Upstart 0.5.0, /proc, and telinit

[Garrett Cooper]

On Mon, Feb 23, 2009 at 7:06 PM, Jeff Oliver
<Jeffrey.Oliver at panasonic.aero> wrote:
> First off, I would say that there is no need to mount /proc, if its not
> necessary.  For the most part, job files, wouldn't even require /proc to be
> mounted.  Following on from that logic, there is only 1 place, that I can
> see, that actually attempts to use a /proc file, when upstart changes the
> oom_adj value.  So, first check to see if the job file is even changing that
> value, if not, /proc is not needed and skip that step.  Second, if it does
> need /proc, then what?  You could simply fail the job, quietly ignore the
> error, or mount proc?  I guess that part is up for debate yet.
>
> It should not use system(), it opens itself up for other problems.
>
> Jeff
>
> Garrett Cooper wrote:
>
> On Mon, Feb 23, 2009 at 2:14 PM, Adam Sampson <ats at offog.org> wrote:
>
>
> Scott James Remnant <scott at netsplit.com> writes:
>
>
>
> Nobody has supplied a patch.
>
>
> Well, here's what I'm using happily, but I suspect a little more finesse
> would be useful. ;)
>
> http://offog.org/darcs/garstow/sys/upstart/files/mountproc.diff
>
>
> Uh, that will almost certainly fail if it can't allocate a shell
> because system(3) calls execve(2) and allocates a shell.
>
> Using mount(2), like so, would be better:
>
> #include <string.h>          /* For strerror(3). */
> #include <errno.h>
> #include <sys/mount.h>
> #include <linux/kernel.h> /* For panic(2). */
>
> /* ... */
>
> if (mount("none", "/proc", "proc", 0, NULL) < 0) {
>     /* Change to OOM support not available, etc for this instead of a panic?
> */
>     panic ("Failed to mount /proc: %s\n", strerror(errno));
> }
>
>     You'll potentially need to do similar for /dev and /dev/console
> (we currently do that, but I'm not sure if /dev/console is needed...).
> Combined with proper calls to fstab(5) with getent, this should be
> solved properly in less than 30 lines.
> Cheers,
> -Garrett

Yeah, then I'd definitely just disable the oom adjustment logic then.
Do you agree Scott?
Thanks,
-Garrett