]> git.proxmox.com Git - mirror_qemu.git/blob - block/qcow2-snapshot.c
Merge tag 'win32-pull-request' of https://gitlab.com/marcandre.lureau/qemu into staging
[mirror_qemu.git] / block / qcow2-snapshot.c
1 /*
2 * Block driver for the QCOW version 2 format
3 *
4 * Copyright (c) 2004-2006 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
25 #include "qemu/osdep.h"
26 #include "sysemu/block-backend.h"
27 #include "qapi/error.h"
28 #include "qcow2.h"
29 #include "qemu/bswap.h"
30 #include "qemu/error-report.h"
31 #include "qemu/cutils.h"
32 #include "qemu/memalign.h"
33
34 static void qcow2_free_single_snapshot(BlockDriverState *bs, int i)
35 {
36 BDRVQcow2State *s = bs->opaque;
37
38 assert(i >= 0 && i < s->nb_snapshots);
39 g_free(s->snapshots[i].name);
40 g_free(s->snapshots[i].id_str);
41 g_free(s->snapshots[i].unknown_extra_data);
42 memset(&s->snapshots[i], 0, sizeof(s->snapshots[i]));
43 }
44
45 void qcow2_free_snapshots(BlockDriverState *bs)
46 {
47 BDRVQcow2State *s = bs->opaque;
48 int i;
49
50 for(i = 0; i < s->nb_snapshots; i++) {
51 qcow2_free_single_snapshot(bs, i);
52 }
53 g_free(s->snapshots);
54 s->snapshots = NULL;
55 s->nb_snapshots = 0;
56 }
57
58 /*
59 * If @repair is true, try to repair a broken snapshot table instead
60 * of just returning an error:
61 *
62 * - If the snapshot table was too long, set *nb_clusters_reduced to
63 * the number of snapshots removed off the end.
64 * The caller will update the on-disk nb_snapshots accordingly;
65 * this leaks clusters, but is safe.
66 * (The on-disk information must be updated before
67 * qcow2_check_refcounts(), because that function relies on
68 * s->nb_snapshots to reflect the on-disk value.)
69 *
70 * - If there were snapshots with too much extra metadata, increment
71 * *extra_data_dropped for each.
72 * This requires the caller to eventually rewrite the whole snapshot
73 * table, which requires cluster allocation. Therefore, this should
74 * be done only after qcow2_check_refcounts() made sure the refcount
75 * structures are valid.
76 * (In the meantime, the image is still valid because
77 * qcow2_check_refcounts() does not do anything with snapshots'
78 * extra data.)
79 */
80 static int qcow2_do_read_snapshots(BlockDriverState *bs, bool repair,
81 int *nb_clusters_reduced,
82 int *extra_data_dropped,
83 Error **errp)
84 {
85 BDRVQcow2State *s = bs->opaque;
86 QCowSnapshotHeader h;
87 QCowSnapshotExtraData extra;
88 QCowSnapshot *sn;
89 int i, id_str_size, name_size;
90 int64_t offset, pre_sn_offset;
91 uint64_t table_length = 0;
92 int ret;
93
94 if (!s->nb_snapshots) {
95 s->snapshots = NULL;
96 s->snapshots_size = 0;
97 return 0;
98 }
99
100 offset = s->snapshots_offset;
101 s->snapshots = g_new0(QCowSnapshot, s->nb_snapshots);
102
103 for(i = 0; i < s->nb_snapshots; i++) {
104 bool truncate_unknown_extra_data = false;
105
106 pre_sn_offset = offset;
107 table_length = ROUND_UP(table_length, 8);
108
109 /* Read statically sized part of the snapshot header */
110 offset = ROUND_UP(offset, 8);
111 ret = bdrv_pread(bs->file, offset, sizeof(h), &h, 0);
112 if (ret < 0) {
113 error_setg_errno(errp, -ret, "Failed to read snapshot table");
114 goto fail;
115 }
116
117 offset += sizeof(h);
118 sn = s->snapshots + i;
119 sn->l1_table_offset = be64_to_cpu(h.l1_table_offset);
120 sn->l1_size = be32_to_cpu(h.l1_size);
121 sn->vm_state_size = be32_to_cpu(h.vm_state_size);
122 sn->date_sec = be32_to_cpu(h.date_sec);
123 sn->date_nsec = be32_to_cpu(h.date_nsec);
124 sn->vm_clock_nsec = be64_to_cpu(h.vm_clock_nsec);
125 sn->extra_data_size = be32_to_cpu(h.extra_data_size);
126
127 id_str_size = be16_to_cpu(h.id_str_size);
128 name_size = be16_to_cpu(h.name_size);
129
130 if (sn->extra_data_size > QCOW_MAX_SNAPSHOT_EXTRA_DATA) {
131 if (!repair) {
132 ret = -EFBIG;
133 error_setg(errp, "Too much extra metadata in snapshot table "
134 "entry %i", i);
135 error_append_hint(errp, "You can force-remove this extra "
136 "metadata with qemu-img check -r all\n");
137 goto fail;
138 }
139
140 fprintf(stderr, "Discarding too much extra metadata in snapshot "
141 "table entry %i (%" PRIu32 " > %u)\n",
142 i, sn->extra_data_size, QCOW_MAX_SNAPSHOT_EXTRA_DATA);
143
144 (*extra_data_dropped)++;
145 truncate_unknown_extra_data = true;
146 }
147
148 /* Read known extra data */
149 ret = bdrv_pread(bs->file, offset,
150 MIN(sizeof(extra), sn->extra_data_size), &extra, 0);
151 if (ret < 0) {
152 error_setg_errno(errp, -ret, "Failed to read snapshot table");
153 goto fail;
154 }
155 offset += MIN(sizeof(extra), sn->extra_data_size);
156
157 if (sn->extra_data_size >= endof(QCowSnapshotExtraData,
158 vm_state_size_large)) {
159 sn->vm_state_size = be64_to_cpu(extra.vm_state_size_large);
160 }
161
162 if (sn->extra_data_size >= endof(QCowSnapshotExtraData, disk_size)) {
163 sn->disk_size = be64_to_cpu(extra.disk_size);
164 } else {
165 sn->disk_size = bs->total_sectors * BDRV_SECTOR_SIZE;
166 }
167
168 if (sn->extra_data_size >= endof(QCowSnapshotExtraData, icount)) {
169 sn->icount = be64_to_cpu(extra.icount);
170 } else {
171 sn->icount = -1ULL;
172 }
173
174 if (sn->extra_data_size > sizeof(extra)) {
175 uint64_t extra_data_end;
176 size_t unknown_extra_data_size;
177
178 extra_data_end = offset + sn->extra_data_size - sizeof(extra);
179
180 if (truncate_unknown_extra_data) {
181 sn->extra_data_size = QCOW_MAX_SNAPSHOT_EXTRA_DATA;
182 }
183
184 /* Store unknown extra data */
185 unknown_extra_data_size = sn->extra_data_size - sizeof(extra);
186 sn->unknown_extra_data = g_malloc(unknown_extra_data_size);
187 ret = bdrv_pread(bs->file, offset, unknown_extra_data_size,
188 sn->unknown_extra_data, 0);
189 if (ret < 0) {
190 error_setg_errno(errp, -ret,
191 "Failed to read snapshot table");
192 goto fail;
193 }
194 offset = extra_data_end;
195 }
196
197 /* Read snapshot ID */
198 sn->id_str = g_malloc(id_str_size + 1);
199 ret = bdrv_pread(bs->file, offset, id_str_size, sn->id_str, 0);
200 if (ret < 0) {
201 error_setg_errno(errp, -ret, "Failed to read snapshot table");
202 goto fail;
203 }
204 offset += id_str_size;
205 sn->id_str[id_str_size] = '\0';
206
207 /* Read snapshot name */
208 sn->name = g_malloc(name_size + 1);
209 ret = bdrv_pread(bs->file, offset, name_size, sn->name, 0);
210 if (ret < 0) {
211 error_setg_errno(errp, -ret, "Failed to read snapshot table");
212 goto fail;
213 }
214 offset += name_size;
215 sn->name[name_size] = '\0';
216
217 /* Note that the extra data may have been truncated */
218 table_length += sizeof(h) + sn->extra_data_size + id_str_size +
219 name_size;
220 if (!repair) {
221 assert(table_length == offset - s->snapshots_offset);
222 }
223
224 if (table_length > QCOW_MAX_SNAPSHOTS_SIZE ||
225 offset - s->snapshots_offset > INT_MAX)
226 {
227 if (!repair) {
228 ret = -EFBIG;
229 error_setg(errp, "Snapshot table is too big");
230 error_append_hint(errp, "You can force-remove all %u "
231 "overhanging snapshots with qemu-img check "
232 "-r all\n", s->nb_snapshots - i);
233 goto fail;
234 }
235
236 fprintf(stderr, "Discarding %u overhanging snapshots (snapshot "
237 "table is too big)\n", s->nb_snapshots - i);
238
239 *nb_clusters_reduced += (s->nb_snapshots - i);
240
241 /* Discard current snapshot also */
242 qcow2_free_single_snapshot(bs, i);
243
244 /*
245 * This leaks all the rest of the snapshot table and the
246 * snapshots' clusters, but we run in check -r all mode,
247 * so qcow2_check_refcounts() will take care of it.
248 */
249 s->nb_snapshots = i;
250 offset = pre_sn_offset;
251 break;
252 }
253 }
254
255 assert(offset - s->snapshots_offset <= INT_MAX);
256 s->snapshots_size = offset - s->snapshots_offset;
257 return 0;
258
259 fail:
260 qcow2_free_snapshots(bs);
261 return ret;
262 }
263
264 int qcow2_read_snapshots(BlockDriverState *bs, Error **errp)
265 {
266 return qcow2_do_read_snapshots(bs, false, NULL, NULL, errp);
267 }
268
269 /* add at the end of the file a new list of snapshots */
270 int qcow2_write_snapshots(BlockDriverState *bs)
271 {
272 BDRVQcow2State *s = bs->opaque;
273 QCowSnapshot *sn;
274 QCowSnapshotHeader h;
275 QCowSnapshotExtraData extra;
276 int i, name_size, id_str_size, snapshots_size;
277 struct {
278 uint32_t nb_snapshots;
279 uint64_t snapshots_offset;
280 } QEMU_PACKED header_data;
281 int64_t offset, snapshots_offset = 0;
282 int ret;
283
284 /* compute the size of the snapshots */
285 offset = 0;
286 for(i = 0; i < s->nb_snapshots; i++) {
287 sn = s->snapshots + i;
288 offset = ROUND_UP(offset, 8);
289 offset += sizeof(h);
290 offset += MAX(sizeof(extra), sn->extra_data_size);
291 offset += strlen(sn->id_str);
292 offset += strlen(sn->name);
293
294 if (offset > QCOW_MAX_SNAPSHOTS_SIZE) {
295 ret = -EFBIG;
296 goto fail;
297 }
298 }
299
300 assert(offset <= INT_MAX);
301 snapshots_size = offset;
302
303 /* Allocate space for the new snapshot list */
304 snapshots_offset = qcow2_alloc_clusters(bs, snapshots_size);
305 offset = snapshots_offset;
306 if (offset < 0) {
307 ret = offset;
308 goto fail;
309 }
310 ret = bdrv_flush(bs);
311 if (ret < 0) {
312 goto fail;
313 }
314
315 /* The snapshot list position has not yet been updated, so these clusters
316 * must indeed be completely free */
317 ret = qcow2_pre_write_overlap_check(bs, 0, offset, snapshots_size, false);
318 if (ret < 0) {
319 goto fail;
320 }
321
322
323 /* Write all snapshots to the new list */
324 for(i = 0; i < s->nb_snapshots; i++) {
325 sn = s->snapshots + i;
326 memset(&h, 0, sizeof(h));
327 h.l1_table_offset = cpu_to_be64(sn->l1_table_offset);
328 h.l1_size = cpu_to_be32(sn->l1_size);
329 /* If it doesn't fit in 32 bit, older implementations should treat it
330 * as a disk-only snapshot rather than truncate the VM state */
331 if (sn->vm_state_size <= 0xffffffff) {
332 h.vm_state_size = cpu_to_be32(sn->vm_state_size);
333 }
334 h.date_sec = cpu_to_be32(sn->date_sec);
335 h.date_nsec = cpu_to_be32(sn->date_nsec);
336 h.vm_clock_nsec = cpu_to_be64(sn->vm_clock_nsec);
337 h.extra_data_size = cpu_to_be32(MAX(sizeof(extra),
338 sn->extra_data_size));
339
340 memset(&extra, 0, sizeof(extra));
341 extra.vm_state_size_large = cpu_to_be64(sn->vm_state_size);
342 extra.disk_size = cpu_to_be64(sn->disk_size);
343 extra.icount = cpu_to_be64(sn->icount);
344
345 id_str_size = strlen(sn->id_str);
346 name_size = strlen(sn->name);
347 assert(id_str_size <= UINT16_MAX && name_size <= UINT16_MAX);
348 h.id_str_size = cpu_to_be16(id_str_size);
349 h.name_size = cpu_to_be16(name_size);
350 offset = ROUND_UP(offset, 8);
351
352 ret = bdrv_pwrite(bs->file, offset, sizeof(h), &h, 0);
353 if (ret < 0) {
354 goto fail;
355 }
356 offset += sizeof(h);
357
358 ret = bdrv_pwrite(bs->file, offset, sizeof(extra), &extra, 0);
359 if (ret < 0) {
360 goto fail;
361 }
362 offset += sizeof(extra);
363
364 if (sn->extra_data_size > sizeof(extra)) {
365 size_t unknown_extra_data_size =
366 sn->extra_data_size - sizeof(extra);
367
368 /* qcow2_read_snapshots() ensures no unbounded allocation */
369 assert(unknown_extra_data_size <= BDRV_REQUEST_MAX_BYTES);
370 assert(sn->unknown_extra_data);
371
372 ret = bdrv_pwrite(bs->file, offset, unknown_extra_data_size,
373 sn->unknown_extra_data, 0);
374 if (ret < 0) {
375 goto fail;
376 }
377 offset += unknown_extra_data_size;
378 }
379
380 ret = bdrv_pwrite(bs->file, offset, id_str_size, sn->id_str, 0);
381 if (ret < 0) {
382 goto fail;
383 }
384 offset += id_str_size;
385
386 ret = bdrv_pwrite(bs->file, offset, name_size, sn->name, 0);
387 if (ret < 0) {
388 goto fail;
389 }
390 offset += name_size;
391 }
392
393 /*
394 * Update the header to point to the new snapshot table. This requires the
395 * new table and its refcounts to be stable on disk.
396 */
397 ret = bdrv_flush(bs);
398 if (ret < 0) {
399 goto fail;
400 }
401
402 QEMU_BUILD_BUG_ON(offsetof(QCowHeader, snapshots_offset) !=
403 endof(QCowHeader, nb_snapshots));
404
405 header_data.nb_snapshots = cpu_to_be32(s->nb_snapshots);
406 header_data.snapshots_offset = cpu_to_be64(snapshots_offset);
407
408 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, nb_snapshots),
409 sizeof(header_data), &header_data, 0);
410 if (ret < 0) {
411 goto fail;
412 }
413
414 /* free the old snapshot table */
415 qcow2_free_clusters(bs, s->snapshots_offset, s->snapshots_size,
416 QCOW2_DISCARD_SNAPSHOT);
417 s->snapshots_offset = snapshots_offset;
418 s->snapshots_size = snapshots_size;
419 return 0;
420
421 fail:
422 if (snapshots_offset > 0) {
423 qcow2_free_clusters(bs, snapshots_offset, snapshots_size,
424 QCOW2_DISCARD_ALWAYS);
425 }
426 return ret;
427 }
428
429 int coroutine_fn qcow2_check_read_snapshot_table(BlockDriverState *bs,
430 BdrvCheckResult *result,
431 BdrvCheckMode fix)
432 {
433 BDRVQcow2State *s = bs->opaque;
434 Error *local_err = NULL;
435 int nb_clusters_reduced = 0;
436 int extra_data_dropped = 0;
437 int ret;
438 struct {
439 uint32_t nb_snapshots;
440 uint64_t snapshots_offset;
441 } QEMU_PACKED snapshot_table_pointer;
442
443 /* qcow2_do_open() discards this information in check mode */
444 ret = bdrv_pread(bs->file, offsetof(QCowHeader, nb_snapshots),
445 sizeof(snapshot_table_pointer), &snapshot_table_pointer,
446 0);
447 if (ret < 0) {
448 result->check_errors++;
449 fprintf(stderr, "ERROR failed to read the snapshot table pointer from "
450 "the image header: %s\n", strerror(-ret));
451 return ret;
452 }
453
454 s->snapshots_offset = be64_to_cpu(snapshot_table_pointer.snapshots_offset);
455 s->nb_snapshots = be32_to_cpu(snapshot_table_pointer.nb_snapshots);
456
457 if (s->nb_snapshots > QCOW_MAX_SNAPSHOTS && (fix & BDRV_FIX_ERRORS)) {
458 fprintf(stderr, "Discarding %u overhanging snapshots\n",
459 s->nb_snapshots - QCOW_MAX_SNAPSHOTS);
460
461 nb_clusters_reduced += s->nb_snapshots - QCOW_MAX_SNAPSHOTS;
462 s->nb_snapshots = QCOW_MAX_SNAPSHOTS;
463 }
464
465 ret = qcow2_validate_table(bs, s->snapshots_offset, s->nb_snapshots,
466 sizeof(QCowSnapshotHeader),
467 sizeof(QCowSnapshotHeader) * QCOW_MAX_SNAPSHOTS,
468 "snapshot table", &local_err);
469 if (ret < 0) {
470 result->check_errors++;
471 error_reportf_err(local_err, "ERROR ");
472
473 if (s->nb_snapshots > QCOW_MAX_SNAPSHOTS) {
474 fprintf(stderr, "You can force-remove all %u overhanging snapshots "
475 "with qemu-img check -r all\n",
476 s->nb_snapshots - QCOW_MAX_SNAPSHOTS);
477 }
478
479 /* We did not read the snapshot table, so invalidate this information */
480 s->snapshots_offset = 0;
481 s->nb_snapshots = 0;
482
483 return ret;
484 }
485
486 qemu_co_mutex_unlock(&s->lock);
487 ret = qcow2_do_read_snapshots(bs, fix & BDRV_FIX_ERRORS,
488 &nb_clusters_reduced, &extra_data_dropped,
489 &local_err);
490 qemu_co_mutex_lock(&s->lock);
491 if (ret < 0) {
492 result->check_errors++;
493 error_reportf_err(local_err,
494 "ERROR failed to read the snapshot table: ");
495
496 /* We did not read the snapshot table, so invalidate this information */
497 s->snapshots_offset = 0;
498 s->nb_snapshots = 0;
499
500 return ret;
501 }
502 result->corruptions += nb_clusters_reduced + extra_data_dropped;
503
504 if (nb_clusters_reduced) {
505 /*
506 * Update image header now, because:
507 * (1) qcow2_check_refcounts() relies on s->nb_snapshots to be
508 * the same as what the image header says,
509 * (2) this leaks clusters, but qcow2_check_refcounts() will
510 * fix that.
511 */
512 assert(fix & BDRV_FIX_ERRORS);
513
514 snapshot_table_pointer.nb_snapshots = cpu_to_be32(s->nb_snapshots);
515 ret = bdrv_co_pwrite_sync(bs->file, offsetof(QCowHeader, nb_snapshots),
516 sizeof(snapshot_table_pointer.nb_snapshots),
517 &snapshot_table_pointer.nb_snapshots, 0);
518 if (ret < 0) {
519 result->check_errors++;
520 fprintf(stderr, "ERROR failed to update the snapshot count in the "
521 "image header: %s\n", strerror(-ret));
522 return ret;
523 }
524
525 result->corruptions_fixed += nb_clusters_reduced;
526 result->corruptions -= nb_clusters_reduced;
527 }
528
529 /*
530 * All of v3 images' snapshot table entries need to have at least
531 * 16 bytes of extra data.
532 */
533 if (s->qcow_version >= 3) {
534 int i;
535 for (i = 0; i < s->nb_snapshots; i++) {
536 if (s->snapshots[i].extra_data_size <
537 sizeof_field(QCowSnapshotExtraData, vm_state_size_large) +
538 sizeof_field(QCowSnapshotExtraData, disk_size))
539 {
540 result->corruptions++;
541 fprintf(stderr, "%s snapshot table entry %i is incomplete\n",
542 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
543 }
544 }
545 }
546
547 return 0;
548 }
549
550 int coroutine_fn qcow2_check_fix_snapshot_table(BlockDriverState *bs,
551 BdrvCheckResult *result,
552 BdrvCheckMode fix)
553 {
554 BDRVQcow2State *s = bs->opaque;
555 int ret;
556
557 if (result->corruptions && (fix & BDRV_FIX_ERRORS)) {
558 qemu_co_mutex_unlock(&s->lock);
559 ret = qcow2_write_snapshots(bs);
560 qemu_co_mutex_lock(&s->lock);
561 if (ret < 0) {
562 result->check_errors++;
563 fprintf(stderr, "ERROR failed to update snapshot table: %s\n",
564 strerror(-ret));
565 return ret;
566 }
567
568 result->corruptions_fixed += result->corruptions;
569 result->corruptions = 0;
570 }
571
572 return 0;
573 }
574
575 static void find_new_snapshot_id(BlockDriverState *bs,
576 char *id_str, int id_str_size)
577 {
578 BDRVQcow2State *s = bs->opaque;
579 QCowSnapshot *sn;
580 int i;
581 unsigned long id, id_max = 0;
582
583 for(i = 0; i < s->nb_snapshots; i++) {
584 sn = s->snapshots + i;
585 id = strtoul(sn->id_str, NULL, 10);
586 if (id > id_max)
587 id_max = id;
588 }
589 snprintf(id_str, id_str_size, "%lu", id_max + 1);
590 }
591
592 static int find_snapshot_by_id_and_name(BlockDriverState *bs,
593 const char *id,
594 const char *name)
595 {
596 BDRVQcow2State *s = bs->opaque;
597 int i;
598
599 if (id && name) {
600 for (i = 0; i < s->nb_snapshots; i++) {
601 if (!strcmp(s->snapshots[i].id_str, id) &&
602 !strcmp(s->snapshots[i].name, name)) {
603 return i;
604 }
605 }
606 } else if (id) {
607 for (i = 0; i < s->nb_snapshots; i++) {
608 if (!strcmp(s->snapshots[i].id_str, id)) {
609 return i;
610 }
611 }
612 } else if (name) {
613 for (i = 0; i < s->nb_snapshots; i++) {
614 if (!strcmp(s->snapshots[i].name, name)) {
615 return i;
616 }
617 }
618 }
619
620 return -1;
621 }
622
623 static int find_snapshot_by_id_or_name(BlockDriverState *bs,
624 const char *id_or_name)
625 {
626 int ret;
627
628 ret = find_snapshot_by_id_and_name(bs, id_or_name, NULL);
629 if (ret >= 0) {
630 return ret;
631 }
632 return find_snapshot_by_id_and_name(bs, NULL, id_or_name);
633 }
634
635 /* if no id is provided, a new one is constructed */
636 int qcow2_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info)
637 {
638 BDRVQcow2State *s = bs->opaque;
639 QCowSnapshot *new_snapshot_list = NULL;
640 QCowSnapshot *old_snapshot_list = NULL;
641 QCowSnapshot sn1, *sn = &sn1;
642 int i, ret;
643 uint64_t *l1_table = NULL;
644 int64_t l1_table_offset;
645
646 if (s->nb_snapshots >= QCOW_MAX_SNAPSHOTS) {
647 return -EFBIG;
648 }
649
650 if (has_data_file(bs)) {
651 return -ENOTSUP;
652 }
653
654 memset(sn, 0, sizeof(*sn));
655
656 /* Generate an ID */
657 find_new_snapshot_id(bs, sn_info->id_str, sizeof(sn_info->id_str));
658
659 /* Populate sn with passed data */
660 sn->id_str = g_strdup(sn_info->id_str);
661 sn->name = g_strdup(sn_info->name);
662
663 sn->disk_size = bs->total_sectors * BDRV_SECTOR_SIZE;
664 sn->vm_state_size = sn_info->vm_state_size;
665 sn->date_sec = sn_info->date_sec;
666 sn->date_nsec = sn_info->date_nsec;
667 sn->vm_clock_nsec = sn_info->vm_clock_nsec;
668 sn->icount = sn_info->icount;
669 sn->extra_data_size = sizeof(QCowSnapshotExtraData);
670
671 /* Allocate the L1 table of the snapshot and copy the current one there. */
672 l1_table_offset = qcow2_alloc_clusters(bs, s->l1_size * L1E_SIZE);
673 if (l1_table_offset < 0) {
674 ret = l1_table_offset;
675 goto fail;
676 }
677
678 sn->l1_table_offset = l1_table_offset;
679 sn->l1_size = s->l1_size;
680
681 l1_table = g_try_new(uint64_t, s->l1_size);
682 if (s->l1_size && l1_table == NULL) {
683 ret = -ENOMEM;
684 goto fail;
685 }
686
687 for(i = 0; i < s->l1_size; i++) {
688 l1_table[i] = cpu_to_be64(s->l1_table[i]);
689 }
690
691 ret = qcow2_pre_write_overlap_check(bs, 0, sn->l1_table_offset,
692 s->l1_size * L1E_SIZE, false);
693 if (ret < 0) {
694 goto fail;
695 }
696
697 ret = bdrv_pwrite(bs->file, sn->l1_table_offset, s->l1_size * L1E_SIZE,
698 l1_table, 0);
699 if (ret < 0) {
700 goto fail;
701 }
702
703 g_free(l1_table);
704 l1_table = NULL;
705
706 /*
707 * Increase the refcounts of all clusters and make sure everything is
708 * stable on disk before updating the snapshot table to contain a pointer
709 * to the new L1 table.
710 */
711 ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1);
712 if (ret < 0) {
713 goto fail;
714 }
715
716 /* Append the new snapshot to the snapshot list */
717 new_snapshot_list = g_new(QCowSnapshot, s->nb_snapshots + 1);
718 if (s->snapshots) {
719 memcpy(new_snapshot_list, s->snapshots,
720 s->nb_snapshots * sizeof(QCowSnapshot));
721 old_snapshot_list = s->snapshots;
722 }
723 s->snapshots = new_snapshot_list;
724 s->snapshots[s->nb_snapshots++] = *sn;
725
726 ret = qcow2_write_snapshots(bs);
727 if (ret < 0) {
728 g_free(s->snapshots);
729 s->snapshots = old_snapshot_list;
730 s->nb_snapshots--;
731 goto fail;
732 }
733
734 g_free(old_snapshot_list);
735
736 /* The VM state isn't needed any more in the active L1 table; in fact, it
737 * hurts by causing expensive COW for the next snapshot. */
738 qcow2_cluster_discard(bs, qcow2_vm_state_offset(s),
739 ROUND_UP(sn->vm_state_size, s->cluster_size),
740 QCOW2_DISCARD_NEVER, false);
741
742 #ifdef DEBUG_ALLOC
743 {
744 BdrvCheckResult result = {0};
745 qcow2_check_refcounts(bs, &result, 0);
746 }
747 #endif
748 return 0;
749
750 fail:
751 g_free(sn->id_str);
752 g_free(sn->name);
753 g_free(l1_table);
754
755 return ret;
756 }
757
758 /* copy the snapshot 'snapshot_name' into the current disk image */
759 int qcow2_snapshot_goto(BlockDriverState *bs, const char *snapshot_id)
760 {
761 BDRVQcow2State *s = bs->opaque;
762 QCowSnapshot *sn;
763 Error *local_err = NULL;
764 int i, snapshot_index;
765 int cur_l1_bytes, sn_l1_bytes;
766 int ret;
767 uint64_t *sn_l1_table = NULL;
768
769 if (has_data_file(bs)) {
770 return -ENOTSUP;
771 }
772
773 /* Search the snapshot */
774 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
775 if (snapshot_index < 0) {
776 return -ENOENT;
777 }
778 sn = &s->snapshots[snapshot_index];
779
780 ret = qcow2_validate_table(bs, sn->l1_table_offset, sn->l1_size,
781 L1E_SIZE, QCOW_MAX_L1_SIZE,
782 "Snapshot L1 table", &local_err);
783 if (ret < 0) {
784 error_report_err(local_err);
785 goto fail;
786 }
787
788 if (sn->disk_size != bs->total_sectors * BDRV_SECTOR_SIZE) {
789 BlockBackend *blk = blk_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL,
790 &local_err);
791 if (!blk) {
792 error_report_err(local_err);
793 ret = -ENOTSUP;
794 goto fail;
795 }
796
797 ret = blk_truncate(blk, sn->disk_size, true, PREALLOC_MODE_OFF, 0,
798 &local_err);
799 blk_unref(blk);
800 if (ret < 0) {
801 error_report_err(local_err);
802 goto fail;
803 }
804 }
805
806 /*
807 * Make sure that the current L1 table is big enough to contain the whole
808 * L1 table of the snapshot. If the snapshot L1 table is smaller, the
809 * current one must be padded with zeros.
810 */
811 ret = qcow2_grow_l1_table(bs, sn->l1_size, true);
812 if (ret < 0) {
813 goto fail;
814 }
815
816 cur_l1_bytes = s->l1_size * L1E_SIZE;
817 sn_l1_bytes = sn->l1_size * L1E_SIZE;
818
819 /*
820 * Copy the snapshot L1 table to the current L1 table.
821 *
822 * Before overwriting the old current L1 table on disk, make sure to
823 * increase all refcounts for the clusters referenced by the new one.
824 * Decrease the refcount referenced by the old one only when the L1
825 * table is overwritten.
826 */
827 sn_l1_table = g_try_malloc0(cur_l1_bytes);
828 if (cur_l1_bytes && sn_l1_table == NULL) {
829 ret = -ENOMEM;
830 goto fail;
831 }
832
833 ret = bdrv_pread(bs->file, sn->l1_table_offset, sn_l1_bytes, sn_l1_table,
834 0);
835 if (ret < 0) {
836 goto fail;
837 }
838
839 ret = qcow2_update_snapshot_refcount(bs, sn->l1_table_offset,
840 sn->l1_size, 1);
841 if (ret < 0) {
842 goto fail;
843 }
844
845 ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L1,
846 s->l1_table_offset, cur_l1_bytes,
847 false);
848 if (ret < 0) {
849 goto fail;
850 }
851
852 ret = bdrv_pwrite_sync(bs->file, s->l1_table_offset, cur_l1_bytes,
853 sn_l1_table, 0);
854 if (ret < 0) {
855 goto fail;
856 }
857
858 /*
859 * Decrease refcount of clusters of current L1 table.
860 *
861 * At this point, the in-memory s->l1_table points to the old L1 table,
862 * whereas on disk we already have the new one.
863 *
864 * qcow2_update_snapshot_refcount special cases the current L1 table to use
865 * the in-memory data instead of really using the offset to load a new one,
866 * which is why this works.
867 */
868 ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset,
869 s->l1_size, -1);
870
871 /*
872 * Now update the in-memory L1 table to be in sync with the on-disk one. We
873 * need to do this even if updating refcounts failed.
874 */
875 for(i = 0;i < s->l1_size; i++) {
876 s->l1_table[i] = be64_to_cpu(sn_l1_table[i]);
877 }
878
879 if (ret < 0) {
880 goto fail;
881 }
882
883 g_free(sn_l1_table);
884 sn_l1_table = NULL;
885
886 /*
887 * Update QCOW_OFLAG_COPIED in the active L1 table (it may have changed
888 * when we decreased the refcount of the old snapshot.
889 */
890 ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
891 if (ret < 0) {
892 goto fail;
893 }
894
895 #ifdef DEBUG_ALLOC
896 {
897 BdrvCheckResult result = {0};
898 qcow2_check_refcounts(bs, &result, 0);
899 }
900 #endif
901 return 0;
902
903 fail:
904 g_free(sn_l1_table);
905 return ret;
906 }
907
908 int qcow2_snapshot_delete(BlockDriverState *bs,
909 const char *snapshot_id,
910 const char *name,
911 Error **errp)
912 {
913 BDRVQcow2State *s = bs->opaque;
914 QCowSnapshot sn;
915 int snapshot_index, ret;
916
917 if (has_data_file(bs)) {
918 return -ENOTSUP;
919 }
920
921 /* Search the snapshot */
922 snapshot_index = find_snapshot_by_id_and_name(bs, snapshot_id, name);
923 if (snapshot_index < 0) {
924 error_setg(errp, "Can't find the snapshot");
925 return -ENOENT;
926 }
927 sn = s->snapshots[snapshot_index];
928
929 ret = qcow2_validate_table(bs, sn.l1_table_offset, sn.l1_size,
930 L1E_SIZE, QCOW_MAX_L1_SIZE,
931 "Snapshot L1 table", errp);
932 if (ret < 0) {
933 return ret;
934 }
935
936 /* Remove it from the snapshot list */
937 memmove(s->snapshots + snapshot_index,
938 s->snapshots + snapshot_index + 1,
939 (s->nb_snapshots - snapshot_index - 1) * sizeof(sn));
940 s->nb_snapshots--;
941 ret = qcow2_write_snapshots(bs);
942 if (ret < 0) {
943 error_setg_errno(errp, -ret,
944 "Failed to remove snapshot from snapshot list");
945 return ret;
946 }
947
948 /*
949 * The snapshot is now unused, clean up. If we fail after this point, we
950 * won't recover but just leak clusters.
951 */
952 g_free(sn.unknown_extra_data);
953 g_free(sn.id_str);
954 g_free(sn.name);
955
956 /*
957 * Now decrease the refcounts of clusters referenced by the snapshot and
958 * free the L1 table.
959 */
960 ret = qcow2_update_snapshot_refcount(bs, sn.l1_table_offset,
961 sn.l1_size, -1);
962 if (ret < 0) {
963 error_setg_errno(errp, -ret, "Failed to free the cluster and L1 table");
964 return ret;
965 }
966 qcow2_free_clusters(bs, sn.l1_table_offset, sn.l1_size * L1E_SIZE,
967 QCOW2_DISCARD_SNAPSHOT);
968
969 /* must update the copied flag on the current cluster offsets */
970 ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
971 if (ret < 0) {
972 error_setg_errno(errp, -ret,
973 "Failed to update snapshot status in disk");
974 return ret;
975 }
976
977 #ifdef DEBUG_ALLOC
978 {
979 BdrvCheckResult result = {0};
980 qcow2_check_refcounts(bs, &result, 0);
981 }
982 #endif
983 return 0;
984 }
985
986 int qcow2_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab)
987 {
988 BDRVQcow2State *s = bs->opaque;
989 QEMUSnapshotInfo *sn_tab, *sn_info;
990 QCowSnapshot *sn;
991 int i;
992
993 if (has_data_file(bs)) {
994 return -ENOTSUP;
995 }
996 if (!s->nb_snapshots) {
997 *psn_tab = NULL;
998 return s->nb_snapshots;
999 }
1000
1001 sn_tab = g_new0(QEMUSnapshotInfo, s->nb_snapshots);
1002 for(i = 0; i < s->nb_snapshots; i++) {
1003 sn_info = sn_tab + i;
1004 sn = s->snapshots + i;
1005 pstrcpy(sn_info->id_str, sizeof(sn_info->id_str),
1006 sn->id_str);
1007 pstrcpy(sn_info->name, sizeof(sn_info->name),
1008 sn->name);
1009 sn_info->vm_state_size = sn->vm_state_size;
1010 sn_info->date_sec = sn->date_sec;
1011 sn_info->date_nsec = sn->date_nsec;
1012 sn_info->vm_clock_nsec = sn->vm_clock_nsec;
1013 sn_info->icount = sn->icount;
1014 }
1015 *psn_tab = sn_tab;
1016 return s->nb_snapshots;
1017 }
1018
1019 int qcow2_snapshot_load_tmp(BlockDriverState *bs,
1020 const char *snapshot_id,
1021 const char *name,
1022 Error **errp)
1023 {
1024 int i, snapshot_index;
1025 BDRVQcow2State *s = bs->opaque;
1026 QCowSnapshot *sn;
1027 uint64_t *new_l1_table;
1028 int new_l1_bytes;
1029 int ret;
1030
1031 assert(bdrv_is_read_only(bs));
1032
1033 /* Search the snapshot */
1034 snapshot_index = find_snapshot_by_id_and_name(bs, snapshot_id, name);
1035 if (snapshot_index < 0) {
1036 error_setg(errp,
1037 "Can't find snapshot");
1038 return -ENOENT;
1039 }
1040 sn = &s->snapshots[snapshot_index];
1041
1042 /* Allocate and read in the snapshot's L1 table */
1043 ret = qcow2_validate_table(bs, sn->l1_table_offset, sn->l1_size,
1044 L1E_SIZE, QCOW_MAX_L1_SIZE,
1045 "Snapshot L1 table", errp);
1046 if (ret < 0) {
1047 return ret;
1048 }
1049 new_l1_bytes = sn->l1_size * L1E_SIZE;
1050 new_l1_table = qemu_try_blockalign(bs->file->bs, new_l1_bytes);
1051 if (new_l1_table == NULL) {
1052 return -ENOMEM;
1053 }
1054
1055 ret = bdrv_pread(bs->file, sn->l1_table_offset, new_l1_bytes,
1056 new_l1_table, 0);
1057 if (ret < 0) {
1058 error_setg(errp, "Failed to read l1 table for snapshot");
1059 qemu_vfree(new_l1_table);
1060 return ret;
1061 }
1062
1063 /* Switch the L1 table */
1064 qemu_vfree(s->l1_table);
1065
1066 s->l1_size = sn->l1_size;
1067 s->l1_table_offset = sn->l1_table_offset;
1068 s->l1_table = new_l1_table;
1069
1070 for(i = 0;i < s->l1_size; i++) {
1071 be64_to_cpus(&s->l1_table[i]);
1072 }
1073
1074 return 0;
1075 }