Subject: Re: NetBSD as a Jumpstart server -- a gotcha!
To: None <email@example.com>
From: James Chacon <firstname.lastname@example.org>
Date: 07/20/2000 11:08:22
Actaully 2.6 does this also (I dug it out once way in the past).
If the client doesn't receive the answer it won't stop, but if your netmask
isn't a classfull one it'll break.
Then again the sun code is completely broke because it sends the ICMP request
to the *install_config* host and not the boot server. The install_config
host could be on another planet for all it should care because by that point
you're on the net with a default route and other such stuff. Asking the
jumpstart server what it's netmask is and setting your local one to match
seems a bit wrong. This is why we completely rewrote the boot process for
>Here's one for the FAQ...
>A co-worker was configuring a NetBSD system to act as a Jumpstart server
>for many Sun Netra servers. He was running into a problem that looked
>very much like an NFS interoperability issue -- that is, the client would
>get partway through initialization, then fail to mount the file system
>that contained the Solaris distribution.
>There was much combing through tcpdump output, scratching of heads, and
>far too much anxiety for thorpej.
>Well, it turns out that Solaris (2.7, at least) sends an ICMP MASK REQUEST
>to the network the server is attached to in order to determine the netmask,
>since Sun, in their great wisdom, has yet to fully embrace the wonder that
>is DHCP. Well, if the client doesn't get a reply to this ICMP message, then
>it stops dead in its tracks.
>So, if you ever have the occasion to configure a NetBSD machine as a
>Jumpstart server, make sure to set net.inet.icmp.maskrepl = 1.
> -- Jason R. Thorpe <email@example.com>