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 /* Structures and fields present in the VHDX file */
  27
  28 /* The header section has the following blocks,
  29  * each block is 64KB:
  30  *
  31  * _____________________________________________________________________________
  32  * | File Id. |   Header 1    | Header 2   | Region Table |  Reserved (768KB)  |
  33  * |----------|---------------|------------|--------------|--------------------|
  34  * |          |               |            |              |                    |
  35  * 0.........64KB...........128KB........192KB..........256KB................1MB
  36  */
  37
  38 #define VHDX_HEADER_BLOCK_SIZE      (64 * 1024)
  39
  40 #define VHDX_FILE_ID_OFFSET         0
  41 #define VHDX_HEADER1_OFFSET         (VHDX_HEADER_BLOCK_SIZE * 1)
  42 #define VHDX_HEADER2_OFFSET         (VHDX_HEADER_BLOCK_SIZE * 2)
  43 #define VHDX_REGION_TABLE_OFFSET    (VHDX_HEADER_BLOCK_SIZE * 3)
  44 #define VHDX_REGION_TABLE2_OFFSET   (VHDX_HEADER_BLOCK_SIZE * 4)
  45
  46 #define VHDX_HEADER_SECTION_END     (1 * MiB)
  47 /*
  48  * A note on the use of MS-GUID fields.  For more details on the GUID,
  49  * please see: https://en.wikipedia.org/wiki/Globally_unique_identifier.
  50  *
  51  * The VHDX specification only states that these are MS GUIDs, and which
  52  * bytes are data1-data4. It makes no mention of what algorithm should be used
  53  * to generate the GUID, nor what standard.  However, looking at the specified
  54  * known GUID fields, it appears the GUIDs are:
  55  *  Standard/DCE GUID type  (noted by 10b in the MSB of byte 0 of .data4)
  56  *  Random algorithm        (noted by 0x4XXX for .data3)
  57  */
  58
  59 /* ---- HEADER SECTION STRUCTURES ---- */
  60
  61 /* These structures are ones that are defined in the VHDX specification
  62  * document */
  63
  64 #define VHDX_FILE_SIGNATURE 0x656C696678646876  /* "vhdxfile" in ASCII */
  65 typedef struct VHDXFileIdentifier {
  66     uint64_t    signature;              /* "vhdxfile" in ASCII */
  67     uint16_t    creator[256];           /* optional; utf-16 string to identify
  68                                            the vhdx file creator.  Diagnostic
  69                                            only */
  70 } VHDXFileIdentifier;
  71
  72
  73 /* the guid is a 16 byte unique ID - the definition for this used by
  74  * Microsoft is not just 16 bytes though - it is a structure that is defined,
  75  * so we need to follow it here so that endianness does not trip us up */
  76
  77 typedef struct QEMU_PACKED MSGUID {
  78     uint32_t  data1;
  79     uint16_t  data2;
  80     uint16_t  data3;
  81     uint8_t   data4[8];
  82 } MSGUID;
  83
  84 #define guid_eq(a, b) \
  85     (memcmp(&(a), &(b), sizeof(MSGUID)) == 0)
  86
  87 #define VHDX_HEADER_SIZE (4 * 1024)   /* although the vhdx_header struct in disk
  88                                          is only 582 bytes, for purposes of crc
  89                                          the header is the first 4KB of the 64KB
  90                                          block */
  91
  92 /* The full header is 4KB, although the actual header data is much smaller.
  93  * But for the checksum calculation, it is over the entire 4KB structure,
  94  * not just the defined portion of it */
  95 #define VHDX_HEADER_SIGNATURE 0x64616568
  96 typedef struct QEMU_PACKED VHDXHeader {
  97     uint32_t    signature;              /* "head" in ASCII */
  98     uint32_t    checksum;               /* CRC-32C hash of the whole header */
  99     uint64_t    sequence_number;        /* Seq number of this header.  Each
 100                                            VHDX file has 2 of these headers,
 101                                            and only the header with the highest
 102                                            sequence number is valid */
 103     MSGUID      file_write_guid;        /* 128 bit unique identifier. Must be
 104                                            updated to new, unique value before
 105                                            the first modification is made to
 106                                            file */
 107     MSGUID      data_write_guid;        /* 128 bit unique identifier. Must be
 108                                            updated to new, unique value before
 109                                            the first modification is made to
 110                                            visible data.   Visbile data is
 111                                            defined as:
 112                                                     - system & user metadata
 113                                                     - raw block data
 114                                                     - disk size
 115                                                     - any change that will
 116                                                       cause the virtual disk
 117                                                       sector read to differ
 118
 119                                            This does not need to change if
 120                                            blocks are re-arranged */
 121     MSGUID      log_guid;               /* 128 bit unique identifier. If zero,
 122                                            there is no valid log. If non-zero,
 123                                            log entries with this guid are
 124                                            valid. */
 125     uint16_t    log_version;            /* version of the log format. Must be
 126                                            set to zero */
 127     uint16_t    version;                /* version of the vhdx file.  Currently,
 128                                            only supported version is "1" */
 129     uint32_t    log_length;             /* length of the log.  Must be multiple
 130                                            of 1MB */
 131     uint64_t    log_offset;             /* byte offset in the file of the log.
 132                                            Must also be a multiple of 1MB */
 133 } VHDXHeader;
 134
 135 /* Header for the region table block */
 136 #define VHDX_REGION_SIGNATURE  0x69676572  /* "regi" in ASCII */
 137 typedef struct QEMU_PACKED VHDXRegionTableHeader {
 138     uint32_t    signature;              /* "regi" in ASCII */
 139     uint32_t    checksum;               /* CRC-32C hash of the 64KB table */
 140     uint32_t    entry_count;            /* number of valid entries */
 141     uint32_t    reserved;
 142 } VHDXRegionTableHeader;
 143
 144 /* Individual region table entry.  There may be a maximum of 2047 of these
 145  *
 146  *  There are two known region table properties.  Both are required.
 147  *  BAT (block allocation table):  2DC27766F62342009D64115E9BFD4A08
 148  *  Metadata:                      8B7CA20647904B9AB8FE575F050F886E
 149  */
 150 #define VHDX_REGION_ENTRY_REQUIRED  0x01    /* if set, parser must understand
 151                                                this entry in order to open
 152                                                file */
 153 typedef struct QEMU_PACKED VHDXRegionTableEntry {
 154     MSGUID      guid;                   /* 128-bit unique identifier */
 155     uint64_t    file_offset;            /* offset of the object in the file.
 156                                            Must be multiple of 1MB */
 157     uint32_t    length;                 /* length, in bytes, of the object */
 158     uint32_t    data_bits;
 159 } VHDXRegionTableEntry;
 160
 161
 162 /* ---- LOG ENTRY STRUCTURES ---- */
 163 #define VHDX_LOG_MIN_SIZE (1024 * 1024)
 164 #define VHDX_LOG_SECTOR_SIZE 4096
 165 #define VHDX_LOG_HDR_SIZE 64
 166 #define VHDX_LOG_SIGNATURE 0x65676f6c
 167 typedef struct QEMU_PACKED VHDXLogEntryHeader {
 168     uint32_t    signature;              /* "loge" in ASCII */
 169     uint32_t    checksum;               /* CRC-32C hash of the 64KB table */
 170     uint32_t    entry_length;           /* length in bytes, multiple of 1MB */
 171     uint32_t    tail;                   /* byte offset of first log entry of a
 172                                            seq, where this entry is the last
 173                                            entry */
 174     uint64_t    sequence_number;        /* incremented with each log entry.
 175                                            May not be zero. */
 176     uint32_t    descriptor_count;       /* number of descriptors in this log
 177                                            entry, must be >= 0 */
 178     uint32_t    reserved;
 179     MSGUID      log_guid;               /* value of the log_guid from
 180                                            vhdx_header.  If not found in
 181                                            vhdx_header, it is invalid */
 182     uint64_t    flushed_file_offset;    /* see spec for full details - this
 183                                            should be vhdx file size in bytes */
 184     uint64_t    last_file_offset;       /* size in bytes that all allocated
 185                                            file structures fit into */
 186 } VHDXLogEntryHeader;
 187
 188 #define VHDX_LOG_DESC_SIZE 32
 189 #define VHDX_LOG_DESC_SIGNATURE 0x63736564
 190 #define VHDX_LOG_ZERO_SIGNATURE 0x6f72657a
 191 typedef struct QEMU_PACKED VHDXLogDescriptor {
 192     uint32_t    signature;              /* "zero" or "desc" in ASCII */
 193     union  {
 194         uint32_t    reserved;           /* zero desc */
 195         uint32_t    trailing_bytes;     /* data desc: bytes 4092-4096 of the
 196                                            data sector */
 197     };
 198     union {
 199         uint64_t    zero_length;        /* zero desc: length of the section to
 200                                            zero */
 201         uint64_t    leading_bytes;      /* data desc: bytes 0-7 of the data
 202                                            sector */
 203     };
 204     uint64_t    file_offset;            /* file offset to write zeros - multiple
 205                                            of 4kB */
 206     uint64_t    sequence_number;        /* must match same field in
 207                                            vhdx_log_entry_header */
 208 } VHDXLogDescriptor;
 209
 210 #define VHDX_LOG_DATA_SIGNATURE 0x61746164
 211 typedef struct QEMU_PACKED VHDXLogDataSector {
 212     uint32_t    data_signature;         /* "data" in ASCII */
 213     uint32_t    sequence_high;          /* 4 MSB of 8 byte sequence_number */
 214     uint8_t     data[4084];             /* raw data, bytes 8-4091 (inclusive).
 215                                            see the data descriptor field for the
 216                                            other mising bytes */
 217     uint32_t    sequence_low;           /* 4 LSB of 8 byte sequence_number */
 218 } VHDXLogDataSector;
 219
 220
 221
 222 /* block states - different state values depending on whether it is a
 223  * payload block, or a sector block. */
 224
 225 #define PAYLOAD_BLOCK_NOT_PRESENT       0
 226 #define PAYLOAD_BLOCK_UNDEFINED         1
 227 #define PAYLOAD_BLOCK_ZERO              2
 228 #define PAYLOAD_BLOCK_UNMAPPED          5
 229 #define PAYLOAD_BLOCK_FULLY_PRESENT     6
 230 #define PAYLOAD_BLOCK_PARTIALLY_PRESENT 7
 231
 232 #define SB_BLOCK_NOT_PRESENT    0
 233 #define SB_BLOCK_PRESENT        6
 234
 235 /* per the spec */
 236 #define VHDX_MAX_SECTORS_PER_BLOCK  (1 << 23)
 237
 238 /* upper 44 bits are the file offset in 1MB units lower 3 bits are the state
 239    other bits are reserved */
 240 #define VHDX_BAT_STATE_BIT_MASK 0x07
 241 #define VHDX_BAT_FILE_OFF_MASK  0xFFFFFFFFFFF00000 /* upper 44 bits */
 242 typedef uint64_t VHDXBatEntry;
 243
 244 /* ---- METADATA REGION STRUCTURES ---- */
 245
 246 #define VHDX_METADATA_ENTRY_SIZE 32
 247 #define VHDX_METADATA_MAX_ENTRIES 2047  /* not including the header */
 248 #define VHDX_METADATA_TABLE_MAX_SIZE \
 249     (VHDX_METADATA_ENTRY_SIZE * (VHDX_METADATA_MAX_ENTRIES+1))
 250 #define VHDX_METADATA_SIGNATURE 0x617461646174656D  /* "metadata" in ASCII */
 251 typedef struct QEMU_PACKED VHDXMetadataTableHeader {
 252     uint64_t    signature;              /* "metadata" in ASCII */
 253     uint16_t    reserved;
 254     uint16_t    entry_count;            /* number table entries. <= 2047 */
 255     uint32_t    reserved2[5];
 256 } VHDXMetadataTableHeader;
 257
 258 #define VHDX_META_FLAGS_IS_USER         0x01    /* max 1024 entries */
 259 #define VHDX_META_FLAGS_IS_VIRTUAL_DISK 0x02    /* virtual disk metadata if set,
 260                                                    otherwise file metdata */
 261 #define VHDX_META_FLAGS_IS_REQUIRED     0x04    /* parse must understand this
 262                                                    entry to open the file */
 263 typedef struct QEMU_PACKED VHDXMetadataTableEntry {
 264     MSGUID      item_id;                /* 128-bit identifier for metadata */
 265     uint32_t    offset;                 /* byte offset of the metadata.  At
 266                                            least 64kB.  Relative to start of
 267                                            metadata region */
 268                                         /* note: if length = 0, so is offset */
 269     uint32_t    length;                 /* length of metadata. <= 1MB. */
 270     uint32_t    data_bits;              /* least-significant 3 bits are flags,
 271                                            the rest are reserved (see above) */
 272     uint32_t    reserved2;
 273 } VHDXMetadataTableEntry;
 274
 275 #define VHDX_PARAMS_LEAVE_BLOCKS_ALLOCED 0x01   /* Do not change any blocks to
 276                                                    be BLOCK_NOT_PRESENT.
 277                                                    If set indicates a fixed
 278                                                    size VHDX file */
 279 #define VHDX_PARAMS_HAS_PARENT           0x02    /* has parent / backing file */
 280 #define VHDX_BLOCK_SIZE_MIN             (1   * MiB)
 281 #define VHDX_BLOCK_SIZE_MAX             (256 * MiB)
 282 typedef struct QEMU_PACKED VHDXFileParameters {
 283     uint32_t    block_size;             /* size of each payload block, always
 284                                            power of 2, <= 256MB and >= 1MB. */
 285     uint32_t data_bits;                 /* least-significant 2 bits are flags,
 286                                            the rest are reserved (see above) */
 287 } VHDXFileParameters;
 288
 289 #define VHDX_MAX_IMAGE_SIZE  ((uint64_t) 64 * TiB)
 290 typedef struct QEMU_PACKED VHDXVirtualDiskSize {
 291     uint64_t    virtual_disk_size;      /* Size of the virtual disk, in bytes.
 292                                            Must be multiple of the sector size,
 293                                            max of 64TB */
 294 } VHDXVirtualDiskSize;
 295
 296 typedef struct QEMU_PACKED VHDXPage83Data {
 297     MSGUID      page_83_data;           /* unique id for scsi devices that
 298                                            support page 0x83 */
 299 } VHDXPage83Data;
 300
 301 typedef struct QEMU_PACKED VHDXVirtualDiskLogicalSectorSize {
 302     uint32_t    logical_sector_size;    /* virtual disk sector size (in bytes).
 303                                            Can only be 512 or 4096 bytes */
 304 } VHDXVirtualDiskLogicalSectorSize;
 305
 306 typedef struct QEMU_PACKED VHDXVirtualDiskPhysicalSectorSize {
 307     uint32_t    physical_sector_size;   /* physical sector size (in bytes).
 308                                            Can only be 512 or 4096 bytes */
 309 } VHDXVirtualDiskPhysicalSectorSize;
 310
 311 typedef struct QEMU_PACKED VHDXParentLocatorHeader {
 312     MSGUID      locator_type;           /* type of the parent virtual disk. */
 313     uint16_t    reserved;
 314     uint16_t    key_value_count;        /* number of key/value pairs for this
 315                                            locator */
 316 } VHDXParentLocatorHeader;
 317
 318 /* key and value strings are UNICODE strings, UTF-16 LE encoding, no NULs */
 319 typedef struct QEMU_PACKED VHDXParentLocatorEntry {
 320     uint32_t    key_offset;             /* offset in metadata for key, > 0 */
 321     uint32_t    value_offset;           /* offset in metadata for value, >0 */
 322     uint16_t    key_length;             /* length of entry key, > 0 */
 323     uint16_t    value_length;           /* length of entry value, > 0 */
 324 } VHDXParentLocatorEntry;
 325
 326
 327 /* ----- END VHDX SPECIFICATION STRUCTURES ---- */
 328
 329 typedef struct VHDXMetadataEntries {
 330     VHDXMetadataTableEntry file_parameters_entry;
 331     VHDXMetadataTableEntry virtual_disk_size_entry;
 332     VHDXMetadataTableEntry page83_data_entry;
 333     VHDXMetadataTableEntry logical_sector_size_entry;
 334     VHDXMetadataTableEntry phys_sector_size_entry;
 335     VHDXMetadataTableEntry parent_locator_entry;
 336     uint16_t present;
 337 } VHDXMetadataEntries;
 338
 339 typedef struct VHDXLogEntries {
 340     uint64_t offset;
 341     uint64_t length;
 342     uint32_t write;
 343     uint32_t read;
 344     VHDXLogEntryHeader *hdr;
 345     void *desc_buffer;
 346     uint64_t sequence;
 347     uint32_t tail;
 348 } VHDXLogEntries;
 349
 350 typedef struct VHDXRegionEntry {
 351     uint64_t start;
 352     uint64_t end;
 353     QLIST_ENTRY(VHDXRegionEntry) entries;
 354 } VHDXRegionEntry;
 355
 356 typedef struct BDRVVHDXState {
 357     CoMutex lock;
 358
 359     int curr_header;
 360     VHDXHeader *headers[2];
 361
 362     VHDXRegionTableHeader rt;
 363     VHDXRegionTableEntry bat_rt;         /* region table for the BAT */
 364     VHDXRegionTableEntry metadata_rt;    /* region table for the metadata */
 365
 366     VHDXMetadataTableHeader metadata_hdr;
 367     VHDXMetadataEntries metadata_entries;
 368
 369     VHDXFileParameters params;
 370     uint32_t block_size;
 371     uint32_t block_size_bits;
 372     uint32_t sectors_per_block;
 373     uint32_t sectors_per_block_bits;
 374
 375     uint64_t virtual_disk_size;
 376     uint32_t logical_sector_size;
 377     uint32_t physical_sector_size;
 378
 379     uint64_t chunk_ratio;
 380     uint32_t chunk_ratio_bits;
 381     uint32_t logical_sector_size_bits;
 382
 383     uint32_t bat_entries;
 384     VHDXBatEntry *bat;
 385     uint64_t bat_offset;
 386
 387     bool first_visible_write;
 388     MSGUID session_guid;
 389
 390     VHDXLogEntries log;
 391
 392     VHDXParentLocatorHeader parent_header;
 393     VHDXParentLocatorEntry *parent_entries;
 394
 395     Error *migration_blocker;
 396
 397     QLIST_HEAD(VHDXRegionHead, VHDXRegionEntry) regions;
 398 } BDRVVHDXState;
 399
 400 void vhdx_guid_generate(MSGUID *guid);
 401
 402 int vhdx_update_headers(BlockDriverState *bs, BDRVVHDXState *s, bool rw,
 403                         MSGUID *log_guid);
 404
 405 uint32_t vhdx_update_checksum(uint8_t *buf, size_t size, int crc_offset);
 406 uint32_t vhdx_checksum_calc(uint32_t crc, uint8_t *buf, size_t size,
 407                             int crc_offset);
 408
 409 bool vhdx_checksum_is_valid(uint8_t *buf, size_t size, int crc_offset);
 410
 411 int vhdx_parse_log(BlockDriverState *bs, BDRVVHDXState *s, bool *flushed);
 412
 413 int vhdx_log_write_and_flush(BlockDriverState *bs, BDRVVHDXState *s,
 414                              void *data, uint32_t length, uint64_t offset);
 415
 416 static inline void leguid_to_cpus(MSGUID *guid)
 417 {
 418     le32_to_cpus(&guid->data1);
 419     le16_to_cpus(&guid->data2);
 420     le16_to_cpus(&guid->data3);
 421 }
 422
 423 static inline void cpu_to_leguids(MSGUID *guid)
 424 {
 425     cpu_to_le32s(&guid->data1);
 426     cpu_to_le16s(&guid->data2);
 427     cpu_to_le16s(&guid->data3);
 428 }
 429
 430 void vhdx_header_le_import(VHDXHeader *h);
 431 void vhdx_header_le_export(VHDXHeader *orig_h, VHDXHeader *new_h);
 432 void vhdx_log_desc_le_import(VHDXLogDescriptor *d);
 433 void vhdx_log_desc_le_export(VHDXLogDescriptor *d);
 434 void vhdx_log_data_le_export(VHDXLogDataSector *d);
 435 void vhdx_log_entry_hdr_le_import(VHDXLogEntryHeader *hdr);
 436 void vhdx_log_entry_hdr_le_export(VHDXLogEntryHeader *hdr);
 437 void vhdx_region_header_le_import(VHDXRegionTableHeader *hdr);
 438 void vhdx_region_header_le_export(VHDXRegionTableHeader *hdr);
 439 void vhdx_region_entry_le_import(VHDXRegionTableEntry *e);
 440 void vhdx_region_entry_le_export(VHDXRegionTableEntry *e);
 441 void vhdx_metadata_header_le_import(VHDXMetadataTableHeader *hdr);
 442 void vhdx_metadata_header_le_export(VHDXMetadataTableHeader *hdr);
 443 void vhdx_metadata_entry_le_import(VHDXMetadataTableEntry *e);
 444 void vhdx_metadata_entry_le_export(VHDXMetadataTableEntry *e);
 445 int vhdx_user_visible_write(BlockDriverState *bs, BDRVVHDXState *s);
 446
 447 #endif