Tunnel disconnects with "peer ignored channel window"

valt aran sshdude at outlook.com
Fri Oct 10 05:15:16 AEDT 2025


Config on server:

ip tuntap add dev tunX mode tun
ip addr add A.B.C.D/24 dev tunX
ip link set dev tunX up
(+sometimes changed mtu to 1500)

in the sshd config "PermitTunnel yes" is added and sshd is restarted.
Version is 9.6p1, iirc also tried 8.7p1.

Client:

ip tuntap add mode tun tunY
ip addr add E.F.G.H/24 dev tunY
ip link set dev tunY up

Version is 9.9p1

Tunnel is setup with: ssh root at VPNIP -p 22 -f -w Y:X  -N
OS where Ubuntu and Fedora

I'm going to test it with the latest OpenSSH + patch on both sides.

________________________________________
From: Damien Miller <djm at mindrot.org>
Sent: Tuesday, October 7, 2025 11:30 PM
To: valt aran
Cc: openssh-unix-dev at mindrot.org
Subject: Re: Tunnel disconnects with "peer ignored channel window"

On Tue, 7 Oct 2025, valt aran wrote:

> yeye I have read the html warning when signing up. I disabled it
> but seems outlook has two switch options. One for sending, one for
> viewing. Hopefully it is correct this time. Sorry for the (html)
> noise.
>
> Anyways:
>
> Hi,
>
> I have set up a ssh tunnel. It works fine, but under load it
> disconnects after a while. On the client side I get stuff like:
>
> "channel 0: rcvd too much data 1448, win 0/2097152 (excess 209960)"
>
> Usually there are lots of those message after another, but they don't
> immediately lead to a disconnect. After a while though the tunnel is
> disconnected with
>
> "disconnecting 1.2.3.4 port 4321: channel 0: peer ignored channel
> window"
>
> So this indicates the issue is on the server side?
>
> On the sshd server side I often see: "debug1: channel 0: datagram too
> big"

What exact commands are you using to start the tunnel?

What are the configurations of the tun/tap interfaces?

What OpenSSH version and OS are you using?

It's possible there's an error in the remote_window calculations.
Can you first try reproducing the problem with this patch:

diff --git a/channels.c b/channels.c
index 08636c11a..329815138 100644
--- a/channels.c
+++ b/channels.c
@@ -3055,7 +3055,9 @@ channel_output_poll_input_open(struct ssh *ssh, Channel *c)
                 * backpressure at read time? (i.e. read + discard)
                 */
                if (plen > c->remote_window || plen > c->remote_maxpacket) {
-                       debug("channel %d: datagram too big", c->self);
+                       debug("channel %d: datagram length %zu too big for "
+                           "window %u or maximum packet %u", c->self, plen,
+                            c->remote_window, c->remote_maxpacket);
                        return 0;
                }
                /* Enqueue it */

Which should give some better diagnostics on what's going wrong.

I'm suspicious of this section of code in channel_input_data():

  3498          if (c->datagram)
  3499                  win_len += 4;  /* string length header */

might be double-counting the header and permanently shrinking the
channel window for each tunneled packet received. You could try
removing or commenting it out and seeing if it solves the problem.

-d


More information about the openssh-unix-dev mailing list