Date:	Tue, 29 Jul 2008 21:02:03 -0400
From:	Erez Zadok <ezk@...sunysb.edu>
To:	Andrew Morton <akpm@...ux-foundation.org>
Cc:	Hugh Dickins <hugh@...itas.com>, unionfs@...esystems.org,
	ezk@...sunysb.edu, linux-kernel@...r.kernel.org,
	Bharata B Rao <bharata@...ux.vnet.ibm.com>
Subject: Re: [Unionfs] Re: [PATCH -mm] unionfs: build fixes 

In message <20080729172216.84e958f7.akpm@...ux-foundation.org>, Andrew Morton writes:
> On Tue, 29 Jul 2008 19:12:47 +0100 (BST)
> Hugh Dickins <hugh@...itas.com> wrote:
> 
> > Get unionfs building and working in mmotm with the 2.6.27-rc1 VFS changes:
> > permission() has been replaced by inode_permission() without nameidata arg;
> > unionfs_permission() without nameidata arg; vfs_symlink() without mode arg;
> > LOOKUP_ACCESS no longer defined; and kmem_cache_create() no longer passes
> > kmem_cachep to the init_once() constructor.
> > 
> > Note: while okay for inclusion in -mm for now, unionfs_permission() mods
> > will need review and perhaps correction by Erez: without a nameidata arg,
> > some locking vanishes from unionfs_permission(), and a MNT_NOEXEC check on
> > its lower_inode; I have not studied the VFS changes enough to tell whether
> > that amounts to a real issue for unionfs, or just removal of dead code.
> 
> thanks.
> 
> > This should follow git-unionfs.patch
> > I notice my unionfs-fix-memory-leak.patch
> > and fsstack-fsstack_copy_inode_size-locking.patch
> > are currently commented out, yet I don't recall the
> > mm-commits dispatch rider bringing me a telegram to explain why?
> 
> git-unionfs got commented out because of some upstream git (or build)
> catastrophe.  So its fixes got comemnted out too.  Then git-unionfs was
> restored but I forgot to manually restore the followon fixes.  It
> happens.

Shortly I'm going to post fixes which include Hugh's stuff and more.  Sorry
for the delay.

> I must say that I'm not really sure why we're struggling along with
> unionfs.  Last I heard there were fundamental, unresolveable design
> disagreements with the VFS guys.  Those issues should be where 100% of
> the effort is being devoted, but instead we seem to be cruising along
> in a different direction?

Some of my upcoming patches begin to address this (took longer than
expected):

- extracting all whiteout related code into callable methods in unionfs, so
  that I can "drop in" the new whiteout code that Bharata et al. are
  reportedly working on.  I really hope to see some new whiteout code in -mm
  soon.  Bharata?

- reworking the lookup code to handle vfsmounts: this'll be needed when we
  switch from vfs_* to path_* (Miklos's patches).

As for other fundamental issues, I've been posting some suggestions in
recent months.  For example

- the need for cleaner handling of vma->fault(), a relatively minor patch I
  posted, based on hch's LSF08 suggestions.  Got no response from any of the
  VFS folks.

- a post I made regarding suggestions on how to handle lower f/s changes,
  based on Viro's LSF08 comments: to have a superblock level writers count
  (I suggested that it's a superset of the superblock->s_vfs_rename_mutex,
  and perhaps be elevated to be one).  Again, got no responses from anyone
  on the VFS team.

So I'm not sure how much the VFS guys have time now to review such patches
and help me address these issues.  We can't seem to get through even simpler
issues, nor get simple patches merged (ala the copy_inode_size) despite
repeated attempts.

Suggestions how proceed next are very welcome.

Erez.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists