Re: RFC: Reproducible oops with lockdep on count_matching_names()

[Michael Wu <flamingice@sourmilk.net>]

[-- Attachment #1: Type: text/plain, Size: 1631 bytes --]

On Thursday 01 November 2007 15:17:16 Luis R. Rodriguez wrote:
> mcgrof@pogo:~/devel/wireless-2.6$ git-describe
> v2.6.24-rc1-146-g2280253
>
> So I hit segfault with lockdep on count_matching_names() on the
> strcmp() multiple times now. This is reproducible and with different
> wireless drivers.
>
I've found the problem. It appears to be in lockdep. struct lock_class has a 
const char *name field which points to a statically allocated string that 
comes from the code which uses the lock. If that code/string is in a module 
and gets unloaded, the pointer in |name| is no longer valid. Next time this 
field is dereferenced (count_matching_names, in this case), we crash.

The following patch fixes the issue but there's probably a better way.

-Michael Wu

---

diff --git a/include/linux/lockdep.h b/include/linux/lockdep.h
index 4c4d236..2aa0d35 100644
--- a/include/linux/lockdep.h
+++ b/include/linux/lockdep.h
@@ -114,7 +114,7 @@ struct lock_class {
 	 */
 	unsigned long			ops;
 
-	const char			*name;
+	char				name[128];
 	int				name_version;
 
 #ifdef CONFIG_LOCK_STAT
diff --git a/kernel/lockdep.c b/kernel/lockdep.c
index 55fe0c7..63c4d8f 100644
--- a/kernel/lockdep.c
+++ b/kernel/lockdep.c
@@ -768,7 +768,7 @@ register_lock_class(struct lockdep_map *lock, unsigned int 
subclass, int force)
 	class = lock_classes + nr_lock_classes++;
 	debug_atomic_inc(&nr_unused_locks);
 	class->key = key;
-	class->name = lock->name;
+	strcpy(class->name, lock->name);
 	class->subclass = subclass;
 	INIT_LIST_HEAD(&class->lock_entry);
 	INIT_LIST_HEAD(&class->locks_before);

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 194 bytes --]