less is more Exploring code/process-less techniques and other weird-machine methods to hide code (and how to detect them). Slide URL: http://1drv.ms/1dJX2HI 09.09.99 Major References Win32 Buffer Overflows (p55, 09.09.99) dark spyrit / Barnaby Jack One of the greatest ever RIP Major References • Exploitation and state machines • Thomas Dullien / Halvar Flake • Sergey Bratus, TAOSSA (http://www.azimuthsecurity.com/) • Windows Kernel-mode Payload Fundamentals & A Catalog of Windows Local Kernel-mode Backdoor Techniques • Skape (mmiller@hick.org) (both) & Bugcheck (chris@bugcheck.org) & Skywing (Skywing@valhallalegends.com) • The page-fault weird machine: lessons in instruction-less computation • Julian Bangert, Sergey Bratus, Rebecca Shapiro, Sean W. Smith from WOOT'13 Proceedings of the 7th USENIX conference on Offensive Technologies • Volatility Framework • AAron Walters and open source contributors. Windows hardening • Windows XP does not memset(0) driver .text sections • Random slack can be executed • Updated 2k3+ • KINTERRUPT no longer has huge code templates/glue included as part of it’s structure • KINTERRUPT.DispatchCode is now 4 bytes (and always just points to a registered handler in the module) instead of up too 106 bytes of arbitrary code • Updated Vista+ • Page table entries secured • Win8 no longer has executable page table entries • Kernel 9200+ (8/2012) Kernel Pool (heap) is no longer default executable • This is a MAJOR win!!!!!! • No more huge degree’s of unknown executable memory to inspect Our target • Rootkit can shadow/move itself during dump’s • Issues from dumping memory from a live/physical system is problematic and has lead to an interesting arms race; • Using cold-boot attacks • Purpose built dumping hardware or commodity FireWire type inputs • Cause kernel panic to induce a dump • Windows Kernel 9600 (Windows 8.1/2012R2) • A snapshot from VMWare or Hyper-V • We will ignore dump acquisition issues for now and focus on VM snapshots X64 Kernel Virtual Address Space http://www.codemachine.com/article_x64kvas.html Start End Size Description Notes FFFF0800`00000000 FFFFF67F`FFFFFFFF 238TB Unused System Space WIN9600 NOW USE & CAN CONTAIN +X AREAS FFFFF680`00000000 FFFFF6FF`FFFFFFFF 512GB PTE Space -X used to be executable Win7 FFFFF700`00000000 FFFFF77F`FFFFFFFF 512GB HyperSpace 8.1 seems to have cleaned up here, 9200 had 1 +X page FFFFF780`00000000 FFFFF780`00000FFF 4K Shared System Page FFFFF780`00001000 FFFFF7FF`FFFFFFFF 512GB-4K System Cache Working Set FFFFF800`00000000 FFFFF87F`FFFFFFFF 512GB Initial Loader Mappings FFFFF880`00000000 FFFFF89F`FFFFFFFF 128GB Sys PTEs FFFFF8a0`00000000 FFFFF8bF`FFFFFFFF 128GB Paged Pool Area FFFFF900`00000000 FFFFF97F`FFFFFFFF 512GB Session Space FFFFF980`00000000 FFFFFa70`FFFFFFFF 1TB Dynamic Kernel VA Space FFFFFa80`00000000 *nt!MmNonPagedPoolStart-1 6TB Max PFN Database *nt!MmNonPagedPoolStart *nt!MmNonPagedPoolEnd 512GB Max Non-Paged Pool FFFFFFFF`FFc00000 FFFFFFFF`FFFFFFFF 4MB HAL and Loader Mappings Large Page (2MB) allocations DEFAULT NO EXECUTE Page Table Shellcode weird-machine • Win7 and earlier • Can we emit intended shellcode into PTE area? • Perform some VirtualAlloc from user space => executable memory in kernel • Just reserving memory writes PTE PXE at FFFFF6FB7DBEDF68 PPE at FFFFF6FB7DBEDF88 PDE at FFFFF6FB7DBF1008 PTE at FFFFF6FB7E201EA0 contains 0000000000187063 contains 0000000134C04863 contains 0000000100512863 contains 000000002DC3B863 pfn 187 ---DA--KWEV pfn 134c04 ---DA--KWEV pfn 100512 ---DA--KWEV pfn 2dc3b ---DA--KWEV • Page Table shell-code is non-trivial • Lots of gadgets! fffff6fb`7e201ea0 fffff6fb`7e201eb0 fffff6fb`7e201ec0 fffff6fb`7e201ed0 fffff6fb`7e201ee0 63 63 63 63 63 b8 98 78 58 38 c3 e3 63 53 c3 2d 2d 30 30 2e 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 63 63 63 63 63 a8 88 68 48 28 13 13 d3 a3 83 2f 2f 2e 2e 2e 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 c..-....c../.... c..-....c../.... cxc0....ch...... cXS0....cH...... c8......c(...... Win8 PT Shellcode attempting to run • Seemingly cleanly return • No double fault or bug check • We keep entering KiPageFault after returning from KiPageFault… Child-SP ffffd000`2b34ecf8 ffffd000`2b34ed00 ffffd000`2b34eda0 ffffd000`2b34eee0 ffffd000`2b34f078 ffffd000`2b34f080 ffffd000`2b34f9b0 ffffd000`2b34fb60 ffffd000`2b34fbd0 0000003a`ba9bf8f8 RetAddr fffff800`16066ee1 fffff800`1603f5ad fffff800`1615af2f fffff6fb`77fde37a fffff800`01e423fe fffff800`163ae3e5 fffff800`163aed7a fffff800`1615c4b3 00007ff9`c1b265ea 00007ff9`bef92c83 Call Site nt!LOCK_WORKING_SET nt!MiSystemFault+0x911 nt!MmAccessFault+0x7ed nt!KiPageFault+0x12f 0xfffff6fb`77fde37a SIoctl!SioctlDeviceControl+0x27e nt!IopXxxControlFile+0x845 nt!NtDeviceIoControlFile+0x56 nt!KiSystemServiceCopyEnd+0x13 ntdll!NtDeviceIoControlFile+0xa Defense: Rootkit revealing • Default non-execute pool space helps tremendously • Detect the presence of a rootkit by comparing results from multiple sources/abstraction layers • Physical (page tables) • Logical • Driver LIST_ENTRY • VAD • SECTION’s, … Tool evaluation • Implemented in .NET • Operates on direct physical memory dumps from VM snapshots • Demo script that identifies KVAS physical/logical sections • Transforms/Dumps memory / Generates hashes • Future • More well known blocks (local optimization)? • There’s some weird looking fill patterns often sitting around as exec; More page table checks, CR0.WP etc… Example VMWare ~8GB BlockWatch Folder: t:\BW_DEMO Dumping data from : Clone of Clone of Current Win 8.1 - PRO (2)-66bb942e.vmss Found probable kernel @ fffff800b508c000 Debug symbol being loaded for ntkrnlmp.pdb Kernel build number 9600 Root PT Entries: 16, SubTable Entries: 778263 Unlinked entry count: 1544 UnLinked Section: System Space @: 0xffffd00020180000L, Size: 0x1000L UnLinked Section: Loader Mappings @: 0xfffff800021d0000L, Size: 0x1000L UnLinked Section: HAL and Loader Mappings @: 0xffffffffffd02000L, Size: 0x1000L UnLinked Section: hal @: 0xfffff800b5000000L, Size: 0x200000L UnLinked Section: Loader Mappings @: 0xfffff800b5800000L, Size: 0x200000L …moving on; Attack! To the Unknown! • ffffd000201a0000 appears across Hyper-V & VMWare, reboots • Provides RoP gadgets • Fixed writeable executable memory location kd> !pte ffffd000`201a0000 VA ffffd000201a0000 PXE at FFFFF6FB7DBEDD00 contains 0000000000523863 pfn 523 ---DA--KWEV PPE at FFFFF6FB7DBA0000 contains 0000000000522863 pfn 522 ---DA--KWEV PDE at FFFFF6FB74000800 contains 0000000000527863 pfn 527 ---DA--KWEV PTE at FFFFF6E800100D00 contains 0000000000555963 pfn 555 -G-DA—KWEV !pool ffffd000`201a0000 Pool page ffffd000201a0000 region is Unknown ffffd000201a0000 is not a valid large pool allocation, checking large session pool... Unable to read large session pool table (Session data is not present in mini and kernel-only dumps) ffffd000201a0000 is not valid pool. Checking for freed (or corrupt) pool • Writable/Executable at a fixed address Can you guess what it is? • ?? ffffd000`201a0000 21 01 a0 00 00 00 00 80 42 47 49 4b 00 00 00 80 !.......BGIK.... ffffd000`201a01d0 00 19 00 23 00 01 2e 4c 00 00 00 10 67 6c 79 66 ...#...L....glyf ffffd000`201b2fb0 00 6e 00 74 00 65 00 6e 00 74 00 2e 00 53 00 65 .n.t.e.n.t...S.e ffffd000`201b2fc0 00 67 00 6f 00 65 00 20 00 55 00 49 00 03 00 00 .g.o.e. .U.I.... • Segoe_slboot.ttf • Starts at offset 0x1d0 • Initial bytes some sort of heap tag ? BG* • System boot/load time artifact A little more (past end of font) • Seems to have some basic heap structure pointer’s/allocation sizes • Unfortunately it’s all default executable/writable at a fixed address across systems/rebooting ffffd000`201b3000 ffffd000201b3020 0000000100002000 ffffd000201b3020 8000000000300121 … ffffd000`201b3020 4d 5a 90 00 03 00 00 00 04 00 00 00 ff ff 00 00 MZ.............. ffffd000`201b3070 69 73 20 70 72 6f 67 72 61 6d 20 63 61 6e 6e 6f is program canno … ffffd000`201b46b2 49 00 6e 00 74 00 65 00 72 00 6e 00 61 00 6c 00 4e 00 61 00 ffffd000`201b46c6 6d 00 65 00 00 00 62 00 6f 00 6f 00 74 00 72 00 65 00 73 00 I.n.t.e.r.n.a.l.N.a. m.e...b.o.o.t.r.e.s. … • This leaves a lot of room for RoP gadgets (MZ is only .rsrc, why +x?) BIOS Ranges • Platform specific (vmware in this case) • 2012R2 0xffffd00020500000, 8.1 0xffffd00020600000 • Fixed address across reboots (size is 241,664 - 0x3B000) • Physical system dumps ffffd000`206c5a50 ffffd000`206c5a60 ffffd000`206c5a70 ffffd000`206c5a80 ffffd000`206c5a90 c3 75 72 2c 72 32 61 65 20 74 2e 6c 2c 49 20 30 20 20 6e 32 00 6d 49 63 2e 56 61 6e 2e 30 4d 63 63 20 00 77 68 00 56 90 61 69 56 42 c8 72 6e 20 45 02 65 65 4d 20 00 20 00 20 73 00 76 56 77 75 c4 69 4d 61 70 5e 72 77 72 70 04 74 61 65 6f 33 .2.0.VMware virt ual machine.VMwa re, Inc.V M ware , Inc. VBE suppo rt 2.0.......^.3 3: kd> !pte ffffd000`206c5a50 PXE at FFFFF6FB7DBEDD00 contains 0000000000B22863 pfn b22 ---DA--KWEV VA ffffd000206c5a50 PPE at FFFFF6FB7DBA0000 PDE at FFFFF6FB74000818 contains 0000000000B21863 contains 0000000000852863 pfn b21 ---DA--KWEV pfn 852 ---DA--KWEV PTE at FFFFF6E800103628 contains 00000000000C5963 pfn c5 -G-DA--KWEV Other/More dynamic/Misc Areas • Slack • Audit MDL structures • Session Space • ACPI FACS -- exec • Firmware ACPI Control Structure • Verify ACPI with wite list 0: kd> !pool ffffe00000420000 Pool page ffffe00000420000 region is Nonpaged pool *ffffe00000420000 size: 90 previous size: 0 (Allocated) *VM3D Pooltag VM3D : Volume Manager, Binary : volmgr.sys 0: kd> !pool ffffe000`00418000 Pool page ffffe00000418000 region is Nonpaged pool *ffffe00000418000 size: 90 previous size: 0 (Allocated) *VM3D Pooltag VM3D : Volume Manager, Binary : volmgr.sys • Shim Engine (i.e. handling for drvmain.sdb) • Bootloader artifacts • Volume manager heap Related topics • White list extracted bootmgr.exe • Well-known pages • NULL, all set, GUARD • Interesting/weird fill patterns • 2007, ###### • Iiiiii (0x69;) • Make sure their not gadget’able… Other Gadget Areas • There are other +X areas, in the region, but have small variability in their allocation • Windows Boot manager, network boot support code, more font areas Defense: RoP Detection • Spurious Saved Return Addresses • Sometimes RoP Gadget is just random data present in an executable section!!! • All existing RoP Databases or techniques target arbitrary saved return addresses • https://www.corelan.be/index.php/security/corelanropdb/#advapi32dll_8211_5126005755 • 0x77e25c1f, # POP EAX # RETN • Saved Return should be • Simple/Effective/Very reliable reducing gadget surface area Spurious Saved Return Addresses Validation • Conceptually similar to heap back-checking logical links except we walk the stack • Think Heap/Pool verification • Verify op-code preceding saved return address • Adding into BlockWatch • Our operation is static so performance is no big deal and we like to be current! • Some performance impact if implemented at run time • May not reduce the gadget surface area sufficiently Attack: RoP Compiler • Gadget surface area is reduced by SSRAV • • • Use a gadget compiler from verifiable gadgets only Work done from Codeless Pagefaulting; … a “A one Instruction Computer” with A move-branch-if-zero-or-decrement instruction, short movdbz. ... It has been proven that … is Turing-complete… Defense: Finally • RoP compiler’s are neat, but with adequate surface area reduction of dependable RoP gadgets • Remove EXECUTE from unneeded areas • Decommit/wipe unused • RoP chain will be really huge and will have fairly obvious looking characteristics • Even on AMD64 max stack is 4GB; “the maximum size of a segment (ss).” • Detecting codeless-pagefaulting • Specifically; Scan/Verify GDT and TSS • Generally; Performance drain, counters, clock skew, context switches, accounting, … So What? • Page table verifier identifies hidden areas • White list as much as possible • High 99% • Defend RoP attacks with SSRA checking Comprehensive verification • Forensics • Reduction / Analysis aid • APT Detection • Diffing • White list Dependencies? • Require NX • SMEP Reccomended What about script hosts? • Instrument / Profile scripts to generate white lists • Doable for .NET • More difficult for PHP and company? Demo