Re: [nft PATCH RFC] scanner: nat: Move to own scope

From: Florian Westphal <fw@strlen.de>
To: Phil Sutter <phil@nwl.cc>
Cc: Florian Westphal <fw@strlen.de>,
	Pablo Neira Ayuso <pablo@netfilter.org>,
	netfilter-devel@vger.kernel.org
Subject: Re: [nft PATCH RFC] scanner: nat: Move to own scope
Date: Mon, 9 Aug 2021 17:18:33 +0200	[thread overview]
Message-ID: <20210809151833.GM607@breakpoint.cc> (raw)
In-Reply-To: <20210809140141.18976-1-phil@nwl.cc>

Phil Sutter <phil@nwl.cc> wrote:
> Unify nat, masquerade and redirect statements, they widely share their
> syntax.
> This seemingly valid change breaks the parser with this rule:
> 
> | snat ip prefix to ip saddr map { 10.141.11.0/24 : 192.168.2.0/24 }

Yes.

> Problem is that 'prefix' is not in SC_IP and close_scope_ip called from
> parser_bison.y:5067 is not sufficient. I assumed explicit scope closing
> would eliminate this lookahead problem. Did I find a proof against the
> concept or is there a bug in my patch?

You have to keep 'prefix' in the global scope.
What should work as well is to permit 'prefix' from SCANSTATE_IP(6).

The problem is that the parser can't close the new 'IP' scope until
it has enough tokens available to match a complete bison rule.

So, it is in IP scope, sees 'prefix' (which will be STRING as the
PREFIX scan rule is off) and that ends up in a parser error due to lack
of a 'IP STRING' rule.