Patchwork ipvs: Use atomic operations atomicly

login
register
mail settings
Submitter Simon Horman
Date Aug. 28, 2009, 2:37 a.m.
Message ID <20090828023722.GA12136@verge.net.au>
Download mbox | patch
Permalink /patch/32341/
State Not Applicable
Delegated to: David Miller
Headers show

Comments

Simon Horman - Aug. 28, 2009, 2:37 a.m.
A pointed out by Shin Hong, IPVS doesn't always use atomic operations
in an atomic manner. While this seems unlikely to be manifest in
strange behaviour, it seems appropriate to clean this up.

Cc: 홍신 shin hong <hongshin@gmail.com>
Signed-off-by: Simon Horman <horms@verge.net.au>

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Patrick McHardy - Aug. 31, 2009, 12:22 p.m.
Simon Horman wrote:
> A pointed out by Shin Hong, IPVS doesn't always use atomic operations
> in an atomic manner. While this seems unlikely to be manifest in
> strange behaviour, it seems appropriate to clean this up.
> 
> Cc: 홍신 shin hong <hongshin@gmail.com>
> Signed-off-by: Simon Horman <horms@verge.net.au>

Applied, thanks.

>  	if (af == AF_INET &&
>  	    (ip_vs_sync_state & IP_VS_STATE_MASTER) &&
>  	    (((cp->protocol != IPPROTO_TCP ||
>  	       cp->state == IP_VS_TCP_S_ESTABLISHED) &&
> -	      (atomic_read(&cp->in_pkts) % sysctl_ip_vs_sync_threshold[1]
> +	      (pkts % sysctl_ip_vs_sync_threshold[1]

It seems that proc_do_sync_threshold() should check whether this value
is zero. The current checks also look racy since incorrect values are
first updated, then overwritten again.

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Simon Horman - Aug. 31, 2009, 2:49 p.m.
On Mon, Aug 31, 2009 at 02:22:26PM +0200, Patrick McHardy wrote:
> Simon Horman wrote:
> > A pointed out by Shin Hong, IPVS doesn't always use atomic operations
> > in an atomic manner. While this seems unlikely to be manifest in
> > strange behaviour, it seems appropriate to clean this up.
> > 
> > Cc: 홍신 shin hong <hongshin@gmail.com>
> > Signed-off-by: Simon Horman <horms@verge.net.au>
> 
> Applied, thanks.
> 
> >  	if (af == AF_INET &&
> >  	    (ip_vs_sync_state & IP_VS_STATE_MASTER) &&
> >  	    (((cp->protocol != IPPROTO_TCP ||
> >  	       cp->state == IP_VS_TCP_S_ESTABLISHED) &&
> > -	      (atomic_read(&cp->in_pkts) % sysctl_ip_vs_sync_threshold[1]
> > +	      (pkts % sysctl_ip_vs_sync_threshold[1]
> 
> It seems that proc_do_sync_threshold() should check whether this value
> is zero. The current checks also look racy since incorrect values are
> first updated, then overwritten again.

Thanks, I'll look into that.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

Index: net-next-2.6/net/netfilter/ipvs/ip_vs_core.c
===================================================================
--- net-next-2.6.orig/net/netfilter/ipvs/ip_vs_core.c	2009-08-28 12:07:49.000000000 +1000
+++ net-next-2.6/net/netfilter/ipvs/ip_vs_core.c	2009-08-28 12:33:18.000000000 +1000
@@ -1259,7 +1259,7 @@  ip_vs_in(unsigned int hooknum, struct sk
 	struct ip_vs_iphdr iph;
 	struct ip_vs_protocol *pp;
 	struct ip_vs_conn *cp;
-	int ret, restart, af;
+	int ret, restart, af, pkts;
 
 	af = (skb->protocol == htons(ETH_P_IP)) ? AF_INET : AF_INET6;
 
@@ -1346,12 +1346,12 @@  ip_vs_in(unsigned int hooknum, struct sk
 	 * Sync connection if it is about to close to
 	 * encorage the standby servers to update the connections timeout
 	 */
-	atomic_inc(&cp->in_pkts);
+	pkts = atomic_add_return(1, &cp->in_pkts);
 	if (af == AF_INET &&
 	    (ip_vs_sync_state & IP_VS_STATE_MASTER) &&
 	    (((cp->protocol != IPPROTO_TCP ||
 	       cp->state == IP_VS_TCP_S_ESTABLISHED) &&
-	      (atomic_read(&cp->in_pkts) % sysctl_ip_vs_sync_threshold[1]
+	      (pkts % sysctl_ip_vs_sync_threshold[1]
 	       == sysctl_ip_vs_sync_threshold[0])) ||
 	     ((cp->protocol == IPPROTO_TCP) && (cp->old_state != cp->state) &&
 	      ((cp->state == IP_VS_TCP_S_FIN_WAIT) ||
Index: net-next-2.6/net/netfilter/ipvs/ip_vs_wrr.c
===================================================================
--- net-next-2.6.orig/net/netfilter/ipvs/ip_vs_wrr.c	2009-08-28 12:07:49.000000000 +1000
+++ net-next-2.6/net/netfilter/ipvs/ip_vs_wrr.c	2009-08-28 12:33:18.000000000 +1000
@@ -77,11 +77,12 @@  static int ip_vs_wrr_gcd_weight(struct i
 static int ip_vs_wrr_max_weight(struct ip_vs_service *svc)
 {
 	struct ip_vs_dest *dest;
-	int weight = 0;
+	int new_weight, weight = 0;
 
 	list_for_each_entry(dest, &svc->destinations, n_list) {
-		if (atomic_read(&dest->weight) > weight)
-			weight = atomic_read(&dest->weight);
+		new_weight = atomic_read(&dest->weight);
+		if (new_weight > weight)
+			weight = new_weight;
 	}
 
 	return weight;