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