zfs discuss - Jun 2007 - NFS, nested ZFS filesystems and ownership

Hello,
I''m sure there is a simple solution, but I am unable to figure this one
out.

Assuming I have tank/fs, tank/fs/fs1, tank/fs/fs2, and I set sharenfs=on for
tank/fs (child filesystems are inheriting it as well), and I chown
user:group /tank/fs, /tank/fs/fs1 and /tank/fs/fs2, I see:

ls -la /tank/fs
user:group .
user:group fs1
user:group fs2
user:group some_other_file

If I mount server:/tank/fs /tmp/mount from another machine, I see:


ls -la /tmp/mount
user:group .
root:wheel fs1
root:wheel fs2
user:group some_other_file

How can I get user:group to propagate down the nested ZFS filesystem over
NFS?

Thanks,
Marko
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://mail.opensolaris.org/pipermail/zfs-discuss/attachments/20070626/bb59772c/attachment.html>

I figured out how to get it to work, but I still don''t quite understand
it.
The way i got it to work is to zfs unmount tank/fs/fs1 and tank/fs/fs2, and
then it looked like this:

ls -la /tank/fs
user:group .
root:root fs1
root:root fs2

That is, those mountpoints changed to root:root from user:group that was in
effect while it was mounted. This I don''t understand - what is
determining
this? How did zfs know to change this to user:group after zfs mount -a on
local filesystem? Does ZFS inherit parent directory ownership at time of
mounting, regardless of ownership of mountpoint? Does NFS respect ownership
of underlying mountpoint, regardless of how ZFS is mounting it? I would
appreciate an explanation or pointing to appropriate documentation. In any
case, I would expect that reasonable behavour would be for both local ZFS
filesystem hierarchy and the view of the same over NFS to display same
ownership (user:group in question exists on both machines, and client is Mac
OSX 10.4.9)


Marko

On 6/26/07, Marko Milisavljevic <marko at cognistudio.com>
wrote:
>
> Hello,
>
> I''m sure there is a simple solution, but I am unable to figure
this one
> out.
>
>
> Assuming I have tank/fs, tank/fs/fs1, tank/fs/fs2, and I set sharenfs=on
> for tank/fs (child filesystems are inheriting it as well), and I chown
> user:group /tank/fs, /tank/fs/fs1 and /tank/fs/fs2, I see:
>
>
> ls -la /tank/fs
> user:group .
> user:group fs1
> user:group fs2
> user:group some_other_file
>
> If I mount server:/tank/fs /tmp/mount from another machine, I see:
>
>
> ls -la /tmp/mount
> user:group .
> root:wheel fs1
> root:wheel fs2
> user:group some_other_file
>
>
> How can I get user:group to propagate down the nested ZFS filesystem over
> NFS?
>
>
> Thanks,
> Marko
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://mail.opensolaris.org/pipermail/zfs-discuss/attachments/20070626/83746be1/attachment.html>

Well, I didn''t realize this at first because I was testing with newly
empty
directories and sorry about wasting the bandwidth here, but it apprears NFS
is not showing nested ZFS filesystems *at all*, all I was seing is
mountpoints of the parent filesystem, and their changing ownership as server
was mounting and unmounting child filesystems locally.
I see nothing in zfs or mount commands that would allow me to recursively
propagate my NFS mount to display hierarchy created by ZFS on server side.
Is this at all possible or NFS and nested ZFS filesystems don''t mix?

And it seems to be a bug, but unless zfs set sharenfs=on is explicitly set
on a filesystem (even though it is already inherited as "on", and
shows as
such in zfs list -o name,sharenfs), it does not appear in "share"
command
output nor can it be mounted by the NFS client - gives permission error.

Marko

On 6/26/07, Marko Milisavljevic <marko at cognistudio.com>
wrote:
>
> I figured out how to get it to work, but I still don''t quite
understand
> it.
>
> The way i got it to work is to zfs unmount tank/fs/fs1 and tank/fs/fs2,
> and then it looked like this:
>
>
> ls -la /tank/fs
> user:group .
> root:root fs1
> root:root fs2
>
>
> That is, those mountpoints changed to root:root from user:group that was
> in effect while it was mounted. This I don''t understand - what is
> determining this? How did zfs know to change this to user:group after zfs
> mount -a on local filesystem? Does ZFS inherit parent directory ownership
at
> time of mounting, regardless of ownership of mountpoint? Does NFS respect
> ownership of underlying mountpoint, regardless of how ZFS is mounting it? I
> would appreciate an explanation or pointing to appropriate documentation.
In
> any case, I would expect that reasonable behavour would be for both local
> ZFS filesystem hierarchy and the view of the same over NFS to display same
> ownership (user:group in question exists on both machines, and client is
Mac
> OSX 10.4.9)
>
>
> Marko
>
> On 6/26/07, Marko Milisavljevic <marko at cognistudio.com> wrote:
> >
> > Hello,
> >
> > I''m sure there is a simple solution, but I am unable to
figure this one
> > out.
> >
> >
> > Assuming I have tank/fs, tank/fs/fs1, tank/fs/fs2, and I set
sharenfs=on
> > for tank/fs (child filesystems are inheriting it as well), and I chown
> > user:group /tank/fs, /tank/fs/fs1 and /tank/fs/fs2, I see:
> >
> >
> > ls -la /tank/fs
> > user:group .
> > user:group fs1
> > user:group fs2
> > user:group some_other_file
> >
> > If I mount server:/tank/fs /tmp/mount from another machine, I see:
> >
> >
> > ls -la /tmp/mount
> > user:group .
> > root:wheel fs1
> > root:wheel fs2
> > user:group some_other_file
> >
> >
> > How can I get user:group to propagate down the nested ZFS filesystem
> > over NFS?
> >
> >
> > Thanks,
> > Marko
> >
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://mail.opensolaris.org/pipermail/zfs-discuss/attachments/20070626/80920c45/attachment.html>