On Sat, Apr 2, 2016 at 7:33 PM, Martin KaFai Lau <ka...@fb.com> wrote: > One thing to note is that this patch uses the addresses from the sk > instead of iph when updating sk->sk_dst_cache. It is basically the > same logic that the __ip6_datagram_connect() is doing, so some > refactoring works in the first two patches. > > AFAIK, a UDP socket can become connected after sending out some > datagrams in un-connected state. or It can be connected > multiple times to different destinations. I did some quick > tests but I could be wrong. > > I am thinking if there could be a chance that the skb->data, which > has the original outgoing iph, is not related to the current > connected address. If it is possible, we have to specifically > use the addresses in the sk instead of skb->data (i.e. iph) when > updating the sk->sk_dst_cache. > > If we need to use the sk addresses (and other info) to find out a > new dst for a connected udp socket, it is better not doing it while > the userland is connecting to somewhere else. > > If the above case is impossible, we can keep using the info from iph to > do the dst update for a connected-udp sk without taking the lock.
I see your point, but calling __ip6_datagram_connect() seems overkill here, we don't need to update so many things in the pmtu update context, at least IPv4 doesn't do that either. I don't think you have to do that. So why just updating the dst cache (also some addr cache) here is not enough?
