[arch-dev-public] openssh 5.3p1-4: sshd script kills all sshd processes??

[Allan McRae]

On 10/03/10 09:27, Dan McGee wrote:
> On Tue, Mar 9, 2010 at 5:21 PM, Allan McRae<allan at archlinux.org>  wrote:
>> On 10/03/10 09:03, Thomas Bächler wrote:
>>>
>>> What did whoever commited this smoke? It must have been very good:
>>>
>>> http://repos.archlinux.org/wsvn/packages/?op=comp&compare[]=%2Fopenssh%2Ftrunk@59462&compare[]=%2Fopenssh%2Ftrunk@60182
>>>
>>> Whenever you want to do /etc/rc.d/sshd restart remotely, it will kill
>>> your ssh session. It is an important feature of openssh that all ssh
>>> connections stay OPEN even if you kill the master server. With this
>>> change, you will get kicked and goodbye remote server.
>>
>> That looks like it was committed in response to this bug:
>> http://bugs.archlinux.org/task/17138
>
> Yeah this happened to me, I didn't think twice about it thinking I did
> something wrong but realized now that I have never been booted before.
> Good thing I have a remote terminal.
>
> Did this go through [testing]?

Definitely.  And got more than enough signoffs...

> Just wondering if we could/should have
> caught this earlier; maybe we need certain things done with certain
> packages to certify their OK-ness. Especially with sshd, a remote
> restart is almost always a good check that covers a lot of bases.

This has sort of been bought up before.  More in the context of not 
needing to test every possible combination of bug fixes (instead just a 
basic usage test) to signoff lesser used packages.  So this would 
suggest the opposite for some packages.  That is getting a complicated 
signoff policy.

As an aside, it took 11 days for this bug to be noticed. So considering 
a package "bug free" after a week and giving autosignoffs has issues.

Allan