Separate pre-classification checks for client-side malware and phishing

chrome/browser/safe_browsing/client_side_detection_host.cc

 // Copyright (c) 2012 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "chrome/browser/safe_browsing/client_side_detection_host.h"
 
 #include <vector>
 
 #include "base/logging.h"
 #include "base/memory/ref_counted.h"
 #include "base/memory/scoped_ptr.h"
 #include "base/metrics/histogram.h"
 #include "base/prefs/pref_service.h"
 #include "base/sequenced_task_runner_helpers.h"
 #include "base/strings/utf_string_conversions.h"
 #include "chrome/browser/browser_process.h"
 #include "chrome/browser/profiles/profile.h"
 #include "chrome/browser/safe_browsing/browser_feature_extractor.h"
 #include "chrome/browser/safe_browsing/client_side_detection_service.h"
 #include "chrome/browser/safe_browsing/database_manager.h"
 #include "chrome/browser/safe_browsing/safe_browsing_service.h"
 #include "chrome/common/chrome_switches.h"

[mattm, 2014/03/18 02:19:06]

unused?

[noé, 2014/03/20 17:01:45]

Done.

 #include "chrome/common/chrome_version_info.h"

[mattm, 2014/03/18 02:19:06]

unused?

[noé, 2014/03/20 17:01:45]

Done.

 #include "chrome/common/pref_names.h"
 #include "chrome/common/safe_browsing/csd.pb.h"
 #include "chrome/common/safe_browsing/safebrowsing_messages.h"
 #include "content/public/browser/browser_thread.h"
 #include "content/public/browser/navigation_controller.h"
 #include "content/public/browser/navigation_details.h"
 #include "content/public/browser/navigation_entry.h"
 #include "content/public/browser/notification_details.h"
 #include "content/public/browser/notification_source.h"
 #include "content/public/browser/notification_types.h"
 #include "content/public/browser/render_process_host.h"
 #include "content/public/browser/render_view_host.h"
 #include "content/public/browser/resource_request_details.h"
 #include "content/public/browser/web_contents.h"
 #include "content/public/common/frame_navigate_params.h"
 #include "url/gurl.h"
 
 using content::BrowserThread;
 using content::NavigationEntry;
 using content::ResourceRequestDetails;
 using content::WebContents;
 
 namespace safe_browsing {
 
 const int ClientSideDetectionHost::kMaxUrlsPerIP = 20;
 const int ClientSideDetectionHost::kMaxIPsPerBrowse = 200;
 
 const char kSafeBrowsingMatchKey[] = "safe_browsing_match";
 
+typedef base::Callback<void(bool)> ShouldClassifyUrlCallback;
+
 // This class is instantiated each time a new toplevel URL loads, and
-// asynchronously checks whether the phishing classifier should run for this
-// URL.  If so, it notifies the renderer with a StartPhishingDetection IPC.
-// Objects of this class are ref-counted and will be destroyed once nobody
-// uses it anymore.  If |web_contents|, |csd_service| or |host| go away you need
-// to call Cancel().  We keep the |database_manager| alive in a ref pointer for
-// as long as it takes.
+// asynchronously checks whether the malware and phishing classifiers should run
+// for this URL.  If so, it notifies the host class by calling the provided
+// callback form the UI thread.  Objects of this class are ref-counted and will
+// be destroyed once nobody uses it anymore.  If |web_contents|, |csd_service|
+// or |host| go away you need to call Cancel().  We keep the |database_manager|
+// alive in a ref pointer for as long as it takes.
 class ClientSideDetectionHost::ShouldClassifyUrlRequest
     : public base::RefCountedThreadSafe<
           ClientSideDetectionHost::ShouldClassifyUrlRequest> {
  public:
-  ShouldClassifyUrlRequest(const content::FrameNavigateParams& params,
-                           WebContents* web_contents,
-                           ClientSideDetectionService* csd_service,
-                           SafeBrowsingDatabaseManager* database_manager,
-                           ClientSideDetectionHost* host)
-      : canceled_(false),
-        params_(params),
+  ShouldClassifyUrlRequest(
+      const content::FrameNavigateParams& params,
+      const ShouldClassifyUrlCallback& start_phishing_classification,
+      const ShouldClassifyUrlCallback& start_malware_classification,
+      WebContents* web_contents,
+      ClientSideDetectionService* csd_service,
+      SafeBrowsingDatabaseManager* database_manager,
+      ClientSideDetectionHost* host)
+      : params_(params),
         web_contents_(web_contents),
         csd_service_(csd_service),
         database_manager_(database_manager),
-        host_(host) {
+        host_(host),
+        start_phishing_classification_cb_(start_phishing_classification),
+        start_malware_classification_cb_(start_malware_classification) {
     DCHECK(BrowserThread::CurrentlyOn(BrowserThread::UI));
     DCHECK(web_contents_);
     DCHECK(csd_service_);
     DCHECK(database_manager_.get());
     DCHECK(host_);
   }
 
   void Start() {
     DCHECK(BrowserThread::CurrentlyOn(BrowserThread::UI));
 
     // We start by doing some simple checks that can run on the UI thread.
     UMA_HISTOGRAM_COUNTS("SBClientPhishing.ClassificationStart", 1);
+    UMA_HISTOGRAM_COUNTS("SBClientMalware.ClassificationStart", 1);
 
     // Only classify [X]HTML documents.
     if (params_.contents_mime_type != "text/html" &&
         params_.contents_mime_type != "application/xhtml+xml") {
       VLOG(1) << "Skipping phishing classification for URL: " << params_.url
               << " because it has an unsupported MIME type: "
               << params_.contents_mime_type;
-      UMA_HISTOGRAM_ENUMERATION("SBClientPhishing.PreClassificationCheckFail",
-                                NO_CLASSIFY_UNSUPPORTED_MIME_TYPE,
-                                NO_CLASSIFY_MAX);
-      return;
+      DontClassifyForPhishing(NO_CLASSIFY_UNSUPPORTED_MIME_TYPE);
     }
 
     if (csd_service_->IsPrivateIPAddress(params_.socket_address.host())) {
       VLOG(1) << "Skipping phishing classification for URL: " << params_.url
               << " because of hosting on private IP: "
               << params_.socket_address.host();
-      UMA_HISTOGRAM_ENUMERATION("SBClientPhishing.PreClassificationCheckFail",
-                                NO_CLASSIFY_PRIVATE_IP,
-                                NO_CLASSIFY_MAX);
-      return;
+      DontClassifyForPhishing(NO_CLASSIFY_PRIVATE_IP);
+      DontClassifyForMalware(NO_CLASSIFY_PRIVATE_IP);
     }
 
-    // Don't run the phishing classifier if the tab is incognito.
+    // Don't run any classifier if the tab is incognito.
     if (web_contents_->GetBrowserContext()->IsOffTheRecord()) {
-      VLOG(1) << "Skipping phishing classification for URL: " << params_.url
-              << " because we're browsing incognito.";
-      UMA_HISTOGRAM_ENUMERATION("SBClientPhishing.PreClassificationCheckFail",
-                                NO_CLASSIFY_OFF_THE_RECORD,
-                                NO_CLASSIFY_MAX);
-
-      return;
+      VLOG(1) << "Skipping phishing and malware classification for URL: "
+              << params_.url << " because we're browsing incognito.";
+      DontClassifyForPhishing(NO_CLASSIFY_OFF_THE_RECORD);
+      DontClassifyForMalware(NO_CLASSIFY_OFF_THE_RECORD);
     }
 
     // We lookup the csd-whitelist before we lookup the cache because
     // a URL may have recently been whitelisted.  If the URL matches
-    // the csd-whitelist we won't start classification.  The
+    // the csd-whitelist we won't phishing start classification.  The

[mattm, 2014/03/18 02:19:06]

word order?

[noé, 2014/03/20 17:01:45]

Done.

     // csd-whitelist check has to be done on the IO thread because it
     // uses the SafeBrowsing service class.
-    BrowserThread::PostTask(
-        BrowserThread::IO,
-        FROM_HERE,
-        base::Bind(&ShouldClassifyUrlRequest::CheckCsdWhitelist,
-                   this, params_.url));
+    if (MaybeClassifyForPhishing() || MaybeClassifyForMalware()) {
+      BrowserThread::PostTask(
+          BrowserThread::IO,
+          FROM_HERE,
+          base::Bind(&ShouldClassifyUrlRequest::CheckCsdWhitelist,
+                     this, params_.url));
+    }
   }
 
   void Cancel() {
-    canceled_ = true;
+    DontClassifyForPhishing(NO_CLASSIFY_CANCEL);
+    DontClassifyForMalware(NO_CLASSIFY_CANCEL);
     // Just to make sure we don't do anything stupid we reset all these
     // pointers except for the safebrowsing service class which may be
     // accessed by CheckCsdWhitelist().
     web_contents_ = NULL;
     csd_service_ = NULL;
     host_ = NULL;
   }
 
  private:
   friend class base::RefCountedThreadSafe<
       ClientSideDetectionHost::ShouldClassifyUrlRequest>;
 
   // Enum used to keep stats about why the pre-classification check failed.
   enum PreClassificationCheckFailures {
     OBSOLETE_NO_CLASSIFY_PROXY_FETCH,
     NO_CLASSIFY_PRIVATE_IP,
     NO_CLASSIFY_OFF_THE_RECORD,
     NO_CLASSIFY_MATCH_CSD_WHITELIST,
     NO_CLASSIFY_TOO_MANY_REPORTS,
     NO_CLASSIFY_UNSUPPORTED_MIME_TYPE,
+    NO_CLASSIFY_NO_DATABASE_MANAGER,
+    NO_CLASSIFY_KILLSWITCH,
+    NO_CLASSIFY_CANCEL,
+    NO_CLASSIFY_RESULT_FROM_CACHE,
 
     NO_CLASSIFY_MAX  // Always add new values before this one.
   };
 
   // The destructor can be called either from the UI or the IO thread.
   virtual ~ShouldClassifyUrlRequest() { }
 
+  bool MaybeClassifyForPhishing() const {

[mattm, 2014/03/18 02:19:06]

Calling these MaybeFoo is a bit confusing, usually that means the function
method might do something itself.

[noé, 2014/03/20 17:01:45]

Done.

+    DCHECK(BrowserThread::CurrentlyOn(BrowserThread::UI));
+    return !start_phishing_classification_cb_.is_null();
+  }
+
+  bool MaybeClassifyForMalware() const {
+    DCHECK(BrowserThread::CurrentlyOn(BrowserThread::UI));
+    return !start_malware_classification_cb_.is_null();
+  }
+
+  void DontClassifyForPhishing(PreClassificationCheckFailures reason) {
+    DCHECK(BrowserThread::CurrentlyOn(BrowserThread::UI));
+    if (MaybeClassifyForPhishing()) {
+      // Track the first reason why we stopped classifying for phishing.
+      UMA_HISTOGRAM_ENUMERATION("SBClientPhishing.PreClassificationCheckFail",
+                                reason, NO_CLASSIFY_MAX);
+      VLOG(2) << "Failed phishing pre-classification checks.  Reason: "

[mattm, 2014/03/18 02:19:06]

prefer DVLOG instead of VLOG, unless there is a reason for VLOG

[noé, 2014/03/20 17:01:45]

Done.

+              << reason;
+      start_phishing_classification_cb_.Run(false);
+    }
+    start_phishing_classification_cb_.Reset();
+  }
+
+  void DontClassifyForMalware(PreClassificationCheckFailures reason) {
+    DCHECK(BrowserThread::CurrentlyOn(BrowserThread::UI));
+    if (MaybeClassifyForMalware()) {
+      // Track the first reason why we stopped classifying for malware.
+      UMA_HISTOGRAM_ENUMERATION("SBClientMalware.PreClassificationCheckFail",
+                                reason, NO_CLASSIFY_MAX);
+      VLOG(2) << "Failed malware pre-classification checks.  Reason: "
+              << reason;
+      start_malware_classification_cb_.Run(false);
+    }
+    start_malware_classification_cb_.Reset();
+  }
+
   void CheckCsdWhitelist(const GURL& url) {

[mattm, 2014/03/18 02:19:06]

update name

[noé, 2014/03/20 17:01:45]

Done.

     DCHECK(BrowserThread::CurrentlyOn(BrowserThread::IO));
-    if (!database_manager_.get() ||
-        database_manager_->MatchCsdWhitelistUrl(url)) {
-      // We're done.  There is no point in going back to the UI thread.
-      VLOG(1) << "Skipping phishing classification for URL: " << url
-              << " because it matches the csd whitelist";
-      UMA_HISTOGRAM_ENUMERATION("SBClientPhishing.PreClassificationCheckFail",
-                                NO_CLASSIFY_MATCH_CSD_WHITELIST,
-                                NO_CLASSIFY_MAX);
-      return;
+    // We don't want to call the classification callbacks from the IO
+    // thread so we simply pass the results of this method to CheckCache()
+    // which is called on the UI thread;
+    PreClassificationCheckFailures phishing_reason = NO_CLASSIFY_MAX;
+    PreClassificationCheckFailures malware_reason = NO_CLASSIFY_MAX;
+    if (!database_manager_.get()) {
+      // We cannot check the Safe Browsing whitelists so we stop here
+      // for safety.
+      malware_reason = phishing_reason = NO_CLASSIFY_NO_DATABASE_MANAGER;
+    } else {
+      if (database_manager_->MatchCsdWhitelistUrl(url)) {
+        VLOG(1) << "Skipping phishing classification for URL: " << url
+                << " because it matches the csd whitelist";
+        phishing_reason = NO_CLASSIFY_MATCH_CSD_WHITELIST;
+      }
+      if (database_manager_->IsMalwareKillSwitchOn()) {
+        malware_reason = NO_CLASSIFY_KILLSWITCH;
+      }
     }
-
-    bool malware_killswitch_on = database_manager_->IsMalwareKillSwitchOn();
-
     BrowserThread::PostTask(
         BrowserThread::UI,
         FROM_HERE,
-        base::Bind(&ShouldClassifyUrlRequest::CheckCache, this,
-                   malware_killswitch_on));
+        base::Bind(&ShouldClassifyUrlRequest::CheckCache,
+                   this,
+                   phishing_reason,
+                   malware_reason));
   }
 
-  void CheckCache(bool malware_killswitch_on) {
+  void CheckCache(PreClassificationCheckFailures phishing_reason,
+                  PreClassificationCheckFailures malware_reason) {
     DCHECK(BrowserThread::CurrentlyOn(BrowserThread::UI));
-    if (canceled_) {
-      return;
+    if (phishing_reason != NO_CLASSIFY_MAX)
+      DontClassifyForPhishing(phishing_reason);
+    if (malware_reason != NO_CLASSIFY_MAX)
+      DontClassifyForMalware(malware_reason);
+    if (!MaybeClassifyForMalware() && !MaybeClassifyForPhishing()) {
+      return;  // No point in doing anything else.
     }
-
-    host_->SetMalwareKillSwitch(malware_killswitch_on);
-    // If result is cached, we don't want to run classification again
+    // If result is cached, we don't want to run classification again.
+    // In that case we're just trying to show the warning.
     bool is_phishing;
     if (csd_service_->GetValidCachedResult(params_.url, &is_phishing)) {
       VLOG(1) << "Satisfying request for " << params_.url << " from cache";
       UMA_HISTOGRAM_COUNTS("SBClientPhishing.RequestSatisfiedFromCache", 1);
       // Since we are already on the UI thread, this is safe.
       host_->MaybeShowPhishingWarning(params_.url, is_phishing);
-      return;
+      DontClassifyForPhishing(NO_CLASSIFY_RESULT_FROM_CACHE);
     }
 
     // We want to limit the number of requests, though we will ignore the
     // limit for urls in the cache.  We don't want to start classifying
     // too many pages as phishing, but for those that we already think are
-    // phishing we want to give ourselves a chance to fix false positives.
+    // phishing we want to send a request to the server to give ourselves
+    // a chance to fix misclassifications.
     if (csd_service_->IsInCache(params_.url)) {
       VLOG(1) << "Reporting limit skipped for " << params_.url
               << " as it was in the cache.";
       UMA_HISTOGRAM_COUNTS("SBClientPhishing.ReportLimitSkipped", 1);
     } else if (csd_service_->OverPhishingReportLimit()) {
       VLOG(1) << "Too many report phishing requests sent recently, "
               << "not running classification for " << params_.url;
-      UMA_HISTOGRAM_ENUMERATION("SBClientPhishing.PreClassificationCheckFail",
-                                NO_CLASSIFY_TOO_MANY_REPORTS,
-                                NO_CLASSIFY_MAX);
-      return;
+      DontClassifyForPhishing(NO_CLASSIFY_TOO_MANY_REPORTS);
+    }
+    if (csd_service_->OverMalwareReportLimit()) {
+      DontClassifyForMalware(NO_CLASSIFY_TOO_MANY_REPORTS);
     }
 
     // Everything checks out, so start classification.
     // |web_contents_| is safe to call as we will be destructed
     // before it is.
-    VLOG(1) << "Instruct renderer to start phishing detection for URL: "
-            << params_.url;
-    content::RenderViewHost* rvh = web_contents_->GetRenderViewHost();
-    rvh->Send(new SafeBrowsingMsg_StartPhishingDetection(
-        rvh->GetRoutingID(), params_.url));
+    if (MaybeClassifyForPhishing())
+      start_phishing_classification_cb_.Run(true);
+    if (MaybeClassifyForMalware())
+      start_malware_classification_cb_.Run(true);
   }
 
-  // No need to protect |canceled_| with a lock because it is only read and
-  // written by the UI thread.
-  bool canceled_;
   content::FrameNavigateParams params_;
   WebContents* web_contents_;
   ClientSideDetectionService* csd_service_;
   // We keep a ref pointer here just to make sure the safe browsing
   // database manager stays alive long enough.
   scoped_refptr<SafeBrowsingDatabaseManager> database_manager_;
   ClientSideDetectionHost* host_;
 
+  ShouldClassifyUrlCallback start_phishing_classification_cb_;
+  ShouldClassifyUrlCallback start_malware_classification_cb_;
+
   DISALLOW_COPY_AND_ASSIGN(ShouldClassifyUrlRequest);
 };
 
 // static
 ClientSideDetectionHost* ClientSideDetectionHost::Create(
     WebContents* tab) {
   return new ClientSideDetectionHost(tab);
 }
 
 ClientSideDetectionHost::ClientSideDetectionHost(WebContents* tab)
     : content::WebContentsObserver(tab),
       csd_service_(NULL),
+      classification_request_(NULL),
+      should_extract_malware_features_(true),
+      onload_complete_(false),
       weak_factory_(this),
-      unsafe_unique_page_id_(-1),
-      malware_killswitch_on_(false),
-      malware_report_enabled_(false) {
+      unsafe_unique_page_id_(-1) {
   DCHECK(tab);
   // Note: csd_service_ and sb_service will be NULL here in testing.
   csd_service_ = g_browser_process->safe_browsing_detection_service();
   feature_extractor_.reset(new BrowserFeatureExtractor(tab, this));
   registrar_.Add(this, content::NOTIFICATION_RESOURCE_RESPONSE_STARTED,
                  content::Source<WebContents>(tab));
 
   scoped_refptr<SafeBrowsingService> sb_service =
       g_browser_process->safe_browsing_service();
   if (sb_service.get()) {
     ui_manager_ = sb_service->ui_manager();
     database_manager_ = sb_service->database_manager();
     ui_manager_->AddObserver(this);
   }
-
-  // Only enable the malware bad IP matching and report feature for canary
-  // and dev channel.
-  chrome::VersionInfo::Channel channel = chrome::VersionInfo::GetChannel();
-  malware_report_enabled_ = (
-      channel == chrome::VersionInfo::CHANNEL_DEV ||
-      channel == chrome::VersionInfo::CHANNEL_CANARY);

[mattm, 2014/03/18 02:19:06]

CL description should mention that it is removing the channel restriction.

[noé, 2014/03/20 17:01:45]

Done.

 }
 
 ClientSideDetectionHost::~ClientSideDetectionHost() {
   if (ui_manager_.get())
     ui_manager_->RemoveObserver(this);
 }
 
 bool ClientSideDetectionHost::OnMessageReceived(const IPC::Message& message) {
   bool handled = true;
   IPC_BEGIN_MESSAGE_MAP(ClientSideDetectionHost, message)
     IPC_MESSAGE_HANDLER(SafeBrowsingHostMsg_PhishingDetectionDone,
                         OnPhishingDetectionDone)
     IPC_MESSAGE_UNHANDLED(handled = false)
   IPC_END_MESSAGE_MAP()
   return handled;
 }
 
 void ClientSideDetectionHost::DidNavigateMainFrame(
     const content::LoadCommittedDetails& details,
     const content::FrameNavigateParams& params) {
   // TODO(noelutz): move this DCHECK to WebContents and fix all the unit tests
   // that don't call this method on the UI thread.
   // DCHECK(BrowserThread::CurrentlyOn(BrowserThread::UI));
   if (details.is_in_page) {
     // If the navigation is within the same page, the user isn't really
     // navigating away.  We don't need to cancel a pending callback or
     // begin a new classification.
     return;
   }
+  // Cancel any pending classification request.
+  if (classification_request_.get()) {
+    classification_request_->Cancel();
+  }
   // If we navigate away and there currently is a pending phishing
   // report request we have to cancel it to make sure we don't display
   // an interstitial for the wrong page.  Note that this won't cancel
   // the server ping back but only cancel the showing of the
   // interstial.
   weak_factory_.InvalidateWeakPtrs();
 
   if (!csd_service_) {
     return;
   }
-
-  // Cancel any pending classification request.
-  if (classification_request_.get()) {
-    classification_request_->Cancel();
-  }
   browse_info_.reset(new BrowseInfo);
 
   // Store redirect chain information.
   if (params.url.host() != cur_host_) {
     cur_host_ = params.url.host();
     cur_host_redirects_ = params.redirects;
   }
+  browse_info_->url = params.url;
   browse_info_->host_redirects = cur_host_redirects_;
   browse_info_->url_redirects = params.redirects;
   browse_info_->referrer = params.referrer.url;
   browse_info_->http_status_code = details.http_status_code;
+  browse_info_->page_id = params.page_id;
 
-  // Notify the renderer if it should classify this URL.
+  should_extract_malware_features_ = true;
+  should_classify_for_malware_.reset();
+  onload_complete_ = false;
+
+  // Check whether we can cassify the current URL for phishing or malware.
   classification_request_ = new ShouldClassifyUrlRequest(
-      params, web_contents(), csd_service_, database_manager_.get(), this);
+      params,
+      base::Bind(&ClientSideDetectionHost::OnPhishingPreClassificationDone,
+                 weak_factory_.GetWeakPtr()),
+      base::Bind(&ClientSideDetectionHost::OnMalwarePreClassificationDone,
+                 weak_factory_.GetWeakPtr()),
+      web_contents(), csd_service_, database_manager_.get(), this);
   classification_request_->Start();
 }
 
 void ClientSideDetectionHost::OnSafeBrowsingHit(
     const SafeBrowsingUIManager::UnsafeResource& resource) {
   if (!web_contents() || !web_contents()->GetController().GetActiveEntry())
     return;
 
   // Check that the hit is either malware or phishing.
   if (resource.threat_type != SB_THREAT_TYPE_URL_PHISHING &&
@@ skipping 62 matching lines @@
 void ClientSideDetectionHost::WebContentsDestroyed(WebContents* tab) {
   DCHECK(tab);
   // Tell any pending classification request that it is being canceled.
   if (classification_request_.get()) {
     classification_request_->Cancel();
   }
   // Cancel all pending feature extractions.
   feature_extractor_.reset();
 }
 
+void ClientSideDetectionHost::OnPhishingPreClassificationDone(
+    bool should_classify) {
+  DCHECK(BrowserThread::CurrentlyOn(BrowserThread::UI));
+  if (browse_info_.get() && should_classify) {
+    VLOG(1) << "Instruct renderer to start phishing detection for URL: "
+            << browse_info_->url;
+    content::RenderViewHost* rvh = web_contents()->GetRenderViewHost();
+    rvh->Send(new SafeBrowsingMsg_StartPhishingDetection(
+        rvh->GetRoutingID(), browse_info_->url));
+  }
+}
+
+void ClientSideDetectionHost::OnMalwarePreClassificationDone(
+    bool should_classify) {
+  DCHECK(BrowserThread::CurrentlyOn(BrowserThread::UI));
+  // If classification checks failed we should stop extracting malware features.
+  VLOG(2) << "Malware pre-classification checks done. Should classify: "
+          << should_classify;
+  should_extract_malware_features_ = should_classify;
+  should_classify_for_malware_.reset(new bool(should_classify));
+  MaybeStartMalwareFeatureExtraction();
+}
+
+void ClientSideDetectionHost::DocumentOnLoadCompletedInMainFrame(
+    int32 page_id) {
+  DCHECK(BrowserThread::CurrentlyOn(BrowserThread::UI));
+  if (!csd_service_ || !browse_info_.get())
+    return;
+  VLOG(2) << "Main frame onload hander called.";
+  if (browse_info_->page_id != page_id) {
+    // Something weird is happening here.  The BrowseInfo page ID
+    // should always be the same as the most recent load.
+    UMA_HISTOGRAM_COUNTS("SBClientMalware.UnexpectedPageId", 1);
+    return;
+  }
+  onload_complete_ = true;
+  MaybeStartMalwareFeatureExtraction();
+}
+
+void ClientSideDetectionHost::MaybeStartMalwareFeatureExtraction() {
+  DCHECK(BrowserThread::CurrentlyOn(BrowserThread::UI));
+  if (csd_service_ && browse_info_.get() &&
+      should_classify_for_malware_.get() &&
+      *should_classify_for_malware_ &&
+      onload_complete_) {
+    scoped_ptr<ClientMalwareRequest> malware_verdict(

[mattm, 2014/03/18 02:19:06]

malware_request?

[noé, 2014/03/20 17:01:45]

Done.

+        new ClientMalwareRequest);
+    // Start browser-side malware feature extraction.  Once we're done it will
+    // send the malware client verdict request.
+    malware_verdict->set_url(browse_info_->url.spec());
+    const GURL& referrer = browse_info_->referrer;
+    if (referrer.SchemeIs("http")) {  // Only send http urls.
+      malware_verdict->set_referrer_url(referrer.spec());
+    }
+    // This function doesn't expect browse_info_ to stay around after this
+    // function returns.
+    feature_extractor_->ExtractMalwareFeatures(
+        browse_info_.get(),
+        malware_verdict.release(),
+        base::Bind(&ClientSideDetectionHost::MalwareFeatureExtractionDone,
+                   weak_factory_.GetWeakPtr()));
+  }
+}
+
 void ClientSideDetectionHost::OnPhishingDetectionDone(
     const std::string& verdict_str) {
   DCHECK(BrowserThread::CurrentlyOn(BrowserThread::UI));
   // There is something seriously wrong if there is no service class but
   // this method is called.  The renderer should not start phishing detection
   // if there isn't any service class in the browser.
   DCHECK(csd_service_);
-  // There shouldn't be any pending requests because we revoke them everytime
-  // we navigate away.
-  DCHECK(!weak_factory_.HasWeakPtrs());
   DCHECK(browse_info_.get());
 
   // We parse the protocol buffer here.  If we're unable to parse it we won't
   // send the verdict further.
   scoped_ptr<ClientPhishingRequest> verdict(new ClientPhishingRequest);
   if (csd_service_ &&
-      !weak_factory_.HasWeakPtrs() &&
       browse_info_.get() &&
       verdict->ParseFromString(verdict_str) &&
       verdict->IsInitialized()) {
-    // We do the malware IP matching and request sending if the feature
-    // is enabled.
-    if (malware_report_enabled_ && !MalwareKillSwitchIsOn()) {
-      scoped_ptr<ClientMalwareRequest> malware_verdict(
-          new ClientMalwareRequest);
-      // Start browser-side malware feature extraction.  Once we're done it will
-      // send the malware client verdict request.
-      malware_verdict->set_url(verdict->url());
-      const GURL& referrer = browse_info_->referrer;
-      if (referrer.SchemeIs("http")) {  // Only send http urls.
-          malware_verdict->set_referrer_url(referrer.spec());
-      }
-      // This function doesn't expect browse_info_ to stay around after this
-      // function returns.
-      feature_extractor_->ExtractMalwareFeatures(
-          browse_info_.get(),
-          malware_verdict.release(),
-          base::Bind(&ClientSideDetectionHost::MalwareFeatureExtractionDone,
-                     weak_factory_.GetWeakPtr()));
-    }
-
     // We only send phishing verdict to the server if the verdict is phishing or
     // if a SafeBrowsing interstitial was already shown for this site.  E.g., a
     // malware or phishing interstitial was shown but the user clicked
     // through.
     if (verdict->is_phishing() || DidShowSBInterstitial()) {
       if (DidShowSBInterstitial()) {
         browse_info_->unsafe_resource.reset(unsafe_resource_.release());
       }
       // Start browser-side feature extraction.  Once we're done it will send
       // the client verdict request.
       feature_extractor_->ExtractFeatures(
           browse_info_.get(),
           verdict.release(),
           base::Bind(&ClientSideDetectionHost::FeatureExtractionDone,
                      weak_factory_.GetWeakPtr()));
     }
   }
-  browse_info_.reset();
 }
 
 void ClientSideDetectionHost::MaybeShowPhishingWarning(GURL phishing_url,
                                                        bool is_phishing) {
   DCHECK(BrowserThread::CurrentlyOn(BrowserThread::UI));
   VLOG(2) << "Received server phishing verdict for URL:" << phishing_url
           << " is_phishing:" << is_phishing;
   if (is_phishing) {
     DCHECK(web_contents());
     if (ui_manager_.get()) {
@@ skipping 108 matching lines @@
 }
 
 void ClientSideDetectionHost::Observe(
     int type,
     const content::NotificationSource& source,
     const content::NotificationDetails& details) {
   DCHECK(BrowserThread::CurrentlyOn(BrowserThread::UI));
   DCHECK_EQ(type, content::NOTIFICATION_RESOURCE_RESPONSE_STARTED);
   const ResourceRequestDetails* req = content::Details<ResourceRequestDetails>(
       details).ptr();
-  if (req && browse_info_.get() && malware_report_enabled_ &&
-      !MalwareKillSwitchIsOn()) {
-    if (req->url.is_valid()) {
-      UpdateIPUrlMap(req->socket_address.host() /* ip */,
-                     req->url.spec()  /* url */,
-                     req->method,
-                     req->referrer,
-                     req->resource_type);
-    }
+  if (req && browse_info_.get() &&
+      should_extract_malware_features_ && req->url.is_valid()) {
+    UpdateIPUrlMap(req->socket_address.host() /* ip */,
+                   req->url.spec()  /* url */,
+                   req->method,
+                   req->referrer,
+                   req->resource_type);
   }
 }
 
 bool ClientSideDetectionHost::DidShowSBInterstitial() const {
   if (unsafe_unique_page_id_ <= 0 || !web_contents()) {
     return false;
   }
   const NavigationEntry* nav_entry =
       web_contents()->GetController().GetActiveEntry();
   return (nav_entry && nav_entry->GetUniqueID() == unsafe_unique_page_id_);
@@ skipping 10 matching lines @@
   if (ui_manager_.get())
     ui_manager_->RemoveObserver(this);
 
   ui_manager_ = ui_manager;
   if (ui_manager)
     ui_manager_->AddObserver(this);
 
   database_manager_ = database_manager;
 }
 
-bool ClientSideDetectionHost::MalwareKillSwitchIsOn() {
-  DCHECK(BrowserThread::CurrentlyOn(BrowserThread::UI));
-  return malware_killswitch_on_;
-}
-
-void ClientSideDetectionHost::SetMalwareKillSwitch(bool killswitch_on) {
-  DCHECK(BrowserThread::CurrentlyOn(BrowserThread::UI));
-  malware_killswitch_on_ = killswitch_on;
-}
-
 }  // namespace safe_browsing