dhcp-based netboot over jumbo-frame network?

dhcp-based netboot over jumbo-frame network?

[Chris Friesen]

Hi all,

We have an existing system that loads a bare kernel using tftp, then 
mounts a rootfs via NFS.  The wrinkle is that this is on a jumbo-frame 
enabled network, so somehow the MTU needs to be set before mounting the 
rootfs.

In a previous version we patched ipconfig.c to support the DHCP option 
to specify the MTU value, but I'm wondering about the "proper" way to do 
this on current kernels.

I suppose the obvious answer would be initramfs.  The downsides to that 
solution are that it would increase the size of the kernel file, and we 
would need to change the existing build system.  In the face of a small 
patch to ipconfig, it's not a very attractive solution.

Looking forward to your responses.

Chris

Re: dhcp-based netboot over jumbo-frame network?

[Rick Jones]

Chris Friesen wrote:
> 
> Hi all,
> 
> We have an existing system that loads a bare kernel using tftp, then 
> mounts a rootfs via NFS.  The wrinkle is that this is on a jumbo-frame 
> enabled network, so somehow the MTU needs to be set before mounting the 
> rootfs.

Why?  Is the NFS mount not TCP?  If it is TCP, then the TCP MSS exchange will 
provide you cover as it will (should) cause TCP in the NFS server to segment to 
the MSS option your system with the then-smaller MTU sends.

I would think that only if the NFS mount is UDP that you would have a problem 
with the system initially booting with a small MTU.

> In a previous version we patched ipconfig.c to support the DHCP option 
> to specify the MTU value, but I'm wondering about the "proper" way to do 
> this on current kernels.

Was the change submitted to upstream and rejected?

rick jones

Re: dhcp-based netboot over jumbo-frame network?

[Chris Friesen]

Rick Jones wrote:
> Chris Friesen wrote:

>> We have an existing system that loads a bare kernel using tftp, then 
>> mounts a rootfs via NFS.  The wrinkle is that this is on a jumbo-frame 
>> enabled network, so somehow the MTU needs to be set before mounting the 
>> rootfs.
> 
> Why?  Is the NFS mount not TCP?  If it is TCP, then the TCP MSS exchange will 
> provide you cover as it will (should) cause TCP in the NFS server to segment to 
> the MSS option your system with the then-smaller MTU sends.
> 
> I would think that only if the NFS mount is UDP that you would have a problem 
> with the system initially booting with a small MTU.

Currently it uses the default, which is udp.  I might be able to get 
them to try TCP.  Thanks for the suggestion.

>> In a previous version we patched ipconfig.c to support the DHCP option 
>> to specify the MTU value, but I'm wondering about the "proper" way to do 
>> this on current kernels.
> 
> Was the change submitted to upstream and rejected?

The change was submitted (not by us) back in the 2.6.10 timeframe, and 
was rejected with the suggestion to use initramfs.

Chris