[pingpong.git] / Code / Projects / PacketLevelSignatureExtractor / src / main / java / edu / uci / iotproject / detection / layer2 / Layer2SignatureDetector.java

package edu.uci.iotproject.detection.layer2;

import edu.uci.iotproject.analysis.TriggerTrafficExtractor;
import edu.uci.iotproject.analysis.UserAction;
import edu.uci.iotproject.detection.AbstractClusterMatcher;
import edu.uci.iotproject.detection.ClusterMatcherObserver;
import edu.uci.iotproject.detection.SignatureDetectorObserver;
import edu.uci.iotproject.io.PcapHandleReader;
import edu.uci.iotproject.io.PrintWriterUtils;
import edu.uci.iotproject.trafficreassembly.layer2.Layer2Flow;
import edu.uci.iotproject.trafficreassembly.layer2.Layer2FlowReassembler;
import edu.uci.iotproject.util.PcapPacketUtils;
import edu.uci.iotproject.util.PrintUtils;
import org.jgrapht.GraphPath;
import org.jgrapht.alg.shortestpath.DijkstraShortestPath;
import org.jgrapht.graph.DefaultWeightedEdge;
import org.jgrapht.graph.SimpleDirectedWeightedGraph;
import org.pcap4j.core.*;

import java.io.File;
import java.io.FileWriter;
import java.io.IOException;
import java.io.PrintWriter;
import java.time.Duration;
import java.util.*;
import java.util.function.Function;
import java.util.regex.Pattern;

/**
 * Performs layer 2 signature detection.
 *
 * @author Janus Varmarken {@literal <jvarmark@uci.edu>}
 * @author Rahmadi Trimananda {@literal <rtrimana@uci.edu>}
 */
public class Layer2SignatureDetector implements PacketListener, ClusterMatcherObserver {

    /**
     * If set to {@code true}, output written to the results file is also dumped to standard out.
     */
    private static boolean DUPLICATE_OUTPUT_TO_STD_OUT = true;

    private static List<Function<Layer2Flow, Boolean>> parseSignatureMacFilters(String filtersString) {
        List<Function<Layer2Flow, Boolean>> filters = new ArrayList<>();
        String[] filterRegexes = filtersString.split(";");
        for (String filterRegex : filterRegexes) {
            final Pattern regex = Pattern.compile(filterRegex);
            // Create a filter that includes all flows where one of the two MAC addresses match the regex.
            filters.add(flow -> regex.matcher(flow.getEndpoint1().toString()).matches() || regex.matcher(flow.getEndpoint2().toString()).matches());
        }
        return filters;
    }

    public static void main(String[] args) throws PcapNativeException, NotOpenException, IOException {
        // Parse required parameters.
        if (args.length < 8) {
            String errMsg = String.format("Usage: %s inputPcapFile onAnalysisFile offAnalysisFile onSignatureFile offSignatureFile resultsFile" +
                            "\n  inputPcapFile: the target of the detection" +
                            "\n  onAnalysisFile: the file that contains the ON clusters analysis" +
                            "\n  offAnalysisFile: the file that contains the OFF clusters analysis" +
                            "\n  onSignatureFile: the file that contains the ON signature to search for" +
                            "\n  offSignatureFile: the file that contains the OFF signature to search for" +
                            "\n  resultsFile: where to write the results of the detection" +
                            "\n  signatureDuration: the maximum duration of signature detection",
                    Layer2SignatureDetector.class.getSimpleName());
            System.out.println(errMsg);
            String optParamsExplained = "Above are the required, positional arguments. In addition to these, the " +
                    "following options and associated positional arguments may be used:\n" +
                    "  '-onmacfilters <regex>;<regex>;...;<regex>' which specifies that sequence matching should ONLY" +
                    " be performed on flows where the MAC of one of the two endpoints matches the given regex. Note " +
                    "that you MUST specify a regex for each cluster of the signature. This is to facilitate more " +
                    "aggressive filtering on parts of the signature (e.g., the communication that involves the " +
                    "smart home device itself as one can drop all flows that do not include an endpoint with a MAC " +
                    "that matches the vendor's prefix).\n" +
                    "  '-offmacfilters <regex>;<regex>;...;<regex>' works exactly the same as onmacfilters, but " +
                    "applies to the OFF signature instead of the ON signature.\n" +
                    "  '-sout <boolean literal>' true/false literal indicating if output should also be printed to std out; default is true.";
            System.out.println(optParamsExplained);
            return;
        }
        // TODO: We could take 7 inputs if we decided to use the cluster analyses.
        final String pcapFile = args[0];
        final String onClusterAnalysisFile = args[1];
        final String offClusterAnalysisFile = args[2];
        final String onSignatureFile = args[3];
        final String offSignatureFile = args[4];
        final String resultsFile = args[5];
        final int signatureDuration = Integer.parseInt(args[6]);
        final double eps = Double.parseDouble(args[7]);

//        final String pcapFile = args[0];
//        final String onSignatureFile = args[1];
//        final String offSignatureFile = args[2];
//        final String resultsFile = args[3];
//        final int signatureDuration = Integer.parseInt(args[4]);

        // Parse optional parameters.
        List<Function<Layer2Flow, Boolean>> onSignatureMacFilters = null, offSignatureMacFilters = null;
        final int optParamsStartIdx = 7;
        if (args.length > optParamsStartIdx) {
            for (int i = optParamsStartIdx; i < args.length; i++) {
                if (args[i].equalsIgnoreCase("-onMacFilters")) {
                    // Next argument is the cluster-wise MAC filters (separated by semicolons).
                    onSignatureMacFilters = parseSignatureMacFilters(args[i+1]);
                } else if (args[i].equalsIgnoreCase("-offMacFilters")) {
                    // Next argument is the cluster-wise MAC filters (separated by semicolons).
                    offSignatureMacFilters = parseSignatureMacFilters(args[i+1]);
                } else if (args[i].equalsIgnoreCase("-sout")) {
                    // Next argument is a boolean true/false literal.
                    DUPLICATE_OUTPUT_TO_STD_OUT = Boolean.parseBoolean(args[i+1]);
                }
            }
        }

        // Prepare file outputter.
        File outputFile = new File(resultsFile);
        outputFile.getParentFile().mkdirs();
        final PrintWriter resultsWriter = new PrintWriter(new FileWriter(outputFile));
        // Include metadata as comments at the top
        PrintWriterUtils.println("# Detection results for:", resultsWriter, DUPLICATE_OUTPUT_TO_STD_OUT);
        PrintWriterUtils.println("# - inputPcapFile: " + pcapFile, resultsWriter, DUPLICATE_OUTPUT_TO_STD_OUT);
        PrintWriterUtils.println("# - onAnalysisFile: " + onClusterAnalysisFile, resultsWriter, DUPLICATE_OUTPUT_TO_STD_OUT);
        PrintWriterUtils.println("# - offAnalysisFile: " + offClusterAnalysisFile, resultsWriter, DUPLICATE_OUTPUT_TO_STD_OUT);
        PrintWriterUtils.println("# - onSignatureFile: " + onSignatureFile, resultsWriter, DUPLICATE_OUTPUT_TO_STD_OUT);
        PrintWriterUtils.println("# - offSignatureFile: " + offSignatureFile, resultsWriter, DUPLICATE_OUTPUT_TO_STD_OUT);
        resultsWriter.flush();

        // Create signature detectors and add observers that output their detected events.
        List<List<List<PcapPacket>>> onSignature = PrintUtils.deserializeFromFile(onSignatureFile);
        List<List<List<PcapPacket>>> offSignature = PrintUtils.deserializeFromFile(offSignatureFile);
        // Load signature analyses
        List<List<List<PcapPacket>>> onClusterAnalysis = PrintUtils.deserializeFromFile(onClusterAnalysisFile);
        List<List<List<PcapPacket>>> offClusterAnalysis = PrintUtils.deserializeFromFile(offClusterAnalysisFile);
        // TODO: FOR NOW WE DECIDE PER SIGNATURE AND THEN WE OR THE BOOLEANS
        // TODO: SINCE WE ONLY HAVE 2 SIGNATURES FOR NOW (ON AND OFF), THEN IT IS USUALLY EITHER RANGE-BASED OR
        // TODO: STRICT MATCHING
        // Check if we should use range-based matching
//        boolean isRangeBasedForOn = PcapPacketUtils.isRangeBasedMatching(onSignature, eps, offSignature);
//        boolean isRangeBasedForOff = PcapPacketUtils.isRangeBasedMatching(offSignature, eps, onSignature);
        // TODO: WE DON'T DO RANGE-BASED FOR NOW BECAUSE THE RESULTS ARE TERRIBLE FOR LAYER 2 MATCHING
        // TODO: THIS WOULD ONLY WORK FOR SIGNATURES LONGER THAN 2 PACKETS
        boolean isRangeBasedForOn = false;
        boolean isRangeBasedForOff = false;
        // Update the signature with ranges if it is range-based
        if (isRangeBasedForOn) {
            onSignature = PcapPacketUtils.useRangeBasedMatching(onSignature, onClusterAnalysis);
        }
        if (isRangeBasedForOff) {
            offSignature = PcapPacketUtils.useRangeBasedMatching(offSignature, offClusterAnalysis);
        }
        Layer2SignatureDetector onDetector = onSignatureMacFilters == null ?
                new Layer2SignatureDetector(onSignature, isRangeBasedForOn, eps) :
                new Layer2SignatureDetector(onSignature, onSignatureMacFilters, signatureDuration, isRangeBasedForOn, eps);
        Layer2SignatureDetector offDetector = offSignatureMacFilters == null ?
                new Layer2SignatureDetector(offSignature, isRangeBasedForOff, eps) :
                new Layer2SignatureDetector(offSignature, offSignatureMacFilters, signatureDuration, isRangeBasedForOff, eps);
        final List<UserAction> detectedEvents = new ArrayList<>();
        onDetector.addObserver((signature, match) -> {
            UserAction event = new UserAction(UserAction.Type.TOGGLE_ON, match.get(0).get(0).getTimestamp());
            PrintWriterUtils.println(event, resultsWriter, DUPLICATE_OUTPUT_TO_STD_OUT);
            detectedEvents.add(event);
        });
        offDetector.addObserver((signature, match) -> {
            UserAction event = new UserAction(UserAction.Type.TOGGLE_OFF, match.get(0).get(0).getTimestamp());
            PrintWriterUtils.println(event, resultsWriter, DUPLICATE_OUTPUT_TO_STD_OUT);
            for (PcapPacket pcap : match.get(0)) {
                System.out.println(pcap.length() + " -> " + pcap.getTimestamp());
            }
            detectedEvents.add(event);
        });

        // Load the PCAP file
        PcapHandle handle;
        try {
            handle = Pcaps.openOffline(pcapFile, PcapHandle.TimestampPrecision.NANO);
        } catch (PcapNativeException pne) {
            handle = Pcaps.openOffline(pcapFile);
        }
        PcapHandleReader reader = new PcapHandleReader(handle, p -> true, onDetector, offDetector);
        // Parse the file
        reader.readFromHandle();

        String resultOn = "Number of detected events of type " + UserAction.Type.TOGGLE_ON + ": " +
                detectedEvents.stream().filter(ua -> ua.getType() == UserAction.Type.TOGGLE_ON).count();
        String resultOff = "Number of detected events of type " + UserAction.Type.TOGGLE_OFF + ": " +
                detectedEvents.stream().filter(ua -> ua.getType() == UserAction.Type.TOGGLE_OFF).count();
        PrintWriterUtils.println(resultOn, resultsWriter, DUPLICATE_OUTPUT_TO_STD_OUT);
        PrintWriterUtils.println(resultOff, resultsWriter, DUPLICATE_OUTPUT_TO_STD_OUT);

        // Flush output to results file and close it.
        resultsWriter.flush();
        resultsWriter.close();
    }

    /**
     * The signature that this {@link Layer2SignatureDetector} is searching for.
     */
    private final List<List<List<PcapPacket>>> mSignature;

    /**
     * The {@link Layer2ClusterMatcher}s in charge of detecting each individual sequence of packets that together make
     * up the the signature.
     */
    private final List<Layer2ClusterMatcher> mClusterMatchers;

    /**
     * For each {@code i} ({@code i >= 0 && i < mPendingMatches.length}), {@code mPendingMatches[i]} holds the matches
     * found by the {@link Layer2ClusterMatcher} at {@code mClusterMatchers.get(i)} that have yet to be "consumed",
     * i.e., have yet to be included in a signature detected by this {@link Layer2SignatureDetector} (a signature can
     * be encompassed of multiple packet sequences occurring shortly after one another on multiple connections).
     */
    private final List<List<PcapPacket>>[] mPendingMatches;

    /**
     * Maps a {@link Layer2ClusterMatcher} to its corresponding index in {@link #mPendingMatches}.
     */
    private final Map<Layer2ClusterMatcher, Integer> mClusterMatcherIds;

    /**
     * In charge of reassembling layer 2 packet flows.
     */
    private final Layer2FlowReassembler mFlowReassembler = new Layer2FlowReassembler();

    private final List<SignatureDetectorObserver> mObservers = new ArrayList<>();

    private int mInclusionTimeMillis;

    public Layer2SignatureDetector(List<List<List<PcapPacket>>> searchedSignature, boolean isRangeBased, double eps) {
        this(searchedSignature, null, 0, isRangeBased, eps);
    }

    public Layer2SignatureDetector(List<List<List<PcapPacket>>> searchedSignature, List<Function<Layer2Flow,
            Boolean>> flowFilters, int inclusionTimeMillis, boolean isRangeBased, double eps) {
        if (flowFilters != null && flowFilters.size() != searchedSignature.size()) {
            throw new IllegalArgumentException("If flow filters are used, there must be a flow filter for each cluster " +
                    "of the signature.");
        }
        mSignature = Collections.unmodifiableList(searchedSignature);
        List<Layer2ClusterMatcher> clusterMatchers = new ArrayList<>();
        for (int i = 0; i < mSignature.size(); i++) {
            List<List<PcapPacket>> cluster = mSignature.get(i);
            Layer2ClusterMatcher clusterMatcher = flowFilters == null ?
                    new Layer2ClusterMatcher(cluster, isRangeBased, eps) :
                    new Layer2ClusterMatcher(cluster, flowFilters.get(i), isRangeBased, eps);
            clusterMatcher.addObserver(this);
            clusterMatchers.add(clusterMatcher);
        }
        mClusterMatchers = Collections.unmodifiableList(clusterMatchers);
        mPendingMatches = new List[mClusterMatchers.size()];
        for (int i = 0; i < mPendingMatches.length; i++) {
            mPendingMatches[i] = new ArrayList<>();
        }
        Map<Layer2ClusterMatcher, Integer> clusterMatcherIds = new HashMap<>();
        for (int i = 0; i < mClusterMatchers.size(); i++) {
            clusterMatcherIds.put(mClusterMatchers.get(i), i);
        }
        mClusterMatcherIds = Collections.unmodifiableMap(clusterMatcherIds);
        // Register all cluster matchers to receive a notification whenever a new flow is encountered.
        mClusterMatchers.forEach(cm -> mFlowReassembler.addObserver(cm));
        mInclusionTimeMillis =
                inclusionTimeMillis == 0 ? TriggerTrafficExtractor.INCLUSION_WINDOW_MILLIS : inclusionTimeMillis;
    }

    @Override
    public void gotPacket(PcapPacket packet) {
        // Forward packet processing to the flow reassembler that in turn notifies the cluster matchers as appropriate
        mFlowReassembler.gotPacket(packet);
    }

    @Override
    public void onMatch(AbstractClusterMatcher clusterMatcher, List<PcapPacket> match) {
        // TODO: a cluster matcher found a match
        if (clusterMatcher instanceof Layer2ClusterMatcher) {
            // Add the match at the corresponding index
            mPendingMatches[mClusterMatcherIds.get(clusterMatcher)].add(match);
            checkSignatureMatch();
        }
    }

    public void addObserver(SignatureDetectorObserver observer) {
        mObservers.add(observer);
    }

    public boolean removeObserver(SignatureDetectorObserver observer) {
        return mObservers.remove(observer);
    }


    @SuppressWarnings("Duplicates")
    private void checkSignatureMatch() {
        // << Graph-based approach using Balint's idea. >>
        // This implementation assumes that the packets in the inner lists (the sequences) are ordered by asc timestamp.

        // There cannot be a signature match until each Layer3ClusterMatcher has found a match of its respective sequence.
        if (Arrays.stream(mPendingMatches).noneMatch(l -> l.isEmpty())) {
            // Construct the DAG
            final SimpleDirectedWeightedGraph<Vertex, DefaultWeightedEdge> graph =
                    new SimpleDirectedWeightedGraph<>(DefaultWeightedEdge.class);
            // Add a vertex for each match found by all cluster matchers.
            // And maintain an array to keep track of what cluster matcher each vertex corresponds to
            final List<Vertex>[] vertices = new List[mPendingMatches.length];
            for (int i = 0; i < mPendingMatches.length; i++) {
                vertices[i] = new ArrayList<>();
                for (List<PcapPacket> sequence : mPendingMatches[i]) {
                    Vertex v = new Vertex(sequence);
                    vertices[i].add(v); // retain reference for later when we are to add edges
                    graph.addVertex(v); // add to vertex to graph
                }
            }
            // Add dummy source and sink vertices to facilitate search.
            final Vertex source = new Vertex(null);
            final Vertex sink = new Vertex(null);
            graph.addVertex(source);
            graph.addVertex(sink);
            // The source is connected to all vertices that wrap the sequences detected by cluster matcher at index 0.
            // Note: zero cost edges as this is just a dummy link to facilitate search from a common start node.
            for (Vertex v : vertices[0]) {
                DefaultWeightedEdge edge = graph.addEdge(source, v);
                graph.setEdgeWeight(edge, 0.0);
            }
            // Similarly, all vertices that wrap the sequences detected by the last cluster matcher of the signature
            // are connected to the sink node.
            for (Vertex v : vertices[vertices.length-1]) {
                DefaultWeightedEdge edge = graph.addEdge(v, sink);
                graph.setEdgeWeight(edge, 0.0);
            }
            // Now link sequences detected by the cluster matcher at index i to sequences detected by the cluster
            // matcher at index i+1 if they obey the timestamp constraint (i.e., that the latter is later in time than
            // the former).
            for (int i = 0; i < vertices.length; i++) {
                int j = i + 1;
                if (j < vertices.length) {
                    for (Vertex iv : vertices[i]) {
                        PcapPacket ivLast = iv.sequence.get(iv.sequence.size()-1);
                        for (Vertex jv : vertices[j]) {
                            PcapPacket jvFirst = jv.sequence.get(jv.sequence.size()-1);
                            if (ivLast.getTimestamp().isBefore(jvFirst.getTimestamp())) {
                                DefaultWeightedEdge edge = graph.addEdge(iv, jv);
                                // The weight is the duration of the i'th sequence plus the duration between the i'th
                                // and i+1'th sequence.
                                Duration d = Duration.
                                        between(iv.sequence.get(0).getTimestamp(), jvFirst.getTimestamp());
                                // Unfortunately weights are double values, so must convert from long to double.
                                // TODO: need nano second precision? If so, use d.toNanos().
                                // TODO: risk of overflow when converting from long to double..?
                                graph.setEdgeWeight(edge, Long.valueOf(d.toMillis()).doubleValue());
                            }
                            // Alternative version if we cannot assume that sequences are ordered by timestamp:
//                            if (iv.sequence.stream().max(Comparator.comparing(PcapPacket::getTimestamp)).get()
//                                    .getTimestamp().isBefore(jv.sequence.stream().min(
//                                            Comparator.comparing(PcapPacket::getTimestamp)).get().getTimestamp())) {
//
//                            }
                        }
                    }
                }
            }
            // Graph construction complete, run shortest-path to find a (potential) signature match.
            DijkstraShortestPath<Vertex, DefaultWeightedEdge> dijkstra = new DijkstraShortestPath<>(graph);
            GraphPath<Vertex, DefaultWeightedEdge> shortestPath = dijkstra.getPath(source, sink);
            if (shortestPath != null) {
                // The total weight is the duration between the first packet of the first sequence and the last packet
                // of the last sequence, so we simply have to compare the weight against the timeframe that we allow
                // the signature to span. For now we just use the inclusion window we defined for training purposes.
                // Note however, that we must convert back from double to long as the weight is stored as a double in
                // JGraphT's API.
                if (((long)shortestPath.getWeight()) < mInclusionTimeMillis) {
                    // There's a signature match!
                    // Extract the match from the vertices
                    List<List<PcapPacket>> signatureMatch = new ArrayList<>();
                    for(Vertex v : shortestPath.getVertexList()) {
                        if (v == source || v == sink) {
                            // Skip the dummy source and sink nodes.
                            continue;
                        }
                        signatureMatch.add(v.sequence);
                        // As there is a one-to-one correspondence between vertices[] and pendingMatches[], we know that
                        // the sequence we've "consumed" for index i of the matched signature is also at index i in
                        // pendingMatches. We must remove it from pendingMatches so that we don't use it to construct
                        // another signature match in a later call.
                        mPendingMatches[signatureMatch.size()-1].remove(v.sequence);
                    }
                    // Declare success: notify observers
                    mObservers.forEach(obs -> obs.onSignatureDetected(mSignature,
                            Collections.unmodifiableList(signatureMatch)));
                }
            }
        }
    }

    /**
     * Encapsulates a {@code List<PcapPacket>} so as to allow the list to be used as a vertex in a graph while avoiding
     * the expensive {@link AbstractList#equals(Object)} calls when adding vertices to the graph.
     * Using this wrapper makes the incurred {@code equals(Object)} calls delegate to {@link Object#equals(Object)}
     * instead of {@link AbstractList#equals(Object)}. The net effect is a faster implementation, but the graph will not
     * recognize two lists that contain the same items--from a value and not reference point of view--as the same
     * vertex. However, this is fine for our purposes -- in fact restricting it to reference equality seems more
     * appropriate.
     */
    private static class Vertex {
        private final List<PcapPacket> sequence;
        private Vertex(List<PcapPacket> wrappedSequence) {
            sequence = wrappedSequence;
        }
    }
}