public void runJob() {
boolean wasFF = _facade.floodfillEnabled();
boolean ff = shouldBeFloodfill();
_facade.setFloodfillEnabled(ff);
if (ff != wasFF) {
if (ff) {
getContext().router().eventLog().addEvent(EventLog.BECAME_FLOODFILL);
} else {
getContext().router().eventLog().addEvent(EventLog.NOT_FLOODFILL);
}
getContext().router().rebuildRouterInfo(true);
Job routerInfoFlood = new FloodfillRouterInfoFloodJob(getContext(), _facade);
if (getContext().router().getUptime() < 5 * 60 * 1000) {
// Needed to prevent race if router.floodfillParticipant=true (not auto)
routerInfoFlood.getTiming().setStartAfter(getContext().clock().now() + 5 * 60 * 1000);
getContext().jobQueue().addJob(routerInfoFlood);
if (_log.shouldLog(Log.DEBUG)) {
_log.logAlways(
Log.DEBUG, "Deferring our FloodfillRouterInfoFloodJob run because of low uptime.");
}
} else {
routerInfoFlood.runJob();
if (_log.shouldLog(Log.DEBUG)) {
_log.logAlways(Log.DEBUG, "Running FloodfillRouterInfoFloodJob");
}
}
}
if (_log.shouldLog(Log.INFO)) _log.info("Should we be floodfill? " + ff);
int delay = (REQUEUE_DELAY / 2) + getContext().random().nextInt(REQUEUE_DELAY);
// there's a lot of eligible non-floodfills, keep them from all jumping in at once
// TODO: somehow assess the size of the network to make this adaptive?
if (!ff) delay *= 4; // this was 7, reduced for moar FFs --zab
requeue(delay);
}
/**
* Allow the choice as to whether failed searches should count against the peer (such as if we
* search for a random key)
*/
public FailedJob(RouterContext enclosingContext, RouterInfo peer, boolean penalizePeer) {
super(enclosingContext);
_penalizePeer = penalizePeer;
_peer = peer.getIdentity().getHash();
_sentOn = enclosingContext.clock().now();
_isFloodfill = FloodfillNetworkDatabaseFacade.isFloodfill(peer);
}
/**
* we're (probably) searching for a LeaseSet, so to be (overly) cautious, we're sending the
* request out through a tunnel w/ reply back through another tunnel.
*/
protected void sendLeaseSearch(RouterInfo router) {
Hash to = router.getIdentity().getHash();
TunnelInfo inTunnel = getContext().tunnelManager().selectInboundExploratoryTunnel(to);
if (inTunnel == null) {
_log.warn("No tunnels to get search replies through! wtf!");
getContext().jobQueue().addJob(new FailedJob(getContext(), router));
return;
}
TunnelId inTunnelId = inTunnel.getReceiveTunnelId(0);
// this will fail if we've shitlisted our inbound gateway, but the gw may not necessarily
// be shitlisted by whomever needs to contact them, so we don't need to check this
// RouterInfo inGateway = getContext().netDb().lookupRouterInfoLocally(inTunnel.getPeer(0));
// if (inGateway == null) {
// _log.error("We can't find the gateway to our inbound tunnel?! wtf");
// getContext().jobQueue().addJob(new FailedJob(getContext(), router));
// return;
// }
int timeout = getPerPeerTimeoutMs(to);
long expiration = getContext().clock().now() + timeout;
DatabaseLookupMessage msg = buildMessage(inTunnelId, inTunnel.getPeer(0), expiration);
TunnelInfo outTunnel = getContext().tunnelManager().selectOutboundExploratoryTunnel(to);
if (outTunnel == null) {
_log.warn("No tunnels to send search out through! wtf!");
getContext().jobQueue().addJob(new FailedJob(getContext(), router));
return;
}
TunnelId outTunnelId = outTunnel.getSendTunnelId(0);
if (_log.shouldLog(Log.DEBUG))
_log.debug(
getJobId()
+ ": Sending search to "
+ to
+ " for "
+ msg.getSearchKey().toBase64()
+ " w/ replies through ["
+ msg.getFrom().toBase64()
+ "] via tunnel ["
+ msg.getReplyTunnel()
+ "]");
SearchMessageSelector sel =
new SearchMessageSelector(getContext(), router, _expiration, _state);
SearchUpdateReplyFoundJob reply =
new SearchUpdateReplyFoundJob(
getContext(), router, _state, _facade, this, outTunnel, inTunnel);
if (FloodfillNetworkDatabaseFacade.isFloodfill(router)) _floodfillSearchesOutstanding++;
getContext()
.messageRegistry()
.registerPending(sel, reply, new FailedJob(getContext(), router), timeout);
getContext().tunnelDispatcher().dispatchOutbound(msg, outTunnelId, to);
}
/**
* this is now misnamed, as it is only used to determine whether to return floodfill peers only
*/
static boolean onlyQueryFloodfillPeers(RouterContext ctx) {
// if (isCongested(ctx))
// return true;
// If we are floodfill, we want the FloodfillPeerSelector (in add()) to include
// non-ff peers (if required) in DatabaseSearchReplyMessage responses
// so that Exploration works.
// ExploreJob is disabled if we are floodfill.
// The other two places this was called (one below and one in FNDF)
// have been commented out.
// Returning false essentially enables kademlia as a backup to floodfill for search responses.
if (FloodfillNetworkDatabaseFacade.floodfillEnabled(ctx)) return false;
return Boolean.valueOf(ctx.getProperty("netDb.floodfillOnly", DEFAULT_FLOODFILL_ONLY + ""))
.booleanValue();
}
/**
* Send a series of searches to the next available peers as selected by the routing table, but
* making sure no more than SEARCH_BREDTH are outstanding at any time
*/
protected void continueSearch() {
if (_state.completed()) {
if (_log.shouldLog(Log.DEBUG))
_log.debug(getJobId() + ": Search already completed", new Exception("already completed"));
return;
}
int toCheck = getBredth() - _state.getPending().size();
if (toCheck <= 0) {
// too many already pending
if (_log.shouldLog(Log.INFO))
_log.info(
getJobId()
+ ": Too many searches already pending (pending: "
+ _state.getPending().size()
+ " max: "
+ getBredth()
+ ")");
requeuePending();
return;
}
int sent = 0;
Set attempted = _state.getAttempted();
while (sent <= 0) {
// boolean onlyFloodfill = onlyQueryFloodfillPeers(getContext());
boolean onlyFloodfill = true;
if (_floodfillPeersExhausted && onlyFloodfill && _state.getPending().isEmpty()) {
if (_log.shouldLog(Log.WARN))
_log.warn(
getJobId()
+ ": no non-floodfill peers left, and no more pending. Searched: "
+ _state.getAttempted().size()
+ " failed: "
+ _state.getFailed().size());
fail();
return;
}
List closestHashes = getClosestRouters(_state.getTarget(), toCheck, attempted);
if ((closestHashes == null) || (closestHashes.isEmpty())) {
if (_state.getPending().isEmpty()) {
// we tried to find some peers, but there weren't any and no one else is going to answer
if (_log.shouldLog(Log.INFO))
_log.info(
getJobId()
+ ": No peers left, and none pending! Already searched: "
+ _state.getAttempted().size()
+ " failed: "
+ _state.getFailed().size());
fail();
} else {
// no more to try, but we might get data or close peers from some outstanding requests
if (_log.shouldLog(Log.INFO))
_log.info(
getJobId()
+ ": No peers left, but some are pending! Pending: "
+ _state.getPending().size()
+ " attempted: "
+ _state.getAttempted().size()
+ " failed: "
+ _state.getFailed().size());
requeuePending();
}
return;
} else {
attempted.addAll(closestHashes);
for (Iterator iter = closestHashes.iterator(); iter.hasNext(); ) {
Hash peer = (Hash) iter.next();
DatabaseEntry ds = _facade.getDataStore().get(peer);
if (ds == null) {
if (_log.shouldLog(Log.INFO))
_log.info(
"Next closest peer "
+ peer
+ " was only recently referred to us, sending a search for them");
getContext().netDb().lookupRouterInfo(peer, null, null, _timeoutMs);
} else if (!(ds.getType() == DatabaseEntry.KEY_TYPE_ROUTERINFO)) {
if (_log.shouldLog(Log.WARN))
_log.warn(
getJobId()
+ ": Error selecting closest hash that wasnt a router! "
+ peer
+ " : "
+ ds.getClass().getName());
_state.replyTimeout(peer);
} else {
RouterInfo ri = (RouterInfo) ds;
if (!FloodfillNetworkDatabaseFacade.isFloodfill(ri)) {
_floodfillPeersExhausted = true;
if (onlyFloodfill) continue;
}
if (ri.isHidden()) { // || // allow querying shitlisted, since its indirect
// getContext().shitlist().isShitlisted(peer)) {
// dont bother
} else {
_state.addPending(peer);
sendSearch((RouterInfo) ds);
sent++;
}
}
}
/*
if (sent <= 0) {
// the (potentially) last peers being searched for could not be,
// er, searched for, so lets retry ASAP (causing either another
// peer to be selected, or the whole search to fail)
if (_log.shouldLog(Log.INFO))
_log.info(getJobId() + ": No new peer queued up, so we are going to requeue " +
"ourselves in our search for " + _state.getTarget().toBase64());
requeuePending(0);
}
*/
}
}
}
private boolean shouldBeFloodfill() {
if (!SigType.ECDSA_SHA256_P256.isAvailable()) return false;
// Hidden trumps netDb.floodfillParticipant=true
if (getContext().router().isHidden()) return false;
String enabled = getContext().getProperty(PROP_FLOODFILL_PARTICIPANT, "auto");
if ("true".equals(enabled)) return true;
if ("false".equals(enabled)) return false;
// auto from here down
// Only if not shutting down...
if (getContext().router().gracefulShutdownInProgress()) return false;
// ARM ElG decrypt is too slow
if (SystemVersion.isARM() || SystemVersion.isAndroid()) return false;
if (getContext().getBooleanProperty(UDPTransport.PROP_LAPTOP_MODE)) return false;
if (getContext().commSystem().isInBadCountry()) return false;
String country = getContext().commSystem().getOurCountry();
// anonymous proxy, satellite provider (not in bad country list)
if ("a1".equals(country) || "a2".equals(country)) return false;
// Only if up a while...
if (getContext().router().getUptime() < MIN_UPTIME) return false;
RouterInfo ri = getContext().router().getRouterInfo();
if (ri == null) return false;
char bw = ri.getBandwidthTier().charAt(0);
// Only if class M, N, O, P, X
if (bw != Router.CAPABILITY_BW64
&& bw != Router.CAPABILITY_BW128
&& bw != Router.CAPABILITY_BW256
&& bw != Router.CAPABILITY_BW512
&& bw != Router.CAPABILITY_BW_UNLIMITED) return false;
// This list will not include ourselves...
List<Hash> floodfillPeers = _facade.getFloodfillPeers();
long now = getContext().clock().now();
// We know none at all! Must be our turn...
if (floodfillPeers == null || floodfillPeers.isEmpty()) {
_lastChanged = now;
return true;
}
// Only change status every so often
boolean wasFF = _facade.floodfillEnabled();
if (_lastChanged + MIN_CHANGE_DELAY > now) return wasFF;
// This is similar to the qualification we do in FloodOnlySearchJob.runJob().
// Count the "good" ff peers.
//
// Who's not good?
// the unheard-from, unprofiled, failing, unreachable and banlisted ones.
// We should hear from floodfills pretty frequently so set a 60m time limit.
// If unprofiled we haven't talked to them in a long time.
// We aren't contacting the peer directly, so banlist doesn't strictly matter,
// but it's a bad sign, and we often banlist a peer before we fail it...
//
// Future: use Integration calculation
//
int ffcount = floodfillPeers.size();
int failcount = 0;
long before = now - 60 * 60 * 1000;
for (Hash peer : floodfillPeers) {
PeerProfile profile = getContext().profileOrganizer().getProfile(peer);
if (profile == null
|| profile.getLastHeardFrom() < before
|| profile.getIsFailing()
|| getContext().banlist().isBanlisted(peer)
|| getContext().commSystem().wasUnreachable(peer)) failcount++;
}
if (wasFF) ffcount++;
int good = ffcount - failcount;
boolean happy = getContext().router().getRouterInfo().getCapabilities().indexOf('R') >= 0;
// TODO - limit may still be too high
// For reference, the avg lifetime job lag on my Pi is 6.
// Should we consider avg. dropped ff jobs?
RateStat lagStat = getContext().statManager().getRate("jobQueue.jobLag");
RateStat queueStat = getContext().statManager().getRate("router.tunnelBacklog");
happy = happy && lagStat.getRate(60 * 60 * 1000L).getAvgOrLifetimeAvg() < 25;
happy = happy && queueStat.getRate(60 * 60 * 1000L).getAvgOrLifetimeAvg() < 5;
// Only if we're pretty well integrated...
happy = happy && _facade.getKnownRouters() >= 400;
happy = happy && getContext().commSystem().countActivePeers() >= 50;
happy = happy && getContext().tunnelManager().getParticipatingCount() >= 25;
happy = happy && Math.abs(getContext().clock().getOffset()) < 10 * 1000;
// We need an address and no introducers
if (happy) {
RouterAddress ra = getContext().router().getRouterInfo().getTargetAddress("SSU");
if (ra == null) happy = false;
else {
if (ra.getOption("ihost0") != null) happy = false;
}
}
double elG = 0;
RateStat stat = getContext().statManager().getRate("crypto.elGamal.decrypt");
if (stat != null) {
Rate rate = stat.getRate(60 * 60 * 1000L);
if (rate != null) {
elG = rate.getAvgOrLifetimeAvg();
happy = happy && elG <= 40.0d;
}
}
if (_log.shouldLog(Log.DEBUG)) {
final RouterContext rc = getContext();
final String log =
String.format(
"FF criteria breakdown: happy=%b, capabilities=%s, maxLag=%d, known=%d, "
+ "active=%d, participating=%d, offset=%d, ssuAddr=%s ElG=%f",
happy,
rc.router().getRouterInfo().getCapabilities(),
rc.jobQueue().getMaxLag(),
_facade.getKnownRouters(),
rc.commSystem().countActivePeers(),
rc.tunnelManager().getParticipatingCount(),
Math.abs(rc.clock().getOffset()),
rc.router().getRouterInfo().getTargetAddress("SSU").toString(),
elG);
_log.debug(log);
}
// Too few, and we're reachable, let's volunteer
if (good < MIN_FF && happy) {
if (!wasFF) {
_lastChanged = now;
_log.logAlways(
Log.INFO,
"Only " + good + " ff peers and we want " + MIN_FF + " so we are becoming floodfill");
}
return true;
}
// Too many, or we aren't reachable, let's stop
if (good > MAX_FF || (good > MIN_FF && !happy)) {
if (wasFF) {
_lastChanged = now;
_log.logAlways(
Log.INFO,
"Have "
+ good
+ " ff peers and we need only "
+ MIN_FF
+ " to "
+ MAX_FF
+ " so we are disabling floodfill; reachable? "
+ happy);
}
return false;
}
if (_log.shouldLog(Log.INFO))
_log.info(
"Have " + good + " ff peers, not changing, enabled? " + wasFF + "; reachable? " + happy);
return wasFF;
}