[PATCH net] amt: fix tunnel list corruption on device stop

amt_dev_stop() calls list_del_rcu() on each tunnel without holding
amt->lock. amt_tunnel_expire() does the same under amt->lock. When both
race on the same tunnel, the second list_del_rcu() hits LIST_POISON2 and
panics the kernel. An unprivileged user can trigger this inside its own
user/network namespace.

 list_del corruption, prev is LIST_POISON2 (dead000000000122)
 kernel BUG at lib/list_debug.c:59!
 RIP: 0010:__list_del_entry_valid_or_report+0x13a/0x200
 Call Trace:
  amt_dev_stop+0x2c3/0x500 (drivers/net/amt.c:3097)
  __dev_close_many+0x17e/0x470
  unregister_netdevice_many_notify+0x729/0x1f00

Fix:
1. Quiesce RX: clear sk_user_data and call synchronize_net() to ensure
   no RCU readers are traversing tunnel_list. This makes list_del_init()
   safe (it is not RCU-reader-safe unlike list_del_rcu()).
2. Hold amt->lock when unlinking tunnels in stop, using list_del_init()
   so amt_tunnel_expire() can detect already-claimed tunnels via
   list_empty() and skip them.
3. Use while/list_first_entry instead of list_for_each_entry_safe,
   because cancel_delayed_work_sync() can sleep and the cached next
   pointer may become stale.
4. Use disable_delayed_work_sync() instead of cancel_delayed_work_sync()
   to prevent amt_update_handler() from rearming the GC timer in a rare
   race where a packet arrives before the socket is fully released.

Fixes: cbc21dc1cfe9 ("amt: add data plane of amt interface")
Reported-by: [email protected]
Reported-by: Xiang Mei (Microsoft) <[email protected]>
Reported-by: Cen Zhang (Microsoft) <[email protected]>
Cc: [email protected]
Signed-off-by: Cen Zhang (Microsoft) <[email protected]>
---
 drivers/net/amt.c | 25 ++++++++++++++++++++-----
 1 file changed, 20 insertions(+), 5 deletions(-)

diff --git a/drivers/net/amt.c b/drivers/net/amt.c
index bddc24e1856d..a5db02d81291 100644
--- a/drivers/net/amt.c
+++ b/drivers/net/amt.c
@@ -1348,6 +1348,11 @@ static void amt_tunnel_expire(struct work_struct *work)
 	struct amt_dev *amt = tunnel->amt;
 
 	spin_lock_bh(&amt->lock);
+	/* amt_dev_stop() marks tunnels it owns with list_del_init(). */
+	if (list_empty(&tunnel->list)) {
+		spin_unlock_bh(&amt->lock);
+		return;
+	}
 	rcu_read_lock();
 	list_del_rcu(&tunnel->list);
 	amt->nr_tunnels--;
@@ -3068,7 +3073,7 @@ static int amt_dev_open(struct net_device *dev)
 static int amt_dev_stop(struct net_device *dev)
 {
 	struct amt_dev *amt = netdev_priv(dev);
-	struct amt_tunnel_list *tunnel, *tmp;
+	struct amt_tunnel_list *tunnel;
 	struct sk_buff *skb;
 	struct sock *sk;
 	int i;
@@ -3077,9 +3082,11 @@ static int amt_dev_stop(struct net_device *dev)
 	disable_delayed_work_sync(&amt->discovery_wq);
 	cancel_delayed_work_sync(&amt->secret_wq);
 
-	/* shutdown */
+	/* Quiesce RX path before tearing down tunnels. */
 	sk = rtnl_dereference(amt->sk);
 	RCU_INIT_POINTER(amt->sk, NULL);
+	if (sk)
+		rcu_assign_sk_user_data(sk, NULL);
 	synchronize_net();
 	if (sk)
 		udp_tunnel_sock_release(sk);
@@ -3097,13 +3104,21 @@ static int amt_dev_stop(struct net_device *dev)
 	amt->req_cnt = 0;
 	WRITE_ONCE(amt->remote_ip, 0);
 
-	list_for_each_entry_safe(tunnel, tmp, &amt->tunnel_list, list) {
-		list_del_rcu(&tunnel->list);
+	spin_lock_bh(&amt->lock);
+	while (!list_empty(&amt->tunnel_list)) {
+		tunnel = list_first_entry(&amt->tunnel_list,
+					  struct amt_tunnel_list, list);
+		list_del_init(&tunnel->list);
 		amt->nr_tunnels--;
-		cancel_delayed_work_sync(&tunnel->gc_wq);
+		spin_unlock_bh(&amt->lock);
+
+		disable_delayed_work_sync(&tunnel->gc_wq);
 		amt_clear_groups(tunnel);
 		kfree_rcu(tunnel, rcu);
+
+		spin_lock_bh(&amt->lock);
 	}
+	spin_unlock_bh(&amt->lock);
 
 	return 0;
 }
-- 
2.55.0