From mboxrd@z Thu Jan 1 00:00:00 1970 From: Al Viro Subject: Re: [PATCH 5/6] vfs: reorganize do_lookup Date: Wed, 4 Apr 2012 04:16:29 +0100 Message-ID: <20120404031629.GB6589@ZenIV.linux.org.uk> References: <1332759264-23533-1-git-send-email-miklos@szeredi.hu> <1332759264-23533-6-git-send-email-miklos@szeredi.hu> <20120330190526.GA29050@ZenIV.linux.org.uk> <87hax1s069.fsf@tucsk.pomaz.szeredi.hu> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org, hch@infradead.org To: Miklos Szeredi Return-path: Content-Disposition: inline In-Reply-To: <87hax1s069.fsf@tucsk.pomaz.szeredi.hu> Sender: linux-kernel-owner@vger.kernel.org List-Id: linux-fsdevel.vger.kernel.org On Tue, Apr 03, 2012 at 10:13:18AM +0200, Miklos Szeredi wrote: > The do_last() reorganization in the atomic-open series needs to be split > up, I realize. Do you have any other high level comments about that > series? Yes. a) Please, pull removal of open-from-d_revalidate() as far in front of the queue as possible, *along* *with* -EOPENSTALE stuff. Without the latter the former is simply broken - we might have hit -ESTALE before and LOOKUP_REVAL might have already been set, so just failing ->open() with ESTALE may end up not repeating it. TBH, had that thing been in front of the queue, I would've put it into the last pull request; that particular idiocy (NFS4 doing open from all methods involved, except for ->open()) had been a serious source of annoyance for a long time. It really needs killing... b) opendata is simply bogus. You need to pass caller-allocated struct file in any case, right? So why not use it to pass what you need to pass? We want to be able to tell "it's a symlink, here's the vfsmount/dentry, now sod off and handle it yourself"? Sure, but struct file *already* contains struct path and that's not something that might disappear on future kernel changes ;-)