fix(netboot): keep EL10 diskless nodes from DHCP on every port

xCAT-server/share/xcat/netboot/imgutils/imgutils.pm

@@ -189,6 +189,53 @@ sub mark_uftp_receiver {
     return !$error;
 }
 
+# The initial ramdisk configures $prinic, and bringing it up again breaks an NFS-based boot.
+# EL10 reads no ifcfg files, so its images get keyfiles.
+sub write_nic_profiles {
+    my ($rootimg_dir, $osver, $prinic, $othernics) = @_;
+    my @othernics = grep { length } split(/,/, $othernics // '');
+    my $scripts = "$rootimg_dir/etc/sysconfig/network-scripts";
+    my $keyfiles = "$rootimg_dir/etc/NetworkManager/system-connections";
+    my $major = el_major_version($osver);
+    my %profiles;
+    if (defined($major) && $major >= 10) {
+        if (-d $keyfiles) {
+            # The keyfiles of an earlier build can name interfaces that this build no longer requests.
+            opendir(my $dh, $keyfiles) or die("Error: cannot read $keyfiles: $!\n");
+            foreach my $old (grep { /^xcat-genimage-.*\.nmconnection$/ } readdir($dh)) {
+                unlink("$keyfiles/$old") or die("Error: cannot remove $keyfiles/$old: $!\n");
+            }
+            closedir($dh);
+            $profiles{"$keyfiles/xcat-genimage-$_.nmconnection"} =
+              "[connection]\nid=$_\ntype=ethernet\ninterface-name=$_\n\n[ipv4]\nmethod=auto\n" for @othernics;
+        }
+    } elsif (-d $scripts) {
+        $profiles{"$scripts/ifcfg-$prinic"} = "ONBOOT=no\nBOOTPROTO=dhcp\nDEVICE=$prinic\n" if $prinic;
+        $profiles{"$scripts/ifcfg-$_"} = "ONBOOT=yes\nBOOTPROTO=dhcp\nDEVICE=$_\n" for @othernics;
+    }
+    foreach my $file (sort keys %profiles) {
+        # NetworkManager ignores a keyfile that other users can read.
+        my $mode = $file =~ /\.nmconnection$/ ? oct('600') : oct('644');
+        my $fh;
+        open($fh, '>', $file) && chmod($mode, $file) && print($fh $profiles{$file}) && close($fh)
+          or die("Error: cannot write $file: $!\n");
+    }
+    return;
+}
+
+# NetworkManager otherwise makes a DHCP profile for every Ethernet port without one.
+sub disable_auto_profiles {
+    my ($rootimg_dir, $osver) = @_;
+    my $major = el_major_version($osver);
+    my $confdir = "$rootimg_dir/etc/NetworkManager/conf.d";
+    return unless defined($major) && $major >= 10 && -d $confdir;
+    my $file = "$confdir/xcat.conf";
+    my $fh;
+    open($fh, '>', $file) && print($fh "[main]\nno-auto-default=*\n") && close($fh)
+      or die("Error: cannot write $file: $!\n");
+    return;
+}
+
 sub varsubinline{
     my $line=shift;
     my $refvardict=shift;

xCAT-server/share/xcat/netboot/rh/genimage

@@ -2166,19 +2166,8 @@ sub generic_post { #This function is meant to leave the image in a state approxi
     print $cfgfile "#Dummy resolv.conf to make boot cleaner";
     close($cfgfile);
 
-    # Create the ifcfg-x file for diskless node. But keep the ONBOOT=no
-    # to skip the break of nfs-based boot
-    if ($prinic) {
-        open($cfgfile, ">", "$rootimg_dir/etc/sysconfig/network-scripts/ifcfg-$prinic");
-        print $cfgfile "ONBOOT=no\nBOOTPROTO=dhcp\nDEVICE=$prinic\n";
-        close($cfgfile);
-    }
-    foreach (split /,/, $othernics) {
-        if (/^$/) { next; }
-        open($cfgfile, ">", "$rootimg_dir/etc/sysconfig/network-scripts/ifcfg-$_");
-        print $cfgfile "ONBOOT=yes\nBOOTPROTO=dhcp\nDEVICE=$_\n";
-        close($cfgfile);
-    }
+    imgutils::write_nic_profiles($rootimg_dir, $osver, $prinic, $othernics);
+    imgutils::disable_auto_profiles($rootimg_dir, $osver);
 
     # securetty not needed on s390x
     if ($arch ne "s390x") {

xCAT-test/unit/imgutils_nic_profiles.t

@@ -0,0 +1,87 @@
+use strict;
+use warnings;
+
+use FindBin;
+use lib "$FindBin::Bin/../lib";
+use lib "$FindBin::Bin/../../perl-xCAT", "$FindBin::Bin/../../xCAT-server/lib/perl";
+use Fcntl qw(S_IMODE);
+use File::Path qw(make_path);
+use File::Slurper qw(read_text write_text);
+use File::Temp qw(tempdir);
+use Test::More;
+use XCAT::Test::File qw(repo_path);
+
+$ENV{XCATROOT} = repo_path('xCAT-server');
+require(repo_path('xCAT-server/share/xcat/netboot/imgutils/imgutils.pm'));
+
+my @both = ('etc/sysconfig/network-scripts', 'etc/NetworkManager/system-connections');
+
+sub image {
+    my $root = tempdir(CLEANUP => 1);
+    make_path(map { "$root/$_" } @_);
+    return $root;
+}
+
+sub mode {
+    my ($file) = @_;
+    return -f $file ? sprintf('%04o', S_IMODE((stat $file)[2])) : 'missing';
+}
+
+sub profiles {
+    my ($root) = @_;
+    return [map { substr($_, length($root) + 1) } sort glob("$root/{$both[0],$both[1]}/*")];
+}
+
+for my $case (['rhels10.1', 'etc/NetworkManager/system-connections'], ['alma10.1', @both]) {
+    my ($os, @dirs) = @$case;
+    my $root = image(@dirs);
+    imgutils::write_nic_profiles($root, $os, 'eth0', 'eth1,,eth2');
+    is_deeply(profiles($root), [map { "$both[1]/xcat-genimage-$_.nmconnection" } qw(eth1 eth2)],
+        "a $os image with " . join(' and ', @dirs) . ' gets keyfiles for the requested interfaces only');
+    for my $nic (qw(eth1 eth2)) {
+        my $file = "$root/$both[1]/xcat-genimage-$nic.nmconnection";
+        is(-f $file ? read_text($file) : '',
+            "[connection]\nid=$nic\ntype=ethernet\ninterface-name=$nic\n\n[ipv4]\nmethod=auto\n",
+            "the $os keyfile for $nic asks for DHCP");
+        is(mode($file), '0600', "only root can read the $os keyfile for $nic");
+    }
+}
+
+my $rebuilt = image($both[1]);
+write_text("$rebuilt/$both[1]/site.nmconnection", "site profile\n");
+imgutils::write_nic_profiles($rebuilt, 'rhels10.1', 'eth0', 'eth1,eth2');
+imgutils::write_nic_profiles($rebuilt, 'rhels10.1', 'eth0', 'eth2');
+is_deeply(profiles($rebuilt), ["$both[1]/site.nmconnection", "$both[1]/xcat-genimage-eth2.nmconnection"],
+    'a rebuilt EL10 image drops the keyfile of an interface that is no longer requested and keeps other profiles');
+
+my $el9 = image(@both);
+imgutils::write_nic_profiles($el9, 'rhels9.6', 'eth0', 'eth1');
+is_deeply(profiles($el9), ["$both[0]/ifcfg-eth0", "$both[0]/ifcfg-eth1"], 'an EL9 image gets ifcfg files only');
+is(read_text("$el9/$both[0]/ifcfg-eth0"), "ONBOOT=no\nBOOTPROTO=dhcp\nDEVICE=eth0\n",
+    'the EL9 interface of the initial ramdisk stays off at boot');
+is(read_text("$el9/$both[0]/ifcfg-eth1"), "ONBOOT=yes\nBOOTPROTO=dhcp\nDEVICE=eth1\n",
+    'the EL9 interface eth1 gets DHCP');
+is(mode("$el9/$both[0]/ifcfg-eth1"), '0644', 'the EL9 ifcfg file keeps its permissions');
+
+for my $os (qw(rhels10.1 rhels9.6)) {
+    my $bare = image('etc');
+    imgutils::write_nic_profiles($bare, $os, 'eth0', 'eth1');
+    ok(!-e "$bare/etc/sysconfig" && !-e "$bare/etc/NetworkManager", "a $os image without either directory gets no profile");
+}
+
+for my $case (['rhels10.1', 1], ['rocky10.2', 1], ['rhels9.6', 0]) {
+    my ($os, $disabled) = @$case;
+    my $root = image('etc/NetworkManager/conf.d');
+    imgutils::disable_auto_profiles($root, $os);
+    my $conf = "$root/etc/NetworkManager/conf.d/xcat.conf";
+    if ($disabled) {
+        is(-f $conf ? read_text($conf) : '', "[main]\nno-auto-default=*\n", "a $os image turns off the automatic DHCP profiles");
+    } else {
+        ok(!-e $conf, "a $os image keeps the NetworkManager defaults");
+    }
+}
+my $without = image('etc');
+imgutils::disable_auto_profiles($without, 'rhels10.1');
+ok(!-e "$without/etc/NetworkManager", 'an image without NetworkManager gets no NetworkManager configuration');
+
+done_testing();