stackleak: Register the 'stackleak_cleanup' pass before the '*free_cfg' pass
authorAlexander Popov <alex.popov@linux.com>
Thu, 6 Dec 2018 15:13:07 +0000 (18:13 +0300)
committerKees Cook <keescook@chromium.org>
Thu, 6 Dec 2018 17:10:23 +0000 (09:10 -0800)
Currently the 'stackleak_cleanup' pass deleting a CALL insn is executed
after the 'reload' pass. That allows gcc to do some weird optimization in
function prologues and epilogues, which are generated later [1].

Let's avoid that by registering the 'stackleak_cleanup' pass before
the '*free_cfg' pass. It's the moment when the stack frame size is
already final, function prologues and epilogues are generated, and the
machine-dependent code transformations are not done.

[1] https://www.openwall.com/lists/kernel-hardening/2018/11/23/2

Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Alexander Popov <alex.popov@linux.com>
Signed-off-by: Kees Cook <keescook@chromium.org>
scripts/gcc-plugins/stackleak_plugin.c

index 2f48da98b5d421060da5fc056d1199da22031872..dbd37460c573ec549fe33c7015f77206f4485064 100644 (file)
@@ -363,10 +363,12 @@ __visible int plugin_init(struct plugin_name_args *plugin_info,
                                                PASS_POS_INSERT_BEFORE);
 
        /*
-        * The stackleak_cleanup pass should be executed after the
-        * "reload" pass, when the stack frame size is final.
+        * The stackleak_cleanup pass should be executed before the "*free_cfg"
+        * pass. It's the moment when the stack frame size is already final,
+        * function prologues and epilogues are generated, and the
+        * machine-dependent code transformations are not done.
         */
-       PASS_INFO(stackleak_cleanup, "reload", 1, PASS_POS_INSERT_AFTER);
+       PASS_INFO(stackleak_cleanup, "*free_cfg", 1, PASS_POS_INSERT_BEFORE);
 
        if (!plugin_default_version_check(version, &gcc_version)) {
                error(G_("incompatible gcc/plugin versions"));