Submitter : nathans Status : open
Assigned Engineer : dxm Priority : 4
*Modified Date : 10/02/00 *Modified User : lord
*Modified User Domain : sgi.com *Description :
stick this in a bug so it isn't forgotten...
On Jul 21, 8:40am, Nathan Scott wrote:
> Subject: Re: LVM vs. kiobuf I/O
> > > - Kernel uuid generation support
> > ...
> ...
>
> I think there's a bunch of stuff which can be rationalised
> there too...
.....
==========================
ADDITIONAL INFORMATION (ADD)
From: lord@xxxxxxx
Date: Oct 02 2000 08:25:03AM
[pvnews version: 1.71]
==========================
> ==========================
> ADDITIONAL INFORMATION (ADD)
> From: dxm@engr (BugWorks)
> Date: Oct 01 2000 07:45:40PM
> ==========================
>
> This idea has come up before, but I'm raising it again so
> we can hopefully get a resolution.
>
> We figure that having the kernel change the uuid of a FS on
> mount is a bad idea. The implementation of that at the moment
> is at least a little bit broken.
>
> I've added a command to xfs_db which lets the user change
> the uuid of a FS from userland, so I'd like to drop the
> kernel equivalent.
>
> Note we still need the uuidtab stuff so we can check for
> duplicate mounts, but instead of forcing a new uuid, we'd
> just fail to mount a duplicate FS.
>
> Any objections to this idea?
This does seem like a reasonable thing to do, the one thing to be
aware of is the way SGI internally uses plexed filesystems to
create new ptools clones: Split a leg off a mirror and remount it
somewhere else as a new filesystem. I think Glen added some code
to the irix xfs_db commands to change the uuid as well - is your's
the same code?
Using the above scenario, the current approach will mount the filesystem
without any extra work, using the proposed change the mount would fail.
However, there were other issues with the plex splitting which led Glen
down the same path.
|