Money trouble at OpenBSD [LWN.net]

Money trouble at OpenBSD

Posted Mar 24, 2006 16:08 UTC (Fri) by landley (guest, #6789)
In reply to: Money trouble at OpenBSD by job
Parent article: Money trouble at OpenBSD

OpenSSH still has the following bug:

ssh user@somewhere "sleep 100& echo hello"

It should exit right after printing hello, but it doesn't. It hangs until the background process goes away. Nothing else behaves like that. Telnet doesn't do that. OpenSSH doesn't show this bug when running on OpenBSD.

I ran into this back in 2001: http://www.ussg.iu.edu/hypermail/linux/kernel/0105.0/0039.html

And of course I reported it to the OpenSSH developers. They said it was a known issue, and they refused to fix it because they wanted some obscure TTY behavior or Linux to change to match OpenBSD.

http://www.cs.helsinki.fi/linux/linux-kernel/2001-23/0401.html

I asked the Linux developers and they said the Linux kernel's behavior was correct (presumably off-list because I can't find it in the archives), and they said it was a clear bug in OpenSSH, and Theo was smoking something as usual. This was probably Alan Cox. :)

All this was FIVE YEARS AGO, and OpenSSH still has this bug (only when running on Linux, not when running on OpenBSD), and Theo refuses to fix it because of his personal opinions about Linux.

(Log in to post comments)

Money trouble at OpenBSD

Posted Mar 25, 2006 4:02 UTC (Sat) by landley (guest, #6789) [Link]

There existed a patch back in 2001. (Somebody emailed it to me in
response to my LKML posts.) It had been submitted to the OpenBSD guys
already, and they categorically rejected it because they insisted Linux's
behavior was wrong, not theirs.

Rob

Money trouble at OpenBSD

Posted Mar 30, 2006 7:09 UTC (Thu) by djm (subscriber, #11651) [Link]

Sorry, but your statement of the history of this problem is completely untrue. I am an OpenSSH and OpenBSD developer who spent many hours trying to get to the bottom of this.

Your assertion that "Theo won't fix it because it is only Linux" is simply a lie and demeans those of us who try to make OpenSSH a quality product for all operating systems.

Your LKML link talks about a completely different issue (ssh TCP connection attempts vs SYN timeouts) that was fixed *years* ago. See the ConnectionAttempts and ConnectTimeout options in "man ssh_config".

You can see the history of the real hang on exit bug here: http://bugzilla.mindrot.org/show_bug.cgi?id=52

There have been several simple but incorrect patches proposed for this problem, all of which create other problems. We have *never* asked for Linux kernel behaviour to change, only for someone more knowledgable about the interaction between ttys, grandchild processes and select() to assist us debug the problem.

Only recently has someone actually done this work (patch is in bugzilla above) - if it works out OK then it will probably be committed in the near future. However, very few people have actually stepped forward to test the patch.