Rootkits: What they are and how to find them Part 2

Xeno Kovah 2010 xkovah at gmail

All materials is licensed under a Creative Commons Share Alike license.

[Link]

System Calls Revisited

You need to see the full path, and know that attackers can hook basically everywhere along the path.

Conceptual Separation of Duties

Hardware

ReadSector(123) Kernel

Data

OpenFile([Link])

Memory Mapped File

Userspace
Open Vid-yo-game!
4

User

Conceptual Separation of Duties

Hardware

ReadSector(423) Kernel

Data

ReadKey(HKLM\Bla)

Registry Information

Userspace
Open Vid-yo-game!
5

User

Conceptual System Call Interface

Kernel
NtReadKey() NtWriteFile() NtShutdownSystem() NtOpenSemaphore()

Userspace
ReadKey() WriteFile() ShutdownSystem() OpenSemaphore()

User

Yay Comp-utors!
6

Slightly More Accurate System Call Interface

Kernel
NtReadKey() NtWriteFile() NtShutdownSystem() NtOpenSemaphore()

KiFastCallEntry() or KiSystemService()

[Link]
ReadKey() WriteFile() ShutdownSystem() OpenSemaphore()

User

Yay Comp-utors!
7

Kernel User
[Link] NtWriteFile(){ mov eax, 0x112 int 0x2E OR sysenter }

[Link] WriteFile(){ Call IAT:NtWriteFile() } Start Here [Link] Call IAT:WriteFile()

Hook inline at target (seen it)

Hook IAT (seen it)

The INT 0x2E path is the Win2k path

[Link] KiSystemService(){ }

IDT
From userspace INT 0x2E KiSystemService

Hook IDT (seen it)

Kernel
User
9

The sysenter path is the > Win2k path

[Link] KiFastCallEntry(){ }

KiFastCallEntry != KiSystemService

From userspace sysenter

Kernel
User

Hook sysenter (IA32_SYSENTER_EIP MSR) new

10

unused IIS [Link] (if installed and running) unused Native API
KeServiceDescriptorTable

unused IIS [Link] (if installed and running) [Link] API Native API
KeServiceDescriptorTableShadow

[Link] KiSystemService() or KiFastCallEntry(){ * Consult Thread Info * Extract address of System Service Descriptor Table (SSDT) which is KeServiceDescriptorTable normally or KeServiceDescriptorTableShadow if the process has used any graphical (GDI) routines * Parse eax for specific table entry }

Kernel
User
11

unused IIS [Link] (if installed and running) unused Native API struct SystemServiceDescriptorTable{ PULONG_PTR ServiceTableBase; PULONG ServiceCounterTableBase; ULONG NumberOfServices; PUCHAR ParamTableBase; };
KeServiceDescriptorTable

Index to function mappings change between releases to discourage assumptions and SSDT hooking 0x112 - nt!NtWriteFile 1 - nt!NtAccessCheck 0 - nt!NtAcceptConnectPort

service number = eax = 0x112

00000100010010

2 bits table index

Kernel
User

12 bits service index

12

 0x112 - nt!NtWriteFile

Hook SSDT (new)

[Link] NtWriteFile(){ }

Hook inline at target

1 - nt!NtAccessCheck
0 - nt!NtAcceptConnectPort

13

Kernel User
[Link] NtGdiUpdateColors(){ mov eax, 0x112E int 0x2E OR sysenter } UpdateColors(){ Call IAT:NtGdiUpdateColors () }

Hook inline at target (seen it)

Start Here

[Link] Call IAT:UpdateColors()

Hook IAT (seen it) 14

unused IIS [Link] (if installed and running) unused Native API
KeServiceDescriptorTable

unused IIS [Link] (if installed and running) [Link] API Native API
KeServiceDescriptorTableShadow

[Link] KiSystemService() or KiFastCallEntry(){ * Consult Thread Info * Extract address of System Service Descriptor Table (SSDT) which is KeServiceDescriptorTable normally or KeServiceDescriptorTableShadow if the process has used any graphical (GDI) routines * Parse eax for specific table entry }

Kernel
User
15

unused IIS [Link] (if installed and running) [Link] API struct SystemServiceDescriptorTable{ PULONG_PTR ServiceTableBase; PULONG ServiceCounterTableBase; ULONG NumberOfServices; PUCHAR ParamTableBase; }; Native API
KeServiceDescriptorTableShadow

Index to function mappings change between releases to discourage assumptions and SSDT hooking
0x12E - win32k!NtGdiUpdateColors

1 - win32k!NtGdiAbortPath 0 - win32k!NtGdiAbortDoc

service number = eax = 0x112E

01000100101110

2 bits table index

Kernel
User

12 bits service index

16

Hook SSDT (new)

0x12E - win32k!NtGdiUpdateColors

[Link] NtGdiUpdateColors(){ }

Hook inline at target

1 - win32k!NtGdiAbortPath
0 - win32k!NtGdiAbortDoc

17

Big Picture

win32k
KeServiceDescriptorTable

KeServiceDescriptorTableShadow

2e

nt
KiSystemService KiFastCallEntry

kernel user

IDT

INT 2e (depending on Windows/HW version)

[Link] [Link]

sysenter (depending on Windows\HW version)

18

[Link]

O.M.G. it's Yoshi!!!

win32k
KeServiceDescriptorTable

(and Mario is doing the splits)

KeServiceDescriptorTableShadow

nt
KiSystemService KiFastCallEntry

kernel user

INT 2e

[Link] [Link]

sysenter

[Link]

19

Nu2U - sysenter
We never talked about the sysenter instruction in Intermediate x86, due to lack of time, even though it would work well in that class. Sort of as a background notion when we were talking about interrupts that they were 1) a way for hardware to get the CPU's attention, and 2) a way to get some kernel code to execute (BreakOnThruToTheOtherSide lab, discussion of interrupts underlying debugging.) So "back in the day" systems would implement the "system call table"(*nix) or "system service descriptor table" (Windows) as a way for userspace code to ask the kernel to do specific actions for it. E.g. open a file, allocate some memory, This was achieved by putting a system call number in some register and then calling int 0x80 (linux), or int 0x2e (Windows). The code on the kernel side would then just check the designated register(s) which were input parameters and call the appropriate kernel library function.
20

Book page 279, 321

Out with the old, in with the Nu2U

Intel and AMD introduced a specific instruction for achieving this same sort of system call table capability for kernels, but doing it more efficiently. The instructions for doing this are syscall/sysret on AMD and sysenter/sysexit on Intel. Linux used int 0x80 <= 2.4, and sys* >= 2.5, Windows used int 0x2e <= Win2k, sys* >= XP
21

MiSeRly MiSeRy MiSanthRopy

The syscall/sysenter instructions basically just jump to a predefined location in the kernel ala an interrupt. That location is predefined by using a "Model Specific Register" (MSR) MSRs are special registers which exist on specific models and have specific purposes (not a "general purpose" register like eax, ebx, etc.) You read and write MSRs with "rdmsr" (read msr) and "wrmsr (write MSR) IA32_SYSENTER_EIP = 0x176
22

MiSeRly MiSeRy MiSanthRopy

IA32_SYSENTER_EIP = 0x176 Reading from the MSR
mov ecx, 0x176 rdmsr (eax now contains value that was in the MSR)

Writing a MSR
mov eax, 0xdeadbeef mov ecx, 0x176 wrmsr (IA32_SYSENTER_EIP now holds the value 0xdeadbeef)
23

More about system calls

For more into on int vs sys*, as well as how interrupts work and worked on Windows:
How Do Windows NT System Calls REALLY Work? - [Link] c8035/ System Call Optimization with the SYSENTER Instruction - [Link] 8223

It's going to make a whole lot more sense thanks to Intermediate x86 :)
24

NooTooYoo - SSDT
WinDbg command to print tables: !for_each_thread ".echo Thread: @#Thread; dt nt!_kthread ServiceTable @#Thread" (from [Link]

25

Book page 286, 324

KeAddSystemServiceTable()
He4Hook uses KeAddSystemServiceTable() (which was first talked about in Hoglund's NT Rootkit phrack article) to talk from its userspace component to kernel KeAddSystemServiceTable() adds in one of those SystemServiceDescriptorTable structs onto the table pointed to by KeServiceDescriptorTableShadow
26

He4Hook KeAddSystemServiceTable
ASSUMING NO IIS INSTALLED

unused
unused [Link] API Native API

unused He4Hook table [Link] API

Native API
KeServiceDescriptorTableShadow after KeAddSystemServiceTable()

KeServiceDescriptorTableShadow before KeAddSystemServiceTable()

ASSUMING IIS INSTALLED

unused IIS [Link] [Link] API Native API

He4Hook table IIS [Link] [Link] API Native API

KeServiceDescriptorTableShadow after KeAddSystemServiceTable()
27

KeServiceDescriptorTableShadow before KeAddSystemServiceTable()

SSDT Hook

28

From: [Link]

SSDT Hook

hehehe

29

From: [Link]

SSDT False Positives

(go look at the overall SSDT results again at this point)

How you could determine these are due to symantec and not a rootkit is given in the tiddlywiki file in the class materials

30

NouTouYou - IRP
Windows uses an abstraction called IO Request Packets (IRPs) in order to send events to and from hardware IO devices. Drivers can attach to devices with IoAttachDeviceToDeviceStack(), which is how they indicate that they would like to hear about IRPs to/from a specific device. They can also just not attach to the stack, and instead intercept the calls to someone who's already attached.
31

Book page 325, 457

A tale of two objects

Driver Object
Every driver gets this object passed to it when it's loaded as the first parameter of the required driver entry point function

Device Object
Used to create a linked list which holds the other devices for other drivers which want to hear about IRP activity

32

IRP chain
Driver Objects
"\Driver\Ctrl2Cap"

Device Objects
""

"\Driver\Kbdclass"

"KeyboardClass0"

"\Driver\i8042prt"

""

"\Driver\ACPI"

"00000071"
33

DRIVER_OBJECT struct (on XP)

lkd> dt nt!_DRIVER_OBJECT +0x000 Type : Int2B +0x002 Size : Int2B +0x004 DeviceObject : Ptr32 _DEVICE_OBJECT +0x008 Flags : Uint4B +0x00c DriverStart : Ptr32 Void +0x010 DriverSize : Uint4B +0x014 DriverSection : Ptr32 Void +0x018 DriverExtension : Ptr32 _DRIVER_EXTENSION +0x01c DriverName : _UNICODE_STRING +0x024 HardwareDatabase : Ptr32 _UNICODE_STRING +0x028 FastIoDispatch : Ptr32 _FAST_IO_DISPATCH +0x02c DriverInit : Ptr32 long +0x030 DriverStartIo : Ptr32 void +0x034 DriverUnload : Ptr32 void +0x038 MajorFunction : [28] Ptr32 long

34

DEVICE_OBJECT struct (on XP)

typedef struct _DEVICE_OBJECT { CSHORT Type; USHORT Size; LONG ReferenceCount; struct _DRIVER_OBJECT * DriverObject; struct _DEVICE_OBJECT * NextDevice; struct _DEVICE_OBJECT * AttachedDevice; struct _IRP * CurrentIrp; PIO_TIMER Timer; ULONG Flags; ULONG Characteristics; __volatile PVPB Vpb; PVOID DeviceExtension; DEVICE_TYPE DeviceType; CCHAR StackSize; union { LIST_ENTRY ListEntry; WAIT_CONTEXT_BLOCK Wcb; } Queue; ULONG AlignmentRequirement; KDEVICE_QUEUE DeviceQueue; KDPC Dpc; ULONG ActiveThreadCount; PSECURITY_DESCRIPTOR SecurityDescriptor; KEVENT DeviceLock; USHORT SectorSize; USHORT Spare1; struct _DEVOBJ_EXTENSION * DeviceObjectExtension; PVOID Reserved; } DEVICE_OBJECT, *PDEVICE_OBJECT;

[Link]

"NextDevice: A pointer to the next device object, if any, that was created by the same driver. The I/O manager updates this list at each successful call to IoCreateDevice or IoCreateDeviceSecure." "The device object that is pointed to by the AttachedDevice member typically is the device object of a filter driver, which intercepts I/O requests originally targeted to the device represent by the device object. " 35

IRP struct
kd> dt _IRP ntdll!_IRP +0x000 Type : Int2B +0x002 Size : Uint2B +0x004 MdlAddress : Ptr32 _MDL +0x008 Flags : Uint4B +0x00c AssociatedIrp : __unnamed +0x010 ThreadListEntry : _LIST_ENTRY +0x018 IoStatus : _IO_STATUS_BLOCK +0x020 RequestorMode : Char +0x021 PendingReturned : UChar +0x022 StackCount : Char +0x023 CurrentLocation : Char +0x024 Cancel : UChar +0x025 CancelIrql : UChar +0x026 ApcEnvironment : Char +0x027 AllocationFlags : UChar +0x028 UserIosb : Ptr32 _IO_STATUS_BLOCK +0x02c UserEvent : Ptr32 _KEVENT +0x030 Overlay : __unnamed +0x038 CancelRoutine : Ptr32 void +0x03c UserBuffer : Ptr32 Void +0x040 Tail : __unnamed
(see wdm.h for comments on fields)

36

I am the very model of a modern major function bla

(in parody, it's important to maintain the correct number of syllableswes :P)

MajorFunction[] is an array of callback functions which will be called when IRPs are traversing the chain. This table is the target for function pointer hooking, in both legitimate and illegitimate software. So just like with the SSDT, you have to be aware of what 3rd party software might be hooking it.
37

major functions
(from wdm.h)

[Link] #define IRP_MJ_CREATE 0x00 #define IRP_MJ_CREATE_NAMED_PIPE 0x01 #define IRP_MJ_CLOSE 0x02 #define IRP_MJ_READ 0x03 #define IRP_MJ_WRITE 0x04 #define IRP_MJ_QUERY_INFORMATION 0x05 #define IRP_MJ_SET_INFORMATION 0x06 #define IRP_MJ_QUERY_EA 0x07 #define IRP_MJ_SET_EA 0x08 #define IRP_MJ_FLUSH_BUFFERS 0x09 #define IRP_MJ_QUERY_VOLUME_INFORMATION 0x0a #define IRP_MJ_SET_VOLUME_INFORMATION 0x0b #define IRP_MJ_DIRECTORY_CONTROL 0x0c #define IRP_MJ_FILE_SYSTEM_CONTROL 0x0d #define IRP_MJ_DEVICE_CONTROL 0x0e
38

major functions 2
(from wdm.h)

#define IRP_MJ_INTERNAL_DEVICE_CONTROL 0x0f #define IRP_MJ_SHUTDOWN 0x10 #define IRP_MJ_LOCK_CONTROL 0x11 #define IRP_MJ_CLEANUP 0x12 #define IRP_MJ_CREATE_MAILSLOT 0x13 #define IRP_MJ_QUERY_SECURITY 0x14 #define IRP_MJ_SET_SECURITY 0x15 #define IRP_MJ_POWER 0x16 #define IRP_MJ_SYSTEM_CONTROL 0x17 #define IRP_MJ_DEVICE_CHANGE 0x18 #define IRP_MJ_QUERY_QUOTA 0x19 #define IRP_MJ_SET_QUOTA 0x1a #define IRP_MJ_PNP 0x1b #define IRP_MJ_PNP_POWER IRP_MJ_PNP #define IRP_MJ_MAXIMUM_FUNCTION 0x1b

// Obsolete....
39

DeviceTree
[Link]

40

WinDbg (display device driver stack)

kd> !object \device\keyboardclass0 Object: 814e7d28 Type: (819b8ca0) Device ObjectHeader: 814e7d10 (old version) HandleCount: 0 PointerCount: 3 Directory Object: e1006948 Name: KeyboardClass0 kd> !devstack 814e7d28 !DevObj !DrvObj !DevExt ObjectName > 814e7d28 \Driver\Kbdclass 814e7de0 KeyboardClass0 814e7020 \Driver\i8042prt 814e70d8 8167c030 \Driver\ACPI 819a32e8 00000070 !DevNode 818f7348 : DeviceInst is "ACPI\PNP0303\4&5289e18&0" ServiceName is "i8042prt"
41

WinDbg 2 (display driver object)

kd> !devobj 814e7d28 Device object (814e7d28) is for: KeyboardClass0 \Driver\Kbdclass DriverObject 814ea0b8 Current Irp 00000000 RefCount 0 Type 0000000b Flags 00002044 Dacl e13cf7cc DevExt 814e7de0 DevObjExt 814e7ec0 ExtensionFlags (0000000000) AttachedTo (Lower) 814e7020 \Driver\i8042prt Device queue is not busy. kd> dt nt!_DRIVER_OBJECT 814ea0b8 +0x000 Type :4 +0x002 Size : 168 +0x004 DeviceObject : 0x81872030 _DEVICE_OBJECT +0x008 Flags : 0x12 +0x00c DriverStart : 0xf9c4c000 +0x010 DriverSize : 0x6000 +0x014 DriverSection : 0x819b7aa8 +0x018 DriverExtension : 0x814ea160 _DRIVER_EXTENSION +0x01c DriverName : _UNICODE_STRING "\Driver\Kbdclass" +0x024 HardwareDatabase : 0x80670de0 _UNICODE_STRING "\REGISTRY\MACHINE\HARDWARE\DESCRIPTION\SYSTEM" +0x028 FastIoDispatch : (null) +0x02c DriverInit : 0xf9c50610 long kbdclass!GsDriverEntry+0 +0x030 DriverStartIo : (null) +0x034 DriverUnload : (null) +0x038 MajorFunction : [28] 0xf9c4cdd0 long kbdclass!KeyboardClassCreate+0

42

WinDbg 3 (display next driver object)

kd> !devobj 814e7020 Device object (814e7020) is for: \Driver\i8042prt DriverObject 814ea410 Current Irp 00000000 RefCount 0 Type 00000027 Flags 00002004 DevExt 814e70d8 DevObjExt 814e7368 ExtensionFlags (0000000000) AttachedDevice (Upper) 814e7d28 \Driver\Kbdclass AttachedTo (Lower) 8167c030 \Driver\ACPI Device queue is not busy. kd> dt nt!_DRIVER_OBJECT 814ea410 +0x000 Type :4 +0x002 Size : 168 +0x004 DeviceObject : 0x817dda40 _DEVICE_OBJECT +0x008 Flags : 0x12 +0x00c DriverStart : 0xf9a2c000 +0x010 DriverSize : 0xcd00 +0x014 DriverSection : 0x81973070 +0x018 DriverExtension : 0x814ea4b8 _DRIVER_EXTENSION +0x01c DriverName : _UNICODE_STRING "\Driver\i8042prt" +0x024 HardwareDatabase : 0x80670de0 _UNICODE_STRING "\REGISTRY\MACHINE\HARDWARE\DESCRIPTION\SYSTEM" +0x028 FastIoDispatch : (null) +0x02c DriverInit : 0xf9a35285 long i8042prt!GsDriverEntry+0 +0x030 DriverStartIo : 0xf9a2c910 void i8042prt!I8xStartIo+0 +0x034 DriverUnload : 0xf9a32eb6 void i8042prt!I8xUnload+0 +0x038 MajorFunction : [28] 0xf9a2faa6 long i8042prt!I8xCreate+0

43

WinDbg 4 (print IRP table)

kd> dps 814ea410+0x38 L1C 814ea448 f9a2faa6 i8042prt!I8xCreate 814ea44c 804f355a nt!IopInvalidDeviceRequest 814ea450 f9a32e18 i8042prt!I8xClose 814ea454 804f355a nt!IopInvalidDeviceRequest 814ea458 804f355a nt!IopInvalidDeviceRequest 814ea45c 804f355a nt!IopInvalidDeviceRequest 814ea460 804f355a nt!IopInvalidDeviceRequest 814ea464 804f355a nt!IopInvalidDeviceRequest 814ea468 804f355a nt!IopInvalidDeviceRequest 814ea46c f9a2e1f9 i8042prt!I8xFlush 814ea470 804f355a nt!IopInvalidDeviceRequest 814ea474 804f355a nt!IopInvalidDeviceRequest 814ea478 804f355a nt!IopInvalidDeviceRequest 814ea47c 804f355a nt!IopInvalidDeviceRequest 814ea480 f9a32e4b i8042prt!I8xDeviceControl 814ea484 f9a2c836 i8042prt!I8xInternalDeviceControl 814ea488 804f355a nt!IopInvalidDeviceRequest 814ea48c 804f355a nt!IopInvalidDeviceRequest 814ea490 804f355a nt!IopInvalidDeviceRequest 814ea494 804f355a nt!IopInvalidDeviceRequest 814ea498 804f355a nt!IopInvalidDeviceRequest 814ea49c 804f355a nt!IopInvalidDeviceRequest 814ea4a0 f9a337ea i8042prt!I8xPower 814ea4a4 f9a2fa59 i8042prt!I8xSystemControl 814ea4a8 804f355a nt!IopInvalidDeviceRequest 814ea4ac 804f355a nt!IopInvalidDeviceRequest 814ea4b0 804f355a nt!IopInvalidDeviceRequest 814ea4b4 f9a2f990 i8042prt!I8xPnP

dps = display processor-sized pointer (meaning it decides whether it should be 16-64 bits), as a pointer to a symbol

dds = display dword as a pointer to a symbol

44

Stuxnet IRP filtering

[Link] curity_response/whitepapers/w32_stuxnet_dossier.pdf "The driver scans the following filesystem driver objects: \FileSystem\ntfs \FileSystem\fastfat \FileSystem\cdfs A new device object is created by Stuxnet and attached to the device chain for each device object managed by these driver objects. The [Link] driver will manage this driver object. By inserting such objects, Stuxnet is able to intercept IRP requests (example: writes, reads, to devices NTFS, FAT or CD-ROM devices)."
45

Stuxnet IRP filtering 2

"The driver monitors 'directory control' IRPs, in particular 'directory query' notifications. Such IRPs are sent to the device when a user program is browsing a directory, and requests the list of files it contains for instance."

46

He4Hook code
(from [Link])

NTSTATUS KIrpFilter::IrpHandler(){ if (
dwMajorFn == IRP_MJ_DIRECTORY_CONTROL && dwMinorFn == IRP_MN_QUERY_DIRECTORY

) { NtStatus = OnQueryDirectory(pIrp, pIrpStack, pDrvInfo); } }

47

Direct Kernel Object Manipulation (DKOM)

(It's a joke see? Since Distributed Component Object Model - DCOM - is a MS technology?)

Introduced by Jamie Butler in the FU rootkit. [Link] Recognized the prevailing technique of hooking was easily detected, so wrote a detector ("VICE - Catch the hookers!" ;)) DKOM perpetuates the arms race and shows the importance of information asymmetry for rootkits. The attacker reverse engineers a component he finds to be relevant to his goal. Then, having more understanding of the system than the defender, will likely succeed in having the manipulation go undetected.
48

Book page 428, 436

Canonical DKOM
Exploit the fact that the lists which [Link] or top (on *nix) consult to find running processes, are not the lists that the scheduler uses. Therefore things can be removed from those lists but they will still get to run. Also manipulates security tokens to elevate process or user privileges. Is DKOM, but isn't about hiding. But just like hooking does not automatically imply rootkit, DKOM can be used for non-rootkit ends
49

Process Linked List Before DKOM

50

From: [Link]

Process Linked List After DKOM

51

From: [Link]

Detecting DKOM
Different tools used different means to detect FU's process hiding. F-Secure BlackLight used a bruteforce where it calls OpenProcess() on all possible PIDs (which behind the scenes is just consulting PspCidTable, which has a handle for every open process. These handles are not hidden as part of DKOM.) It then calls CreateToolhelp32Snapshot() as another more traditional way to get a list of processes. Any discrepancy in the lists is deemed a hidden process. So Peter Silberman introduced FUTo ([Link] which bypassed BlackLight by manipulating the PspCidTable. Then Butler and Silberman put out RAIDE to detect the FUTo hiding too (using memory signature searching I believe) Klister by Rutkowska walked the list that the scheduler uses This is an example of "cross view detection"

52

FWIW: turns out

This change: For the record I emailed the GMER author and he said: "The 32-bit hex number that is after PID[0] is the EPROCESS structure pointer. In this case GMER cannot identify the name of process." If you look at my [Link] in the tiddlywiki install proceedure, you will see that I'm using [Link] to hide pid 4 ("system"). Sometimes system is called pid 0, sometimes pid 4 (on XP)

53

OS-Provided Callbacks
(The ones we want to highlight for the moment anyway. Go to the links and go up one level on the side bar to find more.) On registry actions:
CmRegisterCallback{Ex} [Link]

On process creation/deletion:
PsSetCreateProcessNotifyRoutine{Ex} [Link]

On thread creation/deletion:
PsSetCreateThreadNotifyRoutine{Ex} [Link]

On image load (e.g. DLL, EXE, SYS mapped into memory, imports resolved, but entry point not yet called):
PsSetLoadImageNotifyRoutine [Link]

Filesystem becoming active (to attach to with a filesystem filter driver):

IoRegisterFsRegistrationChange [Link]

System Shutdown:
IoRegisterShutdownNotification [Link]

54

Some example rootkit use of callbacks

He4Hook - PoC
Sets a callback with PsSetCreateThreadNotifyRoutine() (see the source)

FUTo - PoC
PsSetCreateProcessNotifyRoutine() (see source)

Black Energy 2, Rustock

[Link] Black Energy Thread notify, Rustock Process notify

HybridHook - PoC
Sets a callback with PsSetLoadImageNotifyRoutine() and does IAT hooking at load time (see the source)

TDSS/TDL3
[Link] PsSetLoadImageNotifyRoutine() to inject DLLs

Stuxnet
[Link] epapers/w32_stuxnet_dossier.pdf "The driver also registers to a filesystem registration callback routine in order to hook newly created filesystem objects on the fly."
55

Listing registered callbacks in WinDbg

[Link] - process/memory image load (PsSetCreateProcessNotifyRoutine[Ex]/PsSetImageLoadNotifyRoutine) [Link] - registry callbacks(CmRegisterCallback[Ex]) Here comes a new challenger! Hadoken! [Link] kd> $$>a<c:\[Link] ************************************************ * This command brought to you by [Link] * ************************************************ ************************************ * Printing image load callbacks... * ************************************ ********************************************** * Printing process notification callbacks... * ********************************************** 814ec008 ff2508605c81 jmp dword ptr ds:[815C6008h]
56

Listing registered callbacks

Newest Virus Blok Ada anti-rootkit has fairly comprehensive coverage.

57

NewTewYew - Master Boot Record (MBR)-infecting rootkits aka "Bootkits"

eEye Bootroot - 2005
Derek Soeder and Ryan Permeh From the [Link] "September 20, 2005: Scott Tenaglia provided a NASM port of the source, included as "[Link]". Many folks were not too keen on the MASM requirement, so this is a big boon to BootRootKit users at large. Thanks Scott!"

VBootkit - 2007, Vbootkit 2 - 2009

Nitin Kumar and Vipin Kumar

Stoned Bootkit - 2009

Peter Kleissner
58

Book page 380

What does an MBR actually look like?

[Link] [Link] [Link] [Link]

59

Bootkit Lives here (from disk), but in order to do anything of consequence it has to keep hooking each subsequent thing to keep control.

60

From [Link]

eEye Boot Root

[Link] y-Center/Research/Tools/BootRoot The first PoC, didn't actually change the MBR on the HD, instead booted from a disk which redirected to the normal boot process, hooking as it went.

61

VBootKit
Coined the term "bootkit" for master boot record infecting rootkits Vbootkit
[Link] First thing supporting Vista

Vbootkit 2
[Link] T2%20-%20Vipin%20and%20Nitin%20Kumar%20%20vbootkit%[Link] First thing to support boot subversion on Windows 7 x64 Payload includes disabling code signing & kernel patch protection (KPP aka PatchGuard)

62

Stoned Bootkit
[Link] Basically a weaponized bootkitso, surprise, surprise, it got used in real malware, and the German cops came knocking (illegal to distribute "hacking tools" in Germany). So now he doesn't distribute the full thing. Highly module to support many possible payloads Has a customization to work in the presence of TrueCrypt
63

bootkits in the wild

[Link]
Good dissection (w comparison to eEye boot root) Also shows adding one nop to asm bypassed MS, Kaspersky, F-Secure, and Sophos detection circa 2008

Mebroot w/ Torpig/Sinowal
[Link] p?docid=2008-010718-3448-99 [Link]
IRP hooking. How droll. How easy peasy

TDSS/TDL3
[Link]
IRP hooking. How droll. How easy peasy
64

Detecting bootkits
[Link] (though I think GMER has that mostly built in now) TrendMicro RootkitBuster claims to have removal Remove with "fixmbr" command from a windows install disk recovery mode Turn on your damn TPM! This, and bios subversion, are the whole reason we have trusted boot!
Get insPeCtoR from Corey Kallenberg or
65

An Integrity Measured Boot Process

OS OS Loader
TIME
Measure and store (PCR-4) OS components, other TCBrelevant stuff System runs normally! Measure OS Components Append PCR-3 with measurement Pass control to OS

BIOS CRTM Hardware

Measure Option ROMs Append PCR-1 with measurement--pass control, get it back Measure OS Loader, append PCR-2, pass control

Core Root of Trust for Measurement, (e.g. BIOS block) Measure itself and BIOS Append PCR-0 with hash of measurement

Clear PCRsnot really configurable

66

Loading code into kernel

Service Control Manager (SCM)
Leaves registry footprint

ZwSetSystemInformation() - [Link] - [Link] ZwLoadDriver() [Link]

Uses SCM, ZwSetSystemInformation, ZwLoadDriver

Windows < Vista used to be able to access \Device\PhysicalMemory

[Link]

67

Book page 198

Autoruns
Sysinternals tool to show the various places on the system that are set to automatically load extra code (either on boot, or when something else is loaded) [Link] Recent article "Analyzing a Stuxnet Infection with the Sysinternals Tools, Part 1"
[Link] hive/2011/03/30/[Link]
68

69