svc/connect.in: Add a new `sabotage' command to test ping-failure actions.
[tripe] / svc / conntrack.in
1 #! @PYTHON@
2 ### -*-python-*-
3 ###
4 ### Service for automatically tracking network connection status
5 ###
6 ### (c) 2010 Straylight/Edgeware
7 ###
8
9 ###----- Licensing notice ---------------------------------------------------
10 ###
11 ### This file is part of Trivial IP Encryption (TrIPE).
12 ###
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.
17 ###
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
21 ### for more details.
22 ###
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/>.
25
26 VERSION = '@VERSION@'
27
28 ###--------------------------------------------------------------------------
29 ### External dependencies.
30
31 from ConfigParser import RawConfigParser
32 from optparse import OptionParser
33 import os as OS
34 import sys as SYS
35 import socket as S
36 import mLib as M
37 import tripe as T
38 import dbus as D
39 for i in ['mainloop', 'mainloop.glib']:
40 __import__('dbus.%s' % i)
41 try: from gi.repository import GLib as G
42 except ImportError: import gobject as G
43 from struct import pack, unpack
44
45 SM = T.svcmgr
46 ##__import__('rmcr').__debug = True
47
48 ###--------------------------------------------------------------------------
49 ### Utilities.
50
51 class struct (object):
52 """A simple container object."""
53 def __init__(me, **kw):
54 me.__dict__.update(kw)
55
56 def toposort(cmp, things):
57 """
58 Generate the THINGS in an order consistent with a given partial order.
59
60 The function CMP(X, Y) should return true if X must precede Y, and false if
61 it doesn't care. If X and Y are equal then it should return false.
62
63 The THINGS may be any finite iterable; it is converted to a list
64 internally.
65 """
66
67 ## Make sure we can index the THINGS, and prepare an ordering table.
68 ## What's going on? The THINGS might not have a helpful equality
69 ## predicate, so it's easier to work with indices. The ordering table will
70 ## remember which THINGS (by index) are considered greater than other
71 ## things.
72 things = list(things)
73 n = len(things)
74 order = [{} for i in xrange(n)]
75 rorder = [{} for i in xrange(n)]
76 for i in xrange(n):
77 for j in xrange(n):
78 if i != j and cmp(things[i], things[j]):
79 order[j][i] = True
80 rorder[i][j] = True
81
82 ## Now we can do the sort.
83 out = []
84 while True:
85 done = True
86 for i in xrange(n):
87 if order[i] is not None:
88 done = False
89 if len(order[i]) == 0:
90 for j in rorder[i]:
91 del order[j][i]
92 yield things[i]
93 order[i] = None
94 if done:
95 break
96
97 ###--------------------------------------------------------------------------
98 ### Parse the configuration file.
99
100 ## Hmm. Should I try to integrate this with the peers database? It's not a
101 ## good fit; it'd need special hacks in tripe-newpeers. And the use case for
102 ## this service are largely going to be satellite notes, I don't think
103 ## scalability's going to be a problem.
104
105 class Config (object):
106 """
107 Represents a configuration file.
108
109 The most interesting thing is probably the `groups' slot, which stores a
110 list of pairs (NAME, PATTERNS); the NAME is a string, and the PATTERNS a
111 list of (TAG, PEER, ADDR, MASK) triples. The implication is that there
112 should be precisely one peer with a name matching NAME-*, and that it
113 should be NAME-TAG, where (TAG, PEER, ADDR, MASK) is the first triple such
114 that the host's primary IP address (if PEER is None -- or the IP address it
115 would use for communicating with PEER) is within the network defined by
116 ADDR/MASK.
117 """
118
119 def __init__(me, file):
120 """
121 Construct a new Config object, reading the given FILE.
122 """
123 me._file = file
124 me._fwatch = M.FWatch(file)
125 me._update()
126
127 def check(me):
128 """
129 See whether the configuration file has been updated.
130 """
131 if me._fwatch.update():
132 me._update()
133
134 def _update(me):
135 """
136 Internal function to update the configuration from the underlying file.
137 """
138
139 ## Read the configuration. We have no need of the fancy substitutions,
140 ## so turn them all off.
141 cp = RawConfigParser()
142 cp.read(me._file)
143 if T._debug: print '# reread config'
144
145 ## Save the test address. Make sure it's vaguely sensible. The default
146 ## is probably good for most cases, in fact, since that address isn't
147 ## actually in use. Note that we never send packets to the test address;
148 ## we just use it to discover routing information.
149 if cp.has_option('DEFAULT', 'test-addr'):
150 testaddr = cp.get('DEFAULT', 'test-addr')
151 S.inet_aton(testaddr)
152 else:
153 testaddr = '1.2.3.4'
154
155 ## Scan the configuration file and build the groups structure.
156 groups = []
157 for sec in cp.sections():
158 pats = []
159 for tag in cp.options(sec):
160 spec = cp.get(sec, tag).split()
161
162 ## Parse the entry into peer and network.
163 if len(spec) == 1:
164 peer = None
165 net = spec[0]
166 else:
167 peer, net = spec
168
169 ## Syntax of a net is ADDRESS/MASK, where ADDRESS is a dotted-quad,
170 ## and MASK is either a dotted-quad or a single integer N indicating
171 ## a mask with N leading ones followed by trailing zeroes.
172 slash = net.index('/')
173 addr, = unpack('>L', S.inet_aton(net[:slash]))
174 if net.find('.', slash + 1) >= 0:
175 mask, = unpack('>L', S.inet_aton(net[:slash]))
176 else:
177 n = int(net[slash + 1:], 10)
178 mask = (1 << 32) - (1 << 32 - n)
179 pats.append((tag, peer, addr & mask, mask))
180
181 ## Annoyingly, RawConfigParser doesn't preserve the order of options.
182 ## In order to make things vaguely sane, we topologically sort the
183 ## patterns so that more specific patterns are checked first.
184 pats = list(toposort(lambda (t, p, a, m), (tt, pp, aa, mm): \
185 (p and not pp) or \
186 (p == pp and m == (m | mm) and aa == (a & mm)),
187 pats))
188 groups.append((sec, pats))
189
190 ## Done.
191 me.testaddr = testaddr
192 me.groups = groups
193
194 ### This will be a configuration file.
195 CF = None
196
197 def straddr(a): return a is None and '#<none>' or S.inet_ntoa(pack('>L', a))
198 def strmask(m):
199 for i in xrange(33):
200 if m == 0xffffffff ^ ((1 << (32 - i)) - 1): return i
201 return straddr(m)
202
203 def cmd_showconfig():
204 T.svcinfo('test-addr=%s' % CF.testaddr)
205 def cmd_showgroups():
206 for sec, pats in CF.groups:
207 T.svcinfo(sec)
208 def cmd_showgroup(g):
209 for s, p in CF.groups:
210 if s == g:
211 pats = p
212 break
213 else:
214 raise T.TripeJobError, 'unknown-group', g
215 for t, p, a, m in pats:
216 T.svcinfo('peer', t,
217 'target', p or '(default)',
218 'net', '%s/%s' % (straddr(a), strmask(m)))
219
220 ###--------------------------------------------------------------------------
221 ### Responding to a network up/down event.
222
223 def localaddr(peer):
224 """
225 Return the local IP address used for talking to PEER.
226 """
227 sk = S.socket(S.AF_INET, S.SOCK_DGRAM)
228 try:
229 try:
230 sk.connect((peer, 1))
231 addr, _ = sk.getsockname()
232 addr, = unpack('>L', S.inet_aton(addr))
233 return addr
234 except S.error:
235 return None
236 finally:
237 sk.close()
238
239 _kick = T.Queue()
240 def kickpeers():
241 while True:
242 upness, reason = _kick.get()
243 if T._debug: print '# kickpeers %s: %s' % (upness, reason)
244 select = []
245
246 ## Make sure the configuration file is up-to-date. Don't worry if we
247 ## can't do anything useful.
248 try:
249 CF.check()
250 except Exception, exc:
251 SM.warn('conntrack', 'config-file-error',
252 exc.__class__.__name__, str(exc))
253
254 ## Find the current list of peers.
255 peers = SM.list()
256
257 ## Work out the primary IP address.
258 if upness:
259 addr = localaddr(CF.testaddr)
260 if addr is None:
261 upness = False
262 else:
263 addr = None
264 if not T._debug: pass
265 elif addr: print '# local address = %s' % straddr(addr)
266 else: print '# offline'
267
268 ## Now decide what to do.
269 changes = []
270 for g, pp in CF.groups:
271 if T._debug: print '# check group %s' % g
272
273 ## Find out which peer in the group ought to be active.
274 ip = None
275 map = {}
276 want = None
277 for t, p, a, m in pp:
278 if p is None or not upness:
279 ipq = addr
280 else:
281 ipq = localaddr(p)
282 if T._debug:
283 info = 'peer=%s; target=%s; net=%s/%s; local=%s' % (
284 t, p or '(default)', straddr(a), strmask(m), straddr(ipq))
285 if upness and ip is None and \
286 ipq is not None and (ipq & m) == a:
287 if T._debug: print '# %s: SELECTED' % info
288 map[t] = 'up'
289 select.append('%s=%s' % (g, t))
290 if t == 'down' or t.startswith('down/'):
291 want = None
292 else:
293 want = t
294 ip = ipq
295 else:
296 map[t] = 'down'
297 if T._debug: print '# %s: skipped' % info
298
299 ## Shut down the wrong ones.
300 found = False
301 if T._debug: print '# peer-map = %r' % map
302 for p in peers:
303 what = map.get(p, 'leave')
304 if what == 'up':
305 found = True
306 if T._debug: print '# peer %s: already up' % p
307 elif what == 'down':
308 def _(p = p):
309 try:
310 SM.kill(p)
311 except T.TripeError, exc:
312 if exc.args[0] == 'unknown-peer':
313 ## Inherently racy; don't worry about this.
314 pass
315 else:
316 raise
317 if T._debug: print '# peer %s: bring down' % p
318 changes.append(_)
319
320 ## Start the right one if necessary.
321 if want is not None and not found:
322 def _(want = want):
323 try:
324 SM.svcsubmit('connect', 'active', want)
325 except T.TripeError, exc:
326 SM.warn('conntrack', 'connect-failed', want, *exc.args)
327 if T._debug: print '# peer %s: bring up' % want
328 changes.append(_)
329
330 ## Commit the changes.
331 if changes:
332 SM.notify('conntrack', upness and 'up' or 'down', *select + reason)
333 for c in changes: c()
334
335 def netupdown(upness, reason):
336 """
337 Add or kill peers according to whether the network is up or down.
338
339 UPNESS is true if the network is up, or false if it's down.
340 """
341
342 _kick.put((upness, reason))
343
344 ###--------------------------------------------------------------------------
345 ### NetworkManager monitor.
346
347 NM_NAME = 'org.freedesktop.NetworkManager'
348 NM_PATH = '/org/freedesktop/NetworkManager'
349 NM_IFACE = NM_NAME
350 NMCA_IFACE = NM_NAME + '.Connection.Active'
351
352 NM_STATE_CONNECTED = 3 #obsolete
353 NM_STATE_CONNECTED_LOCAL = 50
354 NM_STATE_CONNECTED_SITE = 60
355 NM_STATE_CONNECTED_GLOBAL = 70
356 NM_CONNSTATES = set([NM_STATE_CONNECTED,
357 NM_STATE_CONNECTED_LOCAL,
358 NM_STATE_CONNECTED_SITE,
359 NM_STATE_CONNECTED_GLOBAL])
360
361 class NetworkManagerMonitor (object):
362 """
363 Watch NetworkManager signals for changes in network state.
364 """
365
366 ## Strategy. There are two kinds of interesting state transitions for us.
367 ## The first one is the global are-we-connected state, which we'll use to
368 ## toggle network upness on a global level. The second is which connection
369 ## has the default route, which we'll use to tweak which peer in the peer
370 ## group is active. The former is most easily tracked using the signal
371 ## org.freedesktop.NetworkManager.StateChanged; for the latter, we track
372 ## org.freedesktop.NetworkManager.Connection.Active.PropertiesChanged and
373 ## look for when a new connection gains the default route.
374
375 def attach(me, bus):
376 try:
377 nm = bus.get_object(NM_NAME, NM_PATH)
378 state = nm.Get(NM_IFACE, 'State')
379 if state in NM_CONNSTATES:
380 netupdown(True, ['nm', 'initially-connected'])
381 else:
382 netupdown(False, ['nm', 'initially-disconnected'])
383 except D.DBusException:
384 pass
385 bus.add_signal_receiver(me._nm_state, 'StateChanged',
386 NM_IFACE, NM_NAME, NM_PATH)
387 bus.add_signal_receiver(me._nm_connchange, 'PropertiesChanged',
388 NMCA_IFACE, NM_NAME, None)
389
390 def _nm_state(me, state):
391 if state in NM_CONNSTATES:
392 netupdown(True, ['nm', 'connected'])
393 else:
394 netupdown(False, ['nm', 'disconnected'])
395
396 def _nm_connchange(me, props):
397 if props.get('Default', False):
398 netupdown(True, ['nm', 'default-connection-change'])
399
400 ##--------------------------------------------------------------------------
401 ### Connman monitor.
402
403 CM_NAME = 'net.connman'
404 CM_PATH = '/'
405 CM_IFACE = 'net.connman.Manager'
406
407 class ConnManMonitor (object):
408 """
409 Watch ConnMan signls for changes in network state.
410 """
411
412 ## Strategy. Everything seems to be usefully encoded in the `State'
413 ## property. If it's `offline', `idle' or `ready' then we don't expect a
414 ## network connection. During handover from one network to another, the
415 ## property passes through `ready' to `online'.
416
417 def attach(me, bus):
418 try:
419 cm = bus.get_object(CM_NAME, CM_PATH)
420 props = cm.GetProperties(dbus_interface = CM_IFACE)
421 state = props['State']
422 netupdown(state == 'online', ['connman', 'initially-%s' % state])
423 except D.DBusException:
424 pass
425 bus.add_signal_receiver(me._cm_state, 'PropertyChanged',
426 CM_IFACE, CM_NAME, CM_PATH)
427
428 def _cm_state(me, prop, value):
429 if prop != 'State': return
430 netupdown(value == 'online', ['connman', value])
431
432 ###--------------------------------------------------------------------------
433 ### Maemo monitor.
434
435 ICD_NAME = 'com.nokia.icd'
436 ICD_PATH = '/com/nokia/icd'
437 ICD_IFACE = ICD_NAME
438
439 class MaemoICdMonitor (object):
440 """
441 Watch ICd signals for changes in network state.
442 """
443
444 ## Strategy. ICd only handles one connection at a time in steady state,
445 ## though when switching between connections, it tries to bring the new one
446 ## up before shutting down the old one. This makes life a bit easier than
447 ## it is with NetworkManager. On the other hand, the notifications are
448 ## relative to particular connections only, and the indicator that the old
449 ## connection is down (`IDLE') comes /after/ the new one comes up
450 ## (`CONNECTED'), so we have to remember which one is active.
451
452 def attach(me, bus):
453 try:
454 icd = bus.get_object(ICD_NAME, ICD_PATH)
455 try:
456 iap = icd.get_ipinfo(dbus_interface = ICD_IFACE)[0]
457 me._iap = iap
458 netupdown(True, ['icd', 'initially-connected', iap])
459 except D.DBusException:
460 me._iap = None
461 netupdown(False, ['icd', 'initially-disconnected'])
462 except D.DBusException:
463 me._iap = None
464 bus.add_signal_receiver(me._icd_state, 'status_changed', ICD_IFACE,
465 ICD_NAME, ICD_PATH)
466
467 def _icd_state(me, iap, ty, state, hunoz):
468 if state == 'CONNECTED':
469 me._iap = iap
470 netupdown(True, ['icd', 'connected', iap])
471 elif state == 'IDLE' and iap == me._iap:
472 me._iap = None
473 netupdown(False, ['icd', 'idle'])
474
475 ###--------------------------------------------------------------------------
476 ### D-Bus connection tracking.
477
478 class DBusMonitor (object):
479 """
480 Maintains a connection to the system D-Bus, and watches for signals.
481
482 If the connection is initially down, or drops for some reason, we retry
483 periodically (every five seconds at the moment). If the connection
484 resurfaces, we reattach the monitors.
485 """
486
487 def __init__(me):
488 """
489 Initialise the object and try to establish a connection to the bus.
490 """
491 me._mons = []
492 me._loop = D.mainloop.glib.DBusGMainLoop()
493 me._state = 'startup'
494 me._reconnect()
495
496 def addmon(me, mon):
497 """
498 Add a monitor object to watch for signals.
499
500 MON.attach(BUS) is called, with BUS being the connection to the system
501 bus. MON should query its service's current status and watch for
502 relevant signals.
503 """
504 me._mons.append(mon)
505 if me._bus is not None:
506 mon.attach(me._bus)
507
508 def _reconnect(me, hunoz = None):
509 """
510 Start connecting to the bus.
511
512 If we fail the first time, retry periodically.
513 """
514 if me._state == 'startup':
515 T.aside(SM.notify, 'conntrack', 'dbus-connection', 'startup')
516 elif me._state == 'connected':
517 T.aside(SM.notify, 'conntrack', 'dbus-connection', 'lost')
518 else:
519 T.aside(SM.notify, 'conntrack', 'dbus-connection',
520 'state=%s' % me._state)
521 me._state == 'reconnecting'
522 me._bus = None
523 if me._try_connect():
524 G.timeout_add_seconds(5, me._try_connect)
525
526 def _try_connect(me):
527 """
528 Actually make a connection attempt.
529
530 If we succeed, attach the monitors.
531 """
532 try:
533 addr = OS.getenv('TRIPE_CONNTRACK_BUS')
534 if addr == 'SESSION':
535 bus = D.SessionBus(mainloop = me._loop, private = True)
536 elif addr is not None:
537 bus = D.bus.BusConnection(addr, mainloop = me._loop)
538 else:
539 bus = D.SystemBus(mainloop = me._loop, private = True)
540 for m in me._mons:
541 m.attach(bus)
542 except D.DBusException, e:
543 return True
544 me._bus = bus
545 me._state = 'connected'
546 bus.call_on_disconnection(me._reconnect)
547 T.aside(SM.notify, 'conntrack', 'dbus-connection', 'connected')
548 return False
549
550 ###--------------------------------------------------------------------------
551 ### TrIPE service.
552
553 class GIOWatcher (object):
554 """
555 Monitor I/O events using glib.
556 """
557 def __init__(me, conn, mc = G.main_context_default()):
558 me._conn = conn
559 me._watch = None
560 me._mc = mc
561 def connected(me, sock):
562 me._watch = G.io_add_watch(sock, G.IO_IN,
563 lambda *hunoz: me._conn.receive())
564 def disconnected(me):
565 G.source_remove(me._watch)
566 me._watch = None
567 def iterate(me):
568 me._mc.iteration(True)
569
570 SM.iowatch = GIOWatcher(SM)
571
572 def init():
573 """
574 Service initialization.
575
576 Add the D-Bus monitor here, because we might send commands off immediately,
577 and we want to make sure the server connection is up.
578 """
579 global DBM
580 T.Coroutine(kickpeers, name = 'kickpeers').switch()
581 DBM = DBusMonitor()
582 DBM.addmon(NetworkManagerMonitor())
583 DBM.addmon(ConnManMonitor())
584 DBM.addmon(MaemoICdMonitor())
585 G.timeout_add_seconds(30, lambda: (netupdown(True, ['interval-timer'])
586 or True))
587
588 def parse_options():
589 """
590 Parse the command-line options.
591
592 Automatically changes directory to the requested configdir, and turns on
593 debugging. Returns the options object.
594 """
595 op = OptionParser(usage = '%prog [-a FILE] [-d DIR]',
596 version = '%%prog %s' % VERSION)
597
598 op.add_option('-a', '--admin-socket',
599 metavar = 'FILE', dest = 'tripesock', default = T.tripesock,
600 help = 'Select socket to connect to [default %default]')
601 op.add_option('-d', '--directory',
602 metavar = 'DIR', dest = 'dir', default = T.configdir,
603 help = 'Select current diretory [default %default]')
604 op.add_option('-c', '--config',
605 metavar = 'FILE', dest = 'conf', default = 'conntrack.conf',
606 help = 'Select configuration [default %default]')
607 op.add_option('--daemon', dest = 'daemon',
608 default = False, action = 'store_true',
609 help = 'Become a daemon after successful initialization')
610 op.add_option('--debug', dest = 'debug',
611 default = False, action = 'store_true',
612 help = 'Emit debugging trace information')
613 op.add_option('--startup', dest = 'startup',
614 default = False, action = 'store_true',
615 help = 'Being called as part of the server startup')
616
617 opts, args = op.parse_args()
618 if args: op.error('no arguments permitted')
619 OS.chdir(opts.dir)
620 T._debug = opts.debug
621 return opts
622
623 ## Service table, for running manually.
624 def cmd_updown(upness):
625 return lambda *args: T.defer(netupdown, upness, ['manual'] + list(args))
626 service_info = [('conntrack', VERSION, {
627 'up': (0, None, '', cmd_updown(True)),
628 'down': (0, None, '', cmd_updown(False)),
629 'show-config': (0, 0, '', cmd_showconfig),
630 'show-groups': (0, 0, '', cmd_showgroups),
631 'show-group': (1, 1, 'GROUP', cmd_showgroup)
632 })]
633
634 if __name__ == '__main__':
635 opts = parse_options()
636 CF = Config(opts.conf)
637 T.runservices(opts.tripesock, service_info,
638 init = init, daemon = opts.daemon)
639
640 ###----- That's all, folks --------------------------------------------------