zvault/docs/repository_readme.md

21 KiB

ZVault repository

This folder is a zVault remote repository and contains backup data.

The repository contains the following components:

  • The backup bundles in the subfolder bundles. The individual files are organized in subfolders and named after their bundle ids. The structure and names of the files is not important as the files include the bundle id in their headers. Thus the files can be renamed and reorganized.
  • The backup anchor files in the subfolder backups. The names of the files and their structure determine the backup names but are not used otherwise.
  • Active locks in the subfolder locks. This folder only contains lock files when the repository is currently used. If any zVault process crashes, a stale lock file might be left back. Those files can be safely removed if no process is running for sure.

Repository format

In case the zVault software is not available for restoring the backups included in this repository the following sections describe the format of the repository so that its contents can be read without zVault.

Bundle files

The bundle file format consists of 5 parts:

  • A magic header with version
  • A tiny header with encryption information
  • An encoded and encrypted bundle information structure
  • An encoded and encrypted chunk list
  • The chunk data (compressed and encrypted)

The main reason for having those multiple parts is that it is expected that the smaller front parts can be read much faster than the the whole file. So information that is needed more frequently is put into earlier parts and the data that is need the least frequent is put into the latter part so that it does not slow down reading the front parts. Keeping those parts in separate files was also considered but rejected to increase the reliability of the storage.

Magic header with version

The first part of a bundle file contains an 8 byte magic header with version information.

The first 6 bytes of the header consist of the fixed string "zvault", followed by one byte with the fixed value 0x01. Those 7 bytes make up the magic header of the file and serve to identify the file type as a zvault bundle file.

The 8th byte of the first file part is the version of the file format. This value is currently 0x01 and is expected to be increased for any breaking changes in the file format.

Encryption header

The encryption header is the second part of the bundle file format and follows directly after the 8 bytes of the magic header.

The header structure is defined in the appendix as BundleHeader and contains information on how to decrypt the other parts of the bundle as well as the encrypted size of the following bundle information.

Please note that this header even exists when the bundle is not encrypted (the header then contains no encryption method).

Bundle information

The bundle information structure is the third part of the bundle file format and follows directly after the encryption header.

The information structure is defined in the appendix as BundleInfo and contains general information on the bundle's contents and on how to decode the other two parts of the bundle file.

This structure is encrypted using the method described in the previous encryption header since it contains confidential information (the bundle id could be used to identify the data contained in the bundle). The size of the encrypted structure is also stored in the previous header. This structure is not compressed, as it is pretty small.

Encoded chunk list

The chunk list is the forth part of the bundle file and follows directly after the bundle information structure.

The chunk list contains hashes and sizes of all chunks stored in this bundle in the order they are stored. The list is encoded as defined in the appendix as ChunkList.

Since the chunk list contains confidential information (the chunk hashes and sized can be used to identify files) the encoded chunk list is encrypted using the encryption method specified in the encryption header. The bundle information structure contains the full size of the encoded and encrypted chunk list as chunk_list_size which is needed since the encryption could add some bytes for a nonce or an authentication code.

The chunk list is not compressed since the hashes have a very high entropy and do not compress significantly.

The chunk list is not stored in the bundle info structure because it can be pretty big compared to the info structure which needs to be read more often.

Chunk data

The chunk data is the final part of a bundle file and follows after the chunk list. The starting position can be obtained from the bundle info structure as the encoded size of the chunk list is stored there as chunk_list_size.

The chunk data part consists of the data of the chunks contained in this bundle simply concatenated without any separator. The individual chunk sizes can be obtained from the chunk list. The starting position of any chunk can be calculated by summing up the sized of all previous chunks.

The chunk data is compressed as whole (solid archive) and encrypted with the methods specified in the bundle information structure.

Backup format

The repository contains multiple backups that share the data contained in the bundles. The individual backups are encoded in backup files as described in the following section. Those backup files reference a list of chunks in the bundles as a root inode entry. Each inode entry references lists of chunks for its data and potential child entries.

All chunks that are referenced either in the backup files or in the inode entries are contained in one of the bundles and is uniquely identified by its hash. An index, e.g. a hash table, can help to find the correct bundle quickly.

Backup files

Backup files contain information on one specific backup and reference the directory root of that backup.

Backup files consist of 3 parts:

  • A magic header with version
  • A tiny header with encryption information
  • An encoded and encrypted backup information structure
Magic header with version

The first part of a backup file contains an 8 byte magic header with version information.

The first 6 bytes of the header consist of the fixed string "zvault", followed by one byte with the fixed value 0x03. Those 7 bytes make up the magic header of the file and serve to identify the file type as a zvault backup file.

The 8th byte of the first file part is the version of the file format. This value is currently 0x01 and is expected to be increased for any breaking changes in the file format.

Encryption header

The encryption header is the second part of the backup file format and follows directly after the 8 bytes of the magic header.

The header structure is defined in the appendix as BackupHeader and contains information on how to decrypt the rest of the backup file.

Please note that this header even exists when the backup file is not encrypted (the header then contains no encryption method).

Backup information

The backup information structure is the final part of the backup file format and follows directly after the encryption header.

The information structure is defined in the appendix as Backup and contains general information on the backup's contents and references the directory root of the backup tree.

This structure is encrypted using the method described in the previous encryption header since it contains confidential information. This structure is not compressed, as it is pretty small.

Directories & file data

The inode entries are encoded as defined in the appendix as Inode. The inode structure contains all meta information on an inode entry, e.g. its file type, the data size, modification time, permissions and ownership, etc. Also, the structure contains optional information that is specific to the file type. For regular files, the inode structure contains the data of that file either inline (for very small files) or as a reference via a chunk list. For directories, the inode structure contains a mapping of child inode entries with their name as key and a chunk list referring their encoded Inode structure as value. For symlinks, the inode structure contains the target in the field symlink_target.

Starting from the root of the Backup structure, the whole backup file tree can be reconstructed by traversing the children of each inode recursively. Since files can only be retrieved by traversing their parent directories, they contain no back link to their parent directory.

Appendix

MessagePack encoding

Most zvault structures are encoded using the MessagePack encoding as specified at http://www.msgpack.org. The version of MessagePack that is used, is dated to 2013-04-21.

All structure encodings are based on a mapping that associates values to the structure's fields. In order to save space, the structure's fields are not referenced by name but by an assigned number. In the encoding specification, this is written as FIELD: TYPE => NUMBER where FIELD is the field name used to reference the field in the rest of the description, TYPE is the type of the field's values and NUMBER is the number used as key for this field in the mapping.

The simple types used are called null, bool, int, float, string and bytes that correspond to the MessagePack data types (null means Nil, bytes means Binary and the other types are lower case to distinguish them from custom types).

Complex data types are noted as {KEY => VALUE} for mappings and [TYPE] for arrays. Tuples of multiple types e.g. (TYPE1, TYPE2, TYPE3) are also encoded as arrays but regarded as differently as they contain different types and have a fixed length.

If a field is optional, its type is listed as TYPE? which means that either null or the TYPE is expected. If a value of TYPE is given. the option is regarded as being set and if null is given, the option is regarded as not being set.

If a structure contains fields with structures or other complex data types, the values of those fields are encoded as described for those values (often again as a mapping on their own). The encoding specification uses the name of the structure as a field type in this case.

For some structures, there exist a set of default values for the structure's fields. If any field is missing in the encoded mapping, the corresponding value from the defaults will be taken instead.

Constants

The following types are used as named constants. In the encoding, simply the value (mostly a number) is used instead of the name but in the rest of the specification the name is used for clarity.

BundleMode

The BundleMode describes the contents of the chunks of a bundle.

  • Data means that the chunks contain file data

  • Meta means that the chunks either contain encoded chunk lists or encoded inode metadata

    BundleMode { Data => 0, Meta => 1 }

HashMethod

The HashMethod describes the method used to create fingerprint hashes from chunk data. This is not relevant for reading backups.

  • Blake2 means the hash method Blake2b as described in RFC 7693 with the hash length set to 128 bits.

  • Murmur3 means the hash method MurmurHash3 as described at https://en.wikipedia.org/wiki/MurmurHash for the x64 architecture and with the hash length set to 128 bits.

    HashMethod { Blake2 => 1, Murmur3 => 2 }

EncryptionMethod

The EncryptionMethod describes the method used to encrypt (and thus also decrypt) data.

  • Sodium means the crypto_box_seal method of libsodium as specified at http://www.libsodium.org as a combination of X25519 and XSalsa20-Poly1305.

    EncryptionMethod { Sodium => 0 }

CompressionMethod

The CompressionMethod describes a compression method used to compress (and thus also decompress) data.

  • Deflate means the gzip/zlib method (without header) as described in RFC 1951

  • Brotli means the Google Brotli method as described in RFC 7932

  • Lzma means the LZMA method (XZ stream format) as described at http://tukaani.org/xz/

  • Lz4 means the LZ4 method as described at http://www.lz4.org

    CompressionMethod { Deflate => 0, Brotli => 1, Lzma => 2, Lz4 => 3 }

FileType

The FileType describes the type of an inode.

  • File means on ordinary file that contains data

  • Directory means a directory that does not contain data but might have children

  • Symlink means a symlink that points to a target

  • BlockDevice means a block device

  • CharDevice means a character device

  • NamedPipe means a named pipe/fifo

    FileType { File => 0, Directory => 1, Symlink => 2, BlockDevice => 3, CharDevice => 4, NamedPipe => 5 }

Types

The following types are used to simplify the encoding specifications. They can simply be substituted by their definitions. For simplicity, their names will be used in the encoding specifications instead of their definitions.

Encryption

The Encryption is a combination of an EncryptionMethod and a key. The method specifies how the key was used to encrypt the data. For the Sodium method, the key is the public key used to encrypt the data with. The secret key needed for decryption, must correspond to that public key.

Encryption = (EncryptionMethod, bytes)

Compression

The Compression is a micro-structure containing the compression method and the compression level. The level is only used for compression.

Compression {
    method: CompressionMethod => 0,
    level: int => 1
}

BundleHeader encoding

The BundleHeader structure contains information on how to decrypt other parts of a bundle. The structure is encoded using the MessagePack encoding that has been defined in a previous section. The encryption field contains the information needed to decrypt the rest of the bundle parts. If the encryption option is set, the following parts are encrypted using the specified method and key, otherwise the parts are not encrypted. The info_size contains the encrypted size of the following BundleInfo structure.

BundleHeader {
    encryption: Encryption? => 0,
    info_size: int => 1
}

BundeInfo encoding

The BundleInfo structure contains information on a bundle. The structure is encoded using the MessagePack encoding that has been defined in a previous section. If the compression option is set, the chunk data is compressed with the specified method, otherwise it is uncompressed. The encrypted size of the following ChunkList is stored in the chunk_list_size field.

BundeInfo {
    id: bytes => 0,
    mode: BundleMode => 1,
    compression: Compression? => 2,
    hash_method: HashMethod => 4,
    raw_size: int => 6,
    encoded_size: int => 7,
    chunk_count: int => 8,
    chunk_list_size: int => 9
}

This structure is encoded with the following field default values:

  • hash_method: Blake2
  • mode: Data
  • All other fields: 0, null or an empty byte sequence depending on the type.

ChunkList encoding

The ChunkList contains a list of chunk hashes and chunk sizes. This list is NOT encoded using the MessagePack format as a simple binary format is much more efficient in this case.

For each chunk, the hash and its size are encoded in the following way:

  • The hash is encoded as 16 bytes (little-endian).
  • The size is encoded as a 32-bit value (4 bytes) in little-endian. The encoded hash and the size are concatenated (hash first, size second) yielding 20 bytes for each chunk. Those 20 bytes of encoded chunk information are concatenated for all chunks in the list in order or appearance in the list.

Inode encoding

The Inode structure contains information on a backup inode, e.g. a file or a directory. The structure is encoded using the MessagePack encoding that has been defined in a previous section. The name field contains the name of this inode which can be concatenated with the names of all parent inodes (with a platform-dependent seperator) to form the full path of the inode. The size field contains the raw size of the data in bytes (this is 0 for everything except files). The file_type specifies the type of this inode. The mode field specifies the permissions of the inode as a number which is normally interpreted as octal. The user and group fields specify the ownership of the inode in the form of user and group id. The timestamp specifies the modification time of the inode in whole seconds since the UNIX epoch (1970-01-01 12:00 am). The symlink_target specifies the target of symlink inodes and is only set for symlinks. The data specifies the data of a file and is only set for regular files. The data is specified as a tuple of nesting and bytes. If nesting is 0, bytes contains the data of the file. This "inline" format is only used for small files. If nesting is 1, bytes is an encoded ChunkList (as described in a previous section). The concatenated data of those chunks make up the data of the file. If nesting is 2, bytes is also an encoded ChunkList, but the concatenated data of those chunks form again an encoded ChunkList which in turn contains the chunks with the file data. Thus nesting specifies the number of indirection steps via ChunkLists. The children field specifies the child inodes of a directory and is only set for directories. It is a mapping from the name of the child entry to the bytes of the encoded chunklist of the encoded Inode structure of the child. It is important that the names in the mapping correspond with the names in the respective child Inodes and that the mapping is stored in alphabetic order of the names. The cum_size, cum_dirs and cum_files are cumulative values for the inode as well as the whole subtree (including all children recursively). cum_size is the sum of all inode data sizes plus 1000 bytes for each inode (for encoded metadata). cum_dirs and cum_files is the count of directories and non-directories (symlinks and regular files). The xattrs contains a mapping of all extended attributes of the inode. And device contains a tuple with the major and minor device id if the inode is a block or character device.

Inode {
    name: string => 0,
    size: int => 1,
    file_type: FileType => 2,
    mode: int => 3,
    user: int => 4,
    group: int => 5,
    timestamp: int => 7,
    symlink_target: string? => 9,
    data: (int, bytes)? => 10,
    children: {string => bytes}? => 11,
    cum_size: int => 12,
    cum_dirs: int => 13,
    cum_files: int => 14
    xattrs: {string => bytes}? => 15,
    device: (int, int)? => 16
}

This structure is encoded with the following field default values:

  • file_type: File
  • mode: 0o644
  • user and group: 1000
  • All other fields: 0, null or an empty string depending on the type.

BackupHeader encoding

The BackupHeader structure contains information on how to decrypt the rest of the backup file. The structure is encoded using the MessagePack encoding that has been defined in a previous section. The encryption field contains the information needed to decrypt the rest of the backup file. If the encryption option is set, the rest of the backup file is encrypted using the specified method and key, otherwise the rest is not encrypted.

BackupHeader {
    encryption: Encryption? => 0
}

Backup encoding

The Backup structure contains information on one specific backup and references the root of the backup file tree. The structure is encoded using the MessagePack encoding that has been defined in a previous section. The root field contains an encoded ChunkList that references the root of the backup file tree. The fields total_data_size, changed_data_size, deduplicated_data_size and encoded_data_size list the sizes of the backup in various stages in bytes.

  • total_data_size gives the cumulative sizes of all entries in the backup.

  • changed_data_size gives the size of only those entries that changed since the reference backup.

  • deduplicated_data_size gives the cumulative raw size of all new chunks in this backup that have not been stored in the repository yet.

  • encoded_data_size gives the cumulative encoded (and compressed) size of all new bundles that have been written specifically to store this backup. The fields bundle_count and chunk_count contain the number of new bundles and chunks that had to be written to store this backup. avg_chunk_size is the average size of new chunks in this backup. The field date specifies the start of the backup run in seconds since the UNIX epoch and the field duration contains the duration of the backup run in seconds as a floating point number containing also fractions of seconds. The fields file_count and dir_count contain the total number of non-directories and directories in this backup. The host and path field contain the host name and the the path on that host where the root of the backup was located. The field config contains the configuration of zVault during the backup run.

    Backup { root: bytes => 0, total_data_size: int => 1, changed_data_size: int => 2, deduplicated_data_size: int => 3, encoded_data_size: int => 4, bundle_count: int => 5, chunk_count: int => 6, avg_chunk_size: float => 7, date: int => 8, duration: float => 9, file_count: int => 10, dir_count: int => 11, host: string => 12, path: string => 13, config: Config => 14 }