Skip to contentSkip to conversation

compiler_types: Introduce the Clang __preserve_most function attribute

PATCH Superseded Marco Elver 4 participants 12 messages Aug 8, 2023 – Aug 11, 2023

Subsystems (5)

v3 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

PATCH v3 1/3 compiler_types: Introduce the Clang __preserve_most function attribute

[1]: "On X86-64 and AArch64 targets, this attribute changes the calling convention of a function. The preserve_most calling convention attempts to make the code in the caller as unintrusive as possible. This convention behaves identically to the C calling convention on how arguments and return values are passed, but it uses a different set of caller/callee-saved registers. This alleviates the burden of saving and recovering a large register set before and after the call in the caller. If the arguments are passed in callee-saved registers, then they will be preserved by the callee across the call. This doesn't apply for values returned in callee-saved registers.

* On X86-64 the callee preserves all general purpose registers, except for R11. R11 can be used as a scratch register. Floating-point registers (XMMs/YMMs) are not preserved and need to be saved by the caller.

* On AArch64 the callee preserve all general purpose registers, except x0-X8 and X16-X18."

[1] https://clang.llvm.org/docs/AttributeReference.html#preserve-most

Introduce the attribute to compiler_types.h as __preserve_most.

Use of this attribute results in better code generation for calls to very rarely called functions, such as error-reporting functions, or rarely executed slow paths.

Beware that the attribute conflicts with instrumentation calls inserted on function entry which do not use __preserve_most themselves. Notably, function tracing which assumes the normal C calling convention for the given architecture. Where the attribute is supported, __preserve_most will imply notrace. It is recommended to restrict use of the attribute to functions that should or already disable tracing.

The attribute may be supported by a future GCC version (see
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110899).

---
v3:
* Quote more from LLVM documentation about which registers are callee/caller with preserve_most.
* Code comment to restrict use where tracing is meant to be disabled.

v2:
* Imply notrace, to avoid any conflicts with tracing which is inserted on function entry. See added comments.

---
 include/linux/compiler_types.h | 28 ++++++++++++++++++++++++++++
 1 file changed, 28 insertions(+)
include/linux/compiler_types.h +28 -0
index 547ea1ff806e..c88488715a39 100644
@@ -106,6 +106,34 @@ static inline void __chk_io_ptr(const volatile void __iomem *ptr) { }
 #define __cold
 #endif
 
+/*
+ * On x86-64 and arm64 targets, __preserve_most changes the calling convention
+ * of a function to make the code in the caller as unintrusive as possible. This
+ * convention behaves identically to the C calling convention on how arguments
+ * and return values are passed, but uses a different set of caller- and callee-
+ * saved registers.
+ *
+ * The purpose is to alleviates the burden of saving and recovering a large
+ * register set before and after the call in the caller.  This is beneficial for
+ * rarely taken slow paths, such as error-reporting functions that may be called
+ * from hot paths.
+ *
+ * Note: This may conflict with instrumentation inserted on function entry which
+ * does not use __preserve_most or equivalent convention (if in assembly). Since
+ * function tracing assumes the normal C calling convention, where the attribute
+ * is supported, __preserve_most implies notrace.  It is recommended to restrict
+ * use of the attribute to functions that should or already disable tracing.
+ *
+ * Optional: not supported by gcc.
+ *
+ * clang: https://clang.llvm.org/docs/AttributeReference.html#preserve-most
+ */
+#if __has_attribute(__preserve_most__)
+# define __preserve_most notrace __attribute__((__preserve_most__))
+#else
+# define __preserve_most
+#endif
+
 /* Builtins */
 
 /*
Signature · 1 line
2.41.0.640.ga95def55d0-goog
#1 lore raw

PATCH v3 2/3 list_debug: Introduce inline wrappers for debug checks

Turn the list debug checking functions __list_*_valid() into inline functions that wrap the out-of-line functions. Care is taken to ensure the inline wrappers are always inlined, so that additional compiler instrumentation (such as sanitizers) does not result in redundant outlining.

This change is preparation for performing checks in the inline wrappers.

No functional change intended.

---
v3:
* Rename ___list_*_valid() to __list_*_valid_or_report().
* Some documentation.

---
 arch/arm64/kvm/hyp/nvhe/list_debug.c |  6 ++---
 include/linux/list.h                 | 37 +++++++++++++++++++++++++---
 lib/list_debug.c                     | 11 ++++-----
 3 files changed, 41 insertions(+), 13 deletions(-)
arch/arm64/kvm/hyp/nvhe/list_debug.c +3 -3
index d68abd7ea124..16266a939a4c 100644
@@ -26,8 +26,8 @@ static inline __must_check bool nvhe_check_data_corruption(bool v)
 
 /* The predicates checked here are taken from lib/list_debug.c. */
 
-bool __list_add_valid(struct list_head *new, struct list_head *prev,
-		      struct list_head *next)
+bool __list_add_valid_or_report(struct list_head *new, struct list_head *prev,
+				struct list_head *next)
 {
 	if (NVHE_CHECK_DATA_CORRUPTION(next->prev != prev) ||
 	    NVHE_CHECK_DATA_CORRUPTION(prev->next != next) ||
@@ -37,7 +37,7 @@ bool __list_add_valid(struct list_head *new, struct list_head *prev,
 	return true;
 }
 
-bool __list_del_entry_valid(struct list_head *entry)
+bool __list_del_entry_valid_or_report(struct list_head *entry)
 {
 	struct list_head *prev, *next;
 
include/linux/list.h +33 -4
index f10344dbad4d..130c6a1bb45c 100644
@@ -39,10 +39,39 @@ static inline void INIT_LIST_HEAD(struct list_head *list)
 }
 
 #ifdef CONFIG_DEBUG_LIST
-extern bool __list_add_valid(struct list_head *new,
-			      struct list_head *prev,
-			      struct list_head *next);
-extern bool __list_del_entry_valid(struct list_head *entry);
+/*
+ * Performs the full set of list corruption checks before __list_add().
+ * On list corruption reports a warning, and returns false.
+ */
+extern bool __list_add_valid_or_report(struct list_head *new,
+				       struct list_head *prev,
+				       struct list_head *next);
+
+/*
+ * Performs list corruption checks before __list_add(). Returns false if a
+ * corruption is detected, true otherwise.
+ */
+static __always_inline bool __list_add_valid(struct list_head *new,
+					     struct list_head *prev,
+					     struct list_head *next)
+{
+	return __list_add_valid_or_report(new, prev, next);
+}
+
+/*
+ * Performs the full set of list corruption checks before __list_del_entry().
+ * On list corruption reports a warning, and returns false.
+ */
+extern bool __list_del_entry_valid_or_report(struct list_head *entry);
+
+/*
+ * Performs list corruption checks before __list_del_entry(). Returns false if a
+ * corruption is detected, true otherwise.
+ */
+static __always_inline bool __list_del_entry_valid(struct list_head *entry)
+{
+	return __list_del_entry_valid_or_report(entry);
+}
 #else
 static inline bool __list_add_valid(struct list_head *new,
 				struct list_head *prev,
lib/list_debug.c +5 -6
index d98d43f80958..2def33b1491f 100644
@@ -17,8 +17,8 @@
  * attempt).
  */
 
-bool __list_add_valid(struct list_head *new, struct list_head *prev,
-		      struct list_head *next)
+bool __list_add_valid_or_report(struct list_head *new, struct list_head *prev,
+				struct list_head *next)
 {
 	if (CHECK_DATA_CORRUPTION(prev == NULL,
 			"list_add corruption. prev is NULL.\n") ||
@@ -37,9 +37,9 @@ bool __list_add_valid(struct list_head *new, struct list_head *prev,
 
 	return true;
 }
-EXPORT_SYMBOL(__list_add_valid);
+EXPORT_SYMBOL(__list_add_valid_or_report);
 
-bool __list_del_entry_valid(struct list_head *entry)
+bool __list_del_entry_valid_or_report(struct list_head *entry)
 {
 	struct list_head *prev, *next;
 
@@ -65,6 +65,5 @@ bool __list_del_entry_valid(struct list_head *entry)
 		return false;
 
 	return true;
-
 }
-EXPORT_SYMBOL(__list_del_entry_valid);
+EXPORT_SYMBOL(__list_del_entry_valid_or_report);
Signature · 1 line
2.41.0.640.ga95def55d0-goog
#2 lore raw

PATCH v3 3/3 list_debug: Introduce CONFIG_DEBUG_LIST_MINIMAL

In reply to Marco Elver

Numerous production kernel configs (see [1, 2]) are choosing to enable CONFIG_DEBUG_LIST, which is also being recommended by KSPP for hardened configs [3]. The feature has never been designed with performance in mind, yet common list manipulation is happening across hot paths all over the kernel.

Introduce CONFIG_DEBUG_LIST_MINIMAL, which performs list pointer checking inline, and only upon list corruption delegates to the reporting slow path.

To generate optimal machine code with CONFIG_DEBUG_LIST_MINIMAL:

1. Elide checking for pointer values which upon dereference would result in an immediate access fault -- therefore "minimal" checks. The trade-off is lower-quality error reports.

2. Use the newly introduced __preserve_most function attribute (available with Clang, but not yet with GCC) to minimize the code footprint for calling the reporting slow path. As a result, function size of callers is reduced by avoiding saving registers before calling the rarely called reporting slow path.

Note that all TUs in lib/Makefile already disable function tracing, including list_debug.c, and __preserve_most's implied notrace has no effect in this case.

3. Because the inline checks are a subset of the full set of checks in ___list_*_valid(), always return false if the inline checks failed. This avoids redundant compare and conditional branch right after return from the slow path.

As a side-effect of the checks being inline, if the compiler can prove some condition to always be true, it can completely elide some checks.

Running netperf with CONFIG_DEBUG_LIST_MINIMAL (using a Clang compiler with "preserve_most") shows throughput improvements, in my case of ~7% on average (up to 20-30% on some test cases).

---
v3:
* Rename ___list_*_valid() to __list_*_valid_or_report().
* More comments.

v2:
* Note that lib/Makefile disables function tracing for everything and __preserve_most's implied notrace is a noop here.

---
 arch/arm64/kvm/hyp/nvhe/list_debug.c |  2 +
 include/linux/list.h                 | 64 +++++++++++++++++++++++++---
 lib/Kconfig.debug                    | 15 +++++++
 lib/list_debug.c                     |  2 +
 4 files changed, 77 insertions(+), 6 deletions(-)
arch/arm64/kvm/hyp/nvhe/list_debug.c +2 -0
index 16266a939a4c..46a2d4f2b3c6 100644
@@ -26,6 +26,7 @@ static inline __must_check bool nvhe_check_data_corruption(bool v)
 
 /* The predicates checked here are taken from lib/list_debug.c. */
 
+__list_valid_slowpath
 bool __list_add_valid_or_report(struct list_head *new, struct list_head *prev,
 				struct list_head *next)
 {
@@ -37,6 +38,7 @@ bool __list_add_valid_or_report(struct list_head *new, struct list_head *prev,
 	return true;
 }
 
+__list_valid_slowpath
 bool __list_del_entry_valid_or_report(struct list_head *entry)
 {
 	struct list_head *prev, *next;
include/linux/list.h +58 -6

96 lines not on this page · raw message

lib/Kconfig.debug +15 -0
index fbc89baf7de6..e72cf08af0fa 100644
@@ -1680,6 +1680,21 @@ config DEBUG_LIST
 
 	  If unsure, say N.
 
+config DEBUG_LIST_MINIMAL
+	bool "Minimal linked list debug checks"
+	default !DEBUG_KERNEL
+	depends on DEBUG_LIST
+	help
+	  Only perform the minimal set of checks in the linked-list walking
+	  routines to catch corruptions that are not guaranteed to result in an
+	  immediate access fault.
+
+	  This trades lower quality error reports for improved performance: the
+	  generated code should be more optimal and provide trade-offs that may
+	  better serve safety- and performance- critical environments.
+
+	  If unsure, say Y.
+
 config DEBUG_PLIST
 	bool "Debug priority linked list manipulation"
 	depends on DEBUG_KERNEL
lib/list_debug.c +2 -0
index 2def33b1491f..0ff547910dd0 100644
@@ -17,6 +17,7 @@
  * attempt).
  */
 
+__list_valid_slowpath
 bool __list_add_valid_or_report(struct list_head *new, struct list_head *prev,
 				struct list_head *next)
 {
@@ -39,6 +40,7 @@ bool __list_add_valid_or_report(struct list_head *new, struct list_head *prev,
 }
 EXPORT_SYMBOL(__list_add_valid_or_report);
 
+__list_valid_slowpath
 bool __list_del_entry_valid_or_report(struct list_head *entry)
 {
 	struct list_head *prev, *next;
Signature · 1 line
2.41.0.640.ga95def55d0-goog
#3 lore raw

Re: [PATCH v3 3/3] list_debug: Introduce CONFIG_DEBUG_LIST_MINIMAL

Quoted 9 lines · Marco Elver

Numerous production kernel configs (see [1, 2]) are choosing to enable CONFIG_DEBUG_LIST, which is also being recommended by KSPP for hardened configs [3]. The feature has never been designed with performance in mind, yet common list manipulation is happening across hot paths all over the kernel.

Introduce CONFIG_DEBUG_LIST_MINIMAL, which performs list pointer checking inline, and only upon list corruption delegates to the reporting slow path.

I'd really like to get away from calling this "DEBUG", since it's used more for hardening (CONFIG_LIST_HARDENED?). Will Deacon spent some time making this better a while back, but the series never landed. Do you have a bit of time to look through it?

https://github.com/KSPP/linux/issues/10
https://lore.kernel.org/lkml/[email protected]/

-Kees

Signature · 1 line
Kees Cook
#4 lore raw

Re: [PATCH v3 3/3] list_debug: Introduce CONFIG_DEBUG_LIST_MINIMAL

Quoted 12 lines · Kees Cook
Quoted 9 lines · Marco Elver

Numerous production kernel configs (see [1, 2]) are choosing to enable CONFIG_DEBUG_LIST, which is also being recommended by KSPP for hardened configs [3]. The feature has never been designed with performance in mind, yet common list manipulation is happening across hot paths all over the kernel.

Introduce CONFIG_DEBUG_LIST_MINIMAL, which performs list pointer checking inline, and only upon list corruption delegates to the reporting slow path.

Quoted 7 lines

I'd really like to get away from calling this "DEBUG", since it's used more for hardening (CONFIG_LIST_HARDENED?). Will Deacon spent some time making this better a while back, but the series never landed. Do you have a bit of time to look through it?

https://github.com/KSPP/linux/issues/10
https://lore.kernel.org/lkml/[email protected]/

I'm fine renaming this one. But there are other issues that Will's series solves, which I don't want this series to depend on. We can try to sort them out separately.

The main problem here is that DEBUG_LIST has been designed to be friendly for debugging (incl. checking poison values and NULL). Some kernel devs may still want that, but for production use is pointless and wasteful.

So what I can propose is to introduce CONFIG_LIST_HARDENED that doesn't depend on CONFIG_DEBUG_LIST, but instead selects it, because we still use that code to produce a report.

If there are other list types that have similar debug checks, but where we can optimize performance by eliding some and moving them inline, we can do the same (CONFIG_*_HARDENED). If the checks are already optimized, we could just rename it to CONFIG_*_HARDENED and remove the DEBUG-name.

I'm not a big fan of the 2 modes either, but that's what we get if we want to support the debugging and hardening usecases. Inlining the full set of checks does not work for performance and size.

#5 lore raw

Re: [PATCH v3 3/3] list_debug: Introduce CONFIG_DEBUG_LIST_MINIMAL

Quoted 12 lines · Marco Elver
Quoted 7 lines

I'd really like to get away from calling this "DEBUG", since it's used more for hardening (CONFIG_LIST_HARDENED?). Will Deacon spent some time making this better a while back, but the series never landed. Do you have a bit of time to look through it?

https://github.com/KSPP/linux/issues/10
https://lore.kernel.org/lkml/[email protected]/

I'm fine renaming this one. But there are other issues that Will's series solves, which I don't want this series to depend on. We can try to sort them out separately.

Quoted 8 lines

The main problem here is that DEBUG_LIST has been designed to be friendly for debugging (incl. checking poison values and NULL). Some kernel devs may still want that, but for production use is pointless and wasteful.

So what I can propose is to introduce CONFIG_LIST_HARDENED that doesn't depend on CONFIG_DEBUG_LIST, but instead selects it, because we still use that code to produce a report.

How about the below?

We'll add CONFIG_HARDEN_LIST (in Kconfig.hardening), which is independent of CONFIG_DEBUG_LIST. For the implementation it selects DEBUG_LIST, but irrelevant for users.

This will get us the best of both worlds: a version for hardening that should remain as fast as possible, and one for debugging with better reports.

------ >8 ------

From: Marco Elver <[email protected]>
Date: Thu, 27 Jul 2023 22:19:02 +0200
Subject: [PATCH v4 3/3] list: Introduce CONFIG_HARDEN_LIST

Numerous production kernel configs (see [1, 2]) are choosing to enable CONFIG_DEBUG_LIST, which is also being recommended by KSPP for hardened configs [3]. The motivation behind this is that the option can be used as a security hardening feature (e.g. CVE-2019-2215 and CVE-2019-2025 are mitigated by the option [4]).

The feature has never been designed with performance in mind, yet common list manipulation is happening across hot paths all over the kernel.

Introduce CONFIG_HARDEN_LIST, which performs list pointer checking inline, and only upon list corruption calls the reporting slow path.

To generate optimal machine code with CONFIG_HARDEN_LIST:

1. Elide checking for pointer values which upon dereference would result in an immediate access fault -- therefore "minimal" checks. The trade-off is lower-quality error reports.

2. Use the newly introduced __preserve_most function attribute (available with Clang, but not yet with GCC) to minimize the code footprint for calling the reporting slow path. As a result, function size of callers is reduced by avoiding saving registers before calling the rarely called reporting slow path.

Note that all TUs in lib/Makefile already disable function tracing, including list_debug.c, and __preserve_most's implied notrace has no effect in this case.

3. Because the inline checks are a subset of the full set of checks in __list_*_valid_or_report(), always return false if the inline checks failed. This avoids redundant compare and conditional branch right after return from the slow path.

As a side-effect of the checks being inline, if the compiler can prove some condition to always be true, it can completely elide some checks.

Running netperf with CONFIG_HARDEN_LIST (using a Clang compiler with "preserve_most") shows throughput improvements, in my case of ~7% on average (up to 20-30% on some test cases).

---
v4:
* Rename to CONFIG_HARDEN_LIST, which can independently be selected from CONFIG_DEBUG_LIST.

v3:
* Rename ___list_*_valid() to __list_*_valid_or_report().
* More comments.

v2:
* Note that lib/Makefile disables function tracing for everything and __preserve_most's implied notrace is a noop here.

---
 arch/arm64/kvm/hyp/nvhe/list_debug.c |  2 +
 include/linux/list.h                 | 64 +++++++++++++++++++++++++---
 lib/Kconfig.debug                    | 12 ++++--
 lib/list_debug.c                     |  2 +
 security/Kconfig.hardening           | 14 ++++++
 5 files changed, 84 insertions(+), 10 deletions(-)
arch/arm64/kvm/hyp/nvhe/list_debug.c +2 -0
index 16266a939a4c..46a2d4f2b3c6 100644
@@ -26,6 +26,7 @@ static inline __must_check bool nvhe_check_data_corruption(bool v)
 
 /* The predicates checked here are taken from lib/list_debug.c. */
 
+__list_valid_slowpath
 bool __list_add_valid_or_report(struct list_head *new, struct list_head *prev,
 				struct list_head *next)
 {
@@ -37,6 +38,7 @@ bool __list_add_valid_or_report(struct list_head *new, struct list_head *prev,
 	return true;
 }
 
+__list_valid_slowpath
 bool __list_del_entry_valid_or_report(struct list_head *entry)
 {
 	struct list_head *prev, *next;
include/linux/list.h +58 -6

96 lines not on this page · raw message

lib/Kconfig.debug +8 -4
index fbc89baf7de6..6b0de78fb2da 100644
@@ -1672,11 +1672,15 @@ config HAVE_DEBUG_BUGVERBOSE
 menu "Debug kernel data structures"
 
 config DEBUG_LIST
-	bool "Debug linked list manipulation"
-	depends on DEBUG_KERNEL || BUG_ON_DATA_CORRUPTION
+	bool "Debug linked list manipulation" if !HARDEN_LIST
+	depends on DEBUG_KERNEL || BUG_ON_DATA_CORRUPTION || HARDEN_LIST
 	help
-	  Enable this to turn on extended checks in the linked-list
-	  walking routines.
+	  Enable this to turn on extended checks in the linked-list walking
+	  routines.
+
+	  If you care about performance, you should enable CONFIG_HARDEN_LIST
+	  instead.  This option alone trades better quality error reports for
+	  worse performance, and is more suitable for debugging.
 
 	  If unsure, say N.
 
lib/list_debug.c +2 -0
index 2def33b1491f..0ff547910dd0 100644
@@ -17,6 +17,7 @@
  * attempt).
  */
 
+__list_valid_slowpath
 bool __list_add_valid_or_report(struct list_head *new, struct list_head *prev,
 				struct list_head *next)
 {
@@ -39,6 +40,7 @@ bool __list_add_valid_or_report(struct list_head *new, struct list_head *prev,
 }
 EXPORT_SYMBOL(__list_add_valid_or_report);
 
+__list_valid_slowpath
 bool __list_del_entry_valid_or_report(struct list_head *entry)
 {
 	struct list_head *prev, *next;
security/Kconfig.hardening +14 -0
index 0f295961e773..a8aef895f13d 100644
@@ -279,6 +279,20 @@ config ZERO_CALL_USED_REGS
 
 endmenu
 
+menu "Hardening of kernel data structures"
+
+config HARDEN_LIST
+	bool "Check integrity of linked list manipulation"
+	select DEBUG_LIST
+	help
+	  Minimal integrity checking in the linked-list manipulation routines
+	  to catch memory corruptions that are not guaranteed to result in an
+	  immediate access fault.
+
+	  If unsure, say N.
+
+endmenu
+
 config CC_HAS_RANDSTRUCT
 	def_bool $(cc-option,-frandomize-layout-seed-file=/dev/null)
 	# Randstruct was first added in Clang 15, but it isn't safe to use until
Signature · 1 line
2.41.0.640.ga95def55d0-goog
#6 lore raw

Re: [PATCH v3 3/3] list_debug: Introduce CONFIG_DEBUG_LIST_MINIMAL

Quoted 17 lines · On Wed, 9 Aug 2023 11:57:19 +0200 Marco Elver <[email protected]> wrote:

static __always_inline bool __list_add_valid(struct list_head *new,

include/linux/list.h +12 -1
 					     struct list_head *prev,
 					     struct list_head *next)
 {
-	return __list_add_valid_or_report(new, prev, next);
+	bool ret = true;
+
+	if (IS_ENABLED(CONFIG_HARDEN_LIST)) {
+		/*
+		 * With the hardening version, elide checking if next and prev
+		 * are NULL, since the immediate dereference of them below would
+		 * result in a fault if NULL.
+		 *
+		 * With the reduced set of checks, we can afford to inline the
+		 * checks, which also gives the compiler a chance to elide some
+		 * of them completely if they can be proven at compile-time. If
+		 * one of the pre-conditions does not hold, the slow-path will

include/linux/list.h static inline void INIT_LIST_HEAD(struct list_head *list) · line 90

Quoted 10 lines
include/linux/list.h +9 -0
+		 * show a report which pre-condition failed.
+		 */
+		if (likely(next->prev == prev && prev->next == next && new != prev && new != next))
+			return true;
+		ret = false;
+	}
+
+	ret &= __list_add_valid_or_report(new, prev, next);
+	return ret;
 }

I would actually prefer DEBUG_LIST to select HARDEN_LIST and not the other way around. It logically doesn't make sense that HARDEN_LIST would select DEBUG_LIST. That is, I could by default want HARDEN_LIST always on, but not DEBUG_LIST (because who knows, it may add other features I don't want). But then, I may have stumbled over something and want more info, and enable DEBUG_LIST (while still having HARDEN_LIST) enabled.

I think you are looking at this from an implementation perspective and not the normal developer one.

This would mean the above function should get enabled by CONFIG_HARDEN_LIST (and CONFIG_DEBUG would select CONFIG_HARDEN) and would look more like:

static __always_inline bool __list_add_valid(struct list_head *new,
struct list_head *prev, struct list_head *next)

{
	bool ret = true;

	if (!IS_ENABLED(CONFIG_DEBUG_LIST)) {
		/*
		 * With the hardening version, elide checking if next and prev
		 * are NULL, since the immediate dereference of them below would
		 * result in a fault if NULL.
		 *
		 * With the reduced set of checks, we can afford to inline the
		 * checks, which also gives the compiler a chance to elide some
		 * of them completely if they can be proven at compile-time. If
		 * one of the pre-conditions does not hold, the slow-path will
		 * show a report which pre-condition failed.
		 */
		if (likely(next->prev == prev && prev->next == next && new != prev && new != next))
			return true;
		ret = false;
	}

	ret &= __list_add_valid_or_report(new, prev, next);
	return ret;
}

That is, if DEBUG_LIST is enabled, we always call the __list_add_valid_or_report(), but if only HARDEN_LIST is enabled, then we do the shortcut.

-- Steve

#7 lore raw

Re: [PATCH v3 3/3] list_debug: Introduce CONFIG_DEBUG_LIST_MINIMAL

On Wed, Aug 09, 2023 at 11:30AM -0400, Steven Rostedt wrote:
[...]

Quoted 11 lines

I would actually prefer DEBUG_LIST to select HARDEN_LIST and not the other way around. It logically doesn't make sense that HARDEN_LIST would select DEBUG_LIST. That is, I could by default want HARDEN_LIST always on, but not DEBUG_LIST (because who knows, it may add other features I don't want). But then, I may have stumbled over something and want more info, and enable DEBUG_LIST (while still having HARDEN_LIST) enabled.

I think you are looking at this from an implementation perspective and not the normal developer one.

[...]

Quoted 4 lines

That is, if DEBUG_LIST is enabled, we always call the __list_add_valid_or_report(), but if only HARDEN_LIST is enabled, then we do the shortcut.

Good point - I think this is better. See below tentative v4.

Kees: Does that also look more like what you had in mind?

Thanks,
-- Marco

------ >8 ------

From: Marco Elver <[email protected]>
Date: Thu, 27 Jul 2023 22:19:02 +0200
Subject: [PATCH] list: Introduce CONFIG_HARDEN_LIST

Numerous production kernel configs (see [1, 2]) are choosing to enable CONFIG_DEBUG_LIST, which is also being recommended by KSPP for hardened configs [3]. The motivation behind this is that the option can be used as a security hardening feature (e.g. CVE-2019-2215 and CVE-2019-2025 are mitigated by the option [4]).

The feature has never been designed with performance in mind, yet common list manipulation is happening across hot paths all over the kernel.

Introduce CONFIG_HARDEN_LIST, which performs list pointer checking inline, and only upon list corruption calls the reporting slow path.

Since DEBUG_LIST is functionally a superset of HARDEN_LIST, the Kconfig variables are designed to reflect that: DEBUG_LIST selects HARDEN_LIST, whereas HARDEN_LIST itself has no dependency on DEBUG_LIST.

To generate optimal machine code with CONFIG_HARDEN_LIST:

1. Elide checking for pointer values which upon dereference would result in an immediate access fault -- therefore "minimal" checks. The trade-off is lower-quality error reports.

2. Use the newly introduced __preserve_most function attribute (available with Clang, but not yet with GCC) to minimize the code footprint for calling the reporting slow path. As a result, function size of callers is reduced by avoiding saving registers before calling the rarely called reporting slow path.

Note that all TUs in lib/Makefile already disable function tracing, including list_debug.c, and __preserve_most's implied notrace has no effect in this case.

3. Because the inline checks are a subset of the full set of checks in __list_*_valid_or_report(), always return false if the inline checks failed. This avoids redundant compare and conditional branch right after return from the slow path.

As a side-effect of the checks being inline, if the compiler can prove some condition to always be true, it can completely elide some checks.

Running netperf with CONFIG_HARDEN_LIST (using a Clang compiler with "preserve_most") shows throughput improvements, in my case of ~7% on average (up to 20-30% on some test cases).

---
v4:
* Rename to CONFIG_HARDEN_LIST, which can independently be selected from CONFIG_DEBUG_LIST.

v3:
* Rename ___list_*_valid() to __list_*_valid_or_report().
* More comments.

v2:
* Note that lib/Makefile disables function tracing for everything and __preserve_most's implied notrace is a noop here.

---
 arch/arm64/kvm/hyp/nvhe/Makefile     |  2 +-
 arch/arm64/kvm/hyp/nvhe/list_debug.c |  2 +
 include/linux/list.h                 | 64 +++++++++++++++++++++++++---
 lib/Kconfig.debug                    |  9 +++-
 lib/Makefile                         |  2 +-
 lib/list_debug.c                     |  5 ++-
 security/Kconfig.hardening           | 13 ++++++
 7 files changed, 86 insertions(+), 11 deletions(-)
arch/arm64/kvm/hyp/nvhe/Makefile +1 -1
index 9ddc025e4b86..c89c85a41ac4 100644
@@ -25,7 +25,7 @@ hyp-obj-y := timer-sr.o sysreg-sr.o debug-sr.o switch.o tlb.o hyp-init.o host.o
 	 cache.o setup.o mm.o mem_protect.o sys_regs.o pkvm.o stacktrace.o ffa.o
 hyp-obj-y += ../vgic-v3-sr.o ../aarch32.o ../vgic-v2-cpuif-proxy.o ../entry.o \
 	 ../fpsimd.o ../hyp-entry.o ../exception.o ../pgtable.o
-hyp-obj-$(CONFIG_DEBUG_LIST) += list_debug.o
+hyp-obj-$(CONFIG_HARDEN_LIST) += list_debug.o
 hyp-obj-y += $(lib-objs)
 
 ##
arch/arm64/kvm/hyp/nvhe/list_debug.c +2 -0
index 16266a939a4c..46a2d4f2b3c6 100644
@@ -26,6 +26,7 @@ static inline __must_check bool nvhe_check_data_corruption(bool v)
 
 /* The predicates checked here are taken from lib/list_debug.c. */
 
+__list_valid_slowpath
 bool __list_add_valid_or_report(struct list_head *new, struct list_head *prev,
 				struct list_head *next)
 {
@@ -37,6 +38,7 @@ bool __list_add_valid_or_report(struct list_head *new, struct list_head *prev,
 	return true;
 }
 
+__list_valid_slowpath
 bool __list_del_entry_valid_or_report(struct list_head *entry)
 {
 	struct list_head *prev, *next;
include/linux/list.h +58 -6

97 lines not on this page · raw message

lib/Kconfig.debug +7 -2
index fbc89baf7de6..9e38956d6f50 100644
@@ -1674,9 +1674,14 @@ menu "Debug kernel data structures"
 config DEBUG_LIST
 	bool "Debug linked list manipulation"
 	depends on DEBUG_KERNEL || BUG_ON_DATA_CORRUPTION
+	select HARDEN_LIST
 	help
-	  Enable this to turn on extended checks in the linked-list
-	  walking routines.
+	  Enable this to turn on extended checks in the linked-list walking
+	  routines.
+
+	  This option trades better quality error reports for performance, and
+	  is more suitable for kernel debugging. If you care about performance,
+	  you should only enable CONFIG_HARDEN_LIST instead.
 
 	  If unsure, say N.
 
lib/Makefile +1 -1
index 42d307ade225..a7ebc9090f04 100644
@@ -161,7 +161,7 @@ obj-$(CONFIG_BTREE) += btree.o
 obj-$(CONFIG_INTERVAL_TREE) += interval_tree.o
 obj-$(CONFIG_ASSOCIATIVE_ARRAY) += assoc_array.o
 obj-$(CONFIG_DEBUG_PREEMPT) += smp_processor_id.o
-obj-$(CONFIG_DEBUG_LIST) += list_debug.o
+obj-$(CONFIG_HARDEN_LIST) += list_debug.o
 obj-$(CONFIG_DEBUG_OBJECTS) += debugobjects.o
 
 obj-$(CONFIG_BITREVERSE) += bitrev.o
lib/list_debug.c +4 -1
index 2def33b1491f..38ddc7c01eab 100644
@@ -2,7 +2,8 @@
  * Copyright 2006, Red Hat, Inc., Dave Jones
  * Released under the General Public License (GPL).
  *
- * This file contains the linked list validation for DEBUG_LIST.
+ * This file contains the linked list validation and error reporting for
+ * HARDEN_LIST and DEBUG_LIST.
  */
 
 #include <linux/export.h>
@@ -17,6 +18,7 @@
  * attempt).
  */
 
+__list_valid_slowpath
 bool __list_add_valid_or_report(struct list_head *new, struct list_head *prev,
 				struct list_head *next)
 {
@@ -39,6 +41,7 @@ bool __list_add_valid_or_report(struct list_head *new, struct list_head *prev,
 }
 EXPORT_SYMBOL(__list_add_valid_or_report);
 
+__list_valid_slowpath
 bool __list_del_entry_valid_or_report(struct list_head *entry)
 {
 	struct list_head *prev, *next;
security/Kconfig.hardening +13 -0
index 0f295961e773..19aa2b7d2f64 100644
@@ -279,6 +279,19 @@ config ZERO_CALL_USED_REGS
 
 endmenu
 
+menu "Hardening of kernel data structures"
+
+config HARDEN_LIST
+	bool "Check integrity of linked list manipulation"
+	help
+	  Minimal integrity checking in the linked-list manipulation routines
+	  to catch memory corruptions that are not guaranteed to result in an
+	  immediate access fault.
+
+	  If unsure, say N.
+
+endmenu
+
 config CC_HAS_RANDSTRUCT
 	def_bool $(cc-option,-frandomize-layout-seed-file=/dev/null)
 	# Randstruct was first added in Clang 15, but it isn't safe to use until
Signature · 1 line
2.41.0.640.ga95def55d0-goog
#8 lore raw

Re: [PATCH v3 3/3] list_debug: Introduce CONFIG_DEBUG_LIST_MINIMAL

Quoted 15 lines · Marco Elver

On Wed, Aug 09, 2023 at 11:30AM -0400, Steven Rostedt wrote:
[...]

Quoted 11 lines

I would actually prefer DEBUG_LIST to select HARDEN_LIST and not the other way around. It logically doesn't make sense that HARDEN_LIST would select DEBUG_LIST. That is, I could by default want HARDEN_LIST always on, but not DEBUG_LIST (because who knows, it may add other features I don't want). But then, I may have stumbled over something and want more info, and enable DEBUG_LIST (while still having HARDEN_LIST) enabled.

I think you are looking at this from an implementation perspective and not the normal developer one.

[...]

Quoted 1 line
Quoted 7 lines
Quoted 3 lines

That is, if DEBUG_LIST is enabled, we always call the __list_add_valid_or_report(), but if only HARDEN_LIST is enabled, then we do the shortcut.

Good point - I think this is better. See below tentative v4.

Kees: Does that also look more like what you had in mind?

Yeah, this looks good. My only nit would be a naming one. All the other hardening features are named "HARDENED", but perhaps the "ED" is redundant in the others. Still, consistency seems nicer. What do you think of CONFIG_LIST_HARDENED ? (The modern trend for Kconfig naming tends to keep the subsystem name first and then apply optional elements after.)

One note: do the LKDTM list hardening tests still pass? i.e. CORRUPT_LIST_ADD
CORRUPT_LIST_DEL

include/linux/list.h static inline void INIT_LIST_HEAD(struct list_head *list) · line 120

Quoted 8 lines

[...]

include/linux/list.h +7 -0
+		/*
+		 * With the hardening version, elide checking if next and prev
+		 * are NULL, LIST_POISON1 or LIST_POISON2, since the immediate
+		 * dereference of them below would result in a fault.
+		 */
+		if (likely(prev->next == entry && next->prev == entry))
+			return true;

I'm not super excited about skipping those checks, since they are values that can be reached through kernel list management confusion. If an attacker is using a system where the zero-page has been mapped and is accessible (i.e. lacking SMAP etc), then attacks could still be constructed. However, I do recognize this chain of exploitation prerequisites is getting rather long, so probably this is a reasonable trade off on modern systems.

-Kees

Signature · 1 line
Kees Cook
#9 lore raw

Re: [PATCH v3 3/3] list_debug: Introduce CONFIG_DEBUG_LIST_MINIMAL

Quoted 23 lines · Kees Cook
Quoted 21 lines · Marco Elver

On Wed, Aug 09, 2023 at 11:30AM -0400, Steven Rostedt wrote:
[...]

Quoted 11 lines

I would actually prefer DEBUG_LIST to select HARDEN_LIST and not the other way around. It logically doesn't make sense that HARDEN_LIST would select DEBUG_LIST. That is, I could by default want HARDEN_LIST always on, but not DEBUG_LIST (because who knows, it may add other features I don't want). But then, I may have stumbled over something and want more info, and enable DEBUG_LIST (while still having HARDEN_LIST) enabled.

I think you are looking at this from an implementation perspective and not the normal developer one.

[...]

Quoted 4 lines

That is, if DEBUG_LIST is enabled, we always call the __list_add_valid_or_report(), but if only HARDEN_LIST is enabled, then we do the shortcut.

Good point - I think this is better. See below tentative v4.

Quoted 7 lines
Quoted 1 line

Kees: Does that also look more like what you had in mind?

Yeah, this looks good. My only nit would be a naming one. All the other hardening features are named "HARDENED", but perhaps the "ED" is redundant in the others. Still, consistency seems nicer. What do you think of CONFIG_LIST_HARDENED ? (The modern trend for Kconfig naming tends to keep the subsystem name first and then apply optional elements after.)

Naming is a bit all over. :-/
I agree with the <subsystem>_<suboption> scheme, generally. I think initially I tried to keep the name shorter, and also find a good counter-part to DEBUG_<suboption>, therefore HARDEN_LIST.

Let's just change it to CONFIG_LIST_HARDENED, given the existing "HARDENED" options.

I don't have a strong preference.

Quoted 3 lines

One note: do the LKDTM list hardening tests still pass? i.e. CORRUPT_LIST_ADD
CORRUPT_LIST_DEL

Yes, they do. Though I need to also adjust BUG_ON_DATA_CORRUPTION to select LIST_HARDENED, and the test should check for the new option (which is implied by DEBUG_LIST now). There will be an additional patch to adjust that.

Quoted 9 lines

include/linux/list.h static inline void INIT_LIST_HEAD(struct list_head *list) · line 120

Quoted 8 lines

[...]

include/linux/list.h +7 -0
+             /*
+              * With the hardening version, elide checking if next and prev
+              * are NULL, LIST_POISON1 or LIST_POISON2, since the immediate
+              * dereference of them below would result in a fault.
+              */
+             if (likely(prev->next == entry && next->prev == entry))
+                     return true;
Quoted 7 lines

I'm not super excited about skipping those checks, since they are values that can be reached through kernel list management confusion. If an attacker is using a system where the zero-page has been mapped and is accessible (i.e. lacking SMAP etc), then attacks could still be constructed. However, I do recognize this chain of exploitation prerequisites is getting rather long, so probably this is a reasonable trade off on modern systems.

Sure, it's a trade-off for systems which do have the bare minimum of modern hardware security features.

#10 lore raw

Re: [PATCH v3 3/3] list_debug: Introduce CONFIG_DEBUG_LIST_MINIMAL

In reply to Kees Cook
Quoted 9 lines · On Thu, 10 Aug 2023 13:11:58 -0700 Kees Cook <[email protected]> wrote:

include/linux/list.h static inline void INIT_LIST_HEAD(struct list_head *list) · line 120

Quoted 8 lines

[...]

include/linux/list.h +7 -0
+		/*
+		 * With the hardening version, elide checking if next and prev
+		 * are NULL, LIST_POISON1 or LIST_POISON2, since the immediate
+		 * dereference of them below would result in a fault.
+		 */
+		if (likely(prev->next == entry && next->prev == entry))
+			return true;  
Quoted 7 lines

I'm not super excited about skipping those checks, since they are values that can be reached through kernel list management confusion. If an attacker is using a system where the zero-page has been mapped and is accessible (i.e. lacking SMAP etc), then attacks could still be constructed. However, I do recognize this chain of exploitation prerequisites is getting rather long, so probably this is a reasonable trade off on modern systems.

A totally hardened machine is one that doesn't run ;-)

Yes, hopefully that when the kernel is configured with HARDENED it will eliminate steps to a prerequisite attack. I'm sure enabling lockdep would also help harden the system too. But there is a balance between security and performance. The more that adding security harms performance, the less people will use that security.

-- Steve

reviewer Acked-by on v3 1/3
#11 lore raw
In reply to Marco Elver
Quoted 34 lines · Marco Elver

[1]: "On X86-64 and AArch64 targets, this attribute changes the calling convention of a function. The preserve_most calling convention attempts to make the code in the caller as unintrusive as possible. This convention behaves identically to the C calling convention on how arguments and return values are passed, but it uses a different set of caller/callee-saved registers. This alleviates the burden of saving and recovering a large register set before and after the call in the caller. If the arguments are passed in callee-saved registers, then they will be preserved by the callee across the call. This doesn't apply for values returned in callee-saved registers.

* On X86-64 the callee preserves all general purpose registers, except for R11. R11 can be used as a scratch register. Floating-point registers (XMMs/YMMs) are not preserved and need to be saved by the caller.

* On AArch64 the callee preserve all general purpose registers, except x0-X8 and X16-X18."

[1] https://clang.llvm.org/docs/AttributeReference.html#preserve-most

Introduce the attribute to compiler_types.h as __preserve_most.

Use of this attribute results in better code generation for calls to very rarely called functions, such as error-reporting functions, or rarely executed slow paths.

Beware that the attribute conflicts with instrumentation calls inserted on function entry which do not use __preserve_most themselves. Notably, function tracing which assumes the normal C calling convention for the given architecture. Where the attribute is supported, __preserve_most will imply notrace. It is recommended to restrict use of the attribute to functions that should or already disable tracing.

Quoted 6 lines

The attribute may be supported by a future GCC version (see
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110899).

So long as this implies notrace I believe it is safe, so:

Mark.

Quoted 55 lines

---
v3:
* Quote more from LLVM documentation about which registers are callee/caller with preserve_most.
* Code comment to restrict use where tracing is meant to be disabled.

v2:
* Imply notrace, to avoid any conflicts with tracing which is inserted on function entry. See added comments.

---
 include/linux/compiler_types.h | 28 ++++++++++++++++++++++++++++
 1 file changed, 28 insertions(+)
include/linux/compiler_types.h +28 -0
index 547ea1ff806e..c88488715a39 100644
@@ -106,6 +106,34 @@ static inline void __chk_io_ptr(const volatile void __iomem *ptr) { }
 #define __cold
 #endif
 
+/*
+ * On x86-64 and arm64 targets, __preserve_most changes the calling convention
+ * of a function to make the code in the caller as unintrusive as possible. This
+ * convention behaves identically to the C calling convention on how arguments
+ * and return values are passed, but uses a different set of caller- and callee-
+ * saved registers.
+ *
+ * The purpose is to alleviates the burden of saving and recovering a large
+ * register set before and after the call in the caller.  This is beneficial for
+ * rarely taken slow paths, such as error-reporting functions that may be called
+ * from hot paths.
+ *
+ * Note: This may conflict with instrumentation inserted on function entry which
+ * does not use __preserve_most or equivalent convention (if in assembly). Since
+ * function tracing assumes the normal C calling convention, where the attribute
+ * is supported, __preserve_most implies notrace.  It is recommended to restrict
+ * use of the attribute to functions that should or already disable tracing.
+ *
+ * Optional: not supported by gcc.
+ *
+ * clang: https://clang.llvm.org/docs/AttributeReference.html#preserve-most
+ */
+#if __has_attribute(__preserve_most__)
+# define __preserve_most notrace __attribute__((__preserve_most__))
+#else
+# define __preserve_most
+#endif
+
 /* Builtins */
 
 /*
Signature · 1 line
2.41.0.640.ga95def55d0-goog