[PATCH] hardening: Default to INIT_STACK_ALL_ZERO if CC_HAS_AUTO_VAR_INIT_ZERO

CC_HAS_AUTO_VAR_INIT_ZERO requires a supported set of compiler options
distinct from those needed by CC_HAS_AUTO_VAR_INIT_PATTERN, Fix up
the Kconfig dependency for INIT_STACK_ALL_ZERO to test for the former
instead of the latter, as these are the options passed by the top-level
Makefile.

Cc: Kees Cook <redacted>
Cc: Nathan Chancellor <nathan@kernel.org>
Cc: Nick Desaulniers <ndesaulniers@google.com>
Cc: Gustavo A. R. Silva <gustavoars@kernel.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Fixes: dcb7c0b9461c ("hardening: Clarify Kconfig text for auto-var-init")
Signed-off-by: Will Deacon <will@kernel.org>
---

I just noticed this while reading the code and I suspect it doesn't really
matter in practice.

 security/Kconfig.hardening | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/security/Kconfig.hardening b/security/Kconfig.hardening
index 90cbaff86e13..341e2fdcba94 100644
--- a/security/Kconfig.hardening
+++ b/security/Kconfig.hardening

@@ -29,7 +29,7 @@ choice
 	prompt "Initialize kernel stack variables at function entry"
 	default GCC_PLUGIN_STRUCTLEAK_BYREF_ALL if COMPILE_TEST && GCC_PLUGINS
 	default INIT_STACK_ALL_PATTERN if COMPILE_TEST && CC_HAS_AUTO_VAR_INIT_PATTERN
-	default INIT_STACK_ALL_ZERO if CC_HAS_AUTO_VAR_INIT_PATTERN
+	default INIT_STACK_ALL_ZERO if CC_HAS_AUTO_VAR_INIT_ZERO
 	default INIT_STACK_NONE
 	help
 	  This option enables initialization of stack variables at

-- 
2.33.0.309.g3052b89438-goog

On 9/14/2021 3:28 AM, Will Deacon wrote:

CC_HAS_AUTO_VAR_INIT_ZERO requires a supported set of compiler options
distinct from those needed by CC_HAS_AUTO_VAR_INIT_PATTERN, Fix up
the Kconfig dependency for INIT_STACK_ALL_ZERO to test for the former
instead of the latter, as these are the options passed by the top-level
Makefile.

Cc: Kees Cook <redacted>
Cc: Nathan Chancellor <nathan@kernel.org>
Cc: Nick Desaulniers <ndesaulniers@google.com>
Cc: Gustavo A. R. Silva <gustavoars@kernel.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Fixes: dcb7c0b9461c ("hardening: Clarify Kconfig text for auto-var-init")
Signed-off-by: Will Deacon <will@kernel.org>

Reviewed-by: Nathan Chancellor <nathan@kernel.org>

One comment below.

While I think this change is correct in and of itself, 
CONFIG_INIT_STACK_ALL_ZERO is broken with GCC 12.x, as 
CONFIG_CC_HAS_AUTO_VAR_INIT_ZERO won't be set even though GCC now 
supports -ftrivial-auto-var-init=zero because GCC does not implement the 
-enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang 
flag for obvious reasons ;) the cc-option call probably needs to be 
adjusted.

Cheers,
Nathan

On Tue, Sep 14, 2021 at 08:58:12AM -0700, Nathan Chancellor wrote:

On 9/14/2021 3:28 AM, Will Deacon wrote:

quoted

CC_HAS_AUTO_VAR_INIT_ZERO requires a supported set of compiler options
distinct from those needed by CC_HAS_AUTO_VAR_INIT_PATTERN, Fix up
the Kconfig dependency for INIT_STACK_ALL_ZERO to test for the former
instead of the latter, as these are the options passed by the top-level
Makefile.

Cc: Kees Cook <redacted>
Cc: Nathan Chancellor <nathan@kernel.org>
Cc: Nick Desaulniers <ndesaulniers@google.com>
Cc: Gustavo A. R. Silva <gustavoars@kernel.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Fixes: dcb7c0b9461c ("hardening: Clarify Kconfig text for auto-var-init")
Signed-off-by: Will Deacon <will@kernel.org>

Reviewed-by: Nathan Chancellor <nathan@kernel.org>

One comment below.

quoted

---

I just noticed this while reading the code and I suspect it doesn't really
matter in practice.

  security/Kconfig.hardening | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/security/Kconfig.hardening b/security/Kconfig.hardening
index 90cbaff86e13..341e2fdcba94 100644
--- a/security/Kconfig.hardening
+++ b/security/Kconfig.hardening

@@ -29,7 +29,7 @@ choice
  	prompt "Initialize kernel stack variables at function entry"
  	default GCC_PLUGIN_STRUCTLEAK_BYREF_ALL if COMPILE_TEST && GCC_PLUGINS
  	default INIT_STACK_ALL_PATTERN if COMPILE_TEST && CC_HAS_AUTO_VAR_INIT_PATTERN
-	default INIT_STACK_ALL_ZERO if CC_HAS_AUTO_VAR_INIT_PATTERN
+	default INIT_STACK_ALL_ZERO if CC_HAS_AUTO_VAR_INIT_ZERO
  	default INIT_STACK_NONE
  	help
  	  This option enables initialization of stack variables at

While I think this change is correct in and of itself,
CONFIG_INIT_STACK_ALL_ZERO is broken with GCC 12.x, as
CONFIG_CC_HAS_AUTO_VAR_INIT_ZERO won't be set even though GCC now supports
-ftrivial-auto-var-init=zero because GCC does not implement the
-enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang
flag for obvious reasons ;) the cc-option call probably needs to be
adjusted.

GCC silently ignores the -enable flag, so things actually work correctly
as-is. But, yes, it makes the command line long and doesn't make sense.
How about we do this instead:

diff --git a/Makefile b/Makefile
index 34a0afc3a8eb..34439deac939 100644
--- a/Makefile
+++ b/Makefile

@@ -831,12 +831,11 @@ endif
 
 # Initialize all stack variables with a zero value.
 ifdef CONFIG_INIT_STACK_ALL_ZERO
-# Future support for zero initialization is still being debated, see
-# https://bugs.llvm.org/show_bug.cgi?id=45497. These flags are subject to being
-# renamed or dropped.
 KBUILD_CFLAGS	+= -ftrivial-auto-var-init=zero
+ifdef CONFIG_CC_HAS_AUTO_VAR_INIT_ZERO_WITH_ENABLE
 KBUILD_CFLAGS	+= -enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang
 endif
+endif
 
 # While VLAs have been removed, GCC produces unreachable stack probes
 # for the randomize_kstack_offset feature. Disable it for all compilers.

diff --git a/security/Kconfig.hardening b/security/Kconfig.hardening
index 90cbaff86e13..beea81df3081 100644
--- a/security/Kconfig.hardening
+++ b/security/Kconfig.hardening

@@ -22,14 +22,22 @@ menu "Memory initialization"
 config CC_HAS_AUTO_VAR_INIT_PATTERN
 	def_bool $(cc-option,-ftrivial-auto-var-init=pattern)
 
+config CC_HAS_AUTO_VAR_INIT_ZERO_WITHOUT_ENABLE
+	def_bool $(cc-option,-ftrivial-auto-var-init=zero)
+
+config CC_HAS_AUTO_VAR_INIT_ZERO_WITH_ENABLE
+	# https://bugs.llvm.org/show_bug.cgi?id=45497
+	def_bool !CC_HAS_AUTO_VAR_INIT_ZERO_WITHOUT_ENABLE && \
+		 $(cc-option,-ftrivial-auto-var-init=zero -enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang)
+
 config CC_HAS_AUTO_VAR_INIT_ZERO
-	def_bool $(cc-option,-ftrivial-auto-var-init=zero -enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang)
+	def_bool CC_HAS_AUTO_VAR_INIT_ZERO_WITHOUT_ENABLE || CC_HAS_AUTO_VAR_INIT_ZERO_WITH_ENABLE
 
 choice
 	prompt "Initialize kernel stack variables at function entry"
 	default GCC_PLUGIN_STRUCTLEAK_BYREF_ALL if COMPILE_TEST && GCC_PLUGINS
 	default INIT_STACK_ALL_PATTERN if COMPILE_TEST && CC_HAS_AUTO_VAR_INIT_PATTERN
-	default INIT_STACK_ALL_ZERO if CC_HAS_AUTO_VAR_INIT_PATTERN
+	default INIT_STACK_ALL_ZERO if CC_HAS_AUTO_VAR_INIT_ZERO
 	default INIT_STACK_NONE
 	help
 	  This option enables initialization of stack variables at

-- 
Kees Cook

`On Tue, Sep 14, 2021 at 10:21 AM Kees Cook [off-list ref] wrote:

On Tue, Sep 14, 2021 at 08:58:12AM -0700, Nathan Chancellor wrote:

quoted

On 9/14/2021 3:28 AM, Will Deacon wrote:

quoted

CC_HAS_AUTO_VAR_INIT_ZERO requires a supported set of compiler options
distinct from those needed by CC_HAS_AUTO_VAR_INIT_PATTERN, Fix up
the Kconfig dependency for INIT_STACK_ALL_ZERO to test for the former
instead of the latter, as these are the options passed by the top-level
Makefile.

Cc: Kees Cook <redacted>
Cc: Nathan Chancellor <nathan@kernel.org>
Cc: Nick Desaulniers <ndesaulniers@google.com>
Cc: Gustavo A. R. Silva <gustavoars@kernel.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Fixes: dcb7c0b9461c ("hardening: Clarify Kconfig text for auto-var-init")
Signed-off-by: Will Deacon <will@kernel.org>

Reviewed-by: Nathan Chancellor <nathan@kernel.org>

One comment below.

quoted

---

I just noticed this while reading the code and I suspect it doesn't really
matter in practice.

  security/Kconfig.hardening | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/security/Kconfig.hardening b/security/Kconfig.hardening
index 90cbaff86e13..341e2fdcba94 100644
--- a/security/Kconfig.hardening
+++ b/security/Kconfig.hardening

@@ -29,7 +29,7 @@ choice
    prompt "Initialize kernel stack variables at function entry"
    default GCC_PLUGIN_STRUCTLEAK_BYREF_ALL if COMPILE_TEST && GCC_PLUGINS
    default INIT_STACK_ALL_PATTERN if COMPILE_TEST && CC_HAS_AUTO_VAR_INIT_PATTERN
-   default INIT_STACK_ALL_ZERO if CC_HAS_AUTO_VAR_INIT_PATTERN
+   default INIT_STACK_ALL_ZERO if CC_HAS_AUTO_VAR_INIT_ZERO
    default INIT_STACK_NONE
    help
      This option enables initialization of stack variables at

While I think this change is correct in and of itself,
CONFIG_INIT_STACK_ALL_ZERO is broken with GCC 12.x, as
CONFIG_CC_HAS_AUTO_VAR_INIT_ZERO won't be set even though GCC now supports
-ftrivial-auto-var-init=zero because GCC does not implement the
-enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang
flag for obvious reasons ;) the cc-option call probably needs to be
adjusted.

GCC silently ignores the -enable flag, so things actually work correctly
as-is.

So then would that mean that CC_HAS_AUTO_VAR_INIT_ZERO_WITH_ENABLE
evaluates to true then, in your patch below?

Rather than create 2 new kconfigs with 1 new invocation of the
compiler via cc-option, how about just adding an `ifdef
CONFIG_CC_IS_CLANG` guard around adding the obnoxious flag to
`KBUILD_CFLAGS` in the top level Makefile?

-- 
Thanks,
~Nick Desaulniers

On Tue, Sep 14, 2021 at 11:53:38AM -0700, Nick Desaulniers wrote:

`On Tue, Sep 14, 2021 at 10:21 AM Kees Cook [off-list ref] wrote:

quoted

On Tue, Sep 14, 2021 at 08:58:12AM -0700, Nathan Chancellor wrote:

quoted

On 9/14/2021 3:28 AM, Will Deacon wrote:

quoted

CC_HAS_AUTO_VAR_INIT_ZERO requires a supported set of compiler options
distinct from those needed by CC_HAS_AUTO_VAR_INIT_PATTERN, Fix up
the Kconfig dependency for INIT_STACK_ALL_ZERO to test for the former
instead of the latter, as these are the options passed by the top-level
Makefile.

Cc: Kees Cook <redacted>
Cc: Nathan Chancellor <nathan@kernel.org>
Cc: Nick Desaulniers <ndesaulniers@google.com>
Cc: Gustavo A. R. Silva <gustavoars@kernel.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Fixes: dcb7c0b9461c ("hardening: Clarify Kconfig text for auto-var-init")
Signed-off-by: Will Deacon <will@kernel.org>

Reviewed-by: Nathan Chancellor <nathan@kernel.org>

One comment below.

quoted

---

I just noticed this while reading the code and I suspect it doesn't really
matter in practice.

  security/Kconfig.hardening | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/security/Kconfig.hardening b/security/Kconfig.hardening
index 90cbaff86e13..341e2fdcba94 100644
--- a/security/Kconfig.hardening
+++ b/security/Kconfig.hardening

@@ -29,7 +29,7 @@ choice
    prompt "Initialize kernel stack variables at function entry"
    default GCC_PLUGIN_STRUCTLEAK_BYREF_ALL if COMPILE_TEST && GCC_PLUGINS
    default INIT_STACK_ALL_PATTERN if COMPILE_TEST && CC_HAS_AUTO_VAR_INIT_PATTERN
-   default INIT_STACK_ALL_ZERO if CC_HAS_AUTO_VAR_INIT_PATTERN
+   default INIT_STACK_ALL_ZERO if CC_HAS_AUTO_VAR_INIT_ZERO
    default INIT_STACK_NONE
    help
      This option enables initialization of stack variables at

While I think this change is correct in and of itself,
CONFIG_INIT_STACK_ALL_ZERO is broken with GCC 12.x, as
CONFIG_CC_HAS_AUTO_VAR_INIT_ZERO won't be set even though GCC now supports
-ftrivial-auto-var-init=zero because GCC does not implement the
-enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang
flag for obvious reasons ;) the cc-option call probably needs to be
adjusted.

GCC silently ignores the -enable flag, so things actually work correctly
as-is.

So then would that mean that CC_HAS_AUTO_VAR_INIT_ZERO_WITH_ENABLE
evaluates to true then, in your patch below?

No, I exclude it based on the results from
CC_HAS_AUTO_VAR_INIT_ZERO_WITHOUT_ENABLE.

here:

quoted

+       def_bool !CC_HAS_AUTO_VAR_INIT_ZERO_WITHOUT_ENABLE && \

Rather than create 2 new kconfigs with 1 new invocation of the
compiler via cc-option, how about just adding an `ifdef
CONFIG_CC_IS_CLANG` guard around adding the obnoxious flag to
`KBUILD_CFLAGS` in the top level Makefile?

That is a bit more sensible, yes. :) Let me try that...

-- 
Kees Cook

On Tue, Sep 14, 2021 at 11:53:38AM -0700, Nick Desaulniers wrote:

Rather than create 2 new kconfigs with 1 new invocation of the
compiler via cc-option, how about just adding an `ifdef
CONFIG_CC_IS_CLANG` guard around adding the obnoxious flag to
`KBUILD_CFLAGS` in the top level Makefile?

v2:

diff --git a/Makefile b/Makefile
index 34a0afc3a8eb..72d165ffabdb 100644
--- a/Makefile
+++ b/Makefile

@@ -831,12 +831,12 @@ endif
 
 # Initialize all stack variables with a zero value.
 ifdef CONFIG_INIT_STACK_ALL_ZERO
-# Future support for zero initialization is still being debated, see
-# https://bugs.llvm.org/show_bug.cgi?id=45497. These flags are subject to being
-# renamed or dropped.
 KBUILD_CFLAGS	+= -ftrivial-auto-var-init=zero
+ifdef CONFIG_CC_IS_CLANG
+# https://bugs.llvm.org/show_bug.cgi?id=45497
 KBUILD_CFLAGS	+= -enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang
 endif
+endif
 
 # While VLAs have been removed, GCC produces unreachable stack probes
 # for the randomize_kstack_offset feature. Disable it for all compilers.

diff --git a/security/Kconfig.hardening b/security/Kconfig.hardening
index 90cbaff86e13..ded17b8abce2 100644
--- a/security/Kconfig.hardening
+++ b/security/Kconfig.hardening

@@ -23,13 +23,16 @@ config CC_HAS_AUTO_VAR_INIT_PATTERN
 	def_bool $(cc-option,-ftrivial-auto-var-init=pattern)
 
 config CC_HAS_AUTO_VAR_INIT_ZERO
+	# GCC ignores the -enable flag, so we can test for the feature with
+	# a single invocation using the flag, but drop it as appropriate in
+	# the Makefile, depending on the presence of Clang.
 	def_bool $(cc-option,-ftrivial-auto-var-init=zero -enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang)
 
 choice
 	prompt "Initialize kernel stack variables at function entry"
 	default GCC_PLUGIN_STRUCTLEAK_BYREF_ALL if COMPILE_TEST && GCC_PLUGINS
 	default INIT_STACK_ALL_PATTERN if COMPILE_TEST && CC_HAS_AUTO_VAR_INIT_PATTERN
-	default INIT_STACK_ALL_ZERO if CC_HAS_AUTO_VAR_INIT_PATTERN
+	default INIT_STACK_ALL_ZERO if CC_HAS_AUTO_VAR_INIT_ZERO
 	default INIT_STACK_NONE
 	help
 	  This option enables initialization of stack variables at

-- 
Kees Cook

On Tue, Sep 14, 2021 at 12:14 PM Kees Cook [off-list ref] wrote:

On Tue, Sep 14, 2021 at 11:53:38AM -0700, Nick Desaulniers wrote:

quoted

Rather than create 2 new kconfigs with 1 new invocation of the
compiler via cc-option, how about just adding an `ifdef
CONFIG_CC_IS_CLANG` guard around adding the obnoxious flag to
`KBUILD_CFLAGS` in the top level Makefile?

v2:

LGTM
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>

I think the formal patch should have Will's Suggested-by on it.

-- 
Thanks,
~Nick Desaulniers

On Tue, Sep 14, 2021 at 12:14:47PM -0700, Kees Cook wrote:

On Tue, Sep 14, 2021 at 11:53:38AM -0700, Nick Desaulniers wrote:

quoted

Rather than create 2 new kconfigs with 1 new invocation of the
compiler via cc-option, how about just adding an `ifdef
CONFIG_CC_IS_CLANG` guard around adding the obnoxious flag to
`KBUILD_CFLAGS` in the top level Makefile?

v2:

Thanks for clarifying! (since I did not actually test before making the
comment...)

Reviewed-by: Nathan Chancellor <nathan@kernel.org>