Re: tcp: peer stuck in low_rtt_dst table loses window clamping, every new connection stalls on RTOs

[Lynxes Exe <lynxes.exe@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 10438 bytes --]

Hi Stefano,

I apologize for the long wait. The issue took a while to represent itself.
As usual, this was done with the assistance of Claude.
After the issue occurred again the table got extracted to show that in fact
two IPs that were not supposed to be there got in the table (more details
below).

Also, thanks to both of you for the responses.

Original response
---------------------------

Thanks for going through it.

The stalls came back this week, about two weeks after the last pasta
restart, so we could check what you asked on a live instance.

1. The insertion, now demonstrated

The pasta in Docker's rootless bundle is static, non-PIE and not
stripped, so low_rtt_dst sits at a fixed address. We read it as root from
/proc/<pid>/mem while pasta kept running:

  slot 0  10.0.0.181     the host's own LAN address
  slot 1  192.168.5.3    a WireGuard peer, over the Internet
  slot 2  10.0.0.80      the host's second address
  slot 3  192.168.122.1  libvirt bridge, host side
  slot 4  10.0.0.187     a Windows desktop on Wi-Fi, RTT 2-6 ms
  slot 5-7               empty

The desktop was the peer that stalled: through pasta it got 14-33 KB/s,
with retransmissions and backoff inside the namespace, while SSH from
the same machine to the host ran at about 9 MB/s. 10.0.0.1 from the
first report wasn't there, because pasta had been restarted since.

2. Why: the ~0U case, reproduced

For a passively opened socket, the handshake gives no RTT sample if the
SYN-ACK was retransmitted (tcp_synack_rtt_meas() only measures when
req->num_retrans is 0, and a retransmitted SYN from the client also
increments it, via tcp_rtx_synack()). The only fallback is the
timestamp echo in tcp_ack_update_rtt(), so if the client didn't
negotiate timestamps, tcpi_min_rtt is still ~0U when pasta checks it.
Windows doesn't offer timestamps on the connections it opens, by
default, which fits the desktop.

Reproducer: from a Linux client with net.ipv4.tcp_timestamps = 0, drop
incoming SYN-ACKs from the pasta host for 2 seconds, with an nftables
input rule on the client's fixed local port, while connecting to a port
that pasta forwards. The client's SYN retries make the server
retransmit the SYN-ACK, and the connection completes after about 2 s.

Results, on Linux 6.18.57, with pasta built from 8ba4132 plus your
check and some logging:

- A plain listener on the same host, reading TCP_INFO right after
  accept(): tcpi_min_rtt 0xffffffff, tcpi_rtt 0, tcpi_total_retrans 3.
  The same client without the drop: tcpi_min_rtt 5423 us.
- pasta: your check fired, the client stayed out of the table, and
  transfers from it stayed at full speed. Without the check, it would
  have been added.
- With timestamps enabled on the client, the same test gets a measured
  RTT from the timestamp echo, and nothing is skipped.

We never saw 0, so skipping only ~0U looks right to me.

3. Patches

Attached, against current master (7380393):

  0001 tcp: Don't add peers to low RTT table before RTT is measured
  0002 tcp: Log additions to low RTT table

0001 is your condition as you wrote it. 0002 logs only additions, with
flow_dbg(): the skip itself happens on every tcp_prepare_flags() call
until the first sample, which was 28 times for one connection in the
test above, so logging it would be noisy. Each patch builds without
warnings; I haven't run the test suite.

I didn't do the expiry part. I'm happy to try it if you have a
preference between tcp_rtt_dst_check() and tcp_update_seqack_wnd().

We're keeping the patched build running and will report if a remote
peer still gets into the table.

Thanks,
Lynxes


On Wed, Nov 4, 2026 at 2:53 PM Stefano Brivio <sbrivio@redhat.com> wrote:
>
> Hi Lynxes,
>
> Thanks for the report, and sorry for the delay.  Some comments inline:
>
> On Wed, 28 Oct 2026 14:15:59 +0100
> Lynxes Exe <lynxes.exe@gmail.com> wrote:
>
> > Hi,
> >
> > A disclaimer before anything is owed: what you're about to read was
> > written by Claude, and is the result of a bug fixing session that occurred
> > on my personal home infrastructure. It was not intended as a bug hunt.
> > Since it seems like this issue has been unreported and I couldn't find
> > anything similar to it, and since what Claude reported seems potentially
> > plausible, I'm putting it here.
> > If AI assisted/generated bug reports are unwanted, I apologize.
>
> At least as far as I'm concerned, in term of bugs reports, whatever
> makes sense and is usable is welcome.  This report has a few dubious
> aspects (see below) but it might still be valuable, I think.
>
> > Original report
> > --------------------------
> > I think a peer address can get stuck in pasta's low-RTT destination
> > table and then lose window clamping for good.
>
> This part wasn't clearly demonstrated.
>
> > That caused every new
> > TCP connection from our LAN router's address to stall on RTOs until
> > pasta was restarted. Details below.
>
> This looks plausible and somewhat demonstrated.
>
> > I haven't found an existing report on Bugzilla or this list.
>
> Correct, no known issues in this sense so far.
>
> > Version: pasta 2026_10_02.cba3570-3-g8ba4132 (the logic below is
> >          unchanged in current master)
> > Setup:   rootless Docker 29.6.1 via rootlesskit, --net=pasta,
> >          --port-driver=implicit, --mtu=65520 (pasta --tcp-ports=auto
> >          --udp-ports=auto --no-map-gw --ipv4-only ...)
> > Kernel:  6.18.57 (Arch linux-lts), x86_64
> >
> > What we think is happening
> > --------------------------
> > We suspect the low-RTT destination table (low_rtt_dst[],
> > tcp_rtt_dst_check() / tcp_rtt_dst_low()). Our LAN router's address,
> > 10.0.0.1, ended up in it even though the router is several
> > milliseconds away.
>
> Nothing clearly confirms the "ended up in it" part (yet).
>
> > [...]
> >
> > We could not read the table directly (no ptrace into pasta), so
> > "10.0.0.1 is in the table" is inferred from this behaviour and the
> > code path.
>
> ...I guess the problem here is training a language model about the
> difference between "inferred" and "suspected". :)
>
> One could ptrace() as root by the way, but that wouldn't help either.
>
> Could you make sure about this with a print in tcp_rtt_dst_check()?  If
> that's not the case, you might need to further trace what's going on in
> tcp_update_seqack_wnd() instead.
>
> Rebuilding and using a modified version of passt is quick (it just
> depends on a C library):
>
>   git clone git://passt.top/passt && cd passt
>   [edit as needed]
>   make && sudo make install # Or make install as root
>
> Both Podman and Docker / rootlesskit will pick up the local build under
> /usr/local/bin. You can cleanly remove it with 'make uninstall' as root
> once you're done.
>
> > Why we think 10.0.0.1 got into the table
> > ----------------------------------------
> > A real RTT of 10 us or less to the router is impossible here
> > (measured RTT 2-6 ms). Our suspicion comes from reading the kernel
> > source; we have not caught it on a live socket:
> >
> > - tcp_rtt_dst_check() tests
> >   (int)tinfo->tcpi_min_rtt > LOW_RTT_THRESHOLD.
> > - The kernel initialises rtt_min to ~0U, both in tcp_init_sock()
> >   (net/ipv4/tcp.c) and for accepted sockets in
> >   tcp_create_openreq_child() (tcp_minisocks.c). tcpi_min_rtt stays
> >   ~0U until the first valid RTT sample.
> > - tcp_synack_rtt_meas() takes no sample when the SYN-ACK was
> >   retransmitted (rtt_us = -1 unless !req->num_retrans). The same
> >   applies on the active side when the SYN was retransmitted (Karn).
> > - Cast to int, ~0U becomes -1, which passes the "<= 10 us" test.
> > - pasta runs this check in tcp_prepare_flags() when it sends the SYN
> >   (inbound) or SYN-ACK (outbound) to the guest. That is right after
> >   the host-side handshake, before any data ACK could provide a
> >   sample.
>
> ...maybe, yes.
>
> > We measured exactly such handshake retransmissions on this
> > Wi-Fi/NAT path: 6 of 1,800 fresh connections took ~1.05 s, i.e. one
> > 1 s retry. With hundreds of inbound and outbound connections to
> > 10.0.0.1 per day, one retried handshake would be enough to insert it
> > permanently.
> >
> > The night the stalls became severe, the published container was also
> > restarted about ten times while clients kept connecting, which made
> > unusual handshakes more likely. We cannot tell exactly when the
> > insertion happened.
> >
> >
> > Suggested fixes
> > ---------------
> > 1. In tcp_rtt_dst_check(), treat tcpi_min_rtt == ~0U (and 0) as
> >    "no sample" and skip the insertion. Compare as unsigned, or check
> >    tcpi_rtt / tcpi_rcv_rtt as well.
>
> I'm not sure if we actually risk hitting 0: strictly speaking that
> would mean less than 1 µs which looks unlikely (even for simple veth
> pairs or local guests), but in practice there might be rounding errors
> somewhere.
>
> Do we need to skip 0? From a quick glance of the kernel paths involved
> I'd say we don't.
>
> Could you test this perhaps?  Not build tested but I think you could
> modify the initial early return condition to:
>
>         if (!min_rtt_cap ||
>             (int)tinfo->tcpi_min_rtt > LOW_RTT_THRESHOLD ||
>             tinfo->tcpi_min_rtt == ~0U /* not measured yet */)
>                 return;
>
> > 2. Don't trust a single sample forever: give entries a lifetime, or
> >    evict an address when a later connection to it shows a min RTT
> >    above the threshold.
>
> Reasonable.  I guess we could add a timestamp and check if the entry
> expired (minutes?) where we update the RTT in
> tcp_update_seqack_wnd()... or even in tcp_rtt_dst_check().  No strong
> preference on my side, I didn't really think it through.
>
> > 3. Log (at debug level) when an address enters the table. It was
> >    invisible here; the only visible signs were retransmissions
> >    inside the namespace.
>
> Yeah, that would be good.  I added this feature before we started
> logging anything, I think.  Feel free to send a patch for that.
>
> > Workaround: restarting pasta clears the table.
> >
> > Happy to test a patch or collect more data.
>
> Let me know if you can check those few things I suggested, thanks.
>
> --
> Stefano
>

[-- Attachment #2: 0001-tcp-Don-t-add-peers-to-low-RTT-table-before-RTT-is-m.patch --]
[-- Type: application/octet-stream, Size: 2081 bytes --]

From 172138c0b005fe92e8f0220b38912685b63a5882 Mon Sep 17 00:00:00 2001
From: Pietro De Domenico <lynxes.exe@gmail.com>
Date: Thu, 12 Nov 2026 16:56:44 +0100
Subject: [PATCH 1/2] tcp: Don't add peers to low RTT table before RTT is
 measured

The kernel initialises the minimum RTT of a socket to ~0U (see
tcp_init_sock() and tcp_create_openreq_child()), and tcpi_min_rtt keeps
that value until the first RTT sample.

For a passively opened socket, the handshake gives no sample if the
SYN-ACK was retransmitted (Karn's algorithm, see tcp_synack_rtt_meas()),
and the only fallback is the timestamp echo, which isn't there if the
peer didn't negotiate TCP timestamps. Windows clients don't, by default.

tcp_rtt_dst_check() casts tcpi_min_rtt to int, so ~0U becomes -1, the
peer passes the threshold check, and it's added to the low RTT table,
which it never leaves: from then on, every connection with that peer
advertises an unclamped window to the guest and stalls on
retransmissions.

Skip the check while the minimum RTT hasn't been measured yet.

Reproduced on Linux 6.18.57 by dropping SYN-ACKs at a client with
net.ipv4.tcp_timestamps set to 0, for two seconds: tcpi_min_rtt of the
accepted socket is 0xffffffff, with tcpi_total_retrans 3, and pasta
would have added the client to the table.

Link: https://archives.passt.top/passt-dev/CAJPNJUuUYSkW776P0hKmweHnKOqXtsXqwm4HXP-+jorXrSfFFg@mail.gmail.com/
Suggested-by: Stefano Brivio <sbrivio@redhat.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Pietro De Domenico <lynxes.exe@gmail.com>
---
 tcp.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/tcp.c b/tcp.c
index 95f4f3e..d88ebd4 100644
--- a/tcp.c
+++ b/tcp.c
@@ -758,7 +758,8 @@ static void tcp_rtt_dst_check(const struct tcp_tap_conn *conn,
 	int i, hole = -1;
 
 	if (!min_rtt_cap ||
-	    (int)tinfo->tcpi_min_rtt > LOW_RTT_THRESHOLD)
+	    (int)tinfo->tcpi_min_rtt > LOW_RTT_THRESHOLD ||
+	    tinfo->tcpi_min_rtt == ~0U /* not measured yet */)
 		return;
 
 	for (i = 0; i < LOW_RTT_TABLE_SIZE; i++) {
-- 
2.56.0


[-- Attachment #3: 0002-tcp-Log-additions-to-low-RTT-table.patch --]
[-- Type: application/octet-stream, Size: 1387 bytes --]

From 3f71b8ba1255e83c5702d4769230d5007c828a3d Mon Sep 17 00:00:00 2001
From: Pietro De Domenico <lynxes.exe@gmail.com>
Date: Thu, 12 Nov 2026 16:56:44 +0100
Subject: [PATCH 2/2] tcp: Log additions to low RTT table

Entries in the low RTT table never expire, and a wrong one makes every
connection with that peer stall, but nothing shows that a peer was
added: the only visible symptom is retransmissions inside the
namespace.

Print a debug message with the address and the measured minimum RTT
whenever a peer is added.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Pietro De Domenico <lynxes.exe@gmail.com>
---
 tcp.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/tcp.c b/tcp.c
index d88ebd4..4aef854 100644
--- a/tcp.c
+++ b/tcp.c
@@ -755,6 +755,7 @@ static void tcp_rtt_dst_check(const struct tcp_tap_conn *conn,
 			      const struct tcp_info_linux *tinfo)
 {
 	const struct flowside *tapside = TAPFLOW(conn);
+	char buf[INANY_ADDRSTRLEN];
 	int i, hole = -1;
 
 	if (!min_rtt_cap ||
@@ -775,6 +776,10 @@ static void tcp_rtt_dst_check(const struct tcp_tap_conn *conn,
 	if (hole == -1)
 		return;
 
+	flow_dbg(conn, "Adding %s to low RTT table, minimum RTT: %u us",
+		 inany_ntop(&tapside->oaddr, buf, sizeof(buf)),
+		 tinfo->tcpi_min_rtt);
+
 	low_rtt_dst[hole++] = tapside->oaddr;
 	if (hole == LOW_RTT_TABLE_SIZE)
 		hole = 0;
-- 
2.56.0