Re: [WARNING] x86/mm: Found insecure W+X mapping at address ..

[Thomas Gleixner]

On Tue, 23 May 2017, Thomas Gleixner wrote:
> On Tue, 23 May 2017, Kees Cook wrote:
> > On Tue, May 23, 2017 at 8:40 AM, Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:
> > > As of 4.12-rc1 one of my machines triggers the insecure W+X mapping.
> >
> > > It's consistenly 9 entries close to the beginning of the module space,
> > > before the first actual module starts. See below.
> > >
> > > Any ideas which avoid bisecting would be appreciated.
> > 
> > Is this the same as:
> > 
> > https://lkml.org/lkml/2017/5/19/899
> > 
> > ?
> > 
> > The location is very similar.
> 
> CONFIG_KPROBES_SANITY_TEST=n does not make it go away, but I suspect it's
> something in that area, as I recently switched on all of these self tests.
> 
> I'll dig into it later today. Need to walk the dogs first.

It's not KPROBES, it's the new fangled ftrace trampoline code. I added a
few printks. All the leaked W+X mappings are allocated via this callchain:

[    2.620465]  module_alloc+0x8e/0xa0
[    2.620764]  ? __fentry__+0x10/0x10
[    2.621049]  arch_ftrace_update_trampoline+0x9f/0x220
[    2.621453]  ? ftrace_caller+0x64/0x64
[    2.621754]  ? __fentry__+0x10/0x10
[    2.622047]  ftrace_startup+0x90/0x200
[    2.622352]  register_ftrace_function+0x50/0x70
[    2.622725]  function_trace_init+0x6d/0xa0
[    2.623057]  trace_selftest_startup_function+0x63/0x4a8
[    2.623477]  run_tracer_selftest+0xfe/0x16c
[    2.623813]  init_trace_selftests+0x5d/0x103
[    2.624163]  ? set_tracepoint_printk+0x3d/0x3d
[    2.624525]  do_one_initcall+0x44/0x170
[    2.624845]  kernel_init_freeable+0x1ff/0x287
[    2.625199]  ? rest_init+0x140/0x140
[    2.625491]  kernel_init+0xe/0x110
[    2.625775]  ret_from_fork+0x2e/0x40

Thanks,

	tglx