One last netbooting issue

ME dugan at passwall.com
Fri May 24 12:38:52 PDT 2002


On Thu, 23 May 2002, Lincoln Peters wrote:
> >From: ME <dugan at passwall.com>
> >If you have control over the boot floppy, try using the netboot stuff and
> >give an explicit arg for where to find the root, but allow it to grab an
> >IP via their DHCP server.
> 
> The boot floppy is set up with the kernel parameters:
> root=/dev/nfs nfsroot=130.157.139.237:/netboot ip=dhcp
> 
> I would think that that would do it.

Correct, that should allow the kernel to get and use an IP address served
to it in the dhcp response.

Why not test something: see if a disk based client running the same dhcp
client service forced its interface down when it is killed (not kill -9)

If killing your dhcp server brings down the interface, that would explain
why "everything stops" after that message appears.

As another test, see what happens when you kill -9 the dhcp client you are
using (not permitting it to exit gracefully). If the session still works
with the interfaces still up, maybe you can alter your dhcp client
shutdown script to killall -9 dhcpclientname and prevent it from changing
anything. Then, when the scrip runs to kill all processes, the dhcp client
is sllready dead, but the interface still stands. Killing remaiing
processes wont kill something that is dead. :-)

(Not relly a hack, more of a kludge.)

> >Then you provide your kernel with a netboot modification to work with the
> >netboot stuff, set it on the server mentioned explcitily from the disk,
> >and then it dl the modified kernel and hands the kernel the IP addressit
> >should use. The client then has no absolute need for DHCP. Of course, if
> >the DHCP Server should try to change the lease on-the-fly and then do
> >something to revoke the license and hand it to some other machine, then
> >you could have problems....
> 
> Hmmm.  I don't have control of the DHCP server, so I can't be sure whether 
> or not it'll try any on-the-fly stuff.
> If the kernel itself is auto-configuring the network as described above, 
> should it even need dhcpcd to be running?

To be nice, what happens if the dhcp server revokes the lease and renews
to another IP address? What happens if the lease time is 10 minutes, and
after no attempt to renew from the client, the server assumes the
unrenewed lease has expired and the associated IP address is free for
reasignment? Then you end up with more than one machine on the network
using the same IP address. (This was the probem/cost I was mentioning
above at the end of that paragraph in not running a dhcp client service
and exists if you use netboot IP handoff or the kernel based autoconf.)

> >I do not recall right now, but I dont think ours use DHCP. But even if
> >they did, the NFS mount is exported from the server as readOnly so there
> >is notstate. The shutdown (desired) for these NC is to just shut them
> >off! (Not state, noit disk corruption because no write acccess! :-)
> 
> I can't get away with read-only access.  Imagine someone spending a class 
> period writing up something in AbiWord or KWord, and then getting a 
> "read-only filesystem" error when he or she clicks on "Save"!

You can make he root fiesystem read only, and then export a different
"/home" that is read-write on a per-client basis. Then, you only need
to sync the /home system and unmount it when shutting down. If completed
early, then any other errors "don't really matter" (within reason :-)

-ME

-----BEGIN GEEK CODE BLOCK-----
Version: 3.12
GCS/CM$/IT$/LS$/S/O$ !d--(++) !s !a+++(-----) C++$(++++) U++++$(+$) P+$>+++ 
L+++$(++) E W+++$(+) N+ o K w+$>++>+++ O-@ M+$ V-$>- !PS !PE Y+ !PGP
t at -(++) 5+@ X@ R- tv- b++ DI+++ D+ G--@ e+>++>++++ h(++)>+ r*>? z?
------END GEEK CODE BLOCK------
decode: http://www.ebb.org/ungeek/ about: http://www.geekcode.com/geek.html




More information about the talk mailing list