Skip to contentSkip to conversation

string.h: Use ARRAY_SIZE() for memtostr*()/strtomem*()

PATCH Superseded Kees Cook 6 participants 18 messages Feb 6, 2025 – Feb 7, 2025

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Kees Cook

author maintainer
#0 lore raw

cover string.h: Use ARRAY_SIZE() for memtostr*()/strtomem*()

Work around a Clang 14 bug by switching to ARRAY_SIZE() (with the added benefit of explicitly checking for char array arguments) in memtostr*/strtomem*().

-Kees

Kees Cook (3):
compiler.h: Move C string helpers into C-only kernel section compiler.h: Introduce __must_be_char_array() string.h: Use ARRAY_SIZE() for memtostr*()/strtomem*()

 include/linux/compiler.h | 32 +++++++++++++++++++-------------
 include/linux/string.h   | 12 ++++++++----
 2 files changed, 27 insertions(+), 17 deletions(-)
Signature · 1 line
2.34.1
author maintainer
#1 lore raw

PATCH 1/3 compiler.h: Move C string helpers into C-only kernel section

The C kernel helpers for evaluating C Strings were placed outside of the assembly ifdef. Move them to the correct place so future changes won't confuse the assembler.

Fixes d7a516c6eeae ("compiler.h: Fix undefined BUILD_BUG_ON_ZERO()")
Fixes 559048d156ff ("string: Check for "nonstring" attribute on strscpy() arguments")
---
 include/linux/compiler.h | 26 +++++++++++++-------------
 1 file changed, 13 insertions(+), 13 deletions(-)
include/linux/compiler.h +13 -13
index 240c632c5b95..7af999a131cb 100644
@@ -214,6 +214,19 @@ void ftrace_likely_update(struct ftrace_likely_data *f, int val,
 	__v;								\
 })
 
+#ifdef __CHECKER__
+#define __BUILD_BUG_ON_ZERO_MSG(e, msg) (0)
+#else /* __CHECKER__ */
+#define __BUILD_BUG_ON_ZERO_MSG(e, msg) ((int)sizeof(struct {_Static_assert(!(e), msg);}))
+#endif /* __CHECKER__ */
+
+/* &a[0] degrades to a pointer: a different type from an array */
+#define __must_be_array(a)	__BUILD_BUG_ON_ZERO_MSG(__same_type((a), &(a)[0]), "must be array")
+
+/* Require C Strings (i.e. NUL-terminated) lack the "nonstring" attribute. */
+#define __must_be_cstr(p) \
+	__BUILD_BUG_ON_ZERO_MSG(__annotated(p, nonstring), "must be cstr (NUL-terminated)")
+
 #endif /* __KERNEL__ */
 
 /**
@@ -254,19 +267,6 @@ static inline void *offset_to_ptr(const int *off)
 
 #define __ADDRESSABLE_ASM_STR(sym) __stringify(__ADDRESSABLE_ASM(sym))
 
-#ifdef __CHECKER__
-#define __BUILD_BUG_ON_ZERO_MSG(e, msg) (0)
-#else /* __CHECKER__ */
-#define __BUILD_BUG_ON_ZERO_MSG(e, msg) ((int)sizeof(struct {_Static_assert(!(e), msg);}))
-#endif /* __CHECKER__ */
-
-/* &a[0] degrades to a pointer: a different type from an array */
-#define __must_be_array(a)	__BUILD_BUG_ON_ZERO_MSG(__same_type((a), &(a)[0]), "must be array")
-
-/* Require C Strings (i.e. NUL-terminated) lack the "nonstring" attribute. */
-#define __must_be_cstr(p) \
-	__BUILD_BUG_ON_ZERO_MSG(__annotated(p, nonstring), "must be cstr (NUL-terminated)")
-
 /*
  * This returns a constant expression while determining if an argument is
  * a constant expression, most importantly without evaluating the argument.
Signature · 1 line
2.34.1
Reviewed-by on v1 1/3
#2 lore raw

Re: [PATCH 1/3] compiler.h: Move C string helpers into C-only kernel section

Quoted 4 lines · Kees Cook

The C kernel helpers for evaluating C Strings were placed outside of the assembly ifdef. Move them to the correct place so future changes won't confuse the assembler.

This moves it also into the kernel ifdef too -- I assume that is intentional. (Perhaps mention ifndef instead of ifdef?)

I checked that this is indeed a pure move, so:

Cheers,
Miguel

author maintainer
#3 lore raw

Re: [PATCH 1/3] compiler.h: Move C string helpers into C-only kernel section

Quoted 8 lines · Miguel Ojeda
Quoted 4 lines · Kees Cook

The C kernel helpers for evaluating C Strings were placed outside of the assembly ifdef. Move them to the correct place so future changes won't confuse the assembler.

This moves it also into the kernel ifdef too -- I assume that is intentional. (Perhaps mention ifndef instead of ifdef?)

Right, yeah. I guess I only mentioned that in the Subject. I'll clarify.

Quoted 4 lines

I checked that this is indeed a pure move, so:

Thanks!

-Kees

Signature · 1 line
Kees Cook
author maintainer
#4 lore raw

PATCH 2/3 compiler.h: Introduce __must_be_char_array()

In reply to Kees Cook

In preparation for adding stricter type checking to the str/mem*() helpers, provide a way to check that a variable is a character array via __must_be_char_array().

---
 include/linux/compiler.h | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
include/linux/compiler.h +7 -1
index 7af999a131cb..a577fe0b1f8a 100644
@@ -221,7 +221,13 @@ void ftrace_likely_update(struct ftrace_likely_data *f, int val,
 #endif /* __CHECKER__ */
 
 /* &a[0] degrades to a pointer: a different type from an array */
-#define __must_be_array(a)	__BUILD_BUG_ON_ZERO_MSG(__same_type((a), &(a)[0]), "must be array")
+#define __is_array(a)		(!__same_type((a), &(a)[0]))
+#define __must_be_array(a)	__BUILD_BUG_ON_ZERO_MSG(!__is_array(a), \
+							"must be array")
+
+#define __is_char_array(a)	(__is_array(a) && sizeof((a)[0]) == 1)
+#define __must_be_char_array(a)	__BUILD_BUG_ON_ZERO_MSG(!__is_char_array(a), \
+							"must be byte array")
 
 /* Require C Strings (i.e. NUL-terminated) lack the "nonstring" attribute. */
 #define __must_be_cstr(p) \
Signature · 1 line
2.34.1
#5 lore raw

Re: [PATCH 2/3] compiler.h: Introduce __must_be_char_array()

Quoted 4 lines · On Thu, 6 Feb 2025 10:11:29 -0800 Kees Cook <[email protected]> wrote:

In preparation for adding stricter type checking to the str/mem*() helpers, provide a way to check that a variable is a character array via __must_be_char_array().

...

include/linux/compiler.h void ftrace_likely_update(struct ftrace_likely_data *f, int val, · line 230

Quoted 3 lines
include/linux/compiler.h +3 -0
+#define __is_char_array(a)	(__is_array(a) && sizeof((a)[0]) == 1)
+#define __must_be_char_array(a)	__BUILD_BUG_ON_ZERO_MSG(!__is_char_array(a), \
+							"must be byte array")

If you are only checking the size, perhaps __is_byte_array() would be a better name.
(You've even used 'byte' in the error message.)

David

author maintainer
#6 lore raw

Re: [PATCH 2/3] compiler.h: Introduce __must_be_char_array()

Quoted 7 lines · David Laight
Quoted 4 lines · On Thu, 6 Feb 2025 10:11:29 -0800 Kees Cook <[email protected]> wrote:

In preparation for adding stricter type checking to the str/mem*() helpers, provide a way to check that a variable is a character array via __must_be_char_array().

Quoted 8 lines

...

include/linux/compiler.h void ftrace_likely_update(struct ftrace_likely_data *f, int val, · line 230

Quoted 3 lines
include/linux/compiler.h +3 -0
+#define __is_char_array(a)	(__is_array(a) && sizeof((a)[0]) == 1)
+#define __must_be_char_array(a)	__BUILD_BUG_ON_ZERO_MSG(!__is_char_array(a), \
+							"must be byte array")

If you are only checking the size, perhaps __is_byte_array() would be a better name.
(You've even used 'byte' in the error message.)

Yeah, fair point. I went and forth on naming. Fixed for v2.

Signature · 1 line
Kees Cook
#7 lore raw

Re: [PATCH 2/3] compiler.h: Introduce __must_be_char_array()

In reply to Kees Cook
Quoted 5 lines · Kees Cook

In preparation for adding stricter type checking to the str/mem*() helpers, provide a way to check that a variable is a character array via __must_be_char_array().

Suggested-by? :)

Quoted 26 lines
---
 include/linux/compiler.h | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
include/linux/compiler.h +7 -1
index 7af999a131cb..a577fe0b1f8a 100644
@@ -221,7 +221,13 @@ void ftrace_likely_update(struct ftrace_likely_data *f, int val,
 #endif /* __CHECKER__ */
 
 /* &a[0] degrades to a pointer: a different type from an array */
-#define __must_be_array(a)	__BUILD_BUG_ON_ZERO_MSG(__same_type((a), &(a)[0]), "must be array")
+#define __is_array(a)		(!__same_type((a), &(a)[0]))
+#define __must_be_array(a)	__BUILD_BUG_ON_ZERO_MSG(!__is_array(a), \
+							"must be array")
+
+#define __is_char_array(a)	(__is_array(a) && sizeof((a)[0]) == 1)
+#define __must_be_char_array(a)	__BUILD_BUG_ON_ZERO_MSG(!__is_char_array(a), \
+							"must be byte array")
 
 /* Require C Strings (i.e. NUL-terminated) lack the "nonstring" attribute. */
 #define __must_be_cstr(p) \
Signature · 1 line
2.34.1
author maintainer
#8 lore raw

Re: [PATCH 2/3] compiler.h: Introduce __must_be_char_array()

Quoted 8 lines · Kent Overstreet
Quoted 5 lines · Kees Cook

In preparation for adding stricter type checking to the str/mem*() helpers, provide a way to check that a variable is a character array via __must_be_char_array().

Suggested-by? :)

Sure, I'll add that. I did it for the ARRAY_SIZE() and forgot which pieces came from where when I split up the patches. :)

Signature · 1 line
Kees Cook
#9 lore raw

Re: [PATCH 2/3] compiler.h: Introduce __must_be_char_array()

In reply to Kees Cook
Quoted 16 lines · Kees Cook

In preparation for adding stricter type checking to the str/mem*() helpers, provide a way to check that a variable is a character array via __must_be_char_array().

---
 include/linux/compiler.h | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
include/linux/compiler.h +0 -0
index 7af999a131cb..a577fe0b1f8a 100644
@@ -221,7 +221,13 @@ void ftrace_likely_update(struct ftrace_likely_data *f, int val,
 #endif /* __CHECKER__ */
 

include/linux/compiler.h void ftrace_likely_update(struct ftrace_likely_data *f, int val, · line 231

Quoted 10 lines
include/linux/compiler.h +7 -1
 /* &a[0] degrades to a pointer: a different type from an array */
-#define __must_be_array(a)	__BUILD_BUG_ON_ZERO_MSG(__same_type((a), &(a)[0]), "must be array")
+#define __is_array(a)		(!__same_type((a), &(a)[0]))
+#define __must_be_array(a)	__BUILD_BUG_ON_ZERO_MSG(!__is_array(a), \
+							"must be array")
+
+#define __is_char_array(a)	(__is_array(a) && sizeof((a)[0]) == 1)
+#define __must_be_char_array(a)	__BUILD_BUG_ON_ZERO_MSG(!__is_char_array(a), \
+							"must be byte array")
 

It's probably unlikely to ever encounter an array of _Bool or array of structs-with-a-single-char-member in the wild, but it does seem a bit odd to base the test on sizeof(). Why not add a

__is_character_type(t) (__same_type(t, char) ||
__same_type(t, signed char) || __same_type(t, unsigned char) )

helper and write the test using __is_character_type((a)[0])?

Or if you really mean that it must be an array of char, not any of the three "character types", simply replace the sizeof() by __same_type(a[0], char)

Rasmus

#10 lore raw

Re: [PATCH 2/3] compiler.h: Introduce __must_be_char_array()

Quoted 37 lines · On Fri, 07 Feb 2025 09:55:00 +0100 Rasmus Villemoes <[email protected]> wrote:
Quoted 16 lines · Kees Cook

In preparation for adding stricter type checking to the str/mem*() helpers, provide a way to check that a variable is a character array via __must_be_char_array().

---
 include/linux/compiler.h | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
include/linux/compiler.h +0 -0
index 7af999a131cb..a577fe0b1f8a 100644
@@ -221,7 +221,13 @@ void ftrace_likely_update(struct ftrace_likely_data *f, int val,
 #endif /* __CHECKER__ */
 

include/linux/compiler.h void ftrace_likely_update(struct ftrace_likely_data *f, int val, · line 231

Quoted 10 lines
include/linux/compiler.h +7 -1
 /* &a[0] degrades to a pointer: a different type from an array */
-#define __must_be_array(a)	__BUILD_BUG_ON_ZERO_MSG(__same_type((a), &(a)[0]), "must be array")
+#define __is_array(a)		(!__same_type((a), &(a)[0]))
+#define __must_be_array(a)	__BUILD_BUG_ON_ZERO_MSG(!__is_array(a), \
+							"must be array")
+
+#define __is_char_array(a)	(__is_array(a) && sizeof((a)[0]) == 1)
+#define __must_be_char_array(a)	__BUILD_BUG_ON_ZERO_MSG(!__is_char_array(a), \
+							"must be byte array")
 

It's probably unlikely to ever encounter an array of _Bool or array of structs-with-a-single-char-member in the wild, but it does seem a bit odd to base the test on sizeof(). Why not add a

__is_character_type(t) (__same_type(t, char) ||
__same_type(t, signed char) || __same_type(t, unsigned char) )

Quoted 5 lines

helper and write the test using __is_character_type((a)[0])?

Or if you really mean that it must be an array of char, not any of the three "character types", simply replace the sizeof() by __same_type(a[0], char)

I'm not sure it matters enough to slow down the compilation by that much cpp bloat.

David

#11 lore raw

Re: [PATCH 2/3] compiler.h: Introduce __must_be_char_array()

In reply to Kees Cook
Quoted 31 lines · Kees Cook

In preparation for adding stricter type checking to the str/mem*() helpers, provide a way to check that a variable is a character array via __must_be_char_array().

---
 include/linux/compiler.h | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
include/linux/compiler.h +7 -1
index 7af999a131cb..a577fe0b1f8a 100644
@@ -221,7 +221,13 @@ void ftrace_likely_update(struct ftrace_likely_data *f, int val,
 #endif /* __CHECKER__ */
 
 /* &a[0] degrades to a pointer: a different type from an array */
-#define __must_be_array(a)	__BUILD_BUG_ON_ZERO_MSG(__same_type((a), &(a)[0]), "must be array")
+#define __is_array(a)		(!__same_type((a), &(a)[0]))
+#define __must_be_array(a)	__BUILD_BUG_ON_ZERO_MSG(!__is_array(a), \
+							"must be array")
+
+#define __is_char_array(a)	(__is_array(a) && sizeof((a)[0]) == 1)
+#define __must_be_char_array(a)	__BUILD_BUG_ON_ZERO_MSG(!__is_char_array(a), \
+							"must be byte array")
 
 /* Require C Strings (i.e. NUL-terminated) lack the "nonstring" attribute. */
 #define __must_be_cstr(p) \
Signature · 1 line
2.34.1

Why not just use a combination of ARRAY_SIZE() and a function call (to verify that it's a character array) for the typechecking?

author maintainer
#12 lore raw

PATCH 3/3 string.h: Use ARRAY_SIZE() for memtostr*()/strtomem*()

In reply to Kees Cook

The destination argument of memtostr*() and strtomem*() must be a fixed-size char array at compile time, so there is no need to use __builtin_object_size() (which is useful for when an argument is either a pointer or unknown). Instead use ARRAY_SIZE(), which has the benefit of working around a bug in Clang (fixed[1] in 15+) that got __builtin_object_size() wrong sometimes.

---
Cc: Suren Baghdasaryan <[email protected]>
Cc: [email protected]
Cc: Andy Shevchenko <[email protected]>
Cc: [email protected]

---
 include/linux/string.h | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)
include/linux/string.h +8 -4
index 493ac4862c77..01ac26be274d 100644
@@ -411,7 +411,8 @@ void memcpy_and_pad(void *dest, size_t dest_len, const void *src, size_t count,
  * must be discoverable by the compiler.
  */
 #define strtomem_pad(dest, src, pad)	do {				\
-	const size_t _dest_len = __builtin_object_size(dest, 1);	\
+	const size_t _dest_len = __must_be_char_array(dest) +		\
+				ARRAY_SIZE(dest);			\
 	const size_t _src_len = __builtin_object_size(src, 1);		\
 									\
 	BUILD_BUG_ON(!__builtin_constant_p(_dest_len) ||		\
@@ -434,7 +435,8 @@ void memcpy_and_pad(void *dest, size_t dest_len, const void *src, size_t count,
  * must be discoverable by the compiler.
  */
 #define strtomem(dest, src)	do {					\
-	const size_t _dest_len = __builtin_object_size(dest, 1);	\
+	const size_t _dest_len = __must_be_char_array(dest) +		\
+				ARRAY_SIZE(dest);			\
 	const size_t _src_len = __builtin_object_size(src, 1);		\
 									\
 	BUILD_BUG_ON(!__builtin_constant_p(_dest_len) ||		\
@@ -453,7 +455,8 @@ void memcpy_and_pad(void *dest, size_t dest_len, const void *src, size_t count,
  * Note that sizes of @dest and @src must be known at compile-time.
  */
 #define memtostr(dest, src)	do {					\
-	const size_t _dest_len = __builtin_object_size(dest, 1);	\
+	const size_t _dest_len = __must_be_char_array(dest) +		\
+				ARRAY_SIZE(dest);			\
 	const size_t _src_len = __builtin_object_size(src, 1);		\
 	const size_t _src_chars = strnlen(src, _src_len);		\
 	const size_t _copy_len = min(_dest_len - 1, _src_chars);	\
@@ -478,7 +481,8 @@ void memcpy_and_pad(void *dest, size_t dest_len, const void *src, size_t count,
  * Note that sizes of @dest and @src must be known at compile-time.
  */
 #define memtostr_pad(dest, src)		do {				\
-	const size_t _dest_len = __builtin_object_size(dest, 1);	\
+	const size_t _dest_len = __must_be_char_array(dest) +		\
+				ARRAY_SIZE(dest);			\
 	const size_t _src_len = __builtin_object_size(src, 1);		\
 	const size_t _src_chars = strnlen(src, _src_len);		\
 	const size_t _copy_len = min(_dest_len - 1, _src_chars);	\
Signature · 1 line
2.34.1
#13 lore raw
In reply to Kees Cook
Quoted 4 lines · Kees Cook

Work around a Clang 14 bug by switching to ARRAY_SIZE() (with the added benefit of explicitly checking for char array arguments) in memtostr*/strtomem*().

What's the minimum Clang version we build kernel with? 12?

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#14 lore raw
Quoted 2 lines · Andy Shevchenko

What's the minimum Clang version we build kernel with? 12?

13.0.1 for most architectures according to scripts/min-tool-version.sh.

Cheers,
Miguel

#15 lore raw
Quoted 7 lines · Miguel Ojeda
Quoted 2 lines · Andy Shevchenko

What's the minimum Clang version we build kernel with? 12?

13.0.1 for most architectures according to scripts/min-tool-version.sh.

Okay, does it mean 13 has no such a bug?
Otherwise the commit message and other comments might be clearer...

Signature · 2 lines
With Best Regards,
Andy Shevchenko
author maintainer
#16 lore raw
Quoted 11 lines · Andy Shevchenko
Quoted 7 lines · Miguel Ojeda
Quoted 2 lines · Andy Shevchenko

What's the minimum Clang version we build kernel with? 12?

13.0.1 for most architectures according to scripts/min-tool-version.sh.

Okay, does it mean 13 has no such a bug?

It probably did, but Clang 13 didn't support compile-time error messages (added in 14). Regardless, this fixes it there and makes the API more robust.

Quoted 1 line

Otherwise the commit message and other comments might be clearer...

How should I rephrase things? I mention the bug and when it was fixed in patch 3:

...
benefit of working around a bug in Clang (fixed[1] in 15+) that got
...

-Kees

Signature · 1 line
Kees Cook
#17 lore raw
Quoted 16 lines · Kees Cook
Quoted 10 lines · Andy Shevchenko
Quoted 6 lines · Miguel Ojeda
Quoted 2 lines · Andy Shevchenko

What's the minimum Clang version we build kernel with? 12?

13.0.1 for most architectures according to scripts/min-tool-version.sh.

Okay, does it mean 13 has no such a bug?

It probably did, but Clang 13 didn't support compile-time error messages (added in 14). Regardless, this fixes it there and makes the API more robust.

Quoted 8 lines
Quoted 1 line

Otherwise the commit message and other comments might be clearer...

How should I rephrase things? I mention the bug and when it was fixed in patch 3:

...
benefit of working around a bug in Clang (fixed[1] in 15+) that got
...

Ah, the version is mentioned only in cover letter. So, the patch's commit message seems good to me.

Signature · 2 lines
With Best Regards,
Andy Shevchenko