From: Marco Stornelli Subject: Re: [RFC][PATCH] ext2: Resolve i_nlink race in ext2_rename Date: Fri, 25 Feb 2011 08:38:15 +0100 Message-ID: References: <1298528501-32176-1-git-send-email-johunt@akamai.com> <20110224063749.GV22723@ZenIV.linux.org.uk> <20110224112010.GB23042@quack.suse.cz> <4D66BD18.2030207@akamai.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Cc: Jan Kara , Al Viro , "linux-ext4@vger.kernel.org" , "sandeen@redhat.com" , "linux-kernel@vger.kernel.org" , "linux-fsdevel@vger.kernel.org" To: Josh Hunt Return-path: Received: from mail-qy0-f174.google.com ([209.85.216.174]:54742 "EHLO mail-qy0-f174.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755404Ab1BYHiQ (ORCPT ); Fri, 25 Feb 2011 02:38:16 -0500 In-Reply-To: <4D66BD18.2030207@akamai.com> Sender: linux-ext4-owner@vger.kernel.org List-ID: Hi, 2011/2/24 Josh Hunt : > Jan > > I'm not seeing the problem with your patch as was expected since we're > not messing with i_nlink anymore. Al suggested marking the inode as > dirty where we were previously doing the old_inode dec. I believe this > is needed as well since we are updating it's ctime. I've attached a > version marking the inode dirty and it also fixes the comment making > reference to calling inode_dec_link_count(). > > I'm not completely clear on the historical reasons for messing with the > link count of old_inode in the first place. It was just to simulate the > linking and unlinking of the old_inode? > > Thanks > Josh > can we share your test/benchmark? I'd like to add it to my test suite as no-regression test. Marco