]> git.proxmox.com Git - qemu.git/blob - block/qcow2-snapshot.c
Merge remote-tracking branch 'pmaydell/tags/pull-arm-devs-20130812' 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 offset = snapshots_offset;
184 if (offset < 0) {
185 return offset;
186 }
187 ret = bdrv_flush(bs);
188 if (ret < 0) {
189 return ret;
190 }
191
192 /* Write all snapshots to the new list */
193 for(i = 0; i < s->nb_snapshots; i++) {
194 sn = s->snapshots + i;
195 memset(&h, 0, sizeof(h));
196 h.l1_table_offset = cpu_to_be64(sn->l1_table_offset);
197 h.l1_size = cpu_to_be32(sn->l1_size);
198 /* If it doesn't fit in 32 bit, older implementations should treat it
199 * as a disk-only snapshot rather than truncate the VM state */
200 if (sn->vm_state_size <= 0xffffffff) {
201 h.vm_state_size = cpu_to_be32(sn->vm_state_size);
202 }
203 h.date_sec = cpu_to_be32(sn->date_sec);
204 h.date_nsec = cpu_to_be32(sn->date_nsec);
205 h.vm_clock_nsec = cpu_to_be64(sn->vm_clock_nsec);
206 h.extra_data_size = cpu_to_be32(sizeof(extra));
207
208 memset(&extra, 0, sizeof(extra));
209 extra.vm_state_size_large = cpu_to_be64(sn->vm_state_size);
210 extra.disk_size = cpu_to_be64(sn->disk_size);
211
212 id_str_size = strlen(sn->id_str);
213 name_size = strlen(sn->name);
214 h.id_str_size = cpu_to_be16(id_str_size);
215 h.name_size = cpu_to_be16(name_size);
216 offset = align_offset(offset, 8);
217
218 ret = bdrv_pwrite(bs->file, offset, &h, sizeof(h));
219 if (ret < 0) {
220 goto fail;
221 }
222 offset += sizeof(h);
223
224 ret = bdrv_pwrite(bs->file, offset, &extra, sizeof(extra));
225 if (ret < 0) {
226 goto fail;
227 }
228 offset += sizeof(extra);
229
230 ret = bdrv_pwrite(bs->file, offset, sn->id_str, id_str_size);
231 if (ret < 0) {
232 goto fail;
233 }
234 offset += id_str_size;
235
236 ret = bdrv_pwrite(bs->file, offset, sn->name, name_size);
237 if (ret < 0) {
238 goto fail;
239 }
240 offset += name_size;
241 }
242
243 /*
244 * Update the header to point to the new snapshot table. This requires the
245 * new table and its refcounts to be stable on disk.
246 */
247 ret = bdrv_flush(bs);
248 if (ret < 0) {
249 goto fail;
250 }
251
252 QEMU_BUILD_BUG_ON(offsetof(QCowHeader, snapshots_offset) !=
253 offsetof(QCowHeader, nb_snapshots) + sizeof(header_data.nb_snapshots));
254
255 header_data.nb_snapshots = cpu_to_be32(s->nb_snapshots);
256 header_data.snapshots_offset = cpu_to_be64(snapshots_offset);
257
258 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, nb_snapshots),
259 &header_data, sizeof(header_data));
260 if (ret < 0) {
261 goto fail;
262 }
263
264 /* free the old snapshot table */
265 qcow2_free_clusters(bs, s->snapshots_offset, s->snapshots_size,
266 QCOW2_DISCARD_SNAPSHOT);
267 s->snapshots_offset = snapshots_offset;
268 s->snapshots_size = snapshots_size;
269 return 0;
270
271 fail:
272 return ret;
273 }
274
275 static void find_new_snapshot_id(BlockDriverState *bs,
276 char *id_str, int id_str_size)
277 {
278 BDRVQcowState *s = bs->opaque;
279 QCowSnapshot *sn;
280 int i, id, id_max = 0;
281
282 for(i = 0; i < s->nb_snapshots; i++) {
283 sn = s->snapshots + i;
284 id = strtoul(sn->id_str, NULL, 10);
285 if (id > id_max)
286 id_max = id;
287 }
288 snprintf(id_str, id_str_size, "%d", id_max + 1);
289 }
290
291 static int find_snapshot_by_id(BlockDriverState *bs, const char *id_str)
292 {
293 BDRVQcowState *s = bs->opaque;
294 int i;
295
296 for(i = 0; i < s->nb_snapshots; i++) {
297 if (!strcmp(s->snapshots[i].id_str, id_str))
298 return i;
299 }
300 return -1;
301 }
302
303 static int find_snapshot_by_id_or_name(BlockDriverState *bs, const char *name)
304 {
305 BDRVQcowState *s = bs->opaque;
306 int i, ret;
307
308 ret = find_snapshot_by_id(bs, name);
309 if (ret >= 0)
310 return ret;
311 for(i = 0; i < s->nb_snapshots; i++) {
312 if (!strcmp(s->snapshots[i].name, name))
313 return i;
314 }
315 return -1;
316 }
317
318 /* if no id is provided, a new one is constructed */
319 int qcow2_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info)
320 {
321 BDRVQcowState *s = bs->opaque;
322 QCowSnapshot *new_snapshot_list = NULL;
323 QCowSnapshot *old_snapshot_list = NULL;
324 QCowSnapshot sn1, *sn = &sn1;
325 int i, ret;
326 uint64_t *l1_table = NULL;
327 int64_t l1_table_offset;
328
329 memset(sn, 0, sizeof(*sn));
330
331 /* Generate an ID if it wasn't passed */
332 if (sn_info->id_str[0] == '\0') {
333 find_new_snapshot_id(bs, sn_info->id_str, sizeof(sn_info->id_str));
334 }
335
336 /* Check that the ID is unique */
337 if (find_snapshot_by_id(bs, sn_info->id_str) >= 0) {
338 return -EEXIST;
339 }
340
341 /* Populate sn with passed data */
342 sn->id_str = g_strdup(sn_info->id_str);
343 sn->name = g_strdup(sn_info->name);
344
345 sn->disk_size = bs->total_sectors * BDRV_SECTOR_SIZE;
346 sn->vm_state_size = sn_info->vm_state_size;
347 sn->date_sec = sn_info->date_sec;
348 sn->date_nsec = sn_info->date_nsec;
349 sn->vm_clock_nsec = sn_info->vm_clock_nsec;
350
351 /* Allocate the L1 table of the snapshot and copy the current one there. */
352 l1_table_offset = qcow2_alloc_clusters(bs, s->l1_size * sizeof(uint64_t));
353 if (l1_table_offset < 0) {
354 ret = l1_table_offset;
355 goto fail;
356 }
357
358 sn->l1_table_offset = l1_table_offset;
359 sn->l1_size = s->l1_size;
360
361 l1_table = g_malloc(s->l1_size * sizeof(uint64_t));
362 for(i = 0; i < s->l1_size; i++) {
363 l1_table[i] = cpu_to_be64(s->l1_table[i]);
364 }
365
366 ret = bdrv_pwrite(bs->file, sn->l1_table_offset, l1_table,
367 s->l1_size * sizeof(uint64_t));
368 if (ret < 0) {
369 goto fail;
370 }
371
372 g_free(l1_table);
373 l1_table = NULL;
374
375 /*
376 * Increase the refcounts of all clusters and make sure everything is
377 * stable on disk before updating the snapshot table to contain a pointer
378 * to the new L1 table.
379 */
380 ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1);
381 if (ret < 0) {
382 goto fail;
383 }
384
385 /* Append the new snapshot to the snapshot list */
386 new_snapshot_list = g_malloc((s->nb_snapshots + 1) * sizeof(QCowSnapshot));
387 if (s->snapshots) {
388 memcpy(new_snapshot_list, s->snapshots,
389 s->nb_snapshots * sizeof(QCowSnapshot));
390 old_snapshot_list = s->snapshots;
391 }
392 s->snapshots = new_snapshot_list;
393 s->snapshots[s->nb_snapshots++] = *sn;
394
395 ret = qcow2_write_snapshots(bs);
396 if (ret < 0) {
397 g_free(s->snapshots);
398 s->snapshots = old_snapshot_list;
399 goto fail;
400 }
401
402 g_free(old_snapshot_list);
403
404 #ifdef DEBUG_ALLOC
405 {
406 BdrvCheckResult result = {0};
407 qcow2_check_refcounts(bs, &result, 0);
408 }
409 #endif
410 return 0;
411
412 fail:
413 g_free(sn->id_str);
414 g_free(sn->name);
415 g_free(l1_table);
416
417 return ret;
418 }
419
420 /* copy the snapshot 'snapshot_name' into the current disk image */
421 int qcow2_snapshot_goto(BlockDriverState *bs, const char *snapshot_id)
422 {
423 BDRVQcowState *s = bs->opaque;
424 QCowSnapshot *sn;
425 int i, snapshot_index;
426 int cur_l1_bytes, sn_l1_bytes;
427 int ret;
428 uint64_t *sn_l1_table = NULL;
429
430 /* Search the snapshot */
431 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
432 if (snapshot_index < 0) {
433 return -ENOENT;
434 }
435 sn = &s->snapshots[snapshot_index];
436
437 if (sn->disk_size != bs->total_sectors * BDRV_SECTOR_SIZE) {
438 error_report("qcow2: Loading snapshots with different disk "
439 "size is not implemented");
440 ret = -ENOTSUP;
441 goto fail;
442 }
443
444 /*
445 * Make sure that the current L1 table is big enough to contain the whole
446 * L1 table of the snapshot. If the snapshot L1 table is smaller, the
447 * current one must be padded with zeros.
448 */
449 ret = qcow2_grow_l1_table(bs, sn->l1_size, true);
450 if (ret < 0) {
451 goto fail;
452 }
453
454 cur_l1_bytes = s->l1_size * sizeof(uint64_t);
455 sn_l1_bytes = sn->l1_size * sizeof(uint64_t);
456
457 /*
458 * Copy the snapshot L1 table to the current L1 table.
459 *
460 * Before overwriting the old current L1 table on disk, make sure to
461 * increase all refcounts for the clusters referenced by the new one.
462 * Decrease the refcount referenced by the old one only when the L1
463 * table is overwritten.
464 */
465 sn_l1_table = g_malloc0(cur_l1_bytes);
466
467 ret = bdrv_pread(bs->file, sn->l1_table_offset, sn_l1_table, sn_l1_bytes);
468 if (ret < 0) {
469 goto fail;
470 }
471
472 ret = qcow2_update_snapshot_refcount(bs, sn->l1_table_offset,
473 sn->l1_size, 1);
474 if (ret < 0) {
475 goto fail;
476 }
477
478 ret = bdrv_pwrite_sync(bs->file, s->l1_table_offset, sn_l1_table,
479 cur_l1_bytes);
480 if (ret < 0) {
481 goto fail;
482 }
483
484 /*
485 * Decrease refcount of clusters of current L1 table.
486 *
487 * At this point, the in-memory s->l1_table points to the old L1 table,
488 * whereas on disk we already have the new one.
489 *
490 * qcow2_update_snapshot_refcount special cases the current L1 table to use
491 * the in-memory data instead of really using the offset to load a new one,
492 * which is why this works.
493 */
494 ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset,
495 s->l1_size, -1);
496
497 /*
498 * Now update the in-memory L1 table to be in sync with the on-disk one. We
499 * need to do this even if updating refcounts failed.
500 */
501 for(i = 0;i < s->l1_size; i++) {
502 s->l1_table[i] = be64_to_cpu(sn_l1_table[i]);
503 }
504
505 if (ret < 0) {
506 goto fail;
507 }
508
509 g_free(sn_l1_table);
510 sn_l1_table = NULL;
511
512 /*
513 * Update QCOW_OFLAG_COPIED in the active L1 table (it may have changed
514 * when we decreased the refcount of the old snapshot.
515 */
516 ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
517 if (ret < 0) {
518 goto fail;
519 }
520
521 #ifdef DEBUG_ALLOC
522 {
523 BdrvCheckResult result = {0};
524 qcow2_check_refcounts(bs, &result, 0);
525 }
526 #endif
527 return 0;
528
529 fail:
530 g_free(sn_l1_table);
531 return ret;
532 }
533
534 int qcow2_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
535 {
536 BDRVQcowState *s = bs->opaque;
537 QCowSnapshot sn;
538 int snapshot_index, ret;
539
540 /* Search the snapshot */
541 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
542 if (snapshot_index < 0) {
543 return -ENOENT;
544 }
545 sn = s->snapshots[snapshot_index];
546
547 /* Remove it from the snapshot list */
548 memmove(s->snapshots + snapshot_index,
549 s->snapshots + snapshot_index + 1,
550 (s->nb_snapshots - snapshot_index - 1) * sizeof(sn));
551 s->nb_snapshots--;
552 ret = qcow2_write_snapshots(bs);
553 if (ret < 0) {
554 return ret;
555 }
556
557 /*
558 * The snapshot is now unused, clean up. If we fail after this point, we
559 * won't recover but just leak clusters.
560 */
561 g_free(sn.id_str);
562 g_free(sn.name);
563
564 /*
565 * Now decrease the refcounts of clusters referenced by the snapshot and
566 * free the L1 table.
567 */
568 ret = qcow2_update_snapshot_refcount(bs, sn.l1_table_offset,
569 sn.l1_size, -1);
570 if (ret < 0) {
571 return ret;
572 }
573 qcow2_free_clusters(bs, sn.l1_table_offset, sn.l1_size * sizeof(uint64_t),
574 QCOW2_DISCARD_SNAPSHOT);
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 }