Attempting to add router's MAC address as a reference for packet direction in layer...
[pingpong.git] / Code / Projects / PacketLevelSignatureExtractor / src / main / java / edu / uci / iotproject / detection / layer2 / Layer2ClusterMatcher.java
index 5021c3119693bf84b5dc076e79cdf30e614ab815..0bfc6b2990d0b06852373b6e2ef6b3e1dcffe058 100644 (file)
@@ -1,5 +1,6 @@
 package edu.uci.iotproject.detection.layer2;
 
+import edu.uci.iotproject.analysis.TriggerTrafficExtractor;
 import edu.uci.iotproject.trafficreassembly.layer2.Layer2FlowReassembler;
 import edu.uci.iotproject.trafficreassembly.layer2.Layer2Flow;
 import edu.uci.iotproject.trafficreassembly.layer2.Layer2FlowReassemblerObserver;
@@ -27,35 +28,89 @@ public class Layer2ClusterMatcher extends AbstractClusterMatcher implements Laye
      * of {@link #mCluster} and has so far matched {@code j} packets of that particular sequence.
      */
     private final Map<Layer2Flow, Layer2SequenceMatcher[][]> mPerFlowSeqMatchers = new HashMap<>();
+    private final Map<Layer2Flow, List<Layer2RangeMatcher>> mPerFlowRangeMatcher = new HashMap<>();
 
     private final Function<Layer2Flow, Boolean> mFlowFilter;
 
+    /**
+     * Specifying range-based instead of conservative exact matching.
+     */
+    private final boolean mRangeBased;
+
+    /**
+     * Epsilon value used by the DBSCAN algorithm; it is used again for range-based matching here.
+     */
+    private final double mEps;
+
+    private int mInclusionTimeMillis;
+
+    /**
+     * Keeping track of maximum number of skipped packets
+     */
+    private int mMaxSkippedPackets;
+    private List<Integer> mSkippedPackets;
+
+    private int mLimitSkippedPackets;
+
+    /**
+     * Router's WLAN MAC.
+     */
+    private String mTrainingRouterWlanMac;
+    private String mRouterWlanMac;
+
     /**
      * Create a new {@link Layer2ClusterMatcher} that attempts to find occurrences of {@code cluster}'s members.
      * @param cluster The sequence mutations that the new {@link Layer2ClusterMatcher} should search for.
      */
-    public Layer2ClusterMatcher(List<List<PcapPacket>> cluster) {
+    public Layer2ClusterMatcher(List<List<PcapPacket>> cluster, String trainingRouterWlanMac, String routerWlanMac, int inclusionTimeMillis,
+                                boolean isRangeBased, double eps, int limitSkippedPackets) {
         // Consider all flows if no flow filter specified.
-        this(cluster, flow -> true);
+        this(cluster, trainingRouterWlanMac, routerWlanMac, flow -> true, inclusionTimeMillis, isRangeBased, eps,
+                limitSkippedPackets);
     }
 
     /**
      * Create a new {@link Layer2ClusterMatcher} that attempts to find occurrences of {@code cluster}'s members.
      * @param cluster The sequence mutations that the new {@link Layer2ClusterMatcher} should search for.
+     * @param trainingRouterWlanMac The training router's WLAN MAC (used for determining the direction of packets).
+     * @param routerWlanMac The target trace router's WLAN MAC (used for determining the direction of packets).
      * @param flowFilter A filter that defines what {@link Layer2Flow}s the new {@link Layer2ClusterMatcher} should
      *                   search for {@code cluster}'s members in. If {@code flowFilter} returns {@code true}, the flow
      *                   will be included (searched). Note that {@code flowFilter} is only queried once for each flow,
      *                   namely when the {@link Layer2FlowReassembler} notifies the {@link Layer2ClusterMatcher} about
      *                   the new flow. This functionality may for example come in handy when one only wants to search
      *                   for matches in the subset of flows that involves a specific (range of) MAC(s).
+     * @param inclusionTimeMillis Packet inclusion time limit for matching.
+     * @param isRangeBased The boolean that decides if it is range-based vs. strict matching.
+     * @param eps The epsilon value used in the DBSCAN algorithm.
      */
-    public Layer2ClusterMatcher(List<List<PcapPacket>> cluster, Function<Layer2Flow, Boolean> flowFilter) {
-        super(cluster);
+    public Layer2ClusterMatcher(List<List<PcapPacket>> cluster, String trainingRouterWlanMac, String routerWlanMac,
+                                Function<Layer2Flow, Boolean> flowFilter, int inclusionTimeMillis, boolean isRangeBased,
+                                double eps, int limitSkippedPackets) {
+        super(cluster, isRangeBased);
         mFlowFilter = flowFilter;
+        mTrainingRouterWlanMac = trainingRouterWlanMac;
+        mRouterWlanMac = routerWlanMac;
+        mRangeBased = isRangeBased;
+        mEps = eps;
+        mInclusionTimeMillis =
+                inclusionTimeMillis == 0 ? TriggerTrafficExtractor.INCLUSION_WINDOW_MILLIS : inclusionTimeMillis;
+        mMaxSkippedPackets = 0;
+        mSkippedPackets = new ArrayList<>();
+        // Give integer's MAX_VALUE if -1
+        mLimitSkippedPackets = limitSkippedPackets == -1 ? Integer.MAX_VALUE : limitSkippedPackets;
     }
 
     @Override
     public void onNewPacket(Layer2Flow flow, PcapPacket newPacket) {
+        if (mRangeBased) {
+            rangeBasedMatching(flow, newPacket);
+        } else {
+            conservativeMatching(flow, newPacket);
+        }
+    }
+
+    private void conservativeMatching(Layer2Flow flow, PcapPacket newPacket) {
         if (mPerFlowSeqMatchers.get(flow) == null) {
             // If this is the first time we encounter this flow, we need to set up sequence matchers for it.
             // All sequences of the cluster have the same length, so we only need to compute the length of the nested
@@ -65,7 +120,8 @@ public class Layer2ClusterMatcher extends AbstractClusterMatcher implements Laye
             Layer2SequenceMatcher[][] matchers = new Layer2SequenceMatcher[mCluster.size()][mCluster.get(0).size()];
             // Prepare a "state 0" sequence matcher for each sequence variation in the cluster.
             for (int i = 0; i < matchers.length; i++) {
-                matchers[i][0] = new Layer2SequenceMatcher(mCluster.get(i));
+                matchers[i][0] = new Layer2SequenceMatcher(mCluster.get(i), mInclusionTimeMillis, mTrainingRouterWlanMac,
+                        mRouterWlanMac);
             }
             // Associate the new sequence matcher table with the new flow
             mPerFlowSeqMatchers.put(flow, matchers);
@@ -86,8 +142,12 @@ public class Layer2ClusterMatcher extends AbstractClusterMatcher implements Laye
                 boolean matched = sm.matchPacket(newPacket);
                 if (matched) {
                     if (sm.getMatchedPacketsCount() == sm.getTargetSequencePacketCount()) {
-                        // Sequence matcher has a match. Report it to observers.
-                        mObservers.forEach(o -> o.onMatch(this, sm.getMatchedPackets()));
+                        // Update maximum skipped packets
+                        boolean stillMatch = checkMaxSkippedPackets(flow.getPackets(), sm.getMatchedPackets());
+                        if (stillMatch) {
+                            // Sequence matcher has a match. Report it to observers.
+                            mObservers.forEach(o -> o.onMatch(this, sm.getMatchedPackets()));
+                        }
                         // Remove the now terminated sequence matcher.
                         matchers[i][j] = null;
                     } else {
@@ -105,13 +165,110 @@ public class Layer2ClusterMatcher extends AbstractClusterMatcher implements Laye
                     // We always want to have a sequence matcher in state 0, regardless of if the one that advanced
                     // from state zero completed its matching or if it replaced a different one in state 1 or not.
                     if (sm.getMatchedPacketsCount() == 1) {
-                        matchers[i][j] = new Layer2SequenceMatcher(sm.getTargetSequence());
+                        matchers[i][j] = new Layer2SequenceMatcher(sm.getTargetSequence(), mInclusionTimeMillis,
+                                mTrainingRouterWlanMac, mRouterWlanMac);
                     }
                 }
             }
         }
     }
 
+    // Update the maximum number of skipped packets.
+    private boolean checkMaxSkippedPackets(List<PcapPacket> flowPackets, List<PcapPacket> matchedPackets) {
+        // Count number of skipped packets by looking into
+        // the difference of indices of two matched packets.
+        boolean stillMatch = true;
+        for(int i = 1; i < matchedPackets.size(); ++i) {
+            int currIndex = flowPackets.indexOf(matchedPackets.get(i-1));
+            int nextIndex = flowPackets.indexOf(matchedPackets.get(i));
+            int skippedPackets = nextIndex - currIndex;
+            if (mMaxSkippedPackets < skippedPackets) {
+                mMaxSkippedPackets = skippedPackets;
+            }
+            if (mLimitSkippedPackets < skippedPackets) {
+                stillMatch = false;
+            }
+            mSkippedPackets.add(skippedPackets);
+        }
+        return stillMatch;
+    }
+
+    private void rangeBasedMatching(Layer2Flow flow, PcapPacket newPacket) {
+        // TODO: For range-based matching, we need to create a new matcher every time we see the first element of
+        //  the sequence (between lower and upper bounds).
+        if (mPerFlowRangeMatcher.get(flow) == null) {
+            // If this is the first time we encounter this flow, we need to set up a list of sequence matchers.
+            List<Layer2RangeMatcher> listMatchers = new ArrayList<>();
+            // Prepare a "state 0" sequence matcher.
+            Layer2RangeMatcher matcher = new Layer2RangeMatcher(mCluster.get(0), mCluster.get(1),
+                    mInclusionTimeMillis, mEps, mTrainingRouterWlanMac, mRouterWlanMac);
+            listMatchers.add(matcher);
+            // Associate the new sequence matcher table with the new flow.
+            mPerFlowRangeMatcher.put(flow, listMatchers);
+        }
+        // Fetch table that contains sequence matchers for this flow.
+        List<Layer2RangeMatcher> listMatchers = mPerFlowRangeMatcher.get(flow);
+        // Add a new matcher if all matchers have already advanced to the next stage.
+        // We always need a new matcher to match from NO packets.
+        boolean addOneArray = true;
+        for(Layer2RangeMatcher matcher : listMatchers) {
+            if (matcher.getMatchedPacketsCount() == 0) {
+                addOneArray = false;
+            }
+        }
+        // Add the new matcher into the list
+        if (addOneArray) {
+            Layer2RangeMatcher newMatcher = new Layer2RangeMatcher(mCluster.get(0), mCluster.get(1),
+                    mInclusionTimeMillis, mEps, mTrainingRouterWlanMac, mRouterWlanMac);
+            listMatchers.add(newMatcher);
+        }
+        // Present packet to the sequence matchers.
+        // Make a shallow copy of the list so that we can clean up the actual list when a matcher is terminated.
+        // Otherwise, we would get an exception for changing the list while iterating on it.
+        List<Layer2RangeMatcher> listMatchersCopy = new ArrayList<>(listMatchers);
+        Layer2RangeMatcher previousMatcher = null;
+        for(Layer2RangeMatcher matcher : listMatchersCopy) {
+            Layer2RangeMatcher sm = matcher;
+            // Check if no packets are matched yet or if there are matched packets, the next packet to be matched
+            // has to be later than the last matched packet.
+            // In most traces, a small amount of the packets appear out of order (with regards to their timestamp),
+            // which is why this check is required.
+            // Obviously it would not be needed if packets where guaranteed to be processed in timestamp
+            // order here.
+            if (sm.getMatchedPacketsCount() == 0 ||
+                    newPacket.getTimestamp().isAfter(sm.getLastPacket().getTimestamp())) {
+                boolean matched = sm.matchPacket(newPacket);
+                if (matched) {
+                    // BUG: found on May 29, 2019
+                    // We need to remove a previous match if the current match is later in time.
+                    // This is done only if we have matched at least 1 packet (we are about to match the second or
+                    // later packets) and we match for the same packet position in the signature (check the size!).
+                    if (previousMatcher != null && sm.getMatchedPacketsCount() > 1 &&
+                            previousMatcher.getMatchedPacketsCount() == sm.getMatchedPacketsCount()) {
+                        List<PcapPacket> previouslyMatchedPackets = previousMatcher.getMatchedPackets();
+                        List<PcapPacket> currentlyMatchedPackets = sm.getMatchedPackets();
+                        // We need to check 1 packet before the last matched packet from the previous matcher.
+                        int packetIndexToCheck = (sm.getMatchedPacketsCount() - 1) - 1;
+                        if (currentlyMatchedPackets.get(packetIndexToCheck).getTimestamp().
+                            isAfter(previouslyMatchedPackets.get(packetIndexToCheck).getTimestamp())) {
+                            listMatchers.remove(previousMatcher);
+                        }
+                    }
+                    if (sm.getMatchedPacketsCount() == sm.getTargetSequencePacketCount()) {
+                        // Update maximum skipped packets
+                        boolean stillMatch = checkMaxSkippedPackets(flow.getPackets(), sm.getMatchedPackets());
+                        if (stillMatch) {
+                            // Sequence matcher has a match. Report it to observers.
+                            mObservers.forEach(o -> o.onMatch(this, sm.getMatchedPackets()));
+                        }
+                        // Terminate sequence matcher since matching is complete.
+                        listMatchers.remove(matcher);
+                    }
+                    previousMatcher = sm;
+                }
+            }
+        }
+    }
 
     @Override
     protected List<List<PcapPacket>> pruneCluster(List<List<PcapPacket>> cluster) {
@@ -150,4 +307,18 @@ public class Layer2ClusterMatcher extends AbstractClusterMatcher implements Laye
             System.out.println(">>> IGNORING FLOW: " + newFlow + " <<<");
         }
     }
+
+    /**
+      * Return the maximum number of skipped packets.
+      */
+    public int getMaxSkippedPackets() {
+       return mMaxSkippedPackets;
+    }
+
+    /**
+     * Return the numbers of skipped packets.
+     */
+    public List<Integer> getSkippedPackets() {
+        return mSkippedPackets;
+    }
 }