Ensure SR-IOV NIC states are updated on migration

daemon-common/vm.py

@@ -555,6 +555,33 @@ def disable_vm(zkhandler, domain):
     return True, 'Marked VM "{}" as disable.'.format(domain)
 
 
+def update_vm_sriov_nics(zkhandler, dom_uuid, source_node, target_node):
+    # Update all the SR-IOV device states on both nodes, used during migrations but called by the node-side
+    vm_config = zkhandler.read(('domain.xml', dom_uuid))
+    parsed_xml = lxml.objectify.fromstring(vm_config)
+    dnetworks = common.getDomainNetworks(parsed_xml, {})
+    retcode = True
+    retmsg = ''
+    for network in dnetworks:
+        if network['type'] in ['direct', 'hostdev']:
+            # Check if the network is already in use
+            is_used = zkhandler.read(('node.sriov.vf', target_node, 'sriov_vf.used', network['source']))
+            if is_used == 'True':
+                used_by_name = searchClusterByUUID(zkhandler, zkhandler.read(('node.sriov.vf', target_node, 'sriov_vf.used_by', network['source'])))
+                if retcode:
+                    retcode = False
+                    retmsg = 'Attempting to use SR-IOV network "{}" which is already used by VM "{}"'.format(network['source'], used_by_name)
+
+            # We must update the "used" section
+            if retcode:
+                # This conditional ensure that if we failed the is_used check, we don't try to overwrite the information of a VF that belongs to another VM
+                set_sriov_vf_vm(zkhandler, dom_uuid, target_node, network['source'], network['mac'], network['type'])
+            # ... but we still want to free the old node in an case
+            unset_sriov_vf_vm(zkhandler, source_node, network['source'])
+
+    return retcode, retmsg
+
+
 def move_vm(zkhandler, domain, target_node, wait=False, force_live=False):
     # Validate that VM exists in cluster
     dom_uuid = getDomainUUID(zkhandler, domain)

node-daemon/pvcnoded/VMInstance.py

@@ -34,6 +34,8 @@ import pvcnoded.VMConsoleWatcherInstance as VMConsoleWatcherInstance
 
 import daemon_lib.common as daemon_common
 
+from daemon_lib.vm import update_vm_sriov_nics
+
 
 def flush_locks(zkhandler, logger, dom_uuid, this_node=None):
     logger.out('Flushing RBD locks for VM "{}"'.format(dom_uuid), state='i')
@@ -672,6 +674,11 @@ class VMInstance(object):
         self.logger.out('Acquired write lock for synchronization phase D', state='o', prefix='Domain {}'.format(self.domuuid))
         time.sleep(0.5)  # Time for reader to acquire the lock
 
+        # Update any SR-IOV NIC states now
+        sriov_update_result, sriov_update_error = update_vm_sriov_nics(self.zkhandler, self.domuuid, self.last_currentnode, self.node)
+        if not sriov_update_result:
+            self.logger.out('{}; VM will likely fail to start.'.format(sriov_update_error), state='w', prefix='Domain {}'.format(self.domuuid))
+
         self.state = self.zkhandler.read(('domain.state', self.domuuid))
         self.dom = self.lookupByUUID(self.domuuid)
         if self.dom: