Code/Projects/PacketLevelSignatureExtractor/src/main/java/edu/uci/iotproject/detection/layer3/Layer3ClusterMatcher.java

   1 package edu.uci.iotproject.detection.layer3;
   2
   3 import edu.uci.iotproject.detection.AbstractClusterMatcher;
   4 import edu.uci.iotproject.detection.ClusterMatcherObserver;
   5 import edu.uci.iotproject.trafficreassembly.layer3.Conversation;
   6 import edu.uci.iotproject.trafficreassembly.layer3.TcpReassembler;
   7 import edu.uci.iotproject.analysis.TcpConversationUtils;
   8 import edu.uci.iotproject.io.PcapHandleReader;
   9 import edu.uci.iotproject.util.PrintUtils;
  10 import org.pcap4j.core.*;
  11
  12 import java.time.ZoneId;
  13 import java.util.*;
  14 import java.util.stream.Collectors;
  15
  16 import static edu.uci.iotproject.util.PcapPacketUtils.*;
  17
  18 /**
  19  * Searches a traffic trace for sequences of packets "belong to" a given cluster (in other words, attempts to classify
  20  * traffic as pertaining to a given cluster).
  21  *
  22  * @author Janus Varmarken {@literal <jvarmark@uci.edu>}
  23  * @author Rahmadi Trimananda {@literal <rtrimana@uci.edu>}
  24  */
  25 public class Layer3ClusterMatcher extends AbstractClusterMatcher implements PacketListener {
  26
  27     // Test client
  28     public static void main(String[] args) throws PcapNativeException, NotOpenException {
  29
  30 //        String path = "/scratch/July-2018"; // Rahmadi
  31 //        String path = "/Users/varmarken/temp/UCI IoT Project/experiments"; // Janus
  32 //        final String inputPcapFile = path + "/2018-07/dlink/dlink.wlan1.local.pcap";
  33 //        final String signatureFile = path + "/2018-07/dlink/offSignature1.sig";
  34 //
  35 //        List<List<PcapPacket>> signature = PrintUtils.deserializeClustersFromFile(signatureFile);
  36 //        Layer3ClusterMatcher clusterMatcher = new Layer3ClusterMatcher(signature, null,
  37 //                (sig, match) -> System.out.println(
  38 //                        String.format("[ !!! SIGNATURE DETECTED AT %s !!! ]",
  39 //                                match.get(0).getTimestamp().atZone(ZoneId.of("America/Los_Angeles")))
  40 //                )
  41 //        );
  42 //
  43 //        PcapHandle handle;
  44 //        try {
  45 //            handle = Pcaps.openOffline(inputPcapFile, PcapHandle.TimestampPrecision.NANO);
  46 //        } catch (PcapNativeException pne) {
  47 //            handle = Pcaps.openOffline(inputPcapFile);
  48 //        }
  49 //        PcapHandleReader reader = new PcapHandleReader(handle, p -> true, clusterMatcher);
  50 //        reader.readFromHandle();
  51 //        clusterMatcher.performDetection();
  52     }
  53
  54     /**
  55      * The ordered directions of packets in the sequences that make up {@link #mCluster}.
  56      */
  57     private final Conversation.Direction[] mClusterMemberDirections;
  58
  59     /**
  60      * For reassembling the observed traffic into TCP connections.
  61      */
  62     private final TcpReassembler mTcpReassembler = new TcpReassembler();
  63
  64     /**
  65      * IP of the router's WAN port (if analyzed traffic is captured at the ISP's point of view).
  66      */
  67     private final String mRouterWanIp;
  68
  69     /**
  70      * Epsilon value used by the DBSCAN algorithm; it is used again for range-based matching here.
  71      */
  72     private final double mEps;
  73
  74     /**
  75      * Create a {@link Layer3ClusterMatcher}.
  76      * @param cluster The cluster that traffic is matched against.
  77      * @param routerWanIp The router's WAN IP if examining traffic captured at the ISP's point of view (used for
  78      *                    determining the direction of packets).
  79      * @param eps The epsilon value used in the DBSCAN algorithm.
  80      * @param isRangeBased The boolean that decides if it is range-based vs. strict matching.
  81      * @param detectionObservers Client code that wants to get notified whenever the {@link Layer3ClusterMatcher} detects that
  82      *                          (a subset of) the examined traffic is similar to the traffic that makes up
  83      *                          {@code cluster}, i.e., when the examined traffic is classified as pertaining to
  84      *                          {@code cluster}.
  85      */
  86     public Layer3ClusterMatcher(List<List<PcapPacket>> cluster, String routerWanIp, boolean isRangeBased, double eps,
  87                                 ClusterMatcherObserver... detectionObservers) {
  88         super(cluster, isRangeBased);
  89         Objects.requireNonNull(detectionObservers, "detectionObservers cannot be null");
  90         for (ClusterMatcherObserver obs : detectionObservers) {
  91             addObserver(obs);
  92         }
  93         // Build the cluster members' direction sequence.
  94         // Note: assumes that the provided cluster was captured within the local network (routerWanIp is set to null).
  95         mClusterMemberDirections = getPacketDirections(cluster.get(0), null);
  96         /*
  97          * Enforce restriction on cluster members: all representatives must exhibit the same direction pattern and
  98          * contain the same number of packets. Note that this is a somewhat heavy operation, so it may be disabled later
  99          * on in favor of performance. However, it is only run once (at instantiation), so the overhead may be warranted
 100          * in order to ensure correctness, especially during the development/debugging phase.
 101          */
 102         if (!isRangeBased) {    // Only when it is not range-based
 103             if (mCluster.stream().
 104                     anyMatch(inner -> !Arrays.equals(mClusterMemberDirections, getPacketDirections(inner, null)))) {
 105                 throw new IllegalArgumentException(
 106                         "cluster members must contain the same number of packets and exhibit the same packet direction " +
 107                                 "pattern"
 108                 );
 109             }
 110         }
 111         mEps = eps;
 112         mRouterWanIp = routerWanIp;
 113     }
 114
 115     @Override
 116     public void gotPacket(PcapPacket packet) {
 117         // Present packet to TCP reassembler so that it can be mapped to a connection (if it is a TCP packet).
 118         mTcpReassembler.gotPacket(packet);
 119     }
 120
 121     /**
 122      * Get the cluster that describes the packet sequence that this {@link Layer3ClusterMatcher} is searching for.
 123      * @return the cluster that describes the packet sequence that this {@link Layer3ClusterMatcher} is searching for.
 124      */
 125     public List<List<PcapPacket>> getCluster() {
 126         return mCluster;
 127     }
 128
 129     public void performDetectionRangeBased() {
 130         /*
 131          * Let's start out simple by building a version that only works for signatures that do not span across multiple
 132          * TCP conversations...
 133          */
 134         for (Conversation c : mTcpReassembler.getTcpConversations()) {
 135             if (c.isTls() && c.getTlsApplicationDataPackets().isEmpty() || !c.isTls() && c.getPackets().isEmpty()) {
 136                 // Skip empty conversations.
 137                 continue;
 138             }
 139             List<PcapPacket> lowerBound = mCluster.get(0);
 140             List<PcapPacket> upperBound = mCluster.get(1);
 141             if (isTlsSequence(lowerBound) != c.isTls() || isTlsSequence(upperBound) != c.isTls()) {
 142                 // We consider it a mismatch if one is a TLS application data sequence and the other is not.
 143                 continue;
 144             }
 145             // Fetch set of packets to examine based on TLS or not.
 146             List<PcapPacket> cPkts = c.isTls() ? c.getTlsApplicationDataPackets() : c.getPackets();
 147             Optional<List<PcapPacket>> match;
 148             while ((match = findSubsequenceInSequence(lowerBound, upperBound, cPkts, mClusterMemberDirections, null)).
 149                     isPresent()) {
 150                 List<PcapPacket> matchSeq = match.get();
 151                 // Notify observers about the match.
 152                 mObservers.forEach(o -> o.onMatch(Layer3ClusterMatcher.this, matchSeq));
 153                 /*
 154                  * Get the index in cPkts of the last packet in the sequence of packets that matches the searched
 155                  * signature sequence.
 156                  */
 157                 int matchSeqEndIdx = cPkts.indexOf(matchSeq.get(matchSeq.size() - 1));
 158                 // We restart the search for the signature sequence immediately after that index, so truncate cPkts.
 159                 cPkts = cPkts.stream().skip(matchSeqEndIdx + 1).collect(Collectors.toList());
 160             }
 161         }
 162     }
 163
 164     public void performDetectionConservative() {
 165         /*
 166          * Let's start out simple by building a version that only works for signatures that do not span across multiple
 167          * TCP conversations...
 168          */
 169         for (Conversation c : mTcpReassembler.getTcpConversations()) {
 170             if (c.isTls() && c.getTlsApplicationDataPackets().isEmpty() || !c.isTls() && c.getPackets().isEmpty()) {
 171                 // Skip empty conversations.
 172                 continue;
 173             }
 174             for (List<PcapPacket> signatureSequence : mCluster) {
 175                 if (isTlsSequence(signatureSequence) != c.isTls()) {
 176                     // We consider it a mismatch if one is a TLS application data sequence and the other is not.
 177                     continue;
 178                 }
 179                 // Fetch set of packets to examine based on TLS or not.
 180                 List<PcapPacket> cPkts = c.isTls() ? c.getTlsApplicationDataPackets() : c.getPackets();
 181                 /*
 182                  * Note: we embed the attempt to detect the signature sequence in a loop in order to capture those cases
 183                  * where the same signature sequence appears multiple times in one Conversation.
 184                  *
 185                  * Note: since we expect all sequences that together make up the signature to exhibit the same direction
 186                  * pattern, we can simply pass the precomputed direction array for the signature sequence so that it
 187                  * won't have to be recomputed internally in each call to findSubsequenceInSequence().
 188                  */
 189                 Optional<List<PcapPacket>> match;
 190                 while ((match = findSubsequenceInSequence(signatureSequence, cPkts, mClusterMemberDirections, null)).
 191                         isPresent()) {
 192                     List<PcapPacket> matchSeq = match.get();
 193                     // Notify observers about the match.
 194                     mObservers.forEach(o -> o.onMatch(Layer3ClusterMatcher.this, matchSeq));
 195                     /*
 196                      * Get the index in cPkts of the last packet in the sequence of packets that matches the searched
 197                      * signature sequence.
 198                      */
 199                     int matchSeqEndIdx = cPkts.indexOf(matchSeq.get(matchSeq.size() - 1));
 200                     // We restart the search for the signature sequence immediately after that index, so truncate cPkts.
 201                     cPkts = cPkts.stream().skip(matchSeqEndIdx + 1).collect(Collectors.toList());
 202                 }
 203             }
 204
 205             /*
 206              * TODO:
 207              * if no item in cluster matches, also perform a distance-based matching to cover those cases where we did
 208              * not manage to capture every single mutation of the sequence during training.
 209              *
 210              * Need to compute average/centroid of cluster to do so...? Compute within-cluster variance, then check if
 211              * distance between input conversation and cluster average/centroid is smaller than or equal to the computed
 212              * variance?
 213              */
 214         }
 215     }
 216
 217     /**
 218      * Checks if {@code sequence} is a sequence of TLS packets. Note: the current implementation relies on inspection
 219      * of the port numbers when deciding between TLS vs. non-TLS. Therefore, only the first packet of {@code sequence}
 220      * is examined as it is assumed that all packets in {@code sequence} pertain to the same {@link Conversation} and
 221      * hence share the same set of two src/dst port numbers (albeit possibly alternating between which one is the src
 222      * and which one is the dst, as packets in {@code sequence} may be in alternating directions).
 223      * @param sequence The sequence of packets for which it is to be determined if it is a sequence of TLS packets or
 224      *                 non-TLS packets.
 225      * @return {@code true} if {@code sequence} is a sequence of TLS packets, {@code false} otherwise.
 226      */
 227     private boolean isTlsSequence(List<PcapPacket> sequence) {
 228         // NOTE: Assumes ALL packets in sequence pertain to the same TCP connection!
 229         PcapPacket firstPkt = sequence.get(0);
 230         int srcPort = getSourcePort(firstPkt);
 231         int dstPort = getDestinationPort(firstPkt);
 232         return TcpConversationUtils.isTlsPort(srcPort) || TcpConversationUtils.isTlsPort(dstPort);
 233     }
 234
 235     /**
 236      * Examine if a given sequence of packets ({@code sequence}) contains a given shorter sequence of packets
 237      * ({@code subsequence}). Note: the current implementation actually searches for a substring as it does not allow
 238      * for interleaving packets in {@code sequence} that are not in {@code subsequence}; for example, if
 239      * {@code subsequence} consists of packet lengths [2, 3, 5] and {@code sequence} consists of  packet lengths
 240      * [2, 3, 4, 5], the result will be that there is no match (because of the interleaving 4). If we are to allow
 241      * interleaving packets, we need a modified version of
 242      * <a href="https://stackoverflow.com/a/20545604/1214974">this</a>.
 243      *
 244      * @param subsequence The sequence to search for.
 245      * @param sequence The sequence to search.
 246      * @param subsequenceDirections The directions of packets in {@code subsequence} such that for all {@code i},
 247      *                              {@code subsequenceDirections[i]} is the direction of the packet returned by
 248      *                              {@code subsequence.get(i)}. May be set to {@code null}, in which this call will
 249      *                              internally compute the packet directions.
 250      * @param sequenceDirections The directions of packets in {@code sequence} such that for all {@code i},
 251      *                           {@code sequenceDirections[i]} is the direction of the packet returned by
 252      *                           {@code sequence.get(i)}. May be set to {@code null}, in which this call will internally
 253      *                           compute the packet directions.
 254      *
 255      * @return An {@link Optional} containing the part of {@code sequence} that matches {@code subsequence}, or an empty
 256      *         {@link Optional} if no part of {@code sequence} matches {@code subsequence}.
 257      */
 258     private Optional<List<PcapPacket>> findSubsequenceInSequence(List<PcapPacket> subsequence,
 259                                                                  List<PcapPacket> sequence,
 260                                                                  Conversation.Direction[] subsequenceDirections,
 261                                                                  Conversation.Direction[] sequenceDirections) {
 262         if (sequence.size() < subsequence.size()) {
 263             // If subsequence is longer, it cannot be contained in sequence.
 264             return Optional.empty();
 265         }
 266         if (isTlsSequence(subsequence) != isTlsSequence(sequence)) {
 267             // We consider it a mismatch if one is a TLS application data sequence and the other is not.
 268             return Optional.empty();
 269         }
 270         // If packet directions have not been precomputed by calling code, we need to construct them.
 271         if (subsequenceDirections == null) {
 272             subsequenceDirections = getPacketDirections(subsequence, mRouterWanIp);
 273         }
 274         if (sequenceDirections == null) {
 275             sequenceDirections = getPacketDirections(sequence, mRouterWanIp);
 276         }
 277         int subseqIdx = 0;
 278         int seqIdx = 0;
 279         while (seqIdx < sequence.size()) {
 280             PcapPacket subseqPkt = subsequence.get(subseqIdx);
 281             PcapPacket seqPkt = sequence.get(seqIdx);
 282             // We only have a match if packet lengths and directions match.
 283             if (subseqPkt.getOriginalLength() == seqPkt.getOriginalLength() &&
 284                     subsequenceDirections[subseqIdx] == sequenceDirections[seqIdx]) {
 285                 // A match; advance both indices to consider next packet in subsequence vs. next packet in sequence.
 286                 subseqIdx++;
 287                 seqIdx++;
 288                 if (subseqIdx == subsequence.size()) {
 289                     // We managed to match the entire subsequence in sequence.
 290                     // Return the sublist of sequence that matches subsequence.
 291                     /*
 292                      * TODO:
 293                      * ASSUMES THE BACKING LIST (i.e., 'sequence') IS _NOT_ STRUCTURALLY MODIFIED, hence may not work
 294                      * for live traces!
 295                      */
 296                     return Optional.of(sequence.subList(seqIdx - subsequence.size(), seqIdx));
 297                 }
 298             } else {
 299                 // Mismatch.
 300                 if (subseqIdx > 0) {
 301                     /*
 302                      * If we managed to match parts of subsequence, we restart the search for subsequence in sequence at
 303                      * the index of sequence where the current mismatch occurred. I.e., we must reset subseqIdx, but
 304                      * leave seqIdx untouched.
 305                      */
 306                     subseqIdx = 0;
 307                 } else {
 308                     /*
 309                      * First packet of subsequence didn't match packet at seqIdx of sequence, so we move forward in
 310                      * sequence, i.e., we continue the search for subsequence in sequence starting at index seqIdx+1 of
 311                      * sequence.
 312                      */
 313                     seqIdx++;
 314                 }
 315             }
 316         }
 317         return Optional.empty();
 318     }
 319
 320     /**
 321      * Overloading the method {@code findSubsequenceInSequence} for range-based matching. Instead of a sequence,
 322      * we have sequences of lower and upper bounds.
 323      *
 324      * @param lowerBound The lower bound of the sequence we search for.
 325      * @param upperBound The upper bound of the sequence we search for.
 326      * @param subsequenceDirections The directions of packets in {@code subsequence} such that for all {@code i},
 327      *                              {@code subsequenceDirections[i]} is the direction of the packet returned by
 328      *                              {@code subsequence.get(i)}. May be set to {@code null}, in which this call will
 329      *                              internally compute the packet directions.
 330      * @param sequenceDirections The directions of packets in {@code sequence} such that for all {@code i},
 331      *                           {@code sequenceDirections[i]} is the direction of the packet returned by
 332      *                           {@code sequence.get(i)}. May be set to {@code null}, in which this call will internally
 333      *                           compute the packet directions.
 334      *
 335      * @return An {@link Optional} containing the part of {@code sequence} that matches {@code subsequence}, or an empty
 336      *         {@link Optional} if no part of {@code sequence} matches {@code subsequence}.
 337      */
 338     private Optional<List<PcapPacket>> findSubsequenceInSequence(List<PcapPacket> lowerBound,
 339                                                                  List<PcapPacket> upperBound,
 340                                                                  List<PcapPacket> sequence,
 341                                                                  Conversation.Direction[] subsequenceDirections,
 342                                                                  Conversation.Direction[] sequenceDirections) {
 343         // Just do the checks for either lower or upper bound!
 344         // TODO: For now we use just the lower bound
 345         if (sequence.size() < lowerBound.size()) {
 346             // If subsequence is longer, it cannot be contained in sequence.
 347             return Optional.empty();
 348         }
 349         if (isTlsSequence(lowerBound) != isTlsSequence(sequence)) {
 350             // We consider it a mismatch if one is a TLS application data sequence and the other is not.
 351             return Optional.empty();
 352         }
 353         // If packet directions have not been precomputed by calling code, we need to construct them.
 354         if (subsequenceDirections == null) {
 355             subsequenceDirections = getPacketDirections(lowerBound, mRouterWanIp);
 356         }
 357         if (sequenceDirections == null) {
 358             sequenceDirections = getPacketDirections(sequence, mRouterWanIp);
 359         }
 360         int subseqIdx = 0;
 361         int seqIdx = 0;
 362         while (seqIdx < sequence.size()) {
 363             PcapPacket lowBndPkt = lowerBound.get(subseqIdx);
 364             PcapPacket upBndPkt = upperBound.get(subseqIdx);
 365             PcapPacket seqPkt = sequence.get(seqIdx);
 366             // We only have a match if packet lengths and directions match.
 367             // The packet lengths have to be in the range of [lowerBound - eps, upperBound+eps]
 368             // TODO: Maybe we could do better here for the double to integer conversion?
 369             int epsLowerBound = lowBndPkt.length() - (int) mEps;
 370             int epsUpperBound = upBndPkt.length() + (int) mEps;
 371             if (epsLowerBound <= seqPkt.getOriginalLength() &&
 372                     seqPkt.getOriginalLength() <= epsUpperBound &&
 373                     subsequenceDirections[subseqIdx] == sequenceDirections[seqIdx]) {
 374                 // A match; advance both indices to consider next packet in subsequence vs. next packet in sequence.
 375                 subseqIdx++;
 376                 seqIdx++;
 377                 if (subseqIdx == lowerBound.size()) {
 378                     // We managed to match the entire subsequence in sequence.
 379                     // Return the sublist of sequence that matches subsequence.
 380                     /*
 381                      * TODO:
 382                      * ASSUMES THE BACKING LIST (i.e., 'sequence') IS _NOT_ STRUCTURALLY MODIFIED, hence may not work
 383                      * for live traces!
 384                      */
 385                     return Optional.of(sequence.subList(seqIdx - lowerBound.size(), seqIdx));
 386                 }
 387             } else {
 388                 // Mismatch.
 389                 if (subseqIdx > 0) {
 390                     /*
 391                      * If we managed to match parts of subsequence, we restart the search for subsequence in sequence at
 392                      * the index of sequence where the current mismatch occurred. I.e., we must reset subseqIdx, but
 393                      * leave seqIdx untouched.
 394                      */
 395                     subseqIdx = 0;
 396                 } else {
 397                     /*
 398                      * First packet of subsequence didn't match packet at seqIdx of sequence, so we move forward in
 399                      * sequence, i.e., we continue the search for subsequence in sequence starting at index seqIdx+1 of
 400                      * sequence.
 401                      */
 402                     seqIdx++;
 403                 }
 404             }
 405         }
 406         return Optional.empty();
 407     }
 408
 409     /**
 410      * Given a cluster, produces a pruned version of that cluster. In the pruned version, there are no duplicate cluster
 411      * members. Two cluster members are considered identical if their packets lengths and packet directions are
 412      * identical. The resulting pruned cluster is unmodifiable (this applies to both the outermost list as well as the
 413      * nested lists) in order to preserve its integrity when exposed to external code (e.g., through
 414      * {@link #getCluster()}).
 415      *
 416      * @param cluster A cluster to prune.
 417      * @return The resulting pruned cluster.
 418      */
 419     @Override
 420     protected List<List<PcapPacket>> pruneCluster(List<List<PcapPacket>> cluster) {
 421         List<List<PcapPacket>> prunedCluster = new ArrayList<>();
 422         for (List<PcapPacket> originalClusterSeq : cluster) {
 423             boolean alreadyPresent = false;
 424             for (List<PcapPacket> prunedClusterSeq : prunedCluster) {
 425                 Optional<List<PcapPacket>> duplicate = findSubsequenceInSequence(originalClusterSeq, prunedClusterSeq,
 426                         mClusterMemberDirections, mClusterMemberDirections);
 427                 if (duplicate.isPresent()) {
 428                     alreadyPresent = true;
 429                     break;
 430                 }
 431             }
 432             if (!alreadyPresent) {
 433                 prunedCluster.add(Collections.unmodifiableList(originalClusterSeq));
 434             }
 435         }
 436         return Collections.unmodifiableList(prunedCluster);
 437     }
 438
 439     /**
 440      * Given a {@code List<PcapPacket>}, generate a {@code Conversation.Direction[]} such that each entry in the
 441      * resulting {@code Conversation.Direction[]} specifies the direction of the {@link PcapPacket} at the corresponding
 442      * index in the input list.
 443      * @param packets The list of packets for which to construct a corresponding array of packet directions.
 444      * @param routerWanIp The IP of the router's WAN port. This is used for determining the direction of packets when
 445      *                    the traffic is captured just outside the local network (at the ISP side of the router). Set to
 446      *                    {@code null} if {@code packets} stem from traffic captured within the local network.
 447      * @return A {@code Conversation.Direction[]} specifying the direction of the {@link PcapPacket} at the
 448      *         corresponding index in {@code packets}.
 449      */
 450     private static Conversation.Direction[] getPacketDirections(List<PcapPacket> packets, String routerWanIp) {
 451         Conversation.Direction[] directions = new Conversation.Direction[packets.size()];
 452         for (int i = 0; i < packets.size(); i++) {
 453             PcapPacket pkt = packets.get(i);
 454             if (getSourceIp(pkt).equals(getDestinationIp(pkt))) {
 455                 // Sanity check: we shouldn't be processing loopback traffic
 456                 throw new AssertionError("loopback traffic detected");
 457             }
 458             if (isSrcIpLocal(pkt) || getSourceIp(pkt).equals(routerWanIp)) {
 459                 directions[i] = Conversation.Direction.CLIENT_TO_SERVER;
 460             } else if (isDstIpLocal(pkt) || getDestinationIp(pkt).equals(routerWanIp)) {
 461                 directions[i] = Conversation.Direction.SERVER_TO_CLIENT;
 462             } else {
 463                 //throw new IllegalArgumentException("no local IP or router WAN port IP found, can't detect direction");
 464             }
 465         }
 466         return directions;
 467     }
 468
 469 }