qemu-iotests: add infrastructure of fd passing via SCM
[qemu/qmp-unstable.git] / block / qcow2-snapshot.c
blobffead08ca2cdd9a292cb482efe70a54119757a50
1 /*
2 * Block driver for the QCOW version 2 format
4 * Copyright (c) 2004-2006 Fabrice Bellard
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:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
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.
25 #include "qemu-common.h"
26 #include "block/block_int.h"
27 #include "block/qcow2.h"
29 typedef struct QEMU_PACKED QCowSnapshotHeader {
30 /* header is 8 byte aligned */
31 uint64_t l1_table_offset;
33 uint32_t l1_size;
34 uint16_t id_str_size;
35 uint16_t name_size;
37 uint32_t date_sec;
38 uint32_t date_nsec;
40 uint64_t vm_clock_nsec;
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;
49 typedef struct QEMU_PACKED QCowSnapshotExtraData {
50 uint64_t vm_state_size_large;
51 uint64_t disk_size;
52 } QCowSnapshotExtraData;
54 void qcow2_free_snapshots(BlockDriverState *bs)
56 BDRVQcowState *s = bs->opaque;
57 int i;
59 for(i = 0; i < s->nb_snapshots; i++) {
60 g_free(s->snapshots[i].name);
61 g_free(s->snapshots[i].id_str);
63 g_free(s->snapshots);
64 s->snapshots = NULL;
65 s->nb_snapshots = 0;
68 int qcow2_read_snapshots(BlockDriverState *bs)
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;
79 if (!s->nb_snapshots) {
80 s->snapshots = NULL;
81 s->snapshots_size = 0;
82 return 0;
85 offset = s->snapshots_offset;
86 s->snapshots = g_malloc0(s->nb_snapshots * sizeof(QCowSnapshot));
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;
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);
106 id_str_size = be16_to_cpu(h.id_str_size);
107 name_size = be16_to_cpu(h.name_size);
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;
115 offset += extra_data_size;
117 if (extra_data_size >= 8) {
118 sn->vm_state_size = be64_to_cpu(extra.vm_state_size_large);
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;
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;
133 offset += id_str_size;
134 sn->id_str[id_str_size] = '\0';
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;
142 offset += name_size;
143 sn->name[name_size] = '\0';
146 s->snapshots_size = offset - s->snapshots_offset;
147 return 0;
149 fail:
150 qcow2_free_snapshots(bs);
151 return ret;
154 /* add at the end of the file a new list of snapshots */
155 static int qcow2_write_snapshots(BlockDriverState *bs)
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;
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);
179 snapshots_size = offset;
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;
187 ret = bdrv_flush(bs);
188 if (ret < 0) {
189 return ret;
192 /* The snapshot list position has not yet been updated, so these clusters
193 * must indeed be completely free */
194 ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_DEFAULT, offset,
195 s->snapshots_size);
196 if (ret < 0) {
197 return ret;
201 /* Write all snapshots to the new list */
202 for(i = 0; i < s->nb_snapshots; i++) {
203 sn = s->snapshots + i;
204 memset(&h, 0, sizeof(h));
205 h.l1_table_offset = cpu_to_be64(sn->l1_table_offset);
206 h.l1_size = cpu_to_be32(sn->l1_size);
207 /* If it doesn't fit in 32 bit, older implementations should treat it
208 * as a disk-only snapshot rather than truncate the VM state */
209 if (sn->vm_state_size <= 0xffffffff) {
210 h.vm_state_size = cpu_to_be32(sn->vm_state_size);
212 h.date_sec = cpu_to_be32(sn->date_sec);
213 h.date_nsec = cpu_to_be32(sn->date_nsec);
214 h.vm_clock_nsec = cpu_to_be64(sn->vm_clock_nsec);
215 h.extra_data_size = cpu_to_be32(sizeof(extra));
217 memset(&extra, 0, sizeof(extra));
218 extra.vm_state_size_large = cpu_to_be64(sn->vm_state_size);
219 extra.disk_size = cpu_to_be64(sn->disk_size);
221 id_str_size = strlen(sn->id_str);
222 name_size = strlen(sn->name);
223 h.id_str_size = cpu_to_be16(id_str_size);
224 h.name_size = cpu_to_be16(name_size);
225 offset = align_offset(offset, 8);
227 ret = bdrv_pwrite(bs->file, offset, &h, sizeof(h));
228 if (ret < 0) {
229 goto fail;
231 offset += sizeof(h);
233 ret = bdrv_pwrite(bs->file, offset, &extra, sizeof(extra));
234 if (ret < 0) {
235 goto fail;
237 offset += sizeof(extra);
239 ret = bdrv_pwrite(bs->file, offset, sn->id_str, id_str_size);
240 if (ret < 0) {
241 goto fail;
243 offset += id_str_size;
245 ret = bdrv_pwrite(bs->file, offset, sn->name, name_size);
246 if (ret < 0) {
247 goto fail;
249 offset += name_size;
253 * Update the header to point to the new snapshot table. This requires the
254 * new table and its refcounts to be stable on disk.
256 ret = bdrv_flush(bs);
257 if (ret < 0) {
258 goto fail;
261 QEMU_BUILD_BUG_ON(offsetof(QCowHeader, snapshots_offset) !=
262 offsetof(QCowHeader, nb_snapshots) + sizeof(header_data.nb_snapshots));
264 header_data.nb_snapshots = cpu_to_be32(s->nb_snapshots);
265 header_data.snapshots_offset = cpu_to_be64(snapshots_offset);
267 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, nb_snapshots),
268 &header_data, sizeof(header_data));
269 if (ret < 0) {
270 goto fail;
273 /* free the old snapshot table */
274 qcow2_free_clusters(bs, s->snapshots_offset, s->snapshots_size,
275 QCOW2_DISCARD_SNAPSHOT);
276 s->snapshots_offset = snapshots_offset;
277 s->snapshots_size = snapshots_size;
278 return 0;
280 fail:
281 return ret;
284 static void find_new_snapshot_id(BlockDriverState *bs,
285 char *id_str, int id_str_size)
287 BDRVQcowState *s = bs->opaque;
288 QCowSnapshot *sn;
289 int i, id, id_max = 0;
291 for(i = 0; i < s->nb_snapshots; i++) {
292 sn = s->snapshots + i;
293 id = strtoul(sn->id_str, NULL, 10);
294 if (id > id_max)
295 id_max = id;
297 snprintf(id_str, id_str_size, "%d", id_max + 1);
300 static int find_snapshot_by_id(BlockDriverState *bs, const char *id_str)
302 BDRVQcowState *s = bs->opaque;
303 int i;
305 for(i = 0; i < s->nb_snapshots; i++) {
306 if (!strcmp(s->snapshots[i].id_str, id_str))
307 return i;
309 return -1;
312 static int find_snapshot_by_id_or_name(BlockDriverState *bs, const char *name)
314 BDRVQcowState *s = bs->opaque;
315 int i, ret;
317 ret = find_snapshot_by_id(bs, name);
318 if (ret >= 0)
319 return ret;
320 for(i = 0; i < s->nb_snapshots; i++) {
321 if (!strcmp(s->snapshots[i].name, name))
322 return i;
324 return -1;
327 /* if no id is provided, a new one is constructed */
328 int qcow2_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info)
330 BDRVQcowState *s = bs->opaque;
331 QCowSnapshot *new_snapshot_list = NULL;
332 QCowSnapshot *old_snapshot_list = NULL;
333 QCowSnapshot sn1, *sn = &sn1;
334 int i, ret;
335 uint64_t *l1_table = NULL;
336 int64_t l1_table_offset;
338 memset(sn, 0, sizeof(*sn));
340 /* Generate an ID if it wasn't passed */
341 if (sn_info->id_str[0] == '\0') {
342 find_new_snapshot_id(bs, sn_info->id_str, sizeof(sn_info->id_str));
345 /* Check that the ID is unique */
346 if (find_snapshot_by_id(bs, sn_info->id_str) >= 0) {
347 return -EEXIST;
350 /* Populate sn with passed data */
351 sn->id_str = g_strdup(sn_info->id_str);
352 sn->name = g_strdup(sn_info->name);
354 sn->disk_size = bs->total_sectors * BDRV_SECTOR_SIZE;
355 sn->vm_state_size = sn_info->vm_state_size;
356 sn->date_sec = sn_info->date_sec;
357 sn->date_nsec = sn_info->date_nsec;
358 sn->vm_clock_nsec = sn_info->vm_clock_nsec;
360 /* Allocate the L1 table of the snapshot and copy the current one there. */
361 l1_table_offset = qcow2_alloc_clusters(bs, s->l1_size * sizeof(uint64_t));
362 if (l1_table_offset < 0) {
363 ret = l1_table_offset;
364 goto fail;
367 sn->l1_table_offset = l1_table_offset;
368 sn->l1_size = s->l1_size;
370 l1_table = g_malloc(s->l1_size * sizeof(uint64_t));
371 for(i = 0; i < s->l1_size; i++) {
372 l1_table[i] = cpu_to_be64(s->l1_table[i]);
375 ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_DEFAULT,
376 sn->l1_table_offset, s->l1_size * sizeof(uint64_t));
377 if (ret < 0) {
378 goto fail;
381 ret = bdrv_pwrite(bs->file, sn->l1_table_offset, l1_table,
382 s->l1_size * sizeof(uint64_t));
383 if (ret < 0) {
384 goto fail;
387 g_free(l1_table);
388 l1_table = NULL;
391 * Increase the refcounts of all clusters and make sure everything is
392 * stable on disk before updating the snapshot table to contain a pointer
393 * to the new L1 table.
395 ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1);
396 if (ret < 0) {
397 goto fail;
400 /* Append the new snapshot to the snapshot list */
401 new_snapshot_list = g_malloc((s->nb_snapshots + 1) * sizeof(QCowSnapshot));
402 if (s->snapshots) {
403 memcpy(new_snapshot_list, s->snapshots,
404 s->nb_snapshots * sizeof(QCowSnapshot));
405 old_snapshot_list = s->snapshots;
407 s->snapshots = new_snapshot_list;
408 s->snapshots[s->nb_snapshots++] = *sn;
410 ret = qcow2_write_snapshots(bs);
411 if (ret < 0) {
412 g_free(s->snapshots);
413 s->snapshots = old_snapshot_list;
414 goto fail;
417 g_free(old_snapshot_list);
419 /* The VM state isn't needed any more in the active L1 table; in fact, it
420 * hurts by causing expensive COW for the next snapshot. */
421 qcow2_discard_clusters(bs, qcow2_vm_state_offset(s),
422 align_offset(sn->vm_state_size, s->cluster_size)
423 >> BDRV_SECTOR_BITS,
424 QCOW2_DISCARD_NEVER);
426 #ifdef DEBUG_ALLOC
428 BdrvCheckResult result = {0};
429 qcow2_check_refcounts(bs, &result, 0);
431 #endif
432 return 0;
434 fail:
435 g_free(sn->id_str);
436 g_free(sn->name);
437 g_free(l1_table);
439 return ret;
442 /* copy the snapshot 'snapshot_name' into the current disk image */
443 int qcow2_snapshot_goto(BlockDriverState *bs, const char *snapshot_id)
445 BDRVQcowState *s = bs->opaque;
446 QCowSnapshot *sn;
447 int i, snapshot_index;
448 int cur_l1_bytes, sn_l1_bytes;
449 int ret;
450 uint64_t *sn_l1_table = NULL;
452 /* Search the snapshot */
453 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
454 if (snapshot_index < 0) {
455 return -ENOENT;
457 sn = &s->snapshots[snapshot_index];
459 if (sn->disk_size != bs->total_sectors * BDRV_SECTOR_SIZE) {
460 error_report("qcow2: Loading snapshots with different disk "
461 "size is not implemented");
462 ret = -ENOTSUP;
463 goto fail;
467 * Make sure that the current L1 table is big enough to contain the whole
468 * L1 table of the snapshot. If the snapshot L1 table is smaller, the
469 * current one must be padded with zeros.
471 ret = qcow2_grow_l1_table(bs, sn->l1_size, true);
472 if (ret < 0) {
473 goto fail;
476 cur_l1_bytes = s->l1_size * sizeof(uint64_t);
477 sn_l1_bytes = sn->l1_size * sizeof(uint64_t);
480 * Copy the snapshot L1 table to the current L1 table.
482 * Before overwriting the old current L1 table on disk, make sure to
483 * increase all refcounts for the clusters referenced by the new one.
484 * Decrease the refcount referenced by the old one only when the L1
485 * table is overwritten.
487 sn_l1_table = g_malloc0(cur_l1_bytes);
489 ret = bdrv_pread(bs->file, sn->l1_table_offset, sn_l1_table, sn_l1_bytes);
490 if (ret < 0) {
491 goto fail;
494 ret = qcow2_update_snapshot_refcount(bs, sn->l1_table_offset,
495 sn->l1_size, 1);
496 if (ret < 0) {
497 goto fail;
500 ret = qcow2_pre_write_overlap_check(bs,
501 QCOW2_OL_DEFAULT & ~QCOW2_OL_ACTIVE_L1,
502 s->l1_table_offset, cur_l1_bytes);
503 if (ret < 0) {
504 goto fail;
507 ret = bdrv_pwrite_sync(bs->file, s->l1_table_offset, sn_l1_table,
508 cur_l1_bytes);
509 if (ret < 0) {
510 goto fail;
514 * Decrease refcount of clusters of current L1 table.
516 * At this point, the in-memory s->l1_table points to the old L1 table,
517 * whereas on disk we already have the new one.
519 * qcow2_update_snapshot_refcount special cases the current L1 table to use
520 * the in-memory data instead of really using the offset to load a new one,
521 * which is why this works.
523 ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset,
524 s->l1_size, -1);
527 * Now update the in-memory L1 table to be in sync with the on-disk one. We
528 * need to do this even if updating refcounts failed.
530 for(i = 0;i < s->l1_size; i++) {
531 s->l1_table[i] = be64_to_cpu(sn_l1_table[i]);
534 if (ret < 0) {
535 goto fail;
538 g_free(sn_l1_table);
539 sn_l1_table = NULL;
542 * Update QCOW_OFLAG_COPIED in the active L1 table (it may have changed
543 * when we decreased the refcount of the old snapshot.
545 ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
546 if (ret < 0) {
547 goto fail;
550 #ifdef DEBUG_ALLOC
552 BdrvCheckResult result = {0};
553 qcow2_check_refcounts(bs, &result, 0);
555 #endif
556 return 0;
558 fail:
559 g_free(sn_l1_table);
560 return ret;
563 int qcow2_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
565 BDRVQcowState *s = bs->opaque;
566 QCowSnapshot sn;
567 int snapshot_index, ret;
569 /* Search the snapshot */
570 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
571 if (snapshot_index < 0) {
572 return -ENOENT;
574 sn = s->snapshots[snapshot_index];
576 /* Remove it from the snapshot list */
577 memmove(s->snapshots + snapshot_index,
578 s->snapshots + snapshot_index + 1,
579 (s->nb_snapshots - snapshot_index - 1) * sizeof(sn));
580 s->nb_snapshots--;
581 ret = qcow2_write_snapshots(bs);
582 if (ret < 0) {
583 return ret;
587 * The snapshot is now unused, clean up. If we fail after this point, we
588 * won't recover but just leak clusters.
590 g_free(sn.id_str);
591 g_free(sn.name);
594 * Now decrease the refcounts of clusters referenced by the snapshot and
595 * free the L1 table.
597 ret = qcow2_update_snapshot_refcount(bs, sn.l1_table_offset,
598 sn.l1_size, -1);
599 if (ret < 0) {
600 return ret;
602 qcow2_free_clusters(bs, sn.l1_table_offset, sn.l1_size * sizeof(uint64_t),
603 QCOW2_DISCARD_SNAPSHOT);
605 /* must update the copied flag on the current cluster offsets */
606 ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
607 if (ret < 0) {
608 return ret;
611 #ifdef DEBUG_ALLOC
613 BdrvCheckResult result = {0};
614 qcow2_check_refcounts(bs, &result, 0);
616 #endif
617 return 0;
620 int qcow2_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab)
622 BDRVQcowState *s = bs->opaque;
623 QEMUSnapshotInfo *sn_tab, *sn_info;
624 QCowSnapshot *sn;
625 int i;
627 if (!s->nb_snapshots) {
628 *psn_tab = NULL;
629 return s->nb_snapshots;
632 sn_tab = g_malloc0(s->nb_snapshots * sizeof(QEMUSnapshotInfo));
633 for(i = 0; i < s->nb_snapshots; i++) {
634 sn_info = sn_tab + i;
635 sn = s->snapshots + i;
636 pstrcpy(sn_info->id_str, sizeof(sn_info->id_str),
637 sn->id_str);
638 pstrcpy(sn_info->name, sizeof(sn_info->name),
639 sn->name);
640 sn_info->vm_state_size = sn->vm_state_size;
641 sn_info->date_sec = sn->date_sec;
642 sn_info->date_nsec = sn->date_nsec;
643 sn_info->vm_clock_nsec = sn->vm_clock_nsec;
645 *psn_tab = sn_tab;
646 return s->nb_snapshots;
649 int qcow2_snapshot_load_tmp(BlockDriverState *bs, const char *snapshot_name)
651 int i, snapshot_index;
652 BDRVQcowState *s = bs->opaque;
653 QCowSnapshot *sn;
654 uint64_t *new_l1_table;
655 int new_l1_bytes;
656 int ret;
658 assert(bs->read_only);
660 /* Search the snapshot */
661 snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_name);
662 if (snapshot_index < 0) {
663 return -ENOENT;
665 sn = &s->snapshots[snapshot_index];
667 /* Allocate and read in the snapshot's L1 table */
668 new_l1_bytes = s->l1_size * sizeof(uint64_t);
669 new_l1_table = g_malloc0(align_offset(new_l1_bytes, 512));
671 ret = bdrv_pread(bs->file, sn->l1_table_offset, new_l1_table, new_l1_bytes);
672 if (ret < 0) {
673 g_free(new_l1_table);
674 return ret;
677 /* Switch the L1 table */
678 g_free(s->l1_table);
680 s->l1_size = sn->l1_size;
681 s->l1_table_offset = sn->l1_table_offset;
682 s->l1_table = new_l1_table;
684 for(i = 0;i < s->l1_size; i++) {
685 be64_to_cpus(&s->l1_table[i]);
688 return 0;