security: allow using Clang's zero initialization for stack variables
RFC PATCH v2 1/1 security: allow using Clang's zero initialization for stack variables
In addition to -ftrivial-auto-var-init=pattern (used by CONFIG_INIT_STACK_ALL now) Clang also supports zero initialization for locals enabled by -ftrivial-auto-var-init=zero.
The future of this flag is still being debated, see
https://bugs.llvm.org/show_bug.cgi?id=45497
Right now it is guarded by another flag, -enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang, which means it may not be supported by future Clang releases. Another possible resolution is that -ftrivial-auto-var-init=zero will persist (as certain users have already started depending on it), but the name of the guard flag will change.
In the meantime, zero initialization has proven itself as a good production mitigation measure against uninitialized locals. Unlike pattern initialization, which has a higher chance of triggering existing bugs, zero initialization provides safe defaults for strings, pointers, indexes, and sizes. On the other hand, pattern initialization remains safer for return values.
Performance-wise, the difference between pattern and zero initialization is usually negligible, although the generated code for zero initialization is more compact.
This patch renames CONFIG_INIT_STACK_ALL to CONFIG_INIT_STACK_ALL_PATTERN and introduces another config option, CONFIG_INIT_STACK_ALL_ZERO, that enables zero initialization for locals if the corresponding flags are supported by Clang.
--
v2:
- as suggested by Kees Cook, make CONFIG_INIT_STACK_ALL_PATTERN and CONFIG_INIT_STACK_ALL_ZERO separate options.
--- Makefile | 12 ++++++++++-- init/main.c | 6 ++++-- security/Kconfig.hardening | 29 +++++++++++++++++++++++++---- 3 files changed, 39 insertions(+), 8 deletions(-)
Makefile +10 -2
@@ -802,11 +802,19 @@ KBUILD_CFLAGS += -fomit-frame-pointer endif endif -# Initialize all stack variables with a pattern, if desired. -ifdef CONFIG_INIT_STACK_ALL +# Initialize all stack variables with a 0xAA pattern. +ifdef CONFIG_INIT_STACK_ALL_PATTERN KBUILD_CFLAGS += -ftrivial-auto-var-init=pattern endif +# Initialize all stack variables with a zero pattern. +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 -enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang +endif + DEBUG_CFLAGS := $(call cc-option, -fno-var-tracking-assignments) ifdef CONFIG_DEBUG_INFO
init/main.c +4 -2
@@ -779,8 +779,10 @@ static void __init report_meminit(void) {const char *stack; -if (IS_ENABLED (CONFIG_INIT_STACK_ALL )) - stack ="all" ; +if (IS_ENABLED (CONFIG_INIT_STACK_ALL_PATTERN )) + stack ="all (pattern)" ; +else if (IS_ENABLED (CONFIG_INIT_STACK_ALL_ZERO )) + stack ="all (zero)" ;else if (IS_ENABLED (CONFIG_GCC_PLUGIN_STRUCTLEAK_BYREF_ALL )) stack ="byref_all" ;else if (IS_ENABLED (CONFIG_GCC_PLUGIN_STRUCTLEAK_BYREF ))
security/Kconfig.hardening +25 -4
@@ -19,13 +19,16 @@ config GCC_PLUGIN_STRUCTLEAK menu "Memory initialization" -config CC_HAS_AUTO_VAR_INIT +config CC_HAS_AUTO_VAR_INIT_PATTERN def_bool $(cc-option,-ftrivial-auto-var-init=pattern) +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) + choice prompt "Initialize kernel stack variables at function entry" default GCC_PLUGIN_STRUCTLEAK_BYREF_ALL if COMPILE_TEST && GCC_PLUGINS - default INIT_STACK_ALL if COMPILE_TEST && CC_HAS_AUTO_VAR_INIT + default INIT_STACK_ALL_PATTERN if COMPILE_TEST && CC_HAS_AUTO_VAR_INIT_PATTERN default INIT_STACK_NONE help This option enables initialization of stack variables at
@@ -88,15 +91,33 @@ choice of uninitialized stack variable exploits and information exposures. - config INIT_STACK_ALL + config INIT_STACK_ALL_PATTERN bool "0xAA-init everything on the stack (strongest)" - depends on CC_HAS_AUTO_VAR_INIT + depends on CC_HAS_AUTO_VAR_INIT_PATTERN help Initializes everything on the stack with a 0xAA pattern. This is intended to eliminate all classes of uninitialized stack variable exploits and information exposures, even variables that were warned to have been left uninitialized. + Pattern initialization is known to provoke many existing bugs + related to uninitialized locals, e.g. pointers receive + non-NULL values, buffer sizes and indices are very big. + + config INIT_STACK_ALL_ZERO + bool "zero-init everything on the stack (strongest and safest)" + depends on CC_HAS_AUTO_VAR_INIT_ZERO + help + Initializes everything on the stack with a zero + pattern. This is intended to eliminate all classes + of uninitialized stack variable exploits and information + exposures, even variables that were warned to have been + left uninitialized. + Zero initialization provides safe defaults for strings, + pointers, indices and sizes, and is therefore more suitable as + a security mitigation measure. + The corresponding flag isn't officially supported by Clang and + may sooner or later go away or get renamed. endchoice
Signature · 1 line
2.27.0.290.gba653c62da-goog
Quoted 22 lines
In addition to -ftrivial-auto-var-init=pattern (used by
CONFIG_INIT_STACK_ALLnow) Clang also supports zero initialization for locals enabled by -ftrivial-auto-var-init=zero.
The future of this flag is still being debated, see
https://bugs.llvm.org/show_bug.cgi?id=45497
Right now it is guarded by another flag, -enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang, which means it may not be supported by future Clang releases. Another possible resolution is that -ftrivial-auto-var-init=zero will persist (as certain users have already started depending on it), but the name of the guard flag will change.In the meantime, zero initialization has proven itself as a good production mitigation measure against uninitialized locals. Unlike pattern initialization, which has a higher chance of triggering existing bugs, zero initialization provides safe defaults for strings, pointers, indexes, and sizes. On the other hand, pattern initialization remains safer for return values.
Performance-wise, the difference between pattern and zero initialization is usually negligible, although the generated code for zero initialization is more compact.
Quoted 4 lines
This patch renames
CONFIG_INIT_STACK_ALLtoCONFIG_INIT_STACK_ALL_PATTERNand introduces another config option,CONFIG_INIT_STACK_ALL_ZERO, that enables zero initialization for locals if the corresponding flags are supported by Clang.
I'm a great fan of zero initialization as opposed to pattern. I don't understand why clang is refusing to make this a supported option.
Anyway:
Quoted 27 lines · On Tue, Jun 16, 2020 at 10:34:35AM +0200, [email protected] wrote:
In addition to -ftrivial-auto-var-init=pattern (used by
CONFIG_INIT_STACK_ALLnow) Clang also supports zero initialization for locals enabled by -ftrivial-auto-var-init=zero.
The future of this flag is still being debated, see
https://bugs.llvm.org/show_bug.cgi?id=45497
Right now it is guarded by another flag, -enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang, which means it may not be supported by future Clang releases. Another possible resolution is that -ftrivial-auto-var-init=zero will persist (as certain users have already started depending on it), but the name of the guard flag will change.In the meantime, zero initialization has proven itself as a good production mitigation measure against uninitialized locals. Unlike pattern initialization, which has a higher chance of triggering existing bugs, zero initialization provides safe defaults for strings, pointers, indexes, and sizes. On the other hand, pattern initialization remains safer for return values.
Performance-wise, the difference between pattern and zero initialization is usually negligible, although the generated code for zero initialization is more compact.This patch renames
CONFIG_INIT_STACK_ALLtoCONFIG_INIT_STACK_ALL_PATTERNand introduces another config option,CONFIG_INIT_STACK_ALL_ZERO, that enables zero initialization for locals if the corresponding flags are supported by Clang.
Quoted 4 lines
Cc Kees Cook <[email protected]>Cc Nick Desaulniers <[email protected]>Cc Greg Kroah-Hartman <[email protected]>Signed-off-by Alexander Potapenko <[email protected]>
Thanks! I've applied this to my for-next/kspp tree (with a few small tweaks).
Quoted 1 line
--
^^ note, this separator should be "---" for diff tools to do the right thing, etc.
Quoted 114 lines
v2:
- as suggested by Kees Cook, makeCONFIG_INIT_STACK_ALL_PATTERNandCONFIG_INIT_STACK_ALL_ZEROseparate options.--- Makefile | 12 ++++++++++-- init/main.c | 6 ++++-- security/Kconfig.hardening | 29 +++++++++++++++++++++++++---- 3 files changed, 39 insertions(+), 8 deletions(-)Makefile +10
-2@@ -802,11 +802,19 @@ KBUILD_CFLAGS += -fomit-frame-pointer endif endif -# Initialize all stack variables with a pattern, if desired. -ifdef CONFIG_INIT_STACK_ALL +# Initialize all stack variables with a 0xAA pattern. +ifdef CONFIG_INIT_STACK_ALL_PATTERN KBUILD_CFLAGS += -ftrivial-auto-var-init=pattern endif +# Initialize all stack variables with a zero pattern. +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 -enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang +endif + DEBUG_CFLAGS := $(call cc-option, -fno-var-tracking-assignments) ifdef CONFIG_DEBUG_INFOinit/main.c +4
-2@@ -779,8 +779,10 @@ static void __init report_meminit(void) {const char *stack; -if (IS_ENABLED (CONFIG_INIT_STACK_ALL )) - stack ="all" ; +if (IS_ENABLED (CONFIG_INIT_STACK_ALL_PATTERN )) + stack ="all (pattern)" ; +else if (IS_ENABLED (CONFIG_INIT_STACK_ALL_ZERO )) + stack ="all (zero)" ;else if (IS_ENABLED (CONFIG_GCC_PLUGIN_STRUCTLEAK_BYREF_ALL )) stack ="byref_all" ;else if (IS_ENABLED (CONFIG_GCC_PLUGIN_STRUCTLEAK_BYREF ))security/Kconfig.hardening +25
-4@@ -19,13 +19,16 @@ config GCC_PLUGIN_STRUCTLEAK menu "Memory initialization" -config CC_HAS_AUTO_VAR_INIT +config CC_HAS_AUTO_VAR_INIT_PATTERN def_bool $(cc-option,-ftrivial-auto-var-init=pattern) +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) + choice prompt "Initialize kernel stack variables at function entry" default GCC_PLUGIN_STRUCTLEAK_BYREF_ALL if COMPILE_TEST && GCC_PLUGINS - default INIT_STACK_ALL if COMPILE_TEST && CC_HAS_AUTO_VAR_INIT + default INIT_STACK_ALL_PATTERN if COMPILE_TEST && CC_HAS_AUTO_VAR_INIT_PATTERN default INIT_STACK_NONE help This option enables initialization of stack variables at@@ -88,15 +91,33 @@ choice of uninitialized stack variable exploits and information exposures. - config INIT_STACK_ALL + config INIT_STACK_ALL_PATTERN bool "0xAA-init everything on the stack (strongest)" - depends on CC_HAS_AUTO_VAR_INIT + depends on CC_HAS_AUTO_VAR_INIT_PATTERN help Initializes everything on the stack with a 0xAA pattern. This is intended to eliminate all classes of uninitialized stack variable exploits and information exposures, even variables that were warned to have been left uninitialized. + Pattern initialization is known to provoke many existing bugs + related to uninitialized locals, e.g. pointers receive + non-NULL values, buffer sizes and indices are very big. + + config INIT_STACK_ALL_ZERO + bool "zero-init everything on the stack (strongest and safest)" + depends on CC_HAS_AUTO_VAR_INIT_ZERO + help + Initializes everything on the stack with a zero + pattern. This is intended to eliminate all classes + of uninitialized stack variable exploits and information + exposures, even variables that were warned to have been + left uninitialized. + Zero initialization provides safe defaults for strings, + pointers, indices and sizes, and is therefore more suitable as + a security mitigation measure. + The corresponding flag isn't officially supported by Clang and + may sooner or later go away or get renamed. endchoiceSignature · 1 line
2.27.0.290.gba653c62da-goog
Signature · 1 line
Kees Cook
Quoted 27 lines · On Tue, Jun 16, 2020 at 10:34:35AM +0200, [email protected] wrote:
In addition to -ftrivial-auto-var-init=pattern (used by
CONFIG_INIT_STACK_ALLnow) Clang also supports zero initialization for locals enabled by -ftrivial-auto-var-init=zero.
The future of this flag is still being debated, see
https://bugs.llvm.org/show_bug.cgi?id=45497
Right now it is guarded by another flag, -enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang, which means it may not be supported by future Clang releases. Another possible resolution is that -ftrivial-auto-var-init=zero will persist (as certain users have already started depending on it), but the name of the guard flag will change.In the meantime, zero initialization has proven itself as a good production mitigation measure against uninitialized locals. Unlike pattern initialization, which has a higher chance of triggering existing bugs, zero initialization provides safe defaults for strings, pointers, indexes, and sizes. On the other hand, pattern initialization remains safer for return values.
Performance-wise, the difference between pattern and zero initialization is usually negligible, although the generated code for zero initialization is more compact.This patch renames
CONFIG_INIT_STACK_ALLtoCONFIG_INIT_STACK_ALL_PATTERNand introduces another config option,CONFIG_INIT_STACK_ALL_ZERO, that enables zero initialization for locals if the corresponding flags are supported by Clang.
Quoted 4 lines
Cc Kees Cook <[email protected]>Cc Nick Desaulniers <[email protected]>Cc Greg Kroah-Hartman <[email protected]>Signed-off-by Alexander Potapenko <[email protected]>
Nit, your From: line of your email does not match your signed-off-by:
line :(
In the future, you should fix that up so that maintainers don't have to do it for you...
Makefile KBUILD_CFLAGS += -fomit-frame-pointer · line 816
Quoted 7 lines
Makefile +7
-0+# Initialize all stack variables with a zero pattern. +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 -enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang +endif
Gotta love the name...
Anyway, if this is enabled, and clang changes the flag or drops it, does the build suddenly break?
And does gcc have something like this as well, or does that have to come in a compiler plugin?
thanks,
greg k-h
Quoted 2 lines
Nit, your From: line of your email does not match your signed-off-by:
line :(
This is *probably* a matter of configuring git:
git config --global user.name "Alexander Potapenko"
git config --global user.email "[email protected]"
git config --global sendemail.from "Alexander Potapenko <[email protected]>"
Quoted 1 line
Gotta love the name...
I've just read through a long discussion thread on clang dev (got there from this cl's llvm link)...
There's a lot of interesting info in there. But ultimately it seems to point out tons of projects already use this or want to use it. And there's security and stability benefits for production builds, while pattern mode can be used for debug builds.
Quoted 2 lines
Anyway, if this is enabled, and clang changes the flag or drops it, does the build suddenly break?
(my understanding of the patch is that the option does compiler testing before it becomes available...
in at least some of our build systems built around kbuild the option going away would then cause a build error due to its lack in the final .config)
Quoted 2 lines
And does gcc have something like this as well, or does that have to come in a compiler plugin?
Quoted 3 lines · Maciej Żenczykowski
Quoted 3 lines
Nit, your From: line of your email does not match your signed-off-by:
line :(
That's hard to notice while being on the sending end, thanks for pointing that out!
I'll look into this.
Quoted 4 lines
This is *probably* a matter of configuring git:
git config --global user.name "Alexander Potapenko"
git config --global user.email "[email protected]"
git config --global sendemail.from "Alexander Potapenko <[email protected]>"
I do have all these options set, although I don't have them quoted, not sure if it's necessary.
Moreover, when I do git send-email --dry-run the From: line looks correct.
Quoted 4 lines
Makefile KBUILD_CFLAGS += -fomit-frame-pointer · line 816
Quoted 2 lines
Makefile +2
-0+KBUILD_CFLAGS += -ftrivial-auto-var-init=zero -enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang +endifGotta love the name...
This is basically the reason why we've been hesitating to add it to the kernel from the very beginning.
Quoted 2 lines
Anyway, if this is enabled, and clang changes the flag or drops it, does the build suddenly break?
My original intention (see v1 of this patch) was to make zero-initialization a secondary option of INIT_STACK_ALL, so that nothing changes for the existing users.
But I agree with Kees that these options should be made distinct, as people may want to use them for different purposes (think debug vs. release builds).
We could make INIT_STACK_ALL_ZERO fall back to INIT_STACK_ALL_PATTERN if the compiler flag goes away - does this make sense?
Quoted 2 lines
And does gcc have something like this as well, or does that have to come in a compiler plugin?
Kees mentioned someone's plans to implement that in GCC, but I don't think they have done it already.
Quoted 2 lines
We could make
INIT_STACK_ALL_ZEROfall back toINIT_STACK_ALL_PATTERNif the compiler flag goes away - does this make sense?
No, I'm pretty sure failing to build, or at least not setting anything is better.
AFAIK pattern actually introduces new bugs that aren't visible at all with neither of these flags set.
(because in practice the default no flag behaviour seems to zero some stuff [probably padding] that it doesn't with pattern)
Quoted 11 lines · Alexander Potapenko
Quoted 4 lines
Makefile KBUILD_CFLAGS += -fomit-frame-pointer · line 816
Quoted 2 lines
Makefile +2
-0+KBUILD_CFLAGS += -ftrivial-auto-var-init=zero -enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang +endifGotta love the name...
This is basically the reason why we've been hesitating to add it to the kernel from the very beginning.
Quoted 2 lines
Anyway, if this is enabled, and clang changes the flag or drops it, does the build suddenly break?
Quoted 6 lines
My original intention (see v1 of this patch) was to make zero-initialization a secondary option of
INIT_STACK_ALL, so that nothing changes for the existing users.
But I agree with Kees that these options should be made distinct, as people may want to use them for different purposes (think debug vs. release builds).
Yeah, and if the flag changes again, we can adapt. But at this point, it's getting used downstream, so we need to land the config in the kernel.
Quoted 2 lines
We could make
INIT_STACK_ALL_ZEROfall back toINIT_STACK_ALL_PATTERNif the compiler flag goes away - does this make sense?
I don't like this idea -- I'm very hesitant to make security options do something different than what they document. It means the end user can't reason about how their kernel is built when looking at their CONFIGs.
Quoted 5 lines
Quoted 2 lines
And does gcc have something like this as well, or does that have to come in a compiler plugin?
Kees mentioned someone's plans to implement that in GCC, but I don't think they have done it already.
I've had some GCC folks reach about about these features, but I haven't seen any public discussion yet.
Signature · 1 line
Kees Cook