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 "qemu/aes.h"
  29 #include "block/coroutine.h"
  30
  31 //#define DEBUG_ALLOC
  32 //#define DEBUG_ALLOC2
  33 //#define DEBUG_EXT
  34
  35 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
  36
  37 #define QCOW_CRYPT_NONE 0
  38 #define QCOW_CRYPT_AES  1
  39
  40 #define QCOW_MAX_CRYPT_CLUSTERS 32
  41
  42 /* indicate that the refcount of the referenced cluster is exactly one. */
  43 #define QCOW_OFLAG_COPIED     (1LL << 63)
  44 /* indicate that the cluster is compressed (they never have the copied flag) */
  45 #define QCOW_OFLAG_COMPRESSED (1LL << 62)
  46 /* The cluster reads as all zeros */
  47 #define QCOW_OFLAG_ZERO (1LL << 0)
  48
  49 #define REFCOUNT_SHIFT 1 /* refcount size is 2 bytes */
  50
  51 #define MIN_CLUSTER_BITS 9
  52 #define MAX_CLUSTER_BITS 21
  53
  54 #define L2_CACHE_SIZE 16
  55
  56 /* Must be at least 4 to cover all cases of refcount table growth */
  57 #define REFCOUNT_CACHE_SIZE 4
  58
  59 #define DEFAULT_CLUSTER_SIZE 65536
  60
  61
  62 #define QCOW2_OPT_LAZY_REFCOUNTS "lazy_refcounts"
  63 #define QCOW2_OPT_DISCARD_REQUEST "pass_discard_request"
  64 #define QCOW2_OPT_DISCARD_SNAPSHOT "pass_discard_snapshot"
  65 #define QCOW2_OPT_DISCARD_OTHER "pass_discard_other"
  66
  67 typedef struct QCowHeader {
  68     uint32_t magic;
  69     uint32_t version;
  70     uint64_t backing_file_offset;
  71     uint32_t backing_file_size;
  72     uint32_t cluster_bits;
  73     uint64_t size; /* in bytes */
  74     uint32_t crypt_method;
  75     uint32_t l1_size; /* XXX: save number of clusters instead ? */
  76     uint64_t l1_table_offset;
  77     uint64_t refcount_table_offset;
  78     uint32_t refcount_table_clusters;
  79     uint32_t nb_snapshots;
  80     uint64_t snapshots_offset;
  81
  82     /* The following fields are only valid for version >= 3 */
  83     uint64_t incompatible_features;
  84     uint64_t compatible_features;
  85     uint64_t autoclear_features;
  86
  87     uint32_t refcount_order;
  88     uint32_t header_length;
  89 } QCowHeader;
  90
  91 typedef struct QCowSnapshot {
  92     uint64_t l1_table_offset;
  93     uint32_t l1_size;
  94     char *id_str;
  95     char *name;
  96     uint64_t disk_size;
  97     uint64_t vm_state_size;
  98     uint32_t date_sec;
  99     uint32_t date_nsec;
 100     uint64_t vm_clock_nsec;
 101 } QCowSnapshot;
 102
 103 struct Qcow2Cache;
 104 typedef struct Qcow2Cache Qcow2Cache;
 105
 106 typedef struct Qcow2UnknownHeaderExtension {
 107     uint32_t magic;
 108     uint32_t len;
 109     QLIST_ENTRY(Qcow2UnknownHeaderExtension) next;
 110     uint8_t data[];
 111 } Qcow2UnknownHeaderExtension;
 112
 113 enum {
 114     QCOW2_FEAT_TYPE_INCOMPATIBLE    = 0,
 115     QCOW2_FEAT_TYPE_COMPATIBLE      = 1,
 116     QCOW2_FEAT_TYPE_AUTOCLEAR       = 2,
 117 };
 118
 119 /* Incompatible feature bits */
 120 enum {
 121     QCOW2_INCOMPAT_DIRTY_BITNR   = 0,
 122     QCOW2_INCOMPAT_DIRTY         = 1 << QCOW2_INCOMPAT_DIRTY_BITNR,
 123
 124     QCOW2_INCOMPAT_MASK          = QCOW2_INCOMPAT_DIRTY,
 125 };
 126
 127 /* Compatible feature bits */
 128 enum {
 129     QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR = 0,
 130     QCOW2_COMPAT_LAZY_REFCOUNTS       = 1 << QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
 131
 132     QCOW2_COMPAT_FEAT_MASK            = QCOW2_COMPAT_LAZY_REFCOUNTS,
 133 };
 134
 135 enum qcow2_discard_type {
 136     QCOW2_DISCARD_NEVER = 0,
 137     QCOW2_DISCARD_ALWAYS,
 138     QCOW2_DISCARD_REQUEST,
 139     QCOW2_DISCARD_SNAPSHOT,
 140     QCOW2_DISCARD_OTHER,
 141     QCOW2_DISCARD_MAX
 142 };
 143
 144 typedef struct Qcow2Feature {
 145     uint8_t type;
 146     uint8_t bit;
 147     char    name[46];
 148 } QEMU_PACKED Qcow2Feature;
 149
 150 typedef struct Qcow2DiscardRegion {
 151     BlockDriverState *bs;
 152     uint64_t offset;
 153     uint64_t bytes;
 154     QTAILQ_ENTRY(Qcow2DiscardRegion) next;
 155 } Qcow2DiscardRegion;
 156
 157 typedef struct BDRVQcowState {
 158     int cluster_bits;
 159     int cluster_size;
 160     int cluster_sectors;
 161     int l2_bits;
 162     int l2_size;
 163     int l1_size;
 164     int l1_vm_state_index;
 165     int csize_shift;
 166     int csize_mask;
 167     uint64_t cluster_offset_mask;
 168     uint64_t l1_table_offset;
 169     uint64_t *l1_table;
 170
 171     Qcow2Cache* l2_table_cache;
 172     Qcow2Cache* refcount_block_cache;
 173
 174     uint8_t *cluster_cache;
 175     uint8_t *cluster_data;
 176     uint64_t cluster_cache_offset;
 177     QLIST_HEAD(QCowClusterAlloc, QCowL2Meta) cluster_allocs;
 178
 179     uint64_t *refcount_table;
 180     uint64_t refcount_table_offset;
 181     uint32_t refcount_table_size;
 182     int64_t free_cluster_index;
 183     int64_t free_byte_offset;
 184
 185     CoMutex lock;
 186
 187     uint32_t crypt_method; /* current crypt method, 0 if no key yet */
 188     uint32_t crypt_method_header;
 189     AES_KEY aes_encrypt_key;
 190     AES_KEY aes_decrypt_key;
 191     uint64_t snapshots_offset;
 192     int snapshots_size;
 193     int nb_snapshots;
 194     QCowSnapshot *snapshots;
 195
 196     int flags;
 197     int qcow_version;
 198     bool use_lazy_refcounts;
 199
 200     bool discard_passthrough[QCOW2_DISCARD_MAX];
 201
 202     uint64_t incompatible_features;
 203     uint64_t compatible_features;
 204     uint64_t autoclear_features;
 205
 206     size_t unknown_header_fields_size;
 207     void* unknown_header_fields;
 208     QLIST_HEAD(, Qcow2UnknownHeaderExtension) unknown_header_ext;
 209     QTAILQ_HEAD (, Qcow2DiscardRegion) discards;
 210     bool cache_discards;
 211 } BDRVQcowState;
 212
 213 /* XXX: use std qcow open function ? */
 214 typedef struct QCowCreateState {
 215     int cluster_size;
 216     int cluster_bits;
 217     uint16_t *refcount_block;
 218     uint64_t *refcount_table;
 219     int64_t l1_table_offset;
 220     int64_t refcount_table_offset;
 221     int64_t refcount_block_offset;
 222 } QCowCreateState;
 223
 224 struct QCowAIOCB;
 225
 226 typedef struct Qcow2COWRegion {
 227     /**
 228      * Offset of the COW region in bytes from the start of the first cluster
 229      * touched by the request.
 230      */
 231     uint64_t    offset;
 232
 233     /** Number of sectors to copy */
 234     int         nb_sectors;
 235 } Qcow2COWRegion;
 236
 237 /**
 238  * Describes an in-flight (part of a) write request that writes to clusters
 239  * that are not referenced in their L2 table yet.
 240  */
 241 typedef struct QCowL2Meta
 242 {
 243     /** Guest offset of the first newly allocated cluster */
 244     uint64_t offset;
 245
 246     /** Host offset of the first newly allocated cluster */
 247     uint64_t alloc_offset;
 248
 249     /**
 250      * Number of sectors from the start of the first allocated cluster to
 251      * the end of the (possibly shortened) request
 252      */
 253     int nb_available;
 254
 255     /** Number of newly allocated clusters */
 256     int nb_clusters;
 257
 258     /**
 259      * Requests that overlap with this allocation and wait to be restarted
 260      * when the allocating request has completed.
 261      */
 262     CoQueue dependent_requests;
 263
 264     /**
 265      * The COW Region between the start of the first allocated cluster and the
 266      * area the guest actually writes to.
 267      */
 268     Qcow2COWRegion cow_start;
 269
 270     /**
 271      * The COW Region between the area the guest actually writes to and the
 272      * end of the last allocated cluster.
 273      */
 274     Qcow2COWRegion cow_end;
 275
 276     /** Pointer to next L2Meta of the same write request */
 277     struct QCowL2Meta *next;
 278
 279     QLIST_ENTRY(QCowL2Meta) next_in_flight;
 280 } QCowL2Meta;
 281
 282 enum {
 283     QCOW2_CLUSTER_UNALLOCATED,
 284     QCOW2_CLUSTER_NORMAL,
 285     QCOW2_CLUSTER_COMPRESSED,
 286     QCOW2_CLUSTER_ZERO
 287 };
 288
 289 #define L1E_OFFSET_MASK 0x00ffffffffffff00ULL
 290 #define L2E_OFFSET_MASK 0x00ffffffffffff00ULL
 291 #define L2E_COMPRESSED_OFFSET_SIZE_MASK 0x3fffffffffffffffULL
 292
 293 #define REFT_OFFSET_MASK 0xffffffffffffff00ULL
 294
 295 static inline int64_t start_of_cluster(BDRVQcowState *s, int64_t offset)
 296 {
 297     return offset & ~(s->cluster_size - 1);
 298 }
 299
 300 static inline int64_t offset_into_cluster(BDRVQcowState *s, int64_t offset)
 301 {
 302     return offset & (s->cluster_size - 1);
 303 }
 304
 305 static inline int size_to_clusters(BDRVQcowState *s, int64_t size)
 306 {
 307     return (size + (s->cluster_size - 1)) >> s->cluster_bits;
 308 }
 309
 310 static inline int64_t size_to_l1(BDRVQcowState *s, int64_t size)
 311 {
 312     int shift = s->cluster_bits + s->l2_bits;
 313     return (size + (1ULL << shift) - 1) >> shift;
 314 }
 315
 316 static inline int offset_to_l2_index(BDRVQcowState *s, int64_t offset)
 317 {
 318     return (offset >> s->cluster_bits) & (s->l2_size - 1);
 319 }
 320
 321 static inline int64_t align_offset(int64_t offset, int n)
 322 {
 323     offset = (offset + n - 1) & ~(n - 1);
 324     return offset;
 325 }
 326
 327 static inline int qcow2_get_cluster_type(uint64_t l2_entry)
 328 {
 329     if (l2_entry & QCOW_OFLAG_COMPRESSED) {
 330         return QCOW2_CLUSTER_COMPRESSED;
 331     } else if (l2_entry & QCOW_OFLAG_ZERO) {
 332         return QCOW2_CLUSTER_ZERO;
 333     } else if (!(l2_entry & L2E_OFFSET_MASK)) {
 334         return QCOW2_CLUSTER_UNALLOCATED;
 335     } else {
 336         return QCOW2_CLUSTER_NORMAL;
 337     }
 338 }
 339
 340 /* Check whether refcounts are eager or lazy */
 341 static inline bool qcow2_need_accurate_refcounts(BDRVQcowState *s)
 342 {
 343     return !(s->incompatible_features & QCOW2_INCOMPAT_DIRTY);
 344 }
 345
 346 static inline uint64_t l2meta_cow_start(QCowL2Meta *m)
 347 {
 348     return m->offset + m->cow_start.offset;
 349 }
 350
 351 static inline uint64_t l2meta_cow_end(QCowL2Meta *m)
 352 {
 353     return m->offset + m->cow_end.offset
 354         + (m->cow_end.nb_sectors << BDRV_SECTOR_BITS);
 355 }
 356
 357 // FIXME Need qcow2_ prefix to global functions
 358
 359 /* qcow2.c functions */
 360 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
 361                   int64_t sector_num, int nb_sectors);
 362
 363 int qcow2_mark_dirty(BlockDriverState *bs);
 364 int qcow2_update_header(BlockDriverState *bs);
 365
 366 /* qcow2-refcount.c functions */
 367 int qcow2_refcount_init(BlockDriverState *bs);
 368 void qcow2_refcount_close(BlockDriverState *bs);
 369
 370 int64_t qcow2_alloc_clusters(BlockDriverState *bs, int64_t size);
 371 int qcow2_alloc_clusters_at(BlockDriverState *bs, uint64_t offset,
 372     int nb_clusters);
 373 int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size);
 374 void qcow2_free_clusters(BlockDriverState *bs,
 375                           int64_t offset, int64_t size,
 376                           enum qcow2_discard_type type);
 377 void qcow2_free_any_clusters(BlockDriverState *bs, uint64_t l2_entry,
 378                              int nb_clusters, enum qcow2_discard_type type);
 379
 380 int qcow2_update_snapshot_refcount(BlockDriverState *bs,
 381     int64_t l1_table_offset, int l1_size, int addend);
 382
 383 int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
 384                           BdrvCheckMode fix);
 385
 386 void qcow2_process_discards(BlockDriverState *bs, int ret);
 387
 388 /* qcow2-cluster.c functions */
 389 int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size,
 390                         bool exact_size);
 391 void qcow2_l2_cache_reset(BlockDriverState *bs);
 392 int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
 393 void qcow2_encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
 394                      uint8_t *out_buf, const uint8_t *in_buf,
 395                      int nb_sectors, int enc,
 396                      const AES_KEY *key);
 397
 398 int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
 399     int *num, uint64_t *cluster_offset);
 400 int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset,
 401     int n_start, int n_end, int *num, uint64_t *host_offset, QCowL2Meta **m);
 402 uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs,
 403                                          uint64_t offset,
 404                                          int compressed_size);
 405
 406 int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m);
 407 int qcow2_discard_clusters(BlockDriverState *bs, uint64_t offset,
 408     int nb_sectors);
 409 int qcow2_zero_clusters(BlockDriverState *bs, uint64_t offset, int nb_sectors);
 410
 411 /* qcow2-snapshot.c functions */
 412 int qcow2_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info);
 413 int qcow2_snapshot_goto(BlockDriverState *bs, const char *snapshot_id);
 414 int qcow2_snapshot_delete(BlockDriverState *bs, const char *snapshot_id);
 415 int qcow2_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab);
 416 int qcow2_snapshot_load_tmp(BlockDriverState *bs, const char *snapshot_name);
 417
 418 void qcow2_free_snapshots(BlockDriverState *bs);
 419 int qcow2_read_snapshots(BlockDriverState *bs);
 420
 421 /* qcow2-cache.c functions */
 422 Qcow2Cache *qcow2_cache_create(BlockDriverState *bs, int num_tables);
 423 int qcow2_cache_destroy(BlockDriverState* bs, Qcow2Cache *c);
 424
 425 void qcow2_cache_entry_mark_dirty(Qcow2Cache *c, void *table);
 426 int qcow2_cache_flush(BlockDriverState *bs, Qcow2Cache *c);
 427 int qcow2_cache_set_dependency(BlockDriverState *bs, Qcow2Cache *c,
 428     Qcow2Cache *dependency);
 429 void qcow2_cache_depends_on_flush(Qcow2Cache *c);
 430
 431 int qcow2_cache_get(BlockDriverState *bs, Qcow2Cache *c, uint64_t offset,
 432     void **table);
 433 int qcow2_cache_get_empty(BlockDriverState *bs, Qcow2Cache *c, uint64_t offset,
 434     void **table);
 435 int qcow2_cache_put(BlockDriverState *bs, Qcow2Cache *c, void **table);
 436
 437 #endif