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"
28 #include "qemu/error-report.h"
30 void qcow2_free_snapshots(BlockDriverState
*bs
)
32 BDRVQcow2State
*s
= bs
->opaque
;
35 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
36 g_free(s
->snapshots
[i
].name
);
37 g_free(s
->snapshots
[i
].id_str
);
44 int qcow2_read_snapshots(BlockDriverState
*bs
)
46 BDRVQcow2State
*s
= bs
->opaque
;
48 QCowSnapshotExtraData extra
;
50 int i
, id_str_size
, name_size
;
52 uint32_t extra_data_size
;
55 if (!s
->nb_snapshots
) {
57 s
->snapshots_size
= 0;
61 offset
= s
->snapshots_offset
;
62 s
->snapshots
= g_new0(QCowSnapshot
, s
->nb_snapshots
);
64 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
65 /* Read statically sized part of the snapshot header */
66 offset
= align_offset(offset
, 8);
67 ret
= bdrv_pread(bs
->file
->bs
, offset
, &h
, sizeof(h
));
73 sn
= s
->snapshots
+ i
;
74 sn
->l1_table_offset
= be64_to_cpu(h
.l1_table_offset
);
75 sn
->l1_size
= be32_to_cpu(h
.l1_size
);
76 sn
->vm_state_size
= be32_to_cpu(h
.vm_state_size
);
77 sn
->date_sec
= be32_to_cpu(h
.date_sec
);
78 sn
->date_nsec
= be32_to_cpu(h
.date_nsec
);
79 sn
->vm_clock_nsec
= be64_to_cpu(h
.vm_clock_nsec
);
80 extra_data_size
= be32_to_cpu(h
.extra_data_size
);
82 id_str_size
= be16_to_cpu(h
.id_str_size
);
83 name_size
= be16_to_cpu(h
.name_size
);
86 ret
= bdrv_pread(bs
->file
->bs
, offset
, &extra
,
87 MIN(sizeof(extra
), extra_data_size
));
91 offset
+= extra_data_size
;
93 if (extra_data_size
>= 8) {
94 sn
->vm_state_size
= be64_to_cpu(extra
.vm_state_size_large
);
97 if (extra_data_size
>= 16) {
98 sn
->disk_size
= be64_to_cpu(extra
.disk_size
);
100 sn
->disk_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
103 /* Read snapshot ID */
104 sn
->id_str
= g_malloc(id_str_size
+ 1);
105 ret
= bdrv_pread(bs
->file
->bs
, offset
, sn
->id_str
, id_str_size
);
109 offset
+= id_str_size
;
110 sn
->id_str
[id_str_size
] = '\0';
112 /* Read snapshot name */
113 sn
->name
= g_malloc(name_size
+ 1);
114 ret
= bdrv_pread(bs
->file
->bs
, offset
, sn
->name
, name_size
);
119 sn
->name
[name_size
] = '\0';
121 if (offset
- s
->snapshots_offset
> QCOW_MAX_SNAPSHOTS_SIZE
) {
127 assert(offset
- s
->snapshots_offset
<= INT_MAX
);
128 s
->snapshots_size
= offset
- s
->snapshots_offset
;
132 qcow2_free_snapshots(bs
);
136 /* add at the end of the file a new list of snapshots */
137 static int qcow2_write_snapshots(BlockDriverState
*bs
)
139 BDRVQcow2State
*s
= bs
->opaque
;
141 QCowSnapshotHeader h
;
142 QCowSnapshotExtraData extra
;
143 int i
, name_size
, id_str_size
, snapshots_size
;
145 uint32_t nb_snapshots
;
146 uint64_t snapshots_offset
;
147 } QEMU_PACKED header_data
;
148 int64_t offset
, snapshots_offset
= 0;
151 /* compute the size of the snapshots */
153 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
154 sn
= s
->snapshots
+ i
;
155 offset
= align_offset(offset
, 8);
157 offset
+= sizeof(extra
);
158 offset
+= strlen(sn
->id_str
);
159 offset
+= strlen(sn
->name
);
161 if (offset
> QCOW_MAX_SNAPSHOTS_SIZE
) {
167 assert(offset
<= INT_MAX
);
168 snapshots_size
= offset
;
170 /* Allocate space for the new snapshot list */
171 snapshots_offset
= qcow2_alloc_clusters(bs
, snapshots_size
);
172 offset
= snapshots_offset
;
177 ret
= bdrv_flush(bs
);
182 /* The snapshot list position has not yet been updated, so these clusters
183 * must indeed be completely free */
184 ret
= qcow2_pre_write_overlap_check(bs
, 0, offset
, snapshots_size
);
190 /* Write all snapshots to the new list */
191 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
192 sn
= s
->snapshots
+ i
;
193 memset(&h
, 0, sizeof(h
));
194 h
.l1_table_offset
= cpu_to_be64(sn
->l1_table_offset
);
195 h
.l1_size
= cpu_to_be32(sn
->l1_size
);
196 /* If it doesn't fit in 32 bit, older implementations should treat it
197 * as a disk-only snapshot rather than truncate the VM state */
198 if (sn
->vm_state_size
<= 0xffffffff) {
199 h
.vm_state_size
= cpu_to_be32(sn
->vm_state_size
);
201 h
.date_sec
= cpu_to_be32(sn
->date_sec
);
202 h
.date_nsec
= cpu_to_be32(sn
->date_nsec
);
203 h
.vm_clock_nsec
= cpu_to_be64(sn
->vm_clock_nsec
);
204 h
.extra_data_size
= cpu_to_be32(sizeof(extra
));
206 memset(&extra
, 0, sizeof(extra
));
207 extra
.vm_state_size_large
= cpu_to_be64(sn
->vm_state_size
);
208 extra
.disk_size
= cpu_to_be64(sn
->disk_size
);
210 id_str_size
= strlen(sn
->id_str
);
211 name_size
= strlen(sn
->name
);
212 assert(id_str_size
<= UINT16_MAX
&& name_size
<= UINT16_MAX
);
213 h
.id_str_size
= cpu_to_be16(id_str_size
);
214 h
.name_size
= cpu_to_be16(name_size
);
215 offset
= align_offset(offset
, 8);
217 ret
= bdrv_pwrite(bs
->file
->bs
, offset
, &h
, sizeof(h
));
223 ret
= bdrv_pwrite(bs
->file
->bs
, offset
, &extra
, sizeof(extra
));
227 offset
+= sizeof(extra
);
229 ret
= bdrv_pwrite(bs
->file
->bs
, offset
, sn
->id_str
, id_str_size
);
233 offset
+= id_str_size
;
235 ret
= bdrv_pwrite(bs
->file
->bs
, offset
, sn
->name
, name_size
);
243 * Update the header to point to the new snapshot table. This requires the
244 * new table and its refcounts to be stable on disk.
246 ret
= bdrv_flush(bs
);
251 QEMU_BUILD_BUG_ON(offsetof(QCowHeader
, snapshots_offset
) !=
252 offsetof(QCowHeader
, nb_snapshots
) + sizeof(header_data
.nb_snapshots
));
254 header_data
.nb_snapshots
= cpu_to_be32(s
->nb_snapshots
);
255 header_data
.snapshots_offset
= cpu_to_be64(snapshots_offset
);
257 ret
= bdrv_pwrite_sync(bs
->file
->bs
, offsetof(QCowHeader
, nb_snapshots
),
258 &header_data
, sizeof(header_data
));
263 /* free the old snapshot table */
264 qcow2_free_clusters(bs
, s
->snapshots_offset
, s
->snapshots_size
,
265 QCOW2_DISCARD_SNAPSHOT
);
266 s
->snapshots_offset
= snapshots_offset
;
267 s
->snapshots_size
= snapshots_size
;
271 if (snapshots_offset
> 0) {
272 qcow2_free_clusters(bs
, snapshots_offset
, snapshots_size
,
273 QCOW2_DISCARD_ALWAYS
);
278 static void find_new_snapshot_id(BlockDriverState
*bs
,
279 char *id_str
, int id_str_size
)
281 BDRVQcow2State
*s
= bs
->opaque
;
284 unsigned long id
, id_max
= 0;
286 for(i
= 0; i
< s
->nb_snapshots
; i
++) {
287 sn
= s
->snapshots
+ i
;
288 id
= strtoul(sn
->id_str
, NULL
, 10);
292 snprintf(id_str
, id_str_size
, "%lu", id_max
+ 1);
295 static int find_snapshot_by_id_and_name(BlockDriverState
*bs
,
299 BDRVQcow2State
*s
= bs
->opaque
;
303 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
304 if (!strcmp(s
->snapshots
[i
].id_str
, id
) &&
305 !strcmp(s
->snapshots
[i
].name
, name
)) {
310 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
311 if (!strcmp(s
->snapshots
[i
].id_str
, id
)) {
316 for (i
= 0; i
< s
->nb_snapshots
; i
++) {
317 if (!strcmp(s
->snapshots
[i
].name
, name
)) {
326 static int find_snapshot_by_id_or_name(BlockDriverState
*bs
,
327 const char *id_or_name
)
331 ret
= find_snapshot_by_id_and_name(bs
, id_or_name
, NULL
);
335 return find_snapshot_by_id_and_name(bs
, NULL
, id_or_name
);
338 /* if no id is provided, a new one is constructed */
339 int qcow2_snapshot_create(BlockDriverState
*bs
, QEMUSnapshotInfo
*sn_info
)
341 BDRVQcow2State
*s
= bs
->opaque
;
342 QCowSnapshot
*new_snapshot_list
= NULL
;
343 QCowSnapshot
*old_snapshot_list
= NULL
;
344 QCowSnapshot sn1
, *sn
= &sn1
;
346 uint64_t *l1_table
= NULL
;
347 int64_t l1_table_offset
;
349 if (s
->nb_snapshots
>= QCOW_MAX_SNAPSHOTS
) {
353 memset(sn
, 0, sizeof(*sn
));
356 find_new_snapshot_id(bs
, sn_info
->id_str
, sizeof(sn_info
->id_str
));
358 /* Check that the ID is unique */
359 if (find_snapshot_by_id_and_name(bs
, sn_info
->id_str
, NULL
) >= 0) {
363 /* Populate sn with passed data */
364 sn
->id_str
= g_strdup(sn_info
->id_str
);
365 sn
->name
= g_strdup(sn_info
->name
);
367 sn
->disk_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
368 sn
->vm_state_size
= sn_info
->vm_state_size
;
369 sn
->date_sec
= sn_info
->date_sec
;
370 sn
->date_nsec
= sn_info
->date_nsec
;
371 sn
->vm_clock_nsec
= sn_info
->vm_clock_nsec
;
373 /* Allocate the L1 table of the snapshot and copy the current one there. */
374 l1_table_offset
= qcow2_alloc_clusters(bs
, s
->l1_size
* sizeof(uint64_t));
375 if (l1_table_offset
< 0) {
376 ret
= l1_table_offset
;
380 sn
->l1_table_offset
= l1_table_offset
;
381 sn
->l1_size
= s
->l1_size
;
383 l1_table
= g_try_new(uint64_t, s
->l1_size
);
384 if (s
->l1_size
&& l1_table
== NULL
) {
389 for(i
= 0; i
< s
->l1_size
; i
++) {
390 l1_table
[i
] = cpu_to_be64(s
->l1_table
[i
]);
393 ret
= qcow2_pre_write_overlap_check(bs
, 0, sn
->l1_table_offset
,
394 s
->l1_size
* sizeof(uint64_t));
399 ret
= bdrv_pwrite(bs
->file
->bs
, sn
->l1_table_offset
, l1_table
,
400 s
->l1_size
* sizeof(uint64_t));
409 * Increase the refcounts of all clusters and make sure everything is
410 * stable on disk before updating the snapshot table to contain a pointer
411 * to the new L1 table.
413 ret
= qcow2_update_snapshot_refcount(bs
, s
->l1_table_offset
, s
->l1_size
, 1);
418 /* Append the new snapshot to the snapshot list */
419 new_snapshot_list
= g_new(QCowSnapshot
, s
->nb_snapshots
+ 1);
421 memcpy(new_snapshot_list
, s
->snapshots
,
422 s
->nb_snapshots
* sizeof(QCowSnapshot
));
423 old_snapshot_list
= s
->snapshots
;
425 s
->snapshots
= new_snapshot_list
;
426 s
->snapshots
[s
->nb_snapshots
++] = *sn
;
428 ret
= qcow2_write_snapshots(bs
);
430 g_free(s
->snapshots
);
431 s
->snapshots
= old_snapshot_list
;
436 g_free(old_snapshot_list
);
438 /* The VM state isn't needed any more in the active L1 table; in fact, it
439 * hurts by causing expensive COW for the next snapshot. */
440 qcow2_discard_clusters(bs
, qcow2_vm_state_offset(s
),
441 align_offset(sn
->vm_state_size
, s
->cluster_size
)
443 QCOW2_DISCARD_NEVER
, false);
447 BdrvCheckResult result
= {0};
448 qcow2_check_refcounts(bs
, &result
, 0);
461 /* copy the snapshot 'snapshot_name' into the current disk image */
462 int qcow2_snapshot_goto(BlockDriverState
*bs
, const char *snapshot_id
)
464 BDRVQcow2State
*s
= bs
->opaque
;
466 int i
, snapshot_index
;
467 int cur_l1_bytes
, sn_l1_bytes
;
469 uint64_t *sn_l1_table
= NULL
;
471 /* Search the snapshot */
472 snapshot_index
= find_snapshot_by_id_or_name(bs
, snapshot_id
);
473 if (snapshot_index
< 0) {
476 sn
= &s
->snapshots
[snapshot_index
];
478 if (sn
->disk_size
!= bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
479 error_report("qcow2: Loading snapshots with different disk "
480 "size is not implemented");
486 * Make sure that the current L1 table is big enough to contain the whole
487 * L1 table of the snapshot. If the snapshot L1 table is smaller, the
488 * current one must be padded with zeros.
490 ret
= qcow2_grow_l1_table(bs
, sn
->l1_size
, true);
495 cur_l1_bytes
= s
->l1_size
* sizeof(uint64_t);
496 sn_l1_bytes
= sn
->l1_size
* sizeof(uint64_t);
499 * Copy the snapshot L1 table to the current L1 table.
501 * Before overwriting the old current L1 table on disk, make sure to
502 * increase all refcounts for the clusters referenced by the new one.
503 * Decrease the refcount referenced by the old one only when the L1
504 * table is overwritten.
506 sn_l1_table
= g_try_malloc0(cur_l1_bytes
);
507 if (cur_l1_bytes
&& sn_l1_table
== NULL
) {
512 ret
= bdrv_pread(bs
->file
->bs
, sn
->l1_table_offset
,
513 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
->bs
, 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 BDRVQcow2State
*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 BDRVQcow2State
*s
= bs
->opaque
;
655 QEMUSnapshotInfo
*sn_tab
, *sn_info
;
659 if (!s
->nb_snapshots
) {
661 return s
->nb_snapshots
;
664 sn_tab
= g_new0(QEMUSnapshotInfo
, s
->nb_snapshots
);
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 BDRVQcow2State
*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
/ sizeof(uint64_t)) {
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
->bs
,
711 align_offset(new_l1_bytes
, 512));
712 if (new_l1_table
== NULL
) {
716 ret
= bdrv_pread(bs
->file
->bs
, sn
->l1_table_offset
,
717 new_l1_table
, new_l1_bytes
);
719 error_setg(errp
, "Failed to read l1 table for snapshot");
720 qemu_vfree(new_l1_table
);
724 /* Switch the L1 table */
725 qemu_vfree(s
->l1_table
);
727 s
->l1_size
= sn
->l1_size
;
728 s
->l1_table_offset
= sn
->l1_table_offset
;
729 s
->l1_table
= new_l1_table
;
731 for(i
= 0;i
< s
->l1_size
; i
++) {
732 be64_to_cpus(&s
->l1_table
[i
]);