Re: [PATCH] kunit/stackinit: Use fill byte different from Clang i386 pattern

[Justin Stitt]

Hi,

On Tue, Mar 04, 2025 at 02:56:11PM -0800, Kees Cook wrote:
> The byte initialization values used with -ftrivial-auto-var-init=pattern
> (CONFIG_INIT_STACK_ALL_PATTERN=y) depends on the compiler, architecture,
> and byte position relative to struct member types. On i386 with Clang,
> this includes the 0xFF value, which means it looks like nothing changes
> between the leaf byte filling pass and the expected "stack wiping"
> pass of the stackinit test.
> 
> Use the byte fill value of 0x99 instead, fixing the test for i386 Clang
> builds.
> 
> Reported-by: ernsteiswuerfel
> Closes: https://github.com/ClangBuiltLinux/linux/issues/2071
> Fixes: 8c30d32b1a32 ("lib/test_stackinit: Handle Clang auto-initialization pattern")
> Signed-off-by: Kees Cook <kees@xxxxxxxxxx>
> ---
> Cc: Nathan Chancellor <nathan@xxxxxxxxxx>
> Cc: Bill Wendling <morbo@xxxxxxxxxx>
> Cc: Justin Stitt <justinstitt@xxxxxxxxxx>
> Cc: llvm@xxxxxxxxxxxxxxx
> ---
>  lib/tests/stackinit_kunit.c | 30 ++++++++++++++++++++----------

Ah, couldn't find this file at first. Depends on [1].

>  1 file changed, 20 insertions(+), 10 deletions(-)
> 
> diff --git a/lib/tests/stackinit_kunit.c b/lib/tests/stackinit_kunit.c
> index 135322592faf..63aa78e6f5c1 100644
> --- a/lib/tests/stackinit_kunit.c
> +++ b/lib/tests/stackinit_kunit.c
> @@ -184,6 +184,15 @@ static bool stackinit_range_contains(char *haystack_start, size_t haystack_size,
>  #define INIT_UNION_assigned_copy(var_type)		\
>  	INIT_STRUCT_assigned_copy(var_type)
>  
> +/*
> + * The "did we actually fill the stack?" check value needs
> + * to be neither 0 nor any of the "pattern" bytes. The
> + * pattern bytes are compiler, architecture, and type based,
> + * so we have to pick a value that never appears for those
> + * combinations. Use 0x99 which is not 0xFF, 0xFE, nor 0xAA.
> + */
> +#define FILL_BYTE	0x99
> +
>  /*
>   * @name: unique string name for the test
>   * @var_type: type to be tested for zeroing initialization
> @@ -206,12 +215,12 @@ static noinline void test_ ## name (struct kunit *test)		\
>  	ZERO_CLONE_ ## which(zero);				\
>  	/* Clear entire check buffer for 0xFF overlap test. */	\
>  	memset(check_buf, 0x00, sizeof(check_buf));		\
> -	/* Fill stack with 0xFF. */				\
> +	/* Fill stack with FILL_BYTE. */			\
>  	ignored = leaf_ ##name((unsigned long)&ignored, 1,	\
>  				FETCH_ARG_ ## which(zero));	\
> -	/* Verify all bytes overwritten with 0xFF. */		\
> +	/* Verify all bytes overwritten with FILL_BYTE. */	\
>  	for (sum = 0, i = 0; i < target_size; i++)		\
> -		sum += (check_buf[i] != 0xFF);			\
> +		sum += (check_buf[i] != FILL_BYTE);		\
>  	/* Clear entire check buffer for later bit tests. */	\
>  	memset(check_buf, 0x00, sizeof(check_buf));		\
>  	/* Extract stack-defined variable contents. */		\
> @@ -222,7 +231,8 @@ static noinline void test_ ## name (struct kunit *test)		\
>  	 * possible between the two leaf function calls.	\
>  	 */							\
>  	KUNIT_ASSERT_EQ_MSG(test, sum, 0,			\
> -			    "leaf fill was not 0xFF!?\n");	\
> +			    "leaf fill was not 0x%02X!?\n",	\
> +			    FILL_BYTE);				\

  	KUNIT_ASSERT_EQ_MSG(test, sum, 0,			\
 -			    "leaf fill was not 0xFF!?\n");	\
 +			    "leaf fill was not 0x%02X!??!?!?!?!?!?!?!\n",	\
 +			    FILL_BYTE);				\

>  								\
>  	/* Validate that compiler lined up fill and target. */	\
>  	KUNIT_ASSERT_TRUE_MSG(test,				\
> @@ -234,9 +244,9 @@ static noinline void test_ ## name (struct kunit *test)		\
>  		(int)((ssize_t)(uintptr_t)fill_start -		\
>  		      (ssize_t)(uintptr_t)target_start));	\
>  								\
> -	/* Look for any bytes still 0xFF in check region. */	\
> +	/* Validate check region has no FILL_BYTE bytes. */	\
>  	for (sum = 0, i = 0; i < target_size; i++)		\
> -		sum += (check_buf[i] == 0xFF);			\
> +		sum += (check_buf[i] == FILL_BYTE);		\
>  								\
>  	if (sum != 0 && xfail)					\
>  		kunit_skip(test,				\
> @@ -271,12 +281,12 @@ static noinline int leaf_ ## name(unsigned long sp, bool fill,	\
>  	 * stack frame of SOME kind...				\
>  	 */							\
>  	memset(buf, (char)(sp & 0xff), sizeof(buf));		\
> -	/* Fill variable with 0xFF. */				\
> +	/* Fill variable with FILL_BYTE. */			\
>  	if (fill) {						\
>  		fill_start = &var;				\
>  		fill_size = sizeof(var);			\
>  		memset(fill_start,				\
> -		       (char)((sp & 0xff) | forced_mask),	\
> +		       FILL_BYTE & forced_mask,			\
>  		       fill_size);				\
>  	}							\
>  								\
> @@ -469,7 +479,7 @@ static int noinline __leaf_switch_none(int path, bool fill)
>  			fill_start = &var;
>  			fill_size = sizeof(var);
>  
> -			memset(fill_start, forced_mask | 0x55, fill_size);
> +			memset(fill_start, (forced_mask | 0x55) & FILL_BYTE, fill_size);
>  		}
>  		memcpy(check_buf, target_start, target_size);
>  		break;
> @@ -480,7 +490,7 @@ static int noinline __leaf_switch_none(int path, bool fill)
>  			fill_start = &var;
>  			fill_size = sizeof(var);
>  
> -			memset(fill_start, forced_mask | 0xaa, fill_size);
> +			memset(fill_start, (forced_mask | 0xaa) & FILL_BYTE, fill_size);
>  		}
>  		memcpy(check_buf, target_start, target_size);
>  		break;
> -- 
> 2.34.1
> 
>

I don't understand the stack init pattern tendencies enough to give a RB
but I looked at your patch and it seems to fully replace the fill test
values from 0xff to 0x99, so if 0x99 is OK then OK.

[1]: https://lore.kernel.org/all/20250211003136.2860503-3-kees@xxxxxxxxxx/

Justin