[PATCH net-next 0/8] inet: more data-race fixes

[PATCH net-next 0/8] inet: more data-race fixes

[Eric Dumazet]

This series fixes some existing data-races on inet fields:

inet->mc_ttl, inet->pmtudisc, inet->tos, inet->uc_index,
inet->mc_index and inet->mc_addr.

While fixing them, we convert eight socket options
to lockless implementation.

Eric Dumazet (8):
  inet: implement lockless IP_MULTICAST_TTL
  inet: implement lockless IP_MTU_DISCOVER
  inet: implement lockless IP_TOS
  inet: lockless getsockopt(IP_OPTIONS)
  inet: lockless getsockopt(IP_MTU)
  inet: implement lockless getsockopt(IP_UNICAST_IF)
  inet: lockless IP_PKTOPTIONS implementation
  inet: implement lockless getsockopt(IP_MULTICAST_IF)

 include/net/ip.h                              |  14 +-
 net/dccp/ipv4.c                               |   2 +-
 net/ipv4/datagram.c                           |   6 +-
 net/ipv4/inet_diag.c                          |   2 +-
 net/ipv4/ip_output.c                          |  13 +-
 net/ipv4/ip_sockglue.c                        | 192 ++++++++----------
 net/ipv4/ping.c                               |   8 +-
 net/ipv4/raw.c                                |  19 +-
 net/ipv4/tcp_ipv4.c                           |   9 +-
 net/ipv4/udp.c                                |  18 +-
 net/mptcp/sockopt.c                           |   8 +-
 net/netfilter/ipvs/ip_vs_sync.c               |   4 +-
 net/sctp/protocol.c                           |   4 +-
 .../selftests/net/mptcp/mptcp_connect.sh      |   2 +-
 14 files changed, 147 insertions(+), 154 deletions(-)

-- 
2.42.0.459.ge4e396fd5e-goog

[PATCH net-next 1/8] inet: implement lockless IP_MULTICAST_TTL

[Eric Dumazet]

inet->mc_ttl can be read locklessly.

Implement proper lockless reads and writes to inet->mc_ttl

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 net/ipv4/ip_output.c            |  2 +-
 net/ipv4/ip_sockglue.c          | 31 ++++++++++++++++---------------
 net/netfilter/ipvs/ip_vs_sync.c |  2 +-
 3 files changed, 18 insertions(+), 17 deletions(-)

diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c
index 4ab877cf6d35f229761986d5c6a17eb2a3ad4043..adad16f1e872ce20941a087b3965fdb040868d4e 100644
--- a/net/ipv4/ip_output.c
+++ b/net/ipv4/ip_output.c
@@ -1430,7 +1430,7 @@ struct sk_buff *__ip_make_skb(struct sock *sk,
 	if (cork->ttl != 0)
 		ttl = cork->ttl;
 	else if (rt->rt_type == RTN_MULTICAST)
-		ttl = inet->mc_ttl;
+		ttl = READ_ONCE(inet->mc_ttl);
 	else
 		ttl = ip_select_ttl(inet, &rt->dst);
 
diff --git a/net/ipv4/ip_sockglue.c b/net/ipv4/ip_sockglue.c
index cce9cb25f3b31cd57fa883ae0dedb6829d8da2fa..4ad3003378ae6b186513000264f77b54a7babe6d 100644
--- a/net/ipv4/ip_sockglue.c
+++ b/net/ipv4/ip_sockglue.c
@@ -1039,6 +1039,17 @@ int do_ip_setsockopt(struct sock *sk, int level, int optname,
 
 		WRITE_ONCE(inet->min_ttl, val);
 		return 0;
+	case IP_MULTICAST_TTL:
+		if (sk->sk_type == SOCK_STREAM)
+			return -EINVAL;
+		if (optlen < 1)
+			return -EINVAL;
+		if (val == -1)
+			val = 1;
+		if (val < 0 || val > 255)
+			return -EINVAL;
+		WRITE_ONCE(inet->mc_ttl, val);
+		return 0;
 	}
 
 	err = 0;
@@ -1101,17 +1112,6 @@ int do_ip_setsockopt(struct sock *sk, int level, int optname,
 			goto e_inval;
 		inet->pmtudisc = val;
 		break;
-	case IP_MULTICAST_TTL:
-		if (sk->sk_type == SOCK_STREAM)
-			goto e_inval;
-		if (optlen < 1)
-			goto e_inval;
-		if (val == -1)
-			val = 1;
-		if (val < 0 || val > 255)
-			goto e_inval;
-		inet->mc_ttl = val;
-		break;
 	case IP_UNICAST_IF:
 	{
 		struct net_device *dev = NULL;
@@ -1592,6 +1592,9 @@ int do_ip_getsockopt(struct sock *sk, int level, int optname,
 	case IP_MINTTL:
 		val = READ_ONCE(inet->min_ttl);
 		goto copyval;
+	case IP_MULTICAST_TTL:
+		val = READ_ONCE(inet->mc_ttl);
+		goto copyval;
 	}
 
 	if (needs_rtnl)
@@ -1649,9 +1652,6 @@ int do_ip_getsockopt(struct sock *sk, int level, int optname,
 		}
 		break;
 	}
-	case IP_MULTICAST_TTL:
-		val = inet->mc_ttl;
-		break;
 	case IP_UNICAST_IF:
 		val = (__force int)htonl((__u32) inet->uc_index);
 		break;
@@ -1718,7 +1718,8 @@ int do_ip_getsockopt(struct sock *sk, int level, int optname,
 			put_cmsg(&msg, SOL_IP, IP_PKTINFO, sizeof(info), &info);
 		}
 		if (inet_test_bit(TTL, sk)) {
-			int hlim = inet->mc_ttl;
+			int hlim = READ_ONCE(inet->mc_ttl);
+
 			put_cmsg(&msg, SOL_IP, IP_TTL, sizeof(hlim), &hlim);
 		}
 		if (inet_test_bit(TOS, sk)) {
diff --git a/net/netfilter/ipvs/ip_vs_sync.c b/net/netfilter/ipvs/ip_vs_sync.c
index 5820a8156c4701bb163f569d735c389d7a8e3820..3eed1670224888acf639cff06537ddf2505461bb 100644
--- a/net/netfilter/ipvs/ip_vs_sync.c
+++ b/net/netfilter/ipvs/ip_vs_sync.c
@@ -1316,7 +1316,7 @@ static void set_mcast_ttl(struct sock *sk, u_char ttl)
 
 	/* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
 	lock_sock(sk);
-	inet->mc_ttl = ttl;
+	WRITE_ONCE(inet->mc_ttl, ttl);
 #ifdef CONFIG_IP_VS_IPV6
 	if (sk->sk_family == AF_INET6) {
 		struct ipv6_pinfo *np = inet6_sk(sk);
-- 
2.42.0.459.ge4e396fd5e-goog

[PATCH net-next 2/8] inet: implement lockless IP_MTU_DISCOVER

[Eric Dumazet]

inet->pmtudisc can be read locklessly.

Implement proper lockless reads and writes to inet->pmtudisc

ip_sock_set_mtu_discover() can now be called from arbitrary
contexts.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/ip.h                | 13 ++++++++-----
 net/ipv4/ip_output.c            |  7 ++++---
 net/ipv4/ip_sockglue.c          | 17 ++++++-----------
 net/ipv4/ping.c                 |  2 +-
 net/ipv4/raw.c                  |  2 +-
 net/ipv4/udp.c                  |  2 +-
 net/netfilter/ipvs/ip_vs_sync.c |  2 +-
 7 files changed, 22 insertions(+), 23 deletions(-)

diff --git a/include/net/ip.h b/include/net/ip.h
index 3489a1cca5e7bc315ba646f6bc125b2b6ded9416..46933a0d98eac2db40c2e88006125588b8f8143e 100644
--- a/include/net/ip.h
+++ b/include/net/ip.h
@@ -434,19 +434,22 @@ int ip_dont_fragment(const struct sock *sk, const struct dst_entry *dst)
 
 static inline bool ip_sk_accept_pmtu(const struct sock *sk)
 {
-	return inet_sk(sk)->pmtudisc != IP_PMTUDISC_INTERFACE &&
-	       inet_sk(sk)->pmtudisc != IP_PMTUDISC_OMIT;
+	u8 pmtudisc = READ_ONCE(inet_sk(sk)->pmtudisc);
+
+	return pmtudisc != IP_PMTUDISC_INTERFACE &&
+	       pmtudisc != IP_PMTUDISC_OMIT;
 }
 
 static inline bool ip_sk_use_pmtu(const struct sock *sk)
 {
-	return inet_sk(sk)->pmtudisc < IP_PMTUDISC_PROBE;
+	return READ_ONCE(inet_sk(sk)->pmtudisc) < IP_PMTUDISC_PROBE;
 }
 
 static inline bool ip_sk_ignore_df(const struct sock *sk)
 {
-	return inet_sk(sk)->pmtudisc < IP_PMTUDISC_DO ||
-	       inet_sk(sk)->pmtudisc == IP_PMTUDISC_OMIT;
+	u8 pmtudisc = READ_ONCE(inet_sk(sk)->pmtudisc);
+
+	return pmtudisc < IP_PMTUDISC_DO || pmtudisc == IP_PMTUDISC_OMIT;
 }
 
 static inline unsigned int ip_dst_mtu_maybe_forward(const struct dst_entry *dst,
diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c
index adad16f1e872ce20941a087b3965fdb040868d4e..2be281f184a5fe5a695ccd51fabe69fa45bea0b8 100644
--- a/net/ipv4/ip_output.c
+++ b/net/ipv4/ip_output.c
@@ -1387,8 +1387,8 @@ struct sk_buff *__ip_make_skb(struct sock *sk,
 	struct ip_options *opt = NULL;
 	struct rtable *rt = (struct rtable *)cork->dst;
 	struct iphdr *iph;
+	u8 pmtudisc, ttl;
 	__be16 df = 0;
-	__u8 ttl;
 
 	skb = __skb_dequeue(queue);
 	if (!skb)
@@ -1418,8 +1418,9 @@ struct sk_buff *__ip_make_skb(struct sock *sk,
 	/* DF bit is set when we want to see DF on outgoing frames.
 	 * If ignore_df is set too, we still allow to fragment this frame
 	 * locally. */
-	if (inet->pmtudisc == IP_PMTUDISC_DO ||
-	    inet->pmtudisc == IP_PMTUDISC_PROBE ||
+	pmtudisc = READ_ONCE(inet->pmtudisc);
+	if (pmtudisc == IP_PMTUDISC_DO ||
+	    pmtudisc == IP_PMTUDISC_PROBE ||
 	    (skb->len <= dst_mtu(&rt->dst) &&
 	     ip_dont_fragment(sk, &rt->dst)))
 		df = htons(IP_DF);
diff --git a/net/ipv4/ip_sockglue.c b/net/ipv4/ip_sockglue.c
index 4ad3003378ae6b186513000264f77b54a7babe6d..6d874cc03c8b4e88d79ebc50a6db105606b6ae60 100644
--- a/net/ipv4/ip_sockglue.c
+++ b/net/ipv4/ip_sockglue.c
@@ -622,9 +622,7 @@ int ip_sock_set_mtu_discover(struct sock *sk, int val)
 {
 	if (val < IP_PMTUDISC_DONT || val > IP_PMTUDISC_OMIT)
 		return -EINVAL;
-	lock_sock(sk);
-	inet_sk(sk)->pmtudisc = val;
-	release_sock(sk);
+	WRITE_ONCE(inet_sk(sk)->pmtudisc, val);
 	return 0;
 }
 EXPORT_SYMBOL(ip_sock_set_mtu_discover);
@@ -1050,6 +1048,8 @@ int do_ip_setsockopt(struct sock *sk, int level, int optname,
 			return -EINVAL;
 		WRITE_ONCE(inet->mc_ttl, val);
 		return 0;
+	case IP_MTU_DISCOVER:
+		return ip_sock_set_mtu_discover(sk, val);
 	}
 
 	err = 0;
@@ -1107,11 +1107,6 @@ int do_ip_setsockopt(struct sock *sk, int level, int optname,
 	case IP_TOS:	/* This sets both TOS and Precedence */
 		__ip_sock_set_tos(sk, val);
 		break;
-	case IP_MTU_DISCOVER:
-		if (val < IP_PMTUDISC_DONT || val > IP_PMTUDISC_OMIT)
-			goto e_inval;
-		inet->pmtudisc = val;
-		break;
 	case IP_UNICAST_IF:
 	{
 		struct net_device *dev = NULL;
@@ -1595,6 +1590,9 @@ int do_ip_getsockopt(struct sock *sk, int level, int optname,
 	case IP_MULTICAST_TTL:
 		val = READ_ONCE(inet->mc_ttl);
 		goto copyval;
+	case IP_MTU_DISCOVER:
+		val = READ_ONCE(inet->pmtudisc);
+		goto copyval;
 	}
 
 	if (needs_rtnl)
@@ -1634,9 +1632,6 @@ int do_ip_getsockopt(struct sock *sk, int level, int optname,
 	case IP_TOS:
 		val = inet->tos;
 		break;
-	case IP_MTU_DISCOVER:
-		val = inet->pmtudisc;
-		break;
 	case IP_MTU:
 	{
 		struct dst_entry *dst;
diff --git a/net/ipv4/ping.c b/net/ipv4/ping.c
index 4dd809b7b18867154df42bc28809b886913e253c..50d12b0c8d46fdcd9b448c3ebc90395ebf426075 100644
--- a/net/ipv4/ping.c
+++ b/net/ipv4/ping.c
@@ -551,7 +551,7 @@ void ping_err(struct sk_buff *skb, int offset, u32 info)
 		case ICMP_DEST_UNREACH:
 			if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
 				ipv4_sk_update_pmtu(skb, sk, info);
-				if (inet_sock->pmtudisc != IP_PMTUDISC_DONT) {
+				if (READ_ONCE(inet_sock->pmtudisc) != IP_PMTUDISC_DONT) {
 					err = EMSGSIZE;
 					harderr = 1;
 					break;
diff --git a/net/ipv4/raw.c b/net/ipv4/raw.c
index 4b5db5d1edc279df1fd7412af2845a7a79c95ec8..ade1aecd7c71184d753a28a67bc9b30087247db4 100644
--- a/net/ipv4/raw.c
+++ b/net/ipv4/raw.c
@@ -239,7 +239,7 @@ static void raw_err(struct sock *sk, struct sk_buff *skb, u32 info)
 		if (code > NR_ICMP_UNREACH)
 			break;
 		if (code == ICMP_FRAG_NEEDED) {
-			harderr = inet->pmtudisc != IP_PMTUDISC_DONT;
+			harderr = READ_ONCE(inet->pmtudisc) != IP_PMTUDISC_DONT;
 			err = EMSGSIZE;
 		} else {
 			err = icmp_err_convert[code].errno;
diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
index c3ff984b63547daf0ecfb4ab96956aee2f8d589d..731a723dc80816f0b5b0803d7397f7e9e8cd8b09 100644
--- a/net/ipv4/udp.c
+++ b/net/ipv4/udp.c
@@ -750,7 +750,7 @@ int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
 	case ICMP_DEST_UNREACH:
 		if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
 			ipv4_sk_update_pmtu(skb, sk, info);
-			if (inet->pmtudisc != IP_PMTUDISC_DONT) {
+			if (READ_ONCE(inet->pmtudisc) != IP_PMTUDISC_DONT) {
 				err = EMSGSIZE;
 				harderr = 1;
 				break;
diff --git a/net/netfilter/ipvs/ip_vs_sync.c b/net/netfilter/ipvs/ip_vs_sync.c
index 3eed1670224888acf639cff06537ddf2505461bb..4f6c795588fbdbf084154025b8172e0fd2ea7384 100644
--- a/net/netfilter/ipvs/ip_vs_sync.c
+++ b/net/netfilter/ipvs/ip_vs_sync.c
@@ -1335,7 +1335,7 @@ static void set_mcast_pmtudisc(struct sock *sk, int val)
 
 	/* setsockopt(sock, SOL_IP, IP_MTU_DISCOVER, &val, sizeof(val)); */
 	lock_sock(sk);
-	inet->pmtudisc = val;
+	WRITE_ONCE(inet->pmtudisc, val);
 #ifdef CONFIG_IP_VS_IPV6
 	if (sk->sk_family == AF_INET6) {
 		struct ipv6_pinfo *np = inet6_sk(sk);
-- 
2.42.0.459.ge4e396fd5e-goog

[PATCH net-next 3/8] inet: implement lockless IP_TOS

[Eric Dumazet]

Some reads of inet->tos are racy.

Add needed READ_ONCE() annotations and convert IP_TOS option lockless.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/ip.h                              |  1 -
 net/dccp/ipv4.c                               |  2 +-
 net/ipv4/inet_diag.c                          |  2 +-
 net/ipv4/ip_output.c                          |  4 +--
 net/ipv4/ip_sockglue.c                        | 29 ++++++++-----------
 net/ipv4/tcp_ipv4.c                           |  9 +++---
 net/mptcp/sockopt.c                           |  8 ++---
 net/sctp/protocol.c                           |  4 +--
 .../selftests/net/mptcp/mptcp_connect.sh      |  2 +-
 9 files changed, 28 insertions(+), 33 deletions(-)

diff --git a/include/net/ip.h b/include/net/ip.h
index 46933a0d98eac2db40c2e88006125588b8f8143e..8836ee5502669f6d3a5fc35a045695bac1c3b1a9 100644
--- a/include/net/ip.h
+++ b/include/net/ip.h
@@ -810,6 +810,5 @@ int ip_sock_set_mtu_discover(struct sock *sk, int val);
 void ip_sock_set_pktinfo(struct sock *sk);
 void ip_sock_set_recverr(struct sock *sk);
 void ip_sock_set_tos(struct sock *sk, int val);
-void  __ip_sock_set_tos(struct sock *sk, int val);
 
 #endif	/* _IP_H */
diff --git a/net/dccp/ipv4.c b/net/dccp/ipv4.c
index 8f56e8723c7386c9f9344f1376823bfd0077c8c2..ef55e4c99e5109d68da5016e5c30e01fa50722be 100644
--- a/net/dccp/ipv4.c
+++ b/net/dccp/ipv4.c
@@ -516,7 +516,7 @@ static int dccp_v4_send_response(const struct sock *sk, struct request_sock *req
 		err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
 					    ireq->ir_rmt_addr,
 					    rcu_dereference(ireq->ireq_opt),
-					    inet_sk(sk)->tos);
+					    READ_ONCE(inet_sk(sk)->tos));
 		rcu_read_unlock();
 		err = net_xmit_eval(err);
 	}
diff --git a/net/ipv4/inet_diag.c b/net/ipv4/inet_diag.c
index e13a84433413ed88088435ff8e11efeb30fc3cca..1f2d7a8bd060e59baeb00fcb1c6aabfcb3bb213d 100644
--- a/net/ipv4/inet_diag.c
+++ b/net/ipv4/inet_diag.c
@@ -134,7 +134,7 @@ int inet_diag_msg_attrs_fill(struct sock *sk, struct sk_buff *skb,
 	 * hence this needs to be included regardless of socket family.
 	 */
 	if (ext & (1 << (INET_DIAG_TOS - 1)))
-		if (nla_put_u8(skb, INET_DIAG_TOS, inet->tos) < 0)
+		if (nla_put_u8(skb, INET_DIAG_TOS, READ_ONCE(inet->tos)) < 0)
 			goto errout;
 
 #if IS_ENABLED(CONFIG_IPV6)
diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c
index 2be281f184a5fe5a695ccd51fabe69fa45bea0b8..85320f92e8363d59e92c54139044cbab7e0561fa 100644
--- a/net/ipv4/ip_output.c
+++ b/net/ipv4/ip_output.c
@@ -544,7 +544,7 @@ EXPORT_SYMBOL(__ip_queue_xmit);
 
 int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl)
 {
-	return __ip_queue_xmit(sk, skb, fl, inet_sk(sk)->tos);
+	return __ip_queue_xmit(sk, skb, fl, READ_ONCE(inet_sk(sk)->tos));
 }
 EXPORT_SYMBOL(ip_queue_xmit);
 
@@ -1438,7 +1438,7 @@ struct sk_buff *__ip_make_skb(struct sock *sk,
 	iph = ip_hdr(skb);
 	iph->version = 4;
 	iph->ihl = 5;
-	iph->tos = (cork->tos != -1) ? cork->tos : inet->tos;
+	iph->tos = (cork->tos != -1) ? cork->tos : READ_ONCE(inet->tos);
 	iph->frag_off = df;
 	iph->ttl = ttl;
 	iph->protocol = sk->sk_protocol;
diff --git a/net/ipv4/ip_sockglue.c b/net/ipv4/ip_sockglue.c
index 6d874cc03c8b4e88d79ebc50a6db105606b6ae60..50c008efbb6de7303621dd30b178c90cb3f5a2fc 100644
--- a/net/ipv4/ip_sockglue.c
+++ b/net/ipv4/ip_sockglue.c
@@ -585,25 +585,20 @@ int ip_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
 	return err;
 }
 
-void __ip_sock_set_tos(struct sock *sk, int val)
+void ip_sock_set_tos(struct sock *sk, int val)
 {
+	u8 old_tos = READ_ONCE(inet_sk(sk)->tos);
+
 	if (sk->sk_type == SOCK_STREAM) {
 		val &= ~INET_ECN_MASK;
-		val |= inet_sk(sk)->tos & INET_ECN_MASK;
+		val |= old_tos & INET_ECN_MASK;
 	}
-	if (inet_sk(sk)->tos != val) {
-		inet_sk(sk)->tos = val;
+	if (old_tos != val) {
+		WRITE_ONCE(inet_sk(sk)->tos, val);
 		WRITE_ONCE(sk->sk_priority, rt_tos2priority(val));
 		sk_dst_reset(sk);
 	}
 }
-
-void ip_sock_set_tos(struct sock *sk, int val)
-{
-	lock_sock(sk);
-	__ip_sock_set_tos(sk, val);
-	release_sock(sk);
-}
 EXPORT_SYMBOL(ip_sock_set_tos);
 
 void ip_sock_set_freebind(struct sock *sk)
@@ -1050,6 +1045,9 @@ int do_ip_setsockopt(struct sock *sk, int level, int optname,
 		return 0;
 	case IP_MTU_DISCOVER:
 		return ip_sock_set_mtu_discover(sk, val);
+	case IP_TOS:	/* This sets both TOS and Precedence */
+		ip_sock_set_tos(sk, val);
+		return 0;
 	}
 
 	err = 0;
@@ -1104,9 +1102,6 @@ int do_ip_setsockopt(struct sock *sk, int level, int optname,
 			}
 		}
 		break;
-	case IP_TOS:	/* This sets both TOS and Precedence */
-		__ip_sock_set_tos(sk, val);
-		break;
 	case IP_UNICAST_IF:
 	{
 		struct net_device *dev = NULL;
@@ -1593,6 +1588,9 @@ int do_ip_getsockopt(struct sock *sk, int level, int optname,
 	case IP_MTU_DISCOVER:
 		val = READ_ONCE(inet->pmtudisc);
 		goto copyval;
+	case IP_TOS:
+		val = READ_ONCE(inet->tos);
+		goto copyval;
 	}
 
 	if (needs_rtnl)
@@ -1629,9 +1627,6 @@ int do_ip_getsockopt(struct sock *sk, int level, int optname,
 			return -EFAULT;
 		return 0;
 	}
-	case IP_TOS:
-		val = inet->tos;
-		break;
 	case IP_MTU:
 	{
 		struct dst_entry *dst;
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index f13eb7e23d03f3681055257e6ebea0612ae3f9b3..1f89ba58e71eff74d8ed75019de9e70d2f4d5926 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -1024,10 +1024,11 @@ static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
 	if (skb) {
 		__tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
 
-		tos = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos) ?
-				(tcp_rsk(req)->syn_tos & ~INET_ECN_MASK) |
-				(inet_sk(sk)->tos & INET_ECN_MASK) :
-				inet_sk(sk)->tos;
+		tos = READ_ONCE(inet_sk(sk)->tos);
+
+		if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos))
+			tos = (tcp_rsk(req)->syn_tos & ~INET_ECN_MASK) |
+			      (tos & INET_ECN_MASK);
 
 		if (!INET_ECN_is_capable(tos) &&
 		    tcp_bpf_ca_needs_ecn((struct sock *)req))
diff --git a/net/mptcp/sockopt.c b/net/mptcp/sockopt.c
index 8260202c00669fd7d2eed2f94a3c2cf225a0d89c..155e8472ba9b83c35c6f827b2bb35c0be4127917 100644
--- a/net/mptcp/sockopt.c
+++ b/net/mptcp/sockopt.c
@@ -734,11 +734,11 @@ static int mptcp_setsockopt_v4_set_tos(struct mptcp_sock *msk, int optname,
 
 	lock_sock(sk);
 	sockopt_seq_inc(msk);
-	val = inet_sk(sk)->tos;
+	val = READ_ONCE(inet_sk(sk)->tos);
 	mptcp_for_each_subflow(msk, subflow) {
 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 
-		__ip_sock_set_tos(ssk, val);
+		ip_sock_set_tos(ssk, val);
 	}
 	release_sock(sk);
 
@@ -1343,7 +1343,7 @@ static int mptcp_getsockopt_v4(struct mptcp_sock *msk, int optname,
 
 	switch (optname) {
 	case IP_TOS:
-		return mptcp_put_int_option(msk, optval, optlen, inet_sk(sk)->tos);
+		return mptcp_put_int_option(msk, optval, optlen, READ_ONCE(inet_sk(sk)->tos));
 	}
 
 	return -EOPNOTSUPP;
@@ -1411,7 +1411,7 @@ static void sync_socket_options(struct mptcp_sock *msk, struct sock *ssk)
 	ssk->sk_bound_dev_if = sk->sk_bound_dev_if;
 	ssk->sk_incoming_cpu = sk->sk_incoming_cpu;
 	ssk->sk_ipv6only = sk->sk_ipv6only;
-	__ip_sock_set_tos(ssk, inet_sk(sk)->tos);
+	ip_sock_set_tos(ssk, inet_sk(sk)->tos);
 
 	if (sk->sk_userlocks & tx_rx_locks) {
 		ssk->sk_userlocks |= sk->sk_userlocks & tx_rx_locks;
diff --git a/net/sctp/protocol.c b/net/sctp/protocol.c
index 2185f44198deb002bc8ed7f1b0f3fe02d6bb9f09..94c6dd53cd62d1fa6236d07946e8d5ff68eb587d 100644
--- a/net/sctp/protocol.c
+++ b/net/sctp/protocol.c
@@ -426,7 +426,7 @@ static void sctp_v4_get_dst(struct sctp_transport *t, union sctp_addr *saddr,
 	struct dst_entry *dst = NULL;
 	union sctp_addr *daddr = &t->ipaddr;
 	union sctp_addr dst_saddr;
-	__u8 tos = inet_sk(sk)->tos;
+	u8 tos = READ_ONCE(inet_sk(sk)->tos);
 
 	if (t->dscp & SCTP_DSCP_SET_MASK)
 		tos = t->dscp & SCTP_DSCP_VAL_MASK;
@@ -1057,7 +1057,7 @@ static inline int sctp_v4_xmit(struct sk_buff *skb, struct sctp_transport *t)
 	struct flowi4 *fl4 = &t->fl.u.ip4;
 	struct sock *sk = skb->sk;
 	struct inet_sock *inet = inet_sk(sk);
-	__u8 dscp = inet->tos;
+	__u8 dscp = READ_ONCE(inet->tos);
 	__be16 df = 0;
 
 	pr_debug("%s: skb:%p, len:%d, src:%pI4, dst:%pI4\n", __func__, skb,
diff --git a/tools/testing/selftests/net/mptcp/mptcp_connect.sh b/tools/testing/selftests/net/mptcp/mptcp_connect.sh
index b1fc8afd072dc6ddde8d561a675a5549a9a37dba..61a2a1988ce69ffa17e0dd8e629eac550f4f7d99 100755
--- a/tools/testing/selftests/net/mptcp/mptcp_connect.sh
+++ b/tools/testing/selftests/net/mptcp/mptcp_connect.sh
@@ -716,7 +716,7 @@ run_test_transparent()
 	# the required infrastructure in MPTCP sockopt code. To support TOS, the
 	# following function has been exported (T). Not great but better than
 	# checking for a specific kernel version.
-	if ! mptcp_lib_kallsyms_has "T __ip_sock_set_tos$"; then
+	if ! mptcp_lib_kallsyms_has "T ip_sock_set_tos$"; then
 		echo "INFO: ${msg} not supported by the kernel: SKIP"
 		mptcp_lib_result_skip "${TEST_GROUP}"
 		return
-- 
2.42.0.459.ge4e396fd5e-goog

[PATCH net-next 4/8] inet: lockless getsockopt(IP_OPTIONS)

[Eric Dumazet]

inet->inet_opt being RCU protected, we can use RCU instead
of locking the socket.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 net/ipv4/ip_sockglue.c | 20 ++++++++++----------
 1 file changed, 10 insertions(+), 10 deletions(-)

diff --git a/net/ipv4/ip_sockglue.c b/net/ipv4/ip_sockglue.c
index 50c008efbb6de7303621dd30b178c90cb3f5a2fc..45d89487914a12061f05c192004ad79f0abbf756 100644
--- a/net/ipv4/ip_sockglue.c
+++ b/net/ipv4/ip_sockglue.c
@@ -1591,27 +1591,20 @@ int do_ip_getsockopt(struct sock *sk, int level, int optname,
 	case IP_TOS:
 		val = READ_ONCE(inet->tos);
 		goto copyval;
-	}
-
-	if (needs_rtnl)
-		rtnl_lock();
-	sockopt_lock_sock(sk);
-
-	switch (optname) {
 	case IP_OPTIONS:
 	{
 		unsigned char optbuf[sizeof(struct ip_options)+40];
 		struct ip_options *opt = (struct ip_options *)optbuf;
 		struct ip_options_rcu *inet_opt;
 
-		inet_opt = rcu_dereference_protected(inet->inet_opt,
-						     lockdep_sock_is_held(sk));
+		rcu_read_lock();
+		inet_opt = rcu_dereference(inet->inet_opt);
 		opt->optlen = 0;
 		if (inet_opt)
 			memcpy(optbuf, &inet_opt->opt,
 			       sizeof(struct ip_options) +
 			       inet_opt->opt.optlen);
-		sockopt_release_sock(sk);
+		rcu_read_unlock();
 
 		if (opt->optlen == 0) {
 			len = 0;
@@ -1627,6 +1620,13 @@ int do_ip_getsockopt(struct sock *sk, int level, int optname,
 			return -EFAULT;
 		return 0;
 	}
+	}
+
+	if (needs_rtnl)
+		rtnl_lock();
+	sockopt_lock_sock(sk);
+
+	switch (optname) {
 	case IP_MTU:
 	{
 		struct dst_entry *dst;
-- 
2.42.0.459.ge4e396fd5e-goog

[PATCH net-next 5/8] inet: lockless getsockopt(IP_MTU)

[Eric Dumazet]

sk_dst_get() does not require socket lock.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 net/ipv4/ip_sockglue.c | 20 +++++++++-----------
 1 file changed, 9 insertions(+), 11 deletions(-)

diff --git a/net/ipv4/ip_sockglue.c b/net/ipv4/ip_sockglue.c
index 45d89487914a12061f05c192004ad79f0abbf756..04579e390ddd4dadb8a107ef0b5da15e7a60f1ff 100644
--- a/net/ipv4/ip_sockglue.c
+++ b/net/ipv4/ip_sockglue.c
@@ -1620,13 +1620,6 @@ int do_ip_getsockopt(struct sock *sk, int level, int optname,
 			return -EFAULT;
 		return 0;
 	}
-	}
-
-	if (needs_rtnl)
-		rtnl_lock();
-	sockopt_lock_sock(sk);
-
-	switch (optname) {
 	case IP_MTU:
 	{
 		struct dst_entry *dst;
@@ -1636,12 +1629,17 @@ int do_ip_getsockopt(struct sock *sk, int level, int optname,
 			val = dst_mtu(dst);
 			dst_release(dst);
 		}
-		if (!val) {
-			sockopt_release_sock(sk);
+		if (!val)
 			return -ENOTCONN;
-		}
-		break;
+		goto copyval;
+	}
 	}
+
+	if (needs_rtnl)
+		rtnl_lock();
+	sockopt_lock_sock(sk);
+
+	switch (optname) {
 	case IP_UNICAST_IF:
 		val = (__force int)htonl((__u32) inet->uc_index);
 		break;
-- 
2.42.0.459.ge4e396fd5e-goog

[PATCH net-next 6/8] inet: implement lockless getsockopt(IP_UNICAST_IF)

[Eric Dumazet]

Add missing READ_ONCE() annotations when reading inet->uc_index

Implementing getsockopt(IP_UNICAST_IF) locklessly seems possible,
the setsockopt() part might not be possible at the moment.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 net/ipv4/datagram.c    |  2 +-
 net/ipv4/ip_sockglue.c | 10 +++++-----
 net/ipv4/ping.c        |  2 +-
 net/ipv4/raw.c         | 13 +++++++------
 net/ipv4/udp.c         | 12 +++++++-----
 5 files changed, 21 insertions(+), 18 deletions(-)

diff --git a/net/ipv4/datagram.c b/net/ipv4/datagram.c
index cb5dbee9e018fbba1bc1e5705e8bec6c4203af56..1480e9ebdfef445960e1f70f34f33a0e0c52b65b 100644
--- a/net/ipv4/datagram.c
+++ b/net/ipv4/datagram.c
@@ -43,7 +43,7 @@ int __ip4_datagram_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len
 		if (!saddr)
 			saddr = inet->mc_addr;
 	} else if (!oif) {
-		oif = inet->uc_index;
+		oif = READ_ONCE(inet->uc_index);
 	}
 	fl4 = &inet->cork.fl.u.ip4;
 	rt = ip_route_connect(fl4, usin->sin_addr.s_addr, saddr, oif,
diff --git a/net/ipv4/ip_sockglue.c b/net/ipv4/ip_sockglue.c
index 04579e390ddd4dadb8a107ef0b5da15e7a60f1ff..58995526c6e965d613b8cdea61b84916d608a6fb 100644
--- a/net/ipv4/ip_sockglue.c
+++ b/net/ipv4/ip_sockglue.c
@@ -1113,7 +1113,7 @@ int do_ip_setsockopt(struct sock *sk, int level, int optname,
 
 		ifindex = (__force int)ntohl((__force __be32)val);
 		if (ifindex == 0) {
-			inet->uc_index = 0;
+			WRITE_ONCE(inet->uc_index, 0);
 			err = 0;
 			break;
 		}
@@ -1130,7 +1130,7 @@ int do_ip_setsockopt(struct sock *sk, int level, int optname,
 		if (sk->sk_bound_dev_if && midx != sk->sk_bound_dev_if)
 			break;
 
-		inet->uc_index = ifindex;
+		WRITE_ONCE(inet->uc_index, ifindex);
 		err = 0;
 		break;
 	}
@@ -1633,6 +1633,9 @@ int do_ip_getsockopt(struct sock *sk, int level, int optname,
 			return -ENOTCONN;
 		goto copyval;
 	}
+	case IP_UNICAST_IF:
+		val = (__force int)htonl((__u32) READ_ONCE(inet->uc_index));
+		goto copyval;
 	}
 
 	if (needs_rtnl)
@@ -1640,9 +1643,6 @@ int do_ip_getsockopt(struct sock *sk, int level, int optname,
 	sockopt_lock_sock(sk);
 
 	switch (optname) {
-	case IP_UNICAST_IF:
-		val = (__force int)htonl((__u32) inet->uc_index);
-		break;
 	case IP_MULTICAST_IF:
 	{
 		struct in_addr addr;
diff --git a/net/ipv4/ping.c b/net/ipv4/ping.c
index 50d12b0c8d46fdcd9b448c3ebc90395ebf426075..66ad1f95af49f222afe0ee75b9163dd0af0a2c49 100644
--- a/net/ipv4/ping.c
+++ b/net/ipv4/ping.c
@@ -777,7 +777,7 @@ static int ping_v4_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
 		if (!saddr)
 			saddr = inet->mc_addr;
 	} else if (!ipc.oif)
-		ipc.oif = inet->uc_index;
+		ipc.oif = READ_ONCE(inet->uc_index);
 
 	flowi4_init_output(&fl4, ipc.oif, ipc.sockc.mark, tos, scope,
 			   sk->sk_protocol, inet_sk_flowi_flags(sk), faddr,
diff --git a/net/ipv4/raw.c b/net/ipv4/raw.c
index ade1aecd7c71184d753a28a67bc9b30087247db4..e2357d23202e5a39832bb1550c365de9a836c363 100644
--- a/net/ipv4/raw.c
+++ b/net/ipv4/raw.c
@@ -482,7 +482,7 @@ static int raw_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
 	int free = 0;
 	__be32 daddr;
 	__be32 saddr;
-	int err;
+	int uc_index, err;
 	struct ip_options_data opt_copy;
 	struct raw_frag_vec rfv;
 	int hdrincl;
@@ -576,24 +576,25 @@ static int raw_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
 	tos = get_rttos(&ipc, inet);
 	scope = ip_sendmsg_scope(inet, &ipc, msg);
 
+	uc_index = READ_ONCE(inet->uc_index);
 	if (ipv4_is_multicast(daddr)) {
 		if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
 			ipc.oif = inet->mc_index;
 		if (!saddr)
 			saddr = inet->mc_addr;
 	} else if (!ipc.oif) {
-		ipc.oif = inet->uc_index;
-	} else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
+		ipc.oif = uc_index;
+	} else if (ipv4_is_lbcast(daddr) && uc_index) {
 		/* oif is set, packet is to local broadcast
 		 * and uc_index is set. oif is most likely set
 		 * by sk_bound_dev_if. If uc_index != oif check if the
 		 * oif is an L3 master and uc_index is an L3 slave.
 		 * If so, we want to allow the send using the uc_index.
 		 */
-		if (ipc.oif != inet->uc_index &&
+		if (ipc.oif != uc_index &&
 		    ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
-							      inet->uc_index)) {
-			ipc.oif = inet->uc_index;
+							      uc_index)) {
+			ipc.oif = uc_index;
 		}
 	}
 
diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
index 731a723dc80816f0b5b0803d7397f7e9e8cd8b09..1e0c3aba1e5a88c7ba50a28511412a1710f1bab5 100644
--- a/net/ipv4/udp.c
+++ b/net/ipv4/udp.c
@@ -1055,6 +1055,7 @@ int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
 	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
 	struct sk_buff *skb;
 	struct ip_options_data opt_copy;
+	int uc_index;
 
 	if (len > 0xFFFF)
 		return -EMSGSIZE;
@@ -1173,6 +1174,7 @@ int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
 	if (scope == RT_SCOPE_LINK)
 		connected = 0;
 
+	uc_index = READ_ONCE(inet->uc_index);
 	if (ipv4_is_multicast(daddr)) {
 		if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
 			ipc.oif = inet->mc_index;
@@ -1180,18 +1182,18 @@ int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
 			saddr = inet->mc_addr;
 		connected = 0;
 	} else if (!ipc.oif) {
-		ipc.oif = inet->uc_index;
-	} else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
+		ipc.oif = uc_index;
+	} else if (ipv4_is_lbcast(daddr) && uc_index) {
 		/* oif is set, packet is to local broadcast and
 		 * uc_index is set. oif is most likely set
 		 * by sk_bound_dev_if. If uc_index != oif check if the
 		 * oif is an L3 master and uc_index is an L3 slave.
 		 * If so, we want to allow the send using the uc_index.
 		 */
-		if (ipc.oif != inet->uc_index &&
+		if (ipc.oif != uc_index &&
 		    ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
-							      inet->uc_index)) {
-			ipc.oif = inet->uc_index;
+							      uc_index)) {
+			ipc.oif = uc_index;
 		}
 	}
 
-- 
2.42.0.459.ge4e396fd5e-goog

[PATCH net-next 7/8] inet: lockless IP_PKTOPTIONS implementation

[Eric Dumazet]

Current implementation is already lockless, because the socket
lock is released before reading socket fields.

Add missing READ_ONCE() annotations.

Note that corresponding WRITE_ONCE() are needed, the order
of the patches do not really matter.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 net/ipv4/ip_sockglue.c | 76 ++++++++++++++++++++----------------------
 1 file changed, 37 insertions(+), 39 deletions(-)

diff --git a/net/ipv4/ip_sockglue.c b/net/ipv4/ip_sockglue.c
index 58995526c6e965d613b8cdea61b84916d608a6fb..1ee01ff64171c94b6b244589518a53ce807a212d 100644
--- a/net/ipv4/ip_sockglue.c
+++ b/net/ipv4/ip_sockglue.c
@@ -1633,6 +1633,43 @@ int do_ip_getsockopt(struct sock *sk, int level, int optname,
 			return -ENOTCONN;
 		goto copyval;
 	}
+	case IP_PKTOPTIONS:
+	{
+		struct msghdr msg;
+
+		if (sk->sk_type != SOCK_STREAM)
+			return -ENOPROTOOPT;
+
+		if (optval.is_kernel) {
+			msg.msg_control_is_user = false;
+			msg.msg_control = optval.kernel;
+		} else {
+			msg.msg_control_is_user = true;
+			msg.msg_control_user = optval.user;
+		}
+		msg.msg_controllen = len;
+		msg.msg_flags = in_compat_syscall() ? MSG_CMSG_COMPAT : 0;
+
+		if (inet_test_bit(PKTINFO, sk)) {
+			struct in_pktinfo info;
+
+			info.ipi_addr.s_addr = READ_ONCE(inet->inet_rcv_saddr);
+			info.ipi_spec_dst.s_addr = READ_ONCE(inet->inet_rcv_saddr);
+			info.ipi_ifindex = READ_ONCE(inet->mc_index);
+			put_cmsg(&msg, SOL_IP, IP_PKTINFO, sizeof(info), &info);
+		}
+		if (inet_test_bit(TTL, sk)) {
+			int hlim = READ_ONCE(inet->mc_ttl);
+
+			put_cmsg(&msg, SOL_IP, IP_TTL, sizeof(hlim), &hlim);
+		}
+		if (inet_test_bit(TOS, sk)) {
+			int tos = READ_ONCE(inet->rcv_tos);
+			put_cmsg(&msg, SOL_IP, IP_TOS, sizeof(tos), &tos);
+		}
+		len -= msg.msg_controllen;
+		return copy_to_sockptr(optlen, &len, sizeof(int));
+	}
 	case IP_UNICAST_IF:
 		val = (__force int)htonl((__u32) READ_ONCE(inet->uc_index));
 		goto copyval;
@@ -1678,45 +1715,6 @@ int do_ip_getsockopt(struct sock *sk, int level, int optname,
 		else
 			err = ip_get_mcast_msfilter(sk, optval, optlen, len);
 		goto out;
-	case IP_PKTOPTIONS:
-	{
-		struct msghdr msg;
-
-		sockopt_release_sock(sk);
-
-		if (sk->sk_type != SOCK_STREAM)
-			return -ENOPROTOOPT;
-
-		if (optval.is_kernel) {
-			msg.msg_control_is_user = false;
-			msg.msg_control = optval.kernel;
-		} else {
-			msg.msg_control_is_user = true;
-			msg.msg_control_user = optval.user;
-		}
-		msg.msg_controllen = len;
-		msg.msg_flags = in_compat_syscall() ? MSG_CMSG_COMPAT : 0;
-
-		if (inet_test_bit(PKTINFO, sk)) {
-			struct in_pktinfo info;
-
-			info.ipi_addr.s_addr = inet->inet_rcv_saddr;
-			info.ipi_spec_dst.s_addr = inet->inet_rcv_saddr;
-			info.ipi_ifindex = inet->mc_index;
-			put_cmsg(&msg, SOL_IP, IP_PKTINFO, sizeof(info), &info);
-		}
-		if (inet_test_bit(TTL, sk)) {
-			int hlim = READ_ONCE(inet->mc_ttl);
-
-			put_cmsg(&msg, SOL_IP, IP_TTL, sizeof(hlim), &hlim);
-		}
-		if (inet_test_bit(TOS, sk)) {
-			int tos = inet->rcv_tos;
-			put_cmsg(&msg, SOL_IP, IP_TOS, sizeof(tos), &tos);
-		}
-		len -= msg.msg_controllen;
-		return copy_to_sockptr(optlen, &len, sizeof(int));
-	}
 	case IP_LOCAL_PORT_RANGE:
 		val = inet->local_port_range.hi << 16 | inet->local_port_range.lo;
 		break;
-- 
2.42.0.459.ge4e396fd5e-goog

[PATCH net-next 8/8] inet: implement lockless getsockopt(IP_MULTICAST_IF)

[Eric Dumazet]

Add missing annotations to inet->mc_index and inet->mc_addr
to fix data-races.

getsockopt(IP_MULTICAST_IF) can be lockless.

setsockopt() side is left for later.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 net/ipv4/datagram.c    |  4 ++--
 net/ipv4/ip_sockglue.c | 25 ++++++++++++-------------
 net/ipv4/ping.c        |  4 ++--
 net/ipv4/raw.c         |  4 ++--
 net/ipv4/udp.c         |  4 ++--
 5 files changed, 20 insertions(+), 21 deletions(-)

diff --git a/net/ipv4/datagram.c b/net/ipv4/datagram.c
index 1480e9ebdfef445960e1f70f34f33a0e0c52b65b..2cc50cbfc2a31ec91fbdc4a541cb89df689cd9ae 100644
--- a/net/ipv4/datagram.c
+++ b/net/ipv4/datagram.c
@@ -39,9 +39,9 @@ int __ip4_datagram_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len
 	saddr = inet->inet_saddr;
 	if (ipv4_is_multicast(usin->sin_addr.s_addr)) {
 		if (!oif || netif_index_is_l3_master(sock_net(sk), oif))
-			oif = inet->mc_index;
+			oif = READ_ONCE(inet->mc_index);
 		if (!saddr)
-			saddr = inet->mc_addr;
+			saddr = READ_ONCE(inet->mc_addr);
 	} else if (!oif) {
 		oif = READ_ONCE(inet->uc_index);
 	}
diff --git a/net/ipv4/ip_sockglue.c b/net/ipv4/ip_sockglue.c
index 1ee01ff64171c94b6b244589518a53ce807a212d..0b74ac49d6a6f82f5e8ffe5279dba3baf30f874e 100644
--- a/net/ipv4/ip_sockglue.c
+++ b/net/ipv4/ip_sockglue.c
@@ -1168,8 +1168,8 @@ int do_ip_setsockopt(struct sock *sk, int level, int optname,
 
 		if (!mreq.imr_ifindex) {
 			if (mreq.imr_address.s_addr == htonl(INADDR_ANY)) {
-				inet->mc_index = 0;
-				inet->mc_addr  = 0;
+				WRITE_ONCE(inet->mc_index, 0);
+				WRITE_ONCE(inet->mc_addr, 0);
 				err = 0;
 				break;
 			}
@@ -1194,8 +1194,8 @@ int do_ip_setsockopt(struct sock *sk, int level, int optname,
 		    midx != sk->sk_bound_dev_if)
 			break;
 
-		inet->mc_index = mreq.imr_ifindex;
-		inet->mc_addr  = mreq.imr_address.s_addr;
+		WRITE_ONCE(inet->mc_index, mreq.imr_ifindex);
+		WRITE_ONCE(inet->mc_addr, mreq.imr_address.s_addr);
 		err = 0;
 		break;
 	}
@@ -1673,19 +1673,11 @@ int do_ip_getsockopt(struct sock *sk, int level, int optname,
 	case IP_UNICAST_IF:
 		val = (__force int)htonl((__u32) READ_ONCE(inet->uc_index));
 		goto copyval;
-	}
-
-	if (needs_rtnl)
-		rtnl_lock();
-	sockopt_lock_sock(sk);
-
-	switch (optname) {
 	case IP_MULTICAST_IF:
 	{
 		struct in_addr addr;
 		len = min_t(unsigned int, len, sizeof(struct in_addr));
-		addr.s_addr = inet->mc_addr;
-		sockopt_release_sock(sk);
+		addr.s_addr = READ_ONCE(inet->mc_addr);
 
 		if (copy_to_sockptr(optlen, &len, sizeof(int)))
 			return -EFAULT;
@@ -1693,6 +1685,13 @@ int do_ip_getsockopt(struct sock *sk, int level, int optname,
 			return -EFAULT;
 		return 0;
 	}
+	}
+
+	if (needs_rtnl)
+		rtnl_lock();
+	sockopt_lock_sock(sk);
+
+	switch (optname) {
 	case IP_MSFILTER:
 	{
 		struct ip_msfilter msf;
diff --git a/net/ipv4/ping.c b/net/ipv4/ping.c
index 66ad1f95af49f222afe0ee75b9163dd0af0a2c49..2c61f444e1c7d322e75e020c41af02977d8814f0 100644
--- a/net/ipv4/ping.c
+++ b/net/ipv4/ping.c
@@ -773,9 +773,9 @@ static int ping_v4_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
 
 	if (ipv4_is_multicast(daddr)) {
 		if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
-			ipc.oif = inet->mc_index;
+			ipc.oif = READ_ONCE(inet->mc_index);
 		if (!saddr)
-			saddr = inet->mc_addr;
+			saddr = READ_ONCE(inet->mc_addr);
 	} else if (!ipc.oif)
 		ipc.oif = READ_ONCE(inet->uc_index);
 
diff --git a/net/ipv4/raw.c b/net/ipv4/raw.c
index e2357d23202e5a39832bb1550c365de9a836c363..27da9d7294c0b4fb9027bb7feb704063dc6302db 100644
--- a/net/ipv4/raw.c
+++ b/net/ipv4/raw.c
@@ -579,9 +579,9 @@ static int raw_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
 	uc_index = READ_ONCE(inet->uc_index);
 	if (ipv4_is_multicast(daddr)) {
 		if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
-			ipc.oif = inet->mc_index;
+			ipc.oif = READ_ONCE(inet->mc_index);
 		if (!saddr)
-			saddr = inet->mc_addr;
+			saddr = READ_ONCE(inet->mc_addr);
 	} else if (!ipc.oif) {
 		ipc.oif = uc_index;
 	} else if (ipv4_is_lbcast(daddr) && uc_index) {
diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
index 1e0c3aba1e5a88c7ba50a28511412a1710f1bab5..7f7724beca33781f8ff12750d1c9c9ccc420f481 100644
--- a/net/ipv4/udp.c
+++ b/net/ipv4/udp.c
@@ -1177,9 +1177,9 @@ int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
 	uc_index = READ_ONCE(inet->uc_index);
 	if (ipv4_is_multicast(daddr)) {
 		if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
-			ipc.oif = inet->mc_index;
+			ipc.oif = READ_ONCE(inet->mc_index);
 		if (!saddr)
-			saddr = inet->mc_addr;
+			saddr = READ_ONCE(inet->mc_addr);
 		connected = 0;
 	} else if (!ipc.oif) {
 		ipc.oif = uc_index;
-- 
2.42.0.459.ge4e396fd5e-goog

Re: [PATCH net-next 1/8] inet: implement lockless IP_MULTICAST_TTL

[David Ahern]

On 9/21/23 7:30 AM, Eric Dumazet wrote:
> inet->mc_ttl can be read locklessly.
> 
> Implement proper lockless reads and writes to inet->mc_ttl
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> ---
>  net/ipv4/ip_output.c            |  2 +-
>  net/ipv4/ip_sockglue.c          | 31 ++++++++++++++++---------------
>  net/netfilter/ipvs/ip_vs_sync.c |  2 +-
>  3 files changed, 18 insertions(+), 17 deletions(-)
> 

Reviewed-by: David Ahern <dsahern@kernel.org>

Re: [PATCH net-next 2/8] inet: implement lockless IP_MTU_DISCOVER

[David Ahern]

On 9/21/23 7:30 AM, Eric Dumazet wrote:
> inet->pmtudisc can be read locklessly.
> 
> Implement proper lockless reads and writes to inet->pmtudisc
> 
> ip_sock_set_mtu_discover() can now be called from arbitrary
> contexts.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> ---
>  include/net/ip.h                | 13 ++++++++-----
>  net/ipv4/ip_output.c            |  7 ++++---
>  net/ipv4/ip_sockglue.c          | 17 ++++++-----------
>  net/ipv4/ping.c                 |  2 +-
>  net/ipv4/raw.c                  |  2 +-
>  net/ipv4/udp.c                  |  2 +-
>  net/netfilter/ipvs/ip_vs_sync.c |  2 +-
>  7 files changed, 22 insertions(+), 23 deletions(-)
> 

Reviewed-by: David Ahern <dsahern@kernel.org>

Re: [PATCH net-next 3/8] inet: implement lockless IP_TOS

[David Ahern]

On 9/21/23 7:30 AM, Eric Dumazet wrote:
> Some reads of inet->tos are racy.
> 
> Add needed READ_ONCE() annotations and convert IP_TOS option lockless.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> ---
>  include/net/ip.h                              |  1 -
>  net/dccp/ipv4.c                               |  2 +-
>  net/ipv4/inet_diag.c                          |  2 +-
>  net/ipv4/ip_output.c                          |  4 +--
>  net/ipv4/ip_sockglue.c                        | 29 ++++++++-----------
>  net/ipv4/tcp_ipv4.c                           |  9 +++---
>  net/mptcp/sockopt.c                           |  8 ++---
>  net/sctp/protocol.c                           |  4 +--
>  .../selftests/net/mptcp/mptcp_connect.sh      |  2 +-
>  9 files changed, 28 insertions(+), 33 deletions(-)
> 

include/net/route.h dereferences sk tos as well.
net/ipv4/icmp.c has a setting of it

Re: [PATCH net-next 4/8] inet: lockless getsockopt(IP_OPTIONS)

[David Ahern]

On 9/21/23 7:30 AM, Eric Dumazet wrote:
> inet->inet_opt being RCU protected, we can use RCU instead
> of locking the socket.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> ---
>  net/ipv4/ip_sockglue.c | 20 ++++++++++----------
>  1 file changed, 10 insertions(+), 10 deletions(-)
> 

Reviewed-by: David Ahern <dsahern@kernel.org>

Re: [PATCH net-next 5/8] inet: lockless getsockopt(IP_MTU)

[David Ahern]

On 9/21/23 7:30 AM, Eric Dumazet wrote:
> sk_dst_get() does not require socket lock.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> ---
>  net/ipv4/ip_sockglue.c | 20 +++++++++-----------
>  1 file changed, 9 insertions(+), 11 deletions(-)
> 

Reviewed-by: David Ahern <dsahern@kernel.org>

Re: [PATCH net-next 3/8] inet: implement lockless IP_TOS

[Eric Dumazet]

On Thu, Sep 21, 2023 at 9:10 PM David Ahern <dsahern@kernel.org> wrote:
>
> On 9/21/23 7:30 AM, Eric Dumazet wrote:
> > Some reads of inet->tos are racy.
> >
> > Add needed READ_ONCE() annotations and convert IP_TOS option lockless.
> >
> > Signed-off-by: Eric Dumazet <edumazet@google.com>
> > ---
> >  include/net/ip.h                              |  1 -
> >  net/dccp/ipv4.c                               |  2 +-
> >  net/ipv4/inet_diag.c                          |  2 +-
> >  net/ipv4/ip_output.c                          |  4 +--
> >  net/ipv4/ip_sockglue.c                        | 29 ++++++++-----------
> >  net/ipv4/tcp_ipv4.c                           |  9 +++---
> >  net/mptcp/sockopt.c                           |  8 ++---
> >  net/sctp/protocol.c                           |  4 +--
> >  .../selftests/net/mptcp/mptcp_connect.sh      |  2 +-
> >  9 files changed, 28 insertions(+), 33 deletions(-)
> >
>
> include/net/route.h dereferences sk tos as well.

Right, thanks for catching this.

> net/ipv4/icmp.c has a setting of it

This is safe, the socket is private to the current thread ( sk =
icmp_xmit_lock()) and not visible to other threads.

Re: [PATCH net-next 6/8] inet: implement lockless getsockopt(IP_UNICAST_IF)

[David Ahern]

On 9/21/23 7:30 AM, Eric Dumazet wrote:
> Add missing READ_ONCE() annotations when reading inet->uc_index
> 
> Implementing getsockopt(IP_UNICAST_IF) locklessly seems possible,
> the setsockopt() part might not be possible at the moment.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> ---
>  net/ipv4/datagram.c    |  2 +-
>  net/ipv4/ip_sockglue.c | 10 +++++-----
>  net/ipv4/ping.c        |  2 +-
>  net/ipv4/raw.c         | 13 +++++++------
>  net/ipv4/udp.c         | 12 +++++++-----
>  5 files changed, 21 insertions(+), 18 deletions(-)
> 

Reviewed-by: David Ahern <dsahern@kernel.org>

Re: [PATCH net-next 7/8] inet: lockless IP_PKTOPTIONS implementation

[David Ahern]

On 9/21/23 7:30 AM, Eric Dumazet wrote:
> Current implementation is already lockless, because the socket
> lock is released before reading socket fields.
> 
> Add missing READ_ONCE() annotations.
> 
> Note that corresponding WRITE_ONCE() are needed, the order
> of the patches do not really matter.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> ---
>  net/ipv4/ip_sockglue.c | 76 ++++++++++++++++++++----------------------
>  1 file changed, 37 insertions(+), 39 deletions(-)
> 

Reviewed-by: David Ahern <dsahern@kernel.org>

Re: [PATCH net-next 8/8] inet: implement lockless getsockopt(IP_MULTICAST_IF)

[David Ahern]

On 9/21/23 7:30 AM, Eric Dumazet wrote:
> Add missing annotations to inet->mc_index and inet->mc_addr
> to fix data-races.
> 
> getsockopt(IP_MULTICAST_IF) can be lockless.
> 
> setsockopt() side is left for later.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> ---
>  net/ipv4/datagram.c    |  4 ++--
>  net/ipv4/ip_sockglue.c | 25 ++++++++++++-------------
>  net/ipv4/ping.c        |  4 ++--
>  net/ipv4/raw.c         |  4 ++--
>  net/ipv4/udp.c         |  4 ++--
>  5 files changed, 20 insertions(+), 21 deletions(-)
> 


Reviewed-by: David Ahern <dsahern@kernel.org>