Re: [PATCH v2] keys: Fix UAF in key_put()

[Oleg Nesterov <oleg@redhat.com>]

On 03/19, David Howells wrote:
>
> --- a/security/keys/gc.c
> +++ b/security/keys/gc.c
> @@ -218,8 +218,10 @@ static void key_garbage_collector(struct work_struct *work)
>  		key = rb_entry(cursor, struct key, serial_node);
>  		cursor = rb_next(cursor);
>  
> -		if (refcount_read(&key->usage) == 0)
> +		if (test_bit(KEY_FLAG_FINAL_PUT, &key->flags)) {
> +			smp_mb(); /* Clobber key->user after FINAL_PUT seen. */
>  			goto found_unreferenced_key;
> +		}
>  
>  		if (unlikely(gc_state & KEY_GC_REAPING_DEAD_1)) {
>  			if (key->type == key_gc_dead_keytype) {
> diff --git a/security/keys/key.c b/security/keys/key.c
> index 3d7d185019d3..7198cd2ac3a3 100644
> --- a/security/keys/key.c
> +++ b/security/keys/key.c
> @@ -658,6 +658,8 @@ void key_put(struct key *key)
>  				key->user->qnbytes -= key->quotalen;
>  				spin_unlock_irqrestore(&key->user->lock, flags);
>  			}
> +			smp_mb(); /* key->user before FINAL_PUT set. */

Can't resist, smp_mb__before_atomic() should equally work,
but this doesn't really matter, please forget.

> +			set_bit(KEY_FLAG_FINAL_PUT, &key->flags);
>  			schedule_work(&key_gc_work);

I believe this patch is correct,

Reviewed-by: Oleg Nesterov <oleg@redhat.com>