[PATCH v2 2/3] kcsan: Instrument memcpy/memset/memmove with newer Clang

[PATCH v2 2/3] kcsan: Instrument memcpy/memset/memmove with newer Clang

[Marco Elver]

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@vger.kernel.org # v5.10+
Signed-off-by: Marco Elver <elver@google.com>
---
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 | 39 +++++++++++++++++++++++++++++++++++++++
 1 file changed, 39 insertions(+)

diff --git a/kernel/kcsan/core.c b/kernel/kcsan/core.c
index fe12dfe254ec..4015f2a3e7f6 100644
--- a/kernel/kcsan/core.c
+++ b/kernel/kcsan/core.c
@@ -18,6 +18,7 @@
 #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 +1309,41 @@ 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)
+{
+	check_access(s, count, 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)
+{
+	check_access(dst, len, KCSAN_ACCESS_WRITE, _RET_IP_);
+	check_access(src, 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)
+{
+	check_access(dst, len, KCSAN_ACCESS_WRITE, _RET_IP_);
+	check_access(src, 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

Re: [PATCH v2 2/3] kcsan: Instrument memcpy/memset/memmove with newer Clang

[Dmitry Vyukov]

On Fri, 9 Sept 2022 at 09:38, Marco Elver <elver@google.com> 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@vger.kernel.org # v5.10+
> Signed-off-by: Marco Elver <elver@google.com>
> ---
> 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 | 39 +++++++++++++++++++++++++++++++++++++++
>  1 file changed, 39 insertions(+)
>
> diff --git a/kernel/kcsan/core.c b/kernel/kcsan/core.c
> index fe12dfe254ec..4015f2a3e7f6 100644
> --- a/kernel/kcsan/core.c
> +++ b/kernel/kcsan/core.c
> @@ -18,6 +18,7 @@
>  #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 +1309,41 @@ 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)
> +{
> +       check_access(s, count, KCSAN_ACCESS_WRITE, _RET_IP_);

These can use large sizes, does it make sense to truncate it to
MAX_ENCODABLE_SIZE?


> +       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)
> +{
> +       check_access(dst, len, KCSAN_ACCESS_WRITE, _RET_IP_);
> +       check_access(src, 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)
> +{
> +       check_access(dst, len, KCSAN_ACCESS_WRITE, _RET_IP_);
> +       check_access(src, 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
>

Re: [PATCH v2 2/3] kcsan: Instrument memcpy/memset/memmove with newer Clang

[Marco Elver]

On Fri, 9 Sept 2022 at 10:38, Dmitry Vyukov <dvyukov@google.com> wrote:
>
> On Fri, 9 Sept 2022 at 09:38, Marco Elver <elver@google.com> 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@vger.kernel.org # v5.10+
> > Signed-off-by: Marco Elver <elver@google.com>
> > ---
> > 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 | 39 +++++++++++++++++++++++++++++++++++++++
> >  1 file changed, 39 insertions(+)
> >
> > diff --git a/kernel/kcsan/core.c b/kernel/kcsan/core.c
> > index fe12dfe254ec..4015f2a3e7f6 100644
> > --- a/kernel/kcsan/core.c
> > +++ b/kernel/kcsan/core.c
> > @@ -18,6 +18,7 @@
> >  #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 +1309,41 @@ 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)
> > +{
> > +       check_access(s, count, KCSAN_ACCESS_WRITE, _RET_IP_);
>
> These can use large sizes, does it make sense to truncate it to
> MAX_ENCODABLE_SIZE?

Hmm, good point - that way it can still set up watchpoints on them.
I'll do a v3.