Unix dump/restore tape format

The Unix dump command makes a full or incremental backup of a single disk partition to magnetic tapes or files. The corresponding restore command recovers data from such tapes. The tape format is vaguely documented in various man pages, but not to a detail necessary to fully understand it. This primarily describes the formats used by Unix V7, 4BSD, and SunOS around the 1980s.

A dump is logically a series of 1024-byte blocks, grouped ten at a time into tape records. There are no tape marks interspersed between records, just the normal double mark and the end of a tape. A block can be a header, or data. Header blocks hold metadata and inode information.

Dumps will normally span several tapes, each of which is called a volume. All volumes begin with a TS_TAPE block. The last volume ends with one or several TS_END blocks. The first volume will have TS_BITS and TS_CLRI blocks after TS_TAPE. The rest of the tape blocks are of type TS_INODE, storing raw inodes as stored on disk, followed by data blocks. In some cases a TS_INODE header can not store all information needed, in which case it can be extended with TS_ADDR blocks. Normally the first inodes written to tape is the entire directory structure of the whole file system, and file inodes make up the rest of the volumes.

The tape format comes in two versions. The first version is for the older UNIX file system, and the second version is for Berkeley's FFS file system. On top of that, data can be 16 or 32 bits wide, and big or little endian.

Notes:

• Since dump reads low-level file system information like inode numbers, it's limited to a single partition.
• If the first volume has been lost, it's likely that directories and filenames can't be recovered.
• File data blocks may spill over to the next volume, but look out for that TS_TAPE block.

TAPE HEADER

This data is common to all headers, but some of the fields may be valid or ignored depending on the header type.

An int is 16 or 32 bits depending on the host.

The checksum is a plain two's complement sum of all ints in the 1024-byte heder block.

Name	Description	Size	Value (old, new format)
Type	Header type	int	TS_TAPE, TS_BITS, TS_CLRI, TS_INODE, TS_ADDR, TS_END
Date	Date of dump	32 bits	Timestamp
Ddate	Dump from	32 bits	Timestamp from previous incremental backup, or 0 for full
Volume	Tape number	int	Tape in dump, from 1
Tapea	Block number	32 bits	Block in dump across all tapes, from 1
Inumber	Inode number	int	File system inode number
Magic	Format identifier	int	60011	60012
Checksum	Checksum	int	Filled in to make sum 84446 (modulo int)
Inode	Inode data	64 or 128 bytes	As stored on disk
Count	Number of addr bytes or data blocks	int
Addr	Data present	Array of bytes	0 or 1

The inode data looks like this:

Name	Old Size	New Size
Mode	16 bits	16 bits
Nlink	16 bits	16 bits
Uid	16 bits	16 bits
Gid	16 bits	16 bits
Size	32 bits	64 bits
Addr	40 bytes	Not present
Atime	32 bits	32 bits
?	Not present	32 bits
Mtime	32 bits	32 bits
?	Not present	32 bits
Ctime	32 bits	32 bits
?	Not present	92 bytes
Total	64 bytes	128 bytes

TS_TAPE

Header type 1. This header must start all volumes. The Date field must be same on all volumes. The Volume field identifies the individual tape in the dump. There are no data blocks.

TS_BITS and TS_CLRI

Header types 3 and 6. Stores bit arrays from the file system. The Count field specifies the number of data blocks following the header block.

TS_INODE and TS_ADDR

Header types 2 and 4. TS_INODE stores the inode data from disk. After the header block comes data blocks, as per the Addr array. An array item of one signals a data block is present on tape. A zero indicates a 1024-byte hole in the file. If a volume runs out of space during a run of data blocks, the next volume begins with a TS_TAPE block, and then follows the rest of the data blocks.

The Addr field is normally 256 byte at the most. If it can't accommodate a large file, several TS_ADDR may follow. A TS_ADDR header stores the same inode data, but following data blocks are appended to the file.

Ordinary files have bit 100000 (octal) set in the Mode field. Directories have bit 40000 set. Note that large directories may need additional TS_ADDR blocks.

The directory for the old format has fixed-length entries.

Size	Description
16 bits	Inode number
14 bytes	ASCII file name

The directory for the new format has variable-length entries. An entry length of 0 signals the end of the directory.

Size	Description
32 bits	Inode number
16 bits	Entry length
16 bits	Name length
? bytes	ASCII file name

TS_END

Header type 5. This header signals the end of the dump on the last volume. Several TS_END blocks may be written. There are no data blocks.