| // Copyright (c) 2011 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/renderer/safe_browsing/phishing_classifier_delegate.h" |
| |
| #include <memory> |
| #include <set> |
| #include <utility> |
| |
| #include "base/bind.h" |
| #include "base/callback.h" |
| #include "base/lazy_instance.h" |
| #include "base/logging.h" |
| #include "base/metrics/histogram_macros.h" |
| #include "base/no_destructor.h" |
| #include "chrome/renderer/safe_browsing/feature_extractor_clock.h" |
| #include "chrome/renderer/safe_browsing/phishing_classifier.h" |
| #include "chrome/renderer/safe_browsing/scorer.h" |
| #include "components/safe_browsing/proto/csd.pb.h" |
| #include "content/public/renderer/document_state.h" |
| #include "content/public/renderer/render_frame.h" |
| #include "content/public/renderer/render_thread.h" |
| #include "services/service_manager/public/cpp/interface_provider.h" |
| #include "third_party/blink/public/platform/web_url.h" |
| #include "third_party/blink/public/web/web_document.h" |
| #include "third_party/blink/public/web/web_local_frame.h" |
| #include "third_party/blink/public/web/web_view.h" |
| |
| using content::DocumentState; |
| using content::RenderThread; |
| |
| namespace safe_browsing { |
| |
| namespace { |
| |
| GURL StripRef(const GURL& url) { |
| GURL::Replacements replacements; |
| replacements.ClearRef(); |
| return url.ReplaceComponents(replacements); |
| } |
| |
| std::set<PhishingClassifierDelegate*>& PhishingClassifierDelegates() { |
| static base::NoDestructor<std::set<PhishingClassifierDelegate*>> s; |
| return *s; |
| } |
| |
| base::LazyInstance<std::unique_ptr<const safe_browsing::Scorer>>:: |
| DestructorAtExit g_phishing_scorer = LAZY_INSTANCE_INITIALIZER; |
| |
| } // namespace |
| |
| // static |
| void PhishingClassifierFilter::Create( |
| mojom::PhishingModelSetterRequest request) { |
| mojo::MakeStrongBinding(std::make_unique<PhishingClassifierFilter>(), |
| std::move(request)); |
| } |
| |
| PhishingClassifierFilter::PhishingClassifierFilter() {} |
| |
| PhishingClassifierFilter::~PhishingClassifierFilter() {} |
| |
| void PhishingClassifierFilter::SetPhishingModel(const std::string& model) { |
| safe_browsing::Scorer* scorer = NULL; |
| // An empty model string means we should disable client-side phishing |
| // detection. |
| if (!model.empty()) { |
| scorer = safe_browsing::Scorer::Create(model); |
| if (!scorer) { |
| DLOG(ERROR) << "Unable to create a PhishingScorer - corrupt model?"; |
| return; |
| } |
| } |
| for (auto* delegate : PhishingClassifierDelegates()) |
| delegate->SetPhishingScorer(scorer); |
| g_phishing_scorer.Get().reset(scorer); |
| } |
| |
| // static |
| PhishingClassifierDelegate* PhishingClassifierDelegate::Create( |
| content::RenderFrame* render_frame, |
| PhishingClassifier* classifier) { |
| // Private constructor and public static Create() method to facilitate |
| // stubbing out this class for binary-size reduction purposes. |
| return new PhishingClassifierDelegate(render_frame, classifier); |
| } |
| |
| PhishingClassifierDelegate::PhishingClassifierDelegate( |
| content::RenderFrame* render_frame, |
| PhishingClassifier* classifier) |
| : content::RenderFrameObserver(render_frame), |
| last_main_frame_transition_(ui::PAGE_TRANSITION_LINK), |
| have_page_text_(false), |
| is_classifying_(false) { |
| PhishingClassifierDelegates().insert(this); |
| if (!classifier) { |
| classifier = |
| new PhishingClassifier(render_frame, new FeatureExtractorClock()); |
| } |
| |
| classifier_.reset(classifier); |
| |
| if (g_phishing_scorer.Get().get()) |
| SetPhishingScorer(g_phishing_scorer.Get().get()); |
| |
| registry_.AddInterface( |
| base::BindRepeating(&PhishingClassifierDelegate::PhishingDetectorRequest, |
| base::Unretained(this))); |
| } |
| |
| PhishingClassifierDelegate::~PhishingClassifierDelegate() { |
| CancelPendingClassification(SHUTDOWN); |
| PhishingClassifierDelegates().erase(this); |
| } |
| |
| void PhishingClassifierDelegate::SetPhishingScorer( |
| const safe_browsing::Scorer* scorer) { |
| if (is_classifying_) { |
| // If there is a classification going on right now it means we're |
| // actually replacing an existing scorer with a new model. In |
| // this case we simply cancel the current classification. |
| // TODO(noelutz): if this happens too frequently we could also |
| // replace the old scorer with the new one once classification is done |
| // but this would complicate the code somewhat. |
| CancelPendingClassification(NEW_PHISHING_SCORER); |
| } |
| classifier_->set_phishing_scorer(scorer); |
| // Start classifying the current page if all conditions are met. |
| // See MaybeStartClassification() for details. |
| MaybeStartClassification(); |
| } |
| |
| void PhishingClassifierDelegate::PhishingDetectorRequest( |
| mojom::PhishingDetectorRequest request) { |
| phishing_detector_bindings_.AddBinding(this, std::move(request)); |
| } |
| |
| void PhishingClassifierDelegate::OnInterfaceRequestForFrame( |
| const std::string& interface_name, |
| mojo::ScopedMessagePipeHandle* interface_pipe) { |
| registry_.TryBindInterface(interface_name, interface_pipe); |
| } |
| |
| void PhishingClassifierDelegate::StartPhishingDetection( |
| const GURL& url, |
| StartPhishingDetectionCallback callback) { |
| last_url_received_from_browser_ = StripRef(url); |
| callback_ = std::move(callback); |
| // Start classifying the current page if all conditions are met. |
| // See MaybeStartClassification() for details. |
| MaybeStartClassification(); |
| } |
| |
| void PhishingClassifierDelegate::DidCommitProvisionalLoad( |
| bool is_same_document_navigation, |
| ui::PageTransition transition) { |
| blink::WebLocalFrame* frame = render_frame()->GetWebFrame(); |
| // A new page is starting to load, so cancel classificaiton. |
| // |
| // TODO(bryner): We shouldn't need to cancel classification if the navigation |
| // is within the same document. However, if we let classification continue in |
| // this case, we need to properly deal with the fact that PageCaptured will |
| // be called again for the same-document navigation. We need to be sure not |
| // to swap out the page text while the term feature extractor is still |
| // running. |
| CancelPendingClassification(is_same_document_navigation ? NAVIGATE_WITHIN_PAGE |
| : NAVIGATE_AWAY); |
| if (frame->Parent()) |
| return; |
| |
| last_main_frame_transition_ = transition; |
| } |
| |
| void PhishingClassifierDelegate::PageCaptured(base::string16* page_text, |
| bool preliminary_capture) { |
| if (preliminary_capture) { |
| return; |
| } |
| // Make sure there's no classification in progress. We don't want to swap |
| // out the page text string from underneath the term feature extractor. |
| // |
| // Note: Currently, if the url hasn't changed, we won't restart |
| // classification in this case. We may want to adjust this. |
| CancelPendingClassification(PAGE_RECAPTURED); |
| last_finished_load_url_ = render_frame()->GetWebFrame()->GetDocument().Url(); |
| classifier_page_text_.swap(*page_text); |
| have_page_text_ = true; |
| MaybeStartClassification(); |
| } |
| |
| void PhishingClassifierDelegate::CancelPendingClassification( |
| CancelClassificationReason reason) { |
| if (is_classifying_) { |
| UMA_HISTOGRAM_ENUMERATION("SBClientPhishing.CancelClassificationReason", |
| reason, |
| CANCEL_CLASSIFICATION_MAX); |
| is_classifying_ = false; |
| } |
| if (classifier_->is_ready()) { |
| classifier_->CancelPendingClassification(); |
| } |
| classifier_page_text_.clear(); |
| have_page_text_ = false; |
| } |
| |
| void PhishingClassifierDelegate::ClassificationDone( |
| const ClientPhishingRequest& verdict) { |
| // We no longer need the page text. |
| classifier_page_text_.clear(); |
| DVLOG(2) << "Phishy verdict = " << verdict.is_phishing() |
| << " score = " << verdict.client_score(); |
| if (verdict.client_score() != PhishingClassifier::kInvalidScore && |
| !callback_.is_null()) { |
| DCHECK_EQ(last_url_sent_to_classifier_.spec(), verdict.url()); |
| std::move(callback_).Run(verdict.SerializeAsString()); |
| } |
| } |
| |
| void PhishingClassifierDelegate::MaybeStartClassification() { |
| // We can begin phishing classification when the following conditions are |
| // met: |
| // 1. A Scorer has been created |
| // 2. The browser has sent a StartPhishingDetection message for the current |
| // toplevel URL. |
| // 3. The page has finished loading and the page text has been extracted. |
| // 4. The load is a new navigation (not a session history navigation). |
| // 5. The toplevel URL has not already been classified. |
| // |
| // Note that if we determine that this particular navigation should not be |
| // classified at all (as opposed to deferring it until we get an IPC or the |
| // load completes), we discard the page text since it won't be needed. |
| if (!classifier_->is_ready()) { |
| DVLOG(2) << "Not starting classification, no Scorer created."; |
| // Keep classifier_page_text_, in case a Scorer is set later. |
| return; |
| } |
| |
| if (last_main_frame_transition_ & ui::PAGE_TRANSITION_FORWARD_BACK) { |
| // Skip loads from session history navigation. However, update the |
| // last URL sent to the classifier, so that we'll properly detect |
| // same-document navigations. |
| DVLOG(2) << "Not starting classification for back/forward navigation"; |
| last_url_sent_to_classifier_ = last_finished_load_url_; |
| classifier_page_text_.clear(); // we won't need this. |
| have_page_text_ = false; |
| return; |
| } |
| |
| GURL stripped_last_load_url(StripRef(last_finished_load_url_)); |
| if (stripped_last_load_url == StripRef(last_url_sent_to_classifier_)) { |
| // We've already classified this toplevel URL, so this was likely an |
| // same-document navigation or a subframe navigation. The browser should |
| // not send a StartPhishingDetection IPC in this case. |
| DVLOG(2) << "Toplevel URL is unchanged, not starting classification."; |
| classifier_page_text_.clear(); // we won't need this. |
| have_page_text_ = false; |
| return; |
| } |
| |
| if (!have_page_text_) { |
| DVLOG(2) << "Not starting classification, there is no page text ready."; |
| return; |
| } |
| |
| if (last_url_received_from_browser_ != stripped_last_load_url) { |
| // The browser has not yet confirmed that this URL should be classified, |
| // so defer classification for now. Note: the ref does not affect |
| // any of the browser's preclassification checks, so we don't require it |
| // to match. |
| DVLOG(2) << "Not starting classification, last url from browser is " |
| << last_url_received_from_browser_ << ", last finished load is " |
| << last_finished_load_url_; |
| // Keep classifier_page_text_, in case the browser notifies us later that |
| // we should classify the URL. |
| return; |
| } |
| |
| DVLOG(2) << "Starting classification for " << last_finished_load_url_; |
| last_url_sent_to_classifier_ = last_finished_load_url_; |
| is_classifying_ = true; |
| classifier_->BeginClassification( |
| &classifier_page_text_, |
| base::BindOnce(&PhishingClassifierDelegate::ClassificationDone, |
| base::Unretained(this))); |
| } |
| |
| void PhishingClassifierDelegate::OnDestruct() { |
| delete this; |
| } |
| |
| } // namespace safe_browsing |