ipv6: kill sk_dst_lock

While testing the np->opt RCU conversion, I found that UDP/IPv6 was
using a mixture of xchg() and sk_dst_lock to protect concurrent changes
to sk->sk_dst_cache, leading to possible corruptions and crashes.

ip6_sk_dst_lookup_flow() uses sk_dst_check() anyway, so the simplest
way to fix the mess is to remove sk_dst_lock completely, as we did for
IPv4.

__ip6_dst_store() and ip6_dst_store() share same implementation.

sk_setup_caps() being called with socket lock being held or not,
we have to use sk_dst_set() instead of __sk_dst_set()

Note that I had to move the "np->dst_cookie = rt6_get_cookie(rt);"
in ip6_dst_store() before the sk_setup_caps(sk, dst) call.

This is because ip6_dst_store() can be called from process context,
without any lock held.

As soon as the dst is installed in sk->sk_dst_cache, dst can be freed
from another cpu doing a concurrent ip6_dst_store()

Doing the dst dereference before doing the install is needed to make
sure no use after free would trigger.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

6 files changed, 7 insertions, 31 deletions

diff --git a/net/core/sock.c b/net/core/sock.c
index 9d79569935a3..e31dfcee1729 100644
--- a/net/core/sock.c
+++ b/net/core/sock.c
@@ -1530,7 +1530,6 @@ struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
 		skb_queue_head_init(&newsk->sk_receive_queue);
 		skb_queue_head_init(&newsk->sk_write_queue);
 
-		spin_lock_init(&newsk->sk_dst_lock);
 		rwlock_init(&newsk->sk_callback_lock);
 		lockdep_set_class_and_name(&newsk->sk_callback_lock,
 				af_callback_keys + newsk->sk_family,
@@ -1607,7 +1606,7 @@ void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
 {
 	u32 max_segs = 1;
 
-	__sk_dst_set(sk, dst);
+	sk_dst_set(sk, dst);
 	sk->sk_route_caps = dst->dev->features;
 	if (sk->sk_route_caps & NETIF_F_GSO)
 		sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
@@ -2388,7 +2387,6 @@ void sock_init_data(struct socket *sock, struct sock *sk)
 	} else
 		sk->sk_wq	=	NULL;
 
-	spin_lock_init(&sk->sk_dst_lock);
 	rwlock_init(&sk->sk_callback_lock);
 	lockdep_set_class_and_name(&sk->sk_callback_lock,
 			af_callback_keys + sk->sk_family,
diff --git a/net/dccp/ipv6.c b/net/dccp/ipv6.c
index e7e0b9bc2a43..9c6d0508e63a 100644
--- a/net/dccp/ipv6.c
+++ b/net/dccp/ipv6.c
@@ -459,7 +459,7 @@ static struct sock *dccp_v6_request_recv_sock(const struct sock *sk,
 	 * comment in that function for the gory details. -acme
 	 */
 
-	__ip6_dst_store(newsk, dst, NULL, NULL);
+	ip6_dst_store(newsk, dst, NULL, NULL);
 	newsk->sk_route_caps = dst->dev->features & ~(NETIF_F_IP_CSUM |
 						      NETIF_F_TSO);
 	newdp6 = (struct dccp6_sock *)newsk;
@@ -883,7 +883,7 @@ static int dccp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 	np->saddr = *saddr;
 	inet->inet_rcv_saddr = LOOPBACK4_IPV6;
 
-	__ip6_dst_store(sk, dst, NULL, NULL);
+	ip6_dst_store(sk, dst, NULL, NULL);
 
 	icsk->icsk_ext_hdr_len = 0;
 	if (opt)
diff --git a/net/ipv6/af_inet6.c b/net/ipv6/af_inet6.c
index 38d66ddfb937..8ec0df75f1c4 100644
--- a/net/ipv6/af_inet6.c
+++ b/net/ipv6/af_inet6.c
@@ -673,7 +673,7 @@ int inet6_sk_rebuild_header(struct sock *sk)
 			return PTR_ERR(dst);
 		}
 
-		__ip6_dst_store(sk, dst, NULL, NULL);
+		ip6_dst_store(sk, dst, NULL, NULL);
 	}
 
 	return 0;
diff --git a/net/ipv6/icmp.c b/net/ipv6/icmp.c
index 36c5a98b0472..0a37ddc7af51 100644
--- a/net/ipv6/icmp.c
+++ b/net/ipv6/icmp.c
@@ -834,11 +834,6 @@ void icmpv6_flow_init(struct sock *sk, struct flowi6 *fl6,
 	security_sk_classify_flow(sk, flowi6_to_flowi(fl6));
 }
 
-/*
- * Special lock-class for __icmpv6_sk:
- */
-static struct lock_class_key icmpv6_socket_sk_dst_lock_key;
-
 static int __net_init icmpv6_sk_init(struct net *net)
 {
 	struct sock *sk;
@@ -860,15 +855,6 @@ static int __net_init icmpv6_sk_init(struct net *net)
 
 		net->ipv6.icmp_sk[i] = sk;
 
-		/*
-		 * Split off their lock-class, because sk->sk_dst_lock
-		 * gets used from softirqs, which is safe for
-		 * __icmpv6_sk (because those never get directly used
-		 * via userspace syscalls), but unsafe for normal sockets.
-		 */
-		lockdep_set_class(&sk->sk_dst_lock,
-				  &icmpv6_socket_sk_dst_lock_key);
-
 		/* Enough space for 2 64K ICMP packets, including
 		 * sk_buff struct overhead.
 		 */
diff --git a/net/ipv6/inet6_connection_sock.c b/net/ipv6/inet6_connection_sock.c
index 3ff5208772bb..a7ca2cde2ecb 100644
--- a/net/ipv6/inet6_connection_sock.c
+++ b/net/ipv6/inet6_connection_sock.c
@@ -111,14 +111,6 @@ void inet6_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
 EXPORT_SYMBOL_GPL(inet6_csk_addr2sockaddr);
 
 static inline
-void __inet6_csk_dst_store(struct sock *sk, struct dst_entry *dst,
-			   const struct in6_addr *daddr,
-			   const struct in6_addr *saddr)
-{
-	__ip6_dst_store(sk, dst, daddr, saddr);
-}
-
-static inline
 struct dst_entry *__inet6_csk_dst_check(struct sock *sk, u32 cookie)
 {
 	return __sk_dst_check(sk, cookie);
@@ -153,7 +145,7 @@ static struct dst_entry *inet6_csk_route_socket(struct sock *sk,
 		dst = ip6_dst_lookup_flow(sk, fl6, final_p);
 
 		if (!IS_ERR(dst))
-			__inet6_csk_dst_store(sk, dst, NULL, NULL);
+			ip6_dst_store(sk, dst, NULL, NULL);
 	}
 	return dst;
 }
diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
index 6a50bb4a0dae..e7aab561b7b4 100644
--- a/net/ipv6/tcp_ipv6.c
+++ b/net/ipv6/tcp_ipv6.c
@@ -257,7 +257,7 @@ static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 	inet->inet_rcv_saddr = LOOPBACK4_IPV6;
 
 	sk->sk_gso_type = SKB_GSO_TCPV6;
-	__ip6_dst_store(sk, dst, NULL, NULL);
+	ip6_dst_store(sk, dst, NULL, NULL);
 
 	if (tcp_death_row.sysctl_tw_recycle &&
 	    !tp->rx_opt.ts_recent_stamp &&
@@ -1060,7 +1060,7 @@ static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *
 	 */
 
 	newsk->sk_gso_type = SKB_GSO_TCPV6;
-	__ip6_dst_store(newsk, dst, NULL, NULL);
+	ip6_dst_store(newsk, dst, NULL, NULL);
 	inet6_sk_rx_dst_set(newsk, skb);
 
 	newtcp6sk = (struct tcp6_sock *)newsk;