cfq-iosched: fix RCU problem in cfq_cic_lookup()
cfq_cic_lookup() needs to properly protect ioc->ioc_data before
dereferencing it and also exclude updaters of ioc->ioc_data as well.
Also add a number of comments documenting why the existing RCU usage
is OK.
Thanks a lot to "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> for
review and comments!
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
diff --git a/block/cfq-iosched.c b/block/cfq-iosched.c
index 4df3f052..d01b411 100644
--- a/block/cfq-iosched.c
+++ b/block/cfq-iosched.c
@@ -1142,6 +1142,9 @@
kmem_cache_free(cfq_pool, cfqq);
}
+/*
+ * Must always be called with the rcu_read_lock() held
+ */
static void
__call_for_each_cic(struct io_context *ioc,
void (*func)(struct io_context *, struct cfq_io_context *))
@@ -1197,6 +1200,11 @@
cfq_cic_free(cic);
}
+/*
+ * Must be called with rcu_read_lock() held or preemption otherwise disabled.
+ * Only two callers of this - ->dtor() which is called with the rcu_read_lock(),
+ * and ->trim() which is called with the task lock held
+ */
static void cfq_free_io_context(struct io_context *ioc)
{
/*
@@ -1502,20 +1510,24 @@
cfq_cic_lookup(struct cfq_data *cfqd, struct io_context *ioc)
{
struct cfq_io_context *cic;
+ unsigned long flags;
void *k;
if (unlikely(!ioc))
return NULL;
+ rcu_read_lock();
+
/*
* we maintain a last-hit cache, to avoid browsing over the tree
*/
cic = rcu_dereference(ioc->ioc_data);
- if (cic && cic->key == cfqd)
+ if (cic && cic->key == cfqd) {
+ rcu_read_unlock();
return cic;
+ }
do {
- rcu_read_lock();
cic = radix_tree_lookup(&ioc->radix_root, (unsigned long) cfqd);
rcu_read_unlock();
if (!cic)
@@ -1524,10 +1536,13 @@
k = cic->key;
if (unlikely(!k)) {
cfq_drop_dead_cic(cfqd, ioc, cic);
+ rcu_read_lock();
continue;
}
+ spin_lock_irqsave(&ioc->lock, flags);
rcu_assign_pointer(ioc->ioc_data, cic);
+ spin_unlock_irqrestore(&ioc->lock, flags);
break;
} while (1);
@@ -2134,6 +2149,10 @@
static void cfq_slab_kill(void)
{
+ /*
+ * Caller already ensured that pending RCU callbacks are completed,
+ * so we should have no busy allocations at this point.
+ */
if (cfq_pool)
kmem_cache_destroy(cfq_pool);
if (cfq_ioc_pool)
@@ -2292,6 +2311,11 @@
ioc_gone = &all_gone;
/* ioc_gone's update must be visible before reading ioc_count */
smp_wmb();
+
+ /*
+ * this also protects us from entering cfq_slab_kill() with
+ * pending RCU callbacks
+ */
if (elv_ioc_count_read(ioc_count))
wait_for_completion(ioc_gone);
cfq_slab_kill();