[PATCH net-next] bonding: fix send_peer_notif leekage on rtnl lock congestion

[PATCH net-next] bonding: fix send_peer_notif leekage on rtnl lock congestion

[Veaceslav Falico]

In bond_mii_monitor()/bond_activebackup_arp_mon(), we verify if we have any
notifications to be sent before trying to get the rtnl_trylock(), and if
we have - we set should_notify_peers and withdraw a notification. However,
if we fail the get the rtnl_trylock(), we don't send out any notification
and don't put the notification back to bond->send_peer_notif.

Fix it by putting back the notification to send_peer_notif in case of
rtnl_trylock() failure.

CC: Jay Vosburgh <fubar@us.ibm.com>
CC: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Veaceslav Falico <vfalico@redhat.com>
---
 drivers/net/bonding/bond_main.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
index 5697043..fd92738 100644
--- a/drivers/net/bonding/bond_main.c
+++ b/drivers/net/bonding/bond_main.c
@@ -2383,7 +2383,10 @@ void bond_mii_monitor(struct work_struct *work)
 		if (!rtnl_trylock()) {
 			read_lock(&bond->lock);
 			delay = 1;
-			should_notify_peers = false;
+			if (should_notify_peers) {
+				bond->send_peer_notif++;
+				should_notify_peers = false;
+			}
 			goto re_arm;
 		}
 
@@ -2999,7 +3002,10 @@ void bond_activebackup_arp_mon(struct work_struct *work)
 		if (!rtnl_trylock()) {
 			read_lock(&bond->lock);
 			delta_in_ticks = 1;
-			should_notify_peers = false;
+			if (should_notify_peers) {
+				bond->send_peer_notif++;
+				should_notify_peers = false;
+			}
 			goto re_arm;
 		}
 
-- 
1.8.1.4

Re: [PATCH net-next] bonding: fix send_peer_notif leekage on rtnl lock congestion

[David Miller]

From: Veaceslav Falico <vfalico@redhat.com>
Date: Mon,  5 Aug 2013 21:47:33 +0200

> @@ -2999,7 +3002,10 @@ void bond_activebackup_arp_mon(struct work_struct *work)
>  		if (!rtnl_trylock()) {
>  			read_lock(&bond->lock);
>  			delta_in_ticks = 1;
> -			should_notify_peers = false;
> +			if (should_notify_peers) {
> +				bond->send_peer_notif++;

I doubt this increment to a shared datastructure is safe with
the locks you hold here.

You don't hold RTNL and you only have bond->lock as a reader.

Re: [PATCH net-next] bonding: fix send_peer_notif leekage on rtnl lock congestion

[Veaceslav Falico]

On Mon, Aug 05, 2013 at 04:12:12PM -0700, David Miller wrote:
>From: Veaceslav Falico <vfalico@redhat.com>
>Date: Mon,  5 Aug 2013 21:47:33 +0200
>
>> @@ -2999,7 +3002,10 @@ void bond_activebackup_arp_mon(struct work_struct *work)
>>  		if (!rtnl_trylock()) {
>>  			read_lock(&bond->lock);
>>  			delta_in_ticks = 1;
>> -			should_notify_peers = false;
>> +			if (should_notify_peers) {
>> +				bond->send_peer_notif++;
>
>I doubt this increment to a shared datastructure is safe with
>the locks you hold here.
>
>You don't hold RTNL and you only have bond->lock as a reader.

You're right, we can race here. This whole locking juggling should be
changed.

Self-NAK for this, I'll rework it.