Re: [PATCH] tcp: fixing TLP's FIN recovery

[Eric Dumazet <eric.dumazet@gmail.com>]

On Sun, 2014-06-08 at 09:41 +0200, Per Hurtig wrote:
> 
> On sön  8 jun 2014 04:58:25, Eric Dumazet wrote:
> > On Sat, 2014-06-07 at 16:34 +0200, Per Hurtig wrote:
> >> Fix to a problem observed when losing a FIN segment that does not
> >> contain data.  In such situations, TLP is unable to recover from
> >> *any* tail loss and instead adds at least PTO ms to the
> >> retransmission process, i.e., RTO = RTO + PTO.
> >>
> >> Signed-off-by: Per Hurtig <per.hurtig@kau.se>
> >> ---
> >>   net/ipv4/tcp_output.c | 6 ++++--
> >>   1 file changed, 4 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
> >> index d463c35..6573765 100644
> >> --- a/net/ipv4/tcp_output.c
> >> +++ b/net/ipv4/tcp_output.c
> >> @@ -2130,8 +2130,10 @@ void tcp_send_loss_probe(struct sock *sk)
> >>   	if (WARN_ON(!skb || !tcp_skb_pcount(skb)))
> >>   		goto rearm_timer;
> >>
> >> -	/* Probe with zero data doesn't trigger fast recovery. */
> >> -	if (skb->len > 0)
> >> +	/* Probe with zero data doesn't trigger fast recovery, if FIN
> >> +	 * flag is not set.
> >> +	 */
> >> +	if ((skb->len > 0) || (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
> >>   		err = __tcp_retransmit_skb(sk, skb);
> >>
> >>   	/* Record snd_nxt for loss detection. */
> >
> >
> > You know, I believe the test was exactly to avoid sending data less FIN
> > packets.
> >
> > If you write :
> >
> >      if (A  || !A)
> >
> > Better remove the condition, completely ;)
> >
> Obviously, but I don't think that FINs are the only segments
> who are targeted by this condition (or targeted at all given
> the implications of this statement). Furthermore, the comment above
> the if statement would probably have mentioned FINs explicity
> and not zero sized segments in general if this were the case.
> 


I see no other possibilities than FIN segments here, or the WARN_ON(!
tcp_skb_pcount(skb)) right before would trigger.

If we believe it could trigger, then we need to remove the WARN_ON(),
because its far more disruptive than waiting a bit more for the RTO.
Remember : RTO is conservative. 

The if (skb->len > 0) only is true for FIN with no data.

This was exactly the intent : Not sending FIN at this stage.

If pure FIN is OK here, just remove the comment and test, this is so
confusing and useless.