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 void qcow2_free_snapshots(BlockDriverState
*bs
)
31 BDRVQcowState
*s
= bs
->opaque
;
34 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
35 g_free(s
->snapshots
[i
].name
);
36 g_free(s
->snapshots
[i
].id_str
);
43 int qcow2_read_snapshots(BlockDriverState
*bs
)
45 BDRVQcowState
*s
= bs
->opaque
;
47 QCowSnapshotExtraData extra
;
49 int i
, id_str_size
, name_size
;
51 uint32_t extra_data_size
;
54 if (!s
->nb_snapshots
) {
56 s
->snapshots_size
= 0;
60 offset
= s
->snapshots_offset
;
61 s
->snapshots
= g_malloc0(s
->nb_snapshots
* sizeof(QCowSnapshot
));
63 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
64 /* Read statically sized part of the snapshot header */
65 offset
= align_offset(offset
, 8);
66 ret
= bdrv_pread(bs
->file
, offset
, &h
, sizeof(h
));
72 sn
= s
->snapshots
+ i
;
73 sn
->l1_table_offset
= be64_to_cpu(h
.l1_table_offset
);
74 sn
->l1_size
= be32_to_cpu(h
.l1_size
);
75 sn
->vm_state_size
= be32_to_cpu(h
.vm_state_size
);
76 sn
->date_sec
= be32_to_cpu(h
.date_sec
);
77 sn
->date_nsec
= be32_to_cpu(h
.date_nsec
);
78 sn
->vm_clock_nsec
= be64_to_cpu(h
.vm_clock_nsec
);
79 extra_data_size
= be32_to_cpu(h
.extra_data_size
);
81 id_str_size
= be16_to_cpu(h
.id_str_size
);
82 name_size
= be16_to_cpu(h
.name_size
);
85 ret
= bdrv_pread(bs
->file
, offset
, &extra
,
86 MIN(sizeof(extra
), extra_data_size
));
90 offset
+= extra_data_size
;
92 if (extra_data_size
>= 8) {
93 sn
->vm_state_size
= be64_to_cpu(extra
.vm_state_size_large
);
96 if (extra_data_size
>= 16) {
97 sn
->disk_size
= be64_to_cpu(extra
.disk_size
);
99 sn
->disk_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
102 /* Read snapshot ID */
103 sn
->id_str
= g_malloc(id_str_size
+ 1);
104 ret
= bdrv_pread(bs
->file
, offset
, sn
->id_str
, id_str_size
);
108 offset
+= id_str_size
;
109 sn
->id_str
[id_str_size
] = '\0';
111 /* Read snapshot name */
112 sn
->name
= g_malloc(name_size
+ 1);
113 ret
= bdrv_pread(bs
->file
, offset
, sn
->name
, name_size
);
118 sn
->name
[name_size
] = '\0';
120 if (offset
- s
->snapshots_offset
> QCOW_MAX_SNAPSHOTS_SIZE
) {
126 assert(offset
- s
->snapshots_offset
<= INT_MAX
);
127 s
->snapshots_size
= offset
- s
->snapshots_offset
;
131 qcow2_free_snapshots(bs
);
135 /* add at the end of the file a new list of snapshots */
136 static int qcow2_write_snapshots(BlockDriverState
*bs
)
138 BDRVQcowState
*s
= bs
->opaque
;
140 QCowSnapshotHeader h
;
141 QCowSnapshotExtraData extra
;
142 int i
, name_size
, id_str_size
, snapshots_size
;
144 uint32_t nb_snapshots
;
145 uint64_t snapshots_offset
;
146 } QEMU_PACKED header_data
;
147 int64_t offset
, snapshots_offset
= 0;
150 /* compute the size of the snapshots */
152 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
153 sn
= s
->snapshots
+ i
;
154 offset
= align_offset(offset
, 8);
156 offset
+= sizeof(extra
);
157 offset
+= strlen(sn
->id_str
);
158 offset
+= strlen(sn
->name
);
160 if (offset
> QCOW_MAX_SNAPSHOTS_SIZE
) {
166 assert(offset
<= INT_MAX
);
167 snapshots_size
= offset
;
169 /* Allocate space for the new snapshot list */
170 snapshots_offset
= qcow2_alloc_clusters(bs
, snapshots_size
);
171 offset
= snapshots_offset
;
176 ret
= bdrv_flush(bs
);
181 /* The snapshot list position has not yet been updated, so these clusters
182 * must indeed be completely free */
183 ret
= qcow2_pre_write_overlap_check(bs
, 0, offset
, snapshots_size
);
189 /* Write all snapshots to the new list */
190 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
191 sn
= s
->snapshots
+ i
;
192 memset(&h
, 0, sizeof(h
));
193 h
.l1_table_offset
= cpu_to_be64(sn
->l1_table_offset
);
194 h
.l1_size
= cpu_to_be32(sn
->l1_size
);
195 /* If it doesn't fit in 32 bit, older implementations should treat it
196 * as a disk-only snapshot rather than truncate the VM state */
197 if (sn
->vm_state_size
<= 0xffffffff) {
198 h
.vm_state_size
= cpu_to_be32(sn
->vm_state_size
);
200 h
.date_sec
= cpu_to_be32(sn
->date_sec
);
201 h
.date_nsec
= cpu_to_be32(sn
->date_nsec
);
202 h
.vm_clock_nsec
= cpu_to_be64(sn
->vm_clock_nsec
);
203 h
.extra_data_size
= cpu_to_be32(sizeof(extra
));
205 memset(&extra
, 0, sizeof(extra
));
206 extra
.vm_state_size_large
= cpu_to_be64(sn
->vm_state_size
);
207 extra
.disk_size
= cpu_to_be64(sn
->disk_size
);
209 id_str_size
= strlen(sn
->id_str
);
210 name_size
= strlen(sn
->name
);
211 assert(id_str_size
<= UINT16_MAX
&& name_size
<= UINT16_MAX
);
212 h
.id_str_size
= cpu_to_be16(id_str_size
);
213 h
.name_size
= cpu_to_be16(name_size
);
214 offset
= align_offset(offset
, 8);
216 ret
= bdrv_pwrite(bs
->file
, offset
, &h
, sizeof(h
));
222 ret
= bdrv_pwrite(bs
->file
, offset
, &extra
, sizeof(extra
));
226 offset
+= sizeof(extra
);
228 ret
= bdrv_pwrite(bs
->file
, offset
, sn
->id_str
, id_str_size
);
232 offset
+= id_str_size
;
234 ret
= bdrv_pwrite(bs
->file
, offset
, sn
->name
, name_size
);
242 * Update the header to point to the new snapshot table. This requires the
243 * new table and its refcounts to be stable on disk.
245 ret
= bdrv_flush(bs
);
250 QEMU_BUILD_BUG_ON(offsetof(QCowHeader
, snapshots_offset
) !=
251 offsetof(QCowHeader
, nb_snapshots
) + sizeof(header_data
.nb_snapshots
));
253 header_data
.nb_snapshots
= cpu_to_be32(s
->nb_snapshots
);
254 header_data
.snapshots_offset
= cpu_to_be64(snapshots_offset
);
256 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, nb_snapshots
),
257 &header_data
, sizeof(header_data
));
262 /* free the old snapshot table */
263 qcow2_free_clusters(bs
, s
->snapshots_offset
, s
->snapshots_size
,
264 QCOW2_DISCARD_SNAPSHOT
);
265 s
->snapshots_offset
= snapshots_offset
;
266 s
->snapshots_size
= snapshots_size
;
270 if (snapshots_offset
> 0) {
271 qcow2_free_clusters(bs
, snapshots_offset
, snapshots_size
,
272 QCOW2_DISCARD_ALWAYS
);
277 static void find_new_snapshot_id(BlockDriverState
*bs
,
278 char *id_str
, int id_str_size
)
280 BDRVQcowState
*s
= bs
->opaque
;
283 unsigned long id
, id_max
= 0;
285 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
286 sn
= s
->snapshots
+ i
;
287 id
= strtoul(sn
->id_str
, NULL
, 10);
291 snprintf(id_str
, id_str_size
, "%lu", id_max
+ 1);
294 static int find_snapshot_by_id_and_name(BlockDriverState
*bs
,
298 BDRVQcowState
*s
= bs
->opaque
;
302 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
303 if (!strcmp(s
->snapshots
[i
].id_str
, id
) &&
304 !strcmp(s
->snapshots
[i
].name
, name
)) {
309 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
310 if (!strcmp(s
->snapshots
[i
].id_str
, id
)) {
315 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
316 if (!strcmp(s
->snapshots
[i
].name
, name
)) {
325 static int find_snapshot_by_id_or_name(BlockDriverState
*bs
,
326 const char *id_or_name
)
330 ret
= find_snapshot_by_id_and_name(bs
, id_or_name
, NULL
);
334 return find_snapshot_by_id_and_name(bs
, NULL
, id_or_name
);
337 /* if no id is provided, a new one is constructed */
338 int qcow2_snapshot_create(BlockDriverState
*bs
, QEMUSnapshotInfo
*sn_info
)
340 BDRVQcowState
*s
= bs
->opaque
;
341 QCowSnapshot
*new_snapshot_list
= NULL
;
342 QCowSnapshot
*old_snapshot_list
= NULL
;
343 QCowSnapshot sn1
, *sn
= &sn1
;
345 uint64_t *l1_table
= NULL
;
346 int64_t l1_table_offset
;
348 if (s
->nb_snapshots
>= QCOW_MAX_SNAPSHOTS
) {
352 memset(sn
, 0, sizeof(*sn
));
354 /* Generate an ID if it wasn't passed */
355 if (sn_info
->id_str
[0] == '\0') {
356 find_new_snapshot_id(bs
, sn_info
->id_str
, sizeof(sn_info
->id_str
));
359 /* Check that the ID is unique */
360 if (find_snapshot_by_id_and_name(bs
, sn_info
->id_str
, NULL
) >= 0) {
364 /* Populate sn with passed data */
365 sn
->id_str
= g_strdup(sn_info
->id_str
);
366 sn
->name
= g_strdup(sn_info
->name
);
368 sn
->disk_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
369 sn
->vm_state_size
= sn_info
->vm_state_size
;
370 sn
->date_sec
= sn_info
->date_sec
;
371 sn
->date_nsec
= sn_info
->date_nsec
;
372 sn
->vm_clock_nsec
= sn_info
->vm_clock_nsec
;
374 /* Allocate the L1 table of the snapshot and copy the current one there. */
375 l1_table_offset
= qcow2_alloc_clusters(bs
, s
->l1_size
* sizeof(uint64_t));
376 if (l1_table_offset
< 0) {
377 ret
= l1_table_offset
;
381 sn
->l1_table_offset
= l1_table_offset
;
382 sn
->l1_size
= s
->l1_size
;
384 l1_table
= g_try_malloc(s
->l1_size
* sizeof(uint64_t));
385 if (s
->l1_size
&& l1_table
== NULL
) {
390 for(i
= 0; i
< s
->l1_size
; i
++) {
391 l1_table
[i
] = cpu_to_be64(s
->l1_table
[i
]);
394 ret
= qcow2_pre_write_overlap_check(bs
, 0, sn
->l1_table_offset
,
395 s
->l1_size
* sizeof(uint64_t));
400 ret
= bdrv_pwrite(bs
->file
, sn
->l1_table_offset
, l1_table
,
401 s
->l1_size
* sizeof(uint64_t));
410 * Increase the refcounts of all clusters and make sure everything is
411 * stable on disk before updating the snapshot table to contain a pointer
412 * to the new L1 table.
414 ret
= qcow2_update_snapshot_refcount(bs
, s
->l1_table_offset
, s
->l1_size
, 1);
419 /* Append the new snapshot to the snapshot list */
420 new_snapshot_list
= g_malloc((s
->nb_snapshots
+ 1) * sizeof(QCowSnapshot
));
422 memcpy(new_snapshot_list
, s
->snapshots
,
423 s
->nb_snapshots
* sizeof(QCowSnapshot
));
424 old_snapshot_list
= s
->snapshots
;
426 s
->snapshots
= new_snapshot_list
;
427 s
->snapshots
[s
->nb_snapshots
++] = *sn
;
429 ret
= qcow2_write_snapshots(bs
);
431 g_free(s
->snapshots
);
432 s
->snapshots
= old_snapshot_list
;
437 g_free(old_snapshot_list
);
439 /* The VM state isn't needed any more in the active L1 table; in fact, it
440 * hurts by causing expensive COW for the next snapshot. */
441 qcow2_discard_clusters(bs
, qcow2_vm_state_offset(s
),
442 align_offset(sn
->vm_state_size
, s
->cluster_size
)
444 QCOW2_DISCARD_NEVER
);
448 BdrvCheckResult result
= {0};
449 qcow2_check_refcounts(bs
, &result
, 0);
462 /* copy the snapshot 'snapshot_name' into the current disk image */
463 int qcow2_snapshot_goto(BlockDriverState
*bs
, const char *snapshot_id
)
465 BDRVQcowState
*s
= bs
->opaque
;
467 int i
, snapshot_index
;
468 int cur_l1_bytes
, sn_l1_bytes
;
470 uint64_t *sn_l1_table
= NULL
;
472 /* Search the snapshot */
473 snapshot_index
= find_snapshot_by_id_or_name(bs
, snapshot_id
);
474 if (snapshot_index
< 0) {
477 sn
= &s
->snapshots
[snapshot_index
];
479 if (sn
->disk_size
!= bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
480 error_report("qcow2: Loading snapshots with different disk "
481 "size is not implemented");
487 * Make sure that the current L1 table is big enough to contain the whole
488 * L1 table of the snapshot. If the snapshot L1 table is smaller, the
489 * current one must be padded with zeros.
491 ret
= qcow2_grow_l1_table(bs
, sn
->l1_size
, true);
496 cur_l1_bytes
= s
->l1_size
* sizeof(uint64_t);
497 sn_l1_bytes
= sn
->l1_size
* sizeof(uint64_t);
500 * Copy the snapshot L1 table to the current L1 table.
502 * Before overwriting the old current L1 table on disk, make sure to
503 * increase all refcounts for the clusters referenced by the new one.
504 * Decrease the refcount referenced by the old one only when the L1
505 * table is overwritten.
507 sn_l1_table
= g_try_malloc0(cur_l1_bytes
);
508 if (cur_l1_bytes
&& sn_l1_table
== NULL
) {
513 ret
= bdrv_pread(bs
->file
, sn
->l1_table_offset
, sn_l1_table
, sn_l1_bytes
);
518 ret
= qcow2_update_snapshot_refcount(bs
, sn
->l1_table_offset
,
524 ret
= qcow2_pre_write_overlap_check(bs
, QCOW2_OL_ACTIVE_L1
,
525 s
->l1_table_offset
, cur_l1_bytes
);
530 ret
= bdrv_pwrite_sync(bs
->file
, s
->l1_table_offset
, sn_l1_table
,
537 * Decrease refcount of clusters of current L1 table.
539 * At this point, the in-memory s->l1_table points to the old L1 table,
540 * whereas on disk we already have the new one.
542 * qcow2_update_snapshot_refcount special cases the current L1 table to use
543 * the in-memory data instead of really using the offset to load a new one,
544 * which is why this works.
546 ret
= qcow2_update_snapshot_refcount(bs
, s
->l1_table_offset
,
550 * Now update the in-memory L1 table to be in sync with the on-disk one. We
551 * need to do this even if updating refcounts failed.
553 for(i
= 0;i
< s
->l1_size
; i
++) {
554 s
->l1_table
[i
] = be64_to_cpu(sn_l1_table
[i
]);
565 * Update QCOW_OFLAG_COPIED in the active L1 table (it may have changed
566 * when we decreased the refcount of the old snapshot.
568 ret
= qcow2_update_snapshot_refcount(bs
, s
->l1_table_offset
, s
->l1_size
, 0);
575 BdrvCheckResult result
= {0};
576 qcow2_check_refcounts(bs
, &result
, 0);
586 int qcow2_snapshot_delete(BlockDriverState
*bs
,
587 const char *snapshot_id
,
591 BDRVQcowState
*s
= bs
->opaque
;
593 int snapshot_index
, ret
;
595 /* Search the snapshot */
596 snapshot_index
= find_snapshot_by_id_and_name(bs
, snapshot_id
, name
);
597 if (snapshot_index
< 0) {
598 error_setg(errp
, "Can't find the snapshot");
601 sn
= s
->snapshots
[snapshot_index
];
603 /* Remove it from the snapshot list */
604 memmove(s
->snapshots
+ snapshot_index
,
605 s
->snapshots
+ snapshot_index
+ 1,
606 (s
->nb_snapshots
- snapshot_index
- 1) * sizeof(sn
));
608 ret
= qcow2_write_snapshots(bs
);
610 error_setg_errno(errp
, -ret
,
611 "Failed to remove snapshot from snapshot list");
616 * The snapshot is now unused, clean up. If we fail after this point, we
617 * won't recover but just leak clusters.
623 * Now decrease the refcounts of clusters referenced by the snapshot and
626 ret
= qcow2_update_snapshot_refcount(bs
, sn
.l1_table_offset
,
629 error_setg_errno(errp
, -ret
, "Failed to free the cluster and L1 table");
632 qcow2_free_clusters(bs
, sn
.l1_table_offset
, sn
.l1_size
* sizeof(uint64_t),
633 QCOW2_DISCARD_SNAPSHOT
);
635 /* must update the copied flag on the current cluster offsets */
636 ret
= qcow2_update_snapshot_refcount(bs
, s
->l1_table_offset
, s
->l1_size
, 0);
638 error_setg_errno(errp
, -ret
,
639 "Failed to update snapshot status in disk");
645 BdrvCheckResult result
= {0};
646 qcow2_check_refcounts(bs
, &result
, 0);
652 int qcow2_snapshot_list(BlockDriverState
*bs
, QEMUSnapshotInfo
**psn_tab
)
654 BDRVQcowState
*s
= bs
->opaque
;
655 QEMUSnapshotInfo
*sn_tab
, *sn_info
;
659 if (!s
->nb_snapshots
) {
661 return s
->nb_snapshots
;
664 sn_tab
= g_malloc0(s
->nb_snapshots
* sizeof(QEMUSnapshotInfo
));
665 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
666 sn_info
= sn_tab
+ i
;
667 sn
= s
->snapshots
+ i
;
668 pstrcpy(sn_info
->id_str
, sizeof(sn_info
->id_str
),
670 pstrcpy(sn_info
->name
, sizeof(sn_info
->name
),
672 sn_info
->vm_state_size
= sn
->vm_state_size
;
673 sn_info
->date_sec
= sn
->date_sec
;
674 sn_info
->date_nsec
= sn
->date_nsec
;
675 sn_info
->vm_clock_nsec
= sn
->vm_clock_nsec
;
678 return s
->nb_snapshots
;
681 int qcow2_snapshot_load_tmp(BlockDriverState
*bs
,
682 const char *snapshot_id
,
686 int i
, snapshot_index
;
687 BDRVQcowState
*s
= bs
->opaque
;
689 uint64_t *new_l1_table
;
693 assert(bs
->read_only
);
695 /* Search the snapshot */
696 snapshot_index
= find_snapshot_by_id_and_name(bs
, snapshot_id
, name
);
697 if (snapshot_index
< 0) {
699 "Can't find snapshot");
702 sn
= &s
->snapshots
[snapshot_index
];
704 /* Allocate and read in the snapshot's L1 table */
705 if (sn
->l1_size
> QCOW_MAX_L1_SIZE
) {
706 error_setg(errp
, "Snapshot L1 table too large");
709 new_l1_bytes
= sn
->l1_size
* sizeof(uint64_t);
710 new_l1_table
= qemu_try_blockalign(bs
->file
,
711 align_offset(new_l1_bytes
, 512));
712 if (new_l1_table
== NULL
) {
716 ret
= bdrv_pread(bs
->file
, sn
->l1_table_offset
, new_l1_table
, new_l1_bytes
);
718 error_setg(errp
, "Failed to read l1 table for snapshot");
719 qemu_vfree(new_l1_table
);
723 /* Switch the L1 table */
724 qemu_vfree(s
->l1_table
);
726 s
->l1_size
= sn
->l1_size
;
727 s
->l1_table_offset
= sn
->l1_table_offset
;
728 s
->l1_table
= new_l1_table
;
730 for(i
= 0;i
< s
->l1_size
; i
++) {
731 be64_to_cpus(&s
->l1_table
[i
]);