No DHCP after installing NS on top of Centos

mark_nl · February 7, 2021, 1:53pm

NethServer Version: 7.9.2009
Module: nethserver-release

Pretty sure the default behavior prior to 7.9.2009 was if the single active network interface was configured with DHCP under NetworkManger’s control; system-init kept a DHCP configuration for this interface.

Installing nethserver on top of an updateted CentOS 7.9.2009 minimal installation (proxmox VM) results in a FIXED IP setup.

Did see this behavior on the ARM images too, where it is mitigated by stopping / disabling NetworkManager at the end of system-init.
Can confirm this hack works on x86_64 too (edited nethserver-install for this), this results in a DHCP configuration.

Before filing a issue I wonder if my assumption a DHCP configuration should survive system-init is correct?

(may be related to:)

mark_nl · February 7, 2021, 2:41pm

Another non critical (slightly related) observation seen on ARM before:

.... systemd[1]: Network Manager is not active.
.... systemd[1]: Dependency failed for Network Manager Wait Online.
.... systemd[1]: Job NetworkManager-wait-online.service/start failed with result 'dependency'.

NetworkManager-wait-online.service is not disabled by nethserver-install

giacomo · February 8, 2021, 10:52am

The code that parses the network configuration is this one: https://github.com/NethServer/nethserver-base/blob/master/root/etc/e-smith/events/actions/nethserver-base-initialize-db

It received no changes since 2016.

By the way, if at the end the installation has the correct static IP on the green interface, it’s much better than the dynamic one.

mark_nl · February 8, 2021, 11:22am

I know, tried hard to find the origin. My only clue here is NetworkManager / network.service changed behavior.
EDIT:
Unfortunately my Perl knowledge is to limited to fully grasp the line I suspect not handling the alleged changed behavior of stopping Networkmanager. (my lack of knowledge: where does $ifname get it’s value?)

To be frank I do not have a problem with it because here it does end up with a correct fixed IP configuration.
Have read one report it did not work well on a RPI install, it ended up with a fixed fall-back configuration. And the referenced install on a (Hetzner) vps.

And again, pretty sure it alway’s ended up with DHCP config in the past.

EDIT2: Very old closed issue addressing this in the past:

github.com/NethServer/dev

Base: support DHCP on green interfaces

opened 12:25PM - 06 Apr 16 UTC

closed 02:55PM - 08 Jul 16 UTC

gsanchietti

verified

Scenario: VPS provider which offers simple virtual machine with a network interf…ace configured using DHCP. The system actually supports green interfaces configured only with static IPs: if initialization scripts set a static IP on the previous scenario, the server could be unreachable if the lease expires. See: http://community.nethserver.org/t/after-installation-on-arubacloud-server-is-not-reachable/2847

stephdl · February 8, 2021, 1:28pm

the value of $ifname come from as an argument of the function fetch_ifcfg_props

sub fetch_ifcfg_props
{
    my $ifname = shift;
    my %props = ();

    # Note: command returns empty output if interface is DOWN:
    open(ADDR, '-|', "ip -o -4 address show dev $ifname");

so check the line nethserver-base/root/etc/e-smith/events/actions/nethserver-base-initialize-db at bceaf5457b92faeea214e6b1c90db718086b6bda · NethServer/nethserver-base · GitHub

shift in perl takes the first arg: shift in Perl

if we would like to assign different arguments we would like to use Processing command line arguments - @ARGV in Perl

something like :

sub functionToDoSomething('toto','25') {
 my ($name, $age) = @ARGV;

or

sub functionToDoSomething('toto','25') {
my $name = shift;
my $age = shift;

mark_nl · February 8, 2021, 3:12pm

Just thinking out loud here

See output of (this is the real output it did not truncate, linefeed at \)

# systemctl status NetworkManager
● NetworkManager.service - Network Manager
   Loaded: loaded (/usr/lib/systemd/system/NetworkManager.service; enabled; vendor preset: enabled)
   Active: active (running) since Mon 2021-02-08 09:43:44 EST; 30s ago
     Docs: man:NetworkManager(8)
 Main PID: 11224 (NetworkManager)
   CGroup: /system.slice/NetworkManager.service
           ├─11224 /usr/sbin/NetworkManager --no-daemon
           └─11241 /sbin/dhclient -d -q -sf /usr/libexec/nm-dhcp-helper -pf /var/run/dhclient-eth0.pid -lf /var/lib/NetworkManager/dhclient-7fd770b9-4431-43f8-b1f1-01d13608a404-eth0.lease...

<OUTPUT TRUNCATED>

# ip -o -4 address show dev eth0
2: eth0    inet 10.0.0.158/24 brd 10.0.0.255 scope global noprefixroute dynamic eth0\       valid_lft 86362sec preferred_lft 86362sec
# ps -A -o cmd | grep -E '(/| )dhclient .' | grep -q eth0
# echo $?
0


# systemctl stop NetworkManager

# ip -o -4 address show dev eth0
2: eth0    inet 10.0.0.158/24 brd 10.0.0.255 scope global noprefixroute dynamic eth0\       valid_lft 86334sec preferred_lft 86334sec
# ps -A -o cmd | grep -E '(/| )dhclient .' | grep -q eth0
# echo $?
1

With NM running
ps -A -o cmd | grep -E '(/| )dhclient .' | grep -q eth0
exit’s 0, which can explain why the hack stop NM at the end of system-init works

Also note the place of dynamic in the first command,
If I count the output arguments it in in the 10th position. here :

my($id, $dev, $family, $cidr, $brd_family, $brd_address, $scope_lbl, $scope, $dynamic, $end) = split(/\s+/, $_, 10);

(again not knowledgeable with PERL.)
It seems to me dynamic is expected on the 9th position. If this is true this test results in False if NM is not running : $dhclient_running || $dynamic =~ 'dynamic'

EDIT: Why do I see noprefixroute?

Can not read the full content: https://access.redhat.com/solutions/4275681

filippo_carletti · February 8, 2021, 10:22pm

It is being added by NM version 1.10 or later (it came with RHEL 7.5).
I think it’s safe to update nethserver-base-initialize-db to get dynamic from 10th position, but I’d ask @giacomo. We could also find a better way to discover if an IP came from dhcp (playing with NM).

mark_nl · February 9, 2021, 6:48am

Thank you for finding this,
looks like that specific line is broken for a while.

If the result of ip -o -4 address show dev $ifname is inconsistent maybe not making assumptions of the output is better after all.

giacomo · February 9, 2021, 7:47am

I can try to fix it, but the code is working since 5 years.

Are you willing to retest the install on all the following?

virtual machine with DHCP
virtual machine without DHCP
Digital Ocean VPS
NethServer image for Digital Ocean Market
Nethesis box

federico.ballarini · February 9, 2021, 8:05am

Come on Giacomo, we know you have nothing to do!

mark_nl · February 9, 2021, 8:06am

Can do this

what about an defensive (a bit dirty) improvement: (add || $end =~ 'dynamic' )

    if ($dhclient_running || $dynamic =~ 'dynamic' || $end =~ 'dynamic' ) {
        $props{'bootproto'} = 'dhcp';
        $props{'gateway'} = '';
    } else {
        $props{'bootproto'} = 'none';
    }

mark_nl · February 9, 2021, 9:16am

can confirm this solves it on my Centos VM, before system-init edited

# cat /etc/e-smith/events/actions/nethserver-base-initialize-db | grep -5 'dynamic'

    # Note: command returns empty output if interface is DOWN:
    open(ADDR, '-|', "ip -o -4 address show dev $ifname");
    while(<ADDR>) {
        chomp;
        my($id, $dev, $family, $cidr, $brd_family, $brd_address, $scope_lbl, $scope, $dynamic, $end) = split(/\s+/, $_, 10);
        my $dhclient_running = (system("ps -A -o cmd | grep -E '(/| )dhclient .' | grep -q $ifname") == 0);

        if ($dhclient_running || $dynamic =~ 'dynamic' || $end =~ 'dynamic') {
            $props{'bootproto'} = 'dhcp';
            $props{'gateway'} = '';
        } else {
            $props{'bootproto'} = 'none';
        }

Have a dhcp setup after system-init

# db networks show
eth0=ethernet
    bootproto=dhcp
    gateway=
    ipaddr=10.0.0.158
    netmask=255.255.255.0
    role=green
ppp0=xdsl-disabled
    AuthType=auto
    FwInBandwidth=
    FwOutBandwidth=
    Password=
    name=PPPoE
    provider=xDSL provider
    role=red
    user=

EDIT:
It does not break if network is brought up without NM and noprefixroute is not added to the interface:

# ip -o -4 address show dev eth0
2: eth0    inet 10.0.0.158/24 brd 10.0.0.255 scope global dynamic eth0\       valid_lft 86067sec preferred_lft 86067sec

# db networks show
eth0=ethernet
    bootproto=dhcp
    gateway=
    ipaddr=10.0.0.158
    netmask=255.255.255.0
    role=green
ppp0=xdsl-disabled
    AuthType=auto
    FwInBandwidth=
    FwOutBandwidth=
    Password=
    name=PPPoE
    provider=xDSL provider
    role=red
    user=

stephdl · February 9, 2021, 2:26pm

ROTFL

mark_nl · February 9, 2021, 4:02pm

another test in the Centos VM;
without NM fixed IP:

ip -o -4 address show dev eth0
2: eth0    inet 10.0.0.158/24 brd 10.0.0.255 scope global eth0\       valid_lft forever preferred_lft forever

# db networks show
eth0=ethernet
    bootproto=none
    gateway=10.0.0.100
    ipaddr=10.0.0.158
    netmask=255.255.255.0
    role=green
ppp0=xdsl-disabled
    AuthType=auto
    FwInBandwidth=
    FwOutBandwidth=
    Password=
    name=PPPoE
    provider=xDSL provider
    role=red
    user=

Results in a correct FIXED IP configuration.

IMO it is pretty safe to say adding || $end =~ 'dynamic' mitigates NM/DHCP behavior and does not break other network configurations.

giacomo · February 10, 2021, 9:30am

3 posts were split to a new topic: Chat about Perl fun

giacomo · February 10, 2021, 9:34am

Thank you for the effort, the patch has been merged!

Now it comes the hardest part: testing
We also need to build a new iso to check for regressions.

mark_nl · February 10, 2021, 9:41am

Tip for those who want to help with testing,

before running nethserver-install on your CentOS test-setup edit
/etc/yum.repos.d/NethServer.repo
enable nethserver-testing (enabled=1) only for nethserver-base (includepkgs=nethserver-base*):

...
[nethserver-testing]
name=NethServer-$releasever - Testing
mirrorlist=http://mirrorlist.nethserver.org/?release=$releasever&repo=testing&arch=$basearch&nsrelease=$nsrelease
#baseurl=http://mirror.nethserver.org/nethserver/$releasever/testing/$basearch/
gpgcheck=1
repo_gpgcheck=1
gpgkey=file:///etc/pki/rpm-gpg/RPM-GPG-KEY-NethServer-$releasever
enabled=1
enablegroups=0
includepkgs=nethserver-base*

…

giacomo · February 10, 2021, 2:23pm

This is the ISO for testing: http://packages.nethserver.org/6422/nethserver-7.9.2009_issue6422-x86_64.iso

Andy_Wismer · February 10, 2021, 2:51pm

@giacomo

Hi Giacomo

Will try early evening on a VM…

My 2 cents
Andy

giacomo · February 10, 2021, 3:03pm

Thank you! I did a quick test with unattended install and everything seems good