On Wed, Jul 29, 2015 at 2:54 AM, Thomas Graf <tg...@suug.ch> wrote: > On 07/29/15 at 11:29am, Eric Dumazet wrote: >> On Wed, 2015-07-29 at 11:13 +0200, Thomas Graf wrote: >> > On 07/28/15 at 04:02pm, Tom Herbert wrote: >> > > This patch creates sk_set_txhash and eliminates protocol specific >> > > inet_set_txhash and ip6_set_txhash. sk_set_txhash simply sets a >> > > random number instead of performing flow dissection. sk_set_txash >> > > is also allowed to be called multiple times for the same socket, >> > > we'll need this when redoing the hash for negative routing advice. >> > > >> > > Signed-off-by: Tom Herbert <t...@herbertland.com> >> > >> > Doesn't this break TX hashing with SO_REUSEPORT? >> >> >> AFAIK nothing uses sk_txhash yet. > > skb_set_hash_from_sk() > skb_get_hash() > > Am I misreading this? I'm not using SO_REUSEPORT and it might be OK > to assume that different sockets may go to different queues even if > the L4 tuple is identical.
Hi Thomas, The salient property of both sk_txhash and skb->hash is that they provide a uniform distribution over flows. It is incorrect to assume that either of these immutable during the lifetime of a flow, so yes this means that packets of a flow may go to different receive queues when hashes change. SO_REUSEPORT is a process in the receive path but uses ehashfn over the ports. But even with SO_REUSEPORT we provide no guarantee that packets of a "flow" will always hit the same socket, the hashing is not consistent when new reuseport sockets are added or removed-- this is actually a long standing issue with SO_REUSEPORT in the TCP case since it is possible to orphan connections in SYN-RECV. I believe Eric was working toward fixing that, so maybe in the future we can use skb->hash if it is a savings. Thanks, Tom -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
