[Cluster-devel] [PATCH dlm/next] fs: dlm: fix race in nodeid2con

[Alexander Aring <aahringo@redhat.com>]

This patch fixes a race in nodeid2con in cases that we parallel running
a lookup and both will create a connection structure for the same nodeid.
It's a rare case to create a new connection structure to keep reader
lockless we just do a lookup inside the protection area again and drop
previous work if this race happens.

Fixes: a47666eb763cc ("fs: dlm: make connection hash lockless")
Signed-off-by: Alexander Aring <aahringo@redhat.com>
---
 fs/dlm/lowcomms.c | 16 +++++++++++++++-
 1 file changed, 15 insertions(+), 1 deletion(-)

diff --git a/fs/dlm/lowcomms.c b/fs/dlm/lowcomms.c
index b7b7360be609e..79f56f16bc2ce 100644
--- a/fs/dlm/lowcomms.c
+++ b/fs/dlm/lowcomms.c
@@ -175,7 +175,7 @@ static struct connection *__find_con(int nodeid)
  */
 static struct connection *nodeid2con(int nodeid, gfp_t alloc)
 {
-	struct connection *con = NULL;
+	struct connection *con, *tmp;
 	int r;
 
 	con = __find_con(nodeid);
@@ -213,6 +213,20 @@ static struct connection *nodeid2con(int nodeid, gfp_t alloc)
 	r = nodeid_hash(nodeid);
 
 	spin_lock(&connections_lock);
+	/* Because multiple workqueues/threads calls this function it can
+	 * race on multiple cpu's. Instead of locking hot path __find_con()
+	 * we just check in rare cases of recently added nodes again
+	 * under protection of connections_lock. If this is the case we
+	 * abort our connection creation and return the existing connection.
+	 */
+	tmp = __find_con(nodeid);
+	if (tmp) {
+		spin_unlock(&connections_lock);
+		kfree(con->rx_buf);
+		kfree(con);
+		return tmp;
+	}
+
 	hlist_add_head_rcu(&con->list, &connection_hash[r]);
 	spin_unlock(&connections_lock);
 
-- 
2.26.2