tcp: annotate sk->sk_sndbuf lockless reads

For the sake of tcp_poll(), there are few places where we fetch
sk->sk_sndbuf while this field can change from IRQ or other cpu.

We need to add READ_ONCE() annotations, and also make sure write
sides use corresponding WRITE_ONCE() to avoid store-tearing.

Note that other transports probably need similar fixes.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

include/net/sock.h

@@ -883,7 +883,7 @@ static inline int sk_stream_min_wspace(const struct sock *sk)
 
 static inline int sk_stream_wspace(const struct sock *sk)
 {
-	return sk->sk_sndbuf - sk->sk_wmem_queued;
+	return READ_ONCE(sk->sk_sndbuf) - sk->sk_wmem_queued;
 }
 
 void sk_stream_write_space(struct sock *sk);
@@ -1207,7 +1207,7 @@ static inline void sk_refcnt_debug_release(const struct sock *sk)
 
 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
 {
-	if (sk->sk_wmem_queued >= sk->sk_sndbuf)
+	if (sk->sk_wmem_queued >= READ_ONCE(sk->sk_sndbuf))
 		return false;
 
 	return sk->sk_prot->stream_memory_free ?
@@ -2220,10 +2220,14 @@ static inline void sk_wake_async(const struct sock *sk, int how, int band)
 
 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
 {
-	if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
-		sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
-		sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
-	}
+	u32 val;
+
+	if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
+		return;
+
+	val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
+
+	WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
 }
 
 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
@@ -2251,7 +2255,7 @@ bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
  */
 static inline bool sock_writeable(const struct sock *sk)
 {
-	return refcount_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
+	return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
 }
 
 static inline gfp_t gfp_any(void)

net/core/filter.c

@@ -4258,7 +4258,8 @@ BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
 		case SO_SNDBUF:
 			val = min_t(u32, val, sysctl_wmem_max);
 			sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
-			sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
+			WRITE_ONCE(sk->sk_sndbuf,
+				   max_t(int, val * 2, SOCK_MIN_SNDBUF));
 			break;
 		case SO_MAX_PACING_RATE: /* 32bit version */
 			if (val != ~0U)

net/core/sock.c

@@ -785,7 +785,8 @@ int sock_setsockopt(struct socket *sock, int level, int optname,
 		 */
 		val = min_t(int, val, INT_MAX / 2);
 		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
-		sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
+		WRITE_ONCE(sk->sk_sndbuf,
+			   max_t(int, val * 2, SOCK_MIN_SNDBUF));
 		/* Wake up sending tasks if we upped the value. */
 		sk->sk_write_space(sk);
 		break;
@@ -2089,8 +2090,10 @@ EXPORT_SYMBOL(sock_i_ino);
 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
 			     gfp_t priority)
 {
-	if (force || refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
+	if (force ||
+	    refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
 		struct sk_buff *skb = alloc_skb(size, priority);
+
 		if (skb) {
 			skb_set_owner_w(skb, sk);
 			return skb;
@@ -2191,7 +2194,7 @@ static long sock_wait_for_wmem(struct sock *sk, long timeo)
 			break;
 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
-		if (refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
+		if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
 			break;
 		if (sk->sk_shutdown & SEND_SHUTDOWN)
 			break;
@@ -2226,7 +2229,7 @@ struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
 		if (sk->sk_shutdown & SEND_SHUTDOWN)
 			goto failure;
 
-		if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
+		if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
 			break;
 
 		sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
@@ -2807,7 +2810,7 @@ static void sock_def_write_space(struct sock *sk)
 	/* Do not wake up a writer until he can make "significant"
 	 * progress.  --DaveM
 	 */
-	if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
+	if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) {
 		wq = rcu_dereference(sk->sk_wq);
 		if (skwq_has_sleeper(wq))
 			wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
@@ -3207,7 +3210,7 @@ void sk_get_meminfo(const struct sock *sk, u32 *mem)
 	mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
 	mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
 	mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
-	mem[SK_MEMINFO_SNDBUF] = sk->sk_sndbuf;
+	mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
 	mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
 	mem[SK_MEMINFO_WMEM_QUEUED] = sk->sk_wmem_queued;
 	mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);

net/ipv4/tcp.c

@@ -450,7 +450,7 @@ void tcp_init_sock(struct sock *sk)
 
 	icsk->icsk_sync_mss = tcp_sync_mss;
 
-	sk->sk_sndbuf = sock_net(sk)->ipv4.sysctl_tcp_wmem[1];
+	WRITE_ONCE(sk->sk_sndbuf, sock_net(sk)->ipv4.sysctl_tcp_wmem[1]);
 	WRITE_ONCE(sk->sk_rcvbuf, sock_net(sk)->ipv4.sysctl_tcp_rmem[1]);
 
 	sk_sockets_allocated_inc(sk);

net/ipv4/tcp_input.c

@@ -359,7 +359,8 @@ static void tcp_sndbuf_expand(struct sock *sk)
 	sndmem *= nr_segs * per_mss;
 
 	if (sk->sk_sndbuf < sndmem)
-		sk->sk_sndbuf = min(sndmem, sock_net(sk)->ipv4.sysctl_tcp_wmem[2]);
+		WRITE_ONCE(sk->sk_sndbuf,
+			   min(sndmem, sock_net(sk)->ipv4.sysctl_tcp_wmem[2]));
 }
 
 /* 2. Tuning advertised window (window_clamp, rcv_ssthresh)