mptcp: Use correct MSS for space-computation

We have seen the following WARN:
[  413.544376] ------------[ cut here ]------------
[  413.566704] in_flight 0 cwnd 1 wseq 509214375 snxt 509212959 mss_now 1416 cache 1424
[...]
[  414.485683] Call Trace:
[  414.497589]  mptcp_write_xmit+0xc3/0x490
[  414.516761]  __tcp_push_pending_frames+0x38/0xd0
[  414.539402]  tcp_sendmsg_locked+0x8f8/0xd00
[  414.559813]  tcp_sendmsg+0x27/0x40
[  414.576339]  sock_sendmsg+0x36/0x50
[  414.593346]  sock_write_iter+0x8f/0x100
[  414.612081]  __vfs_write+0x112/0x1a0
[  414.629447]  vfs_write+0xad/0x1a0
[  414.645486]  ksys_write+0x5a/0xd0
[  414.661601]  do_syscall_64+0x5b/0x1b0
[  414.679146]  entry_SYSCALL_64_after_hwframe+0x44/0xa9

mss_now can effectively be different from mss_cache, if for example we
need to combine SACK with data. In that case, mptcp_is_temp_unavailable will
make the wrong decision.

We should get the correct mss from tcp_current_mss right away.

Fixes: 988ec13 ("mptcp: Make sure that we don't overfill subflows")
Signed-off-by: Christoph Paasch <[email protected]>
Signed-off-by: Matthieu Baerts <[email protected]>
(cherry picked from commit 9d3f35b)
Signed-off-by: Matthieu Baerts <[email protected]>

net/mptcp/mptcp_sched.c

@@ -71,19 +71,19 @@ static bool mptcp_is_temp_unavailable(struct sock *sk,
 	if (in_flight >= tp->snd_cwnd)
 		return true;
 
+	mss_now = tcp_current_mss(sk);
+
 	/* Now, check if what is queued in the subflow's send-queue
 	 * already fills the cwnd.
 	 */
-	space = (tp->snd_cwnd - in_flight) * tp->mss_cache;
+	space = (tp->snd_cwnd - in_flight) * mss_now;
 
 	if (tp->write_seq - tp->snd_nxt >= space)
 		return true;
 
 	if (zero_wnd_test && !before(tp->write_seq, tcp_wnd_end(tp)))
 		return true;
 
-	mss_now = tcp_current_mss(sk);
-
 	/* Don't send on this subflow if we bypass the allowed send-window at
 	 * the per-subflow level. Similar to tcp_snd_wnd_test, but manually
 	 * calculated end_seq (because here at this point end_seq is still at