Date: Mon, 13 Jul 2015 13:17:51 -0500
From: Ben Myers <bpm@sgi.com>
To: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Linus Torvalds <torvalds@linux-foundation.org>,
        "J. Bruce Fields" <bfields@fieldses.org>, linux-kernel@vger.kernel.org,
        linux-fsdevel@vger.kernel.org
Subject: Re: [RFC] freeing unlinked file indefinitely delayed
Message-ID: <20150713181751.GZ4568@sgi.com>
References: <20150708014237.GC17109@ZenIV.linux.org.uk>
 <20150708154143.GG4015@sgi.com>
 <20150712150035.GJ17109@ZenIV.linux.org.uk>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20150712150035.GJ17109@ZenIV.linux.org.uk>
User-Agent: Mutt/1.5.20 (2009-06-14)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1685
Lines: 38

Hey Al,

On Sun, Jul 12, 2015 at 04:00:35PM +0100, Al Viro wrote:
> On Wed, Jul 08, 2015 at 10:41:43AM -0500, Ben Myers wrote:
> 
> > The bug rings a bell for me so I will stick my neck out instead of
> > lurking.  Don't you need to sample that link count under the filesystems
> > internal lock in order to avoid an unlink/iget race?  I suggest creating
> > a helper to prune disconnected dentries which a filesystem could call in
> > .unlink.  That would avoid the risk of unintended side effects with the
> > d_alloc/d_free/icache approach and have provable link count correctness.
> 
> For one thing, this patch does *not* check for i_nlink at all.

I agree that no checking of i_nlink has the advantage of brevity.
Anyone who is using dentry.d_fsdata with an open_by_handle workload (if
there are any) will be affected.

> For another, there's no such thing as 'filesystems internal lock' for
> i_nlink protection - that's handled by i_mutex...  And what does
> iget() have to do with any of that?

i_mutex is good enough only for local filesystems.
Network/clustered/distributed filesystems need to take an internal lock
to provide exclusion for this .unlink with a .link on another host.
That's where I'm coming from with iget().  

Maybe plumbing i_op.unlink with another argument to return i_nlink is
something to consider?  A helper for the few filesystems that need to do
this might be good enough in the near term.

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