You are browsing the archive for Code Injection.

Code Execution via surgical callback overwrites (e.g. DNS memory functions)

June 12, 2019 in Code Injection, Sysmon

Today I looked at Sysmon v10 and its support for logging DNS queries. It’s a pretty cool feature that intercepts all the DNS requests on a monitored host, and if possible, maps them to the process name making that request. It is a nice addition to Sysmon’s already awesome logging capabilities.

Just for fun, I created a simple POC that used DnsQuery_A API to send a multiline DNS query, because I wanted to see how Sysmon will react to it. It was obviously a non-sensical exercise, but it’s fun to see we can modify the layout of Event Logs by introducing some unexpected, redundant data:

Anyway…

I decided to look at the DnsQuery_A function in IDA as well. I was curious if/what characters it accepts & if there is any limit to the buffer it can process. This was a quick & dirty attempt to see if I could send a query that Sysmon would truncate in a similar fashion as I described in this post.

While digging into the code I noticed an interesting way dnsapi.dll is allocating memory. Instead of a fixed (inline) function it relies on a couple of callbacks. One of them is a memory allocation routine. When the library needs memory, it calls the function, and if it is not set, it relies on its own internal routines.

This immediately caught my attention. If we can find the address of this callback inside a remote process we can use it to execute code next time DNS library asks for memory.

This is the memory allocation function used by DnsQuery_* functions (32-bit):

Under normal circumstances finding callback pointers in a remote process memory is quite hard and noisy (lots of ReadProcessMemory calls, possible disassembling). Unless of course there is an interface we can use to surgically target some specific callback (e.g. using documented windows messages, or SetProp function). As far as I can tell there is no such interface in our case.

I found a surrogate solution that we can try to exploit though.

When I looked at references to the callback function (which I named fnMemAlloc on the listing above) I discovered a exported function called DnsApiHeapReset. It takes 3 arguments and each of them is … a callback replacement:

I quickly analyzed each callback’s role and they just are 3 basic/core memory allocation/reallocation/release primitives.

So…

If we can locate the address of dnsapi.dll in a remote process (easy), find the address of exported DnsApiHeapReset function (easy), then with a basic parsing of its code we can discover the address of each callback (also easy). Then, with a single, surgical WriteProcessMemory call we can modify any of them.

This is not a new code injection trick. It’s just one way to execute code without engaging remote threads, APCs, windows hooks, side-loading, process hollowing, patching API code (e.g. NtClose), etc..

There are of course tons of other callbacks like this, but finding their exact location without any point of reference is hard. Or… not really. Just think of all the Windows Procedures – all of them are callbacks.

Beyond good ol’ Run key, Part 107

June 7, 2019 in Anti-Forensics, Autostart (Persistence), Code Injection, Living off the land, LOLBins

This is a persistence, and a code injection trick in one. It affects only environments where NVIDIA CUDA Toolkit is present. If it is the case, the system will have these two environment variables present:

  • CUDA_INJECTION32_PATH
  • CUDA_INJECTION64_PATH

They typically point to legitimate NVIDIA DLLs, but one could replace them with anything. The DLLs are loaded via LoadLibrary.

This is not a backdoor of any sort – just a legitimate profiler interface.