1 package edu.uci.iotproject.detection.layer2;
3 import edu.uci.iotproject.analysis.TriggerTrafficExtractor;
4 import edu.uci.iotproject.analysis.UserAction;
5 import edu.uci.iotproject.detection.AbstractClusterMatcher;
6 import edu.uci.iotproject.detection.ClusterMatcherObserver;
7 import edu.uci.iotproject.detection.SignatureDetectorObserver;
8 import edu.uci.iotproject.io.PcapHandleReader;
9 import edu.uci.iotproject.io.PrintWriterUtils;
10 import edu.uci.iotproject.trafficreassembly.layer2.Layer2Flow;
11 import edu.uci.iotproject.trafficreassembly.layer2.Layer2FlowReassembler;
12 import edu.uci.iotproject.util.PcapPacketUtils;
13 import edu.uci.iotproject.util.PrintUtils;
14 import org.jgrapht.GraphPath;
15 import org.jgrapht.alg.shortestpath.DijkstraShortestPath;
16 import org.jgrapht.graph.DefaultWeightedEdge;
17 import org.jgrapht.graph.SimpleDirectedWeightedGraph;
18 import org.pcap4j.core.*;
21 import java.io.FileWriter;
22 import java.io.IOException;
23 import java.io.PrintWriter;
24 import java.time.Duration;
26 import java.util.function.Function;
27 import java.util.regex.Pattern;
30 * Performs layer 2 signature detection.
32 * @author Janus Varmarken {@literal <jvarmark@uci.edu>}
33 * @author Rahmadi Trimananda {@literal <rtrimana@uci.edu>}
35 public class Layer2SignatureDetector implements PacketListener, ClusterMatcherObserver {
38 * If set to {@code true}, output written to the results file is also dumped to standard out.
40 private static boolean DUPLICATE_OUTPUT_TO_STD_OUT = true;
42 private static List<Function<Layer2Flow, Boolean>> parseSignatureMacFilters(String filtersString) {
43 List<Function<Layer2Flow, Boolean>> filters = new ArrayList<>();
44 String[] filterRegexes = filtersString.split(";");
45 for (String filterRegex : filterRegexes) {
46 final Pattern regex = Pattern.compile(filterRegex);
47 // Create a filter that includes all flows where one of the two MAC addresses match the regex.
48 filters.add(flow -> regex.matcher(flow.getEndpoint1().toString()).matches() || regex.matcher(flow.getEndpoint2().toString()).matches());
53 public static void main(String[] args) throws PcapNativeException, NotOpenException, IOException {
54 // Parse required parameters.
55 if (args.length < 8) {
56 String errMsg = String.format("SPECTO version 1.0\n" +
57 "Copyright (C) 2018-2019 Janus Varmarken and Rahmadi Trimananda.\n" +
58 "University of California, Irvine.\n" +
59 "All rights reserved.\n\n" +
60 "Usage: %s inputPcapFile onAnalysisFile offAnalysisFile onSignatureFile offSignatureFile resultsFile" +
61 "\n inputPcapFile: the target of the detection" +
62 "\n onAnalysisFile: the file that contains the ON clusters analysis" +
63 "\n offAnalysisFile: the file that contains the OFF clusters analysis" +
64 "\n onSignatureFile: the file that contains the ON signature to search for" +
65 "\n offSignatureFile: the file that contains the OFF signature to search for" +
66 "\n resultsFile: where to write the results of the detection" +
67 "\n signatureDuration: the maximum duration of signature detection",
68 Layer2SignatureDetector.class.getSimpleName());
69 System.out.println(errMsg);
70 String optParamsExplained = "Above are the required, positional arguments. In addition to these, the " +
71 "following options and associated positional arguments may be used:\n" +
72 " '-onmacfilters <regex>;<regex>;...;<regex>' which specifies that sequence matching should ONLY" +
73 " be performed on flows where the MAC of one of the two endpoints matches the given regex. Note " +
74 "that you MUST specify a regex for each cluster of the signature. This is to facilitate more " +
75 "aggressive filtering on parts of the signature (e.g., the communication that involves the " +
76 "smart home device itself as one can drop all flows that do not include an endpoint with a MAC " +
77 "that matches the vendor's prefix).\n" +
78 " '-offmacfilters <regex>;<regex>;...;<regex>' works exactly the same as onmacfilters, but " +
79 "applies to the OFF signature instead of the ON signature.\n" +
80 " '-sout <boolean literal>' true/false literal indicating if output should also be printed to std out; default is true.";
81 System.out.println(optParamsExplained);
84 // TODO: We could take 7 inputs if we decided to use the cluster analyses.
85 final String pcapFile = args[0];
86 final String onClusterAnalysisFile = args[1];
87 final String offClusterAnalysisFile = args[2];
88 final String onSignatureFile = args[3];
89 final String offSignatureFile = args[4];
90 final String resultsFile = args[5];
91 final int signatureDuration = Integer.parseInt(args[6]);
92 final double eps = Double.parseDouble(args[7]);
94 // Parse optional parameters.
95 List<Function<Layer2Flow, Boolean>> onSignatureMacFilters = null, offSignatureMacFilters = null;
96 final int optParamsStartIdx = 7;
97 if (args.length > optParamsStartIdx) {
98 for (int i = optParamsStartIdx; i < args.length; i++) {
99 if (args[i].equalsIgnoreCase("-onMacFilters")) {
100 // Next argument is the cluster-wise MAC filters (separated by semicolons).
101 onSignatureMacFilters = parseSignatureMacFilters(args[i+1]);
102 } else if (args[i].equalsIgnoreCase("-offMacFilters")) {
103 // Next argument is the cluster-wise MAC filters (separated by semicolons).
104 offSignatureMacFilters = parseSignatureMacFilters(args[i+1]);
105 } else if (args[i].equalsIgnoreCase("-sout")) {
106 // Next argument is a boolean true/false literal.
107 DUPLICATE_OUTPUT_TO_STD_OUT = Boolean.parseBoolean(args[i+1]);
112 // Prepare file outputter.
113 File outputFile = new File(resultsFile);
114 outputFile.getParentFile().mkdirs();
115 final PrintWriter resultsWriter = new PrintWriter(new FileWriter(outputFile));
116 // Include metadata as comments at the top
117 PrintWriterUtils.println("# Detection results for:", resultsWriter, DUPLICATE_OUTPUT_TO_STD_OUT);
118 PrintWriterUtils.println("# - inputPcapFile: " + pcapFile, resultsWriter, DUPLICATE_OUTPUT_TO_STD_OUT);
119 PrintWriterUtils.println("# - onAnalysisFile: " + onClusterAnalysisFile, resultsWriter, DUPLICATE_OUTPUT_TO_STD_OUT);
120 PrintWriterUtils.println("# - offAnalysisFile: " + offClusterAnalysisFile, resultsWriter, DUPLICATE_OUTPUT_TO_STD_OUT);
121 PrintWriterUtils.println("# - onSignatureFile: " + onSignatureFile, resultsWriter, DUPLICATE_OUTPUT_TO_STD_OUT);
122 PrintWriterUtils.println("# - offSignatureFile: " + offSignatureFile, resultsWriter, DUPLICATE_OUTPUT_TO_STD_OUT);
123 resultsWriter.flush();
125 // Create signature detectors and add observers that output their detected events.
126 List<List<List<PcapPacket>>> onSignature = PrintUtils.deserializeFromFile(onSignatureFile);
127 List<List<List<PcapPacket>>> offSignature = PrintUtils.deserializeFromFile(offSignatureFile);
128 // Load signature analyses
129 List<List<List<PcapPacket>>> onClusterAnalysis = PrintUtils.deserializeFromFile(onClusterAnalysisFile);
130 List<List<List<PcapPacket>>> offClusterAnalysis = PrintUtils.deserializeFromFile(offClusterAnalysisFile);
131 // TODO: FOR NOW WE DECIDE PER SIGNATURE AND THEN WE OR THE BOOLEANS
132 // TODO: SINCE WE ONLY HAVE 2 SIGNATURES FOR NOW (ON AND OFF), THEN IT IS USUALLY EITHER RANGE-BASED OR
133 // TODO: STRICT MATCHING
134 // Check if we should use range-based matching
135 boolean isRangeBasedForOn = PcapPacketUtils.isRangeBasedMatching(onSignature, eps, offSignature);
136 boolean isRangeBasedForOff = PcapPacketUtils.isRangeBasedMatching(offSignature, eps, onSignature);
137 // TODO: WE DON'T DO RANGE-BASED FOR NOW BECAUSE THE RESULTS ARE TERRIBLE FOR LAYER 2 MATCHING
138 // TODO: THIS WOULD ONLY WORK FOR SIGNATURES LONGER THAN 2 PACKETS
139 // boolean isRangeBasedForOn = false;
140 // boolean isRangeBasedForOff = false;
141 // Update the signature with ranges if it is range-based
142 if (isRangeBasedForOn) {
143 onSignature = PcapPacketUtils.useRangeBasedMatching(onSignature, onClusterAnalysis);
145 if (isRangeBasedForOff) {
146 offSignature = PcapPacketUtils.useRangeBasedMatching(offSignature, offClusterAnalysis);
148 Layer2SignatureDetector onDetector = onSignatureMacFilters == null ?
149 new Layer2SignatureDetector(onSignature, signatureDuration, isRangeBasedForOn, eps) :
150 new Layer2SignatureDetector(onSignature, onSignatureMacFilters, signatureDuration, isRangeBasedForOn, eps);
151 Layer2SignatureDetector offDetector = offSignatureMacFilters == null ?
152 new Layer2SignatureDetector(offSignature, signatureDuration, isRangeBasedForOff, eps) :
153 new Layer2SignatureDetector(offSignature, offSignatureMacFilters, signatureDuration, isRangeBasedForOff, eps);
154 final List<UserAction> detectedEvents = new ArrayList<>();
155 onDetector.addObserver((signature, match) -> {
156 UserAction event = new UserAction(UserAction.Type.TOGGLE_ON, match.get(0).get(0).getTimestamp());
157 PrintWriterUtils.println(event, resultsWriter, DUPLICATE_OUTPUT_TO_STD_OUT);
158 detectedEvents.add(event);
160 offDetector.addObserver((signature, match) -> {
161 UserAction event = new UserAction(UserAction.Type.TOGGLE_OFF, match.get(0).get(0).getTimestamp());
162 PrintWriterUtils.println(event, resultsWriter, DUPLICATE_OUTPUT_TO_STD_OUT);
163 detectedEvents.add(event);
166 // Load the PCAP file
169 handle = Pcaps.openOffline(pcapFile, PcapHandle.TimestampPrecision.NANO);
170 } catch (PcapNativeException pne) {
171 handle = Pcaps.openOffline(pcapFile);
173 PcapHandleReader reader = new PcapHandleReader(handle, p -> true, onDetector, offDetector);
175 reader.readFromHandle();
177 String resultOn = "# Number of detected events of type " + UserAction.Type.TOGGLE_ON + ": " +
178 detectedEvents.stream().filter(ua -> ua.getType() == UserAction.Type.TOGGLE_ON).count();
179 String resultOff = "# Number of detected events of type " + UserAction.Type.TOGGLE_OFF + ": " +
180 detectedEvents.stream().filter(ua -> ua.getType() == UserAction.Type.TOGGLE_OFF).count();
181 String onMaxSkippedPackets = "# Number of skipped packets in ON signature " +
182 Integer.toString(onDetector.getMaxSkippedPackets());
183 String offMaxSkippedPackets = "# Number of skipped packets in OFF signature " +
184 Integer.toString(offDetector.getMaxSkippedPackets());
185 PrintWriterUtils.println(resultOn, resultsWriter, DUPLICATE_OUTPUT_TO_STD_OUT);
186 PrintWriterUtils.println(resultOff, resultsWriter, DUPLICATE_OUTPUT_TO_STD_OUT);
187 PrintWriterUtils.println(onMaxSkippedPackets, resultsWriter, DUPLICATE_OUTPUT_TO_STD_OUT);
188 PrintWriterUtils.println(offMaxSkippedPackets, resultsWriter, DUPLICATE_OUTPUT_TO_STD_OUT);
190 // Flush output to results file and close it.
191 resultsWriter.flush();
192 resultsWriter.close();
196 * The signature that this {@link Layer2SignatureDetector} is searching for.
198 private final List<List<List<PcapPacket>>> mSignature;
201 * The {@link Layer2ClusterMatcher}s in charge of detecting each individual sequence of packets that together make
202 * up the the signature.
204 private final List<Layer2ClusterMatcher> mClusterMatchers;
207 * For each {@code i} ({@code i >= 0 && i < mPendingMatches.length}), {@code mPendingMatches[i]} holds the matches
208 * found by the {@link Layer2ClusterMatcher} at {@code mClusterMatchers.get(i)} that have yet to be "consumed",
209 * i.e., have yet to be included in a signature detected by this {@link Layer2SignatureDetector} (a signature can
210 * be encompassed of multiple packet sequences occurring shortly after one another on multiple connections).
212 private final List<List<PcapPacket>>[] mPendingMatches;
215 * Maps a {@link Layer2ClusterMatcher} to its corresponding index in {@link #mPendingMatches}.
217 private final Map<Layer2ClusterMatcher, Integer> mClusterMatcherIds;
220 * In charge of reassembling layer 2 packet flows.
222 private final Layer2FlowReassembler mFlowReassembler = new Layer2FlowReassembler();
224 private final List<SignatureDetectorObserver> mObservers = new ArrayList<>();
226 private int mInclusionTimeMillis;
228 private int mMaxSkippedPackets;
230 public Layer2SignatureDetector(List<List<List<PcapPacket>>> searchedSignature, int signatureDuration, boolean isRangeBased, double eps) {
231 this(searchedSignature, null, signatureDuration, isRangeBased, eps);
234 public Layer2SignatureDetector(List<List<List<PcapPacket>>> searchedSignature, List<Function<Layer2Flow,
235 Boolean>> flowFilters, int inclusionTimeMillis, boolean isRangeBased, double eps) {
236 if (flowFilters != null && flowFilters.size() != searchedSignature.size()) {
237 throw new IllegalArgumentException("If flow filters are used, there must be a flow filter for each cluster " +
238 "of the signature.");
240 mSignature = Collections.unmodifiableList(searchedSignature);
241 List<Layer2ClusterMatcher> clusterMatchers = new ArrayList<>();
242 for (int i = 0; i < mSignature.size(); i++) {
243 List<List<PcapPacket>> cluster = mSignature.get(i);
244 Layer2ClusterMatcher clusterMatcher = flowFilters == null ?
245 new Layer2ClusterMatcher(cluster, inclusionTimeMillis, isRangeBased, eps) :
246 new Layer2ClusterMatcher(cluster, flowFilters.get(i), inclusionTimeMillis, isRangeBased, eps);
247 clusterMatcher.addObserver(this);
248 clusterMatchers.add(clusterMatcher);
250 mClusterMatchers = Collections.unmodifiableList(clusterMatchers);
251 mPendingMatches = new List[mClusterMatchers.size()];
252 for (int i = 0; i < mPendingMatches.length; i++) {
253 mPendingMatches[i] = new ArrayList<>();
255 Map<Layer2ClusterMatcher, Integer> clusterMatcherIds = new HashMap<>();
256 for (int i = 0; i < mClusterMatchers.size(); i++) {
257 clusterMatcherIds.put(mClusterMatchers.get(i), i);
259 mClusterMatcherIds = Collections.unmodifiableMap(clusterMatcherIds);
260 // Register all cluster matchers to receive a notification whenever a new flow is encountered.
261 mClusterMatchers.forEach(cm -> mFlowReassembler.addObserver(cm));
262 mInclusionTimeMillis =
263 inclusionTimeMillis == 0 ? TriggerTrafficExtractor.INCLUSION_WINDOW_MILLIS : inclusionTimeMillis;
264 mMaxSkippedPackets = 0;
267 public int getMaxSkippedPackets() {
268 return mMaxSkippedPackets;
272 public void gotPacket(PcapPacket packet) {
273 // Forward packet processing to the flow reassembler that in turn notifies the cluster matchers as appropriate
274 mFlowReassembler.gotPacket(packet);
278 public void onMatch(AbstractClusterMatcher clusterMatcher, List<PcapPacket> match, int maxSkippedPackets) {
279 // Update the number of skipped packets
280 if (mMaxSkippedPackets < maxSkippedPackets) {
281 mMaxSkippedPackets = maxSkippedPackets;
283 // TODO: a cluster matcher found a match
284 if (clusterMatcher instanceof Layer2ClusterMatcher) {
285 // Add the match at the corresponding index
286 mPendingMatches[mClusterMatcherIds.get(clusterMatcher)].add(match);
287 checkSignatureMatch();
291 public void addObserver(SignatureDetectorObserver observer) {
292 mObservers.add(observer);
295 public boolean removeObserver(SignatureDetectorObserver observer) {
296 return mObservers.remove(observer);
300 @SuppressWarnings("Duplicates")
301 private void checkSignatureMatch() {
302 // << Graph-based approach using Balint's idea. >>
303 // This implementation assumes that the packets in the inner lists (the sequences) are ordered by asc timestamp.
305 // There cannot be a signature match until each Layer3ClusterMatcher has found a match of its respective sequence.
306 if (Arrays.stream(mPendingMatches).noneMatch(l -> l.isEmpty())) {
308 final SimpleDirectedWeightedGraph<Vertex, DefaultWeightedEdge> graph =
309 new SimpleDirectedWeightedGraph<>(DefaultWeightedEdge.class);
310 // Add a vertex for each match found by all cluster matchers.
311 // And maintain an array to keep track of what cluster matcher each vertex corresponds to
312 final List<Vertex>[] vertices = new List[mPendingMatches.length];
313 for (int i = 0; i < mPendingMatches.length; i++) {
314 vertices[i] = new ArrayList<>();
315 for (List<PcapPacket> sequence : mPendingMatches[i]) {
316 Vertex v = new Vertex(sequence);
317 vertices[i].add(v); // retain reference for later when we are to add edges
318 graph.addVertex(v); // add to vertex to graph
321 // Add dummy source and sink vertices to facilitate search.
322 final Vertex source = new Vertex(null);
323 final Vertex sink = new Vertex(null);
324 graph.addVertex(source);
325 graph.addVertex(sink);
326 // The source is connected to all vertices that wrap the sequences detected by cluster matcher at index 0.
327 // Note: zero cost edges as this is just a dummy link to facilitate search from a common start node.
328 for (Vertex v : vertices[0]) {
329 DefaultWeightedEdge edge = graph.addEdge(source, v);
330 graph.setEdgeWeight(edge, 0.0);
332 // Similarly, all vertices that wrap the sequences detected by the last cluster matcher of the signature
333 // are connected to the sink node.
334 for (Vertex v : vertices[vertices.length-1]) {
335 DefaultWeightedEdge edge = graph.addEdge(v, sink);
336 graph.setEdgeWeight(edge, 0.0);
338 // Now link sequences detected by the cluster matcher at index i to sequences detected by the cluster
339 // matcher at index i+1 if they obey the timestamp constraint (i.e., that the latter is later in time than
341 for (int i = 0; i < vertices.length; i++) {
343 if (j < vertices.length) {
344 for (Vertex iv : vertices[i]) {
345 PcapPacket ivLast = iv.sequence.get(iv.sequence.size()-1);
346 for (Vertex jv : vertices[j]) {
347 PcapPacket jvFirst = jv.sequence.get(jv.sequence.size()-1);
348 if (ivLast.getTimestamp().isBefore(jvFirst.getTimestamp())) {
349 DefaultWeightedEdge edge = graph.addEdge(iv, jv);
350 // The weight is the duration of the i'th sequence plus the duration between the i'th
351 // and i+1'th sequence.
352 Duration d = Duration.
353 between(iv.sequence.get(0).getTimestamp(), jvFirst.getTimestamp());
354 // Unfortunately weights are double values, so must convert from long to double.
355 // TODO: need nano second precision? If so, use d.toNanos().
356 // TODO: risk of overflow when converting from long to double..?
357 graph.setEdgeWeight(edge, Long.valueOf(d.toMillis()).doubleValue());
359 // Alternative version if we cannot assume that sequences are ordered by timestamp:
360 // if (iv.sequence.stream().max(Comparator.comparing(PcapPacket::getTimestamp)).get()
361 // .getTimestamp().isBefore(jv.sequence.stream().min(
362 // Comparator.comparing(PcapPacket::getTimestamp)).get().getTimestamp())) {
369 // Graph construction complete, run shortest-path to find a (potential) signature match.
370 DijkstraShortestPath<Vertex, DefaultWeightedEdge> dijkstra = new DijkstraShortestPath<>(graph);
371 GraphPath<Vertex, DefaultWeightedEdge> shortestPath = dijkstra.getPath(source, sink);
372 if (shortestPath != null) {
373 // The total weight is the duration between the first packet of the first sequence and the last packet
374 // of the last sequence, so we simply have to compare the weight against the timeframe that we allow
375 // the signature to span. For now we just use the inclusion window we defined for training purposes.
376 // Note however, that we must convert back from double to long as the weight is stored as a double in
378 if (((long)shortestPath.getWeight()) < mInclusionTimeMillis) {
379 // There's a signature match!
380 // Extract the match from the vertices
381 List<List<PcapPacket>> signatureMatch = new ArrayList<>();
382 for(Vertex v : shortestPath.getVertexList()) {
383 if (v == source || v == sink) {
384 // Skip the dummy source and sink nodes.
387 signatureMatch.add(v.sequence);
388 // As there is a one-to-one correspondence between vertices[] and pendingMatches[], we know that
389 // the sequence we've "consumed" for index i of the matched signature is also at index i in
390 // pendingMatches. We must remove it from pendingMatches so that we don't use it to construct
391 // another signature match in a later call.
392 mPendingMatches[signatureMatch.size()-1].remove(v.sequence);
394 // Declare success: notify observers
395 mObservers.forEach(obs -> obs.onSignatureDetected(mSignature,
396 Collections.unmodifiableList(signatureMatch)));
403 * Encapsulates a {@code List<PcapPacket>} so as to allow the list to be used as a vertex in a graph while avoiding
404 * the expensive {@link AbstractList#equals(Object)} calls when adding vertices to the graph.
405 * Using this wrapper makes the incurred {@code equals(Object)} calls delegate to {@link Object#equals(Object)}
406 * instead of {@link AbstractList#equals(Object)}. The net effect is a faster implementation, but the graph will not
407 * recognize two lists that contain the same items--from a value and not reference point of view--as the same
408 * vertex. However, this is fine for our purposes -- in fact restricting it to reference equality seems more
411 private static class Vertex {
412 private final List<PcapPacket> sequence;
413 private Vertex(List<PcapPacket> wrappedSequence) {
414 sequence = wrappedSequence;