Re: 2.6.10-mm3 scaling problem with inotify

[Robert Love <rml@novell.com>]

On Thu, 2005-01-13 at 20:31 -0500, John McCutchan wrote:

> No, you covered things well. This code was really just a straight copy
> of the dnotify code. Rob cleaned it up at some point, giving us what we
> have today. The only fix I can think of is the one suggested by Rob --
> copying the dnotify code again.

Oh, did I undo the dnotify optimization? :-)

Eh, no, looks like not exactly: The old code always did a dget().

John Hawkes: Attached patch implements the dget() optimization in the
same manner as dnotify.  Compile-tested but not booted.

Let me know!

	Robert Love


inotify: don't do dget() unless we have to

 drivers/char/inotify.c |   14 +++++++++++---
 1 files changed, 11 insertions(+), 3 deletions(-)

diff -urN linux-2.6.10-mm3/drivers/char/inotify.c linux/drivers/char/inotify.c
--- linux-2.6.10-mm3/drivers/char/inotify.c	2005-01-13 21:04:53.108971856 -0500
+++ linux/drivers/char/inotify.c	2005-01-13 21:25:20.052448096 -0500
@@ -607,10 +607,18 @@
 				       u32 cookie, const char *filename)
 {
 	struct dentry *parent;
+	struct inode *inode;
 
-	parent = dget_parent(dentry);
-	inotify_inode_queue_event(parent->d_inode, mask, cookie, filename);
-	dput(parent);
+	spin_lock(&dentry->d_lock);
+	parent = dentry->d_parent;
+	inode = parent->d_inode;
+	if (inode->inotify_data) {
+		dget(parent);
+		spin_unlock(&dentry->d_lock);
+		inotify_inode_queue_event(inode, mask, cookie, filename);
+		dput(parent);
+	} else
+		spin_unlock(&dentry->d_lock);
 }
 EXPORT_SYMBOL_GPL(inotify_dentry_parent_queue_event);