On Sat, 2017-03-04 at 11:03 +1100, NeilBrown wrote:

On Wed, Dec 21 2016, Jeff Layton wrote:

quoted

@@ -2072,7 +2093,12 @@ inode_cmp_iversion(const struct inode *inode, const u64 old)
 static inline bool
 inode_iversion_need_inc(struct inode *inode)
 {
-	return true;
+	bool ret;
+
+	spin_lock(&inode->i_lock);
+	ret = inode->i_state & I_VERS_BUMP;
+	spin_unlock(&inode->i_lock);
+	return ret;
 }
 

I know this code gets removed, so this isn't really important.
By why do you take the spinlock here?  What are you racing again?

Thanks,
NeilBrown

I think I was worried about I_VERS_BUMP being set or cleared during an
increment or query. It is quite possible that that spinlock is not
necessary.
-- 
Jeff Layton [off-list ref]