ip_list_rcv() question

ip_list_rcv() question

[Stephen Suryaputra]

Hi,

Under what condition that ip_list_rcv() would restart the sublist, i.e.
that the skb in the list is having different skb->dev?

void ip_list_rcv(struct list_head *head, struct packet_type *pt,
		 struct net_device *orig_dev)
{
	...
	list_for_each_entry_safe(skb, next, head, list) {
		...
		if (curr_dev != dev || curr_net != net) {
			/* dispatch old sublist */
			...
			/* start new sublist */
			...
		}
		...
	}
	...
}

In my system, the function gets called from the following call chain:
net_rx_action()->napi_poll()->gro_normal_list()

It seems to me that there is one to one relationship between napi_struct
and net_device, but I saw an archived discussion about it might be one
to many. But I didn't see it.

Thanks in advance for the clarifications.

Regards,
Stephen.

Re: ip_list_rcv() question

[Edward Cree]

On 07/10/2021 13:14, Stephen Suryaputra wrote:
> Under what condition that ip_list_rcv() would restart the sublist, i.e.
> that the skb in the list is having different skb->dev?

IIRC, something earlier in the call chain (possibly
 __netif_receive_skb_core()?) can change skb->dev to something other
 than the device that originally received the packet (orig_dev).  I
 think it's if the packet gets handled/transformed by a software
 netdevice (maybe a VLAN device?).
But really when I wrote ip_list_rcv() I just worked on the basis
 that "I don't know it can't change, so I shall assume it can".

HTH,
-ed

Re: ip_list_rcv() question

[Stephen Suryaputra]

On Wed, Oct 13, 2021 at 05:17:51PM +0100, Edward Cree wrote:
> On 07/10/2021 13:14, Stephen Suryaputra wrote:
> > Under what condition that ip_list_rcv() would restart the sublist, i.e.
> > that the skb in the list is having different skb->dev?
> 
> IIRC, something earlier in the call chain (possibly
>  __netif_receive_skb_core()?) can change skb->dev to something other
>  than the device that originally received the packet (orig_dev).  I
>  think it's if the packet gets handled/transformed by a software
>  netdevice (maybe a VLAN device?).
> But really when I wrote ip_list_rcv() I just worked on the basis
>  that "I don't know it can't change, so I shall assume it can".

I see now that vlan_do_receive() changes the skb->dev. It didn't occur
to me because __netif_receive_skb_list_core() has a similar logic for
dispatching sublists. But still the skb->dev could be different that
orig_dev.

Thanks.