Re: [PATCH net-next] ipv6: Check if route exists before notifying it

[Ido Schimmel <idosch@idosch.org>]

On Wed, Jun 19, 2019 at 01:10:08PM -0600, David Ahern wrote:
> On 6/19/19 11:55 AM, Ido Schimmel wrote:
> > diff --git a/net/ipv6/ip6_fib.c b/net/ipv6/ip6_fib.c
> > index 1d16a01eccf5..241a0e9a07c3 100644
> > --- a/net/ipv6/ip6_fib.c
> > +++ b/net/ipv6/ip6_fib.c
> > @@ -393,6 +393,8 @@ int call_fib6_multipath_entry_notifiers(struct net *net,
> >  		.nsiblings = nsiblings,
> >  	};
> >  
> > +	if (!rt)
> > +		return -EINVAL;
> >  	rt->fib6_table->fib_seq++;
> >  	return call_fib6_notifiers(net, event_type, &info.info);
> >  }
> 
> The call to call_fib6_multipath_entry_notifiers in
> ip6_route_multipath_add happens without rt_notif set because the MPATH
> spec is empty? 

There is a nexthop in the syzbot reproducer, but its length is shorter
than sizeof(struct rtnexthop).

> It seems like that check should be done in ip6_route_multipath_add
> rather than call_fib6_multipath_entry_notifiers with an extack saying
> the reason for the failure.

It seemed consistent with ip6_route_mpath_notify(). We can check if
rt6_nh_list is empty and send a proper error message. I'll do that
tomorrow morning since it's already late here.

> My expectation for call_fib6_multipath_entry_notifiers is any errors are
> only for offload handlers. (And we need to get extack added to that for
> relaying reasons.)

We already have extack there...