2010-10-26 18:25:07

by ShaunR

[permalink] [raw]
Subject: NFSv4 and portmap

Everything i keep reading tells me that NFS4 does not need portmap but
for some reason on the NFS server ( CentOS 5.x ) i cannot get the nfs
services started unless portmap is running first.

On the server i modified /etc/sysconfig/nfs with the following options
MOUNTD_NFS_V1="no"
MOUNTD_NFS_V2="no"
RPCNFSDARGS="-N 2 -N 3 -U"

For some reason if i uncomment MOUNTD_NFS_v3="no" i get errors.

On the client, again a CentOS 5.x machine but with a newer custom built
kernel, If i do not start portmap then when i run mount it takes about 2
minutes for the filesystem to mount and i see the following messages in
/var/log/messages

Oct 26 10:25:30 host kernel: rpcbind: server localhost not responding,
timed out

If i start portmap, the mount command completed in a few seconds.


My goal here to is run/support NFSv4 Only.

~ShaunR



2010-10-26 20:09:37

by Steve Dickson

[permalink] [raw]
Subject: Re: NFSv4 and portmap



On 10/26/2010 02:18 PM, ShaunR wrote:
> Everything i keep reading tells me that NFS4 does not need portmap but
> for some reason on the NFS server ( CentOS 5.x ) i cannot get the nfs
> services started unless portmap is running first.
>
> On the server i modified /etc/sysconfig/nfs with the following options
> MOUNTD_NFS_V1="no"
> MOUNTD_NFS_V2="no"
> RPCNFSDARGS="-N 2 -N 3 -U"
>
> For some reason if i uncomment MOUNTD_NFS_v3="no" i get errors.
>
> On the client, again a CentOS 5.x machine but with a newer custom built
> kernel, If i do not start portmap then when i run mount it takes about 2
> minutes for the filesystem to mount and i see the following messages in
> /var/log/messages
>
> Oct 26 10:25:30 host kernel: rpcbind: server localhost not responding,
> timed out
>
> If i start portmap, the mount command completed in a few seconds.
>
>
> My goal here to is run/support NFSv4 Only.
I think you are running into a known problem of when the v4
server comes it uses some legacy code that blindly does
"unregister all my previous existences" to rpcbind.
So that unregister hangs for a bit if rpcbind is not
running... This was fixed in later releases...

steved.