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/osdep.h"
26 #include "qapi/error.h"
27 #include "block/block_int.h"
28 #include "block/qcow2.h"
29 #include "qemu/error-report.h"
30 #include "qemu/cutils.h"
32 void qcow2_free_snapshots(BlockDriverState
*bs
)
34 BDRVQcow2State
*s
= bs
->opaque
;
37 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
38 g_free(s
->snapshots
[i
].name
);
39 g_free(s
->snapshots
[i
].id_str
);
46 int qcow2_read_snapshots(BlockDriverState
*bs
)
48 BDRVQcow2State
*s
= bs
->opaque
;
50 QCowSnapshotExtraData extra
;
52 int i
, id_str_size
, name_size
;
54 uint32_t extra_data_size
;
57 if (!s
->nb_snapshots
) {
59 s
->snapshots_size
= 0;
63 offset
= s
->snapshots_offset
;
64 s
->snapshots
= g_new0(QCowSnapshot
, s
->nb_snapshots
);
66 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
67 /* Read statically sized part of the snapshot header */
68 offset
= align_offset(offset
, 8);
69 ret
= bdrv_pread(bs
->file
->bs
, offset
, &h
, sizeof(h
));
75 sn
= s
->snapshots
+ i
;
76 sn
->l1_table_offset
= be64_to_cpu(h
.l1_table_offset
);
77 sn
->l1_size
= be32_to_cpu(h
.l1_size
);
78 sn
->vm_state_size
= be32_to_cpu(h
.vm_state_size
);
79 sn
->date_sec
= be32_to_cpu(h
.date_sec
);
80 sn
->date_nsec
= be32_to_cpu(h
.date_nsec
);
81 sn
->vm_clock_nsec
= be64_to_cpu(h
.vm_clock_nsec
);
82 extra_data_size
= be32_to_cpu(h
.extra_data_size
);
84 id_str_size
= be16_to_cpu(h
.id_str_size
);
85 name_size
= be16_to_cpu(h
.name_size
);
88 ret
= bdrv_pread(bs
->file
->bs
, offset
, &extra
,
89 MIN(sizeof(extra
), extra_data_size
));
93 offset
+= extra_data_size
;
95 if (extra_data_size
>= 8) {
96 sn
->vm_state_size
= be64_to_cpu(extra
.vm_state_size_large
);
99 if (extra_data_size
>= 16) {
100 sn
->disk_size
= be64_to_cpu(extra
.disk_size
);
102 sn
->disk_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
105 /* Read snapshot ID */
106 sn
->id_str
= g_malloc(id_str_size
+ 1);
107 ret
= bdrv_pread(bs
->file
->bs
, offset
, sn
->id_str
, id_str_size
);
111 offset
+= id_str_size
;
112 sn
->id_str
[id_str_size
] = '\0';
114 /* Read snapshot name */
115 sn
->name
= g_malloc(name_size
+ 1);
116 ret
= bdrv_pread(bs
->file
->bs
, offset
, sn
->name
, name_size
);
121 sn
->name
[name_size
] = '\0';
123 if (offset
- s
->snapshots_offset
> QCOW_MAX_SNAPSHOTS_SIZE
) {
129 assert(offset
- s
->snapshots_offset
<= INT_MAX
);
130 s
->snapshots_size
= offset
- s
->snapshots_offset
;
134 qcow2_free_snapshots(bs
);
138 /* add at the end of the file a new list of snapshots */
139 static int qcow2_write_snapshots(BlockDriverState
*bs
)
141 BDRVQcow2State
*s
= bs
->opaque
;
143 QCowSnapshotHeader h
;
144 QCowSnapshotExtraData extra
;
145 int i
, name_size
, id_str_size
, snapshots_size
;
147 uint32_t nb_snapshots
;
148 uint64_t snapshots_offset
;
149 } QEMU_PACKED header_data
;
150 int64_t offset
, snapshots_offset
= 0;
153 /* compute the size of the snapshots */
155 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
156 sn
= s
->snapshots
+ i
;
157 offset
= align_offset(offset
, 8);
159 offset
+= sizeof(extra
);
160 offset
+= strlen(sn
->id_str
);
161 offset
+= strlen(sn
->name
);
163 if (offset
> QCOW_MAX_SNAPSHOTS_SIZE
) {
169 assert(offset
<= INT_MAX
);
170 snapshots_size
= offset
;
172 /* Allocate space for the new snapshot list */
173 snapshots_offset
= qcow2_alloc_clusters(bs
, snapshots_size
);
174 offset
= snapshots_offset
;
179 ret
= bdrv_flush(bs
);
184 /* The snapshot list position has not yet been updated, so these clusters
185 * must indeed be completely free */
186 ret
= qcow2_pre_write_overlap_check(bs
, 0, offset
, snapshots_size
);
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
);
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
));
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
);
212 id_str_size
= strlen(sn
->id_str
);
213 name_size
= strlen(sn
->name
);
214 assert(id_str_size
<= UINT16_MAX
&& name_size
<= UINT16_MAX
);
215 h
.id_str_size
= cpu_to_be16(id_str_size
);
216 h
.name_size
= cpu_to_be16(name_size
);
217 offset
= align_offset(offset
, 8);
219 ret
= bdrv_pwrite(bs
->file
->bs
, offset
, &h
, sizeof(h
));
225 ret
= bdrv_pwrite(bs
->file
->bs
, offset
, &extra
, sizeof(extra
));
229 offset
+= sizeof(extra
);
231 ret
= bdrv_pwrite(bs
->file
->bs
, offset
, sn
->id_str
, id_str_size
);
235 offset
+= id_str_size
;
237 ret
= bdrv_pwrite(bs
->file
->bs
, offset
, sn
->name
, name_size
);
245 * Update the header to point to the new snapshot table. This requires the
246 * new table and its refcounts to be stable on disk.
248 ret
= bdrv_flush(bs
);
253 QEMU_BUILD_BUG_ON(offsetof(QCowHeader
, snapshots_offset
) !=
254 offsetof(QCowHeader
, nb_snapshots
) + sizeof(header_data
.nb_snapshots
));
256 header_data
.nb_snapshots
= cpu_to_be32(s
->nb_snapshots
);
257 header_data
.snapshots_offset
= cpu_to_be64(snapshots_offset
);
259 ret
= bdrv_pwrite_sync(bs
->file
->bs
, offsetof(QCowHeader
, nb_snapshots
),
260 &header_data
, sizeof(header_data
));
265 /* free the old snapshot table */
266 qcow2_free_clusters(bs
, s
->snapshots_offset
, s
->snapshots_size
,
267 QCOW2_DISCARD_SNAPSHOT
);
268 s
->snapshots_offset
= snapshots_offset
;
269 s
->snapshots_size
= snapshots_size
;
273 if (snapshots_offset
> 0) {
274 qcow2_free_clusters(bs
, snapshots_offset
, snapshots_size
,
275 QCOW2_DISCARD_ALWAYS
);
280 static void find_new_snapshot_id(BlockDriverState
*bs
,
281 char *id_str
, int id_str_size
)
283 BDRVQcow2State
*s
= bs
->opaque
;
286 unsigned long id
, id_max
= 0;
288 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
289 sn
= s
->snapshots
+ i
;
290 id
= strtoul(sn
->id_str
, NULL
, 10);
294 snprintf(id_str
, id_str_size
, "%lu", id_max
+ 1);
297 static int find_snapshot_by_id_and_name(BlockDriverState
*bs
,
301 BDRVQcow2State
*s
= bs
->opaque
;
305 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
306 if (!strcmp(s
->snapshots
[i
].id_str
, id
) &&
307 !strcmp(s
->snapshots
[i
].name
, name
)) {
312 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
313 if (!strcmp(s
->snapshots
[i
].id_str
, id
)) {
318 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
319 if (!strcmp(s
->snapshots
[i
].name
, name
)) {
328 static int find_snapshot_by_id_or_name(BlockDriverState
*bs
,
329 const char *id_or_name
)
333 ret
= find_snapshot_by_id_and_name(bs
, id_or_name
, NULL
);
337 return find_snapshot_by_id_and_name(bs
, NULL
, id_or_name
);
340 /* if no id is provided, a new one is constructed */
341 int qcow2_snapshot_create(BlockDriverState
*bs
, QEMUSnapshotInfo
*sn_info
)
343 BDRVQcow2State
*s
= bs
->opaque
;
344 QCowSnapshot
*new_snapshot_list
= NULL
;
345 QCowSnapshot
*old_snapshot_list
= NULL
;
346 QCowSnapshot sn1
, *sn
= &sn1
;
348 uint64_t *l1_table
= NULL
;
349 int64_t l1_table_offset
;
351 if (s
->nb_snapshots
>= QCOW_MAX_SNAPSHOTS
) {
355 memset(sn
, 0, sizeof(*sn
));
358 find_new_snapshot_id(bs
, sn_info
->id_str
, sizeof(sn_info
->id_str
));
360 /* Check that the ID is unique */
361 if (find_snapshot_by_id_and_name(bs
, sn_info
->id_str
, NULL
) >= 0) {
365 /* Populate sn with passed data */
366 sn
->id_str
= g_strdup(sn_info
->id_str
);
367 sn
->name
= g_strdup(sn_info
->name
);
369 sn
->disk_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
370 sn
->vm_state_size
= sn_info
->vm_state_size
;
371 sn
->date_sec
= sn_info
->date_sec
;
372 sn
->date_nsec
= sn_info
->date_nsec
;
373 sn
->vm_clock_nsec
= sn_info
->vm_clock_nsec
;
375 /* Allocate the L1 table of the snapshot and copy the current one there. */
376 l1_table_offset
= qcow2_alloc_clusters(bs
, s
->l1_size
* sizeof(uint64_t));
377 if (l1_table_offset
< 0) {
378 ret
= l1_table_offset
;
382 sn
->l1_table_offset
= l1_table_offset
;
383 sn
->l1_size
= s
->l1_size
;
385 l1_table
= g_try_new(uint64_t, s
->l1_size
);
386 if (s
->l1_size
&& l1_table
== NULL
) {
391 for(i
= 0; i
< s
->l1_size
; i
++) {
392 l1_table
[i
] = cpu_to_be64(s
->l1_table
[i
]);
395 ret
= qcow2_pre_write_overlap_check(bs
, 0, sn
->l1_table_offset
,
396 s
->l1_size
* sizeof(uint64_t));
401 ret
= bdrv_pwrite(bs
->file
->bs
, sn
->l1_table_offset
, l1_table
,
402 s
->l1_size
* sizeof(uint64_t));
411 * Increase the refcounts of all clusters and make sure everything is
412 * stable on disk before updating the snapshot table to contain a pointer
413 * to the new L1 table.
415 ret
= qcow2_update_snapshot_refcount(bs
, s
->l1_table_offset
, s
->l1_size
, 1);
420 /* Append the new snapshot to the snapshot list */
421 new_snapshot_list
= g_new(QCowSnapshot
, s
->nb_snapshots
+ 1);
423 memcpy(new_snapshot_list
, s
->snapshots
,
424 s
->nb_snapshots
* sizeof(QCowSnapshot
));
425 old_snapshot_list
= s
->snapshots
;
427 s
->snapshots
= new_snapshot_list
;
428 s
->snapshots
[s
->nb_snapshots
++] = *sn
;
430 ret
= qcow2_write_snapshots(bs
);
432 g_free(s
->snapshots
);
433 s
->snapshots
= old_snapshot_list
;
438 g_free(old_snapshot_list
);
440 /* The VM state isn't needed any more in the active L1 table; in fact, it
441 * hurts by causing expensive COW for the next snapshot. */
442 qcow2_discard_clusters(bs
, qcow2_vm_state_offset(s
),
443 align_offset(sn
->vm_state_size
, s
->cluster_size
)
445 QCOW2_DISCARD_NEVER
, false);
449 BdrvCheckResult result
= {0};
450 qcow2_check_refcounts(bs
, &result
, 0);
463 /* copy the snapshot 'snapshot_name' into the current disk image */
464 int qcow2_snapshot_goto(BlockDriverState
*bs
, const char *snapshot_id
)
466 BDRVQcow2State
*s
= bs
->opaque
;
468 int i
, snapshot_index
;
469 int cur_l1_bytes
, sn_l1_bytes
;
471 uint64_t *sn_l1_table
= NULL
;
473 /* Search the snapshot */
474 snapshot_index
= find_snapshot_by_id_or_name(bs
, snapshot_id
);
475 if (snapshot_index
< 0) {
478 sn
= &s
->snapshots
[snapshot_index
];
480 if (sn
->disk_size
!= bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
481 error_report("qcow2: Loading snapshots with different disk "
482 "size is not implemented");
488 * Make sure that the current L1 table is big enough to contain the whole
489 * L1 table of the snapshot. If the snapshot L1 table is smaller, the
490 * current one must be padded with zeros.
492 ret
= qcow2_grow_l1_table(bs
, sn
->l1_size
, true);
497 cur_l1_bytes
= s
->l1_size
* sizeof(uint64_t);
498 sn_l1_bytes
= sn
->l1_size
* sizeof(uint64_t);
501 * Copy the snapshot L1 table to the current L1 table.
503 * Before overwriting the old current L1 table on disk, make sure to
504 * increase all refcounts for the clusters referenced by the new one.
505 * Decrease the refcount referenced by the old one only when the L1
506 * table is overwritten.
508 sn_l1_table
= g_try_malloc0(cur_l1_bytes
);
509 if (cur_l1_bytes
&& sn_l1_table
== NULL
) {
514 ret
= bdrv_pread(bs
->file
->bs
, sn
->l1_table_offset
,
515 sn_l1_table
, sn_l1_bytes
);
520 ret
= qcow2_update_snapshot_refcount(bs
, sn
->l1_table_offset
,
526 ret
= qcow2_pre_write_overlap_check(bs
, QCOW2_OL_ACTIVE_L1
,
527 s
->l1_table_offset
, cur_l1_bytes
);
532 ret
= bdrv_pwrite_sync(bs
->file
->bs
, s
->l1_table_offset
, sn_l1_table
,
539 * Decrease refcount of clusters of current L1 table.
541 * At this point, the in-memory s->l1_table points to the old L1 table,
542 * whereas on disk we already have the new one.
544 * qcow2_update_snapshot_refcount special cases the current L1 table to use
545 * the in-memory data instead of really using the offset to load a new one,
546 * which is why this works.
548 ret
= qcow2_update_snapshot_refcount(bs
, s
->l1_table_offset
,
552 * Now update the in-memory L1 table to be in sync with the on-disk one. We
553 * need to do this even if updating refcounts failed.
555 for(i
= 0;i
< s
->l1_size
; i
++) {
556 s
->l1_table
[i
] = be64_to_cpu(sn_l1_table
[i
]);
567 * Update QCOW_OFLAG_COPIED in the active L1 table (it may have changed
568 * when we decreased the refcount of the old snapshot.
570 ret
= qcow2_update_snapshot_refcount(bs
, s
->l1_table_offset
, s
->l1_size
, 0);
577 BdrvCheckResult result
= {0};
578 qcow2_check_refcounts(bs
, &result
, 0);
588 int qcow2_snapshot_delete(BlockDriverState
*bs
,
589 const char *snapshot_id
,
593 BDRVQcow2State
*s
= bs
->opaque
;
595 int snapshot_index
, ret
;
597 /* Search the snapshot */
598 snapshot_index
= find_snapshot_by_id_and_name(bs
, snapshot_id
, name
);
599 if (snapshot_index
< 0) {
600 error_setg(errp
, "Can't find the snapshot");
603 sn
= s
->snapshots
[snapshot_index
];
605 /* Remove it from the snapshot list */
606 memmove(s
->snapshots
+ snapshot_index
,
607 s
->snapshots
+ snapshot_index
+ 1,
608 (s
->nb_snapshots
- snapshot_index
- 1) * sizeof(sn
));
610 ret
= qcow2_write_snapshots(bs
);
612 error_setg_errno(errp
, -ret
,
613 "Failed to remove snapshot from snapshot list");
618 * The snapshot is now unused, clean up. If we fail after this point, we
619 * won't recover but just leak clusters.
625 * Now decrease the refcounts of clusters referenced by the snapshot and
628 ret
= qcow2_update_snapshot_refcount(bs
, sn
.l1_table_offset
,
631 error_setg_errno(errp
, -ret
, "Failed to free the cluster and L1 table");
634 qcow2_free_clusters(bs
, sn
.l1_table_offset
, sn
.l1_size
* sizeof(uint64_t),
635 QCOW2_DISCARD_SNAPSHOT
);
637 /* must update the copied flag on the current cluster offsets */
638 ret
= qcow2_update_snapshot_refcount(bs
, s
->l1_table_offset
, s
->l1_size
, 0);
640 error_setg_errno(errp
, -ret
,
641 "Failed to update snapshot status in disk");
647 BdrvCheckResult result
= {0};
648 qcow2_check_refcounts(bs
, &result
, 0);
654 int qcow2_snapshot_list(BlockDriverState
*bs
, QEMUSnapshotInfo
**psn_tab
)
656 BDRVQcow2State
*s
= bs
->opaque
;
657 QEMUSnapshotInfo
*sn_tab
, *sn_info
;
661 if (!s
->nb_snapshots
) {
663 return s
->nb_snapshots
;
666 sn_tab
= g_new0(QEMUSnapshotInfo
, s
->nb_snapshots
);
667 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
668 sn_info
= sn_tab
+ i
;
669 sn
= s
->snapshots
+ i
;
670 pstrcpy(sn_info
->id_str
, sizeof(sn_info
->id_str
),
672 pstrcpy(sn_info
->name
, sizeof(sn_info
->name
),
674 sn_info
->vm_state_size
= sn
->vm_state_size
;
675 sn_info
->date_sec
= sn
->date_sec
;
676 sn_info
->date_nsec
= sn
->date_nsec
;
677 sn_info
->vm_clock_nsec
= sn
->vm_clock_nsec
;
680 return s
->nb_snapshots
;
683 int qcow2_snapshot_load_tmp(BlockDriverState
*bs
,
684 const char *snapshot_id
,
688 int i
, snapshot_index
;
689 BDRVQcow2State
*s
= bs
->opaque
;
691 uint64_t *new_l1_table
;
695 assert(bs
->read_only
);
697 /* Search the snapshot */
698 snapshot_index
= find_snapshot_by_id_and_name(bs
, snapshot_id
, name
);
699 if (snapshot_index
< 0) {
701 "Can't find snapshot");
704 sn
= &s
->snapshots
[snapshot_index
];
706 /* Allocate and read in the snapshot's L1 table */
707 if (sn
->l1_size
> QCOW_MAX_L1_SIZE
/ sizeof(uint64_t)) {
708 error_setg(errp
, "Snapshot L1 table too large");
711 new_l1_bytes
= sn
->l1_size
* sizeof(uint64_t);
712 new_l1_table
= qemu_try_blockalign(bs
->file
->bs
,
713 align_offset(new_l1_bytes
, 512));
714 if (new_l1_table
== NULL
) {
718 ret
= bdrv_pread(bs
->file
->bs
, sn
->l1_table_offset
,
719 new_l1_table
, new_l1_bytes
);
721 error_setg(errp
, "Failed to read l1 table for snapshot");
722 qemu_vfree(new_l1_table
);
726 /* Switch the L1 table */
727 qemu_vfree(s
->l1_table
);
729 s
->l1_size
= sn
->l1_size
;
730 s
->l1_table_offset
= sn
->l1_table_offset
;
731 s
->l1_table
= new_l1_table
;
733 for(i
= 0;i
< s
->l1_size
; i
++) {
734 be64_to_cpus(&s
->l1_table
[i
]);