[PATCH v2] openvswitch: Disable bottom half when grabbing stats lock in userspace context

[PATCH v2] openvswitch: Disable bottom half when grabbing stats lock in userspace context

[Zoltan Kiss]

The megaflow feature introduced per-CPU stats, and the new code did some
refactoring as well [1]. However the new functions doesn't disable bottom halves
when locking the stat, and as they can be called from userspace context, they
can be interrupted by netif_receive_skb.

[1] e298e50570: openvswitch: Per cpu flow stats.

Signed-off-by: Zoltan Kiss <zoltan.kiss@citrix.com>
---
v2: use _bh instead of _irqsave/restore
 net/openvswitch/flow.c |    8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/net/openvswitch/flow.c b/net/openvswitch/flow.c
index 16f4b46..9110cf4 100644
--- a/net/openvswitch/flow.c
+++ b/net/openvswitch/flow.c
@@ -90,13 +90,13 @@ static void stats_read(struct flow_stats *stats,
 		       struct ovs_flow_stats *ovs_stats,
 		       unsigned long *used, __be16 *tcp_flags)
 {
-	spin_lock(&stats->lock);
+	spin_lock_bh(&stats->lock);
 	if (time_after(stats->used, *used))
 		*used = stats->used;
 	*tcp_flags |= stats->tcp_flags;
 	ovs_stats->n_packets += stats->packet_count;
 	ovs_stats->n_bytes += stats->byte_count;
-	spin_unlock(&stats->lock);
+	spin_unlock_bh(&stats->lock);
 }
 
 void ovs_flow_stats_get(struct sw_flow *flow, struct ovs_flow_stats *ovs_stats,
@@ -130,12 +130,12 @@ void ovs_flow_stats_get(struct sw_flow *flow, struct ovs_flow_stats *ovs_stats,
 
 static void stats_reset(struct flow_stats *stats)
 {
-	spin_lock(&stats->lock);
+	spin_lock_bh(&stats->lock);
 	stats->used = 0;
 	stats->packet_count = 0;
 	stats->byte_count = 0;
 	stats->tcp_flags = 0;
-	spin_unlock(&stats->lock);
+	spin_unlock_bh(&stats->lock);
 }
 
 void ovs_flow_stats_clear(struct sw_flow *flow)

[PATCH v2] openvswitch: Disable bottom half when grabbing stats lock in userspace context

[Zoltan Kiss]

The megaflow feature introduced per-CPU stats, and the new code did some
refactoring as well [1]. However the new functions doesn't disable bottom halves
when locking the stat, and as they can be called from userspace context, they
can be interrupted by netif_receive_skb.

[1] e298e50570: openvswitch: Per cpu flow stats.

Signed-off-by: Zoltan Kiss <zoltan.kiss-Sxgqhf6Nn4DQT0dZR+AlfA@public.gmane.org>
---
v2: use _bh instead of _irqsave/restore
 net/openvswitch/flow.c |    8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/net/openvswitch/flow.c b/net/openvswitch/flow.c
index 16f4b46..9110cf4 100644
--- a/net/openvswitch/flow.c
+++ b/net/openvswitch/flow.c
@@ -90,13 +90,13 @@ static void stats_read(struct flow_stats *stats,
 		       struct ovs_flow_stats *ovs_stats,
 		       unsigned long *used, __be16 *tcp_flags)
 {
-	spin_lock(&stats->lock);
+	spin_lock_bh(&stats->lock);
 	if (time_after(stats->used, *used))
 		*used = stats->used;
 	*tcp_flags |= stats->tcp_flags;
 	ovs_stats->n_packets += stats->packet_count;
 	ovs_stats->n_bytes += stats->byte_count;
-	spin_unlock(&stats->lock);
+	spin_unlock_bh(&stats->lock);
 }
 
 void ovs_flow_stats_get(struct sw_flow *flow, struct ovs_flow_stats *ovs_stats,
@@ -130,12 +130,12 @@ void ovs_flow_stats_get(struct sw_flow *flow, struct ovs_flow_stats *ovs_stats,
 
 static void stats_reset(struct flow_stats *stats)
 {
-	spin_lock(&stats->lock);
+	spin_lock_bh(&stats->lock);
 	stats->used = 0;
 	stats->packet_count = 0;
 	stats->byte_count = 0;
 	stats->tcp_flags = 0;
-	spin_unlock(&stats->lock);
+	spin_unlock_bh(&stats->lock);
 }
 
 void ovs_flow_stats_clear(struct sw_flow *flow)

Re: [PATCH v2] openvswitch: Disable bottom half when grabbing stats lock in userspace context

[Pravin Shelar]

On Thu, Feb 27, 2014 at 2:10 PM, Zoltan Kiss <zoltan.kiss@citrix.com> wrote:
> The megaflow feature introduced per-CPU stats, and the new code did some
> refactoring as well [1]. However the new functions doesn't disable bottom halves
> when locking the stat, and as they can be called from userspace context, they
> can be interrupted by netif_receive_skb.
>
> [1] e298e50570: openvswitch: Per cpu flow stats.
>
This bug is fixed in openvswitch repo by commit "datapath: Fix
deadlock during stats update."


Thanks.