Fix XVM's classic networking problem

[invirt/packages/invirt-dhcp.git] / invirt-dhcpserver

diff --git a/invirt-dhcpserver b/invirt-dhcpserver
index e803ecd..20a0816 100755 (executable)
--- a/invirt-dhcpserver
+++ b/invirt-dhcpserver
@@ -1,4 +1,5 @@
 #!/usr/bin/python
 #!/usr/bin/python

+import os.path

 import sys
 import pydhcplib
 import pydhcplib.dhcp_network
 import sys
 import pydhcplib
 import pydhcplib.dhcp_network

@@ -12,6 +13,8 @@ from Queue import Queue
 from threading import Thread
 from subprocess import PIPE, Popen
 import netifaces as ni
 from threading import Thread
 from subprocess import PIPE, Popen
 import netifaces as ni

+sys.path.append('/usr/lib/xen-default/lib/python/')
+from xen.lowlevel import xs

 
 import syslog as s
 
 
 import syslog as s
 

@@ -22,10 +25,43 @@ from invirt.config import structs as config
 dhcp_options = {'domain_name_server': ','.join(config.dhcp.dns),
                 'ip_address_lease_time': config.dhcp.get('leasetime', 60*60*24)}
 
 dhcp_options = {'domain_name_server': ','.join(config.dhcp.dns),
                 'ip_address_lease_time': config.dhcp.get('leasetime', 60*60*24)}
 

+class Interfaces(object):
+    @staticmethod
+    def primary_ip(name):
+        """primary_ip returns an interface's primary IP address.
+
+        This is the first IPv4 address returned by "ip addr show $name"
+        """
+        # TODO(quentin): The netifaces module is a pile of crappy C.
+        # Figure out a way to do this in native Python.
+        return ni.ifaddresses(name)[ni.AF_INET][0]['addr']
+
+    @staticmethod
+    def exists(name):
+        """exists checks if an interface exists.
+
+        Args:
+        name: Interface name
+        """
+        return os.path.exists("/sys/class/net/"+name)
+
+

 class DhcpBackend:
     def __init__(self, queue):
         database.connect()
         self.queue = queue
 class DhcpBackend:
     def __init__(self, queue):
         database.connect()
         self.queue = queue

+        self.main_ip = Interfaces.primary_ip(config.xen.iface)
+    def add_route_and_arp(self, ip, intf, gateway):
+        try:
+            p = Popen(['ip', 'route', 'add', ip, 'dev', intf, 'src', self.main_ip, 'metric', '2' if intf.startswith('vif') else '1'], stdout=PIPE, stderr=PIPE)
+            (out, err) = p.communicate()
+            if p.returncode == 0:
+                s.syslog(s.LOG_INFO, "Added route for IP %s to interface %s" % (ip, intf))
+                self.queue.put((ip, gateway))
+            sys.stderr.write(err)
+            sys.stdout.write(out)
+        except Exception as e:
+            s.syslog(s.LOG_ERR, "Could not add route for IP %s: %s" % (ip, e))

     def findNIC(self, mac):
         database.clear_cache()
         return database.NIC.query.filter_by(mac_addr=mac).first()
     def findNIC(self, mac):
         database.clear_cache()
         return database.NIC.query.filter_by(mac_addr=mac).first()

@@ -42,6 +78,38 @@ class DhcpBackend:
             if parts[1] == ipstr:
                 s.syslog(s.LOG_DEBUG, "find_interface found "+str(nic.ip)+" on "+parts[0])
                 return parts[0]
             if parts[1] == ipstr:
                 s.syslog(s.LOG_DEBUG, "find_interface found "+str(nic.ip)+" on "+parts[0])
                 return parts[0]

+        # Either the machine isn't running, or the route is missing.  We can
+        # fix the latter.
+        try:
+            xsc = xs.xs()
+            domid = xsc.read('', '/vm/%s/device/vif/0/frontend-id' % (nic.machine.uuid))
+            # If we didn't find the domid, the machine is either off or the
+            # UUID in xenstore isn't right.  Try slightly harder.
+            if not domid:
+                for uuid in xsc.ls('', '/vm'):
+                    if xsc.read('', '/vm/%s/name' % (uuid)) == 'd_' + nic.machine.name:
+                        domid = xsc.read('', '/vm/%s/device/vif/0/frontend-id' % (uuid))
+            if not domid:
+                xsc.close()
+                return None
+            for vifnum in xsc.ls('', '/local/domain/0/backend/vif/%s' % (domid)):
+                if xsc.read('', '/local/domain/0/backend/vif/%s/%s/mac' % (domid, vifnum)) == nic.mac_addr:
+                    # Prefer the tap if it exists; paravirtualized HVMs will
+                    # have already unplugged it, so if it's there, it's the one
+                    # in use.
+                    for viftype in ('tap', 'vif'):
+                        vif = '%s%s.%s' % (viftype, domid, vifnum)
+                        if Interfaces.exists(vif):
+                            self.add_route_and_arp(nic.ip, vif, nic.gateway)
+                            xsc.close()
+                            return vif
+            xsc.close()
+        except Exception as e:
+            try:
+                xsc.close()
+            except Exception as e2:
+                s.syslog(s.LOG_ERR, "Could not close connection to xenstore: %s" % (e2))
+            s.syslog(s.LOG_ERR, "Could not find interface and add missing route: %s" % (e))

         return None
                             
     def getParameters(self, **extra):
         return None
                             
     def getParameters(self, **extra):

@@ -149,7 +217,6 @@ class DhcpBackend:
             s.syslog(s.LOG_INFO,"Ack ip "+str(yiaddr)+" for "+str(chaddr))
             n = self.findNIC(str(chaddr))
             intf = self.find_interface_by_nic(n)
             s.syslog(s.LOG_INFO,"Ack ip "+str(yiaddr)+" for "+str(chaddr))
             n = self.findNIC(str(chaddr))
             intf = self.find_interface_by_nic(n)

-            main_ip = ni.ifaddresses(config.xen.iface)[ni.AF_INET][0]['addr']

             s.syslog(s.LOG_ERR, "Interface is %s" % (intf))
             # Don't perform "other" actions if the machine isn't running
             other_action = n.other_action if n.other_action and intf else ''
             s.syslog(s.LOG_ERR, "Interface is %s" % (intf))
             # Don't perform "other" actions if the machine isn't running
             other_action = n.other_action if n.other_action and intf else ''

@@ -171,16 +238,7 @@ class DhcpBackend:
                 # IP needs to be added.  Just try adding both of them, and
                 # arp for whichever of them turns out to be new.
                 for parms in [(n.ip, n.gateway), (n.other_ip, n.other_gateway)]:
                 # IP needs to be added.  Just try adding both of them, and
                 # arp for whichever of them turns out to be new.
                 for parms in [(n.ip, n.gateway), (n.other_ip, n.other_gateway)]:

-                    try:
-                        p = Popen(['ip', 'route', 'add', parms[0], 'dev', intf, 'src', main_ip, 'metric', '2' if intf.startswith('vif') else '1'], stdout=PIPE, stderr=PIPE)
-                        (out, err) = p.communicate()
-                        if p.returncode == 0:
-                            s.syslog(s.LOG_INFO, "Added route for IP %s to interface %s" % (parms[0], intf))
-                            self.queue.put(parms)
-                        sys.stderr.write(err)
-                        sys.stdout.write(out)
-                    except Exception as e:
-                        s.syslog(s.LOG_ERR, "Could not add route for IP %s: %s" % (parms[0], e))
+                    self.add_route_and_arp(parms[0], intf, parms[1])

                 try:
                     # iptables will let you add the same rule again and again;
                     # let's not do that.
                 try:
                     # iptables will let you add the same rule again and again;
                     # let's not do that.

@@ -355,7 +413,7 @@ class ArpspoofWorker(Thread):
         while True:
             (ip, gw) = self.queue.get()
             try:
         while True:
             (ip, gw) = self.queue.get()
             try:

-                p = Popen(['timeout', '5', 'arpspoof', '-i', self.iface, '-t', gw, ip], stdout=PIPE, stderr=PIPE)
+                p = Popen(['timeout', '-s', 'KILL', '5', 'arpspoof', '-i', self.iface, '-t', gw, ip], stdout=PIPE, stderr=PIPE)

                 (out, err) = p.communicate()
                 if p.returncode != 124:
                     s.syslog(s.LOG_ERR, "arpspoof returned %s for IP %s gateway %s" % (p.returncode, ip, gw))
                 (out, err) = p.communicate()
                 if p.returncode != 124:
                     s.syslog(s.LOG_ERR, "arpspoof returned %s for IP %s gateway %s" % (p.returncode, ip, gw))