linux-kernel - Re: [PATCH] net: tcp: Fix a PTO timing granularity issue

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]

Date:	Wed, 27 May 2015 10:40:29 -0400
From:	Ido Yariv <ido@...ery.com>
To:	Eric Dumazet <eric.dumazet@...il.com>
Cc:	David Laight <David.Laight@...LAB.COM>,
	"David S. Miller" <davem@...emloft.net>,
	Alexey Kuznetsov <kuznet@....inr.ac.ru>,
	James Morris <jmorris@...ei.org>,
	Hideaki YOSHIFUJI <yoshfuji@...ux-ipv6.org>,
	Patrick McHardy <kaber@...sh.net>,
	Nandita Dukkipati <nanditad@...gle.com>,
	"netdev@...r.kernel.org" <netdev@...r.kernel.org>,
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
	Ido Yariv <idox.yariv@...el.com>
Subject: Re: [PATCH] net: tcp: Fix a PTO timing granularity issue

Hi Eric,

On Wed, May 27, 2015 at 06:41:17AM -0700, Eric Dumazet wrote:
> On Wed, 2015-05-27 at 11:36 +0000, David Laight wrote:
> > From: Of Ido Yariv
> > > Sent: 26 May 2015 21:17
> > > The Tail Loss Probe RFC specifies that the PTO value should be set to
> > > max(2 * SRTT, 10ms), where SRTT is the smoothed round-trip time.
> > > 
> > > The PTO value is converted to jiffies, so the timer may expire
> > > prematurely.
> > > 
> > > This is especially problematic on systems in which HZ <= 100, so work
> > > around this by setting the timeout to at least 2 jiffies on such
> > > systems.
> > > 
> > > The 10ms figure was originally selected based on tests performed with
> > > the current implementation and HZ = 1000. Thus, leave the behavior on
> > > systems with HZ > 100 unchanged.
> > > 
> > > Signed-off-by: Ido Yariv <idox.yariv@...el.com>
> > > ---
> > >  net/ipv4/tcp_output.c | 3 +++
> > >  1 file changed, 3 insertions(+)
> > > 
> > > diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
> > > index 534e5fd..5321df8 100644
> > > --- a/net/ipv4/tcp_output.c
> > > +++ b/net/ipv4/tcp_output.c
> > > @@ -2208,6 +2208,9 @@ bool tcp_schedule_loss_probe(struct sock *sk)
> > >  		timeout = max_t(u32, timeout,
> > >  				(rtt + (rtt >> 1) + TCP_DELACK_MAX));
> > >  	timeout = max_t(u32, timeout, msecs_to_jiffies(10));
> > > +#if HZ <= 100
> > > +	timeout = max_t(u32, timeout, 2);
> > > +#endif
> > 
> > Why not:
> > 	timeout = max_t(u32, timeout, max_t(u32, 2, msecs_to_jiffies(10)));
> > I think the RH max_t() is a compile time constant.
> > 
> > You need 2 jiffies to guarantee a non-zero timeout.
> > Even if HZ=199 with a 'rounding down' msecs_to_jiffies() you get 1 jiffy
> > and a possible immediate timeout.
> > 
> 
> Have you followed previous discussions ?
> 
> I guess we can have a helper macro, but for the moment only one spot was
> found.
> 
> Its kind of depressing having to deal with HZ=100 issues, with modern
> NO_HZ configurations.
> 
> TCP rtts have now usec resolution, so HZ=100 is pushing TCP to very
> imprecise behavior.

HZ=100 is used on some embedded platforms, so it's still something we
have to deal with unfortunately..

Since the '2' here is a lower bound, and msecs_to_jiffies(10) will
return values greater than 2 for HZ>100 anyway, always ensuring the
2 jiffies lower bound shouldn't impact the behavior when HZ=1000.

However, as far as I can tell, comparing msecs_to_jiffies(10) to 2, or
comparing the whole timeout to 2 doesn't make much difference, since
msecs_to_jiffies isn't inlined.

In other words, keeping the #if shouldn't make much difference in behavior,
but will save the small comparison.

Cheers,
Ido.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/