utmp handing (Was: [RFC][PATCH] Add support for oom_score_adj)

[Marc - A. Dahlhaus]

Am Dienstag, den 22.03.2011, 15:45 +0100 schrieb Petr Lautrbach:
> On 03/18/2011 11:09 PM, Scott James Remnant wrote:
> > Cc'ing Petr
> >
> > On Thu, Mar 17, 2011 at 3:27 PM, Marc - A. Dahlhaus<mad at wol.de>  wrote:
> >> Am 17.03.2011 21:45, schrieb Scott James Remnant:
> >>
> >>> On Thu, Mar 17, 2011 at 11:58 AM, Marc - A. Dahlhaus<mad at wol.de>  wrote:
> >>>
> >>>> BTW is the override file stuff upstreamed into the 1.X branch already?
> >>>>
> >>> No. I wanted to release a stable 1.0 with no new features, James
> >>> needed to get features in for Ubuntu natty - so for now Ubuntu is
> >>> running a fork of Upstart (whichout all the bug fixes of 1.0 + 1.1,
> >>> but with new features).
> >>
> >> while we are at it, i had some problems regarding utmp handling in 1.0 and
> >> while running a gdb session and talking with plauterbach on irc he suggested
> >> adding this patch: http://plautrba.fedorapeople.org/upstart-utmp.patch
> >>
> >> The patch fixed the utmp handling on my box which is running an LFS based
> >> system with unpatched upstream versions of glibc-2.13, util-linux-2.19
> >> (agetty) and shadow-4.1.4.3 (login).
> 
> login from shadow-4.1.4.3 is the only difference I can see here. We use /bin/login from
> util-linux in Fedora and as I've skim through sources, login from shadow packages handles
> non-existing utmp entries slightly different. That's probably why there is no
> opened record after first login.

Thanks for the nice detective work :)

Debian and Ubuntu use /bin/login from shadow AFAICT.

So the patch mentioned above would be missing there also...

> >>
> >> The problem was that upstart couldn't find the first opened utmp record on
> >> logout from local ttys for each tty after a clean boot. The behavior was the
> >> same even on a empty utmp file after a boot. SSH ttys where handled correctly
> >> so a buggy glibc was ruled out in our tests.
> >>
> >> I didn't debugged it further after i found out that the patch mentioned above
> >> fixed the bug for me.
> >>
> 
> 
> -- 
> Petr Lautrbach, Red Hat, Inc.
> 

Marc