Re: [PATCH] tcp: avoid a possible divide by zero

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

Le mardi 07 décembre 2010 à 21:32 +0000, Ben Hutchings a écrit :
> On Tue, 2010-12-07 at 22:28 +0100, Eric Dumazet wrote:
> [...]
> > Thanks
> > 
> > Great, I feel we are going to fix all sysctls, one by one then :(
> > 
> > lkml removed from Cc
> > 
> > 
> > [PATCH] tcp: avoid a possible divide by zero
> > 
> > sysctl_tcp_tso_win_divisor might be set to zero while one cpu runs in
> > tcp_tso_should_defer(). Make sure we dont allow a divide by zero by
> > reading sysctl_tcp_tso_win_divisor once.
> > 
> > Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
> > ---
> >  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 05b1ecf..0281223 100644
> > --- a/net/ipv4/tcp_output.c
> > +++ b/net/ipv4/tcp_output.c
> > @@ -1513,6 +1513,7 @@ static int tcp_tso_should_defer(struct sock *sk, struct sk_buff *skb)
> >  	struct tcp_sock *tp = tcp_sk(sk);
> >  	const struct inet_connection_sock *icsk = inet_csk(sk);
> >  	u32 send_win, cong_win, limit, in_flight;
> > +	int win_divisor;
> >  
> >  	if (TCP_SKB_CB(skb)->flags & TCPHDR_FIN)
> >  		goto send_now;
> > @@ -1544,13 +1545,14 @@ static int tcp_tso_should_defer(struct sock *sk, struct sk_buff *skb)
> >  	if ((skb != tcp_write_queue_tail(sk)) && (limit >= skb->len))
> >  		goto send_now;
> >  
> > -	if (sysctl_tcp_tso_win_divisor) {
> > +	win_divisor = sysctl_tcp_tso_win_divisor;
> 
> You need to use ACCESS_ONCE(sysctl_tcp_tso_win_divisor).  Otherwise the
> compiler may eliminate the local variable and read the global twice.

Yes, I knew that, of course :)

I wonder how many bugs like that we have in sysctls

Thanks

[PATCH v2] tcp: avoid a possible divide by zero

sysctl_tcp_tso_win_divisor might be set to zero while one cpu runs in
tcp_tso_should_defer(). Make sure we dont allow a divide by zero by
reading sysctl_tcp_tso_win_divisor exactly once.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
---
v2: Use ACCESS_ONCE() as Ben suggested

 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 05b1ecf..0464d70 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -1513,6 +1513,7 @@ static int tcp_tso_should_defer(struct sock *sk, struct sk_buff *skb)
 	struct tcp_sock *tp = tcp_sk(sk);
 	const struct inet_connection_sock *icsk = inet_csk(sk);
 	u32 send_win, cong_win, limit, in_flight;
+	int win_divisor;
 
 	if (TCP_SKB_CB(skb)->flags & TCPHDR_FIN)
 		goto send_now;
@@ -1544,13 +1545,14 @@ static int tcp_tso_should_defer(struct sock *sk, struct sk_buff *skb)
 	if ((skb != tcp_write_queue_tail(sk)) && (limit >= skb->len))
 		goto send_now;
 
-	if (sysctl_tcp_tso_win_divisor) {
+	win_divisor = ACCESS_ONCE(sysctl_tcp_tso_win_divisor);
+	if (win_divisor) {
 		u32 chunk = min(tp->snd_wnd, tp->snd_cwnd * tp->mss_cache);
 
 		/* If at least some fraction of a window is available,
 		 * just use it.
 		 */
-		chunk /= sysctl_tcp_tso_win_divisor;
+		chunk /= win_divisor;
 		if (limit >= chunk)
 			goto send_now;
 	} else {