4 ### Connect to remote peers, and keep track of them
6 ### (c) 2007 Straylight/Edgeware
9 ###----- Licensing notice ---------------------------------------------------
11 ### This file is part of Trivial IP Encryption (TrIPE).
13 ### TrIPE is free software: you can redistribute it and/or modify it under
14 ### the terms of the GNU General Public License as published by the Free
15 ### Software Foundation; either version 3 of the License, or (at your
16 ### option) any later version.
18 ### TrIPE is distributed in the hope that it will be useful, but WITHOUT
19 ### ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
20 ### FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
23 ### You should have received a copy of the GNU General Public License
24 ### along with TrIPE. If not, see <https://www.gnu.org/licenses/>.
28 ###--------------------------------------------------------------------------
29 ### External dependencies.
31 from optparse import OptionParser
41 import subprocess as PROC
45 ###--------------------------------------------------------------------------
46 ### Running auxiliary commands.
48 class SelLineQueue (M.SelLineBuffer):
49 """Glues the select-line-buffer into the coroutine queue system."""
51 def __new__(cls, file, queue, tag, kind):
52 """See __init__ for documentation."""
53 return M.SelLineBuffer.__new__(cls, file.fileno())
55 def __init__(me, file, queue, tag, kind):
57 Initialize a new line-reading adaptor.
59 The adaptor reads lines from FILE. Each line is inserted as a message of
60 the stated KIND, bearing the TAG, into the QUEUE. End-of-file is
71 me._q.put((me._tag, me._kind, line))
76 me._q.put((me._tag, me._kind, None))
78 class ErrorWatch (T.Coroutine):
80 An object which watches stderr streams for errors and converts them into
83 WARN connect INFO stderr LINE
85 The INFO is a list of tokens associated with the file when it was
88 Usually there is a single ErrorWatch object, called errorwatch.
92 """Initialization: there are no arguments."""
93 T.Coroutine.__init__(me)
98 def watch(me, file, info):
100 Adds FILE to the collection of files to watch.
102 INFO will be written in the warning messages from this FILE. Returns a
103 sequence number which can be used to unregister the file again.
107 me._map[seq] = info, SelLineQueue(file, me._q, seq, 'stderr')
110 def unwatch(me, seq):
111 """Stop watching the file with sequence number SEQ."""
117 Coroutine function: read items from the queue and report them.
119 Unregisters files automatically when they reach EOF.
122 seq, _, line = me._q.get()
126 S.warn(*['connect'] + me._map[seq][0] + ['stderr', line])
130 Coroutine function: wake up every minute and notice changes to the
131 database. When a change happens, tell the Pinger (q.v.) to rescan its
134 cr = T.Coroutine.getcurrent()
136 fw = M.FWatch(opts.cdb)
138 timer = M.SelTimer(time() + 60, lambda: cr.switch())
142 S.notify('connect', 'peerdb-update')
144 class ChildWatch (M.SelSignal):
146 An object which watches for specified processes exiting and reports
147 terminations by writing items of the form (TAG, 'exit', RESULT) to a queue.
149 There is usually only one ChildWatch object, called childwatch.
153 """Initialize the child-watcher."""
154 return M.SelSignal.__new__(cls, SIG.SIGCHLD)
157 """Initialize the child-watcher."""
161 def watch(me, pid, queue, tag):
163 Register PID as a child to watch. If it exits, write (TAG, 'exit', CODE)
164 to the QUEUE, where CODE is one of
166 * None (successful termination)
167 * ['exit-nonzero', CODE] (CODE is a string!)
168 * ['exit-signal', 'S' + CODE] (CODE is the signal number as a string)
169 * ['exit-unknown', STATUS] (STATUS is the entire exit status, in hex)
171 me._pid[pid] = queue, tag
174 def unwatch(me, pid):
175 """Unregister PID as a child to watch."""
182 Called when child processes exit: collect exit statuses and report
187 pid, status = OS.waitpid(-1, OS.WNOHANG)
189 if exc.errno == E.ECHILD:
193 if pid not in me._pid:
195 queue, tag = me._pid[pid]
196 if OS.WIFEXITED(status):
197 exit = OS.WEXITSTATUS(status)
201 code = ['exit-nonzero', str(exit)]
202 elif OS.WIFSIGNALED(status):
203 code = ['exit-signal', 'S' + str(OS.WTERMSIG(status))]
205 code = ['exit-unknown', hex(status)]
206 queue.put((tag, 'exit', code))
208 class Command (object):
210 Represents a running command.
212 This class is the main interface to the machery provided by the ChildWatch
213 and ErrorWatch objects. See also potwatch.
216 def __init__(me, info, queue, tag, args, env):
218 Start a new child process.
220 The ARGS are a list of arguments to be given to the child process. The
221 ENV is either None or a dictionary of environment variable assignments to
222 override the extant environment. INFO is a list of tokens to be included
223 in warnings about the child's stderr output. If the child writes a line
224 to standard output, put (TAG, 'stdout', LINE) to the QUEUE. When the
225 child exits, write (TAG, 'exit', CODE) to the QUEUE.
230 myenv = OS.environ.copy()
231 if env: myenv.update(env)
232 me._proc = PROC.Popen(args = args, env = myenv, bufsize = 1,
233 stdout = PROC.PIPE, stderr = PROC.PIPE)
234 me._lq = SelLineQueue(me._proc.stdout, queue, tag, 'stdout')
235 errorwatch.watch(me._proc.stderr, info)
236 childwatch.watch(me._proc.pid, queue, tag)
240 If I've been forgotten then stop watching for termination.
242 childwatch.unwatch(me._proc.pid)
244 def potwatch(what, name, q):
246 Watch the queue Q for activity as reported by a Command object.
248 Information from the process's stdout is reported as
250 NOTE WHAT NAME stdout LINE
252 abnormal termination is reported as
256 where CODE is what the ChildWatch wrote.
259 while not deadp or not eofp:
260 _, kind, more = q.get()
265 S.notify('connect', what, name, 'stdout', more)
267 if more: S.warn('connect', what, name, *more)
270 ###--------------------------------------------------------------------------
271 ### Peer database utilities.
273 _magic = ['_magic'] # An object distinct from all others
276 """Representation of a peer in the database."""
278 def __init__(me, peer, cdb = None):
280 Create a new peer, named PEER.
282 Information about the peer is read from the database CDB, or the default
283 one given on the command-line.
286 record = (cdb or CDB.init(opts.cdb))['P' + peer]
287 me.__dict__.update(M.URLDecode(record, semip = True))
289 def get(me, key, default = _magic, filter = None):
291 Get the information stashed under KEY from the peer's database record.
293 If DEFAULT is given, then use it if the database doesn't contain the
294 necessary information. If no DEFAULT is given, then report an error. If
295 a FILTER function is given then apply it to the information from the
296 database before returning it.
298 attr = me.__dict__.get(key, default)
300 raise T.TripeJobError('malformed-peer', me.name, 'missing-key', key)
301 elif filter is not None:
307 Return whether the peer's database record has the KEY.
309 return key in me.__dict__
313 Iterate over the available keys in the peer's database record.
315 return me.__dict__.iterkeys()
318 """Parse VALUE as a boolean."""
319 return value in ['t', 'true', 'y', 'yes', 'on']
321 ###--------------------------------------------------------------------------
322 ### Waking up and watching peers.
324 def run_connect(peer, cmd):
326 Start the job of connecting to the passive PEER.
328 The CMD string is a shell command which will connect to the peer (via some
329 back-channel, say ssh and userv), issue a command
331 SVCSUBMIT connect passive [OPTIONS] USER
333 and write the resulting challenge to standard error.
336 cmd = Command(['connect', peer.name], q, 'connect',
337 ['/bin/sh', '-c', cmd], None)
338 _, kind, more = q.peek()
341 S.warn('connect', 'connect', peer.name, 'unexpected-eof')
344 S.greet(peer.name, chal)
346 potwatch('connect', peer.name, q)
348 def run_disconnect(peer, cmd):
350 Start the job of disconnecting from a passive PEER.
352 The CMD string is a shell command which will disconnect from the peer.
355 cmd = Command(['disconnect', peer.name], q, 'disconnect',
356 ['/bin/sh', '-c', cmd], None)
357 potwatch('disconnect', peer.name, q)
360 class PingPeer (object):
362 Object representing a peer which we are pinging to ensure that it is still
365 PingPeer objects are held by the Pinger (q.v.). The Pinger maintains an
366 event queue -- which saves us from having an enormous swarm of coroutines
367 -- but most of the actual work is done here.
369 In order to avoid confusion between different PingPeer instances for the
370 same actual peer, each PingPeer has a sequence number (its `seq'
371 attribute). Events for the PingPeer are identified by a (PEER, SEQ) pair.
372 (Using the PingPeer instance itself will prevent garbage collection of
373 otherwise defunct instances.)
376 def __init__(me, pinger, queue, peer, pingnow):
378 Create a new PingPeer.
380 The PINGER is the Pinger object we should send the results to. This is
381 used when we remove ourselves, if the peer has been explicitly removed.
383 The QUEUE is the event queue on which timer and ping-command events
386 The PEER is a `Peer' object describing the peer.
388 If PINGNOW is true, then immediately start pinging the peer. Otherwise
389 wait until the usual retry interval.
405 me._min = me._max = '-'
410 me._timer = M.SelTimer(time() + me._every, me._time)
412 def update(me, peer):
414 Refreshes the timer parameters for this peer. We don't, however,
415 immediately reschedule anything: that will happen next time anything
418 if peer is None: peer = Peer(me._peer)
419 assert peer.name == me._peer
420 me._every = peer.get('every', filter = T.timespec, default = 120)
421 me._timeout = peer.get('timeout', filter = T.timespec, default = 10)
422 me._retries = peer.get('retries', filter = int, default = 5)
423 me._connectp = peer.has('connect')
428 Send a ping to the peer; the result is sent to the Pinger's event queue.
430 S.rawcommand(T.TripeAsynchronousCommand(
431 me._q, (me._peer, me.seq),
433 '-background', S.bgtag(),
434 '-timeout', str(me._timeout),
439 peer = Peer(me._peer)
441 S.warn('connect', 'reconnecting', me._peer)
443 T.spawn(run_connect, peer, peer.get('connect'))
444 me._timer = M.SelTimer(time() + me._every, me._time)
449 def event(me, code, stuff):
451 Respond to an event which happened to this peer.
453 Timer events indicate that we should start a new ping. (The server has
454 its own timeout which detects lost packets.)
456 We trap unknown-peer responses and detach from the Pinger.
458 If the ping fails and we run out of retries, we attempt to restart the
465 S.notify('connect', 'ping-failed', me._peer, *stuff)
468 elif stuff[0] == 'unknown-peer':
469 me._pinger.kill(me._peer)
470 elif stuff[0] == 'ping-send-failed':
474 if outcome == 'ping-ok' and me._sabotage:
475 outcome = 'ping-timeout'
476 if outcome == 'ping-ok':
478 S.warn('connect', 'ping-ok', me._peer)
480 me._last = '%.1fms' % t
484 if me._min == '-' or t < me._min: me._min = t
485 if me._max == '-' or t > me._max: me._max = t
486 me._timer = M.SelTimer(time() + me._every, me._time)
487 elif outcome == 'ping-timeout':
490 S.warn('connect', 'ping-timeout', me._peer,
491 'attempt', str(me._failures), 'of', str(me._retries))
492 if me._failures < me._retries:
497 me._last = 'reconnect'
498 elif outcome == 'ping-peer-died':
499 me._pinger.kill(me._peer)
502 """Sabotage the peer, for testing purposes."""
504 if me._timer: me._timer.kill()
511 mean = me._sigma_t/me._nping
512 sd = sqrt(me._sigma_t2/me._nping - mean*mean)
513 n = me._nping + me._nlost
514 if not n: pclost = '-'
515 else: pclost = '%d' % ((100*me._nlost + n//2)//n)
516 return { 'last-ping': me._last,
517 'mean-ping': '%.1fms' % mean,
518 'sd-ping': '%.1fms' % sd,
519 'n-ping': '%d' % me._nping,
520 'n-lost': '%d' % me._nlost,
521 'percent-lost': pclost,
522 'min-ping': '%.1fms' % me._min,
523 'max-ping': '%.1fms' % me._max,
524 'state': me._timer and 'idle' or 'check',
525 'failures': me._failures }
530 Handle timer callbacks by posting a timeout event on the queue.
533 me._q.put(((me._peer, me.seq), 'TIMER', None))
536 return 'PingPeer(%s, %d, f = %d)' % (me._peer, me.seq, me._failures)
540 class Pinger (T.Coroutine):
542 The Pinger keeps track of the peers which we expect to be connected and
543 takes action if they seem to stop responding.
545 There is usually only one Pinger, called pinger.
547 The Pinger maintains a collection of PingPeer objects, and an event queue.
548 The PingPeers direct the results of their pings, and timer events, to the
549 event queue. The Pinger's coroutine picks items off the queue and
550 dispatches them back to the PingPeers as appropriate.
554 """Initialize the Pinger."""
555 T.Coroutine.__init__(me)
561 Coroutine function: reads the pinger queue and sends events to the
562 PingPeer objects they correspond to.
565 (peer, seq), code, stuff = me._q.get()
566 if peer in me._peers and seq == me._peers[peer].seq:
567 me._peers[peer].event(code, stuff)
569 def add(me, peer, pingnow):
571 Add PEER to the collection of peers under the Pinger's watchful eye.
572 The arguments are as for PingPeer: see above.
574 me._peers[peer.name] = PingPeer(me, me._q, peer, pingnow)
577 def kill(me, peername):
578 """Remove PEER from the peers being watched by the Pinger."""
579 try: del me._peers[peername]
580 except KeyError: pass
583 def rescan(me, startup):
585 General resynchronization method.
587 We scan the list of peers (with connect scripts) known at the server.
588 Any which are known to the Pinger but aren't known to the server are
589 removed from our list; newly arrived peers are added. (Note that a peer
590 can change state here either due to the server sneakily changing its list
591 without issuing notifications or, more likely, the database changing its
592 idea of whether a peer is interesting.) Finally, PingPeers which are
593 still present are prodded to update their timing parameters.
595 This method is called once at startup to pick up the peers already
596 installed, and again by the dbwatcher coroutine when it detects a change
599 if T._debug: print '# rescan peers'
602 for name in S.list():
603 try: peer = Peer(name)
604 except KeyError: continue
605 if peer.get('watch', filter = boolean, default = False):
606 if T._debug: print '# interesting peer %s' % peer
607 correct[peer.name] = start[peer.name] = peer
609 if T._debug: print '# peer %s ready for adoption' % peer
610 start[peer.name] = peer
611 for name, obj in me._peers.items():
615 if T._debug: print '# peer %s vanished' % name
619 for name, peer in start.iteritems():
620 if name in me._peers: continue
622 if T._debug: print '# setting up peer %s' % name
623 ifname = S.ifname(name)
625 T.defer(adoptpeer, peer, ifname, *addr)
627 if T._debug: print '# adopting new peer %s' % name
633 Returns the list of peers being watched by the Pinger.
635 return me._peers.keys()
638 """Return the PingPeer with the given name."""
639 return me._peers[name]
641 ###--------------------------------------------------------------------------
644 def encode_envvars(env, prefix, vars):
646 Encode the variables in VARS suitably for including in a program
647 environment. Lowercase letters in variable names are forced to uppercase;
648 runs of non-alphanumeric characters are replaced by single underscores; and
649 the PREFIX is prepended. The resulting variables are written to ENV.
651 for k, v in vars.iteritems():
652 env[prefix + r_bad.sub('_', k.upper())] = v
654 r_bad = RX.compile(r'[\W_]+')
657 Translate the database information for a PEER into a dictionary of
658 environment variables with plausible upper-case names and a P_ prefix.
659 Also collect the crypto information into A_ variables.
662 encode_envvars(env, 'P_', dict([(k, peer.get(k)) for k in peer.list()]))
663 encode_envvars(env, 'A_', S.algs(peer.name))
666 def run_ifupdown(what, peer, *args):
668 Run the interface up/down script for a peer.
670 WHAT is 'ifup' or 'ifdown'. PEER names the peer in question. ARGS is a
671 list of arguments to pass to the script, in addition to the peer name.
673 The command is run and watched in the background by potwatch.
676 c = Command([what, peer.name], q, what,
677 M.split(peer.get(what), quotep = True)[0] +
678 [peer.name] + list(args),
680 potwatch(what, peer.name, q)
682 def adoptpeer(peer, ifname, *addr):
684 Add a new peer to our collection.
686 PEER is the `Peer' object; IFNAME is the interface name for its tunnel; and
687 ADDR is the list of tokens representing its address.
689 We try to bring up the interface and provoke a connection to the peer if
693 T.Coroutine(run_ifupdown, name = 'ifup %s' % peer.name) \
694 .switch('ifup', peer, ifname, *addr)
695 cmd = peer.get('connect', default = None)
697 T.Coroutine(run_connect, name = 'connect %s' % peer.name) \
699 if peer.get('watch', filter = boolean, default = False):
700 pinger.add(peer, False)
702 def disownpeer(peer):
703 """Drop the PEER from the Pinger and put its interface to bed."""
704 try: pinger.kill(peer)
705 except KeyError: pass
706 cmd = peer.get('disconnect', default = None)
708 T.Coroutine(run_disconnect, name = 'disconnect %s' % peer.name) \
710 if peer.has('ifdown'):
711 T.Coroutine(run_ifupdown, name = 'ifdown %s' % peer.name) \
712 .switch('ifdown', peer)
714 def addpeer(peer, addr):
716 Process a connect request from a new peer PEER on address ADDR.
718 Any existing peer with this name is disconnected from the server.
720 if peer.name in S.list():
723 booltrue = ['t', 'true', 'y', 'yes', 'on']
725 tunnel = peer.get('tunnel', None),
726 keepalive = peer.get('keepalive', None),
727 key = peer.get('key', None),
728 priv = peer.get('priv', None),
729 mobile = peer.get('mobile', 'nil') in booltrue,
730 cork = peer.get('cork', 'nil') in booltrue,
732 except T.TripeError, exc:
733 raise T.TripeJobError(*exc.args)
735 ## Dictionary mapping challenges to waiting passive-connection coroutines.
738 def notify(_, code, *rest):
740 Watch for notifications.
742 We trap ADD and KILL notifications, and send them straight to adoptpeer and
743 disownpeer respectively; and dispatch GREET notifications to the
744 corresponding waiting coroutine.
747 try: p = Peer(rest[0])
748 except KeyError: return
749 adoptpeer(p, *rest[1:])
751 try: p = Peer(rest[0])
752 except KeyError: return
753 disownpeer(p, *rest[1:])
754 elif code == 'GREET':
756 try: cr = chalmap[chal]
757 except KeyError: pass
758 else: cr.switch(rest[1:])
760 ###--------------------------------------------------------------------------
761 ### Command implementation.
765 kick NAME: Force a new connection attempt for the NAMEd peer.
767 try: pp = pinger.find(name)
768 except KeyError: raise T.TripeJobError('peer-not-adopted', name)
769 try: peer = Peer(name)
770 except KeyError: raise T.TripeJobError('unknown-peer', name)
771 conn = peer.get('connect', None)
772 if conn: T.spawn(run_connect, peer, peer.get('connect'))
773 else: T.spawn(lambda p: S.forcekx(p.name), peer)
777 adopted: Report a list of adopted peers.
779 for name in pinger.adopted():
782 def cmd_active(name):
784 active NAME: Handle an active connection request for the peer called NAME.
786 The appropriate address is read from the database automatically.
788 try: peer = Peer(name)
789 except KeyError: raise T.TripeJobError('unknown-peer', name)
790 addr = peer.get('peer')
791 if addr == 'PASSIVE':
792 raise T.TripeJobError('passive-peer', name)
793 addpeer(peer, M.split(addr, quotep = True)[0])
795 def cmd_listactive():
797 list: Report a list of the available active peers.
799 cdb = CDB.init(opts.cdb)
800 for key in cdb.keys():
801 if key.startswith('P') and Peer(key[1:]).get('peer', '') != 'PASSIVE':
806 info NAME: Report the database entries for the named peer.
808 try: peer = Peer(name)
809 except KeyError: raise T.TripeJobError('unknown-peer', name)
811 try: pp = pinger.find(name)
812 except KeyError: pass
813 else: d.update(pp.info())
814 items = list(peer.list()) + d.keys()
818 except KeyError: v = peer.get(i)
819 T.svcinfo('%s=%s' % (i, v))
821 def cmd_userpeer(user):
823 userpeer USER: Report the peer name for the named user.
825 try: name = CDB.init(opts.cdb)['U' + user]
826 except KeyError: raise T.TripeJobError('unknown-user', user)
829 def cmd_passive(*args):
831 passive [OPTIONS] USER: Await the arrival of the named USER.
833 Report a challenge; when (and if!) the server receives a greeting quoting
834 this challenge, add the corresponding peer to the server.
837 op = T.OptParse(args, ['-timeout'])
839 if opt == '-timeout':
840 timeout = T.timespec(op.arg())
841 user, = op.rest(1, 1)
842 try: name = CDB.init(opts.cdb)['U' + user]
843 except KeyError: raise T.TripeJobError('unknown-user', user)
844 try: peer = Peer(name)
845 except KeyError: raise T.TripeJobError('unknown-peer', name)
847 cr = T.Coroutine.getcurrent()
848 timer = M.SelTimer(time() + timeout, lambda: cr.switch(None))
852 addr = cr.parent.switch()
854 raise T.TripeJobError('connect-timeout')
859 def cmd_sabotage(name):
861 sabotage NAME: Sabotage the NAMEd peer so that we think it can't be pinged.
863 try: pp = pinger.find(name)
864 except KeyError: raise T.TripeJobError('unknown-peer', name)
867 ###--------------------------------------------------------------------------
874 Register the notification watcher, rescan the peers, and add automatic
877 S.handler['NOTE'] = notify
880 pinger.rescan(opts.startup)
883 cdb = CDB.init(opts.cdb)
888 for name in M.split(autos)[0]:
890 peer = Peer(name, cdb)
891 addpeer(peer, M.split(peer.get('peer'), quotep = True)[0])
892 except T.TripeJobError, err:
893 S.warn('connect', 'auto-add-failed', name, *err.args)
897 Initialization to be done before service startup.
899 global errorwatch, childwatch, pinger
900 errorwatch = ErrorWatch()
901 childwatch = ChildWatch()
903 T.Coroutine(dbwatch, name = 'dbwatch').switch()
909 Parse the command-line options.
911 Automatically changes directory to the requested configdir, and turns on
912 debugging. Returns the options object.
914 op = OptionParser(usage = '%prog [-a FILE] [-d DIR]',
915 version = '%%prog %s' % VERSION)
917 op.add_option('-a', '--admin-socket',
918 metavar = 'FILE', dest = 'tripesock', default = T.tripesock,
919 help = 'Select socket to connect to [default %default]')
920 op.add_option('-d', '--directory',
921 metavar = 'DIR', dest = 'dir', default = T.configdir,
922 help = 'Select current diretory [default %default]')
923 op.add_option('-p', '--peerdb',
924 metavar = 'FILE', dest = 'cdb', default = T.peerdb,
925 help = 'Select peers database [default %default]')
926 op.add_option('--daemon', dest = 'daemon',
927 default = False, action = 'store_true',
928 help = 'Become a daemon after successful initialization')
929 op.add_option('--debug', dest = 'debug',
930 default = False, action = 'store_true',
931 help = 'Emit debugging trace information')
932 op.add_option('--startup', dest = 'startup',
933 default = False, action = 'store_true',
934 help = 'Being called as part of the server startup')
936 opts, args = op.parse_args()
937 if args: op.error('no arguments permitted')
939 T._debug = opts.debug
942 ## Service table, for running manually.
943 service_info = [('connect', T.VERSION, {
944 'adopted': (0, 0, '', cmd_adopted),
945 'kick': (1, 1, 'PEER', cmd_kick),
946 'passive': (1, None, '[OPTIONS] USER', cmd_passive),
947 'active': (1, 1, 'PEER', cmd_active),
948 'info': (1, 1, 'PEER', cmd_info),
949 'list-active': (0, 0, '', cmd_listactive),
950 'userpeer': (1, 1, 'USER', cmd_userpeer),
951 'sabotage': (1, 1, 'PEER', cmd_sabotage)
954 if __name__ == '__main__':
955 opts = parse_options()
956 T.runservices(opts.tripesock, service_info,
957 init = init, setup = setup,
958 daemon = opts.daemon)
960 ###----- That's all, folks --------------------------------------------------