Add config modules for controlling IBM PowerVM RMC. (#584)

Reliable Scalable Cluster Technology (RSCT) is a set of software
components that together provide a comprehensive clustering
environment(RAS features) for IBM PowerVM based virtual machines. RSCT
includes the Resource Monitoring and Control (RMC) subsystem. RMC is a
generalized framework used for managing, monitoring, and manipulating
resources. RMC runs as a daemon process on individual machines and needs
creation of unique node id and restarts during VM boot.

LP: #1895979

Co-authored-by: Scott Moser <smoser@brickies.net>

2 files changed, 302 insertions, 0 deletions

diff --git a/cloudinit/config/cc_refresh_rmc_and_interface.py b/cloudinit/config/cc_refresh_rmc_and_interface.py
new file mode 100644
index 00000000..146758ad
--- /dev/null
+++ b/cloudinit/config/cc_refresh_rmc_and_interface.py
@@ -0,0 +1,159 @@
+# (c) Copyright IBM Corp. 2020 All Rights Reserved
+#
+# Author: Aman Kumar Sinha <amansi26@in.ibm.com>
+#
+# This file is part of cloud-init. See LICENSE file for license information.
+
+"""
+Refresh IPv6 interface and RMC
+------------------------------
+**Summary:** Ensure Network Manager is not managing IPv6 interface
+
+This module is IBM PowerVM Hypervisor specific
+
+Reliable Scalable Cluster Technology (RSCT) is a set of software components
+that together provide a comprehensive clustering environment(RAS features)
+for IBM PowerVM based virtual machines. RSCT includes the Resource
+Monitoring and Control (RMC) subsystem. RMC is a generalized framework used
+for managing, monitoring, and manipulating resources. RMC runs as a daemon
+process on individual machines and needs creation of unique node id and
+restarts during VM boot.
+More details refer
+https://www.ibm.com/support/knowledgecenter/en/SGVKBA_3.2/admin/bl503_ovrv.htm
+
+This module handles
+- Refreshing RMC
+- Disabling NetworkManager from handling IPv6 interface, as IPv6 interface
+  is used for communication between RMC daemon and PowerVM hypervisor.
+
+**Internal name:** ``cc_refresh_rmc_and_interface``
+
+**Module frequency:** per always
+
+**Supported distros:** RHEL
+
+"""
+
+from cloudinit import log as logging
+from cloudinit.settings import PER_ALWAYS
+from cloudinit import util
+from cloudinit import subp
+from cloudinit import netinfo
+
+import errno
+
+frequency = PER_ALWAYS
+
+LOG = logging.getLogger(__name__)
+# Ensure that /opt/rsct/bin has been added to standard PATH of the
+# distro. The symlink to rmcctrl is /usr/sbin/rsct/bin/rmcctrl .
+RMCCTRL = 'rmcctrl'
+
+
+def handle(name, _cfg, _cloud, _log, _args):
+    if not subp.which(RMCCTRL):
+        LOG.debug("No '%s' in path, disabled", RMCCTRL)
+        return
+
+    LOG.debug(
+        'Making the IPv6 up explicitly. '
+        'Ensuring IPv6 interface is not being handled by NetworkManager '
+        'and it is  restarted to re-establish the communication with '
+        'the hypervisor')
+
+    ifaces = find_ipv6_ifaces()
+
+    # Setting NM_CONTROLLED=no for IPv6 interface
+    # making it down and up
+
+    if len(ifaces) == 0:
+        LOG.debug("Did not find any interfaces with ipv6 addresses.")
+    else:
+        for iface in ifaces:
+            refresh_ipv6(iface)
+            disable_ipv6(sysconfig_path(iface))
+        restart_network_manager()
+
+
+def find_ipv6_ifaces():
+    info = netinfo.netdev_info()
+    ifaces = []
+    for iface, data in info.items():
+        if iface == "lo":
+            LOG.debug('Skipping localhost interface')
+        if len(data.get("ipv4", [])) != 0:
+            # skip this interface, as it has ipv4 addrs
+            continue
+        ifaces.append(iface)
+    return ifaces
+
+
+def refresh_ipv6(interface):
+    # IPv6 interface is explicitly brought up, subsequent to which the
+    # RMC services are restarted to re-establish the communication with
+    # the hypervisor.
+    subp.subp(['ip', 'link', 'set', interface, 'down'])
+    subp.subp(['ip', 'link', 'set', interface, 'up'])
+
+
+def sysconfig_path(iface):
+    return '/etc/sysconfig/network-scripts/ifcfg-' + iface
+
+
+def restart_network_manager():
+    subp.subp(['systemctl', 'restart', 'NetworkManager'])
+
+
+def disable_ipv6(iface_file):
+    # Ensuring that the communication b/w the hypervisor and VM is not
+    # interrupted due to NetworkManager. For this purpose, as part of
+    # this function, the NM_CONTROLLED is explicitly set to No for IPV6
+    # interface and NetworkManager is restarted.
+    try:
+        contents = util.load_file(iface_file)
+    except IOError as e:
+        if e.errno == errno.ENOENT:
+            LOG.debug("IPv6 interface file %s does not exist\n",
+                      iface_file)
+        else:
+            raise e
+
+    if 'IPV6INIT' not in contents:
+        LOG.debug("Interface file %s did not have IPV6INIT", iface_file)
+        return
+
+    LOG.debug("Editing interface file %s ", iface_file)
+
+    # Dropping any NM_CONTROLLED or IPV6 lines from IPv6 interface file.
+    lines = contents.splitlines()
+    lines = [line for line in lines if not search(line)]
+    lines.append("NM_CONTROLLED=no")
+
+    with open(iface_file, "w") as fp:
+        fp.write("\n".join(lines) + "\n")
+
+
+def search(contents):
+    # Search for any NM_CONTROLLED or IPV6 lines in IPv6 interface file.
+    return(
+        contents.startswith("IPV6ADDR") or
+        contents.startswith("IPADDR6") or
+        contents.startswith("IPV6INIT") or
+        contents.startswith("NM_CONTROLLED"))
+
+
+def refresh_rmc():
+    # To make a healthy connection between RMC daemon and hypervisor we
+    # refresh RMC. With refreshing RMC we are ensuring that making IPv6
+    # down and up shouldn't impact communication between RMC daemon and
+    # hypervisor.
+    # -z : stop Resource Monitoring & Control subsystem and all resource
+    # managers, but the command does not return control to the user
+    # until the subsystem and all resource managers are stopped.
+    # -s : start Resource Monitoring & Control subsystem.
+    try:
+        subp.subp([RMCCTRL, '-z'])
+        subp.subp([RMCCTRL, '-s'])
+    except Exception:
+        util.logexc(LOG, 'Failed to refresh the RMC subsystem.')
+        raise
diff --git a/cloudinit/config/cc_reset_rmc.py b/cloudinit/config/cc_reset_rmc.py
new file mode 100644
index 00000000..1cd72774
--- /dev/null
+++ b/cloudinit/config/cc_reset_rmc.py
@@ -0,0 +1,143 @@
+# (c) Copyright IBM Corp. 2020 All Rights Reserved
+#
+# Author: Aman Kumar Sinha <amansi26@in.ibm.com>
+#
+# This file is part of cloud-init. See LICENSE file for license information.
+
+
+"""
+Reset RMC
+------------
+**Summary:** reset rsct node id
+
+Reset RMC module is IBM PowerVM Hypervisor specific
+
+Reliable Scalable Cluster Technology (RSCT) is a set of software components,
+that  together provide a comprehensive clustering environment (RAS features)
+for IBM PowerVM based virtual machines. RSCT includes the Resource monitoring
+and control (RMC) subsystem. RMC is a generalized framework used for managing,
+monitoring, and manipulating resources. RMC runs as a daemon process on
+individual machines and needs creation of unique node id and restarts
+during VM boot.
+More details refer
+https://www.ibm.com/support/knowledgecenter/en/SGVKBA_3.2/admin/bl503_ovrv.htm
+
+This module handles
+- creation of the unique RSCT node id to every instance/virtual machine
+  and ensure once set, it isn't changed subsequently by cloud-init.
+  In order to do so, it restarts RSCT service.
+
+Prerequisite of using this module is to install RSCT packages.
+
+**Internal name:** ``cc_reset_rmc``
+
+**Module frequency:** per instance
+
+**Supported distros:** rhel, sles and ubuntu
+
+"""
+import os
+
+from cloudinit import log as logging
+from cloudinit.settings import PER_INSTANCE
+from cloudinit import util
+from cloudinit import subp
+
+frequency = PER_INSTANCE
+
+# RMCCTRL is expected to be in system PATH (/opt/rsct/bin)
+# The symlink for RMCCTRL and RECFGCT are
+# /usr/sbin/rsct/bin/rmcctrl and
+# /usr/sbin/rsct/install/bin/recfgct respectively.
+RSCT_PATH = '/opt/rsct/install/bin'
+RMCCTRL = 'rmcctrl'
+RECFGCT = 'recfgct'
+
+LOG = logging.getLogger(__name__)
+
+NODE_ID_FILE = '/etc/ct_node_id'
+
+
+def handle(name, _cfg, cloud, _log, _args):
+    # Ensuring node id has to be generated only once during first boot
+    if cloud.datasource.platform_type == 'none':
+        LOG.debug('Skipping creation of new ct_node_id node')
+        return
+
+    if not os.path.isdir(RSCT_PATH):
+        LOG.debug("module disabled, RSCT_PATH not present")
+        return
+
+    orig_path = os.environ.get('PATH')
+    try:
+        add_path(orig_path)
+        reset_rmc()
+    finally:
+        if orig_path:
+            os.environ['PATH'] = orig_path
+        else:
+            del os.environ['PATH']
+
+
+def reconfigure_rsct_subsystems():
+    # Reconfigure the RSCT subsystems, which includes removing all RSCT data
+    # under the /var/ct directory, generating a new node ID, and making it
+    # appear as if the RSCT components were just installed
+    try:
+        out = subp.subp([RECFGCT])[0]
+        LOG.debug(out.strip())
+        return out
+    except subp.ProcessExecutionError:
+        util.logexc(LOG, 'Failed to reconfigure the RSCT subsystems.')
+        raise
+
+
+def get_node_id():
+    try:
+        fp = util.load_file(NODE_ID_FILE)
+        node_id = fp.split('\n')[0]
+        return node_id
+    except Exception:
+        util.logexc(LOG, 'Failed to get node ID from file %s.' % NODE_ID_FILE)
+        raise
+
+
+def add_path(orig_path):
+    # Adding the RSCT_PATH to env standard path
+    # So thet cloud init automatically find and
+    # run RECFGCT to create new node_id.
+    suff = ":" + orig_path if orig_path else ""
+    os.environ['PATH'] = RSCT_PATH + suff
+    return os.environ['PATH']
+
+
+def rmcctrl():
+    # Stop the RMC subsystem and all resource managers so that we can make
+    # some changes to it
+    try:
+        return subp.subp([RMCCTRL, '-z'])
+    except Exception:
+        util.logexc(LOG, 'Failed to stop the RMC subsystem.')
+        raise
+
+
+def reset_rmc():
+    LOG.debug('Attempting to reset RMC.')
+
+    node_id_before = get_node_id()
+    LOG.debug('Node ID at beginning of module: %s', node_id_before)
+
+    # Stop the RMC subsystem and all resource managers so that we can make
+    # some changes to it
+    rmcctrl()
+    reconfigure_rsct_subsystems()
+
+    node_id_after = get_node_id()
+    LOG.debug('Node ID at end of module: %s', node_id_after)
+
+    # Check if new node ID is generated or not
+    # by comparing old and new node ID
+    if node_id_after == node_id_before:
+        msg = 'New node ID did not get generated.'
+        LOG.error(msg)
+        raise Exception(msg)