Re: [rfc][possible solution] RCU vfsmounts

[Al Viro <viro@ZenIV.linux.org.uk>]

On Sun, Sep 29, 2013 at 07:10:47PM +0100, Al Viro wrote:

> FWIW, right now I'm reviewing the subset of fs code that can be hit in
> RCU mode.  Not a pretty sight, that... ;-/  First catch: in
[snip]

and another, this one completely unrelated to RCU:
                unsigned long gen = (unsigned long) dentry->d_fsdata;
                unsigned long pgen =
                        OCFS2_I(dentry->d_parent->d_inode)->ip_dir_lock_gen;
in ocfs2_dentry_revalidate() needs dentry->d_lock around fetching pgen, as
in the diff below.  I can put it into the next vfs.git pull request
tonight, or pass the buck to ocfs2 folks.  Folks, which tree would
you prefer that to go through?  I'm fine with either variant.  Strictly
speaking, that's -stable fodder, but the race window is quite narrow,
so it's not something earth-shattering...

diff --git a/fs/ocfs2/dcache.c b/fs/ocfs2/dcache.c
index ef99972..0d3a97d 100644
--- a/fs/ocfs2/dcache.c
+++ b/fs/ocfs2/dcache.c
@@ -70,9 +70,10 @@ static int ocfs2_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 	 */
 	if (inode == NULL) {
 		unsigned long gen = (unsigned long) dentry->d_fsdata;
-		unsigned long pgen =
-			OCFS2_I(dentry->d_parent->d_inode)->ip_dir_lock_gen;
-
+		unsigned long pgen;
+		spin_lock(&dentry->d_lock);
+		pgen = OCFS2_I(dentry->d_parent->d_inode)->ip_dir_lock_gen;
+		spin_unlock(&dentry->d_lock);
 		trace_ocfs2_dentry_revalidate_negative(dentry->d_name.len,
 						       dentry->d_name.name,
 						       pgen, gen);