tcp: fix tp->undo_retrans accounting in tcp_sacktag_one()

commit 4f884f3 upstream.

Commit 10d3be5 ("tcp-tso: do not split TSO packets at retransmit
time") may directly retrans a multiple segments TSO/GSO packet without
split, Since this commit, we can no longer assume that a retransmitted
packet is a single segment.

This patch fixes the tp->undo_retrans accounting in tcp_sacktag_one()
that use the actual segments(pcount) of the retransmitted packet.

Before that commit (10d3be5), the assumption underlying the
tp->undo_retrans-- seems correct.

Fixes: 10d3be5 ("tcp-tso: do not split TSO packets at retransmit time")
Signed-off-by: zhenggy <zhenggy@chinatelecom.cn>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

net/ipv4/tcp_input.c

@@ -1314,7 +1314,7 @@ static u8 tcp_sacktag_one(struct sock *sk,
 	if (dup_sack && (sacked & TCPCB_RETRANS)) {
 		if (tp->undo_marker && tp->undo_retrans > 0 &&
 		    after(end_seq, tp->undo_marker))
-			tp->undo_retrans--;
+			tp->undo_retrans = max_t(int, 0, tp->undo_retrans - pcount);
 		if ((sacked & TCPCB_SACKED_ACKED) &&
 		    before(start_seq, state->reord))
 				state->reord = start_seq;