Re: [PATCH 05/15] x86: Implement function_nocfi

[Sami Tolvanen]

On Fri, Apr 16, 2021 at 2:18 PM Borislav Petkov <bp@xxxxxxxxx> wrote:
>
> On Fri, Apr 16, 2021 at 01:38:34PM -0700, Sami Tolvanen wrote:
> > With CONFIG_CFI_CLANG, the compiler replaces function addresses in
> > instrumented C code with jump table addresses. This change implements
> > the function_nocfi() macro, which returns the actual function address
> > instead.
> >
> > Signed-off-by: Sami Tolvanen <samitolvanen@xxxxxxxxxx>
> > ---
> >  arch/x86/include/asm/page.h | 14 ++++++++++++++
> >  1 file changed, 14 insertions(+)
> >
> > diff --git a/arch/x86/include/asm/page.h b/arch/x86/include/asm/page.h
> > index 7555b48803a8..5499a05c44fc 100644
> > --- a/arch/x86/include/asm/page.h
> > +++ b/arch/x86/include/asm/page.h
> > @@ -71,6 +71,20 @@ static inline void copy_user_page(void *to, void *from, unsigned long vaddr,
> >  extern bool __virt_addr_valid(unsigned long kaddr);
> >  #define virt_addr_valid(kaddr)       __virt_addr_valid((unsigned long) (kaddr))
> >
> > +#ifdef CONFIG_CFI_CLANG
>
> Almost every patch is talking about this magical config symbol but it
> is nowhere to be found. How do I disable it, is there a Kconfig entry
> somewhere, etc, etc?

As I mentioned in the cover letter, this series is based on
linux-next. I forgot to include a link to the original patch series
that adds CONFIG_CFI_CLANG, but I'll be sure to point to it in the
next version. Sorry about the confusion.

> > +/*
> > + * With CONFIG_CFI_CLANG, the compiler replaces function address
> > + * references with the address of the function's CFI jump table
> > + * entry. The function_nocfi macro always returns the address of the
> > + * actual function instead.
> > + */
> > +#define function_nocfi(x) ({                                         \
> > +     void *addr;                                                     \
> > +     asm("leaq " __stringify(x) "(%%rip), %0\n\t" : "=r" (addr));    \
> > +     addr;                                                           \
> > +})
>
> Also, eww.
>
> It seems all these newfangled things we get, mean obfuscating code even
> more. What's wrong with using __nocfi instead? That's nicely out of the
> way so slap that in front of functions.

__nocfi only disables CFI checking in a function, the compiler still
changes function addresses to point to the CFI jump table, which is
why we need function_nocfi().

> Also, did you even build this on, say, 32-bit allmodconfig?
>
> Oh well.
>
> In file included from ./include/linux/ftrace.h:22:0,
>                  from ./include/linux/init_task.h:9,
>                  from init/init_task.c:2:
> ./include/linux/ftrace.h: In function ‘ftrace_init_nop’:
> ./arch/x86/include/asm/ftrace.h:9:40: error: implicit declaration of function ‘function_nocfi’ [-Werror=implicit-function-declaration]

This is defined in linux-next, but I do see another issue, which I'll
fix in v2. Note that CFI_CLANG itself cannot be selected on 32-bit
x86.

Sami