Re: [PATCH] netfilter: use krealloc() in nf_conntrack_extend.c V2

[Patrick McHardy <kaber@trash.net>]

Pekka J Enberg wrote:
> The ksize() API is going away because it is being abused and it doesn't even
> work consistenly across different allocators. Therefore, convert
> net/netfilter/nf_conntrack_extend.c to use krealloc().
>
> Cc: <netfilter-devel@vger.kernel.org>
> Cc: <netdev@vger.kernel.org>
> Cc: Matt Mackall <mpm@selenic.com>
> Cc: Christoph Lameter <clameter@sgi.com>
> Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
> ---
> Patrick, please use this patch instead. The previous one did the moving
> unconditionally which is wrong. This one moves entries around only if
> krealloc() allocated a new buffer.
>
>  net/netfilter/nf_conntrack_extend.c |   10 ++++------
>  1 file changed, 4 insertions(+), 6 deletions(-)
>
> Index: slab-2.6/net/netfilter/nf_conntrack_extend.c
> ===================================================================
> --- slab-2.6.orig/net/netfilter/nf_conntrack_extend.c	2008-05-22 21:52:12.000000000 +0300
> +++ slab-2.6/net/netfilter/nf_conntrack_extend.c	2008-05-22 21:52:26.000000000 +0300
> @@ -88,13 +88,11 @@
>  	newlen = newoff + t->len;
>  	rcu_read_unlock();
>  
> -	if (newlen >= ksize(ct->ext)) {
> -		new = kmalloc(newlen, gfp);
> -		if (!new)
> -			return NULL;
> -
> -		memcpy(new, ct->ext, ct->ext->len);
> +	new = krealloc(ct->ext, newlen, gfp);

Unfortunately this means we'll always have to reallocate, even
if there's still some room left from the previous allocation.
Any chance to avoid that?