Re: [Cocci] Whitelist function in rule

[Paul Chaignon <paul@cilium.io>]

On Tue, Jun 02, 2020 at 02:55:23PM +0200, Julia Lawall wrote:
> 
> 
> On Tue, 2 Jun 2020, Paul Chaignon wrote:
> 
> > On Tue, Jun 02, 2020 at 12:19:50PM +0200, Julia Lawall wrote:
> > >
> > >
> > > On Tue, 2 Jun 2020, Paul Chaignon wrote:
> > >
> > > > Hi all,
> > > >
> > > > I'm trying to find calls to ep_tail_call() that are *not* followed by
> > > > 'return DROP_MISSED_TAIL_CALL'. That works fine (rule1 below), but I then
> > > > want to whitelist one specific function, send_drop_notify().
> > > >
> > > > I couldn't find a way to do that from the first rule, so I whitelisted
> > > > send_drop_notify() with a second rule (rule2 below). That seems rather
> > > > cumbersome and I'm wondering if there's maybe a simpler approach?
> > > >
> > > >   @rule1@
> > > >   position p1;
> > >
> > > In coccinelle/tests/python_poscon.cocci you have an example of how to
> > > attach python code to a metavariable:
> > >
> > > @initialize:python@
> > > @@
> > >
> > > def past_line_4(p, other):
> > >     return int(p[0].line) > 4
> > >
> > > @r@
> > > expression e;
> > > @@
> > >
> > > f(e)
> > >
> > > @@
> > > position p : script:python(r.e) { past_line_4(p, e) };
> > > expression r.e;
> > > @@
> > >
> > > g(
> > > -e@p
> > > +27
> > >  )
> > >
> > > In your case, I think that in the python code you should be able to access
> > > p1[0].current_element.  You can check that that is different than the
> > > name of the function that you want to ignore.  There are some limitations
> > > on what can be put in the {} in such a metavariable declaration, so if you
> > > have problems you can make a function, as shown in the example.  That code
> > > should return true if you want the match to take place and false
> > > otherwise.
> >
> > Thanks! That worked great!
> > And the Python filtering looks super useful; I'm likely going to be
> > using that a lot :-)
> 
> You can refer to other metavariables, as illustrated by the example in
> tests, but they have to be nearby, ie no ... between them, and there
> should be no disjunctions involved.  This is to ensure that the matchig
> that led to th metavariable binding is deterministic.  If you violate
> these constraints, Coccinelle will complain, and you will have to make a
> separate python rule in the normal way.

I see. I often need to whitelist simple cases like the above, so
definitely useful despite limitations.

Thanks again for your help!

Paul

> 
> julia
> 
> 
> >
> > Paul
> >
> > >
> > > julia
> > >
> > >
> > >
> > > >   @@
> > > >
> > > >   (
> > > >     ep_tail_call(...);
> > > >     ... when forall
> > > >     return DROP_MISSED_TAIL_CALL;
> > > >   |
> > > >     ep_tail_call@p1(...);
> > > >   )
> > > >
> > > >   @rule2@
> > > >   position p2, rule1.p1;
> > > >   identifier fn != send_drop_notify;
> > > >   @@
> > > >
> > > >   fn(...) {
> > > >   <+...
> > > >   ep_tail_call@p1@p2(...);
> > > >   ...+>
> > > >   }
> > > >
> > > > When I tried using rule1 as the body of the function in rule2, it just
> > > > didn't match anything anymore.
> > > >
> > > > Thanks,
> > > > Paul
> > > > _______________________________________________
> > > > Cocci mailing list
> > > > Cocci@systeme.lip6.fr
> > > > https://systeme.lip6.fr/mailman/listinfo/cocci
> > > >
> >
_______________________________________________
Cocci mailing list
Cocci@systeme.lip6.fr
https://systeme.lip6.fr/mailman/listinfo/cocci