Control: reassign -1 src:linux 3.15~rc5-1~exp1
Control: found -1 src:linux 3.16.3-2
Control: tag -1 upstream fixed-upstream patch
Control: affects -1 dpkg
On Thu, 2014-10-02 at 07:28 +0200, Guillem Jover wrote:
> Hi!
>
> On Wed, 2014-10-01 at 17:43:33 -0400, Paul Kimoto wrote:
> > Package: dpkg
> > Version: 1.16.15
>
> > My kernel is the wheezy backport 3.16.3-2~bpo70+1
> > (linux-image-3.16-0.bpo.2-486).
> >
> > I tried to update rsyslog from 5.8.11-3 to 5.8.11-3+deb7u1 this morning
> > (Oct 1). Aptitude told me first that rsyslogd couldn't be stopped because
> > it wasn't running, then that the new rsyslogd couldn't be started because
> > rsyslogd was already running.
> >
> > rsyslogd _is_ running, but can't be stopped using either /usr/sbin/service
> > or /sbin/start-stop-daemon.
> >
> > $ /usr/sbin/service rsyslog stop
> > [ ok ] Stopping enhanced syslogd: rsyslogd already stopped.
> > $ /sbin/start-stop-daemon --stop --verbose --test --pidfile \
> > /var/run/rsyslogd.pid --exec /usr/sbin/rsyslogd
> > No /usr/sbin/rsyslogd found running; none killed.
> > $ /bin/ps uw $(/bin/cat /var/run/rsyslogd.pid)
> > USER PID %CPU %MEM VSZ RSS TTY STAT START TIME COMMAND
> > root 2838 0.0 0.1 30252 2532 ? Sl Sep30 0:01 /usr/sbin/rsyslogd -c5
> >
> > I noticed that the executable name in /proc is a little funny:
> >
> > $ /bin/ls -l /proc/$(cat /var/run/rsyslogd.pid)/exe
> > lrwxrwxrwx 1 root root 0 Sep 30 02:16 /proc/2838/exe -> /usr/sbin/rsyslogd.dpkg-new (deleted)
> >
> > The ".dpkg-new" suffix doesn't seem to have been considered by
> > pid_is_exec() in start-stop-daemon.c. Could my problem be a consequence
> > of a recent "File cross-renaming support" feature
> > (http://kernelnewbies.org/Linux_3.15) advertised in Linux 3.15?
>
> Yeah, that seems to be it, also see <https://lwn.net/Articles/614057/>.
>
> So not a dpkg bug, but I guess this will need to be fixed in the
> linux 3.16 packages in unstable first, before it can be pulled into a
> backport.
>
> Kernel team, as I'm not sure if this is fixed in sid already, I've not
> reassigned, do you know off-hand? Should I reassign?
It was only just fixed upstream so is not fixed in Debian yet.
I *think* we need to cherry-pick these two:
commit a28ddb87cdddb0db57466ba7f59f831002f4340c
Author: Linus Torvalds <torvalds@linux-foundation.org>
Date: Wed Sep 24 12:27:39 2014 -0700
fold swapping ->d_name.hash into switch_names()
commit d2fa4a8476b911782f7e5167db18770222ac40c3
Author: Mikhail Efremov <sem@altlinux.org>
Date: Wed Sep 24 22:14:33 2014 +0400
vfs: Don't exchange "short" filenames unconditionally.
Ben.
--
Ben Hutchings
Everything should be made as simple as possible, but not simpler.
- Albert Einstein
Attachment:
signature.asc
Description: This is a digitally signed message part