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 /* Size of L1 table entries */ 103 #define L1E_SIZE (sizeof(uint64_t)) 104 105 /* Size of reftable entries */ 106 #define REFTABLE_ENTRY_SIZE (sizeof(uint64_t)) 107 108 #define MIN_CLUSTER_BITS 9 109 #define MAX_CLUSTER_BITS 21 110 111 /* Defined in the qcow2 spec (compressed cluster descriptor) */ 112 #define QCOW2_COMPRESSED_SECTOR_SIZE 512U 113 #define QCOW2_COMPRESSED_SECTOR_MASK (~(QCOW2_COMPRESSED_SECTOR_SIZE - 1ULL)) 114 115 /* Must be at least 2 to cover COW */ 116 #define MIN_L2_CACHE_SIZE 2 /* cache entries */ 117 118 /* Must be at least 4 to cover all cases of refcount table growth */ 119 #define MIN_REFCOUNT_CACHE_SIZE 4 /* clusters */ 120 121 #ifdef CONFIG_LINUX 122 #define DEFAULT_L2_CACHE_MAX_SIZE (32 * MiB) 123 #define DEFAULT_CACHE_CLEAN_INTERVAL 600 /* seconds */ 124 #else 125 #define DEFAULT_L2_CACHE_MAX_SIZE (8 * MiB) 126 /* Cache clean interval is currently available only on Linux, so must be 0 */ 127 #define DEFAULT_CACHE_CLEAN_INTERVAL 0 128 #endif 129 130 #define DEFAULT_CLUSTER_SIZE 65536 131 132 #define QCOW2_OPT_DATA_FILE "data-file" 133 #define QCOW2_OPT_LAZY_REFCOUNTS "lazy-refcounts" 134 #define QCOW2_OPT_DISCARD_REQUEST "pass-discard-request" 135 #define QCOW2_OPT_DISCARD_SNAPSHOT "pass-discard-snapshot" 136 #define QCOW2_OPT_DISCARD_OTHER "pass-discard-other" 137 #define QCOW2_OPT_OVERLAP "overlap-check" 138 #define QCOW2_OPT_OVERLAP_TEMPLATE "overlap-check.template" 139 #define QCOW2_OPT_OVERLAP_MAIN_HEADER "overlap-check.main-header" 140 #define QCOW2_OPT_OVERLAP_ACTIVE_L1 "overlap-check.active-l1" 141 #define QCOW2_OPT_OVERLAP_ACTIVE_L2 "overlap-check.active-l2" 142 #define QCOW2_OPT_OVERLAP_REFCOUNT_TABLE "overlap-check.refcount-table" 143 #define QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK "overlap-check.refcount-block" 144 #define QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE "overlap-check.snapshot-table" 145 #define QCOW2_OPT_OVERLAP_INACTIVE_L1 "overlap-check.inactive-l1" 146 #define QCOW2_OPT_OVERLAP_INACTIVE_L2 "overlap-check.inactive-l2" 147 #define QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY "overlap-check.bitmap-directory" 148 #define QCOW2_OPT_CACHE_SIZE "cache-size" 149 #define QCOW2_OPT_L2_CACHE_SIZE "l2-cache-size" 150 #define QCOW2_OPT_L2_CACHE_ENTRY_SIZE "l2-cache-entry-size" 151 #define QCOW2_OPT_REFCOUNT_CACHE_SIZE "refcount-cache-size" 152 #define QCOW2_OPT_CACHE_CLEAN_INTERVAL "cache-clean-interval" 153 154 typedef struct QCowHeader { 155 uint32_t magic; 156 uint32_t version; 157 uint64_t backing_file_offset; 158 uint32_t backing_file_size; 159 uint32_t cluster_bits; 160 uint64_t size; /* in bytes */ 161 uint32_t crypt_method; 162 uint32_t l1_size; /* XXX: save number of clusters instead ? */ 163 uint64_t l1_table_offset; 164 uint64_t refcount_table_offset; 165 uint32_t refcount_table_clusters; 166 uint32_t nb_snapshots; 167 uint64_t snapshots_offset; 168 169 /* The following fields are only valid for version >= 3 */ 170 uint64_t incompatible_features; 171 uint64_t compatible_features; 172 uint64_t autoclear_features; 173 174 uint32_t refcount_order; 175 uint32_t header_length; 176 177 /* Additional fields */ 178 uint8_t compression_type; 179 180 /* header must be a multiple of 8 */ 181 uint8_t padding[7]; 182 } QEMU_PACKED QCowHeader; 183 184 QEMU_BUILD_BUG_ON(!QEMU_IS_ALIGNED(sizeof(QCowHeader), 8)); 185 186 typedef struct QEMU_PACKED QCowSnapshotHeader { 187 /* header is 8 byte aligned */ 188 uint64_t l1_table_offset; 189 190 uint32_t l1_size; 191 uint16_t id_str_size; 192 uint16_t name_size; 193 194 uint32_t date_sec; 195 uint32_t date_nsec; 196 197 uint64_t vm_clock_nsec; 198 199 uint32_t vm_state_size; 200 uint32_t extra_data_size; /* for extension */ 201 /* extra data follows */ 202 /* id_str follows */ 203 /* name follows */ 204 } QCowSnapshotHeader; 205 206 typedef struct QEMU_PACKED QCowSnapshotExtraData { 207 uint64_t vm_state_size_large; 208 uint64_t disk_size; 209 } QCowSnapshotExtraData; 210 211 212 typedef struct QCowSnapshot { 213 uint64_t l1_table_offset; 214 uint32_t l1_size; 215 char *id_str; 216 char *name; 217 uint64_t disk_size; 218 uint64_t vm_state_size; 219 uint32_t date_sec; 220 uint32_t date_nsec; 221 uint64_t vm_clock_nsec; 222 /* Size of all extra data, including QCowSnapshotExtraData if available */ 223 uint32_t extra_data_size; 224 /* Data beyond QCowSnapshotExtraData, if any */ 225 void *unknown_extra_data; 226 } QCowSnapshot; 227 228 struct Qcow2Cache; 229 typedef struct Qcow2Cache Qcow2Cache; 230 231 typedef struct Qcow2CryptoHeaderExtension { 232 uint64_t offset; 233 uint64_t length; 234 } QEMU_PACKED Qcow2CryptoHeaderExtension; 235 236 typedef struct Qcow2UnknownHeaderExtension { 237 uint32_t magic; 238 uint32_t len; 239 QLIST_ENTRY(Qcow2UnknownHeaderExtension) next; 240 uint8_t data[]; 241 } Qcow2UnknownHeaderExtension; 242 243 enum { 244 QCOW2_FEAT_TYPE_INCOMPATIBLE = 0, 245 QCOW2_FEAT_TYPE_COMPATIBLE = 1, 246 QCOW2_FEAT_TYPE_AUTOCLEAR = 2, 247 }; 248 249 /* Incompatible feature bits */ 250 enum { 251 QCOW2_INCOMPAT_DIRTY_BITNR = 0, 252 QCOW2_INCOMPAT_CORRUPT_BITNR = 1, 253 QCOW2_INCOMPAT_DATA_FILE_BITNR = 2, 254 QCOW2_INCOMPAT_COMPRESSION_BITNR = 3, 255 QCOW2_INCOMPAT_EXTL2_BITNR = 4, 256 QCOW2_INCOMPAT_DIRTY = 1 << QCOW2_INCOMPAT_DIRTY_BITNR, 257 QCOW2_INCOMPAT_CORRUPT = 1 << QCOW2_INCOMPAT_CORRUPT_BITNR, 258 QCOW2_INCOMPAT_DATA_FILE = 1 << QCOW2_INCOMPAT_DATA_FILE_BITNR, 259 QCOW2_INCOMPAT_COMPRESSION = 1 << QCOW2_INCOMPAT_COMPRESSION_BITNR, 260 QCOW2_INCOMPAT_EXTL2 = 1 << QCOW2_INCOMPAT_EXTL2_BITNR, 261 262 QCOW2_INCOMPAT_MASK = QCOW2_INCOMPAT_DIRTY 263 | QCOW2_INCOMPAT_CORRUPT 264 | QCOW2_INCOMPAT_DATA_FILE 265 | QCOW2_INCOMPAT_COMPRESSION 266 | QCOW2_INCOMPAT_EXTL2, 267 }; 268 269 /* Compatible feature bits */ 270 enum { 271 QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR = 0, 272 QCOW2_COMPAT_LAZY_REFCOUNTS = 1 << QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR, 273 274 QCOW2_COMPAT_FEAT_MASK = QCOW2_COMPAT_LAZY_REFCOUNTS, 275 }; 276 277 /* Autoclear feature bits */ 278 enum { 279 QCOW2_AUTOCLEAR_BITMAPS_BITNR = 0, 280 QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR = 1, 281 QCOW2_AUTOCLEAR_BITMAPS = 1 << QCOW2_AUTOCLEAR_BITMAPS_BITNR, 282 QCOW2_AUTOCLEAR_DATA_FILE_RAW = 1 << QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR, 283 284 QCOW2_AUTOCLEAR_MASK = QCOW2_AUTOCLEAR_BITMAPS 285 | QCOW2_AUTOCLEAR_DATA_FILE_RAW, 286 }; 287 288 enum qcow2_discard_type { 289 QCOW2_DISCARD_NEVER = 0, 290 QCOW2_DISCARD_ALWAYS, 291 QCOW2_DISCARD_REQUEST, 292 QCOW2_DISCARD_SNAPSHOT, 293 QCOW2_DISCARD_OTHER, 294 QCOW2_DISCARD_MAX 295 }; 296 297 typedef struct Qcow2Feature { 298 uint8_t type; 299 uint8_t bit; 300 char name[46]; 301 } QEMU_PACKED Qcow2Feature; 302 303 typedef struct Qcow2DiscardRegion { 304 BlockDriverState *bs; 305 uint64_t offset; 306 uint64_t bytes; 307 QTAILQ_ENTRY(Qcow2DiscardRegion) next; 308 } Qcow2DiscardRegion; 309 310 typedef uint64_t Qcow2GetRefcountFunc(const void *refcount_array, 311 uint64_t index); 312 typedef void Qcow2SetRefcountFunc(void *refcount_array, 313 uint64_t index, uint64_t value); 314 315 typedef struct Qcow2BitmapHeaderExt { 316 uint32_t nb_bitmaps; 317 uint32_t reserved32; 318 uint64_t bitmap_directory_size; 319 uint64_t bitmap_directory_offset; 320 } QEMU_PACKED Qcow2BitmapHeaderExt; 321 322 #define QCOW2_MAX_THREADS 4 323 324 typedef struct BDRVQcow2State { 325 int cluster_bits; 326 int cluster_size; 327 int l2_slice_size; 328 int subcluster_bits; 329 int subcluster_size; 330 int subclusters_per_cluster; 331 int l2_bits; 332 int l2_size; 333 int l1_size; 334 int l1_vm_state_index; 335 int refcount_block_bits; 336 int refcount_block_size; 337 int csize_shift; 338 int csize_mask; 339 uint64_t cluster_offset_mask; 340 uint64_t l1_table_offset; 341 uint64_t *l1_table; 342 343 Qcow2Cache* l2_table_cache; 344 Qcow2Cache* refcount_block_cache; 345 QEMUTimer *cache_clean_timer; 346 unsigned cache_clean_interval; 347 348 QLIST_HEAD(, QCowL2Meta) cluster_allocs; 349 350 uint64_t *refcount_table; 351 uint64_t refcount_table_offset; 352 uint32_t refcount_table_size; 353 uint32_t max_refcount_table_index; /* Last used entry in refcount_table */ 354 uint64_t free_cluster_index; 355 uint64_t free_byte_offset; 356 357 CoMutex lock; 358 359 Qcow2CryptoHeaderExtension crypto_header; /* QCow2 header extension */ 360 QCryptoBlockOpenOptions *crypto_opts; /* Disk encryption runtime options */ 361 QCryptoBlock *crypto; /* Disk encryption format driver */ 362 bool crypt_physical_offset; /* Whether to use virtual or physical offset 363 for encryption initialization vector tweak */ 364 uint32_t crypt_method_header; 365 uint64_t snapshots_offset; 366 int snapshots_size; 367 unsigned int nb_snapshots; 368 QCowSnapshot *snapshots; 369 370 uint32_t nb_bitmaps; 371 uint64_t bitmap_directory_size; 372 uint64_t bitmap_directory_offset; 373 374 int flags; 375 int qcow_version; 376 bool use_lazy_refcounts; 377 int refcount_order; 378 int refcount_bits; 379 uint64_t refcount_max; 380 381 Qcow2GetRefcountFunc *get_refcount; 382 Qcow2SetRefcountFunc *set_refcount; 383 384 bool discard_passthrough[QCOW2_DISCARD_MAX]; 385 386 int overlap_check; /* bitmask of Qcow2MetadataOverlap values */ 387 bool signaled_corruption; 388 389 uint64_t incompatible_features; 390 uint64_t compatible_features; 391 uint64_t autoclear_features; 392 393 size_t unknown_header_fields_size; 394 void* unknown_header_fields; 395 QLIST_HEAD(, Qcow2UnknownHeaderExtension) unknown_header_ext; 396 QTAILQ_HEAD (, Qcow2DiscardRegion) discards; 397 bool cache_discards; 398 399 /* Backing file path and format as stored in the image (this is not the 400 * effective path/format, which may be the result of a runtime option 401 * override) */ 402 char *image_backing_file; 403 char *image_backing_format; 404 char *image_data_file; 405 406 CoQueue thread_task_queue; 407 int nb_threads; 408 409 BdrvChild *data_file; 410 411 bool metadata_preallocation_checked; 412 bool metadata_preallocation; 413 /* 414 * Compression type used for the image. Default: 0 - ZLIB 415 * The image compression type is set on image creation. 416 * For now, the only way to change the compression type 417 * is to convert the image with the desired compression type set. 418 */ 419 Qcow2CompressionType compression_type; 420 } BDRVQcow2State; 421 422 typedef struct Qcow2COWRegion { 423 /** 424 * Offset of the COW region in bytes from the start of the first cluster 425 * touched by the request. 426 */ 427 unsigned offset; 428 429 /** Number of bytes to copy */ 430 unsigned nb_bytes; 431 } Qcow2COWRegion; 432 433 /** 434 * Describes an in-flight (part of a) write request that writes to clusters 435 * that are not referenced in their L2 table yet. 436 */ 437 typedef struct QCowL2Meta 438 { 439 /** Guest offset of the first newly allocated cluster */ 440 uint64_t offset; 441 442 /** Host offset of the first newly allocated cluster */ 443 uint64_t alloc_offset; 444 445 /** Number of newly allocated clusters */ 446 int nb_clusters; 447 448 /** Do not free the old clusters */ 449 bool keep_old_clusters; 450 451 /** 452 * Requests that overlap with this allocation and wait to be restarted 453 * when the allocating request has completed. 454 */ 455 CoQueue dependent_requests; 456 457 /** 458 * The COW Region between the start of the first allocated cluster and the 459 * area the guest actually writes to. 460 */ 461 Qcow2COWRegion cow_start; 462 463 /** 464 * The COW Region between the area the guest actually writes to and the 465 * end of the last allocated cluster. 466 */ 467 Qcow2COWRegion cow_end; 468 469 /* 470 * Indicates that COW regions are already handled and do not require 471 * any more processing. 472 */ 473 bool skip_cow; 474 475 /** 476 * Indicates that this is not a normal write request but a preallocation. 477 * If the image has extended L2 entries this means that no new individual 478 * subclusters will be marked as allocated in the L2 bitmap (but any 479 * existing contents of that bitmap will be kept). 480 */ 481 bool prealloc; 482 483 /** 484 * The I/O vector with the data from the actual guest write request. 485 * If non-NULL, this is meant to be merged together with the data 486 * from @cow_start and @cow_end into one single write operation. 487 */ 488 QEMUIOVector *data_qiov; 489 size_t data_qiov_offset; 490 491 /** Pointer to next L2Meta of the same write request */ 492 struct QCowL2Meta *next; 493 494 QLIST_ENTRY(QCowL2Meta) next_in_flight; 495 } QCowL2Meta; 496 497 /* 498 * In images with standard L2 entries all clusters are treated as if 499 * they had one subcluster so QCow2ClusterType and QCow2SubclusterType 500 * can be mapped to each other and have the exact same meaning 501 * (QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC cannot happen in these images). 502 * 503 * In images with extended L2 entries QCow2ClusterType refers to the 504 * complete cluster and QCow2SubclusterType to each of the individual 505 * subclusters, so there are several possible combinations: 506 * 507 * |--------------+---------------------------| 508 * | Cluster type | Possible subcluster types | 509 * |--------------+---------------------------| 510 * | UNALLOCATED | UNALLOCATED_PLAIN | 511 * | | ZERO_PLAIN | 512 * |--------------+---------------------------| 513 * | NORMAL | UNALLOCATED_ALLOC | 514 * | | ZERO_ALLOC | 515 * | | NORMAL | 516 * |--------------+---------------------------| 517 * | COMPRESSED | COMPRESSED | 518 * |--------------+---------------------------| 519 * 520 * QCOW2_SUBCLUSTER_INVALID means that the L2 entry is incorrect and 521 * the image should be marked corrupt. 522 */ 523 524 typedef enum QCow2ClusterType { 525 QCOW2_CLUSTER_UNALLOCATED, 526 QCOW2_CLUSTER_ZERO_PLAIN, 527 QCOW2_CLUSTER_ZERO_ALLOC, 528 QCOW2_CLUSTER_NORMAL, 529 QCOW2_CLUSTER_COMPRESSED, 530 } QCow2ClusterType; 531 532 typedef enum QCow2SubclusterType { 533 QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN, 534 QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC, 535 QCOW2_SUBCLUSTER_ZERO_PLAIN, 536 QCOW2_SUBCLUSTER_ZERO_ALLOC, 537 QCOW2_SUBCLUSTER_NORMAL, 538 QCOW2_SUBCLUSTER_COMPRESSED, 539 QCOW2_SUBCLUSTER_INVALID, 540 } QCow2SubclusterType; 541 542 typedef enum QCow2MetadataOverlap { 543 QCOW2_OL_MAIN_HEADER_BITNR = 0, 544 QCOW2_OL_ACTIVE_L1_BITNR = 1, 545 QCOW2_OL_ACTIVE_L2_BITNR = 2, 546 QCOW2_OL_REFCOUNT_TABLE_BITNR = 3, 547 QCOW2_OL_REFCOUNT_BLOCK_BITNR = 4, 548 QCOW2_OL_SNAPSHOT_TABLE_BITNR = 5, 549 QCOW2_OL_INACTIVE_L1_BITNR = 6, 550 QCOW2_OL_INACTIVE_L2_BITNR = 7, 551 QCOW2_OL_BITMAP_DIRECTORY_BITNR = 8, 552 553 QCOW2_OL_MAX_BITNR = 9, 554 555 QCOW2_OL_NONE = 0, 556 QCOW2_OL_MAIN_HEADER = (1 << QCOW2_OL_MAIN_HEADER_BITNR), 557 QCOW2_OL_ACTIVE_L1 = (1 << QCOW2_OL_ACTIVE_L1_BITNR), 558 QCOW2_OL_ACTIVE_L2 = (1 << QCOW2_OL_ACTIVE_L2_BITNR), 559 QCOW2_OL_REFCOUNT_TABLE = (1 << QCOW2_OL_REFCOUNT_TABLE_BITNR), 560 QCOW2_OL_REFCOUNT_BLOCK = (1 << QCOW2_OL_REFCOUNT_BLOCK_BITNR), 561 QCOW2_OL_SNAPSHOT_TABLE = (1 << QCOW2_OL_SNAPSHOT_TABLE_BITNR), 562 QCOW2_OL_INACTIVE_L1 = (1 << QCOW2_OL_INACTIVE_L1_BITNR), 563 /* NOTE: Checking overlaps with inactive L2 tables will result in bdrv 564 * reads. */ 565 QCOW2_OL_INACTIVE_L2 = (1 << QCOW2_OL_INACTIVE_L2_BITNR), 566 QCOW2_OL_BITMAP_DIRECTORY = (1 << QCOW2_OL_BITMAP_DIRECTORY_BITNR), 567 } QCow2MetadataOverlap; 568 569 /* Perform all overlap checks which can be done in constant time */ 570 #define QCOW2_OL_CONSTANT \ 571 (QCOW2_OL_MAIN_HEADER | QCOW2_OL_ACTIVE_L1 | QCOW2_OL_REFCOUNT_TABLE | \ 572 QCOW2_OL_SNAPSHOT_TABLE | QCOW2_OL_BITMAP_DIRECTORY) 573 574 /* Perform all overlap checks which don't require disk access */ 575 #define QCOW2_OL_CACHED \ 576 (QCOW2_OL_CONSTANT | QCOW2_OL_ACTIVE_L2 | QCOW2_OL_REFCOUNT_BLOCK | \ 577 QCOW2_OL_INACTIVE_L1) 578 579 /* Perform all overlap checks */ 580 #define QCOW2_OL_ALL \ 581 (QCOW2_OL_CACHED | QCOW2_OL_INACTIVE_L2) 582 583 #define L1E_OFFSET_MASK 0x00fffffffffffe00ULL 584 #define L2E_OFFSET_MASK 0x00fffffffffffe00ULL 585 #define L2E_COMPRESSED_OFFSET_SIZE_MASK 0x3fffffffffffffffULL 586 587 #define REFT_OFFSET_MASK 0xfffffffffffffe00ULL 588 589 #define INV_OFFSET (-1ULL) 590 591 static inline bool has_subclusters(BDRVQcow2State *s) 592 { 593 return s->incompatible_features & QCOW2_INCOMPAT_EXTL2; 594 } 595 596 static inline size_t l2_entry_size(BDRVQcow2State *s) 597 { 598 return has_subclusters(s) ? L2E_SIZE_EXTENDED : L2E_SIZE_NORMAL; 599 } 600 601 static inline uint64_t get_l2_entry(BDRVQcow2State *s, uint64_t *l2_slice, 602 int idx) 603 { 604 idx *= l2_entry_size(s) / sizeof(uint64_t); 605 return be64_to_cpu(l2_slice[idx]); 606 } 607 608 static inline uint64_t get_l2_bitmap(BDRVQcow2State *s, uint64_t *l2_slice, 609 int idx) 610 { 611 if (has_subclusters(s)) { 612 idx *= l2_entry_size(s) / sizeof(uint64_t); 613 return be64_to_cpu(l2_slice[idx + 1]); 614 } else { 615 return 0; /* For convenience only; this value has no meaning. */ 616 } 617 } 618 619 static inline void set_l2_entry(BDRVQcow2State *s, uint64_t *l2_slice, 620 int idx, uint64_t entry) 621 { 622 idx *= l2_entry_size(s) / sizeof(uint64_t); 623 l2_slice[idx] = cpu_to_be64(entry); 624 } 625 626 static inline void set_l2_bitmap(BDRVQcow2State *s, uint64_t *l2_slice, 627 int idx, uint64_t bitmap) 628 { 629 assert(has_subclusters(s)); 630 idx *= l2_entry_size(s) / sizeof(uint64_t); 631 l2_slice[idx + 1] = cpu_to_be64(bitmap); 632 } 633 634 static inline bool has_data_file(BlockDriverState *bs) 635 { 636 BDRVQcow2State *s = bs->opaque; 637 return (s->data_file != bs->file); 638 } 639 640 static inline bool data_file_is_raw(BlockDriverState *bs) 641 { 642 BDRVQcow2State *s = bs->opaque; 643 return !!(s->autoclear_features & QCOW2_AUTOCLEAR_DATA_FILE_RAW); 644 } 645 646 static inline int64_t start_of_cluster(BDRVQcow2State *s, int64_t offset) 647 { 648 return offset & ~(s->cluster_size - 1); 649 } 650 651 static inline int64_t offset_into_cluster(BDRVQcow2State *s, int64_t offset) 652 { 653 return offset & (s->cluster_size - 1); 654 } 655 656 static inline int64_t offset_into_subcluster(BDRVQcow2State *s, int64_t offset) 657 { 658 return offset & (s->subcluster_size - 1); 659 } 660 661 static inline uint64_t size_to_clusters(BDRVQcow2State *s, uint64_t size) 662 { 663 return (size + (s->cluster_size - 1)) >> s->cluster_bits; 664 } 665 666 static inline uint64_t size_to_subclusters(BDRVQcow2State *s, uint64_t size) 667 { 668 return (size + (s->subcluster_size - 1)) >> s->subcluster_bits; 669 } 670 671 static inline int64_t size_to_l1(BDRVQcow2State *s, int64_t size) 672 { 673 int shift = s->cluster_bits + s->l2_bits; 674 return (size + (1ULL << shift) - 1) >> shift; 675 } 676 677 static inline int offset_to_l1_index(BDRVQcow2State *s, uint64_t offset) 678 { 679 return offset >> (s->l2_bits + s->cluster_bits); 680 } 681 682 static inline int offset_to_l2_index(BDRVQcow2State *s, int64_t offset) 683 { 684 return (offset >> s->cluster_bits) & (s->l2_size - 1); 685 } 686 687 static inline int offset_to_l2_slice_index(BDRVQcow2State *s, int64_t offset) 688 { 689 return (offset >> s->cluster_bits) & (s->l2_slice_size - 1); 690 } 691 692 static inline int offset_to_sc_index(BDRVQcow2State *s, int64_t offset) 693 { 694 return (offset >> s->subcluster_bits) & (s->subclusters_per_cluster - 1); 695 } 696 697 static inline int64_t qcow2_vm_state_offset(BDRVQcow2State *s) 698 { 699 return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits); 700 } 701 702 static inline QCow2ClusterType qcow2_get_cluster_type(BlockDriverState *bs, 703 uint64_t l2_entry) 704 { 705 BDRVQcow2State *s = bs->opaque; 706 707 if (l2_entry & QCOW_OFLAG_COMPRESSED) { 708 return QCOW2_CLUSTER_COMPRESSED; 709 } else if ((l2_entry & QCOW_OFLAG_ZERO) && !has_subclusters(s)) { 710 if (l2_entry & L2E_OFFSET_MASK) { 711 return QCOW2_CLUSTER_ZERO_ALLOC; 712 } 713 return QCOW2_CLUSTER_ZERO_PLAIN; 714 } else if (!(l2_entry & L2E_OFFSET_MASK)) { 715 /* Offset 0 generally means unallocated, but it is ambiguous with 716 * external data files because 0 is a valid offset there. However, all 717 * clusters in external data files always have refcount 1, so we can 718 * rely on QCOW_OFLAG_COPIED to disambiguate. */ 719 if (has_data_file(bs) && (l2_entry & QCOW_OFLAG_COPIED)) { 720 return QCOW2_CLUSTER_NORMAL; 721 } else { 722 return QCOW2_CLUSTER_UNALLOCATED; 723 } 724 } else { 725 return QCOW2_CLUSTER_NORMAL; 726 } 727 } 728 729 /* 730 * In an image without subsclusters @l2_bitmap is ignored and 731 * @sc_index must be 0. 732 * Return QCOW2_SUBCLUSTER_INVALID if an invalid l2 entry is detected 733 * (this checks the whole entry and bitmap, not only the bits related 734 * to subcluster @sc_index). 735 */ 736 static inline 737 QCow2SubclusterType qcow2_get_subcluster_type(BlockDriverState *bs, 738 uint64_t l2_entry, 739 uint64_t l2_bitmap, 740 unsigned sc_index) 741 { 742 BDRVQcow2State *s = bs->opaque; 743 QCow2ClusterType type = qcow2_get_cluster_type(bs, l2_entry); 744 assert(sc_index < s->subclusters_per_cluster); 745 746 if (has_subclusters(s)) { 747 switch (type) { 748 case QCOW2_CLUSTER_COMPRESSED: 749 return QCOW2_SUBCLUSTER_COMPRESSED; 750 case QCOW2_CLUSTER_NORMAL: 751 if ((l2_bitmap >> 32) & l2_bitmap) { 752 return QCOW2_SUBCLUSTER_INVALID; 753 } else if (l2_bitmap & QCOW_OFLAG_SUB_ZERO(sc_index)) { 754 return QCOW2_SUBCLUSTER_ZERO_ALLOC; 755 } else if (l2_bitmap & QCOW_OFLAG_SUB_ALLOC(sc_index)) { 756 return QCOW2_SUBCLUSTER_NORMAL; 757 } else { 758 return QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC; 759 } 760 case QCOW2_CLUSTER_UNALLOCATED: 761 if (l2_bitmap & QCOW_L2_BITMAP_ALL_ALLOC) { 762 return QCOW2_SUBCLUSTER_INVALID; 763 } else if (l2_bitmap & QCOW_OFLAG_SUB_ZERO(sc_index)) { 764 return QCOW2_SUBCLUSTER_ZERO_PLAIN; 765 } else { 766 return QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN; 767 } 768 default: 769 g_assert_not_reached(); 770 } 771 } else { 772 switch (type) { 773 case QCOW2_CLUSTER_COMPRESSED: 774 return QCOW2_SUBCLUSTER_COMPRESSED; 775 case QCOW2_CLUSTER_ZERO_PLAIN: 776 return QCOW2_SUBCLUSTER_ZERO_PLAIN; 777 case QCOW2_CLUSTER_ZERO_ALLOC: 778 return QCOW2_SUBCLUSTER_ZERO_ALLOC; 779 case QCOW2_CLUSTER_NORMAL: 780 return QCOW2_SUBCLUSTER_NORMAL; 781 case QCOW2_CLUSTER_UNALLOCATED: 782 return QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN; 783 default: 784 g_assert_not_reached(); 785 } 786 } 787 } 788 789 static inline bool qcow2_cluster_is_allocated(QCow2ClusterType type) 790 { 791 return (type == QCOW2_CLUSTER_COMPRESSED || type == QCOW2_CLUSTER_NORMAL || 792 type == QCOW2_CLUSTER_ZERO_ALLOC); 793 } 794 795 /* Check whether refcounts are eager or lazy */ 796 static inline bool qcow2_need_accurate_refcounts(BDRVQcow2State *s) 797 { 798 return !(s->incompatible_features & QCOW2_INCOMPAT_DIRTY); 799 } 800 801 static inline uint64_t l2meta_cow_start(QCowL2Meta *m) 802 { 803 return m->offset + m->cow_start.offset; 804 } 805 806 static inline uint64_t l2meta_cow_end(QCowL2Meta *m) 807 { 808 return m->offset + m->cow_end.offset + m->cow_end.nb_bytes; 809 } 810 811 static inline uint64_t refcount_diff(uint64_t r1, uint64_t r2) 812 { 813 return r1 > r2 ? r1 - r2 : r2 - r1; 814 } 815 816 static inline 817 uint32_t offset_to_reftable_index(BDRVQcow2State *s, uint64_t offset) 818 { 819 return offset >> (s->refcount_block_bits + s->cluster_bits); 820 } 821 822 /* qcow2.c functions */ 823 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size, 824 int refcount_order, bool generous_increase, 825 uint64_t *refblock_count); 826 827 int qcow2_mark_dirty(BlockDriverState *bs); 828 int qcow2_mark_corrupt(BlockDriverState *bs); 829 int qcow2_mark_consistent(BlockDriverState *bs); 830 int qcow2_update_header(BlockDriverState *bs); 831 832 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset, 833 int64_t size, const char *message_format, ...) 834 GCC_FMT_ATTR(5, 6); 835 836 int qcow2_validate_table(BlockDriverState *bs, uint64_t offset, 837 uint64_t entries, size_t entry_len, 838 int64_t max_size_bytes, const char *table_name, 839 Error **errp); 840 841 /* qcow2-refcount.c functions */ 842 int qcow2_refcount_init(BlockDriverState *bs); 843 void qcow2_refcount_close(BlockDriverState *bs); 844 845 int qcow2_get_refcount(BlockDriverState *bs, int64_t cluster_index, 846 uint64_t *refcount); 847 848 int qcow2_update_cluster_refcount(BlockDriverState *bs, int64_t cluster_index, 849 uint64_t addend, bool decrease, 850 enum qcow2_discard_type type); 851 852 int64_t qcow2_refcount_area(BlockDriverState *bs, uint64_t offset, 853 uint64_t additional_clusters, bool exact_size, 854 int new_refblock_index, 855 uint64_t new_refblock_offset); 856 857 int64_t qcow2_alloc_clusters(BlockDriverState *bs, uint64_t size); 858 int64_t qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset, 859 int64_t nb_clusters); 860 int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size); 861 void qcow2_free_clusters(BlockDriverState *bs, 862 int64_t offset, int64_t size, 863 enum qcow2_discard_type type); 864 void qcow2_free_any_cluster(BlockDriverState *bs, uint64_t l2_entry, 865 enum qcow2_discard_type type); 866 867 int qcow2_update_snapshot_refcount(BlockDriverState *bs, 868 int64_t l1_table_offset, int l1_size, int addend); 869 870 int coroutine_fn qcow2_flush_caches(BlockDriverState *bs); 871 int coroutine_fn qcow2_write_caches(BlockDriverState *bs); 872 int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res, 873 BdrvCheckMode fix); 874 875 void qcow2_process_discards(BlockDriverState *bs, int ret); 876 877 int qcow2_check_metadata_overlap(BlockDriverState *bs, int ign, int64_t offset, 878 int64_t size); 879 int qcow2_pre_write_overlap_check(BlockDriverState *bs, int ign, int64_t offset, 880 int64_t size, bool data_file); 881 int qcow2_inc_refcounts_imrt(BlockDriverState *bs, BdrvCheckResult *res, 882 void **refcount_table, 883 int64_t *refcount_table_size, 884 int64_t offset, int64_t size); 885 886 int qcow2_change_refcount_order(BlockDriverState *bs, int refcount_order, 887 BlockDriverAmendStatusCB *status_cb, 888 void *cb_opaque, Error **errp); 889 int qcow2_shrink_reftable(BlockDriverState *bs); 890 int64_t qcow2_get_last_cluster(BlockDriverState *bs, int64_t size); 891 int qcow2_detect_metadata_preallocation(BlockDriverState *bs); 892 893 /* qcow2-cluster.c functions */ 894 int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size, 895 bool exact_size); 896 int qcow2_shrink_l1_table(BlockDriverState *bs, uint64_t max_size); 897 int qcow2_write_l1_entry(BlockDriverState *bs, int l1_index); 898 int qcow2_encrypt_sectors(BDRVQcow2State *s, int64_t sector_num, 899 uint8_t *buf, int nb_sectors, bool enc, Error **errp); 900 901 int qcow2_get_host_offset(BlockDriverState *bs, uint64_t offset, 902 unsigned int *bytes, uint64_t *host_offset, 903 QCow2SubclusterType *subcluster_type); 904 int qcow2_alloc_host_offset(BlockDriverState *bs, uint64_t offset, 905 unsigned int *bytes, uint64_t *host_offset, 906 QCowL2Meta **m); 907 int qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs, 908 uint64_t offset, 909 int compressed_size, 910 uint64_t *host_offset); 911 912 int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m); 913 void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m); 914 int qcow2_cluster_discard(BlockDriverState *bs, uint64_t offset, 915 uint64_t bytes, enum qcow2_discard_type type, 916 bool full_discard); 917 int qcow2_subcluster_zeroize(BlockDriverState *bs, uint64_t offset, 918 uint64_t bytes, int flags); 919 920 int qcow2_expand_zero_clusters(BlockDriverState *bs, 921 BlockDriverAmendStatusCB *status_cb, 922 void *cb_opaque); 923 924 /* qcow2-snapshot.c functions */ 925 int qcow2_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info); 926 int qcow2_snapshot_goto(BlockDriverState *bs, const char *snapshot_id); 927 int qcow2_snapshot_delete(BlockDriverState *bs, 928 const char *snapshot_id, 929 const char *name, 930 Error **errp); 931 int qcow2_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab); 932 int qcow2_snapshot_load_tmp(BlockDriverState *bs, 933 const char *snapshot_id, 934 const char *name, 935 Error **errp); 936 937 void qcow2_free_snapshots(BlockDriverState *bs); 938 int qcow2_read_snapshots(BlockDriverState *bs, Error **errp); 939 int qcow2_write_snapshots(BlockDriverState *bs); 940 941 int coroutine_fn qcow2_check_read_snapshot_table(BlockDriverState *bs, 942 BdrvCheckResult *result, 943 BdrvCheckMode fix); 944 int coroutine_fn qcow2_check_fix_snapshot_table(BlockDriverState *bs, 945 BdrvCheckResult *result, 946 BdrvCheckMode fix); 947 948 /* qcow2-cache.c functions */ 949 Qcow2Cache *qcow2_cache_create(BlockDriverState *bs, int num_tables, 950 unsigned table_size); 951 int qcow2_cache_destroy(Qcow2Cache *c); 952 953 void qcow2_cache_entry_mark_dirty(Qcow2Cache *c, void *table); 954 int qcow2_cache_flush(BlockDriverState *bs, Qcow2Cache *c); 955 int qcow2_cache_write(BlockDriverState *bs, Qcow2Cache *c); 956 int qcow2_cache_set_dependency(BlockDriverState *bs, Qcow2Cache *c, 957 Qcow2Cache *dependency); 958 void qcow2_cache_depends_on_flush(Qcow2Cache *c); 959 960 void qcow2_cache_clean_unused(Qcow2Cache *c); 961 int qcow2_cache_empty(BlockDriverState *bs, Qcow2Cache *c); 962 963 int qcow2_cache_get(BlockDriverState *bs, Qcow2Cache *c, uint64_t offset, 964 void **table); 965 int qcow2_cache_get_empty(BlockDriverState *bs, Qcow2Cache *c, uint64_t offset, 966 void **table); 967 void qcow2_cache_put(Qcow2Cache *c, void **table); 968 void *qcow2_cache_is_table_offset(Qcow2Cache *c, uint64_t offset); 969 void qcow2_cache_discard(Qcow2Cache *c, void *table); 970 971 /* qcow2-bitmap.c functions */ 972 int qcow2_check_bitmaps_refcounts(BlockDriverState *bs, BdrvCheckResult *res, 973 void **refcount_table, 974 int64_t *refcount_table_size); 975 bool qcow2_load_dirty_bitmaps(BlockDriverState *bs, Error **errp); 976 Qcow2BitmapInfoList *qcow2_get_bitmap_info_list(BlockDriverState *bs, 977 Error **errp); 978 int qcow2_reopen_bitmaps_rw(BlockDriverState *bs, Error **errp); 979 int qcow2_truncate_bitmaps_check(BlockDriverState *bs, Error **errp); 980 void qcow2_store_persistent_dirty_bitmaps(BlockDriverState *bs, 981 bool release_stored, Error **errp); 982 int qcow2_reopen_bitmaps_ro(BlockDriverState *bs, Error **errp); 983 bool qcow2_co_can_store_new_dirty_bitmap(BlockDriverState *bs, 984 const char *name, 985 uint32_t granularity, 986 Error **errp); 987 int qcow2_co_remove_persistent_dirty_bitmap(BlockDriverState *bs, 988 const char *name, 989 Error **errp); 990 bool qcow2_supports_persistent_dirty_bitmap(BlockDriverState *bs); 991 uint64_t qcow2_get_persistent_dirty_bitmap_size(BlockDriverState *bs, 992 uint32_t cluster_size); 993 994 ssize_t coroutine_fn 995 qcow2_co_compress(BlockDriverState *bs, void *dest, size_t dest_size, 996 const void *src, size_t src_size); 997 ssize_t coroutine_fn 998 qcow2_co_decompress(BlockDriverState *bs, void *dest, size_t dest_size, 999 const void *src, size_t src_size); 1000 int coroutine_fn 1001 qcow2_co_encrypt(BlockDriverState *bs, uint64_t host_offset, 1002 uint64_t guest_offset, void *buf, size_t len); 1003 int coroutine_fn 1004 qcow2_co_decrypt(BlockDriverState *bs, uint64_t host_offset, 1005 uint64_t guest_offset, void *buf, size_t len); 1006 1007 #endif 1008