4 ### Watch arrival and departure of peers
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
14 ### it under the terms of the GNU General Public License as published by
15 ### the Free Software Foundation; either version 2 of the License, or
16 ### (at your option) any later version.
18 ### TrIPE is distributed in the hope that it will be useful,
19 ### but WITHOUT ANY WARRANTY; without even the implied warranty of
20 ### MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 ### GNU General Public License for more details.
23 ### You should have received a copy of the GNU General Public License
24 ### along with TrIPE; if not, write to the Free Software Foundation,
25 ### Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
29 ###--------------------------------------------------------------------------
30 ### External dependencies.
32 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 watch 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(*['watch'] + me._map[seq][0] + ['stderr', line])
130 Coroutine function: wake up every second 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() + 1, lambda: cr.switch())
142 S.notify('watch', '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('watch', what, name, 'stdout', more)
267 if more: S.warn('watch', what, name, *more)
270 ###--------------------------------------------------------------------------
271 ### Peer database utilities.
273 def timespec(info, key, default):
274 """Parse INFO[KEY] as a timespec, or return DEFAULT."""
276 return T.timespec(info[key])
277 except (KeyError, T.TripeJobError):
280 def integer(info, key, default):
281 """Parse INFO[KEY] as an integer, or return DEFAULT."""
283 return int(info[key])
284 except (KeyError, ValueError):
287 def boolean(info, key, default):
288 """Parse INFO[KEY] as a boolean, or return DEFAULT."""
290 return info[key] in ['t', 'true', 'y', 'yes', 'on']
291 except (KeyError, ValueError):
296 Return a dictionary containing information about PEER from the database.
298 return dict(M.URLDecode(CDB.init(opts.cdb)['P' + peer], semip = True))
300 ###--------------------------------------------------------------------------
301 ### Waking up and watching peers.
303 def connect(peer, conn = None):
305 Start the job of connecting to the passive PEER.
307 The CONN string is a shell command which will connect to the peer (via some
308 back-channel, say ssh and userv), issue a command
310 SVCSUBMIT connect passive [OPTIONS] USER
312 and write the resulting challenge to standard error.
316 conn = peerinfo(peer)['connect']
320 cmd = Command(['connect', peer], q, 'connect',
321 ['/bin/sh', '-c', conn], None)
322 _, kind, more = q.peek()
325 S.warn('watch', 'connect', peer, 'unexpected-eof')
330 potwatch('connect', peer, q)
333 class PingPeer (object):
335 Object representing a peer which we are pinging to ensure that it is still
338 PingPeer objects are held by the Pinger (q.v.). The Pinger maintains an
339 event queue -- which saves us from having an enormous swarm of coroutines
340 -- but most of the actual work is done here.
342 In order to avoid confusion between different PingPeer instances for the
343 same actual peer, each PingPeer has a sequence number (its `seq'
344 attribute). Events for the PingPeer are identified by a (PEER, SEQ) pair.
345 (Using the PingPeer instance itself will prevent garbage collection of
346 otherwise defunct instances.)
349 def __init__(me, pinger, queue, peer, info, pingnow):
351 Create a new PingPeer.
353 The PINGER is the Pinger object we should send the results to. This is
354 used when we remove ourselves, if the peer has been explicitly removed.
356 The QUEUE is the event queue on which timer and ping-command events
359 The PEER is just the peer's name, as a string.
361 The INFO is the database record for the peer, as a dictionary, or None if
362 it's not readily available. (This is just a tweak to save multiple
363 probes if we don't really need them.)
365 If PINGNOW is true, then immediately start pinging the peer. Otherwise
366 wait until the usual retry interval.
380 me._timer = M.SelTimer(time() + me._every, me._time)
382 def update(me, info):
384 Refreshes the timer parameters for this peer. We don't, however,
385 immediately reschedule anything: that will happen next time anything
389 info = peerinfo(me._peer)
390 me._every = timespec(info, 'every', 120)
391 me._timeout = timespec(info, 'timeout', 10)
392 me._retries = integer(info, 'retries', 5)
393 me._connectp = 'connect' in info
398 Send a ping to the peer; the result is sent to the Pinger's event queue.
400 S.rawcommand(T.TripeAsynchronousCommand(
401 me._q, (me._peer, me.seq),
403 '-background', S.bgtag(),
404 '-timeout', str(me._timeout),
409 info = peerinfo(me._peer)
410 if 'connect' in info:
411 S.warn('watch', 'reconnecting', me._peer)
413 T.spawn(connect, me._peer)
414 me._timer = M.SelTimer(time() + me._every, me._time)
418 def event(me, code, stuff):
420 Respond to an event which happened to this peer.
422 Timer events indicate that we should start a new ping. (The server has
423 its own timeout which detects lost packets.)
425 We trap unknown-peer responses and detach from the Pinger.
427 If the ping fails and we run out of retries, we attempt to restart the
434 S.notify('watch', 'ping-failed', me._peer, *stuff)
437 elif stuff[0] == 'unknown-peer':
438 me._pinger.kill(me._peer)
439 elif stuff[0] == 'ping-send-failed':
442 if stuff[0] == 'ping-ok':
444 S.warn('watch', 'ping-ok', me._peer)
445 me._timer = M.SelTimer(time() + me._every, me._time)
446 elif stuff[0] == 'ping-timeout':
448 S.warn('watch', 'ping-timeout', me._peer,
449 'attempt', str(me._failures), 'of', str(me._retries))
450 if me._failures < me._retries:
454 elif stuff[0] == 'ping-peer-died':
455 me._pinger.kill(me._peer)
460 Handle timer callbacks by posting a timeout event on the queue.
463 me._q.put(((me._peer, me.seq), 'TIMER', None))
466 return 'PingPeer(%s, %d, f = %d)' % (me._peer, me.seq, me._failures)
470 class Pinger (T.Coroutine):
472 The Pinger keeps track of the peers which we expect to be connected and
473 takes action if they seem to stop responding.
475 There is usually only one Pinger, called pinger.
477 The Pinger maintains a collection of PingPeer objects, and an event queue.
478 The PingPeers direct the results of their pings, and timer events, to the
479 event queue. The Pinger's coroutine picks items off the queue and
480 dispatches them back to the PingPeers as appropriate.
484 """Initialize the Pinger."""
485 T.Coroutine.__init__(me)
491 Coroutine function: reads the pinger queue and sends events to the
492 PingPeer objects they correspond to.
495 (peer, seq), code, stuff = me._q.get()
496 if peer in me._peers and seq == me._peers[peer].seq:
497 me._peers[peer].event(code, stuff)
499 def add(me, peer, info, pingnow):
501 Add PEER to the collection of peers under the Pinger's watchful eye.
502 The arguments are as for PingPeer: see above.
504 me._peers[peer] = PingPeer(me, me._q, peer, info, pingnow)
508 """Remove PEER from the peers being watched by the Pinger."""
512 def rescan(me, startup):
514 General resynchronization method.
516 We scan the list of peers (with connect scripts) known at the server.
517 Any which are known to the Pinger but aren't known to the server are
518 removed from our list; newly arrived peers are added. (Note that a peer
519 can change state here either due to the server sneakily changing its list
520 without issuing notifications or, more likely, the database changing its
521 idea of whether a peer is interesting.) Finally, PingPeers which are
522 still present are prodded to update their timing parameters.
524 This method is called once at startup to pick up the peers already
525 installed, and again by the dbwatcher coroutine when it detects a change
528 if T._debug: print '# rescan peers'
530 for peer in S.list():
532 info = peerinfo(peer)
535 if boolean(info, 'watch', False):
536 if T._debug: print '# interesting peer %s' % peer
538 for peer, obj in me._peers.items():
540 obj.update(correct[peer])
542 if T._debug: print '# peer %s vanished' % peer
544 for peer, info in correct.iteritems():
545 if peer not in me._peers:
547 if T._debug: print '# setting up peer %s' % peer
548 ifname = S.ifname(peer)
550 T.defer(addpeer, info, peer, ifname, *addr)
552 if T._debug: print '# adopting new peer %s' % peer
553 me.add(peer, info, True)
558 Returns the list of peers being watched by the Pinger.
560 return me._peers.keys()
562 ###--------------------------------------------------------------------------
565 def encode_envvars(env, prefix, vars):
567 Encode the variables in VARS suitably for including in a program
568 environment. Lowercase letters in variable names are forced to uppercase;
569 runs of non-alphanumeric characters are replaced by single underscores; and
570 the PREFIX is prepended. The resulting variables are written to ENV.
572 for k, v in vars.iteritems():
573 env[prefix + r_bad.sub('_', k.upper())] = v
575 r_bad = RX.compile(r'[\W_]+')
578 Translate the database INFO dictionary for a peer into a dictionary of
579 environment variables with plausible upper-case names and a P_ prefix.
580 Also collect the crypto information into A_ variables.
583 encode_envvars(env, 'P_', info)
584 encode_envvars(env, 'A_', S.algs())
587 def ifupdown(what, peer, info, *args):
589 Run the interface up/down script for a peer.
591 WHAT is 'ifup' or 'ifdown'. PEER names the peer in question. INFO is the
592 database record dictionary. ARGS is a list of arguments to pass to the
593 script, in addition to the peer name.
595 The command is run and watched in the background by potwatch.
598 c = Command([what, peer], q, what,
599 M.split(info[what], quotep = True)[0] +
602 potwatch(what, peer, q)
604 def addpeer(info, peer, ifname, *addr):
606 Add a new peer to our collection.
608 INFO is the peer information dictionary, or None if we don't have one yet.
610 PEER names the peer; IFNAME is the interface name for its tunnel; and ADDR
611 is the list of tokens representing its address.
613 We try to bring up the interface and provoke a connection to the peer if
618 info = peerinfo(peer)
622 T.Coroutine(ifupdown, name = 'ifup %s' % peer) \
623 .switch('ifup', peer, info, ifname, *addr)
624 if 'connect' in info:
625 T.Coroutine(connect, name = 'connect %s' % peer) \
626 .switch(peer, info['connect'])
627 if boolean(info, 'watch', False):
628 pinger.add(peer, info, False)
631 """Drop the PEER from the Pinger and put its interface to bed."""
633 info = peerinfo(peer)
641 T.Coroutine(ifupdown, name = 'ifdown %s' % peer) \
642 .switch('ifdown', peer, info)
644 def notify(_, code, *rest):
646 Watch for notifications.
648 We trap ADD and KILL notifications, and send them straight to addpeer and
649 delpeer respectively.
656 ###--------------------------------------------------------------------------
660 raise T.TripeJobError('not-implemented')
664 kick PEER: Force a new connection attempt for PEER
666 if peer not in pinger.adopted():
667 raise T.TripeJobError('peer-not-adopted', peer)
668 T.spawn(connect, peer)
672 adopted: Report a list of adopted peers.
674 for peer in pinger.adopted():
677 ###--------------------------------------------------------------------------
684 Register the notification watcher, and rescan the peers.
686 S.handler['NOTE'] = notify
688 pinger.rescan(opts.startup)
692 Initialization to be done before service startup.
694 global errorwatch, childwatch, pinger
695 errorwatch = ErrorWatch()
696 childwatch = ChildWatch()
698 T.Coroutine(dbwatch, name = 'dbwatch').switch()
704 Parse the command-line options.
706 Automatically changes directory to the requested configdir, and turns on
707 debugging. Returns the options object.
709 op = OptionParser(usage = '%prog [-a FILE] [-d DIR]',
710 version = '%%prog %s' % VERSION)
712 op.add_option('-a', '--admin-socket',
713 metavar = 'FILE', dest = 'tripesock', default = T.tripesock,
714 help = 'Select socket to connect to [default %default]')
715 op.add_option('-d', '--directory',
716 metavar = 'DIR', dest = 'dir', default = T.configdir,
717 help = 'Select current diretory [default %default]')
718 op.add_option('-p', '--peerdb',
719 metavar = 'FILE', dest = 'cdb', default = T.peerdb,
720 help = 'Select peers database [default %default]')
721 op.add_option('--daemon', dest = 'daemon',
722 default = False, action = 'store_true',
723 help = 'Become a daemon after successful initialization')
724 op.add_option('--debug', dest = 'debug',
725 default = False, action = 'store_true',
726 help = 'Emit debugging trace information')
727 op.add_option('--startup', dest = 'startup',
728 default = False, action = 'store_true',
729 help = 'Being called as part of the server startup')
731 opts, args = op.parse_args()
732 if args: op.error('no arguments permitted')
734 T._debug = opts.debug
737 ## Service table, for running manually.
738 service_info = [('watch', T.VERSION, {
739 'adopted': (0, 0, '', cmd_adopted),
740 'kick': (1, 1, 'PEER', cmd_kick)
743 if __name__ == '__main__':
744 opts = parse_options()
745 T.runservices(opts.tripesock, service_info,
746 init = init, setup = setup,
747 daemon = opts.daemon)
749 ###----- That's all, folks --------------------------------------------------