These changes are a raw update to a vanilla kernel 4.1.10, with the

[kvmfornfv.git] / kernel / net / core / request_sock.c

diff --git a/kernel/net/core/request_sock.c b/kernel/net/core/request_sock.c
index 87b22c0..b42f0e2 100644 (file)
--- a/kernel/net/core/request_sock.c
+++ b/kernel/net/core/request_sock.c
@@ -103,10 +103,16 @@ void reqsk_queue_destroy(struct request_sock_queue *queue)
                        spin_lock_bh(&queue->syn_wait_lock);
                        while ((req = lopt->syn_table[i]) != NULL) {
                                lopt->syn_table[i] = req->dl_next;
+                               /* Because of following del_timer_sync(),
+                                * we must release the spinlock here
+                                * or risk a dead lock.
+                                */
+                               spin_unlock_bh(&queue->syn_wait_lock);
                                atomic_inc(&lopt->qlen_dec);
-                               if (del_timer(&req->rsk_timer))
+                               if (del_timer_sync(&req->rsk_timer))
                                        reqsk_put(req);
                                reqsk_put(req);
+                               spin_lock_bh(&queue->syn_wait_lock);
                        }
                        spin_unlock_bh(&queue->syn_wait_lock);
                }