RDS: Fix locking in rds_send_drop_to()
It seems rds_send_drop_to() called
__rds_rdma_send_complete(rs, rm, RDS_RDMA_CANCELED)
with only rds_sock lock, but not rds_message lock. It raced with
other threads that is attempting to modify the rds_message as well,
such as from within rds_rdma_send_complete().
Signed-off-by: Tina Yang <tina.yang@oracle.com>
Signed-off-by: Andy Grover <andy.grover@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/net/rds/send.c b/net/rds/send.c
index 1a2ef24..192a480 100644
--- a/net/rds/send.c
+++ b/net/rds/send.c
@@ -632,9 +632,6 @@
list_move(&rm->m_sock_item, &list);
rds_send_sndbuf_remove(rs, rm);
clear_bit(RDS_MSG_ON_SOCK, &rm->m_flags);
-
- /* If this is a RDMA operation, notify the app. */
- __rds_rdma_send_complete(rs, rm, RDS_RDMA_CANCELED);
}
/* order flag updates with the rs lock */
@@ -643,9 +640,6 @@
spin_unlock_irqrestore(&rs->rs_lock, flags);
- if (wake)
- rds_wake_sk_sleep(rs);
-
conn = NULL;
/* now remove the messages from the conn list as needed */
@@ -653,6 +647,10 @@
/* We do this here rather than in the loop above, so that
* we don't have to nest m_rs_lock under rs->rs_lock */
spin_lock_irqsave(&rm->m_rs_lock, flags2);
+ /* If this is a RDMA operation, notify the app. */
+ spin_lock(&rs->rs_lock);
+ __rds_rdma_send_complete(rs, rm, RDS_RDMA_CANCELED);
+ spin_unlock(&rs->rs_lock);
rm->m_rs = NULL;
spin_unlock_irqrestore(&rm->m_rs_lock, flags2);
@@ -681,6 +679,9 @@
if (conn)
spin_unlock_irqrestore(&conn->c_lock, flags);
+ if (wake)
+ rds_wake_sk_sleep(rs);
+
while (!list_empty(&list)) {
rm = list_entry(list.next, struct rds_message, m_sock_item);
list_del_init(&rm->m_sock_item);