Trouble: running server from Windows inside a SSH session doesn't work

[lars18th]

Hi @richlegrand ,

I've found a problem that I feel you can't solve. However, it will be interesting to document it:

• When executing bitbang.exe serve shell inside one SSH connection to the Windows machine, all seems to work. However, the client when starting the shell only shows but is not getting the inputs.

To check it do that test:

• On your Windows 10/11 machine install/enable the stock OpenSSH server from Microsoft.
• Execute in a local window terminal the regular bitbang.exe serve shell and try to connect from another machine (all works).
• Open your preferable SSH client and connect to your Windows machine, and inside this session execute the same bitbang.exe serve shell.
• When you connect from another machine using bitbang... You see the regular welcome Microsoft Windows message, but can't execute anything.

This is a bug. But perhaps you can't solve it.
Only FYI.

[richlegrand]

Thanks -- and I don't think this one is unsolvable...

First, what I could rule out. I ran your exact topology on Linux -- ssh in,
start bitbang serve shell inside that SSH session, connect from outside --
with both a CLI client and a browser. Input reaches the shell either way, so
nothing about being inside an SSH session breaks this generally. Whatever it
is, it is Windows.

The obvious suspect there is the Windows console support, since it opens
CONIN$/CONOUT$ and an SSH session is exactly what changes that. But the
shell's input does not go through it: the pseudoconsole is built from anonymous
pipes, so the shell's stdin is a pipe we write to, unrelated to the console the
listener itself is attached to. On paper they are independent.

The timing is still suspicious, though. Windows console support is new in the
release I'm finishing, and yours is a build of that branch. So the first thing
worth knowing is whether this is something I just broke:

1. Does 0.4.7 behave the same way? Same SSH session, the released binary
rather than a branch build. If 0.4.7 is fine, then it is a regression and I
want it fixed before this ships rather than after.

2. What does bitbang connect <url> -- dir do from the other machine?
That runs a one-shot command, which takes a plain-pipe path instead of the
pseudoconsole. If the output comes back, the problem is the pseudoconsole
specifically, which narrows it a lot.

Two more if you have the patience, though the first two tell me most of it:

3. bitbang.exe serve shell < NUL inside the SSH session. Redirecting stdin
   leaves the listener with no console at all. If the remote shell starts
   accepting input, the console is involved after all.
4. bitbang.exe serve shell -shell-cmd powershell.exe, in case this is
   cmd.exe rather than the pty.

One question about the symptom, since it changes where I look: when you type
dir and press Enter, do you see the characters echo back and nothing happens,
or do the characters never appear at all? Echo with no execution and no echo at
all are different failures.

Running a listener on a headless Windows box over SSH is a reasonable thing to
want, so this is worth getting right rather than documenting.

[lars18th]

Hi @richlegrand ,

I'll do all the test, but the next week. Thank you for your patience.

[richlegrand]

No rush, but checking in -- did you get a chance to run those?

The one that decides where I look is the first: whether 0.4.7 does the
same thing in the same SSH session. Windows console support is new in
0.5.0, so if 0.4.7 is fine then I broke this and it's mine to fix rather
than a gap to document. The other three narrow it, but that one splits it
in half.