Re: [RFC PATCH v2 4/4] get rid of busy-waiting in shrink_dcache_tree()

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

On Thu, Apr 02, 2026 at 03:49:39PM -0700, Linus Torvalds wrote:

> In fact, the __hlist_del() *without* the init part would work fine for
> me with the comment.
> 
> But it was really the added
> 
>         if (likely(!hlist_unhashed(&dentry->d_alias)))
> 
> that made me go "why is _that_ needed?"
> 
> Afaik, it's a perfectly find hlist entry at that point, and no need to
> add the hlist_unhashed() check. No?

The incremental I've got right now is

diff --git a/fs/dcache.c b/fs/dcache.c
index 09a80f7f3d05..b83c279f34a0 100644
--- a/fs/dcache.c
+++ b/fs/dcache.c
@@ -455,8 +455,15 @@ static void dentry_unlink_inode(struct dentry * dentry)
 
 	raw_write_seqcount_begin(&dentry->d_seq);
 	__d_clear_type_and_inode(dentry);
-	if (likely(!hlist_unhashed(&dentry->d_alias)))
-		__hlist_del(&dentry->d_alias);
+	hlist_del_init(&dentry->d_alias);
+	/*
+	 * dentry becomes negative, so the space occupied by ->d_alias
+	 * belongs to ->waiters now; we could use __hlist_del() instead
+	 * of hlist_del_init(), if not for the stunt pulled by nfs
+	 * dummy root dentries - positive dentry *not* included into
+	 * the alias list of its inode.  Open-coding hlist_del_init()
+	 * and removing zeroing would be too clumsy...
+	 */
 	dentry->waiters = NULL;
 	raw_write_seqcount_end(&dentry->d_seq);
 	spin_unlock(&dentry->d_lock);