-Wmacro-redefined in include/linux/fortify-string.h

  Hi all,

I am seeing the following set of warnings when building an x86_64
configuration that has CONFIG_FORTIFY_SOURCE=y and CONFIG_KMSAN=y:

  In file included from scripts/mod/devicetable-offsets.c:3:
  In file included from ./include/linux/mod_devicetable.h:13:
  In file included from ./include/linux/uuid.h:12:
  In file included from ./include/linux/string.h:253:
  ./include/linux/fortify-string.h:496:9: error: 'memcpy' macro redefined [-Werror,-Wmacro-redefined]
  #define memcpy(p, q, s)  __fortify_memcpy_chk(p, q, s,                  \
          ^
  ./arch/x86/include/asm/string_64.h:17:9: note: previous definition is here
  #define memcpy __msan_memcpy
          ^
  In file included from scripts/mod/devicetable-offsets.c:3:
  In file included from ./include/linux/mod_devicetable.h:13:
  In file included from ./include/linux/uuid.h:12:
  In file included from ./include/linux/string.h:253:
  ./include/linux/fortify-string.h:500:9: error: 'memmove' macro redefined [-Werror,-Wmacro-redefined]
  #define memmove(p, q, s)  __fortify_memcpy_chk(p, q, s,                 \
          ^
  ./arch/x86/include/asm/string_64.h:73:9: note: previous definition is here
  #define memmove __msan_memmove
          ^
  2 errors generated.

I can see that commit ff901d80fff6 ("x86: kmsan: use __msan_ string
functions where possible.") appears to include a fix up for this warning
with memset() but not memcpy() or memmove(). If I apply a similar fix up
like so:


or is there a different obvious fix that I am missing?

Cheers,
Nathan
  

On Wed, Oct 19, 2022 at 8:37 AM Nathan Chancellor <nathan@kernel.org> wrote:
>
> Hi all,
>
> I am seeing the following set of warnings when building an x86_64
> configuration that has CONFIG_FORTIFY_SOURCE=y and CONFIG_KMSAN=y:

I was also looking into this issue recently, because people start
running into it: https://github.com/google/kmsan/issues/89

I have a solution that redefines __underlying_memXXX to __msan_memXXX
under __SANITIZE_MEMORY__ in fortify-string.h and skips `#define
memXXX __msan_memXXX` in string_64.h, making KMSAN kinda work with
FORTIFY_SOURCE.
Dunno if that's necessary though: KMSAN is a debugging tool anyway,
and supporting it in fortify-string.h sounds excessive.

So I'm fine with disabling FORTIFY_STRING under KMSAN, unless someone objects.

>   In file included from scripts/mod/devicetable-offsets.c:3:
>   In file included from ./include/linux/mod_devicetable.h:13:
>   In file included from ./include/linux/uuid.h:12:
>   In file included from ./include/linux/string.h:253:
>   ./include/linux/fortify-string.h:496:9: error: 'memcpy' macro redefined [-Werror,-Wmacro-redefined]
>   #define memcpy(p, q, s)  __fortify_memcpy_chk(p, q, s,                  \
>           ^
>   ./arch/x86/include/asm/string_64.h:17:9: note: previous definition is here
>   #define memcpy __msan_memcpy
>           ^
>   In file included from scripts/mod/devicetable-offsets.c:3:
>   In file included from ./include/linux/mod_devicetable.h:13:
>   In file included from ./include/linux/uuid.h:12:
>   In file included from ./include/linux/string.h:253:
>   ./include/linux/fortify-string.h:500:9: error: 'memmove' macro redefined [-Werror,-Wmacro-redefined]
>   #define memmove(p, q, s)  __fortify_memcpy_chk(p, q, s,                 \
>           ^
>   ./arch/x86/include/asm/string_64.h:73:9: note: previous definition is here
>   #define memmove __msan_memmove
>           ^
>   2 errors generated.
>
> I can see that commit ff901d80fff6 ("x86: kmsan: use __msan_ string
> functions where possible.") appears to include a fix up for this warning
> with memset() but not memcpy() or memmove(). If I apply a similar fix up
> like so:
>
> diff --git a/include/linux/fortify-string.h b/include/linux/fortify-string.h
> index 4029fe368a4f..718ee17b31e3 100644
> --- a/include/linux/fortify-string.h
> +++ b/include/linux/fortify-string.h
> @@ -493,6 +493,7 @@ __FORTIFY_INLINE bool fortify_memcpy_chk(__kernel_size_t size,
>   * __struct_size() vs __member_size() must be captured here to avoid
>   * evaluating argument side-effects further into the macro layers.
>   */
> +#ifndef CONFIG_KMSAN
>  #define memcpy(p, q, s)  __fortify_memcpy_chk(p, q, s,                 \
>                 __struct_size(p), __struct_size(q),                     \
>                 __member_size(p), __member_size(q),                     \
> @@ -501,6 +502,7 @@ __FORTIFY_INLINE bool fortify_memcpy_chk(__kernel_size_t size,
>                 __struct_size(p), __struct_size(q),                     \
>                 __member_size(p), __member_size(q),                     \
>                 memmove)
> +#endif
>
>  extern void *__real_memscan(void *, int, __kernel_size_t) __RENAME(memscan);
>  __FORTIFY_INLINE void *memscan(void * const POS0 p, int c, __kernel_size_t size)
>
> Then the instances of -Wmacro-redefined disappear but the fortify tests
> no longer pass for somewhat obvious reasons:
>
>   warning: unsafe memcpy() usage lacked '__read_overflow2' symbol in lib/test_fortify/read_overflow2-memcpy.c
>   warning: unsafe memmove() usage lacked '__read_overflow2' symbol in lib/test_fortify/read_overflow2-memmove.c
>   warning: unsafe memcpy() usage lacked '__read_overflow2_field' symbol in lib/test_fortify/read_overflow2_field-memcpy.c
>   warning: unsafe memmove() usage lacked '__read_overflow2_field' symbol in lib/test_fortify/read_overflow2_field-memmove.c
>   warning: unsafe memcpy() usage lacked '__write_overflow' symbol in lib/test_fortify/write_overflow-memcpy.c
>   warning: unsafe memmove() usage lacked '__write_overflow' symbol in lib/test_fortify/write_overflow-memmove.c
>   warning: unsafe memset() usage lacked '__write_overflow' symbol in lib/test_fortify/write_overflow-memset.c
>   warning: unsafe memcpy() usage lacked '__write_overflow_field' symbol in lib/test_fortify/write_overflow_field-memcpy.c
>   warning: unsafe memmove() usage lacked '__write_overflow_field' symbol in lib/test_fortify/write_overflow_field-memmove.c
>   warning: unsafe memset() usage lacked '__write_overflow_field' symbol in lib/test_fortify/write_overflow_field-memset.c
>
> Should CONFIG_KMSAN depend on CONFIG_FORTIFY_SOURCE=n like so? It seems
> like the two features are incompatible if I am reading ff901d80fff6
> correctly.
>
> diff --git a/lib/Kconfig.kmsan b/lib/Kconfig.kmsan
> index b2489dd6503f..6a681621e3c5 100644
> --- a/lib/Kconfig.kmsan
> +++ b/lib/Kconfig.kmsan
> @@ -11,7 +11,7 @@ config HAVE_KMSAN_COMPILER
>  config KMSAN
>         bool "KMSAN: detector of uninitialized values use"
>         depends on HAVE_ARCH_KMSAN && HAVE_KMSAN_COMPILER
> -       depends on SLUB && DEBUG_KERNEL && !KASAN && !KCSAN
> +       depends on SLUB && DEBUG_KERNEL && !KASAN && !KCSAN && !FORTIFY_SOURCE
>         select STACKDEPOT
>         select STACKDEPOT_ALWAYS_INIT
>         help
>
> or is there a different obvious fix that I am missing?
>
> Cheers,
> Nathan



--
Alexander Potapenko
Software Engineer

Google Germany GmbH
Erika-Mann-Straße, 33
80636 München

Geschäftsführer: Paul Manicle, Liana Sebastian
Registergericht und -nummer: Hamburg, HRB 86891
Sitz der Gesellschaft: Hamburg
  

On Wed, Oct 19, 2022 at 08:37:19AM -0700, Nathan Chancellor wrote:
> I am seeing the following set of warnings when building an x86_64
> configuration that has CONFIG_FORTIFY_SOURCE=y and CONFIG_KMSAN=y:
> 
>   In file included from scripts/mod/devicetable-offsets.c:3:
>   In file included from ./include/linux/mod_devicetable.h:13:
>   In file included from ./include/linux/uuid.h:12:
>   In file included from ./include/linux/string.h:253:
>   ./include/linux/fortify-string.h:496:9: error: 'memcpy' macro redefined [-Werror,-Wmacro-redefined]
>   #define memcpy(p, q, s)  __fortify_memcpy_chk(p, q, s,                  \
>           ^
>   ./arch/x86/include/asm/string_64.h:17:9: note: previous definition is here
>   #define memcpy __msan_memcpy
>           ^
>   In file included from scripts/mod/devicetable-offsets.c:3:
>   In file included from ./include/linux/mod_devicetable.h:13:
>   In file included from ./include/linux/uuid.h:12:
>   In file included from ./include/linux/string.h:253:
>   ./include/linux/fortify-string.h:500:9: error: 'memmove' macro redefined [-Werror,-Wmacro-redefined]
>   #define memmove(p, q, s)  __fortify_memcpy_chk(p, q, s,                 \
>           ^
>   ./arch/x86/include/asm/string_64.h:73:9: note: previous definition is here
>   #define memmove __msan_memmove
>           ^
>   2 errors generated.
> 
> I can see that commit ff901d80fff6 ("x86: kmsan: use __msan_ string
> functions where possible.") appears to include a fix up for this warning
> with memset() but not memcpy() or memmove(). If I apply a similar fix up
> like so:
> 
> diff --git a/include/linux/fortify-string.h b/include/linux/fortify-string.h
> index 4029fe368a4f..718ee17b31e3 100644
> --- a/include/linux/fortify-string.h
> +++ b/include/linux/fortify-string.h
> @@ -493,6 +493,7 @@ __FORTIFY_INLINE bool fortify_memcpy_chk(__kernel_size_t size,
>   * __struct_size() vs __member_size() must be captured here to avoid
>   * evaluating argument side-effects further into the macro layers.
>   */
> +#ifndef CONFIG_KMSAN
>  #define memcpy(p, q, s)  __fortify_memcpy_chk(p, q, s,			\
>  		__struct_size(p), __struct_size(q),			\
>  		__member_size(p), __member_size(q),			\
> @@ -501,6 +502,7 @@ __FORTIFY_INLINE bool fortify_memcpy_chk(__kernel_size_t size,
>  		__struct_size(p), __struct_size(q),			\
>  		__member_size(p), __member_size(q),			\
>  		memmove)
> +#endif
>  
>  extern void *__real_memscan(void *, int, __kernel_size_t) __RENAME(memscan);
>  __FORTIFY_INLINE void *memscan(void * const POS0 p, int c, __kernel_size_t size)
> 
> Then the instances of -Wmacro-redefined disappear but the fortify tests
> no longer pass for somewhat obvious reasons:
> 
>   warning: unsafe memcpy() usage lacked '__read_overflow2' symbol in lib/test_fortify/read_overflow2-memcpy.c
>   warning: unsafe memmove() usage lacked '__read_overflow2' symbol in lib/test_fortify/read_overflow2-memmove.c
>   warning: unsafe memcpy() usage lacked '__read_overflow2_field' symbol in lib/test_fortify/read_overflow2_field-memcpy.c
>   warning: unsafe memmove() usage lacked '__read_overflow2_field' symbol in lib/test_fortify/read_overflow2_field-memmove.c
>   warning: unsafe memcpy() usage lacked '__write_overflow' symbol in lib/test_fortify/write_overflow-memcpy.c
>   warning: unsafe memmove() usage lacked '__write_overflow' symbol in lib/test_fortify/write_overflow-memmove.c
>   warning: unsafe memset() usage lacked '__write_overflow' symbol in lib/test_fortify/write_overflow-memset.c
>   warning: unsafe memcpy() usage lacked '__write_overflow_field' symbol in lib/test_fortify/write_overflow_field-memcpy.c
>   warning: unsafe memmove() usage lacked '__write_overflow_field' symbol in lib/test_fortify/write_overflow_field-memmove.c
>   warning: unsafe memset() usage lacked '__write_overflow_field' symbol in lib/test_fortify/write_overflow_field-memset.c
> 
> Should CONFIG_KMSAN depend on CONFIG_FORTIFY_SOURCE=n like so? It seems
> like the two features are incompatible if I am reading ff901d80fff6
> correctly.
> 
> diff --git a/lib/Kconfig.kmsan b/lib/Kconfig.kmsan
> index b2489dd6503f..6a681621e3c5 100644
> --- a/lib/Kconfig.kmsan
> +++ b/lib/Kconfig.kmsan
> @@ -11,7 +11,7 @@ config HAVE_KMSAN_COMPILER
>  config KMSAN
>  	bool "KMSAN: detector of uninitialized values use"
>  	depends on HAVE_ARCH_KMSAN && HAVE_KMSAN_COMPILER
> -	depends on SLUB && DEBUG_KERNEL && !KASAN && !KCSAN
> +	depends on SLUB && DEBUG_KERNEL && !KASAN && !KCSAN && !FORTIFY_SOURCE
>  	select STACKDEPOT
>  	select STACKDEPOT_ALWAYS_INIT
>  	help
> 
> or is there a different obvious fix that I am missing?

Hm, why can't KMSAN use the same thing KASAN does, and compose correctly
with FORTIFY? (i.e. redefine the "__underlaying_mem*" macros?)
  

On Wed, Oct 19, 2022 at 09:48:27AM -0700, Alexander Potapenko wrote:
> On Wed, Oct 19, 2022 at 8:37 AM Nathan Chancellor <nathan@kernel.org> wrote:
> >
> > Hi all,
> >
> > I am seeing the following set of warnings when building an x86_64
> > configuration that has CONFIG_FORTIFY_SOURCE=y and CONFIG_KMSAN=y:
> 
> I was also looking into this issue recently, because people start
> running into it: https://github.com/google/kmsan/issues/89
> 
> I have a solution that redefines __underlying_memXXX to __msan_memXXX
> under __SANITIZE_MEMORY__ in fortify-string.h and skips `#define
> memXXX __msan_memXXX` in string_64.h, making KMSAN kinda work with
> FORTIFY_SOURCE.

Oh good!

> Dunno if that's necessary though: KMSAN is a debugging tool anyway,
> and supporting it in fortify-string.h sounds excessive.

I'd much prefer letting them still work together.