[Cc Al and Dave - email thread starts
http://lkml.kernel.org/r/001a113f996099503a055e793...@google.com]
On Tue 21-11-17 20:11:26, Tetsuo Handa wrote:
> On 2017/11/21 16:35, syzbot wrote:
> > Hello,
> >
> > syzkaller hit the following crash on
> > ca91659962303d4fd5211a5e4e13df5cbb11e744
> > git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/master
> > compiler: gcc (GCC) 7.1.1 20170620
> > .config is attached
> > Raw console output is attached.
> >
> > Unfortunately, I don't have any reproducer for this bug yet.
>
> Fault injection found an unchecked register_shrinker() return code.
> Wow, register_shrinker()/unregister_shinker() is possibly frequently called
> path?
>
>
> struct super_block *sget_userns(struct file_system_type *type,
> int (*test)(struct super_block *,void *),
> int (*set)(struct super_block *,void *),
> int flags, struct user_namespace *user_ns,
> void *data)
> {
> (...snipped...)
> spin_unlock(&sb_lock);
> get_filesystem(type);
> register_shrinker(&s->s_shrink); // Error check required.
> return s;
Yes, this is the case since numa aware shrinkers were introduced. I have
a bit hard time to follow the code flow but why cannot we simply
register the shrinker when we allocate the new super block? We
still have the s_umount held so the shrinker cannot race with the
registration code.
Something like the totally untested and possibly wrong
---
diff --git a/fs/super.c b/fs/super.c
index 994db21f59bf..1eb850413fdf 100644
--- a/fs/super.c
+++ b/fs/super.c
@@ -506,6 +506,11 @@ struct super_block *sget_userns(struct file_system_type
*type,
s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
if (!s)
return ERR_PTR(-ENOMEM);
+ if (register_shrinker(&s->s_shrink)) {
+ up_write(&s->s_umount);
+ destroy_super(s);
+ return ERR_PTR(-ENOMEM);
+ }
goto retry;
}
@@ -522,7 +527,6 @@ struct super_block *sget_userns(struct file_system_type
*type,
hlist_add_head(&s->s_instances, &type->fs_supers);
spin_unlock(&sb_lock);
get_filesystem(type);
- register_shrinker(&s->s_shrink);
return s;
}
--
Michal Hocko
SUSE Labs
