Re: ext3 lockdep warning in 2.6.25-rc6

[David Chinner <dgc@sgi.com>]

On Wed, Apr 02, 2008 at 10:12:15AM +0200, Jan Kara wrote:
> On Tue 01-04-08 17:23:34, Erez Zadok wrote:
>  <snip>
<snip>
> > Jan, I'll be happy to test this, but I don't understand two things about
> > this patch:
> > 
> > 1. Is it safe to unlock and re-lock inode_lock temporarily within the loop?
> > 
> > 2. What's the motivation behind having the second toput_inode pointer?  It
> >    appears that the first iteration through the loop, toput_inode will be
> >    NULL, so we'll be iput'ing a NULL pointer (which is ok).  So you're
> >    trying to iput the previous inode pointer that the list iterated over,
> >    right?  Is that intended?
>   I'll try to explain the locking here:
> 1) We are not allowed to call into __invalidate_mapping_pages() with
> inode_lock held - that it the bug lockdep is complaining about. Moreover it
> leads to rather long waiting times for inode_lock (quite possibly several
> seconds).

When you have tens of millions of cached inodes, it can take somewhat
longer than a few seconds.... :/

Cheers,

Dave.
-- 
Dave Chinner
Principal Engineer
SGI Australian Software Group