kcsan: Instrument memcpy/memset/memmove with newer Clang

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 [email protected] # v5.10+

Signed-off-by Marco Elver <[email protected]>

---
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(+)

kernel/kcsan/core.c +50 -0

index fe12dfe254ec..54d077e1a2dc 100644
@@ -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);

Signature · 1 line

2.37.2.789.g6183377224-goog