Re: [PATCH net] ipv4/route: do not check saddr dev if iif is LOOPBACK_IFINDEX

[Stefano Brivio <sbrivio@redhat.com>]

On Sun, 11 Aug 2019 20:49:18 -0700 (PDT)
David Miller <davem@davemloft.net> wrote:

> From: David Ahern <dsahern@gmail.com>
> Date: Thu, 1 Aug 2019 22:16:00 -0600
> 
> > On 8/1/19 10:13 PM, Hangbin Liu wrote:  
> >> On Thu, Aug 01, 2019 at 01:51:25PM -0600, David Ahern wrote:  
> >>> On 8/1/19 2:29 AM, Hangbin Liu wrote:  
> >>>> Jianlin reported a bug that for IPv4, ip route get from src_addr would fail
> >>>> if src_addr is not an address on local system.
> >>>>
> >>>> \# ip route get 1.1.1.1 from 2.2.2.2
> >>>> RTNETLINK answers: Invalid argument  
> >>>
> >>> so this is a forwarding lookup in which case iif should be set. Based on  
> >> 
> >> with out setting iif in userspace, the kernel set iif to lo by default.  
> > 
> > right, it presumes locally generated traffic.  
> >>   
> >>> the above 'route get' inet_rtm_getroute is doing a lookup as if it is
> >>> locally generated traffic.  
> >> 
> >> yeah... but what about the IPv6 part. That cause a different behavior in
> >> userspace.  
> > 
> > just one of many, many annoying differences between v4 and v6. We could
> > try to catalog it.  
> 
> I think we just have to accept this difference because this change would
> change behavior for all route lookups, not just those done by ip route get.

How so, actually? I don't see how that would happen. On the forwarding
path, 'iif' is set (not to loopback interface), so that's not affected.

Is there any other route lookup possibility I'm missing?

-- 
Stefano