Re: [PATCH v4 3/5] stack: Optionally randomize kernel stack offset each syscall

[Kees Cook]

On Mon, Jun 22, 2020 at 11:42:29PM +0200, Jann Horn wrote:
> No, at least on x86-64 and x86 Linux overrides the normal ABI. From
> arch/x86/Makefile:

Ah! Thanks for the pointer.

> 
> # For gcc stack alignment is specified with -mpreferred-stack-boundary,
> # clang has the option -mstack-alignment for that purpose.
> ifneq ($(call cc-option, -mpreferred-stack-boundary=4),)
>       cc_stack_align4 := -mpreferred-stack-boundary=2
>       cc_stack_align8 := -mpreferred-stack-boundary=3
> else ifneq ($(call cc-option, -mstack-alignment=16),)
>       cc_stack_align4 := -mstack-alignment=4
>       cc_stack_align8 := -mstack-alignment=8
> endif
> [...]
> ifeq ($(CONFIG_X86_32),y)
> [...]
>         # Align the stack to the register width instead of using the default
>         # alignment of 16 bytes. This reduces stack usage and the number of
>         # alignment instructions.
>         KBUILD_CFLAGS += $(call cc-option,$(cc_stack_align4))
> [...]
> else
> [...]
>         # By default gcc and clang use a stack alignment of 16 bytes for x86.
>         # However the standard kernel entry on x86-64 leaves the stack on an
>         # 8-byte boundary. If the compiler isn't informed about the actual
>         # alignment it will generate extra alignment instructions for the
>         # default alignment which keep the stack *mis*aligned.
>         # Furthermore an alignment to the register width reduces stack usage
>         # and the number of alignment instructions.
>         KBUILD_CFLAGS += $(call cc-option,$(cc_stack_align8))
> [...]
> endif

And it seems that only x86 does this. No other architecture specifies
-mpreferred-stack-boundary...

> Normal x86-64 ABI has 16-byte stack alignment; Linux kernel x86-64 ABI
> has 8-byte stack alignment.
> Similarly, the normal Linux 32-bit x86 ABI is 16-byte aligned;
> meanwhile Linux kernel x86 ABI has 4-byte stack alignment.
> 
> This is because userspace code wants the stack to be sufficiently
> aligned for fancy SSE instructions and such; the kernel, on the other
> hand, never uses those in normal code, and cares about stack usage and
> such very much.

This makes it nicer for Clang:


diff --git a/include/linux/randomize_kstack.h b/include/linux/randomize_kstack.h
index 1df0dc52cadc..f7e1f68fb50c 100644
--- a/include/linux/randomize_kstack.h
+++ b/include/linux/randomize_kstack.h
@@ -10,6 +10,14 @@ DECLARE_STATIC_KEY_MAYBE(CONFIG_RANDOMIZE_KSTACK_OFFSET_DEFAULT,
 			 randomize_kstack_offset);
 DECLARE_PER_CPU(u32, kstack_offset);
 
+#ifdef CONFIG_X86_64
+#define ARCH_STACK_ALIGN_MASK	~((1 << 8) - 1)
+#elif defined(CONFIG_X86_32)
+#define ARCH_STACK_ALIGN_MASK	~((1 << 4) - 1)
+#else
+#define ARCH_STACK_ALIGN_MASK	~(0)
+#endif
+
 /*
  * Do not use this anywhere else in the kernel. This is used here because
  * it provides an arch-agnostic way to grow the stack with correct
@@ -23,7 +31,8 @@ void *__builtin_alloca(size_t size);
 	if (static_branch_maybe(CONFIG_RANDOMIZE_KSTACK_OFFSET_DEFAULT,	\
 				&randomize_kstack_offset)) {		\
 		u32 offset = this_cpu_read(kstack_offset);		\
-		u8 *ptr = __builtin_alloca(offset & 0x3FF);		\
+		u8 *ptr = __builtin_alloca(offset & 0x3FF &		\
+					   ARCH_STACK_ALIGN_MASK);	\
 		asm volatile("" : "=m"(*ptr));				\
 	}								\
 } while (0)


But I don't like open-coding the x86-ony stack alignment... it should be
in Kconfig or something, I think?

-- 
Kees Cook