l2tp: don't use l2tp_tunnel_find() in l2tp_ip and l2tp_ip6
authorGuillaume Nault <g.nault@alphalink.fr>
Fri, 3 Nov 2017 15:49:00 +0000 (16:49 +0100)
committerDavid S. Miller <davem@davemloft.net>
Sun, 5 Nov 2017 13:22:15 +0000 (22:22 +0900)
Using l2tp_tunnel_find() in l2tp_ip_recv() is wrong for two reasons:

  * It doesn't take a reference on the returned tunnel, which makes the
    call racy wrt. concurrent tunnel deletion.

  * The lookup is only based on the tunnel identifier, so it can return
    a tunnel that doesn't match the packet's addresses or protocol.

For example, a packet sent to an L2TPv3 over IPv6 tunnel can be
delivered to an L2TPv2 over UDPv4 tunnel. This is worse than a simple
cross-talk: when delivering the packet to an L2TP over UDP tunnel, the
corresponding socket is UDP, where ->sk_backlog_rcv() is NULL. Calling
sk_receive_skb() will then crash the kernel by trying to execute this
callback.

And l2tp_tunnel_find() isn't even needed here. __l2tp_ip_bind_lookup()
properly checks the socket binding and connection settings. It was used
as a fallback mechanism for finding tunnels that didn't have their data
path registered yet. But it's not limited to this case and can be used
to replace l2tp_tunnel_find() in the general case.

Fix l2tp_ip6 in the same way.

Fixes: 0d76751fad77 ("l2tp: Add L2TPv3 IP encapsulation (no UDP) support")
Fixes: a32e0eec7042 ("l2tp: introduce L2TPv3 IP encapsulation support for IPv6")
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/l2tp/l2tp_ip.c
net/l2tp/l2tp_ip6.c

index 4d322c1..e4280b6 100644 (file)
@@ -123,6 +123,7 @@ static int l2tp_ip_recv(struct sk_buff *skb)
        unsigned char *ptr, *optr;
        struct l2tp_session *session;
        struct l2tp_tunnel *tunnel = NULL;
+       struct iphdr *iph;
        int length;
 
        if (!pskb_may_pull(skb, 4))
@@ -178,24 +179,17 @@ pass_up:
                goto discard;
 
        tunnel_id = ntohl(*(__be32 *) &skb->data[4]);
-       tunnel = l2tp_tunnel_find(net, tunnel_id);
-       if (tunnel) {
-               sk = tunnel->sock;
-               sock_hold(sk);
-       } else {
-               struct iphdr *iph = (struct iphdr *) skb_network_header(skb);
-
-               read_lock_bh(&l2tp_ip_lock);
-               sk = __l2tp_ip_bind_lookup(net, iph->daddr, iph->saddr,
-                                          inet_iif(skb), tunnel_id);
-               if (!sk) {
-                       read_unlock_bh(&l2tp_ip_lock);
-                       goto discard;
-               }
+       iph = (struct iphdr *)skb_network_header(skb);
 
-               sock_hold(sk);
+       read_lock_bh(&l2tp_ip_lock);
+       sk = __l2tp_ip_bind_lookup(net, iph->daddr, iph->saddr, inet_iif(skb),
+                                  tunnel_id);
+       if (!sk) {
                read_unlock_bh(&l2tp_ip_lock);
+               goto discard;
        }
+       sock_hold(sk);
+       read_unlock_bh(&l2tp_ip_lock);
 
        if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
                goto discard_put;
index 88b397c..8bcaa97 100644 (file)
@@ -136,6 +136,7 @@ static int l2tp_ip6_recv(struct sk_buff *skb)
        unsigned char *ptr, *optr;
        struct l2tp_session *session;
        struct l2tp_tunnel *tunnel = NULL;
+       struct ipv6hdr *iph;
        int length;
 
        if (!pskb_may_pull(skb, 4))
@@ -192,24 +193,17 @@ pass_up:
                goto discard;
 
        tunnel_id = ntohl(*(__be32 *) &skb->data[4]);
-       tunnel = l2tp_tunnel_find(net, tunnel_id);
-       if (tunnel) {
-               sk = tunnel->sock;
-               sock_hold(sk);
-       } else {
-               struct ipv6hdr *iph = ipv6_hdr(skb);
-
-               read_lock_bh(&l2tp_ip6_lock);
-               sk = __l2tp_ip6_bind_lookup(net, &iph->daddr, &iph->saddr,
-                                           inet6_iif(skb), tunnel_id);
-               if (!sk) {
-                       read_unlock_bh(&l2tp_ip6_lock);
-                       goto discard;
-               }
+       iph = ipv6_hdr(skb);
 
-               sock_hold(sk);
+       read_lock_bh(&l2tp_ip6_lock);
+       sk = __l2tp_ip6_bind_lookup(net, &iph->daddr, &iph->saddr,
+                                   inet6_iif(skb), tunnel_id);
+       if (!sk) {
                read_unlock_bh(&l2tp_ip6_lock);
+               goto discard;
        }
+       sock_hold(sk);
+       read_unlock_bh(&l2tp_ip6_lock);
 
        if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
                goto discard_put;