skb header allocation

skb header allocation

[Chris Ross]

I have a network driver that acts as a Ethernet device and builds up a
series of outer headers on skb(s) it receives from upper layers. I am
currently using the technique that is in ipip.c to ensure I have
enough room to add my header ...

if (skb_headroom(skb) < some_value || skb_shared(skb) ||
         ((skb_cloned(skb) && !skb_clone_writable(skb, 0))))
   {
      if ((skb2 = skb_realloc_headroom(skb, some_value)) == NULL)
         return -1;

      dev_kfree_skb(skb);
      skb = skb2;
   }

Is this the best practice for a high bandwidth scenario?

thanks

-chris

Re: skb header allocation

[Stephen Hemminger]

On Tue, 25 Aug 2009 22:04:10 -0500
Chris Ross <chris@compilednetworks.com> wrote:

> I have a network driver that acts as a Ethernet device and builds up a
> series of outer headers on skb(s) it receives from upper layers. I am
> currently using the technique that is in ipip.c to ensure I have
> enough room to add my header ...
> 
> if (skb_headroom(skb) < some_value || skb_shared(skb) ||
>          ((skb_cloned(skb) && !skb_clone_writable(skb, 0))))
>    {
>       if ((skb2 = skb_realloc_headroom(skb, some_value)) == NULL)
>          return -1;
> 
>       dev_kfree_skb(skb);
>       skb = skb2;
>    }
> 
> Is this the best practice for a high bandwidth scenario?

Define a value of dev->hard_header_len that adds space for what you need.
Use skb_cow_head(skb, headroom) before touching skb header.

Re: skb header allocation

[Frank Blaschka]

Is it save to change dev->hard_header_len to a value other than ETH_HLEN for
an ethernet device?

I rememeber we run in trouble with some kernel components (netfilter?, I'm
not sure can't remember the details ...) when we did this is the past.

Is this not an issue any more?

Thanks

Stephen Hemminger schrieb:
> On Tue, 25 Aug 2009 22:04:10 -0500
> Chris Ross <chris@compilednetworks.com> wrote:
> 
>> I have a network driver that acts as a Ethernet device and builds up a
>> series of outer headers on skb(s) it receives from upper layers. I am
>> currently using the technique that is in ipip.c to ensure I have
>> enough room to add my header ...
>>
>> if (skb_headroom(skb) < some_value || skb_shared(skb) ||
>>          ((skb_cloned(skb) && !skb_clone_writable(skb, 0))))
>>    {
>>       if ((skb2 = skb_realloc_headroom(skb, some_value)) == NULL)
>>          return -1;
>>
>>       dev_kfree_skb(skb);
>>       skb = skb2;
>>    }
>>
>> Is this the best practice for a high bandwidth scenario?
> 
> Define a value of dev->hard_header_len that adds space for what you need.
> Use skb_cow_head(skb, headroom) before touching skb header.
> --
> To unsubscribe from this list: send the line "unsubscribe netdev" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 

Re: skb header allocation

[Stephen Hemminger]

On Wed, 26 Aug 2009 08:01:23 +0200
Frank Blaschka <blaschka@linux.vnet.ibm.com> wrote:

> Is it save to change dev->hard_header_len to a value other than ETH_HLEN for
> an ethernet device?
> 
> I rememeber we run in trouble with some kernel components (netfilter?, I'm
> not sure can't remember the details ...) when we did this is the past.
> 

You can add extra data but need a minimum with ETH_HLEN of space above
packet when receiving.


--