spapr: Workaround for broken radix guests
[qemu.git] / block / qcow2.c
1 /*
2 * Block driver for the QCOW version 2 format
3 *
4 * Copyright (c) 2004-2006 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24 #include "qemu/osdep.h"
25 #include "block/block_int.h"
26 #include "sysemu/block-backend.h"
27 #include "qemu/module.h"
28 #include <zlib.h>
29 #include "block/qcow2.h"
30 #include "qemu/error-report.h"
31 #include "qapi/qmp/qerror.h"
32 #include "qapi/qmp/qbool.h"
33 #include "qapi/util.h"
34 #include "qapi/qmp/types.h"
35 #include "qapi-event.h"
36 #include "trace.h"
37 #include "qemu/option_int.h"
38 #include "qemu/cutils.h"
39 #include "qemu/bswap.h"
40
41 /*
42 Differences with QCOW:
43
44 - Support for multiple incremental snapshots.
45 - Memory management by reference counts.
46 - Clusters which have a reference count of one have the bit
47 QCOW_OFLAG_COPIED to optimize write performance.
48 - Size of compressed clusters is stored in sectors to reduce bit usage
49 in the cluster offsets.
50 - Support for storing additional data (such as the VM state) in the
51 snapshots.
52 - If a backing store is used, the cluster size is not constrained
53 (could be backported to QCOW).
54 - L2 tables have always a size of one cluster.
55 */
56
57
58 typedef struct {
59 uint32_t magic;
60 uint32_t len;
61 } QEMU_PACKED QCowExtension;
62
63 #define QCOW2_EXT_MAGIC_END 0
64 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
65 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
66
67 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
68 {
69 const QCowHeader *cow_header = (const void *)buf;
70
71 if (buf_size >= sizeof(QCowHeader) &&
72 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
73 be32_to_cpu(cow_header->version) >= 2)
74 return 100;
75 else
76 return 0;
77 }
78
79
80 /*
81 * read qcow2 extension and fill bs
82 * start reading from start_offset
83 * finish reading upon magic of value 0 or when end_offset reached
84 * unknown magic is skipped (future extension this version knows nothing about)
85 * return 0 upon success, non-0 otherwise
86 */
87 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
88 uint64_t end_offset, void **p_feature_table,
89 Error **errp)
90 {
91 BDRVQcow2State *s = bs->opaque;
92 QCowExtension ext;
93 uint64_t offset;
94 int ret;
95
96 #ifdef DEBUG_EXT
97 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
98 #endif
99 offset = start_offset;
100 while (offset < end_offset) {
101
102 #ifdef DEBUG_EXT
103 /* Sanity check */
104 if (offset > s->cluster_size)
105 printf("qcow2_read_extension: suspicious offset %lu\n", offset);
106
107 printf("attempting to read extended header in offset %lu\n", offset);
108 #endif
109
110 ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext));
111 if (ret < 0) {
112 error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: "
113 "pread fail from offset %" PRIu64, offset);
114 return 1;
115 }
116 be32_to_cpus(&ext.magic);
117 be32_to_cpus(&ext.len);
118 offset += sizeof(ext);
119 #ifdef DEBUG_EXT
120 printf("ext.magic = 0x%x\n", ext.magic);
121 #endif
122 if (offset > end_offset || ext.len > end_offset - offset) {
123 error_setg(errp, "Header extension too large");
124 return -EINVAL;
125 }
126
127 switch (ext.magic) {
128 case QCOW2_EXT_MAGIC_END:
129 return 0;
130
131 case QCOW2_EXT_MAGIC_BACKING_FORMAT:
132 if (ext.len >= sizeof(bs->backing_format)) {
133 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32
134 " too large (>=%zu)", ext.len,
135 sizeof(bs->backing_format));
136 return 2;
137 }
138 ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len);
139 if (ret < 0) {
140 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: "
141 "Could not read format name");
142 return 3;
143 }
144 bs->backing_format[ext.len] = '\0';
145 s->image_backing_format = g_strdup(bs->backing_format);
146 #ifdef DEBUG_EXT
147 printf("Qcow2: Got format extension %s\n", bs->backing_format);
148 #endif
149 break;
150
151 case QCOW2_EXT_MAGIC_FEATURE_TABLE:
152 if (p_feature_table != NULL) {
153 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
154 ret = bdrv_pread(bs->file, offset , feature_table, ext.len);
155 if (ret < 0) {
156 error_setg_errno(errp, -ret, "ERROR: ext_feature_table: "
157 "Could not read table");
158 return ret;
159 }
160
161 *p_feature_table = feature_table;
162 }
163 break;
164
165 default:
166 /* unknown magic - save it in case we need to rewrite the header */
167 {
168 Qcow2UnknownHeaderExtension *uext;
169
170 uext = g_malloc0(sizeof(*uext) + ext.len);
171 uext->magic = ext.magic;
172 uext->len = ext.len;
173 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
174
175 ret = bdrv_pread(bs->file, offset , uext->data, uext->len);
176 if (ret < 0) {
177 error_setg_errno(errp, -ret, "ERROR: unknown extension: "
178 "Could not read data");
179 return ret;
180 }
181 }
182 break;
183 }
184
185 offset += ((ext.len + 7) & ~7);
186 }
187
188 return 0;
189 }
190
191 static void cleanup_unknown_header_ext(BlockDriverState *bs)
192 {
193 BDRVQcow2State *s = bs->opaque;
194 Qcow2UnknownHeaderExtension *uext, *next;
195
196 QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
197 QLIST_REMOVE(uext, next);
198 g_free(uext);
199 }
200 }
201
202 static void report_unsupported_feature(Error **errp, Qcow2Feature *table,
203 uint64_t mask)
204 {
205 char *features = g_strdup("");
206 char *old;
207
208 while (table && table->name[0] != '\0') {
209 if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
210 if (mask & (1ULL << table->bit)) {
211 old = features;
212 features = g_strdup_printf("%s%s%.46s", old, *old ? ", " : "",
213 table->name);
214 g_free(old);
215 mask &= ~(1ULL << table->bit);
216 }
217 }
218 table++;
219 }
220
221 if (mask) {
222 old = features;
223 features = g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64,
224 old, *old ? ", " : "", mask);
225 g_free(old);
226 }
227
228 error_setg(errp, "Unsupported qcow2 feature(s): %s", features);
229 g_free(features);
230 }
231
232 /*
233 * Sets the dirty bit and flushes afterwards if necessary.
234 *
235 * The incompatible_features bit is only set if the image file header was
236 * updated successfully. Therefore it is not required to check the return
237 * value of this function.
238 */
239 int qcow2_mark_dirty(BlockDriverState *bs)
240 {
241 BDRVQcow2State *s = bs->opaque;
242 uint64_t val;
243 int ret;
244
245 assert(s->qcow_version >= 3);
246
247 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
248 return 0; /* already dirty */
249 }
250
251 val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY);
252 ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features),
253 &val, sizeof(val));
254 if (ret < 0) {
255 return ret;
256 }
257 ret = bdrv_flush(bs->file->bs);
258 if (ret < 0) {
259 return ret;
260 }
261
262 /* Only treat image as dirty if the header was updated successfully */
263 s->incompatible_features |= QCOW2_INCOMPAT_DIRTY;
264 return 0;
265 }
266
267 /*
268 * Clears the dirty bit and flushes before if necessary. Only call this
269 * function when there are no pending requests, it does not guard against
270 * concurrent requests dirtying the image.
271 */
272 static int qcow2_mark_clean(BlockDriverState *bs)
273 {
274 BDRVQcow2State *s = bs->opaque;
275
276 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
277 int ret;
278
279 s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY;
280
281 ret = bdrv_flush(bs);
282 if (ret < 0) {
283 return ret;
284 }
285
286 return qcow2_update_header(bs);
287 }
288 return 0;
289 }
290
291 /*
292 * Marks the image as corrupt.
293 */
294 int qcow2_mark_corrupt(BlockDriverState *bs)
295 {
296 BDRVQcow2State *s = bs->opaque;
297
298 s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT;
299 return qcow2_update_header(bs);
300 }
301
302 /*
303 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
304 * before if necessary.
305 */
306 int qcow2_mark_consistent(BlockDriverState *bs)
307 {
308 BDRVQcow2State *s = bs->opaque;
309
310 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
311 int ret = bdrv_flush(bs);
312 if (ret < 0) {
313 return ret;
314 }
315
316 s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT;
317 return qcow2_update_header(bs);
318 }
319 return 0;
320 }
321
322 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result,
323 BdrvCheckMode fix)
324 {
325 int ret = qcow2_check_refcounts(bs, result, fix);
326 if (ret < 0) {
327 return ret;
328 }
329
330 if (fix && result->check_errors == 0 && result->corruptions == 0) {
331 ret = qcow2_mark_clean(bs);
332 if (ret < 0) {
333 return ret;
334 }
335 return qcow2_mark_consistent(bs);
336 }
337 return ret;
338 }
339
340 static int validate_table_offset(BlockDriverState *bs, uint64_t offset,
341 uint64_t entries, size_t entry_len)
342 {
343 BDRVQcow2State *s = bs->opaque;
344 uint64_t size;
345
346 /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
347 * because values will be passed to qemu functions taking int64_t. */
348 if (entries > INT64_MAX / entry_len) {
349 return -EINVAL;
350 }
351
352 size = entries * entry_len;
353
354 if (INT64_MAX - size < offset) {
355 return -EINVAL;
356 }
357
358 /* Tables must be cluster aligned */
359 if (offset & (s->cluster_size - 1)) {
360 return -EINVAL;
361 }
362
363 return 0;
364 }
365
366 static QemuOptsList qcow2_runtime_opts = {
367 .name = "qcow2",
368 .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head),
369 .desc = {
370 {
371 .name = QCOW2_OPT_LAZY_REFCOUNTS,
372 .type = QEMU_OPT_BOOL,
373 .help = "Postpone refcount updates",
374 },
375 {
376 .name = QCOW2_OPT_DISCARD_REQUEST,
377 .type = QEMU_OPT_BOOL,
378 .help = "Pass guest discard requests to the layer below",
379 },
380 {
381 .name = QCOW2_OPT_DISCARD_SNAPSHOT,
382 .type = QEMU_OPT_BOOL,
383 .help = "Generate discard requests when snapshot related space "
384 "is freed",
385 },
386 {
387 .name = QCOW2_OPT_DISCARD_OTHER,
388 .type = QEMU_OPT_BOOL,
389 .help = "Generate discard requests when other clusters are freed",
390 },
391 {
392 .name = QCOW2_OPT_OVERLAP,
393 .type = QEMU_OPT_STRING,
394 .help = "Selects which overlap checks to perform from a range of "
395 "templates (none, constant, cached, all)",
396 },
397 {
398 .name = QCOW2_OPT_OVERLAP_TEMPLATE,
399 .type = QEMU_OPT_STRING,
400 .help = "Selects which overlap checks to perform from a range of "
401 "templates (none, constant, cached, all)",
402 },
403 {
404 .name = QCOW2_OPT_OVERLAP_MAIN_HEADER,
405 .type = QEMU_OPT_BOOL,
406 .help = "Check for unintended writes into the main qcow2 header",
407 },
408 {
409 .name = QCOW2_OPT_OVERLAP_ACTIVE_L1,
410 .type = QEMU_OPT_BOOL,
411 .help = "Check for unintended writes into the active L1 table",
412 },
413 {
414 .name = QCOW2_OPT_OVERLAP_ACTIVE_L2,
415 .type = QEMU_OPT_BOOL,
416 .help = "Check for unintended writes into an active L2 table",
417 },
418 {
419 .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
420 .type = QEMU_OPT_BOOL,
421 .help = "Check for unintended writes into the refcount table",
422 },
423 {
424 .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
425 .type = QEMU_OPT_BOOL,
426 .help = "Check for unintended writes into a refcount block",
427 },
428 {
429 .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
430 .type = QEMU_OPT_BOOL,
431 .help = "Check for unintended writes into the snapshot table",
432 },
433 {
434 .name = QCOW2_OPT_OVERLAP_INACTIVE_L1,
435 .type = QEMU_OPT_BOOL,
436 .help = "Check for unintended writes into an inactive L1 table",
437 },
438 {
439 .name = QCOW2_OPT_OVERLAP_INACTIVE_L2,
440 .type = QEMU_OPT_BOOL,
441 .help = "Check for unintended writes into an inactive L2 table",
442 },
443 {
444 .name = QCOW2_OPT_CACHE_SIZE,
445 .type = QEMU_OPT_SIZE,
446 .help = "Maximum combined metadata (L2 tables and refcount blocks) "
447 "cache size",
448 },
449 {
450 .name = QCOW2_OPT_L2_CACHE_SIZE,
451 .type = QEMU_OPT_SIZE,
452 .help = "Maximum L2 table cache size",
453 },
454 {
455 .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE,
456 .type = QEMU_OPT_SIZE,
457 .help = "Maximum refcount block cache size",
458 },
459 {
460 .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL,
461 .type = QEMU_OPT_NUMBER,
462 .help = "Clean unused cache entries after this time (in seconds)",
463 },
464 { /* end of list */ }
465 },
466 };
467
468 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = {
469 [QCOW2_OL_MAIN_HEADER_BITNR] = QCOW2_OPT_OVERLAP_MAIN_HEADER,
470 [QCOW2_OL_ACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L1,
471 [QCOW2_OL_ACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L2,
472 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
473 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
474 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
475 [QCOW2_OL_INACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L1,
476 [QCOW2_OL_INACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L2,
477 };
478
479 static void cache_clean_timer_cb(void *opaque)
480 {
481 BlockDriverState *bs = opaque;
482 BDRVQcow2State *s = bs->opaque;
483 qcow2_cache_clean_unused(bs, s->l2_table_cache);
484 qcow2_cache_clean_unused(bs, s->refcount_block_cache);
485 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
486 (int64_t) s->cache_clean_interval * 1000);
487 }
488
489 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context)
490 {
491 BDRVQcow2State *s = bs->opaque;
492 if (s->cache_clean_interval > 0) {
493 s->cache_clean_timer = aio_timer_new(context, QEMU_CLOCK_VIRTUAL,
494 SCALE_MS, cache_clean_timer_cb,
495 bs);
496 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
497 (int64_t) s->cache_clean_interval * 1000);
498 }
499 }
500
501 static void cache_clean_timer_del(BlockDriverState *bs)
502 {
503 BDRVQcow2State *s = bs->opaque;
504 if (s->cache_clean_timer) {
505 timer_del(s->cache_clean_timer);
506 timer_free(s->cache_clean_timer);
507 s->cache_clean_timer = NULL;
508 }
509 }
510
511 static void qcow2_detach_aio_context(BlockDriverState *bs)
512 {
513 cache_clean_timer_del(bs);
514 }
515
516 static void qcow2_attach_aio_context(BlockDriverState *bs,
517 AioContext *new_context)
518 {
519 cache_clean_timer_init(bs, new_context);
520 }
521
522 static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts,
523 uint64_t *l2_cache_size,
524 uint64_t *refcount_cache_size, Error **errp)
525 {
526 BDRVQcow2State *s = bs->opaque;
527 uint64_t combined_cache_size;
528 bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set;
529
530 combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE);
531 l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE);
532 refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
533
534 combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0);
535 *l2_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 0);
536 *refcount_cache_size = qemu_opt_get_size(opts,
537 QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0);
538
539 if (combined_cache_size_set) {
540 if (l2_cache_size_set && refcount_cache_size_set) {
541 error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE
542 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set "
543 "the same time");
544 return;
545 } else if (*l2_cache_size > combined_cache_size) {
546 error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed "
547 QCOW2_OPT_CACHE_SIZE);
548 return;
549 } else if (*refcount_cache_size > combined_cache_size) {
550 error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed "
551 QCOW2_OPT_CACHE_SIZE);
552 return;
553 }
554
555 if (l2_cache_size_set) {
556 *refcount_cache_size = combined_cache_size - *l2_cache_size;
557 } else if (refcount_cache_size_set) {
558 *l2_cache_size = combined_cache_size - *refcount_cache_size;
559 } else {
560 *refcount_cache_size = combined_cache_size
561 / (DEFAULT_L2_REFCOUNT_SIZE_RATIO + 1);
562 *l2_cache_size = combined_cache_size - *refcount_cache_size;
563 }
564 } else {
565 if (!l2_cache_size_set && !refcount_cache_size_set) {
566 *l2_cache_size = MAX(DEFAULT_L2_CACHE_BYTE_SIZE,
567 (uint64_t)DEFAULT_L2_CACHE_CLUSTERS
568 * s->cluster_size);
569 *refcount_cache_size = *l2_cache_size
570 / DEFAULT_L2_REFCOUNT_SIZE_RATIO;
571 } else if (!l2_cache_size_set) {
572 *l2_cache_size = *refcount_cache_size
573 * DEFAULT_L2_REFCOUNT_SIZE_RATIO;
574 } else if (!refcount_cache_size_set) {
575 *refcount_cache_size = *l2_cache_size
576 / DEFAULT_L2_REFCOUNT_SIZE_RATIO;
577 }
578 }
579 }
580
581 typedef struct Qcow2ReopenState {
582 Qcow2Cache *l2_table_cache;
583 Qcow2Cache *refcount_block_cache;
584 bool use_lazy_refcounts;
585 int overlap_check;
586 bool discard_passthrough[QCOW2_DISCARD_MAX];
587 uint64_t cache_clean_interval;
588 } Qcow2ReopenState;
589
590 static int qcow2_update_options_prepare(BlockDriverState *bs,
591 Qcow2ReopenState *r,
592 QDict *options, int flags,
593 Error **errp)
594 {
595 BDRVQcow2State *s = bs->opaque;
596 QemuOpts *opts = NULL;
597 const char *opt_overlap_check, *opt_overlap_check_template;
598 int overlap_check_template = 0;
599 uint64_t l2_cache_size, refcount_cache_size;
600 int i;
601 Error *local_err = NULL;
602 int ret;
603
604 opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort);
605 qemu_opts_absorb_qdict(opts, options, &local_err);
606 if (local_err) {
607 error_propagate(errp, local_err);
608 ret = -EINVAL;
609 goto fail;
610 }
611
612 /* get L2 table/refcount block cache size from command line options */
613 read_cache_sizes(bs, opts, &l2_cache_size, &refcount_cache_size,
614 &local_err);
615 if (local_err) {
616 error_propagate(errp, local_err);
617 ret = -EINVAL;
618 goto fail;
619 }
620
621 l2_cache_size /= s->cluster_size;
622 if (l2_cache_size < MIN_L2_CACHE_SIZE) {
623 l2_cache_size = MIN_L2_CACHE_SIZE;
624 }
625 if (l2_cache_size > INT_MAX) {
626 error_setg(errp, "L2 cache size too big");
627 ret = -EINVAL;
628 goto fail;
629 }
630
631 refcount_cache_size /= s->cluster_size;
632 if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) {
633 refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE;
634 }
635 if (refcount_cache_size > INT_MAX) {
636 error_setg(errp, "Refcount cache size too big");
637 ret = -EINVAL;
638 goto fail;
639 }
640
641 /* alloc new L2 table/refcount block cache, flush old one */
642 if (s->l2_table_cache) {
643 ret = qcow2_cache_flush(bs, s->l2_table_cache);
644 if (ret) {
645 error_setg_errno(errp, -ret, "Failed to flush the L2 table cache");
646 goto fail;
647 }
648 }
649
650 if (s->refcount_block_cache) {
651 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
652 if (ret) {
653 error_setg_errno(errp, -ret,
654 "Failed to flush the refcount block cache");
655 goto fail;
656 }
657 }
658
659 r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size);
660 r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size);
661 if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) {
662 error_setg(errp, "Could not allocate metadata caches");
663 ret = -ENOMEM;
664 goto fail;
665 }
666
667 /* New interval for cache cleanup timer */
668 r->cache_clean_interval =
669 qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL,
670 s->cache_clean_interval);
671 #ifndef CONFIG_LINUX
672 if (r->cache_clean_interval != 0) {
673 error_setg(errp, QCOW2_OPT_CACHE_CLEAN_INTERVAL
674 " not supported on this host");
675 ret = -EINVAL;
676 goto fail;
677 }
678 #endif
679 if (r->cache_clean_interval > UINT_MAX) {
680 error_setg(errp, "Cache clean interval too big");
681 ret = -EINVAL;
682 goto fail;
683 }
684
685 /* lazy-refcounts; flush if going from enabled to disabled */
686 r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS,
687 (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS));
688 if (r->use_lazy_refcounts && s->qcow_version < 3) {
689 error_setg(errp, "Lazy refcounts require a qcow2 image with at least "
690 "qemu 1.1 compatibility level");
691 ret = -EINVAL;
692 goto fail;
693 }
694
695 if (s->use_lazy_refcounts && !r->use_lazy_refcounts) {
696 ret = qcow2_mark_clean(bs);
697 if (ret < 0) {
698 error_setg_errno(errp, -ret, "Failed to disable lazy refcounts");
699 goto fail;
700 }
701 }
702
703 /* Overlap check options */
704 opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP);
705 opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE);
706 if (opt_overlap_check_template && opt_overlap_check &&
707 strcmp(opt_overlap_check_template, opt_overlap_check))
708 {
709 error_setg(errp, "Conflicting values for qcow2 options '"
710 QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
711 "' ('%s')", opt_overlap_check, opt_overlap_check_template);
712 ret = -EINVAL;
713 goto fail;
714 }
715 if (!opt_overlap_check) {
716 opt_overlap_check = opt_overlap_check_template ?: "cached";
717 }
718
719 if (!strcmp(opt_overlap_check, "none")) {
720 overlap_check_template = 0;
721 } else if (!strcmp(opt_overlap_check, "constant")) {
722 overlap_check_template = QCOW2_OL_CONSTANT;
723 } else if (!strcmp(opt_overlap_check, "cached")) {
724 overlap_check_template = QCOW2_OL_CACHED;
725 } else if (!strcmp(opt_overlap_check, "all")) {
726 overlap_check_template = QCOW2_OL_ALL;
727 } else {
728 error_setg(errp, "Unsupported value '%s' for qcow2 option "
729 "'overlap-check'. Allowed are any of the following: "
730 "none, constant, cached, all", opt_overlap_check);
731 ret = -EINVAL;
732 goto fail;
733 }
734
735 r->overlap_check = 0;
736 for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) {
737 /* overlap-check defines a template bitmask, but every flag may be
738 * overwritten through the associated boolean option */
739 r->overlap_check |=
740 qemu_opt_get_bool(opts, overlap_bool_option_names[i],
741 overlap_check_template & (1 << i)) << i;
742 }
743
744 r->discard_passthrough[QCOW2_DISCARD_NEVER] = false;
745 r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true;
746 r->discard_passthrough[QCOW2_DISCARD_REQUEST] =
747 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST,
748 flags & BDRV_O_UNMAP);
749 r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] =
750 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true);
751 r->discard_passthrough[QCOW2_DISCARD_OTHER] =
752 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false);
753
754 ret = 0;
755 fail:
756 qemu_opts_del(opts);
757 opts = NULL;
758 return ret;
759 }
760
761 static void qcow2_update_options_commit(BlockDriverState *bs,
762 Qcow2ReopenState *r)
763 {
764 BDRVQcow2State *s = bs->opaque;
765 int i;
766
767 if (s->l2_table_cache) {
768 qcow2_cache_destroy(bs, s->l2_table_cache);
769 }
770 if (s->refcount_block_cache) {
771 qcow2_cache_destroy(bs, s->refcount_block_cache);
772 }
773 s->l2_table_cache = r->l2_table_cache;
774 s->refcount_block_cache = r->refcount_block_cache;
775
776 s->overlap_check = r->overlap_check;
777 s->use_lazy_refcounts = r->use_lazy_refcounts;
778
779 for (i = 0; i < QCOW2_DISCARD_MAX; i++) {
780 s->discard_passthrough[i] = r->discard_passthrough[i];
781 }
782
783 if (s->cache_clean_interval != r->cache_clean_interval) {
784 cache_clean_timer_del(bs);
785 s->cache_clean_interval = r->cache_clean_interval;
786 cache_clean_timer_init(bs, bdrv_get_aio_context(bs));
787 }
788 }
789
790 static void qcow2_update_options_abort(BlockDriverState *bs,
791 Qcow2ReopenState *r)
792 {
793 if (r->l2_table_cache) {
794 qcow2_cache_destroy(bs, r->l2_table_cache);
795 }
796 if (r->refcount_block_cache) {
797 qcow2_cache_destroy(bs, r->refcount_block_cache);
798 }
799 }
800
801 static int qcow2_update_options(BlockDriverState *bs, QDict *options,
802 int flags, Error **errp)
803 {
804 Qcow2ReopenState r = {};
805 int ret;
806
807 ret = qcow2_update_options_prepare(bs, &r, options, flags, errp);
808 if (ret >= 0) {
809 qcow2_update_options_commit(bs, &r);
810 } else {
811 qcow2_update_options_abort(bs, &r);
812 }
813
814 return ret;
815 }
816
817 static int qcow2_do_open(BlockDriverState *bs, QDict *options, int flags,
818 Error **errp)
819 {
820 BDRVQcow2State *s = bs->opaque;
821 unsigned int len, i;
822 int ret = 0;
823 QCowHeader header;
824 Error *local_err = NULL;
825 uint64_t ext_end;
826 uint64_t l1_vm_state_index;
827
828 ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
829 if (ret < 0) {
830 error_setg_errno(errp, -ret, "Could not read qcow2 header");
831 goto fail;
832 }
833 be32_to_cpus(&header.magic);
834 be32_to_cpus(&header.version);
835 be64_to_cpus(&header.backing_file_offset);
836 be32_to_cpus(&header.backing_file_size);
837 be64_to_cpus(&header.size);
838 be32_to_cpus(&header.cluster_bits);
839 be32_to_cpus(&header.crypt_method);
840 be64_to_cpus(&header.l1_table_offset);
841 be32_to_cpus(&header.l1_size);
842 be64_to_cpus(&header.refcount_table_offset);
843 be32_to_cpus(&header.refcount_table_clusters);
844 be64_to_cpus(&header.snapshots_offset);
845 be32_to_cpus(&header.nb_snapshots);
846
847 if (header.magic != QCOW_MAGIC) {
848 error_setg(errp, "Image is not in qcow2 format");
849 ret = -EINVAL;
850 goto fail;
851 }
852 if (header.version < 2 || header.version > 3) {
853 error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version);
854 ret = -ENOTSUP;
855 goto fail;
856 }
857
858 s->qcow_version = header.version;
859
860 /* Initialise cluster size */
861 if (header.cluster_bits < MIN_CLUSTER_BITS ||
862 header.cluster_bits > MAX_CLUSTER_BITS) {
863 error_setg(errp, "Unsupported cluster size: 2^%" PRIu32,
864 header.cluster_bits);
865 ret = -EINVAL;
866 goto fail;
867 }
868
869 s->cluster_bits = header.cluster_bits;
870 s->cluster_size = 1 << s->cluster_bits;
871 s->cluster_sectors = 1 << (s->cluster_bits - 9);
872
873 /* Initialise version 3 header fields */
874 if (header.version == 2) {
875 header.incompatible_features = 0;
876 header.compatible_features = 0;
877 header.autoclear_features = 0;
878 header.refcount_order = 4;
879 header.header_length = 72;
880 } else {
881 be64_to_cpus(&header.incompatible_features);
882 be64_to_cpus(&header.compatible_features);
883 be64_to_cpus(&header.autoclear_features);
884 be32_to_cpus(&header.refcount_order);
885 be32_to_cpus(&header.header_length);
886
887 if (header.header_length < 104) {
888 error_setg(errp, "qcow2 header too short");
889 ret = -EINVAL;
890 goto fail;
891 }
892 }
893
894 if (header.header_length > s->cluster_size) {
895 error_setg(errp, "qcow2 header exceeds cluster size");
896 ret = -EINVAL;
897 goto fail;
898 }
899
900 if (header.header_length > sizeof(header)) {
901 s->unknown_header_fields_size = header.header_length - sizeof(header);
902 s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
903 ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields,
904 s->unknown_header_fields_size);
905 if (ret < 0) {
906 error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
907 "fields");
908 goto fail;
909 }
910 }
911
912 if (header.backing_file_offset > s->cluster_size) {
913 error_setg(errp, "Invalid backing file offset");
914 ret = -EINVAL;
915 goto fail;
916 }
917
918 if (header.backing_file_offset) {
919 ext_end = header.backing_file_offset;
920 } else {
921 ext_end = 1 << header.cluster_bits;
922 }
923
924 /* Handle feature bits */
925 s->incompatible_features = header.incompatible_features;
926 s->compatible_features = header.compatible_features;
927 s->autoclear_features = header.autoclear_features;
928
929 if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
930 void *feature_table = NULL;
931 qcow2_read_extensions(bs, header.header_length, ext_end,
932 &feature_table, NULL);
933 report_unsupported_feature(errp, feature_table,
934 s->incompatible_features &
935 ~QCOW2_INCOMPAT_MASK);
936 ret = -ENOTSUP;
937 g_free(feature_table);
938 goto fail;
939 }
940
941 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
942 /* Corrupt images may not be written to unless they are being repaired
943 */
944 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
945 error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
946 "read/write");
947 ret = -EACCES;
948 goto fail;
949 }
950 }
951
952 /* Check support for various header values */
953 if (header.refcount_order > 6) {
954 error_setg(errp, "Reference count entry width too large; may not "
955 "exceed 64 bits");
956 ret = -EINVAL;
957 goto fail;
958 }
959 s->refcount_order = header.refcount_order;
960 s->refcount_bits = 1 << s->refcount_order;
961 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
962 s->refcount_max += s->refcount_max - 1;
963
964 if (header.crypt_method > QCOW_CRYPT_AES) {
965 error_setg(errp, "Unsupported encryption method: %" PRIu32,
966 header.crypt_method);
967 ret = -EINVAL;
968 goto fail;
969 }
970 if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128,
971 QCRYPTO_CIPHER_MODE_CBC)) {
972 error_setg(errp, "AES cipher not available");
973 ret = -EINVAL;
974 goto fail;
975 }
976 s->crypt_method_header = header.crypt_method;
977 if (s->crypt_method_header) {
978 if (bdrv_uses_whitelist() &&
979 s->crypt_method_header == QCOW_CRYPT_AES) {
980 error_setg(errp,
981 "Use of AES-CBC encrypted qcow2 images is no longer "
982 "supported in system emulators");
983 error_append_hint(errp,
984 "You can use 'qemu-img convert' to convert your "
985 "image to an alternative supported format, such "
986 "as unencrypted qcow2, or raw with the LUKS "
987 "format instead.\n");
988 ret = -ENOSYS;
989 goto fail;
990 }
991
992 bs->encrypted = true;
993 }
994
995 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
996 s->l2_size = 1 << s->l2_bits;
997 /* 2^(s->refcount_order - 3) is the refcount width in bytes */
998 s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
999 s->refcount_block_size = 1 << s->refcount_block_bits;
1000 bs->total_sectors = header.size / 512;
1001 s->csize_shift = (62 - (s->cluster_bits - 8));
1002 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
1003 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
1004
1005 s->refcount_table_offset = header.refcount_table_offset;
1006 s->refcount_table_size =
1007 header.refcount_table_clusters << (s->cluster_bits - 3);
1008
1009 if (header.refcount_table_clusters > qcow2_max_refcount_clusters(s)) {
1010 error_setg(errp, "Reference count table too large");
1011 ret = -EINVAL;
1012 goto fail;
1013 }
1014
1015 ret = validate_table_offset(bs, s->refcount_table_offset,
1016 s->refcount_table_size, sizeof(uint64_t));
1017 if (ret < 0) {
1018 error_setg(errp, "Invalid reference count table offset");
1019 goto fail;
1020 }
1021
1022 /* Snapshot table offset/length */
1023 if (header.nb_snapshots > QCOW_MAX_SNAPSHOTS) {
1024 error_setg(errp, "Too many snapshots");
1025 ret = -EINVAL;
1026 goto fail;
1027 }
1028
1029 ret = validate_table_offset(bs, header.snapshots_offset,
1030 header.nb_snapshots,
1031 sizeof(QCowSnapshotHeader));
1032 if (ret < 0) {
1033 error_setg(errp, "Invalid snapshot table offset");
1034 goto fail;
1035 }
1036
1037 /* read the level 1 table */
1038 if (header.l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) {
1039 error_setg(errp, "Active L1 table too large");
1040 ret = -EFBIG;
1041 goto fail;
1042 }
1043 s->l1_size = header.l1_size;
1044
1045 l1_vm_state_index = size_to_l1(s, header.size);
1046 if (l1_vm_state_index > INT_MAX) {
1047 error_setg(errp, "Image is too big");
1048 ret = -EFBIG;
1049 goto fail;
1050 }
1051 s->l1_vm_state_index = l1_vm_state_index;
1052
1053 /* the L1 table must contain at least enough entries to put
1054 header.size bytes */
1055 if (s->l1_size < s->l1_vm_state_index) {
1056 error_setg(errp, "L1 table is too small");
1057 ret = -EINVAL;
1058 goto fail;
1059 }
1060
1061 ret = validate_table_offset(bs, header.l1_table_offset,
1062 header.l1_size, sizeof(uint64_t));
1063 if (ret < 0) {
1064 error_setg(errp, "Invalid L1 table offset");
1065 goto fail;
1066 }
1067 s->l1_table_offset = header.l1_table_offset;
1068
1069
1070 if (s->l1_size > 0) {
1071 s->l1_table = qemu_try_blockalign(bs->file->bs,
1072 align_offset(s->l1_size * sizeof(uint64_t), 512));
1073 if (s->l1_table == NULL) {
1074 error_setg(errp, "Could not allocate L1 table");
1075 ret = -ENOMEM;
1076 goto fail;
1077 }
1078 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
1079 s->l1_size * sizeof(uint64_t));
1080 if (ret < 0) {
1081 error_setg_errno(errp, -ret, "Could not read L1 table");
1082 goto fail;
1083 }
1084 for(i = 0;i < s->l1_size; i++) {
1085 be64_to_cpus(&s->l1_table[i]);
1086 }
1087 }
1088
1089 /* Parse driver-specific options */
1090 ret = qcow2_update_options(bs, options, flags, errp);
1091 if (ret < 0) {
1092 goto fail;
1093 }
1094
1095 s->cluster_cache = g_malloc(s->cluster_size);
1096 /* one more sector for decompressed data alignment */
1097 s->cluster_data = qemu_try_blockalign(bs->file->bs, QCOW_MAX_CRYPT_CLUSTERS
1098 * s->cluster_size + 512);
1099 if (s->cluster_data == NULL) {
1100 error_setg(errp, "Could not allocate temporary cluster buffer");
1101 ret = -ENOMEM;
1102 goto fail;
1103 }
1104
1105 s->cluster_cache_offset = -1;
1106 s->flags = flags;
1107
1108 ret = qcow2_refcount_init(bs);
1109 if (ret != 0) {
1110 error_setg_errno(errp, -ret, "Could not initialize refcount handling");
1111 goto fail;
1112 }
1113
1114 QLIST_INIT(&s->cluster_allocs);
1115 QTAILQ_INIT(&s->discards);
1116
1117 /* read qcow2 extensions */
1118 if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
1119 &local_err)) {
1120 error_propagate(errp, local_err);
1121 ret = -EINVAL;
1122 goto fail;
1123 }
1124
1125 /* read the backing file name */
1126 if (header.backing_file_offset != 0) {
1127 len = header.backing_file_size;
1128 if (len > MIN(1023, s->cluster_size - header.backing_file_offset) ||
1129 len >= sizeof(bs->backing_file)) {
1130 error_setg(errp, "Backing file name too long");
1131 ret = -EINVAL;
1132 goto fail;
1133 }
1134 ret = bdrv_pread(bs->file, header.backing_file_offset,
1135 bs->backing_file, len);
1136 if (ret < 0) {
1137 error_setg_errno(errp, -ret, "Could not read backing file name");
1138 goto fail;
1139 }
1140 bs->backing_file[len] = '\0';
1141 s->image_backing_file = g_strdup(bs->backing_file);
1142 }
1143
1144 /* Internal snapshots */
1145 s->snapshots_offset = header.snapshots_offset;
1146 s->nb_snapshots = header.nb_snapshots;
1147
1148 ret = qcow2_read_snapshots(bs);
1149 if (ret < 0) {
1150 error_setg_errno(errp, -ret, "Could not read snapshots");
1151 goto fail;
1152 }
1153
1154 /* Clear unknown autoclear feature bits */
1155 if (!bs->read_only && !(flags & BDRV_O_INACTIVE) && s->autoclear_features) {
1156 s->autoclear_features = 0;
1157 ret = qcow2_update_header(bs);
1158 if (ret < 0) {
1159 error_setg_errno(errp, -ret, "Could not update qcow2 header");
1160 goto fail;
1161 }
1162 }
1163
1164 /* Initialise locks */
1165 qemu_co_mutex_init(&s->lock);
1166 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
1167
1168 /* Repair image if dirty */
1169 if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only &&
1170 (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
1171 BdrvCheckResult result = {0};
1172
1173 ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
1174 if (ret < 0) {
1175 error_setg_errno(errp, -ret, "Could not repair dirty image");
1176 goto fail;
1177 }
1178 }
1179
1180 #ifdef DEBUG_ALLOC
1181 {
1182 BdrvCheckResult result = {0};
1183 qcow2_check_refcounts(bs, &result, 0);
1184 }
1185 #endif
1186 return ret;
1187
1188 fail:
1189 g_free(s->unknown_header_fields);
1190 cleanup_unknown_header_ext(bs);
1191 qcow2_free_snapshots(bs);
1192 qcow2_refcount_close(bs);
1193 qemu_vfree(s->l1_table);
1194 /* else pre-write overlap checks in cache_destroy may crash */
1195 s->l1_table = NULL;
1196 cache_clean_timer_del(bs);
1197 if (s->l2_table_cache) {
1198 qcow2_cache_destroy(bs, s->l2_table_cache);
1199 }
1200 if (s->refcount_block_cache) {
1201 qcow2_cache_destroy(bs, s->refcount_block_cache);
1202 }
1203 g_free(s->cluster_cache);
1204 qemu_vfree(s->cluster_data);
1205 return ret;
1206 }
1207
1208 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
1209 Error **errp)
1210 {
1211 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
1212 false, errp);
1213 if (!bs->file) {
1214 return -EINVAL;
1215 }
1216
1217 return qcow2_do_open(bs, options, flags, errp);
1218 }
1219
1220 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1221 {
1222 BDRVQcow2State *s = bs->opaque;
1223
1224 if (bs->encrypted) {
1225 /* Encryption works on a sector granularity */
1226 bs->bl.request_alignment = BDRV_SECTOR_SIZE;
1227 }
1228 bs->bl.pwrite_zeroes_alignment = s->cluster_size;
1229 bs->bl.pdiscard_alignment = s->cluster_size;
1230 }
1231
1232 static int qcow2_set_key(BlockDriverState *bs, const char *key)
1233 {
1234 BDRVQcow2State *s = bs->opaque;
1235 uint8_t keybuf[16];
1236 int len, i;
1237 Error *err = NULL;
1238
1239 memset(keybuf, 0, 16);
1240 len = strlen(key);
1241 if (len > 16)
1242 len = 16;
1243 /* XXX: we could compress the chars to 7 bits to increase
1244 entropy */
1245 for(i = 0;i < len;i++) {
1246 keybuf[i] = key[i];
1247 }
1248 assert(bs->encrypted);
1249
1250 qcrypto_cipher_free(s->cipher);
1251 s->cipher = qcrypto_cipher_new(
1252 QCRYPTO_CIPHER_ALG_AES_128,
1253 QCRYPTO_CIPHER_MODE_CBC,
1254 keybuf, G_N_ELEMENTS(keybuf),
1255 &err);
1256
1257 if (!s->cipher) {
1258 /* XXX would be nice if errors in this method could
1259 * be properly propagate to the caller. Would need
1260 * the bdrv_set_key() API signature to be fixed. */
1261 error_free(err);
1262 return -1;
1263 }
1264 return 0;
1265 }
1266
1267 static int qcow2_reopen_prepare(BDRVReopenState *state,
1268 BlockReopenQueue *queue, Error **errp)
1269 {
1270 Qcow2ReopenState *r;
1271 int ret;
1272
1273 r = g_new0(Qcow2ReopenState, 1);
1274 state->opaque = r;
1275
1276 ret = qcow2_update_options_prepare(state->bs, r, state->options,
1277 state->flags, errp);
1278 if (ret < 0) {
1279 goto fail;
1280 }
1281
1282 /* We need to write out any unwritten data if we reopen read-only. */
1283 if ((state->flags & BDRV_O_RDWR) == 0) {
1284 ret = bdrv_flush(state->bs);
1285 if (ret < 0) {
1286 goto fail;
1287 }
1288
1289 ret = qcow2_mark_clean(state->bs);
1290 if (ret < 0) {
1291 goto fail;
1292 }
1293 }
1294
1295 return 0;
1296
1297 fail:
1298 qcow2_update_options_abort(state->bs, r);
1299 g_free(r);
1300 return ret;
1301 }
1302
1303 static void qcow2_reopen_commit(BDRVReopenState *state)
1304 {
1305 qcow2_update_options_commit(state->bs, state->opaque);
1306 g_free(state->opaque);
1307 }
1308
1309 static void qcow2_reopen_abort(BDRVReopenState *state)
1310 {
1311 qcow2_update_options_abort(state->bs, state->opaque);
1312 g_free(state->opaque);
1313 }
1314
1315 static void qcow2_join_options(QDict *options, QDict *old_options)
1316 {
1317 bool has_new_overlap_template =
1318 qdict_haskey(options, QCOW2_OPT_OVERLAP) ||
1319 qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE);
1320 bool has_new_total_cache_size =
1321 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE);
1322 bool has_all_cache_options;
1323
1324 /* New overlap template overrides all old overlap options */
1325 if (has_new_overlap_template) {
1326 qdict_del(old_options, QCOW2_OPT_OVERLAP);
1327 qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE);
1328 qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER);
1329 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1);
1330 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2);
1331 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE);
1332 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK);
1333 qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE);
1334 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1);
1335 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2);
1336 }
1337
1338 /* New total cache size overrides all old options */
1339 if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) {
1340 qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE);
1341 qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1342 }
1343
1344 qdict_join(options, old_options, false);
1345
1346 /*
1347 * If after merging all cache size options are set, an old total size is
1348 * overwritten. Do keep all options, however, if all three are new. The
1349 * resulting error message is what we want to happen.
1350 */
1351 has_all_cache_options =
1352 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) ||
1353 qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) ||
1354 qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1355
1356 if (has_all_cache_options && !has_new_total_cache_size) {
1357 qdict_del(options, QCOW2_OPT_CACHE_SIZE);
1358 }
1359 }
1360
1361 static int64_t coroutine_fn qcow2_co_get_block_status(BlockDriverState *bs,
1362 int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
1363 {
1364 BDRVQcow2State *s = bs->opaque;
1365 uint64_t cluster_offset;
1366 int index_in_cluster, ret;
1367 unsigned int bytes;
1368 int64_t status = 0;
1369
1370 bytes = MIN(INT_MAX, nb_sectors * BDRV_SECTOR_SIZE);
1371 qemu_co_mutex_lock(&s->lock);
1372 ret = qcow2_get_cluster_offset(bs, sector_num << 9, &bytes,
1373 &cluster_offset);
1374 qemu_co_mutex_unlock(&s->lock);
1375 if (ret < 0) {
1376 return ret;
1377 }
1378
1379 *pnum = bytes >> BDRV_SECTOR_BITS;
1380
1381 if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED &&
1382 !s->cipher) {
1383 index_in_cluster = sector_num & (s->cluster_sectors - 1);
1384 cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
1385 *file = bs->file->bs;
1386 status |= BDRV_BLOCK_OFFSET_VALID | cluster_offset;
1387 }
1388 if (ret == QCOW2_CLUSTER_ZERO) {
1389 status |= BDRV_BLOCK_ZERO;
1390 } else if (ret != QCOW2_CLUSTER_UNALLOCATED) {
1391 status |= BDRV_BLOCK_DATA;
1392 }
1393 return status;
1394 }
1395
1396 /* handle reading after the end of the backing file */
1397 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
1398 int64_t offset, int bytes)
1399 {
1400 uint64_t bs_size = bs->total_sectors * BDRV_SECTOR_SIZE;
1401 int n1;
1402
1403 if ((offset + bytes) <= bs_size) {
1404 return bytes;
1405 }
1406
1407 if (offset >= bs_size) {
1408 n1 = 0;
1409 } else {
1410 n1 = bs_size - offset;
1411 }
1412
1413 qemu_iovec_memset(qiov, n1, 0, bytes - n1);
1414
1415 return n1;
1416 }
1417
1418 static coroutine_fn int qcow2_co_preadv(BlockDriverState *bs, uint64_t offset,
1419 uint64_t bytes, QEMUIOVector *qiov,
1420 int flags)
1421 {
1422 BDRVQcow2State *s = bs->opaque;
1423 int offset_in_cluster, n1;
1424 int ret;
1425 unsigned int cur_bytes; /* number of bytes in current iteration */
1426 uint64_t cluster_offset = 0;
1427 uint64_t bytes_done = 0;
1428 QEMUIOVector hd_qiov;
1429 uint8_t *cluster_data = NULL;
1430
1431 qemu_iovec_init(&hd_qiov, qiov->niov);
1432
1433 qemu_co_mutex_lock(&s->lock);
1434
1435 while (bytes != 0) {
1436
1437 /* prepare next request */
1438 cur_bytes = MIN(bytes, INT_MAX);
1439 if (s->cipher) {
1440 cur_bytes = MIN(cur_bytes,
1441 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1442 }
1443
1444 ret = qcow2_get_cluster_offset(bs, offset, &cur_bytes, &cluster_offset);
1445 if (ret < 0) {
1446 goto fail;
1447 }
1448
1449 offset_in_cluster = offset_into_cluster(s, offset);
1450
1451 qemu_iovec_reset(&hd_qiov);
1452 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes);
1453
1454 switch (ret) {
1455 case QCOW2_CLUSTER_UNALLOCATED:
1456
1457 if (bs->backing) {
1458 /* read from the base image */
1459 n1 = qcow2_backing_read1(bs->backing->bs, &hd_qiov,
1460 offset, cur_bytes);
1461 if (n1 > 0) {
1462 QEMUIOVector local_qiov;
1463
1464 qemu_iovec_init(&local_qiov, hd_qiov.niov);
1465 qemu_iovec_concat(&local_qiov, &hd_qiov, 0, n1);
1466
1467 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1468 qemu_co_mutex_unlock(&s->lock);
1469 ret = bdrv_co_preadv(bs->backing, offset, n1,
1470 &local_qiov, 0);
1471 qemu_co_mutex_lock(&s->lock);
1472
1473 qemu_iovec_destroy(&local_qiov);
1474
1475 if (ret < 0) {
1476 goto fail;
1477 }
1478 }
1479 } else {
1480 /* Note: in this case, no need to wait */
1481 qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes);
1482 }
1483 break;
1484
1485 case QCOW2_CLUSTER_ZERO:
1486 qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes);
1487 break;
1488
1489 case QCOW2_CLUSTER_COMPRESSED:
1490 /* add AIO support for compressed blocks ? */
1491 ret = qcow2_decompress_cluster(bs, cluster_offset);
1492 if (ret < 0) {
1493 goto fail;
1494 }
1495
1496 qemu_iovec_from_buf(&hd_qiov, 0,
1497 s->cluster_cache + offset_in_cluster,
1498 cur_bytes);
1499 break;
1500
1501 case QCOW2_CLUSTER_NORMAL:
1502 if ((cluster_offset & 511) != 0) {
1503 ret = -EIO;
1504 goto fail;
1505 }
1506
1507 if (bs->encrypted) {
1508 assert(s->cipher);
1509
1510 /*
1511 * For encrypted images, read everything into a temporary
1512 * contiguous buffer on which the AES functions can work.
1513 */
1514 if (!cluster_data) {
1515 cluster_data =
1516 qemu_try_blockalign(bs->file->bs,
1517 QCOW_MAX_CRYPT_CLUSTERS
1518 * s->cluster_size);
1519 if (cluster_data == NULL) {
1520 ret = -ENOMEM;
1521 goto fail;
1522 }
1523 }
1524
1525 assert(cur_bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1526 qemu_iovec_reset(&hd_qiov);
1527 qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes);
1528 }
1529
1530 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
1531 qemu_co_mutex_unlock(&s->lock);
1532 ret = bdrv_co_preadv(bs->file,
1533 cluster_offset + offset_in_cluster,
1534 cur_bytes, &hd_qiov, 0);
1535 qemu_co_mutex_lock(&s->lock);
1536 if (ret < 0) {
1537 goto fail;
1538 }
1539 if (bs->encrypted) {
1540 assert(s->cipher);
1541 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
1542 assert((cur_bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
1543 Error *err = NULL;
1544 if (qcow2_encrypt_sectors(s, offset >> BDRV_SECTOR_BITS,
1545 cluster_data, cluster_data,
1546 cur_bytes >> BDRV_SECTOR_BITS,
1547 false, &err) < 0) {
1548 error_free(err);
1549 ret = -EIO;
1550 goto fail;
1551 }
1552 qemu_iovec_from_buf(qiov, bytes_done, cluster_data, cur_bytes);
1553 }
1554 break;
1555
1556 default:
1557 g_assert_not_reached();
1558 ret = -EIO;
1559 goto fail;
1560 }
1561
1562 bytes -= cur_bytes;
1563 offset += cur_bytes;
1564 bytes_done += cur_bytes;
1565 }
1566 ret = 0;
1567
1568 fail:
1569 qemu_co_mutex_unlock(&s->lock);
1570
1571 qemu_iovec_destroy(&hd_qiov);
1572 qemu_vfree(cluster_data);
1573
1574 return ret;
1575 }
1576
1577 static coroutine_fn int qcow2_co_pwritev(BlockDriverState *bs, uint64_t offset,
1578 uint64_t bytes, QEMUIOVector *qiov,
1579 int flags)
1580 {
1581 BDRVQcow2State *s = bs->opaque;
1582 int offset_in_cluster;
1583 int ret;
1584 unsigned int cur_bytes; /* number of sectors in current iteration */
1585 uint64_t cluster_offset;
1586 QEMUIOVector hd_qiov;
1587 uint64_t bytes_done = 0;
1588 uint8_t *cluster_data = NULL;
1589 QCowL2Meta *l2meta = NULL;
1590
1591 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes);
1592
1593 qemu_iovec_init(&hd_qiov, qiov->niov);
1594
1595 s->cluster_cache_offset = -1; /* disable compressed cache */
1596
1597 qemu_co_mutex_lock(&s->lock);
1598
1599 while (bytes != 0) {
1600
1601 l2meta = NULL;
1602
1603 trace_qcow2_writev_start_part(qemu_coroutine_self());
1604 offset_in_cluster = offset_into_cluster(s, offset);
1605 cur_bytes = MIN(bytes, INT_MAX);
1606 if (bs->encrypted) {
1607 cur_bytes = MIN(cur_bytes,
1608 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
1609 - offset_in_cluster);
1610 }
1611
1612 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
1613 &cluster_offset, &l2meta);
1614 if (ret < 0) {
1615 goto fail;
1616 }
1617
1618 assert((cluster_offset & 511) == 0);
1619
1620 qemu_iovec_reset(&hd_qiov);
1621 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes);
1622
1623 if (bs->encrypted) {
1624 Error *err = NULL;
1625 assert(s->cipher);
1626 if (!cluster_data) {
1627 cluster_data = qemu_try_blockalign(bs->file->bs,
1628 QCOW_MAX_CRYPT_CLUSTERS
1629 * s->cluster_size);
1630 if (cluster_data == NULL) {
1631 ret = -ENOMEM;
1632 goto fail;
1633 }
1634 }
1635
1636 assert(hd_qiov.size <=
1637 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1638 qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size);
1639
1640 if (qcow2_encrypt_sectors(s, offset >> BDRV_SECTOR_BITS,
1641 cluster_data, cluster_data,
1642 cur_bytes >>BDRV_SECTOR_BITS,
1643 true, &err) < 0) {
1644 error_free(err);
1645 ret = -EIO;
1646 goto fail;
1647 }
1648
1649 qemu_iovec_reset(&hd_qiov);
1650 qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes);
1651 }
1652
1653 ret = qcow2_pre_write_overlap_check(bs, 0,
1654 cluster_offset + offset_in_cluster, cur_bytes);
1655 if (ret < 0) {
1656 goto fail;
1657 }
1658
1659 qemu_co_mutex_unlock(&s->lock);
1660 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
1661 trace_qcow2_writev_data(qemu_coroutine_self(),
1662 cluster_offset + offset_in_cluster);
1663 ret = bdrv_co_pwritev(bs->file,
1664 cluster_offset + offset_in_cluster,
1665 cur_bytes, &hd_qiov, 0);
1666 qemu_co_mutex_lock(&s->lock);
1667 if (ret < 0) {
1668 goto fail;
1669 }
1670
1671 while (l2meta != NULL) {
1672 QCowL2Meta *next;
1673
1674 ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
1675 if (ret < 0) {
1676 goto fail;
1677 }
1678
1679 /* Take the request off the list of running requests */
1680 if (l2meta->nb_clusters != 0) {
1681 QLIST_REMOVE(l2meta, next_in_flight);
1682 }
1683
1684 qemu_co_queue_restart_all(&l2meta->dependent_requests);
1685
1686 next = l2meta->next;
1687 g_free(l2meta);
1688 l2meta = next;
1689 }
1690
1691 bytes -= cur_bytes;
1692 offset += cur_bytes;
1693 bytes_done += cur_bytes;
1694 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes);
1695 }
1696 ret = 0;
1697
1698 fail:
1699 qemu_co_mutex_unlock(&s->lock);
1700
1701 while (l2meta != NULL) {
1702 QCowL2Meta *next;
1703
1704 if (l2meta->nb_clusters != 0) {
1705 QLIST_REMOVE(l2meta, next_in_flight);
1706 }
1707 qemu_co_queue_restart_all(&l2meta->dependent_requests);
1708
1709 next = l2meta->next;
1710 g_free(l2meta);
1711 l2meta = next;
1712 }
1713
1714 qemu_iovec_destroy(&hd_qiov);
1715 qemu_vfree(cluster_data);
1716 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
1717
1718 return ret;
1719 }
1720
1721 static int qcow2_inactivate(BlockDriverState *bs)
1722 {
1723 BDRVQcow2State *s = bs->opaque;
1724 int ret, result = 0;
1725
1726 ret = qcow2_cache_flush(bs, s->l2_table_cache);
1727 if (ret) {
1728 result = ret;
1729 error_report("Failed to flush the L2 table cache: %s",
1730 strerror(-ret));
1731 }
1732
1733 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
1734 if (ret) {
1735 result = ret;
1736 error_report("Failed to flush the refcount block cache: %s",
1737 strerror(-ret));
1738 }
1739
1740 if (result == 0) {
1741 qcow2_mark_clean(bs);
1742 }
1743
1744 return result;
1745 }
1746
1747 static void qcow2_close(BlockDriverState *bs)
1748 {
1749 BDRVQcow2State *s = bs->opaque;
1750 qemu_vfree(s->l1_table);
1751 /* else pre-write overlap checks in cache_destroy may crash */
1752 s->l1_table = NULL;
1753
1754 if (!(s->flags & BDRV_O_INACTIVE)) {
1755 qcow2_inactivate(bs);
1756 }
1757
1758 cache_clean_timer_del(bs);
1759 qcow2_cache_destroy(bs, s->l2_table_cache);
1760 qcow2_cache_destroy(bs, s->refcount_block_cache);
1761
1762 qcrypto_cipher_free(s->cipher);
1763 s->cipher = NULL;
1764
1765 g_free(s->unknown_header_fields);
1766 cleanup_unknown_header_ext(bs);
1767
1768 g_free(s->image_backing_file);
1769 g_free(s->image_backing_format);
1770
1771 g_free(s->cluster_cache);
1772 qemu_vfree(s->cluster_data);
1773 qcow2_refcount_close(bs);
1774 qcow2_free_snapshots(bs);
1775 }
1776
1777 static void qcow2_invalidate_cache(BlockDriverState *bs, Error **errp)
1778 {
1779 BDRVQcow2State *s = bs->opaque;
1780 int flags = s->flags;
1781 QCryptoCipher *cipher = NULL;
1782 QDict *options;
1783 Error *local_err = NULL;
1784 int ret;
1785
1786 /*
1787 * Backing files are read-only which makes all of their metadata immutable,
1788 * that means we don't have to worry about reopening them here.
1789 */
1790
1791 cipher = s->cipher;
1792 s->cipher = NULL;
1793
1794 qcow2_close(bs);
1795
1796 memset(s, 0, sizeof(BDRVQcow2State));
1797 options = qdict_clone_shallow(bs->options);
1798
1799 flags &= ~BDRV_O_INACTIVE;
1800 ret = qcow2_do_open(bs, options, flags, &local_err);
1801 QDECREF(options);
1802 if (local_err) {
1803 error_propagate(errp, local_err);
1804 error_prepend(errp, "Could not reopen qcow2 layer: ");
1805 bs->drv = NULL;
1806 return;
1807 } else if (ret < 0) {
1808 error_setg_errno(errp, -ret, "Could not reopen qcow2 layer");
1809 bs->drv = NULL;
1810 return;
1811 }
1812
1813 s->cipher = cipher;
1814 }
1815
1816 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
1817 size_t len, size_t buflen)
1818 {
1819 QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
1820 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
1821
1822 if (buflen < ext_len) {
1823 return -ENOSPC;
1824 }
1825
1826 *ext_backing_fmt = (QCowExtension) {
1827 .magic = cpu_to_be32(magic),
1828 .len = cpu_to_be32(len),
1829 };
1830
1831 if (len) {
1832 memcpy(buf + sizeof(QCowExtension), s, len);
1833 }
1834
1835 return ext_len;
1836 }
1837
1838 /*
1839 * Updates the qcow2 header, including the variable length parts of it, i.e.
1840 * the backing file name and all extensions. qcow2 was not designed to allow
1841 * such changes, so if we run out of space (we can only use the first cluster)
1842 * this function may fail.
1843 *
1844 * Returns 0 on success, -errno in error cases.
1845 */
1846 int qcow2_update_header(BlockDriverState *bs)
1847 {
1848 BDRVQcow2State *s = bs->opaque;
1849 QCowHeader *header;
1850 char *buf;
1851 size_t buflen = s->cluster_size;
1852 int ret;
1853 uint64_t total_size;
1854 uint32_t refcount_table_clusters;
1855 size_t header_length;
1856 Qcow2UnknownHeaderExtension *uext;
1857
1858 buf = qemu_blockalign(bs, buflen);
1859
1860 /* Header structure */
1861 header = (QCowHeader*) buf;
1862
1863 if (buflen < sizeof(*header)) {
1864 ret = -ENOSPC;
1865 goto fail;
1866 }
1867
1868 header_length = sizeof(*header) + s->unknown_header_fields_size;
1869 total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
1870 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
1871
1872 *header = (QCowHeader) {
1873 /* Version 2 fields */
1874 .magic = cpu_to_be32(QCOW_MAGIC),
1875 .version = cpu_to_be32(s->qcow_version),
1876 .backing_file_offset = 0,
1877 .backing_file_size = 0,
1878 .cluster_bits = cpu_to_be32(s->cluster_bits),
1879 .size = cpu_to_be64(total_size),
1880 .crypt_method = cpu_to_be32(s->crypt_method_header),
1881 .l1_size = cpu_to_be32(s->l1_size),
1882 .l1_table_offset = cpu_to_be64(s->l1_table_offset),
1883 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset),
1884 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
1885 .nb_snapshots = cpu_to_be32(s->nb_snapshots),
1886 .snapshots_offset = cpu_to_be64(s->snapshots_offset),
1887
1888 /* Version 3 fields */
1889 .incompatible_features = cpu_to_be64(s->incompatible_features),
1890 .compatible_features = cpu_to_be64(s->compatible_features),
1891 .autoclear_features = cpu_to_be64(s->autoclear_features),
1892 .refcount_order = cpu_to_be32(s->refcount_order),
1893 .header_length = cpu_to_be32(header_length),
1894 };
1895
1896 /* For older versions, write a shorter header */
1897 switch (s->qcow_version) {
1898 case 2:
1899 ret = offsetof(QCowHeader, incompatible_features);
1900 break;
1901 case 3:
1902 ret = sizeof(*header);
1903 break;
1904 default:
1905 ret = -EINVAL;
1906 goto fail;
1907 }
1908
1909 buf += ret;
1910 buflen -= ret;
1911 memset(buf, 0, buflen);
1912
1913 /* Preserve any unknown field in the header */
1914 if (s->unknown_header_fields_size) {
1915 if (buflen < s->unknown_header_fields_size) {
1916 ret = -ENOSPC;
1917 goto fail;
1918 }
1919
1920 memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
1921 buf += s->unknown_header_fields_size;
1922 buflen -= s->unknown_header_fields_size;
1923 }
1924
1925 /* Backing file format header extension */
1926 if (s->image_backing_format) {
1927 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
1928 s->image_backing_format,
1929 strlen(s->image_backing_format),
1930 buflen);
1931 if (ret < 0) {
1932 goto fail;
1933 }
1934
1935 buf += ret;
1936 buflen -= ret;
1937 }
1938
1939 /* Feature table */
1940 if (s->qcow_version >= 3) {
1941 Qcow2Feature features[] = {
1942 {
1943 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1944 .bit = QCOW2_INCOMPAT_DIRTY_BITNR,
1945 .name = "dirty bit",
1946 },
1947 {
1948 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1949 .bit = QCOW2_INCOMPAT_CORRUPT_BITNR,
1950 .name = "corrupt bit",
1951 },
1952 {
1953 .type = QCOW2_FEAT_TYPE_COMPATIBLE,
1954 .bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
1955 .name = "lazy refcounts",
1956 },
1957 };
1958
1959 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
1960 features, sizeof(features), buflen);
1961 if (ret < 0) {
1962 goto fail;
1963 }
1964 buf += ret;
1965 buflen -= ret;
1966 }
1967
1968 /* Keep unknown header extensions */
1969 QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
1970 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
1971 if (ret < 0) {
1972 goto fail;
1973 }
1974
1975 buf += ret;
1976 buflen -= ret;
1977 }
1978
1979 /* End of header extensions */
1980 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
1981 if (ret < 0) {
1982 goto fail;
1983 }
1984
1985 buf += ret;
1986 buflen -= ret;
1987
1988 /* Backing file name */
1989 if (s->image_backing_file) {
1990 size_t backing_file_len = strlen(s->image_backing_file);
1991
1992 if (buflen < backing_file_len) {
1993 ret = -ENOSPC;
1994 goto fail;
1995 }
1996
1997 /* Using strncpy is ok here, since buf is not NUL-terminated. */
1998 strncpy(buf, s->image_backing_file, buflen);
1999
2000 header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
2001 header->backing_file_size = cpu_to_be32(backing_file_len);
2002 }
2003
2004 /* Write the new header */
2005 ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
2006 if (ret < 0) {
2007 goto fail;
2008 }
2009
2010 ret = 0;
2011 fail:
2012 qemu_vfree(header);
2013 return ret;
2014 }
2015
2016 static int qcow2_change_backing_file(BlockDriverState *bs,
2017 const char *backing_file, const char *backing_fmt)
2018 {
2019 BDRVQcow2State *s = bs->opaque;
2020
2021 if (backing_file && strlen(backing_file) > 1023) {
2022 return -EINVAL;
2023 }
2024
2025 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
2026 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
2027
2028 g_free(s->image_backing_file);
2029 g_free(s->image_backing_format);
2030
2031 s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
2032 s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
2033
2034 return qcow2_update_header(bs);
2035 }
2036
2037 static int preallocate(BlockDriverState *bs)
2038 {
2039 uint64_t bytes;
2040 uint64_t offset;
2041 uint64_t host_offset = 0;
2042 unsigned int cur_bytes;
2043 int ret;
2044 QCowL2Meta *meta;
2045
2046 bytes = bdrv_getlength(bs);
2047 offset = 0;
2048
2049 while (bytes) {
2050 cur_bytes = MIN(bytes, INT_MAX);
2051 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
2052 &host_offset, &meta);
2053 if (ret < 0) {
2054 return ret;
2055 }
2056
2057 while (meta) {
2058 QCowL2Meta *next = meta->next;
2059
2060 ret = qcow2_alloc_cluster_link_l2(bs, meta);
2061 if (ret < 0) {
2062 qcow2_free_any_clusters(bs, meta->alloc_offset,
2063 meta->nb_clusters, QCOW2_DISCARD_NEVER);
2064 return ret;
2065 }
2066
2067 /* There are no dependent requests, but we need to remove our
2068 * request from the list of in-flight requests */
2069 QLIST_REMOVE(meta, next_in_flight);
2070
2071 g_free(meta);
2072 meta = next;
2073 }
2074
2075 /* TODO Preallocate data if requested */
2076
2077 bytes -= cur_bytes;
2078 offset += cur_bytes;
2079 }
2080
2081 /*
2082 * It is expected that the image file is large enough to actually contain
2083 * all of the allocated clusters (otherwise we get failing reads after
2084 * EOF). Extend the image to the last allocated sector.
2085 */
2086 if (host_offset != 0) {
2087 uint8_t data = 0;
2088 ret = bdrv_pwrite(bs->file, (host_offset + cur_bytes) - 1,
2089 &data, 1);
2090 if (ret < 0) {
2091 return ret;
2092 }
2093 }
2094
2095 return 0;
2096 }
2097
2098 static int qcow2_create2(const char *filename, int64_t total_size,
2099 const char *backing_file, const char *backing_format,
2100 int flags, size_t cluster_size, PreallocMode prealloc,
2101 QemuOpts *opts, int version, int refcount_order,
2102 Error **errp)
2103 {
2104 int cluster_bits;
2105 QDict *options;
2106
2107 /* Calculate cluster_bits */
2108 cluster_bits = ctz32(cluster_size);
2109 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
2110 (1 << cluster_bits) != cluster_size)
2111 {
2112 error_setg(errp, "Cluster size must be a power of two between %d and "
2113 "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
2114 return -EINVAL;
2115 }
2116
2117 /*
2118 * Open the image file and write a minimal qcow2 header.
2119 *
2120 * We keep things simple and start with a zero-sized image. We also
2121 * do without refcount blocks or a L1 table for now. We'll fix the
2122 * inconsistency later.
2123 *
2124 * We do need a refcount table because growing the refcount table means
2125 * allocating two new refcount blocks - the seconds of which would be at
2126 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
2127 * size for any qcow2 image.
2128 */
2129 BlockBackend *blk;
2130 QCowHeader *header;
2131 uint64_t* refcount_table;
2132 Error *local_err = NULL;
2133 int ret;
2134
2135 if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
2136 /* Note: The following calculation does not need to be exact; if it is a
2137 * bit off, either some bytes will be "leaked" (which is fine) or we
2138 * will need to increase the file size by some bytes (which is fine,
2139 * too, as long as the bulk is allocated here). Therefore, using
2140 * floating point arithmetic is fine. */
2141 int64_t meta_size = 0;
2142 uint64_t nreftablee, nrefblocke, nl1e, nl2e;
2143 int64_t aligned_total_size = align_offset(total_size, cluster_size);
2144 int refblock_bits, refblock_size;
2145 /* refcount entry size in bytes */
2146 double rces = (1 << refcount_order) / 8.;
2147
2148 /* see qcow2_open() */
2149 refblock_bits = cluster_bits - (refcount_order - 3);
2150 refblock_size = 1 << refblock_bits;
2151
2152 /* header: 1 cluster */
2153 meta_size += cluster_size;
2154
2155 /* total size of L2 tables */
2156 nl2e = aligned_total_size / cluster_size;
2157 nl2e = align_offset(nl2e, cluster_size / sizeof(uint64_t));
2158 meta_size += nl2e * sizeof(uint64_t);
2159
2160 /* total size of L1 tables */
2161 nl1e = nl2e * sizeof(uint64_t) / cluster_size;
2162 nl1e = align_offset(nl1e, cluster_size / sizeof(uint64_t));
2163 meta_size += nl1e * sizeof(uint64_t);
2164
2165 /* total size of refcount blocks
2166 *
2167 * note: every host cluster is reference-counted, including metadata
2168 * (even refcount blocks are recursively included).
2169 * Let:
2170 * a = total_size (this is the guest disk size)
2171 * m = meta size not including refcount blocks and refcount tables
2172 * c = cluster size
2173 * y1 = number of refcount blocks entries
2174 * y2 = meta size including everything
2175 * rces = refcount entry size in bytes
2176 * then,
2177 * y1 = (y2 + a)/c
2178 * y2 = y1 * rces + y1 * rces * sizeof(u64) / c + m
2179 * we can get y1:
2180 * y1 = (a + m) / (c - rces - rces * sizeof(u64) / c)
2181 */
2182 nrefblocke = (aligned_total_size + meta_size + cluster_size)
2183 / (cluster_size - rces - rces * sizeof(uint64_t)
2184 / cluster_size);
2185 meta_size += DIV_ROUND_UP(nrefblocke, refblock_size) * cluster_size;
2186
2187 /* total size of refcount tables */
2188 nreftablee = nrefblocke / refblock_size;
2189 nreftablee = align_offset(nreftablee, cluster_size / sizeof(uint64_t));
2190 meta_size += nreftablee * sizeof(uint64_t);
2191
2192 qemu_opt_set_number(opts, BLOCK_OPT_SIZE,
2193 aligned_total_size + meta_size, &error_abort);
2194 qemu_opt_set(opts, BLOCK_OPT_PREALLOC, PreallocMode_lookup[prealloc],
2195 &error_abort);
2196 }
2197
2198 ret = bdrv_create_file(filename, opts, &local_err);
2199 if (ret < 0) {
2200 error_propagate(errp, local_err);
2201 return ret;
2202 }
2203
2204 blk = blk_new_open(filename, NULL, NULL,
2205 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
2206 &local_err);
2207 if (blk == NULL) {
2208 error_propagate(errp, local_err);
2209 return -EIO;
2210 }
2211
2212 blk_set_allow_write_beyond_eof(blk, true);
2213
2214 /* Write the header */
2215 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
2216 header = g_malloc0(cluster_size);
2217 *header = (QCowHeader) {
2218 .magic = cpu_to_be32(QCOW_MAGIC),
2219 .version = cpu_to_be32(version),
2220 .cluster_bits = cpu_to_be32(cluster_bits),
2221 .size = cpu_to_be64(0),
2222 .l1_table_offset = cpu_to_be64(0),
2223 .l1_size = cpu_to_be32(0),
2224 .refcount_table_offset = cpu_to_be64(cluster_size),
2225 .refcount_table_clusters = cpu_to_be32(1),
2226 .refcount_order = cpu_to_be32(refcount_order),
2227 .header_length = cpu_to_be32(sizeof(*header)),
2228 };
2229
2230 if (flags & BLOCK_FLAG_ENCRYPT) {
2231 header->crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
2232 } else {
2233 header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
2234 }
2235
2236 if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) {
2237 header->compatible_features |=
2238 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
2239 }
2240
2241 ret = blk_pwrite(blk, 0, header, cluster_size, 0);
2242 g_free(header);
2243 if (ret < 0) {
2244 error_setg_errno(errp, -ret, "Could not write qcow2 header");
2245 goto out;
2246 }
2247
2248 /* Write a refcount table with one refcount block */
2249 refcount_table = g_malloc0(2 * cluster_size);
2250 refcount_table[0] = cpu_to_be64(2 * cluster_size);
2251 ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0);
2252 g_free(refcount_table);
2253
2254 if (ret < 0) {
2255 error_setg_errno(errp, -ret, "Could not write refcount table");
2256 goto out;
2257 }
2258
2259 blk_unref(blk);
2260 blk = NULL;
2261
2262 /*
2263 * And now open the image and make it consistent first (i.e. increase the
2264 * refcount of the cluster that is occupied by the header and the refcount
2265 * table)
2266 */
2267 options = qdict_new();
2268 qdict_put(options, "driver", qstring_from_str("qcow2"));
2269 blk = blk_new_open(filename, NULL, options,
2270 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH,
2271 &local_err);
2272 if (blk == NULL) {
2273 error_propagate(errp, local_err);
2274 ret = -EIO;
2275 goto out;
2276 }
2277
2278 ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size);
2279 if (ret < 0) {
2280 error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
2281 "header and refcount table");
2282 goto out;
2283
2284 } else if (ret != 0) {
2285 error_report("Huh, first cluster in empty image is already in use?");
2286 abort();
2287 }
2288
2289 /* Create a full header (including things like feature table) */
2290 ret = qcow2_update_header(blk_bs(blk));
2291 if (ret < 0) {
2292 error_setg_errno(errp, -ret, "Could not update qcow2 header");
2293 goto out;
2294 }
2295
2296 /* Okay, now that we have a valid image, let's give it the right size */
2297 ret = blk_truncate(blk, total_size);
2298 if (ret < 0) {
2299 error_setg_errno(errp, -ret, "Could not resize image");
2300 goto out;
2301 }
2302
2303 /* Want a backing file? There you go.*/
2304 if (backing_file) {
2305 ret = bdrv_change_backing_file(blk_bs(blk), backing_file, backing_format);
2306 if (ret < 0) {
2307 error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
2308 "with format '%s'", backing_file, backing_format);
2309 goto out;
2310 }
2311 }
2312
2313 /* And if we're supposed to preallocate metadata, do that now */
2314 if (prealloc != PREALLOC_MODE_OFF) {
2315 BDRVQcow2State *s = blk_bs(blk)->opaque;
2316 qemu_co_mutex_lock(&s->lock);
2317 ret = preallocate(blk_bs(blk));
2318 qemu_co_mutex_unlock(&s->lock);
2319 if (ret < 0) {
2320 error_setg_errno(errp, -ret, "Could not preallocate metadata");
2321 goto out;
2322 }
2323 }
2324
2325 blk_unref(blk);
2326 blk = NULL;
2327
2328 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning */
2329 options = qdict_new();
2330 qdict_put(options, "driver", qstring_from_str("qcow2"));
2331 blk = blk_new_open(filename, NULL, options,
2332 BDRV_O_RDWR | BDRV_O_NO_BACKING, &local_err);
2333 if (blk == NULL) {
2334 error_propagate(errp, local_err);
2335 ret = -EIO;
2336 goto out;
2337 }
2338
2339 ret = 0;
2340 out:
2341 if (blk) {
2342 blk_unref(blk);
2343 }
2344 return ret;
2345 }
2346
2347 static int qcow2_create(const char *filename, QemuOpts *opts, Error **errp)
2348 {
2349 char *backing_file = NULL;
2350 char *backing_fmt = NULL;
2351 char *buf = NULL;
2352 uint64_t size = 0;
2353 int flags = 0;
2354 size_t cluster_size = DEFAULT_CLUSTER_SIZE;
2355 PreallocMode prealloc;
2356 int version = 3;
2357 uint64_t refcount_bits = 16;
2358 int refcount_order;
2359 Error *local_err = NULL;
2360 int ret;
2361
2362 /* Read out options */
2363 size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2364 BDRV_SECTOR_SIZE);
2365 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2366 backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
2367 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
2368 flags |= BLOCK_FLAG_ENCRYPT;
2369 }
2370 cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
2371 DEFAULT_CLUSTER_SIZE);
2372 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
2373 prealloc = qapi_enum_parse(PreallocMode_lookup, buf,
2374 PREALLOC_MODE__MAX, PREALLOC_MODE_OFF,
2375 &local_err);
2376 if (local_err) {
2377 error_propagate(errp, local_err);
2378 ret = -EINVAL;
2379 goto finish;
2380 }
2381 g_free(buf);
2382 buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
2383 if (!buf) {
2384 /* keep the default */
2385 } else if (!strcmp(buf, "0.10")) {
2386 version = 2;
2387 } else if (!strcmp(buf, "1.1")) {
2388 version = 3;
2389 } else {
2390 error_setg(errp, "Invalid compatibility level: '%s'", buf);
2391 ret = -EINVAL;
2392 goto finish;
2393 }
2394
2395 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_LAZY_REFCOUNTS, false)) {
2396 flags |= BLOCK_FLAG_LAZY_REFCOUNTS;
2397 }
2398
2399 if (backing_file && prealloc != PREALLOC_MODE_OFF) {
2400 error_setg(errp, "Backing file and preallocation cannot be used at "
2401 "the same time");
2402 ret = -EINVAL;
2403 goto finish;
2404 }
2405
2406 if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) {
2407 error_setg(errp, "Lazy refcounts only supported with compatibility "
2408 "level 1.1 and above (use compat=1.1 or greater)");
2409 ret = -EINVAL;
2410 goto finish;
2411 }
2412
2413 refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS,
2414 refcount_bits);
2415 if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
2416 error_setg(errp, "Refcount width must be a power of two and may not "
2417 "exceed 64 bits");
2418 ret = -EINVAL;
2419 goto finish;
2420 }
2421
2422 if (version < 3 && refcount_bits != 16) {
2423 error_setg(errp, "Different refcount widths than 16 bits require "
2424 "compatibility level 1.1 or above (use compat=1.1 or "
2425 "greater)");
2426 ret = -EINVAL;
2427 goto finish;
2428 }
2429
2430 refcount_order = ctz32(refcount_bits);
2431
2432 ret = qcow2_create2(filename, size, backing_file, backing_fmt, flags,
2433 cluster_size, prealloc, opts, version, refcount_order,
2434 &local_err);
2435 error_propagate(errp, local_err);
2436
2437 finish:
2438 g_free(backing_file);
2439 g_free(backing_fmt);
2440 g_free(buf);
2441 return ret;
2442 }
2443
2444
2445 static bool is_zero_sectors(BlockDriverState *bs, int64_t start,
2446 uint32_t count)
2447 {
2448 int nr;
2449 BlockDriverState *file;
2450 int64_t res;
2451
2452 if (!count) {
2453 return true;
2454 }
2455 res = bdrv_get_block_status_above(bs, NULL, start, count,
2456 &nr, &file);
2457 return res >= 0 && (res & BDRV_BLOCK_ZERO) && nr == count;
2458 }
2459
2460 static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs,
2461 int64_t offset, int count, BdrvRequestFlags flags)
2462 {
2463 int ret;
2464 BDRVQcow2State *s = bs->opaque;
2465
2466 uint32_t head = offset % s->cluster_size;
2467 uint32_t tail = (offset + count) % s->cluster_size;
2468
2469 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, count);
2470
2471 if (head || tail) {
2472 int64_t cl_start = (offset - head) >> BDRV_SECTOR_BITS;
2473 uint64_t off;
2474 unsigned int nr;
2475
2476 assert(head + count <= s->cluster_size);
2477
2478 /* check whether remainder of cluster already reads as zero */
2479 if (!(is_zero_sectors(bs, cl_start,
2480 DIV_ROUND_UP(head, BDRV_SECTOR_SIZE)) &&
2481 is_zero_sectors(bs, (offset + count) >> BDRV_SECTOR_BITS,
2482 DIV_ROUND_UP(-tail & (s->cluster_size - 1),
2483 BDRV_SECTOR_SIZE)))) {
2484 return -ENOTSUP;
2485 }
2486
2487 qemu_co_mutex_lock(&s->lock);
2488 /* We can have new write after previous check */
2489 offset = cl_start << BDRV_SECTOR_BITS;
2490 count = s->cluster_size;
2491 nr = s->cluster_size;
2492 ret = qcow2_get_cluster_offset(bs, offset, &nr, &off);
2493 if (ret != QCOW2_CLUSTER_UNALLOCATED && ret != QCOW2_CLUSTER_ZERO) {
2494 qemu_co_mutex_unlock(&s->lock);
2495 return -ENOTSUP;
2496 }
2497 } else {
2498 qemu_co_mutex_lock(&s->lock);
2499 }
2500
2501 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, count);
2502
2503 /* Whatever is left can use real zero clusters */
2504 ret = qcow2_zero_clusters(bs, offset, count >> BDRV_SECTOR_BITS, flags);
2505 qemu_co_mutex_unlock(&s->lock);
2506
2507 return ret;
2508 }
2509
2510 static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs,
2511 int64_t offset, int count)
2512 {
2513 int ret;
2514 BDRVQcow2State *s = bs->opaque;
2515
2516 if (!QEMU_IS_ALIGNED(offset | count, s->cluster_size)) {
2517 assert(count < s->cluster_size);
2518 return -ENOTSUP;
2519 }
2520
2521 qemu_co_mutex_lock(&s->lock);
2522 ret = qcow2_discard_clusters(bs, offset, count >> BDRV_SECTOR_BITS,
2523 QCOW2_DISCARD_REQUEST, false);
2524 qemu_co_mutex_unlock(&s->lock);
2525 return ret;
2526 }
2527
2528 static int qcow2_truncate(BlockDriverState *bs, int64_t offset)
2529 {
2530 BDRVQcow2State *s = bs->opaque;
2531 int64_t new_l1_size;
2532 int ret;
2533
2534 if (offset & 511) {
2535 error_report("The new size must be a multiple of 512");
2536 return -EINVAL;
2537 }
2538
2539 /* cannot proceed if image has snapshots */
2540 if (s->nb_snapshots) {
2541 error_report("Can't resize an image which has snapshots");
2542 return -ENOTSUP;
2543 }
2544
2545 /* shrinking is currently not supported */
2546 if (offset < bs->total_sectors * 512) {
2547 error_report("qcow2 doesn't support shrinking images yet");
2548 return -ENOTSUP;
2549 }
2550
2551 new_l1_size = size_to_l1(s, offset);
2552 ret = qcow2_grow_l1_table(bs, new_l1_size, true);
2553 if (ret < 0) {
2554 return ret;
2555 }
2556
2557 /* write updated header.size */
2558 offset = cpu_to_be64(offset);
2559 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
2560 &offset, sizeof(uint64_t));
2561 if (ret < 0) {
2562 return ret;
2563 }
2564
2565 s->l1_vm_state_index = new_l1_size;
2566 return 0;
2567 }
2568
2569 /* XXX: put compressed sectors first, then all the cluster aligned
2570 tables to avoid losing bytes in alignment */
2571 static coroutine_fn int
2572 qcow2_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
2573 uint64_t bytes, QEMUIOVector *qiov)
2574 {
2575 BDRVQcow2State *s = bs->opaque;
2576 QEMUIOVector hd_qiov;
2577 struct iovec iov;
2578 z_stream strm;
2579 int ret, out_len;
2580 uint8_t *buf, *out_buf;
2581 uint64_t cluster_offset;
2582
2583 if (bytes == 0) {
2584 /* align end of file to a sector boundary to ease reading with
2585 sector based I/Os */
2586 cluster_offset = bdrv_getlength(bs->file->bs);
2587 return bdrv_truncate(bs->file, cluster_offset);
2588 }
2589
2590 buf = qemu_blockalign(bs, s->cluster_size);
2591 if (bytes != s->cluster_size) {
2592 if (bytes > s->cluster_size ||
2593 offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
2594 {
2595 qemu_vfree(buf);
2596 return -EINVAL;
2597 }
2598 /* Zero-pad last write if image size is not cluster aligned */
2599 memset(buf + bytes, 0, s->cluster_size - bytes);
2600 }
2601 qemu_iovec_to_buf(qiov, 0, buf, bytes);
2602
2603 out_buf = g_malloc(s->cluster_size);
2604
2605 /* best compression, small window, no zlib header */
2606 memset(&strm, 0, sizeof(strm));
2607 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
2608 Z_DEFLATED, -12,
2609 9, Z_DEFAULT_STRATEGY);
2610 if (ret != 0) {
2611 ret = -EINVAL;
2612 goto fail;
2613 }
2614
2615 strm.avail_in = s->cluster_size;
2616 strm.next_in = (uint8_t *)buf;
2617 strm.avail_out = s->cluster_size;
2618 strm.next_out = out_buf;
2619
2620 ret = deflate(&strm, Z_FINISH);
2621 if (ret != Z_STREAM_END && ret != Z_OK) {
2622 deflateEnd(&strm);
2623 ret = -EINVAL;
2624 goto fail;
2625 }
2626 out_len = strm.next_out - out_buf;
2627
2628 deflateEnd(&strm);
2629
2630 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
2631 /* could not compress: write normal cluster */
2632 ret = qcow2_co_pwritev(bs, offset, bytes, qiov, 0);
2633 if (ret < 0) {
2634 goto fail;
2635 }
2636 goto success;
2637 }
2638
2639 qemu_co_mutex_lock(&s->lock);
2640 cluster_offset =
2641 qcow2_alloc_compressed_cluster_offset(bs, offset, out_len);
2642 if (!cluster_offset) {
2643 qemu_co_mutex_unlock(&s->lock);
2644 ret = -EIO;
2645 goto fail;
2646 }
2647 cluster_offset &= s->cluster_offset_mask;
2648
2649 ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len);
2650 qemu_co_mutex_unlock(&s->lock);
2651 if (ret < 0) {
2652 goto fail;
2653 }
2654
2655 iov = (struct iovec) {
2656 .iov_base = out_buf,
2657 .iov_len = out_len,
2658 };
2659 qemu_iovec_init_external(&hd_qiov, &iov, 1);
2660
2661 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
2662 ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0);
2663 if (ret < 0) {
2664 goto fail;
2665 }
2666 success:
2667 ret = 0;
2668 fail:
2669 qemu_vfree(buf);
2670 g_free(out_buf);
2671 return ret;
2672 }
2673
2674 static int make_completely_empty(BlockDriverState *bs)
2675 {
2676 BDRVQcow2State *s = bs->opaque;
2677 int ret, l1_clusters;
2678 int64_t offset;
2679 uint64_t *new_reftable = NULL;
2680 uint64_t rt_entry, l1_size2;
2681 struct {
2682 uint64_t l1_offset;
2683 uint64_t reftable_offset;
2684 uint32_t reftable_clusters;
2685 } QEMU_PACKED l1_ofs_rt_ofs_cls;
2686
2687 ret = qcow2_cache_empty(bs, s->l2_table_cache);
2688 if (ret < 0) {
2689 goto fail;
2690 }
2691
2692 ret = qcow2_cache_empty(bs, s->refcount_block_cache);
2693 if (ret < 0) {
2694 goto fail;
2695 }
2696
2697 /* Refcounts will be broken utterly */
2698 ret = qcow2_mark_dirty(bs);
2699 if (ret < 0) {
2700 goto fail;
2701 }
2702
2703 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
2704
2705 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
2706 l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t);
2707
2708 /* After this call, neither the in-memory nor the on-disk refcount
2709 * information accurately describe the actual references */
2710
2711 ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset,
2712 l1_clusters * s->cluster_size, 0);
2713 if (ret < 0) {
2714 goto fail_broken_refcounts;
2715 }
2716 memset(s->l1_table, 0, l1_size2);
2717
2718 BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE);
2719
2720 /* Overwrite enough clusters at the beginning of the sectors to place
2721 * the refcount table, a refcount block and the L1 table in; this may
2722 * overwrite parts of the existing refcount and L1 table, which is not
2723 * an issue because the dirty flag is set, complete data loss is in fact
2724 * desired and partial data loss is consequently fine as well */
2725 ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size,
2726 (2 + l1_clusters) * s->cluster_size, 0);
2727 /* This call (even if it failed overall) may have overwritten on-disk
2728 * refcount structures; in that case, the in-memory refcount information
2729 * will probably differ from the on-disk information which makes the BDS
2730 * unusable */
2731 if (ret < 0) {
2732 goto fail_broken_refcounts;
2733 }
2734
2735 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
2736 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
2737
2738 /* "Create" an empty reftable (one cluster) directly after the image
2739 * header and an empty L1 table three clusters after the image header;
2740 * the cluster between those two will be used as the first refblock */
2741 l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size);
2742 l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size);
2743 l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1);
2744 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset),
2745 &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls));
2746 if (ret < 0) {
2747 goto fail_broken_refcounts;
2748 }
2749
2750 s->l1_table_offset = 3 * s->cluster_size;
2751
2752 new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t));
2753 if (!new_reftable) {
2754 ret = -ENOMEM;
2755 goto fail_broken_refcounts;
2756 }
2757
2758 s->refcount_table_offset = s->cluster_size;
2759 s->refcount_table_size = s->cluster_size / sizeof(uint64_t);
2760 s->max_refcount_table_index = 0;
2761
2762 g_free(s->refcount_table);
2763 s->refcount_table = new_reftable;
2764 new_reftable = NULL;
2765
2766 /* Now the in-memory refcount information again corresponds to the on-disk
2767 * information (reftable is empty and no refblocks (the refblock cache is
2768 * empty)); however, this means some clusters (e.g. the image header) are
2769 * referenced, but not refcounted, but the normal qcow2 code assumes that
2770 * the in-memory information is always correct */
2771
2772 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
2773
2774 /* Enter the first refblock into the reftable */
2775 rt_entry = cpu_to_be64(2 * s->cluster_size);
2776 ret = bdrv_pwrite_sync(bs->file, s->cluster_size,
2777 &rt_entry, sizeof(rt_entry));
2778 if (ret < 0) {
2779 goto fail_broken_refcounts;
2780 }
2781 s->refcount_table[0] = 2 * s->cluster_size;
2782
2783 s->free_cluster_index = 0;
2784 assert(3 + l1_clusters <= s->refcount_block_size);
2785 offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2);
2786 if (offset < 0) {
2787 ret = offset;
2788 goto fail_broken_refcounts;
2789 } else if (offset > 0) {
2790 error_report("First cluster in emptied image is in use");
2791 abort();
2792 }
2793
2794 /* Now finally the in-memory information corresponds to the on-disk
2795 * structures and is correct */
2796 ret = qcow2_mark_clean(bs);
2797 if (ret < 0) {
2798 goto fail;
2799 }
2800
2801 ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size);
2802 if (ret < 0) {
2803 goto fail;
2804 }
2805
2806 return 0;
2807
2808 fail_broken_refcounts:
2809 /* The BDS is unusable at this point. If we wanted to make it usable, we
2810 * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
2811 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
2812 * again. However, because the functions which could have caused this error
2813 * path to be taken are used by those functions as well, it's very likely
2814 * that that sequence will fail as well. Therefore, just eject the BDS. */
2815 bs->drv = NULL;
2816
2817 fail:
2818 g_free(new_reftable);
2819 return ret;
2820 }
2821
2822 static int qcow2_make_empty(BlockDriverState *bs)
2823 {
2824 BDRVQcow2State *s = bs->opaque;
2825 uint64_t start_sector;
2826 int sector_step = (QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size) /
2827 BDRV_SECTOR_SIZE);
2828 int l1_clusters, ret = 0;
2829
2830 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
2831
2832 if (s->qcow_version >= 3 && !s->snapshots &&
2833 3 + l1_clusters <= s->refcount_block_size) {
2834 /* The following function only works for qcow2 v3 images (it requires
2835 * the dirty flag) and only as long as there are no snapshots (because
2836 * it completely empties the image). Furthermore, the L1 table and three
2837 * additional clusters (image header, refcount table, one refcount
2838 * block) have to fit inside one refcount block. */
2839 return make_completely_empty(bs);
2840 }
2841
2842 /* This fallback code simply discards every active cluster; this is slow,
2843 * but works in all cases */
2844 for (start_sector = 0; start_sector < bs->total_sectors;
2845 start_sector += sector_step)
2846 {
2847 /* As this function is generally used after committing an external
2848 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
2849 * default action for this kind of discard is to pass the discard,
2850 * which will ideally result in an actually smaller image file, as
2851 * is probably desired. */
2852 ret = qcow2_discard_clusters(bs, start_sector * BDRV_SECTOR_SIZE,
2853 MIN(sector_step,
2854 bs->total_sectors - start_sector),
2855 QCOW2_DISCARD_SNAPSHOT, true);
2856 if (ret < 0) {
2857 break;
2858 }
2859 }
2860
2861 return ret;
2862 }
2863
2864 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
2865 {
2866 BDRVQcow2State *s = bs->opaque;
2867 int ret;
2868
2869 qemu_co_mutex_lock(&s->lock);
2870 ret = qcow2_cache_write(bs, s->l2_table_cache);
2871 if (ret < 0) {
2872 qemu_co_mutex_unlock(&s->lock);
2873 return ret;
2874 }
2875
2876 if (qcow2_need_accurate_refcounts(s)) {
2877 ret = qcow2_cache_write(bs, s->refcount_block_cache);
2878 if (ret < 0) {
2879 qemu_co_mutex_unlock(&s->lock);
2880 return ret;
2881 }
2882 }
2883 qemu_co_mutex_unlock(&s->lock);
2884
2885 return 0;
2886 }
2887
2888 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2889 {
2890 BDRVQcow2State *s = bs->opaque;
2891 bdi->unallocated_blocks_are_zero = true;
2892 bdi->can_write_zeroes_with_unmap = (s->qcow_version >= 3);
2893 bdi->cluster_size = s->cluster_size;
2894 bdi->vm_state_offset = qcow2_vm_state_offset(s);
2895 return 0;
2896 }
2897
2898 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs)
2899 {
2900 BDRVQcow2State *s = bs->opaque;
2901 ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1);
2902
2903 *spec_info = (ImageInfoSpecific){
2904 .type = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
2905 .u.qcow2.data = g_new(ImageInfoSpecificQCow2, 1),
2906 };
2907 if (s->qcow_version == 2) {
2908 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
2909 .compat = g_strdup("0.10"),
2910 .refcount_bits = s->refcount_bits,
2911 };
2912 } else if (s->qcow_version == 3) {
2913 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
2914 .compat = g_strdup("1.1"),
2915 .lazy_refcounts = s->compatible_features &
2916 QCOW2_COMPAT_LAZY_REFCOUNTS,
2917 .has_lazy_refcounts = true,
2918 .corrupt = s->incompatible_features &
2919 QCOW2_INCOMPAT_CORRUPT,
2920 .has_corrupt = true,
2921 .refcount_bits = s->refcount_bits,
2922 };
2923 } else {
2924 /* if this assertion fails, this probably means a new version was
2925 * added without having it covered here */
2926 assert(false);
2927 }
2928
2929 return spec_info;
2930 }
2931
2932 #if 0
2933 static void dump_refcounts(BlockDriverState *bs)
2934 {
2935 BDRVQcow2State *s = bs->opaque;
2936 int64_t nb_clusters, k, k1, size;
2937 int refcount;
2938
2939 size = bdrv_getlength(bs->file->bs);
2940 nb_clusters = size_to_clusters(s, size);
2941 for(k = 0; k < nb_clusters;) {
2942 k1 = k;
2943 refcount = get_refcount(bs, k);
2944 k++;
2945 while (k < nb_clusters && get_refcount(bs, k) == refcount)
2946 k++;
2947 printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
2948 k - k1);
2949 }
2950 }
2951 #endif
2952
2953 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
2954 int64_t pos)
2955 {
2956 BDRVQcow2State *s = bs->opaque;
2957
2958 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
2959 return bs->drv->bdrv_co_pwritev(bs, qcow2_vm_state_offset(s) + pos,
2960 qiov->size, qiov, 0);
2961 }
2962
2963 static int qcow2_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
2964 int64_t pos)
2965 {
2966 BDRVQcow2State *s = bs->opaque;
2967
2968 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
2969 return bs->drv->bdrv_co_preadv(bs, qcow2_vm_state_offset(s) + pos,
2970 qiov->size, qiov, 0);
2971 }
2972
2973 /*
2974 * Downgrades an image's version. To achieve this, any incompatible features
2975 * have to be removed.
2976 */
2977 static int qcow2_downgrade(BlockDriverState *bs, int target_version,
2978 BlockDriverAmendStatusCB *status_cb, void *cb_opaque)
2979 {
2980 BDRVQcow2State *s = bs->opaque;
2981 int current_version = s->qcow_version;
2982 int ret;
2983
2984 if (target_version == current_version) {
2985 return 0;
2986 } else if (target_version > current_version) {
2987 return -EINVAL;
2988 } else if (target_version != 2) {
2989 return -EINVAL;
2990 }
2991
2992 if (s->refcount_order != 4) {
2993 error_report("compat=0.10 requires refcount_bits=16");
2994 return -ENOTSUP;
2995 }
2996
2997 /* clear incompatible features */
2998 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
2999 ret = qcow2_mark_clean(bs);
3000 if (ret < 0) {
3001 return ret;
3002 }
3003 }
3004
3005 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
3006 * the first place; if that happens nonetheless, returning -ENOTSUP is the
3007 * best thing to do anyway */
3008
3009 if (s->incompatible_features) {
3010 return -ENOTSUP;
3011 }
3012
3013 /* since we can ignore compatible features, we can set them to 0 as well */
3014 s->compatible_features = 0;
3015 /* if lazy refcounts have been used, they have already been fixed through
3016 * clearing the dirty flag */
3017
3018 /* clearing autoclear features is trivial */
3019 s->autoclear_features = 0;
3020
3021 ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque);
3022 if (ret < 0) {
3023 return ret;
3024 }
3025
3026 s->qcow_version = target_version;
3027 ret = qcow2_update_header(bs);
3028 if (ret < 0) {
3029 s->qcow_version = current_version;
3030 return ret;
3031 }
3032 return 0;
3033 }
3034
3035 typedef enum Qcow2AmendOperation {
3036 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
3037 * statically initialized to so that the helper CB can discern the first
3038 * invocation from an operation change */
3039 QCOW2_NO_OPERATION = 0,
3040
3041 QCOW2_CHANGING_REFCOUNT_ORDER,
3042 QCOW2_DOWNGRADING,
3043 } Qcow2AmendOperation;
3044
3045 typedef struct Qcow2AmendHelperCBInfo {
3046 /* The code coordinating the amend operations should only modify
3047 * these four fields; the rest will be managed by the CB */
3048 BlockDriverAmendStatusCB *original_status_cb;
3049 void *original_cb_opaque;
3050
3051 Qcow2AmendOperation current_operation;
3052
3053 /* Total number of operations to perform (only set once) */
3054 int total_operations;
3055
3056 /* The following fields are managed by the CB */
3057
3058 /* Number of operations completed */
3059 int operations_completed;
3060
3061 /* Cumulative offset of all completed operations */
3062 int64_t offset_completed;
3063
3064 Qcow2AmendOperation last_operation;
3065 int64_t last_work_size;
3066 } Qcow2AmendHelperCBInfo;
3067
3068 static void qcow2_amend_helper_cb(BlockDriverState *bs,
3069 int64_t operation_offset,
3070 int64_t operation_work_size, void *opaque)
3071 {
3072 Qcow2AmendHelperCBInfo *info = opaque;
3073 int64_t current_work_size;
3074 int64_t projected_work_size;
3075
3076 if (info->current_operation != info->last_operation) {
3077 if (info->last_operation != QCOW2_NO_OPERATION) {
3078 info->offset_completed += info->last_work_size;
3079 info->operations_completed++;
3080 }
3081
3082 info->last_operation = info->current_operation;
3083 }
3084
3085 assert(info->total_operations > 0);
3086 assert(info->operations_completed < info->total_operations);
3087
3088 info->last_work_size = operation_work_size;
3089
3090 current_work_size = info->offset_completed + operation_work_size;
3091
3092 /* current_work_size is the total work size for (operations_completed + 1)
3093 * operations (which includes this one), so multiply it by the number of
3094 * operations not covered and divide it by the number of operations
3095 * covered to get a projection for the operations not covered */
3096 projected_work_size = current_work_size * (info->total_operations -
3097 info->operations_completed - 1)
3098 / (info->operations_completed + 1);
3099
3100 info->original_status_cb(bs, info->offset_completed + operation_offset,
3101 current_work_size + projected_work_size,
3102 info->original_cb_opaque);
3103 }
3104
3105 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,
3106 BlockDriverAmendStatusCB *status_cb,
3107 void *cb_opaque)
3108 {
3109 BDRVQcow2State *s = bs->opaque;
3110 int old_version = s->qcow_version, new_version = old_version;
3111 uint64_t new_size = 0;
3112 const char *backing_file = NULL, *backing_format = NULL;
3113 bool lazy_refcounts = s->use_lazy_refcounts;
3114 const char *compat = NULL;
3115 uint64_t cluster_size = s->cluster_size;
3116 bool encrypt;
3117 int refcount_bits = s->refcount_bits;
3118 Error *local_err = NULL;
3119 int ret;
3120 QemuOptDesc *desc = opts->list->desc;
3121 Qcow2AmendHelperCBInfo helper_cb_info;
3122
3123 while (desc && desc->name) {
3124 if (!qemu_opt_find(opts, desc->name)) {
3125 /* only change explicitly defined options */
3126 desc++;
3127 continue;
3128 }
3129
3130 if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) {
3131 compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL);
3132 if (!compat) {
3133 /* preserve default */
3134 } else if (!strcmp(compat, "0.10")) {
3135 new_version = 2;
3136 } else if (!strcmp(compat, "1.1")) {
3137 new_version = 3;
3138 } else {
3139 error_report("Unknown compatibility level %s", compat);
3140 return -EINVAL;
3141 }
3142 } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) {
3143 error_report("Cannot change preallocation mode");
3144 return -ENOTSUP;
3145 } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) {
3146 new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
3147 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) {
3148 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
3149 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) {
3150 backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
3151 } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) {
3152 encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT,
3153 !!s->cipher);
3154
3155 if (encrypt != !!s->cipher) {
3156 error_report("Changing the encryption flag is not supported");
3157 return -ENOTSUP;
3158 }
3159 } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) {
3160 cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE,
3161 cluster_size);
3162 if (cluster_size != s->cluster_size) {
3163 error_report("Changing the cluster size is not supported");
3164 return -ENOTSUP;
3165 }
3166 } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
3167 lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS,
3168 lazy_refcounts);
3169 } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) {
3170 refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS,
3171 refcount_bits);
3172
3173 if (refcount_bits <= 0 || refcount_bits > 64 ||
3174 !is_power_of_2(refcount_bits))
3175 {
3176 error_report("Refcount width must be a power of two and may "
3177 "not exceed 64 bits");
3178 return -EINVAL;
3179 }
3180 } else {
3181 /* if this point is reached, this probably means a new option was
3182 * added without having it covered here */
3183 abort();
3184 }
3185
3186 desc++;
3187 }
3188
3189 helper_cb_info = (Qcow2AmendHelperCBInfo){
3190 .original_status_cb = status_cb,
3191 .original_cb_opaque = cb_opaque,
3192 .total_operations = (new_version < old_version)
3193 + (s->refcount_bits != refcount_bits)
3194 };
3195
3196 /* Upgrade first (some features may require compat=1.1) */
3197 if (new_version > old_version) {
3198 s->qcow_version = new_version;
3199 ret = qcow2_update_header(bs);
3200 if (ret < 0) {
3201 s->qcow_version = old_version;
3202 return ret;
3203 }
3204 }
3205
3206 if (s->refcount_bits != refcount_bits) {
3207 int refcount_order = ctz32(refcount_bits);
3208 Error *local_error = NULL;
3209
3210 if (new_version < 3 && refcount_bits != 16) {
3211 error_report("Different refcount widths than 16 bits require "
3212 "compatibility level 1.1 or above (use compat=1.1 or "
3213 "greater)");
3214 return -EINVAL;
3215 }
3216
3217 helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER;
3218 ret = qcow2_change_refcount_order(bs, refcount_order,
3219 &qcow2_amend_helper_cb,
3220 &helper_cb_info, &local_error);
3221 if (ret < 0) {
3222 error_report_err(local_error);
3223 return ret;
3224 }
3225 }
3226
3227 if (backing_file || backing_format) {
3228 ret = qcow2_change_backing_file(bs,
3229 backing_file ?: s->image_backing_file,
3230 backing_format ?: s->image_backing_format);
3231 if (ret < 0) {
3232 return ret;
3233 }
3234 }
3235
3236 if (s->use_lazy_refcounts != lazy_refcounts) {
3237 if (lazy_refcounts) {
3238 if (new_version < 3) {
3239 error_report("Lazy refcounts only supported with compatibility "
3240 "level 1.1 and above (use compat=1.1 or greater)");
3241 return -EINVAL;
3242 }
3243 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
3244 ret = qcow2_update_header(bs);
3245 if (ret < 0) {
3246 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
3247 return ret;
3248 }
3249 s->use_lazy_refcounts = true;
3250 } else {
3251 /* make image clean first */
3252 ret = qcow2_mark_clean(bs);
3253 if (ret < 0) {
3254 return ret;
3255 }
3256 /* now disallow lazy refcounts */
3257 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
3258 ret = qcow2_update_header(bs);
3259 if (ret < 0) {
3260 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
3261 return ret;
3262 }
3263 s->use_lazy_refcounts = false;
3264 }
3265 }
3266
3267 if (new_size) {
3268 BlockBackend *blk = blk_new(BLK_PERM_RESIZE, BLK_PERM_ALL);
3269 ret = blk_insert_bs(blk, bs, &local_err);
3270 if (ret < 0) {
3271 error_report_err(local_err);
3272 blk_unref(blk);
3273 return ret;
3274 }
3275
3276 ret = blk_truncate(blk, new_size);
3277 blk_unref(blk);
3278 if (ret < 0) {
3279 return ret;
3280 }
3281 }
3282
3283 /* Downgrade last (so unsupported features can be removed before) */
3284 if (new_version < old_version) {
3285 helper_cb_info.current_operation = QCOW2_DOWNGRADING;
3286 ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb,
3287 &helper_cb_info);
3288 if (ret < 0) {
3289 return ret;
3290 }
3291 }
3292
3293 return 0;
3294 }
3295
3296 /*
3297 * If offset or size are negative, respectively, they will not be included in
3298 * the BLOCK_IMAGE_CORRUPTED event emitted.
3299 * fatal will be ignored for read-only BDS; corruptions found there will always
3300 * be considered non-fatal.
3301 */
3302 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
3303 int64_t size, const char *message_format, ...)
3304 {
3305 BDRVQcow2State *s = bs->opaque;
3306 const char *node_name;
3307 char *message;
3308 va_list ap;
3309
3310 fatal = fatal && !bs->read_only;
3311
3312 if (s->signaled_corruption &&
3313 (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
3314 {
3315 return;
3316 }
3317
3318 va_start(ap, message_format);
3319 message = g_strdup_vprintf(message_format, ap);
3320 va_end(ap);
3321
3322 if (fatal) {
3323 fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
3324 "corruption events will be suppressed\n", message);
3325 } else {
3326 fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
3327 "corruption events will be suppressed\n", message);
3328 }
3329
3330 node_name = bdrv_get_node_name(bs);
3331 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs),
3332 *node_name != '\0', node_name,
3333 message, offset >= 0, offset,
3334 size >= 0, size,
3335 fatal, &error_abort);
3336 g_free(message);
3337
3338 if (fatal) {
3339 qcow2_mark_corrupt(bs);
3340 bs->drv = NULL; /* make BDS unusable */
3341 }
3342
3343 s->signaled_corruption = true;
3344 }
3345
3346 static QemuOptsList qcow2_create_opts = {
3347 .name = "qcow2-create-opts",
3348 .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
3349 .desc = {
3350 {
3351 .name = BLOCK_OPT_SIZE,
3352 .type = QEMU_OPT_SIZE,
3353 .help = "Virtual disk size"
3354 },
3355 {
3356 .name = BLOCK_OPT_COMPAT_LEVEL,
3357 .type = QEMU_OPT_STRING,
3358 .help = "Compatibility level (0.10 or 1.1)"
3359 },
3360 {
3361 .name = BLOCK_OPT_BACKING_FILE,
3362 .type = QEMU_OPT_STRING,
3363 .help = "File name of a base image"
3364 },
3365 {
3366 .name = BLOCK_OPT_BACKING_FMT,
3367 .type = QEMU_OPT_STRING,
3368 .help = "Image format of the base image"
3369 },
3370 {
3371 .name = BLOCK_OPT_ENCRYPT,
3372 .type = QEMU_OPT_BOOL,
3373 .help = "Encrypt the image",
3374 .def_value_str = "off"
3375 },
3376 {
3377 .name = BLOCK_OPT_CLUSTER_SIZE,
3378 .type = QEMU_OPT_SIZE,
3379 .help = "qcow2 cluster size",
3380 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
3381 },
3382 {
3383 .name = BLOCK_OPT_PREALLOC,
3384 .type = QEMU_OPT_STRING,
3385 .help = "Preallocation mode (allowed values: off, metadata, "
3386 "falloc, full)"
3387 },
3388 {
3389 .name = BLOCK_OPT_LAZY_REFCOUNTS,
3390 .type = QEMU_OPT_BOOL,
3391 .help = "Postpone refcount updates",