Paul Wallace wrote: > > > Evan Layton wrote: >> Can you provide the output from >> zfs list >> zfs list -o name,mountpoint,mounted > I also include 'beadm list' which shows that "opensolaris-2" > is the active BE. "opensolaris-3" was the new one created during > image-update but it was destroyed following failed update. > > Looks like "opensolaris2" lives in a strange place "/tmp/mnt3509" > which I guess was created when I performed the interim > update to b101b a while ago (which has since been very stable)
I'm not sure what may have mounted it at /tmp/mnt3509 but normally the temporary mountpoint will be something more like "/tmp/.be.LQaGAd". > > I am guessing that this causes "opensolaris-3" to be mounted in the > wrong place which then fails at the image-update step - hence suggestion > that I try to build a new BE manually in a sensible place. > (although it looks like it tried to mount in "/tmp/.be.LQaGAd" > which seems sensible enough > The problem is since /opt in the currently active BE is not empty it won't be empty in the new BE since it's a clone. Because it's not empty the rpool/ROOT/opensolaris-3/opt can't be mounted there since it's relative to the mountpoint of the root dataset (rpool/ROOT/opensolaris-3 mounted at /tmp/.be.LQaGAd) > > # beadm list > BE Active Mountpoint Space Policy Created > -- ------ ---------- ----- ------ ------- > opensolaris - - 10.23M static 2008-06-03 12:44 > opensolaris-1 - - 87.43M static 2008-06-25 17:42 > opensolaris-2 NR / 12.15G static 2008-12-08 13:41 > > # zfs list > NAME USED AVAIL REFER MOUNTPOINT > rpool 24.1G 32.5G 62K /rpool > rpool/ROOT 9.98G 32.5G 18K /rpool/ROOT > rpool/ROOT/opensolaris 9.71M 32.5G 2.46G legacy > rpool/ROOT/opensolaris-1 86.9M 32.5G 2.81G legacy > rpool/ROOT/opensolaris-1/opt 515K 32.5G 516M /opt > rpool/ROOT/opensolaris-2 9.89G 32.5G 5.77G /tmp/mnt3509 > rpool/ROOT/opensolaris-2/opt 932M 32.5G 484M /tmp/mnt3509/opt Ok it is what I was referring to and 1156 is the cause. You can see that the mountpoints for the currently active BE are not correct and you can see below that rpool/ROOT/opensolaris-2/opt is not mounted at all. > rpool/ROOT/opensolaris/opt 531K 32.5G 516M /opt > rpool/export 14.1G 32.5G 19K /export > rpool/export/home 14.1G 32.5G 14.1G /export/home > > # zfs list -o name,mountpoint,mounted > NAME MOUNTPOINT MOUNTED > rpool /rpool yes > rpool/ROOT /rpool/ROOT yes > rpool/ROOT/opensolaris legacy no > rpool/ROOT/opensolaris-1 legacy no > rpool/ROOT/opensolaris-1/opt /opt no > rpool/ROOT/opensolaris-2 /tmp/mnt3509 yes > rpool/ROOT/opensolaris-2/opt /tmp/mnt3509/opt no > rpool/ROOT/opensolaris/opt /opt no > rpool/export /export yes > rpool/export/home /export/home yes > root at paw-opensolaris-2008.05:~# > >> Also if the dataset rpool/ROOT/opensolaris-3/opt is not currently >> mounted is there any data in /opt of your currenty booted BE >> (opensolaris-3). > BE "opensolaris-3" was created and then destroyed during the failed update. > It's not clear why a new BE created by image-update would not be created > correctly and would fail in this way. No obvious problems with disk space > or anything like that. This is because the new BE is a clone of the current BE and the current BE is not correct due to it being left mounted over a reboot. It is a documented requirement that a BE that has been mounted be unmounted before rebooting or booting to that BE. > >> http://defect.opensolaris.org/bz/show_bug.cgi?id=1156 > Don't think this helps, it doesn't get as far as booting, > the image-update step fails. It already had booted the BE that has the problem. You're hitting this because the currectly active BE is the one that has the issue. At some point int eh past opensolaris-2 was mounted at /tmp/mnt3509 and never unmounted, leaving the mountpoint properties set to these temporary mount points. Thankfully this is less of an issue for installations going back to 2008.11 since there is no separate /opt dataset. Due to this these newer installations are much less likely to hit this issue. To fix this we're going to have to play around with the datasets a bit... The following steps should resolve the issue: # beadm activate opensolaris-1 # reboot When booted to opensolaris-1 run: # beadm umount opensolaris-2 # zfs set mountpoint=/a rpool/ROOT/opensolaris-2 # zfs mount rpool/ROOT/opensolaris-2 # zfs set mountpoint=/tmp/mnt rpool/ROOT/opensolaris-2/opt # zfs mount rpool/ROOT/opensolaris-2/opt # cp -R /a/opt/* /tmp/mnt # zfs umount rpool/ROOT/opensolaris-2/opt # zfs set mountpoint=/opt rpool/ROOT/opensolaris-2/opt # mv /a/opt /a/old-opt (so we can still get back to this if we need to) # zfs umount rpool/ROOT/opensolaris-2 # zfs set mountpoint=legacy rpool/ROOT/opensolaris-2 # zfs list (to check that the mount points are now correct) # beadm activate opensolaris-2 # reboot At this point you should be able to do the image-update. -evan
