From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1760155AbYDPUk4 (ORCPT ); Wed, 16 Apr 2008 16:40:56 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1752174AbYDPUks (ORCPT ); Wed, 16 Apr 2008 16:40:48 -0400 Received: from pat.uio.no ([129.240.10.15]:34656 "EHLO pat.uio.no" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752171AbYDPUkr (ORCPT ); Wed, 16 Apr 2008 16:40:47 -0400 Subject: Re: [patch 1/3] NFS: fix potential NULL pointer dereference From: Trond Myklebust To: Cyrill Gorcunov Cc: bfields@fieldses.org, neilb@suse.de, ibm-acpi@hmh.eng.br, len.brown@intel.com, kkeil@suse.de, akpm@linux-foundation.org, linux-kernel@vger.kernel.org In-Reply-To: References: <20080416174421.442716301@gmail.com> <48063bc9.2234440a.747d.09ea@mx.google.com> <1208369492.5376.9.camel@heimdal.trondhjem.org> <20080416181336.GA7657@cvg> <1208372123.5376.43.camel@heimdal.trondhjem.org> Content-Type: text/plain Date: Wed, 16 Apr 2008 16:40:28 -0400 Message-Id: <1208378428.8598.12.camel@heimdal.trondhjem.org> Mime-Version: 1.0 X-Mailer: Evolution 2.12.1 Content-Transfer-Encoding: 7bit X-UiO-Resend: resent X-UiO-Spam-info: not spam, SpamAssassin (score=0.0, required=5.0, autolearn=disabled, none) X-UiO-Scanned: 791CA48C070271DC957EFDF87FE195D375C9D105 X-UiO-SR-test: 76FB64D364420248AA5403F1959829A58147DC44 X-UiO-SPAM-Test: remote_host: 129.240.10.9 spam_score: 0 maxlevel 200 minaction 2 bait 0 mail/h: 640 total 7929005 max/h 8345 blacklist 0 greylist 0 ratelimit 0 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Thu, 2008-04-17 at 00:19 +0400, Cyrill Gorcunov wrote: > Trond, I've just pointed the problem and its solution (which is seems > to be a bit ugly, according to the rest nfs coding principle). So if > you prefer to have such a check in 'walk_path' function - just say me > that. You choose :) Thanks for comments > > So? The defensive coding principle is that you perform validity checks > > when the pointer is created. Otherwise, we could equally well have added > > the NULL deref check to nfs4_path_walk()... No, your fix was correct, it was just incomplete. The point I was making above was that defensive programming means that _all_ these validity/NULL pointer checks should really be done in nfs4_validate_mount_data and nfs_validate_mount_data. We shouldn't rely on checks in other parts of the code. In fact, as an example: it looks to me as if the lack of a nfs_server.hostname, leads to a lack of nfs_client->cl_hostname, which will eventually cause an Oops if you 'cat /proc/fs/nfsfs/servers', or if you hit the printk in nfs_update_inode(), or various other dprintk()s. Trond