I've borked my ZFS system upgrading to -STABLE

[Steve Bertrand]

Glen Barber wrote:
> On Wed, May 27, 2009 at 12:49 PM, Steve Bertrand <steve at ibctech.ca> wrote:

>> However, during the upgrade, something didn't go quite right, and now
>> when I boot, I end up with:
>>
>> /libexec/ld-elf.so.1: Shared object "libc.so.7" not found required by sh
>>
>> ...at the "Enter full path of shell or RETURN for /bin/sh#" prompt.
>>
>> Hitting return repeats the cycle.

> I've seen this before.  My situation was different, as I had just
> installed the machine, finished a 'co' on the source tree and rebuilt
> world/kernel.  Since there was nothing, production-wise, on the
> machine, I thought nothing of it and reinstalled.  Had there been data
> on the machine, I would have dug deeper.

I've ran into issues as such quite a few times in the past, but have
always been able to work around the issue in order to get the problem
resolved (and therefore built up quite a confidence level upgrading
production boxes).

The difference here is ZFS. I am extremely confident that all of the
data is fine, it's just the actual running system that is the problem.
Unfortunately, I have no experience (yet) in getting the pool mounted in
a secondary system to fix/re-upgrade it ;)

>> I can boot the box using alternate media, but I can't figure out the
>> procedure to mount the ZFS pool into the filesystem.
>>
> Can you boot into single-user mode from the USB device?

No. Whether I boot "normally" or to single-user, the effect is the same.

>> This box is a running-backup box, so I'm not overly concerned about the
>> data, but I would like to learn how to properly get to the data if this
>> situation arises in the future.
>>
>> Can someone provide some pointers on how to do this without destroying
>> the pool?
>>
> 
> Unfortunately, I haven't played with ZFS (because of lack of hardware
> to run it on)... My reason for the post was to suggest single-user
> mode, if you haven't already tried.

I appreciate all the feedback I can get, but in this case, I have tried
single-user mode.

> Also, to stop the repeated reboot, you could use /rescue/sh in place
> of /bin/sh  (hopefully).

The box is literally on a bench beside my desk, so... *running over to
try it*...

...it appears as though entering "/rescue/sh" drops me into a prompt.
However, whenever I try to use any other command (eg. df, clear etc),
the same error above appears, but with "...required by df" instead.

...going further, you've now got me smelling progress. Never having had
the need to use anything under /rescue, I can verify that my root
filesystem (ZFS) is mounted by doing:

# /rescue/ls /

My normal approach at this point would be to simply mount the system
under /mnt on an existing running system, but my original problem is
that I don't know how to do that with a ZFS pool.

Do you have any more suggestions as to how I can recover the existing
system from here? (I'm going to have a look as to what tools are
available under rescue, to see if I can simply re-do the upgrade procedure).

Thanks!

Steve
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/x-pkcs7-signature
Size: 3233 bytes
Desc: S/MIME Cryptographic Signature
Url : http://lists.freebsd.org/pipermail/freebsd-stable/attachments/20090527/7fc7dd6c/smime.bin