Primary Access Token Manipulation
Defense Evasion, Privilege Escalation by stealing an re-using security access tokens.
Last updated
Defense Evasion, Privilege Escalation by stealing an re-using security access tokens.
Last updated
One of the techniques of token manipulation is creating a new process with a token "stolen" from another process. This is when a token of an already existing access token present in one of the running processes on the victim host, is retrieved, duplicated and then used for creating a new process, making the new process assume the privileges of that stolen token.
A high level process of the token stealing that will be carried out in this lab is as follows:
Step
Win32 API
Open a process with access token you want to steal
OpenProcess
Get a handle to the access token of that process
OpenProcesToken
Make a duplicate of the access token present in that process
DuplicateTokenEx
Create a new process with the newly aquired access token
CreateProcessWithTokenW
Below is the C++ code implementing the above process. Note the variable PID_TO_IMPERSONATE that has a value of 3060 This is a process ID that we want to impersonate/steal the token from, since it is running as a domain admin and makes it for a good target:
Note the line 16, which specifies the executable that should be launched with an impersonated token, which in our case effectively is a simple netcat reverse shell calling back to the attacking system:
This is the code if you want to compile and try it yourself:
Launching Tokens.exe from the powershell console spawns a reverse shell that the attacker catches. Note how the powershell.exe - the parent process of Tokens.exe and Tokens.exe itself are running under PC-Mantvydas\mantvydas, but the newly spawned shell is running under OFFENSE\Administrator - this is because of the successful token theft:
The logon for OFFESNE\administrator in the above test was of logon type 2 (interactive logon, meaning I launched a new process on the victim system using a runas /user:administrator@offense cmd command).
Another quick test that I wanted to do was a theft of an access token that was present in the system due to a network logon (i.e psexec, winexec, pth-winexe, etc), so I spawned a cmd shell remotely from the attacking machine to the victim machine using:
which created a new process on the victim system with a PID of 4780:
Enumerating all the access tokens on the victim system with PowerSploit:
...gives the below. Note the available token (highlighted) - it is the cmd.exe from above screenshot and its logon type is as expected - 3 - a network logon:
This token again can be stolen the same way we did it earlier. Let's change the PID in Tokens.cpp of the process we want to impersonate to 4780:
Running the compiled code invokes a new process with the newly stolen token:
note the cmd.exe has a PID 5188 - if we rerun the Invoke-TokenManipulation, we can see the new process is using the access token with logon type 3:
Imagine you were investigating the host we stole the tokens from, because it exhibited some anomalous behaviour. In this particularly contrived example, since Tokens.exe was written to the disk on the victim system, you could have a quick look at its dissasembly and conclude it is attempting to manipulate access tokens - note that we can see the victim process PID and the CMDLINE arguments:
As suggested by the above, you should think about API monitoring if you want to detect these token manipulations on endpoints, but beware - this can be quite noisy.
Windows event logs of IDs 4672 and 4674 may be helpful for you as a defender also - below shows a network logon of a pth-winexe //10.0.0.2 -U offense/administrator%pass cmd and then later, a netcat reverse shell originating from the same logon session:
