Em Wed, Jul 01, 2020 at 08:21:13AM -0700, Alexei Starovoitov escreveu:
> On Wed, Jul 1, 2020 at 2:34 AM Daniel Borkmann <[email protected]> wrote:
> >
> > +1, I think augmenting mid-term would be the best given
> > check_sleepable_blacklist()
> > is rather a very fragile workaround^hack and it's also a generic lsm/sec
> > hooks issue
>
> I tried to make that crystal clear back in march during bpf virtual
> conference.
> imo whitelist is just as fragile as blacklist. Underlying
> implementation can change.
>
> > (at least for BPF_PROG_TYPE_LSM type & for the sake of documenting it for
> > other LSMs).
> > Perhaps there are function attributes that could be used and later
> > retrieved via BTF?
>
> Even if we convince gcc folks to add another function attribute it
> won't appear in dwarf.
Warning, hack ahead!
Perhaps we could do that with some sort of convention, i.e. define some
type and make a function returning that type to have the desired
attribute?
I.e.
typedef __attribute__foo__int int;
__attribute__foo__int function_bla(...)
{
}
?
> So we won't be able to convert it to BTF in pahole.
> Looking at how we failed to extend address_space() attribute to
> support existing __rcu
> and __user annotations I don't have high hopes of achieving annotations
> via compiler (either gcc or clang). So plan B is to engage with sparse folks
> and
> make it emit BTF with __rcu, __user and other annotations.
--
- Arnaldo
