Re: [net-next PATCH] net: reduce cycles spend on ICMP replies that gets rate limited

[Eric Dumazet]

On Fri, 2017-01-06 at 11:40 -0800, Eric Dumazet wrote:
> On Fri, 2017-01-06 at 18:39 +0100, Jesper Dangaard Brouer wrote:
> 
> 
> > @@ -648,13 +668,17 @@ void icmp_send(struct sk_buff *skb_in, int type, int 
> > code, __be32 info)
> >             }
> >     }
> >  
> > -   icmp_param = kmalloc(sizeof(*icmp_param), GFP_ATOMIC);
> > -   if (!icmp_param)
> > -           return;
> > -
> >     sk = icmp_xmit_lock(net);
> >     if (!sk)
> > -           goto out_free;
> > +           goto out;
> > +
> > +   /* Check global sysctl_icmp_msgs_per_sec ratelimit */
> > +   if (!icmpv4_global_allow(net, type, code))
> > +           goto out_unlock;
> > +
> > +   icmp_param = kmalloc(sizeof(*icmp_param), GFP_ATOMIC);
> > +   if (!icmp_param)
> > +           goto out_unlock;
> 

You could call icmp_xmit_lock() _after_ checking global limit perhaps. 

That would remove one atomic op.

if (!icmpv4_global_allow(net, type, code))
    goto out;

sk = icmp_xmit_lock(net);
if (!sk)
    goto out;