[Cu-wireless] dhcp bug needs fixing

[David Young]

I need some help solving a bug in our node software.

While a C-U Wireless node boots, it tries to discover an Internet
gateway on each ethernet by running the DHCP client, dhclient, on each
ethernet interface. A node gives the -1 option to dhclient, which
means to try a few times to get a lease, and if no lease is acquired,
exit with code -2.

If dhclient quits with code -2, a node takes it to mean that there is
no gateway on that ethernet---or, at least, there is no DHCP server. So
the node starts a DHCP server on that interface.

There are a few things wrong with this approach.

  1 if a subscriber has not yet plugged his node into his Internet gateway
  before it boots, then his node will choose to run a DHCP server on
  every interface.

  2 if a subscriber's node gets a DHCP lease at boot, but at some time his
  gateway does not renew the node's DHCP lease, then dhclient will quit.
  Once dhclient quits, it does not start up again.

  3 once a node starts the DHCP server on a port, it will not use an
  Internet gateway on that port until it reboots.

Scenario #2 occurred the other day at the IMC, because IMC's router
crashed. Even after it was restored, the wireless network did not
re-establish a route to the Internet, because dhclient had quit.

The upshot is this: a node is inflexible about the time that its Internet
gateway is connected, and it will not tolerate a temporary failure of its
gateway. Also, once an ethernet port is assigned the "subscriber port"
role, it will not change from that role, even if after a gateway w/
DHCP server has been attached.

This problem affects the reliability and usability of nodes.  Can somebody
come up with a solution?

Dave

-- 
David Young             OJC Technologies
dyoung at ojctech.com      Engineering from the Right Brain
                        Urbana, IL * (217) 278-3933