Re: [nft PATCH RFC v2 3/4] Implement --echo option

[Phil Sutter <phil@nwl.cc>]

On Thu, Jul 27, 2017 at 06:56:47PM +0200, Pablo Neira Ayuso wrote:
> On Thu, Jul 27, 2017 at 06:54:35PM +0200, Phil Sutter wrote:
> > On Thu, Jul 27, 2017 at 06:43:00PM +0200, Pablo Neira Ayuso wrote:
> > > On Thu, Jul 27, 2017 at 06:21:43PM +0200, Phil Sutter wrote:
> > > > diff --git a/include/linux/netfilter/nf_tables.h b/include/linux/netfilter/nf_tables.h
> > > > index 40096de04e963..70404f724808a 100644
> > > > --- a/include/linux/netfilter/nf_tables.h
> > > > +++ b/include/linux/netfilter/nf_tables.h
> > > > @@ -114,6 +114,7 @@ enum nf_tables_msg_types {
> > > >  	NFT_MSG_GETOBJ,
> > > >  	NFT_MSG_DELOBJ,
> > > >  	NFT_MSG_GETOBJ_RESET,
> > > > +	NFT_MSG_META_ECHO,
> > > 
> > > Why this?
> > 
> > See netlink_echo_callback(): I want to reuse monitor code as much as
> > possible, and therefore introduced that function as common callback
> > dispatching to monitor callbacks. In order to make them print data like
> > I want them to, I need to have an indicator. The simplest way was to
> > introduce this fake type which in callbacks acts as third option to
> > NFT_MSG_NEW* or NFT_MSG_DEL*.
> 
> Hm, there must be a way to reuse this code without adding a new
> message type. From a netlink perspective, users should just set on the
> NLM_ECHO flag to get echo reports. I don't want to have a new message
> type for this.

Ah, maybe you misinterpreted it's scope: I gave it this weird name
because that message type actually doesn't exist from kernel PoV. It's
solely used in between netlink_echo_callback() and the
netlink_events_*_cb() functions it calls. It's pretty much just a hack,
but allowed me to get by without bigger changes to the callback
functions shared with monitor code.

Cheers, Phil