Re: [PATCH v6 1/4] xattr: extract simple_xattr code from tmpfs

[Aristeu Rozanski <aris-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>]

On Wed, Aug 22, 2012 at 01:25:06PM -0700, Hugh Dickins wrote:
> > > Probably more important would be to remove spin_lock() and spin_unlock()
> > > (and INIT_LIST_HEAD) from simple_xattrs_free() - those are unnecessary
> > > in shmem_evict_inode(), and wouldn't they be unnecessary whenever
> > > simple_xattrs_free() gets called?
> > 
> > Removing INIT_LIST_HEAD() it's possible by actually unlinking each xattr
> > inside the loop before freeing them. still, it'll have to check if the list is
> > empty or not, which might end up being the same?
> > 
> > About the locking, I'm not sure, I'm investigating it.
> 
> I think we have a misunderstanding.
> 
> INIT_LIST_HEAD() is not expensive, I just meant to remove it because
> I thought it unnecessary by that point.

ah, I see.

> Do you envisage anywhere that would call simple_xattrs_free() except
> a filesystem's evict_inode()?

cgroup does it differently and it's called in d_iput() path (see cgroup_diput),
because it needs to selectively remove files upon remount.

> By that point, the inode is on its way out of the system: nothing
> much (yes, I am being a bit vague there ;) can get to it any more,
> there's no need to reinitialize the list head and there's no need for
> locking, because nothing else can be playing with those xattrs now.

I agree with you. That's why I'm looking into it because I'm pretty sure
I removed it at some point in the past and decided to put it back after
investigating the easily reproducible oops. Sadly I managed to forget
the analisys I did at the time.

-- 
Aristeu