Re: [gentoo-hardened] Confused about winbindd and SELinux

[Chris PeBenito]

On Thu, 2008-03-20 at 15:37 -0400, Mike Edenfield wrote:
> Chris PeBenito wrote:
> > On Thu, 2008-03-20 at 11:17 -0400, Mike Edenfield wrote:
> >> Chris PeBenito wrote:
> >
> > Thats because, as I said in my previous response, the interface is for
> > connecting over a unix domain socket.  If we look at the macro
> > stream_connect_pattern(), it has:
> >
> > 	allow $1 $2:dir search_dir_perms;
> > 	allow $1 $3:sock_file { getattr write };
> > 	allow $1 $4:unix_stream_socket connectto;
> >
> > no mention of pipes (fifo_files).  So to fix up the policy I have to
> > make a new interface for rw on winbind_tmp_t fifo_files, and then make
> > the appropriate domains call that interface.
>
> Ok, sorry for being so confused here.  I'm obviously using the word
> "pipe" much too loosely and inaccurately.  The file in /tmp/.winbind is
> also a unix domain socket, not a named pipe (though the winbindd(8) man
> page calls it a "pipe"):
>
> # ls -lFZ /tmp/.winbindd
> srwxrwxrwx+ 1 root root system_u:object_r:winbind_tmp_t 0 Mar 18 17:12 pipe=

Ah, now things are making sense.  Its unfortunate that they name the
socket "pipe".

> The messages I'm getting look like this:
>
> type=AVC msg=audit(1205875201.198:114): avc:  denied  { getattr } for
> pid=4806 comm="cron" path="/tmp/.winbindd" dev=hda3 ino=4374539
> scontext=system_u:system_r:crond_t
> tcontext=system_u:object_r:winbind_tmp_t tclass=dir
> type=AVC msg=audit(1205875201.198:115): avc:  denied  { search } for
> pid=4806 comm="cron" name=".winbindd" dev=hda3 ino=4374539
> scontext=system_u:system_r:crond_t
> tcontext=system_u:object_r:winbind_tmp_t tclass=dir
> type=AVC msg=audit(1205875201.198:115): avc:  denied  { getattr } for
> pid=4806 comm="cron" path="/tmp/.winbindd/pipe" dev=hda3 ino=4374540
> scontext=system_u:system_r:crond_t
> tcontext=system_u:object_r:winbind_tmp_t tclass=sock_file
> type=AVC msg=audit(1205875201.198:116): avc:  denied  { write } for
> pid=4806 comm="cron" name="pipe" dev=hda3 ino=4374540
> scontext=system_u:system_r:crond_t
> tcontext=system_u:object_r:winbind_tmp_t tclass=sock_file
>
>
> >> But to actually address your email :), so far I've gotten AVC's from
> >> these domains that I think have a legitimate reason to access winbind:
> >>
> >> crond_t, newrole_t, semanage_t (for genhomedircon), sshd_t, and the
> >> various *_sudo_t domains.
> >>
> >> I also got warnings from portage_t.sandbox, because it runs tar.  I can
> >> see allow rules already in place for portage_t.sandbox -> winbind_tmp_t
> >> for objects of type file, dir, and lnk_file, but I'm seeing messages for
> >> winbind_tmp_t:sock_file as well.
> >>
> >> There was one from run_init_t, which appears to be when it runs the
> >> samba startup script, and I'm not sure why it's accessing the winbind
> >> pipe before it transitions into the samba domains.
> >
> > Are you using pam_winbind, pam_smbpass or nss_winbind on this system
> > too?
>
> Yes, I'm using both pam_winbind and nss_winbind.  pam_winbind is in my
> system-auth pam configuration, and nss_winbind is being used by nss for
> passwd, group, and shadow.

Now that things make sense, fixing up that interface that you mentioned
earlier in the thread should do it.

-- 
Chris PeBenito
<pebenito@g.o>
Developer,
Hardened Gentoo Linux
 
Public Key: http://pgp.mit.edu:11371/pks/lookup?op=get&search=0xE6AF9243
Key fingerprint = B0E6 877A 883F A57A 8E6A  CB00 BC8E E42D E6AF 9243