Dear Mr. Elkner,

this is tracked with CR 6891553. There exists an escalation to get
a Solaris 10 patch for this. So it will also get fixed in Nevada.

> --- usr/lib/lu/lumount_zones.orig     Wed Aug  5 01:53:12 2009
> +++ usr/lib/lu/lumount_zones  Mon Oct 19 07:58:26 2009
> @@ -81,7 +81,7 @@
>                   mountopt more
>               do
>                       if lulib_is_subdir $mountpoint "$dir"; then
> -                             lulib_unmount_pathname $mountpoint
> +                             /sbin/umount $mountpoint
>                       fi
>               done
>               rmdir "$dir" || exit 1
>   
the existing IDR is identical to your proposed fix.

Thank you very much
Michael Brug

On 19.10.09 08:53, Jens Elkner wrote:
> Hi,
>
> just a "heads up" to prevent inclusion of LU u8 aka 121430-42/121431-43
> bugs.  The ludo bug wrt. the vfstab is probably already known. However
> there is another bug, which prevents proper BE umounts. I.e. luumount
> and friends will fail, if a dataset is mounted in the global zone and
> inherited to a non-global zone. E.g. for a "lumount $BE /mnt":
>
> pool1/web/foo/sites - /pool1/web/foo/sites zfs - no 
> rw,devices,setuid,nonbmand,exec,xattr,noatime
> /pool1/web/foo/sites - /pool1/zones/foo/root/data/sites lofs - no rw,nodevices
> /pool1/web/foo/sites - /pool1/zones/foo/root/mnt/data/sites lofs - no 
> rw,nodevices
> /pool1/zones/foo-s10u8/root - /pool1/zones/foo/root/mnt lofs - no
>
> The root cause is, that lumount_zones uses "lulib_unmount_pathname $mntpoint"
> instead of /sbin/umount $mntpoint. Because 
> "lulib::lulib_get_zfs_dataset_from_mntpt $mntpoint" now (since the mentioned
> patches) returns the corresponding dataset indeed, this results into a
> "/sbin/zfs umount $dataset" or wrt. example "zfs umount pool1/web/foo/sites"!
>
> zfs seems not to return an error code != 0 - however it isn't checked
> anyway. The consequence is, that the zonepath can not be umounted (in
> the example /pool1/zones/foo/root/mnt) and thus can not be removed.
> This leads to a break and umounting pathes from other non-global zones
> is skipped completely, which leaves a lot of lofs mounted FS and finally
> breaks luumount and friends.
>
> So, please fix before including the changes into Nevada.
>
> Proposed fix:
> http://iws.cs.uni-magdeburg.de/~elkner/luc/lumount_zones-5.10u8.patch
>
> Regards,
> jel.
>   


-- 

Mit freundlichen Gruessen / Best regards

        Michael Brug
--
*******************************************************************
Michael Brug                    TSC Systems Engineer CC Solaris
Solaris Competence Center       Tel: +49-89-46008-3681
Sun Microsystems GmbH           Fax: +49-2102-4511-672
Brandenburger-Str. 2            Michael.Brug at SUN.COM
D-40880 Ratingen                http://www.sun.de
*******************************************************************

Sitz der Gesellschaft: Sun Microsystems GmbH, Sonnenallee 1, D-85551 
Kirchheim-Heimstetten
Amtsgericht M?nchen: HRB 161028
Gesch?ftsf?hrer: Thomas Schr?der, Wolfgang Engels, Wolf Frenkel
Vorsitzender des Aufsichtsrates: Martin H?ring


  • [instal... Jens Elkner
    • [i... Michael Brug - Sun Germany Technical Solution Center - Duesseldorf
      • ... Jens Elkner

Reply via email to