Moritz Willers wrote:
> Pavel,
>
> They are followed if the NFS v3 request is the mount request. You'll
> even find it documented in share_nfs(1M):
>
> If the file system being shared is a symbolic link to a
> ...
>
I don't share the link, but the destination of the link is already shared.
> Not every OS does it that way, but it does come in handy at times.
>
Yes.
> Kyle,
>
> from what I understand it is something that stops to exist with NFS
> v4. I've seen it disappear in any vendor's NFS v4 implementation,
> where they used to have that behaviour with NFS v3.
>
So what's the benefit of v4 again? It's only been complicating things
for me. and don't list mirror mounts. Yuck - That's what the automounter
is for.
Now sharemgr won't even let you share a filesystem as -o vers=3. So the
only way to fix this is to globally limit the max version to 3 in
/etc/default/nfs.
> On 9 Dec 2008, at 16:15, Pavel Filipensky wrote:
>> Hi Kyle,
>>
>> symbolic links are never followed on the NFS server, they must be
>> processed on the client.
>> This is part of the NFSv{2,3,4} protocol specification.
>>
I know links within the filesystem are followed by the client, but the
mount protocol has for as long as I can remember (always?) returned a
handle to the destination of the link, when the link was passed in the
request.
I'm talking about:
Server:
/ex/Inst -> /export/Install
share -F nfs -o ... /export/Install
Client:
mount server:/ex/Inst
This Works with version3. When a mount for /ex/Inst is requested, The
mount protocol returns an NFS handle for /export/Install. Once the
client has the handle NFSd doesn't have to know about the link, or that
it was mounted with a different path.
With Version 4 there is no mountd, or mount protocol. I haven't looked
far enough yet to see why the v4 nfsd can't do the same... maybe because
that would make it follow all soft links????
-Kyle
>> to which the symlinks
>> are pointing to. And why nfsv4 client cannot mount them. Check the snoop
>> output.
>>
>> --Pavel
>>
>>
>> Kyle McDonald wrote:
>>> I know it may not be the best idea to do this, but in order to shorten
>>> the length of pathnames in DHCP options I ended up making soflinks to
>>> directories much lower in the filesystem.
>>>
>>> With NFSv3, if the client had permission to mount the direcotry the
>>> link
>>> pointed at, everything was good. Now with NFSv4 as the default, all
>>> those clients are getting denied access.
>>>
>>> Is there some easy way to get the NFSv4 server to follow them?
>>>
>>> I know I can selectively, or globally force shares from this server to
>>> only be vers=3, but I think I'd rather just enable the old behavior if
>>> possible.
>>>
>>> -Kyle
>>>
>>> _______________________________________________
>>> nfs-discuss mailing list
>>> nfs-discuss at opensolaris.org
>>>
>>
>> _______________________________________________
>> nfs-discuss mailing list
>> nfs-discuss at opensolaris.org
>
