Date: Wed, 15 Jun 2011 07:00:03 -0600 From: "Justin T. Gibbs" <gibbs@scsiguy.com> To: Pawel Jakub Dawidek <pjd@FreeBSD.org> Cc: fs@FreeBSD.org Subject: Re: [CFR][ZFS] Add "zpool labelclear" command. Message-ID: <4DF8ACD3.1070202@scsiguy.com> In-Reply-To: <20110615120524.GI1975@garage.freebsd.pl> References: <4DF7CDD0.8040108@scsiguy.com> <20110615120524.GI1975@garage.freebsd.pl>
next in thread | previous in thread | raw e-mail | index | archive | help
On 6/15/11 6:05 AM, Pawel Jakub Dawidek wrote: > On Tue, Jun 14, 2011 at 03:08:32PM -0600, Justin T. Gibbs wrote: > > ZFS rightfully has a lot of safety belts in place to ward off unintended > > data loss. But in some scenarios, the safety belts are so restrictive, > > the only way to proceed is to wipe the label information off of a drive. > > > > Here's an example: > > > > Pull a drive that is active in a pool on one system and stick it into > > another system. ZFS will correctly reject this drive as a member of > > a new pool or as the argument of a replace command. But if you really > > want to use that drive, how do you clear it's "potentially active" > > status? If the pool were imported, you could destroy it, but ZFS wont > > allow you to import a pool unless there are sufficient members for it > > to serve I/O (I know about the undocumented -F option for import, > > but users aren't going to find that). You can use dd to wipe the label > > data off, but where exactly does ZFS keep its for copies of the label? > > In most of the cases like that you can use -f switch, eg. you can create > pool or replace vdev using one that is active when you use that switch. > I'm sure you are aware of this, so I guess it doesn't always work for you? Most of my testing has been on v15, so perhaps the situation is better on v28? On v15, "replace -f" certainly didn't work. Even if "replace -f" does work in v28 (or is made to work), what would be the correct way to just delete the label off of such a drive in the current zpool command set? At Spectra Logic, we've found it very useful in our drive fault testing to be able to easily restore a drive to an unlabeled state in order to verify that ZFSD does the right thing with both labeled and unlabeled drives. If our use case is considered rare, I don't need to push this change back into FreeBSD. However, a quick search indicates that at least some Solaris users have desired a similar command: http://opensolaris.org/jive/thread.jspa?messageID=462337 -- Justin
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4DF8ACD3.1070202>