Dear Maintainer,
The recent change in logging format brought this problem to my attention
again by means of long logcheck mails arriving in my mailbox. I believe
this issue has been around as long as I have been using the setup below.
It's just that I've been ignoring it until now because nothing breaks.
Here's my setup:
- eth0 uses a static IP (10.11.18.2/24)
- eth1 uses a dynamic address from 10.11.43.0/24
- DHCP server lives at 10.11.10.11
$ ip route
default via 10.11.18.1 dev eth0
10.11.18.0/24 dev eth0 proto kernel scope link src 10.11.18.2
10.11.43.0/24 dev eth1 proto kernel scope link src 10.11.43.128
Networking works without any trouble but when time comes up to renew the
lease on eth1, I get reams of messages in my logs. Below is an excerpt,
starting at the previous DHCPREQUEST that got DHCPACK'd until the next
DHCPACK. This cycle repeats every 3.5 hours (corresponding to the time
difference between the expire and renew times in the lease).
Oct 5 03:03:09 helix dhclient[1450]: DHCPREQUEST on eth1 to 10.11.10.11 port 67
Oct 5 03:03:28 helix dhclient[1450]: DHCPREQUEST on eth1 to 255.255.255.255 port 67
Oct 5 03:03:28 helix dhclient[1450]: DHCPACK from 10.11.43.1
Oct 5 03:03:28 helix dhclient[1450]: bound to 10.11.43.128 -- renewal in 5624 seconds.
Oct 5 04:37:12 helix dhclient[1450]: DHCPREQUEST on eth1 to 10.11.10.11 port 67
Oct 5 04:37:18 helix dhclient[1450]: DHCPREQUEST on eth1 to 10.11.10.11 port 67
Oct 5 04:37:24 helix dhclient[1450]: DHCPREQUEST on eth1 to 10.11.10.11 port 67
[ this repeats every 5 to 20 seconds for about 500 times ]
Oct 5 06:33:08 helix dhclient[1450]: DHCPREQUEST on eth1 to 10.11.10.11 port 67
Oct 5 06:33:28 helix dhclient[1450]: DHCPREQUEST on eth1 to 10.11.10.11 port 67
Oct 5 06:33:40 helix dhclient[1450]: DHCPREQUEST on eth1 to 255.255.255.255 port 67
Oct 5 06:33:40 helix dhclient[1450]: DHCPACK from 10.11.43.1
Oct 5 06:33:40 helix dhclient[1450]: bound to 10.11.43.128 -- renewal in 6885 seconds.
The switch from 10.11.10.11 to the 255.255.255.255 broadcast happens at
the first point in time that exceeds the time given in the lease for a
rebind.
My iptables implement a DROP policy but ESTABLISHED,RELATED traffic is
ACCEPTed. I've tried adding explicit rules to ACCEPT all UDP and TCP
traffic on ports 67 and 68 for every interface but that did not change
anything.
I captured network packets with wireshark and can see the DCHPREQUESTs
leave from 10.11.18.2 (as per routing table) and the DCHPACKs come in
from 10.11.10.11. Yet, eth1 does not seem able to renew its lease from
that. When eth1 finally does renew its lease, it is via the DCHPACK
from the subnet's gateway (10.11.43.1). The lease contains 10.11.10.11
as the DCHP server (and 10.11.43.1 as the router).
I would have expected eth1 to be able to renew its lease directly from
10.11.10.11 and not for it to log two hours worth of DHCPREQUESTs before
it is finally able to renew (or rebind?) via the gateway.