As explained in commit 51d7d5205d33 ("powerpc: Add smp_mb() to arch_spin_is_locked()", for some architectures the ACQUIRE during spin_lock only applies to loading the lock, not to storing the lock state.
nf_conntrack_lock() does not handle this correctly: /* 1) Acquire the lock */ spin_lock(lock); while (unlikely(nf_conntrack_locks_all)) { spin_unlock(lock); spinlock_store_acquire() is missing between spin_lock and reading nf_conntrack_locks_all. In addition, reading nf_conntrack_locks_all needs ACQUIRE memory ordering. 2nd, minor issue: If there would be many nf_conntrack_all_lock() callers, then nf_conntrack_lock() would loop forever. Therefore: Change nf_conntrack_lock and nf_conntract_lock_all() to the approach used by ipc/sem.c: - add spinlock_store_acquire() - add smp_load_acquire() - for nf_conntrack_lock, use spin_lock(&global_lock) instead of spin_unlock_wait(&global_lock) and loop backward. - use smp_store_mb() instead of a raw smp_mb() Signed-off-by: Manfred Spraul <manf...@colorfullife.com> Cc: Pablo Neira Ayuso <pa...@netfilter.org> Cc: netfilter-de...@vger.kernel.org --- Question: Should I split this patch? First a patch that uses smp_mb(), with Cc: stable. The replace the smp_mb() with spinlock_store_acquire, not for stable net/netfilter/nf_conntrack_core.c | 36 ++++++++++++++++++++++-------------- 1 file changed, 22 insertions(+), 14 deletions(-) diff --git a/net/netfilter/nf_conntrack_core.c b/net/netfilter/nf_conntrack_core.c index 7d90a5d..f840b0b 100644 --- a/net/netfilter/nf_conntrack_core.c +++ b/net/netfilter/nf_conntrack_core.c @@ -79,20 +79,29 @@ static __read_mostly bool nf_conntrack_locks_all; void nf_conntrack_lock(spinlock_t *lock) __acquires(lock) { + /* 1) Acquire the lock */ spin_lock(lock); - while (unlikely(nf_conntrack_locks_all)) { - spin_unlock(lock); - /* - * Order the 'nf_conntrack_locks_all' load vs. the - * spin_unlock_wait() loads below, to ensure - * that 'nf_conntrack_locks_all_lock' is indeed held: - */ - smp_rmb(); /* spin_lock(&nf_conntrack_locks_all_lock) */ - spin_unlock_wait(&nf_conntrack_locks_all_lock); - spin_lock(lock); - } + /* 2) Order storing the lock and reading nf_conntrack_locks_all */ + spinlock_store_acquire(); + + /* 3) read nf_conntrack_locks_all, with ACQUIRE semantics */ + if (likely(smp_load_acquire(&nf_conntrack_locks_all) == false)) + return; + + /* fast path failed, unlock */ + spin_unlock(lock); + + /* Slow path 1) get global lock */ + spin_lock(&nf_conntrack_locks_all_lock); + + /* Slow path 2) get the lock we want */ + spin_lock(lock); + + /* Slow path 3) release the global lock */ + spin_unlock(&nf_conntrack_locks_all_lock); } + EXPORT_SYMBOL_GPL(nf_conntrack_lock); static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2) @@ -132,15 +141,14 @@ static void nf_conntrack_all_lock(void) int i; spin_lock(&nf_conntrack_locks_all_lock); - nf_conntrack_locks_all = true; /* - * Order the above store of 'nf_conntrack_locks_all' against + * Order the store of 'nf_conntrack_locks_all' against * the spin_unlock_wait() loads below, such that if * nf_conntrack_lock() observes 'nf_conntrack_locks_all' * we must observe nf_conntrack_locks[] held: */ - smp_mb(); /* spin_lock(&nf_conntrack_locks_all_lock) */ + smp_store_mb(nf_conntrack_locks_all, true); for (i = 0; i < CONNTRACK_LOCKS; i++) { spin_unlock_wait(&nf_conntrack_locks[i]); -- 2.7.4