Return-Path: Received: from mx1.redhat.com ([209.132.183.28]:1597 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752722Ab1IVRcQ (ORCPT ); Thu, 22 Sep 2011 13:32:16 -0400 Date: Thu, 22 Sep 2011 13:35:29 -0400 From: Jeff Layton To: Ian Kent Cc: Linus Torvalds , David Howells , miklos@szeredi.hu, viro@zeniv.linux.org.uk, gregkh@suse.de, linux-nfs@vger.kernel.org, leonardo.lists@gmail.com Subject: Re: [PATCH] VFS: Suppress automount on [l]stat, [l]getxattr, etc. Message-ID: <20110922133529.6d3ea8de@barsoom.rdu.redhat.com> In-Reply-To: <1316709935.3346.48.camel@perseus.themaw.net> References: <20110922134510.24683.14576.stgit@warthog.procyon.org.uk> <1316707443.3346.44.camel@perseus.themaw.net> <1316709935.3346.48.camel@perseus.themaw.net> Content-Type: text/plain; charset=US-ASCII Sender: linux-nfs-owner@vger.kernel.org List-ID: MIME-Version: 1.0 On Fri, 23 Sep 2011 00:45:35 +0800 Ian Kent wrote: > On Thu, 2011-09-22 at 09:30 -0700, Linus Torvalds wrote: > > On Thu, Sep 22, 2011 at 9:04 AM, Ian Kent wrote: > > > > > > I haven't checked what the side effects would be but using the > > > LOOKUP_DIRECTORY flag in walks that get caught by the removal of the > > > LOOKUP_FOLLOW test in follow_automount() and retaining Miklos's original > > > patch is probably a more natural solution IMO. > > > > Ok, I do have to agree with that. > > > > So instead of adding a new flag, let's just document a few *logical* > > rules for what causes auto-mounting: > > > > - opening the mount-point itself with LOOKUP_DIRECTORY does so > > > > Logic: when you use LOOKUP_DIRECTORY, you expect to see the > > *contents* of the mount-point. > > > > - looking something up *under* the mount-point does so. > > > > This may be obvious, but it actually has a non-obvious special > > case: what about the pathname "mountpoint" vs "mountpoint/" > > > > And I think the "LOOKUP_DIRECTORY" rule ends up automatically also > > resolving that special case: when we have a slash at the end of the > > last component, it not only implies that we care about the contents, > > it will also automatically set LOOKUP_DIRECTORY. > > > > So I'm getting more and more convinced that LOOKUP_DIRECTORY is > > actually the right thing to trigger on. It automatically means that > > "opendir()" on the mountpoint will do the right thing (because > > O_DIRECTORY results in LOOKUP_DIRECTORY), and it automatically - and > > very naturally - gives user processes the ability to choose whether > > they want to see auto-monting or not ("path" doesn't get auto-mounted, > > but "path/" does). > > > > Yet at the same time it keeps the "stupid default behavior for > > processes that only look at each directory entry and don't even think > > about automounting" be the "don't auto-mount when not necessary" > > behavior. > > > > So: no new flag. Just make nfs4 use LOOKUP_DIRECTORY, and let's add > > big documentation notes about this. > > Yes, that's what I think is best. > > It's late here so I'll survey the code and see if I can find any other > instances of this and post a patch, tomorrow. > > Jeff, could you test adding LOOKUP_DIRECTORY to the walk in > nfs_follow_remote_path() please, just in case I've got this all wrong. > Yep, adding LOOKUP_DIRECTORY also fixes the problem. I'll go ahead and send a patch to Trond... Thanks, -- Jeff Layton