RAR (Roshal Archive)

RAR is an archive format: it packs many files and folders into one file, compresses them, and can encrypt them with a password. It’s common for software and media downloads, comic books (.cbr) and backups. Its strengths are a good compression ratio, archives split into volumes, and a recovery record that repairs damaged archives.

RAR is proprietary. Only WinRAR and the rar command line tool from RARLAB create RAR files. Extraction is open: RARLAB publishes the UnRAR source code, which can be used in any program that extracts RAR, but not to build a RAR-compatible archiver. All open-source libraries on this site only read RAR.

When to use RAR#

• Use to extract archives you received: all major platforms have free tools that open RAR.
• Use when you already own WinRAR and need volumes plus a recovery record (to survive damaged downloads or disks) in one archive.
• Avoid for archives you create for other people: ZIP opens everywhere without extra software, 7z compresses as well or better with an open format.
• Avoid in programs that must create archives: no open-source library writes RAR. Use ZIP, 7z or TAR with gzip, xz or zstd.

RAR4 vs RAR5#

RAR has 2 incompatible archive formats. Both use the .rar extension; the signature tells them apart.

Current WinRAR creates RAR5 by default and creates RAR4 only on request. Archives with a dictionary larger than 4 GB (a RAR 7.0 option) need UnRAR 7.0 or newer to extract.

File structure#

A RAR file is a signature followed by a sequence of blocks (headers). Each header may be followed by a data area, such as the compressed bytes of a file. An optional SFX module (an executable that extracts the archive) can come before the signature; readers search for the signature in the first megabyte or so.

RAR5 layout#

[SFX module]                optional, self-extracting archives
signature                   52 61 72 21 1A 07 01 00
[archive encryption header] only when file names are encrypted
main archive header
file header + data          one per file (and per volume part)
service header + data       optional: comment (CMT), NTFS streams (STM), quick open (QO)
[recovery record]           service header named RR
end of archive header

Every RAR5 header starts with the same fields. Sizes and most numbers are vint: variable-length integers with 7 bits per byte, low bits first, and the high bit set on every byte except the last.

The start of a typical RAR5 file:

52 61 72 21 1A 07 01 00   signature "Rar!\x1A\x07\x01\x00"
xx xx xx xx               header CRC32
0A                        header size (vint): 10 bytes follow
01                        header type 1: main archive header
05                        header flags: extra area present, skip if unknown
06                        extra area size: 6
00                        archive flags: not a volume, not solid
...                       extra area (locator record), then file headers (type 2)

Main archive header flags: 0x0001 volume, 0x0002 volume number present (all volumes except the first), 0x0004 solid, 0x0008 recovery record present, 0x0010 locked (can’t be modified).

File header fields after the common part: file flags (directory, mtime present, CRC32 present, unknown size), unpacked size, attributes, modification time, CRC-32, compression info, host OS (0 Windows, 1 Unix), and the UTF-8 name. The compression info vint holds the method in bits 7-9 (0 = store, 1-5 = fastest to best) and the dictionary size in bits 10-14. Extra records in a file header add encryption parameters, a BLAKE2sp hash, high-precision times, symlinks and hard links, and Unix owner names.

RAR4 layout#

RAR4 uses fixed-size fields in little-endian order. Every block starts with a 7-byte header:

The 7-byte signature is itself a marker block (0x72). A file block (0x74) stores packed size, unpacked size, host OS, CRC-32, DOS date and time, the version needed to extract, the method (0x30 store to 0x35 best), name length, attributes, the name, and an 8-byte salt for encrypted files. Files over 4 GB add 32-bit high parts of the sizes.

Solid archives#

In a solid archive the files are compressed as one continuous stream, so similar files share the dictionary and compress much better. The cost: extracting one file means decompressing all files before it, and damage in one place breaks every file after it. Libraries that extract solid archives are much faster when they process entries in order.

Volumes#

A large archive can be split into volumes of a fixed size. RAR5 archives and most newer RAR4 archives name them name.part1.rar, name.part2.rar; old RAR4 archives use name.rar, name.r00, name.r01. Open the first volume to extract: tools find the next ones by name. A file that spans volumes has a file header in each volume with the “continues” flags set.

Encryption#

RAR encrypts file data, and optionally all headers too (WinRAR’s “Encrypt file names”). Without the password, an archive with encrypted headers doesn’t even show its file list. RAR5 uses AES-256 in CBC mode with a key from PBKDF2-HMAC-SHA256 (2¹⁵ iterations by default), and stores a check value so a wrong password is detected before extraction. RAR4 uses AES-128.

Recovery record and recovery volumes#

A recovery record is extra redundancy data (a percentage of the archive size) stored inside the archive. WinRAR’s “Repair” command uses it to fix damaged sectors. Recovery volumes (.rev) do the same for whole missing volumes. Extraction libraries ignore both.

Specifications and documentation#

• RAR 5.0 archive format (RARLAB): the official description of RAR5 headers, without the compression algorithm
• UnRAR source code (RARLAB): the reference decompressor for RAR4 and RAR5, the only complete description of the compression
• UnRAR 7.0 implementation notes: what changed for RAR 7.0 large dictionaries
• RAR 4 format description: RAR4 block layout, from the bitjs project
• application/vnd.rar media type registration (IANA)
• RAR on Wikipedia

Software that opens RAR files#

Libraries#

Open-source libraries extract RAR files; none create them. See all RAR libraries for a comparison.

Related formats#

• ZIP: the most compatible archive format, built into every OS, open specification
• 7z: open format with LZMA2, usually as small as or smaller than RAR
• TAR: Unix archive format, compressed with gzip, xz or zstd
• CBR: a comic book is a RAR archive of images renamed to .cbr; CBZ is the ZIP version