[CRX] Add fallback PDF detection for Chrome 127-

extensions/chromium/contentscript.js

@@ -23,6 +23,9 @@ function getViewerURL(pdf_url) {
 }
 
 document.addEventListener("animationstart", onAnimationStart, true);
+if (document.contentType === "application/pdf") {
+  chrome.runtime.sendMessage({ action: "canRequestBody" }, maybeRenderPdfDoc);
+}
 
 function onAnimationStart(event) {
   if (event.animationName === "pdfjs-detected-object-or-embed") {
@@ -221,3 +224,22 @@ function getEmbeddedViewerURL(path) {
   path = a.href;
   return getViewerURL(path) + fragment;
 }
+
+function maybeRenderPdfDoc(isNotPOST) {
+  if (!isNotPOST) {
+    // The document was loaded through a POST request, but we cannot access the
+    // original response body, nor safely send a new request to fetch the PDF.
+    // Until #4483 is fixed, POST requests should be ignored.
+    return;
+  }
+
+  // Detected PDF that was not redirected by the declarativeNetRequest rules.
+  // Maybe because this was served without Content-Type and sniffed as PDF.
+  // Or because this is Chrome 127-, which does not support responseHeaders
+  // condition in declarativeNetRequest (DNR), and PDF requests are therefore
+  // not redirected via DNR.
+
+  // In any case, load the viewer.
+  console.log(`Detected PDF via document, opening viewer for ${document.URL}`);
+  location.href = getEmbeddedViewerURL(document.URL);
+}

extensions/chromium/pdfHandler.js

@@ -14,6 +14,8 @@ See the License for the specific language governing permissions and
 limitations under the License.
 */
 
+/* globals canRequestBody */ // From preserve-referer.js
+
 "use strict";
 
 var VIEWER_URL = chrome.runtime.getURL("content/web/viewer.html");
@@ -202,7 +204,7 @@ async function registerPdfRedirectRule() {
     // https://github.com/w3c/webextensions/issues/638#issuecomment-2181124486
     //
     // We do not bother with detecting that because we fall back to catching
-    // PDF documents via the content script.
+    // PDF documents via maybeRenderPdfDoc in contentscript.js.
   } catch (e) {
     console.error("Failed to register rules to redirect PDF requests.");
     console.error(e);
@@ -303,6 +305,11 @@ chrome.runtime.onMessage.addListener(function (message, sender, sendResponse) {
         url,
       });
     }
+    return undefined;
+  }
+  if (message && message.action === "canRequestBody") {
+    sendResponse(canRequestBody(sender.tab.id, sender.frameId));
+    return undefined;
   }
   return undefined;
 });

extensions/chromium/preserve-referer.js

@@ -30,6 +30,8 @@ limitations under the License.
  * See setReferer in chromecom.js for more explanation of this logic.
  */
 
+/* exported canRequestBody */ // Used in pdfHandler.js
+
 // g_referrers[tabId][frameId] = referrer of PDF frame.
 var g_referrers = {};
 var g_referrerTimers = {};
@@ -38,14 +40,18 @@ var g_referrerTimers = {};
 // from being kept in memory for too long, cap the data duration to 5 minutes.
 var REFERRER_IN_MEMORY_TIME = 300000;
 
+// g_postRequests[tabId] = Set of frameId that were loaded via POST.
+var g_postRequests = {};
+
 var rIsReferer = /^referer$/i;
 chrome.webRequest.onSendHeaders.addListener(
   function saveReferer(details) {
-    const { tabId, frameId, requestHeaders } = details;
+    const { tabId, frameId, requestHeaders, method } = details;
     g_referrers[tabId] ??= {};
     g_referrers[tabId][frameId] = requestHeaders.find(h =>
       rIsReferer.test(h.name)
     )?.value;
+    setCanRequestBody(tabId, frameId, method !== "GET");
     forgetReferrerEventually(tabId);
   },
   { urls: ["*://*/*"], types: ["main_frame", "sub_frame"] },
@@ -59,9 +65,30 @@ function forgetReferrerEventually(tabId) {
   g_referrerTimers[tabId] = setTimeout(() => {
     delete g_referrers[tabId];
     delete g_referrerTimers[tabId];
+    delete g_postRequests[tabId];
   }, REFERRER_IN_MEMORY_TIME);
 }
 
+// Keeps track of whether a document in tabId + frameId is loaded through a
+// POST form submission. Although this logic has nothing to do with referrer
+// tracking, it is still here to enable re-use of the webRequest listener above.
+function setCanRequestBody(tabId, frameId, isPOST) {
+  if (isPOST) {
+    g_postRequests[tabId] ??= new Set();
+    g_postRequests[tabId].add(frameId);
+  } else {
+    g_postRequests[tabId]?.delete(frameId);
+  }
+}
+
+function canRequestBody(tabId, frameId) {
+  // Returns true unless the frame is known to be loaded through a POST request.
+  // If the background suspends, the information may be lost. This is acceptable
+  // because the information is only potentially needed shortly after document
+  // load, by contentscript.js.
+  return !g_postRequests[tabId]?.has(frameId);
+}
+
 // This method binds a webRequest event handler which adds the Referer header
 // to matching PDF resource requests (only if the Referer is non-empty). The
 // handler is removed as soon as the PDF viewer frame is unloaded.