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
29 #include "block/coroutine.h"
32 //#define DEBUG_ALLOC2
35 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
37 #define QCOW_CRYPT_NONE 0
38 #define QCOW_CRYPT_AES 1
40 #define QCOW_MAX_CRYPT_CLUSTERS 32
41 #define QCOW_MAX_SNAPSHOTS 65536
43 /* 8 MB refcount table is enough for 2 PB images at 64k cluster size
44 * (128 GB for 512 byte clusters, 2 EB for 2 MB clusters) */
45 #define QCOW_MAX_REFTABLE_SIZE 0x800000
47 /* 32 MB L1 table is enough for 2 PB images at 64k cluster size
48 * (128 GB for 512 byte clusters, 2 EB for 2 MB clusters) */
49 #define QCOW_MAX_L1_SIZE 0x2000000
51 /* Allow for an average of 1k per snapshot table entry, should be plenty of
52 * space for snapshot names and IDs */
53 #define QCOW_MAX_SNAPSHOTS_SIZE (1024 * QCOW_MAX_SNAPSHOTS)
55 /* indicate that the refcount of the referenced cluster is exactly one. */
56 #define QCOW_OFLAG_COPIED (1ULL << 63)
57 /* indicate that the cluster is compressed (they never have the copied flag) */
58 #define QCOW_OFLAG_COMPRESSED (1ULL << 62)
59 /* The cluster reads as all zeros */
60 #define QCOW_OFLAG_ZERO (1ULL << 0)
62 #define REFCOUNT_SHIFT 1 /* refcount size is 2 bytes */
64 #define MIN_CLUSTER_BITS 9
65 #define MAX_CLUSTER_BITS 21
67 #define MIN_L2_CACHE_SIZE 1 /* cluster */
69 /* Must be at least 4 to cover all cases of refcount table growth */
70 #define MIN_REFCOUNT_CACHE_SIZE 4 /* clusters */
72 #define DEFAULT_L2_CACHE_BYTE_SIZE 1048576 /* bytes */
74 /* The refblock cache needs only a fourth of the L2 cache size to cover as many
76 #define DEFAULT_L2_REFCOUNT_SIZE_RATIO 4
78 #define DEFAULT_CLUSTER_SIZE 65536
81 #define QCOW2_OPT_LAZY_REFCOUNTS "lazy-refcounts"
82 #define QCOW2_OPT_DISCARD_REQUEST "pass-discard-request"
83 #define QCOW2_OPT_DISCARD_SNAPSHOT "pass-discard-snapshot"
84 #define QCOW2_OPT_DISCARD_OTHER "pass-discard-other"
85 #define QCOW2_OPT_OVERLAP "overlap-check"
86 #define QCOW2_OPT_OVERLAP_MAIN_HEADER "overlap-check.main-header"
87 #define QCOW2_OPT_OVERLAP_ACTIVE_L1 "overlap-check.active-l1"
88 #define QCOW2_OPT_OVERLAP_ACTIVE_L2 "overlap-check.active-l2"
89 #define QCOW2_OPT_OVERLAP_REFCOUNT_TABLE "overlap-check.refcount-table"
90 #define QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK "overlap-check.refcount-block"
91 #define QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE "overlap-check.snapshot-table"
92 #define QCOW2_OPT_OVERLAP_INACTIVE_L1 "overlap-check.inactive-l1"
93 #define QCOW2_OPT_OVERLAP_INACTIVE_L2 "overlap-check.inactive-l2"
94 #define QCOW2_OPT_CACHE_SIZE "cache-size"
95 #define QCOW2_OPT_L2_CACHE_SIZE "l2-cache-size"
96 #define QCOW2_OPT_REFCOUNT_CACHE_SIZE "refcount-cache-size"
98 typedef struct QCowHeader
{
101 uint64_t backing_file_offset
;
102 uint32_t backing_file_size
;
103 uint32_t cluster_bits
;
104 uint64_t size
; /* in bytes */
105 uint32_t crypt_method
;
106 uint32_t l1_size
; /* XXX: save number of clusters instead ? */
107 uint64_t l1_table_offset
;
108 uint64_t refcount_table_offset
;
109 uint32_t refcount_table_clusters
;
110 uint32_t nb_snapshots
;
111 uint64_t snapshots_offset
;
113 /* The following fields are only valid for version >= 3 */
114 uint64_t incompatible_features
;
115 uint64_t compatible_features
;
116 uint64_t autoclear_features
;
118 uint32_t refcount_order
;
119 uint32_t header_length
;
120 } QEMU_PACKED QCowHeader
;
122 typedef struct QEMU_PACKED QCowSnapshotHeader
{
123 /* header is 8 byte aligned */
124 uint64_t l1_table_offset
;
127 uint16_t id_str_size
;
133 uint64_t vm_clock_nsec
;
135 uint32_t vm_state_size
;
136 uint32_t extra_data_size
; /* for extension */
137 /* extra data follows */
140 } QCowSnapshotHeader
;
142 typedef struct QEMU_PACKED QCowSnapshotExtraData
{
143 uint64_t vm_state_size_large
;
145 } QCowSnapshotExtraData
;
148 typedef struct QCowSnapshot
{
149 uint64_t l1_table_offset
;
154 uint64_t vm_state_size
;
157 uint64_t vm_clock_nsec
;
161 typedef struct Qcow2Cache Qcow2Cache
;
163 typedef struct Qcow2UnknownHeaderExtension
{
166 QLIST_ENTRY(Qcow2UnknownHeaderExtension
) next
;
168 } Qcow2UnknownHeaderExtension
;
171 QCOW2_FEAT_TYPE_INCOMPATIBLE
= 0,
172 QCOW2_FEAT_TYPE_COMPATIBLE
= 1,
173 QCOW2_FEAT_TYPE_AUTOCLEAR
= 2,
176 /* Incompatible feature bits */
178 QCOW2_INCOMPAT_DIRTY_BITNR
= 0,
179 QCOW2_INCOMPAT_CORRUPT_BITNR
= 1,
180 QCOW2_INCOMPAT_DIRTY
= 1 << QCOW2_INCOMPAT_DIRTY_BITNR
,
181 QCOW2_INCOMPAT_CORRUPT
= 1 << QCOW2_INCOMPAT_CORRUPT_BITNR
,
183 QCOW2_INCOMPAT_MASK
= QCOW2_INCOMPAT_DIRTY
184 | QCOW2_INCOMPAT_CORRUPT
,
187 /* Compatible feature bits */
189 QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR
= 0,
190 QCOW2_COMPAT_LAZY_REFCOUNTS
= 1 << QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR
,
192 QCOW2_COMPAT_FEAT_MASK
= QCOW2_COMPAT_LAZY_REFCOUNTS
,
195 enum qcow2_discard_type
{
196 QCOW2_DISCARD_NEVER
= 0,
197 QCOW2_DISCARD_ALWAYS
,
198 QCOW2_DISCARD_REQUEST
,
199 QCOW2_DISCARD_SNAPSHOT
,
204 typedef struct Qcow2Feature
{
208 } QEMU_PACKED Qcow2Feature
;
210 typedef struct Qcow2DiscardRegion
{
211 BlockDriverState
*bs
;
214 QTAILQ_ENTRY(Qcow2DiscardRegion
) next
;
215 } Qcow2DiscardRegion
;
217 typedef struct BDRVQcowState
{
224 int l1_vm_state_index
;
227 uint64_t cluster_offset_mask
;
228 uint64_t l1_table_offset
;
231 Qcow2Cache
* l2_table_cache
;
232 Qcow2Cache
* refcount_block_cache
;
234 uint8_t *cluster_cache
;
235 uint8_t *cluster_data
;
236 uint64_t cluster_cache_offset
;
237 QLIST_HEAD(QCowClusterAlloc
, QCowL2Meta
) cluster_allocs
;
239 uint64_t *refcount_table
;
240 uint64_t refcount_table_offset
;
241 uint32_t refcount_table_size
;
242 uint64_t free_cluster_index
;
243 uint64_t free_byte_offset
;
247 uint32_t crypt_method
; /* current crypt method, 0 if no key yet */
248 uint32_t crypt_method_header
;
249 AES_KEY aes_encrypt_key
;
250 AES_KEY aes_decrypt_key
;
251 uint64_t snapshots_offset
;
253 unsigned int nb_snapshots
;
254 QCowSnapshot
*snapshots
;
258 bool use_lazy_refcounts
;
261 bool discard_passthrough
[QCOW2_DISCARD_MAX
];
263 int overlap_check
; /* bitmask of Qcow2MetadataOverlap values */
265 uint64_t incompatible_features
;
266 uint64_t compatible_features
;
267 uint64_t autoclear_features
;
269 size_t unknown_header_fields_size
;
270 void* unknown_header_fields
;
271 QLIST_HEAD(, Qcow2UnknownHeaderExtension
) unknown_header_ext
;
272 QTAILQ_HEAD (, Qcow2DiscardRegion
) discards
;
276 /* XXX: use std qcow open function ? */
277 typedef struct QCowCreateState
{
280 uint16_t *refcount_block
;
281 uint64_t *refcount_table
;
282 int64_t l1_table_offset
;
283 int64_t refcount_table_offset
;
284 int64_t refcount_block_offset
;
289 typedef struct Qcow2COWRegion
{
291 * Offset of the COW region in bytes from the start of the first cluster
292 * touched by the request.
296 /** Number of sectors to copy */
301 * Describes an in-flight (part of a) write request that writes to clusters
302 * that are not referenced in their L2 table yet.
304 typedef struct QCowL2Meta
306 /** Guest offset of the first newly allocated cluster */
309 /** Host offset of the first newly allocated cluster */
310 uint64_t alloc_offset
;
313 * Number of sectors from the start of the first allocated cluster to
314 * the end of the (possibly shortened) request
318 /** Number of newly allocated clusters */
322 * Requests that overlap with this allocation and wait to be restarted
323 * when the allocating request has completed.
325 CoQueue dependent_requests
;
328 * The COW Region between the start of the first allocated cluster and the
329 * area the guest actually writes to.
331 Qcow2COWRegion cow_start
;
334 * The COW Region between the area the guest actually writes to and the
335 * end of the last allocated cluster.
337 Qcow2COWRegion cow_end
;
339 /** Pointer to next L2Meta of the same write request */
340 struct QCowL2Meta
*next
;
342 QLIST_ENTRY(QCowL2Meta
) next_in_flight
;
346 QCOW2_CLUSTER_UNALLOCATED
,
347 QCOW2_CLUSTER_NORMAL
,
348 QCOW2_CLUSTER_COMPRESSED
,
352 typedef enum QCow2MetadataOverlap
{
353 QCOW2_OL_MAIN_HEADER_BITNR
= 0,
354 QCOW2_OL_ACTIVE_L1_BITNR
= 1,
355 QCOW2_OL_ACTIVE_L2_BITNR
= 2,
356 QCOW2_OL_REFCOUNT_TABLE_BITNR
= 3,
357 QCOW2_OL_REFCOUNT_BLOCK_BITNR
= 4,
358 QCOW2_OL_SNAPSHOT_TABLE_BITNR
= 5,
359 QCOW2_OL_INACTIVE_L1_BITNR
= 6,
360 QCOW2_OL_INACTIVE_L2_BITNR
= 7,
362 QCOW2_OL_MAX_BITNR
= 8,
365 QCOW2_OL_MAIN_HEADER
= (1 << QCOW2_OL_MAIN_HEADER_BITNR
),
366 QCOW2_OL_ACTIVE_L1
= (1 << QCOW2_OL_ACTIVE_L1_BITNR
),
367 QCOW2_OL_ACTIVE_L2
= (1 << QCOW2_OL_ACTIVE_L2_BITNR
),
368 QCOW2_OL_REFCOUNT_TABLE
= (1 << QCOW2_OL_REFCOUNT_TABLE_BITNR
),
369 QCOW2_OL_REFCOUNT_BLOCK
= (1 << QCOW2_OL_REFCOUNT_BLOCK_BITNR
),
370 QCOW2_OL_SNAPSHOT_TABLE
= (1 << QCOW2_OL_SNAPSHOT_TABLE_BITNR
),
371 QCOW2_OL_INACTIVE_L1
= (1 << QCOW2_OL_INACTIVE_L1_BITNR
),
372 /* NOTE: Checking overlaps with inactive L2 tables will result in bdrv
374 QCOW2_OL_INACTIVE_L2
= (1 << QCOW2_OL_INACTIVE_L2_BITNR
),
375 } QCow2MetadataOverlap
;
377 /* Perform all overlap checks which can be done in constant time */
378 #define QCOW2_OL_CONSTANT \
379 (QCOW2_OL_MAIN_HEADER | QCOW2_OL_ACTIVE_L1 | QCOW2_OL_REFCOUNT_TABLE | \
380 QCOW2_OL_SNAPSHOT_TABLE)
382 /* Perform all overlap checks which don't require disk access */
383 #define QCOW2_OL_CACHED \
384 (QCOW2_OL_CONSTANT | QCOW2_OL_ACTIVE_L2 | QCOW2_OL_REFCOUNT_BLOCK | \
385 QCOW2_OL_INACTIVE_L1)
387 /* Perform all overlap checks */
388 #define QCOW2_OL_ALL \
389 (QCOW2_OL_CACHED | QCOW2_OL_INACTIVE_L2)
391 #define L1E_OFFSET_MASK 0x00fffffffffffe00ULL
392 #define L2E_OFFSET_MASK 0x00fffffffffffe00ULL
393 #define L2E_COMPRESSED_OFFSET_SIZE_MASK 0x3fffffffffffffffULL
395 #define REFT_OFFSET_MASK 0xfffffffffffffe00ULL
397 static inline int64_t start_of_cluster(BDRVQcowState
*s
, int64_t offset
)
399 return offset
& ~(s
->cluster_size
- 1);
402 static inline int64_t offset_into_cluster(BDRVQcowState
*s
, int64_t offset
)
404 return offset
& (s
->cluster_size
- 1);
407 static inline int size_to_clusters(BDRVQcowState
*s
, int64_t size
)
409 return (size
+ (s
->cluster_size
- 1)) >> s
->cluster_bits
;
412 static inline int64_t size_to_l1(BDRVQcowState
*s
, int64_t size
)
414 int shift
= s
->cluster_bits
+ s
->l2_bits
;
415 return (size
+ (1ULL << shift
) - 1) >> shift
;
418 static inline int offset_to_l2_index(BDRVQcowState
*s
, int64_t offset
)
420 return (offset
>> s
->cluster_bits
) & (s
->l2_size
- 1);
423 static inline int64_t align_offset(int64_t offset
, int n
)
425 offset
= (offset
+ n
- 1) & ~(n
- 1);
429 static inline int64_t qcow2_vm_state_offset(BDRVQcowState
*s
)
431 return (int64_t)s
->l1_vm_state_index
<< (s
->cluster_bits
+ s
->l2_bits
);
434 static inline uint64_t qcow2_max_refcount_clusters(BDRVQcowState
*s
)
436 return QCOW_MAX_REFTABLE_SIZE
>> s
->cluster_bits
;
439 static inline int qcow2_get_cluster_type(uint64_t l2_entry
)
441 if (l2_entry
& QCOW_OFLAG_COMPRESSED
) {
442 return QCOW2_CLUSTER_COMPRESSED
;
443 } else if (l2_entry
& QCOW_OFLAG_ZERO
) {
444 return QCOW2_CLUSTER_ZERO
;
445 } else if (!(l2_entry
& L2E_OFFSET_MASK
)) {
446 return QCOW2_CLUSTER_UNALLOCATED
;
448 return QCOW2_CLUSTER_NORMAL
;
452 /* Check whether refcounts are eager or lazy */
453 static inline bool qcow2_need_accurate_refcounts(BDRVQcowState
*s
)
455 return !(s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
);
458 static inline uint64_t l2meta_cow_start(QCowL2Meta
*m
)
460 return m
->offset
+ m
->cow_start
.offset
;
463 static inline uint64_t l2meta_cow_end(QCowL2Meta
*m
)
465 return m
->offset
+ m
->cow_end
.offset
466 + (m
->cow_end
.nb_sectors
<< BDRV_SECTOR_BITS
);
469 // FIXME Need qcow2_ prefix to global functions
471 /* qcow2.c functions */
472 int qcow2_backing_read1(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
473 int64_t sector_num
, int nb_sectors
);
475 int qcow2_mark_dirty(BlockDriverState
*bs
);
476 int qcow2_mark_corrupt(BlockDriverState
*bs
);
477 int qcow2_mark_consistent(BlockDriverState
*bs
);
478 int qcow2_update_header(BlockDriverState
*bs
);
480 /* qcow2-refcount.c functions */
481 int qcow2_refcount_init(BlockDriverState
*bs
);
482 void qcow2_refcount_close(BlockDriverState
*bs
);
484 int qcow2_update_cluster_refcount(BlockDriverState
*bs
, int64_t cluster_index
,
485 int addend
, enum qcow2_discard_type type
);
487 int64_t qcow2_alloc_clusters(BlockDriverState
*bs
, uint64_t size
);
488 int qcow2_alloc_clusters_at(BlockDriverState
*bs
, uint64_t offset
,
490 int64_t qcow2_alloc_bytes(BlockDriverState
*bs
, int size
);
491 void qcow2_free_clusters(BlockDriverState
*bs
,
492 int64_t offset
, int64_t size
,
493 enum qcow2_discard_type type
);
494 void qcow2_free_any_clusters(BlockDriverState
*bs
, uint64_t l2_entry
,
495 int nb_clusters
, enum qcow2_discard_type type
);
497 int qcow2_update_snapshot_refcount(BlockDriverState
*bs
,
498 int64_t l1_table_offset
, int l1_size
, int addend
);
500 int qcow2_check_refcounts(BlockDriverState
*bs
, BdrvCheckResult
*res
,
503 void qcow2_process_discards(BlockDriverState
*bs
, int ret
);
505 int qcow2_check_metadata_overlap(BlockDriverState
*bs
, int ign
, int64_t offset
,
507 int qcow2_pre_write_overlap_check(BlockDriverState
*bs
, int ign
, int64_t offset
,
510 /* qcow2-cluster.c functions */
511 int qcow2_grow_l1_table(BlockDriverState
*bs
, uint64_t min_size
,
513 int qcow2_write_l1_entry(BlockDriverState
*bs
, int l1_index
);
514 void qcow2_l2_cache_reset(BlockDriverState
*bs
);
515 int qcow2_decompress_cluster(BlockDriverState
*bs
, uint64_t cluster_offset
);
516 void qcow2_encrypt_sectors(BDRVQcowState
*s
, int64_t sector_num
,
517 uint8_t *out_buf
, const uint8_t *in_buf
,
518 int nb_sectors
, int enc
,
521 int qcow2_get_cluster_offset(BlockDriverState
*bs
, uint64_t offset
,
522 int *num
, uint64_t *cluster_offset
);
523 int qcow2_alloc_cluster_offset(BlockDriverState
*bs
, uint64_t offset
,
524 int *num
, uint64_t *host_offset
, QCowL2Meta
**m
);
525 uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState
*bs
,
527 int compressed_size
);
529 int qcow2_alloc_cluster_link_l2(BlockDriverState
*bs
, QCowL2Meta
*m
);
530 int qcow2_discard_clusters(BlockDriverState
*bs
, uint64_t offset
,
531 int nb_sectors
, enum qcow2_discard_type type
);
532 int qcow2_zero_clusters(BlockDriverState
*bs
, uint64_t offset
, int nb_sectors
);
534 int qcow2_expand_zero_clusters(BlockDriverState
*bs
);
536 /* qcow2-snapshot.c functions */
537 int qcow2_snapshot_create(BlockDriverState
*bs
, QEMUSnapshotInfo
*sn_info
);
538 int qcow2_snapshot_goto(BlockDriverState
*bs
, const char *snapshot_id
);
539 int qcow2_snapshot_delete(BlockDriverState
*bs
,
540 const char *snapshot_id
,
543 int qcow2_snapshot_list(BlockDriverState
*bs
, QEMUSnapshotInfo
**psn_tab
);
544 int qcow2_snapshot_load_tmp(BlockDriverState
*bs
,
545 const char *snapshot_id
,
549 void qcow2_free_snapshots(BlockDriverState
*bs
);
550 int qcow2_read_snapshots(BlockDriverState
*bs
);
552 /* qcow2-cache.c functions */
553 Qcow2Cache
*qcow2_cache_create(BlockDriverState
*bs
, int num_tables
);
554 int qcow2_cache_destroy(BlockDriverState
* bs
, Qcow2Cache
*c
);
556 void qcow2_cache_entry_mark_dirty(Qcow2Cache
*c
, void *table
);
557 int qcow2_cache_flush(BlockDriverState
*bs
, Qcow2Cache
*c
);
558 int qcow2_cache_set_dependency(BlockDriverState
*bs
, Qcow2Cache
*c
,
559 Qcow2Cache
*dependency
);
560 void qcow2_cache_depends_on_flush(Qcow2Cache
*c
);
562 int qcow2_cache_empty(BlockDriverState
*bs
, Qcow2Cache
*c
);
564 int qcow2_cache_get(BlockDriverState
*bs
, Qcow2Cache
*c
, uint64_t offset
,
566 int qcow2_cache_get_empty(BlockDriverState
*bs
, Qcow2Cache
*c
, uint64_t offset
,
568 int qcow2_cache_put(BlockDriverState
*bs
, Qcow2Cache
*c
, void **table
);