ssh using gssapi will enforce FILE: credentials cache

[Toby Blake]

Hi,

ssh connections from a client with the following in ssh_config...

GSSAPIAuthentication yes
GSSAPIDelegateCredentials yes

... to an ubuntu 20.04 machine result in KRB5CCNAME being set to 'FILE:/tmp/krb5cc_[uid]_[random]' despite the following in /etc/krb5.conf:

[libdefaults]
 ...
 default_ccache_name = KEYRING:persistent:%{uid}

This means that we cannot enforce a policy to use KEYRING ccaches across our systems. Authentications which go via the pam stack (e.g. login to the machine at the console or over ssh using a password) can be configured to use a KEYRING ccache, via libpam-krb5 settings in /etc/krb5.conf.

The FILE: setting seems to be hard-coded in the openssh code (auth-krb5.c). It would be great if ssh(gssapi-with-mic) connections either (a) set KRB5CCNAME to the default_ccache_name value, if set in /etc/krb5.conf, or (b) didn't set KRB5CCNAME at all, so the system default is used.

Many thanks
Toby Blake
School of Informatics
University of Edinburgh

[Christian Ehrhardt  (paelzer)]

Hi Toby,

It seems that is an ongoing topic for years, I've found this discussed from the KRB POV [1] and on openssh [2]. Especially following [1] it seems things aren't too easy but there are a few workarounds/hints that might or might not help your use case.

In general having this configurable instead of hard-coded in ssh sounds right to me, but would then be an upstream feature request that you could report at [3]. If you happen to do so it would be awesome to report the ID back here so that we can link the bugs and track what upstream thinks/says about it.

One thing thou - you write explicitly "to a 20.04 machine" is that behavior in any way a regression to the former versions?

[1]: http://kerberos.996246.n3.nabble.com/KRB5CCNAME-and-sshd-td13395.html
[2]: https://lists.mindrot.org/pipermail/openssh-unix-dev/2014-December/033217.html
[3]: https://bugzilla.mindrot.org/show_bug.cgi

[Toby Blake (toby-inf)]

Hi Christian,

Thanks for your reply (and for the links). I agree that a configurable option coming from upstream would be the preferred option. I'll submit a bug upstream accordingly and note the ID here.

I note there's a patch referenced in the openssh mailing list link, unfortunately the link is a 404 - I'll see if I can track it down via other means. On this topic, we've been using ssh on scientific linux for a number of years now and that is patched to use the default ccache name. It would be interesting to see if it's the same or similar as the 404 patch above.

Depending on the outcome of my upstream bug report, I'll see if I can port any patches we use to the current ubuntu/upstream code base.

As to your final question re: regression - I don't think so, looking at the code, but we are only now in the process of converting desktop machines to ubuntu.

Cheers
Toby

[Christian Ehrhardt  (paelzer)]

Due to that hint with SciLinux I have fetched
http://ftp.scientificlinux.org/linux/scientific/7.8/SRPMS/vendor/openssh-7.4p1-21.el7.src.rpm

I can't see it but that is https://bugzilla.redhat.com/show_bug.cgi?id=991186
I can see follow on issues referring to it https://bugzilla.redhat.com/show_bug.cgi?id=1161073 thou.

# use default_ccache_name from /etc/krb5.conf (#991186)
Patch902: openssh-6.3p1-krb5-use-default_ccache_name.patch

The patch is not/no-more in https://git.centos.org/rpms/openssh/blob/c8/f/SOURCES
But in https://git.centos.org/rpms/openssh/blob/c7/f/SOURCES/openssh-6.3p1-krb5-use-default_ccache_name.patch

That is suspicious, there must be a reason it is gone right?
Maybe it was hard to maintain or somewhat bad and they only could drop it on the major version change.

In the v8 spec I find:

# Improve ccache handling in openssh (#991186, #1199363, #1566494)
# https://bugzilla.mindrot.org/show_bug.cgi?id=2775
Patch804: openssh-7.7p1-gssapi-new-unique.patch
# Respect k5login_directory option in krk5.conf (#1328243)
Patch805: openssh-7.2p2-k5login_directory.patch

FYI navigate from here https://git.centos.org/rpms/openssh/blob/c8/f/SPECS/openssh.spec

I'm not sure on this, but maybe carrying more of the RH sauce back to upstream might help.
Definitely activity on the new or old upstream bug will be needed.

Keep us in the loop here what happens.

[Toby Blake (toby-inf)]

Hi Christian,

Again, thanks for the above.

https://bugzilla.mindrot.org/show_bug.cgi?id=2775, in particular, looks interesting, as it seems to be an attempt to bring the relevant ccache patches up to date for version 8. e.g. we have been patching our SL systems additionally for https://bugzilla.redhat.com/show_bug.cgi?id=1199363.

I'll give this a try and report back. I'll hold off on reporting this as a bug upstream until I've tried the patch(es).

Cheers
Toby

[Toby Blake (toby-inf)]

Hi there,

I'm afraid I haven't had much time to look properly into this recently, but it remains on my list.

In the meantime, I've submitted an enhancement request upstream:

https://bugzilla.mindrot.org/show_bug.cgi?id=3203

Cheers
Toby

[Andreas Hasenack (ahasenack)]

Toby, you are mostly interested in this because you have some sort of policy, perhaps one that doesn't allow secrets to be stored on disk in clear text and protected just by filesystem permissions?