Re: [PATCH v3] tcp: fix tp->undo_retrans accounting in tcp_sacktag_one()

On Mon, Sep 13, 2021 at 8:49 AM Eric Dumazet [off-list ref] wrote:

On Mon, Sep 13, 2021 at 3:51 AM zhenggy [off-list ref] wrote:

quoted

Commit 10d3be569243 ("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 (10d3be569243), the assumption underlying the
tp->undo_retrans-- seems correct.

Fixes: 10d3be569243 ("tcp-tso: do not split TSO packets at retransmit time")

nit: We normally do not add an empty line between Fixes: tag and others.

Reviewed-by: Eric Dumazet <redacted>

Acked-by: Yuchung Cheng <redacted>

This is a nice fix that would increase the effectiveness of TCP undo
in high-speed networks.

quoted

Signed-off-by: zhenggy <redacted>
---
 net/ipv4/tcp_input.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 3f7bd7a..141e85e 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c

@@ -1346,7 +1346,7 @@ 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;
--

1.8.3.1