Re: Oops on 2.4.x invalid procfs i_ino value
From: Marcelo Tosatti
Date: Wed Dec 22 2004 - 13:07:24 EST
On Mon, Dec 20, 2004 at 04:35:18PM -0600, Brent Casavant wrote:
> On Fri, 17 Dec 2004, William Lee Irwin III wrote:
> > On Fri, Dec 17, 2004 at 04:49:44PM -0600, Brent Casavant wrote:
> > >> On a related note, if it matters, on about half the crash dumps I've
> > >> looked at, I see a pid of 0 has been assigned to a user process,
> > >> tripping this same problem. I suspect there's another bug somewhere
> > >> that's allowing a pid of 0 to be chosen in the first place -- but I
> > >> don't totally discount that this problem may lay in SGI's patches to
> > >> this particular kernel -- I'll need to take a more thorough look.
> > On Fri, Dec 17, 2004 at 04:38:35PM -0800, William Lee Irwin III wrote:
> > > That's rather ominous. I'll pore over pid.c and see what's going on.
> > > Also, does the pid.c in your kernel version match 2.6.x-CURRENT?
> > Ouch, 2.4.21; this will be trouble. So next, what patches atop 2.4.21?
> I wouldn't worry about the pid=0 issue -- I think it's most likely
> due to the PAGG patches (http://oss.sgi.com/projects/pagg) causing
> some sort of problem at process teardown (all the pid=0 processes are
> in the process of exiting).
> I'm more concerned about the (0 == pid & 0xffff) bug, which is present
> in the unpatched mainline 2.4.x kernel. It seems that the easiest fix
> is marking such pids as in-use at pidmap allocation, so that they are
> never assigned to real tasks. I've got the code almost done, but need
> to port it to top-of-tree before submitting a patch.
Wouldnt it be feasible to have another "procfs inode type" to indicate such
lower 16-bit zeroed pid's with a new type PROC_PID_INO_ZERO16BIT (or a better
name) and have fake_ino() handle these case by then using the upper 16-bits on
the inode for this "special" pid's.
And have proc_pid_make_inode() and related code handle this new type? No?
I'm not a big fan of making such pids unuseable for real tasks, so it would be
nice if we could come up a fix for the buggy proc inode logic.
Thanks for finding this out!
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/