If we run hundreds of socket parallel we may run in to situation
where last socket is about to unregister old priv and the new one
is taking it half usable.
Signed-off-by: Oleksij Rempel <o.rempel@xxxxxxxxxxxxxx>
---
net/can/j1939/main.c | 7 ++++---
1 file changed, 4 insertions(+), 3 deletions(-)
diff --git a/net/can/j1939/main.c b/net/can/j1939/main.c
index ac026a8408c3..ac786bde5948 100644
--- a/net/can/j1939/main.c
+++ b/net/can/j1939/main.c
@@ -185,6 +185,7 @@ static void __j1939_rx_release(struct kref *kref)
j1939_can_rx_unregister(priv);
j1939_ecu_unmap_all(priv);
j1939_priv_set(priv->ndev, NULL);
+ spin_unlock(&j1939_netdev_lock);
}
/* get pointer to priv without increasing ref counter */
@@ -215,9 +216,9 @@ struct j1939_priv *j1939_priv_get_by_ndev(struct net_device *ndev)
{
struct j1939_priv *priv;
- spin_lock_bh(&j1939_netdev_lock);
+ spin_lock(&j1939_netdev_lock);
priv = j1939_priv_get_by_ndev_locked(ndev);
- spin_unlock_bh(&j1939_netdev_lock);
+ spin_unlock(&j1939_netdev_lock);
return priv;
}
@@ -272,7 +273,7 @@ struct j1939_priv *j1939_netdev_start(struct net *net, struct net_device *ndev)
void j1939_netdev_stop(struct j1939_priv *priv)
{
- kref_put(&priv->rx_kref, __j1939_rx_release);
+ kref_put_lock(&priv->rx_kref, __j1939_rx_release, &j1939_netdev_lock);
j1939_priv_put(priv);
}
--
2.20.1
