[PATCH] lkdtm: Mark lkdtm_rodata_do_nothing() notrace

lkdtm_rodata_do_nothing() is an empty function which is generated in
order to test the non-executability of rodata.

Currently if function tracing is enabled then an mcount callsite will be
generated for lkdtm_rodata_do_nothing(), and it will appear in the list
of available functions for function tracing (available_filter_functions).

Given it's purpose purely as a test function, it seems preferable for
lkdtm_rodata_do_nothing() to be marked notrace, so it doesn't appear as
traceable.

This also avoids triggering a linker bug on powerpc:

  https://sourceware.org/bugzilla/show_bug.cgi?id=20428

When the linker sees code that needs to generate a call stub, eg. a
branch to mcount(), it assumes the section is executable and
dereferences a NULL pointer leading to a linker segfault. Marking
lkdtm_rodata_do_nothing() notrace avoids triggering the bug because the
function contains no other function calls.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
---
 drivers/misc/lkdtm_rodata.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/misc/lkdtm_rodata.c b/drivers/misc/lkdtm_rodata.c
index 166b1db3969f..3564477b8c2d 100644
--- a/drivers/misc/lkdtm_rodata.c
+++ b/drivers/misc/lkdtm_rodata.c

@@ -4,7 +4,7 @@
  */
 #include "lkdtm.h"
 
-void lkdtm_rodata_do_nothing(void)
+void notrace lkdtm_rodata_do_nothing(void)
 {
 	/* Does nothing. We just want an architecture agnostic "return". */
 }

-- 
2.7.4

On Tue, Aug 2, 2016 at 9:59 PM, Michael Ellerman [off-list ref] wrote:

lkdtm_rodata_do_nothing() is an empty function which is generated in
order to test the non-executability of rodata.

Currently if function tracing is enabled then an mcount callsite will be
generated for lkdtm_rodata_do_nothing(), and it will appear in the list
of available functions for function tracing (available_filter_functions).

Given it's purpose purely as a test function, it seems preferable for
lkdtm_rodata_do_nothing() to be marked notrace, so it doesn't appear as
traceable.

This also avoids triggering a linker bug on powerpc:

  https://sourceware.org/bugzilla/show_bug.cgi?id=20428

When the linker sees code that needs to generate a call stub, eg. a
branch to mcount(), it assumes the section is executable and
dereferences a NULL pointer leading to a linker segfault. Marking
lkdtm_rodata_do_nothing() notrace avoids triggering the bug because the
function contains no other function calls.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

Awesome! Thanks for tracking this down. I've applied it to my tree, it
should get picked up by Greg on my next pull request.

-Kees

-- 
Kees Cook
Brillo & Chrome OS Security