Known Issues with other programs


== cryptsetup — awkward input processing ==

Some people create their crypto partition using a command like

	openssl ... | cryptsetup create ...

Without any extra arguments, input is processed as if it were
interactive, that is, everything starting from the first newline is
ignored. This is standard behavior for stdin. Other truncations to
binary characters may happen.

pam_mount's mount.crypt makes sure that libcryptsetup uses the entire
key material, including newlines, NUL bytes or other characters.
However, since you created your crypto volume with a truncated key
that is different from the real one, mounting may fail unexpectedly.


== cryptsetup — key truncation ==

cryptsetup implicitly assumes -s 256, which either pads or truncates
the key material after it has gone through cryptsetup's hashing (-h),
if any. This means that

	cryptsetup create -h sha512 ...

will hash the input with SHA-512, then truncate it down to 256 bits,
unless -s 512 was explicitly specified.

pam_mount won't do this sort of key weakening when a key file is used.
Remember that a key file is supposed to already contain the _final_ key
used for the filesystem, i.e. no extra hashing. (This is why pam_mount
also passes -h plain to cryptsetup by default.) Thus, pam_mount defaults
to using the key file's length (when decrypted) as the cipher size.


== shell — key expansion ==

Some HOWTOs suggest manual key generation for encrypted volumes, however
they fail to guard against shell semantics, such as:

	KEY=$(head -c79 /dev/urandom)

At least bash strips all \x00 bytes from the input. There might be worse
behavior. Furthermore,

	echo $KEY | openssl ...

implicitly adds a newline into the stream, which is unwanted for
key generation. Please use the pmt-ehd tool to create PLAIN-type
encrypted volumes.


== gksu & kdesu ==

gksu interprets any output on stderr as an error. pam_mount writes
debug output to stderr, so this combination will only work if debugging
is disabled in pam_mount, or gksu gets fixed.


== sshd — various ==

The "UsePAM" configuration option is required to be enabled to make
sshd go through the PAM stacks.

When "PrivilegeSeparation" is enabled in OpenSSH versions before 4.9,
ssh will not run correctly through the PAM stacks. In 4.9 and later,
this is fixed.

When public key authentication is used, the PAM auth stage is entirely
skipped. The same goes for Challenge Response Authentication.

So pam_mount would normally ask for a password in the session stage,
but in any OpenSSH to date, PAM modules do not seem to be able to ask
for a password in the session stage, "conversation" always fails:
https://bugzilla.mindrot.org/show_bug.cgi?id=926#c35
https://bugzilla.mindrot.org/show_bug.cgi?id=688

"UseLogin yes" may be used to enable pam_mount -- irrespective of
public key authentification, privilege separation or UsePAM=no. sshd
itself will not do anything useful w.r.t. pam_mount, but it will call
/bin/login which will then run through the PAM session stage, where
pam_mount can ask your for a password. Read the sshd documentation
about possible pitfalls involved using UseLogin.


== su, probably others — privilege drop ==

I sometimes get reports about unmount failing because of insufficient
privileges. Some programs and/or distributions and/or pam
configurations seem to drop the root privileges after successful
authentification. This goes counter to pam_mount which needs these
privileges for umount. (May not apply for FUSE mounts.)

Known constellations include

	* su from coreutils, on some distros
	* GDM on Ubuntu


== sudo ==

sudo has an internal bug (def_prompt is NULL) that leads to a crash
when a PAM module tries to invoke the conversation function.

Seen with at least 1.6.9p17.
Reference: http://bugs.debian.org/492333


== truecrypt ==

The scriptable interface of Truecrypt 5 and upwards is broken and
cannot be used by pam_mount.


== vsftpd — not using PAM ==

vsftpd does not run through the PAM session code, hence will never
call pam_mount's mounting functions.
It also appears to drop privileges so that there would be a
unmounting problems.


# right-margin: 72