block: Catch backing files assigned to non-COW drivers
[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-common.h"
25 #include "block/block_int.h"
26 #include "qemu/module.h"
27 #include <zlib.h>
28 #include "qemu/aes.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 "trace.h"
34 #include "qemu/option_int.h"
35
36 /*
37 Differences with QCOW:
38
39 - Support for multiple incremental snapshots.
40 - Memory management by reference counts.
41 - Clusters which have a reference count of one have the bit
42 QCOW_OFLAG_COPIED to optimize write performance.
43 - Size of compressed clusters is stored in sectors to reduce bit usage
44 in the cluster offsets.
45 - Support for storing additional data (such as the VM state) in the
46 snapshots.
47 - If a backing store is used, the cluster size is not constrained
48 (could be backported to QCOW).
49 - L2 tables have always a size of one cluster.
50 */
51
52
53 typedef struct {
54 uint32_t magic;
55 uint32_t len;
56 } QEMU_PACKED QCowExtension;
57
58 #define QCOW2_EXT_MAGIC_END 0
59 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
60 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
61
62 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
63 {
64 const QCowHeader *cow_header = (const void *)buf;
65
66 if (buf_size >= sizeof(QCowHeader) &&
67 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
68 be32_to_cpu(cow_header->version) >= 2)
69 return 100;
70 else
71 return 0;
72 }
73
74
75 /*
76 * read qcow2 extension and fill bs
77 * start reading from start_offset
78 * finish reading upon magic of value 0 or when end_offset reached
79 * unknown magic is skipped (future extension this version knows nothing about)
80 * return 0 upon success, non-0 otherwise
81 */
82 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
83 uint64_t end_offset, void **p_feature_table,
84 Error **errp)
85 {
86 BDRVQcowState *s = bs->opaque;
87 QCowExtension ext;
88 uint64_t offset;
89 int ret;
90
91 #ifdef DEBUG_EXT
92 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
93 #endif
94 offset = start_offset;
95 while (offset < end_offset) {
96
97 #ifdef DEBUG_EXT
98 /* Sanity check */
99 if (offset > s->cluster_size)
100 printf("qcow2_read_extension: suspicious offset %lu\n", offset);
101
102 printf("attempting to read extended header in offset %lu\n", offset);
103 #endif
104
105 ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext));
106 if (ret < 0) {
107 error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: "
108 "pread fail from offset %" PRIu64, offset);
109 return 1;
110 }
111 be32_to_cpus(&ext.magic);
112 be32_to_cpus(&ext.len);
113 offset += sizeof(ext);
114 #ifdef DEBUG_EXT
115 printf("ext.magic = 0x%x\n", ext.magic);
116 #endif
117 if (ext.len > end_offset - offset) {
118 error_setg(errp, "Header extension too large");
119 return -EINVAL;
120 }
121
122 switch (ext.magic) {
123 case QCOW2_EXT_MAGIC_END:
124 return 0;
125
126 case QCOW2_EXT_MAGIC_BACKING_FORMAT:
127 if (ext.len >= sizeof(bs->backing_format)) {
128 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32
129 " too large (>=%zu)", ext.len,
130 sizeof(bs->backing_format));
131 return 2;
132 }
133 ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len);
134 if (ret < 0) {
135 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: "
136 "Could not read format name");
137 return 3;
138 }
139 bs->backing_format[ext.len] = '\0';
140 #ifdef DEBUG_EXT
141 printf("Qcow2: Got format extension %s\n", bs->backing_format);
142 #endif
143 break;
144
145 case QCOW2_EXT_MAGIC_FEATURE_TABLE:
146 if (p_feature_table != NULL) {
147 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
148 ret = bdrv_pread(bs->file, offset , feature_table, ext.len);
149 if (ret < 0) {
150 error_setg_errno(errp, -ret, "ERROR: ext_feature_table: "
151 "Could not read table");
152 return ret;
153 }
154
155 *p_feature_table = feature_table;
156 }
157 break;
158
159 default:
160 /* unknown magic - save it in case we need to rewrite the header */
161 {
162 Qcow2UnknownHeaderExtension *uext;
163
164 uext = g_malloc0(sizeof(*uext) + ext.len);
165 uext->magic = ext.magic;
166 uext->len = ext.len;
167 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
168
169 ret = bdrv_pread(bs->file, offset , uext->data, uext->len);
170 if (ret < 0) {
171 error_setg_errno(errp, -ret, "ERROR: unknown extension: "
172 "Could not read data");
173 return ret;
174 }
175 }
176 break;
177 }
178
179 offset += ((ext.len + 7) & ~7);
180 }
181
182 return 0;
183 }
184
185 static void cleanup_unknown_header_ext(BlockDriverState *bs)
186 {
187 BDRVQcowState *s = bs->opaque;
188 Qcow2UnknownHeaderExtension *uext, *next;
189
190 QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
191 QLIST_REMOVE(uext, next);
192 g_free(uext);
193 }
194 }
195
196 static void GCC_FMT_ATTR(3, 4) report_unsupported(BlockDriverState *bs,
197 Error **errp, const char *fmt, ...)
198 {
199 char msg[64];
200 va_list ap;
201
202 va_start(ap, fmt);
203 vsnprintf(msg, sizeof(msg), fmt, ap);
204 va_end(ap);
205
206 error_set(errp, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE, bs->device_name, "qcow2",
207 msg);
208 }
209
210 static void report_unsupported_feature(BlockDriverState *bs,
211 Error **errp, Qcow2Feature *table, uint64_t mask)
212 {
213 while (table && table->name[0] != '\0') {
214 if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
215 if (mask & (1 << table->bit)) {
216 report_unsupported(bs, errp, "%.46s", table->name);
217 mask &= ~(1 << table->bit);
218 }
219 }
220 table++;
221 }
222
223 if (mask) {
224 report_unsupported(bs, errp, "Unknown incompatible feature: %" PRIx64,
225 mask);
226 }
227 }
228
229 /*
230 * Sets the dirty bit and flushes afterwards if necessary.
231 *
232 * The incompatible_features bit is only set if the image file header was
233 * updated successfully. Therefore it is not required to check the return
234 * value of this function.
235 */
236 int qcow2_mark_dirty(BlockDriverState *bs)
237 {
238 BDRVQcowState *s = bs->opaque;
239 uint64_t val;
240 int ret;
241
242 assert(s->qcow_version >= 3);
243
244 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
245 return 0; /* already dirty */
246 }
247
248 val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY);
249 ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features),
250 &val, sizeof(val));
251 if (ret < 0) {
252 return ret;
253 }
254 ret = bdrv_flush(bs->file);
255 if (ret < 0) {
256 return ret;
257 }
258
259 /* Only treat image as dirty if the header was updated successfully */
260 s->incompatible_features |= QCOW2_INCOMPAT_DIRTY;
261 return 0;
262 }
263
264 /*
265 * Clears the dirty bit and flushes before if necessary. Only call this
266 * function when there are no pending requests, it does not guard against
267 * concurrent requests dirtying the image.
268 */
269 static int qcow2_mark_clean(BlockDriverState *bs)
270 {
271 BDRVQcowState *s = bs->opaque;
272
273 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
274 int ret;
275
276 s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY;
277
278 ret = bdrv_flush(bs);
279 if (ret < 0) {
280 return ret;
281 }
282
283 return qcow2_update_header(bs);
284 }
285 return 0;
286 }
287
288 /*
289 * Marks the image as corrupt.
290 */
291 int qcow2_mark_corrupt(BlockDriverState *bs)
292 {
293 BDRVQcowState *s = bs->opaque;
294
295 s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT;
296 return qcow2_update_header(bs);
297 }
298
299 /*
300 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
301 * before if necessary.
302 */
303 int qcow2_mark_consistent(BlockDriverState *bs)
304 {
305 BDRVQcowState *s = bs->opaque;
306
307 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
308 int ret = bdrv_flush(bs);
309 if (ret < 0) {
310 return ret;
311 }
312
313 s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT;
314 return qcow2_update_header(bs);
315 }
316 return 0;
317 }
318
319 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result,
320 BdrvCheckMode fix)
321 {
322 int ret = qcow2_check_refcounts(bs, result, fix);
323 if (ret < 0) {
324 return ret;
325 }
326
327 if (fix && result->check_errors == 0 && result->corruptions == 0) {
328 ret = qcow2_mark_clean(bs);
329 if (ret < 0) {
330 return ret;
331 }
332 return qcow2_mark_consistent(bs);
333 }
334 return ret;
335 }
336
337 static int validate_table_offset(BlockDriverState *bs, uint64_t offset,
338 uint64_t entries, size_t entry_len)
339 {
340 BDRVQcowState *s = bs->opaque;
341 uint64_t size;
342
343 /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
344 * because values will be passed to qemu functions taking int64_t. */
345 if (entries > INT64_MAX / entry_len) {
346 return -EINVAL;
347 }
348
349 size = entries * entry_len;
350
351 if (INT64_MAX - size < offset) {
352 return -EINVAL;
353 }
354
355 /* Tables must be cluster aligned */
356 if (offset & (s->cluster_size - 1)) {
357 return -EINVAL;
358 }
359
360 return 0;
361 }
362
363 static QemuOptsList qcow2_runtime_opts = {
364 .name = "qcow2",
365 .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head),
366 .desc = {
367 {
368 .name = QCOW2_OPT_LAZY_REFCOUNTS,
369 .type = QEMU_OPT_BOOL,
370 .help = "Postpone refcount updates",
371 },
372 {
373 .name = QCOW2_OPT_DISCARD_REQUEST,
374 .type = QEMU_OPT_BOOL,
375 .help = "Pass guest discard requests to the layer below",
376 },
377 {
378 .name = QCOW2_OPT_DISCARD_SNAPSHOT,
379 .type = QEMU_OPT_BOOL,
380 .help = "Generate discard requests when snapshot related space "
381 "is freed",
382 },
383 {
384 .name = QCOW2_OPT_DISCARD_OTHER,
385 .type = QEMU_OPT_BOOL,
386 .help = "Generate discard requests when other clusters are freed",
387 },
388 {
389 .name = QCOW2_OPT_OVERLAP,
390 .type = QEMU_OPT_STRING,
391 .help = "Selects which overlap checks to perform from a range of "
392 "templates (none, constant, cached, all)",
393 },
394 {
395 .name = QCOW2_OPT_OVERLAP_MAIN_HEADER,
396 .type = QEMU_OPT_BOOL,
397 .help = "Check for unintended writes into the main qcow2 header",
398 },
399 {
400 .name = QCOW2_OPT_OVERLAP_ACTIVE_L1,
401 .type = QEMU_OPT_BOOL,
402 .help = "Check for unintended writes into the active L1 table",
403 },
404 {
405 .name = QCOW2_OPT_OVERLAP_ACTIVE_L2,
406 .type = QEMU_OPT_BOOL,
407 .help = "Check for unintended writes into an active L2 table",
408 },
409 {
410 .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
411 .type = QEMU_OPT_BOOL,
412 .help = "Check for unintended writes into the refcount table",
413 },
414 {
415 .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
416 .type = QEMU_OPT_BOOL,
417 .help = "Check for unintended writes into a refcount block",
418 },
419 {
420 .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
421 .type = QEMU_OPT_BOOL,
422 .help = "Check for unintended writes into the snapshot table",
423 },
424 {
425 .name = QCOW2_OPT_OVERLAP_INACTIVE_L1,
426 .type = QEMU_OPT_BOOL,
427 .help = "Check for unintended writes into an inactive L1 table",
428 },
429 {
430 .name = QCOW2_OPT_OVERLAP_INACTIVE_L2,
431 .type = QEMU_OPT_BOOL,
432 .help = "Check for unintended writes into an inactive L2 table",
433 },
434 { /* end of list */ }
435 },
436 };
437
438 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = {
439 [QCOW2_OL_MAIN_HEADER_BITNR] = QCOW2_OPT_OVERLAP_MAIN_HEADER,
440 [QCOW2_OL_ACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L1,
441 [QCOW2_OL_ACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L2,
442 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
443 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
444 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
445 [QCOW2_OL_INACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L1,
446 [QCOW2_OL_INACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L2,
447 };
448
449 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
450 Error **errp)
451 {
452 BDRVQcowState *s = bs->opaque;
453 unsigned int len, i;
454 int ret = 0;
455 QCowHeader header;
456 QemuOpts *opts;
457 Error *local_err = NULL;
458 uint64_t ext_end;
459 uint64_t l1_vm_state_index;
460 const char *opt_overlap_check;
461 int overlap_check_template = 0;
462
463 ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
464 if (ret < 0) {
465 error_setg_errno(errp, -ret, "Could not read qcow2 header");
466 goto fail;
467 }
468 be32_to_cpus(&header.magic);
469 be32_to_cpus(&header.version);
470 be64_to_cpus(&header.backing_file_offset);
471 be32_to_cpus(&header.backing_file_size);
472 be64_to_cpus(&header.size);
473 be32_to_cpus(&header.cluster_bits);
474 be32_to_cpus(&header.crypt_method);
475 be64_to_cpus(&header.l1_table_offset);
476 be32_to_cpus(&header.l1_size);
477 be64_to_cpus(&header.refcount_table_offset);
478 be32_to_cpus(&header.refcount_table_clusters);
479 be64_to_cpus(&header.snapshots_offset);
480 be32_to_cpus(&header.nb_snapshots);
481
482 if (header.magic != QCOW_MAGIC) {
483 error_setg(errp, "Image is not in qcow2 format");
484 ret = -EINVAL;
485 goto fail;
486 }
487 if (header.version < 2 || header.version > 3) {
488 report_unsupported(bs, errp, "QCOW version %" PRIu32, header.version);
489 ret = -ENOTSUP;
490 goto fail;
491 }
492
493 s->qcow_version = header.version;
494
495 /* Initialise cluster size */
496 if (header.cluster_bits < MIN_CLUSTER_BITS ||
497 header.cluster_bits > MAX_CLUSTER_BITS) {
498 error_setg(errp, "Unsupported cluster size: 2^%" PRIu32,
499 header.cluster_bits);
500 ret = -EINVAL;
501 goto fail;
502 }
503
504 s->cluster_bits = header.cluster_bits;
505 s->cluster_size = 1 << s->cluster_bits;
506 s->cluster_sectors = 1 << (s->cluster_bits - 9);
507
508 /* Initialise version 3 header fields */
509 if (header.version == 2) {
510 header.incompatible_features = 0;
511 header.compatible_features = 0;
512 header.autoclear_features = 0;
513 header.refcount_order = 4;
514 header.header_length = 72;
515 } else {
516 be64_to_cpus(&header.incompatible_features);
517 be64_to_cpus(&header.compatible_features);
518 be64_to_cpus(&header.autoclear_features);
519 be32_to_cpus(&header.refcount_order);
520 be32_to_cpus(&header.header_length);
521
522 if (header.header_length < 104) {
523 error_setg(errp, "qcow2 header too short");
524 ret = -EINVAL;
525 goto fail;
526 }
527 }
528
529 if (header.header_length > s->cluster_size) {
530 error_setg(errp, "qcow2 header exceeds cluster size");
531 ret = -EINVAL;
532 goto fail;
533 }
534
535 if (header.header_length > sizeof(header)) {
536 s->unknown_header_fields_size = header.header_length - sizeof(header);
537 s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
538 ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields,
539 s->unknown_header_fields_size);
540 if (ret < 0) {
541 error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
542 "fields");
543 goto fail;
544 }
545 }
546
547 if (header.backing_file_offset > s->cluster_size) {
548 error_setg(errp, "Invalid backing file offset");
549 ret = -EINVAL;
550 goto fail;
551 }
552
553 if (header.backing_file_offset) {
554 ext_end = header.backing_file_offset;
555 } else {
556 ext_end = 1 << header.cluster_bits;
557 }
558
559 /* Handle feature bits */
560 s->incompatible_features = header.incompatible_features;
561 s->compatible_features = header.compatible_features;
562 s->autoclear_features = header.autoclear_features;
563
564 if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
565 void *feature_table = NULL;
566 qcow2_read_extensions(bs, header.header_length, ext_end,
567 &feature_table, NULL);
568 report_unsupported_feature(bs, errp, feature_table,
569 s->incompatible_features &
570 ~QCOW2_INCOMPAT_MASK);
571 ret = -ENOTSUP;
572 g_free(feature_table);
573 goto fail;
574 }
575
576 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
577 /* Corrupt images may not be written to unless they are being repaired
578 */
579 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
580 error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
581 "read/write");
582 ret = -EACCES;
583 goto fail;
584 }
585 }
586
587 /* Check support for various header values */
588 if (header.refcount_order != 4) {
589 report_unsupported(bs, errp, "%d bit reference counts",
590 1 << header.refcount_order);
591 ret = -ENOTSUP;
592 goto fail;
593 }
594 s->refcount_order = header.refcount_order;
595
596 if (header.crypt_method > QCOW_CRYPT_AES) {
597 error_setg(errp, "Unsupported encryption method: %" PRIu32,
598 header.crypt_method);
599 ret = -EINVAL;
600 goto fail;
601 }
602 s->crypt_method_header = header.crypt_method;
603 if (s->crypt_method_header) {
604 bs->encrypted = 1;
605 }
606
607 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
608 s->l2_size = 1 << s->l2_bits;
609 bs->total_sectors = header.size / 512;
610 s->csize_shift = (62 - (s->cluster_bits - 8));
611 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
612 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
613
614 s->refcount_table_offset = header.refcount_table_offset;
615 s->refcount_table_size =
616 header.refcount_table_clusters << (s->cluster_bits - 3);
617
618 if (header.refcount_table_clusters > qcow2_max_refcount_clusters(s)) {
619 error_setg(errp, "Reference count table too large");
620 ret = -EINVAL;
621 goto fail;
622 }
623
624 ret = validate_table_offset(bs, s->refcount_table_offset,
625 s->refcount_table_size, sizeof(uint64_t));
626 if (ret < 0) {
627 error_setg(errp, "Invalid reference count table offset");
628 goto fail;
629 }
630
631 /* Snapshot table offset/length */
632 if (header.nb_snapshots > QCOW_MAX_SNAPSHOTS) {
633 error_setg(errp, "Too many snapshots");
634 ret = -EINVAL;
635 goto fail;
636 }
637
638 ret = validate_table_offset(bs, header.snapshots_offset,
639 header.nb_snapshots,
640 sizeof(QCowSnapshotHeader));
641 if (ret < 0) {
642 error_setg(errp, "Invalid snapshot table offset");
643 goto fail;
644 }
645
646 /* read the level 1 table */
647 if (header.l1_size > QCOW_MAX_L1_SIZE) {
648 error_setg(errp, "Active L1 table too large");
649 ret = -EFBIG;
650 goto fail;
651 }
652 s->l1_size = header.l1_size;
653
654 l1_vm_state_index = size_to_l1(s, header.size);
655 if (l1_vm_state_index > INT_MAX) {
656 error_setg(errp, "Image is too big");
657 ret = -EFBIG;
658 goto fail;
659 }
660 s->l1_vm_state_index = l1_vm_state_index;
661
662 /* the L1 table must contain at least enough entries to put
663 header.size bytes */
664 if (s->l1_size < s->l1_vm_state_index) {
665 error_setg(errp, "L1 table is too small");
666 ret = -EINVAL;
667 goto fail;
668 }
669
670 ret = validate_table_offset(bs, header.l1_table_offset,
671 header.l1_size, sizeof(uint64_t));
672 if (ret < 0) {
673 error_setg(errp, "Invalid L1 table offset");
674 goto fail;
675 }
676 s->l1_table_offset = header.l1_table_offset;
677
678
679 if (s->l1_size > 0) {
680 s->l1_table = g_malloc0(
681 align_offset(s->l1_size * sizeof(uint64_t), 512));
682 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
683 s->l1_size * sizeof(uint64_t));
684 if (ret < 0) {
685 error_setg_errno(errp, -ret, "Could not read L1 table");
686 goto fail;
687 }
688 for(i = 0;i < s->l1_size; i++) {
689 be64_to_cpus(&s->l1_table[i]);
690 }
691 }
692
693 /* alloc L2 table/refcount block cache */
694 s->l2_table_cache = qcow2_cache_create(bs, L2_CACHE_SIZE);
695 s->refcount_block_cache = qcow2_cache_create(bs, REFCOUNT_CACHE_SIZE);
696
697 s->cluster_cache = g_malloc(s->cluster_size);
698 /* one more sector for decompressed data alignment */
699 s->cluster_data = qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
700 + 512);
701 s->cluster_cache_offset = -1;
702 s->flags = flags;
703
704 ret = qcow2_refcount_init(bs);
705 if (ret != 0) {
706 error_setg_errno(errp, -ret, "Could not initialize refcount handling");
707 goto fail;
708 }
709
710 QLIST_INIT(&s->cluster_allocs);
711 QTAILQ_INIT(&s->discards);
712
713 /* read qcow2 extensions */
714 if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
715 &local_err)) {
716 error_propagate(errp, local_err);
717 ret = -EINVAL;
718 goto fail;
719 }
720
721 /* read the backing file name */
722 if (header.backing_file_offset != 0) {
723 len = header.backing_file_size;
724 if (len > MIN(1023, s->cluster_size - header.backing_file_offset)) {
725 error_setg(errp, "Backing file name too long");
726 ret = -EINVAL;
727 goto fail;
728 }
729 ret = bdrv_pread(bs->file, header.backing_file_offset,
730 bs->backing_file, len);
731 if (ret < 0) {
732 error_setg_errno(errp, -ret, "Could not read backing file name");
733 goto fail;
734 }
735 bs->backing_file[len] = '\0';
736 }
737
738 /* Internal snapshots */
739 s->snapshots_offset = header.snapshots_offset;
740 s->nb_snapshots = header.nb_snapshots;
741
742 ret = qcow2_read_snapshots(bs);
743 if (ret < 0) {
744 error_setg_errno(errp, -ret, "Could not read snapshots");
745 goto fail;
746 }
747
748 /* Clear unknown autoclear feature bits */
749 if (!bs->read_only && !(flags & BDRV_O_INCOMING) && s->autoclear_features) {
750 s->autoclear_features = 0;
751 ret = qcow2_update_header(bs);
752 if (ret < 0) {
753 error_setg_errno(errp, -ret, "Could not update qcow2 header");
754 goto fail;
755 }
756 }
757
758 /* Initialise locks */
759 qemu_co_mutex_init(&s->lock);
760
761 /* Repair image if dirty */
762 if (!(flags & (BDRV_O_CHECK | BDRV_O_INCOMING)) && !bs->read_only &&
763 (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
764 BdrvCheckResult result = {0};
765
766 ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS);
767 if (ret < 0) {
768 error_setg_errno(errp, -ret, "Could not repair dirty image");
769 goto fail;
770 }
771 }
772
773 /* Enable lazy_refcounts according to image and command line options */
774 opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort);
775 qemu_opts_absorb_qdict(opts, options, &local_err);
776 if (local_err) {
777 error_propagate(errp, local_err);
778 ret = -EINVAL;
779 goto fail;
780 }
781
782 s->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS,
783 (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS));
784
785 s->discard_passthrough[QCOW2_DISCARD_NEVER] = false;
786 s->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true;
787 s->discard_passthrough[QCOW2_DISCARD_REQUEST] =
788 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST,
789 flags & BDRV_O_UNMAP);
790 s->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] =
791 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true);
792 s->discard_passthrough[QCOW2_DISCARD_OTHER] =
793 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false);
794
795 opt_overlap_check = qemu_opt_get(opts, "overlap-check") ?: "cached";
796 if (!strcmp(opt_overlap_check, "none")) {
797 overlap_check_template = 0;
798 } else if (!strcmp(opt_overlap_check, "constant")) {
799 overlap_check_template = QCOW2_OL_CONSTANT;
800 } else if (!strcmp(opt_overlap_check, "cached")) {
801 overlap_check_template = QCOW2_OL_CACHED;
802 } else if (!strcmp(opt_overlap_check, "all")) {
803 overlap_check_template = QCOW2_OL_ALL;
804 } else {
805 error_setg(errp, "Unsupported value '%s' for qcow2 option "
806 "'overlap-check'. Allowed are either of the following: "
807 "none, constant, cached, all", opt_overlap_check);
808 qemu_opts_del(opts);
809 ret = -EINVAL;
810 goto fail;
811 }
812
813 s->overlap_check = 0;
814 for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) {
815 /* overlap-check defines a template bitmask, but every flag may be
816 * overwritten through the associated boolean option */
817 s->overlap_check |=
818 qemu_opt_get_bool(opts, overlap_bool_option_names[i],
819 overlap_check_template & (1 << i)) << i;
820 }
821
822 qemu_opts_del(opts);
823
824 if (s->use_lazy_refcounts && s->qcow_version < 3) {
825 error_setg(errp, "Lazy refcounts require a qcow2 image with at least "
826 "qemu 1.1 compatibility level");
827 ret = -EINVAL;
828 goto fail;
829 }
830
831 #ifdef DEBUG_ALLOC
832 {
833 BdrvCheckResult result = {0};
834 qcow2_check_refcounts(bs, &result, 0);
835 }
836 #endif
837 return ret;
838
839 fail:
840 g_free(s->unknown_header_fields);
841 cleanup_unknown_header_ext(bs);
842 qcow2_free_snapshots(bs);
843 qcow2_refcount_close(bs);
844 g_free(s->l1_table);
845 /* else pre-write overlap checks in cache_destroy may crash */
846 s->l1_table = NULL;
847 if (s->l2_table_cache) {
848 qcow2_cache_destroy(bs, s->l2_table_cache);
849 }
850 if (s->refcount_block_cache) {
851 qcow2_cache_destroy(bs, s->refcount_block_cache);
852 }
853 g_free(s->cluster_cache);
854 qemu_vfree(s->cluster_data);
855 return ret;
856 }
857
858 static int qcow2_refresh_limits(BlockDriverState *bs)
859 {
860 BDRVQcowState *s = bs->opaque;
861
862 bs->bl.write_zeroes_alignment = s->cluster_sectors;
863
864 return 0;
865 }
866
867 static int qcow2_set_key(BlockDriverState *bs, const char *key)
868 {
869 BDRVQcowState *s = bs->opaque;
870 uint8_t keybuf[16];
871 int len, i;
872
873 memset(keybuf, 0, 16);
874 len = strlen(key);
875 if (len > 16)
876 len = 16;
877 /* XXX: we could compress the chars to 7 bits to increase
878 entropy */
879 for(i = 0;i < len;i++) {
880 keybuf[i] = key[i];
881 }
882 s->crypt_method = s->crypt_method_header;
883
884 if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
885 return -1;
886 if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
887 return -1;
888 #if 0
889 /* test */
890 {
891 uint8_t in[16];
892 uint8_t out[16];
893 uint8_t tmp[16];
894 for(i=0;i<16;i++)
895 in[i] = i;
896 AES_encrypt(in, tmp, &s->aes_encrypt_key);
897 AES_decrypt(tmp, out, &s->aes_decrypt_key);
898 for(i = 0; i < 16; i++)
899 printf(" %02x", tmp[i]);
900 printf("\n");
901 for(i = 0; i < 16; i++)
902 printf(" %02x", out[i]);
903 printf("\n");
904 }
905 #endif
906 return 0;
907 }
908
909 /* We have no actual commit/abort logic for qcow2, but we need to write out any
910 * unwritten data if we reopen read-only. */
911 static int qcow2_reopen_prepare(BDRVReopenState *state,
912 BlockReopenQueue *queue, Error **errp)
913 {
914 int ret;
915
916 if ((state->flags & BDRV_O_RDWR) == 0) {
917 ret = bdrv_flush(state->bs);
918 if (ret < 0) {
919 return ret;
920 }
921
922 ret = qcow2_mark_clean(state->bs);
923 if (ret < 0) {
924 return ret;
925 }
926 }
927
928 return 0;
929 }
930
931 static int64_t coroutine_fn qcow2_co_get_block_status(BlockDriverState *bs,
932 int64_t sector_num, int nb_sectors, int *pnum)
933 {
934 BDRVQcowState *s = bs->opaque;
935 uint64_t cluster_offset;
936 int index_in_cluster, ret;
937 int64_t status = 0;
938
939 *pnum = nb_sectors;
940 qemu_co_mutex_lock(&s->lock);
941 ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset);
942 qemu_co_mutex_unlock(&s->lock);
943 if (ret < 0) {
944 return ret;
945 }
946
947 if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED &&
948 !s->crypt_method) {
949 index_in_cluster = sector_num & (s->cluster_sectors - 1);
950 cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
951 status |= BDRV_BLOCK_OFFSET_VALID | cluster_offset;
952 }
953 if (ret == QCOW2_CLUSTER_ZERO) {
954 status |= BDRV_BLOCK_ZERO;
955 } else if (ret != QCOW2_CLUSTER_UNALLOCATED) {
956 status |= BDRV_BLOCK_DATA;
957 }
958 return status;
959 }
960
961 /* handle reading after the end of the backing file */
962 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
963 int64_t sector_num, int nb_sectors)
964 {
965 int n1;
966 if ((sector_num + nb_sectors) <= bs->total_sectors)
967 return nb_sectors;
968 if (sector_num >= bs->total_sectors)
969 n1 = 0;
970 else
971 n1 = bs->total_sectors - sector_num;
972
973 qemu_iovec_memset(qiov, 512 * n1, 0, 512 * (nb_sectors - n1));
974
975 return n1;
976 }
977
978 static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num,
979 int remaining_sectors, QEMUIOVector *qiov)
980 {
981 BDRVQcowState *s = bs->opaque;
982 int index_in_cluster, n1;
983 int ret;
984 int cur_nr_sectors; /* number of sectors in current iteration */
985 uint64_t cluster_offset = 0;
986 uint64_t bytes_done = 0;
987 QEMUIOVector hd_qiov;
988 uint8_t *cluster_data = NULL;
989
990 qemu_iovec_init(&hd_qiov, qiov->niov);
991
992 qemu_co_mutex_lock(&s->lock);
993
994 while (remaining_sectors != 0) {
995
996 /* prepare next request */
997 cur_nr_sectors = remaining_sectors;
998 if (s->crypt_method) {
999 cur_nr_sectors = MIN(cur_nr_sectors,
1000 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
1001 }
1002
1003 ret = qcow2_get_cluster_offset(bs, sector_num << 9,
1004 &cur_nr_sectors, &cluster_offset);
1005 if (ret < 0) {
1006 goto fail;
1007 }
1008
1009 index_in_cluster = sector_num & (s->cluster_sectors - 1);
1010
1011 qemu_iovec_reset(&hd_qiov);
1012 qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1013 cur_nr_sectors * 512);
1014
1015 switch (ret) {
1016 case QCOW2_CLUSTER_UNALLOCATED:
1017
1018 if (bs->backing_hd) {
1019 /* read from the base image */
1020 n1 = qcow2_backing_read1(bs->backing_hd, &hd_qiov,
1021 sector_num, cur_nr_sectors);
1022 if (n1 > 0) {
1023 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1024 qemu_co_mutex_unlock(&s->lock);
1025 ret = bdrv_co_readv(bs->backing_hd, sector_num,
1026 n1, &hd_qiov);
1027 qemu_co_mutex_lock(&s->lock);
1028 if (ret < 0) {
1029 goto fail;
1030 }
1031 }
1032 } else {
1033 /* Note: in this case, no need to wait */
1034 qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
1035 }
1036 break;
1037
1038 case QCOW2_CLUSTER_ZERO:
1039 qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
1040 break;
1041
1042 case QCOW2_CLUSTER_COMPRESSED:
1043 /* add AIO support for compressed blocks ? */
1044 ret = qcow2_decompress_cluster(bs, cluster_offset);
1045 if (ret < 0) {
1046 goto fail;
1047 }
1048
1049 qemu_iovec_from_buf(&hd_qiov, 0,
1050 s->cluster_cache + index_in_cluster * 512,
1051 512 * cur_nr_sectors);
1052 break;
1053
1054 case QCOW2_CLUSTER_NORMAL:
1055 if ((cluster_offset & 511) != 0) {
1056 ret = -EIO;
1057 goto fail;
1058 }
1059
1060 if (s->crypt_method) {
1061 /*
1062 * For encrypted images, read everything into a temporary
1063 * contiguous buffer on which the AES functions can work.
1064 */
1065 if (!cluster_data) {
1066 cluster_data =
1067 qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1068 }
1069
1070 assert(cur_nr_sectors <=
1071 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
1072 qemu_iovec_reset(&hd_qiov);
1073 qemu_iovec_add(&hd_qiov, cluster_data,
1074 512 * cur_nr_sectors);
1075 }
1076
1077 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
1078 qemu_co_mutex_unlock(&s->lock);
1079 ret = bdrv_co_readv(bs->file,
1080 (cluster_offset >> 9) + index_in_cluster,
1081 cur_nr_sectors, &hd_qiov);
1082 qemu_co_mutex_lock(&s->lock);
1083 if (ret < 0) {
1084 goto fail;
1085 }
1086 if (s->crypt_method) {
1087 qcow2_encrypt_sectors(s, sector_num, cluster_data,
1088 cluster_data, cur_nr_sectors, 0, &s->aes_decrypt_key);
1089 qemu_iovec_from_buf(qiov, bytes_done,
1090 cluster_data, 512 * cur_nr_sectors);
1091 }
1092 break;
1093
1094 default:
1095 g_assert_not_reached();
1096 ret = -EIO;
1097 goto fail;
1098 }
1099
1100 remaining_sectors -= cur_nr_sectors;
1101 sector_num += cur_nr_sectors;
1102 bytes_done += cur_nr_sectors * 512;
1103 }
1104 ret = 0;
1105
1106 fail:
1107 qemu_co_mutex_unlock(&s->lock);
1108
1109 qemu_iovec_destroy(&hd_qiov);
1110 qemu_vfree(cluster_data);
1111
1112 return ret;
1113 }
1114
1115 static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
1116 int64_t sector_num,
1117 int remaining_sectors,
1118 QEMUIOVector *qiov)
1119 {
1120 BDRVQcowState *s = bs->opaque;
1121 int index_in_cluster;
1122 int ret;
1123 int cur_nr_sectors; /* number of sectors in current iteration */
1124 uint64_t cluster_offset;
1125 QEMUIOVector hd_qiov;
1126 uint64_t bytes_done = 0;
1127 uint8_t *cluster_data = NULL;
1128 QCowL2Meta *l2meta = NULL;
1129
1130 trace_qcow2_writev_start_req(qemu_coroutine_self(), sector_num,
1131 remaining_sectors);
1132
1133 qemu_iovec_init(&hd_qiov, qiov->niov);
1134
1135 s->cluster_cache_offset = -1; /* disable compressed cache */
1136
1137 qemu_co_mutex_lock(&s->lock);
1138
1139 while (remaining_sectors != 0) {
1140
1141 l2meta = NULL;
1142
1143 trace_qcow2_writev_start_part(qemu_coroutine_self());
1144 index_in_cluster = sector_num & (s->cluster_sectors - 1);
1145 cur_nr_sectors = remaining_sectors;
1146 if (s->crypt_method &&
1147 cur_nr_sectors >
1148 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors - index_in_cluster) {
1149 cur_nr_sectors =
1150 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors - index_in_cluster;
1151 }
1152
1153 ret = qcow2_alloc_cluster_offset(bs, sector_num << 9,
1154 &cur_nr_sectors, &cluster_offset, &l2meta);
1155 if (ret < 0) {
1156 goto fail;
1157 }
1158
1159 assert((cluster_offset & 511) == 0);
1160
1161 qemu_iovec_reset(&hd_qiov);
1162 qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1163 cur_nr_sectors * 512);
1164
1165 if (s->crypt_method) {
1166 if (!cluster_data) {
1167 cluster_data = qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS *
1168 s->cluster_size);
1169 }
1170
1171 assert(hd_qiov.size <=
1172 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1173 qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size);
1174
1175 qcow2_encrypt_sectors(s, sector_num, cluster_data,
1176 cluster_data, cur_nr_sectors, 1, &s->aes_encrypt_key);
1177
1178 qemu_iovec_reset(&hd_qiov);
1179 qemu_iovec_add(&hd_qiov, cluster_data,
1180 cur_nr_sectors * 512);
1181 }
1182
1183 ret = qcow2_pre_write_overlap_check(bs, 0,
1184 cluster_offset + index_in_cluster * BDRV_SECTOR_SIZE,
1185 cur_nr_sectors * BDRV_SECTOR_SIZE);
1186 if (ret < 0) {
1187 goto fail;
1188 }
1189
1190 qemu_co_mutex_unlock(&s->lock);
1191 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
1192 trace_qcow2_writev_data(qemu_coroutine_self(),
1193 (cluster_offset >> 9) + index_in_cluster);
1194 ret = bdrv_co_writev(bs->file,
1195 (cluster_offset >> 9) + index_in_cluster,
1196 cur_nr_sectors, &hd_qiov);
1197 qemu_co_mutex_lock(&s->lock);
1198 if (ret < 0) {
1199 goto fail;
1200 }
1201
1202 while (l2meta != NULL) {
1203 QCowL2Meta *next;
1204
1205 ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
1206 if (ret < 0) {
1207 goto fail;
1208 }
1209
1210 /* Take the request off the list of running requests */
1211 if (l2meta->nb_clusters != 0) {
1212 QLIST_REMOVE(l2meta, next_in_flight);
1213 }
1214
1215 qemu_co_queue_restart_all(&l2meta->dependent_requests);
1216
1217 next = l2meta->next;
1218 g_free(l2meta);
1219 l2meta = next;
1220 }
1221
1222 remaining_sectors -= cur_nr_sectors;
1223 sector_num += cur_nr_sectors;
1224 bytes_done += cur_nr_sectors * 512;
1225 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_nr_sectors);
1226 }
1227 ret = 0;
1228
1229 fail:
1230 qemu_co_mutex_unlock(&s->lock);
1231
1232 while (l2meta != NULL) {
1233 QCowL2Meta *next;
1234
1235 if (l2meta->nb_clusters != 0) {
1236 QLIST_REMOVE(l2meta, next_in_flight);
1237 }
1238 qemu_co_queue_restart_all(&l2meta->dependent_requests);
1239
1240 next = l2meta->next;
1241 g_free(l2meta);
1242 l2meta = next;
1243 }
1244
1245 qemu_iovec_destroy(&hd_qiov);
1246 qemu_vfree(cluster_data);
1247 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
1248
1249 return ret;
1250 }
1251
1252 static void qcow2_close(BlockDriverState *bs)
1253 {
1254 BDRVQcowState *s = bs->opaque;
1255 g_free(s->l1_table);
1256 /* else pre-write overlap checks in cache_destroy may crash */
1257 s->l1_table = NULL;
1258
1259 if (!(bs->open_flags & BDRV_O_INCOMING)) {
1260 qcow2_cache_flush(bs, s->l2_table_cache);
1261 qcow2_cache_flush(bs, s->refcount_block_cache);
1262
1263 qcow2_mark_clean(bs);
1264 }
1265
1266 qcow2_cache_destroy(bs, s->l2_table_cache);
1267 qcow2_cache_destroy(bs, s->refcount_block_cache);
1268
1269 g_free(s->unknown_header_fields);
1270 cleanup_unknown_header_ext(bs);
1271
1272 g_free(s->cluster_cache);
1273 qemu_vfree(s->cluster_data);
1274 qcow2_refcount_close(bs);
1275 qcow2_free_snapshots(bs);
1276 }
1277
1278 static void qcow2_invalidate_cache(BlockDriverState *bs, Error **errp)
1279 {
1280 BDRVQcowState *s = bs->opaque;
1281 int flags = s->flags;
1282 AES_KEY aes_encrypt_key;
1283 AES_KEY aes_decrypt_key;
1284 uint32_t crypt_method = 0;
1285 QDict *options;
1286 Error *local_err = NULL;
1287 int ret;
1288
1289 /*
1290 * Backing files are read-only which makes all of their metadata immutable,
1291 * that means we don't have to worry about reopening them here.
1292 */
1293
1294 if (s->crypt_method) {
1295 crypt_method = s->crypt_method;
1296 memcpy(&aes_encrypt_key, &s->aes_encrypt_key, sizeof(aes_encrypt_key));
1297 memcpy(&aes_decrypt_key, &s->aes_decrypt_key, sizeof(aes_decrypt_key));
1298 }
1299
1300 qcow2_close(bs);
1301
1302 bdrv_invalidate_cache(bs->file, &local_err);
1303 if (local_err) {
1304 error_propagate(errp, local_err);
1305 return;
1306 }
1307
1308 memset(s, 0, sizeof(BDRVQcowState));
1309 options = qdict_clone_shallow(bs->options);
1310
1311 ret = qcow2_open(bs, options, flags, &local_err);
1312 QDECREF(options);
1313 if (local_err) {
1314 error_setg(errp, "Could not reopen qcow2 layer: %s",
1315 error_get_pretty(local_err));
1316 error_free(local_err);
1317 return;
1318 } else if (ret < 0) {
1319 error_setg_errno(errp, -ret, "Could not reopen qcow2 layer");
1320 return;
1321 }
1322
1323 if (crypt_method) {
1324 s->crypt_method = crypt_method;
1325 memcpy(&s->aes_encrypt_key, &aes_encrypt_key, sizeof(aes_encrypt_key));
1326 memcpy(&s->aes_decrypt_key, &aes_decrypt_key, sizeof(aes_decrypt_key));
1327 }
1328 }
1329
1330 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
1331 size_t len, size_t buflen)
1332 {
1333 QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
1334 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
1335
1336 if (buflen < ext_len) {
1337 return -ENOSPC;
1338 }
1339
1340 *ext_backing_fmt = (QCowExtension) {
1341 .magic = cpu_to_be32(magic),
1342 .len = cpu_to_be32(len),
1343 };
1344 memcpy(buf + sizeof(QCowExtension), s, len);
1345
1346 return ext_len;
1347 }
1348
1349 /*
1350 * Updates the qcow2 header, including the variable length parts of it, i.e.
1351 * the backing file name and all extensions. qcow2 was not designed to allow
1352 * such changes, so if we run out of space (we can only use the first cluster)
1353 * this function may fail.
1354 *
1355 * Returns 0 on success, -errno in error cases.
1356 */
1357 int qcow2_update_header(BlockDriverState *bs)
1358 {
1359 BDRVQcowState *s = bs->opaque;
1360 QCowHeader *header;
1361 char *buf;
1362 size_t buflen = s->cluster_size;
1363 int ret;
1364 uint64_t total_size;
1365 uint32_t refcount_table_clusters;
1366 size_t header_length;
1367 Qcow2UnknownHeaderExtension *uext;
1368
1369 buf = qemu_blockalign(bs, buflen);
1370
1371 /* Header structure */
1372 header = (QCowHeader*) buf;
1373
1374 if (buflen < sizeof(*header)) {
1375 ret = -ENOSPC;
1376 goto fail;
1377 }
1378
1379 header_length = sizeof(*header) + s->unknown_header_fields_size;
1380 total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
1381 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
1382
1383 *header = (QCowHeader) {
1384 /* Version 2 fields */
1385 .magic = cpu_to_be32(QCOW_MAGIC),
1386 .version = cpu_to_be32(s->qcow_version),
1387 .backing_file_offset = 0,
1388 .backing_file_size = 0,
1389 .cluster_bits = cpu_to_be32(s->cluster_bits),
1390 .size = cpu_to_be64(total_size),
1391 .crypt_method = cpu_to_be32(s->crypt_method_header),
1392 .l1_size = cpu_to_be32(s->l1_size),
1393 .l1_table_offset = cpu_to_be64(s->l1_table_offset),
1394 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset),
1395 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
1396 .nb_snapshots = cpu_to_be32(s->nb_snapshots),
1397 .snapshots_offset = cpu_to_be64(s->snapshots_offset),
1398
1399 /* Version 3 fields */
1400 .incompatible_features = cpu_to_be64(s->incompatible_features),
1401 .compatible_features = cpu_to_be64(s->compatible_features),
1402 .autoclear_features = cpu_to_be64(s->autoclear_features),
1403 .refcount_order = cpu_to_be32(s->refcount_order),
1404 .header_length = cpu_to_be32(header_length),
1405 };
1406
1407 /* For older versions, write a shorter header */
1408 switch (s->qcow_version) {
1409 case 2:
1410 ret = offsetof(QCowHeader, incompatible_features);
1411 break;
1412 case 3:
1413 ret = sizeof(*header);
1414 break;
1415 default:
1416 ret = -EINVAL;
1417 goto fail;
1418 }
1419
1420 buf += ret;
1421 buflen -= ret;
1422 memset(buf, 0, buflen);
1423
1424 /* Preserve any unknown field in the header */
1425 if (s->unknown_header_fields_size) {
1426 if (buflen < s->unknown_header_fields_size) {
1427 ret = -ENOSPC;
1428 goto fail;
1429 }
1430
1431 memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
1432 buf += s->unknown_header_fields_size;
1433 buflen -= s->unknown_header_fields_size;
1434 }
1435
1436 /* Backing file format header extension */
1437 if (*bs->backing_format) {
1438 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
1439 bs->backing_format, strlen(bs->backing_format),
1440 buflen);
1441 if (ret < 0) {
1442 goto fail;
1443 }
1444
1445 buf += ret;
1446 buflen -= ret;
1447 }
1448
1449 /* Feature table */
1450 Qcow2Feature features[] = {
1451 {
1452 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1453 .bit = QCOW2_INCOMPAT_DIRTY_BITNR,
1454 .name = "dirty bit",
1455 },
1456 {
1457 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1458 .bit = QCOW2_INCOMPAT_CORRUPT_BITNR,
1459 .name = "corrupt bit",
1460 },
1461 {
1462 .type = QCOW2_FEAT_TYPE_COMPATIBLE,
1463 .bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
1464 .name = "lazy refcounts",
1465 },
1466 };
1467
1468 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
1469 features, sizeof(features), buflen);
1470 if (ret < 0) {
1471 goto fail;
1472 }
1473 buf += ret;
1474 buflen -= ret;
1475
1476 /* Keep unknown header extensions */
1477 QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
1478 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
1479 if (ret < 0) {
1480 goto fail;
1481 }
1482
1483 buf += ret;
1484 buflen -= ret;
1485 }
1486
1487 /* End of header extensions */
1488 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
1489 if (ret < 0) {
1490 goto fail;
1491 }
1492
1493 buf += ret;
1494 buflen -= ret;
1495
1496 /* Backing file name */
1497 if (*bs->backing_file) {
1498 size_t backing_file_len = strlen(bs->backing_file);
1499
1500 if (buflen < backing_file_len) {
1501 ret = -ENOSPC;
1502 goto fail;
1503 }
1504
1505 /* Using strncpy is ok here, since buf is not NUL-terminated. */
1506 strncpy(buf, bs->backing_file, buflen);
1507
1508 header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
1509 header->backing_file_size = cpu_to_be32(backing_file_len);
1510 }
1511
1512 /* Write the new header */
1513 ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
1514 if (ret < 0) {
1515 goto fail;
1516 }
1517
1518 ret = 0;
1519 fail:
1520 qemu_vfree(header);
1521 return ret;
1522 }
1523
1524 static int qcow2_change_backing_file(BlockDriverState *bs,
1525 const char *backing_file, const char *backing_fmt)
1526 {
1527 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
1528 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
1529
1530 return qcow2_update_header(bs);
1531 }
1532
1533 static int preallocate(BlockDriverState *bs)
1534 {
1535 uint64_t nb_sectors;
1536 uint64_t offset;
1537 uint64_t host_offset = 0;
1538 int num;
1539 int ret;
1540 QCowL2Meta *meta;
1541
1542 nb_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
1543 offset = 0;
1544
1545 while (nb_sectors) {
1546 num = MIN(nb_sectors, INT_MAX >> BDRV_SECTOR_BITS);
1547 ret = qcow2_alloc_cluster_offset(bs, offset, &num,
1548 &host_offset, &meta);
1549 if (ret < 0) {
1550 return ret;
1551 }
1552
1553 while (meta) {
1554 QCowL2Meta *next = meta->next;
1555
1556 ret = qcow2_alloc_cluster_link_l2(bs, meta);
1557 if (ret < 0) {
1558 qcow2_free_any_clusters(bs, meta->alloc_offset,
1559 meta->nb_clusters, QCOW2_DISCARD_NEVER);
1560 return ret;
1561 }
1562
1563 /* There are no dependent requests, but we need to remove our
1564 * request from the list of in-flight requests */
1565 QLIST_REMOVE(meta, next_in_flight);
1566
1567 g_free(meta);
1568 meta = next;
1569 }
1570
1571 /* TODO Preallocate data if requested */
1572
1573 nb_sectors -= num;
1574 offset += num << BDRV_SECTOR_BITS;
1575 }
1576
1577 /*
1578 * It is expected that the image file is large enough to actually contain
1579 * all of the allocated clusters (otherwise we get failing reads after
1580 * EOF). Extend the image to the last allocated sector.
1581 */
1582 if (host_offset != 0) {
1583 uint8_t buf[BDRV_SECTOR_SIZE];
1584 memset(buf, 0, BDRV_SECTOR_SIZE);
1585 ret = bdrv_write(bs->file, (host_offset >> BDRV_SECTOR_BITS) + num - 1,
1586 buf, 1);
1587 if (ret < 0) {
1588 return ret;
1589 }
1590 }
1591
1592 return 0;
1593 }
1594
1595 static int qcow2_create2(const char *filename, int64_t total_size,
1596 const char *backing_file, const char *backing_format,
1597 int flags, size_t cluster_size, int prealloc,
1598 QemuOpts *opts, int version,
1599 Error **errp)
1600 {
1601 /* Calculate cluster_bits */
1602 int cluster_bits;
1603 cluster_bits = ffs(cluster_size) - 1;
1604 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
1605 (1 << cluster_bits) != cluster_size)
1606 {
1607 error_setg(errp, "Cluster size must be a power of two between %d and "
1608 "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
1609 return -EINVAL;
1610 }
1611
1612 /*
1613 * Open the image file and write a minimal qcow2 header.
1614 *
1615 * We keep things simple and start with a zero-sized image. We also
1616 * do without refcount blocks or a L1 table for now. We'll fix the
1617 * inconsistency later.
1618 *
1619 * We do need a refcount table because growing the refcount table means
1620 * allocating two new refcount blocks - the seconds of which would be at
1621 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
1622 * size for any qcow2 image.
1623 */
1624 BlockDriverState* bs;
1625 QCowHeader *header;
1626 uint64_t* refcount_table;
1627 Error *local_err = NULL;
1628 int ret;
1629
1630 ret = bdrv_create_file(filename, opts, &local_err);
1631 if (ret < 0) {
1632 error_propagate(errp, local_err);
1633 return ret;
1634 }
1635
1636 bs = NULL;
1637 ret = bdrv_open(&bs, filename, NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL,
1638 NULL, &local_err);
1639 if (ret < 0) {
1640 error_propagate(errp, local_err);
1641 return ret;
1642 }
1643
1644 /* Write the header */
1645 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
1646 header = g_malloc0(cluster_size);
1647 *header = (QCowHeader) {
1648 .magic = cpu_to_be32(QCOW_MAGIC),
1649 .version = cpu_to_be32(version),
1650 .cluster_bits = cpu_to_be32(cluster_bits),
1651 .size = cpu_to_be64(0),
1652 .l1_table_offset = cpu_to_be64(0),
1653 .l1_size = cpu_to_be32(0),
1654 .refcount_table_offset = cpu_to_be64(cluster_size),
1655 .refcount_table_clusters = cpu_to_be32(1),
1656 .refcount_order = cpu_to_be32(3 + REFCOUNT_SHIFT),
1657 .header_length = cpu_to_be32(sizeof(*header)),
1658 };
1659
1660 if (flags & BLOCK_FLAG_ENCRYPT) {
1661 header->crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
1662 } else {
1663 header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
1664 }
1665
1666 if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) {
1667 header->compatible_features |=
1668 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
1669 }
1670
1671 ret = bdrv_pwrite(bs, 0, header, cluster_size);
1672 g_free(header);
1673 if (ret < 0) {
1674 error_setg_errno(errp, -ret, "Could not write qcow2 header");
1675 goto out;
1676 }
1677
1678 /* Write a refcount table with one refcount block */
1679 refcount_table = g_malloc0(2 * cluster_size);
1680 refcount_table[0] = cpu_to_be64(2 * cluster_size);
1681 ret = bdrv_pwrite(bs, cluster_size, refcount_table, 2 * cluster_size);
1682 g_free(refcount_table);
1683
1684 if (ret < 0) {
1685 error_setg_errno(errp, -ret, "Could not write refcount table");
1686 goto out;
1687 }
1688
1689 bdrv_unref(bs);
1690 bs = NULL;
1691
1692 /*
1693 * And now open the image and make it consistent first (i.e. increase the
1694 * refcount of the cluster that is occupied by the header and the refcount
1695 * table)
1696 */
1697 BlockDriver* drv = bdrv_find_format("qcow2");
1698 assert(drv != NULL);
1699 ret = bdrv_open(&bs, filename, NULL, NULL,
1700 BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_FLUSH, drv, &local_err);
1701 if (ret < 0) {
1702 error_propagate(errp, local_err);
1703 goto out;
1704 }
1705
1706 ret = qcow2_alloc_clusters(bs, 3 * cluster_size);
1707 if (ret < 0) {
1708 error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
1709 "header and refcount table");
1710 goto out;
1711
1712 } else if (ret != 0) {
1713 error_report("Huh, first cluster in empty image is already in use?");
1714 abort();
1715 }
1716
1717 /* Okay, now that we have a valid image, let's give it the right size */
1718 ret = bdrv_truncate(bs, total_size * BDRV_SECTOR_SIZE);
1719 if (ret < 0) {
1720 error_setg_errno(errp, -ret, "Could not resize image");
1721 goto out;
1722 }
1723
1724 /* Want a backing file? There you go.*/
1725 if (backing_file) {
1726 ret = bdrv_change_backing_file(bs, backing_file, backing_format);
1727 if (ret < 0) {
1728 error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
1729 "with format '%s'", backing_file, backing_format);
1730 goto out;
1731 }
1732 }
1733
1734 /* And if we're supposed to preallocate metadata, do that now */
1735 if (prealloc) {
1736 BDRVQcowState *s = bs->opaque;
1737 qemu_co_mutex_lock(&s->lock);
1738 ret = preallocate(bs);
1739 qemu_co_mutex_unlock(&s->lock);
1740 if (ret < 0) {
1741 error_setg_errno(errp, -ret, "Could not preallocate metadata");
1742 goto out;
1743 }
1744 }
1745
1746 bdrv_unref(bs);
1747 bs = NULL;
1748
1749 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning */
1750 ret = bdrv_open(&bs, filename, NULL, NULL,
1751 BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_BACKING,
1752 drv, &local_err);
1753 if (local_err) {
1754 error_propagate(errp, local_err);
1755 goto out;
1756 }
1757
1758 ret = 0;
1759 out:
1760 if (bs) {
1761 bdrv_unref(bs);
1762 }
1763 return ret;
1764 }
1765
1766 static int qcow2_create(const char *filename, QemuOpts *opts, Error **errp)
1767 {
1768 char *backing_file = NULL;
1769 char *backing_fmt = NULL;
1770 char *buf = NULL;
1771 uint64_t sectors = 0;
1772 int flags = 0;
1773 size_t cluster_size = DEFAULT_CLUSTER_SIZE;
1774 int prealloc = 0;
1775 int version = 3;
1776 Error *local_err = NULL;
1777 int ret;
1778
1779 /* Read out options */
1780 sectors = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0) / 512;
1781 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
1782 backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
1783 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
1784 flags |= BLOCK_FLAG_ENCRYPT;
1785 }
1786 cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
1787 DEFAULT_CLUSTER_SIZE);
1788 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1789 if (!buf || !strcmp(buf, "off")) {
1790 prealloc = 0;
1791 } else if (!strcmp(buf, "metadata")) {
1792 prealloc = 1;
1793 } else {
1794 error_setg(errp, "Invalid preallocation mode: '%s'", buf);
1795 ret = -EINVAL;
1796 goto finish;
1797 }
1798 g_free(buf);
1799 buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
1800 if (!buf) {
1801 /* keep the default */
1802 } else if (!strcmp(buf, "0.10")) {
1803 version = 2;
1804 } else if (!strcmp(buf, "1.1")) {
1805 version = 3;
1806 } else {
1807 error_setg(errp, "Invalid compatibility level: '%s'", buf);
1808 ret = -EINVAL;
1809 goto finish;
1810 }
1811
1812 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_LAZY_REFCOUNTS, false)) {
1813 flags |= BLOCK_FLAG_LAZY_REFCOUNTS;
1814 }
1815
1816 if (backing_file && prealloc) {
1817 error_setg(errp, "Backing file and preallocation cannot be used at "
1818 "the same time");
1819 ret = -EINVAL;
1820 goto finish;
1821 }
1822
1823 if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) {
1824 error_setg(errp, "Lazy refcounts only supported with compatibility "
1825 "level 1.1 and above (use compat=1.1 or greater)");
1826 ret = -EINVAL;
1827 goto finish;
1828 }
1829
1830 ret = qcow2_create2(filename, sectors, backing_file, backing_fmt, flags,
1831 cluster_size, prealloc, opts, version, &local_err);
1832 if (local_err) {
1833 error_propagate(errp, local_err);
1834 }
1835
1836 finish:
1837 g_free(backing_file);
1838 g_free(backing_fmt);
1839 g_free(buf);
1840 return ret;
1841 }
1842
1843 static coroutine_fn int qcow2_co_write_zeroes(BlockDriverState *bs,
1844 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
1845 {
1846 int ret;
1847 BDRVQcowState *s = bs->opaque;
1848
1849 /* Emulate misaligned zero writes */
1850 if (sector_num % s->cluster_sectors || nb_sectors % s->cluster_sectors) {
1851 return -ENOTSUP;
1852 }
1853
1854 /* Whatever is left can use real zero clusters */
1855 qemu_co_mutex_lock(&s->lock);
1856 ret = qcow2_zero_clusters(bs, sector_num << BDRV_SECTOR_BITS,
1857 nb_sectors);
1858 qemu_co_mutex_unlock(&s->lock);
1859
1860 return ret;
1861 }
1862
1863 static coroutine_fn int qcow2_co_discard(BlockDriverState *bs,
1864 int64_t sector_num, int nb_sectors)
1865 {
1866 int ret;
1867 BDRVQcowState *s = bs->opaque;
1868
1869 qemu_co_mutex_lock(&s->lock);
1870 ret = qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS,
1871 nb_sectors, QCOW2_DISCARD_REQUEST);
1872 qemu_co_mutex_unlock(&s->lock);
1873 return ret;
1874 }
1875
1876 static int qcow2_truncate(BlockDriverState *bs, int64_t offset)
1877 {
1878 BDRVQcowState *s = bs->opaque;
1879 int64_t new_l1_size;
1880 int ret;
1881
1882 if (offset & 511) {
1883 error_report("The new size must be a multiple of 512");
1884 return -EINVAL;
1885 }
1886
1887 /* cannot proceed if image has snapshots */
1888 if (s->nb_snapshots) {
1889 error_report("Can't resize an image which has snapshots");
1890 return -ENOTSUP;
1891 }
1892
1893 /* shrinking is currently not supported */
1894 if (offset < bs->total_sectors * 512) {
1895 error_report("qcow2 doesn't support shrinking images yet");
1896 return -ENOTSUP;
1897 }
1898
1899 new_l1_size = size_to_l1(s, offset);
1900 ret = qcow2_grow_l1_table(bs, new_l1_size, true);
1901 if (ret < 0) {
1902 return ret;
1903 }
1904
1905 /* write updated header.size */
1906 offset = cpu_to_be64(offset);
1907 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
1908 &offset, sizeof(uint64_t));
1909 if (ret < 0) {
1910 return ret;
1911 }
1912
1913 s->l1_vm_state_index = new_l1_size;
1914 return 0;
1915 }
1916
1917 /* XXX: put compressed sectors first, then all the cluster aligned
1918 tables to avoid losing bytes in alignment */
1919 static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num,
1920 const uint8_t *buf, int nb_sectors)
1921 {
1922 BDRVQcowState *s = bs->opaque;
1923 z_stream strm;
1924 int ret, out_len;
1925 uint8_t *out_buf;
1926 uint64_t cluster_offset;
1927
1928 if (nb_sectors == 0) {
1929 /* align end of file to a sector boundary to ease reading with
1930 sector based I/Os */
1931 cluster_offset = bdrv_getlength(bs->file);
1932 cluster_offset = (cluster_offset + 511) & ~511;
1933 bdrv_truncate(bs->file, cluster_offset);
1934 return 0;
1935 }
1936
1937 if (nb_sectors != s->cluster_sectors) {
1938 ret = -EINVAL;
1939
1940 /* Zero-pad last write if image size is not cluster aligned */
1941 if (sector_num + nb_sectors == bs->total_sectors &&
1942 nb_sectors < s->cluster_sectors) {
1943 uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size);
1944 memset(pad_buf, 0, s->cluster_size);
1945 memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE);
1946 ret = qcow2_write_compressed(bs, sector_num,
1947 pad_buf, s->cluster_sectors);
1948 qemu_vfree(pad_buf);
1949 }
1950 return ret;
1951 }
1952
1953 out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1954
1955 /* best compression, small window, no zlib header */
1956 memset(&strm, 0, sizeof(strm));
1957 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1958 Z_DEFLATED, -12,
1959 9, Z_DEFAULT_STRATEGY);
1960 if (ret != 0) {
1961 ret = -EINVAL;
1962 goto fail;
1963 }
1964
1965 strm.avail_in = s->cluster_size;
1966 strm.next_in = (uint8_t *)buf;
1967 strm.avail_out = s->cluster_size;
1968 strm.next_out = out_buf;
1969
1970 ret = deflate(&strm, Z_FINISH);
1971 if (ret != Z_STREAM_END && ret != Z_OK) {
1972 deflateEnd(&strm);
1973 ret = -EINVAL;
1974 goto fail;
1975 }
1976 out_len = strm.next_out - out_buf;
1977
1978 deflateEnd(&strm);
1979
1980 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1981 /* could not compress: write normal cluster */
1982 ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
1983 if (ret < 0) {
1984 goto fail;
1985 }
1986 } else {
1987 cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
1988 sector_num << 9, out_len);
1989 if (!cluster_offset) {
1990 ret = -EIO;
1991 goto fail;
1992 }
1993 cluster_offset &= s->cluster_offset_mask;
1994
1995 ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len);
1996 if (ret < 0) {
1997 goto fail;
1998 }
1999
2000 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
2001 ret = bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len);
2002 if (ret < 0) {
2003 goto fail;
2004 }
2005 }
2006
2007 ret = 0;
2008 fail:
2009 g_free(out_buf);
2010 return ret;
2011 }
2012
2013 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
2014 {
2015 BDRVQcowState *s = bs->opaque;
2016 int ret;
2017
2018 qemu_co_mutex_lock(&s->lock);
2019 ret = qcow2_cache_flush(bs, s->l2_table_cache);
2020 if (ret < 0) {
2021 qemu_co_mutex_unlock(&s->lock);
2022 return ret;
2023 }
2024
2025 if (qcow2_need_accurate_refcounts(s)) {
2026 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2027 if (ret < 0) {
2028 qemu_co_mutex_unlock(&s->lock);
2029 return ret;
2030 }
2031 }
2032 qemu_co_mutex_unlock(&s->lock);
2033
2034 return 0;
2035 }
2036
2037 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2038 {
2039 BDRVQcowState *s = bs->opaque;
2040 bdi->unallocated_blocks_are_zero = true;
2041 bdi->can_write_zeroes_with_unmap = (s->qcow_version >= 3);
2042 bdi->cluster_size = s->cluster_size;
2043 bdi->vm_state_offset = qcow2_vm_state_offset(s);
2044 return 0;
2045 }
2046
2047 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs)
2048 {
2049 BDRVQcowState *s = bs->opaque;
2050 ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1);
2051
2052 *spec_info = (ImageInfoSpecific){
2053 .kind = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
2054 {
2055 .qcow2 = g_new(ImageInfoSpecificQCow2, 1),
2056 },
2057 };
2058 if (s->qcow_version == 2) {
2059 *spec_info->qcow2 = (ImageInfoSpecificQCow2){
2060 .compat = g_strdup("0.10"),
2061 };
2062 } else if (s->qcow_version == 3) {
2063 *spec_info->qcow2 = (ImageInfoSpecificQCow2){
2064 .compat = g_strdup("1.1"),
2065 .lazy_refcounts = s->compatible_features &
2066 QCOW2_COMPAT_LAZY_REFCOUNTS,
2067 .has_lazy_refcounts = true,
2068 };
2069 }
2070
2071 return spec_info;
2072 }
2073
2074 #if 0
2075 static void dump_refcounts(BlockDriverState *bs)
2076 {
2077 BDRVQcowState *s = bs->opaque;
2078 int64_t nb_clusters, k, k1, size;
2079 int refcount;
2080
2081 size = bdrv_getlength(bs->file);
2082 nb_clusters = size_to_clusters(s, size);
2083 for(k = 0; k < nb_clusters;) {
2084 k1 = k;
2085 refcount = get_refcount(bs, k);
2086 k++;
2087 while (k < nb_clusters && get_refcount(bs, k) == refcount)
2088 k++;
2089 printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
2090 k - k1);
2091 }
2092 }
2093 #endif
2094
2095 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
2096 int64_t pos)
2097 {
2098 BDRVQcowState *s = bs->opaque;
2099 int64_t total_sectors = bs->total_sectors;
2100 int growable = bs->growable;
2101 bool zero_beyond_eof = bs->zero_beyond_eof;
2102 int ret;
2103
2104 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
2105 bs->growable = 1;
2106 bs->zero_beyond_eof = false;
2107 ret = bdrv_pwritev(bs, qcow2_vm_state_offset(s) + pos, qiov);
2108 bs->growable = growable;
2109 bs->zero_beyond_eof = zero_beyond_eof;
2110
2111 /* bdrv_co_do_writev will have increased the total_sectors value to include
2112 * the VM state - the VM state is however not an actual part of the block
2113 * device, therefore, we need to restore the old value. */
2114 bs->total_sectors = total_sectors;
2115
2116 return ret;
2117 }
2118
2119 static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf,
2120 int64_t pos, int size)
2121 {
2122 BDRVQcowState *s = bs->opaque;
2123 int growable = bs->growable;
2124 bool zero_beyond_eof = bs->zero_beyond_eof;
2125 int ret;
2126
2127 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
2128 bs->growable = 1;
2129 bs->zero_beyond_eof = false;
2130 ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size);
2131 bs->growable = growable;
2132 bs->zero_beyond_eof = zero_beyond_eof;
2133
2134 return ret;
2135 }
2136
2137 /*
2138 * Downgrades an image's version. To achieve this, any incompatible features
2139 * have to be removed.
2140 */
2141 static int qcow2_downgrade(BlockDriverState *bs, int target_version)
2142 {
2143 BDRVQcowState *s = bs->opaque;
2144 int current_version = s->qcow_version;
2145 int ret;
2146
2147 if (target_version == current_version) {
2148 return 0;
2149 } else if (target_version > current_version) {
2150 return -EINVAL;
2151 } else if (target_version != 2) {
2152 return -EINVAL;
2153 }
2154
2155 if (s->refcount_order != 4) {
2156 /* we would have to convert the image to a refcount_order == 4 image
2157 * here; however, since qemu (at the time of writing this) does not
2158 * support anything different than 4 anyway, there is no point in doing
2159 * so right now; however, we should error out (if qemu supports this in
2160 * the future and this code has not been adapted) */
2161 error_report("qcow2_downgrade: Image refcount orders other than 4 are "
2162 "currently not supported.");
2163 return -ENOTSUP;
2164 }
2165
2166 /* clear incompatible features */
2167 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
2168 ret = qcow2_mark_clean(bs);
2169 if (ret < 0) {
2170 return ret;
2171 }
2172 }
2173
2174 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
2175 * the first place; if that happens nonetheless, returning -ENOTSUP is the
2176 * best thing to do anyway */
2177
2178 if (s->incompatible_features) {
2179 return -ENOTSUP;
2180 }
2181
2182 /* since we can ignore compatible features, we can set them to 0 as well */
2183 s->compatible_features = 0;
2184 /* if lazy refcounts have been used, they have already been fixed through
2185 * clearing the dirty flag */
2186
2187 /* clearing autoclear features is trivial */
2188 s->autoclear_features = 0;
2189
2190 ret = qcow2_expand_zero_clusters(bs);
2191 if (ret < 0) {
2192 return ret;
2193 }
2194
2195 s->qcow_version = target_version;
2196 ret = qcow2_update_header(bs);
2197 if (ret < 0) {
2198 s->qcow_version = current_version;
2199 return ret;
2200 }
2201 return 0;
2202 }
2203
2204 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts)
2205 {
2206 BDRVQcowState *s = bs->opaque;
2207 int old_version = s->qcow_version, new_version = old_version;
2208 uint64_t new_size = 0;
2209 const char *backing_file = NULL, *backing_format = NULL;
2210 bool lazy_refcounts = s->use_lazy_refcounts;
2211 const char *compat = NULL;
2212 uint64_t cluster_size = s->cluster_size;
2213 bool encrypt;
2214 int ret;
2215 QemuOptDesc *desc = opts->list->desc;
2216
2217 while (desc && desc->name) {
2218 if (!qemu_opt_find(opts, desc->name)) {
2219 /* only change explicitly defined options */
2220 desc++;
2221 continue;
2222 }
2223
2224 if (!strcmp(desc->name, "compat")) {
2225 compat = qemu_opt_get(opts, "compat");
2226 if (!compat) {
2227 /* preserve default */
2228 } else if (!strcmp(compat, "0.10")) {
2229 new_version = 2;
2230 } else if (!strcmp(compat, "1.1")) {
2231 new_version = 3;
2232 } else {
2233 fprintf(stderr, "Unknown compatibility level %s.\n", compat);
2234 return -EINVAL;
2235 }
2236 } else if (!strcmp(desc->name, "preallocation")) {
2237 fprintf(stderr, "Cannot change preallocation mode.\n");
2238 return -ENOTSUP;
2239 } else if (!strcmp(desc->name, "size")) {
2240 new_size = qemu_opt_get_size(opts, "size", 0);
2241 } else if (!strcmp(desc->name, "backing_file")) {
2242 backing_file = qemu_opt_get(opts, "backing_file");
2243 } else if (!strcmp(desc->name, "backing_fmt")) {
2244 backing_format = qemu_opt_get(opts, "backing_fmt");
2245 } else if (!strcmp(desc->name, "encryption")) {
2246 encrypt = qemu_opt_get_bool(opts, "encryption", s->crypt_method);
2247 if (encrypt != !!s->crypt_method) {
2248 fprintf(stderr, "Changing the encryption flag is not "
2249 "supported.\n");
2250 return -ENOTSUP;
2251 }
2252 } else if (!strcmp(desc->name, "cluster_size")) {
2253 cluster_size = qemu_opt_get_size(opts, "cluster_size",
2254 cluster_size);
2255 if (cluster_size != s->cluster_size) {
2256 fprintf(stderr, "Changing the cluster size is not "
2257 "supported.\n");
2258 return -ENOTSUP;
2259 }
2260 } else if (!strcmp(desc->name, "lazy_refcounts")) {
2261 lazy_refcounts = qemu_opt_get_bool(opts, "lazy_refcounts",
2262 lazy_refcounts);
2263 } else {
2264 /* if this assertion fails, this probably means a new option was
2265 * added without having it covered here */
2266 assert(false);
2267 }
2268
2269 desc++;
2270 }
2271
2272 if (new_version != old_version) {
2273 if (new_version > old_version) {
2274 /* Upgrade */
2275 s->qcow_version = new_version;
2276 ret = qcow2_update_header(bs);
2277 if (ret < 0) {
2278 s->qcow_version = old_version;
2279 return ret;
2280 }
2281 } else {
2282 ret = qcow2_downgrade(bs, new_version);
2283 if (ret < 0) {
2284 return ret;
2285 }
2286 }
2287 }
2288
2289 if (backing_file || backing_format) {
2290 ret = qcow2_change_backing_file(bs, backing_file ?: bs->backing_file,
2291 backing_format ?: bs->backing_format);
2292 if (ret < 0) {
2293 return ret;
2294 }
2295 }
2296
2297 if (s->use_lazy_refcounts != lazy_refcounts) {
2298 if (lazy_refcounts) {
2299 if (s->qcow_version < 3) {
2300 fprintf(stderr, "Lazy refcounts only supported with compatibility "
2301 "level 1.1 and above (use compat=1.1 or greater)\n");
2302 return -EINVAL;
2303 }
2304 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
2305 ret = qcow2_update_header(bs);
2306 if (ret < 0) {
2307 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
2308 return ret;
2309 }
2310 s->use_lazy_refcounts = true;
2311 } else {
2312 /* make image clean first */
2313 ret = qcow2_mark_clean(bs);
2314 if (ret < 0) {
2315 return ret;
2316 }
2317 /* now disallow lazy refcounts */
2318 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
2319 ret = qcow2_update_header(bs);
2320 if (ret < 0) {
2321 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
2322 return ret;
2323 }
2324 s->use_lazy_refcounts = false;
2325 }
2326 }
2327
2328 if (new_size) {
2329 ret = bdrv_truncate(bs, new_size);
2330 if (ret < 0) {
2331 return ret;
2332 }
2333 }
2334
2335 return 0;
2336 }
2337
2338 static QemuOptsList qcow2_create_opts = {
2339 .name = "qcow2-create-opts",
2340 .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
2341 .desc = {
2342 {
2343 .name = BLOCK_OPT_SIZE,
2344 .type = QEMU_OPT_SIZE,
2345 .help = "Virtual disk size"
2346 },
2347 {
2348 .name = BLOCK_OPT_COMPAT_LEVEL,
2349 .type = QEMU_OPT_STRING,
2350 .help = "Compatibility level (0.10 or 1.1)"
2351 },
2352 {
2353 .name = BLOCK_OPT_BACKING_FILE,
2354 .type = QEMU_OPT_STRING,
2355 .help = "File name of a base image"
2356 },
2357 {
2358 .name = BLOCK_OPT_BACKING_FMT,
2359 .type = QEMU_OPT_STRING,
2360 .help = "Image format of the base image"
2361 },
2362 {
2363 .name = BLOCK_OPT_ENCRYPT,
2364 .type = QEMU_OPT_BOOL,
2365 .help = "Encrypt the image",
2366 .def_value_str = "off"
2367 },
2368 {
2369 .name = BLOCK_OPT_CLUSTER_SIZE,
2370 .type = QEMU_OPT_SIZE,
2371 .help = "qcow2 cluster size",
2372 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
2373 },
2374 {
2375 .name = BLOCK_OPT_PREALLOC,
2376 .type = QEMU_OPT_STRING,
2377 .help = "Preallocation mode (allowed values: off, metadata)"
2378 },
2379 {
2380 .name = BLOCK_OPT_LAZY_REFCOUNTS,
2381 .type = QEMU_OPT_BOOL,
2382 .help = "Postpone refcount updates",
2383 .def_value_str = "off"
2384 },
2385 { /* end of list */ }
2386 }
2387 };
2388
2389 static BlockDriver bdrv_qcow2 = {
2390 .format_name = "qcow2",
2391 .instance_size = sizeof(BDRVQcowState),
2392 .bdrv_probe = qcow2_probe,
2393 .bdrv_open = qcow2_open,
2394 .bdrv_close = qcow2_close,
2395 .bdrv_reopen_prepare = qcow2_reopen_prepare,
2396 .bdrv_create = qcow2_create,
2397 .bdrv_has_zero_init = bdrv_has_zero_init_1,
2398 .bdrv_co_get_block_status = qcow2_co_get_block_status,
2399 .bdrv_set_key = qcow2_set_key,
2400
2401 .bdrv_co_readv = qcow2_co_readv,
2402 .bdrv_co_writev = qcow2_co_writev,
2403 .bdrv_co_flush_to_os = qcow2_co_flush_to_os,
2404
2405 .bdrv_co_write_zeroes = qcow2_co_write_zeroes,
2406 .bdrv_co_discard = qcow2_co_discard,
2407 .bdrv_truncate = qcow2_truncate,
2408 .bdrv_write_compressed = qcow2_write_compressed,
2409
2410 .bdrv_snapshot_create = qcow2_snapshot_create,
2411 .bdrv_snapshot_goto = qcow2_snapshot_goto,
2412 .bdrv_snapshot_delete = qcow2_snapshot_delete,
2413 .bdrv_snapshot_list = qcow2_snapshot_list,
2414 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
2415 .bdrv_get_info = qcow2_get_info,
2416 .bdrv_get_specific_info = qcow2_get_specific_info,
2417
2418 .bdrv_save_vmstate = qcow2_save_vmstate,
2419 .bdrv_load_vmstate = qcow2_load_vmstate,
2420
2421 .supports_backing = true,
2422 .bdrv_change_backing_file = qcow2_change_backing_file,
2423
2424 .bdrv_refresh_limits = qcow2_refresh_limits,
2425 .bdrv_invalidate_cache = qcow2_invalidate_cache,
2426
2427 .create_opts = &qcow2_create_opts,
2428 .bdrv_check = qcow2_check,
2429 .bdrv_amend_options = qcow2_amend_options,
2430 };
2431
2432 static void bdrv_qcow2_init(void)
2433 {
2434 bdrv_register(&bdrv_qcow2);
2435 }
2436
2437 block_init(bdrv_qcow2_init);