On (03/11/08 18:45), Nicolas Williams wrote:
> 
> On Tue, Mar 11, 2008 at 06:51:11PM -0400, [EMAIL PROTECTED]
> wrote:
> > Moreover, as we discussed in various hallway conversations,
> > it's a little odd to configure ipsec policy in one way (using
> > ipsecconf) and to configure md5 in another (via on/off switches). 
> 
> Except that a *lot* of IPsec configuration should be specifiable
> through
> IP_SEC_OPT-like interfaces.  (Not necessarily things like, oh, say,
> trust anchors.)

Well, it's easy enough to do this via a setsockopt, but as
Jim pointed out, it's meaningless to turn on md5 if there are no
keys. 

> No, not PF_KEY.  If you want sanitized ways to deal with keys then use
> tokens (including soft tokens).  PF_KEY is not an API that we can
> expect
> simple apps to use.

Actually, as others have pointed out, tcp-md5 itself is such
a weak security model that it's only needed in some corner cases
like bgp-daemons, and simple apps would be better off using 
full-blown ipsec. Besides, openbsd (which is one of the few
implementations that actually gets the server side semantics right)
uses PF_KEY sockets. 

OTOH, the benefit to using ipseckey/ipsecconf is that we
don't have any Interface changes to third party routing daemons-
they "just work". 

--Sowmini

_______________________________________________
networking-discuss mailing list
[email protected]

Reply via email to