Re: [PATCH] mm: Revert x86_64 and arm64 ELF_ET_DYN_BASE base

[Dmitry Vyukov]

On Sat, 17 Feb 2024 at 07:50, Kees Cook <keescook@xxxxxxxxxxxx> wrote:
>
> *extreme thread[1] necromancy*
>
> On Mon, Aug 07, 2017 at 01:15:42PM -0700, Kees Cook wrote:
> > Moving the x86_64 and arm64 PIE base from 0x555555554000 to 0x000100000000
> > broke AddressSanitizer. This is a partial revert of:
> >
> >   commit eab09532d400 ("binfmt_elf: use ELF_ET_DYN_BASE only for PIE")
> >   commit 02445990a96e ("arm64: move ELF_ET_DYN_BASE to 4GB / 4MB")
> >
> > The AddressSanitizer tool has hard-coded expectations about where
> > executable mappings are loaded. The motivation for changing the PIE
> > base in the above commits was to avoid the Stack-Clash CVEs that
> > allowed executable mappings to get too close to heap and stack. This
> > was mainly a problem on 32-bit, but the 64-bit bases were moved too,
> > in an effort to proactively protect those systems (proofs of concept
> > do exist that show 64-bit collisions, but other recent changes to fix
> > stack accounting and setuid behaviors will minimize the impact).
>
> I happened to be looking at this again today, and wondered where things
> stood. It seems like ASan's mappings are documented here:
> https://github.com/google/sanitizers/wiki/AddressSanitizerAlgorithm#64-bit
>
> This implies that it would be safe to move the ELF_ET_DYN_BASE from
> 0x555555554000 down to 0x200000000000, since the shadow map ends at
> 0x10007fff7fff. (Well, anything above there would work, I was just
> picking a "round" number above it. We could just as well use
> 0x100080000000, I think.)
>
> Is this correct? I'd like to open up some more room between mmap and
> stack...

Note that there is also TSAN and MSAN with their own mappings.
These are also different per-arch, e.g. TSAN/Linux/x86_64:
https://github.com/llvm/llvm-project/blob/d2a26a7bd5fc7cc5752337b7f4f999642feb37dc/compiler-rt/lib/tsan/rtl/tsan_platform.h#L48-L58

Search "linux/" in that file for other arches, e.g.:
https://github.com/llvm/llvm-project/blob/d2a26a7bd5fc7cc5752337b7f4f999642feb37dc/compiler-rt/lib/tsan/rtl/tsan_platform.h#L156-L165
https://github.com/llvm/llvm-project/blob/d2a26a7bd5fc7cc5752337b7f4f999642feb37dc/compiler-rt/lib/tsan/rtl/tsan_platform.h#L187-L196
https://github.com/llvm/llvm-project/blob/d2a26a7bd5fc7cc5752337b7f4f999642feb37dc/compiler-rt/lib/tsan/rtl/tsan_platform.h#L218-L227
https://github.com/llvm/llvm-project/blob/d2a26a7bd5fc7cc5752337b7f4f999642feb37dc/compiler-rt/lib/tsan/rtl/tsan_platform.h#L252-L263

And MSAN mappings:
https://github.com/llvm/llvm-project/blob/d2a26a7bd5fc7cc5752337b7f4f999642feb37dc/compiler-rt/lib/msan/msan.h#L44-L61


> Thanks!
>
> -Kees
>
> [1] https://lore.kernel.org/lkml/20170807201542.GA21271@beast/
>
> >
> > The new 32-bit PIE base is fine for ASan (since it matches the ET_EXEC
> > base), so only the 64-bit PIE base needs to be reverted to let x86 and
> > arm64 ASan binaries run again. Future changes to the 64-bit PIE base on
> > these architectures can be made optional once a more dynamic method for
> > dealing with AddressSanitizer is found. (e.g. always loading PIE into
> > the mmap region for marked binaries.)
> >
> > Reported-by: Kostya Serebryany <kcc@xxxxxxxxxx>
> > Cc: stable@xxxxxxxxxxxxxxx
> > Signed-off-by: Kees Cook <keescook@xxxxxxxxxxxx>
> > ---
> >  arch/arm64/include/asm/elf.h | 4 ++--
> >  arch/x86/include/asm/elf.h   | 4 ++--
> >  2 files changed, 4 insertions(+), 4 deletions(-)
> >
> > diff --git a/arch/arm64/include/asm/elf.h b/arch/arm64/include/asm/elf.h
> > index acae781f7359..3288c2b36731 100644
> > --- a/arch/arm64/include/asm/elf.h
> > +++ b/arch/arm64/include/asm/elf.h
> > @@ -114,10 +114,10 @@
> >
> >  /*
> >   * This is the base location for PIE (ET_DYN with INTERP) loads. On
> > - * 64-bit, this is raised to 4GB to leave the entire 32-bit address
> > + * 64-bit, this is above 4GB to leave the entire 32-bit address
> >   * space open for things that want to use the area for 32-bit pointers.
> >   */
> > -#define ELF_ET_DYN_BASE              0x100000000UL
> > +#define ELF_ET_DYN_BASE              (2 * TASK_SIZE_64 / 3)
> >
> >  #ifndef __ASSEMBLY__
> >
> > diff --git a/arch/x86/include/asm/elf.h b/arch/x86/include/asm/elf.h
> > index 1c18d83d3f09..9aeb91935ce0 100644
> > --- a/arch/x86/include/asm/elf.h
> > +++ b/arch/x86/include/asm/elf.h
> > @@ -247,11 +247,11 @@ extern int force_personality32;
> >
> >  /*
> >   * This is the base location for PIE (ET_DYN with INTERP) loads. On
> > - * 64-bit, this is raised to 4GB to leave the entire 32-bit address
> > + * 64-bit, this is above 4GB to leave the entire 32-bit address
> >   * space open for things that want to use the area for 32-bit pointers.
> >   */
> >  #define ELF_ET_DYN_BASE              (mmap_is_ia32() ? 0x000400000UL : \
> > -                                               0x100000000UL)
> > +                                               (TASK_SIZE / 3 * 2))
> >
> >  /* This yields a mask that user programs can use to figure out what
> >     instruction set this CPU supports.  This could be done in user space,
> > --
> > 2.7.4
> >
> >
> > --
> > Kees Cook
> > Pixel Security
>
> --
> Kees Cook