From owner-freebsd-current@FreeBSD.ORG Wed Oct 29 10:30:44 2008 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 436DE1065672 for ; Wed, 29 Oct 2008 10:30:44 +0000 (UTC) (envelope-from joe@tao.org.uk) Received: from mailhost.tao.org.uk (tao.uscs.susx.ac.uk [139.184.131.101]) by mx1.freebsd.org (Postfix) with ESMTP id 09E658FC1C for ; Wed, 29 Oct 2008 10:30:43 +0000 (UTC) (envelope-from joe@tao.org.uk) Received: from [192.168.0.80] (unknown [91.84.151.241]) (Authenticated sender: joemail@transwarp.tao.org.uk) by mailhost.tao.org.uk (Postfix) with ESMTP id 313DCA0BC; Wed, 29 Oct 2008 10:13:01 +0000 (GMT) Message-ID: <49083134.9050701@tao.org.uk> Date: Wed, 29 Oct 2008 09:47:32 +0000 From: Dr Josef Karthauser User-Agent: Thunderbird 2.0.0.17 (Windows/20080914) MIME-Version: 1.0 To: Peter Schuller References: <0A8710FF880BBB4687AEC66F9DCF7D0C26B462@SERVER1.emerytelcom.com> <5f67a8c40810142200o6074cd42hec0ff6f2b87d3648@mail.gmail.com> <20081017175424.GA73168@hyperion.scode.org> In-Reply-To: <20081017175424.GA73168@hyperion.scode.org> Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Cc: Zaphod Beeblebrox , Daniel Jameson , freebsd-current@freebsd.org Subject: Re: ZFS attaching direct access drives as Scsibus#target#drive# 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: Wed, 29 Oct 2008 10:30:44 -0000 Peter Schuller wrote: >> glabel can have some funny effects due to the geom taste order. If >> something is part of a mirror (for instance) and the label tastes before the >> mirror, I've seen items of the mirror come up with glabel instead of >> attaching to the mirror. > > Does gmirror perhaps keep meta-data at the beginning of the underlying > device? > > Something which can easily happen is that you have, for example, a > slice covering the entire device (exactly the entire device). You then > glabel the slice, and when you reboot the underlying device is labeled > instead of the slice. Similar issues can happen in other cases, such > as glabel:ing an iSCSI target on another host, if that iSCSI target is > backed by, for example, a zvol. > > For these reasons I always make sure that the "contained" device ends > earlier than the "containing" device (and the other way around). This > way there is never any GEOM (disk, slice, partition, glabel, gmirror, > etc) which starts or ends at the exact same spot as another GEOM. This > removes the ambiguity. > > At least this is my interpretation of the situation; I haven't gone > and proven it. The meta data is kept at the end of the device; I always use the -h flag when labelling to get around this problem, as it records the name of the "contained" device in the meta block and never gets confused that way. Joe