[prev in list] [next in list] [prev in thread] [next in thread] 

List:       kernel-hardening
Subject:    [kernel-hardening] Re: [lkp-robot] [printk] 7f7c60e066: BUG:KASAN:slab-out-of-bounds
From:       Kees Cook <keescook () chromium ! org>
Date:       2017-10-31 2:47:04
Message-ID: CAGXu5jJk3nm98URfa+r-HwHA_qH-GOTrHCahDup4u17HiGcORA () mail ! gmail ! com
[Download RAW message or body]

On Mon, Oct 30, 2017 at 7:39 PM, Tobin C. Harding <me@tobin.cc> wrote:
> On Mon, Oct 30, 2017 at 05:14:59PM -0700, Kees Cook wrote:
> > On Wed, Oct 25, 2017 at 9:22 AM, kernel test robot
> > <xiaolong.ye@intel.com> wrote:
> 
> thanks for looking at this, I was at a loss as to what (if any) action I
> needed to take.

I have finally learned after several instances that if 0day produces a
report, it's real. :) Even the crazy stuff I haven't been able to
reproduce on my own has come back to bite me later, so now I'm very
attentive to 0day report. ;)

> > > FYI, we noticed the following commit (built with gcc-4.9):
> > > 
> > > commit: 7f7c60e0663645e757e520245606fde9c6e326bb ("printk: hash addresses \
> > >                 printed with %p")
> > > url: https://github.com/0day-ci/linux/commits/Tobin-C-Harding/printk-hash-addresses-printed-with-p/20171024-231922
> > > 
> > 
> > It's not clear to me which of the various versions this test ran
> > against, but it seems like the printf self-tests got very confused by
> > the results:
> > 
> > > [   40.275423] test_printf: kvasprintf(..., "%p %p", ...) returned '3cf9adbe \
> > > eff717bf', expected '0000000001234567 fffffffffedcba98' [   40.296739] \
> > > test_printf: vsnprintf(buf, 256, "|%-*p|%*p|", ...) returned 19, expected 39 [  \
> > > 40.322776] test_printf: vsnprintf(buf, 16, "|%-*p|%*p|", ...) returned 19, \
> > > expected 39 [   40.334834] test_printf: vsnprintf(buf, 0, "|%-*p|%*p|", ...) \
> > > returned 19, expected 39
> > 
> > I assume v10 will fix the width issues, but probably not the value tests...
> 
> Oh, so I need to update lib/test_printf.c to cover hashed %p.

Yeah, looks like it.

> > And it claims a use-after-free, too:
> > 
> > > [   39.757461] The buggy address belongs to the object at 22cb34bb
> > > [   39.757461]  which belongs to the cache kmalloc-32 of size 32
> > > [   39.757461] The buggy address is located 0 bytes inside of
> > > [   39.757461]  32-byte region [22cb34bb, 24ac3a60)
> > 
> > Which becomes rather unreadable, since the address got hashed. :P
> 
> So I think we need to patch mm/kasan/report to use %pK instead of %p.

Yeah, looks like it.

> I don't know what I should be doing about
> 
> [   39.757461] BUG: KASAN: slab-out-of-bounds in __test+0xee/0x13f

Yeah, I have no idea where this error came from. It does seem
triggered by the printf self-test code, though, so if a KASAN bulid
doesn't show it, maybe a full SLAB DEBUG build will?

-Kees

-- 
Kees Cook
Pixel Security


[prev in list] [next in list] [prev in thread] [next in thread]