[PATCH 3.2 65/77] ipv6: add complete rcu protection around np->opt

[Ben Hutchings <ben@decadent.org.uk>]

3.2.75-rc1 review patch.  If anyone has any objections, please let me know.

------------------

From: Eric Dumazet <edumazet@google.com>

[ Upstream commit 45f6fad84cc305103b28d73482b344d7f5b76f39 ]

This patch addresses multiple problems :

UDP/RAW sendmsg() need to get a stable struct ipv6_txoptions
while socket is not locked : Other threads can change np->opt
concurrently. Dmitry posted a syzkaller
(http://github.com/google/syzkaller) program desmonstrating
use-after-free.

Starting with TCP/DCCP lockless listeners, tcp_v6_syn_recv_sock()
and dccp_v6_request_recv_sock() also need to use RCU protection
to dereference np->opt once (before calling ipv6_dup_options())

This patch adds full RCU protection to np->opt

Reported-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
[bwh: Backported to 3.2:
 - Drop changes to l2tp
 - Fix an additional use of np->opt in tcp_v6_send_synack()
 - Fold in commit 43264e0bd963 ("ipv6: remove unnecessary codes in tcp_ipv6.c")
 - Adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
---
--- a/include/linux/ipv6.h
+++ b/include/linux/ipv6.h
@@ -369,7 +369,7 @@ struct ipv6_pinfo {
 	struct ipv6_ac_socklist	*ipv6_ac_list;
 	struct ipv6_fl_socklist *ipv6_fl_list;
 
-	struct ipv6_txoptions	*opt;
+	struct ipv6_txoptions __rcu	*opt;
 	struct sk_buff		*pktoptions;
 	struct sk_buff		*rxpmtu;
 	struct {
--- a/include/net/ipv6.h
+++ b/include/net/ipv6.h
@@ -191,6 +191,7 @@ extern rwlock_t ip6_ra_lock;
  */
 
 struct ipv6_txoptions {
+	atomic_t		refcnt;
 	/* Length of this structure */
 	int			tot_len;
 
@@ -203,7 +204,7 @@ struct ipv6_txoptions {
 	struct ipv6_opt_hdr	*dst0opt;
 	struct ipv6_rt_hdr	*srcrt;	/* Routing Header */
 	struct ipv6_opt_hdr	*dst1opt;
-
+	struct rcu_head		rcu;
 	/* Option buffer, as read by IPV6_PKTOPTIONS, starts here. */
 };
 
@@ -229,6 +230,24 @@ struct ipv6_fl_socklist {
 	struct ip6_flowlabel	*fl;
 };
 
+static inline struct ipv6_txoptions *txopt_get(const struct ipv6_pinfo *np)
+{
+	struct ipv6_txoptions *opt;
+
+	rcu_read_lock();
+	opt = rcu_dereference(np->opt);
+	if (opt && !atomic_inc_not_zero(&opt->refcnt))
+		opt = NULL;
+	rcu_read_unlock();
+	return opt;
+}
+
+static inline void txopt_put(struct ipv6_txoptions *opt)
+{
+	if (opt && atomic_dec_and_test(&opt->refcnt))
+		kfree_rcu(opt, rcu);
+}
+
 extern struct ip6_flowlabel	*fl6_sock_lookup(struct sock *sk, __be32 label);
 extern struct ipv6_txoptions	*fl6_merge_options(struct ipv6_txoptions * opt_space,
 						   struct ip6_flowlabel * fl,
--- a/net/dccp/ipv6.c
+++ b/net/dccp/ipv6.c
@@ -252,7 +252,9 @@ static int dccp_v6_send_response(struct
 	security_req_classify_flow(req, flowi6_to_flowi(&fl6));
 
 
-	final_p = fl6_update_dst(&fl6, np->opt, &final);
+	rcu_read_lock();
+	final_p = fl6_update_dst(&fl6, rcu_dereference(np->opt), &final);
+	rcu_read_unlock();
 
 	dst = ip6_dst_lookup_flow(sk, &fl6, final_p, false);
 	if (IS_ERR(dst)) {
@@ -269,7 +271,10 @@ static int dccp_v6_send_response(struct
 							 &ireq6->loc_addr,
 							 &ireq6->rmt_addr);
 		ipv6_addr_copy(&fl6.daddr, &ireq6->rmt_addr);
-		err = ip6_xmit(sk, skb, &fl6, np->opt, np->tclass);
+		rcu_read_lock();
+		err = ip6_xmit(sk, skb, &fl6, rcu_dereference(np->opt),
+			       np->tclass);
+		rcu_read_unlock();
 		err = net_xmit_eval(err);
 	}
 
@@ -463,6 +468,7 @@ static struct sock *dccp_v6_request_recv
 {
 	struct inet6_request_sock *ireq6 = inet6_rsk(req);
 	struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
+	struct ipv6_txoptions *opt;
 	struct inet_sock *newinet;
 	struct dccp6_sock *newdp6;
 	struct sock *newsk;
@@ -586,13 +592,15 @@ static struct sock *dccp_v6_request_recv
 	 * Yes, keeping reference count would be much more clever, but we make
 	 * one more one thing there: reattach optmem to newsk.
 	 */
-	if (np->opt != NULL)
-		newnp->opt = ipv6_dup_options(newsk, np->opt);
-
+	opt = rcu_dereference(np->opt);
+	if (opt) {
+		opt = ipv6_dup_options(newsk, opt);
+		RCU_INIT_POINTER(newnp->opt, opt);
+	}
 	inet_csk(newsk)->icsk_ext_hdr_len = 0;
-	if (newnp->opt != NULL)
-		inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
-						     newnp->opt->opt_flen);
+	if (opt)
+		inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
+						    opt->opt_flen;
 
 	dccp_sync_mss(newsk, dst_mtu(dst));
 
@@ -844,6 +852,7 @@ static int dccp_v6_connect(struct sock *
 	struct ipv6_pinfo *np = inet6_sk(sk);
 	struct dccp_sock *dp = dccp_sk(sk);
 	struct in6_addr *saddr = NULL, *final_p, final;
+	struct ipv6_txoptions *opt;
 	struct flowi6 fl6;
 	struct dst_entry *dst;
 	int addr_type;
@@ -946,7 +955,8 @@ static int dccp_v6_connect(struct sock *
 	fl6.fl6_sport = inet->inet_sport;
 	security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
 
-	final_p = fl6_update_dst(&fl6, np->opt, &final);
+	opt = rcu_dereference_protected(np->opt, sock_owned_by_user(sk));
+	final_p = fl6_update_dst(&fl6, opt, &final);
 
 	dst = ip6_dst_lookup_flow(sk, &fl6, final_p, true);
 	if (IS_ERR(dst)) {
@@ -966,9 +976,8 @@ static int dccp_v6_connect(struct sock *
 	__ip6_dst_store(sk, dst, NULL, NULL);
 
 	icsk->icsk_ext_hdr_len = 0;
-	if (np->opt != NULL)
-		icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
-					  np->opt->opt_nflen);
+	if (opt)
+		icsk->icsk_ext_hdr_len = opt->opt_flen + opt->opt_nflen;
 
 	inet->inet_dport = usin->sin6_port;
 
--- a/net/ipv6/af_inet6.c
+++ b/net/ipv6/af_inet6.c
@@ -430,8 +430,11 @@ void inet6_destroy_sock(struct sock *sk)
 
 	/* Free tx options */
 
-	if ((opt = xchg(&np->opt, NULL)) != NULL)
-		sock_kfree_s(sk, opt, opt->tot_len);
+	opt = xchg((__force struct ipv6_txoptions **)&np->opt, NULL);
+	if (opt) {
+		atomic_sub(opt->tot_len, &sk->sk_omem_alloc);
+		txopt_put(opt);
+	}
 }
 
 EXPORT_SYMBOL_GPL(inet6_destroy_sock);
@@ -669,7 +672,10 @@ int inet6_sk_rebuild_header(struct sock
 		fl6.fl6_sport = inet->inet_sport;
 		security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
 
-		final_p = fl6_update_dst(&fl6, np->opt, &final);
+		rcu_read_lock();
+		final_p = fl6_update_dst(&fl6, rcu_dereference(np->opt),
+					 &final);
+		rcu_read_unlock();
 
 		dst = ip6_dst_lookup_flow(sk, &fl6, final_p, false);
 		if (IS_ERR(dst)) {
--- a/net/ipv6/datagram.c
+++ b/net/ipv6/datagram.c
@@ -166,8 +166,10 @@ ipv4_connected:
 
 	security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
 
-	opt = flowlabel ? flowlabel->opt : np->opt;
+	rcu_read_lock();
+	opt = flowlabel ? flowlabel->opt : rcu_dereference(np->opt);
 	final_p = fl6_update_dst(&fl6, opt, &final);
+	rcu_read_unlock();
 
 	dst = ip6_dst_lookup_flow(sk, &fl6, final_p, true);
 	err = 0;
--- a/net/ipv6/exthdrs.c
+++ b/net/ipv6/exthdrs.c
@@ -748,6 +748,7 @@ ipv6_dup_options(struct sock *sk, struct
 			*((char**)&opt2->dst1opt) += dif;
 		if (opt2->srcrt)
 			*((char**)&opt2->srcrt) += dif;
+		atomic_set(&opt2->refcnt, 1);
 	}
 	return opt2;
 }
@@ -812,7 +813,7 @@ ipv6_renew_options(struct sock *sk, stru
 		return ERR_PTR(-ENOBUFS);
 
 	memset(opt2, 0, tot_len);
-
+	atomic_set(&opt2->refcnt, 1);
 	opt2->tot_len = tot_len;
 	p = (char *)(opt2 + 1);
 
--- a/net/ipv6/inet6_connection_sock.c
+++ b/net/ipv6/inet6_connection_sock.c
@@ -66,7 +66,9 @@ struct dst_entry *inet6_csk_route_req(st
 	memset(&fl6, 0, sizeof(fl6));
 	fl6.flowi6_proto = IPPROTO_TCP;
 	ipv6_addr_copy(&fl6.daddr, &treq->rmt_addr);
-	final_p = fl6_update_dst(&fl6, np->opt, &final);
+	rcu_read_lock();
+	final_p = fl6_update_dst(&fl6, rcu_dereference(np->opt), &final);
+	rcu_read_unlock();
 	ipv6_addr_copy(&fl6.saddr, &treq->loc_addr);
 	fl6.flowi6_oif = sk->sk_bound_dev_if;
 	fl6.flowi6_mark = sk->sk_mark;
@@ -225,7 +227,9 @@ int inet6_csk_xmit(struct sk_buff *skb,
 	fl6.fl6_dport = inet->inet_dport;
 	security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
 
-	final_p = fl6_update_dst(&fl6, np->opt, &final);
+	rcu_read_lock();
+	final_p = fl6_update_dst(&fl6, rcu_dereference(np->opt), &final);
+	rcu_read_unlock();
 
 	dst = __inet6_csk_dst_check(sk, np->dst_cookie);
 
@@ -248,7 +252,8 @@ int inet6_csk_xmit(struct sk_buff *skb,
 	/* Restore final destination back after routing done */
 	ipv6_addr_copy(&fl6.daddr, &np->daddr);
 
-	res = ip6_xmit(sk, skb, &fl6, np->opt, np->tclass);
+	res = ip6_xmit(sk, skb, &fl6, rcu_dereference(np->opt),
+		       np->tclass);
 	rcu_read_unlock();
 	return res;
 }
--- a/net/ipv6/ipv6_sockglue.c
+++ b/net/ipv6/ipv6_sockglue.c
@@ -110,10 +110,12 @@ struct ipv6_txoptions *ipv6_update_optio
 			icsk->icsk_ext_hdr_len = opt->opt_flen + opt->opt_nflen;
 			icsk->icsk_sync_mss(sk, icsk->icsk_pmtu_cookie);
 		}
-		opt = xchg(&inet6_sk(sk)->opt, opt);
+		opt = xchg((__force struct ipv6_txoptions **)&inet6_sk(sk)->opt,
+			   opt);
 	} else {
 		spin_lock(&sk->sk_dst_lock);
-		opt = xchg(&inet6_sk(sk)->opt, opt);
+		opt = xchg((__force struct ipv6_txoptions **)&inet6_sk(sk)->opt,
+			   opt);
 		spin_unlock(&sk->sk_dst_lock);
 	}
 	sk_dst_reset(sk);
@@ -213,9 +215,12 @@ static int do_ipv6_setsockopt(struct soc
 				sk->sk_socket->ops = &inet_dgram_ops;
 				sk->sk_family = PF_INET;
 			}
-			opt = xchg(&np->opt, NULL);
-			if (opt)
-				sock_kfree_s(sk, opt, opt->tot_len);
+			opt = xchg((__force struct ipv6_txoptions **)&np->opt,
+				   NULL);
+			if (opt) {
+				atomic_sub(opt->tot_len, &sk->sk_omem_alloc);
+				txopt_put(opt);
+			}
 			pktopt = xchg(&np->pktoptions, NULL);
 			kfree_skb(pktopt);
 
@@ -384,7 +389,8 @@ static int do_ipv6_setsockopt(struct soc
 		if (optname != IPV6_RTHDR && !capable(CAP_NET_RAW))
 			break;
 
-		opt = ipv6_renew_options(sk, np->opt, optname,
+		opt = rcu_dereference_protected(np->opt, sock_owned_by_user(sk));
+		opt = ipv6_renew_options(sk, opt, optname,
 					 (struct ipv6_opt_hdr __user *)optval,
 					 optlen);
 		if (IS_ERR(opt)) {
@@ -413,8 +419,10 @@ static int do_ipv6_setsockopt(struct soc
 		retv = 0;
 		opt = ipv6_update_options(sk, opt);
 sticky_done:
-		if (opt)
-			sock_kfree_s(sk, opt, opt->tot_len);
+		if (opt) {
+			atomic_sub(opt->tot_len, &sk->sk_omem_alloc);
+			txopt_put(opt);
+		}
 		break;
 	}
 
@@ -467,6 +475,7 @@ sticky_done:
 			break;
 
 		memset(opt, 0, sizeof(*opt));
+		atomic_set(&opt->refcnt, 1);
 		opt->tot_len = sizeof(*opt) + optlen;
 		retv = -EFAULT;
 		if (copy_from_user(opt+1, optval, optlen))
@@ -483,8 +492,10 @@ update:
 		retv = 0;
 		opt = ipv6_update_options(sk, opt);
 done:
-		if (opt)
-			sock_kfree_s(sk, opt, opt->tot_len);
+		if (opt) {
+			atomic_sub(opt->tot_len, &sk->sk_omem_alloc);
+			txopt_put(opt);
+		}
 		break;
 	}
 	case IPV6_UNICAST_HOPS:
@@ -1053,10 +1064,11 @@ static int do_ipv6_getsockopt(struct soc
 	case IPV6_RTHDR:
 	case IPV6_DSTOPTS:
 	{
+		struct ipv6_txoptions *opt;
 
 		lock_sock(sk);
-		len = ipv6_getsockopt_sticky(sk, np->opt,
-					     optname, optval, len);
+		opt = rcu_dereference_protected(np->opt, sock_owned_by_user(sk));
+		len = ipv6_getsockopt_sticky(sk, opt, optname, optval, len);
 		release_sock(sk);
 		/* check if ipv6_getsockopt_sticky() returns err code */
 		if (len < 0)
--- a/net/ipv6/raw.c
+++ b/net/ipv6/raw.c
@@ -724,6 +724,7 @@ static int rawv6_probe_proto_opt(struct
 static int rawv6_sendmsg(struct kiocb *iocb, struct sock *sk,
 		   struct msghdr *msg, size_t len)
 {
+	struct ipv6_txoptions *opt_to_free = NULL;
 	struct ipv6_txoptions opt_space;
 	struct sockaddr_in6 * sin6 = (struct sockaddr_in6 *) msg->msg_name;
 	struct in6_addr *daddr, *final_p, final;
@@ -830,8 +831,10 @@ static int rawv6_sendmsg(struct kiocb *i
 		if (!(opt->opt_nflen|opt->opt_flen))
 			opt = NULL;
 	}
-	if (opt == NULL)
-		opt = np->opt;
+	if (!opt) {
+		opt = txopt_get(np);
+		opt_to_free = opt;
+	}
 	if (flowlabel)
 		opt = fl6_merge_options(&opt_space, flowlabel, opt);
 	opt = ipv6_fixup_options(&opt_space, opt);
@@ -896,6 +899,7 @@ done:
 	dst_release(dst);
 out:
 	fl6_sock_release(flowlabel);
+	txopt_put(opt_to_free);
 	return err<0?err:len;
 do_confirm:
 	dst_confirm(dst);
--- a/net/ipv6/syncookies.c
+++ b/net/ipv6/syncookies.c
@@ -238,7 +238,7 @@ struct sock *cookie_v6_check(struct sock
 		memset(&fl6, 0, sizeof(fl6));
 		fl6.flowi6_proto = IPPROTO_TCP;
 		ipv6_addr_copy(&fl6.daddr, &ireq6->rmt_addr);
-		final_p = fl6_update_dst(&fl6, np->opt, &final);
+		final_p = fl6_update_dst(&fl6, rcu_dereference(np->opt), &final);
 		ipv6_addr_copy(&fl6.saddr, &ireq6->loc_addr);
 		fl6.flowi6_oif = sk->sk_bound_dev_if;
 		fl6.flowi6_mark = sk->sk_mark;
--- a/net/ipv6/tcp_ipv6.c
+++ b/net/ipv6/tcp_ipv6.c
@@ -131,6 +131,7 @@ static int tcp_v6_connect(struct sock *s
 	struct ipv6_pinfo *np = inet6_sk(sk);
 	struct tcp_sock *tp = tcp_sk(sk);
 	struct in6_addr *saddr = NULL, *final_p, final;
+	struct ipv6_txoptions *opt;
 	struct rt6_info *rt;
 	struct flowi6 fl6;
 	struct dst_entry *dst;
@@ -252,7 +253,8 @@ static int tcp_v6_connect(struct sock *s
 	fl6.fl6_dport = usin->sin6_port;
 	fl6.fl6_sport = inet->inet_sport;
 
-	final_p = fl6_update_dst(&fl6, np->opt, &final);
+	opt = rcu_dereference_protected(np->opt, sock_owned_by_user(sk));
+	final_p = fl6_update_dst(&fl6, opt, &final);
 
 	security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
 
@@ -295,9 +297,9 @@ static int tcp_v6_connect(struct sock *s
 	}
 
 	icsk->icsk_ext_hdr_len = 0;
-	if (np->opt)
-		icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
-					  np->opt->opt_nflen);
+	if (opt)
+		icsk->icsk_ext_hdr_len = opt->opt_flen +
+					 opt->opt_nflen;
 
 	tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
 
@@ -481,7 +483,6 @@ static int tcp_v6_send_synack(struct soc
 	struct inet6_request_sock *treq = inet6_rsk(req);
 	struct ipv6_pinfo *np = inet6_sk(sk);
 	struct sk_buff * skb;
-	struct ipv6_txoptions *opt = NULL;
 	struct in6_addr * final_p, final;
 	struct flowi6 fl6;
 	struct dst_entry *dst;
@@ -498,8 +499,7 @@ static int tcp_v6_send_synack(struct soc
 	fl6.fl6_sport = inet_rsk(req)->loc_port;
 	security_req_classify_flow(req, flowi6_to_flowi(&fl6));
 
-	opt = np->opt;
-	final_p = fl6_update_dst(&fl6, opt, &final);
+	final_p = fl6_update_dst(&fl6, rcu_dereference(np->opt), &final);
 
 	dst = ip6_dst_lookup_flow(sk, &fl6, final_p, false);
 	if (IS_ERR(dst)) {
@@ -513,13 +513,12 @@ static int tcp_v6_send_synack(struct soc
 		__tcp_v6_send_check(skb, &treq->loc_addr, &treq->rmt_addr);
 
 		ipv6_addr_copy(&fl6.daddr, &treq->rmt_addr);
-		err = ip6_xmit(sk, skb, &fl6, opt, np->tclass);
+		err = ip6_xmit(sk, skb, &fl6, rcu_dereference(np->opt),
+			       np->tclass);
 		err = net_xmit_eval(err);
 	}
 
 done:
-	if (opt && opt != np->opt)
-		sock_kfree_s(sk, opt, opt->tot_len);
 	dst_release(dst);
 	return err;
 }
@@ -1408,7 +1407,6 @@ static struct sock * tcp_v6_syn_recv_soc
 	}
 
 	treq = inet6_rsk(req);
-	opt = np->opt;
 
 	if (sk_acceptq_is_full(sk))
 		goto out_overflow;
@@ -1476,16 +1474,15 @@ static struct sock * tcp_v6_syn_recv_soc
 	   but we make one more one thing there: reattach optmem
 	   to newsk.
 	 */
+	opt = rcu_dereference(np->opt);
 	if (opt) {
-		newnp->opt = ipv6_dup_options(newsk, opt);
-		if (opt != np->opt)
-			sock_kfree_s(sk, opt, opt->tot_len);
+		opt = ipv6_dup_options(newsk, opt);
+		RCU_INIT_POINTER(newnp->opt, opt);
 	}
-
-	inet_csk(newsk)->icsk_ext_hdr_len = 0;
-	if (newnp->opt)
-		inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
-						     newnp->opt->opt_flen);
+ 	inet_csk(newsk)->icsk_ext_hdr_len = 0;
+	if (opt)
+		inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
+						    opt->opt_flen;
 
 	tcp_mtup_init(newsk);
 	tcp_sync_mss(newsk, dst_mtu(dst));
@@ -1530,8 +1527,6 @@ static struct sock * tcp_v6_syn_recv_soc
 out_overflow:
 	NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 out_nonewsk:
-	if (opt && opt != np->opt)
-		sock_kfree_s(sk, opt, opt->tot_len);
 	dst_release(dst);
 out:
 	NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
--- a/net/ipv6/udp.c
+++ b/net/ipv6/udp.c
@@ -954,6 +954,7 @@ int udpv6_sendmsg(struct kiocb *iocb, st
 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) msg->msg_name;
 	struct in6_addr *daddr, *final_p, final;
 	struct ipv6_txoptions *opt = NULL;
+	struct ipv6_txoptions *opt_to_free = NULL;
 	struct ip6_flowlabel *flowlabel = NULL;
 	struct flowi6 fl6;
 	struct dst_entry *dst;
@@ -1107,8 +1108,10 @@ do_udp_sendmsg:
 			opt = NULL;
 		connected = 0;
 	}
-	if (opt == NULL)
-		opt = np->opt;
+	if (!opt) {
+		opt = txopt_get(np);
+		opt_to_free = opt;
+	}
 	if (flowlabel)
 		opt = fl6_merge_options(&opt_space, flowlabel, opt);
 	opt = ipv6_fixup_options(&opt_space, opt);
@@ -1208,6 +1211,7 @@ do_append_data:
 out:
 	dst_release(dst);
 	fl6_sock_release(flowlabel);
+	txopt_put(opt_to_free);
 	if (!err)
 		return len;
 	/*