Looking at do_output, it appears that we silently drop a packet if a bad
out_port was specified because vport will be NULL from ovs_vport_rcu and we
just free the skb and return from thereon. Instead, an ideal behavior should
be to log this in user space via log_execute_message and also perhaps into
dmesg as a warning.
In 2.3, we had do_output return an error, but it was never propagated to
user space through do_execute_actions, but seems like fe90efd9d converted
do_output to a 'void' function. I'm thinking if we should bring back return
value of do_output and then have do_execute_actions return it to userspace.
Please suggest..
A patch on top of 2.3 should look like this:
diff --git a/datapath/actions.c b/datapath/actions.c
index ddacbc6..41b25aa 100644
--- a/datapath/actions.c
+++ b/datapath/actions.c
@@ -402,6 +402,7 @@ static int do_output(struct datapath *dp, struct sk_buff
*skb, int out_port)
vport = ovs_vport_rcu(dp, out_port);
if (unlikely(!vport)) {
kfree_skb(skb);
+ pr_warn("invalid port number specified: %d\n", out_port);
return -ENODEV;
}
@@ -584,13 +585,14 @@ static int do_execute_actions(struct datapath *dp, struct
sk_buff *skb,
int prev_port = -1;
const struct nlattr *a;
int rem;
+ int err;
for (a = attr, rem = len; rem > 0;
a = nla_next(a, &rem)) {
int err = 0;
if (prev_port != -1) {
- do_output(dp, skb_clone(skb, GFP_ATOMIC), prev_port);
+ err = do_output(dp, skb_clone(skb, GFP_ATOMIC),
prev_port);
prev_port = -1;
}
@@ -651,11 +653,11 @@ static int do_execute_actions(struct datapath *dp, struct
sk_buff *skb,
}
if (prev_port != -1)
- do_output(dp, skb, prev_port);
+ err = do_output(dp, skb, prev_port);
else
consume_skb(skb);
- return 0;
+ return err;
}
/* We limit the number of times that we pass into execute_actions()
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
