Marius Nnnerich wrote:
> On Tue, Feb 3, 2009 at 2:06 PM, Ivan Voras <ivoras@freebsd.org> wrote:
>> Marius Nnnerich wrote:
>>
>>> I'm not happy with the symlinks either. When someone is manipulating a
>>> partition table she should be able to live with the consequences. I
>>> would rather go for the UUID in UFS header approach if there is enough
>>> room. BTW I implemented GPT UUID glabels a while ago please see:
>>> http://www.freebsd.org/cgi/query-pr.cgi?pr=128398
>> I have a patch for UFS "GUID" labels (not exactly GUIDs, but every UFS
>> file system has a reasonably unique ID associated with it) but have
>> encountered what seems a bug in GEOM slicers - two dev entries pointing
>> to the same device don't work well with orphaning/tasting. Have you
>> encountered something similar perhaps?
>
> I haven't encountered strange behaviour there but I will test it again.
I've read the code in the referenced PR and I see you've taken a
different way to implement it than I did, but the problem might still be
present; try the following sequence of operations:
* create a geom (GPT slice in your case) that has two /dev entries on it
- one from the GPT label and one from GPT GUID
* mount one of those /dev entries (i.e. open it for exclusive access) -
the other one should disappear
* unmount it
* see if the labels are correctly created (tasted) again.
* repeat with the base device (on which GPT is created) - both labels
should be spoiled and detected again
If you see what I'm seeing (labels not correctly tasted) then it could
be a problem. If not, it's my problem :)
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
iD4DBQFJibRBldnAQVacBcgRAoniAJjDbD+x9xWQisUBfYVNtwMiT+iwAJ9nSRvg
ZXQj4lkQ0GxiGtetT+ZYZQ==
=8K4f
-----END PGP SIGNATURE-----