[PATCH net 0/2] VXLAN/geneve RCU fixes

[PATCH net 0/2] VXLAN/geneve RCU fixes

[Jakub Kicinski]

Hi!

VXLAN and GENEVE need to take RCU lock explicitly because TX path
only has the _bh() flavour of RCU locked.  Making the reconfiguration
path wait for both normal and _bh() RCU would be bigger hassle so
just acquire the lock, as suggested by Pravin:

https://www.mail-archive.com/netdev@vger.kernel.org/msg155583.html

Jakub Kicinski (2):
  vxlan: lock RCU on TX path
  geneve: lock RCU on TX path

 drivers/net/geneve.c | 2 ++
 drivers/net/vxlan.c  | 8 ++++++--
 2 files changed, 8 insertions(+), 2 deletions(-)

-- 
2.11.0

[PATCH net 1/2] vxlan: lock RCU on TX path

[Jakub Kicinski]

There is no guarantees that callers of the TX path will hold
the RCU lock.  Grab it explicitly.

Fixes: c6fcc4fc5f8b ("vxlan: avoid using stale vxlan socket.")
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 drivers/net/vxlan.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/drivers/net/vxlan.c b/drivers/net/vxlan.c
index 4e27c5b09600..e53ee696c962 100644
--- a/drivers/net/vxlan.c
+++ b/drivers/net/vxlan.c
@@ -2108,6 +2108,7 @@ static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
 	src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
 				     vxlan->cfg.port_max, true);
 
+	rcu_read_lock();
 	if (dst->sa.sa_family == AF_INET) {
 		struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
 		struct rtable *rt;
@@ -2130,7 +2131,7 @@ static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
 						    dst_port, vni, &rt->dst,
 						    rt->rt_flags);
 			if (err)
-				return;
+				goto out_unlock;
 		} else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
 			df = htons(IP_DF);
 		}
@@ -2169,7 +2170,7 @@ static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
 						    dst_port, vni, ndst,
 						    rt6i_flags);
 			if (err)
-				return;
+				goto out_unlock;
 		}
 
 		tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
@@ -2186,6 +2187,8 @@ static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
 				     label, src_port, dst_port, !udp_sum);
 #endif
 	}
+out_unlock:
+	rcu_read_unlock();
 	return;
 
 drop:
@@ -2194,6 +2197,7 @@ static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
 	return;
 
 tx_error:
+	rcu_read_unlock();
 	if (err == -ELOOP)
 		dev->stats.collisions++;
 	else if (err == -ENETUNREACH)
-- 
2.11.0

[PATCH net 2/2] geneve: lock RCU on TX path

[Jakub Kicinski]

There is no guarantees that callers of the TX path will hold
the RCU lock.  Grab it explicitly.

Fixes: fceb9c3e3825 ("geneve: avoid using stale geneve socket.")
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 drivers/net/geneve.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/net/geneve.c b/drivers/net/geneve.c
index 45301cb98bc1..7074b40ebd7f 100644
--- a/drivers/net/geneve.c
+++ b/drivers/net/geneve.c
@@ -881,12 +881,14 @@ static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
 		info = &geneve->info;
 	}
 
+	rcu_read_lock();
 #if IS_ENABLED(CONFIG_IPV6)
 	if (info->mode & IP_TUNNEL_INFO_IPV6)
 		err = geneve6_xmit_skb(skb, dev, geneve, info);
 	else
 #endif
 		err = geneve_xmit_skb(skb, dev, geneve, info);
+	rcu_read_unlock();
 
 	if (likely(!err))
 		return NETDEV_TX_OK;
-- 
2.11.0

Re: [PATCH net 0/2] VXLAN/geneve RCU fixes

[David Miller]

From: Jakub Kicinski <jakub.kicinski@netronome.com>
Date: Fri, 24 Feb 2017 11:43:35 -0800

> VXLAN and GENEVE need to take RCU lock explicitly because TX path
> only has the _bh() flavour of RCU locked.  Making the reconfiguration
> path wait for both normal and _bh() RCU would be bigger hassle so
> just acquire the lock, as suggested by Pravin:
> 
> https://www.mail-archive.com/netdev@vger.kernel.org/msg155583.html

Series applied and queued up for -stable, thanks.