From: Eric Dumazet <eric.duma...@gmail.com> Date: Mon, 14 Apr 2014 10:40:04 -0700
> On Mon, 2014-04-14 at 13:34 -0400, David Miller wrote: >> From: Eric Dumazet <eric.duma...@gmail.com> >> Date: Mon, 14 Apr 2014 10:12:29 -0700 >> >> > Hmm, it seems commit 31c70d5956fc l2tp: keep original skb ownership >> > is the problem. >> > >> > ip_queue_xmit() assumes the socket attached to skb is an inet socket. >> >> This is similar to the "send over AF_PACKET" issue we were discussing >> the other week. >> >> It seems we need a real resolution to this issue. >> >> To recap: >> >> 1) We want to charge memory to the top-level socket. >> >> 2) However during encapsulations etc. we can end up in IP stack >> which expects only IP sockets to be attached to skb->sk >> >> I suspect that in the short term we may have to bit the bullet and >> compromise #2, and do flow control via the tunnel's socket. > > Or add a sk parameter to ip_queue_xmit(), to break the assumption "sk = > skb->sk", which happened to be generally true, but is ambiguous for IP > tunnels. Yep that would handle the l2tp cases, but vxlan needs something different since it goes through iptunnel_xmit(). So perhaps as a quick fix we can add an 'sk' arg to both ip_queue_xmit() and ip_local_out(). -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
