Re: [PATCH] all arches, signal: Move restart_block to struct task_struct

["H. Peter Anvin" <hpa@zytor.com>]

On x86 we have discussed to combine thread_info and task_struct into a single allocation with a percpu variable lining pointng at it.

On October 26, 2014 11:09:39 AM PDT, Al Viro <viro@ZenIV.linux.org.uk> wrote:
>On Sun, Oct 26, 2014 at 10:36:45AM -0700, Andy Lutomirski wrote:
>
>> I never said it was the *only* juicy target, but we can fix the rest,
>> too.  Also, I suspect that overwriting task could be harder to
>> exploit.  First, you need to avoid crashing, and second, on systems
>> with SMAP or similar protection, you need to make task point
>somewhere
>> that contains a useful exploit payload.
>> 
>> We could probably get rid of thread_info's task pointer on x86, too
>--
>> it's not used by get_current() any more.
>
>Huh?  If you can overwrite that pointer, you can bloody well overwrite
>->task itself, making it point into the overwritten part of stack right
>next to thread_info.
>
>Again, on most of the architectures the _only_ way to reach task_struct
>is via thread_info:
>	* everything that uses asm-generic/current.h - arm, arm64, blackfin,
>c6x, hexagon, metag, mips, openrisc, sh, um, unicore32
>	* everything that should be using it - alpha, avr32, cris, m32r,
>parisc, score, tile.  These guys can simply add generic-y += current.h
>into their asm/Kbuild and remove asm/current.h.
>	* nearly the same situation - xtensa (there's an asm variant of
>the same thing + copy of asm-generic/current.h for C)
>	* sparc32
>	* m68k-noMMU
>	* mn10300-SMP
>
>It's a strong majority.  Check arch/*/asm/current.h and see for
>yourself.

-- 
Sent from my mobile phone.  Please pardon brevity and lack of formatting.