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 #if !defined(KiB)
  22 #define KiB              (1 * 1024)
  23 #endif
  24 #if !defined(MiB)
  25 #define MiB            (KiB * 1024)
  26 #endif
  27 #define GiB            (MiB * 1024)
  28 #define TiB ((uint64_t) GiB * 1024)
  29
  30 /* Structures and fields present in the VHDX file */
  31
  32 /* The header section has the following blocks,
  33  * each block is 64KB:
  34  *
  35  * _____________________________________________________________________________
  36  * | File Id. |   Header 1    | Header 2   | Region Table |  Reserved (768KB)  |
  37  * |----------|---------------|------------|--------------|--------------------|
  38  * |          |               |            |              |                    |
  39  * 0.........64KB...........128KB........192KB..........256KB................1MB
  40  */
  41
  42 #define VHDX_HEADER_BLOCK_SIZE      (64 * 1024)
  43
  44 #define VHDX_FILE_ID_OFFSET         0
  45 #define VHDX_HEADER1_OFFSET         (VHDX_HEADER_BLOCK_SIZE * 1)
  46 #define VHDX_HEADER2_OFFSET         (VHDX_HEADER_BLOCK_SIZE * 2)
  47 #define VHDX_REGION_TABLE_OFFSET    (VHDX_HEADER_BLOCK_SIZE * 3)
  48 #define VHDX_REGION_TABLE2_OFFSET   (VHDX_HEADER_BLOCK_SIZE * 4)
  49
  50 #define VHDX_HEADER_SECTION_END     (1 * MiB)
  51 /*
  52  * A note on the use of MS-GUID fields.  For more details on the GUID,
  53  * please see: https://en.wikipedia.org/wiki/Globally_unique_identifier.
  54  *
  55  * The VHDX specification only states that these are MS GUIDs, and which
  56  * bytes are data1-data4. It makes no mention of what algorithm should be used
  57  * to generate the GUID, nor what standard.  However, looking at the specified
  58  * known GUID fields, it appears the GUIDs are:
  59  *  Standard/DCE GUID type  (noted by 10b in the MSB of byte 0 of .data4)
  60  *  Random algorithm        (noted by 0x4XXX for .data3)
  61  */
  62
  63 /* ---- HEADER SECTION STRUCTURES ---- */
  64
  65 /* These structures are ones that are defined in the VHDX specification
  66  * document */
  67
  68 #define VHDX_FILE_SIGNATURE 0x656C696678646876  /* "vhdxfile" in ASCII */
  69 typedef struct VHDXFileIdentifier {
  70     uint64_t    signature;              /* "vhdxfile" in ASCII */
  71     uint16_t    creator[256];           /* optional; utf-16 string to identify
  72                                            the vhdx file creator.  Diagnostic
  73                                            only */
  74 } VHDXFileIdentifier;
  75
  76
  77 /* the guid is a 16 byte unique ID - the definition for this used by
  78  * Microsoft is not just 16 bytes though - it is a structure that is defined,
  79  * so we need to follow it here so that endianness does not trip us up */
  80
  81 typedef struct QEMU_PACKED MSGUID {
  82     uint32_t  data1;
  83     uint16_t  data2;
  84     uint16_t  data3;
  85     uint8_t   data4[8];
  86 } MSGUID;
  87
  88 #define guid_eq(a, b) \
  89     (memcmp(&(a), &(b), sizeof(MSGUID)) == 0)
  90
  91 #define VHDX_HEADER_SIZE (4 * 1024)   /* although the vhdx_header struct in disk
  92                                          is only 582 bytes, for purposes of crc
  93                                          the header is the first 4KB of the 64KB
  94                                          block */
  95
  96 /* The full header is 4KB, although the actual header data is much smaller.
  97  * But for the checksum calculation, it is over the entire 4KB structure,
  98  * not just the defined portion of it */
  99 #define VHDX_HEADER_SIGNATURE 0x64616568
 100 typedef struct QEMU_PACKED VHDXHeader {
 101     uint32_t    signature;              /* "head" in ASCII */
 102     uint32_t    checksum;               /* CRC-32C hash of the whole header */
 103     uint64_t    sequence_number;        /* Seq number of this header.  Each
 104                                            VHDX file has 2 of these headers,
 105                                            and only the header with the highest
 106                                            sequence number is valid */
 107     MSGUID      file_write_guid;        /* 128 bit unique identifier. Must be
 108                                            updated to new, unique value before
 109                                            the first modification is made to
 110                                            file */
 111     MSGUID      data_write_guid;        /* 128 bit unique identifier. Must be
 112                                            updated to new, unique value before
 113                                            the first modification is made to
 114                                            visible data.   Visbile data is
 115                                            defined as:
 116                                                     - system & user metadata
 117                                                     - raw block data
 118                                                     - disk size
 119                                                     - any change that will
 120                                                       cause the virtual disk
 121                                                       sector read to differ
 122
 123                                            This does not need to change if
 124                                            blocks are re-arranged */
 125     MSGUID      log_guid;               /* 128 bit unique identifier. If zero,
 126                                            there is no valid log. If non-zero,
 127                                            log entries with this guid are
 128                                            valid. */
 129     uint16_t    log_version;            /* version of the log format. Must be
 130                                            set to zero */
 131     uint16_t    version;                /* version of the vhdx file.  Currently,
 132                                            only supported version is "1" */
 133     uint32_t    log_length;             /* length of the log.  Must be multiple
 134                                            of 1MB */
 135     uint64_t    log_offset;             /* byte offset in the file of the log.
 136                                            Must also be a multiple of 1MB */
 137 } VHDXHeader;
 138
 139 /* Header for the region table block */
 140 #define VHDX_REGION_SIGNATURE  0x69676572  /* "regi" in ASCII */
 141 typedef struct QEMU_PACKED VHDXRegionTableHeader {
 142     uint32_t    signature;              /* "regi" in ASCII */
 143     uint32_t    checksum;               /* CRC-32C hash of the 64KB table */
 144     uint32_t    entry_count;            /* number of valid entries */
 145     uint32_t    reserved;
 146 } VHDXRegionTableHeader;
 147
 148 /* Individual region table entry.  There may be a maximum of 2047 of these
 149  *
 150  *  There are two known region table properties.  Both are required.
 151  *  BAT (block allocation table):  2DC27766F62342009D64115E9BFD4A08
 152  *  Metadata:                      8B7CA20647904B9AB8FE575F050F886E
 153  */
 154 #define VHDX_REGION_ENTRY_REQUIRED  0x01    /* if set, parser must understand
 155                                                this entry in order to open
 156                                                file */
 157 typedef struct QEMU_PACKED VHDXRegionTableEntry {
 158     MSGUID      guid;                   /* 128-bit unique identifier */
 159     uint64_t    file_offset;            /* offset of the object in the file.
 160                                            Must be multiple of 1MB */
 161     uint32_t    length;                 /* length, in bytes, of the object */
 162     uint32_t    data_bits;
 163 } VHDXRegionTableEntry;
 164
 165
 166 /* ---- LOG ENTRY STRUCTURES ---- */
 167 #define VHDX_LOG_MIN_SIZE (1024 * 1024)
 168 #define VHDX_LOG_SECTOR_SIZE 4096
 169 #define VHDX_LOG_HDR_SIZE 64
 170 #define VHDX_LOG_SIGNATURE 0x65676f6c
 171 typedef struct QEMU_PACKED VHDXLogEntryHeader {
 172     uint32_t    signature;              /* "loge" in ASCII */
 173     uint32_t    checksum;               /* CRC-32C hash of the 64KB table */
 174     uint32_t    entry_length;           /* length in bytes, multiple of 1MB */
 175     uint32_t    tail;                   /* byte offset of first log entry of a
 176                                            seq, where this entry is the last
 177                                            entry */
 178     uint64_t    sequence_number;        /* incremented with each log entry.
 179                                            May not be zero. */
 180     uint32_t    descriptor_count;       /* number of descriptors in this log
 181                                            entry, must be >= 0 */
 182     uint32_t    reserved;
 183     MSGUID      log_guid;               /* value of the log_guid from
 184                                            vhdx_header.  If not found in
 185                                            vhdx_header, it is invalid */
 186     uint64_t    flushed_file_offset;    /* see spec for full details - this
 187                                            should be vhdx file size in bytes */
 188     uint64_t    last_file_offset;       /* size in bytes that all allocated
 189                                            file structures fit into */
 190 } VHDXLogEntryHeader;
 191
 192 #define VHDX_LOG_DESC_SIZE 32
 193 #define VHDX_LOG_DESC_SIGNATURE 0x63736564
 194 #define VHDX_LOG_ZERO_SIGNATURE 0x6f72657a
 195 typedef struct QEMU_PACKED VHDXLogDescriptor {
 196     uint32_t    signature;              /* "zero" or "desc" in ASCII */
 197     union  {
 198         uint32_t    reserved;           /* zero desc */
 199         uint32_t    trailing_bytes;     /* data desc: bytes 4092-4096 of the
 200                                            data sector */
 201     };
 202     union {
 203         uint64_t    zero_length;        /* zero desc: length of the section to
 204                                            zero */
 205         uint64_t    leading_bytes;      /* data desc: bytes 0-7 of the data
 206                                            sector */
 207     };
 208     uint64_t    file_offset;            /* file offset to write zeros - multiple
 209                                            of 4kB */
 210     uint64_t    sequence_number;        /* must match same field in
 211                                            vhdx_log_entry_header */
 212 } VHDXLogDescriptor;
 213
 214 #define VHDX_LOG_DATA_SIGNATURE 0x61746164
 215 typedef struct QEMU_PACKED VHDXLogDataSector {
 216     uint32_t    data_signature;         /* "data" in ASCII */
 217     uint32_t    sequence_high;          /* 4 MSB of 8 byte sequence_number */
 218     uint8_t     data[4084];             /* raw data, bytes 8-4091 (inclusive).
 219                                            see the data descriptor field for the
 220                                            other mising bytes */
 221     uint32_t    sequence_low;           /* 4 LSB of 8 byte sequence_number */
 222 } VHDXLogDataSector;
 223
 224
 225
 226 /* block states - different state values depending on whether it is a
 227  * payload block, or a sector block. */
 228
 229 #define PAYLOAD_BLOCK_NOT_PRESENT       0
 230 #define PAYLOAD_BLOCK_UNDEFINED         1
 231 #define PAYLOAD_BLOCK_ZERO              2
 232 #define PAYLOAD_BLOCK_UNMAPPED          5
 233 #define PAYLOAD_BLOCK_FULLY_PRESENT     6
 234 #define PAYLOAD_BLOCK_PARTIALLY_PRESENT 7
 235
 236 #define SB_BLOCK_NOT_PRESENT    0
 237 #define SB_BLOCK_PRESENT        6
 238
 239 /* per the spec */
 240 #define VHDX_MAX_SECTORS_PER_BLOCK  (1 << 23)
 241
 242 /* upper 44 bits are the file offset in 1MB units lower 3 bits are the state
 243    other bits are reserved */
 244 #define VHDX_BAT_STATE_BIT_MASK 0x07
 245 #define VHDX_BAT_FILE_OFF_MASK  0xFFFFFFFFFFF00000 /* upper 44 bits */
 246 typedef uint64_t VHDXBatEntry;
 247
 248 /* ---- METADATA REGION STRUCTURES ---- */
 249
 250 #define VHDX_METADATA_ENTRY_SIZE 32
 251 #define VHDX_METADATA_MAX_ENTRIES 2047  /* not including the header */
 252 #define VHDX_METADATA_TABLE_MAX_SIZE \
 253     (VHDX_METADATA_ENTRY_SIZE * (VHDX_METADATA_MAX_ENTRIES+1))
 254 #define VHDX_METADATA_SIGNATURE 0x617461646174656D  /* "metadata" in ASCII */
 255 typedef struct QEMU_PACKED VHDXMetadataTableHeader {
 256     uint64_t    signature;              /* "metadata" in ASCII */
 257     uint16_t    reserved;
 258     uint16_t    entry_count;            /* number table entries. <= 2047 */
 259     uint32_t    reserved2[5];
 260 } VHDXMetadataTableHeader;
 261
 262 #define VHDX_META_FLAGS_IS_USER         0x01    /* max 1024 entries */
 263 #define VHDX_META_FLAGS_IS_VIRTUAL_DISK 0x02    /* virtual disk metadata if set,
 264                                                    otherwise file metdata */
 265 #define VHDX_META_FLAGS_IS_REQUIRED     0x04    /* parse must understand this
 266                                                    entry to open the file */
 267 typedef struct QEMU_PACKED VHDXMetadataTableEntry {
 268     MSGUID      item_id;                /* 128-bit identifier for metadata */
 269     uint32_t    offset;                 /* byte offset of the metadata.  At
 270                                            least 64kB.  Relative to start of
 271                                            metadata region */
 272                                         /* note: if length = 0, so is offset */
 273     uint32_t    length;                 /* length of metadata. <= 1MB. */
 274     uint32_t    data_bits;              /* least-significant 3 bits are flags,
 275                                            the rest are reserved (see above) */
 276     uint32_t    reserved2;
 277 } VHDXMetadataTableEntry;
 278
 279 #define VHDX_PARAMS_LEAVE_BLOCKS_ALLOCED 0x01   /* Do not change any blocks to
 280                                                    be BLOCK_NOT_PRESENT.
 281                                                    If set indicates a fixed
 282                                                    size VHDX file */
 283 #define VHDX_PARAMS_HAS_PARENT           0x02    /* has parent / backing file */
 284 #define VHDX_BLOCK_SIZE_MIN             (1   * MiB)
 285 #define VHDX_BLOCK_SIZE_MAX             (256 * MiB)
 286 typedef struct QEMU_PACKED VHDXFileParameters {
 287     uint32_t    block_size;             /* size of each payload block, always
 288                                            power of 2, <= 256MB and >= 1MB. */
 289     uint32_t data_bits;                 /* least-significant 2 bits are flags,
 290                                            the rest are reserved (see above) */
 291 } VHDXFileParameters;
 292
 293 #define VHDX_MAX_IMAGE_SIZE  ((uint64_t) 64 * TiB)
 294 typedef struct QEMU_PACKED VHDXVirtualDiskSize {
 295     uint64_t    virtual_disk_size;      /* Size of the virtual disk, in bytes.
 296                                            Must be multiple of the sector size,
 297                                            max of 64TB */
 298 } VHDXVirtualDiskSize;
 299
 300 typedef struct QEMU_PACKED VHDXPage83Data {
 301     MSGUID      page_83_data;           /* unique id for scsi devices that
 302                                            support page 0x83 */
 303 } VHDXPage83Data;
 304
 305 typedef struct QEMU_PACKED VHDXVirtualDiskLogicalSectorSize {
 306     uint32_t    logical_sector_size;    /* virtual disk sector size (in bytes).
 307                                            Can only be 512 or 4096 bytes */
 308 } VHDXVirtualDiskLogicalSectorSize;
 309
 310 typedef struct QEMU_PACKED VHDXVirtualDiskPhysicalSectorSize {
 311     uint32_t    physical_sector_size;   /* physical sector size (in bytes).
 312                                            Can only be 512 or 4096 bytes */
 313 } VHDXVirtualDiskPhysicalSectorSize;
 314
 315 typedef struct QEMU_PACKED VHDXParentLocatorHeader {
 316     MSGUID      locator_type;           /* type of the parent virtual disk. */
 317     uint16_t    reserved;
 318     uint16_t    key_value_count;        /* number of key/value pairs for this
 319                                            locator */
 320 } VHDXParentLocatorHeader;
 321
 322 /* key and value strings are UNICODE strings, UTF-16 LE encoding, no NULs */
 323 typedef struct QEMU_PACKED VHDXParentLocatorEntry {
 324     uint32_t    key_offset;             /* offset in metadata for key, > 0 */
 325     uint32_t    value_offset;           /* offset in metadata for value, >0 */
 326     uint16_t    key_length;             /* length of entry key, > 0 */
 327     uint16_t    value_length;           /* length of entry value, > 0 */
 328 } VHDXParentLocatorEntry;
 329
 330
 331 /* ----- END VHDX SPECIFICATION STRUCTURES ---- */
 332
 333 typedef struct VHDXMetadataEntries {
 334     VHDXMetadataTableEntry file_parameters_entry;
 335     VHDXMetadataTableEntry virtual_disk_size_entry;
 336     VHDXMetadataTableEntry page83_data_entry;
 337     VHDXMetadataTableEntry logical_sector_size_entry;
 338     VHDXMetadataTableEntry phys_sector_size_entry;
 339     VHDXMetadataTableEntry parent_locator_entry;
 340     uint16_t present;
 341 } VHDXMetadataEntries;
 342
 343 typedef struct VHDXLogEntries {
 344     uint64_t offset;
 345     uint64_t length;
 346     uint32_t write;
 347     uint32_t read;
 348     VHDXLogEntryHeader *hdr;
 349     void *desc_buffer;
 350     uint64_t sequence;
 351     uint32_t tail;
 352 } VHDXLogEntries;
 353
 354 typedef struct VHDXRegionEntry {
 355     uint64_t start;
 356     uint64_t end;
 357     QLIST_ENTRY(VHDXRegionEntry) entries;
 358 } VHDXRegionEntry;
 359
 360 typedef struct BDRVVHDXState {
 361     CoMutex lock;
 362
 363     int curr_header;
 364     VHDXHeader *headers[2];
 365
 366     VHDXRegionTableHeader rt;
 367     VHDXRegionTableEntry bat_rt;         /* region table for the BAT */
 368     VHDXRegionTableEntry metadata_rt;    /* region table for the metadata */
 369
 370     VHDXMetadataTableHeader metadata_hdr;
 371     VHDXMetadataEntries metadata_entries;
 372
 373     VHDXFileParameters params;
 374     uint32_t block_size;
 375     uint32_t block_size_bits;
 376     uint32_t sectors_per_block;
 377     uint32_t sectors_per_block_bits;
 378
 379     uint64_t virtual_disk_size;
 380     uint32_t logical_sector_size;
 381     uint32_t physical_sector_size;
 382
 383     uint64_t chunk_ratio;
 384     uint32_t chunk_ratio_bits;
 385     uint32_t logical_sector_size_bits;
 386
 387     uint32_t bat_entries;
 388     VHDXBatEntry *bat;
 389     uint64_t bat_offset;
 390
 391     bool first_visible_write;
 392     MSGUID session_guid;
 393
 394     VHDXLogEntries log;
 395
 396     VHDXParentLocatorHeader parent_header;
 397     VHDXParentLocatorEntry *parent_entries;
 398
 399     Error *migration_blocker;
 400
 401     QLIST_HEAD(VHDXRegionHead, 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
 417 int vhdx_log_write_and_flush(BlockDriverState *bs, BDRVVHDXState *s,
 418                              void *data, uint32_t length, uint64_t offset);
 419
 420 static inline void leguid_to_cpus(MSGUID *guid)
 421 {
 422     le32_to_cpus(&guid->data1);
 423     le16_to_cpus(&guid->data2);
 424     le16_to_cpus(&guid->data3);
 425 }
 426
 427 static inline void cpu_to_leguids(MSGUID *guid)
 428 {
 429     cpu_to_le32s(&guid->data1);
 430     cpu_to_le16s(&guid->data2);
 431     cpu_to_le16s(&guid->data3);
 432 }
 433
 434 void vhdx_header_le_import(VHDXHeader *h);
 435 void vhdx_header_le_export(VHDXHeader *orig_h, VHDXHeader *new_h);
 436 void vhdx_log_desc_le_import(VHDXLogDescriptor *d);
 437 void vhdx_log_desc_le_export(VHDXLogDescriptor *d);
 438 void vhdx_log_data_le_export(VHDXLogDataSector *d);
 439 void vhdx_log_entry_hdr_le_import(VHDXLogEntryHeader *hdr);
 440 void vhdx_log_entry_hdr_le_export(VHDXLogEntryHeader *hdr);
 441 void vhdx_region_header_le_import(VHDXRegionTableHeader *hdr);
 442 void vhdx_region_header_le_export(VHDXRegionTableHeader *hdr);
 443 void vhdx_region_entry_le_import(VHDXRegionTableEntry *e);
 444 void vhdx_region_entry_le_export(VHDXRegionTableEntry *e);
 445 void vhdx_metadata_header_le_import(VHDXMetadataTableHeader *hdr);
 446 void vhdx_metadata_header_le_export(VHDXMetadataTableHeader *hdr);
 447 void vhdx_metadata_entry_le_import(VHDXMetadataTableEntry *e);
 448 void vhdx_metadata_entry_le_export(VHDXMetadataTableEntry *e);
 449 int vhdx_user_visible_write(BlockDriverState *bs, BDRVVHDXState *s);
 450
 451 #endif