My planned work on networking stack
Bruce M Simpson
bms at spc.org
Tue Mar 2 11:38:38 PST 2004
On Tue, Mar 02, 2004 at 03:14:06PM +0100, Andre Oppermann wrote:
> > I've been fielding suggestions from individuals who feel using a multi-bit
> > trie might be more suitable for achieving higher PPS rates.
> Yes. Which one should not matter. I want to make an API for the IPv4
> routing code. Different routing implementations then can be loaded or
> changed at runtime or boot time.
This sounds like an excellent idea.
I agree that the PF_ROUTE interface is long in need of an overhaul (see
previous mail to this effect in archives). The bit mask extraction sockaddr
fandango we do to get things like netmasks in and out is nothing short of evil.
The routing code could also benefit from some style cleanup and the use
of the UMA zone allocator.
> > > would the policy-routing optioned table sort of similar to VRF's or
> > > different routing instances that could potentially be tied to userlands
> > > like Quagga?
> > That's the plan, I believe, anyway... It would be nice if Quagga could be
> > taught about how to add TCP-MD5 keys to both FreeBSD and OpenBSD SADBs.
> What is the relationship tcp-md5 --> policy-routing?
It's another one of those cross OS features which, whilst deceptively
simple on the surface, requires some hacking of the routing daemon to
grok PF_KEY messages (a well defined interface). We need a well-defined
interface which is hopefully cross OS for policy routing.
More information about the freebsd-net