[PATCH net] tcp: Fix integer-overflows in TCP vegas

[PATCH net] tcp: Fix integer-overflows in TCP vegas

[Christoph Paasch]

In vegas we do a multiplication of the cwnd and the rtt. This
may overflow and thus their result is stored in a u64. However, we first
need to cast the cwnd so that actually 64-bit arithmetic is done.

Cc: Doug Leith <doug.leith@nuim.ie>
Fixes: 8d3a564da34e (tcp: tcp_vegas cong avoid fix)
Signed-off-by: Christoph Paasch <christoph.paasch@uclouvain.be>
---
 net/ipv4/tcp_vegas.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/ipv4/tcp_vegas.c b/net/ipv4/tcp_vegas.c
index 9a5e05f27f4f..6a4bdea2a0fb 100644
--- a/net/ipv4/tcp_vegas.c
+++ b/net/ipv4/tcp_vegas.c
@@ -218,7 +218,7 @@ static void tcp_vegas_cong_avoid(struct sock *sk, u32 ack, u32 acked)
 			 * This is:
 			 *     (actual rate in segments) * baseRTT
 			 */
-			target_cwnd = tp->snd_cwnd * vegas->baseRTT / rtt;
+			target_cwnd = (u64)tp->snd_cwnd * vegas->baseRTT / rtt;
 
 			/* Calculate the difference between the window we had,
 			 * and the window we would like to have. This quantity
-- 
1.9.3

Re: [PATCH net] tcp: Fix integer-overflows in TCP vegas

[Neal Cardwell]

On Tue, Jul 22, 2014 at 5:36 PM, Christoph Paasch
<christoph.paasch@uclouvain.be> wrote:
> In vegas we do a multiplication of the cwnd and the rtt. This
> may overflow and thus their result is stored in a u64. However, we first
> need to cast the cwnd so that actually 64-bit arithmetic is done.
>
> Cc: Doug Leith <doug.leith@nuim.ie>
> Fixes: 8d3a564da34e (tcp: tcp_vegas cong avoid fix)
> Signed-off-by: Christoph Paasch <christoph.paasch@uclouvain.be>
> ---
>  net/ipv4/tcp_vegas.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)

Acked-by: Neal Cardwell <ncardwell@google.com>

Thanks!

neal

RE: [PATCH net] tcp: Fix integer-overflows in TCP vegas

[David Laight]

From: Of Christoph Paasch
> In vegas we do a multiplication of the cwnd and the rtt. This
> may overflow and thus their result is stored in a u64. However, we first
> need to cast the cwnd so that actually 64-bit arithmetic is done.
> 
> Cc: Doug Leith <doug.leith@nuim.ie>
> Fixes: 8d3a564da34e (tcp: tcp_vegas cong avoid fix)
> Signed-off-by: Christoph Paasch <christoph.paasch@uclouvain.be>
> ---
>  net/ipv4/tcp_vegas.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/net/ipv4/tcp_vegas.c b/net/ipv4/tcp_vegas.c
> index 9a5e05f27f4f..6a4bdea2a0fb 100644
> --- a/net/ipv4/tcp_vegas.c
> +++ b/net/ipv4/tcp_vegas.c
> @@ -218,7 +218,7 @@ static void tcp_vegas_cong_avoid(struct sock *sk, u32 ack, u32 acked)
>  			 * This is:
>  			 *     (actual rate in segments) * baseRTT
>  			 */
> -			target_cwnd = tp->snd_cwnd * vegas->baseRTT / rtt;
> +			target_cwnd = (u64)tp->snd_cwnd * vegas->baseRTT / rtt;

Won't that add a reference to the 64bit divide function?
If snd_cwnd is small then maybe:
	target_cwnd = (256u * vegas->baseRTT) / rtt * tp->snd_cwnd / 256u;
If large I think low bits are always zero so:
	target_cwnd = (256u * vegas->baseRTT) / rtt * (tp->snd_cwnd / 256u);
Possibly with a different power of 2...

	David

Re: [PATCH net] tcp: Fix integer-overflows in TCP vegas

['Christoph Paasch']

On 23/07/14 - 08:38:30, David Laight wrote:
> From: Of Christoph Paasch
> > In vegas we do a multiplication of the cwnd and the rtt. This
> > may overflow and thus their result is stored in a u64. However, we first
> > need to cast the cwnd so that actually 64-bit arithmetic is done.
> > 
> > Cc: Doug Leith <doug.leith@nuim.ie>
> > Fixes: 8d3a564da34e (tcp: tcp_vegas cong avoid fix)
> > Signed-off-by: Christoph Paasch <christoph.paasch@uclouvain.be>
> > ---
> >  net/ipv4/tcp_vegas.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/net/ipv4/tcp_vegas.c b/net/ipv4/tcp_vegas.c
> > index 9a5e05f27f4f..6a4bdea2a0fb 100644
> > --- a/net/ipv4/tcp_vegas.c
> > +++ b/net/ipv4/tcp_vegas.c
> > @@ -218,7 +218,7 @@ static void tcp_vegas_cong_avoid(struct sock *sk, u32 ack, u32 acked)
> >  			 * This is:
> >  			 *     (actual rate in segments) * baseRTT
> >  			 */
> > -			target_cwnd = tp->snd_cwnd * vegas->baseRTT / rtt;
> > +			target_cwnd = (u64)tp->snd_cwnd * vegas->baseRTT / rtt;
> 
> Won't that add a reference to the 64bit divide function?

Yes, you are right. Sorry... (that's what happens if one only tests on a
64-bit system... :S )

> If snd_cwnd is small then maybe:
> 	target_cwnd = (256u * vegas->baseRTT) / rtt * tp->snd_cwnd / 256u;
> If large I think low bits are always zero so:
> 	target_cwnd = (256u * vegas->baseRTT) / rtt * (tp->snd_cwnd / 256u);
> Possibly with a different power of 2...

I realize that baseRTT is always smaller or equal to rtt.

Thus, target_cwnd will always be <= snd_cwnd. So, I think target_cwnd does not
need to be a u64.


I think that it would be safer to do do_div(). That way no matter what the
difference between baseRTT and rtt, we always set the correct value. Otherwise
we might set target_cwnd to 0, while with a do_div it would not.

But a do_div might be more costly than your solution.


What do you (or others) think?


Cheers,
Christoph

Re: [PATCH net] tcp: Fix integer-overflows in TCP vegas

[David Miller]

From: 'Christoph Paasch' <christoph.paasch@uclouvain.be>
Date: Wed, 23 Jul 2014 13:18:02 +0200

> What do you (or others) think?

I think we should avoid the 64-bit operation altogether.

If you can't say for sure whether or not constraints guarantee
no overflows if you code this a certain way, add assertions.