Re: [PATCH] selinux: convert struct sidtab count to refcount_t

[Jann Horn <jannh@google.com>]

On Mon, Jul 22, 2019 at 3:44 PM Ondrej Mosnacek <omosnace@redhat.com> wrote:
> On Mon, Jul 22, 2019 at 1:35 PM NitinGote <nitin.r.gote@intel.com> 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.
> >
> > Signed-off-by: NitinGote <nitin.r.gote@intel.com>
>
> Nack.
>
> The 'count' variable is not used as a reference counter here. It
> tracks the number of entries in sidtab, which is a very specific
> lookup table that can only grow (the count never decreases). I only
> made it atomic because the variable is read outside of the sidtab's
> spin lock and thus the reads and writes to it need to be guaranteed to
> be atomic. The counter is only updated under the spin lock, so
> insertions do not race with each other.

Probably shouldn't even be atomic_t... quoting Documentation/atomic_t.txt:

| SEMANTICS
| ---------
|
| Non-RMW ops:
|
| The non-RMW ops are (typically) regular LOADs and STOREs and are canonically
| implemented using READ_ONCE(), WRITE_ONCE(), smp_load_acquire() and
| smp_store_release() respectively. Therefore, if you find yourself only using
| the Non-RMW operations of atomic_t, you do not in fact need atomic_t at all
| and are doing it wrong.

So I think what you actually want here is a plain "int count", and then:
 - for unlocked reads, either READ_ONCE()+smp_rmb() or smp_load_acquire()
 - for writes, either smp_wmb()+WRITE_ONCE() or smp_store_release()

smp_load_acquire() and smp_store_release() are probably the nicest
here, since they are semantically clearer than smp_rmb()/smp_wmb().