[PATCH] Fix -fsanitize=address -fstack-protector* (PR sanitizer/88333)

[PATCH] Fix -fsanitize=address -fstack-protector* (PR sanitizer/88333)

[Jakub Jelinek]

Hi!

The current asan.c code requires that the whole block of vars starts and
ends on ASAN_RED_ZONE_SIZE (i.e. 32 byte) boundary, so that it is on 4 byte
boundary in the shadow memory.  Normally it is, when frame_offset starts at
0, but with -fstack-protector there is the guard above it and in that case
following patch is needed to realign the end of the block.

Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

2018-12-04  Jakub Jelinek  <jakub@redhat.com>

	PR sanitizer/88333
	* cfgexpand.c (expand_stack_vars): If asan_vec is empty, start with
	aligning frame offset to ASAN_RED_ZONE_SIZE bytes.

	* c-c++-common/asan/pr88333.c: New test.

--- gcc/cfgexpand.c.jj	2018-11-30 19:59:59.676789914 +0100
+++ gcc/cfgexpand.c	2018-12-04 11:25:55.549672029 +0100
@@ -1124,6 +1124,11 @@ expand_stack_vars (bool (*pred) (size_t)
 	      && frame_offset.is_constant (&prev_offset)
 	      && stack_vars[i].size.is_constant ())
 	    {
+	      if (data->asan_vec.is_empty ())
+		{
+		  alloc_stack_frame_space (0, ASAN_RED_ZONE_SIZE);
+		  prev_offset = frame_offset.to_constant ();
+		}
 	      prev_offset = align_base (prev_offset,
 					MAX (alignb, ASAN_MIN_RED_ZONE_SIZE),
 					!FRAME_GROWS_DOWNWARD);
--- gcc/testsuite/c-c++-common/asan/pr88333.c.jj	2018-12-04 12:14:44.329877625 +0100
+++ gcc/testsuite/c-c++-common/asan/pr88333.c	2018-12-04 12:14:34.804033303 +0100
@@ -0,0 +1,12 @@
+/* PR sanitizer/88333 */
+/* { dg-do compile { target fstack_protector } } */
+/* { dg-options "-fstack-protector-strong -fsanitize=address" } */
+
+void bar (int *);
+
+void
+foo (void)
+{
+  int c;
+  bar (&c);
+}

	Jakub

Re: [PATCH] Fix -fsanitize=address -fstack-protector* (PR sanitizer/88333)

[Jeff Law]

On 12/4/18 4:32 PM, Jakub Jelinek wrote:
> Hi!
> 
> The current asan.c code requires that the whole block of vars starts and
> ends on ASAN_RED_ZONE_SIZE (i.e. 32 byte) boundary, so that it is on 4 byte
> boundary in the shadow memory.  Normally it is, when frame_offset starts at
> 0, but with -fstack-protector there is the guard above it and in that case
> following patch is needed to realign the end of the block.
> 
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
> 
> 2018-12-04  Jakub Jelinek  <jakub@redhat.com>
> 
> 	PR sanitizer/88333
> 	* cfgexpand.c (expand_stack_vars): If asan_vec is empty, start with
> 	aligning frame offset to ASAN_RED_ZONE_SIZE bytes.
> 
> 	* c-c++-common/asan/pr88333.c: New test.
OK
jeff