tcp: do not recycle cloned skbs

It is illegal to change arbitrary fields in skb_shared_info if the
skb is cloned.

Before calling skb_zcopy_clear() we need to ensure this rule,
therefore we need to move the test from sk_stream_alloc_skb()
to sk_wmem_free_skb()

Fixes: 4f66154 ("tcp: fix zerocopy and notsent_lowat issues")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Diagnosed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

include/net/sock.h

@@ -1473,7 +1473,7 @@ static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
 	sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
 	sk->sk_wmem_queued -= skb->truesize;
 	sk_mem_uncharge(sk, skb->truesize);
-	if (!sk->sk_tx_skb_cache) {
+	if (!sk->sk_tx_skb_cache && !skb_cloned(skb)) {
 		skb_zcopy_clear(skb, true);
 		sk->sk_tx_skb_cache = skb;
 		return;

net/ipv4/tcp.c

@@ -855,7 +855,7 @@ struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
 
 	if (likely(!size)) {
 		skb = sk->sk_tx_skb_cache;
-		if (skb && !skb_cloned(skb)) {
+		if (skb) {
 			skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
 			sk->sk_tx_skb_cache = NULL;
 			pskb_trim(skb, 0);