torbutton/en/design/design.xml (759dcea9e) - tor-webwml.git

design.xml

<?xml version="1.0" encoding="ISO-8859-1"?>
<!DOCTYPE article PUBLIC "-//OASIS//DTD DocBook XML V4.4//EN"
     "file:///usr/share/sgml/docbook/xml-dtd-4.4-1.0-30.1/docbookx.dtd">

<article id="design">
 <articleinfo>
  <title>Torbutton Design Documentation</title>
   <author>
    <firstname>Mike</firstname><surname>Perry</surname>
    <affiliation>
     <address><email>mikeperry.fscked/org</email></address>
    </affiliation>
   </author>
   <pubdate>Apr 3 2011</pubdate>
 </articleinfo>

<sect1>
  <title>Introduction</title>
  <para>

This document describes the goals, operation, and testing procedures of the
Torbutton Firefox extension. It is current as of Torbutton 1.3.2.

</para>
  <sect2 id="adversary">
   <title>Adversary Model</title>
   <para>

A Tor web browser adversary has a number of goals, capabilities, and attack
types that can be used to guide us towards a set of requirements for the
Torbutton extension. Let's start with the goals.

</para>
   <sect3 id="adversarygoals">
    <title>Adversary Goals</title>
    <orderedlist>

     <listitem><command>Bypassing proxy settings</command>
     <para>The adversary's primary goal is direct compromise and bypass of 
Tor, causing the user to directly connect to an IP of the adversary's
choosing.</para>
     </listitem>
     <listitem><command>Correlation of Tor vs Non-Tor Activity</command>
     <para>If direct proxy bypass is not possible, the adversary will likely
happily settle for the ability to correlate something a user did via Tor with
their non-Tor activity. This can be done with cookies, cache identifiers,
javascript events, and even CSS. Sometimes the fact that a user uses Tor may
be enough for some authorities.</para>
     </listitem>
     <listitem><command>History disclosure</command>
     <para>
The adversary may also be interested in history disclosure: the ability to
query a user's history to see if they have issued certain censored search
queries, or visited censored sites.
     </para>
     </listitem>
     <listitem><command>Location information</command>
     <para>

Location information such as timezone and locality can be useful for the
adversary to determine if a user is in fact originating from one of the
regions they are attempting to control, or to zero-in on the geographical
location of a particular dissident or whistleblower.

</para>
     </listitem>
     <listitem><command>Miscellaneous anonymity set reduction</command>
     <para>

Anonymity set reduction is also useful in attempting to zero in on a
particular individual. If the dissident or whistleblower is using a rare build
of Firefox for an obscure operating system, this can be very useful
information for tracking them down, or at least <link
linkend="fingerprinting">tracking their activities</link>.

</para>
     </listitem>
     <listitem><command>History records and other on-disk
information</command>
     <para>
In some cases, the adversary may opt for a heavy-handed approach, such as
seizing the computers of all Tor users in an area (especially after narrowing
the field by the above two pieces of information). History records and cache
data are the primary goals here.
     </para>
     </listitem>
    </orderedlist>
   </sect3>

<sect3 id="adversarypositioning">
    <title>Adversary Capabilities - Positioning</title>
    <para>
The adversary can position themselves at a number of different locations in
order to execute their attacks.
    </para>
    <orderedlist>
     <listitem><command>Exit Node or Upstream Router</command>
     <para>
The adversary can run exit nodes, or alternatively, they may control routers
upstream of exit nodes. Both of these scenarios have been observed in the
wild.
     </para>
     </listitem>
     <listitem><command>Adservers and/or Malicious Websites</command>
     <para>
The adversary can also run websites, or more likely, they can contract out
ad space from a number of different adservers and inject content that way. For
some users, the adversary may be the adservers themselves. It is not
inconceivable that adservers may try to subvert or reduce a user's anonymity 
through Tor for marketing purposes.
     </para>
     </listitem>
     <listitem><command>Local Network/ISP/Upstream Router</command>
     <para>
The adversary can also inject malicious content at the user's upstream router
when they have Tor disabled, in an attempt to correlate their Tor and Non-Tor
activity.
     </para>
     </listitem>
     <listitem><command>Physical Access</command>
     <para>
Some users face adversaries with intermittent or constant physical access.
Users in Internet cafes, for example, face such a threat. In addition, in
countries where simply using tools like Tor is illegal, users may face
confiscation of their computer equipment for excessive Tor usage or just
general suspicion.
     </para>
     </listitem>
    </orderedlist>
   </sect3>

<sect3 id="attacks">
    <title>Adversary Capabilities - Attacks</title>
    <para>

The adversary can perform the following attacks from a number of different 
positions to accomplish various aspects of their goals. It should be noted
that many of these attacks (especially those involving IP address leakage) are
often performed by accident by websites that simply have Javascript, dynamic 
CSS elements, and plugins. Others are performed by adservers seeking to
correlate users' activity across different IP addresses, and still others are
performed by malicious agents on the Tor network and at national firewalls.

</para>
    <orderedlist>
     <listitem><command>Inserting Javascript</command>
     <para>
If not properly disabled, Javascript event handlers and timers
can cause the browser to perform network activity after Tor has been disabled,
thus allowing the adversary to correlate Tor and Non-Tor activity and reveal
a user's non-Tor IP address. Javascript
also allows the adversary to execute <ulink
url="http://whattheinternetknowsaboutyou.com/">history disclosure attacks</ulink>:
to query the history via the different attributes of 'visited' links to search
for particular Google queries, sites, or even to <ulink
url="http://www.mikeonads.com/2008/07/13/using-your-browser-url-history-estimate-gender/">profile
users based on gender and other classifications</ulink>. Finally,
Javascript can be used to query the user's timezone via the
<function>Date()</function> object, and to reduce the anonymity set by querying
the <function>navigator</function> object for operating system, CPU, locale, 
and user agent information.
     </para>
     </listitem>

<listitem><command>Inserting Plugins</command>
     <para>

Plugins are abysmal at obeying the proxy settings of the browser. Every plugin
capable of performing network activity that the author has
investigated is also capable of performing network activity independent of
browser proxy settings - and often independent of its own proxy settings.
Sites that have plugin content don't even have to be malicious to obtain a
user's
Non-Tor IP (it usually leaks by itself), though <ulink
url="http://decloak.net">plenty of active
exploits</ulink> are possible as well. In addition, plugins can be used to store unique identifiers that are more
difficult to clear than standard cookies. 
<ulink url="http://epic.org/privacy/cookies/flash.html">Flash-based
cookies</ulink> fall into this category, but there are likely numerous other
examples.

</para>
     </listitem>
     <listitem><command>Inserting CSS</command>
     <para>

CSS can also be used to correlate Tor and Non-Tor activity and reveal a user's
Non-Tor IP address, via the usage of
<ulink url="http://www.tjkdesign.com/articles/css%20pop%20ups/">CSS
popups</ulink> - essentially CSS-based event handlers that fetch content via
CSS's onmouseover attribute. If these popups are allowed to perform network
activity in a different Tor state than they were loaded in, they can easily
correlate Tor and Non-Tor activity and reveal a user's IP address. In
addition, CSS can also be used without Javascript to perform <ulink
url="http://ha.ckers.org/weird/CSS-history.cgi">CSS-only history disclosure
attacks</ulink>.
     </para>
     </listitem>
     <listitem><command>Read and insert cookies</command>
     <para>

An adversary in a position to perform MITM content alteration can inject
document content elements to both read and inject cookies for
arbitrary domains. In fact, many "SSL secured" websites are vulnerable to this
sort of <ulink url="http://seclists.org/bugtraq/2007/Aug/0070.html">active
sidejacking</ulink>.

</para>
     </listitem>
     <listitem><command>Create arbitrary cached content</command>
     <para>

Likewise, the browser cache can also be used to <ulink
url="http://crypto.stanford.edu/sameorigin/safecachetest.html">store unique
identifiers</ulink>. Since by default the cache has no same-origin policy,
these identifiers can be read by any domain, making them an ideal target for
adserver-class adversaries.

</para>
     </listitem>

<listitem id="fingerprinting"><command>Fingerprint users based on browser
attributes</command>
<para>

There is an absurd amount of information available to websites via attributes
of the browser. This information can be used to reduce anonymity set, or even
<ulink url="http://mandark.fr/0x000000/articles/Total_Recall_On_Firefox..html">uniquely
fingerprint individual users</ulink>. </para>
<para>
For illustration, let's perform a
back-of-the-envelope calculation on the number of anonymity sets for just the
resolution information available in the <ulink
url="http://developer.mozilla.org/en/docs/DOM:window">window</ulink> and
<ulink
url="http://developer.mozilla.org/en/docs/DOM:window.screen">window.screen</ulink>
objects.

Browser window resolution information provides something like
(1280-640)*(1024-480)=348160 different anonymity sets. Desktop resolution
information contributes about another factor of 5 (for about 5 resolutions in
typical use). In addition, the dimensions and position of the desktop taskbar
are available, which can reveal hints on OS information. This boosts the count
by a factor of 5 (for each of the major desktop taskbars - Windows, OSX, KDE
and Gnome, and None). Subtracting the browser content window
size from the browser outer window size provide yet more information.
Firefox toolbar presence gives about a factor of 8 (3 toolbars on/off give
2<superscript>3</superscript>=8). Interface effects such as title bar font size
and window manager settings gives a factor of about 9 (say 3 common font sizes
for the title bar and 3 common sizes for browser GUI element fonts).
Multiply this all out, and you have (1280-640)*(1024-480)*5*5*8*9 ~=
2<superscript>29</superscript>, or a 29 bit identifier based on resolution
information alone. </para>

<para>

Of course, this space is non-uniform in user density and prone to incremental
changes. The <ulink
url="https://wiki.mozilla.org/Fingerprinting#Data">Panopticlick study
done</ulink> by the EFF attempts to measure the actual entropy - the number of
identifying bits of information encoded in browser properties.  Their result
data is definitely useful, and the metric is probably the appropriate one for
determining how identifying a particular browser property is. However, some
quirks of their study means that they do not extract as much information as
they could from display information: they only use desktop resolution (which
Torbutton reports as the window resolution) and do not attempt to infer the
size of toolbars.

</para>
<!--
FIXME: This is no longer true. Only certain addons are now discoverable, and
only if they want to be:
http://webdevwonders.com/detecting-firefox-add-ons/
https://developer.mozilla.org/en/Updating_web_applications_for_Firefox_3#section_7

<para>

To add insult to injury, <ulink
url="http://pseudo-flaw.net/content/tor/torbutton/">chrome URL disclosure
attacks</ulink> mean that each and every extension on <ulink
url="https://addons.mozilla.org">addons.mozilla.org</ulink> adds another bit
to that 2<superscript>29</superscript>. With hundreds of popular extensions
and thousands of extensions total, it is easy to see that this sort of
information is an impressively powerful identifier if used properly by a
competent and determined adversary such as an ad network.  Again, a
nearest-neighbor bit vector space approach here would also gracefully handle
incremental changes to installed extensions.

</para>
-->
     </listitem>
     <listitem><command>Remotely or locally exploit browser and/or
OS</command>
     <para>
Last, but definitely not least, the adversary can exploit either general 
browser vulnerabilities, plugin vulnerabilities, or OS vulnerabilities to
install malware and surveillance software. An adversary with physical access
can perform similar actions. Regrettably, this last attack capability is
outside of Torbutton's ability to defend against, but it is worth mentioning
for completeness.
     </para>
     </listitem>
    </orderedlist>
   </sect3>

</sect2>

<sect2 id="requirements">
   <title>Torbutton Requirements</title>
<note>

Since many settings satisfy multiple requirements, this design document is
organized primarily by Torbutton components and settings. However, if you are
the type that would rather read the document from the requirements
perspective, it is in fact possible to search for each of the following
requirement phrases in the text to find the relevant features that help meet
that requirement.

</note>
   <para>

From the above Adversary Model, a number of requirements become clear.

</para>

<orderedlist> 

 <listitem id="proxy"><command>Proxy Obedience</command> 
 <para>The browser
MUST NOT bypass Tor proxy settings for any content.</para></listitem>
 <listitem id="isolation"><command>Network Isolation</command>
 <para>Pages MUST NOT perform any network activity in a Tor state different
 from the state they were originally loaded in.</para></listitem>
 <listitem id="state"><command>State Separation</command>
 <para>Browser state (cookies, cache, history, 'DOM storage'), accumulated in
 one Tor state MUST NOT be accessible via the network in
 another Tor state.</para></listitem>
 <listitem id="undiscoverability"><command>Tor Undiscoverability</command><para>With
the advent of bridge support in Tor 0.2.0.x, there are now a class of Tor
users whose network fingerprint does not obviously betray the fact that they
are using Tor. This should extend to the browser as well - Torbutton MUST NOT 
reveal its presence while Tor is disabled.</para></listitem>
 <listitem id="disk"><command>Disk Avoidance</command><para>The browser SHOULD NOT write any Tor-related state to disk, or store it
 in memory beyond the duration of one Tor toggle.</para></listitem>
 <listitem id="location"><command>Location Neutrality</command><para>The browser SHOULD NOT leak location-specific information, such as
 timezone or locale via Tor.</para></listitem>
 <listitem id="setpreservation"><command>Anonymity Set
Preservation</command><para>The browser SHOULD NOT leak any other anonymity set reducing information 
 (such as user agent, extension presence, and resolution information)
automatically via Tor. The assessment of the attacks above should make it clear
that anonymity set reduction is a very powerful method of tracking and
eventually identifying anonymous users.
</para></listitem>
 <listitem id="updates"><command>Update Safety</command><para>The browser
SHOULD NOT perform unauthenticated updates or upgrades via Tor.</para></listitem>
 <listitem id="interoperate"><command>Interoperability</command><para>Torbutton SHOULD interoperate with third-party proxy switchers that
 enable the user to switch between a number of different proxies. It MUST
 provide full Tor protection in the event a third-party proxy switcher has
 enabled the Tor proxy settings.</para></listitem>
</orderedlist>
  </sect2>
  <sect2 id="layout">
   <title>Extension Layout</title>

<para>Firefox extensions consist of two main categories of code: 'Components' and
'Chrome'. Components are a fancy name for classes that implement a given
interface or interfaces. In Firefox, components <ulink
url="https://developer.mozilla.org/en/XPCOM">can be
written</ulink> in C++,
Javascript, or a mixture of both. Components have two identifiers: their
'<ulink
url="http://www.mozilla.org/projects/xpcom/book/cxc/html/quicktour2.html#1005005">Contract
ID</ulink>' (a human readable path-like string), and their '<ulink
url="http://www.mozilla.org/projects/xpcom/book/cxc/html/quicktour2.html#1005329">Class
ID</ulink>' (a GUID hex-string). In addition, the interfaces they implement each have a hex
'Interface ID'. It is possible to 'hook' system components - to reimplement
their interface members with your own wrappers - but only if the rest of the
browser refers to the component by its Contract ID. If the browser refers to
the component by Class ID, it bypasses your hooks in that use case.
Technically, it may be possible to hook Class IDs by unregistering the
original component, and then re-registering your own, but this relies on
obsolete and deprecated interfaces and has proved to be less than
stable.</para>

<para>'Chrome' is a combination of XML and Javascript used to describe a window.
Extensions are allowed to create 'overlays' that are 'bound' to existing XML
window definitions, or they can create their own windows. The DTD for this XML
is called <ulink
url="http://developer.mozilla.org/en/docs/XUL_Reference">XUL</ulink>.</para>
  </sect2>
</sect1>
<sect1 id="components">
  <title>Components</title>
  <para>

Torbutton installs components for two purposes: hooking existing components to
reimplement their interfaces; and creating new components that provide
services to other pieces of the extension.

</para>

<sect2 id="hookedxpcom">
   <title>Hooked Components</title>

<para>Torbutton makes extensive use of Contract ID hooking, and implements some
of its own standalone components as well.  Let's discuss the hooked components
first.</para>

<sect3 id="appblocker">
 <title><ulink
url="http://www.oxymoronical.com/experiments/xpcomref/applications/Firefox/3.5/components/%40mozilla.org/uriloader/external-protocol-service%3B1">@mozilla.org/uriloader/external-protocol-service;1
</ulink>, <ulink
url="http://www.oxymoronical.com/experiments/xpcomref/applications/Firefox/3.5/components/%40mozilla.org/uriloader/external-helper-app-service%3B1">@mozilla.org/uriloader/external-helper-app-service;1</ulink>,
and <ulink url="http://www.oxymoronical.com/experiments/xpcomref/applications/Firefox/3.5/components/%40mozilla.org/mime%3B1">@mozilla.org/mime;1</ulink>
- <ulink
  url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/components/external-app-blocker.js">components/external-app-blocker.js</ulink></title>
 <para>
Due to <link linkend="FirefoxBugs">Firefox Bug</link> <ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=440892">440892</ulink> allowing Firefox 3.x to automatically launch some
applications without user intervention, Torbutton had to wrap the three
components involved in launching external applications to provide user
confirmation before doing so while Tor is enabled. Since external applications
do not obey proxy settings, they can be manipulated to automatically connect
back to arbitrary servers outside of Tor with no user intervention. Fixing
this issue helps to satisfy Torbutton's <link linkend="proxy">Proxy
Obedience</link> Requirement.
 </para>
</sect3>
<sect3>
<title><ulink url="http://www.oxymoronical.com/experiments/xpcomref/applications/Firefox/3.5/components/%40mozilla.org/browser/global-history;2">@mozilla.org/browser/global-history;2</ulink>
- <ulink
  url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/components/ignore-history.js">components/ignore-history.js</ulink></title>

<para>This component was contributed by <ulink
url="http://www.collinjackson.com/">Collin Jackson</ulink> as a method for defeating
CSS and Javascript-based methods of history disclosure. The global-history
component is what is used by Firefox to determine if a link was visited or not
(to apply the appropriate style to the link). By hooking the <ulink
url="https://developer.mozilla.org/en/nsIGlobalHistory2#isVisited.28.29">isVisited</ulink>
and <ulink 
url="https://developer.mozilla.org/en/nsIGlobalHistory2#addURI.28.29">addURI</ulink>
methods, Torbutton is able to selectively prevent history items from being
added or being displayed as visited, depending on the Tor state and the user's
preferences.
</para>
<para>
This component helps satisfy the <link linkend="state">State Separation</link>
and <link linkend="disk">Disk Avoidance</link> requirements of Torbutton. It
is only needed for Firefox 3.x. On Firefox 4, we omit this component in favor
of the <ulink
url="https://developer.mozilla.org/en/CSS/Privacy_and_the_%3avisited_selector">built-in
history protections</ulink>.
</para>
</sect3>
<sect3 id="livemarks">
<title><ulink
url="http://www.oxymoronical.com/experiments/xpcomref/applications/Firefox/3.5/components/%40mozilla.org/browser/livemark-service;2">@mozilla.org/browser/livemark-service;2</ulink>
- <ulink
  url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/components/block-livemarks.js">components/block-livemarks.js</ulink></title>
<para>

The <ulink
url="http://www.mozilla.com/en-US/firefox/livebookmarks.html">livemark</ulink> service
is started by a timer that runs 5 seconds after Firefox
startup. As a result, we cannot simply call the stopUpdateLivemarks() method to
disable it. We must wrap the component to prevent this start() call from
firing in the event the browser starts in Tor mode.

</para>
<para>
This component helps satisfy the <link linkend="isolation">Network
Isolation</link> and <link linkend="setpreservation">Anonymity Set
Preservation</link> requirements.
</para>
</sect3>
</sect2>
<sect2>
<title>New Components</title>

<para>Torbutton creates four new components that are used throughout the
extension. These components do not hook any interfaces, nor are they used
anywhere besides Torbutton itself.</para>

<sect3 id="cookiejar">
<title><ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/components/cookie-jar-selector.js">@torproject.org/cookie-jar-selector;2
- components/cookie-jar-selector.js</ulink></title>

<para>The cookie jar selector (also based on code from <ulink
url="http://www.collinjackson.com/">Collin
Jackson</ulink>) is used by the Torbutton chrome to switch between
Tor and Non-Tor cookies. It stores an XML representation of the current
cookie state in memory and/or on disk. When Tor is toggled, it syncs the
current cookies to this XML store, and then loads the cookies for the other
state from the XML store.
</para>

<para>
This component helps to address the <link linkend="state">State
Isolation</link> requirement of Torbutton.
</para>

</sect3>
<sect3>
<title><ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/components/torbutton-logger.js">@torproject.org/torbutton-logger;1
- components/torbutton-logger.js</ulink></title>

<para>The torbutton logger component allows on-the-fly redirection of torbutton
logging messages to either Firefox stderr
(<command>extensions.torbutton.logmethod=0</command>), the Javascript error console
(<command>extensions.torbutton.logmethod=1</command>), or the DebugLogger extension (if
available - <command>extensions.torbutton.logmethod=2</command>). It also allows you to
change the loglevel on the fly by changing
<command>extensions.torbutton.loglevel</command> (1-5, 1 is most verbose).
</para>
</sect3>
<sect3 id="windowmapper">

<title><ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/components/window-mapper.js">@torproject.org/content-window-mapper;1
- components/window-mapper.js</ulink></title>

<para>Torbutton tags Firefox <ulink
url="https://developer.mozilla.org/en/XUL_Tutorial/Tabboxes">tabs</ulink> with a special variable that indicates the Tor
state the tab was most recently used under to fetch a page. The problem is
that for many Firefox events, it is not possible to determine the tab that is
actually receiving the event. The Torbutton window mapper allows the Torbutton
chrome and other components to look up a <ulink
url="https://developer.mozilla.org/en/XUL/tabbrowser">browser
tab</ulink> for a given <ulink
url="https://developer.mozilla.org/en/nsIDOMWindow">HTML content
window</ulink>. It does this by traversing all windows and all browsers, until it
finds the browser with the requested <ulink
url="https://developer.mozilla.org/en/XUL/tabbrowser#p-contentWindow">contentWindow</ulink> element. Since the content policy
and page loading in general can generate hundreds of these lookups, this
result is cached inside the component.
</para>
</sect3>
<sect3 id="crashobserver">
 <title><ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/components/crash-observer.js">@torproject.org/crash-observer;1</ulink></title>
  <para>

This component detects when Firefox crashes by altering Firefox prefs during
runtime and checking for the same values at startup. It <ulink
url="https://developer.mozilla.org/en/XPCOM_Interface_Reference/nsIPrefService#savePrefFile()">synchronizes
the preference service</ulink> to ensure the altered prefs are written to disk
immediately.

</para>
</sect3>
<sect3 id="tbsessionstore">
 <title><ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/components/tbSessionStore.js">@torproject.org/torbutton-ss-blocker;1</ulink></title>
  <para>

This component subscribes to the Firefox <ulink
url="https://developer.mozilla.org/en/Observer_Notifications#Session_Store">sessionstore-state-write</ulink>
observer event to filter out URLs from tabs loaded during Tor, to prevent them
from being written to disk. To do this, it checks the
<command>__tb_tor_fetched</command> tag of tab objects before writing them out. If
the tag is from a blocked Tor state, the tab is not written to disk.  This is
a rather expensive operation that involves potentially very large JSON
evaluations and object tree traversals, but it preferable to replacing the
Firefox session store with our own implementation, which is what was done in
years past.

</para>
</sect3>

<sect3 id="refspoofer">
 <title><ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/components/torRefSpoofer.js">@torproject.org/torRefSpoofer;1</ulink></title>
 <para>
This component handles optional referrer spoofing for Torbuton. It implements a
form of "smart" referer spoofing using <ulink
url="https://developer.mozilla.org/en/Setting_HTTP_request_headers">http-on-modify-request</ulink>
to modify the Referrer header. The code sends the default browser referrer
header only if the destination domain is a suffix of the source, or if the
source is a suffix of the destination. Otherwise, it sends no referer. This
strange suffix logic is used as a heuristic: some rare sites on the web block
requests without proper referer headers, and this logic is an attempt to cater
to them. Unfortunately, it may not be enough. For example, google.fr will not
send a referer to google.com using this logic. Hence, it is off by default.
 </para>
</sect3>

<sect3 id="contentpolicy">
<title><ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/components/cssblocker.js">@torproject.org/cssblocker;1
- components/cssblocker.js</ulink></title>

<para>This is a key component to Torbutton's security measures. When Tor is
toggled, Javascript is disabled, and pages are instructed to stop loading.
However, CSS is still able to perform network operations by loading styles for
onmouseover events and other operations. In addition, favicons can still be
loaded by the browser. The cssblocker component prevents this by implementing
and registering an <ulink
url="https://developer.mozilla.org/en/nsIContentPolicy">nsIContentPolicy</ulink>.
When an nsIContentPolicy is registered, Firefox checks every attempted network
request against its <ulink
url="https://developer.mozilla.org/en/nsIContentPolicy#shouldLoad()">shouldLoad</ulink>
member function to determine if the load should proceed. In Torbutton's case,
the content policy looks up the appropriate browser tab using the <link
linkend="windowmapper">window mapper</link>,
and checks that tab's load tag against the current Tor state. If the tab was
loaded in a different state than the current state, the fetch is denied.
Otherwise, it is allowed.</para> This helps to achieve the <link
linkend="isolation">Network
Isolation</link> requirements of Torbutton.

<para>In addition, the content policy also blocks website javascript from
<ulink
url="http://webdevwonders.com/detecting-firefox-add-ons/">querying for
versions and existence of extension chrome</ulink> while Tor is enabled, and
also masks the presence of Torbutton to website javascript while Tor is
disabled. </para>

<para>

Finally, some of the work that logically belongs to the content policy is
instead handled by the <command>torbutton_http_observer</command> and
<command>torbutton_weblistener</command> in <ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/chrome/content/torbutton.js">torbutton.js</ulink>. These two objects handle blocking of
Firefox 3 favicon loads, popups, and full page plugins, which for whatever
reason are not passed to the Firefox content policy itself (see Firefox Bugs 
<ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=437014">437014</ulink> and 
<ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=401296">401296</ulink>).

</para>

<!-- 
FIXME: Hrmm, the content policy doesn't really lend itself well to display 
this way.. People looking for this much detail should consult the source.

<para>
    <table rowheader="firstcol" frame='all'><title>Access Permissions Table</title>
    <tgroup cols='5' align='left' colsep='1' rowsep='1'>
       <tbody>
       <row>
         <entry></entry>
         <entry>chrome/resource</entry>
         <entry>a3</entry>
         <entry>a4</entry>
         <entry>a5</entry>
       </row>
       <row>
         <entry>file</entry>
         <entry>b2</entry>
         <entry>b3</entry>
         <entry>b4</entry>
         <entry>b5</entry>
       </row>
       <row>
         <entry>c1</entry>
         <entry>c2</entry>
         <entry>c3</entry>
         <entry>c4</entry>
         <entry>c5</entry>
       </row>
       <row>
         <entry>d1</entry>
         <entry>d2</entry>
         <entry>d3</entry>
         <entry>d4</entry>
         <entry>d5</entry>
       </row>
       </tbody>
       </tgroup>
       </table>
</para>
-->

<para>

This helps to fulfill both the <link
linkend="setpreservation">Anonymity Set Preservation</link> and the <link
linkend="undiscoverability">Tor Undiscoverability</link> requirements of
Torbutton.</para>

</sect3>
</sect2>
</sect1>
<sect1>
 <title>Chrome</title>

<para>The chrome is where all the torbutton graphical elements and windows are
located. </para>
<sect2>
 <title>XUL Windows and Overlays</title>
<para>
Each window is described as an <ulink
url="http://developer.mozilla.org/en/docs/XUL_Reference">XML file</ulink>, with zero or more Javascript
files attached. The scope of these Javascript files is their containing
window. XUL files that add new elements and script to existing Firefox windows
are called overlays.</para>

<sect3 id="browseroverlay">
<title>Browser Overlay - <ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/chrome/content/torbutton.xul">torbutton.xul</ulink></title>

<para>The browser overlay, torbutton.xul, defines the toolbar button, the status
bar, and events for toggling the button. The overlay code is in <ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/chrome/content/torbutton.js">chrome/content/torbutton.js</ulink>.
It contains event handlers for preference update, shutdown, upgrade, and
location change events.</para>

</sect3>
<sect3>
 <title>Preferences Window - <ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/chrome/content/preferences.xul">preferences.xul</ulink></title>

<para>The preferences window of course lays out the Torbutton preferences, with
handlers located in <ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/chrome/content/preferences.js">chrome/content/preferences.js</ulink>.</para>
</sect3>
<sect3>
 <title>Other Windows</title>

<para>There are additional windows that describe popups for right clicking on
the status bar, the toolbutton, and the about page.</para>

</sect3>
</sect2>
<sect2>
 <title>Major Chrome Observers</title>
 <para>
In addition to the <link linkend="components">components described
above</link>, Torbutton also instantiates several observers in the browser
overlay window. These mostly grew due to scoping convenience, and many should
probably be relocated into their own components.
 </para>
  <orderedlist>
   <listitem><command>torbutton_window_pref_observer</command>
    <para>
This is an observer that listens for Torbutton state changes, for the purposes
of updating the Torbutton button graphic as the Tor state changes.
    </para>
   </listitem>

<listitem><command>torbutton_unique_pref_observer</command>
    <para>

This is an observer that only runs in one window, called the main window. It
listens for changes to all of the Torbutton preferences, as well as Torbutton
controlled Firefox preferences. It is what carries out the toggle path when
the proxy settings change. When the main window is closed, the
torbutton_close_window event handler runs to dub a new window the "main
window".

</para>
   </listitem>

<listitem><command>tbHistoryListener</command>
    <para>
The tbHistoryListener exists to prevent client window Javascript from
interacting with window.history to forcibly navigate a user to a tab session
history entry from a different Tor state. It also expunges the window.history
entries during toggle. This listener helps Torbutton
satisfy the <link linkend="isolation">Network Isolation</link> requirement as
well as the <link linkend="state">State Separation</link> requirement.

</para>
   </listitem>

<listitem><command>torbutton_http_observer</command>
    <para>

The torbutton_http_observer performs some of the work that logically belongs
to the content policy. This handles blocking of
Firefox 3 favicon loads, which for whatever
reason are not passed to the Firefox content policy itself (see Firefox Bugs
<ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=437014">437014</ulink> and
<ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=401296">401296</ulink>).

</para>
    <para>
The observer is also responsible for redirecting users to alternate
search engines when Google presents them with a Captcha, as well as copying
Google Captcha-related cookies between international Google domains.
    </para>
   </listitem>

<listitem><command>torbutton_proxyservice</command>
    <para>
The Torbutton proxy service handles redirecting Torbutton-related update
checks on addons.mozilla.org through Tor. This is done to help satisfy the
<link linkend="undiscoverability">Tor Undiscoverability</link> requirement.
    </para>
   </listitem>

<listitem><command>torbutton_weblistener</command>
<para>The <ulink
url="https://developer.mozilla.org/en/nsIWebProgressListener#onLocationChange">location
change</ulink> <ulink
url="https://developer.mozilla.org/en/nsIWebProgress">webprogress
listener</ulink>, <command>torbutton_weblistener</command> is one of the most
important parts of the chrome from a security standpoint. It is a <ulink
url="https://developer.mozilla.org/en/nsIWebProgressListener">webprogress
listener</ulink> that handles receiving an event every time a page load or
iframe load occurs. This class eventually calls down to
<function>torbutton_update_tags()</function> and
<function>torbutton_hookdoc()</function>, which apply the browser Tor load
state tags, plugin permissions, and install the Javascript hooks to hook the
<ulink
url="https://developer.mozilla.org/en/DOM/window.screen">window.screen</ulink>
object to obfuscate browser and desktop resolution information.

</para>
   </listitem>

</orderedlist>
 </sect2>
</sect1>

<sect1>
 <title>Toggle Code Path</title>
 <para>

The act of toggling is connected to <function>torbutton_toggle()</function>
via the <ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/chrome/content/torbutton.xul">torbutton.xul</ulink>
and <ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/chrome/content/popup.xul">popup.xul</ulink>
overlay files. Most of the work in the toggling process is present in <ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/chrome/content/torbutton.js">torbutton.js</ulink>

</para>
<para>

Toggling is a 3 stage process: Button Click, Proxy Update, and
Settings Update. These stages are reflected in the prefs
<command>extensions.torbutton.tor_enabled</command>,
<command>extensions.torbutton.proxies_applied</command>, and
<command>extensions.torbutton.settings_applied</command>. The reason for the
three stage preference update is to ensure immediate enforcement of <link
linkend="isolation">Network Isolation</link> via the <link
linkend="contentpolicy">content policy</link>. Since the content window
javascript runs on a different thread than the chrome javascript, it is
important to properly convey the stages to the content policy to avoid race
conditions and leakage, especially with <ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=409737">Firefox Bug 
409737</ulink> unfixed. The content policy does not allow any network activity
whatsoever during this three stage transition.

</para>
 <sect2>
  <title>Button Click</title>
  <para>

This is the first step in the toggling process. When the user clicks the
toggle button or the toolbar, <function>torbutton_toggle()</function> is
called. This function checks the current Tor status by comparing the current
proxy settings to the selected Tor settings, and then sets the proxy settings
to the opposite state, and sets the pref
<command>extensions.torbutton.tor_enabled</command> to reflect the new state.
It is this proxy pref update that gives notification via the <ulink
url="https://developer.mozilla.org/en/NsIPrefBranch2#addObserver.28.29">pref
observer</ulink>
<command>torbutton_unique_pref_observer</command> to perform the rest of the
toggle.

</para>
 </sect2>
 <sect2>
  <title>Proxy Update</title>
  <para>

When Torbutton receives any proxy change notifications via its
<command>torbutton_unique_pref_observer</command>, it calls
<function>torbutton_set_status()</function> which checks against the Tor
settings to see if the Tor proxy settings match the current settings. If so,
it calls <function>torbutton_update_status()</function>, which determines if
the Tor state has actually changed, and sets
<command>extensions.torbutton.proxies_applied</command> to the appropriate Tor
state value, and ensures that
<command>extensions.torbutton.tor_enabled</command> is also set to the correct
value. This is decoupled from the button click functionality via the pref
observer so that other addons (such as SwitchProxy) can switch the proxy
settings between multiple proxies.

</para>
 </sect2>

 <sect2>
  <title>Settings Update</title>
  <para>

The next stage is also handled by
<function>torbutton_update_status()</function>. This function sets scores of
Firefox preferences, saving the original values to prefs under
<command>extensions.torbutton.saved.*</command>, and performs the <link
linkend="cookiejar">cookie jarring</link>, state clearing (such as window.name
and DOM storage), and <link linkend="preferences">preference
toggling</link>. At the
end of its work, it sets
<command>extensions.torbutton.settings_applied</command>, which signifies the
completion of the toggle operation to the <link
linkend="contentpolicy">content policy</link>.

</para>
 </sect2>
<sect2 id="preferences">
<title>Firefox preferences touched during Toggle</title>
<para>
There are also a number of Firefox preferences set in
<function>torbutton_update_status()</function> that aren't governed by any
Torbutton setting. These are:
</para>
<orderedlist>

<listitem><ulink
url="http://kb.mozillazine.org/Network.security.ports.banned">network.security.ports.banned</ulink>
 <para>
Torbutton sets this setting to add ports 8123, 8118, 9050 and 9051 (which it
reads from <command>extensions.torbutton.banned_ports</command>) to the list
of ports Firefox is forbidden to access. These ports are Polipo, Privoxy, Tor,
and the Tor control port, respectively. This is set for both Tor and Non-Tor
usage, and prevents websites from attempting to do http fetches from these
ports to see if they are open, which addresses the <link
linkend="undiscoverability">Tor Undiscoverability</link> requirement.
 </para>
 </listitem>
 <listitem><ulink url="http://kb.mozillazine.org/Browser.send_pings">browser.send_pings</ulink>
 <para>
This setting is currently always disabled. If anyone ever complains saying
that they *want* their browser to be able to send ping notifications to a
page or arbitrary link, I'll make this a pref or Tor-only. But I'm not holding
my breath. I haven't checked if the content policy is called for pings, but if
not, this setting helps with meeting the <link linkend="isolation">Network
Isolation</link> requirement.
 </para>
 </listitem>
 <listitem><ulink
url="http://kb.mozillazine.org/Browser.safebrowsing.remoteLookups">browser.safebrowsing.remoteLookups</ulink>
 <para>
Likewise for this setting. I find it hard to imagine anyone who wants to ask
Google in real time if each URL they visit is safe, especially when the list
of unsafe URLs is downloaded anyway. This helps fulfill the <link
linkend="disk">Disk Avoidance</link> requirement, by preventing your entire
browsing history from ending up on Google's disks.
 </para>
 </listitem>
 <listitem><ulink
url="http://kb.mozillazine.org/Browser.safebrowsing.enabled">browser.safebrowsing.enabled</ulink>
 <para>
Safebrowsing does <ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=360387">unauthenticated
updates under Firefox 2</ulink>, so it is disabled during Tor usage. 
This helps fulfill the <link linkend="updates">Update
Safety</link> requirement. Firefox 3 has the fix for that bug, and so
safebrowsing updates are enabled during Tor usage.
 </para>
 </listitem>
 <listitem><ulink
url="http://kb.mozillazine.org/Network.protocol-handler.warn-external.%28protocol%29">network.protocol-handler.warn-external.(protocol)</ulink>
 <para>
If Tor is enabled, we need to prevent random external applications from
launching without at least warning the user. This group of settings only
partially accomplishes this, however. Applications can still be launched via
plugins. The mechanisms for handling this are described under the "Disable
Plugins During Tor Usage" preference. This helps fulfill the <link
linkend="proxy">Proxy Obedience</link> requirement, by preventing external
applications from accessing network resources at the command of Tor-fetched
pages. Unfortunately, due to <link linkend="FirefoxBugs">Firefox Bug</link>
<ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=440892">440892</ulink>,
these prefs are no longer obeyed. They are set still anyway out of respect for
the dead.
 </para>
</listitem>
  <listitem><ulink
url="http://kb.mozillazine.org/Browser.sessionstore.max_tabs_undo">browser.sessionstore.max_tabs_undo</ulink>
   <para>

To help satisfy the Torbutton <link linkend="state">State Separation</link>
and <link linkend="isolation">Network Isolation</link> requirements,
Torbutton needs to purge the Undo Tab history on toggle to prevent repeat
"Undo Close" operations from accidentally restoring tabs from a different Tor
State. This purge is accomplished by setting this preference to 0 and then
restoring it to the previous user value upon toggle.

</para>
  </listitem>

<listitem><command>security.enable_ssl2</command> or <ulink
url="http://www.oxymoronical.com/experiments/xpcomref/applications/Firefox/3.5/interfaces/nsIDOMCrypto">nsIDOMCrypto::logout()</ulink>
   <para>
TLS Session IDs can persist for an indefinite duration, providing an
identifier that is sent to TLS sites that can be used to link activity. This
is particularly troublesome now that we have certificate verification in place
in Firefox 3: The OCSP server can use this Session ID to build a history of
TLS sites someone visits, and also correlate their activity as users move from
network to network (such as home to work to coffee shop, etc), inside and
outside of Tor. To handle this and to help satisfy our <link
linkend="state">State Separation Requirement</link>, we call the logout()
function of nsIDOMCrypto. Since this may be absent, or may fail, we fall back
to toggling
<command>security.enable_ssl2</command>, which clears the SSL Session ID
cache via the pref observer at <ulink
url="http://mxr.mozilla.org/security/source/security/manager/ssl/src/nsNSSComponent.cpp">nsNSSComponent.cpp</ulink>.
   </para>
  </listitem>
  <listitem><command>security.OCSP.enabled</command>
   <para>
Similarly, we toggle <command>security.OCSP.enabled</command>, which clears the OCSP certificate
validation cache via the pref observer at <ulink
url="http://mxr.mozilla.org/security/source/security/manager/ssl/src/nsNSSComponent.cpp">nsNSSComponent.cpp</ulink>.
In this way, exit nodes will not be able to fingerprint you
based the fact that non-Tor OCSP lookups were obviously previously cached.
To handle this and to help satisfy our <link
linkend="state">State Separation Requirement</link>,
   </para>
  </listitem>
  <listitem><command><ulink
url="http://kb.mozillazine.org/Updating_extensions#Disabling_update_checks_for_individual_add-ons_-_Advanced_users">extensions.e0204bd5-9d31-402b-a99d-a6aa8ffebdca.getAddons.cache.enabled</ulink></command>
  <para>
We permanently disable addon usage statistic reporting to the
addons.mozilla.org statistics engine. These statistics send version
information about Torbutton users via non-Tor, allowing their Tor use to be
uncovered. Disabling this reporting helps Torbutton to satisfy its <link
linkend="undiscoverability">Tor Undiscoverability</link> requirement.

</para>
  </listitem>

<listitem><command><ulink url="http://www.mozilla.com/en-US/firefox/geolocation/">geo.enabled</ulink></command>
   <para>

Torbutton disables Geolocation support in Firefox 3.5 and above whenever tor
is enabled. This helps Torbutton maintain its
<link linkend="location">Location Neutrality</link> requirement.
While Firefox does prompt before divulging geolocational information,
the assumption is that Tor users will never want to give their
location away during Tor usage, and even allowing websites to prompt
them to do so will only cause confusion and accidents to happen. Moreover,
just because users may approve a site to know their location in non-Tor mode
does not mean they want it divulged during Tor mode.

</para>
  </listitem>

<listitem><command><ulink
url="http://kb.mozillazine.org/Browser.zoom.siteSpecific">browser.zoom.siteSpecific</ulink></command>
   <para>

Firefox actually remembers your zoom settings for certain sites. CSS
and Javascript rule can use this to recognize previous visitors to a site.
This helps Torbutton fulfill its <link linkend="state">State Separation</link>
requirement.

</para>
  </listitem>

<listitem><command><ulink
url="https://developer.mozilla.org/en/controlling_dns_prefetching">network.dns.disablePrefetch</ulink></command>
   <para>

Firefox 3.5 and above implement prefetching of DNS resolution for hostnames in
links on a page to decrease page load latency. While Firefox does typically
disable this behavior when proxies are enabled, we set this pref for added
safety during Tor usage. Additionally, to prevent Tor-loaded tabs from having
their links prefetched after a toggle to Non-Tor mode occurs,
we also set the docShell attribute
<ulink
url="http://www.oxymoronical.com/experiments/apidocs/interface/nsIDocShell">
allowDNSPrefetch</ulink> to false on Tor loaded tabs. This happens in the same
positions in the code as those for disabling plugins via the allowPlugins
docShell attribute. This helps Torbutton fulfill its <link
linkend="isolation">Network Isolation</link> requirement.

</para>
  </listitem>

<listitem><command><ulink
url="http://kb.mozillazine.org/Browser.cache.offline.enable">browser.cache.offline.enable</ulink></command>
   <para>

Firefox has the ability to store web applications in a special cache to allow
them to continue to operate while the user is offline. Since this subsystem
is actually different than the normal disk cache, it must be dealt with
separately. Thus, Torbutton sets this preference to false whenever Tor is
enabled. This helps Torbutton fulfill its <link linkend="disk">Disk
Avoidance</link> and <link linkend="state">State Separation</link>
requirements.

</para>
  </listitem>

</orderedlist>
</sect2>

</sect1>

<sect1>
 <title>Description of Options</title>
<para>This section provides a detailed description of Torbutton's options. Each
option is presented as the string from the preferences window, a summary, the
preferences it touches, and the effect this has on the components, chrome, and
browser properties.</para>

 <sect2>
  <title>Proxy Settings</title>
 <sect3>
  <title>Test Settings</title>
  <para>
This button under the Proxy Settings tab provides a way to verify that the 
proxy settings are correct, and actually do route through the Tor network. It
performs this check by issuing an <ulink
url="http://developer.mozilla.org/en/docs/XMLHttpRequest">XMLHTTPRequest</ulink>
for <ulink
url="https://check.torproject.org/?TorButton=True">https://check.torproject.org/?Torbutton=True</ulink>.
This is a special page that returns very simple, yet well-formed XHTML that
Torbutton can easily inspect for a hidden link with an id of
<command>TorCheckResult</command> and a target of <command>success</command>
or <command>failure</command> to indicate if the
user hit the page from a Tor IP, a non-Tor IP. This check is handled in
<function>torbutton_test_settings()</function> in <ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/chrome/content/torbutton.js">torbutton.js</ulink>.
Presenting the results to the user is handled by the <ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/chrome/content/preferences.xul">preferences
window</ulink>
callback <function>torbutton_prefs_test_settings()</function> in <ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/chrome/content/preferences.js">preferences.js</ulink>.

</para>
 </sect3>
 </sect2>
 <sect2>
  <title>Dynamic Content Settings</title>
 <sect3 id="plugins">
  <title>Disable plugins on Tor Usage (crucial)</title>
 <para>Option: <command>extensions.torbutton.no_tor_plugins</command></para>

<para>Java and plugins <ulink
url="http://java.sun.com/j2se/1.5.0/docs/api/java/net/class-use/NetworkInterface.html">can query</ulink> the <ulink
url="http://www.rgagnon.com/javadetails/java-0095.html">local IP
address</ulink> and report it back to the
remote site. They can also <ulink
url="http://decloak.net">bypass proxy settings</ulink> and directly connect to a
remote site without Tor. Every browser plugin we have tested with Firefox has
some form of network capability, and every one ignores proxy settings or worse - only
partially obeys them. This includes but is not limited to:
QuickTime, Windows Media Player, RealPlayer, mplayerplug-in, AcroRead, and
Flash.

</para>
 <para>
Enabling this preference causes the above mentioned Torbutton chrome web progress
 listener <command>torbutton_weblistener</command> to disable Java via <command>security.enable_java</command> and to disable
 plugins via the browser <ulink
 url="https://developer.mozilla.org/en/XUL%3aProperty%3adocShell">docShell</ulink>
 attribute <command>allowPlugins</command>. These flags are set every time a new window is
 created (<function>torbutton_tag_new_browser()</function>), every time a web
load
event occurs
 (<function>torbutton_update_tags()</function>), and every time the tor state is changed
 (<function>torbutton_update_status()</function>). As a backup measure, plugins are also
 prevented from loading by the content policy in <ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/components/cssblocker.js">@torproject.org/cssblocker;1</ulink> if Tor is
 enabled and this option is set.
 </para>

<para>All of this turns out to be insufficient if the user directly clicks
on a plugin-handled mime-type. <ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=401296">In this case</ulink>,
the browser decides that maybe it should ignore all these other settings and
load the plugin anyways, because maybe the user really did want to load it
(never mind this same load-style could happen automatically  with meta-refresh
or any number of other ways..). To handle these cases, Torbutton stores a list
of plugin-handled mime-types, and sets the pref
<command>plugin.disable_full_page_plugin_for_types</command> to this list.
Additionally, (since nothing can be assumed when relying on Firefox
preferences and internals) if it detects a load of one of them from the web
progress listener, it cancels the request, tells the associated DOMWindow to
stop loading, clears the document, AND throws an exception. Anything short of
all this and the plugin managed to find some way to load.
 </para>

<!--

FIXME: Hrmm, technically this behavior is not covered by this pref.

<para>
Furthermore, with version 3.0 and above, Firefox
<ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=440892">began ignoring</ulink>

<ulink
url="http://kb.mozillazine.org/Network.protocol-handler.warn-external.%28protocol%29">network.protocol-handler.warn-external.(protocol)</ulink>
prefs, which caused us to have to <link linkend="appblocker">wrap the external
app launcher components</link> to prevent external apps from being loaded to
bypass proxy settings.
 </para>
-->

<para>
 All this could be avoided, of course, if Firefox would either <ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=401296">obey
 allowPlugins</ulink> for directly visited URLs, or notify its content policy for such
 loads either <ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=309524">via</ulink> <ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=380556">shouldProcess</ulink> or shouldLoad. The fact that it does not is
 not very encouraging.
 </para>

<para>

Since most plugins completely ignore browser proxy settings, the actions
performed by this setting are crucial to satisfying the <link
linkend="proxy">Proxy Obedience</link> requirement.

</para>
</sect3>
<sect3>
 <title>Isolate Dynamic Content to Tor State (crucial)</title>

<para>Option: <command>extensions.torbutton.isolate_content</command></para>

<para>Enabling this preference is what enables the <ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/components/cssblocker.js">@torproject.org/cssblocker;1</ulink> content policy
mentioned above, and causes it to block content load attempts in pages an
opposite Tor state from the current state. Freshly loaded <ulink
url="https://developer.mozilla.org/en/XUL/tabbrowser">browser
tabs</ulink> are tagged
with a <command>__tb_load_state</command> member in
<function>torbutton_update_tags()</function> and this
value is compared against the current tor state in the content policy.</para>

<para>It also kills all Javascript in each page loaded under that state by
toggling the <command>allowJavascript</command> <ulink
url="https://developer.mozilla.org/en/XUL%3aProperty%3adocShell">docShell</ulink> property, and issues a
<ulink
url="https://developer.mozilla.org/en/XPCOM_Interface_Reference/nsIWebNavigation#stop()">webNavigation.stop(webNavigation.STOP_ALL)</ulink> to each browser tab (the
equivalent of hitting the STOP button).</para>

<para>

Unfortunately, <ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=409737">Firefox bug
409737</ulink> prevents <command>docShell.allowJavascript</command> from killing
all event handlers, and event handlers registered with <ulink
url="http://developer.mozilla.org/en/docs/DOM:element.addEventListener">addEventListener()</ulink>
are still able to execute. The <link linkend="contentpolicy">Torbutton Content
Policy</link> should prevent such code from performing network activity within
the current tab, but activity that happens via a popup window or via a
Javascript redirect can still slip by. For this reason, Torbutton blocks
popups by checking for a valid <ulink
url="http://developer.mozilla.org/en/docs/DOM:window.opener">window.opener</ulink>
attribute in <function>torbutton_check_progress()</function>. If the window
has an opener from a different Tor state, its load is blocked. The content
policy also takes similar action to prevent Javascript redirects. This also
has the side effect/feature of preventing the user from following any links
from a page loaded in an opposite Tor state.

</para>

<para>
This setting is responsible for satisfying the <link
linkend="isolation">Network Isolation</link> requirement.
</para>

</sect3>
<sect3 id="jshooks">

<title>Hook Dangerous Javascript</title>

<para>Option: <command>extensions.torbutton.kill_bad_js</command></para>

<para>This setting enables injection of the <ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/chrome/content/jshooks.js">Javascript
hooking code</ulink>. This is done in the chrome in
<function>torbutton_hookdoc()</function>, which is called ultimately by both the 
<ulink
url="https://developer.mozilla.org/en/nsIWebProgressListener">webprogress
listener</ulink> <command>torbutton_weblistener</command> and the <link
linkend="contentpolicy">content policy</link> (the latter being a hack to handle
javascript: urls).

In the Firefox 2 days, this option did a lot more than
it does now. It used to be responsible for timezone and improved useragent
spoofing, and history object cloaking. However, now it only provides
obfuscation of the <ulink
url="https://developer.mozilla.org/en/DOM/window.screen">window.screen</ulink>
object to mask your browser and desktop resolution.
The resolution hooks
effectively make the Firefox browser window appear to websites as if the renderable area
takes up the entire desktop, has no toolbar or other GUI element space, and
the desktop itself has no toolbars.
These hooks drastically reduce the amount of information available to do <link
linkend="fingerprinting">anonymity set reduction attacks</link> and help to
meet the <link linkend="setpreservation">Anonymity Set Preservation</link>
requirements. Unfortunately, Gregory Fleischer discovered it is still possible
to retrieve the original screen values by using <ulink
url="http://pseudo-flaw.net/tor/torbutton/unmask-sandbox-xpcnativewrapper.html">XPCNativeWrapper</ulink>
or <ulink
url="http://pseudo-flaw.net/tor/torbutton/unmask-components-lookupmethod.html">Components.lookupMethod</ulink>.
We are still looking for a workaround as of Torbutton 1.3.2.

</para>
</sect3>
<sect3>
<title>Resize windows to multiples of 50px during Tor usage (recommended)</title>

<para>Option: <command>extensions.torbutton.resize_windows</command></para>

<para>

This option drastically cuts down on the number of distinct anonymity sets
that divide the Tor web userbase. Without this setting, the dimensions for a
typical browser window range from 600-1200 horizontal pixels and 400-1000
vertical pixels, or about 600x600 = 360000 different sets. Resizing the
browser window to multiples of 50 on each side reduces the number of sets by
50^2, bringing the total number of sets to 144. Of course, the distribution
among these sets are not uniform, but scaling by 50 will improve the situation
due to this non-uniformity for users in the less common resolutions.
Obviously the ideal situation would be to lie entirely about the browser
window size, but this will likely cause all sorts of rendering issues, and is
also not implementable in a foolproof way from extension land.

</para>
<para>

The implementation of this setting is spread across a couple of different
locations in the Torbutton javascript <link linkend="browseroverlay">browser
overlay</link>. Since resizing minimized windows causes them to be restored,
and since maximized windows remember their previous size to the pixel, windows
must be resized before every document load (at the time of browser tagging)
via <function>torbutton_check_round()</function>, called by
<function>torbutton_update_tags()</function>. To prevent drift, the extension
tracks the original values of the windows and uses this to perform the
rounding on document load. In addition, to prevent the user from resizing a
window to a non-50px multiple, a resize listener
(<function>torbutton_do_resize()</function>) is installed on every new browser
window to record the new size and round it to a 50px multiple while Tor is
enabled. In all cases, the browser's contentWindow.innerWidth and innerHeight
are set. This ensures that there is no discrepancy between the 50 pixel cutoff
and the actual renderable area of the browser (so that it is not possible to
infer toolbar size/presence by the distance to the nearest 50 pixel roundoff).

</para>
<para>
This setting helps to meet the <link
linkend="setpreservation">Anonymity Set Preservation</link> requirements.
</para>
</sect3>
<sect3>

<title>Disable Search Suggestions during Tor (recommended)</title>

<para>Option: <command>extensions.torbutton.no_search</command></para>

<para>
This setting causes Torbutton to disable <ulink
url="http://kb.mozillazine.org/Browser.search.suggest.enabled"><command>browser.search.suggest.enabled</command></ulink>
during Tor usage.
This governs if you get Google search suggestions during Tor
usage. Your Google cookie is transmitted with google search suggestions, hence
this is recommended to be disabled.

</para>
<para>
While this setting doesn't satisfy any Torbutton requirements, the fact that
cookies are transmitted for partially typed queries does not seem desirable
for Tor usage.
</para>
</sect3>

<sect3>
<title>Disable Updates During Tor</title>

<para>Option: <command>extensions.torbutton.no_updates</command></para>

<para>This setting causes Torbutton to disable the four <ulink
url="http://wiki.mozilla.org/Update:Users/Checking_For_Updates#Preference_Controls_and_State">Firefox
update settings</ulink> during Tor
  usage: <command>extensions.update.enabled</command>,
<command>app.update.enabled</command>,
  <command>app.update.auto</command>, and
<command>browser.search.update</command>.  These prevent the
  browser from updating extensions, checking for Firefox upgrades, and
  checking for search plugin updates while Tor is enabled.
  </para>
<para>
This setting satisfies the <link
linkend="updates">Update Safety</link> requirement.
</para>
</sect3>
<sect3>
<title>Redirect Torbutton Updates Via Tor (recommended)</title>

<para>Option: <command>extensions.torbutton.update_torbutton_via_tor</command></para>

<para>This setting causes Torbutton to install an

<ulink
url="https://developer.mozilla.org/en/nsIProtocolProxyFilter">nsIProtocolProxyFilter</ulink>
in order to redirect all version update checks and Torbutton update downloads
via Tor, regardless of if Tor is enabled or not. This was done both to address
concerns about data retention done by <ulink
url="https://www.addons.mozilla.org">addons.mozilla.org</ulink>, as well as to
help censored users meet the <link linkend="undiscoverability">Tor
Undiscoverability</link> requirement.

</para>
</sect3>

<sect3>
<title>Disable livemarks updates during Tor usage (recommended)</title>
  <para>Option:
   <simplelist>
   <member><command>extensions.torbutton.disable_livemarks</command></member>
   </simplelist>
  </para>

<para>

This option causes Torbutton to prevent Firefox from loading <ulink
url="http://www.mozilla.com/firefox/livebookmarks.html">Livemarks</ulink> during
Tor usage. Because people often have very personalized Livemarks (such as RSS
feeds of Wikipedia articles they maintain, etc). This is accomplished both by
<link linkend="livemarks">wrapping the livemark-service component</link> and
by calling stopUpdateLivemarks() on the <ulink
url="http://www.oxymoronical.com/experiments/xpcomref/applications/Firefox/3.5/components/%40mozilla.org/browser/livemark-service;2">Livemark
service</ulink> when Tor is enabled.

</para>

<para>
This helps satisfy the <link linkend="isolation">Network
Isolation</link> and <link linkend="setpreservation">Anonymity Set
Preservation</link> requirements.
</para>

</sect3>
<sect3>
<title>Block Tor/Non-Tor access to network from file:// urls (recommended)</title>
  <para>Options:
   <simplelist>
   <member><command>extensions.torbutton.block_tor_file_net</command></member>
   <member><command>extensions.torbutton.block_nontor_file_net</command></member>
   </simplelist>
  </para>

<para>

These settings prevent file urls from performing network operations during the
respective Tor states. Firefox 2's implementation of same origin policy allows
file urls to read and <ulink
url="http://www.gnucitizen.org/blog/content-disposition-hacking/">submit
arbitrary files from the local filesystem</ulink> to arbitrary websites. To
make matters worse, the 'Content-Disposition' header can be injected
arbitrarily by exit nodes to trick users into running arbitrary html files in
the local context. These preferences cause the <link
linkend="contentpolicy">content policy</link> to block access to any network
resources from File urls during the appropriate Tor state.

</para>
<para>

This preference helps to ensure Tor's <link linkend="isolation">Network
Isolation</link> requirement, by preventing file urls from executing network
operations in opposite Tor states. Also, allowing pages to submit arbitrary
files to arbitrary sites just generally seems like a bad idea.

</para>
</sect3>

<sect3>

<title>Close all Tor/Non-Tor tabs and windows on toggle (optional)</title>

<para>Options:
   <simplelist>
   <member><command>extensions.torbutton.close_nontor</command></member>
   <member><command>extensions.torbutton.close_tor</command></member>
   </simplelist>
  </para>

<para>

These settings cause Torbutton to enumerate through all windows and close all
tabs in each window for the appropriate Tor state. This code can be found in
<function>torbutton_update_status()</function>.  The main reason these settings
exist is as a backup mechanism in the event of any Javascript or content policy
leaks due to <ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=409737">Firefox Bug
409737</ulink>.  Torbutton currently tries to block all Javascript network
activity via the content policy, but until that bug is fixed, there is some
risk that there are alternate ways to bypass the policy. This option is
available as an extra assurance of <link linkend="isolation">Network
Isolation</link> for those who would like to be sure that when Tor is toggled
all page activity has ceased. It also serves as a potential future workaround
in the event a content policy failure is discovered, and provides an additional
level of protection for the <link linkend="disk">Disk Avoidance</link>
protection so that browser state is not sitting around waiting to be swapped
out longer than necessary.

</para>
<para>
While this setting doesn't satisfy any Torbutton requirements, the fact that
cookies are transmitted for partially typed queries does not seem desirable
for Tor usage.
</para>
</sect3>
 </sect2>
 <sect2>
  <title>History and Forms Settings</title>
<sect3>
<title>Isolate Access to History navigation to Tor state (crucial)</title>
  <para>Option: <command>extensions.torbutton.block_js_history</command></para>
  <para>
This setting determines if Torbutton installs an <ulink
url="http://www.oxymoronical.com/experiments/apidocs/interface/nsISHistoryListener">nsISHistoryListener</ulink>
attached to the <ulink
url="http://www.oxymoronical.com/experiments/apidocs/interface/nsISHistory">sessionHistory</ulink> of 
of each browser's <ulink
url="https://developer.mozilla.org/en/XUL%3aProperty%3awebNavigation">webNavigatator</ulink>.
The nsIShistoryListener is instantiated with a reference to the containing
browser window and blocks the back, forward, and reload buttons on the browser
navigation bar when Tor is in an opposite state than the one to load the
current tab. In addition, Tor clears the session history during a new document
load if this setting is enabled.

</para>
  <para>

This is marked as a crucial setting in part
because Javascript access to the history object is indistinguishable from 
user clicks, and because
<ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=409737">Firefox Bug
409737</ulink> allows javascript to execute in opposite Tor states, javascript
can issue reloads after Tor toggle to reveal your original IP. Even without
this bug, however, Javascript is still able to access previous pages in your
session history that may have been loaded under a different Tor state, to
attempt to correlate your activity.

</para>
   <para>

This setting helps to fulfill Torbutton's <link linkend="state">State
Separation</link> and (until Bug 409737 is fixed) <link linkend="isolation">Network Isolation</link>
requirements.

</para>
</sect3>

<sect3>
<title>History Access Settings</title>

<para>Options:
  <simplelist>
   <member><command>extensions.torbutton.block_thread</command></member>
   <member><command>extensions.torbutton.block_nthread</command></member>
   <member><command>extensions.torbutton.block_thwrite</command></member>
   <member><command>extensions.torbutton.block_nthwrite</command></member>
  </simplelist>
  </para>

<para>On Firefox 3.x, these four settings govern the behavior of the <ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/components/ignore-history.js">components/ignore-history.js</ulink>
history blocker component mentioned above. By hooking the browser's view of
the history itself via the <ulink
url="http://www.oxymoronical.com/experiments/xpcomref/applications/Firefox/3.5/components/%40mozilla.org/browser/global-history;2">@mozilla.org/browser/global-history;2</ulink>
and <ulink
url="http://www.oxymoronical.com/experiments/xpcomref/applications/Firefox/3.5/components/%40mozilla.org/browser/nav-history-service;1">@mozilla.org/browser/nav-history-service;1</ulink>
components, this mechanism defeats all document-based <ulink
url="http://whattheinternetknowsaboutyou.com/">history disclosure
attacks</ulink>, including <ulink
url="http://ha.ckers.org/weird/CSS-history.cgi">CSS-only attacks</ulink>.

The component also hooks functions involved in writing history to disk via
both the <ulink
url="http://developer.mozilla.org/en/docs/Places_migration_guide#History">Places
Database</ulink> and the older Firefox 2 mechanisms.

</para>

<para>
On Firefox 4, Mozilla finally <ulink
url="https://developer.mozilla.org/en/CSS/Privacy_and_the_%3avisited_selector">addressed
these issues</ulink>, so we can effectively ignore the "read" pair of the
above prefs. We then only need to link the write prefs to
<command>places.history.enabled</command>, which disabled writing to the
history store while set.
</para>

<para>
This setting helps to satisfy the <link
linkend="state">State Separation</link> and <link
linkend="disk">Disk Avoidance</link> requirements.
</para>

</sect3>
<sect3>

<title>Clear History During Tor Toggle (optional)</title>

<para>Option: <command>extensions.torbutton.clear_history</command></para>

<para>This setting governs if Torbutton calls
<ulink
url="https://developer.mozilla.org/en/nsIBrowserHistory#removeAllPages.28.29">nsIBrowserHistory.removeAllPages</ulink>
and <ulink
url="http://www.oxymoronical.com/experiments/apidocs/interface/nsISHistory">nsISHistory.PurgeHistory</ulink>
for each tab on Tor toggle.</para>
<para>
This setting is an optional way to help satisfy the <link
linkend="state">State Separation</link> requirement.
</para>

</sect3>
<sect3>
<title>Block Password+Form saving during Tor/Non-Tor</title>

<para>Options:
  <simplelist>
  <member><command>extensions.torbutton.block_tforms</command></member>
  <member><command>extensions.torbutton.block_ntforms</command></member>
  </simplelist>
  </para>

<para>These settings govern if Torbutton disables
<command>browser.formfill.enable</command>
and <command>signon.rememberSignons</command> during Tor and Non-Tor usage.
Since form fields can be read at any time by Javascript, this setting is a lot
more important than it seems.
</para>

<para>
This setting helps to satisfy the <link
linkend="state">State Separation</link> and <link
linkend="disk">Disk Avoidance</link> requirements.
</para>

</sect3>
 </sect2>
 <sect2>
  <title>Cache Settings</title>
<sect3>
  <title>Block Tor disk cache and clear all cache on Tor Toggle</title>

<para>Option: <command>extensions.torbutton.clear_cache</command>
  </para>

<para>This option causes Torbutton to call <ulink
url="https://developer.mozilla.org/en/nsICacheService#evictEntries.28.29">nsICacheService.evictEntries(0)</ulink>
on Tor toggle to remove all entries from the cache. In addition, this setting
causes Torbutton to set <ulink
url="http://kb.mozillazine.org/Browser.cache.disk.enable">browser.cache.disk.enable</ulink> to false.
</para>
<para>
This setting helps to satisfy the <link
linkend="state">State Separation</link> and <link
linkend="disk">Disk Avoidance</link> requirements.
</para>

</sect3>
<sect3>
  <title>Block disk and memory cache during Tor</title>

<para>Option: <command>extensions.torbutton.block_cache</command></para>

<para>This setting
causes Torbutton to set <ulink
url="http://kb.mozillazine.org/Browser.cache.memory.enable">browser.cache.memory.enable</ulink>,
<ulink
url="http://kb.mozillazine.org/Browser.cache.disk.enable">browser.cache.disk.enable</ulink> and
<ulink
url="http://kb.mozillazine.org/Network.http.use-cache">network.http.use-cache</ulink> to false during tor usage.
</para>
<para>
This setting helps to satisfy the <link
linkend="state">State Separation</link> and <link
linkend="disk">Disk Avoidance</link> requirements.
</para>

</sect3>
 </sect2>
 <sect2>
  <title>Cookie and Auth Settings</title>
<sect3>
  <title>Clear Cookies on Tor Toggle</title>

<para>Option: <command>extensions.torbutton.clear_cookies</command>
  </para>

<para>

This setting causes Torbutton to call <ulink
url="https://developer.mozilla.org/en/nsICookieManager#removeAll.28.29">nsICookieManager.removeAll()</ulink> on
every Tor toggle. In addition, this sets <ulink
url="http://kb.mozillazine.org/Network.cookie.lifetimePolicy">network.cookie.lifetimePolicy</ulink>
to 2 for Tor usage, which causes all cookies to be demoted to session cookies,
which prevents them from being written to disk.

</para>
<para>
This setting helps to satisfy the <link
linkend="state">State Separation</link> and <link
linkend="disk">Disk Avoidance</link> requirements.
</para>

</sect3>
<sect3>
  
  <title>Store Non-Tor cookies in a protected jar</title>

<para>Option: <command>extensions.torbutton.cookie_jars</command>
  </para>

<para>

This setting causes Torbutton to use <ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/components/cookie-jar-selector.js">@torproject.org/cookie-jar-selector;2</ulink> to store
non-tor cookies in a cookie jar during Tor usage, and clear the Tor cookies
before restoring the jar.
</para>
<para>
This setting also sets <ulink
url="http://kb.mozillazine.org/Network.cookie.lifetimePolicy">network.cookie.lifetimePolicy</ulink>
to 2 for Tor usage, which causes all cookies to be demoted to session cookies,
which prevents them from being written to disk.

</para>

<para>
This setting helps to satisfy the <link
linkend="state">State Separation</link> and <link
linkend="disk">Disk Avoidance</link> requirements.
</para>

</sect3>
<sect3>

<title>Store both Non-Tor and Tor cookies in a protected jar (dangerous)</title>

<para>Option: <command>extensions.torbutton.dual_cookie_jars</command>
  </para>

<para>

<para>
This setting helps to satisfy the <link
linkend="state">State Separation</link> requirement.
</para>

</sect3>

<sect3>

<title>Manage My Own Cookies (dangerous)</title>

<para>Options: None</para>
<para>This setting disables all Torbutton cookie handling by setting the above
cookie prefs all to false.</para>
</sect3>
<sect3>

<sect3>
  <title>Do not write Tor/Non-Tor cookies to disk</title>
  <para>Options:
  <simplelist>
  <member><command>extensions.torbutton.tor_memory_jar</command></member>
  <member><command>extensions.torbutton.nontor_memory_jar</command></member>
  </simplelist>
  </para>

<para>
These settings (contributed by arno) cause Torbutton to set <ulink
url="http://kb.mozillazine.org/Network.cookie.lifetimePolicy">network.cookie.lifetimePolicy</ulink>
to 2 during the appropriate Tor state, and to store cookies acquired in that
state into a Javascript
<ulink
url="http://developer.mozilla.org/en/docs/Core_JavaScript_1.5_Guide:Processing_XML_with_E4X">E4X</ulink>
object as opposed to writing them to disk.
</para>

<para>
This allows Torbutton to provide an option to preserve a user's 
cookies while still satisfying the <link linkend="disk">Disk Avoidance</link>
requirement.
</para>
</sect3>

<title>Disable DOM Storage during Tor usage (crucial)</title>

<para>Option: <command>extensions.torbutton.disable_domstorage</command>
  </para>

<para>

This setting causes Torbutton to toggle <command>dom.storage.enabled</command> during Tor
usage to prevent 
<ulink
  url="http://developer.mozilla.org/en/docs/DOM:Storage">DOM Storage</ulink> from
  being used to store persistent information across Tor states.</para>
<para>
This setting helps to satisfy the <link
linkend="state">State Separation</link> requirement.
</para>

</sect3>

<sect3>
  <title>Clear HTTP Auth on Tor Toggle (recommended)</title>
<para>Option: <command>extensions.torbutton.clear_http_auth</command>
  </para>

<para>
This setting causes Torbutton to call <ulink
url="http://www.oxymoronical.com/experiments/apidocs/interface/nsIHttpAuthManager">nsIHttpAuthManager.clearAll()</ulink>
every time Tor is toggled.
</para>

<para>
This setting helps to satisfy the <link
linkend="state">State Separation</link> requirement.
</para>
</sect3>
 </sect2>
 <sect2>
  <title>Startup Settings</title>
<sect3>
  <title>On Browser Startup, set Tor state to: Tor, Non-Tor</title>
  <para>Options:
   <command>extensions.torbutton.restore_tor</command>
  </para>

<para>This option governs what Tor state tor is loaded in to.
<function>torbutton_set_initial_state()</function> covers the case where the
browser did not crash, and <function>torbutton_crash_recover()</function>
covers the case where the <link linkend="crashobserver">crash observer</link>
detected a crash.
</para>
<para>

Since the Tor state after a Firefox crash is unknown/indeterminate, this
setting helps to satisfy the <link linkend="state">State Separation</link>
requirement in the event of Firefox crashes by ensuring all cookies,
settings and saved sessions are reloaded from a fixed Tor state.
 
</para>
</sect3>

<sect3>
  <title>Prevent session store from saving Non-Tor/Tor-loaded tabs</title>

<para>Options: 
  <simplelist>
    <member><command>extensions.torbutton.nonontor_sessionstore</command></member>
    <member><command>extensions.torbutton.notor_sessionstore</command></member>
  </simplelist>
  </para>

<para>If these options are enabled, the <link
linkend="tbsessionstore">tbSessionStore.js</link> component uses the session
store listeners to filter out the appropriate tabs before writing the session
store data to disk.
</para>
<para>
This setting helps to satisfy the <link linkend="disk">Disk Avoidance</link>
requirement, and also helps to satisfy the <link
linkend="state">State Separation</link> requirement in the event of Firefox
crashes.

</para>

</sect3>
 </sect2>
 <sect2>
  <title>Shutdown Settings</title>
<sect3>

<title>Clear cookies on Tor/Non-Tor shutdown</title>

<para>Option: <command>extensions.torbutton.shutdown_method</command>
  </para>

<para> This option variable can actually take 3 values: 0, 1, and 2. 0 means no
cookie clearing, 1 means clear only during Tor-enabled shutdown, and 2 means
clear for both Tor and Non-Tor shutdown. When set to 1 or 2, Torbutton listens
for the <ulink
url="http://developer.mozilla.org/en/docs/Observer_Notifications#Application_shutdown">quit-application-granted</ulink> event in
<link linkend="crashobserver">crash-observer.js</link> and use <ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/components/cookie-jar-selector.js">@torproject.org/cookie-jar-selector;2</ulink>
to clear out all cookies and all cookie jars upon shutdown.
</para>
<para>
This setting helps to satisfy the <link
linkend="state">State Separation</link> requirement.
</para>

</sect3>
 </sect2>
 <sect2>
  <title>Header Settings</title>
<sect3>

<title>Set user agent during Tor usage (crucial)</title>
  <para>Options:
   <simplelist>
    <member><command>extensions.torbutton.set_uagent</command></member>
    <member><command>extensions.torbutton.platform_override</command></member>
    <member><command>extensions.torbutton.oscpu_override</command></member>
    <member><command>extensions.torbutton.buildID_override</command></member>
    <member><command>extensions.torbutton.productsub_override</command></member>
    <member><command>extensions.torbutton.appname_override</command></member>
    <member><command>extensions.torbutton.appversion_override</command></member>
    <member><command>extensions.torbutton.useragent_override</command></member>
    <member><command>extensions.torbutton.useragent_vendor</command></member>
    <member><command>extensions.torbutton.useragent_vendorSub</command></member>
  </simplelist>
   </para>

<para>On face, user agent switching appears to be straight-forward in Firefox.
It provides several options for controlling the browser user agent string:
<command>general.appname.override</command>,
<command>general.appversion.override</command>,
<command>general.platform.override</command>,
<command>general.oscpu.override</command>,
<command>general.productSub.override</command>,
<command>general.buildID.override</command>,
<command>general.useragent.override</command>,
<command>general.useragent.vendor</command>, and
<command>general.useragent.vendorSub</command>. If
the Torbutton preference <command>extensions.torbutton.set_uagent</command> is
true, Torbutton copies all of the other above prefs into their corresponding
browser preferences during Tor usage.</para>

<para>

It also turns out that it is possible to detect the original Firefox version
by <ulink url="http://ha.ckers.org/blog/20070516/read-firefox-settings-poc/">inspecting
certain resource:// files</ulink>. These cases are handled by Torbutton's
<link linkend="contentpolicy">content policy</link>.

</para>

<para>
This setting helps to satisfy the <link
linkend="setpreservation">Anonymity Set Preservation</link> requirement.
</para>

</sect3>
<sect3>

<title>Spoof US English Browser</title>
<para>Options:
<simplelist>
 <member><command>extensions.torbutton.spoof_english</command></member>
 <member><command>extensions.torbutton.spoof_charset</command></member>
 <member><command>extensions.torbutton.spoof_language</command></member>
</simplelist>
</para>

<para> This option causes Torbutton to set
<command>general.useragent.locale</command>
<command>intl.accept_languages</command> to the value specified in
<command>extensions.torbutton.spoof_locale</command>,
<command>extensions.torbutton.spoof_charset</command> and
<command>extensions.torbutton.spoof_language</command> during Tor usage, as
well as hooking <command>navigator.language</command> via its <link
linkend="jshooks">javascript hooks</link>.
 </para>
<para>
This setting helps to satisfy the <link
linkend="setpreservation">Anonymity Set Preservation</link> and <link
linkend="location">Location Neutrality</link> requirements.
</para>

</sect3>

<sect3>
  <title>Referer Spoofing Options</title>

<para>Option: <command>extensions.torbutton.refererspoof</command>
</para>

<para>
This option variable has three values. If it is 0, "smart" referer spoofing is
enabled. If it is 1, the referer behaves as normal. If it is 2, no referer is
sent. The default value is 1. The smart referer spoofing is implemented by the
<link linkend="refspoofer">torRefSpoofer</link> component.

</para>
<para>
This setting also does not directly satisfy any Torbutton requirement, but
some may desire to mask their referrer for general privacy concerns.
</para>
</sect3>

<sect3>
  <title>Strip platform and language off of Google Search Box queries</title>

<para>Option: <command>extensions.torbutton.fix_google_srch</command>
</para>

<para>

This option causes Torbutton to use the <ulink
url="https://wiki.mozilla.org/Search_Service:API">@mozilla.org/browser/search-service;1</ulink>
component to wrap the Google search plugin. On many platforms, notably Debian
and Ubuntu, the Google search plugin is set to reveal a lot of language and
platform information. This setting strips off that info while Tor is enabled.

</para>
<para>
This setting helps Torbutton to fulfill its <link
linkend="setpreservation">Anonymity Set Preservation</link> requirement.
</para>
</sect3>

<sect3>
  <title>Automatically use an alternate search engine when presented with a
Google Captcha</title>

<para>Options:
<simplelist>
 <member><command>extensions.torbutton.asked_google_captcha</command></member>
 <member><command>extensions.torbutton.dodge_google_captcha</command></member>
 <member><command>extensions.torbutton.google_redir_url</command></member>
</simplelist>
</para>

<para>

Google's search engine has rate limiting features that cause it to
<ulink
url="http://googleonlinesecurity.blogspot.com/2007/07/reason-behind-were-sorry-message.html">present
captchas</ulink> and sometimes even outright ban IPs that issue large numbers
of search queries, especially if a lot of these queries appear to be searching
for software vulnerabilities or unprotected comment areas.

</para>
<para>

Despite multiple discussions with Google, we were unable to come to a solution
or any form of compromise that would reduce the number of captchas and
outright bans seen by Tor users issuing regular queries.

</para>
<para>
As a result, we've implemented this option as an <ulink
url="https://developer.mozilla.org/en/XUL_School/Intercepting_Page_Loads#HTTP_Observers">'http-on-modify-request'</ulink>
http observer to optionally redirect banned or captcha-triggering Google
queries to search engines that do not rate limit Tor users. The current
options are duckduckgo.com, ixquick.com, bing.com, yahoo.com and scroogle.org. These are
encoded in the preferences
<command>extensions.torbutton.redir_url.[1-5]</command>.

</para>
</sect3>

<sect3>

<title>Store SSL/CA Certs in separate jars for Tor/Non-Tor (recommended)</title>

<para>Options:
<simplelist>
 <member><command>extensions.torbutton.jar_certs</command></member>
 <member><command>extensions.torbutton.jar_ca_certs</command></member>
</simplelist>
</para>
<para>

These settings govern if Torbutton attempts to isolate the user's SSL
certificates into separate jars for each Tor state. This isolation is
implemented in <function>torbutton_jar_certs()</function> in <ulink
url="https://gitweb.torproject.org/torbutton.git/blob_plain/HEAD:/src/chrome/content/torbutton.js">chrome/content/torbutton.js</ulink>,
which calls <function>torbutton_jar_cert_type()</function> and
<function>torbutton_unjar_cert_type()</function> for each certificate type in
the <ulink
url="http://www.oxymoronical.com/experiments/xpcomref/applications/Firefox/3.5/components/%40mozilla.org/security/nsscertcache;1">@mozilla.org/security/nsscertcache;1</ulink>.
Certificates are deleted from and imported to the <ulink
url="http://www.oxymoronical.com/experiments/xpcomref/applications/Firefox/3.5/components/%40mozilla.org/security/x509certdb;1">@mozilla.org/security/x509certdb;1</ulink>.
</para>

<para>
The first time this pref is used, a backup of the user's certificates is
created in their profile directory under the name
<filename>cert8.db.bak</filename>. This file can be copied back to
<filename>cert8.db</filename> to fully restore the original state of the
user's certificates in the event of any error.
</para>

<para>
Since exit nodes and malicious sites can insert content elements sourced to
specific SSL sites to query if a user has a certain certificate,
this setting helps to satisfy the <link linkend="state">State
Separation</link> requirement of Torbutton. Unfortunately, <ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=435159">Firefox Bug
435159</ulink> prevents it from functioning correctly in the event of rapid Tor toggle, so it
is currently not exposed via the preferences UI.

</para>

</sect3>

</sect2>
</sect1>

<sect1 id="FirefoxBugs">
  <title>Relevant Firefox Bugs</title>
  <para>

</para>
  <sect2 id="FirefoxSecurity">
   <title>Bugs impacting security</title>
   <para>

Torbutton has to work around a number of Firefox bugs that impact its
security. Most of these are mentioned elsewhere in this document, but they
have also been gathered here for reference. In order of decreasing severity,
they are:

</para>
   <orderedlist>
    <listitem><ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=429070">Bug 429070 - exposing
Components.interfaces to untrusted content leaks information about installed
extensions</ulink>
     <para>
<ulink url="http://pseudo-flaw.net/">Gregory Fleischer</ulink> demonstrated at Defcon 17 that these interfaces can
also be used to <ulink
url="http://pseudo-flaw.net/tor/torbutton/fingerprint-firefox.html">fingerprint
Firefox down the to the minor version</ulink>. Note that his test has not been
updated since 3.5.3, hence it reports 3.5.3 for more recent Firefoxes. This
bug interferes with Torbutton's ability to satisfy its <link
linkend="setpreservation">Anonymity Set Preservation</link> requirement.
     </para>
    </listitem>

<listitem><ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=280661">Bug 280661 - SOCKS proxy server
connection timeout hard-coded</ulink>
    <para>

This bug prevents us from using the Firefox SOCKS layer directly, and
currently requires us to ship an auxiliary HTTP proxy called <ulink
url="http://www.pps.jussieu.fr/~jch/software/polipo/">Polipo</ulink>. If this
patch were landed, we would no longer need to ship Polipo, which has a number
of privacy and security issues of its own (in addition to being unmaintained).

</para>
   </listitem>
   <listitem><ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=418986">Bug 418986 - window.screen
provides a large amount of identifiable information</ulink>
   <para>

As <link linkend="fingerprinting">mentioned above</link>, a large amount of
information is available from <ulink
url="http://developer.mozilla.org/en/docs/DOM:window.screen">window.screen</ulink>.
The most sensitive data to anonymity is actually that which is not used in
rendering - such as desktop resolution, and window decoration size.
Currently, there is no way to obscure this information without Javascript
hooking. In addition, many of this same desktop and window decoration
resolution information is available via <ulink
url="https://developer.mozilla.org/En/CSS/Media_queries">CSS Media
Queries</ulink>, so perhaps some more lower-level rendering controls or
preferences need to be provided. These issues interfere with Torbutton's
ability to fulfill its <link linkend="setpreservation">Anonymity Set
Preservation</link> requirement.

</para>
   </listitem>
   <listitem><ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=435159">Bug 435159 -
nsNSSCertificateDB::DeleteCertificate has race conditions</ulink>
      <para>

In Torbutton 1.2.0rc1, code was added to attempt to isolate SSL certificates
the user has installed. Unfortunately, the method call to delete a certificate
from the current certificate database acts lazily: it only sets a variable
that marks a cert for deletion later, and it is not cleared if that
certificate is re-added. This means that if the Tor state is toggled quickly,
that certificate could remain present until it is re-inserted (causing an
error dialog), and worse, it would still be deleted after that.  The lack of
this functionality is considered a Torbutton security bug because cert
isolation is considered a <link linkend="state">State Separation</link>
feature.

</para>
     </listitem>
     <listitem>Give more visibility into and control over TLS
negotiation
     <para>

There are several <ulink
url="https://trac.torproject.org/projects/tor/ticket/2482">TLS issues
impacting Torbutton security</ulink>. It is not clear if these should be one
Firefox bug or several, but in particular we need better control over various
aspects of TLS connections. Firefox currently provides no observer capable of
extracting TLS parameters or certificates early enough to cancel a TLS
request. We would like to be able to provide <ulink
url="https://www.eff.org/https-everywhere">HTTPS-Everywhere</ulink> users with
the ability to <ulink
url="https://trac.torproject.org/projects/tor/wiki/HTTPSEverywhere/SSLObservatorySubmission">have
their certificates audited</ulink> by a <ulink
url="http://www.networknotary.org/">Perspectives</ulink>-style set of
notaries. The problem with this is that the API observer points do not exist
for any Firefox addon to actually block authentication token submission over a
TLS channel, so every addon to date (including Perspectives) is actually
providing users with notification *after* their authentication tokens have
already been compromised. This obviously needs to be fixed.
     </para>
     </listitem>
     <listitem><ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=575230">Bug 575230 - Provide option to
reduce precision of Date()</ulink>
      <para>

Currently it is possible to <ulink
url="http://arstechnica.com/tech-policy/news/2010/02/firm-uses-typing-cadence-to-finger-unauthorized-users.ars">fingerprint
users based on their typing cadence</ulink> using the high precision timer
available to javascript. Using this same precision, it is possible to compute
an identifier based upon the clock drift of the client from some nominal
source. The latter is not much of a concern for Tor users, as the variable
delay to load and run a page is measured on the order of seconds, but the high
precision timer can still be used to fingerprint aspects of a browser's
javascript engine and processor, and apparently also a user's typing cadence.
This bug hinders Torbutton's ability to satisfy its <link
linkend="setpreservation">Anonymity Set Preservation</link> requirement.

</para>
     </listitem>
    <listitem><ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=122752">SOCKS
Username/Password Support</ulink>
    <para>
We need <ulink url="https://developer.mozilla.org/en/nsIProxyInfo">Firefox
APIs</ulink> or about:config settings to control the SOCKS Username and
Password fields. The reason why we need this support is to utilize an (as yet
unimplemented) scheme to separate Tor traffic based <ulink
url="https://gitweb.torproject.org/torspec.git/blob_plain/HEAD:/proposals/171-separate-streams.txt">on
SOCKS username/password</ulink>.
    </para>
    </listitem>

<listitem><ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=409737">Bug 409737 -
javascript.enabled and docShell.allowJavascript do not disable all event
handlers</ulink>
     <para>

This bug allows pages to execute javascript via addEventListener and perhaps
other callbacks. In order to prevent this bug from enabling an attacker to
break the <link linkend="isolation">Network Isolation</link> requirement,
Torbutton 1.1.13 began blocking popups and history manipulation from different
Tor states.  So long as there are no ways to open popups or redirect the user
to a new page, the <link linkend="contentpolicy">Torbutton content
policy</link> should block Javascript network access. However, if there are
ways to open popups or perform redirects such that Torbutton cannot block
them, pages may still have free reign to break that requirement and reveal a
user's original IP address.

</para>
     </listitem>
     <listitem><ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=448743">Bug 448743 -
Decouple general.useragent.locale from spoofing of navigator.language</ulink>
     <para>

Currently, Torbutton spoofs the <command>navigator.language</command>
attribute via <link linkend="jshooks">Javascript hooks</link>. Unfortunately,
these do not work on Firefox 3. It would be ideal to have
a pref to set this value (something like a
<command>general.useragent.override.locale</command>),
to avoid fragmenting the anonymity set of users of foreign locales. This issue
impedes Torbutton from fully meeting its <link
linkend="setpreservation">Anonymity Set Preservation</link>
requirement on Firefox 3.

</para>
     </listitem>
    </orderedlist>
  </sect2>
  <sect2 id="FirefoxWishlist">
   <title>Bugs blocking functionality</title>
   <para>
The following bugs impact Torbutton and similar extensions' functionality.
   </para>

<listitem><ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=445696">Bug 445696 -
Extensions cannot determine if Firefox is full screen</ulink>
   <para>

The windowState property of <ulink
url="https://developer.mozilla.org/en/XUL/window">ChromeWindows</ulink> does not accurately reflect the true
state of the window in some cases on Linux. This causes Torbutton to attempt
to resize maximized and minimized windows when it should not.

</para>
   </listitem>
   <listitem><ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=629820">nsIContentPolicy::shouldLoad not
called for web request in Firefox Mobile</ulink>
    <para>

The new <ulink
url="https://wiki.mozilla.org/Mobile/Fennec/Extensions/Electrolysis">Electrolysis</ulink>
multiprocess system appears to have some pretty rough edge cases with respect
to registering XPCOM category managers such as the nsIContentPolicy, which
make it difficult to do a straight-forward port of Torbutton or
HTTPS-Everywhere to Firefox Mobile.  It probably also has similar issues with
wrapping existing <link linkend="hookedxpcom">Firefox XPCOM components</link>,
which will also cause more problems for porting Torbutton.

</para>
   </listitem>
   <listitem><ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=290456">Bug 290456 -
Block/clear Flash MX "cookies" as well</ulink>
   <para>

Today, it is possible to allow plugins if you have a transparent proxy such as
<ulink url="http://anonymityanywhere.com/incognito/">Incognito</ulink> to prevent proxy bypass. However, flash cookies can still be used to
link your Tor and Non-Tor activity, and this reveal your IP to an adversary
that does so. This can be solved by manually removing your flash cookies (like
<ulink
url="https://addons.mozilla.org/en-US/firefox/addon/6623">BetterPrivacy</ulink> does), but
it would be nice if there was a standard way to do this from a Firefox API.

</para>
   </listitem>
   <listitem><ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=417869">Bug 417869 -
Browser context is difficult to obtain from many XPCOM callbacks</ulink>
   <para>

It is difficult to determine which tabbrowser many XPCOM callbacks originate
from, and in some cases absolutely no context information is provided at all.
While this doesn't have much of an effect on Torbutton, it does make writing
extensions that would like to do per-tab settings and content filters (such as
FoxyProxy) difficult to impossible to implement securely.

</para>
   </listitem>
<!--
FIXME: This doesn't really apply anymore.
   <listitem><ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=418321">Bug 418321 -
Components do not expose disk interfaces</ulink>
   <para>

Several components currently provide no way of reimplementing their disk
access to easily satisfy Torbutton's <link linkend="disk">Disk
Avoidance</link> requirements. Workarounds exist, but they are <link
linkend="sessionstore">clunky</link>, and
some of them involve disabling functionality during Tor usage.

</para>
   </listitem>
-->

<!--
FIXME: Need to use new observer methods if possible
   <listitem><ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=448741">Bug 448741 -
nsISessionStore uses private methods and is not extensible</ulink>
   <para>

Similar to the above bug, in the specific case of the sessionstore component,
the API is not amenable to Contract ID hooking, and this requires that
Torbutton include modified copies of this component for Firefox 2 and 3, which
has <ulink
url="https://bugs.torproject.org/flyspray/index.php?do=details&amp;id=722">raised
objections</ulink> from some developers.

</para>
   </listitem>
   <listitem><ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=439384">Bug 439384 -
"profile-do-change" event does not cause cookie table reload</ulink>
   <para>

In Firefox 3, the change to the new SQLlite database for cookie storage has a
bug that prevents Torbutton's cookie jaring from working properly. The
"profile-do-change" observer event no longer properly causes either a sync or
reload of the cookie database from disk after it is copied into place.
Torbutton currently works around this by issuing the SQLLite queries manually
to store and rebuild the cookie database.

</para>
   </listitem>

<listitem><ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=248970">Bug 248970 (PrivateBrowsing) - Private Browsing mode (global toggle for
saving/caching everything)</ulink>
   <para>

This bug catalogs the discussion of a 'Private Mode' in Firefox that would
perform many, but not all, of the activities of Torbutton. It would be useful
to leverage the resulting setting to simplify Torbutton. This bug is listed so
we can track this progress and ensure that it doesn't end up defining
behaviors contrary to and incompatible with Torbutton's requirements (though a
subset of the <link linkend="requirements">requirements</link> is of course fine).

</para>
   </listitem>
-->

</orderedlist>
  </sect2>
  <sect2 id="FirefoxMiscBugs">
   <title>Low Priority Bugs</title>
   <para>
The following bugs have an effect upon Torbutton, but are superseded by more
practical and more easily fixable variant bugs above; or have stable, simple
workarounds.
  </para>

<orderedlist>
    <listitem><ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=435151">Bug 435151 - XPCSafeJSObjectWrapper breaks evalInSandbox</ulink>
     <para>

Under Firefox 3, the XPCSafeJSObjectWrapper breaks when you try to use
constructors of classes defined from within the scope of the sandbox, among
other things. This prevents Torbutton from applying the Timezone hooks under
Firefox 3, but a better solution for Torbutton's specific date hooking needs 
would be a fix for the above mentioned Bug 392274. Of course, many more
extensions may be interested in the sandbox hooking functionality working
properly though.

</para>
     </listitem>
     <listitem><ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=440892">Bug 440892 -
network.protocol-handler.warn-external are ignored</ulink>
     <para>

Sometime in the Firefox 3 development cycle, the preferences that governed
warning a user when external apps were launched got disconnected from the code
that does the launching. Torbutton depended on these prefs to prevent websites
from launching specially crafted documents and application arguments that
caused Proxy Bypass. We currently work around this issue by <link
linkend="appblocker">wrapping the app launching components</link> to present a
popup before launching external apps while Tor is enabled. While this works,
it would be nice if these prefs were either fixed or removed.

</para>
     </listitem>
    <listitem><ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=437014">Bug 437014 -
nsIContentPolicy::shouldLoad no longer called for favicons</ulink>
    <para>

Firefox 3.0 stopped calling the shouldLoad call of content policy for favicon
loads. Torbutton had relied on this call to block favicon loads for opposite
Tor states. The workaround it employs for Firefox 3 is to cancel the request
when it arrives in the <command>torbutton_http_observer</command> used for
blocking full page plugin loads. This seems to work just fine, but is a bit
dirty.

</para>
    </listitem>
<!--
    <listitem><ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=437016">Bug 437016 -
nsIContentPolicy::shouldLoad not called for livemarks</ulink>
    <para>

An alternative fix for the livemarks bug above would be to block livemarks
fetches from the content policy. Unfortunately shouldLoad is not called for
livemarks fetches.

</para>
    </listitem>
-->
 
     <listitem><ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=309524">Bug 309524</ulink>
and <ulink url="https://bugzilla.mozilla.org/show_bug.cgi?id=380556">Bug
380556</ulink> - nsIContentPolicy::shouldProcess is not called.
     <para>

This is a call that would be useful to develop a better workaround for the
allowPlugins issue above. If the content policy were called before a URL was
handed over to a plugin or helper app, it would make the workaround for the
above allowPlugins bug a lot cleaner. Obviously this bug is not as severe as
the others though, but it might be nice to have this API as a backup.

</para>
     </listitem>

<listitem><ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=401296">Bug 401296 - docShell.allowPlugins
not honored for direct links</ulink> (Perhaps subset of <ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=282106">Bug 282106</ulink>?)
     <para>

Similar to the javascript plugin disabling attribute, the plugin disabling
attribute is also not perfect &mdash; it is ignored for direct links to plugin
handled content, as well as meta-refreshes to plugin handled content.  This
requires Torbutton to listen to a number of different http events to intercept
plugin-related mime type URLs and cancel their requests. Again, since plugins
are quite horrible about obeying proxy settings, loading a plugin pretty much
ensures a way to break the <link linkend="isolation">Network Isolation</link>
requirement and reveal a user's original IP address. Torbutton's code to
perform this workaround has been subverted at least once already by Kyle
Williams.

</para>
     </listitem>
   <listitem><ulink
url="https://bugzilla.mozilla.org/show_bug.cgi?id=419598">Bug 419598 - 'var
Date' is deletable</ulink>
     <para>

Based on Page 62 of the <ulink
url="http://www.ecma-international.org/publications/files/ECMA-ST/Ecma-262.pdf">ECMA-262
Javascript spec</ulink>, it seems like it should be possible to do something
like the following to prevent the Date object from being unmasked:
<screen>
with(window) {
    var Date = fakeDate;
    var otherVariable = 42;
}

delete window.Date; // Should fail. Instead succeeds, revealing original Date.
delete window.otherVariable; // Fails, leaving window.otherVariable set to 42.
</screen>

From the ECMA-262 spec:

<blockquote>
If the variable statement occurs inside a FunctionDeclaration, the variables
are defined with function-local scope in that function, as described in
s10.1.3. Otherwise, they are defined with global scope (that is, they are
created as members of the global object, as described in 10.1.3) using
property attributes { DontDelete }. Variables are created when the execution
scope is entered. A Block does not define a new execution scope. Only Program
and FunctionDeclaration produce a new scope. Variables are initialized to
undefined when created. A variable with an Initialiser is assigned the value
of its AssignmentExpression when the VariableStatement is executed, not when
the variable is created.
</blockquote>

In fact, this is exactly how the with statement with a variable declaration
behaves <emphasis>for all other variables other than ones that shadow system
variables</emphasis>. Some variables (such as
<command>window.screen</command>, and <command>window.history</command>) can't
even be shadowed in this way, and give an error about lacking a setter. If
such shadowing were possible, it would greatly simplify the Javascript hooking
code, which currently relies on undocumented semantics of
<command>__proto__</command> to copy the original values in the event of a
delete. This <command>__proto__</command> hack unfortunately does not work for
the Date object though.

</para>
    </listitem>

</orderedlist>
  </sect2>
</sect1>

<sect1 id="TestPlan">
  <title>Testing</title>
  <para>

The purpose of this section is to cover all the known ways that Tor browser
security can be subverted from a penetration testing perspective. The hope
is that it will be useful both for creating a &quot;Tor Safety Check&quot;
page, and for developing novel tests and actively attacking Torbutton with the
goal of finding vulnerabilities in either it or the Mozilla components,
interfaces and settings upon which it relies.

</para>
  <sect2 id="SingleStateTesting">
   <title>Single state testing</title>
   <para>

Torbutton is a complicated piece of software. During development, changes to
one component can affect a whole slough of unrelated features.  A number of
aggregated test suites exist that can be used to test for regressions in
Torbutton and to help aid in the development of Torbutton-like addons and
other privacy modifications of other browsers. Some of these test suites exist
as a single automated page, while others are a series of pages you must visit
individually. They are provided here for reference and future regression
testing, and also in the hope that some brave soul will one day decide to
combine them into a comprehensive automated test suite.

<orderedlist>
      <listitem><ulink url="http://decloak.net/">Decloak.net</ulink>
       <para>

Decloak.net is the canonical source of plugin and external-application based
proxy-bypass exploits. It is a fully automated test suite maintained by <ulink
url="http://digitaloffense.net/">HD Moore</ulink> as a service for people to
use to test their anonymity systems.

</para>
      </listitem>
      <listitem><ulink url="http://deanonymizer.com/">Deanonymizer.com</ulink>
       <para>

Deanonymizer.com is another automated test suite that tests for proxy bypass
and other information disclosure vulnerabilities. It is maintained by Kyle
Williams, the author of <ulink url="http://www.janusvm.com/">JanusVM</ulink>
and <ulink url="http://www.januspa.com/">JanusPA</ulink>.

</para>
      </listitem>
      <listitem><ulink url="https://www.jondos.de/en/anontest">JonDos
AnonTest</ulink>
       <para>

The <ulink url="https://www.jondos.de">JonDos people</ulink> also provide an
anonymity tester. It is more focused on HTTP headers than plugin bypass, and
points out a couple of headers Torbutton could do a better job with
obfuscating.

</para>
      </listitem>
      <listitem><ulink url="http://browserspy.dk">Browserspy.dk</ulink>
       <para>

Browserspy.dk provides a tremendous collection of browser fingerprinting and
general privacy tests. Unfortunately they are only available one page at a
time, and there is not really solid feedback on good vs bad behavior in
the test results.

</para>
      </listitem>
      <listitem><ulink url="http://analyze.privacy.net/">Privacy
Analyzer</ulink>
       <para>

The Privacy Analyzer provides a dump of all sorts of browser attributes and
settings that it detects, including some information on your origin IP
address. Its page layout and lack of good vs bad test result feedback makes it
not as useful as a user-facing testing tool, but it does provide some
interesting checks in a single page.

</para>
      </listitem>
      <listitem><ulink url="http://ha.ckers.org/mr-t/">Mr. T</ulink>
       <para>

Mr. T is a collection of browser fingerprinting and deanonymization exploits
discovered by the <ulink url="http://ha.ckers.org">ha.ckers.org</ulink> crew
and others. It is also not as user friendly as some of the above tests, but it
is a useful collection.

</para>
      </listitem>
      <listitem>Gregory Fleischer's <ulink
url="http://pseudo-flaw.net/content/tor/torbutton/">Torbutton</ulink> and
<ulink
url="http://pseudo-flaw.net/content/defcon/dc-17-demos/d.html">Defcon
17</ulink> Test Cases
       <para>

Gregory Fleischer has been hacking and testing Firefox and Torbutton privacy
issues for the past 2 years. He has an excellent collection of all his test
cases that can be used for regression testing. In his Defcon work, he
demonstrates ways to infer Firefox version based on arcane browser properties.
We are still trying to determine the best way to address some of those test
cases.

</para>
      </listitem>
      <listitem><ulink url="https://torcheck.xenobite.eu/index.php">Xenobite's
TorCheck Page</ulink>
       <para>

This page checks to ensure you are using a valid Tor exit node and checks for
some basic browser properties related to privacy. It is not very fine-grained
or complete, but it is automated and could be turned into something useful
with a bit of work.

</para>
      </listitem>
     </orderedlist>
    </para>
  </sect2>
  <sect2>
   <title>Multi-state testing</title>
   <para>

The tests in this section are geared towards a page that would instruct the
user to toggle their Tor state after the fetch and perform some operations:
mouseovers, stray clicks, and potentially reloads.

</para>
   <sect3>
    <title>Cookies and Cache Correlation</title>
    <para>
The most obvious test is to set a cookie, ask the user to toggle tor, and then
have them reload the page. The cookie should no longer be set if they are
using the default Torbutton settings. In addition, it is possible to leverage
the cache to <ulink
url="http://crypto.stanford.edu/sameorigin/safecachetest.html">store unique
identifiers</ulink>. The default settings of Torbutton should also protect
against these from persisting across Tor Toggle.

</para>
   </sect3>
   <sect3>
    <title>Javascript timers and event handlers</title>
    <para>

Javascript can set timers and register event handlers in the hopes of fetching
URLs after the user has toggled Torbutton. 
    </para>
   </sect3>
   <sect3>
    <title>CSS Popups and non-script Dynamic Content</title>
    <para>

Even if Javascript is disabled, CSS is still able to 
<ulink url="http://www.tjkdesign.com/articles/css%20pop%20ups/">create popup-like
windows</ulink>
via the 'onmouseover' CSS attribute, which can cause arbitrary browser
activity as soon as the mouse enters into the content window. It is also
possible for meta-refresh tags to set timers long enough to make it likely
that the user has toggled Tor before fetching content.

</para>
   </sect3>
  </sect2>
  <sect2 id="HackTorbutton">
   <title>Active testing (aka How to Hack Torbutton)</title>
   <para>

The idea behind active testing is to discover vulnerabilities in Torbutton to
bypass proxy settings, run script in an opposite Tor state, store unique
identifiers, leak location information, or otherwise violate <link
linkend="requirements">its requirements</link>. Torbutton has ventured out
into a strange and new security landscape. It depends on Firefox mechanisms
that haven't necessarily been audited for security, certainly not for the
threat model that Torbutton seeks to address. As such, it and the interfaces
it depends upon still need a 'trial by fire' typical of new technologies. This
section of the document was written with the intention of making that period
as fast as possible. Please help us get through this period by considering
these attacks, playing with them, and reporting what you find (and potentially
submitting the test cases back to be run in the standard batch of Torbutton
tests.

</para>
   <sect3>
    <title>Some suggested vectors to investigate</title>
    <para>
    <itemizedlist>
     <listitem>Strange ways to register Javascript <ulink
url="http://en.wikipedia.org/wiki/DOM_Events">events</ulink> and <ulink
url="http://www.devshed.com/c/a/JavaScript/Using-Timers-in-JavaScript/">timeouts</ulink> should
be verified to actually be ineffective after Tor has been toggled.</listitem>
     <listitem>Other ways to cause Javascript to be executed after
<command>javascript.enabled</command> has been toggled off.</listitem>
     <listitem>Odd ways to attempt to load plugins. Kyle Williams has had
some success with direct loads/meta-refreshes of plugin-handled URLs.</listitem>
     <listitem>The Date and Timezone hooks should be verified to work with
crazy combinations of iframes, nested iframes, iframes in frames, frames in
iframes, and popups being loaded and
reloaded in rapid succession, and/or from one another. Think race conditions and deep, 
parallel nesting, involving iframes from both <ulink
url="http://en.wikipedia.org/wiki/Same_origin_policy">same-origin and
non-same-origin</ulink> domains.</listitem>
     <listitem>In addition, there may be alternate ways and other
methods to query the timezone, or otherwise use some of the Date object's
methods in combination to deduce the timezone offset. Of course, the author
tried his best to cover all the methods he could foresee, but it's always good
to have another set of eyes try it out.</listitem>
     <listitem>Similarly, is there any way to confuse the <link
linkend="contentpolicy">content policy</link>
mentioned above to cause it to allow certain types of page fetches? For
example, it was recently discovered that favicons are not fetched by the
content, but the chrome itself, hence the content policy did not look up the
correct window to determine the current Tor tag for the favicon fetch. Are
there other things that can do this? Popups? Bookmarklets? Active bookmarks? </listitem>
     <listitem>Alternate ways to store and fetch unique identifiers. For example, <ulink
url="http://developer.mozilla.org/en/docs/DOM:Storage">DOM Storage</ulink>
caught us off guard. 
It was
also discovered by <ulink url="http://pseudo-flaw.net">Gregory
Fleischer</ulink> that <ulink
url="http://pseudo-flaw.net/content/tor/torbutton/">content window access to
chrome</ulink> can be used to build <link linkend="fingerprinting">unique
identifiers</link>. 
Are there any other
arcane or experimental ways that Firefox provides to create and store unique
identifiers? Or perhaps unique identifiers can be queried or derived from
properties of the machine/browser that Javascript has access to? How unique
can these identifiers be?
     </listitem>
     <listitem>Is it possible to get the browser to write some history to disk
(aside from swap) that can be retrieved later? By default, Torbutton should
write no history, cookie, or other browsing activity information to the
harddisk.</listitem>
     <listitem>Do popup windows make it easier to break any of the above
behavior? Are javascript events still canceled in popups? What about recursive
popups from Javascript, data, and other funky URL types? What about CSS
popups? Are they still blocked after Tor is toggled?</listitem>
     <listitem>Chrome-escalation attacks. The interaction between the
Torbutton chrome Javascript and the client content window javascript is pretty
well-defined and carefully constructed, but perhaps there is a way to smuggle
javascript back in a return value, or otherwise inject network-loaded
javascript into the chrome (and thus gain complete control of the browser).
</listitem>
</itemizedlist>

</para>
   </sect3>
  </sect2>
</sect1>
</article>

tor-webwml.git

Update Torbutton design doc.