[mpls-linux-general] Re: MPLS + VRF

[James R. L. <jl...@mi...>]

On Mon, Oct 04, 2004 at 09:54:54AM +0000, Massimo Magnani wrote:
> Hi James, I am Massimo Magnani from Italy, and I am testing linux MPLS 
> and VRF.
> 
> First of all, sorry if I mail you directly.

That is ok.  I would prefer you contact me via the mailing list, but that
is just so that others can see what you're working on.
(I've CC'ed the lists)

I plan on creating a development tree which will contain the latest MPLS, VRF,
and L2CC code.  I will probably be releasing this as a seperate RPM and
patch.

I the mean time, change mpls4_nexthop_resolve() to look like this:

static int mpls4_nexthop_resolve(struct neighbour **np, struct sockaddr *sock_addr, int oif)
{
        struct sockaddr_in *addr = (struct sockaddr_in *) sock_addr;
        u32 saddr = addr->sin_addr.s_addr;
        struct flowi fl = { .oif = oif,
                            .nl_u = { .ip4_u = { .daddr = saddr } } };
        struct rtable *rp;
	unsigned char vrf = 0;
        int err;

        if (addr->sin_family != AF_INET)
                return -EINVAL;

	if ((dev = dev_get_by_index(oif)))
		vrf = dev->vrf;

        if (inet_addr_type(vrf, saddr) != RTN_UNICAST)
                return -EINVAL;

	fl.vrf = vrf;
        err = __ip_route_output_key(&rp, &fl);
        if (err)
                return err;

        /* If routing cache stops to automatically bind output routes
         * to ARP cache, we will have to make arp_bind_neighbour() ourselves.
         */
        if (!rp->u.dst.neighbour)
                BUG();

        *np = neigh_clone(rp->u.dst.neighbour);
        ip_rt_put(rp);

        return err;
}

Even with the above hack, you will lack the ability to map traffic from an
LSP into a VRF.  For that you will need to implement an 'instruction'.  Look
in linux/net/mpls/mpls_opcode.c for examples.  The new instruction will
need to set the 'vrf' field on a skb.

> I patched my kernel 2.6.8.1 with your mpls patch and everything has been 
> working fine. Then I read that you also have written some code to 
> implement VRF on linux, so I wanted to test this functionality too.
> 
> I patched my kernel again, and I had a compilation problem with mpls4.c, 
> maybe because in include/net/route.h the struct inet_addr_type now has 
> also a vrf 'field' (sorry but I don't know almost anything about C 
> programming...)
> 
> I modified the mpls4.c file - I added a 'unsigned char vrf' in line 289 
> 'static int mpls4_nexthop_resolve(struct neighbour **np, struct sockaddr 
> *sock_addr, int oif, unsigned char vrf), and in line 303 if 
> (inet_add_type(vrf, saddr) != RTN_UNICAST)  - and now I can compile the 
> kernel again.
> 
> I don't know if these two modifications are correct, and moreover, I 
> think that I could fill the vrf variable with some value, but I don't 
> know where to take that value!
> 
> I mean I think I should have a vrf = something, but I don't know what 
> 'something' is!
> 
> Thank you in advance, sorry to have bothered you.
> 
> Best regards.
> 
> Massimo Magnani

-- 
James R. Leu
jl...@mi...