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 #define DEFAULT_LOG_SIZE 1048576 /* 1MiB */
  31 /* Structures and fields present in the VHDX file */
  32
  33 /* The header section has the following blocks,
  34  * each block is 64KB:
  35  *
  36  * _____________________________________________________________________________
  37  * | File Id. |   Header 1    | Header 2   | Region Table |  Reserved (768KB)  |
  38  * |----------|---------------|------------|--------------|--------------------|
  39  * |          |               |            |              |                    |
  40  * 0.........64KB...........128KB........192KB..........256KB................1MB
  41  */
  42
  43 #define VHDX_HEADER_BLOCK_SIZE      (64 * 1024)
  44
  45 #define VHDX_FILE_ID_OFFSET         0
  46 #define VHDX_HEADER1_OFFSET         (VHDX_HEADER_BLOCK_SIZE * 1)
  47 #define VHDX_HEADER2_OFFSET         (VHDX_HEADER_BLOCK_SIZE * 2)
  48 #define VHDX_REGION_TABLE_OFFSET    (VHDX_HEADER_BLOCK_SIZE * 3)
  49 #define VHDX_REGION_TABLE2_OFFSET   (VHDX_HEADER_BLOCK_SIZE * 4)
  50
  51 #define VHDX_HEADER_SECTION_END     (1 * MiB)
  52 /*
  53  * A note on the use of MS-GUID fields.  For more details on the GUID,
  54  * please see: https://en.wikipedia.org/wiki/Globally_unique_identifier.
  55  *
  56  * The VHDX specification only states that these are MS GUIDs, and which
  57  * bytes are data1-data4. It makes no mention of what algorithm should be used
  58  * to generate the GUID, nor what standard.  However, looking at the specified
  59  * known GUID fields, it appears the GUIDs are:
  60  *  Standard/DCE GUID type  (noted by 10b in the MSB of byte 0 of .data4)
  61  *  Random algorithm        (noted by 0x4XXX for .data3)
  62  */
  63
  64 /* ---- HEADER SECTION STRUCTURES ---- */
  65
  66 /* These structures are ones that are defined in the VHDX specification
  67  * document */
  68
  69 #define VHDX_FILE_SIGNATURE 0x656C696678646876ULL  /* "vhdxfile" in ASCII */
  70 typedef struct VHDXFileIdentifier {
  71     uint64_t    signature;              /* "vhdxfile" in ASCII */
  72     uint16_t    creator[256];           /* optional; utf-16 string to identify
  73                                            the vhdx file creator.  Diagnostic
  74                                            only */
  75 } VHDXFileIdentifier;
  76
  77
  78 /* the guid is a 16 byte unique ID - the definition for this used by
  79  * Microsoft is not just 16 bytes though - it is a structure that is defined,
  80  * so we need to follow it here so that endianness does not trip us up */
  81
  82 typedef struct QEMU_PACKED MSGUID {
  83     uint32_t  data1;
  84     uint16_t  data2;
  85     uint16_t  data3;
  86     uint8_t   data4[8];
  87 } MSGUID;
  88
  89 #define guid_eq(a, b) \
  90     (memcmp(&(a), &(b), sizeof(MSGUID)) == 0)
  91
  92 #define VHDX_HEADER_SIZE (4 * 1024)   /* although the vhdx_header struct in disk
  93                                          is only 582 bytes, for purposes of crc
  94                                          the header is the first 4KB of the 64KB
  95                                          block */
  96
  97 /* The full header is 4KB, although the actual header data is much smaller.
  98  * But for the checksum calculation, it is over the entire 4KB structure,
  99  * not just the defined portion of it */
 100 #define VHDX_HEADER_SIGNATURE 0x64616568
 101 typedef struct QEMU_PACKED VHDXHeader {
 102     uint32_t    signature;              /* "head" in ASCII */
 103     uint32_t    checksum;               /* CRC-32C hash of the whole header */
 104     uint64_t    sequence_number;        /* Seq number of this header.  Each
 105                                            VHDX file has 2 of these headers,
 106                                            and only the header with the highest
 107                                            sequence number is valid */
 108     MSGUID      file_write_guid;        /* 128 bit unique identifier. Must be
 109                                            updated to new, unique value before
 110                                            the first modification is made to
 111                                            file */
 112     MSGUID      data_write_guid;        /* 128 bit unique identifier. Must be
 113                                            updated to new, unique value before
 114                                            the first modification is made to
 115                                            visible data.   Visbile data is
 116                                            defined as:
 117                                                     - system & user metadata
 118                                                     - raw block data
 119                                                     - disk size
 120                                                     - any change that will
 121                                                       cause the virtual disk
 122                                                       sector read to differ
 123
 124                                            This does not need to change if
 125                                            blocks are re-arranged */
 126     MSGUID      log_guid;               /* 128 bit unique identifier. If zero,
 127                                            there is no valid log. If non-zero,
 128                                            log entries with this guid are
 129                                            valid. */
 130     uint16_t    log_version;            /* version of the log format. Must be
 131                                            set to zero */
 132     uint16_t    version;                /* version of the vhdx file.  Currently,
 133                                            only supported version is "1" */
 134     uint32_t    log_length;             /* length of the log.  Must be multiple
 135                                            of 1MB */
 136     uint64_t    log_offset;             /* byte offset in the file of the log.
 137                                            Must also be a multiple of 1MB */
 138 } VHDXHeader;
 139
 140 /* Header for the region table block */
 141 #define VHDX_REGION_SIGNATURE  0x69676572  /* "regi" in ASCII */
 142 typedef struct QEMU_PACKED VHDXRegionTableHeader {
 143     uint32_t    signature;              /* "regi" in ASCII */
 144     uint32_t    checksum;               /* CRC-32C hash of the 64KB table */
 145     uint32_t    entry_count;            /* number of valid entries */
 146     uint32_t    reserved;
 147 } VHDXRegionTableHeader;
 148
 149 /* Individual region table entry.  There may be a maximum of 2047 of these
 150  *
 151  *  There are two known region table properties.  Both are required.
 152  *  BAT (block allocation table):  2DC27766F62342009D64115E9BFD4A08
 153  *  Metadata:                      8B7CA20647904B9AB8FE575F050F886E
 154  */
 155 #define VHDX_REGION_ENTRY_REQUIRED  0x01    /* if set, parser must understand
 156                                                this entry in order to open
 157                                                file */
 158 typedef struct QEMU_PACKED VHDXRegionTableEntry {
 159     MSGUID      guid;                   /* 128-bit unique identifier */
 160     uint64_t    file_offset;            /* offset of the object in the file.
 161                                            Must be multiple of 1MB */
 162     uint32_t    length;                 /* length, in bytes, of the object */
 163     uint32_t    data_bits;
 164 } VHDXRegionTableEntry;
 165
 166
 167 /* ---- LOG ENTRY STRUCTURES ---- */
 168 #define VHDX_LOG_MIN_SIZE (1024 * 1024)
 169 #define VHDX_LOG_SECTOR_SIZE 4096
 170 #define VHDX_LOG_HDR_SIZE 64
 171 #define VHDX_LOG_SIGNATURE 0x65676f6c
 172 typedef struct QEMU_PACKED VHDXLogEntryHeader {
 173     uint32_t    signature;              /* "loge" in ASCII */
 174     uint32_t    checksum;               /* CRC-32C hash of the 64KB table */
 175     uint32_t    entry_length;           /* length in bytes, multiple of 1MB */
 176     uint32_t    tail;                   /* byte offset of first log entry of a
 177                                            seq, where this entry is the last
 178                                            entry */
 179     uint64_t    sequence_number;        /* incremented with each log entry.
 180                                            May not be zero. */
 181     uint32_t    descriptor_count;       /* number of descriptors in this log
 182                                            entry, must be >= 0 */
 183     uint32_t    reserved;
 184     MSGUID      log_guid;               /* value of the log_guid from
 185                                            vhdx_header.  If not found in
 186                                            vhdx_header, it is invalid */
 187     uint64_t    flushed_file_offset;    /* see spec for full details - this
 188                                            should be vhdx file size in bytes */
 189     uint64_t    last_file_offset;       /* size in bytes that all allocated
 190                                            file structures fit into */
 191 } VHDXLogEntryHeader;
 192
 193 #define VHDX_LOG_DESC_SIZE 32
 194 #define VHDX_LOG_DESC_SIGNATURE 0x63736564
 195 #define VHDX_LOG_ZERO_SIGNATURE 0x6f72657a
 196 typedef struct QEMU_PACKED VHDXLogDescriptor {
 197     uint32_t    signature;              /* "zero" or "desc" in ASCII */
 198     union  {
 199         uint32_t    reserved;           /* zero desc */
 200         uint32_t    trailing_bytes;     /* data desc: bytes 4092-4096 of the
 201                                            data sector */
 202     };
 203     union {
 204         uint64_t    zero_length;        /* zero desc: length of the section to
 205                                            zero */
 206         uint64_t    leading_bytes;      /* data desc: bytes 0-7 of the data
 207                                            sector */
 208     };
 209     uint64_t    file_offset;            /* file offset to write zeros - multiple
 210                                            of 4kB */
 211     uint64_t    sequence_number;        /* must match same field in
 212                                            vhdx_log_entry_header */
 213 } VHDXLogDescriptor;
 214
 215 #define VHDX_LOG_DATA_SIGNATURE 0x61746164
 216 typedef struct QEMU_PACKED VHDXLogDataSector {
 217     uint32_t    data_signature;         /* "data" in ASCII */
 218     uint32_t    sequence_high;          /* 4 MSB of 8 byte sequence_number */
 219     uint8_t     data[4084];             /* raw data, bytes 8-4091 (inclusive).
 220                                            see the data descriptor field for the
 221                                            other mising bytes */
 222     uint32_t    sequence_low;           /* 4 LSB of 8 byte sequence_number */
 223 } VHDXLogDataSector;
 224
 225
 226
 227 /* block states - different state values depending on whether it is a
 228  * payload block, or a sector block. */
 229
 230 #define PAYLOAD_BLOCK_NOT_PRESENT       0
 231 #define PAYLOAD_BLOCK_UNDEFINED         1
 232 #define PAYLOAD_BLOCK_ZERO              2
 233 #define PAYLOAD_BLOCK_UNMAPPED          3
 234 #define PAYLOAD_BLOCK_UNMAPPED_v095     5
 235 #define PAYLOAD_BLOCK_FULLY_PRESENT     6
 236 #define PAYLOAD_BLOCK_PARTIALLY_PRESENT 7
 237
 238 #define SB_BLOCK_NOT_PRESENT    0
 239 #define SB_BLOCK_PRESENT        6
 240
 241 /* per the spec */
 242 #define VHDX_MAX_SECTORS_PER_BLOCK  (1 << 23)
 243
 244 /* upper 44 bits are the file offset in 1MB units lower 3 bits are the state
 245    other bits are reserved */
 246 #define VHDX_BAT_STATE_BIT_MASK 0x07
 247 #define VHDX_BAT_FILE_OFF_MASK  0xFFFFFFFFFFF00000ULL /* upper 44 bits */
 248 typedef uint64_t VHDXBatEntry;
 249
 250 /* ---- METADATA REGION STRUCTURES ---- */
 251
 252 #define VHDX_METADATA_ENTRY_SIZE 32
 253 #define VHDX_METADATA_MAX_ENTRIES 2047  /* not including the header */
 254 #define VHDX_METADATA_TABLE_MAX_SIZE \
 255     (VHDX_METADATA_ENTRY_SIZE * (VHDX_METADATA_MAX_ENTRIES+1))
 256 #define VHDX_METADATA_SIGNATURE 0x617461646174656DULL  /* "metadata" in ASCII */
 257 typedef struct QEMU_PACKED VHDXMetadataTableHeader {
 258     uint64_t    signature;              /* "metadata" in ASCII */
 259     uint16_t    reserved;
 260     uint16_t    entry_count;            /* number table entries. <= 2047 */
 261     uint32_t    reserved2[5];
 262 } VHDXMetadataTableHeader;
 263
 264 #define VHDX_META_FLAGS_IS_USER         0x01    /* max 1024 entries */
 265 #define VHDX_META_FLAGS_IS_VIRTUAL_DISK 0x02    /* virtual disk metadata if set,
 266                                                    otherwise file metdata */
 267 #define VHDX_META_FLAGS_IS_REQUIRED     0x04    /* parse must understand this
 268                                                    entry to open the file */
 269 typedef struct QEMU_PACKED VHDXMetadataTableEntry {
 270     MSGUID      item_id;                /* 128-bit identifier for metadata */
 271     uint32_t    offset;                 /* byte offset of the metadata.  At
 272                                            least 64kB.  Relative to start of
 273                                            metadata region */
 274                                         /* note: if length = 0, so is offset */
 275     uint32_t    length;                 /* length of metadata. <= 1MB. */
 276     uint32_t    data_bits;              /* least-significant 3 bits are flags,
 277                                            the rest are reserved (see above) */
 278     uint32_t    reserved2;
 279 } VHDXMetadataTableEntry;
 280
 281 #define VHDX_PARAMS_LEAVE_BLOCKS_ALLOCED 0x01   /* Do not change any blocks to
 282                                                    be BLOCK_NOT_PRESENT.
 283                                                    If set indicates a fixed
 284                                                    size VHDX file */
 285 #define VHDX_PARAMS_HAS_PARENT           0x02    /* has parent / backing file */
 286 #define VHDX_BLOCK_SIZE_MIN             (1   * MiB)
 287 #define VHDX_BLOCK_SIZE_MAX             (256 * MiB)
 288 typedef struct QEMU_PACKED VHDXFileParameters {
 289     uint32_t    block_size;             /* size of each payload block, always
 290                                            power of 2, <= 256MB and >= 1MB. */
 291     uint32_t data_bits;                 /* least-significant 2 bits are flags,
 292                                            the rest are reserved (see above) */
 293 } VHDXFileParameters;
 294
 295 #define VHDX_MAX_IMAGE_SIZE  ((uint64_t) 64 * TiB)
 296 typedef struct QEMU_PACKED VHDXVirtualDiskSize {
 297     uint64_t    virtual_disk_size;      /* Size of the virtual disk, in bytes.
 298                                            Must be multiple of the sector size,
 299                                            max of 64TB */
 300 } VHDXVirtualDiskSize;
 301
 302 typedef struct QEMU_PACKED VHDXPage83Data {
 303     MSGUID      page_83_data;           /* unique id for scsi devices that
 304                                            support page 0x83 */
 305 } VHDXPage83Data;
 306
 307 typedef struct QEMU_PACKED VHDXVirtualDiskLogicalSectorSize {
 308     uint32_t    logical_sector_size;    /* virtual disk sector size (in bytes).
 309                                            Can only be 512 or 4096 bytes */
 310 } VHDXVirtualDiskLogicalSectorSize;
 311
 312 typedef struct QEMU_PACKED VHDXVirtualDiskPhysicalSectorSize {
 313     uint32_t    physical_sector_size;   /* physical sector size (in bytes).
 314                                            Can only be 512 or 4096 bytes */
 315 } VHDXVirtualDiskPhysicalSectorSize;
 316
 317 typedef struct QEMU_PACKED VHDXParentLocatorHeader {
 318     MSGUID      locator_type;           /* type of the parent virtual disk. */
 319     uint16_t    reserved;
 320     uint16_t    key_value_count;        /* number of key/value pairs for this
 321                                            locator */
 322 } VHDXParentLocatorHeader;
 323
 324 /* key and value strings are UNICODE strings, UTF-16 LE encoding, no NULs */
 325 typedef struct QEMU_PACKED VHDXParentLocatorEntry {
 326     uint32_t    key_offset;             /* offset in metadata for key, > 0 */
 327     uint32_t    value_offset;           /* offset in metadata for value, >0 */
 328     uint16_t    key_length;             /* length of entry key, > 0 */
 329     uint16_t    value_length;           /* length of entry value, > 0 */
 330 } VHDXParentLocatorEntry;
 331
 332
 333 /* ----- END VHDX SPECIFICATION STRUCTURES ---- */
 334
 335 typedef struct VHDXMetadataEntries {
 336     VHDXMetadataTableEntry file_parameters_entry;
 337     VHDXMetadataTableEntry virtual_disk_size_entry;
 338     VHDXMetadataTableEntry page83_data_entry;
 339     VHDXMetadataTableEntry logical_sector_size_entry;
 340     VHDXMetadataTableEntry phys_sector_size_entry;
 341     VHDXMetadataTableEntry parent_locator_entry;
 342     uint16_t present;
 343 } VHDXMetadataEntries;
 344
 345 typedef struct VHDXLogEntries {
 346     uint64_t offset;
 347     uint64_t length;
 348     uint32_t write;
 349     uint32_t read;
 350     VHDXLogEntryHeader *hdr;
 351     void *desc_buffer;
 352     uint64_t sequence;
 353     uint32_t tail;
 354 } VHDXLogEntries;
 355
 356 typedef struct VHDXRegionEntry {
 357     uint64_t start;
 358     uint64_t end;
 359     QLIST_ENTRY(VHDXRegionEntry) entries;
 360 } VHDXRegionEntry;
 361
 362 typedef struct BDRVVHDXState {
 363     CoMutex lock;
 364
 365     int curr_header;
 366     VHDXHeader *headers[2];
 367
 368     VHDXRegionTableHeader rt;
 369     VHDXRegionTableEntry bat_rt;         /* region table for the BAT */
 370     VHDXRegionTableEntry metadata_rt;    /* region table for the metadata */
 371
 372     VHDXMetadataTableHeader metadata_hdr;
 373     VHDXMetadataEntries metadata_entries;
 374
 375     VHDXFileParameters params;
 376     uint32_t block_size;
 377     uint32_t block_size_bits;
 378     uint32_t sectors_per_block;
 379     uint32_t sectors_per_block_bits;
 380
 381     uint64_t virtual_disk_size;
 382     uint32_t logical_sector_size;
 383     uint32_t physical_sector_size;
 384
 385     uint64_t chunk_ratio;
 386     uint32_t chunk_ratio_bits;
 387     uint32_t logical_sector_size_bits;
 388
 389     uint32_t bat_entries;
 390     VHDXBatEntry *bat;
 391     uint64_t bat_offset;
 392
 393     bool first_visible_write;
 394     MSGUID session_guid;
 395
 396     VHDXLogEntries log;
 397
 398     VHDXParentLocatorHeader parent_header;
 399     VHDXParentLocatorEntry *parent_entries;
 400
 401     Error *migration_blocker;
 402
 403     bool log_replayed_on_open;
 404
 405     QLIST_HEAD(VHDXRegionHead, VHDXRegionEntry) regions;
 406 } BDRVVHDXState;
 407
 408 void vhdx_guid_generate(MSGUID *guid);
 409
 410 int vhdx_update_headers(BlockDriverState *bs, BDRVVHDXState *s, bool rw,
 411                         MSGUID *log_guid);
 412
 413 uint32_t vhdx_update_checksum(uint8_t *buf, size_t size, int crc_offset);
 414 uint32_t vhdx_checksum_calc(uint32_t crc, uint8_t *buf, size_t size,
 415                             int crc_offset);
 416
 417 bool vhdx_checksum_is_valid(uint8_t *buf, size_t size, int crc_offset);
 418
 419 int vhdx_parse_log(BlockDriverState *bs, BDRVVHDXState *s, bool *flushed,
 420                    Error **errp);
 421
 422 int vhdx_log_write_and_flush(BlockDriverState *bs, BDRVVHDXState *s,
 423                              void *data, uint32_t length, uint64_t offset);
 424
 425 static inline void leguid_to_cpus(MSGUID *guid)
 426 {
 427     le32_to_cpus(&guid->data1);
 428     le16_to_cpus(&guid->data2);
 429     le16_to_cpus(&guid->data3);
 430 }
 431
 432 static inline void cpu_to_leguids(MSGUID *guid)
 433 {
 434     cpu_to_le32s(&guid->data1);
 435     cpu_to_le16s(&guid->data2);
 436     cpu_to_le16s(&guid->data3);
 437 }
 438
 439 void vhdx_header_le_import(VHDXHeader *h);
 440 void vhdx_header_le_export(VHDXHeader *orig_h, VHDXHeader *new_h);
 441 void vhdx_log_desc_le_import(VHDXLogDescriptor *d);
 442 void vhdx_log_desc_le_export(VHDXLogDescriptor *d);
 443 void vhdx_log_data_le_import(VHDXLogDataSector *d);
 444 void vhdx_log_data_le_export(VHDXLogDataSector *d);
 445 void vhdx_log_entry_hdr_le_import(VHDXLogEntryHeader *hdr);
 446 void vhdx_log_entry_hdr_le_export(VHDXLogEntryHeader *hdr);
 447 void vhdx_region_header_le_import(VHDXRegionTableHeader *hdr);
 448 void vhdx_region_header_le_export(VHDXRegionTableHeader *hdr);
 449 void vhdx_region_entry_le_import(VHDXRegionTableEntry *e);
 450 void vhdx_region_entry_le_export(VHDXRegionTableEntry *e);
 451 void vhdx_metadata_header_le_import(VHDXMetadataTableHeader *hdr);
 452 void vhdx_metadata_header_le_export(VHDXMetadataTableHeader *hdr);
 453 void vhdx_metadata_entry_le_import(VHDXMetadataTableEntry *e);
 454 void vhdx_metadata_entry_le_export(VHDXMetadataTableEntry *e);
 455 int vhdx_user_visible_write(BlockDriverState *bs, BDRVVHDXState *s);
 456
 457 #endif