Re: [PATCH] core_pattern: set core helpers root and namespace tocrashing process

[Neil Horman]

On Thu, Dec 13, 2012 at 02:25:34PM -0800, Andrew Morton wrote:
> On Thu, 13 Dec 2012 13:12:20 -0500
> Neil Horman <nhorman@xxxxxxxxxxxxx> wrote:
> 
> > On Thu, Dec 13, 2012 at 06:20:48AM -0600, Serge Hallyn wrote:
> > > Quoting Neil Horman (nhorman@xxxxxxxxxxxxx):
> > > > Theres one problem I currently see with it, and that is that I'm not sure we can
> > > > change the current behavior of how the root fs is set for the pipe reader, lest
> > > > we break some user space expectations. As such, I've added a sysctl in this
> > > > patch to allow administrators to globally select if a core reader specified via
> > > > /proc/sys/kernel/core_pattern should use the global rootfs, or the (possibly)
> > > > chrooted fs of the crashing process.
> > > 
> > > Practical question:  How is the admin to make an educated decision on
> > > how to set the sysctl?
> 
> By reading the documentation which Neil didn't include?
> 
Yeah, that was stupid of me, I'll respin this with docs.

> > My thought was that the admin typically wouldn't touch this at all.  I really
> > added it as a backwards compatibility option only.  Setting the user space
> > helper task to the root of the crashing parent has the possibility of breaking
> > existing installs because the core_pattern helper might be expecting global file
> > system access.  Moving forward, my expectation would be that core_pattern
> > helpers would be written with the default setting in mind, and we could
> > eventually deprecate the control entirely.
> > 
> > If you have a better mechanism in mind however (or if you think that removing
> > the control is a resaonable approach), I'm certainly open to that.
> 
> Yeah, this is a tiresome patch but I can't think of a better way.
> 
> Except, perhaps, adding a new token to the core_pattern which says
> "switch namespaces"?
>
I like that idea, perhaps '||' instead of '|' as the leading token can indicate
"use the namespace root" vs. "use the global root".  Thoughts?

> Is there any propect that the core_pattern itself will later become a
> per-namespace containerised thing?  I guess that if the per-container
> core_pattern has been configured, we can implicitly do the namespace
> switch as well.
Yes, that makes sense.  Unfortunately, I don't see proc containerization
happening any time soon.  I suppose if we do the above tokenization, that can be
used despite any future containerization that takes place.

I'll respin the patch with documentation, and replace the extra sysctl with the
above tokenization in the AM.

Best
Neil
 
> 
> 
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/