On Mon, 2017-05-01 at 16:56 -0700, Stephen Hemminger wrote:
> On Mon, 01 May 2017 15:29:48 -0700
> Eric Dumazet <eric.duma...@gmail.com> wrote:
>
> > From: Eric Dumazet <eduma...@google.com>
> >
> > Be careful when comparing tcp_time_stamp to some u32 quantity,
> > otherwise result can be surprising.
> >
> > Fixes: 7c106d7e782b ("[TCP]: TCP Low Priority congestion control")
> > Signed-off-by: Eric Dumazet <eduma...@google.com>
> > ---
> > net/ipv4/tcp_lp.c | 6 ++++--
> > 1 file changed, 4 insertions(+), 2 deletions(-)
> >
> > diff --git a/net/ipv4/tcp_lp.c b/net/ipv4/tcp_lp.c
> > index
> > 046fd3910873306d74207615d6997e1c847ea361..d6fb6c067af4641f232b94e7c590c212648e8173
> > 100644
> > --- a/net/ipv4/tcp_lp.c
> > +++ b/net/ipv4/tcp_lp.c
> > @@ -264,13 +264,15 @@ static void tcp_lp_pkts_acked(struct sock *sk, const
> > struct ack_sample *sample)
> > {
> > struct tcp_sock *tp = tcp_sk(sk);
> > struct lp *lp = inet_csk_ca(sk);
> > + u32 delta;
> >
> > if (sample->rtt_us > 0)
> > tcp_lp_rtt_sample(sk, sample->rtt_us);
> >
> > /* calc inference */
> > - if (tcp_time_stamp > tp->rx_opt.rcv_tsecr)
> > - lp->inference = 3 * (tcp_time_stamp - tp->rx_opt.rcv_tsecr);
> > + delta = tcp_time_stamp - tp->rx_opt.rcv_tsecr;
> > + if ((s32)delta > 0)
> > + lp->inference = 3 * delta;
>
> Agreed time wraparound would cause problems.
> But why not use existing time_after() macro here?
>
Simply to not perform (tcp_time_stamp - tp->rx_opt.rcv_tsecr) twice.
jiffies being volatile, this can not be optimized by the compiler.
I have a patch series (for linux-4.13) that will switch TCP stack to 1ms
TS options, regardless of CONFIG_HZ value, and when cooking it I found
this bug.
