Re: [PATCH 3/8] net: Add Generic Receive Offload infrastructure

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Mon, Dec 15, 2008 at 03:39:20PM -0800, David Miller wrote:
> From: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
> Date: Mon, 15 Dec 2008 15:29:42 -0800
> 
> > On Sat, Dec 13, 2008 at 12:35:21PM +1100, Herbert Xu wrote:
> > > +static int napi_gro_complete(struct sk_buff *skb)
> > > +{
> > > +	struct packet_type *ptype;
> > > +	__be16 type = skb->protocol;
> > > +	struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
> > > +	int err = -ENOENT;
> > > +
> > > +	if (!skb_shinfo(skb)->frag_list)
> > > +		goto out;
> > > +
> > > +	rcu_read_lock();
> > > +	list_for_each_entry_rcu(ptype, head, list) {
> > > +		if (ptype->type != type || ptype->dev || !ptype->gro_complete)
> > > +			continue;
> > > +
> > > +		err = ptype->gro_complete(skb);
> > > +		break;
> > > +	}
> > > +	rcu_read_unlock();
> > > +
> > > +	if (err) {
> > > +		WARN_ON(&ptype->list == head);
> > 
> > Presumably ptype_base[] is a static array rather than a dynamically
> > allocated array that is resized under RCU protection, right?  Otherwise,
> > you could get in trouble if the above raced with the resize operation due
> > to the fact that you are outside of the RCU read-side critical section.
> 
> Yes, and we've been using RCU this way for this table for quite
> some time.  From net/core/dev.c:
> 
> #define PTYPE_HASH_SIZE	(16)
> #define PTYPE_HASH_MASK	(PTYPE_HASH_SIZE - 1)
> 
> static DEFINE_SPINLOCK(ptype_lock);
> static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
> static struct list_head ptype_all __read_mostly;	/* Taps */

Works for me, then!

							Thanx, Paul