-# $Cambridge: exim/src/src/EDITME,v 1.17 2006/10/02 13:38:18 ph10 Exp $
+# $Cambridge: exim/src/src/EDITME,v 1.18 2006/12/05 11:35:28 ph10 Exp $
##################################################
# The Exim mail transport agent #
# configuration of an authenticator for use with SMTP AUTH.) At least one
# operating system has an extended function called crypt16(), which uses up to
# 16 characters of a password (the normal crypt() uses only the first 8). Exim
-# supports the use of crypt16() as well as crypt().
+# supports the use of crypt16() as well as crypt() but note the warning below.
# You can always indicate a crypt16-encrypted password by preceding it with
# "{crypt16}". If you want the default handling (without any preceding
# an encrypted password with "{crypt}". For more details, see the description
# of the "crypteq" condition in the manual chapter on string expansions.
-# Since most operating systems do not include a crypt16() function (yet?), Exim
-# has one of its own, which it uses unless HAVE_CRYPT16 is defined. Normally,
-# that will be set in an OS-specific Makefile for the OS that have such a
-# function, so you should not need to bother with it.
+# Some operating systems do not include a crypt16() function, so Exim has one
+# of its own, which it uses unless HAVE_CRYPT16 is defined. Normally, that will
+# be set in an OS-specific Makefile for the OS that have such a function, so
+# you should not need to bother with it.
+
+# *** WARNING *** WARNING *** WARNING *** WARNING *** WARNING ***
+# It turns out that the above is not entirely accurate. As well as crypt16()
+# there is a function called bigcrypt() that some operating systems have. This
+# may or may not use the same algorithm, and both of them may be different to
+# Exim's built-in crypt16() that is used unless HAVE_CRYPT16 is defined.
+#
+# However, since there is now a move away from the traditional crypt()
+# functions towards using SHA1 and other algorithms, tidying up this area of
+# Exim is seen as very low priority. In practice, if you need to, you can
+# define DEFAULT_CRYPT to the name of any function that has the same interface
+# as the traditional crypt() function.
+# *** WARNING *** WARNING *** WARNING *** WARNING *** WARNING ***
#------------------------------------------------------------------------------
-/* $Cambridge: exim/src/src/expand.c,v 1.72 2006/11/13 12:32:58 ph10 Exp $ */
+/* $Cambridge: exim/src/src/expand.c,v 1.73 2006/12/05 11:35:28 ph10 Exp $ */
/*************************************************
* Exim - an Internet mail transport agent *
#include "exim.h"
+/* Recursively called function */
+
+static uschar *expand_string_internal(uschar *, BOOL, uschar **, BOOL);
+
#ifdef STAND_ALONE
#ifndef SUPPORT_CRYPTEQ
#define SUPPORT_CRYPTEQ
#endif
#endif
+#ifdef LOOKUP_LDAP
+#include "lookups/ldap.h"
+#endif
+
#ifdef SUPPORT_CRYPTEQ
#ifdef CRYPT_H
#include <crypt.h>
#endif
#endif
-#ifdef LOOKUP_LDAP
-#include "lookups/ldap.h"
-#endif
-
-
-
-/* Recursively called function */
+/* The handling of crypt16() is a mess. I will record below the analysis of the
+mess that was sent to me. We decided, however, to make changing this very low
+priority, because in practice people are moving away from the crypt()
+algorithms nowadays, so it doesn't seem worth it.
+
+<quote>
+There is an algorithm named "crypt16" in Ultrix and Tru64. It crypts
+the first 8 characters of the password using a 20-round version of crypt
+(standard crypt does 25 rounds). It then crypts the next 8 characters,
+or an empty block if the password is less than 9 characters, using a
+20-round version of crypt and the same salt as was used for the first
+block. Charaters after the first 16 are ignored. It always generates
+a 16-byte hash, which is expressed together with the salt as a string
+of 24 base 64 digits. Here are some links to peruse:
+
+ http://cvs.pld.org.pl/pam/pamcrypt/crypt16.c?rev=1.2
+ http://seclists.org/bugtraq/1999/Mar/0076.html
+
+There's a different algorithm named "bigcrypt" in HP-UX, Digital Unix,
+and OSF/1. This is the same as the standard crypt if given a password
+of 8 characters or less. If given more, it first does the same as crypt
+using the first 8 characters, then crypts the next 8 (the 9th to 16th)
+using as salt the first two base 64 digits from the first hash block.
+If the password is more than 16 characters then it crypts the 17th to 24th
+characters using as salt the first two base 64 digits from the second hash
+block. And so on: I've seen references to it cutting off the password at
+40 characters (5 blocks), 80 (10 blocks), or 128 (16 blocks). Some links:
+
+ http://cvs.pld.org.pl/pam/pamcrypt/bigcrypt.c?rev=1.2
+ http://seclists.org/bugtraq/1999/Mar/0109.html
+ http://h30097.www3.hp.com/docs/base_doc/DOCUMENTATION/HTML/AA-Q0R2D-
+ TET1_html/sec.c222.html#no_id_208
+
+Exim has something it calls "crypt16". It will either use a native
+crypt16 or its own implementation. A native crypt16 will presumably
+be the one that I called "crypt16" above. The internal "crypt16"
+function, however, is a two-block-maximum implementation of what I called
+"bigcrypt". The documentation matches the internal code.
+
+I suspect that whoever did the "crypt16" stuff for Exim didn't realise
+that crypt16 and bigcrypt were different things.
+
+Exim uses the LDAP-style scheme identifier "{crypt16}" to refer
+to whatever it is using under that name. This unfortunately sets a
+precedent for using "{crypt16}" to identify two incompatible algorithms
+whose output can't be distinguished. With "{crypt16}" thus rendered
+ambiguous, I suggest you deprecate it and invent two new identifiers
+for the two algorithms.
+
+Both crypt16 and bigcrypt are very poor algorithms, btw. Hashing parts
+of the password separately means they can be cracked separately, so
+the double-length hash only doubles the cracking effort instead of
+squaring it. I recommend salted SHA-1 ({SSHA}), or the Blowfish-based
+bcrypt ({CRYPT}$2a$).
+</quote>
+*/
-static uschar *expand_string_internal(uschar *, BOOL, uschar **, BOOL);
git://git.exim.org / exim.git / commitdiff