On Fri, 2018-03-09 at 19:26 +0100, Guillaume Nault wrote: > On Fri, Mar 09, 2018 at 06:58:00PM +0100, Paolo Abeni wrote: > > The single threaded reproducer does not trigger anymore after 1/2, > > _but_ if ask syzbot to test 1/2 that will trigger another splat, > > because syzbot will do also multi threaded tests and we will hit the > > race between connect(tunnel->fd) and l2tp_tunnel_create(), > > > > Ok, and this case is handled by the sk_state test in l2tp_xmit_skb(), > right?
We need both such test and checking for v4mapped address in l2tp_xmit_skb() > I just want to be sure that I didn't miss anything and that patch 1/2 > combined with the socket state check in l2tp_xmit_skb() are enough to > fix the bug. And that overriding ->inet_*addr can be removed entirely > (and safely!). I tested the above in vs the repro and in some real use case, but any additinal pair of eyes are welcome! I'll send v3 soon. Cheers, Paolo