tcp: symmetric challenge ACK for SEG.ACK > SND.NXT

Commit 354e4aa391ed ("tcp: RFC 5961 5.2 Blind Data Injection Attack Mitigation") quotes RFC 5961 Section 5.2 in full, which requires that any incoming segment whose ACK value falls outside [SND.UNA - MAX.SND.WND, SND.NXT] MUST be discarded and an ACK sent back. Linux currently sends that challenge ACK only on the lower edge (SEG.ACK < SND.UNA - MAX.SND.WND); on the symmetric upper edge (SEG.ACK > SND.NXT) the segment is silently dropped with SKB_DROP_REASON_TCP_ACK_UNSENT_DATA.

Patch 1 completes the mitigation by emitting a rate-limited challenge ACK on that branch, reusing tcp_send_challenge_ack() and honouring FLAG_NO_CHALLENGE_ACK for consistency with the lower-edge case. It also updates the existing tcp_ts_recent_invalid_ack.pkt selftest, which drives this exact path, to consume the new challenge ACK so bisect stays clean.

Patch 2 adds a new packetdrill selftest that exercises RFC 5961 Section 5.2 on both edges of the acceptable window, filling a gap in the selftests tree (neither edge had dedicated coverage before).

---

Changelog
=========
v2 -> v3:
- I don't think the AI's concern holds, but I think I can split write(2000) into write(1000) so the test doesn't rely on kernel internals.
sashiko: https://sashiko.dev/#/patchset/20260421014128.289362-1-jiayuan.chen%40linux.dev
v1 -> v2:
- Add Reviewed-by tag.
- Fold the tcp_ts_recent_invalid_ack.pkt update into patch 1 so
that bisect stays clean and the fix is self-contained for
backport.
- Extend the new selftest to cover both edges of RFC 5961
Section 5.2 (SEG.ACK > SND.NXT and SEG.ACK < SND.UNA -
MAX.SND.WND) in a single connection, and rename it to tcp_rfc5961_ack-out-of-window.pkt. Neither edge had explicit
packetdrill coverage before.

v1: https://lore.kernel.org/netdev/[email protected]/

Jiayuan Chen (2):
tcp: send a challenge ACK on SEG.ACK > SND.NXT selftests/net: packetdrill: cover RFC 5961 5.2 challenge ACK on both
edges

 net/ipv4/tcp_input.c                          | 10 ++--
 .../tcp_rfc5961_ack-out-of-window.pkt         | 48 +++++++++++++++++++
 .../packetdrill/tcp_ts_recent_invalid_ack.pkt |  4 +-
 3 files changed, 58 insertions(+), 4 deletions(-)
 create mode 100644 tools/testing/selftests/net/packetdrill/tcp_rfc5961_ack-out-of-window.pkt

2.43.0

RFC 5961 Section 5.2 validates an incoming segment's ACK value against the range [SND.UNA - MAX.SND.WND, SND.NXT] and states:

"All incoming segments whose ACK value doesn't satisfy the above
condition MUST be discarded and an ACK sent back."

Commit 354e4aa391ed ("tcp: RFC 5961 5.2 Blind Data Injection Attack Mitigation") opted Linux into this mitigation and implements the challenge ACK on the lower side (SEG.ACK < SND.UNA - MAX.SND.WND), but the symmetric upper side (SEG.ACK > SND.NXT) still takes the pre-RFC-5961 path and silently returns SKB_DROP_REASON_TCP_ACK_UNSENT_DATA, even though RFC 793 Section 3.9 (now RFC 9293 Section 3.10.7.4) has always required:

"If the ACK acknowledges something not yet sent (SEG.ACK > SND.NXT)
then send an ACK, drop the segment, and return."

Complete the mitigation by sending a challenge ACK on that branch, reusing the existing tcp_send_challenge_ack() path which already enforces the per-socket RFC 5961 Section 7 rate limit via __tcp_oow_rate_limited(). FLAG_NO_CHALLENGE_ACK is honoured for symmetry with the lower-edge case.

Update the existing tcp_ts_recent_invalid_ack.pkt selftest, which drives this exact path, to consume the new challenge ACK.

---
 net/ipv4/tcp_input.c                                   | 10 +++++++---
 .../net/packetdrill/tcp_ts_recent_invalid_ack.pkt      |  4 +++-
 2 files changed, 10 insertions(+), 4 deletions(-)

index 021f745747c5..c2b6f05acdfa 100644
@@ -4284,11 +4284,15 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag)
 		goto old_ack;
 	}
 
-	/* If the ack includes data we haven't sent yet, discard
-	 * this segment (RFC793 Section 3.9).
+	/* If the ack includes data we haven't sent yet, drop the
+	 * segment.  RFC 793 Section 3.9 and RFC 5961 Section 5.2
+	 * require us to send an ACK back in that case.
 	 */
-	if (after(ack, tp->snd_nxt))
+	if (after(ack, tp->snd_nxt)) {
+		if (!(flag & FLAG_NO_CHALLENGE_ACK))
+			tcp_send_challenge_ack(sk, false);
 		return -SKB_DROP_REASON_TCP_ACK_UNSENT_DATA;
+	}
 
 	if (after(ack, prior_snd_una)) {
 		flag |= FLAG_SND_UNA_ADVANCED;

index 174ce9a1bfc0..ee6baf7c36cf 100644
@@ -19,7 +19,9 @@
 
 // bad packet with high tsval (its ACK sequence is above our sndnxt)
    +0 < F. 1:1(0) ack 9999 win 20000 <nop,nop,TS val 200000 ecr 100>
-
+// Challenge ACK for SEG.ACK > SND.NXT (RFC 5961 5.2 / RFC 793 3.9).
+// ecr=200 (not 200000) proves ts_recent was not updated from the bad packet.
+   +0 > . 1:1(0) ack 1 <nop,nop,TS val 200 ecr 200>
 
    +0 < . 1:1001(1000) ack 1 win 20000 <nop,nop,TS val 201 ecr 100>
    +0 > . 1:1(0) ack 1001 <nop,nop,TS val 200 ecr 201>

2.43.0

RFC 5961 Section 5.2 / RFC 793 Section 3.9 require a challenge ACK whenever an incoming SEG.ACK falls outside
[SND.UNA - MAX.SND.WND, SND.NXT]. There is currently no packetdrill coverage for either edge.

Add tcp_rfc5961_ack-out-of-window.pkt, which in a single passive-open connection exercises:

- Upper edge (SEG.ACK > SND.NXT): peer ACKs data that was never sent before the server has transmitted anything.
- Lower edge (SEG.ACK < SND.UNA - MAX.SND.WND): after the server has sent 2000 bytes (the peer-advertised rwnd forces two 1000-byte segments, both acknowledged), peer sends an ACK that is older than the acceptable window.

Both cases must elicit a challenge ACK
<SEQ = SND.NXT, ACK = RCV.NXT, CTL = ACK>. The per-socket RFC 5961 Section 7 rate limit is disabled for the duration of the test so that both challenge ACKs can fire back-to-back.

---
 .../tcp_rfc5961_ack-out-of-window.pkt         | 48 +++++++++++++++++++
 1 file changed, 48 insertions(+)
 create mode 100644 tools/testing/selftests/net/packetdrill/tcp_rfc5961_ack-out-of-window.pkt

new file mode 100644
index 000000000000..2776b8728085
@@ -0,0 +1,48 @@
+// SPDX-License-Identifier: GPL-2.0
+//
+// RFC 5961 Section 5.2 / RFC 793 Section 3.9: an incoming segment's
+// ACK value must lie in [SND.UNA - MAX.SND.WND, SND.NXT]; otherwise
+// the receiver MUST discard the segment and send a challenge ACK
+// back.  Exercise both edges of that window in a single connection.
+
+`./defaults.sh
+sysctl -q net.ipv4.tcp_invalid_ratelimit=0
+`
+
+   0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
+  +0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
+  +0 bind(3, ..., ...) = 0
+  +0 listen(3, 1) = 0
+
+// Three-way handshake.  Peer advertises rwnd = 1000 (no wscale), so
+// MAX.SND.WND is tracked as 1000.
+  +0 < S 0:0(0) win 1000 <mss 1000,sackOK,nop,nop,nop,wscale 0>
+  +0 > S. 0:0(0) ack 1 <...>
++.1 < . 1:1(0) ack 1 win 1000
+  +0 accept(3, ..., ...) = 4
+
+// ---- Upper edge: SEG.ACK > SND.NXT --------------------------------
+// Server has sent nothing yet, so SND.UNA = SND.NXT = 1.
+// Peer sends a pure ACK with SEG.ACK = 2, beyond SND.NXT.
+  +0 < . 1:1(0) ack 2 win 1000
+// Expect a challenge ACK: <SEQ = SND.NXT = 1, ACK = RCV.NXT = 1>.
+  +0 > . 1:1(0) ack 1
+
+// Advance SND.UNA past MAX.SND.WND so that the lower edge becomes
+// reachable.  Issue two 1-MSS writes so each skb is exactly one MSS
+// and PSH is set by tcp_push() at the end of each sendmsg, keeping
+// the setup independent of the TSO / tcp_fragment split path.
+  +0 write(4, ..., 1000) = 1000
+  +0 > P. 1:1001(1000) ack 1
++.01 < . 1:1(0) ack 1001 win 1000
+  +0 write(4, ..., 1000) = 1000
+  +0 > P. 1001:2001(1000) ack 1
++.01 < . 1:1(0) ack 2001 win 1000
+// Now SND.UNA = SND.NXT = 2001, MAX.SND.WND = 1000, bytes_acked = 2000.
+
+// ---- Lower edge: SEG.ACK < SND.UNA - MAX.SND.WND ------------------
+// SND.UNA - MAX.SND.WND = 2001 - 1000 = 1001, so SEG.ACK = 1000 falls
+// below the acceptable range.
+  +0 < . 1:1(0) ack 1000 win 1000
+// Expect a challenge ACK: <SEQ = SND.NXT = 2001, ACK = RCV.NXT = 1>.
+  +0 > . 2001:2001(0) ack 1

2.43.0