From owner-freebsd-current@FreeBSD.ORG Sat Sep 26 07:21:59 2009 Return-Path: Delivered-To: freebsd-current@freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2001:4f8:fff6::34]) by hub.freebsd.org (Postfix) with ESMTP id E71A01065676; Sat, 26 Sep 2009 07:21:59 +0000 (UTC) (envelope-from dfr@rabson.org) Received: from itchy.rabson.org (router.rabson.org [80.177.232.241]) by mx1.freebsd.org (Postfix) with ESMTP id A343C8FC17; Sat, 26 Sep 2009 07:21:59 +0000 (UTC) Received: from [IPv6:2001:470:909f:1:225:ff:feed:9426] (unknown [IPv6:2001:470:909f:1:225:ff:feed:9426]) by itchy.rabson.org (Postfix) with ESMTP id 8F9E15D12; Sat, 26 Sep 2009 08:21:27 +0100 (BST) Mime-Version: 1.0 (Apple Message framework v1076) Content-Type: text/plain; charset=us-ascii; format=flowed; delsp=yes From: Doug Rabson In-Reply-To: Date: Sat, 26 Sep 2009 08:21:27 +0100 Content-Transfer-Encoding: 7bit Message-Id: <290F8258-F9F7-42C8-83EB-88266B2992E4@rabson.org> References: <20090925154125.GD48707@roberto-al.eurocontrol.fr> To: Artem Belevich X-Mailer: Apple Mail (2.1076) Cc: Ollivier Robert , dfr@freebsd.org, freebsd-current@freebsd.org, Stefan Bethke Subject: Re: Booting off raidz *mostly* work X-BeenThere: freebsd-current@freebsd.org X-Mailman-Version: 2.1.5 Precedence: list List-Id: Discussions about the use of FreeBSD-current List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Sat, 26 Sep 2009 07:22:00 -0000 On 25 Sep 2009, at 22:49, Artem Belevich wrote: > Our bootloader implementation does not know how to read gang blocks. > Compare the code from OpenSolaris' variant of grub with what we have > and find the differences: > > Search for zio_read_data here: > http://bitbucket.org/osunix/osunix-gate/src/1cca4617b3a8/usr/src/grub/grub-0.97/stage2/fsys_zfs.c > > And compare with our zio_read here: > http://fxr.watson.org/fxr/source/boot/zfs/zfsimpl.c?im=excerpts#L900 > > We seem to be missing gang block check. That would explain why zfsboot > works only sometimes. If all the files are written using regular > blocks, then we're OK. However, some time later (i.e. during upgrade > or after an edit), some data can end up written to disk in a gang > block and that would make zfsboot choke. This sounds plausible - I never managed to see a gang block while I was working on this.