Re: [PATCH 4.4 00/37] 4.4.110-stable review

[Greg Kroah-Hartman]

On Fri, Jan 05, 2018 at 02:12:33AM -0800, Kees Cook wrote:
> On Thu, Jan 4, 2018 at 9:33 PM, Andy Lutomirski <luto@xxxxxxxxxxxxxx> wrote:
> > On Thu, Jan 4, 2018 at 12:43 PM, Andy Lutomirski <luto@xxxxxxxxxxxxxx> wrote:
> >>
> >>> On Jan 4, 2018, at 12:29 PM, Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> >>>
> >>>> On Thu, Jan 4, 2018 at 12:16 PM, Thomas Voegtle <tv@xxxxxxxx> wrote:
> >>>>
> >>>> Attached a screenshot.
> >>>> Is that useful? Are there some debug options I can add?
> >>>
> >>> Not much of an oops, because the SIGSEGV happens in user space. The
> >>> only reason you get any kernel stack printout at all is because 'init'
> >>> dying will make the kernel print that out.
> >>>
> >>> The segfault address for init looks like the fixmap area to me (first
> >>> byte in the last page of the fixmap?). "Error 5" means that it's a
> >>> user-space read that got a protection fault. So it's not a LDT of GDT
> >>> update or anything like that, it's a normal access from user space (or
> >>> a qemu emulation bug, but that sounds unlikely).
> >>>
> >>> Is that the vsyscall page?
> >>>
> >>> Adding Luto to the participants. I think he noticed one of the
> >>> vsyscall patches missing earlier in the 4.9 series. Maybe the 4.4
> >>> series had something similar..
> >>>
> >>
> >> That's almost certainly it.
> >>
> >> I'll try to find some time today or tomorrow to add a proper selftest.
> >>
> >
> > Give this a shot:
> >
> > https://git.kernel.org/pub/scm/linux/kernel/git/luto/linux.git/commit/?h=x86/pti&id=17c5ebeb2e00879b0af1a9c32bf37ecdd9b9b31b
> >
> > Boot with each of vsyscall=none, vsyscall=native, and vsyscall=emulate
> > and run both the 32-bit and 64-bit variants of that test.  All six
> > combinations should pass.  But I bet they don't on 4.4.
> 
> With my 4.4.110-rc1 under QEMU -cpu=host (Xeon E5-2690 v3)
> 
> vsyscall=emulate:
> 
> # ./test_vsyscall_64
> ...
> [RUN]   Checking read access to the vsyscall page
> [FAIL]  We don't have read access, but we should
> 
> vsyscall=native:
> 
> # ./test_vsyscall_64
> ...
> [RUN]   Checking read access to the vsyscall page
> [FAIL]  We don't have read access, but we should
> 
> Everything else passes.

I get this same error with the latest 4.9-rc tree as well, but it works
just fine on 4.15-rc6.

I'll look at the proposed patches now for this...

thanks so much for the test tool.

greg k-h