1 package edu.uci.iotproject.detection.layer2;
3 import edu.uci.iotproject.analysis.TriggerTrafficExtractor;
4 import edu.uci.iotproject.trafficreassembly.layer2.Layer2FlowReassembler;
5 import edu.uci.iotproject.trafficreassembly.layer2.Layer2Flow;
6 import edu.uci.iotproject.trafficreassembly.layer2.Layer2FlowReassemblerObserver;
7 import edu.uci.iotproject.detection.AbstractClusterMatcher;
8 import edu.uci.iotproject.trafficreassembly.layer2.Layer2FlowObserver;
9 import org.pcap4j.core.*;
11 import java.util.ArrayList;
12 import java.util.HashMap;
13 import java.util.List;
15 import java.util.function.Function;
18 * Attempts to detect members of a cluster (packet sequence mutations) in layer 2 flows.
20 * @author Janus Varmarken {@literal <jvarmark@uci.edu>}
21 * @author Rahmadi Trimananda {@literal <rtrimana@uci.edu>}
23 public class Layer2ClusterMatcher extends AbstractClusterMatcher implements Layer2FlowReassemblerObserver, Layer2FlowObserver {
26 * Maps from a flow to a table of {@link Layer2SequenceMatcher}s for that particular flow. The table {@code t} is
27 * structured such that {@code t[i][j]} is a {@link Layer2SequenceMatcher} that attempts to match member {@code i}
28 * of {@link #mCluster} and has so far matched {@code j} packets of that particular sequence.
30 private final Map<Layer2Flow, Layer2SequenceMatcher[][]> mPerFlowSeqMatchers = new HashMap<>();
31 // private final Map<Layer2Flow, Layer2RangeMatcher[]> mPerFlowRangeMatcher = new HashMap<>();
32 private final Map<Layer2Flow, List<Layer2RangeMatcher>> mPerFlowRangeMatcher = new HashMap<>();
34 private final Function<Layer2Flow, Boolean> mFlowFilter;
37 * Specifying range-based instead of conservative exact matching.
39 private final boolean mRangeBased;
42 * Epsilon value used by the DBSCAN algorithm; it is used again for range-based matching here.
44 private final double mEps;
46 private int mInclusionTimeMillis;
49 * Keeping track of maximum number of skipped packets
51 //private int mMaxSkippedPackets;
52 private List<Integer> mMaxSkippedPackets;
55 * Create a new {@link Layer2ClusterMatcher} that attempts to find occurrences of {@code cluster}'s members.
56 * @param cluster The sequence mutations that the new {@link Layer2ClusterMatcher} should search for.
58 public Layer2ClusterMatcher(List<List<PcapPacket>> cluster, int inclusionTimeMillis,
59 boolean isRangeBased, double eps) {
60 // Consider all flows if no flow filter specified.
61 this(cluster, flow -> true, inclusionTimeMillis, isRangeBased, eps);
65 * Create a new {@link Layer2ClusterMatcher} that attempts to find occurrences of {@code cluster}'s members.
66 * @param cluster The sequence mutations that the new {@link Layer2ClusterMatcher} should search for.
67 * @param flowFilter A filter that defines what {@link Layer2Flow}s the new {@link Layer2ClusterMatcher} should
68 * search for {@code cluster}'s members in. If {@code flowFilter} returns {@code true}, the flow
69 * will be included (searched). Note that {@code flowFilter} is only queried once for each flow,
70 * namely when the {@link Layer2FlowReassembler} notifies the {@link Layer2ClusterMatcher} about
71 * the new flow. This functionality may for example come in handy when one only wants to search
72 * for matches in the subset of flows that involves a specific (range of) MAC(s).
73 * @param inclusionTimeMillis Packet inclusion time limit for matching.
74 * @param isRangeBased The boolean that decides if it is range-based vs. strict matching.
75 * @param eps The epsilon value used in the DBSCAN algorithm.
77 public Layer2ClusterMatcher(List<List<PcapPacket>> cluster, Function<Layer2Flow, Boolean> flowFilter,
78 int inclusionTimeMillis, boolean isRangeBased, double eps) {
79 super(cluster, isRangeBased);
80 mFlowFilter = flowFilter;
81 mRangeBased = isRangeBased;
83 mInclusionTimeMillis =
84 inclusionTimeMillis == 0 ? TriggerTrafficExtractor.INCLUSION_WINDOW_MILLIS : inclusionTimeMillis;
85 //mMaxSkippedPackets = 0;
86 mMaxSkippedPackets = new ArrayList<>();
90 public void onNewPacket(Layer2Flow flow, PcapPacket newPacket) {
92 rangeBasedMatching(flow, newPacket);
94 conservativeMatching(flow, newPacket);
98 private void conservativeMatching(Layer2Flow flow, PcapPacket newPacket) {
99 if (mPerFlowSeqMatchers.get(flow) == null) {
100 // If this is the first time we encounter this flow, we need to set up sequence matchers for it.
101 // All sequences of the cluster have the same length, so we only need to compute the length of the nested
102 // arrays once. We want to make room for a cluster matcher in each state, including the initial empty state
103 // but excluding the final "full match" state (as there is no point in keeping a terminated sequence matcher
104 // around), so the length of the inner array is simply the sequence length.
105 Layer2SequenceMatcher[][] matchers = new Layer2SequenceMatcher[mCluster.size()][mCluster.get(0).size()];
106 // Prepare a "state 0" sequence matcher for each sequence variation in the cluster.
107 for (int i = 0; i < matchers.length; i++) {
108 matchers[i][0] = new Layer2SequenceMatcher(mCluster.get(i), mInclusionTimeMillis);
110 // Associate the new sequence matcher table with the new flow
111 mPerFlowSeqMatchers.put(flow, matchers);
113 // Fetch table that contains sequence matchers for this flow.
114 Layer2SequenceMatcher[][] matchers = mPerFlowSeqMatchers.get(flow);
115 // Present the packet to all sequence matchers.
116 for (int i = 0; i < matchers.length; i++) {
117 // Present packet to the sequence matchers that has advanced the most first. This is to prevent discarding
118 // the sequence matchers that have advanced the most in the special case where the searched sequence
119 // contains two packets of the same length going in the same direction.
120 for (int j = matchers[i].length - 1; j >= 0 ; j--) {
121 Layer2SequenceMatcher sm = matchers[i][j];
123 // There is currently no sequence matcher that has managed to match j packets.
126 boolean matched = sm.matchPacket(newPacket);
128 if (sm.getMatchedPacketsCount() == sm.getTargetSequencePacketCount()) {
129 // Update maximum skipped packets
130 updateMaxSkippedPackets(flow.getPackets(), sm.getMatchedPackets());
131 // Sequence matcher has a match. Report it to observers.
132 mObservers.forEach(o -> o.onMatch(this, sm.getMatchedPackets()));
133 // Remove the now terminated sequence matcher.
134 matchers[i][j] = null;
136 // Sequence matcher advanced one step, so move it to its corresponding new position iff the
137 // packet that advanced it has a later timestamp than that of the last matched packet of the
138 // sequence matcher at the new index, if any. In most traces, a small amount of the packets
139 // appear out of order (with regards to their timestamp), which is why this check is required.
140 // Obviously it would not be needed if packets where guaranteed to be processed in timestamp
142 if (matchers[i][j+1] == null ||
143 newPacket.getTimestamp().isAfter(matchers[i][j+1].getLastPacket().getTimestamp())) {
144 matchers[i][j+1] = sm;
147 // We always want to have a sequence matcher in state 0, regardless of if the one that advanced
148 // from state zero completed its matching or if it replaced a different one in state 1 or not.
149 if (sm.getMatchedPacketsCount() == 1) {
150 matchers[i][j] = new Layer2SequenceMatcher(sm.getTargetSequence(), mInclusionTimeMillis);
157 // Update the maximum number of skipped packets
158 private void updateMaxSkippedPackets(List<PcapPacket> flowPackets, List<PcapPacket> matchedPackets) {
159 // Count number of skipped packets by looking into
160 // the difference of indices of two matched packets
161 for(int i = 1; i < matchedPackets.size(); ++i) {
162 int currIndex = flowPackets.indexOf(matchedPackets.get(i-1));
163 int nextIndex = flowPackets.indexOf(matchedPackets.get(i));
164 int skippedPackets = nextIndex - currIndex;
165 // if (mMaxSkippedPackets < skippedPackets) {
166 // mMaxSkippedPackets = skippedPackets;
168 mMaxSkippedPackets.add(skippedPackets);
172 private void rangeBasedMatching(Layer2Flow flow, PcapPacket newPacket) {
173 // TODO: For range-based matching, we need to create a new matcher every time we see the first element of
174 // the sequence (between lower and upper bounds).
175 if (mPerFlowRangeMatcher.get(flow) == null) {
176 // If this is the first time we encounter this flow, we need to set up a list of sequence matchers.
177 List<Layer2RangeMatcher> listMatchers = new ArrayList<>();
178 // Prepare a "state 0" sequence matcher.
179 Layer2RangeMatcher matcher = new Layer2RangeMatcher(mCluster.get(0), mCluster.get(1),
180 mInclusionTimeMillis, mEps);
181 listMatchers.add(matcher);
182 // Associate the new sequence matcher table with the new flow.
183 mPerFlowRangeMatcher.put(flow, listMatchers);
185 // Fetch table that contains sequence matchers for this flow.
186 List<Layer2RangeMatcher> listMatchers = mPerFlowRangeMatcher.get(flow);
187 // Add a new matcher if all matchers have already advanced to the next stage.
188 // We always need a new matcher to match from NO packets.
189 boolean addOneArray = true;
190 for(Layer2RangeMatcher matcher : listMatchers) {
191 if (matcher.getMatchedPacketsCount() == 0) {
195 // Add the new matcher into the list
197 Layer2RangeMatcher newMatcher = new Layer2RangeMatcher(mCluster.get(0), mCluster.get(1),
198 mInclusionTimeMillis, mEps);
199 listMatchers.add(newMatcher);
201 // Present packet to the sequence matchers.
202 // Make a shallow copy of the list so that we can clean up the actual list when a matcher is terminated.
203 // Otherwise, we would get an exception for changing the list while iterating on it.
204 List<Layer2RangeMatcher> listMatchersCopy = new ArrayList<>(listMatchers);
205 for(Layer2RangeMatcher matcher : listMatchersCopy) {
206 Layer2RangeMatcher sm = matcher;
207 // Check if no packets are matched yet or if there are matched packets, the next packet to be matched
208 // has to be later than the last matched packet.
209 // In most traces, a small amount of the packets appear out of order (with regards to their timestamp),
210 // which is why this check is required.
211 // Obviously it would not be needed if packets where guaranteed to be processed in timestamp
213 if (sm.getMatchedPacketsCount() == 0 ||
214 newPacket.getTimestamp().isAfter(sm.getLastPacket().getTimestamp())) {
215 boolean matched = sm.matchPacket(newPacket);
217 if (sm.getMatchedPacketsCount() == sm.getTargetSequencePacketCount()) {
218 // Update maximum skipped packets
219 updateMaxSkippedPackets(flow.getPackets(), sm.getMatchedPackets());
220 // Sequence matcher has a match. Report it to observers.
221 mObservers.forEach(o -> o.onMatch(this, sm.getMatchedPackets()));
222 // Terminate sequence matcher since matching is complete.
223 listMatchers.remove(matcher);
231 protected List<List<PcapPacket>> pruneCluster(List<List<PcapPacket>> cluster) {
232 // Note: we assume that all sequences in the input cluster are of the same length and that their packet
233 // directions are identical.
234 List<List<PcapPacket>> prunedCluster = new ArrayList<>();
235 for (List<PcapPacket> originalClusterSeq : cluster) {
236 boolean alreadyPresent = prunedCluster.stream().anyMatch(pcPkts -> {
237 for (int i = 0; i < pcPkts.size(); i++) {
238 if (pcPkts.get(i).getOriginalLength() != originalClusterSeq.get(i).getOriginalLength()) {
244 if (!alreadyPresent) {
245 // Add the sequence if not already present in the pruned cluster.
246 prunedCluster.add(originalClusterSeq);
249 return prunedCluster;
252 private static final boolean DEBUG = false;
255 public void onNewFlow(Layer2FlowReassembler reassembler, Layer2Flow newFlow) {
256 // New flow detected. Check if we should consider it when searching for cluster member matches.
257 if (mFlowFilter.apply(newFlow)) {
259 System.out.println(">>> ACCEPTING FLOW: " + newFlow + " <<<");
261 // Subscribe to the new flow to get updates whenever a new packet pertaining to the flow is processed.
262 newFlow.addFlowObserver(this);
264 System.out.println(">>> IGNORING FLOW: " + newFlow + " <<<");
269 * Return the maximum number of skipped packets.
271 // public int getMaxSkippedPackets() {
272 // return mMaxSkippedPackets;
274 public List<Integer> getMaxSkippedPackets() {
275 return mMaxSkippedPackets;