04/19/2005 Note: We've tried to include most of the topics that will be covered in our presentation. However, many of these things will be discussed, but will not be our main concentration. The majority of our presentation will be around topic V. C, our new post-exploitation advances, and how they will be designed and integrated into Metasploit 3. Metasploit 3 won't be finished in time for Blackhat, but more important than the software itself is the work we've done on developing the individual components that it will be built from. This outline is pretty indepth, so it's hard to pick out all the nuggets of coolness. We've developed many really useful technologies in the area of payload development. These are things we've created since last year's Blackhat, including many advances in polymorphic shellcode, and high-permutation attacks. I. Introduction A. Who are we? 1. spoonm 2. skape B. Exploit Technology 1. Three Phases of Exploitation a. Pre-exploitation b. Exploitation c. Post-exploitation 2. Pre-exploitation - Preparing the payload a. NOP generation b. Payload encoders c. Connection handler initialization 3. Exploitation - Leveraging the vulnerability a. Stack overflows c. Heap overflows b. SEH overwrites d. Arbitrary pointer overwrites 4. Post-Exploitation - Manipulating the target a. Command shell redirection b. Arbitrary command execution c. Pivoting payloads d. Advanced payload interaction B. Where do we stand? 1. Pre-exploitation a. Robust and elegant encoders do exist i. SkyLined's alpha-numeric encoder ii. Spoonm's Shikata Ga Nai b. Payload encoders generally taken for granted i. Most encoders are static with a variable key ii. IDS able to signature static decoder stubs c. NOP generation hasn't publically changed much i. PoC exploits generally use predictable nops, if any ii. ADMmutate easily signatured by most NIDS (Snort, Fnord) iii. Not considered very important to many researchers d. NIDS deployments are playing chase the tail i. The mouse always has the advantage; NIDS is reactive ii. Advanced nops and encoders push NIDS to its limits 2. Exploitation a. Techniques have become very mature i. Linux/BSD exploitation techniques largely unchanged ii. Win32 heap overflows now more reliable (oded/shok) iii. Win32 SEH overwrites make exploitation easy, even on XPSP2 b. Exploitation topics have been beaten to death 3. Post-exploitation a. Common payloads are limited i. Command shell interaction has poor automation support ii. Limited to the command set that the interpreter provides iii. Bounded by the utilities installed on the target machine iv. Restrictive environments (chroot) can hinder command execution v. HIPS vendors becoming more adept at detecting basic payloads - LoadLibraryA from the stack, etc (mcafee 8.0i) b. Communication vectors largely unchanged i. Reverse and port-bind payloads still the most common ii. Findsock style payloads still unused by PoC exploits iii. Alternative communication vectors rarely discussed c. Pivoting technology exists i. Core ST described system call proxying in 2002 ii. Metasploit's (2.3) meterpreter provides basic network pivoting C. What will we discuss? 1. Pre-exploitation Research a. NOP generation i. Opty2 b. Encoders i. Additive feedback encoders ii. Shikata Ga Nai 2. Post-Exploitation Research a. Library Injection i. Facilitates things like Meterpreter and VNC injection b. VNC Injection i. Introduced at blackhat04 with the Metasploit 2.2 release c. Meterpreter & dN i. The cross-platform post-exploitation suite 3. Payload Research a. PassiveX i. Taking advantage of ActiveX controls b. Ordinal Stagers i. Tiny network capable stagers c. Egghunt (maybe?) i. Small payload capable of locating a larger payload ii. Useful for exploits with limited space II. Pre-exploitation A. OptyNop2 1. Creation and benefits of multi-byte nopsleds B. Additive feedback encoders C. Shikata Ga Nai 1. High permutation shellcode through simple dependency mappings III. Post-exploitation A. Library Injection 1. Overview a. Paper published in 04/2004 b. Provides advanced code execution c. Code can be written in any language that can compile down to a shared object. d. Allows developer to use all of the APIs exposed to a normal application. e. Detailed explanation can be found in included resources (include lib inject paper) 2. Two types of library injection a. On-Disk i. Library is loaded from the target's harddrive or via a file share. ii. Can be detected by AV products due to fs access b. In-Memory i. Library is uploaded to the target and loaded from memory without touching the disk ii. Evades file system filter drivers, such as those provided by AV companies iii. Not touching the disk means no forensic trace iv. VirtualLock prevents swapping to disk, but requires admin 3. In-memory library injection on Windows a. System calls used by the library loader are hooked i. NtCreateFile ii. NtMapViewOfSection iii. etc b. Unique image name is used to identify the image to image that is in memory c. System call hooks are removed so that future injectino can occur d. Alternative approaches i. Could do client-side relocations, but would need to handle import processing 4. In-memory library injection on Linux/BSD a. No known public implementations b. Requires alternate approach i. Hooking API routines not always possible -- symtab not mapped into memory ii. libc symbol version mismatches lead to linking nightmares c. Client-side relocations seem most feasible i. Remote side maps a region of memory and sends the client the base address ii. Client processes relocations and transmits the relocated image as its mapped segment would appear iii. Requires locating rtld base so that PLT lookups will work B. VNC Injection 1. Implements VNC as an injectable DLL a. Uses RealVNC as the code-base b. VNC communication uses the exploit connection c. No physical trace is left of the VNC server d. Can operate regardless of existing VNC installations 2. Easy way to illustrate insecurities a. Suits understand mouse movement better than black box command prompts C. Meterpreter 1. Generic post-exploitation suite a. Based on library injection b. Uses the established exploit connection i. Especially powerful with findsock payloads; no new connection c. Executes entirely from memory d. No new processes or file access required for the payload to succeed e. Detailed explanation can be found in included resources (include meterpreter paper) 2. Extension system provides advanced automation support a. No need to hand write tedious assembly b. Existing native code can be ported to a meterpreter extension 3. Architecture a. Design goals i. Very flexible protocol; should adapt to extension requirements ii. Exposure of a channelized communication system to extensions iii. Stealth operation iv. Should be portable to various operating systems v. Client from one platform should work with server on another platform b. Protocol i. Uses TLVs (Type-Length-Value) to support arbitrary data passing. ii. TLVs allow the packet parser to be oblivious to the structure of the value field iii. Type field is broken down into meta types 4. Core interface a. Overview i. Minimal interface to support the loading of extensions ii. Implements the basic packet transmission and dispatching facilities iii. Exposes channel allocation and management to extensions b. Advanced features i. Migrating the server instance between processes 5. The ``stdapi'' extension a. Overview i. Included in Metasploit 3.0 ii. Provides access to some of the common subsystems of the target operating system iii. Allows for easy automation and implementation of robust post-exploitation scripts b. File System i. File and directory interaction ii. Files can be uploaded and downloaded between the attacker and the target c. Network i. Transparent network pivoting ii. Route table enumeration and manipulation iii. Local interface enumeration d. Process i. Process execution, optionally with channelized IO ii. Enumeration of running processes iii. Modification of arbitrary memory iv. Creation and modification of running threads v. Loading and interacting with shared object files e. Registry i. General registry API access ii. Opening, creating, and removing registry keys iii. Setting, querying, and deleting registry values iv. Enumeration of both keys and values f. User interface i. Disabling local user interaction via the keyboard and/or mouse (similar to VNC) ii. Idle timeout checking to see how long it's been since the user did something interactive 6. The ``priv'' extension a. Still in development b. Exposes an interface to escalating local privileges through local vulnerabilities c. SAM dumping support similar to pwdump3 D. dN 1. Simple low-footprint post-exploitation tool 2. Useful to scope out execution enviroment, then bootstrap other tools IV. Payload Research A. PassiveX 1. Overview a. Post-exploitation payload b. A derivative of On-Disk library injection that uses ActiveX controls c. Supports arbitrary DLL injection in any language that can be compiled as an ActiveX control (C++, VB, etc) d. Detailed analysis can be found in included resources (include passivex paper) 2. Payload Implementation a. Disables iexplore Internet zone restrictions on ActiveX controls i. Modifies four registry values that are stored per-user b. Launches a hidden iexplore at a URL with an embedded OCX i. The OCX does not have to be signed ii. No user interaction is required iii. OCX is automatically downloaded, registered, and loaded into the browser's context 3. Sample HTTP tunneling ActiveX control a. HTTP GET/POST used to build tunnel to attacker i. Uses the WININET API ii. Outbound traffic from target machine encapsulated in POST request iii. Inbound traffic from attacker encapsulated in GET response iv. Proxy configuration automatically inherited v. Requires HTTP server capable of performing encap/decap on the HTTP packets b. Streaming connection through HTTP tunnel can be created i. socketpair doesn't exist natively on win32 but can be simulated with a local listener ii. Streaming abstraction allows advanced payloads to transparently use the HTTP transport (Meterpreter, VNC) iii. Local listener less covert, but highly beneficial 4. Pros a. Bypasses restrictive outbound filters b. Re-uses proxy configuration c. Looks like normal user traffic d. Allows full access to the win32 API like all forms of library injection 5. Cons a. Touches the disk b. Requires administrative access i. Internet Explorer prohibits the downloading of ActiveX controls as non-admin B. Windows Ordinal Stagers 1. Overview a. Technique from Oded's lightning talk from core04 b. Uses static ordinals to locate winsock symbol addresses c. Compatible with all versions of Windows d. 92 byte reverse stager, 93 byte findsock stager e. Detailed explanation can be found in included resources (include spoonm ordinal paper) 2. Reverse Ordinal Stager a. Walks InitOrderModuleList searching for ws2_32 b. Uses static ordinals to index the export table c. Creates fd with socket d. Chains connect and recv frames e. Returns into buffer read in from file descriptor C. Egghunting 1. Overview a. Small stub payload that can search for a larger payload b. Useful for exploits that have limited payload space but can stash more payload elsewhere in memory c. Example exploits include the IE object type vulnerability. d. Goal is to safely search target address space for the larger payload. e. Larger payload is located by searching for an egg that is prepended to it. f. Detailed analysis can be found in included resources (include egghunt paper) 2. Two primary methods of egghunting on Windows a. SEH 1. 60 bytes in size, searches for an 8 byte egg, compatible with all versions of Windows (including 9x) 2. Installs a custom exception handler 3. Begins walking the address space i. When a bad address is encounter, the current pointer is incremented by PAGE_SIZE ii. When a mismatch of the egg is encountered, the current pointer is incremented by one 4. Each address is compared against the 8-byte egg i. Bad addresses will throw exceptions which will be handled by the custom exception handler 5. When the egg is found, jump past the egg into the larger payload b. System call 1. Very small stub (32 bytes), searches for an 8 byte egg, only compatible with NT+ 2. Uses a non-intrusive system call (NtDisplayString) to search validate addresses i. Same page walking logic as SEH is used 3. When the egg is found, jump past the egg into the larger payload 3. One primary method of egghunting on Linux a. System call 1. 30 bytes in size, searches for an 8 byte egg 2. System call technique applies to other UNIX variants as well 3. Uses the sigaction system call to validate 16 bytes at once 4. Pros i. Very small 5. Cons i. Corner cases may cause it to be unstable V. The Complete Picture - Metasploit 3.0 A. The power of pre-exploitation prepartion 1. Abstract NOP generation 2. Abstract payload encoding 3. Abstract exploit connection handlers i. Bind handler ii. Reverse handler iii. Findsock handler B. The power of exploit generalization 1. Standardized exploit creation interface i. Robust target definitions ii. Detailed exploit implementation information iii. Entirely payload independent, no hardcoding paylaods C. The power of post-exploitation automation 1. Complicated tasks made simple with scripting i. Want to download a targets entire harddrive? ii. Want to disable local user interaction? iii. Want to upload and play an mp3 on the target? iv. Want to 'be' on the target's network? 2. Standard interface means cross platform support VI. Conclusion - where do we go from here? A. Future Post-exploitation research 1. Mesh network support 2. Expansion of the stdapi described in this document included resources: http://www.hick.org/code/skape/papers/egghunt-shellcode.pdf http://www.hick.org/code/skape/papers/meterpreter.pdf http://www.hick.org/code/skape/papers/remote-library-injection.pdf http://www.hick.org/code/skape/papers/passivex.pdf [not released yet] http://www.metasploit.com/users/spoonm/ordinals.txt cited material: STRIDE: http://www.ics.forth.gr/carv/acs/ACS/Activities/papers/stride-IFIP-SEC05.pdf syscall proxy: http://www.coresecurity.com/files/files/11/SyscallProxying.pdf