Detecting and Defending Against
Third-Party Tracking on the Web
Franziska Roesner, Tadayoshi Kohno, David Wetherall
April 26, 2012
NSDI
Third-Party Web Tracking
Advertise.com
Tracker.com
Bigger browsing profiles
= increased value for trackers
= reduced privacy for users
(Hypothetical tracking relationships only.)
April 26, 2012
Franziska Roesner
2
Tracking is Complicated
• Much discussion of tracking, but limited
understanding of how it actually works.
• Our goals:
– Understand the tracking ecosystem.
• How is tracking actually done in the wild?
• What kinds of browsing profiles do trackers compile?
• How effective are defenses available to users?
– Address gaps with new defense (ShareMeNot).
April 26, 2012
Franziska Roesner
3
Outline
• How Tracking Works
– Tracking Mechanisms
– Tracking Taxonomy
• Measurements
• Defenses
April 26, 2012
Franziska Roesner
4
Mechanisms Required By Trackers
• Ability to store user identity in the browser
– Browser cookies
– HTML5 LocalStorage and Flash cookies (LSOs)
– Not considering more exotic storage mechanisms or
approximate fingerprinting
• Ability to communicate visited page and user
identity back to tracker
– Identity: Cookies attached to requests
– Visited page: HTTP referrers
– Both: scripts that embed information in URLs
April 26, 2012
Franziska Roesner
5
Tracking: The Simple Version
• Within-Site: First-party cookies are used to track
repeat visits to a site.
• Cross-Site: Third-party cookies are used by trackers
included in other sites to create profiles.
tracker.com
http://site1.com
http://site2.com
Cookie Database
tracker.com:
id=789
cookie:
id=789
<iframe
src=tracker.com/a
d.html>
ad
cookie:
id=789
</iframe>
April 26, 2012
Franziska Roesner
processing engine
9:30am: user 789
visited site1.com
9:31am: user 789
visited site2.com logs
6
Our Tracking Taxonomy
Name
N/A
Scope
User Visits
Directly?
Within-Site Yes
Site does its own on-site analytics.
Evolution: Embedding analytics libraries
Analytics
Within-Site No
Vanilla
Cross-Site No
Evolution: Third-party cookie blocking
Forced
Cross-Site Yes
(forced)
Overview
Site uses third-party analytics engine (e.g.,
Google Analytics).
Site embeds third-party tracker that uses
third-party storage (e.g., Doubleclick).
Site embeds third-party tracker that forced
the user to visit directly (e.g., via popup).
Referred
Cross-Site
No
Tracker relies on another cross-site tracker to
leak unique identifier values.
Personal
Cross-Site
Yes
Site embeds third-party tracker that the user
otherwise visits directly (e.g., Facebook).
April 26, 2012
Franziska Roesner
7
Quirks of Third-Party Cookie Blocking
• Option blocks the setting of third-party
cookies: all browsers
• Option blocks the sending of third-party
cookies: only Firefox
• Result: Once a third-party cookie is somehow
set, it can be used (in most browsers).
April 26, 2012
Franziska Roesner
8
Forced Tracking
http://site1.com
<iframe
src=tracker.com/a
d.html>
tracker.com
processing engine
High-level point:
On most browsers,adif a tracker can ever set a cookie,
tracker.com:
</iframe> popup
id=321
third-party cookie
blocking is rendered ineffective.
Cookie Database
http://site2.com
<iframe
src=tracker.com/a
d.html>
ad
logs
1:30pm:
site1.com: user
321
1:31pm:
site2.com: user
321
</iframe>
April 26, 2012
Franziska Roesner
9
Our Tracking Taxonomy
Type (Name)
N/A
Scope
User Visits
Directly?
Within-Site Yes
Site does its own on-site analytics.
Evolution: Embedding analytics libraries
Analytics
Within-Site No
Vanilla
Cross-Site No
Evolution: Third-party cookie blocking
Forced
Cross-Site Yes
(forced)
Evolution: Complex ad networks
Overview
Site uses third-party analytics engine (e.g.,
Google Analytics).
Site embeds third-party tracker that uses
third-party storage (e.g., Doubleclick).
Site embeds third-party tracker that forced
the user to visit directly (e.g., via popup).
Referred
Cross-Site
No
Tracker relies on another cross-site tracker to
leak unique identifier values.
Personal
Cross-Site
Yes
Site embeds third-party tracker that the user
otherwise visits directly (e.g., Facebook).
April 26, 2012
Franziska Roesner
10
Referred Tracking
http://site1.com
tracker.com
processing engine
<iframe
src=tracker.com/a
d.html>
High-level point:
ad
One tracker with
client-side state can enable
tracker.com:
</iframe>
id=522tracking by
partners without
client-side state.
othertracker.com/tra
Cookie Database
http://site2.com
<iframe
src=tracker.com/a
d.html>
ck?id=522&referrer=
site1.com
processing engine
ad
</iframe>
othertracker.com/tra
ck?id=522&referrer
=site2.com
April 26, 2012
Franziska Roesner
othertracker.com
logs
2:34pm:
site1.com: user
522
2:35pm:
site2.com: user
522
11
Our Tracking Taxonomy
Type (Name) Scope
N/A
User Visits Overview
Directly?
Within-Site Yes
Site does its own on-site analytics.
Evolution: Embedding analytics libraries
Analytics
Within-Site No
Vanilla
Cross-Site No
Evolution: Third-party cookie blocking
Forced
Cross-Site Yes
(forced)
Evolution: Complex ad networks
Site uses third-party analytics engine (e.g.,
Google Analytics).
Site embeds third-party tracker that uses
third-party storage (e.g., Doubleclick).
Site embeds third-party tracker that forced
the user to visit directly (e.g., via popup).
Referred
Cross-Site
No
Tracker relies on another cross-site tracker to
leak unique identifier values.
Personal
Cross-Site
Yes
Site embeds third-party tracker that the user
otherwise visits directly (e.g., Facebook).
Evolution: Social networks
April 26, 2012
Franziska Roesner
12
Personal Tracking
• Just loading these buttons (not clicking on
them) enables tracking.
• Users visit these sites directly.
• This tracking is often not anonymous (linked
to accounts).
April 26, 2012
Franziska Roesner
13
Our Tracking Taxonomy
Type (Name) Scope
N/A
User Visits Overview
Directly?
Within-Site Yes
Site does its own on-site analytics.
Evolution: Embedding analytics libraries
Analytics
Within-Site No
Vanilla
Cross-Site No
Evolution: Third-party cookie blocking
Forced
Cross-Site Yes
(forced)
Evolution: Complex ad networks
Site uses third-party analytics engine (e.g.,
Google Analytics).
Site embeds third-party tracker that uses
third-party storage (e.g., Doubleclick).
Site embeds third-party tracker that forced
Anonymous
the user to visit directly (e.g., via popup).
Referred
Cross-Site
No
Tracker relies on another cross-site tracker to
leak unique identifier values.
Personal
Cross-Site
Yes
Site embeds third-party tracker that the user
otherwise visits directly (e.g., Facebook).
Evolution: Social networks
April 26, 2012
Franziska Roesner
14
Outline
• How Tracking Works
– Tracking mechanisms
– Tracking taxonomy
• Measurements
• Defenses
April 26, 2012
Franziska Roesner
15
Measurement Tool: TrackingTracker
•
•
•
•
Firefox add-on
Based on taxonomy of client-side mechanisms
Crawls the web, automatically categorizing trackers
Monitors:
– Third-party requests
– Cookies, HTML5 LocalStorage, Flash LSOs
(considers state that changes across clean measurement runs)
– Identifier leaks
April 26, 2012
Franziska Roesner
16
Measurement Study
• 3 data sets
– Alexa Top 500
• 5 pages per domain: main page and up to 4 links
– Alexa Non-Top 500
• Sites ranked #501, #601, #701, etc.
• 5 pages per domain: main page and up to 4 links
– AOL search logs
• 300 unique queries for 35 random users
April 26, 2012
Franziska Roesner
17
Tracking Prevalence (Top 500)
• 524 unique trackers on 500 domains
457 domains (91%) embed at least one tracker.
(97% of those include at least one cross-site tracker.)
50% of domains embed
between 4 and 5 trackers.
One domain
includes
43 trackers.
April 26, 2012
Franziska Roesner
18
-,
./
01
21
9- .34
: ; 56
. 75
?1 / 5.< - 8
5/ 5= $
65
;
72
-@
/ 5 , - - = / >$
1@ - 75
9@ - , - 8
/ .
B: 6/ 1 / 75- $
12 @5A 8
>6 75 $
/@ - 8
>D C/ 75 $
<E 3</ 1 / @ 8 $
.9 >9 75<8 8 1 8 > 8 $
@D 21 75-@ ,/ 8
.9 @75 $
D <9 8
/
19 @/ 75- $
C/ C6 8 $
@4 5<7
6<2 2/
19 , 7 >$
9> 5A< 8
<2 19 675- $
C<> 2 8
/ 8 F67 $
6/ / 5@C 9<1 8
<2 75 $
,
3- 0636 8 $
19 : >: 759> ; / 8 $
A< 75
65 - 8
; . 927 $
8 : / = 5- 8
/9 1
<1 <75- $
G. 8
/F $
758
$
,-
! "#$%&"'( "&) #*&+$&',- '. / 0 #1+23'
( %%$
) %%$
*%%$
! &%$
! %%$
&%$
! / 4'56'! "#$%&"2'/ +'! / 4'766'. / 0 #1+2'
) &%$
H <>A<20I <>/ $
*#+$
J@
- 660I <>/ $KL/ @
6- 21.M
$
J@
- 660I <>/ $KN2- 238 - : 6M
$
*&%$
! " #$
! &( $! ( #$
! %#$! %&$
#) $ " ! $
' %$
( &$ ( ( $( %$ ) ( $ ) ) $
April 26, 2012
Franziska Roesner
) *$) *$ ) %$*#$ *+$ *' $
%$
19
*+
,
45 - . ,/
.- .0
8.
+) 12
)9
) 0 - 3"
-.
7
59 ) ) 1. )
(9 7 6
,"
; * 8- 5 1. )
52 9. : 6 "
38 1.
-9 )6
3= <- 1. "
/> ) 6
?/- 5 - 91 "
,( 3( . ) 6
/6 6 5 6 3 "
9= 25 1. )
) 9 7- 6 "
,( 91.
=
/( ) 6
"
5( 9- 1. )
<- <8 6 "
9@ . /12
8
5( /271 3"
( 3 .)
:/ 6"
/2 5( 81. )
</3 2 6
- A8 "
8- 6 - 1. )
9< ( /5 6 "
/2 1.
7
)
?) B8?8 6 "
5( *3* 1. )
( 3 +- 6 "
: / 1.
8. ) 6
+ , ( 21 "
6 *- 0 . ) 6
-( 5
/5 /1. ) "
C, 6
-A "
1. )
#) 6 "
D1
23"
()
! "#$%&'( )*&'+, '! - . &/0'
12 3'4 /&"/5'! "#$%&'( )*&/'67'8#9': ; '<"#//=( )>&'8"- ?@&"/'
#! ! "
' &! "
' %! "
' $! "
' #! "
' !!"
&! "
%! "
$! "
#! "
!"
Doubleclick:
Avg 39% (Max 66%)
Facebook:
Avg 23% (Max 45%)
April 26, 2012
Franziska Roesner
Each line
represents
one user.
Google:
Avg 21% (Max 61%)
20
LocalStorage and Flash Cookies
• Surprisingly little use of these mechanisms!
• Of 524 trackers on Alexa Top 500:
– Only 5 set unique identifiers in LocalStorage
– 35 set unique identifiers in Flash cookies
• Respawning:
– LS  Cookie: 1 case; Cookie  LS: 3 cases
– Flash Cookie: 6 cases; Cookie  Flash: 7 cases
April 26, 2012
Franziska Roesner
21
Outline
• How Tracking Works
– Tracking mechanisms
– Tracking taxonomy
• Measurements
• Defenses
April 26, 2012
Franziska Roesner
22
Defenses to Reduce Tracking
• We explore several in the paper:
– Third-party cookie blocking
– Do Not Track header
– Popup blocking
– Clearing client-side state
– Disabling JavaScript
– Private browsing mode
April 26, 2012
Franziska Roesner
23
&%$
"#$%&"'(
&"'( "&)
"&) #*
#*&+$&',&+$&',- '.'. // 00 #1
#1+23'
+23'
!! "#$%
%%$
//4'56'!
"#$%
&
"2'/+'!
+'!&
/"#$%
4'766'.
#1
+2',!
81
/#1
1
&1&
2'>*
/ $%
&93'
! /!!4'56'!
"#$%
"2'/
/"#$%
4'766'.
// 0/+'!
+2',!
81"9:(
"9:(/#";<'=/
#";<'=/
/%%
2'>*
/ $%
&93'
4'56'!
"#$%
"2'/
+'!
4'766'.
0//#1
4'56'!
&
"2'/
+'!
4'766'.
#1
+2'
!! //&4'56'!
&
"2'/
//4'766'.
00 +2'
+2'
(%
%%
%$$
&**#
(&++#
(no cookie blocking)
&%
&%
%
+#
$$
()()%
*#
@
-- JI660
<$
>/
L/
621.M
$ #$
@
660
I$<K#>/
#$K@
L/
6-21.O
#
J@
-LJ660
<->/IK>
K</L/
@
621.M
$21.M
@
660
M
N/
@
6-@
!*#+$
' *#
*#+$
)**#
%%
%$$ !*#+$
)
(
%
++#
(
' +#
%%$
!*&%
%*#
+#
$$
!*&%
%
&%$
%%$
&%$
,-
-,
%$
#$
!! ""#$
! " #$ $%! #!$%
!&(
&(!#$#!$&'
$!(( #$
#$
#
$&'
! &%
&%
$
!$%
$%
+#
$
*#! &( $
Referred Tracker
Forced Tracker
! ( #$
%#$
#$
!!%
%&$
&$"#)&#$
!!%
"
&#
#) $ ""!!$$
!
%
%
$
!$++#
%%$
$**#
! %#$! %&$
%$$
'' %
#) $ " ! $
&$(((($$((%
((&$
%$$)()(&#
&%
&%
%
+#
$$
($$))))$$))*$)
*#$
%
*#
*$)*$
*$!)))%%#$$!*#$
" # *+$*'*'$$
( &# !$&#
" # ! ) #' !%*#
$ $&# $&# $! # $! #$$# $$# $*#$*#$*# " # ' *+$
# '#
( &$ ( ( $( %$ ) ( $ ) ) $
%$$
%
+#
) *$) *$ ) %$*#$ *+$ *' $
*#
%$$
!*%
++#
%
!*%
**#
- @ >@AE/ 5--88 #
8
A
<2 1G9@/ <67755-- 8 #$$
C<>< 266 8 #
/ 18- F6775- #$
6/ /: 1 5- 8
@C 9<-17 8 #
<2 5 755- #$
, 02H - 8
6 H 8#
3
1A 1 - : 367755- #$
C. 91.> >: DD - 88 #
I @ 9>8 ; /75 #$
>/ A<618 75- 8
2A 5 1 - 8 #
; F. 2 92775- #$
8 : /A=/ @5- 88 #
/ 9 H 1 75 #$
<1G/ A <75-- 8
./- 75 8 #
F75- #$
- 88 #
#$
%%$
J@
- J660
<>/I$<K>/
N2238 @
-M
$
@
-I660
#$KN>A/
#$: 6M
@
- I660
#M
P 238
<B/ @
O
J@
-L660
<>/K>
$K</N2-#: 6M
$
./
65
01
21
-@
/5
99-1 .34
1@
:5;/ 567
9@
75
/ 6 ?1 .:/-5-.< 5-- 88
B: 65 / 1 5,/-; ;57=57 ##$
12- @/ @5A --, -. -28/
>6 51 75<,=- /=7755 >##$
/ @@9 - 8>-? -- 8
C/ @/16A $,/./@7 8 ##$
B1:A75- /A1< 755-- 8
B5> 8
1A2<B8 @
3</
>6>6 $ A67755- ##$
.9
AC2 - 88 #
3- /5@
8
>CD< /7755- #$
<8
1
<CE: - 88
@D 213
- @ <,/</.@> 1: >1 //@7755- ##$
: 9> .9 < 9@
D8 D 87511-.9C A8C>75 88 ##$
<9@DC 2816> 75-- 8
/ -71- 2 ,$-/2 8 #
5@
19
7
@-.<996/ / @
8DD@<E @77555---88 #$
C/
@4
1A $<99///775 8 ##$$
6<2119 @-/ : 755---8
,97CC.>//2@; C6/575 88 ##$$
5
F 8-@844/66A<2>2 <7-28/ #
<
>
= 1969/ $2,,77755- #$
&%$
>A<
<>/
H I<20
>A<
20
H <>A<
<>/I 20I
$</I$<#>/ $#
JH <>20
<B>
20K>
April 26, 2012
Franziska Roesner
24
Personal Tracking Revisited
• Most popular, based on measurements:
– Facebook, Google, Twitter, AddThis, YouTube,
LinkedIn, Digg, Stumbleupon
• Third-party cookie blocking is ineffective.
• Existing browser extension solutions remove the
buttons (undesirable to some users).
• Can we reduce tracking but allow use?
April 26, 2012
Franziska Roesner
25
ShareMeNot
http://sharemenot.cs.washington.edu
• A browser extension that protects against
tracking from third-party social media buttons
while still allowing them to be used.
• Firefox version: removes cookies from relevant
requests until user clicks button.
– Similar: Priv3 Firefox add-on
• Chrome version: replace buttons with local
stand-in button until user click.
April 26, 2012
Franziska Roesner
NEW
26
Effectiveness of ShareMeNot (Top 500)
Tracker
Facebook
Without ShareMeNot With ShareMeNot
154
9
Google
Twitter
AddThis
149
93
34
15
0
0
YouTube
LinkedIn
Digg
Stumbleupon
30
22
8
6
0
0
0
0
April 26, 2012
Franziska Roesner
27
Summary
• Introduced taxonomy of tracking behavior for any
client-side identifiers.
– Analytics, Vanilla, Forced, Referred, Personal
• Studied tracking in the wild with browser
measurements.
– Revealed rich tracking ecosystem.
– Results can assist informed broader discussions.
• Developed ShareMeNot, a new privacyenhancing defense for personal tracking.
http://sharemenot.cs.washington.edu/
April 26, 2012
Franziska Roesner
28