On 9 April 2013 14:28, Antonio Quartulli ordex@autistici.org wrote:

Hi Mihail,

On Tue, Apr 09, 2013 at 01:47:52PM +0300, Mihail Costea wrote:

...

Hi,

In the file distributed-arp-table.c, function batadv_dat_snoop_incoming_arp_request, after creation of ARP reply (skb_new = arp_create(...)) and testing if the creation was successful, shouldn't exist a call to skb_reset_mac_header(skb_new), just like in batadv_dat_snoop_outgoing_arp_request?

I've done something similar for NDP and I think that this is the line that might generate a kernel panic (not sure yet, I might have to do more testing).

As I said, I'm not sure yet, and maybe my code might generate it. But my question is why isn't here a skb_reset_mac_header as it is in the first function.

skb_reset_mac_header() is invoked in batadv_send_skb_packet() right before accessing the ethernet header. This is where DAT needs it, no further accesses to the Ethernet header are done on its side.

I'd say that invocation at distributed-arp-table.c:843 can probably be avoided (and there are other similar cases in the rest of the code...)

skb_reset_mac_header() has to be invoked at the right moment in order to correctly set the skb->mac_header pointer. In batman-adv we do this when skb->data points at the beginning of the batman-adv header when we need to access the Ethernet header for some reasons.

Be aware that some of the skb_* functions do invoke skb_reset_mac_header() on their own (e.g. eth_type_trans).

Cheers,

-- Antonio Quartulli

..each of us alone is worth nothing.. Ernesto "Che" Guevara

Understood. Thanks.

-- Mihail Costea