lkml.org 
[lkml]   [2009]   [Jun]   [25]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: kmemleak false positive?
From
Date
On Thu, 2009-06-25 at 11:40 -0400, Dave Jones wrote:
> On Thu, Jun 25, 2009 at 04:25:39PM +0100, Catalin Marinas wrote:
> > > Hmm, it's pretty noisy, and everything it's found so far looks to be
> > > a false positive.
> >
> > In this case, it would make sense to enable task stacks scanning by
> > default:
> >
> > diff --git a/mm/kmemleak.c b/mm/kmemleak.c
> > index 17096d1..a38418a 100644
> > --- a/mm/kmemleak.c
> > +++ b/mm/kmemleak.c
> > @@ -194,7 +194,7 @@ static unsigned long jiffies_min_age;
> > /* delay between automatic memory scannings */
> > static signed long jiffies_scan_wait;
> > /* enables or disables the task stacks scanning */
> > -static int kmemleak_stack_scan;
> > +static int kmemleak_stack_scan = 1;
>
> heh, I just did the same patch for the rawhide kernel builds.
>
> > > > You can mount debugfs on /sys/kerne/debug and read the kmemleak file in
> > > > there (it triggers a new scan as well).
> > >
> > > Currently prints the acpi traces I already posted.
> >
> > If they are still consistently shown with stack=on, it could be a leak.
>
> Could be, though as you mentioned, with ACPI it's really hard to tell.
>
> Here's another case (with stack scanning on btw) which looks odd..
>
> kmemleak: unreferenced object 0xd86ba000 (size 16):
> kmemleak: comm "init", pid 1, jiffies 4294683556
> kmemleak: backtrace:
> kmemleak: [<c04fd8b3>] kmemleak_alloc+0x193/0x2b8
> kmemleak: [<c04f5e73>] kmem_cache_alloc+0x11e/0x174
> kmemleak: [<c05cdfdc>] avtab_insertf+0xd6/0x140
> kmemleak: [<c05ce3d7>] avtab_read_item+0x26a/0x284
> kmemleak: [<c05ce5a5>] avtab_read+0x82/0xe5
> kmemleak: [<c05d0618>] policydb_read+0x40c/0x1028
> kmemleak: [<c05d459d>] security_load_policy+0x57/0x37c
> kmemleak: [<c05c9995>] sel_write_load+0xb2/0x54a
> kmemleak: [<c0500186>] vfs_write+0x9f/0x10f
> kmemleak: [<c05002e1>] sys_write+0x58/0x8d
> kmemleak: [<c040a8eb>] sysenter_do_call+0x12/0x38
> kmemleak: [<ffffffff>] 0xffffffff
>
> I looked over the SELinux code, and couldn't see an obvious leak.
> Eric Paris came to the same conclusion.

I suspect it is a false positive caused by the current odd way in which
we update the policydb. So I would expect it to go away when we get
around to rewriting that code, already on our todo list.

However, KaiGai Kohei noticed that /sys/kernel/slab/avtab_node/objects
seems to grow upon repeated load_policy invocations (of the same policy)
for some kernels (e.g. F11 kernel) while remaining constant for the
rawhide kernel.

# for i in `seq 1 100`
> do
> load_policy
> cat /sys/kernel/slab/avtab_node/objects
> done

--
Stephen Smalley
National Security Agency



\
 
 \ /
  Last update: 2009-06-25 21:53    [from the cache]
©2003-2011 Jasper Spaans