From mboxrd@z Thu Jan 1 00:00:00 1970 From: Steven Whitehouse Subject: Re: [Cluster-devel] [RFC 07/11] 9p: Stop using the generic xattr_handler infrastructure Date: Fri, 21 Aug 2015 09:35:56 +0100 Message-ID: <55D6E2EC.6030306@redhat.com> References: <1440094798-1411-1-git-send-email-agruenba@redhat.com> <1440094798-1411-8-git-send-email-agruenba@redhat.com> <20150821064645.GA24276@infradead.org> Mime-Version: 1.0 Content-Type: text/plain; charset=windows-1252; format=flowed Content-Transfer-Encoding: 7bit Cc: "J. Bruce Fields" , cluster-devel@redhat.com, David Quigley , Eric Paris , linux-security-module@vger.kernel.org, Alexander Viro , linux-fsdevel@vger.kernel.org, "Aneesh Kumar K.V" To: Christoph Hellwig , Andreas Gruenbacher Return-path: Received: from mx1.redhat.com ([209.132.183.28]:51931 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753244AbbHUIgB (ORCPT ); Fri, 21 Aug 2015 04:36:01 -0400 In-Reply-To: <20150821064645.GA24276@infradead.org> Sender: linux-fsdevel-owner@vger.kernel.org List-ID: Hi, On 21/08/15 07:46, Christoph Hellwig wrote: > On Thu, Aug 20, 2015 at 08:19:54PM +0200, Andreas Gruenbacher wrote: >> Of all the file systems that use the generic xattr handler infrastructure, 9p >> is the only one that needs dentries inside the xattr file system code. Open >> code the xattr handling code in 9p so that we can then convert the generic code >> to pass down inodes instead of dentries; this actually takes only little >> additional code. > NAK. The rest of the series looks somewhat sensible but this is wrong. > > We need want to converge to one interface, probably more similar to the > handlers, so we need to decided if we want to pass dentries or inodes to > both of them. > > Personally I'm in favor of inodes, and that's how it used to be, but > that makes life more complicated for path based file systems like cifs > or 9p. > We did look at this in detail some time ago, and Al came to the conclusion that 9p and CIFS both required the dentries and that it would not be possible to pass just an inode. In the SELinux refresh case, the dentry is not always available, and I don't think we have any choice about that. I know it is a bit of a mess, so if there is a better way to do it, then that would be good. At the moment though, it is not clear quite how we could get around this issue, Steve.