Re: [PATCH 1/2 nf-next] netfilter: nf_tables: Introduce stateful object update operation

[Florian Westphal <fw@strlen.de>]

Fernando Fernandez Mancera <ffmancera@riseup.net> wrote:
> This patch adds the infrastructure needed for the stateful object update
> support.
> 
> Signed-off-by: Fernando Fernandez Mancera <ffmancera@riseup.net>
> ---
>  include/net/netfilter/nf_tables.h |  6 +++
>  net/netfilter/nf_tables_api.c     | 71 ++++++++++++++++++++++++++++---
>  2 files changed, 70 insertions(+), 7 deletions(-)
> 
> diff --git a/include/net/netfilter/nf_tables.h b/include/net/netfilter/nf_tables.h
> index dc301e3d6739..dc4e32040ea9 100644
> --- a/include/net/netfilter/nf_tables.h
> +++ b/include/net/netfilter/nf_tables.h
> @@ -1123,6 +1123,9 @@ struct nft_object_ops {
>  	int				(*dump)(struct sk_buff *skb,
>  						struct nft_object *obj,
>  						bool reset);
> +	int				(*update)(const struct nft_ctx *ctx,
> +						  const struct nlattr *const tb[],
> +						  struct nft_object *obj);

maybe adda 'bool commit' argument here.

> +	err = obj->ops->update(ctx, (const struct nlattr * const *)tb, obj);

Then, set it to 'false' here.
You would have to keep 'tb' allocated and place it on the 'trans'
object.

> +	nft_trans_obj_update(trans) = true;

	nft_trans_obj_update_tb(trans) = tb;

> -			nft_clear(net, nft_trans_obj(trans));
> -			nf_tables_obj_notify(&trans->ctx, nft_trans_obj(trans),
> -					     NFT_MSG_NEWOBJ);
> -			nft_trans_destroy(trans);
> +			if (nft_trans_obj_update(trans)) {

				nft_trans_obj(trans)->ops->update(&trans->ctx,
					      nft_trans_obj_update_tb(trans),
					      nft_trans_obj(trans),
					      true);

				kfree(nft_trans_obj_update_tb(trans));


Because otherwise we will update objects while we're not yet sure that
we can process/handle the entire batch.

I think we should, if possible, only update once we've made it to
the commit phase.