Re: [Patch bpf] bpf: fix a potential use-after-free in bpf_link_free()

[Cong Wang <xiyou.wangcong@gmail.com>]

On Mon, Jun 03, 2024 at 06:15:55PM +0200, Daniel Borkmann wrote:
> On 6/2/24 9:32 PM, Jiri Olsa wrote:
> > On Sun, Jun 02, 2024 at 11:27:03AM -0700, Cong Wang wrote:
> > > From: Cong Wang <cong.wang@bytedance.com>
> > > 
> > > After commit 1a80dbcb2dba, bpf_link can be freed by
> > > link->ops->dealloc_deferred, but the code still tests and uses
> > > link->ops->dealloc afterward, which leads to a use-after-free as
> > > reported by syzbot. Actually, one of them should be sufficient, so
> > > just call one of them instead of both. Also add a WARN_ON() in case
> > > of any problematic implementation.
> > > 
> > > Reported-by: syzbot+1989ee16d94720836244@syzkaller.appspotmail.com
> > > Fixes: 1a80dbcb2dba ("bpf: support deferring bpf_link dealloc to after RCU grace period")
> > > Cc: Andrii Nakryiko <andrii@kernel.org>
> > > Signed-off-by: Cong Wang <cong.wang@bytedance.com>
> > > ---
> > >   kernel/bpf/syscall.c | 4 ++--
> > >   1 file changed, 2 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
> > > index 2222c3ff88e7..d8f244069495 100644
> > > --- a/kernel/bpf/syscall.c
> > > +++ b/kernel/bpf/syscall.c
> > > @@ -2998,6 +2998,7 @@ static int bpf_obj_get(const union bpf_attr *attr)
> > >   void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
> > >   		   const struct bpf_link_ops *ops, struct bpf_prog *prog)
> > >   {
> > > +	WARN_ON(ops->dealloc && ops->dealloc_deferred);
> > >   	atomic64_set(&link->refcnt, 1);
> > >   	link->type = type;
> > >   	link->id = 0;
> > > @@ -3074,8 +3075,7 @@ static void bpf_link_free(struct bpf_link *link)
> > >   			call_rcu_tasks_trace(&link->rcu, bpf_link_defer_dealloc_mult_rcu_gp);
> > >   		else
> > >   			call_rcu(&link->rcu, bpf_link_defer_dealloc_rcu_gp);
> > > -	}
> > > -	if (link->ops->dealloc)
> > > +	} else if (link->ops->dealloc)
> > >   		link->ops->dealloc(link);
> > 
> > nice catch
> 
> +1, thanks Cong !
> 
> > Acked-by: Jiri Olsa <jolsa@kernel.org>
> 
> I think it would also be slightly nicer to just fetch the ops once, which
> wouldn't have caused the issue if it was done back then in the first place.
> Do you mind if I squash this in and then apply it to bpf tree? Looks as
> follows :
> 

Sounds good to me.

Thanks.