Skip to contentSkip to conversation

mm: kmsan: export kmsan_copy_page_meta()

PATCH Applied Alexander Potapenko 2 participants 7 messages Oct 24, 2022 – Oct 27, 2022

Subsystems (6)
lore
Share & tools
lore

Alexander Potapenko

author maintainer
#0 lore raw

PATCH 1/5 mm: kmsan: export kmsan_copy_page_meta()

Certain modules call copy_user_highpage(), which calls kmsan_copy_page_meta() under KMSAN, so we need to export the latter.

Cc Andrew Morton <[email protected]>
Fixes b073d7f8aee4 ("mm: kmsan: maintain KMSAN metadata for page operations")
---
 mm/kmsan/shadow.c | 1 +
 1 file changed, 1 insertion(+)
mm/kmsan/shadow.c +1 -0
index 21e3e196ec3cf..a787c04e9583c 100644
@@ -167,6 +167,7 @@ void kmsan_copy_page_meta(struct page *dst, struct page *src)
 	__memcpy(origin_ptr_for(dst), origin_ptr_for(src), PAGE_SIZE);
 	kmsan_leave_runtime();
 }
+EXPORT_SYMBOL(kmsan_copy_page_meta);
 
 void kmsan_alloc_page(struct page *page, unsigned int order, gfp_t flags)
 {
Signature · 1 line
2.38.0.135.g90850a2211-goog
author maintainer
#1 lore raw

PATCH 2/5 x86/purgatory: disable KMSAN instrumentation

The stand-alone purgatory.ro does not contain the KMSAN runtime, therefore it can't be built with KMSAN compiler instrumentation.

---
 arch/x86/purgatory/Makefile | 1 +
 1 file changed, 1 insertion(+)
arch/x86/purgatory/Makefile +1 -0
index 58a200dc762d6..17f09dc263811 100644
@@ -26,6 +26,7 @@ GCOV_PROFILE	:= n
 KASAN_SANITIZE	:= n
 UBSAN_SANITIZE	:= n
 KCSAN_SANITIZE	:= n
+KMSAN_SANITIZE	:= n
 KCOV_INSTRUMENT := n
 
 # These are adjustments to the compiler flags used for objects that
Signature · 1 line
2.38.0.135.g90850a2211-goog
author maintainer
#2 lore raw

PATCH 3/5 Kconfig.debug: disable CONFIG_FRAME_WARN for KMSAN by default

In reply to Alexander Potapenko

KMSAN adds a lot of instrumentation to the code, which results in increased stack usage (up to 2048 bytes and more in some cases). It's hard to predict how big the stack frames can be, so we disable the warnings for KMSAN instead.

---
 lib/Kconfig.debug | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
lib/Kconfig.debug +2 -1
index 3fc7abffc7aa2..29280072dc0e4 100644
@@ -400,8 +400,9 @@ config FRAME_WARN
 	default 1536 if (!64BIT && XTENSA)
 	default 1024 if !64BIT
 	default 2048 if 64BIT
+	default 0 if KMSAN
 	help
-	  Tell gcc to warn at build time for stack frames larger than this.
+	  Tell the compiler to warn at build time for stack frames larger than this.
 	  Setting this too low will cause a lot of warnings.
 	  Setting it to 0 disables the warning.
 
Signature · 1 line
2.38.0.135.g90850a2211-goog
#3 lore raw

Re: [PATCH 3/5] Kconfig.debug: disable CONFIG_FRAME_WARN for KMSAN by default

Quoted 22 lines · Alexander Potapenko

KMSAN adds a lot of instrumentation to the code, which results in increased stack usage (up to 2048 bytes and more in some cases). It's hard to predict how big the stack frames can be, so we disable the warnings for KMSAN instead.

---
 lib/Kconfig.debug | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
lib/Kconfig.debug +0 -0
index 3fc7abffc7aa2..29280072dc0e4 100644

lib/Kconfig.debug config FRAME_WARN · line 403

Quoted 4 lines
lib/Kconfig.debug +1 -0
@@ -400,8 +400,9 @@ config FRAME_WARN
        default 1536 if (!64BIT && XTENSA)
        default 1024 if !64BIT
        default 2048 if 64BIT
+       default 0 if KMSAN

This is wrong.

Kconfig picks up the first default entry which has true 'if' condition.

Since (!64BIT || 64BIT) covers all the possible cases, this patch is just adding dead code.

Quoted 9 lines
lib/Kconfig.debug +1 -2
        help
-         Tell gcc to warn at build time for stack frames larger than this.
+         Tell the compiler to warn at build time for stack frames larger than this.
          Setting this too low will cause a lot of warnings.
          Setting it to 0 disables the warning.
 
--

2.38.0.135.g90850a2211-goog

Signature · 2 lines
Best Regards
Masahiro Yamada
author maintainer
#4 lore raw

Re: [PATCH 3/5] Kconfig.debug: disable CONFIG_FRAME_WARN for KMSAN by default

Quoted 33 lines · Masahiro Yamada
Quoted 22 lines · Alexander Potapenko

KMSAN adds a lot of instrumentation to the code, which results in increased stack usage (up to 2048 bytes and more in some cases). It's hard to predict how big the stack frames can be, so we disable the warnings for KMSAN instead.

---
 lib/Kconfig.debug | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
lib/Kconfig.debug +0 -0
index 3fc7abffc7aa2..29280072dc0e4 100644

lib/Kconfig.debug config FRAME_WARN · line 403

Quoted 4 lines
lib/Kconfig.debug +1 -0
@@ -400,8 +400,9 @@ config FRAME_WARN
        default 1536 if (!64BIT && XTENSA)
        default 1024 if !64BIT
        default 2048 if 64BIT
+       default 0 if KMSAN

This is wrong.

Quoted 7 lines

Kconfig picks up the first default entry which has true 'if' condition.

Since (!64BIT || 64BIT) covers all the possible cases, this patch is just adding dead code.

Thanks for clarifying!
What we actually need is to forcefully disable CONFIG_FRAME_WARN under KMSAN, but adding constructs like "select FRAME_WARN 0" or "select FRAME_WARN=0" doesn't work, right?

I'll move "default 0 if KMSAN" to the beginning of FRAME_WARN declaration then.

Quoted 19 lines
Quoted 12 lines
lib/Kconfig.debug +1 -2
        help
-         Tell gcc to warn at build time for stack frames larger than this.
+         Tell the compiler to warn at build time for stack frames larger than this.
          Setting this too low will cause a lot of warnings.
          Setting it to 0 disables the warning.
 
--
Quoted 2 lines

2.38.0.135.g90850a2211-goog

--
Best Regards
Masahiro Yamada

Signature · 10 lines
Alexander Potapenko
Software Engineer

Google Germany GmbH
Erika-Mann-Straße, 33
80636 München

Geschäftsführer: Paul Manicle, Liana Sebastian
Registergericht und -nummer: Hamburg, HRB 86891
Sitz der Gesellschaft: Hamburg
author maintainer
#5 lore raw

PATCH 4/5 x86: asm: make sure __put_user_size() evaluates pointer once

In reply to Alexander Potapenko

User access macros must ensure their arguments are evaluated only once if they are used more than once in the macro body. Adding instrument_put_user() to __put_user_size() resulted in double evaluation of the ptr argument, which led to correctness issues when performing e.g. unsafe_put_user(..., p++, ...).
To fix those issues, evaluate the ptr argument of __put_user_size() at the beginning of the macro.

Cc Andrew Morton <[email protected]>
Cc Borislav Petkov <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Thomas Gleixner <[email protected]>
Fixes 888f84a6da4d ("x86: asm: instrument usercopy in get_user() and put_user()")
---
 arch/x86/include/asm/uaccess.h | 13 +++++++------
 1 file changed, 7 insertions(+), 6 deletions(-)
arch/x86/include/asm/uaccess.h +7 -6
index 8bc614cfe21b9..1cc756eafa447 100644
@@ -254,24 +254,25 @@ extern void __put_user_nocheck_8(void);
 #define __put_user_size(x, ptr, size, label)				\
 do {									\
 	__typeof__(*(ptr)) __x = (x); /* eval x once */			\
-	__chk_user_ptr(ptr);						\
+	__typeof__(ptr) __ptr = (ptr); /* eval ptr once */		\
+	__chk_user_ptr(__ptr);						\
 	switch (size) {							\
 	case 1:								\
-		__put_user_goto(__x, ptr, "b", "iq", label);		\
+		__put_user_goto(__x, __ptr, "b", "iq", label);		\
 		break;							\
 	case 2:								\
-		__put_user_goto(__x, ptr, "w", "ir", label);		\
+		__put_user_goto(__x, __ptr, "w", "ir", label);		\
 		break;							\
 	case 4:								\
-		__put_user_goto(__x, ptr, "l", "ir", label);		\
+		__put_user_goto(__x, __ptr, "l", "ir", label);		\
 		break;							\
 	case 8:								\
-		__put_user_goto_u64(__x, ptr, label);			\
+		__put_user_goto_u64(__x, __ptr, label);			\
 		break;							\
 	default:							\
 		__put_user_bad();					\
 	}								\
-	instrument_put_user(__x, ptr, size);				\
+	instrument_put_user(__x, __ptr, size);				\
 } while (0)
 
 #ifdef CONFIG_CC_HAS_ASM_GOTO_OUTPUT
Signature · 1 line
2.38.0.135.g90850a2211-goog
author maintainer
#6 lore raw

PATCH 5/5 x86: fortify: kmsan: fix KMSAN fortify builds

In reply to Alexander Potapenko

Ensure that KMSAN builds replace memset/memcpy/memmove calls with the respective __msan_XXX functions, and that none of the macros are redefined twice. This should allow building kernel with both CONFIG_KMSAN and CONFIG_FORTIFY_SOURCE.

---
 arch/x86/include/asm/string_64.h | 11 +++++++----
 include/linux/fortify-string.h   | 17 +++++++++++++++--
 include/linux/kmsan_string.h     | 21 +++++++++++++++++++++
 mm/kmsan/instrumentation.c       |  1 +
 4 files changed, 44 insertions(+), 6 deletions(-)
 create mode 100644 include/linux/kmsan_string.h
arch/x86/include/asm/string_64.h +7 -4
index 3b87d889b6e16..888731ccf1f67 100644
@@ -10,10 +10,13 @@
 /* Even with __builtin_ the compiler may decide to use the out of line
    function. */
 
+#if defined(__SANITIZE_MEMORY__) && defined(__NO_FORTIFY)
+#include <linux/kmsan_string.h>
+#endif
+
 #define __HAVE_ARCH_MEMCPY 1
-#if defined(__SANITIZE_MEMORY__)
+#if defined(__SANITIZE_MEMORY__) && defined(__NO_FORTIFY)
 #undef memcpy
-void *__msan_memcpy(void *dst, const void *src, size_t size);
 #define memcpy __msan_memcpy
 #else
 extern void *memcpy(void *to, const void *from, size_t len);
@@ -21,7 +24,7 @@ extern void *memcpy(void *to, const void *from, size_t len);
 extern void *__memcpy(void *to, const void *from, size_t len);
 
 #define __HAVE_ARCH_MEMSET
-#if defined(__SANITIZE_MEMORY__)
+#if defined(__SANITIZE_MEMORY__) && defined(__NO_FORTIFY)
 extern void *__msan_memset(void *s, int c, size_t n);
 #undef memset
 #define memset __msan_memset
@@ -67,7 +70,7 @@ static inline void *memset64(uint64_t *s, uint64_t v, size_t n)
 }
 
 #define __HAVE_ARCH_MEMMOVE
-#if defined(__SANITIZE_MEMORY__)
+#if defined(__SANITIZE_MEMORY__) && defined(__NO_FORTIFY)
 #undef memmove
 void *__msan_memmove(void *dest, const void *src, size_t len);
 #define memmove __msan_memmove
include/linux/fortify-string.h +15 -2
index 4029fe368a4f6..18a31b125f9d6 100644
@@ -43,11 +43,24 @@ extern __kernel_size_t __underlying_strlen(const char *p) __RENAME(strlen);
 extern char *__underlying_strncat(char *p, const char *q, __kernel_size_t count) __RENAME(strncat);
 extern char *__underlying_strncpy(char *p, const char *q, __kernel_size_t size) __RENAME(strncpy);
 #else
-#define __underlying_memchr	__builtin_memchr
-#define __underlying_memcmp	__builtin_memcmp
+
+#if defined(__SANITIZE_MEMORY__)
+/*
+ * For KMSAN builds all memcpy/memset/memmove calls should be replaced by the
+ * corresponding __msan_XXX functions.
+ */
+#include <linux/kmsan_string.h>
+#define __underlying_memcpy	__msan_memcpy
+#define __underlying_memmove	__msan_memmove
+#define __underlying_memset	__msan_memset
+#else
 #define __underlying_memcpy	__builtin_memcpy
 #define __underlying_memmove	__builtin_memmove
 #define __underlying_memset	__builtin_memset
+#endif
+
+#define __underlying_memchr	__builtin_memchr
+#define __underlying_memcmp	__builtin_memcmp
 #define __underlying_strcat	__builtin_strcat
 #define __underlying_strcpy	__builtin_strcpy
 #define __underlying_strlen	__builtin_strlen
include/linux/kmsan_string.h new file +21 -0
new file mode 100644
index 0000000000000..7287da6f52eff
@@ -0,0 +1,21 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * KMSAN string functions API used in other headers.
+ *
+ * Copyright (C) 2022 Google LLC
+ * Author: Alexander Potapenko <[email protected]>
+ *
+ */
+#ifndef _LINUX_KMSAN_STRING_H
+#define _LINUX_KMSAN_STRING_H
+
+/*
+ * KMSAN overrides the default memcpy/memset/memmove implementations in the
+ * kernel, which requires having __msan_XXX function prototypes in several other
+ * headers. Keep them in one place instead of open-coding.
+ */
+void *__msan_memcpy(void *dst, const void *src, size_t size);
+void *__msan_memset(void *s, int c, size_t n);
+void *__msan_memmove(void *dest, const void *src, size_t len);
+
+#endif /* _LINUX_KMSAN_STRING_H */
mm/kmsan/instrumentation.c +1 -0
index 280d154132684..271f135f97a16 100644
@@ -14,6 +14,7 @@
 
 #include "kmsan.h"
 #include <linux/gfp.h>
+#include <linux/kmsan_string.h>
 #include <linux/mm.h>
 #include <linux/uaccess.h>
 
Signature · 1 line
2.38.0.135.g90850a2211-goog