zfs discuss - Dec 2008 - [Fwd: Re: [indiana-discuss] build 100 image-update: cannot boot to previous BEs]

zfs folks,

I sent the following to indiana-discuss at opensolaris.org, but perhaps
someone here can get to the bottom of this.  Why must zfs trash my
system so often with this hostid nonsense?  How do I recover from this
situation?  (I have no OpenSolaris boot CD with me at the moment, so
zpool import while booted off of the CD isn''t an option)

-Seb

-------- Forwarded Message --------
From: Sebastien Roy <Sebastien.Roy at Sun.COM>
To: David.Comay at Sun.COM
Cc: Indiana Discuss <indiana-discuss at opensolaris.org>
Subject: Re: [indiana-discuss] build 100 image-update: cannot boot to
previous BEs
Date: Sat, 13 Dec 2008 10:54:34 -0500

David,

On Thu, 2008-10-30 at 19:06 -0700, David.Comay at Sun.COM
wrote:
> > After an image-update to build 100, I can no longer boot to my
previous
> > boot environments.  The system successfully boots into build 100, but
my
> > build <= 99 boot environments all crash when mounting zfs root like
this
> > (pardon the lack of a more detailed stack, I scribbled this on a piece
> > of paper):
> 
> Seb, can you reboot your build 100 BE one additional time?  After you
> do this, the hostid of the system should be restored to what it was
> originally and your build 99 BE should then boot.
While this seemed to work for an update from 99 to 100, I''m having this
same problem again, and this time, it''s not resolvable with subsequent
reboots.

The issue is that I had a 2008.11 BE, and created another BE for
testing.  I rebooted over to this "test" BE and bfu''ed it
with test
archives.  I can boot this "test" BE just fine, and I''m now
done my
testing.  I now can''t boot _any_ of my other BE''s that were
created
prior to the "test" BE, including 2008.11.  They all panic as I
initially described:

<garbage address>
mutex_owner_running()
lookuppnat()
vn_removeat()
vn_remove()
zfs''spa_config_write()
zfs''spa_config_sync()
zfs''spa_open_common()
zfs''spa_open()
zfs''dsl_dlobj_to_dsname()
zfs''zfs_parse_bootfs()
zfs''zfs_mountroot()
rootconf()
vfs_mountroot()
main()
_locore_start()

Is there another way to get my 2008.11 BE back?  Is there a bug filed
for this issue, either with ZFS boot, with bfu, or whatever it is that
decides to trash my system?  The issue was originally described as a
"hostid" issue.  Is panicing the best way to handle whatever problem
this is?

Thanks,
-Seb


_______________________________________________
indiana-discuss mailing list
indiana-discuss at opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/indiana-discuss

On Sat, 2008-12-13 at 12:18 -0500, Sebastien Roy wrote:
> I sent the following to indiana-discuss at opensolaris.org, but perhaps
> someone here can get to the bottom of this.  Why must zfs trash my
> system so often with this hostid nonsense?  How do I recover from this
> situation?  (I have no OpenSolaris boot CD with me at the moment, so
> zpool import while booted off of the CD isn''t an option)
The problem turned out to be not related to the hostid, but because I
had done a "zpool upgrade" while booted into ON build > 103
(bringing
the zpool version to 14).  The zpool version is newer than what is
supported in OpenSolaris, so none of my BEs can boot.

Perhaps it would be nice if "zpool upgrade" could say something like,
"Upgrading the pool will cause other boot environments using this pool
to no longer boot.  Are you sure you want to do this?"

-Seb