summaryrefslogtreecommitdiff
path: root/README
diff options
context:
space:
mode:
Diffstat (limited to 'README')
-rw-r--r--README18
1 files changed, 17 insertions, 1 deletions
diff --git a/README b/README
index 32b005d..bf9eacd 100644
--- a/README
+++ b/README
@@ -25,7 +25,7 @@ options
* debug: produce a bit of debug output
* nodelay: don't tell the PAM stack to cause a delay on auth failure
* flock: use a shared (read) advisory lock on pwdfile, you should better move new versions into place instead
-* legacy_crypt: turns on bigcrypt and "broken md5_crypt", you will only need that if you use password hashes from another system that uses those algorithms
+* legacy_crypt: see section LEGACY CRYPT
PASSWORD FILE
@@ -36,3 +36,19 @@ First field contains the username, the second the crypt()ed password.
Other fields are optional.
crypt()ed passwords in various formats can be generated with mkpasswd from the whois package.
+
+
+LEGACY CRYPT
+============
+
+There are two crypt types that are disabled by default: bigcrypt and broken md5_crypt.
+They are disabled because they use static buffers which is bad when doing PAM authentication using this module in a multithreaded server.
+All the other crypt types are checked via the systems crypt_r function if available, else with the normal crypt function and the same static-buffer-problem.
+
+bigcrypt was used on DEC systems to allow for longer passwords.
+You can check if your passwd file contains any of these with `cut -d: -f2 passwd-file | egrep '^[^$].{13}'`.
+
+Broken md5_crypt is a speciality of big-endian systems.
+An early implementation of md5_crypt got the byte order wrong here and produced different crypt outputs.
+You might have some of these crypt hashes in your passwd file only if you created them on a big-endian system.
+If an md5_crypt hash also worked on a little-endian system (up to and including libpam-pwdfile 0.99) it isn't broken md5_crypt.