]> git.proxmox.com Git - qemu.git/blob - block/qcow2-snapshot.c
Merge remote-tracking branch 'qemu-kvm/uq/master' into staging
[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-common.h"
26 #include "block/block_int.h"
27 #include "block/qcow2.h"
28
29 typedef struct QEMU_PACKED QCowSnapshotHeader {
30 /* header is 8 byte aligned */
31 uint64_t l1_table_offset;
32
33 uint32_t l1_size;
34 uint16_t id_str_size;
35 uint16_t name_size;
36
37 uint32_t date_sec;
38 uint32_t date_nsec;
39
40 uint64_t vm_clock_nsec;
41
42 uint32_t vm_state_size;
43 uint32_t extra_data_size; /* for extension */
44 /* extra data follows */
45 /* id_str follows */
46 /* name follows */
47 } QCowSnapshotHeader;
48
49 typedef struct QEMU_PACKED QCowSnapshotExtraData {
50 uint64_t vm_state_size_large;
51 uint64_t disk_size;
52 } QCowSnapshotExtraData;
53
54 void qcow2_free_snapshots(BlockDriverState *bs)
55 {
56 BDRVQcowState *s = bs->opaque;
57 int i;
58
59 for(i = 0; i < s->nb_snapshots; i++) {
60 g_free(s->snapshots[i].name);
61 g_free(s->snapshots[i].id_str);
62 }
63 g_free(s->snapshots);
64 s->snapshots = NULL;
65 s->nb_snapshots = 0;
66 }
67
68 int qcow2_read_snapshots(BlockDriverState *bs)
69 {
70 BDRVQcowState *s = bs->opaque;
71 QCowSnapshotHeader h;
72 QCowSnapshotExtraData extra;
73 QCowSnapshot *sn;
74 int i, id_str_size, name_size;
75 int64_t offset;
76 uint32_t extra_data_size;
77 int ret;
78
79 if (!s->nb_snapshots) {
80 s->snapshots = NULL;
81 s->snapshots_size = 0;
82 return 0;
83 }
84
85 offset = s->snapshots_offset;
86 s->snapshots = g_malloc0(s->nb_snapshots * sizeof(QCowSnapshot));
87
88 for(i = 0; i < s->nb_snapshots; i++) {
89 /* Read statically sized part of the snapshot header */
90 offset = align_offset(offset, 8);
91 ret = bdrv_pread(bs->file, offset, &h, sizeof(h));
92 if (ret < 0) {
93 goto fail;
94 }
95
96 offset += sizeof(h);
97 sn = s->snapshots + i;
98 sn->l1_table_offset = be64_to_cpu(h.l1_table_offset);
99 sn->l1_size = be32_to_cpu(h.l1_size);
100 sn->vm_state_size = be32_to_cpu(h.vm_state_size);
101 sn->date_sec = be32_to_cpu(h.date_sec);
102 sn->date_nsec = be32_to_cpu(h.date_nsec);
103 sn->vm_clock_nsec = be64_to_cpu(h.vm_clock_nsec);
104 extra_data_size = be32_to_cpu(h.extra_data_size);
105
106 id_str_size = be16_to_cpu(h.id_str_size);
107 name_size = be16_to_cpu(h.name_size);
108
109 /* Read extra data */
110 ret = bdrv_pread(bs->file, offset, &extra,
111 MIN(sizeof(extra), extra_data_size));
112 if (ret < 0) {
113 goto fail;
114 }
115 offset += extra_data_size;
116
117 if (extra_data_size >= 8) {
118 sn->vm_state_size = be64_to_cpu(extra.vm_state_size_large);
119 }
120
121 if (extra_data_size >= 16) {
122 sn->disk_size = be64_to_cpu(extra.disk_size);
123 } else {
124 sn->disk_size = bs->total_sectors * BDRV_SECTOR_SIZE;
125 }
126
127 /* Read snapshot ID */
128 sn->id_str = g_malloc(id_str_size + 1);
129 ret = bdrv_pread(bs->file, offset, sn->id_str, id_str_size);
130 if (ret < 0) {
131 goto fail;
132 }
133 offset += id_str_size;
134 sn->id_str[id_str_size] = '\0';
135
136 /* Read snapshot name */
137 sn->name = g_malloc(name_size + 1);
138 ret = bdrv_pread(bs->file, offset, sn->name, name_size);
139 if (ret < 0) {
140 goto fail;
141 }
142 offset += name_size;
143 sn->name[name_size] = '\0';
144 }
145
146 s->snapshots_size = offset - s->snapshots_offset;
147 return 0;
148
149 fail:
150 qcow2_free_snapshots(bs);
151 return ret;
152 }
153
154 /* add at the end of the file a new list of snapshots */
155 static int qcow2_write_snapshots(BlockDriverState *bs)
156 {
157 BDRVQcowState *s = bs->opaque;
158 QCowSnapshot *sn;
159 QCowSnapshotHeader h;
160 QCowSnapshotExtraData extra;
161 int i, name_size, id_str_size, snapshots_size;
162 struct {
163 uint32_t nb_snapshots;
164 uint64_t snapshots_offset;
165 } QEMU_PACKED header_data;
166 int64_t offset, snapshots_offset;
167 int ret;
168
169 /* compute the size of the snapshots */
170 offset = 0;
171 for(i = 0; i < s->nb_snapshots; i++) {
172 sn = s->snapshots + i;
173 offset = align_offset(offset, 8);
174 offset += sizeof(h);
175 offset += sizeof(extra);
176 offset += strlen(sn->id_str);
177 offset += strlen(sn->name);
178 }
179 snapshots_size = offset;
180
181 /* Allocate space for the new snapshot list */
182 snapshots_offset = qcow2_alloc_clusters(bs, snapshots_size);
183 bdrv_flush(bs->file);
184 offset = snapshots_offset;
185 if (offset < 0) {
186 return offset;
187 }
188
189 /* Write all snapshots to the new list */
190 for(i = 0; i < s->nb_snapshots; i++) {
191 sn = s->snapshots + i;
192 memset(&h, 0, sizeof(h));
193 h.l1_table_offset = cpu_to_be64(sn->l1_table_offset);
194 h.l1_size = cpu_to_be32(sn->l1_size);
195 /* If it doesn't fit in 32 bit, older implementations should treat it
196 * as a disk-only snapshot rather than truncate the VM state */
197 if (sn->vm_state_size <= 0xffffffff) {
198 h.vm_state_size = cpu_to_be32(sn->vm_state_size);
199 }
200 h.date_sec = cpu_to_be32(sn->date_sec);
201 h.date_nsec = cpu_to_be32(sn->date_nsec);
202 h.vm_clock_nsec = cpu_to_be64(sn->vm_clock_nsec);
203 h.extra_data_size = cpu_to_be32(sizeof(extra));
204
205 memset(&extra, 0, sizeof(extra));
206 extra.vm_state_size_large = cpu_to_be64(sn->vm_state_size);
207 extra.disk_size = cpu_to_be64(sn->disk_size);
208
209 id_str_size = strlen(sn->id_str);
210 name_size = strlen(sn->name);
211 h.id_str_size = cpu_to_be16(id_str_size);
212 h.name_size = cpu_to_be16(name_size);
213 offset = align_offset(offset, 8);
214
215 ret = bdrv_pwrite(bs->file, offset, &h, sizeof(h));
216 if (ret < 0) {
217 goto fail;
218 }
219 offset += sizeof(h);
220
221 ret = bdrv_pwrite(bs->file, offset, &extra, sizeof(extra));
222 if (ret < 0) {
223 goto fail;
224 }
225 offset += sizeof(extra);
226
227 ret = bdrv_pwrite(bs->file, offset, sn->id_str, id_str_size);
228 if (ret < 0) {
229 goto fail;
230 }
231 offset += id_str_size;
232
233 ret = bdrv_pwrite(bs->file, offset, sn->name, name_size);
234 if (ret < 0) {
235 goto fail;
236 }
237 offset += name_size;
238 }
239
240 /*
241 * Update the header to point to the new snapshot table. This requires the
242 * new table and its refcounts to be stable on disk.
243 */
244 ret = bdrv_flush(bs);
245 if (ret < 0) {
246 goto fail;
247 }
248
249 QEMU_BUILD_BUG_ON(offsetof(QCowHeader, snapshots_offset) !=
250 offsetof(QCowHeader, nb_snapshots) + sizeof(header_data.nb_snapshots));
251
252 header_data.nb_snapshots = cpu_to_be32(s->nb_snapshots);
253 header_data.snapshots_offset = cpu_to_be64(snapshots_offset);
254
255 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, nb_snapshots),
256 &header_data, sizeof(header_data));
257 if (ret < 0) {
258 goto fail;
259 }
260
261 /* free the old snapshot table */
262 qcow2_free_clusters(bs, s->snapshots_offset, s->snapshots_size);
263 s->snapshots_offset = snapshots_offset;
264 s->snapshots_size = snapshots_size;
265 return 0;
266
267 fail:
268 return ret;
269 }
270
271 static void find_new_snapshot_id(BlockDriverState *bs,
272 char *id_str, int id_str_size)
273 {
274 BDRVQcowState *s = bs->opaque;
275 QCowSnapshot *sn;
276 int i, id, id_max = 0;
277
278 for(i = 0; i < s->nb_snapshots; i++) {
279 sn = s->snapshots + i;
280 id = strtoul(sn->id_str, NULL, 10);
281 if (id > id_max)
282 id_max = id;
283 }
284 snprintf(id_str, id_str_size, "%d", id_max + 1);
285 }
286
287 static int find_snapshot_by_id(BlockDriverState *bs, const char *id_str)
288 {
289 BDRVQcowState *s = bs->opaque;
290 int i;
291
292 for(i = 0; i < s->nb_snapshots; i++) {
293 if (!strcmp(s->snapshots[i].id_str, id_str))
294 return i;
295 }
296 return -1;
297 }
298
299 static int find_snapshot_by_id_or_name(BlockDriverState *bs, const char *name)
300 {
301 BDRVQcowState *s = bs->opaque;
302 int i, ret;
303
304 ret = find_snapshot_by_id(bs, name);
305 if (ret >= 0)
306 return ret;
307 for(i = 0; i < s->nb_snapshots; i++) {
308 if (!strcmp(s->snapshots[i].name, name))
309 return i;
310 }
311 return -1;
312 }
313
314 /* if no id is provided, a new one is constructed */
315 int qcow2_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info)
316 {
317 BDRVQcowState *s = bs->opaque;
318 QCowSnapshot *new_snapshot_list = NULL;
319 QCowSnapshot *old_snapshot_list = NULL;
320 QCowSnapshot sn1, *sn = &sn1;
321 int i, ret;
322 uint64_t *l1_table = NULL;
323 int64_t l1_table_offset;
324
325 memset(sn, 0, sizeof(*sn));
326
327 /* Generate an ID if it wasn't passed */
328 if (sn_info->id_str[0] == '\0') {
329 find_new_snapshot_id(bs, sn_info->id_str, sizeof(sn_info->id_str));
330 }
331
332 /* Check that the ID is unique */
333 if (find_snapshot_by_id(bs, sn_info->id_str) >= 0) {
334 return -EEXIST;
335 }
336
337 /* Populate sn with passed data */
338 sn->id_str = g_strdup(sn_info->id_str);
339 sn->name = g_strdup(sn_info->name);
340
341 sn->disk_size = bs->total_sectors * BDRV_SECTOR_SIZE;
342 sn->vm_state_size = sn_info->vm_state_size;
343 sn->date_sec = sn_info->date_sec;
344 sn->date_nsec = sn_info->date_nsec;
345 sn->vm_clock_nsec = sn_info->vm_clock_nsec;
346
347 /* Allocate the L1 table of the snapshot and copy the current one there. */
348 l1_table_offset = qcow2_alloc_clusters(bs, s->l1_size * sizeof(uint64_t));
349 if (l1_table_offset < 0) {
350 ret = l1_table_offset;
351 goto fail;
352 }
353
354 sn->l1_table_offset = l1_table_offset;
355 sn->l1_size = s->l1_size;
356
357 l1_table = g_malloc(s->l1_size * sizeof(uint64_t));
358 for(i = 0; i < s->l1_size; i++) {
359 l1_table[i] = cpu_to_be64(s->l1_table[i]);
360 }
361
362 ret = bdrv_pwrite(bs->file, sn->l1_table_offset, l1_table,
363 s->l1_size * sizeof(uint64_t));
364 if (ret < 0) {
365 goto fail;
366 }
367
368 g_free(l1_table);
369 l1_table = NULL;
370
371 /*
372 * Increase the refcounts of all clusters and make sure everything is
373 * stable on disk before updating the snapshot table to contain a pointer
374 * to the new L1 table.
375 */
376 ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1);
377 if (ret < 0) {
378 goto fail;
379 }
380
381 ret = bdrv_flush(bs);
382 if (ret < 0) {
383 goto fail;
384 }
385
386 /* Append the new snapshot to the snapshot list */
387 new_snapshot_list = g_malloc((s->nb_snapshots + 1) * sizeof(QCowSnapshot));
388 if (s->snapshots) {
389 memcpy(new_snapshot_list, s->snapshots,
390 s->nb_snapshots * sizeof(QCowSnapshot));
391 old_snapshot_list = s->snapshots;
392 }
393 s->snapshots = new_snapshot_list;
394 s->snapshots[s->nb_snapshots++] = *sn;
395
396 ret = qcow2_write_snapshots(bs);
397 if (ret < 0) {
398 g_free(s->snapshots);
399 s->snapshots = old_snapshot_list;
400 goto fail;
401 }
402
403 g_free(old_snapshot_list);
404
405 #ifdef DEBUG_ALLOC
406 {
407 BdrvCheckResult result = {0};
408 qcow2_check_refcounts(bs, &result, 0);
409 }
410 #endif
411 return 0;
412
413 fail:
414 g_free(sn->id_str);
415 g_free(sn->name);
416 g_free(l1_table);
417
418 return ret;
419 }
420
421 /* copy the snapshot 'snapshot_name' into the current disk image */
422 int qcow2_snapshot_goto(BlockDriverState *bs, const char *snapshot_id)
423 {
424 BDRVQcowState *s = bs->opaque;
425 QCowSnapshot *sn;
426 int i, snapshot_index;
427 int cur_l1_bytes, sn_l1_bytes;
428 int ret;
429 uint64_t *sn_l1_table = NULL;
430
431 /* Search the snapshot */
432 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
433 if (snapshot_index < 0) {
434 return -ENOENT;
435 }
436 sn = &s->snapshots[snapshot_index];
437
438 if (sn->disk_size != bs->total_sectors * BDRV_SECTOR_SIZE) {
439 error_report("qcow2: Loading snapshots with different disk "
440 "size is not implemented");
441 ret = -ENOTSUP;
442 goto fail;
443 }
444
445 /*
446 * Make sure that the current L1 table is big enough to contain the whole
447 * L1 table of the snapshot. If the snapshot L1 table is smaller, the
448 * current one must be padded with zeros.
449 */
450 ret = qcow2_grow_l1_table(bs, sn->l1_size, true);
451 if (ret < 0) {
452 goto fail;
453 }
454
455 cur_l1_bytes = s->l1_size * sizeof(uint64_t);
456 sn_l1_bytes = sn->l1_size * sizeof(uint64_t);
457
458 /*
459 * Copy the snapshot L1 table to the current L1 table.
460 *
461 * Before overwriting the old current L1 table on disk, make sure to
462 * increase all refcounts for the clusters referenced by the new one.
463 * Decrease the refcount referenced by the old one only when the L1
464 * table is overwritten.
465 */
466 sn_l1_table = g_malloc0(cur_l1_bytes);
467
468 ret = bdrv_pread(bs->file, sn->l1_table_offset, sn_l1_table, sn_l1_bytes);
469 if (ret < 0) {
470 goto fail;
471 }
472
473 ret = qcow2_update_snapshot_refcount(bs, sn->l1_table_offset,
474 sn->l1_size, 1);
475 if (ret < 0) {
476 goto fail;
477 }
478
479 ret = bdrv_pwrite_sync(bs->file, s->l1_table_offset, sn_l1_table,
480 cur_l1_bytes);
481 if (ret < 0) {
482 goto fail;
483 }
484
485 /*
486 * Decrease refcount of clusters of current L1 table.
487 *
488 * At this point, the in-memory s->l1_table points to the old L1 table,
489 * whereas on disk we already have the new one.
490 *
491 * qcow2_update_snapshot_refcount special cases the current L1 table to use
492 * the in-memory data instead of really using the offset to load a new one,
493 * which is why this works.
494 */
495 ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset,
496 s->l1_size, -1);
497
498 /*
499 * Now update the in-memory L1 table to be in sync with the on-disk one. We
500 * need to do this even if updating refcounts failed.
501 */
502 for(i = 0;i < s->l1_size; i++) {
503 s->l1_table[i] = be64_to_cpu(sn_l1_table[i]);
504 }
505
506 if (ret < 0) {
507 goto fail;
508 }
509
510 g_free(sn_l1_table);
511 sn_l1_table = NULL;
512
513 /*
514 * Update QCOW_OFLAG_COPIED in the active L1 table (it may have changed
515 * when we decreased the refcount of the old snapshot.
516 */
517 ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
518 if (ret < 0) {
519 goto fail;
520 }
521
522 #ifdef DEBUG_ALLOC
523 {
524 BdrvCheckResult result = {0};
525 qcow2_check_refcounts(bs, &result, 0);
526 }
527 #endif
528 return 0;
529
530 fail:
531 g_free(sn_l1_table);
532 return ret;
533 }
534
535 int qcow2_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
536 {
537 BDRVQcowState *s = bs->opaque;
538 QCowSnapshot sn;
539 int snapshot_index, ret;
540
541 /* Search the snapshot */
542 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
543 if (snapshot_index < 0) {
544 return -ENOENT;
545 }
546 sn = s->snapshots[snapshot_index];
547
548 /* Remove it from the snapshot list */
549 memmove(s->snapshots + snapshot_index,
550 s->snapshots + snapshot_index + 1,
551 (s->nb_snapshots - snapshot_index - 1) * sizeof(sn));
552 s->nb_snapshots--;
553 ret = qcow2_write_snapshots(bs);
554 if (ret < 0) {
555 return ret;
556 }
557
558 /*
559 * The snapshot is now unused, clean up. If we fail after this point, we
560 * won't recover but just leak clusters.
561 */
562 g_free(sn.id_str);
563 g_free(sn.name);
564
565 /*
566 * Now decrease the refcounts of clusters referenced by the snapshot and
567 * free the L1 table.
568 */
569 ret = qcow2_update_snapshot_refcount(bs, sn.l1_table_offset,
570 sn.l1_size, -1);
571 if (ret < 0) {
572 return ret;
573 }
574 qcow2_free_clusters(bs, sn.l1_table_offset, sn.l1_size * sizeof(uint64_t));
575
576 /* must update the copied flag on the current cluster offsets */
577 ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
578 if (ret < 0) {
579 return ret;
580 }
581
582 #ifdef DEBUG_ALLOC
583 {
584 BdrvCheckResult result = {0};
585 qcow2_check_refcounts(bs, &result, 0);
586 }
587 #endif
588 return 0;
589 }
590
591 int qcow2_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab)
592 {
593 BDRVQcowState *s = bs->opaque;
594 QEMUSnapshotInfo *sn_tab, *sn_info;
595 QCowSnapshot *sn;
596 int i;
597
598 if (!s->nb_snapshots) {
599 *psn_tab = NULL;
600 return s->nb_snapshots;
601 }
602
603 sn_tab = g_malloc0(s->nb_snapshots * sizeof(QEMUSnapshotInfo));
604 for(i = 0; i < s->nb_snapshots; i++) {
605 sn_info = sn_tab + i;
606 sn = s->snapshots + i;
607 pstrcpy(sn_info->id_str, sizeof(sn_info->id_str),
608 sn->id_str);
609 pstrcpy(sn_info->name, sizeof(sn_info->name),
610 sn->name);
611 sn_info->vm_state_size = sn->vm_state_size;
612 sn_info->date_sec = sn->date_sec;
613 sn_info->date_nsec = sn->date_nsec;
614 sn_info->vm_clock_nsec = sn->vm_clock_nsec;
615 }
616 *psn_tab = sn_tab;
617 return s->nb_snapshots;
618 }
619
620 int qcow2_snapshot_load_tmp(BlockDriverState *bs, const char *snapshot_name)
621 {
622 int i, snapshot_index;
623 BDRVQcowState *s = bs->opaque;
624 QCowSnapshot *sn;
625 uint64_t *new_l1_table;
626 int new_l1_bytes;
627 int ret;
628
629 assert(bs->read_only);
630
631 /* Search the snapshot */
632 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_name);
633 if (snapshot_index < 0) {
634 return -ENOENT;
635 }
636 sn = &s->snapshots[snapshot_index];
637
638 /* Allocate and read in the snapshot's L1 table */
639 new_l1_bytes = s->l1_size * sizeof(uint64_t);
640 new_l1_table = g_malloc0(align_offset(new_l1_bytes, 512));
641
642 ret = bdrv_pread(bs->file, sn->l1_table_offset, new_l1_table, new_l1_bytes);
643 if (ret < 0) {
644 g_free(new_l1_table);
645 return ret;
646 }
647
648 /* Switch the L1 table */
649 g_free(s->l1_table);
650
651 s->l1_size = sn->l1_size;
652 s->l1_table_offset = sn->l1_table_offset;
653 s->l1_table = new_l1_table;
654
655 for(i = 0;i < s->l1_size; i++) {
656 be64_to_cpus(&s->l1_table[i]);
657 }
658
659 return 0;
660 }