openssh unix dev - Oct 2010 - killing remote commands

Dear ssh gurus,


Here's the version I'm testing on :
    flavien :/$ ssh -V
    OpenSSH_5.3p1 Debian-3ubuntu4, OpenSSL 0.9.8k 25 Mar 2009

I launch a remote command :
    flavien$ ssh -o ControlMaster=yes -o ControlPath=/tmp/ssh-control localhost
'echo pid:$$ ...sleeping...; sleep 2803'
    flavien at localhost's password: 
    pid:11565 ...sleeping...

On another shell, I kill the ssh client :
    flavien$ ssh -o ControlPath=/tmp/ssh-control -O exit localhost
    Exit request sent.
    flavien$

The ssh client is killed in the original terminal. Fine. However, the
shell at the other end is not killed :
    flavien$ ps -ef | grep 2803
    flavien  11565     1  0 10:37 ?        00:00:00 bash -c echo pid:$$
...sleeping...; sleep 2803
    flavien  11566 11565  0 10:37 ?        00:00:00 sleep 2803

If the remote process is an interactive shell, however, it is killed
once the ssh client terminates. I suspect that's because the shell
gets an EOF on its stdin. Is this true ?

For reading this ml for a few years, I have in mind some sort of
"signal forwarding" feature that would send a SIGINT to the remote
processes. Am I making this up ?

TIA,

Flavien.

Flavien,

you are right, this sounds like an old problem. I can redirect you to 
this thread 
http://www.derkeiler.com/Newsgroups/comp.security.ssh/2007-06/msg00139.html 
and this bug https://bugzilla.mindrot.org/show_bug.cgi?id=396 in which 
it is discussed and a patch proposed for 4.6p1. I think that modifying 
the proposed patch for the latest version should be straightforward.

IMHO, having the proposed option "RemoteCommandCleanup" in the main 
branch would be really interesting/necessary.

Regards,
Matthieu

Flavien a ?crit :
> Dear ssh gurus,
>
>
> Here's the version I'm testing on :
>     flavien :/$ ssh -V
>     OpenSSH_5.3p1 Debian-3ubuntu4, OpenSSL 0.9.8k 25 Mar 2009
>
> I launch a remote command :
>     flavien$ ssh -o ControlMaster=yes -o ControlPath=/tmp/ssh-control
localhost 'echo pid:$$ ...sleeping...; sleep 2803'
>     flavien at localhost's password: 
>     pid:11565 ...sleeping...
>
> On another shell, I kill the ssh client :
>     flavien$ ssh -o ControlPath=/tmp/ssh-control -O exit localhost
>     Exit request sent.
>     flavien$
>
> The ssh client is killed in the original terminal. Fine. However, the
> shell at the other end is not killed :
>     flavien$ ps -ef | grep 2803
>     flavien  11565     1  0 10:37 ?        00:00:00 bash -c echo pid:$$
...sleeping...; sleep 2803
>     flavien  11566 11565  0 10:37 ?        00:00:00 sleep 2803
>
> If the remote process is an interactive shell, however, it is killed
> once the ssh client terminates. I suspect that's because the shell
> gets an EOF on its stdin. Is this true ?
>
> For reading this ml for a few years, I have in mind some sort of
> "signal forwarding" feature that would send a SIGINT to the
remote
> processes. Am I making this up ?
>
> TIA,
>
> Flavien.
> _______________________________________________
> openssh-unix-dev mailing list
> openssh-unix-dev at mindrot.org
> https://lists.mindrot.org/mailman/listinfo/openssh-unix-dev
>