block/vhdx.h

   1 /*
   2  * Block driver for Hyper-V VHDX Images
   3  *
   4  * Copyright (c) 2013 Red Hat, Inc.,
   5  *
   6  * Authors:
   7  *  Jeff Cody <jcody@redhat.com>
   8  *
   9  *  This is based on the "VHDX Format Specification v1.00", published 8/25/2012
  10  *  by Microsoft:
  11  *      https://www.microsoft.com/en-us/download/details.aspx?id=34750
  12  *
  13  * This work is licensed under the terms of the GNU LGPL, version 2 or later.
  14  * See the COPYING.LIB file in the top-level directory.
  15  *
  16  */
  17
  18 #ifndef BLOCK_VHDX_H
  19 #define BLOCK_VHDX_H
  20
  21 #define KiB              (1 * 1024)
  22 #define MiB            (KiB * 1024)
  23 #define GiB            (MiB * 1024)
  24 #define TiB ((uint64_t) GiB * 1024)
  25
  26 #define DEFAULT_LOG_SIZE 1048576 /* 1MiB */
  27 /* Structures and fields present in the VHDX file */
  28
  29 /* The header section has the following blocks,
  30  * each block is 64KB:
  31  *
  32  * _____________________________________________________________________________
  33  * | File Id. |   Header 1    | Header 2   | Region Table |  Reserved (768KB)  |
  34  * |----------|---------------|------------|--------------|--------------------|
  35  * |          |               |            |              |                    |
  36  * 0.........64KB...........128KB........192KB..........256KB................1MB
  37  */
  38
  39 #define VHDX_HEADER_BLOCK_SIZE      (64 * 1024)
  40
  41 #define VHDX_FILE_ID_OFFSET         0
  42 #define VHDX_HEADER1_OFFSET         (VHDX_HEADER_BLOCK_SIZE * 1)
  43 #define VHDX_HEADER2_OFFSET         (VHDX_HEADER_BLOCK_SIZE * 2)
  44 #define VHDX_REGION_TABLE_OFFSET    (VHDX_HEADER_BLOCK_SIZE * 3)
  45 #define VHDX_REGION_TABLE2_OFFSET   (VHDX_HEADER_BLOCK_SIZE * 4)
  46
  47 #define VHDX_HEADER_SECTION_END     (1 * MiB)
  48 /*
  49  * A note on the use of MS-GUID fields.  For more details on the GUID,
  50  * please see: https://en.wikipedia.org/wiki/Globally_unique_identifier.
  51  *
  52  * The VHDX specification only states that these are MS GUIDs, and which
  53  * bytes are data1-data4. It makes no mention of what algorithm should be used
  54  * to generate the GUID, nor what standard.  However, looking at the specified
  55  * known GUID fields, it appears the GUIDs are:
  56  *  Standard/DCE GUID type  (noted by 10b in the MSB of byte 0 of .data4)
  57  *  Random algorithm        (noted by 0x4XXX for .data3)
  58  */
  59
  60 /* ---- HEADER SECTION STRUCTURES ---- */
  61
  62 /* These structures are ones that are defined in the VHDX specification
  63  * document */
  64
  65 #define VHDX_FILE_SIGNATURE 0x656C696678646876ULL  /* "vhdxfile" in ASCII */
  66 typedef struct VHDXFileIdentifier {
  67     uint64_t    signature;              /* "vhdxfile" in ASCII */
  68     uint16_t    creator[256];           /* optional; utf-16 string to identify
  69                                            the vhdx file creator.  Diagnostic
  70                                            only */
  71 } VHDXFileIdentifier;
  72
  73
  74 /* the guid is a 16 byte unique ID - the definition for this used by
  75  * Microsoft is not just 16 bytes though - it is a structure that is defined,
  76  * so we need to follow it here so that endianness does not trip us up */
  77
  78 typedef struct QEMU_PACKED MSGUID {
  79     uint32_t  data1;
  80     uint16_t  data2;
  81     uint16_t  data3;
  82     uint8_t   data4[8];
  83 } MSGUID;
  84
  85 #define guid_eq(a, b) \
  86     (memcmp(&(a), &(b), sizeof(MSGUID)) == 0)
  87
  88 #define VHDX_HEADER_SIZE (4 * 1024)   /* although the vhdx_header struct in disk
  89                                          is only 582 bytes, for purposes of crc
  90                                          the header is the first 4KB of the 64KB
  91                                          block */
  92
  93 /* The full header is 4KB, although the actual header data is much smaller.
  94  * But for the checksum calculation, it is over the entire 4KB structure,
  95  * not just the defined portion of it */
  96 #define VHDX_HEADER_SIGNATURE 0x64616568
  97 typedef struct QEMU_PACKED VHDXHeader {
  98     uint32_t    signature;              /* "head" in ASCII */
  99     uint32_t    checksum;               /* CRC-32C hash of the whole header */
 100     uint64_t    sequence_number;        /* Seq number of this header.  Each
 101                                            VHDX file has 2 of these headers,
 102                                            and only the header with the highest
 103                                            sequence number is valid */
 104     MSGUID      file_write_guid;        /* 128 bit unique identifier. Must be
 105                                            updated to new, unique value before
 106                                            the first modification is made to
 107                                            file */
 108     MSGUID      data_write_guid;        /* 128 bit unique identifier. Must be
 109                                            updated to new, unique value before
 110                                            the first modification is made to
 111                                            visible data.   Visbile data is
 112                                            defined as:
 113                                                     - system & user metadata
 114                                                     - raw block data
 115                                                     - disk size
 116                                                     - any change that will
 117                                                       cause the virtual disk
 118                                                       sector read to differ
 119
 120                                            This does not need to change if
 121                                            blocks are re-arranged */
 122     MSGUID      log_guid;               /* 128 bit unique identifier. If zero,
 123                                            there is no valid log. If non-zero,
 124                                            log entries with this guid are
 125                                            valid. */
 126     uint16_t    log_version;            /* version of the log format. Must be
 127                                            set to zero */
 128     uint16_t    version;                /* version of the vhdx file.  Currently,
 129                                            only supported version is "1" */
 130     uint32_t    log_length;             /* length of the log.  Must be multiple
 131                                            of 1MB */
 132     uint64_t    log_offset;             /* byte offset in the file of the log.
 133                                            Must also be a multiple of 1MB */
 134 } VHDXHeader;
 135
 136 /* Header for the region table block */
 137 #define VHDX_REGION_SIGNATURE  0x69676572  /* "regi" in ASCII */
 138 typedef struct QEMU_PACKED VHDXRegionTableHeader {
 139     uint32_t    signature;              /* "regi" in ASCII */
 140     uint32_t    checksum;               /* CRC-32C hash of the 64KB table */
 141     uint32_t    entry_count;            /* number of valid entries */
 142     uint32_t    reserved;
 143 } VHDXRegionTableHeader;
 144
 145 /* Individual region table entry.  There may be a maximum of 2047 of these
 146  *
 147  *  There are two known region table properties.  Both are required.
 148  *  BAT (block allocation table):  2DC27766F62342009D64115E9BFD4A08
 149  *  Metadata:                      8B7CA20647904B9AB8FE575F050F886E
 150  */
 151 #define VHDX_REGION_ENTRY_REQUIRED  0x01    /* if set, parser must understand
 152                                                this entry in order to open
 153                                                file */
 154 typedef struct QEMU_PACKED VHDXRegionTableEntry {
 155     MSGUID      guid;                   /* 128-bit unique identifier */
 156     uint64_t    file_offset;            /* offset of the object in the file.
 157                                            Must be multiple of 1MB */
 158     uint32_t    length;                 /* length, in bytes, of the object */
 159     uint32_t    data_bits;
 160 } VHDXRegionTableEntry;
 161
 162
 163 /* ---- LOG ENTRY STRUCTURES ---- */
 164 #define VHDX_LOG_MIN_SIZE (1024 * 1024)
 165 #define VHDX_LOG_SECTOR_SIZE 4096
 166 #define VHDX_LOG_HDR_SIZE 64
 167 #define VHDX_LOG_SIGNATURE 0x65676f6c
 168 typedef struct QEMU_PACKED VHDXLogEntryHeader {
 169     uint32_t    signature;              /* "loge" in ASCII */
 170     uint32_t    checksum;               /* CRC-32C hash of the 64KB table */
 171     uint32_t    entry_length;           /* length in bytes, multiple of 1MB */
 172     uint32_t    tail;                   /* byte offset of first log entry of a
 173                                            seq, where this entry is the last
 174                                            entry */
 175     uint64_t    sequence_number;        /* incremented with each log entry.
 176                                            May not be zero. */
 177     uint32_t    descriptor_count;       /* number of descriptors in this log
 178                                            entry, must be >= 0 */
 179     uint32_t    reserved;
 180     MSGUID      log_guid;               /* value of the log_guid from
 181                                            vhdx_header.  If not found in
 182                                            vhdx_header, it is invalid */
 183     uint64_t    flushed_file_offset;    /* see spec for full details - this
 184                                            should be vhdx file size in bytes */
 185     uint64_t    last_file_offset;       /* size in bytes that all allocated
 186                                            file structures fit into */
 187 } VHDXLogEntryHeader;
 188
 189 #define VHDX_LOG_DESC_SIZE 32
 190 #define VHDX_LOG_DESC_SIGNATURE 0x63736564
 191 #define VHDX_LOG_ZERO_SIGNATURE 0x6f72657a
 192 typedef struct QEMU_PACKED VHDXLogDescriptor {
 193     uint32_t    signature;              /* "zero" or "desc" in ASCII */
 194     union  {
 195         uint32_t    reserved;           /* zero desc */
 196         uint32_t    trailing_bytes;     /* data desc: bytes 4092-4096 of the
 197                                            data sector */
 198     };
 199     union {
 200         uint64_t    zero_length;        /* zero desc: length of the section to
 201                                            zero */
 202         uint64_t    leading_bytes;      /* data desc: bytes 0-7 of the data
 203                                            sector */
 204     };
 205     uint64_t    file_offset;            /* file offset to write zeros - multiple
 206                                            of 4kB */
 207     uint64_t    sequence_number;        /* must match same field in
 208                                            vhdx_log_entry_header */
 209 } VHDXLogDescriptor;
 210
 211 #define VHDX_LOG_DATA_SIGNATURE 0x61746164
 212 typedef struct QEMU_PACKED VHDXLogDataSector {
 213     uint32_t    data_signature;         /* "data" in ASCII */
 214     uint32_t    sequence_high;          /* 4 MSB of 8 byte sequence_number */
 215     uint8_t     data[4084];             /* raw data, bytes 8-4091 (inclusive).
 216                                            see the data descriptor field for the
 217                                            other mising bytes */
 218     uint32_t    sequence_low;           /* 4 LSB of 8 byte sequence_number */
 219 } VHDXLogDataSector;
 220
 221
 222
 223 /* block states - different state values depending on whether it is a
 224  * payload block, or a sector block. */
 225
 226 #define PAYLOAD_BLOCK_NOT_PRESENT       0
 227 #define PAYLOAD_BLOCK_UNDEFINED         1
 228 #define PAYLOAD_BLOCK_ZERO              2
 229 #define PAYLOAD_BLOCK_UNMAPPED          3
 230 #define PAYLOAD_BLOCK_UNMAPPED_v095     5
 231 #define PAYLOAD_BLOCK_FULLY_PRESENT     6
 232 #define PAYLOAD_BLOCK_PARTIALLY_PRESENT 7
 233
 234 #define SB_BLOCK_NOT_PRESENT    0
 235 #define SB_BLOCK_PRESENT        6
 236
 237 /* per the spec */
 238 #define VHDX_MAX_SECTORS_PER_BLOCK  (1 << 23)
 239
 240 /* upper 44 bits are the file offset in 1MB units lower 3 bits are the state
 241    other bits are reserved */
 242 #define VHDX_BAT_STATE_BIT_MASK 0x07
 243 #define VHDX_BAT_FILE_OFF_MASK  0xFFFFFFFFFFF00000ULL /* upper 44 bits */
 244 typedef uint64_t VHDXBatEntry;
 245
 246 /* ---- METADATA REGION STRUCTURES ---- */
 247
 248 #define VHDX_METADATA_ENTRY_SIZE 32
 249 #define VHDX_METADATA_MAX_ENTRIES 2047  /* not including the header */
 250 #define VHDX_METADATA_TABLE_MAX_SIZE \
 251     (VHDX_METADATA_ENTRY_SIZE * (VHDX_METADATA_MAX_ENTRIES+1))
 252 #define VHDX_METADATA_SIGNATURE 0x617461646174656DULL  /* "metadata" in ASCII */
 253 typedef struct QEMU_PACKED VHDXMetadataTableHeader {
 254     uint64_t    signature;              /* "metadata" in ASCII */
 255     uint16_t    reserved;
 256     uint16_t    entry_count;            /* number table entries. <= 2047 */
 257     uint32_t    reserved2[5];
 258 } VHDXMetadataTableHeader;
 259
 260 #define VHDX_META_FLAGS_IS_USER         0x01    /* max 1024 entries */
 261 #define VHDX_META_FLAGS_IS_VIRTUAL_DISK 0x02    /* virtual disk metadata if set,
 262                                                    otherwise file metdata */
 263 #define VHDX_META_FLAGS_IS_REQUIRED     0x04    /* parse must understand this
 264                                                    entry to open the file */
 265 typedef struct QEMU_PACKED VHDXMetadataTableEntry {
 266     MSGUID      item_id;                /* 128-bit identifier for metadata */
 267     uint32_t    offset;                 /* byte offset of the metadata.  At
 268                                            least 64kB.  Relative to start of
 269                                            metadata region */
 270                                         /* note: if length = 0, so is offset */
 271     uint32_t    length;                 /* length of metadata. <= 1MB. */
 272     uint32_t    data_bits;              /* least-significant 3 bits are flags,
 273                                            the rest are reserved (see above) */
 274     uint32_t    reserved2;
 275 } VHDXMetadataTableEntry;
 276
 277 #define VHDX_PARAMS_LEAVE_BLOCKS_ALLOCED 0x01   /* Do not change any blocks to
 278                                                    be BLOCK_NOT_PRESENT.
 279                                                    If set indicates a fixed
 280                                                    size VHDX file */
 281 #define VHDX_PARAMS_HAS_PARENT           0x02    /* has parent / backing file */
 282 #define VHDX_BLOCK_SIZE_MIN             (1   * MiB)
 283 #define VHDX_BLOCK_SIZE_MAX             (256 * MiB)
 284 typedef struct QEMU_PACKED VHDXFileParameters {
 285     uint32_t    block_size;             /* size of each payload block, always
 286                                            power of 2, <= 256MB and >= 1MB. */
 287     uint32_t data_bits;                 /* least-significant 2 bits are flags,
 288                                            the rest are reserved (see above) */
 289 } VHDXFileParameters;
 290
 291 #define VHDX_MAX_IMAGE_SIZE  ((uint64_t) 64 * TiB)
 292 typedef struct QEMU_PACKED VHDXVirtualDiskSize {
 293     uint64_t    virtual_disk_size;      /* Size of the virtual disk, in bytes.
 294                                            Must be multiple of the sector size,
 295                                            max of 64TB */
 296 } VHDXVirtualDiskSize;
 297
 298 typedef struct QEMU_PACKED VHDXPage83Data {
 299     MSGUID      page_83_data;           /* unique id for scsi devices that
 300                                            support page 0x83 */
 301 } VHDXPage83Data;
 302
 303 typedef struct QEMU_PACKED VHDXVirtualDiskLogicalSectorSize {
 304     uint32_t    logical_sector_size;    /* virtual disk sector size (in bytes).
 305                                            Can only be 512 or 4096 bytes */
 306 } VHDXVirtualDiskLogicalSectorSize;
 307
 308 typedef struct QEMU_PACKED VHDXVirtualDiskPhysicalSectorSize {
 309     uint32_t    physical_sector_size;   /* physical sector size (in bytes).
 310                                            Can only be 512 or 4096 bytes */
 311 } VHDXVirtualDiskPhysicalSectorSize;
 312
 313 typedef struct QEMU_PACKED VHDXParentLocatorHeader {
 314     MSGUID      locator_type;           /* type of the parent virtual disk. */
 315     uint16_t    reserved;
 316     uint16_t    key_value_count;        /* number of key/value pairs for this
 317                                            locator */
 318 } VHDXParentLocatorHeader;
 319
 320 /* key and value strings are UNICODE strings, UTF-16 LE encoding, no NULs */
 321 typedef struct QEMU_PACKED VHDXParentLocatorEntry {
 322     uint32_t    key_offset;             /* offset in metadata for key, > 0 */
 323     uint32_t    value_offset;           /* offset in metadata for value, >0 */
 324     uint16_t    key_length;             /* length of entry key, > 0 */
 325     uint16_t    value_length;           /* length of entry value, > 0 */
 326 } VHDXParentLocatorEntry;
 327
 328
 329 /* ----- END VHDX SPECIFICATION STRUCTURES ---- */
 330
 331 typedef struct VHDXMetadataEntries {
 332     VHDXMetadataTableEntry file_parameters_entry;
 333     VHDXMetadataTableEntry virtual_disk_size_entry;
 334     VHDXMetadataTableEntry page83_data_entry;
 335     VHDXMetadataTableEntry logical_sector_size_entry;
 336     VHDXMetadataTableEntry phys_sector_size_entry;
 337     VHDXMetadataTableEntry parent_locator_entry;
 338     uint16_t present;
 339 } VHDXMetadataEntries;
 340
 341 typedef struct VHDXLogEntries {
 342     uint64_t offset;
 343     uint64_t length;
 344     uint32_t write;
 345     uint32_t read;
 346     VHDXLogEntryHeader *hdr;
 347     void *desc_buffer;
 348     uint64_t sequence;
 349     uint32_t tail;
 350 } VHDXLogEntries;
 351
 352 typedef struct VHDXRegionEntry {
 353     uint64_t start;
 354     uint64_t end;
 355     QLIST_ENTRY(VHDXRegionEntry) entries;
 356 } VHDXRegionEntry;
 357
 358 typedef struct BDRVVHDXState {
 359     CoMutex lock;
 360
 361     int curr_header;
 362     VHDXHeader *headers[2];
 363
 364     VHDXRegionTableHeader rt;
 365     VHDXRegionTableEntry bat_rt;         /* region table for the BAT */
 366     VHDXRegionTableEntry metadata_rt;    /* region table for the metadata */
 367
 368     VHDXMetadataTableHeader metadata_hdr;
 369     VHDXMetadataEntries metadata_entries;
 370
 371     VHDXFileParameters params;
 372     uint32_t block_size;
 373     uint32_t block_size_bits;
 374     uint32_t sectors_per_block;
 375     uint32_t sectors_per_block_bits;
 376
 377     uint64_t virtual_disk_size;
 378     uint32_t logical_sector_size;
 379     uint32_t physical_sector_size;
 380
 381     uint64_t chunk_ratio;
 382     uint32_t chunk_ratio_bits;
 383     uint32_t logical_sector_size_bits;
 384
 385     uint32_t bat_entries;
 386     VHDXBatEntry *bat;
 387     uint64_t bat_offset;
 388
 389     bool first_visible_write;
 390     MSGUID session_guid;
 391
 392     VHDXLogEntries log;
 393
 394     VHDXParentLocatorHeader parent_header;
 395     VHDXParentLocatorEntry *parent_entries;
 396
 397     Error *migration_blocker;
 398
 399     bool log_replayed_on_open;
 400
 401     QLIST_HEAD(, VHDXRegionEntry) regions;
 402 } BDRVVHDXState;
 403
 404 void vhdx_guid_generate(MSGUID *guid);
 405
 406 int vhdx_update_headers(BlockDriverState *bs, BDRVVHDXState *s, bool rw,
 407                         MSGUID *log_guid);
 408
 409 uint32_t vhdx_update_checksum(uint8_t *buf, size_t size, int crc_offset);
 410 uint32_t vhdx_checksum_calc(uint32_t crc, uint8_t *buf, size_t size,
 411                             int crc_offset);
 412
 413 bool vhdx_checksum_is_valid(uint8_t *buf, size_t size, int crc_offset);
 414
 415 int vhdx_parse_log(BlockDriverState *bs, BDRVVHDXState *s, bool *flushed,
 416                    Error **errp);
 417
 418 int vhdx_log_write_and_flush(BlockDriverState *bs, BDRVVHDXState *s,
 419                              void *data, uint32_t length, uint64_t offset);
 420
 421 static inline void leguid_to_cpus(MSGUID *guid)
 422 {
 423     guid->data1 = le32_to_cpu(guid->data1);
 424     guid->data2 = le16_to_cpu(guid->data2);
 425     guid->data3 = le16_to_cpu(guid->data3);
 426 }
 427
 428 static inline void cpu_to_leguids(MSGUID *guid)
 429 {
 430     guid->data1 = cpu_to_le32(guid->data1);
 431     guid->data2 = cpu_to_le16(guid->data2);
 432     guid->data3 = cpu_to_le16(guid->data3);
 433 }
 434
 435 void vhdx_header_le_import(VHDXHeader *h);
 436 void vhdx_header_le_export(VHDXHeader *orig_h, VHDXHeader *new_h);
 437 void vhdx_log_desc_le_import(VHDXLogDescriptor *d);
 438 void vhdx_log_desc_le_export(VHDXLogDescriptor *d);
 439 void vhdx_log_data_le_import(VHDXLogDataSector *d);
 440 void vhdx_log_data_le_export(VHDXLogDataSector *d);
 441 void vhdx_log_entry_hdr_le_import(VHDXLogEntryHeader *hdr);
 442 void vhdx_log_entry_hdr_le_export(VHDXLogEntryHeader *hdr);
 443 void vhdx_region_header_le_import(VHDXRegionTableHeader *hdr);
 444 void vhdx_region_header_le_export(VHDXRegionTableHeader *hdr);
 445 void vhdx_region_entry_le_import(VHDXRegionTableEntry *e);
 446 void vhdx_region_entry_le_export(VHDXRegionTableEntry *e);
 447 void vhdx_metadata_header_le_import(VHDXMetadataTableHeader *hdr);
 448 void vhdx_metadata_header_le_export(VHDXMetadataTableHeader *hdr);
 449 void vhdx_metadata_entry_le_import(VHDXMetadataTableEntry *e);
 450 void vhdx_metadata_entry_le_export(VHDXMetadataTableEntry *e);
 451 int vhdx_user_visible_write(BlockDriverState *bs, BDRVVHDXState *s);
 452
 453 #endif