Re: [Qemu-devel] [PATCH 0/3] iotests: Fix test 039

[Michael Mueller <mimu@linux.vnet.ibm.com>]

On Tue, 25 Nov 2014 15:04:38 +0100
Markus Armbruster <armbru@redhat.com> wrote:

> Max Reitz <mreitz@redhat.com> writes:
> 
> > On 2014-11-25 at 14:48, Markus Armbruster wrote:
> >> Max Reitz <mreitz@redhat.com> writes:
> >>
> >>> On 2014-11-25 at 14:20, Markus Armbruster wrote:
> >>>> Max Reitz <mreitz@redhat.com> writes:
> >>>>
> >>>>> On 2014-11-25 at 13:21, Markus Armbruster wrote:
> >>>>>> Max Reitz <mreitz@redhat.com> writes:
> >>>>>>
> >>>>>>> Test 039 used to fail
> >>>>>> I'm confused: "used to" suggests it doesn't anymore, but you sending a
> >>>>>> patches strongly suggests something's broken.
> >>>>> Well, it used to fail before this series. :-P
> >>>>>
> >>>>> You're right, this sounds bad. Currently, 039 does fail, at least on
> >>>>> any system with a /proc/sys/kernel/core_pattern passing the dump to
> >>>>> another program. After this series, it does no longer.
> >>>>>
> >>>>>>>                          because qemu-io -c abort may generate core dumps
> >>>>>>> even with ulimit -c 0 (and the output then contains "(core dumped)").
> >>>>>> How?
> >>>>> See the patches[1][2] by Mao Chuan Li. If
> >>>>> /proc/sys/kernel/core_pattern passes the dump to another program,
> >>>>> ulimit -c 0 does not matter.
> >>>>>
> >>>>> [1] http://lists.nongnu.org/archive/html/qemu-devel/2014-11/msg02092.html
> >>>>> [2] http://lists.nongnu.org/archive/html/qemu-devel/2014-11/msg02093.html
> >>>>>
> >>>>> The problem with those patches is that they require access to
> >>>>> /proc/sys/kernel/core_pattern. I don't like having to run the iotests
> >>>>> as root.
> >>>> To me, this sounds like a case of "doctor, it hurts when I do this".
> >>> What do you mean? That I don't want the iotests to run as root? Or
> >>> that I don't want to go the alternative of filtering out the "(core
> >>> dumped)" message?
> >> I mean:
> >>
> >>      Doctor, it hurts when I write weird stuff to
> >>      /proc/sys/kernel/core_pattern.
> >>
> >>      Don't do that then.
> >>
> >> If you want to be a nicer doc than me, go right ahead.
> >
> > I don't write weird stuff there. My default system configuration does
> > (and mine is not the only one):
> >
> > $ uname -r
> > 3.17.3-200.fc20.x86_64
> > $ cat /proc/sys/kernel/core_pattern
> > |/usr/sbin/chroot /proc/%P/root /usr/libexec/abrt-hook-ccpp %s %c %p
> > %u %g %t e
> 
> abrt is one of the things I kill with prejudice on all my development
> machines.

Markus,

we use non-personal test machines that shall automatically start tests like iotests when patches
are committed, thus this series really helps... Thanks.

Michael

>