[redhat-lspp] labeled ipsec status

Eric Paris eparis at redhat.com
Mon Jan 8 20:59:57 UTC 2007


On Mon, 2007-01-08 at 15:45 -0500, Paul Moore wrote:
> On Monday, January 8 2007 3:31 pm, Eric Paris wrote:
> > > 3. Toggle to accept or reject unlabeled packets.
> > > Dan has completed this. He added a boolean, allow_unlabeled_packets,
> > > to selinux policy. Currently, because of a problem in lspp60
> > > kernel, boolean does not work. I tested the boolean on
> > > upstream kernel from kernel.org, 2.6.20-rc3-git4 and the boolean
> > > worked great and as expected. (See #5 below as to why
> > > it did not work in lspp60.)
> >
> > can paul make sure this works for NetLabel as well (since 5 shouldn't be
> > applicable to NetLabel)?
> 
> I'll verify that this still works on lspp.60 but I have no reason to believe 
> it wouldn't.  The way NetLabel allows/denies non-NetLabel packets is 
> different from IPsec.
> 
> When SELinux receives a packet it queries NetLabel to see if there are any 
> NetLabel related security attributes attached to the packet; there are three 
> possible results from this query:
> 
> 1. security attributes are present - query function returns success populates 
> a structure with the NetLabel security attributes
> 
> 2. security attributes are not present and the unlabeled flag is set to 
> allow - query function returns success and the security attribute structure 
> is cleared
> 
> 3. security attributes are not present and the unlabeled flag is set to deny - 
> query function returns failure
> 
> We can go into all the pros/cons of such an approach vs a granular policy 
> approach if you would like but when I lost the argument to use 
> SECINITSID_NETMSG as the default NetLabel packet type we lost the ability to 
> distinguish between NetLabel'd and non-NetLabel'd packets in SELinux policy 
> (NetLabel uses SECINITSID_UNLABELED/unlabeled_t for incoming traffic).

I guess my real question is how do we say "reject all unlabeled packets"
on a NetLabel system if the boolean in question is not enforcing such
policy with respect to netlabel.  Or am I as usual just lost?




More information about the redhat-lspp mailing list