xref: /openbmc/qemu/block/qcow2.h (revision 64552b6b)
1 /*
2  * Block driver for the QCOW version 2 format
3  *
4  * Copyright (c) 2004-2006 Fabrice Bellard
5  *
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:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
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
22  * THE SOFTWARE.
23  */
24 
25 #ifndef BLOCK_QCOW2_H
26 #define BLOCK_QCOW2_H
27 
28 #include "crypto/block.h"
29 #include "qemu/coroutine.h"
30 #include "qemu/units.h"
31 #include "block/block_int.h"
32 
33 //#define DEBUG_ALLOC
34 //#define DEBUG_ALLOC2
35 //#define DEBUG_EXT
36 
37 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
38 
39 #define QCOW_CRYPT_NONE 0
40 #define QCOW_CRYPT_AES  1
41 #define QCOW_CRYPT_LUKS 2
42 
43 #define QCOW_MAX_CRYPT_CLUSTERS 32
44 #define QCOW_MAX_SNAPSHOTS 65536
45 
46 /* Field widths in qcow2 mean normal cluster offsets cannot reach
47  * 64PB; depending on cluster size, compressed clusters can have a
48  * smaller limit (64PB for up to 16k clusters, then ramps down to
49  * 512TB for 2M clusters).  */
50 #define QCOW_MAX_CLUSTER_OFFSET ((1ULL << 56) - 1)
51 
52 /* 8 MB refcount table is enough for 2 PB images at 64k cluster size
53  * (128 GB for 512 byte clusters, 2 EB for 2 MB clusters) */
54 #define QCOW_MAX_REFTABLE_SIZE (8 * MiB)
55 
56 /* 32 MB L1 table is enough for 2 PB images at 64k cluster size
57  * (128 GB for 512 byte clusters, 2 EB for 2 MB clusters) */
58 #define QCOW_MAX_L1_SIZE (32 * MiB)
59 
60 /* Allow for an average of 1k per snapshot table entry, should be plenty of
61  * space for snapshot names and IDs */
62 #define QCOW_MAX_SNAPSHOTS_SIZE (1024 * QCOW_MAX_SNAPSHOTS)
63 
64 /* Bitmap header extension constraints */
65 #define QCOW2_MAX_BITMAPS 65535
66 #define QCOW2_MAX_BITMAP_DIRECTORY_SIZE (1024 * QCOW2_MAX_BITMAPS)
67 
68 /* indicate that the refcount of the referenced cluster is exactly one. */
69 #define QCOW_OFLAG_COPIED     (1ULL << 63)
70 /* indicate that the cluster is compressed (they never have the copied flag) */
71 #define QCOW_OFLAG_COMPRESSED (1ULL << 62)
72 /* The cluster reads as all zeros */
73 #define QCOW_OFLAG_ZERO (1ULL << 0)
74 
75 #define MIN_CLUSTER_BITS 9
76 #define MAX_CLUSTER_BITS 21
77 
78 /* Defined in the qcow2 spec (compressed cluster descriptor) */
79 #define QCOW2_COMPRESSED_SECTOR_SIZE 512U
80 #define QCOW2_COMPRESSED_SECTOR_MASK (~(QCOW2_COMPRESSED_SECTOR_SIZE - 1))
81 
82 /* Must be at least 2 to cover COW */
83 #define MIN_L2_CACHE_SIZE 2 /* cache entries */
84 
85 /* Must be at least 4 to cover all cases of refcount table growth */
86 #define MIN_REFCOUNT_CACHE_SIZE 4 /* clusters */
87 
88 #ifdef CONFIG_LINUX
89 #define DEFAULT_L2_CACHE_MAX_SIZE (32 * MiB)
90 #define DEFAULT_CACHE_CLEAN_INTERVAL 600  /* seconds */
91 #else
92 #define DEFAULT_L2_CACHE_MAX_SIZE (8 * MiB)
93 /* Cache clean interval is currently available only on Linux, so must be 0 */
94 #define DEFAULT_CACHE_CLEAN_INTERVAL 0
95 #endif
96 
97 #define DEFAULT_CLUSTER_SIZE 65536
98 
99 #define QCOW2_OPT_DATA_FILE "data-file"
100 #define QCOW2_OPT_LAZY_REFCOUNTS "lazy-refcounts"
101 #define QCOW2_OPT_DISCARD_REQUEST "pass-discard-request"
102 #define QCOW2_OPT_DISCARD_SNAPSHOT "pass-discard-snapshot"
103 #define QCOW2_OPT_DISCARD_OTHER "pass-discard-other"
104 #define QCOW2_OPT_OVERLAP "overlap-check"
105 #define QCOW2_OPT_OVERLAP_TEMPLATE "overlap-check.template"
106 #define QCOW2_OPT_OVERLAP_MAIN_HEADER "overlap-check.main-header"
107 #define QCOW2_OPT_OVERLAP_ACTIVE_L1 "overlap-check.active-l1"
108 #define QCOW2_OPT_OVERLAP_ACTIVE_L2 "overlap-check.active-l2"
109 #define QCOW2_OPT_OVERLAP_REFCOUNT_TABLE "overlap-check.refcount-table"
110 #define QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK "overlap-check.refcount-block"
111 #define QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE "overlap-check.snapshot-table"
112 #define QCOW2_OPT_OVERLAP_INACTIVE_L1 "overlap-check.inactive-l1"
113 #define QCOW2_OPT_OVERLAP_INACTIVE_L2 "overlap-check.inactive-l2"
114 #define QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY "overlap-check.bitmap-directory"
115 #define QCOW2_OPT_CACHE_SIZE "cache-size"
116 #define QCOW2_OPT_L2_CACHE_SIZE "l2-cache-size"
117 #define QCOW2_OPT_L2_CACHE_ENTRY_SIZE "l2-cache-entry-size"
118 #define QCOW2_OPT_REFCOUNT_CACHE_SIZE "refcount-cache-size"
119 #define QCOW2_OPT_CACHE_CLEAN_INTERVAL "cache-clean-interval"
120 
121 typedef struct QCowHeader {
122     uint32_t magic;
123     uint32_t version;
124     uint64_t backing_file_offset;
125     uint32_t backing_file_size;
126     uint32_t cluster_bits;
127     uint64_t size; /* in bytes */
128     uint32_t crypt_method;
129     uint32_t l1_size; /* XXX: save number of clusters instead ? */
130     uint64_t l1_table_offset;
131     uint64_t refcount_table_offset;
132     uint32_t refcount_table_clusters;
133     uint32_t nb_snapshots;
134     uint64_t snapshots_offset;
135 
136     /* The following fields are only valid for version >= 3 */
137     uint64_t incompatible_features;
138     uint64_t compatible_features;
139     uint64_t autoclear_features;
140 
141     uint32_t refcount_order;
142     uint32_t header_length;
143 } QEMU_PACKED QCowHeader;
144 
145 typedef struct QEMU_PACKED QCowSnapshotHeader {
146     /* header is 8 byte aligned */
147     uint64_t l1_table_offset;
148 
149     uint32_t l1_size;
150     uint16_t id_str_size;
151     uint16_t name_size;
152 
153     uint32_t date_sec;
154     uint32_t date_nsec;
155 
156     uint64_t vm_clock_nsec;
157 
158     uint32_t vm_state_size;
159     uint32_t extra_data_size; /* for extension */
160     /* extra data follows */
161     /* id_str follows */
162     /* name follows  */
163 } QCowSnapshotHeader;
164 
165 typedef struct QEMU_PACKED QCowSnapshotExtraData {
166     uint64_t vm_state_size_large;
167     uint64_t disk_size;
168 } QCowSnapshotExtraData;
169 
170 
171 typedef struct QCowSnapshot {
172     uint64_t l1_table_offset;
173     uint32_t l1_size;
174     char *id_str;
175     char *name;
176     uint64_t disk_size;
177     uint64_t vm_state_size;
178     uint32_t date_sec;
179     uint32_t date_nsec;
180     uint64_t vm_clock_nsec;
181 } QCowSnapshot;
182 
183 struct Qcow2Cache;
184 typedef struct Qcow2Cache Qcow2Cache;
185 
186 typedef struct Qcow2CryptoHeaderExtension {
187     uint64_t offset;
188     uint64_t length;
189 } QEMU_PACKED Qcow2CryptoHeaderExtension;
190 
191 typedef struct Qcow2UnknownHeaderExtension {
192     uint32_t magic;
193     uint32_t len;
194     QLIST_ENTRY(Qcow2UnknownHeaderExtension) next;
195     uint8_t data[];
196 } Qcow2UnknownHeaderExtension;
197 
198 enum {
199     QCOW2_FEAT_TYPE_INCOMPATIBLE    = 0,
200     QCOW2_FEAT_TYPE_COMPATIBLE      = 1,
201     QCOW2_FEAT_TYPE_AUTOCLEAR       = 2,
202 };
203 
204 /* Incompatible feature bits */
205 enum {
206     QCOW2_INCOMPAT_DIRTY_BITNR      = 0,
207     QCOW2_INCOMPAT_CORRUPT_BITNR    = 1,
208     QCOW2_INCOMPAT_DATA_FILE_BITNR  = 2,
209     QCOW2_INCOMPAT_DIRTY            = 1 << QCOW2_INCOMPAT_DIRTY_BITNR,
210     QCOW2_INCOMPAT_CORRUPT          = 1 << QCOW2_INCOMPAT_CORRUPT_BITNR,
211     QCOW2_INCOMPAT_DATA_FILE        = 1 << QCOW2_INCOMPAT_DATA_FILE_BITNR,
212 
213     QCOW2_INCOMPAT_MASK             = QCOW2_INCOMPAT_DIRTY
214                                     | QCOW2_INCOMPAT_CORRUPT
215                                     | QCOW2_INCOMPAT_DATA_FILE,
216 };
217 
218 /* Compatible feature bits */
219 enum {
220     QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR = 0,
221     QCOW2_COMPAT_LAZY_REFCOUNTS       = 1 << QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
222 
223     QCOW2_COMPAT_FEAT_MASK            = QCOW2_COMPAT_LAZY_REFCOUNTS,
224 };
225 
226 /* Autoclear feature bits */
227 enum {
228     QCOW2_AUTOCLEAR_BITMAPS_BITNR       = 0,
229     QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR = 1,
230     QCOW2_AUTOCLEAR_BITMAPS             = 1 << QCOW2_AUTOCLEAR_BITMAPS_BITNR,
231     QCOW2_AUTOCLEAR_DATA_FILE_RAW       = 1 << QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR,
232 
233     QCOW2_AUTOCLEAR_MASK                = QCOW2_AUTOCLEAR_BITMAPS
234                                         | QCOW2_AUTOCLEAR_DATA_FILE_RAW,
235 };
236 
237 enum qcow2_discard_type {
238     QCOW2_DISCARD_NEVER = 0,
239     QCOW2_DISCARD_ALWAYS,
240     QCOW2_DISCARD_REQUEST,
241     QCOW2_DISCARD_SNAPSHOT,
242     QCOW2_DISCARD_OTHER,
243     QCOW2_DISCARD_MAX
244 };
245 
246 typedef struct Qcow2Feature {
247     uint8_t type;
248     uint8_t bit;
249     char    name[46];
250 } QEMU_PACKED Qcow2Feature;
251 
252 typedef struct Qcow2DiscardRegion {
253     BlockDriverState *bs;
254     uint64_t offset;
255     uint64_t bytes;
256     QTAILQ_ENTRY(Qcow2DiscardRegion) next;
257 } Qcow2DiscardRegion;
258 
259 typedef uint64_t Qcow2GetRefcountFunc(const void *refcount_array,
260                                       uint64_t index);
261 typedef void Qcow2SetRefcountFunc(void *refcount_array,
262                                   uint64_t index, uint64_t value);
263 
264 typedef struct Qcow2BitmapHeaderExt {
265     uint32_t nb_bitmaps;
266     uint32_t reserved32;
267     uint64_t bitmap_directory_size;
268     uint64_t bitmap_directory_offset;
269 } QEMU_PACKED Qcow2BitmapHeaderExt;
270 
271 #define QCOW2_MAX_THREADS 4
272 
273 typedef struct BDRVQcow2State {
274     int cluster_bits;
275     int cluster_size;
276     int l2_slice_size;
277     int l2_bits;
278     int l2_size;
279     int l1_size;
280     int l1_vm_state_index;
281     int refcount_block_bits;
282     int refcount_block_size;
283     int csize_shift;
284     int csize_mask;
285     uint64_t cluster_offset_mask;
286     uint64_t l1_table_offset;
287     uint64_t *l1_table;
288 
289     Qcow2Cache* l2_table_cache;
290     Qcow2Cache* refcount_block_cache;
291     QEMUTimer *cache_clean_timer;
292     unsigned cache_clean_interval;
293 
294     uint8_t *cluster_cache;
295     uint8_t *cluster_data;
296     uint64_t cluster_cache_offset;
297     QLIST_HEAD(, QCowL2Meta) cluster_allocs;
298 
299     uint64_t *refcount_table;
300     uint64_t refcount_table_offset;
301     uint32_t refcount_table_size;
302     uint32_t max_refcount_table_index; /* Last used entry in refcount_table */
303     uint64_t free_cluster_index;
304     uint64_t free_byte_offset;
305 
306     CoMutex lock;
307 
308     Qcow2CryptoHeaderExtension crypto_header; /* QCow2 header extension */
309     QCryptoBlockOpenOptions *crypto_opts; /* Disk encryption runtime options */
310     QCryptoBlock *crypto; /* Disk encryption format driver */
311     bool crypt_physical_offset; /* Whether to use virtual or physical offset
312                                    for encryption initialization vector tweak */
313     uint32_t crypt_method_header;
314     uint64_t snapshots_offset;
315     int snapshots_size;
316     unsigned int nb_snapshots;
317     QCowSnapshot *snapshots;
318 
319     uint32_t nb_bitmaps;
320     uint64_t bitmap_directory_size;
321     uint64_t bitmap_directory_offset;
322 
323     int flags;
324     int qcow_version;
325     bool use_lazy_refcounts;
326     int refcount_order;
327     int refcount_bits;
328     uint64_t refcount_max;
329 
330     Qcow2GetRefcountFunc *get_refcount;
331     Qcow2SetRefcountFunc *set_refcount;
332 
333     bool discard_passthrough[QCOW2_DISCARD_MAX];
334 
335     int overlap_check; /* bitmask of Qcow2MetadataOverlap values */
336     bool signaled_corruption;
337 
338     uint64_t incompatible_features;
339     uint64_t compatible_features;
340     uint64_t autoclear_features;
341 
342     size_t unknown_header_fields_size;
343     void* unknown_header_fields;
344     QLIST_HEAD(, Qcow2UnknownHeaderExtension) unknown_header_ext;
345     QTAILQ_HEAD (, Qcow2DiscardRegion) discards;
346     bool cache_discards;
347 
348     /* Backing file path and format as stored in the image (this is not the
349      * effective path/format, which may be the result of a runtime option
350      * override) */
351     char *image_backing_file;
352     char *image_backing_format;
353     char *image_data_file;
354 
355     CoQueue thread_task_queue;
356     int nb_threads;
357 
358     BdrvChild *data_file;
359 
360     bool metadata_preallocation_checked;
361     bool metadata_preallocation;
362 } BDRVQcow2State;
363 
364 typedef struct Qcow2COWRegion {
365     /**
366      * Offset of the COW region in bytes from the start of the first cluster
367      * touched by the request.
368      */
369     unsigned    offset;
370 
371     /** Number of bytes to copy */
372     unsigned    nb_bytes;
373 } Qcow2COWRegion;
374 
375 /**
376  * Describes an in-flight (part of a) write request that writes to clusters
377  * that are not referenced in their L2 table yet.
378  */
379 typedef struct QCowL2Meta
380 {
381     /** Guest offset of the first newly allocated cluster */
382     uint64_t offset;
383 
384     /** Host offset of the first newly allocated cluster */
385     uint64_t alloc_offset;
386 
387     /** Number of newly allocated clusters */
388     int nb_clusters;
389 
390     /** Do not free the old clusters */
391     bool keep_old_clusters;
392 
393     /**
394      * Requests that overlap with this allocation and wait to be restarted
395      * when the allocating request has completed.
396      */
397     CoQueue dependent_requests;
398 
399     /**
400      * The COW Region between the start of the first allocated cluster and the
401      * area the guest actually writes to.
402      */
403     Qcow2COWRegion cow_start;
404 
405     /**
406      * The COW Region between the area the guest actually writes to and the
407      * end of the last allocated cluster.
408      */
409     Qcow2COWRegion cow_end;
410 
411     /*
412      * Indicates that COW regions are already handled and do not require
413      * any more processing.
414      */
415     bool skip_cow;
416 
417     /**
418      * The I/O vector with the data from the actual guest write request.
419      * If non-NULL, this is meant to be merged together with the data
420      * from @cow_start and @cow_end into one single write operation.
421      */
422     QEMUIOVector *data_qiov;
423 
424     /** Pointer to next L2Meta of the same write request */
425     struct QCowL2Meta *next;
426 
427     QLIST_ENTRY(QCowL2Meta) next_in_flight;
428 } QCowL2Meta;
429 
430 typedef enum QCow2ClusterType {
431     QCOW2_CLUSTER_UNALLOCATED,
432     QCOW2_CLUSTER_ZERO_PLAIN,
433     QCOW2_CLUSTER_ZERO_ALLOC,
434     QCOW2_CLUSTER_NORMAL,
435     QCOW2_CLUSTER_COMPRESSED,
436 } QCow2ClusterType;
437 
438 typedef enum QCow2MetadataOverlap {
439     QCOW2_OL_MAIN_HEADER_BITNR      = 0,
440     QCOW2_OL_ACTIVE_L1_BITNR        = 1,
441     QCOW2_OL_ACTIVE_L2_BITNR        = 2,
442     QCOW2_OL_REFCOUNT_TABLE_BITNR   = 3,
443     QCOW2_OL_REFCOUNT_BLOCK_BITNR   = 4,
444     QCOW2_OL_SNAPSHOT_TABLE_BITNR   = 5,
445     QCOW2_OL_INACTIVE_L1_BITNR      = 6,
446     QCOW2_OL_INACTIVE_L2_BITNR      = 7,
447     QCOW2_OL_BITMAP_DIRECTORY_BITNR = 8,
448 
449     QCOW2_OL_MAX_BITNR              = 9,
450 
451     QCOW2_OL_NONE             = 0,
452     QCOW2_OL_MAIN_HEADER      = (1 << QCOW2_OL_MAIN_HEADER_BITNR),
453     QCOW2_OL_ACTIVE_L1        = (1 << QCOW2_OL_ACTIVE_L1_BITNR),
454     QCOW2_OL_ACTIVE_L2        = (1 << QCOW2_OL_ACTIVE_L2_BITNR),
455     QCOW2_OL_REFCOUNT_TABLE   = (1 << QCOW2_OL_REFCOUNT_TABLE_BITNR),
456     QCOW2_OL_REFCOUNT_BLOCK   = (1 << QCOW2_OL_REFCOUNT_BLOCK_BITNR),
457     QCOW2_OL_SNAPSHOT_TABLE   = (1 << QCOW2_OL_SNAPSHOT_TABLE_BITNR),
458     QCOW2_OL_INACTIVE_L1      = (1 << QCOW2_OL_INACTIVE_L1_BITNR),
459     /* NOTE: Checking overlaps with inactive L2 tables will result in bdrv
460      * reads. */
461     QCOW2_OL_INACTIVE_L2      = (1 << QCOW2_OL_INACTIVE_L2_BITNR),
462     QCOW2_OL_BITMAP_DIRECTORY = (1 << QCOW2_OL_BITMAP_DIRECTORY_BITNR),
463 } QCow2MetadataOverlap;
464 
465 /* Perform all overlap checks which can be done in constant time */
466 #define QCOW2_OL_CONSTANT \
467     (QCOW2_OL_MAIN_HEADER | QCOW2_OL_ACTIVE_L1 | QCOW2_OL_REFCOUNT_TABLE | \
468      QCOW2_OL_SNAPSHOT_TABLE | QCOW2_OL_BITMAP_DIRECTORY)
469 
470 /* Perform all overlap checks which don't require disk access */
471 #define QCOW2_OL_CACHED \
472     (QCOW2_OL_CONSTANT | QCOW2_OL_ACTIVE_L2 | QCOW2_OL_REFCOUNT_BLOCK | \
473      QCOW2_OL_INACTIVE_L1)
474 
475 /* Perform all overlap checks */
476 #define QCOW2_OL_ALL \
477     (QCOW2_OL_CACHED | QCOW2_OL_INACTIVE_L2)
478 
479 #define L1E_OFFSET_MASK 0x00fffffffffffe00ULL
480 #define L2E_OFFSET_MASK 0x00fffffffffffe00ULL
481 #define L2E_COMPRESSED_OFFSET_SIZE_MASK 0x3fffffffffffffffULL
482 
483 #define REFT_OFFSET_MASK 0xfffffffffffffe00ULL
484 
485 #define INV_OFFSET (-1ULL)
486 
487 static inline bool has_data_file(BlockDriverState *bs)
488 {
489     BDRVQcow2State *s = bs->opaque;
490     return (s->data_file != bs->file);
491 }
492 
493 static inline bool data_file_is_raw(BlockDriverState *bs)
494 {
495     BDRVQcow2State *s = bs->opaque;
496     return !!(s->autoclear_features & QCOW2_AUTOCLEAR_DATA_FILE_RAW);
497 }
498 
499 static inline int64_t start_of_cluster(BDRVQcow2State *s, int64_t offset)
500 {
501     return offset & ~(s->cluster_size - 1);
502 }
503 
504 static inline int64_t offset_into_cluster(BDRVQcow2State *s, int64_t offset)
505 {
506     return offset & (s->cluster_size - 1);
507 }
508 
509 static inline uint64_t size_to_clusters(BDRVQcow2State *s, uint64_t size)
510 {
511     return (size + (s->cluster_size - 1)) >> s->cluster_bits;
512 }
513 
514 static inline int64_t size_to_l1(BDRVQcow2State *s, int64_t size)
515 {
516     int shift = s->cluster_bits + s->l2_bits;
517     return (size + (1ULL << shift) - 1) >> shift;
518 }
519 
520 static inline int offset_to_l1_index(BDRVQcow2State *s, uint64_t offset)
521 {
522     return offset >> (s->l2_bits + s->cluster_bits);
523 }
524 
525 static inline int offset_to_l2_index(BDRVQcow2State *s, int64_t offset)
526 {
527     return (offset >> s->cluster_bits) & (s->l2_size - 1);
528 }
529 
530 static inline int offset_to_l2_slice_index(BDRVQcow2State *s, int64_t offset)
531 {
532     return (offset >> s->cluster_bits) & (s->l2_slice_size - 1);
533 }
534 
535 static inline int64_t qcow2_vm_state_offset(BDRVQcow2State *s)
536 {
537     return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits);
538 }
539 
540 static inline QCow2ClusterType qcow2_get_cluster_type(BlockDriverState *bs,
541                                                       uint64_t l2_entry)
542 {
543     if (l2_entry & QCOW_OFLAG_COMPRESSED) {
544         return QCOW2_CLUSTER_COMPRESSED;
545     } else if (l2_entry & QCOW_OFLAG_ZERO) {
546         if (l2_entry & L2E_OFFSET_MASK) {
547             return QCOW2_CLUSTER_ZERO_ALLOC;
548         }
549         return QCOW2_CLUSTER_ZERO_PLAIN;
550     } else if (!(l2_entry & L2E_OFFSET_MASK)) {
551         /* Offset 0 generally means unallocated, but it is ambiguous with
552          * external data files because 0 is a valid offset there. However, all
553          * clusters in external data files always have refcount 1, so we can
554          * rely on QCOW_OFLAG_COPIED to disambiguate. */
555         if (has_data_file(bs) && (l2_entry & QCOW_OFLAG_COPIED)) {
556             return QCOW2_CLUSTER_NORMAL;
557         } else {
558             return QCOW2_CLUSTER_UNALLOCATED;
559         }
560     } else {
561         return QCOW2_CLUSTER_NORMAL;
562     }
563 }
564 
565 /* Check whether refcounts are eager or lazy */
566 static inline bool qcow2_need_accurate_refcounts(BDRVQcow2State *s)
567 {
568     return !(s->incompatible_features & QCOW2_INCOMPAT_DIRTY);
569 }
570 
571 static inline uint64_t l2meta_cow_start(QCowL2Meta *m)
572 {
573     return m->offset + m->cow_start.offset;
574 }
575 
576 static inline uint64_t l2meta_cow_end(QCowL2Meta *m)
577 {
578     return m->offset + m->cow_end.offset + m->cow_end.nb_bytes;
579 }
580 
581 static inline uint64_t refcount_diff(uint64_t r1, uint64_t r2)
582 {
583     return r1 > r2 ? r1 - r2 : r2 - r1;
584 }
585 
586 static inline
587 uint32_t offset_to_reftable_index(BDRVQcow2State *s, uint64_t offset)
588 {
589     return offset >> (s->refcount_block_bits + s->cluster_bits);
590 }
591 
592 /* qcow2.c functions */
593 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size,
594                                      int refcount_order, bool generous_increase,
595                                      uint64_t *refblock_count);
596 
597 int qcow2_mark_dirty(BlockDriverState *bs);
598 int qcow2_mark_corrupt(BlockDriverState *bs);
599 int qcow2_mark_consistent(BlockDriverState *bs);
600 int qcow2_update_header(BlockDriverState *bs);
601 
602 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
603                              int64_t size, const char *message_format, ...)
604                              GCC_FMT_ATTR(5, 6);
605 
606 int qcow2_validate_table(BlockDriverState *bs, uint64_t offset,
607                          uint64_t entries, size_t entry_len,
608                          int64_t max_size_bytes, const char *table_name,
609                          Error **errp);
610 
611 /* qcow2-refcount.c functions */
612 int qcow2_refcount_init(BlockDriverState *bs);
613 void qcow2_refcount_close(BlockDriverState *bs);
614 
615 int qcow2_get_refcount(BlockDriverState *bs, int64_t cluster_index,
616                        uint64_t *refcount);
617 
618 int qcow2_update_cluster_refcount(BlockDriverState *bs, int64_t cluster_index,
619                                   uint64_t addend, bool decrease,
620                                   enum qcow2_discard_type type);
621 
622 int64_t qcow2_refcount_area(BlockDriverState *bs, uint64_t offset,
623                             uint64_t additional_clusters, bool exact_size,
624                             int new_refblock_index,
625                             uint64_t new_refblock_offset);
626 
627 int64_t qcow2_alloc_clusters(BlockDriverState *bs, uint64_t size);
628 int64_t qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset,
629                                 int64_t nb_clusters);
630 int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size);
631 void qcow2_free_clusters(BlockDriverState *bs,
632                           int64_t offset, int64_t size,
633                           enum qcow2_discard_type type);
634 void qcow2_free_any_clusters(BlockDriverState *bs, uint64_t l2_entry,
635                              int nb_clusters, enum qcow2_discard_type type);
636 
637 int qcow2_update_snapshot_refcount(BlockDriverState *bs,
638     int64_t l1_table_offset, int l1_size, int addend);
639 
640 int coroutine_fn qcow2_flush_caches(BlockDriverState *bs);
641 int coroutine_fn qcow2_write_caches(BlockDriverState *bs);
642 int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
643                           BdrvCheckMode fix);
644 
645 void qcow2_process_discards(BlockDriverState *bs, int ret);
646 
647 int qcow2_check_metadata_overlap(BlockDriverState *bs, int ign, int64_t offset,
648                                  int64_t size);
649 int qcow2_pre_write_overlap_check(BlockDriverState *bs, int ign, int64_t offset,
650                                   int64_t size, bool data_file);
651 int qcow2_inc_refcounts_imrt(BlockDriverState *bs, BdrvCheckResult *res,
652                              void **refcount_table,
653                              int64_t *refcount_table_size,
654                              int64_t offset, int64_t size);
655 
656 int qcow2_change_refcount_order(BlockDriverState *bs, int refcount_order,
657                                 BlockDriverAmendStatusCB *status_cb,
658                                 void *cb_opaque, Error **errp);
659 int qcow2_shrink_reftable(BlockDriverState *bs);
660 int64_t qcow2_get_last_cluster(BlockDriverState *bs, int64_t size);
661 int qcow2_detect_metadata_preallocation(BlockDriverState *bs);
662 
663 /* qcow2-cluster.c functions */
664 int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size,
665                         bool exact_size);
666 int qcow2_shrink_l1_table(BlockDriverState *bs, uint64_t max_size);
667 int qcow2_write_l1_entry(BlockDriverState *bs, int l1_index);
668 int qcow2_encrypt_sectors(BDRVQcow2State *s, int64_t sector_num,
669                           uint8_t *buf, int nb_sectors, bool enc, Error **errp);
670 
671 int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
672                              unsigned int *bytes, uint64_t *cluster_offset);
673 int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset,
674                                unsigned int *bytes, uint64_t *host_offset,
675                                QCowL2Meta **m);
676 int qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs,
677                                           uint64_t offset,
678                                           int compressed_size,
679                                           uint64_t *host_offset);
680 
681 int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m);
682 void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m);
683 int qcow2_cluster_discard(BlockDriverState *bs, uint64_t offset,
684                           uint64_t bytes, enum qcow2_discard_type type,
685                           bool full_discard);
686 int qcow2_cluster_zeroize(BlockDriverState *bs, uint64_t offset,
687                           uint64_t bytes, int flags);
688 
689 int qcow2_expand_zero_clusters(BlockDriverState *bs,
690                                BlockDriverAmendStatusCB *status_cb,
691                                void *cb_opaque);
692 
693 /* qcow2-snapshot.c functions */
694 int qcow2_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info);
695 int qcow2_snapshot_goto(BlockDriverState *bs, const char *snapshot_id);
696 int qcow2_snapshot_delete(BlockDriverState *bs,
697                           const char *snapshot_id,
698                           const char *name,
699                           Error **errp);
700 int qcow2_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab);
701 int qcow2_snapshot_load_tmp(BlockDriverState *bs,
702                             const char *snapshot_id,
703                             const char *name,
704                             Error **errp);
705 
706 void qcow2_free_snapshots(BlockDriverState *bs);
707 int qcow2_read_snapshots(BlockDriverState *bs);
708 
709 /* qcow2-cache.c functions */
710 Qcow2Cache *qcow2_cache_create(BlockDriverState *bs, int num_tables,
711                                unsigned table_size);
712 int qcow2_cache_destroy(Qcow2Cache *c);
713 
714 void qcow2_cache_entry_mark_dirty(Qcow2Cache *c, void *table);
715 int qcow2_cache_flush(BlockDriverState *bs, Qcow2Cache *c);
716 int qcow2_cache_write(BlockDriverState *bs, Qcow2Cache *c);
717 int qcow2_cache_set_dependency(BlockDriverState *bs, Qcow2Cache *c,
718     Qcow2Cache *dependency);
719 void qcow2_cache_depends_on_flush(Qcow2Cache *c);
720 
721 void qcow2_cache_clean_unused(Qcow2Cache *c);
722 int qcow2_cache_empty(BlockDriverState *bs, Qcow2Cache *c);
723 
724 int qcow2_cache_get(BlockDriverState *bs, Qcow2Cache *c, uint64_t offset,
725     void **table);
726 int qcow2_cache_get_empty(BlockDriverState *bs, Qcow2Cache *c, uint64_t offset,
727     void **table);
728 void qcow2_cache_put(Qcow2Cache *c, void **table);
729 void *qcow2_cache_is_table_offset(Qcow2Cache *c, uint64_t offset);
730 void qcow2_cache_discard(Qcow2Cache *c, void *table);
731 
732 /* qcow2-bitmap.c functions */
733 int qcow2_check_bitmaps_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
734                                   void **refcount_table,
735                                   int64_t *refcount_table_size);
736 bool qcow2_load_dirty_bitmaps(BlockDriverState *bs, Error **errp);
737 Qcow2BitmapInfoList *qcow2_get_bitmap_info_list(BlockDriverState *bs,
738                                                 Error **errp);
739 int qcow2_reopen_bitmaps_rw_hint(BlockDriverState *bs, bool *header_updated,
740                                  Error **errp);
741 int qcow2_reopen_bitmaps_rw(BlockDriverState *bs, Error **errp);
742 int qcow2_truncate_bitmaps_check(BlockDriverState *bs, Error **errp);
743 void qcow2_store_persistent_dirty_bitmaps(BlockDriverState *bs, Error **errp);
744 int qcow2_reopen_bitmaps_ro(BlockDriverState *bs, Error **errp);
745 bool qcow2_can_store_new_dirty_bitmap(BlockDriverState *bs,
746                                       const char *name,
747                                       uint32_t granularity,
748                                       Error **errp);
749 void qcow2_remove_persistent_dirty_bitmap(BlockDriverState *bs,
750                                           const char *name,
751                                           Error **errp);
752 
753 ssize_t coroutine_fn
754 qcow2_co_compress(BlockDriverState *bs, void *dest, size_t dest_size,
755                   const void *src, size_t src_size);
756 ssize_t coroutine_fn
757 qcow2_co_decompress(BlockDriverState *bs, void *dest, size_t dest_size,
758                     const void *src, size_t src_size);
759 int coroutine_fn
760 qcow2_co_encrypt(BlockDriverState *bs, uint64_t file_cluster_offset,
761                  uint64_t offset, void *buf, size_t len);
762 int coroutine_fn
763 qcow2_co_decrypt(BlockDriverState *bs, uint64_t file_cluster_offset,
764                  uint64_t offset, void *buf, size_t len);
765 
766 #endif
767