Session Fixation Vulnerability
Session Fixation Vulnerability
Session Fixation Vulnerability
December 2002
(Revised February 2007 – the Acknowledgments section)
Current copy available at http://www.acrossecurity.com/papers/session_fixation.pdf
1. Abstract
Many web-based applications employ some kind of session management to create a
user-friendly environment. Sessions are stored on server and associated with
respective users by session identifiers (IDs). Naturally, session IDs present an
attractive target for attackers, who, by obtaining them, effectively hijack users’
identities. Knowing that, web servers are employing techniques for protecting session
IDs from three classes of attacks: interception, prediction and brute-force attacks.
This paper reveals a fourth class of attacks against session IDs: session fixation
attacks. In a session fixation attack, the attacker fixes the user’s session ID before the
user even logs into the target server, thereby eliminating the need to obtain the user’s
session ID afterwards. There are many ways for the attacker to perform a session
fixation attack, depending on the session ID transport mechanism (URL arguments,
hidden form fields, cookies) and the vulnerabilities available in the target system or its
immediate environment. The paper provides detailed information about exploiting
vulnerable systems as well as recommendations for protecting them against session
fixation attacks.
2. Introduction
Web-based applications frequently use sessions to provide a friendly environment to
their users. HTTP [1] is a stateless protocol, which means that it provides no
integrated way for a web server to maintain states throughout user’s subsequent
requests. In order to overcome this problem, web servers – or sometimes web
applications – implement various kinds of session management. The basic idea behind
web session management is that the server generates a session identifier (ID) at
some early point in user interaction, sends this ID to the user’s browser and makes
sure that this same ID will be sent back by the browser along with each subsequent
request. Session IDs thereby become identification tokens for users, and servers can
use them to maintain session data (e.g., variables) and create a session-like
experience to the users.
There are three widely used methods for maintaining sessions in web environment:
URL arguments, hidden form fields and cookies [2]. While each of them has its
benefits and shortcomings, cookies have proven to be the most convenient and also
the least insecure of the three. From security perspective, most – if not all - known
attacks against cookie-based session maintenance schemes can also be used against
URL- or hidden form fields-based schemes, while the converse is not true. This makes
cookies the best choice security-wise.
Very often, session IDs are not only identification tokens, but also authenticators. This
means that upon login, users are authenticated based on their credentials (e.g.,
usernames/passwords or digital certificates) and issued session IDs that will
effectively serve as temporary static passwords for accessing their sessions.
This makes session IDs a very appealing target for attackers. In many cases, an
attacker who manages to obtain a valid ID of user’s session can use it to directly enter
that session – often without arising user’s suspicion. Interestingly, most cross-site
scripting [3] proof-of-concept exploits focus on obtaining the session ID stored in
browser’s cookie storage. This class of attacks, where the attacker gains access to the
user’s session by obtaining his session ID, is called session hijacking [4].
Web session security is focused on preventing three types of attacks against session
IDs: interception, prediction and brute-force attacks. Encrypted communication
effectively protects against interception1. Using cryptographically strong pseudo-
random number generators and carefully chosen seeds that don’t leak from the server
prevents prediction of session IDs. Finally, session IDs are immune to brute-force
methods if their effective bit-length is large enough with respect to the number of
simultaneous sessions2.
Proposals have been made for mitigating the threat of stolen session IDs [6], and
some products already implement such ideas (e.g., RSA Security’s ACE/Agents for
web servers).
3. Session fixation
As mentioned above, web session security is mainly focused on preventing the
attacker from obtaining – either intercepting, predicting or brute-forcing - a session ID
issued by the web server (also called “target server” in this paper) to the user’s
browser.
This approach, however, ignores one possibility: namely the possibility of the attacker
“issuing” a session ID to the user’s browser, thereby forcing the browser into using a
chosen session. We’ll call this class of attacks “session fixation” attacks, because the
user’s session ID has been fixed in advance instead of having been generated
randomly at login time.
1
Although forgetting to mark session ID cookies as »secure« keeps the attacker's foot in the
door [7].
2
David Endler of iDefense wrote a very interesting article [5] on this topic.
In a session fixation attack, the attacker fixes the user’s session ID before
the user even logs into the target server, thereby eliminating the need to
obtain the user’s session ID afterwards.
Let’s take a look at a simple example of a session fixation attack. Figure 1 shows a
web server online.worldbank.dom that hosts a session-aware web banking
application. Session IDs are transported from browser to server within a URL
argument sessionid.
First, the attacker – who in this case is also a legitimate user of the system – logs in
to the server (1) and is issued a session ID 1234 (2). She then sends a hyperlink
http://online.worldbank.dom/login.jsp?sessionid=1234 to the user, trying
to lure him into clicking on it (3). The user (how convenient for our example) clicks on
the link, which opens the server’s login page in his browser (4). Note that upon
receipt of the request for login.jsp?sessionid=1234, the web application has
established that a session already exists for this user and a new one need not be
created. Finally, the user provides his credentials to the login script (5) and the server
grants him access to his bank account. However, at this point, knowing the session ID,
the attacker can also access the user’s account via account.jsp?sessionid=1234
(6).
Since the session has already been fixed before the user logged in, we say that the
user logged into the attacker’s session.
attacker 6
Log
3 in
GE s e ssi
T /a oni
cco d=1
unt 234
.jsp
?s e
ssi
http://online.worldbank.dom oni
d=1
/login.jsp?sessionid=1234 234
GET /login.jsp?sessionid=1234
4
username & password
5
user online.worldbank.dom
The above example is the simplest – and the least dangerous - form of a session
fixation attack and has many shortcomings (for the attacker), such as: she has to be a
legitimate user on the target server and she has to trick the user into logging in
through the hyperlink she provided.
The following chapters will describe various methods for making session fixation more
reliable, less detectable and available to “outside” attackers (those that are not
legitimate users on the target server). But first, let’s examine the attack process step
by step.
4. Attack process
Generally, session fixation attack is a three-step process, as shown in Figure 2:
1. Session setup: First, the attacker either sets up a so-called “trap session” on
the target server and obtains that session’s ID, or selects a – usually arbitrary
– session ID to be used in the attack. In some cases, the established trap
session needs to be maintained (kept alive) by repeatedly sending requests
referencing it to avoid idle session timeout.
2. Session fixation: Next, the attacker needs to introduce her session ID to the
user’s browser, thereby fixing his session.
3. Session entrance: Finally, the attacker has to wait until the user logs in to
the target server using the previously fixed session ID and then enter the
user’s session.
1 Session setup
1a Session maintenance
2 Session fixation
3 Session entrance
For a permissive system, the attacker only needs to make up a random trap session
ID, remember it and use it for the attack.
In a strict system, the attacker will have to actually establish a trap session with the
target server, extract the trap session ID, remember it and use it for the attack.
Depending on the session management logic, this session will need to be kept alive at
least until the user logs into it. Generally, sessions don’t exist indefinitely; rather,
servers automatically destroy them upon idle or absolute timeout. Idle timeout can
easily be avoided by automatically sending periodic requests for the trap session.
Absolute timeout, while less frequently employed, presents a tougher obstacle and in
most cases defines the maximum timeframe for the entire attack process.
Furthermore, restarting the web server can destroy all active sessions, requiring the
attacker to return to the session setup step.
A permissive system requires no trap session maintenance.
In this step the attacker tries to transport the trap session ID to the user’s browser.
Methods for accomplishing that differ depending on the session ID transport
mechanism. Let’s examine each one of them.
Session ID in a cookie
Cookies are a predominant session ID transport mechanism, partly also due to their
security in comparison to URL arguments and hidden form fields. Ironically, on the
other hand, cookies provide the most convenient, covert, effective and durable means
of exploiting session fixation vulnerabilities.
What the attacker needs to do is install the trap session ID cookie on the user’s
browser. According to RFC2965 [2], the browser will only accept a cookie assigned
either to the issuing server or the issuing server’s domain. Consequently, even though
it would provide a great attack avenue, the attacker’s web server www.attacker.com
can’t set a cookie for the target web server online.worldbank.dom.
The attacker can choose among the three available methods for issuing a cookie to the
browser:
A. using a client-side script that sets a cookie on the browser;
B. using the HTML <META> tag with Set-Cookie attribute;
C. using the Set-Cookie HTTP response header.
We’ll analyze each of them in terms of their usability in a session fixation attack.
Cross-site scripting
The attacker can exploit a cross-site scripting vulnerability on the
online.worldbank.dom server in order to have that server issue the desired session
ID cookie itself. An example URL for exploiting the cross-site scripting vulnerability in
Microsoft’s Internet Information Server [8] is:
http://online.worldbank.dom/<script>document.cookie="sessioni
d=1234”;</script>.idc
The attacker could introduce this malicious URL to the user’s browser by sending the
user an HTML-formatted e-mail message including a small (invisible) frame originating
from this URL. Alternatively, the attacker could send the user only a hyperlink to this
URL (which can be disguised as something benign3) and try to lure the user into
clicking on it.
Persistent cookies
The above cross-site scripting exploit can be enhanced by setting a persistent cookie
instead of a temporary one (which is a default), thereby fixing the user’s session for a
longer period of time. Example:
3
Microsoft Outlook, for example, doesn't make it easy for user to see where a hyperlink in a
message is actually pointing to.
http://online.worldbank.dom/<script>document.cookie="sessioni
d=1234;%20Expires=Friday,%201-Jan-2010%2000:00:00%20GMT”;</sc
ript>.idc
Opening this URL in the user’s browser will fix his session ID to 1234 until the year
2010. Unless the user clears his persistent cookies or reinstalls his computer, chances
are that the attacker will gain a long-term access to the user’s bank account. Clearly,
this attack will be more effective on permissive systems, which don’t require the
attacker to maintain the trap session.
Domain cookies
Furthermore, domain cookies can expand the attack area from the target server to the
entire target server’s domain. Domain cookies are cookies with their domain attribute
set to the issuing server’s domain (e.g., ".worldbank.dom"). This attribute instructs
the browser to not only send the cookie back to the issuing server but also to any
other server in the specified domain. An example of exploiting a cross-site scripting
vulnerability for issuing a domain cookie is this:
http://online.worldbank.dom/<script>document.cookie="sessioni
d=1234;domain=.worldbank.dom”;</script>.idc
In a session fixation attack, the added value of domain cookies becomes clear in the
following scenario (see Figure 3):
Attacker exploits a cross-site scripting vulnerability on server www.worldbank.dom
(which is in the same domain as the target sever online.worldbank.dom) for
generating a cookie-issuing script that sets a domain trap session ID cookie for the
.worldbank.dom domain. She provides such cookie-issuing URL to the user’s
browser (1). Upon visiting the provided URL (2), the user’s browser will accept a
cookie from www.worldbank.dom (3) and will – since it’s a domain cookie – send it
to online.worldbank.dom when the user later decides to log in to his bank account
(4).
GET /account.jsp
5
Cookie: sessionid=1234
attacker
1
online.worldbank.dom
http://www.worldbank.dom/ p
n.js 234
<script>document.cookie= / l ogi nid=1
T
...</script>.idc GE sessio
ie :
ok
Co
4
GET /<script>document.cookie=...</script>.idc
2
document.cookie="sessionid=1234; domain=.worldbank.dom"
3
user
www.worldbank.dom
Figure 3: Session fixation using a cross-site scripting vulnerability on another server in domain
After the user’s successful login, the attacker will be able to access his bank account
using the fixed session ID (5).
A very effective attack scenario could employ a persistent domain trap session ID
cookie.
Method B: Issuing a cookie using the <META> tag with Set-Cookie attribute
The server can also issue a cookie to the browser by including an appropriate <META>
tag in the returned HTML document, for example:
<meta http-equiv=Set-Cookie content="sessionid=1234">
Now, the attacker wants the target server - or any other web server in the target
server’s domain - to return a specific, trap session ID cookie-issuing <META> tag to
the user’s browser. This can be achieved by using a “meta tag injection” method.
http://online.worldbank.dom/<meta%20http-equiv=Set-Cookie%20c
ontent="sessionid=1234;%20Expires=Friday,%201-Jan-2010%2000:0
0:00%20GMT”>.idc
In contrast to client-side scripting4, the processing of <META> tags can’t be disabled
on today’s browsers5, which makes the meta tag injection method superior to the
client-side scripting method in a session fixation attack.
Session adoption
Some servers (e.g., JRun) accept any session ID in a URL and issue it back as a cookie
to the browser. For example, requesting:
http://online.worldbank.dom/?jsessionid=1234
sets the session cookie JSESSIONID to 1234 and creates a new session with that ID
on the server6. We’ll call such behavior “session adoption”, due to the fact that the
server effectively “adopts” a session ID that was generated by someone else.
4
For example, client-side scripting is disabled in the Internet Explorer's Restricted sites zone.
5
With the exception of the META REFRESH tag in Internet Explorer, which can be disabled.
6
In JRun, it is required that a session with the proposed ID doesn’t exist on the server yet in
order for the server to return the session ID cookie. This may be different in other systems with
similar behavior.
attacker
4 DNS
2 .d om
nk user's DNS server
ld ba
or
k.w 6.6
6
h ac 6.6
http://hack.worldbank.dom s t : . 6
ue 6
eq :6
/setcookie.cgi
N Sr o nse
D es p
Sr
DN
GET /setcookie.cgi
5
user Set-Cookie: sessionid=1234; domain=.worldbank.dom
6
First, she sets up a web server with a simple server-side script that issues a domain
trap session ID cookie for the domain .worldbank.dom. She then adds a record to
the user’s DNS server (1)7, mapping the hostname hack.worldbank.dom to her web
server’s IP address. To fix the user’s session, she needs to make the user’s browser
request the cookie-issuing script on the hack.worldbank.dom server (2). The user’s
browser will ask the DNS server about the IP address of the hack.worldbank.dom
server (3) and the DNS server will reply with the IP address of the attacker’s web
server (4). Thereafter, the browser will send a request for the cookie-issuing script to
the attacker’s web server (5) and will be issued a trap session ID cookie for the
.worldbank.dom domain (6). Since the cookie will have been issued by a host in the
.worldbank.dom domain, the browser will be forced to accept it, whereby the user’s
session would be fixed to the attacker’s session ID.
Furthermore, the ability to add a host record to the user’s DNS server also provides
the attacker with an opportunity to introduce the malicious cookie-issuing URL to the
user’s browser in a more covert fashion (than via e-mail). By modifying one or more
other records on the user’s DNS server (e.g., for www.yahoo.com), she can redirect
the user’s browser to her web server when the user decides to visit the Yahoo pages.
In this case, the fake www.yahoo.com server would first redirect the browser to the
7
For example, using some DNS cache poisoning vulnerability.
Network-based attack
Finally, the attacker with ability to sniff and modify network traffic coming to and from
the user’s browser can also avoid sending the user a potentially suspicious e-mail and
can perform the session fixation in an ultimately covert fashion. Namely, the attacker
can inject a small (invisible) image in any web server’s response to the browser – for
example when the user is reading Yahoo news. This image would originate from any
web server in the .worldbank.dom domain. Upon requesting the image content, the
browser would connect to this web server and the attacker, intercepting the request,
could send a fake response by the web server, including a Set-Cookie header,
thereby fixing the user’s session.
Notes: It’s important to realize that using an encrypted communication between the
user’s browser and the target web server has literally no effect on the exploitability of
session fixation vulnerabilities. It’s also important to note that with cookie-based
session mechanisms, the session fixation problem isn’t solved automatically by fixing
all cross-site scripting vulnerabilities on the target web server (which can solve the
session hijacking problem, by the way). In fact, any vulnerable server in the target
server’s domain or a vulnerable DNS server can provide a session fixation opportunity
for the attacker. For example, a seemingly irrelevant cross-site scripting vulnerability
on one server can jeopardize a sensitive web server in the same domain. Finally, the
HttpOnly cookie attribute [9], recently introduced by Microsoft mainly for the
purpose of making session hijacking attacks more difficult, has little or no effect on
session fixation attacks.
After the user has logged in to the trap session and before he has logged out, the
attacker can enter the trap session and assume the user’s identity. In many systems,
the attacker will be able to use the session without the user noticing anything suspect.
In case the user doesn’t log out of the system, the attacker has an opportunity to
keep the session alive – and thereby the access to the user’s identity – for a long
time.
Summarizing this section, Figure 5 graphically presents the session fixation process.
5. Countermeasures
First of all, we need to make it clear that preventing session fixation attacks is mainly
the responsibility of the web application, and not the underlying web server. The web
server – which usually provides the session management API to applications – should
make sure that session IDs can’t be intercepted, predicted or brute-forced. But as far
as session fixation is concerned, only the web application can implement effective
protection.
There is one common denominator to all session fixation attacks and scenarios: the
user logs in to a session with an attacker-chosen ID, instead of having been issued a
newly generated session ID by the server. Since there seems to be no compelling
reason for web applications to explicitly allow this to happen - and seems more like a
side effect of current implementations –, we propose forceful prevention of logging
into a chosen session. Web applications must ignore any session ID provided by the
user’s browser at login and must always generate a new session to which the user will
log in if successfully authenticated.
If possible, a web application on a strict system should only issue session IDs of newly
generated sessions to users after they have successfully authenticated (as opposed to
issuing them along with the login form). This means that an attacker who isn’t a
legitimate user of the system will not be able to get a valid session ID and will
therefore be unable to perform a session fixation attack.
Most methods for mitigating the threat of stolen session IDs are also applicable to
session fixation. Some of them are listed below.
• Binding the session ID to the browser’s network address (as seen by the
server)
1 Session setup
yes
Idle session
1a Session maintenance timeout
2 Session fixation
Lure the user into logging Lure the user into logging
in through a malicious URL cookies in through a malicious login form
Exploit cross-site scripting Exploit session adoption Modify the response from any
for issuing a cookie for issuing a cookie server in domain to issue a cookie
3 Session entrance
• Binding the session ID to the user’s SSL client certificate - very important and
often overlooked issue in highly critical applications: each server-side script
must first check whether the proposed session was actually established using
the supplied certificate.
• Session destruction, either due to logging out or timeout, must take place on
the server (deleting session), not just on the browser (deleting the session
cookie).
• The user must have an option to log out – thereby destroying not just his
current session, but also any previous sessions that may still exist (in order to
prevent the attacker from using an old session the user forgot to log out from).
• Absolute session timeouts prevent attackers from both maintaining a trap
session as well as maintaining an already entered user’s session for a long
period of time.
6. Conclusion
The session fixation vulnerability seems to be present in many session-enabled web-
based applications. In fact, it has been present in almost all web-based systems
(including many high profile web banking systems) that we’ve ever come across.
Understandably, we can’t list any vulnerable real-world systems here but there are
many out there - also among the largest ones. It would be an impossible job to try to
make an exhaustive list of all publicly available vulnerable applications and inform
their vendors about this vulnerability. Therefore we’ll have to leave this job to these
vendors and the worldwide security community in hope that as few as possible
malicious exploits will see the light of day. And for all the guys out there doing the
security reviews: this is just another checkbox to add to your checklist.
Timing
Session fixation Attacker attacks the user’s browser before he logs in to the
target server.
Session hijacking Attacker attacks the user’s browser after he logs in to the target
server.
Impact Duration
Session fixation Attacker gains one-time, temporary or long-term access to the
user’s session(s).
Session hijacking Attacker usually gains one-time access to the user’s session and
has to repeat the attack in order to gain access to another one.
Session Maintenance
Session fixation Can require the attacker to maintain the trap session until the
user logs into it.
Session hijacking Requires no session maintenance.
Attack Vectors
Session fixation 1. Tricking the user to log in through a malicious hyperlink or a
malicious login form
2. Exploiting a cross-site scripting vulnerability on any web
server in the target server’s domain
3. Exploiting a meta tag injection vulnerability on any web
server in the target server’s domain
4. Exploiting the “session adoption” feature of some web servers
5. Breaking into any host in the target server’s domain
6. Adding a domain cookie-issuing server to the target server’s
domain in the user’s DNS server
7. Network traffic modification
Session hijacking 1. Exploiting a cross-site scripting vulnerability on the target
server
2. Obtaining the session ID in the HTTP Referer header sent to
another web server
3. Network traffic sniffing (only works with an unencrypted link
to the target server)
8. Acknowledgments
Many thanks to Kevin Fu of MIT, Saša Kos, Aljoša Ocepek and Stanka Šalamun for
their invaluable contribution to this paper.
In February 2007 I was made aware of a Vuln-Dev posting [11] by Jeff Jancula of
Wachovia (then called First Union), describing the concept of session fixation and
dating back in August 2001 – i.e., predating the first publication of this paper.
Obviously, massive credits go to Jeff and his colleagues for publishing what seems, at
the time of this writing, the first public record of the idea described herein. I wish I
had found their posting when first publishing this paper so that I could acknowledge
them from the beginning, but I hope this revision sets things straight. Thanks to Amit
Klein as well for notifying me about Jeff’s posting.
9. References
[1] IETF, »RFC2616: Hypertext Transfer Protocol -- HTTP/1.1«
http://www.ietf.org/rfc/rfc2616.txt
[2] IETF, »RFC2109: HTTP State Management Mechanism«
http://www.ietf.org/rfc/rfc2109.txt
[3] The Open Web Application Security Project, »Cross Site Scripting«
http://www.owasp.org/asac/input_validation/css.shtml
[4] The Open Web Application Security Project, »Session Hijacking«
http://www.owasp.org/asac/auth-session/hijack.shtml
[5] David Endler, »Brute-Force Exploitation of Web Application Session IDs«
http://online.securityfocus.com/data/library/SessionIDs.pdf
[6] Kevin Fu, Emil Sit, Kendra Smith, Nick Feamster, »Dos and Don’ts of Client
Authentication on the Web«
http://pdos.lcs.mit.edu/cookies/pubs/webauth:tr.pdf
[7] ACROS, »Remote Retrieval Of IIS Session Cookies From Web Browsers«
http://www.acrossecurity.com/aspr/ASPR-2000-07-22-1-PUB.txt
[8] SecurityFocus, »Microsoft IIS IDC Extension Cross Site Scripting Vulnerability«
http://online.securityfocus.com/bid/5900/info/
[9] Michael Howard, Microsoft, »Some Bad News and Some Good News«
http://msdn.microsoft.com/library/en-us/dncode/html/secure10102002.asp
[10] CERT, »CERT© Advisory CA-2000-02: Malicious HTML Tags Embedded in Client
Web Requests«
http://www.cert.org/advisories/CA-2000-02.html
[11] Jeff Jancula, »Security Problems With Web Servers' Session Tracking
Mechanisms«
http://archives.neohapsis.com/archives/vuln-dev/2001-q3/0548.html