diff --git a/include/net/tcp.h b/include/net/tcp.h
index 3e4a0a06b00c..68a22a171c2b 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -1423,6 +1423,23 @@ static inline void tcp_init_send_head(struct sock *sk)
 	sk->sk_send_head = NULL;
 }
 
+static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
+{
+	struct sk_buff *skb = tcp_write_queue_head(sk);
+
+	if (skb == tcp_send_head(sk))
+		skb = NULL;
+
+	return skb;
+}
+
+static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
+{
+	struct sk_buff *skb = tcp_send_head(sk);
+
+	return skb ? tcp_write_queue_prev(sk, skb) : tcp_write_queue_tail(sk);
+}
+
 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
 {
 	__skb_queue_tail(&sk->sk_write_queue, skb);
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 15143d31a82a..88b8bf1a28d8 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -1063,6 +1063,7 @@ int tcp_fragment(struct sock *sk, struct sk_buff *skb, u32 len,
 	struct tcp_sock *tp = tcp_sk(sk);
 	struct sk_buff *buff;
 	int nsize, old_factor;
+	long limit;
 	int nlen;
 	u8 flags;
 
@@ -1073,7 +1074,15 @@ int tcp_fragment(struct sock *sk, struct sk_buff *skb, u32 len,
 	if (nsize < 0)
 		nsize = 0;
 
-	if (unlikely((sk->sk_wmem_queued >> 1) > sk->sk_sndbuf + 0x20000)) {
+	/* tcp_sendmsg() can overshoot sk_wmem_queued by one full size skb.
+	 * We need some allowance to not penalize applications setting small
+	 * SO_SNDBUF values.
+	 * Also allow first and last skb in retransmit queue to be split.
+	 */
+	limit = sk->sk_sndbuf + 2 * SKB_TRUESIZE(GSO_MAX_SIZE);
+	if (unlikely((sk->sk_wmem_queued >> 1) > limit &&
+		     skb != tcp_rtx_queue_head(sk) &&
+		     skb != tcp_rtx_queue_tail(sk))) {
 		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPWQUEUETOOBIG);
 		return -ENOMEM;
 	}