Re: booting using nfs over tcp
Re: booting using nfs over tcp
- Subject: Re: booting using nfs over tcp
- From: Mike Mackovitch <email@hidden>
- Date: Wed, 6 Oct 2004 19:49:30 -0700
On Thu, Oct 07, 2004 at 03:07:49AM +0200, Maarten Thibaut wrote:
> Hi,
>
> Is it possible to boot the mach kernel using nfs over tcp? We need this
> because the nfs server is an HA cluster (the ip address of the server
> resides on one of the physical boxes in the cluster).
>
> In this setup the UDP replies from the server arrive from a different
> address than the one the nfs call was sent to. The client then responds
> with ICMP packets telling the server it doesn't want the packets.
>
> Can we configure the macosx client to either use TCP or allow the UDP
> replies from a different machine than the one it's contacting?
For normal NFS mounts, you can work around the problem where a cluster
(or any system with multiple IP addresses) responds from a different
address than the address the request was sent to by either using TCP
(mount_nfs -T) or by specifying the UDP "noconn" option (mount_nfs -c).
Unfortunately, when netbooting you do not get a chance to specify any
of these options. The use of UDP is currently hardcoded into the NFS
root mounting code. So you are out of luck at the moment.
A bug has been filed to change the netboot code to attempt TCP mounts
first and fallback to UDP if TCP is not available. This will make
netbooting from clusters work.
Thanks
--macko
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Darwin-kernel mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden