Hi Eric, Thanks for the review. On Thu, Aug 15, 2019 at 11:16:58AM +0200, Eric Dumazet wrote: > > diff --git a/net/ipv4/ip_tunnel.c b/net/ipv4/ip_tunnel.c > > index 38c02bb62e2c..c6713c7287df 100644 > > --- a/net/ipv4/ip_tunnel.c > > +++ b/net/ipv4/ip_tunnel.c > > @@ -597,6 +597,9 @@ void ip_md_tunnel_xmit(struct sk_buff *skb, struct > > net_device *dev, > > goto tx_error; > > } > > > > + if (skb_dst(skb) && !skb_dst(skb)->dev) > > + skb_dst(skb)->dev = rt->dst.dev; > > + > > > IMO this looks wrong. > This dst seems shared.
If the dst is shared, it may cause some problem. Could you point me where the dst may be shared possibly? > Once set, we will reuse the same dev ? If yes, how about just set the skb dst to rt->dst, as the iptunnel_xmit would do later. skb_dst_drop(skb); skb_dst_set(skb, &rt->dst); or do you have any other idea? > > If intended, why not doing this in __metadata_dst_init() instead of in the > fast path ? I'm afraid we couldn't do this, I didn't find a way to init dev in __metadata_dst_init(). Do you? Thanks Hangbin
