARD();
4304 BDRVQcow2State *s = bs->opaque;
4305 uint64_t old_length;
4306 int64_t new_l1_size;
4307 int ret;
4308 QDict *options;
4309
4310 if (prealloc != PREALLOC_MODE_OFF && prealloc != PREALLOC_MODE_METADATA &&
4311 prealloc != PREALLOC_MODE_FALLOC && prealloc != PREALLOC_MODE_FULL)
4312 {
4313 error_setg(errp, "Unsupported preallocation mode '%s'",
4314 PreallocMode_str(prealloc));
4315 return -ENOTSUP;
4316 }
4317
4318 if (!QEMU_IS_ALIGNED(offset, BDRV_SECTOR_SIZE)) {
4319 error_setg(errp, "The new size must be a multiple of %u",
4320 (unsigned) BDRV_SECTOR_SIZE);
4321 return -EINVAL;
4322 }
4323
4324 qemu_co_mutex_lock(&s->lock);
4325
4326 /*
4327 * Even though we store snapshot size for all images, it was not
4328 * required until v3, so it is not safe to proceed for v2.
4329 */
4330 if (s->nb_snapshots && s->qcow_version < 3) {
4331 error_setg(errp, "Can't resize a v2 image which has snapshots");
4332 ret = -ENOTSUP;
4333 goto fail;
4334 }
4335
4336 /* See qcow2-bitmap.c for which bitmap scenarios prevent a resize. */
4337 if (qcow2_truncate_bitmaps_check(bs, errp)) {
4338 ret = -ENOTSUP;
4339 goto fail;
4340 }
4341
4342 old_length = bs->total_sectors * BDRV_SECTOR_SIZE;
4343 new_l1_size = size_to_l1(s, offset);
4344
4345 if (offset < old_length) {
4346 int64_t last_cluster, old_file_size;
4347 if (prealloc != PREALLOC_MODE_OFF) {
4348 error_setg(errp,
4349 "Preallocation can't be used for shrinking an image");
4350 ret = -EINVAL;
4351 goto fail;
4352 }
4353
4354 ret = qcow2_cluster_discard(bs, ROUND_UP(offset, s->cluster_size),
4355 old_length - ROUND_UP(offset,
4356 s->cluster_size),
4357 QCOW2_DISCARD_ALWAYS, true);
4358 if (ret < 0) {
4359 error_setg_errno(errp, -ret, "Failed to discard cropped clusters");
4360 goto fail;
4361 }
4362
4363 ret = qcow2_shrink_l1_table(bs, new_l1_size);
4364 if (ret < 0) {
4365 error_setg_errno(errp, -ret,
4366 "Failed to reduce the number of L2 tables");
4367 goto fail;
4368 }
4369
4370 ret = qcow2_shrink_reftable(bs);
4371 if (ret < 0) {
4372 error_setg_errno(errp, -ret,
4373 "Failed to discard unused refblocks");
4374 goto fail;
4375 }
4376
4377 old_file_size = bdrv_co_getlength(bs->file->bs);
4378 if (old_file_size < 0) {
4379 error_setg_errno(errp, -old_file_size,
4380 "Failed to inquire current file length");
4381 ret = old_file_size;
4382 goto fail;
4383 }
4384 last_cluster = qcow2_get_last_cluster(bs, old_file_size);
4385 if (last_cluster < 0) {
4386 error_setg_errno(errp, -last_cluster,
4387 "Failed to find the last cluster");
4388 ret = last_cluster;
4389 goto fail;
4390 }
4391 if ((last_cluster + 1) * s->cluster_size < old_file_size) {
4392 Error *local_err = NULL;
4393
4394 /*
4395 * Do not pass @exact here: It will not help the user if
4396 * we get an error here just because they wanted to shrink
4397 * their qcow2 image (on a block device) with qemu-img.
4398 * (And on the qcow2 layer, the @exact requirement is
4399 * always fulfilled, so there is no need to pass it on.)
4400 */
4401 bdrv_co_truncate(bs->file, (last_cluster + 1) * s->cluster_size,
4402 false, PREALLOC_MODE_OFF, 0, &local_err);
4403 if (local_err) {
4404 warn_reportf_err(local_err,
4405 "Failed to truncate the tail of the image: ");
4406 }
4407 }
4408 } else {
4409 ret = qcow2_grow_l1_table(bs, new_l1_size, true);
4410 if (ret < 0) {
4411 error_setg_errno(errp, -ret, "Failed to grow the L1 table");
4412 goto fail;
4413 }
4414
4415 if (data_file_is_raw(bs) && prealloc == PREALLOC_MODE_OFF) {
4416 /*
4417 * When creating a qcow2 image with data-file-raw, we enforce
4418 * at least prealloc=metadata, so that the L1/L2 tables are
4419 * fully allocated and reading from the data file will return
4420 * the same data as reading from the qcow2 image. When the
4421 * image is grown, we must consequently preallocate the
4422 * metadata structures to cover the added area.
4423 */
4424 prealloc = PREALLOC_MODE_METADATA;
4425 }
4426 }
4427
4428 switch (prealloc) {
4429 case PREALLOC_MODE_OFF:
4430 if (has_data_file(bs)) {
4431 /*
4432 * If the caller wants an exact resize, the external data
4433 * file should be resized to the exact target size, too,
4434 * so we pass @exact here.
4435 */
4436 ret = bdrv_co_truncate(s->data_file, offset, exact, prealloc, 0,
4437 errp);
4438 if (ret < 0) {
4439 goto fail;
4440 }
4441 }
4442 break;
4443
4444 case PREALLOC_MODE_METADATA:
4445 ret = preallocate_co(bs, old_length, offset, prealloc, errp);
4446 if (ret < 0) {
4447 goto fail;
4448 }
4449 break;
4450
4451 case PREALLOC_MODE_FALLOC:
4452 case PREALLOC_MODE_FULL:
4453 {
4454 int64_t allocation_start, host_offset, guest_offset;
4455 int64_t clusters_allocated;
4456 int64_t old_file_size, last_cluster, new_file_size;
4457 uint64_t nb_new_data_clusters, nb_new_l2_tables;
4458 bool subclusters_need_allocation = false;
4459
4460 /* With a data file, preallocation means just allocating the metadata
4461 * and forwarding the truncate request to the data file */
4462 if (has_data_file(bs)) {
4463 ret = preallocate_co(bs, old_length, offset, prealloc, errp);
4464 if (ret < 0) {
4465 goto fail;
4466 }
4467 break;
4468 }
4469
4470 old_file_size = bdrv_co_getlength(bs->file->bs);
4471 if (old_file_size < 0) {
4472 error_setg_errno(errp, -old_file_size,
4473 "Failed to inquire current file length");
4474 ret = old_file_size;
4475 goto fail;
4476 }
4477
4478 last_cluster = qcow2_get_last_cluster(bs, old_file_size);
4479 if (last_cluster >= 0) {
4480 old_file_size = (last_cluster + 1) * s->cluster_size;
4481 } else {
4482 old_file_size = ROUND_UP(old_file_size, s->cluster_size);
4483 }
4484
4485 nb_new_data_clusters = (ROUND_UP(offset, s->cluster_size) -
4486 start_of_cluster(s, old_length)) >> s->cluster_bits;
4487
4488 /* This is an overestimation; we will not actually allocate space for
4489 * these in the file but just make sure the new refcount structures are
4490 * able to cover them so we will not have to allocate new refblocks
4491 * while entering the data blocks in the potentially new L2 tables.
4492 * (We do not actually care where the L2 tables are placed. Maybe they
4493 * are already allocated or they can be placed somewhere before
4494 * @old_file_size. It does not matter because they will be fully
4495 * allocated automatically, so they do not need to be covered by the
4496 * preallocation. All that matters is that we will not have to allocate
4497 * new refcount structures for them.) */
4498 nb_new_l2_tables = DIV_ROUND_UP(nb_new_data_clusters,
4499 s->cluster_size / l2_entry_size(s));
4500 /* The cluster range may not be aligned to L2 boundaries, so add one L2
4501 * table for a potential head/tail */
4502 nb_new_l2_tables++;
4503
4504 allocation_start = qcow2_refcount_area(bs, old_file_size,
4505 nb_new_data_clusters +
4506 nb_new_l2_tables,
4507 true, 0, 0);
4508 if (allocation_start < 0) {
4509 error_setg_errno(errp, -allocation_start,
4510 "Failed to resize refcount structures");
4511 ret = allocation_start;
4512 goto fail;
4513 }
4514
4515 clusters_allocated = qcow2_alloc_clusters_at(bs, allocation_start,
4516 nb_new_data_clusters);
4517 if (clusters_allocated < 0) {
4518 error_setg_errno(errp, -clusters_allocated,
4519 "Failed to allocate data clusters");
4520 ret = clusters_allocated;
4521 goto fail;
4522 }
4523
4524 assert(clusters_allocated == nb_new_data_clusters);
4525
4526 /* Allocate the data area */
4527 new_file_size = allocation_start +
4528 nb_new_data_clusters * s->cluster_size;
4529 /*
4530 * Image file grows, so @exact does not matter.
4531 *
4532 * If we need to zero out the new area, try first whether the protocol
4533 * driver can already take care of this.
4534 */
4535 if (flags & BDRV_REQ_ZERO_WRITE) {
4536 ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc,
4537 BDRV_REQ_ZERO_WRITE, NULL);
4538 if (ret >= 0) {
4539 flags &= ~BDRV_REQ_ZERO_WRITE;
4540 /* Ensure that we read zeroes and not backing file data */
4541 subclusters_need_allocation = true;
4542 }
4543 } else {
4544 ret = -1;
4545 }
4546 if (ret < 0) {
4547 ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc, 0,
4548 errp);
4549 }
4550 if (ret < 0) {
4551 error_prepend(errp, "Failed to resize underlying file: ");
4552 qcow2_free_clusters(bs, allocation_start,
4553 nb_new_data_clusters * s->cluster_size,
4554 QCOW2_DISCARD_OTHER);
4555 goto fail;
4556 }
4557
4558 /* Create the necessary L2 entries */
4559 host_offset = allocation_start;
4560 guest_offset = old_length;
4561 while (nb_new_data_clusters) {
4562 int64_t nb_clusters = MIN(
4563 nb_new_data_clusters,
4564 s->l2_slice_size - offset_to_l2_slice_index(s, guest_offset));
4565 unsigned cow_start_length = offset_into_cluster(s, guest_offset);
4566 QCowL2Meta allocation;
4567 guest_offset = start_of_cluster(s, guest_offset);
4568 allocation = (QCowL2Meta) {
4569 .offset = guest_offset,
4570 .alloc_offset = host_offset,
4571 .nb_clusters = nb_clusters,
4572 .cow_start = {
4573 .offset = 0,
4574 .nb_bytes = cow_start_length,
4575 },
4576 .cow_end = {
4577 .offset = nb_clusters << s->cluster_bits,
4578 .nb_bytes = 0,
4579 },
4580 .prealloc = !subclusters_need_allocation,
4581 };
4582 qemu_co_queue_init(&allocation.dependent_requests);
4583
4584 ret = qcow2_alloc_cluster_link_l2(bs, &allocation);
4585 if (ret < 0) {
4586 error_setg_errno(errp, -ret, "Failed to update L2 tables");
4587 qcow2_free_clusters(bs, host_offset,
4588 nb_new_data_clusters * s->cluster_size,
4589 QCOW2_DISCARD_OTHER);
4590 goto fail;
4591 }
4592
4593 guest_offset += nb_clusters * s->cluster_size;
4594 host_offset += nb_clusters * s->cluster_size;
4595 nb_new_data_clusters -= nb_clusters;
4596 }
4597 break;
4598 }
4599
4600 default:
4601 g_assert_not_reached();
4602 }
4603
4604 if ((flags & BDRV_REQ_ZERO_WRITE) && offset > old_length) {
4605 uint64_t zero_start = QEMU_ALIGN_UP(old_length, s->subcluster_size);
4606
4607 /*
4608 * Use zero clusters as much as we can. qcow2_subcluster_zeroize()
4609 * requires a subcluster-aligned start. The end may be unaligned if
4610 * it is at the end of the image (which it is here).
4611 */
4612 if (offset > zero_start) {
4613 ret = qcow2_subcluster_zeroize(bs, zero_start, offset - zero_start,
4614 0);
4615 if (ret < 0) {
4616 error_setg_errno(errp, -ret, "Failed to zero out new clusters");
4617 goto fail;
4618 }
4619 }
4620
4621 /* Write explicit zeros for the unaligned head */
4622 if (zero_start > old_length) {
4623 uint64_t len = MIN(zero_start, offset) - old_length;
4624 uint8_t *buf = qemu_blockalign0(bs, len);
4625 QEMUIOVector qiov;
4626 qemu_iovec_init_buf(&qiov, buf, len);
4627
4628 qemu_co_mutex_unlock(&s->lock);
4629 ret = qcow2_co_pwritev_part(bs, old_length, len, &qiov, 0, 0);
4630 qemu_co_mutex_lock(&s->lock);
4631
4632 qemu_vfree(buf);
4633 if (ret < 0) {
4634 error_setg_errno(errp, -ret, "Failed to zero out the new area");
4635 goto fail;
4636 }
4637 }
4638 }
4639
4640 if (prealloc != PREALLOC_MODE_OFF) {
4641 /* Flush metadata before actually changing the image size */
4642 ret = qcow2_write_caches(bs);
4643 if (ret < 0) {
4644 error_setg_errno(errp, -ret,
4645 "Failed to flush the preallocated area to disk");
4646 goto fail;
4647 }
4648 }
4649
4650 bs->total_sectors = offset / BDRV_SECTOR_SIZE;
4651
4652 /* write updated header.size */
4653 offset = cpu_to_be64(offset);
4654 ret = bdrv_co_pwrite_sync(bs->file, offsetof(QCowHeader, size),
4655 sizeof(offset), &offset, 0);
4656 if (ret < 0) {
4657 error_setg_errno(errp, -ret, "Failed to update the image size");
4658 goto fail;
4659 }
4660
4661 s->l1_vm_state_index = new_l1_size;
4662
4663 /* Update cache sizes */
4664 options = qdict_clone_shallow(bs->options);
4665 ret = qcow2_update_options(bs, options, s->flags, errp);
4666 qobject_unref(options);
4667 if (ret < 0) {
4668 goto fail;
4669 }
4670 ret = 0;
4671 fail:
4672 qemu_co_mutex_unlock(&s->lock);
4673 return ret;
4674 }
4675
4676 static int coroutine_fn GRAPH_RDLOCK
qcow2_co_pwritev_compressed_task(BlockDriverState * bs,uint64_t offset,uint64_t bytes,QEMUIOVector * qiov,size_t qiov_offset)4677 qcow2_co_pwritev_compressed_task(BlockDriverState *bs,
4678 uint64_t offset, uint64_t bytes,
4679 QEMUIOVector *qiov, size_t qiov_offset)
4680 {
4681 BDRVQcow2State *s = bs->opaque;
4682 int ret;
4683 ssize_t out_len;
4684 uint8_t *buf, *out_buf;
4685 uint64_t cluster_offset;
4686
4687 assert(bytes == s->cluster_size || (bytes < s->cluster_size &&
4688 (offset + bytes == bs->total_sectors << BDRV_SECTOR_BITS)));
4689
4690 buf = qemu_blockalign(bs, s->cluster_size);
4691 if (bytes < s->cluster_size) {
4692 /* Zero-pad last write if image size is not cluster aligned */
4693 memset(buf + bytes, 0, s->cluster_size - bytes);
4694 }
4695 qemu_iovec_to_buf(qiov, qiov_offset, buf, bytes);
4696
4697 out_buf = g_malloc(s->cluster_size);
4698
4699 out_len = qcow2_co_compress(bs, out_buf, s->cluster_size - 1,
4700 buf, s->cluster_size);
4701 if (out_len == -ENOMEM) {
4702 /* could not compress: write normal cluster */
4703 ret = qcow2_co_pwritev_part(bs, offset, bytes, qiov, qiov_offset, 0);
4704 if (ret < 0) {
4705 goto fail;
4706 }
4707 goto success;
4708 } else if (out_len < 0) {
4709 ret = -EINVAL;
4710 goto fail;
4711 }
4712
4713 qemu_co_mutex_lock(&s->lock);
4714 ret = qcow2_alloc_compressed_cluster_offset(bs, offset, out_len,
4715 &cluster_offset);
4716 if (ret < 0) {
4717 qemu_co_mutex_unlock(&s->lock);
4718 goto fail;
4719 }
4720
4721 ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len, true);
4722 qemu_co_mutex_unlock(&s->lock);
4723 if (ret < 0) {
4724 goto fail;
4725 }
4726
4727 BLKDBG_CO_EVENT(s->data_file, BLKDBG_WRITE_COMPRESSED);
4728 ret = bdrv_co_pwrite(s->data_file, cluster_offset, out_len, out_buf, 0);
4729 if (ret < 0) {
4730 goto fail;
4731 }
4732 success:
4733 ret = 0;
4734 fail:
4735 qemu_vfree(buf);
4736 g_free(out_buf);
4737 return ret;
4738 }
4739
4740 /*
4741 * This function can count as GRAPH_RDLOCK because
4742 * qcow2_co_pwritev_compressed_part() holds the graph lock and keeps it until
4743 * this coroutine has terminated.
4744 */
4745 static int coroutine_fn GRAPH_RDLOCK
qcow2_co_pwritev_compressed_task_entry(AioTask * task)4746 qcow2_co_pwritev_compressed_task_entry(AioTask *task)
4747 {
4748 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
4749
4750 assert(!t->subcluster_type && !t->l2meta);
4751
4752 return qcow2_co_pwritev_compressed_task(t->bs, t->offset, t->bytes, t->qiov,
4753 t->qiov_offset);
4754 }
4755
4756 /*
4757 * XXX: put compressed sectors first, then all the cluster aligned
4758 * tables to avoid losing bytes in alignment
4759 */
4760 static int coroutine_fn GRAPH_RDLOCK
qcow2_co_pwritev_compressed_part(BlockDriverState * bs,int64_t offset,int64_t bytes,QEMUIOVector * qiov,size_t qiov_offset)4761 qcow2_co_pwritev_compressed_part(BlockDriverState *bs,
4762 int64_t offset, int64_t bytes,
4763 QEMUIOVector *qiov, size_t qiov_offset)
4764 {
4765 BDRVQcow2State *s = bs->opaque;
4766 AioTaskPool *aio = NULL;
4767 int ret = 0;
4768
4769 if (has_data_file(bs)) {
4770 return -ENOTSUP;
4771 }
4772
4773 if (bytes == 0) {
4774 /*
4775 * align end of file to a sector boundary to ease reading with
4776 * sector based I/Os
4777 */
4778 int64_t len = bdrv_co_getlength(bs->file->bs);
4779 if (len < 0) {
4780 return len;
4781 }
4782 return bdrv_co_truncate(bs->file, len, false, PREALLOC_MODE_OFF, 0,
4783 NULL);
4784 }
4785
4786 if (offset_into_cluster(s, offset)) {
4787 return -EINVAL;
4788 }
4789
4790 if (offset_into_cluster(s, bytes) &&
4791 (offset + bytes) != (bs->total_sectors << BDRV_SECTOR_BITS)) {
4792 return -EINVAL;
4793 }
4794
4795 while (bytes && aio_task_pool_status(aio) == 0) {
4796 uint64_t chunk_size = MIN(bytes, s->cluster_size);
4797
4798 if (!aio && chunk_size != bytes) {
4799 aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
4800 }
4801
4802 ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_compressed_task_entry,
4803 0, 0, offset, chunk_size, qiov, qiov_offset, NULL);
4804 if (ret < 0) {
4805 break;
4806 }
4807 qiov_offset += chunk_size;
4808 offset += chunk_size;
4809 bytes -= chunk_size;
4810 }
4811
4812 if (aio) {
4813 aio_task_pool_wait_all(aio);
4814 if (ret == 0) {
4815 ret = aio_task_pool_status(aio);
4816 }
4817 g_free(aio);
4818 }
4819
4820 return ret;
4821 }
4822
4823 static int coroutine_fn GRAPH_RDLOCK
qcow2_co_preadv_compressed(BlockDriverState * bs,uint64_t l2_entry,uint64_t offset,uint64_t bytes,QEMUIOVector * qiov,size_t qiov_offset)4824 qcow2_co_preadv_compressed(BlockDriverState *bs,
4825 uint64_t l2_entry,
4826 uint64_t offset,
4827 uint64_t bytes,
4828 QEMUIOVector *qiov,
4829 size_t qiov_offset)
4830 {
4831 BDRVQcow2State *s = bs->opaque;
4832 int ret = 0, csize;
4833 uint64_t coffset;
4834 uint8_t *buf, *out_buf;
4835 int offset_in_cluster = offset_into_cluster(s, offset);
4836
4837 qcow2_parse_compressed_l2_entry(bs, l2_entry, &coffset, &csize);
4838
4839 buf = g_try_malloc(csize);
4840 if (!buf) {
4841 return -ENOMEM;
4842 }
4843
4844 out_buf = qemu_blockalign(bs, s->cluster_size);
4845
4846 BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
4847 ret = bdrv_co_pread(bs->file, coffset, csize, buf, 0);
4848 if (ret < 0) {
4849 goto fail;
4850 }
4851
4852 if (qcow2_co_decompress(bs, out_buf, s->cluster_size, buf, csize) < 0) {
4853 ret = -EIO;
4854 goto fail;
4855 }
4856
4857 qemu_iovec_from_buf(qiov, qiov_offset, out_buf + offset_in_cluster, bytes);
4858
4859 fail:
4860 qemu_vfree(out_buf);
4861 g_free(buf);
4862
4863 return ret;
4864 }
4865
make_completely_empty(BlockDriverState * bs)4866 static int GRAPH_RDLOCK make_completely_empty(BlockDriverState *bs)
4867 {
4868 BDRVQcow2State *s = bs->opaque;
4869 Error *local_err = NULL;
4870 int ret, l1_clusters;
4871 int64_t offset;
4872 uint64_t *new_reftable = NULL;
4873 uint64_t rt_entry, l1_size2;
4874 struct {
4875 uint64_t l1_offset;
4876 uint64_t reftable_offset;
4877 uint32_t reftable_clusters;
4878 } QEMU_PACKED l1_ofs_rt_ofs_cls;
4879
4880 ret = qcow2_cache_empty(bs, s->l2_table_cache);
4881 if (ret < 0) {
4882 goto fail;
4883 }
4884
4885 ret = qcow2_cache_empty(bs, s->refcount_block_cache);
4886 if (ret < 0) {
4887 goto fail;
4888 }
4889
4890 /* Refcounts will be broken utterly */
4891 ret = qcow2_mark_dirty(bs);
4892 if (ret < 0) {
4893 goto fail;
4894 }
4895
4896 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
4897
4898 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / L1E_SIZE);
4899 l1_size2 = (uint64_t)s->l1_size * L1E_SIZE;
4900
4901 /* After this call, neither the in-memory nor the on-disk refcount
4902 * information accurately describe the actual references */
4903
4904 ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset,
4905 l1_clusters * s->cluster_size, 0);
4906 if (ret < 0) {
4907 goto fail_broken_refcounts;
4908 }
4909 memset(s->l1_table, 0, l1_size2);
4910
4911 BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE);
4912
4913 /* Overwrite enough clusters at the beginning of the sectors to place
4914 * the refcount table, a refcount block and the L1 table in; this may
4915 * overwrite parts of the existing refcount and L1 table, which is not
4916 * an issue because the dirty flag is set, complete data loss is in fact
4917 * desired and partial data loss is consequently fine as well */
4918 ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size,
4919 (2 + l1_clusters) * s->cluster_size, 0);
4920 /* This call (even if it failed overall) may have overwritten on-disk
4921 * refcount structures; in that case, the in-memory refcount information
4922 * will probably differ from the on-disk information which makes the BDS
4923 * unusable */
4924 if (ret < 0) {
4925 goto fail_broken_refcounts;
4926 }
4927
4928 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
4929 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
4930
4931 /* "Create" an empty reftable (one cluster) directly after the image
4932 * header and an empty L1 table three clusters after the image header;
4933 * the cluster between those two will be used as the first refblock */
4934 l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size);
4935 l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size);
4936 l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1);
4937 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset),
4938 sizeof(l1_ofs_rt_ofs_cls), &l1_ofs_rt_ofs_cls, 0);
4939 if (ret < 0) {
4940 goto fail_broken_refcounts;
4941 }
4942
4943 s->l1_table_offset = 3 * s->cluster_size;
4944
4945 new_reftable = g_try_new0(uint64_t, s->cluster_size / REFTABLE_ENTRY_SIZE);
4946 if (!new_reftable) {
4947 ret = -ENOMEM;
4948 goto fail_broken_refcounts;
4949 }
4950
4951 s->refcount_table_offset = s->cluster_size;
4952 s->refcount_table_size = s->cluster_size / REFTABLE_ENTRY_SIZE;
4953 s->max_refcount_table_index = 0;
4954
4955 g_free(s->refcount_table);
4956 s->refcount_table = new_reftable;
4957 new_reftable = NULL;
4958
4959 /* Now the in-memory refcount information again corresponds to the on-disk
4960 * information (reftable is empty and no refblocks (the refblock cache is
4961 * empty)); however, this means some clusters (e.g. the image header) are
4962 * referenced, but not refcounted, but the normal qcow2 code assumes that
4963 * the in-memory information is always correct */
4964
4965 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
4966
4967 /* Enter the first refblock into the reftable */
4968 rt_entry = cpu_to_be64(2 * s->cluster_size);
4969 ret = bdrv_pwrite_sync(bs->file, s->cluster_size, sizeof(rt_entry),
4970 &rt_entry, 0);
4971 if (ret < 0) {
4972 goto fail_broken_refcounts;
4973 }
4974 s->refcount_table[0] = 2 * s->cluster_size;
4975
4976 s->free_cluster_index = 0;
4977 assert(3 + l1_clusters <= s->refcount_block_size);
4978 offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2);
4979 if (offset < 0) {
4980 ret = offset;
4981 goto fail_broken_refcounts;
4982 } else if (offset > 0) {
4983 error_report("First cluster in emptied image is in use");
4984 abort();
4985 }
4986
4987 /* Now finally the in-memory information corresponds to the on-disk
4988 * structures and is correct */
4989 ret = qcow2_mark_clean(bs);
4990 if (ret < 0) {
4991 goto fail;
4992 }
4993
4994 ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size, false,
4995 PREALLOC_MODE_OFF, 0, &local_err);
4996 if (ret < 0) {
4997 error_report_err(local_err);
4998 goto fail;
4999 }
5000
5001 return 0;
5002
5003 fail_broken_refcounts:
5004 /* The BDS is unusable at this point. If we wanted to make it usable, we
5005 * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
5006 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
5007 * again. However, because the functions which could have caused this error
5008 * path to be taken are used by those functions as well, it's very likely
5009 * that that sequence will fail as well. Therefore, just eject the BDS. */
5010 bs->drv = NULL;
5011
5012 fail:
5013 g_free(new_reftable);
5014 return ret;
5015 }
5016
qcow2_make_empty(BlockDriverState * bs)5017 static int GRAPH_RDLOCK qcow2_make_empty(BlockDriverState *bs)
5018 {
5019 BDRVQcow2State *s = bs->opaque;
5020 uint64_t offset, end_offset;
5021 int step = QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size);
5022 int l1_clusters, ret = 0;
5023
5024 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / L1E_SIZE);
5025
5026 if (s->qcow_version >= 3 && !s->snapshots && !s->nb_bitmaps &&
5027 3 + l1_clusters <= s->refcount_block_size &&
5028 s->crypt_method_header != QCOW_CRYPT_LUKS &&
5029 !has_data_file(bs)) {
5030 /* The following function only works for qcow2 v3 images (it
5031 * requires the dirty flag) and only as long as there are no
5032 * features that reserve extra clusters (such as snapshots,
5033 * LUKS header, or persistent bitmaps), because it completely
5034 * empties the image. Furthermore, the L1 table and three
5035 * additional clusters (image header, refcount table, one
5036 * refcount block) have to fit inside one refcount block. It
5037 * only resets the image file, i.e. does not work with an
5038 * external data file. */
5039 return make_completely_empty(bs);
5040 }
5041
5042 /* This fallback code simply discards every active cluster; this is slow,
5043 * but works in all cases */
5044 end_offset = bs->total_sectors * BDRV_SECTOR_SIZE;
5045 for (offset = 0; offset < end_offset; offset += step) {
5046 /* As this function is generally used after committing an external
5047 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
5048 * default action for this kind of discard is to pass the discard,
5049 * which will ideally result in an actually smaller image file, as
5050 * is probably desired. */
5051 ret = qcow2_cluster_discard(bs, offset, MIN(step, end_offset - offset),
5052 QCOW2_DISCARD_SNAPSHOT, true);
5053 if (ret < 0) {
5054 break;
5055 }
5056 }
5057
5058 return ret;
5059 }
5060
qcow2_co_flush_to_os(BlockDriverState * bs)5061 static coroutine_fn GRAPH_RDLOCK int qcow2_co_flush_to_os(BlockDriverState *bs)
5062 {
5063 BDRVQcow2State *s = bs->opaque;
5064 int ret;
5065
5066 qemu_co_mutex_lock(&s->lock);
5067 ret = qcow2_write_caches(bs);
5068 qemu_co_mutex_unlock(&s->lock);
5069
5070 return ret;
5071 }
5072
qcow2_measure(QemuOpts * opts,BlockDriverState * in_bs,Error ** errp)5073 static BlockMeasureInfo *qcow2_measure(QemuOpts *opts, BlockDriverState *in_bs,
5074 Error **errp)
5075 {
5076 Error *local_err = NULL;
5077 BlockMeasureInfo *info;
5078 uint64_t required = 0; /* bytes that contribute to required size */
5079 uint64_t virtual_size; /* disk size as seen by guest */
5080 uint64_t refcount_bits;
5081 uint64_t l2_tables;
5082 uint64_t luks_payload_size = 0;
5083 size_t cluster_size;
5084 int version;
5085 char *optstr;
5086 PreallocMode prealloc;
5087 bool has_backing_file;
5088 bool has_luks;
5089 bool extended_l2;
5090 size_t l2e_size;
5091
5092 /* Parse image creation options */
5093 extended_l2 = qemu_opt_get_bool_del(opts, BLOCK_OPT_EXTL2, false);
5094
5095 cluster_size = qcow2_opt_get_cluster_size_del(opts, extended_l2,
5096 &local_err);
5097 if (local_err) {
5098 goto err;
5099 }
5100
5101 version = qcow2_opt_get_version_del(opts, &local_err);
5102 if (local_err) {
5103 goto err;
5104 }
5105
5106 refcount_bits = qcow2_opt_get_refcount_bits_del(opts, version, &local_err);
5107 if (local_err) {
5108 goto err;
5109 }
5110
5111 optstr = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
5112 prealloc = qapi_enum_parse(&PreallocMode_lookup, optstr,
5113 PREALLOC_MODE_OFF, &local_err);
5114 g_free(optstr);
5115 if (local_err) {
5116 goto err;
5117 }
5118
5119 optstr = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
5120 has_backing_file = !!optstr;
5121 g_free(optstr);
5122
5123 optstr = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT);
5124 has_luks = optstr && strcmp(optstr, "luks") == 0;
5125 g_free(optstr);
5126
5127 if (has_luks) {
5128 g_autoptr(QCryptoBlockCreateOptions) create_opts = NULL;
5129 QDict *cryptoopts = qcow2_extract_crypto_opts(opts, "luks", errp);
5130 size_t headerlen;
5131
5132 create_opts = block_crypto_create_opts_init(cryptoopts, errp);
5133 qobject_unref(cryptoopts);
5134 if (!create_opts) {
5135 goto err;
5136 }
5137
5138 if (!qcrypto_block_calculate_payload_offset(create_opts,
5139 "encrypt.",
5140 &headerlen,
5141 &local_err)) {
5142 goto err;
5143 }
5144
5145 luks_payload_size = ROUND_UP(headerlen, cluster_size);
5146 }
5147
5148 virtual_size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
5149 virtual_size = ROUND_UP(virtual_size, cluster_size);
5150
5151 /* Check that virtual disk size is valid */
5152 l2e_size = extended_l2 ? L2E_SIZE_EXTENDED : L2E_SIZE_NORMAL;
5153 l2_tables = DIV_ROUND_UP(virtual_size / cluster_size,
5154 cluster_size / l2e_size);
5155 if (l2_tables * L1E_SIZE > QCOW_MAX_L1_SIZE) {
5156 error_setg(&local_err, "The image size is too large "
5157 "(try using a larger cluster size)");
5158 goto err;
5159 }
5160
5161 /* Account for input image */
5162 if (in_bs) {
5163 int64_t ssize = bdrv_getlength(in_bs);
5164 if (ssize < 0) {
5165 error_setg_errno(&local_err, -ssize,
5166 "Unable to get image virtual_size");
5167 goto err;
5168 }
5169
5170 virtual_size = ROUND_UP(ssize, cluster_size);
5171
5172 if (has_backing_file) {
5173 /* We don't how much of the backing chain is shared by the input
5174 * image and the new image file. In the worst case the new image's
5175 * backing file has nothing in common with the input image. Be
5176 * conservative and assume all clusters need to be written.
5177 */
5178 required = virtual_size;
5179 } else {
5180 int64_t offset;
5181 int64_t pnum = 0;
5182
5183 for (offset = 0; offset < ssize; offset += pnum) {
5184 int ret;
5185
5186 ret = bdrv_block_status_above(in_bs, NULL, offset,
5187 ssize - offset, &pnum, NULL,
5188 NULL);
5189 if (ret < 0) {
5190 error_setg_errno(&local_err, -ret,
5191 "Unable to get block status");
5192 goto err;
5193 }
5194
5195 if (ret & BDRV_BLOCK_ZERO) {
5196 /* Skip zero regions (safe with no backing file) */
5197 } else if ((ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) ==
5198 (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) {
5199 /* Extend pnum to end of cluster for next iteration */
5200 pnum = ROUND_UP(offset + pnum, cluster_size) - offset;
5201
5202 /* Count clusters we've seen */
5203 required += offset % cluster_size + pnum;
5204 }
5205 }
5206 }
5207 }
5208
5209 /* Take into account preallocation. Nothing special is needed for
5210 * PREALLOC_MODE_METADATA since metadata is always counted.
5211 */
5212 if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
5213 required = virtual_size;
5214 }
5215
5216 info = g_new0(BlockMeasureInfo, 1);
5217 info->fully_allocated = luks_payload_size +
5218 qcow2_calc_prealloc_size(virtual_size, cluster_size,
5219 ctz32(refcount_bits), extended_l2);
5220
5221 /*
5222 * Remove data clusters that are not required. This overestimates the
5223 * required size because metadata needed for the fully allocated file is
5224 * still counted. Show bitmaps only if both source and destination
5225 * would support them.
5226 */
5227 info->required = info->fully_allocated - virtual_size + required;
5228 info->has_bitmaps = version >= 3 && in_bs &&
5229 bdrv_supports_persistent_dirty_bitmap(in_bs);
5230 if (info->has_bitmaps) {
5231 info->bitmaps = qcow2_get_persistent_dirty_bitmap_size(in_bs,
5232 cluster_size);
5233 }
5234 return info;
5235
5236 err:
5237 error_propagate(errp, local_err);
5238 return NULL;
5239 }
5240
5241 static int coroutine_fn
qcow2_co_get_info(BlockDriverState * bs,BlockDriverInfo * bdi)5242 qcow2_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
5243 {
5244 BDRVQcow2State *s = bs->opaque;
5245 bdi->cluster_size = s->cluster_size;
5246 bdi->subcluster_size = s->subcluster_size;
5247 bdi->vm_state_offset = qcow2_vm_state_offset(s);
5248 bdi->is_dirty = s->incompatible_features & QCOW2_INCOMPAT_DIRTY;
5249 return 0;
5250 }
5251
5252 static ImageInfoSpecific * GRAPH_RDLOCK
qcow2_get_specific_info(BlockDriverState * bs,Error ** errp)5253 qcow2_get_specific_info(BlockDriverState *bs, Error **errp)
5254 {
5255 BDRVQcow2State *s = bs->opaque;
5256 ImageInfoSpecific *spec_info;
5257 QCryptoBlockInfo *encrypt_info = NULL;
5258
5259 if (s->crypto != NULL) {
5260 encrypt_info = qcrypto_block_get_info(s->crypto, errp);
5261 if (!encrypt_info) {
5262 return NULL;
5263 }
5264 }
5265
5266 spec_info = g_new(ImageInfoSpecific, 1);
5267 *spec_info = (ImageInfoSpecific){
5268 .type = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
5269 .u.qcow2.data = g_new0(ImageInfoSpecificQCow2, 1),
5270 };
5271 if (s->qcow_version == 2) {
5272 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
5273 .compat = g_strdup("0.10"),
5274 .refcount_bits = s->refcount_bits,
5275 };
5276 } else if (s->qcow_version == 3) {
5277 Qcow2BitmapInfoList *bitmaps;
5278 if (!qcow2_get_bitmap_info_list(bs, &bitmaps, errp)) {
5279 qapi_free_ImageInfoSpecific(spec_info);
5280 qapi_free_QCryptoBlockInfo(encrypt_info);
5281 return NULL;
5282 }
5283 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
5284 .compat = g_strdup("1.1"),
5285 .lazy_refcounts = s->compatible_features &
5286 QCOW2_COMPAT_LAZY_REFCOUNTS,
5287 .has_lazy_refcounts = true,
5288 .corrupt = s->incompatible_features &
5289 QCOW2_INCOMPAT_CORRUPT,
5290 .has_corrupt = true,
5291 .has_extended_l2 = true,
5292 .extended_l2 = has_subclusters(s),
5293 .refcount_bits = s->refcount_bits,
5294 .has_bitmaps = !!bitmaps,
5295 .bitmaps = bitmaps,
5296 .data_file = g_strdup(s->image_data_file),
5297 .has_data_file_raw = has_data_file(bs),
5298 .data_file_raw = data_file_is_raw(bs),
5299 .compression_type = s->compression_type,
5300 };
5301 } else {
5302 /* if this assertion fails, this probably means a new version was
5303 * added without having it covered here */
5304 g_assert_not_reached();
5305 }
5306
5307 if (encrypt_info) {
5308 ImageInfoSpecificQCow2Encryption *qencrypt =
5309 g_new(ImageInfoSpecificQCow2Encryption, 1);
5310 switch (encrypt_info->format) {
5311 case QCRYPTO_BLOCK_FORMAT_QCOW:
5312 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES;
5313 break;
5314 case QCRYPTO_BLOCK_FORMAT_LUKS:
5315 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS;
5316 qencrypt->u.luks = encrypt_info->u.luks;
5317 break;
5318 default:
5319 abort();
5320 }
5321 /* Since we did shallow copy above, erase any pointers
5322 * in the original info */
5323 memset(&encrypt_info->u, 0, sizeof(encrypt_info->u));
5324 qapi_free_QCryptoBlockInfo(encrypt_info);
5325
5326 spec_info->u.qcow2.data->encrypt = qencrypt;
5327 }
5328
5329 return spec_info;
5330 }
5331
5332 static int coroutine_mixed_fn GRAPH_RDLOCK
qcow2_has_zero_init(BlockDriverState * bs)5333 qcow2_has_zero_init(BlockDriverState *bs)
5334 {
5335 BDRVQcow2State *s = bs->opaque;
5336 bool preallocated;
5337
5338 if (qemu_in_coroutine()) {
5339 qemu_co_mutex_lock(&s->lock);
5340 }
5341 /*
5342 * Check preallocation status: Preallocated images have all L2
5343 * tables allocated, nonpreallocated images have none. It is
5344 * therefore enough to check the first one.
5345 */
5346 preallocated = s->l1_size > 0 && s->l1_table[0] != 0;
5347 if (qemu_in_coroutine()) {
5348 qemu_co_mutex_unlock(&s->lock);
5349 }
5350
5351 if (!preallocated) {
5352 return 1;
5353 } else if (bs->encrypted) {
5354 return 0;
5355 } else {
5356 return bdrv_has_zero_init(s->data_file->bs);
5357 }
5358 }
5359
5360 /*
5361 * Check the request to vmstate. On success return
5362 * qcow2_vm_state_offset(bs) + @pos
5363 */
qcow2_check_vmstate_request(BlockDriverState * bs,QEMUIOVector * qiov,int64_t pos)5364 static int64_t qcow2_check_vmstate_request(BlockDriverState *bs,
5365 QEMUIOVector *qiov, int64_t pos)
5366 {
5367 BDRVQcow2State *s = bs->opaque;
5368 int64_t vmstate_offset = qcow2_vm_state_offset(s);
5369 int ret;
5370
5371 /* Incoming requests must be OK */
5372 bdrv_check_qiov_request(pos, qiov->size, qiov, 0, &error_abort);
5373
5374 if (INT64_MAX - pos < vmstate_offset) {
5375 return -EIO;
5376 }
5377
5378 pos += vmstate_offset;
5379 ret = bdrv_check_qiov_request(pos, qiov->size, qiov, 0, NULL);
5380 if (ret < 0) {
5381 return ret;
5382 }
5383
5384 return pos;
5385 }
5386
5387 static int coroutine_fn GRAPH_RDLOCK
qcow2_co_save_vmstate(BlockDriverState * bs,QEMUIOVector * qiov,int64_t pos)5388 qcow2_co_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
5389 {
5390 int64_t offset = qcow2_check_vmstate_request(bs, qiov, pos);
5391 if (offset < 0) {
5392 return offset;
5393 }
5394
5395 BLKDBG_CO_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
5396 return bs->drv->bdrv_co_pwritev_part(bs, offset, qiov->size, qiov, 0, 0);
5397 }
5398
5399 static int coroutine_fn GRAPH_RDLOCK
qcow2_co_load_vmstate(BlockDriverState * bs,QEMUIOVector * qiov,int64_t pos)5400 qcow2_co_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
5401 {
5402 int64_t offset = qcow2_check_vmstate_request(bs, qiov, pos);
5403 if (offset < 0) {
5404 return offset;
5405 }
5406
5407 BLKDBG_CO_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
5408 return bs->drv->bdrv_co_preadv_part(bs, offset, qiov->size, qiov, 0, 0);
5409 }
5410
qcow2_has_compressed_clusters(BlockDriverState * bs)5411 static int GRAPH_RDLOCK qcow2_has_compressed_clusters(BlockDriverState *bs)
5412 {
5413 int64_t offset = 0;
5414 int64_t bytes = bdrv_getlength(bs);
5415
5416 if (bytes < 0) {
5417 return bytes;
5418 }
5419
5420 while (bytes != 0) {
5421 int ret;
5422 QCow2SubclusterType type;
5423 unsigned int cur_bytes = MIN(INT_MAX, bytes);
5424 uint64_t host_offset;
5425
5426 ret = qcow2_get_host_offset(bs, offset, &cur_bytes, &host_offset,
5427 &type);
5428 if (ret < 0) {
5429 return ret;
5430 }
5431
5432 if (type == QCOW2_SUBCLUSTER_COMPRESSED) {
5433 return 1;
5434 }
5435
5436 offset += cur_bytes;
5437 bytes -= cur_bytes;
5438 }
5439
5440 return 0;
5441 }
5442
5443 /*
5444 * Downgrades an image's version. To achieve this, any incompatible features
5445 * have to be removed.
5446 */
5447 static int GRAPH_RDLOCK
qcow2_downgrade(BlockDriverState * bs,int target_version,BlockDriverAmendStatusCB * status_cb,void * cb_opaque,Error ** errp)5448 qcow2_downgrade(BlockDriverState *bs, int target_version,
5449 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
5450 Error **errp)
5451 {
5452 BDRVQcow2State *s = bs->opaque;
5453 int current_version = s->qcow_version;
5454 int ret;
5455 int i;
5456
5457 /* This is qcow2_downgrade(), not qcow2_upgrade() */
5458 assert(target_version < current_version);
5459
5460 /* There are no other versions (now) that you can downgrade to */
5461 assert(target_version == 2);
5462
5463 if (s->refcount_order != 4) {
5464 error_setg(errp, "compat=0.10 requires refcount_bits=16");
5465 return -ENOTSUP;
5466 }
5467
5468 if (has_data_file(bs)) {
5469 error_setg(errp, "Cannot downgrade an image with a data file");
5470 return -ENOTSUP;
5471 }
5472
5473 /*
5474 * If any internal snapshot has a different size than the current
5475 * image size, or VM state size that exceeds 32 bits, downgrading
5476 * is unsafe. Even though we would still use v3-compliant output
5477 * to preserve that data, other v2 programs might not realize
5478 * those optional fields are important.
5479 */
5480 for (i = 0; i < s->nb_snapshots; i++) {
5481 if (s->snapshots[i].vm_state_size > UINT32_MAX ||
5482 s->snapshots[i].disk_size != bs->total_sectors * BDRV_SECTOR_SIZE) {
5483 error_setg(errp, "Internal snapshots prevent downgrade of image");
5484 return -ENOTSUP;
5485 }
5486 }
5487
5488 /* clear incompatible features */
5489 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
5490 ret = qcow2_mark_clean(bs);
5491 if (ret < 0) {
5492 error_setg_errno(errp, -ret, "Failed to make the image clean");
5493 return ret;
5494 }
5495 }
5496
5497 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
5498 * the first place; if that happens nonetheless, returning -ENOTSUP is the
5499 * best thing to do anyway */
5500
5501 if (s->incompatible_features & ~QCOW2_INCOMPAT_COMPRESSION) {
5502 error_setg(errp, "Cannot downgrade an image with incompatible features "
5503 "0x%" PRIx64 " set",
5504 s->incompatible_features & ~QCOW2_INCOMPAT_COMPRESSION);
5505 return -ENOTSUP;
5506 }
5507
5508 /* since we can ignore compatible features, we can set them to 0 as well */
5509 s->compatible_features = 0;
5510 /* if lazy refcounts have been used, they have already been fixed through
5511 * clearing the dirty flag */
5512
5513 /* clearing autoclear features is trivial */
5514 s->autoclear_features = 0;
5515
5516 ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque);
5517 if (ret < 0) {
5518 error_setg_errno(errp, -ret, "Failed to turn zero into data clusters");
5519 return ret;
5520 }
5521
5522 if (s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION) {
5523 ret = qcow2_has_compressed_clusters(bs);
5524 if (ret < 0) {
5525 error_setg(errp, "Failed to check block status");
5526 return -EINVAL;
5527 }
5528 if (ret) {
5529 error_setg(errp, "Cannot downgrade an image with zstd compression "
5530 "type and existing compressed clusters");
5531 return -ENOTSUP;
5532 }
5533 /*
5534 * No compressed clusters for now, so just chose default zlib
5535 * compression.
5536 */
5537 s->incompatible_features &= ~QCOW2_INCOMPAT_COMPRESSION;
5538 s->compression_type = QCOW2_COMPRESSION_TYPE_ZLIB;
5539 }
5540
5541 assert(s->incompatible_features == 0);
5542
5543 s->qcow_version = target_version;
5544 ret = qcow2_update_header(bs);
5545 if (ret < 0) {
5546 s->qcow_version = current_version;
5547 error_setg_errno(errp, -ret, "Failed to update the image header");
5548 return ret;
5549 }
5550 return 0;
5551 }
5552
5553 /*
5554 * Upgrades an image's version. While newer versions encompass all
5555 * features of older versions, some things may have to be presented
5556 * differently.
5557 */
5558 static int GRAPH_RDLOCK
qcow2_upgrade(BlockDriverState * bs,int target_version,BlockDriverAmendStatusCB * status_cb,void * cb_opaque,Error ** errp)5559 qcow2_upgrade(BlockDriverState *bs, int target_version,
5560 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
5561 Error **errp)
5562 {
5563 BDRVQcow2State *s = bs->opaque;
5564 bool need_snapshot_update;
5565 int current_version = s->qcow_version;
5566 int i;
5567 int ret;
5568
5569 /* This is qcow2_upgrade(), not qcow2_downgrade() */
5570 assert(target_version > current_version);
5571
5572 /* There are no other versions (yet) that you can upgrade to */
5573 assert(target_version == 3);
5574
5575 status_cb(bs, 0, 2, cb_opaque);
5576
5577 /*
5578 * In v2, snapshots do not need to have extra data. v3 requires
5579 * the 64-bit VM state size and the virtual disk size to be
5580 * present.
5581 * qcow2_write_snapshots() will always write the list in the
5582 * v3-compliant format.
5583 */
5584 need_snapshot_update = false;
5585 for (i = 0; i < s->nb_snapshots; i++) {
5586 if (s->snapshots[i].extra_data_size <
5587 sizeof_field(QCowSnapshotExtraData, vm_state_size_large) +
5588 sizeof_field(QCowSnapshotExtraData, disk_size))
5589 {
5590 need_snapshot_update = true;
5591 break;
5592 }
5593 }
5594 if (need_snapshot_update) {
5595 ret = qcow2_write_snapshots(bs);
5596 if (ret < 0) {
5597 error_setg_errno(errp, -ret, "Failed to update the snapshot table");
5598 return ret;
5599 }
5600 }
5601 status_cb(bs, 1, 2, cb_opaque);
5602
5603 s->qcow_version = target_version;
5604 ret = qcow2_update_header(bs);
5605 if (ret < 0) {
5606 s->qcow_version = current_version;
5607 error_setg_errno(errp, -ret, "Failed to update the image header");
5608 return ret;
5609 }
5610 status_cb(bs, 2, 2, cb_opaque);
5611
5612 return 0;
5613 }
5614
5615 typedef enum Qcow2AmendOperation {
5616 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
5617 * statically initialized to so that the helper CB can discern the first
5618 * invocation from an operation change */
5619 QCOW2_NO_OPERATION = 0,
5620
5621 QCOW2_UPGRADING,
5622 QCOW2_UPDATING_ENCRYPTION,
5623 QCOW2_CHANGING_REFCOUNT_ORDER,
5624 QCOW2_DOWNGRADING,
5625 } Qcow2AmendOperation;
5626
5627 typedef struct Qcow2AmendHelperCBInfo {
5628 /* The code coordinating the amend operations should only modify
5629 * these four fields; the rest will be managed by the CB */
5630 BlockDriverAmendStatusCB *original_status_cb;
5631 void *original_cb_opaque;
5632
5633 Qcow2AmendOperation current_operation;
5634
5635 /* Total number of operations to perform (only set once) */
5636 int total_operations;
5637
5638 /* The following fields are managed by the CB */
5639
5640 /* Number of operations completed */
5641 int operations_completed;
5642
5643 /* Cumulative offset of all completed operations */
5644 int64_t offset_completed;
5645
5646 Qcow2AmendOperation last_operation;
5647 int64_t last_work_size;
5648 } Qcow2AmendHelperCBInfo;
5649
qcow2_amend_helper_cb(BlockDriverState * bs,int64_t operation_offset,int64_t operation_work_size,void * opaque)5650 static void qcow2_amend_helper_cb(BlockDriverState *bs,
5651 int64_t operation_offset,
5652 int64_t operation_work_size, void *opaque)
5653 {
5654 Qcow2AmendHelperCBInfo *info = opaque;
5655 int64_t current_work_size;
5656 int64_t projected_work_size;
5657
5658 if (info->current_operation != info->last_operation) {
5659 if (info->last_operation != QCOW2_NO_OPERATION) {
5660 info->offset_completed += info->last_work_size;
5661 info->operations_completed++;
5662 }
5663
5664 info->last_operation = info->current_operation;
5665 }
5666
5667 assert(info->total_operations > 0);
5668 assert(info->operations_completed < info->total_operations);
5669
5670 info->last_work_size = operation_work_size;
5671
5672 current_work_size = info->offset_completed + operation_work_size;
5673
5674 /* current_work_size is the total work size for (operations_completed + 1)
5675 * operations (which includes this one), so multiply it by the number of
5676 * operations not covered and divide it by the number of operations
5677 * covered to get a projection for the operations not covered */
5678 projected_work_size = current_work_size * (info->total_operations -
5679 info->operations_completed - 1)
5680 / (info->operations_completed + 1);
5681
5682 info->original_status_cb(bs, info->offset_completed + operation_offset,
5683 current_work_size + projected_work_size,
5684 info->original_cb_opaque);
5685 }
5686
5687 static int GRAPH_RDLOCK
qcow2_amend_options(BlockDriverState * bs,QemuOpts * opts,BlockDriverAmendStatusCB * status_cb,void * cb_opaque,bool force,Error ** errp)5688 qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,
5689 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
5690 bool force, Error **errp)
5691 {
5692 BDRVQcow2State *s = bs->opaque;
5693 int old_version = s->qcow_version, new_version = old_version;
5694 uint64_t new_size = 0;
5695 const char *backing_file = NULL, *backing_format = NULL, *data_file = NULL;
5696 bool lazy_refcounts = s->use_lazy_refcounts;
5697 bool data_file_raw = data_file_is_raw(bs);
5698 const char *compat = NULL;
5699 int refcount_bits = s->refcount_bits;
5700 int ret;
5701 QemuOptDesc *desc = opts->list->desc;
5702 Qcow2AmendHelperCBInfo helper_cb_info;
5703 bool encryption_update = false;
5704
5705 while (desc && desc->name) {
5706 if (!qemu_opt_find(opts, desc->name)) {
5707 /* only change explicitly defined options */
5708 desc++;
5709 continue;
5710 }
5711
5712 if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) {
5713 compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL);
5714 if (!compat) {
5715 /* preserve default */
5716 } else if (!strcmp(compat, "0.10") || !strcmp(compat, "v2")) {
5717 new_version = 2;
5718 } else if (!strcmp(compat, "1.1") || !strcmp(compat, "v3")) {
5719 new_version = 3;
5720 } else {
5721 error_setg(errp, "Unknown compatibility level %s", compat);
5722 return -EINVAL;
5723 }
5724 } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) {
5725 new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
5726 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) {
5727 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
5728 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) {
5729 backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
5730 } else if (g_str_has_prefix(desc->name, "encrypt.")) {
5731 if (!s->crypto) {
5732 error_setg(errp,
5733 "Can't amend encryption options - encryption not present");
5734 return -EINVAL;
5735 }
5736 if (s->crypt_method_header != QCOW_CRYPT_LUKS) {
5737 error_setg(errp,
5738 "Only LUKS encryption options can be amended");
5739 return -ENOTSUP;
5740 }
5741 encryption_update = true;
5742 } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
5743 lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS,
5744 lazy_refcounts);
5745 } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) {
5746 refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS,
5747 refcount_bits);
5748
5749 if (refcount_bits <= 0 || refcount_bits > 64 ||
5750 !is_power_of_2(refcount_bits))
5751 {
5752 error_setg(errp, "Refcount width must be a power of two and "
5753 "may not exceed 64 bits");
5754 return -EINVAL;
5755 }
5756 } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE)) {
5757 data_file = qemu_opt_get(opts, BLOCK_OPT_DATA_FILE);
5758 if (data_file && !has_data_file(bs)) {
5759 error_setg(errp, "data-file can only be set for images that "
5760 "use an external data file");
5761 return -EINVAL;
5762 }
5763 } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE_RAW)) {
5764 data_file_raw = qemu_opt_get_bool(opts, BLOCK_OPT_DATA_FILE_RAW,
5765 data_file_raw);
5766 if (data_file_raw && !data_file_is_raw(bs)) {
5767 error_setg(errp, "data-file-raw cannot be set on existing "
5768 "images");
5769 return -EINVAL;
5770 }
5771 } else {
5772 /* if this point is reached, this probably means a new option was
5773 * added without having it covered here */
5774 abort();
5775 }
5776
5777 desc++;
5778 }
5779
5780 helper_cb_info = (Qcow2AmendHelperCBInfo){
5781 .original_status_cb = status_cb,
5782 .original_cb_opaque = cb_opaque,
5783 .total_operations = (new_version != old_version)
5784 + (s->refcount_bits != refcount_bits) +
5785 (encryption_update == true)
5786 };
5787
5788 /* Upgrade first (some features may require compat=1.1) */
5789 if (new_version > old_version) {
5790 helper_cb_info.current_operation = QCOW2_UPGRADING;
5791 ret = qcow2_upgrade(bs, new_version, &qcow2_amend_helper_cb,
5792 &helper_cb_info, errp);
5793 if (ret < 0) {
5794 return ret;
5795 }
5796 }
5797
5798 if (encryption_update) {
5799 QDict *amend_opts_dict;
5800 QCryptoBlockAmendOptions *amend_opts;
5801
5802 helper_cb_info.current_operation = QCOW2_UPDATING_ENCRYPTION;
5803 amend_opts_dict = qcow2_extract_crypto_opts(opts, "luks", errp);
5804 if (!amend_opts_dict) {
5805 return -EINVAL;
5806 }
5807 amend_opts = block_crypto_amend_opts_init(amend_opts_dict, errp);
5808 qobject_unref(amend_opts_dict);
5809 if (!amend_opts) {
5810 return -EINVAL;
5811 }
5812 ret = qcrypto_block_amend_options(s->crypto,
5813 qcow2_crypto_hdr_read_func,
5814 qcow2_crypto_hdr_write_func,
5815 bs,
5816 amend_opts,
5817 force,
5818 errp);
5819 qapi_free_QCryptoBlockAmendOptions(amend_opts);
5820 if (ret < 0) {
5821 return ret;
5822 }
5823 }
5824
5825 if (s->refcount_bits != refcount_bits) {
5826 int refcount_order = ctz32(refcount_bits);
5827
5828 if (new_version < 3 && refcount_bits != 16) {
5829 error_setg(errp, "Refcount widths other than 16 bits require "
5830 "compatibility level 1.1 or above (use compat=1.1 or "
5831 "greater)");
5832 return -EINVAL;
5833 }
5834
5835 helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER;
5836 ret = qcow2_change_refcount_order(bs, refcount_order,
5837 &qcow2_amend_helper_cb,
5838 &helper_cb_info, errp);
5839 if (ret < 0) {
5840 return ret;
5841 }
5842 }
5843
5844 /* data-file-raw blocks backing files, so clear it first if requested */
5845 if (data_file_raw) {
5846 s->autoclear_features |= QCOW2_AUTOCLEAR_DATA_FILE_RAW;
5847 } else {
5848 s->autoclear_features &= ~QCOW2_AUTOCLEAR_DATA_FILE_RAW;
5849 }
5850
5851 if (data_file) {
5852 g_free(s->image_data_file);
5853 s->image_data_file = *data_file ? g_strdup(data_file) : NULL;
5854 }
5855
5856 ret = qcow2_update_header(bs);
5857 if (ret < 0) {
5858 error_setg_errno(errp, -ret, "Failed to update the image header");
5859 return ret;
5860 }
5861
5862 if (backing_file || backing_format) {
5863 if (g_strcmp0(backing_file, s->image_backing_file) ||
5864 g_strcmp0(backing_format, s->image_backing_format)) {
5865 error_setg(errp, "Cannot amend the backing file");
5866 error_append_hint(errp,
5867 "You can use 'qemu-img rebase' instead.\n");
5868 return -EINVAL;
5869 }
5870 }
5871
5872 if (s->use_lazy_refcounts != lazy_refcounts) {
5873 if (lazy_refcounts) {
5874 if (new_version < 3) {
5875 error_setg(errp, "Lazy refcounts only supported with "
5876 "compatibility level 1.1 and above (use compat=1.1 "
5877 "or greater)");
5878 return -EINVAL;
5879 }
5880 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
5881 ret = qcow2_update_header(bs);
5882 if (ret < 0) {
5883 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
5884 error_setg_errno(errp, -ret, "Failed to update the image header");
5885 return ret;
5886 }
5887 s->use_lazy_refcounts = true;
5888 } else {
5889 /* make image clean first */
5890 ret = qcow2_mark_clean(bs);
5891 if (ret < 0) {
5892 error_setg_errno(errp, -ret, "Failed to make the image clean");
5893 return ret;
5894 }
5895 /* now disallow lazy refcounts */
5896 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
5897 ret = qcow2_update_header(bs);
5898 if (ret < 0) {
5899 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
5900 error_setg_errno(errp, -ret, "Failed to update the image header");
5901 return ret;
5902 }
5903 s->use_lazy_refcounts = false;
5904 }
5905 }
5906
5907 if (new_size) {
5908 BlockBackend *blk = blk_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL,
5909 errp);
5910 if (!blk) {
5911 return -EPERM;
5912 }
5913
5914 /*
5915 * Amending image options should ensure that the image has
5916 * exactly the given new values, so pass exact=true here.
5917 */
5918 ret = blk_truncate(blk, new_size, true, PREALLOC_MODE_OFF, 0, errp);
5919 blk_unref(blk);
5920 if (ret < 0) {
5921 return ret;
5922 }
5923 }
5924
5925 /* Downgrade last (so unsupported features can be removed before) */
5926 if (new_version < old_version) {
5927 helper_cb_info.current_operation = QCOW2_DOWNGRADING;
5928 ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb,
5929 &helper_cb_info, errp);
5930 if (ret < 0) {
5931 return ret;
5932 }
5933 }
5934
5935 return 0;
5936 }
5937
qcow2_co_amend(BlockDriverState * bs,BlockdevAmendOptions * opts,bool force,Error ** errp)5938 static int coroutine_fn qcow2_co_amend(BlockDriverState *bs,
5939 BlockdevAmendOptions *opts,
5940 bool force,
5941 Error **errp)
5942 {
5943 BlockdevAmendOptionsQcow2 *qopts = &opts->u.qcow2;
5944 BDRVQcow2State *s = bs->opaque;
5945 int ret = 0;
5946
5947 if (qopts->encrypt) {
5948 if (!s->crypto) {
5949 error_setg(errp, "image is not encrypted, can't amend");
5950 return -EOPNOTSUPP;
5951 }
5952
5953 if (qopts->encrypt->format != QCRYPTO_BLOCK_FORMAT_LUKS) {
5954 error_setg(errp,
5955 "Amend can't be used to change the qcow2 encryption format");
5956 return -EOPNOTSUPP;
5957 }
5958
5959 if (s->crypt_method_header != QCOW_CRYPT_LUKS) {
5960 error_setg(errp,
5961 "Only LUKS encryption options can be amended for qcow2 with blockdev-amend");
5962 return -EOPNOTSUPP;
5963 }
5964
5965 ret = qcrypto_block_amend_options(s->crypto,
5966 qcow2_crypto_hdr_read_func,
5967 qcow2_crypto_hdr_write_func,
5968 bs,
5969 qopts->encrypt,
5970 force,
5971 errp);
5972 }
5973 return ret;
5974 }
5975
5976 /*
5977 * If offset or size are negative, respectively, they will not be included in
5978 * the BLOCK_IMAGE_CORRUPTED event emitted.
5979 * fatal will be ignored for read-only BDS; corruptions found there will always
5980 * be considered non-fatal.
5981 */
qcow2_signal_corruption(BlockDriverState * bs,bool fatal,int64_t offset,int64_t size,const char * message_format,...)5982 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
5983 int64_t size, const char *message_format, ...)
5984 {
5985 BDRVQcow2State *s = bs->opaque;
5986 const char *node_name;
5987 char *message;
5988 va_list ap;
5989
5990 fatal = fatal && bdrv_is_writable(bs);
5991
5992 if (s->signaled_corruption &&
5993 (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
5994 {
5995 return;
5996 }
5997
5998 va_start(ap, message_format);
5999 message = g_strdup_vprintf(message_format, ap);
6000 va_end(ap);
6001
6002 if (fatal) {
6003 fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
6004 "corruption events will be suppressed\n", message);
6005 } else {
6006 fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
6007 "corruption events will be suppressed\n", message);
6008 }
6009
6010 node_name = bdrv_get_node_name(bs);
6011 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs),
6012 *node_name ? node_name : NULL,
6013 message, offset >= 0, offset,
6014 size >= 0, size,
6015 fatal);
6016 g_free(message);
6017
6018 if (fatal) {
6019 qcow2_mark_corrupt(bs);
6020 bs->drv = NULL; /* make BDS unusable */
6021 }
6022
6023 s->signaled_corruption = true;
6024 }
6025
6026 #define QCOW_COMMON_OPTIONS \
6027 { \
6028 .name = BLOCK_OPT_SIZE, \
6029 .type = QEMU_OPT_SIZE, \
6030 .help = "Virtual disk size" \
6031 }, \
6032 { \
6033 .name = BLOCK_OPT_COMPAT_LEVEL, \
6034 .type = QEMU_OPT_STRING, \
6035 .help = "Compatibility level (v2 [0.10] or v3 [1.1])" \
6036 }, \
6037 { \
6038 .name = BLOCK_OPT_BACKING_FILE, \
6039 .type = QEMU_OPT_STRING, \
6040 .help = "File name of a base image" \
6041 }, \
6042 { \
6043 .name = BLOCK_OPT_BACKING_FMT, \
6044 .type = QEMU_OPT_STRING, \
6045 .help = "Image format of the base image" \
6046 }, \
6047 { \
6048 .name = BLOCK_OPT_DATA_FILE, \
6049 .type = QEMU_OPT_STRING, \
6050 .help = "File name of an external data file" \
6051 }, \
6052 { \
6053 .name = BLOCK_OPT_DATA_FILE_RAW, \
6054 .type = QEMU_OPT_BOOL, \
6055 .help = "The external data file must stay valid " \
6056 "as a raw image" \
6057 }, \
6058 { \
6059 .name = BLOCK_OPT_LAZY_REFCOUNTS, \
6060 .type = QEMU_OPT_BOOL, \
6061 .help = "Postpone refcount updates", \
6062 .def_value_str = "off" \
6063 }, \
6064 { \
6065 .name = BLOCK_OPT_REFCOUNT_BITS, \
6066 .type = QEMU_OPT_NUMBER, \
6067 .help = "Width of a reference count entry in bits", \
6068 .def_value_str = "16" \
6069 }
6070
6071 static QemuOptsList qcow2_create_opts = {
6072 .name = "qcow2-create-opts",
6073 .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
6074 .desc = {
6075 { \
6076 .name = BLOCK_OPT_ENCRYPT, \
6077 .type = QEMU_OPT_BOOL, \
6078 .help = "Encrypt the image with format 'aes'. (Deprecated " \
6079 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)", \
6080 }, \
6081 { \
6082 .name = BLOCK_OPT_ENCRYPT_FORMAT, \
6083 .type = QEMU_OPT_STRING, \
6084 .help = "Encrypt the image, format choices: 'aes', 'luks'", \
6085 }, \
6086 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", \
6087 "ID of secret providing qcow AES key or LUKS passphrase"), \
6088 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."), \
6089 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."), \
6090 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."), \
6091 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."), \
6092 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."), \
6093 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."), \
6094 { \
6095 .name = BLOCK_OPT_CLUSTER_SIZE, \
6096 .type = QEMU_OPT_SIZE, \
6097 .help = "qcow2 cluster size", \
6098 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE) \
6099 }, \
6100 { \
6101 .name = BLOCK_OPT_EXTL2, \
6102 .type = QEMU_OPT_BOOL, \
6103 .help = "Extended L2 tables", \
6104 .def_value_str = "off" \
6105 }, \
6106 { \
6107 .name = BLOCK_OPT_PREALLOC, \
6108 .type = QEMU_OPT_STRING, \
6109 .help = "Preallocation mode (allowed values: off, " \
6110 "metadata, falloc, full)" \
6111 }, \
6112 { \
6113 .name = BLOCK_OPT_COMPRESSION_TYPE, \
6114 .type = QEMU_OPT_STRING, \
6115 .help = "Compression method used for image cluster " \
6116 "compression", \
6117 .def_value_str = "zlib" \
6118 },
6119 QCOW_COMMON_OPTIONS,
6120 { /* end of list */ }
6121 }
6122 };
6123
6124 static QemuOptsList qcow2_amend_opts = {
6125 .name = "qcow2-amend-opts",
6126 .head = QTAILQ_HEAD_INITIALIZER(qcow2_amend_opts.head),
6127 .desc = {
6128 BLOCK_CRYPTO_OPT_DEF_LUKS_STATE("encrypt."),
6129 BLOCK_CRYPTO_OPT_DEF_LUKS_KEYSLOT("encrypt."),
6130 BLOCK_CRYPTO_OPT_DEF_LUKS_OLD_SECRET("encrypt."),
6131 BLOCK_CRYPTO_OPT_DEF_LUKS_NEW_SECRET("encrypt."),
6132 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
6133 QCOW_COMMON_OPTIONS,
6134 { /* end of list */ }
6135 }
6136 };
6137
6138 static const char *const qcow2_strong_runtime_opts[] = {
6139 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET,
6140
6141 NULL
6142 };
6143
6144 BlockDriver bdrv_qcow2 = {
6145 .format_name = "qcow2",
6146 .instance_size = sizeof(BDRVQcow2State),
6147 .bdrv_probe = qcow2_probe,
6148 .bdrv_open = qcow2_open,
6149 .bdrv_close = qcow2_close,
6150 .bdrv_reopen_prepare = qcow2_reopen_prepare,
6151 .bdrv_reopen_commit = qcow2_reopen_commit,
6152 .bdrv_reopen_commit_post = qcow2_reopen_commit_post,
6153 .bdrv_reopen_abort = qcow2_reopen_abort,
6154 .bdrv_join_options = qcow2_join_options,
6155 .bdrv_child_perm = bdrv_default_perms,
6156 .bdrv_co_create_opts = qcow2_co_create_opts,
6157 .bdrv_co_create = qcow2_co_create,
6158 .bdrv_has_zero_init = qcow2_has_zero_init,
6159 .bdrv_co_block_status = qcow2_co_block_status,
6160
6161 .bdrv_co_preadv_part = qcow2_co_preadv_part,
6162 .bdrv_co_pwritev_part = qcow2_co_pwritev_part,
6163 .bdrv_co_flush_to_os = qcow2_co_flush_to_os,
6164
6165 .bdrv_co_pwrite_zeroes = qcow2_co_pwrite_zeroes,
6166 .bdrv_co_pdiscard = qcow2_co_pdiscard,
6167 .bdrv_co_copy_range_from = qcow2_co_copy_range_from,
6168 .bdrv_co_copy_range_to = qcow2_co_copy_range_to,
6169 .bdrv_co_truncate = qcow2_co_truncate,
6170 .bdrv_co_pwritev_compressed_part = qcow2_co_pwritev_compressed_part,
6171 .bdrv_make_empty = qcow2_make_empty,
6172
6173 .bdrv_snapshot_create = qcow2_snapshot_create,
6174 .bdrv_snapshot_goto = qcow2_snapshot_goto,
6175 .bdrv_snapshot_delete = qcow2_snapshot_delete,
6176 .bdrv_snapshot_list = qcow2_snapshot_list,
6177 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
6178 .bdrv_measure = qcow2_measure,
6179 .bdrv_co_get_info = qcow2_co_get_info,
6180 .bdrv_get_specific_info = qcow2_get_specific_info,
6181
6182 .bdrv_co_save_vmstate = qcow2_co_save_vmstate,
6183 .bdrv_co_load_vmstate = qcow2_co_load_vmstate,
6184
6185 .is_format = true,
6186 .supports_backing = true,
6187 .bdrv_co_change_backing_file = qcow2_co_change_backing_file,
6188
6189 .bdrv_refresh_limits = qcow2_refresh_limits,
6190 .bdrv_co_invalidate_cache = qcow2_co_invalidate_cache,
6191 .bdrv_inactivate = qcow2_inactivate,
6192
6193 .create_opts = &qcow2_create_opts,
6194 .amend_opts = &qcow2_amend_opts,
6195 .strong_runtime_opts = qcow2_strong_runtime_opts,
6196 .mutable_opts = mutable_opts,
6197 .bdrv_co_check = qcow2_co_check,
6198 .bdrv_amend_options = qcow2_amend_options,
6199 .bdrv_co_amend = qcow2_co_amend,
6200
6201 .bdrv_detach_aio_context = qcow2_detach_aio_context,
6202 .bdrv_attach_aio_context = qcow2_attach_aio_context,
6203
6204 .bdrv_supports_persistent_dirty_bitmap =
6205 qcow2_supports_persistent_dirty_bitmap,
6206 .bdrv_co_can_store_new_dirty_bitmap = qcow2_co_can_store_new_dirty_bitmap,
6207 .bdrv_co_remove_persistent_dirty_bitmap =
6208 qcow2_co_remove_persistent_dirty_bitmap,
6209 };
6210
bdrv_qcow2_init(void)6211 static void bdrv_qcow2_init(void)
6212 {
6213 bdrv_register(&bdrv_qcow2);
6214 }
6215
6216 block_init(bdrv_qcow2_init);
6217