Home | History | Annotate | Download | only in safe_browsing
      1 // Copyright (c) 2012 The Chromium Authors. All rights reserved.
      2 // Use of this source code is governed by a BSD-style license that can be
      3 // found in the LICENSE file.
      4 //
      5 // This class handles the process of extracting all of the features from a
      6 // page and computing a phishyness score.  The basic steps are:
      7 //  - Run each feature extractor over the page, building up a FeatureMap of
      8 //    feature -> value.
      9 //  - SHA-256 hash all of the feature names in the map so that they match the
     10 //    supplied model.
     11 //  - Hand the hashed map off to a Scorer, which computes the probability that
     12 //    the page is phishy.
     13 //  - If the page is phishy, run the supplied callback.
     14 //
     15 // For more details, see phishing_*_feature_extractor.h, scorer.h, and
     16 // client_model.proto.
     17 
     18 #ifndef CHROME_RENDERER_SAFE_BROWSING_PHISHING_CLASSIFIER_H_
     19 #define CHROME_RENDERER_SAFE_BROWSING_PHISHING_CLASSIFIER_H_
     20 
     21 #include "base/basictypes.h"
     22 #include "base/callback.h"
     23 #include "base/memory/scoped_ptr.h"
     24 #include "base/memory/weak_ptr.h"
     25 #include "base/strings/string16.h"
     26 
     27 namespace content {
     28 class RenderView;
     29 }
     30 
     31 namespace safe_browsing {
     32 class ClientPhishingRequest;
     33 class FeatureExtractorClock;
     34 class FeatureMap;
     35 class PhishingDOMFeatureExtractor;
     36 class PhishingTermFeatureExtractor;
     37 class PhishingUrlFeatureExtractor;
     38 class Scorer;
     39 
     40 class PhishingClassifier {
     41  public:
     42   // Callback to be run when phishing classification finishes. The verdict
     43   // is a ClientPhishingRequest which contains the verdict computed by the
     44   // classifier as well as the extracted features.  If the verdict.is_phishing()
     45   // is true, the page is considered phishy by the client-side model,
     46   // and the browser should ping back to get a final verdict.  The
     47   // verdict.client_score() is set to kInvalidScore if classification failed.
     48   typedef base::Callback<void(const ClientPhishingRequest& /* verdict */)>
     49       DoneCallback;
     50 
     51   static const float kInvalidScore;
     52 
     53   // Creates a new PhishingClassifier object that will operate on
     54   // |render_view|.  |clock| is used to time feature extractor operations, and
     55   // the PhishingClassifier takes ownership of this object.  Note that the
     56   // classifier will not be 'ready' until set_phishing_scorer() is called.
     57   PhishingClassifier(content::RenderView* render_view,
     58                      FeatureExtractorClock* clock);
     59   virtual ~PhishingClassifier();
     60 
     61   // Sets a scorer for the classifier to use in computing the phishiness score.
     62   // This must live at least as long as the PhishingClassifier.  The caller is
     63   // expected to cancel any pending classification before setting a phishing
     64   // scorer.
     65   void set_phishing_scorer(const Scorer* scorer);
     66 
     67   // Returns true if the classifier is ready to classify pages, i.e. it
     68   // has had a scorer set via set_phishing_scorer().
     69   bool is_ready() const;
     70 
     71   // Called by the RenderView when a page has finished loading.  This begins
     72   // the feature extraction and scoring process. |page_text| should contain
     73   // the plain text of a web page, including any subframes, as returned by
     74   // RenderView::CaptureText().  |page_text| is owned by the caller, and must
     75   // not be destroyed until either |done_callback| is run or
     76   // CancelPendingClassification() is called.
     77   //
     78   // To avoid blocking the render thread for too long, phishing classification
     79   // may run in several chunks of work, posting a task to the current
     80   // MessageLoop to continue processing.  Once the scoring process is complete,
     81   // |done_callback| is run on the current thread.  PhishingClassifier takes
     82   // ownership of the callback.
     83   //
     84   // It is an error to call BeginClassification if the classifier is not yet
     85   // ready.
     86   virtual void BeginClassification(const string16* page_text,
     87                                    const DoneCallback& callback);
     88 
     89   // Called by the RenderView (on the render thread) when a page is unloading
     90   // or the RenderView is being destroyed.  This cancels any extraction that
     91   // is in progress.  It is an error to call CancelPendingClassification if
     92   // the classifier is not yet ready.
     93   virtual void CancelPendingClassification();
     94 
     95  private:
     96   // Any score equal to or above this value is considered phishy.
     97   static const float kPhishyThreshold;
     98 
     99   // Begins the feature extraction process, by extracting URL features and
    100   // beginning DOM feature extraction.
    101   void BeginFeatureExtraction();
    102 
    103   // Callback to be run when DOM feature extraction is complete.
    104   // If it was successful, begins term feature extraction, otherwise
    105   // runs the DoneCallback with a non-phishy verdict.
    106   void DOMExtractionFinished(bool success);
    107 
    108   // Callback to be run when term feature extraction is complete.
    109   // If it was successful, computes a score and runs the DoneCallback.
    110   // If extraction was unsuccessful, runs the DoneCallback with a
    111   // non-phishy verdict.
    112   void TermExtractionFinished(bool success);
    113 
    114   // Helper to verify that there is no pending phishing classification.  Dies
    115   // in debug builds if the state is not as expected.  This is a no-op in
    116   // release builds.
    117   void CheckNoPendingClassification();
    118 
    119   // Helper method to run the DoneCallback and clear the state.
    120   void RunCallback(const ClientPhishingRequest& verdict);
    121 
    122   // Helper to run the DoneCallback when feature extraction has failed.
    123   // This always signals a non-phishy verdict for the page, with kInvalidScore.
    124   void RunFailureCallback();
    125 
    126   // Clears the current state of the PhishingClassifier.
    127   void Clear();
    128 
    129   content::RenderView* render_view_;  // owns us
    130   const Scorer* scorer_;  // owned by the caller
    131   scoped_ptr<FeatureExtractorClock> clock_;
    132   scoped_ptr<PhishingUrlFeatureExtractor> url_extractor_;
    133   scoped_ptr<PhishingDOMFeatureExtractor> dom_extractor_;
    134   scoped_ptr<PhishingTermFeatureExtractor> term_extractor_;
    135 
    136   // State for any in-progress extraction.
    137   scoped_ptr<FeatureMap> features_;
    138   const string16* page_text_;  // owned by the caller
    139   DoneCallback done_callback_;
    140 
    141   // Used in scheduling BeginFeatureExtraction tasks.
    142   // These pointers are invalidated if classification is cancelled.
    143   base::WeakPtrFactory<PhishingClassifier> weak_factory_;
    144 
    145   DISALLOW_COPY_AND_ASSIGN(PhishingClassifier);
    146 };
    147 
    148 }  // namespace safe_browsing
    149 
    150 #endif  // CHROME_RENDERER_SAFE_BROWSING_PHISHING_CLASSIFIER_H_
    151