block/qcow2.h

   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 /* Maximum amount of extra data per snapshot table entry to accept */
  65 #define QCOW_MAX_SNAPSHOT_EXTRA_DATA 1024
  66
  67 /* Bitmap header extension constraints */
  68 #define QCOW2_MAX_BITMAPS 65535
  69 #define QCOW2_MAX_BITMAP_DIRECTORY_SIZE (1024 * QCOW2_MAX_BITMAPS)
  70
  71 /* Maximum of parallel sub-request per guest request */
  72 #define QCOW2_MAX_WORKERS 8
  73
  74 /* indicate that the refcount of the referenced cluster is exactly one. */
  75 #define QCOW_OFLAG_COPIED     (1ULL << 63)
  76 /* indicate that the cluster is compressed (they never have the copied flag) */
  77 #define QCOW_OFLAG_COMPRESSED (1ULL << 62)
  78 /* The cluster reads as all zeros */
  79 #define QCOW_OFLAG_ZERO (1ULL << 0)
  80
  81 #define MIN_CLUSTER_BITS 9
  82 #define MAX_CLUSTER_BITS 21
  83
  84 /* Defined in the qcow2 spec (compressed cluster descriptor) */
  85 #define QCOW2_COMPRESSED_SECTOR_SIZE 512U
  86 #define QCOW2_COMPRESSED_SECTOR_MASK (~(QCOW2_COMPRESSED_SECTOR_SIZE - 1ULL))
  87
  88 /* Must be at least 2 to cover COW */
  89 #define MIN_L2_CACHE_SIZE 2 /* cache entries */
  90
  91 /* Must be at least 4 to cover all cases of refcount table growth */
  92 #define MIN_REFCOUNT_CACHE_SIZE 4 /* clusters */
  93
  94 #ifdef CONFIG_LINUX
  95 #define DEFAULT_L2_CACHE_MAX_SIZE (32 * MiB)
  96 #define DEFAULT_CACHE_CLEAN_INTERVAL 600  /* seconds */
  97 #else
  98 #define DEFAULT_L2_CACHE_MAX_SIZE (8 * MiB)
  99 /* Cache clean interval is currently available only on Linux, so must be 0 */
 100 #define DEFAULT_CACHE_CLEAN_INTERVAL 0
 101 #endif
 102
 103 #define DEFAULT_CLUSTER_SIZE 65536
 104
 105 #define QCOW2_OPT_DATA_FILE "data-file"
 106 #define QCOW2_OPT_LAZY_REFCOUNTS "lazy-refcounts"
 107 #define QCOW2_OPT_DISCARD_REQUEST "pass-discard-request"
 108 #define QCOW2_OPT_DISCARD_SNAPSHOT "pass-discard-snapshot"
 109 #define QCOW2_OPT_DISCARD_OTHER "pass-discard-other"
 110 #define QCOW2_OPT_OVERLAP "overlap-check"
 111 #define QCOW2_OPT_OVERLAP_TEMPLATE "overlap-check.template"
 112 #define QCOW2_OPT_OVERLAP_MAIN_HEADER "overlap-check.main-header"
 113 #define QCOW2_OPT_OVERLAP_ACTIVE_L1 "overlap-check.active-l1"
 114 #define QCOW2_OPT_OVERLAP_ACTIVE_L2 "overlap-check.active-l2"
 115 #define QCOW2_OPT_OVERLAP_REFCOUNT_TABLE "overlap-check.refcount-table"
 116 #define QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK "overlap-check.refcount-block"
 117 #define QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE "overlap-check.snapshot-table"
 118 #define QCOW2_OPT_OVERLAP_INACTIVE_L1 "overlap-check.inactive-l1"
 119 #define QCOW2_OPT_OVERLAP_INACTIVE_L2 "overlap-check.inactive-l2"
 120 #define QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY "overlap-check.bitmap-directory"
 121 #define QCOW2_OPT_CACHE_SIZE "cache-size"
 122 #define QCOW2_OPT_L2_CACHE_SIZE "l2-cache-size"
 123 #define QCOW2_OPT_L2_CACHE_ENTRY_SIZE "l2-cache-entry-size"
 124 #define QCOW2_OPT_REFCOUNT_CACHE_SIZE "refcount-cache-size"
 125 #define QCOW2_OPT_CACHE_CLEAN_INTERVAL "cache-clean-interval"
 126
 127 typedef struct QCowHeader {
 128     uint32_t magic;
 129     uint32_t version;
 130     uint64_t backing_file_offset;
 131     uint32_t backing_file_size;
 132     uint32_t cluster_bits;
 133     uint64_t size; /* in bytes */
 134     uint32_t crypt_method;
 135     uint32_t l1_size; /* XXX: save number of clusters instead ? */
 136     uint64_t l1_table_offset;
 137     uint64_t refcount_table_offset;
 138     uint32_t refcount_table_clusters;
 139     uint32_t nb_snapshots;
 140     uint64_t snapshots_offset;
 141
 142     /* The following fields are only valid for version >= 3 */
 143     uint64_t incompatible_features;
 144     uint64_t compatible_features;
 145     uint64_t autoclear_features;
 146
 147     uint32_t refcount_order;
 148     uint32_t header_length;
 149
 150     /* Additional fields */
 151     uint8_t compression_type;
 152
 153     /* header must be a multiple of 8 */
 154     uint8_t padding[7];
 155 } QEMU_PACKED QCowHeader;
 156
 157 QEMU_BUILD_BUG_ON(!QEMU_IS_ALIGNED(sizeof(QCowHeader), 8));
 158
 159 typedef struct QEMU_PACKED QCowSnapshotHeader {
 160     /* header is 8 byte aligned */
 161     uint64_t l1_table_offset;
 162
 163     uint32_t l1_size;
 164     uint16_t id_str_size;
 165     uint16_t name_size;
 166
 167     uint32_t date_sec;
 168     uint32_t date_nsec;
 169
 170     uint64_t vm_clock_nsec;
 171
 172     uint32_t vm_state_size;
 173     uint32_t extra_data_size; /* for extension */
 174     /* extra data follows */
 175     /* id_str follows */
 176     /* name follows  */
 177 } QCowSnapshotHeader;
 178
 179 typedef struct QEMU_PACKED QCowSnapshotExtraData {
 180     uint64_t vm_state_size_large;
 181     uint64_t disk_size;
 182 } QCowSnapshotExtraData;
 183
 184
 185 typedef struct QCowSnapshot {
 186     uint64_t l1_table_offset;
 187     uint32_t l1_size;
 188     char *id_str;
 189     char *name;
 190     uint64_t disk_size;
 191     uint64_t vm_state_size;
 192     uint32_t date_sec;
 193     uint32_t date_nsec;
 194     uint64_t vm_clock_nsec;
 195     /* Size of all extra data, including QCowSnapshotExtraData if available */
 196     uint32_t extra_data_size;
 197     /* Data beyond QCowSnapshotExtraData, if any */
 198     void *unknown_extra_data;
 199 } QCowSnapshot;
 200
 201 struct Qcow2Cache;
 202 typedef struct Qcow2Cache Qcow2Cache;
 203
 204 typedef struct Qcow2CryptoHeaderExtension {
 205     uint64_t offset;
 206     uint64_t length;
 207 } QEMU_PACKED Qcow2CryptoHeaderExtension;
 208
 209 typedef struct Qcow2UnknownHeaderExtension {
 210     uint32_t magic;
 211     uint32_t len;
 212     QLIST_ENTRY(Qcow2UnknownHeaderExtension) next;
 213     uint8_t data[];
 214 } Qcow2UnknownHeaderExtension;
 215
 216 enum {
 217     QCOW2_FEAT_TYPE_INCOMPATIBLE    = 0,
 218     QCOW2_FEAT_TYPE_COMPATIBLE      = 1,
 219     QCOW2_FEAT_TYPE_AUTOCLEAR       = 2,
 220 };
 221
 222 /* Incompatible feature bits */
 223 enum {
 224     QCOW2_INCOMPAT_DIRTY_BITNR      = 0,
 225     QCOW2_INCOMPAT_CORRUPT_BITNR    = 1,
 226     QCOW2_INCOMPAT_DATA_FILE_BITNR  = 2,
 227     QCOW2_INCOMPAT_COMPRESSION_BITNR = 3,
 228     QCOW2_INCOMPAT_DIRTY            = 1 << QCOW2_INCOMPAT_DIRTY_BITNR,
 229     QCOW2_INCOMPAT_CORRUPT          = 1 << QCOW2_INCOMPAT_CORRUPT_BITNR,
 230     QCOW2_INCOMPAT_DATA_FILE        = 1 << QCOW2_INCOMPAT_DATA_FILE_BITNR,
 231     QCOW2_INCOMPAT_COMPRESSION      = 1 << QCOW2_INCOMPAT_COMPRESSION_BITNR,
 232
 233     QCOW2_INCOMPAT_MASK             = QCOW2_INCOMPAT_DIRTY
 234                                     | QCOW2_INCOMPAT_CORRUPT
 235                                     | QCOW2_INCOMPAT_DATA_FILE
 236                                     | QCOW2_INCOMPAT_COMPRESSION,
 237 };
 238
 239 /* Compatible feature bits */
 240 enum {
 241     QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR = 0,
 242     QCOW2_COMPAT_LAZY_REFCOUNTS       = 1 << QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
 243
 244     QCOW2_COMPAT_FEAT_MASK            = QCOW2_COMPAT_LAZY_REFCOUNTS,
 245 };
 246
 247 /* Autoclear feature bits */
 248 enum {
 249     QCOW2_AUTOCLEAR_BITMAPS_BITNR       = 0,
 250     QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR = 1,
 251     QCOW2_AUTOCLEAR_BITMAPS             = 1 << QCOW2_AUTOCLEAR_BITMAPS_BITNR,
 252     QCOW2_AUTOCLEAR_DATA_FILE_RAW       = 1 << QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR,
 253
 254     QCOW2_AUTOCLEAR_MASK                = QCOW2_AUTOCLEAR_BITMAPS
 255                                         | QCOW2_AUTOCLEAR_DATA_FILE_RAW,
 256 };
 257
 258 enum qcow2_discard_type {
 259     QCOW2_DISCARD_NEVER = 0,
 260     QCOW2_DISCARD_ALWAYS,
 261     QCOW2_DISCARD_REQUEST,
 262     QCOW2_DISCARD_SNAPSHOT,
 263     QCOW2_DISCARD_OTHER,
 264     QCOW2_DISCARD_MAX
 265 };
 266
 267 typedef struct Qcow2Feature {
 268     uint8_t type;
 269     uint8_t bit;
 270     char    name[46];
 271 } QEMU_PACKED Qcow2Feature;
 272
 273 typedef struct Qcow2DiscardRegion {
 274     BlockDriverState *bs;
 275     uint64_t offset;
 276     uint64_t bytes;
 277     QTAILQ_ENTRY(Qcow2DiscardRegion) next;
 278 } Qcow2DiscardRegion;
 279
 280 typedef uint64_t Qcow2GetRefcountFunc(const void *refcount_array,
 281                                       uint64_t index);
 282 typedef void Qcow2SetRefcountFunc(void *refcount_array,
 283                                   uint64_t index, uint64_t value);
 284
 285 typedef struct Qcow2BitmapHeaderExt {
 286     uint32_t nb_bitmaps;
 287     uint32_t reserved32;
 288     uint64_t bitmap_directory_size;
 289     uint64_t bitmap_directory_offset;
 290 } QEMU_PACKED Qcow2BitmapHeaderExt;
 291
 292 #define QCOW2_MAX_THREADS 4
 293
 294 typedef struct BDRVQcow2State {
 295     int cluster_bits;
 296     int cluster_size;
 297     int l2_slice_size;
 298     int l2_bits;
 299     int l2_size;
 300     int l1_size;
 301     int l1_vm_state_index;
 302     int refcount_block_bits;
 303     int refcount_block_size;
 304     int csize_shift;
 305     int csize_mask;
 306     uint64_t cluster_offset_mask;
 307     uint64_t l1_table_offset;
 308     uint64_t *l1_table;
 309
 310     Qcow2Cache* l2_table_cache;
 311     Qcow2Cache* refcount_block_cache;
 312     QEMUTimer *cache_clean_timer;
 313     unsigned cache_clean_interval;
 314
 315     QLIST_HEAD(, QCowL2Meta) cluster_allocs;
 316
 317     uint64_t *refcount_table;
 318     uint64_t refcount_table_offset;
 319     uint32_t refcount_table_size;
 320     uint32_t max_refcount_table_index; /* Last used entry in refcount_table */
 321     uint64_t free_cluster_index;
 322     uint64_t free_byte_offset;
 323
 324     CoMutex lock;
 325
 326     Qcow2CryptoHeaderExtension crypto_header; /* QCow2 header extension */
 327     QCryptoBlockOpenOptions *crypto_opts; /* Disk encryption runtime options */
 328     QCryptoBlock *crypto; /* Disk encryption format driver */
 329     bool crypt_physical_offset; /* Whether to use virtual or physical offset
 330                                    for encryption initialization vector tweak */
 331     uint32_t crypt_method_header;
 332     uint64_t snapshots_offset;
 333     int snapshots_size;
 334     unsigned int nb_snapshots;
 335     QCowSnapshot *snapshots;
 336
 337     uint32_t nb_bitmaps;
 338     uint64_t bitmap_directory_size;
 339     uint64_t bitmap_directory_offset;
 340
 341     int flags;
 342     int qcow_version;
 343     bool use_lazy_refcounts;
 344     int refcount_order;
 345     int refcount_bits;
 346     uint64_t refcount_max;
 347
 348     Qcow2GetRefcountFunc *get_refcount;
 349     Qcow2SetRefcountFunc *set_refcount;
 350
 351     bool discard_passthrough[QCOW2_DISCARD_MAX];
 352
 353     int overlap_check; /* bitmask of Qcow2MetadataOverlap values */
 354     bool signaled_corruption;
 355
 356     uint64_t incompatible_features;
 357     uint64_t compatible_features;
 358     uint64_t autoclear_features;
 359
 360     size_t unknown_header_fields_size;
 361     void* unknown_header_fields;
 362     QLIST_HEAD(, Qcow2UnknownHeaderExtension) unknown_header_ext;
 363     QTAILQ_HEAD (, Qcow2DiscardRegion) discards;
 364     bool cache_discards;
 365
 366     /* Backing file path and format as stored in the image (this is not the
 367      * effective path/format, which may be the result of a runtime option
 368      * override) */
 369     char *image_backing_file;
 370     char *image_backing_format;
 371     char *image_data_file;
 372
 373     CoQueue thread_task_queue;
 374     int nb_threads;
 375
 376     BdrvChild *data_file;
 377
 378     bool metadata_preallocation_checked;
 379     bool metadata_preallocation;
 380     /*
 381      * Compression type used for the image. Default: 0 - ZLIB
 382      * The image compression type is set on image creation.
 383      * For now, the only way to change the compression type
 384      * is to convert the image with the desired compression type set.
 385      */
 386     Qcow2CompressionType compression_type;
 387 } BDRVQcow2State;
 388
 389 typedef struct Qcow2COWRegion {
 390     /**
 391      * Offset of the COW region in bytes from the start of the first cluster
 392      * touched by the request.
 393      */
 394     unsigned    offset;
 395
 396     /** Number of bytes to copy */
 397     unsigned    nb_bytes;
 398 } Qcow2COWRegion;
 399
 400 /**
 401  * Describes an in-flight (part of a) write request that writes to clusters
 402  * that are not referenced in their L2 table yet.
 403  */
 404 typedef struct QCowL2Meta
 405 {
 406     /** Guest offset of the first newly allocated cluster */
 407     uint64_t offset;
 408
 409     /** Host offset of the first newly allocated cluster */
 410     uint64_t alloc_offset;
 411
 412     /** Number of newly allocated clusters */
 413     int nb_clusters;
 414
 415     /** Do not free the old clusters */
 416     bool keep_old_clusters;
 417
 418     /**
 419      * Requests that overlap with this allocation and wait to be restarted
 420      * when the allocating request has completed.
 421      */
 422     CoQueue dependent_requests;
 423
 424     /**
 425      * The COW Region between the start of the first allocated cluster and the
 426      * area the guest actually writes to.
 427      */
 428     Qcow2COWRegion cow_start;
 429
 430     /**
 431      * The COW Region between the area the guest actually writes to and the
 432      * end of the last allocated cluster.
 433      */
 434     Qcow2COWRegion cow_end;
 435
 436     /*
 437      * Indicates that COW regions are already handled and do not require
 438      * any more processing.
 439      */
 440     bool skip_cow;
 441
 442     /**
 443      * The I/O vector with the data from the actual guest write request.
 444      * If non-NULL, this is meant to be merged together with the data
 445      * from @cow_start and @cow_end into one single write operation.
 446      */
 447     QEMUIOVector *data_qiov;
 448     size_t data_qiov_offset;
 449
 450     /** Pointer to next L2Meta of the same write request */
 451     struct QCowL2Meta *next;
 452
 453     QLIST_ENTRY(QCowL2Meta) next_in_flight;
 454 } QCowL2Meta;
 455
 456 typedef enum QCow2ClusterType {
 457     QCOW2_CLUSTER_UNALLOCATED,
 458     QCOW2_CLUSTER_ZERO_PLAIN,
 459     QCOW2_CLUSTER_ZERO_ALLOC,
 460     QCOW2_CLUSTER_NORMAL,
 461     QCOW2_CLUSTER_COMPRESSED,
 462 } QCow2ClusterType;
 463
 464 typedef enum QCow2MetadataOverlap {
 465     QCOW2_OL_MAIN_HEADER_BITNR      = 0,
 466     QCOW2_OL_ACTIVE_L1_BITNR        = 1,
 467     QCOW2_OL_ACTIVE_L2_BITNR        = 2,
 468     QCOW2_OL_REFCOUNT_TABLE_BITNR   = 3,
 469     QCOW2_OL_REFCOUNT_BLOCK_BITNR   = 4,
 470     QCOW2_OL_SNAPSHOT_TABLE_BITNR   = 5,
 471     QCOW2_OL_INACTIVE_L1_BITNR      = 6,
 472     QCOW2_OL_INACTIVE_L2_BITNR      = 7,
 473     QCOW2_OL_BITMAP_DIRECTORY_BITNR = 8,
 474
 475     QCOW2_OL_MAX_BITNR              = 9,
 476
 477     QCOW2_OL_NONE             = 0,
 478     QCOW2_OL_MAIN_HEADER      = (1 << QCOW2_OL_MAIN_HEADER_BITNR),
 479     QCOW2_OL_ACTIVE_L1        = (1 << QCOW2_OL_ACTIVE_L1_BITNR),
 480     QCOW2_OL_ACTIVE_L2        = (1 << QCOW2_OL_ACTIVE_L2_BITNR),
 481     QCOW2_OL_REFCOUNT_TABLE   = (1 << QCOW2_OL_REFCOUNT_TABLE_BITNR),
 482     QCOW2_OL_REFCOUNT_BLOCK   = (1 << QCOW2_OL_REFCOUNT_BLOCK_BITNR),
 483     QCOW2_OL_SNAPSHOT_TABLE   = (1 << QCOW2_OL_SNAPSHOT_TABLE_BITNR),
 484     QCOW2_OL_INACTIVE_L1      = (1 << QCOW2_OL_INACTIVE_L1_BITNR),
 485     /* NOTE: Checking overlaps with inactive L2 tables will result in bdrv
 486      * reads. */
 487     QCOW2_OL_INACTIVE_L2      = (1 << QCOW2_OL_INACTIVE_L2_BITNR),
 488     QCOW2_OL_BITMAP_DIRECTORY = (1 << QCOW2_OL_BITMAP_DIRECTORY_BITNR),
 489 } QCow2MetadataOverlap;
 490
 491 /* Perform all overlap checks which can be done in constant time */
 492 #define QCOW2_OL_CONSTANT \
 493     (QCOW2_OL_MAIN_HEADER | QCOW2_OL_ACTIVE_L1 | QCOW2_OL_REFCOUNT_TABLE | \
 494      QCOW2_OL_SNAPSHOT_TABLE | QCOW2_OL_BITMAP_DIRECTORY)
 495
 496 /* Perform all overlap checks which don't require disk access */
 497 #define QCOW2_OL_CACHED \
 498     (QCOW2_OL_CONSTANT | QCOW2_OL_ACTIVE_L2 | QCOW2_OL_REFCOUNT_BLOCK | \
 499      QCOW2_OL_INACTIVE_L1)
 500
 501 /* Perform all overlap checks */
 502 #define QCOW2_OL_ALL \
 503     (QCOW2_OL_CACHED | QCOW2_OL_INACTIVE_L2)
 504
 505 #define L1E_OFFSET_MASK 0x00fffffffffffe00ULL
 506 #define L2E_OFFSET_MASK 0x00fffffffffffe00ULL
 507 #define L2E_COMPRESSED_OFFSET_SIZE_MASK 0x3fffffffffffffffULL
 508
 509 #define REFT_OFFSET_MASK 0xfffffffffffffe00ULL
 510
 511 #define INV_OFFSET (-1ULL)
 512
 513 static inline bool has_data_file(BlockDriverState *bs)
 514 {
 515     BDRVQcow2State *s = bs->opaque;
 516     return (s->data_file != bs->file);
 517 }
 518
 519 static inline bool data_file_is_raw(BlockDriverState *bs)
 520 {
 521     BDRVQcow2State *s = bs->opaque;
 522     return !!(s->autoclear_features & QCOW2_AUTOCLEAR_DATA_FILE_RAW);
 523 }
 524
 525 static inline int64_t start_of_cluster(BDRVQcow2State *s, int64_t offset)
 526 {
 527     return offset & ~(s->cluster_size - 1);
 528 }
 529
 530 static inline int64_t offset_into_cluster(BDRVQcow2State *s, int64_t offset)
 531 {
 532     return offset & (s->cluster_size - 1);
 533 }
 534
 535 static inline uint64_t size_to_clusters(BDRVQcow2State *s, uint64_t size)
 536 {
 537     return (size + (s->cluster_size - 1)) >> s->cluster_bits;
 538 }
 539
 540 static inline int64_t size_to_l1(BDRVQcow2State *s, int64_t size)
 541 {
 542     int shift = s->cluster_bits + s->l2_bits;
 543     return (size + (1ULL << shift) - 1) >> shift;
 544 }
 545
 546 static inline int offset_to_l1_index(BDRVQcow2State *s, uint64_t offset)
 547 {
 548     return offset >> (s->l2_bits + s->cluster_bits);
 549 }
 550
 551 static inline int offset_to_l2_index(BDRVQcow2State *s, int64_t offset)
 552 {
 553     return (offset >> s->cluster_bits) & (s->l2_size - 1);
 554 }
 555
 556 static inline int offset_to_l2_slice_index(BDRVQcow2State *s, int64_t offset)
 557 {
 558     return (offset >> s->cluster_bits) & (s->l2_slice_size - 1);
 559 }
 560
 561 static inline int64_t qcow2_vm_state_offset(BDRVQcow2State *s)
 562 {
 563     return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits);
 564 }
 565
 566 static inline QCow2ClusterType qcow2_get_cluster_type(BlockDriverState *bs,
 567                                                       uint64_t l2_entry)
 568 {
 569     if (l2_entry & QCOW_OFLAG_COMPRESSED) {
 570         return QCOW2_CLUSTER_COMPRESSED;
 571     } else if (l2_entry & QCOW_OFLAG_ZERO) {
 572         if (l2_entry & L2E_OFFSET_MASK) {
 573             return QCOW2_CLUSTER_ZERO_ALLOC;
 574         }
 575         return QCOW2_CLUSTER_ZERO_PLAIN;
 576     } else if (!(l2_entry & L2E_OFFSET_MASK)) {
 577         /* Offset 0 generally means unallocated, but it is ambiguous with
 578          * external data files because 0 is a valid offset there. However, all
 579          * clusters in external data files always have refcount 1, so we can
 580          * rely on QCOW_OFLAG_COPIED to disambiguate. */
 581         if (has_data_file(bs) && (l2_entry & QCOW_OFLAG_COPIED)) {
 582             return QCOW2_CLUSTER_NORMAL;
 583         } else {
 584             return QCOW2_CLUSTER_UNALLOCATED;
 585         }
 586     } else {
 587         return QCOW2_CLUSTER_NORMAL;
 588     }
 589 }
 590
 591 /* Check whether refcounts are eager or lazy */
 592 static inline bool qcow2_need_accurate_refcounts(BDRVQcow2State *s)
 593 {
 594     return !(s->incompatible_features & QCOW2_INCOMPAT_DIRTY);
 595 }
 596
 597 static inline uint64_t l2meta_cow_start(QCowL2Meta *m)
 598 {
 599     return m->offset + m->cow_start.offset;
 600 }
 601
 602 static inline uint64_t l2meta_cow_end(QCowL2Meta *m)
 603 {
 604     return m->offset + m->cow_end.offset + m->cow_end.nb_bytes;
 605 }
 606
 607 static inline uint64_t refcount_diff(uint64_t r1, uint64_t r2)
 608 {
 609     return r1 > r2 ? r1 - r2 : r2 - r1;
 610 }
 611
 612 static inline
 613 uint32_t offset_to_reftable_index(BDRVQcow2State *s, uint64_t offset)
 614 {
 615     return offset >> (s->refcount_block_bits + s->cluster_bits);
 616 }
 617
 618 /* qcow2.c functions */
 619 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size,
 620                                      int refcount_order, bool generous_increase,
 621                                      uint64_t *refblock_count);
 622
 623 int qcow2_mark_dirty(BlockDriverState *bs);
 624 int qcow2_mark_corrupt(BlockDriverState *bs);
 625 int qcow2_mark_consistent(BlockDriverState *bs);
 626 int qcow2_update_header(BlockDriverState *bs);
 627
 628 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
 629                              int64_t size, const char *message_format, ...)
 630                              GCC_FMT_ATTR(5, 6);
 631
 632 int qcow2_validate_table(BlockDriverState *bs, uint64_t offset,
 633                          uint64_t entries, size_t entry_len,
 634                          int64_t max_size_bytes, const char *table_name,
 635                          Error **errp);
 636
 637 /* qcow2-refcount.c functions */
 638 int qcow2_refcount_init(BlockDriverState *bs);
 639 void qcow2_refcount_close(BlockDriverState *bs);
 640
 641 int qcow2_get_refcount(BlockDriverState *bs, int64_t cluster_index,
 642                        uint64_t *refcount);
 643
 644 int qcow2_update_cluster_refcount(BlockDriverState *bs, int64_t cluster_index,
 645                                   uint64_t addend, bool decrease,
 646                                   enum qcow2_discard_type type);
 647
 648 int64_t qcow2_refcount_area(BlockDriverState *bs, uint64_t offset,
 649                             uint64_t additional_clusters, bool exact_size,
 650                             int new_refblock_index,
 651                             uint64_t new_refblock_offset);
 652
 653 int64_t qcow2_alloc_clusters(BlockDriverState *bs, uint64_t size);
 654 int64_t qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset,
 655                                 int64_t nb_clusters);
 656 int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size);
 657 void qcow2_free_clusters(BlockDriverState *bs,
 658                           int64_t offset, int64_t size,
 659                           enum qcow2_discard_type type);
 660 void qcow2_free_any_clusters(BlockDriverState *bs, uint64_t l2_entry,
 661                              int nb_clusters, enum qcow2_discard_type type);
 662
 663 int qcow2_update_snapshot_refcount(BlockDriverState *bs,
 664     int64_t l1_table_offset, int l1_size, int addend);
 665
 666 int coroutine_fn qcow2_flush_caches(BlockDriverState *bs);
 667 int coroutine_fn qcow2_write_caches(BlockDriverState *bs);
 668 int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
 669                           BdrvCheckMode fix);
 670
 671 void qcow2_process_discards(BlockDriverState *bs, int ret);
 672
 673 int qcow2_check_metadata_overlap(BlockDriverState *bs, int ign, int64_t offset,
 674                                  int64_t size);
 675 int qcow2_pre_write_overlap_check(BlockDriverState *bs, int ign, int64_t offset,
 676                                   int64_t size, bool data_file);
 677 int qcow2_inc_refcounts_imrt(BlockDriverState *bs, BdrvCheckResult *res,
 678                              void **refcount_table,
 679                              int64_t *refcount_table_size,
 680                              int64_t offset, int64_t size);
 681
 682 int qcow2_change_refcount_order(BlockDriverState *bs, int refcount_order,
 683                                 BlockDriverAmendStatusCB *status_cb,
 684                                 void *cb_opaque, Error **errp);
 685 int qcow2_shrink_reftable(BlockDriverState *bs);
 686 int64_t qcow2_get_last_cluster(BlockDriverState *bs, int64_t size);
 687 int qcow2_detect_metadata_preallocation(BlockDriverState *bs);
 688
 689 /* qcow2-cluster.c functions */
 690 int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size,
 691                         bool exact_size);
 692 int qcow2_shrink_l1_table(BlockDriverState *bs, uint64_t max_size);
 693 int qcow2_write_l1_entry(BlockDriverState *bs, int l1_index);
 694 int qcow2_encrypt_sectors(BDRVQcow2State *s, int64_t sector_num,
 695                           uint8_t *buf, int nb_sectors, bool enc, Error **errp);
 696
 697 int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
 698                              unsigned int *bytes, uint64_t *cluster_offset);
 699 int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset,
 700                                unsigned int *bytes, uint64_t *host_offset,
 701                                QCowL2Meta **m);
 702 int qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs,
 703                                           uint64_t offset,
 704                                           int compressed_size,
 705                                           uint64_t *host_offset);
 706
 707 int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m);
 708 void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m);
 709 int qcow2_cluster_discard(BlockDriverState *bs, uint64_t offset,
 710                           uint64_t bytes, enum qcow2_discard_type type,
 711                           bool full_discard);
 712 int qcow2_cluster_zeroize(BlockDriverState *bs, uint64_t offset,
 713                           uint64_t bytes, int flags);
 714
 715 int qcow2_expand_zero_clusters(BlockDriverState *bs,
 716                                BlockDriverAmendStatusCB *status_cb,
 717                                void *cb_opaque);
 718
 719 /* qcow2-snapshot.c functions */
 720 int qcow2_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info);
 721 int qcow2_snapshot_goto(BlockDriverState *bs, const char *snapshot_id);
 722 int qcow2_snapshot_delete(BlockDriverState *bs,
 723                           const char *snapshot_id,
 724                           const char *name,
 725                           Error **errp);
 726 int qcow2_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab);
 727 int qcow2_snapshot_load_tmp(BlockDriverState *bs,
 728                             const char *snapshot_id,
 729                             const char *name,
 730                             Error **errp);
 731
 732 void qcow2_free_snapshots(BlockDriverState *bs);
 733 int qcow2_read_snapshots(BlockDriverState *bs, Error **errp);
 734 int qcow2_write_snapshots(BlockDriverState *bs);
 735
 736 int coroutine_fn qcow2_check_read_snapshot_table(BlockDriverState *bs,
 737                                                  BdrvCheckResult *result,
 738                                                  BdrvCheckMode fix);
 739 int coroutine_fn qcow2_check_fix_snapshot_table(BlockDriverState *bs,
 740                                                 BdrvCheckResult *result,
 741                                                 BdrvCheckMode fix);
 742
 743 /* qcow2-cache.c functions */
 744 Qcow2Cache *qcow2_cache_create(BlockDriverState *bs, int num_tables,
 745                                unsigned table_size);
 746 int qcow2_cache_destroy(Qcow2Cache *c);
 747
 748 void qcow2_cache_entry_mark_dirty(Qcow2Cache *c, void *table);
 749 int qcow2_cache_flush(BlockDriverState *bs, Qcow2Cache *c);
 750 int qcow2_cache_write(BlockDriverState *bs, Qcow2Cache *c);
 751 int qcow2_cache_set_dependency(BlockDriverState *bs, Qcow2Cache *c,
 752     Qcow2Cache *dependency);
 753 void qcow2_cache_depends_on_flush(Qcow2Cache *c);
 754
 755 void qcow2_cache_clean_unused(Qcow2Cache *c);
 756 int qcow2_cache_empty(BlockDriverState *bs, Qcow2Cache *c);
 757
 758 int qcow2_cache_get(BlockDriverState *bs, Qcow2Cache *c, uint64_t offset,
 759     void **table);
 760 int qcow2_cache_get_empty(BlockDriverState *bs, Qcow2Cache *c, uint64_t offset,
 761     void **table);
 762 void qcow2_cache_put(Qcow2Cache *c, void **table);
 763 void *qcow2_cache_is_table_offset(Qcow2Cache *c, uint64_t offset);
 764 void qcow2_cache_discard(Qcow2Cache *c, void *table);
 765
 766 /* qcow2-bitmap.c functions */
 767 int qcow2_check_bitmaps_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
 768                                   void **refcount_table,
 769                                   int64_t *refcount_table_size);
 770 bool qcow2_load_dirty_bitmaps(BlockDriverState *bs, Error **errp);
 771 Qcow2BitmapInfoList *qcow2_get_bitmap_info_list(BlockDriverState *bs,
 772                                                 Error **errp);
 773 int qcow2_reopen_bitmaps_rw(BlockDriverState *bs, Error **errp);
 774 int qcow2_truncate_bitmaps_check(BlockDriverState *bs, Error **errp);
 775 void qcow2_store_persistent_dirty_bitmaps(BlockDriverState *bs,
 776                                           bool release_stored, Error **errp);
 777 int qcow2_reopen_bitmaps_ro(BlockDriverState *bs, Error **errp);
 778 bool qcow2_co_can_store_new_dirty_bitmap(BlockDriverState *bs,
 779                                          const char *name,
 780                                          uint32_t granularity,
 781                                          Error **errp);
 782 int qcow2_co_remove_persistent_dirty_bitmap(BlockDriverState *bs,
 783                                             const char *name,
 784                                             Error **errp);
 785 bool qcow2_supports_persistent_dirty_bitmap(BlockDriverState *bs);
 786 uint64_t qcow2_get_persistent_dirty_bitmap_size(BlockDriverState *bs,
 787                                                 uint32_t cluster_size);
 788
 789 ssize_t coroutine_fn
 790 qcow2_co_compress(BlockDriverState *bs, void *dest, size_t dest_size,
 791                   const void *src, size_t src_size);
 792 ssize_t coroutine_fn
 793 qcow2_co_decompress(BlockDriverState *bs, void *dest, size_t dest_size,
 794                     const void *src, size_t src_size);
 795 int coroutine_fn
 796 qcow2_co_encrypt(BlockDriverState *bs, uint64_t host_offset,
 797                  uint64_t guest_offset, void *buf, size_t len);
 798 int coroutine_fn
 799 qcow2_co_decrypt(BlockDriverState *bs, uint64_t host_offset,
 800                  uint64_t guest_offset, void *buf, size_t len);
 801
 802 #endif