On Tue, May 24, 2016 at 04:27:26PM +0200, Peter Zijlstra wrote:
> nf_conntrack_lock{,_all}() is borken as it misses a bunch of memory
> barriers to order the whole global vs local locks scheme.
>
> Even x86 (and other TSO archs) are affected.
>
> Signed-off-by: Peter Zijlstra (Intel) <pet...@infradead.org>
> ---
> net/netfilter/nf_conntrack_core.c | 30 +++++++++++++++++++++++++++++-
> 1 file changed, 29 insertions(+), 1 deletion(-)
>
> --- a/net/netfilter/nf_conntrack_core.c
> +++ b/net/netfilter/nf_conntrack_core.c
> @@ -74,7 +74,18 @@ void nf_conntrack_lock(spinlock_t *lock)
> spin_lock(lock);
> while (unlikely(nf_conntrack_locks_all)) {
And note that we can replace nf_conntrack_locks_all with
spin_is_locked(nf_conntrack_locks_all_lock), since that is the exact
same state.
But I didn't want to do too much in one go.
> spin_unlock(lock);
> + /*
> + * Order the nf_contrack_locks_all load vs the
> spin_unlock_wait()
> + * loads below, to ensure locks_all is indeed held.
> + */
> + smp_rmb(); /* spin_lock(locks_all) */
> spin_unlock_wait(&nf_conntrack_locks_all_lock);
> + /*
> + * The control dependency's LOAD->STORE order is enough to
> + * guarantee the spin_lock() is ordered after the above
> + * unlock_wait(). And the ACQUIRE of the lock ensures we are
> + * fully ordered against the spin_unlock() of locks_all.
> + */
> spin_lock(lock);
> }
> }
