Controller: remove default '-v' flag!!

For the moment, we are deprecating using localhost. We may allow
it again in the future.

CONTRIBUTORS

@@ -29,14 +29,16 @@ Andrew Ferguson
 Eder Leao Fernandes
 Gregory Gee
 Jon Hall
+Roan Huang
 Vitaly Ivanov
+Babis Kaidos
 Rich Lane
+Rémy Léone
 Zi Shen Lim
+David Mahler
 Murphy McCauley
 José Pedro Oliveira
 James Page
-Rich Lane
-Rémy Léone
 Angad Singh
 Piyush Srivastava
 Ed Swierk

INSTALL

@@ -2,7 +2,7 @@
 Mininet Installation/Configuration Notes
 ----------------------------------------
 
-Mininet 2.2.0
+Mininet 2.2.1
 ---
 
 The supported installation methods for Mininet are 1) using a

LICENSE

@@ -1,6 +1,6 @@
-Mininet 2.2.0 License
+Mininet 2.2.1 License
 
-Copyright (c) 2013-2014 Open Networking Laboratory
+Copyright (c) 2013-2015 Open Networking Laboratory
 Copyright (c) 2009-2012 Bob Lantz and The Board of Trustees of
 The Leland Stanford Junior University
 

README.md

@@ -3,7 +3,7 @@ Mininet: Rapid Prototyping for Software Defined Networks
 
 *The best way to emulate almost any network on your laptop!*
 
-Mininet 2.2.0
+Mininet 2.2.1
 
 ### What is Mininet?
 
@@ -68,44 +68,25 @@ Mininet includes:
 
 ### New features in this release
 
-This release provides a number of bug fixes as well as
-several new features, including:
+This is primarily a performance improvement and bug fix release.
 
-* Improved OpenFlow 1.3 support
+- Batch startup has been implemented for Open vSwitch, improving
+  startup performance.
 
-	- `mn --switch ovs,protocols=openflow13` starts OVS in 1.3 mode
-	- `install.sh -w` installs a 1.3-compatible Wireshark dissector using
-	  Loxigen
-	- `install.sh -y` installs the Ryu 1.3-compatible controller
+- OVS patch links have been implemented via OVSLink and --link ovs
 
-* A new `nodelib.py` node library, and new `Node` types including
-  `LinuxBridge`, `OVSBridge`, `LinuxRouter` (see `examples/`)
-  and `NAT`
+  Warning! These links have *serious limitations* compared to
+  virtual Ethernet pairs: they are not attached to real Linux
+  interfaces so you cannot use tcpdump or wireshark with them;
+  they also cannot be used in long chains - we don't recommend more
+  than 64 OVSLinks, for example --linear,64. However, they can offer
+  significantly better performance than veth pairs, for certain
+  configurations.
 
-* A `--nat` option which connects a Mininet network to your LAN using NAT
-  (For this to work correctly, Mininet's `--ipbase` subnet should not 
-  overlap with any external or internet IP addresses you wish to use)
+- You can now easily install Mininet on a Raspberry Pi ;-)
 
-* An improved MiniEdit GUI (`examples/miniedit.py`) - thanks to
-  Gregory Gee
-
-* Support for multiple `--custom` arguments to `mn`
-
-* Experimental cluster support - consult the 
-  [documentation](http://docs.mininet.org) for details -
-  as well as `examples/cluster.py` and an experimental `--cluster`
-  option for topologies built with the default `Host` and `OVSSwitch`
-  classes:
-
-  `mn --cluster localhost,server1,server2`
-
-Note that examples contain experimental features which might
-"graduate" into mainline Mininet in the future, but they should 
-not be considered a stable part of the Mininet API!
-
-A number of bugs have also been fixed, most notably multiple link
-support in `Topo()`. See github issues and the release notes on
-the Mininet wiki for additional information.
+- Additional information for this release and previous releases
+  may be found in the release notes on docs.mininet.org
 
 ### Installation
 
@@ -129,6 +110,8 @@ Mininet mailing list, `mininet-discuss` at:
 
 ### Join Us
 
+Thanks again to all of the Mininet contributors!
+
 Mininet is an open source project and is currently hosted
 at <https://github.com/mininet>.  You are encouraged to download
 the code, examine it, modify it, and submit bug reports, bug fixes,
@@ -142,12 +125,5 @@ hard work that Mininet continues to grow and improve.
 Best wishes, and we look forward to seeing what you can do with
 Mininet to change the networking world!
 
-The Mininet Core Team:
-
-* Bob Lantz
-* Brian O'Connor
-* Cody Burkard
-
-Thanks again to all of the Mininet contributors, particularly Gregory
-Gee for his work on MiniEdit.
-
+Bob Lantz
+Mininet Core Team

bin/mn

@@ -25,30 +25,31 @@ from mininet.cli import CLI
 from mininet.log import lg, LEVELS, info, debug, warn, error
 from mininet.net import Mininet, MininetWithControlNet, VERSION
 from mininet.node import ( Host, CPULimitedHost, Controller, OVSController,
-                           RYU, NOX, RemoteController, findController,
-                           DefaultController,
+                           Ryu, NOX, RemoteController, findController,
+                           DefaultController, NullController,
                            UserSwitch, OVSSwitch, OVSBridge,
-                           OVSLegacyKernelSwitch, IVSSwitch )
-from mininet.nodelib import LinuxBridge
-from mininet.link import Link, TCLink
-from mininet.topo import SingleSwitchTopo, LinearTopo, SingleSwitchReversedTopo
+                           IVSSwitch )
+from mininet.nodelib import LinuxBridge, Server
+from mininet.link import Link, TCLink, OVSLink
+from mininet.topo import ( SingleSwitchTopo, LinearTopo,
+                           SingleSwitchReversedTopo, MinimalTopo )
 from mininet.topolib import TreeTopo, TorusTopo
-from mininet.util import customConstructor, splitArgs
+from mininet.util import customClass, specialClass, splitArgs
 from mininet.util import buildTopo
 
 from functools import partial
 
 # Experimental! cluster edition prototype
-from mininet.examples.cluster import ( MininetCluster, RemoteHost,
-                                       RemoteOVSSwitch, RemoteLink,
-                                       SwitchBinPlacer, RandomPlacer )
+from mininet.examples.cluster import ( MininetCluster,
+                                       SwitchBinPlacer, RandomPlacer,
+                                       ClusterCleanup )
 from mininet.examples.clustercli import ClusterCLI
 
 PLACEMENT = { 'block': SwitchBinPlacer, 'random': RandomPlacer }
 
 # built in topologies, created only when run
 TOPODEF = 'minimal'
-TOPOS = { 'minimal': lambda: SingleSwitchTopo( k=2 ),
+TOPOS = { 'minimal': MinimalTopo,
           'linear': LinearTopo,
           'reversed': SingleSwitchReversedTopo,
           'single': SingleSwitchTopo,
@@ -61,28 +62,30 @@ SWITCHES = { 'user': UserSwitch,
              'ovsbr' : OVSBridge,
              # Keep ovsk for compatibility with 2.0
              'ovsk': OVSSwitch,
-             'ovsl': OVSLegacyKernelSwitch,
              'ivs': IVSSwitch,
              'lxbr': LinuxBridge,
              'default': OVSSwitch }
 
 HOSTDEF = 'proc'
 HOSTS = { 'proc': Host,
-          'rt': partial( CPULimitedHost, sched='rt' ),
-          'cfs': partial( CPULimitedHost, sched='cfs' ) }
+          'server': Server,
+          'rt': specialClass( CPULimitedHost, defaults=dict( sched='rt' ) ),
+          'cfs': specialClass( CPULimitedHost, defaults=dict( sched='cfs' ) ) }
 
 CONTROLLERDEF = 'default'
+
 CONTROLLERS = { 'ref': Controller,
                 'ovsc': OVSController,
                 'nox': NOX,
                 'remote': RemoteController,
-                'ryu': RYU,
+                'ryu': Ryu,
                 'default': DefaultController,  # Note: replaced below
-                'none': lambda name: None }
+                'none': NullController }
 
 LINKDEF = 'default'
 LINKS = { 'default': Link,
-          'tc': TCLink }
+          'tc': TCLink,
+          'ovs': OVSLink }
 
 
 # optional tests to run
@@ -95,24 +98,21 @@ ALTSPELLING = { 'pingall': 'pingAll',
                 'iperfUDP': 'iperfUdp' }
 
 
-def addDictOption( opts, choicesDict, default, name, helpStr=None ):
+def addDictOption( opts, choicesDict, default, name, **kwargs ):
     """Convenience function to add choices dicts to OptionParser.
        opts: OptionParser instance
        choicesDict: dictionary of valid choices, must include default
        default: default choice key
        name: long option name
-       help: string"""
-    if default not in choicesDict:
-        raise Exception( 'Invalid  default %s for choices dict: %s' %
-                         ( default, name ) )
-    if not helpStr:
-        helpStr = ( '|'.join( sorted( choicesDict.keys() ) ) +
-                    '[,param=value...]' )
-    opts.add_option( '--' + name,
-                     type='string',
-                     default = default,
-                     help = helpStr )
-
+       kwargs: additional arguments to add_option"""
+    helpStr = ( '|'.join( sorted( choicesDict.keys() ) ) +
+                '[,param=value...]' )
+    helpList = [ '%s=%s' % ( k, v.__name__ )
+                 for k, v in choicesDict.items() ]
+    helpStr += ' ' + ( ' '.join( helpList ) )
+    params = dict( type='string', default=default, help=helpStr )
+    params.update( **kwargs )
+    opts.add_option( '--' + name, **params )
 
 def version( *_args ):
     "Print Mininet version and exit"
@@ -158,7 +158,7 @@ class MininetRunner( object ):
 
     def setCustom( self, name, value ):
         "Set custom parameters for MininetRunner."
-        if name in ( 'topos', 'switches', 'hosts', 'controllers' ):
+        if name in ( 'topos', 'switches', 'hosts', 'controllers', 'links' ):
             # Update dictionaries
             param = name.upper()
             globals()[ param ].update( value )
@@ -197,7 +197,7 @@ class MininetRunner( object ):
         opts = OptionParser( description=desc, usage=usage )
         addDictOption( opts, SWITCHES, SWITCHDEF, 'switch' )
         addDictOption( opts, HOSTS, HOSTDEF, 'host' )
-        addDictOption( opts, CONTROLLERS, CONTROLLERDEF, 'controller' )
+        addDictOption( opts, CONTROLLERS, [], 'controller', action='append' )
         addDictOption( opts, LINKS, LINKDEF, 'link' )
         addDictOption( opts, TOPOS, TOPODEF, 'topo' )
 
@@ -238,7 +238,7 @@ class MininetRunner( object ):
                          default=False, help="pin hosts to CPU cores "
                          "(requires --host cfs or --host rt)" )
         opts.add_option( '--nat', action='callback', callback=self.setNat,
-                         help="adds a NAT to the topology that"
+                         help="[option=val...] adds a NAT to the topology that"
                          " connects Mininet hosts to the physical network."
                          " Warning: This may route any traffic on the machine"
                          " that uses Mininet's"
@@ -280,34 +280,39 @@ class MininetRunner( object ):
     def begin( self ):
         "Create and run mininet."
 
+        if self.options.cluster:
+            servers = self.options.cluster.split( ',' )
+            for server in servers:
+                ClusterCleanup.add( server )
+
         if self.options.clean:
             cleanup()
             exit()
 
         start = time.time()
 
-        if self.options.controller == 'default':
+        if not self.options.controller:
             # Update default based on available controllers
             CONTROLLERS[ 'default' ] = findController()
-            if CONTROLLERS[ 'default' ] is None:
+            self.options.controller = [ 'default' ]
+            if not CONTROLLERS[ 'default' ]:
+                self.options.controller = [ 'none' ]
                 if self.options.switch == 'default':
                     info( '*** No default OpenFlow controller found '
                           'for default switch!\n' )
                     info( '*** Falling back to OVS Bridge\n' )
                     self.options.switch = 'ovsbr'
-                    self.options.controller = 'none'
-                elif self.options.switch in ( 'ovsbr', 'lxbr' ):
-                    self.options.controller = 'none'
-                else:
+                elif self.options.switch not in ( 'ovsbr', 'lxbr' ):
                     raise Exception( "Could not find a default controller "
                                      "for switch %s" %
                                      self.options.switch )
 
         topo = buildTopo( TOPOS, self.options.topo )
-        switch = customConstructor( SWITCHES, self.options.switch )
-        host = customConstructor( HOSTS, self.options.host )
-        controller = customConstructor( CONTROLLERS, self.options.controller )
-        link = customConstructor( LINKS, self.options.link )
+        switch = customClass( SWITCHES, self.options.switch )
+        host = customClass( HOSTS, self.options.host )
+        controller = [ customClass( CONTROLLERS, c )
+                       for c in self.options.controller ]
+        link = customClass( LINKS, self.options.link )
 
         if self.validate:
             self.validate( self.options )
@@ -330,10 +335,8 @@ class MininetRunner( object ):
         Net = MininetWithControlNet if inNamespace else Mininet
         cli = ClusterCLI if cluster else CLI
         if cluster:
-            warn( '*** WARNING: Experimental cluster mode!\n'
-                  '*** Using RemoteHost, RemoteOVSSwitch, RemoteLink\n' )
-            host, switch, link = RemoteHost, RemoteOVSSwitch, RemoteLink
-            Net = partial( MininetCluster, servers=cluster.split( ',' ),
+            warn( '*** WARNING: Experimental cluster mode!\n' )
+            Net = partial( MininetCluster, servers=servers,
                            placement=PLACEMENT[ self.options.placement ] )
 
         mn = Net( topo=topo,

examples/cluster-old.py

@@ -0,0 +1,918 @@
+#!/usr/bin/python
+
+"""
+cluster.py: prototyping/experimentation for distributed Mininet,
+            aka Mininet: Cluster Edition
+
+Author: Bob Lantz
+
+Core classes:
+
+    RemoteNode: a Node() running on a remote server
+    RemoteOVSSwitch(): an OVSSwitch() running on a remote server
+    RemoteLink: a Link() on a remote server
+    Tunnel: a Link() between a local Node() and a RemoteNode()
+
+These are largely interoperable with local objects.
+
+- One Mininet to rule them all
+
+It is important that the same topologies, APIs, and CLI can be used
+with minimal or no modification in both local and distributed environments.
+
+- Multiple placement models
+
+Placement should be as easy as possible. We should provide basic placement
+support and also allow for explicit placement.
+
+Questions:
+
+What is the basic communication mechanism?
+
+To start with? Probably a single multiplexed ssh connection between each
+pair of mininet servers that needs to communicate.
+
+How are tunnels created?
+
+We have several options including ssh, GRE, OF capsulator, socat, VDE, l2tp,
+etc..  It's not clear what the best one is.  For now, we use ssh tunnels since
+they are encrypted and semi-automatically shared.  We will probably want to
+support GRE as well because it's very easy to set up with OVS.
+
+How are tunnels destroyed?
+
+They are destroyed when the links are deleted in Mininet.stop()
+
+How does RemoteNode.popen() work?
+
+It opens a shared ssh connection to the remote server and attaches to
+the namespace using mnexec -a -g.
+
+Is there any value to using Paramiko vs. raw ssh?
+
+Maybe, but it doesn't seem to support L2 tunneling.
+
+Should we preflight the entire network, including all server-to-server
+connections?
+
+Yes! We don't yet do this with remote server-to-server connections yet.
+
+Should we multiplex the link ssh connections?
+
+Yes, this is done automatically with ControlMaster=auto.
+
+Note on ssh and DNS:
+Please add UseDNS: no to your /etc/ssh/sshd_config!!!
+
+Things to do:
+
+- asynchronous/pipelined/parallel startup
+- ssh debugging/profiling
+- make connections into real objects
+- support for other tunneling schemes
+- tests and benchmarks
+- hifi support (e.g. delay compensation)
+"""
+
+from mininet.node import Node, Host, OVSSwitch, Controller
+from mininet.link import Link, Intf
+from mininet.net import Mininet
+from mininet.topo import LinearTopo
+from mininet.topolib import TreeTopo
+from mininet.util import quietRun, errRun
+from mininet.examples.clustercli import CLI
+from mininet.log import setLogLevel, debug, info, error
+from mininet.clean import addCleanupCallback
+
+from signal import signal, SIGINT, SIG_IGN
+from subprocess import Popen, PIPE, STDOUT
+import os
+from random import randrange
+import sys
+import re
+from itertools import groupby
+from operator import attrgetter
+from distutils.version import StrictVersion
+
+
+def findUser():
+    "Try to return logged-in (usually non-root) user"
+    return (
+            # If we're running sudo
+            os.environ.get( 'SUDO_USER', False ) or
+            # Logged-in user (if we have a tty)
+            ( quietRun( 'who am i' ).split() or [ False ] )[ 0 ] or
+            # Give up and return effective user
+            quietRun( 'whoami' ).strip() )
+
+
+class ClusterCleanup( object ):
+    "Cleanup callback"
+
+    inited = False
+    serveruser = {}
+
+    @classmethod
+    def add( cls, server, user='' ):
+        "Add an entry to server: user dict"
+        if not cls.inited:
+            addCleanupCallback( cls.cleanup )
+        if not user:
+            user = findUser()
+        cls.serveruser[ server ] = user
+
+    @classmethod
+    def cleanup( cls ):
+        "Clean up"
+        info( '*** Cleaning up cluster\n' )
+        for server, user in cls.serveruser.iteritems():
+            if server == 'localhost':
+                # Handled by mininet.clean.cleanup()
+                continue
+            else:
+                cmd = [ 'su', user, '-c',
+                        'ssh %s@%s sudo mn -c' % ( user, server ) ]
+                info( cmd, '\n' )
+                info( quietRun( cmd ) )
+
+# BL note: so little code is required for remote nodes,
+# we will probably just want to update the main Node()
+# class to enable it for remote access! However, there
+# are a large number of potential failure conditions with
+# remote nodes which we may want to detect and handle.
+# Another interesting point is that we could put everything
+# in a mix-in class and easily add cluster mode to 2.0.
+
+class RemoteMixin( object ):
+
+    "A mix-in class to turn local nodes into remote nodes"
+
+    # ssh base command
+    # -q: don't print stupid diagnostic messages
+    # BatchMode yes: don't ask for password
+    # ForwardAgent yes: forward authentication credentials
+    sshbase = [ 'ssh', '-q',
+                '-o', 'BatchMode=yes',
+                '-o', 'ForwardAgent=yes', '-tt' ]
+
+    def __init__( self, name, server='localhost', user=None, serverIP=None,
+                  controlPath=True, splitInit=False, **kwargs):
+        """Instantiate a remote node
+           name: name of remote node
+           server: remote server (optional)
+           user: user on remote server (optional)
+           controlPath: specify shared ssh control path (optional)
+           splitInit: split initialization?
+           **kwargs: see Node()"""
+        # We connect to servers by IP address
+        self.server = server if server else 'localhost'
+        self.serverIP = ( serverIP if serverIP
+                          else self.findServerIP( self.server ) )
+        self.user = user if user else findUser()
+        ClusterCleanup.add( server=server, user=user )
+        if controlPath is True:
+            # Set a default control path for shared SSH connections
+            controlPath = '/tmp/mn-%r@%h:%p'
+        self.controlPath = controlPath
+        self.splitInit = splitInit
+        if self.user and self.server != 'localhost':
+            self.dest = '%s@%s' % ( self.user, self.serverIP )
+            self.sshcmd = [ 'sudo', '-E', '-u', self.user ] + self.sshbase
+            if self.controlPath:
+                self.sshcmd += [ '-o', 'ControlPath=' + self.controlPath,
+                                 '-o', 'ControlMaster=auto',
+                                 '-o', 'ControlPersist=' + '1' ]
+            self.sshcmd += [ self.dest ]
+            self.isRemote = True
+        else:
+            self.dest = None
+            self.sshcmd = []
+            self.isRemote = False
+        # Satisfy pylint
+        self.shell, self.pid = None, None
+        super( RemoteMixin, self ).__init__( name, **kwargs )
+
+    # Determine IP address of local host
+    _ipMatchRegex = re.compile( r'\d+\.\d+\.\d+\.\d+' )
+
+    @classmethod
+    def findServerIP( cls, server ):
+        "Return our server's IP address"
+        # First, check for an IP address
+        ipmatch = cls._ipMatchRegex.findall( server )
+        if ipmatch:
+            return ipmatch[ 0 ]
+        # Otherwise, look up remote server
+        output = quietRun( 'getent ahostsv4 %s' % server )
+        ips = cls._ipMatchRegex.findall( output )
+        ip = ips[ 0 ] if ips else None
+        return ip
+
+    # Command support via shell process in namespace
+    def startShell( self, *args, **kwargs ):
+        "Start a shell process for running commands"
+        if self.isRemote:
+            kwargs.update( mnopts='-cp' )
+        super( RemoteMixin, self ).startShell( *args, **kwargs )
+        # Optional split initialization
+        self.sendCmd( 'echo $$' )
+        if not self.splitInit:
+            self.finishInit()
+
+    def finishInit( self ):
+        "Wait for split initialization to complete"
+        self.pid = int( self.waitOutput() )
+
+    def rpopen( self, *cmd, **opts ):
+        "Return a Popen object on underlying server in root namespace"
+        params = { 'stdin': PIPE,
+                   'stdout': PIPE,
+                   'stderr': STDOUT,
+                   'sudo': True }
+        params.update( opts )
+        return self._popen( *cmd, **params )
+
+    def rcmd( self, *cmd, **opts):
+        """rcmd: run a command on underlying server
+           in root namespace
+           args: string or list of strings
+           returns: stdout and stderr"""
+        popen = self.rpopen( *cmd, **opts )
+        # print 'RCMD: POPEN:', popen
+        # These loops are tricky to get right.
+        # Once the process exits, we can read
+        # EOF twice if necessary.
+        result = ''
+        while True:
+            poll = popen.poll()
+            result += popen.stdout.read()
+            if poll is not None:
+                break
+        return result
+
+    @staticmethod
+    def _ignoreSignal():
+        "Detach from process group to ignore all signals"
+        os.setpgrp()
+
+    def _popen( self, cmd, sudo=True, tt=True, **params):
+        """Spawn a process on a remote node
+            cmd: remote command to run (list)
+            **params: parameters to Popen()
+            returns: Popen() object"""
+        if type( cmd ) is str:
+            cmd = cmd.split()
+        if self.isRemote:
+            if sudo:
+                cmd = [ 'sudo', '-E' ] + cmd
+            if tt:
+                cmd = self.sshcmd + cmd
+            else:
+                # Hack: remove -tt
+                sshcmd = list( self.sshcmd )
+                sshcmd.remove( '-tt' )
+                cmd = sshcmd + cmd
+        else:
+            if self.user and not sudo:
+                # Drop privileges
+                cmd = [ 'sudo', '-E', '-u', self.user ] + cmd
+        params.update( preexec_fn=self._ignoreSignal )
+        debug( '_popen', cmd, '\n' )
+        popen = super( RemoteMixin, self )._popen( cmd, **params )
+        return popen
+
+    def popen( self, *args, **kwargs ):
+        "Override: disable -tt"
+        return super( RemoteMixin, self).popen( *args, tt=False, **kwargs )
+
+    def addIntf( self, *args, **kwargs ):
+        "Override: use RemoteLink.moveIntf"
+        kwargs.update( moveIntfFn=RemoteLink.moveIntf )
+        return super( RemoteMixin, self).addIntf( *args, **kwargs )
+
+
+class RemoteNode( RemoteMixin, Node ):
+    "A node on a remote server"
+    pass
+
+
+class RemoteHost( RemoteNode ):
+    "A RemoteHost is simply a RemoteNode"
+    pass
+
+
+class RemoteOVSSwitch( RemoteMixin, OVSSwitch ):
+    "Remote instance of Open vSwitch"
+
+    OVSVersions = {}
+
+    def __init__( self, *args, **kwargs ):
+        # No batch startup yet
+        kwargs.update( batch=True )
+        super( RemoteOVSSwitch, self ).__init__( *args, **kwargs )
+
+    def isOldOVS( self ):
+        "Is remote switch using an old OVS version?"
+        cls = type( self )
+        if self.server not in cls.OVSVersions:
+            # pylint: disable=not-callable
+            vers = self.cmd( 'ovs-vsctl --version' )
+            # pylint: enable=not-callable
+            cls.OVSVersions[ self.server ] = re.findall(
+                r'\d+\.\d+', vers )[ 0 ]
+        return ( StrictVersion( cls.OVSVersions[ self.server ] ) <
+                 StrictVersion( '1.10' ) )
+
+    @classmethod
+    def batchStartup( cls, switches, **_kwargs ):
+        "Start up switches in per-server batches"
+        key = attrgetter( 'server' )
+        for server, switchGroup in groupby( sorted( switches, key=key ), key ):
+            info( '(%s)' % server )
+            group = tuple( switchGroup )
+            switch = group[ 0 ]
+            OVSSwitch.batchStartup( group, run=switch.cmd )
+        return switches
+
+    @classmethod
+    def batchShutdown( cls, switches, **_kwargs ):
+        "Stop switches in per-server batches"
+        key = attrgetter( 'server' )
+        for server, switchGroup in groupby( sorted( switches, key=key ), key ):
+            info( '(%s)' % server )
+            group = tuple( switchGroup )
+            switch = group[ 0 ]
+            OVSSwitch.batchShutdown( group, run=switch.rcmd )
+        return switches
+
+
+class RemoteLink( Link ):
+    "A RemoteLink is a link between nodes which may be on different servers"
+
+    def __init__( self, node1, node2, **kwargs ):
+        """Initialize a RemoteLink
+           see Link() for parameters"""
+        # Create links on remote node
+        self.node1 = node1
+        self.node2 = node2
+        self.tunnel = None
+        kwargs.setdefault( 'params1', {} )
+        kwargs.setdefault( 'params2', {} )
+        self.cmd = None  # satisfy pylint
+        Link.__init__( self, node1, node2, **kwargs )
+
+    def stop( self ):
+        "Stop this link"
+        if self.tunnel:
+            self.tunnel.terminate()
+            self.intf1.delete()
+            self.intf2.delete()
+        else:
+            Link.stop( self )
+        self.tunnel = None
+
+    def makeIntfPair( self, intfname1, intfname2, addr1=None, addr2=None,
+                      node1=None, node2=None, deleteIntfs=True   ):
+        """Create pair of interfaces
+            intfname1: name of interface 1
+            intfname2: name of interface 2
+            (override this method [and possibly delete()]
+            to change link type)"""
+        node1 = self.node1 if node1 is None else node1
+        node2 = self.node2 if node2 is None else node2
+        server1 = getattr( node1, 'server', 'localhost' )
+        server2 = getattr( node2, 'server', 'localhost' )
+        if server1 == server2:
+            # Link within same server
+            return Link.makeIntfPair( intfname1, intfname2, addr1, addr2,
+                                      node1, node2, deleteIntfs=deleteIntfs,
+                                      runCmd=None )
+        # Otherwise, make a tunnel
+        self.tunnel = self.makeTunnel( node1, node2, intfname1, intfname2,
+                                       addr1, addr2 )
+        return self.tunnel
+
+    @staticmethod
+    def moveIntf( intf, node, printError=True ):
+        """Move remote interface from root ns to node
+            intf: string, interface
+            dstNode: destination Node
+            srcNode: source Node or None (default) for root ns
+            printError: if true, print error"""
+        intf = str( intf )
+        cmd = 'ip link set %s netns %s' % ( intf, node.pid )
+        node.rcmd( cmd )
+        links = node.cmd( 'ip link show' )
+        if not ' %s:' % intf in links:
+            if printError:
+                error( '*** Error: RemoteLink.moveIntf: ' + intf +
+                       ' not successfully moved to ' + node.name + '\n' )
+            return False
+        return True
+
+    def makeTunnel( self, node1, node2, intfname1, intfname2,
+                    addr1=None, addr2=None ):
+        "Make a tunnel across switches on different servers"
+        # We should never try to create a tunnel to ourselves!
+        assert node1.server != 'localhost' or node2.server != 'localhost'
+        # And we can't ssh into this server remotely as 'localhost',
+        # so try again swappping node1 and node2
+        if node2.server == 'localhost':
+            return self.makeTunnel( node2, node1, intfname2, intfname1,
+                                    addr2, addr1 )
+        # 1. Create tap interfaces
+        for node in node1, node2:
+            # For now we are hard-wiring tap9, which we will rename
+            cmd = 'ip tuntap add dev tap9 mode tap user ' + node.user
+            result = node.rcmd( cmd )
+            if result:
+                raise Exception( 'error creating tap9 on %s: %s' %
+                                 ( node, result ) )
+        # 2. Create ssh tunnel between tap interfaces
+        # -n: close stdin
+        dest = '%s@%s' % ( node2.user, node2.serverIP )
+        cmd = [ 'ssh', '-n', '-o', 'Tunnel=Ethernet', '-w', '9:9',
+                dest, 'echo @' ]
+        self.cmd = cmd
+        tunnel = node1.rpopen( cmd, sudo=False )
+        # When we receive the character '@', it means that our
+        # tunnel should be set up
+        debug( 'Waiting for tunnel to come up...\n' )
+        ch = tunnel.stdout.read( 1 )
+        if ch != '@':
+            raise Exception( 'makeTunnel:\n',
+                             'Tunnel setup failed for',
+                             '%s:%s' % ( node1, node1.dest ), 'to',
+                             '%s:%s\n' % ( node2, node2.dest ),
+                             'command was:', cmd, '\n' )
+        # 3. Move interfaces if necessary
+        for node in node1, node2:
+            if not self.moveIntf( 'tap9', node ):
+                raise Exception( 'interface move failed on node %s' % node )
+        # 4. Rename tap interfaces to desired names
+        for node, intf, addr in ( ( node1, intfname1, addr1 ),
+                                  ( node2, intfname2, addr2 ) ):
+            if not addr:
+                result = node.cmd( 'ip link set tap9 name', intf )
+            else:
+                result = node.cmd( 'ip link set tap9 name', intf,
+                                   'address', addr )
+            if result:
+                raise Exception( 'error renaming %s: %s' % ( intf, result ) )
+        return tunnel
+
+    def status( self ):
+        "Detailed representation of link"
+        if self.tunnel:
+            if self.tunnel.poll() is not None:
+                status = "Tunnel EXITED %s" % self.tunnel.returncode
+            else:
+                status = "Tunnel Running (%s: %s)" % (
+                    self.tunnel.pid, self.cmd )
+        else:
+            status = "OK"
+        result = "%s %s" % ( Link.status( self ), status )
+        return result
+
+
+# Some simple placement algorithms for MininetCluster
+
+class Placer( object ):
+    "Node placement algorithm for MininetCluster"
+
+    def __init__( self, servers=None, nodes=None, hosts=None,
+                  switches=None, controllers=None, links=None ):
+        """Initialize placement object
+           servers: list of servers
+           nodes: list of all nodes
+           hosts: list of hosts
+           switches: list of switches
+           controllers: list of controllers
+           links: list of links
+           (all arguments are optional)
+           returns: server"""
+        self.servers = servers or []
+        self.nodes = nodes or []
+        self.hosts = hosts or []
+        self.switches = switches or []
+        self.controllers = controllers or []
+        self.links = links or []
+
+    def place( self, node ):
+        "Return server for a given node"
+        assert self, node  # satisfy pylint
+        # Default placement: run locally
+        return 'localhost'
+
+
+class RandomPlacer( Placer ):
+    "Random placement"
+    def place( self, nodename ):
+        """Random placement function
+            nodename: node name"""
+        assert nodename  # please pylint
+        # This may be slow with lots of servers
+        return self.servers[ randrange( 0, len( self.servers ) ) ]
+
+
+class RoundRobinPlacer( Placer ):
+    """Round-robin placement
+       Note this will usually result in cross-server links between
+       hosts and switches"""
+
+    def __init__( self, *args, **kwargs ):
+        Placer.__init__( self, *args, **kwargs )
+        self.next = 0
+
+    def place( self, nodename ):
+        """Round-robin placement function
+            nodename: node name"""
+        assert nodename  # please pylint
+        # This may be slow with lots of servers
+        server = self.servers[ self.next ]
+        self.next = ( self.next + 1 ) % len( self.servers )
+        return server
+
+
+class SwitchBinPlacer( Placer ):
+    """Place switches (and controllers) into evenly-sized bins,
+       and attempt to co-locate hosts and switches"""
+
+    def __init__( self, *args, **kwargs ):
+        Placer.__init__( self, *args, **kwargs )
+        # Easy lookup for servers and node sets
+        self.servdict = dict( enumerate( self.servers ) )
+        self.hset = frozenset( self.hosts )
+        self.sset = frozenset( self.switches )
+        self.cset = frozenset( self.controllers )
+        # Server and switch placement indices
+        self.placement = self.calculatePlacement()
+
+    @staticmethod
+    def bin( nodes, servers ):
+        "Distribute nodes evenly over servers"
+        # Calculate base bin size
+        nlen = len( nodes )
+        slen = len( servers )
+        # Basic bin size
+        quotient = int( nlen / slen )
+        binsizes = { server: quotient for server in servers }
+        # Distribute remainder
+        remainder = nlen % slen
+        for server in servers[ 0 : remainder ]:
+            binsizes[ server ] += 1
+        # Create binsize[ server ] tickets for each server
+        tickets = sum( [ binsizes[ server ] * [ server ]
+                         for server in servers ], [] )
+        # And assign one ticket to each node
+        return { node: ticket for node, ticket in zip( nodes, tickets ) }
+
+    def calculatePlacement( self ):
+        "Pre-calculate node placement"
+        placement = {}
+        # Create host-switch connectivity map,
+        # associating host with last switch that it's
+        # connected to
+        switchFor = {}
+        for src, dst in self.links:
+            if src in self.hset and dst in self.sset:
+                switchFor[ src ] = dst
+            if dst in self.hset and src in self.sset:
+                switchFor[ dst ] = src
+        # Place switches
+        placement = self.bin( self.switches, self.servers )
+        # Place controllers and merge into placement dict
+        placement.update( self.bin( self.controllers, self.servers ) )
+        # Co-locate hosts with their switches
+        for h in self.hosts:
+            if h in placement:
+                # Host is already placed - leave it there
+                continue
+            if h in switchFor:
+                placement[ h ] = placement[ switchFor[ h ] ]
+            else:
+                raise Exception(
+                        "SwitchBinPlacer: cannot place isolated host " + h )
+        return placement
+
+    def place( self, node ):
+        """Simple placement algorithm:
+           place switches into evenly sized bins,
+           and place hosts near their switches"""
+        return self.placement[ node ]
+
+
+class HostSwitchBinPlacer( Placer ):
+    """Place switches *and hosts* into evenly-sized bins
+       Note that this will usually result in cross-server
+       links between hosts and switches"""
+
+    def __init__( self, *args, **kwargs ):
+        Placer.__init__( self, *args, **kwargs )
+        # Calculate bin sizes
+        scount = len( self.servers )
+        self.hbin = max( int( len( self.hosts ) / scount ), 1 )
+        self.sbin = max( int( len( self.switches ) / scount ), 1 )
+        self.cbin = max( int( len( self.controllers ) / scount ), 1 )
+        info( 'scount:', scount )
+        info( 'bins:', self.hbin, self.sbin, self.cbin, '\n' )
+        self.servdict = dict( enumerate( self.servers ) )
+        self.hset = frozenset( self.hosts )
+        self.sset = frozenset( self.switches )
+        self.cset = frozenset( self.controllers )
+        self.hind, self.sind, self.cind = 0, 0, 0
+
+    def place( self, nodename ):
+        """Simple placement algorithm:
+            place nodes into evenly sized bins"""
+        # Place nodes into bins
+        if nodename in self.hset:
+            server = self.servdict[ self.hind / self.hbin ]
+            self.hind += 1
+        elif nodename in self.sset:
+            server = self.servdict[ self.sind / self.sbin ]
+            self.sind += 1
+        elif nodename in self.cset:
+            server = self.servdict[ self.cind / self.cbin ]
+            self.cind += 1
+        else:
+            info( 'warning: unknown node', nodename )
+            server = self.servdict[ 0 ]
+        return server
+
+
+# The MininetCluster class is not strictly necessary.
+# However, it has several purposes:
+# 1. To set up ssh connection sharing/multiplexing
+# 2. To pre-flight the system so that everything is more likely to work
+# 3. To allow connection/connectivity monitoring
+# 4. To support pluggable placement algorithms
+
+class MininetCluster( Mininet ):
+
+    "Cluster-enhanced version of Mininet class"
+
+    # Default ssh command
+    # BatchMode yes: don't ask for password
+    # ForwardAgent yes: forward authentication credentials
+    sshcmd = [ 'ssh', '-o', 'BatchMode=yes', '-o', 'ForwardAgent=yes' ]
+
+    def __init__( self, *args, **kwargs ):
+        """servers: a list of servers to use (note: include
+           localhost or None to use local system as well)
+           user: user name for server ssh
+           placement: Placer() subclass"""
+        params = { 'host': RemoteHost,
+                   'switch': RemoteOVSSwitch,
+                   'link': RemoteLink,
+                   'precheck': True }
+        params.update( kwargs )
+        servers = params.pop( 'servers', [ 'localhost' ] )
+        servers = [ s if s else 'localhost' for s in servers ]
+        self.servers = servers
+        self.serverIP = params.pop( 'serverIP', {} )
+        if not self.serverIP:
+            self.serverIP = { server: RemoteMixin.findServerIP( server )
+                              for server in self.servers }
+        self.user = params.pop( 'user', findUser() )
+        if params.pop( 'precheck' ):
+            self.precheck()
+        self.connections = {}
+        self.placement = params.pop( 'placement', SwitchBinPlacer )
+        # Make sure control directory exists
+        self.cdir = os.environ[ 'HOME' ] + '/.ssh/mn'
+        errRun( [ 'mkdir', '-p', self.cdir ] )
+        Mininet.__init__( self, *args, **params )
+
+    def popen( self, cmd ):
+        "Popen() for server connections"
+        assert self  # please pylint
+        old = signal( SIGINT, SIG_IGN )
+        conn = Popen( cmd, stdin=PIPE, stdout=PIPE, close_fds=True )
+        signal( SIGINT, old )
+        return conn
+
+    def baddLink( self, *args, **kwargs ):
+        "break addlink for testing"
+        pass
+
+    def precheck( self ):
+        """Pre-check to make sure connection works and that
+           we can call sudo without a password"""
+        result = 0
+        info( '*** Checking servers\n' )
+        for server in self.servers:
+            ip = self.serverIP[ server ]
+            if not server or server == 'localhost':
+                continue
+            info( server, '' )
+            dest = '%s@%s' % ( self.user, ip )
+            cmd = [ 'sudo', '-E', '-u', self.user ]
+            cmd += self.sshcmd + [ '-n', dest, 'sudo true' ]
+            debug( ' '.join( cmd ), '\n' )
+            _out, _err, code = errRun( cmd )
+            if code != 0:
+                error( '\nstartConnection: server connection check failed '
+                       'to %s using command:\n%s\n'
+                        % ( server, ' '.join( cmd ) ) )
+            result |= code
+        if result:
+            error( '*** Server precheck failed.\n'
+                   '*** Make sure that the above ssh command works'
+                   ' correctly.\n'
+                   '*** You may also need to run mn -c on all nodes, and/or\n'
+                   '*** use sudo -E.\n' )
+            sys.exit( 1 )
+        info( '\n' )
+
+    def modifiedaddHost( self, *args, **kwargs ):
+        "Slightly modify addHost"
+        assert self  # please pylint
+        kwargs[ 'splitInit' ] = True
+        return Mininet.addHost( *args, **kwargs )
+
+    def placeNodes( self ):
+        """Place nodes on servers (if they don't have a server), and
+           start shell processes"""
+        if not self.servers or not self.topo:
+            # No shirt, no shoes, no service
+            return
+        nodes = self.topo.nodes()
+        placer = self.placement( servers=self.servers,
+                                 nodes=self.topo.nodes(),
+                                 hosts=self.topo.hosts(),
+                                 switches=self.topo.switches(),
+                                 links=self.topo.links() )
+        for node in nodes:
+            config = self.topo.nodeInfo( node )
+            # keep local server name consistent accross nodes
+            if 'server' in config.keys() and config[ 'server' ] is None:
+                config[ 'server' ] = 'localhost'
+            server = config.setdefault( 'server', placer.place( node ) )
+            if server:
+                config.setdefault( 'serverIP', self.serverIP[ server ] )
+            info( '%s:%s ' % ( node, server ) )
+            key = ( None, server )
+            _dest, cfile, _conn = self.connections.get(
+                        key, ( None, None, None ) )
+            if cfile:
+                config.setdefault( 'controlPath', cfile )
+
+    def addController( self, *args, **kwargs ):
+        "Patch to update IP address to global IP address"
+        controller = Mininet.addController( self, *args, **kwargs )
+        # Update IP address for controller that may not be local
+        if ( isinstance( controller, Controller )
+             and controller.IP() == '127.0.0.1' ):
+            links = controller.cmd( 'ip link show' )
+            eth0 = re.findall( ' (.*eth0):', links )
+            if not eth0:
+                raise Exception( 'Cannot find IP address for controller eth0' )
+            Intf( eth0[ 0 ], node=controller ).updateIP()
+        return controller
+
+    def buildFromTopo( self, *args, **kwargs ):
+        "Start network"
+        info( '*** Placing nodes\n' )
+        self.placeNodes()
+        info( '\n' )
+        Mininet.buildFromTopo( self, *args, **kwargs )
+
+
+def testNsTunnels():
+    "Test tunnels between nodes in namespaces"
+    net = Mininet( host=RemoteHost, link=RemoteLink )
+    h1 = net.addHost( 'h1' )
+    h2 = net.addHost( 'h2', server='ubuntu2' )
+    net.addLink( h1, h2 )
+    net.start()
+    net.pingAll()
+    net.stop()
+
+# Manual topology creation with net.add*()
+#
+# This shows how node options may be used to manage
+# cluster placement using the net.add*() API
+
+def testRemoteNet( remote='ubuntu2' ):
+    "Test remote Node classes"
+    print '*** Remote Node Test'
+    net = Mininet( host=RemoteHost, switch=RemoteOVSSwitch,
+                   link=RemoteLink )
+    c0 = net.addController( 'c0' )
+    # Make sure controller knows its non-loopback address
+    Intf( 'eth0', node=c0 ).updateIP()
+    print "*** Creating local h1"
+    h1 = net.addHost( 'h1' )
+    print "*** Creating remote h2"
+    h2 = net.addHost( 'h2', server=remote )
+    print "*** Creating local s1"
+    s1 = net.addSwitch( 's1' )
+    print "*** Creating remote s2"
+    s2 = net.addSwitch( 's2', server=remote )
+    print "*** Adding links"
+    net.addLink( h1, s1 )
+    net.addLink( s1, s2 )
+    net.addLink( h2, s2 )
+    net.start()
+    print 'Mininet is running on', quietRun( 'hostname' ).strip()
+    for node in c0, h1, h2, s1, s2:
+        print 'Node', node, 'is running on', node.cmd( 'hostname' ).strip()
+    net.pingAll()
+    CLI( net )
+    net.stop()
+
+
+# High-level/Topo API example
+#
+# This shows how existing Mininet topologies may be used in cluster
+# mode by creating node placement functions and a controller which
+# can be accessed remotely. This implements a very compatible version
+# of cluster edition with a minimum of code!
+
+remoteHosts = [ 'h2' ]
+remoteSwitches = [ 's2' ]
+remoteServer = 'ubuntu2'
+
+def HostPlacer( name, *args, **params ):
+    "Custom Host() constructor which places hosts on servers"
+    if name in remoteHosts:
+        return RemoteHost( name, *args, server=remoteServer, **params )
+    else:
+        return Host( name, *args, **params )
+
+def SwitchPlacer( name, *args, **params ):
+    "Custom Switch() constructor which places switches on servers"
+    if name in remoteSwitches:
+        return RemoteOVSSwitch( name, *args, server=remoteServer, **params )
+    else:
+        return RemoteOVSSwitch( name, *args, **params )
+
+def ClusterController( *args, **kwargs):
+    "Custom Controller() constructor which updates its eth0 IP address"
+    controller = Controller( *args, **kwargs )
+    # Find out its IP address so that cluster switches can connect
+    Intf( 'eth0', node=controller ).updateIP()
+    return controller
+
+def testRemoteTopo():
+    "Test remote Node classes using Mininet()/Topo() API"
+    topo = LinearTopo( 2 )
+    net = Mininet( topo=topo, host=HostPlacer, switch=SwitchPlacer,
+                   link=RemoteLink, controller=ClusterController )
+    net.start()
+    net.pingAll()
+    net.stop()
+
+# Need to test backwards placement, where each host is on
+# a server other than its switch!! But seriously we could just
+# do random switch placement rather than completely random
+# host placement.
+
+def testRemoteSwitches():
+    "Test with local hosts and remote switches"
+    servers = [ 'localhost', 'ubuntu2']
+    topo = TreeTopo( depth=4, fanout=2 )
+    net = MininetCluster( topo=topo, servers=servers,
+                          placement=RoundRobinPlacer )
+    net.start()
+    net.pingAll()
+    net.stop()
+
+
+#
+# For testing and demo purposes it would be nice to draw the
+# network graph and color it based on server.
+
+# The MininetCluster() class integrates pluggable placement
+# functions, for maximum ease of use. MininetCluster() also
+# pre-flights and multiplexes server connections.
+
+def testMininetCluster():
+    "Test MininetCluster()"
+    servers = [ 'localhost', 'ubuntu2' ]
+    topo = TreeTopo( depth=3, fanout=3 )
+    net = MininetCluster( topo=topo, servers=servers,
+                          placement=SwitchBinPlacer )
+    net.start()
+    net.pingAll()
+    net.stop()
+
+def signalTest():
+    "Make sure hosts are robust to signals"
+    h = RemoteHost( 'h0', server='ubuntu1' )
+    h.shell.send_signal( SIGINT )
+    h.shell.poll()
+    if h.shell.returncode is None:
+        print 'OK: ', h, 'has not exited'
+    else:
+        print 'FAILURE:', h, 'exited with code', h.shell.returncode
+    h.stop()
+
+if __name__ == '__main__':
+    setLogLevel( 'info' )
+    # testRemoteTopo()
+    # testRemoteNet()
+    # testMininetCluster()
+    # testRemoteSwitches()
+    signalTest()

examples/clustercli.py

@@ -19,8 +19,7 @@ class ClusterCLI( CLI ):
                    'magenta', 'pink', 'grey', 'brown',
                    'white' ]
         slen, clen = len( seq ), len( colors )
-        reps = max( 1, slen / clen )
-        colors = colors * reps
+        colors = colors * ( slen / clen + 1 )
         colors = colors[ 0 : slen ]
         return colors
 

examples/controlnet.py

@@ -27,11 +27,18 @@ from mininet.log import setLogLevel, info
 
 class DataController( Controller ):
     """Data Network Controller.
-       patched to avoid checkListening error"""
+       patched to avoid checkListening error and to delete intfs"""
+
     def checkListening( self ):
         "Ignore spurious error"
         pass
 
+    def stop( self, *args, **kwargs ):
+        "Make sure intfs are deleted"
+        kwargs.update( deleteIntfs=True )
+        super( DataController, self ).stop( *args, **kwargs )
+
+
 class MininetFacade( object ):
     """Mininet object facade that allows a single CLI to
        talk to one or more networks"""

examples/fakecluster.py

@@ -0,0 +1,89 @@
+#!/usr/bin/python
+
+"""
+fakecluster.py: a fake cluster for testing Mininet cluster edition!!!
+
+We are going to self-host Mininet by creating a virtual cluster
+for cluster edition.
+
+Note: ssh is kind of a mess - you end up having to do things
+like h1 sudo -E -u openflow ssh 10.2
+"""
+
+from mininet.net import Mininet
+from mininet.nodelib import LinuxBridge, Server
+from mininet.cli import CLI
+from mininet.topo import Topo, SingleSwitchTopo
+from mininet.log import setLogLevel, warn
+from mininet.util import errRun, quietRun
+from mininet.link import Link
+
+from functools import partial
+from sys import argv
+
+class MininetServer( Server ):
+    "A server (for nested Mininet) that runs ssh and ovs"
+
+    privateDirs = [ '/var/run/sshd', '/etc/openvswitch',
+                    '/var/run/openvswitch', '/var/log/openvswitch' ]
+
+    def __init__( self, *args, **kwargs ):
+        "Turn on ovs by default"
+        kwargs.setdefault( 'ovs', True )
+        super( MininetServer, self ).__init__( *args, **kwargs )
+
+    def config( self, **kwargs ):
+        """Configure/start sshd and other stuff
+           ovs: start Open vSwitch?"""
+        self.ovs = kwargs.get( 'ovs' )
+        super( MininetServer, self ).config( **kwargs )
+        if self.ovs:
+            self.service( 'openvswitch-switch start' )
+
+    def terminate( self, *args, **kwargs ):
+        "Shut down services and terminate server"
+        if self.ovs:
+            self.service( 'openvswitch-switch stop' )
+        super( MininetServer, self ).terminate( *args, **kwargs )
+
+
+class ServerLink( Link ):
+    def intfName( self, node, n ):
+        "Override to avoid destruction by cleanup!"
+        # This is kind of ugly... for some reason 'eth0' fails so
+        # we just use 'm1eth0'; however, this should nest reasonably.
+        return ( node.name + 'eth' + repr( n ) if isinstance( node, Server )
+                 else node.name + '-eth' + repr( n ) )
+    def makeIntfPair( self, *args, **kwargs ):
+        "Override to use quietRun"
+        kwargs.update( runCmd=quietRun )
+        super( ServerLink, self ).makeIntfPair( *args, **kwargs )
+
+class ClusterTopo( Topo ):
+    "Cluster topology: m1..mN"
+    def build( self, n ):
+        ms1 = self.addSwitch( 'ms1' )
+        for i in range( 1, n + 1 ):
+            h = self.addHost( 'm%d' % i )
+            self.addLink( h, ms1, cls=ServerLink )
+
+
+def test( serverCount ):
+    "Test this setup"
+    setLogLevel( 'info' )
+    topo = ClusterTopo( serverCount )
+    host = partial( MininetServer, ssh=True, ovs=True)
+    net = Mininet( topo=topo, host=host, switch=LinuxBridge, ipBase='10.0.1.0/24' )
+    MininetServer.updateHostsFiles( net.hosts )
+    # addNAT().configDefault() also connects root namespace to Mininet
+    net.addNAT().configDefault()
+    net.start()
+    CLI( net )
+    net.stop()
+
+hosts = { 'mn': MininetServer }
+links = { 'mn': ServerLink }
+
+if __name__ == '__main__':
+    n = 12 if len( argv ) != 2 else int( argv[ 1 ] )
+    test( n )

examples/linuxrouter.py

@@ -6,20 +6,25 @@ linuxrouter.py: Example network with Linux IP router
 This example converts a Node into a router using IP forwarding
 already built into Linux.
 
-The topology contains a router with three IP subnets:
- - 192.168.1.0/24 (interface IP: 192.168.1.1)
- - 172.16.0.0/12 (interface IP: 172.16.0.1)
- - 10.0.0.0/8 (interface IP: 10.0.0.1)
+The example topology creates a router and three IP subnets:
 
- It also contains three hosts, one in each subnet:
- - h1 (IP: 192.168.1.100)
- - h2 (IP: 172.16.0.100)
- - h3 (IP: 10.0.0.100)
+    - 192.168.1.0/24 (r0-eth1, IP: 192.168.1.1)
+    - 172.16.0.0/12 (r0-eth2, IP: 172.16.0.1)
+    - 10.0.0.0/8 (r0-eth3, IP: 10.0.0.1)
 
- Routing entries can be added to the routing tables of the
- hosts or router using the "ip route add" or "route add" command.
- See the man pages for more details.
+Each subnet consists of a single host connected to
+a single switch:
 
+    r0-eth1 - s1-eth1 - h1-eth0 (IP: 192.168.1.100)
+    r0-eth2 - s2-eth1 - h2-eth0 (IP: 172.16.0.100)
+    r0-eth3 - s3-eth1 - h3-eth0 (IP: 10.0.0.100)
+
+The example relies on default routing entries that are
+automatically created for each router interface, as well
+as 'defaultRoute' parameters for the host interfaces.
+
+Additional routes may be added to the router or hosts by
+executing 'ip route' or 'route' commands on the router or hosts.
 """
 
 from mininet.topo import Topo
@@ -42,29 +47,39 @@ class LinuxRouter( Node ):
 
 
 class NetworkTopo( Topo ):
-    "A simple topology of a router with three subnets (one host in each)."
+    "A LinuxRouter connecting three IP subnets"
 
     def build( self, **_opts ):
-        router = self.addNode( 'r0', cls=LinuxRouter, ip='192.168.1.1/24' )
+
+        defaultIP = '192.168.1.1/24'  # IP address for r0-eth1
+        router = self.addNode( 'r0', cls=LinuxRouter, ip=defaultIP )
+
+        s1, s2, s3 = [ self.addSwitch( s ) for s in 's1', 's2', 's3' ]
+
+        self.addLink( s1, router, intfName2='r0-eth1',
+                      params2={ 'ip' : defaultIP } )  # for clarity
+        self.addLink( s2, router, intfName2='r0-eth2',
+                      params2={ 'ip' : '172.16.0.1/12' } )
+        self.addLink( s3, router, intfName2='r0-eth3',
+                      params2={ 'ip' : '10.0.0.1/8' } )
+
         h1 = self.addHost( 'h1', ip='192.168.1.100/24',
                            defaultRoute='via 192.168.1.1' )
         h2 = self.addHost( 'h2', ip='172.16.0.100/12',
                            defaultRoute='via 172.16.0.1' )
         h3 = self.addHost( 'h3', ip='10.0.0.100/8',
                            defaultRoute='via 10.0.0.1' )
-        self.addLink( h1, router, intfName2='r0-eth1',
-                      params2={ 'ip' : '192.168.1.1/24' } )
-        self.addLink( h2, router, intfName2='r0-eth2',
-                      params2={ 'ip' : '172.16.0.1/12' } )
-        self.addLink( h3, router, intfName2='r0-eth3',
-                      params2={ 'ip' : '10.0.0.1/8' } )
+
+        for h, s in [ (h1, s1), (h2, s2), (h3, s3) ]:
+            self.addLink( h, s )
+
 
 def run():
     "Test linux router"
     topo = NetworkTopo()
-    net = Mininet( topo=topo, controller=None )  # no controller needed
+    net = Mininet( topo=topo )  # controller is used by s1-s3
     net.start()
-    info( '*** Routing Table on Router\n' )
+    info( '*** Routing Table on Router:\n' )
     print net[ 'r0' ].cmd( 'route' )
     CLI( net )
     net.stop()

examples/miniedit.py

@@ -49,7 +49,7 @@ from mininet.log import info, setLogLevel
 from mininet.net import Mininet, VERSION
 from mininet.util import netParse, ipAdd, quietRun
 from mininet.util import buildTopo
-from mininet.util import custom, customConstructor
+from mininet.util import custom, customClass
 from mininet.term import makeTerm, cleanUpScreens
 from mininet.node import Controller, RemoteController, NOX, OVSController
 from mininet.node import CPULimitedHost, Host, Node
@@ -3222,7 +3222,7 @@ class MiniEdit( Frame ):
             return
         self.newTopology()
         topo = buildTopo( TOPOS, self.options.topo )
-        link = customConstructor( LINKS, self.options.link )
+        link = customClass( LINKS, self.options.link )
         importNet = Mininet(topo=topo, build=False, link=link)
         importNet.build()
 

examples/nat.py

@@ -2,11 +2,6 @@
 
 """
 Example to create a Mininet topology and connect it to the internet via NAT
-through eth0 on the host.
-
-Glen Gibb, February 2011
-
-(slight modifications by BL, 5/13)
 """
 
 from mininet.cli import CLI
@@ -14,99 +9,14 @@ from mininet.log import lg
 from mininet.node import Node
 from mininet.topolib import TreeNet
 
-#################################
-def startNAT( root, inetIntf='eth0', subnet='10.0/8' ):
-    """Start NAT/forwarding between Mininet and external network
-    root: node to access iptables from
-    inetIntf: interface for internet access
-    subnet: Mininet subnet (default 10.0/8)="""
-
-    # Identify the interface connecting to the mininet network
-    localIntf = root.defaultIntf()
-
-    # Flush any currently active rules
-    root.cmd( 'iptables -F' )
-    root.cmd( 'iptables -t nat -F' )
-
-    # Create default entries for unmatched traffic
-    root.cmd( 'iptables -P INPUT ACCEPT' )
-    root.cmd( 'iptables -P OUTPUT ACCEPT' )
-    root.cmd( 'iptables -P FORWARD DROP' )
-
-    # Configure NAT
-    root.cmd( 'iptables -I FORWARD -i', localIntf, '-d', subnet, '-j DROP' )
-    root.cmd( 'iptables -A FORWARD -i', localIntf, '-s', subnet, '-j ACCEPT' )
-    root.cmd( 'iptables -A FORWARD -i', inetIntf, '-d', subnet, '-j ACCEPT' )
-    root.cmd( 'iptables -t nat -A POSTROUTING -o ', inetIntf, '-j MASQUERADE' )
-
-    # Instruct the kernel to perform forwarding
-    root.cmd( 'sysctl net.ipv4.ip_forward=1' )
-
-def stopNAT( root ):
-    """Stop NAT/forwarding between Mininet and external network"""
-    # Flush any currently active rules
-    root.cmd( 'iptables -F' )
-    root.cmd( 'iptables -t nat -F' )
-
-    # Instruct the kernel to stop forwarding
-    root.cmd( 'sysctl net.ipv4.ip_forward=0' )
-
-def fixNetworkManager( root, intf ):
-    """Prevent network-manager from messing with our interface,
-       by specifying manual configuration in /etc/network/interfaces
-       root: a node in the root namespace (for running commands)
-       intf: interface name"""
-    cfile = '/etc/network/interfaces'
-    line = '\niface %s inet manual\n' % intf
-    config = open( cfile ).read()
-    if line not in config:
-        print '*** Adding', line.strip(), 'to', cfile
-        with open( cfile, 'a' ) as f:
-            f.write( line )
-        # Probably need to restart network-manager to be safe -
-        # hopefully this won't disconnect you
-        root.cmd( 'service network-manager restart' )
-
-def connectToInternet( network, switch='s1', rootip='10.254', subnet='10.0/8'):
-    """Connect the network to the internet
-       switch: switch to connect to root namespace
-       rootip: address for interface in root namespace
-       subnet: Mininet subnet"""
-    switch = network.get( switch )
-    prefixLen = subnet.split( '/' )[ 1 ]
-
-    # Create a node in root namespace
-    root = Node( 'root', inNamespace=False )
-
-    # Prevent network-manager from interfering with our interface
-    fixNetworkManager( root, 'root-eth0' )
-
-    # Create link between root NS and switch
-    link = network.addLink( root, switch )
-    link.intf1.setIP( rootip, prefixLen )
-
-    # Start network that now includes link to root namespace
-    network.start()
-
-    # Start NAT and establish forwarding
-    startNAT( root )
-
-    # Establish routes from end hosts
-    for host in network.hosts:
-        host.cmd( 'ip route flush root 0/0' )
-        host.cmd( 'route add -net', subnet, 'dev', host.defaultIntf() )
-        host.cmd( 'route add default gw', rootip )
-
-    return root
-
 if __name__ == '__main__':
     lg.setLogLevel( 'info')
     net = TreeNet( depth=1, fanout=4 )
-    # Configure and start NATted connectivity
-    rootnode = connectToInternet( net )
+    # Add NAT connectivity
+    net.addNAT().configDefault()
+    net.start()
     print "*** Hosts are running and should have internet connectivity"
     print "*** Type 'exit' or control-D to shut down network"
     CLI( net )
     # Shut down NAT
-    stopNAT( rootnode )
     net.stop()

examples/sshd.py

@@ -23,7 +23,6 @@ from mininet.cli import CLI
 from mininet.log import lg
 from mininet.node import Node
 from mininet.topolib import TreeTopo
-from mininet.link import Link
 from mininet.util import waitListening
 
 def TreeNet( depth=1, fanout=2, **kwargs ):
@@ -39,7 +38,7 @@ def connectToRootNS( network, switch, ip, routes ):
       routes: host networks to route to"""
     # Create a node in root namespace and link to switch 0
     root = Node( 'root', inNamespace=False )
-    intf = Link( root, switch ).intf1
+    intf = network.addLink( root, switch ).intf1
     root.setIP( ip, intf=intf )
     # Start network that now includes link to root namespace
     network.start()

examples/tree1024.py

@@ -9,10 +9,10 @@ and running sysctl -p. Check util/sysctl_addon.
 
 from mininet.cli import CLI
 from mininet.log import setLogLevel
-from mininet.node import OVSKernelSwitch
+from mininet.node import OVSSwitch
 from mininet.topolib import TreeNet
 
 if __name__ == '__main__':
     setLogLevel( 'info' )
-    network = TreeNet( depth=2, fanout=32, switch=OVSKernelSwitch )
+    network = TreeNet( depth=2, fanout=32, switch=OVSSwitch )
     network.run( CLI, network )

mininet/clean.py

@@ -25,73 +25,102 @@ def sh( cmd ):
 
 def killprocs( pattern ):
     "Reliably terminate processes matching a pattern (including args)"
-    sh( 'pkill -9 -f %s' % pattern )
-    # Make sure they are gone
     while True:
         try:
-            pids = co( [ 'pgrep', '-f', pattern ] )
+            pids = co( [ 'pgrep', '-f', pattern ] ).split( '\n' )
         except CalledProcessError:
-            pids = ''
+            pids = []
+        # Don't kill init
+        pids = [ pid for pid in pids if pid and pid != '1' ]
         if pids:
-            sh( 'pkill -9 -f %s' % pattern )
+            sh( "kill -9 %s" % ' '.join( pids ) )
             time.sleep( .5 )
         else:
             break
 
-def cleanup():
-    """Clean up junk which might be left over from old runs;
-       do fast stuff before slow dp and link removal!"""
+class Cleanup( object ):
+    "Wrapper for cleanup()"
 
-    info("*** Removing excess controllers/ofprotocols/ofdatapaths/pings/noxes"
-         "\n")
-    zombies = 'controller ofprotocol ofdatapath ping nox_core lt-nox_core '
-    zombies += 'ovs-openflowd ovs-controller udpbwtest mnexec ivs'
-    # Note: real zombie processes can't actually be killed, since they
-    # are already (un)dead. Then again,
-    # you can't connect to them either, so they're mostly harmless.
-    # Send SIGTERM first to give processes a chance to shutdown cleanly.
-    sh( 'killall ' + zombies + ' 2> /dev/null' )
-    time.sleep( 1 )
-    sh( 'killall -9 ' + zombies + ' 2> /dev/null' )
+    callbacks = []
 
-    # And kill off sudo mnexec
-    sh( 'pkill -9 -f "sudo mnexec"')
+    @classmethod
+    def cleanup( cls):
+        """Clean up junk which might be left over from old runs;
+           do fast stuff before slow dp and link removal!"""
 
-    info( "*** Removing junk from /tmp\n" )
-    sh( 'rm -f /tmp/vconn* /tmp/vlogs* /tmp/*.out /tmp/*.log' )
+        info( "*** Removing excess controllers/ofprotocols/ofdatapaths/"
+              "pings/noxes\n" )
+        zombies = 'controller ofprotocol ofdatapath ping nox_core lt-nox_core '
+        zombies += 'ovs-openflowd ovs-controller udpbwtest mnexec ivs'
+        # Note: real zombie processes can't actually be killed, since they
+        # are already (un)dead. Then again,
+        # you can't connect to them either, so they're mostly harmless.
+        # Send SIGTERM first to give processes a chance to shutdown cleanly.
+        sh( 'killall ' + zombies + ' 2> /dev/null' )
+        time.sleep( 1 )
+        sh( 'killall -9 ' + zombies + ' 2> /dev/null' )
 
-    info( "*** Removing old X11 tunnels\n" )
-    cleanUpScreens()
+        # And kill off sudo mnexec
+        sh( 'pkill -9 -f "sudo mnexec"')
 
-    info( "*** Removing excess kernel datapaths\n" )
-    dps = sh( "ps ax | egrep -o 'dp[0-9]+' | sed 's/dp/nl:/'" ).splitlines()
-    for dp in dps:
-        if dp:
-            sh( 'dpctl deldp ' + dp )
+        info( "*** Removing junk from /tmp\n" )
+        sh( 'rm -f /tmp/vconn* /tmp/vlogs* /tmp/*.out /tmp/*.log' )
 
-    info( "***  Removing OVS datapaths" )
-    dps = sh("ovs-vsctl --timeout=1 list-br").strip().splitlines()
-    if dps:
-        sh( "ovs-vsctl " + " -- ".join( "--if-exists del-br " + dp
-                                        for dp in dps if dp ) )
-    # And in case the above didn't work...
-    dps = sh("ovs-vsctl --timeout=1 list-br").strip().splitlines()
-    for dp in dps:
-        sh( 'ovs-vsctl del-br ' + dp )
+        info( "*** Removing old X11 tunnels\n" )
+        cleanUpScreens()
 
-    info( "*** Removing all links of the pattern foo-ethX\n" )
-    links = sh( "ip link show | "
-                "egrep -o '([-_.[:alnum:]]+-eth[[:digit:]]+)'" ).splitlines()
-    for link in links:
-        if link:
-            sh( "ip link del " + link )
+        info( "*** Removing excess kernel datapaths\n" )
+        dps = sh( "ps ax | egrep -o 'dp[0-9]+' | sed 's/dp/nl:/'"
+                  ).splitlines()
+        for dp in dps:
+            if dp:
+                sh( 'dpctl deldp ' + dp )
 
-    info( "*** Killing stale mininet node processes\n" )
-    killprocs( 'mininet:' )
+        info( "***  Removing OVS datapaths\n" )
+        dps = sh("ovs-vsctl --timeout=1 list-br").strip().splitlines()
+        if dps:
+            sh( "ovs-vsctl " + " -- ".join( "--if-exists del-br " + dp
+                                            for dp in dps if dp ) )
+        # And in case the above didn't work...
+        dps = sh( "ovs-vsctl --timeout=1 list-br" ).strip().splitlines()
+        for dp in dps:
+            sh( 'ovs-vsctl del-br ' + dp )
 
-    info( "*** Shutting down stale tunnels\n" )
-    killprocs( 'Tunnel=Ethernet' )
-    killprocs( '.ssh/mn')
-    sh( 'rm -f ~/.ssh/mn/*' )
+        info( "*** Removing all links of the pattern foo-ethX\n" )
+        links = sh( "ip link show | "
+                    "egrep -o '([-_.[:alnum:]]+-eth[[:digit:]]+)'"
+                    ).splitlines()
+        # Delete blocks of links
+        n = 1000  # chunk size
+        for i in xrange( 0, len( links ), n ):
+            cmd = ';'.join( 'ip link del %s' % link
+                             for link in links[ i : i + n ] )
+            sh( '( %s ) 2> /dev/null' % cmd )
 
-    info( "*** Cleanup complete.\n" )
+        if 'tap9' in sh( 'ip link show' ):
+            info( "*** Removing tap9 - assuming it's from cluster edition\n" )
+            sh( 'ip link del tap9' )
+
+        info( "*** Killing stale mininet node processes\n" )
+        killprocs( 'mininet:' )
+
+        info( "*** Shutting down stale tunnels\n" )
+        killprocs( 'Tunnel=Ethernet' )
+        killprocs( '.ssh/mn')
+        sh( 'rm -f ~/.ssh/mn/*' )
+
+        # Call any additional cleanup code if necessary
+        for callback in cls.callbacks:
+            callback()
+
+        info( "*** Cleanup complete.\n" )
+
+    @classmethod
+    def addCleanupCallback( cls, callback ):
+        "Add cleanup callback"
+        if callback not in cls.callbacks:
+            cls.callbacks.append( callback )
+
+
+cleanup = Cleanup.cleanup
+addCleanupCallback = Cleanup.addCleanupCallback

mininet/cli.py

@@ -45,6 +45,10 @@ class CLI( Cmd ):
     prompt = 'mininet> '
 
     def __init__( self, mininet, stdin=sys.stdin, script=None ):
+        """Start and run interactive or batch mode CLI
+           mininet: Mininet network object
+           stdin: standard input for CLI
+           script: script to run in batch mode"""
         self.mn = mininet
         # Local variable bindings for py command
         self.locals = { 'net': mininet }
@@ -56,33 +60,54 @@ class CLI( Cmd ):
         Cmd.__init__( self )
         info( '*** Starting CLI:\n' )
 
-        # Set up history if readline is available
-        try:
-            import readline
-        except ImportError:
-            pass
-        else:
-            history_path = os.path.expanduser('~/.mininet_history')
-            if os.path.isfile(history_path):
-                readline.read_history_file(history_path)
-            atexit.register(lambda: readline.write_history_file(history_path))
-
         if self.inputFile:
             self.do_source( self.inputFile )
             return
+
+        self.initReadline()
+        self.run()
+
+    readlineInited = False
+
+    @classmethod
+    def initReadline( cls ):
+        "Set up history if readline is available"
+        # Only set up readline once to prevent multiplying the history file
+        if cls.readlineInited:
+            return
+        cls.readlineInited = True
+        try:
+            from readline import read_history_file, write_history_file
+        except ImportError:
+            pass
+        else:
+            history_path = os.path.expanduser( '~/.mininet_history' )
+            if os.path.isfile( history_path ):
+                read_history_file( history_path )
+            atexit.register( lambda: write_history_file( history_path ) )
+
+    def run( self ):
+        "Run our cmdloop(), catching KeyboardInterrupt"
         while True:
             try:
                 # Make sure no nodes are still waiting
                 for node in self.mn.values():
                     while node.waiting:
+                        info( 'stopping', node, '\n' )
                         node.sendInt()
                         node.waitOutput()
                 if self.isatty():
-                    quietRun( 'stty echo sane intr "^C"' )
+                    quietRun( 'stty echo sane intr ^C' )
                 self.cmdloop()
                 break
             except KeyboardInterrupt:
-                output( '\nInterrupt\n' )
+                # Output a message - unless it's also interrupted
+                # pylint: disable=broad-except
+                try:
+                    output( '\nInterrupt\n' )
+                except Exception:
+                    pass
+                # pylint: enable=broad-except
 
     def emptyline( self ):
         "Don't repeat last command when you hit return."

mininet/link.py

@@ -26,6 +26,7 @@ Link: basic link class for creating veth pairs
 
 from mininet.log import info, error, debug
 from mininet.util import makeIntfPair, quietRun
+import mininet.node
 import re
 
 class Intf( object ):
@@ -49,7 +50,11 @@ class Intf( object ):
         if self.name == 'lo':
             self.ip = '127.0.0.1'
         # Add to node (and move ourselves if necessary )
-        node.addIntf( self, port=port )
+        moveIntfFn = params.pop( 'moveIntfFn', None )
+        if moveIntfFn:
+            node.addIntf( self, port=port, moveIntfFn=moveIntfFn )
+        else:
+            node.addIntf( self, port=port )
         # Save params for future reference
         self.params = params
         self.config( **params )
@@ -192,9 +197,10 @@ class Intf( object ):
     def delete( self ):
         "Delete interface"
         self.cmd( 'ip link del ' + self.name )
-        if self.node.inNamespace:
-            # Link may have been dumped into root NS
-            quietRun( 'ip link del ' + self.name )
+        # We used to do this, but it slows us down:
+        # if self.node.inNamespace:
+        # Link may have been dumped into root NS
+        # quietRun( 'ip link del ' + self.name )
 
     def status( self ):
         "Return intf status as a string"
@@ -216,15 +222,19 @@ class TCIntf( Intf ):
        Allows specification of bandwidth limits (various methods)
        as well as delay, loss and max queue length"""
 
+    # The parameters we use seem to work reasonably up to 1 Gb/sec
+    # For higher data rates, we will probably need to change them.
+    bwParamMax = 1000
+
     def bwCmds( self, bw=None, speedup=0, use_hfsc=False, use_tbf=False,
                 latency_ms=None, enable_ecn=False, enable_red=False ):
         "Return tc commands to set bandwidth"
 
         cmds, parent = [], ' root '
 
-        if bw and ( bw < 0 or bw > 1000 ):
-            error( 'Bandwidth', bw, 'is outside range 0..1000 Mbps\n' )
-
+        if bw and ( bw < 0 or bw > self.bwParamMax ):
+            error( 'Bandwidth limit', bw, 'is outside supported range 0..%d'
+                   % self.bwParamMax, '- ignoring\n' )
         elif bw is not None:
             # BL: this seems a bit brittle...
             if ( speedup > 0 and
@@ -367,10 +377,11 @@ class Link( object ):
     """A basic link is just a veth pair.
        Other types of links could be tunnels, link emulators, etc.."""
 
+    # pylint: disable=too-many-branches
     def __init__( self, node1, node2, port1=None, port2=None,
                   intfName1=None, intfName2=None, addr1=None, addr2=None,
                   intf=Intf, cls1=None, cls2=None, params1=None,
-                  params2=None ):
+                  params2=None, fast=True ):
         """Create veth link to another node, making two new interfaces.
            node1: first node
            node2: second node
@@ -405,7 +416,14 @@ class Link( object ):
         if not intfName2:
             intfName2 = self.intfName( node2, params2[ 'port' ] )
 
-        self.makeIntfPair( intfName1, intfName2, addr1, addr2 )
+        self.fast = fast
+        if fast:
+            params1.setdefault( 'moveIntfFn', self._ignore )
+            params2.setdefault( 'moveIntfFn', self._ignore )
+            self.makeIntfPair( intfName1, intfName2, addr1, addr2,
+                               node1, node2, deleteIntfs=False )
+        else:
+            self.makeIntfPair( intfName1, intfName2, addr1, addr2 )
 
         if not cls1:
             cls1 = intf
@@ -419,6 +437,12 @@ class Link( object ):
 
         # All we are is dust in the wind, and our two interfaces
         self.intf1, self.intf2 = intf1, intf2
+    # pylint: enable=too-many-branches
+
+    @staticmethod
+    def _ignore( *args, **kwargs ):
+        "Ignore any arguments"
+        pass
 
     def intfName( self, node, n ):
         "Construct a canonical interface name node-ethN for interface n."
@@ -427,24 +451,33 @@ class Link( object ):
         return node.name + '-eth' + repr( n )
 
     @classmethod
-    def makeIntfPair( cls, intfname1, intfname2, addr1=None, addr2=None ):
+    def makeIntfPair( cls, intfname1, intfname2, addr1=None, addr2=None,
+                      node1=None, node2=None, deleteIntfs=True,
+                      runCmd=quietRun ):
         """Create pair of interfaces
-           intfname1: name of interface 1
-           intfname2: name of interface 2
+           intfname1: name for interface 1
+           intfname2: name for interface 2
+           addr1: MAC address for interface 1 (optional)
+           addr2: MAC address for interface 2 (optional)
+           node1: home node for interface 1 (optional)
+           node2: home node for interface 2 (optional)
            (override this method [and possibly delete()]
            to change link type)"""
         # Leave this as a class method for now
         assert cls
-        return makeIntfPair( intfname1, intfname2, addr1, addr2 )
+        return makeIntfPair( intfname1, intfname2, addr1, addr2, node1, node2,
+                             deleteIntfs=deleteIntfs, runCmd=runCmd )
 
     def delete( self ):
         "Delete this link"
         self.intf1.delete()
-        self.intf2.delete()
+        # We only need to delete one side, though this doesn't seem to
+        # cost us much and might help subclasses.
+        # self.intf2.delete()
 
     def stop( self ):
         "Override to stop and clean up link as needed"
-        pass
+        self.delete()
 
     def status( self ):
         "Return link status as a string"
@@ -453,6 +486,41 @@ class Link( object ):
     def __str__( self ):
         return '%s<->%s' % ( self.intf1, self.intf2 )
 
+
+class OVSIntf( Intf ):
+    "Patch interface on an OVSSwitch"
+
+    def ifconfig( self, *args ):
+        cmd = ' '.join( args )
+        if cmd == 'up':
+            # OVSIntf is always up
+            return
+        else:
+            raise Exception( 'OVSIntf cannot do ifconfig ' + cmd )
+
+
+class OVSLink( Link ):
+    """Link that makes patch links between OVSSwitches
+       Warning: in testing we have found that no more
+       than ~64 OVS patch links should be used in row."""
+
+    def __init__( self, node1, node2, **kwargs ):
+        "See Link.__init__() for options"
+        self.isPatchLink = False
+        if ( isinstance( node1, mininet.node.OVSSwitch ) and
+             isinstance( node2, mininet.node.OVSSwitch ) ):
+            self.isPatchLink = True
+            kwargs.update( cls1=OVSIntf, cls2=OVSIntf )
+        Link.__init__( self, node1, node2, **kwargs )
+
+    def makeIntfPair( self, *args, **kwargs ):
+        "Usually delegated to OVSSwitch"
+        if self.isPatchLink:
+            return None, None
+        else:
+            return Link.makeIntfPair( *args, **kwargs )
+
+
 class TCLink( Link ):
     "Link with symmetric TC interfaces configured via opts"
     def __init__( self, node1, node2, port1=None, port2=None,

mininet/net.py

@@ -102,12 +102,13 @@ from mininet.node import ( Node, Host, OVSKernelSwitch, DefaultController,
                            Controller )
 from mininet.nodelib import NAT
 from mininet.link import Link, Intf
-from mininet.util import quietRun, fixLimits, numCores, ensureRoot
-from mininet.util import macColonHex, ipStr, ipParse, netParse, ipAdd
+from mininet.util import ( quietRun, fixLimits, numCores, ensureRoot,
+                           macColonHex, ipStr, ipParse, netParse, ipAdd,
+                           waitListening )
 from mininet.term import cleanUpScreens, makeTerms
 
 # Mininet version: should be consistent with README and LICENSE
-VERSION = "2.2.0"
+VERSION = "2.2.1"
 
 class Mininet( object ):
     "Network emulation with hosts spawned in network namespaces."
@@ -356,6 +357,8 @@ class Mininet( object ):
             options.setdefault( 'port1', port1 )
         if port2 is not None:
             options.setdefault( 'port2', port2 )
+        if self.intf is not None:
+            options.setdefault( 'intf', self.intf )
         # Set default MAC - this should probably be in Link
         options.setdefault( 'addr1', self.randMac() )
         options.setdefault( 'addr2', self.randMac() )
@@ -401,7 +404,7 @@ class Mininet( object ):
             if not isinstance( classes, list ):
                 classes = [ classes ]
             for i, cls in enumerate( classes ):
-                # Allow Controller objects because nobody understands currying
+                # Allow Controller objects because nobody understands partial()
                 if isinstance( cls, Controller ):
                     self.addController( cls )
                 else:
@@ -414,7 +417,12 @@ class Mininet( object ):
 
         info( '\n*** Adding switches:\n' )
         for switchName in topo.switches():
-            self.addSwitch( switchName, **topo.nodeInfo( switchName) )
+            # A bit ugly: add batch parameter if appropriate
+            params = topo.nodeInfo( switchName)
+            cls = params.get( 'cls', self.switch )
+            if hasattr( cls, 'batchStartup' ):
+                params.setdefault( 'batch', True )
+            self.addSwitch( switchName, **params )
             info( switchName + ' ' )
 
         info( '\n*** Adding links:\n' )
@@ -458,7 +466,7 @@ class Mininet( object ):
     def stopXterms( self ):
         "Kill each xterm."
         for term in self.terms:
-            os.kill( term.pid, signal.SIGKILL )
+            term.send_signal( signal.SIGKILL )
         cleanUpScreens()
 
     def staticArp( self ):
@@ -481,6 +489,13 @@ class Mininet( object ):
         for switch in self.switches:
             info( switch.name + ' ')
             switch.start( self.controllers )
+        started = {}
+        for swclass, switches in groupby(
+                sorted( self.switches, key=type ), type ):
+            switches = tuple( switches )
+            if hasattr( swclass, 'batchStartup' ):
+                success = swclass.batchStartup( switches )
+                started.update( { s: s for s in success } )
         info( '\n' )
         if self.waitConn:
             self.waitConnected()
@@ -495,20 +510,25 @@ class Mininet( object ):
         if self.terms:
             info( '*** Stopping %i terms\n' % len( self.terms ) )
             self.stopXterms()
-        info( '*** Stopping %i switches\n' % len( self.switches ) )
-        for swclass, switches in groupby(
-                sorted( self.switches, key=type ), type ):
-            if hasattr( swclass, 'batchShutdown' ):
-                swclass.batchShutdown( switches )
-        for switch in self.switches:
-            info( switch.name + ' ' )
-            switch.stop()
-            switch.terminate()
-        info( '\n' )
         info( '*** Stopping %i links\n' % len( self.links ) )
         for link in self.links:
+            info( '.' )
             link.stop()
         info( '\n' )
+        info( '*** Stopping %i switches\n' % len( self.switches ) )
+        stopped = {}
+        for swclass, switches in groupby(
+                sorted( self.switches, key=type ), type ):
+            switches = tuple( switches )
+            if hasattr( swclass, 'batchShutdown' ):
+                success = swclass.batchShutdown( switches )
+                stopped.update( { s: s for s in success } )
+        for switch in self.switches:
+            info( switch.name + ' ' )
+            if switch not in stopped:
+                switch.stop()
+            switch.terminate()
+        info( '\n' )
         info( '*** Stopping %i hosts\n' % len( self.hosts ) )
         for host in self.hosts:
             info( host.name + ' ' )
@@ -712,27 +732,25 @@ class Mininet( object ):
     # XXX This should be cleaned up
 
     def iperf( self, hosts=None, l4Type='TCP', udpBw='10M', fmt=None,
-               seconds=5):
+               seconds=5, port=5001):
         """Run iperf between two hosts.
            hosts: list of hosts; if None, uses first and last hosts
            l4Type: string, one of [ TCP, UDP ]
            udpBw: bandwidth target for UDP test
            fmt: iperf format argument if any
            seconds: iperf time to transmit
+           port: iperf port
            returns: two-element array of [ server, client ] speeds
            note: send() is buffered, so client rate can be much higher than
            the actual transmission rate; on an unloaded system, server
            rate should be much closer to the actual receive rate"""
-        if not quietRun( 'which telnet' ):
-            error( 'Cannot find telnet in $PATH - required for iperf test' )
-            return
         hosts = hosts or [ self.hosts[ 0 ], self.hosts[ -1 ] ]
         assert len( hosts ) == 2
         client, server = hosts
-        output( '*** Iperf: testing ' + l4Type + ' bandwidth between ' )
-        output( "%s and %s\n" % ( client.name, server.name ) )
+        output( '*** Iperf: testing', l4Type, 'bandwidth between',
+                client, 'and', server, '\n' )
         server.cmd( 'killall -9 iperf' )
-        iperfArgs = 'iperf '
+        iperfArgs = 'iperf -p %d ' % port
         bwArgs = ''
         if l4Type == 'UDP':
             iperfArgs += '-u '
@@ -741,20 +759,16 @@ class Mininet( object ):
             raise Exception( 'Unexpected l4 type: %s' % l4Type )
         if fmt:
             iperfArgs += '-f %s ' % fmt
-        server.sendCmd( iperfArgs + '-s', printPid=True )
-        servout = ''
-        while server.lastPid is None:
-            servout += server.monitor()
+        server.sendCmd( iperfArgs + '-s' )
         if l4Type == 'TCP':
-            while 'Connected' not in client.cmd(
-                    'sh -c "echo A | telnet -e A %s 5001"' % server.IP()):
-                info( 'Waiting for iperf to start up...' )
-                sleep(.5)
+            if not waitListening( client, server.IP(), port ):
+                raise Exception( 'Could not connect to iperf on port %d'
+                                 % port )
         cliout = client.cmd( iperfArgs + '-t %d -c ' % seconds +
                              server.IP() + ' ' + bwArgs )
         debug( 'Client output: %s\n' % cliout )
         server.sendInt()
-        servout += server.waitOutput()
+        servout = server.waitOutput()
         debug( 'Server output: %s\n' % servout )
         result = [ self._parseIperf( servout ), self._parseIperf( cliout ) ]
         if l4Type == 'UDP':

mininet/node.py

@@ -23,20 +23,26 @@ Switch: superclass for switch nodes.
 UserSwitch: a switch using the user-space switch from the OpenFlow
     reference implementation.
 
-KernelSwitch: a switch using the kernel switch from the OpenFlow reference
-    implementation.
-
-OVSSwitch: a switch using the OpenVSwitch OpenFlow-compatible switch
+OVSSwitch: a switch using the Open vSwitch OpenFlow-compatible switch
     implementation (openvswitch.org).
 
+OVSBridge: an Ethernet bridge implemented using Open vSwitch.
+    Supports STP.
+
+IVSSwitch: OpenFlow switch using the Indigo Virtual Switch.
+
 Controller: superclass for OpenFlow controllers. The default controller
     is controller(8) from the reference implementation.
 
+OVSController: The test controller from Open vSwitch.
+
 NOXController: a controller node using NOX (noxrepo.org).
 
+Ryu: The Ryu controller (https://osrg.github.io/ryu/)
+
 RemoteController: a remote controller node, which may use any
     arbitrary OpenFlow-compatible controller, and which is not
-    created or managed by mininet.
+    created or managed by Mininet.
 
 Future enhancements:
 
@@ -57,8 +63,8 @@ from time import sleep
 from mininet.log import info, error, warn, debug
 from mininet.util import ( quietRun, errRun, errFail, moveIntf, isShellBuiltin,
                            numCores, retry, mountCgroups )
-from mininet.moduledeps import moduleDeps, pathCheck, OVS_KMOD, OF_KMOD, TUN
-from mininet.link import Link, Intf, TCIntf
+from mininet.moduledeps import moduleDeps, pathCheck, TUN
+from mininet.link import Link, Intf, TCIntf, OVSIntf
 from re import findall
 from distutils.version import StrictVersion
 
@@ -68,9 +74,9 @@ class Node( object ):
 
     portBase = 0  # Nodes always start with eth0/port0, even in OF 1.0
 
-    def __init__( self, name, inNamespace=True, **params ):
+    def __init__( self, name, **params ):
         """name: name of node
-           inNamespace: in network namespace?
+           ns: private namespaces to use ['net','mnt']
            privateDirs: list of private directory strings or tuples
            params: Node parameters (see config() for details)"""
 
@@ -79,7 +85,14 @@ class Node( object ):
 
         self.name = params.get( 'name', name )
         self.privateDirs = params.get( 'privateDirs', [] )
-        self.inNamespace = params.get( 'inNamespace', inNamespace )
+        self.overlayDirs = params.get( 'overlayDirs', [] )
+
+        # Support old inNamespace param
+        self.ns = params.get( 'ns', ( 'net', 'mnt' ) )
+        inNamespace = params.get( 'inNamespace', True )
+        if not inNamespace:
+            self.ns = []
+        self.inNamespace = 'net' in self.ns
 
         # Stash configuration parameters for future reference
         self.params = params
@@ -96,8 +109,9 @@ class Node( object ):
         self.waiting = False
         self.readbuf = ''
 
-        # Start command interpreter shell
+        # Start command interpreter shell and mount any local dirs
         self.startShell()
+        self.mountOverlayDirs()
         self.mountPrivateDirs()
 
     # File descriptor to node mapping support
@@ -114,22 +128,26 @@ class Node( object ):
         node = cls.outToNode.get( fd )
         return node or cls.inToNode.get( fd )
 
+    _marker = re.compile( chr( 1 ) + r'(\d+)\r\n' )
+
     # Command support via shell process in namespace
     def startShell( self, mnopts=None ):
         "Start a shell process for running commands"
         if self.shell:
             error( "%s: shell is already running\n" % self.name )
             return
-        # mnexec: (c)lose descriptors, (d)etach from tty,
-        # (p)rint pid, and run in (n)amespace
-        opts = '-cd' if mnopts is None else mnopts
-        if self.inNamespace:
-            opts += 'n'
-        # bash -m: enable job control, i: force interactive
+        # mnexec: (c)lose descriptors
+        # (p)rint pid, and run in (n)etwork and (m)ount namespace
+        opts = '-cdp' if mnopts is None else mnopts
+        # Handle additional namespaces if specified
+        nsmap = { 'pid': 'P', 'mnt': 'm', 'net': 'n', 'uts': 'u' }
+        chars = [ nsmap.get( ns, '' ) for ns in self.ns ]
+        opts += ''.join( chars )
+        # bash -i: force interactive
         # -s: pass $* to shell, and make process easy to find in ps
         # prompt is set to sentinel chr( 127 )
         cmd = [ 'mnexec', opts, 'env', 'PS1=' + chr( 127 ),
-                'bash', '--norc', '-mis', 'mininet:' + self.name ]
+                'bash', '--norc', '-is', 'mininet:' + self.name ]
         # Spawn a shell subprocess in a pseudo-tty, to disable buffering
         # in the subprocess and insulate it from signals (e.g. SIGINT)
         # received by the parent
@@ -151,17 +169,18 @@ class Node( object ):
         self.lastPid = None
         self.readbuf = ''
         # Wait for prompt
-        while True:
-            data = self.read( 1024 )
-            if data[ -1 ] == chr( 127 ):
-                break
-            self.pollOut.poll()
-        self.waiting = False
-        self.cmd( 'stty -echo' )
-        self.cmd( 'set +m' )
+        self.waiting = True
+        self.waitOutput()
+        if 'P' in opts:
+            assert self.lastPid is not None
+            self.pid = self.lastPid
+        # +m: disable job control notification
+        self.cmd( 'unset HISTFILE; stty -echo; set +m' )
 
     def mountPrivateDirs( self ):
-        "mount private directories"
+        "Mount private directories"
+        # Avoid expanding a string into a list of chars
+        assert not isinstance( self.privateDirs, basestring )
         for directory in self.privateDirs:
             if isinstance( directory, tuple ):
                 # mount given private directory
@@ -177,13 +196,61 @@ class Node( object ):
                 self.cmd( 'mount -n -t tmpfs tmpfs %s' % directory )
 
     def unmountPrivateDirs( self ):
-        "mount private directories"
+        "Unmount private and overlay directories"
         for directory in self.privateDirs:
             if isinstance( directory, tuple ):
                 self.cmd( 'umount ', directory[ 0 ] )
             else:
                 self.cmd( 'umount ', directory )
 
+    # XXX We should make overlayDirs as consistent as possible
+    # with privateDirs.
+
+    def _overlayFrom( self, entry ):
+        "Helper function: return mountpaint, overlay, tmpfs from entry"
+        if type( entry ) is str:
+            # '/mountpoint'
+            mountpoint, overlay = entry, None
+        elif len( entry ) is 1:
+            # [ '/mountpoint' ]
+            mountpoint, overlay = entry[ 0 ], None
+        else:
+            # [ '/mountpoint', '/overlay' ]
+            mountpoint, overlay = entry
+        tmpfs = None if overlay else '/tmp/%s/%s' % ( self, mountpoint )
+        return mountpoint, overlay, tmpfs
+
+    def mountOverlayDirs( self ):
+        """Mount overlay directories. Overlay directories are similar
+            to private directories except they are copy-on-write copies
+            of directories in the host file system.
+            overlayDirs is of the form ((mountpoint,overlaydir), ...)
+            much like privateDirs. If overlaydir doesn't exist, we
+            mount a tmpfs at the specified mount point."""
+        # Avoid expanding a string into a list of chars
+        assert not isinstance( self.overlayDirs, basestring )
+        for entry in self.overlayDirs:
+            mountpoint, overlay, tmpfs  = self._overlayFrom( entry )
+            # Create tmpfs if overlay dir is not specified
+            if not overlay:
+                overlay = tmpfs
+                self.cmd( 'mkdir -p', overlay )
+                self.cmd( 'mount -t tmpfs tmpfs', overlay )
+            # Mount overlay dir at mount point
+            self.cmd( ( 'mount -t overlayfs -o upperdir=%s,lowerdir=%s'
+                       ' overlayfs %s' ) % ( overlay, mountpoint, mountpoint ) )
+
+    def unmountOverlayDirs( self ):
+        "Unmount overlay directories"
+        for entry in self.overlayDirs:
+            mountpoint, overlay, tmpfs = self._overlayFrom( entry )
+            # Unfortunately these umounts can fail if the mount point
+            # is in use, possibly leaving tmpfs garbage in the root
+            # mount namespace / file system
+            self.cmd( 'umount', mountpoint )
+            if not overlay:
+                self.cmd( 'umount', tmpfs )
+
     def _popen( self, cmd, **params ):
         """Internal method: spawn and return a process
             cmd: command to run (list)
@@ -194,10 +261,11 @@ class Node( object ):
 
     def cleanup( self ):
         "Help python collect its garbage."
+        # We used to do this, but it slows us down:
         # Intfs may end up in root NS
-        for intfName in self.intfNames():
-            if self.name in intfName:
-                quietRun( 'ip link del ' + intfName )
+        # for intfName in self.intfNames():
+        # if self.name in intfName:
+        # quietRun( 'ip link del ' + intfName )
         self.shell = None
 
     # Subshell I/O, commands and control
@@ -236,6 +304,7 @@ class Node( object ):
     def terminate( self ):
         "Send kill signal to Node and clean up after it."
         self.unmountPrivateDirs()
+        self.unmountOverlayDirs()
         if self.shell:
             if self.shell.poll() is None:
                 os.killpg( self.shell.pid, signal.SIGHUP )
@@ -258,9 +327,9 @@ class Node( object ):
         """Send a command, followed by a command to echo a sentinel,
            and return without waiting for the command to complete.
            args: command and arguments, or string
-           printPid: print command's PID?"""
-        assert not self.waiting
-        printPid = kwargs.get( 'printPid', True )
+           printPid: print command's PID? (False)"""
+        assert self.shell and not self.waiting
+        printPid = kwargs.get( 'printPid', False )
         # Allow sendCmd( [ list ] )
         if len( args ) == 1 and isinstance( args[ 0 ], list ):
             cmd = args[ 0 ]
@@ -284,10 +353,10 @@ class Node( object ):
         self.lastPid = None
         self.waiting = True
 
-    def sendInt( self, intr=chr( 3 ) ):
+    def sendInt( self, signal=signal.SIGINT ):
         "Interrupt running command."
-        debug( 'sendInt: writing chr(%d)\n' % ord( intr ) )
-        self.write( intr )
+        debug( "sending signal %d to pgrp %d" % ( signal, self.pid ) )
+        os.killpg( self.pid, signal )
 
     def monitor( self, timeoutms=None, findPid=True ):
         """Monitor and return the output of a command.
@@ -298,18 +367,18 @@ class Node( object ):
         data = self.read( 1024 )
         pidre = r'\[\d+\] \d+\r\n'
         # Look for PID
-        marker = chr( 1 ) + r'\d+\r\n'
         if findPid and chr( 1 ) in data:
             # suppress the job and PID of a backgrounded command
             if re.findall( pidre, data ):
                 data = re.sub( pidre, '', data )
             # Marker can be read in chunks; continue until all of it is read
-            while not re.findall( marker, data ):
+            while True:
+                markers = self._marker.findall( data )
+                if markers:
+                    self.lastPid = int( markers[ -1 ] )
+                    data = self._marker.sub( '', data )
+                    break
                 data += self.read( 1024 )
-            markers = re.findall( marker, data )
-            if markers:
-                self.lastPid = int( markers[ 0 ][ 1: ] )
-                data = re.sub( marker, '', data )
         # Look for sentinel/EOF
         if len( data ) > 0 and data[ -1 ] == chr( 127 ):
             self.waiting = False
@@ -339,8 +408,11 @@ class Node( object ):
         verbose = kwargs.get( 'verbose', False )
         log = info if verbose else debug
         log( '*** %s : %s\n' % ( self.name, args ) )
-        self.sendCmd( *args, **kwargs )
-        return self.waitOutput( verbose )
+        if self.shell:
+            self.sendCmd( *args, **kwargs )
+            return self.waitOutput( verbose )
+        else:
+            warn( '(%s exited - ignoring cmd%s)\n' % ( self, args ) )
 
     def cmdPrint( self, *args):
         """Call cmd and printing its output
@@ -372,6 +444,7 @@ class Node( object ):
         # Shell requires a string, not a list!
         if defaults.get( 'shell', False ):
             cmd = ' '.join( cmd )
+        debug( cmd, defaults )
         popen = self._popen( cmd, **defaults )
         return popen
 
@@ -503,15 +576,13 @@ class Node( object ):
            mac: MAC address as string"""
         return self.intf( intf ).setMAC( mac )
 
-    def setIP( self, ip, prefixLen=8, intf=None ):
+    def setIP( self, ip, prefixLen=8, intf=None, **kwargs ):
         """Set the IP address for an interface.
            intf: intf or intf name
            ip: IP address as a string
-           prefixLen: prefix length, e.g. 8 for /8 or 16M addrs"""
-        # This should probably be rethought
-        if '/' not in ip:
-            ip = '%s/%s' % ( ip, prefixLen )
-        return self.intf( intf ).setIP( ip )
+           prefixLen: prefix length, e.g. 8 for /8 or 16M addrs
+           kwargs: any additional arguments for intf.setIP"""
+        return self.intf( intf ).setIP( ip, prefixLen, **kwargs )
 
     def IP( self, intf=None ):
         "Return IP address of a node or specific interface."
@@ -671,7 +742,7 @@ class CPULimitedHost( Host ):
         "Clean up our cgroup"
         # info( '*** deleting cgroup', self.cgroup, '\n' )
         _out, _err, exitcode = errRun( 'cgdelete -r ' + self.cgroup )
-        return exitcode != 0
+        return exitcode == 0  # success condition
 
     def popen( self, *args, **kwargs ):
         """Return a Popen() object in node's namespace
@@ -848,7 +919,7 @@ class Switch( Node ):
         self.dpid = self.defaultDpid( dpid )
         self.opts = opts
         self.listenPort = listenPort
-        if not self.inNamespace:
+        if 'net' not in self.ns:
             self.controlIntf = Intf( 'lo', self, port=0 )
 
     def defaultDpid( self, dpid=None ):
@@ -893,6 +964,12 @@ class Switch( Node ):
         debug( 'Assuming', repr( self ), 'is connected to a controller\n' )
         return True
 
+    def stop( self, deleteIntfs=True ):
+        """Stop switch
+           deleteIntfs: delete interfaces? (True)"""
+        if deleteIntfs:
+            self.deleteIntfs()
+
     def __repr__( self ):
         "More informative string representation"
         intfs = ( ','.join( [ '%s:%s' % ( i.name, i.IP() )
@@ -999,73 +1076,32 @@ class UserSwitch( Switch ):
         self.cmd( 'kill %ofprotocol' )
         super( UserSwitch, self ).stop( deleteIntfs )
 
-class OVSLegacyKernelSwitch( Switch ):
-    """Open VSwitch legacy kernel-space switch using ovs-openflowd.
-       Currently only works in the root namespace."""
-
-    def __init__( self, name, dp=None, **kwargs ):
-        """Init.
-           name: name for switch
-           dp: netlink id (0, 1, 2, ...)
-           defaultMAC: default MAC as unsigned int; random value if None"""
-        Switch.__init__( self, name, **kwargs )
-        self.dp = dp if dp else self.name
-        self.intf = self.dp
-        if self.inNamespace:
-            error( "OVSKernelSwitch currently only works"
-                   " in the root namespace.\n" )
-            exit( 1 )
-
-    @classmethod
-    def setup( cls ):
-        "Ensure any dependencies are loaded; if not, try to load them."
-        pathCheck( 'ovs-dpctl', 'ovs-openflowd',
-                   moduleName='Open vSwitch (openvswitch.org)')
-        moduleDeps( subtract=OF_KMOD, add=OVS_KMOD )
-
-    def start( self, controllers ):
-        "Start up kernel datapath."
-        ofplog = '/tmp/' + self.name + '-ofp.log'
-        # Delete local datapath if it exists;
-        # then create a new one monitoring the given interfaces
-        self.cmd( 'ovs-dpctl del-dp ' + self.dp )
-        self.cmd( 'ovs-dpctl add-dp ' + self.dp )
-        intfs = [ str( i ) for i in self.intfList() if not i.IP() ]
-        self.cmd( 'ovs-dpctl', 'add-if', self.dp, ' '.join( intfs ) )
-        # Run protocol daemon
-        clist = ','.join( [ 'tcp:%s:%d' % ( c.IP(), c.port )
-                            for c in controllers ] )
-        self.cmd( 'ovs-openflowd ' + self.dp +
-                  ' ' + clist +
-                  ' --fail=secure ' + self.opts +
-                  ' --datapath-id=' + self.dpid +
-                  ' 1>' + ofplog + ' 2>' + ofplog + '&' )
-        self.execed = False
-
-    def stop( self, deleteIntfs=True ):
-        """Terminate kernel datapath."
-           deleteIntfs: delete interfaces? (True)"""
-        quietRun( 'ovs-dpctl del-dp ' + self.dp )
-        self.cmd( 'kill %ovs-openflowd' )
-        super( OVSLegacyKernelSwitch, self ).stop( deleteIntfs )
-
 
 class OVSSwitch( Switch ):
     "Open vSwitch switch. Depends on ovs-vsctl."
 
     def __init__( self, name, failMode='secure', datapath='kernel',
-                  inband=False, protocols=None, **params ):
-        """Init.
-           name: name for switch
+                  inband=False, protocols=None,
+                  reconnectms=1000, stp=False, batch=False, **params ):
+        """name: name for switch
            failMode: controller loss behavior (secure|open)
            datapath: userspace or kernel mode (kernel|user)
-           inband: use in-band control (False)"""
+           inband: use in-band control (False)
+           protocols: use specific OpenFlow version(s) (e.g. OpenFlow13)
+                      Unspecified (or old OVS version) uses OVS default
+           reconnectms: max reconnect timeout in ms (0/None for default)
+           stp: enable STP (False, requires failMode=standalone)
+           batch: enable batch startup (False)"""
         Switch.__init__( self, name, **params )
         self.failMode = failMode
         self.datapath = datapath
         self.inband = inband
         self.protocols = protocols
+        self.reconnectms = reconnectms
+        self.stp = stp
         self._uuids = []  # controller UUIDs
+        self.batch = batch
+        self.commands = []  # saved commands for batch startup
 
     @classmethod
     def setup( cls ):
@@ -1089,23 +1125,23 @@ class OVSSwitch( Switch ):
         version = quietRun( 'ovs-vsctl --version' )
         cls.OVSVersion = findall( r'\d+\.\d+', version )[ 0 ]
 
-    @classmethod
-    def isOldOVS( cls ):
+    def isOldOVS( self ):
         "Is OVS ersion < 1.10?"
-        return ( StrictVersion( cls.OVSVersion ) <
+        return ( StrictVersion( self.OVSVersion ) <
                  StrictVersion( '1.10' ) )
 
-    @classmethod
-    def batchShutdown( cls, switches ):
-        "Call ovs-vsctl del-br on all OVSSwitches in a list"
-        quietRun( 'ovs-vsctl ' +
-                  ' -- '.join( '--if-exists del-br %s' % s
-                               for s in switches ) )
-
     def dpctl( self, *args ):
         "Run ovs-ofctl command"
         return self.cmd( 'ovs-ofctl', args[ 0 ], self, *args[ 1: ] )
 
+    def vsctl( self, *args, **kwargs ):
+        "Run ovs-vsctl command (or queue for later execution)"
+        if self.batch:
+            cmd = ' '.join( str( arg ).strip() for arg in args )
+            self.commands.append( cmd )
+        else:
+            return self.cmd( 'ovs-vsctl', *args, **kwargs )
+
     @staticmethod
     def TCReapply( intf ):
         """Unfortunately OVS and Mininet are fighting
@@ -1116,13 +1152,13 @@ class OVSSwitch( Switch ):
 
     def attach( self, intf ):
         "Connect a data port"
-        self.cmd( 'ovs-vsctl add-port', self, intf )
+        self.vsctl( 'add-port', self, intf )
         self.cmd( 'ifconfig', intf, 'up' )
         self.TCReapply( intf )
 
     def detach( self, intf ):
         "Disconnect a data port"
-        self.cmd( 'ovs-vsctl del-port', self, intf )
+        self.vsctl( 'del-port', self, intf )
 
     def controllerUUIDs( self, update=False ):
         """Return ovsdb UUIDs for our controllers
@@ -1140,64 +1176,109 @@ class OVSSwitch( Switch ):
     def connected( self ):
         "Are we connected to at least one of our controllers?"
         for uuid in self.controllerUUIDs():
-            if 'true' in self.cmd( 'ovs-vsctl -- get Controller',
-                                            uuid, 'is_connected' ):
+            if 'true' in self.vsctl( '-- get Controller',
+                                     uuid, 'is_connected' ):
                 return True
         return self.failMode == 'standalone'
 
+    def intfOpts( self, intf ):
+        "Return OVS interface options for intf"
+        opts = ''
+        if not self.isOldOVS():
+            # ofport_request is not supported on old OVS
+            opts += ' ofport_request=%s' % self.ports[ intf ]
+            # Patch ports don't work well with old OVS
+            if isinstance( intf, OVSIntf ):
+                intf1, intf2 = intf.link.intf1, intf.link.intf2
+                peer = intf1 if intf1 != intf else intf2
+                opts += ' type=patch options:peer=%s' % peer
+        return '' if not opts else ' -- set Interface %s' % intf + opts
+
+    def bridgeOpts( self ):
+        "Return OVS bridge options"
+        opts = ( ' other_config:datapath-id=%s' % self.dpid +
+                 ' fail_mode=%s' % self.failMode )
+        if not self.inband:
+            opts += ' other-config:disable-in-band=true'
+        if self.datapath == 'user':
+            opts += ' datapath_type=netdev'
+        if self.protocols and not self.isOldOVS():
+            opts += ' protocols=%s' % self.protocols
+        if self.stp and self.failMode == 'standalone':
+            opts += ' stp_enable=true'
+        return opts
+
     def start( self, controllers ):
         "Start up a new OVS OpenFlow switch using ovs-vsctl"
         if self.inNamespace:
             raise Exception(
-                'OVS kernel switch does not work in a namespace' )
-        # Annoyingly, --if-exists option seems not to work
-        self.cmd( 'ovs-vsctl del-br', self )
+                'OVS kernel switch does not work in a network namespace' )
         int( self.dpid, 16 )  # DPID must be a hex string
-        # Interfaces and controllers
-        intfs = ' '.join( '-- add-port %s %s ' % ( self, intf ) +
-                          '-- set Interface %s ' % intf +
-                          'ofport_request=%s ' % self.ports[ intf ]
-                          for intf in self.intfList()
-                          if self.ports[ intf ] and not intf.IP() )
-        clist = ' '.join( '%s:%s:%d' % ( c.protocol, c.IP(), c.port )
-                          for c in controllers )
+        # Command to add interfaces
+        intfs = ''.join( ' -- add-port %s %s' % ( self, intf ) +
+                         self.intfOpts( intf )
+                         for intf in self.intfList()
+                         if self.ports[ intf ] and not intf.IP() )
+        # Command to create controller entries
+        clist = [ ( self.name + c.name, '%s:%s:%d' %
+                  ( c.protocol, c.IP(), c.port ) )
+                  for c in controllers ]
         if self.listenPort:
-            clist += ' ptcp:%s' % self.listenPort
-        # Construct big ovs-vsctl command for new versions of OVS
+            clist.append( ( self.name + '-listen',
+                            'ptcp:%s' % self.listenPort ) )
+        ccmd = '-- --id=@%s create Controller target=\\"%s\\"'
+        if self.reconnectms:
+            ccmd += ' max_backoff=%d' % self.reconnectms
+        cargs = ' '.join( ccmd % ( name, target )
+                          for name, target in clist )
+        # Controller ID list
+        cids = ','.join( '@%s' % name for name, _target in clist )
+        # Try to delete any existing bridges with the same name
         if not self.isOldOVS():
-            cmd = ( 'ovs-vsctl add-br %s ' % self +
-                    '-- set Bridge %s ' % self +
-                    'other_config:datapath-id=%s ' % self.dpid +
-                    '-- set-fail-mode %s %s ' % ( self, self.failMode ) +
-                    intfs +
-                    '-- set-controller %s %s ' % ( self, clist ) )
-        # Construct ovs-vsctl commands for old versions of OVS
-        else:
-            self.cmd( 'ovs-vsctl add-br', self )
-            for intf in self.intfList():
-                if not intf.IP():
-                    self.cmd( 'ovs-vsctl add-port', self, intf )
-            cmd = ( 'ovs-vsctl set Bridge %s ' % self +
-                    'other_config:datapath-id=%s ' % self.dpid +
-                    '-- set-fail-mode %s %s ' % ( self, self.failMode ) +
-                    '-- set-controller %s %s ' % ( self, clist ) )
-        if not self.inband:
-            cmd += ( '-- set bridge %s '
-                     'other-config:disable-in-band=true ' % self )
-        if self.datapath == 'user':
-            cmd += '-- set bridge %s datapath_type=netdev ' % self
-        if self.protocols:
-            cmd += '-- set bridge %s protocols=%s' % ( self, self.protocols )
-        # Do it!!
-        self.cmd( cmd )
-        # Reconnect quickly to controllers (1s vs. 15s max_backoff)
-        uuids = [ '-- set Controller %s max_backoff=1000' % uuid
-                  for uuid in self.controllerUUIDs() ]
-        if uuids:
-            self.cmd( 'ovs-vsctl', *uuids )
+            cargs += ' -- --if-exists del-br %s' % self
+        # One ovs-vsctl command to rule them all!
+        self.vsctl( cargs +
+                    ' -- add-br %s' % self +
+                    ' -- set bridge %s controller=[%s]' % ( self, cids  ) +
+                    self.bridgeOpts() +
+                    intfs )
         # If necessary, restore TC config overwritten by OVS
-        for intf in self.intfList():
-            self.TCReapply( intf )
+        if not self.batch:
+            for intf in self.intfList():
+                self.TCReapply( intf )
+
+    # This should be ~ int( quietRun( 'getconf ARG_MAX' ) ),
+    # but the real limit seems to be much lower
+    argmax = 128000
+
+    @classmethod
+    def batchStartup( cls, switches, run=errRun ):
+        """Batch startup for OVS
+           switches: switches to start up
+           run: function to run commands (errRun)"""
+        info( '...' )
+        cmds = 'ovs-vsctl'
+        for switch in switches:
+            if switch.isOldOVS():
+                # Ideally we'd optimize this also
+                run( 'ovs-vsctl del-br %s' % switch )
+            for cmd in switch.commands:
+                cmd = cmd.strip()
+                # Don't exceed ARG_MAX
+                if len( cmds ) + len( cmd ) >= cls.argmax:
+                    run( cmds, shell=True )
+                    cmds = 'ovs-vsctl'
+                cmds += ' ' + cmd
+                switch.cmds = []
+                switch.batch = False
+        if cmds:
+            run( cmds, shell=True )
+        # Reapply link config if necessary...
+        for switch in switches:
+            for intf in switch.intfs.itervalues():
+                if isinstance( intf, TCIntf ):
+                    intf.config( **intf.params )
+        return switches
 
     def stop( self, deleteIntfs=True ):
         """Terminate OVS switch.
@@ -1207,6 +1288,22 @@ class OVSSwitch( Switch ):
             self.cmd( 'ip link del', self )
         super( OVSSwitch, self ).stop( deleteIntfs )
 
+    @classmethod
+    def batchShutdown( cls, switches, run=errRun ):
+        "Shut down a list of OVS switches"
+        delcmd = 'del-br %s'
+        if switches and not switches[ 0 ].isOldOVS():
+            delcmd = '--if-exists ' + delcmd
+        # First, delete them all from ovsdb
+        run( 'ovs-vsctl ' +
+             ' -- '.join( delcmd % s for s in switches ) )
+        # Next, shut down all of the processes
+        pids = ' '.join( str( switch.pid ) for switch in switches )
+        run( 'kill -HUP ' + pids )
+        for switch in switches:
+            switch.shell = None
+        return switches
+
 
 OVSKernelSwitch = OVSSwitch
 
@@ -1214,13 +1311,24 @@ OVSKernelSwitch = OVSSwitch
 class OVSBridge( OVSSwitch ):
     "OVSBridge is an OVSSwitch in standalone/bridge mode"
 
-    def __init__( self, args, **kwargs ):
+    def __init__( self, *args, **kwargs ):
+        """stp: enable Spanning Tree Protocol (False)
+           see OVSSwitch for other options"""
         kwargs.update( failMode='standalone' )
-        OVSSwitch.__init__( self, args, **kwargs )
+        OVSSwitch.__init__( self, *args, **kwargs )
 
     def start( self, controllers ):
+        "Start bridge, ignoring controllers argument"
         OVSSwitch.start( self, controllers=[] )
 
+    def connected( self ):
+        "Are we forwarding yet?"
+        if self.stp:
+            status = self.dpctl( 'show' )
+            return 'STP_FORWARD' in status and not 'STP_LEARN' in status
+        else:
+            return True
+
 
 class IVSSwitch( Switch ):
     "Indigo Virtual Switch"
@@ -1247,6 +1355,7 @@ class IVSSwitch( Switch ):
         "Kill each IVS switch, to be waited on later in stop()"
         for switch in switches:
             switch.cmd( 'kill %ivs' )
+        return switches
 
     def start( self, controllers ):
         "Start up a new IVS switch"
@@ -1296,11 +1405,15 @@ class Controller( Node ):
        OpenFlow controller."""
 
     def __init__( self, name, inNamespace=False, command='controller',
-                  cargs='-v ptcp:%d', cdir=None, ip="127.0.0.1",
+                  cargs='ptcp:%d', cdir=None, ip="127.0.0.1",
                   port=6633, protocol='tcp', **params ):
         self.command = command
         self.cargs = cargs
         self.cdir = cdir
+        # Accept 'ip:port' syntax as shorthand
+        if ':' in ip:
+            ip, port = ip.split( ':' )
+            port = int( port )
         self.ip = ip
         self.port = port
         self.protocol = protocol
@@ -1400,7 +1513,7 @@ class NOX( Controller ):
                              cdir=noxCoreDir,
                              **kwargs )
 
-class RYU( Controller ):
+class Ryu( Controller ):
     "Controller to run Ryu application"
     def __init__( self, name, *ryuArgs, **kwargs ):
         """Init.
@@ -1422,6 +1535,7 @@ class RYU( Controller ):
                              cdir=ryuCoreDir,
                              **kwargs )
 
+
 class RemoteController( Controller ):
     "Controller running outside of Mininet's control."
 
@@ -1465,3 +1579,7 @@ def DefaultController( name, controllers=DefaultControllers, **kwargs ):
     if not controller:
         raise Exception( 'Could not find a default OpenFlow controller' )
     return controller( name, **kwargs )
+
+def NullController( *_args, **_kwargs ):
+    "Nonexistent controller - simply returns None"
+    return None

mininet/nodelib.py

@@ -4,11 +4,13 @@ Node Library for Mininet
 This contains additional Node types which you may find to be useful.
 """
 
-from mininet.node import Node, Switch
+from mininet.node import Node, Host, Switch
 from mininet.log import info, warn
 from mininet.moduledeps import pathCheck
+from mininet.util import quietRun
 
 import re
+from tempfile import NamedTemporaryFile
 
 class LinuxBridge( Switch ):
     "Linux Bridge (with optional spanning tree)"
@@ -59,23 +61,30 @@ class LinuxBridge( Switch ):
 
     @classmethod
     def setup( cls ):
-        "Make sure our class dependencies are available"
+        "Check dependencies and warn about firewalling"
         pathCheck( 'brctl', moduleName='bridge-utils' )
+        # Disable Linux bridge firewalling so that traffic can flow!
+        for table in 'arp', 'ip', 'ip6':
+            cmd = 'sysctl net.bridge.bridge-nf-call-%stables' % table
+            out = quietRun( cmd ).strip()
+            if out.endswith( '1' ):
+                warn( 'Warning: Linux bridge may not work with', out, '\n' )
 
 
 class NAT( Node ):
     "NAT: Provides connectivity to external network"
 
-    def __init__( self, name, inetIntf=None, subnet='10.0/8',
-                  localIntf=None, **params):
+    def __init__( self, name, subnet='10.0/8',
+                  localIntf=None, flush=False, **params):
         """Start NAT/forwarding between Mininet and external network
-           inetIntf: interface for internet access
-           subnet: Mininet subnet (default 10.0/8)="""
+           subnet: Mininet subnet (default 10.0/8)
+           flush: flush iptables before installing NAT rules"""
         super( NAT, self ).__init__( name, **params )
 
-        self.inetIntf = inetIntf if inetIntf else self.getGatewayIntf()
         self.subnet = subnet
         self.localIntf = localIntf
+        self.flush = flush
+        self.forwardState = self.cmd( 'sysctl -n net.ipv4.ip_forward' ).strip()
 
     def config( self, **params ):
         """Configure the NAT and iptables"""
@@ -84,27 +93,24 @@ class NAT( Node ):
         if not self.localIntf:
             self.localIntf = self.defaultIntf()
 
-        self.cmd( 'sysctl net.ipv4.ip_forward=0' )
+        if self.flush:
+            self.cmd( 'sysctl net.ipv4.ip_forward=0' )
+            self.cmd( 'iptables -F' )
+            self.cmd( 'iptables -t nat -F' )
+            # Create default entries for unmatched traffic
+            self.cmd( 'iptables -P INPUT ACCEPT' )
+            self.cmd( 'iptables -P OUTPUT ACCEPT' )
+            self.cmd( 'iptables -P FORWARD DROP' )
 
-        # Flush any currently active rules
-        # TODO: is this safe?
-        self.cmd( 'iptables -F' )
-        self.cmd( 'iptables -t nat -F' )
-
-        # Create default entries for unmatched traffic
-        self.cmd( 'iptables -P INPUT ACCEPT' )
-        self.cmd( 'iptables -P OUTPUT ACCEPT' )
-        self.cmd( 'iptables -P FORWARD DROP' )
-
-        # Configure NAT
+        # Install NAT rules
         self.cmd( 'iptables -I FORWARD',
                   '-i', self.localIntf, '-d', self.subnet, '-j DROP' )
         self.cmd( 'iptables -A FORWARD',
                   '-i', self.localIntf, '-s', self.subnet, '-j ACCEPT' )
         self.cmd( 'iptables -A FORWARD',
-                  '-i', self.inetIntf, '-d', self.subnet, '-j ACCEPT' )
+                  '-o', self.localIntf, '-d', self.subnet,'-j ACCEPT' )
         self.cmd( 'iptables -t nat -A POSTROUTING',
-                  '-o', self.inetIntf, '-s', self.subnet, '-j MASQUERADE' )
+                  '-s', self.subnet, "'!'", '-d', self.subnet, '-j MASQUERADE' )
 
         # Instruct the kernel to perform forwarding
         self.cmd( 'sysctl net.ipv4.ip_forward=1' )
@@ -123,26 +129,100 @@ class NAT( Node ):
         # hopefully this won't disconnect you
         self.cmd( 'service network-manager restart' )
 
-    def getGatewayIntf( self, fallback='eth0' ):
-        """Return gateway interface name
-           fallback: default device to fall back to"""
-        routes = self.cmd( 'ip route show' )
-        match = re.search( r'default via \S+ dev (\S+)', routes )
-        if match:
-            return match.group( 1 )
-        else:
-            warn( 'There is no default route set.',
-                  'Using', fallback, 'as gateway interface...\n' )
-            return fallback
-
     def terminate( self ):
-        """Stop NAT/forwarding between Mininet and external network"""
-        # Flush any currently active rules
-        # TODO: is this safe?
-        self.cmd( 'iptables -F' )
-        self.cmd( 'iptables -t nat -F' )
-
-        # Instruct the kernel to stop forwarding
-        self.cmd( 'sysctl net.ipv4.ip_forward=0' )
-
+        "Stop NAT/forwarding between Mininet and external network"
+        # Remote NAT rules
+        self.cmd( 'iptables -D FORWARD',
+                   '-i', self.localIntf, '-d', self.subnet, '-j DROP' )
+        self.cmd( 'iptables -D FORWARD',
+                  '-i', self.localIntf, '-s', self.subnet, '-j ACCEPT' )
+        self.cmd( 'iptables -D FORWARD',
+                  '-o', self.localIntf, '-d', self.subnet,'-j ACCEPT' )
+        self.cmd( 'iptables -t nat -D POSTROUTING',
+                  '-s', self.subnet, '\'!\'', '-d', self.subnet, '-j MASQUERADE' )
+        # Put the forwarding state back to what it was
+        self.cmd( 'sysctl net.ipv4.ip_forward=%s' % self.forwardState )
         super( NAT, self ).terminate()
+
+
+class Server( Host ):
+    """Run sshd in a net/mnt/pid/uts namespace, with private /etc/hosts
+       WARNING!!! KNOWN ISSUES:
+       - control-c does not work in Mininet CLI with pid namespace
+       - xterm does not work from Mininet CLI
+       We may be able to address these issues in the future."""
+
+    ns = [ 'net', 'mnt', 'pid', 'uts' ]
+    overlayDirs = [ '/etc', '/var/run', '/var/log' ]
+    privateDirs = [ '/var/run/sshd', ]
+
+    def __init__( self, *args, **kwargs ):
+        """Add overlay dirs and private dirs, and change permissions
+           ssh: run sshd? (True)"""
+        kwargs.setdefault( 'inNamespace', True )
+        kwargs.setdefault( 'ns', self.ns )
+        kwargs.setdefault( 'privateDirs', self.privateDirs )
+        kwargs.setdefault( 'overlayDirs', self.overlayDirs )
+        kwargs.setdefault( 'ssh', True )
+        super( Server, self ).__init__( *args, **kwargs )
+        # Change permissions, mainly for ssh
+        for pdir in self.privateDirs:
+            self.cmd( 'chown root:root', pdir )
+            self.cmd( 'chmod 755', pdir )
+
+    @staticmethod
+    def updateHostsFiles( servers, tmpdir='/tmp' ):
+        """Update local hosts files on a list of servers
+            servers: list of servers
+            tmpdir: tmp dir shared between mn and servers"""
+        # This scales as n^2, so for a large configuration it's
+        # more efficient to use a DNS server
+        for s in servers:
+            dirs = ( getattr( s, 'overlayDirs', [] ) +
+            getattr( s, 'privateDirs', [] ))
+            if '/etc' in dirs:
+                with NamedTemporaryFile( dir=tmpdir ) as tmpfile:
+                      tmpfile.write( '# Mininet hosts file\n' )
+                      tmpfile.write( '127.0.0.1 localhost %s\n' % s )
+                      for t  in servers:
+                        tmpfile.write( '%s %s\n' % ( t.IP(), t ) )
+                      tmpfile.flush()
+                      s.cmd( 'cp', tmpfile.name, '/etc/hosts' )
+            else:
+                warn( 'not updating hosts file on %s\n' % s )
+
+    def service( self, cmd ):
+        """Start or stop a service
+           usage: service( 'ssh stop' )"""
+        self.cmd( '/etc/init.d/%s' % cmd )
+
+    def motd( self ):
+        "Return login message as a string"
+        return 'Welcome to Mininet host %s at %s' % ( self, self.IP() )
+
+    def startSSH( self, motdPath='/var/run/motd.dynamic' ):
+        "Update motd, clear out utmp/wtmp/btmp, and start sshd"
+        # Note: /var/run and /var/log must be overlays!
+        assert ( '/var/run' in ( self.overlayDirs + self.privateDirs ) and
+                 '/var/log' in ( self.overlayDirs + self.privateDirs ) )
+        self.cmd( "echo  '%s' > %s" % ( self.motd(), motdPath ) )
+        self.cmd( 'truncate -s0 /var/run/utmp /var/log/wtmp* /var/log/btmp*' )
+        # sshd.pid should really be in /var/run/sshd instead of /var/run
+        self.cmd( 'rm /var/run/sshd.pid' )
+        self.cmd( '/etc/init.d/ssh start' )
+
+    def config( self, **kwargs ):
+        """Configure/start sshd and other stuff
+            ssh: start sshd? (True )"""
+        super( Server, self ).config( **kwargs )
+        self.ssh = kwargs.get( 'ssh' )
+        if self.ssh:
+            self.startSSH()
+        if 'uts' in self.ns:
+            self.cmd( 'hostname', self )
+
+    def terminate( self, *args, **kwargs ):
+        "Shut down services and terminate server"
+        if self.ssh:
+            self.service( 'ssh stop' )
+        super( Server, self ).terminate( *args, **kwargs )

mininet/term.py

@@ -5,12 +5,13 @@ Utility functions to run a terminal (connected via socat(1)) on each host.
 Requires socat(1) and xterm(1).
 Optionally uses gnome-terminal.
 """
-
-from os import environ
-
 from mininet.log import error
 from mininet.util import quietRun, errRun
 
+from os import environ, getpid, path
+from subprocess import Popen
+from tempfile import NamedTemporaryFile
+
 def tunnelX11( node, display=None):
     """Create an X11 tunnel from node:6000 to the root host
        display: display on root host (optional)
@@ -28,14 +29,65 @@ def tunnelX11( node, display=None):
         quietRun( 'xhost +si:localuser:root' )
         return display, None
     else:
-        # Create a tunnel for the TCP connection
+        # XXX Need to handle case where node is in UTS namespace
+        # in this case, we need to set XAUTHORITY to a private
+        # xauth file
+        if False and 'uts' in node.inNamespace and  not hasattr( node, 'xauthFile' ):
+            node.xauthFile = NamedTemporaryFile()
+            quietRun( 'xauth extract $DISPLAY | xauth -f %s merge' % node.xauthFile.name )
         port = 6000 + int( float( screen ) )
-        connection = r'TCP\:%s\:%s' % ( host, port )
-        cmd = [ "socat", "TCP-LISTEN:%d,fork,reuseaddr" % port,
-                "EXEC:'mnexec -a 1 socat STDIO %s'" % connection ]
+        # This can conflict if we are running nested Mininet
+        # in a pid namespace
+        socket = '/tmp/mininet.x11.%d' % getpid()
+        if not path.exists( socket ):
+            cmd = 'socat unix-listen:%s,fork tcp:localhost:%d' % ( socket, port  )
+            # Should be shut down when mn shuts down
+            tunnelX11.socket = Popen( cmd, shell=True )
+        # Create a tunnel for the TCP connection
+        cmd = 'socat tcp-listen:%d,fork,reuseaddr unix:%s' % ( port, socket )
+
     return 'localhost:' + screen, node.popen( cmd )
 
-def makeTerm( node, title='Node', term='xterm', display=None ):
+"""
+
+With pid namespaces, we can't easily escape the pid jail using mnexec
+(or can we?)
+
+What we can do, however, is create a unix socket in /tmp which connects
+to our x server, and a tcp listener in the host that connects to our
+unix socket!
+
+We just have to make sure that we clean up our processes and sockets
+when we quit.
+
+If we're using UTS namespaces, then xauth will get confused because
+our hostname has changed. So, we use a private XAUTHORITY file per
+host. We need to initialize this file with the key of our X server;
+this is a bit hard to figure out because the usual xauth list $DISPLAY
+may fail even if xlib can figure out a backup cookie to use.
+
+But what about namespace conflicts? This could certainly be very
+annoying for nested mininet!! In this case, our nested mininet servers
+should have a private /tmp that they can use.... except that conflicts
+with the shared unix domain socket in /tmp! ;-p
+
+We could also use a random name for the socket, to avoid the namespace
+conflict, although it's not clear what to do for cleanup to avoid
+blasting this....
+
+Perhaps the best idea is to use a canonical name (mininet.x11) and if the
+first name fails, try a second name??
+
+Another idea is to create a tmp dir for Mininet based on the pid of the
+mn process.....
+
+recommendation: for now, use mininet.x11.1234 as the socket.
+
+"""
+
+
+
+def makeTerm( node, title='Node', term='xterm', display=None, cmd='bash'):
     """Create an X11 tunnel to the node and start up a terminal.
        node: Node object
        title: base title
@@ -54,7 +106,8 @@ def makeTerm( node, title='Node', term='xterm', display=None ):
     display, tunnel = tunnelX11( node, display )
     if display is None:
         return []
-    term = node.popen( cmds[ term ] + [ display, '-e', 'env TERM=ansi bash'] )
+    term = node.popen( cmds[ term ] +
+                       [ display, '-e', 'env TERM=ansi %s' % cmd ] )
     return [ tunnel, term ] if tunnel else [ term ]
 
 def runX11( node, cmd ):

mininet/test/test_switchdpidassignment.py

@@ -8,8 +8,7 @@ import sys
 
 from mininet.net import Mininet
 from mininet.node import Host, Controller
-from mininet.node import ( UserSwitch, OVSSwitch, OVSLegacyKernelSwitch,
-                           IVSSwitch )
+from mininet.node import ( UserSwitch, OVSSwitch, IVSSwitch )
 from mininet.topo import Topo
 from mininet.log import setLogLevel
 from mininet.util import quietRun
@@ -81,12 +80,6 @@ class testSwitchOVSUser( TestSwitchDpidAssignmentOVS ):
     "Test dpid assignnment of OVS User Switch."
     switchClass = OVSUser
 
-@unittest.skipUnless( quietRun( 'which ovs-openflowd' ),
-                      'OVS Legacy Kernel switch is not installed' )
-class testSwitchOVSLegacyKernel( TestSwitchDpidAssignmentOVS ):
-    "Test dpid assignnment of OVS Legacy Kernel Switch."
-    switchClass = OVSLegacyKernelSwitch
-
 @unittest.skipUnless( quietRun( 'which ivs-ctl' ),
                       'IVS switch is not installed' )
 class testSwitchIVS( TestSwitchDpidAssignmentOVS ):

mininet/topo.py

@@ -318,6 +318,12 @@ class SingleSwitchReversedTopo( Topo ):
                           port1=0, port2=( k - h + 1 ) )
 
 
+class MinimalTopo( SingleSwitchTopo ):
+    "Minimal topology with two hosts and one switch"
+    def build( self ):
+        return SingleSwitchTopo.build( self, k=2 )
+
+
 class LinearTopo( Topo ):
     "Linear topology of k switches, with n hosts per switch."
 

mininet/topolib.py

@@ -45,10 +45,18 @@ class TorusTopo( Topo ):
        without STP turned on! It can be used with STP, e.g.:
        # mn --topo torus,3,3 --switch lxbr,stp=1 --test pingall"""
 
-    def build( self, x, y ):
+    def build( self, x, y, n=1 ):
+        """x: dimension of torus in x-direction
+           y: dimension of torus in y-direction
+           n: number of hosts per switch"""
         if x < 3 or y < 3:
             raise Exception( 'Please use 3x3 or greater for compatibility '
                              'with 2.1' )
+        if n == 1:
+            genHostName = lambda loc, k: 'h%s' % ( loc )
+        else:
+            genHostName = lambda loc, k: 'h%sx%d' % ( loc, k )
+
         hosts, switches, dpid = {}, {}, 0
         # Create and wire interior
         for i in range( 0, x ):
@@ -57,9 +65,10 @@ class TorusTopo( Topo ):
                 # dpid cannot be zero for OVS
                 dpid = ( i + 1 ) * 256 + ( j + 1 )
                 switch = switches[ i, j ] = self.addSwitch(
-                    's' + loc, dpid='%016x' % dpid )
-                host = hosts[ i, j ] = self.addHost( 'h' + loc )
-                self.addLink( host, switch )
+                    's' + loc, dpid='%x' % dpid )
+                for k in range( 0, n ):
+                    host = hosts[ i, j, k ] = self.addHost( genHostName( loc, k + 1 ) )
+                    self.addLink( host, switch )
         # Connect switches
         for i in range( 0, x ):
             for j in range( 0, y ):

mininet/util.py

@@ -56,6 +56,7 @@ def oldQuietRun( *cmd ):
 # This is a bit complicated, but it enables us to
 # monitor command output as it is happening
 
+# pylint: disable=too-many-branches
 def errRun( *cmd, **kwargs ):
     """Run a command and return stdout, stderr and return code
        cmd: string or list of command and args
@@ -77,6 +78,7 @@ def errRun( *cmd, **kwargs ):
         cmd = [ str( arg ) for arg in cmd ]
     elif isinstance( cmd, list ) and shell:
         cmd = " ".join( arg for arg in cmd )
+    debug( '*** errRun:', cmd, '\n' )
     popen = Popen( cmd, stdout=PIPE, stderr=stderr, shell=shell )
     # We use poll() because select() doesn't work with large fd numbers,
     # and thus communicate() doesn't work either
@@ -91,21 +93,31 @@ def errRun( *cmd, **kwargs ):
         errDone = False
     while not outDone or not errDone:
         readable = poller.poll()
-        for fd, _event in readable:
+        for fd, event in readable:
             f = fdtofile[ fd ]
-            data = f.read( 1024 )
-            if echo:
-                output( data )
-            if f == popen.stdout:
-                out += data
-                if data == '':
+            if event & POLLIN:
+                data = f.read( 1024 )
+                if echo:
+                    output( data )
+                if f == popen.stdout:
+                    out += data
+                    if data == '':
+                        outDone = True
+                elif f == popen.stderr:
+                    err += data
+                    if data == '':
+                        errDone = True
+            else:  # POLLHUP or something unexpected
+                if f == popen.stdout:
                     outDone = True
-            elif f == popen.stderr:
-                err += data
-                if data == '':
+                elif f == popen.stderr:
                     errDone = True
+                poller.unregister( fd )
+
     returncode = popen.wait()
+    debug( out, err, returncode )
     return out, err, returncode
+# pylint: enable=too-many-branches
 
 def errFail( *cmd, **kwargs ):
     "Run a command using errRun and raise exception on nonzero exit"
@@ -145,27 +157,42 @@ isShellBuiltin.builtIns = None
 # live in the root namespace and thus do not have to be
 # explicitly moved.
 
-def makeIntfPair( intf1, intf2, addr1=None, addr2=None, runCmd=quietRun ):
-    """Make a veth pair connecting intf1 and intf2.
-       intf1: string, interface
-       intf2: string, interface
+def makeIntfPair( intf1, intf2, addr1=None, addr2=None, node1=None, node2=None,
+                  deleteIntfs=True, runCmd=None ):
+    """Make a veth pair connnecting new interfaces intf1 and intf2
+       intf1: name for interface 1
+       intf2: name for interface 2
+       addr1: MAC address for interface 1 (optional)
+       addr2: MAC address for interface 2 (optional)
+       node1: home node for interface 1 (optional)
+       node2: home node for interface 2 (optional)
+       deleteIntfs: delete intfs before creating them
        runCmd: function to run shell commands (quietRun)
-       returns: ip link add result"""
-    # Delete any old interfaces with the same names
-    runCmd( 'ip link del ' + intf1 )
-    runCmd( 'ip link del ' + intf2 )
+       raises Exception on failure"""
+    if not runCmd:
+        runCmd = quietRun if not node1 else node1.cmd
+        runCmd2 = quietRun if not node2 else node2.cmd
+    if deleteIntfs:
+        # Delete any old interfaces with the same names
+        runCmd( 'ip link del ' + intf1 )
+        runCmd2( 'ip link del ' + intf2 )
     # Create new pair
+    ns1 = 1 if not node1 else node1.pid
+    ns2 = 1 if not node2 else node2.pid
     if addr1 is None and addr2 is None:
-        cmd = 'ip link add name ' + intf1 + ' type veth peer name ' + intf2
+        cmdOutput = runCmd( 'ip link add name %s netns %s '
+                            'type veth peer name %s '
+                            'netns %s' % ( intf1, ns1, intf2, ns2 ) )
     else:
-        cmd = ( 'ip link add name ' + intf1 + ' address ' + addr1 +
-                ' type veth peer name ' + intf2 + ' address ' + addr2 )
-    cmdOutput = runCmd( cmd )
-    if cmdOutput == '':
-        return True
-    else:
-        error( "Error creating interface pair: %s " % cmdOutput )
-        return False
+        cmdOutput = runCmd( 'ip link add name %s '
+                            'address %s netns %s '
+                            'type veth peer name %s '
+                            'address %s '
+                            'netns %s' %
+                            (  intf1, addr1, ns1, intf2, addr2, ns2 ) )
+    if cmdOutput:
+        raise Exception( "Error creating interface pair (%s,%s): %s " %
+                         ( intf1, intf2, cmdOutput ) )
 
 def retry( retries, delaySecs, fn, *args, **keywords ):
     """Try something several times before giving up.
@@ -481,47 +508,91 @@ def custom( cls, **params ):
     customized.__name__ = 'custom(%s,%s)' % ( cls, params )
     return customized
 
-def splitArgs( argstr ):
-    """Split argument string into usable python arguments
-       argstr: argument string with format fn,arg2,kw1=arg3...
-       returns: fn, args, kwargs"""
-    split = argstr.split( ',' )
-    fn = split[ 0 ]
-    params = split[ 1: ]
+
+
+def parseArgs( argstr ):
+    """Parse argument string
+       returns args, kwargs"""
+    # One step at a time: support param=[a,b,c]
+    paramre = r'\[[^\]]*\]|[^,\[\]]+'
+    paramre = r'\w+=\[[^\]]*\]|\w+\[^,\[\]]+' + paramre
+    params = re.findall( paramre, argstr )
+    # Parse lists
+    for i, arg in enumerate( params ):
+        if arg.startswith( '[' ) and arg.endswith( ']' ):
+            arg = arg[ 1 : -1 ]
+            print "recurse on", arg
+            params[ i ] = parseArgs( args )
     # Convert int and float args; removes the need for function
     # to be flexible with input arg formats.
     args = [ makeNumeric( s ) for s in params if '=' not in s ]
     kwargs = {}
     for s in [ p for p in params if '=' in p ]:
-        key, val = s.split( '=', 1 )
-        kwargs[ key ] = makeNumeric( val )
-    return fn, args, kwargs
+        key, arg = s.split( '=', 1 )
+        if arg.startswith( '[' ) and arg.endswith( ']' ):
+            arg = arg[ 1 : -1 ]
+            print 'recurse on', arg
+            arg = parseArgs( arg )[ 0 ]
+        else:
+            arg = makeNumeric( arg )
+        kwargs[ key ] = arg
+    return args, kwargs
 
-def customConstructor( constructors, argStr ):
-    """Return custom constructor based on argStr
-    The args and key/val pairs in argsStr will be automatically applied
-    when the generated constructor is later used.
+def splitArgs( argstr ):
+    """Split argument string into usable python arguments
+       argstr: argument string with format fn,arg2,kw1=arg3...
+       returns: fn, args, kwargs"""
+    args, kwargs = parseArgs( argstr )
+    return args[ 0 ], args[ 1: ], kwargs
+
+def customClass( classes, argStr ):
+    """Return customized class based on argStr
+    The args and key/val pairs in argStr will be automatically applied
+    when the generated class is later used.
     """
-    cname, newargs, kwargs = splitArgs( argStr )
-    constructor = constructors.get( cname, None )
-
-    if not constructor:
+    cname, args, kwargs = splitArgs( argStr )
+    cls = classes.get( cname, None )
+    if not cls:
         raise Exception( "error: %s is unknown - please specify one of %s" %
-                         ( cname, constructors.keys() ) )
+                         ( cname, classes.keys() ) )
+    if not args and not kwargs:
+        return cls
 
-    def customized( name, *args, **params ):
-        "Customized constructor, useful for Node, Link, and other classes"
-        params = params.copy()
-        params.update( kwargs )
-        if not newargs:
-            return constructor( name, *args, **params )
-        if args:
-            warn( 'warning: %s replacing %s with %s\n' % (
-                  constructor, args, newargs ) )
-        return constructor( name, *newargs, **params )
+    return specialClass( cls, append=args, defaults=kwargs )
+
+def specialClass( cls, prepend=None, append=None,
+                  defaults=None, override=None ):
+    """Like functools.partial, but it returns a class
+       prepend: arguments to prepend to argument list
+       append: arguments to append to argument list
+       defaults: default values for keyword arguments
+       override: keyword arguments to override"""
+
+    if prepend is None:
+        prepend = []
+
+    if append is None:
+        append = []
+
+    if defaults is None:
+        defaults = {}
+
+    if override is None:
+        override = {}
+
+    class CustomClass( cls ):
+        "Customized subclass with preset args/params"
+        def __init__( self, *args, **params ):
+            newparams = defaults.copy()
+            newparams.update( params )
+            newparams.update( override )
+            cls.__init__( self, *( list( prepend ) + list( args ) +
+                                   list( append ) ),
+                          **newparams )
+
+    CustomClass.__name__ = '%s%s' % ( cls.__name__, defaults )
+    return CustomClass
 
-    customized.__name__ = 'customConstructor(%s)' % argStr
-    return customized
 
 def buildTopo( topos, topoStr ):
     """Create topology from string with format (object, arg1, arg2,...).
@@ -551,18 +622,20 @@ def waitListening( client=None, server='127.0.0.1', port=80, timeout=None ):
         raise Exception('Could not find telnet' )
     # pylint: disable=maybe-no-member
     serverIP = server if isinstance( server, basestring ) else server.IP()
-    cmd = ( 'sh -c "echo A | telnet -e A %s %s"' %
-            ( serverIP, port ) )
+    cmd = ( 'echo A | telnet -e A %s %s' % ( serverIP, port ) )
     time = 0
-    while 'Connected' not in runCmd( cmd ):
-        if timeout:
-            print time
-            if time >= timeout:
-                error( 'could not connect to %s on port %d\n'
-                       % ( server, port ) )
-                return False
-        output('waiting for', server,
-               'to listen on port', port, '\n')
+    result = runCmd( cmd )
+    while 'Connected' not in result:
+        if 'No route' in result:
+            rtable = runCmd( 'route' )
+            error( 'no route to %s:\n%s' % ( server, rtable ) )
+            return False
+        if timeout and time >= timeout:
+            error( 'could not connect to %s on port %d\n' % ( server, port ) )
+            return False
+        debug( 'waiting for', server, 'to listen on port', port, '\n' )
+        info( '.' )
         sleep( .5 )
         time += .5
+        result = runCmd( cmd )
     return True

mnexec.c

@@ -5,9 +5,9 @@
  *
  *  - closing all file descriptors except stdin/out/error
  *  - detaching from a controlling tty using setsid
- *  - running in network and mount namespaces
+ *  - running in network and other namespaces
  *  - printing out the pid of a process so we can identify it later
- *  - attaching to a namespace and cgroup
+ *  - attaching to namespace(s) and cgroup
  *  - setting RT scheduling
  *
  * Partially based on public domain setsid(1)
@@ -24,21 +24,26 @@
 #include <sched.h>
 #include <ctype.h>
 #include <sys/mount.h>
+#include <sys/wait.h>
 
 #if !defined(VERSION)
 #define VERSION "(devel)"
 #endif
 
+
 void usage(char *name)
 {
     printf("Execution utility for Mininet\n\n"
-           "Usage: %s [-cdnp] [-a pid] [-g group] [-r rtprio] cmd args...\n\n"
+           "Usage: %s [-cdmnPpu] [-a pid] [-g group] [-r rtprio] cmd args...\n\n"
            "Options:\n"
            "  -c: close all file descriptors except stdin/out/error\n"
            "  -d: detach from tty by calling setsid()\n"
-           "  -n: run in new network and mount namespaces\n"
+           "  -m: run in a new mount namespace\n"
+           "  -n: run in a new network namespace\n"
+           "  -P: run in a new pid namespace\n"
+           "  -u: run in a new UTS (ipc, hostname) namespace\n"
            "  -p: print ^A + pid\n"
-           "  -a pid: attach to pid's network and mount namespaces\n"
+           "  -a pid: attach to pid's namespaces\n"
            "  -g group: add to cgroup\n"
            "  -r rtprio: run with SCHED_RR (usually requires -g)\n"
            "  -v: print version\n",
@@ -51,6 +56,7 @@ int setns(int fd, int nstype)
     return syscall(__NR_setns, fd, nstype);
 }
 
+
 /* Validate alphanumeric path foo1/bar2/baz */
 void validate(char *path)
 {
@@ -63,6 +69,7 @@ void validate(char *path)
     }
 }
 
+
 /* Add our pid to cgroup */
 void cgroup(char *gname)
 {
@@ -92,107 +99,181 @@ void cgroup(char *gname)
     }
 }
 
+/* Attach to ns 'name' if present */
+int attachns( pid_t pid, char *name ) {
+    char path[PATH_MAX];
+    int nsid;
+
+    sprintf(path, "/proc/%d/ns/%s", pid, name) ;
+
+    if ((nsid = open(path, O_RDONLY)) < 0)
+        return nsid;
+
+    if (setns(nsid, 0) != 0) {
+        perror("setns");
+        return 1;
+    }
+
+    return 0;
+}
+
+/* Attach to pid's namespaces - returns true if pidns */
+int attach(int pid) {
+
+    char *cwd = get_current_dir_name();
+    char path[PATH_MAX];
+    int pidns = 0;
+
+    attachns(pid, "net");
+    attachns(pid, "uts");
+    if ( attachns(pid, "pid") == 0 )
+        pidns = 1;
+
+    if (attachns(pid, "mnt") != 0) {
+        /* Plan B: chroot into pid's root file system */
+        sprintf(path, "/proc/%d/root", pid);
+        if (chroot(path) < 0) {
+            perror(path);
+        }
+    }
+
+    /* chdir to correct working directory */
+    if (chdir(cwd) != 0) {
+        perror(cwd);
+    }
+
+    return pidns;
+}
+
+
 int main(int argc, char *argv[])
 {
     int c;
-    int fd;
-    char path[PATH_MAX];
-    int nsid;
-    int pid;
-    char *cwd = get_current_dir_name();
 
-    static struct sched_param sp;
-    while ((c = getopt(argc, argv, "+cdnpa:g:r:vh")) != -1)
+    /* Argument flags */
+    int flags = 0;
+    int closefds = 0;
+    int attachpid = 0;
+    char *cgrouparg = NULL;
+    int detachtty = 0;
+    int printpid = 0;
+    int rtprio = 0;
+    int pidns = 0;
+
+    while ((c = getopt(argc, argv, "+cdmnPpa:g:r:uvh")) != -1)
         switch(c) {
-        case 'c':
-            /* close file descriptors except stdin/out/error */
-            for (fd = getdtablesize(); fd > 2; fd--)
-                close(fd);
-            break;
-        case 'd':
-            /* detach from tty */
-            if (getpgrp() == getpid()) {
-                switch(fork()) {
-                    case -1:
-                        perror("fork");
-                        return 1;
-                    case 0:     /* child */
-                        break;
-                    default:    /* parent */
-                        return 0;
-                }
-            }
-            setsid();
-            break;
-        case 'n':
-            /* run in network and mount namespaces */
-            if (unshare(CLONE_NEWNET|CLONE_NEWNS) == -1) {
-                perror("unshare");
+            case 'c': closefds = 1; break;
+            case 'd':   detachtty = 1; break;
+            case 'm':   flags |= CLONE_NEWNS; break;
+            case 'n':   flags |= CLONE_NEWNET; break;
+            case 'p':   printpid = 1; break;
+            case 'P':   flags |= CLONE_NEWPID; break;
+            case 'a':   attachpid = atoi(optarg);break;
+            case 'g':   cgrouparg = optarg ; break;
+            case 'r':   rtprio = atoi(optarg); break;
+            case 'u':   flags |= CLONE_NEWUTS; break;
+            case 'v':   printf("%s\n", VERSION); exit(0);
+            case 'h':   usage(argv[0]); exit(0);
+            default:    usage(argv[0]); exit(1);
+    }
+
+    if (closefds) {
+        /* close file descriptors except stdin/out/error */
+        int fd;
+        for (fd = getdtablesize(); fd > 2; fd--) close(fd);
+    }
+
+    /* XXX We should not fork twice if we don't need to!! */
+    if (detachtty) {
+        /* detach from tty */
+        if (getpgrp() == getpid()) {
+            switch(fork()) {
+            case -1:
+                perror("fork");
                 return 1;
+            case 0:     /* child */
+                break;
+            default:    /* parent */
+                return 0;
             }
-            /* mount sysfs to pick up the new network namespace */
-            if (mount("sysfs", "/sys", "sysfs", MS_MGC_VAL, NULL) == -1) {
-                perror("mount");
-                return 1;
-            }
-            break;
-        case 'p':
-            /* print pid */
-            printf("\001%d\n", getpid());
-            fflush(stdout);
-            break;
-        case 'a':
-            /* Attach to pid's network namespace and mount namespace */
-            pid = atoi(optarg);
-            sprintf(path, "/proc/%d/ns/net", pid);
-            nsid = open(path, O_RDONLY);
-            if (nsid < 0) {
-                perror(path);
-                return 1;
-            }
-            if (setns(nsid, 0) != 0) {
-                perror("setns");
-                return 1;
-            }
-            /* Plan A: call setns() to attach to mount namespace */
-            sprintf(path, "/proc/%d/ns/mnt", pid);
-            nsid = open(path, O_RDONLY);
-            if (nsid < 0 || setns(nsid, 0) != 0) {
-                /* Plan B: chroot/chdir into pid's root file system */
-                sprintf(path, "/proc/%d/root", pid);
-                if (chroot(path) < 0) {
-                    perror(path);
-                    return 1;
-                }
-            }
-            /* chdir to correct working directory */
-            if (chdir(cwd) != 0) {
-                perror(cwd);
-                return 1;
-            }
-            break;
-        case 'g':
-            /* Attach to cgroup */
-            cgroup(optarg);
-            break;
-        case 'r':
-            /* Set RT scheduling priority */
-            sp.sched_priority = atoi(optarg);
-            if (sched_setscheduler(getpid(), SCHED_RR, &sp) < 0) {
-                perror("sched_setscheduler");
-                return 1;
-            }
-            break;
-        case 'v':
-            printf("%s\n", VERSION);
-            exit(0);
-        case 'h':
-            usage(argv[0]);
-            exit(0);
-        default:
-            usage(argv[0]);
-            exit(1);
         }
+        setsid();
+    }
+
+    if (attachpid) {
+        /* Attach to existing namespace(s) */
+      pidns = attach(attachpid);
+    }
+    else {
+        /* Create new namespace(s) */
+        if (unshare(flags) == -1) {
+            perror("unshare");
+            return 1;
+        }
+    }
+
+    /* Use a new process group so we can use killpg */
+    setpgid( 0, 0 );
+
+    if ( flags & CLONE_NEWPID || pidns ) {
+        /* For pid namespace, we need to fork and wait for child ;-( */
+        pid_t pid = fork();
+        switch(pid) {
+            int status;
+            case -1:
+                perror("fork");
+                return 1;
+            case 0:
+                /* child continues below */
+                break;
+            default:
+                /* We print the *child pid* if needed */
+                if (printpid) {
+                    printf("\001%d\n", pid);
+                    fflush(stdout);
+                }
+                /* Parent needs to wait for child and exit */
+                wait(&status);
+                return 0;
+        }
+    }
+
+    else if (printpid) {
+        /* If we're in a pid namespace, parent prints our pid instead */
+        printf("\001%d\n", getpid());
+        fflush(stdout);
+    }
+
+
+    if (flags & CLONE_NEWNS && !attachpid) {
+        /* Child remounts /proc for ps */
+        if (mount("proc", "/proc", "proc", MS_MGC_VAL, NULL) == -1) {
+            perror("mountproc");
+        }
+    }
+
+    if (cgrouparg) {
+        /* Attach to cgroup */
+        cgroup(cgrouparg);
+    }
+
+    if (flags & CLONE_NEWNET & CLONE_NEWNS) {
+        /* Mount sysfs to pick up the new network namespace */
+        if (mount("sysfs", "/sys", "sysfs", MS_MGC_VAL, NULL) == -1) {
+            perror("mount");
+            return 1;
+        }
+    }
+
+    if (rtprio != 0) {
+        /* Set RT scheduling priority */
+        static struct sched_param sp;
+        sp.sched_priority = atoi(optarg);
+        if (sched_setscheduler(getpid(), SCHED_RR, &sp) < 0) {
+            perror("sched_setscheduler");
+            return 1;
+        }
+    }
 
     if (optind < argc) {
         execvp(argv[optind], &argv[optind]);

util/clustersetup.sh

@@ -25,28 +25,40 @@ clean=false
 declare -a hosts=()
 user=$(whoami)
 SSHDIR=/tmp/mn/ssh
-USERDIR=/home/$user/.ssh
-usage=$'./clustersetup.sh [ -p|h|c ] [ host1 ] [ host2 ] ...\n
+USERDIR=$HOME/.ssh
+usage="./clustersetup.sh [ -p|h|c ] [ host1 ] [ host2 ] ...\n
         Authenticate yourself and other cluster nodes to each other
         via ssh for mininet cluster edition. By default, we use a
         temporary ssh setup. An ssh directory is mounted over
-        /home/user/.ssh on each machine in the cluster.
-        
+        $USERDIR on each machine in the cluster.
+
                 -h: display this help
                 -p: create a persistent ssh setup. This will add
                     new ssh keys and known_hosts to each nodes
-                    /home/user/.ssh directory
+                    $USERDIR directory
                 -c: method to clean up a temporary ssh setup.
                     Any hosts taken as arguments will be cleaned
-        '
+        "
+
 
 persistentSetup() {
     echo "***creating key pair"
-    ssh-keygen -t rsa -C "Cluster_Edition_Key" -f $USERDIR/cluster_key -N '' &> /dev/null
-    cat $USERDIR/cluster_key.pub >> $USERDIR/authorized_keys
+    if ! ssh-keygen -t rsa -C "Cluster_Edition_Key" -f $USERDIR/cluster_key -N ''; then
+        echo "key generation returned error" $?
+    fi
+
+    echo "***adding to authorized_keys"
+    if ! grep -f $USERDIR/cluster_key.pub $USERDIR/authorized_keys > /dev/null; then
+        cat $USERDIR/cluster_key.pub >> $USERDIR/authorized_keys
+    else
+        echo "$USERDIR/cluster_key.pub already in $USERDIR/authorized_keys"
+    fi
+
     echo "***configuring ssh"
-    echo "IdentityFile $USERDIR/cluster_key" >> $USERDIR/config
-    echo "IdentityFile $USERDIR/id_rsa" >> $USERDIR/config
+    if ! grep cluster_key $USERDIR/config; then
+        echo "IdentityFile $USERDIR/cluster_key" >> $USERDIR/config
+        echo "IdentityFile $USERDIR/id_rsa" >> $USERDIR/config
+    fi
 
     for host in $hosts; do
         echo "***copying public key to $host"
@@ -56,58 +68,65 @@ persistentSetup() {
         scp $USERDIR/cluster_key.pub $user@$host:$USERDIR
         echo "***configuring remote host"
         ssh -o ForwardAgent=yes  $user@$host "
-        echo 'IdentityFile $USERDIR/cluster_key' >> $USERDIR/config
-        echo 'IdentityFile $USERDIR/id_rsa' >> $USERDIR/config"
+        if ! grep cluster_key $USERDIR/config > /dev/null; then
+            echo 'IdentityFile $USERDIR/cluster_key' >> $USERDIR/config
+            echo 'IdentityFile $USERDIR/id_rsa' >> $USERDIR/config;
+        fi"
     done
 
     for host in $hosts; do
         echo "***copying known_hosts to $host"
         scp $USERDIR/known_hosts $user@$host:$USERDIR/cluster_known_hosts
+        echo "***appending"
         ssh $user@$host "
-        cat $USERDIR/cluster_known_hosts >> $USERDIR/known_hosts
+        if ! grep -f $USERDIR/cluster_known_hosts $USERDIR/known_hosts > /dev/null; then
+            cat $USERDIR/cluster_known_hosts >> $USERDIR/known_hosts;
+        fi;
         rm $USERDIR/cluster_known_hosts"
     done
+
+    echo "***done with persistent setup"
 }
 
 tempSetup() {
-    
+
     echo "***creating temporary ssh directory"
-    mkdir -p $SSHDIR 
+    mkdir -p $SSHDIR
     echo "***creating key pair"
-    ssh-keygen -t rsa -C "Cluster_Edition_Key" -f /tmp/mn/ssh/id_rsa -N '' &> /dev/null
+    ssh-keygen -t rsa -C "Cluster_Edition_Key" -f $SSHDIR/id_rsa -N '' &> /dev/null
 
     echo "***mounting temporary ssh directory"
-    sudo mount --bind $SSHDIR /home/$user/.ssh
+    sudo mount --bind $SSHDIR $USERDIR
     cp $SSHDIR/id_rsa.pub $SSHDIR/authorized_keys
 
-for host in $hosts; do
-    echo "***copying public key to $host"
-    ssh-copy-id $user@$host &> /dev/null
-    echo "***mounting remote temporary ssh directory for $host"
-    ssh -o ForwardAgent=yes  $user@$host "
-    mkdir -p /tmp/mn/ssh
-    cp /home/$user/.ssh/authorized_keys $SSHDIR/authorized_keys
-    sudo mount --bind $SSHDIR /home/$user/.ssh"
-    echo "***copying key pair to $host"
-    scp $SSHDIR/{id_rsa,id_rsa.pub} $user@$host:$SSHDIR
-done
+    for host in $hosts; do
+        echo "***copying public key to $host"
+        ssh-copy-id $user@$host &> /dev/null
+        echo "***mounting remote temporary ssh directory for $host"
+        ssh -o ForwardAgent=yes  $user@$host "
+        mkdir -p $SSHDIR
+        cp $USERDIR/authorized_keys $SSHDIR/authorized_keys
+        sudo mount --bind $SSHDIR $USERDIR"
+        echo "***copying key pair to $host"
+        scp $SSHDIR/{id_rsa,id_rsa.pub} $user@$host:$SSHDIR
+    done
 
-for host in $hosts; do
-    echo "***copying known_hosts to $host"
-    scp $SSHDIR/known_hosts $user@$host:$SSHDIR
-done
+    for host in $hosts; do
+        echo "***copying known_hosts to $host"
+        scp $SSHDIR/known_hosts $user@$host:$SSHDIR
+    done
 }
 
 cleanup() {
-    
+
     for host in $hosts; do
     echo "***cleaning up $host"
-    ssh $user@$host "sudo umount /home/$user/.ssh
+    ssh $user@$host "sudo umount $USERDIR
                           sudo rm -rf $SSHDIR"
     done
 
     echo "**unmounting local directories"
-    sudo umount /home/$user/.ssh
+    sudo umount $USERDIR
     echo "***removing temporary ssh directory"
     sudo rm -rf $SSHDIR
     echo "done!"
@@ -145,9 +164,8 @@ if $showHelp; then
 fi
 
 for i in "$@"; do
-    output=$(getent ahostsv4 "$i")
-    if [ -z "$output" ]; then
-        echo '***WARNING: could not find hostname "$i"'
+    if ! getent ahostsv4 $i > /dev/null; then
+        echo "***WARNING: could not find hostname $i"
         echo ""
     else
         hosts+="$i "

util/install.sh

@@ -1,6 +1,6 @@
 #!/usr/bin/env bash
 
-# Mininet install script for Ubuntu (and Debian Lenny)
+# Mininet install script for Ubuntu (and Debian Wheezy+)
 # Brandon Heller (brandonh@stanford.edu)
 
 # Fail on error
@@ -102,7 +102,10 @@ OF13_SWITCH_REV=${OF13_SWITCH_REV:-""}
 function kernel {
     echo "Install Mininet-compatible kernel if necessary"
     sudo apt-get update
-    $install linux-image-$KERNEL_NAME
+    if ! $install linux-image-$KERNEL_NAME; then
+        echo "Could not install linux-image-$KERNEL_NAME"
+        echo "Skipping - assuming installed kernel is OK."
+    fi
 }
 
 function kernel_clean {
@@ -189,17 +192,24 @@ function of13 {
     fi
 
     # Install netbee
-    NBEESRC="nbeesrc-jan-10-2013"
+    if [ "$DIST" = "Ubuntu" ] && version_ge $RELEASE 14.04; then
+        NBEESRC="nbeesrc-feb-24-2015"
+        NBEEDIR="netbee"
+    else
+        NBEESRC="nbeesrc-jan-10-2013"
+        NBEEDIR="nbeesrc-jan-10-2013"
+    fi
+
     NBEEURL=${NBEEURL:-http://www.nbee.org/download/}
     wget -nc ${NBEEURL}${NBEESRC}.zip
     unzip ${NBEESRC}.zip
-    cd ${NBEESRC}/src
+    cd ${NBEEDIR}/src
     cmake .
     make
     cd $BUILD_DIR/
-    sudo cp ${NBEESRC}/bin/libn*.so /usr/local/lib
+    sudo cp ${NBEEDIR}/bin/libn*.so /usr/local/lib
     sudo ldconfig
-    sudo cp -R ${NBEESRC}/include/ /usr/
+    sudo cp -R ${NBEEDIR}/include/ /usr/
 
     # Resume the install:
     cd $BUILD_DIR/ofsoftswitch13
@@ -258,53 +268,63 @@ function ubuntuOvs {
     OVS_SRC=$BUILD_DIR/openvswitch
     OVS_TARBALL_LOC=http://openvswitch.org/releases
 
-    if [ "$DIST" = "Ubuntu" ] && version_ge $RELEASE 12.04; then
-        rm -rf $OVS_SRC
-        mkdir -p $OVS_SRC
-        cd $OVS_SRC
+    if ! echo "$DIST" | egrep "Ubuntu|Debian" > /dev/null; then
+        echo "OS must be Ubuntu or Debian"
+        $cd BUILD_DIR
+        return
+    fi
+    if [ "$DIST" = "Ubuntu" ] && ! version_ge $RELEASE 12.04; then
+        echo "Ubuntu version must be >= 12.04"
+        cd $BUILD_DIR
+        return
+    fi
+    if [ "$DIST" = "Debian" ] && ! version_ge $RELEASE 7.0; then
+        echo "Debian version must be >= 7.0"
+        cd $BUILD_DIR
+        return
+    fi
 
-        if wget $OVS_TARBALL_LOC/openvswitch-$OVS_RELEASE.tar.gz 2> /dev/null; then
-            tar xzf openvswitch-$OVS_RELEASE.tar.gz
-        else
-            echo "Failed to find OVS at $OVS_TARBALL_LOC/openvswitch-$OVS_RELEASE.tar.gz"
-            cd $BUILD_DIR
-            return
-        fi
+    rm -rf $OVS_SRC
+    mkdir -p $OVS_SRC
+    cd $OVS_SRC
 
-        # Remove any old packages
-        $remove openvswitch-common openvswitch-datapath-dkms openvswitch-controller \
-                openvswitch-pki openvswitch-switch
-
-        # Get build deps
-        $install build-essential fakeroot debhelper autoconf automake libssl-dev \
-                 pkg-config bzip2 openssl python-all procps python-qt4 \
-                 python-zopeinterface python-twisted-conch dkms
-
-        # Build OVS
-        cd $BUILD_DIR/openvswitch/openvswitch-$OVS_RELEASE
-                DEB_BUILD_OPTIONS='parallel=2 nocheck' fakeroot debian/rules binary
-        cd ..
-        $pkginst openvswitch-common_$OVS_RELEASE*.deb openvswitch-datapath-dkms_$OVS_RELEASE*.deb \
-                 openvswitch-pki_$OVS_RELEASE*.deb openvswitch-switch_$OVS_RELEASE*.deb
-        if $pkginst openvswitch-controller_$OVS_RELEASE*.deb; then
-            echo "Ignoring error installing openvswitch-controller"
-        fi
-
-        modinfo openvswitch
-        sudo ovs-vsctl show
-        # Switch can run on its own, but
-        # Mininet should control the controller
-        # This appears to only be an issue on Ubuntu/Debian
-        if sudo service openvswitch-controller stop; then
-            echo "Stopped running controller"
-        fi
-        if [ -e /etc/init.d/openvswitch-controller ]; then
-            sudo update-rc.d openvswitch-controller disable
-        fi
+    if wget $OVS_TARBALL_LOC/openvswitch-$OVS_RELEASE.tar.gz 2> /dev/null; then
+        tar xzf openvswitch-$OVS_RELEASE.tar.gz
     else
-        echo "Failed to install Open vSwitch.  OS must be Ubuntu >= 12.04"
-            cd $BUILD_DIR
-            return
+        echo "Failed to find OVS at $OVS_TARBALL_LOC/openvswitch-$OVS_RELEASE.tar.gz"
+        cd $BUILD_DIR
+        return
+    fi
+
+    # Remove any old packages
+    $remove openvswitch-common openvswitch-datapath-dkms openvswitch-controller \
+            openvswitch-pki openvswitch-switch
+
+    # Get build deps
+    $install build-essential fakeroot debhelper autoconf automake libssl-dev \
+             pkg-config bzip2 openssl python-all procps python-qt4 \
+             python-zopeinterface python-twisted-conch dkms
+
+    # Build OVS
+    cd $BUILD_DIR/openvswitch/openvswitch-$OVS_RELEASE
+            DEB_BUILD_OPTIONS='parallel=2 nocheck' fakeroot debian/rules binary
+    cd ..
+    $pkginst openvswitch-common_$OVS_RELEASE*.deb openvswitch-datapath-dkms_$OVS_RELEASE*.deb \
+             openvswitch-pki_$OVS_RELEASE*.deb openvswitch-switch_$OVS_RELEASE*.deb
+    if $pkginst openvswitch-controller_$OVS_RELEASE*.deb; then
+        echo "Ignoring error installing openvswitch-controller"
+    fi
+
+    modinfo openvswitch
+    sudo ovs-vsctl show
+    # Switch can run on its own, but
+    # Mininet should control the controller
+    # This appears to only be an issue on Ubuntu/Debian
+    if sudo service openvswitch-controller stop; then
+        echo "Stopped running controller"
+    fi
+    if [ -e /etc/init.d/openvswitch-controller ]; then
+        sudo update-rc.d openvswitch-controller disable
     fi
 }
 
@@ -319,14 +339,17 @@ function ovs {
         return
     fi
 
-    # Manually installing openvswitch-datapath may be necessary
-    # for manually built kernel .debs using Debian's defective kernel
-    # packaging, which doesn't yield usable headers.
-    if ! dpkg --get-selections | grep openvswitch-datapath; then
-        # If you've already installed a datapath, assume you
-        # know what you're doing and don't need dkms datapath.
-        # Otherwise, install it.
-        $install openvswitch-datapath-dkms
+    if [ "$DIST" = "Ubuntu" ] && ! version_ge $RELEASE 14.04; then
+        # Older Ubuntu versions need openvswitch-datapath/-dkms
+        # Manually installing openvswitch-datapath may be necessary
+        # for manually built kernel .debs using Debian's defective kernel
+        # packaging, which doesn't yield usable headers.
+        if ! dpkg --get-selections | grep openvswitch-datapath; then
+            # If you've already installed a datapath, assume you
+            # know what you're doing and don't need dkms datapath.
+            # Otherwise, install it.
+            $install openvswitch-datapath-dkms
+        fi
     fi
 
     $install openvswitch-switch
@@ -343,7 +366,7 @@ function ovs {
     else
         echo "Attempting to install openvswitch-testcontroller"
         if ! $install openvswitch-testcontroller; then
-            echo "Failed - giving up"
+            echo "Failed - skipping openvswitch-testcontroller"
         fi
     fi
 
@@ -567,7 +590,7 @@ net.ipv6.conf.lo.disable_ipv6 = 1' | sudo tee -a /etc/sysctl.conf > /dev/null
     # Since the above doesn't disable neighbor discovery, also do this:
     if ! grep 'ipv6.disable' /etc/default/grub; then
         sudo sed -i -e \
-        's/GRUB_CMDLINE_LINUX_DEFAULT="/GRUB_CMDLINE_LINUX_DEFAULT="ipv6.disable=1/' \
+        's/GRUB_CMDLINE_LINUX_DEFAULT="/GRUB_CMDLINE_LINUX_DEFAULT="ipv6.disable=1 /' \
         /etc/default/grub
         sudo update-grub
     fi

util/m

@@ -40,5 +40,5 @@ if [ -d $rootdir -a -x $rootdir/bin/bash ]; then
     cmd="chroot $rootdir /bin/bash -c $cmd"
 fi
 
-cmd="exec sudo mnexec $cg -a $pid $cmd"
+cmd="exec sudo -E mnexec $cg -a $pid $cmd"
 eval $cmd

util/versioncheck.py

@@ -8,7 +8,7 @@ version = 'Mininet ' + co( 'PYTHONPATH=. bin/mn --version', shell=True )
 version = version.strip()
 
 # Find all Mininet path references
-lines = co( "egrep -or 'Mininet [0-9\.]+\w*' *", shell=True )
+lines = co( "egrep -or 'Mininet [0-9\.\+]+\w*' *", shell=True )
 
 error = False
 

util/vm/build.py

@@ -323,7 +323,7 @@ zerombr yes
 clearpart --all --initlabel
 #Automatic partitioning
 autopart
-#System authorization infomation
+#System authorization information
 auth  --useshadow  --enablemd5
 #Firewall configuration
 firewall --disabled