Improved DHCP client messaging

Dockerfile

@@ -1,7 +1,9 @@
 FROM python:3-alpine
 
 COPY requirements.txt /opt/
-RUN pip install -r /opt/requirements.txt
+RUN apk --no-cache add gcc musl-dev && \
+    pip install -r /opt/requirements.txt && \
+    apk --no-cache del gcc musl-dev
 
 RUN mkdir -p /opt/plugin /run/docker/plugins /var/run/docker/netns
 COPY net-dhcp/ /opt/plugin/net_dhcp

net-dhcp/network.py

@@ -222,10 +222,11 @@ def join():
 
     ipv6 = ipv6_enabled(req['NetworkID'])
     for route in bridge.routes:
-        if route['type'] != rtypes['RTN_UNICAST'] or (route['family'] == socket.AF_INET6 and not ipv6):
+        if route['type'] != rtypes['RTN_UNICAST'] or \
+            (route['family'] == socket.AF_INET6 and not ipv6):
             continue
 
-        if route['dst'] == '' or route['dst'] == '/0':
+        if route['dst'] in ('', '/0'):
             if route['family'] == socket.AF_INET and 'Gateway' not in res:
                 logger.info('Adding IPv4 gateway %s', route['gateway'])
                 res['Gateway'] = route['gateway']
@@ -253,15 +254,38 @@ class ContainerDHCPManager:
     def __init__(self, network, endpoint):
         self.network = network
         self.endpoint = endpoint
+        self.ipv6 = ipv6_enabled(network)
 
         self._thread = threading.Thread(target=self.run)
         self._thread.start()
 
+    def _on_event(self, dhcp, event_type, _args):
+        if event_type != udhcpc.EventType.RENEW:
+            return
+
+        for route in dhcp.iface.routes:
+            if route['type'] != rtypes['RTN_UNICAST'] or \
+                (route['family'] == socket.AF_INET6 and not self.ipv6) or \
+                route['dst'] not in ('', '/0'):
+                continue
+
+            # Needed because Route.remove() doesn't like a blank destination
+            logger.info('Removing default route via %s', route['gateway'])
+            route['dst'] = '::' if route['family'] == socket.AF_INET6 else '0.0.0.0'
+            (route
+                .remove()
+                .commit())
+
+        logger.info('Adding default route via %s', dhcp.gateway)
+        (dhcp.iface.routes.add({'gateway': dhcp.gateway})
+            .commit())
+
     def run(self):
         iface = endpoint_container_iface(self.network, self.endpoint)
         self.dhcp = udhcpc.DHCPClient(iface)
         logger.info('Starting DHCP client on %s in container namespace %s', iface['ifname'], \
             self.dhcp.netns)
+
     def stop(self):
         logger.info('Shutting down DHCP client on %s in container namespace %s', \
             self.dhcp.iface['ifname'], self.dhcp.netns)

net-dhcp/udhcpc.py

@@ -2,15 +2,15 @@ from enum import Enum
 import ipaddress
 import os
 from os import path
-import fcntl
-import time
+from select import select
 import threading
 import subprocess
 import logging
 
+from eventfd import EventFD
+import posix_ipc
 from pyroute2.netns.process.proxy import NSPopen
 
-EVENT_PREFIX = '__event'
 HANDLER_SCRIPT = path.join(path.dirname(__file__), 'udhcpc_handler.py')
 AWAIT_INTERVAL = 0.1
 
@@ -26,13 +26,7 @@ class DHCPClientError(Exception):
     pass
 
 def _nspopen_wrapper(netns):
-    def _wrapper(*args, **kwargs):
-        # We have to set O_NONBLOCK on stdout since NSPopen uses a global lock
-        # on the object (e.g. deadlock if we try to readline() and terminate())
-        proc = NSPopen(netns, *args, **kwargs)
-        proc.stdout.fcntl(fcntl.F_SETFL, os.O_NONBLOCK)
-        return proc
-    return _wrapper
+    return lambda *args, **kwargs: NSPopen(netns, *args, **kwargs)
 class DHCPClient:
     def __init__(self, iface, once=False, event_listener=None):
         self.iface = iface
@@ -49,14 +43,17 @@ class DHCPClient:
         Popen = _nspopen_wrapper(self.netns) if self.netns else subprocess.Popen
         cmdline = ['/sbin/udhcpc', '-s', HANDLER_SCRIPT, '-i', iface['ifname'], '-f']
         cmdline.append('-q' if once else '-R')
-        self.proc = Popen(cmdline, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, encoding='utf-8')
+
+        self._event_queue = posix_ipc.MessageQueue(f'/udhcpc_{iface["address"].replace(":", "_")}', \
+            flags=os.O_CREAT | os.O_EXCL)
+        self.proc = Popen(cmdline, env={'EVENT_QUEUE': self._event_queue.name})
 
         self._has_lease = threading.Event()
         self.ip = None
         self.gateway = None
         self.domain = None
 
-        self._running = True
+        self._shutdown_event = EventFD()
         self._event_thread = threading.Thread(target=self._read_events)
         self._event_thread.start()
 
@@ -73,21 +70,13 @@ class DHCPClient:
             self.domain = None
 
     def _read_events(self):
-        while self._running:
-            line = self.proc.stdout.readline().strip()
-            if not line:
-                # stdout will be O_NONBLOCK if udhcpc is in a netns
-                # We can't use select() since the file descriptor is from
-                # the NSPopen proxy
-                if self.netns and self._running:
-                    time.sleep(0.1)
-                continue
+        while True:
+            r, _w, _e = select([self._shutdown_event, self._event_queue.mqd], [], [])
+            if self._shutdown_event in r:
+                break
 
-            if not line.startswith(EVENT_PREFIX):
-                logger.debug('[udhcpc#%d] %s', self.proc.pid, line)
-                continue
-
-            args = line.split(' ')[1:]
+            msg, _priority = self._event_queue.receive()
+            args = msg.decode('utf-8').split(' ')
             try:
                 event_type = EventType(args[0])
             except ValueError:
@@ -114,7 +103,10 @@ class DHCPClient:
             raise DHCPClientError(f'udhcpc exited with non-zero exit code {self.proc.returncode}')
         if self.netns:
             self.proc.release()
-        self._running = False
+
+        self._shutdown_event.set()
         self._event_thread.join()
+        self._event_queue.close()
+        self._event_queue.unlink()
 
         return self.ip

net-dhcp/udhcpc_handler.py

@@ -2,7 +2,7 @@
 import sys
 from os import environ as env
 
-EVENT_PREFIX = '__event'
+import posix_ipc
 
 if __name__ != '__main__':
     print('You shouldn\'t be importing this script!')
@@ -10,9 +10,15 @@ if __name__ != '__main__':
 
 event_type = sys.argv[1]
 if event_type in ('bound', 'renew'):
-    print(f'{EVENT_PREFIX} {event_type} {env["ip"]}/{env["mask"]} {env["router"]} {env["domain"]}')
+    event = f'{event_type} {env["ip"]}/{env["mask"]} {env["router"]} {env["domain"]}'
 elif event_type in ('deconfig', 'leasefail', 'nak'):
-    print(f'{EVENT_PREFIX} {event_type}')
+    event = event_type
 else:
     print(f'unknown udhcpc event "{event_type}"')
     sys.exit(1)
+
+#with open(env['EVENT_FILE'], 'a') as event_file:
+#    event_file.write(event + '\n')
+queue = posix_ipc.MessageQueue(env['EVENT_QUEUE'])
+queue.send(event)
+queue.close()

requirements.txt

@@ -2,3 +2,5 @@ flask==1.1.1
 gunicorn==19.9.0
 pyroute2==0.5.6
 docker==4.0.2
+eventfd==0.2
+posix_ipc==1.0.4