|
1
|
/**
|
|
2
|
* This file is part of Haketilo.
|
|
3
|
*
|
|
4
|
* Function: Main content script that runs in all frames.
|
|
5
|
*
|
|
6
|
* Copyright (C) 2021 Wojtek Kosior
|
|
7
|
* Copyright (C) 2021 jahoti
|
|
8
|
* Redistribution terms are gathered in the `copyright' file.
|
|
9
|
*/
|
|
10
|
|
|
11
|
/*
|
|
12
|
* IMPORTS_START
|
|
13
|
* IMPORT handle_page_actions
|
|
14
|
* IMPORT gen_nonce
|
|
15
|
* IMPORT is_privileged_url
|
|
16
|
* IMPORT browser
|
|
17
|
* IMPORT is_chrome
|
|
18
|
* IMPORT is_mozilla
|
|
19
|
* IMPORT start_activity_info_server
|
|
20
|
* IMPORT make_csp_rule
|
|
21
|
* IMPORT csp_header_regex
|
|
22
|
* IMPORT report_settings
|
|
23
|
* IMPORTS_END
|
|
24
|
*/
|
|
25
|
|
|
26
|
document.content_loaded = document.readyState === "complete";
|
|
27
|
const wait_loaded = e => e.content_loaded ? Promise.resolve() :
|
|
28
|
new Promise(c => e.addEventListener("DOMContentLoaded", c, {once: true}));
|
|
29
|
|
|
30
|
wait_loaded(document).then(() => document.content_loaded = true);
|
|
31
|
|
|
32
|
/*
|
|
33
|
* In the case of HTML documents:
|
|
34
|
* 1. When injecting some payload we need to sanitize <meta> CSP tags before
|
|
35
|
* they reach the document.
|
|
36
|
* 2. Only <meta> tags inside <head> are considered valid by the browser and
|
|
37
|
* need to be considered.
|
|
38
|
* 3. We want to detach <html> from document, wait until its <head> completes
|
|
39
|
* loading, sanitize it and re-attach <html>.
|
|
40
|
* 4. We shall wait for anything to appear in or after <body> and take that as
|
|
41
|
* a sign <head> has finished loading.
|
|
42
|
* 5. Otherwise, getting the `DOMContentLoaded' event on the document shall also
|
|
43
|
* be a sign that <head> is fully loaded.
|
|
44
|
*/
|
|
45
|
|
|
46
|
function make_body_start_observer(DOM_element, waiting)
|
|
47
|
{
|
|
48
|
const observer = new MutationObserver(() => try_body_started(waiting));
|
|
49
|
observer.observe(DOM_element, {childList: true});
|
|
50
|
return observer;
|
|
51
|
}
|
|
52
|
|
|
53
|
function try_body_started(waiting)
|
|
54
|
{
|
|
55
|
const body = waiting.detached_html.querySelector("body");
|
|
56
|
|
|
57
|
if ((body && (body.firstChild || body.nextSibling)) ||
|
|
58
|
waiting.doc.documentElement.nextSibling) {
|
|
59
|
finish_waiting(waiting);
|
|
60
|
return true;
|
|
61
|
}
|
|
62
|
|
|
63
|
if (body && waiting.observers.length < 2)
|
|
64
|
waiting.observers.push(make_body_start_observer(body, waiting));
|
|
65
|
}
|
|
66
|
|
|
67
|
function finish_waiting(waiting)
|
|
68
|
{
|
|
69
|
if (waiting.finished)
|
|
70
|
return;
|
|
71
|
waiting.finished = true;
|
|
72
|
waiting.observers.forEach(observer => observer.disconnect());
|
|
73
|
setTimeout(waiting.callback, 0);
|
|
74
|
}
|
|
75
|
|
|
76
|
function _wait_for_head(doc, detached_html, callback)
|
|
77
|
{
|
|
78
|
const waiting = {doc, detached_html, callback, observers: []};
|
|
79
|
|
|
80
|
if (try_body_started(waiting))
|
|
81
|
return;
|
|
82
|
|
|
83
|
waiting.observers = [make_body_start_observer(detached_html, waiting)];
|
|
84
|
|
|
85
|
wait_loaded(doc).then(() => finish_waiting(waiting));
|
|
86
|
}
|
|
87
|
|
|
88
|
function wait_for_head(doc, detached_html)
|
|
89
|
{
|
|
90
|
return new Promise(cb => _wait_for_head(doc, detached_html, cb));
|
|
91
|
}
|
|
92
|
|
|
93
|
const blocked_str = "blocked";
|
|
94
|
|
|
95
|
function block_attribute(node, attr, ns=null)
|
|
96
|
{
|
|
97
|
const [hasa, geta, seta, rema] = ["has", "get", "set", "remove"]
|
|
98
|
.map(m => (n, ...args) => typeof ns === "string" ?
|
|
99
|
n[`${m}AttributeNS`](ns, ...args) : n[`${m}Attribute`](...args));
|
|
100
|
/*
|
|
101
|
* Disabling attributes by prepending `-blocked' allows them to still be
|
|
102
|
* relatively easily accessed in case they contain some useful data.
|
|
103
|
*/
|
|
104
|
const construct_name = [attr];
|
|
105
|
while (hasa(node, construct_name.join("")))
|
|
106
|
construct_name.unshift(blocked_str);
|
|
107
|
|
|
108
|
while (construct_name.length > 1) {
|
|
109
|
construct_name.shift();
|
|
110
|
const name = construct_name.join("");
|
|
111
|
seta(node, `${blocked_str}-${name}`, geta(node, name));
|
|
112
|
}
|
|
113
|
|
|
114
|
rema(node, attr);
|
|
115
|
}
|
|
116
|
|
|
117
|
/*
|
|
118
|
* Used to disable `<script>'s and `<meta>'s that have not yet been added to
|
|
119
|
* live DOM (doesn't work for those already added).
|
|
120
|
*/
|
|
121
|
function sanitize_meta(meta)
|
|
122
|
{
|
|
123
|
if (csp_header_regex.test(meta.httpEquiv) && meta.content)
|
|
124
|
block_attribute(meta, "content");
|
|
125
|
}
|
|
126
|
|
|
127
|
function sanitize_script(script)
|
|
128
|
{
|
|
129
|
script.haketilo_blocked_type = script.getAttribute("type");
|
|
130
|
script.type = "text/plain";
|
|
131
|
}
|
|
132
|
|
|
133
|
/*
|
|
134
|
* Executed after `<script>' has been connected to the DOM, when it is no longer
|
|
135
|
* eligible for being executed by the browser.
|
|
136
|
*/
|
|
137
|
function desanitize_script(script)
|
|
138
|
{
|
|
139
|
script.setAttribute("type", script.haketilo_blocked_type);
|
|
140
|
|
|
141
|
if ([null, undefined].includes(script.haketilo_blocked_type))
|
|
142
|
script.removeAttribute("type");
|
|
143
|
|
|
144
|
delete script.haketilo_blocked_type;
|
|
145
|
}
|
|
146
|
|
|
147
|
const bad_url_reg = /^data:([^,;]*ml|unknown-content-type)/i;
|
|
148
|
function sanitize_urls(element)
|
|
149
|
{
|
|
150
|
for (const attr of [...element.attributes || []]
|
|
151
|
.filter(attr => /^(href|src|data)$/i.test(attr.localName))
|
|
152
|
.filter(attr => bad_url_reg.test(attr.value)))
|
|
153
|
block_attribute(element, attr.localName, attr.namespaceURI);
|
|
154
|
}
|
|
155
|
|
|
156
|
function start_data_urls_sanitizing(doc)
|
|
157
|
{
|
|
158
|
doc.querySelectorAll("*[href], *[src], *[data]").forEach(sanitize_urls);
|
|
159
|
if (!doc.content_loaded) {
|
|
160
|
const mutation_handler = m => m.addedNodes.forEach(sanitize_urls);
|
|
161
|
const mo = new MutationObserver(ms => ms.forEach(mutation_handler));
|
|
162
|
mo.observe(doc, {childList: true, subtree: true});
|
|
163
|
wait_loaded(doc).then(() => mo.disconnect());
|
|
164
|
}
|
|
165
|
}
|
|
166
|
|
|
167
|
/*
|
|
168
|
* Normally, we block scripts with CSP. However, Mozilla does optimizations that
|
|
169
|
* cause part of the DOM to be loaded when our content scripts get to run. Thus,
|
|
170
|
* before the CSP rules we inject (for non-HTTP pages) become effective, we need
|
|
171
|
* to somehow block the execution of `<script>'s and intrinsics that were
|
|
172
|
* already there. Additionally, some browsers (IceCat 60) seem to have problems
|
|
173
|
* applying this CSP to non-inline `<scripts>' in certain scenarios.
|
|
174
|
*/
|
|
175
|
function prevent_script_execution(event)
|
|
176
|
{
|
|
177
|
if (!event.target.haketilo_payload)
|
|
178
|
event.preventDefault();
|
|
179
|
}
|
|
180
|
|
|
181
|
function mozilla_initial_block(doc)
|
|
182
|
{
|
|
183
|
doc.addEventListener("beforescriptexecute", prevent_script_execution);
|
|
184
|
|
|
185
|
for (const elem of doc.querySelectorAll("*")) {
|
|
186
|
[...elem.attributes].map(attr => attr.localName)
|
|
187
|
.filter(attr => /^on/.test(attr) && elem.wrappedJSObject[attr])
|
|
188
|
.forEach(attr => elem.wrappedJSObject[attr] = null);
|
|
189
|
}
|
|
190
|
}
|
|
191
|
|
|
192
|
/*
|
|
193
|
* Here we block all scripts of a document which might be either and
|
|
194
|
* HTMLDocument or an XMLDocument. Modifying an XML document might disrupt
|
|
195
|
* Mozilla's XML preview. This is an unfortunate thing we have to accept for
|
|
196
|
* now. XML documents *have to* be sanitized as well because they might
|
|
197
|
* contain `<script>' tags (or on* attributes) with namespace declared as
|
|
198
|
* "http://www.w3.org/1999/xhtml" or "http://www.w3.org/2000/svg" which allows
|
|
199
|
* javascript execution.
|
|
200
|
*/
|
|
201
|
async function sanitize_document(doc, policy)
|
|
202
|
{
|
|
203
|
/*
|
|
204
|
* Blocking of scripts that are in the DOM from the beginning. Needed for
|
|
205
|
* Mozilla.
|
|
206
|
*/
|
|
207
|
if (is_mozilla)
|
|
208
|
mozilla_initial_block(doc);
|
|
209
|
|
|
210
|
/*
|
|
211
|
* Ensure our CSP rules are employed from the beginning. This CSP injection
|
|
212
|
* method is, when possible, going to be applied together with CSP rules
|
|
213
|
* injected using webRequest.
|
|
214
|
* Using elements namespaced as HTML makes this CSP injection also work for
|
|
215
|
* non-HTML documents.
|
|
216
|
*/
|
|
217
|
const html = new DOMParser().parseFromString(`<html><head><meta \
|
|
218
|
http-equiv="Content-Security-Policy" content="${make_csp_rule(policy)}"\
|
|
219
|
/></head><body>Loading...</body></html>`, "text/html").documentElement;
|
|
220
|
|
|
221
|
/*
|
|
222
|
* Root node gets hijacked now, to be re-attached after <head> is loaded
|
|
223
|
* and sanitized.
|
|
224
|
*/
|
|
225
|
const root = doc.documentElement;
|
|
226
|
root.replaceWith(html);
|
|
227
|
|
|
228
|
/*
|
|
229
|
* When we don't inject payload, we neither block document's CSP `<meta>'
|
|
230
|
* tags nor wait for `<head>' to be parsed.
|
|
231
|
*/
|
|
232
|
if (policy.has_payload) {
|
|
233
|
await wait_for_head(doc, root);
|
|
234
|
|
|
235
|
root.querySelectorAll("head meta")
|
|
236
|
.forEach(m => sanitize_meta(m, policy));
|
|
237
|
}
|
|
238
|
|
|
239
|
root.querySelectorAll("script").forEach(s => sanitize_script(s, policy));
|
|
240
|
html.replaceWith(root);
|
|
241
|
root.querySelectorAll("script").forEach(s => desanitize_script(s, policy));
|
|
242
|
|
|
243
|
start_data_urls_sanitizing(doc);
|
|
244
|
}
|
|
245
|
|
|
246
|
async function _disable_service_workers()
|
|
247
|
{
|
|
248
|
if (!navigator.serviceWorker)
|
|
249
|
return;
|
|
250
|
|
|
251
|
const registrations = await navigator.serviceWorker.getRegistrations();
|
|
252
|
if (registrations.length === 0)
|
|
253
|
return;
|
|
254
|
|
|
255
|
console.warn("Service Workers detected on this page! Unregistering and reloading.");
|
|
256
|
|
|
257
|
try {
|
|
258
|
await Promise.all(registrations.map(r => r.unregister()));
|
|
259
|
} finally {
|
|
260
|
location.reload();
|
|
261
|
}
|
|
262
|
|
|
263
|
/* Never actually return! */
|
|
264
|
return new Promise(() => 0);
|
|
265
|
}
|
|
266
|
|
|
267
|
/*
|
|
268
|
* Trying to use servce workers APIs might result in exceptions, for example
|
|
269
|
* when in a non-HTML document. Because of this, we wrap the function that does
|
|
270
|
* the actual work in a try {} block.
|
|
271
|
*/
|
|
272
|
async function disable_service_workers()
|
|
273
|
{
|
|
274
|
try {
|
|
275
|
await _disable_service_workers()
|
|
276
|
} catch (e) {
|
|
277
|
console.debug("Exception thrown during an attempt to detect and disable service workers.", e);
|
|
278
|
}
|
|
279
|
}
|
|
280
|
|
|
281
|
function synchronously_get_policy(url)
|
|
282
|
{
|
|
283
|
const encoded_url = encodeURIComponent(url);
|
|
284
|
const request_url = `${browser.runtime.getURL("dummy")}?url=${encoded_url}`;
|
|
285
|
|
|
286
|
try {
|
|
287
|
var xhttp = new XMLHttpRequest();
|
|
288
|
xhttp.open("GET", request_url, false);
|
|
289
|
xhttp.send();
|
|
290
|
} catch(e) {
|
|
291
|
console.error("Failure to synchronously fetch policy for url.", e);
|
|
292
|
return {allow: false};
|
|
293
|
}
|
|
294
|
|
|
295
|
const policy = /^[^?]*\?settings=(.*)$/.exec(xhttp.responseURL)[1];
|
|
296
|
return JSON.parse(decodeURIComponent(policy));
|
|
297
|
}
|
|
298
|
|
|
299
|
if (!is_privileged_url(document.URL)) {
|
|
300
|
const policy = synchronously_get_policy(document.URL);
|
|
301
|
|
|
302
|
if (!(document instanceof HTMLDocument))
|
|
303
|
delete policy.payload;
|
|
304
|
|
|
305
|
console.debug("current policy", policy);
|
|
306
|
|
|
307
|
report_settings(policy);
|
|
308
|
|
|
309
|
policy.nonce = gen_nonce();
|
|
310
|
|
|
311
|
const doc_ready = Promise.all([
|
|
312
|
policy.allow ? Promise.resolve() : sanitize_document(document, policy),
|
|
313
|
policy.allow ? Promise.resolve() : disable_service_workers(),
|
|
314
|
wait_loaded(document)
|
|
315
|
]);
|
|
316
|
|
|
317
|
handle_page_actions(policy, doc_ready);
|
|
318
|
|
|
319
|
start_activity_info_server();
|
|
320
|
}
|