]> git.proxmox.com Git - qemu.git/blob - block/qcow2-snapshot.c
qcow2: flush refcount cache correctly in qcow2_write_snapshots()
[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 s->snapshots_offset = snapshots_offset;
267 s->snapshots_size = snapshots_size;
268 return 0;
269
270 fail:
271 return ret;
272 }
273
274 static void find_new_snapshot_id(BlockDriverState *bs,
275 char *id_str, int id_str_size)
276 {
277 BDRVQcowState *s = bs->opaque;
278 QCowSnapshot *sn;
279 int i, id, id_max = 0;
280
281 for(i = 0; i < s->nb_snapshots; i++) {
282 sn = s->snapshots + i;
283 id = strtoul(sn->id_str, NULL, 10);
284 if (id > id_max)
285 id_max = id;
286 }
287 snprintf(id_str, id_str_size, "%d", id_max + 1);
288 }
289
290 static int find_snapshot_by_id(BlockDriverState *bs, const char *id_str)
291 {
292 BDRVQcowState *s = bs->opaque;
293 int i;
294
295 for(i = 0; i < s->nb_snapshots; i++) {
296 if (!strcmp(s->snapshots[i].id_str, id_str))
297 return i;
298 }
299 return -1;
300 }
301
302 static int find_snapshot_by_id_or_name(BlockDriverState *bs, const char *name)
303 {
304 BDRVQcowState *s = bs->opaque;
305 int i, ret;
306
307 ret = find_snapshot_by_id(bs, name);
308 if (ret >= 0)
309 return ret;
310 for(i = 0; i < s->nb_snapshots; i++) {
311 if (!strcmp(s->snapshots[i].name, name))
312 return i;
313 }
314 return -1;
315 }
316
317 /* if no id is provided, a new one is constructed */
318 int qcow2_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info)
319 {
320 BDRVQcowState *s = bs->opaque;
321 QCowSnapshot *new_snapshot_list = NULL;
322 QCowSnapshot *old_snapshot_list = NULL;
323 QCowSnapshot sn1, *sn = &sn1;
324 int i, ret;
325 uint64_t *l1_table = NULL;
326 int64_t l1_table_offset;
327
328 memset(sn, 0, sizeof(*sn));
329
330 /* Generate an ID if it wasn't passed */
331 if (sn_info->id_str[0] == '\0') {
332 find_new_snapshot_id(bs, sn_info->id_str, sizeof(sn_info->id_str));
333 }
334
335 /* Check that the ID is unique */
336 if (find_snapshot_by_id(bs, sn_info->id_str) >= 0) {
337 return -EEXIST;
338 }
339
340 /* Populate sn with passed data */
341 sn->id_str = g_strdup(sn_info->id_str);
342 sn->name = g_strdup(sn_info->name);
343
344 sn->disk_size = bs->total_sectors * BDRV_SECTOR_SIZE;
345 sn->vm_state_size = sn_info->vm_state_size;
346 sn->date_sec = sn_info->date_sec;
347 sn->date_nsec = sn_info->date_nsec;
348 sn->vm_clock_nsec = sn_info->vm_clock_nsec;
349
350 /* Allocate the L1 table of the snapshot and copy the current one there. */
351 l1_table_offset = qcow2_alloc_clusters(bs, s->l1_size * sizeof(uint64_t));
352 if (l1_table_offset < 0) {
353 ret = l1_table_offset;
354 goto fail;
355 }
356
357 sn->l1_table_offset = l1_table_offset;
358 sn->l1_size = s->l1_size;
359
360 l1_table = g_malloc(s->l1_size * sizeof(uint64_t));
361 for(i = 0; i < s->l1_size; i++) {
362 l1_table[i] = cpu_to_be64(s->l1_table[i]);
363 }
364
365 ret = bdrv_pwrite(bs->file, sn->l1_table_offset, l1_table,
366 s->l1_size * sizeof(uint64_t));
367 if (ret < 0) {
368 goto fail;
369 }
370
371 g_free(l1_table);
372 l1_table = NULL;
373
374 /*
375 * Increase the refcounts of all clusters and make sure everything is
376 * stable on disk before updating the snapshot table to contain a pointer
377 * to the new L1 table.
378 */
379 ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1);
380 if (ret < 0) {
381 goto fail;
382 }
383
384 ret = bdrv_flush(bs);
385 if (ret < 0) {
386 goto fail;
387 }
388
389 /* Append the new snapshot to the snapshot list */
390 new_snapshot_list = g_malloc((s->nb_snapshots + 1) * sizeof(QCowSnapshot));
391 if (s->snapshots) {
392 memcpy(new_snapshot_list, s->snapshots,
393 s->nb_snapshots * sizeof(QCowSnapshot));
394 old_snapshot_list = s->snapshots;
395 }
396 s->snapshots = new_snapshot_list;
397 s->snapshots[s->nb_snapshots++] = *sn;
398
399 ret = qcow2_write_snapshots(bs);
400 if (ret < 0) {
401 g_free(s->snapshots);
402 s->snapshots = old_snapshot_list;
403 goto fail;
404 }
405
406 g_free(old_snapshot_list);
407
408 #ifdef DEBUG_ALLOC
409 {
410 BdrvCheckResult result = {0};
411 qcow2_check_refcounts(bs, &result, 0);
412 }
413 #endif
414 return 0;
415
416 fail:
417 g_free(sn->id_str);
418 g_free(sn->name);
419 g_free(l1_table);
420
421 return ret;
422 }
423
424 /* copy the snapshot 'snapshot_name' into the current disk image */
425 int qcow2_snapshot_goto(BlockDriverState *bs, const char *snapshot_id)
426 {
427 BDRVQcowState *s = bs->opaque;
428 QCowSnapshot *sn;
429 int i, snapshot_index;
430 int cur_l1_bytes, sn_l1_bytes;
431 int ret;
432 uint64_t *sn_l1_table = NULL;
433
434 /* Search the snapshot */
435 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
436 if (snapshot_index < 0) {
437 return -ENOENT;
438 }
439 sn = &s->snapshots[snapshot_index];
440
441 if (sn->disk_size != bs->total_sectors * BDRV_SECTOR_SIZE) {
442 error_report("qcow2: Loading snapshots with different disk "
443 "size is not implemented");
444 ret = -ENOTSUP;
445 goto fail;
446 }
447
448 /*
449 * Make sure that the current L1 table is big enough to contain the whole
450 * L1 table of the snapshot. If the snapshot L1 table is smaller, the
451 * current one must be padded with zeros.
452 */
453 ret = qcow2_grow_l1_table(bs, sn->l1_size, true);
454 if (ret < 0) {
455 goto fail;
456 }
457
458 cur_l1_bytes = s->l1_size * sizeof(uint64_t);
459 sn_l1_bytes = sn->l1_size * sizeof(uint64_t);
460
461 /*
462 * Copy the snapshot L1 table to the current L1 table.
463 *
464 * Before overwriting the old current L1 table on disk, make sure to
465 * increase all refcounts for the clusters referenced by the new one.
466 * Decrease the refcount referenced by the old one only when the L1
467 * table is overwritten.
468 */
469 sn_l1_table = g_malloc0(cur_l1_bytes);
470
471 ret = bdrv_pread(bs->file, sn->l1_table_offset, sn_l1_table, sn_l1_bytes);
472 if (ret < 0) {
473 goto fail;
474 }
475
476 ret = qcow2_update_snapshot_refcount(bs, sn->l1_table_offset,
477 sn->l1_size, 1);
478 if (ret < 0) {
479 goto fail;
480 }
481
482 ret = bdrv_pwrite_sync(bs->file, s->l1_table_offset, sn_l1_table,
483 cur_l1_bytes);
484 if (ret < 0) {
485 goto fail;
486 }
487
488 /*
489 * Decrease refcount of clusters of current L1 table.
490 *
491 * At this point, the in-memory s->l1_table points to the old L1 table,
492 * whereas on disk we already have the new one.
493 *
494 * qcow2_update_snapshot_refcount special cases the current L1 table to use
495 * the in-memory data instead of really using the offset to load a new one,
496 * which is why this works.
497 */
498 ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset,
499 s->l1_size, -1);
500
501 /*
502 * Now update the in-memory L1 table to be in sync with the on-disk one. We
503 * need to do this even if updating refcounts failed.
504 */
505 for(i = 0;i < s->l1_size; i++) {
506 s->l1_table[i] = be64_to_cpu(sn_l1_table[i]);
507 }
508
509 if (ret < 0) {
510 goto fail;
511 }
512
513 g_free(sn_l1_table);
514 sn_l1_table = NULL;
515
516 /*
517 * Update QCOW_OFLAG_COPIED in the active L1 table (it may have changed
518 * when we decreased the refcount of the old snapshot.
519 */
520 ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
521 if (ret < 0) {
522 goto fail;
523 }
524
525 #ifdef DEBUG_ALLOC
526 {
527 BdrvCheckResult result = {0};
528 qcow2_check_refcounts(bs, &result, 0);
529 }
530 #endif
531 return 0;
532
533 fail:
534 g_free(sn_l1_table);
535 return ret;
536 }
537
538 int qcow2_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
539 {
540 BDRVQcowState *s = bs->opaque;
541 QCowSnapshot sn;
542 int snapshot_index, ret;
543
544 /* Search the snapshot */
545 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
546 if (snapshot_index < 0) {
547 return -ENOENT;
548 }
549 sn = s->snapshots[snapshot_index];
550
551 /* Remove it from the snapshot list */
552 memmove(s->snapshots + snapshot_index,
553 s->snapshots + snapshot_index + 1,
554 (s->nb_snapshots - snapshot_index - 1) * sizeof(sn));
555 s->nb_snapshots--;
556 ret = qcow2_write_snapshots(bs);
557 if (ret < 0) {
558 return ret;
559 }
560
561 /*
562 * The snapshot is now unused, clean up. If we fail after this point, we
563 * won't recover but just leak clusters.
564 */
565 g_free(sn.id_str);
566 g_free(sn.name);
567
568 /*
569 * Now decrease the refcounts of clusters referenced by the snapshot and
570 * free the L1 table.
571 */
572 ret = qcow2_update_snapshot_refcount(bs, sn.l1_table_offset,
573 sn.l1_size, -1);
574 if (ret < 0) {
575 return ret;
576 }
577 qcow2_free_clusters(bs, sn.l1_table_offset, sn.l1_size * sizeof(uint64_t));
578
579 /* must update the copied flag on the current cluster offsets */
580 ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
581 if (ret < 0) {
582 return ret;
583 }
584
585 #ifdef DEBUG_ALLOC
586 {
587 BdrvCheckResult result = {0};
588 qcow2_check_refcounts(bs, &result, 0);
589 }
590 #endif
591 return 0;
592 }
593
594 int qcow2_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab)
595 {
596 BDRVQcowState *s = bs->opaque;
597 QEMUSnapshotInfo *sn_tab, *sn_info;
598 QCowSnapshot *sn;
599 int i;
600
601 if (!s->nb_snapshots) {
602 *psn_tab = NULL;
603 return s->nb_snapshots;
604 }
605
606 sn_tab = g_malloc0(s->nb_snapshots * sizeof(QEMUSnapshotInfo));
607 for(i = 0; i < s->nb_snapshots; i++) {
608 sn_info = sn_tab + i;
609 sn = s->snapshots + i;
610 pstrcpy(sn_info->id_str, sizeof(sn_info->id_str),
611 sn->id_str);
612 pstrcpy(sn_info->name, sizeof(sn_info->name),
613 sn->name);
614 sn_info->vm_state_size = sn->vm_state_size;
615 sn_info->date_sec = sn->date_sec;
616 sn_info->date_nsec = sn->date_nsec;
617 sn_info->vm_clock_nsec = sn->vm_clock_nsec;
618 }
619 *psn_tab = sn_tab;
620 return s->nb_snapshots;
621 }
622
623 int qcow2_snapshot_load_tmp(BlockDriverState *bs, const char *snapshot_name)
624 {
625 int i, snapshot_index;
626 BDRVQcowState *s = bs->opaque;
627 QCowSnapshot *sn;
628 uint64_t *new_l1_table;
629 int new_l1_bytes;
630 int ret;
631
632 assert(bs->read_only);
633
634 /* Search the snapshot */
635 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_name);
636 if (snapshot_index < 0) {
637 return -ENOENT;
638 }
639 sn = &s->snapshots[snapshot_index];
640
641 /* Allocate and read in the snapshot's L1 table */
642 new_l1_bytes = s->l1_size * sizeof(uint64_t);
643 new_l1_table = g_malloc0(align_offset(new_l1_bytes, 512));
644
645 ret = bdrv_pread(bs->file, sn->l1_table_offset, new_l1_table, new_l1_bytes);
646 if (ret < 0) {
647 g_free(new_l1_table);
648 return ret;
649 }
650
651 /* Switch the L1 table */
652 g_free(s->l1_table);
653
654 s->l1_size = sn->l1_size;
655 s->l1_table_offset = sn->l1_table_offset;
656 s->l1_table = new_l1_table;
657
658 for(i = 0;i < s->l1_size; i++) {
659 be64_to_cpus(&s->l1_table[i]);
660 }
661
662 return 0;
663 }