I am writing this blog in response to a need I felt when I was new to authentication in Windows. The blog does not intend to reinvent the wheel, instead to help someone new to Windows authentication to get started and find the resources one needs to understand authentication. I’ll briefly describe concepts; especially the ones that I felt were not described in plain English and then direct the reader to the appropriate resource. I am only going to describe network authentication and will not make any attempt to describe local authentication or Windows logon process.
In a nutshell, to prove to a service you really are who you claim to be. A service is typically running on a separate computer where you want to access the resources offered by the service, e.g. a file server is a service and the files are resources. If the owner of the file(s) restricts the access to a select few, the owner will put in place restrictions as to who can access the file(s) and what they are allowed to do with the files. The “who” question is answered by authentication and the “what” question is answered by authorization.
To make this blog more hands on and less boring, let’s look how authentication is performed in SMB/SMB2. In SMB/SMB2, authentication takes place in SESSION SETUP message. If you look at session setup message in a network trace, you will find something called security blob. I’ll use netmon trace for this blog (wireshark trace will also show this info albeit in a different format).
Frame: Number = 137, Captured Frame Length = 1514, MediaType = ETHERNET
+ Ethernet: Etype = Internet IP (IPv4),DestinationAddress:[00-15-5D-49-34-12],SourceAddress:[00-15-5D-49-34-11]
+ Ipv4: Src = 192.168.0.201, Dest = 192.168.0.1, Next Protocol = TCP, Packet ID = 270, Total IP Length = 1500
+ Tcp: Flags=...A...., SrcPort=49163, DstPort=Microsoft-DS(445), PayloadLen=1460, Seq=368368040 - 368369500, Ack=2088786732, Win=512 (scale factor 0x8) = 131072
+ SMBOverTCP: Length = 3148
- SMB2: C SESSION SETUP (0x1)
+ SMB2Header: C SESSION SETUP (0x1),TID=0x0000, MID=0x0001, PID=0xFEFF, SID=0x0000
StructureSize: 25 (0x19)
VcNumber: 0 (0x0)
+ SecurityMode: 2 (0x2)
+ Capabilities: 0x1
Channel: 0 (0x0)
SecurityBufferOffset: 88 (0x58)
SecurityBufferLength: 3060 (0xBF4)
PreviousSessionId: 0 (0x0)
By expanding securityBlob, one can see the details of authentication protocol. As we will see shortly.
Authentication generally does not cause network traffic on its own (Kerberos is a notable exception and I’ll describe that in a little bit). Authentication is also optional and only depends on service, e.g. the most used service in the world, HTTP, does not use authentication normally. You don’t need to provide a username/password for www.yahoo.com, for example.
As shown above, securityBlob is an adjunct to a protocol message. When looking at a network trace of a protocol, authentication can be found in a message somewhere at the beginning part of the message exchange. A little knowledge of the protocol is very helpful as every protocol sends security blob in a message unique to the protocol. For example, in LDAP, the authentication takes place in bind message.
Let’s expand securityBlob and see what’s inside:
So, what is this GSSAPI? We explore that next.
GSSAPI stands for Generic Security Services Application Programming Interface. It is specified in RFC2078, which documents version 2 of this API. Please consult the RFC for older or newer versions of this API at http://tools.ietf.org/html/rfc2078. There are two updates to version 2 at the writing of this blog.
Microsoft’s implementation of GSSAPI is called Security Support Provider Interface (SSPI). The details of SSPI are available at http://msdn.microsoft.com/en-us/library/aa380493(VS.85).aspx
GSSAPI is an attempt to make the job of application designer easier. This goal is achieved by providing function calls that are independent of the authentication protocol.
An application programmer calls the method GSS_Init_sec_context ( InitializeSecurityContext aka ISC in SSPI) with the choice of the protocol he/she wants to use e.g. Kerberos. The call is successful if the return code is either GSS_S_COMPLETE or GSS_S_CONTINUE_NEEDED. If GSS_S_CONTINUE_NEEDED is returned, the method also returns a blob that is sent as securityBlob in SESSION SETUP message. When server receives the session setup messages, it calls the GSSAPI method GSS_Accept_sec_context ( AcceptSecurityContext aka ASC in SSPI). The return code signifies if the authentication is complete. If the return code is not zero, server sends the securityBlob that ASC returned to client. The Client calls ISC again with the blob server send and this goes on till the method calls returns GSS_S_COMPLETE which signals the successful completion of authentication.
As you may have noticed, SSPI method calls have different names than GSSAPI names e.g. GSS_Init_sec_context is called InitializeSecurityContext in SSPI. There is a very helpful list available at SSPI/Kerberos Interoperability with GSSAPI that describe a scenario when an SSPI client interoperates with a GSSAPI server. This does not provide a complete one to one mapping but enough info that a designer can do the mapping on his/her own.
To understand SSPI a little better, compile and run sample code called sample SSPI code at http://msdn.microsoft.com/en-us/library/aa379449(v=VS.85).aspx.
The code enables you to select Negotiate (more about this later), NTLM or Kerberos authentication protocol. By running this application and observing network traces, much can be learned about SSPI. Also study the code to see how a typical application uses SSPI. Highly recommended.
If we expand GSSAPI in the Netmon trace, we can see the authentication protocol being used by GSSAPI. In this case it is SPNEGO.
+ ThisMech: SpnegoToken (22.214.171.124.5.5.2)
+ InnerContextToken: 0x1
What is SPNEGO and how did GSSAPI select this authentication protocol?
Well, in the GSSAPI call, GSS_Init_sec_context , there is an input parameter called mech_type. This specifies the authentication mechanism (protocol) that GSSAPI will use. While GSSAPI shields the software designer from the details of authentication protocol, it still requires the designer to select the authentication protocol e.g. Kerberos, NTLM etc. This has two disadvantages; the designer needs to know what authentication protocols are available and once the software is written, the protocol is fixed. If a new protocol becomes available, in the worst case the software needs to be modified and recompiled. In a more reasonable case, some configuration needed to be change to select the new protocol, if the designer foresaw the possibility of new protocol.
To make the selection of authentication protocol dynamic, all one need to do it to use just one protocol, SPNEGO. This seemingly self-contradictory sentence is actually correct. SPNEGO is an authentication protocol that is not an authentication protocol in itself. SPNEGO stands for “the simple and protected generic security service application program interface negotiation” and is specified in RFC 4178 .
To use SPNEGO, mech_type is set to “Negotiate” in SSPI for InitializeSecurityContext call. In the resulting token that client sends to the server, SPNEGO presents a list of authentication mechanisms to the server. The list of mechanisms is in the order of most desirable to least desirable.
Let’s expand the SPNEGO token in the Netmon trace we are studying in this blog:
- InnerContextToken: 0x1
- SpnegoToken: 0x1
- MechTypes: Prefer MsKerberosToken (1.2.840.48018.1.2.2)
+ MechType: MsKerberosToken (1.2.840.48018.1.2.2)
+ MechType: KerberosToken (1.2.840.1135126.96.36.199)
+ MechType: Negoex (188.8.131.52.4.1.3184.108.40.206)
+ MechType: NLMP (220.127.116.11.4.1.318.104.22.168)
+ MechToken: 0x1
As one can see, SPNEGO presents a list of mechanisms (authentication protocols) that client supports and lists them in the order of preference with MsKerberos being most desirable and NLMP (NTLM) being the least desirable in this example. To reduce the number of round trips, SPNEGO also includes the initial token for the most desirable mechanism in MechToken field. This is called optimistic token and is sent in the hope that server will also select the same mechanism as client.
If server selects the same protocol as client, the security blob in SESSION SETUP response will contain a response token for the same protocol and authentication will continue or complete as per selected protocol.
If the server does not support the most preferred mechanism, it will send its choice in the SESSION SETUP response and the client will send the first token of the server selected mechanism and messages will go back and forth till authentication is complete.
The important point here is that as new mechanisms become available, list of mechanism will change dynamically and there is no need to change software or provisioning.
If you look at the scheme that SPNEGO uses to make mechanism selection dynamic, you will notice one security concern. What if a malicious hacker proposes a list of very week authentication protocols with the intent to authenticate with previously used token that he/she intercepted?
To make sure that the list of mechanisms is coming from a user with the knowledge of the credentials, MechListMIC token is used. For details, please consult the blog mechListMIC token.
Let’s expand MechToken and see what’s in there:
- MechToken: 0x1
- MsKerberosToken: 0x1
+ ThisMech: KerberosToken (1.2.840.113522.214.171.124)
- InnerContextToken: 0x1
- KerberosToken: 0x1
TokId: Krb5ApReq (0x100)
- ApReq: KRB_AP_REQ (14)
+ PvNo: 5
+ MsgType: KRB_AP_REQ (14)
+ Tag2: 0x1
+ Ticket: Realm: NWTRADERS.MSFT, Sname: cifs/hyper-v-dc1.nwtraders.msft
This is Kerberos AP_REQ token. If the most preferred protocol were NLMP, ThisMech would have NTLM Negotiate message in it. For GSSAPI, SPNEGO is the authentication protocol but inside the SPNEGO token, the messages of actual authentication protocol are sent.
Let’s see how server responds to the above SESSION SETUP request:
+ NegState: accept-completed (0)
+ SupportedMech: MsKerberosToken (1.2.840.48018.1.2.2)
- ResponseToken: 0x1
- MsKerberosToken: 0x1
+ ThisMech: KerberosToken (1.2.840.1135126.96.36.199)
- InnerContextToken: 0x1
- KerberosToken: 0x1
TokId: Krb5ApRep (0x200)
- ApRep: KRB_AP_REP (15)
+ PvNo: 5
+ MsgType: KRB_AP_REP (15)
+ Tag2: 0x1
This is Kerberos AP_REP token which is wrapped in an SPNEGO token. The response token in SPNEGO is called NegTokenResp and the state of negotiation is accept-completed. The authentication is now complete.
[RFC4178] : The Simple and Protected Generic Security Service Application Program Interface (GSS-API) Negotiation Mechanism
[MS-SPNG]: Simple and Protected GSS-API Negotiation Mechanism (SPNEGO) Extension
Kerberos is an authentication mechanism that is extensively used in Microsoft technologies and is the one preferred by Microsoft. Current version ofKerberos is version 5 and is defined in RFC4120.
I’ll not even try to explain Kerberos. The subject is too big and there is tons of information available on this subject. The Evolution of Kerberos Authentication Service has a great introduction to Kerberos. If clicking on the link does not work, just bing the title and you will find it.
You may have noticed that the protocol in the network trace is actually called “MsKerberos” and in the list of mechanisms there are two Kerberos mechanisms: Kerberos and MsKerboros. So, what’s the difference? Well, the number you see next to the name of mechanism (1.2.840.1135188.8.131.52) is called OID (object ID). For some reason Microsoft used 1.2.840.1135184.108.40.206 for Kerberos instead of the official Kerberos OID (1.2.840.1135220.127.116.11). So the one Microsoft used is called MsKerberos. By the way, both of them are same and use of either one is okay.
Kerberos is unique in the way that it has its own port (port 88) and it causes network traffic on its own. This happens when a client get a ticket granting ticket (TGT) that is subsequently used to get service tickets from TGS (ticket granting server). So two exchanges, AS_REQ & AS_RES and TGS_REQ and TGS_RES are sent on their own and are not sent as a part of other protocol message.
AP_REQ & AP_RES are the messages that are sent as part of messages of other protocols that use Kerberos for authentication. For example, for CIFS//SMB/SMB2, AP_REQ is sent in SESSION SETUP request and AP_RES is sent in SESSION SETUP response, as I showed earlier.
Microsoft extensions of Kerberos are specified in MS-KILE. Extensions are typically additions to protocols that are defined by an RFC’s. MS-NLMP is not an extension as the document describes the whole protocol.
RFC4120 The Kerberos Network Authentication Service (V5)
[MS-KILE]: Kerberos Protocol Extensions
The Evolution of Kerberos Authentication Service
Designing an Authentication System: a Dialogue in Four Scenes is an interesting play that describes why’s of Kerberos.
There are tons of resources available on Kerberos but my feeling is that none of them are geared towards a novice. I found “The Evolution of Kerberos Authentication Service” to be the most helpful for understanding. “Designing an Authentication System” story is an honest effort to make Kerberos easy to understand but I find it rather long. I’ll recommend reading it at least twice to get something out of it.
NTLM is an old Microsoft authentication protocol that refuses to die. One major reason is that to use Kerberos, your server and client should be joined to a domain (also called Kerberos realm). In a situation where your computers are part of a workgroup, there is no KDC (key distribution center) and Kerberos cannot be used.
NTLM stands for NT LAN Manager. NTLM has gone through several changes and as such has multiple versions. NTLMv1, NTLM with Extended Session Security and NTLMv2. Sometimes NTLM with extended session security is called NTLM2 (note there is no v). Before Microsoft made NTLM specification public as MS-NLMP, people had already studied the protocol to the degree where they were able to write applications that were able to use NTLM to authenticate against Windows services. The study of NTLM by independent developers resulted in some documents describing NTLM protocol and along the way established a terminology for the protocol. This terminology is different from the terminology that Microsoft uses. NTLM2 is a result of that. Please consult The NTLM Authentication Protocol and Security Support Provider to gain an understanding of non-Microsoft NTLM terminology. Such is the popularity of the above mentioned reference that I wrote a blog that explains the terminology of MS-NLMP in terms of “The NTLM Authentication Protocol and Security Support Provider”. The blog is available at NTLM Terminology: MS-NLMP vs. http://davenport.sourceforge.net/ntlm.html.
NTLM does not rely on a third party (KDC) to authenticate a client to a server. So, wherever Kerberos cannot be used, authentication falls back to NTLM. NTLM relies on a shared secret (password) for authentication that client and authenticating server share.
In NTLM authentication, a client sends a NEGOTIATE_MESSAGE to server where client expresses its capabilities in terms of the version and options of NTLM that it supports. In response to that, server send a CHELLENGE_MESSAGE to client which contains a random byte string. Client uses that challenge in conjunction with its knowledge of user password to create LM an NT challenge responses and send those with the options it used to create the response to server in an AUTHENTICATE_MESSAGE. On the receipt of AUTHENTICATE_MESSAGE, server calculates LM and NT challenge responses based on its knowledge of user password, challenge and the options client used. If the responses match, authentication succeeds.
The above is a super-simplified picture of NTLM authentication. You should consult the resources at the end of this section for a deeper understanding of this protocol.
NTLM is also used in domain environment where server and client don’t share a secret. In this scenario, since server does not know the password of the user, it sends the challenge response to the domain controller for verification. This type of authentication is called pass-through authentication. Pass‑through authentication is described in MS-APDS and MS-NRPC.
The NTLM Authentication Protocol and Security Support Provider
MS-NLMP: NT LAN Manager (NTLM) Authentication Protocol Specification
MS-APDS: Authentication Protocol Domain Support Specification
MS-NRPC: Netlogon Remote Protocol Specification
NTLM Terminology: MS-NLMP vs. http://davenport.sourceforge.net/ntlm.html
NTLM keys and sundry stuff
In addition to Negotiate (SPNEGO), Kerberos and NTLM, there are other authentication protocols that are used in Microsoft technologies. They are not as common as the three I mentioned. MS-AUTHSO and MS-SECO are great documents for an overview of all the authentication protocols used by Microsoft. MS‑AUTHSO and MS-SECO are overview documents and describe protocols briefly but every protocol has a reference to the complete specification of the protocol.
To learn about the protocol from an independent perspective, I would highly recommend “Network Security: Private Communication in a Public World” by Charlie Kaufman, Radia Perlman & Mike Speciner. The book explains authentication techniques as well as encryption algorithms. The book is mathematical in nature but there is a lot that can be learned without a degree in mathematics.
[KAUFMAN] Kaufman, C., Perlman, R., and M. Speciner, "Network Security: Private Communication in a Public World, Second Edition", Prentice Hall, 2002, ISBN: 0130460192.
[MS-AUTHSO]: Windows Authentication Services System Overview
[MS-SECO]: Windows Security Overview
Authentication Reference http://msdn.microsoft.com/en-us/library/aa374736(v=VS.85).aspx