"unlocking" stale nfs? adding -t to running nfsd?

[Palle Girgensohn]

Thanks for the reply!

--On Sunday, June 13, 2004 15:00:47 -0500 Dan Nelson 
<dnelson at allantgroup.com> wrote:

> In the last episode (Jun 13), Palle Girgensohn said:
>> I have an nfs mount mounted without -i or -s (stoopid me!), just
>> plain mount server:/fs /lfs. This was over a WAN connection, and of
>> course the connection server<->client broke somehow, and now the
>> mount is stale. This naturally means that I cannot do ls -l / , since
>> it hangs forever. Now the question: is there any way to unstale this,
>> so the machine can go back to normal again, without a reboot?
>
> umount -f /mountpoint, and remount it.  The only thing I know of that
> can cause an entire mountpoint to go stale is if the server gets
> rebooted with a new kernel and it can't determine which filesystem an
> incoming request is for.  Connectivity issues shouldn't cause this.

hmm nfs over WAN genererally sucks... I actually had to reboot the client. 
:(

>> I should really do this mount with tcp, of course, but found no way
>> to get a running nfsd to also start accepting tcp (nfsd runs with "-n
>> 6 -u", no -t). Is there a way to tell a running nfsd to start
>> accepting tcp connections?
>
> Just bounce nfsd after changing nfs_server_flags in rc.conf.

bounce, you mean like kill -USR1 ? Surely, nfsd does not read rc.conf, so 
kill -USR1 #pid && nfsd -t ...? Is that safe when the server has active 
clients?



/Palle