Re: [PATCH v3 1/2] kcsan: Instrument memcpy/memset/memmove with newer Clang

[Paul E. McKenney]

On Mon, Sep 12, 2022 at 11:45:40AM +0200, Marco Elver wrote:
> With Clang version 16+, -fsanitize=thread will turn
> memcpy/memset/memmove calls in instrumented functions into
> __tsan_memcpy/__tsan_memset/__tsan_memmove calls respectively.
> 
> Add these functions to the core KCSAN runtime, so that we (a) catch data
> races with mem* functions, and (b) won't run into linker errors with
> such newer compilers.
> 
> Cc: stable@xxxxxxxxxxxxxxx # v5.10+
> Signed-off-by: Marco Elver <elver@xxxxxxxxxx>

Queued and pushed, thank you!

							Thanx, Paul

> ---
> v3:
> * Truncate sizes larger than MAX_ENCODABLE_SIZE, so we still set up
>   watchpoints on them. Iterating through MAX_ENCODABLE_SIZE blocks may
>   result in pathological cases where performance would seriously suffer.
>   So let's avoid that for now.
> * Just use memcpy/memset/memmove instead of __mem*() functions. Many
>   architectures that already support KCSAN don't define them (mips,
>   s390), and having both __mem* and mem versions of the functions
>   provides little benefit elsewhere; and backporting would become more
>   difficult, too. The compiler should not inline them given all
>   parameters are non-constants here.
> 
> v2:
> * Fix for architectures which do not provide their own
>   memcpy/memset/memmove and instead use the generic versions in
>   lib/string. In this case we'll just alias the __tsan_ variants.
> ---
>  kernel/kcsan/core.c | 50 +++++++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 50 insertions(+)
> 
> diff --git a/kernel/kcsan/core.c b/kernel/kcsan/core.c
> index fe12dfe254ec..54d077e1a2dc 100644
> --- a/kernel/kcsan/core.c
> +++ b/kernel/kcsan/core.c
> @@ -14,10 +14,12 @@
>  #include <linux/init.h>
>  #include <linux/kernel.h>
>  #include <linux/list.h>
> +#include <linux/minmax.h>
>  #include <linux/moduleparam.h>
>  #include <linux/percpu.h>
>  #include <linux/preempt.h>
>  #include <linux/sched.h>
> +#include <linux/string.h>
>  #include <linux/uaccess.h>
>  
>  #include "encoding.h"
> @@ -1308,3 +1310,51 @@ noinline void __tsan_atomic_signal_fence(int memorder)
>  	}
>  }
>  EXPORT_SYMBOL(__tsan_atomic_signal_fence);
> +
> +#ifdef __HAVE_ARCH_MEMSET
> +void *__tsan_memset(void *s, int c, size_t count);
> +noinline void *__tsan_memset(void *s, int c, size_t count)
> +{
> +	/*
> +	 * Instead of not setting up watchpoints where accessed size is greater
> +	 * than MAX_ENCODABLE_SIZE, truncate checked size to MAX_ENCODABLE_SIZE.
> +	 */
> +	size_t check_len = min_t(size_t, count, MAX_ENCODABLE_SIZE);
> +
> +	check_access(s, check_len, KCSAN_ACCESS_WRITE, _RET_IP_);
> +	return memset(s, c, count);
> +}
> +#else
> +void *__tsan_memset(void *s, int c, size_t count) __alias(memset);
> +#endif
> +EXPORT_SYMBOL(__tsan_memset);
> +
> +#ifdef __HAVE_ARCH_MEMMOVE
> +void *__tsan_memmove(void *dst, const void *src, size_t len);
> +noinline void *__tsan_memmove(void *dst, const void *src, size_t len)
> +{
> +	size_t check_len = min_t(size_t, len, MAX_ENCODABLE_SIZE);
> +
> +	check_access(dst, check_len, KCSAN_ACCESS_WRITE, _RET_IP_);
> +	check_access(src, check_len, 0, _RET_IP_);
> +	return memmove(dst, src, len);
> +}
> +#else
> +void *__tsan_memmove(void *dst, const void *src, size_t len) __alias(memmove);
> +#endif
> +EXPORT_SYMBOL(__tsan_memmove);
> +
> +#ifdef __HAVE_ARCH_MEMCPY
> +void *__tsan_memcpy(void *dst, const void *src, size_t len);
> +noinline void *__tsan_memcpy(void *dst, const void *src, size_t len)
> +{
> +	size_t check_len = min_t(size_t, len, MAX_ENCODABLE_SIZE);
> +
> +	check_access(dst, check_len, KCSAN_ACCESS_WRITE, _RET_IP_);
> +	check_access(src, check_len, 0, _RET_IP_);
> +	return memcpy(dst, src, len);
> +}
> +#else
> +void *__tsan_memcpy(void *dst, const void *src, size_t len) __alias(memcpy);
> +#endif
> +EXPORT_SYMBOL(__tsan_memcpy);
> -- 
> 2.37.2.789.g6183377224-goog
>