27

I'm experiencing a weird issue where my Ubuntu 18.04 (server) box gets issued a wrong IP address during boot from the DHCP server. Running dhclient after boot on the interface results in the right IP being added to the interface.

The DHCP Server is a Windows box where a reservation was manually configured using the MAC address shown by ip addr in ubuntu (without colons):

5: eno4: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000
    link/ether 00:26:b9:82:44:27 brd ff:ff:ff:ff:ff:ff
    inet 10.10.11.162/23 brd 10.10.11.255 scope global dynamic eno4
       valid_lft 689861sec preferred_lft 689861sec
    inet6 fe80::226:b9ff:fe82:4427/64 scope link
       valid_lft forever preferred_lft forever

My 50-courtin-networking.cfg (cloud-init cfg)

network:
  version: 2
  ethernets:

    bcm:
      match:
        name: eno*
      dhcp4: true
      dhcp6: false

Journalctl entries for DHCP:

#journalctl | grep -Ei 'dhcp'`
Jul 12 10:10:56 skprov2 systemd-networkd[1160]: eno1: DHCP lease lost
Jul 12 10:10:57 skprov2 systemd-networkd[1160]: eno4: DHCP lease lost
Jul 12 10:11:00 skprov2 systemd-networkd[1160]: eno1: DHCPv4 address 10.10.11.157/23 via 10.10.10.254
Jul 12 10:11:02 skprov2 systemd-networkd[1160]: eno4: DHCPv4 address 10.10.11.162/23 via 10.10.10.254

Manually calling dhclient after login (verbose):

# dhclient -v eno4
Internet Systems Consortium DHCP Client 4.3.5
Copyright 2004-2016 Internet Systems Consortium.
All rights reserved.
For info, please visit https://www.isc.org/software/dhcp/

Listening on LPF/eno4/00:26:b9:82:44:27
Sending on   LPF/eno4/00:26:b9:82:44:27
Sending on   Socket/fallback
DHCPREQUEST of 10.10.10.40 on eno4 to 255.255.255.255 port 67 (xid=0x4cb8a62d)
DHCPACK of 10.10.10.40 from 10.10.10.10
bound to 10.10.10.40 -- renewal in 294538 seconds.

10.10.10.10 is the correct DHCP server, and 10.10.10.40 is the IP configured on it. On the Windows DHCP, the wrong lease (.162) shows a long "Unique ID" that does not contain any MAC address present on the ubuntu box: 032e827c00020000ab11d0fc617dced58a43

What's the right way to avoid this? Deny leases for the long UID? Where does that UID come from in the first place? The NIC is onboard in a Dell PowerEdge R710 server.

Improve this question

3 Answers 3

Reset to default
47

The cause of the problem is that the built-in network config of Ubuntu 18.04 no longer uses the NIC Mac address as the default id for DHCP requests.

The traditional (and I believe "sensible") behavior can be restored by adding dhcp-identifier: mac to the configuration in the /etc/netplan/xxx.yaml (cloud-init) file as follows:

network:
    renderer: networkd
    version: 2
    ethernets:
        nicdevicename:
            dhcp4: true
            dhcp-identifier: mac

Where "nicdevicename" is the name of your network device

Use 

sudo netplan apply

to try the new configuration. If you get any errors, please note that precise indentation is very important in .yaml files..

Improve this answer
8
  • Nice one. BTW it's documented in the official netplan docs/reference: netplan.io/reference
    – NoMad
    Commented Oct 29, 2018 at 9:24
  • For those who might have different problem than the question: On my work place computer the ethernet was working on and off lately. I was getting proper IP (checked with ip addr), cable was working, I could ping some computers in the network but no route to outside, no one could ping me. The MAC was correctly registered in the DHCP white list and was not banned by any means and yet no way out. This answer solved my problem.
    – Mehrad
    Commented Apr 23, 2019 at 9:42
  • 6
    this is such WTF
    – mati kepa
    Commented Oct 30, 2019 at 11:17
  • 1
    The default is an RFC4361 client-id (stored in a file). The idea is that the device will present itself with the same id to the network, even if the network hardware is changed. Nice idea, but in my opinion this should be optional behaviour, since the mac id was (and hence should remain) the default.
    – anneb
    Commented Jan 8, 2020 at 4:50
  • 6
    How this is supposed to be working with cloning of VMs? Every VM will get the same IP as result
    – Gregory Suvalian
    Commented Feb 23, 2020 at 19:59
8

Denying the lease won't work. There's no way networkd could know why it's being denied, so it won't just magically switch to a different ID type if you do so. You have to do that manually.

If your systemd version is recent enough and if you have direct control over the config files written out by cloud-init, you can tell systemd-networkd to send a MAC-address-based client ID via the *.network file:

[DHCP]
ClientIdentifier=mac

But if you know that systemd-networkd will always be used, you can just assign the correct lease to client ID 032e827c00020000ab11d0fc617dced58a43, because that's what systemd-networkd will always send for that machine. (It generates the ID based on /etc/machine-id.)

Mos DHCP clients, including dhclient, supply a client-ID field of type '01' (MAC-based). Another common type is '00' (domain name). However, by default, systemd-networkd supplies an "opaque" client-ID that was generated from the contents of /etc/machine-id.

According to the DHCP protocol, leases are chosen by client ID first (as long as the client supplies a "client ID" option, which may or may not be MAC-based), then by the MAC address only if the client didn't send an ID.

So when you're configuring a reservation, all good DHCP servers will allow you to enter either the client ID or the MAC address. If you enter just the MAC address, then I suppose that a type-'01' (MAC-based) client ID is automatically implied. There may be a checkbox named "Ignore client ID", which is convenient for you but technically violates the DHCP spec.

(For example, I have two Wi-Fi adapters with different MACs, but I've configured the OS to send the same client ID no matter which adapter is connected. This way I get the same address via both.)

Improve this answer
6
  • 2
    So it was networkd after all... I didn't know networkd doesn't use the MAC address by default and thought maybe this is an ID generated by some Dell Firmware before actually booting for system management (or something). I'm in the process of testing a reservation for that GUID now.
    – NoMad
    Commented Jul 12, 2018 at 10:00
  • 1
    On that note, Dell iDRAC certainly might be doing DHCP, but it has its own MAC address separate from the actual server. (It also boots up as soon as you connect the power, even if the entire server is shut down.)
    – grawity_u1686
    Commented Jul 12, 2018 at 10:25
  • iDRAC currently is (and was) disabled. The ubuntu box now gets the right IP with the reservation for that GUID, but after systemctl restart systemd-networkd none of the interfaces are rechable by ping. networkd seems to screw up routes...
    – NoMad
    Commented Jul 12, 2018 at 10:50
  • Why would the people behind networkd decide to drop using the MAC address as the default ID? Using /etc/machine-id sounds like a bad idea for machines with more than one NIC and for cloned instances? Can anyone point to the rationale behind this decision?
    – anneb
    Commented Oct 23, 2018 at 9:21
  • @anneb: I don't know the rationale. But for cloned instances, just as /etc/ssh/ssh_keys are supposed to be reset after clone, so is the machine-id. (Afaik, systemd is even able to pick up the virtual machine ID from KVM/HyperV/etc.) Networkd's DHCPv4 client ID is generated similarly to DHCPv6 DUIDs, and has the interface ID hashed into it for multi-interface systems.
    – grawity_u1686
    Commented Oct 23, 2018 at 9:26
5

On vSphere it has been noted that, if a template contain the machine-id then any VMs cloned from the template get same ip as DHCP use the machine-id not the MAC address. Solution is to remove the machine-id from the file /etc/machine-id in the template so new machine-id generated during the cloning.

echo -n > /etc/machine-id
Improve this answer
1

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .