Problem with disklabel and filesystem over iSCSI

[Danny Braniss]

> ok, i started the system in single-user-mode and brought up the iSCSI 
> connection. I tried to install the ufs first, after each step I synced. 
>   No change - problem remains the same (newfs is failing).
> 
> Then I installed the ZFS again, also synced after each step, then 
> unmounted the iSCSI device, synced again (always at least three times). 
> Then rebooted the system (going to single-user-mode again) and the ZFS 
> is corrupted again with the same message.
> 
in the case of zfs, you have to
	 /etc/rc.d/zfs stop

> I don't think the buffers are the problem.
> 
are you sure that the iscsi target is only used by the freebsd host?

> Erik
> 
> -----
> ArgonSoft GmbH      | Im Ermlisgrund 3      | 76337 Waldbronn
> Tel: +49 7243 71520 | Fax: +49 7243 715222  | http://www.argonsoft.de
> Umsatzsteuer-Identnummer: DE205762306       | Handelsregister: HRB2372E
> Geschäftsführer: Erik Scholtz
> 
> Danny Braniss wrote:
>  >
>  > ok, so the problem is on shutdown/reboot. Buffers don't seem to be
>  > flushed.
>  > To check if this is correct, try shutdown, then under single user
>  > unmount the iscsi, sync, sync, reboot
>  >
>  > danny
>  >
> >> Hi,
> >>
> >> I tested the default iSCSI_initiator shipped with 7.0, 7.1RC1 and 
> >> 7.1RC2. Additionally i changed it with that version from 
> >> ftp://ftp.cs.huji.ac.il/users/danny/freebsd/iscsi-2.1.tar.gz with each 
> >> release. The effect is with all combinations the same:
> >>
> >> 7.0 native / 7.0 + iscsi-2.1
> >> 7.1RC1 native / 7.1RC1 + iscsi-2.1
> >> 7.2RC2 native / 7.2RC2 + iscsi-2.1
> >>
> >>
> >> Additionally info + tests:
> >> --------------------------
> >> 1) I also checked it for ufs in dangerously dedicated mode - also the 
> >> same effect.
> >>
> >> 2) After the reboot, the iSCSI device is back on /dev/da0 as expected.
> >>
> >> 3) The SAN system is a Hardware-SAN (iStor / GigaStor), that works 
> >> without any problems under Ubuntu, CenOS, MacOS X, Windows and RedHat 
> >> (all tested the last days)
> >>
> >> 4) I could get ufs to work with the following (terribly wrong) partition 
> >> map:
> >>
> >>           0         40         39        -     12     unused        0
> >>          40     409600     409639    da0s1    165    FreeBSD        0
> >>      409640 1928708016 1929117655    da0s2    165    FreeBSD        0
> >> 1929117656     262184 1929379839        -     12     unused        0
> >>
> >> With this partition map, newfs runs without any failure. The filesystem 
> >> is heavily damaged and can be repaired with fsck. After repairing, the 
> >> fs can be mounted and used as normal. But when running a fsck, thousands 
> >> of errors must be corrected and the result is an empty disk again (when 
> >> answering all with YES - I ran it with -y flag, since there are too many 
> >> questions to be answered manually, even when copying only three big 
> >> files).