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
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
;
40 uint64_t vm_clock_nsec
;
42 uint32_t vm_state_size
;
43 uint32_t extra_data_size
; /* for extension */
44 /* extra data follows */
49 typedef struct QEMU_PACKED QCowSnapshotExtraData
{
50 uint64_t vm_state_size_large
;
52 } QCowSnapshotExtraData
;
54 void qcow2_free_snapshots(BlockDriverState
*bs
)
56 BDRVQcowState
*s
= bs
->opaque
;
59 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
60 g_free(s
->snapshots
[i
].name
);
61 g_free(s
->snapshots
[i
].id_str
);
68 int qcow2_read_snapshots(BlockDriverState
*bs
)
70 BDRVQcowState
*s
= bs
->opaque
;
72 QCowSnapshotExtraData extra
;
74 int i
, id_str_size
, name_size
;
76 uint32_t extra_data_size
;
79 if (!s
->nb_snapshots
) {
81 s
->snapshots_size
= 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
));
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
));
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
);
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
);
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
);
143 sn
->name
[name_size
] = '\0';
146 s
->snapshots_size
= offset
- s
->snapshots_offset
;
150 qcow2_free_snapshots(bs
);
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
;
159 QCowSnapshotHeader h
;
160 QCowSnapshotExtraData extra
;
161 int i
, name_size
, id_str_size
, snapshots_size
;
163 uint32_t nb_snapshots
;
164 uint64_t snapshots_offset
;
165 } QEMU_PACKED header_data
;
166 int64_t offset
, snapshots_offset
;
169 /* compute the size of the snapshots */
171 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
172 sn
= s
->snapshots
+ i
;
173 offset
= align_offset(offset
, 8);
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
;
187 ret
= bdrv_flush(bs
);
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
,
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
));
233 ret
= bdrv_pwrite(bs
->file
, offset
, &extra
, sizeof(extra
));
237 offset
+= sizeof(extra
);
239 ret
= bdrv_pwrite(bs
->file
, offset
, sn
->id_str
, id_str_size
);
243 offset
+= id_str_size
;
245 ret
= bdrv_pwrite(bs
->file
, offset
, sn
->name
, 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
);
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
));
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
;
284 static void find_new_snapshot_id(BlockDriverState
*bs
,
285 char *id_str
, int id_str_size
)
287 BDRVQcowState
*s
= bs
->opaque
;
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);
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
;
305 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
306 if (!strcmp(s
->snapshots
[i
].id_str
, id_str
))
312 static int find_snapshot_by_id_or_name(BlockDriverState
*bs
, const char *name
)
314 BDRVQcowState
*s
= bs
->opaque
;
317 ret
= find_snapshot_by_id(bs
, name
);
320 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
321 if (!strcmp(s
->snapshots
[i
].name
, name
))
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
;
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) {
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
;
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));
381 ret
= bdrv_pwrite(bs
->file
, sn
->l1_table_offset
, l1_table
,
382 s
->l1_size
* sizeof(uint64_t));
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);
400 /* Append the new snapshot to the snapshot list */
401 new_snapshot_list
= g_malloc((s
->nb_snapshots
+ 1) * sizeof(QCowSnapshot
));
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
);
412 g_free(s
->snapshots
);
413 s
->snapshots
= old_snapshot_list
;
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
)
424 QCOW2_DISCARD_NEVER
);
428 BdrvCheckResult result
= {0};
429 qcow2_check_refcounts(bs
, &result
, 0);
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
;
447 int i
, snapshot_index
;
448 int cur_l1_bytes
, sn_l1_bytes
;
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) {
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");
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);
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
);
494 ret
= qcow2_update_snapshot_refcount(bs
, sn
->l1_table_offset
,
500 ret
= qcow2_pre_write_overlap_check(bs
,
501 QCOW2_OL_DEFAULT
& ~QCOW2_OL_ACTIVE_L1
,
502 s
->l1_table_offset
, cur_l1_bytes
);
507 ret
= bdrv_pwrite_sync(bs
->file
, s
->l1_table_offset
, sn_l1_table
,
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
,
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
]);
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);
552 BdrvCheckResult result
= {0};
553 qcow2_check_refcounts(bs
, &result
, 0);
563 int qcow2_snapshot_delete(BlockDriverState
*bs
, const char *snapshot_id
)
565 BDRVQcowState
*s
= bs
->opaque
;
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) {
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
));
581 ret
= qcow2_write_snapshots(bs
);
587 * The snapshot is now unused, clean up. If we fail after this point, we
588 * won't recover but just leak clusters.
594 * Now decrease the refcounts of clusters referenced by the snapshot and
597 ret
= qcow2_update_snapshot_refcount(bs
, sn
.l1_table_offset
,
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);
613 BdrvCheckResult result
= {0};
614 qcow2_check_refcounts(bs
, &result
, 0);
620 int qcow2_snapshot_list(BlockDriverState
*bs
, QEMUSnapshotInfo
**psn_tab
)
622 BDRVQcowState
*s
= bs
->opaque
;
623 QEMUSnapshotInfo
*sn_tab
, *sn_info
;
627 if (!s
->nb_snapshots
) {
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
),
638 pstrcpy(sn_info
->name
, sizeof(sn_info
->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
;
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
;
654 uint64_t *new_l1_table
;
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) {
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
);
673 g_free(new_l1_table
);
677 /* Switch the 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
]);