Hello world and first question concerning Munin

[Stephen Smalley]

On Fri, 2008-09-12 at 14:35 +0200, Gabriele Pohl wrote:
> Hi all,
> 
> my name is Gabriele Pohl. I live in Bonn, Germany
> and use Fedora for a few years (starting with
> Core 4 and upgraded to 9 some months ago)
> 
> I use Munin (http://munin.projects.linpro.no/)
> to monitor my computers hardware and services.
> After upgrading to Fedora 9 I decided to use SELinux
> in mode *enforce* and run into lots of problems 
> concerning SELinux and Munin-Plugins, that need 
> high system privileges to access block devices a.s.o.
> 
> I would like to solve this issues in a good
> manner and therefore subscribed to this list
> to ask the experts, how to do it.
> 
> Now my first question:
> 
> Plugin smart_ is written in Python.
> It calls "smartctl" from the smartmontools package
> (http://smartmontools.sourceforge.net/) to read the
> values of the SMART-Attributes from the harddisks.
> 
> To activate the plugin, one has to create a link
> within the service directory. 
> 
> Actually the link looks like this:
> lrwxrwxrwx  root root unconfined_u:object_r:munin_etc_t:s0 smart_sda
> -> /usr/share/munin/plugins/smart_
> 
> The plugins file looks like this:
> -rwxr-xr-x  root root
> system_u:object_r:munin_exec_t:s0 /usr/share/munin/plugins/smart_
> 
> Executable smartctl looks like this:
> -rwxr-xr-x  root root
> system_u:object_r:fsadm_exec_t:s0 /usr/sbin/smartctl
> 
> It needs access to the disks block device /dev/sda
> that looks like this:
> brw-rw----  root disk system_u:object_r:fixed_disk_device_t:s0 /dev/sda 
> 
> I have policy type targeted active and 
> policy module munin 1.4.0 installed.
> 
> I get the following raw audit messages, when 
> calling smart_sda:
> 
> host=calex.dipohl.com type=AVC msg=audit(1221221404.542:709): avc:
> denied { getattr } for pid=18327 comm="python" path="/dev/sda" dev=tmpfs
> ino=298 scontext=unconfined_u:system_r:munin_t:s0
> tcontext=system_u:object_r:fixed_disk_device_t:s0 tclass=blk_file
> 
> host=calex.dipohl.com type=SYSCALL msg=audit(1221221404.542:709):
> arch=40000003 syscall=195 success=no exit=-13 a0=8fbe278 a1=bfcdf038
> a2=3e8ff4 a3=8f481b8 items=0 ppid=18220 pid=18327 auid=500 uid=0 gid=491
> euid=0 suid=0 fsuid=0 egid=491 sgid=491 fsgid=491 tty=(none) ses=1
> comm="python" exe="/usr/bin/python"
> subj=unconfined_u:system_r:munin_t:s0 key=(null) 
> 
> As the FAQ said, I fed these messages into audit2allow:
> audit2allow -M mine < avcs
> 
> and get the following mine.te:
> -------------------------------
> module mine 1.0;
> 
> require {
> 	type munin_t;
> 	type fixed_disk_device_t;
> 	class blk_file getattr;
> }
> require {
> 	type munin_t;
> 	type fixed_disk_device_t;
> 	class blk_file getattr;
> }
> 
> #============= munin_t ==============
> allow munin_t fixed_disk_device_t:blk_file getattr;
> -------------------------------
> 
> and a mine.pp
> 
> Will it be ok to load that into the kernel using 
> semodule -i mine.pp ?
> 
> And why are there two identical *require* structs?
> Can / Should I delete one of them?
> What shall I do with the message of type "SYSCALL"
> if it were wrong to put it into the avcs-File?
> 
> Should I make adjustments to the files above
> (service-link, plugin-file)
> 
> Anything else, that you can advise?

Ideally the munin_t domain itself shouldn't need any access to the raw
device - it should transition into the existing domain for smartd
(fsdaemon_t) upon executing the smartctl program.  I don't know offhand
if the existing munin policy module has such a domain transition rule.

However, mere getattr access (i.e. the ability to stat the file) isn't a
big deal, so you could likely grant that one w/o difficulty.  What would
be more problematic is allowing read or write access to the raw device.

The duplicate require blocks look like a bug in audit2allow.

-- 
Stephen Smalley
National Security Agency