.NET malware is well-known by security analysts, but even existing many tools such as dnSpy,.NET Reflector, de4dot and so on to make the analysis easier, most professionals have used them as a black box tool, without concerning to .NET internals, structures, MSIL coding and details. In critical cases, it is necessary have enough knowledge about internal mechanisms and to debug these .NET threats using WinDbg.
Unfortunately, .NET malware samples have become very challenger because it is so complicated to deobfuscated associated resources, as unpacking and dumping them from memory. Furthermore, most GUI debugging tools does an inside view of mechanisms such as CRL Loader, Managed Heap, Synchronization issues and Garbage Collection.
In the other side, .NET malware threats are incredibly interesting when analyzed from the MSIL instruction code, which allows to see code injections using .MSIL and attempts to compromise .NET Runtime keep being a real concern.
The purpose of this presentation is to help professionals to understand .NET malware threats and techniques by explaining concepts about .NET internals, mechanisms and few reversing techniques.
.NET MALWARE THREAT: INTERNALS AND REVERSING DEF CON USA 2019
1. 1
.NET MALWARE THREAT:
INTERNALS AND
REVERSING
DEF CON USA 2019
DEF CON USA 2019
by Alexandre Borges
ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
2. DEF CON USA 2019
2
Malware and Security Researcher.
Speaker at DEF CON USA 2018
Speaker at DEF CON China 2019
Speaker at CONFidence Conference
2019 (Poland)
Speaker at HITB 2019 Amsterdam
Speaker at BSIDES
2019/2018/2017/2016
Speaker at H2HC 2016/2015
Speaker at BHACK 2018
Consultant, Instructor and Speaker on
Malware Analysis, Memory Analysis,
Digital Forensics and Rootkits.
Reviewer member of the The Journal
of Digital Forensics, Security and Law.
Referee on Digital Investigation: The
International Journal of Digital
Forensics & Incident Response
Agenda:
Introduction
Managed executable structures
CLR and Assembly Loader details
.NET internals metadata
Modules, assemblies and manifest
.NET program structures
Malicious code through MSIL
.NET debugging
Few GC and synchronous aspects
Conclusion
ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
3. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
3
Last talks in conferences:
CONFidence Conference 2019:
https://confidence-conference.org/2019/bio.html#id=37486
slides:
http://www.blackstormsecurity.com/CONFIDENCE_2019_ALEXANDRE.pdf
DEF CON China 2019:
https://www.defcon.org/html/dc-china-1/dc-cn-1-speakers.html#Borges
slides:
http://www.blackstormsecurity.com/docs/DEFCON_CHINA_ALEXANDRE.pdf
HITB Amsterdam 2019:
https://conference.hitb.org/hitbsecconf2019ams/speakers/alexandre-borges/
slides: http://www.blackstormsecurity.com/docs/HITB_AMS_2019.pdf
DEF CON USA 2018:
https://www.defcon.org/html/defcon-26/dc-26-speakers.html#Borges
slides: http://www.blackstormsecurity.com/docs/DEFCON2018.pdf
Malwoverview Tool: https://github.com/alexandreborges/malwoverview
5. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
5
Motivations to this talk about .NET reversing and internals:
Most of the time, professionals are interested in unpacking embedded
resources from a .NET sample.
In another moment, the concern is dumping the unpacked binary from
memory.
Sometimes, we have looked for any unpacking routine to dynamically unpack
the encrypted content.
All of these actions are correct and recommended.
However....
Many people don’t understand .NET metadata components.
Most people based their analysis on the decompiled code, but never on IL.
Malware’s authors have manipulated the IL to attack the system and even the
.NET runtime.
6. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019 6
There are many available methods to infect a system using .NET malware. Most of
the time, a .NET code decrypts and loads a native code (or injects a code into a
target process).
However, there are few approaches that use indirect techniques:
An e-mail comes from the Internet and a first dropper is downloaded.
This dropper fetches a encrypted payload, which contains a native payload
and a managed code.
The payload 1 executes and injects a DLL into a remote chosen process.
This DLL loads (and sometime decrypts) the malicious managed code.
The malicious managed code drops the payload 2 (real and advanced).
The true infection starts.
dropper
(unmanaged)
payload 1
(unmanaged)
vector
(managed)
inject a DLL in
a remote
process
payload 2
Infection
7. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
7
It is not necessary to comment about how to inject a code because the steps are
the same ever-sequence:
CreateToolhelp32Snapshot( ) Module32First( ) Module32Next( )
comparison (wcscmp( ))
VirtualAllocEx( ) WriteProcessMemory( ) CreateRemoteThread( )
WaitForSingleObject VirtualFreeEx( ).
Find the offset of injected DLL from the base module (any testing module).
Use this offset to invoke functions from any injected remote process through
GetProcessAddress( ) + CreateRemoteThread( ).
Thi injected DLL can load the next stage and, eventually, decrypt it.
Obviously, the .NET managed code can be loaded from any process or, even
worse, from an natived injected code (DLL).
After loading it, it is easy to execute it. Our simple case above.
8. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
8
We should remember that a typical native application can also load a .NET runtime
and execute a managed code:
CLRCreateInstance( ): provides the ICLRMetaHost interface.
ICLRMetaHost::GetRunTime( ): gets the ICLRRuntimeInfo.
ICLRRuntimeInfo::GetInterface( ): Loads the CLR into the current process and
returns runtime interface pointers.
ICLRRuntimeHost::ExecuteApplication( ): specifies the application to be
activated in a new domain.
ICLRRuntimeHost::Start( ): starts the the runtime.
ICLRRuntimeHost::ExecuteInDefaultAppDomain( ): invokes a method in the
.NET managed assembly (this steps does not work for all .NET assembly’s
method). Thus, in this case, starts the managed assembly.
Finally, the real infection starts.
9. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
9
The .NET framework is composed by:
CLR (Common Language Runtime), which is the .NET engine.
Libraries (System.IO, System.Reflection, System.Collections, ...).
Basically:
source code is written in C#, F#, VB.NET and Powershell.
compiled to CLI (Common Language Infrastruture Code).
executed by the CLR.
Tools used to reverse and analyze .NET malware threats are completely different
than ones used to reverse native language:
dnSpy (excellent)
ILSpy (excellent)
RedGate .NET Reflector
De4dot (deobfuscator)
Microsoft Visual Studio
WinDbg (including SOS.dll extension)
DotPeek
IDA Pro
Microsoft ILASM/ILDASM (Intermediate
Language Assembly/Disassembler)
10. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
10
Other interesting tools to analyze and understand .NET runtime are available:
MemoScope.Net: https://github.com/fremag/MemoScope.Net
Shed -- a .NET runtime inspector: https://github.com/enkomio/shed
SuperDump, for automated crash dump analysis:
https://github.com/Dynatrace/superdump
DumpMiner: https://github.com/dudikeleti/DumpMiner
MemAnalyzer: https://github.com/Alois-xx/MemAnalyzer
Sharplab: https://sharplab.io/
ObjectLayoutInspector to analyze internal structures of the CLR types at
runtime (https://github.com/SergeyTeplyakov/ObjectLayoutInspector)
Tools are excellent to help us, but most .NET malware threats have deployed the
same tricks from native code to make our job harder: packers, obfuscation and
anti-reversing techniques.
.NET Reactor
Salamander .NET Obfuscator
Dotfuscator
Smart Assembly
CryptoObfuscator for .NET
Agile
ArmDot
babelfor.NET
Eazfuscator.NET
Spice.Net
Skater.NET
VM Protect 3.40
11. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
11
There are many obfuscators, which perform:
Control flow obfuscation and dead/junk code insertion.
Renaming: methods signatures, fields, methods implementation, namespaces,
metadata and external references.
Re-encoding: changing printable to unprintable characters
Simple encryption of methods and strings.
Cross reference obfuscation.
Yes, I know... I’ve already talked about de-obfuscation in DEF CON China 2019.
Most time, the real and encoded malicious code (payload) is downloaded and
decrypted/loaded into the memory for execution:
System.Reflection.Assembly.Load( )
System.Reflection.Assembly.LoadFile()
System.Reflection.MethodInfo.Invoke( )
As we already know, Load( )/LoadFile( ) function are usually followed by:
GetType ( ) GetMethod( ) Invoke( ) (this is a typical Reflection approach)
12. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
12
Another possible approach would be:
GetAssemblyName( ) + GetType( ) + GetMethod( ) + Invoke( )
Some “encrypted content” is loaded from as a resource, so it is usual finding the
following sequence:
FindResource( ) + SizeOfResource( ) + LoadResource( ) + LockResource( )
Resources.ResourceManager.GetObject( )
Additionally, we’ve seen techniques using embedded references such as DLLs as
resources through a sequence of calls using:
AssemblyLoader.Attach( ) + AssemblyLoader.ResolveAssembly( ).
As you’ve guessed, AssemblyLoader.ResolveAssembly( ) is used to resolving
assemblies that are not available at the exact time of calling other methods, which
are external references to the binary itself.
17. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
17
As every single malware code, this one is using Reflection to retrieve information in
runtime. In this case also calls the GetExecutingAssembly( ) method to get the
Assembly object, which represents the current assembly.
18. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
18
Therefore, we can extract these resources (DLLs, for example) by using either dnSpy
or ILSpy , decrypt and load them again into the managed code.
Of course, in this case, we’ll be able to see all “hidden” references, finally.
To load the “decrypted” resources into the managed code, we can use ILSpy +
Reflexil plugin (http://reflexil.net/).
Finally, it is necessary to remove the “old” references to the embedded resources
(performed by AssemblyLoader.Attach( )) from the initializer (or removing the whole
initializer) because, at this time, they are “decrypted”.
By the way, Reflexil is able to handle different obfuscators such as Babel NET,
CodeFort, Skater NET, SmartAssembly, Spices Net, Xenocode, Eazfuscator NET,
Goliath NET, ILProtector, MaxtoCode, MPRESS, Rummage, CodeVeil,CodeWall,
CryptoObfuscator, DeepSea, Dotfuscator, dotNET Reactor, CliSecure and so on.
At end, gaining knowledge in .NET internals and metadata can be interesting.
19. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
19
Most time, there are module/type initializers (similar to TLS in native code)
executing before classes and entry point methods.
.NET protectors hardly change the entry point and, usually, the trick is in the
initializer.
.cctor( ) method is a static class constructor:
called before the Main( ) method (usually set as entry point), for example.
when the module has a .cctor (<Module>::.cctor( )), so it is run before
executing any other class initializers or even an entry point.
It is common finding unpackers, decrypters and hooks in the .cctor( )
method.
Hijacking the ICorJitCompiler::compileMethod( ) is an interesting and useful way to
take the control of the JIT engine because this method is used to create a native
code, so we find managed and native code together.
In this case: .cctor( ) hooking compileMethod( ) hiding/encryting user code.
21. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
21
Metadata works as descriptors for each structure component of the application:
classes, attributes, members, and so on.
Remember that a .NET application is composed by:
managed executable files, which each one contains metadata
managed code (optionally)
.NET Assembly: managed .NET application (modules) + class libraries + resources
files (more information later)
CLR runtime environment: loaders + JIT compiler.
.NET source code .NET compiler module (IL + metadata) CLR ( loaders +
JIT compiler) native instruction Execution Engine
22. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
22
Managed module is composed by:
PE header: If the module contains only IL code, so most of information of
header is ignored. However, if the module also contains native code, so things
are different.
CLR header: contains the version of the CLR, token of Main( ) (natural entry
poiint), resources and so on.
Metadata: describe types and members. Additionally, it helps the GC to track
the life time of objects.
IL (Intermediate Language) code: the managed code.
Managed Modules
Resource Files
Compiler (C#, VB, F#)
+ Linker
Managed Modules
Resource Files
Manifest
.NET Assembly
(.exe or .dll)
23. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
23
PE Header
Native Code /
Data
CLR Header
CLR Data
(ILcode, metadata,
managed resources)
DOS Header
PE Header
Data Directories
(size and location of CLR header)
Section Headers
.text
(includes MSIL and metadata)
.idata
.data
Remaining sections
Managed resources in contained into .text section (and not .rsrc section).
24. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
24
Metadata is composed by tables such as:
Definition tables: ModuleDef, TypeDef, MethodDef, FieldDef, ParamDef,
PropertyDef and EventDef
Reference tables: AssemblyRef, ModuleRef, TypeRef and MemberRef.
Manifest tables: AssemblyDef, FileDef, ManifestResourceDef and
ExportedTypesDef.
Most malicious .NET malware samples have:
Used code manipulation (encryption/decryption) in .ctor( )/.cctor( )/Finalize( )
Called unmanaged functions from DLLs using P/Invoke.
Used COM components (very usual).
27. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019 27
Metadata describes all declared or referenced data in a module such as classes,
members, attributes, properties and relationships.
Metadata is organized as a relational database using cross-references and making
possible to find what class each method comes from.
Metadata are represented by named streams, which are classified as metadata
heaps and metadata tables.
slot 1: Class A -- methods at slot 1
slot 2: Class B -- methods at slot 3
slot 3: Class C -- methods at slot 5
slot 4: Class D -- methods at slot 6
slot 5: Class E -- methods at slot 8
slot 1: Method 1 - Classe A
slot 2: Method 2 - Classe A
slot 3: Method 1 - Classe B
slot 4: Method 2 - Classe B
slot 5: Method 1 - Classe C
slot 6: Method 1 - Classe D
slot 7: Method 2 - Classe D
slot 8: Method 1 - Classe E
28. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
28
Metadata heaps:
GUID heap: contains objects of size equal to 16 bytes.
String heap: contains strings.
Blog heap: contains arbitrary binary objects aligned on 4-byte boundary.
There can be 6 named streams:
#GUID: contains global unique identifiers.
#Strings: contains names of classes, methods, and so on.
#US: contains user defined strings.
#~: contains compressed metadata stream.
#-: contains uncompressed metadata stream.
Blob: contains metadata from binary objects.
An important note: compressed and uncompressed named streams are
mutually exclusive.
Metadata tables:
The schema defines the metadata tables by usings a descriptor.
There are more than 40 metadata tables.
29. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
29
Tokens have 4 bytes, which the first byte determines the metadata table and the
three remaining bytes are the RID.
RID (record identifiers) are used as row indexes in metadata tables.
Tokens determines which metadata tables are being referred.
Unfortunately, tokens don’t cover all tables (auxiliary tables, which are hardcoded).
0(0x0): Module
1(0x1): TypeRef
2(0x2): TypeDef
3(0x3): FieldPtr
4(0x4): Field
5(0x5): MethodPtr
6(0x6): Method
7(0x7): ParamPtr
8(0x8): Param
9(0x9): InterfaceImpl
10(0xa): MemberRef
11(0xb): Constant
12(0xc): CustomAttribute
13(0xd): FieldMarshal
14(0xe): DeclSecurity
15(0xf): ClassLayout
16(0x10): FieldLayout
17(0x11): StandAloneSig
18(0x12): EventMap
19(0x13): EventPtr
20(0x14): Event
21(0x15): PropertyMap
22(0x16): PropertyPtr
23(0x17): Property
24(0x18): MethodSemantics
25(0x19): MethodImpl
26(0x1a): ModuleRef
27(0x1b): TypeSpec
28(0x1c): ImplMap
29(0x1d): FieldRVA
30(0x1e): ENCLog
31(0x1f): ENCMap
32(0x20): Assembly
33(0x21): AssemblyProcessor
34(0x22): AssemblyOS
35(0x23): AssemblyRef
36(0x24): AssemblyRefProcessor
37(0x25): AssemblyRefOS
38(0x26): File
39(0x27): ExportedType
40(0x28): ManifestResource
41(0x29): NestedClass
42(0x2a): GenericParam
43(0x2b): MethodSpec
44(0x2c): GenericParamConstraint
32. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
32
Check the installed .NET version:
Subdirectories under C:WindowsMicrosoft.NET
clrver.exe
clrver.exe -all
Programming directly in IL (Intermediate Language) can be interesting because:
IL is stack based, so we don’t find any instruction related to register
manipulation.
Ngen.exe can be used to compile IL instructions to native code.
Eventually, malware threats have attacked the .NET runtime to subvert the
system.
Assemblies can be classified as:
private: it is specific of an application and deployed at same directory.
shared: it is shared and used by other applications.
33. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
33
In .NET applications:
.NET Assembly:
In malware samples, we usually find that resources are encrypted
binaries and DLLs.
Remember that the application can download assembly files from a URL
(codeBase element).
.NET malware have used multi-file assemblies, partitioning types over
different files. Unfortunately, it is only possible to create multfile
assembly in the command line.
Few malware authors have create .NET malware containing different
types: such as C# and VB in the same assembly.
34. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
34
Compile multi-file .NET malware is pretty easy:
csc.exe /t:module hooking.cs
csc.exe /t:module injection.cs
csc.exe /out:malwarelib.dll /t:library /addmodule:hooking.netmodule
/addmodule:injection.netmodule Defcon.cs
In this case, we have a multi-file assembly:
includes a managed module named hooking.netmodule and
injection.netmodule. The output file is a DLL named malwarelib.dll
a manifest file wrapping everything.
This compiling command add the hooking.mod file to the FileDef manifest
metadata table and the its exported types to the ExportedTypeDef manifest
metadata table.
To check: ILDasm View MetaInfo Show! and look for the FileDef and
ExportedTypeDef tables.
35. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019 35
Native modules referred by the
assembly. The module name is
in the ModuleRef.
External assemblies that
referred by the assembly
(AssemblyRef table).
Manifest
Used when a strong
assembly is specified.
38. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
38
Of course, we could have a “big malware module” to use in projects:
al.exe /out: BigMalwareLib.dll /t:library hooking. netmodule injection.
netmodule
csc.exe /t:module /r:BigMalwareLib.dll Defcon.cs
al /out:Defcon.exe /t:exe /main:Defcon.Main Defcon.netmodule
In this case, the __EntryPoint() global function will contain the Defcon::Main( )
function call (check the IL code to confirm it).
It is not necessary to mention that malware’s authors usually don’t write strong
assemblies, which as signed with the private/public key pair from the publisher.
Unless that this key pair has been stolen...
csc.exe /out:TestProgram.exe /t:exe Program.cs
sn.exe -k AlexandreBorges.snk
sn.exe -p AlexandreBorges.snk AlexandreBorges.PublicKey sha256
Sn.exe -tp AlexandreBorges.PublicKey
csc.exe /out:TestProgram.exe /t:exe /keyfile:AlexandreBorges.snk Program.cs
40. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
40
Once the system is compromised through a native malware and we have access to
the system as administrator, so it is possible to copy our .NET assembly to the
Global Assembly Cache (GAC). The Registry is not changed.
Once a malicious .NET assembly (first stage, as a resource library) is copied to GAC,
so it can be accessed by other assemblies.
Thus, other malicious .NET malware samples (second stage) can access methods
and types from the first stage.
Only strong assemblies (signed) can be copied to the GAC (located at
C:WindowsMicrosoft.NETassembly) by using GACUtil.exe /i command.
Futhermore, including /r option integrates the assembly with the Windows install
engine.
Unfortunately, the GACUtil.exe is not available in home-user systems, but it is
possible to use the MSI to install the malware threat into the GAC.
At end, it is still feasible to using delay signing, which is a partial signing only using
the public key. Therefore, private key is not used (and there isn’t real protection).
41. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
41
The delay signing allows that the malicious assembly to be installed into the GAC
and, worse, other assemblies can make reference to it.
csc.exe /out:malware.dll /t:exe Program.cs
sn.exe -k AlexandreBorges.snk
sn.exe -p AlexandreBorges.snk AlexandreBorges.PublicKey sha256
sn.exe -tp AlexandreBorges.PublicKey
csc.exe /out:malware.dll /t:exe /keyfile:AlexandreBorges.PublicKey /delaysign
Program.cs
sn.exe -Vr malware.dll (CLR trust in the assembly without using the hash).
Using csc.exe /resource makes simple to add resources (generated by resgen.exe ,
for example). It updates the the ManifestResourceDef table.
It is not so hard to perform a supply-chain attack because, when a file is specified
as reference in the csc.exe compiler using /reference switch, it looks at:
the working directory
csc.exe directory
directory specified by the /lib switch
directory specified by the LIB environment variable.
42. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
42
Several malware samples have been modified or written directly in ILAsm to
bypass common tools.
While ILAsm is not complicated, maybe it is still recommended to remember few
directives and instructions.
.assembly DefCon { }: identifies the current assembly as being DefCon.
.assembly extern <assemblyname>: determines the external managed assembly
used by the program. For example, .assembly extern <mscorlib>
.module malware.dll: identifies the current module.
.namespace Conference: identities the namespace, but it does not represent a
metadata.
.class public auto ansi Hacker entends [mscorlib]System.Object. Its keywords;
.class: identifies the current class (Hacker)
public: specifies the visibility. For example, it could be “private”.
auto: determines the class layout style. It could be “explicit” and
“sequencial”.
ansi: string encode while communicating to unmaged code. Other values are
autochar and unicode.
extends: determines its base class
43. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
43
Other flags for .class directive are:
private: used with private classes, which are not visible outside the current
assembly.
sealed: the current class can’t be derived from this class.
abstract: the current class can’t be instantiated (it holds abstract methods).
explicit: the loader preserve the order of fields in the memory.
sequential: the loader preserves the order of the instance fields as specified
in the class.
nested family: the class is visible from the descendants of the current class
only.
nested assembly: the class is visible only from the current assembly.
nested famandassem: the class is visible from the descendants of the current
class, but residing in the same assembly only.
windowsruntime: the class is a Windows runtime type.
.class public enum Exam: declares a class enumeration named “Exam”.
.ctor( ): instance constructor, which is related to instance fields.
.cctor( ): class constructor (known as type initializer), which is related to static
fields.
44. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
44
During malware analysis, take care: there can be a global .cctor directive, which it
is related to global fields.
call: call a method. Its possible keywords:
return type: void, int32, and so on.
vararg: variable number of arguments
calli: directive used to call methods indirectly by taking arguments + function
pointer.
ldc.i4.0
ldc.i4.1
ldc.i4.2
ldftn void DefCon::Test(int32, int32, int32)
calli void(int32, int32, int32)
(method reference):
call instance void DefCon::Exam(int32, int32, int32)
call instance [.module malware.dll]::Hooking(int32, int32, native int)
45. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
45
.field: specifies a variable of any type declared directly in the class (or struct). Its
main keywords can be:
public / assembly / family (accessed by any decending class) / private
static (shared by all instances of the referred class).
.method: specifies the method declaration. Its main keywords (flags) can be:
public / static: similar meaning as especified in “field” explanation above.
cil managed: it means this method is represented in managed code.
newslot: creates a new slot in the virtual table of the class to prevent that a
existing method (same name and signature) to be overriden in a derived class.
native unmanaged: it means this method is represented in a native code.
abstract: of course, no implementation is provided.
final: as known, the method can’t be overridden.
virtual: method can be “redefined” in derived classes.
strict: this method can only be overridden whether it is accessible from the
class that is overriding it. Of course, the method must be virtual.
noinline: it is not allowed to replace calls to this method by an inline version.
pinvokeimpl: declares an unmanaged method from a managed code (it’s is
also known as P/Invoke mechanism).
46. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
46
.method public hidebysig static pinvokeimpl("user32.dll" winapi) int32
FindWindow(string,string) cil managed preservesig
preservesig: return of method must be preserved.
FindWindows(string,string): function invoked from the “user32.dll” and that
returns a int32 value.
.class public DefCon implements InterfaceA,InterfaceB {
.method void virtual int32 IfB_Speaker(string) {
.override InterfaceB::Speaker
...
}
.class public DefConChina extends DefCon {
.method public specialname void .ctor( ) {
ldarg.0
call instance void DefCon::.ctor( )
ret }
callvirt instance void DefCon::IfB_Speaker( )
47. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
47
.entrypoint: identifies the method as the entry point of the assembly.
.maxstack: defines the maximum stack depth used by the function code
.locals int: defines the local variable of the current method and the “init” keyword
is initializing the variable with “zero” (for example, a integer variable).
.data <var_1>: defines a data segment named “var_1”.
stloc <var>: retrieves the value returned by the call and stores into the “var”
variable.
ldarg.0: Load argument 0 onto the stack.
ldloc <var>: copies the value of “var” onto the stack. Variants, after optmization
and run, such as ldloc.0, ldloc.1, ldloc2 and ldloc3 (representing the first local
variables) are possible.
ldstr: loads the reference to a string onto stack.
ldsflda: loads the reference of a static field onto the stack.
ldsfld: loads the value of a static field onto the stack.
ldc.i4 8: loads the constant value 8 onto the stack.
48. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019 48
br Borges: its unconditional jump similar to “jmp” in native assembly. In this case,
jumping to “Borges” label.
brtrue DefCon: takes an item from stack and, if it is zero, so jumps to “Alex”
branch. Similar to jz instruction.
brfalse Alex: takes an item from stack and, if it is one, so jumps to “Alex” branch.
Similar to jnz instruction.
.this: it is a reference to the current class (not instance of the class like C++).
.base: it is a reference to the parent of the current class.
.typedef: creates a alias to a type.
.try / catch: the same meaning of traditional C language.
49. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
49
auto: loader defines the
“best lay out” in the
memory”
nested and sealed class!
specialname flag helps
the loader to
understand this is a
special function
(constructor)
Remember that a field is a variable of any type that is
declared directly in a class or struct.
50. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
50
instance
constructor
load argument 0 onto the stack.
reserving 8 slots for arguments.
push 0 onto stack as int32.
loads a string reference onto stack.
replaces the value of a field with a value from stack.
Using a custom attribute statement to set
the value of CompilerGeneratedAttribute .
52. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
52
It calls a virtual method, which can
be overriden the the derived class.
It loads -1 onto the stack.
Calls a static method named GetCurrentProcess( ) from
Process class (within namespace System.Diagnostics) and
returning an instance of Process class.
Duplicate the value of
the top of the stack.
legal instructions to way
out of a “try block.”
54. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
54
family: can be accessed by any class
descending from the current one.
ldfld: loads the instance field onto the stack.
ldsfld: loads the static field onto the stack.
Event declaration. We should
remember that all events must
have a subscribing method
(.addon ) and a unsubscribing
method (.removeon), at least.
Cleaning up
55. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019 55
Declaring three local class variables in
three different slots: 0, 1 and 2. We should
remember that, eventually, slots of same
type can be reused. However, it is another
talk...
56. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019 56
Finally, when the publisher calls the Invoke method of the
aggregate delegate, so the event is raised.
Delegates are references representing “type-safe”
function pointers. Thus, Combine( ) adds callback
function pointers to an aggregate, which is used to
implement the event.
Generic Delegate! Compares the second and third arguments and, if
it’s equal, replace the first argument (!!0&).
58. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019 58
pushes a null references onto the stack
converts to int64 and pushes it onto the stack
converts to int32 and throws an exception when overflow
declares and initializes the local variable
Function used to decrypt strings
loads the local variable onto stack.
59. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
59
DLL loaded from the Global Assembly Cache can and need to be monitored to
detect strange behavior. Tools to log the DLL loading such as Fuslogvw.exe
(Assembly Bind Log Viewer) and common applications such as Process Monitor can
help us.
Of course, .NET malware threats can try to compromise the .NET runtime class
libraries and JIT, which would cause a deep infection in the system and demand a
detailed investigation because:
changing the runtime library (at IL code) can be lethal to many applications.
it is feasible to change (hooking)/replace a runtime library.
Changing JIT cause same problems, but it is harder.
Remember about basics:
copy DLL from GAC dnSpy/Reflector + Reflexil ildasm change ilasm
Ngen copy back to GAC (malware dropper can accomplish this task)
60. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
60
Of course, nothing is so simple:
If the malware’s target is a DLL from .NET runtime, so it is digitally signed and
it would be necessary to have the private key to sign it. Unfortunately, we
don’t have.
Another option would be to generate a new pair of keys and re-sign all the
DLL Framework. Unfortunately, it is so much work.
Copying a modified runtime DLL over the existing one can be difficult or
almost impossible because other programs can be using it. Thus, we should
stop programs and services to accomplish this task.
Eventually, it is necessary to reboot the machine (urgh!) to perform this copy
from a script.
Using the new and modified DLL can be tricky: uninstall the existing native
library (ngen uninstall <dll>) and remove it from its respective directory
under NativeImages_<version> directory.
There are other many tricks such as dropping an assembly into C:WindowsSystem32
or Syswow64)TasksTasks.dll (hint from Casey Smith)
61. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
61
An alternative would be change the Registry. In this case, the GAC continue
being associated to the original (and untouched) assembly, while its
associated native image is changed to the modified version.
In this case:
HKEY_LOCAL_MACHINESOFTWAREMicrosoftFusionNativeImagesInd
exv2.0.50727_64IL key holds information (name + signature key)
about the original assembly.
HKEY_LOCAL_MACHINESOFTWAREMicrosoftFusionNativeImagesInd
exv2.0.50727_64NI key would hold information (name + MVID) about
the modified native image.
Using the MVID from NI key makes simple to locate the native image.
Thus, we can either override the native image with a modified version or
change the MVID entry to point to another native image.
GAC (old .NET assemblies) / GAC_32 (IL and x86) / GAC_64 (IL and x64) /
GAC_MSIL (IL code) directories are under C:WindowsAssembly directory.
62. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
62
Most of the time, .NET malwares attacking the .NET libraries try either to remove
some check or introduce hooking points at entry or exit of a method. In this
case, System.Reflection is commonly used.
Additionally, there are cases of .NET malware threats attacking applications and
the service management offered by System.ServiceProcess.ServiceBase class and
their associated method such as OnStart( ), OnStop( ), Run( ), ServiceMain( ) and
so on.
Modifying a target code for changing the execution flow demands inserting
references (.assembly extern directive) to signed libraries (version + public key)
to be able to access member and call methods.
Of course, we should remember to increase the stack (.maxstack).
At end, we have multiple types of attacks from a malicious managed code by
establishing a C2, intercepting communication with trusted websites, opening
shells, gathering system information, launching native second stage droppers,
intercepting filesystem communication, stealing information and so on.
63. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
63
WinDbg is always an excellent tool to understand a .NET malware in a better way
or even getting a basic understanding, at least.
Install SOSEX extension:
Download it from http://www.stevestechspot.com/downloads/sosex_64.zip
or http://www.stevestechspot.com/downloads/sosex_32.zip
Unpack it and copy to your WinDbg installation directory. For example:
C:Program Files (x86)Windows Kits10Debuggersx64|x86
Attach the WinDbg to either a running application (the .NET malware) or a saved
dump.
Remember that the CLR process is composed by:
System Domain
Shared Domain
Default Domain
code running at this domains can’t access resources from another
application domain.
64. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019 64
0x25B05A: Entry Point from dumpbin /headers malware1.exe
Remember:
Malware executes
Win loaders find the PE’s entry point
Jump to mscoree.dll
Call to CorExeMain( )
Return to assembly’s entry point.
Disassembling CorExeMain( ) from the start.
We could have used before this point: sxe ld mscorwks.dll ; g
69. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019 69
Get a list of managed threads. Of
course, we could used the -special
option to get additional information.
Checks the
unmanaged stack
trace for this thread.
COM Threading Model:
STA: Single Thread Apartment
MTA: Multi Thread Apartment
Threat state:
(0x0) Newly
initialized thread.
(0x020) It can enter
a Join.
(0x200) background
thread.
71. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
71
Checks the managed stack
Display information about the MethodDesc structure
Check if the instruction pointer address belongs to the JIT code.
Method definition.
Remember: Metadata token is
composed by a Table Reference (1 byte)
and a Table Index (3 byte).
74. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019 74
Dump information about the Method Table
Dump information about the Method
Table and display a list of all methods.
Code is PreJIT compiled
Type definition
EEClass data structure is similar to
the method table, but it stores
fields that are less frequently used.
75. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019 75
Dump information about a specific module
Dump information about the assembly (as shown previously)
Data accessed and/or updated less frequently
Data accessed and/or updated very frequently
Data used to help COM operations
79. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019 79
Shows information about the EEClass structure
Set up a breakpoint on a code that is not JIT yet.
Displays the MethodTable
structure and EEClass structure
of test.Client.Verbunden method.
Displays the MethodDesc
structure information
82. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019 82
As a general overview, during allocation requests:
If the maximum expected memory for the Gen0 is
exceeded, collect non-rooted objects and promote
rooted objects to Gen 1.
The same approach is valid when collecting objects
from Gen 1 and Gen 2.
If Gen 2 is exceeded, so GC adds a new segment to
Gen 2.
Objects in Gen 0 and 1 are short-lived.
Reference chain
to the object
from stack...
from handle tables...
from the previous slide
83. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
83
The Finalization Queue contains
objects with finalizers (Finalize( )).
When an object in Finalization
Queue becomes rootless, so the
GC put it into the f-reachable
queue, which are considered
garbage (but alive).
85. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019 85
Dumps the process
to later analysis
Look for the string in
the managed heap.
It shows information about locks
Make easier to find deadlocked threads
Displays information
about a type or variable
It could seems unbelievable,
but some malware samples
don’t work because deadlocks
If there is some deadlock, so
use the DumpObj command to
find additional information
about the thread.
CCW: COM Callable Wrapper
RCW: Runtime Callable Wrapper,
which intercepts, manage the
object’s lifetime and the
transition between managed
code and native code.
86. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019 86
Remember that an event works as
synchronization object.
When an event happens (going from
non-signaled state to signaled state), the
waiting thread (WaitForSingleObject( ))
starts its execution.
Auto reset: If the event is signaled, so
allows the thread being release and it is
automatically reset to non-signaled state.
Manual reset: the event remains in
signaled state until being intentionally
reset.
Other synchonization techniques could
be Semaphores, ReaderWriterLock,
Mutex and so on...
It shows specific-object handle information
88. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019 88
Get objects (and their respective metadata) stored
in the heap. To a short output, use !DumpHeap -stat
Dumps the heap, but
limit the output to the
specified type name.
Class !
90. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019 90
Boxing turns a value type
into an object reference
(reference type)
Unboxing turns a object
reference into a value type
!DumpIL displays the IL
instructions of a method
92. ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
DEF CON USA 2019
92
Other possible WinDbg breakpoints that could be used to gather further
information:
How to log API calls:
bp mscorwks!MethodTable::MapMethodDeclToMethodImpl
bp clr!MethodTable::MapMethodDeclToMethodImpl
How to get possible strings:
!bpmd mscorlib.dll System.String.CreateStringFromEncoding
!bpmd mscorlib.dll System.String.Intern
!bpmd mscorlib.dll System.Text.StringBuilder.ToString
bp mscorwks!GlobalStringLiteralMap::GetStringLiteral
bp clr!StringLiteralMap::GetstringLiteral
How to examine loaded assemblies:
bp mscorwks!CLRMapViewOfFileEx
bp clr!AssemblyNative::LoadFromBuffer
96. DEF CON USA 2019 96
ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER
Acknowledgments to:
DEF CON staff, who have been always very kind with
me.
You, who have reserved some time attend my talk.
Security is like a drunk: while walking back-and-forth, he
always proceeds halfway through the remaining distance,
but he never gets there.
Remember: the best of this life are people.
97. DEF CON USA 2019
97
Malware and Security Researcher.
Speaker at DEF CON USA 2018
Speaker at DEF CON China 2019
Speaker at CONFidence Conference
2019 (Poland)
Speaker at HITB 2019 Amsterdam
Speaker at BSIDES
2019/2018/2017/2016
Speaker at H2HC 2016/2015
Speaker at BHACK 2018
Consultant, Instructor and Speaker on
Malware Analysis, Memory Analysis,
Digital Forensics and Rootkits.
Reviewer member of the The Journal
of Digital Forensics, Security and Law.
Referee on Digital Investigation: The
International Journal of Digital
Forensics & Incident Response
THANK YOU FOR
ATTENDING MY TALK.
Twitter:
@ale_sp_brazil
@blackstormsecbr
Website: http://www.blackstormsecurity.com
LinkedIn: http://www.linkedin.com/in/aleborges
E-mail: alexandreborges@blackstormsecurity.com
ALEXANDREBORGES–MALWAREANDSECURITYRESEARCHER