sunrpc/cache: drop reference when sunrpc_cache_pipe_upcall() detects a race
authorNeilBrown <neilb@suse.com>
Fri, 4 Mar 2016 06:20:13 +0000 (17:20 +1100)
committerJ. Bruce Fields <bfields@redhat.com>
Thu, 17 Mar 2016 18:57:15 +0000 (14:57 -0400)
sunrpc_cache_pipe_upcall() can detect a race if CACHE_PENDING is no longer
set.  In this case it aborts the queuing of the upcall.
However it has already taken a new counted reference on "h" and
doesn't "put" it, even though it frees the data structure holding the reference.

So let's delay the "cache_get" until we know we need it.

Fixes: f9e1aedc6c79 ("sunrpc/cache: remove races with queuing an upcall.")
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
net/sunrpc/cache.c

index 273bc3a35425ba99351eedb3f99244f0f9c07ab7..008c25d1b9f9c8c955e0da6fe631a2207966f2bb 100644 (file)
@@ -1182,14 +1182,14 @@ int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
        }
 
        crq->q.reader = 0;
-       crq->item = cache_get(h);
        crq->buf = buf;
        crq->len = 0;
        crq->readers = 0;
        spin_lock(&queue_lock);
-       if (test_bit(CACHE_PENDING, &h->flags))
+       if (test_bit(CACHE_PENDING, &h->flags)) {
+               crq->item = cache_get(h);
                list_add_tail(&crq->q.list, &detail->queue);
-       else
+       else
                /* Lost a race, no longer PENDING, so don't enqueue */
                ret = -EAGAIN;
        spin_unlock(&queue_lock);