Re: geom_vinum problems (crashes and lockups)

From: Pawel Jakub Dawidek <pjd_at_FreeBSD.org>
Date: Sat, 26 Jun 2004 18:08:21 +0200
On Sat, Jun 26, 2004 at 05:45:51PM +0200, Lukas Ertl wrote:
+> On Sat, 26 Jun 2004, Pawel Jakub Dawidek wrote:
+> 
+> >On Sat, Jun 26, 2004 at 02:07:03PM +0200, Lukas Ertl wrote:
+> >+> This is a situation that I currently consider 'unsupported'.  The 
+> >problem
+> >+> is that the 'd' partition has no offset, thus it is (from the point of
+> >+> view of a geom_vinum drive geom) the same as the 'c' partition.  This is
+> >+> the first problem.  The second problem is the on-disk format of a
+> >+> bsdlabel.  As far as I understand it, this meta-data is stored inside
+> >+> the first partition, so when you open the drive for writing (i.e.
+> >+> mount, fsck) you trigger a spoil event which cause all kinds of 
+> >confusion
+> >+> for geom_vinum.
+> >
+> >I don't know exactly what problems you got, but maybe it can help you:
+> >To avoid spoiling and metadata changes under me, I'm opening every
+> >provider, which I want to use in geom_mirror (r1w1e1), even if provider
+> >which I create isn't yet opened.
+> 
+> This was a good tip, but unfortunately it still doesn't work out.  I 
+> always run into the 'spoiled but dcr = %d' KASSERT.  Unfortunately, kernel 
+> debugging is currently rather a 'guessing' since kgdb doesn't work.  The 
+> next problem is that you might specify /dev/da1s1a as vinum drive but get 
+> /dev/da1s1c or even /dev/da1s1.  'Classic vinum' looked at the type field 
+> in the disklabel, but we don't do this anymore.

This is a race in GEOM. I'm using a patch for this when I running
geom_mirror. I sent it to phk more than week ago (AFAIR) and I still
didn't get an answer.

Try if it works you:

	http://people.freebsd.org/~pjd/patches/geom_subr.c.16.patch

-- 
Pawel Jakub Dawidek                       http://www.FreeBSD.org
pjd_at_FreeBSD.org                           http://garage.freebsd.pl
FreeBSD committer                         Am I Evil? Yes, I Am!

Received on Sat Jun 26 2004 - 14:08:25 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:37:59 UTC