Description of problem: It is not possible to enroll ipa replica on IPv6 only ipa client. Version-Release number of selected component (if applicable): ipa-client-4.10.1-8.el9_2.x86_64 ipa-client-common-4.10.1-8.el9_2.noarch ipa-common-4.10.1-8.el9_2.noarch ipa-healthcheck-core-0.12-1.el9.noarch ipa-selinux-4.10.1-8.el9_2.noarch ipa-server-4.10.1-8.el9_2.x86_64 ipa-server-common-4.10.1-8.el9_2.noarch ipa-server-dns-4.10.1-8.el9_2.noarch ipa-server-trust-ad-4.10.1-8.el9_2.x86_64 How reproducible: Always. Steps to Reproduce: filippa:~# ipa-replica-install \ > --no-ui-redirect \ > --no-ssh \ > --no-sshd \ > --unattended \ > \ > --setup-ca \ > \ > --setup-dns \ > --no-forwarders \ > --allow-zone-overlap \ > --no-dns-sshfp \ > --setup-adtrust Connection to https://triss.jot23.net/ipa/json failed with Insufficient access: SASL(-1): generic failure: GSSAPI Error: Unspecified GSS failure. Minor code may provide more information (Credential cache is empty) Lookup failed: Preferred host filippa.jot23.net does not provide DNS. WARNING: The smb.conf already exists. Running ipa-adtrust-install will break your existing samba configuration. Run connection check to master Your system may be partly configured. Run /usr/sbin/ipa-server-install --uninstall to clean up. Connection check failed! See /var/log/ipareplica-conncheck.log for more information. If the check results are not valid it can be skipped with --skip-conncheck parameter. The ipa-replica-install command failed. See /var/log/ipareplica-install.log for more information The excerpt from /var/log/ipareplica-install.log: 2023-08-07T15:15:30Z DEBUG args=['/usr/sbin/ipa-replica-conncheck', '--master', 'triss.jot23.net', '--auto-master-check', '--realm', 'IPA.JOT23.NET', '--hostname', 'filippa.jot23.net', '--ca-cert-file', '/etc/ipa/ca.crt'] 2023-08-07T15:15:31Z DEBUG Process finished, return code=1 2023-08-07T15:15:31Z DEBUG stdout= 2023-08-07T15:15:31Z DEBUG stderr=Check connection from replica to remote master 'triss.jot23.net': Failed to connect to port 389 tcp on 192.168.16.10 Directory Service: Unsecure port (389): FAILED Failed to connect to port 636 tcp on 192.168.16.10 Directory Service: Secure port (636): FAILED Failed to connect to port 88 tcp on 192.168.16.10 Kerberos KDC: TCP (88): FAILED Failed to connect to port 464 tcp on 192.168.16.10 Kerberos Kpasswd: TCP (464): FAILED Failed to connect to port 80 tcp on 192.168.16.10 HTTP Server: Unsecure port (80): FAILED Failed to connect to port 443 tcp on 192.168.16.10 HTTP Server: Secure port (443): FAILED ERROR: Port check failed! Inaccessible port(s): 389 (TCP), 636 (TCP), 88 (TCP), 464 (TCP), 80 (TCP), 443 (TCP) The host is IPv6 only: filippa:~# ifconfig enp1s0: flags=4163<UP,BROADCAST,RUNNING,MULTICAST> mtu 1500 inet6 fe80::5054:ff:fea2:1097 prefixlen 64 scopeid 0x20<link> inet6 2a06:a005:1991:1100::20 prefixlen 64 scopeid 0x0<global> ether 52:54:00:a2:10:97 txqueuelen 1000 (Ethernet) RX packets 86772 bytes 69805321 (66.5 MiB) RX errors 0 dropped 34778 overruns 0 frame 0 TX packets 33819 bytes 4954541 (4.7 MiB) TX errors 0 dropped 0 overruns 0 carrier 0 collisions 0 lo: flags=73<UP,LOOPBACK,RUNNING> mtu 65536 inet 127.0.0.1 netmask 255.0.0.0 inet6 ::1 prefixlen 128 scopeid 0x10<host> loop txqueuelen 1000 (Local Loopback) RX packets 46 bytes 2696 (2.6 KiB) RX errors 0 dropped 0 overruns 0 frame 0 TX packets 46 bytes 2696 (2.6 KiB) TX errors 0 dropped 0 overruns 0 carrier 0 collisions 0 ipa-replica-conncheck is NOT compliant with RFC6724. See https://datatracker.ietf.org/doc/html/rfc6724#section-6 where Rule 1 clearly states: Avoid unusable destinations. The system in question has AAAA record which should have been used in the first place: filippa:~# dig triss.jot23.net aaaa +short 2aXX:XXXX:XXXX:1000::10 Actual results: Cannot enroll IPv6 only replica. Expected results: Can enroll IPv6 only replica. Additional info: It is possible to skip the check with --skip-conncheck option and proceed. It should not be needed though. The system being enrolled is on RHEL 9.2. Other replicas are on AlmaLinux 9.2.
Notes: We need to extend IPA installers to be smarter and recognize IPv6-only or IPv4-only setups. This means also taking into account gai.conf configuration (default one, when /etc/gai.conf does not exist, is to prefer IPv6 over IPv4). When only IPv6 routing is available: - when detecting IPv6-only configuration, filter out link-local IPv6 routes as these will not lead to working IPv6 setup - filter out IPv4 hosts from the list of available replicas. Assume system resolver is taking gai.conf preferences. When only IPv4 routing is available: - when detecting IPv4-only configuration, filter out link-local IPv6 routes as these will not lead to working IPv6 setup - filter out IPv6 hosts from the list of available replicas. Assume system resolver is already taking gai.conf preferences into account. In both cases it should be done for both client and replica deployments. Additional consideration on the client side is to filter out IPv4 hosts from IPv6-only environment (and vice-versa) and issue a warning that hosts were dropped out. Perhaps, we would need to add a way to tune this through /etc/ipa/installer.conf to allow either behavior (drop/not drop) in order to avoid adding new command line options.
The interesting part is that the installation from dual stack system against IPv6 only master works: the connection checker is able to resolve IPv6 only destination. Please make sure to validate connection checking both ways. Connection checker seems to check connection from client to master first and then seems to ask the master to check the other way - this can be prone to same issue as described here. Additionally I noted that CA enrolment at least sometimes fails when CA resolves to both IPv4 and IPv6 but is not reachable by IPv4 (RFC1918 involved at least on one side of the connection). But this likely is another issue (please let me know if I should look into it a bit more and file a bug if confirmed).