[Fwd: Re: [patch] inotify: lock avoidance with parent watch status in dentry]

[Stephen Smalley]

FYI.

-------- Forwarded Message --------
From: Andrew Morton <akpm at osdl.org>
To: Nick Piggin <nickpiggin at yahoo.com.au>
Cc: holt at sgi.com, john at johnmccutchan.com, linux-kernel at vger.kernel.org,
rml at novell.com, hch at lst.de, linux-fsdevel at vger.kernel.org, Al Viro
<viro at ftp.linux.org.uk>
Subject: Re: [patch] inotify: lock avoidance with parent watch status in
dentry
Date: 	Mon, 27 Feb 2006 16:48:12 -0800

Nick Piggin <nickpiggin at yahoo.com.au> wrote:
>
>  Previous inotify work avoidance is good when inotify is completely
>  unused, but it breaks down if even a single watch is in place anywhere
>  in the system. Robin Holt notices that udev is one such culprit - it
>  slows down a 512-thread application on a 512 CPU system from 6 seconds
>  to 22 minutes.

A problem is that the audit tree (believe it or not) adds a pile of new
inotify functionality.  I don't know what those changes do and they might
conflict with the changes you've made (apart from giving us two copies of
inotify_inode_watched()) and the audit changes were apparently only
socialised on the linux-audit mailing list and my twice-sent patch to make
the audit tree compile has been ignored for a couple of weeks.

So I'm going to bitbucket the audit tree until a) it compiles and b) its
inotify changes have been explained and reviewed and c) we've reviewed
those changes against your optimisations.  I think fixes will be needed.