Re: [PATCH bpf-next v4 1/4] bpf: Prevent updating extended prog to prog_array map

[Eduard Zingerman <eddyz87@gmail.com>]

On Tue, 2024-10-01 at 21:20 +0800, Leon Hwang wrote:

[...]

> > > diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
> > > index a8f1808a1ca54..db17c52fa35db 100644
> > > --- a/kernel/bpf/syscall.c
> > > +++ b/kernel/bpf/syscall.c
> > > @@ -3212,14 +3212,23 @@ static void bpf_tracing_link_release(struct bpf_link *link)
> > >  {
> > >  	struct bpf_tracing_link *tr_link =
> > >  		container_of(link, struct bpf_tracing_link, link.link);
> > > -
> > > -	WARN_ON_ONCE(bpf_trampoline_unlink_prog(&tr_link->link,
> > > -						tr_link->trampoline));
> > > +	struct bpf_prog *tgt_prog = tr_link->tgt_prog;
> > > +
> > > +	if (link->prog->type == BPF_PROG_TYPE_EXT) {
> > > +		mutex_lock(&tgt_prog->aux->ext_mutex);
> > > +		WARN_ON_ONCE(bpf_trampoline_unlink_prog(&tr_link->link,
> > > +							tr_link->trampoline));
> > > +		tgt_prog->aux->is_extended = false;
> > 
> > In case if unlink fails is_extended should not be reset.
> > 
> 
> Nope.
> 
> In bpf_trampoline_unlink_prog(), 'tr->extension_prog = NULL;' always no
> matter whether fail to unlink.
> 
> So, it should reset is_extended always too.

Hm, you are correct, sorry for the noise.
It is unfortunate that these updates are separated in the code, tbh.

[...]