Re: [PATCH] locking/refcounts, x86/asm: Use unique .text section for refcount exceptions

[Ard Biesheuvel]

On 2 September 2017 at 03:58, Kees Cook <keescook@xxxxxxxxxxxx> wrote:
> On Fri, Sep 1, 2017 at 2:43 PM, Ard Biesheuvel
> <ard.biesheuvel@xxxxxxxxxx> wrote:
>> On 1 September 2017 at 21:22, Kees Cook <keescook@xxxxxxxxxxxx> wrote:
>>> Using .text.unlikely for refcount exceptions isn't safe because gcc may
>>> move entire functions into .text.unlikely (e.g. in6_dev_get()), which
>>> would cause any uses of a protected refcount_t function to stay inline
>>> with the function, triggering the protection unconditionally:
>>>
>>>         .section        .text.unlikely,"ax",@progbits
>>>         .type   in6_dev_get, @function
>>> in6_dev_getx:
>>> .LFB4673:
>>>         .loc 2 4128 0
>>>         .cfi_startproc
>>> ...
>>>         lock; incl 480(%rbx)
>>>         js 111f
>>>         .pushsection .text.unlikely
>>> 111:    lea 480(%rbx), %rcx
>>> 112:    .byte 0x0f, 0xff
>>> .popsection
>>> 113:
>>>
>>> This creates a unique .text section and adds an additional test to the
>>> exception handler to WARN in the case of having none of OF, SF, nor ZF
>>> set so we can see things like this more easily in the future.
>>>
>>> Reported-by: Mike Galbraith <efault@xxxxxx>
>>> Fixes: 7a46ec0e2f48 ("locking/refcounts, x86/asm: Implement fast refcount overflow protection")
>>> Signed-off-by: Kees Cook <keescook@xxxxxxxxxxxx>
>>> ---
>>>  arch/x86/Kconfig                  | 2 +-
>>>  arch/x86/include/asm/refcount.h   | 2 +-
>>>  arch/x86/mm/extable.c             | 7 ++++++-
>>>  include/asm-generic/vmlinux.lds.h | 1 +
>>>  4 files changed, 9 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
>>> index eaa8ff41f424..c6acdcdb3fc6 100644
>>> --- a/arch/x86/Kconfig
>>> +++ b/arch/x86/Kconfig
>>> @@ -56,7 +56,7 @@ config X86
>>>         select ARCH_HAS_MMIO_FLUSH
>>>         select ARCH_HAS_PMEM_API                if X86_64
>>>         # Causing hangs/crashes, see the commit that added this change for details.
>>> -       select ARCH_HAS_REFCOUNT                if BROKEN
>>> +       select ARCH_HAS_REFCOUNT
>>>         select ARCH_HAS_UACCESS_FLUSHCACHE      if X86_64
>>>         select ARCH_HAS_SET_MEMORY
>>>         select ARCH_HAS_SG_CHAIN
>>> diff --git a/arch/x86/include/asm/refcount.h b/arch/x86/include/asm/refcount.h
>>> index ff871210b9f2..4e44250e7d0d 100644
>>> --- a/arch/x86/include/asm/refcount.h
>>> +++ b/arch/x86/include/asm/refcount.h
>>> @@ -15,7 +15,7 @@
>>>   * back to the regular execution flow in .text.
>>>   */
>>>  #define _REFCOUNT_EXCEPTION                            \
>>> -       ".pushsection .text.unlikely\n"                 \
>>> +       ".pushsection .text..refcount\n"                \
>>
>> You could also use a .subsection here: those are always emitted out of
>> line, but into the same section.
>
> I considered it (especially after looking to see how you handled it in
> the arm64 port) but decided against it in favor of collecting them all
> at the end with .text.unlikely.
>

OK. On arm64, I needed this because the out of line sequence refers
back to the inline code, which resulted in section mismatches if the
refcount API is used from __init code. If that problem does not exist
on x86, I guess it doesn't matter.