Re: [PATCH net 1/1] ovpn: avoid caching stale IPv6 dst after FIB changes

[David Ahern <dsahern@kernel.org>]

On 9/18/26 12:51 AM, Ralf Lici wrote:
> On Thu, 17 Sep 2026 13:09:10 -0600, David Ahern <dsahern@kernel.org> wrote:
>> On 9/17/26 4:09 AM, Ralf Lici wrote:
>>> diff --git a/drivers/net/ovpn/udp.c b/drivers/net/ovpn/udp.c
>>> index 7f69e8890b5b..a28974cc36e8 100644
>>> --- a/drivers/net/ovpn/udp.c
>>> +++ b/drivers/net/ovpn/udp.c
>>> @@ -250,7 +252,11 @@ static int ovpn_udp6_output(struct ovpn_peer *peer, struct ovpn_bind *bind,
>>>  		dst_cache_reset(cache);
>>>  	}
>>>  
>>> -	dst = ip6_dst_lookup_flow(sock_net(sk), sk, &fl, NULL);
>>> +	gen0 = rt_genid_ipv6(net);
>>
>> this detail about routes should not be buried here. rt_genid_ipv6 really
>> should be local to net/ipv6/route.c
>>
>>
> 
> I agree. Keeping this detail inside the IPv6 routing code is exactly why
> I sent the RFC linked from the cover letter, and I would appreciate your
> thoughts on the proposed approach:
> 
> https://lore.kernel.org/netdev/20260901122501.482920-1-ralf@mandelbit.com/
> 
> The pattern of obtaining a dst and only later sampling its validation
> cookie when publishing it into a persistent cache is widespread in both
> drivers and the networking core. It is racy for IPv6 routes with
> rt6_info::sernum == 0, because rt6_get_cookie then reads the current
> fib6_node::fn_sernum, not necessarily the value corresponding to the
> lookup which returned that dst.
> 

the patch is focused on detecting a genid change at the time the dst is
cached. Why? It can just as easily change immediately after the double
genid check, so you are not really solving the problem.

Staleness is irrelevant until it is used again (ie, genid can change
many times between dst_cache_set and next use). That is the proper time
to recheck the genid and refresh the dst as needed.