Skip to contentSkip to conversation

netfilter: bridge: reset skb->pkt_type after NF_INET_POST_ROUTING traversal

PATCH In review Antoine Tenart 3 participants 5 messages Nov 23, 2020 – Nov 28, 2020

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 netfilter: bridge: reset skb->pkt_type after NF_INET_POST_ROUTING traversal

Netfilter changes PACKET_OTHERHOST to PACKET_HOST before invoking the hooks as, while it's an expected value for a bridge, routing expects PACKET_HOST. The change is undone later on after hook traversal. This can be seen with pairs of functions updating skb>pkt_type and then reverting it to its original value:

For hook NF_INET_PRE_ROUTING:
setup_pre_routing / br_nf_pre_routing_finish

For hook NF_INET_FORWARD:
br_nf_forward_ip / br_nf_forward_finish

But the third case where netfilter does this, for hook NF_INET_POST_ROUTING, the packet type is changed in br_nf_post_routing but never reverted. A comment says:

/* We assume any code from br_dev_queue_push_xmit onwards doesn't care
* about the value of skb->pkt_type. */

But when having a tunnel (say vxlan) attached to a bridge we have the following call trace:

br_nf_pre_routing
br_nf_pre_routing_ipv6
br_nf_pre_routing_finish
br_nf_forward_ip
br_nf_forward_finish
br_nf_post_routing <- pkt_type is updated to PACKET_HOST
br_nf_dev_queue_xmit <- but not reverted to its original value
vxlan_xmit
vxlan_xmit_one
skb_tunnel_check_pmtu <- a check on pkt_type is performed

In this specific case, this creates issues such as when an ICMPv6 PTB should be sent back. When CONFIG_BRIDGE_NETFILTER is enabled, the PTB isn't sent (as skb_tunnel_check_pmtu checks if pkt_type is PACKET_HOST and returns early).

If the comment is right and no one cares about the value of skb->pkt_type after br_dev_queue_push_xmit (which isn't true), resetting it to its original value should be safe.

---
 net/bridge/br_netfilter_hooks.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
net/bridge/br_netfilter_hooks.c +5 -2
index 04c3f9a82650..8edfb98ae1d5 100644
@@ -735,6 +735,11 @@ static int br_nf_dev_queue_xmit(struct net *net, struct sock *sk, struct sk_buff
 	mtu_reserved = nf_bridge_mtu_reduction(skb);
 	mtu = skb->dev->mtu;
 
+	if (nf_bridge->pkt_otherhost) {
+		skb->pkt_type = PACKET_OTHERHOST;
+		nf_bridge->pkt_otherhost = false;
+	}
+
 	if (nf_bridge->frag_max_size && nf_bridge->frag_max_size < mtu)
 		mtu = nf_bridge->frag_max_size;
 
@@ -835,8 +840,6 @@ static unsigned int br_nf_post_routing(void *priv,
 	else
 		return NF_ACCEPT;
 
-	/* We assume any code from br_dev_queue_push_xmit onwards doesn't care
-	 * about the value of skb->pkt_type. */
 	if (skb->pkt_type == PACKET_OTHERHOST) {
 		skb->pkt_type = PACKET_HOST;
 		nf_bridge->pkt_otherhost = true;
Signature · 1 line
2.28.0
maintainer Reviewed-by on v1 1/1
#1 lore raw
Quoted 18 lines · Antoine Tenart

Netfilter changes PACKET_OTHERHOST to PACKET_HOST before invoking the hooks as, while it's an expected value for a bridge, routing expects PACKET_HOST. The change is undone later on after hook traversal. This can be seen with pairs of functions updating skb>pkt_type and then reverting it to its original value:

For hook NF_INET_PRE_ROUTING:
setup_pre_routing / br_nf_pre_routing_finish

For hook NF_INET_FORWARD:
br_nf_forward_ip / br_nf_forward_finish

But the third case where netfilter does this, for hook NF_INET_POST_ROUTING, the packet type is changed in br_nf_post_routing but never reverted. A comment says:

/* We assume any code from br_dev_queue_push_xmit onwards doesn't care
* about the value of skb->pkt_type. */

[..]

Quoted 11 lines

But when having a tunnel (say vxlan) attached to a bridge we have the following call trace:

In this specific case, this creates issues such as when an ICMPv6 PTB should be sent back. When CONFIG_BRIDGE_NETFILTER is enabled, the PTB isn't sent (as skb_tunnel_check_pmtu checks if pkt_type is PACKET_HOST and returns early).

If the comment is right and no one cares about the value of skb->pkt_type after br_dev_queue_push_xmit (which isn't true), resetting it to its original value should be safe.

That comment is 18 years old, safe bet noone thought of ipv6-in-tunnel-interface-added-as-bridge-port back then.

maintainer
#2 lore raw
Quoted 29 lines · On Mon, 23 Nov 2020 19:32:53 +0100 Florian Westphal wrote:
Quoted 18 lines · Antoine Tenart

Netfilter changes PACKET_OTHERHOST to PACKET_HOST before invoking the hooks as, while it's an expected value for a bridge, routing expects PACKET_HOST. The change is undone later on after hook traversal. This can be seen with pairs of functions updating skb>pkt_type and then reverting it to its original value:

For hook NF_INET_PRE_ROUTING:
setup_pre_routing / br_nf_pre_routing_finish

For hook NF_INET_FORWARD:
br_nf_forward_ip / br_nf_forward_finish

But the third case where netfilter does this, for hook NF_INET_POST_ROUTING, the packet type is changed in br_nf_post_routing but never reverted. A comment says:

/* We assume any code from br_dev_queue_push_xmit onwards doesn't care
* about the value of skb->pkt_type. */

[..]

Quoted 8 lines

But when having a tunnel (say vxlan) attached to a bridge we have the following call trace:

In this specific case, this creates issues such as when an ICMPv6 PTB should be sent back. When CONFIG_BRIDGE_NETFILTER is enabled, the PTB isn't sent (as skb_tunnel_check_pmtu checks if pkt_type is PACKET_HOST and returns early).

Quoted 8 lines
Quoted 3 lines

If the comment is right and no one cares about the value of skb->pkt_type after br_dev_queue_push_xmit (which isn't true), resetting it to its original value should be safe.

That comment is 18 years old, safe bet noone thought of ipv6-in-tunnel-interface-added-as-bridge-port back then.

Sounds like a fix. Probably hard to pin point which commit to blame, but this should go to net, not net-next, right?

maintainer
#3 lore raw
Quoted 8 lines · Jakub Kicinski
Quoted 4 lines · On Mon, 23 Nov 2020 19:32:53 +0100 Florian Westphal wrote:

That comment is 18 years old, safe bet noone thought of ipv6-in-tunnel-interface-added-as-bridge-port back then.

Sounds like a fix. Probably hard to pin point which commit to blame, but this should go to net, not net-next, right?

The commit predates git history, so probably a good idea to add

Fixes 1da177e4c3f4 ("Linux-2.6.12-rc2")

... and apply it to net tree.

maintainer
#4 lore raw
Quoted 7 lines · On Sat, 28 Nov 2020 10:59:29 +0100 Florian Westphal wrote:
Quoted 6 lines · Jakub Kicinski
Quoted 4 lines · On Mon, 23 Nov 2020 19:32:53 +0100 Florian Westphal wrote:

That comment is 18 years old, safe bet noone thought of ipv6-in-tunnel-interface-added-as-bridge-port back then.

Quoted 7 lines
Quoted 2 lines

Sounds like a fix. Probably hard to pin point which commit to blame, but this should go to net, not net-next, right?

The commit predates git history, so probably a good idea to add

Fixes 1da177e4c3f4 ("Linux-2.6.12-rc2")

... and apply it to net tree.

Done, thanks!