don't double-modify response headers retrieved from cache

2 files changed, 223 insertions, 33 deletions

diff --git a/background/webrequest.js b/background/webrequest.js
index a523772..5ec7b7f 100644
--- a/background/webrequest.js
+++ b/background/webrequest.js
@@ -3,7 +3,7 @@
  *
  * Function: Modify HTTP traffic usng webRequest API.
  *
- * Copyright (C) 2021 Wojtek Kosior <koszko@koszko.org>
+ * Copyright (C) 2021, 2022 Wojtek Kosior <koszko@koszko.org>
  *
  * This program is free software: you can redistribute it and/or modify
  * it under the terms of the GNU General Public License as published by
@@ -41,53 +41,145 @@
  * proprietary program, I am not going to enforce this in court.
  */
 
-#IMPORT common/indexeddb.js        AS haketilodb
+#IMPORT common/indexeddb.js AS haketilodb
+
 #IF MOZILLA
 #IMPORT background/stream_filter.js
 #ENDIF
 
 #FROM common/browser.js IMPORT browser
-#FROM common/misc.js    IMPORT is_privileged_url, csp_header_regex
+#FROM common/misc.js    IMPORT is_privileged_url, csp_header_regex, \
+                               sha256_async AS sha256
 #FROM common/policy.js  IMPORT decide_policy
 
 #FROM background/patterns_query_manager.js IMPORT tree, default_allow
 
 let secret;
 
-function on_headers_received(details)
-{
+#IF MOZILLA
+/*
+ * Under Mozilla-based browsers, responses are cached together with headers as
+ * they appear *after* modifications by Haketilo. This means Haketilo's CSP
+ * script-blocking headers might be present in responses loaded from cache. In
+ * the meantime the user might have changes Haketilo settings to instead allow
+ * the scripts on the page in question. This causes a problem and creates the
+ * need to somehow restore the response headers to the state in which they
+ * arrived from the server.
+ * To cope with this, Haketilo will inject some additional headers with private
+ * data. Those will include a hard-to-guess value derived from extension's
+ * internal ID. It is assumed the internal ID has a longer lifetime than cached
+ * responses.
+ */
+
+const settings_page_url = browser.runtime.getURL("html/settings.html");
+const header_prefix_prom = sha256(settings_page_url)
+      .then(hash => `X-Haketilo-${hash}`);
+
+/*
+ * Mozilla, unlike Chrome, allows webRequest callbacks to return promises. Here
+ * we leverage that to be able to use asynchronous sha256 computation.
+ */
+async function on_headers_received(details) {
+#IF NEVER
+} /* Help auto-indent in editors. */
+#ENDIF
+#ELSE
+function on_headers_received(details) {
+#ENDIF
     const url = details.url;
     if (is_privileged_url(details.url))
 	return;
 
     let headers = details.responseHeaders;
 
+#IF MOZILLA
+    const prefix = await header_prefix_prom;
+
+    /*
+     * We assume that the original CSP headers of a response are always
+     * preserved under names of the form:
+     *     X-Haketilo-<some_secret>-<original_name>
+     * In some cases the original response may contain no CSP headers. To still
+     * be able to tell whether the headers we were provided were modified by
+     * Haketilo in the past, all modifications are accompanied by addition of an
+     * extra header with name:
+     *     X-Haketilo-<some_secret>
+     */
+
+    const restore_old_headers = details.fromCache &&
+	  !!headers.filter(h => h.name === prefix).length;
+
+    if (restore_old_headers) {
+	const restored_headers = [];
+
+	for (const h of headers) {
+	    if (csp_header_regex.test(h.name) || h.name === prefix)
+		continue;
+
+	    if (h.name.startsWith(prefix)) {
+		restored_headers.push({
+		    name:  h.name.substring(prefix.length + 1),
+		    value: h.value
+		});
+	    } else {
+		restored_headers.push(h);
+	    }
+	}
+
+	headers = restored_headers;
+    }
+#ENDIF
+
     const policy =
 	  decide_policy(tree, details.url, !!default_allow.value, secret);
-    if (policy.allow)
-	return;
 
-    if (policy.payload)
-	headers = headers.filter(h => !csp_header_regex.test(h.name));
+    if (!policy.allow) {
+#IF MOZILLA
+	const to_append = [{name: prefix, value: ":)"}];
+
+	for (const h of headers.filter(h => csp_header_regex.test(h.name))) {
+	    if (!policy.payload)
+		to_append.push(Object.assign({}, h));
+
+	    h.name = `${prefix}-${h.name}`;
+	}
 
-    headers.push({name: "Content-Security-Policy", value: policy.csp});
+	headers.push(...to_append);
+#ELSE
+	if (policy.payload)
+	    headers = headers.filter(h => !csp_header_regex.test(h.name));
+#ENDIF
+
+	headers.push({name: "Content-Security-Policy", value: policy.csp});
+    }
 
 #IF MOZILLA
-    let skip = false;
-    for (const header of headers) {
-	if (header.name.toLowerCase().trim() !== "content-disposition")
-	    continue;
-
-	if (/^\s*attachment\s*(;.*)$/i.test(header.value)) {
-		skip = true;
-	} else {
-	    skip = false;
-	    break;
+    /*
+     * When page is meant to be viewed in the browser, use streamFilter to
+     * inject a dummy <script> at the very beginning of it. This <script>
+     * will cause extension's content scripts to run before page's first <meta>
+     * tag is rendered so that they can prevent CSP rules inside <meta> tags
+     * from blocking the payload we want to inject.
+     */
+
+    let use_stream_filter = !!policy.payload;
+    if (use_stream_filter) {
+	for (const header of headers) {
+	    if (header.name.toLowerCase().trim() !== "content-disposition")
+		continue;
+
+	    if (/^\s*attachment\s*(;.*)$/i.test(header.value)) {
+		use_stream_filter = false;
+	    } else {
+		use_stream_filter = true;
+		break;
+	    }
 	}
     }
-    skip = skip || (details.statusCode >= 300 && details.statusCode < 400);
+    use_stream_filter = use_stream_filter &&
+	(details.statusCode < 300 || details.statusCode >= 400);
 
-    if (!skip)
+    if (use_stream_filter)
 	headers = stream_filter.apply(details, headers, policy);
 #ENDIF
 
diff --git a/test/haketilo_test/unit/test_webrequest.py b/test/haketilo_test/unit/test_webrequest.py
index fb24b3d..1244117 100644
--- a/test/haketilo_test/unit/test_webrequest.py
+++ b/test/haketilo_test/unit/test_webrequest.py
@@ -24,6 +24,10 @@ import pytest
 from ..script_loader import load_script
 from .utils import are_scripts_allowed
 
+allowed_url = 'https://site.with.scripts.allow.ed/'
+blocked_url = 'https://site.with.scripts.block.ed/'
+payload_url = 'https://site.with.paylo.ad/'
+
 def webrequest_js():
     return (load_script('background/webrequest.js',
                         '#IMPORT common/patterns_query_tree.js AS pqt') +
@@ -34,24 +38,118 @@ def webrequest_js():
             default_allow = {name: "default_allow", value: true};
 
             // Rule to block scripts.
-            pqt.register(tree, "https://site.with.scripts.block.ed/***",
+            pqt.register(tree, "%(blocked)s***",
                          "~allow", 0);
 
             // Rule to allow scripts, but overridden by payload assignment.
-            pqt.register(tree, "https://site.with.paylo.ad/***", "~allow", 1);
-            pqt.register(tree, "https://site.with.paylo.ad/***",
-                         "somemapping", {identifier: "someresource"});
+            pqt.register(tree, "%(payload)s***", "~allow", 1);
+            pqt.register(tree, "%(payload)s***", "somemapping",
+                         {identifier: "someresource"});
 
             // Mock stream_filter.
             stream_filter.apply = (details, headers, policy) => headers;
+            ''' % {'blocked': blocked_url, 'payload': payload_url})
+
+def webrequest_js_start_called():
+    return webrequest_js() + ';\nstart("somesecret");'
+
+ext_url = 'moz-extension://49de6ce9-49fc-49e1-8102-7ef35286389c/html/settings.html'
+prefix = 'X-Haketilo-' + sha256(ext_url.encode()).digest().hex()
+
+# Prepare a list of headers as could be sent by a website.
+sample_csp_header = {
+    'name': 'Content-Security-Policy',
+    'value': "script-src 'self';"
+}
+sample_csp_header_idx = 7
+
+sample_headers = [
+    {'name': 'Content-Type',     'value': 'text/html;charset=utf-8'},
+    {'name': 'Content-Length',   'value': '61954'},
+    {'name': 'Content-Language', 'value': 'en'},
+    {'name': 'Expires',          'value': 'Mon, 12 Mar 2012 11:04...'},
+    {'name': 'Last-Modified',    'value': 'Fri, 26 Jul 2013 22:50...'},
+    {'name': 'Cache-Control',    'value': 'max-age=0, s-maxage=86...'},
+    {'name': 'Age',              'value': '224'},
+    {'name': 'Server',           'value': 'nginx/1.1.19'},
+    {'name': 'Date',             'value': 'Thu, 10 Mar 2022 12:09...'}
+]
+
+sample_headers.insert(sample_csp_header_idx, sample_csp_header)
+
+# Prepare a list of headers as would be crafted by Haketilo when there is a
+# payload to inject.
+nonce_source = f'somemapping:someresource:{payload_url}:somesecret'.encode()
+nonce = f'nonce-{sha256(nonce_source).digest().hex()}'
+
+payload_csp_header = {
+    'name': f'Content-Security-Policy',
+    'value': ("prefetch-src 'none'; script-src-attr 'none'; "
+              f"script-src '{nonce}'; script-src-elem '{nonce}';")
+}
+
+sample_payload_headers = [
+    *sample_headers,
+    {'name': prefix, 'value': ':)'},
+    payload_csp_header
+]
+
+sample_payload_headers[sample_csp_header_idx] = {
+    **sample_csp_header,
+    'name': f'{prefix}-{sample_csp_header["name"]}',
+}
+
+# Prepare a list of headers as would be crafted by Haketilo when scripts are
+# blocked.
+sample_blocked_headers = [*sample_payload_headers]
+sample_blocked_headers.pop()
+sample_blocked_headers.append(sample_csp_header)
+sample_blocked_headers.append({
+    'name': f'Content-Security-Policy',
+    'value': ("prefetch-src 'none'; script-src-attr 'none'; "
+              f"script-src 'none'; script-src-elem 'none';")
+})
+
+@pytest.mark.get_page('https://gotmyowndoma.in')
+@pytest.mark.parametrize('params', [
+    (sample_headers,         allowed_url),
+    (sample_blocked_headers, blocked_url),
+    (sample_payload_headers, payload_url),
+])
+def test_webrequest_on_headers_received(driver, execute_in_page, params):
+    """Unit-test the on_headers_received() function."""
+    headers_out, url = params
+
+    execute_in_page(
+        '''{
+        // Mock browser object.
+        const url = arguments[0];
+        this.browser = {runtime: {getURL: () => url}};
+        }''',
+        ext_url)
+
+    execute_in_page(webrequest_js())
+
+    execute_in_page('secret = "somesecret";')
+
+    for headers_in in [
+            sample_headers,
+            sample_blocked_headers,
+            sample_payload_headers
+    ]:
+        details = {'url': url, 'responseHeaders': headers_in, 'fromCache': True}
+        res = execute_in_page('returnval(on_headers_received(arguments[0]));',
+                              details)
 
-            // Mock secret and start webrequest operations.
-            start("somesecret");
-            ''')
+        assert res == {'responseHeaders': headers_out}
 
-@pytest.mark.ext_data({'background_script': webrequest_js})
+@pytest.mark.ext_data({'background_script': webrequest_js_start_called})
 @pytest.mark.usefixtures('webextension')
-def test_on_headers_received(driver, execute_in_page):
+def test_webrequest_real_pages(driver, execute_in_page):
+    """
+    Test webRequest-based header modifications by loading actual pages and
+    attempting to run scripts within them.
+    """
     for attempt in range(10):
         driver.get('https://site.with.scripts.block.ed/')
 
@@ -59,10 +157,10 @@ def test_on_headers_received(driver, execute_in_page):
             break
         assert attempt != 9
 
-    driver.get('https://site.with.scripts.allow.ed/')
+    driver.get(allowed_url)
     assert are_scripts_allowed(driver)
 
-    driver.get('https://site.with.paylo.ad/')
+    driver.get(payload_url)
     assert not are_scripts_allowed(driver)
     source = 'somemapping:someresource:https://site.with.paylo.ad/index.html:somesecret'
     assert are_scripts_allowed(driver, sha256(source.encode()).digest().hex())