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 QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER 32 82 83 /* The subcluster X [0..31] is allocated */ 84 #define QCOW_OFLAG_SUB_ALLOC(X) (1ULL << (X)) 85 /* The subcluster X [0..31] reads as zeroes */ 86 #define QCOW_OFLAG_SUB_ZERO(X) (QCOW_OFLAG_SUB_ALLOC(X) << 32) 87 /* Subclusters [X, Y) (0 <= X <= Y <= 32) are allocated */ 88 #define QCOW_OFLAG_SUB_ALLOC_RANGE(X, Y) \ 89 (QCOW_OFLAG_SUB_ALLOC(Y) - QCOW_OFLAG_SUB_ALLOC(X)) 90 /* Subclusters [X, Y) (0 <= X <= Y <= 32) read as zeroes */ 91 #define QCOW_OFLAG_SUB_ZERO_RANGE(X, Y) \ 92 (QCOW_OFLAG_SUB_ALLOC_RANGE(X, Y) << 32) 93 /* L2 entry bitmap with all allocation bits set */ 94 #define QCOW_L2_BITMAP_ALL_ALLOC (QCOW_OFLAG_SUB_ALLOC_RANGE(0, 32)) 95 /* L2 entry bitmap with all "read as zeroes" bits set */ 96 #define QCOW_L2_BITMAP_ALL_ZEROES (QCOW_OFLAG_SUB_ZERO_RANGE(0, 32)) 97 98 /* Size of normal and extended L2 entries */ 99 #define L2E_SIZE_NORMAL (sizeof(uint64_t)) 100 #define L2E_SIZE_EXTENDED (sizeof(uint64_t) * 2) 101 102 #define MIN_CLUSTER_BITS 9 103 #define MAX_CLUSTER_BITS 21 104 105 /* Defined in the qcow2 spec (compressed cluster descriptor) */ 106 #define QCOW2_COMPRESSED_SECTOR_SIZE 512U 107 #define QCOW2_COMPRESSED_SECTOR_MASK (~(QCOW2_COMPRESSED_SECTOR_SIZE - 1ULL)) 108 109 /* Must be at least 2 to cover COW */ 110 #define MIN_L2_CACHE_SIZE 2 /* cache entries */ 111 112 /* Must be at least 4 to cover all cases of refcount table growth */ 113 #define MIN_REFCOUNT_CACHE_SIZE 4 /* clusters */ 114 115 #ifdef CONFIG_LINUX 116 #define DEFAULT_L2_CACHE_MAX_SIZE (32 * MiB) 117 #define DEFAULT_CACHE_CLEAN_INTERVAL 600 /* seconds */ 118 #else 119 #define DEFAULT_L2_CACHE_MAX_SIZE (8 * MiB) 120 /* Cache clean interval is currently available only on Linux, so must be 0 */ 121 #define DEFAULT_CACHE_CLEAN_INTERVAL 0 122 #endif 123 124 #define DEFAULT_CLUSTER_SIZE 65536 125 126 #define QCOW2_OPT_DATA_FILE "data-file" 127 #define QCOW2_OPT_LAZY_REFCOUNTS "lazy-refcounts" 128 #define QCOW2_OPT_DISCARD_REQUEST "pass-discard-request" 129 #define QCOW2_OPT_DISCARD_SNAPSHOT "pass-discard-snapshot" 130 #define QCOW2_OPT_DISCARD_OTHER "pass-discard-other" 131 #define QCOW2_OPT_OVERLAP "overlap-check" 132 #define QCOW2_OPT_OVERLAP_TEMPLATE "overlap-check.template" 133 #define QCOW2_OPT_OVERLAP_MAIN_HEADER "overlap-check.main-header" 134 #define QCOW2_OPT_OVERLAP_ACTIVE_L1 "overlap-check.active-l1" 135 #define QCOW2_OPT_OVERLAP_ACTIVE_L2 "overlap-check.active-l2" 136 #define QCOW2_OPT_OVERLAP_REFCOUNT_TABLE "overlap-check.refcount-table" 137 #define QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK "overlap-check.refcount-block" 138 #define QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE "overlap-check.snapshot-table" 139 #define QCOW2_OPT_OVERLAP_INACTIVE_L1 "overlap-check.inactive-l1" 140 #define QCOW2_OPT_OVERLAP_INACTIVE_L2 "overlap-check.inactive-l2" 141 #define QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY "overlap-check.bitmap-directory" 142 #define QCOW2_OPT_CACHE_SIZE "cache-size" 143 #define QCOW2_OPT_L2_CACHE_SIZE "l2-cache-size" 144 #define QCOW2_OPT_L2_CACHE_ENTRY_SIZE "l2-cache-entry-size" 145 #define QCOW2_OPT_REFCOUNT_CACHE_SIZE "refcount-cache-size" 146 #define QCOW2_OPT_CACHE_CLEAN_INTERVAL "cache-clean-interval" 147 148 typedef struct QCowHeader { 149 uint32_t magic; 150 uint32_t version; 151 uint64_t backing_file_offset; 152 uint32_t backing_file_size; 153 uint32_t cluster_bits; 154 uint64_t size; /* in bytes */ 155 uint32_t crypt_method; 156 uint32_t l1_size; /* XXX: save number of clusters instead ? */ 157 uint64_t l1_table_offset; 158 uint64_t refcount_table_offset; 159 uint32_t refcount_table_clusters; 160 uint32_t nb_snapshots; 161 uint64_t snapshots_offset; 162 163 /* The following fields are only valid for version >= 3 */ 164 uint64_t incompatible_features; 165 uint64_t compatible_features; 166 uint64_t autoclear_features; 167 168 uint32_t refcount_order; 169 uint32_t header_length; 170 171 /* Additional fields */ 172 uint8_t compression_type; 173 174 /* header must be a multiple of 8 */ 175 uint8_t padding[7]; 176 } QEMU_PACKED QCowHeader; 177 178 QEMU_BUILD_BUG_ON(!QEMU_IS_ALIGNED(sizeof(QCowHeader), 8)); 179 180 typedef struct QEMU_PACKED QCowSnapshotHeader { 181 /* header is 8 byte aligned */ 182 uint64_t l1_table_offset; 183 184 uint32_t l1_size; 185 uint16_t id_str_size; 186 uint16_t name_size; 187 188 uint32_t date_sec; 189 uint32_t date_nsec; 190 191 uint64_t vm_clock_nsec; 192 193 uint32_t vm_state_size; 194 uint32_t extra_data_size; /* for extension */ 195 /* extra data follows */ 196 /* id_str follows */ 197 /* name follows */ 198 } QCowSnapshotHeader; 199 200 typedef struct QEMU_PACKED QCowSnapshotExtraData { 201 uint64_t vm_state_size_large; 202 uint64_t disk_size; 203 } QCowSnapshotExtraData; 204 205 206 typedef struct QCowSnapshot { 207 uint64_t l1_table_offset; 208 uint32_t l1_size; 209 char *id_str; 210 char *name; 211 uint64_t disk_size; 212 uint64_t vm_state_size; 213 uint32_t date_sec; 214 uint32_t date_nsec; 215 uint64_t vm_clock_nsec; 216 /* Size of all extra data, including QCowSnapshotExtraData if available */ 217 uint32_t extra_data_size; 218 /* Data beyond QCowSnapshotExtraData, if any */ 219 void *unknown_extra_data; 220 } QCowSnapshot; 221 222 struct Qcow2Cache; 223 typedef struct Qcow2Cache Qcow2Cache; 224 225 typedef struct Qcow2CryptoHeaderExtension { 226 uint64_t offset; 227 uint64_t length; 228 } QEMU_PACKED Qcow2CryptoHeaderExtension; 229 230 typedef struct Qcow2UnknownHeaderExtension { 231 uint32_t magic; 232 uint32_t len; 233 QLIST_ENTRY(Qcow2UnknownHeaderExtension) next; 234 uint8_t data[]; 235 } Qcow2UnknownHeaderExtension; 236 237 enum { 238 QCOW2_FEAT_TYPE_INCOMPATIBLE = 0, 239 QCOW2_FEAT_TYPE_COMPATIBLE = 1, 240 QCOW2_FEAT_TYPE_AUTOCLEAR = 2, 241 }; 242 243 /* Incompatible feature bits */ 244 enum { 245 QCOW2_INCOMPAT_DIRTY_BITNR = 0, 246 QCOW2_INCOMPAT_CORRUPT_BITNR = 1, 247 QCOW2_INCOMPAT_DATA_FILE_BITNR = 2, 248 QCOW2_INCOMPAT_COMPRESSION_BITNR = 3, 249 QCOW2_INCOMPAT_EXTL2_BITNR = 4, 250 QCOW2_INCOMPAT_DIRTY = 1 << QCOW2_INCOMPAT_DIRTY_BITNR, 251 QCOW2_INCOMPAT_CORRUPT = 1 << QCOW2_INCOMPAT_CORRUPT_BITNR, 252 QCOW2_INCOMPAT_DATA_FILE = 1 << QCOW2_INCOMPAT_DATA_FILE_BITNR, 253 QCOW2_INCOMPAT_COMPRESSION = 1 << QCOW2_INCOMPAT_COMPRESSION_BITNR, 254 QCOW2_INCOMPAT_EXTL2 = 1 << QCOW2_INCOMPAT_EXTL2_BITNR, 255 256 QCOW2_INCOMPAT_MASK = QCOW2_INCOMPAT_DIRTY 257 | QCOW2_INCOMPAT_CORRUPT 258 | QCOW2_INCOMPAT_DATA_FILE 259 | QCOW2_INCOMPAT_COMPRESSION 260 | QCOW2_INCOMPAT_EXTL2, 261 }; 262 263 /* Compatible feature bits */ 264 enum { 265 QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR = 0, 266 QCOW2_COMPAT_LAZY_REFCOUNTS = 1 << QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR, 267 268 QCOW2_COMPAT_FEAT_MASK = QCOW2_COMPAT_LAZY_REFCOUNTS, 269 }; 270 271 /* Autoclear feature bits */ 272 enum { 273 QCOW2_AUTOCLEAR_BITMAPS_BITNR = 0, 274 QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR = 1, 275 QCOW2_AUTOCLEAR_BITMAPS = 1 << QCOW2_AUTOCLEAR_BITMAPS_BITNR, 276 QCOW2_AUTOCLEAR_DATA_FILE_RAW = 1 << QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR, 277 278 QCOW2_AUTOCLEAR_MASK = QCOW2_AUTOCLEAR_BITMAPS 279 | QCOW2_AUTOCLEAR_DATA_FILE_RAW, 280 }; 281 282 enum qcow2_discard_type { 283 QCOW2_DISCARD_NEVER = 0, 284 QCOW2_DISCARD_ALWAYS, 285 QCOW2_DISCARD_REQUEST, 286 QCOW2_DISCARD_SNAPSHOT, 287 QCOW2_DISCARD_OTHER, 288 QCOW2_DISCARD_MAX 289 }; 290 291 typedef struct Qcow2Feature { 292 uint8_t type; 293 uint8_t bit; 294 char name[46]; 295 } QEMU_PACKED Qcow2Feature; 296 297 typedef struct Qcow2DiscardRegion { 298 BlockDriverState *bs; 299 uint64_t offset; 300 uint64_t bytes; 301 QTAILQ_ENTRY(Qcow2DiscardRegion) next; 302 } Qcow2DiscardRegion; 303 304 typedef uint64_t Qcow2GetRefcountFunc(const void *refcount_array, 305 uint64_t index); 306 typedef void Qcow2SetRefcountFunc(void *refcount_array, 307 uint64_t index, uint64_t value); 308 309 typedef struct Qcow2BitmapHeaderExt { 310 uint32_t nb_bitmaps; 311 uint32_t reserved32; 312 uint64_t bitmap_directory_size; 313 uint64_t bitmap_directory_offset; 314 } QEMU_PACKED Qcow2BitmapHeaderExt; 315 316 #define QCOW2_MAX_THREADS 4 317 318 typedef struct BDRVQcow2State { 319 int cluster_bits; 320 int cluster_size; 321 int l2_slice_size; 322 int subcluster_bits; 323 int subcluster_size; 324 int subclusters_per_cluster; 325 int l2_bits; 326 int l2_size; 327 int l1_size; 328 int l1_vm_state_index; 329 int refcount_block_bits; 330 int refcount_block_size; 331 int csize_shift; 332 int csize_mask; 333 uint64_t cluster_offset_mask; 334 uint64_t l1_table_offset; 335 uint64_t *l1_table; 336 337 Qcow2Cache* l2_table_cache; 338 Qcow2Cache* refcount_block_cache; 339 QEMUTimer *cache_clean_timer; 340 unsigned cache_clean_interval; 341 342 QLIST_HEAD(, QCowL2Meta) cluster_allocs; 343 344 uint64_t *refcount_table; 345 uint64_t refcount_table_offset; 346 uint32_t refcount_table_size; 347 uint32_t max_refcount_table_index; /* Last used entry in refcount_table */ 348 uint64_t free_cluster_index; 349 uint64_t free_byte_offset; 350 351 CoMutex lock; 352 353 Qcow2CryptoHeaderExtension crypto_header; /* QCow2 header extension */ 354 QCryptoBlockOpenOptions *crypto_opts; /* Disk encryption runtime options */ 355 QCryptoBlock *crypto; /* Disk encryption format driver */ 356 bool crypt_physical_offset; /* Whether to use virtual or physical offset 357 for encryption initialization vector tweak */ 358 uint32_t crypt_method_header; 359 uint64_t snapshots_offset; 360 int snapshots_size; 361 unsigned int nb_snapshots; 362 QCowSnapshot *snapshots; 363 364 uint32_t nb_bitmaps; 365 uint64_t bitmap_directory_size; 366 uint64_t bitmap_directory_offset; 367 368 int flags; 369 int qcow_version; 370 bool use_lazy_refcounts; 371 int refcount_order; 372 int refcount_bits; 373 uint64_t refcount_max; 374 375 Qcow2GetRefcountFunc *get_refcount; 376 Qcow2SetRefcountFunc *set_refcount; 377 378 bool discard_passthrough[QCOW2_DISCARD_MAX]; 379 380 int overlap_check; /* bitmask of Qcow2MetadataOverlap values */ 381 bool signaled_corruption; 382 383 uint64_t incompatible_features; 384 uint64_t compatible_features; 385 uint64_t autoclear_features; 386 387 size_t unknown_header_fields_size; 388 void* unknown_header_fields; 389 QLIST_HEAD(, Qcow2UnknownHeaderExtension) unknown_header_ext; 390 QTAILQ_HEAD (, Qcow2DiscardRegion) discards; 391 bool cache_discards; 392 393 /* Backing file path and format as stored in the image (this is not the 394 * effective path/format, which may be the result of a runtime option 395 * override) */ 396 char *image_backing_file; 397 char *image_backing_format; 398 char *image_data_file; 399 400 CoQueue thread_task_queue; 401 int nb_threads; 402 403 BdrvChild *data_file; 404 405 bool metadata_preallocation_checked; 406 bool metadata_preallocation; 407 /* 408 * Compression type used for the image. Default: 0 - ZLIB 409 * The image compression type is set on image creation. 410 * For now, the only way to change the compression type 411 * is to convert the image with the desired compression type set. 412 */ 413 Qcow2CompressionType compression_type; 414 } BDRVQcow2State; 415 416 typedef struct Qcow2COWRegion { 417 /** 418 * Offset of the COW region in bytes from the start of the first cluster 419 * touched by the request. 420 */ 421 unsigned offset; 422 423 /** Number of bytes to copy */ 424 unsigned nb_bytes; 425 } Qcow2COWRegion; 426 427 /** 428 * Describes an in-flight (part of a) write request that writes to clusters 429 * that are not referenced in their L2 table yet. 430 */ 431 typedef struct QCowL2Meta 432 { 433 /** Guest offset of the first newly allocated cluster */ 434 uint64_t offset; 435 436 /** Host offset of the first newly allocated cluster */ 437 uint64_t alloc_offset; 438 439 /** Number of newly allocated clusters */ 440 int nb_clusters; 441 442 /** Do not free the old clusters */ 443 bool keep_old_clusters; 444 445 /** 446 * Requests that overlap with this allocation and wait to be restarted 447 * when the allocating request has completed. 448 */ 449 CoQueue dependent_requests; 450 451 /** 452 * The COW Region between the start of the first allocated cluster and the 453 * area the guest actually writes to. 454 */ 455 Qcow2COWRegion cow_start; 456 457 /** 458 * The COW Region between the area the guest actually writes to and the 459 * end of the last allocated cluster. 460 */ 461 Qcow2COWRegion cow_end; 462 463 /* 464 * Indicates that COW regions are already handled and do not require 465 * any more processing. 466 */ 467 bool skip_cow; 468 469 /** 470 * Indicates that this is not a normal write request but a preallocation. 471 * If the image has extended L2 entries this means that no new individual 472 * subclusters will be marked as allocated in the L2 bitmap (but any 473 * existing contents of that bitmap will be kept). 474 */ 475 bool prealloc; 476 477 /** 478 * The I/O vector with the data from the actual guest write request. 479 * If non-NULL, this is meant to be merged together with the data 480 * from @cow_start and @cow_end into one single write operation. 481 */ 482 QEMUIOVector *data_qiov; 483 size_t data_qiov_offset; 484 485 /** Pointer to next L2Meta of the same write request */ 486 struct QCowL2Meta *next; 487 488 QLIST_ENTRY(QCowL2Meta) next_in_flight; 489 } QCowL2Meta; 490 491 /* 492 * In images with standard L2 entries all clusters are treated as if 493 * they had one subcluster so QCow2ClusterType and QCow2SubclusterType 494 * can be mapped to each other and have the exact same meaning 495 * (QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC cannot happen in these images). 496 * 497 * In images with extended L2 entries QCow2ClusterType refers to the 498 * complete cluster and QCow2SubclusterType to each of the individual 499 * subclusters, so there are several possible combinations: 500 * 501 * |--------------+---------------------------| 502 * | Cluster type | Possible subcluster types | 503 * |--------------+---------------------------| 504 * | UNALLOCATED | UNALLOCATED_PLAIN | 505 * | | ZERO_PLAIN | 506 * |--------------+---------------------------| 507 * | NORMAL | UNALLOCATED_ALLOC | 508 * | | ZERO_ALLOC | 509 * | | NORMAL | 510 * |--------------+---------------------------| 511 * | COMPRESSED | COMPRESSED | 512 * |--------------+---------------------------| 513 * 514 * QCOW2_SUBCLUSTER_INVALID means that the L2 entry is incorrect and 515 * the image should be marked corrupt. 516 */ 517 518 typedef enum QCow2ClusterType { 519 QCOW2_CLUSTER_UNALLOCATED, 520 QCOW2_CLUSTER_ZERO_PLAIN, 521 QCOW2_CLUSTER_ZERO_ALLOC, 522 QCOW2_CLUSTER_NORMAL, 523 QCOW2_CLUSTER_COMPRESSED, 524 } QCow2ClusterType; 525 526 typedef enum QCow2SubclusterType { 527 QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN, 528 QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC, 529 QCOW2_SUBCLUSTER_ZERO_PLAIN, 530 QCOW2_SUBCLUSTER_ZERO_ALLOC, 531 QCOW2_SUBCLUSTER_NORMAL, 532 QCOW2_SUBCLUSTER_COMPRESSED, 533 QCOW2_SUBCLUSTER_INVALID, 534 } QCow2SubclusterType; 535 536 typedef enum QCow2MetadataOverlap { 537 QCOW2_OL_MAIN_HEADER_BITNR = 0, 538 QCOW2_OL_ACTIVE_L1_BITNR = 1, 539 QCOW2_OL_ACTIVE_L2_BITNR = 2, 540 QCOW2_OL_REFCOUNT_TABLE_BITNR = 3, 541 QCOW2_OL_REFCOUNT_BLOCK_BITNR = 4, 542 QCOW2_OL_SNAPSHOT_TABLE_BITNR = 5, 543 QCOW2_OL_INACTIVE_L1_BITNR = 6, 544 QCOW2_OL_INACTIVE_L2_BITNR = 7, 545 QCOW2_OL_BITMAP_DIRECTORY_BITNR = 8, 546 547 QCOW2_OL_MAX_BITNR = 9, 548 549 QCOW2_OL_NONE = 0, 550 QCOW2_OL_MAIN_HEADER = (1 << QCOW2_OL_MAIN_HEADER_BITNR), 551 QCOW2_OL_ACTIVE_L1 = (1 << QCOW2_OL_ACTIVE_L1_BITNR), 552 QCOW2_OL_ACTIVE_L2 = (1 << QCOW2_OL_ACTIVE_L2_BITNR), 553 QCOW2_OL_REFCOUNT_TABLE = (1 << QCOW2_OL_REFCOUNT_TABLE_BITNR), 554 QCOW2_OL_REFCOUNT_BLOCK = (1 << QCOW2_OL_REFCOUNT_BLOCK_BITNR), 555 QCOW2_OL_SNAPSHOT_TABLE = (1 << QCOW2_OL_SNAPSHOT_TABLE_BITNR), 556 QCOW2_OL_INACTIVE_L1 = (1 << QCOW2_OL_INACTIVE_L1_BITNR), 557 /* NOTE: Checking overlaps with inactive L2 tables will result in bdrv 558 * reads. */ 559 QCOW2_OL_INACTIVE_L2 = (1 << QCOW2_OL_INACTIVE_L2_BITNR), 560 QCOW2_OL_BITMAP_DIRECTORY = (1 << QCOW2_OL_BITMAP_DIRECTORY_BITNR), 561 } QCow2MetadataOverlap; 562 563 /* Perform all overlap checks which can be done in constant time */ 564 #define QCOW2_OL_CONSTANT \ 565 (QCOW2_OL_MAIN_HEADER | QCOW2_OL_ACTIVE_L1 | QCOW2_OL_REFCOUNT_TABLE | \ 566 QCOW2_OL_SNAPSHOT_TABLE | QCOW2_OL_BITMAP_DIRECTORY) 567 568 /* Perform all overlap checks which don't require disk access */ 569 #define QCOW2_OL_CACHED \ 570 (QCOW2_OL_CONSTANT | QCOW2_OL_ACTIVE_L2 | QCOW2_OL_REFCOUNT_BLOCK | \ 571 QCOW2_OL_INACTIVE_L1) 572 573 /* Perform all overlap checks */ 574 #define QCOW2_OL_ALL \ 575 (QCOW2_OL_CACHED | QCOW2_OL_INACTIVE_L2) 576 577 #define L1E_OFFSET_MASK 0x00fffffffffffe00ULL 578 #define L2E_OFFSET_MASK 0x00fffffffffffe00ULL 579 #define L2E_COMPRESSED_OFFSET_SIZE_MASK 0x3fffffffffffffffULL 580 581 #define REFT_OFFSET_MASK 0xfffffffffffffe00ULL 582 583 #define INV_OFFSET (-1ULL) 584 585 static inline bool has_subclusters(BDRVQcow2State *s) 586 { 587 return s->incompatible_features & QCOW2_INCOMPAT_EXTL2; 588 } 589 590 static inline size_t l2_entry_size(BDRVQcow2State *s) 591 { 592 return has_subclusters(s) ? L2E_SIZE_EXTENDED : L2E_SIZE_NORMAL; 593 } 594 595 static inline uint64_t get_l2_entry(BDRVQcow2State *s, uint64_t *l2_slice, 596 int idx) 597 { 598 idx *= l2_entry_size(s) / sizeof(uint64_t); 599 return be64_to_cpu(l2_slice[idx]); 600 } 601 602 static inline uint64_t get_l2_bitmap(BDRVQcow2State *s, uint64_t *l2_slice, 603 int idx) 604 { 605 if (has_subclusters(s)) { 606 idx *= l2_entry_size(s) / sizeof(uint64_t); 607 return be64_to_cpu(l2_slice[idx + 1]); 608 } else { 609 return 0; /* For convenience only; this value has no meaning. */ 610 } 611 } 612 613 static inline void set_l2_entry(BDRVQcow2State *s, uint64_t *l2_slice, 614 int idx, uint64_t entry) 615 { 616 idx *= l2_entry_size(s) / sizeof(uint64_t); 617 l2_slice[idx] = cpu_to_be64(entry); 618 } 619 620 static inline void set_l2_bitmap(BDRVQcow2State *s, uint64_t *l2_slice, 621 int idx, uint64_t bitmap) 622 { 623 assert(has_subclusters(s)); 624 idx *= l2_entry_size(s) / sizeof(uint64_t); 625 l2_slice[idx + 1] = cpu_to_be64(bitmap); 626 } 627 628 static inline bool has_data_file(BlockDriverState *bs) 629 { 630 BDRVQcow2State *s = bs->opaque; 631 return (s->data_file != bs->file); 632 } 633 634 static inline bool data_file_is_raw(BlockDriverState *bs) 635 { 636 BDRVQcow2State *s = bs->opaque; 637 return !!(s->autoclear_features & QCOW2_AUTOCLEAR_DATA_FILE_RAW); 638 } 639 640 static inline int64_t start_of_cluster(BDRVQcow2State *s, int64_t offset) 641 { 642 return offset & ~(s->cluster_size - 1); 643 } 644 645 static inline int64_t offset_into_cluster(BDRVQcow2State *s, int64_t offset) 646 { 647 return offset & (s->cluster_size - 1); 648 } 649 650 static inline int64_t offset_into_subcluster(BDRVQcow2State *s, int64_t offset) 651 { 652 return offset & (s->subcluster_size - 1); 653 } 654 655 static inline uint64_t size_to_clusters(BDRVQcow2State *s, uint64_t size) 656 { 657 return (size + (s->cluster_size - 1)) >> s->cluster_bits; 658 } 659 660 static inline uint64_t size_to_subclusters(BDRVQcow2State *s, uint64_t size) 661 { 662 return (size + (s->subcluster_size - 1)) >> s->subcluster_bits; 663 } 664 665 static inline int64_t size_to_l1(BDRVQcow2State *s, int64_t size) 666 { 667 int shift = s->cluster_bits + s->l2_bits; 668 return (size + (1ULL << shift) - 1) >> shift; 669 } 670 671 static inline int offset_to_l1_index(BDRVQcow2State *s, uint64_t offset) 672 { 673 return offset >> (s->l2_bits + s->cluster_bits); 674 } 675 676 static inline int offset_to_l2_index(BDRVQcow2State *s, int64_t offset) 677 { 678 return (offset >> s->cluster_bits) & (s->l2_size - 1); 679 } 680 681 static inline int offset_to_l2_slice_index(BDRVQcow2State *s, int64_t offset) 682 { 683 return (offset >> s->cluster_bits) & (s->l2_slice_size - 1); 684 } 685 686 static inline int offset_to_sc_index(BDRVQcow2State *s, int64_t offset) 687 { 688 return (offset >> s->subcluster_bits) & (s->subclusters_per_cluster - 1); 689 } 690 691 static inline int64_t qcow2_vm_state_offset(BDRVQcow2State *s) 692 { 693 return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits); 694 } 695 696 static inline QCow2ClusterType qcow2_get_cluster_type(BlockDriverState *bs, 697 uint64_t l2_entry) 698 { 699 BDRVQcow2State *s = bs->opaque; 700 701 if (l2_entry & QCOW_OFLAG_COMPRESSED) { 702 return QCOW2_CLUSTER_COMPRESSED; 703 } else if ((l2_entry & QCOW_OFLAG_ZERO) && !has_subclusters(s)) { 704 if (l2_entry & L2E_OFFSET_MASK) { 705 return QCOW2_CLUSTER_ZERO_ALLOC; 706 } 707 return QCOW2_CLUSTER_ZERO_PLAIN; 708 } else if (!(l2_entry & L2E_OFFSET_MASK)) { 709 /* Offset 0 generally means unallocated, but it is ambiguous with 710 * external data files because 0 is a valid offset there. However, all 711 * clusters in external data files always have refcount 1, so we can 712 * rely on QCOW_OFLAG_COPIED to disambiguate. */ 713 if (has_data_file(bs) && (l2_entry & QCOW_OFLAG_COPIED)) { 714 return QCOW2_CLUSTER_NORMAL; 715 } else { 716 return QCOW2_CLUSTER_UNALLOCATED; 717 } 718 } else { 719 return QCOW2_CLUSTER_NORMAL; 720 } 721 } 722 723 /* 724 * In an image without subsclusters @l2_bitmap is ignored and 725 * @sc_index must be 0. 726 * Return QCOW2_SUBCLUSTER_INVALID if an invalid l2 entry is detected 727 * (this checks the whole entry and bitmap, not only the bits related 728 * to subcluster @sc_index). 729 */ 730 static inline 731 QCow2SubclusterType qcow2_get_subcluster_type(BlockDriverState *bs, 732 uint64_t l2_entry, 733 uint64_t l2_bitmap, 734 unsigned sc_index) 735 { 736 BDRVQcow2State *s = bs->opaque; 737 QCow2ClusterType type = qcow2_get_cluster_type(bs, l2_entry); 738 assert(sc_index < s->subclusters_per_cluster); 739 740 if (has_subclusters(s)) { 741 switch (type) { 742 case QCOW2_CLUSTER_COMPRESSED: 743 return QCOW2_SUBCLUSTER_COMPRESSED; 744 case QCOW2_CLUSTER_NORMAL: 745 if ((l2_bitmap >> 32) & l2_bitmap) { 746 return QCOW2_SUBCLUSTER_INVALID; 747 } else if (l2_bitmap & QCOW_OFLAG_SUB_ZERO(sc_index)) { 748 return QCOW2_SUBCLUSTER_ZERO_ALLOC; 749 } else if (l2_bitmap & QCOW_OFLAG_SUB_ALLOC(sc_index)) { 750 return QCOW2_SUBCLUSTER_NORMAL; 751 } else { 752 return QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC; 753 } 754 case QCOW2_CLUSTER_UNALLOCATED: 755 if (l2_bitmap & QCOW_L2_BITMAP_ALL_ALLOC) { 756 return QCOW2_SUBCLUSTER_INVALID; 757 } else if (l2_bitmap & QCOW_OFLAG_SUB_ZERO(sc_index)) { 758 return QCOW2_SUBCLUSTER_ZERO_PLAIN; 759 } else { 760 return QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN; 761 } 762 default: 763 g_assert_not_reached(); 764 } 765 } else { 766 switch (type) { 767 case QCOW2_CLUSTER_COMPRESSED: 768 return QCOW2_SUBCLUSTER_COMPRESSED; 769 case QCOW2_CLUSTER_ZERO_PLAIN: 770 return QCOW2_SUBCLUSTER_ZERO_PLAIN; 771 case QCOW2_CLUSTER_ZERO_ALLOC: 772 return QCOW2_SUBCLUSTER_ZERO_ALLOC; 773 case QCOW2_CLUSTER_NORMAL: 774 return QCOW2_SUBCLUSTER_NORMAL; 775 case QCOW2_CLUSTER_UNALLOCATED: 776 return QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN; 777 default: 778 g_assert_not_reached(); 779 } 780 } 781 } 782 783 static inline bool qcow2_cluster_is_allocated(QCow2ClusterType type) 784 { 785 return (type == QCOW2_CLUSTER_COMPRESSED || type == QCOW2_CLUSTER_NORMAL || 786 type == QCOW2_CLUSTER_ZERO_ALLOC); 787 } 788 789 /* Check whether refcounts are eager or lazy */ 790 static inline bool qcow2_need_accurate_refcounts(BDRVQcow2State *s) 791 { 792 return !(s->incompatible_features & QCOW2_INCOMPAT_DIRTY); 793 } 794 795 static inline uint64_t l2meta_cow_start(QCowL2Meta *m) 796 { 797 return m->offset + m->cow_start.offset; 798 } 799 800 static inline uint64_t l2meta_cow_end(QCowL2Meta *m) 801 { 802 return m->offset + m->cow_end.offset + m->cow_end.nb_bytes; 803 } 804 805 static inline uint64_t refcount_diff(uint64_t r1, uint64_t r2) 806 { 807 return r1 > r2 ? r1 - r2 : r2 - r1; 808 } 809 810 static inline 811 uint32_t offset_to_reftable_index(BDRVQcow2State *s, uint64_t offset) 812 { 813 return offset >> (s->refcount_block_bits + s->cluster_bits); 814 } 815 816 /* qcow2.c functions */ 817 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size, 818 int refcount_order, bool generous_increase, 819 uint64_t *refblock_count); 820 821 int qcow2_mark_dirty(BlockDriverState *bs); 822 int qcow2_mark_corrupt(BlockDriverState *bs); 823 int qcow2_mark_consistent(BlockDriverState *bs); 824 int qcow2_update_header(BlockDriverState *bs); 825 826 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset, 827 int64_t size, const char *message_format, ...) 828 GCC_FMT_ATTR(5, 6); 829 830 int qcow2_validate_table(BlockDriverState *bs, uint64_t offset, 831 uint64_t entries, size_t entry_len, 832 int64_t max_size_bytes, const char *table_name, 833 Error **errp); 834 835 /* qcow2-refcount.c functions */ 836 int qcow2_refcount_init(BlockDriverState *bs); 837 void qcow2_refcount_close(BlockDriverState *bs); 838 839 int qcow2_get_refcount(BlockDriverState *bs, int64_t cluster_index, 840 uint64_t *refcount); 841 842 int qcow2_update_cluster_refcount(BlockDriverState *bs, int64_t cluster_index, 843 uint64_t addend, bool decrease, 844 enum qcow2_discard_type type); 845 846 int64_t qcow2_refcount_area(BlockDriverState *bs, uint64_t offset, 847 uint64_t additional_clusters, bool exact_size, 848 int new_refblock_index, 849 uint64_t new_refblock_offset); 850 851 int64_t qcow2_alloc_clusters(BlockDriverState *bs, uint64_t size); 852 int64_t qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset, 853 int64_t nb_clusters); 854 int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size); 855 void qcow2_free_clusters(BlockDriverState *bs, 856 int64_t offset, int64_t size, 857 enum qcow2_discard_type type); 858 void qcow2_free_any_clusters(BlockDriverState *bs, uint64_t l2_entry, 859 int nb_clusters, enum qcow2_discard_type type); 860 861 int qcow2_update_snapshot_refcount(BlockDriverState *bs, 862 int64_t l1_table_offset, int l1_size, int addend); 863 864 int coroutine_fn qcow2_flush_caches(BlockDriverState *bs); 865 int coroutine_fn qcow2_write_caches(BlockDriverState *bs); 866 int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res, 867 BdrvCheckMode fix); 868 869 void qcow2_process_discards(BlockDriverState *bs, int ret); 870 871 int qcow2_check_metadata_overlap(BlockDriverState *bs, int ign, int64_t offset, 872 int64_t size); 873 int qcow2_pre_write_overlap_check(BlockDriverState *bs, int ign, int64_t offset, 874 int64_t size, bool data_file); 875 int qcow2_inc_refcounts_imrt(BlockDriverState *bs, BdrvCheckResult *res, 876 void **refcount_table, 877 int64_t *refcount_table_size, 878 int64_t offset, int64_t size); 879 880 int qcow2_change_refcount_order(BlockDriverState *bs, int refcount_order, 881 BlockDriverAmendStatusCB *status_cb, 882 void *cb_opaque, Error **errp); 883 int qcow2_shrink_reftable(BlockDriverState *bs); 884 int64_t qcow2_get_last_cluster(BlockDriverState *bs, int64_t size); 885 int qcow2_detect_metadata_preallocation(BlockDriverState *bs); 886 887 /* qcow2-cluster.c functions */ 888 int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size, 889 bool exact_size); 890 int qcow2_shrink_l1_table(BlockDriverState *bs, uint64_t max_size); 891 int qcow2_write_l1_entry(BlockDriverState *bs, int l1_index); 892 int qcow2_encrypt_sectors(BDRVQcow2State *s, int64_t sector_num, 893 uint8_t *buf, int nb_sectors, bool enc, Error **errp); 894 895 int qcow2_get_host_offset(BlockDriverState *bs, uint64_t offset, 896 unsigned int *bytes, uint64_t *host_offset, 897 QCow2SubclusterType *subcluster_type); 898 int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset, 899 unsigned int *bytes, uint64_t *host_offset, 900 QCowL2Meta **m); 901 int qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs, 902 uint64_t offset, 903 int compressed_size, 904 uint64_t *host_offset); 905 906 int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m); 907 void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m); 908 int qcow2_cluster_discard(BlockDriverState *bs, uint64_t offset, 909 uint64_t bytes, enum qcow2_discard_type type, 910 bool full_discard); 911 int qcow2_subcluster_zeroize(BlockDriverState *bs, uint64_t offset, 912 uint64_t bytes, int flags); 913 914 int qcow2_expand_zero_clusters(BlockDriverState *bs, 915 BlockDriverAmendStatusCB *status_cb, 916 void *cb_opaque); 917 918 /* qcow2-snapshot.c functions */ 919 int qcow2_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info); 920 int qcow2_snapshot_goto(BlockDriverState *bs, const char *snapshot_id); 921 int qcow2_snapshot_delete(BlockDriverState *bs, 922 const char *snapshot_id, 923 const char *name, 924 Error **errp); 925 int qcow2_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab); 926 int qcow2_snapshot_load_tmp(BlockDriverState *bs, 927 const char *snapshot_id, 928 const char *name, 929 Error **errp); 930 931 void qcow2_free_snapshots(BlockDriverState *bs); 932 int qcow2_read_snapshots(BlockDriverState *bs, Error **errp); 933 int qcow2_write_snapshots(BlockDriverState *bs); 934 935 int coroutine_fn qcow2_check_read_snapshot_table(BlockDriverState *bs, 936 BdrvCheckResult *result, 937 BdrvCheckMode fix); 938 int coroutine_fn qcow2_check_fix_snapshot_table(BlockDriverState *bs, 939 BdrvCheckResult *result, 940 BdrvCheckMode fix); 941 942 /* qcow2-cache.c functions */ 943 Qcow2Cache *qcow2_cache_create(BlockDriverState *bs, int num_tables, 944 unsigned table_size); 945 int qcow2_cache_destroy(Qcow2Cache *c); 946 947 void qcow2_cache_entry_mark_dirty(Qcow2Cache *c, void *table); 948 int qcow2_cache_flush(BlockDriverState *bs, Qcow2Cache *c); 949 int qcow2_cache_write(BlockDriverState *bs, Qcow2Cache *c); 950 int qcow2_cache_set_dependency(BlockDriverState *bs, Qcow2Cache *c, 951 Qcow2Cache *dependency); 952 void qcow2_cache_depends_on_flush(Qcow2Cache *c); 953 954 void qcow2_cache_clean_unused(Qcow2Cache *c); 955 int qcow2_cache_empty(BlockDriverState *bs, Qcow2Cache *c); 956 957 int qcow2_cache_get(BlockDriverState *bs, Qcow2Cache *c, uint64_t offset, 958 void **table); 959 int qcow2_cache_get_empty(BlockDriverState *bs, Qcow2Cache *c, uint64_t offset, 960 void **table); 961 void qcow2_cache_put(Qcow2Cache *c, void **table); 962 void *qcow2_cache_is_table_offset(Qcow2Cache *c, uint64_t offset); 963 void qcow2_cache_discard(Qcow2Cache *c, void *table); 964 965 /* qcow2-bitmap.c functions */ 966 int qcow2_check_bitmaps_refcounts(BlockDriverState *bs, BdrvCheckResult *res, 967 void **refcount_table, 968 int64_t *refcount_table_size); 969 bool qcow2_load_dirty_bitmaps(BlockDriverState *bs, Error **errp); 970 Qcow2BitmapInfoList *qcow2_get_bitmap_info_list(BlockDriverState *bs, 971 Error **errp); 972 int qcow2_reopen_bitmaps_rw(BlockDriverState *bs, Error **errp); 973 int qcow2_truncate_bitmaps_check(BlockDriverState *bs, Error **errp); 974 void qcow2_store_persistent_dirty_bitmaps(BlockDriverState *bs, 975 bool release_stored, Error **errp); 976 int qcow2_reopen_bitmaps_ro(BlockDriverState *bs, Error **errp); 977 bool qcow2_co_can_store_new_dirty_bitmap(BlockDriverState *bs, 978 const char *name, 979 uint32_t granularity, 980 Error **errp); 981 int qcow2_co_remove_persistent_dirty_bitmap(BlockDriverState *bs, 982 const char *name, 983 Error **errp); 984 bool qcow2_supports_persistent_dirty_bitmap(BlockDriverState *bs); 985 uint64_t qcow2_get_persistent_dirty_bitmap_size(BlockDriverState *bs, 986 uint32_t cluster_size); 987 988 ssize_t coroutine_fn 989 qcow2_co_compress(BlockDriverState *bs, void *dest, size_t dest_size, 990 const void *src, size_t src_size); 991 ssize_t coroutine_fn 992 qcow2_co_decompress(BlockDriverState *bs, void *dest, size_t dest_size, 993 const void *src, size_t src_size); 994 int coroutine_fn 995 qcow2_co_encrypt(BlockDriverState *bs, uint64_t host_offset, 996 uint64_t guest_offset, void *buf, size_t len); 997 int coroutine_fn 998 qcow2_co_decrypt(BlockDriverState *bs, uint64_t host_offset, 999 uint64_t guest_offset, void *buf, size_t len); 1000 1001 #endif 1002