Re: [B.A.T.M.A.N.] [RFC] batman-adv: dirty hack to recompute mac_len in the rx path

[Simon Wunderlich]

Hello Marek,

On Tue, Oct 02, 2012 at 01:16:40PM +0800, Marek Lindner wrote:
> On Tuesday, October 02, 2012 04:53:28 Simon Wunderlich wrote:
> > +#define skb_share_check(skb, b) \
> > +       skb_share_check(skb, b); \
> > +       if (skb) \
> > +               skb_reset_mac_len(skb)
> > +
> > +#endif /* < KERNEL_VERSION(3, 8, 0) */
> 
> Has this patch been tested ? Our skb_share_check() call is this:
> skb = skb_share_check(skb, GFP_ATOMIC);
> 
> Now we replace this function call with 2 function calls and 2 return values ?

I have not tested in a real machine, but only the first function will
return the skb. The second part is a separate statement. I've checked
it with gcc -E (preprocessor only), these lines in main.c will expand to:

 hard_iface = ({ const typeof( ((struct batadv_hard_iface 
*)0)->batman_adv_ptype ) *__mptr = (ptype); (struct batadv_hard_iface *)( (char 
*)__mptr - __builtin_offsetof(struct batadv_hard_iface,batman_adv_ptype) );})
                       ;
 skb = skb_share_check(skb, ((( gfp_t)0x20u))); if (skb) skb_reset_mac_len(skb);


 if (!skb)
  goto err_out;

As you can see, the skb = skb_share_check() statement stays at it is, and
the reset_mac_len() is done afterwards.

Cheers,
        Simon

signature.asc
Description: Digital signature