netdev - Re: [PATCH] tcp: don't use timestamp from repaired skb-s to calculate RTT

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <1407849283.10122.70.camel@edumazet-glaptop2.roam.corp.google.com>
Date:	Tue, 12 Aug 2014 06:14:43 -0700
From:	Eric Dumazet <eric.dumazet@...il.com>
To:	Andrew Vagin <avagin@...allels.com>
Cc:	Andrey Vagin <avagin@...nvz.org>, netdev@...r.kernel.org,
	linux-kernel@...r.kernel.org, Eric Dumazet <edumazet@...gle.com>,
	Pavel Emelyanov <xemul@...allels.com>,
	"David S. Miller" <davem@...emloft.net>
Subject: Re: [PATCH] tcp: don't use timestamp from repaired skb-s to
 calculate RTT

On Tue, 2014-08-12 at 16:33 +0400, Andrew Vagin wrote:
> On Tue, Aug 12, 2014 at 05:15:01AM -0700, Eric Dumazet wrote:
> > On Tue, 2014-08-12 at 13:45 +0400, Andrey Vagin wrote:
> > > We don't know right timestamp for repaired skb-s. Wrong RTT estimations
> > > isn't good, because some congestion modules heavily depends on it.
> > > 
> > > This patch adds the TCPCB_REPAIRED flag, which is included in
> > > TCPCB_RETRANS.
> > 
> > ...
> > 
> > > +
> > > +				/* All packets are restored as if they have
> > > +				 * already been sent. skb_mstamp isn't set to
> > > +				 * avoid wrong rtt estimation.
> > > +				 */
> > > +				if (tp->repair) {
> > > +					TCP_SKB_CB(skb)->sacked |= TCPCB_REPAIRED;
> > > +					TCP_SKB_CB(skb)->when = tcp_time_stamp;
> > > +				}
> > >  			}
> > >  
> > >  			/* Try to append data to the end of skb. */
> > 
> > 
> > Are you sure TCP_SKB_CB(skb)->when needs to be set ?
> 
> It's used in tcp_rearm_rto() for calculating a retransmit timeout.
> ...
> 	const u32 rto_time_stamp = TCP_SKB_CB(skb)->when + rto;
> 	s32 delta = (s32)(rto_time_stamp - tcp_time_stamp);
> ...
> 
> "when" is used as a start point, so I think it's acceptable here.
> 
> I will add a comment. Thanks.

tcp_rearm_rto() does the following :


        if (!tp->packets_out) {
                inet_csk_clear_xmit_timer(sk, ICSK_TIME_RETRANS);
        } else {
                u32 rto = inet_csk(sk)->icsk_rto;
                /* Offset the time elapsed after installing regular RTO */
                if (icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
                    icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
                        struct sk_buff *skb = tcp_write_queue_head(sk);
                        const u32 rto_time_stamp = TCP_SKB_CB(skb)->when + rto;

This means that : at least one packet was transmitted (packets_out is not 0)

Since we timestamp all packets we transmit (look at tcp_transmit_skb() callers, all doing :

TCP_SKB_CB(skb)->when = tcp_time_stamp; 

Then, write queue head was timestamped properly at the time packet was sent,
not at the time tcp repair code reinjected skbs into the write queue.




--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html