Re: Problem with inotify

[John McCutchan <ttb@tentacle.dhs.org>]

On Tue, Jul 05, 2005 at 06:06:56PM +0100, Anton Altaparmakov wrote:
> On Tue, 5 Jul 2005, John McCutchan wrote:
> > On Tue, Jul 05, 2005 at 08:56:15AM +0100, Anton Altaparmakov wrote:
> > > On Mon, 4 Jul 2005, John McCutchan wrote:
> > > > On Mon, 2005-07-04 at 20:09 +0100, Anton Altaparmakov wrote:
> > > > > On Mon, 4 Jul 2005, Daniel Drake wrote:
> > > > > > Anton Altaparmakov wrote:
> > > > > > > )-:  I have addressed the only things I can think off that could cause
> > > > > > > the oops and below is the resulting patch.  Could you please test it?
> > > > > > 
> > > > > > Yeah!! After removing I_WILL_FREE stuff, that fixed both the oops *and* the
> > > > > > hang. Everything works nicely now.
> > > > > 
> > > > > Great!  Thanks a lot for testing!  I will send a patch to Robert and 
> > > > > Andrew in a minute with more comments added.
> > > > 
> > > > Nice work, I am going to have a closer look at the patch soon. Could you
> > > > post the final patch at http://bugzilla.kernel.org/show_bug.cgi?id=4796
> > > 
> > > Thanks.  Now done.  But I am not sure about the white space.  I can't get 
> > > anything sensible out of IE on Mac OS 9 which I am on at the moment.
> > 
> > Originally inotify had 2 functions that handled this. One that would
> > build up a list of inodes to call remove_watch on, the other function
> > would do the actual calling of remove_watch. This mirrored the other
> > unmount paths. I'm wondering if it wouldn't be cleaner to revert back to
> > that way?
> 
> That is certainly possible.  Out of curiosity, how did you anchor the 
> inodes to your private list?  Or did you just have a dynamically allocated 
> array of pointers to inodes?
> 

The only reason I suggested it, is I'm afraid that maybe we are still
missing a corner case. Even with your fix.

We just added a new list_head in the inode struct. If you look at older
versions of inotify, maybe around 0.15 you can see for yourself.

John