Re: [PATCH 07/17] net: convert sock.sk_refcnt from atomic_t to refcount_t

From: Eric Dumazet
Date: Thu Mar 16 2017 - 12:59:16 EST


On Thu, 2017-03-16 at 17:28 +0200, Elena Reshetova wrote:
> refcount_t type and corresponding API should be
> used instead of atomic_t when the variable is used as
> a reference counter. This allows to avoid accidental
> refcounter overflows that might lead to use-after-free
> situations.


...

> static __always_inline void sock_hold(struct sock *sk)
> {
> - atomic_inc(&sk->sk_refcnt);
> + refcount_inc(&sk->sk_refcnt);
> }
>

While I certainly see the value of these refcount_t, we have a very
different behavior on these atomic_inc() which were doing a single
inlined LOCK RMW on x86.

We now call an external function performing a
atomic_read(), various ops/tests, then atomic_cmpxchg_relaxed(), in a
loop, loosing the nice ability for x86 of preventing live locks.

Looks a lot of bloat, just to be able to chase hypothetical bugs in the
kernel.

I would love to have a way to enable extra debugging when I want a debug
kernel, like LOCKDEP or KASAN.

By adding all this bloat, we assert linux kernel is terminally buggy and
every atomic_inc() we did was suspicious, and need to be always
instrumented/validated.