Kerberos

A Kerberos realm is a logical group of networked computers that share a common authentication database. The authentication database is used to store the Kerberos tickets that are issued to users and services when they authenticate to the network.

In a Kerberos environment, each networked computer is a member of a realm. The realm is responsible for authenticating users and services and for issuing and managing Kerberos tickets.

A Kerberos realm can be implemented on any type of network, including networks that are not running Windows. In fact, the Kerberos protocol was developed as an open standard and is used by many different types of operating systems and networks.

On a Windows network, a Kerberos realm is typically equivalent to a domain. Each domain in a Windows network is a member of a realm, and the domain controller is responsible for authenticating users and services and for issuing and managing Kerberos tickets.

However, on a non-Windows network, a Kerberos realm can be implemented independently of any domain structure. In this case, the Kerberos server is responsible for authenticating users and services and for issuing and managing Kerberos tickets.

The Hacker Recipes mainly covers Kerberos attacks in a Windows context, but some attacks can also be conducted in non-Windows networks.

Tickets

Kerberos is an authentication protocol based on tickets. It basically works like this (simplified process):

  1. Client asks the KDC (Key Distribution Center, usually is a domain controller) for a TGT (Ticket Granting Ticket). One of the requesting user's keys is used for pre-authentication. The TGT is provided by the Authentication Service (AS). The client request is called AS-REQ, the answer is called AS-REP.

  2. Client uses the TGT to ask the KDC for a ST (Service Ticket). That ticket is provided by the Ticket Granting Service (TGS). The client request is called TGS-REQ, the answer is called TGS-REP.

  3. Client uses the ST (Service Ticket) to access a service. The client request to the service is called AP-REQ, the service answer is called AP-REP.

  4. Both tickets (TGT and ST) usually contain an encrypted PAC (Privilege Authentication Certificate), a set of information that the target service will read to decide if the authentication user can access the service or not (user ID, group memberships and so on).

A Service Ticket (ST) allows access to a specific service.

When requesting a service ticket, the client (cname) specifies the service it wants to obtain access to by supplying it's sname, which can be one of 9 types (RPC 4120 section 6.2). Shortly put, the following formats are supported:

  • servicePrincipalName

  • userPrincipalName

  • sAMAccountName

  • sAMAccountName@DomainNetBIOSName

  • sAMAccountName@DomainFQDN

  • DomainNetBIOSName\sAMAccountName

  • DomainFQDN\sAMAccountName

Note that if you use the SRV01 string as a sAMAccountName, and the SRV01 account does not exist, and the SRV01$ account exists, this name will be treated as a principal name of the SRV01$ account.

(swarm.ptsecurity.com)

On a good note, if the service name is specified as something else than an SPN (i.e. SAN, UPN), Kerberos will basically deliver service tickets if the requested service

  • has a trailing $ in the requested SAN (sAMAccountName)

  • or has at least one SPN (servicePrincipalName)

If the service ticket is requested through a specific U2U (User-to-User) request, then neither of the conditions above will be required, the target service user can be specified by its UPN (userPrincipalName).

(sources: Twitter and [MS-KILE] section 3.3.5.1.1).

The TGT is used to ask for STs. TGTs can be obtained when supplying a valid secret key. That key can be one of the following (read more).

Key name (a.k.a. etype)Details on key calculation

DES

Key derivated from user's password

RC4

Key == NT hash

AES128

Key derivated from user's password (with salt)

AES256

Key derivated from user's password (with salt)

By default, the salt is always

  • For users: uppercase FQDN + case sensitive username = DOMAIN.LOCALuser

  • For computers: uppercase FQDN + host + lowercase FQDN hostname without the trailing $ = DOMAIN.LOCALhostcomputer.domain.local

(Kerberos keys calculation)

Again, Microsoft has poorly implemented the zero-knowledge proof concept in Kerberos. An attacker knowing a user's NT hash could use it to ask the KDC for a TGT (if RC4 key is accepted). This is called Overpass-the-hash.

pagePass the key

Users are not the only ones whose NT hashes can be used to abuse Kerberos.

  • A TGT is encrypted with the krbtgt's NT hash. An attacker knowing the krbtgt's NT hash can forge TGTs impersonating a domain admin. He can then request STs as a domain admin for any service. The attacker would have access to everything. This forged TGT is called a Golden ticket.

  • A ST is encrypted with the service account's NT hash. An attacker knowing a service account's NT hash can use it to forge a Service ticket and obtain access to that service. This forged Service ticket is called a Silver ticket.

pageForged tickets

Overpass-the-hash, silver ticket and golden ticket attacks are used by attackers to obtain illegitimate tickets that can then be used to access services using Kerberos without knowing any password. This is called Pass-the-ticket.

urlhttps://github.com/ShutdownRepo/The-Hacker-Recipes/blob/master/ad/movement/kerberos/broken-reference/README.md

Roasting

If Kerberos preauthentication is disabled for a user, it is possible to request a TGT for that specific user without knowing any credentials. When the TGT is requested, the KDC sends it along with a session key in the KRB_AS_REP message to the requesting client. The session key being encrypted with the requested user's NT hash, it is possible to crack that session key offline in a an attempt to find the user's password. This is called ASREProasting.

urlhttps://github.com/ShutdownRepo/The-Hacker-Recipes/blob/master/ad/movement/kerberos/broken-reference/README.md

If an attacker finds himself in a man-in-the-middle position, effectively capturing Kerberos messages, he could capture KRB_AS_REQ messages and operate a similar cracking attempt.

pageASREQroast

When attackers have a foothold in the domain (i.e. valid domain credentials), they have the (intended) ability to request a service ticket (ST) for any valid SPN (ServicePrincipalName), or SAN (samAccountName). The ST being encrypted with the service account's NT hash, when that service account's password is weak, it is then possible to crack the ST offline in an attempt to find the password. This is called Kerberoasting. On a side note, obtaining a service ticket for a service specified by its SAN in an attempt to Kerberoast the account will only work if the service has at least one SPN.

urlhttps://github.com/ShutdownRepo/The-Hacker-Recipes/blob/master/ad/movement/kerberos/broken-reference/README.md

Delegations

Kerberos delegations allow services to access other services on behalf of domain users. For instance, this allows services to require access to other services' data on the authenticated user's behalf in order to pull data that only the said user is supposed to have access to.

In some situations, Kerberos delegations can be abused by attackers to operate lateral movement or privilege escalation.

pageDelegations

In some cases, the delegation will not work. Depending on the context, the bronze bit vulnerability (CVE-2020-17049) can be used to try to bypass restrictions.

pageBronze Bit

Service-for-User extensions

Kerberos delegations can be abused by attackers to obtain access to valuable assets and sometimes even escalate to domain admin privileges. Regarding constrained delegations and rbcd, those types of delegation rely on Kerberos extensions called S4U2Self and S4U2Proxy.

  • Service for User to Self (S4U2self): allows a service to obtain a Service Ticket, on behalf of another user (called "principal"), to itself.

S4U2self requirements

There are no particular requirements to this extension. The service does not necessarily have to have an SPN, as long as the request's sname field is formatted correctly (see Tickets). This requirement applies when combining S4U2self with U2U.

The resulting Service Ticket is forwardable (i.e. can be used with S4U2Proxy to access another service) if, and only if:

  • the service is configured for constrained delegation (KCD) with protocol transition

  • the principal is not "sensitive for delegation"

  • the principal is not a member of the Protected Users group

  • Service for User to Proxy (S4U2proxy): allows a service to obtain a Service Ticket, on behalf of a user to a different service.

S4U2proxy requirements

For this extension to work properly, the service needs to supply a Service Ticket as "additional-ticket" (i.e. used as an evidence that the service using S4U2proxy has the authority to do it on behalf of a user).

For an S4U2proxy request to work and have the KDC issue a Service Ticket:

  • the ST used as "additional-ticket" must have the forwardable flag set.

  • alternatively, in the TGS-REQ, in the pre-authentication data, the PA-PAC-OPTIONS structure must contain a padata value with the resource-based constrained delegation bit set (doc). nota bene 1: this only applies if the resource-based constrained delegation (RBCD) is actually possible and authorized in the proper AD objects attributes. nota bene 2: Rubeus and Impacket's getST always set that bit when doing S4U2proxy.

On a side note, S4U2Proxy always results in a forwardable ST, even when the ticket used as evidence wasn't forwardable.

More technical notes

S4U2self and S4U2proxy are variations of Service Ticket requests (KRB_TGS_REQ). Below is what differentiates a S4U2self from a S4U2proxy from a standard KRB_TGS_REQ.​

  • S4U2self

    • Request contains a PA-FOR-USER padata type structure containing the name and the realm of the user to impersonate (doc).

    • the cname (user name authenticating) and the sname (service name being requested) are the same. In order to succeed and not have the KDC throw an KDC_ERR_S_PRINCIPAL_UNKNOWN, the sname should refer to an account that has at least one SPN (Service Principal Name) set.

  • S4U2proxy

    • Request contains an additional-tickets field containing a service ticket. In order to succeed and not have the KDC throw an KDC_ERR_BADOPTION, the ticket should have the forwardable flag set. In a standard constrained delegation or rbcd scenario, the ticket added in the additional-tickets field is the one obtained through S4U2self.

    • Request contains the CNAME-IN-ADDL-TKT flag in the kdc-options field, indicating S4U2proxy is used (doc).

User-to-User authentication

U2U has nothing to do with S4U mechanisms and plays no part in delegation internals.

[U2U] allows users to host secure application services on their desktop machines. [...] In the user-to-user protocol, one user acts as a server, and the other user acts as a client. (Frequently Asked Questions about Kerberos).

More technical notes

A U2U request is a variation of a common Service Ticket request (KRB_TGS_REQ). Below is what differentiates a U2U from a standard KRB_TGS_REQ. It allows a user to request a service ticket to another user.

  • Request contains an additional-tickets field containing the target user TGT.

  • Request contains the ENC-TKT-IN-SKEY flag in the kdc-options field, indicating that the ticket for the end server is to be encrypted in the session key from the additional TGT provided (doc).

  • The sname refers to a UPN (User Principal Name) of an account that doesn't necessarily have to have an SPN set.

A ticket produced with u2u is encrypted with the session key of the TGT included as additional-ticket instead of the target service long-term key (i.e. DES, RC4 or AES128/256 key).

S4U2self + U2U

In some specific scenarios, S4U2self and U2U can be combined, in which case the flags and structures both mechanisms include in their requests are combined.

This allows to

Last updated