Re: [PATCH net] openvswitch: Fix ovs_flow_key_update()
From: Yi-Hung Wei <hidden>
Date: 2017-03-30 18:39:59
Also in:
lkml
On Thu, Mar 30, 2017 at 6:22 AM, Jiri Benc [off-list ref] wrote:
On Wed, 29 Mar 2017 17:14:10 -0700, Yi-Hung Wei wrote:quoted
ovs_flow_key_update() is called when the flow key is invalid, and it is used to update and revalidate the flow key. Commit 329f45bc4f19 ("openvswitch: add mac_proto field to the flow key") introduces mac_proto field to flow key and use it to determine whether the flow key is valid. However, the commit does not update the code path in ovs_flow_key_update() to revalidate the flow key which may cause BUG_ON() on execute_recirc(). This patch addresses the aforementioned issue.Good catch.quoted
int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key) { + int res; + + res = key_extract_mac_proto(skb); + if (res < 0) + return res; + key->mac_proto = res; + return key_extract(skb, key); }But this should just reset the SW_FLOW_KEY_INVALID flag, there's no need to recompute mac_proto. Something like this: int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key) { - return key_extract(skb, key); + int res; + + res = key_extract(skb, key); + if (!res) + key->mac_proto &= ~SW_FLOW_KEY_INVALID; + return res; }
Hi Jiri,
One case that I worry is that key_extract() currently relies on mac_proto to
decide whether to process the link layer. So if we update key->mac_proto
after key_extract(), wouldn't we run into a problem like the following?
If we invalidate a flow key of a L3 packet, the flow's mac_proto is like this
(MAC_PROTO_NONE | SW_FLOW_KEY_INVALID), then key_extract() will
process the link layer of this L3 packet since mac_proto !=MAC_PROTO_NONE?
In this case, shall we update key_extract() like this
static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
/* Link layer. */
clear_vlan(key);
- if (key->mac_proto == MAC_PROTO_NONE) {
+ if (key->mac_proto & MAC_PROTO_NONE) {
if (unlikely(eth_type_vlan(skb->protocol)))
return -EINVAL;
Thanks,
-Yi-Hung
Thanks, Jiri