ejabberd and name_bind

[Randy Barlow]

Greetings!

The ejabberd Fedora package has its own SELinux policy module that it
ships[0]. A user has reported an issue with an SELinux denial with the
default ejabberd config[1].

I spent some time trying to modify the policy to allow the name_bind on
the port, but it seems that my attempts result in it still being
denied:

allow ejabberd_t unreserved_port_t:udp_socket name_bind;

As I commented on the ticket, I also found that setting the nis_enabled
bool on my system to true would solve the problem.

However, I think it would be ideal if I could adjust the ejabberd
module to do this on the users' behalf, as it is not obvious to the
average user (or even to me) that this boolean could be the solution to
the problem.

Is there something I could adjust in the ejabberd policy that would
resolve this issue? Thanks.


[0]
https://src.fedoraproject.org/rpms/ejabberd/blob/rawhide/f/ejabberd.te
[1] https://bugzilla.redhat.com/show_bug.cgi?id=1901466

[Casper]

Randy Barlow a écrit :
>  Greetings!
>  
>  The ejabberd Fedora package has its own SELinux policy module that it
>  ships[0]. A user has reported an issue with an SELinux denial with the
>  default ejabberd config[1].
>  
>  I spent some time trying to modify the policy to allow the name_bind on
>  the port, but it seems that my attempts result in it still being
>  denied:
>  
>  allow ejabberd_t unreserved_port_t:udp_socket name_bind;
>   
Hi Randy,

Thank you so much for your work! I'm spending time every year to fix
AVCs for ejabberd (on my systems) without going deep in this
issue. But I stored all .te files, so I'm happy to be able to compare
with your .te file :)



File: ejabberd-udp-unreserved_port-fedora-33.te

"""
module ejabberd-udp-unreserved_port-fedora-33 1.0;

require {
	type unreserved_port_t;
	type ejabberd_t;
	class udp_socket name_bind;
}

#============= ejabberd_t ==============

#!!!! This avc can be allowed using the boolean 'nis_enabled'
allow ejabberd_t unreserved_port_t:udp_socket name_bind;
"""

>  As I commented on the ticket, I also found that setting the
> nis_enabled
>  bool on my system to true would solve the problem.
>   
How did you do that... I mean, you have found the Graal...

>  However, I think it would be ideal if I could adjust the ejabberd
>  module to do this on the users' behalf, as it is not obvious to the
>  average user (or even to me) that this boolean could be the solution to
>  the problem.
>   
The Graal, I said :)

>  Is there something I could adjust in the ejabberd policy that would
>  resolve this issue? Thanks.
>   
On my side, I will make a fresh install on fresh box to see what is
exactly required or not, then compare, then send you PR :)

I also want to see what is required with the default ejabberd config
and with my "advanced" config file.


Best regards,
Casper
-- 
GnuPG: AE157E0B29F0BEF2 at keys.openpgp.org
CA Cert: https://dl.casperlefantom.net/pub/ssl/root.der
Jabber/XMPP Messaging: casper(a)casperlefantom.net

[Randy Barlow]

>   
Hey Casper!

I ended up finding out what my problem was. The policy I had mentioned
trying in my first post ended up being correct after all. It turned out
that I had a rogue ejabberd policy in the system in addition to the
policy I was testing, and the rogue policy had higher priority than the
one provided by the RPM ☹

I must have installed this policy by hand some time in the past and
forgotten to remove it. After removing it, I found that I just needed
one more line on the policy to address this issue:

https://src.fedoraproject.org/rpms/ejabberd/c/d48066216cac7e2ec1626d65376...

With this policy I am able to run ejabberd with STUN/TURN enabled and
with the nis_enabled bool disabled.

There are updates for F33-36 in Bodhi now:

https://bodhi.fedoraproject.org/updates/?packages=ejabberd