The golden rule for many sandboxes is to attach the monitor to a every piece of executable code that is spawn by the analyzed sample. This approach has its obvious merits – many samples use lots of trickery and ‘seeing it all’ is a definite plus. It is also an unique selling point for some sandboxes to be able to ‘catch’ the most trickiest evasions that could otherwise potentially sneak-in under the radar of the monitor and do the evil thing while the sandbox would report nada…
I’d like to suggest a small, configurable optimization change to the sandbox behavior. One that should be relatively easy to implement for the most common use cases, and one that may be quite beneficial for readability and processing purposes.
Its name is ‘Unfollow’.
Let’s have a quick look at the below example:
Retrieves Module handle (via GetModuleHandleA): kernel32 Retrieves Procedure Address (via GetProcAddress): KERNEL32.dll, CreateProcessA Retrieves Module handle (via GetModuleHandleA): kernel32 Retrieves Procedure Address (via GetProcAddress): KERNEL32.dll, GetModuleFileNameA Loads Library (via LoadLibraryA): Shell32.dll Retrieves Procedure Address (via GetProcAddress): SHELL32.dll, SHGetFolderPathA Retrieves Module handle (via GetModuleHandleA): kernel32 Retrieves Procedure Address (via GetProcAddress): KERNEL32.dll, CreateProcessA Creates Process (via CreateProcessA): , cmd=sc stop WinDefend, flags= Retrieves Module handle (via GetModuleHandleA): kernel32 Retrieves Procedure Address (via GetProcAddress): KERNEL32.dll, CreateProcessA Creates Process (via CreateProcessA): , cmd=sc config WinDefend start= disabled, flags= Retrieves Module handle (via GetModuleHandleA): kernel32 Retrieves Procedure Address (via GetProcAddress): KERNEL32.dll, CreateProcessA Creates Process (via CreateProcessA): , cmd=net stop msmpsvc, flags=
The story the log tells us is pretty clear – the sample is launching ‘sc’ and ‘net’ commands to kill/change the config of security services:
- sc stop WinDefend
- sc config WinDefend start= disabled
- net stop msmpsvc
Look how clear the malicious behavior it is, and how easy it is to cherry-pick it from the logs, even in a textual format! Running a full-blown monitor over the spawn ‘utilities’ would be completely unnecessary… (unless of course, you want full report on IOCs, etc.).
In this particular case, my monitor just continues w/o following the programs the malware spawns i.e. recognizes their presence, but ‘unfollows’ them.
And what would happen if these were to be followed?
Let’s look at another example – the malware is executed, it connects to SCM, and first tries to Open, and if it doesn’t exist, Create the service called ‘vom’:
Connects to SCM database (via OpenSCMManagerA) Opens a service (via OpenServiceA): vom Creates Service (via CreateServiceA): vom
– at this moment the monitor would need to start monitoring the ‘services.exe’ that is responsible for service creation:
services.exe
...
Creates/Opens Registry Key: vom
Sets Registry Value (via NtSetValueKey): HKEY_LOCAL_MACHINE\SYSTEM\
ControlSet001\Services\vom\Type, REG_DWORD, 1
Sets Registry Value (via NtSetValueKey): HKEY_LOCAL_MACHINE\SYSTEM\
ControlSet001\Services\vom\Start, REG_DWORD, 0
Sets Registry Value (via NtSetValueKey): HKEY_LOCAL_MACHINE\SYSTEM\
ControlSet001\Services\vom\ErrorControl, REG_DWORD, 1
Sets Registry Value (via NtSetValueKey): HKEY_LOCAL_MACHINE\SYSTEM\
ControlSet001\Services\vom\ImagePath, REG_EXPAND_SZ,
system32\drivers\vom.sys
Sets Registry Value (via NtSetValueKey): HKEY_LOCAL_MACHINE\SYSTEM\
ControlSet001\Services\vom\DisplayName, REG_SZ, vom
Creates/Opens Registry Key: Security
Sets Registry Value (via NtSetValueKey): HKEY_LOCAL_MACHINE\SYSTEM\
ControlSet001\Services\vom\Security\Security, REG_BINARY,
...
and many many logs for all services affected
From the full-IOC report perspective – it makes sense, but if we talk about in-depth analysis where very specific high-level info is needed it actually adds a lot of noise to the report. We actually want to know what the sample does. Not how the service creation works (of course, it’s interesting, but not at this stage!).
Again, I emphasize it could be an optional setting – one that could enable individual analysts to speed up the log analysis by removing a lot of clutter from the final report, and perhaps enable the sandbox to ‘see’ more (as CPU cycles required by the ‘utility’ process monitoring and logging can be fully delegated to the main malware during – the typically short – session time). Of course, many modern reports can be ‘collapsed’ to hide the artifacts that are not that interesting + the process hierarchy is typically clearly shown on a graph, or a tree, but still – following processes is quite CPU-expensive and not always necessary.