Re: [nft PATCH 0/2] A bit of non-constant binop follow-up

[Phil Sutter <phil@nwl.cc>]

On Fri, Apr 03, 2026 at 04:53:14PM +0100, Jeremy Sowden wrote:
> On 2026-04-02, at 20:43:18 +0200, Phil Sutter wrote:
> > When asked about how to translate ebtables' --arp-gratuitous match, I
> > noticed that basically everything is there already but the parser
> > rejects it.
> > 
> > While we can't do a simple 'arp saddr ip == arp daddr ip' because cmp
> > expression requires for one side of the equation to be constant, using
> > XOR on LHS we can work around this limitation:
> > 
> > arp saddr ip ^ arp daddr ip == 0.0.0.0
> > 
> > Thanks to Jeremy's work on bitwise expression (which one might want to
> > repeat for cmp),
> 
> I'll take a look. :)

Cool, thanks! The ability for cmp to operate on two registers instead of
one register and payload ("data reg") would allow user space to
implement the above as 'arp saddr ip == arp daddr ip' (without the need
for an internal conversion into XOR). It is not a short-term solution
though due to the needed kernel support.

Cheers, Phil