tcp: really ignore MSG_ZEROCOPY if no SO_ZEROCOPY

According to the documentation in msg_zerocopy.rst, the SO_ZEROCOPY
flag was introduced because send(2) ignores unknown message flags and
any legacy application which was accidentally passing the equivalent of
MSG_ZEROCOPY earlier should not see any new behaviour.

Before commit f214f91 ("tcp: enable MSG_ZEROCOPY"), a send(2) call
which passed the equivalent of MSG_ZEROCOPY without setting SO_ZEROCOPY
would succeed.  However, after that commit, it fails with -ENOBUFS.  So
it appears that the SO_ZEROCOPY flag fails to fulfill its intended
purpose.  Fix it.

Fixes: f214f91 ("tcp: enable MSG_ZEROCOPY")
Signed-off-by: Vincent Whitchurch <vincent.whitchurch@axis.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/core/skbuff.c

@@ -939,9 +939,6 @@ struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size)
 
 	WARN_ON_ONCE(!in_task());
 
-	if (!sock_flag(sk, SOCK_ZEROCOPY))
-		return NULL;
-
 	skb = sock_omalloc(sk, 0, GFP_KERNEL);
 	if (!skb)
 		return NULL;

net/ipv4/tcp.c

@@ -1185,7 +1185,7 @@ int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)
 
 	flags = msg->msg_flags;
 
-	if (flags & MSG_ZEROCOPY && size) {
+	if (flags & MSG_ZEROCOPY && size && sock_flag(sk, SOCK_ZEROCOPY)) {
 		if (sk->sk_state != TCP_ESTABLISHED) {
 			err = -EINVAL;
 			goto out_err;