path: root/doc/crypt.5

.\" Written and revised by Solar Designer <solar at openwall.com> in 2000-2011.
.\" Revised by Zack Weinberg <zackw at panix.com> in 2017.
.\" Converted to mdoc format by Zack Weinberg in 2018.
.\"
.\" No copyright is claimed, and this man page is hereby placed in the public
.\" domain.  In case this attempt to disclaim copyright and place the man page
.\" in the public domain is deemed null and void, then the man page is
.\" Copyright 2000-2011 Solar Designer, 2017 Zack Weinberg, and it is
.\" hereby released to the general public under the following terms:
.\"
.\" Redistribution and use in source and binary forms, with or without
.\" modification, are permitted.
.\"
.\" There's ABSOLUTELY NO WARRANTY, express or implied.
.\"
.Dd October 11, 2017
.Dt CRYPT 5
.Os "Openwall Project"
.Sh NAME
.Nm crypt
.Nd storage format for hashed passphrases and available hashing methods
.Sh DESCRIPTION
The hashing methods implemented by
.Xr crypt 3
are designed only to process user passphrases for storage and authentication;
they are not suitable for use as general-purpose cryptographic hashes.
.Pp
Passphrase hashing is not a replacement for strong passphrases.
It is always possible
for an attacker with access to the hashed passphrases
to guess and check possible cleartext passphrases.
However, with a strong hashing method,
guessing will be too slow for the attacker
to discover a strong passphrase.
.Pp
All of the hashing methods use a
.Dq salt
to perturb the hash function,
so that the same passphrase may produce many possible hashes.
Newer methods accept longer salt strings.
The salt should be chosen at random for each user.
Salt defeats a number of attacks:
.Bl -enum
.It
It is not possible to hash a passphrase once
and then test it against each account's stored hash;
the hash calculation must be repeated for each account.
.It
It is not possible to tell whether two accounts use the same passphrase
without successfully guessing one of the phrases.
.It
Tables of precalculated hashes of commonly used passphrases
must have an entry for each possible salt,
which makes them impractically large.
.El
.Pp
All of the hashing methods are also deliberately engineered to be slow;
they use many iterations of an underlying cryptographic primitive
to increase the cost of each guess.
The newer hashing methods allow the number of iterations to be adjusted,
using the
.Dq CPU time cost
parameter to
.Xr crypt_gensalt 3 .
This makes it possible to keep the hash slow as hardware improves.
.Sh FORMAT OF HASHED PASSPHRASES
All of the hashing methods supported by
.Xr crypt 3
produce a hashed passphrase which consists of four components:
.Ar prefix ,
.Ar options ,
.Ar salt ,
and
.Ar hash .
The prefix controls which hashing method is to be used, and is the
appropriate string to pass to
.Xr crypt_gensalt 3
to select that method.
The contents of
.Ar options ,
.Ar salt ,
and
.Ar hash
are up to the method.
Depending on the method, the
.Ar prefix
and
.Ar options
components may be empty.
.Pp
The
.Fa setting
argument to
.Xr crypt 3
must begin with the first three components of a valid hashed passphrase,
but anything after that is ignored.
This makes authentication simple:
hash the input passphrase using the stored passphrase as the setting,
and then compare the result to the stored passphrase.
.Pp
Hashed passphrases are always entirely printable ASCII,
and do not contain any whitespace
or the characters
.Sq Li \&: ,
.Sq Li \&; ,
.Sq Li \&* ,
.Sq Li \&! ,
or
.Sq Li \&\e .
(These characters are used as delimiters and special markers in the
.Xr passwd 5
and
.Xr shadow 5
files.)
.Pp
The syntax of each component of a hashed passphrase
is up to the hashing method.
.Sq Li \&$
characters usually delimit components,
and the salt and hash are usually encoded as numerals in base 64.
The details of this base-64 encoding vary among hashing methods.
The common
.Dq base64
encoding specified by RFC 4648 is usually
.Em not
used.
.Sh AVAILABLE HASHING METHODS
This is a list of
.Em all
the hashing methods supported by
.Xr crypt 3 ,
in decreasing order of strength.
Many of the older methods
are now considered too weak to use for new passphrases.
The hashed passphrase format is expressed
with extended regular expressions (see
.Xr regex 7 )
and does not show the division into prefix, options, salt, and hash.
.de hash
.Bl -tag -width 2n
.It Sy Prefix
.\" mandoc bug: .Qq comes out with curly quotes.
.\" mandoc bug: .Li is hyperlinked to itself for no apparent reason.
.Bf Li
"\\$1"
.Ef
.if "\\$1"" (empty string)
.It Sy Hashed passphrase format
.\" mandoc bug: .Li is hyperlinked to itself for no apparent reason.
.Bf -literal
\&\\$2
.Ef
.It Sy Maximum passphrase length
.ie "\\$3"unlimited" unlimited
.el \\$3 characters
.if "\\$4"7" (ignores 8th bit)
.It Sy Hash size
\\$6 bits
.if !"\\$5"\\$6" \{\
.It Sy Effective key size
\&\\$5 bits
.\}
.It Sy Salt size
\\$7 bits
.It Sy CPU time cost parameter
\\$8
.El
..
.Ss yescrypt
yescrypt is a scalable passphrase hashing scheme designed by Solar Designer,
which is based on Colin Percival's scrypt.
Recommended for new hashes.
.hash "$y$" "\e$y\e$[./A-Za-z0-9]+\e$[./A-Za-z0-9]{,86}\e$[./A-Za-z0-9]{43}" unlimited 8 256 256 "up to 512 (128+ recommended)" "1 to 11 (logarithmic)"
.Ss gost-yescrypt
gost-yescrypt uses the output from the yescrypt hashing method in place of a
hmac message.  Thus, the yescrypt crypto properties are superseded by the
GOST R 34.11-2012 (Streebog) hash function with a 256 bit digest.
This hashing method is useful in applications that need modern passphrase
hashing methods, but require to rely on the cryptographic properties of GOST
algorithms.
The GOST R 34.11-2012 (Streebog) hash function has been published by the IETF
as RFC 6986.
Recommended for new hashes.
.hash "$gy$" "\e$gy\e$[./A-Za-z0-9]+\e$[./A-Za-z0-9]{,86}\e$[./A-Za-z0-9]{43}" unlimited 8 256 256 "up to 512 (128+ recommended)" "1 to 11 (logarithmic)"
.Ss scrypt
scrypt is a password-based key derivation function created by Colin Percival,
originally for the Tarsnap online backup service.
The algorithm was specifically designed to make it costly to perform
large-scale custom hardware attacks by requiring large amounts of memory.
In 2016, the scrypt algorithm was published by IETF as RFC 7914.
.hash "$7$" "\e$7\e$[./A-Za-z0-9]{11,97}\e$[./A-Za-z0-9]{43}" unlimited 8 256 256 "up to 512 (128+ recommended)" "6 to 11 (logarithmic)"
.Ss bcrypt
A hash based on the Blowfish block cipher,
modified to have an extra-expensive key schedule.
Originally developed by Niels Provos and David Mazieres for OpenBSD
and also supported on recent versions of FreeBSD and NetBSD,
on Solaris 10 and newer, and on several GNU/*/Linux distributions.
.hash "$2b$" "\e$2[abxy]\e$[0-9]{2}\e$[./A-Za-z0-9]{53}" 72 8 184 184 128 "4 to 31 (logarithmic)"
.Pp
The alternative prefix "$2y$" is equivalent to "$2b$".
It exists for historical reasons only.
The alternative prefixes "$2a$" and "$2x$"
provide bug-compatibility with crypt_blowfish 1.0.4 and earlier,
which incorrectly processed characters with the 8th bit set.
.Ss sha512crypt
A hash based on SHA-2 with 512-bit output,
originally developed by Ulrich Drepper for GNU libc.
Supported on Linux but not common elsewhere.
Acceptable for new hashes.
The default CPU time cost parameter is 5000,
which is too low for modern hardware.
.hash "$6$" "\e$6\e$(rounds=[1-9][0-9]+\e$)?[^$:\(rsn]{1,16}\e$[./0-9A-Za-z]{86}" unlimited 8 512 512 "6 to 96" "1000 to 999,999,999"
.Ss sha256crypt
A hash based on SHA-2 with 256-bit output,
originally developed by Ulrich Drepper for GNU libc.
Supported on Linux but not common elsewhere.
Acceptable for new hashes.
The default CPU time cost parameter is 5000,
which is too low for modern hardware.
.hash "$5$" "\e$5\e$(rounds=[1-9][0-9]+\e$)?[^$:\(rsn]{1,16}\e$[./0-9A-Za-z]{43}" unlimited 8 256 256 "6 to 96" "1000 to 999,999,999"
.Ss sha1crypt
A hash based on HMAC-SHA1.
Originally developed by Simon Gerraty for NetBSD.
Not as weak as the DES-based hashes below,
but SHA1 is so cheap on modern hardware
that it should not be used for new hashes.
.hash "$sha1" "\e$sha1\e$[1-9][0-9]+\e$[./0-9A-Za-z]{1,64}\e$[./0-9A-Za-z]{8,64}[./0-9A-Za-z]{32}" unlimited 8 160 160 "6 to 384" "4 to 4,294,967,295"
.Ss SunMD5
A hash based on the MD5 algorithm,
with additional cleverness to make precomputation difficult,
originally developed by Alec David Muffet for Solaris.
Not adopted elsewhere, to our knowledge.
Not as weak as the DES-based hashes below,
but MD5 is so cheap on modern hardware
that it should not be used for new hashes.
.hash "$md5" "\e$md5(,rounds=[1-9][0-9]+)?\e$[./0-9A-Za-z]{8}\e${1,2}[./0-9A-Za-z]{22}" unlimited 8 128 128 48 "4096 to 4,294,963,199"
.Ss md5crypt
A hash based on the MD5 algorithm, originally developed by
Poul-Henning Kamp for FreeBSD.
Supported on most free Unixes and newer versions of Solaris.
Not as weak as the DES-based hashes below,
but MD5 is so cheap on modern hardware
that it should not be used for new hashes.
CPU time cost is not adjustable.
.hash "$1$" "\e$1\e$[^$:\(rsn]{1,8}\e$[./0-9A-Za-z]{22}" unlimited 8 128 128 "6 to 48" 1000
.Ss bsdicrypt (BSDI extended DES)
A weak extension of traditional DES,
which eliminates the length limit,
increases the salt size,
and makes the time cost tunable.
It originates with BSDI
and is also available on at least NetBSD, OpenBSD, and FreeBSD
due to the use of David Burren's FreeSec library.
It is better than bigcrypt and traditional DES,
but still should not be used for new hashes.
.hash _ "_[./0-9A-Za-z]{19}" unlimited 7 56 64 24 "1 to 16,777,215 (must be odd)"
.Ss bigcrypt
A weak extension of traditional DES,
available on some System V-derived Unixes.
All it does is raise the length limit from 8 to 128 characters,
and it does this in a crude way that allows attackers to
guess chunks of a long passphrase in parallel.
It should not be used for new hashes.
.hash "" "[./0-9A-Za-z]{13,178}" 128 7 "up to 896" "up to 1024" 12 25
.Ss descrypt (Traditional DES)
The original hashing method from Unix V7, based on the DES block cipher.
Because DES is cheap on modern hardware,
because there are only 4096 possible salts and 2**56 possible hashes,
and because it truncates passphrases to 8 characters,
it is feasible to discover
.Em any
passphrase hashed with this method.
It should only be used if you absolutely have to generate hashes
that will work on an old operating system that supports nothing else.
.hash "" "[./0-9A-Za-z]{13}" 8 7 56 64 12 25
.Ss NT
The hashing method used for network authentication
in some versions of the SMB/CIFS protocol.
Available, for cross-compatibility's sake, on FreeBSD.
Based on MD4.
Has no salt or tunable cost parameter.
Like traditional DES, it is so weak that
.Em any
passphrase hashed with this method is guessable.
It should only be used if you absolutely have to generate hashes
that will work on an old operating system that supports nothing else.
.hash "$3$" "\e$3\e$\e$[0-9a-f]{32}" unlimited 8 256 256 0 1
.Sh SEE ALSO
.Xr crypt 3 ,
.Xr crypt_gensalt 3 ,
.Xr getpwent 3 ,
.Xr passwd 5 ,
.Xr shadow 5 ,
.Xr pam 8
.Rs
.%A Niels Provos
.%A David Mazieres
.%T A Future-Adaptable Password Scheme
.%B Proceedings of the 1999 USENIX Annual Technical Conference
.%D June 1999
.%U https://www.usenix.org/events/usenix99/provos.html
.Re
.Rs
.%A Robert Morris
.%A Ken Thompson
.%T Password Security: A Case History
.%J Communications of the ACM
.%V 22
.%N 11
.%D 1979
.%U http://wolfram.schneider.org/bsd/7thEdManVol2/password/password.pdf
.Re